xref: /openbmc/linux/kernel/trace/ftrace.c (revision 293d5b43)
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 	function_profile_call(trace->func, 0, NULL, NULL);
876 	return 1;
877 }
878 
879 static void profile_graph_return(struct ftrace_graph_ret *trace)
880 {
881 	struct ftrace_profile_stat *stat;
882 	unsigned long long calltime;
883 	struct ftrace_profile *rec;
884 	unsigned long flags;
885 
886 	local_irq_save(flags);
887 	stat = this_cpu_ptr(&ftrace_profile_stats);
888 	if (!stat->hash || !ftrace_profile_enabled)
889 		goto out;
890 
891 	/* If the calltime was zero'd ignore it */
892 	if (!trace->calltime)
893 		goto out;
894 
895 	calltime = trace->rettime - trace->calltime;
896 
897 	if (!fgraph_graph_time) {
898 		int index;
899 
900 		index = trace->depth;
901 
902 		/* Append this call time to the parent time to subtract */
903 		if (index)
904 			current->ret_stack[index - 1].subtime += calltime;
905 
906 		if (current->ret_stack[index].subtime < calltime)
907 			calltime -= current->ret_stack[index].subtime;
908 		else
909 			calltime = 0;
910 	}
911 
912 	rec = ftrace_find_profiled_func(stat, trace->func);
913 	if (rec) {
914 		rec->time += calltime;
915 		rec->time_squared += calltime * calltime;
916 	}
917 
918  out:
919 	local_irq_restore(flags);
920 }
921 
922 static int register_ftrace_profiler(void)
923 {
924 	return register_ftrace_graph(&profile_graph_return,
925 				     &profile_graph_entry);
926 }
927 
928 static void unregister_ftrace_profiler(void)
929 {
930 	unregister_ftrace_graph();
931 }
932 #else
933 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
934 	.func		= function_profile_call,
935 	.flags		= FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
936 	INIT_OPS_HASH(ftrace_profile_ops)
937 };
938 
939 static int register_ftrace_profiler(void)
940 {
941 	return register_ftrace_function(&ftrace_profile_ops);
942 }
943 
944 static void unregister_ftrace_profiler(void)
945 {
946 	unregister_ftrace_function(&ftrace_profile_ops);
947 }
948 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
949 
950 static ssize_t
951 ftrace_profile_write(struct file *filp, const char __user *ubuf,
952 		     size_t cnt, loff_t *ppos)
953 {
954 	unsigned long val;
955 	int ret;
956 
957 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
958 	if (ret)
959 		return ret;
960 
961 	val = !!val;
962 
963 	mutex_lock(&ftrace_profile_lock);
964 	if (ftrace_profile_enabled ^ val) {
965 		if (val) {
966 			ret = ftrace_profile_init();
967 			if (ret < 0) {
968 				cnt = ret;
969 				goto out;
970 			}
971 
972 			ret = register_ftrace_profiler();
973 			if (ret < 0) {
974 				cnt = ret;
975 				goto out;
976 			}
977 			ftrace_profile_enabled = 1;
978 		} else {
979 			ftrace_profile_enabled = 0;
980 			/*
981 			 * unregister_ftrace_profiler calls stop_machine
982 			 * so this acts like an synchronize_sched.
983 			 */
984 			unregister_ftrace_profiler();
985 		}
986 	}
987  out:
988 	mutex_unlock(&ftrace_profile_lock);
989 
990 	*ppos += cnt;
991 
992 	return cnt;
993 }
994 
995 static ssize_t
996 ftrace_profile_read(struct file *filp, char __user *ubuf,
997 		     size_t cnt, loff_t *ppos)
998 {
999 	char buf[64];		/* big enough to hold a number */
1000 	int r;
1001 
1002 	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1003 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1004 }
1005 
1006 static const struct file_operations ftrace_profile_fops = {
1007 	.open		= tracing_open_generic,
1008 	.read		= ftrace_profile_read,
1009 	.write		= ftrace_profile_write,
1010 	.llseek		= default_llseek,
1011 };
1012 
1013 /* used to initialize the real stat files */
1014 static struct tracer_stat function_stats __initdata = {
1015 	.name		= "functions",
1016 	.stat_start	= function_stat_start,
1017 	.stat_next	= function_stat_next,
1018 	.stat_cmp	= function_stat_cmp,
1019 	.stat_headers	= function_stat_headers,
1020 	.stat_show	= function_stat_show
1021 };
1022 
1023 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1024 {
1025 	struct ftrace_profile_stat *stat;
1026 	struct dentry *entry;
1027 	char *name;
1028 	int ret;
1029 	int cpu;
1030 
1031 	for_each_possible_cpu(cpu) {
1032 		stat = &per_cpu(ftrace_profile_stats, cpu);
1033 
1034 		name = kasprintf(GFP_KERNEL, "function%d", cpu);
1035 		if (!name) {
1036 			/*
1037 			 * The files created are permanent, if something happens
1038 			 * we still do not free memory.
1039 			 */
1040 			WARN(1,
1041 			     "Could not allocate stat file for cpu %d\n",
1042 			     cpu);
1043 			return;
1044 		}
1045 		stat->stat = function_stats;
1046 		stat->stat.name = name;
1047 		ret = register_stat_tracer(&stat->stat);
1048 		if (ret) {
1049 			WARN(1,
1050 			     "Could not register function stat for cpu %d\n",
1051 			     cpu);
1052 			kfree(name);
1053 			return;
1054 		}
1055 	}
1056 
1057 	entry = tracefs_create_file("function_profile_enabled", 0644,
1058 				    d_tracer, NULL, &ftrace_profile_fops);
1059 	if (!entry)
1060 		pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1061 }
1062 
1063 #else /* CONFIG_FUNCTION_PROFILER */
1064 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1065 {
1066 }
1067 #endif /* CONFIG_FUNCTION_PROFILER */
1068 
1069 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1070 
1071 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1072 static int ftrace_graph_active;
1073 #else
1074 # define ftrace_graph_active 0
1075 #endif
1076 
1077 #ifdef CONFIG_DYNAMIC_FTRACE
1078 
1079 static struct ftrace_ops *removed_ops;
1080 
1081 /*
1082  * Set when doing a global update, like enabling all recs or disabling them.
1083  * It is not set when just updating a single ftrace_ops.
1084  */
1085 static bool update_all_ops;
1086 
1087 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1088 # error Dynamic ftrace depends on MCOUNT_RECORD
1089 #endif
1090 
1091 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1092 
1093 struct ftrace_func_probe {
1094 	struct hlist_node	node;
1095 	struct ftrace_probe_ops	*ops;
1096 	unsigned long		flags;
1097 	unsigned long		ip;
1098 	void			*data;
1099 	struct list_head	free_list;
1100 };
1101 
1102 struct ftrace_func_entry {
1103 	struct hlist_node hlist;
1104 	unsigned long ip;
1105 };
1106 
1107 struct ftrace_hash {
1108 	unsigned long		size_bits;
1109 	struct hlist_head	*buckets;
1110 	unsigned long		count;
1111 	struct rcu_head		rcu;
1112 };
1113 
1114 /*
1115  * We make these constant because no one should touch them,
1116  * but they are used as the default "empty hash", to avoid allocating
1117  * it all the time. These are in a read only section such that if
1118  * anyone does try to modify it, it will cause an exception.
1119  */
1120 static const struct hlist_head empty_buckets[1];
1121 static const struct ftrace_hash empty_hash = {
1122 	.buckets = (struct hlist_head *)empty_buckets,
1123 };
1124 #define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1125 
1126 static struct ftrace_ops global_ops = {
1127 	.func				= ftrace_stub,
1128 	.local_hash.notrace_hash	= EMPTY_HASH,
1129 	.local_hash.filter_hash		= EMPTY_HASH,
1130 	INIT_OPS_HASH(global_ops)
1131 	.flags				= FTRACE_OPS_FL_RECURSION_SAFE |
1132 					  FTRACE_OPS_FL_INITIALIZED |
1133 					  FTRACE_OPS_FL_PID,
1134 };
1135 
1136 /*
1137  * This is used by __kernel_text_address() to return true if the
1138  * address is on a dynamically allocated trampoline that would
1139  * not return true for either core_kernel_text() or
1140  * is_module_text_address().
1141  */
1142 bool is_ftrace_trampoline(unsigned long addr)
1143 {
1144 	struct ftrace_ops *op;
1145 	bool ret = false;
1146 
1147 	/*
1148 	 * Some of the ops may be dynamically allocated,
1149 	 * they are freed after a synchronize_sched().
1150 	 */
1151 	preempt_disable_notrace();
1152 
1153 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1154 		/*
1155 		 * This is to check for dynamically allocated trampolines.
1156 		 * Trampolines that are in kernel text will have
1157 		 * core_kernel_text() return true.
1158 		 */
1159 		if (op->trampoline && op->trampoline_size)
1160 			if (addr >= op->trampoline &&
1161 			    addr < op->trampoline + op->trampoline_size) {
1162 				ret = true;
1163 				goto out;
1164 			}
1165 	} while_for_each_ftrace_op(op);
1166 
1167  out:
1168 	preempt_enable_notrace();
1169 
1170 	return ret;
1171 }
1172 
1173 struct ftrace_page {
1174 	struct ftrace_page	*next;
1175 	struct dyn_ftrace	*records;
1176 	int			index;
1177 	int			size;
1178 };
1179 
1180 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1181 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1182 
1183 /* estimate from running different kernels */
1184 #define NR_TO_INIT		10000
1185 
1186 static struct ftrace_page	*ftrace_pages_start;
1187 static struct ftrace_page	*ftrace_pages;
1188 
1189 static bool __always_inline ftrace_hash_empty(struct ftrace_hash *hash)
1190 {
1191 	return !hash || !hash->count;
1192 }
1193 
1194 static struct ftrace_func_entry *
1195 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1196 {
1197 	unsigned long key;
1198 	struct ftrace_func_entry *entry;
1199 	struct hlist_head *hhd;
1200 
1201 	if (ftrace_hash_empty(hash))
1202 		return NULL;
1203 
1204 	if (hash->size_bits > 0)
1205 		key = hash_long(ip, hash->size_bits);
1206 	else
1207 		key = 0;
1208 
1209 	hhd = &hash->buckets[key];
1210 
1211 	hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1212 		if (entry->ip == ip)
1213 			return entry;
1214 	}
1215 	return NULL;
1216 }
1217 
1218 static void __add_hash_entry(struct ftrace_hash *hash,
1219 			     struct ftrace_func_entry *entry)
1220 {
1221 	struct hlist_head *hhd;
1222 	unsigned long key;
1223 
1224 	if (hash->size_bits)
1225 		key = hash_long(entry->ip, hash->size_bits);
1226 	else
1227 		key = 0;
1228 
1229 	hhd = &hash->buckets[key];
1230 	hlist_add_head(&entry->hlist, hhd);
1231 	hash->count++;
1232 }
1233 
1234 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1235 {
1236 	struct ftrace_func_entry *entry;
1237 
1238 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1239 	if (!entry)
1240 		return -ENOMEM;
1241 
1242 	entry->ip = ip;
1243 	__add_hash_entry(hash, entry);
1244 
1245 	return 0;
1246 }
1247 
1248 static void
1249 free_hash_entry(struct ftrace_hash *hash,
1250 		  struct ftrace_func_entry *entry)
1251 {
1252 	hlist_del(&entry->hlist);
1253 	kfree(entry);
1254 	hash->count--;
1255 }
1256 
1257 static void
1258 remove_hash_entry(struct ftrace_hash *hash,
1259 		  struct ftrace_func_entry *entry)
1260 {
1261 	hlist_del(&entry->hlist);
1262 	hash->count--;
1263 }
1264 
1265 static void ftrace_hash_clear(struct ftrace_hash *hash)
1266 {
1267 	struct hlist_head *hhd;
1268 	struct hlist_node *tn;
1269 	struct ftrace_func_entry *entry;
1270 	int size = 1 << hash->size_bits;
1271 	int i;
1272 
1273 	if (!hash->count)
1274 		return;
1275 
1276 	for (i = 0; i < size; i++) {
1277 		hhd = &hash->buckets[i];
1278 		hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1279 			free_hash_entry(hash, entry);
1280 	}
1281 	FTRACE_WARN_ON(hash->count);
1282 }
1283 
1284 static void free_ftrace_hash(struct ftrace_hash *hash)
1285 {
1286 	if (!hash || hash == EMPTY_HASH)
1287 		return;
1288 	ftrace_hash_clear(hash);
1289 	kfree(hash->buckets);
1290 	kfree(hash);
1291 }
1292 
1293 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1294 {
1295 	struct ftrace_hash *hash;
1296 
1297 	hash = container_of(rcu, struct ftrace_hash, rcu);
1298 	free_ftrace_hash(hash);
1299 }
1300 
1301 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1302 {
1303 	if (!hash || hash == EMPTY_HASH)
1304 		return;
1305 	call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1306 }
1307 
1308 void ftrace_free_filter(struct ftrace_ops *ops)
1309 {
1310 	ftrace_ops_init(ops);
1311 	free_ftrace_hash(ops->func_hash->filter_hash);
1312 	free_ftrace_hash(ops->func_hash->notrace_hash);
1313 }
1314 
1315 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1316 {
1317 	struct ftrace_hash *hash;
1318 	int size;
1319 
1320 	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1321 	if (!hash)
1322 		return NULL;
1323 
1324 	size = 1 << size_bits;
1325 	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1326 
1327 	if (!hash->buckets) {
1328 		kfree(hash);
1329 		return NULL;
1330 	}
1331 
1332 	hash->size_bits = size_bits;
1333 
1334 	return hash;
1335 }
1336 
1337 static struct ftrace_hash *
1338 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1339 {
1340 	struct ftrace_func_entry *entry;
1341 	struct ftrace_hash *new_hash;
1342 	int size;
1343 	int ret;
1344 	int i;
1345 
1346 	new_hash = alloc_ftrace_hash(size_bits);
1347 	if (!new_hash)
1348 		return NULL;
1349 
1350 	/* Empty hash? */
1351 	if (ftrace_hash_empty(hash))
1352 		return new_hash;
1353 
1354 	size = 1 << hash->size_bits;
1355 	for (i = 0; i < size; i++) {
1356 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1357 			ret = add_hash_entry(new_hash, entry->ip);
1358 			if (ret < 0)
1359 				goto free_hash;
1360 		}
1361 	}
1362 
1363 	FTRACE_WARN_ON(new_hash->count != hash->count);
1364 
1365 	return new_hash;
1366 
1367  free_hash:
1368 	free_ftrace_hash(new_hash);
1369 	return NULL;
1370 }
1371 
1372 static void
1373 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1374 static void
1375 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1376 
1377 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1378 				       struct ftrace_hash *new_hash);
1379 
1380 static int
1381 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1382 		 struct ftrace_hash **dst, struct ftrace_hash *src)
1383 {
1384 	struct ftrace_func_entry *entry;
1385 	struct hlist_node *tn;
1386 	struct hlist_head *hhd;
1387 	struct ftrace_hash *new_hash;
1388 	int size = src->count;
1389 	int bits = 0;
1390 	int ret;
1391 	int i;
1392 
1393 	/* Reject setting notrace hash on IPMODIFY ftrace_ops */
1394 	if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1395 		return -EINVAL;
1396 
1397 	/*
1398 	 * If the new source is empty, just free dst and assign it
1399 	 * the empty_hash.
1400 	 */
1401 	if (!src->count) {
1402 		new_hash = EMPTY_HASH;
1403 		goto update;
1404 	}
1405 
1406 	/*
1407 	 * Make the hash size about 1/2 the # found
1408 	 */
1409 	for (size /= 2; size; size >>= 1)
1410 		bits++;
1411 
1412 	/* Don't allocate too much */
1413 	if (bits > FTRACE_HASH_MAX_BITS)
1414 		bits = FTRACE_HASH_MAX_BITS;
1415 
1416 	new_hash = alloc_ftrace_hash(bits);
1417 	if (!new_hash)
1418 		return -ENOMEM;
1419 
1420 	size = 1 << src->size_bits;
1421 	for (i = 0; i < size; i++) {
1422 		hhd = &src->buckets[i];
1423 		hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1424 			remove_hash_entry(src, entry);
1425 			__add_hash_entry(new_hash, entry);
1426 		}
1427 	}
1428 
1429 update:
1430 	/* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1431 	if (enable) {
1432 		/* IPMODIFY should be updated only when filter_hash updating */
1433 		ret = ftrace_hash_ipmodify_update(ops, new_hash);
1434 		if (ret < 0) {
1435 			free_ftrace_hash(new_hash);
1436 			return ret;
1437 		}
1438 	}
1439 
1440 	/*
1441 	 * Remove the current set, update the hash and add
1442 	 * them back.
1443 	 */
1444 	ftrace_hash_rec_disable_modify(ops, enable);
1445 
1446 	rcu_assign_pointer(*dst, new_hash);
1447 
1448 	ftrace_hash_rec_enable_modify(ops, enable);
1449 
1450 	return 0;
1451 }
1452 
1453 static bool hash_contains_ip(unsigned long ip,
1454 			     struct ftrace_ops_hash *hash)
1455 {
1456 	/*
1457 	 * The function record is a match if it exists in the filter
1458 	 * hash and not in the notrace hash. Note, an emty hash is
1459 	 * considered a match for the filter hash, but an empty
1460 	 * notrace hash is considered not in the notrace hash.
1461 	 */
1462 	return (ftrace_hash_empty(hash->filter_hash) ||
1463 		ftrace_lookup_ip(hash->filter_hash, ip)) &&
1464 		(ftrace_hash_empty(hash->notrace_hash) ||
1465 		 !ftrace_lookup_ip(hash->notrace_hash, ip));
1466 }
1467 
1468 /*
1469  * Test the hashes for this ops to see if we want to call
1470  * the ops->func or not.
1471  *
1472  * It's a match if the ip is in the ops->filter_hash or
1473  * the filter_hash does not exist or is empty,
1474  *  AND
1475  * the ip is not in the ops->notrace_hash.
1476  *
1477  * This needs to be called with preemption disabled as
1478  * the hashes are freed with call_rcu_sched().
1479  */
1480 static int
1481 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1482 {
1483 	struct ftrace_ops_hash hash;
1484 	int ret;
1485 
1486 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1487 	/*
1488 	 * There's a small race when adding ops that the ftrace handler
1489 	 * that wants regs, may be called without them. We can not
1490 	 * allow that handler to be called if regs is NULL.
1491 	 */
1492 	if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1493 		return 0;
1494 #endif
1495 
1496 	hash.filter_hash = rcu_dereference_raw_notrace(ops->func_hash->filter_hash);
1497 	hash.notrace_hash = rcu_dereference_raw_notrace(ops->func_hash->notrace_hash);
1498 
1499 	if (hash_contains_ip(ip, &hash))
1500 		ret = 1;
1501 	else
1502 		ret = 0;
1503 
1504 	return ret;
1505 }
1506 
1507 /*
1508  * This is a double for. Do not use 'break' to break out of the loop,
1509  * you must use a goto.
1510  */
1511 #define do_for_each_ftrace_rec(pg, rec)					\
1512 	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1513 		int _____i;						\
1514 		for (_____i = 0; _____i < pg->index; _____i++) {	\
1515 			rec = &pg->records[_____i];
1516 
1517 #define while_for_each_ftrace_rec()		\
1518 		}				\
1519 	}
1520 
1521 
1522 static int ftrace_cmp_recs(const void *a, const void *b)
1523 {
1524 	const struct dyn_ftrace *key = a;
1525 	const struct dyn_ftrace *rec = b;
1526 
1527 	if (key->flags < rec->ip)
1528 		return -1;
1529 	if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1530 		return 1;
1531 	return 0;
1532 }
1533 
1534 /**
1535  * ftrace_location_range - return the first address of a traced location
1536  *	if it touches the given ip range
1537  * @start: start of range to search.
1538  * @end: end of range to search (inclusive). @end points to the last byte
1539  *	to check.
1540  *
1541  * Returns rec->ip if the related ftrace location is a least partly within
1542  * the given address range. That is, the first address of the instruction
1543  * that is either a NOP or call to the function tracer. It checks the ftrace
1544  * internal tables to determine if the address belongs or not.
1545  */
1546 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1547 {
1548 	struct ftrace_page *pg;
1549 	struct dyn_ftrace *rec;
1550 	struct dyn_ftrace key;
1551 
1552 	key.ip = start;
1553 	key.flags = end;	/* overload flags, as it is unsigned long */
1554 
1555 	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1556 		if (end < pg->records[0].ip ||
1557 		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1558 			continue;
1559 		rec = bsearch(&key, pg->records, pg->index,
1560 			      sizeof(struct dyn_ftrace),
1561 			      ftrace_cmp_recs);
1562 		if (rec)
1563 			return rec->ip;
1564 	}
1565 
1566 	return 0;
1567 }
1568 
1569 /**
1570  * ftrace_location - return true if the ip giving is a traced location
1571  * @ip: the instruction pointer to check
1572  *
1573  * Returns rec->ip if @ip given is a pointer to a ftrace location.
1574  * That is, the instruction that is either a NOP or call to
1575  * the function tracer. It checks the ftrace internal tables to
1576  * determine if the address belongs or not.
1577  */
1578 unsigned long ftrace_location(unsigned long ip)
1579 {
1580 	return ftrace_location_range(ip, ip);
1581 }
1582 
1583 /**
1584  * ftrace_text_reserved - return true if range contains an ftrace location
1585  * @start: start of range to search
1586  * @end: end of range to search (inclusive). @end points to the last byte to check.
1587  *
1588  * Returns 1 if @start and @end contains a ftrace location.
1589  * That is, the instruction that is either a NOP or call to
1590  * the function tracer. It checks the ftrace internal tables to
1591  * determine if the address belongs or not.
1592  */
1593 int ftrace_text_reserved(const void *start, const void *end)
1594 {
1595 	unsigned long ret;
1596 
1597 	ret = ftrace_location_range((unsigned long)start,
1598 				    (unsigned long)end);
1599 
1600 	return (int)!!ret;
1601 }
1602 
1603 /* Test if ops registered to this rec needs regs */
1604 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1605 {
1606 	struct ftrace_ops *ops;
1607 	bool keep_regs = false;
1608 
1609 	for (ops = ftrace_ops_list;
1610 	     ops != &ftrace_list_end; ops = ops->next) {
1611 		/* pass rec in as regs to have non-NULL val */
1612 		if (ftrace_ops_test(ops, rec->ip, rec)) {
1613 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1614 				keep_regs = true;
1615 				break;
1616 			}
1617 		}
1618 	}
1619 
1620 	return  keep_regs;
1621 }
1622 
1623 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1624 				     int filter_hash,
1625 				     bool inc)
1626 {
1627 	struct ftrace_hash *hash;
1628 	struct ftrace_hash *other_hash;
1629 	struct ftrace_page *pg;
1630 	struct dyn_ftrace *rec;
1631 	bool update = false;
1632 	int count = 0;
1633 	int all = 0;
1634 
1635 	/* Only update if the ops has been registered */
1636 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1637 		return false;
1638 
1639 	/*
1640 	 * In the filter_hash case:
1641 	 *   If the count is zero, we update all records.
1642 	 *   Otherwise we just update the items in the hash.
1643 	 *
1644 	 * In the notrace_hash case:
1645 	 *   We enable the update in the hash.
1646 	 *   As disabling notrace means enabling the tracing,
1647 	 *   and enabling notrace means disabling, the inc variable
1648 	 *   gets inversed.
1649 	 */
1650 	if (filter_hash) {
1651 		hash = ops->func_hash->filter_hash;
1652 		other_hash = ops->func_hash->notrace_hash;
1653 		if (ftrace_hash_empty(hash))
1654 			all = 1;
1655 	} else {
1656 		inc = !inc;
1657 		hash = ops->func_hash->notrace_hash;
1658 		other_hash = ops->func_hash->filter_hash;
1659 		/*
1660 		 * If the notrace hash has no items,
1661 		 * then there's nothing to do.
1662 		 */
1663 		if (ftrace_hash_empty(hash))
1664 			return false;
1665 	}
1666 
1667 	do_for_each_ftrace_rec(pg, rec) {
1668 		int in_other_hash = 0;
1669 		int in_hash = 0;
1670 		int match = 0;
1671 
1672 		if (rec->flags & FTRACE_FL_DISABLED)
1673 			continue;
1674 
1675 		if (all) {
1676 			/*
1677 			 * Only the filter_hash affects all records.
1678 			 * Update if the record is not in the notrace hash.
1679 			 */
1680 			if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1681 				match = 1;
1682 		} else {
1683 			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1684 			in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1685 
1686 			/*
1687 			 * If filter_hash is set, we want to match all functions
1688 			 * that are in the hash but not in the other hash.
1689 			 *
1690 			 * If filter_hash is not set, then we are decrementing.
1691 			 * That means we match anything that is in the hash
1692 			 * and also in the other_hash. That is, we need to turn
1693 			 * off functions in the other hash because they are disabled
1694 			 * by this hash.
1695 			 */
1696 			if (filter_hash && in_hash && !in_other_hash)
1697 				match = 1;
1698 			else if (!filter_hash && in_hash &&
1699 				 (in_other_hash || ftrace_hash_empty(other_hash)))
1700 				match = 1;
1701 		}
1702 		if (!match)
1703 			continue;
1704 
1705 		if (inc) {
1706 			rec->flags++;
1707 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1708 				return false;
1709 
1710 			/*
1711 			 * If there's only a single callback registered to a
1712 			 * function, and the ops has a trampoline registered
1713 			 * for it, then we can call it directly.
1714 			 */
1715 			if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1716 				rec->flags |= FTRACE_FL_TRAMP;
1717 			else
1718 				/*
1719 				 * If we are adding another function callback
1720 				 * to this function, and the previous had a
1721 				 * custom trampoline in use, then we need to go
1722 				 * back to the default trampoline.
1723 				 */
1724 				rec->flags &= ~FTRACE_FL_TRAMP;
1725 
1726 			/*
1727 			 * If any ops wants regs saved for this function
1728 			 * then all ops will get saved regs.
1729 			 */
1730 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1731 				rec->flags |= FTRACE_FL_REGS;
1732 		} else {
1733 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1734 				return false;
1735 			rec->flags--;
1736 
1737 			/*
1738 			 * If the rec had REGS enabled and the ops that is
1739 			 * being removed had REGS set, then see if there is
1740 			 * still any ops for this record that wants regs.
1741 			 * If not, we can stop recording them.
1742 			 */
1743 			if (ftrace_rec_count(rec) > 0 &&
1744 			    rec->flags & FTRACE_FL_REGS &&
1745 			    ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1746 				if (!test_rec_ops_needs_regs(rec))
1747 					rec->flags &= ~FTRACE_FL_REGS;
1748 			}
1749 
1750 			/*
1751 			 * If the rec had TRAMP enabled, then it needs to
1752 			 * be cleared. As TRAMP can only be enabled iff
1753 			 * there is only a single ops attached to it.
1754 			 * In otherwords, always disable it on decrementing.
1755 			 * In the future, we may set it if rec count is
1756 			 * decremented to one, and the ops that is left
1757 			 * has a trampoline.
1758 			 */
1759 			rec->flags &= ~FTRACE_FL_TRAMP;
1760 
1761 			/*
1762 			 * flags will be cleared in ftrace_check_record()
1763 			 * if rec count is zero.
1764 			 */
1765 		}
1766 		count++;
1767 
1768 		/* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1769 		update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1770 
1771 		/* Shortcut, if we handled all records, we are done. */
1772 		if (!all && count == hash->count)
1773 			return update;
1774 	} while_for_each_ftrace_rec();
1775 
1776 	return update;
1777 }
1778 
1779 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1780 				    int filter_hash)
1781 {
1782 	return __ftrace_hash_rec_update(ops, filter_hash, 0);
1783 }
1784 
1785 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1786 				   int filter_hash)
1787 {
1788 	return __ftrace_hash_rec_update(ops, filter_hash, 1);
1789 }
1790 
1791 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1792 					  int filter_hash, int inc)
1793 {
1794 	struct ftrace_ops *op;
1795 
1796 	__ftrace_hash_rec_update(ops, filter_hash, inc);
1797 
1798 	if (ops->func_hash != &global_ops.local_hash)
1799 		return;
1800 
1801 	/*
1802 	 * If the ops shares the global_ops hash, then we need to update
1803 	 * all ops that are enabled and use this hash.
1804 	 */
1805 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1806 		/* Already done */
1807 		if (op == ops)
1808 			continue;
1809 		if (op->func_hash == &global_ops.local_hash)
1810 			__ftrace_hash_rec_update(op, filter_hash, inc);
1811 	} while_for_each_ftrace_op(op);
1812 }
1813 
1814 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1815 					   int filter_hash)
1816 {
1817 	ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1818 }
1819 
1820 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1821 					  int filter_hash)
1822 {
1823 	ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1824 }
1825 
1826 /*
1827  * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1828  * or no-needed to update, -EBUSY if it detects a conflict of the flag
1829  * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1830  * Note that old_hash and new_hash has below meanings
1831  *  - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1832  *  - If the hash is EMPTY_HASH, it hits nothing
1833  *  - Anything else hits the recs which match the hash entries.
1834  */
1835 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1836 					 struct ftrace_hash *old_hash,
1837 					 struct ftrace_hash *new_hash)
1838 {
1839 	struct ftrace_page *pg;
1840 	struct dyn_ftrace *rec, *end = NULL;
1841 	int in_old, in_new;
1842 
1843 	/* Only update if the ops has been registered */
1844 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1845 		return 0;
1846 
1847 	if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1848 		return 0;
1849 
1850 	/*
1851 	 * Since the IPMODIFY is a very address sensitive action, we do not
1852 	 * allow ftrace_ops to set all functions to new hash.
1853 	 */
1854 	if (!new_hash || !old_hash)
1855 		return -EINVAL;
1856 
1857 	/* Update rec->flags */
1858 	do_for_each_ftrace_rec(pg, rec) {
1859 		/* We need to update only differences of filter_hash */
1860 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1861 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1862 		if (in_old == in_new)
1863 			continue;
1864 
1865 		if (in_new) {
1866 			/* New entries must ensure no others are using it */
1867 			if (rec->flags & FTRACE_FL_IPMODIFY)
1868 				goto rollback;
1869 			rec->flags |= FTRACE_FL_IPMODIFY;
1870 		} else /* Removed entry */
1871 			rec->flags &= ~FTRACE_FL_IPMODIFY;
1872 	} while_for_each_ftrace_rec();
1873 
1874 	return 0;
1875 
1876 rollback:
1877 	end = rec;
1878 
1879 	/* Roll back what we did above */
1880 	do_for_each_ftrace_rec(pg, rec) {
1881 		if (rec == end)
1882 			goto err_out;
1883 
1884 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1885 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1886 		if (in_old == in_new)
1887 			continue;
1888 
1889 		if (in_new)
1890 			rec->flags &= ~FTRACE_FL_IPMODIFY;
1891 		else
1892 			rec->flags |= FTRACE_FL_IPMODIFY;
1893 	} while_for_each_ftrace_rec();
1894 
1895 err_out:
1896 	return -EBUSY;
1897 }
1898 
1899 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1900 {
1901 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1902 
1903 	if (ftrace_hash_empty(hash))
1904 		hash = NULL;
1905 
1906 	return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1907 }
1908 
1909 /* Disabling always succeeds */
1910 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1911 {
1912 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1913 
1914 	if (ftrace_hash_empty(hash))
1915 		hash = NULL;
1916 
1917 	__ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1918 }
1919 
1920 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1921 				       struct ftrace_hash *new_hash)
1922 {
1923 	struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1924 
1925 	if (ftrace_hash_empty(old_hash))
1926 		old_hash = NULL;
1927 
1928 	if (ftrace_hash_empty(new_hash))
1929 		new_hash = NULL;
1930 
1931 	return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
1932 }
1933 
1934 static void print_ip_ins(const char *fmt, const unsigned char *p)
1935 {
1936 	int i;
1937 
1938 	printk(KERN_CONT "%s", fmt);
1939 
1940 	for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1941 		printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1942 }
1943 
1944 static struct ftrace_ops *
1945 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1946 static struct ftrace_ops *
1947 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1948 
1949 enum ftrace_bug_type ftrace_bug_type;
1950 const void *ftrace_expected;
1951 
1952 static void print_bug_type(void)
1953 {
1954 	switch (ftrace_bug_type) {
1955 	case FTRACE_BUG_UNKNOWN:
1956 		break;
1957 	case FTRACE_BUG_INIT:
1958 		pr_info("Initializing ftrace call sites\n");
1959 		break;
1960 	case FTRACE_BUG_NOP:
1961 		pr_info("Setting ftrace call site to NOP\n");
1962 		break;
1963 	case FTRACE_BUG_CALL:
1964 		pr_info("Setting ftrace call site to call ftrace function\n");
1965 		break;
1966 	case FTRACE_BUG_UPDATE:
1967 		pr_info("Updating ftrace call site to call a different ftrace function\n");
1968 		break;
1969 	}
1970 }
1971 
1972 /**
1973  * ftrace_bug - report and shutdown function tracer
1974  * @failed: The failed type (EFAULT, EINVAL, EPERM)
1975  * @rec: The record that failed
1976  *
1977  * The arch code that enables or disables the function tracing
1978  * can call ftrace_bug() when it has detected a problem in
1979  * modifying the code. @failed should be one of either:
1980  * EFAULT - if the problem happens on reading the @ip address
1981  * EINVAL - if what is read at @ip is not what was expected
1982  * EPERM - if the problem happens on writting to the @ip address
1983  */
1984 void ftrace_bug(int failed, struct dyn_ftrace *rec)
1985 {
1986 	unsigned long ip = rec ? rec->ip : 0;
1987 
1988 	switch (failed) {
1989 	case -EFAULT:
1990 		FTRACE_WARN_ON_ONCE(1);
1991 		pr_info("ftrace faulted on modifying ");
1992 		print_ip_sym(ip);
1993 		break;
1994 	case -EINVAL:
1995 		FTRACE_WARN_ON_ONCE(1);
1996 		pr_info("ftrace failed to modify ");
1997 		print_ip_sym(ip);
1998 		print_ip_ins(" actual:   ", (unsigned char *)ip);
1999 		pr_cont("\n");
2000 		if (ftrace_expected) {
2001 			print_ip_ins(" expected: ", ftrace_expected);
2002 			pr_cont("\n");
2003 		}
2004 		break;
2005 	case -EPERM:
2006 		FTRACE_WARN_ON_ONCE(1);
2007 		pr_info("ftrace faulted on writing ");
2008 		print_ip_sym(ip);
2009 		break;
2010 	default:
2011 		FTRACE_WARN_ON_ONCE(1);
2012 		pr_info("ftrace faulted on unknown error ");
2013 		print_ip_sym(ip);
2014 	}
2015 	print_bug_type();
2016 	if (rec) {
2017 		struct ftrace_ops *ops = NULL;
2018 
2019 		pr_info("ftrace record flags: %lx\n", rec->flags);
2020 		pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2021 			rec->flags & FTRACE_FL_REGS ? " R" : "  ");
2022 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
2023 			ops = ftrace_find_tramp_ops_any(rec);
2024 			if (ops) {
2025 				do {
2026 					pr_cont("\ttramp: %pS (%pS)",
2027 						(void *)ops->trampoline,
2028 						(void *)ops->func);
2029 					ops = ftrace_find_tramp_ops_next(rec, ops);
2030 				} while (ops);
2031 			} else
2032 				pr_cont("\ttramp: ERROR!");
2033 
2034 		}
2035 		ip = ftrace_get_addr_curr(rec);
2036 		pr_cont("\n expected tramp: %lx\n", ip);
2037 	}
2038 }
2039 
2040 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2041 {
2042 	unsigned long flag = 0UL;
2043 
2044 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2045 
2046 	if (rec->flags & FTRACE_FL_DISABLED)
2047 		return FTRACE_UPDATE_IGNORE;
2048 
2049 	/*
2050 	 * If we are updating calls:
2051 	 *
2052 	 *   If the record has a ref count, then we need to enable it
2053 	 *   because someone is using it.
2054 	 *
2055 	 *   Otherwise we make sure its disabled.
2056 	 *
2057 	 * If we are disabling calls, then disable all records that
2058 	 * are enabled.
2059 	 */
2060 	if (enable && ftrace_rec_count(rec))
2061 		flag = FTRACE_FL_ENABLED;
2062 
2063 	/*
2064 	 * If enabling and the REGS flag does not match the REGS_EN, or
2065 	 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2066 	 * this record. Set flags to fail the compare against ENABLED.
2067 	 */
2068 	if (flag) {
2069 		if (!(rec->flags & FTRACE_FL_REGS) !=
2070 		    !(rec->flags & FTRACE_FL_REGS_EN))
2071 			flag |= FTRACE_FL_REGS;
2072 
2073 		if (!(rec->flags & FTRACE_FL_TRAMP) !=
2074 		    !(rec->flags & FTRACE_FL_TRAMP_EN))
2075 			flag |= FTRACE_FL_TRAMP;
2076 	}
2077 
2078 	/* If the state of this record hasn't changed, then do nothing */
2079 	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2080 		return FTRACE_UPDATE_IGNORE;
2081 
2082 	if (flag) {
2083 		/* Save off if rec is being enabled (for return value) */
2084 		flag ^= rec->flags & FTRACE_FL_ENABLED;
2085 
2086 		if (update) {
2087 			rec->flags |= FTRACE_FL_ENABLED;
2088 			if (flag & FTRACE_FL_REGS) {
2089 				if (rec->flags & FTRACE_FL_REGS)
2090 					rec->flags |= FTRACE_FL_REGS_EN;
2091 				else
2092 					rec->flags &= ~FTRACE_FL_REGS_EN;
2093 			}
2094 			if (flag & FTRACE_FL_TRAMP) {
2095 				if (rec->flags & FTRACE_FL_TRAMP)
2096 					rec->flags |= FTRACE_FL_TRAMP_EN;
2097 				else
2098 					rec->flags &= ~FTRACE_FL_TRAMP_EN;
2099 			}
2100 		}
2101 
2102 		/*
2103 		 * If this record is being updated from a nop, then
2104 		 *   return UPDATE_MAKE_CALL.
2105 		 * Otherwise,
2106 		 *   return UPDATE_MODIFY_CALL to tell the caller to convert
2107 		 *   from the save regs, to a non-save regs function or
2108 		 *   vice versa, or from a trampoline call.
2109 		 */
2110 		if (flag & FTRACE_FL_ENABLED) {
2111 			ftrace_bug_type = FTRACE_BUG_CALL;
2112 			return FTRACE_UPDATE_MAKE_CALL;
2113 		}
2114 
2115 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2116 		return FTRACE_UPDATE_MODIFY_CALL;
2117 	}
2118 
2119 	if (update) {
2120 		/* If there's no more users, clear all flags */
2121 		if (!ftrace_rec_count(rec))
2122 			rec->flags = 0;
2123 		else
2124 			/*
2125 			 * Just disable the record, but keep the ops TRAMP
2126 			 * and REGS states. The _EN flags must be disabled though.
2127 			 */
2128 			rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2129 					FTRACE_FL_REGS_EN);
2130 	}
2131 
2132 	ftrace_bug_type = FTRACE_BUG_NOP;
2133 	return FTRACE_UPDATE_MAKE_NOP;
2134 }
2135 
2136 /**
2137  * ftrace_update_record, set a record that now is tracing or not
2138  * @rec: the record to update
2139  * @enable: set to 1 if the record is tracing, zero to force disable
2140  *
2141  * The records that represent all functions that can be traced need
2142  * to be updated when tracing has been enabled.
2143  */
2144 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2145 {
2146 	return ftrace_check_record(rec, enable, 1);
2147 }
2148 
2149 /**
2150  * ftrace_test_record, check if the record has been enabled or not
2151  * @rec: the record to test
2152  * @enable: set to 1 to check if enabled, 0 if it is disabled
2153  *
2154  * The arch code may need to test if a record is already set to
2155  * tracing to determine how to modify the function code that it
2156  * represents.
2157  */
2158 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2159 {
2160 	return ftrace_check_record(rec, enable, 0);
2161 }
2162 
2163 static struct ftrace_ops *
2164 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2165 {
2166 	struct ftrace_ops *op;
2167 	unsigned long ip = rec->ip;
2168 
2169 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2170 
2171 		if (!op->trampoline)
2172 			continue;
2173 
2174 		if (hash_contains_ip(ip, op->func_hash))
2175 			return op;
2176 	} while_for_each_ftrace_op(op);
2177 
2178 	return NULL;
2179 }
2180 
2181 static struct ftrace_ops *
2182 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2183 			   struct ftrace_ops *op)
2184 {
2185 	unsigned long ip = rec->ip;
2186 
2187 	while_for_each_ftrace_op(op) {
2188 
2189 		if (!op->trampoline)
2190 			continue;
2191 
2192 		if (hash_contains_ip(ip, op->func_hash))
2193 			return op;
2194 	}
2195 
2196 	return NULL;
2197 }
2198 
2199 static struct ftrace_ops *
2200 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2201 {
2202 	struct ftrace_ops *op;
2203 	unsigned long ip = rec->ip;
2204 
2205 	/*
2206 	 * Need to check removed ops first.
2207 	 * If they are being removed, and this rec has a tramp,
2208 	 * and this rec is in the ops list, then it would be the
2209 	 * one with the tramp.
2210 	 */
2211 	if (removed_ops) {
2212 		if (hash_contains_ip(ip, &removed_ops->old_hash))
2213 			return removed_ops;
2214 	}
2215 
2216 	/*
2217 	 * Need to find the current trampoline for a rec.
2218 	 * Now, a trampoline is only attached to a rec if there
2219 	 * was a single 'ops' attached to it. But this can be called
2220 	 * when we are adding another op to the rec or removing the
2221 	 * current one. Thus, if the op is being added, we can
2222 	 * ignore it because it hasn't attached itself to the rec
2223 	 * yet.
2224 	 *
2225 	 * If an ops is being modified (hooking to different functions)
2226 	 * then we don't care about the new functions that are being
2227 	 * added, just the old ones (that are probably being removed).
2228 	 *
2229 	 * If we are adding an ops to a function that already is using
2230 	 * a trampoline, it needs to be removed (trampolines are only
2231 	 * for single ops connected), then an ops that is not being
2232 	 * modified also needs to be checked.
2233 	 */
2234 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2235 
2236 		if (!op->trampoline)
2237 			continue;
2238 
2239 		/*
2240 		 * If the ops is being added, it hasn't gotten to
2241 		 * the point to be removed from this tree yet.
2242 		 */
2243 		if (op->flags & FTRACE_OPS_FL_ADDING)
2244 			continue;
2245 
2246 
2247 		/*
2248 		 * If the ops is being modified and is in the old
2249 		 * hash, then it is probably being removed from this
2250 		 * function.
2251 		 */
2252 		if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2253 		    hash_contains_ip(ip, &op->old_hash))
2254 			return op;
2255 		/*
2256 		 * If the ops is not being added or modified, and it's
2257 		 * in its normal filter hash, then this must be the one
2258 		 * we want!
2259 		 */
2260 		if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2261 		    hash_contains_ip(ip, op->func_hash))
2262 			return op;
2263 
2264 	} while_for_each_ftrace_op(op);
2265 
2266 	return NULL;
2267 }
2268 
2269 static struct ftrace_ops *
2270 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2271 {
2272 	struct ftrace_ops *op;
2273 	unsigned long ip = rec->ip;
2274 
2275 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2276 		/* pass rec in as regs to have non-NULL val */
2277 		if (hash_contains_ip(ip, op->func_hash))
2278 			return op;
2279 	} while_for_each_ftrace_op(op);
2280 
2281 	return NULL;
2282 }
2283 
2284 /**
2285  * ftrace_get_addr_new - Get the call address to set to
2286  * @rec:  The ftrace record descriptor
2287  *
2288  * If the record has the FTRACE_FL_REGS set, that means that it
2289  * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2290  * is not not set, then it wants to convert to the normal callback.
2291  *
2292  * Returns the address of the trampoline to set to
2293  */
2294 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2295 {
2296 	struct ftrace_ops *ops;
2297 
2298 	/* Trampolines take precedence over regs */
2299 	if (rec->flags & FTRACE_FL_TRAMP) {
2300 		ops = ftrace_find_tramp_ops_new(rec);
2301 		if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2302 			pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2303 				(void *)rec->ip, (void *)rec->ip, rec->flags);
2304 			/* Ftrace is shutting down, return anything */
2305 			return (unsigned long)FTRACE_ADDR;
2306 		}
2307 		return ops->trampoline;
2308 	}
2309 
2310 	if (rec->flags & FTRACE_FL_REGS)
2311 		return (unsigned long)FTRACE_REGS_ADDR;
2312 	else
2313 		return (unsigned long)FTRACE_ADDR;
2314 }
2315 
2316 /**
2317  * ftrace_get_addr_curr - Get the call address that is already there
2318  * @rec:  The ftrace record descriptor
2319  *
2320  * The FTRACE_FL_REGS_EN is set when the record already points to
2321  * a function that saves all the regs. Basically the '_EN' version
2322  * represents the current state of the function.
2323  *
2324  * Returns the address of the trampoline that is currently being called
2325  */
2326 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2327 {
2328 	struct ftrace_ops *ops;
2329 
2330 	/* Trampolines take precedence over regs */
2331 	if (rec->flags & FTRACE_FL_TRAMP_EN) {
2332 		ops = ftrace_find_tramp_ops_curr(rec);
2333 		if (FTRACE_WARN_ON(!ops)) {
2334 			pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2335 				(void *)rec->ip, (void *)rec->ip);
2336 			/* Ftrace is shutting down, return anything */
2337 			return (unsigned long)FTRACE_ADDR;
2338 		}
2339 		return ops->trampoline;
2340 	}
2341 
2342 	if (rec->flags & FTRACE_FL_REGS_EN)
2343 		return (unsigned long)FTRACE_REGS_ADDR;
2344 	else
2345 		return (unsigned long)FTRACE_ADDR;
2346 }
2347 
2348 static int
2349 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2350 {
2351 	unsigned long ftrace_old_addr;
2352 	unsigned long ftrace_addr;
2353 	int ret;
2354 
2355 	ftrace_addr = ftrace_get_addr_new(rec);
2356 
2357 	/* This needs to be done before we call ftrace_update_record */
2358 	ftrace_old_addr = ftrace_get_addr_curr(rec);
2359 
2360 	ret = ftrace_update_record(rec, enable);
2361 
2362 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2363 
2364 	switch (ret) {
2365 	case FTRACE_UPDATE_IGNORE:
2366 		return 0;
2367 
2368 	case FTRACE_UPDATE_MAKE_CALL:
2369 		ftrace_bug_type = FTRACE_BUG_CALL;
2370 		return ftrace_make_call(rec, ftrace_addr);
2371 
2372 	case FTRACE_UPDATE_MAKE_NOP:
2373 		ftrace_bug_type = FTRACE_BUG_NOP;
2374 		return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2375 
2376 	case FTRACE_UPDATE_MODIFY_CALL:
2377 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2378 		return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2379 	}
2380 
2381 	return -1; /* unknow ftrace bug */
2382 }
2383 
2384 void __weak ftrace_replace_code(int enable)
2385 {
2386 	struct dyn_ftrace *rec;
2387 	struct ftrace_page *pg;
2388 	int failed;
2389 
2390 	if (unlikely(ftrace_disabled))
2391 		return;
2392 
2393 	do_for_each_ftrace_rec(pg, rec) {
2394 		failed = __ftrace_replace_code(rec, enable);
2395 		if (failed) {
2396 			ftrace_bug(failed, rec);
2397 			/* Stop processing */
2398 			return;
2399 		}
2400 	} while_for_each_ftrace_rec();
2401 }
2402 
2403 struct ftrace_rec_iter {
2404 	struct ftrace_page	*pg;
2405 	int			index;
2406 };
2407 
2408 /**
2409  * ftrace_rec_iter_start, start up iterating over traced functions
2410  *
2411  * Returns an iterator handle that is used to iterate over all
2412  * the records that represent address locations where functions
2413  * are traced.
2414  *
2415  * May return NULL if no records are available.
2416  */
2417 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2418 {
2419 	/*
2420 	 * We only use a single iterator.
2421 	 * Protected by the ftrace_lock mutex.
2422 	 */
2423 	static struct ftrace_rec_iter ftrace_rec_iter;
2424 	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2425 
2426 	iter->pg = ftrace_pages_start;
2427 	iter->index = 0;
2428 
2429 	/* Could have empty pages */
2430 	while (iter->pg && !iter->pg->index)
2431 		iter->pg = iter->pg->next;
2432 
2433 	if (!iter->pg)
2434 		return NULL;
2435 
2436 	return iter;
2437 }
2438 
2439 /**
2440  * ftrace_rec_iter_next, get the next record to process.
2441  * @iter: The handle to the iterator.
2442  *
2443  * Returns the next iterator after the given iterator @iter.
2444  */
2445 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2446 {
2447 	iter->index++;
2448 
2449 	if (iter->index >= iter->pg->index) {
2450 		iter->pg = iter->pg->next;
2451 		iter->index = 0;
2452 
2453 		/* Could have empty pages */
2454 		while (iter->pg && !iter->pg->index)
2455 			iter->pg = iter->pg->next;
2456 	}
2457 
2458 	if (!iter->pg)
2459 		return NULL;
2460 
2461 	return iter;
2462 }
2463 
2464 /**
2465  * ftrace_rec_iter_record, get the record at the iterator location
2466  * @iter: The current iterator location
2467  *
2468  * Returns the record that the current @iter is at.
2469  */
2470 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2471 {
2472 	return &iter->pg->records[iter->index];
2473 }
2474 
2475 static int
2476 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2477 {
2478 	int ret;
2479 
2480 	if (unlikely(ftrace_disabled))
2481 		return 0;
2482 
2483 	ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2484 	if (ret) {
2485 		ftrace_bug_type = FTRACE_BUG_INIT;
2486 		ftrace_bug(ret, rec);
2487 		return 0;
2488 	}
2489 	return 1;
2490 }
2491 
2492 /*
2493  * archs can override this function if they must do something
2494  * before the modifying code is performed.
2495  */
2496 int __weak ftrace_arch_code_modify_prepare(void)
2497 {
2498 	return 0;
2499 }
2500 
2501 /*
2502  * archs can override this function if they must do something
2503  * after the modifying code is performed.
2504  */
2505 int __weak ftrace_arch_code_modify_post_process(void)
2506 {
2507 	return 0;
2508 }
2509 
2510 void ftrace_modify_all_code(int command)
2511 {
2512 	int update = command & FTRACE_UPDATE_TRACE_FUNC;
2513 	int err = 0;
2514 
2515 	/*
2516 	 * If the ftrace_caller calls a ftrace_ops func directly,
2517 	 * we need to make sure that it only traces functions it
2518 	 * expects to trace. When doing the switch of functions,
2519 	 * we need to update to the ftrace_ops_list_func first
2520 	 * before the transition between old and new calls are set,
2521 	 * as the ftrace_ops_list_func will check the ops hashes
2522 	 * to make sure the ops are having the right functions
2523 	 * traced.
2524 	 */
2525 	if (update) {
2526 		err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2527 		if (FTRACE_WARN_ON(err))
2528 			return;
2529 	}
2530 
2531 	if (command & FTRACE_UPDATE_CALLS)
2532 		ftrace_replace_code(1);
2533 	else if (command & FTRACE_DISABLE_CALLS)
2534 		ftrace_replace_code(0);
2535 
2536 	if (update && ftrace_trace_function != ftrace_ops_list_func) {
2537 		function_trace_op = set_function_trace_op;
2538 		smp_wmb();
2539 		/* If irqs are disabled, we are in stop machine */
2540 		if (!irqs_disabled())
2541 			smp_call_function(ftrace_sync_ipi, NULL, 1);
2542 		err = ftrace_update_ftrace_func(ftrace_trace_function);
2543 		if (FTRACE_WARN_ON(err))
2544 			return;
2545 	}
2546 
2547 	if (command & FTRACE_START_FUNC_RET)
2548 		err = ftrace_enable_ftrace_graph_caller();
2549 	else if (command & FTRACE_STOP_FUNC_RET)
2550 		err = ftrace_disable_ftrace_graph_caller();
2551 	FTRACE_WARN_ON(err);
2552 }
2553 
2554 static int __ftrace_modify_code(void *data)
2555 {
2556 	int *command = data;
2557 
2558 	ftrace_modify_all_code(*command);
2559 
2560 	return 0;
2561 }
2562 
2563 /**
2564  * ftrace_run_stop_machine, go back to the stop machine method
2565  * @command: The command to tell ftrace what to do
2566  *
2567  * If an arch needs to fall back to the stop machine method, the
2568  * it can call this function.
2569  */
2570 void ftrace_run_stop_machine(int command)
2571 {
2572 	stop_machine(__ftrace_modify_code, &command, NULL);
2573 }
2574 
2575 /**
2576  * arch_ftrace_update_code, modify the code to trace or not trace
2577  * @command: The command that needs to be done
2578  *
2579  * Archs can override this function if it does not need to
2580  * run stop_machine() to modify code.
2581  */
2582 void __weak arch_ftrace_update_code(int command)
2583 {
2584 	ftrace_run_stop_machine(command);
2585 }
2586 
2587 static void ftrace_run_update_code(int command)
2588 {
2589 	int ret;
2590 
2591 	ret = ftrace_arch_code_modify_prepare();
2592 	FTRACE_WARN_ON(ret);
2593 	if (ret)
2594 		return;
2595 
2596 	/*
2597 	 * By default we use stop_machine() to modify the code.
2598 	 * But archs can do what ever they want as long as it
2599 	 * is safe. The stop_machine() is the safest, but also
2600 	 * produces the most overhead.
2601 	 */
2602 	arch_ftrace_update_code(command);
2603 
2604 	ret = ftrace_arch_code_modify_post_process();
2605 	FTRACE_WARN_ON(ret);
2606 }
2607 
2608 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2609 				   struct ftrace_ops_hash *old_hash)
2610 {
2611 	ops->flags |= FTRACE_OPS_FL_MODIFYING;
2612 	ops->old_hash.filter_hash = old_hash->filter_hash;
2613 	ops->old_hash.notrace_hash = old_hash->notrace_hash;
2614 	ftrace_run_update_code(command);
2615 	ops->old_hash.filter_hash = NULL;
2616 	ops->old_hash.notrace_hash = NULL;
2617 	ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2618 }
2619 
2620 static ftrace_func_t saved_ftrace_func;
2621 static int ftrace_start_up;
2622 
2623 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2624 {
2625 }
2626 
2627 static void per_cpu_ops_free(struct ftrace_ops *ops)
2628 {
2629 	free_percpu(ops->disabled);
2630 }
2631 
2632 static void ftrace_startup_enable(int command)
2633 {
2634 	if (saved_ftrace_func != ftrace_trace_function) {
2635 		saved_ftrace_func = ftrace_trace_function;
2636 		command |= FTRACE_UPDATE_TRACE_FUNC;
2637 	}
2638 
2639 	if (!command || !ftrace_enabled)
2640 		return;
2641 
2642 	ftrace_run_update_code(command);
2643 }
2644 
2645 static void ftrace_startup_all(int command)
2646 {
2647 	update_all_ops = true;
2648 	ftrace_startup_enable(command);
2649 	update_all_ops = false;
2650 }
2651 
2652 static int ftrace_startup(struct ftrace_ops *ops, int command)
2653 {
2654 	int ret;
2655 
2656 	if (unlikely(ftrace_disabled))
2657 		return -ENODEV;
2658 
2659 	ret = __register_ftrace_function(ops);
2660 	if (ret)
2661 		return ret;
2662 
2663 	ftrace_start_up++;
2664 
2665 	/*
2666 	 * Note that ftrace probes uses this to start up
2667 	 * and modify functions it will probe. But we still
2668 	 * set the ADDING flag for modification, as probes
2669 	 * do not have trampolines. If they add them in the
2670 	 * future, then the probes will need to distinguish
2671 	 * between adding and updating probes.
2672 	 */
2673 	ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2674 
2675 	ret = ftrace_hash_ipmodify_enable(ops);
2676 	if (ret < 0) {
2677 		/* Rollback registration process */
2678 		__unregister_ftrace_function(ops);
2679 		ftrace_start_up--;
2680 		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2681 		return ret;
2682 	}
2683 
2684 	if (ftrace_hash_rec_enable(ops, 1))
2685 		command |= FTRACE_UPDATE_CALLS;
2686 
2687 	ftrace_startup_enable(command);
2688 
2689 	ops->flags &= ~FTRACE_OPS_FL_ADDING;
2690 
2691 	return 0;
2692 }
2693 
2694 static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2695 {
2696 	int ret;
2697 
2698 	if (unlikely(ftrace_disabled))
2699 		return -ENODEV;
2700 
2701 	ret = __unregister_ftrace_function(ops);
2702 	if (ret)
2703 		return ret;
2704 
2705 	ftrace_start_up--;
2706 	/*
2707 	 * Just warn in case of unbalance, no need to kill ftrace, it's not
2708 	 * critical but the ftrace_call callers may be never nopped again after
2709 	 * further ftrace uses.
2710 	 */
2711 	WARN_ON_ONCE(ftrace_start_up < 0);
2712 
2713 	/* Disabling ipmodify never fails */
2714 	ftrace_hash_ipmodify_disable(ops);
2715 
2716 	if (ftrace_hash_rec_disable(ops, 1))
2717 		command |= FTRACE_UPDATE_CALLS;
2718 
2719 	ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2720 
2721 	if (saved_ftrace_func != ftrace_trace_function) {
2722 		saved_ftrace_func = ftrace_trace_function;
2723 		command |= FTRACE_UPDATE_TRACE_FUNC;
2724 	}
2725 
2726 	if (!command || !ftrace_enabled) {
2727 		/*
2728 		 * If these are per_cpu ops, they still need their
2729 		 * per_cpu field freed. Since, function tracing is
2730 		 * not currently active, we can just free them
2731 		 * without synchronizing all CPUs.
2732 		 */
2733 		if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2734 			per_cpu_ops_free(ops);
2735 		return 0;
2736 	}
2737 
2738 	/*
2739 	 * If the ops uses a trampoline, then it needs to be
2740 	 * tested first on update.
2741 	 */
2742 	ops->flags |= FTRACE_OPS_FL_REMOVING;
2743 	removed_ops = ops;
2744 
2745 	/* The trampoline logic checks the old hashes */
2746 	ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2747 	ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2748 
2749 	ftrace_run_update_code(command);
2750 
2751 	/*
2752 	 * If there's no more ops registered with ftrace, run a
2753 	 * sanity check to make sure all rec flags are cleared.
2754 	 */
2755 	if (ftrace_ops_list == &ftrace_list_end) {
2756 		struct ftrace_page *pg;
2757 		struct dyn_ftrace *rec;
2758 
2759 		do_for_each_ftrace_rec(pg, rec) {
2760 			if (FTRACE_WARN_ON_ONCE(rec->flags))
2761 				pr_warn("  %pS flags:%lx\n",
2762 					(void *)rec->ip, rec->flags);
2763 		} while_for_each_ftrace_rec();
2764 	}
2765 
2766 	ops->old_hash.filter_hash = NULL;
2767 	ops->old_hash.notrace_hash = NULL;
2768 
2769 	removed_ops = NULL;
2770 	ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2771 
2772 	/*
2773 	 * Dynamic ops may be freed, we must make sure that all
2774 	 * callers are done before leaving this function.
2775 	 * The same goes for freeing the per_cpu data of the per_cpu
2776 	 * ops.
2777 	 *
2778 	 * Again, normal synchronize_sched() is not good enough.
2779 	 * We need to do a hard force of sched synchronization.
2780 	 * This is because we use preempt_disable() to do RCU, but
2781 	 * the function tracers can be called where RCU is not watching
2782 	 * (like before user_exit()). We can not rely on the RCU
2783 	 * infrastructure to do the synchronization, thus we must do it
2784 	 * ourselves.
2785 	 */
2786 	if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) {
2787 		schedule_on_each_cpu(ftrace_sync);
2788 
2789 		arch_ftrace_trampoline_free(ops);
2790 
2791 		if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2792 			per_cpu_ops_free(ops);
2793 	}
2794 
2795 	return 0;
2796 }
2797 
2798 static void ftrace_startup_sysctl(void)
2799 {
2800 	int command;
2801 
2802 	if (unlikely(ftrace_disabled))
2803 		return;
2804 
2805 	/* Force update next time */
2806 	saved_ftrace_func = NULL;
2807 	/* ftrace_start_up is true if we want ftrace running */
2808 	if (ftrace_start_up) {
2809 		command = FTRACE_UPDATE_CALLS;
2810 		if (ftrace_graph_active)
2811 			command |= FTRACE_START_FUNC_RET;
2812 		ftrace_startup_enable(command);
2813 	}
2814 }
2815 
2816 static void ftrace_shutdown_sysctl(void)
2817 {
2818 	int command;
2819 
2820 	if (unlikely(ftrace_disabled))
2821 		return;
2822 
2823 	/* ftrace_start_up is true if ftrace is running */
2824 	if (ftrace_start_up) {
2825 		command = FTRACE_DISABLE_CALLS;
2826 		if (ftrace_graph_active)
2827 			command |= FTRACE_STOP_FUNC_RET;
2828 		ftrace_run_update_code(command);
2829 	}
2830 }
2831 
2832 static cycle_t		ftrace_update_time;
2833 unsigned long		ftrace_update_tot_cnt;
2834 
2835 static inline int ops_traces_mod(struct ftrace_ops *ops)
2836 {
2837 	/*
2838 	 * Filter_hash being empty will default to trace module.
2839 	 * But notrace hash requires a test of individual module functions.
2840 	 */
2841 	return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2842 		ftrace_hash_empty(ops->func_hash->notrace_hash);
2843 }
2844 
2845 /*
2846  * Check if the current ops references the record.
2847  *
2848  * If the ops traces all functions, then it was already accounted for.
2849  * If the ops does not trace the current record function, skip it.
2850  * If the ops ignores the function via notrace filter, skip it.
2851  */
2852 static inline bool
2853 ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2854 {
2855 	/* If ops isn't enabled, ignore it */
2856 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2857 		return 0;
2858 
2859 	/* If ops traces all then it includes this function */
2860 	if (ops_traces_mod(ops))
2861 		return 1;
2862 
2863 	/* The function must be in the filter */
2864 	if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2865 	    !ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2866 		return 0;
2867 
2868 	/* If in notrace hash, we ignore it too */
2869 	if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2870 		return 0;
2871 
2872 	return 1;
2873 }
2874 
2875 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2876 {
2877 	struct ftrace_page *pg;
2878 	struct dyn_ftrace *p;
2879 	cycle_t start, stop;
2880 	unsigned long update_cnt = 0;
2881 	unsigned long rec_flags = 0;
2882 	int i;
2883 
2884 	start = ftrace_now(raw_smp_processor_id());
2885 
2886 	/*
2887 	 * When a module is loaded, this function is called to convert
2888 	 * the calls to mcount in its text to nops, and also to create
2889 	 * an entry in the ftrace data. Now, if ftrace is activated
2890 	 * after this call, but before the module sets its text to
2891 	 * read-only, the modification of enabling ftrace can fail if
2892 	 * the read-only is done while ftrace is converting the calls.
2893 	 * To prevent this, the module's records are set as disabled
2894 	 * and will be enabled after the call to set the module's text
2895 	 * to read-only.
2896 	 */
2897 	if (mod)
2898 		rec_flags |= FTRACE_FL_DISABLED;
2899 
2900 	for (pg = new_pgs; pg; pg = pg->next) {
2901 
2902 		for (i = 0; i < pg->index; i++) {
2903 
2904 			/* If something went wrong, bail without enabling anything */
2905 			if (unlikely(ftrace_disabled))
2906 				return -1;
2907 
2908 			p = &pg->records[i];
2909 			p->flags = rec_flags;
2910 
2911 			/*
2912 			 * Do the initial record conversion from mcount jump
2913 			 * to the NOP instructions.
2914 			 */
2915 			if (!ftrace_code_disable(mod, p))
2916 				break;
2917 
2918 			update_cnt++;
2919 		}
2920 	}
2921 
2922 	stop = ftrace_now(raw_smp_processor_id());
2923 	ftrace_update_time = stop - start;
2924 	ftrace_update_tot_cnt += update_cnt;
2925 
2926 	return 0;
2927 }
2928 
2929 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2930 {
2931 	int order;
2932 	int cnt;
2933 
2934 	if (WARN_ON(!count))
2935 		return -EINVAL;
2936 
2937 	order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2938 
2939 	/*
2940 	 * We want to fill as much as possible. No more than a page
2941 	 * may be empty.
2942 	 */
2943 	while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2944 		order--;
2945 
2946  again:
2947 	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2948 
2949 	if (!pg->records) {
2950 		/* if we can't allocate this size, try something smaller */
2951 		if (!order)
2952 			return -ENOMEM;
2953 		order >>= 1;
2954 		goto again;
2955 	}
2956 
2957 	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2958 	pg->size = cnt;
2959 
2960 	if (cnt > count)
2961 		cnt = count;
2962 
2963 	return cnt;
2964 }
2965 
2966 static struct ftrace_page *
2967 ftrace_allocate_pages(unsigned long num_to_init)
2968 {
2969 	struct ftrace_page *start_pg;
2970 	struct ftrace_page *pg;
2971 	int order;
2972 	int cnt;
2973 
2974 	if (!num_to_init)
2975 		return 0;
2976 
2977 	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2978 	if (!pg)
2979 		return NULL;
2980 
2981 	/*
2982 	 * Try to allocate as much as possible in one continues
2983 	 * location that fills in all of the space. We want to
2984 	 * waste as little space as possible.
2985 	 */
2986 	for (;;) {
2987 		cnt = ftrace_allocate_records(pg, num_to_init);
2988 		if (cnt < 0)
2989 			goto free_pages;
2990 
2991 		num_to_init -= cnt;
2992 		if (!num_to_init)
2993 			break;
2994 
2995 		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2996 		if (!pg->next)
2997 			goto free_pages;
2998 
2999 		pg = pg->next;
3000 	}
3001 
3002 	return start_pg;
3003 
3004  free_pages:
3005 	pg = start_pg;
3006 	while (pg) {
3007 		order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3008 		free_pages((unsigned long)pg->records, order);
3009 		start_pg = pg->next;
3010 		kfree(pg);
3011 		pg = start_pg;
3012 	}
3013 	pr_info("ftrace: FAILED to allocate memory for functions\n");
3014 	return NULL;
3015 }
3016 
3017 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3018 
3019 struct ftrace_iterator {
3020 	loff_t				pos;
3021 	loff_t				func_pos;
3022 	struct ftrace_page		*pg;
3023 	struct dyn_ftrace		*func;
3024 	struct ftrace_func_probe	*probe;
3025 	struct trace_parser		parser;
3026 	struct ftrace_hash		*hash;
3027 	struct ftrace_ops		*ops;
3028 	int				hidx;
3029 	int				idx;
3030 	unsigned			flags;
3031 };
3032 
3033 static void *
3034 t_hash_next(struct seq_file *m, loff_t *pos)
3035 {
3036 	struct ftrace_iterator *iter = m->private;
3037 	struct hlist_node *hnd = NULL;
3038 	struct hlist_head *hhd;
3039 
3040 	(*pos)++;
3041 	iter->pos = *pos;
3042 
3043 	if (iter->probe)
3044 		hnd = &iter->probe->node;
3045  retry:
3046 	if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
3047 		return NULL;
3048 
3049 	hhd = &ftrace_func_hash[iter->hidx];
3050 
3051 	if (hlist_empty(hhd)) {
3052 		iter->hidx++;
3053 		hnd = NULL;
3054 		goto retry;
3055 	}
3056 
3057 	if (!hnd)
3058 		hnd = hhd->first;
3059 	else {
3060 		hnd = hnd->next;
3061 		if (!hnd) {
3062 			iter->hidx++;
3063 			goto retry;
3064 		}
3065 	}
3066 
3067 	if (WARN_ON_ONCE(!hnd))
3068 		return NULL;
3069 
3070 	iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
3071 
3072 	return iter;
3073 }
3074 
3075 static void *t_hash_start(struct seq_file *m, loff_t *pos)
3076 {
3077 	struct ftrace_iterator *iter = m->private;
3078 	void *p = NULL;
3079 	loff_t l;
3080 
3081 	if (!(iter->flags & FTRACE_ITER_DO_HASH))
3082 		return NULL;
3083 
3084 	if (iter->func_pos > *pos)
3085 		return NULL;
3086 
3087 	iter->hidx = 0;
3088 	for (l = 0; l <= (*pos - iter->func_pos); ) {
3089 		p = t_hash_next(m, &l);
3090 		if (!p)
3091 			break;
3092 	}
3093 	if (!p)
3094 		return NULL;
3095 
3096 	/* Only set this if we have an item */
3097 	iter->flags |= FTRACE_ITER_HASH;
3098 
3099 	return iter;
3100 }
3101 
3102 static int
3103 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
3104 {
3105 	struct ftrace_func_probe *rec;
3106 
3107 	rec = iter->probe;
3108 	if (WARN_ON_ONCE(!rec))
3109 		return -EIO;
3110 
3111 	if (rec->ops->print)
3112 		return rec->ops->print(m, rec->ip, rec->ops, rec->data);
3113 
3114 	seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
3115 
3116 	if (rec->data)
3117 		seq_printf(m, ":%p", rec->data);
3118 	seq_putc(m, '\n');
3119 
3120 	return 0;
3121 }
3122 
3123 static void *
3124 t_next(struct seq_file *m, void *v, loff_t *pos)
3125 {
3126 	struct ftrace_iterator *iter = m->private;
3127 	struct ftrace_ops *ops = iter->ops;
3128 	struct dyn_ftrace *rec = NULL;
3129 
3130 	if (unlikely(ftrace_disabled))
3131 		return NULL;
3132 
3133 	if (iter->flags & FTRACE_ITER_HASH)
3134 		return t_hash_next(m, pos);
3135 
3136 	(*pos)++;
3137 	iter->pos = iter->func_pos = *pos;
3138 
3139 	if (iter->flags & FTRACE_ITER_PRINTALL)
3140 		return t_hash_start(m, pos);
3141 
3142  retry:
3143 	if (iter->idx >= iter->pg->index) {
3144 		if (iter->pg->next) {
3145 			iter->pg = iter->pg->next;
3146 			iter->idx = 0;
3147 			goto retry;
3148 		}
3149 	} else {
3150 		rec = &iter->pg->records[iter->idx++];
3151 		if (((iter->flags & FTRACE_ITER_FILTER) &&
3152 		     !(ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))) ||
3153 
3154 		    ((iter->flags & FTRACE_ITER_NOTRACE) &&
3155 		     !ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip)) ||
3156 
3157 		    ((iter->flags & FTRACE_ITER_ENABLED) &&
3158 		     !(rec->flags & FTRACE_FL_ENABLED))) {
3159 
3160 			rec = NULL;
3161 			goto retry;
3162 		}
3163 	}
3164 
3165 	if (!rec)
3166 		return t_hash_start(m, pos);
3167 
3168 	iter->func = rec;
3169 
3170 	return iter;
3171 }
3172 
3173 static void reset_iter_read(struct ftrace_iterator *iter)
3174 {
3175 	iter->pos = 0;
3176 	iter->func_pos = 0;
3177 	iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
3178 }
3179 
3180 static void *t_start(struct seq_file *m, loff_t *pos)
3181 {
3182 	struct ftrace_iterator *iter = m->private;
3183 	struct ftrace_ops *ops = iter->ops;
3184 	void *p = NULL;
3185 	loff_t l;
3186 
3187 	mutex_lock(&ftrace_lock);
3188 
3189 	if (unlikely(ftrace_disabled))
3190 		return NULL;
3191 
3192 	/*
3193 	 * If an lseek was done, then reset and start from beginning.
3194 	 */
3195 	if (*pos < iter->pos)
3196 		reset_iter_read(iter);
3197 
3198 	/*
3199 	 * For set_ftrace_filter reading, if we have the filter
3200 	 * off, we can short cut and just print out that all
3201 	 * functions are enabled.
3202 	 */
3203 	if ((iter->flags & FTRACE_ITER_FILTER &&
3204 	     ftrace_hash_empty(ops->func_hash->filter_hash)) ||
3205 	    (iter->flags & FTRACE_ITER_NOTRACE &&
3206 	     ftrace_hash_empty(ops->func_hash->notrace_hash))) {
3207 		if (*pos > 0)
3208 			return t_hash_start(m, pos);
3209 		iter->flags |= FTRACE_ITER_PRINTALL;
3210 		/* reset in case of seek/pread */
3211 		iter->flags &= ~FTRACE_ITER_HASH;
3212 		return iter;
3213 	}
3214 
3215 	if (iter->flags & FTRACE_ITER_HASH)
3216 		return t_hash_start(m, pos);
3217 
3218 	/*
3219 	 * Unfortunately, we need to restart at ftrace_pages_start
3220 	 * every time we let go of the ftrace_mutex. This is because
3221 	 * those pointers can change without the lock.
3222 	 */
3223 	iter->pg = ftrace_pages_start;
3224 	iter->idx = 0;
3225 	for (l = 0; l <= *pos; ) {
3226 		p = t_next(m, p, &l);
3227 		if (!p)
3228 			break;
3229 	}
3230 
3231 	if (!p)
3232 		return t_hash_start(m, pos);
3233 
3234 	return iter;
3235 }
3236 
3237 static void t_stop(struct seq_file *m, void *p)
3238 {
3239 	mutex_unlock(&ftrace_lock);
3240 }
3241 
3242 void * __weak
3243 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3244 {
3245 	return NULL;
3246 }
3247 
3248 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3249 				struct dyn_ftrace *rec)
3250 {
3251 	void *ptr;
3252 
3253 	ptr = arch_ftrace_trampoline_func(ops, rec);
3254 	if (ptr)
3255 		seq_printf(m, " ->%pS", ptr);
3256 }
3257 
3258 static int t_show(struct seq_file *m, void *v)
3259 {
3260 	struct ftrace_iterator *iter = m->private;
3261 	struct dyn_ftrace *rec;
3262 
3263 	if (iter->flags & FTRACE_ITER_HASH)
3264 		return t_hash_show(m, iter);
3265 
3266 	if (iter->flags & FTRACE_ITER_PRINTALL) {
3267 		if (iter->flags & FTRACE_ITER_NOTRACE)
3268 			seq_puts(m, "#### no functions disabled ####\n");
3269 		else
3270 			seq_puts(m, "#### all functions enabled ####\n");
3271 		return 0;
3272 	}
3273 
3274 	rec = iter->func;
3275 
3276 	if (!rec)
3277 		return 0;
3278 
3279 	seq_printf(m, "%ps", (void *)rec->ip);
3280 	if (iter->flags & FTRACE_ITER_ENABLED) {
3281 		struct ftrace_ops *ops;
3282 
3283 		seq_printf(m, " (%ld)%s%s",
3284 			   ftrace_rec_count(rec),
3285 			   rec->flags & FTRACE_FL_REGS ? " R" : "  ",
3286 			   rec->flags & FTRACE_FL_IPMODIFY ? " I" : "  ");
3287 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
3288 			ops = ftrace_find_tramp_ops_any(rec);
3289 			if (ops) {
3290 				do {
3291 					seq_printf(m, "\ttramp: %pS (%pS)",
3292 						   (void *)ops->trampoline,
3293 						   (void *)ops->func);
3294 					add_trampoline_func(m, ops, rec);
3295 					ops = ftrace_find_tramp_ops_next(rec, ops);
3296 				} while (ops);
3297 			} else
3298 				seq_puts(m, "\ttramp: ERROR!");
3299 		} else {
3300 			add_trampoline_func(m, NULL, rec);
3301 		}
3302 	}
3303 
3304 	seq_putc(m, '\n');
3305 
3306 	return 0;
3307 }
3308 
3309 static const struct seq_operations show_ftrace_seq_ops = {
3310 	.start = t_start,
3311 	.next = t_next,
3312 	.stop = t_stop,
3313 	.show = t_show,
3314 };
3315 
3316 static int
3317 ftrace_avail_open(struct inode *inode, struct file *file)
3318 {
3319 	struct ftrace_iterator *iter;
3320 
3321 	if (unlikely(ftrace_disabled))
3322 		return -ENODEV;
3323 
3324 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3325 	if (iter) {
3326 		iter->pg = ftrace_pages_start;
3327 		iter->ops = &global_ops;
3328 	}
3329 
3330 	return iter ? 0 : -ENOMEM;
3331 }
3332 
3333 static int
3334 ftrace_enabled_open(struct inode *inode, struct file *file)
3335 {
3336 	struct ftrace_iterator *iter;
3337 
3338 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3339 	if (iter) {
3340 		iter->pg = ftrace_pages_start;
3341 		iter->flags = FTRACE_ITER_ENABLED;
3342 		iter->ops = &global_ops;
3343 	}
3344 
3345 	return iter ? 0 : -ENOMEM;
3346 }
3347 
3348 /**
3349  * ftrace_regex_open - initialize function tracer filter files
3350  * @ops: The ftrace_ops that hold the hash filters
3351  * @flag: The type of filter to process
3352  * @inode: The inode, usually passed in to your open routine
3353  * @file: The file, usually passed in to your open routine
3354  *
3355  * ftrace_regex_open() initializes the filter files for the
3356  * @ops. Depending on @flag it may process the filter hash or
3357  * the notrace hash of @ops. With this called from the open
3358  * routine, you can use ftrace_filter_write() for the write
3359  * routine if @flag has FTRACE_ITER_FILTER set, or
3360  * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3361  * tracing_lseek() should be used as the lseek routine, and
3362  * release must call ftrace_regex_release().
3363  */
3364 int
3365 ftrace_regex_open(struct ftrace_ops *ops, int flag,
3366 		  struct inode *inode, struct file *file)
3367 {
3368 	struct ftrace_iterator *iter;
3369 	struct ftrace_hash *hash;
3370 	int ret = 0;
3371 
3372 	ftrace_ops_init(ops);
3373 
3374 	if (unlikely(ftrace_disabled))
3375 		return -ENODEV;
3376 
3377 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3378 	if (!iter)
3379 		return -ENOMEM;
3380 
3381 	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3382 		kfree(iter);
3383 		return -ENOMEM;
3384 	}
3385 
3386 	iter->ops = ops;
3387 	iter->flags = flag;
3388 
3389 	mutex_lock(&ops->func_hash->regex_lock);
3390 
3391 	if (flag & FTRACE_ITER_NOTRACE)
3392 		hash = ops->func_hash->notrace_hash;
3393 	else
3394 		hash = ops->func_hash->filter_hash;
3395 
3396 	if (file->f_mode & FMODE_WRITE) {
3397 		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3398 
3399 		if (file->f_flags & O_TRUNC)
3400 			iter->hash = alloc_ftrace_hash(size_bits);
3401 		else
3402 			iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3403 
3404 		if (!iter->hash) {
3405 			trace_parser_put(&iter->parser);
3406 			kfree(iter);
3407 			ret = -ENOMEM;
3408 			goto out_unlock;
3409 		}
3410 	}
3411 
3412 	if (file->f_mode & FMODE_READ) {
3413 		iter->pg = ftrace_pages_start;
3414 
3415 		ret = seq_open(file, &show_ftrace_seq_ops);
3416 		if (!ret) {
3417 			struct seq_file *m = file->private_data;
3418 			m->private = iter;
3419 		} else {
3420 			/* Failed */
3421 			free_ftrace_hash(iter->hash);
3422 			trace_parser_put(&iter->parser);
3423 			kfree(iter);
3424 		}
3425 	} else
3426 		file->private_data = iter;
3427 
3428  out_unlock:
3429 	mutex_unlock(&ops->func_hash->regex_lock);
3430 
3431 	return ret;
3432 }
3433 
3434 static int
3435 ftrace_filter_open(struct inode *inode, struct file *file)
3436 {
3437 	struct ftrace_ops *ops = inode->i_private;
3438 
3439 	return ftrace_regex_open(ops,
3440 			FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
3441 			inode, file);
3442 }
3443 
3444 static int
3445 ftrace_notrace_open(struct inode *inode, struct file *file)
3446 {
3447 	struct ftrace_ops *ops = inode->i_private;
3448 
3449 	return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3450 				 inode, file);
3451 }
3452 
3453 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3454 struct ftrace_glob {
3455 	char *search;
3456 	unsigned len;
3457 	int type;
3458 };
3459 
3460 /*
3461  * If symbols in an architecture don't correspond exactly to the user-visible
3462  * name of what they represent, it is possible to define this function to
3463  * perform the necessary adjustments.
3464 */
3465 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3466 {
3467 	return str;
3468 }
3469 
3470 static int ftrace_match(char *str, struct ftrace_glob *g)
3471 {
3472 	int matched = 0;
3473 	int slen;
3474 
3475 	str = arch_ftrace_match_adjust(str, g->search);
3476 
3477 	switch (g->type) {
3478 	case MATCH_FULL:
3479 		if (strcmp(str, g->search) == 0)
3480 			matched = 1;
3481 		break;
3482 	case MATCH_FRONT_ONLY:
3483 		if (strncmp(str, g->search, g->len) == 0)
3484 			matched = 1;
3485 		break;
3486 	case MATCH_MIDDLE_ONLY:
3487 		if (strstr(str, g->search))
3488 			matched = 1;
3489 		break;
3490 	case MATCH_END_ONLY:
3491 		slen = strlen(str);
3492 		if (slen >= g->len &&
3493 		    memcmp(str + slen - g->len, g->search, g->len) == 0)
3494 			matched = 1;
3495 		break;
3496 	}
3497 
3498 	return matched;
3499 }
3500 
3501 static int
3502 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3503 {
3504 	struct ftrace_func_entry *entry;
3505 	int ret = 0;
3506 
3507 	entry = ftrace_lookup_ip(hash, rec->ip);
3508 	if (clear_filter) {
3509 		/* Do nothing if it doesn't exist */
3510 		if (!entry)
3511 			return 0;
3512 
3513 		free_hash_entry(hash, entry);
3514 	} else {
3515 		/* Do nothing if it exists */
3516 		if (entry)
3517 			return 0;
3518 
3519 		ret = add_hash_entry(hash, rec->ip);
3520 	}
3521 	return ret;
3522 }
3523 
3524 static int
3525 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3526 		struct ftrace_glob *mod_g, int exclude_mod)
3527 {
3528 	char str[KSYM_SYMBOL_LEN];
3529 	char *modname;
3530 
3531 	kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3532 
3533 	if (mod_g) {
3534 		int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3535 
3536 		/* blank module name to match all modules */
3537 		if (!mod_g->len) {
3538 			/* blank module globbing: modname xor exclude_mod */
3539 			if ((!exclude_mod) != (!modname))
3540 				goto func_match;
3541 			return 0;
3542 		}
3543 
3544 		/* not matching the module */
3545 		if (!modname || !mod_matches) {
3546 			if (exclude_mod)
3547 				goto func_match;
3548 			else
3549 				return 0;
3550 		}
3551 
3552 		if (mod_matches && exclude_mod)
3553 			return 0;
3554 
3555 func_match:
3556 		/* blank search means to match all funcs in the mod */
3557 		if (!func_g->len)
3558 			return 1;
3559 	}
3560 
3561 	return ftrace_match(str, func_g);
3562 }
3563 
3564 static int
3565 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3566 {
3567 	struct ftrace_page *pg;
3568 	struct dyn_ftrace *rec;
3569 	struct ftrace_glob func_g = { .type = MATCH_FULL };
3570 	struct ftrace_glob mod_g = { .type = MATCH_FULL };
3571 	struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3572 	int exclude_mod = 0;
3573 	int found = 0;
3574 	int ret;
3575 	int clear_filter;
3576 
3577 	if (func) {
3578 		func_g.type = filter_parse_regex(func, len, &func_g.search,
3579 						 &clear_filter);
3580 		func_g.len = strlen(func_g.search);
3581 	}
3582 
3583 	if (mod) {
3584 		mod_g.type = filter_parse_regex(mod, strlen(mod),
3585 				&mod_g.search, &exclude_mod);
3586 		mod_g.len = strlen(mod_g.search);
3587 	}
3588 
3589 	mutex_lock(&ftrace_lock);
3590 
3591 	if (unlikely(ftrace_disabled))
3592 		goto out_unlock;
3593 
3594 	do_for_each_ftrace_rec(pg, rec) {
3595 		if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3596 			ret = enter_record(hash, rec, clear_filter);
3597 			if (ret < 0) {
3598 				found = ret;
3599 				goto out_unlock;
3600 			}
3601 			found = 1;
3602 		}
3603 	} while_for_each_ftrace_rec();
3604  out_unlock:
3605 	mutex_unlock(&ftrace_lock);
3606 
3607 	return found;
3608 }
3609 
3610 static int
3611 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3612 {
3613 	return match_records(hash, buff, len, NULL);
3614 }
3615 
3616 
3617 /*
3618  * We register the module command as a template to show others how
3619  * to register the a command as well.
3620  */
3621 
3622 static int
3623 ftrace_mod_callback(struct ftrace_hash *hash,
3624 		    char *func, char *cmd, char *module, int enable)
3625 {
3626 	int ret;
3627 
3628 	/*
3629 	 * cmd == 'mod' because we only registered this func
3630 	 * for the 'mod' ftrace_func_command.
3631 	 * But if you register one func with multiple commands,
3632 	 * you can tell which command was used by the cmd
3633 	 * parameter.
3634 	 */
3635 	ret = match_records(hash, func, strlen(func), module);
3636 	if (!ret)
3637 		return -EINVAL;
3638 	if (ret < 0)
3639 		return ret;
3640 	return 0;
3641 }
3642 
3643 static struct ftrace_func_command ftrace_mod_cmd = {
3644 	.name			= "mod",
3645 	.func			= ftrace_mod_callback,
3646 };
3647 
3648 static int __init ftrace_mod_cmd_init(void)
3649 {
3650 	return register_ftrace_command(&ftrace_mod_cmd);
3651 }
3652 core_initcall(ftrace_mod_cmd_init);
3653 
3654 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
3655 				      struct ftrace_ops *op, struct pt_regs *pt_regs)
3656 {
3657 	struct ftrace_func_probe *entry;
3658 	struct hlist_head *hhd;
3659 	unsigned long key;
3660 
3661 	key = hash_long(ip, FTRACE_HASH_BITS);
3662 
3663 	hhd = &ftrace_func_hash[key];
3664 
3665 	if (hlist_empty(hhd))
3666 		return;
3667 
3668 	/*
3669 	 * Disable preemption for these calls to prevent a RCU grace
3670 	 * period. This syncs the hash iteration and freeing of items
3671 	 * on the hash. rcu_read_lock is too dangerous here.
3672 	 */
3673 	preempt_disable_notrace();
3674 	hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3675 		if (entry->ip == ip)
3676 			entry->ops->func(ip, parent_ip, &entry->data);
3677 	}
3678 	preempt_enable_notrace();
3679 }
3680 
3681 static struct ftrace_ops trace_probe_ops __read_mostly =
3682 {
3683 	.func		= function_trace_probe_call,
3684 	.flags		= FTRACE_OPS_FL_INITIALIZED,
3685 	INIT_OPS_HASH(trace_probe_ops)
3686 };
3687 
3688 static int ftrace_probe_registered;
3689 
3690 static void __enable_ftrace_function_probe(struct ftrace_ops_hash *old_hash)
3691 {
3692 	int ret;
3693 	int i;
3694 
3695 	if (ftrace_probe_registered) {
3696 		/* still need to update the function call sites */
3697 		if (ftrace_enabled)
3698 			ftrace_run_modify_code(&trace_probe_ops, FTRACE_UPDATE_CALLS,
3699 					       old_hash);
3700 		return;
3701 	}
3702 
3703 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3704 		struct hlist_head *hhd = &ftrace_func_hash[i];
3705 		if (hhd->first)
3706 			break;
3707 	}
3708 	/* Nothing registered? */
3709 	if (i == FTRACE_FUNC_HASHSIZE)
3710 		return;
3711 
3712 	ret = ftrace_startup(&trace_probe_ops, 0);
3713 
3714 	ftrace_probe_registered = 1;
3715 }
3716 
3717 static void __disable_ftrace_function_probe(void)
3718 {
3719 	int i;
3720 
3721 	if (!ftrace_probe_registered)
3722 		return;
3723 
3724 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3725 		struct hlist_head *hhd = &ftrace_func_hash[i];
3726 		if (hhd->first)
3727 			return;
3728 	}
3729 
3730 	/* no more funcs left */
3731 	ftrace_shutdown(&trace_probe_ops, 0);
3732 
3733 	ftrace_probe_registered = 0;
3734 }
3735 
3736 
3737 static void ftrace_free_entry(struct ftrace_func_probe *entry)
3738 {
3739 	if (entry->ops->free)
3740 		entry->ops->free(entry->ops, entry->ip, &entry->data);
3741 	kfree(entry);
3742 }
3743 
3744 int
3745 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3746 			      void *data)
3747 {
3748 	struct ftrace_ops_hash old_hash_ops;
3749 	struct ftrace_func_probe *entry;
3750 	struct ftrace_glob func_g;
3751 	struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3752 	struct ftrace_hash *old_hash = *orig_hash;
3753 	struct ftrace_hash *hash;
3754 	struct ftrace_page *pg;
3755 	struct dyn_ftrace *rec;
3756 	int not;
3757 	unsigned long key;
3758 	int count = 0;
3759 	int ret;
3760 
3761 	func_g.type = filter_parse_regex(glob, strlen(glob),
3762 			&func_g.search, &not);
3763 	func_g.len = strlen(func_g.search);
3764 
3765 	/* we do not support '!' for function probes */
3766 	if (WARN_ON(not))
3767 		return -EINVAL;
3768 
3769 	mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3770 
3771 	old_hash_ops.filter_hash = old_hash;
3772 	/* Probes only have filters */
3773 	old_hash_ops.notrace_hash = NULL;
3774 
3775 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
3776 	if (!hash) {
3777 		count = -ENOMEM;
3778 		goto out;
3779 	}
3780 
3781 	if (unlikely(ftrace_disabled)) {
3782 		count = -ENODEV;
3783 		goto out;
3784 	}
3785 
3786 	mutex_lock(&ftrace_lock);
3787 
3788 	do_for_each_ftrace_rec(pg, rec) {
3789 
3790 		if (!ftrace_match_record(rec, &func_g, NULL, 0))
3791 			continue;
3792 
3793 		entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3794 		if (!entry) {
3795 			/* If we did not process any, then return error */
3796 			if (!count)
3797 				count = -ENOMEM;
3798 			goto out_unlock;
3799 		}
3800 
3801 		count++;
3802 
3803 		entry->data = data;
3804 
3805 		/*
3806 		 * The caller might want to do something special
3807 		 * for each function we find. We call the callback
3808 		 * to give the caller an opportunity to do so.
3809 		 */
3810 		if (ops->init) {
3811 			if (ops->init(ops, rec->ip, &entry->data) < 0) {
3812 				/* caller does not like this func */
3813 				kfree(entry);
3814 				continue;
3815 			}
3816 		}
3817 
3818 		ret = enter_record(hash, rec, 0);
3819 		if (ret < 0) {
3820 			kfree(entry);
3821 			count = ret;
3822 			goto out_unlock;
3823 		}
3824 
3825 		entry->ops = ops;
3826 		entry->ip = rec->ip;
3827 
3828 		key = hash_long(entry->ip, FTRACE_HASH_BITS);
3829 		hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3830 
3831 	} while_for_each_ftrace_rec();
3832 
3833 	ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3834 
3835 	__enable_ftrace_function_probe(&old_hash_ops);
3836 
3837 	if (!ret)
3838 		free_ftrace_hash_rcu(old_hash);
3839 	else
3840 		count = ret;
3841 
3842  out_unlock:
3843 	mutex_unlock(&ftrace_lock);
3844  out:
3845 	mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3846 	free_ftrace_hash(hash);
3847 
3848 	return count;
3849 }
3850 
3851 enum {
3852 	PROBE_TEST_FUNC		= 1,
3853 	PROBE_TEST_DATA		= 2
3854 };
3855 
3856 static void
3857 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3858 				  void *data, int flags)
3859 {
3860 	struct ftrace_func_entry *rec_entry;
3861 	struct ftrace_func_probe *entry;
3862 	struct ftrace_func_probe *p;
3863 	struct ftrace_glob func_g;
3864 	struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3865 	struct ftrace_hash *old_hash = *orig_hash;
3866 	struct list_head free_list;
3867 	struct ftrace_hash *hash;
3868 	struct hlist_node *tmp;
3869 	char str[KSYM_SYMBOL_LEN];
3870 	int i, ret;
3871 
3872 	if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3873 		func_g.search = NULL;
3874 	else if (glob) {
3875 		int not;
3876 
3877 		func_g.type = filter_parse_regex(glob, strlen(glob),
3878 						 &func_g.search, &not);
3879 		func_g.len = strlen(func_g.search);
3880 		func_g.search = glob;
3881 
3882 		/* we do not support '!' for function probes */
3883 		if (WARN_ON(not))
3884 			return;
3885 	}
3886 
3887 	mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3888 
3889 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3890 	if (!hash)
3891 		/* Hmm, should report this somehow */
3892 		goto out_unlock;
3893 
3894 	INIT_LIST_HEAD(&free_list);
3895 
3896 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3897 		struct hlist_head *hhd = &ftrace_func_hash[i];
3898 
3899 		hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3900 
3901 			/* break up if statements for readability */
3902 			if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3903 				continue;
3904 
3905 			if ((flags & PROBE_TEST_DATA) && entry->data != data)
3906 				continue;
3907 
3908 			/* do this last, since it is the most expensive */
3909 			if (func_g.search) {
3910 				kallsyms_lookup(entry->ip, NULL, NULL,
3911 						NULL, str);
3912 				if (!ftrace_match(str, &func_g))
3913 					continue;
3914 			}
3915 
3916 			rec_entry = ftrace_lookup_ip(hash, entry->ip);
3917 			/* It is possible more than one entry had this ip */
3918 			if (rec_entry)
3919 				free_hash_entry(hash, rec_entry);
3920 
3921 			hlist_del_rcu(&entry->node);
3922 			list_add(&entry->free_list, &free_list);
3923 		}
3924 	}
3925 	mutex_lock(&ftrace_lock);
3926 	__disable_ftrace_function_probe();
3927 	/*
3928 	 * Remove after the disable is called. Otherwise, if the last
3929 	 * probe is removed, a null hash means *all enabled*.
3930 	 */
3931 	ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3932 	synchronize_sched();
3933 	if (!ret)
3934 		free_ftrace_hash_rcu(old_hash);
3935 
3936 	list_for_each_entry_safe(entry, p, &free_list, free_list) {
3937 		list_del(&entry->free_list);
3938 		ftrace_free_entry(entry);
3939 	}
3940 	mutex_unlock(&ftrace_lock);
3941 
3942  out_unlock:
3943 	mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3944 	free_ftrace_hash(hash);
3945 }
3946 
3947 void
3948 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3949 				void *data)
3950 {
3951 	__unregister_ftrace_function_probe(glob, ops, data,
3952 					  PROBE_TEST_FUNC | PROBE_TEST_DATA);
3953 }
3954 
3955 void
3956 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3957 {
3958 	__unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3959 }
3960 
3961 void unregister_ftrace_function_probe_all(char *glob)
3962 {
3963 	__unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3964 }
3965 
3966 static LIST_HEAD(ftrace_commands);
3967 static DEFINE_MUTEX(ftrace_cmd_mutex);
3968 
3969 /*
3970  * Currently we only register ftrace commands from __init, so mark this
3971  * __init too.
3972  */
3973 __init int register_ftrace_command(struct ftrace_func_command *cmd)
3974 {
3975 	struct ftrace_func_command *p;
3976 	int ret = 0;
3977 
3978 	mutex_lock(&ftrace_cmd_mutex);
3979 	list_for_each_entry(p, &ftrace_commands, list) {
3980 		if (strcmp(cmd->name, p->name) == 0) {
3981 			ret = -EBUSY;
3982 			goto out_unlock;
3983 		}
3984 	}
3985 	list_add(&cmd->list, &ftrace_commands);
3986  out_unlock:
3987 	mutex_unlock(&ftrace_cmd_mutex);
3988 
3989 	return ret;
3990 }
3991 
3992 /*
3993  * Currently we only unregister ftrace commands from __init, so mark
3994  * this __init too.
3995  */
3996 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
3997 {
3998 	struct ftrace_func_command *p, *n;
3999 	int ret = -ENODEV;
4000 
4001 	mutex_lock(&ftrace_cmd_mutex);
4002 	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4003 		if (strcmp(cmd->name, p->name) == 0) {
4004 			ret = 0;
4005 			list_del_init(&p->list);
4006 			goto out_unlock;
4007 		}
4008 	}
4009  out_unlock:
4010 	mutex_unlock(&ftrace_cmd_mutex);
4011 
4012 	return ret;
4013 }
4014 
4015 static int ftrace_process_regex(struct ftrace_hash *hash,
4016 				char *buff, int len, int enable)
4017 {
4018 	char *func, *command, *next = buff;
4019 	struct ftrace_func_command *p;
4020 	int ret = -EINVAL;
4021 
4022 	func = strsep(&next, ":");
4023 
4024 	if (!next) {
4025 		ret = ftrace_match_records(hash, func, len);
4026 		if (!ret)
4027 			ret = -EINVAL;
4028 		if (ret < 0)
4029 			return ret;
4030 		return 0;
4031 	}
4032 
4033 	/* command found */
4034 
4035 	command = strsep(&next, ":");
4036 
4037 	mutex_lock(&ftrace_cmd_mutex);
4038 	list_for_each_entry(p, &ftrace_commands, list) {
4039 		if (strcmp(p->name, command) == 0) {
4040 			ret = p->func(hash, func, command, next, enable);
4041 			goto out_unlock;
4042 		}
4043 	}
4044  out_unlock:
4045 	mutex_unlock(&ftrace_cmd_mutex);
4046 
4047 	return ret;
4048 }
4049 
4050 static ssize_t
4051 ftrace_regex_write(struct file *file, const char __user *ubuf,
4052 		   size_t cnt, loff_t *ppos, int enable)
4053 {
4054 	struct ftrace_iterator *iter;
4055 	struct trace_parser *parser;
4056 	ssize_t ret, read;
4057 
4058 	if (!cnt)
4059 		return 0;
4060 
4061 	if (file->f_mode & FMODE_READ) {
4062 		struct seq_file *m = file->private_data;
4063 		iter = m->private;
4064 	} else
4065 		iter = file->private_data;
4066 
4067 	if (unlikely(ftrace_disabled))
4068 		return -ENODEV;
4069 
4070 	/* iter->hash is a local copy, so we don't need regex_lock */
4071 
4072 	parser = &iter->parser;
4073 	read = trace_get_user(parser, ubuf, cnt, ppos);
4074 
4075 	if (read >= 0 && trace_parser_loaded(parser) &&
4076 	    !trace_parser_cont(parser)) {
4077 		ret = ftrace_process_regex(iter->hash, parser->buffer,
4078 					   parser->idx, enable);
4079 		trace_parser_clear(parser);
4080 		if (ret < 0)
4081 			goto out;
4082 	}
4083 
4084 	ret = read;
4085  out:
4086 	return ret;
4087 }
4088 
4089 ssize_t
4090 ftrace_filter_write(struct file *file, const char __user *ubuf,
4091 		    size_t cnt, loff_t *ppos)
4092 {
4093 	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4094 }
4095 
4096 ssize_t
4097 ftrace_notrace_write(struct file *file, const char __user *ubuf,
4098 		     size_t cnt, loff_t *ppos)
4099 {
4100 	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4101 }
4102 
4103 static int
4104 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4105 {
4106 	struct ftrace_func_entry *entry;
4107 
4108 	if (!ftrace_location(ip))
4109 		return -EINVAL;
4110 
4111 	if (remove) {
4112 		entry = ftrace_lookup_ip(hash, ip);
4113 		if (!entry)
4114 			return -ENOENT;
4115 		free_hash_entry(hash, entry);
4116 		return 0;
4117 	}
4118 
4119 	return add_hash_entry(hash, ip);
4120 }
4121 
4122 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4123 				   struct ftrace_ops_hash *old_hash)
4124 {
4125 	struct ftrace_ops *op;
4126 
4127 	if (!ftrace_enabled)
4128 		return;
4129 
4130 	if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4131 		ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4132 		return;
4133 	}
4134 
4135 	/*
4136 	 * If this is the shared global_ops filter, then we need to
4137 	 * check if there is another ops that shares it, is enabled.
4138 	 * If so, we still need to run the modify code.
4139 	 */
4140 	if (ops->func_hash != &global_ops.local_hash)
4141 		return;
4142 
4143 	do_for_each_ftrace_op(op, ftrace_ops_list) {
4144 		if (op->func_hash == &global_ops.local_hash &&
4145 		    op->flags & FTRACE_OPS_FL_ENABLED) {
4146 			ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4147 			/* Only need to do this once */
4148 			return;
4149 		}
4150 	} while_for_each_ftrace_op(op);
4151 }
4152 
4153 static int
4154 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4155 		unsigned long ip, int remove, int reset, int enable)
4156 {
4157 	struct ftrace_hash **orig_hash;
4158 	struct ftrace_ops_hash old_hash_ops;
4159 	struct ftrace_hash *old_hash;
4160 	struct ftrace_hash *hash;
4161 	int ret;
4162 
4163 	if (unlikely(ftrace_disabled))
4164 		return -ENODEV;
4165 
4166 	mutex_lock(&ops->func_hash->regex_lock);
4167 
4168 	if (enable)
4169 		orig_hash = &ops->func_hash->filter_hash;
4170 	else
4171 		orig_hash = &ops->func_hash->notrace_hash;
4172 
4173 	if (reset)
4174 		hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4175 	else
4176 		hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4177 
4178 	if (!hash) {
4179 		ret = -ENOMEM;
4180 		goto out_regex_unlock;
4181 	}
4182 
4183 	if (buf && !ftrace_match_records(hash, buf, len)) {
4184 		ret = -EINVAL;
4185 		goto out_regex_unlock;
4186 	}
4187 	if (ip) {
4188 		ret = ftrace_match_addr(hash, ip, remove);
4189 		if (ret < 0)
4190 			goto out_regex_unlock;
4191 	}
4192 
4193 	mutex_lock(&ftrace_lock);
4194 	old_hash = *orig_hash;
4195 	old_hash_ops.filter_hash = ops->func_hash->filter_hash;
4196 	old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
4197 	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
4198 	if (!ret) {
4199 		ftrace_ops_update_code(ops, &old_hash_ops);
4200 		free_ftrace_hash_rcu(old_hash);
4201 	}
4202 	mutex_unlock(&ftrace_lock);
4203 
4204  out_regex_unlock:
4205 	mutex_unlock(&ops->func_hash->regex_lock);
4206 
4207 	free_ftrace_hash(hash);
4208 	return ret;
4209 }
4210 
4211 static int
4212 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4213 		int reset, int enable)
4214 {
4215 	return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4216 }
4217 
4218 /**
4219  * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4220  * @ops - the ops to set the filter with
4221  * @ip - the address to add to or remove from the filter.
4222  * @remove - non zero to remove the ip from the filter
4223  * @reset - non zero to reset all filters before applying this filter.
4224  *
4225  * Filters denote which functions should be enabled when tracing is enabled
4226  * If @ip is NULL, it failes to update filter.
4227  */
4228 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4229 			 int remove, int reset)
4230 {
4231 	ftrace_ops_init(ops);
4232 	return ftrace_set_addr(ops, ip, remove, reset, 1);
4233 }
4234 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4235 
4236 static int
4237 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4238 		 int reset, int enable)
4239 {
4240 	return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4241 }
4242 
4243 /**
4244  * ftrace_set_filter - set a function to filter on in ftrace
4245  * @ops - the ops to set the filter with
4246  * @buf - the string that holds the function filter text.
4247  * @len - the length of the string.
4248  * @reset - non zero to reset all filters before applying this filter.
4249  *
4250  * Filters denote which functions should be enabled when tracing is enabled.
4251  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4252  */
4253 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4254 		       int len, int reset)
4255 {
4256 	ftrace_ops_init(ops);
4257 	return ftrace_set_regex(ops, buf, len, reset, 1);
4258 }
4259 EXPORT_SYMBOL_GPL(ftrace_set_filter);
4260 
4261 /**
4262  * ftrace_set_notrace - set a function to not trace in ftrace
4263  * @ops - the ops to set the notrace filter with
4264  * @buf - the string that holds the function notrace text.
4265  * @len - the length of the string.
4266  * @reset - non zero to reset all filters before applying this filter.
4267  *
4268  * Notrace Filters denote which functions should not be enabled when tracing
4269  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4270  * for tracing.
4271  */
4272 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4273 			int len, int reset)
4274 {
4275 	ftrace_ops_init(ops);
4276 	return ftrace_set_regex(ops, buf, len, reset, 0);
4277 }
4278 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4279 /**
4280  * ftrace_set_global_filter - set a function to filter on with global tracers
4281  * @buf - the string that holds the function filter text.
4282  * @len - the length of the string.
4283  * @reset - non zero to reset all filters before applying this filter.
4284  *
4285  * Filters denote which functions should be enabled when tracing is enabled.
4286  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4287  */
4288 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4289 {
4290 	ftrace_set_regex(&global_ops, buf, len, reset, 1);
4291 }
4292 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4293 
4294 /**
4295  * ftrace_set_global_notrace - set a function to not trace with global tracers
4296  * @buf - the string that holds the function notrace text.
4297  * @len - the length of the string.
4298  * @reset - non zero to reset all filters before applying this filter.
4299  *
4300  * Notrace Filters denote which functions should not be enabled when tracing
4301  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4302  * for tracing.
4303  */
4304 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4305 {
4306 	ftrace_set_regex(&global_ops, buf, len, reset, 0);
4307 }
4308 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4309 
4310 /*
4311  * command line interface to allow users to set filters on boot up.
4312  */
4313 #define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
4314 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4315 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4316 
4317 /* Used by function selftest to not test if filter is set */
4318 bool ftrace_filter_param __initdata;
4319 
4320 static int __init set_ftrace_notrace(char *str)
4321 {
4322 	ftrace_filter_param = true;
4323 	strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4324 	return 1;
4325 }
4326 __setup("ftrace_notrace=", set_ftrace_notrace);
4327 
4328 static int __init set_ftrace_filter(char *str)
4329 {
4330 	ftrace_filter_param = true;
4331 	strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4332 	return 1;
4333 }
4334 __setup("ftrace_filter=", set_ftrace_filter);
4335 
4336 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4337 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4338 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4339 static int ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer);
4340 
4341 static unsigned long save_global_trampoline;
4342 static unsigned long save_global_flags;
4343 
4344 static int __init set_graph_function(char *str)
4345 {
4346 	strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4347 	return 1;
4348 }
4349 __setup("ftrace_graph_filter=", set_graph_function);
4350 
4351 static int __init set_graph_notrace_function(char *str)
4352 {
4353 	strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4354 	return 1;
4355 }
4356 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
4357 
4358 static void __init set_ftrace_early_graph(char *buf, int enable)
4359 {
4360 	int ret;
4361 	char *func;
4362 	unsigned long *table = ftrace_graph_funcs;
4363 	int *count = &ftrace_graph_count;
4364 
4365 	if (!enable) {
4366 		table = ftrace_graph_notrace_funcs;
4367 		count = &ftrace_graph_notrace_count;
4368 	}
4369 
4370 	while (buf) {
4371 		func = strsep(&buf, ",");
4372 		/* we allow only one expression at a time */
4373 		ret = ftrace_set_func(table, count, FTRACE_GRAPH_MAX_FUNCS, func);
4374 		if (ret)
4375 			printk(KERN_DEBUG "ftrace: function %s not "
4376 					  "traceable\n", func);
4377 	}
4378 }
4379 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4380 
4381 void __init
4382 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
4383 {
4384 	char *func;
4385 
4386 	ftrace_ops_init(ops);
4387 
4388 	while (buf) {
4389 		func = strsep(&buf, ",");
4390 		ftrace_set_regex(ops, func, strlen(func), 0, enable);
4391 	}
4392 }
4393 
4394 static void __init set_ftrace_early_filters(void)
4395 {
4396 	if (ftrace_filter_buf[0])
4397 		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
4398 	if (ftrace_notrace_buf[0])
4399 		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
4400 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4401 	if (ftrace_graph_buf[0])
4402 		set_ftrace_early_graph(ftrace_graph_buf, 1);
4403 	if (ftrace_graph_notrace_buf[0])
4404 		set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
4405 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4406 }
4407 
4408 int ftrace_regex_release(struct inode *inode, struct file *file)
4409 {
4410 	struct seq_file *m = (struct seq_file *)file->private_data;
4411 	struct ftrace_ops_hash old_hash_ops;
4412 	struct ftrace_iterator *iter;
4413 	struct ftrace_hash **orig_hash;
4414 	struct ftrace_hash *old_hash;
4415 	struct trace_parser *parser;
4416 	int filter_hash;
4417 	int ret;
4418 
4419 	if (file->f_mode & FMODE_READ) {
4420 		iter = m->private;
4421 		seq_release(inode, file);
4422 	} else
4423 		iter = file->private_data;
4424 
4425 	parser = &iter->parser;
4426 	if (trace_parser_loaded(parser)) {
4427 		parser->buffer[parser->idx] = 0;
4428 		ftrace_match_records(iter->hash, parser->buffer, parser->idx);
4429 	}
4430 
4431 	trace_parser_put(parser);
4432 
4433 	mutex_lock(&iter->ops->func_hash->regex_lock);
4434 
4435 	if (file->f_mode & FMODE_WRITE) {
4436 		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
4437 
4438 		if (filter_hash)
4439 			orig_hash = &iter->ops->func_hash->filter_hash;
4440 		else
4441 			orig_hash = &iter->ops->func_hash->notrace_hash;
4442 
4443 		mutex_lock(&ftrace_lock);
4444 		old_hash = *orig_hash;
4445 		old_hash_ops.filter_hash = iter->ops->func_hash->filter_hash;
4446 		old_hash_ops.notrace_hash = iter->ops->func_hash->notrace_hash;
4447 		ret = ftrace_hash_move(iter->ops, filter_hash,
4448 				       orig_hash, iter->hash);
4449 		if (!ret) {
4450 			ftrace_ops_update_code(iter->ops, &old_hash_ops);
4451 			free_ftrace_hash_rcu(old_hash);
4452 		}
4453 		mutex_unlock(&ftrace_lock);
4454 	}
4455 
4456 	mutex_unlock(&iter->ops->func_hash->regex_lock);
4457 	free_ftrace_hash(iter->hash);
4458 	kfree(iter);
4459 
4460 	return 0;
4461 }
4462 
4463 static const struct file_operations ftrace_avail_fops = {
4464 	.open = ftrace_avail_open,
4465 	.read = seq_read,
4466 	.llseek = seq_lseek,
4467 	.release = seq_release_private,
4468 };
4469 
4470 static const struct file_operations ftrace_enabled_fops = {
4471 	.open = ftrace_enabled_open,
4472 	.read = seq_read,
4473 	.llseek = seq_lseek,
4474 	.release = seq_release_private,
4475 };
4476 
4477 static const struct file_operations ftrace_filter_fops = {
4478 	.open = ftrace_filter_open,
4479 	.read = seq_read,
4480 	.write = ftrace_filter_write,
4481 	.llseek = tracing_lseek,
4482 	.release = ftrace_regex_release,
4483 };
4484 
4485 static const struct file_operations ftrace_notrace_fops = {
4486 	.open = ftrace_notrace_open,
4487 	.read = seq_read,
4488 	.write = ftrace_notrace_write,
4489 	.llseek = tracing_lseek,
4490 	.release = ftrace_regex_release,
4491 };
4492 
4493 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4494 
4495 static DEFINE_MUTEX(graph_lock);
4496 
4497 int ftrace_graph_count;
4498 int ftrace_graph_notrace_count;
4499 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4500 unsigned long ftrace_graph_notrace_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4501 
4502 struct ftrace_graph_data {
4503 	unsigned long *table;
4504 	size_t size;
4505 	int *count;
4506 	const struct seq_operations *seq_ops;
4507 };
4508 
4509 static void *
4510 __g_next(struct seq_file *m, loff_t *pos)
4511 {
4512 	struct ftrace_graph_data *fgd = m->private;
4513 
4514 	if (*pos >= *fgd->count)
4515 		return NULL;
4516 	return &fgd->table[*pos];
4517 }
4518 
4519 static void *
4520 g_next(struct seq_file *m, void *v, loff_t *pos)
4521 {
4522 	(*pos)++;
4523 	return __g_next(m, pos);
4524 }
4525 
4526 static void *g_start(struct seq_file *m, loff_t *pos)
4527 {
4528 	struct ftrace_graph_data *fgd = m->private;
4529 
4530 	mutex_lock(&graph_lock);
4531 
4532 	/* Nothing, tell g_show to print all functions are enabled */
4533 	if (!*fgd->count && !*pos)
4534 		return (void *)1;
4535 
4536 	return __g_next(m, pos);
4537 }
4538 
4539 static void g_stop(struct seq_file *m, void *p)
4540 {
4541 	mutex_unlock(&graph_lock);
4542 }
4543 
4544 static int g_show(struct seq_file *m, void *v)
4545 {
4546 	unsigned long *ptr = v;
4547 
4548 	if (!ptr)
4549 		return 0;
4550 
4551 	if (ptr == (unsigned long *)1) {
4552 		struct ftrace_graph_data *fgd = m->private;
4553 
4554 		if (fgd->table == ftrace_graph_funcs)
4555 			seq_puts(m, "#### all functions enabled ####\n");
4556 		else
4557 			seq_puts(m, "#### no functions disabled ####\n");
4558 		return 0;
4559 	}
4560 
4561 	seq_printf(m, "%ps\n", (void *)*ptr);
4562 
4563 	return 0;
4564 }
4565 
4566 static const struct seq_operations ftrace_graph_seq_ops = {
4567 	.start = g_start,
4568 	.next = g_next,
4569 	.stop = g_stop,
4570 	.show = g_show,
4571 };
4572 
4573 static int
4574 __ftrace_graph_open(struct inode *inode, struct file *file,
4575 		    struct ftrace_graph_data *fgd)
4576 {
4577 	int ret = 0;
4578 
4579 	mutex_lock(&graph_lock);
4580 	if ((file->f_mode & FMODE_WRITE) &&
4581 	    (file->f_flags & O_TRUNC)) {
4582 		*fgd->count = 0;
4583 		memset(fgd->table, 0, fgd->size * sizeof(*fgd->table));
4584 	}
4585 	mutex_unlock(&graph_lock);
4586 
4587 	if (file->f_mode & FMODE_READ) {
4588 		ret = seq_open(file, fgd->seq_ops);
4589 		if (!ret) {
4590 			struct seq_file *m = file->private_data;
4591 			m->private = fgd;
4592 		}
4593 	} else
4594 		file->private_data = fgd;
4595 
4596 	return ret;
4597 }
4598 
4599 static int
4600 ftrace_graph_open(struct inode *inode, struct file *file)
4601 {
4602 	struct ftrace_graph_data *fgd;
4603 
4604 	if (unlikely(ftrace_disabled))
4605 		return -ENODEV;
4606 
4607 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4608 	if (fgd == NULL)
4609 		return -ENOMEM;
4610 
4611 	fgd->table = ftrace_graph_funcs;
4612 	fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4613 	fgd->count = &ftrace_graph_count;
4614 	fgd->seq_ops = &ftrace_graph_seq_ops;
4615 
4616 	return __ftrace_graph_open(inode, file, fgd);
4617 }
4618 
4619 static int
4620 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
4621 {
4622 	struct ftrace_graph_data *fgd;
4623 
4624 	if (unlikely(ftrace_disabled))
4625 		return -ENODEV;
4626 
4627 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4628 	if (fgd == NULL)
4629 		return -ENOMEM;
4630 
4631 	fgd->table = ftrace_graph_notrace_funcs;
4632 	fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4633 	fgd->count = &ftrace_graph_notrace_count;
4634 	fgd->seq_ops = &ftrace_graph_seq_ops;
4635 
4636 	return __ftrace_graph_open(inode, file, fgd);
4637 }
4638 
4639 static int
4640 ftrace_graph_release(struct inode *inode, struct file *file)
4641 {
4642 	if (file->f_mode & FMODE_READ) {
4643 		struct seq_file *m = file->private_data;
4644 
4645 		kfree(m->private);
4646 		seq_release(inode, file);
4647 	} else {
4648 		kfree(file->private_data);
4649 	}
4650 
4651 	return 0;
4652 }
4653 
4654 static int
4655 ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer)
4656 {
4657 	struct ftrace_glob func_g;
4658 	struct dyn_ftrace *rec;
4659 	struct ftrace_page *pg;
4660 	int fail = 1;
4661 	int not;
4662 	bool exists;
4663 	int i;
4664 
4665 	/* decode regex */
4666 	func_g.type = filter_parse_regex(buffer, strlen(buffer),
4667 					 &func_g.search, &not);
4668 	if (!not && *idx >= size)
4669 		return -EBUSY;
4670 
4671 	func_g.len = strlen(func_g.search);
4672 
4673 	mutex_lock(&ftrace_lock);
4674 
4675 	if (unlikely(ftrace_disabled)) {
4676 		mutex_unlock(&ftrace_lock);
4677 		return -ENODEV;
4678 	}
4679 
4680 	do_for_each_ftrace_rec(pg, rec) {
4681 
4682 		if (ftrace_match_record(rec, &func_g, NULL, 0)) {
4683 			/* if it is in the array */
4684 			exists = false;
4685 			for (i = 0; i < *idx; i++) {
4686 				if (array[i] == rec->ip) {
4687 					exists = true;
4688 					break;
4689 				}
4690 			}
4691 
4692 			if (!not) {
4693 				fail = 0;
4694 				if (!exists) {
4695 					array[(*idx)++] = rec->ip;
4696 					if (*idx >= size)
4697 						goto out;
4698 				}
4699 			} else {
4700 				if (exists) {
4701 					array[i] = array[--(*idx)];
4702 					array[*idx] = 0;
4703 					fail = 0;
4704 				}
4705 			}
4706 		}
4707 	} while_for_each_ftrace_rec();
4708 out:
4709 	mutex_unlock(&ftrace_lock);
4710 
4711 	if (fail)
4712 		return -EINVAL;
4713 
4714 	return 0;
4715 }
4716 
4717 static ssize_t
4718 ftrace_graph_write(struct file *file, const char __user *ubuf,
4719 		   size_t cnt, loff_t *ppos)
4720 {
4721 	struct trace_parser parser;
4722 	ssize_t read, ret = 0;
4723 	struct ftrace_graph_data *fgd = file->private_data;
4724 
4725 	if (!cnt)
4726 		return 0;
4727 
4728 	if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX))
4729 		return -ENOMEM;
4730 
4731 	read = trace_get_user(&parser, ubuf, cnt, ppos);
4732 
4733 	if (read >= 0 && trace_parser_loaded((&parser))) {
4734 		parser.buffer[parser.idx] = 0;
4735 
4736 		mutex_lock(&graph_lock);
4737 
4738 		/* we allow only one expression at a time */
4739 		ret = ftrace_set_func(fgd->table, fgd->count, fgd->size,
4740 				      parser.buffer);
4741 
4742 		mutex_unlock(&graph_lock);
4743 	}
4744 
4745 	if (!ret)
4746 		ret = read;
4747 
4748 	trace_parser_put(&parser);
4749 
4750 	return ret;
4751 }
4752 
4753 static const struct file_operations ftrace_graph_fops = {
4754 	.open		= ftrace_graph_open,
4755 	.read		= seq_read,
4756 	.write		= ftrace_graph_write,
4757 	.llseek		= tracing_lseek,
4758 	.release	= ftrace_graph_release,
4759 };
4760 
4761 static const struct file_operations ftrace_graph_notrace_fops = {
4762 	.open		= ftrace_graph_notrace_open,
4763 	.read		= seq_read,
4764 	.write		= ftrace_graph_write,
4765 	.llseek		= tracing_lseek,
4766 	.release	= ftrace_graph_release,
4767 };
4768 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4769 
4770 void ftrace_create_filter_files(struct ftrace_ops *ops,
4771 				struct dentry *parent)
4772 {
4773 
4774 	trace_create_file("set_ftrace_filter", 0644, parent,
4775 			  ops, &ftrace_filter_fops);
4776 
4777 	trace_create_file("set_ftrace_notrace", 0644, parent,
4778 			  ops, &ftrace_notrace_fops);
4779 }
4780 
4781 /*
4782  * The name "destroy_filter_files" is really a misnomer. Although
4783  * in the future, it may actualy delete the files, but this is
4784  * really intended to make sure the ops passed in are disabled
4785  * and that when this function returns, the caller is free to
4786  * free the ops.
4787  *
4788  * The "destroy" name is only to match the "create" name that this
4789  * should be paired with.
4790  */
4791 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
4792 {
4793 	mutex_lock(&ftrace_lock);
4794 	if (ops->flags & FTRACE_OPS_FL_ENABLED)
4795 		ftrace_shutdown(ops, 0);
4796 	ops->flags |= FTRACE_OPS_FL_DELETED;
4797 	mutex_unlock(&ftrace_lock);
4798 }
4799 
4800 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
4801 {
4802 
4803 	trace_create_file("available_filter_functions", 0444,
4804 			d_tracer, NULL, &ftrace_avail_fops);
4805 
4806 	trace_create_file("enabled_functions", 0444,
4807 			d_tracer, NULL, &ftrace_enabled_fops);
4808 
4809 	ftrace_create_filter_files(&global_ops, d_tracer);
4810 
4811 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4812 	trace_create_file("set_graph_function", 0444, d_tracer,
4813 				    NULL,
4814 				    &ftrace_graph_fops);
4815 	trace_create_file("set_graph_notrace", 0444, d_tracer,
4816 				    NULL,
4817 				    &ftrace_graph_notrace_fops);
4818 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4819 
4820 	return 0;
4821 }
4822 
4823 static int ftrace_cmp_ips(const void *a, const void *b)
4824 {
4825 	const unsigned long *ipa = a;
4826 	const unsigned long *ipb = b;
4827 
4828 	if (*ipa > *ipb)
4829 		return 1;
4830 	if (*ipa < *ipb)
4831 		return -1;
4832 	return 0;
4833 }
4834 
4835 static int ftrace_process_locs(struct module *mod,
4836 			       unsigned long *start,
4837 			       unsigned long *end)
4838 {
4839 	struct ftrace_page *start_pg;
4840 	struct ftrace_page *pg;
4841 	struct dyn_ftrace *rec;
4842 	unsigned long count;
4843 	unsigned long *p;
4844 	unsigned long addr;
4845 	unsigned long flags = 0; /* Shut up gcc */
4846 	int ret = -ENOMEM;
4847 
4848 	count = end - start;
4849 
4850 	if (!count)
4851 		return 0;
4852 
4853 	sort(start, count, sizeof(*start),
4854 	     ftrace_cmp_ips, NULL);
4855 
4856 	start_pg = ftrace_allocate_pages(count);
4857 	if (!start_pg)
4858 		return -ENOMEM;
4859 
4860 	mutex_lock(&ftrace_lock);
4861 
4862 	/*
4863 	 * Core and each module needs their own pages, as
4864 	 * modules will free them when they are removed.
4865 	 * Force a new page to be allocated for modules.
4866 	 */
4867 	if (!mod) {
4868 		WARN_ON(ftrace_pages || ftrace_pages_start);
4869 		/* First initialization */
4870 		ftrace_pages = ftrace_pages_start = start_pg;
4871 	} else {
4872 		if (!ftrace_pages)
4873 			goto out;
4874 
4875 		if (WARN_ON(ftrace_pages->next)) {
4876 			/* Hmm, we have free pages? */
4877 			while (ftrace_pages->next)
4878 				ftrace_pages = ftrace_pages->next;
4879 		}
4880 
4881 		ftrace_pages->next = start_pg;
4882 	}
4883 
4884 	p = start;
4885 	pg = start_pg;
4886 	while (p < end) {
4887 		addr = ftrace_call_adjust(*p++);
4888 		/*
4889 		 * Some architecture linkers will pad between
4890 		 * the different mcount_loc sections of different
4891 		 * object files to satisfy alignments.
4892 		 * Skip any NULL pointers.
4893 		 */
4894 		if (!addr)
4895 			continue;
4896 
4897 		if (pg->index == pg->size) {
4898 			/* We should have allocated enough */
4899 			if (WARN_ON(!pg->next))
4900 				break;
4901 			pg = pg->next;
4902 		}
4903 
4904 		rec = &pg->records[pg->index++];
4905 		rec->ip = addr;
4906 	}
4907 
4908 	/* We should have used all pages */
4909 	WARN_ON(pg->next);
4910 
4911 	/* Assign the last page to ftrace_pages */
4912 	ftrace_pages = pg;
4913 
4914 	/*
4915 	 * We only need to disable interrupts on start up
4916 	 * because we are modifying code that an interrupt
4917 	 * may execute, and the modification is not atomic.
4918 	 * But for modules, nothing runs the code we modify
4919 	 * until we are finished with it, and there's no
4920 	 * reason to cause large interrupt latencies while we do it.
4921 	 */
4922 	if (!mod)
4923 		local_irq_save(flags);
4924 	ftrace_update_code(mod, start_pg);
4925 	if (!mod)
4926 		local_irq_restore(flags);
4927 	ret = 0;
4928  out:
4929 	mutex_unlock(&ftrace_lock);
4930 
4931 	return ret;
4932 }
4933 
4934 #ifdef CONFIG_MODULES
4935 
4936 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4937 
4938 static int referenced_filters(struct dyn_ftrace *rec)
4939 {
4940 	struct ftrace_ops *ops;
4941 	int cnt = 0;
4942 
4943 	for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
4944 		if (ops_references_rec(ops, rec))
4945 		    cnt++;
4946 	}
4947 
4948 	return cnt;
4949 }
4950 
4951 void ftrace_release_mod(struct module *mod)
4952 {
4953 	struct dyn_ftrace *rec;
4954 	struct ftrace_page **last_pg;
4955 	struct ftrace_page *pg;
4956 	int order;
4957 
4958 	mutex_lock(&ftrace_lock);
4959 
4960 	if (ftrace_disabled)
4961 		goto out_unlock;
4962 
4963 	/*
4964 	 * Each module has its own ftrace_pages, remove
4965 	 * them from the list.
4966 	 */
4967 	last_pg = &ftrace_pages_start;
4968 	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4969 		rec = &pg->records[0];
4970 		if (within_module_core(rec->ip, mod)) {
4971 			/*
4972 			 * As core pages are first, the first
4973 			 * page should never be a module page.
4974 			 */
4975 			if (WARN_ON(pg == ftrace_pages_start))
4976 				goto out_unlock;
4977 
4978 			/* Check if we are deleting the last page */
4979 			if (pg == ftrace_pages)
4980 				ftrace_pages = next_to_ftrace_page(last_pg);
4981 
4982 			*last_pg = pg->next;
4983 			order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4984 			free_pages((unsigned long)pg->records, order);
4985 			kfree(pg);
4986 		} else
4987 			last_pg = &pg->next;
4988 	}
4989  out_unlock:
4990 	mutex_unlock(&ftrace_lock);
4991 }
4992 
4993 void ftrace_module_enable(struct module *mod)
4994 {
4995 	struct dyn_ftrace *rec;
4996 	struct ftrace_page *pg;
4997 
4998 	mutex_lock(&ftrace_lock);
4999 
5000 	if (ftrace_disabled)
5001 		goto out_unlock;
5002 
5003 	/*
5004 	 * If the tracing is enabled, go ahead and enable the record.
5005 	 *
5006 	 * The reason not to enable the record immediatelly is the
5007 	 * inherent check of ftrace_make_nop/ftrace_make_call for
5008 	 * correct previous instructions.  Making first the NOP
5009 	 * conversion puts the module to the correct state, thus
5010 	 * passing the ftrace_make_call check.
5011 	 *
5012 	 * We also delay this to after the module code already set the
5013 	 * text to read-only, as we now need to set it back to read-write
5014 	 * so that we can modify the text.
5015 	 */
5016 	if (ftrace_start_up)
5017 		ftrace_arch_code_modify_prepare();
5018 
5019 	do_for_each_ftrace_rec(pg, rec) {
5020 		int cnt;
5021 		/*
5022 		 * do_for_each_ftrace_rec() is a double loop.
5023 		 * module text shares the pg. If a record is
5024 		 * not part of this module, then skip this pg,
5025 		 * which the "break" will do.
5026 		 */
5027 		if (!within_module_core(rec->ip, mod))
5028 			break;
5029 
5030 		cnt = 0;
5031 
5032 		/*
5033 		 * When adding a module, we need to check if tracers are
5034 		 * currently enabled and if they are, and can trace this record,
5035 		 * we need to enable the module functions as well as update the
5036 		 * reference counts for those function records.
5037 		 */
5038 		if (ftrace_start_up)
5039 			cnt += referenced_filters(rec);
5040 
5041 		/* This clears FTRACE_FL_DISABLED */
5042 		rec->flags = cnt;
5043 
5044 		if (ftrace_start_up && cnt) {
5045 			int failed = __ftrace_replace_code(rec, 1);
5046 			if (failed) {
5047 				ftrace_bug(failed, rec);
5048 				goto out_loop;
5049 			}
5050 		}
5051 
5052 	} while_for_each_ftrace_rec();
5053 
5054  out_loop:
5055 	if (ftrace_start_up)
5056 		ftrace_arch_code_modify_post_process();
5057 
5058  out_unlock:
5059 	mutex_unlock(&ftrace_lock);
5060 }
5061 
5062 void ftrace_module_init(struct module *mod)
5063 {
5064 	if (ftrace_disabled || !mod->num_ftrace_callsites)
5065 		return;
5066 
5067 	ftrace_process_locs(mod, mod->ftrace_callsites,
5068 			    mod->ftrace_callsites + mod->num_ftrace_callsites);
5069 }
5070 #endif /* CONFIG_MODULES */
5071 
5072 void __init ftrace_init(void)
5073 {
5074 	extern unsigned long __start_mcount_loc[];
5075 	extern unsigned long __stop_mcount_loc[];
5076 	unsigned long count, flags;
5077 	int ret;
5078 
5079 	local_irq_save(flags);
5080 	ret = ftrace_dyn_arch_init();
5081 	local_irq_restore(flags);
5082 	if (ret)
5083 		goto failed;
5084 
5085 	count = __stop_mcount_loc - __start_mcount_loc;
5086 	if (!count) {
5087 		pr_info("ftrace: No functions to be traced?\n");
5088 		goto failed;
5089 	}
5090 
5091 	pr_info("ftrace: allocating %ld entries in %ld pages\n",
5092 		count, count / ENTRIES_PER_PAGE + 1);
5093 
5094 	last_ftrace_enabled = ftrace_enabled = 1;
5095 
5096 	ret = ftrace_process_locs(NULL,
5097 				  __start_mcount_loc,
5098 				  __stop_mcount_loc);
5099 
5100 	set_ftrace_early_filters();
5101 
5102 	return;
5103  failed:
5104 	ftrace_disabled = 1;
5105 }
5106 
5107 /* Do nothing if arch does not support this */
5108 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
5109 {
5110 }
5111 
5112 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5113 {
5114 
5115 /*
5116  * Currently there's no safe way to free a trampoline when the kernel
5117  * is configured with PREEMPT. That is because a task could be preempted
5118  * when it jumped to the trampoline, it may be preempted for a long time
5119  * depending on the system load, and currently there's no way to know
5120  * when it will be off the trampoline. If the trampoline is freed
5121  * too early, when the task runs again, it will be executing on freed
5122  * memory and crash.
5123  */
5124 #ifdef CONFIG_PREEMPT
5125 	/* Currently, only non dynamic ops can have a trampoline */
5126 	if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
5127 		return;
5128 #endif
5129 
5130 	arch_ftrace_update_trampoline(ops);
5131 }
5132 
5133 #else
5134 
5135 static struct ftrace_ops global_ops = {
5136 	.func			= ftrace_stub,
5137 	.flags			= FTRACE_OPS_FL_RECURSION_SAFE |
5138 				  FTRACE_OPS_FL_INITIALIZED |
5139 				  FTRACE_OPS_FL_PID,
5140 };
5141 
5142 static int __init ftrace_nodyn_init(void)
5143 {
5144 	ftrace_enabled = 1;
5145 	return 0;
5146 }
5147 core_initcall(ftrace_nodyn_init);
5148 
5149 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
5150 static inline void ftrace_startup_enable(int command) { }
5151 static inline void ftrace_startup_all(int command) { }
5152 /* Keep as macros so we do not need to define the commands */
5153 # define ftrace_startup(ops, command)					\
5154 	({								\
5155 		int ___ret = __register_ftrace_function(ops);		\
5156 		if (!___ret)						\
5157 			(ops)->flags |= FTRACE_OPS_FL_ENABLED;		\
5158 		___ret;							\
5159 	})
5160 # define ftrace_shutdown(ops, command)					\
5161 	({								\
5162 		int ___ret = __unregister_ftrace_function(ops);		\
5163 		if (!___ret)						\
5164 			(ops)->flags &= ~FTRACE_OPS_FL_ENABLED;		\
5165 		___ret;							\
5166 	})
5167 
5168 # define ftrace_startup_sysctl()	do { } while (0)
5169 # define ftrace_shutdown_sysctl()	do { } while (0)
5170 
5171 static inline int
5172 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
5173 {
5174 	return 1;
5175 }
5176 
5177 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5178 {
5179 }
5180 
5181 #endif /* CONFIG_DYNAMIC_FTRACE */
5182 
5183 __init void ftrace_init_global_array_ops(struct trace_array *tr)
5184 {
5185 	tr->ops = &global_ops;
5186 	tr->ops->private = tr;
5187 }
5188 
5189 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
5190 {
5191 	/* If we filter on pids, update to use the pid function */
5192 	if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
5193 		if (WARN_ON(tr->ops->func != ftrace_stub))
5194 			printk("ftrace ops had %pS for function\n",
5195 			       tr->ops->func);
5196 	}
5197 	tr->ops->func = func;
5198 	tr->ops->private = tr;
5199 }
5200 
5201 void ftrace_reset_array_ops(struct trace_array *tr)
5202 {
5203 	tr->ops->func = ftrace_stub;
5204 }
5205 
5206 static inline void
5207 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5208 		       struct ftrace_ops *ignored, struct pt_regs *regs)
5209 {
5210 	struct ftrace_ops *op;
5211 	int bit;
5212 
5213 	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5214 	if (bit < 0)
5215 		return;
5216 
5217 	/*
5218 	 * Some of the ops may be dynamically allocated,
5219 	 * they must be freed after a synchronize_sched().
5220 	 */
5221 	preempt_disable_notrace();
5222 
5223 	do_for_each_ftrace_op(op, ftrace_ops_list) {
5224 		/*
5225 		 * Check the following for each ops before calling their func:
5226 		 *  if RCU flag is set, then rcu_is_watching() must be true
5227 		 *  if PER_CPU is set, then ftrace_function_local_disable()
5228 		 *                          must be false
5229 		 *  Otherwise test if the ip matches the ops filter
5230 		 *
5231 		 * If any of the above fails then the op->func() is not executed.
5232 		 */
5233 		if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
5234 		    (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5235 		     !ftrace_function_local_disabled(op)) &&
5236 		    ftrace_ops_test(op, ip, regs)) {
5237 
5238 			if (FTRACE_WARN_ON(!op->func)) {
5239 				pr_warn("op=%p %pS\n", op, op);
5240 				goto out;
5241 			}
5242 			op->func(ip, parent_ip, op, regs);
5243 		}
5244 	} while_for_each_ftrace_op(op);
5245 out:
5246 	preempt_enable_notrace();
5247 	trace_clear_recursion(bit);
5248 }
5249 
5250 /*
5251  * Some archs only support passing ip and parent_ip. Even though
5252  * the list function ignores the op parameter, we do not want any
5253  * C side effects, where a function is called without the caller
5254  * sending a third parameter.
5255  * Archs are to support both the regs and ftrace_ops at the same time.
5256  * If they support ftrace_ops, it is assumed they support regs.
5257  * If call backs want to use regs, they must either check for regs
5258  * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
5259  * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
5260  * An architecture can pass partial regs with ftrace_ops and still
5261  * set the ARCH_SUPPORTS_FTRACE_OPS.
5262  */
5263 #if ARCH_SUPPORTS_FTRACE_OPS
5264 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5265 				 struct ftrace_ops *op, struct pt_regs *regs)
5266 {
5267 	__ftrace_ops_list_func(ip, parent_ip, NULL, regs);
5268 }
5269 #else
5270 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
5271 {
5272 	__ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
5273 }
5274 #endif
5275 
5276 /*
5277  * If there's only one function registered but it does not support
5278  * recursion, needs RCU protection and/or requires per cpu handling, then
5279  * this function will be called by the mcount trampoline.
5280  */
5281 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
5282 				   struct ftrace_ops *op, struct pt_regs *regs)
5283 {
5284 	int bit;
5285 
5286 	if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
5287 		return;
5288 
5289 	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5290 	if (bit < 0)
5291 		return;
5292 
5293 	preempt_disable_notrace();
5294 
5295 	if (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5296 	    !ftrace_function_local_disabled(op)) {
5297 		op->func(ip, parent_ip, op, regs);
5298 	}
5299 
5300 	preempt_enable_notrace();
5301 	trace_clear_recursion(bit);
5302 }
5303 
5304 /**
5305  * ftrace_ops_get_func - get the function a trampoline should call
5306  * @ops: the ops to get the function for
5307  *
5308  * Normally the mcount trampoline will call the ops->func, but there
5309  * are times that it should not. For example, if the ops does not
5310  * have its own recursion protection, then it should call the
5311  * ftrace_ops_recurs_func() instead.
5312  *
5313  * Returns the function that the trampoline should call for @ops.
5314  */
5315 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
5316 {
5317 	/*
5318 	 * If the function does not handle recursion, needs to be RCU safe,
5319 	 * or does per cpu logic, then we need to call the assist handler.
5320 	 */
5321 	if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
5322 	    ops->flags & (FTRACE_OPS_FL_RCU | FTRACE_OPS_FL_PER_CPU))
5323 		return ftrace_ops_assist_func;
5324 
5325 	return ops->func;
5326 }
5327 
5328 static void
5329 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
5330 		    struct task_struct *prev, struct task_struct *next)
5331 {
5332 	struct trace_array *tr = data;
5333 	struct trace_pid_list *pid_list;
5334 
5335 	pid_list = rcu_dereference_sched(tr->function_pids);
5336 
5337 	this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
5338 		       trace_ignore_this_task(pid_list, next));
5339 }
5340 
5341 static void clear_ftrace_pids(struct trace_array *tr)
5342 {
5343 	struct trace_pid_list *pid_list;
5344 	int cpu;
5345 
5346 	pid_list = rcu_dereference_protected(tr->function_pids,
5347 					     lockdep_is_held(&ftrace_lock));
5348 	if (!pid_list)
5349 		return;
5350 
5351 	unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
5352 
5353 	for_each_possible_cpu(cpu)
5354 		per_cpu_ptr(tr->trace_buffer.data, cpu)->ftrace_ignore_pid = false;
5355 
5356 	rcu_assign_pointer(tr->function_pids, NULL);
5357 
5358 	/* Wait till all users are no longer using pid filtering */
5359 	synchronize_sched();
5360 
5361 	trace_free_pid_list(pid_list);
5362 }
5363 
5364 static void ftrace_pid_reset(struct trace_array *tr)
5365 {
5366 	mutex_lock(&ftrace_lock);
5367 	clear_ftrace_pids(tr);
5368 
5369 	ftrace_update_pid_func();
5370 	ftrace_startup_all(0);
5371 
5372 	mutex_unlock(&ftrace_lock);
5373 }
5374 
5375 /* Greater than any max PID */
5376 #define FTRACE_NO_PIDS		(void *)(PID_MAX_LIMIT + 1)
5377 
5378 static void *fpid_start(struct seq_file *m, loff_t *pos)
5379 	__acquires(RCU)
5380 {
5381 	struct trace_pid_list *pid_list;
5382 	struct trace_array *tr = m->private;
5383 
5384 	mutex_lock(&ftrace_lock);
5385 	rcu_read_lock_sched();
5386 
5387 	pid_list = rcu_dereference_sched(tr->function_pids);
5388 
5389 	if (!pid_list)
5390 		return !(*pos) ? FTRACE_NO_PIDS : NULL;
5391 
5392 	return trace_pid_start(pid_list, pos);
5393 }
5394 
5395 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
5396 {
5397 	struct trace_array *tr = m->private;
5398 	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
5399 
5400 	if (v == FTRACE_NO_PIDS)
5401 		return NULL;
5402 
5403 	return trace_pid_next(pid_list, v, pos);
5404 }
5405 
5406 static void fpid_stop(struct seq_file *m, void *p)
5407 	__releases(RCU)
5408 {
5409 	rcu_read_unlock_sched();
5410 	mutex_unlock(&ftrace_lock);
5411 }
5412 
5413 static int fpid_show(struct seq_file *m, void *v)
5414 {
5415 	if (v == FTRACE_NO_PIDS) {
5416 		seq_puts(m, "no pid\n");
5417 		return 0;
5418 	}
5419 
5420 	return trace_pid_show(m, v);
5421 }
5422 
5423 static const struct seq_operations ftrace_pid_sops = {
5424 	.start = fpid_start,
5425 	.next = fpid_next,
5426 	.stop = fpid_stop,
5427 	.show = fpid_show,
5428 };
5429 
5430 static int
5431 ftrace_pid_open(struct inode *inode, struct file *file)
5432 {
5433 	struct trace_array *tr = inode->i_private;
5434 	struct seq_file *m;
5435 	int ret = 0;
5436 
5437 	if (trace_array_get(tr) < 0)
5438 		return -ENODEV;
5439 
5440 	if ((file->f_mode & FMODE_WRITE) &&
5441 	    (file->f_flags & O_TRUNC))
5442 		ftrace_pid_reset(tr);
5443 
5444 	ret = seq_open(file, &ftrace_pid_sops);
5445 	if (ret < 0) {
5446 		trace_array_put(tr);
5447 	} else {
5448 		m = file->private_data;
5449 		/* copy tr over to seq ops */
5450 		m->private = tr;
5451 	}
5452 
5453 	return ret;
5454 }
5455 
5456 static void ignore_task_cpu(void *data)
5457 {
5458 	struct trace_array *tr = data;
5459 	struct trace_pid_list *pid_list;
5460 
5461 	/*
5462 	 * This function is called by on_each_cpu() while the
5463 	 * event_mutex is held.
5464 	 */
5465 	pid_list = rcu_dereference_protected(tr->function_pids,
5466 					     mutex_is_locked(&ftrace_lock));
5467 
5468 	this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
5469 		       trace_ignore_this_task(pid_list, current));
5470 }
5471 
5472 static ssize_t
5473 ftrace_pid_write(struct file *filp, const char __user *ubuf,
5474 		   size_t cnt, loff_t *ppos)
5475 {
5476 	struct seq_file *m = filp->private_data;
5477 	struct trace_array *tr = m->private;
5478 	struct trace_pid_list *filtered_pids = NULL;
5479 	struct trace_pid_list *pid_list;
5480 	ssize_t ret;
5481 
5482 	if (!cnt)
5483 		return 0;
5484 
5485 	mutex_lock(&ftrace_lock);
5486 
5487 	filtered_pids = rcu_dereference_protected(tr->function_pids,
5488 					     lockdep_is_held(&ftrace_lock));
5489 
5490 	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
5491 	if (ret < 0)
5492 		goto out;
5493 
5494 	rcu_assign_pointer(tr->function_pids, pid_list);
5495 
5496 	if (filtered_pids) {
5497 		synchronize_sched();
5498 		trace_free_pid_list(filtered_pids);
5499 	} else if (pid_list) {
5500 		/* Register a probe to set whether to ignore the tracing of a task */
5501 		register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
5502 	}
5503 
5504 	/*
5505 	 * Ignoring of pids is done at task switch. But we have to
5506 	 * check for those tasks that are currently running.
5507 	 * Always do this in case a pid was appended or removed.
5508 	 */
5509 	on_each_cpu(ignore_task_cpu, tr, 1);
5510 
5511 	ftrace_update_pid_func();
5512 	ftrace_startup_all(0);
5513  out:
5514 	mutex_unlock(&ftrace_lock);
5515 
5516 	if (ret > 0)
5517 		*ppos += ret;
5518 
5519 	return ret;
5520 }
5521 
5522 static int
5523 ftrace_pid_release(struct inode *inode, struct file *file)
5524 {
5525 	struct trace_array *tr = inode->i_private;
5526 
5527 	trace_array_put(tr);
5528 
5529 	return seq_release(inode, file);
5530 }
5531 
5532 static const struct file_operations ftrace_pid_fops = {
5533 	.open		= ftrace_pid_open,
5534 	.write		= ftrace_pid_write,
5535 	.read		= seq_read,
5536 	.llseek		= tracing_lseek,
5537 	.release	= ftrace_pid_release,
5538 };
5539 
5540 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
5541 {
5542 	trace_create_file("set_ftrace_pid", 0644, d_tracer,
5543 			    tr, &ftrace_pid_fops);
5544 }
5545 
5546 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
5547 					 struct dentry *d_tracer)
5548 {
5549 	/* Only the top level directory has the dyn_tracefs and profile */
5550 	WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
5551 
5552 	ftrace_init_dyn_tracefs(d_tracer);
5553 	ftrace_profile_tracefs(d_tracer);
5554 }
5555 
5556 /**
5557  * ftrace_kill - kill ftrace
5558  *
5559  * This function should be used by panic code. It stops ftrace
5560  * but in a not so nice way. If you need to simply kill ftrace
5561  * from a non-atomic section, use ftrace_kill.
5562  */
5563 void ftrace_kill(void)
5564 {
5565 	ftrace_disabled = 1;
5566 	ftrace_enabled = 0;
5567 	clear_ftrace_function();
5568 }
5569 
5570 /**
5571  * Test if ftrace is dead or not.
5572  */
5573 int ftrace_is_dead(void)
5574 {
5575 	return ftrace_disabled;
5576 }
5577 
5578 /**
5579  * register_ftrace_function - register a function for profiling
5580  * @ops - ops structure that holds the function for profiling.
5581  *
5582  * Register a function to be called by all functions in the
5583  * kernel.
5584  *
5585  * Note: @ops->func and all the functions it calls must be labeled
5586  *       with "notrace", otherwise it will go into a
5587  *       recursive loop.
5588  */
5589 int register_ftrace_function(struct ftrace_ops *ops)
5590 {
5591 	int ret = -1;
5592 
5593 	ftrace_ops_init(ops);
5594 
5595 	mutex_lock(&ftrace_lock);
5596 
5597 	ret = ftrace_startup(ops, 0);
5598 
5599 	mutex_unlock(&ftrace_lock);
5600 
5601 	return ret;
5602 }
5603 EXPORT_SYMBOL_GPL(register_ftrace_function);
5604 
5605 /**
5606  * unregister_ftrace_function - unregister a function for profiling.
5607  * @ops - ops structure that holds the function to unregister
5608  *
5609  * Unregister a function that was added to be called by ftrace profiling.
5610  */
5611 int unregister_ftrace_function(struct ftrace_ops *ops)
5612 {
5613 	int ret;
5614 
5615 	mutex_lock(&ftrace_lock);
5616 	ret = ftrace_shutdown(ops, 0);
5617 	mutex_unlock(&ftrace_lock);
5618 
5619 	return ret;
5620 }
5621 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
5622 
5623 int
5624 ftrace_enable_sysctl(struct ctl_table *table, int write,
5625 		     void __user *buffer, size_t *lenp,
5626 		     loff_t *ppos)
5627 {
5628 	int ret = -ENODEV;
5629 
5630 	mutex_lock(&ftrace_lock);
5631 
5632 	if (unlikely(ftrace_disabled))
5633 		goto out;
5634 
5635 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
5636 
5637 	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
5638 		goto out;
5639 
5640 	last_ftrace_enabled = !!ftrace_enabled;
5641 
5642 	if (ftrace_enabled) {
5643 
5644 		/* we are starting ftrace again */
5645 		if (ftrace_ops_list != &ftrace_list_end)
5646 			update_ftrace_function();
5647 
5648 		ftrace_startup_sysctl();
5649 
5650 	} else {
5651 		/* stopping ftrace calls (just send to ftrace_stub) */
5652 		ftrace_trace_function = ftrace_stub;
5653 
5654 		ftrace_shutdown_sysctl();
5655 	}
5656 
5657  out:
5658 	mutex_unlock(&ftrace_lock);
5659 	return ret;
5660 }
5661 
5662 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5663 
5664 static struct ftrace_ops graph_ops = {
5665 	.func			= ftrace_stub,
5666 	.flags			= FTRACE_OPS_FL_RECURSION_SAFE |
5667 				   FTRACE_OPS_FL_INITIALIZED |
5668 				   FTRACE_OPS_FL_PID |
5669 				   FTRACE_OPS_FL_STUB,
5670 #ifdef FTRACE_GRAPH_TRAMP_ADDR
5671 	.trampoline		= FTRACE_GRAPH_TRAMP_ADDR,
5672 	/* trampoline_size is only needed for dynamically allocated tramps */
5673 #endif
5674 	ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
5675 };
5676 
5677 void ftrace_graph_sleep_time_control(bool enable)
5678 {
5679 	fgraph_sleep_time = enable;
5680 }
5681 
5682 void ftrace_graph_graph_time_control(bool enable)
5683 {
5684 	fgraph_graph_time = enable;
5685 }
5686 
5687 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
5688 {
5689 	return 0;
5690 }
5691 
5692 /* The callbacks that hook a function */
5693 trace_func_graph_ret_t ftrace_graph_return =
5694 			(trace_func_graph_ret_t)ftrace_stub;
5695 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
5696 static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
5697 
5698 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
5699 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
5700 {
5701 	int i;
5702 	int ret = 0;
5703 	int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
5704 	struct task_struct *g, *t;
5705 
5706 	for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
5707 		ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
5708 					* sizeof(struct ftrace_ret_stack),
5709 					GFP_KERNEL);
5710 		if (!ret_stack_list[i]) {
5711 			start = 0;
5712 			end = i;
5713 			ret = -ENOMEM;
5714 			goto free;
5715 		}
5716 	}
5717 
5718 	read_lock(&tasklist_lock);
5719 	do_each_thread(g, t) {
5720 		if (start == end) {
5721 			ret = -EAGAIN;
5722 			goto unlock;
5723 		}
5724 
5725 		if (t->ret_stack == NULL) {
5726 			atomic_set(&t->tracing_graph_pause, 0);
5727 			atomic_set(&t->trace_overrun, 0);
5728 			t->curr_ret_stack = -1;
5729 			/* Make sure the tasks see the -1 first: */
5730 			smp_wmb();
5731 			t->ret_stack = ret_stack_list[start++];
5732 		}
5733 	} while_each_thread(g, t);
5734 
5735 unlock:
5736 	read_unlock(&tasklist_lock);
5737 free:
5738 	for (i = start; i < end; i++)
5739 		kfree(ret_stack_list[i]);
5740 	return ret;
5741 }
5742 
5743 static void
5744 ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
5745 			struct task_struct *prev, struct task_struct *next)
5746 {
5747 	unsigned long long timestamp;
5748 	int index;
5749 
5750 	/*
5751 	 * Does the user want to count the time a function was asleep.
5752 	 * If so, do not update the time stamps.
5753 	 */
5754 	if (fgraph_sleep_time)
5755 		return;
5756 
5757 	timestamp = trace_clock_local();
5758 
5759 	prev->ftrace_timestamp = timestamp;
5760 
5761 	/* only process tasks that we timestamped */
5762 	if (!next->ftrace_timestamp)
5763 		return;
5764 
5765 	/*
5766 	 * Update all the counters in next to make up for the
5767 	 * time next was sleeping.
5768 	 */
5769 	timestamp -= next->ftrace_timestamp;
5770 
5771 	for (index = next->curr_ret_stack; index >= 0; index--)
5772 		next->ret_stack[index].calltime += timestamp;
5773 }
5774 
5775 /* Allocate a return stack for each task */
5776 static int start_graph_tracing(void)
5777 {
5778 	struct ftrace_ret_stack **ret_stack_list;
5779 	int ret, cpu;
5780 
5781 	ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
5782 				sizeof(struct ftrace_ret_stack *),
5783 				GFP_KERNEL);
5784 
5785 	if (!ret_stack_list)
5786 		return -ENOMEM;
5787 
5788 	/* The cpu_boot init_task->ret_stack will never be freed */
5789 	for_each_online_cpu(cpu) {
5790 		if (!idle_task(cpu)->ret_stack)
5791 			ftrace_graph_init_idle_task(idle_task(cpu), cpu);
5792 	}
5793 
5794 	do {
5795 		ret = alloc_retstack_tasklist(ret_stack_list);
5796 	} while (ret == -EAGAIN);
5797 
5798 	if (!ret) {
5799 		ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5800 		if (ret)
5801 			pr_info("ftrace_graph: Couldn't activate tracepoint"
5802 				" probe to kernel_sched_switch\n");
5803 	}
5804 
5805 	kfree(ret_stack_list);
5806 	return ret;
5807 }
5808 
5809 /*
5810  * Hibernation protection.
5811  * The state of the current task is too much unstable during
5812  * suspend/restore to disk. We want to protect against that.
5813  */
5814 static int
5815 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
5816 							void *unused)
5817 {
5818 	switch (state) {
5819 	case PM_HIBERNATION_PREPARE:
5820 		pause_graph_tracing();
5821 		break;
5822 
5823 	case PM_POST_HIBERNATION:
5824 		unpause_graph_tracing();
5825 		break;
5826 	}
5827 	return NOTIFY_DONE;
5828 }
5829 
5830 static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
5831 {
5832 	if (!ftrace_ops_test(&global_ops, trace->func, NULL))
5833 		return 0;
5834 	return __ftrace_graph_entry(trace);
5835 }
5836 
5837 /*
5838  * The function graph tracer should only trace the functions defined
5839  * by set_ftrace_filter and set_ftrace_notrace. If another function
5840  * tracer ops is registered, the graph tracer requires testing the
5841  * function against the global ops, and not just trace any function
5842  * that any ftrace_ops registered.
5843  */
5844 static void update_function_graph_func(void)
5845 {
5846 	struct ftrace_ops *op;
5847 	bool do_test = false;
5848 
5849 	/*
5850 	 * The graph and global ops share the same set of functions
5851 	 * to test. If any other ops is on the list, then
5852 	 * the graph tracing needs to test if its the function
5853 	 * it should call.
5854 	 */
5855 	do_for_each_ftrace_op(op, ftrace_ops_list) {
5856 		if (op != &global_ops && op != &graph_ops &&
5857 		    op != &ftrace_list_end) {
5858 			do_test = true;
5859 			/* in double loop, break out with goto */
5860 			goto out;
5861 		}
5862 	} while_for_each_ftrace_op(op);
5863  out:
5864 	if (do_test)
5865 		ftrace_graph_entry = ftrace_graph_entry_test;
5866 	else
5867 		ftrace_graph_entry = __ftrace_graph_entry;
5868 }
5869 
5870 static struct notifier_block ftrace_suspend_notifier = {
5871 	.notifier_call = ftrace_suspend_notifier_call,
5872 };
5873 
5874 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
5875 			trace_func_graph_ent_t entryfunc)
5876 {
5877 	int ret = 0;
5878 
5879 	mutex_lock(&ftrace_lock);
5880 
5881 	/* we currently allow only one tracer registered at a time */
5882 	if (ftrace_graph_active) {
5883 		ret = -EBUSY;
5884 		goto out;
5885 	}
5886 
5887 	register_pm_notifier(&ftrace_suspend_notifier);
5888 
5889 	ftrace_graph_active++;
5890 	ret = start_graph_tracing();
5891 	if (ret) {
5892 		ftrace_graph_active--;
5893 		goto out;
5894 	}
5895 
5896 	ftrace_graph_return = retfunc;
5897 
5898 	/*
5899 	 * Update the indirect function to the entryfunc, and the
5900 	 * function that gets called to the entry_test first. Then
5901 	 * call the update fgraph entry function to determine if
5902 	 * the entryfunc should be called directly or not.
5903 	 */
5904 	__ftrace_graph_entry = entryfunc;
5905 	ftrace_graph_entry = ftrace_graph_entry_test;
5906 	update_function_graph_func();
5907 
5908 	ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
5909 out:
5910 	mutex_unlock(&ftrace_lock);
5911 	return ret;
5912 }
5913 
5914 void unregister_ftrace_graph(void)
5915 {
5916 	mutex_lock(&ftrace_lock);
5917 
5918 	if (unlikely(!ftrace_graph_active))
5919 		goto out;
5920 
5921 	ftrace_graph_active--;
5922 	ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
5923 	ftrace_graph_entry = ftrace_graph_entry_stub;
5924 	__ftrace_graph_entry = ftrace_graph_entry_stub;
5925 	ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
5926 	unregister_pm_notifier(&ftrace_suspend_notifier);
5927 	unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5928 
5929 #ifdef CONFIG_DYNAMIC_FTRACE
5930 	/*
5931 	 * Function graph does not allocate the trampoline, but
5932 	 * other global_ops do. We need to reset the ALLOC_TRAMP flag
5933 	 * if one was used.
5934 	 */
5935 	global_ops.trampoline = save_global_trampoline;
5936 	if (save_global_flags & FTRACE_OPS_FL_ALLOC_TRAMP)
5937 		global_ops.flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
5938 #endif
5939 
5940  out:
5941 	mutex_unlock(&ftrace_lock);
5942 }
5943 
5944 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
5945 
5946 static void
5947 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
5948 {
5949 	atomic_set(&t->tracing_graph_pause, 0);
5950 	atomic_set(&t->trace_overrun, 0);
5951 	t->ftrace_timestamp = 0;
5952 	/* make curr_ret_stack visible before we add the ret_stack */
5953 	smp_wmb();
5954 	t->ret_stack = ret_stack;
5955 }
5956 
5957 /*
5958  * Allocate a return stack for the idle task. May be the first
5959  * time through, or it may be done by CPU hotplug online.
5960  */
5961 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
5962 {
5963 	t->curr_ret_stack = -1;
5964 	/*
5965 	 * The idle task has no parent, it either has its own
5966 	 * stack or no stack at all.
5967 	 */
5968 	if (t->ret_stack)
5969 		WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
5970 
5971 	if (ftrace_graph_active) {
5972 		struct ftrace_ret_stack *ret_stack;
5973 
5974 		ret_stack = per_cpu(idle_ret_stack, cpu);
5975 		if (!ret_stack) {
5976 			ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5977 					    * sizeof(struct ftrace_ret_stack),
5978 					    GFP_KERNEL);
5979 			if (!ret_stack)
5980 				return;
5981 			per_cpu(idle_ret_stack, cpu) = ret_stack;
5982 		}
5983 		graph_init_task(t, ret_stack);
5984 	}
5985 }
5986 
5987 /* Allocate a return stack for newly created task */
5988 void ftrace_graph_init_task(struct task_struct *t)
5989 {
5990 	/* Make sure we do not use the parent ret_stack */
5991 	t->ret_stack = NULL;
5992 	t->curr_ret_stack = -1;
5993 
5994 	if (ftrace_graph_active) {
5995 		struct ftrace_ret_stack *ret_stack;
5996 
5997 		ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5998 				* sizeof(struct ftrace_ret_stack),
5999 				GFP_KERNEL);
6000 		if (!ret_stack)
6001 			return;
6002 		graph_init_task(t, ret_stack);
6003 	}
6004 }
6005 
6006 void ftrace_graph_exit_task(struct task_struct *t)
6007 {
6008 	struct ftrace_ret_stack	*ret_stack = t->ret_stack;
6009 
6010 	t->ret_stack = NULL;
6011 	/* NULL must become visible to IRQs before we free it: */
6012 	barrier();
6013 
6014 	kfree(ret_stack);
6015 }
6016 #endif
6017