Loading...
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/sched/task.h>
19#include <linux/kallsyms.h>
20#include <linux/seq_file.h>
21#include <linux/suspend.h>
22#include <linux/tracefs.h>
23#include <linux/hardirq.h>
24#include <linux/kthread.h>
25#include <linux/uaccess.h>
26#include <linux/bsearch.h>
27#include <linux/module.h>
28#include <linux/ftrace.h>
29#include <linux/sysctl.h>
30#include <linux/slab.h>
31#include <linux/ctype.h>
32#include <linux/sort.h>
33#include <linux/list.h>
34#include <linux/hash.h>
35#include <linux/rcupdate.h>
36
37#include <trace/events/sched.h>
38
39#include <asm/sections.h>
40#include <asm/setup.h>
41
42#include "trace_output.h"
43#include "trace_stat.h"
44
45#define FTRACE_WARN_ON(cond) \
46 ({ \
47 int ___r = cond; \
48 if (WARN_ON(___r)) \
49 ftrace_kill(); \
50 ___r; \
51 })
52
53#define FTRACE_WARN_ON_ONCE(cond) \
54 ({ \
55 int ___r = cond; \
56 if (WARN_ON_ONCE(___r)) \
57 ftrace_kill(); \
58 ___r; \
59 })
60
61/* hash bits for specific function selection */
62#define FTRACE_HASH_BITS 7
63#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
64#define FTRACE_HASH_DEFAULT_BITS 10
65#define FTRACE_HASH_MAX_BITS 12
66
67#ifdef CONFIG_DYNAMIC_FTRACE
68#define INIT_OPS_HASH(opsname) \
69 .func_hash = &opsname.local_hash, \
70 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
71#define ASSIGN_OPS_HASH(opsname, val) \
72 .func_hash = val, \
73 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
74#else
75#define INIT_OPS_HASH(opsname)
76#define ASSIGN_OPS_HASH(opsname, val)
77#endif
78
79static struct ftrace_ops ftrace_list_end __read_mostly = {
80 .func = ftrace_stub,
81 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
82 INIT_OPS_HASH(ftrace_list_end)
83};
84
85/* ftrace_enabled is a method to turn ftrace on or off */
86int ftrace_enabled __read_mostly;
87static int last_ftrace_enabled;
88
89/* Current function tracing op */
90struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
91/* What to set function_trace_op to */
92static struct ftrace_ops *set_function_trace_op;
93
94static bool ftrace_pids_enabled(struct ftrace_ops *ops)
95{
96 struct trace_array *tr;
97
98 if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
99 return false;
100
101 tr = ops->private;
102
103 return tr->function_pids != NULL;
104}
105
106static void ftrace_update_trampoline(struct ftrace_ops *ops);
107
108/*
109 * ftrace_disabled is set when an anomaly is discovered.
110 * ftrace_disabled is much stronger than ftrace_enabled.
111 */
112static int ftrace_disabled __read_mostly;
113
114static DEFINE_MUTEX(ftrace_lock);
115
116static struct ftrace_ops __rcu *ftrace_ops_list __read_mostly = &ftrace_list_end;
117ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
118static struct ftrace_ops global_ops;
119
120#if ARCH_SUPPORTS_FTRACE_OPS
121static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
122 struct ftrace_ops *op, struct pt_regs *regs);
123#else
124/* See comment below, where ftrace_ops_list_func is defined */
125static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
126#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
127#endif
128
129/*
130 * Traverse the ftrace_global_list, invoking all entries. The reason that we
131 * can use rcu_dereference_raw_notrace() is that elements removed from this list
132 * are simply leaked, so there is no need to interact with a grace-period
133 * mechanism. The rcu_dereference_raw_notrace() calls are needed to handle
134 * concurrent insertions into the ftrace_global_list.
135 *
136 * Silly Alpha and silly pointer-speculation compiler optimizations!
137 */
138#define do_for_each_ftrace_op(op, list) \
139 op = rcu_dereference_raw_notrace(list); \
140 do
141
142/*
143 * Optimized for just a single item in the list (as that is the normal case).
144 */
145#define while_for_each_ftrace_op(op) \
146 while (likely(op = rcu_dereference_raw_notrace((op)->next)) && \
147 unlikely((op) != &ftrace_list_end))
148
149static inline void ftrace_ops_init(struct ftrace_ops *ops)
150{
151#ifdef CONFIG_DYNAMIC_FTRACE
152 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
153 mutex_init(&ops->local_hash.regex_lock);
154 ops->func_hash = &ops->local_hash;
155 ops->flags |= FTRACE_OPS_FL_INITIALIZED;
156 }
157#endif
158}
159
160/**
161 * ftrace_nr_registered_ops - return number of ops registered
162 *
163 * Returns the number of ftrace_ops registered and tracing functions
164 */
165int ftrace_nr_registered_ops(void)
166{
167 struct ftrace_ops *ops;
168 int cnt = 0;
169
170 mutex_lock(&ftrace_lock);
171
172 for (ops = rcu_dereference_protected(ftrace_ops_list,
173 lockdep_is_held(&ftrace_lock));
174 ops != &ftrace_list_end;
175 ops = rcu_dereference_protected(ops->next,
176 lockdep_is_held(&ftrace_lock)))
177 cnt++;
178
179 mutex_unlock(&ftrace_lock);
180
181 return cnt;
182}
183
184static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
185 struct ftrace_ops *op, struct pt_regs *regs)
186{
187 struct trace_array *tr = op->private;
188
189 if (tr && this_cpu_read(tr->trace_buffer.data->ftrace_ignore_pid))
190 return;
191
192 op->saved_func(ip, parent_ip, op, regs);
193}
194
195/**
196 * clear_ftrace_function - reset the ftrace function
197 *
198 * This NULLs the ftrace function and in essence stops
199 * tracing. There may be lag
200 */
201void clear_ftrace_function(void)
202{
203 ftrace_trace_function = ftrace_stub;
204}
205
206static void ftrace_sync(struct work_struct *work)
207{
208 /*
209 * This function is just a stub to implement a hard force
210 * of synchronize_sched(). This requires synchronizing
211 * tasks even in userspace and idle.
212 *
213 * Yes, function tracing is rude.
214 */
215}
216
217static void ftrace_sync_ipi(void *data)
218{
219 /* Probably not needed, but do it anyway */
220 smp_rmb();
221}
222
223#ifdef CONFIG_FUNCTION_GRAPH_TRACER
224static void update_function_graph_func(void);
225
226/* Both enabled by default (can be cleared by function_graph tracer flags */
227static bool fgraph_sleep_time = true;
228static bool fgraph_graph_time = true;
229
230#else
231static inline void update_function_graph_func(void) { }
232#endif
233
234
235static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
236{
237 /*
238 * If this is a dynamic, RCU, or per CPU ops, or we force list func,
239 * then it needs to call the list anyway.
240 */
241 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_RCU) ||
242 FTRACE_FORCE_LIST_FUNC)
243 return ftrace_ops_list_func;
244
245 return ftrace_ops_get_func(ops);
246}
247
248static void update_ftrace_function(void)
249{
250 ftrace_func_t func;
251
252 /*
253 * Prepare the ftrace_ops that the arch callback will use.
254 * If there's only one ftrace_ops registered, the ftrace_ops_list
255 * will point to the ops we want.
256 */
257 set_function_trace_op = rcu_dereference_protected(ftrace_ops_list,
258 lockdep_is_held(&ftrace_lock));
259
260 /* If there's no ftrace_ops registered, just call the stub function */
261 if (set_function_trace_op == &ftrace_list_end) {
262 func = ftrace_stub;
263
264 /*
265 * If we are at the end of the list and this ops is
266 * recursion safe and not dynamic and the arch supports passing ops,
267 * then have the mcount trampoline call the function directly.
268 */
269 } else if (rcu_dereference_protected(ftrace_ops_list->next,
270 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
271 func = ftrace_ops_get_list_func(ftrace_ops_list);
272
273 } else {
274 /* Just use the default ftrace_ops */
275 set_function_trace_op = &ftrace_list_end;
276 func = ftrace_ops_list_func;
277 }
278
279 update_function_graph_func();
280
281 /* If there's no change, then do nothing more here */
282 if (ftrace_trace_function == func)
283 return;
284
285 /*
286 * If we are using the list function, it doesn't care
287 * about the function_trace_ops.
288 */
289 if (func == ftrace_ops_list_func) {
290 ftrace_trace_function = func;
291 /*
292 * Don't even bother setting function_trace_ops,
293 * it would be racy to do so anyway.
294 */
295 return;
296 }
297
298#ifndef CONFIG_DYNAMIC_FTRACE
299 /*
300 * For static tracing, we need to be a bit more careful.
301 * The function change takes affect immediately. Thus,
302 * we need to coorditate the setting of the function_trace_ops
303 * with the setting of the ftrace_trace_function.
304 *
305 * Set the function to the list ops, which will call the
306 * function we want, albeit indirectly, but it handles the
307 * ftrace_ops and doesn't depend on function_trace_op.
308 */
309 ftrace_trace_function = ftrace_ops_list_func;
310 /*
311 * Make sure all CPUs see this. Yes this is slow, but static
312 * tracing is slow and nasty to have enabled.
313 */
314 schedule_on_each_cpu(ftrace_sync);
315 /* Now all cpus are using the list ops. */
316 function_trace_op = set_function_trace_op;
317 /* Make sure the function_trace_op is visible on all CPUs */
318 smp_wmb();
319 /* Nasty way to force a rmb on all cpus */
320 smp_call_function(ftrace_sync_ipi, NULL, 1);
321 /* OK, we are all set to update the ftrace_trace_function now! */
322#endif /* !CONFIG_DYNAMIC_FTRACE */
323
324 ftrace_trace_function = func;
325}
326
327int using_ftrace_ops_list_func(void)
328{
329 return ftrace_trace_function == ftrace_ops_list_func;
330}
331
332static void add_ftrace_ops(struct ftrace_ops __rcu **list,
333 struct ftrace_ops *ops)
334{
335 rcu_assign_pointer(ops->next, *list);
336
337 /*
338 * We are entering ops into the list but another
339 * CPU might be walking that list. We need to make sure
340 * the ops->next pointer is valid before another CPU sees
341 * the ops pointer included into the list.
342 */
343 rcu_assign_pointer(*list, ops);
344}
345
346static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
347 struct ftrace_ops *ops)
348{
349 struct ftrace_ops **p;
350
351 /*
352 * If we are removing the last function, then simply point
353 * to the ftrace_stub.
354 */
355 if (rcu_dereference_protected(*list,
356 lockdep_is_held(&ftrace_lock)) == ops &&
357 rcu_dereference_protected(ops->next,
358 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
359 *list = &ftrace_list_end;
360 return 0;
361 }
362
363 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
364 if (*p == ops)
365 break;
366
367 if (*p != ops)
368 return -1;
369
370 *p = (*p)->next;
371 return 0;
372}
373
374static void ftrace_update_trampoline(struct ftrace_ops *ops);
375
376static int __register_ftrace_function(struct ftrace_ops *ops)
377{
378 if (ops->flags & FTRACE_OPS_FL_DELETED)
379 return -EINVAL;
380
381 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
382 return -EBUSY;
383
384#ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
385 /*
386 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
387 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
388 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
389 */
390 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
391 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
392 return -EINVAL;
393
394 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
395 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
396#endif
397
398 if (!core_kernel_data((unsigned long)ops))
399 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
400
401 add_ftrace_ops(&ftrace_ops_list, ops);
402
403 /* Always save the function, and reset at unregistering */
404 ops->saved_func = ops->func;
405
406 if (ftrace_pids_enabled(ops))
407 ops->func = ftrace_pid_func;
408
409 ftrace_update_trampoline(ops);
410
411 if (ftrace_enabled)
412 update_ftrace_function();
413
414 return 0;
415}
416
417static int __unregister_ftrace_function(struct ftrace_ops *ops)
418{
419 int ret;
420
421 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
422 return -EBUSY;
423
424 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
425
426 if (ret < 0)
427 return ret;
428
429 if (ftrace_enabled)
430 update_ftrace_function();
431
432 ops->func = ops->saved_func;
433
434 return 0;
435}
436
437static void ftrace_update_pid_func(void)
438{
439 struct ftrace_ops *op;
440
441 /* Only do something if we are tracing something */
442 if (ftrace_trace_function == ftrace_stub)
443 return;
444
445 do_for_each_ftrace_op(op, ftrace_ops_list) {
446 if (op->flags & FTRACE_OPS_FL_PID) {
447 op->func = ftrace_pids_enabled(op) ?
448 ftrace_pid_func : op->saved_func;
449 ftrace_update_trampoline(op);
450 }
451 } while_for_each_ftrace_op(op);
452
453 update_ftrace_function();
454}
455
456#ifdef CONFIG_FUNCTION_PROFILER
457struct ftrace_profile {
458 struct hlist_node node;
459 unsigned long ip;
460 unsigned long counter;
461#ifdef CONFIG_FUNCTION_GRAPH_TRACER
462 unsigned long long time;
463 unsigned long long time_squared;
464#endif
465};
466
467struct ftrace_profile_page {
468 struct ftrace_profile_page *next;
469 unsigned long index;
470 struct ftrace_profile records[];
471};
472
473struct ftrace_profile_stat {
474 atomic_t disabled;
475 struct hlist_head *hash;
476 struct ftrace_profile_page *pages;
477 struct ftrace_profile_page *start;
478 struct tracer_stat stat;
479};
480
481#define PROFILE_RECORDS_SIZE \
482 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
483
484#define PROFILES_PER_PAGE \
485 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
486
487static int ftrace_profile_enabled __read_mostly;
488
489/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
490static DEFINE_MUTEX(ftrace_profile_lock);
491
492static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
493
494#define FTRACE_PROFILE_HASH_BITS 10
495#define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
496
497static void *
498function_stat_next(void *v, int idx)
499{
500 struct ftrace_profile *rec = v;
501 struct ftrace_profile_page *pg;
502
503 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
504
505 again:
506 if (idx != 0)
507 rec++;
508
509 if ((void *)rec >= (void *)&pg->records[pg->index]) {
510 pg = pg->next;
511 if (!pg)
512 return NULL;
513 rec = &pg->records[0];
514 if (!rec->counter)
515 goto again;
516 }
517
518 return rec;
519}
520
521static void *function_stat_start(struct tracer_stat *trace)
522{
523 struct ftrace_profile_stat *stat =
524 container_of(trace, struct ftrace_profile_stat, stat);
525
526 if (!stat || !stat->start)
527 return NULL;
528
529 return function_stat_next(&stat->start->records[0], 0);
530}
531
532#ifdef CONFIG_FUNCTION_GRAPH_TRACER
533/* function graph compares on total time */
534static int function_stat_cmp(void *p1, void *p2)
535{
536 struct ftrace_profile *a = p1;
537 struct ftrace_profile *b = p2;
538
539 if (a->time < b->time)
540 return -1;
541 if (a->time > b->time)
542 return 1;
543 else
544 return 0;
545}
546#else
547/* not function graph compares against hits */
548static int function_stat_cmp(void *p1, void *p2)
549{
550 struct ftrace_profile *a = p1;
551 struct ftrace_profile *b = p2;
552
553 if (a->counter < b->counter)
554 return -1;
555 if (a->counter > b->counter)
556 return 1;
557 else
558 return 0;
559}
560#endif
561
562static int function_stat_headers(struct seq_file *m)
563{
564#ifdef CONFIG_FUNCTION_GRAPH_TRACER
565 seq_puts(m, " Function "
566 "Hit Time Avg s^2\n"
567 " -------- "
568 "--- ---- --- ---\n");
569#else
570 seq_puts(m, " Function Hit\n"
571 " -------- ---\n");
572#endif
573 return 0;
574}
575
576static int function_stat_show(struct seq_file *m, void *v)
577{
578 struct ftrace_profile *rec = v;
579 char str[KSYM_SYMBOL_LEN];
580 int ret = 0;
581#ifdef CONFIG_FUNCTION_GRAPH_TRACER
582 static struct trace_seq s;
583 unsigned long long avg;
584 unsigned long long stddev;
585#endif
586 mutex_lock(&ftrace_profile_lock);
587
588 /* we raced with function_profile_reset() */
589 if (unlikely(rec->counter == 0)) {
590 ret = -EBUSY;
591 goto out;
592 }
593
594#ifdef CONFIG_FUNCTION_GRAPH_TRACER
595 avg = rec->time;
596 do_div(avg, rec->counter);
597 if (tracing_thresh && (avg < tracing_thresh))
598 goto out;
599#endif
600
601 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
602 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
603
604#ifdef CONFIG_FUNCTION_GRAPH_TRACER
605 seq_puts(m, " ");
606
607 /* Sample standard deviation (s^2) */
608 if (rec->counter <= 1)
609 stddev = 0;
610 else {
611 /*
612 * Apply Welford's method:
613 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
614 */
615 stddev = rec->counter * rec->time_squared -
616 rec->time * rec->time;
617
618 /*
619 * Divide only 1000 for ns^2 -> us^2 conversion.
620 * trace_print_graph_duration will divide 1000 again.
621 */
622 do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
623 }
624
625 trace_seq_init(&s);
626 trace_print_graph_duration(rec->time, &s);
627 trace_seq_puts(&s, " ");
628 trace_print_graph_duration(avg, &s);
629 trace_seq_puts(&s, " ");
630 trace_print_graph_duration(stddev, &s);
631 trace_print_seq(m, &s);
632#endif
633 seq_putc(m, '\n');
634out:
635 mutex_unlock(&ftrace_profile_lock);
636
637 return ret;
638}
639
640static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
641{
642 struct ftrace_profile_page *pg;
643
644 pg = stat->pages = stat->start;
645
646 while (pg) {
647 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
648 pg->index = 0;
649 pg = pg->next;
650 }
651
652 memset(stat->hash, 0,
653 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
654}
655
656int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
657{
658 struct ftrace_profile_page *pg;
659 int functions;
660 int pages;
661 int i;
662
663 /* If we already allocated, do nothing */
664 if (stat->pages)
665 return 0;
666
667 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
668 if (!stat->pages)
669 return -ENOMEM;
670
671#ifdef CONFIG_DYNAMIC_FTRACE
672 functions = ftrace_update_tot_cnt;
673#else
674 /*
675 * We do not know the number of functions that exist because
676 * dynamic tracing is what counts them. With past experience
677 * we have around 20K functions. That should be more than enough.
678 * It is highly unlikely we will execute every function in
679 * the kernel.
680 */
681 functions = 20000;
682#endif
683
684 pg = stat->start = stat->pages;
685
686 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
687
688 for (i = 1; i < pages; i++) {
689 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
690 if (!pg->next)
691 goto out_free;
692 pg = pg->next;
693 }
694
695 return 0;
696
697 out_free:
698 pg = stat->start;
699 while (pg) {
700 unsigned long tmp = (unsigned long)pg;
701
702 pg = pg->next;
703 free_page(tmp);
704 }
705
706 stat->pages = NULL;
707 stat->start = NULL;
708
709 return -ENOMEM;
710}
711
712static int ftrace_profile_init_cpu(int cpu)
713{
714 struct ftrace_profile_stat *stat;
715 int size;
716
717 stat = &per_cpu(ftrace_profile_stats, cpu);
718
719 if (stat->hash) {
720 /* If the profile is already created, simply reset it */
721 ftrace_profile_reset(stat);
722 return 0;
723 }
724
725 /*
726 * We are profiling all functions, but usually only a few thousand
727 * functions are hit. We'll make a hash of 1024 items.
728 */
729 size = FTRACE_PROFILE_HASH_SIZE;
730
731 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
732
733 if (!stat->hash)
734 return -ENOMEM;
735
736 /* Preallocate the function profiling pages */
737 if (ftrace_profile_pages_init(stat) < 0) {
738 kfree(stat->hash);
739 stat->hash = NULL;
740 return -ENOMEM;
741 }
742
743 return 0;
744}
745
746static int ftrace_profile_init(void)
747{
748 int cpu;
749 int ret = 0;
750
751 for_each_possible_cpu(cpu) {
752 ret = ftrace_profile_init_cpu(cpu);
753 if (ret)
754 break;
755 }
756
757 return ret;
758}
759
760/* interrupts must be disabled */
761static struct ftrace_profile *
762ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
763{
764 struct ftrace_profile *rec;
765 struct hlist_head *hhd;
766 unsigned long key;
767
768 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
769 hhd = &stat->hash[key];
770
771 if (hlist_empty(hhd))
772 return NULL;
773
774 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
775 if (rec->ip == ip)
776 return rec;
777 }
778
779 return NULL;
780}
781
782static void ftrace_add_profile(struct ftrace_profile_stat *stat,
783 struct ftrace_profile *rec)
784{
785 unsigned long key;
786
787 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
788 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
789}
790
791/*
792 * The memory is already allocated, this simply finds a new record to use.
793 */
794static struct ftrace_profile *
795ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
796{
797 struct ftrace_profile *rec = NULL;
798
799 /* prevent recursion (from NMIs) */
800 if (atomic_inc_return(&stat->disabled) != 1)
801 goto out;
802
803 /*
804 * Try to find the function again since an NMI
805 * could have added it
806 */
807 rec = ftrace_find_profiled_func(stat, ip);
808 if (rec)
809 goto out;
810
811 if (stat->pages->index == PROFILES_PER_PAGE) {
812 if (!stat->pages->next)
813 goto out;
814 stat->pages = stat->pages->next;
815 }
816
817 rec = &stat->pages->records[stat->pages->index++];
818 rec->ip = ip;
819 ftrace_add_profile(stat, rec);
820
821 out:
822 atomic_dec(&stat->disabled);
823
824 return rec;
825}
826
827static void
828function_profile_call(unsigned long ip, unsigned long parent_ip,
829 struct ftrace_ops *ops, struct pt_regs *regs)
830{
831 struct ftrace_profile_stat *stat;
832 struct ftrace_profile *rec;
833 unsigned long flags;
834
835 if (!ftrace_profile_enabled)
836 return;
837
838 local_irq_save(flags);
839
840 stat = this_cpu_ptr(&ftrace_profile_stats);
841 if (!stat->hash || !ftrace_profile_enabled)
842 goto out;
843
844 rec = ftrace_find_profiled_func(stat, ip);
845 if (!rec) {
846 rec = ftrace_profile_alloc(stat, ip);
847 if (!rec)
848 goto out;
849 }
850
851 rec->counter++;
852 out:
853 local_irq_restore(flags);
854}
855
856#ifdef CONFIG_FUNCTION_GRAPH_TRACER
857static int profile_graph_entry(struct ftrace_graph_ent *trace)
858{
859 int index = trace->depth;
860
861 function_profile_call(trace->func, 0, NULL, NULL);
862
863 /* If function graph is shutting down, ret_stack can be NULL */
864 if (!current->ret_stack)
865 return 0;
866
867 if (index >= 0 && index < FTRACE_RETFUNC_DEPTH)
868 current->ret_stack[index].subtime = 0;
869
870 return 1;
871}
872
873static void profile_graph_return(struct ftrace_graph_ret *trace)
874{
875 struct ftrace_profile_stat *stat;
876 unsigned long long calltime;
877 struct ftrace_profile *rec;
878 unsigned long flags;
879
880 local_irq_save(flags);
881 stat = this_cpu_ptr(&ftrace_profile_stats);
882 if (!stat->hash || !ftrace_profile_enabled)
883 goto out;
884
885 /* If the calltime was zero'd ignore it */
886 if (!trace->calltime)
887 goto out;
888
889 calltime = trace->rettime - trace->calltime;
890
891 if (!fgraph_graph_time) {
892 int index;
893
894 index = trace->depth;
895
896 /* Append this call time to the parent time to subtract */
897 if (index)
898 current->ret_stack[index - 1].subtime += calltime;
899
900 if (current->ret_stack[index].subtime < calltime)
901 calltime -= current->ret_stack[index].subtime;
902 else
903 calltime = 0;
904 }
905
906 rec = ftrace_find_profiled_func(stat, trace->func);
907 if (rec) {
908 rec->time += calltime;
909 rec->time_squared += calltime * calltime;
910 }
911
912 out:
913 local_irq_restore(flags);
914}
915
916static int register_ftrace_profiler(void)
917{
918 return register_ftrace_graph(&profile_graph_return,
919 &profile_graph_entry);
920}
921
922static void unregister_ftrace_profiler(void)
923{
924 unregister_ftrace_graph();
925}
926#else
927static struct ftrace_ops ftrace_profile_ops __read_mostly = {
928 .func = function_profile_call,
929 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
930 INIT_OPS_HASH(ftrace_profile_ops)
931};
932
933static int register_ftrace_profiler(void)
934{
935 return register_ftrace_function(&ftrace_profile_ops);
936}
937
938static void unregister_ftrace_profiler(void)
939{
940 unregister_ftrace_function(&ftrace_profile_ops);
941}
942#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
943
944static ssize_t
945ftrace_profile_write(struct file *filp, const char __user *ubuf,
946 size_t cnt, loff_t *ppos)
947{
948 unsigned long val;
949 int ret;
950
951 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
952 if (ret)
953 return ret;
954
955 val = !!val;
956
957 mutex_lock(&ftrace_profile_lock);
958 if (ftrace_profile_enabled ^ val) {
959 if (val) {
960 ret = ftrace_profile_init();
961 if (ret < 0) {
962 cnt = ret;
963 goto out;
964 }
965
966 ret = register_ftrace_profiler();
967 if (ret < 0) {
968 cnt = ret;
969 goto out;
970 }
971 ftrace_profile_enabled = 1;
972 } else {
973 ftrace_profile_enabled = 0;
974 /*
975 * unregister_ftrace_profiler calls stop_machine
976 * so this acts like an synchronize_sched.
977 */
978 unregister_ftrace_profiler();
979 }
980 }
981 out:
982 mutex_unlock(&ftrace_profile_lock);
983
984 *ppos += cnt;
985
986 return cnt;
987}
988
989static ssize_t
990ftrace_profile_read(struct file *filp, char __user *ubuf,
991 size_t cnt, loff_t *ppos)
992{
993 char buf[64]; /* big enough to hold a number */
994 int r;
995
996 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
997 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
998}
999
1000static const struct file_operations ftrace_profile_fops = {
1001 .open = tracing_open_generic,
1002 .read = ftrace_profile_read,
1003 .write = ftrace_profile_write,
1004 .llseek = default_llseek,
1005};
1006
1007/* used to initialize the real stat files */
1008static struct tracer_stat function_stats __initdata = {
1009 .name = "functions",
1010 .stat_start = function_stat_start,
1011 .stat_next = function_stat_next,
1012 .stat_cmp = function_stat_cmp,
1013 .stat_headers = function_stat_headers,
1014 .stat_show = function_stat_show
1015};
1016
1017static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1018{
1019 struct ftrace_profile_stat *stat;
1020 struct dentry *entry;
1021 char *name;
1022 int ret;
1023 int cpu;
1024
1025 for_each_possible_cpu(cpu) {
1026 stat = &per_cpu(ftrace_profile_stats, cpu);
1027
1028 name = kasprintf(GFP_KERNEL, "function%d", cpu);
1029 if (!name) {
1030 /*
1031 * The files created are permanent, if something happens
1032 * we still do not free memory.
1033 */
1034 WARN(1,
1035 "Could not allocate stat file for cpu %d\n",
1036 cpu);
1037 return;
1038 }
1039 stat->stat = function_stats;
1040 stat->stat.name = name;
1041 ret = register_stat_tracer(&stat->stat);
1042 if (ret) {
1043 WARN(1,
1044 "Could not register function stat for cpu %d\n",
1045 cpu);
1046 kfree(name);
1047 return;
1048 }
1049 }
1050
1051 entry = tracefs_create_file("function_profile_enabled", 0644,
1052 d_tracer, NULL, &ftrace_profile_fops);
1053 if (!entry)
1054 pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1055}
1056
1057#else /* CONFIG_FUNCTION_PROFILER */
1058static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1059{
1060}
1061#endif /* CONFIG_FUNCTION_PROFILER */
1062
1063static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1064
1065#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1066static int ftrace_graph_active;
1067#else
1068# define ftrace_graph_active 0
1069#endif
1070
1071#ifdef CONFIG_DYNAMIC_FTRACE
1072
1073static struct ftrace_ops *removed_ops;
1074
1075/*
1076 * Set when doing a global update, like enabling all recs or disabling them.
1077 * It is not set when just updating a single ftrace_ops.
1078 */
1079static bool update_all_ops;
1080
1081#ifndef CONFIG_FTRACE_MCOUNT_RECORD
1082# error Dynamic ftrace depends on MCOUNT_RECORD
1083#endif
1084
1085struct ftrace_func_entry {
1086 struct hlist_node hlist;
1087 unsigned long ip;
1088};
1089
1090struct ftrace_func_probe {
1091 struct ftrace_probe_ops *probe_ops;
1092 struct ftrace_ops ops;
1093 struct trace_array *tr;
1094 struct list_head list;
1095 void *data;
1096 int ref;
1097};
1098
1099/*
1100 * We make these constant because no one should touch them,
1101 * but they are used as the default "empty hash", to avoid allocating
1102 * it all the time. These are in a read only section such that if
1103 * anyone does try to modify it, it will cause an exception.
1104 */
1105static const struct hlist_head empty_buckets[1];
1106static const struct ftrace_hash empty_hash = {
1107 .buckets = (struct hlist_head *)empty_buckets,
1108};
1109#define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1110
1111static struct ftrace_ops global_ops = {
1112 .func = ftrace_stub,
1113 .local_hash.notrace_hash = EMPTY_HASH,
1114 .local_hash.filter_hash = EMPTY_HASH,
1115 INIT_OPS_HASH(global_ops)
1116 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
1117 FTRACE_OPS_FL_INITIALIZED |
1118 FTRACE_OPS_FL_PID,
1119};
1120
1121/*
1122 * Used by the stack undwinder to know about dynamic ftrace trampolines.
1123 */
1124struct ftrace_ops *ftrace_ops_trampoline(unsigned long addr)
1125{
1126 struct ftrace_ops *op = NULL;
1127
1128 /*
1129 * Some of the ops may be dynamically allocated,
1130 * they are freed after a synchronize_sched().
1131 */
1132 preempt_disable_notrace();
1133
1134 do_for_each_ftrace_op(op, ftrace_ops_list) {
1135 /*
1136 * This is to check for dynamically allocated trampolines.
1137 * Trampolines that are in kernel text will have
1138 * core_kernel_text() return true.
1139 */
1140 if (op->trampoline && op->trampoline_size)
1141 if (addr >= op->trampoline &&
1142 addr < op->trampoline + op->trampoline_size) {
1143 preempt_enable_notrace();
1144 return op;
1145 }
1146 } while_for_each_ftrace_op(op);
1147 preempt_enable_notrace();
1148
1149 return NULL;
1150}
1151
1152/*
1153 * This is used by __kernel_text_address() to return true if the
1154 * address is on a dynamically allocated trampoline that would
1155 * not return true for either core_kernel_text() or
1156 * is_module_text_address().
1157 */
1158bool is_ftrace_trampoline(unsigned long addr)
1159{
1160 return ftrace_ops_trampoline(addr) != NULL;
1161}
1162
1163struct ftrace_page {
1164 struct ftrace_page *next;
1165 struct dyn_ftrace *records;
1166 int index;
1167 int size;
1168};
1169
1170#define ENTRY_SIZE sizeof(struct dyn_ftrace)
1171#define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1172
1173/* estimate from running different kernels */
1174#define NR_TO_INIT 10000
1175
1176static struct ftrace_page *ftrace_pages_start;
1177static struct ftrace_page *ftrace_pages;
1178
1179static __always_inline unsigned long
1180ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1181{
1182 if (hash->size_bits > 0)
1183 return hash_long(ip, hash->size_bits);
1184
1185 return 0;
1186}
1187
1188/* Only use this function if ftrace_hash_empty() has already been tested */
1189static __always_inline struct ftrace_func_entry *
1190__ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1191{
1192 unsigned long key;
1193 struct ftrace_func_entry *entry;
1194 struct hlist_head *hhd;
1195
1196 key = ftrace_hash_key(hash, ip);
1197 hhd = &hash->buckets[key];
1198
1199 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1200 if (entry->ip == ip)
1201 return entry;
1202 }
1203 return NULL;
1204}
1205
1206/**
1207 * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1208 * @hash: The hash to look at
1209 * @ip: The instruction pointer to test
1210 *
1211 * Search a given @hash to see if a given instruction pointer (@ip)
1212 * exists in it.
1213 *
1214 * Returns the entry that holds the @ip if found. NULL otherwise.
1215 */
1216struct ftrace_func_entry *
1217ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1218{
1219 if (ftrace_hash_empty(hash))
1220 return NULL;
1221
1222 return __ftrace_lookup_ip(hash, ip);
1223}
1224
1225static void __add_hash_entry(struct ftrace_hash *hash,
1226 struct ftrace_func_entry *entry)
1227{
1228 struct hlist_head *hhd;
1229 unsigned long key;
1230
1231 key = ftrace_hash_key(hash, entry->ip);
1232 hhd = &hash->buckets[key];
1233 hlist_add_head(&entry->hlist, hhd);
1234 hash->count++;
1235}
1236
1237static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1238{
1239 struct ftrace_func_entry *entry;
1240
1241 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1242 if (!entry)
1243 return -ENOMEM;
1244
1245 entry->ip = ip;
1246 __add_hash_entry(hash, entry);
1247
1248 return 0;
1249}
1250
1251static void
1252free_hash_entry(struct ftrace_hash *hash,
1253 struct ftrace_func_entry *entry)
1254{
1255 hlist_del(&entry->hlist);
1256 kfree(entry);
1257 hash->count--;
1258}
1259
1260static void
1261remove_hash_entry(struct ftrace_hash *hash,
1262 struct ftrace_func_entry *entry)
1263{
1264 hlist_del_rcu(&entry->hlist);
1265 hash->count--;
1266}
1267
1268static void ftrace_hash_clear(struct ftrace_hash *hash)
1269{
1270 struct hlist_head *hhd;
1271 struct hlist_node *tn;
1272 struct ftrace_func_entry *entry;
1273 int size = 1 << hash->size_bits;
1274 int i;
1275
1276 if (!hash->count)
1277 return;
1278
1279 for (i = 0; i < size; i++) {
1280 hhd = &hash->buckets[i];
1281 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1282 free_hash_entry(hash, entry);
1283 }
1284 FTRACE_WARN_ON(hash->count);
1285}
1286
1287static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1288{
1289 list_del(&ftrace_mod->list);
1290 kfree(ftrace_mod->module);
1291 kfree(ftrace_mod->func);
1292 kfree(ftrace_mod);
1293}
1294
1295static void clear_ftrace_mod_list(struct list_head *head)
1296{
1297 struct ftrace_mod_load *p, *n;
1298
1299 /* stack tracer isn't supported yet */
1300 if (!head)
1301 return;
1302
1303 mutex_lock(&ftrace_lock);
1304 list_for_each_entry_safe(p, n, head, list)
1305 free_ftrace_mod(p);
1306 mutex_unlock(&ftrace_lock);
1307}
1308
1309static void free_ftrace_hash(struct ftrace_hash *hash)
1310{
1311 if (!hash || hash == EMPTY_HASH)
1312 return;
1313 ftrace_hash_clear(hash);
1314 kfree(hash->buckets);
1315 kfree(hash);
1316}
1317
1318static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1319{
1320 struct ftrace_hash *hash;
1321
1322 hash = container_of(rcu, struct ftrace_hash, rcu);
1323 free_ftrace_hash(hash);
1324}
1325
1326static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1327{
1328 if (!hash || hash == EMPTY_HASH)
1329 return;
1330 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1331}
1332
1333void ftrace_free_filter(struct ftrace_ops *ops)
1334{
1335 ftrace_ops_init(ops);
1336 free_ftrace_hash(ops->func_hash->filter_hash);
1337 free_ftrace_hash(ops->func_hash->notrace_hash);
1338}
1339
1340static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1341{
1342 struct ftrace_hash *hash;
1343 int size;
1344
1345 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1346 if (!hash)
1347 return NULL;
1348
1349 size = 1 << size_bits;
1350 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1351
1352 if (!hash->buckets) {
1353 kfree(hash);
1354 return NULL;
1355 }
1356
1357 hash->size_bits = size_bits;
1358
1359 return hash;
1360}
1361
1362
1363static int ftrace_add_mod(struct trace_array *tr,
1364 const char *func, const char *module,
1365 int enable)
1366{
1367 struct ftrace_mod_load *ftrace_mod;
1368 struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1369
1370 ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1371 if (!ftrace_mod)
1372 return -ENOMEM;
1373
1374 ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1375 ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1376 ftrace_mod->enable = enable;
1377
1378 if (!ftrace_mod->func || !ftrace_mod->module)
1379 goto out_free;
1380
1381 list_add(&ftrace_mod->list, mod_head);
1382
1383 return 0;
1384
1385 out_free:
1386 free_ftrace_mod(ftrace_mod);
1387
1388 return -ENOMEM;
1389}
1390
1391static struct ftrace_hash *
1392alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1393{
1394 struct ftrace_func_entry *entry;
1395 struct ftrace_hash *new_hash;
1396 int size;
1397 int ret;
1398 int i;
1399
1400 new_hash = alloc_ftrace_hash(size_bits);
1401 if (!new_hash)
1402 return NULL;
1403
1404 if (hash)
1405 new_hash->flags = hash->flags;
1406
1407 /* Empty hash? */
1408 if (ftrace_hash_empty(hash))
1409 return new_hash;
1410
1411 size = 1 << hash->size_bits;
1412 for (i = 0; i < size; i++) {
1413 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1414 ret = add_hash_entry(new_hash, entry->ip);
1415 if (ret < 0)
1416 goto free_hash;
1417 }
1418 }
1419
1420 FTRACE_WARN_ON(new_hash->count != hash->count);
1421
1422 return new_hash;
1423
1424 free_hash:
1425 free_ftrace_hash(new_hash);
1426 return NULL;
1427}
1428
1429static void
1430ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1431static void
1432ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1433
1434static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1435 struct ftrace_hash *new_hash);
1436
1437static struct ftrace_hash *
1438__ftrace_hash_move(struct ftrace_hash *src)
1439{
1440 struct ftrace_func_entry *entry;
1441 struct hlist_node *tn;
1442 struct hlist_head *hhd;
1443 struct ftrace_hash *new_hash;
1444 int size = src->count;
1445 int bits = 0;
1446 int i;
1447
1448 /*
1449 * If the new source is empty, just return the empty_hash.
1450 */
1451 if (ftrace_hash_empty(src))
1452 return EMPTY_HASH;
1453
1454 /*
1455 * Make the hash size about 1/2 the # found
1456 */
1457 for (size /= 2; size; size >>= 1)
1458 bits++;
1459
1460 /* Don't allocate too much */
1461 if (bits > FTRACE_HASH_MAX_BITS)
1462 bits = FTRACE_HASH_MAX_BITS;
1463
1464 new_hash = alloc_ftrace_hash(bits);
1465 if (!new_hash)
1466 return NULL;
1467
1468 new_hash->flags = src->flags;
1469
1470 size = 1 << src->size_bits;
1471 for (i = 0; i < size; i++) {
1472 hhd = &src->buckets[i];
1473 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1474 remove_hash_entry(src, entry);
1475 __add_hash_entry(new_hash, entry);
1476 }
1477 }
1478
1479 return new_hash;
1480}
1481
1482static int
1483ftrace_hash_move(struct ftrace_ops *ops, int enable,
1484 struct ftrace_hash **dst, struct ftrace_hash *src)
1485{
1486 struct ftrace_hash *new_hash;
1487 int ret;
1488
1489 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1490 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1491 return -EINVAL;
1492
1493 new_hash = __ftrace_hash_move(src);
1494 if (!new_hash)
1495 return -ENOMEM;
1496
1497 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1498 if (enable) {
1499 /* IPMODIFY should be updated only when filter_hash updating */
1500 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1501 if (ret < 0) {
1502 free_ftrace_hash(new_hash);
1503 return ret;
1504 }
1505 }
1506
1507 /*
1508 * Remove the current set, update the hash and add
1509 * them back.
1510 */
1511 ftrace_hash_rec_disable_modify(ops, enable);
1512
1513 rcu_assign_pointer(*dst, new_hash);
1514
1515 ftrace_hash_rec_enable_modify(ops, enable);
1516
1517 return 0;
1518}
1519
1520static bool hash_contains_ip(unsigned long ip,
1521 struct ftrace_ops_hash *hash)
1522{
1523 /*
1524 * The function record is a match if it exists in the filter
1525 * hash and not in the notrace hash. Note, an emty hash is
1526 * considered a match for the filter hash, but an empty
1527 * notrace hash is considered not in the notrace hash.
1528 */
1529 return (ftrace_hash_empty(hash->filter_hash) ||
1530 __ftrace_lookup_ip(hash->filter_hash, ip)) &&
1531 (ftrace_hash_empty(hash->notrace_hash) ||
1532 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1533}
1534
1535/*
1536 * Test the hashes for this ops to see if we want to call
1537 * the ops->func or not.
1538 *
1539 * It's a match if the ip is in the ops->filter_hash or
1540 * the filter_hash does not exist or is empty,
1541 * AND
1542 * the ip is not in the ops->notrace_hash.
1543 *
1544 * This needs to be called with preemption disabled as
1545 * the hashes are freed with call_rcu_sched().
1546 */
1547static int
1548ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1549{
1550 struct ftrace_ops_hash hash;
1551 int ret;
1552
1553#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1554 /*
1555 * There's a small race when adding ops that the ftrace handler
1556 * that wants regs, may be called without them. We can not
1557 * allow that handler to be called if regs is NULL.
1558 */
1559 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1560 return 0;
1561#endif
1562
1563 rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1564 rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1565
1566 if (hash_contains_ip(ip, &hash))
1567 ret = 1;
1568 else
1569 ret = 0;
1570
1571 return ret;
1572}
1573
1574/*
1575 * This is a double for. Do not use 'break' to break out of the loop,
1576 * you must use a goto.
1577 */
1578#define do_for_each_ftrace_rec(pg, rec) \
1579 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1580 int _____i; \
1581 for (_____i = 0; _____i < pg->index; _____i++) { \
1582 rec = &pg->records[_____i];
1583
1584#define while_for_each_ftrace_rec() \
1585 } \
1586 }
1587
1588
1589static int ftrace_cmp_recs(const void *a, const void *b)
1590{
1591 const struct dyn_ftrace *key = a;
1592 const struct dyn_ftrace *rec = b;
1593
1594 if (key->flags < rec->ip)
1595 return -1;
1596 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1597 return 1;
1598 return 0;
1599}
1600
1601/**
1602 * ftrace_location_range - return the first address of a traced location
1603 * if it touches the given ip range
1604 * @start: start of range to search.
1605 * @end: end of range to search (inclusive). @end points to the last byte
1606 * to check.
1607 *
1608 * Returns rec->ip if the related ftrace location is a least partly within
1609 * the given address range. That is, the first address of the instruction
1610 * that is either a NOP or call to the function tracer. It checks the ftrace
1611 * internal tables to determine if the address belongs or not.
1612 */
1613unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1614{
1615 struct ftrace_page *pg;
1616 struct dyn_ftrace *rec;
1617 struct dyn_ftrace key;
1618
1619 key.ip = start;
1620 key.flags = end; /* overload flags, as it is unsigned long */
1621
1622 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1623 if (end < pg->records[0].ip ||
1624 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1625 continue;
1626 rec = bsearch(&key, pg->records, pg->index,
1627 sizeof(struct dyn_ftrace),
1628 ftrace_cmp_recs);
1629 if (rec)
1630 return rec->ip;
1631 }
1632
1633 return 0;
1634}
1635
1636/**
1637 * ftrace_location - return true if the ip giving is a traced location
1638 * @ip: the instruction pointer to check
1639 *
1640 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1641 * That is, the instruction that is either a NOP or call to
1642 * the function tracer. It checks the ftrace internal tables to
1643 * determine if the address belongs or not.
1644 */
1645unsigned long ftrace_location(unsigned long ip)
1646{
1647 return ftrace_location_range(ip, ip);
1648}
1649
1650/**
1651 * ftrace_text_reserved - return true if range contains an ftrace location
1652 * @start: start of range to search
1653 * @end: end of range to search (inclusive). @end points to the last byte to check.
1654 *
1655 * Returns 1 if @start and @end contains a ftrace location.
1656 * That is, the instruction that is either a NOP or call to
1657 * the function tracer. It checks the ftrace internal tables to
1658 * determine if the address belongs or not.
1659 */
1660int ftrace_text_reserved(const void *start, const void *end)
1661{
1662 unsigned long ret;
1663
1664 ret = ftrace_location_range((unsigned long)start,
1665 (unsigned long)end);
1666
1667 return (int)!!ret;
1668}
1669
1670/* Test if ops registered to this rec needs regs */
1671static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1672{
1673 struct ftrace_ops *ops;
1674 bool keep_regs = false;
1675
1676 for (ops = ftrace_ops_list;
1677 ops != &ftrace_list_end; ops = ops->next) {
1678 /* pass rec in as regs to have non-NULL val */
1679 if (ftrace_ops_test(ops, rec->ip, rec)) {
1680 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1681 keep_regs = true;
1682 break;
1683 }
1684 }
1685 }
1686
1687 return keep_regs;
1688}
1689
1690static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1691 int filter_hash,
1692 bool inc)
1693{
1694 struct ftrace_hash *hash;
1695 struct ftrace_hash *other_hash;
1696 struct ftrace_page *pg;
1697 struct dyn_ftrace *rec;
1698 bool update = false;
1699 int count = 0;
1700 int all = false;
1701
1702 /* Only update if the ops has been registered */
1703 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1704 return false;
1705
1706 /*
1707 * In the filter_hash case:
1708 * If the count is zero, we update all records.
1709 * Otherwise we just update the items in the hash.
1710 *
1711 * In the notrace_hash case:
1712 * We enable the update in the hash.
1713 * As disabling notrace means enabling the tracing,
1714 * and enabling notrace means disabling, the inc variable
1715 * gets inversed.
1716 */
1717 if (filter_hash) {
1718 hash = ops->func_hash->filter_hash;
1719 other_hash = ops->func_hash->notrace_hash;
1720 if (ftrace_hash_empty(hash))
1721 all = true;
1722 } else {
1723 inc = !inc;
1724 hash = ops->func_hash->notrace_hash;
1725 other_hash = ops->func_hash->filter_hash;
1726 /*
1727 * If the notrace hash has no items,
1728 * then there's nothing to do.
1729 */
1730 if (ftrace_hash_empty(hash))
1731 return false;
1732 }
1733
1734 do_for_each_ftrace_rec(pg, rec) {
1735 int in_other_hash = 0;
1736 int in_hash = 0;
1737 int match = 0;
1738
1739 if (rec->flags & FTRACE_FL_DISABLED)
1740 continue;
1741
1742 if (all) {
1743 /*
1744 * Only the filter_hash affects all records.
1745 * Update if the record is not in the notrace hash.
1746 */
1747 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1748 match = 1;
1749 } else {
1750 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1751 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1752
1753 /*
1754 * If filter_hash is set, we want to match all functions
1755 * that are in the hash but not in the other hash.
1756 *
1757 * If filter_hash is not set, then we are decrementing.
1758 * That means we match anything that is in the hash
1759 * and also in the other_hash. That is, we need to turn
1760 * off functions in the other hash because they are disabled
1761 * by this hash.
1762 */
1763 if (filter_hash && in_hash && !in_other_hash)
1764 match = 1;
1765 else if (!filter_hash && in_hash &&
1766 (in_other_hash || ftrace_hash_empty(other_hash)))
1767 match = 1;
1768 }
1769 if (!match)
1770 continue;
1771
1772 if (inc) {
1773 rec->flags++;
1774 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1775 return false;
1776
1777 /*
1778 * If there's only a single callback registered to a
1779 * function, and the ops has a trampoline registered
1780 * for it, then we can call it directly.
1781 */
1782 if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1783 rec->flags |= FTRACE_FL_TRAMP;
1784 else
1785 /*
1786 * If we are adding another function callback
1787 * to this function, and the previous had a
1788 * custom trampoline in use, then we need to go
1789 * back to the default trampoline.
1790 */
1791 rec->flags &= ~FTRACE_FL_TRAMP;
1792
1793 /*
1794 * If any ops wants regs saved for this function
1795 * then all ops will get saved regs.
1796 */
1797 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1798 rec->flags |= FTRACE_FL_REGS;
1799 } else {
1800 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1801 return false;
1802 rec->flags--;
1803
1804 /*
1805 * If the rec had REGS enabled and the ops that is
1806 * being removed had REGS set, then see if there is
1807 * still any ops for this record that wants regs.
1808 * If not, we can stop recording them.
1809 */
1810 if (ftrace_rec_count(rec) > 0 &&
1811 rec->flags & FTRACE_FL_REGS &&
1812 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1813 if (!test_rec_ops_needs_regs(rec))
1814 rec->flags &= ~FTRACE_FL_REGS;
1815 }
1816
1817 /*
1818 * If the rec had TRAMP enabled, then it needs to
1819 * be cleared. As TRAMP can only be enabled iff
1820 * there is only a single ops attached to it.
1821 * In otherwords, always disable it on decrementing.
1822 * In the future, we may set it if rec count is
1823 * decremented to one, and the ops that is left
1824 * has a trampoline.
1825 */
1826 rec->flags &= ~FTRACE_FL_TRAMP;
1827
1828 /*
1829 * flags will be cleared in ftrace_check_record()
1830 * if rec count is zero.
1831 */
1832 }
1833 count++;
1834
1835 /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1836 update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1837
1838 /* Shortcut, if we handled all records, we are done. */
1839 if (!all && count == hash->count)
1840 return update;
1841 } while_for_each_ftrace_rec();
1842
1843 return update;
1844}
1845
1846static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1847 int filter_hash)
1848{
1849 return __ftrace_hash_rec_update(ops, filter_hash, 0);
1850}
1851
1852static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1853 int filter_hash)
1854{
1855 return __ftrace_hash_rec_update(ops, filter_hash, 1);
1856}
1857
1858static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1859 int filter_hash, int inc)
1860{
1861 struct ftrace_ops *op;
1862
1863 __ftrace_hash_rec_update(ops, filter_hash, inc);
1864
1865 if (ops->func_hash != &global_ops.local_hash)
1866 return;
1867
1868 /*
1869 * If the ops shares the global_ops hash, then we need to update
1870 * all ops that are enabled and use this hash.
1871 */
1872 do_for_each_ftrace_op(op, ftrace_ops_list) {
1873 /* Already done */
1874 if (op == ops)
1875 continue;
1876 if (op->func_hash == &global_ops.local_hash)
1877 __ftrace_hash_rec_update(op, filter_hash, inc);
1878 } while_for_each_ftrace_op(op);
1879}
1880
1881static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1882 int filter_hash)
1883{
1884 ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1885}
1886
1887static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1888 int filter_hash)
1889{
1890 ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1891}
1892
1893/*
1894 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1895 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1896 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1897 * Note that old_hash and new_hash has below meanings
1898 * - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1899 * - If the hash is EMPTY_HASH, it hits nothing
1900 * - Anything else hits the recs which match the hash entries.
1901 */
1902static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1903 struct ftrace_hash *old_hash,
1904 struct ftrace_hash *new_hash)
1905{
1906 struct ftrace_page *pg;
1907 struct dyn_ftrace *rec, *end = NULL;
1908 int in_old, in_new;
1909
1910 /* Only update if the ops has been registered */
1911 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1912 return 0;
1913
1914 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1915 return 0;
1916
1917 /*
1918 * Since the IPMODIFY is a very address sensitive action, we do not
1919 * allow ftrace_ops to set all functions to new hash.
1920 */
1921 if (!new_hash || !old_hash)
1922 return -EINVAL;
1923
1924 /* Update rec->flags */
1925 do_for_each_ftrace_rec(pg, rec) {
1926
1927 if (rec->flags & FTRACE_FL_DISABLED)
1928 continue;
1929
1930 /* We need to update only differences of filter_hash */
1931 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1932 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1933 if (in_old == in_new)
1934 continue;
1935
1936 if (in_new) {
1937 /* New entries must ensure no others are using it */
1938 if (rec->flags & FTRACE_FL_IPMODIFY)
1939 goto rollback;
1940 rec->flags |= FTRACE_FL_IPMODIFY;
1941 } else /* Removed entry */
1942 rec->flags &= ~FTRACE_FL_IPMODIFY;
1943 } while_for_each_ftrace_rec();
1944
1945 return 0;
1946
1947rollback:
1948 end = rec;
1949
1950 /* Roll back what we did above */
1951 do_for_each_ftrace_rec(pg, rec) {
1952
1953 if (rec->flags & FTRACE_FL_DISABLED)
1954 continue;
1955
1956 if (rec == end)
1957 goto err_out;
1958
1959 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1960 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1961 if (in_old == in_new)
1962 continue;
1963
1964 if (in_new)
1965 rec->flags &= ~FTRACE_FL_IPMODIFY;
1966 else
1967 rec->flags |= FTRACE_FL_IPMODIFY;
1968 } while_for_each_ftrace_rec();
1969
1970err_out:
1971 return -EBUSY;
1972}
1973
1974static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1975{
1976 struct ftrace_hash *hash = ops->func_hash->filter_hash;
1977
1978 if (ftrace_hash_empty(hash))
1979 hash = NULL;
1980
1981 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1982}
1983
1984/* Disabling always succeeds */
1985static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1986{
1987 struct ftrace_hash *hash = ops->func_hash->filter_hash;
1988
1989 if (ftrace_hash_empty(hash))
1990 hash = NULL;
1991
1992 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1993}
1994
1995static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1996 struct ftrace_hash *new_hash)
1997{
1998 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1999
2000 if (ftrace_hash_empty(old_hash))
2001 old_hash = NULL;
2002
2003 if (ftrace_hash_empty(new_hash))
2004 new_hash = NULL;
2005
2006 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
2007}
2008
2009static void print_ip_ins(const char *fmt, const unsigned char *p)
2010{
2011 int i;
2012
2013 printk(KERN_CONT "%s", fmt);
2014
2015 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
2016 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
2017}
2018
2019static struct ftrace_ops *
2020ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
2021static struct ftrace_ops *
2022ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
2023
2024enum ftrace_bug_type ftrace_bug_type;
2025const void *ftrace_expected;
2026
2027static void print_bug_type(void)
2028{
2029 switch (ftrace_bug_type) {
2030 case FTRACE_BUG_UNKNOWN:
2031 break;
2032 case FTRACE_BUG_INIT:
2033 pr_info("Initializing ftrace call sites\n");
2034 break;
2035 case FTRACE_BUG_NOP:
2036 pr_info("Setting ftrace call site to NOP\n");
2037 break;
2038 case FTRACE_BUG_CALL:
2039 pr_info("Setting ftrace call site to call ftrace function\n");
2040 break;
2041 case FTRACE_BUG_UPDATE:
2042 pr_info("Updating ftrace call site to call a different ftrace function\n");
2043 break;
2044 }
2045}
2046
2047/**
2048 * ftrace_bug - report and shutdown function tracer
2049 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2050 * @rec: The record that failed
2051 *
2052 * The arch code that enables or disables the function tracing
2053 * can call ftrace_bug() when it has detected a problem in
2054 * modifying the code. @failed should be one of either:
2055 * EFAULT - if the problem happens on reading the @ip address
2056 * EINVAL - if what is read at @ip is not what was expected
2057 * EPERM - if the problem happens on writting to the @ip address
2058 */
2059void ftrace_bug(int failed, struct dyn_ftrace *rec)
2060{
2061 unsigned long ip = rec ? rec->ip : 0;
2062
2063 switch (failed) {
2064 case -EFAULT:
2065 FTRACE_WARN_ON_ONCE(1);
2066 pr_info("ftrace faulted on modifying ");
2067 print_ip_sym(ip);
2068 break;
2069 case -EINVAL:
2070 FTRACE_WARN_ON_ONCE(1);
2071 pr_info("ftrace failed to modify ");
2072 print_ip_sym(ip);
2073 print_ip_ins(" actual: ", (unsigned char *)ip);
2074 pr_cont("\n");
2075 if (ftrace_expected) {
2076 print_ip_ins(" expected: ", ftrace_expected);
2077 pr_cont("\n");
2078 }
2079 break;
2080 case -EPERM:
2081 FTRACE_WARN_ON_ONCE(1);
2082 pr_info("ftrace faulted on writing ");
2083 print_ip_sym(ip);
2084 break;
2085 default:
2086 FTRACE_WARN_ON_ONCE(1);
2087 pr_info("ftrace faulted on unknown error ");
2088 print_ip_sym(ip);
2089 }
2090 print_bug_type();
2091 if (rec) {
2092 struct ftrace_ops *ops = NULL;
2093
2094 pr_info("ftrace record flags: %lx\n", rec->flags);
2095 pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2096 rec->flags & FTRACE_FL_REGS ? " R" : " ");
2097 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2098 ops = ftrace_find_tramp_ops_any(rec);
2099 if (ops) {
2100 do {
2101 pr_cont("\ttramp: %pS (%pS)",
2102 (void *)ops->trampoline,
2103 (void *)ops->func);
2104 ops = ftrace_find_tramp_ops_next(rec, ops);
2105 } while (ops);
2106 } else
2107 pr_cont("\ttramp: ERROR!");
2108
2109 }
2110 ip = ftrace_get_addr_curr(rec);
2111 pr_cont("\n expected tramp: %lx\n", ip);
2112 }
2113}
2114
2115static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2116{
2117 unsigned long flag = 0UL;
2118
2119 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2120
2121 if (rec->flags & FTRACE_FL_DISABLED)
2122 return FTRACE_UPDATE_IGNORE;
2123
2124 /*
2125 * If we are updating calls:
2126 *
2127 * If the record has a ref count, then we need to enable it
2128 * because someone is using it.
2129 *
2130 * Otherwise we make sure its disabled.
2131 *
2132 * If we are disabling calls, then disable all records that
2133 * are enabled.
2134 */
2135 if (enable && ftrace_rec_count(rec))
2136 flag = FTRACE_FL_ENABLED;
2137
2138 /*
2139 * If enabling and the REGS flag does not match the REGS_EN, or
2140 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2141 * this record. Set flags to fail the compare against ENABLED.
2142 */
2143 if (flag) {
2144 if (!(rec->flags & FTRACE_FL_REGS) !=
2145 !(rec->flags & FTRACE_FL_REGS_EN))
2146 flag |= FTRACE_FL_REGS;
2147
2148 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2149 !(rec->flags & FTRACE_FL_TRAMP_EN))
2150 flag |= FTRACE_FL_TRAMP;
2151 }
2152
2153 /* If the state of this record hasn't changed, then do nothing */
2154 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2155 return FTRACE_UPDATE_IGNORE;
2156
2157 if (flag) {
2158 /* Save off if rec is being enabled (for return value) */
2159 flag ^= rec->flags & FTRACE_FL_ENABLED;
2160
2161 if (update) {
2162 rec->flags |= FTRACE_FL_ENABLED;
2163 if (flag & FTRACE_FL_REGS) {
2164 if (rec->flags & FTRACE_FL_REGS)
2165 rec->flags |= FTRACE_FL_REGS_EN;
2166 else
2167 rec->flags &= ~FTRACE_FL_REGS_EN;
2168 }
2169 if (flag & FTRACE_FL_TRAMP) {
2170 if (rec->flags & FTRACE_FL_TRAMP)
2171 rec->flags |= FTRACE_FL_TRAMP_EN;
2172 else
2173 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2174 }
2175 }
2176
2177 /*
2178 * If this record is being updated from a nop, then
2179 * return UPDATE_MAKE_CALL.
2180 * Otherwise,
2181 * return UPDATE_MODIFY_CALL to tell the caller to convert
2182 * from the save regs, to a non-save regs function or
2183 * vice versa, or from a trampoline call.
2184 */
2185 if (flag & FTRACE_FL_ENABLED) {
2186 ftrace_bug_type = FTRACE_BUG_CALL;
2187 return FTRACE_UPDATE_MAKE_CALL;
2188 }
2189
2190 ftrace_bug_type = FTRACE_BUG_UPDATE;
2191 return FTRACE_UPDATE_MODIFY_CALL;
2192 }
2193
2194 if (update) {
2195 /* If there's no more users, clear all flags */
2196 if (!ftrace_rec_count(rec))
2197 rec->flags = 0;
2198 else
2199 /*
2200 * Just disable the record, but keep the ops TRAMP
2201 * and REGS states. The _EN flags must be disabled though.
2202 */
2203 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2204 FTRACE_FL_REGS_EN);
2205 }
2206
2207 ftrace_bug_type = FTRACE_BUG_NOP;
2208 return FTRACE_UPDATE_MAKE_NOP;
2209}
2210
2211/**
2212 * ftrace_update_record, set a record that now is tracing or not
2213 * @rec: the record to update
2214 * @enable: set to 1 if the record is tracing, zero to force disable
2215 *
2216 * The records that represent all functions that can be traced need
2217 * to be updated when tracing has been enabled.
2218 */
2219int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2220{
2221 return ftrace_check_record(rec, enable, 1);
2222}
2223
2224/**
2225 * ftrace_test_record, check if the record has been enabled or not
2226 * @rec: the record to test
2227 * @enable: set to 1 to check if enabled, 0 if it is disabled
2228 *
2229 * The arch code may need to test if a record is already set to
2230 * tracing to determine how to modify the function code that it
2231 * represents.
2232 */
2233int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2234{
2235 return ftrace_check_record(rec, enable, 0);
2236}
2237
2238static struct ftrace_ops *
2239ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2240{
2241 struct ftrace_ops *op;
2242 unsigned long ip = rec->ip;
2243
2244 do_for_each_ftrace_op(op, ftrace_ops_list) {
2245
2246 if (!op->trampoline)
2247 continue;
2248
2249 if (hash_contains_ip(ip, op->func_hash))
2250 return op;
2251 } while_for_each_ftrace_op(op);
2252
2253 return NULL;
2254}
2255
2256static struct ftrace_ops *
2257ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2258 struct ftrace_ops *op)
2259{
2260 unsigned long ip = rec->ip;
2261
2262 while_for_each_ftrace_op(op) {
2263
2264 if (!op->trampoline)
2265 continue;
2266
2267 if (hash_contains_ip(ip, op->func_hash))
2268 return op;
2269 }
2270
2271 return NULL;
2272}
2273
2274static struct ftrace_ops *
2275ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2276{
2277 struct ftrace_ops *op;
2278 unsigned long ip = rec->ip;
2279
2280 /*
2281 * Need to check removed ops first.
2282 * If they are being removed, and this rec has a tramp,
2283 * and this rec is in the ops list, then it would be the
2284 * one with the tramp.
2285 */
2286 if (removed_ops) {
2287 if (hash_contains_ip(ip, &removed_ops->old_hash))
2288 return removed_ops;
2289 }
2290
2291 /*
2292 * Need to find the current trampoline for a rec.
2293 * Now, a trampoline is only attached to a rec if there
2294 * was a single 'ops' attached to it. But this can be called
2295 * when we are adding another op to the rec or removing the
2296 * current one. Thus, if the op is being added, we can
2297 * ignore it because it hasn't attached itself to the rec
2298 * yet.
2299 *
2300 * If an ops is being modified (hooking to different functions)
2301 * then we don't care about the new functions that are being
2302 * added, just the old ones (that are probably being removed).
2303 *
2304 * If we are adding an ops to a function that already is using
2305 * a trampoline, it needs to be removed (trampolines are only
2306 * for single ops connected), then an ops that is not being
2307 * modified also needs to be checked.
2308 */
2309 do_for_each_ftrace_op(op, ftrace_ops_list) {
2310
2311 if (!op->trampoline)
2312 continue;
2313
2314 /*
2315 * If the ops is being added, it hasn't gotten to
2316 * the point to be removed from this tree yet.
2317 */
2318 if (op->flags & FTRACE_OPS_FL_ADDING)
2319 continue;
2320
2321
2322 /*
2323 * If the ops is being modified and is in the old
2324 * hash, then it is probably being removed from this
2325 * function.
2326 */
2327 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2328 hash_contains_ip(ip, &op->old_hash))
2329 return op;
2330 /*
2331 * If the ops is not being added or modified, and it's
2332 * in its normal filter hash, then this must be the one
2333 * we want!
2334 */
2335 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2336 hash_contains_ip(ip, op->func_hash))
2337 return op;
2338
2339 } while_for_each_ftrace_op(op);
2340
2341 return NULL;
2342}
2343
2344static struct ftrace_ops *
2345ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2346{
2347 struct ftrace_ops *op;
2348 unsigned long ip = rec->ip;
2349
2350 do_for_each_ftrace_op(op, ftrace_ops_list) {
2351 /* pass rec in as regs to have non-NULL val */
2352 if (hash_contains_ip(ip, op->func_hash))
2353 return op;
2354 } while_for_each_ftrace_op(op);
2355
2356 return NULL;
2357}
2358
2359/**
2360 * ftrace_get_addr_new - Get the call address to set to
2361 * @rec: The ftrace record descriptor
2362 *
2363 * If the record has the FTRACE_FL_REGS set, that means that it
2364 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2365 * is not not set, then it wants to convert to the normal callback.
2366 *
2367 * Returns the address of the trampoline to set to
2368 */
2369unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2370{
2371 struct ftrace_ops *ops;
2372
2373 /* Trampolines take precedence over regs */
2374 if (rec->flags & FTRACE_FL_TRAMP) {
2375 ops = ftrace_find_tramp_ops_new(rec);
2376 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2377 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2378 (void *)rec->ip, (void *)rec->ip, rec->flags);
2379 /* Ftrace is shutting down, return anything */
2380 return (unsigned long)FTRACE_ADDR;
2381 }
2382 return ops->trampoline;
2383 }
2384
2385 if (rec->flags & FTRACE_FL_REGS)
2386 return (unsigned long)FTRACE_REGS_ADDR;
2387 else
2388 return (unsigned long)FTRACE_ADDR;
2389}
2390
2391/**
2392 * ftrace_get_addr_curr - Get the call address that is already there
2393 * @rec: The ftrace record descriptor
2394 *
2395 * The FTRACE_FL_REGS_EN is set when the record already points to
2396 * a function that saves all the regs. Basically the '_EN' version
2397 * represents the current state of the function.
2398 *
2399 * Returns the address of the trampoline that is currently being called
2400 */
2401unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2402{
2403 struct ftrace_ops *ops;
2404
2405 /* Trampolines take precedence over regs */
2406 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2407 ops = ftrace_find_tramp_ops_curr(rec);
2408 if (FTRACE_WARN_ON(!ops)) {
2409 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2410 (void *)rec->ip, (void *)rec->ip);
2411 /* Ftrace is shutting down, return anything */
2412 return (unsigned long)FTRACE_ADDR;
2413 }
2414 return ops->trampoline;
2415 }
2416
2417 if (rec->flags & FTRACE_FL_REGS_EN)
2418 return (unsigned long)FTRACE_REGS_ADDR;
2419 else
2420 return (unsigned long)FTRACE_ADDR;
2421}
2422
2423static int
2424__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2425{
2426 unsigned long ftrace_old_addr;
2427 unsigned long ftrace_addr;
2428 int ret;
2429
2430 ftrace_addr = ftrace_get_addr_new(rec);
2431
2432 /* This needs to be done before we call ftrace_update_record */
2433 ftrace_old_addr = ftrace_get_addr_curr(rec);
2434
2435 ret = ftrace_update_record(rec, enable);
2436
2437 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2438
2439 switch (ret) {
2440 case FTRACE_UPDATE_IGNORE:
2441 return 0;
2442
2443 case FTRACE_UPDATE_MAKE_CALL:
2444 ftrace_bug_type = FTRACE_BUG_CALL;
2445 return ftrace_make_call(rec, ftrace_addr);
2446
2447 case FTRACE_UPDATE_MAKE_NOP:
2448 ftrace_bug_type = FTRACE_BUG_NOP;
2449 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2450
2451 case FTRACE_UPDATE_MODIFY_CALL:
2452 ftrace_bug_type = FTRACE_BUG_UPDATE;
2453 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2454 }
2455
2456 return -1; /* unknow ftrace bug */
2457}
2458
2459void __weak ftrace_replace_code(int enable)
2460{
2461 struct dyn_ftrace *rec;
2462 struct ftrace_page *pg;
2463 int failed;
2464
2465 if (unlikely(ftrace_disabled))
2466 return;
2467
2468 do_for_each_ftrace_rec(pg, rec) {
2469
2470 if (rec->flags & FTRACE_FL_DISABLED)
2471 continue;
2472
2473 failed = __ftrace_replace_code(rec, enable);
2474 if (failed) {
2475 ftrace_bug(failed, rec);
2476 /* Stop processing */
2477 return;
2478 }
2479 } while_for_each_ftrace_rec();
2480}
2481
2482struct ftrace_rec_iter {
2483 struct ftrace_page *pg;
2484 int index;
2485};
2486
2487/**
2488 * ftrace_rec_iter_start, start up iterating over traced functions
2489 *
2490 * Returns an iterator handle that is used to iterate over all
2491 * the records that represent address locations where functions
2492 * are traced.
2493 *
2494 * May return NULL if no records are available.
2495 */
2496struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2497{
2498 /*
2499 * We only use a single iterator.
2500 * Protected by the ftrace_lock mutex.
2501 */
2502 static struct ftrace_rec_iter ftrace_rec_iter;
2503 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2504
2505 iter->pg = ftrace_pages_start;
2506 iter->index = 0;
2507
2508 /* Could have empty pages */
2509 while (iter->pg && !iter->pg->index)
2510 iter->pg = iter->pg->next;
2511
2512 if (!iter->pg)
2513 return NULL;
2514
2515 return iter;
2516}
2517
2518/**
2519 * ftrace_rec_iter_next, get the next record to process.
2520 * @iter: The handle to the iterator.
2521 *
2522 * Returns the next iterator after the given iterator @iter.
2523 */
2524struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2525{
2526 iter->index++;
2527
2528 if (iter->index >= iter->pg->index) {
2529 iter->pg = iter->pg->next;
2530 iter->index = 0;
2531
2532 /* Could have empty pages */
2533 while (iter->pg && !iter->pg->index)
2534 iter->pg = iter->pg->next;
2535 }
2536
2537 if (!iter->pg)
2538 return NULL;
2539
2540 return iter;
2541}
2542
2543/**
2544 * ftrace_rec_iter_record, get the record at the iterator location
2545 * @iter: The current iterator location
2546 *
2547 * Returns the record that the current @iter is at.
2548 */
2549struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2550{
2551 return &iter->pg->records[iter->index];
2552}
2553
2554static int
2555ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2556{
2557 int ret;
2558
2559 if (unlikely(ftrace_disabled))
2560 return 0;
2561
2562 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2563 if (ret) {
2564 ftrace_bug_type = FTRACE_BUG_INIT;
2565 ftrace_bug(ret, rec);
2566 return 0;
2567 }
2568 return 1;
2569}
2570
2571/*
2572 * archs can override this function if they must do something
2573 * before the modifying code is performed.
2574 */
2575int __weak ftrace_arch_code_modify_prepare(void)
2576{
2577 return 0;
2578}
2579
2580/*
2581 * archs can override this function if they must do something
2582 * after the modifying code is performed.
2583 */
2584int __weak ftrace_arch_code_modify_post_process(void)
2585{
2586 return 0;
2587}
2588
2589void ftrace_modify_all_code(int command)
2590{
2591 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2592 int err = 0;
2593
2594 /*
2595 * If the ftrace_caller calls a ftrace_ops func directly,
2596 * we need to make sure that it only traces functions it
2597 * expects to trace. When doing the switch of functions,
2598 * we need to update to the ftrace_ops_list_func first
2599 * before the transition between old and new calls are set,
2600 * as the ftrace_ops_list_func will check the ops hashes
2601 * to make sure the ops are having the right functions
2602 * traced.
2603 */
2604 if (update) {
2605 err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2606 if (FTRACE_WARN_ON(err))
2607 return;
2608 }
2609
2610 if (command & FTRACE_UPDATE_CALLS)
2611 ftrace_replace_code(1);
2612 else if (command & FTRACE_DISABLE_CALLS)
2613 ftrace_replace_code(0);
2614
2615 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2616 function_trace_op = set_function_trace_op;
2617 smp_wmb();
2618 /* If irqs are disabled, we are in stop machine */
2619 if (!irqs_disabled())
2620 smp_call_function(ftrace_sync_ipi, NULL, 1);
2621 err = ftrace_update_ftrace_func(ftrace_trace_function);
2622 if (FTRACE_WARN_ON(err))
2623 return;
2624 }
2625
2626 if (command & FTRACE_START_FUNC_RET)
2627 err = ftrace_enable_ftrace_graph_caller();
2628 else if (command & FTRACE_STOP_FUNC_RET)
2629 err = ftrace_disable_ftrace_graph_caller();
2630 FTRACE_WARN_ON(err);
2631}
2632
2633static int __ftrace_modify_code(void *data)
2634{
2635 int *command = data;
2636
2637 ftrace_modify_all_code(*command);
2638
2639 return 0;
2640}
2641
2642/**
2643 * ftrace_run_stop_machine, go back to the stop machine method
2644 * @command: The command to tell ftrace what to do
2645 *
2646 * If an arch needs to fall back to the stop machine method, the
2647 * it can call this function.
2648 */
2649void ftrace_run_stop_machine(int command)
2650{
2651 stop_machine(__ftrace_modify_code, &command, NULL);
2652}
2653
2654/**
2655 * arch_ftrace_update_code, modify the code to trace or not trace
2656 * @command: The command that needs to be done
2657 *
2658 * Archs can override this function if it does not need to
2659 * run stop_machine() to modify code.
2660 */
2661void __weak arch_ftrace_update_code(int command)
2662{
2663 ftrace_run_stop_machine(command);
2664}
2665
2666static void ftrace_run_update_code(int command)
2667{
2668 int ret;
2669
2670 ret = ftrace_arch_code_modify_prepare();
2671 FTRACE_WARN_ON(ret);
2672 if (ret)
2673 return;
2674
2675 /*
2676 * By default we use stop_machine() to modify the code.
2677 * But archs can do what ever they want as long as it
2678 * is safe. The stop_machine() is the safest, but also
2679 * produces the most overhead.
2680 */
2681 arch_ftrace_update_code(command);
2682
2683 ret = ftrace_arch_code_modify_post_process();
2684 FTRACE_WARN_ON(ret);
2685}
2686
2687static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2688 struct ftrace_ops_hash *old_hash)
2689{
2690 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2691 ops->old_hash.filter_hash = old_hash->filter_hash;
2692 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2693 ftrace_run_update_code(command);
2694 ops->old_hash.filter_hash = NULL;
2695 ops->old_hash.notrace_hash = NULL;
2696 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2697}
2698
2699static ftrace_func_t saved_ftrace_func;
2700static int ftrace_start_up;
2701
2702void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2703{
2704}
2705
2706static void ftrace_startup_enable(int command)
2707{
2708 if (saved_ftrace_func != ftrace_trace_function) {
2709 saved_ftrace_func = ftrace_trace_function;
2710 command |= FTRACE_UPDATE_TRACE_FUNC;
2711 }
2712
2713 if (!command || !ftrace_enabled)
2714 return;
2715
2716 ftrace_run_update_code(command);
2717}
2718
2719static void ftrace_startup_all(int command)
2720{
2721 update_all_ops = true;
2722 ftrace_startup_enable(command);
2723 update_all_ops = false;
2724}
2725
2726static int ftrace_startup(struct ftrace_ops *ops, int command)
2727{
2728 int ret;
2729
2730 if (unlikely(ftrace_disabled))
2731 return -ENODEV;
2732
2733 ret = __register_ftrace_function(ops);
2734 if (ret)
2735 return ret;
2736
2737 ftrace_start_up++;
2738
2739 /*
2740 * Note that ftrace probes uses this to start up
2741 * and modify functions it will probe. But we still
2742 * set the ADDING flag for modification, as probes
2743 * do not have trampolines. If they add them in the
2744 * future, then the probes will need to distinguish
2745 * between adding and updating probes.
2746 */
2747 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2748
2749 ret = ftrace_hash_ipmodify_enable(ops);
2750 if (ret < 0) {
2751 /* Rollback registration process */
2752 __unregister_ftrace_function(ops);
2753 ftrace_start_up--;
2754 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2755 return ret;
2756 }
2757
2758 if (ftrace_hash_rec_enable(ops, 1))
2759 command |= FTRACE_UPDATE_CALLS;
2760
2761 ftrace_startup_enable(command);
2762
2763 ops->flags &= ~FTRACE_OPS_FL_ADDING;
2764
2765 return 0;
2766}
2767
2768static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2769{
2770 int ret;
2771
2772 if (unlikely(ftrace_disabled))
2773 return -ENODEV;
2774
2775 ret = __unregister_ftrace_function(ops);
2776 if (ret)
2777 return ret;
2778
2779 ftrace_start_up--;
2780 /*
2781 * Just warn in case of unbalance, no need to kill ftrace, it's not
2782 * critical but the ftrace_call callers may be never nopped again after
2783 * further ftrace uses.
2784 */
2785 WARN_ON_ONCE(ftrace_start_up < 0);
2786
2787 /* Disabling ipmodify never fails */
2788 ftrace_hash_ipmodify_disable(ops);
2789
2790 if (ftrace_hash_rec_disable(ops, 1))
2791 command |= FTRACE_UPDATE_CALLS;
2792
2793 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2794
2795 if (saved_ftrace_func != ftrace_trace_function) {
2796 saved_ftrace_func = ftrace_trace_function;
2797 command |= FTRACE_UPDATE_TRACE_FUNC;
2798 }
2799
2800 if (!command || !ftrace_enabled) {
2801 /*
2802 * If these are dynamic or per_cpu ops, they still
2803 * need their data freed. Since, function tracing is
2804 * not currently active, we can just free them
2805 * without synchronizing all CPUs.
2806 */
2807 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
2808 goto free_ops;
2809
2810 return 0;
2811 }
2812
2813 /*
2814 * If the ops uses a trampoline, then it needs to be
2815 * tested first on update.
2816 */
2817 ops->flags |= FTRACE_OPS_FL_REMOVING;
2818 removed_ops = ops;
2819
2820 /* The trampoline logic checks the old hashes */
2821 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2822 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2823
2824 ftrace_run_update_code(command);
2825
2826 /*
2827 * If there's no more ops registered with ftrace, run a
2828 * sanity check to make sure all rec flags are cleared.
2829 */
2830 if (rcu_dereference_protected(ftrace_ops_list,
2831 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
2832 struct ftrace_page *pg;
2833 struct dyn_ftrace *rec;
2834
2835 do_for_each_ftrace_rec(pg, rec) {
2836 if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_FL_DISABLED))
2837 pr_warn(" %pS flags:%lx\n",
2838 (void *)rec->ip, rec->flags);
2839 } while_for_each_ftrace_rec();
2840 }
2841
2842 ops->old_hash.filter_hash = NULL;
2843 ops->old_hash.notrace_hash = NULL;
2844
2845 removed_ops = NULL;
2846 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2847
2848 /*
2849 * Dynamic ops may be freed, we must make sure that all
2850 * callers are done before leaving this function.
2851 * The same goes for freeing the per_cpu data of the per_cpu
2852 * ops.
2853 */
2854 if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
2855 /*
2856 * We need to do a hard force of sched synchronization.
2857 * This is because we use preempt_disable() to do RCU, but
2858 * the function tracers can be called where RCU is not watching
2859 * (like before user_exit()). We can not rely on the RCU
2860 * infrastructure to do the synchronization, thus we must do it
2861 * ourselves.
2862 */
2863 schedule_on_each_cpu(ftrace_sync);
2864
2865 /*
2866 * When the kernel is preeptive, tasks can be preempted
2867 * while on a ftrace trampoline. Just scheduling a task on
2868 * a CPU is not good enough to flush them. Calling
2869 * synchornize_rcu_tasks() will wait for those tasks to
2870 * execute and either schedule voluntarily or enter user space.
2871 */
2872 if (IS_ENABLED(CONFIG_PREEMPT))
2873 synchronize_rcu_tasks();
2874
2875 free_ops:
2876 arch_ftrace_trampoline_free(ops);
2877 }
2878
2879 return 0;
2880}
2881
2882static void ftrace_startup_sysctl(void)
2883{
2884 int command;
2885
2886 if (unlikely(ftrace_disabled))
2887 return;
2888
2889 /* Force update next time */
2890 saved_ftrace_func = NULL;
2891 /* ftrace_start_up is true if we want ftrace running */
2892 if (ftrace_start_up) {
2893 command = FTRACE_UPDATE_CALLS;
2894 if (ftrace_graph_active)
2895 command |= FTRACE_START_FUNC_RET;
2896 ftrace_startup_enable(command);
2897 }
2898}
2899
2900static void ftrace_shutdown_sysctl(void)
2901{
2902 int command;
2903
2904 if (unlikely(ftrace_disabled))
2905 return;
2906
2907 /* ftrace_start_up is true if ftrace is running */
2908 if (ftrace_start_up) {
2909 command = FTRACE_DISABLE_CALLS;
2910 if (ftrace_graph_active)
2911 command |= FTRACE_STOP_FUNC_RET;
2912 ftrace_run_update_code(command);
2913 }
2914}
2915
2916static u64 ftrace_update_time;
2917unsigned long ftrace_update_tot_cnt;
2918
2919static inline int ops_traces_mod(struct ftrace_ops *ops)
2920{
2921 /*
2922 * Filter_hash being empty will default to trace module.
2923 * But notrace hash requires a test of individual module functions.
2924 */
2925 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2926 ftrace_hash_empty(ops->func_hash->notrace_hash);
2927}
2928
2929/*
2930 * Check if the current ops references the record.
2931 *
2932 * If the ops traces all functions, then it was already accounted for.
2933 * If the ops does not trace the current record function, skip it.
2934 * If the ops ignores the function via notrace filter, skip it.
2935 */
2936static inline bool
2937ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2938{
2939 /* If ops isn't enabled, ignore it */
2940 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2941 return 0;
2942
2943 /* If ops traces all then it includes this function */
2944 if (ops_traces_mod(ops))
2945 return 1;
2946
2947 /* The function must be in the filter */
2948 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2949 !__ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2950 return 0;
2951
2952 /* If in notrace hash, we ignore it too */
2953 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2954 return 0;
2955
2956 return 1;
2957}
2958
2959static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2960{
2961 struct ftrace_page *pg;
2962 struct dyn_ftrace *p;
2963 u64 start, stop;
2964 unsigned long update_cnt = 0;
2965 unsigned long rec_flags = 0;
2966 int i;
2967
2968 start = ftrace_now(raw_smp_processor_id());
2969
2970 /*
2971 * When a module is loaded, this function is called to convert
2972 * the calls to mcount in its text to nops, and also to create
2973 * an entry in the ftrace data. Now, if ftrace is activated
2974 * after this call, but before the module sets its text to
2975 * read-only, the modification of enabling ftrace can fail if
2976 * the read-only is done while ftrace is converting the calls.
2977 * To prevent this, the module's records are set as disabled
2978 * and will be enabled after the call to set the module's text
2979 * to read-only.
2980 */
2981 if (mod)
2982 rec_flags |= FTRACE_FL_DISABLED;
2983
2984 for (pg = new_pgs; pg; pg = pg->next) {
2985
2986 for (i = 0; i < pg->index; i++) {
2987
2988 /* If something went wrong, bail without enabling anything */
2989 if (unlikely(ftrace_disabled))
2990 return -1;
2991
2992 p = &pg->records[i];
2993 p->flags = rec_flags;
2994
2995 /*
2996 * Do the initial record conversion from mcount jump
2997 * to the NOP instructions.
2998 */
2999 if (!ftrace_code_disable(mod, p))
3000 break;
3001
3002 update_cnt++;
3003 }
3004 }
3005
3006 stop = ftrace_now(raw_smp_processor_id());
3007 ftrace_update_time = stop - start;
3008 ftrace_update_tot_cnt += update_cnt;
3009
3010 return 0;
3011}
3012
3013static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3014{
3015 int order;
3016 int cnt;
3017
3018 if (WARN_ON(!count))
3019 return -EINVAL;
3020
3021 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
3022
3023 /*
3024 * We want to fill as much as possible. No more than a page
3025 * may be empty.
3026 */
3027 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
3028 order--;
3029
3030 again:
3031 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3032
3033 if (!pg->records) {
3034 /* if we can't allocate this size, try something smaller */
3035 if (!order)
3036 return -ENOMEM;
3037 order >>= 1;
3038 goto again;
3039 }
3040
3041 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3042 pg->size = cnt;
3043
3044 if (cnt > count)
3045 cnt = count;
3046
3047 return cnt;
3048}
3049
3050static struct ftrace_page *
3051ftrace_allocate_pages(unsigned long num_to_init)
3052{
3053 struct ftrace_page *start_pg;
3054 struct ftrace_page *pg;
3055 int order;
3056 int cnt;
3057
3058 if (!num_to_init)
3059 return 0;
3060
3061 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3062 if (!pg)
3063 return NULL;
3064
3065 /*
3066 * Try to allocate as much as possible in one continues
3067 * location that fills in all of the space. We want to
3068 * waste as little space as possible.
3069 */
3070 for (;;) {
3071 cnt = ftrace_allocate_records(pg, num_to_init);
3072 if (cnt < 0)
3073 goto free_pages;
3074
3075 num_to_init -= cnt;
3076 if (!num_to_init)
3077 break;
3078
3079 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3080 if (!pg->next)
3081 goto free_pages;
3082
3083 pg = pg->next;
3084 }
3085
3086 return start_pg;
3087
3088 free_pages:
3089 pg = start_pg;
3090 while (pg) {
3091 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3092 free_pages((unsigned long)pg->records, order);
3093 start_pg = pg->next;
3094 kfree(pg);
3095 pg = start_pg;
3096 }
3097 pr_info("ftrace: FAILED to allocate memory for functions\n");
3098 return NULL;
3099}
3100
3101#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3102
3103struct ftrace_iterator {
3104 loff_t pos;
3105 loff_t func_pos;
3106 loff_t mod_pos;
3107 struct ftrace_page *pg;
3108 struct dyn_ftrace *func;
3109 struct ftrace_func_probe *probe;
3110 struct ftrace_func_entry *probe_entry;
3111 struct trace_parser parser;
3112 struct ftrace_hash *hash;
3113 struct ftrace_ops *ops;
3114 struct trace_array *tr;
3115 struct list_head *mod_list;
3116 int pidx;
3117 int idx;
3118 unsigned flags;
3119};
3120
3121static void *
3122t_probe_next(struct seq_file *m, loff_t *pos)
3123{
3124 struct ftrace_iterator *iter = m->private;
3125 struct trace_array *tr = iter->ops->private;
3126 struct list_head *func_probes;
3127 struct ftrace_hash *hash;
3128 struct list_head *next;
3129 struct hlist_node *hnd = NULL;
3130 struct hlist_head *hhd;
3131 int size;
3132
3133 (*pos)++;
3134 iter->pos = *pos;
3135
3136 if (!tr)
3137 return NULL;
3138
3139 func_probes = &tr->func_probes;
3140 if (list_empty(func_probes))
3141 return NULL;
3142
3143 if (!iter->probe) {
3144 next = func_probes->next;
3145 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3146 }
3147
3148 if (iter->probe_entry)
3149 hnd = &iter->probe_entry->hlist;
3150
3151 hash = iter->probe->ops.func_hash->filter_hash;
3152 size = 1 << hash->size_bits;
3153
3154 retry:
3155 if (iter->pidx >= size) {
3156 if (iter->probe->list.next == func_probes)
3157 return NULL;
3158 next = iter->probe->list.next;
3159 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3160 hash = iter->probe->ops.func_hash->filter_hash;
3161 size = 1 << hash->size_bits;
3162 iter->pidx = 0;
3163 }
3164
3165 hhd = &hash->buckets[iter->pidx];
3166
3167 if (hlist_empty(hhd)) {
3168 iter->pidx++;
3169 hnd = NULL;
3170 goto retry;
3171 }
3172
3173 if (!hnd)
3174 hnd = hhd->first;
3175 else {
3176 hnd = hnd->next;
3177 if (!hnd) {
3178 iter->pidx++;
3179 goto retry;
3180 }
3181 }
3182
3183 if (WARN_ON_ONCE(!hnd))
3184 return NULL;
3185
3186 iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3187
3188 return iter;
3189}
3190
3191static void *t_probe_start(struct seq_file *m, loff_t *pos)
3192{
3193 struct ftrace_iterator *iter = m->private;
3194 void *p = NULL;
3195 loff_t l;
3196
3197 if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3198 return NULL;
3199
3200 if (iter->mod_pos > *pos)
3201 return NULL;
3202
3203 iter->probe = NULL;
3204 iter->probe_entry = NULL;
3205 iter->pidx = 0;
3206 for (l = 0; l <= (*pos - iter->mod_pos); ) {
3207 p = t_probe_next(m, &l);
3208 if (!p)
3209 break;
3210 }
3211 if (!p)
3212 return NULL;
3213
3214 /* Only set this if we have an item */
3215 iter->flags |= FTRACE_ITER_PROBE;
3216
3217 return iter;
3218}
3219
3220static int
3221t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3222{
3223 struct ftrace_func_entry *probe_entry;
3224 struct ftrace_probe_ops *probe_ops;
3225 struct ftrace_func_probe *probe;
3226
3227 probe = iter->probe;
3228 probe_entry = iter->probe_entry;
3229
3230 if (WARN_ON_ONCE(!probe || !probe_entry))
3231 return -EIO;
3232
3233 probe_ops = probe->probe_ops;
3234
3235 if (probe_ops->print)
3236 return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3237
3238 seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3239 (void *)probe_ops->func);
3240
3241 return 0;
3242}
3243
3244static void *
3245t_mod_next(struct seq_file *m, loff_t *pos)
3246{
3247 struct ftrace_iterator *iter = m->private;
3248 struct trace_array *tr = iter->tr;
3249
3250 (*pos)++;
3251 iter->pos = *pos;
3252
3253 iter->mod_list = iter->mod_list->next;
3254
3255 if (iter->mod_list == &tr->mod_trace ||
3256 iter->mod_list == &tr->mod_notrace) {
3257 iter->flags &= ~FTRACE_ITER_MOD;
3258 return NULL;
3259 }
3260
3261 iter->mod_pos = *pos;
3262
3263 return iter;
3264}
3265
3266static void *t_mod_start(struct seq_file *m, loff_t *pos)
3267{
3268 struct ftrace_iterator *iter = m->private;
3269 void *p = NULL;
3270 loff_t l;
3271
3272 if (iter->func_pos > *pos)
3273 return NULL;
3274
3275 iter->mod_pos = iter->func_pos;
3276
3277 /* probes are only available if tr is set */
3278 if (!iter->tr)
3279 return NULL;
3280
3281 for (l = 0; l <= (*pos - iter->func_pos); ) {
3282 p = t_mod_next(m, &l);
3283 if (!p)
3284 break;
3285 }
3286 if (!p) {
3287 iter->flags &= ~FTRACE_ITER_MOD;
3288 return t_probe_start(m, pos);
3289 }
3290
3291 /* Only set this if we have an item */
3292 iter->flags |= FTRACE_ITER_MOD;
3293
3294 return iter;
3295}
3296
3297static int
3298t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
3299{
3300 struct ftrace_mod_load *ftrace_mod;
3301 struct trace_array *tr = iter->tr;
3302
3303 if (WARN_ON_ONCE(!iter->mod_list) ||
3304 iter->mod_list == &tr->mod_trace ||
3305 iter->mod_list == &tr->mod_notrace)
3306 return -EIO;
3307
3308 ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
3309
3310 if (ftrace_mod->func)
3311 seq_printf(m, "%s", ftrace_mod->func);
3312 else
3313 seq_putc(m, '*');
3314
3315 seq_printf(m, ":mod:%s\n", ftrace_mod->module);
3316
3317 return 0;
3318}
3319
3320static void *
3321t_func_next(struct seq_file *m, loff_t *pos)
3322{
3323 struct ftrace_iterator *iter = m->private;
3324 struct dyn_ftrace *rec = NULL;
3325
3326 (*pos)++;
3327
3328 retry:
3329 if (iter->idx >= iter->pg->index) {
3330 if (iter->pg->next) {
3331 iter->pg = iter->pg->next;
3332 iter->idx = 0;
3333 goto retry;
3334 }
3335 } else {
3336 rec = &iter->pg->records[iter->idx++];
3337 if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3338 !ftrace_lookup_ip(iter->hash, rec->ip)) ||
3339
3340 ((iter->flags & FTRACE_ITER_ENABLED) &&
3341 !(rec->flags & FTRACE_FL_ENABLED))) {
3342
3343 rec = NULL;
3344 goto retry;
3345 }
3346 }
3347
3348 if (!rec)
3349 return NULL;
3350
3351 iter->pos = iter->func_pos = *pos;
3352 iter->func = rec;
3353
3354 return iter;
3355}
3356
3357static void *
3358t_next(struct seq_file *m, void *v, loff_t *pos)
3359{
3360 struct ftrace_iterator *iter = m->private;
3361 loff_t l = *pos; /* t_probe_start() must use original pos */
3362 void *ret;
3363
3364 if (unlikely(ftrace_disabled))
3365 return NULL;
3366
3367 if (iter->flags & FTRACE_ITER_PROBE)
3368 return t_probe_next(m, pos);
3369
3370 if (iter->flags & FTRACE_ITER_MOD)
3371 return t_mod_next(m, pos);
3372
3373 if (iter->flags & FTRACE_ITER_PRINTALL) {
3374 /* next must increment pos, and t_probe_start does not */
3375 (*pos)++;
3376 return t_mod_start(m, &l);
3377 }
3378
3379 ret = t_func_next(m, pos);
3380
3381 if (!ret)
3382 return t_mod_start(m, &l);
3383
3384 return ret;
3385}
3386
3387static void reset_iter_read(struct ftrace_iterator *iter)
3388{
3389 iter->pos = 0;
3390 iter->func_pos = 0;
3391 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
3392}
3393
3394static void *t_start(struct seq_file *m, loff_t *pos)
3395{
3396 struct ftrace_iterator *iter = m->private;
3397 void *p = NULL;
3398 loff_t l;
3399
3400 mutex_lock(&ftrace_lock);
3401
3402 if (unlikely(ftrace_disabled))
3403 return NULL;
3404
3405 /*
3406 * If an lseek was done, then reset and start from beginning.
3407 */
3408 if (*pos < iter->pos)
3409 reset_iter_read(iter);
3410
3411 /*
3412 * For set_ftrace_filter reading, if we have the filter
3413 * off, we can short cut and just print out that all
3414 * functions are enabled.
3415 */
3416 if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3417 ftrace_hash_empty(iter->hash)) {
3418 iter->func_pos = 1; /* Account for the message */
3419 if (*pos > 0)
3420 return t_mod_start(m, pos);
3421 iter->flags |= FTRACE_ITER_PRINTALL;
3422 /* reset in case of seek/pread */
3423 iter->flags &= ~FTRACE_ITER_PROBE;
3424 return iter;
3425 }
3426
3427 if (iter->flags & FTRACE_ITER_MOD)
3428 return t_mod_start(m, pos);
3429
3430 /*
3431 * Unfortunately, we need to restart at ftrace_pages_start
3432 * every time we let go of the ftrace_mutex. This is because
3433 * those pointers can change without the lock.
3434 */
3435 iter->pg = ftrace_pages_start;
3436 iter->idx = 0;
3437 for (l = 0; l <= *pos; ) {
3438 p = t_func_next(m, &l);
3439 if (!p)
3440 break;
3441 }
3442
3443 if (!p)
3444 return t_mod_start(m, pos);
3445
3446 return iter;
3447}
3448
3449static void t_stop(struct seq_file *m, void *p)
3450{
3451 mutex_unlock(&ftrace_lock);
3452}
3453
3454void * __weak
3455arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3456{
3457 return NULL;
3458}
3459
3460static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3461 struct dyn_ftrace *rec)
3462{
3463 void *ptr;
3464
3465 ptr = arch_ftrace_trampoline_func(ops, rec);
3466 if (ptr)
3467 seq_printf(m, " ->%pS", ptr);
3468}
3469
3470static int t_show(struct seq_file *m, void *v)
3471{
3472 struct ftrace_iterator *iter = m->private;
3473 struct dyn_ftrace *rec;
3474
3475 if (iter->flags & FTRACE_ITER_PROBE)
3476 return t_probe_show(m, iter);
3477
3478 if (iter->flags & FTRACE_ITER_MOD)
3479 return t_mod_show(m, iter);
3480
3481 if (iter->flags & FTRACE_ITER_PRINTALL) {
3482 if (iter->flags & FTRACE_ITER_NOTRACE)
3483 seq_puts(m, "#### no functions disabled ####\n");
3484 else
3485 seq_puts(m, "#### all functions enabled ####\n");
3486 return 0;
3487 }
3488
3489 rec = iter->func;
3490
3491 if (!rec)
3492 return 0;
3493
3494 seq_printf(m, "%ps", (void *)rec->ip);
3495 if (iter->flags & FTRACE_ITER_ENABLED) {
3496 struct ftrace_ops *ops;
3497
3498 seq_printf(m, " (%ld)%s%s",
3499 ftrace_rec_count(rec),
3500 rec->flags & FTRACE_FL_REGS ? " R" : " ",
3501 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ");
3502 if (rec->flags & FTRACE_FL_TRAMP_EN) {
3503 ops = ftrace_find_tramp_ops_any(rec);
3504 if (ops) {
3505 do {
3506 seq_printf(m, "\ttramp: %pS (%pS)",
3507 (void *)ops->trampoline,
3508 (void *)ops->func);
3509 add_trampoline_func(m, ops, rec);
3510 ops = ftrace_find_tramp_ops_next(rec, ops);
3511 } while (ops);
3512 } else
3513 seq_puts(m, "\ttramp: ERROR!");
3514 } else {
3515 add_trampoline_func(m, NULL, rec);
3516 }
3517 }
3518
3519 seq_putc(m, '\n');
3520
3521 return 0;
3522}
3523
3524static const struct seq_operations show_ftrace_seq_ops = {
3525 .start = t_start,
3526 .next = t_next,
3527 .stop = t_stop,
3528 .show = t_show,
3529};
3530
3531static int
3532ftrace_avail_open(struct inode *inode, struct file *file)
3533{
3534 struct ftrace_iterator *iter;
3535
3536 if (unlikely(ftrace_disabled))
3537 return -ENODEV;
3538
3539 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3540 if (!iter)
3541 return -ENOMEM;
3542
3543 iter->pg = ftrace_pages_start;
3544 iter->ops = &global_ops;
3545
3546 return 0;
3547}
3548
3549static int
3550ftrace_enabled_open(struct inode *inode, struct file *file)
3551{
3552 struct ftrace_iterator *iter;
3553
3554 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3555 if (!iter)
3556 return -ENOMEM;
3557
3558 iter->pg = ftrace_pages_start;
3559 iter->flags = FTRACE_ITER_ENABLED;
3560 iter->ops = &global_ops;
3561
3562 return 0;
3563}
3564
3565/**
3566 * ftrace_regex_open - initialize function tracer filter files
3567 * @ops: The ftrace_ops that hold the hash filters
3568 * @flag: The type of filter to process
3569 * @inode: The inode, usually passed in to your open routine
3570 * @file: The file, usually passed in to your open routine
3571 *
3572 * ftrace_regex_open() initializes the filter files for the
3573 * @ops. Depending on @flag it may process the filter hash or
3574 * the notrace hash of @ops. With this called from the open
3575 * routine, you can use ftrace_filter_write() for the write
3576 * routine if @flag has FTRACE_ITER_FILTER set, or
3577 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3578 * tracing_lseek() should be used as the lseek routine, and
3579 * release must call ftrace_regex_release().
3580 */
3581int
3582ftrace_regex_open(struct ftrace_ops *ops, int flag,
3583 struct inode *inode, struct file *file)
3584{
3585 struct ftrace_iterator *iter;
3586 struct ftrace_hash *hash;
3587 struct list_head *mod_head;
3588 struct trace_array *tr = ops->private;
3589 int ret = 0;
3590
3591 ftrace_ops_init(ops);
3592
3593 if (unlikely(ftrace_disabled))
3594 return -ENODEV;
3595
3596 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3597 if (!iter)
3598 return -ENOMEM;
3599
3600 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3601 kfree(iter);
3602 return -ENOMEM;
3603 }
3604
3605 iter->ops = ops;
3606 iter->flags = flag;
3607 iter->tr = tr;
3608
3609 mutex_lock(&ops->func_hash->regex_lock);
3610
3611 if (flag & FTRACE_ITER_NOTRACE) {
3612 hash = ops->func_hash->notrace_hash;
3613 mod_head = tr ? &tr->mod_notrace : NULL;
3614 } else {
3615 hash = ops->func_hash->filter_hash;
3616 mod_head = tr ? &tr->mod_trace : NULL;
3617 }
3618
3619 iter->mod_list = mod_head;
3620
3621 if (file->f_mode & FMODE_WRITE) {
3622 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3623
3624 if (file->f_flags & O_TRUNC) {
3625 iter->hash = alloc_ftrace_hash(size_bits);
3626 clear_ftrace_mod_list(mod_head);
3627 } else {
3628 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3629 }
3630
3631 if (!iter->hash) {
3632 trace_parser_put(&iter->parser);
3633 kfree(iter);
3634 ret = -ENOMEM;
3635 goto out_unlock;
3636 }
3637 } else
3638 iter->hash = hash;
3639
3640 if (file->f_mode & FMODE_READ) {
3641 iter->pg = ftrace_pages_start;
3642
3643 ret = seq_open(file, &show_ftrace_seq_ops);
3644 if (!ret) {
3645 struct seq_file *m = file->private_data;
3646 m->private = iter;
3647 } else {
3648 /* Failed */
3649 free_ftrace_hash(iter->hash);
3650 trace_parser_put(&iter->parser);
3651 kfree(iter);
3652 }
3653 } else
3654 file->private_data = iter;
3655
3656 out_unlock:
3657 mutex_unlock(&ops->func_hash->regex_lock);
3658
3659 return ret;
3660}
3661
3662static int
3663ftrace_filter_open(struct inode *inode, struct file *file)
3664{
3665 struct ftrace_ops *ops = inode->i_private;
3666
3667 return ftrace_regex_open(ops,
3668 FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
3669 inode, file);
3670}
3671
3672static int
3673ftrace_notrace_open(struct inode *inode, struct file *file)
3674{
3675 struct ftrace_ops *ops = inode->i_private;
3676
3677 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3678 inode, file);
3679}
3680
3681/* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3682struct ftrace_glob {
3683 char *search;
3684 unsigned len;
3685 int type;
3686};
3687
3688/*
3689 * If symbols in an architecture don't correspond exactly to the user-visible
3690 * name of what they represent, it is possible to define this function to
3691 * perform the necessary adjustments.
3692*/
3693char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3694{
3695 return str;
3696}
3697
3698static int ftrace_match(char *str, struct ftrace_glob *g)
3699{
3700 int matched = 0;
3701 int slen;
3702
3703 str = arch_ftrace_match_adjust(str, g->search);
3704
3705 switch (g->type) {
3706 case MATCH_FULL:
3707 if (strcmp(str, g->search) == 0)
3708 matched = 1;
3709 break;
3710 case MATCH_FRONT_ONLY:
3711 if (strncmp(str, g->search, g->len) == 0)
3712 matched = 1;
3713 break;
3714 case MATCH_MIDDLE_ONLY:
3715 if (strstr(str, g->search))
3716 matched = 1;
3717 break;
3718 case MATCH_END_ONLY:
3719 slen = strlen(str);
3720 if (slen >= g->len &&
3721 memcmp(str + slen - g->len, g->search, g->len) == 0)
3722 matched = 1;
3723 break;
3724 case MATCH_GLOB:
3725 if (glob_match(g->search, str))
3726 matched = 1;
3727 break;
3728 }
3729
3730 return matched;
3731}
3732
3733static int
3734enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3735{
3736 struct ftrace_func_entry *entry;
3737 int ret = 0;
3738
3739 entry = ftrace_lookup_ip(hash, rec->ip);
3740 if (clear_filter) {
3741 /* Do nothing if it doesn't exist */
3742 if (!entry)
3743 return 0;
3744
3745 free_hash_entry(hash, entry);
3746 } else {
3747 /* Do nothing if it exists */
3748 if (entry)
3749 return 0;
3750
3751 ret = add_hash_entry(hash, rec->ip);
3752 }
3753 return ret;
3754}
3755
3756static int
3757ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3758 struct ftrace_glob *mod_g, int exclude_mod)
3759{
3760 char str[KSYM_SYMBOL_LEN];
3761 char *modname;
3762
3763 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3764
3765 if (mod_g) {
3766 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3767
3768 /* blank module name to match all modules */
3769 if (!mod_g->len) {
3770 /* blank module globbing: modname xor exclude_mod */
3771 if (!exclude_mod != !modname)
3772 goto func_match;
3773 return 0;
3774 }
3775
3776 /*
3777 * exclude_mod is set to trace everything but the given
3778 * module. If it is set and the module matches, then
3779 * return 0. If it is not set, and the module doesn't match
3780 * also return 0. Otherwise, check the function to see if
3781 * that matches.
3782 */
3783 if (!mod_matches == !exclude_mod)
3784 return 0;
3785func_match:
3786 /* blank search means to match all funcs in the mod */
3787 if (!func_g->len)
3788 return 1;
3789 }
3790
3791 return ftrace_match(str, func_g);
3792}
3793
3794static int
3795match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3796{
3797 struct ftrace_page *pg;
3798 struct dyn_ftrace *rec;
3799 struct ftrace_glob func_g = { .type = MATCH_FULL };
3800 struct ftrace_glob mod_g = { .type = MATCH_FULL };
3801 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3802 int exclude_mod = 0;
3803 int found = 0;
3804 int ret;
3805 int clear_filter = 0;
3806
3807 if (func) {
3808 func_g.type = filter_parse_regex(func, len, &func_g.search,
3809 &clear_filter);
3810 func_g.len = strlen(func_g.search);
3811 }
3812
3813 if (mod) {
3814 mod_g.type = filter_parse_regex(mod, strlen(mod),
3815 &mod_g.search, &exclude_mod);
3816 mod_g.len = strlen(mod_g.search);
3817 }
3818
3819 mutex_lock(&ftrace_lock);
3820
3821 if (unlikely(ftrace_disabled))
3822 goto out_unlock;
3823
3824 do_for_each_ftrace_rec(pg, rec) {
3825
3826 if (rec->flags & FTRACE_FL_DISABLED)
3827 continue;
3828
3829 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3830 ret = enter_record(hash, rec, clear_filter);
3831 if (ret < 0) {
3832 found = ret;
3833 goto out_unlock;
3834 }
3835 found = 1;
3836 }
3837 } while_for_each_ftrace_rec();
3838 out_unlock:
3839 mutex_unlock(&ftrace_lock);
3840
3841 return found;
3842}
3843
3844static int
3845ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3846{
3847 return match_records(hash, buff, len, NULL);
3848}
3849
3850static void ftrace_ops_update_code(struct ftrace_ops *ops,
3851 struct ftrace_ops_hash *old_hash)
3852{
3853 struct ftrace_ops *op;
3854
3855 if (!ftrace_enabled)
3856 return;
3857
3858 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
3859 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
3860 return;
3861 }
3862
3863 /*
3864 * If this is the shared global_ops filter, then we need to
3865 * check if there is another ops that shares it, is enabled.
3866 * If so, we still need to run the modify code.
3867 */
3868 if (ops->func_hash != &global_ops.local_hash)
3869 return;
3870
3871 do_for_each_ftrace_op(op, ftrace_ops_list) {
3872 if (op->func_hash == &global_ops.local_hash &&
3873 op->flags & FTRACE_OPS_FL_ENABLED) {
3874 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
3875 /* Only need to do this once */
3876 return;
3877 }
3878 } while_for_each_ftrace_op(op);
3879}
3880
3881static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3882 struct ftrace_hash **orig_hash,
3883 struct ftrace_hash *hash,
3884 int enable)
3885{
3886 struct ftrace_ops_hash old_hash_ops;
3887 struct ftrace_hash *old_hash;
3888 int ret;
3889
3890 old_hash = *orig_hash;
3891 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3892 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3893 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3894 if (!ret) {
3895 ftrace_ops_update_code(ops, &old_hash_ops);
3896 free_ftrace_hash_rcu(old_hash);
3897 }
3898 return ret;
3899}
3900
3901static bool module_exists(const char *module)
3902{
3903 /* All modules have the symbol __this_module */
3904 const char this_mod[] = "__this_module";
3905 char modname[MAX_PARAM_PREFIX_LEN + sizeof(this_mod) + 2];
3906 unsigned long val;
3907 int n;
3908
3909 n = snprintf(modname, sizeof(modname), "%s:%s", module, this_mod);
3910
3911 if (n > sizeof(modname) - 1)
3912 return false;
3913
3914 val = module_kallsyms_lookup_name(modname);
3915 return val != 0;
3916}
3917
3918static int cache_mod(struct trace_array *tr,
3919 const char *func, char *module, int enable)
3920{
3921 struct ftrace_mod_load *ftrace_mod, *n;
3922 struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
3923 int ret;
3924
3925 mutex_lock(&ftrace_lock);
3926
3927 /* We do not cache inverse filters */
3928 if (func[0] == '!') {
3929 func++;
3930 ret = -EINVAL;
3931
3932 /* Look to remove this hash */
3933 list_for_each_entry_safe(ftrace_mod, n, head, list) {
3934 if (strcmp(ftrace_mod->module, module) != 0)
3935 continue;
3936
3937 /* no func matches all */
3938 if (strcmp(func, "*") == 0 ||
3939 (ftrace_mod->func &&
3940 strcmp(ftrace_mod->func, func) == 0)) {
3941 ret = 0;
3942 free_ftrace_mod(ftrace_mod);
3943 continue;
3944 }
3945 }
3946 goto out;
3947 }
3948
3949 ret = -EINVAL;
3950 /* We only care about modules that have not been loaded yet */
3951 if (module_exists(module))
3952 goto out;
3953
3954 /* Save this string off, and execute it when the module is loaded */
3955 ret = ftrace_add_mod(tr, func, module, enable);
3956 out:
3957 mutex_unlock(&ftrace_lock);
3958
3959 return ret;
3960}
3961
3962static int
3963ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3964 int reset, int enable);
3965
3966#ifdef CONFIG_MODULES
3967static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
3968 char *mod, bool enable)
3969{
3970 struct ftrace_mod_load *ftrace_mod, *n;
3971 struct ftrace_hash **orig_hash, *new_hash;
3972 LIST_HEAD(process_mods);
3973 char *func;
3974 int ret;
3975
3976 mutex_lock(&ops->func_hash->regex_lock);
3977
3978 if (enable)
3979 orig_hash = &ops->func_hash->filter_hash;
3980 else
3981 orig_hash = &ops->func_hash->notrace_hash;
3982
3983 new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
3984 *orig_hash);
3985 if (!new_hash)
3986 goto out; /* warn? */
3987
3988 mutex_lock(&ftrace_lock);
3989
3990 list_for_each_entry_safe(ftrace_mod, n, head, list) {
3991
3992 if (strcmp(ftrace_mod->module, mod) != 0)
3993 continue;
3994
3995 if (ftrace_mod->func)
3996 func = kstrdup(ftrace_mod->func, GFP_KERNEL);
3997 else
3998 func = kstrdup("*", GFP_KERNEL);
3999
4000 if (!func) /* warn? */
4001 continue;
4002
4003 list_del(&ftrace_mod->list);
4004 list_add(&ftrace_mod->list, &process_mods);
4005
4006 /* Use the newly allocated func, as it may be "*" */
4007 kfree(ftrace_mod->func);
4008 ftrace_mod->func = func;
4009 }
4010
4011 mutex_unlock(&ftrace_lock);
4012
4013 list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
4014
4015 func = ftrace_mod->func;
4016
4017 /* Grabs ftrace_lock, which is why we have this extra step */
4018 match_records(new_hash, func, strlen(func), mod);
4019 free_ftrace_mod(ftrace_mod);
4020 }
4021
4022 if (enable && list_empty(head))
4023 new_hash->flags &= ~FTRACE_HASH_FL_MOD;
4024
4025 mutex_lock(&ftrace_lock);
4026
4027 ret = ftrace_hash_move_and_update_ops(ops, orig_hash,
4028 new_hash, enable);
4029 mutex_unlock(&ftrace_lock);
4030
4031 out:
4032 mutex_unlock(&ops->func_hash->regex_lock);
4033
4034 free_ftrace_hash(new_hash);
4035}
4036
4037static void process_cached_mods(const char *mod_name)
4038{
4039 struct trace_array *tr;
4040 char *mod;
4041
4042 mod = kstrdup(mod_name, GFP_KERNEL);
4043 if (!mod)
4044 return;
4045
4046 mutex_lock(&trace_types_lock);
4047 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
4048 if (!list_empty(&tr->mod_trace))
4049 process_mod_list(&tr->mod_trace, tr->ops, mod, true);
4050 if (!list_empty(&tr->mod_notrace))
4051 process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
4052 }
4053 mutex_unlock(&trace_types_lock);
4054
4055 kfree(mod);
4056}
4057#endif
4058
4059/*
4060 * We register the module command as a template to show others how
4061 * to register the a command as well.
4062 */
4063
4064static int
4065ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
4066 char *func_orig, char *cmd, char *module, int enable)
4067{
4068 char *func;
4069 int ret;
4070
4071 /* match_records() modifies func, and we need the original */
4072 func = kstrdup(func_orig, GFP_KERNEL);
4073 if (!func)
4074 return -ENOMEM;
4075
4076 /*
4077 * cmd == 'mod' because we only registered this func
4078 * for the 'mod' ftrace_func_command.
4079 * But if you register one func with multiple commands,
4080 * you can tell which command was used by the cmd
4081 * parameter.
4082 */
4083 ret = match_records(hash, func, strlen(func), module);
4084 kfree(func);
4085
4086 if (!ret)
4087 return cache_mod(tr, func_orig, module, enable);
4088 if (ret < 0)
4089 return ret;
4090 return 0;
4091}
4092
4093static struct ftrace_func_command ftrace_mod_cmd = {
4094 .name = "mod",
4095 .func = ftrace_mod_callback,
4096};
4097
4098static int __init ftrace_mod_cmd_init(void)
4099{
4100 return register_ftrace_command(&ftrace_mod_cmd);
4101}
4102core_initcall(ftrace_mod_cmd_init);
4103
4104static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
4105 struct ftrace_ops *op, struct pt_regs *pt_regs)
4106{
4107 struct ftrace_probe_ops *probe_ops;
4108 struct ftrace_func_probe *probe;
4109
4110 probe = container_of(op, struct ftrace_func_probe, ops);
4111 probe_ops = probe->probe_ops;
4112
4113 /*
4114 * Disable preemption for these calls to prevent a RCU grace
4115 * period. This syncs the hash iteration and freeing of items
4116 * on the hash. rcu_read_lock is too dangerous here.
4117 */
4118 preempt_disable_notrace();
4119 probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
4120 preempt_enable_notrace();
4121}
4122
4123struct ftrace_func_map {
4124 struct ftrace_func_entry entry;
4125 void *data;
4126};
4127
4128struct ftrace_func_mapper {
4129 struct ftrace_hash hash;
4130};
4131
4132/**
4133 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
4134 *
4135 * Returns a ftrace_func_mapper descriptor that can be used to map ips to data.
4136 */
4137struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
4138{
4139 struct ftrace_hash *hash;
4140
4141 /*
4142 * The mapper is simply a ftrace_hash, but since the entries
4143 * in the hash are not ftrace_func_entry type, we define it
4144 * as a separate structure.
4145 */
4146 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4147 return (struct ftrace_func_mapper *)hash;
4148}
4149
4150/**
4151 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
4152 * @mapper: The mapper that has the ip maps
4153 * @ip: the instruction pointer to find the data for
4154 *
4155 * Returns the data mapped to @ip if found otherwise NULL. The return
4156 * is actually the address of the mapper data pointer. The address is
4157 * returned for use cases where the data is no bigger than a long, and
4158 * the user can use the data pointer as its data instead of having to
4159 * allocate more memory for the reference.
4160 */
4161void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
4162 unsigned long ip)
4163{
4164 struct ftrace_func_entry *entry;
4165 struct ftrace_func_map *map;
4166
4167 entry = ftrace_lookup_ip(&mapper->hash, ip);
4168 if (!entry)
4169 return NULL;
4170
4171 map = (struct ftrace_func_map *)entry;
4172 return &map->data;
4173}
4174
4175/**
4176 * ftrace_func_mapper_add_ip - Map some data to an ip
4177 * @mapper: The mapper that has the ip maps
4178 * @ip: The instruction pointer address to map @data to
4179 * @data: The data to map to @ip
4180 *
4181 * Returns 0 on succes otherwise an error.
4182 */
4183int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
4184 unsigned long ip, void *data)
4185{
4186 struct ftrace_func_entry *entry;
4187 struct ftrace_func_map *map;
4188
4189 entry = ftrace_lookup_ip(&mapper->hash, ip);
4190 if (entry)
4191 return -EBUSY;
4192
4193 map = kmalloc(sizeof(*map), GFP_KERNEL);
4194 if (!map)
4195 return -ENOMEM;
4196
4197 map->entry.ip = ip;
4198 map->data = data;
4199
4200 __add_hash_entry(&mapper->hash, &map->entry);
4201
4202 return 0;
4203}
4204
4205/**
4206 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
4207 * @mapper: The mapper that has the ip maps
4208 * @ip: The instruction pointer address to remove the data from
4209 *
4210 * Returns the data if it is found, otherwise NULL.
4211 * Note, if the data pointer is used as the data itself, (see
4212 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
4213 * if the data pointer was set to zero.
4214 */
4215void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
4216 unsigned long ip)
4217{
4218 struct ftrace_func_entry *entry;
4219 struct ftrace_func_map *map;
4220 void *data;
4221
4222 entry = ftrace_lookup_ip(&mapper->hash, ip);
4223 if (!entry)
4224 return NULL;
4225
4226 map = (struct ftrace_func_map *)entry;
4227 data = map->data;
4228
4229 remove_hash_entry(&mapper->hash, entry);
4230 kfree(entry);
4231
4232 return data;
4233}
4234
4235/**
4236 * free_ftrace_func_mapper - free a mapping of ips and data
4237 * @mapper: The mapper that has the ip maps
4238 * @free_func: A function to be called on each data item.
4239 *
4240 * This is used to free the function mapper. The @free_func is optional
4241 * and can be used if the data needs to be freed as well.
4242 */
4243void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
4244 ftrace_mapper_func free_func)
4245{
4246 struct ftrace_func_entry *entry;
4247 struct ftrace_func_map *map;
4248 struct hlist_head *hhd;
4249 int size = 1 << mapper->hash.size_bits;
4250 int i;
4251
4252 if (free_func && mapper->hash.count) {
4253 for (i = 0; i < size; i++) {
4254 hhd = &mapper->hash.buckets[i];
4255 hlist_for_each_entry(entry, hhd, hlist) {
4256 map = (struct ftrace_func_map *)entry;
4257 free_func(map);
4258 }
4259 }
4260 }
4261 free_ftrace_hash(&mapper->hash);
4262}
4263
4264static void release_probe(struct ftrace_func_probe *probe)
4265{
4266 struct ftrace_probe_ops *probe_ops;
4267
4268 mutex_lock(&ftrace_lock);
4269
4270 WARN_ON(probe->ref <= 0);
4271
4272 /* Subtract the ref that was used to protect this instance */
4273 probe->ref--;
4274
4275 if (!probe->ref) {
4276 probe_ops = probe->probe_ops;
4277 /*
4278 * Sending zero as ip tells probe_ops to free
4279 * the probe->data itself
4280 */
4281 if (probe_ops->free)
4282 probe_ops->free(probe_ops, probe->tr, 0, probe->data);
4283 list_del(&probe->list);
4284 kfree(probe);
4285 }
4286 mutex_unlock(&ftrace_lock);
4287}
4288
4289static void acquire_probe_locked(struct ftrace_func_probe *probe)
4290{
4291 /*
4292 * Add one ref to keep it from being freed when releasing the
4293 * ftrace_lock mutex.
4294 */
4295 probe->ref++;
4296}
4297
4298int
4299register_ftrace_function_probe(char *glob, struct trace_array *tr,
4300 struct ftrace_probe_ops *probe_ops,
4301 void *data)
4302{
4303 struct ftrace_func_entry *entry;
4304 struct ftrace_func_probe *probe;
4305 struct ftrace_hash **orig_hash;
4306 struct ftrace_hash *old_hash;
4307 struct ftrace_hash *hash;
4308 int count = 0;
4309 int size;
4310 int ret;
4311 int i;
4312
4313 if (WARN_ON(!tr))
4314 return -EINVAL;
4315
4316 /* We do not support '!' for function probes */
4317 if (WARN_ON(glob[0] == '!'))
4318 return -EINVAL;
4319
4320
4321 mutex_lock(&ftrace_lock);
4322 /* Check if the probe_ops is already registered */
4323 list_for_each_entry(probe, &tr->func_probes, list) {
4324 if (probe->probe_ops == probe_ops)
4325 break;
4326 }
4327 if (&probe->list == &tr->func_probes) {
4328 probe = kzalloc(sizeof(*probe), GFP_KERNEL);
4329 if (!probe) {
4330 mutex_unlock(&ftrace_lock);
4331 return -ENOMEM;
4332 }
4333 probe->probe_ops = probe_ops;
4334 probe->ops.func = function_trace_probe_call;
4335 probe->tr = tr;
4336 ftrace_ops_init(&probe->ops);
4337 list_add(&probe->list, &tr->func_probes);
4338 }
4339
4340 acquire_probe_locked(probe);
4341
4342 mutex_unlock(&ftrace_lock);
4343
4344 mutex_lock(&probe->ops.func_hash->regex_lock);
4345
4346 orig_hash = &probe->ops.func_hash->filter_hash;
4347 old_hash = *orig_hash;
4348 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4349
4350 ret = ftrace_match_records(hash, glob, strlen(glob));
4351
4352 /* Nothing found? */
4353 if (!ret)
4354 ret = -EINVAL;
4355
4356 if (ret < 0)
4357 goto out;
4358
4359 size = 1 << hash->size_bits;
4360 for (i = 0; i < size; i++) {
4361 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4362 if (ftrace_lookup_ip(old_hash, entry->ip))
4363 continue;
4364 /*
4365 * The caller might want to do something special
4366 * for each function we find. We call the callback
4367 * to give the caller an opportunity to do so.
4368 */
4369 if (probe_ops->init) {
4370 ret = probe_ops->init(probe_ops, tr,
4371 entry->ip, data,
4372 &probe->data);
4373 if (ret < 0) {
4374 if (probe_ops->free && count)
4375 probe_ops->free(probe_ops, tr,
4376 0, probe->data);
4377 probe->data = NULL;
4378 goto out;
4379 }
4380 }
4381 count++;
4382 }
4383 }
4384
4385 mutex_lock(&ftrace_lock);
4386
4387 if (!count) {
4388 /* Nothing was added? */
4389 ret = -EINVAL;
4390 goto out_unlock;
4391 }
4392
4393 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4394 hash, 1);
4395 if (ret < 0)
4396 goto err_unlock;
4397
4398 /* One ref for each new function traced */
4399 probe->ref += count;
4400
4401 if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
4402 ret = ftrace_startup(&probe->ops, 0);
4403
4404 out_unlock:
4405 mutex_unlock(&ftrace_lock);
4406
4407 if (!ret)
4408 ret = count;
4409 out:
4410 mutex_unlock(&probe->ops.func_hash->regex_lock);
4411 free_ftrace_hash(hash);
4412
4413 release_probe(probe);
4414
4415 return ret;
4416
4417 err_unlock:
4418 if (!probe_ops->free || !count)
4419 goto out_unlock;
4420
4421 /* Failed to do the move, need to call the free functions */
4422 for (i = 0; i < size; i++) {
4423 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4424 if (ftrace_lookup_ip(old_hash, entry->ip))
4425 continue;
4426 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4427 }
4428 }
4429 goto out_unlock;
4430}
4431
4432int
4433unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
4434 struct ftrace_probe_ops *probe_ops)
4435{
4436 struct ftrace_ops_hash old_hash_ops;
4437 struct ftrace_func_entry *entry;
4438 struct ftrace_func_probe *probe;
4439 struct ftrace_glob func_g;
4440 struct ftrace_hash **orig_hash;
4441 struct ftrace_hash *old_hash;
4442 struct ftrace_hash *hash = NULL;
4443 struct hlist_node *tmp;
4444 struct hlist_head hhd;
4445 char str[KSYM_SYMBOL_LEN];
4446 int count = 0;
4447 int i, ret = -ENODEV;
4448 int size;
4449
4450 if (!glob || !strlen(glob) || !strcmp(glob, "*"))
4451 func_g.search = NULL;
4452 else {
4453 int not;
4454
4455 func_g.type = filter_parse_regex(glob, strlen(glob),
4456 &func_g.search, ¬);
4457 func_g.len = strlen(func_g.search);
4458
4459 /* we do not support '!' for function probes */
4460 if (WARN_ON(not))
4461 return -EINVAL;
4462 }
4463
4464 mutex_lock(&ftrace_lock);
4465 /* Check if the probe_ops is already registered */
4466 list_for_each_entry(probe, &tr->func_probes, list) {
4467 if (probe->probe_ops == probe_ops)
4468 break;
4469 }
4470 if (&probe->list == &tr->func_probes)
4471 goto err_unlock_ftrace;
4472
4473 ret = -EINVAL;
4474 if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
4475 goto err_unlock_ftrace;
4476
4477 acquire_probe_locked(probe);
4478
4479 mutex_unlock(&ftrace_lock);
4480
4481 mutex_lock(&probe->ops.func_hash->regex_lock);
4482
4483 orig_hash = &probe->ops.func_hash->filter_hash;
4484 old_hash = *orig_hash;
4485
4486 if (ftrace_hash_empty(old_hash))
4487 goto out_unlock;
4488
4489 old_hash_ops.filter_hash = old_hash;
4490 /* Probes only have filters */
4491 old_hash_ops.notrace_hash = NULL;
4492
4493 ret = -ENOMEM;
4494 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4495 if (!hash)
4496 goto out_unlock;
4497
4498 INIT_HLIST_HEAD(&hhd);
4499
4500 size = 1 << hash->size_bits;
4501 for (i = 0; i < size; i++) {
4502 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
4503
4504 if (func_g.search) {
4505 kallsyms_lookup(entry->ip, NULL, NULL,
4506 NULL, str);
4507 if (!ftrace_match(str, &func_g))
4508 continue;
4509 }
4510 count++;
4511 remove_hash_entry(hash, entry);
4512 hlist_add_head(&entry->hlist, &hhd);
4513 }
4514 }
4515
4516 /* Nothing found? */
4517 if (!count) {
4518 ret = -EINVAL;
4519 goto out_unlock;
4520 }
4521
4522 mutex_lock(&ftrace_lock);
4523
4524 WARN_ON(probe->ref < count);
4525
4526 probe->ref -= count;
4527
4528 if (ftrace_hash_empty(hash))
4529 ftrace_shutdown(&probe->ops, 0);
4530
4531 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4532 hash, 1);
4533
4534 /* still need to update the function call sites */
4535 if (ftrace_enabled && !ftrace_hash_empty(hash))
4536 ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
4537 &old_hash_ops);
4538 synchronize_sched();
4539
4540 hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
4541 hlist_del(&entry->hlist);
4542 if (probe_ops->free)
4543 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4544 kfree(entry);
4545 }
4546 mutex_unlock(&ftrace_lock);
4547
4548 out_unlock:
4549 mutex_unlock(&probe->ops.func_hash->regex_lock);
4550 free_ftrace_hash(hash);
4551
4552 release_probe(probe);
4553
4554 return ret;
4555
4556 err_unlock_ftrace:
4557 mutex_unlock(&ftrace_lock);
4558 return ret;
4559}
4560
4561void clear_ftrace_function_probes(struct trace_array *tr)
4562{
4563 struct ftrace_func_probe *probe, *n;
4564
4565 list_for_each_entry_safe(probe, n, &tr->func_probes, list)
4566 unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
4567}
4568
4569static LIST_HEAD(ftrace_commands);
4570static DEFINE_MUTEX(ftrace_cmd_mutex);
4571
4572/*
4573 * Currently we only register ftrace commands from __init, so mark this
4574 * __init too.
4575 */
4576__init int register_ftrace_command(struct ftrace_func_command *cmd)
4577{
4578 struct ftrace_func_command *p;
4579 int ret = 0;
4580
4581 mutex_lock(&ftrace_cmd_mutex);
4582 list_for_each_entry(p, &ftrace_commands, list) {
4583 if (strcmp(cmd->name, p->name) == 0) {
4584 ret = -EBUSY;
4585 goto out_unlock;
4586 }
4587 }
4588 list_add(&cmd->list, &ftrace_commands);
4589 out_unlock:
4590 mutex_unlock(&ftrace_cmd_mutex);
4591
4592 return ret;
4593}
4594
4595/*
4596 * Currently we only unregister ftrace commands from __init, so mark
4597 * this __init too.
4598 */
4599__init int unregister_ftrace_command(struct ftrace_func_command *cmd)
4600{
4601 struct ftrace_func_command *p, *n;
4602 int ret = -ENODEV;
4603
4604 mutex_lock(&ftrace_cmd_mutex);
4605 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4606 if (strcmp(cmd->name, p->name) == 0) {
4607 ret = 0;
4608 list_del_init(&p->list);
4609 goto out_unlock;
4610 }
4611 }
4612 out_unlock:
4613 mutex_unlock(&ftrace_cmd_mutex);
4614
4615 return ret;
4616}
4617
4618static int ftrace_process_regex(struct ftrace_iterator *iter,
4619 char *buff, int len, int enable)
4620{
4621 struct ftrace_hash *hash = iter->hash;
4622 struct trace_array *tr = iter->ops->private;
4623 char *func, *command, *next = buff;
4624 struct ftrace_func_command *p;
4625 int ret = -EINVAL;
4626
4627 func = strsep(&next, ":");
4628
4629 if (!next) {
4630 ret = ftrace_match_records(hash, func, len);
4631 if (!ret)
4632 ret = -EINVAL;
4633 if (ret < 0)
4634 return ret;
4635 return 0;
4636 }
4637
4638 /* command found */
4639
4640 command = strsep(&next, ":");
4641
4642 mutex_lock(&ftrace_cmd_mutex);
4643 list_for_each_entry(p, &ftrace_commands, list) {
4644 if (strcmp(p->name, command) == 0) {
4645 ret = p->func(tr, hash, func, command, next, enable);
4646 goto out_unlock;
4647 }
4648 }
4649 out_unlock:
4650 mutex_unlock(&ftrace_cmd_mutex);
4651
4652 return ret;
4653}
4654
4655static ssize_t
4656ftrace_regex_write(struct file *file, const char __user *ubuf,
4657 size_t cnt, loff_t *ppos, int enable)
4658{
4659 struct ftrace_iterator *iter;
4660 struct trace_parser *parser;
4661 ssize_t ret, read;
4662
4663 if (!cnt)
4664 return 0;
4665
4666 if (file->f_mode & FMODE_READ) {
4667 struct seq_file *m = file->private_data;
4668 iter = m->private;
4669 } else
4670 iter = file->private_data;
4671
4672 if (unlikely(ftrace_disabled))
4673 return -ENODEV;
4674
4675 /* iter->hash is a local copy, so we don't need regex_lock */
4676
4677 parser = &iter->parser;
4678 read = trace_get_user(parser, ubuf, cnt, ppos);
4679
4680 if (read >= 0 && trace_parser_loaded(parser) &&
4681 !trace_parser_cont(parser)) {
4682 ret = ftrace_process_regex(iter, parser->buffer,
4683 parser->idx, enable);
4684 trace_parser_clear(parser);
4685 if (ret < 0)
4686 goto out;
4687 }
4688
4689 ret = read;
4690 out:
4691 return ret;
4692}
4693
4694ssize_t
4695ftrace_filter_write(struct file *file, const char __user *ubuf,
4696 size_t cnt, loff_t *ppos)
4697{
4698 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4699}
4700
4701ssize_t
4702ftrace_notrace_write(struct file *file, const char __user *ubuf,
4703 size_t cnt, loff_t *ppos)
4704{
4705 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4706}
4707
4708static int
4709ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4710{
4711 struct ftrace_func_entry *entry;
4712
4713 if (!ftrace_location(ip))
4714 return -EINVAL;
4715
4716 if (remove) {
4717 entry = ftrace_lookup_ip(hash, ip);
4718 if (!entry)
4719 return -ENOENT;
4720 free_hash_entry(hash, entry);
4721 return 0;
4722 }
4723
4724 return add_hash_entry(hash, ip);
4725}
4726
4727static int
4728ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4729 unsigned long ip, int remove, int reset, int enable)
4730{
4731 struct ftrace_hash **orig_hash;
4732 struct ftrace_hash *hash;
4733 int ret;
4734
4735 if (unlikely(ftrace_disabled))
4736 return -ENODEV;
4737
4738 mutex_lock(&ops->func_hash->regex_lock);
4739
4740 if (enable)
4741 orig_hash = &ops->func_hash->filter_hash;
4742 else
4743 orig_hash = &ops->func_hash->notrace_hash;
4744
4745 if (reset)
4746 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4747 else
4748 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4749
4750 if (!hash) {
4751 ret = -ENOMEM;
4752 goto out_regex_unlock;
4753 }
4754
4755 if (buf && !ftrace_match_records(hash, buf, len)) {
4756 ret = -EINVAL;
4757 goto out_regex_unlock;
4758 }
4759 if (ip) {
4760 ret = ftrace_match_addr(hash, ip, remove);
4761 if (ret < 0)
4762 goto out_regex_unlock;
4763 }
4764
4765 mutex_lock(&ftrace_lock);
4766 ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4767 mutex_unlock(&ftrace_lock);
4768
4769 out_regex_unlock:
4770 mutex_unlock(&ops->func_hash->regex_lock);
4771
4772 free_ftrace_hash(hash);
4773 return ret;
4774}
4775
4776static int
4777ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4778 int reset, int enable)
4779{
4780 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4781}
4782
4783/**
4784 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4785 * @ops - the ops to set the filter with
4786 * @ip - the address to add to or remove from the filter.
4787 * @remove - non zero to remove the ip from the filter
4788 * @reset - non zero to reset all filters before applying this filter.
4789 *
4790 * Filters denote which functions should be enabled when tracing is enabled
4791 * If @ip is NULL, it failes to update filter.
4792 */
4793int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4794 int remove, int reset)
4795{
4796 ftrace_ops_init(ops);
4797 return ftrace_set_addr(ops, ip, remove, reset, 1);
4798}
4799EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4800
4801/**
4802 * ftrace_ops_set_global_filter - setup ops to use global filters
4803 * @ops - the ops which will use the global filters
4804 *
4805 * ftrace users who need global function trace filtering should call this.
4806 * It can set the global filter only if ops were not initialized before.
4807 */
4808void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
4809{
4810 if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
4811 return;
4812
4813 ftrace_ops_init(ops);
4814 ops->func_hash = &global_ops.local_hash;
4815}
4816EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
4817
4818static int
4819ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4820 int reset, int enable)
4821{
4822 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4823}
4824
4825/**
4826 * ftrace_set_filter - set a function to filter on in ftrace
4827 * @ops - the ops to set the filter with
4828 * @buf - the string that holds the function filter text.
4829 * @len - the length of the string.
4830 * @reset - non zero to reset all filters before applying this filter.
4831 *
4832 * Filters denote which functions should be enabled when tracing is enabled.
4833 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4834 */
4835int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4836 int len, int reset)
4837{
4838 ftrace_ops_init(ops);
4839 return ftrace_set_regex(ops, buf, len, reset, 1);
4840}
4841EXPORT_SYMBOL_GPL(ftrace_set_filter);
4842
4843/**
4844 * ftrace_set_notrace - set a function to not trace in ftrace
4845 * @ops - the ops to set the notrace filter with
4846 * @buf - the string that holds the function notrace text.
4847 * @len - the length of the string.
4848 * @reset - non zero to reset all filters before applying this filter.
4849 *
4850 * Notrace Filters denote which functions should not be enabled when tracing
4851 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4852 * for tracing.
4853 */
4854int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4855 int len, int reset)
4856{
4857 ftrace_ops_init(ops);
4858 return ftrace_set_regex(ops, buf, len, reset, 0);
4859}
4860EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4861/**
4862 * ftrace_set_global_filter - set a function to filter on with global tracers
4863 * @buf - the string that holds the function filter text.
4864 * @len - the length of the string.
4865 * @reset - non zero to reset all filters before applying this filter.
4866 *
4867 * Filters denote which functions should be enabled when tracing is enabled.
4868 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4869 */
4870void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4871{
4872 ftrace_set_regex(&global_ops, buf, len, reset, 1);
4873}
4874EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4875
4876/**
4877 * ftrace_set_global_notrace - set a function to not trace with global tracers
4878 * @buf - the string that holds the function notrace text.
4879 * @len - the length of the string.
4880 * @reset - non zero to reset all filters before applying this filter.
4881 *
4882 * Notrace Filters denote which functions should not be enabled when tracing
4883 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4884 * for tracing.
4885 */
4886void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4887{
4888 ftrace_set_regex(&global_ops, buf, len, reset, 0);
4889}
4890EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4891
4892/*
4893 * command line interface to allow users to set filters on boot up.
4894 */
4895#define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
4896static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4897static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4898
4899/* Used by function selftest to not test if filter is set */
4900bool ftrace_filter_param __initdata;
4901
4902static int __init set_ftrace_notrace(char *str)
4903{
4904 ftrace_filter_param = true;
4905 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4906 return 1;
4907}
4908__setup("ftrace_notrace=", set_ftrace_notrace);
4909
4910static int __init set_ftrace_filter(char *str)
4911{
4912 ftrace_filter_param = true;
4913 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4914 return 1;
4915}
4916__setup("ftrace_filter=", set_ftrace_filter);
4917
4918#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4919static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4920static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4921static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
4922
4923static int __init set_graph_function(char *str)
4924{
4925 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4926 return 1;
4927}
4928__setup("ftrace_graph_filter=", set_graph_function);
4929
4930static int __init set_graph_notrace_function(char *str)
4931{
4932 strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4933 return 1;
4934}
4935__setup("ftrace_graph_notrace=", set_graph_notrace_function);
4936
4937static int __init set_graph_max_depth_function(char *str)
4938{
4939 if (!str)
4940 return 0;
4941 fgraph_max_depth = simple_strtoul(str, NULL, 0);
4942 return 1;
4943}
4944__setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
4945
4946static void __init set_ftrace_early_graph(char *buf, int enable)
4947{
4948 int ret;
4949 char *func;
4950 struct ftrace_hash *hash;
4951
4952 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4953 if (WARN_ON(!hash))
4954 return;
4955
4956 while (buf) {
4957 func = strsep(&buf, ",");
4958 /* we allow only one expression at a time */
4959 ret = ftrace_graph_set_hash(hash, func);
4960 if (ret)
4961 printk(KERN_DEBUG "ftrace: function %s not "
4962 "traceable\n", func);
4963 }
4964
4965 if (enable)
4966 ftrace_graph_hash = hash;
4967 else
4968 ftrace_graph_notrace_hash = hash;
4969}
4970#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4971
4972void __init
4973ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
4974{
4975 char *func;
4976
4977 ftrace_ops_init(ops);
4978
4979 while (buf) {
4980 func = strsep(&buf, ",");
4981 ftrace_set_regex(ops, func, strlen(func), 0, enable);
4982 }
4983}
4984
4985static void __init set_ftrace_early_filters(void)
4986{
4987 if (ftrace_filter_buf[0])
4988 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
4989 if (ftrace_notrace_buf[0])
4990 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
4991#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4992 if (ftrace_graph_buf[0])
4993 set_ftrace_early_graph(ftrace_graph_buf, 1);
4994 if (ftrace_graph_notrace_buf[0])
4995 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
4996#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4997}
4998
4999int ftrace_regex_release(struct inode *inode, struct file *file)
5000{
5001 struct seq_file *m = (struct seq_file *)file->private_data;
5002 struct ftrace_iterator *iter;
5003 struct ftrace_hash **orig_hash;
5004 struct trace_parser *parser;
5005 int filter_hash;
5006 int ret;
5007
5008 if (file->f_mode & FMODE_READ) {
5009 iter = m->private;
5010 seq_release(inode, file);
5011 } else
5012 iter = file->private_data;
5013
5014 parser = &iter->parser;
5015 if (trace_parser_loaded(parser)) {
5016 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
5017 }
5018
5019 trace_parser_put(parser);
5020
5021 mutex_lock(&iter->ops->func_hash->regex_lock);
5022
5023 if (file->f_mode & FMODE_WRITE) {
5024 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
5025
5026 if (filter_hash) {
5027 orig_hash = &iter->ops->func_hash->filter_hash;
5028 if (iter->tr && !list_empty(&iter->tr->mod_trace))
5029 iter->hash->flags |= FTRACE_HASH_FL_MOD;
5030 } else
5031 orig_hash = &iter->ops->func_hash->notrace_hash;
5032
5033 mutex_lock(&ftrace_lock);
5034 ret = ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
5035 iter->hash, filter_hash);
5036 mutex_unlock(&ftrace_lock);
5037 } else {
5038 /* For read only, the hash is the ops hash */
5039 iter->hash = NULL;
5040 }
5041
5042 mutex_unlock(&iter->ops->func_hash->regex_lock);
5043 free_ftrace_hash(iter->hash);
5044 kfree(iter);
5045
5046 return 0;
5047}
5048
5049static const struct file_operations ftrace_avail_fops = {
5050 .open = ftrace_avail_open,
5051 .read = seq_read,
5052 .llseek = seq_lseek,
5053 .release = seq_release_private,
5054};
5055
5056static const struct file_operations ftrace_enabled_fops = {
5057 .open = ftrace_enabled_open,
5058 .read = seq_read,
5059 .llseek = seq_lseek,
5060 .release = seq_release_private,
5061};
5062
5063static const struct file_operations ftrace_filter_fops = {
5064 .open = ftrace_filter_open,
5065 .read = seq_read,
5066 .write = ftrace_filter_write,
5067 .llseek = tracing_lseek,
5068 .release = ftrace_regex_release,
5069};
5070
5071static const struct file_operations ftrace_notrace_fops = {
5072 .open = ftrace_notrace_open,
5073 .read = seq_read,
5074 .write = ftrace_notrace_write,
5075 .llseek = tracing_lseek,
5076 .release = ftrace_regex_release,
5077};
5078
5079#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5080
5081static DEFINE_MUTEX(graph_lock);
5082
5083struct ftrace_hash *ftrace_graph_hash = EMPTY_HASH;
5084struct ftrace_hash *ftrace_graph_notrace_hash = EMPTY_HASH;
5085
5086enum graph_filter_type {
5087 GRAPH_FILTER_NOTRACE = 0,
5088 GRAPH_FILTER_FUNCTION,
5089};
5090
5091#define FTRACE_GRAPH_EMPTY ((void *)1)
5092
5093struct ftrace_graph_data {
5094 struct ftrace_hash *hash;
5095 struct ftrace_func_entry *entry;
5096 int idx; /* for hash table iteration */
5097 enum graph_filter_type type;
5098 struct ftrace_hash *new_hash;
5099 const struct seq_operations *seq_ops;
5100 struct trace_parser parser;
5101};
5102
5103static void *
5104__g_next(struct seq_file *m, loff_t *pos)
5105{
5106 struct ftrace_graph_data *fgd = m->private;
5107 struct ftrace_func_entry *entry = fgd->entry;
5108 struct hlist_head *head;
5109 int i, idx = fgd->idx;
5110
5111 if (*pos >= fgd->hash->count)
5112 return NULL;
5113
5114 if (entry) {
5115 hlist_for_each_entry_continue(entry, hlist) {
5116 fgd->entry = entry;
5117 return entry;
5118 }
5119
5120 idx++;
5121 }
5122
5123 for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
5124 head = &fgd->hash->buckets[i];
5125 hlist_for_each_entry(entry, head, hlist) {
5126 fgd->entry = entry;
5127 fgd->idx = i;
5128 return entry;
5129 }
5130 }
5131 return NULL;
5132}
5133
5134static void *
5135g_next(struct seq_file *m, void *v, loff_t *pos)
5136{
5137 (*pos)++;
5138 return __g_next(m, pos);
5139}
5140
5141static void *g_start(struct seq_file *m, loff_t *pos)
5142{
5143 struct ftrace_graph_data *fgd = m->private;
5144
5145 mutex_lock(&graph_lock);
5146
5147 if (fgd->type == GRAPH_FILTER_FUNCTION)
5148 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5149 lockdep_is_held(&graph_lock));
5150 else
5151 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5152 lockdep_is_held(&graph_lock));
5153
5154 /* Nothing, tell g_show to print all functions are enabled */
5155 if (ftrace_hash_empty(fgd->hash) && !*pos)
5156 return FTRACE_GRAPH_EMPTY;
5157
5158 fgd->idx = 0;
5159 fgd->entry = NULL;
5160 return __g_next(m, pos);
5161}
5162
5163static void g_stop(struct seq_file *m, void *p)
5164{
5165 mutex_unlock(&graph_lock);
5166}
5167
5168static int g_show(struct seq_file *m, void *v)
5169{
5170 struct ftrace_func_entry *entry = v;
5171
5172 if (!entry)
5173 return 0;
5174
5175 if (entry == FTRACE_GRAPH_EMPTY) {
5176 struct ftrace_graph_data *fgd = m->private;
5177
5178 if (fgd->type == GRAPH_FILTER_FUNCTION)
5179 seq_puts(m, "#### all functions enabled ####\n");
5180 else
5181 seq_puts(m, "#### no functions disabled ####\n");
5182 return 0;
5183 }
5184
5185 seq_printf(m, "%ps\n", (void *)entry->ip);
5186
5187 return 0;
5188}
5189
5190static const struct seq_operations ftrace_graph_seq_ops = {
5191 .start = g_start,
5192 .next = g_next,
5193 .stop = g_stop,
5194 .show = g_show,
5195};
5196
5197static int
5198__ftrace_graph_open(struct inode *inode, struct file *file,
5199 struct ftrace_graph_data *fgd)
5200{
5201 int ret = 0;
5202 struct ftrace_hash *new_hash = NULL;
5203
5204 if (file->f_mode & FMODE_WRITE) {
5205 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
5206
5207 if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
5208 return -ENOMEM;
5209
5210 if (file->f_flags & O_TRUNC)
5211 new_hash = alloc_ftrace_hash(size_bits);
5212 else
5213 new_hash = alloc_and_copy_ftrace_hash(size_bits,
5214 fgd->hash);
5215 if (!new_hash) {
5216 ret = -ENOMEM;
5217 goto out;
5218 }
5219 }
5220
5221 if (file->f_mode & FMODE_READ) {
5222 ret = seq_open(file, &ftrace_graph_seq_ops);
5223 if (!ret) {
5224 struct seq_file *m = file->private_data;
5225 m->private = fgd;
5226 } else {
5227 /* Failed */
5228 free_ftrace_hash(new_hash);
5229 new_hash = NULL;
5230 }
5231 } else
5232 file->private_data = fgd;
5233
5234out:
5235 if (ret < 0 && file->f_mode & FMODE_WRITE)
5236 trace_parser_put(&fgd->parser);
5237
5238 fgd->new_hash = new_hash;
5239
5240 /*
5241 * All uses of fgd->hash must be taken with the graph_lock
5242 * held. The graph_lock is going to be released, so force
5243 * fgd->hash to be reinitialized when it is taken again.
5244 */
5245 fgd->hash = NULL;
5246
5247 return ret;
5248}
5249
5250static int
5251ftrace_graph_open(struct inode *inode, struct file *file)
5252{
5253 struct ftrace_graph_data *fgd;
5254 int ret;
5255
5256 if (unlikely(ftrace_disabled))
5257 return -ENODEV;
5258
5259 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5260 if (fgd == NULL)
5261 return -ENOMEM;
5262
5263 mutex_lock(&graph_lock);
5264
5265 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5266 lockdep_is_held(&graph_lock));
5267 fgd->type = GRAPH_FILTER_FUNCTION;
5268 fgd->seq_ops = &ftrace_graph_seq_ops;
5269
5270 ret = __ftrace_graph_open(inode, file, fgd);
5271 if (ret < 0)
5272 kfree(fgd);
5273
5274 mutex_unlock(&graph_lock);
5275 return ret;
5276}
5277
5278static int
5279ftrace_graph_notrace_open(struct inode *inode, struct file *file)
5280{
5281 struct ftrace_graph_data *fgd;
5282 int ret;
5283
5284 if (unlikely(ftrace_disabled))
5285 return -ENODEV;
5286
5287 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5288 if (fgd == NULL)
5289 return -ENOMEM;
5290
5291 mutex_lock(&graph_lock);
5292
5293 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5294 lockdep_is_held(&graph_lock));
5295 fgd->type = GRAPH_FILTER_NOTRACE;
5296 fgd->seq_ops = &ftrace_graph_seq_ops;
5297
5298 ret = __ftrace_graph_open(inode, file, fgd);
5299 if (ret < 0)
5300 kfree(fgd);
5301
5302 mutex_unlock(&graph_lock);
5303 return ret;
5304}
5305
5306static int
5307ftrace_graph_release(struct inode *inode, struct file *file)
5308{
5309 struct ftrace_graph_data *fgd;
5310 struct ftrace_hash *old_hash, *new_hash;
5311 struct trace_parser *parser;
5312 int ret = 0;
5313
5314 if (file->f_mode & FMODE_READ) {
5315 struct seq_file *m = file->private_data;
5316
5317 fgd = m->private;
5318 seq_release(inode, file);
5319 } else {
5320 fgd = file->private_data;
5321 }
5322
5323
5324 if (file->f_mode & FMODE_WRITE) {
5325
5326 parser = &fgd->parser;
5327
5328 if (trace_parser_loaded((parser))) {
5329 ret = ftrace_graph_set_hash(fgd->new_hash,
5330 parser->buffer);
5331 }
5332
5333 trace_parser_put(parser);
5334
5335 new_hash = __ftrace_hash_move(fgd->new_hash);
5336 if (!new_hash) {
5337 ret = -ENOMEM;
5338 goto out;
5339 }
5340
5341 mutex_lock(&graph_lock);
5342
5343 if (fgd->type == GRAPH_FILTER_FUNCTION) {
5344 old_hash = rcu_dereference_protected(ftrace_graph_hash,
5345 lockdep_is_held(&graph_lock));
5346 rcu_assign_pointer(ftrace_graph_hash, new_hash);
5347 } else {
5348 old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5349 lockdep_is_held(&graph_lock));
5350 rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
5351 }
5352
5353 mutex_unlock(&graph_lock);
5354
5355 /* Wait till all users are no longer using the old hash */
5356 synchronize_sched();
5357
5358 free_ftrace_hash(old_hash);
5359 }
5360
5361 out:
5362 free_ftrace_hash(fgd->new_hash);
5363 kfree(fgd);
5364
5365 return ret;
5366}
5367
5368static int
5369ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
5370{
5371 struct ftrace_glob func_g;
5372 struct dyn_ftrace *rec;
5373 struct ftrace_page *pg;
5374 struct ftrace_func_entry *entry;
5375 int fail = 1;
5376 int not;
5377
5378 /* decode regex */
5379 func_g.type = filter_parse_regex(buffer, strlen(buffer),
5380 &func_g.search, ¬);
5381
5382 func_g.len = strlen(func_g.search);
5383
5384 mutex_lock(&ftrace_lock);
5385
5386 if (unlikely(ftrace_disabled)) {
5387 mutex_unlock(&ftrace_lock);
5388 return -ENODEV;
5389 }
5390
5391 do_for_each_ftrace_rec(pg, rec) {
5392
5393 if (rec->flags & FTRACE_FL_DISABLED)
5394 continue;
5395
5396 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
5397 entry = ftrace_lookup_ip(hash, rec->ip);
5398
5399 if (!not) {
5400 fail = 0;
5401
5402 if (entry)
5403 continue;
5404 if (add_hash_entry(hash, rec->ip) < 0)
5405 goto out;
5406 } else {
5407 if (entry) {
5408 free_hash_entry(hash, entry);
5409 fail = 0;
5410 }
5411 }
5412 }
5413 } while_for_each_ftrace_rec();
5414out:
5415 mutex_unlock(&ftrace_lock);
5416
5417 if (fail)
5418 return -EINVAL;
5419
5420 return 0;
5421}
5422
5423static ssize_t
5424ftrace_graph_write(struct file *file, const char __user *ubuf,
5425 size_t cnt, loff_t *ppos)
5426{
5427 ssize_t read, ret = 0;
5428 struct ftrace_graph_data *fgd = file->private_data;
5429 struct trace_parser *parser;
5430
5431 if (!cnt)
5432 return 0;
5433
5434 /* Read mode uses seq functions */
5435 if (file->f_mode & FMODE_READ) {
5436 struct seq_file *m = file->private_data;
5437 fgd = m->private;
5438 }
5439
5440 parser = &fgd->parser;
5441
5442 read = trace_get_user(parser, ubuf, cnt, ppos);
5443
5444 if (read >= 0 && trace_parser_loaded(parser) &&
5445 !trace_parser_cont(parser)) {
5446
5447 ret = ftrace_graph_set_hash(fgd->new_hash,
5448 parser->buffer);
5449 trace_parser_clear(parser);
5450 }
5451
5452 if (!ret)
5453 ret = read;
5454
5455 return ret;
5456}
5457
5458static const struct file_operations ftrace_graph_fops = {
5459 .open = ftrace_graph_open,
5460 .read = seq_read,
5461 .write = ftrace_graph_write,
5462 .llseek = tracing_lseek,
5463 .release = ftrace_graph_release,
5464};
5465
5466static const struct file_operations ftrace_graph_notrace_fops = {
5467 .open = ftrace_graph_notrace_open,
5468 .read = seq_read,
5469 .write = ftrace_graph_write,
5470 .llseek = tracing_lseek,
5471 .release = ftrace_graph_release,
5472};
5473#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5474
5475void ftrace_create_filter_files(struct ftrace_ops *ops,
5476 struct dentry *parent)
5477{
5478
5479 trace_create_file("set_ftrace_filter", 0644, parent,
5480 ops, &ftrace_filter_fops);
5481
5482 trace_create_file("set_ftrace_notrace", 0644, parent,
5483 ops, &ftrace_notrace_fops);
5484}
5485
5486/*
5487 * The name "destroy_filter_files" is really a misnomer. Although
5488 * in the future, it may actualy delete the files, but this is
5489 * really intended to make sure the ops passed in are disabled
5490 * and that when this function returns, the caller is free to
5491 * free the ops.
5492 *
5493 * The "destroy" name is only to match the "create" name that this
5494 * should be paired with.
5495 */
5496void ftrace_destroy_filter_files(struct ftrace_ops *ops)
5497{
5498 mutex_lock(&ftrace_lock);
5499 if (ops->flags & FTRACE_OPS_FL_ENABLED)
5500 ftrace_shutdown(ops, 0);
5501 ops->flags |= FTRACE_OPS_FL_DELETED;
5502 mutex_unlock(&ftrace_lock);
5503}
5504
5505static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
5506{
5507
5508 trace_create_file("available_filter_functions", 0444,
5509 d_tracer, NULL, &ftrace_avail_fops);
5510
5511 trace_create_file("enabled_functions", 0444,
5512 d_tracer, NULL, &ftrace_enabled_fops);
5513
5514 ftrace_create_filter_files(&global_ops, d_tracer);
5515
5516#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5517 trace_create_file("set_graph_function", 0644, d_tracer,
5518 NULL,
5519 &ftrace_graph_fops);
5520 trace_create_file("set_graph_notrace", 0644, d_tracer,
5521 NULL,
5522 &ftrace_graph_notrace_fops);
5523#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5524
5525 return 0;
5526}
5527
5528static int ftrace_cmp_ips(const void *a, const void *b)
5529{
5530 const unsigned long *ipa = a;
5531 const unsigned long *ipb = b;
5532
5533 if (*ipa > *ipb)
5534 return 1;
5535 if (*ipa < *ipb)
5536 return -1;
5537 return 0;
5538}
5539
5540static int ftrace_process_locs(struct module *mod,
5541 unsigned long *start,
5542 unsigned long *end)
5543{
5544 struct ftrace_page *start_pg;
5545 struct ftrace_page *pg;
5546 struct dyn_ftrace *rec;
5547 unsigned long count;
5548 unsigned long *p;
5549 unsigned long addr;
5550 unsigned long flags = 0; /* Shut up gcc */
5551 int ret = -ENOMEM;
5552
5553 count = end - start;
5554
5555 if (!count)
5556 return 0;
5557
5558 sort(start, count, sizeof(*start),
5559 ftrace_cmp_ips, NULL);
5560
5561 start_pg = ftrace_allocate_pages(count);
5562 if (!start_pg)
5563 return -ENOMEM;
5564
5565 mutex_lock(&ftrace_lock);
5566
5567 /*
5568 * Core and each module needs their own pages, as
5569 * modules will free them when they are removed.
5570 * Force a new page to be allocated for modules.
5571 */
5572 if (!mod) {
5573 WARN_ON(ftrace_pages || ftrace_pages_start);
5574 /* First initialization */
5575 ftrace_pages = ftrace_pages_start = start_pg;
5576 } else {
5577 if (!ftrace_pages)
5578 goto out;
5579
5580 if (WARN_ON(ftrace_pages->next)) {
5581 /* Hmm, we have free pages? */
5582 while (ftrace_pages->next)
5583 ftrace_pages = ftrace_pages->next;
5584 }
5585
5586 ftrace_pages->next = start_pg;
5587 }
5588
5589 p = start;
5590 pg = start_pg;
5591 while (p < end) {
5592 addr = ftrace_call_adjust(*p++);
5593 /*
5594 * Some architecture linkers will pad between
5595 * the different mcount_loc sections of different
5596 * object files to satisfy alignments.
5597 * Skip any NULL pointers.
5598 */
5599 if (!addr)
5600 continue;
5601
5602 if (pg->index == pg->size) {
5603 /* We should have allocated enough */
5604 if (WARN_ON(!pg->next))
5605 break;
5606 pg = pg->next;
5607 }
5608
5609 rec = &pg->records[pg->index++];
5610 rec->ip = addr;
5611 }
5612
5613 /* We should have used all pages */
5614 WARN_ON(pg->next);
5615
5616 /* Assign the last page to ftrace_pages */
5617 ftrace_pages = pg;
5618
5619 /*
5620 * We only need to disable interrupts on start up
5621 * because we are modifying code that an interrupt
5622 * may execute, and the modification is not atomic.
5623 * But for modules, nothing runs the code we modify
5624 * until we are finished with it, and there's no
5625 * reason to cause large interrupt latencies while we do it.
5626 */
5627 if (!mod)
5628 local_irq_save(flags);
5629 ftrace_update_code(mod, start_pg);
5630 if (!mod)
5631 local_irq_restore(flags);
5632 ret = 0;
5633 out:
5634 mutex_unlock(&ftrace_lock);
5635
5636 return ret;
5637}
5638
5639struct ftrace_mod_func {
5640 struct list_head list;
5641 char *name;
5642 unsigned long ip;
5643 unsigned int size;
5644};
5645
5646struct ftrace_mod_map {
5647 struct rcu_head rcu;
5648 struct list_head list;
5649 struct module *mod;
5650 unsigned long start_addr;
5651 unsigned long end_addr;
5652 struct list_head funcs;
5653 unsigned int num_funcs;
5654};
5655
5656#ifdef CONFIG_MODULES
5657
5658#define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
5659
5660static LIST_HEAD(ftrace_mod_maps);
5661
5662static int referenced_filters(struct dyn_ftrace *rec)
5663{
5664 struct ftrace_ops *ops;
5665 int cnt = 0;
5666
5667 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
5668 if (ops_references_rec(ops, rec))
5669 cnt++;
5670 }
5671
5672 return cnt;
5673}
5674
5675static void
5676clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
5677{
5678 struct ftrace_func_entry *entry;
5679 struct dyn_ftrace *rec;
5680 int i;
5681
5682 if (ftrace_hash_empty(hash))
5683 return;
5684
5685 for (i = 0; i < pg->index; i++) {
5686 rec = &pg->records[i];
5687 entry = __ftrace_lookup_ip(hash, rec->ip);
5688 /*
5689 * Do not allow this rec to match again.
5690 * Yeah, it may waste some memory, but will be removed
5691 * if/when the hash is modified again.
5692 */
5693 if (entry)
5694 entry->ip = 0;
5695 }
5696}
5697
5698/* Clear any records from hashs */
5699static void clear_mod_from_hashes(struct ftrace_page *pg)
5700{
5701 struct trace_array *tr;
5702
5703 mutex_lock(&trace_types_lock);
5704 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5705 if (!tr->ops || !tr->ops->func_hash)
5706 continue;
5707 mutex_lock(&tr->ops->func_hash->regex_lock);
5708 clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
5709 clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
5710 mutex_unlock(&tr->ops->func_hash->regex_lock);
5711 }
5712 mutex_unlock(&trace_types_lock);
5713}
5714
5715static void ftrace_free_mod_map(struct rcu_head *rcu)
5716{
5717 struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
5718 struct ftrace_mod_func *mod_func;
5719 struct ftrace_mod_func *n;
5720
5721 /* All the contents of mod_map are now not visible to readers */
5722 list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
5723 kfree(mod_func->name);
5724 list_del(&mod_func->list);
5725 kfree(mod_func);
5726 }
5727
5728 kfree(mod_map);
5729}
5730
5731void ftrace_release_mod(struct module *mod)
5732{
5733 struct ftrace_mod_map *mod_map;
5734 struct ftrace_mod_map *n;
5735 struct dyn_ftrace *rec;
5736 struct ftrace_page **last_pg;
5737 struct ftrace_page *tmp_page = NULL;
5738 struct ftrace_page *pg;
5739 int order;
5740
5741 mutex_lock(&ftrace_lock);
5742
5743 if (ftrace_disabled)
5744 goto out_unlock;
5745
5746 list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
5747 if (mod_map->mod == mod) {
5748 list_del_rcu(&mod_map->list);
5749 call_rcu_sched(&mod_map->rcu, ftrace_free_mod_map);
5750 break;
5751 }
5752 }
5753
5754 /*
5755 * Each module has its own ftrace_pages, remove
5756 * them from the list.
5757 */
5758 last_pg = &ftrace_pages_start;
5759 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
5760 rec = &pg->records[0];
5761 if (within_module_core(rec->ip, mod) ||
5762 within_module_init(rec->ip, mod)) {
5763 /*
5764 * As core pages are first, the first
5765 * page should never be a module page.
5766 */
5767 if (WARN_ON(pg == ftrace_pages_start))
5768 goto out_unlock;
5769
5770 /* Check if we are deleting the last page */
5771 if (pg == ftrace_pages)
5772 ftrace_pages = next_to_ftrace_page(last_pg);
5773
5774 ftrace_update_tot_cnt -= pg->index;
5775 *last_pg = pg->next;
5776
5777 pg->next = tmp_page;
5778 tmp_page = pg;
5779 } else
5780 last_pg = &pg->next;
5781 }
5782 out_unlock:
5783 mutex_unlock(&ftrace_lock);
5784
5785 for (pg = tmp_page; pg; pg = tmp_page) {
5786
5787 /* Needs to be called outside of ftrace_lock */
5788 clear_mod_from_hashes(pg);
5789
5790 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
5791 free_pages((unsigned long)pg->records, order);
5792 tmp_page = pg->next;
5793 kfree(pg);
5794 }
5795}
5796
5797void ftrace_module_enable(struct module *mod)
5798{
5799 struct dyn_ftrace *rec;
5800 struct ftrace_page *pg;
5801
5802 mutex_lock(&ftrace_lock);
5803
5804 if (ftrace_disabled)
5805 goto out_unlock;
5806
5807 /*
5808 * If the tracing is enabled, go ahead and enable the record.
5809 *
5810 * The reason not to enable the record immediatelly is the
5811 * inherent check of ftrace_make_nop/ftrace_make_call for
5812 * correct previous instructions. Making first the NOP
5813 * conversion puts the module to the correct state, thus
5814 * passing the ftrace_make_call check.
5815 *
5816 * We also delay this to after the module code already set the
5817 * text to read-only, as we now need to set it back to read-write
5818 * so that we can modify the text.
5819 */
5820 if (ftrace_start_up)
5821 ftrace_arch_code_modify_prepare();
5822
5823 do_for_each_ftrace_rec(pg, rec) {
5824 int cnt;
5825 /*
5826 * do_for_each_ftrace_rec() is a double loop.
5827 * module text shares the pg. If a record is
5828 * not part of this module, then skip this pg,
5829 * which the "break" will do.
5830 */
5831 if (!within_module_core(rec->ip, mod) &&
5832 !within_module_init(rec->ip, mod))
5833 break;
5834
5835 cnt = 0;
5836
5837 /*
5838 * When adding a module, we need to check if tracers are
5839 * currently enabled and if they are, and can trace this record,
5840 * we need to enable the module functions as well as update the
5841 * reference counts for those function records.
5842 */
5843 if (ftrace_start_up)
5844 cnt += referenced_filters(rec);
5845
5846 /* This clears FTRACE_FL_DISABLED */
5847 rec->flags = cnt;
5848
5849 if (ftrace_start_up && cnt) {
5850 int failed = __ftrace_replace_code(rec, 1);
5851 if (failed) {
5852 ftrace_bug(failed, rec);
5853 goto out_loop;
5854 }
5855 }
5856
5857 } while_for_each_ftrace_rec();
5858
5859 out_loop:
5860 if (ftrace_start_up)
5861 ftrace_arch_code_modify_post_process();
5862
5863 out_unlock:
5864 mutex_unlock(&ftrace_lock);
5865
5866 process_cached_mods(mod->name);
5867}
5868
5869void ftrace_module_init(struct module *mod)
5870{
5871 if (ftrace_disabled || !mod->num_ftrace_callsites)
5872 return;
5873
5874 ftrace_process_locs(mod, mod->ftrace_callsites,
5875 mod->ftrace_callsites + mod->num_ftrace_callsites);
5876}
5877
5878static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
5879 struct dyn_ftrace *rec)
5880{
5881 struct ftrace_mod_func *mod_func;
5882 unsigned long symsize;
5883 unsigned long offset;
5884 char str[KSYM_SYMBOL_LEN];
5885 char *modname;
5886 const char *ret;
5887
5888 ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
5889 if (!ret)
5890 return;
5891
5892 mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
5893 if (!mod_func)
5894 return;
5895
5896 mod_func->name = kstrdup(str, GFP_KERNEL);
5897 if (!mod_func->name) {
5898 kfree(mod_func);
5899 return;
5900 }
5901
5902 mod_func->ip = rec->ip - offset;
5903 mod_func->size = symsize;
5904
5905 mod_map->num_funcs++;
5906
5907 list_add_rcu(&mod_func->list, &mod_map->funcs);
5908}
5909
5910static struct ftrace_mod_map *
5911allocate_ftrace_mod_map(struct module *mod,
5912 unsigned long start, unsigned long end)
5913{
5914 struct ftrace_mod_map *mod_map;
5915
5916 mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
5917 if (!mod_map)
5918 return NULL;
5919
5920 mod_map->mod = mod;
5921 mod_map->start_addr = start;
5922 mod_map->end_addr = end;
5923 mod_map->num_funcs = 0;
5924
5925 INIT_LIST_HEAD_RCU(&mod_map->funcs);
5926
5927 list_add_rcu(&mod_map->list, &ftrace_mod_maps);
5928
5929 return mod_map;
5930}
5931
5932static const char *
5933ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
5934 unsigned long addr, unsigned long *size,
5935 unsigned long *off, char *sym)
5936{
5937 struct ftrace_mod_func *found_func = NULL;
5938 struct ftrace_mod_func *mod_func;
5939
5940 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
5941 if (addr >= mod_func->ip &&
5942 addr < mod_func->ip + mod_func->size) {
5943 found_func = mod_func;
5944 break;
5945 }
5946 }
5947
5948 if (found_func) {
5949 if (size)
5950 *size = found_func->size;
5951 if (off)
5952 *off = addr - found_func->ip;
5953 if (sym)
5954 strlcpy(sym, found_func->name, KSYM_NAME_LEN);
5955
5956 return found_func->name;
5957 }
5958
5959 return NULL;
5960}
5961
5962const char *
5963ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
5964 unsigned long *off, char **modname, char *sym)
5965{
5966 struct ftrace_mod_map *mod_map;
5967 const char *ret = NULL;
5968
5969 /* mod_map is freed via call_rcu_sched() */
5970 preempt_disable();
5971 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
5972 ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
5973 if (ret) {
5974 if (modname)
5975 *modname = mod_map->mod->name;
5976 break;
5977 }
5978 }
5979 preempt_enable();
5980
5981 return ret;
5982}
5983
5984int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
5985 char *type, char *name,
5986 char *module_name, int *exported)
5987{
5988 struct ftrace_mod_map *mod_map;
5989 struct ftrace_mod_func *mod_func;
5990
5991 preempt_disable();
5992 list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
5993
5994 if (symnum >= mod_map->num_funcs) {
5995 symnum -= mod_map->num_funcs;
5996 continue;
5997 }
5998
5999 list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
6000 if (symnum > 1) {
6001 symnum--;
6002 continue;
6003 }
6004
6005 *value = mod_func->ip;
6006 *type = 'T';
6007 strlcpy(name, mod_func->name, KSYM_NAME_LEN);
6008 strlcpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
6009 *exported = 1;
6010 preempt_enable();
6011 return 0;
6012 }
6013 WARN_ON(1);
6014 break;
6015 }
6016 preempt_enable();
6017 return -ERANGE;
6018}
6019
6020#else
6021static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
6022 struct dyn_ftrace *rec) { }
6023static inline struct ftrace_mod_map *
6024allocate_ftrace_mod_map(struct module *mod,
6025 unsigned long start, unsigned long end)
6026{
6027 return NULL;
6028}
6029#endif /* CONFIG_MODULES */
6030
6031struct ftrace_init_func {
6032 struct list_head list;
6033 unsigned long ip;
6034};
6035
6036/* Clear any init ips from hashes */
6037static void
6038clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
6039{
6040 struct ftrace_func_entry *entry;
6041
6042 if (ftrace_hash_empty(hash))
6043 return;
6044
6045 entry = __ftrace_lookup_ip(hash, func->ip);
6046
6047 /*
6048 * Do not allow this rec to match again.
6049 * Yeah, it may waste some memory, but will be removed
6050 * if/when the hash is modified again.
6051 */
6052 if (entry)
6053 entry->ip = 0;
6054}
6055
6056static void
6057clear_func_from_hashes(struct ftrace_init_func *func)
6058{
6059 struct trace_array *tr;
6060
6061 mutex_lock(&trace_types_lock);
6062 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6063 if (!tr->ops || !tr->ops->func_hash)
6064 continue;
6065 mutex_lock(&tr->ops->func_hash->regex_lock);
6066 clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
6067 clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
6068 mutex_unlock(&tr->ops->func_hash->regex_lock);
6069 }
6070 mutex_unlock(&trace_types_lock);
6071}
6072
6073static void add_to_clear_hash_list(struct list_head *clear_list,
6074 struct dyn_ftrace *rec)
6075{
6076 struct ftrace_init_func *func;
6077
6078 func = kmalloc(sizeof(*func), GFP_KERNEL);
6079 if (!func) {
6080 WARN_ONCE(1, "alloc failure, ftrace filter could be stale\n");
6081 return;
6082 }
6083
6084 func->ip = rec->ip;
6085 list_add(&func->list, clear_list);
6086}
6087
6088void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
6089{
6090 unsigned long start = (unsigned long)(start_ptr);
6091 unsigned long end = (unsigned long)(end_ptr);
6092 struct ftrace_page **last_pg = &ftrace_pages_start;
6093 struct ftrace_page *pg;
6094 struct dyn_ftrace *rec;
6095 struct dyn_ftrace key;
6096 struct ftrace_mod_map *mod_map = NULL;
6097 struct ftrace_init_func *func, *func_next;
6098 struct list_head clear_hash;
6099 int order;
6100
6101 INIT_LIST_HEAD(&clear_hash);
6102
6103 key.ip = start;
6104 key.flags = end; /* overload flags, as it is unsigned long */
6105
6106 mutex_lock(&ftrace_lock);
6107
6108 /*
6109 * If we are freeing module init memory, then check if
6110 * any tracer is active. If so, we need to save a mapping of
6111 * the module functions being freed with the address.
6112 */
6113 if (mod && ftrace_ops_list != &ftrace_list_end)
6114 mod_map = allocate_ftrace_mod_map(mod, start, end);
6115
6116 for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
6117 if (end < pg->records[0].ip ||
6118 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
6119 continue;
6120 again:
6121 rec = bsearch(&key, pg->records, pg->index,
6122 sizeof(struct dyn_ftrace),
6123 ftrace_cmp_recs);
6124 if (!rec)
6125 continue;
6126
6127 /* rec will be cleared from hashes after ftrace_lock unlock */
6128 add_to_clear_hash_list(&clear_hash, rec);
6129
6130 if (mod_map)
6131 save_ftrace_mod_rec(mod_map, rec);
6132
6133 pg->index--;
6134 ftrace_update_tot_cnt--;
6135 if (!pg->index) {
6136 *last_pg = pg->next;
6137 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
6138 free_pages((unsigned long)pg->records, order);
6139 kfree(pg);
6140 pg = container_of(last_pg, struct ftrace_page, next);
6141 if (!(*last_pg))
6142 ftrace_pages = pg;
6143 continue;
6144 }
6145 memmove(rec, rec + 1,
6146 (pg->index - (rec - pg->records)) * sizeof(*rec));
6147 /* More than one function may be in this block */
6148 goto again;
6149 }
6150 mutex_unlock(&ftrace_lock);
6151
6152 list_for_each_entry_safe(func, func_next, &clear_hash, list) {
6153 clear_func_from_hashes(func);
6154 kfree(func);
6155 }
6156}
6157
6158void __init ftrace_free_init_mem(void)
6159{
6160 void *start = (void *)(&__init_begin);
6161 void *end = (void *)(&__init_end);
6162
6163 ftrace_free_mem(NULL, start, end);
6164}
6165
6166void __init ftrace_init(void)
6167{
6168 extern unsigned long __start_mcount_loc[];
6169 extern unsigned long __stop_mcount_loc[];
6170 unsigned long count, flags;
6171 int ret;
6172
6173 local_irq_save(flags);
6174 ret = ftrace_dyn_arch_init();
6175 local_irq_restore(flags);
6176 if (ret)
6177 goto failed;
6178
6179 count = __stop_mcount_loc - __start_mcount_loc;
6180 if (!count) {
6181 pr_info("ftrace: No functions to be traced?\n");
6182 goto failed;
6183 }
6184
6185 pr_info("ftrace: allocating %ld entries in %ld pages\n",
6186 count, count / ENTRIES_PER_PAGE + 1);
6187
6188 last_ftrace_enabled = ftrace_enabled = 1;
6189
6190 ret = ftrace_process_locs(NULL,
6191 __start_mcount_loc,
6192 __stop_mcount_loc);
6193
6194 set_ftrace_early_filters();
6195
6196 return;
6197 failed:
6198 ftrace_disabled = 1;
6199}
6200
6201/* Do nothing if arch does not support this */
6202void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
6203{
6204}
6205
6206static void ftrace_update_trampoline(struct ftrace_ops *ops)
6207{
6208 arch_ftrace_update_trampoline(ops);
6209}
6210
6211void ftrace_init_trace_array(struct trace_array *tr)
6212{
6213 INIT_LIST_HEAD(&tr->func_probes);
6214 INIT_LIST_HEAD(&tr->mod_trace);
6215 INIT_LIST_HEAD(&tr->mod_notrace);
6216}
6217#else
6218
6219static struct ftrace_ops global_ops = {
6220 .func = ftrace_stub,
6221 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
6222 FTRACE_OPS_FL_INITIALIZED |
6223 FTRACE_OPS_FL_PID,
6224};
6225
6226static int __init ftrace_nodyn_init(void)
6227{
6228 ftrace_enabled = 1;
6229 return 0;
6230}
6231core_initcall(ftrace_nodyn_init);
6232
6233static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
6234static inline void ftrace_startup_enable(int command) { }
6235static inline void ftrace_startup_all(int command) { }
6236/* Keep as macros so we do not need to define the commands */
6237# define ftrace_startup(ops, command) \
6238 ({ \
6239 int ___ret = __register_ftrace_function(ops); \
6240 if (!___ret) \
6241 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
6242 ___ret; \
6243 })
6244# define ftrace_shutdown(ops, command) \
6245 ({ \
6246 int ___ret = __unregister_ftrace_function(ops); \
6247 if (!___ret) \
6248 (ops)->flags &= ~FTRACE_OPS_FL_ENABLED; \
6249 ___ret; \
6250 })
6251
6252# define ftrace_startup_sysctl() do { } while (0)
6253# define ftrace_shutdown_sysctl() do { } while (0)
6254
6255static inline int
6256ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
6257{
6258 return 1;
6259}
6260
6261static void ftrace_update_trampoline(struct ftrace_ops *ops)
6262{
6263}
6264
6265#endif /* CONFIG_DYNAMIC_FTRACE */
6266
6267__init void ftrace_init_global_array_ops(struct trace_array *tr)
6268{
6269 tr->ops = &global_ops;
6270 tr->ops->private = tr;
6271 ftrace_init_trace_array(tr);
6272}
6273
6274void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
6275{
6276 /* If we filter on pids, update to use the pid function */
6277 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
6278 if (WARN_ON(tr->ops->func != ftrace_stub))
6279 printk("ftrace ops had %pS for function\n",
6280 tr->ops->func);
6281 }
6282 tr->ops->func = func;
6283 tr->ops->private = tr;
6284}
6285
6286void ftrace_reset_array_ops(struct trace_array *tr)
6287{
6288 tr->ops->func = ftrace_stub;
6289}
6290
6291static inline void
6292__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6293 struct ftrace_ops *ignored, struct pt_regs *regs)
6294{
6295 struct ftrace_ops *op;
6296 int bit;
6297
6298 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6299 if (bit < 0)
6300 return;
6301
6302 /*
6303 * Some of the ops may be dynamically allocated,
6304 * they must be freed after a synchronize_sched().
6305 */
6306 preempt_disable_notrace();
6307
6308 do_for_each_ftrace_op(op, ftrace_ops_list) {
6309 /*
6310 * Check the following for each ops before calling their func:
6311 * if RCU flag is set, then rcu_is_watching() must be true
6312 * if PER_CPU is set, then ftrace_function_local_disable()
6313 * must be false
6314 * Otherwise test if the ip matches the ops filter
6315 *
6316 * If any of the above fails then the op->func() is not executed.
6317 */
6318 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
6319 ftrace_ops_test(op, ip, regs)) {
6320 if (FTRACE_WARN_ON(!op->func)) {
6321 pr_warn("op=%p %pS\n", op, op);
6322 goto out;
6323 }
6324 op->func(ip, parent_ip, op, regs);
6325 }
6326 } while_for_each_ftrace_op(op);
6327out:
6328 preempt_enable_notrace();
6329 trace_clear_recursion(bit);
6330}
6331
6332/*
6333 * Some archs only support passing ip and parent_ip. Even though
6334 * the list function ignores the op parameter, we do not want any
6335 * C side effects, where a function is called without the caller
6336 * sending a third parameter.
6337 * Archs are to support both the regs and ftrace_ops at the same time.
6338 * If they support ftrace_ops, it is assumed they support regs.
6339 * If call backs want to use regs, they must either check for regs
6340 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
6341 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
6342 * An architecture can pass partial regs with ftrace_ops and still
6343 * set the ARCH_SUPPORTS_FTRACE_OPS.
6344 */
6345#if ARCH_SUPPORTS_FTRACE_OPS
6346static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6347 struct ftrace_ops *op, struct pt_regs *regs)
6348{
6349 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
6350}
6351#else
6352static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
6353{
6354 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
6355}
6356#endif
6357
6358/*
6359 * If there's only one function registered but it does not support
6360 * recursion, needs RCU protection and/or requires per cpu handling, then
6361 * this function will be called by the mcount trampoline.
6362 */
6363static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
6364 struct ftrace_ops *op, struct pt_regs *regs)
6365{
6366 int bit;
6367
6368 if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
6369 return;
6370
6371 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6372 if (bit < 0)
6373 return;
6374
6375 preempt_disable_notrace();
6376
6377 op->func(ip, parent_ip, op, regs);
6378
6379 preempt_enable_notrace();
6380 trace_clear_recursion(bit);
6381}
6382
6383/**
6384 * ftrace_ops_get_func - get the function a trampoline should call
6385 * @ops: the ops to get the function for
6386 *
6387 * Normally the mcount trampoline will call the ops->func, but there
6388 * are times that it should not. For example, if the ops does not
6389 * have its own recursion protection, then it should call the
6390 * ftrace_ops_assist_func() instead.
6391 *
6392 * Returns the function that the trampoline should call for @ops.
6393 */
6394ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
6395{
6396 /*
6397 * If the function does not handle recursion, needs to be RCU safe,
6398 * or does per cpu logic, then we need to call the assist handler.
6399 */
6400 if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
6401 ops->flags & FTRACE_OPS_FL_RCU)
6402 return ftrace_ops_assist_func;
6403
6404 return ops->func;
6405}
6406
6407static void
6408ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
6409 struct task_struct *prev, struct task_struct *next)
6410{
6411 struct trace_array *tr = data;
6412 struct trace_pid_list *pid_list;
6413
6414 pid_list = rcu_dereference_sched(tr->function_pids);
6415
6416 this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6417 trace_ignore_this_task(pid_list, next));
6418}
6419
6420static void
6421ftrace_pid_follow_sched_process_fork(void *data,
6422 struct task_struct *self,
6423 struct task_struct *task)
6424{
6425 struct trace_pid_list *pid_list;
6426 struct trace_array *tr = data;
6427
6428 pid_list = rcu_dereference_sched(tr->function_pids);
6429 trace_filter_add_remove_task(pid_list, self, task);
6430}
6431
6432static void
6433ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
6434{
6435 struct trace_pid_list *pid_list;
6436 struct trace_array *tr = data;
6437
6438 pid_list = rcu_dereference_sched(tr->function_pids);
6439 trace_filter_add_remove_task(pid_list, NULL, task);
6440}
6441
6442void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
6443{
6444 if (enable) {
6445 register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6446 tr);
6447 register_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6448 tr);
6449 } else {
6450 unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6451 tr);
6452 unregister_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6453 tr);
6454 }
6455}
6456
6457static void clear_ftrace_pids(struct trace_array *tr)
6458{
6459 struct trace_pid_list *pid_list;
6460 int cpu;
6461
6462 pid_list = rcu_dereference_protected(tr->function_pids,
6463 lockdep_is_held(&ftrace_lock));
6464 if (!pid_list)
6465 return;
6466
6467 unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6468
6469 for_each_possible_cpu(cpu)
6470 per_cpu_ptr(tr->trace_buffer.data, cpu)->ftrace_ignore_pid = false;
6471
6472 rcu_assign_pointer(tr->function_pids, NULL);
6473
6474 /* Wait till all users are no longer using pid filtering */
6475 synchronize_sched();
6476
6477 trace_free_pid_list(pid_list);
6478}
6479
6480void ftrace_clear_pids(struct trace_array *tr)
6481{
6482 mutex_lock(&ftrace_lock);
6483
6484 clear_ftrace_pids(tr);
6485
6486 mutex_unlock(&ftrace_lock);
6487}
6488
6489static void ftrace_pid_reset(struct trace_array *tr)
6490{
6491 mutex_lock(&ftrace_lock);
6492 clear_ftrace_pids(tr);
6493
6494 ftrace_update_pid_func();
6495 ftrace_startup_all(0);
6496
6497 mutex_unlock(&ftrace_lock);
6498}
6499
6500/* Greater than any max PID */
6501#define FTRACE_NO_PIDS (void *)(PID_MAX_LIMIT + 1)
6502
6503static void *fpid_start(struct seq_file *m, loff_t *pos)
6504 __acquires(RCU)
6505{
6506 struct trace_pid_list *pid_list;
6507 struct trace_array *tr = m->private;
6508
6509 mutex_lock(&ftrace_lock);
6510 rcu_read_lock_sched();
6511
6512 pid_list = rcu_dereference_sched(tr->function_pids);
6513
6514 if (!pid_list)
6515 return !(*pos) ? FTRACE_NO_PIDS : NULL;
6516
6517 return trace_pid_start(pid_list, pos);
6518}
6519
6520static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
6521{
6522 struct trace_array *tr = m->private;
6523 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
6524
6525 if (v == FTRACE_NO_PIDS)
6526 return NULL;
6527
6528 return trace_pid_next(pid_list, v, pos);
6529}
6530
6531static void fpid_stop(struct seq_file *m, void *p)
6532 __releases(RCU)
6533{
6534 rcu_read_unlock_sched();
6535 mutex_unlock(&ftrace_lock);
6536}
6537
6538static int fpid_show(struct seq_file *m, void *v)
6539{
6540 if (v == FTRACE_NO_PIDS) {
6541 seq_puts(m, "no pid\n");
6542 return 0;
6543 }
6544
6545 return trace_pid_show(m, v);
6546}
6547
6548static const struct seq_operations ftrace_pid_sops = {
6549 .start = fpid_start,
6550 .next = fpid_next,
6551 .stop = fpid_stop,
6552 .show = fpid_show,
6553};
6554
6555static int
6556ftrace_pid_open(struct inode *inode, struct file *file)
6557{
6558 struct trace_array *tr = inode->i_private;
6559 struct seq_file *m;
6560 int ret = 0;
6561
6562 if (trace_array_get(tr) < 0)
6563 return -ENODEV;
6564
6565 if ((file->f_mode & FMODE_WRITE) &&
6566 (file->f_flags & O_TRUNC))
6567 ftrace_pid_reset(tr);
6568
6569 ret = seq_open(file, &ftrace_pid_sops);
6570 if (ret < 0) {
6571 trace_array_put(tr);
6572 } else {
6573 m = file->private_data;
6574 /* copy tr over to seq ops */
6575 m->private = tr;
6576 }
6577
6578 return ret;
6579}
6580
6581static void ignore_task_cpu(void *data)
6582{
6583 struct trace_array *tr = data;
6584 struct trace_pid_list *pid_list;
6585
6586 /*
6587 * This function is called by on_each_cpu() while the
6588 * event_mutex is held.
6589 */
6590 pid_list = rcu_dereference_protected(tr->function_pids,
6591 mutex_is_locked(&ftrace_lock));
6592
6593 this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6594 trace_ignore_this_task(pid_list, current));
6595}
6596
6597static ssize_t
6598ftrace_pid_write(struct file *filp, const char __user *ubuf,
6599 size_t cnt, loff_t *ppos)
6600{
6601 struct seq_file *m = filp->private_data;
6602 struct trace_array *tr = m->private;
6603 struct trace_pid_list *filtered_pids = NULL;
6604 struct trace_pid_list *pid_list;
6605 ssize_t ret;
6606
6607 if (!cnt)
6608 return 0;
6609
6610 mutex_lock(&ftrace_lock);
6611
6612 filtered_pids = rcu_dereference_protected(tr->function_pids,
6613 lockdep_is_held(&ftrace_lock));
6614
6615 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
6616 if (ret < 0)
6617 goto out;
6618
6619 rcu_assign_pointer(tr->function_pids, pid_list);
6620
6621 if (filtered_pids) {
6622 synchronize_sched();
6623 trace_free_pid_list(filtered_pids);
6624 } else if (pid_list) {
6625 /* Register a probe to set whether to ignore the tracing of a task */
6626 register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6627 }
6628
6629 /*
6630 * Ignoring of pids is done at task switch. But we have to
6631 * check for those tasks that are currently running.
6632 * Always do this in case a pid was appended or removed.
6633 */
6634 on_each_cpu(ignore_task_cpu, tr, 1);
6635
6636 ftrace_update_pid_func();
6637 ftrace_startup_all(0);
6638 out:
6639 mutex_unlock(&ftrace_lock);
6640
6641 if (ret > 0)
6642 *ppos += ret;
6643
6644 return ret;
6645}
6646
6647static int
6648ftrace_pid_release(struct inode *inode, struct file *file)
6649{
6650 struct trace_array *tr = inode->i_private;
6651
6652 trace_array_put(tr);
6653
6654 return seq_release(inode, file);
6655}
6656
6657static const struct file_operations ftrace_pid_fops = {
6658 .open = ftrace_pid_open,
6659 .write = ftrace_pid_write,
6660 .read = seq_read,
6661 .llseek = tracing_lseek,
6662 .release = ftrace_pid_release,
6663};
6664
6665void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
6666{
6667 trace_create_file("set_ftrace_pid", 0644, d_tracer,
6668 tr, &ftrace_pid_fops);
6669}
6670
6671void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
6672 struct dentry *d_tracer)
6673{
6674 /* Only the top level directory has the dyn_tracefs and profile */
6675 WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
6676
6677 ftrace_init_dyn_tracefs(d_tracer);
6678 ftrace_profile_tracefs(d_tracer);
6679}
6680
6681/**
6682 * ftrace_kill - kill ftrace
6683 *
6684 * This function should be used by panic code. It stops ftrace
6685 * but in a not so nice way. If you need to simply kill ftrace
6686 * from a non-atomic section, use ftrace_kill.
6687 */
6688void ftrace_kill(void)
6689{
6690 ftrace_disabled = 1;
6691 ftrace_enabled = 0;
6692 clear_ftrace_function();
6693}
6694
6695/**
6696 * Test if ftrace is dead or not.
6697 */
6698int ftrace_is_dead(void)
6699{
6700 return ftrace_disabled;
6701}
6702
6703/**
6704 * register_ftrace_function - register a function for profiling
6705 * @ops - ops structure that holds the function for profiling.
6706 *
6707 * Register a function to be called by all functions in the
6708 * kernel.
6709 *
6710 * Note: @ops->func and all the functions it calls must be labeled
6711 * with "notrace", otherwise it will go into a
6712 * recursive loop.
6713 */
6714int register_ftrace_function(struct ftrace_ops *ops)
6715{
6716 int ret = -1;
6717
6718 ftrace_ops_init(ops);
6719
6720 mutex_lock(&ftrace_lock);
6721
6722 ret = ftrace_startup(ops, 0);
6723
6724 mutex_unlock(&ftrace_lock);
6725
6726 return ret;
6727}
6728EXPORT_SYMBOL_GPL(register_ftrace_function);
6729
6730/**
6731 * unregister_ftrace_function - unregister a function for profiling.
6732 * @ops - ops structure that holds the function to unregister
6733 *
6734 * Unregister a function that was added to be called by ftrace profiling.
6735 */
6736int unregister_ftrace_function(struct ftrace_ops *ops)
6737{
6738 int ret;
6739
6740 mutex_lock(&ftrace_lock);
6741 ret = ftrace_shutdown(ops, 0);
6742 mutex_unlock(&ftrace_lock);
6743
6744 return ret;
6745}
6746EXPORT_SYMBOL_GPL(unregister_ftrace_function);
6747
6748int
6749ftrace_enable_sysctl(struct ctl_table *table, int write,
6750 void __user *buffer, size_t *lenp,
6751 loff_t *ppos)
6752{
6753 int ret = -ENODEV;
6754
6755 mutex_lock(&ftrace_lock);
6756
6757 if (unlikely(ftrace_disabled))
6758 goto out;
6759
6760 ret = proc_dointvec(table, write, buffer, lenp, ppos);
6761
6762 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
6763 goto out;
6764
6765 last_ftrace_enabled = !!ftrace_enabled;
6766
6767 if (ftrace_enabled) {
6768
6769 /* we are starting ftrace again */
6770 if (rcu_dereference_protected(ftrace_ops_list,
6771 lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
6772 update_ftrace_function();
6773
6774 ftrace_startup_sysctl();
6775
6776 } else {
6777 /* stopping ftrace calls (just send to ftrace_stub) */
6778 ftrace_trace_function = ftrace_stub;
6779
6780 ftrace_shutdown_sysctl();
6781 }
6782
6783 out:
6784 mutex_unlock(&ftrace_lock);
6785 return ret;
6786}
6787
6788#ifdef CONFIG_FUNCTION_GRAPH_TRACER
6789
6790static struct ftrace_ops graph_ops = {
6791 .func = ftrace_stub,
6792 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
6793 FTRACE_OPS_FL_INITIALIZED |
6794 FTRACE_OPS_FL_PID |
6795 FTRACE_OPS_FL_STUB,
6796#ifdef FTRACE_GRAPH_TRAMP_ADDR
6797 .trampoline = FTRACE_GRAPH_TRAMP_ADDR,
6798 /* trampoline_size is only needed for dynamically allocated tramps */
6799#endif
6800 ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
6801};
6802
6803void ftrace_graph_sleep_time_control(bool enable)
6804{
6805 fgraph_sleep_time = enable;
6806}
6807
6808void ftrace_graph_graph_time_control(bool enable)
6809{
6810 fgraph_graph_time = enable;
6811}
6812
6813int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
6814{
6815 return 0;
6816}
6817
6818/* The callbacks that hook a function */
6819trace_func_graph_ret_t ftrace_graph_return =
6820 (trace_func_graph_ret_t)ftrace_stub;
6821trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
6822static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
6823
6824/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
6825static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
6826{
6827 int i;
6828 int ret = 0;
6829 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
6830 struct task_struct *g, *t;
6831
6832 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
6833 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
6834 * sizeof(struct ftrace_ret_stack),
6835 GFP_KERNEL);
6836 if (!ret_stack_list[i]) {
6837 start = 0;
6838 end = i;
6839 ret = -ENOMEM;
6840 goto free;
6841 }
6842 }
6843
6844 read_lock(&tasklist_lock);
6845 do_each_thread(g, t) {
6846 if (start == end) {
6847 ret = -EAGAIN;
6848 goto unlock;
6849 }
6850
6851 if (t->ret_stack == NULL) {
6852 atomic_set(&t->tracing_graph_pause, 0);
6853 atomic_set(&t->trace_overrun, 0);
6854 t->curr_ret_stack = -1;
6855 /* Make sure the tasks see the -1 first: */
6856 smp_wmb();
6857 t->ret_stack = ret_stack_list[start++];
6858 }
6859 } while_each_thread(g, t);
6860
6861unlock:
6862 read_unlock(&tasklist_lock);
6863free:
6864 for (i = start; i < end; i++)
6865 kfree(ret_stack_list[i]);
6866 return ret;
6867}
6868
6869static void
6870ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
6871 struct task_struct *prev, struct task_struct *next)
6872{
6873 unsigned long long timestamp;
6874 int index;
6875
6876 /*
6877 * Does the user want to count the time a function was asleep.
6878 * If so, do not update the time stamps.
6879 */
6880 if (fgraph_sleep_time)
6881 return;
6882
6883 timestamp = trace_clock_local();
6884
6885 prev->ftrace_timestamp = timestamp;
6886
6887 /* only process tasks that we timestamped */
6888 if (!next->ftrace_timestamp)
6889 return;
6890
6891 /*
6892 * Update all the counters in next to make up for the
6893 * time next was sleeping.
6894 */
6895 timestamp -= next->ftrace_timestamp;
6896
6897 for (index = next->curr_ret_stack; index >= 0; index--)
6898 next->ret_stack[index].calltime += timestamp;
6899}
6900
6901/* Allocate a return stack for each task */
6902static int start_graph_tracing(void)
6903{
6904 struct ftrace_ret_stack **ret_stack_list;
6905 int ret, cpu;
6906
6907 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
6908 sizeof(struct ftrace_ret_stack *),
6909 GFP_KERNEL);
6910
6911 if (!ret_stack_list)
6912 return -ENOMEM;
6913
6914 /* The cpu_boot init_task->ret_stack will never be freed */
6915 for_each_online_cpu(cpu) {
6916 if (!idle_task(cpu)->ret_stack)
6917 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
6918 }
6919
6920 do {
6921 ret = alloc_retstack_tasklist(ret_stack_list);
6922 } while (ret == -EAGAIN);
6923
6924 if (!ret) {
6925 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
6926 if (ret)
6927 pr_info("ftrace_graph: Couldn't activate tracepoint"
6928 " probe to kernel_sched_switch\n");
6929 }
6930
6931 kfree(ret_stack_list);
6932 return ret;
6933}
6934
6935/*
6936 * Hibernation protection.
6937 * The state of the current task is too much unstable during
6938 * suspend/restore to disk. We want to protect against that.
6939 */
6940static int
6941ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
6942 void *unused)
6943{
6944 switch (state) {
6945 case PM_HIBERNATION_PREPARE:
6946 pause_graph_tracing();
6947 break;
6948
6949 case PM_POST_HIBERNATION:
6950 unpause_graph_tracing();
6951 break;
6952 }
6953 return NOTIFY_DONE;
6954}
6955
6956static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
6957{
6958 if (!ftrace_ops_test(&global_ops, trace->func, NULL))
6959 return 0;
6960 return __ftrace_graph_entry(trace);
6961}
6962
6963/*
6964 * The function graph tracer should only trace the functions defined
6965 * by set_ftrace_filter and set_ftrace_notrace. If another function
6966 * tracer ops is registered, the graph tracer requires testing the
6967 * function against the global ops, and not just trace any function
6968 * that any ftrace_ops registered.
6969 */
6970static void update_function_graph_func(void)
6971{
6972 struct ftrace_ops *op;
6973 bool do_test = false;
6974
6975 /*
6976 * The graph and global ops share the same set of functions
6977 * to test. If any other ops is on the list, then
6978 * the graph tracing needs to test if its the function
6979 * it should call.
6980 */
6981 do_for_each_ftrace_op(op, ftrace_ops_list) {
6982 if (op != &global_ops && op != &graph_ops &&
6983 op != &ftrace_list_end) {
6984 do_test = true;
6985 /* in double loop, break out with goto */
6986 goto out;
6987 }
6988 } while_for_each_ftrace_op(op);
6989 out:
6990 if (do_test)
6991 ftrace_graph_entry = ftrace_graph_entry_test;
6992 else
6993 ftrace_graph_entry = __ftrace_graph_entry;
6994}
6995
6996static struct notifier_block ftrace_suspend_notifier = {
6997 .notifier_call = ftrace_suspend_notifier_call,
6998};
6999
7000int register_ftrace_graph(trace_func_graph_ret_t retfunc,
7001 trace_func_graph_ent_t entryfunc)
7002{
7003 int ret = 0;
7004
7005 mutex_lock(&ftrace_lock);
7006
7007 /* we currently allow only one tracer registered at a time */
7008 if (ftrace_graph_active) {
7009 ret = -EBUSY;
7010 goto out;
7011 }
7012
7013 register_pm_notifier(&ftrace_suspend_notifier);
7014
7015 ftrace_graph_active++;
7016 ret = start_graph_tracing();
7017 if (ret) {
7018 ftrace_graph_active--;
7019 goto out;
7020 }
7021
7022 ftrace_graph_return = retfunc;
7023
7024 /*
7025 * Update the indirect function to the entryfunc, and the
7026 * function that gets called to the entry_test first. Then
7027 * call the update fgraph entry function to determine if
7028 * the entryfunc should be called directly or not.
7029 */
7030 __ftrace_graph_entry = entryfunc;
7031 ftrace_graph_entry = ftrace_graph_entry_test;
7032 update_function_graph_func();
7033
7034 ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
7035out:
7036 mutex_unlock(&ftrace_lock);
7037 return ret;
7038}
7039
7040void unregister_ftrace_graph(void)
7041{
7042 mutex_lock(&ftrace_lock);
7043
7044 if (unlikely(!ftrace_graph_active))
7045 goto out;
7046
7047 ftrace_graph_active--;
7048 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
7049 ftrace_graph_entry = ftrace_graph_entry_stub;
7050 __ftrace_graph_entry = ftrace_graph_entry_stub;
7051 ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
7052 unregister_pm_notifier(&ftrace_suspend_notifier);
7053 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
7054
7055 out:
7056 mutex_unlock(&ftrace_lock);
7057}
7058
7059static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
7060
7061static void
7062graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
7063{
7064 atomic_set(&t->tracing_graph_pause, 0);
7065 atomic_set(&t->trace_overrun, 0);
7066 t->ftrace_timestamp = 0;
7067 /* make curr_ret_stack visible before we add the ret_stack */
7068 smp_wmb();
7069 t->ret_stack = ret_stack;
7070}
7071
7072/*
7073 * Allocate a return stack for the idle task. May be the first
7074 * time through, or it may be done by CPU hotplug online.
7075 */
7076void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
7077{
7078 t->curr_ret_stack = -1;
7079 /*
7080 * The idle task has no parent, it either has its own
7081 * stack or no stack at all.
7082 */
7083 if (t->ret_stack)
7084 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
7085
7086 if (ftrace_graph_active) {
7087 struct ftrace_ret_stack *ret_stack;
7088
7089 ret_stack = per_cpu(idle_ret_stack, cpu);
7090 if (!ret_stack) {
7091 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
7092 * sizeof(struct ftrace_ret_stack),
7093 GFP_KERNEL);
7094 if (!ret_stack)
7095 return;
7096 per_cpu(idle_ret_stack, cpu) = ret_stack;
7097 }
7098 graph_init_task(t, ret_stack);
7099 }
7100}
7101
7102/* Allocate a return stack for newly created task */
7103void ftrace_graph_init_task(struct task_struct *t)
7104{
7105 /* Make sure we do not use the parent ret_stack */
7106 t->ret_stack = NULL;
7107 t->curr_ret_stack = -1;
7108
7109 if (ftrace_graph_active) {
7110 struct ftrace_ret_stack *ret_stack;
7111
7112 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
7113 * sizeof(struct ftrace_ret_stack),
7114 GFP_KERNEL);
7115 if (!ret_stack)
7116 return;
7117 graph_init_task(t, ret_stack);
7118 }
7119}
7120
7121void ftrace_graph_exit_task(struct task_struct *t)
7122{
7123 struct ftrace_ret_stack *ret_stack = t->ret_stack;
7124
7125 t->ret_stack = NULL;
7126 /* NULL must become visible to IRQs before we free it: */
7127 barrier();
7128
7129 kfree(ret_stack);
7130}
7131#endif
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
77static 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 */
84int ftrace_enabled __read_mostly;
85static int last_ftrace_enabled;
86
87/* Current function tracing op */
88struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
89/* What to set function_trace_op to */
90static struct ftrace_ops *set_function_trace_op;
91
92/* List for set_ftrace_pid's pids. */
93LIST_HEAD(ftrace_pids);
94struct ftrace_pid {
95 struct list_head list;
96 struct pid *pid;
97};
98
99static bool ftrace_pids_enabled(void)
100{
101 return !list_empty(&ftrace_pids);
102}
103
104static 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 */
110static int ftrace_disabled __read_mostly;
111
112static DEFINE_MUTEX(ftrace_lock);
113
114static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
115ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
116static struct ftrace_ops global_ops;
117
118#if ARCH_SUPPORTS_FTRACE_OPS
119static 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 */
123static 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
147static 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 */
163int 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
179static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
180 struct ftrace_ops *op, struct pt_regs *regs)
181{
182 if (!test_tsk_trace_trace(current))
183 return;
184
185 op->saved_func(ip, parent_ip, op, regs);
186}
187
188/**
189 * clear_ftrace_function - reset the ftrace function
190 *
191 * This NULLs the ftrace function and in essence stops
192 * tracing. There may be lag
193 */
194void clear_ftrace_function(void)
195{
196 ftrace_trace_function = ftrace_stub;
197}
198
199static void per_cpu_ops_disable_all(struct ftrace_ops *ops)
200{
201 int cpu;
202
203 for_each_possible_cpu(cpu)
204 *per_cpu_ptr(ops->disabled, cpu) = 1;
205}
206
207static int per_cpu_ops_alloc(struct ftrace_ops *ops)
208{
209 int __percpu *disabled;
210
211 if (WARN_ON_ONCE(!(ops->flags & FTRACE_OPS_FL_PER_CPU)))
212 return -EINVAL;
213
214 disabled = alloc_percpu(int);
215 if (!disabled)
216 return -ENOMEM;
217
218 ops->disabled = disabled;
219 per_cpu_ops_disable_all(ops);
220 return 0;
221}
222
223static void ftrace_sync(struct work_struct *work)
224{
225 /*
226 * This function is just a stub to implement a hard force
227 * of synchronize_sched(). This requires synchronizing
228 * tasks even in userspace and idle.
229 *
230 * Yes, function tracing is rude.
231 */
232}
233
234static void ftrace_sync_ipi(void *data)
235{
236 /* Probably not needed, but do it anyway */
237 smp_rmb();
238}
239
240#ifdef CONFIG_FUNCTION_GRAPH_TRACER
241static void update_function_graph_func(void);
242
243/* Both enabled by default (can be cleared by function_graph tracer flags */
244static bool fgraph_sleep_time = true;
245static bool fgraph_graph_time = true;
246
247#else
248static inline void update_function_graph_func(void) { }
249#endif
250
251
252static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
253{
254 /*
255 * If this is a dynamic, RCU, or per CPU ops, or we force list func,
256 * then it needs to call the list anyway.
257 */
258 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU |
259 FTRACE_OPS_FL_RCU) || FTRACE_FORCE_LIST_FUNC)
260 return ftrace_ops_list_func;
261
262 return ftrace_ops_get_func(ops);
263}
264
265static void update_ftrace_function(void)
266{
267 ftrace_func_t func;
268
269 /*
270 * Prepare the ftrace_ops that the arch callback will use.
271 * If there's only one ftrace_ops registered, the ftrace_ops_list
272 * will point to the ops we want.
273 */
274 set_function_trace_op = ftrace_ops_list;
275
276 /* If there's no ftrace_ops registered, just call the stub function */
277 if (ftrace_ops_list == &ftrace_list_end) {
278 func = ftrace_stub;
279
280 /*
281 * If we are at the end of the list and this ops is
282 * recursion safe and not dynamic and the arch supports passing ops,
283 * then have the mcount trampoline call the function directly.
284 */
285 } else if (ftrace_ops_list->next == &ftrace_list_end) {
286 func = ftrace_ops_get_list_func(ftrace_ops_list);
287
288 } else {
289 /* Just use the default ftrace_ops */
290 set_function_trace_op = &ftrace_list_end;
291 func = ftrace_ops_list_func;
292 }
293
294 update_function_graph_func();
295
296 /* If there's no change, then do nothing more here */
297 if (ftrace_trace_function == func)
298 return;
299
300 /*
301 * If we are using the list function, it doesn't care
302 * about the function_trace_ops.
303 */
304 if (func == ftrace_ops_list_func) {
305 ftrace_trace_function = func;
306 /*
307 * Don't even bother setting function_trace_ops,
308 * it would be racy to do so anyway.
309 */
310 return;
311 }
312
313#ifndef CONFIG_DYNAMIC_FTRACE
314 /*
315 * For static tracing, we need to be a bit more careful.
316 * The function change takes affect immediately. Thus,
317 * we need to coorditate the setting of the function_trace_ops
318 * with the setting of the ftrace_trace_function.
319 *
320 * Set the function to the list ops, which will call the
321 * function we want, albeit indirectly, but it handles the
322 * ftrace_ops and doesn't depend on function_trace_op.
323 */
324 ftrace_trace_function = ftrace_ops_list_func;
325 /*
326 * Make sure all CPUs see this. Yes this is slow, but static
327 * tracing is slow and nasty to have enabled.
328 */
329 schedule_on_each_cpu(ftrace_sync);
330 /* Now all cpus are using the list ops. */
331 function_trace_op = set_function_trace_op;
332 /* Make sure the function_trace_op is visible on all CPUs */
333 smp_wmb();
334 /* Nasty way to force a rmb on all cpus */
335 smp_call_function(ftrace_sync_ipi, NULL, 1);
336 /* OK, we are all set to update the ftrace_trace_function now! */
337#endif /* !CONFIG_DYNAMIC_FTRACE */
338
339 ftrace_trace_function = func;
340}
341
342int using_ftrace_ops_list_func(void)
343{
344 return ftrace_trace_function == ftrace_ops_list_func;
345}
346
347static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
348{
349 ops->next = *list;
350 /*
351 * We are entering ops into the list but another
352 * CPU might be walking that list. We need to make sure
353 * the ops->next pointer is valid before another CPU sees
354 * the ops pointer included into the list.
355 */
356 rcu_assign_pointer(*list, ops);
357}
358
359static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
360{
361 struct ftrace_ops **p;
362
363 /*
364 * If we are removing the last function, then simply point
365 * to the ftrace_stub.
366 */
367 if (*list == ops && ops->next == &ftrace_list_end) {
368 *list = &ftrace_list_end;
369 return 0;
370 }
371
372 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
373 if (*p == ops)
374 break;
375
376 if (*p != ops)
377 return -1;
378
379 *p = (*p)->next;
380 return 0;
381}
382
383static void ftrace_update_trampoline(struct ftrace_ops *ops);
384
385static int __register_ftrace_function(struct ftrace_ops *ops)
386{
387 if (ops->flags & FTRACE_OPS_FL_DELETED)
388 return -EINVAL;
389
390 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
391 return -EBUSY;
392
393#ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
394 /*
395 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
396 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
397 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
398 */
399 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
400 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
401 return -EINVAL;
402
403 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
404 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
405#endif
406
407 if (!core_kernel_data((unsigned long)ops))
408 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
409
410 if (ops->flags & FTRACE_OPS_FL_PER_CPU) {
411 if (per_cpu_ops_alloc(ops))
412 return -ENOMEM;
413 }
414
415 add_ftrace_ops(&ftrace_ops_list, ops);
416
417 /* Always save the function, and reset at unregistering */
418 ops->saved_func = ops->func;
419
420 if (ops->flags & FTRACE_OPS_FL_PID && ftrace_pids_enabled())
421 ops->func = ftrace_pid_func;
422
423 ftrace_update_trampoline(ops);
424
425 if (ftrace_enabled)
426 update_ftrace_function();
427
428 return 0;
429}
430
431static int __unregister_ftrace_function(struct ftrace_ops *ops)
432{
433 int ret;
434
435 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
436 return -EBUSY;
437
438 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
439
440 if (ret < 0)
441 return ret;
442
443 if (ftrace_enabled)
444 update_ftrace_function();
445
446 ops->func = ops->saved_func;
447
448 return 0;
449}
450
451static void ftrace_update_pid_func(void)
452{
453 bool enabled = ftrace_pids_enabled();
454 struct ftrace_ops *op;
455
456 /* Only do something if we are tracing something */
457 if (ftrace_trace_function == ftrace_stub)
458 return;
459
460 do_for_each_ftrace_op(op, ftrace_ops_list) {
461 if (op->flags & FTRACE_OPS_FL_PID) {
462 op->func = enabled ? ftrace_pid_func :
463 op->saved_func;
464 ftrace_update_trampoline(op);
465 }
466 } while_for_each_ftrace_op(op);
467
468 update_ftrace_function();
469}
470
471#ifdef CONFIG_FUNCTION_PROFILER
472struct ftrace_profile {
473 struct hlist_node node;
474 unsigned long ip;
475 unsigned long counter;
476#ifdef CONFIG_FUNCTION_GRAPH_TRACER
477 unsigned long long time;
478 unsigned long long time_squared;
479#endif
480};
481
482struct ftrace_profile_page {
483 struct ftrace_profile_page *next;
484 unsigned long index;
485 struct ftrace_profile records[];
486};
487
488struct ftrace_profile_stat {
489 atomic_t disabled;
490 struct hlist_head *hash;
491 struct ftrace_profile_page *pages;
492 struct ftrace_profile_page *start;
493 struct tracer_stat stat;
494};
495
496#define PROFILE_RECORDS_SIZE \
497 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
498
499#define PROFILES_PER_PAGE \
500 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
501
502static int ftrace_profile_enabled __read_mostly;
503
504/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
505static DEFINE_MUTEX(ftrace_profile_lock);
506
507static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
508
509#define FTRACE_PROFILE_HASH_BITS 10
510#define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
511
512static void *
513function_stat_next(void *v, int idx)
514{
515 struct ftrace_profile *rec = v;
516 struct ftrace_profile_page *pg;
517
518 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
519
520 again:
521 if (idx != 0)
522 rec++;
523
524 if ((void *)rec >= (void *)&pg->records[pg->index]) {
525 pg = pg->next;
526 if (!pg)
527 return NULL;
528 rec = &pg->records[0];
529 if (!rec->counter)
530 goto again;
531 }
532
533 return rec;
534}
535
536static void *function_stat_start(struct tracer_stat *trace)
537{
538 struct ftrace_profile_stat *stat =
539 container_of(trace, struct ftrace_profile_stat, stat);
540
541 if (!stat || !stat->start)
542 return NULL;
543
544 return function_stat_next(&stat->start->records[0], 0);
545}
546
547#ifdef CONFIG_FUNCTION_GRAPH_TRACER
548/* function graph compares on total time */
549static int function_stat_cmp(void *p1, void *p2)
550{
551 struct ftrace_profile *a = p1;
552 struct ftrace_profile *b = p2;
553
554 if (a->time < b->time)
555 return -1;
556 if (a->time > b->time)
557 return 1;
558 else
559 return 0;
560}
561#else
562/* not function graph compares against hits */
563static int function_stat_cmp(void *p1, void *p2)
564{
565 struct ftrace_profile *a = p1;
566 struct ftrace_profile *b = p2;
567
568 if (a->counter < b->counter)
569 return -1;
570 if (a->counter > b->counter)
571 return 1;
572 else
573 return 0;
574}
575#endif
576
577static int function_stat_headers(struct seq_file *m)
578{
579#ifdef CONFIG_FUNCTION_GRAPH_TRACER
580 seq_puts(m, " Function "
581 "Hit Time Avg s^2\n"
582 " -------- "
583 "--- ---- --- ---\n");
584#else
585 seq_puts(m, " Function Hit\n"
586 " -------- ---\n");
587#endif
588 return 0;
589}
590
591static int function_stat_show(struct seq_file *m, void *v)
592{
593 struct ftrace_profile *rec = v;
594 char str[KSYM_SYMBOL_LEN];
595 int ret = 0;
596#ifdef CONFIG_FUNCTION_GRAPH_TRACER
597 static struct trace_seq s;
598 unsigned long long avg;
599 unsigned long long stddev;
600#endif
601 mutex_lock(&ftrace_profile_lock);
602
603 /* we raced with function_profile_reset() */
604 if (unlikely(rec->counter == 0)) {
605 ret = -EBUSY;
606 goto out;
607 }
608
609#ifdef CONFIG_FUNCTION_GRAPH_TRACER
610 avg = rec->time;
611 do_div(avg, rec->counter);
612 if (tracing_thresh && (avg < tracing_thresh))
613 goto out;
614#endif
615
616 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
617 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
618
619#ifdef CONFIG_FUNCTION_GRAPH_TRACER
620 seq_puts(m, " ");
621
622 /* Sample standard deviation (s^2) */
623 if (rec->counter <= 1)
624 stddev = 0;
625 else {
626 /*
627 * Apply Welford's method:
628 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
629 */
630 stddev = rec->counter * rec->time_squared -
631 rec->time * rec->time;
632
633 /*
634 * Divide only 1000 for ns^2 -> us^2 conversion.
635 * trace_print_graph_duration will divide 1000 again.
636 */
637 do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
638 }
639
640 trace_seq_init(&s);
641 trace_print_graph_duration(rec->time, &s);
642 trace_seq_puts(&s, " ");
643 trace_print_graph_duration(avg, &s);
644 trace_seq_puts(&s, " ");
645 trace_print_graph_duration(stddev, &s);
646 trace_print_seq(m, &s);
647#endif
648 seq_putc(m, '\n');
649out:
650 mutex_unlock(&ftrace_profile_lock);
651
652 return ret;
653}
654
655static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
656{
657 struct ftrace_profile_page *pg;
658
659 pg = stat->pages = stat->start;
660
661 while (pg) {
662 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
663 pg->index = 0;
664 pg = pg->next;
665 }
666
667 memset(stat->hash, 0,
668 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
669}
670
671int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
672{
673 struct ftrace_profile_page *pg;
674 int functions;
675 int pages;
676 int i;
677
678 /* If we already allocated, do nothing */
679 if (stat->pages)
680 return 0;
681
682 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
683 if (!stat->pages)
684 return -ENOMEM;
685
686#ifdef CONFIG_DYNAMIC_FTRACE
687 functions = ftrace_update_tot_cnt;
688#else
689 /*
690 * We do not know the number of functions that exist because
691 * dynamic tracing is what counts them. With past experience
692 * we have around 20K functions. That should be more than enough.
693 * It is highly unlikely we will execute every function in
694 * the kernel.
695 */
696 functions = 20000;
697#endif
698
699 pg = stat->start = stat->pages;
700
701 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
702
703 for (i = 1; i < pages; i++) {
704 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
705 if (!pg->next)
706 goto out_free;
707 pg = pg->next;
708 }
709
710 return 0;
711
712 out_free:
713 pg = stat->start;
714 while (pg) {
715 unsigned long tmp = (unsigned long)pg;
716
717 pg = pg->next;
718 free_page(tmp);
719 }
720
721 stat->pages = NULL;
722 stat->start = NULL;
723
724 return -ENOMEM;
725}
726
727static int ftrace_profile_init_cpu(int cpu)
728{
729 struct ftrace_profile_stat *stat;
730 int size;
731
732 stat = &per_cpu(ftrace_profile_stats, cpu);
733
734 if (stat->hash) {
735 /* If the profile is already created, simply reset it */
736 ftrace_profile_reset(stat);
737 return 0;
738 }
739
740 /*
741 * We are profiling all functions, but usually only a few thousand
742 * functions are hit. We'll make a hash of 1024 items.
743 */
744 size = FTRACE_PROFILE_HASH_SIZE;
745
746 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
747
748 if (!stat->hash)
749 return -ENOMEM;
750
751 /* Preallocate the function profiling pages */
752 if (ftrace_profile_pages_init(stat) < 0) {
753 kfree(stat->hash);
754 stat->hash = NULL;
755 return -ENOMEM;
756 }
757
758 return 0;
759}
760
761static int ftrace_profile_init(void)
762{
763 int cpu;
764 int ret = 0;
765
766 for_each_possible_cpu(cpu) {
767 ret = ftrace_profile_init_cpu(cpu);
768 if (ret)
769 break;
770 }
771
772 return ret;
773}
774
775/* interrupts must be disabled */
776static struct ftrace_profile *
777ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
778{
779 struct ftrace_profile *rec;
780 struct hlist_head *hhd;
781 unsigned long key;
782
783 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
784 hhd = &stat->hash[key];
785
786 if (hlist_empty(hhd))
787 return NULL;
788
789 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
790 if (rec->ip == ip)
791 return rec;
792 }
793
794 return NULL;
795}
796
797static void ftrace_add_profile(struct ftrace_profile_stat *stat,
798 struct ftrace_profile *rec)
799{
800 unsigned long key;
801
802 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
803 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
804}
805
806/*
807 * The memory is already allocated, this simply finds a new record to use.
808 */
809static struct ftrace_profile *
810ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
811{
812 struct ftrace_profile *rec = NULL;
813
814 /* prevent recursion (from NMIs) */
815 if (atomic_inc_return(&stat->disabled) != 1)
816 goto out;
817
818 /*
819 * Try to find the function again since an NMI
820 * could have added it
821 */
822 rec = ftrace_find_profiled_func(stat, ip);
823 if (rec)
824 goto out;
825
826 if (stat->pages->index == PROFILES_PER_PAGE) {
827 if (!stat->pages->next)
828 goto out;
829 stat->pages = stat->pages->next;
830 }
831
832 rec = &stat->pages->records[stat->pages->index++];
833 rec->ip = ip;
834 ftrace_add_profile(stat, rec);
835
836 out:
837 atomic_dec(&stat->disabled);
838
839 return rec;
840}
841
842static void
843function_profile_call(unsigned long ip, unsigned long parent_ip,
844 struct ftrace_ops *ops, struct pt_regs *regs)
845{
846 struct ftrace_profile_stat *stat;
847 struct ftrace_profile *rec;
848 unsigned long flags;
849
850 if (!ftrace_profile_enabled)
851 return;
852
853 local_irq_save(flags);
854
855 stat = this_cpu_ptr(&ftrace_profile_stats);
856 if (!stat->hash || !ftrace_profile_enabled)
857 goto out;
858
859 rec = ftrace_find_profiled_func(stat, ip);
860 if (!rec) {
861 rec = ftrace_profile_alloc(stat, ip);
862 if (!rec)
863 goto out;
864 }
865
866 rec->counter++;
867 out:
868 local_irq_restore(flags);
869}
870
871#ifdef CONFIG_FUNCTION_GRAPH_TRACER
872static int profile_graph_entry(struct ftrace_graph_ent *trace)
873{
874 function_profile_call(trace->func, 0, NULL, NULL);
875 return 1;
876}
877
878static void profile_graph_return(struct ftrace_graph_ret *trace)
879{
880 struct ftrace_profile_stat *stat;
881 unsigned long long calltime;
882 struct ftrace_profile *rec;
883 unsigned long flags;
884
885 local_irq_save(flags);
886 stat = this_cpu_ptr(&ftrace_profile_stats);
887 if (!stat->hash || !ftrace_profile_enabled)
888 goto out;
889
890 /* If the calltime was zero'd ignore it */
891 if (!trace->calltime)
892 goto out;
893
894 calltime = trace->rettime - trace->calltime;
895
896 if (!fgraph_graph_time) {
897 int index;
898
899 index = trace->depth;
900
901 /* Append this call time to the parent time to subtract */
902 if (index)
903 current->ret_stack[index - 1].subtime += calltime;
904
905 if (current->ret_stack[index].subtime < calltime)
906 calltime -= current->ret_stack[index].subtime;
907 else
908 calltime = 0;
909 }
910
911 rec = ftrace_find_profiled_func(stat, trace->func);
912 if (rec) {
913 rec->time += calltime;
914 rec->time_squared += calltime * calltime;
915 }
916
917 out:
918 local_irq_restore(flags);
919}
920
921static int register_ftrace_profiler(void)
922{
923 return register_ftrace_graph(&profile_graph_return,
924 &profile_graph_entry);
925}
926
927static void unregister_ftrace_profiler(void)
928{
929 unregister_ftrace_graph();
930}
931#else
932static struct ftrace_ops ftrace_profile_ops __read_mostly = {
933 .func = function_profile_call,
934 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
935 INIT_OPS_HASH(ftrace_profile_ops)
936};
937
938static int register_ftrace_profiler(void)
939{
940 return register_ftrace_function(&ftrace_profile_ops);
941}
942
943static void unregister_ftrace_profiler(void)
944{
945 unregister_ftrace_function(&ftrace_profile_ops);
946}
947#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
948
949static ssize_t
950ftrace_profile_write(struct file *filp, const char __user *ubuf,
951 size_t cnt, loff_t *ppos)
952{
953 unsigned long val;
954 int ret;
955
956 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
957 if (ret)
958 return ret;
959
960 val = !!val;
961
962 mutex_lock(&ftrace_profile_lock);
963 if (ftrace_profile_enabled ^ val) {
964 if (val) {
965 ret = ftrace_profile_init();
966 if (ret < 0) {
967 cnt = ret;
968 goto out;
969 }
970
971 ret = register_ftrace_profiler();
972 if (ret < 0) {
973 cnt = ret;
974 goto out;
975 }
976 ftrace_profile_enabled = 1;
977 } else {
978 ftrace_profile_enabled = 0;
979 /*
980 * unregister_ftrace_profiler calls stop_machine
981 * so this acts like an synchronize_sched.
982 */
983 unregister_ftrace_profiler();
984 }
985 }
986 out:
987 mutex_unlock(&ftrace_profile_lock);
988
989 *ppos += cnt;
990
991 return cnt;
992}
993
994static ssize_t
995ftrace_profile_read(struct file *filp, char __user *ubuf,
996 size_t cnt, loff_t *ppos)
997{
998 char buf[64]; /* big enough to hold a number */
999 int r;
1000
1001 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1002 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1003}
1004
1005static const struct file_operations ftrace_profile_fops = {
1006 .open = tracing_open_generic,
1007 .read = ftrace_profile_read,
1008 .write = ftrace_profile_write,
1009 .llseek = default_llseek,
1010};
1011
1012/* used to initialize the real stat files */
1013static struct tracer_stat function_stats __initdata = {
1014 .name = "functions",
1015 .stat_start = function_stat_start,
1016 .stat_next = function_stat_next,
1017 .stat_cmp = function_stat_cmp,
1018 .stat_headers = function_stat_headers,
1019 .stat_show = function_stat_show
1020};
1021
1022static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1023{
1024 struct ftrace_profile_stat *stat;
1025 struct dentry *entry;
1026 char *name;
1027 int ret;
1028 int cpu;
1029
1030 for_each_possible_cpu(cpu) {
1031 stat = &per_cpu(ftrace_profile_stats, cpu);
1032
1033 name = kasprintf(GFP_KERNEL, "function%d", cpu);
1034 if (!name) {
1035 /*
1036 * The files created are permanent, if something happens
1037 * we still do not free memory.
1038 */
1039 WARN(1,
1040 "Could not allocate stat file for cpu %d\n",
1041 cpu);
1042 return;
1043 }
1044 stat->stat = function_stats;
1045 stat->stat.name = name;
1046 ret = register_stat_tracer(&stat->stat);
1047 if (ret) {
1048 WARN(1,
1049 "Could not register function stat for cpu %d\n",
1050 cpu);
1051 kfree(name);
1052 return;
1053 }
1054 }
1055
1056 entry = tracefs_create_file("function_profile_enabled", 0644,
1057 d_tracer, NULL, &ftrace_profile_fops);
1058 if (!entry)
1059 pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1060}
1061
1062#else /* CONFIG_FUNCTION_PROFILER */
1063static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1064{
1065}
1066#endif /* CONFIG_FUNCTION_PROFILER */
1067
1068static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1069
1070#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1071static int ftrace_graph_active;
1072#else
1073# define ftrace_graph_active 0
1074#endif
1075
1076#ifdef CONFIG_DYNAMIC_FTRACE
1077
1078static struct ftrace_ops *removed_ops;
1079
1080/*
1081 * Set when doing a global update, like enabling all recs or disabling them.
1082 * It is not set when just updating a single ftrace_ops.
1083 */
1084static bool update_all_ops;
1085
1086#ifndef CONFIG_FTRACE_MCOUNT_RECORD
1087# error Dynamic ftrace depends on MCOUNT_RECORD
1088#endif
1089
1090static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1091
1092struct ftrace_func_probe {
1093 struct hlist_node node;
1094 struct ftrace_probe_ops *ops;
1095 unsigned long flags;
1096 unsigned long ip;
1097 void *data;
1098 struct list_head free_list;
1099};
1100
1101struct ftrace_func_entry {
1102 struct hlist_node hlist;
1103 unsigned long ip;
1104};
1105
1106struct ftrace_hash {
1107 unsigned long size_bits;
1108 struct hlist_head *buckets;
1109 unsigned long count;
1110 struct rcu_head rcu;
1111};
1112
1113/*
1114 * We make these constant because no one should touch them,
1115 * but they are used as the default "empty hash", to avoid allocating
1116 * it all the time. These are in a read only section such that if
1117 * anyone does try to modify it, it will cause an exception.
1118 */
1119static const struct hlist_head empty_buckets[1];
1120static const struct ftrace_hash empty_hash = {
1121 .buckets = (struct hlist_head *)empty_buckets,
1122};
1123#define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1124
1125static struct ftrace_ops global_ops = {
1126 .func = ftrace_stub,
1127 .local_hash.notrace_hash = EMPTY_HASH,
1128 .local_hash.filter_hash = EMPTY_HASH,
1129 INIT_OPS_HASH(global_ops)
1130 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
1131 FTRACE_OPS_FL_INITIALIZED |
1132 FTRACE_OPS_FL_PID,
1133};
1134
1135/*
1136 * This is used by __kernel_text_address() to return true if the
1137 * address is on a dynamically allocated trampoline that would
1138 * not return true for either core_kernel_text() or
1139 * is_module_text_address().
1140 */
1141bool is_ftrace_trampoline(unsigned long addr)
1142{
1143 struct ftrace_ops *op;
1144 bool ret = false;
1145
1146 /*
1147 * Some of the ops may be dynamically allocated,
1148 * they are freed after a synchronize_sched().
1149 */
1150 preempt_disable_notrace();
1151
1152 do_for_each_ftrace_op(op, ftrace_ops_list) {
1153 /*
1154 * This is to check for dynamically allocated trampolines.
1155 * Trampolines that are in kernel text will have
1156 * core_kernel_text() return true.
1157 */
1158 if (op->trampoline && op->trampoline_size)
1159 if (addr >= op->trampoline &&
1160 addr < op->trampoline + op->trampoline_size) {
1161 ret = true;
1162 goto out;
1163 }
1164 } while_for_each_ftrace_op(op);
1165
1166 out:
1167 preempt_enable_notrace();
1168
1169 return ret;
1170}
1171
1172struct ftrace_page {
1173 struct ftrace_page *next;
1174 struct dyn_ftrace *records;
1175 int index;
1176 int size;
1177};
1178
1179#define ENTRY_SIZE sizeof(struct dyn_ftrace)
1180#define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1181
1182/* estimate from running different kernels */
1183#define NR_TO_INIT 10000
1184
1185static struct ftrace_page *ftrace_pages_start;
1186static struct ftrace_page *ftrace_pages;
1187
1188static bool __always_inline ftrace_hash_empty(struct ftrace_hash *hash)
1189{
1190 return !hash || !hash->count;
1191}
1192
1193static struct ftrace_func_entry *
1194ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1195{
1196 unsigned long key;
1197 struct ftrace_func_entry *entry;
1198 struct hlist_head *hhd;
1199
1200 if (ftrace_hash_empty(hash))
1201 return NULL;
1202
1203 if (hash->size_bits > 0)
1204 key = hash_long(ip, hash->size_bits);
1205 else
1206 key = 0;
1207
1208 hhd = &hash->buckets[key];
1209
1210 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1211 if (entry->ip == ip)
1212 return entry;
1213 }
1214 return NULL;
1215}
1216
1217static void __add_hash_entry(struct ftrace_hash *hash,
1218 struct ftrace_func_entry *entry)
1219{
1220 struct hlist_head *hhd;
1221 unsigned long key;
1222
1223 if (hash->size_bits)
1224 key = hash_long(entry->ip, hash->size_bits);
1225 else
1226 key = 0;
1227
1228 hhd = &hash->buckets[key];
1229 hlist_add_head(&entry->hlist, hhd);
1230 hash->count++;
1231}
1232
1233static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1234{
1235 struct ftrace_func_entry *entry;
1236
1237 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1238 if (!entry)
1239 return -ENOMEM;
1240
1241 entry->ip = ip;
1242 __add_hash_entry(hash, entry);
1243
1244 return 0;
1245}
1246
1247static void
1248free_hash_entry(struct ftrace_hash *hash,
1249 struct ftrace_func_entry *entry)
1250{
1251 hlist_del(&entry->hlist);
1252 kfree(entry);
1253 hash->count--;
1254}
1255
1256static void
1257remove_hash_entry(struct ftrace_hash *hash,
1258 struct ftrace_func_entry *entry)
1259{
1260 hlist_del(&entry->hlist);
1261 hash->count--;
1262}
1263
1264static void ftrace_hash_clear(struct ftrace_hash *hash)
1265{
1266 struct hlist_head *hhd;
1267 struct hlist_node *tn;
1268 struct ftrace_func_entry *entry;
1269 int size = 1 << hash->size_bits;
1270 int i;
1271
1272 if (!hash->count)
1273 return;
1274
1275 for (i = 0; i < size; i++) {
1276 hhd = &hash->buckets[i];
1277 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1278 free_hash_entry(hash, entry);
1279 }
1280 FTRACE_WARN_ON(hash->count);
1281}
1282
1283static void free_ftrace_hash(struct ftrace_hash *hash)
1284{
1285 if (!hash || hash == EMPTY_HASH)
1286 return;
1287 ftrace_hash_clear(hash);
1288 kfree(hash->buckets);
1289 kfree(hash);
1290}
1291
1292static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1293{
1294 struct ftrace_hash *hash;
1295
1296 hash = container_of(rcu, struct ftrace_hash, rcu);
1297 free_ftrace_hash(hash);
1298}
1299
1300static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1301{
1302 if (!hash || hash == EMPTY_HASH)
1303 return;
1304 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1305}
1306
1307void ftrace_free_filter(struct ftrace_ops *ops)
1308{
1309 ftrace_ops_init(ops);
1310 free_ftrace_hash(ops->func_hash->filter_hash);
1311 free_ftrace_hash(ops->func_hash->notrace_hash);
1312}
1313
1314static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1315{
1316 struct ftrace_hash *hash;
1317 int size;
1318
1319 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1320 if (!hash)
1321 return NULL;
1322
1323 size = 1 << size_bits;
1324 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1325
1326 if (!hash->buckets) {
1327 kfree(hash);
1328 return NULL;
1329 }
1330
1331 hash->size_bits = size_bits;
1332
1333 return hash;
1334}
1335
1336static struct ftrace_hash *
1337alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1338{
1339 struct ftrace_func_entry *entry;
1340 struct ftrace_hash *new_hash;
1341 int size;
1342 int ret;
1343 int i;
1344
1345 new_hash = alloc_ftrace_hash(size_bits);
1346 if (!new_hash)
1347 return NULL;
1348
1349 /* Empty hash? */
1350 if (ftrace_hash_empty(hash))
1351 return new_hash;
1352
1353 size = 1 << hash->size_bits;
1354 for (i = 0; i < size; i++) {
1355 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1356 ret = add_hash_entry(new_hash, entry->ip);
1357 if (ret < 0)
1358 goto free_hash;
1359 }
1360 }
1361
1362 FTRACE_WARN_ON(new_hash->count != hash->count);
1363
1364 return new_hash;
1365
1366 free_hash:
1367 free_ftrace_hash(new_hash);
1368 return NULL;
1369}
1370
1371static void
1372ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1373static void
1374ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1375
1376static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1377 struct ftrace_hash *new_hash);
1378
1379static int
1380ftrace_hash_move(struct ftrace_ops *ops, int enable,
1381 struct ftrace_hash **dst, struct ftrace_hash *src)
1382{
1383 struct ftrace_func_entry *entry;
1384 struct hlist_node *tn;
1385 struct hlist_head *hhd;
1386 struct ftrace_hash *new_hash;
1387 int size = src->count;
1388 int bits = 0;
1389 int ret;
1390 int i;
1391
1392 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1393 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1394 return -EINVAL;
1395
1396 /*
1397 * If the new source is empty, just free dst and assign it
1398 * the empty_hash.
1399 */
1400 if (!src->count) {
1401 new_hash = EMPTY_HASH;
1402 goto update;
1403 }
1404
1405 /*
1406 * Make the hash size about 1/2 the # found
1407 */
1408 for (size /= 2; size; size >>= 1)
1409 bits++;
1410
1411 /* Don't allocate too much */
1412 if (bits > FTRACE_HASH_MAX_BITS)
1413 bits = FTRACE_HASH_MAX_BITS;
1414
1415 new_hash = alloc_ftrace_hash(bits);
1416 if (!new_hash)
1417 return -ENOMEM;
1418
1419 size = 1 << src->size_bits;
1420 for (i = 0; i < size; i++) {
1421 hhd = &src->buckets[i];
1422 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1423 remove_hash_entry(src, entry);
1424 __add_hash_entry(new_hash, entry);
1425 }
1426 }
1427
1428update:
1429 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1430 if (enable) {
1431 /* IPMODIFY should be updated only when filter_hash updating */
1432 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1433 if (ret < 0) {
1434 free_ftrace_hash(new_hash);
1435 return ret;
1436 }
1437 }
1438
1439 /*
1440 * Remove the current set, update the hash and add
1441 * them back.
1442 */
1443 ftrace_hash_rec_disable_modify(ops, enable);
1444
1445 rcu_assign_pointer(*dst, new_hash);
1446
1447 ftrace_hash_rec_enable_modify(ops, enable);
1448
1449 return 0;
1450}
1451
1452static bool hash_contains_ip(unsigned long ip,
1453 struct ftrace_ops_hash *hash)
1454{
1455 /*
1456 * The function record is a match if it exists in the filter
1457 * hash and not in the notrace hash. Note, an emty hash is
1458 * considered a match for the filter hash, but an empty
1459 * notrace hash is considered not in the notrace hash.
1460 */
1461 return (ftrace_hash_empty(hash->filter_hash) ||
1462 ftrace_lookup_ip(hash->filter_hash, ip)) &&
1463 (ftrace_hash_empty(hash->notrace_hash) ||
1464 !ftrace_lookup_ip(hash->notrace_hash, ip));
1465}
1466
1467/*
1468 * Test the hashes for this ops to see if we want to call
1469 * the ops->func or not.
1470 *
1471 * It's a match if the ip is in the ops->filter_hash or
1472 * the filter_hash does not exist or is empty,
1473 * AND
1474 * the ip is not in the ops->notrace_hash.
1475 *
1476 * This needs to be called with preemption disabled as
1477 * the hashes are freed with call_rcu_sched().
1478 */
1479static int
1480ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1481{
1482 struct ftrace_ops_hash hash;
1483 int ret;
1484
1485#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1486 /*
1487 * There's a small race when adding ops that the ftrace handler
1488 * that wants regs, may be called without them. We can not
1489 * allow that handler to be called if regs is NULL.
1490 */
1491 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1492 return 0;
1493#endif
1494
1495 hash.filter_hash = rcu_dereference_raw_notrace(ops->func_hash->filter_hash);
1496 hash.notrace_hash = rcu_dereference_raw_notrace(ops->func_hash->notrace_hash);
1497
1498 if (hash_contains_ip(ip, &hash))
1499 ret = 1;
1500 else
1501 ret = 0;
1502
1503 return ret;
1504}
1505
1506/*
1507 * This is a double for. Do not use 'break' to break out of the loop,
1508 * you must use a goto.
1509 */
1510#define do_for_each_ftrace_rec(pg, rec) \
1511 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1512 int _____i; \
1513 for (_____i = 0; _____i < pg->index; _____i++) { \
1514 rec = &pg->records[_____i];
1515
1516#define while_for_each_ftrace_rec() \
1517 } \
1518 }
1519
1520
1521static int ftrace_cmp_recs(const void *a, const void *b)
1522{
1523 const struct dyn_ftrace *key = a;
1524 const struct dyn_ftrace *rec = b;
1525
1526 if (key->flags < rec->ip)
1527 return -1;
1528 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1529 return 1;
1530 return 0;
1531}
1532
1533static unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1534{
1535 struct ftrace_page *pg;
1536 struct dyn_ftrace *rec;
1537 struct dyn_ftrace key;
1538
1539 key.ip = start;
1540 key.flags = end; /* overload flags, as it is unsigned long */
1541
1542 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1543 if (end < pg->records[0].ip ||
1544 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1545 continue;
1546 rec = bsearch(&key, pg->records, pg->index,
1547 sizeof(struct dyn_ftrace),
1548 ftrace_cmp_recs);
1549 if (rec)
1550 return rec->ip;
1551 }
1552
1553 return 0;
1554}
1555
1556/**
1557 * ftrace_location - return true if the ip giving is a traced location
1558 * @ip: the instruction pointer to check
1559 *
1560 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1561 * That is, the instruction that is either a NOP or call to
1562 * the function tracer. It checks the ftrace internal tables to
1563 * determine if the address belongs or not.
1564 */
1565unsigned long ftrace_location(unsigned long ip)
1566{
1567 return ftrace_location_range(ip, ip);
1568}
1569
1570/**
1571 * ftrace_text_reserved - return true if range contains an ftrace location
1572 * @start: start of range to search
1573 * @end: end of range to search (inclusive). @end points to the last byte to check.
1574 *
1575 * Returns 1 if @start and @end contains a ftrace location.
1576 * That is, the instruction that is either a NOP or call to
1577 * the function tracer. It checks the ftrace internal tables to
1578 * determine if the address belongs or not.
1579 */
1580int ftrace_text_reserved(const void *start, const void *end)
1581{
1582 unsigned long ret;
1583
1584 ret = ftrace_location_range((unsigned long)start,
1585 (unsigned long)end);
1586
1587 return (int)!!ret;
1588}
1589
1590/* Test if ops registered to this rec needs regs */
1591static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1592{
1593 struct ftrace_ops *ops;
1594 bool keep_regs = false;
1595
1596 for (ops = ftrace_ops_list;
1597 ops != &ftrace_list_end; ops = ops->next) {
1598 /* pass rec in as regs to have non-NULL val */
1599 if (ftrace_ops_test(ops, rec->ip, rec)) {
1600 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1601 keep_regs = true;
1602 break;
1603 }
1604 }
1605 }
1606
1607 return keep_regs;
1608}
1609
1610static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1611 int filter_hash,
1612 bool inc)
1613{
1614 struct ftrace_hash *hash;
1615 struct ftrace_hash *other_hash;
1616 struct ftrace_page *pg;
1617 struct dyn_ftrace *rec;
1618 bool update = false;
1619 int count = 0;
1620 int all = 0;
1621
1622 /* Only update if the ops has been registered */
1623 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1624 return false;
1625
1626 /*
1627 * In the filter_hash case:
1628 * If the count is zero, we update all records.
1629 * Otherwise we just update the items in the hash.
1630 *
1631 * In the notrace_hash case:
1632 * We enable the update in the hash.
1633 * As disabling notrace means enabling the tracing,
1634 * and enabling notrace means disabling, the inc variable
1635 * gets inversed.
1636 */
1637 if (filter_hash) {
1638 hash = ops->func_hash->filter_hash;
1639 other_hash = ops->func_hash->notrace_hash;
1640 if (ftrace_hash_empty(hash))
1641 all = 1;
1642 } else {
1643 inc = !inc;
1644 hash = ops->func_hash->notrace_hash;
1645 other_hash = ops->func_hash->filter_hash;
1646 /*
1647 * If the notrace hash has no items,
1648 * then there's nothing to do.
1649 */
1650 if (ftrace_hash_empty(hash))
1651 return false;
1652 }
1653
1654 do_for_each_ftrace_rec(pg, rec) {
1655 int in_other_hash = 0;
1656 int in_hash = 0;
1657 int match = 0;
1658
1659 if (rec->flags & FTRACE_FL_DISABLED)
1660 continue;
1661
1662 if (all) {
1663 /*
1664 * Only the filter_hash affects all records.
1665 * Update if the record is not in the notrace hash.
1666 */
1667 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1668 match = 1;
1669 } else {
1670 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1671 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1672
1673 /*
1674 * If filter_hash is set, we want to match all functions
1675 * that are in the hash but not in the other hash.
1676 *
1677 * If filter_hash is not set, then we are decrementing.
1678 * That means we match anything that is in the hash
1679 * and also in the other_hash. That is, we need to turn
1680 * off functions in the other hash because they are disabled
1681 * by this hash.
1682 */
1683 if (filter_hash && in_hash && !in_other_hash)
1684 match = 1;
1685 else if (!filter_hash && in_hash &&
1686 (in_other_hash || ftrace_hash_empty(other_hash)))
1687 match = 1;
1688 }
1689 if (!match)
1690 continue;
1691
1692 if (inc) {
1693 rec->flags++;
1694 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1695 return false;
1696
1697 /*
1698 * If there's only a single callback registered to a
1699 * function, and the ops has a trampoline registered
1700 * for it, then we can call it directly.
1701 */
1702 if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1703 rec->flags |= FTRACE_FL_TRAMP;
1704 else
1705 /*
1706 * If we are adding another function callback
1707 * to this function, and the previous had a
1708 * custom trampoline in use, then we need to go
1709 * back to the default trampoline.
1710 */
1711 rec->flags &= ~FTRACE_FL_TRAMP;
1712
1713 /*
1714 * If any ops wants regs saved for this function
1715 * then all ops will get saved regs.
1716 */
1717 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1718 rec->flags |= FTRACE_FL_REGS;
1719 } else {
1720 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1721 return false;
1722 rec->flags--;
1723
1724 /*
1725 * If the rec had REGS enabled and the ops that is
1726 * being removed had REGS set, then see if there is
1727 * still any ops for this record that wants regs.
1728 * If not, we can stop recording them.
1729 */
1730 if (ftrace_rec_count(rec) > 0 &&
1731 rec->flags & FTRACE_FL_REGS &&
1732 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1733 if (!test_rec_ops_needs_regs(rec))
1734 rec->flags &= ~FTRACE_FL_REGS;
1735 }
1736
1737 /*
1738 * If the rec had TRAMP enabled, then it needs to
1739 * be cleared. As TRAMP can only be enabled iff
1740 * there is only a single ops attached to it.
1741 * In otherwords, always disable it on decrementing.
1742 * In the future, we may set it if rec count is
1743 * decremented to one, and the ops that is left
1744 * has a trampoline.
1745 */
1746 rec->flags &= ~FTRACE_FL_TRAMP;
1747
1748 /*
1749 * flags will be cleared in ftrace_check_record()
1750 * if rec count is zero.
1751 */
1752 }
1753 count++;
1754
1755 /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1756 update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1757
1758 /* Shortcut, if we handled all records, we are done. */
1759 if (!all && count == hash->count)
1760 return update;
1761 } while_for_each_ftrace_rec();
1762
1763 return update;
1764}
1765
1766static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1767 int filter_hash)
1768{
1769 return __ftrace_hash_rec_update(ops, filter_hash, 0);
1770}
1771
1772static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1773 int filter_hash)
1774{
1775 return __ftrace_hash_rec_update(ops, filter_hash, 1);
1776}
1777
1778static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1779 int filter_hash, int inc)
1780{
1781 struct ftrace_ops *op;
1782
1783 __ftrace_hash_rec_update(ops, filter_hash, inc);
1784
1785 if (ops->func_hash != &global_ops.local_hash)
1786 return;
1787
1788 /*
1789 * If the ops shares the global_ops hash, then we need to update
1790 * all ops that are enabled and use this hash.
1791 */
1792 do_for_each_ftrace_op(op, ftrace_ops_list) {
1793 /* Already done */
1794 if (op == ops)
1795 continue;
1796 if (op->func_hash == &global_ops.local_hash)
1797 __ftrace_hash_rec_update(op, filter_hash, inc);
1798 } while_for_each_ftrace_op(op);
1799}
1800
1801static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1802 int filter_hash)
1803{
1804 ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1805}
1806
1807static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1808 int filter_hash)
1809{
1810 ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1811}
1812
1813/*
1814 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1815 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1816 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1817 * Note that old_hash and new_hash has below meanings
1818 * - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1819 * - If the hash is EMPTY_HASH, it hits nothing
1820 * - Anything else hits the recs which match the hash entries.
1821 */
1822static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1823 struct ftrace_hash *old_hash,
1824 struct ftrace_hash *new_hash)
1825{
1826 struct ftrace_page *pg;
1827 struct dyn_ftrace *rec, *end = NULL;
1828 int in_old, in_new;
1829
1830 /* Only update if the ops has been registered */
1831 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1832 return 0;
1833
1834 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1835 return 0;
1836
1837 /*
1838 * Since the IPMODIFY is a very address sensitive action, we do not
1839 * allow ftrace_ops to set all functions to new hash.
1840 */
1841 if (!new_hash || !old_hash)
1842 return -EINVAL;
1843
1844 /* Update rec->flags */
1845 do_for_each_ftrace_rec(pg, rec) {
1846 /* We need to update only differences of filter_hash */
1847 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1848 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1849 if (in_old == in_new)
1850 continue;
1851
1852 if (in_new) {
1853 /* New entries must ensure no others are using it */
1854 if (rec->flags & FTRACE_FL_IPMODIFY)
1855 goto rollback;
1856 rec->flags |= FTRACE_FL_IPMODIFY;
1857 } else /* Removed entry */
1858 rec->flags &= ~FTRACE_FL_IPMODIFY;
1859 } while_for_each_ftrace_rec();
1860
1861 return 0;
1862
1863rollback:
1864 end = rec;
1865
1866 /* Roll back what we did above */
1867 do_for_each_ftrace_rec(pg, rec) {
1868 if (rec == end)
1869 goto err_out;
1870
1871 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1872 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1873 if (in_old == in_new)
1874 continue;
1875
1876 if (in_new)
1877 rec->flags &= ~FTRACE_FL_IPMODIFY;
1878 else
1879 rec->flags |= FTRACE_FL_IPMODIFY;
1880 } while_for_each_ftrace_rec();
1881
1882err_out:
1883 return -EBUSY;
1884}
1885
1886static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1887{
1888 struct ftrace_hash *hash = ops->func_hash->filter_hash;
1889
1890 if (ftrace_hash_empty(hash))
1891 hash = NULL;
1892
1893 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1894}
1895
1896/* Disabling always succeeds */
1897static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1898{
1899 struct ftrace_hash *hash = ops->func_hash->filter_hash;
1900
1901 if (ftrace_hash_empty(hash))
1902 hash = NULL;
1903
1904 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1905}
1906
1907static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1908 struct ftrace_hash *new_hash)
1909{
1910 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1911
1912 if (ftrace_hash_empty(old_hash))
1913 old_hash = NULL;
1914
1915 if (ftrace_hash_empty(new_hash))
1916 new_hash = NULL;
1917
1918 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
1919}
1920
1921static void print_ip_ins(const char *fmt, const unsigned char *p)
1922{
1923 int i;
1924
1925 printk(KERN_CONT "%s", fmt);
1926
1927 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1928 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1929}
1930
1931static struct ftrace_ops *
1932ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1933static struct ftrace_ops *
1934ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1935
1936enum ftrace_bug_type ftrace_bug_type;
1937const void *ftrace_expected;
1938
1939static void print_bug_type(void)
1940{
1941 switch (ftrace_bug_type) {
1942 case FTRACE_BUG_UNKNOWN:
1943 break;
1944 case FTRACE_BUG_INIT:
1945 pr_info("Initializing ftrace call sites\n");
1946 break;
1947 case FTRACE_BUG_NOP:
1948 pr_info("Setting ftrace call site to NOP\n");
1949 break;
1950 case FTRACE_BUG_CALL:
1951 pr_info("Setting ftrace call site to call ftrace function\n");
1952 break;
1953 case FTRACE_BUG_UPDATE:
1954 pr_info("Updating ftrace call site to call a different ftrace function\n");
1955 break;
1956 }
1957}
1958
1959/**
1960 * ftrace_bug - report and shutdown function tracer
1961 * @failed: The failed type (EFAULT, EINVAL, EPERM)
1962 * @rec: The record that failed
1963 *
1964 * The arch code that enables or disables the function tracing
1965 * can call ftrace_bug() when it has detected a problem in
1966 * modifying the code. @failed should be one of either:
1967 * EFAULT - if the problem happens on reading the @ip address
1968 * EINVAL - if what is read at @ip is not what was expected
1969 * EPERM - if the problem happens on writting to the @ip address
1970 */
1971void ftrace_bug(int failed, struct dyn_ftrace *rec)
1972{
1973 unsigned long ip = rec ? rec->ip : 0;
1974
1975 switch (failed) {
1976 case -EFAULT:
1977 FTRACE_WARN_ON_ONCE(1);
1978 pr_info("ftrace faulted on modifying ");
1979 print_ip_sym(ip);
1980 break;
1981 case -EINVAL:
1982 FTRACE_WARN_ON_ONCE(1);
1983 pr_info("ftrace failed to modify ");
1984 print_ip_sym(ip);
1985 print_ip_ins(" actual: ", (unsigned char *)ip);
1986 pr_cont("\n");
1987 if (ftrace_expected) {
1988 print_ip_ins(" expected: ", ftrace_expected);
1989 pr_cont("\n");
1990 }
1991 break;
1992 case -EPERM:
1993 FTRACE_WARN_ON_ONCE(1);
1994 pr_info("ftrace faulted on writing ");
1995 print_ip_sym(ip);
1996 break;
1997 default:
1998 FTRACE_WARN_ON_ONCE(1);
1999 pr_info("ftrace faulted on unknown error ");
2000 print_ip_sym(ip);
2001 }
2002 print_bug_type();
2003 if (rec) {
2004 struct ftrace_ops *ops = NULL;
2005
2006 pr_info("ftrace record flags: %lx\n", rec->flags);
2007 pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2008 rec->flags & FTRACE_FL_REGS ? " R" : " ");
2009 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2010 ops = ftrace_find_tramp_ops_any(rec);
2011 if (ops) {
2012 do {
2013 pr_cont("\ttramp: %pS (%pS)",
2014 (void *)ops->trampoline,
2015 (void *)ops->func);
2016 ops = ftrace_find_tramp_ops_next(rec, ops);
2017 } while (ops);
2018 } else
2019 pr_cont("\ttramp: ERROR!");
2020
2021 }
2022 ip = ftrace_get_addr_curr(rec);
2023 pr_cont("\n expected tramp: %lx\n", ip);
2024 }
2025}
2026
2027static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2028{
2029 unsigned long flag = 0UL;
2030
2031 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2032
2033 if (rec->flags & FTRACE_FL_DISABLED)
2034 return FTRACE_UPDATE_IGNORE;
2035
2036 /*
2037 * If we are updating calls:
2038 *
2039 * If the record has a ref count, then we need to enable it
2040 * because someone is using it.
2041 *
2042 * Otherwise we make sure its disabled.
2043 *
2044 * If we are disabling calls, then disable all records that
2045 * are enabled.
2046 */
2047 if (enable && ftrace_rec_count(rec))
2048 flag = FTRACE_FL_ENABLED;
2049
2050 /*
2051 * If enabling and the REGS flag does not match the REGS_EN, or
2052 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2053 * this record. Set flags to fail the compare against ENABLED.
2054 */
2055 if (flag) {
2056 if (!(rec->flags & FTRACE_FL_REGS) !=
2057 !(rec->flags & FTRACE_FL_REGS_EN))
2058 flag |= FTRACE_FL_REGS;
2059
2060 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2061 !(rec->flags & FTRACE_FL_TRAMP_EN))
2062 flag |= FTRACE_FL_TRAMP;
2063 }
2064
2065 /* If the state of this record hasn't changed, then do nothing */
2066 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2067 return FTRACE_UPDATE_IGNORE;
2068
2069 if (flag) {
2070 /* Save off if rec is being enabled (for return value) */
2071 flag ^= rec->flags & FTRACE_FL_ENABLED;
2072
2073 if (update) {
2074 rec->flags |= FTRACE_FL_ENABLED;
2075 if (flag & FTRACE_FL_REGS) {
2076 if (rec->flags & FTRACE_FL_REGS)
2077 rec->flags |= FTRACE_FL_REGS_EN;
2078 else
2079 rec->flags &= ~FTRACE_FL_REGS_EN;
2080 }
2081 if (flag & FTRACE_FL_TRAMP) {
2082 if (rec->flags & FTRACE_FL_TRAMP)
2083 rec->flags |= FTRACE_FL_TRAMP_EN;
2084 else
2085 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2086 }
2087 }
2088
2089 /*
2090 * If this record is being updated from a nop, then
2091 * return UPDATE_MAKE_CALL.
2092 * Otherwise,
2093 * return UPDATE_MODIFY_CALL to tell the caller to convert
2094 * from the save regs, to a non-save regs function or
2095 * vice versa, or from a trampoline call.
2096 */
2097 if (flag & FTRACE_FL_ENABLED) {
2098 ftrace_bug_type = FTRACE_BUG_CALL;
2099 return FTRACE_UPDATE_MAKE_CALL;
2100 }
2101
2102 ftrace_bug_type = FTRACE_BUG_UPDATE;
2103 return FTRACE_UPDATE_MODIFY_CALL;
2104 }
2105
2106 if (update) {
2107 /* If there's no more users, clear all flags */
2108 if (!ftrace_rec_count(rec))
2109 rec->flags = 0;
2110 else
2111 /*
2112 * Just disable the record, but keep the ops TRAMP
2113 * and REGS states. The _EN flags must be disabled though.
2114 */
2115 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2116 FTRACE_FL_REGS_EN);
2117 }
2118
2119 ftrace_bug_type = FTRACE_BUG_NOP;
2120 return FTRACE_UPDATE_MAKE_NOP;
2121}
2122
2123/**
2124 * ftrace_update_record, set a record that now is tracing or not
2125 * @rec: the record to update
2126 * @enable: set to 1 if the record is tracing, zero to force disable
2127 *
2128 * The records that represent all functions that can be traced need
2129 * to be updated when tracing has been enabled.
2130 */
2131int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2132{
2133 return ftrace_check_record(rec, enable, 1);
2134}
2135
2136/**
2137 * ftrace_test_record, check if the record has been enabled or not
2138 * @rec: the record to test
2139 * @enable: set to 1 to check if enabled, 0 if it is disabled
2140 *
2141 * The arch code may need to test if a record is already set to
2142 * tracing to determine how to modify the function code that it
2143 * represents.
2144 */
2145int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2146{
2147 return ftrace_check_record(rec, enable, 0);
2148}
2149
2150static struct ftrace_ops *
2151ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2152{
2153 struct ftrace_ops *op;
2154 unsigned long ip = rec->ip;
2155
2156 do_for_each_ftrace_op(op, ftrace_ops_list) {
2157
2158 if (!op->trampoline)
2159 continue;
2160
2161 if (hash_contains_ip(ip, op->func_hash))
2162 return op;
2163 } while_for_each_ftrace_op(op);
2164
2165 return NULL;
2166}
2167
2168static struct ftrace_ops *
2169ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2170 struct ftrace_ops *op)
2171{
2172 unsigned long ip = rec->ip;
2173
2174 while_for_each_ftrace_op(op) {
2175
2176 if (!op->trampoline)
2177 continue;
2178
2179 if (hash_contains_ip(ip, op->func_hash))
2180 return op;
2181 }
2182
2183 return NULL;
2184}
2185
2186static struct ftrace_ops *
2187ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2188{
2189 struct ftrace_ops *op;
2190 unsigned long ip = rec->ip;
2191
2192 /*
2193 * Need to check removed ops first.
2194 * If they are being removed, and this rec has a tramp,
2195 * and this rec is in the ops list, then it would be the
2196 * one with the tramp.
2197 */
2198 if (removed_ops) {
2199 if (hash_contains_ip(ip, &removed_ops->old_hash))
2200 return removed_ops;
2201 }
2202
2203 /*
2204 * Need to find the current trampoline for a rec.
2205 * Now, a trampoline is only attached to a rec if there
2206 * was a single 'ops' attached to it. But this can be called
2207 * when we are adding another op to the rec or removing the
2208 * current one. Thus, if the op is being added, we can
2209 * ignore it because it hasn't attached itself to the rec
2210 * yet.
2211 *
2212 * If an ops is being modified (hooking to different functions)
2213 * then we don't care about the new functions that are being
2214 * added, just the old ones (that are probably being removed).
2215 *
2216 * If we are adding an ops to a function that already is using
2217 * a trampoline, it needs to be removed (trampolines are only
2218 * for single ops connected), then an ops that is not being
2219 * modified also needs to be checked.
2220 */
2221 do_for_each_ftrace_op(op, ftrace_ops_list) {
2222
2223 if (!op->trampoline)
2224 continue;
2225
2226 /*
2227 * If the ops is being added, it hasn't gotten to
2228 * the point to be removed from this tree yet.
2229 */
2230 if (op->flags & FTRACE_OPS_FL_ADDING)
2231 continue;
2232
2233
2234 /*
2235 * If the ops is being modified and is in the old
2236 * hash, then it is probably being removed from this
2237 * function.
2238 */
2239 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2240 hash_contains_ip(ip, &op->old_hash))
2241 return op;
2242 /*
2243 * If the ops is not being added or modified, and it's
2244 * in its normal filter hash, then this must be the one
2245 * we want!
2246 */
2247 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2248 hash_contains_ip(ip, op->func_hash))
2249 return op;
2250
2251 } while_for_each_ftrace_op(op);
2252
2253 return NULL;
2254}
2255
2256static struct ftrace_ops *
2257ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2258{
2259 struct ftrace_ops *op;
2260 unsigned long ip = rec->ip;
2261
2262 do_for_each_ftrace_op(op, ftrace_ops_list) {
2263 /* pass rec in as regs to have non-NULL val */
2264 if (hash_contains_ip(ip, op->func_hash))
2265 return op;
2266 } while_for_each_ftrace_op(op);
2267
2268 return NULL;
2269}
2270
2271/**
2272 * ftrace_get_addr_new - Get the call address to set to
2273 * @rec: The ftrace record descriptor
2274 *
2275 * If the record has the FTRACE_FL_REGS set, that means that it
2276 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2277 * is not not set, then it wants to convert to the normal callback.
2278 *
2279 * Returns the address of the trampoline to set to
2280 */
2281unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2282{
2283 struct ftrace_ops *ops;
2284
2285 /* Trampolines take precedence over regs */
2286 if (rec->flags & FTRACE_FL_TRAMP) {
2287 ops = ftrace_find_tramp_ops_new(rec);
2288 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2289 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2290 (void *)rec->ip, (void *)rec->ip, rec->flags);
2291 /* Ftrace is shutting down, return anything */
2292 return (unsigned long)FTRACE_ADDR;
2293 }
2294 return ops->trampoline;
2295 }
2296
2297 if (rec->flags & FTRACE_FL_REGS)
2298 return (unsigned long)FTRACE_REGS_ADDR;
2299 else
2300 return (unsigned long)FTRACE_ADDR;
2301}
2302
2303/**
2304 * ftrace_get_addr_curr - Get the call address that is already there
2305 * @rec: The ftrace record descriptor
2306 *
2307 * The FTRACE_FL_REGS_EN is set when the record already points to
2308 * a function that saves all the regs. Basically the '_EN' version
2309 * represents the current state of the function.
2310 *
2311 * Returns the address of the trampoline that is currently being called
2312 */
2313unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2314{
2315 struct ftrace_ops *ops;
2316
2317 /* Trampolines take precedence over regs */
2318 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2319 ops = ftrace_find_tramp_ops_curr(rec);
2320 if (FTRACE_WARN_ON(!ops)) {
2321 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2322 (void *)rec->ip, (void *)rec->ip);
2323 /* Ftrace is shutting down, return anything */
2324 return (unsigned long)FTRACE_ADDR;
2325 }
2326 return ops->trampoline;
2327 }
2328
2329 if (rec->flags & FTRACE_FL_REGS_EN)
2330 return (unsigned long)FTRACE_REGS_ADDR;
2331 else
2332 return (unsigned long)FTRACE_ADDR;
2333}
2334
2335static int
2336__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2337{
2338 unsigned long ftrace_old_addr;
2339 unsigned long ftrace_addr;
2340 int ret;
2341
2342 ftrace_addr = ftrace_get_addr_new(rec);
2343
2344 /* This needs to be done before we call ftrace_update_record */
2345 ftrace_old_addr = ftrace_get_addr_curr(rec);
2346
2347 ret = ftrace_update_record(rec, enable);
2348
2349 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2350
2351 switch (ret) {
2352 case FTRACE_UPDATE_IGNORE:
2353 return 0;
2354
2355 case FTRACE_UPDATE_MAKE_CALL:
2356 ftrace_bug_type = FTRACE_BUG_CALL;
2357 return ftrace_make_call(rec, ftrace_addr);
2358
2359 case FTRACE_UPDATE_MAKE_NOP:
2360 ftrace_bug_type = FTRACE_BUG_NOP;
2361 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2362
2363 case FTRACE_UPDATE_MODIFY_CALL:
2364 ftrace_bug_type = FTRACE_BUG_UPDATE;
2365 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2366 }
2367
2368 return -1; /* unknow ftrace bug */
2369}
2370
2371void __weak ftrace_replace_code(int enable)
2372{
2373 struct dyn_ftrace *rec;
2374 struct ftrace_page *pg;
2375 int failed;
2376
2377 if (unlikely(ftrace_disabled))
2378 return;
2379
2380 do_for_each_ftrace_rec(pg, rec) {
2381 failed = __ftrace_replace_code(rec, enable);
2382 if (failed) {
2383 ftrace_bug(failed, rec);
2384 /* Stop processing */
2385 return;
2386 }
2387 } while_for_each_ftrace_rec();
2388}
2389
2390struct ftrace_rec_iter {
2391 struct ftrace_page *pg;
2392 int index;
2393};
2394
2395/**
2396 * ftrace_rec_iter_start, start up iterating over traced functions
2397 *
2398 * Returns an iterator handle that is used to iterate over all
2399 * the records that represent address locations where functions
2400 * are traced.
2401 *
2402 * May return NULL if no records are available.
2403 */
2404struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2405{
2406 /*
2407 * We only use a single iterator.
2408 * Protected by the ftrace_lock mutex.
2409 */
2410 static struct ftrace_rec_iter ftrace_rec_iter;
2411 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2412
2413 iter->pg = ftrace_pages_start;
2414 iter->index = 0;
2415
2416 /* Could have empty pages */
2417 while (iter->pg && !iter->pg->index)
2418 iter->pg = iter->pg->next;
2419
2420 if (!iter->pg)
2421 return NULL;
2422
2423 return iter;
2424}
2425
2426/**
2427 * ftrace_rec_iter_next, get the next record to process.
2428 * @iter: The handle to the iterator.
2429 *
2430 * Returns the next iterator after the given iterator @iter.
2431 */
2432struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2433{
2434 iter->index++;
2435
2436 if (iter->index >= iter->pg->index) {
2437 iter->pg = iter->pg->next;
2438 iter->index = 0;
2439
2440 /* Could have empty pages */
2441 while (iter->pg && !iter->pg->index)
2442 iter->pg = iter->pg->next;
2443 }
2444
2445 if (!iter->pg)
2446 return NULL;
2447
2448 return iter;
2449}
2450
2451/**
2452 * ftrace_rec_iter_record, get the record at the iterator location
2453 * @iter: The current iterator location
2454 *
2455 * Returns the record that the current @iter is at.
2456 */
2457struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2458{
2459 return &iter->pg->records[iter->index];
2460}
2461
2462static int
2463ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2464{
2465 int ret;
2466
2467 if (unlikely(ftrace_disabled))
2468 return 0;
2469
2470 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2471 if (ret) {
2472 ftrace_bug_type = FTRACE_BUG_INIT;
2473 ftrace_bug(ret, rec);
2474 return 0;
2475 }
2476 return 1;
2477}
2478
2479/*
2480 * archs can override this function if they must do something
2481 * before the modifying code is performed.
2482 */
2483int __weak ftrace_arch_code_modify_prepare(void)
2484{
2485 return 0;
2486}
2487
2488/*
2489 * archs can override this function if they must do something
2490 * after the modifying code is performed.
2491 */
2492int __weak ftrace_arch_code_modify_post_process(void)
2493{
2494 return 0;
2495}
2496
2497void ftrace_modify_all_code(int command)
2498{
2499 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2500 int err = 0;
2501
2502 /*
2503 * If the ftrace_caller calls a ftrace_ops func directly,
2504 * we need to make sure that it only traces functions it
2505 * expects to trace. When doing the switch of functions,
2506 * we need to update to the ftrace_ops_list_func first
2507 * before the transition between old and new calls are set,
2508 * as the ftrace_ops_list_func will check the ops hashes
2509 * to make sure the ops are having the right functions
2510 * traced.
2511 */
2512 if (update) {
2513 err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2514 if (FTRACE_WARN_ON(err))
2515 return;
2516 }
2517
2518 if (command & FTRACE_UPDATE_CALLS)
2519 ftrace_replace_code(1);
2520 else if (command & FTRACE_DISABLE_CALLS)
2521 ftrace_replace_code(0);
2522
2523 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2524 function_trace_op = set_function_trace_op;
2525 smp_wmb();
2526 /* If irqs are disabled, we are in stop machine */
2527 if (!irqs_disabled())
2528 smp_call_function(ftrace_sync_ipi, NULL, 1);
2529 err = ftrace_update_ftrace_func(ftrace_trace_function);
2530 if (FTRACE_WARN_ON(err))
2531 return;
2532 }
2533
2534 if (command & FTRACE_START_FUNC_RET)
2535 err = ftrace_enable_ftrace_graph_caller();
2536 else if (command & FTRACE_STOP_FUNC_RET)
2537 err = ftrace_disable_ftrace_graph_caller();
2538 FTRACE_WARN_ON(err);
2539}
2540
2541static int __ftrace_modify_code(void *data)
2542{
2543 int *command = data;
2544
2545 ftrace_modify_all_code(*command);
2546
2547 return 0;
2548}
2549
2550/**
2551 * ftrace_run_stop_machine, go back to the stop machine method
2552 * @command: The command to tell ftrace what to do
2553 *
2554 * If an arch needs to fall back to the stop machine method, the
2555 * it can call this function.
2556 */
2557void ftrace_run_stop_machine(int command)
2558{
2559 stop_machine(__ftrace_modify_code, &command, NULL);
2560}
2561
2562/**
2563 * arch_ftrace_update_code, modify the code to trace or not trace
2564 * @command: The command that needs to be done
2565 *
2566 * Archs can override this function if it does not need to
2567 * run stop_machine() to modify code.
2568 */
2569void __weak arch_ftrace_update_code(int command)
2570{
2571 ftrace_run_stop_machine(command);
2572}
2573
2574static void ftrace_run_update_code(int command)
2575{
2576 int ret;
2577
2578 ret = ftrace_arch_code_modify_prepare();
2579 FTRACE_WARN_ON(ret);
2580 if (ret)
2581 return;
2582
2583 /*
2584 * By default we use stop_machine() to modify the code.
2585 * But archs can do what ever they want as long as it
2586 * is safe. The stop_machine() is the safest, but also
2587 * produces the most overhead.
2588 */
2589 arch_ftrace_update_code(command);
2590
2591 ret = ftrace_arch_code_modify_post_process();
2592 FTRACE_WARN_ON(ret);
2593}
2594
2595static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2596 struct ftrace_ops_hash *old_hash)
2597{
2598 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2599 ops->old_hash.filter_hash = old_hash->filter_hash;
2600 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2601 ftrace_run_update_code(command);
2602 ops->old_hash.filter_hash = NULL;
2603 ops->old_hash.notrace_hash = NULL;
2604 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2605}
2606
2607static ftrace_func_t saved_ftrace_func;
2608static int ftrace_start_up;
2609
2610void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2611{
2612}
2613
2614static void per_cpu_ops_free(struct ftrace_ops *ops)
2615{
2616 free_percpu(ops->disabled);
2617}
2618
2619static void ftrace_startup_enable(int command)
2620{
2621 if (saved_ftrace_func != ftrace_trace_function) {
2622 saved_ftrace_func = ftrace_trace_function;
2623 command |= FTRACE_UPDATE_TRACE_FUNC;
2624 }
2625
2626 if (!command || !ftrace_enabled)
2627 return;
2628
2629 ftrace_run_update_code(command);
2630}
2631
2632static void ftrace_startup_all(int command)
2633{
2634 update_all_ops = true;
2635 ftrace_startup_enable(command);
2636 update_all_ops = false;
2637}
2638
2639static int ftrace_startup(struct ftrace_ops *ops, int command)
2640{
2641 int ret;
2642
2643 if (unlikely(ftrace_disabled))
2644 return -ENODEV;
2645
2646 ret = __register_ftrace_function(ops);
2647 if (ret)
2648 return ret;
2649
2650 ftrace_start_up++;
2651
2652 /*
2653 * Note that ftrace probes uses this to start up
2654 * and modify functions it will probe. But we still
2655 * set the ADDING flag for modification, as probes
2656 * do not have trampolines. If they add them in the
2657 * future, then the probes will need to distinguish
2658 * between adding and updating probes.
2659 */
2660 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2661
2662 ret = ftrace_hash_ipmodify_enable(ops);
2663 if (ret < 0) {
2664 /* Rollback registration process */
2665 __unregister_ftrace_function(ops);
2666 ftrace_start_up--;
2667 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2668 return ret;
2669 }
2670
2671 if (ftrace_hash_rec_enable(ops, 1))
2672 command |= FTRACE_UPDATE_CALLS;
2673
2674 ftrace_startup_enable(command);
2675
2676 ops->flags &= ~FTRACE_OPS_FL_ADDING;
2677
2678 return 0;
2679}
2680
2681static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2682{
2683 int ret;
2684
2685 if (unlikely(ftrace_disabled))
2686 return -ENODEV;
2687
2688 ret = __unregister_ftrace_function(ops);
2689 if (ret)
2690 return ret;
2691
2692 ftrace_start_up--;
2693 /*
2694 * Just warn in case of unbalance, no need to kill ftrace, it's not
2695 * critical but the ftrace_call callers may be never nopped again after
2696 * further ftrace uses.
2697 */
2698 WARN_ON_ONCE(ftrace_start_up < 0);
2699
2700 /* Disabling ipmodify never fails */
2701 ftrace_hash_ipmodify_disable(ops);
2702
2703 if (ftrace_hash_rec_disable(ops, 1))
2704 command |= FTRACE_UPDATE_CALLS;
2705
2706 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2707
2708 if (saved_ftrace_func != ftrace_trace_function) {
2709 saved_ftrace_func = ftrace_trace_function;
2710 command |= FTRACE_UPDATE_TRACE_FUNC;
2711 }
2712
2713 if (!command || !ftrace_enabled) {
2714 /*
2715 * If these are per_cpu ops, they still need their
2716 * per_cpu field freed. Since, function tracing is
2717 * not currently active, we can just free them
2718 * without synchronizing all CPUs.
2719 */
2720 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2721 per_cpu_ops_free(ops);
2722 return 0;
2723 }
2724
2725 /*
2726 * If the ops uses a trampoline, then it needs to be
2727 * tested first on update.
2728 */
2729 ops->flags |= FTRACE_OPS_FL_REMOVING;
2730 removed_ops = ops;
2731
2732 /* The trampoline logic checks the old hashes */
2733 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2734 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2735
2736 ftrace_run_update_code(command);
2737
2738 /*
2739 * If there's no more ops registered with ftrace, run a
2740 * sanity check to make sure all rec flags are cleared.
2741 */
2742 if (ftrace_ops_list == &ftrace_list_end) {
2743 struct ftrace_page *pg;
2744 struct dyn_ftrace *rec;
2745
2746 do_for_each_ftrace_rec(pg, rec) {
2747 if (FTRACE_WARN_ON_ONCE(rec->flags))
2748 pr_warn(" %pS flags:%lx\n",
2749 (void *)rec->ip, rec->flags);
2750 } while_for_each_ftrace_rec();
2751 }
2752
2753 ops->old_hash.filter_hash = NULL;
2754 ops->old_hash.notrace_hash = NULL;
2755
2756 removed_ops = NULL;
2757 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2758
2759 /*
2760 * Dynamic ops may be freed, we must make sure that all
2761 * callers are done before leaving this function.
2762 * The same goes for freeing the per_cpu data of the per_cpu
2763 * ops.
2764 *
2765 * Again, normal synchronize_sched() is not good enough.
2766 * We need to do a hard force of sched synchronization.
2767 * This is because we use preempt_disable() to do RCU, but
2768 * the function tracers can be called where RCU is not watching
2769 * (like before user_exit()). We can not rely on the RCU
2770 * infrastructure to do the synchronization, thus we must do it
2771 * ourselves.
2772 */
2773 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) {
2774 schedule_on_each_cpu(ftrace_sync);
2775
2776 arch_ftrace_trampoline_free(ops);
2777
2778 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2779 per_cpu_ops_free(ops);
2780 }
2781
2782 return 0;
2783}
2784
2785static void ftrace_startup_sysctl(void)
2786{
2787 int command;
2788
2789 if (unlikely(ftrace_disabled))
2790 return;
2791
2792 /* Force update next time */
2793 saved_ftrace_func = NULL;
2794 /* ftrace_start_up is true if we want ftrace running */
2795 if (ftrace_start_up) {
2796 command = FTRACE_UPDATE_CALLS;
2797 if (ftrace_graph_active)
2798 command |= FTRACE_START_FUNC_RET;
2799 ftrace_startup_enable(command);
2800 }
2801}
2802
2803static void ftrace_shutdown_sysctl(void)
2804{
2805 int command;
2806
2807 if (unlikely(ftrace_disabled))
2808 return;
2809
2810 /* ftrace_start_up is true if ftrace is running */
2811 if (ftrace_start_up) {
2812 command = FTRACE_DISABLE_CALLS;
2813 if (ftrace_graph_active)
2814 command |= FTRACE_STOP_FUNC_RET;
2815 ftrace_run_update_code(command);
2816 }
2817}
2818
2819static cycle_t ftrace_update_time;
2820unsigned long ftrace_update_tot_cnt;
2821
2822static inline int ops_traces_mod(struct ftrace_ops *ops)
2823{
2824 /*
2825 * Filter_hash being empty will default to trace module.
2826 * But notrace hash requires a test of individual module functions.
2827 */
2828 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2829 ftrace_hash_empty(ops->func_hash->notrace_hash);
2830}
2831
2832/*
2833 * Check if the current ops references the record.
2834 *
2835 * If the ops traces all functions, then it was already accounted for.
2836 * If the ops does not trace the current record function, skip it.
2837 * If the ops ignores the function via notrace filter, skip it.
2838 */
2839static inline bool
2840ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2841{
2842 /* If ops isn't enabled, ignore it */
2843 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2844 return 0;
2845
2846 /* If ops traces all then it includes this function */
2847 if (ops_traces_mod(ops))
2848 return 1;
2849
2850 /* The function must be in the filter */
2851 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2852 !ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2853 return 0;
2854
2855 /* If in notrace hash, we ignore it too */
2856 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2857 return 0;
2858
2859 return 1;
2860}
2861
2862static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2863{
2864 struct ftrace_page *pg;
2865 struct dyn_ftrace *p;
2866 cycle_t start, stop;
2867 unsigned long update_cnt = 0;
2868 unsigned long rec_flags = 0;
2869 int i;
2870
2871 start = ftrace_now(raw_smp_processor_id());
2872
2873 /*
2874 * When a module is loaded, this function is called to convert
2875 * the calls to mcount in its text to nops, and also to create
2876 * an entry in the ftrace data. Now, if ftrace is activated
2877 * after this call, but before the module sets its text to
2878 * read-only, the modification of enabling ftrace can fail if
2879 * the read-only is done while ftrace is converting the calls.
2880 * To prevent this, the module's records are set as disabled
2881 * and will be enabled after the call to set the module's text
2882 * to read-only.
2883 */
2884 if (mod)
2885 rec_flags |= FTRACE_FL_DISABLED;
2886
2887 for (pg = new_pgs; pg; pg = pg->next) {
2888
2889 for (i = 0; i < pg->index; i++) {
2890
2891 /* If something went wrong, bail without enabling anything */
2892 if (unlikely(ftrace_disabled))
2893 return -1;
2894
2895 p = &pg->records[i];
2896 p->flags = rec_flags;
2897
2898 /*
2899 * Do the initial record conversion from mcount jump
2900 * to the NOP instructions.
2901 */
2902 if (!ftrace_code_disable(mod, p))
2903 break;
2904
2905 update_cnt++;
2906 }
2907 }
2908
2909 stop = ftrace_now(raw_smp_processor_id());
2910 ftrace_update_time = stop - start;
2911 ftrace_update_tot_cnt += update_cnt;
2912
2913 return 0;
2914}
2915
2916static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2917{
2918 int order;
2919 int cnt;
2920
2921 if (WARN_ON(!count))
2922 return -EINVAL;
2923
2924 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2925
2926 /*
2927 * We want to fill as much as possible. No more than a page
2928 * may be empty.
2929 */
2930 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2931 order--;
2932
2933 again:
2934 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2935
2936 if (!pg->records) {
2937 /* if we can't allocate this size, try something smaller */
2938 if (!order)
2939 return -ENOMEM;
2940 order >>= 1;
2941 goto again;
2942 }
2943
2944 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2945 pg->size = cnt;
2946
2947 if (cnt > count)
2948 cnt = count;
2949
2950 return cnt;
2951}
2952
2953static struct ftrace_page *
2954ftrace_allocate_pages(unsigned long num_to_init)
2955{
2956 struct ftrace_page *start_pg;
2957 struct ftrace_page *pg;
2958 int order;
2959 int cnt;
2960
2961 if (!num_to_init)
2962 return 0;
2963
2964 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2965 if (!pg)
2966 return NULL;
2967
2968 /*
2969 * Try to allocate as much as possible in one continues
2970 * location that fills in all of the space. We want to
2971 * waste as little space as possible.
2972 */
2973 for (;;) {
2974 cnt = ftrace_allocate_records(pg, num_to_init);
2975 if (cnt < 0)
2976 goto free_pages;
2977
2978 num_to_init -= cnt;
2979 if (!num_to_init)
2980 break;
2981
2982 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2983 if (!pg->next)
2984 goto free_pages;
2985
2986 pg = pg->next;
2987 }
2988
2989 return start_pg;
2990
2991 free_pages:
2992 pg = start_pg;
2993 while (pg) {
2994 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2995 free_pages((unsigned long)pg->records, order);
2996 start_pg = pg->next;
2997 kfree(pg);
2998 pg = start_pg;
2999 }
3000 pr_info("ftrace: FAILED to allocate memory for functions\n");
3001 return NULL;
3002}
3003
3004#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3005
3006struct ftrace_iterator {
3007 loff_t pos;
3008 loff_t func_pos;
3009 struct ftrace_page *pg;
3010 struct dyn_ftrace *func;
3011 struct ftrace_func_probe *probe;
3012 struct trace_parser parser;
3013 struct ftrace_hash *hash;
3014 struct ftrace_ops *ops;
3015 int hidx;
3016 int idx;
3017 unsigned flags;
3018};
3019
3020static void *
3021t_hash_next(struct seq_file *m, loff_t *pos)
3022{
3023 struct ftrace_iterator *iter = m->private;
3024 struct hlist_node *hnd = NULL;
3025 struct hlist_head *hhd;
3026
3027 (*pos)++;
3028 iter->pos = *pos;
3029
3030 if (iter->probe)
3031 hnd = &iter->probe->node;
3032 retry:
3033 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
3034 return NULL;
3035
3036 hhd = &ftrace_func_hash[iter->hidx];
3037
3038 if (hlist_empty(hhd)) {
3039 iter->hidx++;
3040 hnd = NULL;
3041 goto retry;
3042 }
3043
3044 if (!hnd)
3045 hnd = hhd->first;
3046 else {
3047 hnd = hnd->next;
3048 if (!hnd) {
3049 iter->hidx++;
3050 goto retry;
3051 }
3052 }
3053
3054 if (WARN_ON_ONCE(!hnd))
3055 return NULL;
3056
3057 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
3058
3059 return iter;
3060}
3061
3062static void *t_hash_start(struct seq_file *m, loff_t *pos)
3063{
3064 struct ftrace_iterator *iter = m->private;
3065 void *p = NULL;
3066 loff_t l;
3067
3068 if (!(iter->flags & FTRACE_ITER_DO_HASH))
3069 return NULL;
3070
3071 if (iter->func_pos > *pos)
3072 return NULL;
3073
3074 iter->hidx = 0;
3075 for (l = 0; l <= (*pos - iter->func_pos); ) {
3076 p = t_hash_next(m, &l);
3077 if (!p)
3078 break;
3079 }
3080 if (!p)
3081 return NULL;
3082
3083 /* Only set this if we have an item */
3084 iter->flags |= FTRACE_ITER_HASH;
3085
3086 return iter;
3087}
3088
3089static int
3090t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
3091{
3092 struct ftrace_func_probe *rec;
3093
3094 rec = iter->probe;
3095 if (WARN_ON_ONCE(!rec))
3096 return -EIO;
3097
3098 if (rec->ops->print)
3099 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
3100
3101 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
3102
3103 if (rec->data)
3104 seq_printf(m, ":%p", rec->data);
3105 seq_putc(m, '\n');
3106
3107 return 0;
3108}
3109
3110static void *
3111t_next(struct seq_file *m, void *v, loff_t *pos)
3112{
3113 struct ftrace_iterator *iter = m->private;
3114 struct ftrace_ops *ops = iter->ops;
3115 struct dyn_ftrace *rec = NULL;
3116
3117 if (unlikely(ftrace_disabled))
3118 return NULL;
3119
3120 if (iter->flags & FTRACE_ITER_HASH)
3121 return t_hash_next(m, pos);
3122
3123 (*pos)++;
3124 iter->pos = iter->func_pos = *pos;
3125
3126 if (iter->flags & FTRACE_ITER_PRINTALL)
3127 return t_hash_start(m, pos);
3128
3129 retry:
3130 if (iter->idx >= iter->pg->index) {
3131 if (iter->pg->next) {
3132 iter->pg = iter->pg->next;
3133 iter->idx = 0;
3134 goto retry;
3135 }
3136 } else {
3137 rec = &iter->pg->records[iter->idx++];
3138 if (((iter->flags & FTRACE_ITER_FILTER) &&
3139 !(ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))) ||
3140
3141 ((iter->flags & FTRACE_ITER_NOTRACE) &&
3142 !ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip)) ||
3143
3144 ((iter->flags & FTRACE_ITER_ENABLED) &&
3145 !(rec->flags & FTRACE_FL_ENABLED))) {
3146
3147 rec = NULL;
3148 goto retry;
3149 }
3150 }
3151
3152 if (!rec)
3153 return t_hash_start(m, pos);
3154
3155 iter->func = rec;
3156
3157 return iter;
3158}
3159
3160static void reset_iter_read(struct ftrace_iterator *iter)
3161{
3162 iter->pos = 0;
3163 iter->func_pos = 0;
3164 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
3165}
3166
3167static void *t_start(struct seq_file *m, loff_t *pos)
3168{
3169 struct ftrace_iterator *iter = m->private;
3170 struct ftrace_ops *ops = iter->ops;
3171 void *p = NULL;
3172 loff_t l;
3173
3174 mutex_lock(&ftrace_lock);
3175
3176 if (unlikely(ftrace_disabled))
3177 return NULL;
3178
3179 /*
3180 * If an lseek was done, then reset and start from beginning.
3181 */
3182 if (*pos < iter->pos)
3183 reset_iter_read(iter);
3184
3185 /*
3186 * For set_ftrace_filter reading, if we have the filter
3187 * off, we can short cut and just print out that all
3188 * functions are enabled.
3189 */
3190 if ((iter->flags & FTRACE_ITER_FILTER &&
3191 ftrace_hash_empty(ops->func_hash->filter_hash)) ||
3192 (iter->flags & FTRACE_ITER_NOTRACE &&
3193 ftrace_hash_empty(ops->func_hash->notrace_hash))) {
3194 if (*pos > 0)
3195 return t_hash_start(m, pos);
3196 iter->flags |= FTRACE_ITER_PRINTALL;
3197 /* reset in case of seek/pread */
3198 iter->flags &= ~FTRACE_ITER_HASH;
3199 return iter;
3200 }
3201
3202 if (iter->flags & FTRACE_ITER_HASH)
3203 return t_hash_start(m, pos);
3204
3205 /*
3206 * Unfortunately, we need to restart at ftrace_pages_start
3207 * every time we let go of the ftrace_mutex. This is because
3208 * those pointers can change without the lock.
3209 */
3210 iter->pg = ftrace_pages_start;
3211 iter->idx = 0;
3212 for (l = 0; l <= *pos; ) {
3213 p = t_next(m, p, &l);
3214 if (!p)
3215 break;
3216 }
3217
3218 if (!p)
3219 return t_hash_start(m, pos);
3220
3221 return iter;
3222}
3223
3224static void t_stop(struct seq_file *m, void *p)
3225{
3226 mutex_unlock(&ftrace_lock);
3227}
3228
3229void * __weak
3230arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3231{
3232 return NULL;
3233}
3234
3235static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3236 struct dyn_ftrace *rec)
3237{
3238 void *ptr;
3239
3240 ptr = arch_ftrace_trampoline_func(ops, rec);
3241 if (ptr)
3242 seq_printf(m, " ->%pS", ptr);
3243}
3244
3245static int t_show(struct seq_file *m, void *v)
3246{
3247 struct ftrace_iterator *iter = m->private;
3248 struct dyn_ftrace *rec;
3249
3250 if (iter->flags & FTRACE_ITER_HASH)
3251 return t_hash_show(m, iter);
3252
3253 if (iter->flags & FTRACE_ITER_PRINTALL) {
3254 if (iter->flags & FTRACE_ITER_NOTRACE)
3255 seq_puts(m, "#### no functions disabled ####\n");
3256 else
3257 seq_puts(m, "#### all functions enabled ####\n");
3258 return 0;
3259 }
3260
3261 rec = iter->func;
3262
3263 if (!rec)
3264 return 0;
3265
3266 seq_printf(m, "%ps", (void *)rec->ip);
3267 if (iter->flags & FTRACE_ITER_ENABLED) {
3268 struct ftrace_ops *ops;
3269
3270 seq_printf(m, " (%ld)%s%s",
3271 ftrace_rec_count(rec),
3272 rec->flags & FTRACE_FL_REGS ? " R" : " ",
3273 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ");
3274 if (rec->flags & FTRACE_FL_TRAMP_EN) {
3275 ops = ftrace_find_tramp_ops_any(rec);
3276 if (ops) {
3277 do {
3278 seq_printf(m, "\ttramp: %pS (%pS)",
3279 (void *)ops->trampoline,
3280 (void *)ops->func);
3281 add_trampoline_func(m, ops, rec);
3282 ops = ftrace_find_tramp_ops_next(rec, ops);
3283 } while (ops);
3284 } else
3285 seq_puts(m, "\ttramp: ERROR!");
3286 } else {
3287 add_trampoline_func(m, NULL, rec);
3288 }
3289 }
3290
3291 seq_putc(m, '\n');
3292
3293 return 0;
3294}
3295
3296static const struct seq_operations show_ftrace_seq_ops = {
3297 .start = t_start,
3298 .next = t_next,
3299 .stop = t_stop,
3300 .show = t_show,
3301};
3302
3303static int
3304ftrace_avail_open(struct inode *inode, struct file *file)
3305{
3306 struct ftrace_iterator *iter;
3307
3308 if (unlikely(ftrace_disabled))
3309 return -ENODEV;
3310
3311 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3312 if (iter) {
3313 iter->pg = ftrace_pages_start;
3314 iter->ops = &global_ops;
3315 }
3316
3317 return iter ? 0 : -ENOMEM;
3318}
3319
3320static int
3321ftrace_enabled_open(struct inode *inode, struct file *file)
3322{
3323 struct ftrace_iterator *iter;
3324
3325 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3326 if (iter) {
3327 iter->pg = ftrace_pages_start;
3328 iter->flags = FTRACE_ITER_ENABLED;
3329 iter->ops = &global_ops;
3330 }
3331
3332 return iter ? 0 : -ENOMEM;
3333}
3334
3335/**
3336 * ftrace_regex_open - initialize function tracer filter files
3337 * @ops: The ftrace_ops that hold the hash filters
3338 * @flag: The type of filter to process
3339 * @inode: The inode, usually passed in to your open routine
3340 * @file: The file, usually passed in to your open routine
3341 *
3342 * ftrace_regex_open() initializes the filter files for the
3343 * @ops. Depending on @flag it may process the filter hash or
3344 * the notrace hash of @ops. With this called from the open
3345 * routine, you can use ftrace_filter_write() for the write
3346 * routine if @flag has FTRACE_ITER_FILTER set, or
3347 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3348 * tracing_lseek() should be used as the lseek routine, and
3349 * release must call ftrace_regex_release().
3350 */
3351int
3352ftrace_regex_open(struct ftrace_ops *ops, int flag,
3353 struct inode *inode, struct file *file)
3354{
3355 struct ftrace_iterator *iter;
3356 struct ftrace_hash *hash;
3357 int ret = 0;
3358
3359 ftrace_ops_init(ops);
3360
3361 if (unlikely(ftrace_disabled))
3362 return -ENODEV;
3363
3364 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3365 if (!iter)
3366 return -ENOMEM;
3367
3368 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3369 kfree(iter);
3370 return -ENOMEM;
3371 }
3372
3373 iter->ops = ops;
3374 iter->flags = flag;
3375
3376 mutex_lock(&ops->func_hash->regex_lock);
3377
3378 if (flag & FTRACE_ITER_NOTRACE)
3379 hash = ops->func_hash->notrace_hash;
3380 else
3381 hash = ops->func_hash->filter_hash;
3382
3383 if (file->f_mode & FMODE_WRITE) {
3384 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3385
3386 if (file->f_flags & O_TRUNC)
3387 iter->hash = alloc_ftrace_hash(size_bits);
3388 else
3389 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3390
3391 if (!iter->hash) {
3392 trace_parser_put(&iter->parser);
3393 kfree(iter);
3394 ret = -ENOMEM;
3395 goto out_unlock;
3396 }
3397 }
3398
3399 if (file->f_mode & FMODE_READ) {
3400 iter->pg = ftrace_pages_start;
3401
3402 ret = seq_open(file, &show_ftrace_seq_ops);
3403 if (!ret) {
3404 struct seq_file *m = file->private_data;
3405 m->private = iter;
3406 } else {
3407 /* Failed */
3408 free_ftrace_hash(iter->hash);
3409 trace_parser_put(&iter->parser);
3410 kfree(iter);
3411 }
3412 } else
3413 file->private_data = iter;
3414
3415 out_unlock:
3416 mutex_unlock(&ops->func_hash->regex_lock);
3417
3418 return ret;
3419}
3420
3421static int
3422ftrace_filter_open(struct inode *inode, struct file *file)
3423{
3424 struct ftrace_ops *ops = inode->i_private;
3425
3426 return ftrace_regex_open(ops,
3427 FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
3428 inode, file);
3429}
3430
3431static int
3432ftrace_notrace_open(struct inode *inode, struct file *file)
3433{
3434 struct ftrace_ops *ops = inode->i_private;
3435
3436 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3437 inode, file);
3438}
3439
3440/* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3441struct ftrace_glob {
3442 char *search;
3443 unsigned len;
3444 int type;
3445};
3446
3447static int ftrace_match(char *str, struct ftrace_glob *g)
3448{
3449 int matched = 0;
3450 int slen;
3451
3452 switch (g->type) {
3453 case MATCH_FULL:
3454 if (strcmp(str, g->search) == 0)
3455 matched = 1;
3456 break;
3457 case MATCH_FRONT_ONLY:
3458 if (strncmp(str, g->search, g->len) == 0)
3459 matched = 1;
3460 break;
3461 case MATCH_MIDDLE_ONLY:
3462 if (strstr(str, g->search))
3463 matched = 1;
3464 break;
3465 case MATCH_END_ONLY:
3466 slen = strlen(str);
3467 if (slen >= g->len &&
3468 memcmp(str + slen - g->len, g->search, g->len) == 0)
3469 matched = 1;
3470 break;
3471 }
3472
3473 return matched;
3474}
3475
3476static int
3477enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3478{
3479 struct ftrace_func_entry *entry;
3480 int ret = 0;
3481
3482 entry = ftrace_lookup_ip(hash, rec->ip);
3483 if (clear_filter) {
3484 /* Do nothing if it doesn't exist */
3485 if (!entry)
3486 return 0;
3487
3488 free_hash_entry(hash, entry);
3489 } else {
3490 /* Do nothing if it exists */
3491 if (entry)
3492 return 0;
3493
3494 ret = add_hash_entry(hash, rec->ip);
3495 }
3496 return ret;
3497}
3498
3499static int
3500ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3501 struct ftrace_glob *mod_g, int exclude_mod)
3502{
3503 char str[KSYM_SYMBOL_LEN];
3504 char *modname;
3505
3506 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3507
3508 if (mod_g) {
3509 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3510
3511 /* blank module name to match all modules */
3512 if (!mod_g->len) {
3513 /* blank module globbing: modname xor exclude_mod */
3514 if ((!exclude_mod) != (!modname))
3515 goto func_match;
3516 return 0;
3517 }
3518
3519 /* not matching the module */
3520 if (!modname || !mod_matches) {
3521 if (exclude_mod)
3522 goto func_match;
3523 else
3524 return 0;
3525 }
3526
3527 if (mod_matches && exclude_mod)
3528 return 0;
3529
3530func_match:
3531 /* blank search means to match all funcs in the mod */
3532 if (!func_g->len)
3533 return 1;
3534 }
3535
3536 return ftrace_match(str, func_g);
3537}
3538
3539static int
3540match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3541{
3542 struct ftrace_page *pg;
3543 struct dyn_ftrace *rec;
3544 struct ftrace_glob func_g = { .type = MATCH_FULL };
3545 struct ftrace_glob mod_g = { .type = MATCH_FULL };
3546 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3547 int exclude_mod = 0;
3548 int found = 0;
3549 int ret;
3550 int clear_filter;
3551
3552 if (func) {
3553 func_g.type = filter_parse_regex(func, len, &func_g.search,
3554 &clear_filter);
3555 func_g.len = strlen(func_g.search);
3556 }
3557
3558 if (mod) {
3559 mod_g.type = filter_parse_regex(mod, strlen(mod),
3560 &mod_g.search, &exclude_mod);
3561 mod_g.len = strlen(mod_g.search);
3562 }
3563
3564 mutex_lock(&ftrace_lock);
3565
3566 if (unlikely(ftrace_disabled))
3567 goto out_unlock;
3568
3569 do_for_each_ftrace_rec(pg, rec) {
3570 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3571 ret = enter_record(hash, rec, clear_filter);
3572 if (ret < 0) {
3573 found = ret;
3574 goto out_unlock;
3575 }
3576 found = 1;
3577 }
3578 } while_for_each_ftrace_rec();
3579 out_unlock:
3580 mutex_unlock(&ftrace_lock);
3581
3582 return found;
3583}
3584
3585static int
3586ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3587{
3588 return match_records(hash, buff, len, NULL);
3589}
3590
3591
3592/*
3593 * We register the module command as a template to show others how
3594 * to register the a command as well.
3595 */
3596
3597static int
3598ftrace_mod_callback(struct ftrace_hash *hash,
3599 char *func, char *cmd, char *module, int enable)
3600{
3601 int ret;
3602
3603 /*
3604 * cmd == 'mod' because we only registered this func
3605 * for the 'mod' ftrace_func_command.
3606 * But if you register one func with multiple commands,
3607 * you can tell which command was used by the cmd
3608 * parameter.
3609 */
3610 ret = match_records(hash, func, strlen(func), module);
3611 if (!ret)
3612 return -EINVAL;
3613 if (ret < 0)
3614 return ret;
3615 return 0;
3616}
3617
3618static struct ftrace_func_command ftrace_mod_cmd = {
3619 .name = "mod",
3620 .func = ftrace_mod_callback,
3621};
3622
3623static int __init ftrace_mod_cmd_init(void)
3624{
3625 return register_ftrace_command(&ftrace_mod_cmd);
3626}
3627core_initcall(ftrace_mod_cmd_init);
3628
3629static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
3630 struct ftrace_ops *op, struct pt_regs *pt_regs)
3631{
3632 struct ftrace_func_probe *entry;
3633 struct hlist_head *hhd;
3634 unsigned long key;
3635
3636 key = hash_long(ip, FTRACE_HASH_BITS);
3637
3638 hhd = &ftrace_func_hash[key];
3639
3640 if (hlist_empty(hhd))
3641 return;
3642
3643 /*
3644 * Disable preemption for these calls to prevent a RCU grace
3645 * period. This syncs the hash iteration and freeing of items
3646 * on the hash. rcu_read_lock is too dangerous here.
3647 */
3648 preempt_disable_notrace();
3649 hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3650 if (entry->ip == ip)
3651 entry->ops->func(ip, parent_ip, &entry->data);
3652 }
3653 preempt_enable_notrace();
3654}
3655
3656static struct ftrace_ops trace_probe_ops __read_mostly =
3657{
3658 .func = function_trace_probe_call,
3659 .flags = FTRACE_OPS_FL_INITIALIZED,
3660 INIT_OPS_HASH(trace_probe_ops)
3661};
3662
3663static int ftrace_probe_registered;
3664
3665static void __enable_ftrace_function_probe(struct ftrace_ops_hash *old_hash)
3666{
3667 int ret;
3668 int i;
3669
3670 if (ftrace_probe_registered) {
3671 /* still need to update the function call sites */
3672 if (ftrace_enabled)
3673 ftrace_run_modify_code(&trace_probe_ops, FTRACE_UPDATE_CALLS,
3674 old_hash);
3675 return;
3676 }
3677
3678 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3679 struct hlist_head *hhd = &ftrace_func_hash[i];
3680 if (hhd->first)
3681 break;
3682 }
3683 /* Nothing registered? */
3684 if (i == FTRACE_FUNC_HASHSIZE)
3685 return;
3686
3687 ret = ftrace_startup(&trace_probe_ops, 0);
3688
3689 ftrace_probe_registered = 1;
3690}
3691
3692static void __disable_ftrace_function_probe(void)
3693{
3694 int i;
3695
3696 if (!ftrace_probe_registered)
3697 return;
3698
3699 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3700 struct hlist_head *hhd = &ftrace_func_hash[i];
3701 if (hhd->first)
3702 return;
3703 }
3704
3705 /* no more funcs left */
3706 ftrace_shutdown(&trace_probe_ops, 0);
3707
3708 ftrace_probe_registered = 0;
3709}
3710
3711
3712static void ftrace_free_entry(struct ftrace_func_probe *entry)
3713{
3714 if (entry->ops->free)
3715 entry->ops->free(entry->ops, entry->ip, &entry->data);
3716 kfree(entry);
3717}
3718
3719int
3720register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3721 void *data)
3722{
3723 struct ftrace_ops_hash old_hash_ops;
3724 struct ftrace_func_probe *entry;
3725 struct ftrace_glob func_g;
3726 struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3727 struct ftrace_hash *old_hash = *orig_hash;
3728 struct ftrace_hash *hash;
3729 struct ftrace_page *pg;
3730 struct dyn_ftrace *rec;
3731 int not;
3732 unsigned long key;
3733 int count = 0;
3734 int ret;
3735
3736 func_g.type = filter_parse_regex(glob, strlen(glob),
3737 &func_g.search, ¬);
3738 func_g.len = strlen(func_g.search);
3739
3740 /* we do not support '!' for function probes */
3741 if (WARN_ON(not))
3742 return -EINVAL;
3743
3744 mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3745
3746 old_hash_ops.filter_hash = old_hash;
3747 /* Probes only have filters */
3748 old_hash_ops.notrace_hash = NULL;
3749
3750 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
3751 if (!hash) {
3752 count = -ENOMEM;
3753 goto out;
3754 }
3755
3756 if (unlikely(ftrace_disabled)) {
3757 count = -ENODEV;
3758 goto out;
3759 }
3760
3761 mutex_lock(&ftrace_lock);
3762
3763 do_for_each_ftrace_rec(pg, rec) {
3764
3765 if (!ftrace_match_record(rec, &func_g, NULL, 0))
3766 continue;
3767
3768 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3769 if (!entry) {
3770 /* If we did not process any, then return error */
3771 if (!count)
3772 count = -ENOMEM;
3773 goto out_unlock;
3774 }
3775
3776 count++;
3777
3778 entry->data = data;
3779
3780 /*
3781 * The caller might want to do something special
3782 * for each function we find. We call the callback
3783 * to give the caller an opportunity to do so.
3784 */
3785 if (ops->init) {
3786 if (ops->init(ops, rec->ip, &entry->data) < 0) {
3787 /* caller does not like this func */
3788 kfree(entry);
3789 continue;
3790 }
3791 }
3792
3793 ret = enter_record(hash, rec, 0);
3794 if (ret < 0) {
3795 kfree(entry);
3796 count = ret;
3797 goto out_unlock;
3798 }
3799
3800 entry->ops = ops;
3801 entry->ip = rec->ip;
3802
3803 key = hash_long(entry->ip, FTRACE_HASH_BITS);
3804 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3805
3806 } while_for_each_ftrace_rec();
3807
3808 ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3809
3810 __enable_ftrace_function_probe(&old_hash_ops);
3811
3812 if (!ret)
3813 free_ftrace_hash_rcu(old_hash);
3814 else
3815 count = ret;
3816
3817 out_unlock:
3818 mutex_unlock(&ftrace_lock);
3819 out:
3820 mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3821 free_ftrace_hash(hash);
3822
3823 return count;
3824}
3825
3826enum {
3827 PROBE_TEST_FUNC = 1,
3828 PROBE_TEST_DATA = 2
3829};
3830
3831static void
3832__unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3833 void *data, int flags)
3834{
3835 struct ftrace_func_entry *rec_entry;
3836 struct ftrace_func_probe *entry;
3837 struct ftrace_func_probe *p;
3838 struct ftrace_glob func_g;
3839 struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3840 struct ftrace_hash *old_hash = *orig_hash;
3841 struct list_head free_list;
3842 struct ftrace_hash *hash;
3843 struct hlist_node *tmp;
3844 char str[KSYM_SYMBOL_LEN];
3845 int i, ret;
3846
3847 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3848 func_g.search = NULL;
3849 else if (glob) {
3850 int not;
3851
3852 func_g.type = filter_parse_regex(glob, strlen(glob),
3853 &func_g.search, ¬);
3854 func_g.len = strlen(func_g.search);
3855 func_g.search = glob;
3856
3857 /* we do not support '!' for function probes */
3858 if (WARN_ON(not))
3859 return;
3860 }
3861
3862 mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3863
3864 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3865 if (!hash)
3866 /* Hmm, should report this somehow */
3867 goto out_unlock;
3868
3869 INIT_LIST_HEAD(&free_list);
3870
3871 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3872 struct hlist_head *hhd = &ftrace_func_hash[i];
3873
3874 hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3875
3876 /* break up if statements for readability */
3877 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3878 continue;
3879
3880 if ((flags & PROBE_TEST_DATA) && entry->data != data)
3881 continue;
3882
3883 /* do this last, since it is the most expensive */
3884 if (func_g.search) {
3885 kallsyms_lookup(entry->ip, NULL, NULL,
3886 NULL, str);
3887 if (!ftrace_match(str, &func_g))
3888 continue;
3889 }
3890
3891 rec_entry = ftrace_lookup_ip(hash, entry->ip);
3892 /* It is possible more than one entry had this ip */
3893 if (rec_entry)
3894 free_hash_entry(hash, rec_entry);
3895
3896 hlist_del_rcu(&entry->node);
3897 list_add(&entry->free_list, &free_list);
3898 }
3899 }
3900 mutex_lock(&ftrace_lock);
3901 __disable_ftrace_function_probe();
3902 /*
3903 * Remove after the disable is called. Otherwise, if the last
3904 * probe is removed, a null hash means *all enabled*.
3905 */
3906 ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3907 synchronize_sched();
3908 if (!ret)
3909 free_ftrace_hash_rcu(old_hash);
3910
3911 list_for_each_entry_safe(entry, p, &free_list, free_list) {
3912 list_del(&entry->free_list);
3913 ftrace_free_entry(entry);
3914 }
3915 mutex_unlock(&ftrace_lock);
3916
3917 out_unlock:
3918 mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3919 free_ftrace_hash(hash);
3920}
3921
3922void
3923unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3924 void *data)
3925{
3926 __unregister_ftrace_function_probe(glob, ops, data,
3927 PROBE_TEST_FUNC | PROBE_TEST_DATA);
3928}
3929
3930void
3931unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3932{
3933 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3934}
3935
3936void unregister_ftrace_function_probe_all(char *glob)
3937{
3938 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3939}
3940
3941static LIST_HEAD(ftrace_commands);
3942static DEFINE_MUTEX(ftrace_cmd_mutex);
3943
3944/*
3945 * Currently we only register ftrace commands from __init, so mark this
3946 * __init too.
3947 */
3948__init int register_ftrace_command(struct ftrace_func_command *cmd)
3949{
3950 struct ftrace_func_command *p;
3951 int ret = 0;
3952
3953 mutex_lock(&ftrace_cmd_mutex);
3954 list_for_each_entry(p, &ftrace_commands, list) {
3955 if (strcmp(cmd->name, p->name) == 0) {
3956 ret = -EBUSY;
3957 goto out_unlock;
3958 }
3959 }
3960 list_add(&cmd->list, &ftrace_commands);
3961 out_unlock:
3962 mutex_unlock(&ftrace_cmd_mutex);
3963
3964 return ret;
3965}
3966
3967/*
3968 * Currently we only unregister ftrace commands from __init, so mark
3969 * this __init too.
3970 */
3971__init int unregister_ftrace_command(struct ftrace_func_command *cmd)
3972{
3973 struct ftrace_func_command *p, *n;
3974 int ret = -ENODEV;
3975
3976 mutex_lock(&ftrace_cmd_mutex);
3977 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3978 if (strcmp(cmd->name, p->name) == 0) {
3979 ret = 0;
3980 list_del_init(&p->list);
3981 goto out_unlock;
3982 }
3983 }
3984 out_unlock:
3985 mutex_unlock(&ftrace_cmd_mutex);
3986
3987 return ret;
3988}
3989
3990static int ftrace_process_regex(struct ftrace_hash *hash,
3991 char *buff, int len, int enable)
3992{
3993 char *func, *command, *next = buff;
3994 struct ftrace_func_command *p;
3995 int ret = -EINVAL;
3996
3997 func = strsep(&next, ":");
3998
3999 if (!next) {
4000 ret = ftrace_match_records(hash, func, len);
4001 if (!ret)
4002 ret = -EINVAL;
4003 if (ret < 0)
4004 return ret;
4005 return 0;
4006 }
4007
4008 /* command found */
4009
4010 command = strsep(&next, ":");
4011
4012 mutex_lock(&ftrace_cmd_mutex);
4013 list_for_each_entry(p, &ftrace_commands, list) {
4014 if (strcmp(p->name, command) == 0) {
4015 ret = p->func(hash, func, command, next, enable);
4016 goto out_unlock;
4017 }
4018 }
4019 out_unlock:
4020 mutex_unlock(&ftrace_cmd_mutex);
4021
4022 return ret;
4023}
4024
4025static ssize_t
4026ftrace_regex_write(struct file *file, const char __user *ubuf,
4027 size_t cnt, loff_t *ppos, int enable)
4028{
4029 struct ftrace_iterator *iter;
4030 struct trace_parser *parser;
4031 ssize_t ret, read;
4032
4033 if (!cnt)
4034 return 0;
4035
4036 if (file->f_mode & FMODE_READ) {
4037 struct seq_file *m = file->private_data;
4038 iter = m->private;
4039 } else
4040 iter = file->private_data;
4041
4042 if (unlikely(ftrace_disabled))
4043 return -ENODEV;
4044
4045 /* iter->hash is a local copy, so we don't need regex_lock */
4046
4047 parser = &iter->parser;
4048 read = trace_get_user(parser, ubuf, cnt, ppos);
4049
4050 if (read >= 0 && trace_parser_loaded(parser) &&
4051 !trace_parser_cont(parser)) {
4052 ret = ftrace_process_regex(iter->hash, parser->buffer,
4053 parser->idx, enable);
4054 trace_parser_clear(parser);
4055 if (ret < 0)
4056 goto out;
4057 }
4058
4059 ret = read;
4060 out:
4061 return ret;
4062}
4063
4064ssize_t
4065ftrace_filter_write(struct file *file, const char __user *ubuf,
4066 size_t cnt, loff_t *ppos)
4067{
4068 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4069}
4070
4071ssize_t
4072ftrace_notrace_write(struct file *file, const char __user *ubuf,
4073 size_t cnt, loff_t *ppos)
4074{
4075 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4076}
4077
4078static int
4079ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4080{
4081 struct ftrace_func_entry *entry;
4082
4083 if (!ftrace_location(ip))
4084 return -EINVAL;
4085
4086 if (remove) {
4087 entry = ftrace_lookup_ip(hash, ip);
4088 if (!entry)
4089 return -ENOENT;
4090 free_hash_entry(hash, entry);
4091 return 0;
4092 }
4093
4094 return add_hash_entry(hash, ip);
4095}
4096
4097static void ftrace_ops_update_code(struct ftrace_ops *ops,
4098 struct ftrace_ops_hash *old_hash)
4099{
4100 struct ftrace_ops *op;
4101
4102 if (!ftrace_enabled)
4103 return;
4104
4105 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4106 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4107 return;
4108 }
4109
4110 /*
4111 * If this is the shared global_ops filter, then we need to
4112 * check if there is another ops that shares it, is enabled.
4113 * If so, we still need to run the modify code.
4114 */
4115 if (ops->func_hash != &global_ops.local_hash)
4116 return;
4117
4118 do_for_each_ftrace_op(op, ftrace_ops_list) {
4119 if (op->func_hash == &global_ops.local_hash &&
4120 op->flags & FTRACE_OPS_FL_ENABLED) {
4121 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4122 /* Only need to do this once */
4123 return;
4124 }
4125 } while_for_each_ftrace_op(op);
4126}
4127
4128static int
4129ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4130 unsigned long ip, int remove, int reset, int enable)
4131{
4132 struct ftrace_hash **orig_hash;
4133 struct ftrace_ops_hash old_hash_ops;
4134 struct ftrace_hash *old_hash;
4135 struct ftrace_hash *hash;
4136 int ret;
4137
4138 if (unlikely(ftrace_disabled))
4139 return -ENODEV;
4140
4141 mutex_lock(&ops->func_hash->regex_lock);
4142
4143 if (enable)
4144 orig_hash = &ops->func_hash->filter_hash;
4145 else
4146 orig_hash = &ops->func_hash->notrace_hash;
4147
4148 if (reset)
4149 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4150 else
4151 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4152
4153 if (!hash) {
4154 ret = -ENOMEM;
4155 goto out_regex_unlock;
4156 }
4157
4158 if (buf && !ftrace_match_records(hash, buf, len)) {
4159 ret = -EINVAL;
4160 goto out_regex_unlock;
4161 }
4162 if (ip) {
4163 ret = ftrace_match_addr(hash, ip, remove);
4164 if (ret < 0)
4165 goto out_regex_unlock;
4166 }
4167
4168 mutex_lock(&ftrace_lock);
4169 old_hash = *orig_hash;
4170 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
4171 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
4172 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
4173 if (!ret) {
4174 ftrace_ops_update_code(ops, &old_hash_ops);
4175 free_ftrace_hash_rcu(old_hash);
4176 }
4177 mutex_unlock(&ftrace_lock);
4178
4179 out_regex_unlock:
4180 mutex_unlock(&ops->func_hash->regex_lock);
4181
4182 free_ftrace_hash(hash);
4183 return ret;
4184}
4185
4186static int
4187ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4188 int reset, int enable)
4189{
4190 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4191}
4192
4193/**
4194 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4195 * @ops - the ops to set the filter with
4196 * @ip - the address to add to or remove from the filter.
4197 * @remove - non zero to remove the ip from the filter
4198 * @reset - non zero to reset all filters before applying this filter.
4199 *
4200 * Filters denote which functions should be enabled when tracing is enabled
4201 * If @ip is NULL, it failes to update filter.
4202 */
4203int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4204 int remove, int reset)
4205{
4206 ftrace_ops_init(ops);
4207 return ftrace_set_addr(ops, ip, remove, reset, 1);
4208}
4209EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4210
4211static int
4212ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4213 int reset, int enable)
4214{
4215 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4216}
4217
4218/**
4219 * ftrace_set_filter - set a function to filter on in ftrace
4220 * @ops - the ops to set the filter with
4221 * @buf - the string that holds the function filter text.
4222 * @len - the length of the string.
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 @buf is NULL and reset is set, all functions will be enabled for tracing.
4227 */
4228int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4229 int len, int reset)
4230{
4231 ftrace_ops_init(ops);
4232 return ftrace_set_regex(ops, buf, len, reset, 1);
4233}
4234EXPORT_SYMBOL_GPL(ftrace_set_filter);
4235
4236/**
4237 * ftrace_set_notrace - set a function to not trace in ftrace
4238 * @ops - the ops to set the notrace filter with
4239 * @buf - the string that holds the function notrace text.
4240 * @len - the length of the string.
4241 * @reset - non zero to reset all filters before applying this filter.
4242 *
4243 * Notrace Filters denote which functions should not be enabled when tracing
4244 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4245 * for tracing.
4246 */
4247int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4248 int len, int reset)
4249{
4250 ftrace_ops_init(ops);
4251 return ftrace_set_regex(ops, buf, len, reset, 0);
4252}
4253EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4254/**
4255 * ftrace_set_global_filter - set a function to filter on with global tracers
4256 * @buf - the string that holds the function filter text.
4257 * @len - the length of the string.
4258 * @reset - non zero to reset all filters before applying this filter.
4259 *
4260 * Filters denote which functions should be enabled when tracing is enabled.
4261 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4262 */
4263void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4264{
4265 ftrace_set_regex(&global_ops, buf, len, reset, 1);
4266}
4267EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4268
4269/**
4270 * ftrace_set_global_notrace - set a function to not trace with global tracers
4271 * @buf - the string that holds the function notrace text.
4272 * @len - the length of the string.
4273 * @reset - non zero to reset all filters before applying this filter.
4274 *
4275 * Notrace Filters denote which functions should not be enabled when tracing
4276 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4277 * for tracing.
4278 */
4279void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4280{
4281 ftrace_set_regex(&global_ops, buf, len, reset, 0);
4282}
4283EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4284
4285/*
4286 * command line interface to allow users to set filters on boot up.
4287 */
4288#define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
4289static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4290static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4291
4292/* Used by function selftest to not test if filter is set */
4293bool ftrace_filter_param __initdata;
4294
4295static int __init set_ftrace_notrace(char *str)
4296{
4297 ftrace_filter_param = true;
4298 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4299 return 1;
4300}
4301__setup("ftrace_notrace=", set_ftrace_notrace);
4302
4303static int __init set_ftrace_filter(char *str)
4304{
4305 ftrace_filter_param = true;
4306 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4307 return 1;
4308}
4309__setup("ftrace_filter=", set_ftrace_filter);
4310
4311#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4312static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4313static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4314static int ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer);
4315
4316static unsigned long save_global_trampoline;
4317static unsigned long save_global_flags;
4318
4319static int __init set_graph_function(char *str)
4320{
4321 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4322 return 1;
4323}
4324__setup("ftrace_graph_filter=", set_graph_function);
4325
4326static int __init set_graph_notrace_function(char *str)
4327{
4328 strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4329 return 1;
4330}
4331__setup("ftrace_graph_notrace=", set_graph_notrace_function);
4332
4333static void __init set_ftrace_early_graph(char *buf, int enable)
4334{
4335 int ret;
4336 char *func;
4337 unsigned long *table = ftrace_graph_funcs;
4338 int *count = &ftrace_graph_count;
4339
4340 if (!enable) {
4341 table = ftrace_graph_notrace_funcs;
4342 count = &ftrace_graph_notrace_count;
4343 }
4344
4345 while (buf) {
4346 func = strsep(&buf, ",");
4347 /* we allow only one expression at a time */
4348 ret = ftrace_set_func(table, count, FTRACE_GRAPH_MAX_FUNCS, func);
4349 if (ret)
4350 printk(KERN_DEBUG "ftrace: function %s not "
4351 "traceable\n", func);
4352 }
4353}
4354#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4355
4356void __init
4357ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
4358{
4359 char *func;
4360
4361 ftrace_ops_init(ops);
4362
4363 while (buf) {
4364 func = strsep(&buf, ",");
4365 ftrace_set_regex(ops, func, strlen(func), 0, enable);
4366 }
4367}
4368
4369static void __init set_ftrace_early_filters(void)
4370{
4371 if (ftrace_filter_buf[0])
4372 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
4373 if (ftrace_notrace_buf[0])
4374 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
4375#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4376 if (ftrace_graph_buf[0])
4377 set_ftrace_early_graph(ftrace_graph_buf, 1);
4378 if (ftrace_graph_notrace_buf[0])
4379 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
4380#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4381}
4382
4383int ftrace_regex_release(struct inode *inode, struct file *file)
4384{
4385 struct seq_file *m = (struct seq_file *)file->private_data;
4386 struct ftrace_ops_hash old_hash_ops;
4387 struct ftrace_iterator *iter;
4388 struct ftrace_hash **orig_hash;
4389 struct ftrace_hash *old_hash;
4390 struct trace_parser *parser;
4391 int filter_hash;
4392 int ret;
4393
4394 if (file->f_mode & FMODE_READ) {
4395 iter = m->private;
4396 seq_release(inode, file);
4397 } else
4398 iter = file->private_data;
4399
4400 parser = &iter->parser;
4401 if (trace_parser_loaded(parser)) {
4402 parser->buffer[parser->idx] = 0;
4403 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
4404 }
4405
4406 trace_parser_put(parser);
4407
4408 mutex_lock(&iter->ops->func_hash->regex_lock);
4409
4410 if (file->f_mode & FMODE_WRITE) {
4411 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
4412
4413 if (filter_hash)
4414 orig_hash = &iter->ops->func_hash->filter_hash;
4415 else
4416 orig_hash = &iter->ops->func_hash->notrace_hash;
4417
4418 mutex_lock(&ftrace_lock);
4419 old_hash = *orig_hash;
4420 old_hash_ops.filter_hash = iter->ops->func_hash->filter_hash;
4421 old_hash_ops.notrace_hash = iter->ops->func_hash->notrace_hash;
4422 ret = ftrace_hash_move(iter->ops, filter_hash,
4423 orig_hash, iter->hash);
4424 if (!ret) {
4425 ftrace_ops_update_code(iter->ops, &old_hash_ops);
4426 free_ftrace_hash_rcu(old_hash);
4427 }
4428 mutex_unlock(&ftrace_lock);
4429 }
4430
4431 mutex_unlock(&iter->ops->func_hash->regex_lock);
4432 free_ftrace_hash(iter->hash);
4433 kfree(iter);
4434
4435 return 0;
4436}
4437
4438static const struct file_operations ftrace_avail_fops = {
4439 .open = ftrace_avail_open,
4440 .read = seq_read,
4441 .llseek = seq_lseek,
4442 .release = seq_release_private,
4443};
4444
4445static const struct file_operations ftrace_enabled_fops = {
4446 .open = ftrace_enabled_open,
4447 .read = seq_read,
4448 .llseek = seq_lseek,
4449 .release = seq_release_private,
4450};
4451
4452static const struct file_operations ftrace_filter_fops = {
4453 .open = ftrace_filter_open,
4454 .read = seq_read,
4455 .write = ftrace_filter_write,
4456 .llseek = tracing_lseek,
4457 .release = ftrace_regex_release,
4458};
4459
4460static const struct file_operations ftrace_notrace_fops = {
4461 .open = ftrace_notrace_open,
4462 .read = seq_read,
4463 .write = ftrace_notrace_write,
4464 .llseek = tracing_lseek,
4465 .release = ftrace_regex_release,
4466};
4467
4468#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4469
4470static DEFINE_MUTEX(graph_lock);
4471
4472int ftrace_graph_count;
4473int ftrace_graph_notrace_count;
4474unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4475unsigned long ftrace_graph_notrace_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4476
4477struct ftrace_graph_data {
4478 unsigned long *table;
4479 size_t size;
4480 int *count;
4481 const struct seq_operations *seq_ops;
4482};
4483
4484static void *
4485__g_next(struct seq_file *m, loff_t *pos)
4486{
4487 struct ftrace_graph_data *fgd = m->private;
4488
4489 if (*pos >= *fgd->count)
4490 return NULL;
4491 return &fgd->table[*pos];
4492}
4493
4494static void *
4495g_next(struct seq_file *m, void *v, loff_t *pos)
4496{
4497 (*pos)++;
4498 return __g_next(m, pos);
4499}
4500
4501static void *g_start(struct seq_file *m, loff_t *pos)
4502{
4503 struct ftrace_graph_data *fgd = m->private;
4504
4505 mutex_lock(&graph_lock);
4506
4507 /* Nothing, tell g_show to print all functions are enabled */
4508 if (!*fgd->count && !*pos)
4509 return (void *)1;
4510
4511 return __g_next(m, pos);
4512}
4513
4514static void g_stop(struct seq_file *m, void *p)
4515{
4516 mutex_unlock(&graph_lock);
4517}
4518
4519static int g_show(struct seq_file *m, void *v)
4520{
4521 unsigned long *ptr = v;
4522
4523 if (!ptr)
4524 return 0;
4525
4526 if (ptr == (unsigned long *)1) {
4527 struct ftrace_graph_data *fgd = m->private;
4528
4529 if (fgd->table == ftrace_graph_funcs)
4530 seq_puts(m, "#### all functions enabled ####\n");
4531 else
4532 seq_puts(m, "#### no functions disabled ####\n");
4533 return 0;
4534 }
4535
4536 seq_printf(m, "%ps\n", (void *)*ptr);
4537
4538 return 0;
4539}
4540
4541static const struct seq_operations ftrace_graph_seq_ops = {
4542 .start = g_start,
4543 .next = g_next,
4544 .stop = g_stop,
4545 .show = g_show,
4546};
4547
4548static int
4549__ftrace_graph_open(struct inode *inode, struct file *file,
4550 struct ftrace_graph_data *fgd)
4551{
4552 int ret = 0;
4553
4554 mutex_lock(&graph_lock);
4555 if ((file->f_mode & FMODE_WRITE) &&
4556 (file->f_flags & O_TRUNC)) {
4557 *fgd->count = 0;
4558 memset(fgd->table, 0, fgd->size * sizeof(*fgd->table));
4559 }
4560 mutex_unlock(&graph_lock);
4561
4562 if (file->f_mode & FMODE_READ) {
4563 ret = seq_open(file, fgd->seq_ops);
4564 if (!ret) {
4565 struct seq_file *m = file->private_data;
4566 m->private = fgd;
4567 }
4568 } else
4569 file->private_data = fgd;
4570
4571 return ret;
4572}
4573
4574static int
4575ftrace_graph_open(struct inode *inode, struct file *file)
4576{
4577 struct ftrace_graph_data *fgd;
4578
4579 if (unlikely(ftrace_disabled))
4580 return -ENODEV;
4581
4582 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4583 if (fgd == NULL)
4584 return -ENOMEM;
4585
4586 fgd->table = ftrace_graph_funcs;
4587 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4588 fgd->count = &ftrace_graph_count;
4589 fgd->seq_ops = &ftrace_graph_seq_ops;
4590
4591 return __ftrace_graph_open(inode, file, fgd);
4592}
4593
4594static int
4595ftrace_graph_notrace_open(struct inode *inode, struct file *file)
4596{
4597 struct ftrace_graph_data *fgd;
4598
4599 if (unlikely(ftrace_disabled))
4600 return -ENODEV;
4601
4602 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4603 if (fgd == NULL)
4604 return -ENOMEM;
4605
4606 fgd->table = ftrace_graph_notrace_funcs;
4607 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4608 fgd->count = &ftrace_graph_notrace_count;
4609 fgd->seq_ops = &ftrace_graph_seq_ops;
4610
4611 return __ftrace_graph_open(inode, file, fgd);
4612}
4613
4614static int
4615ftrace_graph_release(struct inode *inode, struct file *file)
4616{
4617 if (file->f_mode & FMODE_READ) {
4618 struct seq_file *m = file->private_data;
4619
4620 kfree(m->private);
4621 seq_release(inode, file);
4622 } else {
4623 kfree(file->private_data);
4624 }
4625
4626 return 0;
4627}
4628
4629static int
4630ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer)
4631{
4632 struct ftrace_glob func_g;
4633 struct dyn_ftrace *rec;
4634 struct ftrace_page *pg;
4635 int fail = 1;
4636 int not;
4637 bool exists;
4638 int i;
4639
4640 /* decode regex */
4641 func_g.type = filter_parse_regex(buffer, strlen(buffer),
4642 &func_g.search, ¬);
4643 if (!not && *idx >= size)
4644 return -EBUSY;
4645
4646 func_g.len = strlen(func_g.search);
4647
4648 mutex_lock(&ftrace_lock);
4649
4650 if (unlikely(ftrace_disabled)) {
4651 mutex_unlock(&ftrace_lock);
4652 return -ENODEV;
4653 }
4654
4655 do_for_each_ftrace_rec(pg, rec) {
4656
4657 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
4658 /* if it is in the array */
4659 exists = false;
4660 for (i = 0; i < *idx; i++) {
4661 if (array[i] == rec->ip) {
4662 exists = true;
4663 break;
4664 }
4665 }
4666
4667 if (!not) {
4668 fail = 0;
4669 if (!exists) {
4670 array[(*idx)++] = rec->ip;
4671 if (*idx >= size)
4672 goto out;
4673 }
4674 } else {
4675 if (exists) {
4676 array[i] = array[--(*idx)];
4677 array[*idx] = 0;
4678 fail = 0;
4679 }
4680 }
4681 }
4682 } while_for_each_ftrace_rec();
4683out:
4684 mutex_unlock(&ftrace_lock);
4685
4686 if (fail)
4687 return -EINVAL;
4688
4689 return 0;
4690}
4691
4692static ssize_t
4693ftrace_graph_write(struct file *file, const char __user *ubuf,
4694 size_t cnt, loff_t *ppos)
4695{
4696 struct trace_parser parser;
4697 ssize_t read, ret = 0;
4698 struct ftrace_graph_data *fgd = file->private_data;
4699
4700 if (!cnt)
4701 return 0;
4702
4703 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX))
4704 return -ENOMEM;
4705
4706 read = trace_get_user(&parser, ubuf, cnt, ppos);
4707
4708 if (read >= 0 && trace_parser_loaded((&parser))) {
4709 parser.buffer[parser.idx] = 0;
4710
4711 mutex_lock(&graph_lock);
4712
4713 /* we allow only one expression at a time */
4714 ret = ftrace_set_func(fgd->table, fgd->count, fgd->size,
4715 parser.buffer);
4716
4717 mutex_unlock(&graph_lock);
4718 }
4719
4720 if (!ret)
4721 ret = read;
4722
4723 trace_parser_put(&parser);
4724
4725 return ret;
4726}
4727
4728static const struct file_operations ftrace_graph_fops = {
4729 .open = ftrace_graph_open,
4730 .read = seq_read,
4731 .write = ftrace_graph_write,
4732 .llseek = tracing_lseek,
4733 .release = ftrace_graph_release,
4734};
4735
4736static const struct file_operations ftrace_graph_notrace_fops = {
4737 .open = ftrace_graph_notrace_open,
4738 .read = seq_read,
4739 .write = ftrace_graph_write,
4740 .llseek = tracing_lseek,
4741 .release = ftrace_graph_release,
4742};
4743#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4744
4745void ftrace_create_filter_files(struct ftrace_ops *ops,
4746 struct dentry *parent)
4747{
4748
4749 trace_create_file("set_ftrace_filter", 0644, parent,
4750 ops, &ftrace_filter_fops);
4751
4752 trace_create_file("set_ftrace_notrace", 0644, parent,
4753 ops, &ftrace_notrace_fops);
4754}
4755
4756/*
4757 * The name "destroy_filter_files" is really a misnomer. Although
4758 * in the future, it may actualy delete the files, but this is
4759 * really intended to make sure the ops passed in are disabled
4760 * and that when this function returns, the caller is free to
4761 * free the ops.
4762 *
4763 * The "destroy" name is only to match the "create" name that this
4764 * should be paired with.
4765 */
4766void ftrace_destroy_filter_files(struct ftrace_ops *ops)
4767{
4768 mutex_lock(&ftrace_lock);
4769 if (ops->flags & FTRACE_OPS_FL_ENABLED)
4770 ftrace_shutdown(ops, 0);
4771 ops->flags |= FTRACE_OPS_FL_DELETED;
4772 mutex_unlock(&ftrace_lock);
4773}
4774
4775static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
4776{
4777
4778 trace_create_file("available_filter_functions", 0444,
4779 d_tracer, NULL, &ftrace_avail_fops);
4780
4781 trace_create_file("enabled_functions", 0444,
4782 d_tracer, NULL, &ftrace_enabled_fops);
4783
4784 ftrace_create_filter_files(&global_ops, d_tracer);
4785
4786#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4787 trace_create_file("set_graph_function", 0444, d_tracer,
4788 NULL,
4789 &ftrace_graph_fops);
4790 trace_create_file("set_graph_notrace", 0444, d_tracer,
4791 NULL,
4792 &ftrace_graph_notrace_fops);
4793#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4794
4795 return 0;
4796}
4797
4798static int ftrace_cmp_ips(const void *a, const void *b)
4799{
4800 const unsigned long *ipa = a;
4801 const unsigned long *ipb = b;
4802
4803 if (*ipa > *ipb)
4804 return 1;
4805 if (*ipa < *ipb)
4806 return -1;
4807 return 0;
4808}
4809
4810static int ftrace_process_locs(struct module *mod,
4811 unsigned long *start,
4812 unsigned long *end)
4813{
4814 struct ftrace_page *start_pg;
4815 struct ftrace_page *pg;
4816 struct dyn_ftrace *rec;
4817 unsigned long count;
4818 unsigned long *p;
4819 unsigned long addr;
4820 unsigned long flags = 0; /* Shut up gcc */
4821 int ret = -ENOMEM;
4822
4823 count = end - start;
4824
4825 if (!count)
4826 return 0;
4827
4828 sort(start, count, sizeof(*start),
4829 ftrace_cmp_ips, NULL);
4830
4831 start_pg = ftrace_allocate_pages(count);
4832 if (!start_pg)
4833 return -ENOMEM;
4834
4835 mutex_lock(&ftrace_lock);
4836
4837 /*
4838 * Core and each module needs their own pages, as
4839 * modules will free them when they are removed.
4840 * Force a new page to be allocated for modules.
4841 */
4842 if (!mod) {
4843 WARN_ON(ftrace_pages || ftrace_pages_start);
4844 /* First initialization */
4845 ftrace_pages = ftrace_pages_start = start_pg;
4846 } else {
4847 if (!ftrace_pages)
4848 goto out;
4849
4850 if (WARN_ON(ftrace_pages->next)) {
4851 /* Hmm, we have free pages? */
4852 while (ftrace_pages->next)
4853 ftrace_pages = ftrace_pages->next;
4854 }
4855
4856 ftrace_pages->next = start_pg;
4857 }
4858
4859 p = start;
4860 pg = start_pg;
4861 while (p < end) {
4862 addr = ftrace_call_adjust(*p++);
4863 /*
4864 * Some architecture linkers will pad between
4865 * the different mcount_loc sections of different
4866 * object files to satisfy alignments.
4867 * Skip any NULL pointers.
4868 */
4869 if (!addr)
4870 continue;
4871
4872 if (pg->index == pg->size) {
4873 /* We should have allocated enough */
4874 if (WARN_ON(!pg->next))
4875 break;
4876 pg = pg->next;
4877 }
4878
4879 rec = &pg->records[pg->index++];
4880 rec->ip = addr;
4881 }
4882
4883 /* We should have used all pages */
4884 WARN_ON(pg->next);
4885
4886 /* Assign the last page to ftrace_pages */
4887 ftrace_pages = pg;
4888
4889 /*
4890 * We only need to disable interrupts on start up
4891 * because we are modifying code that an interrupt
4892 * may execute, and the modification is not atomic.
4893 * But for modules, nothing runs the code we modify
4894 * until we are finished with it, and there's no
4895 * reason to cause large interrupt latencies while we do it.
4896 */
4897 if (!mod)
4898 local_irq_save(flags);
4899 ftrace_update_code(mod, start_pg);
4900 if (!mod)
4901 local_irq_restore(flags);
4902 ret = 0;
4903 out:
4904 mutex_unlock(&ftrace_lock);
4905
4906 return ret;
4907}
4908
4909#ifdef CONFIG_MODULES
4910
4911#define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4912
4913static int referenced_filters(struct dyn_ftrace *rec)
4914{
4915 struct ftrace_ops *ops;
4916 int cnt = 0;
4917
4918 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
4919 if (ops_references_rec(ops, rec))
4920 cnt++;
4921 }
4922
4923 return cnt;
4924}
4925
4926void ftrace_release_mod(struct module *mod)
4927{
4928 struct dyn_ftrace *rec;
4929 struct ftrace_page **last_pg;
4930 struct ftrace_page *pg;
4931 int order;
4932
4933 mutex_lock(&ftrace_lock);
4934
4935 if (ftrace_disabled)
4936 goto out_unlock;
4937
4938 /*
4939 * Each module has its own ftrace_pages, remove
4940 * them from the list.
4941 */
4942 last_pg = &ftrace_pages_start;
4943 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4944 rec = &pg->records[0];
4945 if (within_module_core(rec->ip, mod)) {
4946 /*
4947 * As core pages are first, the first
4948 * page should never be a module page.
4949 */
4950 if (WARN_ON(pg == ftrace_pages_start))
4951 goto out_unlock;
4952
4953 /* Check if we are deleting the last page */
4954 if (pg == ftrace_pages)
4955 ftrace_pages = next_to_ftrace_page(last_pg);
4956
4957 *last_pg = pg->next;
4958 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4959 free_pages((unsigned long)pg->records, order);
4960 kfree(pg);
4961 } else
4962 last_pg = &pg->next;
4963 }
4964 out_unlock:
4965 mutex_unlock(&ftrace_lock);
4966}
4967
4968void ftrace_module_enable(struct module *mod)
4969{
4970 struct dyn_ftrace *rec;
4971 struct ftrace_page *pg;
4972
4973 mutex_lock(&ftrace_lock);
4974
4975 if (ftrace_disabled)
4976 goto out_unlock;
4977
4978 /*
4979 * If the tracing is enabled, go ahead and enable the record.
4980 *
4981 * The reason not to enable the record immediatelly is the
4982 * inherent check of ftrace_make_nop/ftrace_make_call for
4983 * correct previous instructions. Making first the NOP
4984 * conversion puts the module to the correct state, thus
4985 * passing the ftrace_make_call check.
4986 *
4987 * We also delay this to after the module code already set the
4988 * text to read-only, as we now need to set it back to read-write
4989 * so that we can modify the text.
4990 */
4991 if (ftrace_start_up)
4992 ftrace_arch_code_modify_prepare();
4993
4994 do_for_each_ftrace_rec(pg, rec) {
4995 int cnt;
4996 /*
4997 * do_for_each_ftrace_rec() is a double loop.
4998 * module text shares the pg. If a record is
4999 * not part of this module, then skip this pg,
5000 * which the "break" will do.
5001 */
5002 if (!within_module_core(rec->ip, mod))
5003 break;
5004
5005 cnt = 0;
5006
5007 /*
5008 * When adding a module, we need to check if tracers are
5009 * currently enabled and if they are, and can trace this record,
5010 * we need to enable the module functions as well as update the
5011 * reference counts for those function records.
5012 */
5013 if (ftrace_start_up)
5014 cnt += referenced_filters(rec);
5015
5016 /* This clears FTRACE_FL_DISABLED */
5017 rec->flags = cnt;
5018
5019 if (ftrace_start_up && cnt) {
5020 int failed = __ftrace_replace_code(rec, 1);
5021 if (failed) {
5022 ftrace_bug(failed, rec);
5023 goto out_loop;
5024 }
5025 }
5026
5027 } while_for_each_ftrace_rec();
5028
5029 out_loop:
5030 if (ftrace_start_up)
5031 ftrace_arch_code_modify_post_process();
5032
5033 out_unlock:
5034 mutex_unlock(&ftrace_lock);
5035}
5036
5037void ftrace_module_init(struct module *mod)
5038{
5039 if (ftrace_disabled || !mod->num_ftrace_callsites)
5040 return;
5041
5042 ftrace_process_locs(mod, mod->ftrace_callsites,
5043 mod->ftrace_callsites + mod->num_ftrace_callsites);
5044}
5045#endif /* CONFIG_MODULES */
5046
5047void __init ftrace_init(void)
5048{
5049 extern unsigned long __start_mcount_loc[];
5050 extern unsigned long __stop_mcount_loc[];
5051 unsigned long count, flags;
5052 int ret;
5053
5054 local_irq_save(flags);
5055 ret = ftrace_dyn_arch_init();
5056 local_irq_restore(flags);
5057 if (ret)
5058 goto failed;
5059
5060 count = __stop_mcount_loc - __start_mcount_loc;
5061 if (!count) {
5062 pr_info("ftrace: No functions to be traced?\n");
5063 goto failed;
5064 }
5065
5066 pr_info("ftrace: allocating %ld entries in %ld pages\n",
5067 count, count / ENTRIES_PER_PAGE + 1);
5068
5069 last_ftrace_enabled = ftrace_enabled = 1;
5070
5071 ret = ftrace_process_locs(NULL,
5072 __start_mcount_loc,
5073 __stop_mcount_loc);
5074
5075 set_ftrace_early_filters();
5076
5077 return;
5078 failed:
5079 ftrace_disabled = 1;
5080}
5081
5082/* Do nothing if arch does not support this */
5083void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
5084{
5085}
5086
5087static void ftrace_update_trampoline(struct ftrace_ops *ops)
5088{
5089
5090/*
5091 * Currently there's no safe way to free a trampoline when the kernel
5092 * is configured with PREEMPT. That is because a task could be preempted
5093 * when it jumped to the trampoline, it may be preempted for a long time
5094 * depending on the system load, and currently there's no way to know
5095 * when it will be off the trampoline. If the trampoline is freed
5096 * too early, when the task runs again, it will be executing on freed
5097 * memory and crash.
5098 */
5099#ifdef CONFIG_PREEMPT
5100 /* Currently, only non dynamic ops can have a trampoline */
5101 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
5102 return;
5103#endif
5104
5105 arch_ftrace_update_trampoline(ops);
5106}
5107
5108#else
5109
5110static struct ftrace_ops global_ops = {
5111 .func = ftrace_stub,
5112 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
5113 FTRACE_OPS_FL_INITIALIZED |
5114 FTRACE_OPS_FL_PID,
5115};
5116
5117static int __init ftrace_nodyn_init(void)
5118{
5119 ftrace_enabled = 1;
5120 return 0;
5121}
5122core_initcall(ftrace_nodyn_init);
5123
5124static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
5125static inline void ftrace_startup_enable(int command) { }
5126static inline void ftrace_startup_all(int command) { }
5127/* Keep as macros so we do not need to define the commands */
5128# define ftrace_startup(ops, command) \
5129 ({ \
5130 int ___ret = __register_ftrace_function(ops); \
5131 if (!___ret) \
5132 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
5133 ___ret; \
5134 })
5135# define ftrace_shutdown(ops, command) \
5136 ({ \
5137 int ___ret = __unregister_ftrace_function(ops); \
5138 if (!___ret) \
5139 (ops)->flags &= ~FTRACE_OPS_FL_ENABLED; \
5140 ___ret; \
5141 })
5142
5143# define ftrace_startup_sysctl() do { } while (0)
5144# define ftrace_shutdown_sysctl() do { } while (0)
5145
5146static inline int
5147ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
5148{
5149 return 1;
5150}
5151
5152static void ftrace_update_trampoline(struct ftrace_ops *ops)
5153{
5154}
5155
5156#endif /* CONFIG_DYNAMIC_FTRACE */
5157
5158__init void ftrace_init_global_array_ops(struct trace_array *tr)
5159{
5160 tr->ops = &global_ops;
5161 tr->ops->private = tr;
5162}
5163
5164void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
5165{
5166 /* If we filter on pids, update to use the pid function */
5167 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
5168 if (WARN_ON(tr->ops->func != ftrace_stub))
5169 printk("ftrace ops had %pS for function\n",
5170 tr->ops->func);
5171 }
5172 tr->ops->func = func;
5173 tr->ops->private = tr;
5174}
5175
5176void ftrace_reset_array_ops(struct trace_array *tr)
5177{
5178 tr->ops->func = ftrace_stub;
5179}
5180
5181static inline void
5182__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5183 struct ftrace_ops *ignored, struct pt_regs *regs)
5184{
5185 struct ftrace_ops *op;
5186 int bit;
5187
5188 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5189 if (bit < 0)
5190 return;
5191
5192 /*
5193 * Some of the ops may be dynamically allocated,
5194 * they must be freed after a synchronize_sched().
5195 */
5196 preempt_disable_notrace();
5197
5198 do_for_each_ftrace_op(op, ftrace_ops_list) {
5199 /*
5200 * Check the following for each ops before calling their func:
5201 * if RCU flag is set, then rcu_is_watching() must be true
5202 * if PER_CPU is set, then ftrace_function_local_disable()
5203 * must be false
5204 * Otherwise test if the ip matches the ops filter
5205 *
5206 * If any of the above fails then the op->func() is not executed.
5207 */
5208 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
5209 (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5210 !ftrace_function_local_disabled(op)) &&
5211 ftrace_ops_test(op, ip, regs)) {
5212
5213 if (FTRACE_WARN_ON(!op->func)) {
5214 pr_warn("op=%p %pS\n", op, op);
5215 goto out;
5216 }
5217 op->func(ip, parent_ip, op, regs);
5218 }
5219 } while_for_each_ftrace_op(op);
5220out:
5221 preempt_enable_notrace();
5222 trace_clear_recursion(bit);
5223}
5224
5225/*
5226 * Some archs only support passing ip and parent_ip. Even though
5227 * the list function ignores the op parameter, we do not want any
5228 * C side effects, where a function is called without the caller
5229 * sending a third parameter.
5230 * Archs are to support both the regs and ftrace_ops at the same time.
5231 * If they support ftrace_ops, it is assumed they support regs.
5232 * If call backs want to use regs, they must either check for regs
5233 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
5234 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
5235 * An architecture can pass partial regs with ftrace_ops and still
5236 * set the ARCH_SUPPORTS_FTRACE_OPS.
5237 */
5238#if ARCH_SUPPORTS_FTRACE_OPS
5239static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5240 struct ftrace_ops *op, struct pt_regs *regs)
5241{
5242 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
5243}
5244#else
5245static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
5246{
5247 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
5248}
5249#endif
5250
5251/*
5252 * If there's only one function registered but it does not support
5253 * recursion, needs RCU protection and/or requires per cpu handling, then
5254 * this function will be called by the mcount trampoline.
5255 */
5256static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
5257 struct ftrace_ops *op, struct pt_regs *regs)
5258{
5259 int bit;
5260
5261 if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
5262 return;
5263
5264 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5265 if (bit < 0)
5266 return;
5267
5268 preempt_disable_notrace();
5269
5270 if (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5271 !ftrace_function_local_disabled(op)) {
5272 op->func(ip, parent_ip, op, regs);
5273 }
5274
5275 preempt_enable_notrace();
5276 trace_clear_recursion(bit);
5277}
5278
5279/**
5280 * ftrace_ops_get_func - get the function a trampoline should call
5281 * @ops: the ops to get the function for
5282 *
5283 * Normally the mcount trampoline will call the ops->func, but there
5284 * are times that it should not. For example, if the ops does not
5285 * have its own recursion protection, then it should call the
5286 * ftrace_ops_recurs_func() instead.
5287 *
5288 * Returns the function that the trampoline should call for @ops.
5289 */
5290ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
5291{
5292 /*
5293 * If the function does not handle recursion, needs to be RCU safe,
5294 * or does per cpu logic, then we need to call the assist handler.
5295 */
5296 if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
5297 ops->flags & (FTRACE_OPS_FL_RCU | FTRACE_OPS_FL_PER_CPU))
5298 return ftrace_ops_assist_func;
5299
5300 return ops->func;
5301}
5302
5303static void clear_ftrace_swapper(void)
5304{
5305 struct task_struct *p;
5306 int cpu;
5307
5308 get_online_cpus();
5309 for_each_online_cpu(cpu) {
5310 p = idle_task(cpu);
5311 clear_tsk_trace_trace(p);
5312 }
5313 put_online_cpus();
5314}
5315
5316static void set_ftrace_swapper(void)
5317{
5318 struct task_struct *p;
5319 int cpu;
5320
5321 get_online_cpus();
5322 for_each_online_cpu(cpu) {
5323 p = idle_task(cpu);
5324 set_tsk_trace_trace(p);
5325 }
5326 put_online_cpus();
5327}
5328
5329static void clear_ftrace_pid(struct pid *pid)
5330{
5331 struct task_struct *p;
5332
5333 rcu_read_lock();
5334 do_each_pid_task(pid, PIDTYPE_PID, p) {
5335 clear_tsk_trace_trace(p);
5336 } while_each_pid_task(pid, PIDTYPE_PID, p);
5337 rcu_read_unlock();
5338
5339 put_pid(pid);
5340}
5341
5342static void set_ftrace_pid(struct pid *pid)
5343{
5344 struct task_struct *p;
5345
5346 rcu_read_lock();
5347 do_each_pid_task(pid, PIDTYPE_PID, p) {
5348 set_tsk_trace_trace(p);
5349 } while_each_pid_task(pid, PIDTYPE_PID, p);
5350 rcu_read_unlock();
5351}
5352
5353static void clear_ftrace_pid_task(struct pid *pid)
5354{
5355 if (pid == ftrace_swapper_pid)
5356 clear_ftrace_swapper();
5357 else
5358 clear_ftrace_pid(pid);
5359}
5360
5361static void set_ftrace_pid_task(struct pid *pid)
5362{
5363 if (pid == ftrace_swapper_pid)
5364 set_ftrace_swapper();
5365 else
5366 set_ftrace_pid(pid);
5367}
5368
5369static int ftrace_pid_add(int p)
5370{
5371 struct pid *pid;
5372 struct ftrace_pid *fpid;
5373 int ret = -EINVAL;
5374
5375 mutex_lock(&ftrace_lock);
5376
5377 if (!p)
5378 pid = ftrace_swapper_pid;
5379 else
5380 pid = find_get_pid(p);
5381
5382 if (!pid)
5383 goto out;
5384
5385 ret = 0;
5386
5387 list_for_each_entry(fpid, &ftrace_pids, list)
5388 if (fpid->pid == pid)
5389 goto out_put;
5390
5391 ret = -ENOMEM;
5392
5393 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
5394 if (!fpid)
5395 goto out_put;
5396
5397 list_add(&fpid->list, &ftrace_pids);
5398 fpid->pid = pid;
5399
5400 set_ftrace_pid_task(pid);
5401
5402 ftrace_update_pid_func();
5403
5404 ftrace_startup_all(0);
5405
5406 mutex_unlock(&ftrace_lock);
5407 return 0;
5408
5409out_put:
5410 if (pid != ftrace_swapper_pid)
5411 put_pid(pid);
5412
5413out:
5414 mutex_unlock(&ftrace_lock);
5415 return ret;
5416}
5417
5418static void ftrace_pid_reset(void)
5419{
5420 struct ftrace_pid *fpid, *safe;
5421
5422 mutex_lock(&ftrace_lock);
5423 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
5424 struct pid *pid = fpid->pid;
5425
5426 clear_ftrace_pid_task(pid);
5427
5428 list_del(&fpid->list);
5429 kfree(fpid);
5430 }
5431
5432 ftrace_update_pid_func();
5433 ftrace_startup_all(0);
5434
5435 mutex_unlock(&ftrace_lock);
5436}
5437
5438static void *fpid_start(struct seq_file *m, loff_t *pos)
5439{
5440 mutex_lock(&ftrace_lock);
5441
5442 if (!ftrace_pids_enabled() && (!*pos))
5443 return (void *) 1;
5444
5445 return seq_list_start(&ftrace_pids, *pos);
5446}
5447
5448static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
5449{
5450 if (v == (void *)1)
5451 return NULL;
5452
5453 return seq_list_next(v, &ftrace_pids, pos);
5454}
5455
5456static void fpid_stop(struct seq_file *m, void *p)
5457{
5458 mutex_unlock(&ftrace_lock);
5459}
5460
5461static int fpid_show(struct seq_file *m, void *v)
5462{
5463 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
5464
5465 if (v == (void *)1) {
5466 seq_puts(m, "no pid\n");
5467 return 0;
5468 }
5469
5470 if (fpid->pid == ftrace_swapper_pid)
5471 seq_puts(m, "swapper tasks\n");
5472 else
5473 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
5474
5475 return 0;
5476}
5477
5478static const struct seq_operations ftrace_pid_sops = {
5479 .start = fpid_start,
5480 .next = fpid_next,
5481 .stop = fpid_stop,
5482 .show = fpid_show,
5483};
5484
5485static int
5486ftrace_pid_open(struct inode *inode, struct file *file)
5487{
5488 int ret = 0;
5489
5490 if ((file->f_mode & FMODE_WRITE) &&
5491 (file->f_flags & O_TRUNC))
5492 ftrace_pid_reset();
5493
5494 if (file->f_mode & FMODE_READ)
5495 ret = seq_open(file, &ftrace_pid_sops);
5496
5497 return ret;
5498}
5499
5500static ssize_t
5501ftrace_pid_write(struct file *filp, const char __user *ubuf,
5502 size_t cnt, loff_t *ppos)
5503{
5504 char buf[64], *tmp;
5505 long val;
5506 int ret;
5507
5508 if (cnt >= sizeof(buf))
5509 return -EINVAL;
5510
5511 if (copy_from_user(&buf, ubuf, cnt))
5512 return -EFAULT;
5513
5514 buf[cnt] = 0;
5515
5516 /*
5517 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
5518 * to clean the filter quietly.
5519 */
5520 tmp = strstrip(buf);
5521 if (strlen(tmp) == 0)
5522 return 1;
5523
5524 ret = kstrtol(tmp, 10, &val);
5525 if (ret < 0)
5526 return ret;
5527
5528 ret = ftrace_pid_add(val);
5529
5530 return ret ? ret : cnt;
5531}
5532
5533static int
5534ftrace_pid_release(struct inode *inode, struct file *file)
5535{
5536 if (file->f_mode & FMODE_READ)
5537 seq_release(inode, file);
5538
5539 return 0;
5540}
5541
5542static const struct file_operations ftrace_pid_fops = {
5543 .open = ftrace_pid_open,
5544 .write = ftrace_pid_write,
5545 .read = seq_read,
5546 .llseek = tracing_lseek,
5547 .release = ftrace_pid_release,
5548};
5549
5550static __init int ftrace_init_tracefs(void)
5551{
5552 struct dentry *d_tracer;
5553
5554 d_tracer = tracing_init_dentry();
5555 if (IS_ERR(d_tracer))
5556 return 0;
5557
5558 ftrace_init_dyn_tracefs(d_tracer);
5559
5560 trace_create_file("set_ftrace_pid", 0644, d_tracer,
5561 NULL, &ftrace_pid_fops);
5562
5563 ftrace_profile_tracefs(d_tracer);
5564
5565 return 0;
5566}
5567fs_initcall(ftrace_init_tracefs);
5568
5569/**
5570 * ftrace_kill - kill ftrace
5571 *
5572 * This function should be used by panic code. It stops ftrace
5573 * but in a not so nice way. If you need to simply kill ftrace
5574 * from a non-atomic section, use ftrace_kill.
5575 */
5576void ftrace_kill(void)
5577{
5578 ftrace_disabled = 1;
5579 ftrace_enabled = 0;
5580 clear_ftrace_function();
5581}
5582
5583/**
5584 * Test if ftrace is dead or not.
5585 */
5586int ftrace_is_dead(void)
5587{
5588 return ftrace_disabled;
5589}
5590
5591/**
5592 * register_ftrace_function - register a function for profiling
5593 * @ops - ops structure that holds the function for profiling.
5594 *
5595 * Register a function to be called by all functions in the
5596 * kernel.
5597 *
5598 * Note: @ops->func and all the functions it calls must be labeled
5599 * with "notrace", otherwise it will go into a
5600 * recursive loop.
5601 */
5602int register_ftrace_function(struct ftrace_ops *ops)
5603{
5604 int ret = -1;
5605
5606 ftrace_ops_init(ops);
5607
5608 mutex_lock(&ftrace_lock);
5609
5610 ret = ftrace_startup(ops, 0);
5611
5612 mutex_unlock(&ftrace_lock);
5613
5614 return ret;
5615}
5616EXPORT_SYMBOL_GPL(register_ftrace_function);
5617
5618/**
5619 * unregister_ftrace_function - unregister a function for profiling.
5620 * @ops - ops structure that holds the function to unregister
5621 *
5622 * Unregister a function that was added to be called by ftrace profiling.
5623 */
5624int unregister_ftrace_function(struct ftrace_ops *ops)
5625{
5626 int ret;
5627
5628 mutex_lock(&ftrace_lock);
5629 ret = ftrace_shutdown(ops, 0);
5630 mutex_unlock(&ftrace_lock);
5631
5632 return ret;
5633}
5634EXPORT_SYMBOL_GPL(unregister_ftrace_function);
5635
5636int
5637ftrace_enable_sysctl(struct ctl_table *table, int write,
5638 void __user *buffer, size_t *lenp,
5639 loff_t *ppos)
5640{
5641 int ret = -ENODEV;
5642
5643 mutex_lock(&ftrace_lock);
5644
5645 if (unlikely(ftrace_disabled))
5646 goto out;
5647
5648 ret = proc_dointvec(table, write, buffer, lenp, ppos);
5649
5650 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
5651 goto out;
5652
5653 last_ftrace_enabled = !!ftrace_enabled;
5654
5655 if (ftrace_enabled) {
5656
5657 /* we are starting ftrace again */
5658 if (ftrace_ops_list != &ftrace_list_end)
5659 update_ftrace_function();
5660
5661 ftrace_startup_sysctl();
5662
5663 } else {
5664 /* stopping ftrace calls (just send to ftrace_stub) */
5665 ftrace_trace_function = ftrace_stub;
5666
5667 ftrace_shutdown_sysctl();
5668 }
5669
5670 out:
5671 mutex_unlock(&ftrace_lock);
5672 return ret;
5673}
5674
5675#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5676
5677static struct ftrace_ops graph_ops = {
5678 .func = ftrace_stub,
5679 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
5680 FTRACE_OPS_FL_INITIALIZED |
5681 FTRACE_OPS_FL_PID |
5682 FTRACE_OPS_FL_STUB,
5683#ifdef FTRACE_GRAPH_TRAMP_ADDR
5684 .trampoline = FTRACE_GRAPH_TRAMP_ADDR,
5685 /* trampoline_size is only needed for dynamically allocated tramps */
5686#endif
5687 ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
5688};
5689
5690void ftrace_graph_sleep_time_control(bool enable)
5691{
5692 fgraph_sleep_time = enable;
5693}
5694
5695void ftrace_graph_graph_time_control(bool enable)
5696{
5697 fgraph_graph_time = enable;
5698}
5699
5700int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
5701{
5702 return 0;
5703}
5704
5705/* The callbacks that hook a function */
5706trace_func_graph_ret_t ftrace_graph_return =
5707 (trace_func_graph_ret_t)ftrace_stub;
5708trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
5709static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
5710
5711/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
5712static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
5713{
5714 int i;
5715 int ret = 0;
5716 unsigned long flags;
5717 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
5718 struct task_struct *g, *t;
5719
5720 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
5721 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
5722 * sizeof(struct ftrace_ret_stack),
5723 GFP_KERNEL);
5724 if (!ret_stack_list[i]) {
5725 start = 0;
5726 end = i;
5727 ret = -ENOMEM;
5728 goto free;
5729 }
5730 }
5731
5732 read_lock_irqsave(&tasklist_lock, flags);
5733 do_each_thread(g, t) {
5734 if (start == end) {
5735 ret = -EAGAIN;
5736 goto unlock;
5737 }
5738
5739 if (t->ret_stack == NULL) {
5740 atomic_set(&t->tracing_graph_pause, 0);
5741 atomic_set(&t->trace_overrun, 0);
5742 t->curr_ret_stack = -1;
5743 /* Make sure the tasks see the -1 first: */
5744 smp_wmb();
5745 t->ret_stack = ret_stack_list[start++];
5746 }
5747 } while_each_thread(g, t);
5748
5749unlock:
5750 read_unlock_irqrestore(&tasklist_lock, flags);
5751free:
5752 for (i = start; i < end; i++)
5753 kfree(ret_stack_list[i]);
5754 return ret;
5755}
5756
5757static void
5758ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
5759 struct task_struct *prev, struct task_struct *next)
5760{
5761 unsigned long long timestamp;
5762 int index;
5763
5764 /*
5765 * Does the user want to count the time a function was asleep.
5766 * If so, do not update the time stamps.
5767 */
5768 if (fgraph_sleep_time)
5769 return;
5770
5771 timestamp = trace_clock_local();
5772
5773 prev->ftrace_timestamp = timestamp;
5774
5775 /* only process tasks that we timestamped */
5776 if (!next->ftrace_timestamp)
5777 return;
5778
5779 /*
5780 * Update all the counters in next to make up for the
5781 * time next was sleeping.
5782 */
5783 timestamp -= next->ftrace_timestamp;
5784
5785 for (index = next->curr_ret_stack; index >= 0; index--)
5786 next->ret_stack[index].calltime += timestamp;
5787}
5788
5789/* Allocate a return stack for each task */
5790static int start_graph_tracing(void)
5791{
5792 struct ftrace_ret_stack **ret_stack_list;
5793 int ret, cpu;
5794
5795 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
5796 sizeof(struct ftrace_ret_stack *),
5797 GFP_KERNEL);
5798
5799 if (!ret_stack_list)
5800 return -ENOMEM;
5801
5802 /* The cpu_boot init_task->ret_stack will never be freed */
5803 for_each_online_cpu(cpu) {
5804 if (!idle_task(cpu)->ret_stack)
5805 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
5806 }
5807
5808 do {
5809 ret = alloc_retstack_tasklist(ret_stack_list);
5810 } while (ret == -EAGAIN);
5811
5812 if (!ret) {
5813 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5814 if (ret)
5815 pr_info("ftrace_graph: Couldn't activate tracepoint"
5816 " probe to kernel_sched_switch\n");
5817 }
5818
5819 kfree(ret_stack_list);
5820 return ret;
5821}
5822
5823/*
5824 * Hibernation protection.
5825 * The state of the current task is too much unstable during
5826 * suspend/restore to disk. We want to protect against that.
5827 */
5828static int
5829ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
5830 void *unused)
5831{
5832 switch (state) {
5833 case PM_HIBERNATION_PREPARE:
5834 pause_graph_tracing();
5835 break;
5836
5837 case PM_POST_HIBERNATION:
5838 unpause_graph_tracing();
5839 break;
5840 }
5841 return NOTIFY_DONE;
5842}
5843
5844static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
5845{
5846 if (!ftrace_ops_test(&global_ops, trace->func, NULL))
5847 return 0;
5848 return __ftrace_graph_entry(trace);
5849}
5850
5851/*
5852 * The function graph tracer should only trace the functions defined
5853 * by set_ftrace_filter and set_ftrace_notrace. If another function
5854 * tracer ops is registered, the graph tracer requires testing the
5855 * function against the global ops, and not just trace any function
5856 * that any ftrace_ops registered.
5857 */
5858static void update_function_graph_func(void)
5859{
5860 struct ftrace_ops *op;
5861 bool do_test = false;
5862
5863 /*
5864 * The graph and global ops share the same set of functions
5865 * to test. If any other ops is on the list, then
5866 * the graph tracing needs to test if its the function
5867 * it should call.
5868 */
5869 do_for_each_ftrace_op(op, ftrace_ops_list) {
5870 if (op != &global_ops && op != &graph_ops &&
5871 op != &ftrace_list_end) {
5872 do_test = true;
5873 /* in double loop, break out with goto */
5874 goto out;
5875 }
5876 } while_for_each_ftrace_op(op);
5877 out:
5878 if (do_test)
5879 ftrace_graph_entry = ftrace_graph_entry_test;
5880 else
5881 ftrace_graph_entry = __ftrace_graph_entry;
5882}
5883
5884static struct notifier_block ftrace_suspend_notifier = {
5885 .notifier_call = ftrace_suspend_notifier_call,
5886};
5887
5888int register_ftrace_graph(trace_func_graph_ret_t retfunc,
5889 trace_func_graph_ent_t entryfunc)
5890{
5891 int ret = 0;
5892
5893 mutex_lock(&ftrace_lock);
5894
5895 /* we currently allow only one tracer registered at a time */
5896 if (ftrace_graph_active) {
5897 ret = -EBUSY;
5898 goto out;
5899 }
5900
5901 register_pm_notifier(&ftrace_suspend_notifier);
5902
5903 ftrace_graph_active++;
5904 ret = start_graph_tracing();
5905 if (ret) {
5906 ftrace_graph_active--;
5907 goto out;
5908 }
5909
5910 ftrace_graph_return = retfunc;
5911
5912 /*
5913 * Update the indirect function to the entryfunc, and the
5914 * function that gets called to the entry_test first. Then
5915 * call the update fgraph entry function to determine if
5916 * the entryfunc should be called directly or not.
5917 */
5918 __ftrace_graph_entry = entryfunc;
5919 ftrace_graph_entry = ftrace_graph_entry_test;
5920 update_function_graph_func();
5921
5922 ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
5923out:
5924 mutex_unlock(&ftrace_lock);
5925 return ret;
5926}
5927
5928void unregister_ftrace_graph(void)
5929{
5930 mutex_lock(&ftrace_lock);
5931
5932 if (unlikely(!ftrace_graph_active))
5933 goto out;
5934
5935 ftrace_graph_active--;
5936 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
5937 ftrace_graph_entry = ftrace_graph_entry_stub;
5938 __ftrace_graph_entry = ftrace_graph_entry_stub;
5939 ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
5940 unregister_pm_notifier(&ftrace_suspend_notifier);
5941 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5942
5943#ifdef CONFIG_DYNAMIC_FTRACE
5944 /*
5945 * Function graph does not allocate the trampoline, but
5946 * other global_ops do. We need to reset the ALLOC_TRAMP flag
5947 * if one was used.
5948 */
5949 global_ops.trampoline = save_global_trampoline;
5950 if (save_global_flags & FTRACE_OPS_FL_ALLOC_TRAMP)
5951 global_ops.flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
5952#endif
5953
5954 out:
5955 mutex_unlock(&ftrace_lock);
5956}
5957
5958static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
5959
5960static void
5961graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
5962{
5963 atomic_set(&t->tracing_graph_pause, 0);
5964 atomic_set(&t->trace_overrun, 0);
5965 t->ftrace_timestamp = 0;
5966 /* make curr_ret_stack visible before we add the ret_stack */
5967 smp_wmb();
5968 t->ret_stack = ret_stack;
5969}
5970
5971/*
5972 * Allocate a return stack for the idle task. May be the first
5973 * time through, or it may be done by CPU hotplug online.
5974 */
5975void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
5976{
5977 t->curr_ret_stack = -1;
5978 /*
5979 * The idle task has no parent, it either has its own
5980 * stack or no stack at all.
5981 */
5982 if (t->ret_stack)
5983 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
5984
5985 if (ftrace_graph_active) {
5986 struct ftrace_ret_stack *ret_stack;
5987
5988 ret_stack = per_cpu(idle_ret_stack, cpu);
5989 if (!ret_stack) {
5990 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5991 * sizeof(struct ftrace_ret_stack),
5992 GFP_KERNEL);
5993 if (!ret_stack)
5994 return;
5995 per_cpu(idle_ret_stack, cpu) = ret_stack;
5996 }
5997 graph_init_task(t, ret_stack);
5998 }
5999}
6000
6001/* Allocate a return stack for newly created task */
6002void ftrace_graph_init_task(struct task_struct *t)
6003{
6004 /* Make sure we do not use the parent ret_stack */
6005 t->ret_stack = NULL;
6006 t->curr_ret_stack = -1;
6007
6008 if (ftrace_graph_active) {
6009 struct ftrace_ret_stack *ret_stack;
6010
6011 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6012 * sizeof(struct ftrace_ret_stack),
6013 GFP_KERNEL);
6014 if (!ret_stack)
6015 return;
6016 graph_init_task(t, ret_stack);
6017 }
6018}
6019
6020void ftrace_graph_exit_task(struct task_struct *t)
6021{
6022 struct ftrace_ret_stack *ret_stack = t->ret_stack;
6023
6024 t->ret_stack = NULL;
6025 /* NULL must become visible to IRQs before we free it: */
6026 barrier();
6027
6028 kfree(ret_stack);
6029}
6030#endif