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