]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/trace/ftrace.c
Merge tag 'powerpc-4.13-8' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[mirror_ubuntu-artful-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/sections.h>
40 #include <asm/setup.h>
41
42 #include "trace_output.h"
43 #include "trace_stat.h"
44
45 #define FTRACE_WARN_ON(cond) \
46 ({ \
47 int ___r = cond; \
48 if (WARN_ON(___r)) \
49 ftrace_kill(); \
50 ___r; \
51 })
52
53 #define FTRACE_WARN_ON_ONCE(cond) \
54 ({ \
55 int ___r = cond; \
56 if (WARN_ON_ONCE(___r)) \
57 ftrace_kill(); \
58 ___r; \
59 })
60
61 /* hash bits for specific function selection */
62 #define FTRACE_HASH_BITS 7
63 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
64 #define FTRACE_HASH_DEFAULT_BITS 10
65 #define FTRACE_HASH_MAX_BITS 12
66
67 #ifdef CONFIG_DYNAMIC_FTRACE
68 #define INIT_OPS_HASH(opsname) \
69 .func_hash = &opsname.local_hash, \
70 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
71 #define ASSIGN_OPS_HASH(opsname, val) \
72 .func_hash = val, \
73 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
74 #else
75 #define INIT_OPS_HASH(opsname)
76 #define ASSIGN_OPS_HASH(opsname, val)
77 #endif
78
79 static struct ftrace_ops ftrace_list_end __read_mostly = {
80 .func = ftrace_stub,
81 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
82 INIT_OPS_HASH(ftrace_list_end)
83 };
84
85 /* ftrace_enabled is a method to turn ftrace on or off */
86 int ftrace_enabled __read_mostly;
87 static int last_ftrace_enabled;
88
89 /* Current function tracing op */
90 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
91 /* What to set function_trace_op to */
92 static struct ftrace_ops *set_function_trace_op;
93
94 static bool ftrace_pids_enabled(struct ftrace_ops *ops)
95 {
96 struct trace_array *tr;
97
98 if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
99 return false;
100
101 tr = ops->private;
102
103 return tr->function_pids != NULL;
104 }
105
106 static void ftrace_update_trampoline(struct ftrace_ops *ops);
107
108 /*
109 * ftrace_disabled is set when an anomaly is discovered.
110 * ftrace_disabled is much stronger than ftrace_enabled.
111 */
112 static int ftrace_disabled __read_mostly;
113
114 static DEFINE_MUTEX(ftrace_lock);
115
116 static struct ftrace_ops __rcu *ftrace_ops_list __read_mostly = &ftrace_list_end;
117 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
118 static struct ftrace_ops global_ops;
119
120 #if ARCH_SUPPORTS_FTRACE_OPS
121 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
122 struct ftrace_ops *op, struct pt_regs *regs);
123 #else
124 /* See comment below, where ftrace_ops_list_func is defined */
125 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
126 #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
127 #endif
128
129 /*
130 * Traverse the ftrace_global_list, invoking all entries. The reason that we
131 * can use rcu_dereference_raw_notrace() is that elements removed from this list
132 * are simply leaked, so there is no need to interact with a grace-period
133 * mechanism. The rcu_dereference_raw_notrace() calls are needed to handle
134 * concurrent insertions into the ftrace_global_list.
135 *
136 * Silly Alpha and silly pointer-speculation compiler optimizations!
137 */
138 #define do_for_each_ftrace_op(op, list) \
139 op = rcu_dereference_raw_notrace(list); \
140 do
141
142 /*
143 * Optimized for just a single item in the list (as that is the normal case).
144 */
145 #define while_for_each_ftrace_op(op) \
146 while (likely(op = rcu_dereference_raw_notrace((op)->next)) && \
147 unlikely((op) != &ftrace_list_end))
148
149 static inline void ftrace_ops_init(struct ftrace_ops *ops)
150 {
151 #ifdef CONFIG_DYNAMIC_FTRACE
152 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
153 mutex_init(&ops->local_hash.regex_lock);
154 ops->func_hash = &ops->local_hash;
155 ops->flags |= FTRACE_OPS_FL_INITIALIZED;
156 }
157 #endif
158 }
159
160 /**
161 * ftrace_nr_registered_ops - return number of ops registered
162 *
163 * Returns the number of ftrace_ops registered and tracing functions
164 */
165 int ftrace_nr_registered_ops(void)
166 {
167 struct ftrace_ops *ops;
168 int cnt = 0;
169
170 mutex_lock(&ftrace_lock);
171
172 for (ops = rcu_dereference_protected(ftrace_ops_list,
173 lockdep_is_held(&ftrace_lock));
174 ops != &ftrace_list_end;
175 ops = rcu_dereference_protected(ops->next,
176 lockdep_is_held(&ftrace_lock)))
177 cnt++;
178
179 mutex_unlock(&ftrace_lock);
180
181 return cnt;
182 }
183
184 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
185 struct ftrace_ops *op, struct pt_regs *regs)
186 {
187 struct trace_array *tr = op->private;
188
189 if (tr && this_cpu_read(tr->trace_buffer.data->ftrace_ignore_pid))
190 return;
191
192 op->saved_func(ip, parent_ip, op, regs);
193 }
194
195 /**
196 * clear_ftrace_function - reset the ftrace function
197 *
198 * This NULLs the ftrace function and in essence stops
199 * tracing. There may be lag
200 */
201 void clear_ftrace_function(void)
202 {
203 ftrace_trace_function = ftrace_stub;
204 }
205
206 static void per_cpu_ops_disable_all(struct ftrace_ops *ops)
207 {
208 int cpu;
209
210 for_each_possible_cpu(cpu)
211 *per_cpu_ptr(ops->disabled, cpu) = 1;
212 }
213
214 static int per_cpu_ops_alloc(struct ftrace_ops *ops)
215 {
216 int __percpu *disabled;
217
218 if (WARN_ON_ONCE(!(ops->flags & FTRACE_OPS_FL_PER_CPU)))
219 return -EINVAL;
220
221 disabled = alloc_percpu(int);
222 if (!disabled)
223 return -ENOMEM;
224
225 ops->disabled = disabled;
226 per_cpu_ops_disable_all(ops);
227 return 0;
228 }
229
230 static void ftrace_sync(struct work_struct *work)
231 {
232 /*
233 * This function is just a stub to implement a hard force
234 * of synchronize_sched(). This requires synchronizing
235 * tasks even in userspace and idle.
236 *
237 * Yes, function tracing is rude.
238 */
239 }
240
241 static void ftrace_sync_ipi(void *data)
242 {
243 /* Probably not needed, but do it anyway */
244 smp_rmb();
245 }
246
247 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
248 static void update_function_graph_func(void);
249
250 /* Both enabled by default (can be cleared by function_graph tracer flags */
251 static bool fgraph_sleep_time = true;
252 static bool fgraph_graph_time = true;
253
254 #else
255 static inline void update_function_graph_func(void) { }
256 #endif
257
258
259 static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
260 {
261 /*
262 * If this is a dynamic, RCU, or per CPU ops, or we force list func,
263 * then it needs to call the list anyway.
264 */
265 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU |
266 FTRACE_OPS_FL_RCU) || FTRACE_FORCE_LIST_FUNC)
267 return ftrace_ops_list_func;
268
269 return ftrace_ops_get_func(ops);
270 }
271
272 static void update_ftrace_function(void)
273 {
274 ftrace_func_t func;
275
276 /*
277 * Prepare the ftrace_ops that the arch callback will use.
278 * If there's only one ftrace_ops registered, the ftrace_ops_list
279 * will point to the ops we want.
280 */
281 set_function_trace_op = rcu_dereference_protected(ftrace_ops_list,
282 lockdep_is_held(&ftrace_lock));
283
284 /* If there's no ftrace_ops registered, just call the stub function */
285 if (set_function_trace_op == &ftrace_list_end) {
286 func = ftrace_stub;
287
288 /*
289 * If we are at the end of the list and this ops is
290 * recursion safe and not dynamic and the arch supports passing ops,
291 * then have the mcount trampoline call the function directly.
292 */
293 } else if (rcu_dereference_protected(ftrace_ops_list->next,
294 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
295 func = ftrace_ops_get_list_func(ftrace_ops_list);
296
297 } else {
298 /* Just use the default ftrace_ops */
299 set_function_trace_op = &ftrace_list_end;
300 func = ftrace_ops_list_func;
301 }
302
303 update_function_graph_func();
304
305 /* If there's no change, then do nothing more here */
306 if (ftrace_trace_function == func)
307 return;
308
309 /*
310 * If we are using the list function, it doesn't care
311 * about the function_trace_ops.
312 */
313 if (func == ftrace_ops_list_func) {
314 ftrace_trace_function = func;
315 /*
316 * Don't even bother setting function_trace_ops,
317 * it would be racy to do so anyway.
318 */
319 return;
320 }
321
322 #ifndef CONFIG_DYNAMIC_FTRACE
323 /*
324 * For static tracing, we need to be a bit more careful.
325 * The function change takes affect immediately. Thus,
326 * we need to coorditate the setting of the function_trace_ops
327 * with the setting of the ftrace_trace_function.
328 *
329 * Set the function to the list ops, which will call the
330 * function we want, albeit indirectly, but it handles the
331 * ftrace_ops and doesn't depend on function_trace_op.
332 */
333 ftrace_trace_function = ftrace_ops_list_func;
334 /*
335 * Make sure all CPUs see this. Yes this is slow, but static
336 * tracing is slow and nasty to have enabled.
337 */
338 schedule_on_each_cpu(ftrace_sync);
339 /* Now all cpus are using the list ops. */
340 function_trace_op = set_function_trace_op;
341 /* Make sure the function_trace_op is visible on all CPUs */
342 smp_wmb();
343 /* Nasty way to force a rmb on all cpus */
344 smp_call_function(ftrace_sync_ipi, NULL, 1);
345 /* OK, we are all set to update the ftrace_trace_function now! */
346 #endif /* !CONFIG_DYNAMIC_FTRACE */
347
348 ftrace_trace_function = func;
349 }
350
351 int using_ftrace_ops_list_func(void)
352 {
353 return ftrace_trace_function == ftrace_ops_list_func;
354 }
355
356 static void add_ftrace_ops(struct ftrace_ops __rcu **list,
357 struct ftrace_ops *ops)
358 {
359 rcu_assign_pointer(ops->next, *list);
360
361 /*
362 * We are entering ops into the list but another
363 * CPU might be walking that list. We need to make sure
364 * the ops->next pointer is valid before another CPU sees
365 * the ops pointer included into the list.
366 */
367 rcu_assign_pointer(*list, ops);
368 }
369
370 static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
371 struct ftrace_ops *ops)
372 {
373 struct ftrace_ops **p;
374
375 /*
376 * If we are removing the last function, then simply point
377 * to the ftrace_stub.
378 */
379 if (rcu_dereference_protected(*list,
380 lockdep_is_held(&ftrace_lock)) == ops &&
381 rcu_dereference_protected(ops->next,
382 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
383 *list = &ftrace_list_end;
384 return 0;
385 }
386
387 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
388 if (*p == ops)
389 break;
390
391 if (*p != ops)
392 return -1;
393
394 *p = (*p)->next;
395 return 0;
396 }
397
398 static void ftrace_update_trampoline(struct ftrace_ops *ops);
399
400 static int __register_ftrace_function(struct ftrace_ops *ops)
401 {
402 if (ops->flags & FTRACE_OPS_FL_DELETED)
403 return -EINVAL;
404
405 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
406 return -EBUSY;
407
408 #ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
409 /*
410 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
411 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
412 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
413 */
414 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
415 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
416 return -EINVAL;
417
418 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
419 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
420 #endif
421
422 if (!core_kernel_data((unsigned long)ops))
423 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
424
425 if (ops->flags & FTRACE_OPS_FL_PER_CPU) {
426 if (per_cpu_ops_alloc(ops))
427 return -ENOMEM;
428 }
429
430 add_ftrace_ops(&ftrace_ops_list, ops);
431
432 /* Always save the function, and reset at unregistering */
433 ops->saved_func = ops->func;
434
435 if (ftrace_pids_enabled(ops))
436 ops->func = ftrace_pid_func;
437
438 ftrace_update_trampoline(ops);
439
440 if (ftrace_enabled)
441 update_ftrace_function();
442
443 return 0;
444 }
445
446 static int __unregister_ftrace_function(struct ftrace_ops *ops)
447 {
448 int ret;
449
450 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
451 return -EBUSY;
452
453 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
454
455 if (ret < 0)
456 return ret;
457
458 if (ftrace_enabled)
459 update_ftrace_function();
460
461 ops->func = ops->saved_func;
462
463 return 0;
464 }
465
466 static void ftrace_update_pid_func(void)
467 {
468 struct ftrace_ops *op;
469
470 /* Only do something if we are tracing something */
471 if (ftrace_trace_function == ftrace_stub)
472 return;
473
474 do_for_each_ftrace_op(op, ftrace_ops_list) {
475 if (op->flags & FTRACE_OPS_FL_PID) {
476 op->func = ftrace_pids_enabled(op) ?
477 ftrace_pid_func : op->saved_func;
478 ftrace_update_trampoline(op);
479 }
480 } while_for_each_ftrace_op(op);
481
482 update_ftrace_function();
483 }
484
485 #ifdef CONFIG_FUNCTION_PROFILER
486 struct ftrace_profile {
487 struct hlist_node node;
488 unsigned long ip;
489 unsigned long counter;
490 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
491 unsigned long long time;
492 unsigned long long time_squared;
493 #endif
494 };
495
496 struct ftrace_profile_page {
497 struct ftrace_profile_page *next;
498 unsigned long index;
499 struct ftrace_profile records[];
500 };
501
502 struct ftrace_profile_stat {
503 atomic_t disabled;
504 struct hlist_head *hash;
505 struct ftrace_profile_page *pages;
506 struct ftrace_profile_page *start;
507 struct tracer_stat stat;
508 };
509
510 #define PROFILE_RECORDS_SIZE \
511 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
512
513 #define PROFILES_PER_PAGE \
514 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
515
516 static int ftrace_profile_enabled __read_mostly;
517
518 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
519 static DEFINE_MUTEX(ftrace_profile_lock);
520
521 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
522
523 #define FTRACE_PROFILE_HASH_BITS 10
524 #define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
525
526 static void *
527 function_stat_next(void *v, int idx)
528 {
529 struct ftrace_profile *rec = v;
530 struct ftrace_profile_page *pg;
531
532 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
533
534 again:
535 if (idx != 0)
536 rec++;
537
538 if ((void *)rec >= (void *)&pg->records[pg->index]) {
539 pg = pg->next;
540 if (!pg)
541 return NULL;
542 rec = &pg->records[0];
543 if (!rec->counter)
544 goto again;
545 }
546
547 return rec;
548 }
549
550 static void *function_stat_start(struct tracer_stat *trace)
551 {
552 struct ftrace_profile_stat *stat =
553 container_of(trace, struct ftrace_profile_stat, stat);
554
555 if (!stat || !stat->start)
556 return NULL;
557
558 return function_stat_next(&stat->start->records[0], 0);
559 }
560
561 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
562 /* function graph compares on total time */
563 static int function_stat_cmp(void *p1, void *p2)
564 {
565 struct ftrace_profile *a = p1;
566 struct ftrace_profile *b = p2;
567
568 if (a->time < b->time)
569 return -1;
570 if (a->time > b->time)
571 return 1;
572 else
573 return 0;
574 }
575 #else
576 /* not function graph compares against hits */
577 static int function_stat_cmp(void *p1, void *p2)
578 {
579 struct ftrace_profile *a = p1;
580 struct ftrace_profile *b = p2;
581
582 if (a->counter < b->counter)
583 return -1;
584 if (a->counter > b->counter)
585 return 1;
586 else
587 return 0;
588 }
589 #endif
590
591 static int function_stat_headers(struct seq_file *m)
592 {
593 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
594 seq_puts(m, " Function "
595 "Hit Time Avg s^2\n"
596 " -------- "
597 "--- ---- --- ---\n");
598 #else
599 seq_puts(m, " Function Hit\n"
600 " -------- ---\n");
601 #endif
602 return 0;
603 }
604
605 static int function_stat_show(struct seq_file *m, void *v)
606 {
607 struct ftrace_profile *rec = v;
608 char str[KSYM_SYMBOL_LEN];
609 int ret = 0;
610 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
611 static struct trace_seq s;
612 unsigned long long avg;
613 unsigned long long stddev;
614 #endif
615 mutex_lock(&ftrace_profile_lock);
616
617 /* we raced with function_profile_reset() */
618 if (unlikely(rec->counter == 0)) {
619 ret = -EBUSY;
620 goto out;
621 }
622
623 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
624 avg = rec->time;
625 do_div(avg, rec->counter);
626 if (tracing_thresh && (avg < tracing_thresh))
627 goto out;
628 #endif
629
630 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
631 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
632
633 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
634 seq_puts(m, " ");
635
636 /* Sample standard deviation (s^2) */
637 if (rec->counter <= 1)
638 stddev = 0;
639 else {
640 /*
641 * Apply Welford's method:
642 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
643 */
644 stddev = rec->counter * rec->time_squared -
645 rec->time * rec->time;
646
647 /*
648 * Divide only 1000 for ns^2 -> us^2 conversion.
649 * trace_print_graph_duration will divide 1000 again.
650 */
651 do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
652 }
653
654 trace_seq_init(&s);
655 trace_print_graph_duration(rec->time, &s);
656 trace_seq_puts(&s, " ");
657 trace_print_graph_duration(avg, &s);
658 trace_seq_puts(&s, " ");
659 trace_print_graph_duration(stddev, &s);
660 trace_print_seq(m, &s);
661 #endif
662 seq_putc(m, '\n');
663 out:
664 mutex_unlock(&ftrace_profile_lock);
665
666 return ret;
667 }
668
669 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
670 {
671 struct ftrace_profile_page *pg;
672
673 pg = stat->pages = stat->start;
674
675 while (pg) {
676 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
677 pg->index = 0;
678 pg = pg->next;
679 }
680
681 memset(stat->hash, 0,
682 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
683 }
684
685 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
686 {
687 struct ftrace_profile_page *pg;
688 int functions;
689 int pages;
690 int i;
691
692 /* If we already allocated, do nothing */
693 if (stat->pages)
694 return 0;
695
696 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
697 if (!stat->pages)
698 return -ENOMEM;
699
700 #ifdef CONFIG_DYNAMIC_FTRACE
701 functions = ftrace_update_tot_cnt;
702 #else
703 /*
704 * We do not know the number of functions that exist because
705 * dynamic tracing is what counts them. With past experience
706 * we have around 20K functions. That should be more than enough.
707 * It is highly unlikely we will execute every function in
708 * the kernel.
709 */
710 functions = 20000;
711 #endif
712
713 pg = stat->start = stat->pages;
714
715 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
716
717 for (i = 1; i < pages; i++) {
718 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
719 if (!pg->next)
720 goto out_free;
721 pg = pg->next;
722 }
723
724 return 0;
725
726 out_free:
727 pg = stat->start;
728 while (pg) {
729 unsigned long tmp = (unsigned long)pg;
730
731 pg = pg->next;
732 free_page(tmp);
733 }
734
735 stat->pages = NULL;
736 stat->start = NULL;
737
738 return -ENOMEM;
739 }
740
741 static int ftrace_profile_init_cpu(int cpu)
742 {
743 struct ftrace_profile_stat *stat;
744 int size;
745
746 stat = &per_cpu(ftrace_profile_stats, cpu);
747
748 if (stat->hash) {
749 /* If the profile is already created, simply reset it */
750 ftrace_profile_reset(stat);
751 return 0;
752 }
753
754 /*
755 * We are profiling all functions, but usually only a few thousand
756 * functions are hit. We'll make a hash of 1024 items.
757 */
758 size = FTRACE_PROFILE_HASH_SIZE;
759
760 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
761
762 if (!stat->hash)
763 return -ENOMEM;
764
765 /* Preallocate the function profiling pages */
766 if (ftrace_profile_pages_init(stat) < 0) {
767 kfree(stat->hash);
768 stat->hash = NULL;
769 return -ENOMEM;
770 }
771
772 return 0;
773 }
774
775 static int ftrace_profile_init(void)
776 {
777 int cpu;
778 int ret = 0;
779
780 for_each_possible_cpu(cpu) {
781 ret = ftrace_profile_init_cpu(cpu);
782 if (ret)
783 break;
784 }
785
786 return ret;
787 }
788
789 /* interrupts must be disabled */
790 static struct ftrace_profile *
791 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
792 {
793 struct ftrace_profile *rec;
794 struct hlist_head *hhd;
795 unsigned long key;
796
797 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
798 hhd = &stat->hash[key];
799
800 if (hlist_empty(hhd))
801 return NULL;
802
803 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
804 if (rec->ip == ip)
805 return rec;
806 }
807
808 return NULL;
809 }
810
811 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
812 struct ftrace_profile *rec)
813 {
814 unsigned long key;
815
816 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
817 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
818 }
819
820 /*
821 * The memory is already allocated, this simply finds a new record to use.
822 */
823 static struct ftrace_profile *
824 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
825 {
826 struct ftrace_profile *rec = NULL;
827
828 /* prevent recursion (from NMIs) */
829 if (atomic_inc_return(&stat->disabled) != 1)
830 goto out;
831
832 /*
833 * Try to find the function again since an NMI
834 * could have added it
835 */
836 rec = ftrace_find_profiled_func(stat, ip);
837 if (rec)
838 goto out;
839
840 if (stat->pages->index == PROFILES_PER_PAGE) {
841 if (!stat->pages->next)
842 goto out;
843 stat->pages = stat->pages->next;
844 }
845
846 rec = &stat->pages->records[stat->pages->index++];
847 rec->ip = ip;
848 ftrace_add_profile(stat, rec);
849
850 out:
851 atomic_dec(&stat->disabled);
852
853 return rec;
854 }
855
856 static void
857 function_profile_call(unsigned long ip, unsigned long parent_ip,
858 struct ftrace_ops *ops, struct pt_regs *regs)
859 {
860 struct ftrace_profile_stat *stat;
861 struct ftrace_profile *rec;
862 unsigned long flags;
863
864 if (!ftrace_profile_enabled)
865 return;
866
867 local_irq_save(flags);
868
869 stat = this_cpu_ptr(&ftrace_profile_stats);
870 if (!stat->hash || !ftrace_profile_enabled)
871 goto out;
872
873 rec = ftrace_find_profiled_func(stat, ip);
874 if (!rec) {
875 rec = ftrace_profile_alloc(stat, ip);
876 if (!rec)
877 goto out;
878 }
879
880 rec->counter++;
881 out:
882 local_irq_restore(flags);
883 }
884
885 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
886 static int profile_graph_entry(struct ftrace_graph_ent *trace)
887 {
888 int index = trace->depth;
889
890 function_profile_call(trace->func, 0, NULL, NULL);
891
892 /* If function graph is shutting down, ret_stack can be NULL */
893 if (!current->ret_stack)
894 return 0;
895
896 if (index >= 0 && index < FTRACE_RETFUNC_DEPTH)
897 current->ret_stack[index].subtime = 0;
898
899 return 1;
900 }
901
902 static void profile_graph_return(struct ftrace_graph_ret *trace)
903 {
904 struct ftrace_profile_stat *stat;
905 unsigned long long calltime;
906 struct ftrace_profile *rec;
907 unsigned long flags;
908
909 local_irq_save(flags);
910 stat = this_cpu_ptr(&ftrace_profile_stats);
911 if (!stat->hash || !ftrace_profile_enabled)
912 goto out;
913
914 /* If the calltime was zero'd ignore it */
915 if (!trace->calltime)
916 goto out;
917
918 calltime = trace->rettime - trace->calltime;
919
920 if (!fgraph_graph_time) {
921 int index;
922
923 index = trace->depth;
924
925 /* Append this call time to the parent time to subtract */
926 if (index)
927 current->ret_stack[index - 1].subtime += calltime;
928
929 if (current->ret_stack[index].subtime < calltime)
930 calltime -= current->ret_stack[index].subtime;
931 else
932 calltime = 0;
933 }
934
935 rec = ftrace_find_profiled_func(stat, trace->func);
936 if (rec) {
937 rec->time += calltime;
938 rec->time_squared += calltime * calltime;
939 }
940
941 out:
942 local_irq_restore(flags);
943 }
944
945 static int register_ftrace_profiler(void)
946 {
947 return register_ftrace_graph(&profile_graph_return,
948 &profile_graph_entry);
949 }
950
951 static void unregister_ftrace_profiler(void)
952 {
953 unregister_ftrace_graph();
954 }
955 #else
956 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
957 .func = function_profile_call,
958 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
959 INIT_OPS_HASH(ftrace_profile_ops)
960 };
961
962 static int register_ftrace_profiler(void)
963 {
964 return register_ftrace_function(&ftrace_profile_ops);
965 }
966
967 static void unregister_ftrace_profiler(void)
968 {
969 unregister_ftrace_function(&ftrace_profile_ops);
970 }
971 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
972
973 static ssize_t
974 ftrace_profile_write(struct file *filp, const char __user *ubuf,
975 size_t cnt, loff_t *ppos)
976 {
977 unsigned long val;
978 int ret;
979
980 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
981 if (ret)
982 return ret;
983
984 val = !!val;
985
986 mutex_lock(&ftrace_profile_lock);
987 if (ftrace_profile_enabled ^ val) {
988 if (val) {
989 ret = ftrace_profile_init();
990 if (ret < 0) {
991 cnt = ret;
992 goto out;
993 }
994
995 ret = register_ftrace_profiler();
996 if (ret < 0) {
997 cnt = ret;
998 goto out;
999 }
1000 ftrace_profile_enabled = 1;
1001 } else {
1002 ftrace_profile_enabled = 0;
1003 /*
1004 * unregister_ftrace_profiler calls stop_machine
1005 * so this acts like an synchronize_sched.
1006 */
1007 unregister_ftrace_profiler();
1008 }
1009 }
1010 out:
1011 mutex_unlock(&ftrace_profile_lock);
1012
1013 *ppos += cnt;
1014
1015 return cnt;
1016 }
1017
1018 static ssize_t
1019 ftrace_profile_read(struct file *filp, char __user *ubuf,
1020 size_t cnt, loff_t *ppos)
1021 {
1022 char buf[64]; /* big enough to hold a number */
1023 int r;
1024
1025 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1026 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1027 }
1028
1029 static const struct file_operations ftrace_profile_fops = {
1030 .open = tracing_open_generic,
1031 .read = ftrace_profile_read,
1032 .write = ftrace_profile_write,
1033 .llseek = default_llseek,
1034 };
1035
1036 /* used to initialize the real stat files */
1037 static struct tracer_stat function_stats __initdata = {
1038 .name = "functions",
1039 .stat_start = function_stat_start,
1040 .stat_next = function_stat_next,
1041 .stat_cmp = function_stat_cmp,
1042 .stat_headers = function_stat_headers,
1043 .stat_show = function_stat_show
1044 };
1045
1046 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1047 {
1048 struct ftrace_profile_stat *stat;
1049 struct dentry *entry;
1050 char *name;
1051 int ret;
1052 int cpu;
1053
1054 for_each_possible_cpu(cpu) {
1055 stat = &per_cpu(ftrace_profile_stats, cpu);
1056
1057 name = kasprintf(GFP_KERNEL, "function%d", cpu);
1058 if (!name) {
1059 /*
1060 * The files created are permanent, if something happens
1061 * we still do not free memory.
1062 */
1063 WARN(1,
1064 "Could not allocate stat file for cpu %d\n",
1065 cpu);
1066 return;
1067 }
1068 stat->stat = function_stats;
1069 stat->stat.name = name;
1070 ret = register_stat_tracer(&stat->stat);
1071 if (ret) {
1072 WARN(1,
1073 "Could not register function stat for cpu %d\n",
1074 cpu);
1075 kfree(name);
1076 return;
1077 }
1078 }
1079
1080 entry = tracefs_create_file("function_profile_enabled", 0644,
1081 d_tracer, NULL, &ftrace_profile_fops);
1082 if (!entry)
1083 pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1084 }
1085
1086 #else /* CONFIG_FUNCTION_PROFILER */
1087 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1088 {
1089 }
1090 #endif /* CONFIG_FUNCTION_PROFILER */
1091
1092 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1093
1094 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1095 static int ftrace_graph_active;
1096 #else
1097 # define ftrace_graph_active 0
1098 #endif
1099
1100 #ifdef CONFIG_DYNAMIC_FTRACE
1101
1102 static struct ftrace_ops *removed_ops;
1103
1104 /*
1105 * Set when doing a global update, like enabling all recs or disabling them.
1106 * It is not set when just updating a single ftrace_ops.
1107 */
1108 static bool update_all_ops;
1109
1110 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1111 # error Dynamic ftrace depends on MCOUNT_RECORD
1112 #endif
1113
1114 struct ftrace_func_entry {
1115 struct hlist_node hlist;
1116 unsigned long ip;
1117 };
1118
1119 struct ftrace_func_probe {
1120 struct ftrace_probe_ops *probe_ops;
1121 struct ftrace_ops ops;
1122 struct trace_array *tr;
1123 struct list_head list;
1124 void *data;
1125 int ref;
1126 };
1127
1128 /*
1129 * We make these constant because no one should touch them,
1130 * but they are used as the default "empty hash", to avoid allocating
1131 * it all the time. These are in a read only section such that if
1132 * anyone does try to modify it, it will cause an exception.
1133 */
1134 static const struct hlist_head empty_buckets[1];
1135 static const struct ftrace_hash empty_hash = {
1136 .buckets = (struct hlist_head *)empty_buckets,
1137 };
1138 #define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1139
1140 static struct ftrace_ops global_ops = {
1141 .func = ftrace_stub,
1142 .local_hash.notrace_hash = EMPTY_HASH,
1143 .local_hash.filter_hash = EMPTY_HASH,
1144 INIT_OPS_HASH(global_ops)
1145 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
1146 FTRACE_OPS_FL_INITIALIZED |
1147 FTRACE_OPS_FL_PID,
1148 };
1149
1150 /*
1151 * This is used by __kernel_text_address() to return true if the
1152 * address is on a dynamically allocated trampoline that would
1153 * not return true for either core_kernel_text() or
1154 * is_module_text_address().
1155 */
1156 bool is_ftrace_trampoline(unsigned long addr)
1157 {
1158 struct ftrace_ops *op;
1159 bool ret = false;
1160
1161 /*
1162 * Some of the ops may be dynamically allocated,
1163 * they are freed after a synchronize_sched().
1164 */
1165 preempt_disable_notrace();
1166
1167 do_for_each_ftrace_op(op, ftrace_ops_list) {
1168 /*
1169 * This is to check for dynamically allocated trampolines.
1170 * Trampolines that are in kernel text will have
1171 * core_kernel_text() return true.
1172 */
1173 if (op->trampoline && op->trampoline_size)
1174 if (addr >= op->trampoline &&
1175 addr < op->trampoline + op->trampoline_size) {
1176 ret = true;
1177 goto out;
1178 }
1179 } while_for_each_ftrace_op(op);
1180
1181 out:
1182 preempt_enable_notrace();
1183
1184 return ret;
1185 }
1186
1187 struct ftrace_page {
1188 struct ftrace_page *next;
1189 struct dyn_ftrace *records;
1190 int index;
1191 int size;
1192 };
1193
1194 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1195 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1196
1197 /* estimate from running different kernels */
1198 #define NR_TO_INIT 10000
1199
1200 static struct ftrace_page *ftrace_pages_start;
1201 static struct ftrace_page *ftrace_pages;
1202
1203 static __always_inline unsigned long
1204 ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1205 {
1206 if (hash->size_bits > 0)
1207 return hash_long(ip, hash->size_bits);
1208
1209 return 0;
1210 }
1211
1212 /* Only use this function if ftrace_hash_empty() has already been tested */
1213 static __always_inline struct ftrace_func_entry *
1214 __ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1215 {
1216 unsigned long key;
1217 struct ftrace_func_entry *entry;
1218 struct hlist_head *hhd;
1219
1220 key = ftrace_hash_key(hash, ip);
1221 hhd = &hash->buckets[key];
1222
1223 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1224 if (entry->ip == ip)
1225 return entry;
1226 }
1227 return NULL;
1228 }
1229
1230 /**
1231 * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1232 * @hash: The hash to look at
1233 * @ip: The instruction pointer to test
1234 *
1235 * Search a given @hash to see if a given instruction pointer (@ip)
1236 * exists in it.
1237 *
1238 * Returns the entry that holds the @ip if found. NULL otherwise.
1239 */
1240 struct ftrace_func_entry *
1241 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1242 {
1243 if (ftrace_hash_empty(hash))
1244 return NULL;
1245
1246 return __ftrace_lookup_ip(hash, ip);
1247 }
1248
1249 static void __add_hash_entry(struct ftrace_hash *hash,
1250 struct ftrace_func_entry *entry)
1251 {
1252 struct hlist_head *hhd;
1253 unsigned long key;
1254
1255 key = ftrace_hash_key(hash, entry->ip);
1256 hhd = &hash->buckets[key];
1257 hlist_add_head(&entry->hlist, hhd);
1258 hash->count++;
1259 }
1260
1261 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1262 {
1263 struct ftrace_func_entry *entry;
1264
1265 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1266 if (!entry)
1267 return -ENOMEM;
1268
1269 entry->ip = ip;
1270 __add_hash_entry(hash, entry);
1271
1272 return 0;
1273 }
1274
1275 static void
1276 free_hash_entry(struct ftrace_hash *hash,
1277 struct ftrace_func_entry *entry)
1278 {
1279 hlist_del(&entry->hlist);
1280 kfree(entry);
1281 hash->count--;
1282 }
1283
1284 static void
1285 remove_hash_entry(struct ftrace_hash *hash,
1286 struct ftrace_func_entry *entry)
1287 {
1288 hlist_del_rcu(&entry->hlist);
1289 hash->count--;
1290 }
1291
1292 static void ftrace_hash_clear(struct ftrace_hash *hash)
1293 {
1294 struct hlist_head *hhd;
1295 struct hlist_node *tn;
1296 struct ftrace_func_entry *entry;
1297 int size = 1 << hash->size_bits;
1298 int i;
1299
1300 if (!hash->count)
1301 return;
1302
1303 for (i = 0; i < size; i++) {
1304 hhd = &hash->buckets[i];
1305 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1306 free_hash_entry(hash, entry);
1307 }
1308 FTRACE_WARN_ON(hash->count);
1309 }
1310
1311 static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1312 {
1313 list_del(&ftrace_mod->list);
1314 kfree(ftrace_mod->module);
1315 kfree(ftrace_mod->func);
1316 kfree(ftrace_mod);
1317 }
1318
1319 static void clear_ftrace_mod_list(struct list_head *head)
1320 {
1321 struct ftrace_mod_load *p, *n;
1322
1323 /* stack tracer isn't supported yet */
1324 if (!head)
1325 return;
1326
1327 mutex_lock(&ftrace_lock);
1328 list_for_each_entry_safe(p, n, head, list)
1329 free_ftrace_mod(p);
1330 mutex_unlock(&ftrace_lock);
1331 }
1332
1333 static void free_ftrace_hash(struct ftrace_hash *hash)
1334 {
1335 if (!hash || hash == EMPTY_HASH)
1336 return;
1337 ftrace_hash_clear(hash);
1338 kfree(hash->buckets);
1339 kfree(hash);
1340 }
1341
1342 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1343 {
1344 struct ftrace_hash *hash;
1345
1346 hash = container_of(rcu, struct ftrace_hash, rcu);
1347 free_ftrace_hash(hash);
1348 }
1349
1350 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1351 {
1352 if (!hash || hash == EMPTY_HASH)
1353 return;
1354 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1355 }
1356
1357 void ftrace_free_filter(struct ftrace_ops *ops)
1358 {
1359 ftrace_ops_init(ops);
1360 free_ftrace_hash(ops->func_hash->filter_hash);
1361 free_ftrace_hash(ops->func_hash->notrace_hash);
1362 }
1363
1364 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1365 {
1366 struct ftrace_hash *hash;
1367 int size;
1368
1369 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1370 if (!hash)
1371 return NULL;
1372
1373 size = 1 << size_bits;
1374 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1375
1376 if (!hash->buckets) {
1377 kfree(hash);
1378 return NULL;
1379 }
1380
1381 hash->size_bits = size_bits;
1382
1383 return hash;
1384 }
1385
1386
1387 static int ftrace_add_mod(struct trace_array *tr,
1388 const char *func, const char *module,
1389 int enable)
1390 {
1391 struct ftrace_mod_load *ftrace_mod;
1392 struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1393
1394 ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1395 if (!ftrace_mod)
1396 return -ENOMEM;
1397
1398 ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1399 ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1400 ftrace_mod->enable = enable;
1401
1402 if (!ftrace_mod->func || !ftrace_mod->module)
1403 goto out_free;
1404
1405 list_add(&ftrace_mod->list, mod_head);
1406
1407 return 0;
1408
1409 out_free:
1410 free_ftrace_mod(ftrace_mod);
1411
1412 return -ENOMEM;
1413 }
1414
1415 static struct ftrace_hash *
1416 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1417 {
1418 struct ftrace_func_entry *entry;
1419 struct ftrace_hash *new_hash;
1420 int size;
1421 int ret;
1422 int i;
1423
1424 new_hash = alloc_ftrace_hash(size_bits);
1425 if (!new_hash)
1426 return NULL;
1427
1428 if (hash)
1429 new_hash->flags = hash->flags;
1430
1431 /* Empty hash? */
1432 if (ftrace_hash_empty(hash))
1433 return new_hash;
1434
1435 size = 1 << hash->size_bits;
1436 for (i = 0; i < size; i++) {
1437 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1438 ret = add_hash_entry(new_hash, entry->ip);
1439 if (ret < 0)
1440 goto free_hash;
1441 }
1442 }
1443
1444 FTRACE_WARN_ON(new_hash->count != hash->count);
1445
1446 return new_hash;
1447
1448 free_hash:
1449 free_ftrace_hash(new_hash);
1450 return NULL;
1451 }
1452
1453 static void
1454 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1455 static void
1456 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1457
1458 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1459 struct ftrace_hash *new_hash);
1460
1461 static struct ftrace_hash *
1462 __ftrace_hash_move(struct ftrace_hash *src)
1463 {
1464 struct ftrace_func_entry *entry;
1465 struct hlist_node *tn;
1466 struct hlist_head *hhd;
1467 struct ftrace_hash *new_hash;
1468 int size = src->count;
1469 int bits = 0;
1470 int i;
1471
1472 /*
1473 * If the new source is empty, just return the empty_hash.
1474 */
1475 if (ftrace_hash_empty(src))
1476 return EMPTY_HASH;
1477
1478 /*
1479 * Make the hash size about 1/2 the # found
1480 */
1481 for (size /= 2; size; size >>= 1)
1482 bits++;
1483
1484 /* Don't allocate too much */
1485 if (bits > FTRACE_HASH_MAX_BITS)
1486 bits = FTRACE_HASH_MAX_BITS;
1487
1488 new_hash = alloc_ftrace_hash(bits);
1489 if (!new_hash)
1490 return NULL;
1491
1492 new_hash->flags = src->flags;
1493
1494 size = 1 << src->size_bits;
1495 for (i = 0; i < size; i++) {
1496 hhd = &src->buckets[i];
1497 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1498 remove_hash_entry(src, entry);
1499 __add_hash_entry(new_hash, entry);
1500 }
1501 }
1502
1503 return new_hash;
1504 }
1505
1506 static int
1507 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1508 struct ftrace_hash **dst, struct ftrace_hash *src)
1509 {
1510 struct ftrace_hash *new_hash;
1511 int ret;
1512
1513 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1514 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1515 return -EINVAL;
1516
1517 new_hash = __ftrace_hash_move(src);
1518 if (!new_hash)
1519 return -ENOMEM;
1520
1521 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1522 if (enable) {
1523 /* IPMODIFY should be updated only when filter_hash updating */
1524 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1525 if (ret < 0) {
1526 free_ftrace_hash(new_hash);
1527 return ret;
1528 }
1529 }
1530
1531 /*
1532 * Remove the current set, update the hash and add
1533 * them back.
1534 */
1535 ftrace_hash_rec_disable_modify(ops, enable);
1536
1537 rcu_assign_pointer(*dst, new_hash);
1538
1539 ftrace_hash_rec_enable_modify(ops, enable);
1540
1541 return 0;
1542 }
1543
1544 static bool hash_contains_ip(unsigned long ip,
1545 struct ftrace_ops_hash *hash)
1546 {
1547 /*
1548 * The function record is a match if it exists in the filter
1549 * hash and not in the notrace hash. Note, an emty hash is
1550 * considered a match for the filter hash, but an empty
1551 * notrace hash is considered not in the notrace hash.
1552 */
1553 return (ftrace_hash_empty(hash->filter_hash) ||
1554 __ftrace_lookup_ip(hash->filter_hash, ip)) &&
1555 (ftrace_hash_empty(hash->notrace_hash) ||
1556 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1557 }
1558
1559 /*
1560 * Test the hashes for this ops to see if we want to call
1561 * the ops->func or not.
1562 *
1563 * It's a match if the ip is in the ops->filter_hash or
1564 * the filter_hash does not exist or is empty,
1565 * AND
1566 * the ip is not in the ops->notrace_hash.
1567 *
1568 * This needs to be called with preemption disabled as
1569 * the hashes are freed with call_rcu_sched().
1570 */
1571 static int
1572 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1573 {
1574 struct ftrace_ops_hash hash;
1575 int ret;
1576
1577 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1578 /*
1579 * There's a small race when adding ops that the ftrace handler
1580 * that wants regs, may be called without them. We can not
1581 * allow that handler to be called if regs is NULL.
1582 */
1583 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1584 return 0;
1585 #endif
1586
1587 rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1588 rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1589
1590 if (hash_contains_ip(ip, &hash))
1591 ret = 1;
1592 else
1593 ret = 0;
1594
1595 return ret;
1596 }
1597
1598 /*
1599 * This is a double for. Do not use 'break' to break out of the loop,
1600 * you must use a goto.
1601 */
1602 #define do_for_each_ftrace_rec(pg, rec) \
1603 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1604 int _____i; \
1605 for (_____i = 0; _____i < pg->index; _____i++) { \
1606 rec = &pg->records[_____i];
1607
1608 #define while_for_each_ftrace_rec() \
1609 } \
1610 }
1611
1612
1613 static int ftrace_cmp_recs(const void *a, const void *b)
1614 {
1615 const struct dyn_ftrace *key = a;
1616 const struct dyn_ftrace *rec = b;
1617
1618 if (key->flags < rec->ip)
1619 return -1;
1620 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1621 return 1;
1622 return 0;
1623 }
1624
1625 /**
1626 * ftrace_location_range - return the first address of a traced location
1627 * if it touches the given ip range
1628 * @start: start of range to search.
1629 * @end: end of range to search (inclusive). @end points to the last byte
1630 * to check.
1631 *
1632 * Returns rec->ip if the related ftrace location is a least partly within
1633 * the given address range. That is, the first address of the instruction
1634 * that is either a NOP or call to the function tracer. It checks the ftrace
1635 * internal tables to determine if the address belongs or not.
1636 */
1637 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1638 {
1639 struct ftrace_page *pg;
1640 struct dyn_ftrace *rec;
1641 struct dyn_ftrace key;
1642
1643 key.ip = start;
1644 key.flags = end; /* overload flags, as it is unsigned long */
1645
1646 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1647 if (end < pg->records[0].ip ||
1648 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1649 continue;
1650 rec = bsearch(&key, pg->records, pg->index,
1651 sizeof(struct dyn_ftrace),
1652 ftrace_cmp_recs);
1653 if (rec)
1654 return rec->ip;
1655 }
1656
1657 return 0;
1658 }
1659
1660 /**
1661 * ftrace_location - return true if the ip giving is a traced location
1662 * @ip: the instruction pointer to check
1663 *
1664 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1665 * That is, the instruction that is either a NOP or call to
1666 * the function tracer. It checks the ftrace internal tables to
1667 * determine if the address belongs or not.
1668 */
1669 unsigned long ftrace_location(unsigned long ip)
1670 {
1671 return ftrace_location_range(ip, ip);
1672 }
1673
1674 /**
1675 * ftrace_text_reserved - return true if range contains an ftrace location
1676 * @start: start of range to search
1677 * @end: end of range to search (inclusive). @end points to the last byte to check.
1678 *
1679 * Returns 1 if @start and @end contains a ftrace location.
1680 * That is, the instruction that is either a NOP or call to
1681 * the function tracer. It checks the ftrace internal tables to
1682 * determine if the address belongs or not.
1683 */
1684 int ftrace_text_reserved(const void *start, const void *end)
1685 {
1686 unsigned long ret;
1687
1688 ret = ftrace_location_range((unsigned long)start,
1689 (unsigned long)end);
1690
1691 return (int)!!ret;
1692 }
1693
1694 /* Test if ops registered to this rec needs regs */
1695 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1696 {
1697 struct ftrace_ops *ops;
1698 bool keep_regs = false;
1699
1700 for (ops = ftrace_ops_list;
1701 ops != &ftrace_list_end; ops = ops->next) {
1702 /* pass rec in as regs to have non-NULL val */
1703 if (ftrace_ops_test(ops, rec->ip, rec)) {
1704 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1705 keep_regs = true;
1706 break;
1707 }
1708 }
1709 }
1710
1711 return keep_regs;
1712 }
1713
1714 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1715 int filter_hash,
1716 bool inc)
1717 {
1718 struct ftrace_hash *hash;
1719 struct ftrace_hash *other_hash;
1720 struct ftrace_page *pg;
1721 struct dyn_ftrace *rec;
1722 bool update = false;
1723 int count = 0;
1724 int all = false;
1725
1726 /* Only update if the ops has been registered */
1727 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1728 return false;
1729
1730 /*
1731 * In the filter_hash case:
1732 * If the count is zero, we update all records.
1733 * Otherwise we just update the items in the hash.
1734 *
1735 * In the notrace_hash case:
1736 * We enable the update in the hash.
1737 * As disabling notrace means enabling the tracing,
1738 * and enabling notrace means disabling, the inc variable
1739 * gets inversed.
1740 */
1741 if (filter_hash) {
1742 hash = ops->func_hash->filter_hash;
1743 other_hash = ops->func_hash->notrace_hash;
1744 if (ftrace_hash_empty(hash))
1745 all = true;
1746 } else {
1747 inc = !inc;
1748 hash = ops->func_hash->notrace_hash;
1749 other_hash = ops->func_hash->filter_hash;
1750 /*
1751 * If the notrace hash has no items,
1752 * then there's nothing to do.
1753 */
1754 if (ftrace_hash_empty(hash))
1755 return false;
1756 }
1757
1758 do_for_each_ftrace_rec(pg, rec) {
1759 int in_other_hash = 0;
1760 int in_hash = 0;
1761 int match = 0;
1762
1763 if (rec->flags & FTRACE_FL_DISABLED)
1764 continue;
1765
1766 if (all) {
1767 /*
1768 * Only the filter_hash affects all records.
1769 * Update if the record is not in the notrace hash.
1770 */
1771 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1772 match = 1;
1773 } else {
1774 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1775 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1776
1777 /*
1778 * If filter_hash is set, we want to match all functions
1779 * that are in the hash but not in the other hash.
1780 *
1781 * If filter_hash is not set, then we are decrementing.
1782 * That means we match anything that is in the hash
1783 * and also in the other_hash. That is, we need to turn
1784 * off functions in the other hash because they are disabled
1785 * by this hash.
1786 */
1787 if (filter_hash && in_hash && !in_other_hash)
1788 match = 1;
1789 else if (!filter_hash && in_hash &&
1790 (in_other_hash || ftrace_hash_empty(other_hash)))
1791 match = 1;
1792 }
1793 if (!match)
1794 continue;
1795
1796 if (inc) {
1797 rec->flags++;
1798 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1799 return false;
1800
1801 /*
1802 * If there's only a single callback registered to a
1803 * function, and the ops has a trampoline registered
1804 * for it, then we can call it directly.
1805 */
1806 if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1807 rec->flags |= FTRACE_FL_TRAMP;
1808 else
1809 /*
1810 * If we are adding another function callback
1811 * to this function, and the previous had a
1812 * custom trampoline in use, then we need to go
1813 * back to the default trampoline.
1814 */
1815 rec->flags &= ~FTRACE_FL_TRAMP;
1816
1817 /*
1818 * If any ops wants regs saved for this function
1819 * then all ops will get saved regs.
1820 */
1821 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1822 rec->flags |= FTRACE_FL_REGS;
1823 } else {
1824 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1825 return false;
1826 rec->flags--;
1827
1828 /*
1829 * If the rec had REGS enabled and the ops that is
1830 * being removed had REGS set, then see if there is
1831 * still any ops for this record that wants regs.
1832 * If not, we can stop recording them.
1833 */
1834 if (ftrace_rec_count(rec) > 0 &&
1835 rec->flags & FTRACE_FL_REGS &&
1836 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1837 if (!test_rec_ops_needs_regs(rec))
1838 rec->flags &= ~FTRACE_FL_REGS;
1839 }
1840
1841 /*
1842 * If the rec had TRAMP enabled, then it needs to
1843 * be cleared. As TRAMP can only be enabled iff
1844 * there is only a single ops attached to it.
1845 * In otherwords, always disable it on decrementing.
1846 * In the future, we may set it if rec count is
1847 * decremented to one, and the ops that is left
1848 * has a trampoline.
1849 */
1850 rec->flags &= ~FTRACE_FL_TRAMP;
1851
1852 /*
1853 * flags will be cleared in ftrace_check_record()
1854 * if rec count is zero.
1855 */
1856 }
1857 count++;
1858
1859 /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1860 update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1861
1862 /* Shortcut, if we handled all records, we are done. */
1863 if (!all && count == hash->count)
1864 return update;
1865 } while_for_each_ftrace_rec();
1866
1867 return update;
1868 }
1869
1870 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1871 int filter_hash)
1872 {
1873 return __ftrace_hash_rec_update(ops, filter_hash, 0);
1874 }
1875
1876 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1877 int filter_hash)
1878 {
1879 return __ftrace_hash_rec_update(ops, filter_hash, 1);
1880 }
1881
1882 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1883 int filter_hash, int inc)
1884 {
1885 struct ftrace_ops *op;
1886
1887 __ftrace_hash_rec_update(ops, filter_hash, inc);
1888
1889 if (ops->func_hash != &global_ops.local_hash)
1890 return;
1891
1892 /*
1893 * If the ops shares the global_ops hash, then we need to update
1894 * all ops that are enabled and use this hash.
1895 */
1896 do_for_each_ftrace_op(op, ftrace_ops_list) {
1897 /* Already done */
1898 if (op == ops)
1899 continue;
1900 if (op->func_hash == &global_ops.local_hash)
1901 __ftrace_hash_rec_update(op, filter_hash, inc);
1902 } while_for_each_ftrace_op(op);
1903 }
1904
1905 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1906 int filter_hash)
1907 {
1908 ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1909 }
1910
1911 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1912 int filter_hash)
1913 {
1914 ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1915 }
1916
1917 /*
1918 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1919 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1920 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1921 * Note that old_hash and new_hash has below meanings
1922 * - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1923 * - If the hash is EMPTY_HASH, it hits nothing
1924 * - Anything else hits the recs which match the hash entries.
1925 */
1926 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1927 struct ftrace_hash *old_hash,
1928 struct ftrace_hash *new_hash)
1929 {
1930 struct ftrace_page *pg;
1931 struct dyn_ftrace *rec, *end = NULL;
1932 int in_old, in_new;
1933
1934 /* Only update if the ops has been registered */
1935 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1936 return 0;
1937
1938 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1939 return 0;
1940
1941 /*
1942 * Since the IPMODIFY is a very address sensitive action, we do not
1943 * allow ftrace_ops to set all functions to new hash.
1944 */
1945 if (!new_hash || !old_hash)
1946 return -EINVAL;
1947
1948 /* Update rec->flags */
1949 do_for_each_ftrace_rec(pg, rec) {
1950
1951 if (rec->flags & FTRACE_FL_DISABLED)
1952 continue;
1953
1954 /* We need to update only differences of filter_hash */
1955 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1956 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1957 if (in_old == in_new)
1958 continue;
1959
1960 if (in_new) {
1961 /* New entries must ensure no others are using it */
1962 if (rec->flags & FTRACE_FL_IPMODIFY)
1963 goto rollback;
1964 rec->flags |= FTRACE_FL_IPMODIFY;
1965 } else /* Removed entry */
1966 rec->flags &= ~FTRACE_FL_IPMODIFY;
1967 } while_for_each_ftrace_rec();
1968
1969 return 0;
1970
1971 rollback:
1972 end = rec;
1973
1974 /* Roll back what we did above */
1975 do_for_each_ftrace_rec(pg, rec) {
1976
1977 if (rec->flags & FTRACE_FL_DISABLED)
1978 continue;
1979
1980 if (rec == end)
1981 goto err_out;
1982
1983 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1984 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1985 if (in_old == in_new)
1986 continue;
1987
1988 if (in_new)
1989 rec->flags &= ~FTRACE_FL_IPMODIFY;
1990 else
1991 rec->flags |= FTRACE_FL_IPMODIFY;
1992 } while_for_each_ftrace_rec();
1993
1994 err_out:
1995 return -EBUSY;
1996 }
1997
1998 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1999 {
2000 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2001
2002 if (ftrace_hash_empty(hash))
2003 hash = NULL;
2004
2005 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
2006 }
2007
2008 /* Disabling always succeeds */
2009 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
2010 {
2011 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2012
2013 if (ftrace_hash_empty(hash))
2014 hash = NULL;
2015
2016 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
2017 }
2018
2019 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
2020 struct ftrace_hash *new_hash)
2021 {
2022 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
2023
2024 if (ftrace_hash_empty(old_hash))
2025 old_hash = NULL;
2026
2027 if (ftrace_hash_empty(new_hash))
2028 new_hash = NULL;
2029
2030 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
2031 }
2032
2033 static void print_ip_ins(const char *fmt, const unsigned char *p)
2034 {
2035 int i;
2036
2037 printk(KERN_CONT "%s", fmt);
2038
2039 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
2040 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
2041 }
2042
2043 static struct ftrace_ops *
2044 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
2045 static struct ftrace_ops *
2046 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
2047
2048 enum ftrace_bug_type ftrace_bug_type;
2049 const void *ftrace_expected;
2050
2051 static void print_bug_type(void)
2052 {
2053 switch (ftrace_bug_type) {
2054 case FTRACE_BUG_UNKNOWN:
2055 break;
2056 case FTRACE_BUG_INIT:
2057 pr_info("Initializing ftrace call sites\n");
2058 break;
2059 case FTRACE_BUG_NOP:
2060 pr_info("Setting ftrace call site to NOP\n");
2061 break;
2062 case FTRACE_BUG_CALL:
2063 pr_info("Setting ftrace call site to call ftrace function\n");
2064 break;
2065 case FTRACE_BUG_UPDATE:
2066 pr_info("Updating ftrace call site to call a different ftrace function\n");
2067 break;
2068 }
2069 }
2070
2071 /**
2072 * ftrace_bug - report and shutdown function tracer
2073 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2074 * @rec: The record that failed
2075 *
2076 * The arch code that enables or disables the function tracing
2077 * can call ftrace_bug() when it has detected a problem in
2078 * modifying the code. @failed should be one of either:
2079 * EFAULT - if the problem happens on reading the @ip address
2080 * EINVAL - if what is read at @ip is not what was expected
2081 * EPERM - if the problem happens on writting to the @ip address
2082 */
2083 void ftrace_bug(int failed, struct dyn_ftrace *rec)
2084 {
2085 unsigned long ip = rec ? rec->ip : 0;
2086
2087 switch (failed) {
2088 case -EFAULT:
2089 FTRACE_WARN_ON_ONCE(1);
2090 pr_info("ftrace faulted on modifying ");
2091 print_ip_sym(ip);
2092 break;
2093 case -EINVAL:
2094 FTRACE_WARN_ON_ONCE(1);
2095 pr_info("ftrace failed to modify ");
2096 print_ip_sym(ip);
2097 print_ip_ins(" actual: ", (unsigned char *)ip);
2098 pr_cont("\n");
2099 if (ftrace_expected) {
2100 print_ip_ins(" expected: ", ftrace_expected);
2101 pr_cont("\n");
2102 }
2103 break;
2104 case -EPERM:
2105 FTRACE_WARN_ON_ONCE(1);
2106 pr_info("ftrace faulted on writing ");
2107 print_ip_sym(ip);
2108 break;
2109 default:
2110 FTRACE_WARN_ON_ONCE(1);
2111 pr_info("ftrace faulted on unknown error ");
2112 print_ip_sym(ip);
2113 }
2114 print_bug_type();
2115 if (rec) {
2116 struct ftrace_ops *ops = NULL;
2117
2118 pr_info("ftrace record flags: %lx\n", rec->flags);
2119 pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2120 rec->flags & FTRACE_FL_REGS ? " R" : " ");
2121 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2122 ops = ftrace_find_tramp_ops_any(rec);
2123 if (ops) {
2124 do {
2125 pr_cont("\ttramp: %pS (%pS)",
2126 (void *)ops->trampoline,
2127 (void *)ops->func);
2128 ops = ftrace_find_tramp_ops_next(rec, ops);
2129 } while (ops);
2130 } else
2131 pr_cont("\ttramp: ERROR!");
2132
2133 }
2134 ip = ftrace_get_addr_curr(rec);
2135 pr_cont("\n expected tramp: %lx\n", ip);
2136 }
2137 }
2138
2139 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2140 {
2141 unsigned long flag = 0UL;
2142
2143 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2144
2145 if (rec->flags & FTRACE_FL_DISABLED)
2146 return FTRACE_UPDATE_IGNORE;
2147
2148 /*
2149 * If we are updating calls:
2150 *
2151 * If the record has a ref count, then we need to enable it
2152 * because someone is using it.
2153 *
2154 * Otherwise we make sure its disabled.
2155 *
2156 * If we are disabling calls, then disable all records that
2157 * are enabled.
2158 */
2159 if (enable && ftrace_rec_count(rec))
2160 flag = FTRACE_FL_ENABLED;
2161
2162 /*
2163 * If enabling and the REGS flag does not match the REGS_EN, or
2164 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2165 * this record. Set flags to fail the compare against ENABLED.
2166 */
2167 if (flag) {
2168 if (!(rec->flags & FTRACE_FL_REGS) !=
2169 !(rec->flags & FTRACE_FL_REGS_EN))
2170 flag |= FTRACE_FL_REGS;
2171
2172 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2173 !(rec->flags & FTRACE_FL_TRAMP_EN))
2174 flag |= FTRACE_FL_TRAMP;
2175 }
2176
2177 /* If the state of this record hasn't changed, then do nothing */
2178 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2179 return FTRACE_UPDATE_IGNORE;
2180
2181 if (flag) {
2182 /* Save off if rec is being enabled (for return value) */
2183 flag ^= rec->flags & FTRACE_FL_ENABLED;
2184
2185 if (update) {
2186 rec->flags |= FTRACE_FL_ENABLED;
2187 if (flag & FTRACE_FL_REGS) {
2188 if (rec->flags & FTRACE_FL_REGS)
2189 rec->flags |= FTRACE_FL_REGS_EN;
2190 else
2191 rec->flags &= ~FTRACE_FL_REGS_EN;
2192 }
2193 if (flag & FTRACE_FL_TRAMP) {
2194 if (rec->flags & FTRACE_FL_TRAMP)
2195 rec->flags |= FTRACE_FL_TRAMP_EN;
2196 else
2197 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2198 }
2199 }
2200
2201 /*
2202 * If this record is being updated from a nop, then
2203 * return UPDATE_MAKE_CALL.
2204 * Otherwise,
2205 * return UPDATE_MODIFY_CALL to tell the caller to convert
2206 * from the save regs, to a non-save regs function or
2207 * vice versa, or from a trampoline call.
2208 */
2209 if (flag & FTRACE_FL_ENABLED) {
2210 ftrace_bug_type = FTRACE_BUG_CALL;
2211 return FTRACE_UPDATE_MAKE_CALL;
2212 }
2213
2214 ftrace_bug_type = FTRACE_BUG_UPDATE;
2215 return FTRACE_UPDATE_MODIFY_CALL;
2216 }
2217
2218 if (update) {
2219 /* If there's no more users, clear all flags */
2220 if (!ftrace_rec_count(rec))
2221 rec->flags = 0;
2222 else
2223 /*
2224 * Just disable the record, but keep the ops TRAMP
2225 * and REGS states. The _EN flags must be disabled though.
2226 */
2227 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2228 FTRACE_FL_REGS_EN);
2229 }
2230
2231 ftrace_bug_type = FTRACE_BUG_NOP;
2232 return FTRACE_UPDATE_MAKE_NOP;
2233 }
2234
2235 /**
2236 * ftrace_update_record, set a record that now is tracing or not
2237 * @rec: the record to update
2238 * @enable: set to 1 if the record is tracing, zero to force disable
2239 *
2240 * The records that represent all functions that can be traced need
2241 * to be updated when tracing has been enabled.
2242 */
2243 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2244 {
2245 return ftrace_check_record(rec, enable, 1);
2246 }
2247
2248 /**
2249 * ftrace_test_record, check if the record has been enabled or not
2250 * @rec: the record to test
2251 * @enable: set to 1 to check if enabled, 0 if it is disabled
2252 *
2253 * The arch code may need to test if a record is already set to
2254 * tracing to determine how to modify the function code that it
2255 * represents.
2256 */
2257 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2258 {
2259 return ftrace_check_record(rec, enable, 0);
2260 }
2261
2262 static struct ftrace_ops *
2263 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2264 {
2265 struct ftrace_ops *op;
2266 unsigned long ip = rec->ip;
2267
2268 do_for_each_ftrace_op(op, ftrace_ops_list) {
2269
2270 if (!op->trampoline)
2271 continue;
2272
2273 if (hash_contains_ip(ip, op->func_hash))
2274 return op;
2275 } while_for_each_ftrace_op(op);
2276
2277 return NULL;
2278 }
2279
2280 static struct ftrace_ops *
2281 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2282 struct ftrace_ops *op)
2283 {
2284 unsigned long ip = rec->ip;
2285
2286 while_for_each_ftrace_op(op) {
2287
2288 if (!op->trampoline)
2289 continue;
2290
2291 if (hash_contains_ip(ip, op->func_hash))
2292 return op;
2293 }
2294
2295 return NULL;
2296 }
2297
2298 static struct ftrace_ops *
2299 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2300 {
2301 struct ftrace_ops *op;
2302 unsigned long ip = rec->ip;
2303
2304 /*
2305 * Need to check removed ops first.
2306 * If they are being removed, and this rec has a tramp,
2307 * and this rec is in the ops list, then it would be the
2308 * one with the tramp.
2309 */
2310 if (removed_ops) {
2311 if (hash_contains_ip(ip, &removed_ops->old_hash))
2312 return removed_ops;
2313 }
2314
2315 /*
2316 * Need to find the current trampoline for a rec.
2317 * Now, a trampoline is only attached to a rec if there
2318 * was a single 'ops' attached to it. But this can be called
2319 * when we are adding another op to the rec or removing the
2320 * current one. Thus, if the op is being added, we can
2321 * ignore it because it hasn't attached itself to the rec
2322 * yet.
2323 *
2324 * If an ops is being modified (hooking to different functions)
2325 * then we don't care about the new functions that are being
2326 * added, just the old ones (that are probably being removed).
2327 *
2328 * If we are adding an ops to a function that already is using
2329 * a trampoline, it needs to be removed (trampolines are only
2330 * for single ops connected), then an ops that is not being
2331 * modified also needs to be checked.
2332 */
2333 do_for_each_ftrace_op(op, ftrace_ops_list) {
2334
2335 if (!op->trampoline)
2336 continue;
2337
2338 /*
2339 * If the ops is being added, it hasn't gotten to
2340 * the point to be removed from this tree yet.
2341 */
2342 if (op->flags & FTRACE_OPS_FL_ADDING)
2343 continue;
2344
2345
2346 /*
2347 * If the ops is being modified and is in the old
2348 * hash, then it is probably being removed from this
2349 * function.
2350 */
2351 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2352 hash_contains_ip(ip, &op->old_hash))
2353 return op;
2354 /*
2355 * If the ops is not being added or modified, and it's
2356 * in its normal filter hash, then this must be the one
2357 * we want!
2358 */
2359 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2360 hash_contains_ip(ip, op->func_hash))
2361 return op;
2362
2363 } while_for_each_ftrace_op(op);
2364
2365 return NULL;
2366 }
2367
2368 static struct ftrace_ops *
2369 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2370 {
2371 struct ftrace_ops *op;
2372 unsigned long ip = rec->ip;
2373
2374 do_for_each_ftrace_op(op, ftrace_ops_list) {
2375 /* pass rec in as regs to have non-NULL val */
2376 if (hash_contains_ip(ip, op->func_hash))
2377 return op;
2378 } while_for_each_ftrace_op(op);
2379
2380 return NULL;
2381 }
2382
2383 /**
2384 * ftrace_get_addr_new - Get the call address to set to
2385 * @rec: The ftrace record descriptor
2386 *
2387 * If the record has the FTRACE_FL_REGS set, that means that it
2388 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2389 * is not not set, then it wants to convert to the normal callback.
2390 *
2391 * Returns the address of the trampoline to set to
2392 */
2393 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2394 {
2395 struct ftrace_ops *ops;
2396
2397 /* Trampolines take precedence over regs */
2398 if (rec->flags & FTRACE_FL_TRAMP) {
2399 ops = ftrace_find_tramp_ops_new(rec);
2400 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2401 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2402 (void *)rec->ip, (void *)rec->ip, rec->flags);
2403 /* Ftrace is shutting down, return anything */
2404 return (unsigned long)FTRACE_ADDR;
2405 }
2406 return ops->trampoline;
2407 }
2408
2409 if (rec->flags & FTRACE_FL_REGS)
2410 return (unsigned long)FTRACE_REGS_ADDR;
2411 else
2412 return (unsigned long)FTRACE_ADDR;
2413 }
2414
2415 /**
2416 * ftrace_get_addr_curr - Get the call address that is already there
2417 * @rec: The ftrace record descriptor
2418 *
2419 * The FTRACE_FL_REGS_EN is set when the record already points to
2420 * a function that saves all the regs. Basically the '_EN' version
2421 * represents the current state of the function.
2422 *
2423 * Returns the address of the trampoline that is currently being called
2424 */
2425 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2426 {
2427 struct ftrace_ops *ops;
2428
2429 /* Trampolines take precedence over regs */
2430 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2431 ops = ftrace_find_tramp_ops_curr(rec);
2432 if (FTRACE_WARN_ON(!ops)) {
2433 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2434 (void *)rec->ip, (void *)rec->ip);
2435 /* Ftrace is shutting down, return anything */
2436 return (unsigned long)FTRACE_ADDR;
2437 }
2438 return ops->trampoline;
2439 }
2440
2441 if (rec->flags & FTRACE_FL_REGS_EN)
2442 return (unsigned long)FTRACE_REGS_ADDR;
2443 else
2444 return (unsigned long)FTRACE_ADDR;
2445 }
2446
2447 static int
2448 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2449 {
2450 unsigned long ftrace_old_addr;
2451 unsigned long ftrace_addr;
2452 int ret;
2453
2454 ftrace_addr = ftrace_get_addr_new(rec);
2455
2456 /* This needs to be done before we call ftrace_update_record */
2457 ftrace_old_addr = ftrace_get_addr_curr(rec);
2458
2459 ret = ftrace_update_record(rec, enable);
2460
2461 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2462
2463 switch (ret) {
2464 case FTRACE_UPDATE_IGNORE:
2465 return 0;
2466
2467 case FTRACE_UPDATE_MAKE_CALL:
2468 ftrace_bug_type = FTRACE_BUG_CALL;
2469 return ftrace_make_call(rec, ftrace_addr);
2470
2471 case FTRACE_UPDATE_MAKE_NOP:
2472 ftrace_bug_type = FTRACE_BUG_NOP;
2473 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2474
2475 case FTRACE_UPDATE_MODIFY_CALL:
2476 ftrace_bug_type = FTRACE_BUG_UPDATE;
2477 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2478 }
2479
2480 return -1; /* unknow ftrace bug */
2481 }
2482
2483 void __weak ftrace_replace_code(int enable)
2484 {
2485 struct dyn_ftrace *rec;
2486 struct ftrace_page *pg;
2487 int failed;
2488
2489 if (unlikely(ftrace_disabled))
2490 return;
2491
2492 do_for_each_ftrace_rec(pg, rec) {
2493
2494 if (rec->flags & FTRACE_FL_DISABLED)
2495 continue;
2496
2497 failed = __ftrace_replace_code(rec, enable);
2498 if (failed) {
2499 ftrace_bug(failed, rec);
2500 /* Stop processing */
2501 return;
2502 }
2503 } while_for_each_ftrace_rec();
2504 }
2505
2506 struct ftrace_rec_iter {
2507 struct ftrace_page *pg;
2508 int index;
2509 };
2510
2511 /**
2512 * ftrace_rec_iter_start, start up iterating over traced functions
2513 *
2514 * Returns an iterator handle that is used to iterate over all
2515 * the records that represent address locations where functions
2516 * are traced.
2517 *
2518 * May return NULL if no records are available.
2519 */
2520 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2521 {
2522 /*
2523 * We only use a single iterator.
2524 * Protected by the ftrace_lock mutex.
2525 */
2526 static struct ftrace_rec_iter ftrace_rec_iter;
2527 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2528
2529 iter->pg = ftrace_pages_start;
2530 iter->index = 0;
2531
2532 /* Could have empty pages */
2533 while (iter->pg && !iter->pg->index)
2534 iter->pg = iter->pg->next;
2535
2536 if (!iter->pg)
2537 return NULL;
2538
2539 return iter;
2540 }
2541
2542 /**
2543 * ftrace_rec_iter_next, get the next record to process.
2544 * @iter: The handle to the iterator.
2545 *
2546 * Returns the next iterator after the given iterator @iter.
2547 */
2548 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2549 {
2550 iter->index++;
2551
2552 if (iter->index >= iter->pg->index) {
2553 iter->pg = iter->pg->next;
2554 iter->index = 0;
2555
2556 /* Could have empty pages */
2557 while (iter->pg && !iter->pg->index)
2558 iter->pg = iter->pg->next;
2559 }
2560
2561 if (!iter->pg)
2562 return NULL;
2563
2564 return iter;
2565 }
2566
2567 /**
2568 * ftrace_rec_iter_record, get the record at the iterator location
2569 * @iter: The current iterator location
2570 *
2571 * Returns the record that the current @iter is at.
2572 */
2573 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2574 {
2575 return &iter->pg->records[iter->index];
2576 }
2577
2578 static int
2579 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2580 {
2581 int ret;
2582
2583 if (unlikely(ftrace_disabled))
2584 return 0;
2585
2586 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2587 if (ret) {
2588 ftrace_bug_type = FTRACE_BUG_INIT;
2589 ftrace_bug(ret, rec);
2590 return 0;
2591 }
2592 return 1;
2593 }
2594
2595 /*
2596 * archs can override this function if they must do something
2597 * before the modifying code is performed.
2598 */
2599 int __weak ftrace_arch_code_modify_prepare(void)
2600 {
2601 return 0;
2602 }
2603
2604 /*
2605 * archs can override this function if they must do something
2606 * after the modifying code is performed.
2607 */
2608 int __weak ftrace_arch_code_modify_post_process(void)
2609 {
2610 return 0;
2611 }
2612
2613 void ftrace_modify_all_code(int command)
2614 {
2615 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2616 int err = 0;
2617
2618 /*
2619 * If the ftrace_caller calls a ftrace_ops func directly,
2620 * we need to make sure that it only traces functions it
2621 * expects to trace. When doing the switch of functions,
2622 * we need to update to the ftrace_ops_list_func first
2623 * before the transition between old and new calls are set,
2624 * as the ftrace_ops_list_func will check the ops hashes
2625 * to make sure the ops are having the right functions
2626 * traced.
2627 */
2628 if (update) {
2629 err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2630 if (FTRACE_WARN_ON(err))
2631 return;
2632 }
2633
2634 if (command & FTRACE_UPDATE_CALLS)
2635 ftrace_replace_code(1);
2636 else if (command & FTRACE_DISABLE_CALLS)
2637 ftrace_replace_code(0);
2638
2639 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2640 function_trace_op = set_function_trace_op;
2641 smp_wmb();
2642 /* If irqs are disabled, we are in stop machine */
2643 if (!irqs_disabled())
2644 smp_call_function(ftrace_sync_ipi, NULL, 1);
2645 err = ftrace_update_ftrace_func(ftrace_trace_function);
2646 if (FTRACE_WARN_ON(err))
2647 return;
2648 }
2649
2650 if (command & FTRACE_START_FUNC_RET)
2651 err = ftrace_enable_ftrace_graph_caller();
2652 else if (command & FTRACE_STOP_FUNC_RET)
2653 err = ftrace_disable_ftrace_graph_caller();
2654 FTRACE_WARN_ON(err);
2655 }
2656
2657 static int __ftrace_modify_code(void *data)
2658 {
2659 int *command = data;
2660
2661 ftrace_modify_all_code(*command);
2662
2663 return 0;
2664 }
2665
2666 /**
2667 * ftrace_run_stop_machine, go back to the stop machine method
2668 * @command: The command to tell ftrace what to do
2669 *
2670 * If an arch needs to fall back to the stop machine method, the
2671 * it can call this function.
2672 */
2673 void ftrace_run_stop_machine(int command)
2674 {
2675 stop_machine(__ftrace_modify_code, &command, NULL);
2676 }
2677
2678 /**
2679 * arch_ftrace_update_code, modify the code to trace or not trace
2680 * @command: The command that needs to be done
2681 *
2682 * Archs can override this function if it does not need to
2683 * run stop_machine() to modify code.
2684 */
2685 void __weak arch_ftrace_update_code(int command)
2686 {
2687 ftrace_run_stop_machine(command);
2688 }
2689
2690 static void ftrace_run_update_code(int command)
2691 {
2692 int ret;
2693
2694 ret = ftrace_arch_code_modify_prepare();
2695 FTRACE_WARN_ON(ret);
2696 if (ret)
2697 return;
2698
2699 /*
2700 * By default we use stop_machine() to modify the code.
2701 * But archs can do what ever they want as long as it
2702 * is safe. The stop_machine() is the safest, but also
2703 * produces the most overhead.
2704 */
2705 arch_ftrace_update_code(command);
2706
2707 ret = ftrace_arch_code_modify_post_process();
2708 FTRACE_WARN_ON(ret);
2709 }
2710
2711 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2712 struct ftrace_ops_hash *old_hash)
2713 {
2714 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2715 ops->old_hash.filter_hash = old_hash->filter_hash;
2716 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2717 ftrace_run_update_code(command);
2718 ops->old_hash.filter_hash = NULL;
2719 ops->old_hash.notrace_hash = NULL;
2720 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2721 }
2722
2723 static ftrace_func_t saved_ftrace_func;
2724 static int ftrace_start_up;
2725
2726 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2727 {
2728 }
2729
2730 static void per_cpu_ops_free(struct ftrace_ops *ops)
2731 {
2732 free_percpu(ops->disabled);
2733 }
2734
2735 static void ftrace_startup_enable(int command)
2736 {
2737 if (saved_ftrace_func != ftrace_trace_function) {
2738 saved_ftrace_func = ftrace_trace_function;
2739 command |= FTRACE_UPDATE_TRACE_FUNC;
2740 }
2741
2742 if (!command || !ftrace_enabled)
2743 return;
2744
2745 ftrace_run_update_code(command);
2746 }
2747
2748 static void ftrace_startup_all(int command)
2749 {
2750 update_all_ops = true;
2751 ftrace_startup_enable(command);
2752 update_all_ops = false;
2753 }
2754
2755 static int ftrace_startup(struct ftrace_ops *ops, int command)
2756 {
2757 int ret;
2758
2759 if (unlikely(ftrace_disabled))
2760 return -ENODEV;
2761
2762 ret = __register_ftrace_function(ops);
2763 if (ret)
2764 return ret;
2765
2766 ftrace_start_up++;
2767
2768 /*
2769 * Note that ftrace probes uses this to start up
2770 * and modify functions it will probe. But we still
2771 * set the ADDING flag for modification, as probes
2772 * do not have trampolines. If they add them in the
2773 * future, then the probes will need to distinguish
2774 * between adding and updating probes.
2775 */
2776 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2777
2778 ret = ftrace_hash_ipmodify_enable(ops);
2779 if (ret < 0) {
2780 /* Rollback registration process */
2781 __unregister_ftrace_function(ops);
2782 ftrace_start_up--;
2783 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2784 return ret;
2785 }
2786
2787 if (ftrace_hash_rec_enable(ops, 1))
2788 command |= FTRACE_UPDATE_CALLS;
2789
2790 ftrace_startup_enable(command);
2791
2792 ops->flags &= ~FTRACE_OPS_FL_ADDING;
2793
2794 return 0;
2795 }
2796
2797 static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2798 {
2799 int ret;
2800
2801 if (unlikely(ftrace_disabled))
2802 return -ENODEV;
2803
2804 ret = __unregister_ftrace_function(ops);
2805 if (ret)
2806 return ret;
2807
2808 ftrace_start_up--;
2809 /*
2810 * Just warn in case of unbalance, no need to kill ftrace, it's not
2811 * critical but the ftrace_call callers may be never nopped again after
2812 * further ftrace uses.
2813 */
2814 WARN_ON_ONCE(ftrace_start_up < 0);
2815
2816 /* Disabling ipmodify never fails */
2817 ftrace_hash_ipmodify_disable(ops);
2818
2819 if (ftrace_hash_rec_disable(ops, 1))
2820 command |= FTRACE_UPDATE_CALLS;
2821
2822 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2823
2824 if (saved_ftrace_func != ftrace_trace_function) {
2825 saved_ftrace_func = ftrace_trace_function;
2826 command |= FTRACE_UPDATE_TRACE_FUNC;
2827 }
2828
2829 if (!command || !ftrace_enabled) {
2830 /*
2831 * If these are per_cpu ops, they still need their
2832 * per_cpu field freed. Since, function tracing is
2833 * not currently active, we can just free them
2834 * without synchronizing all CPUs.
2835 */
2836 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2837 per_cpu_ops_free(ops);
2838 return 0;
2839 }
2840
2841 /*
2842 * If the ops uses a trampoline, then it needs to be
2843 * tested first on update.
2844 */
2845 ops->flags |= FTRACE_OPS_FL_REMOVING;
2846 removed_ops = ops;
2847
2848 /* The trampoline logic checks the old hashes */
2849 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2850 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2851
2852 ftrace_run_update_code(command);
2853
2854 /*
2855 * If there's no more ops registered with ftrace, run a
2856 * sanity check to make sure all rec flags are cleared.
2857 */
2858 if (rcu_dereference_protected(ftrace_ops_list,
2859 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
2860 struct ftrace_page *pg;
2861 struct dyn_ftrace *rec;
2862
2863 do_for_each_ftrace_rec(pg, rec) {
2864 if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_FL_DISABLED))
2865 pr_warn(" %pS flags:%lx\n",
2866 (void *)rec->ip, rec->flags);
2867 } while_for_each_ftrace_rec();
2868 }
2869
2870 ops->old_hash.filter_hash = NULL;
2871 ops->old_hash.notrace_hash = NULL;
2872
2873 removed_ops = NULL;
2874 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2875
2876 /*
2877 * Dynamic ops may be freed, we must make sure that all
2878 * callers are done before leaving this function.
2879 * The same goes for freeing the per_cpu data of the per_cpu
2880 * ops.
2881 */
2882 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) {
2883 /*
2884 * We need to do a hard force of sched synchronization.
2885 * This is because we use preempt_disable() to do RCU, but
2886 * the function tracers can be called where RCU is not watching
2887 * (like before user_exit()). We can not rely on the RCU
2888 * infrastructure to do the synchronization, thus we must do it
2889 * ourselves.
2890 */
2891 schedule_on_each_cpu(ftrace_sync);
2892
2893 /*
2894 * When the kernel is preeptive, tasks can be preempted
2895 * while on a ftrace trampoline. Just scheduling a task on
2896 * a CPU is not good enough to flush them. Calling
2897 * synchornize_rcu_tasks() will wait for those tasks to
2898 * execute and either schedule voluntarily or enter user space.
2899 */
2900 if (IS_ENABLED(CONFIG_PREEMPT))
2901 synchronize_rcu_tasks();
2902
2903 arch_ftrace_trampoline_free(ops);
2904
2905 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2906 per_cpu_ops_free(ops);
2907 }
2908
2909 return 0;
2910 }
2911
2912 static void ftrace_startup_sysctl(void)
2913 {
2914 int command;
2915
2916 if (unlikely(ftrace_disabled))
2917 return;
2918
2919 /* Force update next time */
2920 saved_ftrace_func = NULL;
2921 /* ftrace_start_up is true if we want ftrace running */
2922 if (ftrace_start_up) {
2923 command = FTRACE_UPDATE_CALLS;
2924 if (ftrace_graph_active)
2925 command |= FTRACE_START_FUNC_RET;
2926 ftrace_startup_enable(command);
2927 }
2928 }
2929
2930 static void ftrace_shutdown_sysctl(void)
2931 {
2932 int command;
2933
2934 if (unlikely(ftrace_disabled))
2935 return;
2936
2937 /* ftrace_start_up is true if ftrace is running */
2938 if (ftrace_start_up) {
2939 command = FTRACE_DISABLE_CALLS;
2940 if (ftrace_graph_active)
2941 command |= FTRACE_STOP_FUNC_RET;
2942 ftrace_run_update_code(command);
2943 }
2944 }
2945
2946 static u64 ftrace_update_time;
2947 unsigned long ftrace_update_tot_cnt;
2948
2949 static inline int ops_traces_mod(struct ftrace_ops *ops)
2950 {
2951 /*
2952 * Filter_hash being empty will default to trace module.
2953 * But notrace hash requires a test of individual module functions.
2954 */
2955 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2956 ftrace_hash_empty(ops->func_hash->notrace_hash);
2957 }
2958
2959 /*
2960 * Check if the current ops references the record.
2961 *
2962 * If the ops traces all functions, then it was already accounted for.
2963 * If the ops does not trace the current record function, skip it.
2964 * If the ops ignores the function via notrace filter, skip it.
2965 */
2966 static inline bool
2967 ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2968 {
2969 /* If ops isn't enabled, ignore it */
2970 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2971 return 0;
2972
2973 /* If ops traces all then it includes this function */
2974 if (ops_traces_mod(ops))
2975 return 1;
2976
2977 /* The function must be in the filter */
2978 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2979 !__ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2980 return 0;
2981
2982 /* If in notrace hash, we ignore it too */
2983 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2984 return 0;
2985
2986 return 1;
2987 }
2988
2989 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2990 {
2991 struct ftrace_page *pg;
2992 struct dyn_ftrace *p;
2993 u64 start, stop;
2994 unsigned long update_cnt = 0;
2995 unsigned long rec_flags = 0;
2996 int i;
2997
2998 start = ftrace_now(raw_smp_processor_id());
2999
3000 /*
3001 * When a module is loaded, this function is called to convert
3002 * the calls to mcount in its text to nops, and also to create
3003 * an entry in the ftrace data. Now, if ftrace is activated
3004 * after this call, but before the module sets its text to
3005 * read-only, the modification of enabling ftrace can fail if
3006 * the read-only is done while ftrace is converting the calls.
3007 * To prevent this, the module's records are set as disabled
3008 * and will be enabled after the call to set the module's text
3009 * to read-only.
3010 */
3011 if (mod)
3012 rec_flags |= FTRACE_FL_DISABLED;
3013
3014 for (pg = new_pgs; pg; pg = pg->next) {
3015
3016 for (i = 0; i < pg->index; i++) {
3017
3018 /* If something went wrong, bail without enabling anything */
3019 if (unlikely(ftrace_disabled))
3020 return -1;
3021
3022 p = &pg->records[i];
3023 p->flags = rec_flags;
3024
3025 /*
3026 * Do the initial record conversion from mcount jump
3027 * to the NOP instructions.
3028 */
3029 if (!ftrace_code_disable(mod, p))
3030 break;
3031
3032 update_cnt++;
3033 }
3034 }
3035
3036 stop = ftrace_now(raw_smp_processor_id());
3037 ftrace_update_time = stop - start;
3038 ftrace_update_tot_cnt += update_cnt;
3039
3040 return 0;
3041 }
3042
3043 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3044 {
3045 int order;
3046 int cnt;
3047
3048 if (WARN_ON(!count))
3049 return -EINVAL;
3050
3051 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
3052
3053 /*
3054 * We want to fill as much as possible. No more than a page
3055 * may be empty.
3056 */
3057 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
3058 order--;
3059
3060 again:
3061 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3062
3063 if (!pg->records) {
3064 /* if we can't allocate this size, try something smaller */
3065 if (!order)
3066 return -ENOMEM;
3067 order >>= 1;
3068 goto again;
3069 }
3070
3071 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3072 pg->size = cnt;
3073
3074 if (cnt > count)
3075 cnt = count;
3076
3077 return cnt;
3078 }
3079
3080 static struct ftrace_page *
3081 ftrace_allocate_pages(unsigned long num_to_init)
3082 {
3083 struct ftrace_page *start_pg;
3084 struct ftrace_page *pg;
3085 int order;
3086 int cnt;
3087
3088 if (!num_to_init)
3089 return 0;
3090
3091 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3092 if (!pg)
3093 return NULL;
3094
3095 /*
3096 * Try to allocate as much as possible in one continues
3097 * location that fills in all of the space. We want to
3098 * waste as little space as possible.
3099 */
3100 for (;;) {
3101 cnt = ftrace_allocate_records(pg, num_to_init);
3102 if (cnt < 0)
3103 goto free_pages;
3104
3105 num_to_init -= cnt;
3106 if (!num_to_init)
3107 break;
3108
3109 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3110 if (!pg->next)
3111 goto free_pages;
3112
3113 pg = pg->next;
3114 }
3115
3116 return start_pg;
3117
3118 free_pages:
3119 pg = start_pg;
3120 while (pg) {
3121 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3122 free_pages((unsigned long)pg->records, order);
3123 start_pg = pg->next;
3124 kfree(pg);
3125 pg = start_pg;
3126 }
3127 pr_info("ftrace: FAILED to allocate memory for functions\n");
3128 return NULL;
3129 }
3130
3131 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3132
3133 struct ftrace_iterator {
3134 loff_t pos;
3135 loff_t func_pos;
3136 loff_t mod_pos;
3137 struct ftrace_page *pg;
3138 struct dyn_ftrace *func;
3139 struct ftrace_func_probe *probe;
3140 struct ftrace_func_entry *probe_entry;
3141 struct trace_parser parser;
3142 struct ftrace_hash *hash;
3143 struct ftrace_ops *ops;
3144 struct trace_array *tr;
3145 struct list_head *mod_list;
3146 int pidx;
3147 int idx;
3148 unsigned flags;
3149 };
3150
3151 static void *
3152 t_probe_next(struct seq_file *m, loff_t *pos)
3153 {
3154 struct ftrace_iterator *iter = m->private;
3155 struct trace_array *tr = iter->ops->private;
3156 struct list_head *func_probes;
3157 struct ftrace_hash *hash;
3158 struct list_head *next;
3159 struct hlist_node *hnd = NULL;
3160 struct hlist_head *hhd;
3161 int size;
3162
3163 (*pos)++;
3164 iter->pos = *pos;
3165
3166 if (!tr)
3167 return NULL;
3168
3169 func_probes = &tr->func_probes;
3170 if (list_empty(func_probes))
3171 return NULL;
3172
3173 if (!iter->probe) {
3174 next = func_probes->next;
3175 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3176 }
3177
3178 if (iter->probe_entry)
3179 hnd = &iter->probe_entry->hlist;
3180
3181 hash = iter->probe->ops.func_hash->filter_hash;
3182 size = 1 << hash->size_bits;
3183
3184 retry:
3185 if (iter->pidx >= size) {
3186 if (iter->probe->list.next == func_probes)
3187 return NULL;
3188 next = iter->probe->list.next;
3189 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3190 hash = iter->probe->ops.func_hash->filter_hash;
3191 size = 1 << hash->size_bits;
3192 iter->pidx = 0;
3193 }
3194
3195 hhd = &hash->buckets[iter->pidx];
3196
3197 if (hlist_empty(hhd)) {
3198 iter->pidx++;
3199 hnd = NULL;
3200 goto retry;
3201 }
3202
3203 if (!hnd)
3204 hnd = hhd->first;
3205 else {
3206 hnd = hnd->next;
3207 if (!hnd) {
3208 iter->pidx++;
3209 goto retry;
3210 }
3211 }
3212
3213 if (WARN_ON_ONCE(!hnd))
3214 return NULL;
3215
3216 iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3217
3218 return iter;
3219 }
3220
3221 static void *t_probe_start(struct seq_file *m, loff_t *pos)
3222 {
3223 struct ftrace_iterator *iter = m->private;
3224 void *p = NULL;
3225 loff_t l;
3226
3227 if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3228 return NULL;
3229
3230 if (iter->mod_pos > *pos)
3231 return NULL;
3232
3233 iter->probe = NULL;
3234 iter->probe_entry = NULL;
3235 iter->pidx = 0;
3236 for (l = 0; l <= (*pos - iter->mod_pos); ) {
3237 p = t_probe_next(m, &l);
3238 if (!p)
3239 break;
3240 }
3241 if (!p)
3242 return NULL;
3243
3244 /* Only set this if we have an item */
3245 iter->flags |= FTRACE_ITER_PROBE;
3246
3247 return iter;
3248 }
3249
3250 static int
3251 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3252 {
3253 struct ftrace_func_entry *probe_entry;
3254 struct ftrace_probe_ops *probe_ops;
3255 struct ftrace_func_probe *probe;
3256
3257 probe = iter->probe;
3258 probe_entry = iter->probe_entry;
3259
3260 if (WARN_ON_ONCE(!probe || !probe_entry))
3261 return -EIO;
3262
3263 probe_ops = probe->probe_ops;
3264
3265 if (probe_ops->print)
3266 return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3267
3268 seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3269 (void *)probe_ops->func);
3270
3271 return 0;
3272 }
3273
3274 static void *
3275 t_mod_next(struct seq_file *m, loff_t *pos)
3276 {
3277 struct ftrace_iterator *iter = m->private;
3278 struct trace_array *tr = iter->tr;
3279
3280 (*pos)++;
3281 iter->pos = *pos;
3282
3283 iter->mod_list = iter->mod_list->next;
3284
3285 if (iter->mod_list == &tr->mod_trace ||
3286 iter->mod_list == &tr->mod_notrace) {
3287 iter->flags &= ~FTRACE_ITER_MOD;
3288 return NULL;
3289 }
3290
3291 iter->mod_pos = *pos;
3292
3293 return iter;
3294 }
3295
3296 static void *t_mod_start(struct seq_file *m, loff_t *pos)
3297 {
3298 struct ftrace_iterator *iter = m->private;
3299 void *p = NULL;
3300 loff_t l;
3301
3302 if (iter->func_pos > *pos)
3303 return NULL;
3304
3305 iter->mod_pos = iter->func_pos;
3306
3307 /* probes are only available if tr is set */
3308 if (!iter->tr)
3309 return NULL;
3310
3311 for (l = 0; l <= (*pos - iter->func_pos); ) {
3312 p = t_mod_next(m, &l);
3313 if (!p)
3314 break;
3315 }
3316 if (!p) {
3317 iter->flags &= ~FTRACE_ITER_MOD;
3318 return t_probe_start(m, pos);
3319 }
3320
3321 /* Only set this if we have an item */
3322 iter->flags |= FTRACE_ITER_MOD;
3323
3324 return iter;
3325 }
3326
3327 static int
3328 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
3329 {
3330 struct ftrace_mod_load *ftrace_mod;
3331 struct trace_array *tr = iter->tr;
3332
3333 if (WARN_ON_ONCE(!iter->mod_list) ||
3334 iter->mod_list == &tr->mod_trace ||
3335 iter->mod_list == &tr->mod_notrace)
3336 return -EIO;
3337
3338 ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
3339
3340 if (ftrace_mod->func)
3341 seq_printf(m, "%s", ftrace_mod->func);
3342 else
3343 seq_putc(m, '*');
3344
3345 seq_printf(m, ":mod:%s\n", ftrace_mod->module);
3346
3347 return 0;
3348 }
3349
3350 static void *
3351 t_func_next(struct seq_file *m, loff_t *pos)
3352 {
3353 struct ftrace_iterator *iter = m->private;
3354 struct dyn_ftrace *rec = NULL;
3355
3356 (*pos)++;
3357
3358 retry:
3359 if (iter->idx >= iter->pg->index) {
3360 if (iter->pg->next) {
3361 iter->pg = iter->pg->next;
3362 iter->idx = 0;
3363 goto retry;
3364 }
3365 } else {
3366 rec = &iter->pg->records[iter->idx++];
3367 if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3368 !ftrace_lookup_ip(iter->hash, rec->ip)) ||
3369
3370 ((iter->flags & FTRACE_ITER_ENABLED) &&
3371 !(rec->flags & FTRACE_FL_ENABLED))) {
3372
3373 rec = NULL;
3374 goto retry;
3375 }
3376 }
3377
3378 if (!rec)
3379 return NULL;
3380
3381 iter->pos = iter->func_pos = *pos;
3382 iter->func = rec;
3383
3384 return iter;
3385 }
3386
3387 static void *
3388 t_next(struct seq_file *m, void *v, loff_t *pos)
3389 {
3390 struct ftrace_iterator *iter = m->private;
3391 loff_t l = *pos; /* t_probe_start() must use original pos */
3392 void *ret;
3393
3394 if (unlikely(ftrace_disabled))
3395 return NULL;
3396
3397 if (iter->flags & FTRACE_ITER_PROBE)
3398 return t_probe_next(m, pos);
3399
3400 if (iter->flags & FTRACE_ITER_MOD)
3401 return t_mod_next(m, pos);
3402
3403 if (iter->flags & FTRACE_ITER_PRINTALL) {
3404 /* next must increment pos, and t_probe_start does not */
3405 (*pos)++;
3406 return t_mod_start(m, &l);
3407 }
3408
3409 ret = t_func_next(m, pos);
3410
3411 if (!ret)
3412 return t_mod_start(m, &l);
3413
3414 return ret;
3415 }
3416
3417 static void reset_iter_read(struct ftrace_iterator *iter)
3418 {
3419 iter->pos = 0;
3420 iter->func_pos = 0;
3421 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
3422 }
3423
3424 static void *t_start(struct seq_file *m, loff_t *pos)
3425 {
3426 struct ftrace_iterator *iter = m->private;
3427 void *p = NULL;
3428 loff_t l;
3429
3430 mutex_lock(&ftrace_lock);
3431
3432 if (unlikely(ftrace_disabled))
3433 return NULL;
3434
3435 /*
3436 * If an lseek was done, then reset and start from beginning.
3437 */
3438 if (*pos < iter->pos)
3439 reset_iter_read(iter);
3440
3441 /*
3442 * For set_ftrace_filter reading, if we have the filter
3443 * off, we can short cut and just print out that all
3444 * functions are enabled.
3445 */
3446 if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3447 ftrace_hash_empty(iter->hash)) {
3448 iter->func_pos = 1; /* Account for the message */
3449 if (*pos > 0)
3450 return t_mod_start(m, pos);
3451 iter->flags |= FTRACE_ITER_PRINTALL;
3452 /* reset in case of seek/pread */
3453 iter->flags &= ~FTRACE_ITER_PROBE;
3454 return iter;
3455 }
3456
3457 if (iter->flags & FTRACE_ITER_MOD)
3458 return t_mod_start(m, pos);
3459
3460 /*
3461 * Unfortunately, we need to restart at ftrace_pages_start
3462 * every time we let go of the ftrace_mutex. This is because
3463 * those pointers can change without the lock.
3464 */
3465 iter->pg = ftrace_pages_start;
3466 iter->idx = 0;
3467 for (l = 0; l <= *pos; ) {
3468 p = t_func_next(m, &l);
3469 if (!p)
3470 break;
3471 }
3472
3473 if (!p)
3474 return t_mod_start(m, pos);
3475
3476 return iter;
3477 }
3478
3479 static void t_stop(struct seq_file *m, void *p)
3480 {
3481 mutex_unlock(&ftrace_lock);
3482 }
3483
3484 void * __weak
3485 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3486 {
3487 return NULL;
3488 }
3489
3490 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3491 struct dyn_ftrace *rec)
3492 {
3493 void *ptr;
3494
3495 ptr = arch_ftrace_trampoline_func(ops, rec);
3496 if (ptr)
3497 seq_printf(m, " ->%pS", ptr);
3498 }
3499
3500 static int t_show(struct seq_file *m, void *v)
3501 {
3502 struct ftrace_iterator *iter = m->private;
3503 struct dyn_ftrace *rec;
3504
3505 if (iter->flags & FTRACE_ITER_PROBE)
3506 return t_probe_show(m, iter);
3507
3508 if (iter->flags & FTRACE_ITER_MOD)
3509 return t_mod_show(m, iter);
3510
3511 if (iter->flags & FTRACE_ITER_PRINTALL) {
3512 if (iter->flags & FTRACE_ITER_NOTRACE)
3513 seq_puts(m, "#### no functions disabled ####\n");
3514 else
3515 seq_puts(m, "#### all functions enabled ####\n");
3516 return 0;
3517 }
3518
3519 rec = iter->func;
3520
3521 if (!rec)
3522 return 0;
3523
3524 seq_printf(m, "%ps", (void *)rec->ip);
3525 if (iter->flags & FTRACE_ITER_ENABLED) {
3526 struct ftrace_ops *ops;
3527
3528 seq_printf(m, " (%ld)%s%s",
3529 ftrace_rec_count(rec),
3530 rec->flags & FTRACE_FL_REGS ? " R" : " ",
3531 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ");
3532 if (rec->flags & FTRACE_FL_TRAMP_EN) {
3533 ops = ftrace_find_tramp_ops_any(rec);
3534 if (ops) {
3535 do {
3536 seq_printf(m, "\ttramp: %pS (%pS)",
3537 (void *)ops->trampoline,
3538 (void *)ops->func);
3539 add_trampoline_func(m, ops, rec);
3540 ops = ftrace_find_tramp_ops_next(rec, ops);
3541 } while (ops);
3542 } else
3543 seq_puts(m, "\ttramp: ERROR!");
3544 } else {
3545 add_trampoline_func(m, NULL, rec);
3546 }
3547 }
3548
3549 seq_putc(m, '\n');
3550
3551 return 0;
3552 }
3553
3554 static const struct seq_operations show_ftrace_seq_ops = {
3555 .start = t_start,
3556 .next = t_next,
3557 .stop = t_stop,
3558 .show = t_show,
3559 };
3560
3561 static int
3562 ftrace_avail_open(struct inode *inode, struct file *file)
3563 {
3564 struct ftrace_iterator *iter;
3565
3566 if (unlikely(ftrace_disabled))
3567 return -ENODEV;
3568
3569 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3570 if (!iter)
3571 return -ENOMEM;
3572
3573 iter->pg = ftrace_pages_start;
3574 iter->ops = &global_ops;
3575
3576 return 0;
3577 }
3578
3579 static int
3580 ftrace_enabled_open(struct inode *inode, struct file *file)
3581 {
3582 struct ftrace_iterator *iter;
3583
3584 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3585 if (!iter)
3586 return -ENOMEM;
3587
3588 iter->pg = ftrace_pages_start;
3589 iter->flags = FTRACE_ITER_ENABLED;
3590 iter->ops = &global_ops;
3591
3592 return 0;
3593 }
3594
3595 /**
3596 * ftrace_regex_open - initialize function tracer filter files
3597 * @ops: The ftrace_ops that hold the hash filters
3598 * @flag: The type of filter to process
3599 * @inode: The inode, usually passed in to your open routine
3600 * @file: The file, usually passed in to your open routine
3601 *
3602 * ftrace_regex_open() initializes the filter files for the
3603 * @ops. Depending on @flag it may process the filter hash or
3604 * the notrace hash of @ops. With this called from the open
3605 * routine, you can use ftrace_filter_write() for the write
3606 * routine if @flag has FTRACE_ITER_FILTER set, or
3607 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3608 * tracing_lseek() should be used as the lseek routine, and
3609 * release must call ftrace_regex_release().
3610 */
3611 int
3612 ftrace_regex_open(struct ftrace_ops *ops, int flag,
3613 struct inode *inode, struct file *file)
3614 {
3615 struct ftrace_iterator *iter;
3616 struct ftrace_hash *hash;
3617 struct list_head *mod_head;
3618 struct trace_array *tr = ops->private;
3619 int ret = 0;
3620
3621 ftrace_ops_init(ops);
3622
3623 if (unlikely(ftrace_disabled))
3624 return -ENODEV;
3625
3626 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3627 if (!iter)
3628 return -ENOMEM;
3629
3630 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3631 kfree(iter);
3632 return -ENOMEM;
3633 }
3634
3635 iter->ops = ops;
3636 iter->flags = flag;
3637 iter->tr = tr;
3638
3639 mutex_lock(&ops->func_hash->regex_lock);
3640
3641 if (flag & FTRACE_ITER_NOTRACE) {
3642 hash = ops->func_hash->notrace_hash;
3643 mod_head = tr ? &tr->mod_notrace : NULL;
3644 } else {
3645 hash = ops->func_hash->filter_hash;
3646 mod_head = tr ? &tr->mod_trace : NULL;
3647 }
3648
3649 iter->mod_list = mod_head;
3650
3651 if (file->f_mode & FMODE_WRITE) {
3652 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3653
3654 if (file->f_flags & O_TRUNC) {
3655 iter->hash = alloc_ftrace_hash(size_bits);
3656 clear_ftrace_mod_list(mod_head);
3657 } else {
3658 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3659 }
3660
3661 if (!iter->hash) {
3662 trace_parser_put(&iter->parser);
3663 kfree(iter);
3664 ret = -ENOMEM;
3665 goto out_unlock;
3666 }
3667 } else
3668 iter->hash = hash;
3669
3670 if (file->f_mode & FMODE_READ) {
3671 iter->pg = ftrace_pages_start;
3672
3673 ret = seq_open(file, &show_ftrace_seq_ops);
3674 if (!ret) {
3675 struct seq_file *m = file->private_data;
3676 m->private = iter;
3677 } else {
3678 /* Failed */
3679 free_ftrace_hash(iter->hash);
3680 trace_parser_put(&iter->parser);
3681 kfree(iter);
3682 }
3683 } else
3684 file->private_data = iter;
3685
3686 out_unlock:
3687 mutex_unlock(&ops->func_hash->regex_lock);
3688
3689 return ret;
3690 }
3691
3692 static int
3693 ftrace_filter_open(struct inode *inode, struct file *file)
3694 {
3695 struct ftrace_ops *ops = inode->i_private;
3696
3697 return ftrace_regex_open(ops,
3698 FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
3699 inode, file);
3700 }
3701
3702 static int
3703 ftrace_notrace_open(struct inode *inode, struct file *file)
3704 {
3705 struct ftrace_ops *ops = inode->i_private;
3706
3707 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3708 inode, file);
3709 }
3710
3711 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3712 struct ftrace_glob {
3713 char *search;
3714 unsigned len;
3715 int type;
3716 };
3717
3718 /*
3719 * If symbols in an architecture don't correspond exactly to the user-visible
3720 * name of what they represent, it is possible to define this function to
3721 * perform the necessary adjustments.
3722 */
3723 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3724 {
3725 return str;
3726 }
3727
3728 static int ftrace_match(char *str, struct ftrace_glob *g)
3729 {
3730 int matched = 0;
3731 int slen;
3732
3733 str = arch_ftrace_match_adjust(str, g->search);
3734
3735 switch (g->type) {
3736 case MATCH_FULL:
3737 if (strcmp(str, g->search) == 0)
3738 matched = 1;
3739 break;
3740 case MATCH_FRONT_ONLY:
3741 if (strncmp(str, g->search, g->len) == 0)
3742 matched = 1;
3743 break;
3744 case MATCH_MIDDLE_ONLY:
3745 if (strstr(str, g->search))
3746 matched = 1;
3747 break;
3748 case MATCH_END_ONLY:
3749 slen = strlen(str);
3750 if (slen >= g->len &&
3751 memcmp(str + slen - g->len, g->search, g->len) == 0)
3752 matched = 1;
3753 break;
3754 case MATCH_GLOB:
3755 if (glob_match(g->search, str))
3756 matched = 1;
3757 break;
3758 }
3759
3760 return matched;
3761 }
3762
3763 static int
3764 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3765 {
3766 struct ftrace_func_entry *entry;
3767 int ret = 0;
3768
3769 entry = ftrace_lookup_ip(hash, rec->ip);
3770 if (clear_filter) {
3771 /* Do nothing if it doesn't exist */
3772 if (!entry)
3773 return 0;
3774
3775 free_hash_entry(hash, entry);
3776 } else {
3777 /* Do nothing if it exists */
3778 if (entry)
3779 return 0;
3780
3781 ret = add_hash_entry(hash, rec->ip);
3782 }
3783 return ret;
3784 }
3785
3786 static int
3787 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3788 struct ftrace_glob *mod_g, int exclude_mod)
3789 {
3790 char str[KSYM_SYMBOL_LEN];
3791 char *modname;
3792
3793 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3794
3795 if (mod_g) {
3796 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3797
3798 /* blank module name to match all modules */
3799 if (!mod_g->len) {
3800 /* blank module globbing: modname xor exclude_mod */
3801 if (!exclude_mod != !modname)
3802 goto func_match;
3803 return 0;
3804 }
3805
3806 /*
3807 * exclude_mod is set to trace everything but the given
3808 * module. If it is set and the module matches, then
3809 * return 0. If it is not set, and the module doesn't match
3810 * also return 0. Otherwise, check the function to see if
3811 * that matches.
3812 */
3813 if (!mod_matches == !exclude_mod)
3814 return 0;
3815 func_match:
3816 /* blank search means to match all funcs in the mod */
3817 if (!func_g->len)
3818 return 1;
3819 }
3820
3821 return ftrace_match(str, func_g);
3822 }
3823
3824 static int
3825 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3826 {
3827 struct ftrace_page *pg;
3828 struct dyn_ftrace *rec;
3829 struct ftrace_glob func_g = { .type = MATCH_FULL };
3830 struct ftrace_glob mod_g = { .type = MATCH_FULL };
3831 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3832 int exclude_mod = 0;
3833 int found = 0;
3834 int ret;
3835 int clear_filter = 0;
3836
3837 if (func) {
3838 func_g.type = filter_parse_regex(func, len, &func_g.search,
3839 &clear_filter);
3840 func_g.len = strlen(func_g.search);
3841 }
3842
3843 if (mod) {
3844 mod_g.type = filter_parse_regex(mod, strlen(mod),
3845 &mod_g.search, &exclude_mod);
3846 mod_g.len = strlen(mod_g.search);
3847 }
3848
3849 mutex_lock(&ftrace_lock);
3850
3851 if (unlikely(ftrace_disabled))
3852 goto out_unlock;
3853
3854 do_for_each_ftrace_rec(pg, rec) {
3855
3856 if (rec->flags & FTRACE_FL_DISABLED)
3857 continue;
3858
3859 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3860 ret = enter_record(hash, rec, clear_filter);
3861 if (ret < 0) {
3862 found = ret;
3863 goto out_unlock;
3864 }
3865 found = 1;
3866 }
3867 } while_for_each_ftrace_rec();
3868 out_unlock:
3869 mutex_unlock(&ftrace_lock);
3870
3871 return found;
3872 }
3873
3874 static int
3875 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3876 {
3877 return match_records(hash, buff, len, NULL);
3878 }
3879
3880 static void ftrace_ops_update_code(struct ftrace_ops *ops,
3881 struct ftrace_ops_hash *old_hash)
3882 {
3883 struct ftrace_ops *op;
3884
3885 if (!ftrace_enabled)
3886 return;
3887
3888 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
3889 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
3890 return;
3891 }
3892
3893 /*
3894 * If this is the shared global_ops filter, then we need to
3895 * check if there is another ops that shares it, is enabled.
3896 * If so, we still need to run the modify code.
3897 */
3898 if (ops->func_hash != &global_ops.local_hash)
3899 return;
3900
3901 do_for_each_ftrace_op(op, ftrace_ops_list) {
3902 if (op->func_hash == &global_ops.local_hash &&
3903 op->flags & FTRACE_OPS_FL_ENABLED) {
3904 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
3905 /* Only need to do this once */
3906 return;
3907 }
3908 } while_for_each_ftrace_op(op);
3909 }
3910
3911 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3912 struct ftrace_hash **orig_hash,
3913 struct ftrace_hash *hash,
3914 int enable)
3915 {
3916 struct ftrace_ops_hash old_hash_ops;
3917 struct ftrace_hash *old_hash;
3918 int ret;
3919
3920 old_hash = *orig_hash;
3921 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3922 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3923 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3924 if (!ret) {
3925 ftrace_ops_update_code(ops, &old_hash_ops);
3926 free_ftrace_hash_rcu(old_hash);
3927 }
3928 return ret;
3929 }
3930
3931 static bool module_exists(const char *module)
3932 {
3933 /* All modules have the symbol __this_module */
3934 const char this_mod[] = "__this_module";
3935 const int modname_size = MAX_PARAM_PREFIX_LEN + sizeof(this_mod) + 1;
3936 char modname[modname_size + 1];
3937 unsigned long val;
3938 int n;
3939
3940 n = snprintf(modname, modname_size + 1, "%s:%s", module, this_mod);
3941
3942 if (n > modname_size)
3943 return false;
3944
3945 val = module_kallsyms_lookup_name(modname);
3946 return val != 0;
3947 }
3948
3949 static int cache_mod(struct trace_array *tr,
3950 const char *func, char *module, int enable)
3951 {
3952 struct ftrace_mod_load *ftrace_mod, *n;
3953 struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
3954 int ret;
3955
3956 mutex_lock(&ftrace_lock);
3957
3958 /* We do not cache inverse filters */
3959 if (func[0] == '!') {
3960 func++;
3961 ret = -EINVAL;
3962
3963 /* Look to remove this hash */
3964 list_for_each_entry_safe(ftrace_mod, n, head, list) {
3965 if (strcmp(ftrace_mod->module, module) != 0)
3966 continue;
3967
3968 /* no func matches all */
3969 if (strcmp(func, "*") == 0 ||
3970 (ftrace_mod->func &&
3971 strcmp(ftrace_mod->func, func) == 0)) {
3972 ret = 0;
3973 free_ftrace_mod(ftrace_mod);
3974 continue;
3975 }
3976 }
3977 goto out;
3978 }
3979
3980 ret = -EINVAL;
3981 /* We only care about modules that have not been loaded yet */
3982 if (module_exists(module))
3983 goto out;
3984
3985 /* Save this string off, and execute it when the module is loaded */
3986 ret = ftrace_add_mod(tr, func, module, enable);
3987 out:
3988 mutex_unlock(&ftrace_lock);
3989
3990 return ret;
3991 }
3992
3993 static int
3994 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3995 int reset, int enable);
3996
3997 #ifdef CONFIG_MODULES
3998 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
3999 char *mod, bool enable)
4000 {
4001 struct ftrace_mod_load *ftrace_mod, *n;
4002 struct ftrace_hash **orig_hash, *new_hash;
4003 LIST_HEAD(process_mods);
4004 char *func;
4005 int ret;
4006
4007 mutex_lock(&ops->func_hash->regex_lock);
4008
4009 if (enable)
4010 orig_hash = &ops->func_hash->filter_hash;
4011 else
4012 orig_hash = &ops->func_hash->notrace_hash;
4013
4014 new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
4015 *orig_hash);
4016 if (!new_hash)
4017 goto out; /* warn? */
4018
4019 mutex_lock(&ftrace_lock);
4020
4021 list_for_each_entry_safe(ftrace_mod, n, head, list) {
4022
4023 if (strcmp(ftrace_mod->module, mod) != 0)
4024 continue;
4025
4026 if (ftrace_mod->func)
4027 func = kstrdup(ftrace_mod->func, GFP_KERNEL);
4028 else
4029 func = kstrdup("*", GFP_KERNEL);
4030
4031 if (!func) /* warn? */
4032 continue;
4033
4034 list_del(&ftrace_mod->list);
4035 list_add(&ftrace_mod->list, &process_mods);
4036
4037 /* Use the newly allocated func, as it may be "*" */
4038 kfree(ftrace_mod->func);
4039 ftrace_mod->func = func;
4040 }
4041
4042 mutex_unlock(&ftrace_lock);
4043
4044 list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
4045
4046 func = ftrace_mod->func;
4047
4048 /* Grabs ftrace_lock, which is why we have this extra step */
4049 match_records(new_hash, func, strlen(func), mod);
4050 free_ftrace_mod(ftrace_mod);
4051 }
4052
4053 if (enable && list_empty(head))
4054 new_hash->flags &= ~FTRACE_HASH_FL_MOD;
4055
4056 mutex_lock(&ftrace_lock);
4057
4058 ret = ftrace_hash_move_and_update_ops(ops, orig_hash,
4059 new_hash, enable);
4060 mutex_unlock(&ftrace_lock);
4061
4062 out:
4063 mutex_unlock(&ops->func_hash->regex_lock);
4064
4065 free_ftrace_hash(new_hash);
4066 }
4067
4068 static void process_cached_mods(const char *mod_name)
4069 {
4070 struct trace_array *tr;
4071 char *mod;
4072
4073 mod = kstrdup(mod_name, GFP_KERNEL);
4074 if (!mod)
4075 return;
4076
4077 mutex_lock(&trace_types_lock);
4078 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
4079 if (!list_empty(&tr->mod_trace))
4080 process_mod_list(&tr->mod_trace, tr->ops, mod, true);
4081 if (!list_empty(&tr->mod_notrace))
4082 process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
4083 }
4084 mutex_unlock(&trace_types_lock);
4085
4086 kfree(mod);
4087 }
4088 #endif
4089
4090 /*
4091 * We register the module command as a template to show others how
4092 * to register the a command as well.
4093 */
4094
4095 static int
4096 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
4097 char *func_orig, char *cmd, char *module, int enable)
4098 {
4099 char *func;
4100 int ret;
4101
4102 /* match_records() modifies func, and we need the original */
4103 func = kstrdup(func_orig, GFP_KERNEL);
4104 if (!func)
4105 return -ENOMEM;
4106
4107 /*
4108 * cmd == 'mod' because we only registered this func
4109 * for the 'mod' ftrace_func_command.
4110 * But if you register one func with multiple commands,
4111 * you can tell which command was used by the cmd
4112 * parameter.
4113 */
4114 ret = match_records(hash, func, strlen(func), module);
4115 kfree(func);
4116
4117 if (!ret)
4118 return cache_mod(tr, func_orig, module, enable);
4119 if (ret < 0)
4120 return ret;
4121 return 0;
4122 }
4123
4124 static struct ftrace_func_command ftrace_mod_cmd = {
4125 .name = "mod",
4126 .func = ftrace_mod_callback,
4127 };
4128
4129 static int __init ftrace_mod_cmd_init(void)
4130 {
4131 return register_ftrace_command(&ftrace_mod_cmd);
4132 }
4133 core_initcall(ftrace_mod_cmd_init);
4134
4135 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
4136 struct ftrace_ops *op, struct pt_regs *pt_regs)
4137 {
4138 struct ftrace_probe_ops *probe_ops;
4139 struct ftrace_func_probe *probe;
4140
4141 probe = container_of(op, struct ftrace_func_probe, ops);
4142 probe_ops = probe->probe_ops;
4143
4144 /*
4145 * Disable preemption for these calls to prevent a RCU grace
4146 * period. This syncs the hash iteration and freeing of items
4147 * on the hash. rcu_read_lock is too dangerous here.
4148 */
4149 preempt_disable_notrace();
4150 probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
4151 preempt_enable_notrace();
4152 }
4153
4154 struct ftrace_func_map {
4155 struct ftrace_func_entry entry;
4156 void *data;
4157 };
4158
4159 struct ftrace_func_mapper {
4160 struct ftrace_hash hash;
4161 };
4162
4163 /**
4164 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
4165 *
4166 * Returns a ftrace_func_mapper descriptor that can be used to map ips to data.
4167 */
4168 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
4169 {
4170 struct ftrace_hash *hash;
4171
4172 /*
4173 * The mapper is simply a ftrace_hash, but since the entries
4174 * in the hash are not ftrace_func_entry type, we define it
4175 * as a separate structure.
4176 */
4177 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4178 return (struct ftrace_func_mapper *)hash;
4179 }
4180
4181 /**
4182 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
4183 * @mapper: The mapper that has the ip maps
4184 * @ip: the instruction pointer to find the data for
4185 *
4186 * Returns the data mapped to @ip if found otherwise NULL. The return
4187 * is actually the address of the mapper data pointer. The address is
4188 * returned for use cases where the data is no bigger than a long, and
4189 * the user can use the data pointer as its data instead of having to
4190 * allocate more memory for the reference.
4191 */
4192 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
4193 unsigned long ip)
4194 {
4195 struct ftrace_func_entry *entry;
4196 struct ftrace_func_map *map;
4197
4198 entry = ftrace_lookup_ip(&mapper->hash, ip);
4199 if (!entry)
4200 return NULL;
4201
4202 map = (struct ftrace_func_map *)entry;
4203 return &map->data;
4204 }
4205
4206 /**
4207 * ftrace_func_mapper_add_ip - Map some data to an ip
4208 * @mapper: The mapper that has the ip maps
4209 * @ip: The instruction pointer address to map @data to
4210 * @data: The data to map to @ip
4211 *
4212 * Returns 0 on succes otherwise an error.
4213 */
4214 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
4215 unsigned long ip, void *data)
4216 {
4217 struct ftrace_func_entry *entry;
4218 struct ftrace_func_map *map;
4219
4220 entry = ftrace_lookup_ip(&mapper->hash, ip);
4221 if (entry)
4222 return -EBUSY;
4223
4224 map = kmalloc(sizeof(*map), GFP_KERNEL);
4225 if (!map)
4226 return -ENOMEM;
4227
4228 map->entry.ip = ip;
4229 map->data = data;
4230
4231 __add_hash_entry(&mapper->hash, &map->entry);
4232
4233 return 0;
4234 }
4235
4236 /**
4237 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
4238 * @mapper: The mapper that has the ip maps
4239 * @ip: The instruction pointer address to remove the data from
4240 *
4241 * Returns the data if it is found, otherwise NULL.
4242 * Note, if the data pointer is used as the data itself, (see
4243 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
4244 * if the data pointer was set to zero.
4245 */
4246 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
4247 unsigned long ip)
4248 {
4249 struct ftrace_func_entry *entry;
4250 struct ftrace_func_map *map;
4251 void *data;
4252
4253 entry = ftrace_lookup_ip(&mapper->hash, ip);
4254 if (!entry)
4255 return NULL;
4256
4257 map = (struct ftrace_func_map *)entry;
4258 data = map->data;
4259
4260 remove_hash_entry(&mapper->hash, entry);
4261 kfree(entry);
4262
4263 return data;
4264 }
4265
4266 /**
4267 * free_ftrace_func_mapper - free a mapping of ips and data
4268 * @mapper: The mapper that has the ip maps
4269 * @free_func: A function to be called on each data item.
4270 *
4271 * This is used to free the function mapper. The @free_func is optional
4272 * and can be used if the data needs to be freed as well.
4273 */
4274 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
4275 ftrace_mapper_func free_func)
4276 {
4277 struct ftrace_func_entry *entry;
4278 struct ftrace_func_map *map;
4279 struct hlist_head *hhd;
4280 int size = 1 << mapper->hash.size_bits;
4281 int i;
4282
4283 if (free_func && mapper->hash.count) {
4284 for (i = 0; i < size; i++) {
4285 hhd = &mapper->hash.buckets[i];
4286 hlist_for_each_entry(entry, hhd, hlist) {
4287 map = (struct ftrace_func_map *)entry;
4288 free_func(map);
4289 }
4290 }
4291 }
4292 free_ftrace_hash(&mapper->hash);
4293 }
4294
4295 static void release_probe(struct ftrace_func_probe *probe)
4296 {
4297 struct ftrace_probe_ops *probe_ops;
4298
4299 mutex_lock(&ftrace_lock);
4300
4301 WARN_ON(probe->ref <= 0);
4302
4303 /* Subtract the ref that was used to protect this instance */
4304 probe->ref--;
4305
4306 if (!probe->ref) {
4307 probe_ops = probe->probe_ops;
4308 /*
4309 * Sending zero as ip tells probe_ops to free
4310 * the probe->data itself
4311 */
4312 if (probe_ops->free)
4313 probe_ops->free(probe_ops, probe->tr, 0, probe->data);
4314 list_del(&probe->list);
4315 kfree(probe);
4316 }
4317 mutex_unlock(&ftrace_lock);
4318 }
4319
4320 static void acquire_probe_locked(struct ftrace_func_probe *probe)
4321 {
4322 /*
4323 * Add one ref to keep it from being freed when releasing the
4324 * ftrace_lock mutex.
4325 */
4326 probe->ref++;
4327 }
4328
4329 int
4330 register_ftrace_function_probe(char *glob, struct trace_array *tr,
4331 struct ftrace_probe_ops *probe_ops,
4332 void *data)
4333 {
4334 struct ftrace_func_entry *entry;
4335 struct ftrace_func_probe *probe;
4336 struct ftrace_hash **orig_hash;
4337 struct ftrace_hash *old_hash;
4338 struct ftrace_hash *hash;
4339 int count = 0;
4340 int size;
4341 int ret;
4342 int i;
4343
4344 if (WARN_ON(!tr))
4345 return -EINVAL;
4346
4347 /* We do not support '!' for function probes */
4348 if (WARN_ON(glob[0] == '!'))
4349 return -EINVAL;
4350
4351
4352 mutex_lock(&ftrace_lock);
4353 /* Check if the probe_ops is already registered */
4354 list_for_each_entry(probe, &tr->func_probes, list) {
4355 if (probe->probe_ops == probe_ops)
4356 break;
4357 }
4358 if (&probe->list == &tr->func_probes) {
4359 probe = kzalloc(sizeof(*probe), GFP_KERNEL);
4360 if (!probe) {
4361 mutex_unlock(&ftrace_lock);
4362 return -ENOMEM;
4363 }
4364 probe->probe_ops = probe_ops;
4365 probe->ops.func = function_trace_probe_call;
4366 probe->tr = tr;
4367 ftrace_ops_init(&probe->ops);
4368 list_add(&probe->list, &tr->func_probes);
4369 }
4370
4371 acquire_probe_locked(probe);
4372
4373 mutex_unlock(&ftrace_lock);
4374
4375 mutex_lock(&probe->ops.func_hash->regex_lock);
4376
4377 orig_hash = &probe->ops.func_hash->filter_hash;
4378 old_hash = *orig_hash;
4379 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4380
4381 ret = ftrace_match_records(hash, glob, strlen(glob));
4382
4383 /* Nothing found? */
4384 if (!ret)
4385 ret = -EINVAL;
4386
4387 if (ret < 0)
4388 goto out;
4389
4390 size = 1 << hash->size_bits;
4391 for (i = 0; i < size; i++) {
4392 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4393 if (ftrace_lookup_ip(old_hash, entry->ip))
4394 continue;
4395 /*
4396 * The caller might want to do something special
4397 * for each function we find. We call the callback
4398 * to give the caller an opportunity to do so.
4399 */
4400 if (probe_ops->init) {
4401 ret = probe_ops->init(probe_ops, tr,
4402 entry->ip, data,
4403 &probe->data);
4404 if (ret < 0) {
4405 if (probe_ops->free && count)
4406 probe_ops->free(probe_ops, tr,
4407 0, probe->data);
4408 probe->data = NULL;
4409 goto out;
4410 }
4411 }
4412 count++;
4413 }
4414 }
4415
4416 mutex_lock(&ftrace_lock);
4417
4418 if (!count) {
4419 /* Nothing was added? */
4420 ret = -EINVAL;
4421 goto out_unlock;
4422 }
4423
4424 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4425 hash, 1);
4426 if (ret < 0)
4427 goto err_unlock;
4428
4429 /* One ref for each new function traced */
4430 probe->ref += count;
4431
4432 if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
4433 ret = ftrace_startup(&probe->ops, 0);
4434
4435 out_unlock:
4436 mutex_unlock(&ftrace_lock);
4437
4438 if (!ret)
4439 ret = count;
4440 out:
4441 mutex_unlock(&probe->ops.func_hash->regex_lock);
4442 free_ftrace_hash(hash);
4443
4444 release_probe(probe);
4445
4446 return ret;
4447
4448 err_unlock:
4449 if (!probe_ops->free || !count)
4450 goto out_unlock;
4451
4452 /* Failed to do the move, need to call the free functions */
4453 for (i = 0; i < size; i++) {
4454 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4455 if (ftrace_lookup_ip(old_hash, entry->ip))
4456 continue;
4457 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4458 }
4459 }
4460 goto out_unlock;
4461 }
4462
4463 int
4464 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
4465 struct ftrace_probe_ops *probe_ops)
4466 {
4467 struct ftrace_ops_hash old_hash_ops;
4468 struct ftrace_func_entry *entry;
4469 struct ftrace_func_probe *probe;
4470 struct ftrace_glob func_g;
4471 struct ftrace_hash **orig_hash;
4472 struct ftrace_hash *old_hash;
4473 struct ftrace_hash *hash = NULL;
4474 struct hlist_node *tmp;
4475 struct hlist_head hhd;
4476 char str[KSYM_SYMBOL_LEN];
4477 int count = 0;
4478 int i, ret = -ENODEV;
4479 int size;
4480
4481 if (!glob || !strlen(glob) || !strcmp(glob, "*"))
4482 func_g.search = NULL;
4483 else {
4484 int not;
4485
4486 func_g.type = filter_parse_regex(glob, strlen(glob),
4487 &func_g.search, &not);
4488 func_g.len = strlen(func_g.search);
4489 func_g.search = glob;
4490
4491 /* we do not support '!' for function probes */
4492 if (WARN_ON(not))
4493 return -EINVAL;
4494 }
4495
4496 mutex_lock(&ftrace_lock);
4497 /* Check if the probe_ops is already registered */
4498 list_for_each_entry(probe, &tr->func_probes, list) {
4499 if (probe->probe_ops == probe_ops)
4500 break;
4501 }
4502 if (&probe->list == &tr->func_probes)
4503 goto err_unlock_ftrace;
4504
4505 ret = -EINVAL;
4506 if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
4507 goto err_unlock_ftrace;
4508
4509 acquire_probe_locked(probe);
4510
4511 mutex_unlock(&ftrace_lock);
4512
4513 mutex_lock(&probe->ops.func_hash->regex_lock);
4514
4515 orig_hash = &probe->ops.func_hash->filter_hash;
4516 old_hash = *orig_hash;
4517
4518 if (ftrace_hash_empty(old_hash))
4519 goto out_unlock;
4520
4521 old_hash_ops.filter_hash = old_hash;
4522 /* Probes only have filters */
4523 old_hash_ops.notrace_hash = NULL;
4524
4525 ret = -ENOMEM;
4526 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4527 if (!hash)
4528 goto out_unlock;
4529
4530 INIT_HLIST_HEAD(&hhd);
4531
4532 size = 1 << hash->size_bits;
4533 for (i = 0; i < size; i++) {
4534 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
4535
4536 if (func_g.search) {
4537 kallsyms_lookup(entry->ip, NULL, NULL,
4538 NULL, str);
4539 if (!ftrace_match(str, &func_g))
4540 continue;
4541 }
4542 count++;
4543 remove_hash_entry(hash, entry);
4544 hlist_add_head(&entry->hlist, &hhd);
4545 }
4546 }
4547
4548 /* Nothing found? */
4549 if (!count) {
4550 ret = -EINVAL;
4551 goto out_unlock;
4552 }
4553
4554 mutex_lock(&ftrace_lock);
4555
4556 WARN_ON(probe->ref < count);
4557
4558 probe->ref -= count;
4559
4560 if (ftrace_hash_empty(hash))
4561 ftrace_shutdown(&probe->ops, 0);
4562
4563 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4564 hash, 1);
4565
4566 /* still need to update the function call sites */
4567 if (ftrace_enabled && !ftrace_hash_empty(hash))
4568 ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
4569 &old_hash_ops);
4570 synchronize_sched();
4571
4572 hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
4573 hlist_del(&entry->hlist);
4574 if (probe_ops->free)
4575 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4576 kfree(entry);
4577 }
4578 mutex_unlock(&ftrace_lock);
4579
4580 out_unlock:
4581 mutex_unlock(&probe->ops.func_hash->regex_lock);
4582 free_ftrace_hash(hash);
4583
4584 release_probe(probe);
4585
4586 return ret;
4587
4588 err_unlock_ftrace:
4589 mutex_unlock(&ftrace_lock);
4590 return ret;
4591 }
4592
4593 void clear_ftrace_function_probes(struct trace_array *tr)
4594 {
4595 struct ftrace_func_probe *probe, *n;
4596
4597 list_for_each_entry_safe(probe, n, &tr->func_probes, list)
4598 unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
4599 }
4600
4601 static LIST_HEAD(ftrace_commands);
4602 static DEFINE_MUTEX(ftrace_cmd_mutex);
4603
4604 /*
4605 * Currently we only register ftrace commands from __init, so mark this
4606 * __init too.
4607 */
4608 __init int register_ftrace_command(struct ftrace_func_command *cmd)
4609 {
4610 struct ftrace_func_command *p;
4611 int ret = 0;
4612
4613 mutex_lock(&ftrace_cmd_mutex);
4614 list_for_each_entry(p, &ftrace_commands, list) {
4615 if (strcmp(cmd->name, p->name) == 0) {
4616 ret = -EBUSY;
4617 goto out_unlock;
4618 }
4619 }
4620 list_add(&cmd->list, &ftrace_commands);
4621 out_unlock:
4622 mutex_unlock(&ftrace_cmd_mutex);
4623
4624 return ret;
4625 }
4626
4627 /*
4628 * Currently we only unregister ftrace commands from __init, so mark
4629 * this __init too.
4630 */
4631 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
4632 {
4633 struct ftrace_func_command *p, *n;
4634 int ret = -ENODEV;
4635
4636 mutex_lock(&ftrace_cmd_mutex);
4637 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4638 if (strcmp(cmd->name, p->name) == 0) {
4639 ret = 0;
4640 list_del_init(&p->list);
4641 goto out_unlock;
4642 }
4643 }
4644 out_unlock:
4645 mutex_unlock(&ftrace_cmd_mutex);
4646
4647 return ret;
4648 }
4649
4650 static int ftrace_process_regex(struct ftrace_iterator *iter,
4651 char *buff, int len, int enable)
4652 {
4653 struct ftrace_hash *hash = iter->hash;
4654 struct trace_array *tr = iter->ops->private;
4655 char *func, *command, *next = buff;
4656 struct ftrace_func_command *p;
4657 int ret = -EINVAL;
4658
4659 func = strsep(&next, ":");
4660
4661 if (!next) {
4662 ret = ftrace_match_records(hash, func, len);
4663 if (!ret)
4664 ret = -EINVAL;
4665 if (ret < 0)
4666 return ret;
4667 return 0;
4668 }
4669
4670 /* command found */
4671
4672 command = strsep(&next, ":");
4673
4674 mutex_lock(&ftrace_cmd_mutex);
4675 list_for_each_entry(p, &ftrace_commands, list) {
4676 if (strcmp(p->name, command) == 0) {
4677 ret = p->func(tr, hash, func, command, next, enable);
4678 goto out_unlock;
4679 }
4680 }
4681 out_unlock:
4682 mutex_unlock(&ftrace_cmd_mutex);
4683
4684 return ret;
4685 }
4686
4687 static ssize_t
4688 ftrace_regex_write(struct file *file, const char __user *ubuf,
4689 size_t cnt, loff_t *ppos, int enable)
4690 {
4691 struct ftrace_iterator *iter;
4692 struct trace_parser *parser;
4693 ssize_t ret, read;
4694
4695 if (!cnt)
4696 return 0;
4697
4698 if (file->f_mode & FMODE_READ) {
4699 struct seq_file *m = file->private_data;
4700 iter = m->private;
4701 } else
4702 iter = file->private_data;
4703
4704 if (unlikely(ftrace_disabled))
4705 return -ENODEV;
4706
4707 /* iter->hash is a local copy, so we don't need regex_lock */
4708
4709 parser = &iter->parser;
4710 read = trace_get_user(parser, ubuf, cnt, ppos);
4711
4712 if (read >= 0 && trace_parser_loaded(parser) &&
4713 !trace_parser_cont(parser)) {
4714 ret = ftrace_process_regex(iter, parser->buffer,
4715 parser->idx, enable);
4716 trace_parser_clear(parser);
4717 if (ret < 0)
4718 goto out;
4719 }
4720
4721 ret = read;
4722 out:
4723 return ret;
4724 }
4725
4726 ssize_t
4727 ftrace_filter_write(struct file *file, const char __user *ubuf,
4728 size_t cnt, loff_t *ppos)
4729 {
4730 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4731 }
4732
4733 ssize_t
4734 ftrace_notrace_write(struct file *file, const char __user *ubuf,
4735 size_t cnt, loff_t *ppos)
4736 {
4737 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4738 }
4739
4740 static int
4741 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4742 {
4743 struct ftrace_func_entry *entry;
4744
4745 if (!ftrace_location(ip))
4746 return -EINVAL;
4747
4748 if (remove) {
4749 entry = ftrace_lookup_ip(hash, ip);
4750 if (!entry)
4751 return -ENOENT;
4752 free_hash_entry(hash, entry);
4753 return 0;
4754 }
4755
4756 return add_hash_entry(hash, ip);
4757 }
4758
4759 static int
4760 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4761 unsigned long ip, int remove, int reset, int enable)
4762 {
4763 struct ftrace_hash **orig_hash;
4764 struct ftrace_hash *hash;
4765 int ret;
4766
4767 if (unlikely(ftrace_disabled))
4768 return -ENODEV;
4769
4770 mutex_lock(&ops->func_hash->regex_lock);
4771
4772 if (enable)
4773 orig_hash = &ops->func_hash->filter_hash;
4774 else
4775 orig_hash = &ops->func_hash->notrace_hash;
4776
4777 if (reset)
4778 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4779 else
4780 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4781
4782 if (!hash) {
4783 ret = -ENOMEM;
4784 goto out_regex_unlock;
4785 }
4786
4787 if (buf && !ftrace_match_records(hash, buf, len)) {
4788 ret = -EINVAL;
4789 goto out_regex_unlock;
4790 }
4791 if (ip) {
4792 ret = ftrace_match_addr(hash, ip, remove);
4793 if (ret < 0)
4794 goto out_regex_unlock;
4795 }
4796
4797 mutex_lock(&ftrace_lock);
4798 ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4799 mutex_unlock(&ftrace_lock);
4800
4801 out_regex_unlock:
4802 mutex_unlock(&ops->func_hash->regex_lock);
4803
4804 free_ftrace_hash(hash);
4805 return ret;
4806 }
4807
4808 static int
4809 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4810 int reset, int enable)
4811 {
4812 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4813 }
4814
4815 /**
4816 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4817 * @ops - the ops to set the filter with
4818 * @ip - the address to add to or remove from the filter.
4819 * @remove - non zero to remove the ip from the filter
4820 * @reset - non zero to reset all filters before applying this filter.
4821 *
4822 * Filters denote which functions should be enabled when tracing is enabled
4823 * If @ip is NULL, it failes to update filter.
4824 */
4825 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4826 int remove, int reset)
4827 {
4828 ftrace_ops_init(ops);
4829 return ftrace_set_addr(ops, ip, remove, reset, 1);
4830 }
4831 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4832
4833 /**
4834 * ftrace_ops_set_global_filter - setup ops to use global filters
4835 * @ops - the ops which will use the global filters
4836 *
4837 * ftrace users who need global function trace filtering should call this.
4838 * It can set the global filter only if ops were not initialized before.
4839 */
4840 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
4841 {
4842 if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
4843 return;
4844
4845 ftrace_ops_init(ops);
4846 ops->func_hash = &global_ops.local_hash;
4847 }
4848 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
4849
4850 static int
4851 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4852 int reset, int enable)
4853 {
4854 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4855 }
4856
4857 /**
4858 * ftrace_set_filter - set a function to filter on in ftrace
4859 * @ops - the ops to set the filter with
4860 * @buf - the string that holds the function filter text.
4861 * @len - the length of the string.
4862 * @reset - non zero to reset all filters before applying this filter.
4863 *
4864 * Filters denote which functions should be enabled when tracing is enabled.
4865 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4866 */
4867 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4868 int len, int reset)
4869 {
4870 ftrace_ops_init(ops);
4871 return ftrace_set_regex(ops, buf, len, reset, 1);
4872 }
4873 EXPORT_SYMBOL_GPL(ftrace_set_filter);
4874
4875 /**
4876 * ftrace_set_notrace - set a function to not trace in ftrace
4877 * @ops - the ops to set the notrace filter with
4878 * @buf - the string that holds the function notrace text.
4879 * @len - the length of the string.
4880 * @reset - non zero to reset all filters before applying this filter.
4881 *
4882 * Notrace Filters denote which functions should not be enabled when tracing
4883 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4884 * for tracing.
4885 */
4886 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4887 int len, int reset)
4888 {
4889 ftrace_ops_init(ops);
4890 return ftrace_set_regex(ops, buf, len, reset, 0);
4891 }
4892 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4893 /**
4894 * ftrace_set_global_filter - set a function to filter on with global tracers
4895 * @buf - the string that holds the function filter text.
4896 * @len - the length of the string.
4897 * @reset - non zero to reset all filters before applying this filter.
4898 *
4899 * Filters denote which functions should be enabled when tracing is enabled.
4900 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4901 */
4902 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4903 {
4904 ftrace_set_regex(&global_ops, buf, len, reset, 1);
4905 }
4906 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4907
4908 /**
4909 * ftrace_set_global_notrace - set a function to not trace with global tracers
4910 * @buf - the string that holds the function notrace text.
4911 * @len - the length of the string.
4912 * @reset - non zero to reset all filters before applying this filter.
4913 *
4914 * Notrace Filters denote which functions should not be enabled when tracing
4915 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4916 * for tracing.
4917 */
4918 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4919 {
4920 ftrace_set_regex(&global_ops, buf, len, reset, 0);
4921 }
4922 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4923
4924 /*
4925 * command line interface to allow users to set filters on boot up.
4926 */
4927 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
4928 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4929 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4930
4931 /* Used by function selftest to not test if filter is set */
4932 bool ftrace_filter_param __initdata;
4933
4934 static int __init set_ftrace_notrace(char *str)
4935 {
4936 ftrace_filter_param = true;
4937 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4938 return 1;
4939 }
4940 __setup("ftrace_notrace=", set_ftrace_notrace);
4941
4942 static int __init set_ftrace_filter(char *str)
4943 {
4944 ftrace_filter_param = true;
4945 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4946 return 1;
4947 }
4948 __setup("ftrace_filter=", set_ftrace_filter);
4949
4950 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4951 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4952 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4953 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
4954
4955 static unsigned long save_global_trampoline;
4956 static unsigned long save_global_flags;
4957
4958 static int __init set_graph_function(char *str)
4959 {
4960 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4961 return 1;
4962 }
4963 __setup("ftrace_graph_filter=", set_graph_function);
4964
4965 static int __init set_graph_notrace_function(char *str)
4966 {
4967 strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4968 return 1;
4969 }
4970 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
4971
4972 static int __init set_graph_max_depth_function(char *str)
4973 {
4974 if (!str)
4975 return 0;
4976 fgraph_max_depth = simple_strtoul(str, NULL, 0);
4977 return 1;
4978 }
4979 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
4980
4981 static void __init set_ftrace_early_graph(char *buf, int enable)
4982 {
4983 int ret;
4984 char *func;
4985 struct ftrace_hash *hash;
4986
4987 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4988 if (WARN_ON(!hash))
4989 return;
4990
4991 while (buf) {
4992 func = strsep(&buf, ",");
4993 /* we allow only one expression at a time */
4994 ret = ftrace_graph_set_hash(hash, func);
4995 if (ret)
4996 printk(KERN_DEBUG "ftrace: function %s not "
4997 "traceable\n", func);
4998 }
4999
5000 if (enable)
5001 ftrace_graph_hash = hash;
5002 else
5003 ftrace_graph_notrace_hash = hash;
5004 }
5005 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5006
5007 void __init
5008 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
5009 {
5010 char *func;
5011
5012 ftrace_ops_init(ops);
5013
5014 while (buf) {
5015 func = strsep(&buf, ",");
5016 ftrace_set_regex(ops, func, strlen(func), 0, enable);
5017 }
5018 }
5019
5020 static void __init set_ftrace_early_filters(void)
5021 {
5022 if (ftrace_filter_buf[0])
5023 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
5024 if (ftrace_notrace_buf[0])
5025 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
5026 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5027 if (ftrace_graph_buf[0])
5028 set_ftrace_early_graph(ftrace_graph_buf, 1);
5029 if (ftrace_graph_notrace_buf[0])
5030 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
5031 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5032 }
5033
5034 int ftrace_regex_release(struct inode *inode, struct file *file)
5035 {
5036 struct seq_file *m = (struct seq_file *)file->private_data;
5037 struct ftrace_iterator *iter;
5038 struct ftrace_hash **orig_hash;
5039 struct trace_parser *parser;
5040 int filter_hash;
5041 int ret;
5042
5043 if (file->f_mode & FMODE_READ) {
5044 iter = m->private;
5045 seq_release(inode, file);
5046 } else
5047 iter = file->private_data;
5048
5049 parser = &iter->parser;
5050 if (trace_parser_loaded(parser)) {
5051 parser->buffer[parser->idx] = 0;
5052 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
5053 }
5054
5055 trace_parser_put(parser);
5056
5057 mutex_lock(&iter->ops->func_hash->regex_lock);
5058
5059 if (file->f_mode & FMODE_WRITE) {
5060 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
5061
5062 if (filter_hash) {
5063 orig_hash = &iter->ops->func_hash->filter_hash;
5064 if (iter->tr && !list_empty(&iter->tr->mod_trace))
5065 iter->hash->flags |= FTRACE_HASH_FL_MOD;
5066 } else
5067 orig_hash = &iter->ops->func_hash->notrace_hash;
5068
5069 mutex_lock(&ftrace_lock);
5070 ret = ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
5071 iter->hash, filter_hash);
5072 mutex_unlock(&ftrace_lock);
5073 } else {
5074 /* For read only, the hash is the ops hash */
5075 iter->hash = NULL;
5076 }
5077
5078 mutex_unlock(&iter->ops->func_hash->regex_lock);
5079 free_ftrace_hash(iter->hash);
5080 kfree(iter);
5081
5082 return 0;
5083 }
5084
5085 static const struct file_operations ftrace_avail_fops = {
5086 .open = ftrace_avail_open,
5087 .read = seq_read,
5088 .llseek = seq_lseek,
5089 .release = seq_release_private,
5090 };
5091
5092 static const struct file_operations ftrace_enabled_fops = {
5093 .open = ftrace_enabled_open,
5094 .read = seq_read,
5095 .llseek = seq_lseek,
5096 .release = seq_release_private,
5097 };
5098
5099 static const struct file_operations ftrace_filter_fops = {
5100 .open = ftrace_filter_open,
5101 .read = seq_read,
5102 .write = ftrace_filter_write,
5103 .llseek = tracing_lseek,
5104 .release = ftrace_regex_release,
5105 };
5106
5107 static const struct file_operations ftrace_notrace_fops = {
5108 .open = ftrace_notrace_open,
5109 .read = seq_read,
5110 .write = ftrace_notrace_write,
5111 .llseek = tracing_lseek,
5112 .release = ftrace_regex_release,
5113 };
5114
5115 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5116
5117 static DEFINE_MUTEX(graph_lock);
5118
5119 struct ftrace_hash *ftrace_graph_hash = EMPTY_HASH;
5120 struct ftrace_hash *ftrace_graph_notrace_hash = EMPTY_HASH;
5121
5122 enum graph_filter_type {
5123 GRAPH_FILTER_NOTRACE = 0,
5124 GRAPH_FILTER_FUNCTION,
5125 };
5126
5127 #define FTRACE_GRAPH_EMPTY ((void *)1)
5128
5129 struct ftrace_graph_data {
5130 struct ftrace_hash *hash;
5131 struct ftrace_func_entry *entry;
5132 int idx; /* for hash table iteration */
5133 enum graph_filter_type type;
5134 struct ftrace_hash *new_hash;
5135 const struct seq_operations *seq_ops;
5136 struct trace_parser parser;
5137 };
5138
5139 static void *
5140 __g_next(struct seq_file *m, loff_t *pos)
5141 {
5142 struct ftrace_graph_data *fgd = m->private;
5143 struct ftrace_func_entry *entry = fgd->entry;
5144 struct hlist_head *head;
5145 int i, idx = fgd->idx;
5146
5147 if (*pos >= fgd->hash->count)
5148 return NULL;
5149
5150 if (entry) {
5151 hlist_for_each_entry_continue(entry, hlist) {
5152 fgd->entry = entry;
5153 return entry;
5154 }
5155
5156 idx++;
5157 }
5158
5159 for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
5160 head = &fgd->hash->buckets[i];
5161 hlist_for_each_entry(entry, head, hlist) {
5162 fgd->entry = entry;
5163 fgd->idx = i;
5164 return entry;
5165 }
5166 }
5167 return NULL;
5168 }
5169
5170 static void *
5171 g_next(struct seq_file *m, void *v, loff_t *pos)
5172 {
5173 (*pos)++;
5174 return __g_next(m, pos);
5175 }
5176
5177 static void *g_start(struct seq_file *m, loff_t *pos)
5178 {
5179 struct ftrace_graph_data *fgd = m->private;
5180
5181 mutex_lock(&graph_lock);
5182
5183 if (fgd->type == GRAPH_FILTER_FUNCTION)
5184 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5185 lockdep_is_held(&graph_lock));
5186 else
5187 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5188 lockdep_is_held(&graph_lock));
5189
5190 /* Nothing, tell g_show to print all functions are enabled */
5191 if (ftrace_hash_empty(fgd->hash) && !*pos)
5192 return FTRACE_GRAPH_EMPTY;
5193
5194 fgd->idx = 0;
5195 fgd->entry = NULL;
5196 return __g_next(m, pos);
5197 }
5198
5199 static void g_stop(struct seq_file *m, void *p)
5200 {
5201 mutex_unlock(&graph_lock);
5202 }
5203
5204 static int g_show(struct seq_file *m, void *v)
5205 {
5206 struct ftrace_func_entry *entry = v;
5207
5208 if (!entry)
5209 return 0;
5210
5211 if (entry == FTRACE_GRAPH_EMPTY) {
5212 struct ftrace_graph_data *fgd = m->private;
5213
5214 if (fgd->type == GRAPH_FILTER_FUNCTION)
5215 seq_puts(m, "#### all functions enabled ####\n");
5216 else
5217 seq_puts(m, "#### no functions disabled ####\n");
5218 return 0;
5219 }
5220
5221 seq_printf(m, "%ps\n", (void *)entry->ip);
5222
5223 return 0;
5224 }
5225
5226 static const struct seq_operations ftrace_graph_seq_ops = {
5227 .start = g_start,
5228 .next = g_next,
5229 .stop = g_stop,
5230 .show = g_show,
5231 };
5232
5233 static int
5234 __ftrace_graph_open(struct inode *inode, struct file *file,
5235 struct ftrace_graph_data *fgd)
5236 {
5237 int ret = 0;
5238 struct ftrace_hash *new_hash = NULL;
5239
5240 if (file->f_mode & FMODE_WRITE) {
5241 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
5242
5243 if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
5244 return -ENOMEM;
5245
5246 if (file->f_flags & O_TRUNC)
5247 new_hash = alloc_ftrace_hash(size_bits);
5248 else
5249 new_hash = alloc_and_copy_ftrace_hash(size_bits,
5250 fgd->hash);
5251 if (!new_hash) {
5252 ret = -ENOMEM;
5253 goto out;
5254 }
5255 }
5256
5257 if (file->f_mode & FMODE_READ) {
5258 ret = seq_open(file, &ftrace_graph_seq_ops);
5259 if (!ret) {
5260 struct seq_file *m = file->private_data;
5261 m->private = fgd;
5262 } else {
5263 /* Failed */
5264 free_ftrace_hash(new_hash);
5265 new_hash = NULL;
5266 }
5267 } else
5268 file->private_data = fgd;
5269
5270 out:
5271 if (ret < 0 && file->f_mode & FMODE_WRITE)
5272 trace_parser_put(&fgd->parser);
5273
5274 fgd->new_hash = new_hash;
5275
5276 /*
5277 * All uses of fgd->hash must be taken with the graph_lock
5278 * held. The graph_lock is going to be released, so force
5279 * fgd->hash to be reinitialized when it is taken again.
5280 */
5281 fgd->hash = NULL;
5282
5283 return ret;
5284 }
5285
5286 static int
5287 ftrace_graph_open(struct inode *inode, struct file *file)
5288 {
5289 struct ftrace_graph_data *fgd;
5290 int ret;
5291
5292 if (unlikely(ftrace_disabled))
5293 return -ENODEV;
5294
5295 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5296 if (fgd == NULL)
5297 return -ENOMEM;
5298
5299 mutex_lock(&graph_lock);
5300
5301 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5302 lockdep_is_held(&graph_lock));
5303 fgd->type = GRAPH_FILTER_FUNCTION;
5304 fgd->seq_ops = &ftrace_graph_seq_ops;
5305
5306 ret = __ftrace_graph_open(inode, file, fgd);
5307 if (ret < 0)
5308 kfree(fgd);
5309
5310 mutex_unlock(&graph_lock);
5311 return ret;
5312 }
5313
5314 static int
5315 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
5316 {
5317 struct ftrace_graph_data *fgd;
5318 int ret;
5319
5320 if (unlikely(ftrace_disabled))
5321 return -ENODEV;
5322
5323 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5324 if (fgd == NULL)
5325 return -ENOMEM;
5326
5327 mutex_lock(&graph_lock);
5328
5329 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5330 lockdep_is_held(&graph_lock));
5331 fgd->type = GRAPH_FILTER_NOTRACE;
5332 fgd->seq_ops = &ftrace_graph_seq_ops;
5333
5334 ret = __ftrace_graph_open(inode, file, fgd);
5335 if (ret < 0)
5336 kfree(fgd);
5337
5338 mutex_unlock(&graph_lock);
5339 return ret;
5340 }
5341
5342 static int
5343 ftrace_graph_release(struct inode *inode, struct file *file)
5344 {
5345 struct ftrace_graph_data *fgd;
5346 struct ftrace_hash *old_hash, *new_hash;
5347 struct trace_parser *parser;
5348 int ret = 0;
5349
5350 if (file->f_mode & FMODE_READ) {
5351 struct seq_file *m = file->private_data;
5352
5353 fgd = m->private;
5354 seq_release(inode, file);
5355 } else {
5356 fgd = file->private_data;
5357 }
5358
5359
5360 if (file->f_mode & FMODE_WRITE) {
5361
5362 parser = &fgd->parser;
5363
5364 if (trace_parser_loaded((parser))) {
5365 parser->buffer[parser->idx] = 0;
5366 ret = ftrace_graph_set_hash(fgd->new_hash,
5367 parser->buffer);
5368 }
5369
5370 trace_parser_put(parser);
5371
5372 new_hash = __ftrace_hash_move(fgd->new_hash);
5373 if (!new_hash) {
5374 ret = -ENOMEM;
5375 goto out;
5376 }
5377
5378 mutex_lock(&graph_lock);
5379
5380 if (fgd->type == GRAPH_FILTER_FUNCTION) {
5381 old_hash = rcu_dereference_protected(ftrace_graph_hash,
5382 lockdep_is_held(&graph_lock));
5383 rcu_assign_pointer(ftrace_graph_hash, new_hash);
5384 } else {
5385 old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5386 lockdep_is_held(&graph_lock));
5387 rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
5388 }
5389
5390 mutex_unlock(&graph_lock);
5391
5392 /* Wait till all users are no longer using the old hash */
5393 synchronize_sched();
5394
5395 free_ftrace_hash(old_hash);
5396 }
5397
5398 out:
5399 free_ftrace_hash(fgd->new_hash);
5400 kfree(fgd);
5401
5402 return ret;
5403 }
5404
5405 static int
5406 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
5407 {
5408 struct ftrace_glob func_g;
5409 struct dyn_ftrace *rec;
5410 struct ftrace_page *pg;
5411 struct ftrace_func_entry *entry;
5412 int fail = 1;
5413 int not;
5414
5415 /* decode regex */
5416 func_g.type = filter_parse_regex(buffer, strlen(buffer),
5417 &func_g.search, &not);
5418
5419 func_g.len = strlen(func_g.search);
5420
5421 mutex_lock(&ftrace_lock);
5422
5423 if (unlikely(ftrace_disabled)) {
5424 mutex_unlock(&ftrace_lock);
5425 return -ENODEV;
5426 }
5427
5428 do_for_each_ftrace_rec(pg, rec) {
5429
5430 if (rec->flags & FTRACE_FL_DISABLED)
5431 continue;
5432
5433 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
5434 entry = ftrace_lookup_ip(hash, rec->ip);
5435
5436 if (!not) {
5437 fail = 0;
5438
5439 if (entry)
5440 continue;
5441 if (add_hash_entry(hash, rec->ip) < 0)
5442 goto out;
5443 } else {
5444 if (entry) {
5445 free_hash_entry(hash, entry);
5446 fail = 0;
5447 }
5448 }
5449 }
5450 } while_for_each_ftrace_rec();
5451 out:
5452 mutex_unlock(&ftrace_lock);
5453
5454 if (fail)
5455 return -EINVAL;
5456
5457 return 0;
5458 }
5459
5460 static ssize_t
5461 ftrace_graph_write(struct file *file, const char __user *ubuf,
5462 size_t cnt, loff_t *ppos)
5463 {
5464 ssize_t read, ret = 0;
5465 struct ftrace_graph_data *fgd = file->private_data;
5466 struct trace_parser *parser;
5467
5468 if (!cnt)
5469 return 0;
5470
5471 /* Read mode uses seq functions */
5472 if (file->f_mode & FMODE_READ) {
5473 struct seq_file *m = file->private_data;
5474 fgd = m->private;
5475 }
5476
5477 parser = &fgd->parser;
5478
5479 read = trace_get_user(parser, ubuf, cnt, ppos);
5480
5481 if (read >= 0 && trace_parser_loaded(parser) &&
5482 !trace_parser_cont(parser)) {
5483
5484 ret = ftrace_graph_set_hash(fgd->new_hash,
5485 parser->buffer);
5486 trace_parser_clear(parser);
5487 }
5488
5489 if (!ret)
5490 ret = read;
5491
5492 return ret;
5493 }
5494
5495 static const struct file_operations ftrace_graph_fops = {
5496 .open = ftrace_graph_open,
5497 .read = seq_read,
5498 .write = ftrace_graph_write,
5499 .llseek = tracing_lseek,
5500 .release = ftrace_graph_release,
5501 };
5502
5503 static const struct file_operations ftrace_graph_notrace_fops = {
5504 .open = ftrace_graph_notrace_open,
5505 .read = seq_read,
5506 .write = ftrace_graph_write,
5507 .llseek = tracing_lseek,
5508 .release = ftrace_graph_release,
5509 };
5510 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5511
5512 void ftrace_create_filter_files(struct ftrace_ops *ops,
5513 struct dentry *parent)
5514 {
5515
5516 trace_create_file("set_ftrace_filter", 0644, parent,
5517 ops, &ftrace_filter_fops);
5518
5519 trace_create_file("set_ftrace_notrace", 0644, parent,
5520 ops, &ftrace_notrace_fops);
5521 }
5522
5523 /*
5524 * The name "destroy_filter_files" is really a misnomer. Although
5525 * in the future, it may actualy delete the files, but this is
5526 * really intended to make sure the ops passed in are disabled
5527 * and that when this function returns, the caller is free to
5528 * free the ops.
5529 *
5530 * The "destroy" name is only to match the "create" name that this
5531 * should be paired with.
5532 */
5533 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
5534 {
5535 mutex_lock(&ftrace_lock);
5536 if (ops->flags & FTRACE_OPS_FL_ENABLED)
5537 ftrace_shutdown(ops, 0);
5538 ops->flags |= FTRACE_OPS_FL_DELETED;
5539 mutex_unlock(&ftrace_lock);
5540 }
5541
5542 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
5543 {
5544
5545 trace_create_file("available_filter_functions", 0444,
5546 d_tracer, NULL, &ftrace_avail_fops);
5547
5548 trace_create_file("enabled_functions", 0444,
5549 d_tracer, NULL, &ftrace_enabled_fops);
5550
5551 ftrace_create_filter_files(&global_ops, d_tracer);
5552
5553 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5554 trace_create_file("set_graph_function", 0444, d_tracer,
5555 NULL,
5556 &ftrace_graph_fops);
5557 trace_create_file("set_graph_notrace", 0444, d_tracer,
5558 NULL,
5559 &ftrace_graph_notrace_fops);
5560 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5561
5562 return 0;
5563 }
5564
5565 static int ftrace_cmp_ips(const void *a, const void *b)
5566 {
5567 const unsigned long *ipa = a;
5568 const unsigned long *ipb = b;
5569
5570 if (*ipa > *ipb)
5571 return 1;
5572 if (*ipa < *ipb)
5573 return -1;
5574 return 0;
5575 }
5576
5577 static int ftrace_process_locs(struct module *mod,
5578 unsigned long *start,
5579 unsigned long *end)
5580 {
5581 struct ftrace_page *start_pg;
5582 struct ftrace_page *pg;
5583 struct dyn_ftrace *rec;
5584 unsigned long count;
5585 unsigned long *p;
5586 unsigned long addr;
5587 unsigned long flags = 0; /* Shut up gcc */
5588 int ret = -ENOMEM;
5589
5590 count = end - start;
5591
5592 if (!count)
5593 return 0;
5594
5595 sort(start, count, sizeof(*start),
5596 ftrace_cmp_ips, NULL);
5597
5598 start_pg = ftrace_allocate_pages(count);
5599 if (!start_pg)
5600 return -ENOMEM;
5601
5602 mutex_lock(&ftrace_lock);
5603
5604 /*
5605 * Core and each module needs their own pages, as
5606 * modules will free them when they are removed.
5607 * Force a new page to be allocated for modules.
5608 */
5609 if (!mod) {
5610 WARN_ON(ftrace_pages || ftrace_pages_start);
5611 /* First initialization */
5612 ftrace_pages = ftrace_pages_start = start_pg;
5613 } else {
5614 if (!ftrace_pages)
5615 goto out;
5616
5617 if (WARN_ON(ftrace_pages->next)) {
5618 /* Hmm, we have free pages? */
5619 while (ftrace_pages->next)
5620 ftrace_pages = ftrace_pages->next;
5621 }
5622
5623 ftrace_pages->next = start_pg;
5624 }
5625
5626 p = start;
5627 pg = start_pg;
5628 while (p < end) {
5629 addr = ftrace_call_adjust(*p++);
5630 /*
5631 * Some architecture linkers will pad between
5632 * the different mcount_loc sections of different
5633 * object files to satisfy alignments.
5634 * Skip any NULL pointers.
5635 */
5636 if (!addr)
5637 continue;
5638
5639 if (pg->index == pg->size) {
5640 /* We should have allocated enough */
5641 if (WARN_ON(!pg->next))
5642 break;
5643 pg = pg->next;
5644 }
5645
5646 rec = &pg->records[pg->index++];
5647 rec->ip = addr;
5648 }
5649
5650 /* We should have used all pages */
5651 WARN_ON(pg->next);
5652
5653 /* Assign the last page to ftrace_pages */
5654 ftrace_pages = pg;
5655
5656 /*
5657 * We only need to disable interrupts on start up
5658 * because we are modifying code that an interrupt
5659 * may execute, and the modification is not atomic.
5660 * But for modules, nothing runs the code we modify
5661 * until we are finished with it, and there's no
5662 * reason to cause large interrupt latencies while we do it.
5663 */
5664 if (!mod)
5665 local_irq_save(flags);
5666 ftrace_update_code(mod, start_pg);
5667 if (!mod)
5668 local_irq_restore(flags);
5669 ret = 0;
5670 out:
5671 mutex_unlock(&ftrace_lock);
5672
5673 return ret;
5674 }
5675
5676 #ifdef CONFIG_MODULES
5677
5678 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
5679
5680 static int referenced_filters(struct dyn_ftrace *rec)
5681 {
5682 struct ftrace_ops *ops;
5683 int cnt = 0;
5684
5685 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
5686 if (ops_references_rec(ops, rec))
5687 cnt++;
5688 }
5689
5690 return cnt;
5691 }
5692
5693 void ftrace_release_mod(struct module *mod)
5694 {
5695 struct dyn_ftrace *rec;
5696 struct ftrace_page **last_pg;
5697 struct ftrace_page *pg;
5698 int order;
5699
5700 mutex_lock(&ftrace_lock);
5701
5702 if (ftrace_disabled)
5703 goto out_unlock;
5704
5705 /*
5706 * Each module has its own ftrace_pages, remove
5707 * them from the list.
5708 */
5709 last_pg = &ftrace_pages_start;
5710 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
5711 rec = &pg->records[0];
5712 if (within_module_core(rec->ip, mod)) {
5713 /*
5714 * As core pages are first, the first
5715 * page should never be a module page.
5716 */
5717 if (WARN_ON(pg == ftrace_pages_start))
5718 goto out_unlock;
5719
5720 /* Check if we are deleting the last page */
5721 if (pg == ftrace_pages)
5722 ftrace_pages = next_to_ftrace_page(last_pg);
5723
5724 ftrace_update_tot_cnt -= pg->index;
5725 *last_pg = pg->next;
5726 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
5727 free_pages((unsigned long)pg->records, order);
5728 kfree(pg);
5729 } else
5730 last_pg = &pg->next;
5731 }
5732 out_unlock:
5733 mutex_unlock(&ftrace_lock);
5734 }
5735
5736 void ftrace_module_enable(struct module *mod)
5737 {
5738 struct dyn_ftrace *rec;
5739 struct ftrace_page *pg;
5740
5741 mutex_lock(&ftrace_lock);
5742
5743 if (ftrace_disabled)
5744 goto out_unlock;
5745
5746 /*
5747 * If the tracing is enabled, go ahead and enable the record.
5748 *
5749 * The reason not to enable the record immediatelly is the
5750 * inherent check of ftrace_make_nop/ftrace_make_call for
5751 * correct previous instructions. Making first the NOP
5752 * conversion puts the module to the correct state, thus
5753 * passing the ftrace_make_call check.
5754 *
5755 * We also delay this to after the module code already set the
5756 * text to read-only, as we now need to set it back to read-write
5757 * so that we can modify the text.
5758 */
5759 if (ftrace_start_up)
5760 ftrace_arch_code_modify_prepare();
5761
5762 do_for_each_ftrace_rec(pg, rec) {
5763 int cnt;
5764 /*
5765 * do_for_each_ftrace_rec() is a double loop.
5766 * module text shares the pg. If a record is
5767 * not part of this module, then skip this pg,
5768 * which the "break" will do.
5769 */
5770 if (!within_module_core(rec->ip, mod))
5771 break;
5772
5773 cnt = 0;
5774
5775 /*
5776 * When adding a module, we need to check if tracers are
5777 * currently enabled and if they are, and can trace this record,
5778 * we need to enable the module functions as well as update the
5779 * reference counts for those function records.
5780 */
5781 if (ftrace_start_up)
5782 cnt += referenced_filters(rec);
5783
5784 /* This clears FTRACE_FL_DISABLED */
5785 rec->flags = cnt;
5786
5787 if (ftrace_start_up && cnt) {
5788 int failed = __ftrace_replace_code(rec, 1);
5789 if (failed) {
5790 ftrace_bug(failed, rec);
5791 goto out_loop;
5792 }
5793 }
5794
5795 } while_for_each_ftrace_rec();
5796
5797 out_loop:
5798 if (ftrace_start_up)
5799 ftrace_arch_code_modify_post_process();
5800
5801 out_unlock:
5802 mutex_unlock(&ftrace_lock);
5803
5804 process_cached_mods(mod->name);
5805 }
5806
5807 void ftrace_module_init(struct module *mod)
5808 {
5809 if (ftrace_disabled || !mod->num_ftrace_callsites)
5810 return;
5811
5812 ftrace_process_locs(mod, mod->ftrace_callsites,
5813 mod->ftrace_callsites + mod->num_ftrace_callsites);
5814 }
5815 #endif /* CONFIG_MODULES */
5816
5817 void __init ftrace_free_init_mem(void)
5818 {
5819 unsigned long start = (unsigned long)(&__init_begin);
5820 unsigned long end = (unsigned long)(&__init_end);
5821 struct ftrace_page **last_pg = &ftrace_pages_start;
5822 struct ftrace_page *pg;
5823 struct dyn_ftrace *rec;
5824 struct dyn_ftrace key;
5825 int order;
5826
5827 key.ip = start;
5828 key.flags = end; /* overload flags, as it is unsigned long */
5829
5830 mutex_lock(&ftrace_lock);
5831
5832 for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
5833 if (end < pg->records[0].ip ||
5834 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
5835 continue;
5836 again:
5837 rec = bsearch(&key, pg->records, pg->index,
5838 sizeof(struct dyn_ftrace),
5839 ftrace_cmp_recs);
5840 if (!rec)
5841 continue;
5842 pg->index--;
5843 ftrace_update_tot_cnt--;
5844 if (!pg->index) {
5845 *last_pg = pg->next;
5846 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
5847 free_pages((unsigned long)pg->records, order);
5848 kfree(pg);
5849 pg = container_of(last_pg, struct ftrace_page, next);
5850 if (!(*last_pg))
5851 ftrace_pages = pg;
5852 continue;
5853 }
5854 memmove(rec, rec + 1,
5855 (pg->index - (rec - pg->records)) * sizeof(*rec));
5856 /* More than one function may be in this block */
5857 goto again;
5858 }
5859 mutex_unlock(&ftrace_lock);
5860 }
5861
5862 void __init ftrace_init(void)
5863 {
5864 extern unsigned long __start_mcount_loc[];
5865 extern unsigned long __stop_mcount_loc[];
5866 unsigned long count, flags;
5867 int ret;
5868
5869 local_irq_save(flags);
5870 ret = ftrace_dyn_arch_init();
5871 local_irq_restore(flags);
5872 if (ret)
5873 goto failed;
5874
5875 count = __stop_mcount_loc - __start_mcount_loc;
5876 if (!count) {
5877 pr_info("ftrace: No functions to be traced?\n");
5878 goto failed;
5879 }
5880
5881 pr_info("ftrace: allocating %ld entries in %ld pages\n",
5882 count, count / ENTRIES_PER_PAGE + 1);
5883
5884 last_ftrace_enabled = ftrace_enabled = 1;
5885
5886 ret = ftrace_process_locs(NULL,
5887 __start_mcount_loc,
5888 __stop_mcount_loc);
5889
5890 set_ftrace_early_filters();
5891
5892 return;
5893 failed:
5894 ftrace_disabled = 1;
5895 }
5896
5897 /* Do nothing if arch does not support this */
5898 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
5899 {
5900 }
5901
5902 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5903 {
5904 arch_ftrace_update_trampoline(ops);
5905 }
5906
5907 void ftrace_init_trace_array(struct trace_array *tr)
5908 {
5909 INIT_LIST_HEAD(&tr->func_probes);
5910 INIT_LIST_HEAD(&tr->mod_trace);
5911 INIT_LIST_HEAD(&tr->mod_notrace);
5912 }
5913 #else
5914
5915 static struct ftrace_ops global_ops = {
5916 .func = ftrace_stub,
5917 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
5918 FTRACE_OPS_FL_INITIALIZED |
5919 FTRACE_OPS_FL_PID,
5920 };
5921
5922 static int __init ftrace_nodyn_init(void)
5923 {
5924 ftrace_enabled = 1;
5925 return 0;
5926 }
5927 core_initcall(ftrace_nodyn_init);
5928
5929 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
5930 static inline void ftrace_startup_enable(int command) { }
5931 static inline void ftrace_startup_all(int command) { }
5932 /* Keep as macros so we do not need to define the commands */
5933 # define ftrace_startup(ops, command) \
5934 ({ \
5935 int ___ret = __register_ftrace_function(ops); \
5936 if (!___ret) \
5937 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
5938 ___ret; \
5939 })
5940 # define ftrace_shutdown(ops, command) \
5941 ({ \
5942 int ___ret = __unregister_ftrace_function(ops); \
5943 if (!___ret) \
5944 (ops)->flags &= ~FTRACE_OPS_FL_ENABLED; \
5945 ___ret; \
5946 })
5947
5948 # define ftrace_startup_sysctl() do { } while (0)
5949 # define ftrace_shutdown_sysctl() do { } while (0)
5950
5951 static inline int
5952 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
5953 {
5954 return 1;
5955 }
5956
5957 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5958 {
5959 }
5960
5961 #endif /* CONFIG_DYNAMIC_FTRACE */
5962
5963 __init void ftrace_init_global_array_ops(struct trace_array *tr)
5964 {
5965 tr->ops = &global_ops;
5966 tr->ops->private = tr;
5967 ftrace_init_trace_array(tr);
5968 }
5969
5970 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
5971 {
5972 /* If we filter on pids, update to use the pid function */
5973 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
5974 if (WARN_ON(tr->ops->func != ftrace_stub))
5975 printk("ftrace ops had %pS for function\n",
5976 tr->ops->func);
5977 }
5978 tr->ops->func = func;
5979 tr->ops->private = tr;
5980 }
5981
5982 void ftrace_reset_array_ops(struct trace_array *tr)
5983 {
5984 tr->ops->func = ftrace_stub;
5985 }
5986
5987 static inline void
5988 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5989 struct ftrace_ops *ignored, struct pt_regs *regs)
5990 {
5991 struct ftrace_ops *op;
5992 int bit;
5993
5994 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5995 if (bit < 0)
5996 return;
5997
5998 /*
5999 * Some of the ops may be dynamically allocated,
6000 * they must be freed after a synchronize_sched().
6001 */
6002 preempt_disable_notrace();
6003
6004 do_for_each_ftrace_op(op, ftrace_ops_list) {
6005 /*
6006 * Check the following for each ops before calling their func:
6007 * if RCU flag is set, then rcu_is_watching() must be true
6008 * if PER_CPU is set, then ftrace_function_local_disable()
6009 * must be false
6010 * Otherwise test if the ip matches the ops filter
6011 *
6012 * If any of the above fails then the op->func() is not executed.
6013 */
6014 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
6015 (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
6016 !ftrace_function_local_disabled(op)) &&
6017 ftrace_ops_test(op, ip, regs)) {
6018
6019 if (FTRACE_WARN_ON(!op->func)) {
6020 pr_warn("op=%p %pS\n", op, op);
6021 goto out;
6022 }
6023 op->func(ip, parent_ip, op, regs);
6024 }
6025 } while_for_each_ftrace_op(op);
6026 out:
6027 preempt_enable_notrace();
6028 trace_clear_recursion(bit);
6029 }
6030
6031 /*
6032 * Some archs only support passing ip and parent_ip. Even though
6033 * the list function ignores the op parameter, we do not want any
6034 * C side effects, where a function is called without the caller
6035 * sending a third parameter.
6036 * Archs are to support both the regs and ftrace_ops at the same time.
6037 * If they support ftrace_ops, it is assumed they support regs.
6038 * If call backs want to use regs, they must either check for regs
6039 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
6040 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
6041 * An architecture can pass partial regs with ftrace_ops and still
6042 * set the ARCH_SUPPORTS_FTRACE_OPS.
6043 */
6044 #if ARCH_SUPPORTS_FTRACE_OPS
6045 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6046 struct ftrace_ops *op, struct pt_regs *regs)
6047 {
6048 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
6049 }
6050 #else
6051 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
6052 {
6053 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
6054 }
6055 #endif
6056
6057 /*
6058 * If there's only one function registered but it does not support
6059 * recursion, needs RCU protection and/or requires per cpu handling, then
6060 * this function will be called by the mcount trampoline.
6061 */
6062 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
6063 struct ftrace_ops *op, struct pt_regs *regs)
6064 {
6065 int bit;
6066
6067 if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
6068 return;
6069
6070 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6071 if (bit < 0)
6072 return;
6073
6074 preempt_disable_notrace();
6075
6076 if (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
6077 !ftrace_function_local_disabled(op)) {
6078 op->func(ip, parent_ip, op, regs);
6079 }
6080
6081 preempt_enable_notrace();
6082 trace_clear_recursion(bit);
6083 }
6084
6085 /**
6086 * ftrace_ops_get_func - get the function a trampoline should call
6087 * @ops: the ops to get the function for
6088 *
6089 * Normally the mcount trampoline will call the ops->func, but there
6090 * are times that it should not. For example, if the ops does not
6091 * have its own recursion protection, then it should call the
6092 * ftrace_ops_assist_func() instead.
6093 *
6094 * Returns the function that the trampoline should call for @ops.
6095 */
6096 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
6097 {
6098 /*
6099 * If the function does not handle recursion, needs to be RCU safe,
6100 * or does per cpu logic, then we need to call the assist handler.
6101 */
6102 if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
6103 ops->flags & (FTRACE_OPS_FL_RCU | FTRACE_OPS_FL_PER_CPU))
6104 return ftrace_ops_assist_func;
6105
6106 return ops->func;
6107 }
6108
6109 static void
6110 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
6111 struct task_struct *prev, struct task_struct *next)
6112 {
6113 struct trace_array *tr = data;
6114 struct trace_pid_list *pid_list;
6115
6116 pid_list = rcu_dereference_sched(tr->function_pids);
6117
6118 this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6119 trace_ignore_this_task(pid_list, next));
6120 }
6121
6122 static void
6123 ftrace_pid_follow_sched_process_fork(void *data,
6124 struct task_struct *self,
6125 struct task_struct *task)
6126 {
6127 struct trace_pid_list *pid_list;
6128 struct trace_array *tr = data;
6129
6130 pid_list = rcu_dereference_sched(tr->function_pids);
6131 trace_filter_add_remove_task(pid_list, self, task);
6132 }
6133
6134 static void
6135 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
6136 {
6137 struct trace_pid_list *pid_list;
6138 struct trace_array *tr = data;
6139
6140 pid_list = rcu_dereference_sched(tr->function_pids);
6141 trace_filter_add_remove_task(pid_list, NULL, task);
6142 }
6143
6144 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
6145 {
6146 if (enable) {
6147 register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6148 tr);
6149 register_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6150 tr);
6151 } else {
6152 unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6153 tr);
6154 unregister_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6155 tr);
6156 }
6157 }
6158
6159 static void clear_ftrace_pids(struct trace_array *tr)
6160 {
6161 struct trace_pid_list *pid_list;
6162 int cpu;
6163
6164 pid_list = rcu_dereference_protected(tr->function_pids,
6165 lockdep_is_held(&ftrace_lock));
6166 if (!pid_list)
6167 return;
6168
6169 unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6170
6171 for_each_possible_cpu(cpu)
6172 per_cpu_ptr(tr->trace_buffer.data, cpu)->ftrace_ignore_pid = false;
6173
6174 rcu_assign_pointer(tr->function_pids, NULL);
6175
6176 /* Wait till all users are no longer using pid filtering */
6177 synchronize_sched();
6178
6179 trace_free_pid_list(pid_list);
6180 }
6181
6182 void ftrace_clear_pids(struct trace_array *tr)
6183 {
6184 mutex_lock(&ftrace_lock);
6185
6186 clear_ftrace_pids(tr);
6187
6188 mutex_unlock(&ftrace_lock);
6189 }
6190
6191 static void ftrace_pid_reset(struct trace_array *tr)
6192 {
6193 mutex_lock(&ftrace_lock);
6194 clear_ftrace_pids(tr);
6195
6196 ftrace_update_pid_func();
6197 ftrace_startup_all(0);
6198
6199 mutex_unlock(&ftrace_lock);
6200 }
6201
6202 /* Greater than any max PID */
6203 #define FTRACE_NO_PIDS (void *)(PID_MAX_LIMIT + 1)
6204
6205 static void *fpid_start(struct seq_file *m, loff_t *pos)
6206 __acquires(RCU)
6207 {
6208 struct trace_pid_list *pid_list;
6209 struct trace_array *tr = m->private;
6210
6211 mutex_lock(&ftrace_lock);
6212 rcu_read_lock_sched();
6213
6214 pid_list = rcu_dereference_sched(tr->function_pids);
6215
6216 if (!pid_list)
6217 return !(*pos) ? FTRACE_NO_PIDS : NULL;
6218
6219 return trace_pid_start(pid_list, pos);
6220 }
6221
6222 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
6223 {
6224 struct trace_array *tr = m->private;
6225 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
6226
6227 if (v == FTRACE_NO_PIDS)
6228 return NULL;
6229
6230 return trace_pid_next(pid_list, v, pos);
6231 }
6232
6233 static void fpid_stop(struct seq_file *m, void *p)
6234 __releases(RCU)
6235 {
6236 rcu_read_unlock_sched();
6237 mutex_unlock(&ftrace_lock);
6238 }
6239
6240 static int fpid_show(struct seq_file *m, void *v)
6241 {
6242 if (v == FTRACE_NO_PIDS) {
6243 seq_puts(m, "no pid\n");
6244 return 0;
6245 }
6246
6247 return trace_pid_show(m, v);
6248 }
6249
6250 static const struct seq_operations ftrace_pid_sops = {
6251 .start = fpid_start,
6252 .next = fpid_next,
6253 .stop = fpid_stop,
6254 .show = fpid_show,
6255 };
6256
6257 static int
6258 ftrace_pid_open(struct inode *inode, struct file *file)
6259 {
6260 struct trace_array *tr = inode->i_private;
6261 struct seq_file *m;
6262 int ret = 0;
6263
6264 if (trace_array_get(tr) < 0)
6265 return -ENODEV;
6266
6267 if ((file->f_mode & FMODE_WRITE) &&
6268 (file->f_flags & O_TRUNC))
6269 ftrace_pid_reset(tr);
6270
6271 ret = seq_open(file, &ftrace_pid_sops);
6272 if (ret < 0) {
6273 trace_array_put(tr);
6274 } else {
6275 m = file->private_data;
6276 /* copy tr over to seq ops */
6277 m->private = tr;
6278 }
6279
6280 return ret;
6281 }
6282
6283 static void ignore_task_cpu(void *data)
6284 {
6285 struct trace_array *tr = data;
6286 struct trace_pid_list *pid_list;
6287
6288 /*
6289 * This function is called by on_each_cpu() while the
6290 * event_mutex is held.
6291 */
6292 pid_list = rcu_dereference_protected(tr->function_pids,
6293 mutex_is_locked(&ftrace_lock));
6294
6295 this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6296 trace_ignore_this_task(pid_list, current));
6297 }
6298
6299 static ssize_t
6300 ftrace_pid_write(struct file *filp, const char __user *ubuf,
6301 size_t cnt, loff_t *ppos)
6302 {
6303 struct seq_file *m = filp->private_data;
6304 struct trace_array *tr = m->private;
6305 struct trace_pid_list *filtered_pids = NULL;
6306 struct trace_pid_list *pid_list;
6307 ssize_t ret;
6308
6309 if (!cnt)
6310 return 0;
6311
6312 mutex_lock(&ftrace_lock);
6313
6314 filtered_pids = rcu_dereference_protected(tr->function_pids,
6315 lockdep_is_held(&ftrace_lock));
6316
6317 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
6318 if (ret < 0)
6319 goto out;
6320
6321 rcu_assign_pointer(tr->function_pids, pid_list);
6322
6323 if (filtered_pids) {
6324 synchronize_sched();
6325 trace_free_pid_list(filtered_pids);
6326 } else if (pid_list) {
6327 /* Register a probe to set whether to ignore the tracing of a task */
6328 register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6329 }
6330
6331 /*
6332 * Ignoring of pids is done at task switch. But we have to
6333 * check for those tasks that are currently running.
6334 * Always do this in case a pid was appended or removed.
6335 */
6336 on_each_cpu(ignore_task_cpu, tr, 1);
6337
6338 ftrace_update_pid_func();
6339 ftrace_startup_all(0);
6340 out:
6341 mutex_unlock(&ftrace_lock);
6342
6343 if (ret > 0)
6344 *ppos += ret;
6345
6346 return ret;
6347 }
6348
6349 static int
6350 ftrace_pid_release(struct inode *inode, struct file *file)
6351 {
6352 struct trace_array *tr = inode->i_private;
6353
6354 trace_array_put(tr);
6355
6356 return seq_release(inode, file);
6357 }
6358
6359 static const struct file_operations ftrace_pid_fops = {
6360 .open = ftrace_pid_open,
6361 .write = ftrace_pid_write,
6362 .read = seq_read,
6363 .llseek = tracing_lseek,
6364 .release = ftrace_pid_release,
6365 };
6366
6367 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
6368 {
6369 trace_create_file("set_ftrace_pid", 0644, d_tracer,
6370 tr, &ftrace_pid_fops);
6371 }
6372
6373 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
6374 struct dentry *d_tracer)
6375 {
6376 /* Only the top level directory has the dyn_tracefs and profile */
6377 WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
6378
6379 ftrace_init_dyn_tracefs(d_tracer);
6380 ftrace_profile_tracefs(d_tracer);
6381 }
6382
6383 /**
6384 * ftrace_kill - kill ftrace
6385 *
6386 * This function should be used by panic code. It stops ftrace
6387 * but in a not so nice way. If you need to simply kill ftrace
6388 * from a non-atomic section, use ftrace_kill.
6389 */
6390 void ftrace_kill(void)
6391 {
6392 ftrace_disabled = 1;
6393 ftrace_enabled = 0;
6394 clear_ftrace_function();
6395 }
6396
6397 /**
6398 * Test if ftrace is dead or not.
6399 */
6400 int ftrace_is_dead(void)
6401 {
6402 return ftrace_disabled;
6403 }
6404
6405 /**
6406 * register_ftrace_function - register a function for profiling
6407 * @ops - ops structure that holds the function for profiling.
6408 *
6409 * Register a function to be called by all functions in the
6410 * kernel.
6411 *
6412 * Note: @ops->func and all the functions it calls must be labeled
6413 * with "notrace", otherwise it will go into a
6414 * recursive loop.
6415 */
6416 int register_ftrace_function(struct ftrace_ops *ops)
6417 {
6418 int ret = -1;
6419
6420 ftrace_ops_init(ops);
6421
6422 mutex_lock(&ftrace_lock);
6423
6424 ret = ftrace_startup(ops, 0);
6425
6426 mutex_unlock(&ftrace_lock);
6427
6428 return ret;
6429 }
6430 EXPORT_SYMBOL_GPL(register_ftrace_function);
6431
6432 /**
6433 * unregister_ftrace_function - unregister a function for profiling.
6434 * @ops - ops structure that holds the function to unregister
6435 *
6436 * Unregister a function that was added to be called by ftrace profiling.
6437 */
6438 int unregister_ftrace_function(struct ftrace_ops *ops)
6439 {
6440 int ret;
6441
6442 mutex_lock(&ftrace_lock);
6443 ret = ftrace_shutdown(ops, 0);
6444 mutex_unlock(&ftrace_lock);
6445
6446 return ret;
6447 }
6448 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
6449
6450 int
6451 ftrace_enable_sysctl(struct ctl_table *table, int write,
6452 void __user *buffer, size_t *lenp,
6453 loff_t *ppos)
6454 {
6455 int ret = -ENODEV;
6456
6457 mutex_lock(&ftrace_lock);
6458
6459 if (unlikely(ftrace_disabled))
6460 goto out;
6461
6462 ret = proc_dointvec(table, write, buffer, lenp, ppos);
6463
6464 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
6465 goto out;
6466
6467 last_ftrace_enabled = !!ftrace_enabled;
6468
6469 if (ftrace_enabled) {
6470
6471 /* we are starting ftrace again */
6472 if (rcu_dereference_protected(ftrace_ops_list,
6473 lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
6474 update_ftrace_function();
6475
6476 ftrace_startup_sysctl();
6477
6478 } else {
6479 /* stopping ftrace calls (just send to ftrace_stub) */
6480 ftrace_trace_function = ftrace_stub;
6481
6482 ftrace_shutdown_sysctl();
6483 }
6484
6485 out:
6486 mutex_unlock(&ftrace_lock);
6487 return ret;
6488 }
6489
6490 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6491
6492 static struct ftrace_ops graph_ops = {
6493 .func = ftrace_stub,
6494 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
6495 FTRACE_OPS_FL_INITIALIZED |
6496 FTRACE_OPS_FL_PID |
6497 FTRACE_OPS_FL_STUB,
6498 #ifdef FTRACE_GRAPH_TRAMP_ADDR
6499 .trampoline = FTRACE_GRAPH_TRAMP_ADDR,
6500 /* trampoline_size is only needed for dynamically allocated tramps */
6501 #endif
6502 ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
6503 };
6504
6505 void ftrace_graph_sleep_time_control(bool enable)
6506 {
6507 fgraph_sleep_time = enable;
6508 }
6509
6510 void ftrace_graph_graph_time_control(bool enable)
6511 {
6512 fgraph_graph_time = enable;
6513 }
6514
6515 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
6516 {
6517 return 0;
6518 }
6519
6520 /* The callbacks that hook a function */
6521 trace_func_graph_ret_t ftrace_graph_return =
6522 (trace_func_graph_ret_t)ftrace_stub;
6523 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
6524 static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
6525
6526 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
6527 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
6528 {
6529 int i;
6530 int ret = 0;
6531 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
6532 struct task_struct *g, *t;
6533
6534 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
6535 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
6536 * sizeof(struct ftrace_ret_stack),
6537 GFP_KERNEL);
6538 if (!ret_stack_list[i]) {
6539 start = 0;
6540 end = i;
6541 ret = -ENOMEM;
6542 goto free;
6543 }
6544 }
6545
6546 read_lock(&tasklist_lock);
6547 do_each_thread(g, t) {
6548 if (start == end) {
6549 ret = -EAGAIN;
6550 goto unlock;
6551 }
6552
6553 if (t->ret_stack == NULL) {
6554 atomic_set(&t->tracing_graph_pause, 0);
6555 atomic_set(&t->trace_overrun, 0);
6556 t->curr_ret_stack = -1;
6557 /* Make sure the tasks see the -1 first: */
6558 smp_wmb();
6559 t->ret_stack = ret_stack_list[start++];
6560 }
6561 } while_each_thread(g, t);
6562
6563 unlock:
6564 read_unlock(&tasklist_lock);
6565 free:
6566 for (i = start; i < end; i++)
6567 kfree(ret_stack_list[i]);
6568 return ret;
6569 }
6570
6571 static void
6572 ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
6573 struct task_struct *prev, struct task_struct *next)
6574 {
6575 unsigned long long timestamp;
6576 int index;
6577
6578 /*
6579 * Does the user want to count the time a function was asleep.
6580 * If so, do not update the time stamps.
6581 */
6582 if (fgraph_sleep_time)
6583 return;
6584
6585 timestamp = trace_clock_local();
6586
6587 prev->ftrace_timestamp = timestamp;
6588
6589 /* only process tasks that we timestamped */
6590 if (!next->ftrace_timestamp)
6591 return;
6592
6593 /*
6594 * Update all the counters in next to make up for the
6595 * time next was sleeping.
6596 */
6597 timestamp -= next->ftrace_timestamp;
6598
6599 for (index = next->curr_ret_stack; index >= 0; index--)
6600 next->ret_stack[index].calltime += timestamp;
6601 }
6602
6603 /* Allocate a return stack for each task */
6604 static int start_graph_tracing(void)
6605 {
6606 struct ftrace_ret_stack **ret_stack_list;
6607 int ret, cpu;
6608
6609 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
6610 sizeof(struct ftrace_ret_stack *),
6611 GFP_KERNEL);
6612
6613 if (!ret_stack_list)
6614 return -ENOMEM;
6615
6616 /* The cpu_boot init_task->ret_stack will never be freed */
6617 for_each_online_cpu(cpu) {
6618 if (!idle_task(cpu)->ret_stack)
6619 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
6620 }
6621
6622 do {
6623 ret = alloc_retstack_tasklist(ret_stack_list);
6624 } while (ret == -EAGAIN);
6625
6626 if (!ret) {
6627 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
6628 if (ret)
6629 pr_info("ftrace_graph: Couldn't activate tracepoint"
6630 " probe to kernel_sched_switch\n");
6631 }
6632
6633 kfree(ret_stack_list);
6634 return ret;
6635 }
6636
6637 /*
6638 * Hibernation protection.
6639 * The state of the current task is too much unstable during
6640 * suspend/restore to disk. We want to protect against that.
6641 */
6642 static int
6643 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
6644 void *unused)
6645 {
6646 switch (state) {
6647 case PM_HIBERNATION_PREPARE:
6648 pause_graph_tracing();
6649 break;
6650
6651 case PM_POST_HIBERNATION:
6652 unpause_graph_tracing();
6653 break;
6654 }
6655 return NOTIFY_DONE;
6656 }
6657
6658 static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
6659 {
6660 if (!ftrace_ops_test(&global_ops, trace->func, NULL))
6661 return 0;
6662 return __ftrace_graph_entry(trace);
6663 }
6664
6665 /*
6666 * The function graph tracer should only trace the functions defined
6667 * by set_ftrace_filter and set_ftrace_notrace. If another function
6668 * tracer ops is registered, the graph tracer requires testing the
6669 * function against the global ops, and not just trace any function
6670 * that any ftrace_ops registered.
6671 */
6672 static void update_function_graph_func(void)
6673 {
6674 struct ftrace_ops *op;
6675 bool do_test = false;
6676
6677 /*
6678 * The graph and global ops share the same set of functions
6679 * to test. If any other ops is on the list, then
6680 * the graph tracing needs to test if its the function
6681 * it should call.
6682 */
6683 do_for_each_ftrace_op(op, ftrace_ops_list) {
6684 if (op != &global_ops && op != &graph_ops &&
6685 op != &ftrace_list_end) {
6686 do_test = true;
6687 /* in double loop, break out with goto */
6688 goto out;
6689 }
6690 } while_for_each_ftrace_op(op);
6691 out:
6692 if (do_test)
6693 ftrace_graph_entry = ftrace_graph_entry_test;
6694 else
6695 ftrace_graph_entry = __ftrace_graph_entry;
6696 }
6697
6698 static struct notifier_block ftrace_suspend_notifier = {
6699 .notifier_call = ftrace_suspend_notifier_call,
6700 };
6701
6702 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
6703 trace_func_graph_ent_t entryfunc)
6704 {
6705 int ret = 0;
6706
6707 mutex_lock(&ftrace_lock);
6708
6709 /* we currently allow only one tracer registered at a time */
6710 if (ftrace_graph_active) {
6711 ret = -EBUSY;
6712 goto out;
6713 }
6714
6715 register_pm_notifier(&ftrace_suspend_notifier);
6716
6717 ftrace_graph_active++;
6718 ret = start_graph_tracing();
6719 if (ret) {
6720 ftrace_graph_active--;
6721 goto out;
6722 }
6723
6724 ftrace_graph_return = retfunc;
6725
6726 /*
6727 * Update the indirect function to the entryfunc, and the
6728 * function that gets called to the entry_test first. Then
6729 * call the update fgraph entry function to determine if
6730 * the entryfunc should be called directly or not.
6731 */
6732 __ftrace_graph_entry = entryfunc;
6733 ftrace_graph_entry = ftrace_graph_entry_test;
6734 update_function_graph_func();
6735
6736 ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
6737 out:
6738 mutex_unlock(&ftrace_lock);
6739 return ret;
6740 }
6741
6742 void unregister_ftrace_graph(void)
6743 {
6744 mutex_lock(&ftrace_lock);
6745
6746 if (unlikely(!ftrace_graph_active))
6747 goto out;
6748
6749 ftrace_graph_active--;
6750 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
6751 ftrace_graph_entry = ftrace_graph_entry_stub;
6752 __ftrace_graph_entry = ftrace_graph_entry_stub;
6753 ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
6754 unregister_pm_notifier(&ftrace_suspend_notifier);
6755 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
6756
6757 #ifdef CONFIG_DYNAMIC_FTRACE
6758 /*
6759 * Function graph does not allocate the trampoline, but
6760 * other global_ops do. We need to reset the ALLOC_TRAMP flag
6761 * if one was used.
6762 */
6763 global_ops.trampoline = save_global_trampoline;
6764 if (save_global_flags & FTRACE_OPS_FL_ALLOC_TRAMP)
6765 global_ops.flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
6766 #endif
6767
6768 out:
6769 mutex_unlock(&ftrace_lock);
6770 }
6771
6772 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
6773
6774 static void
6775 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
6776 {
6777 atomic_set(&t->tracing_graph_pause, 0);
6778 atomic_set(&t->trace_overrun, 0);
6779 t->ftrace_timestamp = 0;
6780 /* make curr_ret_stack visible before we add the ret_stack */
6781 smp_wmb();
6782 t->ret_stack = ret_stack;
6783 }
6784
6785 /*
6786 * Allocate a return stack for the idle task. May be the first
6787 * time through, or it may be done by CPU hotplug online.
6788 */
6789 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
6790 {
6791 t->curr_ret_stack = -1;
6792 /*
6793 * The idle task has no parent, it either has its own
6794 * stack or no stack at all.
6795 */
6796 if (t->ret_stack)
6797 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
6798
6799 if (ftrace_graph_active) {
6800 struct ftrace_ret_stack *ret_stack;
6801
6802 ret_stack = per_cpu(idle_ret_stack, cpu);
6803 if (!ret_stack) {
6804 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6805 * sizeof(struct ftrace_ret_stack),
6806 GFP_KERNEL);
6807 if (!ret_stack)
6808 return;
6809 per_cpu(idle_ret_stack, cpu) = ret_stack;
6810 }
6811 graph_init_task(t, ret_stack);
6812 }
6813 }
6814
6815 /* Allocate a return stack for newly created task */
6816 void ftrace_graph_init_task(struct task_struct *t)
6817 {
6818 /* Make sure we do not use the parent ret_stack */
6819 t->ret_stack = NULL;
6820 t->curr_ret_stack = -1;
6821
6822 if (ftrace_graph_active) {
6823 struct ftrace_ret_stack *ret_stack;
6824
6825 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6826 * sizeof(struct ftrace_ret_stack),
6827 GFP_KERNEL);
6828 if (!ret_stack)
6829 return;
6830 graph_init_task(t, ret_stack);
6831 }
6832 }
6833
6834 void ftrace_graph_exit_task(struct task_struct *t)
6835 {
6836 struct ftrace_ret_stack *ret_stack = t->ret_stack;
6837
6838 t->ret_stack = NULL;
6839 /* NULL must become visible to IRQs before we free it: */
6840 barrier();
6841
6842 kfree(ret_stack);
6843 }
6844 #endif