]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/trace/ftrace.c
Merge remote-tracking branches 'asoc/fix/dpcm', 'asoc/fix/imx', 'asoc/fix/msm8916...
[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 (index >= 0 && index < FTRACE_RETFUNC_DEPTH)
893 current->ret_stack[index].subtime = 0;
894
895 return 1;
896 }
897
898 static void profile_graph_return(struct ftrace_graph_ret *trace)
899 {
900 struct ftrace_profile_stat *stat;
901 unsigned long long calltime;
902 struct ftrace_profile *rec;
903 unsigned long flags;
904
905 local_irq_save(flags);
906 stat = this_cpu_ptr(&ftrace_profile_stats);
907 if (!stat->hash || !ftrace_profile_enabled)
908 goto out;
909
910 /* If the calltime was zero'd ignore it */
911 if (!trace->calltime)
912 goto out;
913
914 calltime = trace->rettime - trace->calltime;
915
916 if (!fgraph_graph_time) {
917 int index;
918
919 index = trace->depth;
920
921 /* Append this call time to the parent time to subtract */
922 if (index)
923 current->ret_stack[index - 1].subtime += calltime;
924
925 if (current->ret_stack[index].subtime < calltime)
926 calltime -= current->ret_stack[index].subtime;
927 else
928 calltime = 0;
929 }
930
931 rec = ftrace_find_profiled_func(stat, trace->func);
932 if (rec) {
933 rec->time += calltime;
934 rec->time_squared += calltime * calltime;
935 }
936
937 out:
938 local_irq_restore(flags);
939 }
940
941 static int register_ftrace_profiler(void)
942 {
943 return register_ftrace_graph(&profile_graph_return,
944 &profile_graph_entry);
945 }
946
947 static void unregister_ftrace_profiler(void)
948 {
949 unregister_ftrace_graph();
950 }
951 #else
952 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
953 .func = function_profile_call,
954 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
955 INIT_OPS_HASH(ftrace_profile_ops)
956 };
957
958 static int register_ftrace_profiler(void)
959 {
960 return register_ftrace_function(&ftrace_profile_ops);
961 }
962
963 static void unregister_ftrace_profiler(void)
964 {
965 unregister_ftrace_function(&ftrace_profile_ops);
966 }
967 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
968
969 static ssize_t
970 ftrace_profile_write(struct file *filp, const char __user *ubuf,
971 size_t cnt, loff_t *ppos)
972 {
973 unsigned long val;
974 int ret;
975
976 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
977 if (ret)
978 return ret;
979
980 val = !!val;
981
982 mutex_lock(&ftrace_profile_lock);
983 if (ftrace_profile_enabled ^ val) {
984 if (val) {
985 ret = ftrace_profile_init();
986 if (ret < 0) {
987 cnt = ret;
988 goto out;
989 }
990
991 ret = register_ftrace_profiler();
992 if (ret < 0) {
993 cnt = ret;
994 goto out;
995 }
996 ftrace_profile_enabled = 1;
997 } else {
998 ftrace_profile_enabled = 0;
999 /*
1000 * unregister_ftrace_profiler calls stop_machine
1001 * so this acts like an synchronize_sched.
1002 */
1003 unregister_ftrace_profiler();
1004 }
1005 }
1006 out:
1007 mutex_unlock(&ftrace_profile_lock);
1008
1009 *ppos += cnt;
1010
1011 return cnt;
1012 }
1013
1014 static ssize_t
1015 ftrace_profile_read(struct file *filp, char __user *ubuf,
1016 size_t cnt, loff_t *ppos)
1017 {
1018 char buf[64]; /* big enough to hold a number */
1019 int r;
1020
1021 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1022 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1023 }
1024
1025 static const struct file_operations ftrace_profile_fops = {
1026 .open = tracing_open_generic,
1027 .read = ftrace_profile_read,
1028 .write = ftrace_profile_write,
1029 .llseek = default_llseek,
1030 };
1031
1032 /* used to initialize the real stat files */
1033 static struct tracer_stat function_stats __initdata = {
1034 .name = "functions",
1035 .stat_start = function_stat_start,
1036 .stat_next = function_stat_next,
1037 .stat_cmp = function_stat_cmp,
1038 .stat_headers = function_stat_headers,
1039 .stat_show = function_stat_show
1040 };
1041
1042 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1043 {
1044 struct ftrace_profile_stat *stat;
1045 struct dentry *entry;
1046 char *name;
1047 int ret;
1048 int cpu;
1049
1050 for_each_possible_cpu(cpu) {
1051 stat = &per_cpu(ftrace_profile_stats, cpu);
1052
1053 name = kasprintf(GFP_KERNEL, "function%d", cpu);
1054 if (!name) {
1055 /*
1056 * The files created are permanent, if something happens
1057 * we still do not free memory.
1058 */
1059 WARN(1,
1060 "Could not allocate stat file for cpu %d\n",
1061 cpu);
1062 return;
1063 }
1064 stat->stat = function_stats;
1065 stat->stat.name = name;
1066 ret = register_stat_tracer(&stat->stat);
1067 if (ret) {
1068 WARN(1,
1069 "Could not register function stat for cpu %d\n",
1070 cpu);
1071 kfree(name);
1072 return;
1073 }
1074 }
1075
1076 entry = tracefs_create_file("function_profile_enabled", 0644,
1077 d_tracer, NULL, &ftrace_profile_fops);
1078 if (!entry)
1079 pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1080 }
1081
1082 #else /* CONFIG_FUNCTION_PROFILER */
1083 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1084 {
1085 }
1086 #endif /* CONFIG_FUNCTION_PROFILER */
1087
1088 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1089
1090 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1091 static int ftrace_graph_active;
1092 #else
1093 # define ftrace_graph_active 0
1094 #endif
1095
1096 #ifdef CONFIG_DYNAMIC_FTRACE
1097
1098 static struct ftrace_ops *removed_ops;
1099
1100 /*
1101 * Set when doing a global update, like enabling all recs or disabling them.
1102 * It is not set when just updating a single ftrace_ops.
1103 */
1104 static bool update_all_ops;
1105
1106 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1107 # error Dynamic ftrace depends on MCOUNT_RECORD
1108 #endif
1109
1110 struct ftrace_func_entry {
1111 struct hlist_node hlist;
1112 unsigned long ip;
1113 };
1114
1115 struct ftrace_func_probe {
1116 struct ftrace_probe_ops *probe_ops;
1117 struct ftrace_ops ops;
1118 struct trace_array *tr;
1119 struct list_head list;
1120 void *data;
1121 int ref;
1122 };
1123
1124 /*
1125 * We make these constant because no one should touch them,
1126 * but they are used as the default "empty hash", to avoid allocating
1127 * it all the time. These are in a read only section such that if
1128 * anyone does try to modify it, it will cause an exception.
1129 */
1130 static const struct hlist_head empty_buckets[1];
1131 static const struct ftrace_hash empty_hash = {
1132 .buckets = (struct hlist_head *)empty_buckets,
1133 };
1134 #define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1135
1136 static struct ftrace_ops global_ops = {
1137 .func = ftrace_stub,
1138 .local_hash.notrace_hash = EMPTY_HASH,
1139 .local_hash.filter_hash = EMPTY_HASH,
1140 INIT_OPS_HASH(global_ops)
1141 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
1142 FTRACE_OPS_FL_INITIALIZED |
1143 FTRACE_OPS_FL_PID,
1144 };
1145
1146 /*
1147 * This is used by __kernel_text_address() to return true if the
1148 * address is on a dynamically allocated trampoline that would
1149 * not return true for either core_kernel_text() or
1150 * is_module_text_address().
1151 */
1152 bool is_ftrace_trampoline(unsigned long addr)
1153 {
1154 struct ftrace_ops *op;
1155 bool ret = false;
1156
1157 /*
1158 * Some of the ops may be dynamically allocated,
1159 * they are freed after a synchronize_sched().
1160 */
1161 preempt_disable_notrace();
1162
1163 do_for_each_ftrace_op(op, ftrace_ops_list) {
1164 /*
1165 * This is to check for dynamically allocated trampolines.
1166 * Trampolines that are in kernel text will have
1167 * core_kernel_text() return true.
1168 */
1169 if (op->trampoline && op->trampoline_size)
1170 if (addr >= op->trampoline &&
1171 addr < op->trampoline + op->trampoline_size) {
1172 ret = true;
1173 goto out;
1174 }
1175 } while_for_each_ftrace_op(op);
1176
1177 out:
1178 preempt_enable_notrace();
1179
1180 return ret;
1181 }
1182
1183 struct ftrace_page {
1184 struct ftrace_page *next;
1185 struct dyn_ftrace *records;
1186 int index;
1187 int size;
1188 };
1189
1190 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1191 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1192
1193 /* estimate from running different kernels */
1194 #define NR_TO_INIT 10000
1195
1196 static struct ftrace_page *ftrace_pages_start;
1197 static struct ftrace_page *ftrace_pages;
1198
1199 static __always_inline unsigned long
1200 ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1201 {
1202 if (hash->size_bits > 0)
1203 return hash_long(ip, hash->size_bits);
1204
1205 return 0;
1206 }
1207
1208 /* Only use this function if ftrace_hash_empty() has already been tested */
1209 static __always_inline struct ftrace_func_entry *
1210 __ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1211 {
1212 unsigned long key;
1213 struct ftrace_func_entry *entry;
1214 struct hlist_head *hhd;
1215
1216 key = ftrace_hash_key(hash, ip);
1217 hhd = &hash->buckets[key];
1218
1219 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1220 if (entry->ip == ip)
1221 return entry;
1222 }
1223 return NULL;
1224 }
1225
1226 /**
1227 * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1228 * @hash: The hash to look at
1229 * @ip: The instruction pointer to test
1230 *
1231 * Search a given @hash to see if a given instruction pointer (@ip)
1232 * exists in it.
1233 *
1234 * Returns the entry that holds the @ip if found. NULL otherwise.
1235 */
1236 struct ftrace_func_entry *
1237 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1238 {
1239 if (ftrace_hash_empty(hash))
1240 return NULL;
1241
1242 return __ftrace_lookup_ip(hash, ip);
1243 }
1244
1245 static void __add_hash_entry(struct ftrace_hash *hash,
1246 struct ftrace_func_entry *entry)
1247 {
1248 struct hlist_head *hhd;
1249 unsigned long key;
1250
1251 key = ftrace_hash_key(hash, entry->ip);
1252 hhd = &hash->buckets[key];
1253 hlist_add_head(&entry->hlist, hhd);
1254 hash->count++;
1255 }
1256
1257 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1258 {
1259 struct ftrace_func_entry *entry;
1260
1261 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1262 if (!entry)
1263 return -ENOMEM;
1264
1265 entry->ip = ip;
1266 __add_hash_entry(hash, entry);
1267
1268 return 0;
1269 }
1270
1271 static void
1272 free_hash_entry(struct ftrace_hash *hash,
1273 struct ftrace_func_entry *entry)
1274 {
1275 hlist_del(&entry->hlist);
1276 kfree(entry);
1277 hash->count--;
1278 }
1279
1280 static void
1281 remove_hash_entry(struct ftrace_hash *hash,
1282 struct ftrace_func_entry *entry)
1283 {
1284 hlist_del_rcu(&entry->hlist);
1285 hash->count--;
1286 }
1287
1288 static void ftrace_hash_clear(struct ftrace_hash *hash)
1289 {
1290 struct hlist_head *hhd;
1291 struct hlist_node *tn;
1292 struct ftrace_func_entry *entry;
1293 int size = 1 << hash->size_bits;
1294 int i;
1295
1296 if (!hash->count)
1297 return;
1298
1299 for (i = 0; i < size; i++) {
1300 hhd = &hash->buckets[i];
1301 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1302 free_hash_entry(hash, entry);
1303 }
1304 FTRACE_WARN_ON(hash->count);
1305 }
1306
1307 static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1308 {
1309 list_del(&ftrace_mod->list);
1310 kfree(ftrace_mod->module);
1311 kfree(ftrace_mod->func);
1312 kfree(ftrace_mod);
1313 }
1314
1315 static void clear_ftrace_mod_list(struct list_head *head)
1316 {
1317 struct ftrace_mod_load *p, *n;
1318
1319 /* stack tracer isn't supported yet */
1320 if (!head)
1321 return;
1322
1323 mutex_lock(&ftrace_lock);
1324 list_for_each_entry_safe(p, n, head, list)
1325 free_ftrace_mod(p);
1326 mutex_unlock(&ftrace_lock);
1327 }
1328
1329 static void free_ftrace_hash(struct ftrace_hash *hash)
1330 {
1331 if (!hash || hash == EMPTY_HASH)
1332 return;
1333 ftrace_hash_clear(hash);
1334 kfree(hash->buckets);
1335 kfree(hash);
1336 }
1337
1338 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1339 {
1340 struct ftrace_hash *hash;
1341
1342 hash = container_of(rcu, struct ftrace_hash, rcu);
1343 free_ftrace_hash(hash);
1344 }
1345
1346 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1347 {
1348 if (!hash || hash == EMPTY_HASH)
1349 return;
1350 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1351 }
1352
1353 void ftrace_free_filter(struct ftrace_ops *ops)
1354 {
1355 ftrace_ops_init(ops);
1356 free_ftrace_hash(ops->func_hash->filter_hash);
1357 free_ftrace_hash(ops->func_hash->notrace_hash);
1358 }
1359
1360 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1361 {
1362 struct ftrace_hash *hash;
1363 int size;
1364
1365 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1366 if (!hash)
1367 return NULL;
1368
1369 size = 1 << size_bits;
1370 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1371
1372 if (!hash->buckets) {
1373 kfree(hash);
1374 return NULL;
1375 }
1376
1377 hash->size_bits = size_bits;
1378
1379 return hash;
1380 }
1381
1382
1383 static int ftrace_add_mod(struct trace_array *tr,
1384 const char *func, const char *module,
1385 int enable)
1386 {
1387 struct ftrace_mod_load *ftrace_mod;
1388 struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1389
1390 ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1391 if (!ftrace_mod)
1392 return -ENOMEM;
1393
1394 ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1395 ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1396 ftrace_mod->enable = enable;
1397
1398 if (!ftrace_mod->func || !ftrace_mod->module)
1399 goto out_free;
1400
1401 list_add(&ftrace_mod->list, mod_head);
1402
1403 return 0;
1404
1405 out_free:
1406 free_ftrace_mod(ftrace_mod);
1407
1408 return -ENOMEM;
1409 }
1410
1411 static struct ftrace_hash *
1412 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1413 {
1414 struct ftrace_func_entry *entry;
1415 struct ftrace_hash *new_hash;
1416 int size;
1417 int ret;
1418 int i;
1419
1420 new_hash = alloc_ftrace_hash(size_bits);
1421 if (!new_hash)
1422 return NULL;
1423
1424 if (hash)
1425 new_hash->flags = hash->flags;
1426
1427 /* Empty hash? */
1428 if (ftrace_hash_empty(hash))
1429 return new_hash;
1430
1431 size = 1 << hash->size_bits;
1432 for (i = 0; i < size; i++) {
1433 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1434 ret = add_hash_entry(new_hash, entry->ip);
1435 if (ret < 0)
1436 goto free_hash;
1437 }
1438 }
1439
1440 FTRACE_WARN_ON(new_hash->count != hash->count);
1441
1442 return new_hash;
1443
1444 free_hash:
1445 free_ftrace_hash(new_hash);
1446 return NULL;
1447 }
1448
1449 static void
1450 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1451 static void
1452 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1453
1454 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1455 struct ftrace_hash *new_hash);
1456
1457 static struct ftrace_hash *
1458 __ftrace_hash_move(struct ftrace_hash *src)
1459 {
1460 struct ftrace_func_entry *entry;
1461 struct hlist_node *tn;
1462 struct hlist_head *hhd;
1463 struct ftrace_hash *new_hash;
1464 int size = src->count;
1465 int bits = 0;
1466 int i;
1467
1468 /*
1469 * If the new source is empty, just return the empty_hash.
1470 */
1471 if (ftrace_hash_empty(src))
1472 return EMPTY_HASH;
1473
1474 /*
1475 * Make the hash size about 1/2 the # found
1476 */
1477 for (size /= 2; size; size >>= 1)
1478 bits++;
1479
1480 /* Don't allocate too much */
1481 if (bits > FTRACE_HASH_MAX_BITS)
1482 bits = FTRACE_HASH_MAX_BITS;
1483
1484 new_hash = alloc_ftrace_hash(bits);
1485 if (!new_hash)
1486 return NULL;
1487
1488 new_hash->flags = src->flags;
1489
1490 size = 1 << src->size_bits;
1491 for (i = 0; i < size; i++) {
1492 hhd = &src->buckets[i];
1493 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1494 remove_hash_entry(src, entry);
1495 __add_hash_entry(new_hash, entry);
1496 }
1497 }
1498
1499 return new_hash;
1500 }
1501
1502 static int
1503 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1504 struct ftrace_hash **dst, struct ftrace_hash *src)
1505 {
1506 struct ftrace_hash *new_hash;
1507 int ret;
1508
1509 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1510 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1511 return -EINVAL;
1512
1513 new_hash = __ftrace_hash_move(src);
1514 if (!new_hash)
1515 return -ENOMEM;
1516
1517 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1518 if (enable) {
1519 /* IPMODIFY should be updated only when filter_hash updating */
1520 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1521 if (ret < 0) {
1522 free_ftrace_hash(new_hash);
1523 return ret;
1524 }
1525 }
1526
1527 /*
1528 * Remove the current set, update the hash and add
1529 * them back.
1530 */
1531 ftrace_hash_rec_disable_modify(ops, enable);
1532
1533 rcu_assign_pointer(*dst, new_hash);
1534
1535 ftrace_hash_rec_enable_modify(ops, enable);
1536
1537 return 0;
1538 }
1539
1540 static bool hash_contains_ip(unsigned long ip,
1541 struct ftrace_ops_hash *hash)
1542 {
1543 /*
1544 * The function record is a match if it exists in the filter
1545 * hash and not in the notrace hash. Note, an emty hash is
1546 * considered a match for the filter hash, but an empty
1547 * notrace hash is considered not in the notrace hash.
1548 */
1549 return (ftrace_hash_empty(hash->filter_hash) ||
1550 __ftrace_lookup_ip(hash->filter_hash, ip)) &&
1551 (ftrace_hash_empty(hash->notrace_hash) ||
1552 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1553 }
1554
1555 /*
1556 * Test the hashes for this ops to see if we want to call
1557 * the ops->func or not.
1558 *
1559 * It's a match if the ip is in the ops->filter_hash or
1560 * the filter_hash does not exist or is empty,
1561 * AND
1562 * the ip is not in the ops->notrace_hash.
1563 *
1564 * This needs to be called with preemption disabled as
1565 * the hashes are freed with call_rcu_sched().
1566 */
1567 static int
1568 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1569 {
1570 struct ftrace_ops_hash hash;
1571 int ret;
1572
1573 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1574 /*
1575 * There's a small race when adding ops that the ftrace handler
1576 * that wants regs, may be called without them. We can not
1577 * allow that handler to be called if regs is NULL.
1578 */
1579 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1580 return 0;
1581 #endif
1582
1583 rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1584 rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1585
1586 if (hash_contains_ip(ip, &hash))
1587 ret = 1;
1588 else
1589 ret = 0;
1590
1591 return ret;
1592 }
1593
1594 /*
1595 * This is a double for. Do not use 'break' to break out of the loop,
1596 * you must use a goto.
1597 */
1598 #define do_for_each_ftrace_rec(pg, rec) \
1599 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1600 int _____i; \
1601 for (_____i = 0; _____i < pg->index; _____i++) { \
1602 rec = &pg->records[_____i];
1603
1604 #define while_for_each_ftrace_rec() \
1605 } \
1606 }
1607
1608
1609 static int ftrace_cmp_recs(const void *a, const void *b)
1610 {
1611 const struct dyn_ftrace *key = a;
1612 const struct dyn_ftrace *rec = b;
1613
1614 if (key->flags < rec->ip)
1615 return -1;
1616 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1617 return 1;
1618 return 0;
1619 }
1620
1621 /**
1622 * ftrace_location_range - return the first address of a traced location
1623 * if it touches the given ip range
1624 * @start: start of range to search.
1625 * @end: end of range to search (inclusive). @end points to the last byte
1626 * to check.
1627 *
1628 * Returns rec->ip if the related ftrace location is a least partly within
1629 * the given address range. That is, the first address of the instruction
1630 * that is either a NOP or call to the function tracer. It checks the ftrace
1631 * internal tables to determine if the address belongs or not.
1632 */
1633 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1634 {
1635 struct ftrace_page *pg;
1636 struct dyn_ftrace *rec;
1637 struct dyn_ftrace key;
1638
1639 key.ip = start;
1640 key.flags = end; /* overload flags, as it is unsigned long */
1641
1642 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1643 if (end < pg->records[0].ip ||
1644 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1645 continue;
1646 rec = bsearch(&key, pg->records, pg->index,
1647 sizeof(struct dyn_ftrace),
1648 ftrace_cmp_recs);
1649 if (rec)
1650 return rec->ip;
1651 }
1652
1653 return 0;
1654 }
1655
1656 /**
1657 * ftrace_location - return true if the ip giving is a traced location
1658 * @ip: the instruction pointer to check
1659 *
1660 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1661 * That is, the instruction that is either a NOP or call to
1662 * the function tracer. It checks the ftrace internal tables to
1663 * determine if the address belongs or not.
1664 */
1665 unsigned long ftrace_location(unsigned long ip)
1666 {
1667 return ftrace_location_range(ip, ip);
1668 }
1669
1670 /**
1671 * ftrace_text_reserved - return true if range contains an ftrace location
1672 * @start: start of range to search
1673 * @end: end of range to search (inclusive). @end points to the last byte to check.
1674 *
1675 * Returns 1 if @start and @end contains a ftrace location.
1676 * That is, the instruction that is either a NOP or call to
1677 * the function tracer. It checks the ftrace internal tables to
1678 * determine if the address belongs or not.
1679 */
1680 int ftrace_text_reserved(const void *start, const void *end)
1681 {
1682 unsigned long ret;
1683
1684 ret = ftrace_location_range((unsigned long)start,
1685 (unsigned long)end);
1686
1687 return (int)!!ret;
1688 }
1689
1690 /* Test if ops registered to this rec needs regs */
1691 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1692 {
1693 struct ftrace_ops *ops;
1694 bool keep_regs = false;
1695
1696 for (ops = ftrace_ops_list;
1697 ops != &ftrace_list_end; ops = ops->next) {
1698 /* pass rec in as regs to have non-NULL val */
1699 if (ftrace_ops_test(ops, rec->ip, rec)) {
1700 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1701 keep_regs = true;
1702 break;
1703 }
1704 }
1705 }
1706
1707 return keep_regs;
1708 }
1709
1710 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1711 int filter_hash,
1712 bool inc)
1713 {
1714 struct ftrace_hash *hash;
1715 struct ftrace_hash *other_hash;
1716 struct ftrace_page *pg;
1717 struct dyn_ftrace *rec;
1718 bool update = false;
1719 int count = 0;
1720 int all = false;
1721
1722 /* Only update if the ops has been registered */
1723 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1724 return false;
1725
1726 /*
1727 * In the filter_hash case:
1728 * If the count is zero, we update all records.
1729 * Otherwise we just update the items in the hash.
1730 *
1731 * In the notrace_hash case:
1732 * We enable the update in the hash.
1733 * As disabling notrace means enabling the tracing,
1734 * and enabling notrace means disabling, the inc variable
1735 * gets inversed.
1736 */
1737 if (filter_hash) {
1738 hash = ops->func_hash->filter_hash;
1739 other_hash = ops->func_hash->notrace_hash;
1740 if (ftrace_hash_empty(hash))
1741 all = true;
1742 } else {
1743 inc = !inc;
1744 hash = ops->func_hash->notrace_hash;
1745 other_hash = ops->func_hash->filter_hash;
1746 /*
1747 * If the notrace hash has no items,
1748 * then there's nothing to do.
1749 */
1750 if (ftrace_hash_empty(hash))
1751 return false;
1752 }
1753
1754 do_for_each_ftrace_rec(pg, rec) {
1755 int in_other_hash = 0;
1756 int in_hash = 0;
1757 int match = 0;
1758
1759 if (rec->flags & FTRACE_FL_DISABLED)
1760 continue;
1761
1762 if (all) {
1763 /*
1764 * Only the filter_hash affects all records.
1765 * Update if the record is not in the notrace hash.
1766 */
1767 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1768 match = 1;
1769 } else {
1770 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1771 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1772
1773 /*
1774 * If filter_hash is set, we want to match all functions
1775 * that are in the hash but not in the other hash.
1776 *
1777 * If filter_hash is not set, then we are decrementing.
1778 * That means we match anything that is in the hash
1779 * and also in the other_hash. That is, we need to turn
1780 * off functions in the other hash because they are disabled
1781 * by this hash.
1782 */
1783 if (filter_hash && in_hash && !in_other_hash)
1784 match = 1;
1785 else if (!filter_hash && in_hash &&
1786 (in_other_hash || ftrace_hash_empty(other_hash)))
1787 match = 1;
1788 }
1789 if (!match)
1790 continue;
1791
1792 if (inc) {
1793 rec->flags++;
1794 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1795 return false;
1796
1797 /*
1798 * If there's only a single callback registered to a
1799 * function, and the ops has a trampoline registered
1800 * for it, then we can call it directly.
1801 */
1802 if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1803 rec->flags |= FTRACE_FL_TRAMP;
1804 else
1805 /*
1806 * If we are adding another function callback
1807 * to this function, and the previous had a
1808 * custom trampoline in use, then we need to go
1809 * back to the default trampoline.
1810 */
1811 rec->flags &= ~FTRACE_FL_TRAMP;
1812
1813 /*
1814 * If any ops wants regs saved for this function
1815 * then all ops will get saved regs.
1816 */
1817 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1818 rec->flags |= FTRACE_FL_REGS;
1819 } else {
1820 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1821 return false;
1822 rec->flags--;
1823
1824 /*
1825 * If the rec had REGS enabled and the ops that is
1826 * being removed had REGS set, then see if there is
1827 * still any ops for this record that wants regs.
1828 * If not, we can stop recording them.
1829 */
1830 if (ftrace_rec_count(rec) > 0 &&
1831 rec->flags & FTRACE_FL_REGS &&
1832 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1833 if (!test_rec_ops_needs_regs(rec))
1834 rec->flags &= ~FTRACE_FL_REGS;
1835 }
1836
1837 /*
1838 * If the rec had TRAMP enabled, then it needs to
1839 * be cleared. As TRAMP can only be enabled iff
1840 * there is only a single ops attached to it.
1841 * In otherwords, always disable it on decrementing.
1842 * In the future, we may set it if rec count is
1843 * decremented to one, and the ops that is left
1844 * has a trampoline.
1845 */
1846 rec->flags &= ~FTRACE_FL_TRAMP;
1847
1848 /*
1849 * flags will be cleared in ftrace_check_record()
1850 * if rec count is zero.
1851 */
1852 }
1853 count++;
1854
1855 /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1856 update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1857
1858 /* Shortcut, if we handled all records, we are done. */
1859 if (!all && count == hash->count)
1860 return update;
1861 } while_for_each_ftrace_rec();
1862
1863 return update;
1864 }
1865
1866 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1867 int filter_hash)
1868 {
1869 return __ftrace_hash_rec_update(ops, filter_hash, 0);
1870 }
1871
1872 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1873 int filter_hash)
1874 {
1875 return __ftrace_hash_rec_update(ops, filter_hash, 1);
1876 }
1877
1878 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1879 int filter_hash, int inc)
1880 {
1881 struct ftrace_ops *op;
1882
1883 __ftrace_hash_rec_update(ops, filter_hash, inc);
1884
1885 if (ops->func_hash != &global_ops.local_hash)
1886 return;
1887
1888 /*
1889 * If the ops shares the global_ops hash, then we need to update
1890 * all ops that are enabled and use this hash.
1891 */
1892 do_for_each_ftrace_op(op, ftrace_ops_list) {
1893 /* Already done */
1894 if (op == ops)
1895 continue;
1896 if (op->func_hash == &global_ops.local_hash)
1897 __ftrace_hash_rec_update(op, filter_hash, inc);
1898 } while_for_each_ftrace_op(op);
1899 }
1900
1901 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1902 int filter_hash)
1903 {
1904 ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1905 }
1906
1907 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1908 int filter_hash)
1909 {
1910 ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1911 }
1912
1913 /*
1914 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1915 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1916 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1917 * Note that old_hash and new_hash has below meanings
1918 * - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1919 * - If the hash is EMPTY_HASH, it hits nothing
1920 * - Anything else hits the recs which match the hash entries.
1921 */
1922 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1923 struct ftrace_hash *old_hash,
1924 struct ftrace_hash *new_hash)
1925 {
1926 struct ftrace_page *pg;
1927 struct dyn_ftrace *rec, *end = NULL;
1928 int in_old, in_new;
1929
1930 /* Only update if the ops has been registered */
1931 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1932 return 0;
1933
1934 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1935 return 0;
1936
1937 /*
1938 * Since the IPMODIFY is a very address sensitive action, we do not
1939 * allow ftrace_ops to set all functions to new hash.
1940 */
1941 if (!new_hash || !old_hash)
1942 return -EINVAL;
1943
1944 /* Update rec->flags */
1945 do_for_each_ftrace_rec(pg, rec) {
1946
1947 if (rec->flags & FTRACE_FL_DISABLED)
1948 continue;
1949
1950 /* We need to update only differences of filter_hash */
1951 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1952 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1953 if (in_old == in_new)
1954 continue;
1955
1956 if (in_new) {
1957 /* New entries must ensure no others are using it */
1958 if (rec->flags & FTRACE_FL_IPMODIFY)
1959 goto rollback;
1960 rec->flags |= FTRACE_FL_IPMODIFY;
1961 } else /* Removed entry */
1962 rec->flags &= ~FTRACE_FL_IPMODIFY;
1963 } while_for_each_ftrace_rec();
1964
1965 return 0;
1966
1967 rollback:
1968 end = rec;
1969
1970 /* Roll back what we did above */
1971 do_for_each_ftrace_rec(pg, rec) {
1972
1973 if (rec->flags & FTRACE_FL_DISABLED)
1974 continue;
1975
1976 if (rec == end)
1977 goto err_out;
1978
1979 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1980 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1981 if (in_old == in_new)
1982 continue;
1983
1984 if (in_new)
1985 rec->flags &= ~FTRACE_FL_IPMODIFY;
1986 else
1987 rec->flags |= FTRACE_FL_IPMODIFY;
1988 } while_for_each_ftrace_rec();
1989
1990 err_out:
1991 return -EBUSY;
1992 }
1993
1994 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1995 {
1996 struct ftrace_hash *hash = ops->func_hash->filter_hash;
1997
1998 if (ftrace_hash_empty(hash))
1999 hash = NULL;
2000
2001 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
2002 }
2003
2004 /* Disabling always succeeds */
2005 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
2006 {
2007 struct ftrace_hash *hash = ops->func_hash->filter_hash;
2008
2009 if (ftrace_hash_empty(hash))
2010 hash = NULL;
2011
2012 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
2013 }
2014
2015 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
2016 struct ftrace_hash *new_hash)
2017 {
2018 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
2019
2020 if (ftrace_hash_empty(old_hash))
2021 old_hash = NULL;
2022
2023 if (ftrace_hash_empty(new_hash))
2024 new_hash = NULL;
2025
2026 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
2027 }
2028
2029 static void print_ip_ins(const char *fmt, const unsigned char *p)
2030 {
2031 int i;
2032
2033 printk(KERN_CONT "%s", fmt);
2034
2035 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
2036 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
2037 }
2038
2039 static struct ftrace_ops *
2040 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
2041 static struct ftrace_ops *
2042 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
2043
2044 enum ftrace_bug_type ftrace_bug_type;
2045 const void *ftrace_expected;
2046
2047 static void print_bug_type(void)
2048 {
2049 switch (ftrace_bug_type) {
2050 case FTRACE_BUG_UNKNOWN:
2051 break;
2052 case FTRACE_BUG_INIT:
2053 pr_info("Initializing ftrace call sites\n");
2054 break;
2055 case FTRACE_BUG_NOP:
2056 pr_info("Setting ftrace call site to NOP\n");
2057 break;
2058 case FTRACE_BUG_CALL:
2059 pr_info("Setting ftrace call site to call ftrace function\n");
2060 break;
2061 case FTRACE_BUG_UPDATE:
2062 pr_info("Updating ftrace call site to call a different ftrace function\n");
2063 break;
2064 }
2065 }
2066
2067 /**
2068 * ftrace_bug - report and shutdown function tracer
2069 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2070 * @rec: The record that failed
2071 *
2072 * The arch code that enables or disables the function tracing
2073 * can call ftrace_bug() when it has detected a problem in
2074 * modifying the code. @failed should be one of either:
2075 * EFAULT - if the problem happens on reading the @ip address
2076 * EINVAL - if what is read at @ip is not what was expected
2077 * EPERM - if the problem happens on writting to the @ip address
2078 */
2079 void ftrace_bug(int failed, struct dyn_ftrace *rec)
2080 {
2081 unsigned long ip = rec ? rec->ip : 0;
2082
2083 switch (failed) {
2084 case -EFAULT:
2085 FTRACE_WARN_ON_ONCE(1);
2086 pr_info("ftrace faulted on modifying ");
2087 print_ip_sym(ip);
2088 break;
2089 case -EINVAL:
2090 FTRACE_WARN_ON_ONCE(1);
2091 pr_info("ftrace failed to modify ");
2092 print_ip_sym(ip);
2093 print_ip_ins(" actual: ", (unsigned char *)ip);
2094 pr_cont("\n");
2095 if (ftrace_expected) {
2096 print_ip_ins(" expected: ", ftrace_expected);
2097 pr_cont("\n");
2098 }
2099 break;
2100 case -EPERM:
2101 FTRACE_WARN_ON_ONCE(1);
2102 pr_info("ftrace faulted on writing ");
2103 print_ip_sym(ip);
2104 break;
2105 default:
2106 FTRACE_WARN_ON_ONCE(1);
2107 pr_info("ftrace faulted on unknown error ");
2108 print_ip_sym(ip);
2109 }
2110 print_bug_type();
2111 if (rec) {
2112 struct ftrace_ops *ops = NULL;
2113
2114 pr_info("ftrace record flags: %lx\n", rec->flags);
2115 pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2116 rec->flags & FTRACE_FL_REGS ? " R" : " ");
2117 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2118 ops = ftrace_find_tramp_ops_any(rec);
2119 if (ops) {
2120 do {
2121 pr_cont("\ttramp: %pS (%pS)",
2122 (void *)ops->trampoline,
2123 (void *)ops->func);
2124 ops = ftrace_find_tramp_ops_next(rec, ops);
2125 } while (ops);
2126 } else
2127 pr_cont("\ttramp: ERROR!");
2128
2129 }
2130 ip = ftrace_get_addr_curr(rec);
2131 pr_cont("\n expected tramp: %lx\n", ip);
2132 }
2133 }
2134
2135 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2136 {
2137 unsigned long flag = 0UL;
2138
2139 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2140
2141 if (rec->flags & FTRACE_FL_DISABLED)
2142 return FTRACE_UPDATE_IGNORE;
2143
2144 /*
2145 * If we are updating calls:
2146 *
2147 * If the record has a ref count, then we need to enable it
2148 * because someone is using it.
2149 *
2150 * Otherwise we make sure its disabled.
2151 *
2152 * If we are disabling calls, then disable all records that
2153 * are enabled.
2154 */
2155 if (enable && ftrace_rec_count(rec))
2156 flag = FTRACE_FL_ENABLED;
2157
2158 /*
2159 * If enabling and the REGS flag does not match the REGS_EN, or
2160 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2161 * this record. Set flags to fail the compare against ENABLED.
2162 */
2163 if (flag) {
2164 if (!(rec->flags & FTRACE_FL_REGS) !=
2165 !(rec->flags & FTRACE_FL_REGS_EN))
2166 flag |= FTRACE_FL_REGS;
2167
2168 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2169 !(rec->flags & FTRACE_FL_TRAMP_EN))
2170 flag |= FTRACE_FL_TRAMP;
2171 }
2172
2173 /* If the state of this record hasn't changed, then do nothing */
2174 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2175 return FTRACE_UPDATE_IGNORE;
2176
2177 if (flag) {
2178 /* Save off if rec is being enabled (for return value) */
2179 flag ^= rec->flags & FTRACE_FL_ENABLED;
2180
2181 if (update) {
2182 rec->flags |= FTRACE_FL_ENABLED;
2183 if (flag & FTRACE_FL_REGS) {
2184 if (rec->flags & FTRACE_FL_REGS)
2185 rec->flags |= FTRACE_FL_REGS_EN;
2186 else
2187 rec->flags &= ~FTRACE_FL_REGS_EN;
2188 }
2189 if (flag & FTRACE_FL_TRAMP) {
2190 if (rec->flags & FTRACE_FL_TRAMP)
2191 rec->flags |= FTRACE_FL_TRAMP_EN;
2192 else
2193 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2194 }
2195 }
2196
2197 /*
2198 * If this record is being updated from a nop, then
2199 * return UPDATE_MAKE_CALL.
2200 * Otherwise,
2201 * return UPDATE_MODIFY_CALL to tell the caller to convert
2202 * from the save regs, to a non-save regs function or
2203 * vice versa, or from a trampoline call.
2204 */
2205 if (flag & FTRACE_FL_ENABLED) {
2206 ftrace_bug_type = FTRACE_BUG_CALL;
2207 return FTRACE_UPDATE_MAKE_CALL;
2208 }
2209
2210 ftrace_bug_type = FTRACE_BUG_UPDATE;
2211 return FTRACE_UPDATE_MODIFY_CALL;
2212 }
2213
2214 if (update) {
2215 /* If there's no more users, clear all flags */
2216 if (!ftrace_rec_count(rec))
2217 rec->flags = 0;
2218 else
2219 /*
2220 * Just disable the record, but keep the ops TRAMP
2221 * and REGS states. The _EN flags must be disabled though.
2222 */
2223 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2224 FTRACE_FL_REGS_EN);
2225 }
2226
2227 ftrace_bug_type = FTRACE_BUG_NOP;
2228 return FTRACE_UPDATE_MAKE_NOP;
2229 }
2230
2231 /**
2232 * ftrace_update_record, set a record that now is tracing or not
2233 * @rec: the record to update
2234 * @enable: set to 1 if the record is tracing, zero to force disable
2235 *
2236 * The records that represent all functions that can be traced need
2237 * to be updated when tracing has been enabled.
2238 */
2239 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2240 {
2241 return ftrace_check_record(rec, enable, 1);
2242 }
2243
2244 /**
2245 * ftrace_test_record, check if the record has been enabled or not
2246 * @rec: the record to test
2247 * @enable: set to 1 to check if enabled, 0 if it is disabled
2248 *
2249 * The arch code may need to test if a record is already set to
2250 * tracing to determine how to modify the function code that it
2251 * represents.
2252 */
2253 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2254 {
2255 return ftrace_check_record(rec, enable, 0);
2256 }
2257
2258 static struct ftrace_ops *
2259 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2260 {
2261 struct ftrace_ops *op;
2262 unsigned long ip = rec->ip;
2263
2264 do_for_each_ftrace_op(op, ftrace_ops_list) {
2265
2266 if (!op->trampoline)
2267 continue;
2268
2269 if (hash_contains_ip(ip, op->func_hash))
2270 return op;
2271 } while_for_each_ftrace_op(op);
2272
2273 return NULL;
2274 }
2275
2276 static struct ftrace_ops *
2277 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2278 struct ftrace_ops *op)
2279 {
2280 unsigned long ip = rec->ip;
2281
2282 while_for_each_ftrace_op(op) {
2283
2284 if (!op->trampoline)
2285 continue;
2286
2287 if (hash_contains_ip(ip, op->func_hash))
2288 return op;
2289 }
2290
2291 return NULL;
2292 }
2293
2294 static struct ftrace_ops *
2295 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2296 {
2297 struct ftrace_ops *op;
2298 unsigned long ip = rec->ip;
2299
2300 /*
2301 * Need to check removed ops first.
2302 * If they are being removed, and this rec has a tramp,
2303 * and this rec is in the ops list, then it would be the
2304 * one with the tramp.
2305 */
2306 if (removed_ops) {
2307 if (hash_contains_ip(ip, &removed_ops->old_hash))
2308 return removed_ops;
2309 }
2310
2311 /*
2312 * Need to find the current trampoline for a rec.
2313 * Now, a trampoline is only attached to a rec if there
2314 * was a single 'ops' attached to it. But this can be called
2315 * when we are adding another op to the rec or removing the
2316 * current one. Thus, if the op is being added, we can
2317 * ignore it because it hasn't attached itself to the rec
2318 * yet.
2319 *
2320 * If an ops is being modified (hooking to different functions)
2321 * then we don't care about the new functions that are being
2322 * added, just the old ones (that are probably being removed).
2323 *
2324 * If we are adding an ops to a function that already is using
2325 * a trampoline, it needs to be removed (trampolines are only
2326 * for single ops connected), then an ops that is not being
2327 * modified also needs to be checked.
2328 */
2329 do_for_each_ftrace_op(op, ftrace_ops_list) {
2330
2331 if (!op->trampoline)
2332 continue;
2333
2334 /*
2335 * If the ops is being added, it hasn't gotten to
2336 * the point to be removed from this tree yet.
2337 */
2338 if (op->flags & FTRACE_OPS_FL_ADDING)
2339 continue;
2340
2341
2342 /*
2343 * If the ops is being modified and is in the old
2344 * hash, then it is probably being removed from this
2345 * function.
2346 */
2347 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2348 hash_contains_ip(ip, &op->old_hash))
2349 return op;
2350 /*
2351 * If the ops is not being added or modified, and it's
2352 * in its normal filter hash, then this must be the one
2353 * we want!
2354 */
2355 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2356 hash_contains_ip(ip, op->func_hash))
2357 return op;
2358
2359 } while_for_each_ftrace_op(op);
2360
2361 return NULL;
2362 }
2363
2364 static struct ftrace_ops *
2365 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2366 {
2367 struct ftrace_ops *op;
2368 unsigned long ip = rec->ip;
2369
2370 do_for_each_ftrace_op(op, ftrace_ops_list) {
2371 /* pass rec in as regs to have non-NULL val */
2372 if (hash_contains_ip(ip, op->func_hash))
2373 return op;
2374 } while_for_each_ftrace_op(op);
2375
2376 return NULL;
2377 }
2378
2379 /**
2380 * ftrace_get_addr_new - Get the call address to set to
2381 * @rec: The ftrace record descriptor
2382 *
2383 * If the record has the FTRACE_FL_REGS set, that means that it
2384 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2385 * is not not set, then it wants to convert to the normal callback.
2386 *
2387 * Returns the address of the trampoline to set to
2388 */
2389 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2390 {
2391 struct ftrace_ops *ops;
2392
2393 /* Trampolines take precedence over regs */
2394 if (rec->flags & FTRACE_FL_TRAMP) {
2395 ops = ftrace_find_tramp_ops_new(rec);
2396 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2397 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2398 (void *)rec->ip, (void *)rec->ip, rec->flags);
2399 /* Ftrace is shutting down, return anything */
2400 return (unsigned long)FTRACE_ADDR;
2401 }
2402 return ops->trampoline;
2403 }
2404
2405 if (rec->flags & FTRACE_FL_REGS)
2406 return (unsigned long)FTRACE_REGS_ADDR;
2407 else
2408 return (unsigned long)FTRACE_ADDR;
2409 }
2410
2411 /**
2412 * ftrace_get_addr_curr - Get the call address that is already there
2413 * @rec: The ftrace record descriptor
2414 *
2415 * The FTRACE_FL_REGS_EN is set when the record already points to
2416 * a function that saves all the regs. Basically the '_EN' version
2417 * represents the current state of the function.
2418 *
2419 * Returns the address of the trampoline that is currently being called
2420 */
2421 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2422 {
2423 struct ftrace_ops *ops;
2424
2425 /* Trampolines take precedence over regs */
2426 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2427 ops = ftrace_find_tramp_ops_curr(rec);
2428 if (FTRACE_WARN_ON(!ops)) {
2429 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2430 (void *)rec->ip, (void *)rec->ip);
2431 /* Ftrace is shutting down, return anything */
2432 return (unsigned long)FTRACE_ADDR;
2433 }
2434 return ops->trampoline;
2435 }
2436
2437 if (rec->flags & FTRACE_FL_REGS_EN)
2438 return (unsigned long)FTRACE_REGS_ADDR;
2439 else
2440 return (unsigned long)FTRACE_ADDR;
2441 }
2442
2443 static int
2444 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2445 {
2446 unsigned long ftrace_old_addr;
2447 unsigned long ftrace_addr;
2448 int ret;
2449
2450 ftrace_addr = ftrace_get_addr_new(rec);
2451
2452 /* This needs to be done before we call ftrace_update_record */
2453 ftrace_old_addr = ftrace_get_addr_curr(rec);
2454
2455 ret = ftrace_update_record(rec, enable);
2456
2457 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2458
2459 switch (ret) {
2460 case FTRACE_UPDATE_IGNORE:
2461 return 0;
2462
2463 case FTRACE_UPDATE_MAKE_CALL:
2464 ftrace_bug_type = FTRACE_BUG_CALL;
2465 return ftrace_make_call(rec, ftrace_addr);
2466
2467 case FTRACE_UPDATE_MAKE_NOP:
2468 ftrace_bug_type = FTRACE_BUG_NOP;
2469 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2470
2471 case FTRACE_UPDATE_MODIFY_CALL:
2472 ftrace_bug_type = FTRACE_BUG_UPDATE;
2473 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2474 }
2475
2476 return -1; /* unknow ftrace bug */
2477 }
2478
2479 void __weak ftrace_replace_code(int enable)
2480 {
2481 struct dyn_ftrace *rec;
2482 struct ftrace_page *pg;
2483 int failed;
2484
2485 if (unlikely(ftrace_disabled))
2486 return;
2487
2488 do_for_each_ftrace_rec(pg, rec) {
2489
2490 if (rec->flags & FTRACE_FL_DISABLED)
2491 continue;
2492
2493 failed = __ftrace_replace_code(rec, enable);
2494 if (failed) {
2495 ftrace_bug(failed, rec);
2496 /* Stop processing */
2497 return;
2498 }
2499 } while_for_each_ftrace_rec();
2500 }
2501
2502 struct ftrace_rec_iter {
2503 struct ftrace_page *pg;
2504 int index;
2505 };
2506
2507 /**
2508 * ftrace_rec_iter_start, start up iterating over traced functions
2509 *
2510 * Returns an iterator handle that is used to iterate over all
2511 * the records that represent address locations where functions
2512 * are traced.
2513 *
2514 * May return NULL if no records are available.
2515 */
2516 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2517 {
2518 /*
2519 * We only use a single iterator.
2520 * Protected by the ftrace_lock mutex.
2521 */
2522 static struct ftrace_rec_iter ftrace_rec_iter;
2523 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2524
2525 iter->pg = ftrace_pages_start;
2526 iter->index = 0;
2527
2528 /* Could have empty pages */
2529 while (iter->pg && !iter->pg->index)
2530 iter->pg = iter->pg->next;
2531
2532 if (!iter->pg)
2533 return NULL;
2534
2535 return iter;
2536 }
2537
2538 /**
2539 * ftrace_rec_iter_next, get the next record to process.
2540 * @iter: The handle to the iterator.
2541 *
2542 * Returns the next iterator after the given iterator @iter.
2543 */
2544 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2545 {
2546 iter->index++;
2547
2548 if (iter->index >= iter->pg->index) {
2549 iter->pg = iter->pg->next;
2550 iter->index = 0;
2551
2552 /* Could have empty pages */
2553 while (iter->pg && !iter->pg->index)
2554 iter->pg = iter->pg->next;
2555 }
2556
2557 if (!iter->pg)
2558 return NULL;
2559
2560 return iter;
2561 }
2562
2563 /**
2564 * ftrace_rec_iter_record, get the record at the iterator location
2565 * @iter: The current iterator location
2566 *
2567 * Returns the record that the current @iter is at.
2568 */
2569 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2570 {
2571 return &iter->pg->records[iter->index];
2572 }
2573
2574 static int
2575 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2576 {
2577 int ret;
2578
2579 if (unlikely(ftrace_disabled))
2580 return 0;
2581
2582 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2583 if (ret) {
2584 ftrace_bug_type = FTRACE_BUG_INIT;
2585 ftrace_bug(ret, rec);
2586 return 0;
2587 }
2588 return 1;
2589 }
2590
2591 /*
2592 * archs can override this function if they must do something
2593 * before the modifying code is performed.
2594 */
2595 int __weak ftrace_arch_code_modify_prepare(void)
2596 {
2597 return 0;
2598 }
2599
2600 /*
2601 * archs can override this function if they must do something
2602 * after the modifying code is performed.
2603 */
2604 int __weak ftrace_arch_code_modify_post_process(void)
2605 {
2606 return 0;
2607 }
2608
2609 void ftrace_modify_all_code(int command)
2610 {
2611 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2612 int err = 0;
2613
2614 /*
2615 * If the ftrace_caller calls a ftrace_ops func directly,
2616 * we need to make sure that it only traces functions it
2617 * expects to trace. When doing the switch of functions,
2618 * we need to update to the ftrace_ops_list_func first
2619 * before the transition between old and new calls are set,
2620 * as the ftrace_ops_list_func will check the ops hashes
2621 * to make sure the ops are having the right functions
2622 * traced.
2623 */
2624 if (update) {
2625 err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2626 if (FTRACE_WARN_ON(err))
2627 return;
2628 }
2629
2630 if (command & FTRACE_UPDATE_CALLS)
2631 ftrace_replace_code(1);
2632 else if (command & FTRACE_DISABLE_CALLS)
2633 ftrace_replace_code(0);
2634
2635 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2636 function_trace_op = set_function_trace_op;
2637 smp_wmb();
2638 /* If irqs are disabled, we are in stop machine */
2639 if (!irqs_disabled())
2640 smp_call_function(ftrace_sync_ipi, NULL, 1);
2641 err = ftrace_update_ftrace_func(ftrace_trace_function);
2642 if (FTRACE_WARN_ON(err))
2643 return;
2644 }
2645
2646 if (command & FTRACE_START_FUNC_RET)
2647 err = ftrace_enable_ftrace_graph_caller();
2648 else if (command & FTRACE_STOP_FUNC_RET)
2649 err = ftrace_disable_ftrace_graph_caller();
2650 FTRACE_WARN_ON(err);
2651 }
2652
2653 static int __ftrace_modify_code(void *data)
2654 {
2655 int *command = data;
2656
2657 ftrace_modify_all_code(*command);
2658
2659 return 0;
2660 }
2661
2662 /**
2663 * ftrace_run_stop_machine, go back to the stop machine method
2664 * @command: The command to tell ftrace what to do
2665 *
2666 * If an arch needs to fall back to the stop machine method, the
2667 * it can call this function.
2668 */
2669 void ftrace_run_stop_machine(int command)
2670 {
2671 stop_machine(__ftrace_modify_code, &command, NULL);
2672 }
2673
2674 /**
2675 * arch_ftrace_update_code, modify the code to trace or not trace
2676 * @command: The command that needs to be done
2677 *
2678 * Archs can override this function if it does not need to
2679 * run stop_machine() to modify code.
2680 */
2681 void __weak arch_ftrace_update_code(int command)
2682 {
2683 ftrace_run_stop_machine(command);
2684 }
2685
2686 static void ftrace_run_update_code(int command)
2687 {
2688 int ret;
2689
2690 ret = ftrace_arch_code_modify_prepare();
2691 FTRACE_WARN_ON(ret);
2692 if (ret)
2693 return;
2694
2695 /*
2696 * By default we use stop_machine() to modify the code.
2697 * But archs can do what ever they want as long as it
2698 * is safe. The stop_machine() is the safest, but also
2699 * produces the most overhead.
2700 */
2701 arch_ftrace_update_code(command);
2702
2703 ret = ftrace_arch_code_modify_post_process();
2704 FTRACE_WARN_ON(ret);
2705 }
2706
2707 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2708 struct ftrace_ops_hash *old_hash)
2709 {
2710 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2711 ops->old_hash.filter_hash = old_hash->filter_hash;
2712 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2713 ftrace_run_update_code(command);
2714 ops->old_hash.filter_hash = NULL;
2715 ops->old_hash.notrace_hash = NULL;
2716 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2717 }
2718
2719 static ftrace_func_t saved_ftrace_func;
2720 static int ftrace_start_up;
2721
2722 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2723 {
2724 }
2725
2726 static void per_cpu_ops_free(struct ftrace_ops *ops)
2727 {
2728 free_percpu(ops->disabled);
2729 }
2730
2731 static void ftrace_startup_enable(int command)
2732 {
2733 if (saved_ftrace_func != ftrace_trace_function) {
2734 saved_ftrace_func = ftrace_trace_function;
2735 command |= FTRACE_UPDATE_TRACE_FUNC;
2736 }
2737
2738 if (!command || !ftrace_enabled)
2739 return;
2740
2741 ftrace_run_update_code(command);
2742 }
2743
2744 static void ftrace_startup_all(int command)
2745 {
2746 update_all_ops = true;
2747 ftrace_startup_enable(command);
2748 update_all_ops = false;
2749 }
2750
2751 static int ftrace_startup(struct ftrace_ops *ops, int command)
2752 {
2753 int ret;
2754
2755 if (unlikely(ftrace_disabled))
2756 return -ENODEV;
2757
2758 ret = __register_ftrace_function(ops);
2759 if (ret)
2760 return ret;
2761
2762 ftrace_start_up++;
2763
2764 /*
2765 * Note that ftrace probes uses this to start up
2766 * and modify functions it will probe. But we still
2767 * set the ADDING flag for modification, as probes
2768 * do not have trampolines. If they add them in the
2769 * future, then the probes will need to distinguish
2770 * between adding and updating probes.
2771 */
2772 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2773
2774 ret = ftrace_hash_ipmodify_enable(ops);
2775 if (ret < 0) {
2776 /* Rollback registration process */
2777 __unregister_ftrace_function(ops);
2778 ftrace_start_up--;
2779 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2780 return ret;
2781 }
2782
2783 if (ftrace_hash_rec_enable(ops, 1))
2784 command |= FTRACE_UPDATE_CALLS;
2785
2786 ftrace_startup_enable(command);
2787
2788 ops->flags &= ~FTRACE_OPS_FL_ADDING;
2789
2790 return 0;
2791 }
2792
2793 static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2794 {
2795 int ret;
2796
2797 if (unlikely(ftrace_disabled))
2798 return -ENODEV;
2799
2800 ret = __unregister_ftrace_function(ops);
2801 if (ret)
2802 return ret;
2803
2804 ftrace_start_up--;
2805 /*
2806 * Just warn in case of unbalance, no need to kill ftrace, it's not
2807 * critical but the ftrace_call callers may be never nopped again after
2808 * further ftrace uses.
2809 */
2810 WARN_ON_ONCE(ftrace_start_up < 0);
2811
2812 /* Disabling ipmodify never fails */
2813 ftrace_hash_ipmodify_disable(ops);
2814
2815 if (ftrace_hash_rec_disable(ops, 1))
2816 command |= FTRACE_UPDATE_CALLS;
2817
2818 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2819
2820 if (saved_ftrace_func != ftrace_trace_function) {
2821 saved_ftrace_func = ftrace_trace_function;
2822 command |= FTRACE_UPDATE_TRACE_FUNC;
2823 }
2824
2825 if (!command || !ftrace_enabled) {
2826 /*
2827 * If these are per_cpu ops, they still need their
2828 * per_cpu field freed. Since, function tracing is
2829 * not currently active, we can just free them
2830 * without synchronizing all CPUs.
2831 */
2832 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2833 per_cpu_ops_free(ops);
2834 return 0;
2835 }
2836
2837 /*
2838 * If the ops uses a trampoline, then it needs to be
2839 * tested first on update.
2840 */
2841 ops->flags |= FTRACE_OPS_FL_REMOVING;
2842 removed_ops = ops;
2843
2844 /* The trampoline logic checks the old hashes */
2845 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2846 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2847
2848 ftrace_run_update_code(command);
2849
2850 /*
2851 * If there's no more ops registered with ftrace, run a
2852 * sanity check to make sure all rec flags are cleared.
2853 */
2854 if (rcu_dereference_protected(ftrace_ops_list,
2855 lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
2856 struct ftrace_page *pg;
2857 struct dyn_ftrace *rec;
2858
2859 do_for_each_ftrace_rec(pg, rec) {
2860 if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_FL_DISABLED))
2861 pr_warn(" %pS flags:%lx\n",
2862 (void *)rec->ip, rec->flags);
2863 } while_for_each_ftrace_rec();
2864 }
2865
2866 ops->old_hash.filter_hash = NULL;
2867 ops->old_hash.notrace_hash = NULL;
2868
2869 removed_ops = NULL;
2870 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2871
2872 /*
2873 * Dynamic ops may be freed, we must make sure that all
2874 * callers are done before leaving this function.
2875 * The same goes for freeing the per_cpu data of the per_cpu
2876 * ops.
2877 */
2878 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) {
2879 /*
2880 * We need to do a hard force of sched synchronization.
2881 * This is because we use preempt_disable() to do RCU, but
2882 * the function tracers can be called where RCU is not watching
2883 * (like before user_exit()). We can not rely on the RCU
2884 * infrastructure to do the synchronization, thus we must do it
2885 * ourselves.
2886 */
2887 schedule_on_each_cpu(ftrace_sync);
2888
2889 /*
2890 * When the kernel is preeptive, tasks can be preempted
2891 * while on a ftrace trampoline. Just scheduling a task on
2892 * a CPU is not good enough to flush them. Calling
2893 * synchornize_rcu_tasks() will wait for those tasks to
2894 * execute and either schedule voluntarily or enter user space.
2895 */
2896 if (IS_ENABLED(CONFIG_PREEMPT))
2897 synchronize_rcu_tasks();
2898
2899 arch_ftrace_trampoline_free(ops);
2900
2901 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2902 per_cpu_ops_free(ops);
2903 }
2904
2905 return 0;
2906 }
2907
2908 static void ftrace_startup_sysctl(void)
2909 {
2910 int command;
2911
2912 if (unlikely(ftrace_disabled))
2913 return;
2914
2915 /* Force update next time */
2916 saved_ftrace_func = NULL;
2917 /* ftrace_start_up is true if we want ftrace running */
2918 if (ftrace_start_up) {
2919 command = FTRACE_UPDATE_CALLS;
2920 if (ftrace_graph_active)
2921 command |= FTRACE_START_FUNC_RET;
2922 ftrace_startup_enable(command);
2923 }
2924 }
2925
2926 static void ftrace_shutdown_sysctl(void)
2927 {
2928 int command;
2929
2930 if (unlikely(ftrace_disabled))
2931 return;
2932
2933 /* ftrace_start_up is true if ftrace is running */
2934 if (ftrace_start_up) {
2935 command = FTRACE_DISABLE_CALLS;
2936 if (ftrace_graph_active)
2937 command |= FTRACE_STOP_FUNC_RET;
2938 ftrace_run_update_code(command);
2939 }
2940 }
2941
2942 static u64 ftrace_update_time;
2943 unsigned long ftrace_update_tot_cnt;
2944
2945 static inline int ops_traces_mod(struct ftrace_ops *ops)
2946 {
2947 /*
2948 * Filter_hash being empty will default to trace module.
2949 * But notrace hash requires a test of individual module functions.
2950 */
2951 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2952 ftrace_hash_empty(ops->func_hash->notrace_hash);
2953 }
2954
2955 /*
2956 * Check if the current ops references the record.
2957 *
2958 * If the ops traces all functions, then it was already accounted for.
2959 * If the ops does not trace the current record function, skip it.
2960 * If the ops ignores the function via notrace filter, skip it.
2961 */
2962 static inline bool
2963 ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2964 {
2965 /* If ops isn't enabled, ignore it */
2966 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2967 return 0;
2968
2969 /* If ops traces all then it includes this function */
2970 if (ops_traces_mod(ops))
2971 return 1;
2972
2973 /* The function must be in the filter */
2974 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2975 !__ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2976 return 0;
2977
2978 /* If in notrace hash, we ignore it too */
2979 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2980 return 0;
2981
2982 return 1;
2983 }
2984
2985 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2986 {
2987 struct ftrace_page *pg;
2988 struct dyn_ftrace *p;
2989 u64 start, stop;
2990 unsigned long update_cnt = 0;
2991 unsigned long rec_flags = 0;
2992 int i;
2993
2994 start = ftrace_now(raw_smp_processor_id());
2995
2996 /*
2997 * When a module is loaded, this function is called to convert
2998 * the calls to mcount in its text to nops, and also to create
2999 * an entry in the ftrace data. Now, if ftrace is activated
3000 * after this call, but before the module sets its text to
3001 * read-only, the modification of enabling ftrace can fail if
3002 * the read-only is done while ftrace is converting the calls.
3003 * To prevent this, the module's records are set as disabled
3004 * and will be enabled after the call to set the module's text
3005 * to read-only.
3006 */
3007 if (mod)
3008 rec_flags |= FTRACE_FL_DISABLED;
3009
3010 for (pg = new_pgs; pg; pg = pg->next) {
3011
3012 for (i = 0; i < pg->index; i++) {
3013
3014 /* If something went wrong, bail without enabling anything */
3015 if (unlikely(ftrace_disabled))
3016 return -1;
3017
3018 p = &pg->records[i];
3019 p->flags = rec_flags;
3020
3021 /*
3022 * Do the initial record conversion from mcount jump
3023 * to the NOP instructions.
3024 */
3025 if (!ftrace_code_disable(mod, p))
3026 break;
3027
3028 update_cnt++;
3029 }
3030 }
3031
3032 stop = ftrace_now(raw_smp_processor_id());
3033 ftrace_update_time = stop - start;
3034 ftrace_update_tot_cnt += update_cnt;
3035
3036 return 0;
3037 }
3038
3039 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3040 {
3041 int order;
3042 int cnt;
3043
3044 if (WARN_ON(!count))
3045 return -EINVAL;
3046
3047 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
3048
3049 /*
3050 * We want to fill as much as possible. No more than a page
3051 * may be empty.
3052 */
3053 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
3054 order--;
3055
3056 again:
3057 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3058
3059 if (!pg->records) {
3060 /* if we can't allocate this size, try something smaller */
3061 if (!order)
3062 return -ENOMEM;
3063 order >>= 1;
3064 goto again;
3065 }
3066
3067 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3068 pg->size = cnt;
3069
3070 if (cnt > count)
3071 cnt = count;
3072
3073 return cnt;
3074 }
3075
3076 static struct ftrace_page *
3077 ftrace_allocate_pages(unsigned long num_to_init)
3078 {
3079 struct ftrace_page *start_pg;
3080 struct ftrace_page *pg;
3081 int order;
3082 int cnt;
3083
3084 if (!num_to_init)
3085 return 0;
3086
3087 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3088 if (!pg)
3089 return NULL;
3090
3091 /*
3092 * Try to allocate as much as possible in one continues
3093 * location that fills in all of the space. We want to
3094 * waste as little space as possible.
3095 */
3096 for (;;) {
3097 cnt = ftrace_allocate_records(pg, num_to_init);
3098 if (cnt < 0)
3099 goto free_pages;
3100
3101 num_to_init -= cnt;
3102 if (!num_to_init)
3103 break;
3104
3105 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3106 if (!pg->next)
3107 goto free_pages;
3108
3109 pg = pg->next;
3110 }
3111
3112 return start_pg;
3113
3114 free_pages:
3115 pg = start_pg;
3116 while (pg) {
3117 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3118 free_pages((unsigned long)pg->records, order);
3119 start_pg = pg->next;
3120 kfree(pg);
3121 pg = start_pg;
3122 }
3123 pr_info("ftrace: FAILED to allocate memory for functions\n");
3124 return NULL;
3125 }
3126
3127 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3128
3129 struct ftrace_iterator {
3130 loff_t pos;
3131 loff_t func_pos;
3132 loff_t mod_pos;
3133 struct ftrace_page *pg;
3134 struct dyn_ftrace *func;
3135 struct ftrace_func_probe *probe;
3136 struct ftrace_func_entry *probe_entry;
3137 struct trace_parser parser;
3138 struct ftrace_hash *hash;
3139 struct ftrace_ops *ops;
3140 struct trace_array *tr;
3141 struct list_head *mod_list;
3142 int pidx;
3143 int idx;
3144 unsigned flags;
3145 };
3146
3147 static void *
3148 t_probe_next(struct seq_file *m, loff_t *pos)
3149 {
3150 struct ftrace_iterator *iter = m->private;
3151 struct trace_array *tr = iter->ops->private;
3152 struct list_head *func_probes;
3153 struct ftrace_hash *hash;
3154 struct list_head *next;
3155 struct hlist_node *hnd = NULL;
3156 struct hlist_head *hhd;
3157 int size;
3158
3159 (*pos)++;
3160 iter->pos = *pos;
3161
3162 if (!tr)
3163 return NULL;
3164
3165 func_probes = &tr->func_probes;
3166 if (list_empty(func_probes))
3167 return NULL;
3168
3169 if (!iter->probe) {
3170 next = func_probes->next;
3171 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3172 }
3173
3174 if (iter->probe_entry)
3175 hnd = &iter->probe_entry->hlist;
3176
3177 hash = iter->probe->ops.func_hash->filter_hash;
3178 size = 1 << hash->size_bits;
3179
3180 retry:
3181 if (iter->pidx >= size) {
3182 if (iter->probe->list.next == func_probes)
3183 return NULL;
3184 next = iter->probe->list.next;
3185 iter->probe = list_entry(next, struct ftrace_func_probe, list);
3186 hash = iter->probe->ops.func_hash->filter_hash;
3187 size = 1 << hash->size_bits;
3188 iter->pidx = 0;
3189 }
3190
3191 hhd = &hash->buckets[iter->pidx];
3192
3193 if (hlist_empty(hhd)) {
3194 iter->pidx++;
3195 hnd = NULL;
3196 goto retry;
3197 }
3198
3199 if (!hnd)
3200 hnd = hhd->first;
3201 else {
3202 hnd = hnd->next;
3203 if (!hnd) {
3204 iter->pidx++;
3205 goto retry;
3206 }
3207 }
3208
3209 if (WARN_ON_ONCE(!hnd))
3210 return NULL;
3211
3212 iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3213
3214 return iter;
3215 }
3216
3217 static void *t_probe_start(struct seq_file *m, loff_t *pos)
3218 {
3219 struct ftrace_iterator *iter = m->private;
3220 void *p = NULL;
3221 loff_t l;
3222
3223 if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3224 return NULL;
3225
3226 if (iter->mod_pos > *pos)
3227 return NULL;
3228
3229 iter->probe = NULL;
3230 iter->probe_entry = NULL;
3231 iter->pidx = 0;
3232 for (l = 0; l <= (*pos - iter->mod_pos); ) {
3233 p = t_probe_next(m, &l);
3234 if (!p)
3235 break;
3236 }
3237 if (!p)
3238 return NULL;
3239
3240 /* Only set this if we have an item */
3241 iter->flags |= FTRACE_ITER_PROBE;
3242
3243 return iter;
3244 }
3245
3246 static int
3247 t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3248 {
3249 struct ftrace_func_entry *probe_entry;
3250 struct ftrace_probe_ops *probe_ops;
3251 struct ftrace_func_probe *probe;
3252
3253 probe = iter->probe;
3254 probe_entry = iter->probe_entry;
3255
3256 if (WARN_ON_ONCE(!probe || !probe_entry))
3257 return -EIO;
3258
3259 probe_ops = probe->probe_ops;
3260
3261 if (probe_ops->print)
3262 return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3263
3264 seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3265 (void *)probe_ops->func);
3266
3267 return 0;
3268 }
3269
3270 static void *
3271 t_mod_next(struct seq_file *m, loff_t *pos)
3272 {
3273 struct ftrace_iterator *iter = m->private;
3274 struct trace_array *tr = iter->tr;
3275
3276 (*pos)++;
3277 iter->pos = *pos;
3278
3279 iter->mod_list = iter->mod_list->next;
3280
3281 if (iter->mod_list == &tr->mod_trace ||
3282 iter->mod_list == &tr->mod_notrace) {
3283 iter->flags &= ~FTRACE_ITER_MOD;
3284 return NULL;
3285 }
3286
3287 iter->mod_pos = *pos;
3288
3289 return iter;
3290 }
3291
3292 static void *t_mod_start(struct seq_file *m, loff_t *pos)
3293 {
3294 struct ftrace_iterator *iter = m->private;
3295 void *p = NULL;
3296 loff_t l;
3297
3298 if (iter->func_pos > *pos)
3299 return NULL;
3300
3301 iter->mod_pos = iter->func_pos;
3302
3303 /* probes are only available if tr is set */
3304 if (!iter->tr)
3305 return NULL;
3306
3307 for (l = 0; l <= (*pos - iter->func_pos); ) {
3308 p = t_mod_next(m, &l);
3309 if (!p)
3310 break;
3311 }
3312 if (!p) {
3313 iter->flags &= ~FTRACE_ITER_MOD;
3314 return t_probe_start(m, pos);
3315 }
3316
3317 /* Only set this if we have an item */
3318 iter->flags |= FTRACE_ITER_MOD;
3319
3320 return iter;
3321 }
3322
3323 static int
3324 t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
3325 {
3326 struct ftrace_mod_load *ftrace_mod;
3327 struct trace_array *tr = iter->tr;
3328
3329 if (WARN_ON_ONCE(!iter->mod_list) ||
3330 iter->mod_list == &tr->mod_trace ||
3331 iter->mod_list == &tr->mod_notrace)
3332 return -EIO;
3333
3334 ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
3335
3336 if (ftrace_mod->func)
3337 seq_printf(m, "%s", ftrace_mod->func);
3338 else
3339 seq_putc(m, '*');
3340
3341 seq_printf(m, ":mod:%s\n", ftrace_mod->module);
3342
3343 return 0;
3344 }
3345
3346 static void *
3347 t_func_next(struct seq_file *m, loff_t *pos)
3348 {
3349 struct ftrace_iterator *iter = m->private;
3350 struct dyn_ftrace *rec = NULL;
3351
3352 (*pos)++;
3353
3354 retry:
3355 if (iter->idx >= iter->pg->index) {
3356 if (iter->pg->next) {
3357 iter->pg = iter->pg->next;
3358 iter->idx = 0;
3359 goto retry;
3360 }
3361 } else {
3362 rec = &iter->pg->records[iter->idx++];
3363 if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3364 !ftrace_lookup_ip(iter->hash, rec->ip)) ||
3365
3366 ((iter->flags & FTRACE_ITER_ENABLED) &&
3367 !(rec->flags & FTRACE_FL_ENABLED))) {
3368
3369 rec = NULL;
3370 goto retry;
3371 }
3372 }
3373
3374 if (!rec)
3375 return NULL;
3376
3377 iter->pos = iter->func_pos = *pos;
3378 iter->func = rec;
3379
3380 return iter;
3381 }
3382
3383 static void *
3384 t_next(struct seq_file *m, void *v, loff_t *pos)
3385 {
3386 struct ftrace_iterator *iter = m->private;
3387 loff_t l = *pos; /* t_probe_start() must use original pos */
3388 void *ret;
3389
3390 if (unlikely(ftrace_disabled))
3391 return NULL;
3392
3393 if (iter->flags & FTRACE_ITER_PROBE)
3394 return t_probe_next(m, pos);
3395
3396 if (iter->flags & FTRACE_ITER_MOD)
3397 return t_mod_next(m, pos);
3398
3399 if (iter->flags & FTRACE_ITER_PRINTALL) {
3400 /* next must increment pos, and t_probe_start does not */
3401 (*pos)++;
3402 return t_mod_start(m, &l);
3403 }
3404
3405 ret = t_func_next(m, pos);
3406
3407 if (!ret)
3408 return t_mod_start(m, &l);
3409
3410 return ret;
3411 }
3412
3413 static void reset_iter_read(struct ftrace_iterator *iter)
3414 {
3415 iter->pos = 0;
3416 iter->func_pos = 0;
3417 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
3418 }
3419
3420 static void *t_start(struct seq_file *m, loff_t *pos)
3421 {
3422 struct ftrace_iterator *iter = m->private;
3423 void *p = NULL;
3424 loff_t l;
3425
3426 mutex_lock(&ftrace_lock);
3427
3428 if (unlikely(ftrace_disabled))
3429 return NULL;
3430
3431 /*
3432 * If an lseek was done, then reset and start from beginning.
3433 */
3434 if (*pos < iter->pos)
3435 reset_iter_read(iter);
3436
3437 /*
3438 * For set_ftrace_filter reading, if we have the filter
3439 * off, we can short cut and just print out that all
3440 * functions are enabled.
3441 */
3442 if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3443 ftrace_hash_empty(iter->hash)) {
3444 iter->func_pos = 1; /* Account for the message */
3445 if (*pos > 0)
3446 return t_mod_start(m, pos);
3447 iter->flags |= FTRACE_ITER_PRINTALL;
3448 /* reset in case of seek/pread */
3449 iter->flags &= ~FTRACE_ITER_PROBE;
3450 return iter;
3451 }
3452
3453 if (iter->flags & FTRACE_ITER_MOD)
3454 return t_mod_start(m, pos);
3455
3456 /*
3457 * Unfortunately, we need to restart at ftrace_pages_start
3458 * every time we let go of the ftrace_mutex. This is because
3459 * those pointers can change without the lock.
3460 */
3461 iter->pg = ftrace_pages_start;
3462 iter->idx = 0;
3463 for (l = 0; l <= *pos; ) {
3464 p = t_func_next(m, &l);
3465 if (!p)
3466 break;
3467 }
3468
3469 if (!p)
3470 return t_mod_start(m, pos);
3471
3472 return iter;
3473 }
3474
3475 static void t_stop(struct seq_file *m, void *p)
3476 {
3477 mutex_unlock(&ftrace_lock);
3478 }
3479
3480 void * __weak
3481 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3482 {
3483 return NULL;
3484 }
3485
3486 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3487 struct dyn_ftrace *rec)
3488 {
3489 void *ptr;
3490
3491 ptr = arch_ftrace_trampoline_func(ops, rec);
3492 if (ptr)
3493 seq_printf(m, " ->%pS", ptr);
3494 }
3495
3496 static int t_show(struct seq_file *m, void *v)
3497 {
3498 struct ftrace_iterator *iter = m->private;
3499 struct dyn_ftrace *rec;
3500
3501 if (iter->flags & FTRACE_ITER_PROBE)
3502 return t_probe_show(m, iter);
3503
3504 if (iter->flags & FTRACE_ITER_MOD)
3505 return t_mod_show(m, iter);
3506
3507 if (iter->flags & FTRACE_ITER_PRINTALL) {
3508 if (iter->flags & FTRACE_ITER_NOTRACE)
3509 seq_puts(m, "#### no functions disabled ####\n");
3510 else
3511 seq_puts(m, "#### all functions enabled ####\n");
3512 return 0;
3513 }
3514
3515 rec = iter->func;
3516
3517 if (!rec)
3518 return 0;
3519
3520 seq_printf(m, "%ps", (void *)rec->ip);
3521 if (iter->flags & FTRACE_ITER_ENABLED) {
3522 struct ftrace_ops *ops;
3523
3524 seq_printf(m, " (%ld)%s%s",
3525 ftrace_rec_count(rec),
3526 rec->flags & FTRACE_FL_REGS ? " R" : " ",
3527 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ");
3528 if (rec->flags & FTRACE_FL_TRAMP_EN) {
3529 ops = ftrace_find_tramp_ops_any(rec);
3530 if (ops) {
3531 do {
3532 seq_printf(m, "\ttramp: %pS (%pS)",
3533 (void *)ops->trampoline,
3534 (void *)ops->func);
3535 add_trampoline_func(m, ops, rec);
3536 ops = ftrace_find_tramp_ops_next(rec, ops);
3537 } while (ops);
3538 } else
3539 seq_puts(m, "\ttramp: ERROR!");
3540 } else {
3541 add_trampoline_func(m, NULL, rec);
3542 }
3543 }
3544
3545 seq_putc(m, '\n');
3546
3547 return 0;
3548 }
3549
3550 static const struct seq_operations show_ftrace_seq_ops = {
3551 .start = t_start,
3552 .next = t_next,
3553 .stop = t_stop,
3554 .show = t_show,
3555 };
3556
3557 static int
3558 ftrace_avail_open(struct inode *inode, struct file *file)
3559 {
3560 struct ftrace_iterator *iter;
3561
3562 if (unlikely(ftrace_disabled))
3563 return -ENODEV;
3564
3565 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3566 if (!iter)
3567 return -ENOMEM;
3568
3569 iter->pg = ftrace_pages_start;
3570 iter->ops = &global_ops;
3571
3572 return 0;
3573 }
3574
3575 static int
3576 ftrace_enabled_open(struct inode *inode, struct file *file)
3577 {
3578 struct ftrace_iterator *iter;
3579
3580 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3581 if (!iter)
3582 return -ENOMEM;
3583
3584 iter->pg = ftrace_pages_start;
3585 iter->flags = FTRACE_ITER_ENABLED;
3586 iter->ops = &global_ops;
3587
3588 return 0;
3589 }
3590
3591 /**
3592 * ftrace_regex_open - initialize function tracer filter files
3593 * @ops: The ftrace_ops that hold the hash filters
3594 * @flag: The type of filter to process
3595 * @inode: The inode, usually passed in to your open routine
3596 * @file: The file, usually passed in to your open routine
3597 *
3598 * ftrace_regex_open() initializes the filter files for the
3599 * @ops. Depending on @flag it may process the filter hash or
3600 * the notrace hash of @ops. With this called from the open
3601 * routine, you can use ftrace_filter_write() for the write
3602 * routine if @flag has FTRACE_ITER_FILTER set, or
3603 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3604 * tracing_lseek() should be used as the lseek routine, and
3605 * release must call ftrace_regex_release().
3606 */
3607 int
3608 ftrace_regex_open(struct ftrace_ops *ops, int flag,
3609 struct inode *inode, struct file *file)
3610 {
3611 struct ftrace_iterator *iter;
3612 struct ftrace_hash *hash;
3613 struct list_head *mod_head;
3614 struct trace_array *tr = ops->private;
3615 int ret = 0;
3616
3617 ftrace_ops_init(ops);
3618
3619 if (unlikely(ftrace_disabled))
3620 return -ENODEV;
3621
3622 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3623 if (!iter)
3624 return -ENOMEM;
3625
3626 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3627 kfree(iter);
3628 return -ENOMEM;
3629 }
3630
3631 iter->ops = ops;
3632 iter->flags = flag;
3633 iter->tr = tr;
3634
3635 mutex_lock(&ops->func_hash->regex_lock);
3636
3637 if (flag & FTRACE_ITER_NOTRACE) {
3638 hash = ops->func_hash->notrace_hash;
3639 mod_head = tr ? &tr->mod_notrace : NULL;
3640 } else {
3641 hash = ops->func_hash->filter_hash;
3642 mod_head = tr ? &tr->mod_trace : NULL;
3643 }
3644
3645 iter->mod_list = mod_head;
3646
3647 if (file->f_mode & FMODE_WRITE) {
3648 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3649
3650 if (file->f_flags & O_TRUNC) {
3651 iter->hash = alloc_ftrace_hash(size_bits);
3652 clear_ftrace_mod_list(mod_head);
3653 } else {
3654 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3655 }
3656
3657 if (!iter->hash) {
3658 trace_parser_put(&iter->parser);
3659 kfree(iter);
3660 ret = -ENOMEM;
3661 goto out_unlock;
3662 }
3663 } else
3664 iter->hash = hash;
3665
3666 if (file->f_mode & FMODE_READ) {
3667 iter->pg = ftrace_pages_start;
3668
3669 ret = seq_open(file, &show_ftrace_seq_ops);
3670 if (!ret) {
3671 struct seq_file *m = file->private_data;
3672 m->private = iter;
3673 } else {
3674 /* Failed */
3675 free_ftrace_hash(iter->hash);
3676 trace_parser_put(&iter->parser);
3677 kfree(iter);
3678 }
3679 } else
3680 file->private_data = iter;
3681
3682 out_unlock:
3683 mutex_unlock(&ops->func_hash->regex_lock);
3684
3685 return ret;
3686 }
3687
3688 static int
3689 ftrace_filter_open(struct inode *inode, struct file *file)
3690 {
3691 struct ftrace_ops *ops = inode->i_private;
3692
3693 return ftrace_regex_open(ops,
3694 FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
3695 inode, file);
3696 }
3697
3698 static int
3699 ftrace_notrace_open(struct inode *inode, struct file *file)
3700 {
3701 struct ftrace_ops *ops = inode->i_private;
3702
3703 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3704 inode, file);
3705 }
3706
3707 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3708 struct ftrace_glob {
3709 char *search;
3710 unsigned len;
3711 int type;
3712 };
3713
3714 /*
3715 * If symbols in an architecture don't correspond exactly to the user-visible
3716 * name of what they represent, it is possible to define this function to
3717 * perform the necessary adjustments.
3718 */
3719 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3720 {
3721 return str;
3722 }
3723
3724 static int ftrace_match(char *str, struct ftrace_glob *g)
3725 {
3726 int matched = 0;
3727 int slen;
3728
3729 str = arch_ftrace_match_adjust(str, g->search);
3730
3731 switch (g->type) {
3732 case MATCH_FULL:
3733 if (strcmp(str, g->search) == 0)
3734 matched = 1;
3735 break;
3736 case MATCH_FRONT_ONLY:
3737 if (strncmp(str, g->search, g->len) == 0)
3738 matched = 1;
3739 break;
3740 case MATCH_MIDDLE_ONLY:
3741 if (strstr(str, g->search))
3742 matched = 1;
3743 break;
3744 case MATCH_END_ONLY:
3745 slen = strlen(str);
3746 if (slen >= g->len &&
3747 memcmp(str + slen - g->len, g->search, g->len) == 0)
3748 matched = 1;
3749 break;
3750 case MATCH_GLOB:
3751 if (glob_match(g->search, str))
3752 matched = 1;
3753 break;
3754 }
3755
3756 return matched;
3757 }
3758
3759 static int
3760 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3761 {
3762 struct ftrace_func_entry *entry;
3763 int ret = 0;
3764
3765 entry = ftrace_lookup_ip(hash, rec->ip);
3766 if (clear_filter) {
3767 /* Do nothing if it doesn't exist */
3768 if (!entry)
3769 return 0;
3770
3771 free_hash_entry(hash, entry);
3772 } else {
3773 /* Do nothing if it exists */
3774 if (entry)
3775 return 0;
3776
3777 ret = add_hash_entry(hash, rec->ip);
3778 }
3779 return ret;
3780 }
3781
3782 static int
3783 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3784 struct ftrace_glob *mod_g, int exclude_mod)
3785 {
3786 char str[KSYM_SYMBOL_LEN];
3787 char *modname;
3788
3789 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3790
3791 if (mod_g) {
3792 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3793
3794 /* blank module name to match all modules */
3795 if (!mod_g->len) {
3796 /* blank module globbing: modname xor exclude_mod */
3797 if (!exclude_mod != !modname)
3798 goto func_match;
3799 return 0;
3800 }
3801
3802 /*
3803 * exclude_mod is set to trace everything but the given
3804 * module. If it is set and the module matches, then
3805 * return 0. If it is not set, and the module doesn't match
3806 * also return 0. Otherwise, check the function to see if
3807 * that matches.
3808 */
3809 if (!mod_matches == !exclude_mod)
3810 return 0;
3811 func_match:
3812 /* blank search means to match all funcs in the mod */
3813 if (!func_g->len)
3814 return 1;
3815 }
3816
3817 return ftrace_match(str, func_g);
3818 }
3819
3820 static int
3821 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3822 {
3823 struct ftrace_page *pg;
3824 struct dyn_ftrace *rec;
3825 struct ftrace_glob func_g = { .type = MATCH_FULL };
3826 struct ftrace_glob mod_g = { .type = MATCH_FULL };
3827 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3828 int exclude_mod = 0;
3829 int found = 0;
3830 int ret;
3831 int clear_filter = 0;
3832
3833 if (func) {
3834 func_g.type = filter_parse_regex(func, len, &func_g.search,
3835 &clear_filter);
3836 func_g.len = strlen(func_g.search);
3837 }
3838
3839 if (mod) {
3840 mod_g.type = filter_parse_regex(mod, strlen(mod),
3841 &mod_g.search, &exclude_mod);
3842 mod_g.len = strlen(mod_g.search);
3843 }
3844
3845 mutex_lock(&ftrace_lock);
3846
3847 if (unlikely(ftrace_disabled))
3848 goto out_unlock;
3849
3850 do_for_each_ftrace_rec(pg, rec) {
3851
3852 if (rec->flags & FTRACE_FL_DISABLED)
3853 continue;
3854
3855 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3856 ret = enter_record(hash, rec, clear_filter);
3857 if (ret < 0) {
3858 found = ret;
3859 goto out_unlock;
3860 }
3861 found = 1;
3862 }
3863 } while_for_each_ftrace_rec();
3864 out_unlock:
3865 mutex_unlock(&ftrace_lock);
3866
3867 return found;
3868 }
3869
3870 static int
3871 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3872 {
3873 return match_records(hash, buff, len, NULL);
3874 }
3875
3876 static void ftrace_ops_update_code(struct ftrace_ops *ops,
3877 struct ftrace_ops_hash *old_hash)
3878 {
3879 struct ftrace_ops *op;
3880
3881 if (!ftrace_enabled)
3882 return;
3883
3884 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
3885 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
3886 return;
3887 }
3888
3889 /*
3890 * If this is the shared global_ops filter, then we need to
3891 * check if there is another ops that shares it, is enabled.
3892 * If so, we still need to run the modify code.
3893 */
3894 if (ops->func_hash != &global_ops.local_hash)
3895 return;
3896
3897 do_for_each_ftrace_op(op, ftrace_ops_list) {
3898 if (op->func_hash == &global_ops.local_hash &&
3899 op->flags & FTRACE_OPS_FL_ENABLED) {
3900 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
3901 /* Only need to do this once */
3902 return;
3903 }
3904 } while_for_each_ftrace_op(op);
3905 }
3906
3907 static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3908 struct ftrace_hash **orig_hash,
3909 struct ftrace_hash *hash,
3910 int enable)
3911 {
3912 struct ftrace_ops_hash old_hash_ops;
3913 struct ftrace_hash *old_hash;
3914 int ret;
3915
3916 old_hash = *orig_hash;
3917 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3918 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3919 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3920 if (!ret) {
3921 ftrace_ops_update_code(ops, &old_hash_ops);
3922 free_ftrace_hash_rcu(old_hash);
3923 }
3924 return ret;
3925 }
3926
3927 static bool module_exists(const char *module)
3928 {
3929 /* All modules have the symbol __this_module */
3930 const char this_mod[] = "__this_module";
3931 const int modname_size = MAX_PARAM_PREFIX_LEN + sizeof(this_mod) + 1;
3932 char modname[modname_size + 1];
3933 unsigned long val;
3934 int n;
3935
3936 n = snprintf(modname, modname_size + 1, "%s:%s", module, this_mod);
3937
3938 if (n > modname_size)
3939 return false;
3940
3941 val = module_kallsyms_lookup_name(modname);
3942 return val != 0;
3943 }
3944
3945 static int cache_mod(struct trace_array *tr,
3946 const char *func, char *module, int enable)
3947 {
3948 struct ftrace_mod_load *ftrace_mod, *n;
3949 struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
3950 int ret;
3951
3952 mutex_lock(&ftrace_lock);
3953
3954 /* We do not cache inverse filters */
3955 if (func[0] == '!') {
3956 func++;
3957 ret = -EINVAL;
3958
3959 /* Look to remove this hash */
3960 list_for_each_entry_safe(ftrace_mod, n, head, list) {
3961 if (strcmp(ftrace_mod->module, module) != 0)
3962 continue;
3963
3964 /* no func matches all */
3965 if (strcmp(func, "*") == 0 ||
3966 (ftrace_mod->func &&
3967 strcmp(ftrace_mod->func, func) == 0)) {
3968 ret = 0;
3969 free_ftrace_mod(ftrace_mod);
3970 continue;
3971 }
3972 }
3973 goto out;
3974 }
3975
3976 ret = -EINVAL;
3977 /* We only care about modules that have not been loaded yet */
3978 if (module_exists(module))
3979 goto out;
3980
3981 /* Save this string off, and execute it when the module is loaded */
3982 ret = ftrace_add_mod(tr, func, module, enable);
3983 out:
3984 mutex_unlock(&ftrace_lock);
3985
3986 return ret;
3987 }
3988
3989 static int
3990 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3991 int reset, int enable);
3992
3993 #ifdef CONFIG_MODULES
3994 static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
3995 char *mod, bool enable)
3996 {
3997 struct ftrace_mod_load *ftrace_mod, *n;
3998 struct ftrace_hash **orig_hash, *new_hash;
3999 LIST_HEAD(process_mods);
4000 char *func;
4001 int ret;
4002
4003 mutex_lock(&ops->func_hash->regex_lock);
4004
4005 if (enable)
4006 orig_hash = &ops->func_hash->filter_hash;
4007 else
4008 orig_hash = &ops->func_hash->notrace_hash;
4009
4010 new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
4011 *orig_hash);
4012 if (!new_hash)
4013 goto out; /* warn? */
4014
4015 mutex_lock(&ftrace_lock);
4016
4017 list_for_each_entry_safe(ftrace_mod, n, head, list) {
4018
4019 if (strcmp(ftrace_mod->module, mod) != 0)
4020 continue;
4021
4022 if (ftrace_mod->func)
4023 func = kstrdup(ftrace_mod->func, GFP_KERNEL);
4024 else
4025 func = kstrdup("*", GFP_KERNEL);
4026
4027 if (!func) /* warn? */
4028 continue;
4029
4030 list_del(&ftrace_mod->list);
4031 list_add(&ftrace_mod->list, &process_mods);
4032
4033 /* Use the newly allocated func, as it may be "*" */
4034 kfree(ftrace_mod->func);
4035 ftrace_mod->func = func;
4036 }
4037
4038 mutex_unlock(&ftrace_lock);
4039
4040 list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
4041
4042 func = ftrace_mod->func;
4043
4044 /* Grabs ftrace_lock, which is why we have this extra step */
4045 match_records(new_hash, func, strlen(func), mod);
4046 free_ftrace_mod(ftrace_mod);
4047 }
4048
4049 if (enable && list_empty(head))
4050 new_hash->flags &= ~FTRACE_HASH_FL_MOD;
4051
4052 mutex_lock(&ftrace_lock);
4053
4054 ret = ftrace_hash_move_and_update_ops(ops, orig_hash,
4055 new_hash, enable);
4056 mutex_unlock(&ftrace_lock);
4057
4058 out:
4059 mutex_unlock(&ops->func_hash->regex_lock);
4060
4061 free_ftrace_hash(new_hash);
4062 }
4063
4064 static void process_cached_mods(const char *mod_name)
4065 {
4066 struct trace_array *tr;
4067 char *mod;
4068
4069 mod = kstrdup(mod_name, GFP_KERNEL);
4070 if (!mod)
4071 return;
4072
4073 mutex_lock(&trace_types_lock);
4074 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
4075 if (!list_empty(&tr->mod_trace))
4076 process_mod_list(&tr->mod_trace, tr->ops, mod, true);
4077 if (!list_empty(&tr->mod_notrace))
4078 process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
4079 }
4080 mutex_unlock(&trace_types_lock);
4081
4082 kfree(mod);
4083 }
4084 #endif
4085
4086 /*
4087 * We register the module command as a template to show others how
4088 * to register the a command as well.
4089 */
4090
4091 static int
4092 ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
4093 char *func_orig, char *cmd, char *module, int enable)
4094 {
4095 char *func;
4096 int ret;
4097
4098 /* match_records() modifies func, and we need the original */
4099 func = kstrdup(func_orig, GFP_KERNEL);
4100 if (!func)
4101 return -ENOMEM;
4102
4103 /*
4104 * cmd == 'mod' because we only registered this func
4105 * for the 'mod' ftrace_func_command.
4106 * But if you register one func with multiple commands,
4107 * you can tell which command was used by the cmd
4108 * parameter.
4109 */
4110 ret = match_records(hash, func, strlen(func), module);
4111 kfree(func);
4112
4113 if (!ret)
4114 return cache_mod(tr, func_orig, module, enable);
4115 if (ret < 0)
4116 return ret;
4117 return 0;
4118 }
4119
4120 static struct ftrace_func_command ftrace_mod_cmd = {
4121 .name = "mod",
4122 .func = ftrace_mod_callback,
4123 };
4124
4125 static int __init ftrace_mod_cmd_init(void)
4126 {
4127 return register_ftrace_command(&ftrace_mod_cmd);
4128 }
4129 core_initcall(ftrace_mod_cmd_init);
4130
4131 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
4132 struct ftrace_ops *op, struct pt_regs *pt_regs)
4133 {
4134 struct ftrace_probe_ops *probe_ops;
4135 struct ftrace_func_probe *probe;
4136
4137 probe = container_of(op, struct ftrace_func_probe, ops);
4138 probe_ops = probe->probe_ops;
4139
4140 /*
4141 * Disable preemption for these calls to prevent a RCU grace
4142 * period. This syncs the hash iteration and freeing of items
4143 * on the hash. rcu_read_lock is too dangerous here.
4144 */
4145 preempt_disable_notrace();
4146 probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
4147 preempt_enable_notrace();
4148 }
4149
4150 struct ftrace_func_map {
4151 struct ftrace_func_entry entry;
4152 void *data;
4153 };
4154
4155 struct ftrace_func_mapper {
4156 struct ftrace_hash hash;
4157 };
4158
4159 /**
4160 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
4161 *
4162 * Returns a ftrace_func_mapper descriptor that can be used to map ips to data.
4163 */
4164 struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
4165 {
4166 struct ftrace_hash *hash;
4167
4168 /*
4169 * The mapper is simply a ftrace_hash, but since the entries
4170 * in the hash are not ftrace_func_entry type, we define it
4171 * as a separate structure.
4172 */
4173 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4174 return (struct ftrace_func_mapper *)hash;
4175 }
4176
4177 /**
4178 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
4179 * @mapper: The mapper that has the ip maps
4180 * @ip: the instruction pointer to find the data for
4181 *
4182 * Returns the data mapped to @ip if found otherwise NULL. The return
4183 * is actually the address of the mapper data pointer. The address is
4184 * returned for use cases where the data is no bigger than a long, and
4185 * the user can use the data pointer as its data instead of having to
4186 * allocate more memory for the reference.
4187 */
4188 void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
4189 unsigned long ip)
4190 {
4191 struct ftrace_func_entry *entry;
4192 struct ftrace_func_map *map;
4193
4194 entry = ftrace_lookup_ip(&mapper->hash, ip);
4195 if (!entry)
4196 return NULL;
4197
4198 map = (struct ftrace_func_map *)entry;
4199 return &map->data;
4200 }
4201
4202 /**
4203 * ftrace_func_mapper_add_ip - Map some data to an ip
4204 * @mapper: The mapper that has the ip maps
4205 * @ip: The instruction pointer address to map @data to
4206 * @data: The data to map to @ip
4207 *
4208 * Returns 0 on succes otherwise an error.
4209 */
4210 int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
4211 unsigned long ip, void *data)
4212 {
4213 struct ftrace_func_entry *entry;
4214 struct ftrace_func_map *map;
4215
4216 entry = ftrace_lookup_ip(&mapper->hash, ip);
4217 if (entry)
4218 return -EBUSY;
4219
4220 map = kmalloc(sizeof(*map), GFP_KERNEL);
4221 if (!map)
4222 return -ENOMEM;
4223
4224 map->entry.ip = ip;
4225 map->data = data;
4226
4227 __add_hash_entry(&mapper->hash, &map->entry);
4228
4229 return 0;
4230 }
4231
4232 /**
4233 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
4234 * @mapper: The mapper that has the ip maps
4235 * @ip: The instruction pointer address to remove the data from
4236 *
4237 * Returns the data if it is found, otherwise NULL.
4238 * Note, if the data pointer is used as the data itself, (see
4239 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
4240 * if the data pointer was set to zero.
4241 */
4242 void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
4243 unsigned long ip)
4244 {
4245 struct ftrace_func_entry *entry;
4246 struct ftrace_func_map *map;
4247 void *data;
4248
4249 entry = ftrace_lookup_ip(&mapper->hash, ip);
4250 if (!entry)
4251 return NULL;
4252
4253 map = (struct ftrace_func_map *)entry;
4254 data = map->data;
4255
4256 remove_hash_entry(&mapper->hash, entry);
4257 kfree(entry);
4258
4259 return data;
4260 }
4261
4262 /**
4263 * free_ftrace_func_mapper - free a mapping of ips and data
4264 * @mapper: The mapper that has the ip maps
4265 * @free_func: A function to be called on each data item.
4266 *
4267 * This is used to free the function mapper. The @free_func is optional
4268 * and can be used if the data needs to be freed as well.
4269 */
4270 void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
4271 ftrace_mapper_func free_func)
4272 {
4273 struct ftrace_func_entry *entry;
4274 struct ftrace_func_map *map;
4275 struct hlist_head *hhd;
4276 int size = 1 << mapper->hash.size_bits;
4277 int i;
4278
4279 if (free_func && mapper->hash.count) {
4280 for (i = 0; i < size; i++) {
4281 hhd = &mapper->hash.buckets[i];
4282 hlist_for_each_entry(entry, hhd, hlist) {
4283 map = (struct ftrace_func_map *)entry;
4284 free_func(map);
4285 }
4286 }
4287 }
4288 free_ftrace_hash(&mapper->hash);
4289 }
4290
4291 static void release_probe(struct ftrace_func_probe *probe)
4292 {
4293 struct ftrace_probe_ops *probe_ops;
4294
4295 mutex_lock(&ftrace_lock);
4296
4297 WARN_ON(probe->ref <= 0);
4298
4299 /* Subtract the ref that was used to protect this instance */
4300 probe->ref--;
4301
4302 if (!probe->ref) {
4303 probe_ops = probe->probe_ops;
4304 /*
4305 * Sending zero as ip tells probe_ops to free
4306 * the probe->data itself
4307 */
4308 if (probe_ops->free)
4309 probe_ops->free(probe_ops, probe->tr, 0, probe->data);
4310 list_del(&probe->list);
4311 kfree(probe);
4312 }
4313 mutex_unlock(&ftrace_lock);
4314 }
4315
4316 static void acquire_probe_locked(struct ftrace_func_probe *probe)
4317 {
4318 /*
4319 * Add one ref to keep it from being freed when releasing the
4320 * ftrace_lock mutex.
4321 */
4322 probe->ref++;
4323 }
4324
4325 int
4326 register_ftrace_function_probe(char *glob, struct trace_array *tr,
4327 struct ftrace_probe_ops *probe_ops,
4328 void *data)
4329 {
4330 struct ftrace_func_entry *entry;
4331 struct ftrace_func_probe *probe;
4332 struct ftrace_hash **orig_hash;
4333 struct ftrace_hash *old_hash;
4334 struct ftrace_hash *hash;
4335 int count = 0;
4336 int size;
4337 int ret;
4338 int i;
4339
4340 if (WARN_ON(!tr))
4341 return -EINVAL;
4342
4343 /* We do not support '!' for function probes */
4344 if (WARN_ON(glob[0] == '!'))
4345 return -EINVAL;
4346
4347
4348 mutex_lock(&ftrace_lock);
4349 /* Check if the probe_ops is already registered */
4350 list_for_each_entry(probe, &tr->func_probes, list) {
4351 if (probe->probe_ops == probe_ops)
4352 break;
4353 }
4354 if (&probe->list == &tr->func_probes) {
4355 probe = kzalloc(sizeof(*probe), GFP_KERNEL);
4356 if (!probe) {
4357 mutex_unlock(&ftrace_lock);
4358 return -ENOMEM;
4359 }
4360 probe->probe_ops = probe_ops;
4361 probe->ops.func = function_trace_probe_call;
4362 probe->tr = tr;
4363 ftrace_ops_init(&probe->ops);
4364 list_add(&probe->list, &tr->func_probes);
4365 }
4366
4367 acquire_probe_locked(probe);
4368
4369 mutex_unlock(&ftrace_lock);
4370
4371 mutex_lock(&probe->ops.func_hash->regex_lock);
4372
4373 orig_hash = &probe->ops.func_hash->filter_hash;
4374 old_hash = *orig_hash;
4375 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4376
4377 ret = ftrace_match_records(hash, glob, strlen(glob));
4378
4379 /* Nothing found? */
4380 if (!ret)
4381 ret = -EINVAL;
4382
4383 if (ret < 0)
4384 goto out;
4385
4386 size = 1 << hash->size_bits;
4387 for (i = 0; i < size; i++) {
4388 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4389 if (ftrace_lookup_ip(old_hash, entry->ip))
4390 continue;
4391 /*
4392 * The caller might want to do something special
4393 * for each function we find. We call the callback
4394 * to give the caller an opportunity to do so.
4395 */
4396 if (probe_ops->init) {
4397 ret = probe_ops->init(probe_ops, tr,
4398 entry->ip, data,
4399 &probe->data);
4400 if (ret < 0) {
4401 if (probe_ops->free && count)
4402 probe_ops->free(probe_ops, tr,
4403 0, probe->data);
4404 probe->data = NULL;
4405 goto out;
4406 }
4407 }
4408 count++;
4409 }
4410 }
4411
4412 mutex_lock(&ftrace_lock);
4413
4414 if (!count) {
4415 /* Nothing was added? */
4416 ret = -EINVAL;
4417 goto out_unlock;
4418 }
4419
4420 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4421 hash, 1);
4422 if (ret < 0)
4423 goto err_unlock;
4424
4425 /* One ref for each new function traced */
4426 probe->ref += count;
4427
4428 if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
4429 ret = ftrace_startup(&probe->ops, 0);
4430
4431 out_unlock:
4432 mutex_unlock(&ftrace_lock);
4433
4434 if (!ret)
4435 ret = count;
4436 out:
4437 mutex_unlock(&probe->ops.func_hash->regex_lock);
4438 free_ftrace_hash(hash);
4439
4440 release_probe(probe);
4441
4442 return ret;
4443
4444 err_unlock:
4445 if (!probe_ops->free || !count)
4446 goto out_unlock;
4447
4448 /* Failed to do the move, need to call the free functions */
4449 for (i = 0; i < size; i++) {
4450 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4451 if (ftrace_lookup_ip(old_hash, entry->ip))
4452 continue;
4453 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4454 }
4455 }
4456 goto out_unlock;
4457 }
4458
4459 int
4460 unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
4461 struct ftrace_probe_ops *probe_ops)
4462 {
4463 struct ftrace_ops_hash old_hash_ops;
4464 struct ftrace_func_entry *entry;
4465 struct ftrace_func_probe *probe;
4466 struct ftrace_glob func_g;
4467 struct ftrace_hash **orig_hash;
4468 struct ftrace_hash *old_hash;
4469 struct ftrace_hash *hash = NULL;
4470 struct hlist_node *tmp;
4471 struct hlist_head hhd;
4472 char str[KSYM_SYMBOL_LEN];
4473 int count = 0;
4474 int i, ret = -ENODEV;
4475 int size;
4476
4477 if (!glob || !strlen(glob) || !strcmp(glob, "*"))
4478 func_g.search = NULL;
4479 else {
4480 int not;
4481
4482 func_g.type = filter_parse_regex(glob, strlen(glob),
4483 &func_g.search, &not);
4484 func_g.len = strlen(func_g.search);
4485 func_g.search = glob;
4486
4487 /* we do not support '!' for function probes */
4488 if (WARN_ON(not))
4489 return -EINVAL;
4490 }
4491
4492 mutex_lock(&ftrace_lock);
4493 /* Check if the probe_ops is already registered */
4494 list_for_each_entry(probe, &tr->func_probes, list) {
4495 if (probe->probe_ops == probe_ops)
4496 break;
4497 }
4498 if (&probe->list == &tr->func_probes)
4499 goto err_unlock_ftrace;
4500
4501 ret = -EINVAL;
4502 if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
4503 goto err_unlock_ftrace;
4504
4505 acquire_probe_locked(probe);
4506
4507 mutex_unlock(&ftrace_lock);
4508
4509 mutex_lock(&probe->ops.func_hash->regex_lock);
4510
4511 orig_hash = &probe->ops.func_hash->filter_hash;
4512 old_hash = *orig_hash;
4513
4514 if (ftrace_hash_empty(old_hash))
4515 goto out_unlock;
4516
4517 old_hash_ops.filter_hash = old_hash;
4518 /* Probes only have filters */
4519 old_hash_ops.notrace_hash = NULL;
4520
4521 ret = -ENOMEM;
4522 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4523 if (!hash)
4524 goto out_unlock;
4525
4526 INIT_HLIST_HEAD(&hhd);
4527
4528 size = 1 << hash->size_bits;
4529 for (i = 0; i < size; i++) {
4530 hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
4531
4532 if (func_g.search) {
4533 kallsyms_lookup(entry->ip, NULL, NULL,
4534 NULL, str);
4535 if (!ftrace_match(str, &func_g))
4536 continue;
4537 }
4538 count++;
4539 remove_hash_entry(hash, entry);
4540 hlist_add_head(&entry->hlist, &hhd);
4541 }
4542 }
4543
4544 /* Nothing found? */
4545 if (!count) {
4546 ret = -EINVAL;
4547 goto out_unlock;
4548 }
4549
4550 mutex_lock(&ftrace_lock);
4551
4552 WARN_ON(probe->ref < count);
4553
4554 probe->ref -= count;
4555
4556 if (ftrace_hash_empty(hash))
4557 ftrace_shutdown(&probe->ops, 0);
4558
4559 ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4560 hash, 1);
4561
4562 /* still need to update the function call sites */
4563 if (ftrace_enabled && !ftrace_hash_empty(hash))
4564 ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
4565 &old_hash_ops);
4566 synchronize_sched();
4567
4568 hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
4569 hlist_del(&entry->hlist);
4570 if (probe_ops->free)
4571 probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4572 kfree(entry);
4573 }
4574 mutex_unlock(&ftrace_lock);
4575
4576 out_unlock:
4577 mutex_unlock(&probe->ops.func_hash->regex_lock);
4578 free_ftrace_hash(hash);
4579
4580 release_probe(probe);
4581
4582 return ret;
4583
4584 err_unlock_ftrace:
4585 mutex_unlock(&ftrace_lock);
4586 return ret;
4587 }
4588
4589 void clear_ftrace_function_probes(struct trace_array *tr)
4590 {
4591 struct ftrace_func_probe *probe, *n;
4592
4593 list_for_each_entry_safe(probe, n, &tr->func_probes, list)
4594 unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
4595 }
4596
4597 static LIST_HEAD(ftrace_commands);
4598 static DEFINE_MUTEX(ftrace_cmd_mutex);
4599
4600 /*
4601 * Currently we only register ftrace commands from __init, so mark this
4602 * __init too.
4603 */
4604 __init int register_ftrace_command(struct ftrace_func_command *cmd)
4605 {
4606 struct ftrace_func_command *p;
4607 int ret = 0;
4608
4609 mutex_lock(&ftrace_cmd_mutex);
4610 list_for_each_entry(p, &ftrace_commands, list) {
4611 if (strcmp(cmd->name, p->name) == 0) {
4612 ret = -EBUSY;
4613 goto out_unlock;
4614 }
4615 }
4616 list_add(&cmd->list, &ftrace_commands);
4617 out_unlock:
4618 mutex_unlock(&ftrace_cmd_mutex);
4619
4620 return ret;
4621 }
4622
4623 /*
4624 * Currently we only unregister ftrace commands from __init, so mark
4625 * this __init too.
4626 */
4627 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
4628 {
4629 struct ftrace_func_command *p, *n;
4630 int ret = -ENODEV;
4631
4632 mutex_lock(&ftrace_cmd_mutex);
4633 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4634 if (strcmp(cmd->name, p->name) == 0) {
4635 ret = 0;
4636 list_del_init(&p->list);
4637 goto out_unlock;
4638 }
4639 }
4640 out_unlock:
4641 mutex_unlock(&ftrace_cmd_mutex);
4642
4643 return ret;
4644 }
4645
4646 static int ftrace_process_regex(struct ftrace_iterator *iter,
4647 char *buff, int len, int enable)
4648 {
4649 struct ftrace_hash *hash = iter->hash;
4650 struct trace_array *tr = iter->ops->private;
4651 char *func, *command, *next = buff;
4652 struct ftrace_func_command *p;
4653 int ret = -EINVAL;
4654
4655 func = strsep(&next, ":");
4656
4657 if (!next) {
4658 ret = ftrace_match_records(hash, func, len);
4659 if (!ret)
4660 ret = -EINVAL;
4661 if (ret < 0)
4662 return ret;
4663 return 0;
4664 }
4665
4666 /* command found */
4667
4668 command = strsep(&next, ":");
4669
4670 mutex_lock(&ftrace_cmd_mutex);
4671 list_for_each_entry(p, &ftrace_commands, list) {
4672 if (strcmp(p->name, command) == 0) {
4673 ret = p->func(tr, hash, func, command, next, enable);
4674 goto out_unlock;
4675 }
4676 }
4677 out_unlock:
4678 mutex_unlock(&ftrace_cmd_mutex);
4679
4680 return ret;
4681 }
4682
4683 static ssize_t
4684 ftrace_regex_write(struct file *file, const char __user *ubuf,
4685 size_t cnt, loff_t *ppos, int enable)
4686 {
4687 struct ftrace_iterator *iter;
4688 struct trace_parser *parser;
4689 ssize_t ret, read;
4690
4691 if (!cnt)
4692 return 0;
4693
4694 if (file->f_mode & FMODE_READ) {
4695 struct seq_file *m = file->private_data;
4696 iter = m->private;
4697 } else
4698 iter = file->private_data;
4699
4700 if (unlikely(ftrace_disabled))
4701 return -ENODEV;
4702
4703 /* iter->hash is a local copy, so we don't need regex_lock */
4704
4705 parser = &iter->parser;
4706 read = trace_get_user(parser, ubuf, cnt, ppos);
4707
4708 if (read >= 0 && trace_parser_loaded(parser) &&
4709 !trace_parser_cont(parser)) {
4710 ret = ftrace_process_regex(iter, parser->buffer,
4711 parser->idx, enable);
4712 trace_parser_clear(parser);
4713 if (ret < 0)
4714 goto out;
4715 }
4716
4717 ret = read;
4718 out:
4719 return ret;
4720 }
4721
4722 ssize_t
4723 ftrace_filter_write(struct file *file, const char __user *ubuf,
4724 size_t cnt, loff_t *ppos)
4725 {
4726 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4727 }
4728
4729 ssize_t
4730 ftrace_notrace_write(struct file *file, const char __user *ubuf,
4731 size_t cnt, loff_t *ppos)
4732 {
4733 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4734 }
4735
4736 static int
4737 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4738 {
4739 struct ftrace_func_entry *entry;
4740
4741 if (!ftrace_location(ip))
4742 return -EINVAL;
4743
4744 if (remove) {
4745 entry = ftrace_lookup_ip(hash, ip);
4746 if (!entry)
4747 return -ENOENT;
4748 free_hash_entry(hash, entry);
4749 return 0;
4750 }
4751
4752 return add_hash_entry(hash, ip);
4753 }
4754
4755 static int
4756 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4757 unsigned long ip, int remove, int reset, int enable)
4758 {
4759 struct ftrace_hash **orig_hash;
4760 struct ftrace_hash *hash;
4761 int ret;
4762
4763 if (unlikely(ftrace_disabled))
4764 return -ENODEV;
4765
4766 mutex_lock(&ops->func_hash->regex_lock);
4767
4768 if (enable)
4769 orig_hash = &ops->func_hash->filter_hash;
4770 else
4771 orig_hash = &ops->func_hash->notrace_hash;
4772
4773 if (reset)
4774 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4775 else
4776 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4777
4778 if (!hash) {
4779 ret = -ENOMEM;
4780 goto out_regex_unlock;
4781 }
4782
4783 if (buf && !ftrace_match_records(hash, buf, len)) {
4784 ret = -EINVAL;
4785 goto out_regex_unlock;
4786 }
4787 if (ip) {
4788 ret = ftrace_match_addr(hash, ip, remove);
4789 if (ret < 0)
4790 goto out_regex_unlock;
4791 }
4792
4793 mutex_lock(&ftrace_lock);
4794 ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4795 mutex_unlock(&ftrace_lock);
4796
4797 out_regex_unlock:
4798 mutex_unlock(&ops->func_hash->regex_lock);
4799
4800 free_ftrace_hash(hash);
4801 return ret;
4802 }
4803
4804 static int
4805 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4806 int reset, int enable)
4807 {
4808 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4809 }
4810
4811 /**
4812 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4813 * @ops - the ops to set the filter with
4814 * @ip - the address to add to or remove from the filter.
4815 * @remove - non zero to remove the ip from the filter
4816 * @reset - non zero to reset all filters before applying this filter.
4817 *
4818 * Filters denote which functions should be enabled when tracing is enabled
4819 * If @ip is NULL, it failes to update filter.
4820 */
4821 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4822 int remove, int reset)
4823 {
4824 ftrace_ops_init(ops);
4825 return ftrace_set_addr(ops, ip, remove, reset, 1);
4826 }
4827 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4828
4829 /**
4830 * ftrace_ops_set_global_filter - setup ops to use global filters
4831 * @ops - the ops which will use the global filters
4832 *
4833 * ftrace users who need global function trace filtering should call this.
4834 * It can set the global filter only if ops were not initialized before.
4835 */
4836 void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
4837 {
4838 if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
4839 return;
4840
4841 ftrace_ops_init(ops);
4842 ops->func_hash = &global_ops.local_hash;
4843 }
4844 EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
4845
4846 static int
4847 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4848 int reset, int enable)
4849 {
4850 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4851 }
4852
4853 /**
4854 * ftrace_set_filter - set a function to filter on in ftrace
4855 * @ops - the ops to set the filter with
4856 * @buf - the string that holds the function filter text.
4857 * @len - the length of the string.
4858 * @reset - non zero to reset all filters before applying this filter.
4859 *
4860 * Filters denote which functions should be enabled when tracing is enabled.
4861 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4862 */
4863 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4864 int len, int reset)
4865 {
4866 ftrace_ops_init(ops);
4867 return ftrace_set_regex(ops, buf, len, reset, 1);
4868 }
4869 EXPORT_SYMBOL_GPL(ftrace_set_filter);
4870
4871 /**
4872 * ftrace_set_notrace - set a function to not trace in ftrace
4873 * @ops - the ops to set the notrace filter with
4874 * @buf - the string that holds the function notrace text.
4875 * @len - the length of the string.
4876 * @reset - non zero to reset all filters before applying this filter.
4877 *
4878 * Notrace Filters denote which functions should not be enabled when tracing
4879 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4880 * for tracing.
4881 */
4882 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4883 int len, int reset)
4884 {
4885 ftrace_ops_init(ops);
4886 return ftrace_set_regex(ops, buf, len, reset, 0);
4887 }
4888 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4889 /**
4890 * ftrace_set_global_filter - set a function to filter on with global tracers
4891 * @buf - the string that holds the function filter text.
4892 * @len - the length of the string.
4893 * @reset - non zero to reset all filters before applying this filter.
4894 *
4895 * Filters denote which functions should be enabled when tracing is enabled.
4896 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4897 */
4898 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4899 {
4900 ftrace_set_regex(&global_ops, buf, len, reset, 1);
4901 }
4902 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4903
4904 /**
4905 * ftrace_set_global_notrace - set a function to not trace with global tracers
4906 * @buf - the string that holds the function notrace text.
4907 * @len - the length of the string.
4908 * @reset - non zero to reset all filters before applying this filter.
4909 *
4910 * Notrace Filters denote which functions should not be enabled when tracing
4911 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4912 * for tracing.
4913 */
4914 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4915 {
4916 ftrace_set_regex(&global_ops, buf, len, reset, 0);
4917 }
4918 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4919
4920 /*
4921 * command line interface to allow users to set filters on boot up.
4922 */
4923 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
4924 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4925 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4926
4927 /* Used by function selftest to not test if filter is set */
4928 bool ftrace_filter_param __initdata;
4929
4930 static int __init set_ftrace_notrace(char *str)
4931 {
4932 ftrace_filter_param = true;
4933 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4934 return 1;
4935 }
4936 __setup("ftrace_notrace=", set_ftrace_notrace);
4937
4938 static int __init set_ftrace_filter(char *str)
4939 {
4940 ftrace_filter_param = true;
4941 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4942 return 1;
4943 }
4944 __setup("ftrace_filter=", set_ftrace_filter);
4945
4946 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4947 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4948 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4949 static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
4950
4951 static unsigned long save_global_trampoline;
4952 static unsigned long save_global_flags;
4953
4954 static int __init set_graph_function(char *str)
4955 {
4956 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4957 return 1;
4958 }
4959 __setup("ftrace_graph_filter=", set_graph_function);
4960
4961 static int __init set_graph_notrace_function(char *str)
4962 {
4963 strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4964 return 1;
4965 }
4966 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
4967
4968 static int __init set_graph_max_depth_function(char *str)
4969 {
4970 if (!str)
4971 return 0;
4972 fgraph_max_depth = simple_strtoul(str, NULL, 0);
4973 return 1;
4974 }
4975 __setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
4976
4977 static void __init set_ftrace_early_graph(char *buf, int enable)
4978 {
4979 int ret;
4980 char *func;
4981 struct ftrace_hash *hash;
4982
4983 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4984 if (WARN_ON(!hash))
4985 return;
4986
4987 while (buf) {
4988 func = strsep(&buf, ",");
4989 /* we allow only one expression at a time */
4990 ret = ftrace_graph_set_hash(hash, func);
4991 if (ret)
4992 printk(KERN_DEBUG "ftrace: function %s not "
4993 "traceable\n", func);
4994 }
4995
4996 if (enable)
4997 ftrace_graph_hash = hash;
4998 else
4999 ftrace_graph_notrace_hash = hash;
5000 }
5001 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5002
5003 void __init
5004 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
5005 {
5006 char *func;
5007
5008 ftrace_ops_init(ops);
5009
5010 while (buf) {
5011 func = strsep(&buf, ",");
5012 ftrace_set_regex(ops, func, strlen(func), 0, enable);
5013 }
5014 }
5015
5016 static void __init set_ftrace_early_filters(void)
5017 {
5018 if (ftrace_filter_buf[0])
5019 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
5020 if (ftrace_notrace_buf[0])
5021 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
5022 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5023 if (ftrace_graph_buf[0])
5024 set_ftrace_early_graph(ftrace_graph_buf, 1);
5025 if (ftrace_graph_notrace_buf[0])
5026 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
5027 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5028 }
5029
5030 int ftrace_regex_release(struct inode *inode, struct file *file)
5031 {
5032 struct seq_file *m = (struct seq_file *)file->private_data;
5033 struct ftrace_iterator *iter;
5034 struct ftrace_hash **orig_hash;
5035 struct trace_parser *parser;
5036 int filter_hash;
5037 int ret;
5038
5039 if (file->f_mode & FMODE_READ) {
5040 iter = m->private;
5041 seq_release(inode, file);
5042 } else
5043 iter = file->private_data;
5044
5045 parser = &iter->parser;
5046 if (trace_parser_loaded(parser)) {
5047 parser->buffer[parser->idx] = 0;
5048 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
5049 }
5050
5051 trace_parser_put(parser);
5052
5053 mutex_lock(&iter->ops->func_hash->regex_lock);
5054
5055 if (file->f_mode & FMODE_WRITE) {
5056 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
5057
5058 if (filter_hash) {
5059 orig_hash = &iter->ops->func_hash->filter_hash;
5060 if (iter->tr && !list_empty(&iter->tr->mod_trace))
5061 iter->hash->flags |= FTRACE_HASH_FL_MOD;
5062 } else
5063 orig_hash = &iter->ops->func_hash->notrace_hash;
5064
5065 mutex_lock(&ftrace_lock);
5066 ret = ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
5067 iter->hash, filter_hash);
5068 mutex_unlock(&ftrace_lock);
5069 } else {
5070 /* For read only, the hash is the ops hash */
5071 iter->hash = NULL;
5072 }
5073
5074 mutex_unlock(&iter->ops->func_hash->regex_lock);
5075 free_ftrace_hash(iter->hash);
5076 kfree(iter);
5077
5078 return 0;
5079 }
5080
5081 static const struct file_operations ftrace_avail_fops = {
5082 .open = ftrace_avail_open,
5083 .read = seq_read,
5084 .llseek = seq_lseek,
5085 .release = seq_release_private,
5086 };
5087
5088 static const struct file_operations ftrace_enabled_fops = {
5089 .open = ftrace_enabled_open,
5090 .read = seq_read,
5091 .llseek = seq_lseek,
5092 .release = seq_release_private,
5093 };
5094
5095 static const struct file_operations ftrace_filter_fops = {
5096 .open = ftrace_filter_open,
5097 .read = seq_read,
5098 .write = ftrace_filter_write,
5099 .llseek = tracing_lseek,
5100 .release = ftrace_regex_release,
5101 };
5102
5103 static const struct file_operations ftrace_notrace_fops = {
5104 .open = ftrace_notrace_open,
5105 .read = seq_read,
5106 .write = ftrace_notrace_write,
5107 .llseek = tracing_lseek,
5108 .release = ftrace_regex_release,
5109 };
5110
5111 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5112
5113 static DEFINE_MUTEX(graph_lock);
5114
5115 struct ftrace_hash *ftrace_graph_hash = EMPTY_HASH;
5116 struct ftrace_hash *ftrace_graph_notrace_hash = EMPTY_HASH;
5117
5118 enum graph_filter_type {
5119 GRAPH_FILTER_NOTRACE = 0,
5120 GRAPH_FILTER_FUNCTION,
5121 };
5122
5123 #define FTRACE_GRAPH_EMPTY ((void *)1)
5124
5125 struct ftrace_graph_data {
5126 struct ftrace_hash *hash;
5127 struct ftrace_func_entry *entry;
5128 int idx; /* for hash table iteration */
5129 enum graph_filter_type type;
5130 struct ftrace_hash *new_hash;
5131 const struct seq_operations *seq_ops;
5132 struct trace_parser parser;
5133 };
5134
5135 static void *
5136 __g_next(struct seq_file *m, loff_t *pos)
5137 {
5138 struct ftrace_graph_data *fgd = m->private;
5139 struct ftrace_func_entry *entry = fgd->entry;
5140 struct hlist_head *head;
5141 int i, idx = fgd->idx;
5142
5143 if (*pos >= fgd->hash->count)
5144 return NULL;
5145
5146 if (entry) {
5147 hlist_for_each_entry_continue(entry, hlist) {
5148 fgd->entry = entry;
5149 return entry;
5150 }
5151
5152 idx++;
5153 }
5154
5155 for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
5156 head = &fgd->hash->buckets[i];
5157 hlist_for_each_entry(entry, head, hlist) {
5158 fgd->entry = entry;
5159 fgd->idx = i;
5160 return entry;
5161 }
5162 }
5163 return NULL;
5164 }
5165
5166 static void *
5167 g_next(struct seq_file *m, void *v, loff_t *pos)
5168 {
5169 (*pos)++;
5170 return __g_next(m, pos);
5171 }
5172
5173 static void *g_start(struct seq_file *m, loff_t *pos)
5174 {
5175 struct ftrace_graph_data *fgd = m->private;
5176
5177 mutex_lock(&graph_lock);
5178
5179 if (fgd->type == GRAPH_FILTER_FUNCTION)
5180 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5181 lockdep_is_held(&graph_lock));
5182 else
5183 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5184 lockdep_is_held(&graph_lock));
5185
5186 /* Nothing, tell g_show to print all functions are enabled */
5187 if (ftrace_hash_empty(fgd->hash) && !*pos)
5188 return FTRACE_GRAPH_EMPTY;
5189
5190 fgd->idx = 0;
5191 fgd->entry = NULL;
5192 return __g_next(m, pos);
5193 }
5194
5195 static void g_stop(struct seq_file *m, void *p)
5196 {
5197 mutex_unlock(&graph_lock);
5198 }
5199
5200 static int g_show(struct seq_file *m, void *v)
5201 {
5202 struct ftrace_func_entry *entry = v;
5203
5204 if (!entry)
5205 return 0;
5206
5207 if (entry == FTRACE_GRAPH_EMPTY) {
5208 struct ftrace_graph_data *fgd = m->private;
5209
5210 if (fgd->type == GRAPH_FILTER_FUNCTION)
5211 seq_puts(m, "#### all functions enabled ####\n");
5212 else
5213 seq_puts(m, "#### no functions disabled ####\n");
5214 return 0;
5215 }
5216
5217 seq_printf(m, "%ps\n", (void *)entry->ip);
5218
5219 return 0;
5220 }
5221
5222 static const struct seq_operations ftrace_graph_seq_ops = {
5223 .start = g_start,
5224 .next = g_next,
5225 .stop = g_stop,
5226 .show = g_show,
5227 };
5228
5229 static int
5230 __ftrace_graph_open(struct inode *inode, struct file *file,
5231 struct ftrace_graph_data *fgd)
5232 {
5233 int ret = 0;
5234 struct ftrace_hash *new_hash = NULL;
5235
5236 if (file->f_mode & FMODE_WRITE) {
5237 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
5238
5239 if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
5240 return -ENOMEM;
5241
5242 if (file->f_flags & O_TRUNC)
5243 new_hash = alloc_ftrace_hash(size_bits);
5244 else
5245 new_hash = alloc_and_copy_ftrace_hash(size_bits,
5246 fgd->hash);
5247 if (!new_hash) {
5248 ret = -ENOMEM;
5249 goto out;
5250 }
5251 }
5252
5253 if (file->f_mode & FMODE_READ) {
5254 ret = seq_open(file, &ftrace_graph_seq_ops);
5255 if (!ret) {
5256 struct seq_file *m = file->private_data;
5257 m->private = fgd;
5258 } else {
5259 /* Failed */
5260 free_ftrace_hash(new_hash);
5261 new_hash = NULL;
5262 }
5263 } else
5264 file->private_data = fgd;
5265
5266 out:
5267 if (ret < 0 && file->f_mode & FMODE_WRITE)
5268 trace_parser_put(&fgd->parser);
5269
5270 fgd->new_hash = new_hash;
5271
5272 /*
5273 * All uses of fgd->hash must be taken with the graph_lock
5274 * held. The graph_lock is going to be released, so force
5275 * fgd->hash to be reinitialized when it is taken again.
5276 */
5277 fgd->hash = NULL;
5278
5279 return ret;
5280 }
5281
5282 static int
5283 ftrace_graph_open(struct inode *inode, struct file *file)
5284 {
5285 struct ftrace_graph_data *fgd;
5286 int ret;
5287
5288 if (unlikely(ftrace_disabled))
5289 return -ENODEV;
5290
5291 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5292 if (fgd == NULL)
5293 return -ENOMEM;
5294
5295 mutex_lock(&graph_lock);
5296
5297 fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5298 lockdep_is_held(&graph_lock));
5299 fgd->type = GRAPH_FILTER_FUNCTION;
5300 fgd->seq_ops = &ftrace_graph_seq_ops;
5301
5302 ret = __ftrace_graph_open(inode, file, fgd);
5303 if (ret < 0)
5304 kfree(fgd);
5305
5306 mutex_unlock(&graph_lock);
5307 return ret;
5308 }
5309
5310 static int
5311 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
5312 {
5313 struct ftrace_graph_data *fgd;
5314 int ret;
5315
5316 if (unlikely(ftrace_disabled))
5317 return -ENODEV;
5318
5319 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5320 if (fgd == NULL)
5321 return -ENOMEM;
5322
5323 mutex_lock(&graph_lock);
5324
5325 fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5326 lockdep_is_held(&graph_lock));
5327 fgd->type = GRAPH_FILTER_NOTRACE;
5328 fgd->seq_ops = &ftrace_graph_seq_ops;
5329
5330 ret = __ftrace_graph_open(inode, file, fgd);
5331 if (ret < 0)
5332 kfree(fgd);
5333
5334 mutex_unlock(&graph_lock);
5335 return ret;
5336 }
5337
5338 static int
5339 ftrace_graph_release(struct inode *inode, struct file *file)
5340 {
5341 struct ftrace_graph_data *fgd;
5342 struct ftrace_hash *old_hash, *new_hash;
5343 struct trace_parser *parser;
5344 int ret = 0;
5345
5346 if (file->f_mode & FMODE_READ) {
5347 struct seq_file *m = file->private_data;
5348
5349 fgd = m->private;
5350 seq_release(inode, file);
5351 } else {
5352 fgd = file->private_data;
5353 }
5354
5355
5356 if (file->f_mode & FMODE_WRITE) {
5357
5358 parser = &fgd->parser;
5359
5360 if (trace_parser_loaded((parser))) {
5361 parser->buffer[parser->idx] = 0;
5362 ret = ftrace_graph_set_hash(fgd->new_hash,
5363 parser->buffer);
5364 }
5365
5366 trace_parser_put(parser);
5367
5368 new_hash = __ftrace_hash_move(fgd->new_hash);
5369 if (!new_hash) {
5370 ret = -ENOMEM;
5371 goto out;
5372 }
5373
5374 mutex_lock(&graph_lock);
5375
5376 if (fgd->type == GRAPH_FILTER_FUNCTION) {
5377 old_hash = rcu_dereference_protected(ftrace_graph_hash,
5378 lockdep_is_held(&graph_lock));
5379 rcu_assign_pointer(ftrace_graph_hash, new_hash);
5380 } else {
5381 old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5382 lockdep_is_held(&graph_lock));
5383 rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
5384 }
5385
5386 mutex_unlock(&graph_lock);
5387
5388 /* Wait till all users are no longer using the old hash */
5389 synchronize_sched();
5390
5391 free_ftrace_hash(old_hash);
5392 }
5393
5394 out:
5395 free_ftrace_hash(fgd->new_hash);
5396 kfree(fgd);
5397
5398 return ret;
5399 }
5400
5401 static int
5402 ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
5403 {
5404 struct ftrace_glob func_g;
5405 struct dyn_ftrace *rec;
5406 struct ftrace_page *pg;
5407 struct ftrace_func_entry *entry;
5408 int fail = 1;
5409 int not;
5410
5411 /* decode regex */
5412 func_g.type = filter_parse_regex(buffer, strlen(buffer),
5413 &func_g.search, &not);
5414
5415 func_g.len = strlen(func_g.search);
5416
5417 mutex_lock(&ftrace_lock);
5418
5419 if (unlikely(ftrace_disabled)) {
5420 mutex_unlock(&ftrace_lock);
5421 return -ENODEV;
5422 }
5423
5424 do_for_each_ftrace_rec(pg, rec) {
5425
5426 if (rec->flags & FTRACE_FL_DISABLED)
5427 continue;
5428
5429 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
5430 entry = ftrace_lookup_ip(hash, rec->ip);
5431
5432 if (!not) {
5433 fail = 0;
5434
5435 if (entry)
5436 continue;
5437 if (add_hash_entry(hash, rec->ip) < 0)
5438 goto out;
5439 } else {
5440 if (entry) {
5441 free_hash_entry(hash, entry);
5442 fail = 0;
5443 }
5444 }
5445 }
5446 } while_for_each_ftrace_rec();
5447 out:
5448 mutex_unlock(&ftrace_lock);
5449
5450 if (fail)
5451 return -EINVAL;
5452
5453 return 0;
5454 }
5455
5456 static ssize_t
5457 ftrace_graph_write(struct file *file, const char __user *ubuf,
5458 size_t cnt, loff_t *ppos)
5459 {
5460 ssize_t read, ret = 0;
5461 struct ftrace_graph_data *fgd = file->private_data;
5462 struct trace_parser *parser;
5463
5464 if (!cnt)
5465 return 0;
5466
5467 /* Read mode uses seq functions */
5468 if (file->f_mode & FMODE_READ) {
5469 struct seq_file *m = file->private_data;
5470 fgd = m->private;
5471 }
5472
5473 parser = &fgd->parser;
5474
5475 read = trace_get_user(parser, ubuf, cnt, ppos);
5476
5477 if (read >= 0 && trace_parser_loaded(parser) &&
5478 !trace_parser_cont(parser)) {
5479
5480 ret = ftrace_graph_set_hash(fgd->new_hash,
5481 parser->buffer);
5482 trace_parser_clear(parser);
5483 }
5484
5485 if (!ret)
5486 ret = read;
5487
5488 return ret;
5489 }
5490
5491 static const struct file_operations ftrace_graph_fops = {
5492 .open = ftrace_graph_open,
5493 .read = seq_read,
5494 .write = ftrace_graph_write,
5495 .llseek = tracing_lseek,
5496 .release = ftrace_graph_release,
5497 };
5498
5499 static const struct file_operations ftrace_graph_notrace_fops = {
5500 .open = ftrace_graph_notrace_open,
5501 .read = seq_read,
5502 .write = ftrace_graph_write,
5503 .llseek = tracing_lseek,
5504 .release = ftrace_graph_release,
5505 };
5506 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5507
5508 void ftrace_create_filter_files(struct ftrace_ops *ops,
5509 struct dentry *parent)
5510 {
5511
5512 trace_create_file("set_ftrace_filter", 0644, parent,
5513 ops, &ftrace_filter_fops);
5514
5515 trace_create_file("set_ftrace_notrace", 0644, parent,
5516 ops, &ftrace_notrace_fops);
5517 }
5518
5519 /*
5520 * The name "destroy_filter_files" is really a misnomer. Although
5521 * in the future, it may actualy delete the files, but this is
5522 * really intended to make sure the ops passed in are disabled
5523 * and that when this function returns, the caller is free to
5524 * free the ops.
5525 *
5526 * The "destroy" name is only to match the "create" name that this
5527 * should be paired with.
5528 */
5529 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
5530 {
5531 mutex_lock(&ftrace_lock);
5532 if (ops->flags & FTRACE_OPS_FL_ENABLED)
5533 ftrace_shutdown(ops, 0);
5534 ops->flags |= FTRACE_OPS_FL_DELETED;
5535 mutex_unlock(&ftrace_lock);
5536 }
5537
5538 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
5539 {
5540
5541 trace_create_file("available_filter_functions", 0444,
5542 d_tracer, NULL, &ftrace_avail_fops);
5543
5544 trace_create_file("enabled_functions", 0444,
5545 d_tracer, NULL, &ftrace_enabled_fops);
5546
5547 ftrace_create_filter_files(&global_ops, d_tracer);
5548
5549 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5550 trace_create_file("set_graph_function", 0444, d_tracer,
5551 NULL,
5552 &ftrace_graph_fops);
5553 trace_create_file("set_graph_notrace", 0444, d_tracer,
5554 NULL,
5555 &ftrace_graph_notrace_fops);
5556 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5557
5558 return 0;
5559 }
5560
5561 static int ftrace_cmp_ips(const void *a, const void *b)
5562 {
5563 const unsigned long *ipa = a;
5564 const unsigned long *ipb = b;
5565
5566 if (*ipa > *ipb)
5567 return 1;
5568 if (*ipa < *ipb)
5569 return -1;
5570 return 0;
5571 }
5572
5573 static int ftrace_process_locs(struct module *mod,
5574 unsigned long *start,
5575 unsigned long *end)
5576 {
5577 struct ftrace_page *start_pg;
5578 struct ftrace_page *pg;
5579 struct dyn_ftrace *rec;
5580 unsigned long count;
5581 unsigned long *p;
5582 unsigned long addr;
5583 unsigned long flags = 0; /* Shut up gcc */
5584 int ret = -ENOMEM;
5585
5586 count = end - start;
5587
5588 if (!count)
5589 return 0;
5590
5591 sort(start, count, sizeof(*start),
5592 ftrace_cmp_ips, NULL);
5593
5594 start_pg = ftrace_allocate_pages(count);
5595 if (!start_pg)
5596 return -ENOMEM;
5597
5598 mutex_lock(&ftrace_lock);
5599
5600 /*
5601 * Core and each module needs their own pages, as
5602 * modules will free them when they are removed.
5603 * Force a new page to be allocated for modules.
5604 */
5605 if (!mod) {
5606 WARN_ON(ftrace_pages || ftrace_pages_start);
5607 /* First initialization */
5608 ftrace_pages = ftrace_pages_start = start_pg;
5609 } else {
5610 if (!ftrace_pages)
5611 goto out;
5612
5613 if (WARN_ON(ftrace_pages->next)) {
5614 /* Hmm, we have free pages? */
5615 while (ftrace_pages->next)
5616 ftrace_pages = ftrace_pages->next;
5617 }
5618
5619 ftrace_pages->next = start_pg;
5620 }
5621
5622 p = start;
5623 pg = start_pg;
5624 while (p < end) {
5625 addr = ftrace_call_adjust(*p++);
5626 /*
5627 * Some architecture linkers will pad between
5628 * the different mcount_loc sections of different
5629 * object files to satisfy alignments.
5630 * Skip any NULL pointers.
5631 */
5632 if (!addr)
5633 continue;
5634
5635 if (pg->index == pg->size) {
5636 /* We should have allocated enough */
5637 if (WARN_ON(!pg->next))
5638 break;
5639 pg = pg->next;
5640 }
5641
5642 rec = &pg->records[pg->index++];
5643 rec->ip = addr;
5644 }
5645
5646 /* We should have used all pages */
5647 WARN_ON(pg->next);
5648
5649 /* Assign the last page to ftrace_pages */
5650 ftrace_pages = pg;
5651
5652 /*
5653 * We only need to disable interrupts on start up
5654 * because we are modifying code that an interrupt
5655 * may execute, and the modification is not atomic.
5656 * But for modules, nothing runs the code we modify
5657 * until we are finished with it, and there's no
5658 * reason to cause large interrupt latencies while we do it.
5659 */
5660 if (!mod)
5661 local_irq_save(flags);
5662 ftrace_update_code(mod, start_pg);
5663 if (!mod)
5664 local_irq_restore(flags);
5665 ret = 0;
5666 out:
5667 mutex_unlock(&ftrace_lock);
5668
5669 return ret;
5670 }
5671
5672 #ifdef CONFIG_MODULES
5673
5674 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
5675
5676 static int referenced_filters(struct dyn_ftrace *rec)
5677 {
5678 struct ftrace_ops *ops;
5679 int cnt = 0;
5680
5681 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
5682 if (ops_references_rec(ops, rec))
5683 cnt++;
5684 }
5685
5686 return cnt;
5687 }
5688
5689 void ftrace_release_mod(struct module *mod)
5690 {
5691 struct dyn_ftrace *rec;
5692 struct ftrace_page **last_pg;
5693 struct ftrace_page *pg;
5694 int order;
5695
5696 mutex_lock(&ftrace_lock);
5697
5698 if (ftrace_disabled)
5699 goto out_unlock;
5700
5701 /*
5702 * Each module has its own ftrace_pages, remove
5703 * them from the list.
5704 */
5705 last_pg = &ftrace_pages_start;
5706 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
5707 rec = &pg->records[0];
5708 if (within_module_core(rec->ip, mod)) {
5709 /*
5710 * As core pages are first, the first
5711 * page should never be a module page.
5712 */
5713 if (WARN_ON(pg == ftrace_pages_start))
5714 goto out_unlock;
5715
5716 /* Check if we are deleting the last page */
5717 if (pg == ftrace_pages)
5718 ftrace_pages = next_to_ftrace_page(last_pg);
5719
5720 ftrace_update_tot_cnt -= pg->index;
5721 *last_pg = pg->next;
5722 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
5723 free_pages((unsigned long)pg->records, order);
5724 kfree(pg);
5725 } else
5726 last_pg = &pg->next;
5727 }
5728 out_unlock:
5729 mutex_unlock(&ftrace_lock);
5730 }
5731
5732 void ftrace_module_enable(struct module *mod)
5733 {
5734 struct dyn_ftrace *rec;
5735 struct ftrace_page *pg;
5736
5737 mutex_lock(&ftrace_lock);
5738
5739 if (ftrace_disabled)
5740 goto out_unlock;
5741
5742 /*
5743 * If the tracing is enabled, go ahead and enable the record.
5744 *
5745 * The reason not to enable the record immediatelly is the
5746 * inherent check of ftrace_make_nop/ftrace_make_call for
5747 * correct previous instructions. Making first the NOP
5748 * conversion puts the module to the correct state, thus
5749 * passing the ftrace_make_call check.
5750 *
5751 * We also delay this to after the module code already set the
5752 * text to read-only, as we now need to set it back to read-write
5753 * so that we can modify the text.
5754 */
5755 if (ftrace_start_up)
5756 ftrace_arch_code_modify_prepare();
5757
5758 do_for_each_ftrace_rec(pg, rec) {
5759 int cnt;
5760 /*
5761 * do_for_each_ftrace_rec() is a double loop.
5762 * module text shares the pg. If a record is
5763 * not part of this module, then skip this pg,
5764 * which the "break" will do.
5765 */
5766 if (!within_module_core(rec->ip, mod))
5767 break;
5768
5769 cnt = 0;
5770
5771 /*
5772 * When adding a module, we need to check if tracers are
5773 * currently enabled and if they are, and can trace this record,
5774 * we need to enable the module functions as well as update the
5775 * reference counts for those function records.
5776 */
5777 if (ftrace_start_up)
5778 cnt += referenced_filters(rec);
5779
5780 /* This clears FTRACE_FL_DISABLED */
5781 rec->flags = cnt;
5782
5783 if (ftrace_start_up && cnt) {
5784 int failed = __ftrace_replace_code(rec, 1);
5785 if (failed) {
5786 ftrace_bug(failed, rec);
5787 goto out_loop;
5788 }
5789 }
5790
5791 } while_for_each_ftrace_rec();
5792
5793 out_loop:
5794 if (ftrace_start_up)
5795 ftrace_arch_code_modify_post_process();
5796
5797 out_unlock:
5798 mutex_unlock(&ftrace_lock);
5799
5800 process_cached_mods(mod->name);
5801 }
5802
5803 void ftrace_module_init(struct module *mod)
5804 {
5805 if (ftrace_disabled || !mod->num_ftrace_callsites)
5806 return;
5807
5808 ftrace_process_locs(mod, mod->ftrace_callsites,
5809 mod->ftrace_callsites + mod->num_ftrace_callsites);
5810 }
5811 #endif /* CONFIG_MODULES */
5812
5813 void __init ftrace_free_init_mem(void)
5814 {
5815 unsigned long start = (unsigned long)(&__init_begin);
5816 unsigned long end = (unsigned long)(&__init_end);
5817 struct ftrace_page **last_pg = &ftrace_pages_start;
5818 struct ftrace_page *pg;
5819 struct dyn_ftrace *rec;
5820 struct dyn_ftrace key;
5821 int order;
5822
5823 key.ip = start;
5824 key.flags = end; /* overload flags, as it is unsigned long */
5825
5826 mutex_lock(&ftrace_lock);
5827
5828 for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
5829 if (end < pg->records[0].ip ||
5830 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
5831 continue;
5832 again:
5833 rec = bsearch(&key, pg->records, pg->index,
5834 sizeof(struct dyn_ftrace),
5835 ftrace_cmp_recs);
5836 if (!rec)
5837 continue;
5838 pg->index--;
5839 ftrace_update_tot_cnt--;
5840 if (!pg->index) {
5841 *last_pg = pg->next;
5842 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
5843 free_pages((unsigned long)pg->records, order);
5844 kfree(pg);
5845 pg = container_of(last_pg, struct ftrace_page, next);
5846 if (!(*last_pg))
5847 ftrace_pages = pg;
5848 continue;
5849 }
5850 memmove(rec, rec + 1,
5851 (pg->index - (rec - pg->records)) * sizeof(*rec));
5852 /* More than one function may be in this block */
5853 goto again;
5854 }
5855 mutex_unlock(&ftrace_lock);
5856 }
5857
5858 void __init ftrace_init(void)
5859 {
5860 extern unsigned long __start_mcount_loc[];
5861 extern unsigned long __stop_mcount_loc[];
5862 unsigned long count, flags;
5863 int ret;
5864
5865 local_irq_save(flags);
5866 ret = ftrace_dyn_arch_init();
5867 local_irq_restore(flags);
5868 if (ret)
5869 goto failed;
5870
5871 count = __stop_mcount_loc - __start_mcount_loc;
5872 if (!count) {
5873 pr_info("ftrace: No functions to be traced?\n");
5874 goto failed;
5875 }
5876
5877 pr_info("ftrace: allocating %ld entries in %ld pages\n",
5878 count, count / ENTRIES_PER_PAGE + 1);
5879
5880 last_ftrace_enabled = ftrace_enabled = 1;
5881
5882 ret = ftrace_process_locs(NULL,
5883 __start_mcount_loc,
5884 __stop_mcount_loc);
5885
5886 set_ftrace_early_filters();
5887
5888 return;
5889 failed:
5890 ftrace_disabled = 1;
5891 }
5892
5893 /* Do nothing if arch does not support this */
5894 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
5895 {
5896 }
5897
5898 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5899 {
5900 arch_ftrace_update_trampoline(ops);
5901 }
5902
5903 void ftrace_init_trace_array(struct trace_array *tr)
5904 {
5905 INIT_LIST_HEAD(&tr->func_probes);
5906 INIT_LIST_HEAD(&tr->mod_trace);
5907 INIT_LIST_HEAD(&tr->mod_notrace);
5908 }
5909 #else
5910
5911 static struct ftrace_ops global_ops = {
5912 .func = ftrace_stub,
5913 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
5914 FTRACE_OPS_FL_INITIALIZED |
5915 FTRACE_OPS_FL_PID,
5916 };
5917
5918 static int __init ftrace_nodyn_init(void)
5919 {
5920 ftrace_enabled = 1;
5921 return 0;
5922 }
5923 core_initcall(ftrace_nodyn_init);
5924
5925 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
5926 static inline void ftrace_startup_enable(int command) { }
5927 static inline void ftrace_startup_all(int command) { }
5928 /* Keep as macros so we do not need to define the commands */
5929 # define ftrace_startup(ops, command) \
5930 ({ \
5931 int ___ret = __register_ftrace_function(ops); \
5932 if (!___ret) \
5933 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
5934 ___ret; \
5935 })
5936 # define ftrace_shutdown(ops, command) \
5937 ({ \
5938 int ___ret = __unregister_ftrace_function(ops); \
5939 if (!___ret) \
5940 (ops)->flags &= ~FTRACE_OPS_FL_ENABLED; \
5941 ___ret; \
5942 })
5943
5944 # define ftrace_startup_sysctl() do { } while (0)
5945 # define ftrace_shutdown_sysctl() do { } while (0)
5946
5947 static inline int
5948 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
5949 {
5950 return 1;
5951 }
5952
5953 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5954 {
5955 }
5956
5957 #endif /* CONFIG_DYNAMIC_FTRACE */
5958
5959 __init void ftrace_init_global_array_ops(struct trace_array *tr)
5960 {
5961 tr->ops = &global_ops;
5962 tr->ops->private = tr;
5963 ftrace_init_trace_array(tr);
5964 }
5965
5966 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
5967 {
5968 /* If we filter on pids, update to use the pid function */
5969 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
5970 if (WARN_ON(tr->ops->func != ftrace_stub))
5971 printk("ftrace ops had %pS for function\n",
5972 tr->ops->func);
5973 }
5974 tr->ops->func = func;
5975 tr->ops->private = tr;
5976 }
5977
5978 void ftrace_reset_array_ops(struct trace_array *tr)
5979 {
5980 tr->ops->func = ftrace_stub;
5981 }
5982
5983 static inline void
5984 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5985 struct ftrace_ops *ignored, struct pt_regs *regs)
5986 {
5987 struct ftrace_ops *op;
5988 int bit;
5989
5990 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5991 if (bit < 0)
5992 return;
5993
5994 /*
5995 * Some of the ops may be dynamically allocated,
5996 * they must be freed after a synchronize_sched().
5997 */
5998 preempt_disable_notrace();
5999
6000 do_for_each_ftrace_op(op, ftrace_ops_list) {
6001 /*
6002 * Check the following for each ops before calling their func:
6003 * if RCU flag is set, then rcu_is_watching() must be true
6004 * if PER_CPU is set, then ftrace_function_local_disable()
6005 * must be false
6006 * Otherwise test if the ip matches the ops filter
6007 *
6008 * If any of the above fails then the op->func() is not executed.
6009 */
6010 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
6011 (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
6012 !ftrace_function_local_disabled(op)) &&
6013 ftrace_ops_test(op, ip, regs)) {
6014
6015 if (FTRACE_WARN_ON(!op->func)) {
6016 pr_warn("op=%p %pS\n", op, op);
6017 goto out;
6018 }
6019 op->func(ip, parent_ip, op, regs);
6020 }
6021 } while_for_each_ftrace_op(op);
6022 out:
6023 preempt_enable_notrace();
6024 trace_clear_recursion(bit);
6025 }
6026
6027 /*
6028 * Some archs only support passing ip and parent_ip. Even though
6029 * the list function ignores the op parameter, we do not want any
6030 * C side effects, where a function is called without the caller
6031 * sending a third parameter.
6032 * Archs are to support both the regs and ftrace_ops at the same time.
6033 * If they support ftrace_ops, it is assumed they support regs.
6034 * If call backs want to use regs, they must either check for regs
6035 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
6036 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
6037 * An architecture can pass partial regs with ftrace_ops and still
6038 * set the ARCH_SUPPORTS_FTRACE_OPS.
6039 */
6040 #if ARCH_SUPPORTS_FTRACE_OPS
6041 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6042 struct ftrace_ops *op, struct pt_regs *regs)
6043 {
6044 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
6045 }
6046 #else
6047 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
6048 {
6049 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
6050 }
6051 #endif
6052
6053 /*
6054 * If there's only one function registered but it does not support
6055 * recursion, needs RCU protection and/or requires per cpu handling, then
6056 * this function will be called by the mcount trampoline.
6057 */
6058 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
6059 struct ftrace_ops *op, struct pt_regs *regs)
6060 {
6061 int bit;
6062
6063 if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
6064 return;
6065
6066 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6067 if (bit < 0)
6068 return;
6069
6070 preempt_disable_notrace();
6071
6072 if (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
6073 !ftrace_function_local_disabled(op)) {
6074 op->func(ip, parent_ip, op, regs);
6075 }
6076
6077 preempt_enable_notrace();
6078 trace_clear_recursion(bit);
6079 }
6080
6081 /**
6082 * ftrace_ops_get_func - get the function a trampoline should call
6083 * @ops: the ops to get the function for
6084 *
6085 * Normally the mcount trampoline will call the ops->func, but there
6086 * are times that it should not. For example, if the ops does not
6087 * have its own recursion protection, then it should call the
6088 * ftrace_ops_assist_func() instead.
6089 *
6090 * Returns the function that the trampoline should call for @ops.
6091 */
6092 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
6093 {
6094 /*
6095 * If the function does not handle recursion, needs to be RCU safe,
6096 * or does per cpu logic, then we need to call the assist handler.
6097 */
6098 if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
6099 ops->flags & (FTRACE_OPS_FL_RCU | FTRACE_OPS_FL_PER_CPU))
6100 return ftrace_ops_assist_func;
6101
6102 return ops->func;
6103 }
6104
6105 static void
6106 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
6107 struct task_struct *prev, struct task_struct *next)
6108 {
6109 struct trace_array *tr = data;
6110 struct trace_pid_list *pid_list;
6111
6112 pid_list = rcu_dereference_sched(tr->function_pids);
6113
6114 this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6115 trace_ignore_this_task(pid_list, next));
6116 }
6117
6118 static void
6119 ftrace_pid_follow_sched_process_fork(void *data,
6120 struct task_struct *self,
6121 struct task_struct *task)
6122 {
6123 struct trace_pid_list *pid_list;
6124 struct trace_array *tr = data;
6125
6126 pid_list = rcu_dereference_sched(tr->function_pids);
6127 trace_filter_add_remove_task(pid_list, self, task);
6128 }
6129
6130 static void
6131 ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
6132 {
6133 struct trace_pid_list *pid_list;
6134 struct trace_array *tr = data;
6135
6136 pid_list = rcu_dereference_sched(tr->function_pids);
6137 trace_filter_add_remove_task(pid_list, NULL, task);
6138 }
6139
6140 void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
6141 {
6142 if (enable) {
6143 register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6144 tr);
6145 register_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6146 tr);
6147 } else {
6148 unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6149 tr);
6150 unregister_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6151 tr);
6152 }
6153 }
6154
6155 static void clear_ftrace_pids(struct trace_array *tr)
6156 {
6157 struct trace_pid_list *pid_list;
6158 int cpu;
6159
6160 pid_list = rcu_dereference_protected(tr->function_pids,
6161 lockdep_is_held(&ftrace_lock));
6162 if (!pid_list)
6163 return;
6164
6165 unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6166
6167 for_each_possible_cpu(cpu)
6168 per_cpu_ptr(tr->trace_buffer.data, cpu)->ftrace_ignore_pid = false;
6169
6170 rcu_assign_pointer(tr->function_pids, NULL);
6171
6172 /* Wait till all users are no longer using pid filtering */
6173 synchronize_sched();
6174
6175 trace_free_pid_list(pid_list);
6176 }
6177
6178 void ftrace_clear_pids(struct trace_array *tr)
6179 {
6180 mutex_lock(&ftrace_lock);
6181
6182 clear_ftrace_pids(tr);
6183
6184 mutex_unlock(&ftrace_lock);
6185 }
6186
6187 static void ftrace_pid_reset(struct trace_array *tr)
6188 {
6189 mutex_lock(&ftrace_lock);
6190 clear_ftrace_pids(tr);
6191
6192 ftrace_update_pid_func();
6193 ftrace_startup_all(0);
6194
6195 mutex_unlock(&ftrace_lock);
6196 }
6197
6198 /* Greater than any max PID */
6199 #define FTRACE_NO_PIDS (void *)(PID_MAX_LIMIT + 1)
6200
6201 static void *fpid_start(struct seq_file *m, loff_t *pos)
6202 __acquires(RCU)
6203 {
6204 struct trace_pid_list *pid_list;
6205 struct trace_array *tr = m->private;
6206
6207 mutex_lock(&ftrace_lock);
6208 rcu_read_lock_sched();
6209
6210 pid_list = rcu_dereference_sched(tr->function_pids);
6211
6212 if (!pid_list)
6213 return !(*pos) ? FTRACE_NO_PIDS : NULL;
6214
6215 return trace_pid_start(pid_list, pos);
6216 }
6217
6218 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
6219 {
6220 struct trace_array *tr = m->private;
6221 struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
6222
6223 if (v == FTRACE_NO_PIDS)
6224 return NULL;
6225
6226 return trace_pid_next(pid_list, v, pos);
6227 }
6228
6229 static void fpid_stop(struct seq_file *m, void *p)
6230 __releases(RCU)
6231 {
6232 rcu_read_unlock_sched();
6233 mutex_unlock(&ftrace_lock);
6234 }
6235
6236 static int fpid_show(struct seq_file *m, void *v)
6237 {
6238 if (v == FTRACE_NO_PIDS) {
6239 seq_puts(m, "no pid\n");
6240 return 0;
6241 }
6242
6243 return trace_pid_show(m, v);
6244 }
6245
6246 static const struct seq_operations ftrace_pid_sops = {
6247 .start = fpid_start,
6248 .next = fpid_next,
6249 .stop = fpid_stop,
6250 .show = fpid_show,
6251 };
6252
6253 static int
6254 ftrace_pid_open(struct inode *inode, struct file *file)
6255 {
6256 struct trace_array *tr = inode->i_private;
6257 struct seq_file *m;
6258 int ret = 0;
6259
6260 if (trace_array_get(tr) < 0)
6261 return -ENODEV;
6262
6263 if ((file->f_mode & FMODE_WRITE) &&
6264 (file->f_flags & O_TRUNC))
6265 ftrace_pid_reset(tr);
6266
6267 ret = seq_open(file, &ftrace_pid_sops);
6268 if (ret < 0) {
6269 trace_array_put(tr);
6270 } else {
6271 m = file->private_data;
6272 /* copy tr over to seq ops */
6273 m->private = tr;
6274 }
6275
6276 return ret;
6277 }
6278
6279 static void ignore_task_cpu(void *data)
6280 {
6281 struct trace_array *tr = data;
6282 struct trace_pid_list *pid_list;
6283
6284 /*
6285 * This function is called by on_each_cpu() while the
6286 * event_mutex is held.
6287 */
6288 pid_list = rcu_dereference_protected(tr->function_pids,
6289 mutex_is_locked(&ftrace_lock));
6290
6291 this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6292 trace_ignore_this_task(pid_list, current));
6293 }
6294
6295 static ssize_t
6296 ftrace_pid_write(struct file *filp, const char __user *ubuf,
6297 size_t cnt, loff_t *ppos)
6298 {
6299 struct seq_file *m = filp->private_data;
6300 struct trace_array *tr = m->private;
6301 struct trace_pid_list *filtered_pids = NULL;
6302 struct trace_pid_list *pid_list;
6303 ssize_t ret;
6304
6305 if (!cnt)
6306 return 0;
6307
6308 mutex_lock(&ftrace_lock);
6309
6310 filtered_pids = rcu_dereference_protected(tr->function_pids,
6311 lockdep_is_held(&ftrace_lock));
6312
6313 ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
6314 if (ret < 0)
6315 goto out;
6316
6317 rcu_assign_pointer(tr->function_pids, pid_list);
6318
6319 if (filtered_pids) {
6320 synchronize_sched();
6321 trace_free_pid_list(filtered_pids);
6322 } else if (pid_list) {
6323 /* Register a probe to set whether to ignore the tracing of a task */
6324 register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6325 }
6326
6327 /*
6328 * Ignoring of pids is done at task switch. But we have to
6329 * check for those tasks that are currently running.
6330 * Always do this in case a pid was appended or removed.
6331 */
6332 on_each_cpu(ignore_task_cpu, tr, 1);
6333
6334 ftrace_update_pid_func();
6335 ftrace_startup_all(0);
6336 out:
6337 mutex_unlock(&ftrace_lock);
6338
6339 if (ret > 0)
6340 *ppos += ret;
6341
6342 return ret;
6343 }
6344
6345 static int
6346 ftrace_pid_release(struct inode *inode, struct file *file)
6347 {
6348 struct trace_array *tr = inode->i_private;
6349
6350 trace_array_put(tr);
6351
6352 return seq_release(inode, file);
6353 }
6354
6355 static const struct file_operations ftrace_pid_fops = {
6356 .open = ftrace_pid_open,
6357 .write = ftrace_pid_write,
6358 .read = seq_read,
6359 .llseek = tracing_lseek,
6360 .release = ftrace_pid_release,
6361 };
6362
6363 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
6364 {
6365 trace_create_file("set_ftrace_pid", 0644, d_tracer,
6366 tr, &ftrace_pid_fops);
6367 }
6368
6369 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
6370 struct dentry *d_tracer)
6371 {
6372 /* Only the top level directory has the dyn_tracefs and profile */
6373 WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
6374
6375 ftrace_init_dyn_tracefs(d_tracer);
6376 ftrace_profile_tracefs(d_tracer);
6377 }
6378
6379 /**
6380 * ftrace_kill - kill ftrace
6381 *
6382 * This function should be used by panic code. It stops ftrace
6383 * but in a not so nice way. If you need to simply kill ftrace
6384 * from a non-atomic section, use ftrace_kill.
6385 */
6386 void ftrace_kill(void)
6387 {
6388 ftrace_disabled = 1;
6389 ftrace_enabled = 0;
6390 clear_ftrace_function();
6391 }
6392
6393 /**
6394 * Test if ftrace is dead or not.
6395 */
6396 int ftrace_is_dead(void)
6397 {
6398 return ftrace_disabled;
6399 }
6400
6401 /**
6402 * register_ftrace_function - register a function for profiling
6403 * @ops - ops structure that holds the function for profiling.
6404 *
6405 * Register a function to be called by all functions in the
6406 * kernel.
6407 *
6408 * Note: @ops->func and all the functions it calls must be labeled
6409 * with "notrace", otherwise it will go into a
6410 * recursive loop.
6411 */
6412 int register_ftrace_function(struct ftrace_ops *ops)
6413 {
6414 int ret = -1;
6415
6416 ftrace_ops_init(ops);
6417
6418 mutex_lock(&ftrace_lock);
6419
6420 ret = ftrace_startup(ops, 0);
6421
6422 mutex_unlock(&ftrace_lock);
6423
6424 return ret;
6425 }
6426 EXPORT_SYMBOL_GPL(register_ftrace_function);
6427
6428 /**
6429 * unregister_ftrace_function - unregister a function for profiling.
6430 * @ops - ops structure that holds the function to unregister
6431 *
6432 * Unregister a function that was added to be called by ftrace profiling.
6433 */
6434 int unregister_ftrace_function(struct ftrace_ops *ops)
6435 {
6436 int ret;
6437
6438 mutex_lock(&ftrace_lock);
6439 ret = ftrace_shutdown(ops, 0);
6440 mutex_unlock(&ftrace_lock);
6441
6442 return ret;
6443 }
6444 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
6445
6446 int
6447 ftrace_enable_sysctl(struct ctl_table *table, int write,
6448 void __user *buffer, size_t *lenp,
6449 loff_t *ppos)
6450 {
6451 int ret = -ENODEV;
6452
6453 mutex_lock(&ftrace_lock);
6454
6455 if (unlikely(ftrace_disabled))
6456 goto out;
6457
6458 ret = proc_dointvec(table, write, buffer, lenp, ppos);
6459
6460 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
6461 goto out;
6462
6463 last_ftrace_enabled = !!ftrace_enabled;
6464
6465 if (ftrace_enabled) {
6466
6467 /* we are starting ftrace again */
6468 if (rcu_dereference_protected(ftrace_ops_list,
6469 lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
6470 update_ftrace_function();
6471
6472 ftrace_startup_sysctl();
6473
6474 } else {
6475 /* stopping ftrace calls (just send to ftrace_stub) */
6476 ftrace_trace_function = ftrace_stub;
6477
6478 ftrace_shutdown_sysctl();
6479 }
6480
6481 out:
6482 mutex_unlock(&ftrace_lock);
6483 return ret;
6484 }
6485
6486 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
6487
6488 static struct ftrace_ops graph_ops = {
6489 .func = ftrace_stub,
6490 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
6491 FTRACE_OPS_FL_INITIALIZED |
6492 FTRACE_OPS_FL_PID |
6493 FTRACE_OPS_FL_STUB,
6494 #ifdef FTRACE_GRAPH_TRAMP_ADDR
6495 .trampoline = FTRACE_GRAPH_TRAMP_ADDR,
6496 /* trampoline_size is only needed for dynamically allocated tramps */
6497 #endif
6498 ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
6499 };
6500
6501 void ftrace_graph_sleep_time_control(bool enable)
6502 {
6503 fgraph_sleep_time = enable;
6504 }
6505
6506 void ftrace_graph_graph_time_control(bool enable)
6507 {
6508 fgraph_graph_time = enable;
6509 }
6510
6511 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
6512 {
6513 return 0;
6514 }
6515
6516 /* The callbacks that hook a function */
6517 trace_func_graph_ret_t ftrace_graph_return =
6518 (trace_func_graph_ret_t)ftrace_stub;
6519 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
6520 static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
6521
6522 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
6523 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
6524 {
6525 int i;
6526 int ret = 0;
6527 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
6528 struct task_struct *g, *t;
6529
6530 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
6531 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
6532 * sizeof(struct ftrace_ret_stack),
6533 GFP_KERNEL);
6534 if (!ret_stack_list[i]) {
6535 start = 0;
6536 end = i;
6537 ret = -ENOMEM;
6538 goto free;
6539 }
6540 }
6541
6542 read_lock(&tasklist_lock);
6543 do_each_thread(g, t) {
6544 if (start == end) {
6545 ret = -EAGAIN;
6546 goto unlock;
6547 }
6548
6549 if (t->ret_stack == NULL) {
6550 atomic_set(&t->tracing_graph_pause, 0);
6551 atomic_set(&t->trace_overrun, 0);
6552 t->curr_ret_stack = -1;
6553 /* Make sure the tasks see the -1 first: */
6554 smp_wmb();
6555 t->ret_stack = ret_stack_list[start++];
6556 }
6557 } while_each_thread(g, t);
6558
6559 unlock:
6560 read_unlock(&tasklist_lock);
6561 free:
6562 for (i = start; i < end; i++)
6563 kfree(ret_stack_list[i]);
6564 return ret;
6565 }
6566
6567 static void
6568 ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
6569 struct task_struct *prev, struct task_struct *next)
6570 {
6571 unsigned long long timestamp;
6572 int index;
6573
6574 /*
6575 * Does the user want to count the time a function was asleep.
6576 * If so, do not update the time stamps.
6577 */
6578 if (fgraph_sleep_time)
6579 return;
6580
6581 timestamp = trace_clock_local();
6582
6583 prev->ftrace_timestamp = timestamp;
6584
6585 /* only process tasks that we timestamped */
6586 if (!next->ftrace_timestamp)
6587 return;
6588
6589 /*
6590 * Update all the counters in next to make up for the
6591 * time next was sleeping.
6592 */
6593 timestamp -= next->ftrace_timestamp;
6594
6595 for (index = next->curr_ret_stack; index >= 0; index--)
6596 next->ret_stack[index].calltime += timestamp;
6597 }
6598
6599 /* Allocate a return stack for each task */
6600 static int start_graph_tracing(void)
6601 {
6602 struct ftrace_ret_stack **ret_stack_list;
6603 int ret, cpu;
6604
6605 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
6606 sizeof(struct ftrace_ret_stack *),
6607 GFP_KERNEL);
6608
6609 if (!ret_stack_list)
6610 return -ENOMEM;
6611
6612 /* The cpu_boot init_task->ret_stack will never be freed */
6613 for_each_online_cpu(cpu) {
6614 if (!idle_task(cpu)->ret_stack)
6615 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
6616 }
6617
6618 do {
6619 ret = alloc_retstack_tasklist(ret_stack_list);
6620 } while (ret == -EAGAIN);
6621
6622 if (!ret) {
6623 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
6624 if (ret)
6625 pr_info("ftrace_graph: Couldn't activate tracepoint"
6626 " probe to kernel_sched_switch\n");
6627 }
6628
6629 kfree(ret_stack_list);
6630 return ret;
6631 }
6632
6633 /*
6634 * Hibernation protection.
6635 * The state of the current task is too much unstable during
6636 * suspend/restore to disk. We want to protect against that.
6637 */
6638 static int
6639 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
6640 void *unused)
6641 {
6642 switch (state) {
6643 case PM_HIBERNATION_PREPARE:
6644 pause_graph_tracing();
6645 break;
6646
6647 case PM_POST_HIBERNATION:
6648 unpause_graph_tracing();
6649 break;
6650 }
6651 return NOTIFY_DONE;
6652 }
6653
6654 static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
6655 {
6656 if (!ftrace_ops_test(&global_ops, trace->func, NULL))
6657 return 0;
6658 return __ftrace_graph_entry(trace);
6659 }
6660
6661 /*
6662 * The function graph tracer should only trace the functions defined
6663 * by set_ftrace_filter and set_ftrace_notrace. If another function
6664 * tracer ops is registered, the graph tracer requires testing the
6665 * function against the global ops, and not just trace any function
6666 * that any ftrace_ops registered.
6667 */
6668 static void update_function_graph_func(void)
6669 {
6670 struct ftrace_ops *op;
6671 bool do_test = false;
6672
6673 /*
6674 * The graph and global ops share the same set of functions
6675 * to test. If any other ops is on the list, then
6676 * the graph tracing needs to test if its the function
6677 * it should call.
6678 */
6679 do_for_each_ftrace_op(op, ftrace_ops_list) {
6680 if (op != &global_ops && op != &graph_ops &&
6681 op != &ftrace_list_end) {
6682 do_test = true;
6683 /* in double loop, break out with goto */
6684 goto out;
6685 }
6686 } while_for_each_ftrace_op(op);
6687 out:
6688 if (do_test)
6689 ftrace_graph_entry = ftrace_graph_entry_test;
6690 else
6691 ftrace_graph_entry = __ftrace_graph_entry;
6692 }
6693
6694 static struct notifier_block ftrace_suspend_notifier = {
6695 .notifier_call = ftrace_suspend_notifier_call,
6696 };
6697
6698 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
6699 trace_func_graph_ent_t entryfunc)
6700 {
6701 int ret = 0;
6702
6703 mutex_lock(&ftrace_lock);
6704
6705 /* we currently allow only one tracer registered at a time */
6706 if (ftrace_graph_active) {
6707 ret = -EBUSY;
6708 goto out;
6709 }
6710
6711 register_pm_notifier(&ftrace_suspend_notifier);
6712
6713 ftrace_graph_active++;
6714 ret = start_graph_tracing();
6715 if (ret) {
6716 ftrace_graph_active--;
6717 goto out;
6718 }
6719
6720 ftrace_graph_return = retfunc;
6721
6722 /*
6723 * Update the indirect function to the entryfunc, and the
6724 * function that gets called to the entry_test first. Then
6725 * call the update fgraph entry function to determine if
6726 * the entryfunc should be called directly or not.
6727 */
6728 __ftrace_graph_entry = entryfunc;
6729 ftrace_graph_entry = ftrace_graph_entry_test;
6730 update_function_graph_func();
6731
6732 ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
6733 out:
6734 mutex_unlock(&ftrace_lock);
6735 return ret;
6736 }
6737
6738 void unregister_ftrace_graph(void)
6739 {
6740 mutex_lock(&ftrace_lock);
6741
6742 if (unlikely(!ftrace_graph_active))
6743 goto out;
6744
6745 ftrace_graph_active--;
6746 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
6747 ftrace_graph_entry = ftrace_graph_entry_stub;
6748 __ftrace_graph_entry = ftrace_graph_entry_stub;
6749 ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
6750 unregister_pm_notifier(&ftrace_suspend_notifier);
6751 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
6752
6753 #ifdef CONFIG_DYNAMIC_FTRACE
6754 /*
6755 * Function graph does not allocate the trampoline, but
6756 * other global_ops do. We need to reset the ALLOC_TRAMP flag
6757 * if one was used.
6758 */
6759 global_ops.trampoline = save_global_trampoline;
6760 if (save_global_flags & FTRACE_OPS_FL_ALLOC_TRAMP)
6761 global_ops.flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
6762 #endif
6763
6764 out:
6765 mutex_unlock(&ftrace_lock);
6766 }
6767
6768 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
6769
6770 static void
6771 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
6772 {
6773 atomic_set(&t->tracing_graph_pause, 0);
6774 atomic_set(&t->trace_overrun, 0);
6775 t->ftrace_timestamp = 0;
6776 /* make curr_ret_stack visible before we add the ret_stack */
6777 smp_wmb();
6778 t->ret_stack = ret_stack;
6779 }
6780
6781 /*
6782 * Allocate a return stack for the idle task. May be the first
6783 * time through, or it may be done by CPU hotplug online.
6784 */
6785 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
6786 {
6787 t->curr_ret_stack = -1;
6788 /*
6789 * The idle task has no parent, it either has its own
6790 * stack or no stack at all.
6791 */
6792 if (t->ret_stack)
6793 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
6794
6795 if (ftrace_graph_active) {
6796 struct ftrace_ret_stack *ret_stack;
6797
6798 ret_stack = per_cpu(idle_ret_stack, cpu);
6799 if (!ret_stack) {
6800 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6801 * sizeof(struct ftrace_ret_stack),
6802 GFP_KERNEL);
6803 if (!ret_stack)
6804 return;
6805 per_cpu(idle_ret_stack, cpu) = ret_stack;
6806 }
6807 graph_init_task(t, ret_stack);
6808 }
6809 }
6810
6811 /* Allocate a return stack for newly created task */
6812 void ftrace_graph_init_task(struct task_struct *t)
6813 {
6814 /* Make sure we do not use the parent ret_stack */
6815 t->ret_stack = NULL;
6816 t->curr_ret_stack = -1;
6817
6818 if (ftrace_graph_active) {
6819 struct ftrace_ret_stack *ret_stack;
6820
6821 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6822 * sizeof(struct ftrace_ret_stack),
6823 GFP_KERNEL);
6824 if (!ret_stack)
6825 return;
6826 graph_init_task(t, ret_stack);
6827 }
6828 }
6829
6830 void ftrace_graph_exit_task(struct task_struct *t)
6831 {
6832 struct ftrace_ret_stack *ret_stack = t->ret_stack;
6833
6834 t->ret_stack = NULL;
6835 /* NULL must become visible to IRQs before we free it: */
6836 barrier();
6837
6838 kfree(ret_stack);
6839 }
6840 #endif