projects/linux·blob v3.13 · refs

5134 lines · 113.8 KiB

1/*
2 * Infrastructure for profiling code inserted by 'gcc -pg'.
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <[email protected]>
5 * Copyright (C) 2004-2008 Ingo Molnar <[email protected]>
6 *
7 * Originally ported from the -rt patch by:
8 * Copyright (C) 2007 Arnaldo Carvalho de Melo <[email protected]>
9 *
10 * Based on code in the latency_tracer, that is:
11 *
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15
16#include <linux/stop_machine.h>
17#include <linux/clocksource.h>
18#include <linux/kallsyms.h>
19#include <linux/seq_file.h>
20#include <linux/suspend.h>
21#include <linux/debugfs.h>
22#include <linux/hardirq.h>
23#include <linux/kthread.h>
24#include <linux/uaccess.h>
25#include <linux/bsearch.h>
26#include <linux/module.h>
27#include <linux/ftrace.h>
28#include <linux/sysctl.h>
29#include <linux/slab.h>
30#include <linux/ctype.h>
31#include <linux/sort.h>
32#include <linux/list.h>
33#include <linux/hash.h>
34#include <linux/rcupdate.h>
35
36#include <trace/events/sched.h>
37
38#include <asm/setup.h>
39
40#include "trace_output.h"
41#include "trace_stat.h"
42
43#define FTRACE_WARN_ON(cond) \
44 ({ \
45 int ___r = cond; \
46 if (WARN_ON(___r)) \
47 ftrace_kill(); \
48 ___r; \
49 })
50
51#define FTRACE_WARN_ON_ONCE(cond) \
52 ({ \
53 int ___r = cond; \
54 if (WARN_ON_ONCE(___r)) \
55 ftrace_kill(); \
56 ___r; \
57 })
58
59/* hash bits for specific function selection */
60#define FTRACE_HASH_BITS 7
61#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
62#define FTRACE_HASH_DEFAULT_BITS 10
63#define FTRACE_HASH_MAX_BITS 12
64
65#define FL_GLOBAL_CONTROL_MASK (FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_CONTROL)
66
67#ifdef CONFIG_DYNAMIC_FTRACE
68#define INIT_REGEX_LOCK(opsname) \
69 .regex_lock = __MUTEX_INITIALIZER(opsname.regex_lock),
70#else
71#define INIT_REGEX_LOCK(opsname)
72#endif
73
74static struct ftrace_ops ftrace_list_end __read_mostly = {
75 .func = ftrace_stub,
76 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
77};
78
79/* ftrace_enabled is a method to turn ftrace on or off */
80int ftrace_enabled __read_mostly;
81static int last_ftrace_enabled;
82
83/* Quick disabling of function tracer. */
84int function_trace_stop __read_mostly;
85
86/* Current function tracing op */
87struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
88
89/* List for set_ftrace_pid's pids. */
90LIST_HEAD(ftrace_pids);
91struct ftrace_pid {
92 struct list_head list;
93 struct pid *pid;
94};
95
96/*
97 * ftrace_disabled is set when an anomaly is discovered.
98 * ftrace_disabled is much stronger than ftrace_enabled.
99 */
100static int ftrace_disabled __read_mostly;
101
102static DEFINE_MUTEX(ftrace_lock);
103
104static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
105static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end;
106static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
107ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
108ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
109static struct ftrace_ops global_ops;
110static struct ftrace_ops control_ops;
111
112#if ARCH_SUPPORTS_FTRACE_OPS
113static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
114 struct ftrace_ops *op, struct pt_regs *regs);
115#else
116/* See comment below, where ftrace_ops_list_func is defined */
117static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
118#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
119#endif
120
121/*
122 * Traverse the ftrace_global_list, invoking all entries. The reason that we
123 * can use rcu_dereference_raw_notrace() is that elements removed from this list
124 * are simply leaked, so there is no need to interact with a grace-period
125 * mechanism. The rcu_dereference_raw_notrace() calls are needed to handle
126 * concurrent insertions into the ftrace_global_list.
127 *
128 * Silly Alpha and silly pointer-speculation compiler optimizations!
129 */
130#define do_for_each_ftrace_op(op, list) \
131 op = rcu_dereference_raw_notrace(list); \
132 do
133
134/*
135 * Optimized for just a single item in the list (as that is the normal case).
136 */
137#define while_for_each_ftrace_op(op) \
138 while (likely(op = rcu_dereference_raw_notrace((op)->next)) && \
139 unlikely((op) != &ftrace_list_end))
140
141static inline void ftrace_ops_init(struct ftrace_ops *ops)
142{
143#ifdef CONFIG_DYNAMIC_FTRACE
144 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
145 mutex_init(&ops->regex_lock);
146 ops->flags |= FTRACE_OPS_FL_INITIALIZED;
147 }
148#endif
149}
150
151/**
152 * ftrace_nr_registered_ops - return number of ops registered
153 *
154 * Returns the number of ftrace_ops registered and tracing functions
155 */
156int ftrace_nr_registered_ops(void)
157{
158 struct ftrace_ops *ops;
159 int cnt = 0;
160
161 mutex_lock(&ftrace_lock);
162
163 for (ops = ftrace_ops_list;
164 ops != &ftrace_list_end; ops = ops->next)
165 cnt++;
166
167 mutex_unlock(&ftrace_lock);
168
169 return cnt;
170}
171
172static void
173ftrace_global_list_func(unsigned long ip, unsigned long parent_ip,
174 struct ftrace_ops *op, struct pt_regs *regs)
175{
176 int bit;
177
178 bit = trace_test_and_set_recursion(TRACE_GLOBAL_START, TRACE_GLOBAL_MAX);
179 if (bit < 0)
180 return;
181
182 do_for_each_ftrace_op(op, ftrace_global_list) {
183 op->func(ip, parent_ip, op, regs);
184 } while_for_each_ftrace_op(op);
185
186 trace_clear_recursion(bit);
187}
188
189static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
190 struct ftrace_ops *op, struct pt_regs *regs)
191{
192 if (!test_tsk_trace_trace(current))
193 return;
194
195 ftrace_pid_function(ip, parent_ip, op, regs);
196}
197
198static void set_ftrace_pid_function(ftrace_func_t func)
199{
200 /* do not set ftrace_pid_function to itself! */
201 if (func != ftrace_pid_func)
202 ftrace_pid_function = func;
203}
204
205/**
206 * clear_ftrace_function - reset the ftrace function
207 *
208 * This NULLs the ftrace function and in essence stops
209 * tracing. There may be lag
210 */
211void clear_ftrace_function(void)
212{
213 ftrace_trace_function = ftrace_stub;
214 ftrace_pid_function = ftrace_stub;
215}
216
217static void control_ops_disable_all(struct ftrace_ops *ops)
218{
219 int cpu;
220
221 for_each_possible_cpu(cpu)
222 *per_cpu_ptr(ops->disabled, cpu) = 1;
223}
224
225static int control_ops_alloc(struct ftrace_ops *ops)
226{
227 int __percpu *disabled;
228
229 disabled = alloc_percpu(int);
230 if (!disabled)
231 return -ENOMEM;
232
233 ops->disabled = disabled;
234 control_ops_disable_all(ops);
235 return 0;
236}
237
238static void control_ops_free(struct ftrace_ops *ops)
239{
240 free_percpu(ops->disabled);
241}
242
243static void update_global_ops(void)
244{
245 ftrace_func_t func;
246
247 /*
248 * If there's only one function registered, then call that
249 * function directly. Otherwise, we need to iterate over the
250 * registered callers.
251 */
252 if (ftrace_global_list == &ftrace_list_end ||
253 ftrace_global_list->next == &ftrace_list_end) {
254 func = ftrace_global_list->func;
255 /*
256 * As we are calling the function directly.
257 * If it does not have recursion protection,
258 * the function_trace_op needs to be updated
259 * accordingly.
260 */
261 if (ftrace_global_list->flags & FTRACE_OPS_FL_RECURSION_SAFE)
262 global_ops.flags |= FTRACE_OPS_FL_RECURSION_SAFE;
263 else
264 global_ops.flags &= ~FTRACE_OPS_FL_RECURSION_SAFE;
265 } else {
266 func = ftrace_global_list_func;
267 /* The list has its own recursion protection. */
268 global_ops.flags |= FTRACE_OPS_FL_RECURSION_SAFE;
269 }
270
271
272 /* If we filter on pids, update to use the pid function */
273 if (!list_empty(&ftrace_pids)) {
274 set_ftrace_pid_function(func);
275 func = ftrace_pid_func;
276 }
277
278 global_ops.func = func;
279}
280
281static void update_ftrace_function(void)
282{
283 ftrace_func_t func;
284
285 update_global_ops();
286
287 /*
288 * If we are at the end of the list and this ops is
289 * recursion safe and not dynamic and the arch supports passing ops,
290 * then have the mcount trampoline call the function directly.
291 */
292 if (ftrace_ops_list == &ftrace_list_end ||
293 (ftrace_ops_list->next == &ftrace_list_end &&
294 !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC) &&
295 (ftrace_ops_list->flags & FTRACE_OPS_FL_RECURSION_SAFE) &&
296 !FTRACE_FORCE_LIST_FUNC)) {
297 /* Set the ftrace_ops that the arch callback uses */
298 if (ftrace_ops_list == &global_ops)
299 function_trace_op = ftrace_global_list;
300 else
301 function_trace_op = ftrace_ops_list;
302 func = ftrace_ops_list->func;
303 } else {
304 /* Just use the default ftrace_ops */
305 function_trace_op = &ftrace_list_end;
306 func = ftrace_ops_list_func;
307 }
308
309 ftrace_trace_function = func;
310}
311
312static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
313{
314 ops->next = *list;
315 /*
316 * We are entering ops into the list but another
317 * CPU might be walking that list. We need to make sure
318 * the ops->next pointer is valid before another CPU sees
319 * the ops pointer included into the list.
320 */
321 rcu_assign_pointer(*list, ops);
322}
323
324static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
325{
326 struct ftrace_ops **p;
327
328 /*
329 * If we are removing the last function, then simply point
330 * to the ftrace_stub.
331 */
332 if (*list == ops && ops->next == &ftrace_list_end) {
333 *list = &ftrace_list_end;
334 return 0;
335 }
336
337 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
338 if (*p == ops)
339 break;
340
341 if (*p != ops)
342 return -1;
343
344 *p = (*p)->next;
345 return 0;
346}
347
348static void add_ftrace_list_ops(struct ftrace_ops **list,
349 struct ftrace_ops *main_ops,
350 struct ftrace_ops *ops)
351{
352 int first = *list == &ftrace_list_end;
353 add_ftrace_ops(list, ops);
354 if (first)
355 add_ftrace_ops(&ftrace_ops_list, main_ops);
356}
357
358static int remove_ftrace_list_ops(struct ftrace_ops **list,
359 struct ftrace_ops *main_ops,
360 struct ftrace_ops *ops)
361{
362 int ret = remove_ftrace_ops(list, ops);
363 if (!ret && *list == &ftrace_list_end)
364 ret = remove_ftrace_ops(&ftrace_ops_list, main_ops);
365 return ret;
366}
367
368static int __register_ftrace_function(struct ftrace_ops *ops)
369{
370 if (FTRACE_WARN_ON(ops == &global_ops))
371 return -EINVAL;
372
373 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
374 return -EBUSY;
375
376 /* We don't support both control and global flags set. */
377 if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK)
378 return -EINVAL;
379
380#ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
381 /*
382 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
383 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
384 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
385 */
386 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
387 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
388 return -EINVAL;
389
390 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
391 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
392#endif
393
394 if (!core_kernel_data((unsigned long)ops))
395 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
396
397 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
398 add_ftrace_list_ops(&ftrace_global_list, &global_ops, ops);
399 ops->flags |= FTRACE_OPS_FL_ENABLED;
400 } else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
401 if (control_ops_alloc(ops))
402 return -ENOMEM;
403 add_ftrace_list_ops(&ftrace_control_list, &control_ops, ops);
404 } else
405 add_ftrace_ops(&ftrace_ops_list, ops);
406
407 if (ftrace_enabled)
408 update_ftrace_function();
409
410 return 0;
411}
412
413static void ftrace_sync(struct work_struct *work)
414{
415 /*
416 * This function is just a stub to implement a hard force
417 * of synchronize_sched(). This requires synchronizing
418 * tasks even in userspace and idle.
419 *
420 * Yes, function tracing is rude.
421 */
422}
423
424static int __unregister_ftrace_function(struct ftrace_ops *ops)
425{
426 int ret;
427
428 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
429 return -EBUSY;
430
431 if (FTRACE_WARN_ON(ops == &global_ops))
432 return -EINVAL;
433
434 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
435 ret = remove_ftrace_list_ops(&ftrace_global_list,
436 &global_ops, ops);
437 if (!ret)
438 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
439 } else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
440 ret = remove_ftrace_list_ops(&ftrace_control_list,
441 &control_ops, ops);
442 if (!ret) {
443 /*
444 * The ftrace_ops is now removed from the list,
445 * so there'll be no new users. We must ensure
446 * all current users are done before we free
447 * the control data.
448 * Note synchronize_sched() is not enough, as we
449 * use preempt_disable() to do RCU, but the function
450 * tracer can be called where RCU is not active
451 * (before user_exit()).
452 */
453 schedule_on_each_cpu(ftrace_sync);
454 control_ops_free(ops);
455 }
456 } else
457 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
458
459 if (ret < 0)
460 return ret;
461
462 if (ftrace_enabled)
463 update_ftrace_function();
464
465 /*
466 * Dynamic ops may be freed, we must make sure that all
467 * callers are done before leaving this function.
468 *
469 * Again, normal synchronize_sched() is not good enough.
470 * We need to do a hard force of sched synchronization.
471 */
472 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
473 schedule_on_each_cpu(ftrace_sync);
474
475
476 return 0;
477}
478
479static void ftrace_update_pid_func(void)
480{
481 /* Only do something if we are tracing something */
482 if (ftrace_trace_function == ftrace_stub)
483 return;
484
485 update_ftrace_function();
486}
487
488#ifdef CONFIG_FUNCTION_PROFILER
489struct ftrace_profile {
490 struct hlist_node node;
491 unsigned long ip;
492 unsigned long counter;
493#ifdef CONFIG_FUNCTION_GRAPH_TRACER
494 unsigned long long time;
495 unsigned long long time_squared;
496#endif
497};
498
499struct ftrace_profile_page {
500 struct ftrace_profile_page *next;
501 unsigned long index;
502 struct ftrace_profile records[];
503};
504
505struct ftrace_profile_stat {
506 atomic_t disabled;
507 struct hlist_head *hash;
508 struct ftrace_profile_page *pages;
509 struct ftrace_profile_page *start;
510 struct tracer_stat stat;
511};
512
513#define PROFILE_RECORDS_SIZE \
514 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
515
516#define PROFILES_PER_PAGE \
517 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
518
519static int ftrace_profile_enabled __read_mostly;
520
521/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
522static DEFINE_MUTEX(ftrace_profile_lock);
523
524static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
525
526#define FTRACE_PROFILE_HASH_BITS 10
527#define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
528
529static void *
530function_stat_next(void *v, int idx)
531{
532 struct ftrace_profile *rec = v;
533 struct ftrace_profile_page *pg;
534
535 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
536
537 again:
538 if (idx != 0)
539 rec++;
540
541 if ((void *)rec >= (void *)&pg->records[pg->index]) {
542 pg = pg->next;
543 if (!pg)
544 return NULL;
545 rec = &pg->records[0];
546 if (!rec->counter)
547 goto again;
548 }
549
550 return rec;
551}
552
553static void *function_stat_start(struct tracer_stat *trace)
554{
555 struct ftrace_profile_stat *stat =
556 container_of(trace, struct ftrace_profile_stat, stat);
557
558 if (!stat || !stat->start)
559 return NULL;
560
561 return function_stat_next(&stat->start->records[0], 0);
562}
563
564#ifdef CONFIG_FUNCTION_GRAPH_TRACER
565/* function graph compares on total time */
566static int function_stat_cmp(void *p1, void *p2)
567{
568 struct ftrace_profile *a = p1;
569 struct ftrace_profile *b = p2;
570
571 if (a->time < b->time)
572 return -1;
573 if (a->time > b->time)
574 return 1;
575 else
576 return 0;
577}
578#else
579/* not function graph compares against hits */
580static int function_stat_cmp(void *p1, void *p2)
581{
582 struct ftrace_profile *a = p1;
583 struct ftrace_profile *b = p2;
584
585 if (a->counter < b->counter)
586 return -1;
587 if (a->counter > b->counter)
588 return 1;
589 else
590 return 0;
591}
592#endif
593
594static int function_stat_headers(struct seq_file *m)
595{
596#ifdef CONFIG_FUNCTION_GRAPH_TRACER
597 seq_printf(m, " Function "
598 "Hit Time Avg s^2\n"
599 " -------- "
600 "--- ---- --- ---\n");
601#else
602 seq_printf(m, " Function Hit\n"
603 " -------- ---\n");
604#endif
605 return 0;
606}
607
608static int function_stat_show(struct seq_file *m, void *v)
609{
610 struct ftrace_profile *rec = v;
611 char str[KSYM_SYMBOL_LEN];
612 int ret = 0;
613#ifdef CONFIG_FUNCTION_GRAPH_TRACER
614 static struct trace_seq s;
615 unsigned long long avg;
616 unsigned long long stddev;
617#endif
618 mutex_lock(&ftrace_profile_lock);
619
620 /* we raced with function_profile_reset() */
621 if (unlikely(rec->counter == 0)) {
622 ret = -EBUSY;
623 goto out;
624 }
625
626 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
627 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
628
629#ifdef CONFIG_FUNCTION_GRAPH_TRACER
630 seq_printf(m, " ");
631 avg = rec->time;
632 do_div(avg, rec->counter);
633
634 /* Sample standard deviation (s^2) */
635 if (rec->counter <= 1)
636 stddev = 0;
637 else {
638 /*
639 * Apply Welford's method:
640 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
641 */
642 stddev = rec->counter * rec->time_squared -
643 rec->time * rec->time;
644
645 /*
646 * Divide only 1000 for ns^2 -> us^2 conversion.
647 * trace_print_graph_duration will divide 1000 again.
648 */
649 do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
650 }
651
652 trace_seq_init(&s);
653 trace_print_graph_duration(rec->time, &s);
654 trace_seq_puts(&s, " ");
655 trace_print_graph_duration(avg, &s);
656 trace_seq_puts(&s, " ");
657 trace_print_graph_duration(stddev, &s);
658 trace_print_seq(m, &s);
659#endif
660 seq_putc(m, '\n');
661out:
662 mutex_unlock(&ftrace_profile_lock);
663
664 return ret;
665}
666
667static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
668{
669 struct ftrace_profile_page *pg;
670
671 pg = stat->pages = stat->start;
672
673 while (pg) {
674 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
675 pg->index = 0;
676 pg = pg->next;
677 }
678
679 memset(stat->hash, 0,
680 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
681}
682
683int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
684{
685 struct ftrace_profile_page *pg;
686 int functions;
687 int pages;
688 int i;
689
690 /* If we already allocated, do nothing */
691 if (stat->pages)
692 return 0;
693
694 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
695 if (!stat->pages)
696 return -ENOMEM;
697
698#ifdef CONFIG_DYNAMIC_FTRACE
699 functions = ftrace_update_tot_cnt;
700#else
701 /*
702 * We do not know the number of functions that exist because
703 * dynamic tracing is what counts them. With past experience
704 * we have around 20K functions. That should be more than enough.
705 * It is highly unlikely we will execute every function in
706 * the kernel.
707 */
708 functions = 20000;
709#endif
710
711 pg = stat->start = stat->pages;
712
713 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
714
715 for (i = 1; i < pages; i++) {
716 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
717 if (!pg->next)
718 goto out_free;
719 pg = pg->next;
720 }
721
722 return 0;
723
724 out_free:
725 pg = stat->start;
726 while (pg) {
727 unsigned long tmp = (unsigned long)pg;
728
729 pg = pg->next;
730 free_page(tmp);
731 }
732
733 stat->pages = NULL;
734 stat->start = NULL;
735
736 return -ENOMEM;
737}
738
739static int ftrace_profile_init_cpu(int cpu)
740{
741 struct ftrace_profile_stat *stat;
742 int size;
743
744 stat = &per_cpu(ftrace_profile_stats, cpu);
745
746 if (stat->hash) {
747 /* If the profile is already created, simply reset it */
748 ftrace_profile_reset(stat);
749 return 0;
750 }
751
752 /*
753 * We are profiling all functions, but usually only a few thousand
754 * functions are hit. We'll make a hash of 1024 items.
755 */
756 size = FTRACE_PROFILE_HASH_SIZE;
757
758 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
759
760 if (!stat->hash)
761 return -ENOMEM;
762
763 /* Preallocate the function profiling pages */
764 if (ftrace_profile_pages_init(stat) < 0) {
765 kfree(stat->hash);
766 stat->hash = NULL;
767 return -ENOMEM;
768 }
769
770 return 0;
771}
772
773static int ftrace_profile_init(void)
774{
775 int cpu;
776 int ret = 0;
777
778 for_each_possible_cpu(cpu) {
779 ret = ftrace_profile_init_cpu(cpu);
780 if (ret)
781 break;
782 }
783
784 return ret;
785}
786
787/* interrupts must be disabled */
788static struct ftrace_profile *
789ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
790{
791 struct ftrace_profile *rec;
792 struct hlist_head *hhd;
793 unsigned long key;
794
795 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
796 hhd = &stat->hash[key];
797
798 if (hlist_empty(hhd))
799 return NULL;
800
801 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
802 if (rec->ip == ip)
803 return rec;
804 }
805
806 return NULL;
807}
808
809static void ftrace_add_profile(struct ftrace_profile_stat *stat,
810 struct ftrace_profile *rec)
811{
812 unsigned long key;
813
814 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
815 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
816}
817
818/*
819 * The memory is already allocated, this simply finds a new record to use.
820 */
821static struct ftrace_profile *
822ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
823{
824 struct ftrace_profile *rec = NULL;
825
826 /* prevent recursion (from NMIs) */
827 if (atomic_inc_return(&stat->disabled) != 1)
828 goto out;
829
830 /*
831 * Try to find the function again since an NMI
832 * could have added it
833 */
834 rec = ftrace_find_profiled_func(stat, ip);
835 if (rec)
836 goto out;
837
838 if (stat->pages->index == PROFILES_PER_PAGE) {
839 if (!stat->pages->next)
840 goto out;
841 stat->pages = stat->pages->next;
842 }
843
844 rec = &stat->pages->records[stat->pages->index++];
845 rec->ip = ip;
846 ftrace_add_profile(stat, rec);
847
848 out:
849 atomic_dec(&stat->disabled);
850
851 return rec;
852}
853
854static void
855function_profile_call(unsigned long ip, unsigned long parent_ip,
856 struct ftrace_ops *ops, struct pt_regs *regs)
857{
858 struct ftrace_profile_stat *stat;
859 struct ftrace_profile *rec;
860 unsigned long flags;
861
862 if (!ftrace_profile_enabled)
863 return;
864
865 local_irq_save(flags);
866
867 stat = &__get_cpu_var(ftrace_profile_stats);
868 if (!stat->hash || !ftrace_profile_enabled)
869 goto out;
870
871 rec = ftrace_find_profiled_func(stat, ip);
872 if (!rec) {
873 rec = ftrace_profile_alloc(stat, ip);
874 if (!rec)
875 goto out;
876 }
877
878 rec->counter++;
879 out:
880 local_irq_restore(flags);
881}
882
883#ifdef CONFIG_FUNCTION_GRAPH_TRACER
884static int profile_graph_entry(struct ftrace_graph_ent *trace)
885{
886 function_profile_call(trace->func, 0, NULL, NULL);
887 return 1;
888}
889
890static void profile_graph_return(struct ftrace_graph_ret *trace)
891{
892 struct ftrace_profile_stat *stat;
893 unsigned long long calltime;
894 struct ftrace_profile *rec;
895 unsigned long flags;
896
897 local_irq_save(flags);
898 stat = &__get_cpu_var(ftrace_profile_stats);
899 if (!stat->hash || !ftrace_profile_enabled)
900 goto out;
901
902 /* If the calltime was zero'd ignore it */
903 if (!trace->calltime)
904 goto out;
905
906 calltime = trace->rettime - trace->calltime;
907
908 if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
909 int index;
910
911 index = trace->depth;
912
913 /* Append this call time to the parent time to subtract */
914 if (index)
915 current->ret_stack[index - 1].subtime += calltime;
916
917 if (current->ret_stack[index].subtime < calltime)
918 calltime -= current->ret_stack[index].subtime;
919 else
920 calltime = 0;
921 }
922
923 rec = ftrace_find_profiled_func(stat, trace->func);
924 if (rec) {
925 rec->time += calltime;
926 rec->time_squared += calltime * calltime;
927 }
928
929 out:
930 local_irq_restore(flags);
931}
932
933static int register_ftrace_profiler(void)
934{
935 return register_ftrace_graph(&profile_graph_return,
936 &profile_graph_entry);
937}
938
939static void unregister_ftrace_profiler(void)
940{
941 unregister_ftrace_graph();
942}
943#else
944static struct ftrace_ops ftrace_profile_ops __read_mostly = {
945 .func = function_profile_call,
946 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
947 INIT_REGEX_LOCK(ftrace_profile_ops)
948};
949
950static int register_ftrace_profiler(void)
951{
952 return register_ftrace_function(&ftrace_profile_ops);
953}
954
955static void unregister_ftrace_profiler(void)
956{
957 unregister_ftrace_function(&ftrace_profile_ops);
958}
959#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
960
961static ssize_t
962ftrace_profile_write(struct file *filp, const char __user *ubuf,
963 size_t cnt, loff_t *ppos)
964{
965 unsigned long val;
966 int ret;
967
968 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
969 if (ret)
970 return ret;
971
972 val = !!val;
973
974 mutex_lock(&ftrace_profile_lock);
975 if (ftrace_profile_enabled ^ val) {
976 if (val) {
977 ret = ftrace_profile_init();
978 if (ret < 0) {
979 cnt = ret;
980 goto out;
981 }
982
983 ret = register_ftrace_profiler();
984 if (ret < 0) {
985 cnt = ret;
986 goto out;
987 }
988 ftrace_profile_enabled = 1;
989 } else {
990 ftrace_profile_enabled = 0;
991 /*
992 * unregister_ftrace_profiler calls stop_machine
993 * so this acts like an synchronize_sched.
994 */
995 unregister_ftrace_profiler();
996 }
997 }
998 out:
999 mutex_unlock(&ftrace_profile_lock);
1000
1001 *ppos += cnt;
1002
1003 return cnt;
1004}
1005
1006static ssize_t
1007ftrace_profile_read(struct file *filp, char __user *ubuf,
1008 size_t cnt, loff_t *ppos)
1009{
1010 char buf[64]; /* big enough to hold a number */
1011 int r;
1012
1013 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1014 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1015}
1016
1017static const struct file_operations ftrace_profile_fops = {
1018 .open = tracing_open_generic,
1019 .read = ftrace_profile_read,
1020 .write = ftrace_profile_write,
1021 .llseek = default_llseek,
1022};
1023
1024/* used to initialize the real stat files */
1025static struct tracer_stat function_stats __initdata = {
1026 .name = "functions",
1027 .stat_start = function_stat_start,
1028 .stat_next = function_stat_next,
1029 .stat_cmp = function_stat_cmp,
1030 .stat_headers = function_stat_headers,
1031 .stat_show = function_stat_show
1032};
1033
1034static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1035{
1036 struct ftrace_profile_stat *stat;
1037 struct dentry *entry;
1038 char *name;
1039 int ret;
1040 int cpu;
1041
1042 for_each_possible_cpu(cpu) {
1043 stat = &per_cpu(ftrace_profile_stats, cpu);
1044
1045 /* allocate enough for function name + cpu number */
1046 name = kmalloc(32, GFP_KERNEL);
1047 if (!name) {
1048 /*
1049 * The files created are permanent, if something happens
1050 * we still do not free memory.
1051 */
1052 WARN(1,
1053 "Could not allocate stat file for cpu %d\n",
1054 cpu);
1055 return;
1056 }
1057 stat->stat = function_stats;
1058 snprintf(name, 32, "function%d", cpu);
1059 stat->stat.name = name;
1060 ret = register_stat_tracer(&stat->stat);
1061 if (ret) {
1062 WARN(1,
1063 "Could not register function stat for cpu %d\n",
1064 cpu);
1065 kfree(name);
1066 return;
1067 }
1068 }
1069
1070 entry = debugfs_create_file("function_profile_enabled", 0644,
1071 d_tracer, NULL, &ftrace_profile_fops);
1072 if (!entry)
1073 pr_warning("Could not create debugfs "
1074 "'function_profile_enabled' entry\n");
1075}
1076
1077#else /* CONFIG_FUNCTION_PROFILER */
1078static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1079{
1080}
1081#endif /* CONFIG_FUNCTION_PROFILER */
1082
1083static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1084
1085loff_t
1086ftrace_filter_lseek(struct file *file, loff_t offset, int whence)
1087{
1088 loff_t ret;
1089
1090 if (file->f_mode & FMODE_READ)
1091 ret = seq_lseek(file, offset, whence);
1092 else
1093 file->f_pos = ret = 1;
1094
1095 return ret;
1096}
1097
1098#ifdef CONFIG_DYNAMIC_FTRACE
1099
1100#ifndef CONFIG_FTRACE_MCOUNT_RECORD
1101# error Dynamic ftrace depends on MCOUNT_RECORD
1102#endif
1103
1104static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1105
1106struct ftrace_func_probe {
1107 struct hlist_node node;
1108 struct ftrace_probe_ops *ops;
1109 unsigned long flags;
1110 unsigned long ip;
1111 void *data;
1112 struct list_head free_list;
1113};
1114
1115struct ftrace_func_entry {
1116 struct hlist_node hlist;
1117 unsigned long ip;
1118};
1119
1120struct ftrace_hash {
1121 unsigned long size_bits;
1122 struct hlist_head *buckets;
1123 unsigned long count;
1124 struct rcu_head rcu;
1125};
1126
1127/*
1128 * We make these constant because no one should touch them,
1129 * but they are used as the default "empty hash", to avoid allocating
1130 * it all the time. These are in a read only section such that if
1131 * anyone does try to modify it, it will cause an exception.
1132 */
1133static const struct hlist_head empty_buckets[1];
1134static const struct ftrace_hash empty_hash = {
1135 .buckets = (struct hlist_head *)empty_buckets,
1136};
1137#define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1138
1139static struct ftrace_ops global_ops = {
1140 .func = ftrace_stub,
1141 .notrace_hash = EMPTY_HASH,
1142 .filter_hash = EMPTY_HASH,
1143 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
1144 INIT_REGEX_LOCK(global_ops)
1145};
1146
1147struct ftrace_page {
1148 struct ftrace_page *next;
1149 struct dyn_ftrace *records;
1150 int index;
1151 int size;
1152};
1153
1154static struct ftrace_page *ftrace_new_pgs;
1155
1156#define ENTRY_SIZE sizeof(struct dyn_ftrace)
1157#define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1158
1159/* estimate from running different kernels */
1160#define NR_TO_INIT 10000
1161
1162static struct ftrace_page *ftrace_pages_start;
1163static struct ftrace_page *ftrace_pages;
1164
1165static bool ftrace_hash_empty(struct ftrace_hash *hash)
1166{
1167 return !hash || !hash->count;
1168}
1169
1170static struct ftrace_func_entry *
1171ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1172{
1173 unsigned long key;
1174 struct ftrace_func_entry *entry;
1175 struct hlist_head *hhd;
1176
1177 if (ftrace_hash_empty(hash))
1178 return NULL;
1179
1180 if (hash->size_bits > 0)
1181 key = hash_long(ip, hash->size_bits);
1182 else
1183 key = 0;
1184
1185 hhd = &hash->buckets[key];
1186
1187 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1188 if (entry->ip == ip)
1189 return entry;
1190 }
1191 return NULL;
1192}
1193
1194static void __add_hash_entry(struct ftrace_hash *hash,
1195 struct ftrace_func_entry *entry)
1196{
1197 struct hlist_head *hhd;
1198 unsigned long key;
1199
1200 if (hash->size_bits)
1201 key = hash_long(entry->ip, hash->size_bits);
1202 else
1203 key = 0;
1204
1205 hhd = &hash->buckets[key];
1206 hlist_add_head(&entry->hlist, hhd);
1207 hash->count++;
1208}
1209
1210static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1211{
1212 struct ftrace_func_entry *entry;
1213
1214 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1215 if (!entry)
1216 return -ENOMEM;
1217
1218 entry->ip = ip;
1219 __add_hash_entry(hash, entry);
1220
1221 return 0;
1222}
1223
1224static void
1225free_hash_entry(struct ftrace_hash *hash,
1226 struct ftrace_func_entry *entry)
1227{
1228 hlist_del(&entry->hlist);
1229 kfree(entry);
1230 hash->count--;
1231}
1232
1233static void
1234remove_hash_entry(struct ftrace_hash *hash,
1235 struct ftrace_func_entry *entry)
1236{
1237 hlist_del(&entry->hlist);
1238 hash->count--;
1239}
1240
1241static void ftrace_hash_clear(struct ftrace_hash *hash)
1242{
1243 struct hlist_head *hhd;
1244 struct hlist_node *tn;
1245 struct ftrace_func_entry *entry;
1246 int size = 1 << hash->size_bits;
1247 int i;
1248
1249 if (!hash->count)
1250 return;
1251
1252 for (i = 0; i < size; i++) {
1253 hhd = &hash->buckets[i];
1254 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1255 free_hash_entry(hash, entry);
1256 }
1257 FTRACE_WARN_ON(hash->count);
1258}
1259
1260static void free_ftrace_hash(struct ftrace_hash *hash)
1261{
1262 if (!hash || hash == EMPTY_HASH)
1263 return;
1264 ftrace_hash_clear(hash);
1265 kfree(hash->buckets);
1266 kfree(hash);
1267}
1268
1269static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1270{
1271 struct ftrace_hash *hash;
1272
1273 hash = container_of(rcu, struct ftrace_hash, rcu);
1274 free_ftrace_hash(hash);
1275}
1276
1277static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1278{
1279 if (!hash || hash == EMPTY_HASH)
1280 return;
1281 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1282}
1283
1284void ftrace_free_filter(struct ftrace_ops *ops)
1285{
1286 ftrace_ops_init(ops);
1287 free_ftrace_hash(ops->filter_hash);
1288 free_ftrace_hash(ops->notrace_hash);
1289}
1290
1291static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1292{
1293 struct ftrace_hash *hash;
1294 int size;
1295
1296 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1297 if (!hash)
1298 return NULL;
1299
1300 size = 1 << size_bits;
1301 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1302
1303 if (!hash->buckets) {
1304 kfree(hash);
1305 return NULL;
1306 }
1307
1308 hash->size_bits = size_bits;
1309
1310 return hash;
1311}
1312
1313static struct ftrace_hash *
1314alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1315{
1316 struct ftrace_func_entry *entry;
1317 struct ftrace_hash *new_hash;
1318 int size;
1319 int ret;
1320 int i;
1321
1322 new_hash = alloc_ftrace_hash(size_bits);
1323 if (!new_hash)
1324 return NULL;
1325
1326 /* Empty hash? */
1327 if (ftrace_hash_empty(hash))
1328 return new_hash;
1329
1330 size = 1 << hash->size_bits;
1331 for (i = 0; i < size; i++) {
1332 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1333 ret = add_hash_entry(new_hash, entry->ip);
1334 if (ret < 0)
1335 goto free_hash;
1336 }
1337 }
1338
1339 FTRACE_WARN_ON(new_hash->count != hash->count);
1340
1341 return new_hash;
1342
1343 free_hash:
1344 free_ftrace_hash(new_hash);
1345 return NULL;
1346}
1347
1348static void
1349ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1350static void
1351ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1352
1353static int
1354ftrace_hash_move(struct ftrace_ops *ops, int enable,
1355 struct ftrace_hash **dst, struct ftrace_hash *src)
1356{
1357 struct ftrace_func_entry *entry;
1358 struct hlist_node *tn;
1359 struct hlist_head *hhd;
1360 struct ftrace_hash *old_hash;
1361 struct ftrace_hash *new_hash;
1362 int size = src->count;
1363 int bits = 0;
1364 int ret;
1365 int i;
1366
1367 /*
1368 * Remove the current set, update the hash and add
1369 * them back.
1370 */
1371 ftrace_hash_rec_disable(ops, enable);
1372
1373 /*
1374 * If the new source is empty, just free dst and assign it
1375 * the empty_hash.
1376 */
1377 if (!src->count) {
1378 free_ftrace_hash_rcu(*dst);
1379 rcu_assign_pointer(*dst, EMPTY_HASH);
1380 /* still need to update the function records */
1381 ret = 0;
1382 goto out;
1383 }
1384
1385 /*
1386 * Make the hash size about 1/2 the # found
1387 */
1388 for (size /= 2; size; size >>= 1)
1389 bits++;
1390
1391 /* Don't allocate too much */
1392 if (bits > FTRACE_HASH_MAX_BITS)
1393 bits = FTRACE_HASH_MAX_BITS;
1394
1395 ret = -ENOMEM;
1396 new_hash = alloc_ftrace_hash(bits);
1397 if (!new_hash)
1398 goto out;
1399
1400 size = 1 << src->size_bits;
1401 for (i = 0; i < size; i++) {
1402 hhd = &src->buckets[i];
1403 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1404 remove_hash_entry(src, entry);
1405 __add_hash_entry(new_hash, entry);
1406 }
1407 }
1408
1409 old_hash = *dst;
1410 rcu_assign_pointer(*dst, new_hash);
1411 free_ftrace_hash_rcu(old_hash);
1412
1413 ret = 0;
1414 out:
1415 /*
1416 * Enable regardless of ret:
1417 * On success, we enable the new hash.
1418 * On failure, we re-enable the original hash.
1419 */
1420 ftrace_hash_rec_enable(ops, enable);
1421
1422 return ret;
1423}
1424
1425/*
1426 * Test the hashes for this ops to see if we want to call
1427 * the ops->func or not.
1428 *
1429 * It's a match if the ip is in the ops->filter_hash or
1430 * the filter_hash does not exist or is empty,
1431 * AND
1432 * the ip is not in the ops->notrace_hash.
1433 *
1434 * This needs to be called with preemption disabled as
1435 * the hashes are freed with call_rcu_sched().
1436 */
1437static int
1438ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1439{
1440 struct ftrace_hash *filter_hash;
1441 struct ftrace_hash *notrace_hash;
1442 int ret;
1443
1444#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1445 /*
1446 * There's a small race when adding ops that the ftrace handler
1447 * that wants regs, may be called without them. We can not
1448 * allow that handler to be called if regs is NULL.
1449 */
1450 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1451 return 0;
1452#endif
1453
1454 filter_hash = rcu_dereference_raw_notrace(ops->filter_hash);
1455 notrace_hash = rcu_dereference_raw_notrace(ops->notrace_hash);
1456
1457 if ((ftrace_hash_empty(filter_hash) ||
1458 ftrace_lookup_ip(filter_hash, ip)) &&
1459 (ftrace_hash_empty(notrace_hash) ||
1460 !ftrace_lookup_ip(notrace_hash, ip)))
1461 ret = 1;
1462 else
1463 ret = 0;
1464
1465 return ret;
1466}
1467
1468/*
1469 * This is a double for. Do not use 'break' to break out of the loop,
1470 * you must use a goto.
1471 */
1472#define do_for_each_ftrace_rec(pg, rec) \
1473 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1474 int _____i; \
1475 for (_____i = 0; _____i < pg->index; _____i++) { \
1476 rec = &pg->records[_____i];
1477
1478#define while_for_each_ftrace_rec() \
1479 } \
1480 }
1481
1482
1483static int ftrace_cmp_recs(const void *a, const void *b)
1484{
1485 const struct dyn_ftrace *key = a;
1486 const struct dyn_ftrace *rec = b;
1487
1488 if (key->flags < rec->ip)
1489 return -1;
1490 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1491 return 1;
1492 return 0;
1493}
1494
1495static unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1496{
1497 struct ftrace_page *pg;
1498 struct dyn_ftrace *rec;
1499 struct dyn_ftrace key;
1500
1501 key.ip = start;
1502 key.flags = end; /* overload flags, as it is unsigned long */
1503
1504 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1505 if (end < pg->records[0].ip ||
1506 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1507 continue;
1508 rec = bsearch(&key, pg->records, pg->index,
1509 sizeof(struct dyn_ftrace),
1510 ftrace_cmp_recs);
1511 if (rec)
1512 return rec->ip;
1513 }
1514
1515 return 0;
1516}
1517
1518/**
1519 * ftrace_location - return true if the ip giving is a traced location
1520 * @ip: the instruction pointer to check
1521 *
1522 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1523 * That is, the instruction that is either a NOP or call to
1524 * the function tracer. It checks the ftrace internal tables to
1525 * determine if the address belongs or not.
1526 */
1527unsigned long ftrace_location(unsigned long ip)
1528{
1529 return ftrace_location_range(ip, ip);
1530}
1531
1532/**
1533 * ftrace_text_reserved - return true if range contains an ftrace location
1534 * @start: start of range to search
1535 * @end: end of range to search (inclusive). @end points to the last byte to check.
1536 *
1537 * Returns 1 if @start and @end contains a ftrace location.
1538 * That is, the instruction that is either a NOP or call to
1539 * the function tracer. It checks the ftrace internal tables to
1540 * determine if the address belongs or not.
1541 */
1542int ftrace_text_reserved(void *start, void *end)
1543{
1544 unsigned long ret;
1545
1546 ret = ftrace_location_range((unsigned long)start,
1547 (unsigned long)end);
1548
1549 return (int)!!ret;
1550}
1551
1552static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1553 int filter_hash,
1554 bool inc)
1555{
1556 struct ftrace_hash *hash;
1557 struct ftrace_hash *other_hash;
1558 struct ftrace_page *pg;
1559 struct dyn_ftrace *rec;
1560 int count = 0;
1561 int all = 0;
1562
1563 /* Only update if the ops has been registered */
1564 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1565 return;
1566
1567 /*
1568 * In the filter_hash case:
1569 * If the count is zero, we update all records.
1570 * Otherwise we just update the items in the hash.
1571 *
1572 * In the notrace_hash case:
1573 * We enable the update in the hash.
1574 * As disabling notrace means enabling the tracing,
1575 * and enabling notrace means disabling, the inc variable
1576 * gets inversed.
1577 */
1578 if (filter_hash) {
1579 hash = ops->filter_hash;
1580 other_hash = ops->notrace_hash;
1581 if (ftrace_hash_empty(hash))
1582 all = 1;
1583 } else {
1584 inc = !inc;
1585 hash = ops->notrace_hash;
1586 other_hash = ops->filter_hash;
1587 /*
1588 * If the notrace hash has no items,
1589 * then there's nothing to do.
1590 */
1591 if (ftrace_hash_empty(hash))
1592 return;
1593 }
1594
1595 do_for_each_ftrace_rec(pg, rec) {
1596 int in_other_hash = 0;
1597 int in_hash = 0;
1598 int match = 0;
1599
1600 if (all) {
1601 /*
1602 * Only the filter_hash affects all records.
1603 * Update if the record is not in the notrace hash.
1604 */
1605 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1606 match = 1;
1607 } else {
1608 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1609 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1610
1611 /*
1612 *
1613 */
1614 if (filter_hash && in_hash && !in_other_hash)
1615 match = 1;
1616 else if (!filter_hash && in_hash &&
1617 (in_other_hash || ftrace_hash_empty(other_hash)))
1618 match = 1;
1619 }
1620 if (!match)
1621 continue;
1622
1623 if (inc) {
1624 rec->flags++;
1625 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1626 return;
1627 /*
1628 * If any ops wants regs saved for this function
1629 * then all ops will get saved regs.
1630 */
1631 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1632 rec->flags |= FTRACE_FL_REGS;
1633 } else {
1634 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1635 return;
1636 rec->flags--;
1637 }
1638 count++;
1639 /* Shortcut, if we handled all records, we are done. */
1640 if (!all && count == hash->count)
1641 return;
1642 } while_for_each_ftrace_rec();
1643}
1644
1645static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1646 int filter_hash)
1647{
1648 __ftrace_hash_rec_update(ops, filter_hash, 0);
1649}
1650
1651static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1652 int filter_hash)
1653{
1654 __ftrace_hash_rec_update(ops, filter_hash, 1);
1655}
1656
1657static void print_ip_ins(const char *fmt, unsigned char *p)
1658{
1659 int i;
1660
1661 printk(KERN_CONT "%s", fmt);
1662
1663 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1664 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1665}
1666
1667/**
1668 * ftrace_bug - report and shutdown function tracer
1669 * @failed: The failed type (EFAULT, EINVAL, EPERM)
1670 * @ip: The address that failed
1671 *
1672 * The arch code that enables or disables the function tracing
1673 * can call ftrace_bug() when it has detected a problem in
1674 * modifying the code. @failed should be one of either:
1675 * EFAULT - if the problem happens on reading the @ip address
1676 * EINVAL - if what is read at @ip is not what was expected
1677 * EPERM - if the problem happens on writting to the @ip address
1678 */
1679void ftrace_bug(int failed, unsigned long ip)
1680{
1681 switch (failed) {
1682 case -EFAULT:
1683 FTRACE_WARN_ON_ONCE(1);
1684 pr_info("ftrace faulted on modifying ");
1685 print_ip_sym(ip);
1686 break;
1687 case -EINVAL:
1688 FTRACE_WARN_ON_ONCE(1);
1689 pr_info("ftrace failed to modify ");
1690 print_ip_sym(ip);
1691 print_ip_ins(" actual: ", (unsigned char *)ip);
1692 printk(KERN_CONT "\n");
1693 break;
1694 case -EPERM:
1695 FTRACE_WARN_ON_ONCE(1);
1696 pr_info("ftrace faulted on writing ");
1697 print_ip_sym(ip);
1698 break;
1699 default:
1700 FTRACE_WARN_ON_ONCE(1);
1701 pr_info("ftrace faulted on unknown error ");
1702 print_ip_sym(ip);
1703 }
1704}
1705
1706static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1707{
1708 unsigned long flag = 0UL;
1709
1710 /*
1711 * If we are updating calls:
1712 *
1713 * If the record has a ref count, then we need to enable it
1714 * because someone is using it.
1715 *
1716 * Otherwise we make sure its disabled.
1717 *
1718 * If we are disabling calls, then disable all records that
1719 * are enabled.
1720 */
1721 if (enable && (rec->flags & ~FTRACE_FL_MASK))
1722 flag = FTRACE_FL_ENABLED;
1723
1724 /*
1725 * If enabling and the REGS flag does not match the REGS_EN, then
1726 * do not ignore this record. Set flags to fail the compare against
1727 * ENABLED.
1728 */
1729 if (flag &&
1730 (!(rec->flags & FTRACE_FL_REGS) != !(rec->flags & FTRACE_FL_REGS_EN)))
1731 flag |= FTRACE_FL_REGS;
1732
1733 /* If the state of this record hasn't changed, then do nothing */
1734 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1735 return FTRACE_UPDATE_IGNORE;
1736
1737 if (flag) {
1738 /* Save off if rec is being enabled (for return value) */
1739 flag ^= rec->flags & FTRACE_FL_ENABLED;
1740
1741 if (update) {
1742 rec->flags |= FTRACE_FL_ENABLED;
1743 if (flag & FTRACE_FL_REGS) {
1744 if (rec->flags & FTRACE_FL_REGS)
1745 rec->flags |= FTRACE_FL_REGS_EN;
1746 else
1747 rec->flags &= ~FTRACE_FL_REGS_EN;
1748 }
1749 }
1750
1751 /*
1752 * If this record is being updated from a nop, then
1753 * return UPDATE_MAKE_CALL.
1754 * Otherwise, if the EN flag is set, then return
1755 * UPDATE_MODIFY_CALL_REGS to tell the caller to convert
1756 * from the non-save regs, to a save regs function.
1757 * Otherwise,
1758 * return UPDATE_MODIFY_CALL to tell the caller to convert
1759 * from the save regs, to a non-save regs function.
1760 */
1761 if (flag & FTRACE_FL_ENABLED)
1762 return FTRACE_UPDATE_MAKE_CALL;
1763 else if (rec->flags & FTRACE_FL_REGS_EN)
1764 return FTRACE_UPDATE_MODIFY_CALL_REGS;
1765 else
1766 return FTRACE_UPDATE_MODIFY_CALL;
1767 }
1768
1769 if (update) {
1770 /* If there's no more users, clear all flags */
1771 if (!(rec->flags & ~FTRACE_FL_MASK))
1772 rec->flags = 0;
1773 else
1774 /* Just disable the record (keep REGS state) */
1775 rec->flags &= ~FTRACE_FL_ENABLED;
1776 }
1777
1778 return FTRACE_UPDATE_MAKE_NOP;
1779}
1780
1781/**
1782 * ftrace_update_record, set a record that now is tracing or not
1783 * @rec: the record to update
1784 * @enable: set to 1 if the record is tracing, zero to force disable
1785 *
1786 * The records that represent all functions that can be traced need
1787 * to be updated when tracing has been enabled.
1788 */
1789int ftrace_update_record(struct dyn_ftrace *rec, int enable)
1790{
1791 return ftrace_check_record(rec, enable, 1);
1792}
1793
1794/**
1795 * ftrace_test_record, check if the record has been enabled or not
1796 * @rec: the record to test
1797 * @enable: set to 1 to check if enabled, 0 if it is disabled
1798 *
1799 * The arch code may need to test if a record is already set to
1800 * tracing to determine how to modify the function code that it
1801 * represents.
1802 */
1803int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1804{
1805 return ftrace_check_record(rec, enable, 0);
1806}
1807
1808static int
1809__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1810{
1811 unsigned long ftrace_old_addr;
1812 unsigned long ftrace_addr;
1813 int ret;
1814
1815 ret = ftrace_update_record(rec, enable);
1816
1817 if (rec->flags & FTRACE_FL_REGS)
1818 ftrace_addr = (unsigned long)FTRACE_REGS_ADDR;
1819 else
1820 ftrace_addr = (unsigned long)FTRACE_ADDR;
1821
1822 switch (ret) {
1823 case FTRACE_UPDATE_IGNORE:
1824 return 0;
1825
1826 case FTRACE_UPDATE_MAKE_CALL:
1827 return ftrace_make_call(rec, ftrace_addr);
1828
1829 case FTRACE_UPDATE_MAKE_NOP:
1830 return ftrace_make_nop(NULL, rec, ftrace_addr);
1831
1832 case FTRACE_UPDATE_MODIFY_CALL_REGS:
1833 case FTRACE_UPDATE_MODIFY_CALL:
1834 if (rec->flags & FTRACE_FL_REGS)
1835 ftrace_old_addr = (unsigned long)FTRACE_ADDR;
1836 else
1837 ftrace_old_addr = (unsigned long)FTRACE_REGS_ADDR;
1838
1839 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
1840 }
1841
1842 return -1; /* unknow ftrace bug */
1843}
1844
1845void __weak ftrace_replace_code(int enable)
1846{
1847 struct dyn_ftrace *rec;
1848 struct ftrace_page *pg;
1849 int failed;
1850
1851 if (unlikely(ftrace_disabled))
1852 return;
1853
1854 do_for_each_ftrace_rec(pg, rec) {
1855 failed = __ftrace_replace_code(rec, enable);
1856 if (failed) {
1857 ftrace_bug(failed, rec->ip);
1858 /* Stop processing */
1859 return;
1860 }
1861 } while_for_each_ftrace_rec();
1862}
1863
1864struct ftrace_rec_iter {
1865 struct ftrace_page *pg;
1866 int index;
1867};
1868
1869/**
1870 * ftrace_rec_iter_start, start up iterating over traced functions
1871 *
1872 * Returns an iterator handle that is used to iterate over all
1873 * the records that represent address locations where functions
1874 * are traced.
1875 *
1876 * May return NULL if no records are available.
1877 */
1878struct ftrace_rec_iter *ftrace_rec_iter_start(void)
1879{
1880 /*
1881 * We only use a single iterator.
1882 * Protected by the ftrace_lock mutex.
1883 */
1884 static struct ftrace_rec_iter ftrace_rec_iter;
1885 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
1886
1887 iter->pg = ftrace_pages_start;
1888 iter->index = 0;
1889
1890 /* Could have empty pages */
1891 while (iter->pg && !iter->pg->index)
1892 iter->pg = iter->pg->next;
1893
1894 if (!iter->pg)
1895 return NULL;
1896
1897 return iter;
1898}
1899
1900/**
1901 * ftrace_rec_iter_next, get the next record to process.
1902 * @iter: The handle to the iterator.
1903 *
1904 * Returns the next iterator after the given iterator @iter.
1905 */
1906struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
1907{
1908 iter->index++;
1909
1910 if (iter->index >= iter->pg->index) {
1911 iter->pg = iter->pg->next;
1912 iter->index = 0;
1913
1914 /* Could have empty pages */
1915 while (iter->pg && !iter->pg->index)
1916 iter->pg = iter->pg->next;
1917 }
1918
1919 if (!iter->pg)
1920 return NULL;
1921
1922 return iter;
1923}
1924
1925/**
1926 * ftrace_rec_iter_record, get the record at the iterator location
1927 * @iter: The current iterator location
1928 *
1929 * Returns the record that the current @iter is at.
1930 */
1931struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
1932{
1933 return &iter->pg->records[iter->index];
1934}
1935
1936static int
1937ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1938{
1939 unsigned long ip;
1940 int ret;
1941
1942 ip = rec->ip;
1943
1944 if (unlikely(ftrace_disabled))
1945 return 0;
1946
1947 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1948 if (ret) {
1949 ftrace_bug(ret, ip);
1950 return 0;
1951 }
1952 return 1;
1953}
1954
1955/*
1956 * archs can override this function if they must do something
1957 * before the modifying code is performed.
1958 */
1959int __weak ftrace_arch_code_modify_prepare(void)
1960{
1961 return 0;
1962}
1963
1964/*
1965 * archs can override this function if they must do something
1966 * after the modifying code is performed.
1967 */
1968int __weak ftrace_arch_code_modify_post_process(void)
1969{
1970 return 0;
1971}
1972
1973void ftrace_modify_all_code(int command)
1974{
1975 int update = command & FTRACE_UPDATE_TRACE_FUNC;
1976
1977 /*
1978 * If the ftrace_caller calls a ftrace_ops func directly,
1979 * we need to make sure that it only traces functions it
1980 * expects to trace. When doing the switch of functions,
1981 * we need to update to the ftrace_ops_list_func first
1982 * before the transition between old and new calls are set,
1983 * as the ftrace_ops_list_func will check the ops hashes
1984 * to make sure the ops are having the right functions
1985 * traced.
1986 */
1987 if (update)
1988 ftrace_update_ftrace_func(ftrace_ops_list_func);
1989
1990 if (command & FTRACE_UPDATE_CALLS)
1991 ftrace_replace_code(1);
1992 else if (command & FTRACE_DISABLE_CALLS)
1993 ftrace_replace_code(0);
1994
1995 if (update && ftrace_trace_function != ftrace_ops_list_func)
1996 ftrace_update_ftrace_func(ftrace_trace_function);
1997
1998 if (command & FTRACE_START_FUNC_RET)
1999 ftrace_enable_ftrace_graph_caller();
2000 else if (command & FTRACE_STOP_FUNC_RET)
2001 ftrace_disable_ftrace_graph_caller();
2002}
2003
2004static int __ftrace_modify_code(void *data)
2005{
2006 int *command = data;
2007
2008 ftrace_modify_all_code(*command);
2009
2010 return 0;
2011}
2012
2013/**
2014 * ftrace_run_stop_machine, go back to the stop machine method
2015 * @command: The command to tell ftrace what to do
2016 *
2017 * If an arch needs to fall back to the stop machine method, the
2018 * it can call this function.
2019 */
2020void ftrace_run_stop_machine(int command)
2021{
2022 stop_machine(__ftrace_modify_code, &command, NULL);
2023}
2024
2025/**
2026 * arch_ftrace_update_code, modify the code to trace or not trace
2027 * @command: The command that needs to be done
2028 *
2029 * Archs can override this function if it does not need to
2030 * run stop_machine() to modify code.
2031 */
2032void __weak arch_ftrace_update_code(int command)
2033{
2034 ftrace_run_stop_machine(command);
2035}
2036
2037static void ftrace_run_update_code(int command)
2038{
2039 int ret;
2040
2041 ret = ftrace_arch_code_modify_prepare();
2042 FTRACE_WARN_ON(ret);
2043 if (ret)
2044 return;
2045 /*
2046 * Do not call function tracer while we update the code.
2047 * We are in stop machine.
2048 */
2049 function_trace_stop++;
2050
2051 /*
2052 * By default we use stop_machine() to modify the code.
2053 * But archs can do what ever they want as long as it
2054 * is safe. The stop_machine() is the safest, but also
2055 * produces the most overhead.
2056 */
2057 arch_ftrace_update_code(command);
2058
2059 function_trace_stop--;
2060
2061 ret = ftrace_arch_code_modify_post_process();
2062 FTRACE_WARN_ON(ret);
2063}
2064
2065static ftrace_func_t saved_ftrace_func;
2066static int ftrace_start_up;
2067static int global_start_up;
2068
2069static void ftrace_startup_enable(int command)
2070{
2071 if (saved_ftrace_func != ftrace_trace_function) {
2072 saved_ftrace_func = ftrace_trace_function;
2073 command |= FTRACE_UPDATE_TRACE_FUNC;
2074 }
2075
2076 if (!command || !ftrace_enabled)
2077 return;
2078
2079 ftrace_run_update_code(command);
2080}
2081
2082static int ftrace_startup(struct ftrace_ops *ops, int command)
2083{
2084 bool hash_enable = true;
2085 int ret;
2086
2087 if (unlikely(ftrace_disabled))
2088 return -ENODEV;
2089
2090 ret = __register_ftrace_function(ops);
2091 if (ret)
2092 return ret;
2093
2094 ftrace_start_up++;
2095 command |= FTRACE_UPDATE_CALLS;
2096
2097 /* ops marked global share the filter hashes */
2098 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2099 ops = &global_ops;
2100 /* Don't update hash if global is already set */
2101 if (global_start_up)
2102 hash_enable = false;
2103 global_start_up++;
2104 }
2105
2106 ops->flags |= FTRACE_OPS_FL_ENABLED;
2107 if (hash_enable)
2108 ftrace_hash_rec_enable(ops, 1);
2109
2110 ftrace_startup_enable(command);
2111
2112 return 0;
2113}
2114
2115static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2116{
2117 bool hash_disable = true;
2118 int ret;
2119
2120 if (unlikely(ftrace_disabled))
2121 return -ENODEV;
2122
2123 ret = __unregister_ftrace_function(ops);
2124 if (ret)
2125 return ret;
2126
2127 ftrace_start_up--;
2128 /*
2129 * Just warn in case of unbalance, no need to kill ftrace, it's not
2130 * critical but the ftrace_call callers may be never nopped again after
2131 * further ftrace uses.
2132 */
2133 WARN_ON_ONCE(ftrace_start_up < 0);
2134
2135 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2136 ops = &global_ops;
2137 global_start_up--;
2138 WARN_ON_ONCE(global_start_up < 0);
2139 /* Don't update hash if global still has users */
2140 if (global_start_up) {
2141 WARN_ON_ONCE(!ftrace_start_up);
2142 hash_disable = false;
2143 }
2144 }
2145
2146 if (hash_disable)
2147 ftrace_hash_rec_disable(ops, 1);
2148
2149 if (ops != &global_ops || !global_start_up)
2150 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2151
2152 command |= FTRACE_UPDATE_CALLS;
2153
2154 if (saved_ftrace_func != ftrace_trace_function) {
2155 saved_ftrace_func = ftrace_trace_function;
2156 command |= FTRACE_UPDATE_TRACE_FUNC;
2157 }
2158
2159 if (!command || !ftrace_enabled)
2160 return 0;
2161
2162 ftrace_run_update_code(command);
2163 return 0;
2164}
2165
2166static void ftrace_startup_sysctl(void)
2167{
2168 if (unlikely(ftrace_disabled))
2169 return;
2170
2171 /* Force update next time */
2172 saved_ftrace_func = NULL;
2173 /* ftrace_start_up is true if we want ftrace running */
2174 if (ftrace_start_up)
2175 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
2176}
2177
2178static void ftrace_shutdown_sysctl(void)
2179{
2180 if (unlikely(ftrace_disabled))
2181 return;
2182
2183 /* ftrace_start_up is true if ftrace is running */
2184 if (ftrace_start_up)
2185 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
2186}
2187
2188static cycle_t ftrace_update_time;
2189static unsigned long ftrace_update_cnt;
2190unsigned long ftrace_update_tot_cnt;
2191
2192static inline int ops_traces_mod(struct ftrace_ops *ops)
2193{
2194 /*
2195 * Filter_hash being empty will default to trace module.
2196 * But notrace hash requires a test of individual module functions.
2197 */
2198 return ftrace_hash_empty(ops->filter_hash) &&
2199 ftrace_hash_empty(ops->notrace_hash);
2200}
2201
2202/*
2203 * Check if the current ops references the record.
2204 *
2205 * If the ops traces all functions, then it was already accounted for.
2206 * If the ops does not trace the current record function, skip it.
2207 * If the ops ignores the function via notrace filter, skip it.
2208 */
2209static inline bool
2210ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2211{
2212 /* If ops isn't enabled, ignore it */
2213 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2214 return 0;
2215
2216 /* If ops traces all mods, we already accounted for it */
2217 if (ops_traces_mod(ops))
2218 return 0;
2219
2220 /* The function must be in the filter */
2221 if (!ftrace_hash_empty(ops->filter_hash) &&
2222 !ftrace_lookup_ip(ops->filter_hash, rec->ip))
2223 return 0;
2224
2225 /* If in notrace hash, we ignore it too */
2226 if (ftrace_lookup_ip(ops->notrace_hash, rec->ip))
2227 return 0;
2228
2229 return 1;
2230}
2231
2232static int referenced_filters(struct dyn_ftrace *rec)
2233{
2234 struct ftrace_ops *ops;
2235 int cnt = 0;
2236
2237 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
2238 if (ops_references_rec(ops, rec))
2239 cnt++;
2240 }
2241
2242 return cnt;
2243}
2244
2245static int ftrace_update_code(struct module *mod)
2246{
2247 struct ftrace_page *pg;
2248 struct dyn_ftrace *p;
2249 cycle_t start, stop;
2250 unsigned long ref = 0;
2251 bool test = false;
2252 int i;
2253
2254 /*
2255 * When adding a module, we need to check if tracers are
2256 * currently enabled and if they are set to trace all functions.
2257 * If they are, we need to enable the module functions as well
2258 * as update the reference counts for those function records.
2259 */
2260 if (mod) {
2261 struct ftrace_ops *ops;
2262
2263 for (ops = ftrace_ops_list;
2264 ops != &ftrace_list_end; ops = ops->next) {
2265 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
2266 if (ops_traces_mod(ops))
2267 ref++;
2268 else
2269 test = true;
2270 }
2271 }
2272 }
2273
2274 start = ftrace_now(raw_smp_processor_id());
2275 ftrace_update_cnt = 0;
2276
2277 for (pg = ftrace_new_pgs; pg; pg = pg->next) {
2278
2279 for (i = 0; i < pg->index; i++) {
2280 int cnt = ref;
2281
2282 /* If something went wrong, bail without enabling anything */
2283 if (unlikely(ftrace_disabled))
2284 return -1;
2285
2286 p = &pg->records[i];
2287 if (test)
2288 cnt += referenced_filters(p);
2289 p->flags = cnt;
2290
2291 /*
2292 * Do the initial record conversion from mcount jump
2293 * to the NOP instructions.
2294 */
2295 if (!ftrace_code_disable(mod, p))
2296 break;
2297
2298 ftrace_update_cnt++;
2299
2300 /*
2301 * If the tracing is enabled, go ahead and enable the record.
2302 *
2303 * The reason not to enable the record immediatelly is the
2304 * inherent check of ftrace_make_nop/ftrace_make_call for
2305 * correct previous instructions. Making first the NOP
2306 * conversion puts the module to the correct state, thus
2307 * passing the ftrace_make_call check.
2308 */
2309 if (ftrace_start_up && cnt) {
2310 int failed = __ftrace_replace_code(p, 1);
2311 if (failed)
2312 ftrace_bug(failed, p->ip);
2313 }
2314 }
2315 }
2316
2317 ftrace_new_pgs = NULL;
2318
2319 stop = ftrace_now(raw_smp_processor_id());
2320 ftrace_update_time = stop - start;
2321 ftrace_update_tot_cnt += ftrace_update_cnt;
2322
2323 return 0;
2324}
2325
2326static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2327{
2328 int order;
2329 int cnt;
2330
2331 if (WARN_ON(!count))
2332 return -EINVAL;
2333
2334 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2335
2336 /*
2337 * We want to fill as much as possible. No more than a page
2338 * may be empty.
2339 */
2340 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2341 order--;
2342
2343 again:
2344 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2345
2346 if (!pg->records) {
2347 /* if we can't allocate this size, try something smaller */
2348 if (!order)
2349 return -ENOMEM;
2350 order >>= 1;
2351 goto again;
2352 }
2353
2354 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2355 pg->size = cnt;
2356
2357 if (cnt > count)
2358 cnt = count;
2359
2360 return cnt;
2361}
2362
2363static struct ftrace_page *
2364ftrace_allocate_pages(unsigned long num_to_init)
2365{
2366 struct ftrace_page *start_pg;
2367 struct ftrace_page *pg;
2368 int order;
2369 int cnt;
2370
2371 if (!num_to_init)
2372 return 0;
2373
2374 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2375 if (!pg)
2376 return NULL;
2377
2378 /*
2379 * Try to allocate as much as possible in one continues
2380 * location that fills in all of the space. We want to
2381 * waste as little space as possible.
2382 */
2383 for (;;) {
2384 cnt = ftrace_allocate_records(pg, num_to_init);
2385 if (cnt < 0)
2386 goto free_pages;
2387
2388 num_to_init -= cnt;
2389 if (!num_to_init)
2390 break;
2391
2392 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2393 if (!pg->next)
2394 goto free_pages;
2395
2396 pg = pg->next;
2397 }
2398
2399 return start_pg;
2400
2401 free_pages:
2402 while (start_pg) {
2403 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2404 free_pages((unsigned long)pg->records, order);
2405 start_pg = pg->next;
2406 kfree(pg);
2407 pg = start_pg;
2408 }
2409 pr_info("ftrace: FAILED to allocate memory for functions\n");
2410 return NULL;
2411}
2412
2413static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
2414{
2415 int cnt;
2416
2417 if (!num_to_init) {
2418 pr_info("ftrace: No functions to be traced?\n");
2419 return -1;
2420 }
2421
2422 cnt = num_to_init / ENTRIES_PER_PAGE;
2423 pr_info("ftrace: allocating %ld entries in %d pages\n",
2424 num_to_init, cnt + 1);
2425
2426 return 0;
2427}
2428
2429#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
2430
2431struct ftrace_iterator {
2432 loff_t pos;
2433 loff_t func_pos;
2434 struct ftrace_page *pg;
2435 struct dyn_ftrace *func;
2436 struct ftrace_func_probe *probe;
2437 struct trace_parser parser;
2438 struct ftrace_hash *hash;
2439 struct ftrace_ops *ops;
2440 int hidx;
2441 int idx;
2442 unsigned flags;
2443};
2444
2445static void *
2446t_hash_next(struct seq_file *m, loff_t *pos)
2447{
2448 struct ftrace_iterator *iter = m->private;
2449 struct hlist_node *hnd = NULL;
2450 struct hlist_head *hhd;
2451
2452 (*pos)++;
2453 iter->pos = *pos;
2454
2455 if (iter->probe)
2456 hnd = &iter->probe->node;
2457 retry:
2458 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
2459 return NULL;
2460
2461 hhd = &ftrace_func_hash[iter->hidx];
2462
2463 if (hlist_empty(hhd)) {
2464 iter->hidx++;
2465 hnd = NULL;
2466 goto retry;
2467 }
2468
2469 if (!hnd)
2470 hnd = hhd->first;
2471 else {
2472 hnd = hnd->next;
2473 if (!hnd) {
2474 iter->hidx++;
2475 goto retry;
2476 }
2477 }
2478
2479 if (WARN_ON_ONCE(!hnd))
2480 return NULL;
2481
2482 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
2483
2484 return iter;
2485}
2486
2487static void *t_hash_start(struct seq_file *m, loff_t *pos)
2488{
2489 struct ftrace_iterator *iter = m->private;
2490 void *p = NULL;
2491 loff_t l;
2492
2493 if (!(iter->flags & FTRACE_ITER_DO_HASH))
2494 return NULL;
2495
2496 if (iter->func_pos > *pos)
2497 return NULL;
2498
2499 iter->hidx = 0;
2500 for (l = 0; l <= (*pos - iter->func_pos); ) {
2501 p = t_hash_next(m, &l);
2502 if (!p)
2503 break;
2504 }
2505 if (!p)
2506 return NULL;
2507
2508 /* Only set this if we have an item */
2509 iter->flags |= FTRACE_ITER_HASH;
2510
2511 return iter;
2512}
2513
2514static int
2515t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2516{
2517 struct ftrace_func_probe *rec;
2518
2519 rec = iter->probe;
2520 if (WARN_ON_ONCE(!rec))
2521 return -EIO;
2522
2523 if (rec->ops->print)
2524 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2525
2526 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2527
2528 if (rec->data)
2529 seq_printf(m, ":%p", rec->data);
2530 seq_putc(m, '\n');
2531
2532 return 0;
2533}
2534
2535static void *
2536t_next(struct seq_file *m, void *v, loff_t *pos)
2537{
2538 struct ftrace_iterator *iter = m->private;
2539 struct ftrace_ops *ops = iter->ops;
2540 struct dyn_ftrace *rec = NULL;
2541
2542 if (unlikely(ftrace_disabled))
2543 return NULL;
2544
2545 if (iter->flags & FTRACE_ITER_HASH)
2546 return t_hash_next(m, pos);
2547
2548 (*pos)++;
2549 iter->pos = iter->func_pos = *pos;
2550
2551 if (iter->flags & FTRACE_ITER_PRINTALL)
2552 return t_hash_start(m, pos);
2553
2554 retry:
2555 if (iter->idx >= iter->pg->index) {
2556 if (iter->pg->next) {
2557 iter->pg = iter->pg->next;
2558 iter->idx = 0;
2559 goto retry;
2560 }
2561 } else {
2562 rec = &iter->pg->records[iter->idx++];
2563 if (((iter->flags & FTRACE_ITER_FILTER) &&
2564 !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2565
2566 ((iter->flags & FTRACE_ITER_NOTRACE) &&
2567 !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2568
2569 ((iter->flags & FTRACE_ITER_ENABLED) &&
2570 !(rec->flags & FTRACE_FL_ENABLED))) {
2571
2572 rec = NULL;
2573 goto retry;
2574 }
2575 }
2576
2577 if (!rec)
2578 return t_hash_start(m, pos);
2579
2580 iter->func = rec;
2581
2582 return iter;
2583}
2584
2585static void reset_iter_read(struct ftrace_iterator *iter)
2586{
2587 iter->pos = 0;
2588 iter->func_pos = 0;
2589 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
2590}
2591
2592static void *t_start(struct seq_file *m, loff_t *pos)
2593{
2594 struct ftrace_iterator *iter = m->private;
2595 struct ftrace_ops *ops = iter->ops;
2596 void *p = NULL;
2597 loff_t l;
2598
2599 mutex_lock(&ftrace_lock);
2600
2601 if (unlikely(ftrace_disabled))
2602 return NULL;
2603
2604 /*
2605 * If an lseek was done, then reset and start from beginning.
2606 */
2607 if (*pos < iter->pos)
2608 reset_iter_read(iter);
2609
2610 /*
2611 * For set_ftrace_filter reading, if we have the filter
2612 * off, we can short cut and just print out that all
2613 * functions are enabled.
2614 */
2615 if (iter->flags & FTRACE_ITER_FILTER &&
2616 ftrace_hash_empty(ops->filter_hash)) {
2617 if (*pos > 0)
2618 return t_hash_start(m, pos);
2619 iter->flags |= FTRACE_ITER_PRINTALL;
2620 /* reset in case of seek/pread */
2621 iter->flags &= ~FTRACE_ITER_HASH;
2622 return iter;
2623 }
2624
2625 if (iter->flags & FTRACE_ITER_HASH)
2626 return t_hash_start(m, pos);
2627
2628 /*
2629 * Unfortunately, we need to restart at ftrace_pages_start
2630 * every time we let go of the ftrace_mutex. This is because
2631 * those pointers can change without the lock.
2632 */
2633 iter->pg = ftrace_pages_start;
2634 iter->idx = 0;
2635 for (l = 0; l <= *pos; ) {
2636 p = t_next(m, p, &l);
2637 if (!p)
2638 break;
2639 }
2640
2641 if (!p)
2642 return t_hash_start(m, pos);
2643
2644 return iter;
2645}
2646
2647static void t_stop(struct seq_file *m, void *p)
2648{
2649 mutex_unlock(&ftrace_lock);
2650}
2651
2652static int t_show(struct seq_file *m, void *v)
2653{
2654 struct ftrace_iterator *iter = m->private;
2655 struct dyn_ftrace *rec;
2656
2657 if (iter->flags & FTRACE_ITER_HASH)
2658 return t_hash_show(m, iter);
2659
2660 if (iter->flags & FTRACE_ITER_PRINTALL) {
2661 seq_printf(m, "#### all functions enabled ####\n");
2662 return 0;
2663 }
2664
2665 rec = iter->func;
2666
2667 if (!rec)
2668 return 0;
2669
2670 seq_printf(m, "%ps", (void *)rec->ip);
2671 if (iter->flags & FTRACE_ITER_ENABLED)
2672 seq_printf(m, " (%ld)%s",
2673 rec->flags & ~FTRACE_FL_MASK,
2674 rec->flags & FTRACE_FL_REGS ? " R" : "");
2675 seq_printf(m, "\n");
2676
2677 return 0;
2678}
2679
2680static const struct seq_operations show_ftrace_seq_ops = {
2681 .start = t_start,
2682 .next = t_next,
2683 .stop = t_stop,
2684 .show = t_show,
2685};
2686
2687static int
2688ftrace_avail_open(struct inode *inode, struct file *file)
2689{
2690 struct ftrace_iterator *iter;
2691
2692 if (unlikely(ftrace_disabled))
2693 return -ENODEV;
2694
2695 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2696 if (iter) {
2697 iter->pg = ftrace_pages_start;
2698 iter->ops = &global_ops;
2699 }
2700
2701 return iter ? 0 : -ENOMEM;
2702}
2703
2704static int
2705ftrace_enabled_open(struct inode *inode, struct file *file)
2706{
2707 struct ftrace_iterator *iter;
2708
2709 if (unlikely(ftrace_disabled))
2710 return -ENODEV;
2711
2712 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2713 if (iter) {
2714 iter->pg = ftrace_pages_start;
2715 iter->flags = FTRACE_ITER_ENABLED;
2716 iter->ops = &global_ops;
2717 }
2718
2719 return iter ? 0 : -ENOMEM;
2720}
2721
2722static void ftrace_filter_reset(struct ftrace_hash *hash)
2723{
2724 mutex_lock(&ftrace_lock);
2725 ftrace_hash_clear(hash);
2726 mutex_unlock(&ftrace_lock);
2727}
2728
2729/**
2730 * ftrace_regex_open - initialize function tracer filter files
2731 * @ops: The ftrace_ops that hold the hash filters
2732 * @flag: The type of filter to process
2733 * @inode: The inode, usually passed in to your open routine
2734 * @file: The file, usually passed in to your open routine
2735 *
2736 * ftrace_regex_open() initializes the filter files for the
2737 * @ops. Depending on @flag it may process the filter hash or
2738 * the notrace hash of @ops. With this called from the open
2739 * routine, you can use ftrace_filter_write() for the write
2740 * routine if @flag has FTRACE_ITER_FILTER set, or
2741 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
2742 * ftrace_filter_lseek() should be used as the lseek routine, and
2743 * release must call ftrace_regex_release().
2744 */
2745int
2746ftrace_regex_open(struct ftrace_ops *ops, int flag,
2747 struct inode *inode, struct file *file)
2748{
2749 struct ftrace_iterator *iter;
2750 struct ftrace_hash *hash;
2751 int ret = 0;
2752
2753 ftrace_ops_init(ops);
2754
2755 if (unlikely(ftrace_disabled))
2756 return -ENODEV;
2757
2758 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2759 if (!iter)
2760 return -ENOMEM;
2761
2762 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2763 kfree(iter);
2764 return -ENOMEM;
2765 }
2766
2767 iter->ops = ops;
2768 iter->flags = flag;
2769
2770 mutex_lock(&ops->regex_lock);
2771
2772 if (flag & FTRACE_ITER_NOTRACE)
2773 hash = ops->notrace_hash;
2774 else
2775 hash = ops->filter_hash;
2776
2777 if (file->f_mode & FMODE_WRITE) {
2778 iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2779 if (!iter->hash) {
2780 trace_parser_put(&iter->parser);
2781 kfree(iter);
2782 ret = -ENOMEM;
2783 goto out_unlock;
2784 }
2785 }
2786
2787 if ((file->f_mode & FMODE_WRITE) &&
2788 (file->f_flags & O_TRUNC))
2789 ftrace_filter_reset(iter->hash);
2790
2791 if (file->f_mode & FMODE_READ) {
2792 iter->pg = ftrace_pages_start;
2793
2794 ret = seq_open(file, &show_ftrace_seq_ops);
2795 if (!ret) {
2796 struct seq_file *m = file->private_data;
2797 m->private = iter;
2798 } else {
2799 /* Failed */
2800 free_ftrace_hash(iter->hash);
2801 trace_parser_put(&iter->parser);
2802 kfree(iter);
2803 }
2804 } else
2805 file->private_data = iter;
2806
2807 out_unlock:
2808 mutex_unlock(&ops->regex_lock);
2809
2810 return ret;
2811}
2812
2813static int
2814ftrace_filter_open(struct inode *inode, struct file *file)
2815{
2816 return ftrace_regex_open(&global_ops,
2817 FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
2818 inode, file);
2819}
2820
2821static int
2822ftrace_notrace_open(struct inode *inode, struct file *file)
2823{
2824 return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2825 inode, file);
2826}
2827
2828static int ftrace_match(char *str, char *regex, int len, int type)
2829{
2830 int matched = 0;
2831 int slen;
2832
2833 switch (type) {
2834 case MATCH_FULL:
2835 if (strcmp(str, regex) == 0)
2836 matched = 1;
2837 break;
2838 case MATCH_FRONT_ONLY:
2839 if (strncmp(str, regex, len) == 0)
2840 matched = 1;
2841 break;
2842 case MATCH_MIDDLE_ONLY:
2843 if (strstr(str, regex))
2844 matched = 1;
2845 break;
2846 case MATCH_END_ONLY:
2847 slen = strlen(str);
2848 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2849 matched = 1;
2850 break;
2851 }
2852
2853 return matched;
2854}
2855
2856static int
2857enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2858{
2859 struct ftrace_func_entry *entry;
2860 int ret = 0;
2861
2862 entry = ftrace_lookup_ip(hash, rec->ip);
2863 if (not) {
2864 /* Do nothing if it doesn't exist */
2865 if (!entry)
2866 return 0;
2867
2868 free_hash_entry(hash, entry);
2869 } else {
2870 /* Do nothing if it exists */
2871 if (entry)
2872 return 0;
2873
2874 ret = add_hash_entry(hash, rec->ip);
2875 }
2876 return ret;
2877}
2878
2879static int
2880ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2881 char *regex, int len, int type)
2882{
2883 char str[KSYM_SYMBOL_LEN];
2884 char *modname;
2885
2886 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2887
2888 if (mod) {
2889 /* module lookup requires matching the module */
2890 if (!modname || strcmp(modname, mod))
2891 return 0;
2892
2893 /* blank search means to match all funcs in the mod */
2894 if (!len)
2895 return 1;
2896 }
2897
2898 return ftrace_match(str, regex, len, type);
2899}
2900
2901static int
2902match_records(struct ftrace_hash *hash, char *buff,
2903 int len, char *mod, int not)
2904{
2905 unsigned search_len = 0;
2906 struct ftrace_page *pg;
2907 struct dyn_ftrace *rec;
2908 int type = MATCH_FULL;
2909 char *search = buff;
2910 int found = 0;
2911 int ret;
2912
2913 if (len) {
2914 type = filter_parse_regex(buff, len, &search, &not);
2915 search_len = strlen(search);
2916 }
2917
2918 mutex_lock(&ftrace_lock);
2919
2920 if (unlikely(ftrace_disabled))
2921 goto out_unlock;
2922
2923 do_for_each_ftrace_rec(pg, rec) {
2924 if (ftrace_match_record(rec, mod, search, search_len, type)) {
2925 ret = enter_record(hash, rec, not);
2926 if (ret < 0) {
2927 found = ret;
2928 goto out_unlock;
2929 }
2930 found = 1;
2931 }
2932 } while_for_each_ftrace_rec();
2933 out_unlock:
2934 mutex_unlock(&ftrace_lock);
2935
2936 return found;
2937}
2938
2939static int
2940ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2941{
2942 return match_records(hash, buff, len, NULL, 0);
2943}
2944
2945static int
2946ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2947{
2948 int not = 0;
2949
2950 /* blank or '*' mean the same */
2951 if (strcmp(buff, "*") == 0)
2952 buff[0] = 0;
2953
2954 /* handle the case of 'dont filter this module' */
2955 if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2956 buff[0] = 0;
2957 not = 1;
2958 }
2959
2960 return match_records(hash, buff, strlen(buff), mod, not);
2961}
2962
2963/*
2964 * We register the module command as a template to show others how
2965 * to register the a command as well.
2966 */
2967
2968static int
2969ftrace_mod_callback(struct ftrace_hash *hash,
2970 char *func, char *cmd, char *param, int enable)
2971{
2972 char *mod;
2973 int ret = -EINVAL;
2974
2975 /*
2976 * cmd == 'mod' because we only registered this func
2977 * for the 'mod' ftrace_func_command.
2978 * But if you register one func with multiple commands,
2979 * you can tell which command was used by the cmd
2980 * parameter.
2981 */
2982
2983 /* we must have a module name */
2984 if (!param)
2985 return ret;
2986
2987 mod = strsep(&param, ":");
2988 if (!strlen(mod))
2989 return ret;
2990
2991 ret = ftrace_match_module_records(hash, func, mod);
2992 if (!ret)
2993 ret = -EINVAL;
2994 if (ret < 0)
2995 return ret;
2996
2997 return 0;
2998}
2999
3000static struct ftrace_func_command ftrace_mod_cmd = {
3001 .name = "mod",
3002 .func = ftrace_mod_callback,
3003};
3004
3005static int __init ftrace_mod_cmd_init(void)
3006{
3007 return register_ftrace_command(&ftrace_mod_cmd);
3008}
3009core_initcall(ftrace_mod_cmd_init);
3010
3011static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
3012 struct ftrace_ops *op, struct pt_regs *pt_regs)
3013{
3014 struct ftrace_func_probe *entry;
3015 struct hlist_head *hhd;
3016 unsigned long key;
3017
3018 key = hash_long(ip, FTRACE_HASH_BITS);
3019
3020 hhd = &ftrace_func_hash[key];
3021
3022 if (hlist_empty(hhd))
3023 return;
3024
3025 /*
3026 * Disable preemption for these calls to prevent a RCU grace
3027 * period. This syncs the hash iteration and freeing of items
3028 * on the hash. rcu_read_lock is too dangerous here.
3029 */
3030 preempt_disable_notrace();
3031 hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3032 if (entry->ip == ip)
3033 entry->ops->func(ip, parent_ip, &entry->data);
3034 }
3035 preempt_enable_notrace();
3036}
3037
3038static struct ftrace_ops trace_probe_ops __read_mostly =
3039{
3040 .func = function_trace_probe_call,
3041 .flags = FTRACE_OPS_FL_INITIALIZED,
3042 INIT_REGEX_LOCK(trace_probe_ops)
3043};
3044
3045static int ftrace_probe_registered;
3046
3047static void __enable_ftrace_function_probe(void)
3048{
3049 int ret;
3050 int i;
3051
3052 if (ftrace_probe_registered) {
3053 /* still need to update the function call sites */
3054 if (ftrace_enabled)
3055 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3056 return;
3057 }
3058
3059 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3060 struct hlist_head *hhd = &ftrace_func_hash[i];
3061 if (hhd->first)
3062 break;
3063 }
3064 /* Nothing registered? */
3065 if (i == FTRACE_FUNC_HASHSIZE)
3066 return;
3067
3068 ret = ftrace_startup(&trace_probe_ops, 0);
3069
3070 ftrace_probe_registered = 1;
3071}
3072
3073static void __disable_ftrace_function_probe(void)
3074{
3075 int i;
3076
3077 if (!ftrace_probe_registered)
3078 return;
3079
3080 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3081 struct hlist_head *hhd = &ftrace_func_hash[i];
3082 if (hhd->first)
3083 return;
3084 }
3085
3086 /* no more funcs left */
3087 ftrace_shutdown(&trace_probe_ops, 0);
3088
3089 ftrace_probe_registered = 0;
3090}
3091
3092
3093static void ftrace_free_entry(struct ftrace_func_probe *entry)
3094{
3095 if (entry->ops->free)
3096 entry->ops->free(entry->ops, entry->ip, &entry->data);
3097 kfree(entry);
3098}
3099
3100int
3101register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3102 void *data)
3103{
3104 struct ftrace_func_probe *entry;
3105 struct ftrace_hash **orig_hash = &trace_probe_ops.filter_hash;
3106 struct ftrace_hash *hash;
3107 struct ftrace_page *pg;
3108 struct dyn_ftrace *rec;
3109 int type, len, not;
3110 unsigned long key;
3111 int count = 0;
3112 char *search;
3113 int ret;
3114
3115 type = filter_parse_regex(glob, strlen(glob), &search, &not);
3116 len = strlen(search);
3117
3118 /* we do not support '!' for function probes */
3119 if (WARN_ON(not))
3120 return -EINVAL;
3121
3122 mutex_lock(&trace_probe_ops.regex_lock);
3123
3124 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3125 if (!hash) {
3126 count = -ENOMEM;
3127 goto out;
3128 }
3129
3130 if (unlikely(ftrace_disabled)) {
3131 count = -ENODEV;
3132 goto out;
3133 }
3134
3135 mutex_lock(&ftrace_lock);
3136
3137 do_for_each_ftrace_rec(pg, rec) {
3138
3139 if (!ftrace_match_record(rec, NULL, search, len, type))
3140 continue;
3141
3142 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3143 if (!entry) {
3144 /* If we did not process any, then return error */
3145 if (!count)
3146 count = -ENOMEM;
3147 goto out_unlock;
3148 }
3149
3150 count++;
3151
3152 entry->data = data;
3153
3154 /*
3155 * The caller might want to do something special
3156 * for each function we find. We call the callback
3157 * to give the caller an opportunity to do so.
3158 */
3159 if (ops->init) {
3160 if (ops->init(ops, rec->ip, &entry->data) < 0) {
3161 /* caller does not like this func */
3162 kfree(entry);
3163 continue;
3164 }
3165 }
3166
3167 ret = enter_record(hash, rec, 0);
3168 if (ret < 0) {
3169 kfree(entry);
3170 count = ret;
3171 goto out_unlock;
3172 }
3173
3174 entry->ops = ops;
3175 entry->ip = rec->ip;
3176
3177 key = hash_long(entry->ip, FTRACE_HASH_BITS);
3178 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3179
3180 } while_for_each_ftrace_rec();
3181
3182 ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3183 if (ret < 0)
3184 count = ret;
3185
3186 __enable_ftrace_function_probe();
3187
3188 out_unlock:
3189 mutex_unlock(&ftrace_lock);
3190 out:
3191 mutex_unlock(&trace_probe_ops.regex_lock);
3192 free_ftrace_hash(hash);
3193
3194 return count;
3195}
3196
3197enum {
3198 PROBE_TEST_FUNC = 1,
3199 PROBE_TEST_DATA = 2
3200};
3201
3202static void
3203__unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3204 void *data, int flags)
3205{
3206 struct ftrace_func_entry *rec_entry;
3207 struct ftrace_func_probe *entry;
3208 struct ftrace_func_probe *p;
3209 struct ftrace_hash **orig_hash = &trace_probe_ops.filter_hash;
3210 struct list_head free_list;
3211 struct ftrace_hash *hash;
3212 struct hlist_node *tmp;
3213 char str[KSYM_SYMBOL_LEN];
3214 int type = MATCH_FULL;
3215 int i, len = 0;
3216 char *search;
3217
3218 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3219 glob = NULL;
3220 else if (glob) {
3221 int not;
3222
3223 type = filter_parse_regex(glob, strlen(glob), &search, &not);
3224 len = strlen(search);
3225
3226 /* we do not support '!' for function probes */
3227 if (WARN_ON(not))
3228 return;
3229 }
3230
3231 mutex_lock(&trace_probe_ops.regex_lock);
3232
3233 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3234 if (!hash)
3235 /* Hmm, should report this somehow */
3236 goto out_unlock;
3237
3238 INIT_LIST_HEAD(&free_list);
3239
3240 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3241 struct hlist_head *hhd = &ftrace_func_hash[i];
3242
3243 hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3244
3245 /* break up if statements for readability */
3246 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3247 continue;
3248
3249 if ((flags & PROBE_TEST_DATA) && entry->data != data)
3250 continue;
3251
3252 /* do this last, since it is the most expensive */
3253 if (glob) {
3254 kallsyms_lookup(entry->ip, NULL, NULL,
3255 NULL, str);
3256 if (!ftrace_match(str, glob, len, type))
3257 continue;
3258 }
3259
3260 rec_entry = ftrace_lookup_ip(hash, entry->ip);
3261 /* It is possible more than one entry had this ip */
3262 if (rec_entry)
3263 free_hash_entry(hash, rec_entry);
3264
3265 hlist_del_rcu(&entry->node);
3266 list_add(&entry->free_list, &free_list);
3267 }
3268 }
3269 mutex_lock(&ftrace_lock);
3270 __disable_ftrace_function_probe();
3271 /*
3272 * Remove after the disable is called. Otherwise, if the last
3273 * probe is removed, a null hash means *all enabled*.
3274 */
3275 ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3276 synchronize_sched();
3277 list_for_each_entry_safe(entry, p, &free_list, free_list) {
3278 list_del(&entry->free_list);
3279 ftrace_free_entry(entry);
3280 }
3281 mutex_unlock(&ftrace_lock);
3282
3283 out_unlock:
3284 mutex_unlock(&trace_probe_ops.regex_lock);
3285 free_ftrace_hash(hash);
3286}
3287
3288void
3289unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3290 void *data)
3291{
3292 __unregister_ftrace_function_probe(glob, ops, data,
3293 PROBE_TEST_FUNC | PROBE_TEST_DATA);
3294}
3295
3296void
3297unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3298{
3299 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3300}
3301
3302void unregister_ftrace_function_probe_all(char *glob)
3303{
3304 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3305}
3306
3307static LIST_HEAD(ftrace_commands);
3308static DEFINE_MUTEX(ftrace_cmd_mutex);
3309
3310/*
3311 * Currently we only register ftrace commands from __init, so mark this
3312 * __init too.
3313 */
3314__init int register_ftrace_command(struct ftrace_func_command *cmd)
3315{
3316 struct ftrace_func_command *p;
3317 int ret = 0;
3318
3319 mutex_lock(&ftrace_cmd_mutex);
3320 list_for_each_entry(p, &ftrace_commands, list) {
3321 if (strcmp(cmd->name, p->name) == 0) {
3322 ret = -EBUSY;
3323 goto out_unlock;
3324 }
3325 }
3326 list_add(&cmd->list, &ftrace_commands);
3327 out_unlock:
3328 mutex_unlock(&ftrace_cmd_mutex);
3329
3330 return ret;
3331}
3332
3333/*
3334 * Currently we only unregister ftrace commands from __init, so mark
3335 * this __init too.
3336 */
3337__init int unregister_ftrace_command(struct ftrace_func_command *cmd)
3338{
3339 struct ftrace_func_command *p, *n;
3340 int ret = -ENODEV;
3341
3342 mutex_lock(&ftrace_cmd_mutex);
3343 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3344 if (strcmp(cmd->name, p->name) == 0) {
3345 ret = 0;
3346 list_del_init(&p->list);
3347 goto out_unlock;
3348 }
3349 }
3350 out_unlock:
3351 mutex_unlock(&ftrace_cmd_mutex);
3352
3353 return ret;
3354}
3355
3356static int ftrace_process_regex(struct ftrace_hash *hash,
3357 char *buff, int len, int enable)
3358{
3359 char *func, *command, *next = buff;
3360 struct ftrace_func_command *p;
3361 int ret = -EINVAL;
3362
3363 func = strsep(&next, ":");
3364
3365 if (!next) {
3366 ret = ftrace_match_records(hash, func, len);
3367 if (!ret)
3368 ret = -EINVAL;
3369 if (ret < 0)
3370 return ret;
3371 return 0;
3372 }
3373
3374 /* command found */
3375
3376 command = strsep(&next, ":");
3377
3378 mutex_lock(&ftrace_cmd_mutex);
3379 list_for_each_entry(p, &ftrace_commands, list) {
3380 if (strcmp(p->name, command) == 0) {
3381 ret = p->func(hash, func, command, next, enable);
3382 goto out_unlock;
3383 }
3384 }
3385 out_unlock:
3386 mutex_unlock(&ftrace_cmd_mutex);
3387
3388 return ret;
3389}
3390
3391static ssize_t
3392ftrace_regex_write(struct file *file, const char __user *ubuf,
3393 size_t cnt, loff_t *ppos, int enable)
3394{
3395 struct ftrace_iterator *iter;
3396 struct trace_parser *parser;
3397 ssize_t ret, read;
3398
3399 if (!cnt)
3400 return 0;
3401
3402 if (file->f_mode & FMODE_READ) {
3403 struct seq_file *m = file->private_data;
3404 iter = m->private;
3405 } else
3406 iter = file->private_data;
3407
3408 if (unlikely(ftrace_disabled))
3409 return -ENODEV;
3410
3411 /* iter->hash is a local copy, so we don't need regex_lock */
3412
3413 parser = &iter->parser;
3414 read = trace_get_user(parser, ubuf, cnt, ppos);
3415
3416 if (read >= 0 && trace_parser_loaded(parser) &&
3417 !trace_parser_cont(parser)) {
3418 ret = ftrace_process_regex(iter->hash, parser->buffer,
3419 parser->idx, enable);
3420 trace_parser_clear(parser);
3421 if (ret < 0)
3422 goto out;
3423 }
3424
3425 ret = read;
3426 out:
3427 return ret;
3428}
3429
3430ssize_t
3431ftrace_filter_write(struct file *file, const char __user *ubuf,
3432 size_t cnt, loff_t *ppos)
3433{
3434 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
3435}
3436
3437ssize_t
3438ftrace_notrace_write(struct file *file, const char __user *ubuf,
3439 size_t cnt, loff_t *ppos)
3440{
3441 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
3442}
3443
3444static int
3445ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
3446{
3447 struct ftrace_func_entry *entry;
3448
3449 if (!ftrace_location(ip))
3450 return -EINVAL;
3451
3452 if (remove) {
3453 entry = ftrace_lookup_ip(hash, ip);
3454 if (!entry)
3455 return -ENOENT;
3456 free_hash_entry(hash, entry);
3457 return 0;
3458 }
3459
3460 return add_hash_entry(hash, ip);
3461}
3462
3463static void ftrace_ops_update_code(struct ftrace_ops *ops)
3464{
3465 if (ops->flags & FTRACE_OPS_FL_ENABLED && ftrace_enabled)
3466 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3467}
3468
3469static int
3470ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
3471 unsigned long ip, int remove, int reset, int enable)
3472{
3473 struct ftrace_hash **orig_hash;
3474 struct ftrace_hash *hash;
3475 int ret;
3476
3477 /* All global ops uses the global ops filters */
3478 if (ops->flags & FTRACE_OPS_FL_GLOBAL)
3479 ops = &global_ops;
3480
3481 if (unlikely(ftrace_disabled))
3482 return -ENODEV;
3483
3484 mutex_lock(&ops->regex_lock);
3485
3486 if (enable)
3487 orig_hash = &ops->filter_hash;
3488 else
3489 orig_hash = &ops->notrace_hash;
3490
3491 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3492 if (!hash) {
3493 ret = -ENOMEM;
3494 goto out_regex_unlock;
3495 }
3496
3497 if (reset)
3498 ftrace_filter_reset(hash);
3499 if (buf && !ftrace_match_records(hash, buf, len)) {
3500 ret = -EINVAL;
3501 goto out_regex_unlock;
3502 }
3503 if (ip) {
3504 ret = ftrace_match_addr(hash, ip, remove);
3505 if (ret < 0)
3506 goto out_regex_unlock;
3507 }
3508
3509 mutex_lock(&ftrace_lock);
3510 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3511 if (!ret)
3512 ftrace_ops_update_code(ops);
3513
3514 mutex_unlock(&ftrace_lock);
3515
3516 out_regex_unlock:
3517 mutex_unlock(&ops->regex_lock);
3518
3519 free_ftrace_hash(hash);
3520 return ret;
3521}
3522
3523static int
3524ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
3525 int reset, int enable)
3526{
3527 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
3528}
3529
3530/**
3531 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
3532 * @ops - the ops to set the filter with
3533 * @ip - the address to add to or remove from the filter.
3534 * @remove - non zero to remove the ip from the filter
3535 * @reset - non zero to reset all filters before applying this filter.
3536 *
3537 * Filters denote which functions should be enabled when tracing is enabled
3538 * If @ip is NULL, it failes to update filter.
3539 */
3540int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
3541 int remove, int reset)
3542{
3543 ftrace_ops_init(ops);
3544 return ftrace_set_addr(ops, ip, remove, reset, 1);
3545}
3546EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
3547
3548static int
3549ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3550 int reset, int enable)
3551{
3552 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
3553}
3554
3555/**
3556 * ftrace_set_filter - set a function to filter on in ftrace
3557 * @ops - the ops to set the filter with
3558 * @buf - the string that holds the function filter text.
3559 * @len - the length of the string.
3560 * @reset - non zero to reset all filters before applying this filter.
3561 *
3562 * Filters denote which functions should be enabled when tracing is enabled.
3563 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3564 */
3565int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
3566 int len, int reset)
3567{
3568 ftrace_ops_init(ops);
3569 return ftrace_set_regex(ops, buf, len, reset, 1);
3570}
3571EXPORT_SYMBOL_GPL(ftrace_set_filter);
3572
3573/**
3574 * ftrace_set_notrace - set a function to not trace in ftrace
3575 * @ops - the ops to set the notrace filter with
3576 * @buf - the string that holds the function notrace text.
3577 * @len - the length of the string.
3578 * @reset - non zero to reset all filters before applying this filter.
3579 *
3580 * Notrace Filters denote which functions should not be enabled when tracing
3581 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3582 * for tracing.
3583 */
3584int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
3585 int len, int reset)
3586{
3587 ftrace_ops_init(ops);
3588 return ftrace_set_regex(ops, buf, len, reset, 0);
3589}
3590EXPORT_SYMBOL_GPL(ftrace_set_notrace);
3591/**
3592 * ftrace_set_filter - set a function to filter on in ftrace
3593 * @ops - the ops to set the filter with
3594 * @buf - the string that holds the function filter text.
3595 * @len - the length of the string.
3596 * @reset - non zero to reset all filters before applying this filter.
3597 *
3598 * Filters denote which functions should be enabled when tracing is enabled.
3599 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3600 */
3601void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
3602{
3603 ftrace_set_regex(&global_ops, buf, len, reset, 1);
3604}
3605EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
3606
3607/**
3608 * ftrace_set_notrace - set a function to not trace in ftrace
3609 * @ops - the ops to set the notrace filter with
3610 * @buf - the string that holds the function notrace text.
3611 * @len - the length of the string.
3612 * @reset - non zero to reset all filters before applying this filter.
3613 *
3614 * Notrace Filters denote which functions should not be enabled when tracing
3615 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3616 * for tracing.
3617 */
3618void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
3619{
3620 ftrace_set_regex(&global_ops, buf, len, reset, 0);
3621}
3622EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3623
3624/*
3625 * command line interface to allow users to set filters on boot up.
3626 */
3627#define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
3628static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3629static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3630
3631/* Used by function selftest to not test if filter is set */
3632bool ftrace_filter_param __initdata;
3633
3634static int __init set_ftrace_notrace(char *str)
3635{
3636 ftrace_filter_param = true;
3637 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3638 return 1;
3639}
3640__setup("ftrace_notrace=", set_ftrace_notrace);
3641
3642static int __init set_ftrace_filter(char *str)
3643{
3644 ftrace_filter_param = true;
3645 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3646 return 1;
3647}
3648__setup("ftrace_filter=", set_ftrace_filter);
3649
3650#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3651static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3652static int ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer);
3653
3654static int __init set_graph_function(char *str)
3655{
3656 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3657 return 1;
3658}
3659__setup("ftrace_graph_filter=", set_graph_function);
3660
3661static void __init set_ftrace_early_graph(char *buf)
3662{
3663 int ret;
3664 char *func;
3665
3666 while (buf) {
3667 func = strsep(&buf, ",");
3668 /* we allow only one expression at a time */
3669 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3670 FTRACE_GRAPH_MAX_FUNCS, func);
3671 if (ret)
3672 printk(KERN_DEBUG "ftrace: function %s not "
3673 "traceable\n", func);
3674 }
3675}
3676#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3677
3678void __init
3679ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3680{
3681 char *func;
3682
3683 ftrace_ops_init(ops);
3684
3685 while (buf) {
3686 func = strsep(&buf, ",");
3687 ftrace_set_regex(ops, func, strlen(func), 0, enable);
3688 }
3689}
3690
3691static void __init set_ftrace_early_filters(void)
3692{
3693 if (ftrace_filter_buf[0])
3694 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
3695 if (ftrace_notrace_buf[0])
3696 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
3697#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3698 if (ftrace_graph_buf[0])
3699 set_ftrace_early_graph(ftrace_graph_buf);
3700#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3701}
3702
3703int ftrace_regex_release(struct inode *inode, struct file *file)
3704{
3705 struct seq_file *m = (struct seq_file *)file->private_data;
3706 struct ftrace_iterator *iter;
3707 struct ftrace_hash **orig_hash;
3708 struct trace_parser *parser;
3709 int filter_hash;
3710 int ret;
3711
3712 if (file->f_mode & FMODE_READ) {
3713 iter = m->private;
3714 seq_release(inode, file);
3715 } else
3716 iter = file->private_data;
3717
3718 parser = &iter->parser;
3719 if (trace_parser_loaded(parser)) {
3720 parser->buffer[parser->idx] = 0;
3721 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3722 }
3723
3724 trace_parser_put(parser);
3725
3726 mutex_lock(&iter->ops->regex_lock);
3727
3728 if (file->f_mode & FMODE_WRITE) {
3729 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3730
3731 if (filter_hash)
3732 orig_hash = &iter->ops->filter_hash;
3733 else
3734 orig_hash = &iter->ops->notrace_hash;
3735
3736 mutex_lock(&ftrace_lock);
3737 ret = ftrace_hash_move(iter->ops, filter_hash,
3738 orig_hash, iter->hash);
3739 if (!ret)
3740 ftrace_ops_update_code(iter->ops);
3741
3742 mutex_unlock(&ftrace_lock);
3743 }
3744
3745 mutex_unlock(&iter->ops->regex_lock);
3746 free_ftrace_hash(iter->hash);
3747 kfree(iter);
3748
3749 return 0;
3750}
3751
3752static const struct file_operations ftrace_avail_fops = {
3753 .open = ftrace_avail_open,
3754 .read = seq_read,
3755 .llseek = seq_lseek,
3756 .release = seq_release_private,
3757};
3758
3759static const struct file_operations ftrace_enabled_fops = {
3760 .open = ftrace_enabled_open,
3761 .read = seq_read,
3762 .llseek = seq_lseek,
3763 .release = seq_release_private,
3764};
3765
3766static const struct file_operations ftrace_filter_fops = {
3767 .open = ftrace_filter_open,
3768 .read = seq_read,
3769 .write = ftrace_filter_write,
3770 .llseek = ftrace_filter_lseek,
3771 .release = ftrace_regex_release,
3772};
3773
3774static const struct file_operations ftrace_notrace_fops = {
3775 .open = ftrace_notrace_open,
3776 .read = seq_read,
3777 .write = ftrace_notrace_write,
3778 .llseek = ftrace_filter_lseek,
3779 .release = ftrace_regex_release,
3780};
3781
3782#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3783
3784static DEFINE_MUTEX(graph_lock);
3785
3786int ftrace_graph_count;
3787int ftrace_graph_notrace_count;
3788unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3789unsigned long ftrace_graph_notrace_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3790
3791struct ftrace_graph_data {
3792 unsigned long *table;
3793 size_t size;
3794 int *count;
3795 const struct seq_operations *seq_ops;
3796};
3797
3798static void *
3799__g_next(struct seq_file *m, loff_t *pos)
3800{
3801 struct ftrace_graph_data *fgd = m->private;
3802
3803 if (*pos >= *fgd->count)
3804 return NULL;
3805 return &fgd->table[*pos];
3806}
3807
3808static void *
3809g_next(struct seq_file *m, void *v, loff_t *pos)
3810{
3811 (*pos)++;
3812 return __g_next(m, pos);
3813}
3814
3815static void *g_start(struct seq_file *m, loff_t *pos)
3816{
3817 struct ftrace_graph_data *fgd = m->private;
3818
3819 mutex_lock(&graph_lock);
3820
3821 /* Nothing, tell g_show to print all functions are enabled */
3822 if (!*fgd->count && !*pos)
3823 return (void *)1;
3824
3825 return __g_next(m, pos);
3826}
3827
3828static void g_stop(struct seq_file *m, void *p)
3829{
3830 mutex_unlock(&graph_lock);
3831}
3832
3833static int g_show(struct seq_file *m, void *v)
3834{
3835 unsigned long *ptr = v;
3836
3837 if (!ptr)
3838 return 0;
3839
3840 if (ptr == (unsigned long *)1) {
3841 seq_printf(m, "#### all functions enabled ####\n");
3842 return 0;
3843 }
3844
3845 seq_printf(m, "%ps\n", (void *)*ptr);
3846
3847 return 0;
3848}
3849
3850static const struct seq_operations ftrace_graph_seq_ops = {
3851 .start = g_start,
3852 .next = g_next,
3853 .stop = g_stop,
3854 .show = g_show,
3855};
3856
3857static int
3858__ftrace_graph_open(struct inode *inode, struct file *file,
3859 struct ftrace_graph_data *fgd)
3860{
3861 int ret = 0;
3862
3863 mutex_lock(&graph_lock);
3864 if ((file->f_mode & FMODE_WRITE) &&
3865 (file->f_flags & O_TRUNC)) {
3866 *fgd->count = 0;
3867 memset(fgd->table, 0, fgd->size * sizeof(*fgd->table));
3868 }
3869 mutex_unlock(&graph_lock);
3870
3871 if (file->f_mode & FMODE_READ) {
3872 ret = seq_open(file, fgd->seq_ops);
3873 if (!ret) {
3874 struct seq_file *m = file->private_data;
3875 m->private = fgd;
3876 }
3877 } else
3878 file->private_data = fgd;
3879
3880 return ret;
3881}
3882
3883static int
3884ftrace_graph_open(struct inode *inode, struct file *file)
3885{
3886 struct ftrace_graph_data *fgd;
3887
3888 if (unlikely(ftrace_disabled))
3889 return -ENODEV;
3890
3891 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
3892 if (fgd == NULL)
3893 return -ENOMEM;
3894
3895 fgd->table = ftrace_graph_funcs;
3896 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
3897 fgd->count = &ftrace_graph_count;
3898 fgd->seq_ops = &ftrace_graph_seq_ops;
3899
3900 return __ftrace_graph_open(inode, file, fgd);
3901}
3902
3903static int
3904ftrace_graph_notrace_open(struct inode *inode, struct file *file)
3905{
3906 struct ftrace_graph_data *fgd;
3907
3908 if (unlikely(ftrace_disabled))
3909 return -ENODEV;
3910
3911 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
3912 if (fgd == NULL)
3913 return -ENOMEM;
3914
3915 fgd->table = ftrace_graph_notrace_funcs;
3916 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
3917 fgd->count = &ftrace_graph_notrace_count;
3918 fgd->seq_ops = &ftrace_graph_seq_ops;
3919
3920 return __ftrace_graph_open(inode, file, fgd);
3921}
3922
3923static int
3924ftrace_graph_release(struct inode *inode, struct file *file)
3925{
3926 if (file->f_mode & FMODE_READ) {
3927 struct seq_file *m = file->private_data;
3928
3929 kfree(m->private);
3930 seq_release(inode, file);
3931 } else {
3932 kfree(file->private_data);
3933 }
3934
3935 return 0;
3936}
3937
3938static int
3939ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer)
3940{
3941 struct dyn_ftrace *rec;
3942 struct ftrace_page *pg;
3943 int search_len;
3944 int fail = 1;
3945 int type, not;
3946 char *search;
3947 bool exists;
3948 int i;
3949
3950 /* decode regex */
3951 type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
3952 if (!not && *idx >= size)
3953 return -EBUSY;
3954
3955 search_len = strlen(search);
3956
3957 mutex_lock(&ftrace_lock);
3958
3959 if (unlikely(ftrace_disabled)) {
3960 mutex_unlock(&ftrace_lock);
3961 return -ENODEV;
3962 }
3963
3964 do_for_each_ftrace_rec(pg, rec) {
3965
3966 if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3967 /* if it is in the array */
3968 exists = false;
3969 for (i = 0; i < *idx; i++) {
3970 if (array[i] == rec->ip) {
3971 exists = true;
3972 break;
3973 }
3974 }
3975
3976 if (!not) {
3977 fail = 0;
3978 if (!exists) {
3979 array[(*idx)++] = rec->ip;
3980 if (*idx >= size)
3981 goto out;
3982 }
3983 } else {
3984 if (exists) {
3985 array[i] = array[--(*idx)];
3986 array[*idx] = 0;
3987 fail = 0;
3988 }
3989 }
3990 }
3991 } while_for_each_ftrace_rec();
3992out:
3993 mutex_unlock(&ftrace_lock);
3994
3995 if (fail)
3996 return -EINVAL;
3997
3998 return 0;
3999}
4000
4001static ssize_t
4002ftrace_graph_write(struct file *file, const char __user *ubuf,
4003 size_t cnt, loff_t *ppos)
4004{
4005 struct trace_parser parser;
4006 ssize_t read, ret = 0;
4007 struct ftrace_graph_data *fgd = file->private_data;
4008
4009 if (!cnt)
4010 return 0;
4011
4012 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX))
4013 return -ENOMEM;
4014
4015 read = trace_get_user(&parser, ubuf, cnt, ppos);
4016
4017 if (read >= 0 && trace_parser_loaded((&parser))) {
4018 parser.buffer[parser.idx] = 0;
4019
4020 mutex_lock(&graph_lock);
4021
4022 /* we allow only one expression at a time */
4023 ret = ftrace_set_func(fgd->table, fgd->count, fgd->size,
4024 parser.buffer);
4025
4026 mutex_unlock(&graph_lock);
4027 }
4028
4029 if (!ret)
4030 ret = read;
4031
4032 trace_parser_put(&parser);
4033
4034 return ret;
4035}
4036
4037static const struct file_operations ftrace_graph_fops = {
4038 .open = ftrace_graph_open,
4039 .read = seq_read,
4040 .write = ftrace_graph_write,
4041 .llseek = ftrace_filter_lseek,
4042 .release = ftrace_graph_release,
4043};
4044
4045static const struct file_operations ftrace_graph_notrace_fops = {
4046 .open = ftrace_graph_notrace_open,
4047 .read = seq_read,
4048 .write = ftrace_graph_write,
4049 .llseek = ftrace_filter_lseek,
4050 .release = ftrace_graph_release,
4051};
4052#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4053
4054static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
4055{
4056
4057 trace_create_file("available_filter_functions", 0444,
4058 d_tracer, NULL, &ftrace_avail_fops);
4059
4060 trace_create_file("enabled_functions", 0444,
4061 d_tracer, NULL, &ftrace_enabled_fops);
4062
4063 trace_create_file("set_ftrace_filter", 0644, d_tracer,
4064 NULL, &ftrace_filter_fops);
4065
4066 trace_create_file("set_ftrace_notrace", 0644, d_tracer,
4067 NULL, &ftrace_notrace_fops);
4068
4069#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4070 trace_create_file("set_graph_function", 0444, d_tracer,
4071 NULL,
4072 &ftrace_graph_fops);
4073 trace_create_file("set_graph_notrace", 0444, d_tracer,
4074 NULL,
4075 &ftrace_graph_notrace_fops);
4076#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4077
4078 return 0;
4079}
4080
4081static int ftrace_cmp_ips(const void *a, const void *b)
4082{
4083 const unsigned long *ipa = a;
4084 const unsigned long *ipb = b;
4085
4086 if (*ipa > *ipb)
4087 return 1;
4088 if (*ipa < *ipb)
4089 return -1;
4090 return 0;
4091}
4092
4093static void ftrace_swap_ips(void *a, void *b, int size)
4094{
4095 unsigned long *ipa = a;
4096 unsigned long *ipb = b;
4097 unsigned long t;
4098
4099 t = *ipa;
4100 *ipa = *ipb;
4101 *ipb = t;
4102}
4103
4104static int ftrace_process_locs(struct module *mod,
4105 unsigned long *start,
4106 unsigned long *end)
4107{
4108 struct ftrace_page *start_pg;
4109 struct ftrace_page *pg;
4110 struct dyn_ftrace *rec;
4111 unsigned long count;
4112 unsigned long *p;
4113 unsigned long addr;
4114 unsigned long flags = 0; /* Shut up gcc */
4115 int ret = -ENOMEM;
4116
4117 count = end - start;
4118
4119 if (!count)
4120 return 0;
4121
4122 sort(start, count, sizeof(*start),
4123 ftrace_cmp_ips, ftrace_swap_ips);
4124
4125 start_pg = ftrace_allocate_pages(count);
4126 if (!start_pg)
4127 return -ENOMEM;
4128
4129 mutex_lock(&ftrace_lock);
4130
4131 /*
4132 * Core and each module needs their own pages, as
4133 * modules will free them when they are removed.
4134 * Force a new page to be allocated for modules.
4135 */
4136 if (!mod) {
4137 WARN_ON(ftrace_pages || ftrace_pages_start);
4138 /* First initialization */
4139 ftrace_pages = ftrace_pages_start = start_pg;
4140 } else {
4141 if (!ftrace_pages)
4142 goto out;
4143
4144 if (WARN_ON(ftrace_pages->next)) {
4145 /* Hmm, we have free pages? */
4146 while (ftrace_pages->next)
4147 ftrace_pages = ftrace_pages->next;
4148 }
4149
4150 ftrace_pages->next = start_pg;
4151 }
4152
4153 p = start;
4154 pg = start_pg;
4155 while (p < end) {
4156 addr = ftrace_call_adjust(*p++);
4157 /*
4158 * Some architecture linkers will pad between
4159 * the different mcount_loc sections of different
4160 * object files to satisfy alignments.
4161 * Skip any NULL pointers.
4162 */
4163 if (!addr)
4164 continue;
4165
4166 if (pg->index == pg->size) {
4167 /* We should have allocated enough */
4168 if (WARN_ON(!pg->next))
4169 break;
4170 pg = pg->next;
4171 }
4172
4173 rec = &pg->records[pg->index++];
4174 rec->ip = addr;
4175 }
4176
4177 /* We should have used all pages */
4178 WARN_ON(pg->next);
4179
4180 /* Assign the last page to ftrace_pages */
4181 ftrace_pages = pg;
4182
4183 /* These new locations need to be initialized */
4184 ftrace_new_pgs = start_pg;
4185
4186 /*
4187 * We only need to disable interrupts on start up
4188 * because we are modifying code that an interrupt
4189 * may execute, and the modification is not atomic.
4190 * But for modules, nothing runs the code we modify
4191 * until we are finished with it, and there's no
4192 * reason to cause large interrupt latencies while we do it.
4193 */
4194 if (!mod)
4195 local_irq_save(flags);
4196 ftrace_update_code(mod);
4197 if (!mod)
4198 local_irq_restore(flags);
4199 ret = 0;
4200 out:
4201 mutex_unlock(&ftrace_lock);
4202
4203 return ret;
4204}
4205
4206#ifdef CONFIG_MODULES
4207
4208#define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4209
4210void ftrace_release_mod(struct module *mod)
4211{
4212 struct dyn_ftrace *rec;
4213 struct ftrace_page **last_pg;
4214 struct ftrace_page *pg;
4215 int order;
4216
4217 mutex_lock(&ftrace_lock);
4218
4219 if (ftrace_disabled)
4220 goto out_unlock;
4221
4222 /*
4223 * Each module has its own ftrace_pages, remove
4224 * them from the list.
4225 */
4226 last_pg = &ftrace_pages_start;
4227 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4228 rec = &pg->records[0];
4229 if (within_module_core(rec->ip, mod)) {
4230 /*
4231 * As core pages are first, the first
4232 * page should never be a module page.
4233 */
4234 if (WARN_ON(pg == ftrace_pages_start))
4235 goto out_unlock;
4236
4237 /* Check if we are deleting the last page */
4238 if (pg == ftrace_pages)
4239 ftrace_pages = next_to_ftrace_page(last_pg);
4240
4241 *last_pg = pg->next;
4242 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4243 free_pages((unsigned long)pg->records, order);
4244 kfree(pg);
4245 } else
4246 last_pg = &pg->next;
4247 }
4248 out_unlock:
4249 mutex_unlock(&ftrace_lock);
4250}
4251
4252static void ftrace_init_module(struct module *mod,
4253 unsigned long *start, unsigned long *end)
4254{
4255 if (ftrace_disabled || start == end)
4256 return;
4257 ftrace_process_locs(mod, start, end);
4258}
4259
4260static int ftrace_module_notify_enter(struct notifier_block *self,
4261 unsigned long val, void *data)
4262{
4263 struct module *mod = data;
4264
4265 if (val == MODULE_STATE_COMING)
4266 ftrace_init_module(mod, mod->ftrace_callsites,
4267 mod->ftrace_callsites +
4268 mod->num_ftrace_callsites);
4269 return 0;
4270}
4271
4272static int ftrace_module_notify_exit(struct notifier_block *self,
4273 unsigned long val, void *data)
4274{
4275 struct module *mod = data;
4276
4277 if (val == MODULE_STATE_GOING)
4278 ftrace_release_mod(mod);
4279
4280 return 0;
4281}
4282#else
4283static int ftrace_module_notify_enter(struct notifier_block *self,
4284 unsigned long val, void *data)
4285{
4286 return 0;
4287}
4288static int ftrace_module_notify_exit(struct notifier_block *self,
4289 unsigned long val, void *data)
4290{
4291 return 0;
4292}
4293#endif /* CONFIG_MODULES */
4294
4295struct notifier_block ftrace_module_enter_nb = {
4296 .notifier_call = ftrace_module_notify_enter,
4297 .priority = INT_MAX, /* Run before anything that can use kprobes */
4298};
4299
4300struct notifier_block ftrace_module_exit_nb = {
4301 .notifier_call = ftrace_module_notify_exit,
4302 .priority = INT_MIN, /* Run after anything that can remove kprobes */
4303};
4304
4305extern unsigned long __start_mcount_loc[];
4306extern unsigned long __stop_mcount_loc[];
4307
4308void __init ftrace_init(void)
4309{
4310 unsigned long count, addr, flags;
4311 int ret;
4312
4313 /* Keep the ftrace pointer to the stub */
4314 addr = (unsigned long)ftrace_stub;
4315
4316 local_irq_save(flags);
4317 ftrace_dyn_arch_init(&addr);
4318 local_irq_restore(flags);
4319
4320 /* ftrace_dyn_arch_init places the return code in addr */
4321 if (addr)
4322 goto failed;
4323
4324 count = __stop_mcount_loc - __start_mcount_loc;
4325
4326 ret = ftrace_dyn_table_alloc(count);
4327 if (ret)
4328 goto failed;
4329
4330 last_ftrace_enabled = ftrace_enabled = 1;
4331
4332 ret = ftrace_process_locs(NULL,
4333 __start_mcount_loc,
4334 __stop_mcount_loc);
4335
4336 ret = register_module_notifier(&ftrace_module_enter_nb);
4337 if (ret)
4338 pr_warning("Failed to register trace ftrace module enter notifier\n");
4339
4340 ret = register_module_notifier(&ftrace_module_exit_nb);
4341 if (ret)
4342 pr_warning("Failed to register trace ftrace module exit notifier\n");
4343
4344 set_ftrace_early_filters();
4345
4346 return;
4347 failed:
4348 ftrace_disabled = 1;
4349}
4350
4351#else
4352
4353static struct ftrace_ops global_ops = {
4354 .func = ftrace_stub,
4355 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4356 INIT_REGEX_LOCK(global_ops)
4357};
4358
4359static int __init ftrace_nodyn_init(void)
4360{
4361 ftrace_enabled = 1;
4362 return 0;
4363}
4364core_initcall(ftrace_nodyn_init);
4365
4366static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
4367static inline void ftrace_startup_enable(int command) { }
4368/* Keep as macros so we do not need to define the commands */
4369# define ftrace_startup(ops, command) \
4370 ({ \
4371 int ___ret = __register_ftrace_function(ops); \
4372 if (!___ret) \
4373 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
4374 ___ret; \
4375 })
4376# define ftrace_shutdown(ops, command) __unregister_ftrace_function(ops)
4377
4378# define ftrace_startup_sysctl() do { } while (0)
4379# define ftrace_shutdown_sysctl() do { } while (0)
4380
4381static inline int
4382ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
4383{
4384 return 1;
4385}
4386
4387#endif /* CONFIG_DYNAMIC_FTRACE */
4388
4389static void
4390ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip,
4391 struct ftrace_ops *op, struct pt_regs *regs)
4392{
4393 if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
4394 return;
4395
4396 /*
4397 * Some of the ops may be dynamically allocated,
4398 * they must be freed after a synchronize_sched().
4399 */
4400 preempt_disable_notrace();
4401 trace_recursion_set(TRACE_CONTROL_BIT);
4402
4403 /*
4404 * Control funcs (perf) uses RCU. Only trace if
4405 * RCU is currently active.
4406 */
4407 if (!rcu_is_watching())
4408 goto out;
4409
4410 do_for_each_ftrace_op(op, ftrace_control_list) {
4411 if (!(op->flags & FTRACE_OPS_FL_STUB) &&
4412 !ftrace_function_local_disabled(op) &&
4413 ftrace_ops_test(op, ip, regs))
4414 op->func(ip, parent_ip, op, regs);
4415 } while_for_each_ftrace_op(op);
4416 out:
4417 trace_recursion_clear(TRACE_CONTROL_BIT);
4418 preempt_enable_notrace();
4419}
4420
4421static struct ftrace_ops control_ops = {
4422 .func = ftrace_ops_control_func,
4423 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4424 INIT_REGEX_LOCK(control_ops)
4425};
4426
4427static inline void
4428__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4429 struct ftrace_ops *ignored, struct pt_regs *regs)
4430{
4431 struct ftrace_ops *op;
4432 int bit;
4433
4434 if (function_trace_stop)
4435 return;
4436
4437 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
4438 if (bit < 0)
4439 return;
4440
4441 /*
4442 * Some of the ops may be dynamically allocated,
4443 * they must be freed after a synchronize_sched().
4444 */
4445 preempt_disable_notrace();
4446 do_for_each_ftrace_op(op, ftrace_ops_list) {
4447 if (ftrace_ops_test(op, ip, regs))
4448 op->func(ip, parent_ip, op, regs);
4449 } while_for_each_ftrace_op(op);
4450 preempt_enable_notrace();
4451 trace_clear_recursion(bit);
4452}
4453
4454/*
4455 * Some archs only support passing ip and parent_ip. Even though
4456 * the list function ignores the op parameter, we do not want any
4457 * C side effects, where a function is called without the caller
4458 * sending a third parameter.
4459 * Archs are to support both the regs and ftrace_ops at the same time.
4460 * If they support ftrace_ops, it is assumed they support regs.
4461 * If call backs want to use regs, they must either check for regs
4462 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
4463 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
4464 * An architecture can pass partial regs with ftrace_ops and still
4465 * set the ARCH_SUPPORT_FTARCE_OPS.
4466 */
4467#if ARCH_SUPPORTS_FTRACE_OPS
4468static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4469 struct ftrace_ops *op, struct pt_regs *regs)
4470{
4471 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
4472}
4473#else
4474static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
4475{
4476 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
4477}
4478#endif
4479
4480static void clear_ftrace_swapper(void)
4481{
4482 struct task_struct *p;
4483 int cpu;
4484
4485 get_online_cpus();
4486 for_each_online_cpu(cpu) {
4487 p = idle_task(cpu);
4488 clear_tsk_trace_trace(p);
4489 }
4490 put_online_cpus();
4491}
4492
4493static void set_ftrace_swapper(void)
4494{
4495 struct task_struct *p;
4496 int cpu;
4497
4498 get_online_cpus();
4499 for_each_online_cpu(cpu) {
4500 p = idle_task(cpu);
4501 set_tsk_trace_trace(p);
4502 }
4503 put_online_cpus();
4504}
4505
4506static void clear_ftrace_pid(struct pid *pid)
4507{
4508 struct task_struct *p;
4509
4510 rcu_read_lock();
4511 do_each_pid_task(pid, PIDTYPE_PID, p) {
4512 clear_tsk_trace_trace(p);
4513 } while_each_pid_task(pid, PIDTYPE_PID, p);
4514 rcu_read_unlock();
4515
4516 put_pid(pid);
4517}
4518
4519static void set_ftrace_pid(struct pid *pid)
4520{
4521 struct task_struct *p;
4522
4523 rcu_read_lock();
4524 do_each_pid_task(pid, PIDTYPE_PID, p) {
4525 set_tsk_trace_trace(p);
4526 } while_each_pid_task(pid, PIDTYPE_PID, p);
4527 rcu_read_unlock();
4528}
4529
4530static void clear_ftrace_pid_task(struct pid *pid)
4531{
4532 if (pid == ftrace_swapper_pid)
4533 clear_ftrace_swapper();
4534 else
4535 clear_ftrace_pid(pid);
4536}
4537
4538static void set_ftrace_pid_task(struct pid *pid)
4539{
4540 if (pid == ftrace_swapper_pid)
4541 set_ftrace_swapper();
4542 else
4543 set_ftrace_pid(pid);
4544}
4545
4546static int ftrace_pid_add(int p)
4547{
4548 struct pid *pid;
4549 struct ftrace_pid *fpid;
4550 int ret = -EINVAL;
4551
4552 mutex_lock(&ftrace_lock);
4553
4554 if (!p)
4555 pid = ftrace_swapper_pid;
4556 else
4557 pid = find_get_pid(p);
4558
4559 if (!pid)
4560 goto out;
4561
4562 ret = 0;
4563
4564 list_for_each_entry(fpid, &ftrace_pids, list)
4565 if (fpid->pid == pid)
4566 goto out_put;
4567
4568 ret = -ENOMEM;
4569
4570 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
4571 if (!fpid)
4572 goto out_put;
4573
4574 list_add(&fpid->list, &ftrace_pids);
4575 fpid->pid = pid;
4576
4577 set_ftrace_pid_task(pid);
4578
4579 ftrace_update_pid_func();
4580 ftrace_startup_enable(0);
4581
4582 mutex_unlock(&ftrace_lock);
4583 return 0;
4584
4585out_put:
4586 if (pid != ftrace_swapper_pid)
4587 put_pid(pid);
4588
4589out:
4590 mutex_unlock(&ftrace_lock);
4591 return ret;
4592}
4593
4594static void ftrace_pid_reset(void)
4595{
4596 struct ftrace_pid *fpid, *safe;
4597
4598 mutex_lock(&ftrace_lock);
4599 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
4600 struct pid *pid = fpid->pid;
4601
4602 clear_ftrace_pid_task(pid);
4603
4604 list_del(&fpid->list);
4605 kfree(fpid);
4606 }
4607
4608 ftrace_update_pid_func();
4609 ftrace_startup_enable(0);
4610
4611 mutex_unlock(&ftrace_lock);
4612}
4613
4614static void *fpid_start(struct seq_file *m, loff_t *pos)
4615{
4616 mutex_lock(&ftrace_lock);
4617
4618 if (list_empty(&ftrace_pids) && (!*pos))
4619 return (void *) 1;
4620
4621 return seq_list_start(&ftrace_pids, *pos);
4622}
4623
4624static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
4625{
4626 if (v == (void *)1)
4627 return NULL;
4628
4629 return seq_list_next(v, &ftrace_pids, pos);
4630}
4631
4632static void fpid_stop(struct seq_file *m, void *p)
4633{
4634 mutex_unlock(&ftrace_lock);
4635}
4636
4637static int fpid_show(struct seq_file *m, void *v)
4638{
4639 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
4640
4641 if (v == (void *)1) {
4642 seq_printf(m, "no pid\n");
4643 return 0;
4644 }
4645
4646 if (fpid->pid == ftrace_swapper_pid)
4647 seq_printf(m, "swapper tasks\n");
4648 else
4649 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
4650
4651 return 0;
4652}
4653
4654static const struct seq_operations ftrace_pid_sops = {
4655 .start = fpid_start,
4656 .next = fpid_next,
4657 .stop = fpid_stop,
4658 .show = fpid_show,
4659};
4660
4661static int
4662ftrace_pid_open(struct inode *inode, struct file *file)
4663{
4664 int ret = 0;
4665
4666 if ((file->f_mode & FMODE_WRITE) &&
4667 (file->f_flags & O_TRUNC))
4668 ftrace_pid_reset();
4669
4670 if (file->f_mode & FMODE_READ)
4671 ret = seq_open(file, &ftrace_pid_sops);
4672
4673 return ret;
4674}
4675
4676static ssize_t
4677ftrace_pid_write(struct file *filp, const char __user *ubuf,
4678 size_t cnt, loff_t *ppos)
4679{
4680 char buf[64], *tmp;
4681 long val;
4682 int ret;
4683
4684 if (cnt >= sizeof(buf))
4685 return -EINVAL;
4686
4687 if (copy_from_user(&buf, ubuf, cnt))
4688 return -EFAULT;
4689
4690 buf[cnt] = 0;
4691
4692 /*
4693 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
4694 * to clean the filter quietly.
4695 */
4696 tmp = strstrip(buf);
4697 if (strlen(tmp) == 0)
4698 return 1;
4699
4700 ret = kstrtol(tmp, 10, &val);
4701 if (ret < 0)
4702 return ret;
4703
4704 ret = ftrace_pid_add(val);
4705
4706 return ret ? ret : cnt;
4707}
4708
4709static int
4710ftrace_pid_release(struct inode *inode, struct file *file)
4711{
4712 if (file->f_mode & FMODE_READ)
4713 seq_release(inode, file);
4714
4715 return 0;
4716}
4717
4718static const struct file_operations ftrace_pid_fops = {
4719 .open = ftrace_pid_open,
4720 .write = ftrace_pid_write,
4721 .read = seq_read,
4722 .llseek = ftrace_filter_lseek,
4723 .release = ftrace_pid_release,
4724};
4725
4726static __init int ftrace_init_debugfs(void)
4727{
4728 struct dentry *d_tracer;
4729
4730 d_tracer = tracing_init_dentry();
4731 if (!d_tracer)
4732 return 0;
4733
4734 ftrace_init_dyn_debugfs(d_tracer);
4735
4736 trace_create_file("set_ftrace_pid", 0644, d_tracer,
4737 NULL, &ftrace_pid_fops);
4738
4739 ftrace_profile_debugfs(d_tracer);
4740
4741 return 0;
4742}
4743fs_initcall(ftrace_init_debugfs);
4744
4745/**
4746 * ftrace_kill - kill ftrace
4747 *
4748 * This function should be used by panic code. It stops ftrace
4749 * but in a not so nice way. If you need to simply kill ftrace
4750 * from a non-atomic section, use ftrace_kill.
4751 */
4752void ftrace_kill(void)
4753{
4754 ftrace_disabled = 1;
4755 ftrace_enabled = 0;
4756 clear_ftrace_function();
4757}
4758
4759/**
4760 * Test if ftrace is dead or not.
4761 */
4762int ftrace_is_dead(void)
4763{
4764 return ftrace_disabled;
4765}
4766
4767/**
4768 * register_ftrace_function - register a function for profiling
4769 * @ops - ops structure that holds the function for profiling.
4770 *
4771 * Register a function to be called by all functions in the
4772 * kernel.
4773 *
4774 * Note: @ops->func and all the functions it calls must be labeled
4775 * with "notrace", otherwise it will go into a
4776 * recursive loop.
4777 */
4778int register_ftrace_function(struct ftrace_ops *ops)
4779{
4780 int ret = -1;
4781
4782 ftrace_ops_init(ops);
4783
4784 mutex_lock(&ftrace_lock);
4785
4786 ret = ftrace_startup(ops, 0);
4787
4788 mutex_unlock(&ftrace_lock);
4789
4790 return ret;
4791}
4792EXPORT_SYMBOL_GPL(register_ftrace_function);
4793
4794/**
4795 * unregister_ftrace_function - unregister a function for profiling.
4796 * @ops - ops structure that holds the function to unregister
4797 *
4798 * Unregister a function that was added to be called by ftrace profiling.
4799 */
4800int unregister_ftrace_function(struct ftrace_ops *ops)
4801{
4802 int ret;
4803
4804 mutex_lock(&ftrace_lock);
4805 ret = ftrace_shutdown(ops, 0);
4806 mutex_unlock(&ftrace_lock);
4807
4808 return ret;
4809}
4810EXPORT_SYMBOL_GPL(unregister_ftrace_function);
4811
4812int
4813ftrace_enable_sysctl(struct ctl_table *table, int write,
4814 void __user *buffer, size_t *lenp,
4815 loff_t *ppos)
4816{
4817 int ret = -ENODEV;
4818
4819 mutex_lock(&ftrace_lock);
4820
4821 if (unlikely(ftrace_disabled))
4822 goto out;
4823
4824 ret = proc_dointvec(table, write, buffer, lenp, ppos);
4825
4826 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
4827 goto out;
4828
4829 last_ftrace_enabled = !!ftrace_enabled;
4830
4831 if (ftrace_enabled) {
4832
4833 ftrace_startup_sysctl();
4834
4835 /* we are starting ftrace again */
4836 if (ftrace_ops_list != &ftrace_list_end)
4837 update_ftrace_function();
4838
4839 } else {
4840 /* stopping ftrace calls (just send to ftrace_stub) */
4841 ftrace_trace_function = ftrace_stub;
4842
4843 ftrace_shutdown_sysctl();
4844 }
4845
4846 out:
4847 mutex_unlock(&ftrace_lock);
4848 return ret;
4849}
4850
4851#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4852
4853static int ftrace_graph_active;
4854static struct notifier_block ftrace_suspend_notifier;
4855
4856int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
4857{
4858 return 0;
4859}
4860
4861/* The callbacks that hook a function */
4862trace_func_graph_ret_t ftrace_graph_return =
4863 (trace_func_graph_ret_t)ftrace_stub;
4864trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
4865
4866/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
4867static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
4868{
4869 int i;
4870 int ret = 0;
4871 unsigned long flags;
4872 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
4873 struct task_struct *g, *t;
4874
4875 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
4876 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
4877 * sizeof(struct ftrace_ret_stack),
4878 GFP_KERNEL);
4879 if (!ret_stack_list[i]) {
4880 start = 0;
4881 end = i;
4882 ret = -ENOMEM;
4883 goto free;
4884 }
4885 }
4886
4887 read_lock_irqsave(&tasklist_lock, flags);
4888 do_each_thread(g, t) {
4889 if (start == end) {
4890 ret = -EAGAIN;
4891 goto unlock;
4892 }
4893
4894 if (t->ret_stack == NULL) {
4895 atomic_set(&t->tracing_graph_pause, 0);
4896 atomic_set(&t->trace_overrun, 0);
4897 t->curr_ret_stack = -1;
4898 /* Make sure the tasks see the -1 first: */
4899 smp_wmb();
4900 t->ret_stack = ret_stack_list[start++];
4901 }
4902 } while_each_thread(g, t);
4903
4904unlock:
4905 read_unlock_irqrestore(&tasklist_lock, flags);
4906free:
4907 for (i = start; i < end; i++)
4908 kfree(ret_stack_list[i]);
4909 return ret;
4910}
4911
4912static void
4913ftrace_graph_probe_sched_switch(void *ignore,
4914 struct task_struct *prev, struct task_struct *next)
4915{
4916 unsigned long long timestamp;
4917 int index;
4918
4919 /*
4920 * Does the user want to count the time a function was asleep.
4921 * If so, do not update the time stamps.
4922 */
4923 if (trace_flags & TRACE_ITER_SLEEP_TIME)
4924 return;
4925
4926 timestamp = trace_clock_local();
4927
4928 prev->ftrace_timestamp = timestamp;
4929
4930 /* only process tasks that we timestamped */
4931 if (!next->ftrace_timestamp)
4932 return;
4933
4934 /*
4935 * Update all the counters in next to make up for the
4936 * time next was sleeping.
4937 */
4938 timestamp -= next->ftrace_timestamp;
4939
4940 for (index = next->curr_ret_stack; index >= 0; index--)
4941 next->ret_stack[index].calltime += timestamp;
4942}
4943
4944/* Allocate a return stack for each task */
4945static int start_graph_tracing(void)
4946{
4947 struct ftrace_ret_stack **ret_stack_list;
4948 int ret, cpu;
4949
4950 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
4951 sizeof(struct ftrace_ret_stack *),
4952 GFP_KERNEL);
4953
4954 if (!ret_stack_list)
4955 return -ENOMEM;
4956
4957 /* The cpu_boot init_task->ret_stack will never be freed */
4958 for_each_online_cpu(cpu) {
4959 if (!idle_task(cpu)->ret_stack)
4960 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4961 }
4962
4963 do {
4964 ret = alloc_retstack_tasklist(ret_stack_list);
4965 } while (ret == -EAGAIN);
4966
4967 if (!ret) {
4968 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4969 if (ret)
4970 pr_info("ftrace_graph: Couldn't activate tracepoint"
4971 " probe to kernel_sched_switch\n");
4972 }
4973
4974 kfree(ret_stack_list);
4975 return ret;
4976}
4977
4978/*
4979 * Hibernation protection.
4980 * The state of the current task is too much unstable during
4981 * suspend/restore to disk. We want to protect against that.
4982 */
4983static int
4984ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4985 void *unused)
4986{
4987 switch (state) {
4988 case PM_HIBERNATION_PREPARE:
4989 pause_graph_tracing();
4990 break;
4991
4992 case PM_POST_HIBERNATION:
4993 unpause_graph_tracing();
4994 break;
4995 }
4996 return NOTIFY_DONE;
4997}
4998
4999/* Just a place holder for function graph */
5000static struct ftrace_ops fgraph_ops __read_mostly = {
5001 .func = ftrace_stub,
5002 .flags = FTRACE_OPS_FL_STUB | FTRACE_OPS_FL_GLOBAL |
5003 FTRACE_OPS_FL_RECURSION_SAFE,
5004};
5005
5006int register_ftrace_graph(trace_func_graph_ret_t retfunc,
5007 trace_func_graph_ent_t entryfunc)
5008{
5009 int ret = 0;
5010
5011 mutex_lock(&ftrace_lock);
5012
5013 /* we currently allow only one tracer registered at a time */
5014 if (ftrace_graph_active) {
5015 ret = -EBUSY;
5016 goto out;
5017 }
5018
5019 ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
5020 register_pm_notifier(&ftrace_suspend_notifier);
5021
5022 ftrace_graph_active++;
5023 ret = start_graph_tracing();
5024 if (ret) {
5025 ftrace_graph_active--;
5026 goto out;
5027 }
5028
5029 ftrace_graph_return = retfunc;
5030 ftrace_graph_entry = entryfunc;
5031
5032 ret = ftrace_startup(&fgraph_ops, FTRACE_START_FUNC_RET);
5033
5034out:
5035 mutex_unlock(&ftrace_lock);
5036 return ret;
5037}
5038
5039void unregister_ftrace_graph(void)
5040{
5041 mutex_lock(&ftrace_lock);
5042
5043 if (unlikely(!ftrace_graph_active))
5044 goto out;
5045
5046 ftrace_graph_active--;
5047 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
5048 ftrace_graph_entry = ftrace_graph_entry_stub;
5049 ftrace_shutdown(&fgraph_ops, FTRACE_STOP_FUNC_RET);
5050 unregister_pm_notifier(&ftrace_suspend_notifier);
5051 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5052
5053 out:
5054 mutex_unlock(&ftrace_lock);
5055}
5056
5057static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
5058
5059static void
5060graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
5061{
5062 atomic_set(&t->tracing_graph_pause, 0);
5063 atomic_set(&t->trace_overrun, 0);
5064 t->ftrace_timestamp = 0;
5065 /* make curr_ret_stack visible before we add the ret_stack */
5066 smp_wmb();
5067 t->ret_stack = ret_stack;
5068}
5069
5070/*
5071 * Allocate a return stack for the idle task. May be the first
5072 * time through, or it may be done by CPU hotplug online.
5073 */
5074void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
5075{
5076 t->curr_ret_stack = -1;
5077 /*
5078 * The idle task has no parent, it either has its own
5079 * stack or no stack at all.
5080 */
5081 if (t->ret_stack)
5082 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
5083
5084 if (ftrace_graph_active) {
5085 struct ftrace_ret_stack *ret_stack;
5086
5087 ret_stack = per_cpu(idle_ret_stack, cpu);
5088 if (!ret_stack) {
5089 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5090 * sizeof(struct ftrace_ret_stack),
5091 GFP_KERNEL);
5092 if (!ret_stack)
5093 return;
5094 per_cpu(idle_ret_stack, cpu) = ret_stack;
5095 }
5096 graph_init_task(t, ret_stack);
5097 }
5098}
5099
5100/* Allocate a return stack for newly created task */
5101void ftrace_graph_init_task(struct task_struct *t)
5102{
5103 /* Make sure we do not use the parent ret_stack */
5104 t->ret_stack = NULL;
5105 t->curr_ret_stack = -1;
5106
5107 if (ftrace_graph_active) {
5108 struct ftrace_ret_stack *ret_stack;
5109
5110 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5111 * sizeof(struct ftrace_ret_stack),
5112 GFP_KERNEL);
5113 if (!ret_stack)
5114 return;
5115 graph_init_task(t, ret_stack);
5116 }
5117}
5118
5119void ftrace_graph_exit_task(struct task_struct *t)
5120{
5121 struct ftrace_ret_stack *ret_stack = t->ret_stack;
5122
5123 t->ret_stack = NULL;
5124 /* NULL must become visible to IRQs before we free it: */
5125 barrier();
5126
5127 kfree(ret_stack);
5128}
5129
5130void ftrace_graph_stop(void)
5131{
5132 ftrace_stop();
5133}
5134#endif