ftrace: Fix hash record accounting bug
[linux-2.6.git] / kernel / trace / ftrace.c
blobb1e8943fed1d3a9fd61916527c59c70d57af7d2c
1 /*
2 * Infrastructure for profiling code inserted by 'gcc -pg'.
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
7 * Originally ported from the -rt patch by:
8 * Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code in the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 William Lee Irwin III
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/module.h>
26 #include <linux/ftrace.h>
27 #include <linux/sysctl.h>
28 #include <linux/slab.h>
29 #include <linux/ctype.h>
30 #include <linux/list.h>
31 #include <linux/hash.h>
32 #include <linux/rcupdate.h>
34 #include <trace/events/sched.h>
36 #include <asm/setup.h>
38 #include "trace_output.h"
39 #include "trace_stat.h"
41 #define FTRACE_WARN_ON(cond) \
42 ({ \
43 int ___r = cond; \
44 if (WARN_ON(___r)) \
45 ftrace_kill(); \
46 ___r; \
49 #define FTRACE_WARN_ON_ONCE(cond) \
50 ({ \
51 int ___r = cond; \
52 if (WARN_ON_ONCE(___r)) \
53 ftrace_kill(); \
54 ___r; \
57 /* hash bits for specific function selection */
58 #define FTRACE_HASH_BITS 7
59 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
60 #define FTRACE_HASH_DEFAULT_BITS 10
61 #define FTRACE_HASH_MAX_BITS 12
63 /* ftrace_enabled is a method to turn ftrace on or off */
64 int ftrace_enabled __read_mostly;
65 static int last_ftrace_enabled;
67 /* Quick disabling of function tracer. */
68 int function_trace_stop;
70 /* List for set_ftrace_pid's pids. */
71 LIST_HEAD(ftrace_pids);
72 struct ftrace_pid {
73 struct list_head list;
74 struct pid *pid;
78 * ftrace_disabled is set when an anomaly is discovered.
79 * ftrace_disabled is much stronger than ftrace_enabled.
81 static int ftrace_disabled __read_mostly;
83 static DEFINE_MUTEX(ftrace_lock);
85 static struct ftrace_ops ftrace_list_end __read_mostly = {
86 .func = ftrace_stub,
89 static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
90 static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
91 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
92 static ftrace_func_t __ftrace_trace_function_delay __read_mostly = ftrace_stub;
93 ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
94 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
95 static struct ftrace_ops global_ops;
97 static void
98 ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip);
101 * Traverse the ftrace_global_list, invoking all entries. The reason that we
102 * can use rcu_dereference_raw() is that elements removed from this list
103 * are simply leaked, so there is no need to interact with a grace-period
104 * mechanism. The rcu_dereference_raw() calls are needed to handle
105 * concurrent insertions into the ftrace_global_list.
107 * Silly Alpha and silly pointer-speculation compiler optimizations!
109 static void ftrace_global_list_func(unsigned long ip,
110 unsigned long parent_ip)
112 struct ftrace_ops *op;
114 if (unlikely(trace_recursion_test(TRACE_GLOBAL_BIT)))
115 return;
117 trace_recursion_set(TRACE_GLOBAL_BIT);
118 op = rcu_dereference_raw(ftrace_global_list); /*see above*/
119 while (op != &ftrace_list_end) {
120 op->func(ip, parent_ip);
121 op = rcu_dereference_raw(op->next); /*see above*/
123 trace_recursion_clear(TRACE_GLOBAL_BIT);
126 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
128 if (!test_tsk_trace_trace(current))
129 return;
131 ftrace_pid_function(ip, parent_ip);
134 static void set_ftrace_pid_function(ftrace_func_t func)
136 /* do not set ftrace_pid_function to itself! */
137 if (func != ftrace_pid_func)
138 ftrace_pid_function = func;
142 * clear_ftrace_function - reset the ftrace function
144 * This NULLs the ftrace function and in essence stops
145 * tracing. There may be lag
147 void clear_ftrace_function(void)
149 ftrace_trace_function = ftrace_stub;
150 __ftrace_trace_function = ftrace_stub;
151 __ftrace_trace_function_delay = ftrace_stub;
152 ftrace_pid_function = ftrace_stub;
155 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
157 * For those archs that do not test ftrace_trace_stop in their
158 * mcount call site, we need to do it from C.
160 static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
162 if (function_trace_stop)
163 return;
165 __ftrace_trace_function(ip, parent_ip);
167 #endif
169 static void update_global_ops(void)
171 ftrace_func_t func;
174 * If there's only one function registered, then call that
175 * function directly. Otherwise, we need to iterate over the
176 * registered callers.
178 if (ftrace_global_list == &ftrace_list_end ||
179 ftrace_global_list->next == &ftrace_list_end)
180 func = ftrace_global_list->func;
181 else
182 func = ftrace_global_list_func;
184 /* If we filter on pids, update to use the pid function */
185 if (!list_empty(&ftrace_pids)) {
186 set_ftrace_pid_function(func);
187 func = ftrace_pid_func;
190 global_ops.func = func;
193 static void update_ftrace_function(void)
195 ftrace_func_t func;
197 update_global_ops();
200 * If we are at the end of the list and this ops is
201 * not dynamic, then have the mcount trampoline call
202 * the function directly
204 if (ftrace_ops_list == &ftrace_list_end ||
205 (ftrace_ops_list->next == &ftrace_list_end &&
206 !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC)))
207 func = ftrace_ops_list->func;
208 else
209 func = ftrace_ops_list_func;
211 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
212 ftrace_trace_function = func;
213 #else
214 #ifdef CONFIG_DYNAMIC_FTRACE
215 /* do not update till all functions have been modified */
216 __ftrace_trace_function_delay = func;
217 #else
218 __ftrace_trace_function = func;
219 #endif
220 ftrace_trace_function = ftrace_test_stop_func;
221 #endif
224 static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
226 ops->next = *list;
228 * We are entering ops into the list but another
229 * CPU might be walking that list. We need to make sure
230 * the ops->next pointer is valid before another CPU sees
231 * the ops pointer included into the list.
233 rcu_assign_pointer(*list, ops);
236 static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
238 struct ftrace_ops **p;
241 * If we are removing the last function, then simply point
242 * to the ftrace_stub.
244 if (*list == ops && ops->next == &ftrace_list_end) {
245 *list = &ftrace_list_end;
246 return 0;
249 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
250 if (*p == ops)
251 break;
253 if (*p != ops)
254 return -1;
256 *p = (*p)->next;
257 return 0;
260 static int __register_ftrace_function(struct ftrace_ops *ops)
262 if (ftrace_disabled)
263 return -ENODEV;
265 if (FTRACE_WARN_ON(ops == &global_ops))
266 return -EINVAL;
268 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
269 return -EBUSY;
271 if (!core_kernel_data((unsigned long)ops))
272 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
274 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
275 int first = ftrace_global_list == &ftrace_list_end;
276 add_ftrace_ops(&ftrace_global_list, ops);
277 ops->flags |= FTRACE_OPS_FL_ENABLED;
278 if (first)
279 add_ftrace_ops(&ftrace_ops_list, &global_ops);
280 } else
281 add_ftrace_ops(&ftrace_ops_list, ops);
283 if (ftrace_enabled)
284 update_ftrace_function();
286 return 0;
289 static int __unregister_ftrace_function(struct ftrace_ops *ops)
291 int ret;
293 if (ftrace_disabled)
294 return -ENODEV;
296 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
297 return -EBUSY;
299 if (FTRACE_WARN_ON(ops == &global_ops))
300 return -EINVAL;
302 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
303 ret = remove_ftrace_ops(&ftrace_global_list, ops);
304 if (!ret && ftrace_global_list == &ftrace_list_end)
305 ret = remove_ftrace_ops(&ftrace_ops_list, &global_ops);
306 if (!ret)
307 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
308 } else
309 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
311 if (ret < 0)
312 return ret;
314 if (ftrace_enabled)
315 update_ftrace_function();
318 * Dynamic ops may be freed, we must make sure that all
319 * callers are done before leaving this function.
321 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
322 synchronize_sched();
324 return 0;
327 static void ftrace_update_pid_func(void)
329 /* Only do something if we are tracing something */
330 if (ftrace_trace_function == ftrace_stub)
331 return;
333 update_ftrace_function();
336 #ifdef CONFIG_FUNCTION_PROFILER
337 struct ftrace_profile {
338 struct hlist_node node;
339 unsigned long ip;
340 unsigned long counter;
341 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
342 unsigned long long time;
343 unsigned long long time_squared;
344 #endif
347 struct ftrace_profile_page {
348 struct ftrace_profile_page *next;
349 unsigned long index;
350 struct ftrace_profile records[];
353 struct ftrace_profile_stat {
354 atomic_t disabled;
355 struct hlist_head *hash;
356 struct ftrace_profile_page *pages;
357 struct ftrace_profile_page *start;
358 struct tracer_stat stat;
361 #define PROFILE_RECORDS_SIZE \
362 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
364 #define PROFILES_PER_PAGE \
365 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
367 static int ftrace_profile_bits __read_mostly;
368 static int ftrace_profile_enabled __read_mostly;
370 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
371 static DEFINE_MUTEX(ftrace_profile_lock);
373 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
375 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
377 static void *
378 function_stat_next(void *v, int idx)
380 struct ftrace_profile *rec = v;
381 struct ftrace_profile_page *pg;
383 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
385 again:
386 if (idx != 0)
387 rec++;
389 if ((void *)rec >= (void *)&pg->records[pg->index]) {
390 pg = pg->next;
391 if (!pg)
392 return NULL;
393 rec = &pg->records[0];
394 if (!rec->counter)
395 goto again;
398 return rec;
401 static void *function_stat_start(struct tracer_stat *trace)
403 struct ftrace_profile_stat *stat =
404 container_of(trace, struct ftrace_profile_stat, stat);
406 if (!stat || !stat->start)
407 return NULL;
409 return function_stat_next(&stat->start->records[0], 0);
412 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
413 /* function graph compares on total time */
414 static int function_stat_cmp(void *p1, void *p2)
416 struct ftrace_profile *a = p1;
417 struct ftrace_profile *b = p2;
419 if (a->time < b->time)
420 return -1;
421 if (a->time > b->time)
422 return 1;
423 else
424 return 0;
426 #else
427 /* not function graph compares against hits */
428 static int function_stat_cmp(void *p1, void *p2)
430 struct ftrace_profile *a = p1;
431 struct ftrace_profile *b = p2;
433 if (a->counter < b->counter)
434 return -1;
435 if (a->counter > b->counter)
436 return 1;
437 else
438 return 0;
440 #endif
442 static int function_stat_headers(struct seq_file *m)
444 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
445 seq_printf(m, " Function "
446 "Hit Time Avg s^2\n"
447 " -------- "
448 "--- ---- --- ---\n");
449 #else
450 seq_printf(m, " Function Hit\n"
451 " -------- ---\n");
452 #endif
453 return 0;
456 static int function_stat_show(struct seq_file *m, void *v)
458 struct ftrace_profile *rec = v;
459 char str[KSYM_SYMBOL_LEN];
460 int ret = 0;
461 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
462 static struct trace_seq s;
463 unsigned long long avg;
464 unsigned long long stddev;
465 #endif
466 mutex_lock(&ftrace_profile_lock);
468 /* we raced with function_profile_reset() */
469 if (unlikely(rec->counter == 0)) {
470 ret = -EBUSY;
471 goto out;
474 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
475 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
477 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
478 seq_printf(m, " ");
479 avg = rec->time;
480 do_div(avg, rec->counter);
482 /* Sample standard deviation (s^2) */
483 if (rec->counter <= 1)
484 stddev = 0;
485 else {
486 stddev = rec->time_squared - rec->counter * avg * avg;
488 * Divide only 1000 for ns^2 -> us^2 conversion.
489 * trace_print_graph_duration will divide 1000 again.
491 do_div(stddev, (rec->counter - 1) * 1000);
494 trace_seq_init(&s);
495 trace_print_graph_duration(rec->time, &s);
496 trace_seq_puts(&s, " ");
497 trace_print_graph_duration(avg, &s);
498 trace_seq_puts(&s, " ");
499 trace_print_graph_duration(stddev, &s);
500 trace_print_seq(m, &s);
501 #endif
502 seq_putc(m, '\n');
503 out:
504 mutex_unlock(&ftrace_profile_lock);
506 return ret;
509 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
511 struct ftrace_profile_page *pg;
513 pg = stat->pages = stat->start;
515 while (pg) {
516 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
517 pg->index = 0;
518 pg = pg->next;
521 memset(stat->hash, 0,
522 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
525 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
527 struct ftrace_profile_page *pg;
528 int functions;
529 int pages;
530 int i;
532 /* If we already allocated, do nothing */
533 if (stat->pages)
534 return 0;
536 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
537 if (!stat->pages)
538 return -ENOMEM;
540 #ifdef CONFIG_DYNAMIC_FTRACE
541 functions = ftrace_update_tot_cnt;
542 #else
544 * We do not know the number of functions that exist because
545 * dynamic tracing is what counts them. With past experience
546 * we have around 20K functions. That should be more than enough.
547 * It is highly unlikely we will execute every function in
548 * the kernel.
550 functions = 20000;
551 #endif
553 pg = stat->start = stat->pages;
555 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
557 for (i = 0; i < pages; i++) {
558 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
559 if (!pg->next)
560 goto out_free;
561 pg = pg->next;
564 return 0;
566 out_free:
567 pg = stat->start;
568 while (pg) {
569 unsigned long tmp = (unsigned long)pg;
571 pg = pg->next;
572 free_page(tmp);
575 free_page((unsigned long)stat->pages);
576 stat->pages = NULL;
577 stat->start = NULL;
579 return -ENOMEM;
582 static int ftrace_profile_init_cpu(int cpu)
584 struct ftrace_profile_stat *stat;
585 int size;
587 stat = &per_cpu(ftrace_profile_stats, cpu);
589 if (stat->hash) {
590 /* If the profile is already created, simply reset it */
591 ftrace_profile_reset(stat);
592 return 0;
596 * We are profiling all functions, but usually only a few thousand
597 * functions are hit. We'll make a hash of 1024 items.
599 size = FTRACE_PROFILE_HASH_SIZE;
601 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
603 if (!stat->hash)
604 return -ENOMEM;
606 if (!ftrace_profile_bits) {
607 size--;
609 for (; size; size >>= 1)
610 ftrace_profile_bits++;
613 /* Preallocate the function profiling pages */
614 if (ftrace_profile_pages_init(stat) < 0) {
615 kfree(stat->hash);
616 stat->hash = NULL;
617 return -ENOMEM;
620 return 0;
623 static int ftrace_profile_init(void)
625 int cpu;
626 int ret = 0;
628 for_each_online_cpu(cpu) {
629 ret = ftrace_profile_init_cpu(cpu);
630 if (ret)
631 break;
634 return ret;
637 /* interrupts must be disabled */
638 static struct ftrace_profile *
639 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
641 struct ftrace_profile *rec;
642 struct hlist_head *hhd;
643 struct hlist_node *n;
644 unsigned long key;
646 key = hash_long(ip, ftrace_profile_bits);
647 hhd = &stat->hash[key];
649 if (hlist_empty(hhd))
650 return NULL;
652 hlist_for_each_entry_rcu(rec, n, hhd, node) {
653 if (rec->ip == ip)
654 return rec;
657 return NULL;
660 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
661 struct ftrace_profile *rec)
663 unsigned long key;
665 key = hash_long(rec->ip, ftrace_profile_bits);
666 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
670 * The memory is already allocated, this simply finds a new record to use.
672 static struct ftrace_profile *
673 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
675 struct ftrace_profile *rec = NULL;
677 /* prevent recursion (from NMIs) */
678 if (atomic_inc_return(&stat->disabled) != 1)
679 goto out;
682 * Try to find the function again since an NMI
683 * could have added it
685 rec = ftrace_find_profiled_func(stat, ip);
686 if (rec)
687 goto out;
689 if (stat->pages->index == PROFILES_PER_PAGE) {
690 if (!stat->pages->next)
691 goto out;
692 stat->pages = stat->pages->next;
695 rec = &stat->pages->records[stat->pages->index++];
696 rec->ip = ip;
697 ftrace_add_profile(stat, rec);
699 out:
700 atomic_dec(&stat->disabled);
702 return rec;
705 static void
706 function_profile_call(unsigned long ip, unsigned long parent_ip)
708 struct ftrace_profile_stat *stat;
709 struct ftrace_profile *rec;
710 unsigned long flags;
712 if (!ftrace_profile_enabled)
713 return;
715 local_irq_save(flags);
717 stat = &__get_cpu_var(ftrace_profile_stats);
718 if (!stat->hash || !ftrace_profile_enabled)
719 goto out;
721 rec = ftrace_find_profiled_func(stat, ip);
722 if (!rec) {
723 rec = ftrace_profile_alloc(stat, ip);
724 if (!rec)
725 goto out;
728 rec->counter++;
729 out:
730 local_irq_restore(flags);
733 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
734 static int profile_graph_entry(struct ftrace_graph_ent *trace)
736 function_profile_call(trace->func, 0);
737 return 1;
740 static void profile_graph_return(struct ftrace_graph_ret *trace)
742 struct ftrace_profile_stat *stat;
743 unsigned long long calltime;
744 struct ftrace_profile *rec;
745 unsigned long flags;
747 local_irq_save(flags);
748 stat = &__get_cpu_var(ftrace_profile_stats);
749 if (!stat->hash || !ftrace_profile_enabled)
750 goto out;
752 /* If the calltime was zero'd ignore it */
753 if (!trace->calltime)
754 goto out;
756 calltime = trace->rettime - trace->calltime;
758 if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
759 int index;
761 index = trace->depth;
763 /* Append this call time to the parent time to subtract */
764 if (index)
765 current->ret_stack[index - 1].subtime += calltime;
767 if (current->ret_stack[index].subtime < calltime)
768 calltime -= current->ret_stack[index].subtime;
769 else
770 calltime = 0;
773 rec = ftrace_find_profiled_func(stat, trace->func);
774 if (rec) {
775 rec->time += calltime;
776 rec->time_squared += calltime * calltime;
779 out:
780 local_irq_restore(flags);
783 static int register_ftrace_profiler(void)
785 return register_ftrace_graph(&profile_graph_return,
786 &profile_graph_entry);
789 static void unregister_ftrace_profiler(void)
791 unregister_ftrace_graph();
793 #else
794 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
795 .func = function_profile_call,
798 static int register_ftrace_profiler(void)
800 return register_ftrace_function(&ftrace_profile_ops);
803 static void unregister_ftrace_profiler(void)
805 unregister_ftrace_function(&ftrace_profile_ops);
807 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
809 static ssize_t
810 ftrace_profile_write(struct file *filp, const char __user *ubuf,
811 size_t cnt, loff_t *ppos)
813 unsigned long val;
814 int ret;
816 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
817 if (ret)
818 return ret;
820 val = !!val;
822 mutex_lock(&ftrace_profile_lock);
823 if (ftrace_profile_enabled ^ val) {
824 if (val) {
825 ret = ftrace_profile_init();
826 if (ret < 0) {
827 cnt = ret;
828 goto out;
831 ret = register_ftrace_profiler();
832 if (ret < 0) {
833 cnt = ret;
834 goto out;
836 ftrace_profile_enabled = 1;
837 } else {
838 ftrace_profile_enabled = 0;
840 * unregister_ftrace_profiler calls stop_machine
841 * so this acts like an synchronize_sched.
843 unregister_ftrace_profiler();
846 out:
847 mutex_unlock(&ftrace_profile_lock);
849 *ppos += cnt;
851 return cnt;
854 static ssize_t
855 ftrace_profile_read(struct file *filp, char __user *ubuf,
856 size_t cnt, loff_t *ppos)
858 char buf[64]; /* big enough to hold a number */
859 int r;
861 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
862 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
865 static const struct file_operations ftrace_profile_fops = {
866 .open = tracing_open_generic,
867 .read = ftrace_profile_read,
868 .write = ftrace_profile_write,
869 .llseek = default_llseek,
872 /* used to initialize the real stat files */
873 static struct tracer_stat function_stats __initdata = {
874 .name = "functions",
875 .stat_start = function_stat_start,
876 .stat_next = function_stat_next,
877 .stat_cmp = function_stat_cmp,
878 .stat_headers = function_stat_headers,
879 .stat_show = function_stat_show
882 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
884 struct ftrace_profile_stat *stat;
885 struct dentry *entry;
886 char *name;
887 int ret;
888 int cpu;
890 for_each_possible_cpu(cpu) {
891 stat = &per_cpu(ftrace_profile_stats, cpu);
893 /* allocate enough for function name + cpu number */
894 name = kmalloc(32, GFP_KERNEL);
895 if (!name) {
897 * The files created are permanent, if something happens
898 * we still do not free memory.
900 WARN(1,
901 "Could not allocate stat file for cpu %d\n",
902 cpu);
903 return;
905 stat->stat = function_stats;
906 snprintf(name, 32, "function%d", cpu);
907 stat->stat.name = name;
908 ret = register_stat_tracer(&stat->stat);
909 if (ret) {
910 WARN(1,
911 "Could not register function stat for cpu %d\n",
912 cpu);
913 kfree(name);
914 return;
918 entry = debugfs_create_file("function_profile_enabled", 0644,
919 d_tracer, NULL, &ftrace_profile_fops);
920 if (!entry)
921 pr_warning("Could not create debugfs "
922 "'function_profile_enabled' entry\n");
925 #else /* CONFIG_FUNCTION_PROFILER */
926 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
929 #endif /* CONFIG_FUNCTION_PROFILER */
931 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
933 #ifdef CONFIG_DYNAMIC_FTRACE
935 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
936 # error Dynamic ftrace depends on MCOUNT_RECORD
937 #endif
939 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
941 struct ftrace_func_probe {
942 struct hlist_node node;
943 struct ftrace_probe_ops *ops;
944 unsigned long flags;
945 unsigned long ip;
946 void *data;
947 struct rcu_head rcu;
950 enum {
951 FTRACE_ENABLE_CALLS = (1 << 0),
952 FTRACE_DISABLE_CALLS = (1 << 1),
953 FTRACE_UPDATE_TRACE_FUNC = (1 << 2),
954 FTRACE_START_FUNC_RET = (1 << 3),
955 FTRACE_STOP_FUNC_RET = (1 << 4),
957 struct ftrace_func_entry {
958 struct hlist_node hlist;
959 unsigned long ip;
962 struct ftrace_hash {
963 unsigned long size_bits;
964 struct hlist_head *buckets;
965 unsigned long count;
966 struct rcu_head rcu;
970 * We make these constant because no one should touch them,
971 * but they are used as the default "empty hash", to avoid allocating
972 * it all the time. These are in a read only section such that if
973 * anyone does try to modify it, it will cause an exception.
975 static const struct hlist_head empty_buckets[1];
976 static const struct ftrace_hash empty_hash = {
977 .buckets = (struct hlist_head *)empty_buckets,
979 #define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
981 static struct ftrace_ops global_ops = {
982 .func = ftrace_stub,
983 .notrace_hash = EMPTY_HASH,
984 .filter_hash = EMPTY_HASH,
987 static struct dyn_ftrace *ftrace_new_addrs;
989 static DEFINE_MUTEX(ftrace_regex_lock);
991 struct ftrace_page {
992 struct ftrace_page *next;
993 int index;
994 struct dyn_ftrace records[];
997 #define ENTRIES_PER_PAGE \
998 ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
1000 /* estimate from running different kernels */
1001 #define NR_TO_INIT 10000
1003 static struct ftrace_page *ftrace_pages_start;
1004 static struct ftrace_page *ftrace_pages;
1006 static struct dyn_ftrace *ftrace_free_records;
1008 static struct ftrace_func_entry *
1009 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1011 unsigned long key;
1012 struct ftrace_func_entry *entry;
1013 struct hlist_head *hhd;
1014 struct hlist_node *n;
1016 if (!hash->count)
1017 return NULL;
1019 if (hash->size_bits > 0)
1020 key = hash_long(ip, hash->size_bits);
1021 else
1022 key = 0;
1024 hhd = &hash->buckets[key];
1026 hlist_for_each_entry_rcu(entry, n, hhd, hlist) {
1027 if (entry->ip == ip)
1028 return entry;
1030 return NULL;
1033 static void __add_hash_entry(struct ftrace_hash *hash,
1034 struct ftrace_func_entry *entry)
1036 struct hlist_head *hhd;
1037 unsigned long key;
1039 if (hash->size_bits)
1040 key = hash_long(entry->ip, hash->size_bits);
1041 else
1042 key = 0;
1044 hhd = &hash->buckets[key];
1045 hlist_add_head(&entry->hlist, hhd);
1046 hash->count++;
1049 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1051 struct ftrace_func_entry *entry;
1053 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1054 if (!entry)
1055 return -ENOMEM;
1057 entry->ip = ip;
1058 __add_hash_entry(hash, entry);
1060 return 0;
1063 static void
1064 free_hash_entry(struct ftrace_hash *hash,
1065 struct ftrace_func_entry *entry)
1067 hlist_del(&entry->hlist);
1068 kfree(entry);
1069 hash->count--;
1072 static void
1073 remove_hash_entry(struct ftrace_hash *hash,
1074 struct ftrace_func_entry *entry)
1076 hlist_del(&entry->hlist);
1077 hash->count--;
1080 static void ftrace_hash_clear(struct ftrace_hash *hash)
1082 struct hlist_head *hhd;
1083 struct hlist_node *tp, *tn;
1084 struct ftrace_func_entry *entry;
1085 int size = 1 << hash->size_bits;
1086 int i;
1088 if (!hash->count)
1089 return;
1091 for (i = 0; i < size; i++) {
1092 hhd = &hash->buckets[i];
1093 hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist)
1094 free_hash_entry(hash, entry);
1096 FTRACE_WARN_ON(hash->count);
1099 static void free_ftrace_hash(struct ftrace_hash *hash)
1101 if (!hash || hash == EMPTY_HASH)
1102 return;
1103 ftrace_hash_clear(hash);
1104 kfree(hash->buckets);
1105 kfree(hash);
1108 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1110 struct ftrace_hash *hash;
1112 hash = container_of(rcu, struct ftrace_hash, rcu);
1113 free_ftrace_hash(hash);
1116 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1118 if (!hash || hash == EMPTY_HASH)
1119 return;
1120 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1123 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1125 struct ftrace_hash *hash;
1126 int size;
1128 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1129 if (!hash)
1130 return NULL;
1132 size = 1 << size_bits;
1133 hash->buckets = kzalloc(sizeof(*hash->buckets) * size, GFP_KERNEL);
1135 if (!hash->buckets) {
1136 kfree(hash);
1137 return NULL;
1140 hash->size_bits = size_bits;
1142 return hash;
1145 static struct ftrace_hash *
1146 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1148 struct ftrace_func_entry *entry;
1149 struct ftrace_hash *new_hash;
1150 struct hlist_node *tp;
1151 int size;
1152 int ret;
1153 int i;
1155 new_hash = alloc_ftrace_hash(size_bits);
1156 if (!new_hash)
1157 return NULL;
1159 /* Empty hash? */
1160 if (!hash || !hash->count)
1161 return new_hash;
1163 size = 1 << hash->size_bits;
1164 for (i = 0; i < size; i++) {
1165 hlist_for_each_entry(entry, tp, &hash->buckets[i], hlist) {
1166 ret = add_hash_entry(new_hash, entry->ip);
1167 if (ret < 0)
1168 goto free_hash;
1172 FTRACE_WARN_ON(new_hash->count != hash->count);
1174 return new_hash;
1176 free_hash:
1177 free_ftrace_hash(new_hash);
1178 return NULL;
1181 static void
1182 ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1183 static void
1184 ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1186 static int
1187 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1188 struct ftrace_hash **dst, struct ftrace_hash *src)
1190 struct ftrace_func_entry *entry;
1191 struct hlist_node *tp, *tn;
1192 struct hlist_head *hhd;
1193 struct ftrace_hash *old_hash;
1194 struct ftrace_hash *new_hash;
1195 unsigned long key;
1196 int size = src->count;
1197 int bits = 0;
1198 int ret;
1199 int i;
1202 * Remove the current set, update the hash and add
1203 * them back.
1205 ftrace_hash_rec_disable(ops, enable);
1208 * If the new source is empty, just free dst and assign it
1209 * the empty_hash.
1211 if (!src->count) {
1212 free_ftrace_hash_rcu(*dst);
1213 rcu_assign_pointer(*dst, EMPTY_HASH);
1214 /* still need to update the function records */
1215 ret = 0;
1216 goto out;
1220 * Make the hash size about 1/2 the # found
1222 for (size /= 2; size; size >>= 1)
1223 bits++;
1225 /* Don't allocate too much */
1226 if (bits > FTRACE_HASH_MAX_BITS)
1227 bits = FTRACE_HASH_MAX_BITS;
1229 ret = -ENOMEM;
1230 new_hash = alloc_ftrace_hash(bits);
1231 if (!new_hash)
1232 goto out;
1234 size = 1 << src->size_bits;
1235 for (i = 0; i < size; i++) {
1236 hhd = &src->buckets[i];
1237 hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist) {
1238 if (bits > 0)
1239 key = hash_long(entry->ip, bits);
1240 else
1241 key = 0;
1242 remove_hash_entry(src, entry);
1243 __add_hash_entry(new_hash, entry);
1247 old_hash = *dst;
1248 rcu_assign_pointer(*dst, new_hash);
1249 free_ftrace_hash_rcu(old_hash);
1251 ret = 0;
1252 out:
1254 * Enable regardless of ret:
1255 * On success, we enable the new hash.
1256 * On failure, we re-enable the original hash.
1258 ftrace_hash_rec_enable(ops, enable);
1260 return ret;
1264 * Test the hashes for this ops to see if we want to call
1265 * the ops->func or not.
1267 * It's a match if the ip is in the ops->filter_hash or
1268 * the filter_hash does not exist or is empty,
1269 * AND
1270 * the ip is not in the ops->notrace_hash.
1272 * This needs to be called with preemption disabled as
1273 * the hashes are freed with call_rcu_sched().
1275 static int
1276 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
1278 struct ftrace_hash *filter_hash;
1279 struct ftrace_hash *notrace_hash;
1280 int ret;
1282 filter_hash = rcu_dereference_raw(ops->filter_hash);
1283 notrace_hash = rcu_dereference_raw(ops->notrace_hash);
1285 if ((!filter_hash || !filter_hash->count ||
1286 ftrace_lookup_ip(filter_hash, ip)) &&
1287 (!notrace_hash || !notrace_hash->count ||
1288 !ftrace_lookup_ip(notrace_hash, ip)))
1289 ret = 1;
1290 else
1291 ret = 0;
1293 return ret;
1297 * This is a double for. Do not use 'break' to break out of the loop,
1298 * you must use a goto.
1300 #define do_for_each_ftrace_rec(pg, rec) \
1301 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1302 int _____i; \
1303 for (_____i = 0; _____i < pg->index; _____i++) { \
1304 rec = &pg->records[_____i];
1306 #define while_for_each_ftrace_rec() \
1310 static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1311 int filter_hash,
1312 bool inc)
1314 struct ftrace_hash *hash;
1315 struct ftrace_hash *other_hash;
1316 struct ftrace_page *pg;
1317 struct dyn_ftrace *rec;
1318 int count = 0;
1319 int all = 0;
1321 /* Only update if the ops has been registered */
1322 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1323 return;
1326 * In the filter_hash case:
1327 * If the count is zero, we update all records.
1328 * Otherwise we just update the items in the hash.
1330 * In the notrace_hash case:
1331 * We enable the update in the hash.
1332 * As disabling notrace means enabling the tracing,
1333 * and enabling notrace means disabling, the inc variable
1334 * gets inversed.
1336 if (filter_hash) {
1337 hash = ops->filter_hash;
1338 other_hash = ops->notrace_hash;
1339 if (!hash || !hash->count)
1340 all = 1;
1341 } else {
1342 inc = !inc;
1343 hash = ops->notrace_hash;
1344 other_hash = ops->filter_hash;
1346 * If the notrace hash has no items,
1347 * then there's nothing to do.
1349 if (hash && !hash->count)
1350 return;
1353 do_for_each_ftrace_rec(pg, rec) {
1354 int in_other_hash = 0;
1355 int in_hash = 0;
1356 int match = 0;
1358 if (all) {
1360 * Only the filter_hash affects all records.
1361 * Update if the record is not in the notrace hash.
1363 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1364 match = 1;
1365 } else {
1366 in_hash = hash && !!ftrace_lookup_ip(hash, rec->ip);
1367 in_other_hash = other_hash && !!ftrace_lookup_ip(other_hash, rec->ip);
1372 if (filter_hash && in_hash && !in_other_hash)
1373 match = 1;
1374 else if (!filter_hash && in_hash &&
1375 (in_other_hash || !other_hash->count))
1376 match = 1;
1378 if (!match)
1379 continue;
1381 if (inc) {
1382 rec->flags++;
1383 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1384 return;
1385 } else {
1386 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1387 return;
1388 rec->flags--;
1390 count++;
1391 /* Shortcut, if we handled all records, we are done. */
1392 if (!all && count == hash->count)
1393 return;
1394 } while_for_each_ftrace_rec();
1397 static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1398 int filter_hash)
1400 __ftrace_hash_rec_update(ops, filter_hash, 0);
1403 static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1404 int filter_hash)
1406 __ftrace_hash_rec_update(ops, filter_hash, 1);
1409 static void ftrace_free_rec(struct dyn_ftrace *rec)
1411 rec->freelist = ftrace_free_records;
1412 ftrace_free_records = rec;
1413 rec->flags |= FTRACE_FL_FREE;
1416 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
1418 struct dyn_ftrace *rec;
1420 /* First check for freed records */
1421 if (ftrace_free_records) {
1422 rec = ftrace_free_records;
1424 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
1425 FTRACE_WARN_ON_ONCE(1);
1426 ftrace_free_records = NULL;
1427 return NULL;
1430 ftrace_free_records = rec->freelist;
1431 memset(rec, 0, sizeof(*rec));
1432 return rec;
1435 if (ftrace_pages->index == ENTRIES_PER_PAGE) {
1436 if (!ftrace_pages->next) {
1437 /* allocate another page */
1438 ftrace_pages->next =
1439 (void *)get_zeroed_page(GFP_KERNEL);
1440 if (!ftrace_pages->next)
1441 return NULL;
1443 ftrace_pages = ftrace_pages->next;
1446 return &ftrace_pages->records[ftrace_pages->index++];
1449 static struct dyn_ftrace *
1450 ftrace_record_ip(unsigned long ip)
1452 struct dyn_ftrace *rec;
1454 if (ftrace_disabled)
1455 return NULL;
1457 rec = ftrace_alloc_dyn_node(ip);
1458 if (!rec)
1459 return NULL;
1461 rec->ip = ip;
1462 rec->newlist = ftrace_new_addrs;
1463 ftrace_new_addrs = rec;
1465 return rec;
1468 static void print_ip_ins(const char *fmt, unsigned char *p)
1470 int i;
1472 printk(KERN_CONT "%s", fmt);
1474 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1475 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1478 static void ftrace_bug(int failed, unsigned long ip)
1480 switch (failed) {
1481 case -EFAULT:
1482 FTRACE_WARN_ON_ONCE(1);
1483 pr_info("ftrace faulted on modifying ");
1484 print_ip_sym(ip);
1485 break;
1486 case -EINVAL:
1487 FTRACE_WARN_ON_ONCE(1);
1488 pr_info("ftrace failed to modify ");
1489 print_ip_sym(ip);
1490 print_ip_ins(" actual: ", (unsigned char *)ip);
1491 printk(KERN_CONT "\n");
1492 break;
1493 case -EPERM:
1494 FTRACE_WARN_ON_ONCE(1);
1495 pr_info("ftrace faulted on writing ");
1496 print_ip_sym(ip);
1497 break;
1498 default:
1499 FTRACE_WARN_ON_ONCE(1);
1500 pr_info("ftrace faulted on unknown error ");
1501 print_ip_sym(ip);
1506 /* Return 1 if the address range is reserved for ftrace */
1507 int ftrace_text_reserved(void *start, void *end)
1509 struct dyn_ftrace *rec;
1510 struct ftrace_page *pg;
1512 do_for_each_ftrace_rec(pg, rec) {
1513 if (rec->ip <= (unsigned long)end &&
1514 rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1515 return 1;
1516 } while_for_each_ftrace_rec();
1517 return 0;
1521 static int
1522 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1524 unsigned long ftrace_addr;
1525 unsigned long flag = 0UL;
1527 ftrace_addr = (unsigned long)FTRACE_ADDR;
1530 * If we are enabling tracing:
1532 * If the record has a ref count, then we need to enable it
1533 * because someone is using it.
1535 * Otherwise we make sure its disabled.
1537 * If we are disabling tracing, then disable all records that
1538 * are enabled.
1540 if (enable && (rec->flags & ~FTRACE_FL_MASK))
1541 flag = FTRACE_FL_ENABLED;
1543 /* If the state of this record hasn't changed, then do nothing */
1544 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1545 return 0;
1547 if (flag) {
1548 rec->flags |= FTRACE_FL_ENABLED;
1549 return ftrace_make_call(rec, ftrace_addr);
1552 rec->flags &= ~FTRACE_FL_ENABLED;
1553 return ftrace_make_nop(NULL, rec, ftrace_addr);
1556 static void ftrace_replace_code(int enable)
1558 struct dyn_ftrace *rec;
1559 struct ftrace_page *pg;
1560 int failed;
1562 if (unlikely(ftrace_disabled))
1563 return;
1565 do_for_each_ftrace_rec(pg, rec) {
1566 /* Skip over free records */
1567 if (rec->flags & FTRACE_FL_FREE)
1568 continue;
1570 failed = __ftrace_replace_code(rec, enable);
1571 if (failed) {
1572 ftrace_bug(failed, rec->ip);
1573 /* Stop processing */
1574 return;
1576 } while_for_each_ftrace_rec();
1579 static int
1580 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1582 unsigned long ip;
1583 int ret;
1585 ip = rec->ip;
1587 if (unlikely(ftrace_disabled))
1588 return 0;
1590 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1591 if (ret) {
1592 ftrace_bug(ret, ip);
1593 return 0;
1595 return 1;
1599 * archs can override this function if they must do something
1600 * before the modifying code is performed.
1602 int __weak ftrace_arch_code_modify_prepare(void)
1604 return 0;
1608 * archs can override this function if they must do something
1609 * after the modifying code is performed.
1611 int __weak ftrace_arch_code_modify_post_process(void)
1613 return 0;
1616 static int __ftrace_modify_code(void *data)
1618 int *command = data;
1621 * Do not call function tracer while we update the code.
1622 * We are in stop machine, no worrying about races.
1624 function_trace_stop++;
1626 if (*command & FTRACE_ENABLE_CALLS)
1627 ftrace_replace_code(1);
1628 else if (*command & FTRACE_DISABLE_CALLS)
1629 ftrace_replace_code(0);
1631 if (*command & FTRACE_UPDATE_TRACE_FUNC)
1632 ftrace_update_ftrace_func(ftrace_trace_function);
1634 if (*command & FTRACE_START_FUNC_RET)
1635 ftrace_enable_ftrace_graph_caller();
1636 else if (*command & FTRACE_STOP_FUNC_RET)
1637 ftrace_disable_ftrace_graph_caller();
1639 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
1641 * For archs that call ftrace_test_stop_func(), we must
1642 * wait till after we update all the function callers
1643 * before we update the callback. This keeps different
1644 * ops that record different functions from corrupting
1645 * each other.
1647 __ftrace_trace_function = __ftrace_trace_function_delay;
1648 #endif
1649 function_trace_stop--;
1651 return 0;
1654 static void ftrace_run_update_code(int command)
1656 int ret;
1658 ret = ftrace_arch_code_modify_prepare();
1659 FTRACE_WARN_ON(ret);
1660 if (ret)
1661 return;
1663 stop_machine(__ftrace_modify_code, &command, NULL);
1665 ret = ftrace_arch_code_modify_post_process();
1666 FTRACE_WARN_ON(ret);
1669 static ftrace_func_t saved_ftrace_func;
1670 static int ftrace_start_up;
1671 static int global_start_up;
1673 static void ftrace_startup_enable(int command)
1675 if (saved_ftrace_func != ftrace_trace_function) {
1676 saved_ftrace_func = ftrace_trace_function;
1677 command |= FTRACE_UPDATE_TRACE_FUNC;
1680 if (!command || !ftrace_enabled)
1681 return;
1683 ftrace_run_update_code(command);
1686 static int ftrace_startup(struct ftrace_ops *ops, int command)
1688 bool hash_enable = true;
1690 if (unlikely(ftrace_disabled))
1691 return -ENODEV;
1693 ftrace_start_up++;
1694 command |= FTRACE_ENABLE_CALLS;
1696 /* ops marked global share the filter hashes */
1697 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1698 ops = &global_ops;
1699 /* Don't update hash if global is already set */
1700 if (global_start_up)
1701 hash_enable = false;
1702 global_start_up++;
1705 ops->flags |= FTRACE_OPS_FL_ENABLED;
1706 if (hash_enable)
1707 ftrace_hash_rec_enable(ops, 1);
1709 ftrace_startup_enable(command);
1711 return 0;
1714 static void ftrace_shutdown(struct ftrace_ops *ops, int command)
1716 bool hash_disable = true;
1718 if (unlikely(ftrace_disabled))
1719 return;
1721 ftrace_start_up--;
1723 * Just warn in case of unbalance, no need to kill ftrace, it's not
1724 * critical but the ftrace_call callers may be never nopped again after
1725 * further ftrace uses.
1727 WARN_ON_ONCE(ftrace_start_up < 0);
1729 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1730 ops = &global_ops;
1731 global_start_up--;
1732 WARN_ON_ONCE(global_start_up < 0);
1733 /* Don't update hash if global still has users */
1734 if (global_start_up) {
1735 WARN_ON_ONCE(!ftrace_start_up);
1736 hash_disable = false;
1740 if (hash_disable)
1741 ftrace_hash_rec_disable(ops, 1);
1743 if (ops != &global_ops || !global_start_up)
1744 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
1746 if (!ftrace_start_up)
1747 command |= FTRACE_DISABLE_CALLS;
1749 if (saved_ftrace_func != ftrace_trace_function) {
1750 saved_ftrace_func = ftrace_trace_function;
1751 command |= FTRACE_UPDATE_TRACE_FUNC;
1754 if (!command || !ftrace_enabled)
1755 return;
1757 ftrace_run_update_code(command);
1760 static void ftrace_startup_sysctl(void)
1762 if (unlikely(ftrace_disabled))
1763 return;
1765 /* Force update next time */
1766 saved_ftrace_func = NULL;
1767 /* ftrace_start_up is true if we want ftrace running */
1768 if (ftrace_start_up)
1769 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1772 static void ftrace_shutdown_sysctl(void)
1774 if (unlikely(ftrace_disabled))
1775 return;
1777 /* ftrace_start_up is true if ftrace is running */
1778 if (ftrace_start_up)
1779 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
1782 static cycle_t ftrace_update_time;
1783 static unsigned long ftrace_update_cnt;
1784 unsigned long ftrace_update_tot_cnt;
1786 static int ops_traces_mod(struct ftrace_ops *ops)
1788 struct ftrace_hash *hash;
1790 hash = ops->filter_hash;
1791 return !!(!hash || !hash->count);
1794 static int ftrace_update_code(struct module *mod)
1796 struct dyn_ftrace *p;
1797 cycle_t start, stop;
1798 unsigned long ref = 0;
1801 * When adding a module, we need to check if tracers are
1802 * currently enabled and if they are set to trace all functions.
1803 * If they are, we need to enable the module functions as well
1804 * as update the reference counts for those function records.
1806 if (mod) {
1807 struct ftrace_ops *ops;
1809 for (ops = ftrace_ops_list;
1810 ops != &ftrace_list_end; ops = ops->next) {
1811 if (ops->flags & FTRACE_OPS_FL_ENABLED &&
1812 ops_traces_mod(ops))
1813 ref++;
1817 start = ftrace_now(raw_smp_processor_id());
1818 ftrace_update_cnt = 0;
1820 while (ftrace_new_addrs) {
1822 /* If something went wrong, bail without enabling anything */
1823 if (unlikely(ftrace_disabled))
1824 return -1;
1826 p = ftrace_new_addrs;
1827 ftrace_new_addrs = p->newlist;
1828 p->flags = ref;
1831 * Do the initial record conversion from mcount jump
1832 * to the NOP instructions.
1834 if (!ftrace_code_disable(mod, p)) {
1835 ftrace_free_rec(p);
1836 /* Game over */
1837 break;
1840 ftrace_update_cnt++;
1843 * If the tracing is enabled, go ahead and enable the record.
1845 * The reason not to enable the record immediatelly is the
1846 * inherent check of ftrace_make_nop/ftrace_make_call for
1847 * correct previous instructions. Making first the NOP
1848 * conversion puts the module to the correct state, thus
1849 * passing the ftrace_make_call check.
1851 if (ftrace_start_up && ref) {
1852 int failed = __ftrace_replace_code(p, 1);
1853 if (failed) {
1854 ftrace_bug(failed, p->ip);
1855 ftrace_free_rec(p);
1860 stop = ftrace_now(raw_smp_processor_id());
1861 ftrace_update_time = stop - start;
1862 ftrace_update_tot_cnt += ftrace_update_cnt;
1864 return 0;
1867 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1869 struct ftrace_page *pg;
1870 int cnt;
1871 int i;
1873 /* allocate a few pages */
1874 ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1875 if (!ftrace_pages_start)
1876 return -1;
1879 * Allocate a few more pages.
1881 * TODO: have some parser search vmlinux before
1882 * final linking to find all calls to ftrace.
1883 * Then we can:
1884 * a) know how many pages to allocate.
1885 * and/or
1886 * b) set up the table then.
1888 * The dynamic code is still necessary for
1889 * modules.
1892 pg = ftrace_pages = ftrace_pages_start;
1894 cnt = num_to_init / ENTRIES_PER_PAGE;
1895 pr_info("ftrace: allocating %ld entries in %d pages\n",
1896 num_to_init, cnt + 1);
1898 for (i = 0; i < cnt; i++) {
1899 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1901 /* If we fail, we'll try later anyway */
1902 if (!pg->next)
1903 break;
1905 pg = pg->next;
1908 return 0;
1911 enum {
1912 FTRACE_ITER_FILTER = (1 << 0),
1913 FTRACE_ITER_NOTRACE = (1 << 1),
1914 FTRACE_ITER_PRINTALL = (1 << 2),
1915 FTRACE_ITER_HASH = (1 << 3),
1916 FTRACE_ITER_ENABLED = (1 << 4),
1919 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1921 struct ftrace_iterator {
1922 loff_t pos;
1923 loff_t func_pos;
1924 struct ftrace_page *pg;
1925 struct dyn_ftrace *func;
1926 struct ftrace_func_probe *probe;
1927 struct trace_parser parser;
1928 struct ftrace_hash *hash;
1929 struct ftrace_ops *ops;
1930 int hidx;
1931 int idx;
1932 unsigned flags;
1935 static void *
1936 t_hash_next(struct seq_file *m, loff_t *pos)
1938 struct ftrace_iterator *iter = m->private;
1939 struct hlist_node *hnd = NULL;
1940 struct hlist_head *hhd;
1942 (*pos)++;
1943 iter->pos = *pos;
1945 if (iter->probe)
1946 hnd = &iter->probe->node;
1947 retry:
1948 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1949 return NULL;
1951 hhd = &ftrace_func_hash[iter->hidx];
1953 if (hlist_empty(hhd)) {
1954 iter->hidx++;
1955 hnd = NULL;
1956 goto retry;
1959 if (!hnd)
1960 hnd = hhd->first;
1961 else {
1962 hnd = hnd->next;
1963 if (!hnd) {
1964 iter->hidx++;
1965 goto retry;
1969 if (WARN_ON_ONCE(!hnd))
1970 return NULL;
1972 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
1974 return iter;
1977 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1979 struct ftrace_iterator *iter = m->private;
1980 void *p = NULL;
1981 loff_t l;
1983 if (iter->func_pos > *pos)
1984 return NULL;
1986 iter->hidx = 0;
1987 for (l = 0; l <= (*pos - iter->func_pos); ) {
1988 p = t_hash_next(m, &l);
1989 if (!p)
1990 break;
1992 if (!p)
1993 return NULL;
1995 /* Only set this if we have an item */
1996 iter->flags |= FTRACE_ITER_HASH;
1998 return iter;
2001 static int
2002 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2004 struct ftrace_func_probe *rec;
2006 rec = iter->probe;
2007 if (WARN_ON_ONCE(!rec))
2008 return -EIO;
2010 if (rec->ops->print)
2011 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2013 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2015 if (rec->data)
2016 seq_printf(m, ":%p", rec->data);
2017 seq_putc(m, '\n');
2019 return 0;
2022 static void *
2023 t_next(struct seq_file *m, void *v, loff_t *pos)
2025 struct ftrace_iterator *iter = m->private;
2026 struct ftrace_ops *ops = &global_ops;
2027 struct dyn_ftrace *rec = NULL;
2029 if (unlikely(ftrace_disabled))
2030 return NULL;
2032 if (iter->flags & FTRACE_ITER_HASH)
2033 return t_hash_next(m, pos);
2035 (*pos)++;
2036 iter->pos = iter->func_pos = *pos;
2038 if (iter->flags & FTRACE_ITER_PRINTALL)
2039 return t_hash_start(m, pos);
2041 retry:
2042 if (iter->idx >= iter->pg->index) {
2043 if (iter->pg->next) {
2044 iter->pg = iter->pg->next;
2045 iter->idx = 0;
2046 goto retry;
2048 } else {
2049 rec = &iter->pg->records[iter->idx++];
2050 if ((rec->flags & FTRACE_FL_FREE) ||
2052 ((iter->flags & FTRACE_ITER_FILTER) &&
2053 !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2055 ((iter->flags & FTRACE_ITER_NOTRACE) &&
2056 !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2058 ((iter->flags & FTRACE_ITER_ENABLED) &&
2059 !(rec->flags & ~FTRACE_FL_MASK))) {
2061 rec = NULL;
2062 goto retry;
2066 if (!rec)
2067 return t_hash_start(m, pos);
2069 iter->func = rec;
2071 return iter;
2074 static void reset_iter_read(struct ftrace_iterator *iter)
2076 iter->pos = 0;
2077 iter->func_pos = 0;
2078 iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
2081 static void *t_start(struct seq_file *m, loff_t *pos)
2083 struct ftrace_iterator *iter = m->private;
2084 struct ftrace_ops *ops = &global_ops;
2085 void *p = NULL;
2086 loff_t l;
2088 mutex_lock(&ftrace_lock);
2090 if (unlikely(ftrace_disabled))
2091 return NULL;
2094 * If an lseek was done, then reset and start from beginning.
2096 if (*pos < iter->pos)
2097 reset_iter_read(iter);
2100 * For set_ftrace_filter reading, if we have the filter
2101 * off, we can short cut and just print out that all
2102 * functions are enabled.
2104 if (iter->flags & FTRACE_ITER_FILTER && !ops->filter_hash->count) {
2105 if (*pos > 0)
2106 return t_hash_start(m, pos);
2107 iter->flags |= FTRACE_ITER_PRINTALL;
2108 /* reset in case of seek/pread */
2109 iter->flags &= ~FTRACE_ITER_HASH;
2110 return iter;
2113 if (iter->flags & FTRACE_ITER_HASH)
2114 return t_hash_start(m, pos);
2117 * Unfortunately, we need to restart at ftrace_pages_start
2118 * every time we let go of the ftrace_mutex. This is because
2119 * those pointers can change without the lock.
2121 iter->pg = ftrace_pages_start;
2122 iter->idx = 0;
2123 for (l = 0; l <= *pos; ) {
2124 p = t_next(m, p, &l);
2125 if (!p)
2126 break;
2129 if (!p) {
2130 if (iter->flags & FTRACE_ITER_FILTER)
2131 return t_hash_start(m, pos);
2133 return NULL;
2136 return iter;
2139 static void t_stop(struct seq_file *m, void *p)
2141 mutex_unlock(&ftrace_lock);
2144 static int t_show(struct seq_file *m, void *v)
2146 struct ftrace_iterator *iter = m->private;
2147 struct dyn_ftrace *rec;
2149 if (iter->flags & FTRACE_ITER_HASH)
2150 return t_hash_show(m, iter);
2152 if (iter->flags & FTRACE_ITER_PRINTALL) {
2153 seq_printf(m, "#### all functions enabled ####\n");
2154 return 0;
2157 rec = iter->func;
2159 if (!rec)
2160 return 0;
2162 seq_printf(m, "%ps", (void *)rec->ip);
2163 if (iter->flags & FTRACE_ITER_ENABLED)
2164 seq_printf(m, " (%ld)",
2165 rec->flags & ~FTRACE_FL_MASK);
2166 seq_printf(m, "\n");
2168 return 0;
2171 static const struct seq_operations show_ftrace_seq_ops = {
2172 .start = t_start,
2173 .next = t_next,
2174 .stop = t_stop,
2175 .show = t_show,
2178 static int
2179 ftrace_avail_open(struct inode *inode, struct file *file)
2181 struct ftrace_iterator *iter;
2182 int ret;
2184 if (unlikely(ftrace_disabled))
2185 return -ENODEV;
2187 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2188 if (!iter)
2189 return -ENOMEM;
2191 iter->pg = ftrace_pages_start;
2193 ret = seq_open(file, &show_ftrace_seq_ops);
2194 if (!ret) {
2195 struct seq_file *m = file->private_data;
2197 m->private = iter;
2198 } else {
2199 kfree(iter);
2202 return ret;
2205 static int
2206 ftrace_enabled_open(struct inode *inode, struct file *file)
2208 struct ftrace_iterator *iter;
2209 int ret;
2211 if (unlikely(ftrace_disabled))
2212 return -ENODEV;
2214 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2215 if (!iter)
2216 return -ENOMEM;
2218 iter->pg = ftrace_pages_start;
2219 iter->flags = FTRACE_ITER_ENABLED;
2221 ret = seq_open(file, &show_ftrace_seq_ops);
2222 if (!ret) {
2223 struct seq_file *m = file->private_data;
2225 m->private = iter;
2226 } else {
2227 kfree(iter);
2230 return ret;
2233 static void ftrace_filter_reset(struct ftrace_hash *hash)
2235 mutex_lock(&ftrace_lock);
2236 ftrace_hash_clear(hash);
2237 mutex_unlock(&ftrace_lock);
2240 static int
2241 ftrace_regex_open(struct ftrace_ops *ops, int flag,
2242 struct inode *inode, struct file *file)
2244 struct ftrace_iterator *iter;
2245 struct ftrace_hash *hash;
2246 int ret = 0;
2248 if (unlikely(ftrace_disabled))
2249 return -ENODEV;
2251 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2252 if (!iter)
2253 return -ENOMEM;
2255 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2256 kfree(iter);
2257 return -ENOMEM;
2260 if (flag & FTRACE_ITER_NOTRACE)
2261 hash = ops->notrace_hash;
2262 else
2263 hash = ops->filter_hash;
2265 iter->ops = ops;
2266 iter->flags = flag;
2268 if (file->f_mode & FMODE_WRITE) {
2269 mutex_lock(&ftrace_lock);
2270 iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2271 mutex_unlock(&ftrace_lock);
2273 if (!iter->hash) {
2274 trace_parser_put(&iter->parser);
2275 kfree(iter);
2276 return -ENOMEM;
2280 mutex_lock(&ftrace_regex_lock);
2282 if ((file->f_mode & FMODE_WRITE) &&
2283 (file->f_flags & O_TRUNC))
2284 ftrace_filter_reset(iter->hash);
2286 if (file->f_mode & FMODE_READ) {
2287 iter->pg = ftrace_pages_start;
2289 ret = seq_open(file, &show_ftrace_seq_ops);
2290 if (!ret) {
2291 struct seq_file *m = file->private_data;
2292 m->private = iter;
2293 } else {
2294 /* Failed */
2295 free_ftrace_hash(iter->hash);
2296 trace_parser_put(&iter->parser);
2297 kfree(iter);
2299 } else
2300 file->private_data = iter;
2301 mutex_unlock(&ftrace_regex_lock);
2303 return ret;
2306 static int
2307 ftrace_filter_open(struct inode *inode, struct file *file)
2309 return ftrace_regex_open(&global_ops, FTRACE_ITER_FILTER,
2310 inode, file);
2313 static int
2314 ftrace_notrace_open(struct inode *inode, struct file *file)
2316 return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2317 inode, file);
2320 static loff_t
2321 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
2323 loff_t ret;
2325 if (file->f_mode & FMODE_READ)
2326 ret = seq_lseek(file, offset, origin);
2327 else
2328 file->f_pos = ret = 1;
2330 return ret;
2333 static int ftrace_match(char *str, char *regex, int len, int type)
2335 int matched = 0;
2336 int slen;
2338 switch (type) {
2339 case MATCH_FULL:
2340 if (strcmp(str, regex) == 0)
2341 matched = 1;
2342 break;
2343 case MATCH_FRONT_ONLY:
2344 if (strncmp(str, regex, len) == 0)
2345 matched = 1;
2346 break;
2347 case MATCH_MIDDLE_ONLY:
2348 if (strstr(str, regex))
2349 matched = 1;
2350 break;
2351 case MATCH_END_ONLY:
2352 slen = strlen(str);
2353 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2354 matched = 1;
2355 break;
2358 return matched;
2361 static int
2362 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2364 struct ftrace_func_entry *entry;
2365 int ret = 0;
2367 entry = ftrace_lookup_ip(hash, rec->ip);
2368 if (not) {
2369 /* Do nothing if it doesn't exist */
2370 if (!entry)
2371 return 0;
2373 free_hash_entry(hash, entry);
2374 } else {
2375 /* Do nothing if it exists */
2376 if (entry)
2377 return 0;
2379 ret = add_hash_entry(hash, rec->ip);
2381 return ret;
2384 static int
2385 ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2386 char *regex, int len, int type)
2388 char str[KSYM_SYMBOL_LEN];
2389 char *modname;
2391 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2393 if (mod) {
2394 /* module lookup requires matching the module */
2395 if (!modname || strcmp(modname, mod))
2396 return 0;
2398 /* blank search means to match all funcs in the mod */
2399 if (!len)
2400 return 1;
2403 return ftrace_match(str, regex, len, type);
2406 static int
2407 match_records(struct ftrace_hash *hash, char *buff,
2408 int len, char *mod, int not)
2410 unsigned search_len = 0;
2411 struct ftrace_page *pg;
2412 struct dyn_ftrace *rec;
2413 int type = MATCH_FULL;
2414 char *search = buff;
2415 int found = 0;
2416 int ret;
2418 if (len) {
2419 type = filter_parse_regex(buff, len, &search, &not);
2420 search_len = strlen(search);
2423 mutex_lock(&ftrace_lock);
2425 if (unlikely(ftrace_disabled))
2426 goto out_unlock;
2428 do_for_each_ftrace_rec(pg, rec) {
2430 if (ftrace_match_record(rec, mod, search, search_len, type)) {
2431 ret = enter_record(hash, rec, not);
2432 if (ret < 0) {
2433 found = ret;
2434 goto out_unlock;
2436 found = 1;
2438 } while_for_each_ftrace_rec();
2439 out_unlock:
2440 mutex_unlock(&ftrace_lock);
2442 return found;
2445 static int
2446 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2448 return match_records(hash, buff, len, NULL, 0);
2451 static int
2452 ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2454 int not = 0;
2456 /* blank or '*' mean the same */
2457 if (strcmp(buff, "*") == 0)
2458 buff[0] = 0;
2460 /* handle the case of 'dont filter this module' */
2461 if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2462 buff[0] = 0;
2463 not = 1;
2466 return match_records(hash, buff, strlen(buff), mod, not);
2470 * We register the module command as a template to show others how
2471 * to register the a command as well.
2474 static int
2475 ftrace_mod_callback(struct ftrace_hash *hash,
2476 char *func, char *cmd, char *param, int enable)
2478 char *mod;
2479 int ret = -EINVAL;
2482 * cmd == 'mod' because we only registered this func
2483 * for the 'mod' ftrace_func_command.
2484 * But if you register one func with multiple commands,
2485 * you can tell which command was used by the cmd
2486 * parameter.
2489 /* we must have a module name */
2490 if (!param)
2491 return ret;
2493 mod = strsep(&param, ":");
2494 if (!strlen(mod))
2495 return ret;
2497 ret = ftrace_match_module_records(hash, func, mod);
2498 if (!ret)
2499 ret = -EINVAL;
2500 if (ret < 0)
2501 return ret;
2503 return 0;
2506 static struct ftrace_func_command ftrace_mod_cmd = {
2507 .name = "mod",
2508 .func = ftrace_mod_callback,
2511 static int __init ftrace_mod_cmd_init(void)
2513 return register_ftrace_command(&ftrace_mod_cmd);
2515 device_initcall(ftrace_mod_cmd_init);
2517 static void
2518 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
2520 struct ftrace_func_probe *entry;
2521 struct hlist_head *hhd;
2522 struct hlist_node *n;
2523 unsigned long key;
2525 key = hash_long(ip, FTRACE_HASH_BITS);
2527 hhd = &ftrace_func_hash[key];
2529 if (hlist_empty(hhd))
2530 return;
2533 * Disable preemption for these calls to prevent a RCU grace
2534 * period. This syncs the hash iteration and freeing of items
2535 * on the hash. rcu_read_lock is too dangerous here.
2537 preempt_disable_notrace();
2538 hlist_for_each_entry_rcu(entry, n, hhd, node) {
2539 if (entry->ip == ip)
2540 entry->ops->func(ip, parent_ip, &entry->data);
2542 preempt_enable_notrace();
2545 static struct ftrace_ops trace_probe_ops __read_mostly =
2547 .func = function_trace_probe_call,
2550 static int ftrace_probe_registered;
2552 static void __enable_ftrace_function_probe(void)
2554 int ret;
2555 int i;
2557 if (ftrace_probe_registered)
2558 return;
2560 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2561 struct hlist_head *hhd = &ftrace_func_hash[i];
2562 if (hhd->first)
2563 break;
2565 /* Nothing registered? */
2566 if (i == FTRACE_FUNC_HASHSIZE)
2567 return;
2569 ret = __register_ftrace_function(&trace_probe_ops);
2570 if (!ret)
2571 ret = ftrace_startup(&trace_probe_ops, 0);
2573 ftrace_probe_registered = 1;
2576 static void __disable_ftrace_function_probe(void)
2578 int ret;
2579 int i;
2581 if (!ftrace_probe_registered)
2582 return;
2584 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2585 struct hlist_head *hhd = &ftrace_func_hash[i];
2586 if (hhd->first)
2587 return;
2590 /* no more funcs left */
2591 ret = __unregister_ftrace_function(&trace_probe_ops);
2592 if (!ret)
2593 ftrace_shutdown(&trace_probe_ops, 0);
2595 ftrace_probe_registered = 0;
2599 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2601 struct ftrace_func_probe *entry =
2602 container_of(rhp, struct ftrace_func_probe, rcu);
2604 if (entry->ops->free)
2605 entry->ops->free(&entry->data);
2606 kfree(entry);
2611 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2612 void *data)
2614 struct ftrace_func_probe *entry;
2615 struct ftrace_page *pg;
2616 struct dyn_ftrace *rec;
2617 int type, len, not;
2618 unsigned long key;
2619 int count = 0;
2620 char *search;
2622 type = filter_parse_regex(glob, strlen(glob), &search, &not);
2623 len = strlen(search);
2625 /* we do not support '!' for function probes */
2626 if (WARN_ON(not))
2627 return -EINVAL;
2629 mutex_lock(&ftrace_lock);
2631 if (unlikely(ftrace_disabled))
2632 goto out_unlock;
2634 do_for_each_ftrace_rec(pg, rec) {
2636 if (!ftrace_match_record(rec, NULL, search, len, type))
2637 continue;
2639 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2640 if (!entry) {
2641 /* If we did not process any, then return error */
2642 if (!count)
2643 count = -ENOMEM;
2644 goto out_unlock;
2647 count++;
2649 entry->data = data;
2652 * The caller might want to do something special
2653 * for each function we find. We call the callback
2654 * to give the caller an opportunity to do so.
2656 if (ops->callback) {
2657 if (ops->callback(rec->ip, &entry->data) < 0) {
2658 /* caller does not like this func */
2659 kfree(entry);
2660 continue;
2664 entry->ops = ops;
2665 entry->ip = rec->ip;
2667 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2668 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2670 } while_for_each_ftrace_rec();
2671 __enable_ftrace_function_probe();
2673 out_unlock:
2674 mutex_unlock(&ftrace_lock);
2676 return count;
2679 enum {
2680 PROBE_TEST_FUNC = 1,
2681 PROBE_TEST_DATA = 2
2684 static void
2685 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2686 void *data, int flags)
2688 struct ftrace_func_probe *entry;
2689 struct hlist_node *n, *tmp;
2690 char str[KSYM_SYMBOL_LEN];
2691 int type = MATCH_FULL;
2692 int i, len = 0;
2693 char *search;
2695 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2696 glob = NULL;
2697 else if (glob) {
2698 int not;
2700 type = filter_parse_regex(glob, strlen(glob), &search, &not);
2701 len = strlen(search);
2703 /* we do not support '!' for function probes */
2704 if (WARN_ON(not))
2705 return;
2708 mutex_lock(&ftrace_lock);
2709 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2710 struct hlist_head *hhd = &ftrace_func_hash[i];
2712 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2714 /* break up if statements for readability */
2715 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2716 continue;
2718 if ((flags & PROBE_TEST_DATA) && entry->data != data)
2719 continue;
2721 /* do this last, since it is the most expensive */
2722 if (glob) {
2723 kallsyms_lookup(entry->ip, NULL, NULL,
2724 NULL, str);
2725 if (!ftrace_match(str, glob, len, type))
2726 continue;
2729 hlist_del(&entry->node);
2730 call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2733 __disable_ftrace_function_probe();
2734 mutex_unlock(&ftrace_lock);
2737 void
2738 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2739 void *data)
2741 __unregister_ftrace_function_probe(glob, ops, data,
2742 PROBE_TEST_FUNC | PROBE_TEST_DATA);
2745 void
2746 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2748 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2751 void unregister_ftrace_function_probe_all(char *glob)
2753 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2756 static LIST_HEAD(ftrace_commands);
2757 static DEFINE_MUTEX(ftrace_cmd_mutex);
2759 int register_ftrace_command(struct ftrace_func_command *cmd)
2761 struct ftrace_func_command *p;
2762 int ret = 0;
2764 mutex_lock(&ftrace_cmd_mutex);
2765 list_for_each_entry(p, &ftrace_commands, list) {
2766 if (strcmp(cmd->name, p->name) == 0) {
2767 ret = -EBUSY;
2768 goto out_unlock;
2771 list_add(&cmd->list, &ftrace_commands);
2772 out_unlock:
2773 mutex_unlock(&ftrace_cmd_mutex);
2775 return ret;
2778 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2780 struct ftrace_func_command *p, *n;
2781 int ret = -ENODEV;
2783 mutex_lock(&ftrace_cmd_mutex);
2784 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2785 if (strcmp(cmd->name, p->name) == 0) {
2786 ret = 0;
2787 list_del_init(&p->list);
2788 goto out_unlock;
2791 out_unlock:
2792 mutex_unlock(&ftrace_cmd_mutex);
2794 return ret;
2797 static int ftrace_process_regex(struct ftrace_hash *hash,
2798 char *buff, int len, int enable)
2800 char *func, *command, *next = buff;
2801 struct ftrace_func_command *p;
2802 int ret = -EINVAL;
2804 func = strsep(&next, ":");
2806 if (!next) {
2807 ret = ftrace_match_records(hash, func, len);
2808 if (!ret)
2809 ret = -EINVAL;
2810 if (ret < 0)
2811 return ret;
2812 return 0;
2815 /* command found */
2817 command = strsep(&next, ":");
2819 mutex_lock(&ftrace_cmd_mutex);
2820 list_for_each_entry(p, &ftrace_commands, list) {
2821 if (strcmp(p->name, command) == 0) {
2822 ret = p->func(hash, func, command, next, enable);
2823 goto out_unlock;
2826 out_unlock:
2827 mutex_unlock(&ftrace_cmd_mutex);
2829 return ret;
2832 static ssize_t
2833 ftrace_regex_write(struct file *file, const char __user *ubuf,
2834 size_t cnt, loff_t *ppos, int enable)
2836 struct ftrace_iterator *iter;
2837 struct trace_parser *parser;
2838 ssize_t ret, read;
2840 if (!cnt)
2841 return 0;
2843 mutex_lock(&ftrace_regex_lock);
2845 ret = -ENODEV;
2846 if (unlikely(ftrace_disabled))
2847 goto out_unlock;
2849 if (file->f_mode & FMODE_READ) {
2850 struct seq_file *m = file->private_data;
2851 iter = m->private;
2852 } else
2853 iter = file->private_data;
2855 parser = &iter->parser;
2856 read = trace_get_user(parser, ubuf, cnt, ppos);
2858 if (read >= 0 && trace_parser_loaded(parser) &&
2859 !trace_parser_cont(parser)) {
2860 ret = ftrace_process_regex(iter->hash, parser->buffer,
2861 parser->idx, enable);
2862 trace_parser_clear(parser);
2863 if (ret)
2864 goto out_unlock;
2867 ret = read;
2868 out_unlock:
2869 mutex_unlock(&ftrace_regex_lock);
2871 return ret;
2874 static ssize_t
2875 ftrace_filter_write(struct file *file, const char __user *ubuf,
2876 size_t cnt, loff_t *ppos)
2878 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2881 static ssize_t
2882 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2883 size_t cnt, loff_t *ppos)
2885 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2888 static int
2889 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
2890 int reset, int enable)
2892 struct ftrace_hash **orig_hash;
2893 struct ftrace_hash *hash;
2894 int ret;
2896 /* All global ops uses the global ops filters */
2897 if (ops->flags & FTRACE_OPS_FL_GLOBAL)
2898 ops = &global_ops;
2900 if (unlikely(ftrace_disabled))
2901 return -ENODEV;
2903 if (enable)
2904 orig_hash = &ops->filter_hash;
2905 else
2906 orig_hash = &ops->notrace_hash;
2908 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
2909 if (!hash)
2910 return -ENOMEM;
2912 mutex_lock(&ftrace_regex_lock);
2913 if (reset)
2914 ftrace_filter_reset(hash);
2915 if (buf)
2916 ftrace_match_records(hash, buf, len);
2918 mutex_lock(&ftrace_lock);
2919 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
2920 if (!ret && ops->flags & FTRACE_OPS_FL_ENABLED
2921 && ftrace_enabled)
2922 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2924 mutex_unlock(&ftrace_lock);
2926 mutex_unlock(&ftrace_regex_lock);
2928 free_ftrace_hash(hash);
2929 return ret;
2933 * ftrace_set_filter - set a function to filter on in ftrace
2934 * @ops - the ops to set the filter with
2935 * @buf - the string that holds the function filter text.
2936 * @len - the length of the string.
2937 * @reset - non zero to reset all filters before applying this filter.
2939 * Filters denote which functions should be enabled when tracing is enabled.
2940 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2942 void ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
2943 int len, int reset)
2945 ftrace_set_regex(ops, buf, len, reset, 1);
2947 EXPORT_SYMBOL_GPL(ftrace_set_filter);
2950 * ftrace_set_notrace - set a function to not trace in ftrace
2951 * @ops - the ops to set the notrace filter with
2952 * @buf - the string that holds the function notrace text.
2953 * @len - the length of the string.
2954 * @reset - non zero to reset all filters before applying this filter.
2956 * Notrace Filters denote which functions should not be enabled when tracing
2957 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2958 * for tracing.
2960 void ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
2961 int len, int reset)
2963 ftrace_set_regex(ops, buf, len, reset, 0);
2965 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
2967 * ftrace_set_filter - set a function to filter on in ftrace
2968 * @ops - the ops to set the filter with
2969 * @buf - the string that holds the function filter text.
2970 * @len - the length of the string.
2971 * @reset - non zero to reset all filters before applying this filter.
2973 * Filters denote which functions should be enabled when tracing is enabled.
2974 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2976 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
2978 ftrace_set_regex(&global_ops, buf, len, reset, 1);
2980 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
2983 * ftrace_set_notrace - set a function to not trace in ftrace
2984 * @ops - the ops to set the notrace filter with
2985 * @buf - the string that holds the function notrace text.
2986 * @len - the length of the string.
2987 * @reset - non zero to reset all filters before applying this filter.
2989 * Notrace Filters denote which functions should not be enabled when tracing
2990 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2991 * for tracing.
2993 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
2995 ftrace_set_regex(&global_ops, buf, len, reset, 0);
2997 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3000 * command line interface to allow users to set filters on boot up.
3002 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
3003 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3004 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3006 static int __init set_ftrace_notrace(char *str)
3008 strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3009 return 1;
3011 __setup("ftrace_notrace=", set_ftrace_notrace);
3013 static int __init set_ftrace_filter(char *str)
3015 strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3016 return 1;
3018 __setup("ftrace_filter=", set_ftrace_filter);
3020 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3021 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3022 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
3024 static int __init set_graph_function(char *str)
3026 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3027 return 1;
3029 __setup("ftrace_graph_filter=", set_graph_function);
3031 static void __init set_ftrace_early_graph(char *buf)
3033 int ret;
3034 char *func;
3036 while (buf) {
3037 func = strsep(&buf, ",");
3038 /* we allow only one expression at a time */
3039 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3040 func);
3041 if (ret)
3042 printk(KERN_DEBUG "ftrace: function %s not "
3043 "traceable\n", func);
3046 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3048 static void __init
3049 set_ftrace_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3051 char *func;
3053 while (buf) {
3054 func = strsep(&buf, ",");
3055 ftrace_set_regex(ops, func, strlen(func), 0, enable);
3059 static void __init set_ftrace_early_filters(void)
3061 if (ftrace_filter_buf[0])
3062 set_ftrace_early_filter(&global_ops, ftrace_filter_buf, 1);
3063 if (ftrace_notrace_buf[0])
3064 set_ftrace_early_filter(&global_ops, ftrace_notrace_buf, 0);
3065 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3066 if (ftrace_graph_buf[0])
3067 set_ftrace_early_graph(ftrace_graph_buf);
3068 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3071 static int
3072 ftrace_regex_release(struct inode *inode, struct file *file)
3074 struct seq_file *m = (struct seq_file *)file->private_data;
3075 struct ftrace_iterator *iter;
3076 struct ftrace_hash **orig_hash;
3077 struct trace_parser *parser;
3078 int filter_hash;
3079 int ret;
3081 mutex_lock(&ftrace_regex_lock);
3082 if (file->f_mode & FMODE_READ) {
3083 iter = m->private;
3085 seq_release(inode, file);
3086 } else
3087 iter = file->private_data;
3089 parser = &iter->parser;
3090 if (trace_parser_loaded(parser)) {
3091 parser->buffer[parser->idx] = 0;
3092 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3095 trace_parser_put(parser);
3097 if (file->f_mode & FMODE_WRITE) {
3098 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3100 if (filter_hash)
3101 orig_hash = &iter->ops->filter_hash;
3102 else
3103 orig_hash = &iter->ops->notrace_hash;
3105 mutex_lock(&ftrace_lock);
3106 ret = ftrace_hash_move(iter->ops, filter_hash,
3107 orig_hash, iter->hash);
3108 if (!ret && (iter->ops->flags & FTRACE_OPS_FL_ENABLED)
3109 && ftrace_enabled)
3110 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
3112 mutex_unlock(&ftrace_lock);
3114 free_ftrace_hash(iter->hash);
3115 kfree(iter);
3117 mutex_unlock(&ftrace_regex_lock);
3118 return 0;
3121 static const struct file_operations ftrace_avail_fops = {
3122 .open = ftrace_avail_open,
3123 .read = seq_read,
3124 .llseek = seq_lseek,
3125 .release = seq_release_private,
3128 static const struct file_operations ftrace_enabled_fops = {
3129 .open = ftrace_enabled_open,
3130 .read = seq_read,
3131 .llseek = seq_lseek,
3132 .release = seq_release_private,
3135 static const struct file_operations ftrace_filter_fops = {
3136 .open = ftrace_filter_open,
3137 .read = seq_read,
3138 .write = ftrace_filter_write,
3139 .llseek = ftrace_regex_lseek,
3140 .release = ftrace_regex_release,
3143 static const struct file_operations ftrace_notrace_fops = {
3144 .open = ftrace_notrace_open,
3145 .read = seq_read,
3146 .write = ftrace_notrace_write,
3147 .llseek = ftrace_regex_lseek,
3148 .release = ftrace_regex_release,
3151 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3153 static DEFINE_MUTEX(graph_lock);
3155 int ftrace_graph_count;
3156 int ftrace_graph_filter_enabled;
3157 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3159 static void *
3160 __g_next(struct seq_file *m, loff_t *pos)
3162 if (*pos >= ftrace_graph_count)
3163 return NULL;
3164 return &ftrace_graph_funcs[*pos];
3167 static void *
3168 g_next(struct seq_file *m, void *v, loff_t *pos)
3170 (*pos)++;
3171 return __g_next(m, pos);
3174 static void *g_start(struct seq_file *m, loff_t *pos)
3176 mutex_lock(&graph_lock);
3178 /* Nothing, tell g_show to print all functions are enabled */
3179 if (!ftrace_graph_filter_enabled && !*pos)
3180 return (void *)1;
3182 return __g_next(m, pos);
3185 static void g_stop(struct seq_file *m, void *p)
3187 mutex_unlock(&graph_lock);
3190 static int g_show(struct seq_file *m, void *v)
3192 unsigned long *ptr = v;
3194 if (!ptr)
3195 return 0;
3197 if (ptr == (unsigned long *)1) {
3198 seq_printf(m, "#### all functions enabled ####\n");
3199 return 0;
3202 seq_printf(m, "%ps\n", (void *)*ptr);
3204 return 0;
3207 static const struct seq_operations ftrace_graph_seq_ops = {
3208 .start = g_start,
3209 .next = g_next,
3210 .stop = g_stop,
3211 .show = g_show,
3214 static int
3215 ftrace_graph_open(struct inode *inode, struct file *file)
3217 int ret = 0;
3219 if (unlikely(ftrace_disabled))
3220 return -ENODEV;
3222 mutex_lock(&graph_lock);
3223 if ((file->f_mode & FMODE_WRITE) &&
3224 (file->f_flags & O_TRUNC)) {
3225 ftrace_graph_filter_enabled = 0;
3226 ftrace_graph_count = 0;
3227 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
3229 mutex_unlock(&graph_lock);
3231 if (file->f_mode & FMODE_READ)
3232 ret = seq_open(file, &ftrace_graph_seq_ops);
3234 return ret;
3237 static int
3238 ftrace_graph_release(struct inode *inode, struct file *file)
3240 if (file->f_mode & FMODE_READ)
3241 seq_release(inode, file);
3242 return 0;
3245 static int
3246 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
3248 struct dyn_ftrace *rec;
3249 struct ftrace_page *pg;
3250 int search_len;
3251 int fail = 1;
3252 int type, not;
3253 char *search;
3254 bool exists;
3255 int i;
3257 /* decode regex */
3258 type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
3259 if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
3260 return -EBUSY;
3262 search_len = strlen(search);
3264 mutex_lock(&ftrace_lock);
3266 if (unlikely(ftrace_disabled)) {
3267 mutex_unlock(&ftrace_lock);
3268 return -ENODEV;
3271 do_for_each_ftrace_rec(pg, rec) {
3273 if (rec->flags & FTRACE_FL_FREE)
3274 continue;
3276 if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3277 /* if it is in the array */
3278 exists = false;
3279 for (i = 0; i < *idx; i++) {
3280 if (array[i] == rec->ip) {
3281 exists = true;
3282 break;
3286 if (!not) {
3287 fail = 0;
3288 if (!exists) {
3289 array[(*idx)++] = rec->ip;
3290 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
3291 goto out;
3293 } else {
3294 if (exists) {
3295 array[i] = array[--(*idx)];
3296 array[*idx] = 0;
3297 fail = 0;
3301 } while_for_each_ftrace_rec();
3302 out:
3303 mutex_unlock(&ftrace_lock);
3305 if (fail)
3306 return -EINVAL;
3308 ftrace_graph_filter_enabled = 1;
3309 return 0;
3312 static ssize_t
3313 ftrace_graph_write(struct file *file, const char __user *ubuf,
3314 size_t cnt, loff_t *ppos)
3316 struct trace_parser parser;
3317 ssize_t read, ret;
3319 if (!cnt)
3320 return 0;
3322 mutex_lock(&graph_lock);
3324 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
3325 ret = -ENOMEM;
3326 goto out_unlock;
3329 read = trace_get_user(&parser, ubuf, cnt, ppos);
3331 if (read >= 0 && trace_parser_loaded((&parser))) {
3332 parser.buffer[parser.idx] = 0;
3334 /* we allow only one expression at a time */
3335 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3336 parser.buffer);
3337 if (ret)
3338 goto out_free;
3341 ret = read;
3343 out_free:
3344 trace_parser_put(&parser);
3345 out_unlock:
3346 mutex_unlock(&graph_lock);
3348 return ret;
3351 static const struct file_operations ftrace_graph_fops = {
3352 .open = ftrace_graph_open,
3353 .read = seq_read,
3354 .write = ftrace_graph_write,
3355 .release = ftrace_graph_release,
3356 .llseek = seq_lseek,
3358 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3360 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
3363 trace_create_file("available_filter_functions", 0444,
3364 d_tracer, NULL, &ftrace_avail_fops);
3366 trace_create_file("enabled_functions", 0444,
3367 d_tracer, NULL, &ftrace_enabled_fops);
3369 trace_create_file("set_ftrace_filter", 0644, d_tracer,
3370 NULL, &ftrace_filter_fops);
3372 trace_create_file("set_ftrace_notrace", 0644, d_tracer,
3373 NULL, &ftrace_notrace_fops);
3375 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3376 trace_create_file("set_graph_function", 0444, d_tracer,
3377 NULL,
3378 &ftrace_graph_fops);
3379 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3381 return 0;
3384 static int ftrace_process_locs(struct module *mod,
3385 unsigned long *start,
3386 unsigned long *end)
3388 unsigned long *p;
3389 unsigned long addr;
3390 unsigned long flags = 0; /* Shut up gcc */
3392 mutex_lock(&ftrace_lock);
3393 p = start;
3394 while (p < end) {
3395 addr = ftrace_call_adjust(*p++);
3397 * Some architecture linkers will pad between
3398 * the different mcount_loc sections of different
3399 * object files to satisfy alignments.
3400 * Skip any NULL pointers.
3402 if (!addr)
3403 continue;
3404 ftrace_record_ip(addr);
3408 * We only need to disable interrupts on start up
3409 * because we are modifying code that an interrupt
3410 * may execute, and the modification is not atomic.
3411 * But for modules, nothing runs the code we modify
3412 * until we are finished with it, and there's no
3413 * reason to cause large interrupt latencies while we do it.
3415 if (!mod)
3416 local_irq_save(flags);
3417 ftrace_update_code(mod);
3418 if (!mod)
3419 local_irq_restore(flags);
3420 mutex_unlock(&ftrace_lock);
3422 return 0;
3425 #ifdef CONFIG_MODULES
3426 void ftrace_release_mod(struct module *mod)
3428 struct dyn_ftrace *rec;
3429 struct ftrace_page *pg;
3431 mutex_lock(&ftrace_lock);
3433 if (ftrace_disabled)
3434 goto out_unlock;
3436 do_for_each_ftrace_rec(pg, rec) {
3437 if (within_module_core(rec->ip, mod)) {
3439 * rec->ip is changed in ftrace_free_rec()
3440 * It should not between s and e if record was freed.
3442 FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
3443 ftrace_free_rec(rec);
3445 } while_for_each_ftrace_rec();
3446 out_unlock:
3447 mutex_unlock(&ftrace_lock);
3450 static void ftrace_init_module(struct module *mod,
3451 unsigned long *start, unsigned long *end)
3453 if (ftrace_disabled || start == end)
3454 return;
3455 ftrace_process_locs(mod, start, end);
3458 static int ftrace_module_notify(struct notifier_block *self,
3459 unsigned long val, void *data)
3461 struct module *mod = data;
3463 switch (val) {
3464 case MODULE_STATE_COMING:
3465 ftrace_init_module(mod, mod->ftrace_callsites,
3466 mod->ftrace_callsites +
3467 mod->num_ftrace_callsites);
3468 break;
3469 case MODULE_STATE_GOING:
3470 ftrace_release_mod(mod);
3471 break;
3474 return 0;
3476 #else
3477 static int ftrace_module_notify(struct notifier_block *self,
3478 unsigned long val, void *data)
3480 return 0;
3482 #endif /* CONFIG_MODULES */
3484 struct notifier_block ftrace_module_nb = {
3485 .notifier_call = ftrace_module_notify,
3486 .priority = 0,
3489 extern unsigned long __start_mcount_loc[];
3490 extern unsigned long __stop_mcount_loc[];
3492 void __init ftrace_init(void)
3494 unsigned long count, addr, flags;
3495 int ret;
3497 /* Keep the ftrace pointer to the stub */
3498 addr = (unsigned long)ftrace_stub;
3500 local_irq_save(flags);
3501 ftrace_dyn_arch_init(&addr);
3502 local_irq_restore(flags);
3504 /* ftrace_dyn_arch_init places the return code in addr */
3505 if (addr)
3506 goto failed;
3508 count = __stop_mcount_loc - __start_mcount_loc;
3510 ret = ftrace_dyn_table_alloc(count);
3511 if (ret)
3512 goto failed;
3514 last_ftrace_enabled = ftrace_enabled = 1;
3516 ret = ftrace_process_locs(NULL,
3517 __start_mcount_loc,
3518 __stop_mcount_loc);
3520 ret = register_module_notifier(&ftrace_module_nb);
3521 if (ret)
3522 pr_warning("Failed to register trace ftrace module notifier\n");
3524 set_ftrace_early_filters();
3526 return;
3527 failed:
3528 ftrace_disabled = 1;
3531 #else
3533 static struct ftrace_ops global_ops = {
3534 .func = ftrace_stub,
3537 static int __init ftrace_nodyn_init(void)
3539 ftrace_enabled = 1;
3540 return 0;
3542 device_initcall(ftrace_nodyn_init);
3544 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
3545 static inline void ftrace_startup_enable(int command) { }
3546 /* Keep as macros so we do not need to define the commands */
3547 # define ftrace_startup(ops, command) \
3548 ({ \
3549 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
3550 0; \
3552 # define ftrace_shutdown(ops, command) do { } while (0)
3553 # define ftrace_startup_sysctl() do { } while (0)
3554 # define ftrace_shutdown_sysctl() do { } while (0)
3556 static inline int
3557 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
3559 return 1;
3562 #endif /* CONFIG_DYNAMIC_FTRACE */
3564 static void
3565 ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
3567 struct ftrace_ops *op;
3569 if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT)))
3570 return;
3572 trace_recursion_set(TRACE_INTERNAL_BIT);
3574 * Some of the ops may be dynamically allocated,
3575 * they must be freed after a synchronize_sched().
3577 preempt_disable_notrace();
3578 op = rcu_dereference_raw(ftrace_ops_list);
3579 while (op != &ftrace_list_end) {
3580 if (ftrace_ops_test(op, ip))
3581 op->func(ip, parent_ip);
3582 op = rcu_dereference_raw(op->next);
3584 preempt_enable_notrace();
3585 trace_recursion_clear(TRACE_INTERNAL_BIT);
3588 static void clear_ftrace_swapper(void)
3590 struct task_struct *p;
3591 int cpu;
3593 get_online_cpus();
3594 for_each_online_cpu(cpu) {
3595 p = idle_task(cpu);
3596 clear_tsk_trace_trace(p);
3598 put_online_cpus();
3601 static void set_ftrace_swapper(void)
3603 struct task_struct *p;
3604 int cpu;
3606 get_online_cpus();
3607 for_each_online_cpu(cpu) {
3608 p = idle_task(cpu);
3609 set_tsk_trace_trace(p);
3611 put_online_cpus();
3614 static void clear_ftrace_pid(struct pid *pid)
3616 struct task_struct *p;
3618 rcu_read_lock();
3619 do_each_pid_task(pid, PIDTYPE_PID, p) {
3620 clear_tsk_trace_trace(p);
3621 } while_each_pid_task(pid, PIDTYPE_PID, p);
3622 rcu_read_unlock();
3624 put_pid(pid);
3627 static void set_ftrace_pid(struct pid *pid)
3629 struct task_struct *p;
3631 rcu_read_lock();
3632 do_each_pid_task(pid, PIDTYPE_PID, p) {
3633 set_tsk_trace_trace(p);
3634 } while_each_pid_task(pid, PIDTYPE_PID, p);
3635 rcu_read_unlock();
3638 static void clear_ftrace_pid_task(struct pid *pid)
3640 if (pid == ftrace_swapper_pid)
3641 clear_ftrace_swapper();
3642 else
3643 clear_ftrace_pid(pid);
3646 static void set_ftrace_pid_task(struct pid *pid)
3648 if (pid == ftrace_swapper_pid)
3649 set_ftrace_swapper();
3650 else
3651 set_ftrace_pid(pid);
3654 static int ftrace_pid_add(int p)
3656 struct pid *pid;
3657 struct ftrace_pid *fpid;
3658 int ret = -EINVAL;
3660 mutex_lock(&ftrace_lock);
3662 if (!p)
3663 pid = ftrace_swapper_pid;
3664 else
3665 pid = find_get_pid(p);
3667 if (!pid)
3668 goto out;
3670 ret = 0;
3672 list_for_each_entry(fpid, &ftrace_pids, list)
3673 if (fpid->pid == pid)
3674 goto out_put;
3676 ret = -ENOMEM;
3678 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
3679 if (!fpid)
3680 goto out_put;
3682 list_add(&fpid->list, &ftrace_pids);
3683 fpid->pid = pid;
3685 set_ftrace_pid_task(pid);
3687 ftrace_update_pid_func();
3688 ftrace_startup_enable(0);
3690 mutex_unlock(&ftrace_lock);
3691 return 0;
3693 out_put:
3694 if (pid != ftrace_swapper_pid)
3695 put_pid(pid);
3697 out:
3698 mutex_unlock(&ftrace_lock);
3699 return ret;
3702 static void ftrace_pid_reset(void)
3704 struct ftrace_pid *fpid, *safe;
3706 mutex_lock(&ftrace_lock);
3707 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
3708 struct pid *pid = fpid->pid;
3710 clear_ftrace_pid_task(pid);
3712 list_del(&fpid->list);
3713 kfree(fpid);
3716 ftrace_update_pid_func();
3717 ftrace_startup_enable(0);
3719 mutex_unlock(&ftrace_lock);
3722 static void *fpid_start(struct seq_file *m, loff_t *pos)
3724 mutex_lock(&ftrace_lock);
3726 if (list_empty(&ftrace_pids) && (!*pos))
3727 return (void *) 1;
3729 return seq_list_start(&ftrace_pids, *pos);
3732 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
3734 if (v == (void *)1)
3735 return NULL;
3737 return seq_list_next(v, &ftrace_pids, pos);
3740 static void fpid_stop(struct seq_file *m, void *p)
3742 mutex_unlock(&ftrace_lock);
3745 static int fpid_show(struct seq_file *m, void *v)
3747 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
3749 if (v == (void *)1) {
3750 seq_printf(m, "no pid\n");
3751 return 0;
3754 if (fpid->pid == ftrace_swapper_pid)
3755 seq_printf(m, "swapper tasks\n");
3756 else
3757 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
3759 return 0;
3762 static const struct seq_operations ftrace_pid_sops = {
3763 .start = fpid_start,
3764 .next = fpid_next,
3765 .stop = fpid_stop,
3766 .show = fpid_show,
3769 static int
3770 ftrace_pid_open(struct inode *inode, struct file *file)
3772 int ret = 0;
3774 if ((file->f_mode & FMODE_WRITE) &&
3775 (file->f_flags & O_TRUNC))
3776 ftrace_pid_reset();
3778 if (file->f_mode & FMODE_READ)
3779 ret = seq_open(file, &ftrace_pid_sops);
3781 return ret;
3784 static ssize_t
3785 ftrace_pid_write(struct file *filp, const char __user *ubuf,
3786 size_t cnt, loff_t *ppos)
3788 char buf[64], *tmp;
3789 long val;
3790 int ret;
3792 if (cnt >= sizeof(buf))
3793 return -EINVAL;
3795 if (copy_from_user(&buf, ubuf, cnt))
3796 return -EFAULT;
3798 buf[cnt] = 0;
3801 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
3802 * to clean the filter quietly.
3804 tmp = strstrip(buf);
3805 if (strlen(tmp) == 0)
3806 return 1;
3808 ret = strict_strtol(tmp, 10, &val);
3809 if (ret < 0)
3810 return ret;
3812 ret = ftrace_pid_add(val);
3814 return ret ? ret : cnt;
3817 static int
3818 ftrace_pid_release(struct inode *inode, struct file *file)
3820 if (file->f_mode & FMODE_READ)
3821 seq_release(inode, file);
3823 return 0;
3826 static const struct file_operations ftrace_pid_fops = {
3827 .open = ftrace_pid_open,
3828 .write = ftrace_pid_write,
3829 .read = seq_read,
3830 .llseek = seq_lseek,
3831 .release = ftrace_pid_release,
3834 static __init int ftrace_init_debugfs(void)
3836 struct dentry *d_tracer;
3838 d_tracer = tracing_init_dentry();
3839 if (!d_tracer)
3840 return 0;
3842 ftrace_init_dyn_debugfs(d_tracer);
3844 trace_create_file("set_ftrace_pid", 0644, d_tracer,
3845 NULL, &ftrace_pid_fops);
3847 ftrace_profile_debugfs(d_tracer);
3849 return 0;
3851 fs_initcall(ftrace_init_debugfs);
3854 * ftrace_kill - kill ftrace
3856 * This function should be used by panic code. It stops ftrace
3857 * but in a not so nice way. If you need to simply kill ftrace
3858 * from a non-atomic section, use ftrace_kill.
3860 void ftrace_kill(void)
3862 ftrace_disabled = 1;
3863 ftrace_enabled = 0;
3864 clear_ftrace_function();
3868 * Test if ftrace is dead or not.
3870 int ftrace_is_dead(void)
3872 return ftrace_disabled;
3876 * register_ftrace_function - register a function for profiling
3877 * @ops - ops structure that holds the function for profiling.
3879 * Register a function to be called by all functions in the
3880 * kernel.
3882 * Note: @ops->func and all the functions it calls must be labeled
3883 * with "notrace", otherwise it will go into a
3884 * recursive loop.
3886 int register_ftrace_function(struct ftrace_ops *ops)
3888 int ret = -1;
3890 mutex_lock(&ftrace_lock);
3892 if (unlikely(ftrace_disabled))
3893 goto out_unlock;
3895 ret = __register_ftrace_function(ops);
3896 if (!ret)
3897 ret = ftrace_startup(ops, 0);
3900 out_unlock:
3901 mutex_unlock(&ftrace_lock);
3902 return ret;
3904 EXPORT_SYMBOL_GPL(register_ftrace_function);
3907 * unregister_ftrace_function - unregister a function for profiling.
3908 * @ops - ops structure that holds the function to unregister
3910 * Unregister a function that was added to be called by ftrace profiling.
3912 int unregister_ftrace_function(struct ftrace_ops *ops)
3914 int ret;
3916 mutex_lock(&ftrace_lock);
3917 ret = __unregister_ftrace_function(ops);
3918 if (!ret)
3919 ftrace_shutdown(ops, 0);
3920 mutex_unlock(&ftrace_lock);
3922 return ret;
3924 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
3927 ftrace_enable_sysctl(struct ctl_table *table, int write,
3928 void __user *buffer, size_t *lenp,
3929 loff_t *ppos)
3931 int ret = -ENODEV;
3933 mutex_lock(&ftrace_lock);
3935 if (unlikely(ftrace_disabled))
3936 goto out;
3938 ret = proc_dointvec(table, write, buffer, lenp, ppos);
3940 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3941 goto out;
3943 last_ftrace_enabled = !!ftrace_enabled;
3945 if (ftrace_enabled) {
3947 ftrace_startup_sysctl();
3949 /* we are starting ftrace again */
3950 if (ftrace_ops_list != &ftrace_list_end) {
3951 if (ftrace_ops_list->next == &ftrace_list_end)
3952 ftrace_trace_function = ftrace_ops_list->func;
3953 else
3954 ftrace_trace_function = ftrace_ops_list_func;
3957 } else {
3958 /* stopping ftrace calls (just send to ftrace_stub) */
3959 ftrace_trace_function = ftrace_stub;
3961 ftrace_shutdown_sysctl();
3964 out:
3965 mutex_unlock(&ftrace_lock);
3966 return ret;
3969 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3971 static int ftrace_graph_active;
3972 static struct notifier_block ftrace_suspend_notifier;
3974 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3976 return 0;
3979 /* The callbacks that hook a function */
3980 trace_func_graph_ret_t ftrace_graph_return =
3981 (trace_func_graph_ret_t)ftrace_stub;
3982 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3984 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3985 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3987 int i;
3988 int ret = 0;
3989 unsigned long flags;
3990 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3991 struct task_struct *g, *t;
3993 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3994 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3995 * sizeof(struct ftrace_ret_stack),
3996 GFP_KERNEL);
3997 if (!ret_stack_list[i]) {
3998 start = 0;
3999 end = i;
4000 ret = -ENOMEM;
4001 goto free;
4005 read_lock_irqsave(&tasklist_lock, flags);
4006 do_each_thread(g, t) {
4007 if (start == end) {
4008 ret = -EAGAIN;
4009 goto unlock;
4012 if (t->ret_stack == NULL) {
4013 atomic_set(&t->tracing_graph_pause, 0);
4014 atomic_set(&t->trace_overrun, 0);
4015 t->curr_ret_stack = -1;
4016 /* Make sure the tasks see the -1 first: */
4017 smp_wmb();
4018 t->ret_stack = ret_stack_list[start++];
4020 } while_each_thread(g, t);
4022 unlock:
4023 read_unlock_irqrestore(&tasklist_lock, flags);
4024 free:
4025 for (i = start; i < end; i++)
4026 kfree(ret_stack_list[i]);
4027 return ret;
4030 static void
4031 ftrace_graph_probe_sched_switch(void *ignore,
4032 struct task_struct *prev, struct task_struct *next)
4034 unsigned long long timestamp;
4035 int index;
4038 * Does the user want to count the time a function was asleep.
4039 * If so, do not update the time stamps.
4041 if (trace_flags & TRACE_ITER_SLEEP_TIME)
4042 return;
4044 timestamp = trace_clock_local();
4046 prev->ftrace_timestamp = timestamp;
4048 /* only process tasks that we timestamped */
4049 if (!next->ftrace_timestamp)
4050 return;
4053 * Update all the counters in next to make up for the
4054 * time next was sleeping.
4056 timestamp -= next->ftrace_timestamp;
4058 for (index = next->curr_ret_stack; index >= 0; index--)
4059 next->ret_stack[index].calltime += timestamp;
4062 /* Allocate a return stack for each task */
4063 static int start_graph_tracing(void)
4065 struct ftrace_ret_stack **ret_stack_list;
4066 int ret, cpu;
4068 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
4069 sizeof(struct ftrace_ret_stack *),
4070 GFP_KERNEL);
4072 if (!ret_stack_list)
4073 return -ENOMEM;
4075 /* The cpu_boot init_task->ret_stack will never be freed */
4076 for_each_online_cpu(cpu) {
4077 if (!idle_task(cpu)->ret_stack)
4078 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4081 do {
4082 ret = alloc_retstack_tasklist(ret_stack_list);
4083 } while (ret == -EAGAIN);
4085 if (!ret) {
4086 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4087 if (ret)
4088 pr_info("ftrace_graph: Couldn't activate tracepoint"
4089 " probe to kernel_sched_switch\n");
4092 kfree(ret_stack_list);
4093 return ret;
4097 * Hibernation protection.
4098 * The state of the current task is too much unstable during
4099 * suspend/restore to disk. We want to protect against that.
4101 static int
4102 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4103 void *unused)
4105 switch (state) {
4106 case PM_HIBERNATION_PREPARE:
4107 pause_graph_tracing();
4108 break;
4110 case PM_POST_HIBERNATION:
4111 unpause_graph_tracing();
4112 break;
4114 return NOTIFY_DONE;
4117 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
4118 trace_func_graph_ent_t entryfunc)
4120 int ret = 0;
4122 mutex_lock(&ftrace_lock);
4124 /* we currently allow only one tracer registered at a time */
4125 if (ftrace_graph_active) {
4126 ret = -EBUSY;
4127 goto out;
4130 ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
4131 register_pm_notifier(&ftrace_suspend_notifier);
4133 ftrace_graph_active++;
4134 ret = start_graph_tracing();
4135 if (ret) {
4136 ftrace_graph_active--;
4137 goto out;
4140 ftrace_graph_return = retfunc;
4141 ftrace_graph_entry = entryfunc;
4143 ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
4145 out:
4146 mutex_unlock(&ftrace_lock);
4147 return ret;
4150 void unregister_ftrace_graph(void)
4152 mutex_lock(&ftrace_lock);
4154 if (unlikely(!ftrace_graph_active))
4155 goto out;
4157 ftrace_graph_active--;
4158 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
4159 ftrace_graph_entry = ftrace_graph_entry_stub;
4160 ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
4161 unregister_pm_notifier(&ftrace_suspend_notifier);
4162 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4164 out:
4165 mutex_unlock(&ftrace_lock);
4168 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
4170 static void
4171 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
4173 atomic_set(&t->tracing_graph_pause, 0);
4174 atomic_set(&t->trace_overrun, 0);
4175 t->ftrace_timestamp = 0;
4176 /* make curr_ret_stack visible before we add the ret_stack */
4177 smp_wmb();
4178 t->ret_stack = ret_stack;
4182 * Allocate a return stack for the idle task. May be the first
4183 * time through, or it may be done by CPU hotplug online.
4185 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
4187 t->curr_ret_stack = -1;
4189 * The idle task has no parent, it either has its own
4190 * stack or no stack at all.
4192 if (t->ret_stack)
4193 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
4195 if (ftrace_graph_active) {
4196 struct ftrace_ret_stack *ret_stack;
4198 ret_stack = per_cpu(idle_ret_stack, cpu);
4199 if (!ret_stack) {
4200 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4201 * sizeof(struct ftrace_ret_stack),
4202 GFP_KERNEL);
4203 if (!ret_stack)
4204 return;
4205 per_cpu(idle_ret_stack, cpu) = ret_stack;
4207 graph_init_task(t, ret_stack);
4211 /* Allocate a return stack for newly created task */
4212 void ftrace_graph_init_task(struct task_struct *t)
4214 /* Make sure we do not use the parent ret_stack */
4215 t->ret_stack = NULL;
4216 t->curr_ret_stack = -1;
4218 if (ftrace_graph_active) {
4219 struct ftrace_ret_stack *ret_stack;
4221 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4222 * sizeof(struct ftrace_ret_stack),
4223 GFP_KERNEL);
4224 if (!ret_stack)
4225 return;
4226 graph_init_task(t, ret_stack);
4230 void ftrace_graph_exit_task(struct task_struct *t)
4232 struct ftrace_ret_stack *ret_stack = t->ret_stack;
4234 t->ret_stack = NULL;
4235 /* NULL must become visible to IRQs before we free it: */
4236 barrier();
4238 kfree(ret_stack);
4241 void ftrace_graph_stop(void)
4243 ftrace_stop();
4245 #endif