ftrace: Don't manipulate @pos in t_start()
[linux-2.6.git] / kernel / trace / ftrace.c
1 /*
2  * Infrastructure for profiling code inserted by 'gcc -pg'.
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally ported from the -rt patch by:
8  *   Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code in the latency_tracer, that is:
11  *
12  *  Copyright (C) 2004-2006 Ingo Molnar
13  *  Copyright (C) 2004 William Lee Irwin III
14  */
15
16 #include <linux/stop_machine.h>
17 #include <linux/clocksource.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/suspend.h>
21 #include <linux/debugfs.h>
22 #include <linux/hardirq.h>
23 #include <linux/kthread.h>
24 #include <linux/uaccess.h>
25 #include <linux/kprobes.h>
26 #include <linux/ftrace.h>
27 #include <linux/sysctl.h>
28 #include <linux/ctype.h>
29 #include <linux/list.h>
30 #include <linux/hash.h>
31
32 #include <trace/events/sched.h>
33
34 #include <asm/ftrace.h>
35 #include <asm/setup.h>
36
37 #include "trace_output.h"
38 #include "trace_stat.h"
39
40 #define FTRACE_WARN_ON(cond)                    \
41         do {                                    \
42                 if (WARN_ON(cond))              \
43                         ftrace_kill();          \
44         } while (0)
45
46 #define FTRACE_WARN_ON_ONCE(cond)               \
47         do {                                    \
48                 if (WARN_ON_ONCE(cond))         \
49                         ftrace_kill();          \
50         } while (0)
51
52 /* hash bits for specific function selection */
53 #define FTRACE_HASH_BITS 7
54 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
55
56 /* ftrace_enabled is a method to turn ftrace on or off */
57 int ftrace_enabled __read_mostly;
58 static int last_ftrace_enabled;
59
60 /* Quick disabling of function tracer. */
61 int function_trace_stop;
62
63 /*
64  * ftrace_disabled is set when an anomaly is discovered.
65  * ftrace_disabled is much stronger than ftrace_enabled.
66  */
67 static int ftrace_disabled __read_mostly;
68
69 static DEFINE_MUTEX(ftrace_lock);
70
71 static struct ftrace_ops ftrace_list_end __read_mostly =
72 {
73         .func           = ftrace_stub,
74 };
75
76 static struct ftrace_ops *ftrace_list __read_mostly = &ftrace_list_end;
77 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
78 ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
79 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
80
81 static void ftrace_list_func(unsigned long ip, unsigned long parent_ip)
82 {
83         struct ftrace_ops *op = ftrace_list;
84
85         /* in case someone actually ports this to alpha! */
86         read_barrier_depends();
87
88         while (op != &ftrace_list_end) {
89                 /* silly alpha */
90                 read_barrier_depends();
91                 op->func(ip, parent_ip);
92                 op = op->next;
93         };
94 }
95
96 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
97 {
98         if (!test_tsk_trace_trace(current))
99                 return;
100
101         ftrace_pid_function(ip, parent_ip);
102 }
103
104 static void set_ftrace_pid_function(ftrace_func_t func)
105 {
106         /* do not set ftrace_pid_function to itself! */
107         if (func != ftrace_pid_func)
108                 ftrace_pid_function = func;
109 }
110
111 /**
112  * clear_ftrace_function - reset the ftrace function
113  *
114  * This NULLs the ftrace function and in essence stops
115  * tracing.  There may be lag
116  */
117 void clear_ftrace_function(void)
118 {
119         ftrace_trace_function = ftrace_stub;
120         __ftrace_trace_function = ftrace_stub;
121         ftrace_pid_function = ftrace_stub;
122 }
123
124 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
125 /*
126  * For those archs that do not test ftrace_trace_stop in their
127  * mcount call site, we need to do it from C.
128  */
129 static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
130 {
131         if (function_trace_stop)
132                 return;
133
134         __ftrace_trace_function(ip, parent_ip);
135 }
136 #endif
137
138 static int __register_ftrace_function(struct ftrace_ops *ops)
139 {
140         ops->next = ftrace_list;
141         /*
142          * We are entering ops into the ftrace_list but another
143          * CPU might be walking that list. We need to make sure
144          * the ops->next pointer is valid before another CPU sees
145          * the ops pointer included into the ftrace_list.
146          */
147         smp_wmb();
148         ftrace_list = ops;
149
150         if (ftrace_enabled) {
151                 ftrace_func_t func;
152
153                 if (ops->next == &ftrace_list_end)
154                         func = ops->func;
155                 else
156                         func = ftrace_list_func;
157
158                 if (ftrace_pid_trace) {
159                         set_ftrace_pid_function(func);
160                         func = ftrace_pid_func;
161                 }
162
163                 /*
164                  * For one func, simply call it directly.
165                  * For more than one func, call the chain.
166                  */
167 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
168                 ftrace_trace_function = func;
169 #else
170                 __ftrace_trace_function = func;
171                 ftrace_trace_function = ftrace_test_stop_func;
172 #endif
173         }
174
175         return 0;
176 }
177
178 static int __unregister_ftrace_function(struct ftrace_ops *ops)
179 {
180         struct ftrace_ops **p;
181
182         /*
183          * If we are removing the last function, then simply point
184          * to the ftrace_stub.
185          */
186         if (ftrace_list == ops && ops->next == &ftrace_list_end) {
187                 ftrace_trace_function = ftrace_stub;
188                 ftrace_list = &ftrace_list_end;
189                 return 0;
190         }
191
192         for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next)
193                 if (*p == ops)
194                         break;
195
196         if (*p != ops)
197                 return -1;
198
199         *p = (*p)->next;
200
201         if (ftrace_enabled) {
202                 /* If we only have one func left, then call that directly */
203                 if (ftrace_list->next == &ftrace_list_end) {
204                         ftrace_func_t func = ftrace_list->func;
205
206                         if (ftrace_pid_trace) {
207                                 set_ftrace_pid_function(func);
208                                 func = ftrace_pid_func;
209                         }
210 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
211                         ftrace_trace_function = func;
212 #else
213                         __ftrace_trace_function = func;
214 #endif
215                 }
216         }
217
218         return 0;
219 }
220
221 static void ftrace_update_pid_func(void)
222 {
223         ftrace_func_t func;
224
225         if (ftrace_trace_function == ftrace_stub)
226                 return;
227
228         func = ftrace_trace_function;
229
230         if (ftrace_pid_trace) {
231                 set_ftrace_pid_function(func);
232                 func = ftrace_pid_func;
233         } else {
234                 if (func == ftrace_pid_func)
235                         func = ftrace_pid_function;
236         }
237
238 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
239         ftrace_trace_function = func;
240 #else
241         __ftrace_trace_function = func;
242 #endif
243 }
244
245 #ifdef CONFIG_FUNCTION_PROFILER
246 struct ftrace_profile {
247         struct hlist_node               node;
248         unsigned long                   ip;
249         unsigned long                   counter;
250 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
251         unsigned long long              time;
252 #endif
253 };
254
255 struct ftrace_profile_page {
256         struct ftrace_profile_page      *next;
257         unsigned long                   index;
258         struct ftrace_profile           records[];
259 };
260
261 struct ftrace_profile_stat {
262         atomic_t                        disabled;
263         struct hlist_head               *hash;
264         struct ftrace_profile_page      *pages;
265         struct ftrace_profile_page      *start;
266         struct tracer_stat              stat;
267 };
268
269 #define PROFILE_RECORDS_SIZE                                            \
270         (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
271
272 #define PROFILES_PER_PAGE                                       \
273         (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
274
275 static int ftrace_profile_bits __read_mostly;
276 static int ftrace_profile_enabled __read_mostly;
277
278 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
279 static DEFINE_MUTEX(ftrace_profile_lock);
280
281 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
282
283 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
284
285 static void *
286 function_stat_next(void *v, int idx)
287 {
288         struct ftrace_profile *rec = v;
289         struct ftrace_profile_page *pg;
290
291         pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
292
293  again:
294         rec++;
295         if ((void *)rec >= (void *)&pg->records[pg->index]) {
296                 pg = pg->next;
297                 if (!pg)
298                         return NULL;
299                 rec = &pg->records[0];
300                 if (!rec->counter)
301                         goto again;
302         }
303
304         return rec;
305 }
306
307 static void *function_stat_start(struct tracer_stat *trace)
308 {
309         struct ftrace_profile_stat *stat =
310                 container_of(trace, struct ftrace_profile_stat, stat);
311
312         if (!stat || !stat->start)
313                 return NULL;
314
315         return function_stat_next(&stat->start->records[0], 0);
316 }
317
318 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
319 /* function graph compares on total time */
320 static int function_stat_cmp(void *p1, void *p2)
321 {
322         struct ftrace_profile *a = p1;
323         struct ftrace_profile *b = p2;
324
325         if (a->time < b->time)
326                 return -1;
327         if (a->time > b->time)
328                 return 1;
329         else
330                 return 0;
331 }
332 #else
333 /* not function graph compares against hits */
334 static int function_stat_cmp(void *p1, void *p2)
335 {
336         struct ftrace_profile *a = p1;
337         struct ftrace_profile *b = p2;
338
339         if (a->counter < b->counter)
340                 return -1;
341         if (a->counter > b->counter)
342                 return 1;
343         else
344                 return 0;
345 }
346 #endif
347
348 static int function_stat_headers(struct seq_file *m)
349 {
350 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
351         seq_printf(m, "  Function                               "
352                    "Hit    Time            Avg\n"
353                       "  --------                               "
354                    "---    ----            ---\n");
355 #else
356         seq_printf(m, "  Function                               Hit\n"
357                       "  --------                               ---\n");
358 #endif
359         return 0;
360 }
361
362 static int function_stat_show(struct seq_file *m, void *v)
363 {
364         struct ftrace_profile *rec = v;
365         char str[KSYM_SYMBOL_LEN];
366 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
367         static DEFINE_MUTEX(mutex);
368         static struct trace_seq s;
369         unsigned long long avg;
370 #endif
371
372         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
373         seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
374
375 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
376         seq_printf(m, "    ");
377         avg = rec->time;
378         do_div(avg, rec->counter);
379
380         mutex_lock(&mutex);
381         trace_seq_init(&s);
382         trace_print_graph_duration(rec->time, &s);
383         trace_seq_puts(&s, "    ");
384         trace_print_graph_duration(avg, &s);
385         trace_print_seq(m, &s);
386         mutex_unlock(&mutex);
387 #endif
388         seq_putc(m, '\n');
389
390         return 0;
391 }
392
393 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
394 {
395         struct ftrace_profile_page *pg;
396
397         pg = stat->pages = stat->start;
398
399         while (pg) {
400                 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
401                 pg->index = 0;
402                 pg = pg->next;
403         }
404
405         memset(stat->hash, 0,
406                FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
407 }
408
409 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
410 {
411         struct ftrace_profile_page *pg;
412         int functions;
413         int pages;
414         int i;
415
416         /* If we already allocated, do nothing */
417         if (stat->pages)
418                 return 0;
419
420         stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
421         if (!stat->pages)
422                 return -ENOMEM;
423
424 #ifdef CONFIG_DYNAMIC_FTRACE
425         functions = ftrace_update_tot_cnt;
426 #else
427         /*
428          * We do not know the number of functions that exist because
429          * dynamic tracing is what counts them. With past experience
430          * we have around 20K functions. That should be more than enough.
431          * It is highly unlikely we will execute every function in
432          * the kernel.
433          */
434         functions = 20000;
435 #endif
436
437         pg = stat->start = stat->pages;
438
439         pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
440
441         for (i = 0; i < pages; i++) {
442                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
443                 if (!pg->next)
444                         goto out_free;
445                 pg = pg->next;
446         }
447
448         return 0;
449
450  out_free:
451         pg = stat->start;
452         while (pg) {
453                 unsigned long tmp = (unsigned long)pg;
454
455                 pg = pg->next;
456                 free_page(tmp);
457         }
458
459         free_page((unsigned long)stat->pages);
460         stat->pages = NULL;
461         stat->start = NULL;
462
463         return -ENOMEM;
464 }
465
466 static int ftrace_profile_init_cpu(int cpu)
467 {
468         struct ftrace_profile_stat *stat;
469         int size;
470
471         stat = &per_cpu(ftrace_profile_stats, cpu);
472
473         if (stat->hash) {
474                 /* If the profile is already created, simply reset it */
475                 ftrace_profile_reset(stat);
476                 return 0;
477         }
478
479         /*
480          * We are profiling all functions, but usually only a few thousand
481          * functions are hit. We'll make a hash of 1024 items.
482          */
483         size = FTRACE_PROFILE_HASH_SIZE;
484
485         stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
486
487         if (!stat->hash)
488                 return -ENOMEM;
489
490         if (!ftrace_profile_bits) {
491                 size--;
492
493                 for (; size; size >>= 1)
494                         ftrace_profile_bits++;
495         }
496
497         /* Preallocate the function profiling pages */
498         if (ftrace_profile_pages_init(stat) < 0) {
499                 kfree(stat->hash);
500                 stat->hash = NULL;
501                 return -ENOMEM;
502         }
503
504         return 0;
505 }
506
507 static int ftrace_profile_init(void)
508 {
509         int cpu;
510         int ret = 0;
511
512         for_each_online_cpu(cpu) {
513                 ret = ftrace_profile_init_cpu(cpu);
514                 if (ret)
515                         break;
516         }
517
518         return ret;
519 }
520
521 /* interrupts must be disabled */
522 static struct ftrace_profile *
523 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
524 {
525         struct ftrace_profile *rec;
526         struct hlist_head *hhd;
527         struct hlist_node *n;
528         unsigned long key;
529
530         key = hash_long(ip, ftrace_profile_bits);
531         hhd = &stat->hash[key];
532
533         if (hlist_empty(hhd))
534                 return NULL;
535
536         hlist_for_each_entry_rcu(rec, n, hhd, node) {
537                 if (rec->ip == ip)
538                         return rec;
539         }
540
541         return NULL;
542 }
543
544 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
545                                struct ftrace_profile *rec)
546 {
547         unsigned long key;
548
549         key = hash_long(rec->ip, ftrace_profile_bits);
550         hlist_add_head_rcu(&rec->node, &stat->hash[key]);
551 }
552
553 /*
554  * The memory is already allocated, this simply finds a new record to use.
555  */
556 static struct ftrace_profile *
557 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
558 {
559         struct ftrace_profile *rec = NULL;
560
561         /* prevent recursion (from NMIs) */
562         if (atomic_inc_return(&stat->disabled) != 1)
563                 goto out;
564
565         /*
566          * Try to find the function again since an NMI
567          * could have added it
568          */
569         rec = ftrace_find_profiled_func(stat, ip);
570         if (rec)
571                 goto out;
572
573         if (stat->pages->index == PROFILES_PER_PAGE) {
574                 if (!stat->pages->next)
575                         goto out;
576                 stat->pages = stat->pages->next;
577         }
578
579         rec = &stat->pages->records[stat->pages->index++];
580         rec->ip = ip;
581         ftrace_add_profile(stat, rec);
582
583  out:
584         atomic_dec(&stat->disabled);
585
586         return rec;
587 }
588
589 static void
590 function_profile_call(unsigned long ip, unsigned long parent_ip)
591 {
592         struct ftrace_profile_stat *stat;
593         struct ftrace_profile *rec;
594         unsigned long flags;
595
596         if (!ftrace_profile_enabled)
597                 return;
598
599         local_irq_save(flags);
600
601         stat = &__get_cpu_var(ftrace_profile_stats);
602         if (!stat->hash || !ftrace_profile_enabled)
603                 goto out;
604
605         rec = ftrace_find_profiled_func(stat, ip);
606         if (!rec) {
607                 rec = ftrace_profile_alloc(stat, ip);
608                 if (!rec)
609                         goto out;
610         }
611
612         rec->counter++;
613  out:
614         local_irq_restore(flags);
615 }
616
617 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
618 static int profile_graph_entry(struct ftrace_graph_ent *trace)
619 {
620         function_profile_call(trace->func, 0);
621         return 1;
622 }
623
624 static void profile_graph_return(struct ftrace_graph_ret *trace)
625 {
626         struct ftrace_profile_stat *stat;
627         unsigned long long calltime;
628         struct ftrace_profile *rec;
629         unsigned long flags;
630
631         local_irq_save(flags);
632         stat = &__get_cpu_var(ftrace_profile_stats);
633         if (!stat->hash || !ftrace_profile_enabled)
634                 goto out;
635
636         calltime = trace->rettime - trace->calltime;
637
638         if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
639                 int index;
640
641                 index = trace->depth;
642
643                 /* Append this call time to the parent time to subtract */
644                 if (index)
645                         current->ret_stack[index - 1].subtime += calltime;
646
647                 if (current->ret_stack[index].subtime < calltime)
648                         calltime -= current->ret_stack[index].subtime;
649                 else
650                         calltime = 0;
651         }
652
653         rec = ftrace_find_profiled_func(stat, trace->func);
654         if (rec)
655                 rec->time += calltime;
656
657  out:
658         local_irq_restore(flags);
659 }
660
661 static int register_ftrace_profiler(void)
662 {
663         return register_ftrace_graph(&profile_graph_return,
664                                      &profile_graph_entry);
665 }
666
667 static void unregister_ftrace_profiler(void)
668 {
669         unregister_ftrace_graph();
670 }
671 #else
672 static struct ftrace_ops ftrace_profile_ops __read_mostly =
673 {
674         .func           = function_profile_call,
675 };
676
677 static int register_ftrace_profiler(void)
678 {
679         return register_ftrace_function(&ftrace_profile_ops);
680 }
681
682 static void unregister_ftrace_profiler(void)
683 {
684         unregister_ftrace_function(&ftrace_profile_ops);
685 }
686 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
687
688 static ssize_t
689 ftrace_profile_write(struct file *filp, const char __user *ubuf,
690                      size_t cnt, loff_t *ppos)
691 {
692         unsigned long val;
693         char buf[64];           /* big enough to hold a number */
694         int ret;
695
696         if (cnt >= sizeof(buf))
697                 return -EINVAL;
698
699         if (copy_from_user(&buf, ubuf, cnt))
700                 return -EFAULT;
701
702         buf[cnt] = 0;
703
704         ret = strict_strtoul(buf, 10, &val);
705         if (ret < 0)
706                 return ret;
707
708         val = !!val;
709
710         mutex_lock(&ftrace_profile_lock);
711         if (ftrace_profile_enabled ^ val) {
712                 if (val) {
713                         ret = ftrace_profile_init();
714                         if (ret < 0) {
715                                 cnt = ret;
716                                 goto out;
717                         }
718
719                         ret = register_ftrace_profiler();
720                         if (ret < 0) {
721                                 cnt = ret;
722                                 goto out;
723                         }
724                         ftrace_profile_enabled = 1;
725                 } else {
726                         ftrace_profile_enabled = 0;
727                         /*
728                          * unregister_ftrace_profiler calls stop_machine
729                          * so this acts like an synchronize_sched.
730                          */
731                         unregister_ftrace_profiler();
732                 }
733         }
734  out:
735         mutex_unlock(&ftrace_profile_lock);
736
737         filp->f_pos += cnt;
738
739         return cnt;
740 }
741
742 static ssize_t
743 ftrace_profile_read(struct file *filp, char __user *ubuf,
744                      size_t cnt, loff_t *ppos)
745 {
746         char buf[64];           /* big enough to hold a number */
747         int r;
748
749         r = sprintf(buf, "%u\n", ftrace_profile_enabled);
750         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
751 }
752
753 static const struct file_operations ftrace_profile_fops = {
754         .open           = tracing_open_generic,
755         .read           = ftrace_profile_read,
756         .write          = ftrace_profile_write,
757 };
758
759 /* used to initialize the real stat files */
760 static struct tracer_stat function_stats __initdata = {
761         .name           = "functions",
762         .stat_start     = function_stat_start,
763         .stat_next      = function_stat_next,
764         .stat_cmp       = function_stat_cmp,
765         .stat_headers   = function_stat_headers,
766         .stat_show      = function_stat_show
767 };
768
769 static void ftrace_profile_debugfs(struct dentry *d_tracer)
770 {
771         struct ftrace_profile_stat *stat;
772         struct dentry *entry;
773         char *name;
774         int ret;
775         int cpu;
776
777         for_each_possible_cpu(cpu) {
778                 stat = &per_cpu(ftrace_profile_stats, cpu);
779
780                 /* allocate enough for function name + cpu number */
781                 name = kmalloc(32, GFP_KERNEL);
782                 if (!name) {
783                         /*
784                          * The files created are permanent, if something happens
785                          * we still do not free memory.
786                          */
787                         kfree(stat);
788                         WARN(1,
789                              "Could not allocate stat file for cpu %d\n",
790                              cpu);
791                         return;
792                 }
793                 stat->stat = function_stats;
794                 snprintf(name, 32, "function%d", cpu);
795                 stat->stat.name = name;
796                 ret = register_stat_tracer(&stat->stat);
797                 if (ret) {
798                         WARN(1,
799                              "Could not register function stat for cpu %d\n",
800                              cpu);
801                         kfree(name);
802                         return;
803                 }
804         }
805
806         entry = debugfs_create_file("function_profile_enabled", 0644,
807                                     d_tracer, NULL, &ftrace_profile_fops);
808         if (!entry)
809                 pr_warning("Could not create debugfs "
810                            "'function_profile_enabled' entry\n");
811 }
812
813 #else /* CONFIG_FUNCTION_PROFILER */
814 static void ftrace_profile_debugfs(struct dentry *d_tracer)
815 {
816 }
817 #endif /* CONFIG_FUNCTION_PROFILER */
818
819 /* set when tracing only a pid */
820 struct pid *ftrace_pid_trace;
821 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
822
823 #ifdef CONFIG_DYNAMIC_FTRACE
824
825 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
826 # error Dynamic ftrace depends on MCOUNT_RECORD
827 #endif
828
829 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
830
831 struct ftrace_func_probe {
832         struct hlist_node       node;
833         struct ftrace_probe_ops *ops;
834         unsigned long           flags;
835         unsigned long           ip;
836         void                    *data;
837         struct rcu_head         rcu;
838 };
839
840 enum {
841         FTRACE_ENABLE_CALLS             = (1 << 0),
842         FTRACE_DISABLE_CALLS            = (1 << 1),
843         FTRACE_UPDATE_TRACE_FUNC        = (1 << 2),
844         FTRACE_ENABLE_MCOUNT            = (1 << 3),
845         FTRACE_DISABLE_MCOUNT           = (1 << 4),
846         FTRACE_START_FUNC_RET           = (1 << 5),
847         FTRACE_STOP_FUNC_RET            = (1 << 6),
848 };
849
850 static int ftrace_filtered;
851
852 static struct dyn_ftrace *ftrace_new_addrs;
853
854 static DEFINE_MUTEX(ftrace_regex_lock);
855
856 struct ftrace_page {
857         struct ftrace_page      *next;
858         int                     index;
859         struct dyn_ftrace       records[];
860 };
861
862 #define ENTRIES_PER_PAGE \
863   ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
864
865 /* estimate from running different kernels */
866 #define NR_TO_INIT              10000
867
868 static struct ftrace_page       *ftrace_pages_start;
869 static struct ftrace_page       *ftrace_pages;
870
871 static struct dyn_ftrace *ftrace_free_records;
872
873 /*
874  * This is a double for. Do not use 'break' to break out of the loop,
875  * you must use a goto.
876  */
877 #define do_for_each_ftrace_rec(pg, rec)                                 \
878         for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
879                 int _____i;                                             \
880                 for (_____i = 0; _____i < pg->index; _____i++) {        \
881                         rec = &pg->records[_____i];
882
883 #define while_for_each_ftrace_rec()             \
884                 }                               \
885         }
886
887 #ifdef CONFIG_KPROBES
888
889 static int frozen_record_count;
890
891 static inline void freeze_record(struct dyn_ftrace *rec)
892 {
893         if (!(rec->flags & FTRACE_FL_FROZEN)) {
894                 rec->flags |= FTRACE_FL_FROZEN;
895                 frozen_record_count++;
896         }
897 }
898
899 static inline void unfreeze_record(struct dyn_ftrace *rec)
900 {
901         if (rec->flags & FTRACE_FL_FROZEN) {
902                 rec->flags &= ~FTRACE_FL_FROZEN;
903                 frozen_record_count--;
904         }
905 }
906
907 static inline int record_frozen(struct dyn_ftrace *rec)
908 {
909         return rec->flags & FTRACE_FL_FROZEN;
910 }
911 #else
912 # define freeze_record(rec)                     ({ 0; })
913 # define unfreeze_record(rec)                   ({ 0; })
914 # define record_frozen(rec)                     ({ 0; })
915 #endif /* CONFIG_KPROBES */
916
917 static void ftrace_free_rec(struct dyn_ftrace *rec)
918 {
919         rec->freelist = ftrace_free_records;
920         ftrace_free_records = rec;
921         rec->flags |= FTRACE_FL_FREE;
922 }
923
924 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
925 {
926         struct dyn_ftrace *rec;
927
928         /* First check for freed records */
929         if (ftrace_free_records) {
930                 rec = ftrace_free_records;
931
932                 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
933                         FTRACE_WARN_ON_ONCE(1);
934                         ftrace_free_records = NULL;
935                         return NULL;
936                 }
937
938                 ftrace_free_records = rec->freelist;
939                 memset(rec, 0, sizeof(*rec));
940                 return rec;
941         }
942
943         if (ftrace_pages->index == ENTRIES_PER_PAGE) {
944                 if (!ftrace_pages->next) {
945                         /* allocate another page */
946                         ftrace_pages->next =
947                                 (void *)get_zeroed_page(GFP_KERNEL);
948                         if (!ftrace_pages->next)
949                                 return NULL;
950                 }
951                 ftrace_pages = ftrace_pages->next;
952         }
953
954         return &ftrace_pages->records[ftrace_pages->index++];
955 }
956
957 static struct dyn_ftrace *
958 ftrace_record_ip(unsigned long ip)
959 {
960         struct dyn_ftrace *rec;
961
962         if (ftrace_disabled)
963                 return NULL;
964
965         rec = ftrace_alloc_dyn_node(ip);
966         if (!rec)
967                 return NULL;
968
969         rec->ip = ip;
970         rec->newlist = ftrace_new_addrs;
971         ftrace_new_addrs = rec;
972
973         return rec;
974 }
975
976 static void print_ip_ins(const char *fmt, unsigned char *p)
977 {
978         int i;
979
980         printk(KERN_CONT "%s", fmt);
981
982         for (i = 0; i < MCOUNT_INSN_SIZE; i++)
983                 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
984 }
985
986 static void ftrace_bug(int failed, unsigned long ip)
987 {
988         switch (failed) {
989         case -EFAULT:
990                 FTRACE_WARN_ON_ONCE(1);
991                 pr_info("ftrace faulted on modifying ");
992                 print_ip_sym(ip);
993                 break;
994         case -EINVAL:
995                 FTRACE_WARN_ON_ONCE(1);
996                 pr_info("ftrace failed to modify ");
997                 print_ip_sym(ip);
998                 print_ip_ins(" actual: ", (unsigned char *)ip);
999                 printk(KERN_CONT "\n");
1000                 break;
1001         case -EPERM:
1002                 FTRACE_WARN_ON_ONCE(1);
1003                 pr_info("ftrace faulted on writing ");
1004                 print_ip_sym(ip);
1005                 break;
1006         default:
1007                 FTRACE_WARN_ON_ONCE(1);
1008                 pr_info("ftrace faulted on unknown error ");
1009                 print_ip_sym(ip);
1010         }
1011 }
1012
1013
1014 static int
1015 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1016 {
1017         unsigned long ftrace_addr;
1018         unsigned long ip, fl;
1019
1020         ftrace_addr = (unsigned long)FTRACE_ADDR;
1021
1022         ip = rec->ip;
1023
1024         /*
1025          * If this record is not to be traced and
1026          * it is not enabled then do nothing.
1027          *
1028          * If this record is not to be traced and
1029          * it is enabled then disable it.
1030          *
1031          */
1032         if (rec->flags & FTRACE_FL_NOTRACE) {
1033                 if (rec->flags & FTRACE_FL_ENABLED)
1034                         rec->flags &= ~FTRACE_FL_ENABLED;
1035                 else
1036                         return 0;
1037
1038         } else if (ftrace_filtered && enable) {
1039                 /*
1040                  * Filtering is on:
1041                  */
1042
1043                 fl = rec->flags & (FTRACE_FL_FILTER | FTRACE_FL_ENABLED);
1044
1045                 /* Record is filtered and enabled, do nothing */
1046                 if (fl == (FTRACE_FL_FILTER | FTRACE_FL_ENABLED))
1047                         return 0;
1048
1049                 /* Record is not filtered or enabled, do nothing */
1050                 if (!fl)
1051                         return 0;
1052
1053                 /* Record is not filtered but enabled, disable it */
1054                 if (fl == FTRACE_FL_ENABLED)
1055                         rec->flags &= ~FTRACE_FL_ENABLED;
1056                 else
1057                 /* Otherwise record is filtered but not enabled, enable it */
1058                         rec->flags |= FTRACE_FL_ENABLED;
1059         } else {
1060                 /* Disable or not filtered */
1061
1062                 if (enable) {
1063                         /* if record is enabled, do nothing */
1064                         if (rec->flags & FTRACE_FL_ENABLED)
1065                                 return 0;
1066
1067                         rec->flags |= FTRACE_FL_ENABLED;
1068
1069                 } else {
1070
1071                         /* if record is not enabled, do nothing */
1072                         if (!(rec->flags & FTRACE_FL_ENABLED))
1073                                 return 0;
1074
1075                         rec->flags &= ~FTRACE_FL_ENABLED;
1076                 }
1077         }
1078
1079         if (rec->flags & FTRACE_FL_ENABLED)
1080                 return ftrace_make_call(rec, ftrace_addr);
1081         else
1082                 return ftrace_make_nop(NULL, rec, ftrace_addr);
1083 }
1084
1085 static void ftrace_replace_code(int enable)
1086 {
1087         struct dyn_ftrace *rec;
1088         struct ftrace_page *pg;
1089         int failed;
1090
1091         do_for_each_ftrace_rec(pg, rec) {
1092                 /*
1093                  * Skip over free records, records that have
1094                  * failed and not converted.
1095                  */
1096                 if (rec->flags & FTRACE_FL_FREE ||
1097                     rec->flags & FTRACE_FL_FAILED ||
1098                     !(rec->flags & FTRACE_FL_CONVERTED))
1099                         continue;
1100
1101                 /* ignore updates to this record's mcount site */
1102                 if (get_kprobe((void *)rec->ip)) {
1103                         freeze_record(rec);
1104                         continue;
1105                 } else {
1106                         unfreeze_record(rec);
1107                 }
1108
1109                 failed = __ftrace_replace_code(rec, enable);
1110                 if (failed) {
1111                         rec->flags |= FTRACE_FL_FAILED;
1112                         if ((system_state == SYSTEM_BOOTING) ||
1113                             !core_kernel_text(rec->ip)) {
1114                                 ftrace_free_rec(rec);
1115                                 } else {
1116                                 ftrace_bug(failed, rec->ip);
1117                                         /* Stop processing */
1118                                         return;
1119                                 }
1120                 }
1121         } while_for_each_ftrace_rec();
1122 }
1123
1124 static int
1125 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1126 {
1127         unsigned long ip;
1128         int ret;
1129
1130         ip = rec->ip;
1131
1132         ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1133         if (ret) {
1134                 ftrace_bug(ret, ip);
1135                 rec->flags |= FTRACE_FL_FAILED;
1136                 return 0;
1137         }
1138         return 1;
1139 }
1140
1141 /*
1142  * archs can override this function if they must do something
1143  * before the modifying code is performed.
1144  */
1145 int __weak ftrace_arch_code_modify_prepare(void)
1146 {
1147         return 0;
1148 }
1149
1150 /*
1151  * archs can override this function if they must do something
1152  * after the modifying code is performed.
1153  */
1154 int __weak ftrace_arch_code_modify_post_process(void)
1155 {
1156         return 0;
1157 }
1158
1159 static int __ftrace_modify_code(void *data)
1160 {
1161         int *command = data;
1162
1163         if (*command & FTRACE_ENABLE_CALLS)
1164                 ftrace_replace_code(1);
1165         else if (*command & FTRACE_DISABLE_CALLS)
1166                 ftrace_replace_code(0);
1167
1168         if (*command & FTRACE_UPDATE_TRACE_FUNC)
1169                 ftrace_update_ftrace_func(ftrace_trace_function);
1170
1171         if (*command & FTRACE_START_FUNC_RET)
1172                 ftrace_enable_ftrace_graph_caller();
1173         else if (*command & FTRACE_STOP_FUNC_RET)
1174                 ftrace_disable_ftrace_graph_caller();
1175
1176         return 0;
1177 }
1178
1179 static void ftrace_run_update_code(int command)
1180 {
1181         int ret;
1182
1183         ret = ftrace_arch_code_modify_prepare();
1184         FTRACE_WARN_ON(ret);
1185         if (ret)
1186                 return;
1187
1188         stop_machine(__ftrace_modify_code, &command, NULL);
1189
1190         ret = ftrace_arch_code_modify_post_process();
1191         FTRACE_WARN_ON(ret);
1192 }
1193
1194 static ftrace_func_t saved_ftrace_func;
1195 static int ftrace_start_up;
1196
1197 static void ftrace_startup_enable(int command)
1198 {
1199         if (saved_ftrace_func != ftrace_trace_function) {
1200                 saved_ftrace_func = ftrace_trace_function;
1201                 command |= FTRACE_UPDATE_TRACE_FUNC;
1202         }
1203
1204         if (!command || !ftrace_enabled)
1205                 return;
1206
1207         ftrace_run_update_code(command);
1208 }
1209
1210 static void ftrace_startup(int command)
1211 {
1212         if (unlikely(ftrace_disabled))
1213                 return;
1214
1215         ftrace_start_up++;
1216         command |= FTRACE_ENABLE_CALLS;
1217
1218         ftrace_startup_enable(command);
1219 }
1220
1221 static void ftrace_shutdown(int command)
1222 {
1223         if (unlikely(ftrace_disabled))
1224                 return;
1225
1226         ftrace_start_up--;
1227         /*
1228          * Just warn in case of unbalance, no need to kill ftrace, it's not
1229          * critical but the ftrace_call callers may be never nopped again after
1230          * further ftrace uses.
1231          */
1232         WARN_ON_ONCE(ftrace_start_up < 0);
1233
1234         if (!ftrace_start_up)
1235                 command |= FTRACE_DISABLE_CALLS;
1236
1237         if (saved_ftrace_func != ftrace_trace_function) {
1238                 saved_ftrace_func = ftrace_trace_function;
1239                 command |= FTRACE_UPDATE_TRACE_FUNC;
1240         }
1241
1242         if (!command || !ftrace_enabled)
1243                 return;
1244
1245         ftrace_run_update_code(command);
1246 }
1247
1248 static void ftrace_startup_sysctl(void)
1249 {
1250         int command = FTRACE_ENABLE_MCOUNT;
1251
1252         if (unlikely(ftrace_disabled))
1253                 return;
1254
1255         /* Force update next time */
1256         saved_ftrace_func = NULL;
1257         /* ftrace_start_up is true if we want ftrace running */
1258         if (ftrace_start_up)
1259                 command |= FTRACE_ENABLE_CALLS;
1260
1261         ftrace_run_update_code(command);
1262 }
1263
1264 static void ftrace_shutdown_sysctl(void)
1265 {
1266         int command = FTRACE_DISABLE_MCOUNT;
1267
1268         if (unlikely(ftrace_disabled))
1269                 return;
1270
1271         /* ftrace_start_up is true if ftrace is running */
1272         if (ftrace_start_up)
1273                 command |= FTRACE_DISABLE_CALLS;
1274
1275         ftrace_run_update_code(command);
1276 }
1277
1278 static cycle_t          ftrace_update_time;
1279 static unsigned long    ftrace_update_cnt;
1280 unsigned long           ftrace_update_tot_cnt;
1281
1282 static int ftrace_update_code(struct module *mod)
1283 {
1284         struct dyn_ftrace *p;
1285         cycle_t start, stop;
1286
1287         start = ftrace_now(raw_smp_processor_id());
1288         ftrace_update_cnt = 0;
1289
1290         while (ftrace_new_addrs) {
1291
1292                 /* If something went wrong, bail without enabling anything */
1293                 if (unlikely(ftrace_disabled))
1294                         return -1;
1295
1296                 p = ftrace_new_addrs;
1297                 ftrace_new_addrs = p->newlist;
1298                 p->flags = 0L;
1299
1300                 /* convert record (i.e, patch mcount-call with NOP) */
1301                 if (ftrace_code_disable(mod, p)) {
1302                         p->flags |= FTRACE_FL_CONVERTED;
1303                         ftrace_update_cnt++;
1304                 } else
1305                         ftrace_free_rec(p);
1306         }
1307
1308         stop = ftrace_now(raw_smp_processor_id());
1309         ftrace_update_time = stop - start;
1310         ftrace_update_tot_cnt += ftrace_update_cnt;
1311
1312         return 0;
1313 }
1314
1315 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1316 {
1317         struct ftrace_page *pg;
1318         int cnt;
1319         int i;
1320
1321         /* allocate a few pages */
1322         ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1323         if (!ftrace_pages_start)
1324                 return -1;
1325
1326         /*
1327          * Allocate a few more pages.
1328          *
1329          * TODO: have some parser search vmlinux before
1330          *   final linking to find all calls to ftrace.
1331          *   Then we can:
1332          *    a) know how many pages to allocate.
1333          *     and/or
1334          *    b) set up the table then.
1335          *
1336          *  The dynamic code is still necessary for
1337          *  modules.
1338          */
1339
1340         pg = ftrace_pages = ftrace_pages_start;
1341
1342         cnt = num_to_init / ENTRIES_PER_PAGE;
1343         pr_info("ftrace: allocating %ld entries in %d pages\n",
1344                 num_to_init, cnt + 1);
1345
1346         for (i = 0; i < cnt; i++) {
1347                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1348
1349                 /* If we fail, we'll try later anyway */
1350                 if (!pg->next)
1351                         break;
1352
1353                 pg = pg->next;
1354         }
1355
1356         return 0;
1357 }
1358
1359 enum {
1360         FTRACE_ITER_FILTER      = (1 << 0),
1361         FTRACE_ITER_CONT        = (1 << 1),
1362         FTRACE_ITER_NOTRACE     = (1 << 2),
1363         FTRACE_ITER_FAILURES    = (1 << 3),
1364         FTRACE_ITER_PRINTALL    = (1 << 4),
1365         FTRACE_ITER_HASH        = (1 << 5),
1366 };
1367
1368 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1369
1370 struct ftrace_iterator {
1371         struct ftrace_page      *pg;
1372         int                     hidx;
1373         int                     idx;
1374         unsigned                flags;
1375         unsigned char           buffer[FTRACE_BUFF_MAX+1];
1376         unsigned                buffer_idx;
1377         unsigned                filtered;
1378 };
1379
1380 static void *
1381 t_hash_next(struct seq_file *m, void *v, loff_t *pos)
1382 {
1383         struct ftrace_iterator *iter = m->private;
1384         struct hlist_node *hnd = v;
1385         struct hlist_head *hhd;
1386
1387         WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
1388
1389         (*pos)++;
1390
1391  retry:
1392         if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1393                 return NULL;
1394
1395         hhd = &ftrace_func_hash[iter->hidx];
1396
1397         if (hlist_empty(hhd)) {
1398                 iter->hidx++;
1399                 hnd = NULL;
1400                 goto retry;
1401         }
1402
1403         if (!hnd)
1404                 hnd = hhd->first;
1405         else {
1406                 hnd = hnd->next;
1407                 if (!hnd) {
1408                         iter->hidx++;
1409                         goto retry;
1410                 }
1411         }
1412
1413         return hnd;
1414 }
1415
1416 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1417 {
1418         struct ftrace_iterator *iter = m->private;
1419         void *p = NULL;
1420
1421         iter->flags |= FTRACE_ITER_HASH;
1422
1423         return t_hash_next(m, p, pos);
1424 }
1425
1426 static int t_hash_show(struct seq_file *m, void *v)
1427 {
1428         struct ftrace_func_probe *rec;
1429         struct hlist_node *hnd = v;
1430         char str[KSYM_SYMBOL_LEN];
1431
1432         rec = hlist_entry(hnd, struct ftrace_func_probe, node);
1433
1434         if (rec->ops->print)
1435                 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1436
1437         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1438         seq_printf(m, "%s:", str);
1439
1440         kallsyms_lookup((unsigned long)rec->ops->func, NULL, NULL, NULL, str);
1441         seq_printf(m, "%s", str);
1442
1443         if (rec->data)
1444                 seq_printf(m, ":%p", rec->data);
1445         seq_putc(m, '\n');
1446
1447         return 0;
1448 }
1449
1450 static void *
1451 t_next(struct seq_file *m, void *v, loff_t *pos)
1452 {
1453         struct ftrace_iterator *iter = m->private;
1454         struct dyn_ftrace *rec = NULL;
1455
1456         if (iter->flags & FTRACE_ITER_HASH)
1457                 return t_hash_next(m, v, pos);
1458
1459         (*pos)++;
1460
1461         if (iter->flags & FTRACE_ITER_PRINTALL)
1462                 return NULL;
1463
1464  retry:
1465         if (iter->idx >= iter->pg->index) {
1466                 if (iter->pg->next) {
1467                         iter->pg = iter->pg->next;
1468                         iter->idx = 0;
1469                         goto retry;
1470                 }
1471         } else {
1472                 rec = &iter->pg->records[iter->idx++];
1473                 if ((rec->flags & FTRACE_FL_FREE) ||
1474
1475                     (!(iter->flags & FTRACE_ITER_FAILURES) &&
1476                      (rec->flags & FTRACE_FL_FAILED)) ||
1477
1478                     ((iter->flags & FTRACE_ITER_FAILURES) &&
1479                      !(rec->flags & FTRACE_FL_FAILED)) ||
1480
1481                     ((iter->flags & FTRACE_ITER_FILTER) &&
1482                      !(rec->flags & FTRACE_FL_FILTER)) ||
1483
1484                     ((iter->flags & FTRACE_ITER_NOTRACE) &&
1485                      !(rec->flags & FTRACE_FL_NOTRACE))) {
1486                         rec = NULL;
1487                         goto retry;
1488                 }
1489         }
1490
1491         return rec;
1492 }
1493
1494 static void *t_start(struct seq_file *m, loff_t *pos)
1495 {
1496         struct ftrace_iterator *iter = m->private;
1497         void *p = NULL;
1498         loff_t l;
1499
1500         mutex_lock(&ftrace_lock);
1501         /*
1502          * For set_ftrace_filter reading, if we have the filter
1503          * off, we can short cut and just print out that all
1504          * functions are enabled.
1505          */
1506         if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1507                 if (*pos > 0)
1508                         return t_hash_start(m, pos);
1509                 iter->flags |= FTRACE_ITER_PRINTALL;
1510                 return iter;
1511         }
1512
1513         if (iter->flags & FTRACE_ITER_HASH)
1514                 return t_hash_start(m, pos);
1515
1516         iter->pg = ftrace_pages_start;
1517         iter->idx = 0;
1518         for (l = 0; l <= *pos; ) {
1519                 p = t_next(m, p, &l);
1520                 if (!p)
1521                         break;
1522         }
1523
1524         if (!p && iter->flags & FTRACE_ITER_FILTER)
1525                 return t_hash_start(m, pos);
1526
1527         return p;
1528 }
1529
1530 static void t_stop(struct seq_file *m, void *p)
1531 {
1532         mutex_unlock(&ftrace_lock);
1533 }
1534
1535 static int t_show(struct seq_file *m, void *v)
1536 {
1537         struct ftrace_iterator *iter = m->private;
1538         struct dyn_ftrace *rec = v;
1539         char str[KSYM_SYMBOL_LEN];
1540
1541         if (iter->flags & FTRACE_ITER_HASH)
1542                 return t_hash_show(m, v);
1543
1544         if (iter->flags & FTRACE_ITER_PRINTALL) {
1545                 seq_printf(m, "#### all functions enabled ####\n");
1546                 return 0;
1547         }
1548
1549         if (!rec)
1550                 return 0;
1551
1552         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1553
1554         seq_printf(m, "%s\n", str);
1555
1556         return 0;
1557 }
1558
1559 static struct seq_operations show_ftrace_seq_ops = {
1560         .start = t_start,
1561         .next = t_next,
1562         .stop = t_stop,
1563         .show = t_show,
1564 };
1565
1566 static int
1567 ftrace_avail_open(struct inode *inode, struct file *file)
1568 {
1569         struct ftrace_iterator *iter;
1570         int ret;
1571
1572         if (unlikely(ftrace_disabled))
1573                 return -ENODEV;
1574
1575         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1576         if (!iter)
1577                 return -ENOMEM;
1578
1579         iter->pg = ftrace_pages_start;
1580
1581         ret = seq_open(file, &show_ftrace_seq_ops);
1582         if (!ret) {
1583                 struct seq_file *m = file->private_data;
1584
1585                 m->private = iter;
1586         } else {
1587                 kfree(iter);
1588         }
1589
1590         return ret;
1591 }
1592
1593 int ftrace_avail_release(struct inode *inode, struct file *file)
1594 {
1595         struct seq_file *m = (struct seq_file *)file->private_data;
1596         struct ftrace_iterator *iter = m->private;
1597
1598         seq_release(inode, file);
1599         kfree(iter);
1600
1601         return 0;
1602 }
1603
1604 static int
1605 ftrace_failures_open(struct inode *inode, struct file *file)
1606 {
1607         int ret;
1608         struct seq_file *m;
1609         struct ftrace_iterator *iter;
1610
1611         ret = ftrace_avail_open(inode, file);
1612         if (!ret) {
1613                 m = (struct seq_file *)file->private_data;
1614                 iter = (struct ftrace_iterator *)m->private;
1615                 iter->flags = FTRACE_ITER_FAILURES;
1616         }
1617
1618         return ret;
1619 }
1620
1621
1622 static void ftrace_filter_reset(int enable)
1623 {
1624         struct ftrace_page *pg;
1625         struct dyn_ftrace *rec;
1626         unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1627
1628         mutex_lock(&ftrace_lock);
1629         if (enable)
1630                 ftrace_filtered = 0;
1631         do_for_each_ftrace_rec(pg, rec) {
1632                 if (rec->flags & FTRACE_FL_FAILED)
1633                         continue;
1634                 rec->flags &= ~type;
1635         } while_for_each_ftrace_rec();
1636         mutex_unlock(&ftrace_lock);
1637 }
1638
1639 static int
1640 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1641 {
1642         struct ftrace_iterator *iter;
1643         int ret = 0;
1644
1645         if (unlikely(ftrace_disabled))
1646                 return -ENODEV;
1647
1648         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1649         if (!iter)
1650                 return -ENOMEM;
1651
1652         mutex_lock(&ftrace_regex_lock);
1653         if ((file->f_mode & FMODE_WRITE) &&
1654             !(file->f_flags & O_APPEND))
1655                 ftrace_filter_reset(enable);
1656
1657         if (file->f_mode & FMODE_READ) {
1658                 iter->pg = ftrace_pages_start;
1659                 iter->flags = enable ? FTRACE_ITER_FILTER :
1660                         FTRACE_ITER_NOTRACE;
1661
1662                 ret = seq_open(file, &show_ftrace_seq_ops);
1663                 if (!ret) {
1664                         struct seq_file *m = file->private_data;
1665                         m->private = iter;
1666                 } else
1667                         kfree(iter);
1668         } else
1669                 file->private_data = iter;
1670         mutex_unlock(&ftrace_regex_lock);
1671
1672         return ret;
1673 }
1674
1675 static int
1676 ftrace_filter_open(struct inode *inode, struct file *file)
1677 {
1678         return ftrace_regex_open(inode, file, 1);
1679 }
1680
1681 static int
1682 ftrace_notrace_open(struct inode *inode, struct file *file)
1683 {
1684         return ftrace_regex_open(inode, file, 0);
1685 }
1686
1687 static loff_t
1688 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1689 {
1690         loff_t ret;
1691
1692         if (file->f_mode & FMODE_READ)
1693                 ret = seq_lseek(file, offset, origin);
1694         else
1695                 file->f_pos = ret = 1;
1696
1697         return ret;
1698 }
1699
1700 enum {
1701         MATCH_FULL,
1702         MATCH_FRONT_ONLY,
1703         MATCH_MIDDLE_ONLY,
1704         MATCH_END_ONLY,
1705 };
1706
1707 /*
1708  * (static function - no need for kernel doc)
1709  *
1710  * Pass in a buffer containing a glob and this function will
1711  * set search to point to the search part of the buffer and
1712  * return the type of search it is (see enum above).
1713  * This does modify buff.
1714  *
1715  * Returns enum type.
1716  *  search returns the pointer to use for comparison.
1717  *  not returns 1 if buff started with a '!'
1718  *     0 otherwise.
1719  */
1720 static int
1721 ftrace_setup_glob(char *buff, int len, char **search, int *not)
1722 {
1723         int type = MATCH_FULL;
1724         int i;
1725
1726         if (buff[0] == '!') {
1727                 *not = 1;
1728                 buff++;
1729                 len--;
1730         } else
1731                 *not = 0;
1732
1733         *search = buff;
1734
1735         for (i = 0; i < len; i++) {
1736                 if (buff[i] == '*') {
1737                         if (!i) {
1738                                 *search = buff + 1;
1739                                 type = MATCH_END_ONLY;
1740                         } else {
1741                                 if (type == MATCH_END_ONLY)
1742                                         type = MATCH_MIDDLE_ONLY;
1743                                 else
1744                                         type = MATCH_FRONT_ONLY;
1745                                 buff[i] = 0;
1746                                 break;
1747                         }
1748                 }
1749         }
1750
1751         return type;
1752 }
1753
1754 static int ftrace_match(char *str, char *regex, int len, int type)
1755 {
1756         int matched = 0;
1757         char *ptr;
1758
1759         switch (type) {
1760         case MATCH_FULL:
1761                 if (strcmp(str, regex) == 0)
1762                         matched = 1;
1763                 break;
1764         case MATCH_FRONT_ONLY:
1765                 if (strncmp(str, regex, len) == 0)
1766                         matched = 1;
1767                 break;
1768         case MATCH_MIDDLE_ONLY:
1769                 if (strstr(str, regex))
1770                         matched = 1;
1771                 break;
1772         case MATCH_END_ONLY:
1773                 ptr = strstr(str, regex);
1774                 if (ptr && (ptr[len] == 0))
1775                         matched = 1;
1776                 break;
1777         }
1778
1779         return matched;
1780 }
1781
1782 static int
1783 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1784 {
1785         char str[KSYM_SYMBOL_LEN];
1786
1787         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1788         return ftrace_match(str, regex, len, type);
1789 }
1790
1791 static void ftrace_match_records(char *buff, int len, int enable)
1792 {
1793         unsigned int search_len;
1794         struct ftrace_page *pg;
1795         struct dyn_ftrace *rec;
1796         unsigned long flag;
1797         char *search;
1798         int type;
1799         int not;
1800
1801         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1802         type = ftrace_setup_glob(buff, len, &search, &not);
1803
1804         search_len = strlen(search);
1805
1806         mutex_lock(&ftrace_lock);
1807         do_for_each_ftrace_rec(pg, rec) {
1808
1809                 if (rec->flags & FTRACE_FL_FAILED)
1810                         continue;
1811
1812                 if (ftrace_match_record(rec, search, search_len, type)) {
1813                         if (not)
1814                                 rec->flags &= ~flag;
1815                         else
1816                                 rec->flags |= flag;
1817                 }
1818                 /*
1819                  * Only enable filtering if we have a function that
1820                  * is filtered on.
1821                  */
1822                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1823                         ftrace_filtered = 1;
1824         } while_for_each_ftrace_rec();
1825         mutex_unlock(&ftrace_lock);
1826 }
1827
1828 static int
1829 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1830                            char *regex, int len, int type)
1831 {
1832         char str[KSYM_SYMBOL_LEN];
1833         char *modname;
1834
1835         kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1836
1837         if (!modname || strcmp(modname, mod))
1838                 return 0;
1839
1840         /* blank search means to match all funcs in the mod */
1841         if (len)
1842                 return ftrace_match(str, regex, len, type);
1843         else
1844                 return 1;
1845 }
1846
1847 static void ftrace_match_module_records(char *buff, char *mod, int enable)
1848 {
1849         unsigned search_len = 0;
1850         struct ftrace_page *pg;
1851         struct dyn_ftrace *rec;
1852         int type = MATCH_FULL;
1853         char *search = buff;
1854         unsigned long flag;
1855         int not = 0;
1856
1857         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1858
1859         /* blank or '*' mean the same */
1860         if (strcmp(buff, "*") == 0)
1861                 buff[0] = 0;
1862
1863         /* handle the case of 'dont filter this module' */
1864         if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1865                 buff[0] = 0;
1866                 not = 1;
1867         }
1868
1869         if (strlen(buff)) {
1870                 type = ftrace_setup_glob(buff, strlen(buff), &search, &not);
1871                 search_len = strlen(search);
1872         }
1873
1874         mutex_lock(&ftrace_lock);
1875         do_for_each_ftrace_rec(pg, rec) {
1876
1877                 if (rec->flags & FTRACE_FL_FAILED)
1878                         continue;
1879
1880                 if (ftrace_match_module_record(rec, mod,
1881                                                search, search_len, type)) {
1882                         if (not)
1883                                 rec->flags &= ~flag;
1884                         else
1885                                 rec->flags |= flag;
1886                 }
1887                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1888                         ftrace_filtered = 1;
1889
1890         } while_for_each_ftrace_rec();
1891         mutex_unlock(&ftrace_lock);
1892 }
1893
1894 /*
1895  * We register the module command as a template to show others how
1896  * to register the a command as well.
1897  */
1898
1899 static int
1900 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1901 {
1902         char *mod;
1903
1904         /*
1905          * cmd == 'mod' because we only registered this func
1906          * for the 'mod' ftrace_func_command.
1907          * But if you register one func with multiple commands,
1908          * you can tell which command was used by the cmd
1909          * parameter.
1910          */
1911
1912         /* we must have a module name */
1913         if (!param)
1914                 return -EINVAL;
1915
1916         mod = strsep(&param, ":");
1917         if (!strlen(mod))
1918                 return -EINVAL;
1919
1920         ftrace_match_module_records(func, mod, enable);
1921         return 0;
1922 }
1923
1924 static struct ftrace_func_command ftrace_mod_cmd = {
1925         .name                   = "mod",
1926         .func                   = ftrace_mod_callback,
1927 };
1928
1929 static int __init ftrace_mod_cmd_init(void)
1930 {
1931         return register_ftrace_command(&ftrace_mod_cmd);
1932 }
1933 device_initcall(ftrace_mod_cmd_init);
1934
1935 static void
1936 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1937 {
1938         struct ftrace_func_probe *entry;
1939         struct hlist_head *hhd;
1940         struct hlist_node *n;
1941         unsigned long key;
1942         int resched;
1943
1944         key = hash_long(ip, FTRACE_HASH_BITS);
1945
1946         hhd = &ftrace_func_hash[key];
1947
1948         if (hlist_empty(hhd))
1949                 return;
1950
1951         /*
1952          * Disable preemption for these calls to prevent a RCU grace
1953          * period. This syncs the hash iteration and freeing of items
1954          * on the hash. rcu_read_lock is too dangerous here.
1955          */
1956         resched = ftrace_preempt_disable();
1957         hlist_for_each_entry_rcu(entry, n, hhd, node) {
1958                 if (entry->ip == ip)
1959                         entry->ops->func(ip, parent_ip, &entry->data);
1960         }
1961         ftrace_preempt_enable(resched);
1962 }
1963
1964 static struct ftrace_ops trace_probe_ops __read_mostly =
1965 {
1966         .func           = function_trace_probe_call,
1967 };
1968
1969 static int ftrace_probe_registered;
1970
1971 static void __enable_ftrace_function_probe(void)
1972 {
1973         int i;
1974
1975         if (ftrace_probe_registered)
1976                 return;
1977
1978         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1979                 struct hlist_head *hhd = &ftrace_func_hash[i];
1980                 if (hhd->first)
1981                         break;
1982         }
1983         /* Nothing registered? */
1984         if (i == FTRACE_FUNC_HASHSIZE)
1985                 return;
1986
1987         __register_ftrace_function(&trace_probe_ops);
1988         ftrace_startup(0);
1989         ftrace_probe_registered = 1;
1990 }
1991
1992 static void __disable_ftrace_function_probe(void)
1993 {
1994         int i;
1995
1996         if (!ftrace_probe_registered)
1997                 return;
1998
1999         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2000                 struct hlist_head *hhd = &ftrace_func_hash[i];
2001                 if (hhd->first)
2002                         return;
2003         }
2004
2005         /* no more funcs left */
2006         __unregister_ftrace_function(&trace_probe_ops);
2007         ftrace_shutdown(0);
2008         ftrace_probe_registered = 0;
2009 }
2010
2011
2012 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2013 {
2014         struct ftrace_func_probe *entry =
2015                 container_of(rhp, struct ftrace_func_probe, rcu);
2016
2017         if (entry->ops->free)
2018                 entry->ops->free(&entry->data);
2019         kfree(entry);
2020 }
2021
2022
2023 int
2024 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2025                               void *data)
2026 {
2027         struct ftrace_func_probe *entry;
2028         struct ftrace_page *pg;
2029         struct dyn_ftrace *rec;
2030         int type, len, not;
2031         unsigned long key;
2032         int count = 0;
2033         char *search;
2034
2035         type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
2036         len = strlen(search);
2037
2038         /* we do not support '!' for function probes */
2039         if (WARN_ON(not))
2040                 return -EINVAL;
2041
2042         mutex_lock(&ftrace_lock);
2043         do_for_each_ftrace_rec(pg, rec) {
2044
2045                 if (rec->flags & FTRACE_FL_FAILED)
2046                         continue;
2047
2048                 if (!ftrace_match_record(rec, search, len, type))
2049                         continue;
2050
2051                 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2052                 if (!entry) {
2053                         /* If we did not process any, then return error */
2054                         if (!count)
2055                                 count = -ENOMEM;
2056                         goto out_unlock;
2057                 }
2058
2059                 count++;
2060
2061                 entry->data = data;
2062
2063                 /*
2064                  * The caller might want to do something special
2065                  * for each function we find. We call the callback
2066                  * to give the caller an opportunity to do so.
2067                  */
2068                 if (ops->callback) {
2069                         if (ops->callback(rec->ip, &entry->data) < 0) {
2070                                 /* caller does not like this func */
2071                                 kfree(entry);
2072                                 continue;
2073                         }
2074                 }
2075
2076                 entry->ops = ops;
2077                 entry->ip = rec->ip;
2078
2079                 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2080                 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2081
2082         } while_for_each_ftrace_rec();
2083         __enable_ftrace_function_probe();
2084
2085  out_unlock:
2086         mutex_unlock(&ftrace_lock);
2087
2088         return count;
2089 }
2090
2091 enum {
2092         PROBE_TEST_FUNC         = 1,
2093         PROBE_TEST_DATA         = 2
2094 };
2095
2096 static void
2097 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2098                                   void *data, int flags)
2099 {
2100         struct ftrace_func_probe *entry;
2101         struct hlist_node *n, *tmp;
2102         char str[KSYM_SYMBOL_LEN];
2103         int type = MATCH_FULL;
2104         int i, len = 0;
2105         char *search;
2106
2107         if (glob && (strcmp(glob, "*") || !strlen(glob)))
2108                 glob = NULL;
2109         else {
2110                 int not;
2111
2112                 type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
2113                 len = strlen(search);
2114
2115                 /* we do not support '!' for function probes */
2116                 if (WARN_ON(not))
2117                         return;
2118         }
2119
2120         mutex_lock(&ftrace_lock);
2121         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2122                 struct hlist_head *hhd = &ftrace_func_hash[i];
2123
2124                 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2125
2126                         /* break up if statements for readability */
2127                         if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2128                                 continue;
2129
2130                         if ((flags & PROBE_TEST_DATA) && entry->data != data)
2131                                 continue;
2132
2133                         /* do this last, since it is the most expensive */
2134                         if (glob) {
2135                                 kallsyms_lookup(entry->ip, NULL, NULL,
2136                                                 NULL, str);
2137                                 if (!ftrace_match(str, glob, len, type))
2138                                         continue;
2139                         }
2140
2141                         hlist_del(&entry->node);
2142                         call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2143                 }
2144         }
2145         __disable_ftrace_function_probe();
2146         mutex_unlock(&ftrace_lock);
2147 }
2148
2149 void
2150 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2151                                 void *data)
2152 {
2153         __unregister_ftrace_function_probe(glob, ops, data,
2154                                           PROBE_TEST_FUNC | PROBE_TEST_DATA);
2155 }
2156
2157 void
2158 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2159 {
2160         __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2161 }
2162
2163 void unregister_ftrace_function_probe_all(char *glob)
2164 {
2165         __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2166 }
2167
2168 static LIST_HEAD(ftrace_commands);
2169 static DEFINE_MUTEX(ftrace_cmd_mutex);
2170
2171 int register_ftrace_command(struct ftrace_func_command *cmd)
2172 {
2173         struct ftrace_func_command *p;
2174         int ret = 0;
2175
2176         mutex_lock(&ftrace_cmd_mutex);
2177         list_for_each_entry(p, &ftrace_commands, list) {
2178                 if (strcmp(cmd->name, p->name) == 0) {
2179                         ret = -EBUSY;
2180                         goto out_unlock;
2181                 }
2182         }
2183         list_add(&cmd->list, &ftrace_commands);
2184  out_unlock:
2185         mutex_unlock(&ftrace_cmd_mutex);
2186
2187         return ret;
2188 }
2189
2190 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2191 {
2192         struct ftrace_func_command *p, *n;
2193         int ret = -ENODEV;
2194
2195         mutex_lock(&ftrace_cmd_mutex);
2196         list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2197                 if (strcmp(cmd->name, p->name) == 0) {
2198                         ret = 0;
2199                         list_del_init(&p->list);
2200                         goto out_unlock;
2201                 }
2202         }
2203  out_unlock:
2204         mutex_unlock(&ftrace_cmd_mutex);
2205
2206         return ret;
2207 }
2208
2209 static int ftrace_process_regex(char *buff, int len, int enable)
2210 {
2211         char *func, *command, *next = buff;
2212         struct ftrace_func_command *p;
2213         int ret = -EINVAL;
2214
2215         func = strsep(&next, ":");
2216
2217         if (!next) {
2218                 ftrace_match_records(func, len, enable);
2219                 return 0;
2220         }
2221
2222         /* command found */
2223
2224         command = strsep(&next, ":");
2225
2226         mutex_lock(&ftrace_cmd_mutex);
2227         list_for_each_entry(p, &ftrace_commands, list) {
2228                 if (strcmp(p->name, command) == 0) {
2229                         ret = p->func(func, command, next, enable);
2230                         goto out_unlock;
2231                 }
2232         }
2233  out_unlock:
2234         mutex_unlock(&ftrace_cmd_mutex);
2235
2236         return ret;
2237 }
2238
2239 static ssize_t
2240 ftrace_regex_write(struct file *file, const char __user *ubuf,
2241                    size_t cnt, loff_t *ppos, int enable)
2242 {
2243         struct ftrace_iterator *iter;
2244         char ch;
2245         size_t read = 0;
2246         ssize_t ret;
2247
2248         if (!cnt || cnt < 0)
2249                 return 0;
2250
2251         mutex_lock(&ftrace_regex_lock);
2252
2253         if (file->f_mode & FMODE_READ) {
2254                 struct seq_file *m = file->private_data;
2255                 iter = m->private;
2256         } else
2257                 iter = file->private_data;
2258
2259         if (!*ppos) {
2260                 iter->flags &= ~FTRACE_ITER_CONT;
2261                 iter->buffer_idx = 0;
2262         }
2263
2264         ret = get_user(ch, ubuf++);
2265         if (ret)
2266                 goto out;
2267         read++;
2268         cnt--;
2269
2270         if (!(iter->flags & ~FTRACE_ITER_CONT)) {
2271                 /* skip white space */
2272                 while (cnt && isspace(ch)) {
2273                         ret = get_user(ch, ubuf++);
2274                         if (ret)
2275                                 goto out;
2276                         read++;
2277                         cnt--;
2278                 }
2279
2280                 if (isspace(ch)) {
2281                         file->f_pos += read;
2282                         ret = read;
2283                         goto out;
2284                 }
2285
2286                 iter->buffer_idx = 0;
2287         }
2288
2289         while (cnt && !isspace(ch)) {
2290                 if (iter->buffer_idx < FTRACE_BUFF_MAX)
2291                         iter->buffer[iter->buffer_idx++] = ch;
2292                 else {
2293                         ret = -EINVAL;
2294                         goto out;
2295                 }
2296                 ret = get_user(ch, ubuf++);
2297                 if (ret)
2298                         goto out;
2299                 read++;
2300                 cnt--;
2301         }
2302
2303         if (isspace(ch)) {
2304                 iter->filtered++;
2305                 iter->buffer[iter->buffer_idx] = 0;
2306                 ret = ftrace_process_regex(iter->buffer,
2307                                            iter->buffer_idx, enable);
2308                 if (ret)
2309                         goto out;
2310                 iter->buffer_idx = 0;
2311         } else
2312                 iter->flags |= FTRACE_ITER_CONT;
2313
2314
2315         file->f_pos += read;
2316
2317         ret = read;
2318  out:
2319         mutex_unlock(&ftrace_regex_lock);
2320
2321         return ret;
2322 }
2323
2324 static ssize_t
2325 ftrace_filter_write(struct file *file, const char __user *ubuf,
2326                     size_t cnt, loff_t *ppos)
2327 {
2328         return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2329 }
2330
2331 static ssize_t
2332 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2333                      size_t cnt, loff_t *ppos)
2334 {
2335         return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2336 }
2337
2338 static void
2339 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2340 {
2341         if (unlikely(ftrace_disabled))
2342                 return;
2343
2344         mutex_lock(&ftrace_regex_lock);
2345         if (reset)
2346                 ftrace_filter_reset(enable);
2347         if (buf)
2348                 ftrace_match_records(buf, len, enable);
2349         mutex_unlock(&ftrace_regex_lock);
2350 }
2351
2352 /**
2353  * ftrace_set_filter - set a function to filter on in ftrace
2354  * @buf - the string that holds the function filter text.
2355  * @len - the length of the string.
2356  * @reset - non zero to reset all filters before applying this filter.
2357  *
2358  * Filters denote which functions should be enabled when tracing is enabled.
2359  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2360  */
2361 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2362 {
2363         ftrace_set_regex(buf, len, reset, 1);
2364 }
2365
2366 /**
2367  * ftrace_set_notrace - set a function to not trace in ftrace
2368  * @buf - the string that holds the function notrace text.
2369  * @len - the length of the string.
2370  * @reset - non zero to reset all filters before applying this filter.
2371  *
2372  * Notrace Filters denote which functions should not be enabled when tracing
2373  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2374  * for tracing.
2375  */
2376 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2377 {
2378         ftrace_set_regex(buf, len, reset, 0);
2379 }
2380
2381 /*
2382  * command line interface to allow users to set filters on boot up.
2383  */
2384 #define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
2385 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2386 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2387
2388 static int __init set_ftrace_notrace(char *str)
2389 {
2390         strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2391         return 1;
2392 }
2393 __setup("ftrace_notrace=", set_ftrace_notrace);
2394
2395 static int __init set_ftrace_filter(char *str)
2396 {
2397         strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2398         return 1;
2399 }
2400 __setup("ftrace_filter=", set_ftrace_filter);
2401
2402 static void __init set_ftrace_early_filter(char *buf, int enable)
2403 {
2404         char *func;
2405
2406         while (buf) {
2407                 func = strsep(&buf, ",");
2408                 ftrace_set_regex(func, strlen(func), 0, enable);
2409         }
2410 }
2411
2412 static void __init set_ftrace_early_filters(void)
2413 {
2414         if (ftrace_filter_buf[0])
2415                 set_ftrace_early_filter(ftrace_filter_buf, 1);
2416         if (ftrace_notrace_buf[0])
2417                 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2418 }
2419
2420 static int
2421 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2422 {
2423         struct seq_file *m = (struct seq_file *)file->private_data;
2424         struct ftrace_iterator *iter;
2425
2426         mutex_lock(&ftrace_regex_lock);
2427         if (file->f_mode & FMODE_READ) {
2428                 iter = m->private;
2429
2430                 seq_release(inode, file);
2431         } else
2432                 iter = file->private_data;
2433
2434         if (iter->buffer_idx) {
2435                 iter->filtered++;
2436                 iter->buffer[iter->buffer_idx] = 0;
2437                 ftrace_match_records(iter->buffer, iter->buffer_idx, enable);
2438         }
2439
2440         mutex_lock(&ftrace_lock);
2441         if (ftrace_start_up && ftrace_enabled)
2442                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2443         mutex_unlock(&ftrace_lock);
2444
2445         kfree(iter);
2446         mutex_unlock(&ftrace_regex_lock);
2447         return 0;
2448 }
2449
2450 static int
2451 ftrace_filter_release(struct inode *inode, struct file *file)
2452 {
2453         return ftrace_regex_release(inode, file, 1);
2454 }
2455
2456 static int
2457 ftrace_notrace_release(struct inode *inode, struct file *file)
2458 {
2459         return ftrace_regex_release(inode, file, 0);
2460 }
2461
2462 static const struct file_operations ftrace_avail_fops = {
2463         .open = ftrace_avail_open,
2464         .read = seq_read,
2465         .llseek = seq_lseek,
2466         .release = ftrace_avail_release,
2467 };
2468
2469 static const struct file_operations ftrace_failures_fops = {
2470         .open = ftrace_failures_open,
2471         .read = seq_read,
2472         .llseek = seq_lseek,
2473         .release = ftrace_avail_release,
2474 };
2475
2476 static const struct file_operations ftrace_filter_fops = {
2477         .open = ftrace_filter_open,
2478         .read = seq_read,
2479         .write = ftrace_filter_write,
2480         .llseek = ftrace_regex_lseek,
2481         .release = ftrace_filter_release,
2482 };
2483
2484 static const struct file_operations ftrace_notrace_fops = {
2485         .open = ftrace_notrace_open,
2486         .read = seq_read,
2487         .write = ftrace_notrace_write,
2488         .llseek = ftrace_regex_lseek,
2489         .release = ftrace_notrace_release,
2490 };
2491
2492 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2493
2494 static DEFINE_MUTEX(graph_lock);
2495
2496 int ftrace_graph_count;
2497 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2498
2499 static void *
2500 __g_next(struct seq_file *m, loff_t *pos)
2501 {
2502         unsigned long *array = m->private;
2503
2504         if (*pos >= ftrace_graph_count)
2505                 return NULL;
2506         return &array[*pos];
2507 }
2508
2509 static void *
2510 g_next(struct seq_file *m, void *v, loff_t *pos)
2511 {
2512         (*pos)++;
2513         return __g_next(m, pos);
2514 }
2515
2516 static void *g_start(struct seq_file *m, loff_t *pos)
2517 {
2518         mutex_lock(&graph_lock);
2519
2520         /* Nothing, tell g_show to print all functions are enabled */
2521         if (!ftrace_graph_count && !*pos)
2522                 return (void *)1;
2523
2524         return __g_next(m, pos);
2525 }
2526
2527 static void g_stop(struct seq_file *m, void *p)
2528 {
2529         mutex_unlock(&graph_lock);
2530 }
2531
2532 static int g_show(struct seq_file *m, void *v)
2533 {
2534         unsigned long *ptr = v;
2535         char str[KSYM_SYMBOL_LEN];
2536
2537         if (!ptr)
2538                 return 0;
2539
2540         if (ptr == (unsigned long *)1) {
2541                 seq_printf(m, "#### all functions enabled ####\n");
2542                 return 0;
2543         }
2544
2545         kallsyms_lookup(*ptr, NULL, NULL, NULL, str);
2546
2547         seq_printf(m, "%s\n", str);
2548
2549         return 0;
2550 }
2551
2552 static struct seq_operations ftrace_graph_seq_ops = {
2553         .start = g_start,
2554         .next = g_next,
2555         .stop = g_stop,
2556         .show = g_show,
2557 };
2558
2559 static int
2560 ftrace_graph_open(struct inode *inode, struct file *file)
2561 {
2562         int ret = 0;
2563
2564         if (unlikely(ftrace_disabled))
2565                 return -ENODEV;
2566
2567         mutex_lock(&graph_lock);
2568         if ((file->f_mode & FMODE_WRITE) &&
2569             !(file->f_flags & O_APPEND)) {
2570                 ftrace_graph_count = 0;
2571                 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2572         }
2573
2574         if (file->f_mode & FMODE_READ) {
2575                 ret = seq_open(file, &ftrace_graph_seq_ops);
2576                 if (!ret) {
2577                         struct seq_file *m = file->private_data;
2578                         m->private = ftrace_graph_funcs;
2579                 }
2580         } else
2581                 file->private_data = ftrace_graph_funcs;
2582         mutex_unlock(&graph_lock);
2583
2584         return ret;
2585 }
2586
2587 static int
2588 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2589 {
2590         struct dyn_ftrace *rec;
2591         struct ftrace_page *pg;
2592         int search_len;
2593         int found = 0;
2594         int type, not;
2595         char *search;
2596         bool exists;
2597         int i;
2598
2599         if (ftrace_disabled)
2600                 return -ENODEV;
2601
2602         /* decode regex */
2603         type = ftrace_setup_glob(buffer, strlen(buffer), &search, &not);
2604         if (not)
2605                 return -EINVAL;
2606
2607         search_len = strlen(search);
2608
2609         mutex_lock(&ftrace_lock);
2610         do_for_each_ftrace_rec(pg, rec) {
2611
2612                 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2613                         break;
2614
2615                 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2616                         continue;
2617
2618                 if (ftrace_match_record(rec, search, search_len, type)) {
2619                         /* ensure it is not already in the array */
2620                         exists = false;
2621                         for (i = 0; i < *idx; i++)
2622                                 if (array[i] == rec->ip) {
2623                                         exists = true;
2624                                         break;
2625                                 }
2626                         if (!exists) {
2627                                 array[(*idx)++] = rec->ip;
2628                                 found = 1;
2629                         }
2630                 }
2631         } while_for_each_ftrace_rec();
2632
2633         mutex_unlock(&ftrace_lock);
2634
2635         return found ? 0 : -EINVAL;
2636 }
2637
2638 static ssize_t
2639 ftrace_graph_write(struct file *file, const char __user *ubuf,
2640                    size_t cnt, loff_t *ppos)
2641 {
2642         unsigned char buffer[FTRACE_BUFF_MAX+1];
2643         unsigned long *array;
2644         size_t read = 0;
2645         ssize_t ret;
2646         int index = 0;
2647         char ch;
2648
2649         if (!cnt || cnt < 0)
2650                 return 0;
2651
2652         mutex_lock(&graph_lock);
2653
2654         if (ftrace_graph_count >= FTRACE_GRAPH_MAX_FUNCS) {
2655                 ret = -EBUSY;
2656                 goto out;
2657         }
2658
2659         if (file->f_mode & FMODE_READ) {
2660                 struct seq_file *m = file->private_data;
2661                 array = m->private;
2662         } else
2663                 array = file->private_data;
2664
2665         ret = get_user(ch, ubuf++);
2666         if (ret)
2667                 goto out;
2668         read++;
2669         cnt--;
2670
2671         /* skip white space */
2672         while (cnt && isspace(ch)) {
2673                 ret = get_user(ch, ubuf++);
2674                 if (ret)
2675                         goto out;
2676                 read++;
2677                 cnt--;
2678         }
2679
2680         if (isspace(ch)) {
2681                 *ppos += read;
2682                 ret = read;
2683                 goto out;
2684         }
2685
2686         while (cnt && !isspace(ch)) {
2687                 if (index < FTRACE_BUFF_MAX)
2688                         buffer[index++] = ch;
2689                 else {
2690                         ret = -EINVAL;
2691                         goto out;
2692                 }
2693                 ret = get_user(ch, ubuf++);
2694                 if (ret)
2695                         goto out;
2696                 read++;
2697                 cnt--;
2698         }
2699         buffer[index] = 0;
2700
2701         /* we allow only one expression at a time */
2702         ret = ftrace_set_func(array, &ftrace_graph_count, buffer);
2703         if (ret)
2704                 goto out;
2705
2706         file->f_pos += read;
2707
2708         ret = read;
2709  out:
2710         mutex_unlock(&graph_lock);
2711
2712         return ret;
2713 }
2714
2715 static const struct file_operations ftrace_graph_fops = {
2716         .open = ftrace_graph_open,
2717         .read = seq_read,
2718         .write = ftrace_graph_write,
2719 };
2720 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2721
2722 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2723 {
2724
2725         trace_create_file("available_filter_functions", 0444,
2726                         d_tracer, NULL, &ftrace_avail_fops);
2727
2728         trace_create_file("failures", 0444,
2729                         d_tracer, NULL, &ftrace_failures_fops);
2730
2731         trace_create_file("set_ftrace_filter", 0644, d_tracer,
2732                         NULL, &ftrace_filter_fops);
2733
2734         trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2735                                     NULL, &ftrace_notrace_fops);
2736
2737 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2738         trace_create_file("set_graph_function", 0444, d_tracer,
2739                                     NULL,
2740                                     &ftrace_graph_fops);
2741 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2742
2743         return 0;
2744 }
2745
2746 static int ftrace_convert_nops(struct module *mod,
2747                                unsigned long *start,
2748                                unsigned long *end)
2749 {
2750         unsigned long *p;
2751         unsigned long addr;
2752         unsigned long flags;
2753
2754         mutex_lock(&ftrace_lock);
2755         p = start;
2756         while (p < end) {
2757                 addr = ftrace_call_adjust(*p++);
2758                 /*
2759                  * Some architecture linkers will pad between
2760                  * the different mcount_loc sections of different
2761                  * object files to satisfy alignments.
2762                  * Skip any NULL pointers.
2763                  */
2764                 if (!addr)
2765                         continue;
2766                 ftrace_record_ip(addr);
2767         }
2768
2769         /* disable interrupts to prevent kstop machine */
2770         local_irq_save(flags);
2771         ftrace_update_code(mod);
2772         local_irq_restore(flags);
2773         mutex_unlock(&ftrace_lock);
2774
2775         return 0;
2776 }
2777
2778 #ifdef CONFIG_MODULES
2779 void ftrace_release(void *start, void *end)
2780 {
2781         struct dyn_ftrace *rec;
2782         struct ftrace_page *pg;
2783         unsigned long s = (unsigned long)start;
2784         unsigned long e = (unsigned long)end;
2785
2786         if (ftrace_disabled || !start || start == end)
2787                 return;
2788
2789         mutex_lock(&ftrace_lock);
2790         do_for_each_ftrace_rec(pg, rec) {
2791                 if ((rec->ip >= s) && (rec->ip < e)) {
2792                         /*
2793                          * rec->ip is changed in ftrace_free_rec()
2794                          * It should not between s and e if record was freed.
2795                          */
2796                         FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2797                         ftrace_free_rec(rec);
2798                 }
2799         } while_for_each_ftrace_rec();
2800         mutex_unlock(&ftrace_lock);
2801 }
2802
2803 static void ftrace_init_module(struct module *mod,
2804                                unsigned long *start, unsigned long *end)
2805 {
2806         if (ftrace_disabled || start == end)
2807                 return;
2808         ftrace_convert_nops(mod, start, end);
2809 }
2810
2811 static int ftrace_module_notify(struct notifier_block *self,
2812                                 unsigned long val, void *data)
2813 {
2814         struct module *mod = data;
2815
2816         switch (val) {
2817         case MODULE_STATE_COMING:
2818                 ftrace_init_module(mod, mod->ftrace_callsites,
2819                                    mod->ftrace_callsites +
2820                                    mod->num_ftrace_callsites);
2821                 break;
2822         case MODULE_STATE_GOING:
2823                 ftrace_release(mod->ftrace_callsites,
2824                                mod->ftrace_callsites +
2825                                mod->num_ftrace_callsites);
2826                 break;
2827         }
2828
2829         return 0;
2830 }
2831 #else
2832 static int ftrace_module_notify(struct notifier_block *self,
2833                                 unsigned long val, void *data)
2834 {
2835         return 0;
2836 }
2837 #endif /* CONFIG_MODULES */
2838
2839 struct notifier_block ftrace_module_nb = {
2840         .notifier_call = ftrace_module_notify,
2841         .priority = 0,
2842 };
2843
2844 extern unsigned long __start_mcount_loc[];
2845 extern unsigned long __stop_mcount_loc[];
2846
2847 void __init ftrace_init(void)
2848 {
2849         unsigned long count, addr, flags;
2850         int ret;
2851
2852         /* Keep the ftrace pointer to the stub */
2853         addr = (unsigned long)ftrace_stub;
2854
2855         local_irq_save(flags);
2856         ftrace_dyn_arch_init(&addr);
2857         local_irq_restore(flags);
2858
2859         /* ftrace_dyn_arch_init places the return code in addr */
2860         if (addr)
2861                 goto failed;
2862
2863         count = __stop_mcount_loc - __start_mcount_loc;
2864
2865         ret = ftrace_dyn_table_alloc(count);
2866         if (ret)
2867                 goto failed;
2868
2869         last_ftrace_enabled = ftrace_enabled = 1;
2870
2871         ret = ftrace_convert_nops(NULL,
2872                                   __start_mcount_loc,
2873                                   __stop_mcount_loc);
2874
2875         ret = register_module_notifier(&ftrace_module_nb);
2876         if (ret)
2877                 pr_warning("Failed to register trace ftrace module notifier\n");
2878
2879         set_ftrace_early_filters();
2880
2881         return;
2882  failed:
2883         ftrace_disabled = 1;
2884 }
2885
2886 #else
2887
2888 static int __init ftrace_nodyn_init(void)
2889 {
2890         ftrace_enabled = 1;
2891         return 0;
2892 }
2893 device_initcall(ftrace_nodyn_init);
2894
2895 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2896 static inline void ftrace_startup_enable(int command) { }
2897 /* Keep as macros so we do not need to define the commands */
2898 # define ftrace_startup(command)        do { } while (0)
2899 # define ftrace_shutdown(command)       do { } while (0)
2900 # define ftrace_startup_sysctl()        do { } while (0)
2901 # define ftrace_shutdown_sysctl()       do { } while (0)
2902 #endif /* CONFIG_DYNAMIC_FTRACE */
2903
2904 static ssize_t
2905 ftrace_pid_read(struct file *file, char __user *ubuf,
2906                        size_t cnt, loff_t *ppos)
2907 {
2908         char buf[64];
2909         int r;
2910
2911         if (ftrace_pid_trace == ftrace_swapper_pid)
2912                 r = sprintf(buf, "swapper tasks\n");
2913         else if (ftrace_pid_trace)
2914                 r = sprintf(buf, "%u\n", pid_vnr(ftrace_pid_trace));
2915         else
2916                 r = sprintf(buf, "no pid\n");
2917
2918         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2919 }
2920
2921 static void clear_ftrace_swapper(void)
2922 {
2923         struct task_struct *p;
2924         int cpu;
2925
2926         get_online_cpus();
2927         for_each_online_cpu(cpu) {
2928                 p = idle_task(cpu);
2929                 clear_tsk_trace_trace(p);
2930         }
2931         put_online_cpus();
2932 }
2933
2934 static void set_ftrace_swapper(void)
2935 {
2936         struct task_struct *p;
2937         int cpu;
2938
2939         get_online_cpus();
2940         for_each_online_cpu(cpu) {
2941                 p = idle_task(cpu);
2942                 set_tsk_trace_trace(p);
2943         }
2944         put_online_cpus();
2945 }
2946
2947 static void clear_ftrace_pid(struct pid *pid)
2948 {
2949         struct task_struct *p;
2950
2951         rcu_read_lock();
2952         do_each_pid_task(pid, PIDTYPE_PID, p) {
2953                 clear_tsk_trace_trace(p);
2954         } while_each_pid_task(pid, PIDTYPE_PID, p);
2955         rcu_read_unlock();
2956
2957         put_pid(pid);
2958 }
2959
2960 static void set_ftrace_pid(struct pid *pid)
2961 {
2962         struct task_struct *p;
2963
2964         rcu_read_lock();
2965         do_each_pid_task(pid, PIDTYPE_PID, p) {
2966                 set_tsk_trace_trace(p);
2967         } while_each_pid_task(pid, PIDTYPE_PID, p);
2968         rcu_read_unlock();
2969 }
2970
2971 static void clear_ftrace_pid_task(struct pid **pid)
2972 {
2973         if (*pid == ftrace_swapper_pid)
2974                 clear_ftrace_swapper();
2975         else
2976                 clear_ftrace_pid(*pid);
2977
2978         *pid = NULL;
2979 }
2980
2981 static void set_ftrace_pid_task(struct pid *pid)
2982 {
2983         if (pid == ftrace_swapper_pid)
2984                 set_ftrace_swapper();
2985         else
2986                 set_ftrace_pid(pid);
2987 }
2988
2989 static ssize_t
2990 ftrace_pid_write(struct file *filp, const char __user *ubuf,
2991                    size_t cnt, loff_t *ppos)
2992 {
2993         struct pid *pid;
2994         char buf[64];
2995         long val;
2996         int ret;
2997
2998         if (cnt >= sizeof(buf))
2999                 return -EINVAL;
3000
3001         if (copy_from_user(&buf, ubuf, cnt))
3002                 return -EFAULT;
3003
3004         buf[cnt] = 0;
3005
3006         ret = strict_strtol(buf, 10, &val);
3007         if (ret < 0)
3008                 return ret;
3009
3010         mutex_lock(&ftrace_lock);
3011         if (val < 0) {
3012                 /* disable pid tracing */
3013                 if (!ftrace_pid_trace)
3014                         goto out;
3015
3016                 clear_ftrace_pid_task(&ftrace_pid_trace);
3017
3018         } else {
3019                 /* swapper task is special */
3020                 if (!val) {
3021                         pid = ftrace_swapper_pid;
3022                         if (pid == ftrace_pid_trace)
3023                                 goto out;
3024                 } else {
3025                         pid = find_get_pid(val);
3026
3027                         if (pid == ftrace_pid_trace) {
3028                                 put_pid(pid);
3029                                 goto out;
3030                         }
3031                 }
3032
3033                 if (ftrace_pid_trace)
3034                         clear_ftrace_pid_task(&ftrace_pid_trace);
3035
3036                 if (!pid)
3037                         goto out;
3038
3039                 ftrace_pid_trace = pid;
3040
3041                 set_ftrace_pid_task(ftrace_pid_trace);
3042         }
3043
3044         /* update the function call */
3045         ftrace_update_pid_func();
3046         ftrace_startup_enable(0);
3047
3048  out:
3049         mutex_unlock(&ftrace_lock);
3050
3051         return cnt;
3052 }
3053
3054 static const struct file_operations ftrace_pid_fops = {
3055         .read = ftrace_pid_read,
3056         .write = ftrace_pid_write,
3057 };
3058
3059 static __init int ftrace_init_debugfs(void)
3060 {
3061         struct dentry *d_tracer;
3062
3063         d_tracer = tracing_init_dentry();
3064         if (!d_tracer)
3065                 return 0;
3066
3067         ftrace_init_dyn_debugfs(d_tracer);
3068
3069         trace_create_file("set_ftrace_pid", 0644, d_tracer,
3070                             NULL, &ftrace_pid_fops);
3071
3072         ftrace_profile_debugfs(d_tracer);
3073
3074         return 0;
3075 }
3076 fs_initcall(ftrace_init_debugfs);
3077
3078 /**
3079  * ftrace_kill - kill ftrace
3080  *
3081  * This function should be used by panic code. It stops ftrace
3082  * but in a not so nice way. If you need to simply kill ftrace
3083  * from a non-atomic section, use ftrace_kill.
3084  */
3085 void ftrace_kill(void)
3086 {
3087         ftrace_disabled = 1;
3088         ftrace_enabled = 0;
3089         clear_ftrace_function();
3090 }
3091
3092 /**
3093  * register_ftrace_function - register a function for profiling
3094  * @ops - ops structure that holds the function for profiling.
3095  *
3096  * Register a function to be called by all functions in the
3097  * kernel.
3098  *
3099  * Note: @ops->func and all the functions it calls must be labeled
3100  *       with "notrace", otherwise it will go into a
3101  *       recursive loop.
3102  */
3103 int register_ftrace_function(struct ftrace_ops *ops)
3104 {
3105         int ret;
3106
3107         if (unlikely(ftrace_disabled))
3108                 return -1;
3109
3110         mutex_lock(&ftrace_lock);
3111
3112         ret = __register_ftrace_function(ops);
3113         ftrace_startup(0);
3114
3115         mutex_unlock(&ftrace_lock);
3116         return ret;
3117 }
3118
3119 /**
3120  * unregister_ftrace_function - unregister a function for profiling.
3121  * @ops - ops structure that holds the function to unregister
3122  *
3123  * Unregister a function that was added to be called by ftrace profiling.
3124  */
3125 int unregister_ftrace_function(struct ftrace_ops *ops)
3126 {
3127         int ret;
3128
3129         mutex_lock(&ftrace_lock);
3130         ret = __unregister_ftrace_function(ops);
3131         ftrace_shutdown(0);
3132         mutex_unlock(&ftrace_lock);
3133
3134         return ret;
3135 }
3136
3137 int
3138 ftrace_enable_sysctl(struct ctl_table *table, int write,
3139                      struct file *file, void __user *buffer, size_t *lenp,
3140                      loff_t *ppos)
3141 {
3142         int ret;
3143
3144         if (unlikely(ftrace_disabled))
3145                 return -ENODEV;
3146
3147         mutex_lock(&ftrace_lock);
3148
3149         ret  = proc_dointvec(table, write, file, buffer, lenp, ppos);
3150
3151         if (ret || !write || (last_ftrace_enabled == ftrace_enabled))
3152                 goto out;
3153
3154         last_ftrace_enabled = ftrace_enabled;
3155
3156         if (ftrace_enabled) {
3157
3158                 ftrace_startup_sysctl();
3159
3160                 /* we are starting ftrace again */
3161                 if (ftrace_list != &ftrace_list_end) {
3162                         if (ftrace_list->next == &ftrace_list_end)
3163                                 ftrace_trace_function = ftrace_list->func;
3164                         else
3165                                 ftrace_trace_function = ftrace_list_func;
3166                 }
3167
3168         } else {
3169                 /* stopping ftrace calls (just send to ftrace_stub) */
3170                 ftrace_trace_function = ftrace_stub;
3171
3172                 ftrace_shutdown_sysctl();
3173         }
3174
3175  out:
3176         mutex_unlock(&ftrace_lock);
3177         return ret;
3178 }
3179
3180 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3181
3182 static int ftrace_graph_active;
3183 static struct notifier_block ftrace_suspend_notifier;
3184
3185 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3186 {
3187         return 0;
3188 }
3189
3190 /* The callbacks that hook a function */
3191 trace_func_graph_ret_t ftrace_graph_return =
3192                         (trace_func_graph_ret_t)ftrace_stub;
3193 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3194
3195 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3196 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3197 {
3198         int i;
3199         int ret = 0;
3200         unsigned long flags;
3201         int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3202         struct task_struct *g, *t;
3203
3204         for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3205                 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3206                                         * sizeof(struct ftrace_ret_stack),
3207                                         GFP_KERNEL);
3208                 if (!ret_stack_list[i]) {
3209                         start = 0;
3210                         end = i;
3211                         ret = -ENOMEM;
3212                         goto free;
3213                 }
3214         }
3215
3216         read_lock_irqsave(&tasklist_lock, flags);
3217         do_each_thread(g, t) {
3218                 if (start == end) {
3219                         ret = -EAGAIN;
3220                         goto unlock;
3221                 }
3222
3223                 if (t->ret_stack == NULL) {
3224                         atomic_set(&t->tracing_graph_pause, 0);
3225                         atomic_set(&t->trace_overrun, 0);
3226                         t->curr_ret_stack = -1;
3227                         /* Make sure the tasks see the -1 first: */
3228                         smp_wmb();
3229                         t->ret_stack = ret_stack_list[start++];
3230                 }
3231         } while_each_thread(g, t);
3232
3233 unlock:
3234         read_unlock_irqrestore(&tasklist_lock, flags);
3235 free:
3236         for (i = start; i < end; i++)
3237                 kfree(ret_stack_list[i]);
3238         return ret;
3239 }
3240
3241 static void
3242 ftrace_graph_probe_sched_switch(struct rq *__rq, struct task_struct *prev,
3243                                 struct task_struct *next)
3244 {
3245         unsigned long long timestamp;
3246         int index;
3247
3248         /*
3249          * Does the user want to count the time a function was asleep.
3250          * If so, do not update the time stamps.
3251          */
3252         if (trace_flags & TRACE_ITER_SLEEP_TIME)
3253                 return;
3254
3255         timestamp = trace_clock_local();
3256
3257         prev->ftrace_timestamp = timestamp;
3258
3259         /* only process tasks that we timestamped */
3260         if (!next->ftrace_timestamp)
3261                 return;
3262
3263         /*
3264          * Update all the counters in next to make up for the
3265          * time next was sleeping.
3266          */
3267         timestamp -= next->ftrace_timestamp;
3268
3269         for (index = next->curr_ret_stack; index >= 0; index--)
3270                 next->ret_stack[index].calltime += timestamp;
3271 }
3272
3273 /* Allocate a return stack for each task */
3274 static int start_graph_tracing(void)
3275 {
3276         struct ftrace_ret_stack **ret_stack_list;
3277         int ret, cpu;
3278
3279         ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3280                                 sizeof(struct ftrace_ret_stack *),
3281                                 GFP_KERNEL);
3282
3283         if (!ret_stack_list)
3284                 return -ENOMEM;
3285
3286         /* The cpu_boot init_task->ret_stack will never be freed */
3287         for_each_online_cpu(cpu) {
3288                 if (!idle_task(cpu)->ret_stack)
3289                         ftrace_graph_init_task(idle_task(cpu));
3290         }
3291
3292         do {
3293                 ret = alloc_retstack_tasklist(ret_stack_list);
3294         } while (ret == -EAGAIN);
3295
3296         if (!ret) {
3297                 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch);
3298                 if (ret)
3299                         pr_info("ftrace_graph: Couldn't activate tracepoint"
3300                                 " probe to kernel_sched_switch\n");
3301         }
3302
3303         kfree(ret_stack_list);
3304         return ret;
3305 }
3306
3307 /*
3308  * Hibernation protection.
3309  * The state of the current task is too much unstable during
3310  * suspend/restore to disk. We want to protect against that.
3311  */
3312 static int
3313 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3314                                                         void *unused)
3315 {
3316         switch (state) {
3317         case PM_HIBERNATION_PREPARE:
3318                 pause_graph_tracing();
3319                 break;
3320
3321         case PM_POST_HIBERNATION:
3322                 unpause_graph_tracing();
3323                 break;
3324         }
3325         return NOTIFY_DONE;
3326 }
3327
3328 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3329                         trace_func_graph_ent_t entryfunc)
3330 {
3331         int ret = 0;
3332
3333         mutex_lock(&ftrace_lock);
3334
3335         /* we currently allow only one tracer registered at a time */
3336         if (ftrace_graph_active) {
3337                 ret = -EBUSY;
3338                 goto out;
3339         }
3340
3341         ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3342         register_pm_notifier(&ftrace_suspend_notifier);
3343
3344         ftrace_graph_active++;
3345         ret = start_graph_tracing();
3346         if (ret) {
3347                 ftrace_graph_active--;
3348                 goto out;
3349         }
3350
3351         ftrace_graph_return = retfunc;
3352         ftrace_graph_entry = entryfunc;
3353
3354         ftrace_startup(FTRACE_START_FUNC_RET);
3355
3356 out:
3357         mutex_unlock(&ftrace_lock);
3358         return ret;
3359 }
3360
3361 void unregister_ftrace_graph(void)
3362 {
3363         mutex_lock(&ftrace_lock);
3364
3365         if (unlikely(!ftrace_graph_active))
3366                 goto out;
3367
3368         ftrace_graph_active--;
3369         unregister_trace_sched_switch(ftrace_graph_probe_sched_switch);
3370         ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3371         ftrace_graph_entry = ftrace_graph_entry_stub;
3372         ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3373         unregister_pm_notifier(&ftrace_suspend_notifier);
3374
3375  out:
3376         mutex_unlock(&ftrace_lock);
3377 }
3378
3379 /* Allocate a return stack for newly created task */
3380 void ftrace_graph_init_task(struct task_struct *t)
3381 {
3382         /* Make sure we do not use the parent ret_stack */
3383         t->ret_stack = NULL;
3384
3385         if (ftrace_graph_active) {
3386                 struct ftrace_ret_stack *ret_stack;
3387
3388                 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3389                                 * sizeof(struct ftrace_ret_stack),
3390                                 GFP_KERNEL);
3391                 if (!ret_stack)
3392                         return;
3393                 t->curr_ret_stack = -1;
3394                 atomic_set(&t->tracing_graph_pause, 0);
3395                 atomic_set(&t->trace_overrun, 0);
3396                 t->ftrace_timestamp = 0;
3397                 /* make curr_ret_stack visable before we add the ret_stack */
3398                 smp_wmb();
3399                 t->ret_stack = ret_stack;
3400         }
3401 }
3402
3403 void ftrace_graph_exit_task(struct task_struct *t)
3404 {
3405         struct ftrace_ret_stack *ret_stack = t->ret_stack;
3406
3407         t->ret_stack = NULL;
3408         /* NULL must become visible to IRQs before we free it: */
3409         barrier();
3410
3411         kfree(ret_stack);
3412 }
3413
3414 void ftrace_graph_stop(void)
3415 {
3416         ftrace_stop();
3417 }
3418 #endif
3419