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