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