ftrace: Don't increment @pos in g_start()
[linux-2.6.git] / kernel / trace / ftrace.c
1 /*
2  * Infrastructure for profiling code inserted by 'gcc -pg'.
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally ported from the -rt patch by:
8  *   Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code in the latency_tracer, that is:
11  *
12  *  Copyright (C) 2004-2006 Ingo Molnar
13  *  Copyright (C) 2004 William Lee Irwin III
14  */
15
16 #include <linux/stop_machine.h>
17 #include <linux/clocksource.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/suspend.h>
21 #include <linux/debugfs.h>
22 #include <linux/hardirq.h>
23 #include <linux/kthread.h>
24 #include <linux/uaccess.h>
25 #include <linux/kprobes.h>
26 #include <linux/ftrace.h>
27 #include <linux/sysctl.h>
28 #include <linux/ctype.h>
29 #include <linux/list.h>
30 #include <linux/hash.h>
31
32 #include <trace/events/sched.h>
33
34 #include <asm/ftrace.h>
35 #include <asm/setup.h>
36
37 #include "trace_output.h"
38 #include "trace_stat.h"
39
40 #define FTRACE_WARN_ON(cond)                    \
41         do {                                    \
42                 if (WARN_ON(cond))              \
43                         ftrace_kill();          \
44         } while (0)
45
46 #define FTRACE_WARN_ON_ONCE(cond)               \
47         do {                                    \
48                 if (WARN_ON_ONCE(cond))         \
49                         ftrace_kill();          \
50         } while (0)
51
52 /* hash bits for specific function selection */
53 #define FTRACE_HASH_BITS 7
54 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
55
56 /* ftrace_enabled is a method to turn ftrace on or off */
57 int ftrace_enabled __read_mostly;
58 static int last_ftrace_enabled;
59
60 /* Quick disabling of function tracer. */
61 int function_trace_stop;
62
63 /*
64  * ftrace_disabled is set when an anomaly is discovered.
65  * ftrace_disabled is much stronger than ftrace_enabled.
66  */
67 static int ftrace_disabled __read_mostly;
68
69 static DEFINE_MUTEX(ftrace_lock);
70
71 static struct ftrace_ops ftrace_list_end __read_mostly =
72 {
73         .func           = ftrace_stub,
74 };
75
76 static struct ftrace_ops *ftrace_list __read_mostly = &ftrace_list_end;
77 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
78 ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
79 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
80
81 static void ftrace_list_func(unsigned long ip, unsigned long parent_ip)
82 {
83         struct ftrace_ops *op = ftrace_list;
84
85         /* in case someone actually ports this to alpha! */
86         read_barrier_depends();
87
88         while (op != &ftrace_list_end) {
89                 /* silly alpha */
90                 read_barrier_depends();
91                 op->func(ip, parent_ip);
92                 op = op->next;
93         };
94 }
95
96 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
97 {
98         if (!test_tsk_trace_trace(current))
99                 return;
100
101         ftrace_pid_function(ip, parent_ip);
102 }
103
104 static void set_ftrace_pid_function(ftrace_func_t func)
105 {
106         /* do not set ftrace_pid_function to itself! */
107         if (func != ftrace_pid_func)
108                 ftrace_pid_function = func;
109 }
110
111 /**
112  * clear_ftrace_function - reset the ftrace function
113  *
114  * This NULLs the ftrace function and in essence stops
115  * tracing.  There may be lag
116  */
117 void clear_ftrace_function(void)
118 {
119         ftrace_trace_function = ftrace_stub;
120         __ftrace_trace_function = ftrace_stub;
121         ftrace_pid_function = ftrace_stub;
122 }
123
124 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
125 /*
126  * For those archs that do not test ftrace_trace_stop in their
127  * mcount call site, we need to do it from C.
128  */
129 static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
130 {
131         if (function_trace_stop)
132                 return;
133
134         __ftrace_trace_function(ip, parent_ip);
135 }
136 #endif
137
138 static int __register_ftrace_function(struct ftrace_ops *ops)
139 {
140         ops->next = ftrace_list;
141         /*
142          * We are entering ops into the ftrace_list but another
143          * CPU might be walking that list. We need to make sure
144          * the ops->next pointer is valid before another CPU sees
145          * the ops pointer included into the ftrace_list.
146          */
147         smp_wmb();
148         ftrace_list = ops;
149
150         if (ftrace_enabled) {
151                 ftrace_func_t func;
152
153                 if (ops->next == &ftrace_list_end)
154                         func = ops->func;
155                 else
156                         func = ftrace_list_func;
157
158                 if (ftrace_pid_trace) {
159                         set_ftrace_pid_function(func);
160                         func = ftrace_pid_func;
161                 }
162
163                 /*
164                  * For one func, simply call it directly.
165                  * For more than one func, call the chain.
166                  */
167 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
168                 ftrace_trace_function = func;
169 #else
170                 __ftrace_trace_function = func;
171                 ftrace_trace_function = ftrace_test_stop_func;
172 #endif
173         }
174
175         return 0;
176 }
177
178 static int __unregister_ftrace_function(struct ftrace_ops *ops)
179 {
180         struct ftrace_ops **p;
181
182         /*
183          * If we are removing the last function, then simply point
184          * to the ftrace_stub.
185          */
186         if (ftrace_list == ops && ops->next == &ftrace_list_end) {
187                 ftrace_trace_function = ftrace_stub;
188                 ftrace_list = &ftrace_list_end;
189                 return 0;
190         }
191
192         for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next)
193                 if (*p == ops)
194                         break;
195
196         if (*p != ops)
197                 return -1;
198
199         *p = (*p)->next;
200
201         if (ftrace_enabled) {
202                 /* If we only have one func left, then call that directly */
203                 if (ftrace_list->next == &ftrace_list_end) {
204                         ftrace_func_t func = ftrace_list->func;
205
206                         if (ftrace_pid_trace) {
207                                 set_ftrace_pid_function(func);
208                                 func = ftrace_pid_func;
209                         }
210 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
211                         ftrace_trace_function = func;
212 #else
213                         __ftrace_trace_function = func;
214 #endif
215                 }
216         }
217
218         return 0;
219 }
220
221 static void ftrace_update_pid_func(void)
222 {
223         ftrace_func_t func;
224
225         if (ftrace_trace_function == ftrace_stub)
226                 return;
227
228         func = ftrace_trace_function;
229
230         if (ftrace_pid_trace) {
231                 set_ftrace_pid_function(func);
232                 func = ftrace_pid_func;
233         } else {
234                 if (func == ftrace_pid_func)
235                         func = ftrace_pid_function;
236         }
237
238 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
239         ftrace_trace_function = func;
240 #else
241         __ftrace_trace_function = func;
242 #endif
243 }
244
245 #ifdef CONFIG_FUNCTION_PROFILER
246 struct ftrace_profile {
247         struct hlist_node               node;
248         unsigned long                   ip;
249         unsigned long                   counter;
250 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
251         unsigned long long              time;
252 #endif
253 };
254
255 struct ftrace_profile_page {
256         struct ftrace_profile_page      *next;
257         unsigned long                   index;
258         struct ftrace_profile           records[];
259 };
260
261 struct ftrace_profile_stat {
262         atomic_t                        disabled;
263         struct hlist_head               *hash;
264         struct ftrace_profile_page      *pages;
265         struct ftrace_profile_page      *start;
266         struct tracer_stat              stat;
267 };
268
269 #define PROFILE_RECORDS_SIZE                                            \
270         (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
271
272 #define PROFILES_PER_PAGE                                       \
273         (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
274
275 static int ftrace_profile_bits __read_mostly;
276 static int ftrace_profile_enabled __read_mostly;
277
278 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
279 static DEFINE_MUTEX(ftrace_profile_lock);
280
281 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
282
283 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
284
285 static void *
286 function_stat_next(void *v, int idx)
287 {
288         struct ftrace_profile *rec = v;
289         struct ftrace_profile_page *pg;
290
291         pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
292
293  again:
294         rec++;
295         if ((void *)rec >= (void *)&pg->records[pg->index]) {
296                 pg = pg->next;
297                 if (!pg)
298                         return NULL;
299                 rec = &pg->records[0];
300                 if (!rec->counter)
301                         goto again;
302         }
303
304         return rec;
305 }
306
307 static void *function_stat_start(struct tracer_stat *trace)
308 {
309         struct ftrace_profile_stat *stat =
310                 container_of(trace, struct ftrace_profile_stat, stat);
311
312         if (!stat || !stat->start)
313                 return NULL;
314
315         return function_stat_next(&stat->start->records[0], 0);
316 }
317
318 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
319 /* function graph compares on total time */
320 static int function_stat_cmp(void *p1, void *p2)
321 {
322         struct ftrace_profile *a = p1;
323         struct ftrace_profile *b = p2;
324
325         if (a->time < b->time)
326                 return -1;
327         if (a->time > b->time)
328                 return 1;
329         else
330                 return 0;
331 }
332 #else
333 /* not function graph compares against hits */
334 static int function_stat_cmp(void *p1, void *p2)
335 {
336         struct ftrace_profile *a = p1;
337         struct ftrace_profile *b = p2;
338
339         if (a->counter < b->counter)
340                 return -1;
341         if (a->counter > b->counter)
342                 return 1;
343         else
344                 return 0;
345 }
346 #endif
347
348 static int function_stat_headers(struct seq_file *m)
349 {
350 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
351         seq_printf(m, "  Function                               "
352                    "Hit    Time            Avg\n"
353                       "  --------                               "
354                    "---    ----            ---\n");
355 #else
356         seq_printf(m, "  Function                               Hit\n"
357                       "  --------                               ---\n");
358 #endif
359         return 0;
360 }
361
362 static int function_stat_show(struct seq_file *m, void *v)
363 {
364         struct ftrace_profile *rec = v;
365         char str[KSYM_SYMBOL_LEN];
366 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
367         static DEFINE_MUTEX(mutex);
368         static struct trace_seq s;
369         unsigned long long avg;
370 #endif
371
372         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
373         seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
374
375 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
376         seq_printf(m, "    ");
377         avg = rec->time;
378         do_div(avg, rec->counter);
379
380         mutex_lock(&mutex);
381         trace_seq_init(&s);
382         trace_print_graph_duration(rec->time, &s);
383         trace_seq_puts(&s, "    ");
384         trace_print_graph_duration(avg, &s);
385         trace_print_seq(m, &s);
386         mutex_unlock(&mutex);
387 #endif
388         seq_putc(m, '\n');
389
390         return 0;
391 }
392
393 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
394 {
395         struct ftrace_profile_page *pg;
396
397         pg = stat->pages = stat->start;
398
399         while (pg) {
400                 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
401                 pg->index = 0;
402                 pg = pg->next;
403         }
404
405         memset(stat->hash, 0,
406                FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
407 }
408
409 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
410 {
411         struct ftrace_profile_page *pg;
412         int functions;
413         int pages;
414         int i;
415
416         /* If we already allocated, do nothing */
417         if (stat->pages)
418                 return 0;
419
420         stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
421         if (!stat->pages)
422                 return -ENOMEM;
423
424 #ifdef CONFIG_DYNAMIC_FTRACE
425         functions = ftrace_update_tot_cnt;
426 #else
427         /*
428          * We do not know the number of functions that exist because
429          * dynamic tracing is what counts them. With past experience
430          * we have around 20K functions. That should be more than enough.
431          * It is highly unlikely we will execute every function in
432          * the kernel.
433          */
434         functions = 20000;
435 #endif
436
437         pg = stat->start = stat->pages;
438
439         pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
440
441         for (i = 0; i < pages; i++) {
442                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
443                 if (!pg->next)
444                         goto out_free;
445                 pg = pg->next;
446         }
447
448         return 0;
449
450  out_free:
451         pg = stat->start;
452         while (pg) {
453                 unsigned long tmp = (unsigned long)pg;
454
455                 pg = pg->next;
456                 free_page(tmp);
457         }
458
459         free_page((unsigned long)stat->pages);
460         stat->pages = NULL;
461         stat->start = NULL;
462
463         return -ENOMEM;
464 }
465
466 static int ftrace_profile_init_cpu(int cpu)
467 {
468         struct ftrace_profile_stat *stat;
469         int size;
470
471         stat = &per_cpu(ftrace_profile_stats, cpu);
472
473         if (stat->hash) {
474                 /* If the profile is already created, simply reset it */
475                 ftrace_profile_reset(stat);
476                 return 0;
477         }
478
479         /*
480          * We are profiling all functions, but usually only a few thousand
481          * functions are hit. We'll make a hash of 1024 items.
482          */
483         size = FTRACE_PROFILE_HASH_SIZE;
484
485         stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
486
487         if (!stat->hash)
488                 return -ENOMEM;
489
490         if (!ftrace_profile_bits) {
491                 size--;
492
493                 for (; size; size >>= 1)
494                         ftrace_profile_bits++;
495         }
496
497         /* Preallocate the function profiling pages */
498         if (ftrace_profile_pages_init(stat) < 0) {
499                 kfree(stat->hash);
500                 stat->hash = NULL;
501                 return -ENOMEM;
502         }
503
504         return 0;
505 }
506
507 static int ftrace_profile_init(void)
508 {
509         int cpu;
510         int ret = 0;
511
512         for_each_online_cpu(cpu) {
513                 ret = ftrace_profile_init_cpu(cpu);
514                 if (ret)
515                         break;
516         }
517
518         return ret;
519 }
520
521 /* interrupts must be disabled */
522 static struct ftrace_profile *
523 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
524 {
525         struct ftrace_profile *rec;
526         struct hlist_head *hhd;
527         struct hlist_node *n;
528         unsigned long key;
529
530         key = hash_long(ip, ftrace_profile_bits);
531         hhd = &stat->hash[key];
532
533         if (hlist_empty(hhd))
534                 return NULL;
535
536         hlist_for_each_entry_rcu(rec, n, hhd, node) {
537                 if (rec->ip == ip)
538                         return rec;
539         }
540
541         return NULL;
542 }
543
544 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
545                                struct ftrace_profile *rec)
546 {
547         unsigned long key;
548
549         key = hash_long(rec->ip, ftrace_profile_bits);
550         hlist_add_head_rcu(&rec->node, &stat->hash[key]);
551 }
552
553 /*
554  * The memory is already allocated, this simply finds a new record to use.
555  */
556 static struct ftrace_profile *
557 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
558 {
559         struct ftrace_profile *rec = NULL;
560
561         /* prevent recursion (from NMIs) */
562         if (atomic_inc_return(&stat->disabled) != 1)
563                 goto out;
564
565         /*
566          * Try to find the function again since an NMI
567          * could have added it
568          */
569         rec = ftrace_find_profiled_func(stat, ip);
570         if (rec)
571                 goto out;
572
573         if (stat->pages->index == PROFILES_PER_PAGE) {
574                 if (!stat->pages->next)
575                         goto out;
576                 stat->pages = stat->pages->next;
577         }
578
579         rec = &stat->pages->records[stat->pages->index++];
580         rec->ip = ip;
581         ftrace_add_profile(stat, rec);
582
583  out:
584         atomic_dec(&stat->disabled);
585
586         return rec;
587 }
588
589 static void
590 function_profile_call(unsigned long ip, unsigned long parent_ip)
591 {
592         struct ftrace_profile_stat *stat;
593         struct ftrace_profile *rec;
594         unsigned long flags;
595
596         if (!ftrace_profile_enabled)
597                 return;
598
599         local_irq_save(flags);
600
601         stat = &__get_cpu_var(ftrace_profile_stats);
602         if (!stat->hash || !ftrace_profile_enabled)
603                 goto out;
604
605         rec = ftrace_find_profiled_func(stat, ip);
606         if (!rec) {
607                 rec = ftrace_profile_alloc(stat, ip);
608                 if (!rec)
609                         goto out;
610         }
611
612         rec->counter++;
613  out:
614         local_irq_restore(flags);
615 }
616
617 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
618 static int profile_graph_entry(struct ftrace_graph_ent *trace)
619 {
620         function_profile_call(trace->func, 0);
621         return 1;
622 }
623
624 static void profile_graph_return(struct ftrace_graph_ret *trace)
625 {
626         struct ftrace_profile_stat *stat;
627         unsigned long long calltime;
628         struct ftrace_profile *rec;
629         unsigned long flags;
630
631         local_irq_save(flags);
632         stat = &__get_cpu_var(ftrace_profile_stats);
633         if (!stat->hash || !ftrace_profile_enabled)
634                 goto out;
635
636         calltime = trace->rettime - trace->calltime;
637
638         if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
639                 int index;
640
641                 index = trace->depth;
642
643                 /* Append this call time to the parent time to subtract */
644                 if (index)
645                         current->ret_stack[index - 1].subtime += calltime;
646
647                 if (current->ret_stack[index].subtime < calltime)
648                         calltime -= current->ret_stack[index].subtime;
649                 else
650                         calltime = 0;
651         }
652
653         rec = ftrace_find_profiled_func(stat, trace->func);
654         if (rec)
655                 rec->time += calltime;
656
657  out:
658         local_irq_restore(flags);
659 }
660
661 static int register_ftrace_profiler(void)
662 {
663         return register_ftrace_graph(&profile_graph_return,
664                                      &profile_graph_entry);
665 }
666
667 static void unregister_ftrace_profiler(void)
668 {
669         unregister_ftrace_graph();
670 }
671 #else
672 static struct ftrace_ops ftrace_profile_ops __read_mostly =
673 {
674         .func           = function_profile_call,
675 };
676
677 static int register_ftrace_profiler(void)
678 {
679         return register_ftrace_function(&ftrace_profile_ops);
680 }
681
682 static void unregister_ftrace_profiler(void)
683 {
684         unregister_ftrace_function(&ftrace_profile_ops);
685 }
686 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
687
688 static ssize_t
689 ftrace_profile_write(struct file *filp, const char __user *ubuf,
690                      size_t cnt, loff_t *ppos)
691 {
692         unsigned long val;
693         char buf[64];           /* big enough to hold a number */
694         int ret;
695
696         if (cnt >= sizeof(buf))
697                 return -EINVAL;
698
699         if (copy_from_user(&buf, ubuf, cnt))
700                 return -EFAULT;
701
702         buf[cnt] = 0;
703
704         ret = strict_strtoul(buf, 10, &val);
705         if (ret < 0)
706                 return ret;
707
708         val = !!val;
709
710         mutex_lock(&ftrace_profile_lock);
711         if (ftrace_profile_enabled ^ val) {
712                 if (val) {
713                         ret = ftrace_profile_init();
714                         if (ret < 0) {
715                                 cnt = ret;
716                                 goto out;
717                         }
718
719                         ret = register_ftrace_profiler();
720                         if (ret < 0) {
721                                 cnt = ret;
722                                 goto out;
723                         }
724                         ftrace_profile_enabled = 1;
725                 } else {
726                         ftrace_profile_enabled = 0;
727                         /*
728                          * unregister_ftrace_profiler calls stop_machine
729                          * so this acts like an synchronize_sched.
730                          */
731                         unregister_ftrace_profiler();
732                 }
733         }
734  out:
735         mutex_unlock(&ftrace_profile_lock);
736
737         filp->f_pos += cnt;
738
739         return cnt;
740 }
741
742 static ssize_t
743 ftrace_profile_read(struct file *filp, char __user *ubuf,
744                      size_t cnt, loff_t *ppos)
745 {
746         char buf[64];           /* big enough to hold a number */
747         int r;
748
749         r = sprintf(buf, "%u\n", ftrace_profile_enabled);
750         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
751 }
752
753 static const struct file_operations ftrace_profile_fops = {
754         .open           = tracing_open_generic,
755         .read           = ftrace_profile_read,
756         .write          = ftrace_profile_write,
757 };
758
759 /* used to initialize the real stat files */
760 static struct tracer_stat function_stats __initdata = {
761         .name           = "functions",
762         .stat_start     = function_stat_start,
763         .stat_next      = function_stat_next,
764         .stat_cmp       = function_stat_cmp,
765         .stat_headers   = function_stat_headers,
766         .stat_show      = function_stat_show
767 };
768
769 static void ftrace_profile_debugfs(struct dentry *d_tracer)
770 {
771         struct ftrace_profile_stat *stat;
772         struct dentry *entry;
773         char *name;
774         int ret;
775         int cpu;
776
777         for_each_possible_cpu(cpu) {
778                 stat = &per_cpu(ftrace_profile_stats, cpu);
779
780                 /* allocate enough for function name + cpu number */
781                 name = kmalloc(32, GFP_KERNEL);
782                 if (!name) {
783                         /*
784                          * The files created are permanent, if something happens
785                          * we still do not free memory.
786                          */
787                         kfree(stat);
788                         WARN(1,
789                              "Could not allocate stat file for cpu %d\n",
790                              cpu);
791                         return;
792                 }
793                 stat->stat = function_stats;
794                 snprintf(name, 32, "function%d", cpu);
795                 stat->stat.name = name;
796                 ret = register_stat_tracer(&stat->stat);
797                 if (ret) {
798                         WARN(1,
799                              "Could not register function stat for cpu %d\n",
800                              cpu);
801                         kfree(name);
802                         return;
803                 }
804         }
805
806         entry = debugfs_create_file("function_profile_enabled", 0644,
807                                     d_tracer, NULL, &ftrace_profile_fops);
808         if (!entry)
809                 pr_warning("Could not create debugfs "
810                            "'function_profile_enabled' entry\n");
811 }
812
813 #else /* CONFIG_FUNCTION_PROFILER */
814 static void ftrace_profile_debugfs(struct dentry *d_tracer)
815 {
816 }
817 #endif /* CONFIG_FUNCTION_PROFILER */
818
819 /* set when tracing only a pid */
820 struct pid *ftrace_pid_trace;
821 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
822
823 #ifdef CONFIG_DYNAMIC_FTRACE
824
825 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
826 # error Dynamic ftrace depends on MCOUNT_RECORD
827 #endif
828
829 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
830
831 struct ftrace_func_probe {
832         struct hlist_node       node;
833         struct ftrace_probe_ops *ops;
834         unsigned long           flags;
835         unsigned long           ip;
836         void                    *data;
837         struct rcu_head         rcu;
838 };
839
840 enum {
841         FTRACE_ENABLE_CALLS             = (1 << 0),
842         FTRACE_DISABLE_CALLS            = (1 << 1),
843         FTRACE_UPDATE_TRACE_FUNC        = (1 << 2),
844         FTRACE_ENABLE_MCOUNT            = (1 << 3),
845         FTRACE_DISABLE_MCOUNT           = (1 << 4),
846         FTRACE_START_FUNC_RET           = (1 << 5),
847         FTRACE_STOP_FUNC_RET            = (1 << 6),
848 };
849
850 static int ftrace_filtered;
851
852 static struct dyn_ftrace *ftrace_new_addrs;
853
854 static DEFINE_MUTEX(ftrace_regex_lock);
855
856 struct ftrace_page {
857         struct ftrace_page      *next;
858         int                     index;
859         struct dyn_ftrace       records[];
860 };
861
862 #define ENTRIES_PER_PAGE \
863   ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
864
865 /* estimate from running different kernels */
866 #define NR_TO_INIT              10000
867
868 static struct ftrace_page       *ftrace_pages_start;
869 static struct ftrace_page       *ftrace_pages;
870
871 static struct dyn_ftrace *ftrace_free_records;
872
873 /*
874  * This is a double for. Do not use 'break' to break out of the loop,
875  * you must use a goto.
876  */
877 #define do_for_each_ftrace_rec(pg, rec)                                 \
878         for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
879                 int _____i;                                             \
880                 for (_____i = 0; _____i < pg->index; _____i++) {        \
881                         rec = &pg->records[_____i];
882
883 #define while_for_each_ftrace_rec()             \
884                 }                               \
885         }
886
887 #ifdef CONFIG_KPROBES
888
889 static int frozen_record_count;
890
891 static inline void freeze_record(struct dyn_ftrace *rec)
892 {
893         if (!(rec->flags & FTRACE_FL_FROZEN)) {
894                 rec->flags |= FTRACE_FL_FROZEN;
895                 frozen_record_count++;
896         }
897 }
898
899 static inline void unfreeze_record(struct dyn_ftrace *rec)
900 {
901         if (rec->flags & FTRACE_FL_FROZEN) {
902                 rec->flags &= ~FTRACE_FL_FROZEN;
903                 frozen_record_count--;
904         }
905 }
906
907 static inline int record_frozen(struct dyn_ftrace *rec)
908 {
909         return rec->flags & FTRACE_FL_FROZEN;
910 }
911 #else
912 # define freeze_record(rec)                     ({ 0; })
913 # define unfreeze_record(rec)                   ({ 0; })
914 # define record_frozen(rec)                     ({ 0; })
915 #endif /* CONFIG_KPROBES */
916
917 static void ftrace_free_rec(struct dyn_ftrace *rec)
918 {
919         rec->freelist = ftrace_free_records;
920         ftrace_free_records = rec;
921         rec->flags |= FTRACE_FL_FREE;
922 }
923
924 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
925 {
926         struct dyn_ftrace *rec;
927
928         /* First check for freed records */
929         if (ftrace_free_records) {
930                 rec = ftrace_free_records;
931
932                 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
933                         FTRACE_WARN_ON_ONCE(1);
934                         ftrace_free_records = NULL;
935                         return NULL;
936                 }
937
938                 ftrace_free_records = rec->freelist;
939                 memset(rec, 0, sizeof(*rec));
940                 return rec;
941         }
942
943         if (ftrace_pages->index == ENTRIES_PER_PAGE) {
944                 if (!ftrace_pages->next) {
945                         /* allocate another page */
946                         ftrace_pages->next =
947                                 (void *)get_zeroed_page(GFP_KERNEL);
948                         if (!ftrace_pages->next)
949                                 return NULL;
950                 }
951                 ftrace_pages = ftrace_pages->next;
952         }
953
954         return &ftrace_pages->records[ftrace_pages->index++];
955 }
956
957 static struct dyn_ftrace *
958 ftrace_record_ip(unsigned long ip)
959 {
960         struct dyn_ftrace *rec;
961
962         if (ftrace_disabled)
963                 return NULL;
964
965         rec = ftrace_alloc_dyn_node(ip);
966         if (!rec)
967                 return NULL;
968
969         rec->ip = ip;
970         rec->newlist = ftrace_new_addrs;
971         ftrace_new_addrs = rec;
972
973         return rec;
974 }
975
976 static void print_ip_ins(const char *fmt, unsigned char *p)
977 {
978         int i;
979
980         printk(KERN_CONT "%s", fmt);
981
982         for (i = 0; i < MCOUNT_INSN_SIZE; i++)
983                 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
984 }
985
986 static void ftrace_bug(int failed, unsigned long ip)
987 {
988         switch (failed) {
989         case -EFAULT:
990                 FTRACE_WARN_ON_ONCE(1);
991                 pr_info("ftrace faulted on modifying ");
992                 print_ip_sym(ip);
993                 break;
994         case -EINVAL:
995                 FTRACE_WARN_ON_ONCE(1);
996                 pr_info("ftrace failed to modify ");
997                 print_ip_sym(ip);
998                 print_ip_ins(" actual: ", (unsigned char *)ip);
999                 printk(KERN_CONT "\n");
1000                 break;
1001         case -EPERM:
1002                 FTRACE_WARN_ON_ONCE(1);
1003                 pr_info("ftrace faulted on writing ");
1004                 print_ip_sym(ip);
1005                 break;
1006         default:
1007                 FTRACE_WARN_ON_ONCE(1);
1008                 pr_info("ftrace faulted on unknown error ");
1009                 print_ip_sym(ip);
1010         }
1011 }
1012
1013
1014 static int
1015 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1016 {
1017         unsigned long ftrace_addr;
1018         unsigned long ip, fl;
1019
1020         ftrace_addr = (unsigned long)FTRACE_ADDR;
1021
1022         ip = rec->ip;
1023
1024         /*
1025          * If this record is not to be traced and
1026          * it is not enabled then do nothing.
1027          *
1028          * If this record is not to be traced and
1029          * it is enabled then disable it.
1030          *
1031          */
1032         if (rec->flags & FTRACE_FL_NOTRACE) {
1033                 if (rec->flags & FTRACE_FL_ENABLED)
1034                         rec->flags &= ~FTRACE_FL_ENABLED;
1035                 else
1036                         return 0;
1037
1038         } else if (ftrace_filtered && enable) {
1039                 /*
1040                  * Filtering is on:
1041                  */
1042
1043                 fl = rec->flags & (FTRACE_FL_FILTER | FTRACE_FL_ENABLED);
1044
1045                 /* Record is filtered and enabled, do nothing */
1046                 if (fl == (FTRACE_FL_FILTER | FTRACE_FL_ENABLED))
1047                         return 0;
1048
1049                 /* Record is not filtered or enabled, do nothing */
1050                 if (!fl)
1051                         return 0;
1052
1053                 /* Record is not filtered but enabled, disable it */
1054                 if (fl == FTRACE_FL_ENABLED)
1055                         rec->flags &= ~FTRACE_FL_ENABLED;
1056                 else
1057                 /* Otherwise record is filtered but not enabled, enable it */
1058                         rec->flags |= FTRACE_FL_ENABLED;
1059         } else {
1060                 /* Disable or not filtered */
1061
1062                 if (enable) {
1063                         /* if record is enabled, do nothing */
1064                         if (rec->flags & FTRACE_FL_ENABLED)
1065                                 return 0;
1066
1067                         rec->flags |= FTRACE_FL_ENABLED;
1068
1069                 } else {
1070
1071                         /* if record is not enabled, do nothing */
1072                         if (!(rec->flags & FTRACE_FL_ENABLED))
1073                                 return 0;
1074
1075                         rec->flags &= ~FTRACE_FL_ENABLED;
1076                 }
1077         }
1078
1079         if (rec->flags & FTRACE_FL_ENABLED)
1080                 return ftrace_make_call(rec, ftrace_addr);
1081         else
1082                 return ftrace_make_nop(NULL, rec, ftrace_addr);
1083 }
1084
1085 static void ftrace_replace_code(int enable)
1086 {
1087         struct dyn_ftrace *rec;
1088         struct ftrace_page *pg;
1089         int failed;
1090
1091         do_for_each_ftrace_rec(pg, rec) {
1092                 /*
1093                  * Skip over free records, records that have
1094                  * failed and not converted.
1095                  */
1096                 if (rec->flags & FTRACE_FL_FREE ||
1097                     rec->flags & FTRACE_FL_FAILED ||
1098                     !(rec->flags & FTRACE_FL_CONVERTED))
1099                         continue;
1100
1101                 /* ignore updates to this record's mcount site */
1102                 if (get_kprobe((void *)rec->ip)) {
1103                         freeze_record(rec);
1104                         continue;
1105                 } else {
1106                         unfreeze_record(rec);
1107                 }
1108
1109                 failed = __ftrace_replace_code(rec, enable);
1110                 if (failed) {
1111                         rec->flags |= FTRACE_FL_FAILED;
1112                         if ((system_state == SYSTEM_BOOTING) ||
1113                             !core_kernel_text(rec->ip)) {
1114                                 ftrace_free_rec(rec);
1115                                 } else {
1116                                 ftrace_bug(failed, rec->ip);
1117                                         /* Stop processing */
1118                                         return;
1119                                 }
1120                 }
1121         } while_for_each_ftrace_rec();
1122 }
1123
1124 static int
1125 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1126 {
1127         unsigned long ip;
1128         int ret;
1129
1130         ip = rec->ip;
1131
1132         ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1133         if (ret) {
1134                 ftrace_bug(ret, ip);
1135                 rec->flags |= FTRACE_FL_FAILED;
1136                 return 0;
1137         }
1138         return 1;
1139 }
1140
1141 /*
1142  * archs can override this function if they must do something
1143  * before the modifying code is performed.
1144  */
1145 int __weak ftrace_arch_code_modify_prepare(void)
1146 {
1147         return 0;
1148 }
1149
1150 /*
1151  * archs can override this function if they must do something
1152  * after the modifying code is performed.
1153  */
1154 int __weak ftrace_arch_code_modify_post_process(void)
1155 {
1156         return 0;
1157 }
1158
1159 static int __ftrace_modify_code(void *data)
1160 {
1161         int *command = data;
1162
1163         if (*command & FTRACE_ENABLE_CALLS)
1164                 ftrace_replace_code(1);
1165         else if (*command & FTRACE_DISABLE_CALLS)
1166                 ftrace_replace_code(0);
1167
1168         if (*command & FTRACE_UPDATE_TRACE_FUNC)
1169                 ftrace_update_ftrace_func(ftrace_trace_function);
1170
1171         if (*command & FTRACE_START_FUNC_RET)
1172                 ftrace_enable_ftrace_graph_caller();
1173         else if (*command & FTRACE_STOP_FUNC_RET)
1174                 ftrace_disable_ftrace_graph_caller();
1175
1176         return 0;
1177 }
1178
1179 static void ftrace_run_update_code(int command)
1180 {
1181         int ret;
1182
1183         ret = ftrace_arch_code_modify_prepare();
1184         FTRACE_WARN_ON(ret);
1185         if (ret)
1186                 return;
1187
1188         stop_machine(__ftrace_modify_code, &command, NULL);
1189
1190         ret = ftrace_arch_code_modify_post_process();
1191         FTRACE_WARN_ON(ret);
1192 }
1193
1194 static ftrace_func_t saved_ftrace_func;
1195 static int ftrace_start_up;
1196
1197 static void ftrace_startup_enable(int command)
1198 {
1199         if (saved_ftrace_func != ftrace_trace_function) {
1200                 saved_ftrace_func = ftrace_trace_function;
1201                 command |= FTRACE_UPDATE_TRACE_FUNC;
1202         }
1203
1204         if (!command || !ftrace_enabled)
1205                 return;
1206
1207         ftrace_run_update_code(command);
1208 }
1209
1210 static void ftrace_startup(int command)
1211 {
1212         if (unlikely(ftrace_disabled))
1213                 return;
1214
1215         ftrace_start_up++;
1216         command |= FTRACE_ENABLE_CALLS;
1217
1218         ftrace_startup_enable(command);
1219 }
1220
1221 static void ftrace_shutdown(int command)
1222 {
1223         if (unlikely(ftrace_disabled))
1224                 return;
1225
1226         ftrace_start_up--;
1227         /*
1228          * Just warn in case of unbalance, no need to kill ftrace, it's not
1229          * critical but the ftrace_call callers may be never nopped again after
1230          * further ftrace uses.
1231          */
1232         WARN_ON_ONCE(ftrace_start_up < 0);
1233
1234         if (!ftrace_start_up)
1235                 command |= FTRACE_DISABLE_CALLS;
1236
1237         if (saved_ftrace_func != ftrace_trace_function) {
1238                 saved_ftrace_func = ftrace_trace_function;
1239                 command |= FTRACE_UPDATE_TRACE_FUNC;
1240         }
1241
1242         if (!command || !ftrace_enabled)
1243                 return;
1244
1245         ftrace_run_update_code(command);
1246 }
1247
1248 static void ftrace_startup_sysctl(void)
1249 {
1250         int command = FTRACE_ENABLE_MCOUNT;
1251
1252         if (unlikely(ftrace_disabled))
1253                 return;
1254
1255         /* Force update next time */
1256         saved_ftrace_func = NULL;
1257         /* ftrace_start_up is true if we want ftrace running */
1258         if (ftrace_start_up)
1259                 command |= FTRACE_ENABLE_CALLS;
1260
1261         ftrace_run_update_code(command);
1262 }
1263
1264 static void ftrace_shutdown_sysctl(void)
1265 {
1266         int command = FTRACE_DISABLE_MCOUNT;
1267
1268         if (unlikely(ftrace_disabled))
1269                 return;
1270
1271         /* ftrace_start_up is true if ftrace is running */
1272         if (ftrace_start_up)
1273                 command |= FTRACE_DISABLE_CALLS;
1274
1275         ftrace_run_update_code(command);
1276 }
1277
1278 static cycle_t          ftrace_update_time;
1279 static unsigned long    ftrace_update_cnt;
1280 unsigned long           ftrace_update_tot_cnt;
1281
1282 static int ftrace_update_code(struct module *mod)
1283 {
1284         struct dyn_ftrace *p;
1285         cycle_t start, stop;
1286
1287         start = ftrace_now(raw_smp_processor_id());
1288         ftrace_update_cnt = 0;
1289
1290         while (ftrace_new_addrs) {
1291
1292                 /* If something went wrong, bail without enabling anything */
1293                 if (unlikely(ftrace_disabled))
1294                         return -1;
1295
1296                 p = ftrace_new_addrs;
1297                 ftrace_new_addrs = p->newlist;
1298                 p->flags = 0L;
1299
1300                 /* convert record (i.e, patch mcount-call with NOP) */
1301                 if (ftrace_code_disable(mod, p)) {
1302                         p->flags |= FTRACE_FL_CONVERTED;
1303                         ftrace_update_cnt++;
1304                 } else
1305                         ftrace_free_rec(p);
1306         }
1307
1308         stop = ftrace_now(raw_smp_processor_id());
1309         ftrace_update_time = stop - start;
1310         ftrace_update_tot_cnt += ftrace_update_cnt;
1311
1312         return 0;
1313 }
1314
1315 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1316 {
1317         struct ftrace_page *pg;
1318         int cnt;
1319         int i;
1320
1321         /* allocate a few pages */
1322         ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1323         if (!ftrace_pages_start)
1324                 return -1;
1325
1326         /*
1327          * Allocate a few more pages.
1328          *
1329          * TODO: have some parser search vmlinux before
1330          *   final linking to find all calls to ftrace.
1331          *   Then we can:
1332          *    a) know how many pages to allocate.
1333          *     and/or
1334          *    b) set up the table then.
1335          *
1336          *  The dynamic code is still necessary for
1337          *  modules.
1338          */
1339
1340         pg = ftrace_pages = ftrace_pages_start;
1341
1342         cnt = num_to_init / ENTRIES_PER_PAGE;
1343         pr_info("ftrace: allocating %ld entries in %d pages\n",
1344                 num_to_init, cnt + 1);
1345
1346         for (i = 0; i < cnt; i++) {
1347                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1348
1349                 /* If we fail, we'll try later anyway */
1350                 if (!pg->next)
1351                         break;
1352
1353                 pg = pg->next;
1354         }
1355
1356         return 0;
1357 }
1358
1359 enum {
1360         FTRACE_ITER_FILTER      = (1 << 0),
1361         FTRACE_ITER_CONT        = (1 << 1),
1362         FTRACE_ITER_NOTRACE     = (1 << 2),
1363         FTRACE_ITER_FAILURES    = (1 << 3),
1364         FTRACE_ITER_PRINTALL    = (1 << 4),
1365         FTRACE_ITER_HASH        = (1 << 5),
1366 };
1367
1368 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1369
1370 struct ftrace_iterator {
1371         struct ftrace_page      *pg;
1372         int                     hidx;
1373         int                     idx;
1374         unsigned                flags;
1375         unsigned char           buffer[FTRACE_BUFF_MAX+1];
1376         unsigned                buffer_idx;
1377         unsigned                filtered;
1378 };
1379
1380 static void *
1381 t_hash_next(struct seq_file *m, void *v, loff_t *pos)
1382 {
1383         struct ftrace_iterator *iter = m->private;
1384         struct hlist_node *hnd = v;
1385         struct hlist_head *hhd;
1386
1387         WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
1388
1389         (*pos)++;
1390
1391  retry:
1392         if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1393                 return NULL;
1394
1395         hhd = &ftrace_func_hash[iter->hidx];
1396
1397         if (hlist_empty(hhd)) {
1398                 iter->hidx++;
1399                 hnd = NULL;
1400                 goto retry;
1401         }
1402
1403         if (!hnd)
1404                 hnd = hhd->first;
1405         else {
1406                 hnd = hnd->next;
1407                 if (!hnd) {
1408                         iter->hidx++;
1409                         goto retry;
1410                 }
1411         }
1412
1413         return hnd;
1414 }
1415
1416 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1417 {
1418         struct ftrace_iterator *iter = m->private;
1419         void *p = NULL;
1420
1421         iter->flags |= FTRACE_ITER_HASH;
1422
1423         return t_hash_next(m, p, pos);
1424 }
1425
1426 static int t_hash_show(struct seq_file *m, void *v)
1427 {
1428         struct ftrace_func_probe *rec;
1429         struct hlist_node *hnd = v;
1430         char str[KSYM_SYMBOL_LEN];
1431
1432         rec = hlist_entry(hnd, struct ftrace_func_probe, node);
1433
1434         if (rec->ops->print)
1435                 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1436
1437         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1438         seq_printf(m, "%s:", str);
1439
1440         kallsyms_lookup((unsigned long)rec->ops->func, NULL, NULL, NULL, str);
1441         seq_printf(m, "%s", str);
1442
1443         if (rec->data)
1444                 seq_printf(m, ":%p", rec->data);
1445         seq_putc(m, '\n');
1446
1447         return 0;
1448 }
1449
1450 static void *
1451 t_next(struct seq_file *m, void *v, loff_t *pos)
1452 {
1453         struct ftrace_iterator *iter = m->private;
1454         struct dyn_ftrace *rec = NULL;
1455
1456         if (iter->flags & FTRACE_ITER_HASH)
1457                 return t_hash_next(m, v, pos);
1458
1459         (*pos)++;
1460
1461         if (iter->flags & FTRACE_ITER_PRINTALL)
1462                 return NULL;
1463
1464  retry:
1465         if (iter->idx >= iter->pg->index) {
1466                 if (iter->pg->next) {
1467                         iter->pg = iter->pg->next;
1468                         iter->idx = 0;
1469                         goto retry;
1470                 } else {
1471                         iter->idx = -1;
1472                 }
1473         } else {
1474                 rec = &iter->pg->records[iter->idx++];
1475                 if ((rec->flags & FTRACE_FL_FREE) ||
1476
1477                     (!(iter->flags & FTRACE_ITER_FAILURES) &&
1478                      (rec->flags & FTRACE_FL_FAILED)) ||
1479
1480                     ((iter->flags & FTRACE_ITER_FAILURES) &&
1481                      !(rec->flags & FTRACE_FL_FAILED)) ||
1482
1483                     ((iter->flags & FTRACE_ITER_FILTER) &&
1484                      !(rec->flags & FTRACE_FL_FILTER)) ||
1485
1486                     ((iter->flags & FTRACE_ITER_NOTRACE) &&
1487                      !(rec->flags & FTRACE_FL_NOTRACE))) {
1488                         rec = NULL;
1489                         goto retry;
1490                 }
1491         }
1492
1493         return rec;
1494 }
1495
1496 static void *t_start(struct seq_file *m, loff_t *pos)
1497 {
1498         struct ftrace_iterator *iter = m->private;
1499         void *p = NULL;
1500
1501         mutex_lock(&ftrace_lock);
1502         /*
1503          * For set_ftrace_filter reading, if we have the filter
1504          * off, we can short cut and just print out that all
1505          * functions are enabled.
1506          */
1507         if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1508                 if (*pos > 0)
1509                         return t_hash_start(m, pos);
1510                 iter->flags |= FTRACE_ITER_PRINTALL;
1511                 (*pos)++;
1512                 return iter;
1513         }
1514
1515         if (iter->flags & FTRACE_ITER_HASH)
1516                 return t_hash_start(m, pos);
1517
1518         if (*pos > 0) {
1519                 if (iter->idx < 0)
1520                         return p;
1521                 (*pos)--;
1522                 iter->idx--;
1523         }
1524
1525         p = t_next(m, p, pos);
1526
1527         if (!p)
1528                 return t_hash_start(m, pos);
1529
1530         return p;
1531 }
1532
1533 static void t_stop(struct seq_file *m, void *p)
1534 {
1535         mutex_unlock(&ftrace_lock);
1536 }
1537
1538 static int t_show(struct seq_file *m, void *v)
1539 {
1540         struct ftrace_iterator *iter = m->private;
1541         struct dyn_ftrace *rec = v;
1542         char str[KSYM_SYMBOL_LEN];
1543
1544         if (iter->flags & FTRACE_ITER_HASH)
1545                 return t_hash_show(m, v);
1546
1547         if (iter->flags & FTRACE_ITER_PRINTALL) {
1548                 seq_printf(m, "#### all functions enabled ####\n");
1549                 return 0;
1550         }
1551
1552         if (!rec)
1553                 return 0;
1554
1555         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1556
1557         seq_printf(m, "%s\n", str);
1558
1559         return 0;
1560 }
1561
1562 static struct seq_operations show_ftrace_seq_ops = {
1563         .start = t_start,
1564         .next = t_next,
1565         .stop = t_stop,
1566         .show = t_show,
1567 };
1568
1569 static int
1570 ftrace_avail_open(struct inode *inode, struct file *file)
1571 {
1572         struct ftrace_iterator *iter;
1573         int ret;
1574
1575         if (unlikely(ftrace_disabled))
1576                 return -ENODEV;
1577
1578         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1579         if (!iter)
1580                 return -ENOMEM;
1581
1582         iter->pg = ftrace_pages_start;
1583
1584         ret = seq_open(file, &show_ftrace_seq_ops);
1585         if (!ret) {
1586                 struct seq_file *m = file->private_data;
1587
1588                 m->private = iter;
1589         } else {
1590                 kfree(iter);
1591         }
1592
1593         return ret;
1594 }
1595
1596 int ftrace_avail_release(struct inode *inode, struct file *file)
1597 {
1598         struct seq_file *m = (struct seq_file *)file->private_data;
1599         struct ftrace_iterator *iter = m->private;
1600
1601         seq_release(inode, file);
1602         kfree(iter);
1603
1604         return 0;
1605 }
1606
1607 static int
1608 ftrace_failures_open(struct inode *inode, struct file *file)
1609 {
1610         int ret;
1611         struct seq_file *m;
1612         struct ftrace_iterator *iter;
1613
1614         ret = ftrace_avail_open(inode, file);
1615         if (!ret) {
1616                 m = (struct seq_file *)file->private_data;
1617                 iter = (struct ftrace_iterator *)m->private;
1618                 iter->flags = FTRACE_ITER_FAILURES;
1619         }
1620
1621         return ret;
1622 }
1623
1624
1625 static void ftrace_filter_reset(int enable)
1626 {
1627         struct ftrace_page *pg;
1628         struct dyn_ftrace *rec;
1629         unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1630
1631         mutex_lock(&ftrace_lock);
1632         if (enable)
1633                 ftrace_filtered = 0;
1634         do_for_each_ftrace_rec(pg, rec) {
1635                 if (rec->flags & FTRACE_FL_FAILED)
1636                         continue;
1637                 rec->flags &= ~type;
1638         } while_for_each_ftrace_rec();
1639         mutex_unlock(&ftrace_lock);
1640 }
1641
1642 static int
1643 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1644 {
1645         struct ftrace_iterator *iter;
1646         int ret = 0;
1647
1648         if (unlikely(ftrace_disabled))
1649                 return -ENODEV;
1650
1651         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1652         if (!iter)
1653                 return -ENOMEM;
1654
1655         mutex_lock(&ftrace_regex_lock);
1656         if ((file->f_mode & FMODE_WRITE) &&
1657             !(file->f_flags & O_APPEND))
1658                 ftrace_filter_reset(enable);
1659
1660         if (file->f_mode & FMODE_READ) {
1661                 iter->pg = ftrace_pages_start;
1662                 iter->flags = enable ? FTRACE_ITER_FILTER :
1663                         FTRACE_ITER_NOTRACE;
1664
1665                 ret = seq_open(file, &show_ftrace_seq_ops);
1666                 if (!ret) {
1667                         struct seq_file *m = file->private_data;
1668                         m->private = iter;
1669                 } else
1670                         kfree(iter);
1671         } else
1672                 file->private_data = iter;
1673         mutex_unlock(&ftrace_regex_lock);
1674
1675         return ret;
1676 }
1677
1678 static int
1679 ftrace_filter_open(struct inode *inode, struct file *file)
1680 {
1681         return ftrace_regex_open(inode, file, 1);
1682 }
1683
1684 static int
1685 ftrace_notrace_open(struct inode *inode, struct file *file)
1686 {
1687         return ftrace_regex_open(inode, file, 0);
1688 }
1689
1690 static loff_t
1691 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1692 {
1693         loff_t ret;
1694
1695         if (file->f_mode & FMODE_READ)
1696                 ret = seq_lseek(file, offset, origin);
1697         else
1698                 file->f_pos = ret = 1;
1699
1700         return ret;
1701 }
1702
1703 enum {
1704         MATCH_FULL,
1705         MATCH_FRONT_ONLY,
1706         MATCH_MIDDLE_ONLY,
1707         MATCH_END_ONLY,
1708 };
1709
1710 /*
1711  * (static function - no need for kernel doc)
1712  *
1713  * Pass in a buffer containing a glob and this function will
1714  * set search to point to the search part of the buffer and
1715  * return the type of search it is (see enum above).
1716  * This does modify buff.
1717  *
1718  * Returns enum type.
1719  *  search returns the pointer to use for comparison.
1720  *  not returns 1 if buff started with a '!'
1721  *     0 otherwise.
1722  */
1723 static int
1724 ftrace_setup_glob(char *buff, int len, char **search, int *not)
1725 {
1726         int type = MATCH_FULL;
1727         int i;
1728
1729         if (buff[0] == '!') {
1730                 *not = 1;
1731                 buff++;
1732                 len--;
1733         } else
1734                 *not = 0;
1735
1736         *search = buff;
1737
1738         for (i = 0; i < len; i++) {
1739                 if (buff[i] == '*') {
1740                         if (!i) {
1741                                 *search = buff + 1;
1742                                 type = MATCH_END_ONLY;
1743                         } else {
1744                                 if (type == MATCH_END_ONLY)
1745                                         type = MATCH_MIDDLE_ONLY;
1746                                 else
1747                                         type = MATCH_FRONT_ONLY;
1748                                 buff[i] = 0;
1749                                 break;
1750                         }
1751                 }
1752         }
1753
1754         return type;
1755 }
1756
1757 static int ftrace_match(char *str, char *regex, int len, int type)
1758 {
1759         int matched = 0;
1760         char *ptr;
1761
1762         switch (type) {
1763         case MATCH_FULL:
1764                 if (strcmp(str, regex) == 0)
1765                         matched = 1;
1766                 break;
1767         case MATCH_FRONT_ONLY:
1768                 if (strncmp(str, regex, len) == 0)
1769                         matched = 1;
1770                 break;
1771         case MATCH_MIDDLE_ONLY:
1772                 if (strstr(str, regex))
1773                         matched = 1;
1774                 break;
1775         case MATCH_END_ONLY:
1776                 ptr = strstr(str, regex);
1777                 if (ptr && (ptr[len] == 0))
1778                         matched = 1;
1779                 break;
1780         }
1781
1782         return matched;
1783 }
1784
1785 static int
1786 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1787 {
1788         char str[KSYM_SYMBOL_LEN];
1789
1790         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1791         return ftrace_match(str, regex, len, type);
1792 }
1793
1794 static void ftrace_match_records(char *buff, int len, int enable)
1795 {
1796         unsigned int search_len;
1797         struct ftrace_page *pg;
1798         struct dyn_ftrace *rec;
1799         unsigned long flag;
1800         char *search;
1801         int type;
1802         int not;
1803
1804         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1805         type = ftrace_setup_glob(buff, len, &search, &not);
1806
1807         search_len = strlen(search);
1808
1809         mutex_lock(&ftrace_lock);
1810         do_for_each_ftrace_rec(pg, rec) {
1811
1812                 if (rec->flags & FTRACE_FL_FAILED)
1813                         continue;
1814
1815                 if (ftrace_match_record(rec, search, search_len, type)) {
1816                         if (not)
1817                                 rec->flags &= ~flag;
1818                         else
1819                                 rec->flags |= flag;
1820                 }
1821                 /*
1822                  * Only enable filtering if we have a function that
1823                  * is filtered on.
1824                  */
1825                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1826                         ftrace_filtered = 1;
1827         } while_for_each_ftrace_rec();
1828         mutex_unlock(&ftrace_lock);
1829 }
1830
1831 static int
1832 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1833                            char *regex, int len, int type)
1834 {
1835         char str[KSYM_SYMBOL_LEN];
1836         char *modname;
1837
1838         kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1839
1840         if (!modname || strcmp(modname, mod))
1841                 return 0;
1842
1843         /* blank search means to match all funcs in the mod */
1844         if (len)
1845                 return ftrace_match(str, regex, len, type);
1846         else
1847                 return 1;
1848 }
1849
1850 static void ftrace_match_module_records(char *buff, char *mod, int enable)
1851 {
1852         unsigned search_len = 0;
1853         struct ftrace_page *pg;
1854         struct dyn_ftrace *rec;
1855         int type = MATCH_FULL;
1856         char *search = buff;
1857         unsigned long flag;
1858         int not = 0;
1859
1860         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1861
1862         /* blank or '*' mean the same */
1863         if (strcmp(buff, "*") == 0)
1864                 buff[0] = 0;
1865
1866         /* handle the case of 'dont filter this module' */
1867         if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1868                 buff[0] = 0;
1869                 not = 1;
1870         }
1871
1872         if (strlen(buff)) {
1873                 type = ftrace_setup_glob(buff, strlen(buff), &search, &not);
1874                 search_len = strlen(search);
1875         }
1876
1877         mutex_lock(&ftrace_lock);
1878         do_for_each_ftrace_rec(pg, rec) {
1879
1880                 if (rec->flags & FTRACE_FL_FAILED)
1881                         continue;
1882
1883                 if (ftrace_match_module_record(rec, mod,
1884                                                search, search_len, type)) {
1885                         if (not)
1886                                 rec->flags &= ~flag;
1887                         else
1888                                 rec->flags |= flag;
1889                 }
1890                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1891                         ftrace_filtered = 1;
1892
1893         } while_for_each_ftrace_rec();
1894         mutex_unlock(&ftrace_lock);
1895 }
1896
1897 /*
1898  * We register the module command as a template to show others how
1899  * to register the a command as well.
1900  */
1901
1902 static int
1903 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1904 {
1905         char *mod;
1906
1907         /*
1908          * cmd == 'mod' because we only registered this func
1909          * for the 'mod' ftrace_func_command.
1910          * But if you register one func with multiple commands,
1911          * you can tell which command was used by the cmd
1912          * parameter.
1913          */
1914
1915         /* we must have a module name */
1916         if (!param)
1917                 return -EINVAL;
1918
1919         mod = strsep(&param, ":");
1920         if (!strlen(mod))
1921                 return -EINVAL;
1922
1923         ftrace_match_module_records(func, mod, enable);
1924         return 0;
1925 }
1926
1927 static struct ftrace_func_command ftrace_mod_cmd = {
1928         .name                   = "mod",
1929         .func                   = ftrace_mod_callback,
1930 };
1931
1932 static int __init ftrace_mod_cmd_init(void)
1933 {
1934         return register_ftrace_command(&ftrace_mod_cmd);
1935 }
1936 device_initcall(ftrace_mod_cmd_init);
1937
1938 static void
1939 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1940 {
1941         struct ftrace_func_probe *entry;
1942         struct hlist_head *hhd;
1943         struct hlist_node *n;
1944         unsigned long key;
1945         int resched;
1946
1947         key = hash_long(ip, FTRACE_HASH_BITS);
1948
1949         hhd = &ftrace_func_hash[key];
1950
1951         if (hlist_empty(hhd))
1952                 return;
1953
1954         /*
1955          * Disable preemption for these calls to prevent a RCU grace
1956          * period. This syncs the hash iteration and freeing of items
1957          * on the hash. rcu_read_lock is too dangerous here.
1958          */
1959         resched = ftrace_preempt_disable();
1960         hlist_for_each_entry_rcu(entry, n, hhd, node) {
1961                 if (entry->ip == ip)
1962                         entry->ops->func(ip, parent_ip, &entry->data);
1963         }
1964         ftrace_preempt_enable(resched);
1965 }
1966
1967 static struct ftrace_ops trace_probe_ops __read_mostly =
1968 {
1969         .func           = function_trace_probe_call,
1970 };
1971
1972 static int ftrace_probe_registered;
1973
1974 static void __enable_ftrace_function_probe(void)
1975 {
1976         int i;
1977
1978         if (ftrace_probe_registered)
1979                 return;
1980
1981         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1982                 struct hlist_head *hhd = &ftrace_func_hash[i];
1983                 if (hhd->first)
1984                         break;
1985         }
1986         /* Nothing registered? */
1987         if (i == FTRACE_FUNC_HASHSIZE)
1988                 return;
1989
1990         __register_ftrace_function(&trace_probe_ops);
1991         ftrace_startup(0);
1992         ftrace_probe_registered = 1;
1993 }
1994
1995 static void __disable_ftrace_function_probe(void)
1996 {
1997         int i;
1998
1999         if (!ftrace_probe_registered)
2000                 return;
2001
2002         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2003                 struct hlist_head *hhd = &ftrace_func_hash[i];
2004                 if (hhd->first)
2005                         return;
2006         }
2007
2008         /* no more funcs left */
2009         __unregister_ftrace_function(&trace_probe_ops);
2010         ftrace_shutdown(0);
2011         ftrace_probe_registered = 0;
2012 }
2013
2014
2015 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2016 {
2017         struct ftrace_func_probe *entry =
2018                 container_of(rhp, struct ftrace_func_probe, rcu);
2019
2020         if (entry->ops->free)
2021                 entry->ops->free(&entry->data);
2022         kfree(entry);
2023 }
2024
2025
2026 int
2027 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2028                               void *data)
2029 {
2030         struct ftrace_func_probe *entry;
2031         struct ftrace_page *pg;
2032         struct dyn_ftrace *rec;
2033         int type, len, not;
2034         unsigned long key;
2035         int count = 0;
2036         char *search;
2037
2038         type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
2039         len = strlen(search);
2040
2041         /* we do not support '!' for function probes */
2042         if (WARN_ON(not))
2043                 return -EINVAL;
2044
2045         mutex_lock(&ftrace_lock);
2046         do_for_each_ftrace_rec(pg, rec) {
2047
2048                 if (rec->flags & FTRACE_FL_FAILED)
2049                         continue;
2050
2051                 if (!ftrace_match_record(rec, search, len, type))
2052                         continue;
2053
2054                 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2055                 if (!entry) {
2056                         /* If we did not process any, then return error */
2057                         if (!count)
2058                                 count = -ENOMEM;
2059                         goto out_unlock;
2060                 }
2061
2062                 count++;
2063
2064                 entry->data = data;
2065
2066                 /*
2067                  * The caller might want to do something special
2068                  * for each function we find. We call the callback
2069                  * to give the caller an opportunity to do so.
2070                  */
2071                 if (ops->callback) {
2072                         if (ops->callback(rec->ip, &entry->data) < 0) {
2073                                 /* caller does not like this func */
2074                                 kfree(entry);
2075                                 continue;
2076                         }
2077                 }
2078
2079                 entry->ops = ops;
2080                 entry->ip = rec->ip;
2081
2082                 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2083                 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2084
2085         } while_for_each_ftrace_rec();
2086         __enable_ftrace_function_probe();
2087
2088  out_unlock:
2089         mutex_unlock(&ftrace_lock);
2090
2091         return count;
2092 }
2093
2094 enum {
2095         PROBE_TEST_FUNC         = 1,
2096         PROBE_TEST_DATA         = 2
2097 };
2098
2099 static void
2100 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2101                                   void *data, int flags)
2102 {
2103         struct ftrace_func_probe *entry;
2104         struct hlist_node *n, *tmp;
2105         char str[KSYM_SYMBOL_LEN];
2106         int type = MATCH_FULL;
2107         int i, len = 0;
2108         char *search;
2109
2110         if (glob && (strcmp(glob, "*") || !strlen(glob)))
2111                 glob = NULL;
2112         else {
2113                 int not;
2114
2115                 type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
2116                 len = strlen(search);
2117
2118                 /* we do not support '!' for function probes */
2119                 if (WARN_ON(not))
2120                         return;
2121         }
2122
2123         mutex_lock(&ftrace_lock);
2124         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2125                 struct hlist_head *hhd = &ftrace_func_hash[i];
2126
2127                 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2128
2129                         /* break up if statements for readability */
2130                         if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2131                                 continue;
2132
2133                         if ((flags & PROBE_TEST_DATA) && entry->data != data)
2134                                 continue;
2135
2136                         /* do this last, since it is the most expensive */
2137                         if (glob) {
2138                                 kallsyms_lookup(entry->ip, NULL, NULL,
2139                                                 NULL, str);
2140                                 if (!ftrace_match(str, glob, len, type))
2141                                         continue;
2142                         }
2143
2144                         hlist_del(&entry->node);
2145                         call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2146                 }
2147         }
2148         __disable_ftrace_function_probe();
2149         mutex_unlock(&ftrace_lock);
2150 }
2151
2152 void
2153 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2154                                 void *data)
2155 {
2156         __unregister_ftrace_function_probe(glob, ops, data,
2157                                           PROBE_TEST_FUNC | PROBE_TEST_DATA);
2158 }
2159
2160 void
2161 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2162 {
2163         __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2164 }
2165
2166 void unregister_ftrace_function_probe_all(char *glob)
2167 {
2168         __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2169 }
2170
2171 static LIST_HEAD(ftrace_commands);
2172 static DEFINE_MUTEX(ftrace_cmd_mutex);
2173
2174 int register_ftrace_command(struct ftrace_func_command *cmd)
2175 {
2176         struct ftrace_func_command *p;
2177         int ret = 0;
2178
2179         mutex_lock(&ftrace_cmd_mutex);
2180         list_for_each_entry(p, &ftrace_commands, list) {
2181                 if (strcmp(cmd->name, p->name) == 0) {
2182                         ret = -EBUSY;
2183                         goto out_unlock;
2184                 }
2185         }
2186         list_add(&cmd->list, &ftrace_commands);
2187  out_unlock:
2188         mutex_unlock(&ftrace_cmd_mutex);
2189
2190         return ret;
2191 }
2192
2193 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2194 {
2195         struct ftrace_func_command *p, *n;
2196         int ret = -ENODEV;
2197
2198         mutex_lock(&ftrace_cmd_mutex);
2199         list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2200                 if (strcmp(cmd->name, p->name) == 0) {
2201                         ret = 0;
2202                         list_del_init(&p->list);
2203                         goto out_unlock;
2204                 }
2205         }
2206  out_unlock:
2207         mutex_unlock(&ftrace_cmd_mutex);
2208
2209         return ret;
2210 }
2211
2212 static int ftrace_process_regex(char *buff, int len, int enable)
2213 {
2214         char *func, *command, *next = buff;
2215         struct ftrace_func_command *p;
2216         int ret = -EINVAL;
2217
2218         func = strsep(&next, ":");
2219
2220         if (!next) {
2221                 ftrace_match_records(func, len, enable);
2222                 return 0;
2223         }
2224
2225         /* command found */
2226
2227         command = strsep(&next, ":");
2228
2229         mutex_lock(&ftrace_cmd_mutex);
2230         list_for_each_entry(p, &ftrace_commands, list) {
2231                 if (strcmp(p->name, command) == 0) {
2232                         ret = p->func(func, command, next, enable);
2233                         goto out_unlock;
2234                 }
2235         }
2236  out_unlock:
2237         mutex_unlock(&ftrace_cmd_mutex);
2238
2239         return ret;
2240 }
2241
2242 static ssize_t
2243 ftrace_regex_write(struct file *file, const char __user *ubuf,
2244                    size_t cnt, loff_t *ppos, int enable)
2245 {
2246         struct ftrace_iterator *iter;
2247         char ch;
2248         size_t read = 0;
2249         ssize_t ret;
2250
2251         if (!cnt || cnt < 0)
2252                 return 0;
2253
2254         mutex_lock(&ftrace_regex_lock);
2255
2256         if (file->f_mode & FMODE_READ) {
2257                 struct seq_file *m = file->private_data;
2258                 iter = m->private;
2259         } else
2260                 iter = file->private_data;
2261
2262         if (!*ppos) {
2263                 iter->flags &= ~FTRACE_ITER_CONT;
2264                 iter->buffer_idx = 0;
2265         }
2266
2267         ret = get_user(ch, ubuf++);
2268         if (ret)
2269                 goto out;
2270         read++;
2271         cnt--;
2272
2273         if (!(iter->flags & ~FTRACE_ITER_CONT)) {
2274                 /* skip white space */
2275                 while (cnt && isspace(ch)) {
2276                         ret = get_user(ch, ubuf++);
2277                         if (ret)
2278                                 goto out;
2279                         read++;
2280                         cnt--;
2281                 }
2282
2283                 if (isspace(ch)) {
2284                         file->f_pos += read;
2285                         ret = read;
2286                         goto out;
2287                 }
2288
2289                 iter->buffer_idx = 0;
2290         }
2291
2292         while (cnt && !isspace(ch)) {
2293                 if (iter->buffer_idx < FTRACE_BUFF_MAX)
2294                         iter->buffer[iter->buffer_idx++] = ch;
2295                 else {
2296                         ret = -EINVAL;
2297                         goto out;
2298                 }
2299                 ret = get_user(ch, ubuf++);
2300                 if (ret)
2301                         goto out;
2302                 read++;
2303                 cnt--;
2304         }
2305
2306         if (isspace(ch)) {
2307                 iter->filtered++;
2308                 iter->buffer[iter->buffer_idx] = 0;
2309                 ret = ftrace_process_regex(iter->buffer,
2310                                            iter->buffer_idx, enable);
2311                 if (ret)
2312                         goto out;
2313                 iter->buffer_idx = 0;
2314         } else
2315                 iter->flags |= FTRACE_ITER_CONT;
2316
2317
2318         file->f_pos += read;
2319
2320         ret = read;
2321  out:
2322         mutex_unlock(&ftrace_regex_lock);
2323
2324         return ret;
2325 }
2326
2327 static ssize_t
2328 ftrace_filter_write(struct file *file, const char __user *ubuf,
2329                     size_t cnt, loff_t *ppos)
2330 {
2331         return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2332 }
2333
2334 static ssize_t
2335 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2336                      size_t cnt, loff_t *ppos)
2337 {
2338         return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2339 }
2340
2341 static void
2342 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2343 {
2344         if (unlikely(ftrace_disabled))
2345                 return;
2346
2347         mutex_lock(&ftrace_regex_lock);
2348         if (reset)
2349                 ftrace_filter_reset(enable);
2350         if (buf)
2351                 ftrace_match_records(buf, len, enable);
2352         mutex_unlock(&ftrace_regex_lock);
2353 }
2354
2355 /**
2356  * ftrace_set_filter - set a function to filter on in ftrace
2357  * @buf - the string that holds the function filter text.
2358  * @len - the length of the string.
2359  * @reset - non zero to reset all filters before applying this filter.
2360  *
2361  * Filters denote which functions should be enabled when tracing is enabled.
2362  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2363  */
2364 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2365 {
2366         ftrace_set_regex(buf, len, reset, 1);
2367 }
2368
2369 /**
2370  * ftrace_set_notrace - set a function to not trace in ftrace
2371  * @buf - the string that holds the function notrace text.
2372  * @len - the length of the string.
2373  * @reset - non zero to reset all filters before applying this filter.
2374  *
2375  * Notrace Filters denote which functions should not be enabled when tracing
2376  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2377  * for tracing.
2378  */
2379 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2380 {
2381         ftrace_set_regex(buf, len, reset, 0);
2382 }
2383
2384 /*
2385  * command line interface to allow users to set filters on boot up.
2386  */
2387 #define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
2388 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2389 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2390
2391 static int __init set_ftrace_notrace(char *str)
2392 {
2393         strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2394         return 1;
2395 }
2396 __setup("ftrace_notrace=", set_ftrace_notrace);
2397
2398 static int __init set_ftrace_filter(char *str)
2399 {
2400         strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2401         return 1;
2402 }
2403 __setup("ftrace_filter=", set_ftrace_filter);
2404
2405 static void __init set_ftrace_early_filter(char *buf, int enable)
2406 {
2407         char *func;
2408
2409         while (buf) {
2410                 func = strsep(&buf, ",");
2411                 ftrace_set_regex(func, strlen(func), 0, enable);
2412         }
2413 }
2414
2415 static void __init set_ftrace_early_filters(void)
2416 {
2417         if (ftrace_filter_buf[0])
2418                 set_ftrace_early_filter(ftrace_filter_buf, 1);
2419         if (ftrace_notrace_buf[0])
2420                 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2421 }
2422
2423 static int
2424 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2425 {
2426         struct seq_file *m = (struct seq_file *)file->private_data;
2427         struct ftrace_iterator *iter;
2428
2429         mutex_lock(&ftrace_regex_lock);
2430         if (file->f_mode & FMODE_READ) {
2431                 iter = m->private;
2432
2433                 seq_release(inode, file);
2434         } else
2435                 iter = file->private_data;
2436
2437         if (iter->buffer_idx) {
2438                 iter->filtered++;
2439                 iter->buffer[iter->buffer_idx] = 0;
2440                 ftrace_match_records(iter->buffer, iter->buffer_idx, enable);
2441         }
2442
2443         mutex_lock(&ftrace_lock);
2444         if (ftrace_start_up && ftrace_enabled)
2445                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2446         mutex_unlock(&ftrace_lock);
2447
2448         kfree(iter);
2449         mutex_unlock(&ftrace_regex_lock);
2450         return 0;
2451 }
2452
2453 static int
2454 ftrace_filter_release(struct inode *inode, struct file *file)
2455 {
2456         return ftrace_regex_release(inode, file, 1);
2457 }
2458
2459 static int
2460 ftrace_notrace_release(struct inode *inode, struct file *file)
2461 {
2462         return ftrace_regex_release(inode, file, 0);
2463 }
2464
2465 static const struct file_operations ftrace_avail_fops = {
2466         .open = ftrace_avail_open,
2467         .read = seq_read,
2468         .llseek = seq_lseek,
2469         .release = ftrace_avail_release,
2470 };
2471
2472 static const struct file_operations ftrace_failures_fops = {
2473         .open = ftrace_failures_open,
2474         .read = seq_read,
2475         .llseek = seq_lseek,
2476         .release = ftrace_avail_release,
2477 };
2478
2479 static const struct file_operations ftrace_filter_fops = {
2480         .open = ftrace_filter_open,
2481         .read = seq_read,
2482         .write = ftrace_filter_write,
2483         .llseek = ftrace_regex_lseek,
2484         .release = ftrace_filter_release,
2485 };
2486
2487 static const struct file_operations ftrace_notrace_fops = {
2488         .open = ftrace_notrace_open,
2489         .read = seq_read,
2490         .write = ftrace_notrace_write,
2491         .llseek = ftrace_regex_lseek,
2492         .release = ftrace_notrace_release,
2493 };
2494
2495 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2496
2497 static DEFINE_MUTEX(graph_lock);
2498
2499 int ftrace_graph_count;
2500 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2501
2502 static void *
2503 __g_next(struct seq_file *m, loff_t *pos)
2504 {
2505         unsigned long *array = m->private;
2506
2507         if (*pos >= ftrace_graph_count)
2508                 return NULL;
2509         return &array[*pos];
2510 }
2511
2512 static void *
2513 g_next(struct seq_file *m, void *v, loff_t *pos)
2514 {
2515         (*pos)++;
2516         return __g_next(m, pos);
2517 }
2518
2519 static void *g_start(struct seq_file *m, loff_t *pos)
2520 {
2521         mutex_lock(&graph_lock);
2522
2523         /* Nothing, tell g_show to print all functions are enabled */
2524         if (!ftrace_graph_count && !*pos)
2525                 return (void *)1;
2526
2527         return __g_next(m, pos);
2528 }
2529
2530 static void g_stop(struct seq_file *m, void *p)
2531 {
2532         mutex_unlock(&graph_lock);
2533 }
2534
2535 static int g_show(struct seq_file *m, void *v)
2536 {
2537         unsigned long *ptr = v;
2538         char str[KSYM_SYMBOL_LEN];
2539
2540         if (!ptr)
2541                 return 0;
2542
2543         if (ptr == (unsigned long *)1) {
2544                 seq_printf(m, "#### all functions enabled ####\n");
2545                 return 0;
2546         }
2547
2548         kallsyms_lookup(*ptr, NULL, NULL, NULL, str);
2549
2550         seq_printf(m, "%s\n", str);
2551
2552         return 0;
2553 }
2554
2555 static struct seq_operations ftrace_graph_seq_ops = {
2556         .start = g_start,
2557         .next = g_next,
2558         .stop = g_stop,
2559         .show = g_show,
2560 };
2561
2562 static int
2563 ftrace_graph_open(struct inode *inode, struct file *file)
2564 {
2565         int ret = 0;
2566
2567         if (unlikely(ftrace_disabled))
2568                 return -ENODEV;
2569
2570         mutex_lock(&graph_lock);
2571         if ((file->f_mode & FMODE_WRITE) &&
2572             !(file->f_flags & O_APPEND)) {
2573                 ftrace_graph_count = 0;
2574                 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2575         }
2576
2577         if (file->f_mode & FMODE_READ) {
2578                 ret = seq_open(file, &ftrace_graph_seq_ops);
2579                 if (!ret) {
2580                         struct seq_file *m = file->private_data;
2581                         m->private = ftrace_graph_funcs;
2582                 }
2583         } else
2584                 file->private_data = ftrace_graph_funcs;
2585         mutex_unlock(&graph_lock);
2586
2587         return ret;
2588 }
2589
2590 static int
2591 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2592 {
2593         struct dyn_ftrace *rec;
2594         struct ftrace_page *pg;
2595         int search_len;
2596         int found = 0;
2597         int type, not;
2598         char *search;
2599         bool exists;
2600         int i;
2601
2602         if (ftrace_disabled)
2603                 return -ENODEV;
2604
2605         /* decode regex */
2606         type = ftrace_setup_glob(buffer, strlen(buffer), &search, &not);
2607         if (not)
2608                 return -EINVAL;
2609
2610         search_len = strlen(search);
2611
2612         mutex_lock(&ftrace_lock);
2613         do_for_each_ftrace_rec(pg, rec) {
2614
2615                 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2616                         break;
2617
2618                 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2619                         continue;
2620
2621                 if (ftrace_match_record(rec, search, search_len, type)) {
2622                         /* ensure it is not already in the array */
2623                         exists = false;
2624                         for (i = 0; i < *idx; i++)
2625                                 if (array[i] == rec->ip) {
2626                                         exists = true;
2627                                         break;
2628                                 }
2629                         if (!exists) {
2630                                 array[(*idx)++] = rec->ip;
2631                                 found = 1;
2632                         }
2633                 }
2634         } while_for_each_ftrace_rec();
2635
2636         mutex_unlock(&ftrace_lock);
2637
2638         return found ? 0 : -EINVAL;
2639 }
2640
2641 static ssize_t
2642 ftrace_graph_write(struct file *file, const char __user *ubuf,
2643                    size_t cnt, loff_t *ppos)
2644 {
2645         unsigned char buffer[FTRACE_BUFF_MAX+1];
2646         unsigned long *array;
2647         size_t read = 0;
2648         ssize_t ret;
2649         int index = 0;
2650         char ch;
2651
2652         if (!cnt || cnt < 0)
2653                 return 0;
2654
2655         mutex_lock(&graph_lock);
2656
2657         if (ftrace_graph_count >= FTRACE_GRAPH_MAX_FUNCS) {
2658                 ret = -EBUSY;
2659                 goto out;
2660         }
2661
2662         if (file->f_mode & FMODE_READ) {
2663                 struct seq_file *m = file->private_data;
2664                 array = m->private;
2665         } else
2666                 array = file->private_data;
2667
2668         ret = get_user(ch, ubuf++);
2669         if (ret)
2670                 goto out;
2671         read++;
2672         cnt--;
2673
2674         /* skip white space */
2675         while (cnt && isspace(ch)) {
2676                 ret = get_user(ch, ubuf++);
2677                 if (ret)
2678                         goto out;
2679                 read++;
2680                 cnt--;
2681         }
2682
2683         if (isspace(ch)) {
2684                 *ppos += read;
2685                 ret = read;
2686                 goto out;
2687         }
2688
2689         while (cnt && !isspace(ch)) {
2690                 if (index < FTRACE_BUFF_MAX)
2691                         buffer[index++] = ch;
2692                 else {
2693                         ret = -EINVAL;
2694                         goto out;
2695                 }
2696                 ret = get_user(ch, ubuf++);
2697                 if (ret)
2698                         goto out;
2699                 read++;
2700                 cnt--;
2701         }
2702         buffer[index] = 0;
2703
2704         /* we allow only one expression at a time */
2705         ret = ftrace_set_func(array, &ftrace_graph_count, buffer);
2706         if (ret)
2707                 goto out;
2708
2709         file->f_pos += read;
2710
2711         ret = read;
2712  out:
2713         mutex_unlock(&graph_lock);
2714
2715         return ret;
2716 }
2717
2718 static const struct file_operations ftrace_graph_fops = {
2719         .open = ftrace_graph_open,
2720         .read = seq_read,
2721         .write = ftrace_graph_write,
2722 };
2723 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2724
2725 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2726 {
2727
2728         trace_create_file("available_filter_functions", 0444,
2729                         d_tracer, NULL, &ftrace_avail_fops);
2730
2731         trace_create_file("failures", 0444,
2732                         d_tracer, NULL, &ftrace_failures_fops);
2733
2734         trace_create_file("set_ftrace_filter", 0644, d_tracer,
2735                         NULL, &ftrace_filter_fops);
2736
2737         trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2738                                     NULL, &ftrace_notrace_fops);
2739
2740 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2741         trace_create_file("set_graph_function", 0444, d_tracer,
2742                                     NULL,
2743                                     &ftrace_graph_fops);
2744 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2745
2746         return 0;
2747 }
2748
2749 static int ftrace_convert_nops(struct module *mod,
2750                                unsigned long *start,
2751                                unsigned long *end)
2752 {
2753         unsigned long *p;
2754         unsigned long addr;
2755         unsigned long flags;
2756
2757         mutex_lock(&ftrace_lock);
2758         p = start;
2759         while (p < end) {
2760                 addr = ftrace_call_adjust(*p++);
2761                 /*
2762                  * Some architecture linkers will pad between
2763                  * the different mcount_loc sections of different
2764                  * object files to satisfy alignments.
2765                  * Skip any NULL pointers.
2766                  */
2767                 if (!addr)
2768                         continue;
2769                 ftrace_record_ip(addr);
2770         }
2771
2772         /* disable interrupts to prevent kstop machine */
2773         local_irq_save(flags);
2774         ftrace_update_code(mod);
2775         local_irq_restore(flags);
2776         mutex_unlock(&ftrace_lock);
2777
2778         return 0;
2779 }
2780
2781 #ifdef CONFIG_MODULES
2782 void ftrace_release(void *start, void *end)
2783 {
2784         struct dyn_ftrace *rec;
2785         struct ftrace_page *pg;
2786         unsigned long s = (unsigned long)start;
2787         unsigned long e = (unsigned long)end;
2788
2789         if (ftrace_disabled || !start || start == end)
2790                 return;
2791
2792         mutex_lock(&ftrace_lock);
2793         do_for_each_ftrace_rec(pg, rec) {
2794                 if ((rec->ip >= s) && (rec->ip < e)) {
2795                         /*
2796                          * rec->ip is changed in ftrace_free_rec()
2797                          * It should not between s and e if record was freed.
2798                          */
2799                         FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2800                         ftrace_free_rec(rec);
2801                 }
2802         } while_for_each_ftrace_rec();
2803         mutex_unlock(&ftrace_lock);
2804 }
2805
2806 static void ftrace_init_module(struct module *mod,
2807                                unsigned long *start, unsigned long *end)
2808 {
2809         if (ftrace_disabled || start == end)
2810                 return;
2811         ftrace_convert_nops(mod, start, end);
2812 }
2813
2814 static int ftrace_module_notify(struct notifier_block *self,
2815                                 unsigned long val, void *data)
2816 {
2817         struct module *mod = data;
2818
2819         switch (val) {
2820         case MODULE_STATE_COMING:
2821                 ftrace_init_module(mod, mod->ftrace_callsites,
2822                                    mod->ftrace_callsites +
2823                                    mod->num_ftrace_callsites);
2824                 break;
2825         case MODULE_STATE_GOING:
2826                 ftrace_release(mod->ftrace_callsites,
2827                                mod->ftrace_callsites +
2828                                mod->num_ftrace_callsites);
2829                 break;
2830         }
2831
2832         return 0;
2833 }
2834 #else
2835 static int ftrace_module_notify(struct notifier_block *self,
2836                                 unsigned long val, void *data)
2837 {
2838         return 0;
2839 }
2840 #endif /* CONFIG_MODULES */
2841
2842 struct notifier_block ftrace_module_nb = {
2843         .notifier_call = ftrace_module_notify,
2844         .priority = 0,
2845 };
2846
2847 extern unsigned long __start_mcount_loc[];
2848 extern unsigned long __stop_mcount_loc[];
2849
2850 void __init ftrace_init(void)
2851 {
2852         unsigned long count, addr, flags;
2853         int ret;
2854
2855         /* Keep the ftrace pointer to the stub */
2856         addr = (unsigned long)ftrace_stub;
2857
2858         local_irq_save(flags);
2859         ftrace_dyn_arch_init(&addr);
2860         local_irq_restore(flags);
2861
2862         /* ftrace_dyn_arch_init places the return code in addr */
2863         if (addr)
2864                 goto failed;
2865
2866         count = __stop_mcount_loc - __start_mcount_loc;
2867
2868         ret = ftrace_dyn_table_alloc(count);
2869         if (ret)
2870                 goto failed;
2871
2872         last_ftrace_enabled = ftrace_enabled = 1;
2873
2874         ret = ftrace_convert_nops(NULL,
2875                                   __start_mcount_loc,
2876                                   __stop_mcount_loc);
2877
2878         ret = register_module_notifier(&ftrace_module_nb);
2879         if (ret)
2880                 pr_warning("Failed to register trace ftrace module notifier\n");
2881
2882         set_ftrace_early_filters();
2883
2884         return;
2885  failed:
2886         ftrace_disabled = 1;
2887 }
2888
2889 #else
2890
2891 static int __init ftrace_nodyn_init(void)
2892 {
2893         ftrace_enabled = 1;
2894         return 0;
2895 }
2896 device_initcall(ftrace_nodyn_init);
2897
2898 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2899 static inline void ftrace_startup_enable(int command) { }
2900 /* Keep as macros so we do not need to define the commands */
2901 # define ftrace_startup(command)        do { } while (0)
2902 # define ftrace_shutdown(command)       do { } while (0)
2903 # define ftrace_startup_sysctl()        do { } while (0)
2904 # define ftrace_shutdown_sysctl()       do { } while (0)
2905 #endif /* CONFIG_DYNAMIC_FTRACE */
2906
2907 static ssize_t
2908 ftrace_pid_read(struct file *file, char __user *ubuf,
2909                        size_t cnt, loff_t *ppos)
2910 {
2911         char buf[64];
2912         int r;
2913
2914         if (ftrace_pid_trace == ftrace_swapper_pid)
2915                 r = sprintf(buf, "swapper tasks\n");
2916         else if (ftrace_pid_trace)
2917                 r = sprintf(buf, "%u\n", pid_vnr(ftrace_pid_trace));
2918         else
2919                 r = sprintf(buf, "no pid\n");
2920
2921         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2922 }
2923
2924 static void clear_ftrace_swapper(void)
2925 {
2926         struct task_struct *p;
2927         int cpu;
2928
2929         get_online_cpus();
2930         for_each_online_cpu(cpu) {
2931                 p = idle_task(cpu);
2932                 clear_tsk_trace_trace(p);
2933         }
2934         put_online_cpus();
2935 }
2936
2937 static void set_ftrace_swapper(void)
2938 {
2939         struct task_struct *p;
2940         int cpu;
2941
2942         get_online_cpus();
2943         for_each_online_cpu(cpu) {
2944                 p = idle_task(cpu);
2945                 set_tsk_trace_trace(p);
2946         }
2947         put_online_cpus();
2948 }
2949
2950 static void clear_ftrace_pid(struct pid *pid)
2951 {
2952         struct task_struct *p;
2953
2954         rcu_read_lock();
2955         do_each_pid_task(pid, PIDTYPE_PID, p) {
2956                 clear_tsk_trace_trace(p);
2957         } while_each_pid_task(pid, PIDTYPE_PID, p);
2958         rcu_read_unlock();
2959
2960         put_pid(pid);
2961 }
2962
2963 static void set_ftrace_pid(struct pid *pid)
2964 {
2965         struct task_struct *p;
2966
2967         rcu_read_lock();
2968         do_each_pid_task(pid, PIDTYPE_PID, p) {
2969                 set_tsk_trace_trace(p);
2970         } while_each_pid_task(pid, PIDTYPE_PID, p);
2971         rcu_read_unlock();
2972 }
2973
2974 static void clear_ftrace_pid_task(struct pid **pid)
2975 {
2976         if (*pid == ftrace_swapper_pid)
2977                 clear_ftrace_swapper();
2978         else
2979                 clear_ftrace_pid(*pid);
2980
2981         *pid = NULL;
2982 }
2983
2984 static void set_ftrace_pid_task(struct pid *pid)
2985 {
2986         if (pid == ftrace_swapper_pid)
2987                 set_ftrace_swapper();
2988         else
2989                 set_ftrace_pid(pid);
2990 }
2991
2992 static ssize_t
2993 ftrace_pid_write(struct file *filp, const char __user *ubuf,
2994                    size_t cnt, loff_t *ppos)
2995 {
2996         struct pid *pid;
2997         char buf[64];
2998         long val;
2999         int ret;
3000
3001         if (cnt >= sizeof(buf))
3002                 return -EINVAL;
3003
3004         if (copy_from_user(&buf, ubuf, cnt))
3005                 return -EFAULT;
3006
3007         buf[cnt] = 0;
3008
3009         ret = strict_strtol(buf, 10, &val);
3010         if (ret < 0)
3011                 return ret;
3012
3013         mutex_lock(&ftrace_lock);
3014         if (val < 0) {
3015                 /* disable pid tracing */
3016                 if (!ftrace_pid_trace)
3017                         goto out;
3018
3019                 clear_ftrace_pid_task(&ftrace_pid_trace);
3020
3021         } else {
3022                 /* swapper task is special */
3023                 if (!val) {
3024                         pid = ftrace_swapper_pid;
3025                         if (pid == ftrace_pid_trace)
3026                                 goto out;
3027                 } else {
3028                         pid = find_get_pid(val);
3029
3030                         if (pid == ftrace_pid_trace) {
3031                                 put_pid(pid);
3032                                 goto out;
3033                         }
3034                 }
3035
3036                 if (ftrace_pid_trace)
3037                         clear_ftrace_pid_task(&ftrace_pid_trace);
3038
3039                 if (!pid)
3040                         goto out;
3041
3042                 ftrace_pid_trace = pid;
3043
3044                 set_ftrace_pid_task(ftrace_pid_trace);
3045         }
3046
3047         /* update the function call */
3048         ftrace_update_pid_func();
3049         ftrace_startup_enable(0);
3050
3051  out:
3052         mutex_unlock(&ftrace_lock);
3053
3054         return cnt;
3055 }
3056
3057 static const struct file_operations ftrace_pid_fops = {
3058         .read = ftrace_pid_read,
3059         .write = ftrace_pid_write,
3060 };
3061
3062 static __init int ftrace_init_debugfs(void)
3063 {
3064         struct dentry *d_tracer;
3065
3066         d_tracer = tracing_init_dentry();
3067         if (!d_tracer)
3068                 return 0;
3069
3070         ftrace_init_dyn_debugfs(d_tracer);
3071
3072         trace_create_file("set_ftrace_pid", 0644, d_tracer,
3073                             NULL, &ftrace_pid_fops);
3074
3075         ftrace_profile_debugfs(d_tracer);
3076
3077         return 0;
3078 }
3079 fs_initcall(ftrace_init_debugfs);
3080
3081 /**
3082  * ftrace_kill - kill ftrace
3083  *
3084  * This function should be used by panic code. It stops ftrace
3085  * but in a not so nice way. If you need to simply kill ftrace
3086  * from a non-atomic section, use ftrace_kill.
3087  */
3088 void ftrace_kill(void)
3089 {
3090         ftrace_disabled = 1;
3091         ftrace_enabled = 0;
3092         clear_ftrace_function();
3093 }
3094
3095 /**
3096  * register_ftrace_function - register a function for profiling
3097  * @ops - ops structure that holds the function for profiling.
3098  *
3099  * Register a function to be called by all functions in the
3100  * kernel.
3101  *
3102  * Note: @ops->func and all the functions it calls must be labeled
3103  *       with "notrace", otherwise it will go into a
3104  *       recursive loop.
3105  */
3106 int register_ftrace_function(struct ftrace_ops *ops)
3107 {
3108         int ret;
3109
3110         if (unlikely(ftrace_disabled))
3111                 return -1;
3112
3113         mutex_lock(&ftrace_lock);
3114
3115         ret = __register_ftrace_function(ops);
3116         ftrace_startup(0);
3117
3118         mutex_unlock(&ftrace_lock);
3119         return ret;
3120 }
3121
3122 /**
3123  * unregister_ftrace_function - unregister a function for profiling.
3124  * @ops - ops structure that holds the function to unregister
3125  *
3126  * Unregister a function that was added to be called by ftrace profiling.
3127  */
3128 int unregister_ftrace_function(struct ftrace_ops *ops)
3129 {
3130         int ret;
3131
3132         mutex_lock(&ftrace_lock);
3133         ret = __unregister_ftrace_function(ops);
3134         ftrace_shutdown(0);
3135         mutex_unlock(&ftrace_lock);
3136
3137         return ret;
3138 }
3139
3140 int
3141 ftrace_enable_sysctl(struct ctl_table *table, int write,
3142                      struct file *file, void __user *buffer, size_t *lenp,
3143                      loff_t *ppos)
3144 {
3145         int ret;
3146
3147         if (unlikely(ftrace_disabled))
3148                 return -ENODEV;
3149
3150         mutex_lock(&ftrace_lock);
3151
3152         ret  = proc_dointvec(table, write, file, buffer, lenp, ppos);
3153
3154         if (ret || !write || (last_ftrace_enabled == ftrace_enabled))
3155                 goto out;
3156
3157         last_ftrace_enabled = ftrace_enabled;
3158
3159         if (ftrace_enabled) {
3160
3161                 ftrace_startup_sysctl();
3162
3163                 /* we are starting ftrace again */
3164                 if (ftrace_list != &ftrace_list_end) {
3165                         if (ftrace_list->next == &ftrace_list_end)
3166                                 ftrace_trace_function = ftrace_list->func;
3167                         else
3168                                 ftrace_trace_function = ftrace_list_func;
3169                 }
3170
3171         } else {
3172                 /* stopping ftrace calls (just send to ftrace_stub) */
3173                 ftrace_trace_function = ftrace_stub;
3174
3175                 ftrace_shutdown_sysctl();
3176         }
3177
3178  out:
3179         mutex_unlock(&ftrace_lock);
3180         return ret;
3181 }
3182
3183 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3184
3185 static int ftrace_graph_active;
3186 static struct notifier_block ftrace_suspend_notifier;
3187
3188 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3189 {
3190         return 0;
3191 }
3192
3193 /* The callbacks that hook a function */
3194 trace_func_graph_ret_t ftrace_graph_return =
3195                         (trace_func_graph_ret_t)ftrace_stub;
3196 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3197
3198 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3199 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3200 {
3201         int i;
3202         int ret = 0;
3203         unsigned long flags;
3204         int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3205         struct task_struct *g, *t;
3206
3207         for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3208                 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3209                                         * sizeof(struct ftrace_ret_stack),
3210                                         GFP_KERNEL);
3211                 if (!ret_stack_list[i]) {
3212                         start = 0;
3213                         end = i;
3214                         ret = -ENOMEM;
3215                         goto free;
3216                 }
3217         }
3218
3219         read_lock_irqsave(&tasklist_lock, flags);
3220         do_each_thread(g, t) {
3221                 if (start == end) {
3222                         ret = -EAGAIN;
3223                         goto unlock;
3224                 }
3225
3226                 if (t->ret_stack == NULL) {
3227                         atomic_set(&t->tracing_graph_pause, 0);
3228                         atomic_set(&t->trace_overrun, 0);
3229                         t->curr_ret_stack = -1;
3230                         /* Make sure the tasks see the -1 first: */
3231                         smp_wmb();
3232                         t->ret_stack = ret_stack_list[start++];
3233                 }
3234         } while_each_thread(g, t);
3235
3236 unlock:
3237         read_unlock_irqrestore(&tasklist_lock, flags);
3238 free:
3239         for (i = start; i < end; i++)
3240                 kfree(ret_stack_list[i]);
3241         return ret;
3242 }
3243
3244 static void
3245 ftrace_graph_probe_sched_switch(struct rq *__rq, struct task_struct *prev,
3246                                 struct task_struct *next)
3247 {
3248         unsigned long long timestamp;
3249         int index;
3250
3251         /*
3252          * Does the user want to count the time a function was asleep.
3253          * If so, do not update the time stamps.
3254          */
3255         if (trace_flags & TRACE_ITER_SLEEP_TIME)
3256                 return;
3257
3258         timestamp = trace_clock_local();
3259
3260         prev->ftrace_timestamp = timestamp;
3261
3262         /* only process tasks that we timestamped */
3263         if (!next->ftrace_timestamp)
3264                 return;
3265
3266         /*
3267          * Update all the counters in next to make up for the
3268          * time next was sleeping.
3269          */
3270         timestamp -= next->ftrace_timestamp;
3271
3272         for (index = next->curr_ret_stack; index >= 0; index--)
3273                 next->ret_stack[index].calltime += timestamp;
3274 }
3275
3276 /* Allocate a return stack for each task */
3277 static int start_graph_tracing(void)
3278 {
3279         struct ftrace_ret_stack **ret_stack_list;
3280         int ret, cpu;
3281
3282         ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3283                                 sizeof(struct ftrace_ret_stack *),
3284                                 GFP_KERNEL);
3285
3286         if (!ret_stack_list)
3287                 return -ENOMEM;
3288
3289         /* The cpu_boot init_task->ret_stack will never be freed */
3290         for_each_online_cpu(cpu) {
3291                 if (!idle_task(cpu)->ret_stack)
3292                         ftrace_graph_init_task(idle_task(cpu));
3293         }
3294
3295         do {
3296                 ret = alloc_retstack_tasklist(ret_stack_list);
3297         } while (ret == -EAGAIN);
3298
3299         if (!ret) {
3300                 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch);
3301                 if (ret)
3302                         pr_info("ftrace_graph: Couldn't activate tracepoint"
3303                                 " probe to kernel_sched_switch\n");
3304         }
3305
3306         kfree(ret_stack_list);
3307         return ret;
3308 }
3309
3310 /*
3311  * Hibernation protection.
3312  * The state of the current task is too much unstable during
3313  * suspend/restore to disk. We want to protect against that.
3314  */
3315 static int
3316 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3317                                                         void *unused)
3318 {
3319         switch (state) {
3320         case PM_HIBERNATION_PREPARE:
3321                 pause_graph_tracing();
3322                 break;
3323
3324         case PM_POST_HIBERNATION:
3325                 unpause_graph_tracing();
3326                 break;
3327         }
3328         return NOTIFY_DONE;
3329 }
3330
3331 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3332                         trace_func_graph_ent_t entryfunc)
3333 {
3334         int ret = 0;
3335
3336         mutex_lock(&ftrace_lock);
3337
3338         /* we currently allow only one tracer registered at a time */
3339         if (ftrace_graph_active) {
3340                 ret = -EBUSY;
3341                 goto out;
3342         }
3343
3344         ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3345         register_pm_notifier(&ftrace_suspend_notifier);
3346
3347         ftrace_graph_active++;
3348         ret = start_graph_tracing();
3349         if (ret) {
3350                 ftrace_graph_active--;
3351                 goto out;
3352         }
3353
3354         ftrace_graph_return = retfunc;
3355         ftrace_graph_entry = entryfunc;
3356
3357         ftrace_startup(FTRACE_START_FUNC_RET);
3358
3359 out:
3360         mutex_unlock(&ftrace_lock);
3361         return ret;
3362 }
3363
3364 void unregister_ftrace_graph(void)
3365 {
3366         mutex_lock(&ftrace_lock);
3367
3368         if (unlikely(!ftrace_graph_active))
3369                 goto out;
3370
3371         ftrace_graph_active--;
3372         unregister_trace_sched_switch(ftrace_graph_probe_sched_switch);
3373         ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3374         ftrace_graph_entry = ftrace_graph_entry_stub;
3375         ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3376         unregister_pm_notifier(&ftrace_suspend_notifier);
3377
3378  out:
3379         mutex_unlock(&ftrace_lock);
3380 }
3381
3382 /* Allocate a return stack for newly created task */
3383 void ftrace_graph_init_task(struct task_struct *t)
3384 {
3385         /* Make sure we do not use the parent ret_stack */
3386         t->ret_stack = NULL;
3387
3388         if (ftrace_graph_active) {
3389                 struct ftrace_ret_stack *ret_stack;
3390
3391                 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3392                                 * sizeof(struct ftrace_ret_stack),
3393                                 GFP_KERNEL);
3394                 if (!ret_stack)
3395                         return;
3396                 t->curr_ret_stack = -1;
3397                 atomic_set(&t->tracing_graph_pause, 0);
3398                 atomic_set(&t->trace_overrun, 0);
3399                 t->ftrace_timestamp = 0;
3400                 /* make curr_ret_stack visable before we add the ret_stack */
3401                 smp_wmb();
3402                 t->ret_stack = ret_stack;
3403         }
3404 }
3405
3406 void ftrace_graph_exit_task(struct task_struct *t)
3407 {
3408         struct ftrace_ret_stack *ret_stack = t->ret_stack;
3409
3410         t->ret_stack = NULL;
3411         /* NULL must become visible to IRQs before we free it: */
3412         barrier();
3413
3414         kfree(ret_stack);
3415 }
3416
3417 void ftrace_graph_stop(void)
3418 {
3419         ftrace_stop();
3420 }
3421 #endif
3422