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