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