Loading...
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 Nadia Yvette Chambers
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/bsearch.h>
26#include <linux/module.h>
27#include <linux/ftrace.h>
28#include <linux/sysctl.h>
29#include <linux/slab.h>
30#include <linux/ctype.h>
31#include <linux/sort.h>
32#include <linux/list.h>
33#include <linux/hash.h>
34#include <linux/rcupdate.h>
35
36#include <trace/events/sched.h>
37
38#include <asm/setup.h>
39
40#include "trace_output.h"
41#include "trace_stat.h"
42
43#define FTRACE_WARN_ON(cond) \
44 ({ \
45 int ___r = cond; \
46 if (WARN_ON(___r)) \
47 ftrace_kill(); \
48 ___r; \
49 })
50
51#define FTRACE_WARN_ON_ONCE(cond) \
52 ({ \
53 int ___r = cond; \
54 if (WARN_ON_ONCE(___r)) \
55 ftrace_kill(); \
56 ___r; \
57 })
58
59/* hash bits for specific function selection */
60#define FTRACE_HASH_BITS 7
61#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
62#define FTRACE_HASH_DEFAULT_BITS 10
63#define FTRACE_HASH_MAX_BITS 12
64
65#define FL_GLOBAL_CONTROL_MASK (FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_CONTROL)
66
67#ifdef CONFIG_DYNAMIC_FTRACE
68#define INIT_REGEX_LOCK(opsname) \
69 .regex_lock = __MUTEX_INITIALIZER(opsname.regex_lock),
70#else
71#define INIT_REGEX_LOCK(opsname)
72#endif
73
74static struct ftrace_ops ftrace_list_end __read_mostly = {
75 .func = ftrace_stub,
76 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
77};
78
79/* ftrace_enabled is a method to turn ftrace on or off */
80int ftrace_enabled __read_mostly;
81static int last_ftrace_enabled;
82
83/* Quick disabling of function tracer. */
84int function_trace_stop __read_mostly;
85
86/* Current function tracing op */
87struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
88/* What to set function_trace_op to */
89static struct ftrace_ops *set_function_trace_op;
90
91/* List for set_ftrace_pid's pids. */
92LIST_HEAD(ftrace_pids);
93struct ftrace_pid {
94 struct list_head list;
95 struct pid *pid;
96};
97
98/*
99 * ftrace_disabled is set when an anomaly is discovered.
100 * ftrace_disabled is much stronger than ftrace_enabled.
101 */
102static int ftrace_disabled __read_mostly;
103
104static DEFINE_MUTEX(ftrace_lock);
105
106static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
107static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end;
108static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
109ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
110ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
111static struct ftrace_ops global_ops;
112static struct ftrace_ops control_ops;
113
114#if ARCH_SUPPORTS_FTRACE_OPS
115static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
116 struct ftrace_ops *op, struct pt_regs *regs);
117#else
118/* See comment below, where ftrace_ops_list_func is defined */
119static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
120#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
121#endif
122
123/*
124 * Traverse the ftrace_global_list, invoking all entries. The reason that we
125 * can use rcu_dereference_raw_notrace() is that elements removed from this list
126 * are simply leaked, so there is no need to interact with a grace-period
127 * mechanism. The rcu_dereference_raw_notrace() calls are needed to handle
128 * concurrent insertions into the ftrace_global_list.
129 *
130 * Silly Alpha and silly pointer-speculation compiler optimizations!
131 */
132#define do_for_each_ftrace_op(op, list) \
133 op = rcu_dereference_raw_notrace(list); \
134 do
135
136/*
137 * Optimized for just a single item in the list (as that is the normal case).
138 */
139#define while_for_each_ftrace_op(op) \
140 while (likely(op = rcu_dereference_raw_notrace((op)->next)) && \
141 unlikely((op) != &ftrace_list_end))
142
143static inline void ftrace_ops_init(struct ftrace_ops *ops)
144{
145#ifdef CONFIG_DYNAMIC_FTRACE
146 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
147 mutex_init(&ops->regex_lock);
148 ops->flags |= FTRACE_OPS_FL_INITIALIZED;
149 }
150#endif
151}
152
153/**
154 * ftrace_nr_registered_ops - return number of ops registered
155 *
156 * Returns the number of ftrace_ops registered and tracing functions
157 */
158int ftrace_nr_registered_ops(void)
159{
160 struct ftrace_ops *ops;
161 int cnt = 0;
162
163 mutex_lock(&ftrace_lock);
164
165 for (ops = ftrace_ops_list;
166 ops != &ftrace_list_end; ops = ops->next)
167 cnt++;
168
169 mutex_unlock(&ftrace_lock);
170
171 return cnt;
172}
173
174static void
175ftrace_global_list_func(unsigned long ip, unsigned long parent_ip,
176 struct ftrace_ops *op, struct pt_regs *regs)
177{
178 int bit;
179
180 bit = trace_test_and_set_recursion(TRACE_GLOBAL_START, TRACE_GLOBAL_MAX);
181 if (bit < 0)
182 return;
183
184 do_for_each_ftrace_op(op, ftrace_global_list) {
185 op->func(ip, parent_ip, op, regs);
186 } while_for_each_ftrace_op(op);
187
188 trace_clear_recursion(bit);
189}
190
191static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
192 struct ftrace_ops *op, struct pt_regs *regs)
193{
194 if (!test_tsk_trace_trace(current))
195 return;
196
197 ftrace_pid_function(ip, parent_ip, op, regs);
198}
199
200static void set_ftrace_pid_function(ftrace_func_t func)
201{
202 /* do not set ftrace_pid_function to itself! */
203 if (func != ftrace_pid_func)
204 ftrace_pid_function = func;
205}
206
207/**
208 * clear_ftrace_function - reset the ftrace function
209 *
210 * This NULLs the ftrace function and in essence stops
211 * tracing. There may be lag
212 */
213void clear_ftrace_function(void)
214{
215 ftrace_trace_function = ftrace_stub;
216 ftrace_pid_function = ftrace_stub;
217}
218
219static void control_ops_disable_all(struct ftrace_ops *ops)
220{
221 int cpu;
222
223 for_each_possible_cpu(cpu)
224 *per_cpu_ptr(ops->disabled, cpu) = 1;
225}
226
227static int control_ops_alloc(struct ftrace_ops *ops)
228{
229 int __percpu *disabled;
230
231 disabled = alloc_percpu(int);
232 if (!disabled)
233 return -ENOMEM;
234
235 ops->disabled = disabled;
236 control_ops_disable_all(ops);
237 return 0;
238}
239
240static void update_global_ops(void)
241{
242 ftrace_func_t func = ftrace_global_list_func;
243 void *private = NULL;
244
245 /* The list has its own recursion protection. */
246 global_ops.flags |= FTRACE_OPS_FL_RECURSION_SAFE;
247
248 /*
249 * If there's only one function registered, then call that
250 * function directly. Otherwise, we need to iterate over the
251 * registered callers.
252 */
253 if (ftrace_global_list == &ftrace_list_end ||
254 ftrace_global_list->next == &ftrace_list_end) {
255 func = ftrace_global_list->func;
256 private = ftrace_global_list->private;
257 /*
258 * As we are calling the function directly.
259 * If it does not have recursion protection,
260 * the function_trace_op needs to be updated
261 * accordingly.
262 */
263 if (!(ftrace_global_list->flags & FTRACE_OPS_FL_RECURSION_SAFE))
264 global_ops.flags &= ~FTRACE_OPS_FL_RECURSION_SAFE;
265 }
266
267 /* If we filter on pids, update to use the pid function */
268 if (!list_empty(&ftrace_pids)) {
269 set_ftrace_pid_function(func);
270 func = ftrace_pid_func;
271 }
272
273 global_ops.func = func;
274 global_ops.private = private;
275}
276
277static void ftrace_sync(struct work_struct *work)
278{
279 /*
280 * This function is just a stub to implement a hard force
281 * of synchronize_sched(). This requires synchronizing
282 * tasks even in userspace and idle.
283 *
284 * Yes, function tracing is rude.
285 */
286}
287
288static void ftrace_sync_ipi(void *data)
289{
290 /* Probably not needed, but do it anyway */
291 smp_rmb();
292}
293
294#ifdef CONFIG_FUNCTION_GRAPH_TRACER
295static void update_function_graph_func(void);
296#else
297static inline void update_function_graph_func(void) { }
298#endif
299
300static void update_ftrace_function(void)
301{
302 ftrace_func_t func;
303
304 update_global_ops();
305
306 /*
307 * If we are at the end of the list and this ops is
308 * recursion safe and not dynamic and the arch supports passing ops,
309 * then have the mcount trampoline call the function directly.
310 */
311 if (ftrace_ops_list == &ftrace_list_end ||
312 (ftrace_ops_list->next == &ftrace_list_end &&
313 !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC) &&
314 (ftrace_ops_list->flags & FTRACE_OPS_FL_RECURSION_SAFE) &&
315 !FTRACE_FORCE_LIST_FUNC)) {
316 /* Set the ftrace_ops that the arch callback uses */
317 if (ftrace_ops_list == &global_ops)
318 set_function_trace_op = ftrace_global_list;
319 else
320 set_function_trace_op = ftrace_ops_list;
321 func = ftrace_ops_list->func;
322 } else {
323 /* Just use the default ftrace_ops */
324 set_function_trace_op = &ftrace_list_end;
325 func = ftrace_ops_list_func;
326 }
327
328 /* If there's no change, then do nothing more here */
329 if (ftrace_trace_function == func)
330 return;
331
332 update_function_graph_func();
333
334 /*
335 * If we are using the list function, it doesn't care
336 * about the function_trace_ops.
337 */
338 if (func == ftrace_ops_list_func) {
339 ftrace_trace_function = func;
340 /*
341 * Don't even bother setting function_trace_ops,
342 * it would be racy to do so anyway.
343 */
344 return;
345 }
346
347#ifndef CONFIG_DYNAMIC_FTRACE
348 /*
349 * For static tracing, we need to be a bit more careful.
350 * The function change takes affect immediately. Thus,
351 * we need to coorditate the setting of the function_trace_ops
352 * with the setting of the ftrace_trace_function.
353 *
354 * Set the function to the list ops, which will call the
355 * function we want, albeit indirectly, but it handles the
356 * ftrace_ops and doesn't depend on function_trace_op.
357 */
358 ftrace_trace_function = ftrace_ops_list_func;
359 /*
360 * Make sure all CPUs see this. Yes this is slow, but static
361 * tracing is slow and nasty to have enabled.
362 */
363 schedule_on_each_cpu(ftrace_sync);
364 /* Now all cpus are using the list ops. */
365 function_trace_op = set_function_trace_op;
366 /* Make sure the function_trace_op is visible on all CPUs */
367 smp_wmb();
368 /* Nasty way to force a rmb on all cpus */
369 smp_call_function(ftrace_sync_ipi, NULL, 1);
370 /* OK, we are all set to update the ftrace_trace_function now! */
371#endif /* !CONFIG_DYNAMIC_FTRACE */
372
373 ftrace_trace_function = func;
374}
375
376static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
377{
378 ops->next = *list;
379 /*
380 * We are entering ops into the list but another
381 * CPU might be walking that list. We need to make sure
382 * the ops->next pointer is valid before another CPU sees
383 * the ops pointer included into the list.
384 */
385 rcu_assign_pointer(*list, ops);
386}
387
388static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
389{
390 struct ftrace_ops **p;
391
392 /*
393 * If we are removing the last function, then simply point
394 * to the ftrace_stub.
395 */
396 if (*list == ops && ops->next == &ftrace_list_end) {
397 *list = &ftrace_list_end;
398 return 0;
399 }
400
401 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
402 if (*p == ops)
403 break;
404
405 if (*p != ops)
406 return -1;
407
408 *p = (*p)->next;
409 return 0;
410}
411
412static void add_ftrace_list_ops(struct ftrace_ops **list,
413 struct ftrace_ops *main_ops,
414 struct ftrace_ops *ops)
415{
416 int first = *list == &ftrace_list_end;
417 add_ftrace_ops(list, ops);
418 if (first)
419 add_ftrace_ops(&ftrace_ops_list, main_ops);
420}
421
422static int remove_ftrace_list_ops(struct ftrace_ops **list,
423 struct ftrace_ops *main_ops,
424 struct ftrace_ops *ops)
425{
426 int ret = remove_ftrace_ops(list, ops);
427 if (!ret && *list == &ftrace_list_end)
428 ret = remove_ftrace_ops(&ftrace_ops_list, main_ops);
429 return ret;
430}
431
432static int __register_ftrace_function(struct ftrace_ops *ops)
433{
434 if (ops->flags & FTRACE_OPS_FL_DELETED)
435 return -EINVAL;
436
437 if (FTRACE_WARN_ON(ops == &global_ops))
438 return -EINVAL;
439
440 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
441 return -EBUSY;
442
443 /* We don't support both control and global flags set. */
444 if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK)
445 return -EINVAL;
446
447#ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
448 /*
449 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
450 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
451 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
452 */
453 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
454 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
455 return -EINVAL;
456
457 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
458 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
459#endif
460
461 if (!core_kernel_data((unsigned long)ops))
462 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
463
464 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
465 add_ftrace_list_ops(&ftrace_global_list, &global_ops, ops);
466 ops->flags |= FTRACE_OPS_FL_ENABLED;
467 } else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
468 if (control_ops_alloc(ops))
469 return -ENOMEM;
470 add_ftrace_list_ops(&ftrace_control_list, &control_ops, ops);
471 } else
472 add_ftrace_ops(&ftrace_ops_list, ops);
473
474 if (ftrace_enabled)
475 update_ftrace_function();
476
477 return 0;
478}
479
480static int __unregister_ftrace_function(struct ftrace_ops *ops)
481{
482 int ret;
483
484 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
485 return -EBUSY;
486
487 if (FTRACE_WARN_ON(ops == &global_ops))
488 return -EINVAL;
489
490 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
491 ret = remove_ftrace_list_ops(&ftrace_global_list,
492 &global_ops, ops);
493 if (!ret)
494 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
495 } else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
496 ret = remove_ftrace_list_ops(&ftrace_control_list,
497 &control_ops, ops);
498 } else
499 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
500
501 if (ret < 0)
502 return ret;
503
504 if (ftrace_enabled)
505 update_ftrace_function();
506
507 return 0;
508}
509
510static void ftrace_update_pid_func(void)
511{
512 /* Only do something if we are tracing something */
513 if (ftrace_trace_function == ftrace_stub)
514 return;
515
516 update_ftrace_function();
517}
518
519#ifdef CONFIG_FUNCTION_PROFILER
520struct ftrace_profile {
521 struct hlist_node node;
522 unsigned long ip;
523 unsigned long counter;
524#ifdef CONFIG_FUNCTION_GRAPH_TRACER
525 unsigned long long time;
526 unsigned long long time_squared;
527#endif
528};
529
530struct ftrace_profile_page {
531 struct ftrace_profile_page *next;
532 unsigned long index;
533 struct ftrace_profile records[];
534};
535
536struct ftrace_profile_stat {
537 atomic_t disabled;
538 struct hlist_head *hash;
539 struct ftrace_profile_page *pages;
540 struct ftrace_profile_page *start;
541 struct tracer_stat stat;
542};
543
544#define PROFILE_RECORDS_SIZE \
545 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
546
547#define PROFILES_PER_PAGE \
548 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
549
550static int ftrace_profile_enabled __read_mostly;
551
552/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
553static DEFINE_MUTEX(ftrace_profile_lock);
554
555static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
556
557#define FTRACE_PROFILE_HASH_BITS 10
558#define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
559
560static void *
561function_stat_next(void *v, int idx)
562{
563 struct ftrace_profile *rec = v;
564 struct ftrace_profile_page *pg;
565
566 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
567
568 again:
569 if (idx != 0)
570 rec++;
571
572 if ((void *)rec >= (void *)&pg->records[pg->index]) {
573 pg = pg->next;
574 if (!pg)
575 return NULL;
576 rec = &pg->records[0];
577 if (!rec->counter)
578 goto again;
579 }
580
581 return rec;
582}
583
584static void *function_stat_start(struct tracer_stat *trace)
585{
586 struct ftrace_profile_stat *stat =
587 container_of(trace, struct ftrace_profile_stat, stat);
588
589 if (!stat || !stat->start)
590 return NULL;
591
592 return function_stat_next(&stat->start->records[0], 0);
593}
594
595#ifdef CONFIG_FUNCTION_GRAPH_TRACER
596/* function graph compares on total time */
597static int function_stat_cmp(void *p1, void *p2)
598{
599 struct ftrace_profile *a = p1;
600 struct ftrace_profile *b = p2;
601
602 if (a->time < b->time)
603 return -1;
604 if (a->time > b->time)
605 return 1;
606 else
607 return 0;
608}
609#else
610/* not function graph compares against hits */
611static int function_stat_cmp(void *p1, void *p2)
612{
613 struct ftrace_profile *a = p1;
614 struct ftrace_profile *b = p2;
615
616 if (a->counter < b->counter)
617 return -1;
618 if (a->counter > b->counter)
619 return 1;
620 else
621 return 0;
622}
623#endif
624
625static int function_stat_headers(struct seq_file *m)
626{
627#ifdef CONFIG_FUNCTION_GRAPH_TRACER
628 seq_printf(m, " Function "
629 "Hit Time Avg s^2\n"
630 " -------- "
631 "--- ---- --- ---\n");
632#else
633 seq_printf(m, " Function Hit\n"
634 " -------- ---\n");
635#endif
636 return 0;
637}
638
639static int function_stat_show(struct seq_file *m, void *v)
640{
641 struct ftrace_profile *rec = v;
642 char str[KSYM_SYMBOL_LEN];
643 int ret = 0;
644#ifdef CONFIG_FUNCTION_GRAPH_TRACER
645 static struct trace_seq s;
646 unsigned long long avg;
647 unsigned long long stddev;
648#endif
649 mutex_lock(&ftrace_profile_lock);
650
651 /* we raced with function_profile_reset() */
652 if (unlikely(rec->counter == 0)) {
653 ret = -EBUSY;
654 goto out;
655 }
656
657 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
658 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
659
660#ifdef CONFIG_FUNCTION_GRAPH_TRACER
661 seq_printf(m, " ");
662 avg = rec->time;
663 do_div(avg, rec->counter);
664
665 /* Sample standard deviation (s^2) */
666 if (rec->counter <= 1)
667 stddev = 0;
668 else {
669 /*
670 * Apply Welford's method:
671 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
672 */
673 stddev = rec->counter * rec->time_squared -
674 rec->time * rec->time;
675
676 /*
677 * Divide only 1000 for ns^2 -> us^2 conversion.
678 * trace_print_graph_duration will divide 1000 again.
679 */
680 do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
681 }
682
683 trace_seq_init(&s);
684 trace_print_graph_duration(rec->time, &s);
685 trace_seq_puts(&s, " ");
686 trace_print_graph_duration(avg, &s);
687 trace_seq_puts(&s, " ");
688 trace_print_graph_duration(stddev, &s);
689 trace_print_seq(m, &s);
690#endif
691 seq_putc(m, '\n');
692out:
693 mutex_unlock(&ftrace_profile_lock);
694
695 return ret;
696}
697
698static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
699{
700 struct ftrace_profile_page *pg;
701
702 pg = stat->pages = stat->start;
703
704 while (pg) {
705 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
706 pg->index = 0;
707 pg = pg->next;
708 }
709
710 memset(stat->hash, 0,
711 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
712}
713
714int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
715{
716 struct ftrace_profile_page *pg;
717 int functions;
718 int pages;
719 int i;
720
721 /* If we already allocated, do nothing */
722 if (stat->pages)
723 return 0;
724
725 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
726 if (!stat->pages)
727 return -ENOMEM;
728
729#ifdef CONFIG_DYNAMIC_FTRACE
730 functions = ftrace_update_tot_cnt;
731#else
732 /*
733 * We do not know the number of functions that exist because
734 * dynamic tracing is what counts them. With past experience
735 * we have around 20K functions. That should be more than enough.
736 * It is highly unlikely we will execute every function in
737 * the kernel.
738 */
739 functions = 20000;
740#endif
741
742 pg = stat->start = stat->pages;
743
744 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
745
746 for (i = 1; i < pages; i++) {
747 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
748 if (!pg->next)
749 goto out_free;
750 pg = pg->next;
751 }
752
753 return 0;
754
755 out_free:
756 pg = stat->start;
757 while (pg) {
758 unsigned long tmp = (unsigned long)pg;
759
760 pg = pg->next;
761 free_page(tmp);
762 }
763
764 stat->pages = NULL;
765 stat->start = NULL;
766
767 return -ENOMEM;
768}
769
770static int ftrace_profile_init_cpu(int cpu)
771{
772 struct ftrace_profile_stat *stat;
773 int size;
774
775 stat = &per_cpu(ftrace_profile_stats, cpu);
776
777 if (stat->hash) {
778 /* If the profile is already created, simply reset it */
779 ftrace_profile_reset(stat);
780 return 0;
781 }
782
783 /*
784 * We are profiling all functions, but usually only a few thousand
785 * functions are hit. We'll make a hash of 1024 items.
786 */
787 size = FTRACE_PROFILE_HASH_SIZE;
788
789 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
790
791 if (!stat->hash)
792 return -ENOMEM;
793
794 /* Preallocate the function profiling pages */
795 if (ftrace_profile_pages_init(stat) < 0) {
796 kfree(stat->hash);
797 stat->hash = NULL;
798 return -ENOMEM;
799 }
800
801 return 0;
802}
803
804static int ftrace_profile_init(void)
805{
806 int cpu;
807 int ret = 0;
808
809 for_each_possible_cpu(cpu) {
810 ret = ftrace_profile_init_cpu(cpu);
811 if (ret)
812 break;
813 }
814
815 return ret;
816}
817
818/* interrupts must be disabled */
819static struct ftrace_profile *
820ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
821{
822 struct ftrace_profile *rec;
823 struct hlist_head *hhd;
824 unsigned long key;
825
826 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
827 hhd = &stat->hash[key];
828
829 if (hlist_empty(hhd))
830 return NULL;
831
832 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
833 if (rec->ip == ip)
834 return rec;
835 }
836
837 return NULL;
838}
839
840static void ftrace_add_profile(struct ftrace_profile_stat *stat,
841 struct ftrace_profile *rec)
842{
843 unsigned long key;
844
845 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
846 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
847}
848
849/*
850 * The memory is already allocated, this simply finds a new record to use.
851 */
852static struct ftrace_profile *
853ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
854{
855 struct ftrace_profile *rec = NULL;
856
857 /* prevent recursion (from NMIs) */
858 if (atomic_inc_return(&stat->disabled) != 1)
859 goto out;
860
861 /*
862 * Try to find the function again since an NMI
863 * could have added it
864 */
865 rec = ftrace_find_profiled_func(stat, ip);
866 if (rec)
867 goto out;
868
869 if (stat->pages->index == PROFILES_PER_PAGE) {
870 if (!stat->pages->next)
871 goto out;
872 stat->pages = stat->pages->next;
873 }
874
875 rec = &stat->pages->records[stat->pages->index++];
876 rec->ip = ip;
877 ftrace_add_profile(stat, rec);
878
879 out:
880 atomic_dec(&stat->disabled);
881
882 return rec;
883}
884
885static void
886function_profile_call(unsigned long ip, unsigned long parent_ip,
887 struct ftrace_ops *ops, struct pt_regs *regs)
888{
889 struct ftrace_profile_stat *stat;
890 struct ftrace_profile *rec;
891 unsigned long flags;
892
893 if (!ftrace_profile_enabled)
894 return;
895
896 local_irq_save(flags);
897
898 stat = &__get_cpu_var(ftrace_profile_stats);
899 if (!stat->hash || !ftrace_profile_enabled)
900 goto out;
901
902 rec = ftrace_find_profiled_func(stat, ip);
903 if (!rec) {
904 rec = ftrace_profile_alloc(stat, ip);
905 if (!rec)
906 goto out;
907 }
908
909 rec->counter++;
910 out:
911 local_irq_restore(flags);
912}
913
914#ifdef CONFIG_FUNCTION_GRAPH_TRACER
915static int profile_graph_entry(struct ftrace_graph_ent *trace)
916{
917 function_profile_call(trace->func, 0, NULL, NULL);
918 return 1;
919}
920
921static void profile_graph_return(struct ftrace_graph_ret *trace)
922{
923 struct ftrace_profile_stat *stat;
924 unsigned long long calltime;
925 struct ftrace_profile *rec;
926 unsigned long flags;
927
928 local_irq_save(flags);
929 stat = &__get_cpu_var(ftrace_profile_stats);
930 if (!stat->hash || !ftrace_profile_enabled)
931 goto out;
932
933 /* If the calltime was zero'd ignore it */
934 if (!trace->calltime)
935 goto out;
936
937 calltime = trace->rettime - trace->calltime;
938
939 if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
940 int index;
941
942 index = trace->depth;
943
944 /* Append this call time to the parent time to subtract */
945 if (index)
946 current->ret_stack[index - 1].subtime += calltime;
947
948 if (current->ret_stack[index].subtime < calltime)
949 calltime -= current->ret_stack[index].subtime;
950 else
951 calltime = 0;
952 }
953
954 rec = ftrace_find_profiled_func(stat, trace->func);
955 if (rec) {
956 rec->time += calltime;
957 rec->time_squared += calltime * calltime;
958 }
959
960 out:
961 local_irq_restore(flags);
962}
963
964static int register_ftrace_profiler(void)
965{
966 return register_ftrace_graph(&profile_graph_return,
967 &profile_graph_entry);
968}
969
970static void unregister_ftrace_profiler(void)
971{
972 unregister_ftrace_graph();
973}
974#else
975static struct ftrace_ops ftrace_profile_ops __read_mostly = {
976 .func = function_profile_call,
977 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
978 INIT_REGEX_LOCK(ftrace_profile_ops)
979};
980
981static int register_ftrace_profiler(void)
982{
983 return register_ftrace_function(&ftrace_profile_ops);
984}
985
986static void unregister_ftrace_profiler(void)
987{
988 unregister_ftrace_function(&ftrace_profile_ops);
989}
990#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
991
992static ssize_t
993ftrace_profile_write(struct file *filp, const char __user *ubuf,
994 size_t cnt, loff_t *ppos)
995{
996 unsigned long val;
997 int ret;
998
999 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1000 if (ret)
1001 return ret;
1002
1003 val = !!val;
1004
1005 mutex_lock(&ftrace_profile_lock);
1006 if (ftrace_profile_enabled ^ val) {
1007 if (val) {
1008 ret = ftrace_profile_init();
1009 if (ret < 0) {
1010 cnt = ret;
1011 goto out;
1012 }
1013
1014 ret = register_ftrace_profiler();
1015 if (ret < 0) {
1016 cnt = ret;
1017 goto out;
1018 }
1019 ftrace_profile_enabled = 1;
1020 } else {
1021 ftrace_profile_enabled = 0;
1022 /*
1023 * unregister_ftrace_profiler calls stop_machine
1024 * so this acts like an synchronize_sched.
1025 */
1026 unregister_ftrace_profiler();
1027 }
1028 }
1029 out:
1030 mutex_unlock(&ftrace_profile_lock);
1031
1032 *ppos += cnt;
1033
1034 return cnt;
1035}
1036
1037static ssize_t
1038ftrace_profile_read(struct file *filp, char __user *ubuf,
1039 size_t cnt, loff_t *ppos)
1040{
1041 char buf[64]; /* big enough to hold a number */
1042 int r;
1043
1044 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1045 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1046}
1047
1048static const struct file_operations ftrace_profile_fops = {
1049 .open = tracing_open_generic,
1050 .read = ftrace_profile_read,
1051 .write = ftrace_profile_write,
1052 .llseek = default_llseek,
1053};
1054
1055/* used to initialize the real stat files */
1056static struct tracer_stat function_stats __initdata = {
1057 .name = "functions",
1058 .stat_start = function_stat_start,
1059 .stat_next = function_stat_next,
1060 .stat_cmp = function_stat_cmp,
1061 .stat_headers = function_stat_headers,
1062 .stat_show = function_stat_show
1063};
1064
1065static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1066{
1067 struct ftrace_profile_stat *stat;
1068 struct dentry *entry;
1069 char *name;
1070 int ret;
1071 int cpu;
1072
1073 for_each_possible_cpu(cpu) {
1074 stat = &per_cpu(ftrace_profile_stats, cpu);
1075
1076 /* allocate enough for function name + cpu number */
1077 name = kmalloc(32, GFP_KERNEL);
1078 if (!name) {
1079 /*
1080 * The files created are permanent, if something happens
1081 * we still do not free memory.
1082 */
1083 WARN(1,
1084 "Could not allocate stat file for cpu %d\n",
1085 cpu);
1086 return;
1087 }
1088 stat->stat = function_stats;
1089 snprintf(name, 32, "function%d", cpu);
1090 stat->stat.name = name;
1091 ret = register_stat_tracer(&stat->stat);
1092 if (ret) {
1093 WARN(1,
1094 "Could not register function stat for cpu %d\n",
1095 cpu);
1096 kfree(name);
1097 return;
1098 }
1099 }
1100
1101 entry = debugfs_create_file("function_profile_enabled", 0644,
1102 d_tracer, NULL, &ftrace_profile_fops);
1103 if (!entry)
1104 pr_warning("Could not create debugfs "
1105 "'function_profile_enabled' entry\n");
1106}
1107
1108#else /* CONFIG_FUNCTION_PROFILER */
1109static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1110{
1111}
1112#endif /* CONFIG_FUNCTION_PROFILER */
1113
1114static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1115
1116#ifdef CONFIG_DYNAMIC_FTRACE
1117
1118#ifndef CONFIG_FTRACE_MCOUNT_RECORD
1119# error Dynamic ftrace depends on MCOUNT_RECORD
1120#endif
1121
1122static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1123
1124struct ftrace_func_probe {
1125 struct hlist_node node;
1126 struct ftrace_probe_ops *ops;
1127 unsigned long flags;
1128 unsigned long ip;
1129 void *data;
1130 struct list_head free_list;
1131};
1132
1133struct ftrace_func_entry {
1134 struct hlist_node hlist;
1135 unsigned long ip;
1136};
1137
1138struct ftrace_hash {
1139 unsigned long size_bits;
1140 struct hlist_head *buckets;
1141 unsigned long count;
1142 struct rcu_head rcu;
1143};
1144
1145/*
1146 * We make these constant because no one should touch them,
1147 * but they are used as the default "empty hash", to avoid allocating
1148 * it all the time. These are in a read only section such that if
1149 * anyone does try to modify it, it will cause an exception.
1150 */
1151static const struct hlist_head empty_buckets[1];
1152static const struct ftrace_hash empty_hash = {
1153 .buckets = (struct hlist_head *)empty_buckets,
1154};
1155#define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1156
1157static struct ftrace_ops global_ops = {
1158 .func = ftrace_stub,
1159 .notrace_hash = EMPTY_HASH,
1160 .filter_hash = EMPTY_HASH,
1161 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
1162 INIT_REGEX_LOCK(global_ops)
1163};
1164
1165struct ftrace_page {
1166 struct ftrace_page *next;
1167 struct dyn_ftrace *records;
1168 int index;
1169 int size;
1170};
1171
1172#define ENTRY_SIZE sizeof(struct dyn_ftrace)
1173#define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1174
1175/* estimate from running different kernels */
1176#define NR_TO_INIT 10000
1177
1178static struct ftrace_page *ftrace_pages_start;
1179static struct ftrace_page *ftrace_pages;
1180
1181static bool ftrace_hash_empty(struct ftrace_hash *hash)
1182{
1183 return !hash || !hash->count;
1184}
1185
1186static struct ftrace_func_entry *
1187ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1188{
1189 unsigned long key;
1190 struct ftrace_func_entry *entry;
1191 struct hlist_head *hhd;
1192
1193 if (ftrace_hash_empty(hash))
1194 return NULL;
1195
1196 if (hash->size_bits > 0)
1197 key = hash_long(ip, hash->size_bits);
1198 else
1199 key = 0;
1200
1201 hhd = &hash->buckets[key];
1202
1203 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1204 if (entry->ip == ip)
1205 return entry;
1206 }
1207 return NULL;
1208}
1209
1210static void __add_hash_entry(struct ftrace_hash *hash,
1211 struct ftrace_func_entry *entry)
1212{
1213 struct hlist_head *hhd;
1214 unsigned long key;
1215
1216 if (hash->size_bits)
1217 key = hash_long(entry->ip, hash->size_bits);
1218 else
1219 key = 0;
1220
1221 hhd = &hash->buckets[key];
1222 hlist_add_head(&entry->hlist, hhd);
1223 hash->count++;
1224}
1225
1226static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1227{
1228 struct ftrace_func_entry *entry;
1229
1230 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1231 if (!entry)
1232 return -ENOMEM;
1233
1234 entry->ip = ip;
1235 __add_hash_entry(hash, entry);
1236
1237 return 0;
1238}
1239
1240static void
1241free_hash_entry(struct ftrace_hash *hash,
1242 struct ftrace_func_entry *entry)
1243{
1244 hlist_del(&entry->hlist);
1245 kfree(entry);
1246 hash->count--;
1247}
1248
1249static void
1250remove_hash_entry(struct ftrace_hash *hash,
1251 struct ftrace_func_entry *entry)
1252{
1253 hlist_del(&entry->hlist);
1254 hash->count--;
1255}
1256
1257static void ftrace_hash_clear(struct ftrace_hash *hash)
1258{
1259 struct hlist_head *hhd;
1260 struct hlist_node *tn;
1261 struct ftrace_func_entry *entry;
1262 int size = 1 << hash->size_bits;
1263 int i;
1264
1265 if (!hash->count)
1266 return;
1267
1268 for (i = 0; i < size; i++) {
1269 hhd = &hash->buckets[i];
1270 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1271 free_hash_entry(hash, entry);
1272 }
1273 FTRACE_WARN_ON(hash->count);
1274}
1275
1276static void free_ftrace_hash(struct ftrace_hash *hash)
1277{
1278 if (!hash || hash == EMPTY_HASH)
1279 return;
1280 ftrace_hash_clear(hash);
1281 kfree(hash->buckets);
1282 kfree(hash);
1283}
1284
1285static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1286{
1287 struct ftrace_hash *hash;
1288
1289 hash = container_of(rcu, struct ftrace_hash, rcu);
1290 free_ftrace_hash(hash);
1291}
1292
1293static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1294{
1295 if (!hash || hash == EMPTY_HASH)
1296 return;
1297 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1298}
1299
1300void ftrace_free_filter(struct ftrace_ops *ops)
1301{
1302 ftrace_ops_init(ops);
1303 free_ftrace_hash(ops->filter_hash);
1304 free_ftrace_hash(ops->notrace_hash);
1305}
1306
1307static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1308{
1309 struct ftrace_hash *hash;
1310 int size;
1311
1312 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1313 if (!hash)
1314 return NULL;
1315
1316 size = 1 << size_bits;
1317 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1318
1319 if (!hash->buckets) {
1320 kfree(hash);
1321 return NULL;
1322 }
1323
1324 hash->size_bits = size_bits;
1325
1326 return hash;
1327}
1328
1329static struct ftrace_hash *
1330alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1331{
1332 struct ftrace_func_entry *entry;
1333 struct ftrace_hash *new_hash;
1334 int size;
1335 int ret;
1336 int i;
1337
1338 new_hash = alloc_ftrace_hash(size_bits);
1339 if (!new_hash)
1340 return NULL;
1341
1342 /* Empty hash? */
1343 if (ftrace_hash_empty(hash))
1344 return new_hash;
1345
1346 size = 1 << hash->size_bits;
1347 for (i = 0; i < size; i++) {
1348 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1349 ret = add_hash_entry(new_hash, entry->ip);
1350 if (ret < 0)
1351 goto free_hash;
1352 }
1353 }
1354
1355 FTRACE_WARN_ON(new_hash->count != hash->count);
1356
1357 return new_hash;
1358
1359 free_hash:
1360 free_ftrace_hash(new_hash);
1361 return NULL;
1362}
1363
1364static void
1365ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1366static void
1367ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1368
1369static int
1370ftrace_hash_move(struct ftrace_ops *ops, int enable,
1371 struct ftrace_hash **dst, struct ftrace_hash *src)
1372{
1373 struct ftrace_func_entry *entry;
1374 struct hlist_node *tn;
1375 struct hlist_head *hhd;
1376 struct ftrace_hash *old_hash;
1377 struct ftrace_hash *new_hash;
1378 int size = src->count;
1379 int bits = 0;
1380 int ret;
1381 int i;
1382
1383 /*
1384 * Remove the current set, update the hash and add
1385 * them back.
1386 */
1387 ftrace_hash_rec_disable(ops, enable);
1388
1389 /*
1390 * If the new source is empty, just free dst and assign it
1391 * the empty_hash.
1392 */
1393 if (!src->count) {
1394 free_ftrace_hash_rcu(*dst);
1395 rcu_assign_pointer(*dst, EMPTY_HASH);
1396 /* still need to update the function records */
1397 ret = 0;
1398 goto out;
1399 }
1400
1401 /*
1402 * Make the hash size about 1/2 the # found
1403 */
1404 for (size /= 2; size; size >>= 1)
1405 bits++;
1406
1407 /* Don't allocate too much */
1408 if (bits > FTRACE_HASH_MAX_BITS)
1409 bits = FTRACE_HASH_MAX_BITS;
1410
1411 ret = -ENOMEM;
1412 new_hash = alloc_ftrace_hash(bits);
1413 if (!new_hash)
1414 goto out;
1415
1416 size = 1 << src->size_bits;
1417 for (i = 0; i < size; i++) {
1418 hhd = &src->buckets[i];
1419 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1420 remove_hash_entry(src, entry);
1421 __add_hash_entry(new_hash, entry);
1422 }
1423 }
1424
1425 old_hash = *dst;
1426 rcu_assign_pointer(*dst, new_hash);
1427 free_ftrace_hash_rcu(old_hash);
1428
1429 ret = 0;
1430 out:
1431 /*
1432 * Enable regardless of ret:
1433 * On success, we enable the new hash.
1434 * On failure, we re-enable the original hash.
1435 */
1436 ftrace_hash_rec_enable(ops, enable);
1437
1438 return ret;
1439}
1440
1441/*
1442 * Test the hashes for this ops to see if we want to call
1443 * the ops->func or not.
1444 *
1445 * It's a match if the ip is in the ops->filter_hash or
1446 * the filter_hash does not exist or is empty,
1447 * AND
1448 * the ip is not in the ops->notrace_hash.
1449 *
1450 * This needs to be called with preemption disabled as
1451 * the hashes are freed with call_rcu_sched().
1452 */
1453static int
1454ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1455{
1456 struct ftrace_hash *filter_hash;
1457 struct ftrace_hash *notrace_hash;
1458 int ret;
1459
1460#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1461 /*
1462 * There's a small race when adding ops that the ftrace handler
1463 * that wants regs, may be called without them. We can not
1464 * allow that handler to be called if regs is NULL.
1465 */
1466 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1467 return 0;
1468#endif
1469
1470 filter_hash = rcu_dereference_raw_notrace(ops->filter_hash);
1471 notrace_hash = rcu_dereference_raw_notrace(ops->notrace_hash);
1472
1473 if ((ftrace_hash_empty(filter_hash) ||
1474 ftrace_lookup_ip(filter_hash, ip)) &&
1475 (ftrace_hash_empty(notrace_hash) ||
1476 !ftrace_lookup_ip(notrace_hash, ip)))
1477 ret = 1;
1478 else
1479 ret = 0;
1480
1481 return ret;
1482}
1483
1484/*
1485 * This is a double for. Do not use 'break' to break out of the loop,
1486 * you must use a goto.
1487 */
1488#define do_for_each_ftrace_rec(pg, rec) \
1489 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1490 int _____i; \
1491 for (_____i = 0; _____i < pg->index; _____i++) { \
1492 rec = &pg->records[_____i];
1493
1494#define while_for_each_ftrace_rec() \
1495 } \
1496 }
1497
1498
1499static int ftrace_cmp_recs(const void *a, const void *b)
1500{
1501 const struct dyn_ftrace *key = a;
1502 const struct dyn_ftrace *rec = b;
1503
1504 if (key->flags < rec->ip)
1505 return -1;
1506 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1507 return 1;
1508 return 0;
1509}
1510
1511static unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1512{
1513 struct ftrace_page *pg;
1514 struct dyn_ftrace *rec;
1515 struct dyn_ftrace key;
1516
1517 key.ip = start;
1518 key.flags = end; /* overload flags, as it is unsigned long */
1519
1520 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1521 if (end < pg->records[0].ip ||
1522 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1523 continue;
1524 rec = bsearch(&key, pg->records, pg->index,
1525 sizeof(struct dyn_ftrace),
1526 ftrace_cmp_recs);
1527 if (rec)
1528 return rec->ip;
1529 }
1530
1531 return 0;
1532}
1533
1534/**
1535 * ftrace_location - return true if the ip giving is a traced location
1536 * @ip: the instruction pointer to check
1537 *
1538 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1539 * That is, the instruction that is either a NOP or call to
1540 * the function tracer. It checks the ftrace internal tables to
1541 * determine if the address belongs or not.
1542 */
1543unsigned long ftrace_location(unsigned long ip)
1544{
1545 return ftrace_location_range(ip, ip);
1546}
1547
1548/**
1549 * ftrace_text_reserved - return true if range contains an ftrace location
1550 * @start: start of range to search
1551 * @end: end of range to search (inclusive). @end points to the last byte to check.
1552 *
1553 * Returns 1 if @start and @end contains a ftrace location.
1554 * That is, the instruction that is either a NOP or call to
1555 * the function tracer. It checks the ftrace internal tables to
1556 * determine if the address belongs or not.
1557 */
1558int ftrace_text_reserved(const void *start, const void *end)
1559{
1560 unsigned long ret;
1561
1562 ret = ftrace_location_range((unsigned long)start,
1563 (unsigned long)end);
1564
1565 return (int)!!ret;
1566}
1567
1568static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1569 int filter_hash,
1570 bool inc)
1571{
1572 struct ftrace_hash *hash;
1573 struct ftrace_hash *other_hash;
1574 struct ftrace_page *pg;
1575 struct dyn_ftrace *rec;
1576 int count = 0;
1577 int all = 0;
1578
1579 /* Only update if the ops has been registered */
1580 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1581 return;
1582
1583 /*
1584 * In the filter_hash case:
1585 * If the count is zero, we update all records.
1586 * Otherwise we just update the items in the hash.
1587 *
1588 * In the notrace_hash case:
1589 * We enable the update in the hash.
1590 * As disabling notrace means enabling the tracing,
1591 * and enabling notrace means disabling, the inc variable
1592 * gets inversed.
1593 */
1594 if (filter_hash) {
1595 hash = ops->filter_hash;
1596 other_hash = ops->notrace_hash;
1597 if (ftrace_hash_empty(hash))
1598 all = 1;
1599 } else {
1600 inc = !inc;
1601 hash = ops->notrace_hash;
1602 other_hash = ops->filter_hash;
1603 /*
1604 * If the notrace hash has no items,
1605 * then there's nothing to do.
1606 */
1607 if (ftrace_hash_empty(hash))
1608 return;
1609 }
1610
1611 do_for_each_ftrace_rec(pg, rec) {
1612 int in_other_hash = 0;
1613 int in_hash = 0;
1614 int match = 0;
1615
1616 if (all) {
1617 /*
1618 * Only the filter_hash affects all records.
1619 * Update if the record is not in the notrace hash.
1620 */
1621 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1622 match = 1;
1623 } else {
1624 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1625 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1626
1627 /*
1628 *
1629 */
1630 if (filter_hash && in_hash && !in_other_hash)
1631 match = 1;
1632 else if (!filter_hash && in_hash &&
1633 (in_other_hash || ftrace_hash_empty(other_hash)))
1634 match = 1;
1635 }
1636 if (!match)
1637 continue;
1638
1639 if (inc) {
1640 rec->flags++;
1641 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1642 return;
1643 /*
1644 * If any ops wants regs saved for this function
1645 * then all ops will get saved regs.
1646 */
1647 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1648 rec->flags |= FTRACE_FL_REGS;
1649 } else {
1650 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1651 return;
1652 rec->flags--;
1653 }
1654 count++;
1655 /* Shortcut, if we handled all records, we are done. */
1656 if (!all && count == hash->count)
1657 return;
1658 } while_for_each_ftrace_rec();
1659}
1660
1661static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1662 int filter_hash)
1663{
1664 __ftrace_hash_rec_update(ops, filter_hash, 0);
1665}
1666
1667static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1668 int filter_hash)
1669{
1670 __ftrace_hash_rec_update(ops, filter_hash, 1);
1671}
1672
1673static void print_ip_ins(const char *fmt, unsigned char *p)
1674{
1675 int i;
1676
1677 printk(KERN_CONT "%s", fmt);
1678
1679 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1680 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1681}
1682
1683/**
1684 * ftrace_bug - report and shutdown function tracer
1685 * @failed: The failed type (EFAULT, EINVAL, EPERM)
1686 * @ip: The address that failed
1687 *
1688 * The arch code that enables or disables the function tracing
1689 * can call ftrace_bug() when it has detected a problem in
1690 * modifying the code. @failed should be one of either:
1691 * EFAULT - if the problem happens on reading the @ip address
1692 * EINVAL - if what is read at @ip is not what was expected
1693 * EPERM - if the problem happens on writting to the @ip address
1694 */
1695void ftrace_bug(int failed, unsigned long ip)
1696{
1697 switch (failed) {
1698 case -EFAULT:
1699 FTRACE_WARN_ON_ONCE(1);
1700 pr_info("ftrace faulted on modifying ");
1701 print_ip_sym(ip);
1702 break;
1703 case -EINVAL:
1704 FTRACE_WARN_ON_ONCE(1);
1705 pr_info("ftrace failed to modify ");
1706 print_ip_sym(ip);
1707 print_ip_ins(" actual: ", (unsigned char *)ip);
1708 printk(KERN_CONT "\n");
1709 break;
1710 case -EPERM:
1711 FTRACE_WARN_ON_ONCE(1);
1712 pr_info("ftrace faulted on writing ");
1713 print_ip_sym(ip);
1714 break;
1715 default:
1716 FTRACE_WARN_ON_ONCE(1);
1717 pr_info("ftrace faulted on unknown error ");
1718 print_ip_sym(ip);
1719 }
1720}
1721
1722static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1723{
1724 unsigned long flag = 0UL;
1725
1726 /*
1727 * If we are updating calls:
1728 *
1729 * If the record has a ref count, then we need to enable it
1730 * because someone is using it.
1731 *
1732 * Otherwise we make sure its disabled.
1733 *
1734 * If we are disabling calls, then disable all records that
1735 * are enabled.
1736 */
1737 if (enable && (rec->flags & ~FTRACE_FL_MASK))
1738 flag = FTRACE_FL_ENABLED;
1739
1740 /*
1741 * If enabling and the REGS flag does not match the REGS_EN, then
1742 * do not ignore this record. Set flags to fail the compare against
1743 * ENABLED.
1744 */
1745 if (flag &&
1746 (!(rec->flags & FTRACE_FL_REGS) != !(rec->flags & FTRACE_FL_REGS_EN)))
1747 flag |= FTRACE_FL_REGS;
1748
1749 /* If the state of this record hasn't changed, then do nothing */
1750 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1751 return FTRACE_UPDATE_IGNORE;
1752
1753 if (flag) {
1754 /* Save off if rec is being enabled (for return value) */
1755 flag ^= rec->flags & FTRACE_FL_ENABLED;
1756
1757 if (update) {
1758 rec->flags |= FTRACE_FL_ENABLED;
1759 if (flag & FTRACE_FL_REGS) {
1760 if (rec->flags & FTRACE_FL_REGS)
1761 rec->flags |= FTRACE_FL_REGS_EN;
1762 else
1763 rec->flags &= ~FTRACE_FL_REGS_EN;
1764 }
1765 }
1766
1767 /*
1768 * If this record is being updated from a nop, then
1769 * return UPDATE_MAKE_CALL.
1770 * Otherwise, if the EN flag is set, then return
1771 * UPDATE_MODIFY_CALL_REGS to tell the caller to convert
1772 * from the non-save regs, to a save regs function.
1773 * Otherwise,
1774 * return UPDATE_MODIFY_CALL to tell the caller to convert
1775 * from the save regs, to a non-save regs function.
1776 */
1777 if (flag & FTRACE_FL_ENABLED)
1778 return FTRACE_UPDATE_MAKE_CALL;
1779 else if (rec->flags & FTRACE_FL_REGS_EN)
1780 return FTRACE_UPDATE_MODIFY_CALL_REGS;
1781 else
1782 return FTRACE_UPDATE_MODIFY_CALL;
1783 }
1784
1785 if (update) {
1786 /* If there's no more users, clear all flags */
1787 if (!(rec->flags & ~FTRACE_FL_MASK))
1788 rec->flags = 0;
1789 else
1790 /* Just disable the record (keep REGS state) */
1791 rec->flags &= ~FTRACE_FL_ENABLED;
1792 }
1793
1794 return FTRACE_UPDATE_MAKE_NOP;
1795}
1796
1797/**
1798 * ftrace_update_record, set a record that now is tracing or not
1799 * @rec: the record to update
1800 * @enable: set to 1 if the record is tracing, zero to force disable
1801 *
1802 * The records that represent all functions that can be traced need
1803 * to be updated when tracing has been enabled.
1804 */
1805int ftrace_update_record(struct dyn_ftrace *rec, int enable)
1806{
1807 return ftrace_check_record(rec, enable, 1);
1808}
1809
1810/**
1811 * ftrace_test_record, check if the record has been enabled or not
1812 * @rec: the record to test
1813 * @enable: set to 1 to check if enabled, 0 if it is disabled
1814 *
1815 * The arch code may need to test if a record is already set to
1816 * tracing to determine how to modify the function code that it
1817 * represents.
1818 */
1819int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1820{
1821 return ftrace_check_record(rec, enable, 0);
1822}
1823
1824static int
1825__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1826{
1827 unsigned long ftrace_old_addr;
1828 unsigned long ftrace_addr;
1829 int ret;
1830
1831 ret = ftrace_update_record(rec, enable);
1832
1833 if (rec->flags & FTRACE_FL_REGS)
1834 ftrace_addr = (unsigned long)FTRACE_REGS_ADDR;
1835 else
1836 ftrace_addr = (unsigned long)FTRACE_ADDR;
1837
1838 switch (ret) {
1839 case FTRACE_UPDATE_IGNORE:
1840 return 0;
1841
1842 case FTRACE_UPDATE_MAKE_CALL:
1843 return ftrace_make_call(rec, ftrace_addr);
1844
1845 case FTRACE_UPDATE_MAKE_NOP:
1846 return ftrace_make_nop(NULL, rec, ftrace_addr);
1847
1848 case FTRACE_UPDATE_MODIFY_CALL_REGS:
1849 case FTRACE_UPDATE_MODIFY_CALL:
1850 if (rec->flags & FTRACE_FL_REGS)
1851 ftrace_old_addr = (unsigned long)FTRACE_ADDR;
1852 else
1853 ftrace_old_addr = (unsigned long)FTRACE_REGS_ADDR;
1854
1855 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
1856 }
1857
1858 return -1; /* unknow ftrace bug */
1859}
1860
1861void __weak ftrace_replace_code(int enable)
1862{
1863 struct dyn_ftrace *rec;
1864 struct ftrace_page *pg;
1865 int failed;
1866
1867 if (unlikely(ftrace_disabled))
1868 return;
1869
1870 do_for_each_ftrace_rec(pg, rec) {
1871 failed = __ftrace_replace_code(rec, enable);
1872 if (failed) {
1873 ftrace_bug(failed, rec->ip);
1874 /* Stop processing */
1875 return;
1876 }
1877 } while_for_each_ftrace_rec();
1878}
1879
1880struct ftrace_rec_iter {
1881 struct ftrace_page *pg;
1882 int index;
1883};
1884
1885/**
1886 * ftrace_rec_iter_start, start up iterating over traced functions
1887 *
1888 * Returns an iterator handle that is used to iterate over all
1889 * the records that represent address locations where functions
1890 * are traced.
1891 *
1892 * May return NULL if no records are available.
1893 */
1894struct ftrace_rec_iter *ftrace_rec_iter_start(void)
1895{
1896 /*
1897 * We only use a single iterator.
1898 * Protected by the ftrace_lock mutex.
1899 */
1900 static struct ftrace_rec_iter ftrace_rec_iter;
1901 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
1902
1903 iter->pg = ftrace_pages_start;
1904 iter->index = 0;
1905
1906 /* Could have empty pages */
1907 while (iter->pg && !iter->pg->index)
1908 iter->pg = iter->pg->next;
1909
1910 if (!iter->pg)
1911 return NULL;
1912
1913 return iter;
1914}
1915
1916/**
1917 * ftrace_rec_iter_next, get the next record to process.
1918 * @iter: The handle to the iterator.
1919 *
1920 * Returns the next iterator after the given iterator @iter.
1921 */
1922struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
1923{
1924 iter->index++;
1925
1926 if (iter->index >= iter->pg->index) {
1927 iter->pg = iter->pg->next;
1928 iter->index = 0;
1929
1930 /* Could have empty pages */
1931 while (iter->pg && !iter->pg->index)
1932 iter->pg = iter->pg->next;
1933 }
1934
1935 if (!iter->pg)
1936 return NULL;
1937
1938 return iter;
1939}
1940
1941/**
1942 * ftrace_rec_iter_record, get the record at the iterator location
1943 * @iter: The current iterator location
1944 *
1945 * Returns the record that the current @iter is at.
1946 */
1947struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
1948{
1949 return &iter->pg->records[iter->index];
1950}
1951
1952static int
1953ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1954{
1955 unsigned long ip;
1956 int ret;
1957
1958 ip = rec->ip;
1959
1960 if (unlikely(ftrace_disabled))
1961 return 0;
1962
1963 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1964 if (ret) {
1965 ftrace_bug(ret, ip);
1966 return 0;
1967 }
1968 return 1;
1969}
1970
1971/*
1972 * archs can override this function if they must do something
1973 * before the modifying code is performed.
1974 */
1975int __weak ftrace_arch_code_modify_prepare(void)
1976{
1977 return 0;
1978}
1979
1980/*
1981 * archs can override this function if they must do something
1982 * after the modifying code is performed.
1983 */
1984int __weak ftrace_arch_code_modify_post_process(void)
1985{
1986 return 0;
1987}
1988
1989void ftrace_modify_all_code(int command)
1990{
1991 int update = command & FTRACE_UPDATE_TRACE_FUNC;
1992 int err = 0;
1993
1994 /*
1995 * If the ftrace_caller calls a ftrace_ops func directly,
1996 * we need to make sure that it only traces functions it
1997 * expects to trace. When doing the switch of functions,
1998 * we need to update to the ftrace_ops_list_func first
1999 * before the transition between old and new calls are set,
2000 * as the ftrace_ops_list_func will check the ops hashes
2001 * to make sure the ops are having the right functions
2002 * traced.
2003 */
2004 if (update) {
2005 err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2006 if (FTRACE_WARN_ON(err))
2007 return;
2008 }
2009
2010 if (command & FTRACE_UPDATE_CALLS)
2011 ftrace_replace_code(1);
2012 else if (command & FTRACE_DISABLE_CALLS)
2013 ftrace_replace_code(0);
2014
2015 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2016 function_trace_op = set_function_trace_op;
2017 smp_wmb();
2018 /* If irqs are disabled, we are in stop machine */
2019 if (!irqs_disabled())
2020 smp_call_function(ftrace_sync_ipi, NULL, 1);
2021 err = ftrace_update_ftrace_func(ftrace_trace_function);
2022 if (FTRACE_WARN_ON(err))
2023 return;
2024 }
2025
2026 if (command & FTRACE_START_FUNC_RET)
2027 err = ftrace_enable_ftrace_graph_caller();
2028 else if (command & FTRACE_STOP_FUNC_RET)
2029 err = ftrace_disable_ftrace_graph_caller();
2030 FTRACE_WARN_ON(err);
2031}
2032
2033static int __ftrace_modify_code(void *data)
2034{
2035 int *command = data;
2036
2037 ftrace_modify_all_code(*command);
2038
2039 return 0;
2040}
2041
2042/**
2043 * ftrace_run_stop_machine, go back to the stop machine method
2044 * @command: The command to tell ftrace what to do
2045 *
2046 * If an arch needs to fall back to the stop machine method, the
2047 * it can call this function.
2048 */
2049void ftrace_run_stop_machine(int command)
2050{
2051 stop_machine(__ftrace_modify_code, &command, NULL);
2052}
2053
2054/**
2055 * arch_ftrace_update_code, modify the code to trace or not trace
2056 * @command: The command that needs to be done
2057 *
2058 * Archs can override this function if it does not need to
2059 * run stop_machine() to modify code.
2060 */
2061void __weak arch_ftrace_update_code(int command)
2062{
2063 ftrace_run_stop_machine(command);
2064}
2065
2066static void ftrace_run_update_code(int command)
2067{
2068 int ret;
2069
2070 ret = ftrace_arch_code_modify_prepare();
2071 FTRACE_WARN_ON(ret);
2072 if (ret)
2073 return;
2074 /*
2075 * Do not call function tracer while we update the code.
2076 * We are in stop machine.
2077 */
2078 function_trace_stop++;
2079
2080 /*
2081 * By default we use stop_machine() to modify the code.
2082 * But archs can do what ever they want as long as it
2083 * is safe. The stop_machine() is the safest, but also
2084 * produces the most overhead.
2085 */
2086 arch_ftrace_update_code(command);
2087
2088 function_trace_stop--;
2089
2090 ret = ftrace_arch_code_modify_post_process();
2091 FTRACE_WARN_ON(ret);
2092}
2093
2094static ftrace_func_t saved_ftrace_func;
2095static int ftrace_start_up;
2096static int global_start_up;
2097
2098static void control_ops_free(struct ftrace_ops *ops)
2099{
2100 free_percpu(ops->disabled);
2101}
2102
2103static void ftrace_startup_enable(int command)
2104{
2105 if (saved_ftrace_func != ftrace_trace_function) {
2106 saved_ftrace_func = ftrace_trace_function;
2107 command |= FTRACE_UPDATE_TRACE_FUNC;
2108 }
2109
2110 if (!command || !ftrace_enabled)
2111 return;
2112
2113 ftrace_run_update_code(command);
2114}
2115
2116static int ftrace_startup(struct ftrace_ops *ops, int command)
2117{
2118 bool hash_enable = true;
2119 int ret;
2120
2121 if (unlikely(ftrace_disabled))
2122 return -ENODEV;
2123
2124 ret = __register_ftrace_function(ops);
2125 if (ret)
2126 return ret;
2127
2128 ftrace_start_up++;
2129 command |= FTRACE_UPDATE_CALLS;
2130
2131 /* ops marked global share the filter hashes */
2132 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2133 ops = &global_ops;
2134 /* Don't update hash if global is already set */
2135 if (global_start_up)
2136 hash_enable = false;
2137 global_start_up++;
2138 }
2139
2140 ops->flags |= FTRACE_OPS_FL_ENABLED;
2141 if (hash_enable)
2142 ftrace_hash_rec_enable(ops, 1);
2143
2144 ftrace_startup_enable(command);
2145
2146 return 0;
2147}
2148
2149static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2150{
2151 bool hash_disable = true;
2152 int ret;
2153
2154 if (unlikely(ftrace_disabled))
2155 return -ENODEV;
2156
2157 ret = __unregister_ftrace_function(ops);
2158 if (ret)
2159 return ret;
2160
2161 ftrace_start_up--;
2162 /*
2163 * Just warn in case of unbalance, no need to kill ftrace, it's not
2164 * critical but the ftrace_call callers may be never nopped again after
2165 * further ftrace uses.
2166 */
2167 WARN_ON_ONCE(ftrace_start_up < 0);
2168
2169 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2170 ops = &global_ops;
2171 global_start_up--;
2172 WARN_ON_ONCE(global_start_up < 0);
2173 /* Don't update hash if global still has users */
2174 if (global_start_up) {
2175 WARN_ON_ONCE(!ftrace_start_up);
2176 hash_disable = false;
2177 }
2178 }
2179
2180 if (hash_disable)
2181 ftrace_hash_rec_disable(ops, 1);
2182
2183 if (ops != &global_ops || !global_start_up)
2184 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2185
2186 command |= FTRACE_UPDATE_CALLS;
2187
2188 if (saved_ftrace_func != ftrace_trace_function) {
2189 saved_ftrace_func = ftrace_trace_function;
2190 command |= FTRACE_UPDATE_TRACE_FUNC;
2191 }
2192
2193 if (!command || !ftrace_enabled) {
2194 /*
2195 * If these are control ops, they still need their
2196 * per_cpu field freed. Since, function tracing is
2197 * not currently active, we can just free them
2198 * without synchronizing all CPUs.
2199 */
2200 if (ops->flags & FTRACE_OPS_FL_CONTROL)
2201 control_ops_free(ops);
2202 return 0;
2203 }
2204
2205 ftrace_run_update_code(command);
2206
2207 /*
2208 * Dynamic ops may be freed, we must make sure that all
2209 * callers are done before leaving this function.
2210 * The same goes for freeing the per_cpu data of the control
2211 * ops.
2212 *
2213 * Again, normal synchronize_sched() is not good enough.
2214 * We need to do a hard force of sched synchronization.
2215 * This is because we use preempt_disable() to do RCU, but
2216 * the function tracers can be called where RCU is not watching
2217 * (like before user_exit()). We can not rely on the RCU
2218 * infrastructure to do the synchronization, thus we must do it
2219 * ourselves.
2220 */
2221 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_CONTROL)) {
2222 schedule_on_each_cpu(ftrace_sync);
2223
2224 if (ops->flags & FTRACE_OPS_FL_CONTROL)
2225 control_ops_free(ops);
2226 }
2227
2228 return 0;
2229}
2230
2231static void ftrace_startup_sysctl(void)
2232{
2233 if (unlikely(ftrace_disabled))
2234 return;
2235
2236 /* Force update next time */
2237 saved_ftrace_func = NULL;
2238 /* ftrace_start_up is true if we want ftrace running */
2239 if (ftrace_start_up)
2240 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
2241}
2242
2243static void ftrace_shutdown_sysctl(void)
2244{
2245 if (unlikely(ftrace_disabled))
2246 return;
2247
2248 /* ftrace_start_up is true if ftrace is running */
2249 if (ftrace_start_up)
2250 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
2251}
2252
2253static cycle_t ftrace_update_time;
2254unsigned long ftrace_update_tot_cnt;
2255
2256static inline int ops_traces_mod(struct ftrace_ops *ops)
2257{
2258 /*
2259 * Filter_hash being empty will default to trace module.
2260 * But notrace hash requires a test of individual module functions.
2261 */
2262 return ftrace_hash_empty(ops->filter_hash) &&
2263 ftrace_hash_empty(ops->notrace_hash);
2264}
2265
2266/*
2267 * Check if the current ops references the record.
2268 *
2269 * If the ops traces all functions, then it was already accounted for.
2270 * If the ops does not trace the current record function, skip it.
2271 * If the ops ignores the function via notrace filter, skip it.
2272 */
2273static inline bool
2274ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2275{
2276 /* If ops isn't enabled, ignore it */
2277 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2278 return 0;
2279
2280 /* If ops traces all mods, we already accounted for it */
2281 if (ops_traces_mod(ops))
2282 return 0;
2283
2284 /* The function must be in the filter */
2285 if (!ftrace_hash_empty(ops->filter_hash) &&
2286 !ftrace_lookup_ip(ops->filter_hash, rec->ip))
2287 return 0;
2288
2289 /* If in notrace hash, we ignore it too */
2290 if (ftrace_lookup_ip(ops->notrace_hash, rec->ip))
2291 return 0;
2292
2293 return 1;
2294}
2295
2296static int referenced_filters(struct dyn_ftrace *rec)
2297{
2298 struct ftrace_ops *ops;
2299 int cnt = 0;
2300
2301 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
2302 if (ops_references_rec(ops, rec))
2303 cnt++;
2304 }
2305
2306 return cnt;
2307}
2308
2309static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2310{
2311 struct ftrace_page *pg;
2312 struct dyn_ftrace *p;
2313 cycle_t start, stop;
2314 unsigned long update_cnt = 0;
2315 unsigned long ref = 0;
2316 bool test = false;
2317 int i;
2318
2319 /*
2320 * When adding a module, we need to check if tracers are
2321 * currently enabled and if they are set to trace all functions.
2322 * If they are, we need to enable the module functions as well
2323 * as update the reference counts for those function records.
2324 */
2325 if (mod) {
2326 struct ftrace_ops *ops;
2327
2328 for (ops = ftrace_ops_list;
2329 ops != &ftrace_list_end; ops = ops->next) {
2330 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
2331 if (ops_traces_mod(ops))
2332 ref++;
2333 else
2334 test = true;
2335 }
2336 }
2337 }
2338
2339 start = ftrace_now(raw_smp_processor_id());
2340
2341 for (pg = new_pgs; pg; pg = pg->next) {
2342
2343 for (i = 0; i < pg->index; i++) {
2344 int cnt = ref;
2345
2346 /* If something went wrong, bail without enabling anything */
2347 if (unlikely(ftrace_disabled))
2348 return -1;
2349
2350 p = &pg->records[i];
2351 if (test)
2352 cnt += referenced_filters(p);
2353 p->flags = cnt;
2354
2355 /*
2356 * Do the initial record conversion from mcount jump
2357 * to the NOP instructions.
2358 */
2359 if (!ftrace_code_disable(mod, p))
2360 break;
2361
2362 update_cnt++;
2363
2364 /*
2365 * If the tracing is enabled, go ahead and enable the record.
2366 *
2367 * The reason not to enable the record immediatelly is the
2368 * inherent check of ftrace_make_nop/ftrace_make_call for
2369 * correct previous instructions. Making first the NOP
2370 * conversion puts the module to the correct state, thus
2371 * passing the ftrace_make_call check.
2372 */
2373 if (ftrace_start_up && cnt) {
2374 int failed = __ftrace_replace_code(p, 1);
2375 if (failed)
2376 ftrace_bug(failed, p->ip);
2377 }
2378 }
2379 }
2380
2381 stop = ftrace_now(raw_smp_processor_id());
2382 ftrace_update_time = stop - start;
2383 ftrace_update_tot_cnt += update_cnt;
2384
2385 return 0;
2386}
2387
2388static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2389{
2390 int order;
2391 int cnt;
2392
2393 if (WARN_ON(!count))
2394 return -EINVAL;
2395
2396 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2397
2398 /*
2399 * We want to fill as much as possible. No more than a page
2400 * may be empty.
2401 */
2402 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2403 order--;
2404
2405 again:
2406 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2407
2408 if (!pg->records) {
2409 /* if we can't allocate this size, try something smaller */
2410 if (!order)
2411 return -ENOMEM;
2412 order >>= 1;
2413 goto again;
2414 }
2415
2416 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2417 pg->size = cnt;
2418
2419 if (cnt > count)
2420 cnt = count;
2421
2422 return cnt;
2423}
2424
2425static struct ftrace_page *
2426ftrace_allocate_pages(unsigned long num_to_init)
2427{
2428 struct ftrace_page *start_pg;
2429 struct ftrace_page *pg;
2430 int order;
2431 int cnt;
2432
2433 if (!num_to_init)
2434 return 0;
2435
2436 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2437 if (!pg)
2438 return NULL;
2439
2440 /*
2441 * Try to allocate as much as possible in one continues
2442 * location that fills in all of the space. We want to
2443 * waste as little space as possible.
2444 */
2445 for (;;) {
2446 cnt = ftrace_allocate_records(pg, num_to_init);
2447 if (cnt < 0)
2448 goto free_pages;
2449
2450 num_to_init -= cnt;
2451 if (!num_to_init)
2452 break;
2453
2454 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2455 if (!pg->next)
2456 goto free_pages;
2457
2458 pg = pg->next;
2459 }
2460
2461 return start_pg;
2462
2463 free_pages:
2464 while (start_pg) {
2465 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2466 free_pages((unsigned long)pg->records, order);
2467 start_pg = pg->next;
2468 kfree(pg);
2469 pg = start_pg;
2470 }
2471 pr_info("ftrace: FAILED to allocate memory for functions\n");
2472 return NULL;
2473}
2474
2475#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
2476
2477struct ftrace_iterator {
2478 loff_t pos;
2479 loff_t func_pos;
2480 struct ftrace_page *pg;
2481 struct dyn_ftrace *func;
2482 struct ftrace_func_probe *probe;
2483 struct trace_parser parser;
2484 struct ftrace_hash *hash;
2485 struct ftrace_ops *ops;
2486 int hidx;
2487 int idx;
2488 unsigned flags;
2489};
2490
2491static void *
2492t_hash_next(struct seq_file *m, loff_t *pos)
2493{
2494 struct ftrace_iterator *iter = m->private;
2495 struct hlist_node *hnd = NULL;
2496 struct hlist_head *hhd;
2497
2498 (*pos)++;
2499 iter->pos = *pos;
2500
2501 if (iter->probe)
2502 hnd = &iter->probe->node;
2503 retry:
2504 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
2505 return NULL;
2506
2507 hhd = &ftrace_func_hash[iter->hidx];
2508
2509 if (hlist_empty(hhd)) {
2510 iter->hidx++;
2511 hnd = NULL;
2512 goto retry;
2513 }
2514
2515 if (!hnd)
2516 hnd = hhd->first;
2517 else {
2518 hnd = hnd->next;
2519 if (!hnd) {
2520 iter->hidx++;
2521 goto retry;
2522 }
2523 }
2524
2525 if (WARN_ON_ONCE(!hnd))
2526 return NULL;
2527
2528 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
2529
2530 return iter;
2531}
2532
2533static void *t_hash_start(struct seq_file *m, loff_t *pos)
2534{
2535 struct ftrace_iterator *iter = m->private;
2536 void *p = NULL;
2537 loff_t l;
2538
2539 if (!(iter->flags & FTRACE_ITER_DO_HASH))
2540 return NULL;
2541
2542 if (iter->func_pos > *pos)
2543 return NULL;
2544
2545 iter->hidx = 0;
2546 for (l = 0; l <= (*pos - iter->func_pos); ) {
2547 p = t_hash_next(m, &l);
2548 if (!p)
2549 break;
2550 }
2551 if (!p)
2552 return NULL;
2553
2554 /* Only set this if we have an item */
2555 iter->flags |= FTRACE_ITER_HASH;
2556
2557 return iter;
2558}
2559
2560static int
2561t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2562{
2563 struct ftrace_func_probe *rec;
2564
2565 rec = iter->probe;
2566 if (WARN_ON_ONCE(!rec))
2567 return -EIO;
2568
2569 if (rec->ops->print)
2570 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2571
2572 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2573
2574 if (rec->data)
2575 seq_printf(m, ":%p", rec->data);
2576 seq_putc(m, '\n');
2577
2578 return 0;
2579}
2580
2581static void *
2582t_next(struct seq_file *m, void *v, loff_t *pos)
2583{
2584 struct ftrace_iterator *iter = m->private;
2585 struct ftrace_ops *ops = iter->ops;
2586 struct dyn_ftrace *rec = NULL;
2587
2588 if (unlikely(ftrace_disabled))
2589 return NULL;
2590
2591 if (iter->flags & FTRACE_ITER_HASH)
2592 return t_hash_next(m, pos);
2593
2594 (*pos)++;
2595 iter->pos = iter->func_pos = *pos;
2596
2597 if (iter->flags & FTRACE_ITER_PRINTALL)
2598 return t_hash_start(m, pos);
2599
2600 retry:
2601 if (iter->idx >= iter->pg->index) {
2602 if (iter->pg->next) {
2603 iter->pg = iter->pg->next;
2604 iter->idx = 0;
2605 goto retry;
2606 }
2607 } else {
2608 rec = &iter->pg->records[iter->idx++];
2609 if (((iter->flags & FTRACE_ITER_FILTER) &&
2610 !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2611
2612 ((iter->flags & FTRACE_ITER_NOTRACE) &&
2613 !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2614
2615 ((iter->flags & FTRACE_ITER_ENABLED) &&
2616 !(rec->flags & FTRACE_FL_ENABLED))) {
2617
2618 rec = NULL;
2619 goto retry;
2620 }
2621 }
2622
2623 if (!rec)
2624 return t_hash_start(m, pos);
2625
2626 iter->func = rec;
2627
2628 return iter;
2629}
2630
2631static void reset_iter_read(struct ftrace_iterator *iter)
2632{
2633 iter->pos = 0;
2634 iter->func_pos = 0;
2635 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
2636}
2637
2638static void *t_start(struct seq_file *m, loff_t *pos)
2639{
2640 struct ftrace_iterator *iter = m->private;
2641 struct ftrace_ops *ops = iter->ops;
2642 void *p = NULL;
2643 loff_t l;
2644
2645 mutex_lock(&ftrace_lock);
2646
2647 if (unlikely(ftrace_disabled))
2648 return NULL;
2649
2650 /*
2651 * If an lseek was done, then reset and start from beginning.
2652 */
2653 if (*pos < iter->pos)
2654 reset_iter_read(iter);
2655
2656 /*
2657 * For set_ftrace_filter reading, if we have the filter
2658 * off, we can short cut and just print out that all
2659 * functions are enabled.
2660 */
2661 if (iter->flags & FTRACE_ITER_FILTER &&
2662 ftrace_hash_empty(ops->filter_hash)) {
2663 if (*pos > 0)
2664 return t_hash_start(m, pos);
2665 iter->flags |= FTRACE_ITER_PRINTALL;
2666 /* reset in case of seek/pread */
2667 iter->flags &= ~FTRACE_ITER_HASH;
2668 return iter;
2669 }
2670
2671 if (iter->flags & FTRACE_ITER_HASH)
2672 return t_hash_start(m, pos);
2673
2674 /*
2675 * Unfortunately, we need to restart at ftrace_pages_start
2676 * every time we let go of the ftrace_mutex. This is because
2677 * those pointers can change without the lock.
2678 */
2679 iter->pg = ftrace_pages_start;
2680 iter->idx = 0;
2681 for (l = 0; l <= *pos; ) {
2682 p = t_next(m, p, &l);
2683 if (!p)
2684 break;
2685 }
2686
2687 if (!p)
2688 return t_hash_start(m, pos);
2689
2690 return iter;
2691}
2692
2693static void t_stop(struct seq_file *m, void *p)
2694{
2695 mutex_unlock(&ftrace_lock);
2696}
2697
2698static int t_show(struct seq_file *m, void *v)
2699{
2700 struct ftrace_iterator *iter = m->private;
2701 struct dyn_ftrace *rec;
2702
2703 if (iter->flags & FTRACE_ITER_HASH)
2704 return t_hash_show(m, iter);
2705
2706 if (iter->flags & FTRACE_ITER_PRINTALL) {
2707 seq_printf(m, "#### all functions enabled ####\n");
2708 return 0;
2709 }
2710
2711 rec = iter->func;
2712
2713 if (!rec)
2714 return 0;
2715
2716 seq_printf(m, "%ps", (void *)rec->ip);
2717 if (iter->flags & FTRACE_ITER_ENABLED)
2718 seq_printf(m, " (%ld)%s",
2719 rec->flags & ~FTRACE_FL_MASK,
2720 rec->flags & FTRACE_FL_REGS ? " R" : "");
2721 seq_printf(m, "\n");
2722
2723 return 0;
2724}
2725
2726static const struct seq_operations show_ftrace_seq_ops = {
2727 .start = t_start,
2728 .next = t_next,
2729 .stop = t_stop,
2730 .show = t_show,
2731};
2732
2733static int
2734ftrace_avail_open(struct inode *inode, struct file *file)
2735{
2736 struct ftrace_iterator *iter;
2737
2738 if (unlikely(ftrace_disabled))
2739 return -ENODEV;
2740
2741 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2742 if (iter) {
2743 iter->pg = ftrace_pages_start;
2744 iter->ops = &global_ops;
2745 }
2746
2747 return iter ? 0 : -ENOMEM;
2748}
2749
2750static int
2751ftrace_enabled_open(struct inode *inode, struct file *file)
2752{
2753 struct ftrace_iterator *iter;
2754
2755 if (unlikely(ftrace_disabled))
2756 return -ENODEV;
2757
2758 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2759 if (iter) {
2760 iter->pg = ftrace_pages_start;
2761 iter->flags = FTRACE_ITER_ENABLED;
2762 iter->ops = &global_ops;
2763 }
2764
2765 return iter ? 0 : -ENOMEM;
2766}
2767
2768static void ftrace_filter_reset(struct ftrace_hash *hash)
2769{
2770 mutex_lock(&ftrace_lock);
2771 ftrace_hash_clear(hash);
2772 mutex_unlock(&ftrace_lock);
2773}
2774
2775/**
2776 * ftrace_regex_open - initialize function tracer filter files
2777 * @ops: The ftrace_ops that hold the hash filters
2778 * @flag: The type of filter to process
2779 * @inode: The inode, usually passed in to your open routine
2780 * @file: The file, usually passed in to your open routine
2781 *
2782 * ftrace_regex_open() initializes the filter files for the
2783 * @ops. Depending on @flag it may process the filter hash or
2784 * the notrace hash of @ops. With this called from the open
2785 * routine, you can use ftrace_filter_write() for the write
2786 * routine if @flag has FTRACE_ITER_FILTER set, or
2787 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
2788 * tracing_lseek() should be used as the lseek routine, and
2789 * release must call ftrace_regex_release().
2790 */
2791int
2792ftrace_regex_open(struct ftrace_ops *ops, int flag,
2793 struct inode *inode, struct file *file)
2794{
2795 struct ftrace_iterator *iter;
2796 struct ftrace_hash *hash;
2797 int ret = 0;
2798
2799 ftrace_ops_init(ops);
2800
2801 if (unlikely(ftrace_disabled))
2802 return -ENODEV;
2803
2804 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2805 if (!iter)
2806 return -ENOMEM;
2807
2808 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2809 kfree(iter);
2810 return -ENOMEM;
2811 }
2812
2813 iter->ops = ops;
2814 iter->flags = flag;
2815
2816 mutex_lock(&ops->regex_lock);
2817
2818 if (flag & FTRACE_ITER_NOTRACE)
2819 hash = ops->notrace_hash;
2820 else
2821 hash = ops->filter_hash;
2822
2823 if (file->f_mode & FMODE_WRITE) {
2824 iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2825 if (!iter->hash) {
2826 trace_parser_put(&iter->parser);
2827 kfree(iter);
2828 ret = -ENOMEM;
2829 goto out_unlock;
2830 }
2831 }
2832
2833 if ((file->f_mode & FMODE_WRITE) &&
2834 (file->f_flags & O_TRUNC))
2835 ftrace_filter_reset(iter->hash);
2836
2837 if (file->f_mode & FMODE_READ) {
2838 iter->pg = ftrace_pages_start;
2839
2840 ret = seq_open(file, &show_ftrace_seq_ops);
2841 if (!ret) {
2842 struct seq_file *m = file->private_data;
2843 m->private = iter;
2844 } else {
2845 /* Failed */
2846 free_ftrace_hash(iter->hash);
2847 trace_parser_put(&iter->parser);
2848 kfree(iter);
2849 }
2850 } else
2851 file->private_data = iter;
2852
2853 out_unlock:
2854 mutex_unlock(&ops->regex_lock);
2855
2856 return ret;
2857}
2858
2859static int
2860ftrace_filter_open(struct inode *inode, struct file *file)
2861{
2862 struct ftrace_ops *ops = inode->i_private;
2863
2864 return ftrace_regex_open(ops,
2865 FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
2866 inode, file);
2867}
2868
2869static int
2870ftrace_notrace_open(struct inode *inode, struct file *file)
2871{
2872 struct ftrace_ops *ops = inode->i_private;
2873
2874 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
2875 inode, file);
2876}
2877
2878static int ftrace_match(char *str, char *regex, int len, int type)
2879{
2880 int matched = 0;
2881 int slen;
2882
2883 switch (type) {
2884 case MATCH_FULL:
2885 if (strcmp(str, regex) == 0)
2886 matched = 1;
2887 break;
2888 case MATCH_FRONT_ONLY:
2889 if (strncmp(str, regex, len) == 0)
2890 matched = 1;
2891 break;
2892 case MATCH_MIDDLE_ONLY:
2893 if (strstr(str, regex))
2894 matched = 1;
2895 break;
2896 case MATCH_END_ONLY:
2897 slen = strlen(str);
2898 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2899 matched = 1;
2900 break;
2901 }
2902
2903 return matched;
2904}
2905
2906static int
2907enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2908{
2909 struct ftrace_func_entry *entry;
2910 int ret = 0;
2911
2912 entry = ftrace_lookup_ip(hash, rec->ip);
2913 if (not) {
2914 /* Do nothing if it doesn't exist */
2915 if (!entry)
2916 return 0;
2917
2918 free_hash_entry(hash, entry);
2919 } else {
2920 /* Do nothing if it exists */
2921 if (entry)
2922 return 0;
2923
2924 ret = add_hash_entry(hash, rec->ip);
2925 }
2926 return ret;
2927}
2928
2929static int
2930ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2931 char *regex, int len, int type)
2932{
2933 char str[KSYM_SYMBOL_LEN];
2934 char *modname;
2935
2936 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2937
2938 if (mod) {
2939 /* module lookup requires matching the module */
2940 if (!modname || strcmp(modname, mod))
2941 return 0;
2942
2943 /* blank search means to match all funcs in the mod */
2944 if (!len)
2945 return 1;
2946 }
2947
2948 return ftrace_match(str, regex, len, type);
2949}
2950
2951static int
2952match_records(struct ftrace_hash *hash, char *buff,
2953 int len, char *mod, int not)
2954{
2955 unsigned search_len = 0;
2956 struct ftrace_page *pg;
2957 struct dyn_ftrace *rec;
2958 int type = MATCH_FULL;
2959 char *search = buff;
2960 int found = 0;
2961 int ret;
2962
2963 if (len) {
2964 type = filter_parse_regex(buff, len, &search, ¬);
2965 search_len = strlen(search);
2966 }
2967
2968 mutex_lock(&ftrace_lock);
2969
2970 if (unlikely(ftrace_disabled))
2971 goto out_unlock;
2972
2973 do_for_each_ftrace_rec(pg, rec) {
2974 if (ftrace_match_record(rec, mod, search, search_len, type)) {
2975 ret = enter_record(hash, rec, not);
2976 if (ret < 0) {
2977 found = ret;
2978 goto out_unlock;
2979 }
2980 found = 1;
2981 }
2982 } while_for_each_ftrace_rec();
2983 out_unlock:
2984 mutex_unlock(&ftrace_lock);
2985
2986 return found;
2987}
2988
2989static int
2990ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2991{
2992 return match_records(hash, buff, len, NULL, 0);
2993}
2994
2995static int
2996ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2997{
2998 int not = 0;
2999
3000 /* blank or '*' mean the same */
3001 if (strcmp(buff, "*") == 0)
3002 buff[0] = 0;
3003
3004 /* handle the case of 'dont filter this module' */
3005 if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
3006 buff[0] = 0;
3007 not = 1;
3008 }
3009
3010 return match_records(hash, buff, strlen(buff), mod, not);
3011}
3012
3013/*
3014 * We register the module command as a template to show others how
3015 * to register the a command as well.
3016 */
3017
3018static int
3019ftrace_mod_callback(struct ftrace_hash *hash,
3020 char *func, char *cmd, char *param, int enable)
3021{
3022 char *mod;
3023 int ret = -EINVAL;
3024
3025 /*
3026 * cmd == 'mod' because we only registered this func
3027 * for the 'mod' ftrace_func_command.
3028 * But if you register one func with multiple commands,
3029 * you can tell which command was used by the cmd
3030 * parameter.
3031 */
3032
3033 /* we must have a module name */
3034 if (!param)
3035 return ret;
3036
3037 mod = strsep(¶m, ":");
3038 if (!strlen(mod))
3039 return ret;
3040
3041 ret = ftrace_match_module_records(hash, func, mod);
3042 if (!ret)
3043 ret = -EINVAL;
3044 if (ret < 0)
3045 return ret;
3046
3047 return 0;
3048}
3049
3050static struct ftrace_func_command ftrace_mod_cmd = {
3051 .name = "mod",
3052 .func = ftrace_mod_callback,
3053};
3054
3055static int __init ftrace_mod_cmd_init(void)
3056{
3057 return register_ftrace_command(&ftrace_mod_cmd);
3058}
3059core_initcall(ftrace_mod_cmd_init);
3060
3061static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
3062 struct ftrace_ops *op, struct pt_regs *pt_regs)
3063{
3064 struct ftrace_func_probe *entry;
3065 struct hlist_head *hhd;
3066 unsigned long key;
3067
3068 key = hash_long(ip, FTRACE_HASH_BITS);
3069
3070 hhd = &ftrace_func_hash[key];
3071
3072 if (hlist_empty(hhd))
3073 return;
3074
3075 /*
3076 * Disable preemption for these calls to prevent a RCU grace
3077 * period. This syncs the hash iteration and freeing of items
3078 * on the hash. rcu_read_lock is too dangerous here.
3079 */
3080 preempt_disable_notrace();
3081 hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3082 if (entry->ip == ip)
3083 entry->ops->func(ip, parent_ip, &entry->data);
3084 }
3085 preempt_enable_notrace();
3086}
3087
3088static struct ftrace_ops trace_probe_ops __read_mostly =
3089{
3090 .func = function_trace_probe_call,
3091 .flags = FTRACE_OPS_FL_INITIALIZED,
3092 INIT_REGEX_LOCK(trace_probe_ops)
3093};
3094
3095static int ftrace_probe_registered;
3096
3097static void __enable_ftrace_function_probe(void)
3098{
3099 int ret;
3100 int i;
3101
3102 if (ftrace_probe_registered) {
3103 /* still need to update the function call sites */
3104 if (ftrace_enabled)
3105 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3106 return;
3107 }
3108
3109 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3110 struct hlist_head *hhd = &ftrace_func_hash[i];
3111 if (hhd->first)
3112 break;
3113 }
3114 /* Nothing registered? */
3115 if (i == FTRACE_FUNC_HASHSIZE)
3116 return;
3117
3118 ret = ftrace_startup(&trace_probe_ops, 0);
3119
3120 ftrace_probe_registered = 1;
3121}
3122
3123static void __disable_ftrace_function_probe(void)
3124{
3125 int i;
3126
3127 if (!ftrace_probe_registered)
3128 return;
3129
3130 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3131 struct hlist_head *hhd = &ftrace_func_hash[i];
3132 if (hhd->first)
3133 return;
3134 }
3135
3136 /* no more funcs left */
3137 ftrace_shutdown(&trace_probe_ops, 0);
3138
3139 ftrace_probe_registered = 0;
3140}
3141
3142
3143static void ftrace_free_entry(struct ftrace_func_probe *entry)
3144{
3145 if (entry->ops->free)
3146 entry->ops->free(entry->ops, entry->ip, &entry->data);
3147 kfree(entry);
3148}
3149
3150int
3151register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3152 void *data)
3153{
3154 struct ftrace_func_probe *entry;
3155 struct ftrace_hash **orig_hash = &trace_probe_ops.filter_hash;
3156 struct ftrace_hash *hash;
3157 struct ftrace_page *pg;
3158 struct dyn_ftrace *rec;
3159 int type, len, not;
3160 unsigned long key;
3161 int count = 0;
3162 char *search;
3163 int ret;
3164
3165 type = filter_parse_regex(glob, strlen(glob), &search, ¬);
3166 len = strlen(search);
3167
3168 /* we do not support '!' for function probes */
3169 if (WARN_ON(not))
3170 return -EINVAL;
3171
3172 mutex_lock(&trace_probe_ops.regex_lock);
3173
3174 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3175 if (!hash) {
3176 count = -ENOMEM;
3177 goto out;
3178 }
3179
3180 if (unlikely(ftrace_disabled)) {
3181 count = -ENODEV;
3182 goto out;
3183 }
3184
3185 mutex_lock(&ftrace_lock);
3186
3187 do_for_each_ftrace_rec(pg, rec) {
3188
3189 if (!ftrace_match_record(rec, NULL, search, len, type))
3190 continue;
3191
3192 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3193 if (!entry) {
3194 /* If we did not process any, then return error */
3195 if (!count)
3196 count = -ENOMEM;
3197 goto out_unlock;
3198 }
3199
3200 count++;
3201
3202 entry->data = data;
3203
3204 /*
3205 * The caller might want to do something special
3206 * for each function we find. We call the callback
3207 * to give the caller an opportunity to do so.
3208 */
3209 if (ops->init) {
3210 if (ops->init(ops, rec->ip, &entry->data) < 0) {
3211 /* caller does not like this func */
3212 kfree(entry);
3213 continue;
3214 }
3215 }
3216
3217 ret = enter_record(hash, rec, 0);
3218 if (ret < 0) {
3219 kfree(entry);
3220 count = ret;
3221 goto out_unlock;
3222 }
3223
3224 entry->ops = ops;
3225 entry->ip = rec->ip;
3226
3227 key = hash_long(entry->ip, FTRACE_HASH_BITS);
3228 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3229
3230 } while_for_each_ftrace_rec();
3231
3232 ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3233 if (ret < 0)
3234 count = ret;
3235
3236 __enable_ftrace_function_probe();
3237
3238 out_unlock:
3239 mutex_unlock(&ftrace_lock);
3240 out:
3241 mutex_unlock(&trace_probe_ops.regex_lock);
3242 free_ftrace_hash(hash);
3243
3244 return count;
3245}
3246
3247enum {
3248 PROBE_TEST_FUNC = 1,
3249 PROBE_TEST_DATA = 2
3250};
3251
3252static void
3253__unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3254 void *data, int flags)
3255{
3256 struct ftrace_func_entry *rec_entry;
3257 struct ftrace_func_probe *entry;
3258 struct ftrace_func_probe *p;
3259 struct ftrace_hash **orig_hash = &trace_probe_ops.filter_hash;
3260 struct list_head free_list;
3261 struct ftrace_hash *hash;
3262 struct hlist_node *tmp;
3263 char str[KSYM_SYMBOL_LEN];
3264 int type = MATCH_FULL;
3265 int i, len = 0;
3266 char *search;
3267
3268 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3269 glob = NULL;
3270 else if (glob) {
3271 int not;
3272
3273 type = filter_parse_regex(glob, strlen(glob), &search, ¬);
3274 len = strlen(search);
3275
3276 /* we do not support '!' for function probes */
3277 if (WARN_ON(not))
3278 return;
3279 }
3280
3281 mutex_lock(&trace_probe_ops.regex_lock);
3282
3283 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3284 if (!hash)
3285 /* Hmm, should report this somehow */
3286 goto out_unlock;
3287
3288 INIT_LIST_HEAD(&free_list);
3289
3290 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3291 struct hlist_head *hhd = &ftrace_func_hash[i];
3292
3293 hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3294
3295 /* break up if statements for readability */
3296 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3297 continue;
3298
3299 if ((flags & PROBE_TEST_DATA) && entry->data != data)
3300 continue;
3301
3302 /* do this last, since it is the most expensive */
3303 if (glob) {
3304 kallsyms_lookup(entry->ip, NULL, NULL,
3305 NULL, str);
3306 if (!ftrace_match(str, glob, len, type))
3307 continue;
3308 }
3309
3310 rec_entry = ftrace_lookup_ip(hash, entry->ip);
3311 /* It is possible more than one entry had this ip */
3312 if (rec_entry)
3313 free_hash_entry(hash, rec_entry);
3314
3315 hlist_del_rcu(&entry->node);
3316 list_add(&entry->free_list, &free_list);
3317 }
3318 }
3319 mutex_lock(&ftrace_lock);
3320 __disable_ftrace_function_probe();
3321 /*
3322 * Remove after the disable is called. Otherwise, if the last
3323 * probe is removed, a null hash means *all enabled*.
3324 */
3325 ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3326 synchronize_sched();
3327 list_for_each_entry_safe(entry, p, &free_list, free_list) {
3328 list_del(&entry->free_list);
3329 ftrace_free_entry(entry);
3330 }
3331 mutex_unlock(&ftrace_lock);
3332
3333 out_unlock:
3334 mutex_unlock(&trace_probe_ops.regex_lock);
3335 free_ftrace_hash(hash);
3336}
3337
3338void
3339unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3340 void *data)
3341{
3342 __unregister_ftrace_function_probe(glob, ops, data,
3343 PROBE_TEST_FUNC | PROBE_TEST_DATA);
3344}
3345
3346void
3347unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3348{
3349 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3350}
3351
3352void unregister_ftrace_function_probe_all(char *glob)
3353{
3354 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3355}
3356
3357static LIST_HEAD(ftrace_commands);
3358static DEFINE_MUTEX(ftrace_cmd_mutex);
3359
3360/*
3361 * Currently we only register ftrace commands from __init, so mark this
3362 * __init too.
3363 */
3364__init int register_ftrace_command(struct ftrace_func_command *cmd)
3365{
3366 struct ftrace_func_command *p;
3367 int ret = 0;
3368
3369 mutex_lock(&ftrace_cmd_mutex);
3370 list_for_each_entry(p, &ftrace_commands, list) {
3371 if (strcmp(cmd->name, p->name) == 0) {
3372 ret = -EBUSY;
3373 goto out_unlock;
3374 }
3375 }
3376 list_add(&cmd->list, &ftrace_commands);
3377 out_unlock:
3378 mutex_unlock(&ftrace_cmd_mutex);
3379
3380 return ret;
3381}
3382
3383/*
3384 * Currently we only unregister ftrace commands from __init, so mark
3385 * this __init too.
3386 */
3387__init int unregister_ftrace_command(struct ftrace_func_command *cmd)
3388{
3389 struct ftrace_func_command *p, *n;
3390 int ret = -ENODEV;
3391
3392 mutex_lock(&ftrace_cmd_mutex);
3393 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3394 if (strcmp(cmd->name, p->name) == 0) {
3395 ret = 0;
3396 list_del_init(&p->list);
3397 goto out_unlock;
3398 }
3399 }
3400 out_unlock:
3401 mutex_unlock(&ftrace_cmd_mutex);
3402
3403 return ret;
3404}
3405
3406static int ftrace_process_regex(struct ftrace_hash *hash,
3407 char *buff, int len, int enable)
3408{
3409 char *func, *command, *next = buff;
3410 struct ftrace_func_command *p;
3411 int ret = -EINVAL;
3412
3413 func = strsep(&next, ":");
3414
3415 if (!next) {
3416 ret = ftrace_match_records(hash, func, len);
3417 if (!ret)
3418 ret = -EINVAL;
3419 if (ret < 0)
3420 return ret;
3421 return 0;
3422 }
3423
3424 /* command found */
3425
3426 command = strsep(&next, ":");
3427
3428 mutex_lock(&ftrace_cmd_mutex);
3429 list_for_each_entry(p, &ftrace_commands, list) {
3430 if (strcmp(p->name, command) == 0) {
3431 ret = p->func(hash, func, command, next, enable);
3432 goto out_unlock;
3433 }
3434 }
3435 out_unlock:
3436 mutex_unlock(&ftrace_cmd_mutex);
3437
3438 return ret;
3439}
3440
3441static ssize_t
3442ftrace_regex_write(struct file *file, const char __user *ubuf,
3443 size_t cnt, loff_t *ppos, int enable)
3444{
3445 struct ftrace_iterator *iter;
3446 struct trace_parser *parser;
3447 ssize_t ret, read;
3448
3449 if (!cnt)
3450 return 0;
3451
3452 if (file->f_mode & FMODE_READ) {
3453 struct seq_file *m = file->private_data;
3454 iter = m->private;
3455 } else
3456 iter = file->private_data;
3457
3458 if (unlikely(ftrace_disabled))
3459 return -ENODEV;
3460
3461 /* iter->hash is a local copy, so we don't need regex_lock */
3462
3463 parser = &iter->parser;
3464 read = trace_get_user(parser, ubuf, cnt, ppos);
3465
3466 if (read >= 0 && trace_parser_loaded(parser) &&
3467 !trace_parser_cont(parser)) {
3468 ret = ftrace_process_regex(iter->hash, parser->buffer,
3469 parser->idx, enable);
3470 trace_parser_clear(parser);
3471 if (ret < 0)
3472 goto out;
3473 }
3474
3475 ret = read;
3476 out:
3477 return ret;
3478}
3479
3480ssize_t
3481ftrace_filter_write(struct file *file, const char __user *ubuf,
3482 size_t cnt, loff_t *ppos)
3483{
3484 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
3485}
3486
3487ssize_t
3488ftrace_notrace_write(struct file *file, const char __user *ubuf,
3489 size_t cnt, loff_t *ppos)
3490{
3491 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
3492}
3493
3494static int
3495ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
3496{
3497 struct ftrace_func_entry *entry;
3498
3499 if (!ftrace_location(ip))
3500 return -EINVAL;
3501
3502 if (remove) {
3503 entry = ftrace_lookup_ip(hash, ip);
3504 if (!entry)
3505 return -ENOENT;
3506 free_hash_entry(hash, entry);
3507 return 0;
3508 }
3509
3510 return add_hash_entry(hash, ip);
3511}
3512
3513static void ftrace_ops_update_code(struct ftrace_ops *ops)
3514{
3515 if (ops->flags & FTRACE_OPS_FL_ENABLED && ftrace_enabled)
3516 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3517}
3518
3519static int
3520ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
3521 unsigned long ip, int remove, int reset, int enable)
3522{
3523 struct ftrace_hash **orig_hash;
3524 struct ftrace_hash *hash;
3525 int ret;
3526
3527 /* All global ops uses the global ops filters */
3528 if (ops->flags & FTRACE_OPS_FL_GLOBAL)
3529 ops = &global_ops;
3530
3531 if (unlikely(ftrace_disabled))
3532 return -ENODEV;
3533
3534 mutex_lock(&ops->regex_lock);
3535
3536 if (enable)
3537 orig_hash = &ops->filter_hash;
3538 else
3539 orig_hash = &ops->notrace_hash;
3540
3541 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3542 if (!hash) {
3543 ret = -ENOMEM;
3544 goto out_regex_unlock;
3545 }
3546
3547 if (reset)
3548 ftrace_filter_reset(hash);
3549 if (buf && !ftrace_match_records(hash, buf, len)) {
3550 ret = -EINVAL;
3551 goto out_regex_unlock;
3552 }
3553 if (ip) {
3554 ret = ftrace_match_addr(hash, ip, remove);
3555 if (ret < 0)
3556 goto out_regex_unlock;
3557 }
3558
3559 mutex_lock(&ftrace_lock);
3560 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3561 if (!ret)
3562 ftrace_ops_update_code(ops);
3563
3564 mutex_unlock(&ftrace_lock);
3565
3566 out_regex_unlock:
3567 mutex_unlock(&ops->regex_lock);
3568
3569 free_ftrace_hash(hash);
3570 return ret;
3571}
3572
3573static int
3574ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
3575 int reset, int enable)
3576{
3577 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
3578}
3579
3580/**
3581 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
3582 * @ops - the ops to set the filter with
3583 * @ip - the address to add to or remove from the filter.
3584 * @remove - non zero to remove the ip from the filter
3585 * @reset - non zero to reset all filters before applying this filter.
3586 *
3587 * Filters denote which functions should be enabled when tracing is enabled
3588 * If @ip is NULL, it failes to update filter.
3589 */
3590int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
3591 int remove, int reset)
3592{
3593 ftrace_ops_init(ops);
3594 return ftrace_set_addr(ops, ip, remove, reset, 1);
3595}
3596EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
3597
3598static int
3599ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3600 int reset, int enable)
3601{
3602 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
3603}
3604
3605/**
3606 * ftrace_set_filter - set a function to filter on in ftrace
3607 * @ops - the ops to set the filter with
3608 * @buf - the string that holds the function filter text.
3609 * @len - the length of the string.
3610 * @reset - non zero to reset all filters before applying this filter.
3611 *
3612 * Filters denote which functions should be enabled when tracing is enabled.
3613 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3614 */
3615int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
3616 int len, int reset)
3617{
3618 ftrace_ops_init(ops);
3619 return ftrace_set_regex(ops, buf, len, reset, 1);
3620}
3621EXPORT_SYMBOL_GPL(ftrace_set_filter);
3622
3623/**
3624 * ftrace_set_notrace - set a function to not trace in ftrace
3625 * @ops - the ops to set the notrace filter with
3626 * @buf - the string that holds the function notrace text.
3627 * @len - the length of the string.
3628 * @reset - non zero to reset all filters before applying this filter.
3629 *
3630 * Notrace Filters denote which functions should not be enabled when tracing
3631 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3632 * for tracing.
3633 */
3634int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
3635 int len, int reset)
3636{
3637 ftrace_ops_init(ops);
3638 return ftrace_set_regex(ops, buf, len, reset, 0);
3639}
3640EXPORT_SYMBOL_GPL(ftrace_set_notrace);
3641/**
3642 * ftrace_set_filter - set a function to filter on in ftrace
3643 * @ops - the ops to set the filter with
3644 * @buf - the string that holds the function filter text.
3645 * @len - the length of the string.
3646 * @reset - non zero to reset all filters before applying this filter.
3647 *
3648 * Filters denote which functions should be enabled when tracing is enabled.
3649 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3650 */
3651void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
3652{
3653 ftrace_set_regex(&global_ops, buf, len, reset, 1);
3654}
3655EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
3656
3657/**
3658 * ftrace_set_notrace - set a function to not trace in ftrace
3659 * @ops - the ops to set the notrace filter with
3660 * @buf - the string that holds the function notrace text.
3661 * @len - the length of the string.
3662 * @reset - non zero to reset all filters before applying this filter.
3663 *
3664 * Notrace Filters denote which functions should not be enabled when tracing
3665 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3666 * for tracing.
3667 */
3668void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
3669{
3670 ftrace_set_regex(&global_ops, buf, len, reset, 0);
3671}
3672EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3673
3674/*
3675 * command line interface to allow users to set filters on boot up.
3676 */
3677#define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
3678static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3679static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3680
3681/* Used by function selftest to not test if filter is set */
3682bool ftrace_filter_param __initdata;
3683
3684static int __init set_ftrace_notrace(char *str)
3685{
3686 ftrace_filter_param = true;
3687 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3688 return 1;
3689}
3690__setup("ftrace_notrace=", set_ftrace_notrace);
3691
3692static int __init set_ftrace_filter(char *str)
3693{
3694 ftrace_filter_param = true;
3695 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3696 return 1;
3697}
3698__setup("ftrace_filter=", set_ftrace_filter);
3699
3700#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3701static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3702static int ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer);
3703
3704static int __init set_graph_function(char *str)
3705{
3706 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3707 return 1;
3708}
3709__setup("ftrace_graph_filter=", set_graph_function);
3710
3711static void __init set_ftrace_early_graph(char *buf)
3712{
3713 int ret;
3714 char *func;
3715
3716 while (buf) {
3717 func = strsep(&buf, ",");
3718 /* we allow only one expression at a time */
3719 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3720 FTRACE_GRAPH_MAX_FUNCS, func);
3721 if (ret)
3722 printk(KERN_DEBUG "ftrace: function %s not "
3723 "traceable\n", func);
3724 }
3725}
3726#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3727
3728void __init
3729ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3730{
3731 char *func;
3732
3733 ftrace_ops_init(ops);
3734
3735 while (buf) {
3736 func = strsep(&buf, ",");
3737 ftrace_set_regex(ops, func, strlen(func), 0, enable);
3738 }
3739}
3740
3741static void __init set_ftrace_early_filters(void)
3742{
3743 if (ftrace_filter_buf[0])
3744 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
3745 if (ftrace_notrace_buf[0])
3746 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
3747#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3748 if (ftrace_graph_buf[0])
3749 set_ftrace_early_graph(ftrace_graph_buf);
3750#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3751}
3752
3753int ftrace_regex_release(struct inode *inode, struct file *file)
3754{
3755 struct seq_file *m = (struct seq_file *)file->private_data;
3756 struct ftrace_iterator *iter;
3757 struct ftrace_hash **orig_hash;
3758 struct trace_parser *parser;
3759 int filter_hash;
3760 int ret;
3761
3762 if (file->f_mode & FMODE_READ) {
3763 iter = m->private;
3764 seq_release(inode, file);
3765 } else
3766 iter = file->private_data;
3767
3768 parser = &iter->parser;
3769 if (trace_parser_loaded(parser)) {
3770 parser->buffer[parser->idx] = 0;
3771 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3772 }
3773
3774 trace_parser_put(parser);
3775
3776 mutex_lock(&iter->ops->regex_lock);
3777
3778 if (file->f_mode & FMODE_WRITE) {
3779 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3780
3781 if (filter_hash)
3782 orig_hash = &iter->ops->filter_hash;
3783 else
3784 orig_hash = &iter->ops->notrace_hash;
3785
3786 mutex_lock(&ftrace_lock);
3787 ret = ftrace_hash_move(iter->ops, filter_hash,
3788 orig_hash, iter->hash);
3789 if (!ret)
3790 ftrace_ops_update_code(iter->ops);
3791
3792 mutex_unlock(&ftrace_lock);
3793 }
3794
3795 mutex_unlock(&iter->ops->regex_lock);
3796 free_ftrace_hash(iter->hash);
3797 kfree(iter);
3798
3799 return 0;
3800}
3801
3802static const struct file_operations ftrace_avail_fops = {
3803 .open = ftrace_avail_open,
3804 .read = seq_read,
3805 .llseek = seq_lseek,
3806 .release = seq_release_private,
3807};
3808
3809static const struct file_operations ftrace_enabled_fops = {
3810 .open = ftrace_enabled_open,
3811 .read = seq_read,
3812 .llseek = seq_lseek,
3813 .release = seq_release_private,
3814};
3815
3816static const struct file_operations ftrace_filter_fops = {
3817 .open = ftrace_filter_open,
3818 .read = seq_read,
3819 .write = ftrace_filter_write,
3820 .llseek = tracing_lseek,
3821 .release = ftrace_regex_release,
3822};
3823
3824static const struct file_operations ftrace_notrace_fops = {
3825 .open = ftrace_notrace_open,
3826 .read = seq_read,
3827 .write = ftrace_notrace_write,
3828 .llseek = tracing_lseek,
3829 .release = ftrace_regex_release,
3830};
3831
3832#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3833
3834static DEFINE_MUTEX(graph_lock);
3835
3836int ftrace_graph_count;
3837int ftrace_graph_notrace_count;
3838unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3839unsigned long ftrace_graph_notrace_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3840
3841struct ftrace_graph_data {
3842 unsigned long *table;
3843 size_t size;
3844 int *count;
3845 const struct seq_operations *seq_ops;
3846};
3847
3848static void *
3849__g_next(struct seq_file *m, loff_t *pos)
3850{
3851 struct ftrace_graph_data *fgd = m->private;
3852
3853 if (*pos >= *fgd->count)
3854 return NULL;
3855 return &fgd->table[*pos];
3856}
3857
3858static void *
3859g_next(struct seq_file *m, void *v, loff_t *pos)
3860{
3861 (*pos)++;
3862 return __g_next(m, pos);
3863}
3864
3865static void *g_start(struct seq_file *m, loff_t *pos)
3866{
3867 struct ftrace_graph_data *fgd = m->private;
3868
3869 mutex_lock(&graph_lock);
3870
3871 /* Nothing, tell g_show to print all functions are enabled */
3872 if (!*fgd->count && !*pos)
3873 return (void *)1;
3874
3875 return __g_next(m, pos);
3876}
3877
3878static void g_stop(struct seq_file *m, void *p)
3879{
3880 mutex_unlock(&graph_lock);
3881}
3882
3883static int g_show(struct seq_file *m, void *v)
3884{
3885 unsigned long *ptr = v;
3886
3887 if (!ptr)
3888 return 0;
3889
3890 if (ptr == (unsigned long *)1) {
3891 seq_printf(m, "#### all functions enabled ####\n");
3892 return 0;
3893 }
3894
3895 seq_printf(m, "%ps\n", (void *)*ptr);
3896
3897 return 0;
3898}
3899
3900static const struct seq_operations ftrace_graph_seq_ops = {
3901 .start = g_start,
3902 .next = g_next,
3903 .stop = g_stop,
3904 .show = g_show,
3905};
3906
3907static int
3908__ftrace_graph_open(struct inode *inode, struct file *file,
3909 struct ftrace_graph_data *fgd)
3910{
3911 int ret = 0;
3912
3913 mutex_lock(&graph_lock);
3914 if ((file->f_mode & FMODE_WRITE) &&
3915 (file->f_flags & O_TRUNC)) {
3916 *fgd->count = 0;
3917 memset(fgd->table, 0, fgd->size * sizeof(*fgd->table));
3918 }
3919 mutex_unlock(&graph_lock);
3920
3921 if (file->f_mode & FMODE_READ) {
3922 ret = seq_open(file, fgd->seq_ops);
3923 if (!ret) {
3924 struct seq_file *m = file->private_data;
3925 m->private = fgd;
3926 }
3927 } else
3928 file->private_data = fgd;
3929
3930 return ret;
3931}
3932
3933static int
3934ftrace_graph_open(struct inode *inode, struct file *file)
3935{
3936 struct ftrace_graph_data *fgd;
3937
3938 if (unlikely(ftrace_disabled))
3939 return -ENODEV;
3940
3941 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
3942 if (fgd == NULL)
3943 return -ENOMEM;
3944
3945 fgd->table = ftrace_graph_funcs;
3946 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
3947 fgd->count = &ftrace_graph_count;
3948 fgd->seq_ops = &ftrace_graph_seq_ops;
3949
3950 return __ftrace_graph_open(inode, file, fgd);
3951}
3952
3953static int
3954ftrace_graph_notrace_open(struct inode *inode, struct file *file)
3955{
3956 struct ftrace_graph_data *fgd;
3957
3958 if (unlikely(ftrace_disabled))
3959 return -ENODEV;
3960
3961 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
3962 if (fgd == NULL)
3963 return -ENOMEM;
3964
3965 fgd->table = ftrace_graph_notrace_funcs;
3966 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
3967 fgd->count = &ftrace_graph_notrace_count;
3968 fgd->seq_ops = &ftrace_graph_seq_ops;
3969
3970 return __ftrace_graph_open(inode, file, fgd);
3971}
3972
3973static int
3974ftrace_graph_release(struct inode *inode, struct file *file)
3975{
3976 if (file->f_mode & FMODE_READ) {
3977 struct seq_file *m = file->private_data;
3978
3979 kfree(m->private);
3980 seq_release(inode, file);
3981 } else {
3982 kfree(file->private_data);
3983 }
3984
3985 return 0;
3986}
3987
3988static int
3989ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer)
3990{
3991 struct dyn_ftrace *rec;
3992 struct ftrace_page *pg;
3993 int search_len;
3994 int fail = 1;
3995 int type, not;
3996 char *search;
3997 bool exists;
3998 int i;
3999
4000 /* decode regex */
4001 type = filter_parse_regex(buffer, strlen(buffer), &search, ¬);
4002 if (!not && *idx >= size)
4003 return -EBUSY;
4004
4005 search_len = strlen(search);
4006
4007 mutex_lock(&ftrace_lock);
4008
4009 if (unlikely(ftrace_disabled)) {
4010 mutex_unlock(&ftrace_lock);
4011 return -ENODEV;
4012 }
4013
4014 do_for_each_ftrace_rec(pg, rec) {
4015
4016 if (ftrace_match_record(rec, NULL, search, search_len, type)) {
4017 /* if it is in the array */
4018 exists = false;
4019 for (i = 0; i < *idx; i++) {
4020 if (array[i] == rec->ip) {
4021 exists = true;
4022 break;
4023 }
4024 }
4025
4026 if (!not) {
4027 fail = 0;
4028 if (!exists) {
4029 array[(*idx)++] = rec->ip;
4030 if (*idx >= size)
4031 goto out;
4032 }
4033 } else {
4034 if (exists) {
4035 array[i] = array[--(*idx)];
4036 array[*idx] = 0;
4037 fail = 0;
4038 }
4039 }
4040 }
4041 } while_for_each_ftrace_rec();
4042out:
4043 mutex_unlock(&ftrace_lock);
4044
4045 if (fail)
4046 return -EINVAL;
4047
4048 return 0;
4049}
4050
4051static ssize_t
4052ftrace_graph_write(struct file *file, const char __user *ubuf,
4053 size_t cnt, loff_t *ppos)
4054{
4055 struct trace_parser parser;
4056 ssize_t read, ret = 0;
4057 struct ftrace_graph_data *fgd = file->private_data;
4058
4059 if (!cnt)
4060 return 0;
4061
4062 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX))
4063 return -ENOMEM;
4064
4065 read = trace_get_user(&parser, ubuf, cnt, ppos);
4066
4067 if (read >= 0 && trace_parser_loaded((&parser))) {
4068 parser.buffer[parser.idx] = 0;
4069
4070 mutex_lock(&graph_lock);
4071
4072 /* we allow only one expression at a time */
4073 ret = ftrace_set_func(fgd->table, fgd->count, fgd->size,
4074 parser.buffer);
4075
4076 mutex_unlock(&graph_lock);
4077 }
4078
4079 if (!ret)
4080 ret = read;
4081
4082 trace_parser_put(&parser);
4083
4084 return ret;
4085}
4086
4087static const struct file_operations ftrace_graph_fops = {
4088 .open = ftrace_graph_open,
4089 .read = seq_read,
4090 .write = ftrace_graph_write,
4091 .llseek = tracing_lseek,
4092 .release = ftrace_graph_release,
4093};
4094
4095static const struct file_operations ftrace_graph_notrace_fops = {
4096 .open = ftrace_graph_notrace_open,
4097 .read = seq_read,
4098 .write = ftrace_graph_write,
4099 .llseek = tracing_lseek,
4100 .release = ftrace_graph_release,
4101};
4102#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4103
4104void ftrace_create_filter_files(struct ftrace_ops *ops,
4105 struct dentry *parent)
4106{
4107
4108 trace_create_file("set_ftrace_filter", 0644, parent,
4109 ops, &ftrace_filter_fops);
4110
4111 trace_create_file("set_ftrace_notrace", 0644, parent,
4112 ops, &ftrace_notrace_fops);
4113}
4114
4115/*
4116 * The name "destroy_filter_files" is really a misnomer. Although
4117 * in the future, it may actualy delete the files, but this is
4118 * really intended to make sure the ops passed in are disabled
4119 * and that when this function returns, the caller is free to
4120 * free the ops.
4121 *
4122 * The "destroy" name is only to match the "create" name that this
4123 * should be paired with.
4124 */
4125void ftrace_destroy_filter_files(struct ftrace_ops *ops)
4126{
4127 mutex_lock(&ftrace_lock);
4128 if (ops->flags & FTRACE_OPS_FL_ENABLED)
4129 ftrace_shutdown(ops, 0);
4130 ops->flags |= FTRACE_OPS_FL_DELETED;
4131 mutex_unlock(&ftrace_lock);
4132}
4133
4134static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
4135{
4136
4137 trace_create_file("available_filter_functions", 0444,
4138 d_tracer, NULL, &ftrace_avail_fops);
4139
4140 trace_create_file("enabled_functions", 0444,
4141 d_tracer, NULL, &ftrace_enabled_fops);
4142
4143 ftrace_create_filter_files(&global_ops, d_tracer);
4144
4145#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4146 trace_create_file("set_graph_function", 0444, d_tracer,
4147 NULL,
4148 &ftrace_graph_fops);
4149 trace_create_file("set_graph_notrace", 0444, d_tracer,
4150 NULL,
4151 &ftrace_graph_notrace_fops);
4152#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4153
4154 return 0;
4155}
4156
4157static int ftrace_cmp_ips(const void *a, const void *b)
4158{
4159 const unsigned long *ipa = a;
4160 const unsigned long *ipb = b;
4161
4162 if (*ipa > *ipb)
4163 return 1;
4164 if (*ipa < *ipb)
4165 return -1;
4166 return 0;
4167}
4168
4169static void ftrace_swap_ips(void *a, void *b, int size)
4170{
4171 unsigned long *ipa = a;
4172 unsigned long *ipb = b;
4173 unsigned long t;
4174
4175 t = *ipa;
4176 *ipa = *ipb;
4177 *ipb = t;
4178}
4179
4180static int ftrace_process_locs(struct module *mod,
4181 unsigned long *start,
4182 unsigned long *end)
4183{
4184 struct ftrace_page *start_pg;
4185 struct ftrace_page *pg;
4186 struct dyn_ftrace *rec;
4187 unsigned long count;
4188 unsigned long *p;
4189 unsigned long addr;
4190 unsigned long flags = 0; /* Shut up gcc */
4191 int ret = -ENOMEM;
4192
4193 count = end - start;
4194
4195 if (!count)
4196 return 0;
4197
4198 sort(start, count, sizeof(*start),
4199 ftrace_cmp_ips, ftrace_swap_ips);
4200
4201 start_pg = ftrace_allocate_pages(count);
4202 if (!start_pg)
4203 return -ENOMEM;
4204
4205 mutex_lock(&ftrace_lock);
4206
4207 /*
4208 * Core and each module needs their own pages, as
4209 * modules will free them when they are removed.
4210 * Force a new page to be allocated for modules.
4211 */
4212 if (!mod) {
4213 WARN_ON(ftrace_pages || ftrace_pages_start);
4214 /* First initialization */
4215 ftrace_pages = ftrace_pages_start = start_pg;
4216 } else {
4217 if (!ftrace_pages)
4218 goto out;
4219
4220 if (WARN_ON(ftrace_pages->next)) {
4221 /* Hmm, we have free pages? */
4222 while (ftrace_pages->next)
4223 ftrace_pages = ftrace_pages->next;
4224 }
4225
4226 ftrace_pages->next = start_pg;
4227 }
4228
4229 p = start;
4230 pg = start_pg;
4231 while (p < end) {
4232 addr = ftrace_call_adjust(*p++);
4233 /*
4234 * Some architecture linkers will pad between
4235 * the different mcount_loc sections of different
4236 * object files to satisfy alignments.
4237 * Skip any NULL pointers.
4238 */
4239 if (!addr)
4240 continue;
4241
4242 if (pg->index == pg->size) {
4243 /* We should have allocated enough */
4244 if (WARN_ON(!pg->next))
4245 break;
4246 pg = pg->next;
4247 }
4248
4249 rec = &pg->records[pg->index++];
4250 rec->ip = addr;
4251 }
4252
4253 /* We should have used all pages */
4254 WARN_ON(pg->next);
4255
4256 /* Assign the last page to ftrace_pages */
4257 ftrace_pages = pg;
4258
4259 /*
4260 * We only need to disable interrupts on start up
4261 * because we are modifying code that an interrupt
4262 * may execute, and the modification is not atomic.
4263 * But for modules, nothing runs the code we modify
4264 * until we are finished with it, and there's no
4265 * reason to cause large interrupt latencies while we do it.
4266 */
4267 if (!mod)
4268 local_irq_save(flags);
4269 ftrace_update_code(mod, start_pg);
4270 if (!mod)
4271 local_irq_restore(flags);
4272 ret = 0;
4273 out:
4274 mutex_unlock(&ftrace_lock);
4275
4276 return ret;
4277}
4278
4279#ifdef CONFIG_MODULES
4280
4281#define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4282
4283void ftrace_release_mod(struct module *mod)
4284{
4285 struct dyn_ftrace *rec;
4286 struct ftrace_page **last_pg;
4287 struct ftrace_page *pg;
4288 int order;
4289
4290 mutex_lock(&ftrace_lock);
4291
4292 if (ftrace_disabled)
4293 goto out_unlock;
4294
4295 /*
4296 * Each module has its own ftrace_pages, remove
4297 * them from the list.
4298 */
4299 last_pg = &ftrace_pages_start;
4300 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4301 rec = &pg->records[0];
4302 if (within_module_core(rec->ip, mod)) {
4303 /*
4304 * As core pages are first, the first
4305 * page should never be a module page.
4306 */
4307 if (WARN_ON(pg == ftrace_pages_start))
4308 goto out_unlock;
4309
4310 /* Check if we are deleting the last page */
4311 if (pg == ftrace_pages)
4312 ftrace_pages = next_to_ftrace_page(last_pg);
4313
4314 *last_pg = pg->next;
4315 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4316 free_pages((unsigned long)pg->records, order);
4317 kfree(pg);
4318 } else
4319 last_pg = &pg->next;
4320 }
4321 out_unlock:
4322 mutex_unlock(&ftrace_lock);
4323}
4324
4325static void ftrace_init_module(struct module *mod,
4326 unsigned long *start, unsigned long *end)
4327{
4328 if (ftrace_disabled || start == end)
4329 return;
4330 ftrace_process_locs(mod, start, end);
4331}
4332
4333void ftrace_module_init(struct module *mod)
4334{
4335 ftrace_init_module(mod, mod->ftrace_callsites,
4336 mod->ftrace_callsites +
4337 mod->num_ftrace_callsites);
4338}
4339
4340static int ftrace_module_notify_exit(struct notifier_block *self,
4341 unsigned long val, void *data)
4342{
4343 struct module *mod = data;
4344
4345 if (val == MODULE_STATE_GOING)
4346 ftrace_release_mod(mod);
4347
4348 return 0;
4349}
4350#else
4351static int ftrace_module_notify_exit(struct notifier_block *self,
4352 unsigned long val, void *data)
4353{
4354 return 0;
4355}
4356#endif /* CONFIG_MODULES */
4357
4358struct notifier_block ftrace_module_exit_nb = {
4359 .notifier_call = ftrace_module_notify_exit,
4360 .priority = INT_MIN, /* Run after anything that can remove kprobes */
4361};
4362
4363void __init ftrace_init(void)
4364{
4365 extern unsigned long __start_mcount_loc[];
4366 extern unsigned long __stop_mcount_loc[];
4367 unsigned long count, flags;
4368 int ret;
4369
4370 local_irq_save(flags);
4371 ret = ftrace_dyn_arch_init();
4372 local_irq_restore(flags);
4373 if (ret)
4374 goto failed;
4375
4376 count = __stop_mcount_loc - __start_mcount_loc;
4377 if (!count) {
4378 pr_info("ftrace: No functions to be traced?\n");
4379 goto failed;
4380 }
4381
4382 pr_info("ftrace: allocating %ld entries in %ld pages\n",
4383 count, count / ENTRIES_PER_PAGE + 1);
4384
4385 last_ftrace_enabled = ftrace_enabled = 1;
4386
4387 ret = ftrace_process_locs(NULL,
4388 __start_mcount_loc,
4389 __stop_mcount_loc);
4390
4391 ret = register_module_notifier(&ftrace_module_exit_nb);
4392 if (ret)
4393 pr_warning("Failed to register trace ftrace module exit notifier\n");
4394
4395 set_ftrace_early_filters();
4396
4397 return;
4398 failed:
4399 ftrace_disabled = 1;
4400}
4401
4402#else
4403
4404static struct ftrace_ops global_ops = {
4405 .func = ftrace_stub,
4406 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4407 INIT_REGEX_LOCK(global_ops)
4408};
4409
4410static int __init ftrace_nodyn_init(void)
4411{
4412 ftrace_enabled = 1;
4413 return 0;
4414}
4415core_initcall(ftrace_nodyn_init);
4416
4417static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
4418static inline void ftrace_startup_enable(int command) { }
4419/* Keep as macros so we do not need to define the commands */
4420# define ftrace_startup(ops, command) \
4421 ({ \
4422 int ___ret = __register_ftrace_function(ops); \
4423 if (!___ret) \
4424 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
4425 ___ret; \
4426 })
4427# define ftrace_shutdown(ops, command) \
4428 ({ \
4429 int ___ret = __unregister_ftrace_function(ops); \
4430 if (!___ret) \
4431 (ops)->flags &= ~FTRACE_OPS_FL_ENABLED; \
4432 ___ret; \
4433 })
4434
4435# define ftrace_startup_sysctl() do { } while (0)
4436# define ftrace_shutdown_sysctl() do { } while (0)
4437
4438static inline int
4439ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
4440{
4441 return 1;
4442}
4443
4444#endif /* CONFIG_DYNAMIC_FTRACE */
4445
4446static void
4447ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip,
4448 struct ftrace_ops *op, struct pt_regs *regs)
4449{
4450 if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
4451 return;
4452
4453 /*
4454 * Some of the ops may be dynamically allocated,
4455 * they must be freed after a synchronize_sched().
4456 */
4457 preempt_disable_notrace();
4458 trace_recursion_set(TRACE_CONTROL_BIT);
4459
4460 /*
4461 * Control funcs (perf) uses RCU. Only trace if
4462 * RCU is currently active.
4463 */
4464 if (!rcu_is_watching())
4465 goto out;
4466
4467 do_for_each_ftrace_op(op, ftrace_control_list) {
4468 if (!(op->flags & FTRACE_OPS_FL_STUB) &&
4469 !ftrace_function_local_disabled(op) &&
4470 ftrace_ops_test(op, ip, regs))
4471 op->func(ip, parent_ip, op, regs);
4472 } while_for_each_ftrace_op(op);
4473 out:
4474 trace_recursion_clear(TRACE_CONTROL_BIT);
4475 preempt_enable_notrace();
4476}
4477
4478static struct ftrace_ops control_ops = {
4479 .func = ftrace_ops_control_func,
4480 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4481 INIT_REGEX_LOCK(control_ops)
4482};
4483
4484static inline void
4485__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4486 struct ftrace_ops *ignored, struct pt_regs *regs)
4487{
4488 struct ftrace_ops *op;
4489 int bit;
4490
4491 if (function_trace_stop)
4492 return;
4493
4494 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
4495 if (bit < 0)
4496 return;
4497
4498 /*
4499 * Some of the ops may be dynamically allocated,
4500 * they must be freed after a synchronize_sched().
4501 */
4502 preempt_disable_notrace();
4503 do_for_each_ftrace_op(op, ftrace_ops_list) {
4504 if (ftrace_ops_test(op, ip, regs))
4505 op->func(ip, parent_ip, op, regs);
4506 } while_for_each_ftrace_op(op);
4507 preempt_enable_notrace();
4508 trace_clear_recursion(bit);
4509}
4510
4511/*
4512 * Some archs only support passing ip and parent_ip. Even though
4513 * the list function ignores the op parameter, we do not want any
4514 * C side effects, where a function is called without the caller
4515 * sending a third parameter.
4516 * Archs are to support both the regs and ftrace_ops at the same time.
4517 * If they support ftrace_ops, it is assumed they support regs.
4518 * If call backs want to use regs, they must either check for regs
4519 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
4520 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
4521 * An architecture can pass partial regs with ftrace_ops and still
4522 * set the ARCH_SUPPORT_FTARCE_OPS.
4523 */
4524#if ARCH_SUPPORTS_FTRACE_OPS
4525static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4526 struct ftrace_ops *op, struct pt_regs *regs)
4527{
4528 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
4529}
4530#else
4531static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
4532{
4533 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
4534}
4535#endif
4536
4537static void clear_ftrace_swapper(void)
4538{
4539 struct task_struct *p;
4540 int cpu;
4541
4542 get_online_cpus();
4543 for_each_online_cpu(cpu) {
4544 p = idle_task(cpu);
4545 clear_tsk_trace_trace(p);
4546 }
4547 put_online_cpus();
4548}
4549
4550static void set_ftrace_swapper(void)
4551{
4552 struct task_struct *p;
4553 int cpu;
4554
4555 get_online_cpus();
4556 for_each_online_cpu(cpu) {
4557 p = idle_task(cpu);
4558 set_tsk_trace_trace(p);
4559 }
4560 put_online_cpus();
4561}
4562
4563static void clear_ftrace_pid(struct pid *pid)
4564{
4565 struct task_struct *p;
4566
4567 rcu_read_lock();
4568 do_each_pid_task(pid, PIDTYPE_PID, p) {
4569 clear_tsk_trace_trace(p);
4570 } while_each_pid_task(pid, PIDTYPE_PID, p);
4571 rcu_read_unlock();
4572
4573 put_pid(pid);
4574}
4575
4576static void set_ftrace_pid(struct pid *pid)
4577{
4578 struct task_struct *p;
4579
4580 rcu_read_lock();
4581 do_each_pid_task(pid, PIDTYPE_PID, p) {
4582 set_tsk_trace_trace(p);
4583 } while_each_pid_task(pid, PIDTYPE_PID, p);
4584 rcu_read_unlock();
4585}
4586
4587static void clear_ftrace_pid_task(struct pid *pid)
4588{
4589 if (pid == ftrace_swapper_pid)
4590 clear_ftrace_swapper();
4591 else
4592 clear_ftrace_pid(pid);
4593}
4594
4595static void set_ftrace_pid_task(struct pid *pid)
4596{
4597 if (pid == ftrace_swapper_pid)
4598 set_ftrace_swapper();
4599 else
4600 set_ftrace_pid(pid);
4601}
4602
4603static int ftrace_pid_add(int p)
4604{
4605 struct pid *pid;
4606 struct ftrace_pid *fpid;
4607 int ret = -EINVAL;
4608
4609 mutex_lock(&ftrace_lock);
4610
4611 if (!p)
4612 pid = ftrace_swapper_pid;
4613 else
4614 pid = find_get_pid(p);
4615
4616 if (!pid)
4617 goto out;
4618
4619 ret = 0;
4620
4621 list_for_each_entry(fpid, &ftrace_pids, list)
4622 if (fpid->pid == pid)
4623 goto out_put;
4624
4625 ret = -ENOMEM;
4626
4627 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
4628 if (!fpid)
4629 goto out_put;
4630
4631 list_add(&fpid->list, &ftrace_pids);
4632 fpid->pid = pid;
4633
4634 set_ftrace_pid_task(pid);
4635
4636 ftrace_update_pid_func();
4637 ftrace_startup_enable(0);
4638
4639 mutex_unlock(&ftrace_lock);
4640 return 0;
4641
4642out_put:
4643 if (pid != ftrace_swapper_pid)
4644 put_pid(pid);
4645
4646out:
4647 mutex_unlock(&ftrace_lock);
4648 return ret;
4649}
4650
4651static void ftrace_pid_reset(void)
4652{
4653 struct ftrace_pid *fpid, *safe;
4654
4655 mutex_lock(&ftrace_lock);
4656 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
4657 struct pid *pid = fpid->pid;
4658
4659 clear_ftrace_pid_task(pid);
4660
4661 list_del(&fpid->list);
4662 kfree(fpid);
4663 }
4664
4665 ftrace_update_pid_func();
4666 ftrace_startup_enable(0);
4667
4668 mutex_unlock(&ftrace_lock);
4669}
4670
4671static void *fpid_start(struct seq_file *m, loff_t *pos)
4672{
4673 mutex_lock(&ftrace_lock);
4674
4675 if (list_empty(&ftrace_pids) && (!*pos))
4676 return (void *) 1;
4677
4678 return seq_list_start(&ftrace_pids, *pos);
4679}
4680
4681static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
4682{
4683 if (v == (void *)1)
4684 return NULL;
4685
4686 return seq_list_next(v, &ftrace_pids, pos);
4687}
4688
4689static void fpid_stop(struct seq_file *m, void *p)
4690{
4691 mutex_unlock(&ftrace_lock);
4692}
4693
4694static int fpid_show(struct seq_file *m, void *v)
4695{
4696 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
4697
4698 if (v == (void *)1) {
4699 seq_printf(m, "no pid\n");
4700 return 0;
4701 }
4702
4703 if (fpid->pid == ftrace_swapper_pid)
4704 seq_printf(m, "swapper tasks\n");
4705 else
4706 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
4707
4708 return 0;
4709}
4710
4711static const struct seq_operations ftrace_pid_sops = {
4712 .start = fpid_start,
4713 .next = fpid_next,
4714 .stop = fpid_stop,
4715 .show = fpid_show,
4716};
4717
4718static int
4719ftrace_pid_open(struct inode *inode, struct file *file)
4720{
4721 int ret = 0;
4722
4723 if ((file->f_mode & FMODE_WRITE) &&
4724 (file->f_flags & O_TRUNC))
4725 ftrace_pid_reset();
4726
4727 if (file->f_mode & FMODE_READ)
4728 ret = seq_open(file, &ftrace_pid_sops);
4729
4730 return ret;
4731}
4732
4733static ssize_t
4734ftrace_pid_write(struct file *filp, const char __user *ubuf,
4735 size_t cnt, loff_t *ppos)
4736{
4737 char buf[64], *tmp;
4738 long val;
4739 int ret;
4740
4741 if (cnt >= sizeof(buf))
4742 return -EINVAL;
4743
4744 if (copy_from_user(&buf, ubuf, cnt))
4745 return -EFAULT;
4746
4747 buf[cnt] = 0;
4748
4749 /*
4750 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
4751 * to clean the filter quietly.
4752 */
4753 tmp = strstrip(buf);
4754 if (strlen(tmp) == 0)
4755 return 1;
4756
4757 ret = kstrtol(tmp, 10, &val);
4758 if (ret < 0)
4759 return ret;
4760
4761 ret = ftrace_pid_add(val);
4762
4763 return ret ? ret : cnt;
4764}
4765
4766static int
4767ftrace_pid_release(struct inode *inode, struct file *file)
4768{
4769 if (file->f_mode & FMODE_READ)
4770 seq_release(inode, file);
4771
4772 return 0;
4773}
4774
4775static const struct file_operations ftrace_pid_fops = {
4776 .open = ftrace_pid_open,
4777 .write = ftrace_pid_write,
4778 .read = seq_read,
4779 .llseek = tracing_lseek,
4780 .release = ftrace_pid_release,
4781};
4782
4783static __init int ftrace_init_debugfs(void)
4784{
4785 struct dentry *d_tracer;
4786
4787 d_tracer = tracing_init_dentry();
4788 if (!d_tracer)
4789 return 0;
4790
4791 ftrace_init_dyn_debugfs(d_tracer);
4792
4793 trace_create_file("set_ftrace_pid", 0644, d_tracer,
4794 NULL, &ftrace_pid_fops);
4795
4796 ftrace_profile_debugfs(d_tracer);
4797
4798 return 0;
4799}
4800fs_initcall(ftrace_init_debugfs);
4801
4802/**
4803 * ftrace_kill - kill ftrace
4804 *
4805 * This function should be used by panic code. It stops ftrace
4806 * but in a not so nice way. If you need to simply kill ftrace
4807 * from a non-atomic section, use ftrace_kill.
4808 */
4809void ftrace_kill(void)
4810{
4811 ftrace_disabled = 1;
4812 ftrace_enabled = 0;
4813 clear_ftrace_function();
4814}
4815
4816/**
4817 * Test if ftrace is dead or not.
4818 */
4819int ftrace_is_dead(void)
4820{
4821 return ftrace_disabled;
4822}
4823
4824/**
4825 * register_ftrace_function - register a function for profiling
4826 * @ops - ops structure that holds the function for profiling.
4827 *
4828 * Register a function to be called by all functions in the
4829 * kernel.
4830 *
4831 * Note: @ops->func and all the functions it calls must be labeled
4832 * with "notrace", otherwise it will go into a
4833 * recursive loop.
4834 */
4835int register_ftrace_function(struct ftrace_ops *ops)
4836{
4837 int ret = -1;
4838
4839 ftrace_ops_init(ops);
4840
4841 mutex_lock(&ftrace_lock);
4842
4843 ret = ftrace_startup(ops, 0);
4844
4845 mutex_unlock(&ftrace_lock);
4846
4847 return ret;
4848}
4849EXPORT_SYMBOL_GPL(register_ftrace_function);
4850
4851/**
4852 * unregister_ftrace_function - unregister a function for profiling.
4853 * @ops - ops structure that holds the function to unregister
4854 *
4855 * Unregister a function that was added to be called by ftrace profiling.
4856 */
4857int unregister_ftrace_function(struct ftrace_ops *ops)
4858{
4859 int ret;
4860
4861 mutex_lock(&ftrace_lock);
4862 ret = ftrace_shutdown(ops, 0);
4863 mutex_unlock(&ftrace_lock);
4864
4865 return ret;
4866}
4867EXPORT_SYMBOL_GPL(unregister_ftrace_function);
4868
4869int
4870ftrace_enable_sysctl(struct ctl_table *table, int write,
4871 void __user *buffer, size_t *lenp,
4872 loff_t *ppos)
4873{
4874 int ret = -ENODEV;
4875
4876 mutex_lock(&ftrace_lock);
4877
4878 if (unlikely(ftrace_disabled))
4879 goto out;
4880
4881 ret = proc_dointvec(table, write, buffer, lenp, ppos);
4882
4883 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
4884 goto out;
4885
4886 last_ftrace_enabled = !!ftrace_enabled;
4887
4888 if (ftrace_enabled) {
4889
4890 ftrace_startup_sysctl();
4891
4892 /* we are starting ftrace again */
4893 if (ftrace_ops_list != &ftrace_list_end)
4894 update_ftrace_function();
4895
4896 } else {
4897 /* stopping ftrace calls (just send to ftrace_stub) */
4898 ftrace_trace_function = ftrace_stub;
4899
4900 ftrace_shutdown_sysctl();
4901 }
4902
4903 out:
4904 mutex_unlock(&ftrace_lock);
4905 return ret;
4906}
4907
4908#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4909
4910static int ftrace_graph_active;
4911static struct notifier_block ftrace_suspend_notifier;
4912
4913int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
4914{
4915 return 0;
4916}
4917
4918/* The callbacks that hook a function */
4919trace_func_graph_ret_t ftrace_graph_return =
4920 (trace_func_graph_ret_t)ftrace_stub;
4921trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
4922static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
4923
4924/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
4925static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
4926{
4927 int i;
4928 int ret = 0;
4929 unsigned long flags;
4930 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
4931 struct task_struct *g, *t;
4932
4933 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
4934 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
4935 * sizeof(struct ftrace_ret_stack),
4936 GFP_KERNEL);
4937 if (!ret_stack_list[i]) {
4938 start = 0;
4939 end = i;
4940 ret = -ENOMEM;
4941 goto free;
4942 }
4943 }
4944
4945 read_lock_irqsave(&tasklist_lock, flags);
4946 do_each_thread(g, t) {
4947 if (start == end) {
4948 ret = -EAGAIN;
4949 goto unlock;
4950 }
4951
4952 if (t->ret_stack == NULL) {
4953 atomic_set(&t->tracing_graph_pause, 0);
4954 atomic_set(&t->trace_overrun, 0);
4955 t->curr_ret_stack = -1;
4956 /* Make sure the tasks see the -1 first: */
4957 smp_wmb();
4958 t->ret_stack = ret_stack_list[start++];
4959 }
4960 } while_each_thread(g, t);
4961
4962unlock:
4963 read_unlock_irqrestore(&tasklist_lock, flags);
4964free:
4965 for (i = start; i < end; i++)
4966 kfree(ret_stack_list[i]);
4967 return ret;
4968}
4969
4970static void
4971ftrace_graph_probe_sched_switch(void *ignore,
4972 struct task_struct *prev, struct task_struct *next)
4973{
4974 unsigned long long timestamp;
4975 int index;
4976
4977 /*
4978 * Does the user want to count the time a function was asleep.
4979 * If so, do not update the time stamps.
4980 */
4981 if (trace_flags & TRACE_ITER_SLEEP_TIME)
4982 return;
4983
4984 timestamp = trace_clock_local();
4985
4986 prev->ftrace_timestamp = timestamp;
4987
4988 /* only process tasks that we timestamped */
4989 if (!next->ftrace_timestamp)
4990 return;
4991
4992 /*
4993 * Update all the counters in next to make up for the
4994 * time next was sleeping.
4995 */
4996 timestamp -= next->ftrace_timestamp;
4997
4998 for (index = next->curr_ret_stack; index >= 0; index--)
4999 next->ret_stack[index].calltime += timestamp;
5000}
5001
5002/* Allocate a return stack for each task */
5003static int start_graph_tracing(void)
5004{
5005 struct ftrace_ret_stack **ret_stack_list;
5006 int ret, cpu;
5007
5008 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
5009 sizeof(struct ftrace_ret_stack *),
5010 GFP_KERNEL);
5011
5012 if (!ret_stack_list)
5013 return -ENOMEM;
5014
5015 /* The cpu_boot init_task->ret_stack will never be freed */
5016 for_each_online_cpu(cpu) {
5017 if (!idle_task(cpu)->ret_stack)
5018 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
5019 }
5020
5021 do {
5022 ret = alloc_retstack_tasklist(ret_stack_list);
5023 } while (ret == -EAGAIN);
5024
5025 if (!ret) {
5026 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5027 if (ret)
5028 pr_info("ftrace_graph: Couldn't activate tracepoint"
5029 " probe to kernel_sched_switch\n");
5030 }
5031
5032 kfree(ret_stack_list);
5033 return ret;
5034}
5035
5036/*
5037 * Hibernation protection.
5038 * The state of the current task is too much unstable during
5039 * suspend/restore to disk. We want to protect against that.
5040 */
5041static int
5042ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
5043 void *unused)
5044{
5045 switch (state) {
5046 case PM_HIBERNATION_PREPARE:
5047 pause_graph_tracing();
5048 break;
5049
5050 case PM_POST_HIBERNATION:
5051 unpause_graph_tracing();
5052 break;
5053 }
5054 return NOTIFY_DONE;
5055}
5056
5057/* Just a place holder for function graph */
5058static struct ftrace_ops fgraph_ops __read_mostly = {
5059 .func = ftrace_stub,
5060 .flags = FTRACE_OPS_FL_STUB | FTRACE_OPS_FL_GLOBAL |
5061 FTRACE_OPS_FL_RECURSION_SAFE,
5062};
5063
5064static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
5065{
5066 if (!ftrace_ops_test(&global_ops, trace->func, NULL))
5067 return 0;
5068 return __ftrace_graph_entry(trace);
5069}
5070
5071/*
5072 * The function graph tracer should only trace the functions defined
5073 * by set_ftrace_filter and set_ftrace_notrace. If another function
5074 * tracer ops is registered, the graph tracer requires testing the
5075 * function against the global ops, and not just trace any function
5076 * that any ftrace_ops registered.
5077 */
5078static void update_function_graph_func(void)
5079{
5080 if (ftrace_ops_list == &ftrace_list_end ||
5081 (ftrace_ops_list == &global_ops &&
5082 global_ops.next == &ftrace_list_end))
5083 ftrace_graph_entry = __ftrace_graph_entry;
5084 else
5085 ftrace_graph_entry = ftrace_graph_entry_test;
5086}
5087
5088int register_ftrace_graph(trace_func_graph_ret_t retfunc,
5089 trace_func_graph_ent_t entryfunc)
5090{
5091 int ret = 0;
5092
5093 mutex_lock(&ftrace_lock);
5094
5095 /* we currently allow only one tracer registered at a time */
5096 if (ftrace_graph_active) {
5097 ret = -EBUSY;
5098 goto out;
5099 }
5100
5101 ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
5102 register_pm_notifier(&ftrace_suspend_notifier);
5103
5104 ftrace_graph_active++;
5105 ret = start_graph_tracing();
5106 if (ret) {
5107 ftrace_graph_active--;
5108 goto out;
5109 }
5110
5111 ftrace_graph_return = retfunc;
5112
5113 /*
5114 * Update the indirect function to the entryfunc, and the
5115 * function that gets called to the entry_test first. Then
5116 * call the update fgraph entry function to determine if
5117 * the entryfunc should be called directly or not.
5118 */
5119 __ftrace_graph_entry = entryfunc;
5120 ftrace_graph_entry = ftrace_graph_entry_test;
5121 update_function_graph_func();
5122
5123 ret = ftrace_startup(&fgraph_ops, FTRACE_START_FUNC_RET);
5124
5125out:
5126 mutex_unlock(&ftrace_lock);
5127 return ret;
5128}
5129
5130void unregister_ftrace_graph(void)
5131{
5132 mutex_lock(&ftrace_lock);
5133
5134 if (unlikely(!ftrace_graph_active))
5135 goto out;
5136
5137 ftrace_graph_active--;
5138 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
5139 ftrace_graph_entry = ftrace_graph_entry_stub;
5140 __ftrace_graph_entry = ftrace_graph_entry_stub;
5141 ftrace_shutdown(&fgraph_ops, FTRACE_STOP_FUNC_RET);
5142 unregister_pm_notifier(&ftrace_suspend_notifier);
5143 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5144
5145 out:
5146 mutex_unlock(&ftrace_lock);
5147}
5148
5149static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
5150
5151static void
5152graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
5153{
5154 atomic_set(&t->tracing_graph_pause, 0);
5155 atomic_set(&t->trace_overrun, 0);
5156 t->ftrace_timestamp = 0;
5157 /* make curr_ret_stack visible before we add the ret_stack */
5158 smp_wmb();
5159 t->ret_stack = ret_stack;
5160}
5161
5162/*
5163 * Allocate a return stack for the idle task. May be the first
5164 * time through, or it may be done by CPU hotplug online.
5165 */
5166void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
5167{
5168 t->curr_ret_stack = -1;
5169 /*
5170 * The idle task has no parent, it either has its own
5171 * stack or no stack at all.
5172 */
5173 if (t->ret_stack)
5174 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
5175
5176 if (ftrace_graph_active) {
5177 struct ftrace_ret_stack *ret_stack;
5178
5179 ret_stack = per_cpu(idle_ret_stack, cpu);
5180 if (!ret_stack) {
5181 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5182 * sizeof(struct ftrace_ret_stack),
5183 GFP_KERNEL);
5184 if (!ret_stack)
5185 return;
5186 per_cpu(idle_ret_stack, cpu) = ret_stack;
5187 }
5188 graph_init_task(t, ret_stack);
5189 }
5190}
5191
5192/* Allocate a return stack for newly created task */
5193void ftrace_graph_init_task(struct task_struct *t)
5194{
5195 /* Make sure we do not use the parent ret_stack */
5196 t->ret_stack = NULL;
5197 t->curr_ret_stack = -1;
5198
5199 if (ftrace_graph_active) {
5200 struct ftrace_ret_stack *ret_stack;
5201
5202 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5203 * sizeof(struct ftrace_ret_stack),
5204 GFP_KERNEL);
5205 if (!ret_stack)
5206 return;
5207 graph_init_task(t, ret_stack);
5208 }
5209}
5210
5211void ftrace_graph_exit_task(struct task_struct *t)
5212{
5213 struct ftrace_ret_stack *ret_stack = t->ret_stack;
5214
5215 t->ret_stack = NULL;
5216 /* NULL must become visible to IRQs before we free it: */
5217 barrier();
5218
5219 kfree(ret_stack);
5220}
5221
5222void ftrace_graph_stop(void)
5223{
5224 ftrace_stop();
5225}
5226#endif
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/ftrace.h>
26#include <linux/sysctl.h>
27#include <linux/slab.h>
28#include <linux/ctype.h>
29#include <linux/list.h>
30#include <linux/hash.h>
31#include <linux/rcupdate.h>
32
33#include <trace/events/sched.h>
34
35#include <asm/setup.h>
36
37#include "trace_output.h"
38#include "trace_stat.h"
39
40#define FTRACE_WARN_ON(cond) \
41 ({ \
42 int ___r = cond; \
43 if (WARN_ON(___r)) \
44 ftrace_kill(); \
45 ___r; \
46 })
47
48#define FTRACE_WARN_ON_ONCE(cond) \
49 ({ \
50 int ___r = cond; \
51 if (WARN_ON_ONCE(___r)) \
52 ftrace_kill(); \
53 ___r; \
54 })
55
56/* hash bits for specific function selection */
57#define FTRACE_HASH_BITS 7
58#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
59#define FTRACE_HASH_DEFAULT_BITS 10
60#define FTRACE_HASH_MAX_BITS 12
61
62/* ftrace_enabled is a method to turn ftrace on or off */
63int ftrace_enabled __read_mostly;
64static int last_ftrace_enabled;
65
66/* Quick disabling of function tracer. */
67int function_trace_stop;
68
69/* List for set_ftrace_pid's pids. */
70LIST_HEAD(ftrace_pids);
71struct ftrace_pid {
72 struct list_head list;
73 struct pid *pid;
74};
75
76/*
77 * ftrace_disabled is set when an anomaly is discovered.
78 * ftrace_disabled is much stronger than ftrace_enabled.
79 */
80static int ftrace_disabled __read_mostly;
81
82static DEFINE_MUTEX(ftrace_lock);
83
84static struct ftrace_ops ftrace_list_end __read_mostly = {
85 .func = ftrace_stub,
86};
87
88static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
89static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
90ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
91static ftrace_func_t __ftrace_trace_function_delay __read_mostly = ftrace_stub;
92ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
93ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
94static struct ftrace_ops global_ops;
95
96static void
97ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip);
98
99/*
100 * Traverse the ftrace_global_list, invoking all entries. The reason that we
101 * can use rcu_dereference_raw() is that elements removed from this list
102 * are simply leaked, so there is no need to interact with a grace-period
103 * mechanism. The rcu_dereference_raw() calls are needed to handle
104 * concurrent insertions into the ftrace_global_list.
105 *
106 * Silly Alpha and silly pointer-speculation compiler optimizations!
107 */
108static void ftrace_global_list_func(unsigned long ip,
109 unsigned long parent_ip)
110{
111 struct ftrace_ops *op;
112
113 if (unlikely(trace_recursion_test(TRACE_GLOBAL_BIT)))
114 return;
115
116 trace_recursion_set(TRACE_GLOBAL_BIT);
117 op = rcu_dereference_raw(ftrace_global_list); /*see above*/
118 while (op != &ftrace_list_end) {
119 op->func(ip, parent_ip);
120 op = rcu_dereference_raw(op->next); /*see above*/
121 };
122 trace_recursion_clear(TRACE_GLOBAL_BIT);
123}
124
125static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
126{
127 if (!test_tsk_trace_trace(current))
128 return;
129
130 ftrace_pid_function(ip, parent_ip);
131}
132
133static void set_ftrace_pid_function(ftrace_func_t func)
134{
135 /* do not set ftrace_pid_function to itself! */
136 if (func != ftrace_pid_func)
137 ftrace_pid_function = func;
138}
139
140/**
141 * clear_ftrace_function - reset the ftrace function
142 *
143 * This NULLs the ftrace function and in essence stops
144 * tracing. There may be lag
145 */
146void clear_ftrace_function(void)
147{
148 ftrace_trace_function = ftrace_stub;
149 __ftrace_trace_function = ftrace_stub;
150 __ftrace_trace_function_delay = ftrace_stub;
151 ftrace_pid_function = ftrace_stub;
152}
153
154#undef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
155#ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
156/*
157 * For those archs that do not test ftrace_trace_stop in their
158 * mcount call site, we need to do it from C.
159 */
160static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
161{
162 if (function_trace_stop)
163 return;
164
165 __ftrace_trace_function(ip, parent_ip);
166}
167#endif
168
169static void update_global_ops(void)
170{
171 ftrace_func_t func;
172
173 /*
174 * If there's only one function registered, then call that
175 * function directly. Otherwise, we need to iterate over the
176 * registered callers.
177 */
178 if (ftrace_global_list == &ftrace_list_end ||
179 ftrace_global_list->next == &ftrace_list_end)
180 func = ftrace_global_list->func;
181 else
182 func = ftrace_global_list_func;
183
184 /* If we filter on pids, update to use the pid function */
185 if (!list_empty(&ftrace_pids)) {
186 set_ftrace_pid_function(func);
187 func = ftrace_pid_func;
188 }
189
190 global_ops.func = func;
191}
192
193static void update_ftrace_function(void)
194{
195 ftrace_func_t func;
196
197 update_global_ops();
198
199 /*
200 * If we are at the end of the list and this ops is
201 * not dynamic, then have the mcount trampoline call
202 * the function directly
203 */
204 if (ftrace_ops_list == &ftrace_list_end ||
205 (ftrace_ops_list->next == &ftrace_list_end &&
206 !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC)))
207 func = ftrace_ops_list->func;
208 else
209 func = ftrace_ops_list_func;
210
211#ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
212 ftrace_trace_function = func;
213#else
214#ifdef CONFIG_DYNAMIC_FTRACE
215 /* do not update till all functions have been modified */
216 __ftrace_trace_function_delay = func;
217#else
218 __ftrace_trace_function = func;
219#endif
220 ftrace_trace_function = ftrace_test_stop_func;
221#endif
222}
223
224static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
225{
226 ops->next = *list;
227 /*
228 * We are entering ops into the list but another
229 * CPU might be walking that list. We need to make sure
230 * the ops->next pointer is valid before another CPU sees
231 * the ops pointer included into the list.
232 */
233 rcu_assign_pointer(*list, ops);
234}
235
236static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
237{
238 struct ftrace_ops **p;
239
240 /*
241 * If we are removing the last function, then simply point
242 * to the ftrace_stub.
243 */
244 if (*list == ops && ops->next == &ftrace_list_end) {
245 *list = &ftrace_list_end;
246 return 0;
247 }
248
249 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
250 if (*p == ops)
251 break;
252
253 if (*p != ops)
254 return -1;
255
256 *p = (*p)->next;
257 return 0;
258}
259
260static int __register_ftrace_function(struct ftrace_ops *ops)
261{
262 if (ftrace_disabled)
263 return -ENODEV;
264
265 if (FTRACE_WARN_ON(ops == &global_ops))
266 return -EINVAL;
267
268 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
269 return -EBUSY;
270
271 if (!core_kernel_data((unsigned long)ops))
272 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
273
274 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
275 int first = ftrace_global_list == &ftrace_list_end;
276 add_ftrace_ops(&ftrace_global_list, ops);
277 ops->flags |= FTRACE_OPS_FL_ENABLED;
278 if (first)
279 add_ftrace_ops(&ftrace_ops_list, &global_ops);
280 } else
281 add_ftrace_ops(&ftrace_ops_list, ops);
282
283 if (ftrace_enabled)
284 update_ftrace_function();
285
286 return 0;
287}
288
289static int __unregister_ftrace_function(struct ftrace_ops *ops)
290{
291 int ret;
292
293 if (ftrace_disabled)
294 return -ENODEV;
295
296 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
297 return -EBUSY;
298
299 if (FTRACE_WARN_ON(ops == &global_ops))
300 return -EINVAL;
301
302 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
303 ret = remove_ftrace_ops(&ftrace_global_list, ops);
304 if (!ret && ftrace_global_list == &ftrace_list_end)
305 ret = remove_ftrace_ops(&ftrace_ops_list, &global_ops);
306 if (!ret)
307 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
308 } else
309 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
310
311 if (ret < 0)
312 return ret;
313
314 if (ftrace_enabled)
315 update_ftrace_function();
316
317 /*
318 * Dynamic ops may be freed, we must make sure that all
319 * callers are done before leaving this function.
320 */
321 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
322 synchronize_sched();
323
324 return 0;
325}
326
327static void ftrace_update_pid_func(void)
328{
329 /* Only do something if we are tracing something */
330 if (ftrace_trace_function == ftrace_stub)
331 return;
332
333 update_ftrace_function();
334}
335
336#ifdef CONFIG_FUNCTION_PROFILER
337struct ftrace_profile {
338 struct hlist_node node;
339 unsigned long ip;
340 unsigned long counter;
341#ifdef CONFIG_FUNCTION_GRAPH_TRACER
342 unsigned long long time;
343 unsigned long long time_squared;
344#endif
345};
346
347struct ftrace_profile_page {
348 struct ftrace_profile_page *next;
349 unsigned long index;
350 struct ftrace_profile records[];
351};
352
353struct ftrace_profile_stat {
354 atomic_t disabled;
355 struct hlist_head *hash;
356 struct ftrace_profile_page *pages;
357 struct ftrace_profile_page *start;
358 struct tracer_stat stat;
359};
360
361#define PROFILE_RECORDS_SIZE \
362 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
363
364#define PROFILES_PER_PAGE \
365 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
366
367static int ftrace_profile_bits __read_mostly;
368static int ftrace_profile_enabled __read_mostly;
369
370/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
371static DEFINE_MUTEX(ftrace_profile_lock);
372
373static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
374
375#define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
376
377static void *
378function_stat_next(void *v, int idx)
379{
380 struct ftrace_profile *rec = v;
381 struct ftrace_profile_page *pg;
382
383 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
384
385 again:
386 if (idx != 0)
387 rec++;
388
389 if ((void *)rec >= (void *)&pg->records[pg->index]) {
390 pg = pg->next;
391 if (!pg)
392 return NULL;
393 rec = &pg->records[0];
394 if (!rec->counter)
395 goto again;
396 }
397
398 return rec;
399}
400
401static void *function_stat_start(struct tracer_stat *trace)
402{
403 struct ftrace_profile_stat *stat =
404 container_of(trace, struct ftrace_profile_stat, stat);
405
406 if (!stat || !stat->start)
407 return NULL;
408
409 return function_stat_next(&stat->start->records[0], 0);
410}
411
412#ifdef CONFIG_FUNCTION_GRAPH_TRACER
413/* function graph compares on total time */
414static int function_stat_cmp(void *p1, void *p2)
415{
416 struct ftrace_profile *a = p1;
417 struct ftrace_profile *b = p2;
418
419 if (a->time < b->time)
420 return -1;
421 if (a->time > b->time)
422 return 1;
423 else
424 return 0;
425}
426#else
427/* not function graph compares against hits */
428static int function_stat_cmp(void *p1, void *p2)
429{
430 struct ftrace_profile *a = p1;
431 struct ftrace_profile *b = p2;
432
433 if (a->counter < b->counter)
434 return -1;
435 if (a->counter > b->counter)
436 return 1;
437 else
438 return 0;
439}
440#endif
441
442static int function_stat_headers(struct seq_file *m)
443{
444#ifdef CONFIG_FUNCTION_GRAPH_TRACER
445 seq_printf(m, " Function "
446 "Hit Time Avg s^2\n"
447 " -------- "
448 "--- ---- --- ---\n");
449#else
450 seq_printf(m, " Function Hit\n"
451 " -------- ---\n");
452#endif
453 return 0;
454}
455
456static int function_stat_show(struct seq_file *m, void *v)
457{
458 struct ftrace_profile *rec = v;
459 char str[KSYM_SYMBOL_LEN];
460 int ret = 0;
461#ifdef CONFIG_FUNCTION_GRAPH_TRACER
462 static struct trace_seq s;
463 unsigned long long avg;
464 unsigned long long stddev;
465#endif
466 mutex_lock(&ftrace_profile_lock);
467
468 /* we raced with function_profile_reset() */
469 if (unlikely(rec->counter == 0)) {
470 ret = -EBUSY;
471 goto out;
472 }
473
474 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
475 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
476
477#ifdef CONFIG_FUNCTION_GRAPH_TRACER
478 seq_printf(m, " ");
479 avg = rec->time;
480 do_div(avg, rec->counter);
481
482 /* Sample standard deviation (s^2) */
483 if (rec->counter <= 1)
484 stddev = 0;
485 else {
486 stddev = rec->time_squared - rec->counter * avg * avg;
487 /*
488 * Divide only 1000 for ns^2 -> us^2 conversion.
489 * trace_print_graph_duration will divide 1000 again.
490 */
491 do_div(stddev, (rec->counter - 1) * 1000);
492 }
493
494 trace_seq_init(&s);
495 trace_print_graph_duration(rec->time, &s);
496 trace_seq_puts(&s, " ");
497 trace_print_graph_duration(avg, &s);
498 trace_seq_puts(&s, " ");
499 trace_print_graph_duration(stddev, &s);
500 trace_print_seq(m, &s);
501#endif
502 seq_putc(m, '\n');
503out:
504 mutex_unlock(&ftrace_profile_lock);
505
506 return ret;
507}
508
509static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
510{
511 struct ftrace_profile_page *pg;
512
513 pg = stat->pages = stat->start;
514
515 while (pg) {
516 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
517 pg->index = 0;
518 pg = pg->next;
519 }
520
521 memset(stat->hash, 0,
522 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
523}
524
525int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
526{
527 struct ftrace_profile_page *pg;
528 int functions;
529 int pages;
530 int i;
531
532 /* If we already allocated, do nothing */
533 if (stat->pages)
534 return 0;
535
536 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
537 if (!stat->pages)
538 return -ENOMEM;
539
540#ifdef CONFIG_DYNAMIC_FTRACE
541 functions = ftrace_update_tot_cnt;
542#else
543 /*
544 * We do not know the number of functions that exist because
545 * dynamic tracing is what counts them. With past experience
546 * we have around 20K functions. That should be more than enough.
547 * It is highly unlikely we will execute every function in
548 * the kernel.
549 */
550 functions = 20000;
551#endif
552
553 pg = stat->start = stat->pages;
554
555 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
556
557 for (i = 0; i < pages; i++) {
558 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
559 if (!pg->next)
560 goto out_free;
561 pg = pg->next;
562 }
563
564 return 0;
565
566 out_free:
567 pg = stat->start;
568 while (pg) {
569 unsigned long tmp = (unsigned long)pg;
570
571 pg = pg->next;
572 free_page(tmp);
573 }
574
575 free_page((unsigned long)stat->pages);
576 stat->pages = NULL;
577 stat->start = NULL;
578
579 return -ENOMEM;
580}
581
582static int ftrace_profile_init_cpu(int cpu)
583{
584 struct ftrace_profile_stat *stat;
585 int size;
586
587 stat = &per_cpu(ftrace_profile_stats, cpu);
588
589 if (stat->hash) {
590 /* If the profile is already created, simply reset it */
591 ftrace_profile_reset(stat);
592 return 0;
593 }
594
595 /*
596 * We are profiling all functions, but usually only a few thousand
597 * functions are hit. We'll make a hash of 1024 items.
598 */
599 size = FTRACE_PROFILE_HASH_SIZE;
600
601 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
602
603 if (!stat->hash)
604 return -ENOMEM;
605
606 if (!ftrace_profile_bits) {
607 size--;
608
609 for (; size; size >>= 1)
610 ftrace_profile_bits++;
611 }
612
613 /* Preallocate the function profiling pages */
614 if (ftrace_profile_pages_init(stat) < 0) {
615 kfree(stat->hash);
616 stat->hash = NULL;
617 return -ENOMEM;
618 }
619
620 return 0;
621}
622
623static int ftrace_profile_init(void)
624{
625 int cpu;
626 int ret = 0;
627
628 for_each_online_cpu(cpu) {
629 ret = ftrace_profile_init_cpu(cpu);
630 if (ret)
631 break;
632 }
633
634 return ret;
635}
636
637/* interrupts must be disabled */
638static struct ftrace_profile *
639ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
640{
641 struct ftrace_profile *rec;
642 struct hlist_head *hhd;
643 struct hlist_node *n;
644 unsigned long key;
645
646 key = hash_long(ip, ftrace_profile_bits);
647 hhd = &stat->hash[key];
648
649 if (hlist_empty(hhd))
650 return NULL;
651
652 hlist_for_each_entry_rcu(rec, n, hhd, node) {
653 if (rec->ip == ip)
654 return rec;
655 }
656
657 return NULL;
658}
659
660static void ftrace_add_profile(struct ftrace_profile_stat *stat,
661 struct ftrace_profile *rec)
662{
663 unsigned long key;
664
665 key = hash_long(rec->ip, ftrace_profile_bits);
666 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
667}
668
669/*
670 * The memory is already allocated, this simply finds a new record to use.
671 */
672static struct ftrace_profile *
673ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
674{
675 struct ftrace_profile *rec = NULL;
676
677 /* prevent recursion (from NMIs) */
678 if (atomic_inc_return(&stat->disabled) != 1)
679 goto out;
680
681 /*
682 * Try to find the function again since an NMI
683 * could have added it
684 */
685 rec = ftrace_find_profiled_func(stat, ip);
686 if (rec)
687 goto out;
688
689 if (stat->pages->index == PROFILES_PER_PAGE) {
690 if (!stat->pages->next)
691 goto out;
692 stat->pages = stat->pages->next;
693 }
694
695 rec = &stat->pages->records[stat->pages->index++];
696 rec->ip = ip;
697 ftrace_add_profile(stat, rec);
698
699 out:
700 atomic_dec(&stat->disabled);
701
702 return rec;
703}
704
705static void
706function_profile_call(unsigned long ip, unsigned long parent_ip)
707{
708 struct ftrace_profile_stat *stat;
709 struct ftrace_profile *rec;
710 unsigned long flags;
711
712 if (!ftrace_profile_enabled)
713 return;
714
715 local_irq_save(flags);
716
717 stat = &__get_cpu_var(ftrace_profile_stats);
718 if (!stat->hash || !ftrace_profile_enabled)
719 goto out;
720
721 rec = ftrace_find_profiled_func(stat, ip);
722 if (!rec) {
723 rec = ftrace_profile_alloc(stat, ip);
724 if (!rec)
725 goto out;
726 }
727
728 rec->counter++;
729 out:
730 local_irq_restore(flags);
731}
732
733#ifdef CONFIG_FUNCTION_GRAPH_TRACER
734static int profile_graph_entry(struct ftrace_graph_ent *trace)
735{
736 function_profile_call(trace->func, 0);
737 return 1;
738}
739
740static void profile_graph_return(struct ftrace_graph_ret *trace)
741{
742 struct ftrace_profile_stat *stat;
743 unsigned long long calltime;
744 struct ftrace_profile *rec;
745 unsigned long flags;
746
747 local_irq_save(flags);
748 stat = &__get_cpu_var(ftrace_profile_stats);
749 if (!stat->hash || !ftrace_profile_enabled)
750 goto out;
751
752 /* If the calltime was zero'd ignore it */
753 if (!trace->calltime)
754 goto out;
755
756 calltime = trace->rettime - trace->calltime;
757
758 if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
759 int index;
760
761 index = trace->depth;
762
763 /* Append this call time to the parent time to subtract */
764 if (index)
765 current->ret_stack[index - 1].subtime += calltime;
766
767 if (current->ret_stack[index].subtime < calltime)
768 calltime -= current->ret_stack[index].subtime;
769 else
770 calltime = 0;
771 }
772
773 rec = ftrace_find_profiled_func(stat, trace->func);
774 if (rec) {
775 rec->time += calltime;
776 rec->time_squared += calltime * calltime;
777 }
778
779 out:
780 local_irq_restore(flags);
781}
782
783static int register_ftrace_profiler(void)
784{
785 return register_ftrace_graph(&profile_graph_return,
786 &profile_graph_entry);
787}
788
789static void unregister_ftrace_profiler(void)
790{
791 unregister_ftrace_graph();
792}
793#else
794static struct ftrace_ops ftrace_profile_ops __read_mostly = {
795 .func = function_profile_call,
796};
797
798static int register_ftrace_profiler(void)
799{
800 return register_ftrace_function(&ftrace_profile_ops);
801}
802
803static void unregister_ftrace_profiler(void)
804{
805 unregister_ftrace_function(&ftrace_profile_ops);
806}
807#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
808
809static ssize_t
810ftrace_profile_write(struct file *filp, const char __user *ubuf,
811 size_t cnt, loff_t *ppos)
812{
813 unsigned long val;
814 int ret;
815
816 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
817 if (ret)
818 return ret;
819
820 val = !!val;
821
822 mutex_lock(&ftrace_profile_lock);
823 if (ftrace_profile_enabled ^ val) {
824 if (val) {
825 ret = ftrace_profile_init();
826 if (ret < 0) {
827 cnt = ret;
828 goto out;
829 }
830
831 ret = register_ftrace_profiler();
832 if (ret < 0) {
833 cnt = ret;
834 goto out;
835 }
836 ftrace_profile_enabled = 1;
837 } else {
838 ftrace_profile_enabled = 0;
839 /*
840 * unregister_ftrace_profiler calls stop_machine
841 * so this acts like an synchronize_sched.
842 */
843 unregister_ftrace_profiler();
844 }
845 }
846 out:
847 mutex_unlock(&ftrace_profile_lock);
848
849 *ppos += cnt;
850
851 return cnt;
852}
853
854static ssize_t
855ftrace_profile_read(struct file *filp, char __user *ubuf,
856 size_t cnt, loff_t *ppos)
857{
858 char buf[64]; /* big enough to hold a number */
859 int r;
860
861 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
862 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
863}
864
865static const struct file_operations ftrace_profile_fops = {
866 .open = tracing_open_generic,
867 .read = ftrace_profile_read,
868 .write = ftrace_profile_write,
869 .llseek = default_llseek,
870};
871
872/* used to initialize the real stat files */
873static struct tracer_stat function_stats __initdata = {
874 .name = "functions",
875 .stat_start = function_stat_start,
876 .stat_next = function_stat_next,
877 .stat_cmp = function_stat_cmp,
878 .stat_headers = function_stat_headers,
879 .stat_show = function_stat_show
880};
881
882static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
883{
884 struct ftrace_profile_stat *stat;
885 struct dentry *entry;
886 char *name;
887 int ret;
888 int cpu;
889
890 for_each_possible_cpu(cpu) {
891 stat = &per_cpu(ftrace_profile_stats, cpu);
892
893 /* allocate enough for function name + cpu number */
894 name = kmalloc(32, GFP_KERNEL);
895 if (!name) {
896 /*
897 * The files created are permanent, if something happens
898 * we still do not free memory.
899 */
900 WARN(1,
901 "Could not allocate stat file for cpu %d\n",
902 cpu);
903 return;
904 }
905 stat->stat = function_stats;
906 snprintf(name, 32, "function%d", cpu);
907 stat->stat.name = name;
908 ret = register_stat_tracer(&stat->stat);
909 if (ret) {
910 WARN(1,
911 "Could not register function stat for cpu %d\n",
912 cpu);
913 kfree(name);
914 return;
915 }
916 }
917
918 entry = debugfs_create_file("function_profile_enabled", 0644,
919 d_tracer, NULL, &ftrace_profile_fops);
920 if (!entry)
921 pr_warning("Could not create debugfs "
922 "'function_profile_enabled' entry\n");
923}
924
925#else /* CONFIG_FUNCTION_PROFILER */
926static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
927{
928}
929#endif /* CONFIG_FUNCTION_PROFILER */
930
931static struct pid * const ftrace_swapper_pid = &init_struct_pid;
932
933#ifdef CONFIG_DYNAMIC_FTRACE
934
935#ifndef CONFIG_FTRACE_MCOUNT_RECORD
936# error Dynamic ftrace depends on MCOUNT_RECORD
937#endif
938
939static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
940
941struct ftrace_func_probe {
942 struct hlist_node node;
943 struct ftrace_probe_ops *ops;
944 unsigned long flags;
945 unsigned long ip;
946 void *data;
947 struct rcu_head rcu;
948};
949
950enum {
951 FTRACE_ENABLE_CALLS = (1 << 0),
952 FTRACE_DISABLE_CALLS = (1 << 1),
953 FTRACE_UPDATE_TRACE_FUNC = (1 << 2),
954 FTRACE_START_FUNC_RET = (1 << 3),
955 FTRACE_STOP_FUNC_RET = (1 << 4),
956};
957struct ftrace_func_entry {
958 struct hlist_node hlist;
959 unsigned long ip;
960};
961
962struct ftrace_hash {
963 unsigned long size_bits;
964 struct hlist_head *buckets;
965 unsigned long count;
966 struct rcu_head rcu;
967};
968
969/*
970 * We make these constant because no one should touch them,
971 * but they are used as the default "empty hash", to avoid allocating
972 * it all the time. These are in a read only section such that if
973 * anyone does try to modify it, it will cause an exception.
974 */
975static const struct hlist_head empty_buckets[1];
976static const struct ftrace_hash empty_hash = {
977 .buckets = (struct hlist_head *)empty_buckets,
978};
979#define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
980
981static struct ftrace_ops global_ops = {
982 .func = ftrace_stub,
983 .notrace_hash = EMPTY_HASH,
984 .filter_hash = EMPTY_HASH,
985};
986
987static struct dyn_ftrace *ftrace_new_addrs;
988
989static DEFINE_MUTEX(ftrace_regex_lock);
990
991struct ftrace_page {
992 struct ftrace_page *next;
993 int index;
994 struct dyn_ftrace records[];
995};
996
997#define ENTRIES_PER_PAGE \
998 ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
999
1000/* estimate from running different kernels */
1001#define NR_TO_INIT 10000
1002
1003static struct ftrace_page *ftrace_pages_start;
1004static struct ftrace_page *ftrace_pages;
1005
1006static struct dyn_ftrace *ftrace_free_records;
1007
1008static struct ftrace_func_entry *
1009ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1010{
1011 unsigned long key;
1012 struct ftrace_func_entry *entry;
1013 struct hlist_head *hhd;
1014 struct hlist_node *n;
1015
1016 if (!hash->count)
1017 return NULL;
1018
1019 if (hash->size_bits > 0)
1020 key = hash_long(ip, hash->size_bits);
1021 else
1022 key = 0;
1023
1024 hhd = &hash->buckets[key];
1025
1026 hlist_for_each_entry_rcu(entry, n, hhd, hlist) {
1027 if (entry->ip == ip)
1028 return entry;
1029 }
1030 return NULL;
1031}
1032
1033static void __add_hash_entry(struct ftrace_hash *hash,
1034 struct ftrace_func_entry *entry)
1035{
1036 struct hlist_head *hhd;
1037 unsigned long key;
1038
1039 if (hash->size_bits)
1040 key = hash_long(entry->ip, hash->size_bits);
1041 else
1042 key = 0;
1043
1044 hhd = &hash->buckets[key];
1045 hlist_add_head(&entry->hlist, hhd);
1046 hash->count++;
1047}
1048
1049static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1050{
1051 struct ftrace_func_entry *entry;
1052
1053 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1054 if (!entry)
1055 return -ENOMEM;
1056
1057 entry->ip = ip;
1058 __add_hash_entry(hash, entry);
1059
1060 return 0;
1061}
1062
1063static void
1064free_hash_entry(struct ftrace_hash *hash,
1065 struct ftrace_func_entry *entry)
1066{
1067 hlist_del(&entry->hlist);
1068 kfree(entry);
1069 hash->count--;
1070}
1071
1072static void
1073remove_hash_entry(struct ftrace_hash *hash,
1074 struct ftrace_func_entry *entry)
1075{
1076 hlist_del(&entry->hlist);
1077 hash->count--;
1078}
1079
1080static void ftrace_hash_clear(struct ftrace_hash *hash)
1081{
1082 struct hlist_head *hhd;
1083 struct hlist_node *tp, *tn;
1084 struct ftrace_func_entry *entry;
1085 int size = 1 << hash->size_bits;
1086 int i;
1087
1088 if (!hash->count)
1089 return;
1090
1091 for (i = 0; i < size; i++) {
1092 hhd = &hash->buckets[i];
1093 hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist)
1094 free_hash_entry(hash, entry);
1095 }
1096 FTRACE_WARN_ON(hash->count);
1097}
1098
1099static void free_ftrace_hash(struct ftrace_hash *hash)
1100{
1101 if (!hash || hash == EMPTY_HASH)
1102 return;
1103 ftrace_hash_clear(hash);
1104 kfree(hash->buckets);
1105 kfree(hash);
1106}
1107
1108static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1109{
1110 struct ftrace_hash *hash;
1111
1112 hash = container_of(rcu, struct ftrace_hash, rcu);
1113 free_ftrace_hash(hash);
1114}
1115
1116static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1117{
1118 if (!hash || hash == EMPTY_HASH)
1119 return;
1120 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1121}
1122
1123static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1124{
1125 struct ftrace_hash *hash;
1126 int size;
1127
1128 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1129 if (!hash)
1130 return NULL;
1131
1132 size = 1 << size_bits;
1133 hash->buckets = kzalloc(sizeof(*hash->buckets) * size, GFP_KERNEL);
1134
1135 if (!hash->buckets) {
1136 kfree(hash);
1137 return NULL;
1138 }
1139
1140 hash->size_bits = size_bits;
1141
1142 return hash;
1143}
1144
1145static struct ftrace_hash *
1146alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1147{
1148 struct ftrace_func_entry *entry;
1149 struct ftrace_hash *new_hash;
1150 struct hlist_node *tp;
1151 int size;
1152 int ret;
1153 int i;
1154
1155 new_hash = alloc_ftrace_hash(size_bits);
1156 if (!new_hash)
1157 return NULL;
1158
1159 /* Empty hash? */
1160 if (!hash || !hash->count)
1161 return new_hash;
1162
1163 size = 1 << hash->size_bits;
1164 for (i = 0; i < size; i++) {
1165 hlist_for_each_entry(entry, tp, &hash->buckets[i], hlist) {
1166 ret = add_hash_entry(new_hash, entry->ip);
1167 if (ret < 0)
1168 goto free_hash;
1169 }
1170 }
1171
1172 FTRACE_WARN_ON(new_hash->count != hash->count);
1173
1174 return new_hash;
1175
1176 free_hash:
1177 free_ftrace_hash(new_hash);
1178 return NULL;
1179}
1180
1181static void
1182ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1183static void
1184ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1185
1186static int
1187ftrace_hash_move(struct ftrace_ops *ops, int enable,
1188 struct ftrace_hash **dst, struct ftrace_hash *src)
1189{
1190 struct ftrace_func_entry *entry;
1191 struct hlist_node *tp, *tn;
1192 struct hlist_head *hhd;
1193 struct ftrace_hash *old_hash;
1194 struct ftrace_hash *new_hash;
1195 unsigned long key;
1196 int size = src->count;
1197 int bits = 0;
1198 int ret;
1199 int i;
1200
1201 /*
1202 * Remove the current set, update the hash and add
1203 * them back.
1204 */
1205 ftrace_hash_rec_disable(ops, enable);
1206
1207 /*
1208 * If the new source is empty, just free dst and assign it
1209 * the empty_hash.
1210 */
1211 if (!src->count) {
1212 free_ftrace_hash_rcu(*dst);
1213 rcu_assign_pointer(*dst, EMPTY_HASH);
1214 return 0;
1215 }
1216
1217 /*
1218 * Make the hash size about 1/2 the # found
1219 */
1220 for (size /= 2; size; size >>= 1)
1221 bits++;
1222
1223 /* Don't allocate too much */
1224 if (bits > FTRACE_HASH_MAX_BITS)
1225 bits = FTRACE_HASH_MAX_BITS;
1226
1227 ret = -ENOMEM;
1228 new_hash = alloc_ftrace_hash(bits);
1229 if (!new_hash)
1230 goto out;
1231
1232 size = 1 << src->size_bits;
1233 for (i = 0; i < size; i++) {
1234 hhd = &src->buckets[i];
1235 hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist) {
1236 if (bits > 0)
1237 key = hash_long(entry->ip, bits);
1238 else
1239 key = 0;
1240 remove_hash_entry(src, entry);
1241 __add_hash_entry(new_hash, entry);
1242 }
1243 }
1244
1245 old_hash = *dst;
1246 rcu_assign_pointer(*dst, new_hash);
1247 free_ftrace_hash_rcu(old_hash);
1248
1249 ret = 0;
1250 out:
1251 /*
1252 * Enable regardless of ret:
1253 * On success, we enable the new hash.
1254 * On failure, we re-enable the original hash.
1255 */
1256 ftrace_hash_rec_enable(ops, enable);
1257
1258 return ret;
1259}
1260
1261/*
1262 * Test the hashes for this ops to see if we want to call
1263 * the ops->func or not.
1264 *
1265 * It's a match if the ip is in the ops->filter_hash or
1266 * the filter_hash does not exist or is empty,
1267 * AND
1268 * the ip is not in the ops->notrace_hash.
1269 *
1270 * This needs to be called with preemption disabled as
1271 * the hashes are freed with call_rcu_sched().
1272 */
1273static int
1274ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
1275{
1276 struct ftrace_hash *filter_hash;
1277 struct ftrace_hash *notrace_hash;
1278 int ret;
1279
1280 filter_hash = rcu_dereference_raw(ops->filter_hash);
1281 notrace_hash = rcu_dereference_raw(ops->notrace_hash);
1282
1283 if ((!filter_hash || !filter_hash->count ||
1284 ftrace_lookup_ip(filter_hash, ip)) &&
1285 (!notrace_hash || !notrace_hash->count ||
1286 !ftrace_lookup_ip(notrace_hash, ip)))
1287 ret = 1;
1288 else
1289 ret = 0;
1290
1291 return ret;
1292}
1293
1294/*
1295 * This is a double for. Do not use 'break' to break out of the loop,
1296 * you must use a goto.
1297 */
1298#define do_for_each_ftrace_rec(pg, rec) \
1299 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1300 int _____i; \
1301 for (_____i = 0; _____i < pg->index; _____i++) { \
1302 rec = &pg->records[_____i];
1303
1304#define while_for_each_ftrace_rec() \
1305 } \
1306 }
1307
1308static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1309 int filter_hash,
1310 bool inc)
1311{
1312 struct ftrace_hash *hash;
1313 struct ftrace_hash *other_hash;
1314 struct ftrace_page *pg;
1315 struct dyn_ftrace *rec;
1316 int count = 0;
1317 int all = 0;
1318
1319 /* Only update if the ops has been registered */
1320 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1321 return;
1322
1323 /*
1324 * In the filter_hash case:
1325 * If the count is zero, we update all records.
1326 * Otherwise we just update the items in the hash.
1327 *
1328 * In the notrace_hash case:
1329 * We enable the update in the hash.
1330 * As disabling notrace means enabling the tracing,
1331 * and enabling notrace means disabling, the inc variable
1332 * gets inversed.
1333 */
1334 if (filter_hash) {
1335 hash = ops->filter_hash;
1336 other_hash = ops->notrace_hash;
1337 if (!hash || !hash->count)
1338 all = 1;
1339 } else {
1340 inc = !inc;
1341 hash = ops->notrace_hash;
1342 other_hash = ops->filter_hash;
1343 /*
1344 * If the notrace hash has no items,
1345 * then there's nothing to do.
1346 */
1347 if (hash && !hash->count)
1348 return;
1349 }
1350
1351 do_for_each_ftrace_rec(pg, rec) {
1352 int in_other_hash = 0;
1353 int in_hash = 0;
1354 int match = 0;
1355
1356 if (all) {
1357 /*
1358 * Only the filter_hash affects all records.
1359 * Update if the record is not in the notrace hash.
1360 */
1361 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1362 match = 1;
1363 } else {
1364 in_hash = hash && !!ftrace_lookup_ip(hash, rec->ip);
1365 in_other_hash = other_hash && !!ftrace_lookup_ip(other_hash, rec->ip);
1366
1367 /*
1368 *
1369 */
1370 if (filter_hash && in_hash && !in_other_hash)
1371 match = 1;
1372 else if (!filter_hash && in_hash &&
1373 (in_other_hash || !other_hash->count))
1374 match = 1;
1375 }
1376 if (!match)
1377 continue;
1378
1379 if (inc) {
1380 rec->flags++;
1381 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1382 return;
1383 } else {
1384 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1385 return;
1386 rec->flags--;
1387 }
1388 count++;
1389 /* Shortcut, if we handled all records, we are done. */
1390 if (!all && count == hash->count)
1391 return;
1392 } while_for_each_ftrace_rec();
1393}
1394
1395static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1396 int filter_hash)
1397{
1398 __ftrace_hash_rec_update(ops, filter_hash, 0);
1399}
1400
1401static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1402 int filter_hash)
1403{
1404 __ftrace_hash_rec_update(ops, filter_hash, 1);
1405}
1406
1407static void ftrace_free_rec(struct dyn_ftrace *rec)
1408{
1409 rec->freelist = ftrace_free_records;
1410 ftrace_free_records = rec;
1411 rec->flags |= FTRACE_FL_FREE;
1412}
1413
1414static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
1415{
1416 struct dyn_ftrace *rec;
1417
1418 /* First check for freed records */
1419 if (ftrace_free_records) {
1420 rec = ftrace_free_records;
1421
1422 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
1423 FTRACE_WARN_ON_ONCE(1);
1424 ftrace_free_records = NULL;
1425 return NULL;
1426 }
1427
1428 ftrace_free_records = rec->freelist;
1429 memset(rec, 0, sizeof(*rec));
1430 return rec;
1431 }
1432
1433 if (ftrace_pages->index == ENTRIES_PER_PAGE) {
1434 if (!ftrace_pages->next) {
1435 /* allocate another page */
1436 ftrace_pages->next =
1437 (void *)get_zeroed_page(GFP_KERNEL);
1438 if (!ftrace_pages->next)
1439 return NULL;
1440 }
1441 ftrace_pages = ftrace_pages->next;
1442 }
1443
1444 return &ftrace_pages->records[ftrace_pages->index++];
1445}
1446
1447static struct dyn_ftrace *
1448ftrace_record_ip(unsigned long ip)
1449{
1450 struct dyn_ftrace *rec;
1451
1452 if (ftrace_disabled)
1453 return NULL;
1454
1455 rec = ftrace_alloc_dyn_node(ip);
1456 if (!rec)
1457 return NULL;
1458
1459 rec->ip = ip;
1460 rec->newlist = ftrace_new_addrs;
1461 ftrace_new_addrs = rec;
1462
1463 return rec;
1464}
1465
1466static void print_ip_ins(const char *fmt, unsigned char *p)
1467{
1468 int i;
1469
1470 printk(KERN_CONT "%s", fmt);
1471
1472 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1473 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1474}
1475
1476static void ftrace_bug(int failed, unsigned long ip)
1477{
1478 switch (failed) {
1479 case -EFAULT:
1480 FTRACE_WARN_ON_ONCE(1);
1481 pr_info("ftrace faulted on modifying ");
1482 print_ip_sym(ip);
1483 break;
1484 case -EINVAL:
1485 FTRACE_WARN_ON_ONCE(1);
1486 pr_info("ftrace failed to modify ");
1487 print_ip_sym(ip);
1488 print_ip_ins(" actual: ", (unsigned char *)ip);
1489 printk(KERN_CONT "\n");
1490 break;
1491 case -EPERM:
1492 FTRACE_WARN_ON_ONCE(1);
1493 pr_info("ftrace faulted on writing ");
1494 print_ip_sym(ip);
1495 break;
1496 default:
1497 FTRACE_WARN_ON_ONCE(1);
1498 pr_info("ftrace faulted on unknown error ");
1499 print_ip_sym(ip);
1500 }
1501}
1502
1503
1504/* Return 1 if the address range is reserved for ftrace */
1505int ftrace_text_reserved(void *start, void *end)
1506{
1507 struct dyn_ftrace *rec;
1508 struct ftrace_page *pg;
1509
1510 do_for_each_ftrace_rec(pg, rec) {
1511 if (rec->ip <= (unsigned long)end &&
1512 rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1513 return 1;
1514 } while_for_each_ftrace_rec();
1515 return 0;
1516}
1517
1518
1519static int
1520__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1521{
1522 unsigned long ftrace_addr;
1523 unsigned long flag = 0UL;
1524
1525 ftrace_addr = (unsigned long)FTRACE_ADDR;
1526
1527 /*
1528 * If we are enabling tracing:
1529 *
1530 * If the record has a ref count, then we need to enable it
1531 * because someone is using it.
1532 *
1533 * Otherwise we make sure its disabled.
1534 *
1535 * If we are disabling tracing, then disable all records that
1536 * are enabled.
1537 */
1538 if (enable && (rec->flags & ~FTRACE_FL_MASK))
1539 flag = FTRACE_FL_ENABLED;
1540
1541 /* If the state of this record hasn't changed, then do nothing */
1542 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1543 return 0;
1544
1545 if (flag) {
1546 rec->flags |= FTRACE_FL_ENABLED;
1547 return ftrace_make_call(rec, ftrace_addr);
1548 }
1549
1550 rec->flags &= ~FTRACE_FL_ENABLED;
1551 return ftrace_make_nop(NULL, rec, ftrace_addr);
1552}
1553
1554static void ftrace_replace_code(int enable)
1555{
1556 struct dyn_ftrace *rec;
1557 struct ftrace_page *pg;
1558 int failed;
1559
1560 if (unlikely(ftrace_disabled))
1561 return;
1562
1563 do_for_each_ftrace_rec(pg, rec) {
1564 /* Skip over free records */
1565 if (rec->flags & FTRACE_FL_FREE)
1566 continue;
1567
1568 failed = __ftrace_replace_code(rec, enable);
1569 if (failed) {
1570 ftrace_bug(failed, rec->ip);
1571 /* Stop processing */
1572 return;
1573 }
1574 } while_for_each_ftrace_rec();
1575}
1576
1577static int
1578ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1579{
1580 unsigned long ip;
1581 int ret;
1582
1583 ip = rec->ip;
1584
1585 if (unlikely(ftrace_disabled))
1586 return 0;
1587
1588 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1589 if (ret) {
1590 ftrace_bug(ret, ip);
1591 return 0;
1592 }
1593 return 1;
1594}
1595
1596/*
1597 * archs can override this function if they must do something
1598 * before the modifying code is performed.
1599 */
1600int __weak ftrace_arch_code_modify_prepare(void)
1601{
1602 return 0;
1603}
1604
1605/*
1606 * archs can override this function if they must do something
1607 * after the modifying code is performed.
1608 */
1609int __weak ftrace_arch_code_modify_post_process(void)
1610{
1611 return 0;
1612}
1613
1614static int __ftrace_modify_code(void *data)
1615{
1616 int *command = data;
1617
1618 /*
1619 * Do not call function tracer while we update the code.
1620 * We are in stop machine, no worrying about races.
1621 */
1622 function_trace_stop++;
1623
1624 if (*command & FTRACE_ENABLE_CALLS)
1625 ftrace_replace_code(1);
1626 else if (*command & FTRACE_DISABLE_CALLS)
1627 ftrace_replace_code(0);
1628
1629 if (*command & FTRACE_UPDATE_TRACE_FUNC)
1630 ftrace_update_ftrace_func(ftrace_trace_function);
1631
1632 if (*command & FTRACE_START_FUNC_RET)
1633 ftrace_enable_ftrace_graph_caller();
1634 else if (*command & FTRACE_STOP_FUNC_RET)
1635 ftrace_disable_ftrace_graph_caller();
1636
1637#ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
1638 /*
1639 * For archs that call ftrace_test_stop_func(), we must
1640 * wait till after we update all the function callers
1641 * before we update the callback. This keeps different
1642 * ops that record different functions from corrupting
1643 * each other.
1644 */
1645 __ftrace_trace_function = __ftrace_trace_function_delay;
1646#endif
1647 function_trace_stop--;
1648
1649 return 0;
1650}
1651
1652static void ftrace_run_update_code(int command)
1653{
1654 int ret;
1655
1656 ret = ftrace_arch_code_modify_prepare();
1657 FTRACE_WARN_ON(ret);
1658 if (ret)
1659 return;
1660
1661 stop_machine(__ftrace_modify_code, &command, NULL);
1662
1663 ret = ftrace_arch_code_modify_post_process();
1664 FTRACE_WARN_ON(ret);
1665}
1666
1667static ftrace_func_t saved_ftrace_func;
1668static int ftrace_start_up;
1669static int global_start_up;
1670
1671static void ftrace_startup_enable(int command)
1672{
1673 if (saved_ftrace_func != ftrace_trace_function) {
1674 saved_ftrace_func = ftrace_trace_function;
1675 command |= FTRACE_UPDATE_TRACE_FUNC;
1676 }
1677
1678 if (!command || !ftrace_enabled)
1679 return;
1680
1681 ftrace_run_update_code(command);
1682}
1683
1684static int ftrace_startup(struct ftrace_ops *ops, int command)
1685{
1686 bool hash_enable = true;
1687
1688 if (unlikely(ftrace_disabled))
1689 return -ENODEV;
1690
1691 ftrace_start_up++;
1692 command |= FTRACE_ENABLE_CALLS;
1693
1694 /* ops marked global share the filter hashes */
1695 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1696 ops = &global_ops;
1697 /* Don't update hash if global is already set */
1698 if (global_start_up)
1699 hash_enable = false;
1700 global_start_up++;
1701 }
1702
1703 ops->flags |= FTRACE_OPS_FL_ENABLED;
1704 if (hash_enable)
1705 ftrace_hash_rec_enable(ops, 1);
1706
1707 ftrace_startup_enable(command);
1708
1709 return 0;
1710}
1711
1712static void ftrace_shutdown(struct ftrace_ops *ops, int command)
1713{
1714 bool hash_disable = true;
1715
1716 if (unlikely(ftrace_disabled))
1717 return;
1718
1719 ftrace_start_up--;
1720 /*
1721 * Just warn in case of unbalance, no need to kill ftrace, it's not
1722 * critical but the ftrace_call callers may be never nopped again after
1723 * further ftrace uses.
1724 */
1725 WARN_ON_ONCE(ftrace_start_up < 0);
1726
1727 if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1728 ops = &global_ops;
1729 global_start_up--;
1730 WARN_ON_ONCE(global_start_up < 0);
1731 /* Don't update hash if global still has users */
1732 if (global_start_up) {
1733 WARN_ON_ONCE(!ftrace_start_up);
1734 hash_disable = false;
1735 }
1736 }
1737
1738 if (hash_disable)
1739 ftrace_hash_rec_disable(ops, 1);
1740
1741 if (ops != &global_ops || !global_start_up)
1742 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
1743
1744 if (!ftrace_start_up)
1745 command |= FTRACE_DISABLE_CALLS;
1746
1747 if (saved_ftrace_func != ftrace_trace_function) {
1748 saved_ftrace_func = ftrace_trace_function;
1749 command |= FTRACE_UPDATE_TRACE_FUNC;
1750 }
1751
1752 if (!command || !ftrace_enabled)
1753 return;
1754
1755 ftrace_run_update_code(command);
1756}
1757
1758static void ftrace_startup_sysctl(void)
1759{
1760 if (unlikely(ftrace_disabled))
1761 return;
1762
1763 /* Force update next time */
1764 saved_ftrace_func = NULL;
1765 /* ftrace_start_up is true if we want ftrace running */
1766 if (ftrace_start_up)
1767 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1768}
1769
1770static void ftrace_shutdown_sysctl(void)
1771{
1772 if (unlikely(ftrace_disabled))
1773 return;
1774
1775 /* ftrace_start_up is true if ftrace is running */
1776 if (ftrace_start_up)
1777 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
1778}
1779
1780static cycle_t ftrace_update_time;
1781static unsigned long ftrace_update_cnt;
1782unsigned long ftrace_update_tot_cnt;
1783
1784static int ops_traces_mod(struct ftrace_ops *ops)
1785{
1786 struct ftrace_hash *hash;
1787
1788 hash = ops->filter_hash;
1789 return !!(!hash || !hash->count);
1790}
1791
1792static int ftrace_update_code(struct module *mod)
1793{
1794 struct dyn_ftrace *p;
1795 cycle_t start, stop;
1796 unsigned long ref = 0;
1797
1798 /*
1799 * When adding a module, we need to check if tracers are
1800 * currently enabled and if they are set to trace all functions.
1801 * If they are, we need to enable the module functions as well
1802 * as update the reference counts for those function records.
1803 */
1804 if (mod) {
1805 struct ftrace_ops *ops;
1806
1807 for (ops = ftrace_ops_list;
1808 ops != &ftrace_list_end; ops = ops->next) {
1809 if (ops->flags & FTRACE_OPS_FL_ENABLED &&
1810 ops_traces_mod(ops))
1811 ref++;
1812 }
1813 }
1814
1815 start = ftrace_now(raw_smp_processor_id());
1816 ftrace_update_cnt = 0;
1817
1818 while (ftrace_new_addrs) {
1819
1820 /* If something went wrong, bail without enabling anything */
1821 if (unlikely(ftrace_disabled))
1822 return -1;
1823
1824 p = ftrace_new_addrs;
1825 ftrace_new_addrs = p->newlist;
1826 p->flags = ref;
1827
1828 /*
1829 * Do the initial record conversion from mcount jump
1830 * to the NOP instructions.
1831 */
1832 if (!ftrace_code_disable(mod, p)) {
1833 ftrace_free_rec(p);
1834 /* Game over */
1835 break;
1836 }
1837
1838 ftrace_update_cnt++;
1839
1840 /*
1841 * If the tracing is enabled, go ahead and enable the record.
1842 *
1843 * The reason not to enable the record immediatelly is the
1844 * inherent check of ftrace_make_nop/ftrace_make_call for
1845 * correct previous instructions. Making first the NOP
1846 * conversion puts the module to the correct state, thus
1847 * passing the ftrace_make_call check.
1848 */
1849 if (ftrace_start_up && ref) {
1850 int failed = __ftrace_replace_code(p, 1);
1851 if (failed) {
1852 ftrace_bug(failed, p->ip);
1853 ftrace_free_rec(p);
1854 }
1855 }
1856 }
1857
1858 stop = ftrace_now(raw_smp_processor_id());
1859 ftrace_update_time = stop - start;
1860 ftrace_update_tot_cnt += ftrace_update_cnt;
1861
1862 return 0;
1863}
1864
1865static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1866{
1867 struct ftrace_page *pg;
1868 int cnt;
1869 int i;
1870
1871 /* allocate a few pages */
1872 ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1873 if (!ftrace_pages_start)
1874 return -1;
1875
1876 /*
1877 * Allocate a few more pages.
1878 *
1879 * TODO: have some parser search vmlinux before
1880 * final linking to find all calls to ftrace.
1881 * Then we can:
1882 * a) know how many pages to allocate.
1883 * and/or
1884 * b) set up the table then.
1885 *
1886 * The dynamic code is still necessary for
1887 * modules.
1888 */
1889
1890 pg = ftrace_pages = ftrace_pages_start;
1891
1892 cnt = num_to_init / ENTRIES_PER_PAGE;
1893 pr_info("ftrace: allocating %ld entries in %d pages\n",
1894 num_to_init, cnt + 1);
1895
1896 for (i = 0; i < cnt; i++) {
1897 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1898
1899 /* If we fail, we'll try later anyway */
1900 if (!pg->next)
1901 break;
1902
1903 pg = pg->next;
1904 }
1905
1906 return 0;
1907}
1908
1909enum {
1910 FTRACE_ITER_FILTER = (1 << 0),
1911 FTRACE_ITER_NOTRACE = (1 << 1),
1912 FTRACE_ITER_PRINTALL = (1 << 2),
1913 FTRACE_ITER_HASH = (1 << 3),
1914 FTRACE_ITER_ENABLED = (1 << 4),
1915};
1916
1917#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1918
1919struct ftrace_iterator {
1920 loff_t pos;
1921 loff_t func_pos;
1922 struct ftrace_page *pg;
1923 struct dyn_ftrace *func;
1924 struct ftrace_func_probe *probe;
1925 struct trace_parser parser;
1926 struct ftrace_hash *hash;
1927 struct ftrace_ops *ops;
1928 int hidx;
1929 int idx;
1930 unsigned flags;
1931};
1932
1933static void *
1934t_hash_next(struct seq_file *m, loff_t *pos)
1935{
1936 struct ftrace_iterator *iter = m->private;
1937 struct hlist_node *hnd = NULL;
1938 struct hlist_head *hhd;
1939
1940 (*pos)++;
1941 iter->pos = *pos;
1942
1943 if (iter->probe)
1944 hnd = &iter->probe->node;
1945 retry:
1946 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1947 return NULL;
1948
1949 hhd = &ftrace_func_hash[iter->hidx];
1950
1951 if (hlist_empty(hhd)) {
1952 iter->hidx++;
1953 hnd = NULL;
1954 goto retry;
1955 }
1956
1957 if (!hnd)
1958 hnd = hhd->first;
1959 else {
1960 hnd = hnd->next;
1961 if (!hnd) {
1962 iter->hidx++;
1963 goto retry;
1964 }
1965 }
1966
1967 if (WARN_ON_ONCE(!hnd))
1968 return NULL;
1969
1970 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
1971
1972 return iter;
1973}
1974
1975static void *t_hash_start(struct seq_file *m, loff_t *pos)
1976{
1977 struct ftrace_iterator *iter = m->private;
1978 void *p = NULL;
1979 loff_t l;
1980
1981 if (iter->func_pos > *pos)
1982 return NULL;
1983
1984 iter->hidx = 0;
1985 for (l = 0; l <= (*pos - iter->func_pos); ) {
1986 p = t_hash_next(m, &l);
1987 if (!p)
1988 break;
1989 }
1990 if (!p)
1991 return NULL;
1992
1993 /* Only set this if we have an item */
1994 iter->flags |= FTRACE_ITER_HASH;
1995
1996 return iter;
1997}
1998
1999static int
2000t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2001{
2002 struct ftrace_func_probe *rec;
2003
2004 rec = iter->probe;
2005 if (WARN_ON_ONCE(!rec))
2006 return -EIO;
2007
2008 if (rec->ops->print)
2009 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2010
2011 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2012
2013 if (rec->data)
2014 seq_printf(m, ":%p", rec->data);
2015 seq_putc(m, '\n');
2016
2017 return 0;
2018}
2019
2020static void *
2021t_next(struct seq_file *m, void *v, loff_t *pos)
2022{
2023 struct ftrace_iterator *iter = m->private;
2024 struct ftrace_ops *ops = &global_ops;
2025 struct dyn_ftrace *rec = NULL;
2026
2027 if (unlikely(ftrace_disabled))
2028 return NULL;
2029
2030 if (iter->flags & FTRACE_ITER_HASH)
2031 return t_hash_next(m, pos);
2032
2033 (*pos)++;
2034 iter->pos = iter->func_pos = *pos;
2035
2036 if (iter->flags & FTRACE_ITER_PRINTALL)
2037 return t_hash_start(m, pos);
2038
2039 retry:
2040 if (iter->idx >= iter->pg->index) {
2041 if (iter->pg->next) {
2042 iter->pg = iter->pg->next;
2043 iter->idx = 0;
2044 goto retry;
2045 }
2046 } else {
2047 rec = &iter->pg->records[iter->idx++];
2048 if ((rec->flags & FTRACE_FL_FREE) ||
2049
2050 ((iter->flags & FTRACE_ITER_FILTER) &&
2051 !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2052
2053 ((iter->flags & FTRACE_ITER_NOTRACE) &&
2054 !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2055
2056 ((iter->flags & FTRACE_ITER_ENABLED) &&
2057 !(rec->flags & ~FTRACE_FL_MASK))) {
2058
2059 rec = NULL;
2060 goto retry;
2061 }
2062 }
2063
2064 if (!rec)
2065 return t_hash_start(m, pos);
2066
2067 iter->func = rec;
2068
2069 return iter;
2070}
2071
2072static void reset_iter_read(struct ftrace_iterator *iter)
2073{
2074 iter->pos = 0;
2075 iter->func_pos = 0;
2076 iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
2077}
2078
2079static void *t_start(struct seq_file *m, loff_t *pos)
2080{
2081 struct ftrace_iterator *iter = m->private;
2082 struct ftrace_ops *ops = &global_ops;
2083 void *p = NULL;
2084 loff_t l;
2085
2086 mutex_lock(&ftrace_lock);
2087
2088 if (unlikely(ftrace_disabled))
2089 return NULL;
2090
2091 /*
2092 * If an lseek was done, then reset and start from beginning.
2093 */
2094 if (*pos < iter->pos)
2095 reset_iter_read(iter);
2096
2097 /*
2098 * For set_ftrace_filter reading, if we have the filter
2099 * off, we can short cut and just print out that all
2100 * functions are enabled.
2101 */
2102 if (iter->flags & FTRACE_ITER_FILTER && !ops->filter_hash->count) {
2103 if (*pos > 0)
2104 return t_hash_start(m, pos);
2105 iter->flags |= FTRACE_ITER_PRINTALL;
2106 /* reset in case of seek/pread */
2107 iter->flags &= ~FTRACE_ITER_HASH;
2108 return iter;
2109 }
2110
2111 if (iter->flags & FTRACE_ITER_HASH)
2112 return t_hash_start(m, pos);
2113
2114 /*
2115 * Unfortunately, we need to restart at ftrace_pages_start
2116 * every time we let go of the ftrace_mutex. This is because
2117 * those pointers can change without the lock.
2118 */
2119 iter->pg = ftrace_pages_start;
2120 iter->idx = 0;
2121 for (l = 0; l <= *pos; ) {
2122 p = t_next(m, p, &l);
2123 if (!p)
2124 break;
2125 }
2126
2127 if (!p) {
2128 if (iter->flags & FTRACE_ITER_FILTER)
2129 return t_hash_start(m, pos);
2130
2131 return NULL;
2132 }
2133
2134 return iter;
2135}
2136
2137static void t_stop(struct seq_file *m, void *p)
2138{
2139 mutex_unlock(&ftrace_lock);
2140}
2141
2142static int t_show(struct seq_file *m, void *v)
2143{
2144 struct ftrace_iterator *iter = m->private;
2145 struct dyn_ftrace *rec;
2146
2147 if (iter->flags & FTRACE_ITER_HASH)
2148 return t_hash_show(m, iter);
2149
2150 if (iter->flags & FTRACE_ITER_PRINTALL) {
2151 seq_printf(m, "#### all functions enabled ####\n");
2152 return 0;
2153 }
2154
2155 rec = iter->func;
2156
2157 if (!rec)
2158 return 0;
2159
2160 seq_printf(m, "%ps", (void *)rec->ip);
2161 if (iter->flags & FTRACE_ITER_ENABLED)
2162 seq_printf(m, " (%ld)",
2163 rec->flags & ~FTRACE_FL_MASK);
2164 seq_printf(m, "\n");
2165
2166 return 0;
2167}
2168
2169static const struct seq_operations show_ftrace_seq_ops = {
2170 .start = t_start,
2171 .next = t_next,
2172 .stop = t_stop,
2173 .show = t_show,
2174};
2175
2176static int
2177ftrace_avail_open(struct inode *inode, struct file *file)
2178{
2179 struct ftrace_iterator *iter;
2180 int ret;
2181
2182 if (unlikely(ftrace_disabled))
2183 return -ENODEV;
2184
2185 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2186 if (!iter)
2187 return -ENOMEM;
2188
2189 iter->pg = ftrace_pages_start;
2190
2191 ret = seq_open(file, &show_ftrace_seq_ops);
2192 if (!ret) {
2193 struct seq_file *m = file->private_data;
2194
2195 m->private = iter;
2196 } else {
2197 kfree(iter);
2198 }
2199
2200 return ret;
2201}
2202
2203static int
2204ftrace_enabled_open(struct inode *inode, struct file *file)
2205{
2206 struct ftrace_iterator *iter;
2207 int ret;
2208
2209 if (unlikely(ftrace_disabled))
2210 return -ENODEV;
2211
2212 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2213 if (!iter)
2214 return -ENOMEM;
2215
2216 iter->pg = ftrace_pages_start;
2217 iter->flags = FTRACE_ITER_ENABLED;
2218
2219 ret = seq_open(file, &show_ftrace_seq_ops);
2220 if (!ret) {
2221 struct seq_file *m = file->private_data;
2222
2223 m->private = iter;
2224 } else {
2225 kfree(iter);
2226 }
2227
2228 return ret;
2229}
2230
2231static void ftrace_filter_reset(struct ftrace_hash *hash)
2232{
2233 mutex_lock(&ftrace_lock);
2234 ftrace_hash_clear(hash);
2235 mutex_unlock(&ftrace_lock);
2236}
2237
2238static int
2239ftrace_regex_open(struct ftrace_ops *ops, int flag,
2240 struct inode *inode, struct file *file)
2241{
2242 struct ftrace_iterator *iter;
2243 struct ftrace_hash *hash;
2244 int ret = 0;
2245
2246 if (unlikely(ftrace_disabled))
2247 return -ENODEV;
2248
2249 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2250 if (!iter)
2251 return -ENOMEM;
2252
2253 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2254 kfree(iter);
2255 return -ENOMEM;
2256 }
2257
2258 if (flag & FTRACE_ITER_NOTRACE)
2259 hash = ops->notrace_hash;
2260 else
2261 hash = ops->filter_hash;
2262
2263 iter->ops = ops;
2264 iter->flags = flag;
2265
2266 if (file->f_mode & FMODE_WRITE) {
2267 mutex_lock(&ftrace_lock);
2268 iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2269 mutex_unlock(&ftrace_lock);
2270
2271 if (!iter->hash) {
2272 trace_parser_put(&iter->parser);
2273 kfree(iter);
2274 return -ENOMEM;
2275 }
2276 }
2277
2278 mutex_lock(&ftrace_regex_lock);
2279
2280 if ((file->f_mode & FMODE_WRITE) &&
2281 (file->f_flags & O_TRUNC))
2282 ftrace_filter_reset(iter->hash);
2283
2284 if (file->f_mode & FMODE_READ) {
2285 iter->pg = ftrace_pages_start;
2286
2287 ret = seq_open(file, &show_ftrace_seq_ops);
2288 if (!ret) {
2289 struct seq_file *m = file->private_data;
2290 m->private = iter;
2291 } else {
2292 /* Failed */
2293 free_ftrace_hash(iter->hash);
2294 trace_parser_put(&iter->parser);
2295 kfree(iter);
2296 }
2297 } else
2298 file->private_data = iter;
2299 mutex_unlock(&ftrace_regex_lock);
2300
2301 return ret;
2302}
2303
2304static int
2305ftrace_filter_open(struct inode *inode, struct file *file)
2306{
2307 return ftrace_regex_open(&global_ops, FTRACE_ITER_FILTER,
2308 inode, file);
2309}
2310
2311static int
2312ftrace_notrace_open(struct inode *inode, struct file *file)
2313{
2314 return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2315 inode, file);
2316}
2317
2318static loff_t
2319ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
2320{
2321 loff_t ret;
2322
2323 if (file->f_mode & FMODE_READ)
2324 ret = seq_lseek(file, offset, origin);
2325 else
2326 file->f_pos = ret = 1;
2327
2328 return ret;
2329}
2330
2331static int ftrace_match(char *str, char *regex, int len, int type)
2332{
2333 int matched = 0;
2334 int slen;
2335
2336 switch (type) {
2337 case MATCH_FULL:
2338 if (strcmp(str, regex) == 0)
2339 matched = 1;
2340 break;
2341 case MATCH_FRONT_ONLY:
2342 if (strncmp(str, regex, len) == 0)
2343 matched = 1;
2344 break;
2345 case MATCH_MIDDLE_ONLY:
2346 if (strstr(str, regex))
2347 matched = 1;
2348 break;
2349 case MATCH_END_ONLY:
2350 slen = strlen(str);
2351 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2352 matched = 1;
2353 break;
2354 }
2355
2356 return matched;
2357}
2358
2359static int
2360enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2361{
2362 struct ftrace_func_entry *entry;
2363 int ret = 0;
2364
2365 entry = ftrace_lookup_ip(hash, rec->ip);
2366 if (not) {
2367 /* Do nothing if it doesn't exist */
2368 if (!entry)
2369 return 0;
2370
2371 free_hash_entry(hash, entry);
2372 } else {
2373 /* Do nothing if it exists */
2374 if (entry)
2375 return 0;
2376
2377 ret = add_hash_entry(hash, rec->ip);
2378 }
2379 return ret;
2380}
2381
2382static int
2383ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2384 char *regex, int len, int type)
2385{
2386 char str[KSYM_SYMBOL_LEN];
2387 char *modname;
2388
2389 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2390
2391 if (mod) {
2392 /* module lookup requires matching the module */
2393 if (!modname || strcmp(modname, mod))
2394 return 0;
2395
2396 /* blank search means to match all funcs in the mod */
2397 if (!len)
2398 return 1;
2399 }
2400
2401 return ftrace_match(str, regex, len, type);
2402}
2403
2404static int
2405match_records(struct ftrace_hash *hash, char *buff,
2406 int len, char *mod, int not)
2407{
2408 unsigned search_len = 0;
2409 struct ftrace_page *pg;
2410 struct dyn_ftrace *rec;
2411 int type = MATCH_FULL;
2412 char *search = buff;
2413 int found = 0;
2414 int ret;
2415
2416 if (len) {
2417 type = filter_parse_regex(buff, len, &search, ¬);
2418 search_len = strlen(search);
2419 }
2420
2421 mutex_lock(&ftrace_lock);
2422
2423 if (unlikely(ftrace_disabled))
2424 goto out_unlock;
2425
2426 do_for_each_ftrace_rec(pg, rec) {
2427
2428 if (ftrace_match_record(rec, mod, search, search_len, type)) {
2429 ret = enter_record(hash, rec, not);
2430 if (ret < 0) {
2431 found = ret;
2432 goto out_unlock;
2433 }
2434 found = 1;
2435 }
2436 } while_for_each_ftrace_rec();
2437 out_unlock:
2438 mutex_unlock(&ftrace_lock);
2439
2440 return found;
2441}
2442
2443static int
2444ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2445{
2446 return match_records(hash, buff, len, NULL, 0);
2447}
2448
2449static int
2450ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2451{
2452 int not = 0;
2453
2454 /* blank or '*' mean the same */
2455 if (strcmp(buff, "*") == 0)
2456 buff[0] = 0;
2457
2458 /* handle the case of 'dont filter this module' */
2459 if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2460 buff[0] = 0;
2461 not = 1;
2462 }
2463
2464 return match_records(hash, buff, strlen(buff), mod, not);
2465}
2466
2467/*
2468 * We register the module command as a template to show others how
2469 * to register the a command as well.
2470 */
2471
2472static int
2473ftrace_mod_callback(struct ftrace_hash *hash,
2474 char *func, char *cmd, char *param, int enable)
2475{
2476 char *mod;
2477 int ret = -EINVAL;
2478
2479 /*
2480 * cmd == 'mod' because we only registered this func
2481 * for the 'mod' ftrace_func_command.
2482 * But if you register one func with multiple commands,
2483 * you can tell which command was used by the cmd
2484 * parameter.
2485 */
2486
2487 /* we must have a module name */
2488 if (!param)
2489 return ret;
2490
2491 mod = strsep(¶m, ":");
2492 if (!strlen(mod))
2493 return ret;
2494
2495 ret = ftrace_match_module_records(hash, func, mod);
2496 if (!ret)
2497 ret = -EINVAL;
2498 if (ret < 0)
2499 return ret;
2500
2501 return 0;
2502}
2503
2504static struct ftrace_func_command ftrace_mod_cmd = {
2505 .name = "mod",
2506 .func = ftrace_mod_callback,
2507};
2508
2509static int __init ftrace_mod_cmd_init(void)
2510{
2511 return register_ftrace_command(&ftrace_mod_cmd);
2512}
2513device_initcall(ftrace_mod_cmd_init);
2514
2515static void
2516function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
2517{
2518 struct ftrace_func_probe *entry;
2519 struct hlist_head *hhd;
2520 struct hlist_node *n;
2521 unsigned long key;
2522
2523 key = hash_long(ip, FTRACE_HASH_BITS);
2524
2525 hhd = &ftrace_func_hash[key];
2526
2527 if (hlist_empty(hhd))
2528 return;
2529
2530 /*
2531 * Disable preemption for these calls to prevent a RCU grace
2532 * period. This syncs the hash iteration and freeing of items
2533 * on the hash. rcu_read_lock is too dangerous here.
2534 */
2535 preempt_disable_notrace();
2536 hlist_for_each_entry_rcu(entry, n, hhd, node) {
2537 if (entry->ip == ip)
2538 entry->ops->func(ip, parent_ip, &entry->data);
2539 }
2540 preempt_enable_notrace();
2541}
2542
2543static struct ftrace_ops trace_probe_ops __read_mostly =
2544{
2545 .func = function_trace_probe_call,
2546};
2547
2548static int ftrace_probe_registered;
2549
2550static void __enable_ftrace_function_probe(void)
2551{
2552 int ret;
2553 int i;
2554
2555 if (ftrace_probe_registered)
2556 return;
2557
2558 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2559 struct hlist_head *hhd = &ftrace_func_hash[i];
2560 if (hhd->first)
2561 break;
2562 }
2563 /* Nothing registered? */
2564 if (i == FTRACE_FUNC_HASHSIZE)
2565 return;
2566
2567 ret = __register_ftrace_function(&trace_probe_ops);
2568 if (!ret)
2569 ret = ftrace_startup(&trace_probe_ops, 0);
2570
2571 ftrace_probe_registered = 1;
2572}
2573
2574static void __disable_ftrace_function_probe(void)
2575{
2576 int ret;
2577 int i;
2578
2579 if (!ftrace_probe_registered)
2580 return;
2581
2582 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2583 struct hlist_head *hhd = &ftrace_func_hash[i];
2584 if (hhd->first)
2585 return;
2586 }
2587
2588 /* no more funcs left */
2589 ret = __unregister_ftrace_function(&trace_probe_ops);
2590 if (!ret)
2591 ftrace_shutdown(&trace_probe_ops, 0);
2592
2593 ftrace_probe_registered = 0;
2594}
2595
2596
2597static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2598{
2599 struct ftrace_func_probe *entry =
2600 container_of(rhp, struct ftrace_func_probe, rcu);
2601
2602 if (entry->ops->free)
2603 entry->ops->free(&entry->data);
2604 kfree(entry);
2605}
2606
2607
2608int
2609register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2610 void *data)
2611{
2612 struct ftrace_func_probe *entry;
2613 struct ftrace_page *pg;
2614 struct dyn_ftrace *rec;
2615 int type, len, not;
2616 unsigned long key;
2617 int count = 0;
2618 char *search;
2619
2620 type = filter_parse_regex(glob, strlen(glob), &search, ¬);
2621 len = strlen(search);
2622
2623 /* we do not support '!' for function probes */
2624 if (WARN_ON(not))
2625 return -EINVAL;
2626
2627 mutex_lock(&ftrace_lock);
2628
2629 if (unlikely(ftrace_disabled))
2630 goto out_unlock;
2631
2632 do_for_each_ftrace_rec(pg, rec) {
2633
2634 if (!ftrace_match_record(rec, NULL, search, len, type))
2635 continue;
2636
2637 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2638 if (!entry) {
2639 /* If we did not process any, then return error */
2640 if (!count)
2641 count = -ENOMEM;
2642 goto out_unlock;
2643 }
2644
2645 count++;
2646
2647 entry->data = data;
2648
2649 /*
2650 * The caller might want to do something special
2651 * for each function we find. We call the callback
2652 * to give the caller an opportunity to do so.
2653 */
2654 if (ops->callback) {
2655 if (ops->callback(rec->ip, &entry->data) < 0) {
2656 /* caller does not like this func */
2657 kfree(entry);
2658 continue;
2659 }
2660 }
2661
2662 entry->ops = ops;
2663 entry->ip = rec->ip;
2664
2665 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2666 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2667
2668 } while_for_each_ftrace_rec();
2669 __enable_ftrace_function_probe();
2670
2671 out_unlock:
2672 mutex_unlock(&ftrace_lock);
2673
2674 return count;
2675}
2676
2677enum {
2678 PROBE_TEST_FUNC = 1,
2679 PROBE_TEST_DATA = 2
2680};
2681
2682static void
2683__unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2684 void *data, int flags)
2685{
2686 struct ftrace_func_probe *entry;
2687 struct hlist_node *n, *tmp;
2688 char str[KSYM_SYMBOL_LEN];
2689 int type = MATCH_FULL;
2690 int i, len = 0;
2691 char *search;
2692
2693 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2694 glob = NULL;
2695 else if (glob) {
2696 int not;
2697
2698 type = filter_parse_regex(glob, strlen(glob), &search, ¬);
2699 len = strlen(search);
2700
2701 /* we do not support '!' for function probes */
2702 if (WARN_ON(not))
2703 return;
2704 }
2705
2706 mutex_lock(&ftrace_lock);
2707 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2708 struct hlist_head *hhd = &ftrace_func_hash[i];
2709
2710 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2711
2712 /* break up if statements for readability */
2713 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2714 continue;
2715
2716 if ((flags & PROBE_TEST_DATA) && entry->data != data)
2717 continue;
2718
2719 /* do this last, since it is the most expensive */
2720 if (glob) {
2721 kallsyms_lookup(entry->ip, NULL, NULL,
2722 NULL, str);
2723 if (!ftrace_match(str, glob, len, type))
2724 continue;
2725 }
2726
2727 hlist_del(&entry->node);
2728 call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2729 }
2730 }
2731 __disable_ftrace_function_probe();
2732 mutex_unlock(&ftrace_lock);
2733}
2734
2735void
2736unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2737 void *data)
2738{
2739 __unregister_ftrace_function_probe(glob, ops, data,
2740 PROBE_TEST_FUNC | PROBE_TEST_DATA);
2741}
2742
2743void
2744unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2745{
2746 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2747}
2748
2749void unregister_ftrace_function_probe_all(char *glob)
2750{
2751 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2752}
2753
2754static LIST_HEAD(ftrace_commands);
2755static DEFINE_MUTEX(ftrace_cmd_mutex);
2756
2757int register_ftrace_command(struct ftrace_func_command *cmd)
2758{
2759 struct ftrace_func_command *p;
2760 int ret = 0;
2761
2762 mutex_lock(&ftrace_cmd_mutex);
2763 list_for_each_entry(p, &ftrace_commands, list) {
2764 if (strcmp(cmd->name, p->name) == 0) {
2765 ret = -EBUSY;
2766 goto out_unlock;
2767 }
2768 }
2769 list_add(&cmd->list, &ftrace_commands);
2770 out_unlock:
2771 mutex_unlock(&ftrace_cmd_mutex);
2772
2773 return ret;
2774}
2775
2776int unregister_ftrace_command(struct ftrace_func_command *cmd)
2777{
2778 struct ftrace_func_command *p, *n;
2779 int ret = -ENODEV;
2780
2781 mutex_lock(&ftrace_cmd_mutex);
2782 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2783 if (strcmp(cmd->name, p->name) == 0) {
2784 ret = 0;
2785 list_del_init(&p->list);
2786 goto out_unlock;
2787 }
2788 }
2789 out_unlock:
2790 mutex_unlock(&ftrace_cmd_mutex);
2791
2792 return ret;
2793}
2794
2795static int ftrace_process_regex(struct ftrace_hash *hash,
2796 char *buff, int len, int enable)
2797{
2798 char *func, *command, *next = buff;
2799 struct ftrace_func_command *p;
2800 int ret = -EINVAL;
2801
2802 func = strsep(&next, ":");
2803
2804 if (!next) {
2805 ret = ftrace_match_records(hash, func, len);
2806 if (!ret)
2807 ret = -EINVAL;
2808 if (ret < 0)
2809 return ret;
2810 return 0;
2811 }
2812
2813 /* command found */
2814
2815 command = strsep(&next, ":");
2816
2817 mutex_lock(&ftrace_cmd_mutex);
2818 list_for_each_entry(p, &ftrace_commands, list) {
2819 if (strcmp(p->name, command) == 0) {
2820 ret = p->func(hash, func, command, next, enable);
2821 goto out_unlock;
2822 }
2823 }
2824 out_unlock:
2825 mutex_unlock(&ftrace_cmd_mutex);
2826
2827 return ret;
2828}
2829
2830static ssize_t
2831ftrace_regex_write(struct file *file, const char __user *ubuf,
2832 size_t cnt, loff_t *ppos, int enable)
2833{
2834 struct ftrace_iterator *iter;
2835 struct trace_parser *parser;
2836 ssize_t ret, read;
2837
2838 if (!cnt)
2839 return 0;
2840
2841 mutex_lock(&ftrace_regex_lock);
2842
2843 ret = -ENODEV;
2844 if (unlikely(ftrace_disabled))
2845 goto out_unlock;
2846
2847 if (file->f_mode & FMODE_READ) {
2848 struct seq_file *m = file->private_data;
2849 iter = m->private;
2850 } else
2851 iter = file->private_data;
2852
2853 parser = &iter->parser;
2854 read = trace_get_user(parser, ubuf, cnt, ppos);
2855
2856 if (read >= 0 && trace_parser_loaded(parser) &&
2857 !trace_parser_cont(parser)) {
2858 ret = ftrace_process_regex(iter->hash, parser->buffer,
2859 parser->idx, enable);
2860 trace_parser_clear(parser);
2861 if (ret)
2862 goto out_unlock;
2863 }
2864
2865 ret = read;
2866out_unlock:
2867 mutex_unlock(&ftrace_regex_lock);
2868
2869 return ret;
2870}
2871
2872static ssize_t
2873ftrace_filter_write(struct file *file, const char __user *ubuf,
2874 size_t cnt, loff_t *ppos)
2875{
2876 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2877}
2878
2879static ssize_t
2880ftrace_notrace_write(struct file *file, const char __user *ubuf,
2881 size_t cnt, loff_t *ppos)
2882{
2883 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2884}
2885
2886static int
2887ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
2888 int reset, int enable)
2889{
2890 struct ftrace_hash **orig_hash;
2891 struct ftrace_hash *hash;
2892 int ret;
2893
2894 /* All global ops uses the global ops filters */
2895 if (ops->flags & FTRACE_OPS_FL_GLOBAL)
2896 ops = &global_ops;
2897
2898 if (unlikely(ftrace_disabled))
2899 return -ENODEV;
2900
2901 if (enable)
2902 orig_hash = &ops->filter_hash;
2903 else
2904 orig_hash = &ops->notrace_hash;
2905
2906 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
2907 if (!hash)
2908 return -ENOMEM;
2909
2910 mutex_lock(&ftrace_regex_lock);
2911 if (reset)
2912 ftrace_filter_reset(hash);
2913 if (buf)
2914 ftrace_match_records(hash, buf, len);
2915
2916 mutex_lock(&ftrace_lock);
2917 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
2918 if (!ret && ops->flags & FTRACE_OPS_FL_ENABLED
2919 && ftrace_enabled)
2920 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2921
2922 mutex_unlock(&ftrace_lock);
2923
2924 mutex_unlock(&ftrace_regex_lock);
2925
2926 free_ftrace_hash(hash);
2927 return ret;
2928}
2929
2930/**
2931 * ftrace_set_filter - set a function to filter on in ftrace
2932 * @ops - the ops to set the filter with
2933 * @buf - the string that holds the function filter text.
2934 * @len - the length of the string.
2935 * @reset - non zero to reset all filters before applying this filter.
2936 *
2937 * Filters denote which functions should be enabled when tracing is enabled.
2938 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2939 */
2940void ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
2941 int len, int reset)
2942{
2943 ftrace_set_regex(ops, buf, len, reset, 1);
2944}
2945EXPORT_SYMBOL_GPL(ftrace_set_filter);
2946
2947/**
2948 * ftrace_set_notrace - set a function to not trace in ftrace
2949 * @ops - the ops to set the notrace filter with
2950 * @buf - the string that holds the function notrace text.
2951 * @len - the length of the string.
2952 * @reset - non zero to reset all filters before applying this filter.
2953 *
2954 * Notrace Filters denote which functions should not be enabled when tracing
2955 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2956 * for tracing.
2957 */
2958void ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
2959 int len, int reset)
2960{
2961 ftrace_set_regex(ops, buf, len, reset, 0);
2962}
2963EXPORT_SYMBOL_GPL(ftrace_set_notrace);
2964/**
2965 * ftrace_set_filter - set a function to filter on in ftrace
2966 * @ops - the ops to set the filter with
2967 * @buf - the string that holds the function filter text.
2968 * @len - the length of the string.
2969 * @reset - non zero to reset all filters before applying this filter.
2970 *
2971 * Filters denote which functions should be enabled when tracing is enabled.
2972 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2973 */
2974void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
2975{
2976 ftrace_set_regex(&global_ops, buf, len, reset, 1);
2977}
2978EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
2979
2980/**
2981 * ftrace_set_notrace - set a function to not trace in ftrace
2982 * @ops - the ops to set the notrace filter with
2983 * @buf - the string that holds the function notrace text.
2984 * @len - the length of the string.
2985 * @reset - non zero to reset all filters before applying this filter.
2986 *
2987 * Notrace Filters denote which functions should not be enabled when tracing
2988 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2989 * for tracing.
2990 */
2991void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
2992{
2993 ftrace_set_regex(&global_ops, buf, len, reset, 0);
2994}
2995EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
2996
2997/*
2998 * command line interface to allow users to set filters on boot up.
2999 */
3000#define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
3001static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3002static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3003
3004static int __init set_ftrace_notrace(char *str)
3005{
3006 strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3007 return 1;
3008}
3009__setup("ftrace_notrace=", set_ftrace_notrace);
3010
3011static int __init set_ftrace_filter(char *str)
3012{
3013 strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3014 return 1;
3015}
3016__setup("ftrace_filter=", set_ftrace_filter);
3017
3018#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3019static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3020static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
3021
3022static int __init set_graph_function(char *str)
3023{
3024 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3025 return 1;
3026}
3027__setup("ftrace_graph_filter=", set_graph_function);
3028
3029static void __init set_ftrace_early_graph(char *buf)
3030{
3031 int ret;
3032 char *func;
3033
3034 while (buf) {
3035 func = strsep(&buf, ",");
3036 /* we allow only one expression at a time */
3037 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3038 func);
3039 if (ret)
3040 printk(KERN_DEBUG "ftrace: function %s not "
3041 "traceable\n", func);
3042 }
3043}
3044#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3045
3046static void __init
3047set_ftrace_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3048{
3049 char *func;
3050
3051 while (buf) {
3052 func = strsep(&buf, ",");
3053 ftrace_set_regex(ops, func, strlen(func), 0, enable);
3054 }
3055}
3056
3057static void __init set_ftrace_early_filters(void)
3058{
3059 if (ftrace_filter_buf[0])
3060 set_ftrace_early_filter(&global_ops, ftrace_filter_buf, 1);
3061 if (ftrace_notrace_buf[0])
3062 set_ftrace_early_filter(&global_ops, ftrace_notrace_buf, 0);
3063#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3064 if (ftrace_graph_buf[0])
3065 set_ftrace_early_graph(ftrace_graph_buf);
3066#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3067}
3068
3069static int
3070ftrace_regex_release(struct inode *inode, struct file *file)
3071{
3072 struct seq_file *m = (struct seq_file *)file->private_data;
3073 struct ftrace_iterator *iter;
3074 struct ftrace_hash **orig_hash;
3075 struct trace_parser *parser;
3076 int filter_hash;
3077 int ret;
3078
3079 mutex_lock(&ftrace_regex_lock);
3080 if (file->f_mode & FMODE_READ) {
3081 iter = m->private;
3082
3083 seq_release(inode, file);
3084 } else
3085 iter = file->private_data;
3086
3087 parser = &iter->parser;
3088 if (trace_parser_loaded(parser)) {
3089 parser->buffer[parser->idx] = 0;
3090 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3091 }
3092
3093 trace_parser_put(parser);
3094
3095 if (file->f_mode & FMODE_WRITE) {
3096 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3097
3098 if (filter_hash)
3099 orig_hash = &iter->ops->filter_hash;
3100 else
3101 orig_hash = &iter->ops->notrace_hash;
3102
3103 mutex_lock(&ftrace_lock);
3104 ret = ftrace_hash_move(iter->ops, filter_hash,
3105 orig_hash, iter->hash);
3106 if (!ret && (iter->ops->flags & FTRACE_OPS_FL_ENABLED)
3107 && ftrace_enabled)
3108 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
3109
3110 mutex_unlock(&ftrace_lock);
3111 }
3112 free_ftrace_hash(iter->hash);
3113 kfree(iter);
3114
3115 mutex_unlock(&ftrace_regex_lock);
3116 return 0;
3117}
3118
3119static const struct file_operations ftrace_avail_fops = {
3120 .open = ftrace_avail_open,
3121 .read = seq_read,
3122 .llseek = seq_lseek,
3123 .release = seq_release_private,
3124};
3125
3126static const struct file_operations ftrace_enabled_fops = {
3127 .open = ftrace_enabled_open,
3128 .read = seq_read,
3129 .llseek = seq_lseek,
3130 .release = seq_release_private,
3131};
3132
3133static const struct file_operations ftrace_filter_fops = {
3134 .open = ftrace_filter_open,
3135 .read = seq_read,
3136 .write = ftrace_filter_write,
3137 .llseek = ftrace_regex_lseek,
3138 .release = ftrace_regex_release,
3139};
3140
3141static const struct file_operations ftrace_notrace_fops = {
3142 .open = ftrace_notrace_open,
3143 .read = seq_read,
3144 .write = ftrace_notrace_write,
3145 .llseek = ftrace_regex_lseek,
3146 .release = ftrace_regex_release,
3147};
3148
3149#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3150
3151static DEFINE_MUTEX(graph_lock);
3152
3153int ftrace_graph_count;
3154int ftrace_graph_filter_enabled;
3155unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3156
3157static void *
3158__g_next(struct seq_file *m, loff_t *pos)
3159{
3160 if (*pos >= ftrace_graph_count)
3161 return NULL;
3162 return &ftrace_graph_funcs[*pos];
3163}
3164
3165static void *
3166g_next(struct seq_file *m, void *v, loff_t *pos)
3167{
3168 (*pos)++;
3169 return __g_next(m, pos);
3170}
3171
3172static void *g_start(struct seq_file *m, loff_t *pos)
3173{
3174 mutex_lock(&graph_lock);
3175
3176 /* Nothing, tell g_show to print all functions are enabled */
3177 if (!ftrace_graph_filter_enabled && !*pos)
3178 return (void *)1;
3179
3180 return __g_next(m, pos);
3181}
3182
3183static void g_stop(struct seq_file *m, void *p)
3184{
3185 mutex_unlock(&graph_lock);
3186}
3187
3188static int g_show(struct seq_file *m, void *v)
3189{
3190 unsigned long *ptr = v;
3191
3192 if (!ptr)
3193 return 0;
3194
3195 if (ptr == (unsigned long *)1) {
3196 seq_printf(m, "#### all functions enabled ####\n");
3197 return 0;
3198 }
3199
3200 seq_printf(m, "%ps\n", (void *)*ptr);
3201
3202 return 0;
3203}
3204
3205static const struct seq_operations ftrace_graph_seq_ops = {
3206 .start = g_start,
3207 .next = g_next,
3208 .stop = g_stop,
3209 .show = g_show,
3210};
3211
3212static int
3213ftrace_graph_open(struct inode *inode, struct file *file)
3214{
3215 int ret = 0;
3216
3217 if (unlikely(ftrace_disabled))
3218 return -ENODEV;
3219
3220 mutex_lock(&graph_lock);
3221 if ((file->f_mode & FMODE_WRITE) &&
3222 (file->f_flags & O_TRUNC)) {
3223 ftrace_graph_filter_enabled = 0;
3224 ftrace_graph_count = 0;
3225 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
3226 }
3227 mutex_unlock(&graph_lock);
3228
3229 if (file->f_mode & FMODE_READ)
3230 ret = seq_open(file, &ftrace_graph_seq_ops);
3231
3232 return ret;
3233}
3234
3235static int
3236ftrace_graph_release(struct inode *inode, struct file *file)
3237{
3238 if (file->f_mode & FMODE_READ)
3239 seq_release(inode, file);
3240 return 0;
3241}
3242
3243static int
3244ftrace_set_func(unsigned long *array, int *idx, char *buffer)
3245{
3246 struct dyn_ftrace *rec;
3247 struct ftrace_page *pg;
3248 int search_len;
3249 int fail = 1;
3250 int type, not;
3251 char *search;
3252 bool exists;
3253 int i;
3254
3255 /* decode regex */
3256 type = filter_parse_regex(buffer, strlen(buffer), &search, ¬);
3257 if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
3258 return -EBUSY;
3259
3260 search_len = strlen(search);
3261
3262 mutex_lock(&ftrace_lock);
3263
3264 if (unlikely(ftrace_disabled)) {
3265 mutex_unlock(&ftrace_lock);
3266 return -ENODEV;
3267 }
3268
3269 do_for_each_ftrace_rec(pg, rec) {
3270
3271 if (rec->flags & FTRACE_FL_FREE)
3272 continue;
3273
3274 if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3275 /* if it is in the array */
3276 exists = false;
3277 for (i = 0; i < *idx; i++) {
3278 if (array[i] == rec->ip) {
3279 exists = true;
3280 break;
3281 }
3282 }
3283
3284 if (!not) {
3285 fail = 0;
3286 if (!exists) {
3287 array[(*idx)++] = rec->ip;
3288 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
3289 goto out;
3290 }
3291 } else {
3292 if (exists) {
3293 array[i] = array[--(*idx)];
3294 array[*idx] = 0;
3295 fail = 0;
3296 }
3297 }
3298 }
3299 } while_for_each_ftrace_rec();
3300out:
3301 mutex_unlock(&ftrace_lock);
3302
3303 if (fail)
3304 return -EINVAL;
3305
3306 ftrace_graph_filter_enabled = 1;
3307 return 0;
3308}
3309
3310static ssize_t
3311ftrace_graph_write(struct file *file, const char __user *ubuf,
3312 size_t cnt, loff_t *ppos)
3313{
3314 struct trace_parser parser;
3315 ssize_t read, ret;
3316
3317 if (!cnt)
3318 return 0;
3319
3320 mutex_lock(&graph_lock);
3321
3322 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
3323 ret = -ENOMEM;
3324 goto out_unlock;
3325 }
3326
3327 read = trace_get_user(&parser, ubuf, cnt, ppos);
3328
3329 if (read >= 0 && trace_parser_loaded((&parser))) {
3330 parser.buffer[parser.idx] = 0;
3331
3332 /* we allow only one expression at a time */
3333 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3334 parser.buffer);
3335 if (ret)
3336 goto out_free;
3337 }
3338
3339 ret = read;
3340
3341out_free:
3342 trace_parser_put(&parser);
3343out_unlock:
3344 mutex_unlock(&graph_lock);
3345
3346 return ret;
3347}
3348
3349static const struct file_operations ftrace_graph_fops = {
3350 .open = ftrace_graph_open,
3351 .read = seq_read,
3352 .write = ftrace_graph_write,
3353 .release = ftrace_graph_release,
3354 .llseek = seq_lseek,
3355};
3356#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3357
3358static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
3359{
3360
3361 trace_create_file("available_filter_functions", 0444,
3362 d_tracer, NULL, &ftrace_avail_fops);
3363
3364 trace_create_file("enabled_functions", 0444,
3365 d_tracer, NULL, &ftrace_enabled_fops);
3366
3367 trace_create_file("set_ftrace_filter", 0644, d_tracer,
3368 NULL, &ftrace_filter_fops);
3369
3370 trace_create_file("set_ftrace_notrace", 0644, d_tracer,
3371 NULL, &ftrace_notrace_fops);
3372
3373#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3374 trace_create_file("set_graph_function", 0444, d_tracer,
3375 NULL,
3376 &ftrace_graph_fops);
3377#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3378
3379 return 0;
3380}
3381
3382static int ftrace_process_locs(struct module *mod,
3383 unsigned long *start,
3384 unsigned long *end)
3385{
3386 unsigned long *p;
3387 unsigned long addr;
3388 unsigned long flags = 0; /* Shut up gcc */
3389
3390 mutex_lock(&ftrace_lock);
3391 p = start;
3392 while (p < end) {
3393 addr = ftrace_call_adjust(*p++);
3394 /*
3395 * Some architecture linkers will pad between
3396 * the different mcount_loc sections of different
3397 * object files to satisfy alignments.
3398 * Skip any NULL pointers.
3399 */
3400 if (!addr)
3401 continue;
3402 ftrace_record_ip(addr);
3403 }
3404
3405 /*
3406 * We only need to disable interrupts on start up
3407 * because we are modifying code that an interrupt
3408 * may execute, and the modification is not atomic.
3409 * But for modules, nothing runs the code we modify
3410 * until we are finished with it, and there's no
3411 * reason to cause large interrupt latencies while we do it.
3412 */
3413 if (!mod)
3414 local_irq_save(flags);
3415 ftrace_update_code(mod);
3416 if (!mod)
3417 local_irq_restore(flags);
3418 mutex_unlock(&ftrace_lock);
3419
3420 return 0;
3421}
3422
3423#ifdef CONFIG_MODULES
3424void ftrace_release_mod(struct module *mod)
3425{
3426 struct dyn_ftrace *rec;
3427 struct ftrace_page *pg;
3428
3429 mutex_lock(&ftrace_lock);
3430
3431 if (ftrace_disabled)
3432 goto out_unlock;
3433
3434 do_for_each_ftrace_rec(pg, rec) {
3435 if (within_module_core(rec->ip, mod)) {
3436 /*
3437 * rec->ip is changed in ftrace_free_rec()
3438 * It should not between s and e if record was freed.
3439 */
3440 FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
3441 ftrace_free_rec(rec);
3442 }
3443 } while_for_each_ftrace_rec();
3444 out_unlock:
3445 mutex_unlock(&ftrace_lock);
3446}
3447
3448static void ftrace_init_module(struct module *mod,
3449 unsigned long *start, unsigned long *end)
3450{
3451 if (ftrace_disabled || start == end)
3452 return;
3453 ftrace_process_locs(mod, start, end);
3454}
3455
3456static int ftrace_module_notify(struct notifier_block *self,
3457 unsigned long val, void *data)
3458{
3459 struct module *mod = data;
3460
3461 switch (val) {
3462 case MODULE_STATE_COMING:
3463 ftrace_init_module(mod, mod->ftrace_callsites,
3464 mod->ftrace_callsites +
3465 mod->num_ftrace_callsites);
3466 break;
3467 case MODULE_STATE_GOING:
3468 ftrace_release_mod(mod);
3469 break;
3470 }
3471
3472 return 0;
3473}
3474#else
3475static int ftrace_module_notify(struct notifier_block *self,
3476 unsigned long val, void *data)
3477{
3478 return 0;
3479}
3480#endif /* CONFIG_MODULES */
3481
3482struct notifier_block ftrace_module_nb = {
3483 .notifier_call = ftrace_module_notify,
3484 .priority = 0,
3485};
3486
3487extern unsigned long __start_mcount_loc[];
3488extern unsigned long __stop_mcount_loc[];
3489
3490void __init ftrace_init(void)
3491{
3492 unsigned long count, addr, flags;
3493 int ret;
3494
3495 /* Keep the ftrace pointer to the stub */
3496 addr = (unsigned long)ftrace_stub;
3497
3498 local_irq_save(flags);
3499 ftrace_dyn_arch_init(&addr);
3500 local_irq_restore(flags);
3501
3502 /* ftrace_dyn_arch_init places the return code in addr */
3503 if (addr)
3504 goto failed;
3505
3506 count = __stop_mcount_loc - __start_mcount_loc;
3507
3508 ret = ftrace_dyn_table_alloc(count);
3509 if (ret)
3510 goto failed;
3511
3512 last_ftrace_enabled = ftrace_enabled = 1;
3513
3514 ret = ftrace_process_locs(NULL,
3515 __start_mcount_loc,
3516 __stop_mcount_loc);
3517
3518 ret = register_module_notifier(&ftrace_module_nb);
3519 if (ret)
3520 pr_warning("Failed to register trace ftrace module notifier\n");
3521
3522 set_ftrace_early_filters();
3523
3524 return;
3525 failed:
3526 ftrace_disabled = 1;
3527}
3528
3529#else
3530
3531static struct ftrace_ops global_ops = {
3532 .func = ftrace_stub,
3533};
3534
3535static int __init ftrace_nodyn_init(void)
3536{
3537 ftrace_enabled = 1;
3538 return 0;
3539}
3540device_initcall(ftrace_nodyn_init);
3541
3542static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
3543static inline void ftrace_startup_enable(int command) { }
3544/* Keep as macros so we do not need to define the commands */
3545# define ftrace_startup(ops, command) \
3546 ({ \
3547 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
3548 0; \
3549 })
3550# define ftrace_shutdown(ops, command) do { } while (0)
3551# define ftrace_startup_sysctl() do { } while (0)
3552# define ftrace_shutdown_sysctl() do { } while (0)
3553
3554static inline int
3555ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
3556{
3557 return 1;
3558}
3559
3560#endif /* CONFIG_DYNAMIC_FTRACE */
3561
3562static void
3563ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
3564{
3565 struct ftrace_ops *op;
3566
3567 if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT)))
3568 return;
3569
3570 trace_recursion_set(TRACE_INTERNAL_BIT);
3571 /*
3572 * Some of the ops may be dynamically allocated,
3573 * they must be freed after a synchronize_sched().
3574 */
3575 preempt_disable_notrace();
3576 op = rcu_dereference_raw(ftrace_ops_list);
3577 while (op != &ftrace_list_end) {
3578 if (ftrace_ops_test(op, ip))
3579 op->func(ip, parent_ip);
3580 op = rcu_dereference_raw(op->next);
3581 };
3582 preempt_enable_notrace();
3583 trace_recursion_clear(TRACE_INTERNAL_BIT);
3584}
3585
3586static void clear_ftrace_swapper(void)
3587{
3588 struct task_struct *p;
3589 int cpu;
3590
3591 get_online_cpus();
3592 for_each_online_cpu(cpu) {
3593 p = idle_task(cpu);
3594 clear_tsk_trace_trace(p);
3595 }
3596 put_online_cpus();
3597}
3598
3599static void set_ftrace_swapper(void)
3600{
3601 struct task_struct *p;
3602 int cpu;
3603
3604 get_online_cpus();
3605 for_each_online_cpu(cpu) {
3606 p = idle_task(cpu);
3607 set_tsk_trace_trace(p);
3608 }
3609 put_online_cpus();
3610}
3611
3612static void clear_ftrace_pid(struct pid *pid)
3613{
3614 struct task_struct *p;
3615
3616 rcu_read_lock();
3617 do_each_pid_task(pid, PIDTYPE_PID, p) {
3618 clear_tsk_trace_trace(p);
3619 } while_each_pid_task(pid, PIDTYPE_PID, p);
3620 rcu_read_unlock();
3621
3622 put_pid(pid);
3623}
3624
3625static void set_ftrace_pid(struct pid *pid)
3626{
3627 struct task_struct *p;
3628
3629 rcu_read_lock();
3630 do_each_pid_task(pid, PIDTYPE_PID, p) {
3631 set_tsk_trace_trace(p);
3632 } while_each_pid_task(pid, PIDTYPE_PID, p);
3633 rcu_read_unlock();
3634}
3635
3636static void clear_ftrace_pid_task(struct pid *pid)
3637{
3638 if (pid == ftrace_swapper_pid)
3639 clear_ftrace_swapper();
3640 else
3641 clear_ftrace_pid(pid);
3642}
3643
3644static void set_ftrace_pid_task(struct pid *pid)
3645{
3646 if (pid == ftrace_swapper_pid)
3647 set_ftrace_swapper();
3648 else
3649 set_ftrace_pid(pid);
3650}
3651
3652static int ftrace_pid_add(int p)
3653{
3654 struct pid *pid;
3655 struct ftrace_pid *fpid;
3656 int ret = -EINVAL;
3657
3658 mutex_lock(&ftrace_lock);
3659
3660 if (!p)
3661 pid = ftrace_swapper_pid;
3662 else
3663 pid = find_get_pid(p);
3664
3665 if (!pid)
3666 goto out;
3667
3668 ret = 0;
3669
3670 list_for_each_entry(fpid, &ftrace_pids, list)
3671 if (fpid->pid == pid)
3672 goto out_put;
3673
3674 ret = -ENOMEM;
3675
3676 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
3677 if (!fpid)
3678 goto out_put;
3679
3680 list_add(&fpid->list, &ftrace_pids);
3681 fpid->pid = pid;
3682
3683 set_ftrace_pid_task(pid);
3684
3685 ftrace_update_pid_func();
3686 ftrace_startup_enable(0);
3687
3688 mutex_unlock(&ftrace_lock);
3689 return 0;
3690
3691out_put:
3692 if (pid != ftrace_swapper_pid)
3693 put_pid(pid);
3694
3695out:
3696 mutex_unlock(&ftrace_lock);
3697 return ret;
3698}
3699
3700static void ftrace_pid_reset(void)
3701{
3702 struct ftrace_pid *fpid, *safe;
3703
3704 mutex_lock(&ftrace_lock);
3705 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
3706 struct pid *pid = fpid->pid;
3707
3708 clear_ftrace_pid_task(pid);
3709
3710 list_del(&fpid->list);
3711 kfree(fpid);
3712 }
3713
3714 ftrace_update_pid_func();
3715 ftrace_startup_enable(0);
3716
3717 mutex_unlock(&ftrace_lock);
3718}
3719
3720static void *fpid_start(struct seq_file *m, loff_t *pos)
3721{
3722 mutex_lock(&ftrace_lock);
3723
3724 if (list_empty(&ftrace_pids) && (!*pos))
3725 return (void *) 1;
3726
3727 return seq_list_start(&ftrace_pids, *pos);
3728}
3729
3730static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
3731{
3732 if (v == (void *)1)
3733 return NULL;
3734
3735 return seq_list_next(v, &ftrace_pids, pos);
3736}
3737
3738static void fpid_stop(struct seq_file *m, void *p)
3739{
3740 mutex_unlock(&ftrace_lock);
3741}
3742
3743static int fpid_show(struct seq_file *m, void *v)
3744{
3745 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
3746
3747 if (v == (void *)1) {
3748 seq_printf(m, "no pid\n");
3749 return 0;
3750 }
3751
3752 if (fpid->pid == ftrace_swapper_pid)
3753 seq_printf(m, "swapper tasks\n");
3754 else
3755 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
3756
3757 return 0;
3758}
3759
3760static const struct seq_operations ftrace_pid_sops = {
3761 .start = fpid_start,
3762 .next = fpid_next,
3763 .stop = fpid_stop,
3764 .show = fpid_show,
3765};
3766
3767static int
3768ftrace_pid_open(struct inode *inode, struct file *file)
3769{
3770 int ret = 0;
3771
3772 if ((file->f_mode & FMODE_WRITE) &&
3773 (file->f_flags & O_TRUNC))
3774 ftrace_pid_reset();
3775
3776 if (file->f_mode & FMODE_READ)
3777 ret = seq_open(file, &ftrace_pid_sops);
3778
3779 return ret;
3780}
3781
3782static ssize_t
3783ftrace_pid_write(struct file *filp, const char __user *ubuf,
3784 size_t cnt, loff_t *ppos)
3785{
3786 char buf[64], *tmp;
3787 long val;
3788 int ret;
3789
3790 if (cnt >= sizeof(buf))
3791 return -EINVAL;
3792
3793 if (copy_from_user(&buf, ubuf, cnt))
3794 return -EFAULT;
3795
3796 buf[cnt] = 0;
3797
3798 /*
3799 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
3800 * to clean the filter quietly.
3801 */
3802 tmp = strstrip(buf);
3803 if (strlen(tmp) == 0)
3804 return 1;
3805
3806 ret = strict_strtol(tmp, 10, &val);
3807 if (ret < 0)
3808 return ret;
3809
3810 ret = ftrace_pid_add(val);
3811
3812 return ret ? ret : cnt;
3813}
3814
3815static int
3816ftrace_pid_release(struct inode *inode, struct file *file)
3817{
3818 if (file->f_mode & FMODE_READ)
3819 seq_release(inode, file);
3820
3821 return 0;
3822}
3823
3824static const struct file_operations ftrace_pid_fops = {
3825 .open = ftrace_pid_open,
3826 .write = ftrace_pid_write,
3827 .read = seq_read,
3828 .llseek = seq_lseek,
3829 .release = ftrace_pid_release,
3830};
3831
3832static __init int ftrace_init_debugfs(void)
3833{
3834 struct dentry *d_tracer;
3835
3836 d_tracer = tracing_init_dentry();
3837 if (!d_tracer)
3838 return 0;
3839
3840 ftrace_init_dyn_debugfs(d_tracer);
3841
3842 trace_create_file("set_ftrace_pid", 0644, d_tracer,
3843 NULL, &ftrace_pid_fops);
3844
3845 ftrace_profile_debugfs(d_tracer);
3846
3847 return 0;
3848}
3849fs_initcall(ftrace_init_debugfs);
3850
3851/**
3852 * ftrace_kill - kill ftrace
3853 *
3854 * This function should be used by panic code. It stops ftrace
3855 * but in a not so nice way. If you need to simply kill ftrace
3856 * from a non-atomic section, use ftrace_kill.
3857 */
3858void ftrace_kill(void)
3859{
3860 ftrace_disabled = 1;
3861 ftrace_enabled = 0;
3862 clear_ftrace_function();
3863}
3864
3865/**
3866 * register_ftrace_function - register a function for profiling
3867 * @ops - ops structure that holds the function for profiling.
3868 *
3869 * Register a function to be called by all functions in the
3870 * kernel.
3871 *
3872 * Note: @ops->func and all the functions it calls must be labeled
3873 * with "notrace", otherwise it will go into a
3874 * recursive loop.
3875 */
3876int register_ftrace_function(struct ftrace_ops *ops)
3877{
3878 int ret = -1;
3879
3880 mutex_lock(&ftrace_lock);
3881
3882 if (unlikely(ftrace_disabled))
3883 goto out_unlock;
3884
3885 ret = __register_ftrace_function(ops);
3886 if (!ret)
3887 ret = ftrace_startup(ops, 0);
3888
3889
3890 out_unlock:
3891 mutex_unlock(&ftrace_lock);
3892 return ret;
3893}
3894EXPORT_SYMBOL_GPL(register_ftrace_function);
3895
3896/**
3897 * unregister_ftrace_function - unregister a function for profiling.
3898 * @ops - ops structure that holds the function to unregister
3899 *
3900 * Unregister a function that was added to be called by ftrace profiling.
3901 */
3902int unregister_ftrace_function(struct ftrace_ops *ops)
3903{
3904 int ret;
3905
3906 mutex_lock(&ftrace_lock);
3907 ret = __unregister_ftrace_function(ops);
3908 if (!ret)
3909 ftrace_shutdown(ops, 0);
3910 mutex_unlock(&ftrace_lock);
3911
3912 return ret;
3913}
3914EXPORT_SYMBOL_GPL(unregister_ftrace_function);
3915
3916int
3917ftrace_enable_sysctl(struct ctl_table *table, int write,
3918 void __user *buffer, size_t *lenp,
3919 loff_t *ppos)
3920{
3921 int ret = -ENODEV;
3922
3923 mutex_lock(&ftrace_lock);
3924
3925 if (unlikely(ftrace_disabled))
3926 goto out;
3927
3928 ret = proc_dointvec(table, write, buffer, lenp, ppos);
3929
3930 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3931 goto out;
3932
3933 last_ftrace_enabled = !!ftrace_enabled;
3934
3935 if (ftrace_enabled) {
3936
3937 ftrace_startup_sysctl();
3938
3939 /* we are starting ftrace again */
3940 if (ftrace_ops_list != &ftrace_list_end) {
3941 if (ftrace_ops_list->next == &ftrace_list_end)
3942 ftrace_trace_function = ftrace_ops_list->func;
3943 else
3944 ftrace_trace_function = ftrace_ops_list_func;
3945 }
3946
3947 } else {
3948 /* stopping ftrace calls (just send to ftrace_stub) */
3949 ftrace_trace_function = ftrace_stub;
3950
3951 ftrace_shutdown_sysctl();
3952 }
3953
3954 out:
3955 mutex_unlock(&ftrace_lock);
3956 return ret;
3957}
3958
3959#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3960
3961static int ftrace_graph_active;
3962static struct notifier_block ftrace_suspend_notifier;
3963
3964int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3965{
3966 return 0;
3967}
3968
3969/* The callbacks that hook a function */
3970trace_func_graph_ret_t ftrace_graph_return =
3971 (trace_func_graph_ret_t)ftrace_stub;
3972trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3973
3974/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3975static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3976{
3977 int i;
3978 int ret = 0;
3979 unsigned long flags;
3980 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3981 struct task_struct *g, *t;
3982
3983 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3984 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3985 * sizeof(struct ftrace_ret_stack),
3986 GFP_KERNEL);
3987 if (!ret_stack_list[i]) {
3988 start = 0;
3989 end = i;
3990 ret = -ENOMEM;
3991 goto free;
3992 }
3993 }
3994
3995 read_lock_irqsave(&tasklist_lock, flags);
3996 do_each_thread(g, t) {
3997 if (start == end) {
3998 ret = -EAGAIN;
3999 goto unlock;
4000 }
4001
4002 if (t->ret_stack == NULL) {
4003 atomic_set(&t->tracing_graph_pause, 0);
4004 atomic_set(&t->trace_overrun, 0);
4005 t->curr_ret_stack = -1;
4006 /* Make sure the tasks see the -1 first: */
4007 smp_wmb();
4008 t->ret_stack = ret_stack_list[start++];
4009 }
4010 } while_each_thread(g, t);
4011
4012unlock:
4013 read_unlock_irqrestore(&tasklist_lock, flags);
4014free:
4015 for (i = start; i < end; i++)
4016 kfree(ret_stack_list[i]);
4017 return ret;
4018}
4019
4020static void
4021ftrace_graph_probe_sched_switch(void *ignore,
4022 struct task_struct *prev, struct task_struct *next)
4023{
4024 unsigned long long timestamp;
4025 int index;
4026
4027 /*
4028 * Does the user want to count the time a function was asleep.
4029 * If so, do not update the time stamps.
4030 */
4031 if (trace_flags & TRACE_ITER_SLEEP_TIME)
4032 return;
4033
4034 timestamp = trace_clock_local();
4035
4036 prev->ftrace_timestamp = timestamp;
4037
4038 /* only process tasks that we timestamped */
4039 if (!next->ftrace_timestamp)
4040 return;
4041
4042 /*
4043 * Update all the counters in next to make up for the
4044 * time next was sleeping.
4045 */
4046 timestamp -= next->ftrace_timestamp;
4047
4048 for (index = next->curr_ret_stack; index >= 0; index--)
4049 next->ret_stack[index].calltime += timestamp;
4050}
4051
4052/* Allocate a return stack for each task */
4053static int start_graph_tracing(void)
4054{
4055 struct ftrace_ret_stack **ret_stack_list;
4056 int ret, cpu;
4057
4058 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
4059 sizeof(struct ftrace_ret_stack *),
4060 GFP_KERNEL);
4061
4062 if (!ret_stack_list)
4063 return -ENOMEM;
4064
4065 /* The cpu_boot init_task->ret_stack will never be freed */
4066 for_each_online_cpu(cpu) {
4067 if (!idle_task(cpu)->ret_stack)
4068 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4069 }
4070
4071 do {
4072 ret = alloc_retstack_tasklist(ret_stack_list);
4073 } while (ret == -EAGAIN);
4074
4075 if (!ret) {
4076 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4077 if (ret)
4078 pr_info("ftrace_graph: Couldn't activate tracepoint"
4079 " probe to kernel_sched_switch\n");
4080 }
4081
4082 kfree(ret_stack_list);
4083 return ret;
4084}
4085
4086/*
4087 * Hibernation protection.
4088 * The state of the current task is too much unstable during
4089 * suspend/restore to disk. We want to protect against that.
4090 */
4091static int
4092ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4093 void *unused)
4094{
4095 switch (state) {
4096 case PM_HIBERNATION_PREPARE:
4097 pause_graph_tracing();
4098 break;
4099
4100 case PM_POST_HIBERNATION:
4101 unpause_graph_tracing();
4102 break;
4103 }
4104 return NOTIFY_DONE;
4105}
4106
4107int register_ftrace_graph(trace_func_graph_ret_t retfunc,
4108 trace_func_graph_ent_t entryfunc)
4109{
4110 int ret = 0;
4111
4112 mutex_lock(&ftrace_lock);
4113
4114 /* we currently allow only one tracer registered at a time */
4115 if (ftrace_graph_active) {
4116 ret = -EBUSY;
4117 goto out;
4118 }
4119
4120 ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
4121 register_pm_notifier(&ftrace_suspend_notifier);
4122
4123 ftrace_graph_active++;
4124 ret = start_graph_tracing();
4125 if (ret) {
4126 ftrace_graph_active--;
4127 goto out;
4128 }
4129
4130 ftrace_graph_return = retfunc;
4131 ftrace_graph_entry = entryfunc;
4132
4133 ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
4134
4135out:
4136 mutex_unlock(&ftrace_lock);
4137 return ret;
4138}
4139
4140void unregister_ftrace_graph(void)
4141{
4142 mutex_lock(&ftrace_lock);
4143
4144 if (unlikely(!ftrace_graph_active))
4145 goto out;
4146
4147 ftrace_graph_active--;
4148 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
4149 ftrace_graph_entry = ftrace_graph_entry_stub;
4150 ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
4151 unregister_pm_notifier(&ftrace_suspend_notifier);
4152 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4153
4154 out:
4155 mutex_unlock(&ftrace_lock);
4156}
4157
4158static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
4159
4160static void
4161graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
4162{
4163 atomic_set(&t->tracing_graph_pause, 0);
4164 atomic_set(&t->trace_overrun, 0);
4165 t->ftrace_timestamp = 0;
4166 /* make curr_ret_stack visible before we add the ret_stack */
4167 smp_wmb();
4168 t->ret_stack = ret_stack;
4169}
4170
4171/*
4172 * Allocate a return stack for the idle task. May be the first
4173 * time through, or it may be done by CPU hotplug online.
4174 */
4175void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
4176{
4177 t->curr_ret_stack = -1;
4178 /*
4179 * The idle task has no parent, it either has its own
4180 * stack or no stack at all.
4181 */
4182 if (t->ret_stack)
4183 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
4184
4185 if (ftrace_graph_active) {
4186 struct ftrace_ret_stack *ret_stack;
4187
4188 ret_stack = per_cpu(idle_ret_stack, cpu);
4189 if (!ret_stack) {
4190 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4191 * sizeof(struct ftrace_ret_stack),
4192 GFP_KERNEL);
4193 if (!ret_stack)
4194 return;
4195 per_cpu(idle_ret_stack, cpu) = ret_stack;
4196 }
4197 graph_init_task(t, ret_stack);
4198 }
4199}
4200
4201/* Allocate a return stack for newly created task */
4202void ftrace_graph_init_task(struct task_struct *t)
4203{
4204 /* Make sure we do not use the parent ret_stack */
4205 t->ret_stack = NULL;
4206 t->curr_ret_stack = -1;
4207
4208 if (ftrace_graph_active) {
4209 struct ftrace_ret_stack *ret_stack;
4210
4211 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4212 * sizeof(struct ftrace_ret_stack),
4213 GFP_KERNEL);
4214 if (!ret_stack)
4215 return;
4216 graph_init_task(t, ret_stack);
4217 }
4218}
4219
4220void ftrace_graph_exit_task(struct task_struct *t)
4221{
4222 struct ftrace_ret_stack *ret_stack = t->ret_stack;
4223
4224 t->ret_stack = NULL;
4225 /* NULL must become visible to IRQs before we free it: */
4226 barrier();
4227
4228 kfree(ret_stack);
4229}
4230
4231void ftrace_graph_stop(void)
4232{
4233 ftrace_stop();
4234}
4235#endif