writeback: split writeback_inodes_wb
[linux-2.6/next.git] / kernel / trace / ftrace.c
blob6d2cb14f9449083c9a2e78f507b9c1255c8e7ca2
1 /*
2 * Infrastructure for profiling code inserted by 'gcc -pg'.
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
7 * Originally ported from the -rt patch by:
8 * Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code in the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 William Lee Irwin III
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>
33 #include <trace/events/sched.h>
35 #include <asm/ftrace.h>
36 #include <asm/setup.h>
38 #include "trace_output.h"
39 #include "trace_stat.h"
41 #define FTRACE_WARN_ON(cond) \
42 do { \
43 if (WARN_ON(cond)) \
44 ftrace_kill(); \
45 } while (0)
47 #define FTRACE_WARN_ON_ONCE(cond) \
48 do { \
49 if (WARN_ON_ONCE(cond)) \
50 ftrace_kill(); \
51 } while (0)
53 /* hash bits for specific function selection */
54 #define FTRACE_HASH_BITS 7
55 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
57 /* ftrace_enabled is a method to turn ftrace on or off */
58 int ftrace_enabled __read_mostly;
59 static int last_ftrace_enabled;
61 /* Quick disabling of function tracer. */
62 int function_trace_stop;
64 /* List for set_ftrace_pid's pids. */
65 LIST_HEAD(ftrace_pids);
66 struct ftrace_pid {
67 struct list_head list;
68 struct pid *pid;
72 * ftrace_disabled is set when an anomaly is discovered.
73 * ftrace_disabled is much stronger than ftrace_enabled.
75 static int ftrace_disabled __read_mostly;
77 static DEFINE_MUTEX(ftrace_lock);
79 static struct ftrace_ops ftrace_list_end __read_mostly =
81 .func = ftrace_stub,
84 static struct ftrace_ops *ftrace_list __read_mostly = &ftrace_list_end;
85 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
86 ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
87 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
90 * Traverse the ftrace_list, invoking all entries. The reason that we
91 * can use rcu_dereference_raw() is that elements removed from this list
92 * are simply leaked, so there is no need to interact with a grace-period
93 * mechanism. The rcu_dereference_raw() calls are needed to handle
94 * concurrent insertions into the ftrace_list.
96 * Silly Alpha and silly pointer-speculation compiler optimizations!
98 static void ftrace_list_func(unsigned long ip, unsigned long parent_ip)
100 struct ftrace_ops *op = rcu_dereference_raw(ftrace_list); /*see above*/
102 while (op != &ftrace_list_end) {
103 op->func(ip, parent_ip);
104 op = rcu_dereference_raw(op->next); /*see above*/
108 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
110 if (!test_tsk_trace_trace(current))
111 return;
113 ftrace_pid_function(ip, parent_ip);
116 static void set_ftrace_pid_function(ftrace_func_t func)
118 /* do not set ftrace_pid_function to itself! */
119 if (func != ftrace_pid_func)
120 ftrace_pid_function = func;
124 * clear_ftrace_function - reset the ftrace function
126 * This NULLs the ftrace function and in essence stops
127 * tracing. There may be lag
129 void clear_ftrace_function(void)
131 ftrace_trace_function = ftrace_stub;
132 __ftrace_trace_function = ftrace_stub;
133 ftrace_pid_function = ftrace_stub;
136 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
138 * For those archs that do not test ftrace_trace_stop in their
139 * mcount call site, we need to do it from C.
141 static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
143 if (function_trace_stop)
144 return;
146 __ftrace_trace_function(ip, parent_ip);
148 #endif
150 static int __register_ftrace_function(struct ftrace_ops *ops)
152 ops->next = ftrace_list;
154 * We are entering ops into the ftrace_list but another
155 * CPU might be walking that list. We need to make sure
156 * the ops->next pointer is valid before another CPU sees
157 * the ops pointer included into the ftrace_list.
159 rcu_assign_pointer(ftrace_list, ops);
161 if (ftrace_enabled) {
162 ftrace_func_t func;
164 if (ops->next == &ftrace_list_end)
165 func = ops->func;
166 else
167 func = ftrace_list_func;
169 if (!list_empty(&ftrace_pids)) {
170 set_ftrace_pid_function(func);
171 func = ftrace_pid_func;
175 * For one func, simply call it directly.
176 * For more than one func, call the chain.
178 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
179 ftrace_trace_function = func;
180 #else
181 __ftrace_trace_function = func;
182 ftrace_trace_function = ftrace_test_stop_func;
183 #endif
186 return 0;
189 static int __unregister_ftrace_function(struct ftrace_ops *ops)
191 struct ftrace_ops **p;
194 * If we are removing the last function, then simply point
195 * to the ftrace_stub.
197 if (ftrace_list == ops && ops->next == &ftrace_list_end) {
198 ftrace_trace_function = ftrace_stub;
199 ftrace_list = &ftrace_list_end;
200 return 0;
203 for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next)
204 if (*p == ops)
205 break;
207 if (*p != ops)
208 return -1;
210 *p = (*p)->next;
212 if (ftrace_enabled) {
213 /* If we only have one func left, then call that directly */
214 if (ftrace_list->next == &ftrace_list_end) {
215 ftrace_func_t func = ftrace_list->func;
217 if (!list_empty(&ftrace_pids)) {
218 set_ftrace_pid_function(func);
219 func = ftrace_pid_func;
221 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
222 ftrace_trace_function = func;
223 #else
224 __ftrace_trace_function = func;
225 #endif
229 return 0;
232 static void ftrace_update_pid_func(void)
234 ftrace_func_t func;
236 if (ftrace_trace_function == ftrace_stub)
237 return;
239 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
240 func = ftrace_trace_function;
241 #else
242 func = __ftrace_trace_function;
243 #endif
245 if (!list_empty(&ftrace_pids)) {
246 set_ftrace_pid_function(func);
247 func = ftrace_pid_func;
248 } else {
249 if (func == ftrace_pid_func)
250 func = ftrace_pid_function;
253 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
254 ftrace_trace_function = func;
255 #else
256 __ftrace_trace_function = func;
257 #endif
260 #ifdef CONFIG_FUNCTION_PROFILER
261 struct ftrace_profile {
262 struct hlist_node node;
263 unsigned long ip;
264 unsigned long counter;
265 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
266 unsigned long long time;
267 unsigned long long time_squared;
268 #endif
271 struct ftrace_profile_page {
272 struct ftrace_profile_page *next;
273 unsigned long index;
274 struct ftrace_profile records[];
277 struct ftrace_profile_stat {
278 atomic_t disabled;
279 struct hlist_head *hash;
280 struct ftrace_profile_page *pages;
281 struct ftrace_profile_page *start;
282 struct tracer_stat stat;
285 #define PROFILE_RECORDS_SIZE \
286 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
288 #define PROFILES_PER_PAGE \
289 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
291 static int ftrace_profile_bits __read_mostly;
292 static int ftrace_profile_enabled __read_mostly;
294 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
295 static DEFINE_MUTEX(ftrace_profile_lock);
297 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
299 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
301 static void *
302 function_stat_next(void *v, int idx)
304 struct ftrace_profile *rec = v;
305 struct ftrace_profile_page *pg;
307 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
309 again:
310 if (idx != 0)
311 rec++;
313 if ((void *)rec >= (void *)&pg->records[pg->index]) {
314 pg = pg->next;
315 if (!pg)
316 return NULL;
317 rec = &pg->records[0];
318 if (!rec->counter)
319 goto again;
322 return rec;
325 static void *function_stat_start(struct tracer_stat *trace)
327 struct ftrace_profile_stat *stat =
328 container_of(trace, struct ftrace_profile_stat, stat);
330 if (!stat || !stat->start)
331 return NULL;
333 return function_stat_next(&stat->start->records[0], 0);
336 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
337 /* function graph compares on total time */
338 static int function_stat_cmp(void *p1, void *p2)
340 struct ftrace_profile *a = p1;
341 struct ftrace_profile *b = p2;
343 if (a->time < b->time)
344 return -1;
345 if (a->time > b->time)
346 return 1;
347 else
348 return 0;
350 #else
351 /* not function graph compares against hits */
352 static int function_stat_cmp(void *p1, void *p2)
354 struct ftrace_profile *a = p1;
355 struct ftrace_profile *b = p2;
357 if (a->counter < b->counter)
358 return -1;
359 if (a->counter > b->counter)
360 return 1;
361 else
362 return 0;
364 #endif
366 static int function_stat_headers(struct seq_file *m)
368 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
369 seq_printf(m, " Function "
370 "Hit Time Avg s^2\n"
371 " -------- "
372 "--- ---- --- ---\n");
373 #else
374 seq_printf(m, " Function Hit\n"
375 " -------- ---\n");
376 #endif
377 return 0;
380 static int function_stat_show(struct seq_file *m, void *v)
382 struct ftrace_profile *rec = v;
383 char str[KSYM_SYMBOL_LEN];
384 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
385 static DEFINE_MUTEX(mutex);
386 static struct trace_seq s;
387 unsigned long long avg;
388 unsigned long long stddev;
389 #endif
391 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
392 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
394 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
395 seq_printf(m, " ");
396 avg = rec->time;
397 do_div(avg, rec->counter);
399 /* Sample standard deviation (s^2) */
400 if (rec->counter <= 1)
401 stddev = 0;
402 else {
403 stddev = rec->time_squared - rec->counter * avg * avg;
405 * Divide only 1000 for ns^2 -> us^2 conversion.
406 * trace_print_graph_duration will divide 1000 again.
408 do_div(stddev, (rec->counter - 1) * 1000);
411 mutex_lock(&mutex);
412 trace_seq_init(&s);
413 trace_print_graph_duration(rec->time, &s);
414 trace_seq_puts(&s, " ");
415 trace_print_graph_duration(avg, &s);
416 trace_seq_puts(&s, " ");
417 trace_print_graph_duration(stddev, &s);
418 trace_print_seq(m, &s);
419 mutex_unlock(&mutex);
420 #endif
421 seq_putc(m, '\n');
423 return 0;
426 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
428 struct ftrace_profile_page *pg;
430 pg = stat->pages = stat->start;
432 while (pg) {
433 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
434 pg->index = 0;
435 pg = pg->next;
438 memset(stat->hash, 0,
439 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
442 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
444 struct ftrace_profile_page *pg;
445 int functions;
446 int pages;
447 int i;
449 /* If we already allocated, do nothing */
450 if (stat->pages)
451 return 0;
453 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
454 if (!stat->pages)
455 return -ENOMEM;
457 #ifdef CONFIG_DYNAMIC_FTRACE
458 functions = ftrace_update_tot_cnt;
459 #else
461 * We do not know the number of functions that exist because
462 * dynamic tracing is what counts them. With past experience
463 * we have around 20K functions. That should be more than enough.
464 * It is highly unlikely we will execute every function in
465 * the kernel.
467 functions = 20000;
468 #endif
470 pg = stat->start = stat->pages;
472 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
474 for (i = 0; i < pages; i++) {
475 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
476 if (!pg->next)
477 goto out_free;
478 pg = pg->next;
481 return 0;
483 out_free:
484 pg = stat->start;
485 while (pg) {
486 unsigned long tmp = (unsigned long)pg;
488 pg = pg->next;
489 free_page(tmp);
492 free_page((unsigned long)stat->pages);
493 stat->pages = NULL;
494 stat->start = NULL;
496 return -ENOMEM;
499 static int ftrace_profile_init_cpu(int cpu)
501 struct ftrace_profile_stat *stat;
502 int size;
504 stat = &per_cpu(ftrace_profile_stats, cpu);
506 if (stat->hash) {
507 /* If the profile is already created, simply reset it */
508 ftrace_profile_reset(stat);
509 return 0;
513 * We are profiling all functions, but usually only a few thousand
514 * functions are hit. We'll make a hash of 1024 items.
516 size = FTRACE_PROFILE_HASH_SIZE;
518 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
520 if (!stat->hash)
521 return -ENOMEM;
523 if (!ftrace_profile_bits) {
524 size--;
526 for (; size; size >>= 1)
527 ftrace_profile_bits++;
530 /* Preallocate the function profiling pages */
531 if (ftrace_profile_pages_init(stat) < 0) {
532 kfree(stat->hash);
533 stat->hash = NULL;
534 return -ENOMEM;
537 return 0;
540 static int ftrace_profile_init(void)
542 int cpu;
543 int ret = 0;
545 for_each_online_cpu(cpu) {
546 ret = ftrace_profile_init_cpu(cpu);
547 if (ret)
548 break;
551 return ret;
554 /* interrupts must be disabled */
555 static struct ftrace_profile *
556 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
558 struct ftrace_profile *rec;
559 struct hlist_head *hhd;
560 struct hlist_node *n;
561 unsigned long key;
563 key = hash_long(ip, ftrace_profile_bits);
564 hhd = &stat->hash[key];
566 if (hlist_empty(hhd))
567 return NULL;
569 hlist_for_each_entry_rcu(rec, n, hhd, node) {
570 if (rec->ip == ip)
571 return rec;
574 return NULL;
577 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
578 struct ftrace_profile *rec)
580 unsigned long key;
582 key = hash_long(rec->ip, ftrace_profile_bits);
583 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
587 * The memory is already allocated, this simply finds a new record to use.
589 static struct ftrace_profile *
590 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
592 struct ftrace_profile *rec = NULL;
594 /* prevent recursion (from NMIs) */
595 if (atomic_inc_return(&stat->disabled) != 1)
596 goto out;
599 * Try to find the function again since an NMI
600 * could have added it
602 rec = ftrace_find_profiled_func(stat, ip);
603 if (rec)
604 goto out;
606 if (stat->pages->index == PROFILES_PER_PAGE) {
607 if (!stat->pages->next)
608 goto out;
609 stat->pages = stat->pages->next;
612 rec = &stat->pages->records[stat->pages->index++];
613 rec->ip = ip;
614 ftrace_add_profile(stat, rec);
616 out:
617 atomic_dec(&stat->disabled);
619 return rec;
622 static void
623 function_profile_call(unsigned long ip, unsigned long parent_ip)
625 struct ftrace_profile_stat *stat;
626 struct ftrace_profile *rec;
627 unsigned long flags;
629 if (!ftrace_profile_enabled)
630 return;
632 local_irq_save(flags);
634 stat = &__get_cpu_var(ftrace_profile_stats);
635 if (!stat->hash || !ftrace_profile_enabled)
636 goto out;
638 rec = ftrace_find_profiled_func(stat, ip);
639 if (!rec) {
640 rec = ftrace_profile_alloc(stat, ip);
641 if (!rec)
642 goto out;
645 rec->counter++;
646 out:
647 local_irq_restore(flags);
650 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
651 static int profile_graph_entry(struct ftrace_graph_ent *trace)
653 function_profile_call(trace->func, 0);
654 return 1;
657 static void profile_graph_return(struct ftrace_graph_ret *trace)
659 struct ftrace_profile_stat *stat;
660 unsigned long long calltime;
661 struct ftrace_profile *rec;
662 unsigned long flags;
664 local_irq_save(flags);
665 stat = &__get_cpu_var(ftrace_profile_stats);
666 if (!stat->hash || !ftrace_profile_enabled)
667 goto out;
669 /* If the calltime was zero'd ignore it */
670 if (!trace->calltime)
671 goto out;
673 calltime = trace->rettime - trace->calltime;
675 if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
676 int index;
678 index = trace->depth;
680 /* Append this call time to the parent time to subtract */
681 if (index)
682 current->ret_stack[index - 1].subtime += calltime;
684 if (current->ret_stack[index].subtime < calltime)
685 calltime -= current->ret_stack[index].subtime;
686 else
687 calltime = 0;
690 rec = ftrace_find_profiled_func(stat, trace->func);
691 if (rec) {
692 rec->time += calltime;
693 rec->time_squared += calltime * calltime;
696 out:
697 local_irq_restore(flags);
700 static int register_ftrace_profiler(void)
702 return register_ftrace_graph(&profile_graph_return,
703 &profile_graph_entry);
706 static void unregister_ftrace_profiler(void)
708 unregister_ftrace_graph();
710 #else
711 static struct ftrace_ops ftrace_profile_ops __read_mostly =
713 .func = function_profile_call,
716 static int register_ftrace_profiler(void)
718 return register_ftrace_function(&ftrace_profile_ops);
721 static void unregister_ftrace_profiler(void)
723 unregister_ftrace_function(&ftrace_profile_ops);
725 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
727 static ssize_t
728 ftrace_profile_write(struct file *filp, const char __user *ubuf,
729 size_t cnt, loff_t *ppos)
731 unsigned long val;
732 char buf[64]; /* big enough to hold a number */
733 int ret;
735 if (cnt >= sizeof(buf))
736 return -EINVAL;
738 if (copy_from_user(&buf, ubuf, cnt))
739 return -EFAULT;
741 buf[cnt] = 0;
743 ret = strict_strtoul(buf, 10, &val);
744 if (ret < 0)
745 return ret;
747 val = !!val;
749 mutex_lock(&ftrace_profile_lock);
750 if (ftrace_profile_enabled ^ val) {
751 if (val) {
752 ret = ftrace_profile_init();
753 if (ret < 0) {
754 cnt = ret;
755 goto out;
758 ret = register_ftrace_profiler();
759 if (ret < 0) {
760 cnt = ret;
761 goto out;
763 ftrace_profile_enabled = 1;
764 } else {
765 ftrace_profile_enabled = 0;
767 * unregister_ftrace_profiler calls stop_machine
768 * so this acts like an synchronize_sched.
770 unregister_ftrace_profiler();
773 out:
774 mutex_unlock(&ftrace_profile_lock);
776 *ppos += cnt;
778 return cnt;
781 static ssize_t
782 ftrace_profile_read(struct file *filp, char __user *ubuf,
783 size_t cnt, loff_t *ppos)
785 char buf[64]; /* big enough to hold a number */
786 int r;
788 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
789 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
792 static const struct file_operations ftrace_profile_fops = {
793 .open = tracing_open_generic,
794 .read = ftrace_profile_read,
795 .write = ftrace_profile_write,
798 /* used to initialize the real stat files */
799 static struct tracer_stat function_stats __initdata = {
800 .name = "functions",
801 .stat_start = function_stat_start,
802 .stat_next = function_stat_next,
803 .stat_cmp = function_stat_cmp,
804 .stat_headers = function_stat_headers,
805 .stat_show = function_stat_show
808 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
810 struct ftrace_profile_stat *stat;
811 struct dentry *entry;
812 char *name;
813 int ret;
814 int cpu;
816 for_each_possible_cpu(cpu) {
817 stat = &per_cpu(ftrace_profile_stats, cpu);
819 /* allocate enough for function name + cpu number */
820 name = kmalloc(32, GFP_KERNEL);
821 if (!name) {
823 * The files created are permanent, if something happens
824 * we still do not free memory.
826 WARN(1,
827 "Could not allocate stat file for cpu %d\n",
828 cpu);
829 return;
831 stat->stat = function_stats;
832 snprintf(name, 32, "function%d", cpu);
833 stat->stat.name = name;
834 ret = register_stat_tracer(&stat->stat);
835 if (ret) {
836 WARN(1,
837 "Could not register function stat for cpu %d\n",
838 cpu);
839 kfree(name);
840 return;
844 entry = debugfs_create_file("function_profile_enabled", 0644,
845 d_tracer, NULL, &ftrace_profile_fops);
846 if (!entry)
847 pr_warning("Could not create debugfs "
848 "'function_profile_enabled' entry\n");
851 #else /* CONFIG_FUNCTION_PROFILER */
852 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
855 #endif /* CONFIG_FUNCTION_PROFILER */
857 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
859 #ifdef CONFIG_DYNAMIC_FTRACE
861 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
862 # error Dynamic ftrace depends on MCOUNT_RECORD
863 #endif
865 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
867 struct ftrace_func_probe {
868 struct hlist_node node;
869 struct ftrace_probe_ops *ops;
870 unsigned long flags;
871 unsigned long ip;
872 void *data;
873 struct rcu_head rcu;
876 enum {
877 FTRACE_ENABLE_CALLS = (1 << 0),
878 FTRACE_DISABLE_CALLS = (1 << 1),
879 FTRACE_UPDATE_TRACE_FUNC = (1 << 2),
880 FTRACE_ENABLE_MCOUNT = (1 << 3),
881 FTRACE_DISABLE_MCOUNT = (1 << 4),
882 FTRACE_START_FUNC_RET = (1 << 5),
883 FTRACE_STOP_FUNC_RET = (1 << 6),
886 static int ftrace_filtered;
888 static struct dyn_ftrace *ftrace_new_addrs;
890 static DEFINE_MUTEX(ftrace_regex_lock);
892 struct ftrace_page {
893 struct ftrace_page *next;
894 int index;
895 struct dyn_ftrace records[];
898 #define ENTRIES_PER_PAGE \
899 ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
901 /* estimate from running different kernels */
902 #define NR_TO_INIT 10000
904 static struct ftrace_page *ftrace_pages_start;
905 static struct ftrace_page *ftrace_pages;
907 static struct dyn_ftrace *ftrace_free_records;
910 * This is a double for. Do not use 'break' to break out of the loop,
911 * you must use a goto.
913 #define do_for_each_ftrace_rec(pg, rec) \
914 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
915 int _____i; \
916 for (_____i = 0; _____i < pg->index; _____i++) { \
917 rec = &pg->records[_____i];
919 #define while_for_each_ftrace_rec() \
923 static void ftrace_free_rec(struct dyn_ftrace *rec)
925 rec->freelist = ftrace_free_records;
926 ftrace_free_records = rec;
927 rec->flags |= FTRACE_FL_FREE;
930 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
932 struct dyn_ftrace *rec;
934 /* First check for freed records */
935 if (ftrace_free_records) {
936 rec = ftrace_free_records;
938 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
939 FTRACE_WARN_ON_ONCE(1);
940 ftrace_free_records = NULL;
941 return NULL;
944 ftrace_free_records = rec->freelist;
945 memset(rec, 0, sizeof(*rec));
946 return rec;
949 if (ftrace_pages->index == ENTRIES_PER_PAGE) {
950 if (!ftrace_pages->next) {
951 /* allocate another page */
952 ftrace_pages->next =
953 (void *)get_zeroed_page(GFP_KERNEL);
954 if (!ftrace_pages->next)
955 return NULL;
957 ftrace_pages = ftrace_pages->next;
960 return &ftrace_pages->records[ftrace_pages->index++];
963 static struct dyn_ftrace *
964 ftrace_record_ip(unsigned long ip)
966 struct dyn_ftrace *rec;
968 if (ftrace_disabled)
969 return NULL;
971 rec = ftrace_alloc_dyn_node(ip);
972 if (!rec)
973 return NULL;
975 rec->ip = ip;
976 rec->newlist = ftrace_new_addrs;
977 ftrace_new_addrs = rec;
979 return rec;
982 static void print_ip_ins(const char *fmt, unsigned char *p)
984 int i;
986 printk(KERN_CONT "%s", fmt);
988 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
989 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
992 static void ftrace_bug(int failed, unsigned long ip)
994 switch (failed) {
995 case -EFAULT:
996 FTRACE_WARN_ON_ONCE(1);
997 pr_info("ftrace faulted on modifying ");
998 print_ip_sym(ip);
999 break;
1000 case -EINVAL:
1001 FTRACE_WARN_ON_ONCE(1);
1002 pr_info("ftrace failed to modify ");
1003 print_ip_sym(ip);
1004 print_ip_ins(" actual: ", (unsigned char *)ip);
1005 printk(KERN_CONT "\n");
1006 break;
1007 case -EPERM:
1008 FTRACE_WARN_ON_ONCE(1);
1009 pr_info("ftrace faulted on writing ");
1010 print_ip_sym(ip);
1011 break;
1012 default:
1013 FTRACE_WARN_ON_ONCE(1);
1014 pr_info("ftrace faulted on unknown error ");
1015 print_ip_sym(ip);
1020 /* Return 1 if the address range is reserved for ftrace */
1021 int ftrace_text_reserved(void *start, void *end)
1023 struct dyn_ftrace *rec;
1024 struct ftrace_page *pg;
1026 do_for_each_ftrace_rec(pg, rec) {
1027 if (rec->ip <= (unsigned long)end &&
1028 rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1029 return 1;
1030 } while_for_each_ftrace_rec();
1031 return 0;
1035 static int
1036 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1038 unsigned long ftrace_addr;
1039 unsigned long flag = 0UL;
1041 ftrace_addr = (unsigned long)FTRACE_ADDR;
1044 * If this record is not to be traced or we want to disable it,
1045 * then disable it.
1047 * If we want to enable it and filtering is off, then enable it.
1049 * If we want to enable it and filtering is on, enable it only if
1050 * it's filtered
1052 if (enable && !(rec->flags & FTRACE_FL_NOTRACE)) {
1053 if (!ftrace_filtered || (rec->flags & FTRACE_FL_FILTER))
1054 flag = FTRACE_FL_ENABLED;
1057 /* If the state of this record hasn't changed, then do nothing */
1058 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1059 return 0;
1061 if (flag) {
1062 rec->flags |= FTRACE_FL_ENABLED;
1063 return ftrace_make_call(rec, ftrace_addr);
1066 rec->flags &= ~FTRACE_FL_ENABLED;
1067 return ftrace_make_nop(NULL, rec, ftrace_addr);
1070 static void ftrace_replace_code(int enable)
1072 struct dyn_ftrace *rec;
1073 struct ftrace_page *pg;
1074 int failed;
1076 do_for_each_ftrace_rec(pg, rec) {
1078 * Skip over free records, records that have
1079 * failed and not converted.
1081 if (rec->flags & FTRACE_FL_FREE ||
1082 rec->flags & FTRACE_FL_FAILED ||
1083 !(rec->flags & FTRACE_FL_CONVERTED))
1084 continue;
1086 failed = __ftrace_replace_code(rec, enable);
1087 if (failed) {
1088 rec->flags |= FTRACE_FL_FAILED;
1089 ftrace_bug(failed, rec->ip);
1090 /* Stop processing */
1091 return;
1093 } while_for_each_ftrace_rec();
1096 static int
1097 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1099 unsigned long ip;
1100 int ret;
1102 ip = rec->ip;
1104 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1105 if (ret) {
1106 ftrace_bug(ret, ip);
1107 rec->flags |= FTRACE_FL_FAILED;
1108 return 0;
1110 return 1;
1114 * archs can override this function if they must do something
1115 * before the modifying code is performed.
1117 int __weak ftrace_arch_code_modify_prepare(void)
1119 return 0;
1123 * archs can override this function if they must do something
1124 * after the modifying code is performed.
1126 int __weak ftrace_arch_code_modify_post_process(void)
1128 return 0;
1131 static int __ftrace_modify_code(void *data)
1133 int *command = data;
1135 if (*command & FTRACE_ENABLE_CALLS)
1136 ftrace_replace_code(1);
1137 else if (*command & FTRACE_DISABLE_CALLS)
1138 ftrace_replace_code(0);
1140 if (*command & FTRACE_UPDATE_TRACE_FUNC)
1141 ftrace_update_ftrace_func(ftrace_trace_function);
1143 if (*command & FTRACE_START_FUNC_RET)
1144 ftrace_enable_ftrace_graph_caller();
1145 else if (*command & FTRACE_STOP_FUNC_RET)
1146 ftrace_disable_ftrace_graph_caller();
1148 return 0;
1151 static void ftrace_run_update_code(int command)
1153 int ret;
1155 ret = ftrace_arch_code_modify_prepare();
1156 FTRACE_WARN_ON(ret);
1157 if (ret)
1158 return;
1160 stop_machine(__ftrace_modify_code, &command, NULL);
1162 ret = ftrace_arch_code_modify_post_process();
1163 FTRACE_WARN_ON(ret);
1166 static ftrace_func_t saved_ftrace_func;
1167 static int ftrace_start_up;
1169 static void ftrace_startup_enable(int command)
1171 if (saved_ftrace_func != ftrace_trace_function) {
1172 saved_ftrace_func = ftrace_trace_function;
1173 command |= FTRACE_UPDATE_TRACE_FUNC;
1176 if (!command || !ftrace_enabled)
1177 return;
1179 ftrace_run_update_code(command);
1182 static void ftrace_startup(int command)
1184 if (unlikely(ftrace_disabled))
1185 return;
1187 ftrace_start_up++;
1188 command |= FTRACE_ENABLE_CALLS;
1190 ftrace_startup_enable(command);
1193 static void ftrace_shutdown(int command)
1195 if (unlikely(ftrace_disabled))
1196 return;
1198 ftrace_start_up--;
1200 * Just warn in case of unbalance, no need to kill ftrace, it's not
1201 * critical but the ftrace_call callers may be never nopped again after
1202 * further ftrace uses.
1204 WARN_ON_ONCE(ftrace_start_up < 0);
1206 if (!ftrace_start_up)
1207 command |= FTRACE_DISABLE_CALLS;
1209 if (saved_ftrace_func != ftrace_trace_function) {
1210 saved_ftrace_func = ftrace_trace_function;
1211 command |= FTRACE_UPDATE_TRACE_FUNC;
1214 if (!command || !ftrace_enabled)
1215 return;
1217 ftrace_run_update_code(command);
1220 static void ftrace_startup_sysctl(void)
1222 int command = FTRACE_ENABLE_MCOUNT;
1224 if (unlikely(ftrace_disabled))
1225 return;
1227 /* Force update next time */
1228 saved_ftrace_func = NULL;
1229 /* ftrace_start_up is true if we want ftrace running */
1230 if (ftrace_start_up)
1231 command |= FTRACE_ENABLE_CALLS;
1233 ftrace_run_update_code(command);
1236 static void ftrace_shutdown_sysctl(void)
1238 int command = FTRACE_DISABLE_MCOUNT;
1240 if (unlikely(ftrace_disabled))
1241 return;
1243 /* ftrace_start_up is true if ftrace is running */
1244 if (ftrace_start_up)
1245 command |= FTRACE_DISABLE_CALLS;
1247 ftrace_run_update_code(command);
1250 static cycle_t ftrace_update_time;
1251 static unsigned long ftrace_update_cnt;
1252 unsigned long ftrace_update_tot_cnt;
1254 static int ftrace_update_code(struct module *mod)
1256 struct dyn_ftrace *p;
1257 cycle_t start, stop;
1259 start = ftrace_now(raw_smp_processor_id());
1260 ftrace_update_cnt = 0;
1262 while (ftrace_new_addrs) {
1264 /* If something went wrong, bail without enabling anything */
1265 if (unlikely(ftrace_disabled))
1266 return -1;
1268 p = ftrace_new_addrs;
1269 ftrace_new_addrs = p->newlist;
1270 p->flags = 0L;
1273 * Do the initial record convertion from mcount jump
1274 * to the NOP instructions.
1276 if (!ftrace_code_disable(mod, p)) {
1277 ftrace_free_rec(p);
1278 continue;
1281 p->flags |= FTRACE_FL_CONVERTED;
1282 ftrace_update_cnt++;
1285 * If the tracing is enabled, go ahead and enable the record.
1287 * The reason not to enable the record immediatelly is the
1288 * inherent check of ftrace_make_nop/ftrace_make_call for
1289 * correct previous instructions. Making first the NOP
1290 * conversion puts the module to the correct state, thus
1291 * passing the ftrace_make_call check.
1293 if (ftrace_start_up) {
1294 int failed = __ftrace_replace_code(p, 1);
1295 if (failed) {
1296 ftrace_bug(failed, p->ip);
1297 ftrace_free_rec(p);
1302 stop = ftrace_now(raw_smp_processor_id());
1303 ftrace_update_time = stop - start;
1304 ftrace_update_tot_cnt += ftrace_update_cnt;
1306 return 0;
1309 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1311 struct ftrace_page *pg;
1312 int cnt;
1313 int i;
1315 /* allocate a few pages */
1316 ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1317 if (!ftrace_pages_start)
1318 return -1;
1321 * Allocate a few more pages.
1323 * TODO: have some parser search vmlinux before
1324 * final linking to find all calls to ftrace.
1325 * Then we can:
1326 * a) know how many pages to allocate.
1327 * and/or
1328 * b) set up the table then.
1330 * The dynamic code is still necessary for
1331 * modules.
1334 pg = ftrace_pages = ftrace_pages_start;
1336 cnt = num_to_init / ENTRIES_PER_PAGE;
1337 pr_info("ftrace: allocating %ld entries in %d pages\n",
1338 num_to_init, cnt + 1);
1340 for (i = 0; i < cnt; i++) {
1341 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1343 /* If we fail, we'll try later anyway */
1344 if (!pg->next)
1345 break;
1347 pg = pg->next;
1350 return 0;
1353 enum {
1354 FTRACE_ITER_FILTER = (1 << 0),
1355 FTRACE_ITER_NOTRACE = (1 << 1),
1356 FTRACE_ITER_FAILURES = (1 << 2),
1357 FTRACE_ITER_PRINTALL = (1 << 3),
1358 FTRACE_ITER_HASH = (1 << 4),
1361 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1363 struct ftrace_iterator {
1364 struct ftrace_page *pg;
1365 int hidx;
1366 int idx;
1367 unsigned flags;
1368 struct trace_parser parser;
1371 static void *
1372 t_hash_next(struct seq_file *m, void *v, loff_t *pos)
1374 struct ftrace_iterator *iter = m->private;
1375 struct hlist_node *hnd = v;
1376 struct hlist_head *hhd;
1378 WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
1380 (*pos)++;
1382 retry:
1383 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1384 return NULL;
1386 hhd = &ftrace_func_hash[iter->hidx];
1388 if (hlist_empty(hhd)) {
1389 iter->hidx++;
1390 hnd = NULL;
1391 goto retry;
1394 if (!hnd)
1395 hnd = hhd->first;
1396 else {
1397 hnd = hnd->next;
1398 if (!hnd) {
1399 iter->hidx++;
1400 goto retry;
1404 return hnd;
1407 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1409 struct ftrace_iterator *iter = m->private;
1410 void *p = NULL;
1411 loff_t l;
1413 if (!(iter->flags & FTRACE_ITER_HASH))
1414 *pos = 0;
1416 iter->flags |= FTRACE_ITER_HASH;
1418 iter->hidx = 0;
1419 for (l = 0; l <= *pos; ) {
1420 p = t_hash_next(m, p, &l);
1421 if (!p)
1422 break;
1424 return p;
1427 static int t_hash_show(struct seq_file *m, void *v)
1429 struct ftrace_func_probe *rec;
1430 struct hlist_node *hnd = v;
1432 rec = hlist_entry(hnd, struct ftrace_func_probe, node);
1434 if (rec->ops->print)
1435 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1437 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
1439 if (rec->data)
1440 seq_printf(m, ":%p", rec->data);
1441 seq_putc(m, '\n');
1443 return 0;
1446 static void *
1447 t_next(struct seq_file *m, void *v, loff_t *pos)
1449 struct ftrace_iterator *iter = m->private;
1450 struct dyn_ftrace *rec = NULL;
1452 if (iter->flags & FTRACE_ITER_HASH)
1453 return t_hash_next(m, v, pos);
1455 (*pos)++;
1457 if (iter->flags & FTRACE_ITER_PRINTALL)
1458 return NULL;
1460 retry:
1461 if (iter->idx >= iter->pg->index) {
1462 if (iter->pg->next) {
1463 iter->pg = iter->pg->next;
1464 iter->idx = 0;
1465 goto retry;
1467 } else {
1468 rec = &iter->pg->records[iter->idx++];
1469 if ((rec->flags & FTRACE_FL_FREE) ||
1471 (!(iter->flags & FTRACE_ITER_FAILURES) &&
1472 (rec->flags & FTRACE_FL_FAILED)) ||
1474 ((iter->flags & FTRACE_ITER_FAILURES) &&
1475 !(rec->flags & FTRACE_FL_FAILED)) ||
1477 ((iter->flags & FTRACE_ITER_FILTER) &&
1478 !(rec->flags & FTRACE_FL_FILTER)) ||
1480 ((iter->flags & FTRACE_ITER_NOTRACE) &&
1481 !(rec->flags & FTRACE_FL_NOTRACE))) {
1482 rec = NULL;
1483 goto retry;
1487 return rec;
1490 static void *t_start(struct seq_file *m, loff_t *pos)
1492 struct ftrace_iterator *iter = m->private;
1493 void *p = NULL;
1494 loff_t l;
1496 mutex_lock(&ftrace_lock);
1498 * For set_ftrace_filter reading, if we have the filter
1499 * off, we can short cut and just print out that all
1500 * functions are enabled.
1502 if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1503 if (*pos > 0)
1504 return t_hash_start(m, pos);
1505 iter->flags |= FTRACE_ITER_PRINTALL;
1506 return iter;
1509 if (iter->flags & FTRACE_ITER_HASH)
1510 return t_hash_start(m, pos);
1512 iter->pg = ftrace_pages_start;
1513 iter->idx = 0;
1514 for (l = 0; l <= *pos; ) {
1515 p = t_next(m, p, &l);
1516 if (!p)
1517 break;
1520 if (!p && iter->flags & FTRACE_ITER_FILTER)
1521 return t_hash_start(m, pos);
1523 return p;
1526 static void t_stop(struct seq_file *m, void *p)
1528 mutex_unlock(&ftrace_lock);
1531 static int t_show(struct seq_file *m, void *v)
1533 struct ftrace_iterator *iter = m->private;
1534 struct dyn_ftrace *rec = v;
1536 if (iter->flags & FTRACE_ITER_HASH)
1537 return t_hash_show(m, v);
1539 if (iter->flags & FTRACE_ITER_PRINTALL) {
1540 seq_printf(m, "#### all functions enabled ####\n");
1541 return 0;
1544 if (!rec)
1545 return 0;
1547 seq_printf(m, "%ps\n", (void *)rec->ip);
1549 return 0;
1552 static const struct seq_operations show_ftrace_seq_ops = {
1553 .start = t_start,
1554 .next = t_next,
1555 .stop = t_stop,
1556 .show = t_show,
1559 static int
1560 ftrace_avail_open(struct inode *inode, struct file *file)
1562 struct ftrace_iterator *iter;
1563 int ret;
1565 if (unlikely(ftrace_disabled))
1566 return -ENODEV;
1568 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1569 if (!iter)
1570 return -ENOMEM;
1572 iter->pg = ftrace_pages_start;
1574 ret = seq_open(file, &show_ftrace_seq_ops);
1575 if (!ret) {
1576 struct seq_file *m = file->private_data;
1578 m->private = iter;
1579 } else {
1580 kfree(iter);
1583 return ret;
1586 static int
1587 ftrace_failures_open(struct inode *inode, struct file *file)
1589 int ret;
1590 struct seq_file *m;
1591 struct ftrace_iterator *iter;
1593 ret = ftrace_avail_open(inode, file);
1594 if (!ret) {
1595 m = (struct seq_file *)file->private_data;
1596 iter = (struct ftrace_iterator *)m->private;
1597 iter->flags = FTRACE_ITER_FAILURES;
1600 return ret;
1604 static void ftrace_filter_reset(int enable)
1606 struct ftrace_page *pg;
1607 struct dyn_ftrace *rec;
1608 unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1610 mutex_lock(&ftrace_lock);
1611 if (enable)
1612 ftrace_filtered = 0;
1613 do_for_each_ftrace_rec(pg, rec) {
1614 if (rec->flags & FTRACE_FL_FAILED)
1615 continue;
1616 rec->flags &= ~type;
1617 } while_for_each_ftrace_rec();
1618 mutex_unlock(&ftrace_lock);
1621 static int
1622 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1624 struct ftrace_iterator *iter;
1625 int ret = 0;
1627 if (unlikely(ftrace_disabled))
1628 return -ENODEV;
1630 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1631 if (!iter)
1632 return -ENOMEM;
1634 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
1635 kfree(iter);
1636 return -ENOMEM;
1639 mutex_lock(&ftrace_regex_lock);
1640 if ((file->f_mode & FMODE_WRITE) &&
1641 (file->f_flags & O_TRUNC))
1642 ftrace_filter_reset(enable);
1644 if (file->f_mode & FMODE_READ) {
1645 iter->pg = ftrace_pages_start;
1646 iter->flags = enable ? FTRACE_ITER_FILTER :
1647 FTRACE_ITER_NOTRACE;
1649 ret = seq_open(file, &show_ftrace_seq_ops);
1650 if (!ret) {
1651 struct seq_file *m = file->private_data;
1652 m->private = iter;
1653 } else {
1654 trace_parser_put(&iter->parser);
1655 kfree(iter);
1657 } else
1658 file->private_data = iter;
1659 mutex_unlock(&ftrace_regex_lock);
1661 return ret;
1664 static int
1665 ftrace_filter_open(struct inode *inode, struct file *file)
1667 return ftrace_regex_open(inode, file, 1);
1670 static int
1671 ftrace_notrace_open(struct inode *inode, struct file *file)
1673 return ftrace_regex_open(inode, file, 0);
1676 static loff_t
1677 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1679 loff_t ret;
1681 if (file->f_mode & FMODE_READ)
1682 ret = seq_lseek(file, offset, origin);
1683 else
1684 file->f_pos = ret = 1;
1686 return ret;
1689 static int ftrace_match(char *str, char *regex, int len, int type)
1691 int matched = 0;
1692 int slen;
1694 switch (type) {
1695 case MATCH_FULL:
1696 if (strcmp(str, regex) == 0)
1697 matched = 1;
1698 break;
1699 case MATCH_FRONT_ONLY:
1700 if (strncmp(str, regex, len) == 0)
1701 matched = 1;
1702 break;
1703 case MATCH_MIDDLE_ONLY:
1704 if (strstr(str, regex))
1705 matched = 1;
1706 break;
1707 case MATCH_END_ONLY:
1708 slen = strlen(str);
1709 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
1710 matched = 1;
1711 break;
1714 return matched;
1717 static int
1718 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1720 char str[KSYM_SYMBOL_LEN];
1722 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1723 return ftrace_match(str, regex, len, type);
1726 static int ftrace_match_records(char *buff, int len, int enable)
1728 unsigned int search_len;
1729 struct ftrace_page *pg;
1730 struct dyn_ftrace *rec;
1731 unsigned long flag;
1732 char *search;
1733 int type;
1734 int not;
1735 int found = 0;
1737 flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1738 type = filter_parse_regex(buff, len, &search, &not);
1740 search_len = strlen(search);
1742 mutex_lock(&ftrace_lock);
1743 do_for_each_ftrace_rec(pg, rec) {
1745 if (rec->flags & FTRACE_FL_FAILED)
1746 continue;
1748 if (ftrace_match_record(rec, search, search_len, type)) {
1749 if (not)
1750 rec->flags &= ~flag;
1751 else
1752 rec->flags |= flag;
1753 found = 1;
1756 * Only enable filtering if we have a function that
1757 * is filtered on.
1759 if (enable && (rec->flags & FTRACE_FL_FILTER))
1760 ftrace_filtered = 1;
1761 } while_for_each_ftrace_rec();
1762 mutex_unlock(&ftrace_lock);
1764 return found;
1767 static int
1768 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1769 char *regex, int len, int type)
1771 char str[KSYM_SYMBOL_LEN];
1772 char *modname;
1774 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1776 if (!modname || strcmp(modname, mod))
1777 return 0;
1779 /* blank search means to match all funcs in the mod */
1780 if (len)
1781 return ftrace_match(str, regex, len, type);
1782 else
1783 return 1;
1786 static int ftrace_match_module_records(char *buff, char *mod, int enable)
1788 unsigned search_len = 0;
1789 struct ftrace_page *pg;
1790 struct dyn_ftrace *rec;
1791 int type = MATCH_FULL;
1792 char *search = buff;
1793 unsigned long flag;
1794 int not = 0;
1795 int found = 0;
1797 flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1799 /* blank or '*' mean the same */
1800 if (strcmp(buff, "*") == 0)
1801 buff[0] = 0;
1803 /* handle the case of 'dont filter this module' */
1804 if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1805 buff[0] = 0;
1806 not = 1;
1809 if (strlen(buff)) {
1810 type = filter_parse_regex(buff, strlen(buff), &search, &not);
1811 search_len = strlen(search);
1814 mutex_lock(&ftrace_lock);
1815 do_for_each_ftrace_rec(pg, rec) {
1817 if (rec->flags & FTRACE_FL_FAILED)
1818 continue;
1820 if (ftrace_match_module_record(rec, mod,
1821 search, search_len, type)) {
1822 if (not)
1823 rec->flags &= ~flag;
1824 else
1825 rec->flags |= flag;
1826 found = 1;
1828 if (enable && (rec->flags & FTRACE_FL_FILTER))
1829 ftrace_filtered = 1;
1831 } while_for_each_ftrace_rec();
1832 mutex_unlock(&ftrace_lock);
1834 return found;
1838 * We register the module command as a template to show others how
1839 * to register the a command as well.
1842 static int
1843 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1845 char *mod;
1848 * cmd == 'mod' because we only registered this func
1849 * for the 'mod' ftrace_func_command.
1850 * But if you register one func with multiple commands,
1851 * you can tell which command was used by the cmd
1852 * parameter.
1855 /* we must have a module name */
1856 if (!param)
1857 return -EINVAL;
1859 mod = strsep(&param, ":");
1860 if (!strlen(mod))
1861 return -EINVAL;
1863 if (ftrace_match_module_records(func, mod, enable))
1864 return 0;
1865 return -EINVAL;
1868 static struct ftrace_func_command ftrace_mod_cmd = {
1869 .name = "mod",
1870 .func = ftrace_mod_callback,
1873 static int __init ftrace_mod_cmd_init(void)
1875 return register_ftrace_command(&ftrace_mod_cmd);
1877 device_initcall(ftrace_mod_cmd_init);
1879 static void
1880 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1882 struct ftrace_func_probe *entry;
1883 struct hlist_head *hhd;
1884 struct hlist_node *n;
1885 unsigned long key;
1886 int resched;
1888 key = hash_long(ip, FTRACE_HASH_BITS);
1890 hhd = &ftrace_func_hash[key];
1892 if (hlist_empty(hhd))
1893 return;
1896 * Disable preemption for these calls to prevent a RCU grace
1897 * period. This syncs the hash iteration and freeing of items
1898 * on the hash. rcu_read_lock is too dangerous here.
1900 resched = ftrace_preempt_disable();
1901 hlist_for_each_entry_rcu(entry, n, hhd, node) {
1902 if (entry->ip == ip)
1903 entry->ops->func(ip, parent_ip, &entry->data);
1905 ftrace_preempt_enable(resched);
1908 static struct ftrace_ops trace_probe_ops __read_mostly =
1910 .func = function_trace_probe_call,
1913 static int ftrace_probe_registered;
1915 static void __enable_ftrace_function_probe(void)
1917 int i;
1919 if (ftrace_probe_registered)
1920 return;
1922 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1923 struct hlist_head *hhd = &ftrace_func_hash[i];
1924 if (hhd->first)
1925 break;
1927 /* Nothing registered? */
1928 if (i == FTRACE_FUNC_HASHSIZE)
1929 return;
1931 __register_ftrace_function(&trace_probe_ops);
1932 ftrace_startup(0);
1933 ftrace_probe_registered = 1;
1936 static void __disable_ftrace_function_probe(void)
1938 int i;
1940 if (!ftrace_probe_registered)
1941 return;
1943 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1944 struct hlist_head *hhd = &ftrace_func_hash[i];
1945 if (hhd->first)
1946 return;
1949 /* no more funcs left */
1950 __unregister_ftrace_function(&trace_probe_ops);
1951 ftrace_shutdown(0);
1952 ftrace_probe_registered = 0;
1956 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
1958 struct ftrace_func_probe *entry =
1959 container_of(rhp, struct ftrace_func_probe, rcu);
1961 if (entry->ops->free)
1962 entry->ops->free(&entry->data);
1963 kfree(entry);
1968 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
1969 void *data)
1971 struct ftrace_func_probe *entry;
1972 struct ftrace_page *pg;
1973 struct dyn_ftrace *rec;
1974 int type, len, not;
1975 unsigned long key;
1976 int count = 0;
1977 char *search;
1979 type = filter_parse_regex(glob, strlen(glob), &search, &not);
1980 len = strlen(search);
1982 /* we do not support '!' for function probes */
1983 if (WARN_ON(not))
1984 return -EINVAL;
1986 mutex_lock(&ftrace_lock);
1987 do_for_each_ftrace_rec(pg, rec) {
1989 if (rec->flags & FTRACE_FL_FAILED)
1990 continue;
1992 if (!ftrace_match_record(rec, search, len, type))
1993 continue;
1995 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1996 if (!entry) {
1997 /* If we did not process any, then return error */
1998 if (!count)
1999 count = -ENOMEM;
2000 goto out_unlock;
2003 count++;
2005 entry->data = data;
2008 * The caller might want to do something special
2009 * for each function we find. We call the callback
2010 * to give the caller an opportunity to do so.
2012 if (ops->callback) {
2013 if (ops->callback(rec->ip, &entry->data) < 0) {
2014 /* caller does not like this func */
2015 kfree(entry);
2016 continue;
2020 entry->ops = ops;
2021 entry->ip = rec->ip;
2023 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2024 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2026 } while_for_each_ftrace_rec();
2027 __enable_ftrace_function_probe();
2029 out_unlock:
2030 mutex_unlock(&ftrace_lock);
2032 return count;
2035 enum {
2036 PROBE_TEST_FUNC = 1,
2037 PROBE_TEST_DATA = 2
2040 static void
2041 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2042 void *data, int flags)
2044 struct ftrace_func_probe *entry;
2045 struct hlist_node *n, *tmp;
2046 char str[KSYM_SYMBOL_LEN];
2047 int type = MATCH_FULL;
2048 int i, len = 0;
2049 char *search;
2051 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2052 glob = NULL;
2053 else if (glob) {
2054 int not;
2056 type = filter_parse_regex(glob, strlen(glob), &search, &not);
2057 len = strlen(search);
2059 /* we do not support '!' for function probes */
2060 if (WARN_ON(not))
2061 return;
2064 mutex_lock(&ftrace_lock);
2065 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2066 struct hlist_head *hhd = &ftrace_func_hash[i];
2068 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2070 /* break up if statements for readability */
2071 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2072 continue;
2074 if ((flags & PROBE_TEST_DATA) && entry->data != data)
2075 continue;
2077 /* do this last, since it is the most expensive */
2078 if (glob) {
2079 kallsyms_lookup(entry->ip, NULL, NULL,
2080 NULL, str);
2081 if (!ftrace_match(str, glob, len, type))
2082 continue;
2085 hlist_del(&entry->node);
2086 call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2089 __disable_ftrace_function_probe();
2090 mutex_unlock(&ftrace_lock);
2093 void
2094 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2095 void *data)
2097 __unregister_ftrace_function_probe(glob, ops, data,
2098 PROBE_TEST_FUNC | PROBE_TEST_DATA);
2101 void
2102 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2104 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2107 void unregister_ftrace_function_probe_all(char *glob)
2109 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2112 static LIST_HEAD(ftrace_commands);
2113 static DEFINE_MUTEX(ftrace_cmd_mutex);
2115 int register_ftrace_command(struct ftrace_func_command *cmd)
2117 struct ftrace_func_command *p;
2118 int ret = 0;
2120 mutex_lock(&ftrace_cmd_mutex);
2121 list_for_each_entry(p, &ftrace_commands, list) {
2122 if (strcmp(cmd->name, p->name) == 0) {
2123 ret = -EBUSY;
2124 goto out_unlock;
2127 list_add(&cmd->list, &ftrace_commands);
2128 out_unlock:
2129 mutex_unlock(&ftrace_cmd_mutex);
2131 return ret;
2134 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2136 struct ftrace_func_command *p, *n;
2137 int ret = -ENODEV;
2139 mutex_lock(&ftrace_cmd_mutex);
2140 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2141 if (strcmp(cmd->name, p->name) == 0) {
2142 ret = 0;
2143 list_del_init(&p->list);
2144 goto out_unlock;
2147 out_unlock:
2148 mutex_unlock(&ftrace_cmd_mutex);
2150 return ret;
2153 static int ftrace_process_regex(char *buff, int len, int enable)
2155 char *func, *command, *next = buff;
2156 struct ftrace_func_command *p;
2157 int ret = -EINVAL;
2159 func = strsep(&next, ":");
2161 if (!next) {
2162 if (ftrace_match_records(func, len, enable))
2163 return 0;
2164 return ret;
2167 /* command found */
2169 command = strsep(&next, ":");
2171 mutex_lock(&ftrace_cmd_mutex);
2172 list_for_each_entry(p, &ftrace_commands, list) {
2173 if (strcmp(p->name, command) == 0) {
2174 ret = p->func(func, command, next, enable);
2175 goto out_unlock;
2178 out_unlock:
2179 mutex_unlock(&ftrace_cmd_mutex);
2181 return ret;
2184 static ssize_t
2185 ftrace_regex_write(struct file *file, const char __user *ubuf,
2186 size_t cnt, loff_t *ppos, int enable)
2188 struct ftrace_iterator *iter;
2189 struct trace_parser *parser;
2190 ssize_t ret, read;
2192 if (!cnt)
2193 return 0;
2195 mutex_lock(&ftrace_regex_lock);
2197 if (file->f_mode & FMODE_READ) {
2198 struct seq_file *m = file->private_data;
2199 iter = m->private;
2200 } else
2201 iter = file->private_data;
2203 parser = &iter->parser;
2204 read = trace_get_user(parser, ubuf, cnt, ppos);
2206 if (read >= 0 && trace_parser_loaded(parser) &&
2207 !trace_parser_cont(parser)) {
2208 ret = ftrace_process_regex(parser->buffer,
2209 parser->idx, enable);
2210 trace_parser_clear(parser);
2211 if (ret)
2212 goto out_unlock;
2215 ret = read;
2216 out_unlock:
2217 mutex_unlock(&ftrace_regex_lock);
2219 return ret;
2222 static ssize_t
2223 ftrace_filter_write(struct file *file, const char __user *ubuf,
2224 size_t cnt, loff_t *ppos)
2226 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2229 static ssize_t
2230 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2231 size_t cnt, loff_t *ppos)
2233 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2236 static void
2237 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2239 if (unlikely(ftrace_disabled))
2240 return;
2242 mutex_lock(&ftrace_regex_lock);
2243 if (reset)
2244 ftrace_filter_reset(enable);
2245 if (buf)
2246 ftrace_match_records(buf, len, enable);
2247 mutex_unlock(&ftrace_regex_lock);
2251 * ftrace_set_filter - set a function to filter on in ftrace
2252 * @buf - the string that holds the function filter text.
2253 * @len - the length of the string.
2254 * @reset - non zero to reset all filters before applying this filter.
2256 * Filters denote which functions should be enabled when tracing is enabled.
2257 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2259 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2261 ftrace_set_regex(buf, len, reset, 1);
2265 * ftrace_set_notrace - set a function to not trace in ftrace
2266 * @buf - the string that holds the function notrace text.
2267 * @len - the length of the string.
2268 * @reset - non zero to reset all filters before applying this filter.
2270 * Notrace Filters denote which functions should not be enabled when tracing
2271 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2272 * for tracing.
2274 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2276 ftrace_set_regex(buf, len, reset, 0);
2280 * command line interface to allow users to set filters on boot up.
2282 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
2283 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2284 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2286 static int __init set_ftrace_notrace(char *str)
2288 strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2289 return 1;
2291 __setup("ftrace_notrace=", set_ftrace_notrace);
2293 static int __init set_ftrace_filter(char *str)
2295 strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2296 return 1;
2298 __setup("ftrace_filter=", set_ftrace_filter);
2300 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2301 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
2302 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
2304 static int __init set_graph_function(char *str)
2306 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
2307 return 1;
2309 __setup("ftrace_graph_filter=", set_graph_function);
2311 static void __init set_ftrace_early_graph(char *buf)
2313 int ret;
2314 char *func;
2316 while (buf) {
2317 func = strsep(&buf, ",");
2318 /* we allow only one expression at a time */
2319 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2320 func);
2321 if (ret)
2322 printk(KERN_DEBUG "ftrace: function %s not "
2323 "traceable\n", func);
2326 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2328 static void __init set_ftrace_early_filter(char *buf, int enable)
2330 char *func;
2332 while (buf) {
2333 func = strsep(&buf, ",");
2334 ftrace_set_regex(func, strlen(func), 0, enable);
2338 static void __init set_ftrace_early_filters(void)
2340 if (ftrace_filter_buf[0])
2341 set_ftrace_early_filter(ftrace_filter_buf, 1);
2342 if (ftrace_notrace_buf[0])
2343 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2344 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2345 if (ftrace_graph_buf[0])
2346 set_ftrace_early_graph(ftrace_graph_buf);
2347 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2350 static int
2351 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2353 struct seq_file *m = (struct seq_file *)file->private_data;
2354 struct ftrace_iterator *iter;
2355 struct trace_parser *parser;
2357 mutex_lock(&ftrace_regex_lock);
2358 if (file->f_mode & FMODE_READ) {
2359 iter = m->private;
2361 seq_release(inode, file);
2362 } else
2363 iter = file->private_data;
2365 parser = &iter->parser;
2366 if (trace_parser_loaded(parser)) {
2367 parser->buffer[parser->idx] = 0;
2368 ftrace_match_records(parser->buffer, parser->idx, enable);
2371 mutex_lock(&ftrace_lock);
2372 if (ftrace_start_up && ftrace_enabled)
2373 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2374 mutex_unlock(&ftrace_lock);
2376 trace_parser_put(parser);
2377 kfree(iter);
2379 mutex_unlock(&ftrace_regex_lock);
2380 return 0;
2383 static int
2384 ftrace_filter_release(struct inode *inode, struct file *file)
2386 return ftrace_regex_release(inode, file, 1);
2389 static int
2390 ftrace_notrace_release(struct inode *inode, struct file *file)
2392 return ftrace_regex_release(inode, file, 0);
2395 static const struct file_operations ftrace_avail_fops = {
2396 .open = ftrace_avail_open,
2397 .read = seq_read,
2398 .llseek = seq_lseek,
2399 .release = seq_release_private,
2402 static const struct file_operations ftrace_failures_fops = {
2403 .open = ftrace_failures_open,
2404 .read = seq_read,
2405 .llseek = seq_lseek,
2406 .release = seq_release_private,
2409 static const struct file_operations ftrace_filter_fops = {
2410 .open = ftrace_filter_open,
2411 .read = seq_read,
2412 .write = ftrace_filter_write,
2413 .llseek = ftrace_regex_lseek,
2414 .release = ftrace_filter_release,
2417 static const struct file_operations ftrace_notrace_fops = {
2418 .open = ftrace_notrace_open,
2419 .read = seq_read,
2420 .write = ftrace_notrace_write,
2421 .llseek = ftrace_regex_lseek,
2422 .release = ftrace_notrace_release,
2425 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2427 static DEFINE_MUTEX(graph_lock);
2429 int ftrace_graph_count;
2430 int ftrace_graph_filter_enabled;
2431 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2433 static void *
2434 __g_next(struct seq_file *m, loff_t *pos)
2436 if (*pos >= ftrace_graph_count)
2437 return NULL;
2438 return &ftrace_graph_funcs[*pos];
2441 static void *
2442 g_next(struct seq_file *m, void *v, loff_t *pos)
2444 (*pos)++;
2445 return __g_next(m, pos);
2448 static void *g_start(struct seq_file *m, loff_t *pos)
2450 mutex_lock(&graph_lock);
2452 /* Nothing, tell g_show to print all functions are enabled */
2453 if (!ftrace_graph_filter_enabled && !*pos)
2454 return (void *)1;
2456 return __g_next(m, pos);
2459 static void g_stop(struct seq_file *m, void *p)
2461 mutex_unlock(&graph_lock);
2464 static int g_show(struct seq_file *m, void *v)
2466 unsigned long *ptr = v;
2468 if (!ptr)
2469 return 0;
2471 if (ptr == (unsigned long *)1) {
2472 seq_printf(m, "#### all functions enabled ####\n");
2473 return 0;
2476 seq_printf(m, "%ps\n", (void *)*ptr);
2478 return 0;
2481 static const struct seq_operations ftrace_graph_seq_ops = {
2482 .start = g_start,
2483 .next = g_next,
2484 .stop = g_stop,
2485 .show = g_show,
2488 static int
2489 ftrace_graph_open(struct inode *inode, struct file *file)
2491 int ret = 0;
2493 if (unlikely(ftrace_disabled))
2494 return -ENODEV;
2496 mutex_lock(&graph_lock);
2497 if ((file->f_mode & FMODE_WRITE) &&
2498 (file->f_flags & O_TRUNC)) {
2499 ftrace_graph_filter_enabled = 0;
2500 ftrace_graph_count = 0;
2501 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2503 mutex_unlock(&graph_lock);
2505 if (file->f_mode & FMODE_READ)
2506 ret = seq_open(file, &ftrace_graph_seq_ops);
2508 return ret;
2511 static int
2512 ftrace_graph_release(struct inode *inode, struct file *file)
2514 if (file->f_mode & FMODE_READ)
2515 seq_release(inode, file);
2516 return 0;
2519 static int
2520 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2522 struct dyn_ftrace *rec;
2523 struct ftrace_page *pg;
2524 int search_len;
2525 int fail = 1;
2526 int type, not;
2527 char *search;
2528 bool exists;
2529 int i;
2531 if (ftrace_disabled)
2532 return -ENODEV;
2534 /* decode regex */
2535 type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
2536 if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
2537 return -EBUSY;
2539 search_len = strlen(search);
2541 mutex_lock(&ftrace_lock);
2542 do_for_each_ftrace_rec(pg, rec) {
2544 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2545 continue;
2547 if (ftrace_match_record(rec, search, search_len, type)) {
2548 /* if it is in the array */
2549 exists = false;
2550 for (i = 0; i < *idx; i++) {
2551 if (array[i] == rec->ip) {
2552 exists = true;
2553 break;
2557 if (!not) {
2558 fail = 0;
2559 if (!exists) {
2560 array[(*idx)++] = rec->ip;
2561 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2562 goto out;
2564 } else {
2565 if (exists) {
2566 array[i] = array[--(*idx)];
2567 array[*idx] = 0;
2568 fail = 0;
2572 } while_for_each_ftrace_rec();
2573 out:
2574 mutex_unlock(&ftrace_lock);
2576 if (fail)
2577 return -EINVAL;
2579 ftrace_graph_filter_enabled = 1;
2580 return 0;
2583 static ssize_t
2584 ftrace_graph_write(struct file *file, const char __user *ubuf,
2585 size_t cnt, loff_t *ppos)
2587 struct trace_parser parser;
2588 ssize_t read, ret;
2590 if (!cnt)
2591 return 0;
2593 mutex_lock(&graph_lock);
2595 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
2596 ret = -ENOMEM;
2597 goto out_unlock;
2600 read = trace_get_user(&parser, ubuf, cnt, ppos);
2602 if (read >= 0 && trace_parser_loaded((&parser))) {
2603 parser.buffer[parser.idx] = 0;
2605 /* we allow only one expression at a time */
2606 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2607 parser.buffer);
2608 if (ret)
2609 goto out_free;
2612 ret = read;
2614 out_free:
2615 trace_parser_put(&parser);
2616 out_unlock:
2617 mutex_unlock(&graph_lock);
2619 return ret;
2622 static const struct file_operations ftrace_graph_fops = {
2623 .open = ftrace_graph_open,
2624 .read = seq_read,
2625 .write = ftrace_graph_write,
2626 .release = ftrace_graph_release,
2628 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2630 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2633 trace_create_file("available_filter_functions", 0444,
2634 d_tracer, NULL, &ftrace_avail_fops);
2636 trace_create_file("failures", 0444,
2637 d_tracer, NULL, &ftrace_failures_fops);
2639 trace_create_file("set_ftrace_filter", 0644, d_tracer,
2640 NULL, &ftrace_filter_fops);
2642 trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2643 NULL, &ftrace_notrace_fops);
2645 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2646 trace_create_file("set_graph_function", 0444, d_tracer,
2647 NULL,
2648 &ftrace_graph_fops);
2649 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2651 return 0;
2654 static int ftrace_process_locs(struct module *mod,
2655 unsigned long *start,
2656 unsigned long *end)
2658 unsigned long *p;
2659 unsigned long addr;
2660 unsigned long flags;
2662 mutex_lock(&ftrace_lock);
2663 p = start;
2664 while (p < end) {
2665 addr = ftrace_call_adjust(*p++);
2667 * Some architecture linkers will pad between
2668 * the different mcount_loc sections of different
2669 * object files to satisfy alignments.
2670 * Skip any NULL pointers.
2672 if (!addr)
2673 continue;
2674 ftrace_record_ip(addr);
2677 /* disable interrupts to prevent kstop machine */
2678 local_irq_save(flags);
2679 ftrace_update_code(mod);
2680 local_irq_restore(flags);
2681 mutex_unlock(&ftrace_lock);
2683 return 0;
2686 #ifdef CONFIG_MODULES
2687 void ftrace_release_mod(struct module *mod)
2689 struct dyn_ftrace *rec;
2690 struct ftrace_page *pg;
2692 if (ftrace_disabled)
2693 return;
2695 mutex_lock(&ftrace_lock);
2696 do_for_each_ftrace_rec(pg, rec) {
2697 if (within_module_core(rec->ip, mod)) {
2699 * rec->ip is changed in ftrace_free_rec()
2700 * It should not between s and e if record was freed.
2702 FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2703 ftrace_free_rec(rec);
2705 } while_for_each_ftrace_rec();
2706 mutex_unlock(&ftrace_lock);
2709 static void ftrace_init_module(struct module *mod,
2710 unsigned long *start, unsigned long *end)
2712 if (ftrace_disabled || start == end)
2713 return;
2714 ftrace_process_locs(mod, start, end);
2717 static int ftrace_module_notify(struct notifier_block *self,
2718 unsigned long val, void *data)
2720 struct module *mod = data;
2722 switch (val) {
2723 case MODULE_STATE_COMING:
2724 ftrace_init_module(mod, mod->ftrace_callsites,
2725 mod->ftrace_callsites +
2726 mod->num_ftrace_callsites);
2727 break;
2728 case MODULE_STATE_GOING:
2729 ftrace_release_mod(mod);
2730 break;
2733 return 0;
2735 #else
2736 static int ftrace_module_notify(struct notifier_block *self,
2737 unsigned long val, void *data)
2739 return 0;
2741 #endif /* CONFIG_MODULES */
2743 struct notifier_block ftrace_module_nb = {
2744 .notifier_call = ftrace_module_notify,
2745 .priority = 0,
2748 extern unsigned long __start_mcount_loc[];
2749 extern unsigned long __stop_mcount_loc[];
2751 void __init ftrace_init(void)
2753 unsigned long count, addr, flags;
2754 int ret;
2756 /* Keep the ftrace pointer to the stub */
2757 addr = (unsigned long)ftrace_stub;
2759 local_irq_save(flags);
2760 ftrace_dyn_arch_init(&addr);
2761 local_irq_restore(flags);
2763 /* ftrace_dyn_arch_init places the return code in addr */
2764 if (addr)
2765 goto failed;
2767 count = __stop_mcount_loc - __start_mcount_loc;
2769 ret = ftrace_dyn_table_alloc(count);
2770 if (ret)
2771 goto failed;
2773 last_ftrace_enabled = ftrace_enabled = 1;
2775 ret = ftrace_process_locs(NULL,
2776 __start_mcount_loc,
2777 __stop_mcount_loc);
2779 ret = register_module_notifier(&ftrace_module_nb);
2780 if (ret)
2781 pr_warning("Failed to register trace ftrace module notifier\n");
2783 set_ftrace_early_filters();
2785 return;
2786 failed:
2787 ftrace_disabled = 1;
2790 #else
2792 static int __init ftrace_nodyn_init(void)
2794 ftrace_enabled = 1;
2795 return 0;
2797 device_initcall(ftrace_nodyn_init);
2799 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2800 static inline void ftrace_startup_enable(int command) { }
2801 /* Keep as macros so we do not need to define the commands */
2802 # define ftrace_startup(command) do { } while (0)
2803 # define ftrace_shutdown(command) do { } while (0)
2804 # define ftrace_startup_sysctl() do { } while (0)
2805 # define ftrace_shutdown_sysctl() do { } while (0)
2806 #endif /* CONFIG_DYNAMIC_FTRACE */
2808 static void clear_ftrace_swapper(void)
2810 struct task_struct *p;
2811 int cpu;
2813 get_online_cpus();
2814 for_each_online_cpu(cpu) {
2815 p = idle_task(cpu);
2816 clear_tsk_trace_trace(p);
2818 put_online_cpus();
2821 static void set_ftrace_swapper(void)
2823 struct task_struct *p;
2824 int cpu;
2826 get_online_cpus();
2827 for_each_online_cpu(cpu) {
2828 p = idle_task(cpu);
2829 set_tsk_trace_trace(p);
2831 put_online_cpus();
2834 static void clear_ftrace_pid(struct pid *pid)
2836 struct task_struct *p;
2838 rcu_read_lock();
2839 do_each_pid_task(pid, PIDTYPE_PID, p) {
2840 clear_tsk_trace_trace(p);
2841 } while_each_pid_task(pid, PIDTYPE_PID, p);
2842 rcu_read_unlock();
2844 put_pid(pid);
2847 static void set_ftrace_pid(struct pid *pid)
2849 struct task_struct *p;
2851 rcu_read_lock();
2852 do_each_pid_task(pid, PIDTYPE_PID, p) {
2853 set_tsk_trace_trace(p);
2854 } while_each_pid_task(pid, PIDTYPE_PID, p);
2855 rcu_read_unlock();
2858 static void clear_ftrace_pid_task(struct pid *pid)
2860 if (pid == ftrace_swapper_pid)
2861 clear_ftrace_swapper();
2862 else
2863 clear_ftrace_pid(pid);
2866 static void set_ftrace_pid_task(struct pid *pid)
2868 if (pid == ftrace_swapper_pid)
2869 set_ftrace_swapper();
2870 else
2871 set_ftrace_pid(pid);
2874 static int ftrace_pid_add(int p)
2876 struct pid *pid;
2877 struct ftrace_pid *fpid;
2878 int ret = -EINVAL;
2880 mutex_lock(&ftrace_lock);
2882 if (!p)
2883 pid = ftrace_swapper_pid;
2884 else
2885 pid = find_get_pid(p);
2887 if (!pid)
2888 goto out;
2890 ret = 0;
2892 list_for_each_entry(fpid, &ftrace_pids, list)
2893 if (fpid->pid == pid)
2894 goto out_put;
2896 ret = -ENOMEM;
2898 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
2899 if (!fpid)
2900 goto out_put;
2902 list_add(&fpid->list, &ftrace_pids);
2903 fpid->pid = pid;
2905 set_ftrace_pid_task(pid);
2907 ftrace_update_pid_func();
2908 ftrace_startup_enable(0);
2910 mutex_unlock(&ftrace_lock);
2911 return 0;
2913 out_put:
2914 if (pid != ftrace_swapper_pid)
2915 put_pid(pid);
2917 out:
2918 mutex_unlock(&ftrace_lock);
2919 return ret;
2922 static void ftrace_pid_reset(void)
2924 struct ftrace_pid *fpid, *safe;
2926 mutex_lock(&ftrace_lock);
2927 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
2928 struct pid *pid = fpid->pid;
2930 clear_ftrace_pid_task(pid);
2932 list_del(&fpid->list);
2933 kfree(fpid);
2936 ftrace_update_pid_func();
2937 ftrace_startup_enable(0);
2939 mutex_unlock(&ftrace_lock);
2942 static void *fpid_start(struct seq_file *m, loff_t *pos)
2944 mutex_lock(&ftrace_lock);
2946 if (list_empty(&ftrace_pids) && (!*pos))
2947 return (void *) 1;
2949 return seq_list_start(&ftrace_pids, *pos);
2952 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
2954 if (v == (void *)1)
2955 return NULL;
2957 return seq_list_next(v, &ftrace_pids, pos);
2960 static void fpid_stop(struct seq_file *m, void *p)
2962 mutex_unlock(&ftrace_lock);
2965 static int fpid_show(struct seq_file *m, void *v)
2967 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
2969 if (v == (void *)1) {
2970 seq_printf(m, "no pid\n");
2971 return 0;
2974 if (fpid->pid == ftrace_swapper_pid)
2975 seq_printf(m, "swapper tasks\n");
2976 else
2977 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
2979 return 0;
2982 static const struct seq_operations ftrace_pid_sops = {
2983 .start = fpid_start,
2984 .next = fpid_next,
2985 .stop = fpid_stop,
2986 .show = fpid_show,
2989 static int
2990 ftrace_pid_open(struct inode *inode, struct file *file)
2992 int ret = 0;
2994 if ((file->f_mode & FMODE_WRITE) &&
2995 (file->f_flags & O_TRUNC))
2996 ftrace_pid_reset();
2998 if (file->f_mode & FMODE_READ)
2999 ret = seq_open(file, &ftrace_pid_sops);
3001 return ret;
3004 static ssize_t
3005 ftrace_pid_write(struct file *filp, const char __user *ubuf,
3006 size_t cnt, loff_t *ppos)
3008 char buf[64], *tmp;
3009 long val;
3010 int ret;
3012 if (cnt >= sizeof(buf))
3013 return -EINVAL;
3015 if (copy_from_user(&buf, ubuf, cnt))
3016 return -EFAULT;
3018 buf[cnt] = 0;
3021 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
3022 * to clean the filter quietly.
3024 tmp = strstrip(buf);
3025 if (strlen(tmp) == 0)
3026 return 1;
3028 ret = strict_strtol(tmp, 10, &val);
3029 if (ret < 0)
3030 return ret;
3032 ret = ftrace_pid_add(val);
3034 return ret ? ret : cnt;
3037 static int
3038 ftrace_pid_release(struct inode *inode, struct file *file)
3040 if (file->f_mode & FMODE_READ)
3041 seq_release(inode, file);
3043 return 0;
3046 static const struct file_operations ftrace_pid_fops = {
3047 .open = ftrace_pid_open,
3048 .write = ftrace_pid_write,
3049 .read = seq_read,
3050 .llseek = seq_lseek,
3051 .release = ftrace_pid_release,
3054 static __init int ftrace_init_debugfs(void)
3056 struct dentry *d_tracer;
3058 d_tracer = tracing_init_dentry();
3059 if (!d_tracer)
3060 return 0;
3062 ftrace_init_dyn_debugfs(d_tracer);
3064 trace_create_file("set_ftrace_pid", 0644, d_tracer,
3065 NULL, &ftrace_pid_fops);
3067 ftrace_profile_debugfs(d_tracer);
3069 return 0;
3071 fs_initcall(ftrace_init_debugfs);
3074 * ftrace_kill - kill ftrace
3076 * This function should be used by panic code. It stops ftrace
3077 * but in a not so nice way. If you need to simply kill ftrace
3078 * from a non-atomic section, use ftrace_kill.
3080 void ftrace_kill(void)
3082 ftrace_disabled = 1;
3083 ftrace_enabled = 0;
3084 clear_ftrace_function();
3088 * register_ftrace_function - register a function for profiling
3089 * @ops - ops structure that holds the function for profiling.
3091 * Register a function to be called by all functions in the
3092 * kernel.
3094 * Note: @ops->func and all the functions it calls must be labeled
3095 * with "notrace", otherwise it will go into a
3096 * recursive loop.
3098 int register_ftrace_function(struct ftrace_ops *ops)
3100 int ret;
3102 if (unlikely(ftrace_disabled))
3103 return -1;
3105 mutex_lock(&ftrace_lock);
3107 ret = __register_ftrace_function(ops);
3108 ftrace_startup(0);
3110 mutex_unlock(&ftrace_lock);
3111 return ret;
3115 * unregister_ftrace_function - unregister a function for profiling.
3116 * @ops - ops structure that holds the function to unregister
3118 * Unregister a function that was added to be called by ftrace profiling.
3120 int unregister_ftrace_function(struct ftrace_ops *ops)
3122 int ret;
3124 mutex_lock(&ftrace_lock);
3125 ret = __unregister_ftrace_function(ops);
3126 ftrace_shutdown(0);
3127 mutex_unlock(&ftrace_lock);
3129 return ret;
3133 ftrace_enable_sysctl(struct ctl_table *table, int write,
3134 void __user *buffer, size_t *lenp,
3135 loff_t *ppos)
3137 int ret;
3139 if (unlikely(ftrace_disabled))
3140 return -ENODEV;
3142 mutex_lock(&ftrace_lock);
3144 ret = proc_dointvec(table, write, buffer, lenp, ppos);
3146 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3147 goto out;
3149 last_ftrace_enabled = !!ftrace_enabled;
3151 if (ftrace_enabled) {
3153 ftrace_startup_sysctl();
3155 /* we are starting ftrace again */
3156 if (ftrace_list != &ftrace_list_end) {
3157 if (ftrace_list->next == &ftrace_list_end)
3158 ftrace_trace_function = ftrace_list->func;
3159 else
3160 ftrace_trace_function = ftrace_list_func;
3163 } else {
3164 /* stopping ftrace calls (just send to ftrace_stub) */
3165 ftrace_trace_function = ftrace_stub;
3167 ftrace_shutdown_sysctl();
3170 out:
3171 mutex_unlock(&ftrace_lock);
3172 return ret;
3175 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3177 static int ftrace_graph_active;
3178 static struct notifier_block ftrace_suspend_notifier;
3180 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3182 return 0;
3185 /* The callbacks that hook a function */
3186 trace_func_graph_ret_t ftrace_graph_return =
3187 (trace_func_graph_ret_t)ftrace_stub;
3188 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3190 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3191 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3193 int i;
3194 int ret = 0;
3195 unsigned long flags;
3196 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3197 struct task_struct *g, *t;
3199 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3200 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3201 * sizeof(struct ftrace_ret_stack),
3202 GFP_KERNEL);
3203 if (!ret_stack_list[i]) {
3204 start = 0;
3205 end = i;
3206 ret = -ENOMEM;
3207 goto free;
3211 read_lock_irqsave(&tasklist_lock, flags);
3212 do_each_thread(g, t) {
3213 if (start == end) {
3214 ret = -EAGAIN;
3215 goto unlock;
3218 if (t->ret_stack == NULL) {
3219 atomic_set(&t->tracing_graph_pause, 0);
3220 atomic_set(&t->trace_overrun, 0);
3221 t->curr_ret_stack = -1;
3222 /* Make sure the tasks see the -1 first: */
3223 smp_wmb();
3224 t->ret_stack = ret_stack_list[start++];
3226 } while_each_thread(g, t);
3228 unlock:
3229 read_unlock_irqrestore(&tasklist_lock, flags);
3230 free:
3231 for (i = start; i < end; i++)
3232 kfree(ret_stack_list[i]);
3233 return ret;
3236 static void
3237 ftrace_graph_probe_sched_switch(void *ignore,
3238 struct task_struct *prev, struct task_struct *next)
3240 unsigned long long timestamp;
3241 int index;
3244 * Does the user want to count the time a function was asleep.
3245 * If so, do not update the time stamps.
3247 if (trace_flags & TRACE_ITER_SLEEP_TIME)
3248 return;
3250 timestamp = trace_clock_local();
3252 prev->ftrace_timestamp = timestamp;
3254 /* only process tasks that we timestamped */
3255 if (!next->ftrace_timestamp)
3256 return;
3259 * Update all the counters in next to make up for the
3260 * time next was sleeping.
3262 timestamp -= next->ftrace_timestamp;
3264 for (index = next->curr_ret_stack; index >= 0; index--)
3265 next->ret_stack[index].calltime += timestamp;
3268 /* Allocate a return stack for each task */
3269 static int start_graph_tracing(void)
3271 struct ftrace_ret_stack **ret_stack_list;
3272 int ret, cpu;
3274 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3275 sizeof(struct ftrace_ret_stack *),
3276 GFP_KERNEL);
3278 if (!ret_stack_list)
3279 return -ENOMEM;
3281 /* The cpu_boot init_task->ret_stack will never be freed */
3282 for_each_online_cpu(cpu) {
3283 if (!idle_task(cpu)->ret_stack)
3284 ftrace_graph_init_task(idle_task(cpu));
3287 do {
3288 ret = alloc_retstack_tasklist(ret_stack_list);
3289 } while (ret == -EAGAIN);
3291 if (!ret) {
3292 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3293 if (ret)
3294 pr_info("ftrace_graph: Couldn't activate tracepoint"
3295 " probe to kernel_sched_switch\n");
3298 kfree(ret_stack_list);
3299 return ret;
3303 * Hibernation protection.
3304 * The state of the current task is too much unstable during
3305 * suspend/restore to disk. We want to protect against that.
3307 static int
3308 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3309 void *unused)
3311 switch (state) {
3312 case PM_HIBERNATION_PREPARE:
3313 pause_graph_tracing();
3314 break;
3316 case PM_POST_HIBERNATION:
3317 unpause_graph_tracing();
3318 break;
3320 return NOTIFY_DONE;
3323 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3324 trace_func_graph_ent_t entryfunc)
3326 int ret = 0;
3328 mutex_lock(&ftrace_lock);
3330 /* we currently allow only one tracer registered at a time */
3331 if (ftrace_graph_active) {
3332 ret = -EBUSY;
3333 goto out;
3336 ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3337 register_pm_notifier(&ftrace_suspend_notifier);
3339 ftrace_graph_active++;
3340 ret = start_graph_tracing();
3341 if (ret) {
3342 ftrace_graph_active--;
3343 goto out;
3346 ftrace_graph_return = retfunc;
3347 ftrace_graph_entry = entryfunc;
3349 ftrace_startup(FTRACE_START_FUNC_RET);
3351 out:
3352 mutex_unlock(&ftrace_lock);
3353 return ret;
3356 void unregister_ftrace_graph(void)
3358 mutex_lock(&ftrace_lock);
3360 if (unlikely(!ftrace_graph_active))
3361 goto out;
3363 ftrace_graph_active--;
3364 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3365 ftrace_graph_entry = ftrace_graph_entry_stub;
3366 ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3367 unregister_pm_notifier(&ftrace_suspend_notifier);
3368 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3370 out:
3371 mutex_unlock(&ftrace_lock);
3374 /* Allocate a return stack for newly created task */
3375 void ftrace_graph_init_task(struct task_struct *t)
3377 /* Make sure we do not use the parent ret_stack */
3378 t->ret_stack = NULL;
3379 t->curr_ret_stack = -1;
3381 if (ftrace_graph_active) {
3382 struct ftrace_ret_stack *ret_stack;
3384 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3385 * sizeof(struct ftrace_ret_stack),
3386 GFP_KERNEL);
3387 if (!ret_stack)
3388 return;
3389 atomic_set(&t->tracing_graph_pause, 0);
3390 atomic_set(&t->trace_overrun, 0);
3391 t->ftrace_timestamp = 0;
3392 /* make curr_ret_stack visable before we add the ret_stack */
3393 smp_wmb();
3394 t->ret_stack = ret_stack;
3398 void ftrace_graph_exit_task(struct task_struct *t)
3400 struct ftrace_ret_stack *ret_stack = t->ret_stack;
3402 t->ret_stack = NULL;
3403 /* NULL must become visible to IRQs before we free it: */
3404 barrier();
3406 kfree(ret_stack);
3409 void ftrace_graph_stop(void)
3411 ftrace_stop();
3413 #endif