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 Nadia Yvette Chambers
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>
36 #include <trace/events/sched.h>
38 #include <asm/setup.h>
40 #include "trace_output.h"
41 #include "trace_stat.h"
43 #define FTRACE_WARN_ON(cond) \
51 #define FTRACE_WARN_ON_ONCE(cond) \
54 if (WARN_ON_ONCE(___r)) \
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
65 #define FL_GLOBAL_CONTROL_MASK (FTRACE_OPS_FL_CONTROL)
67 #ifdef CONFIG_DYNAMIC_FTRACE
68 #define INIT_OPS_HASH(opsname) \
69 .func_hash = &opsname.local_hash, \
70 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
72 #define INIT_OPS_HASH(opsname)
75 static struct ftrace_ops ftrace_list_end __read_mostly = {
77 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
78 INIT_OPS_HASH(ftrace_list_end)
81 /* ftrace_enabled is a method to turn ftrace on or off */
82 int ftrace_enabled __read_mostly;
83 static int last_ftrace_enabled;
85 /* Current function tracing op */
86 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
87 /* What to set function_trace_op to */
88 static struct ftrace_ops *set_function_trace_op;
90 /* List for set_ftrace_pid's pids. */
91 LIST_HEAD(ftrace_pids);
93 struct list_head list;
98 * ftrace_disabled is set when an anomaly is discovered.
99 * ftrace_disabled is much stronger than ftrace_enabled.
101 static int ftrace_disabled __read_mostly;
103 static DEFINE_MUTEX(ftrace_lock);
105 static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end;
106 static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
107 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
108 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
109 static struct ftrace_ops global_ops;
110 static struct ftrace_ops control_ops;
112 #if ARCH_SUPPORTS_FTRACE_OPS
113 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
114 struct ftrace_ops *op, struct pt_regs *regs);
116 /* See comment below, where ftrace_ops_list_func is defined */
117 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
118 #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
122 * Traverse the ftrace_global_list, invoking all entries. The reason that we
123 * can use rcu_dereference_raw_notrace() is that elements removed from this list
124 * are simply leaked, so there is no need to interact with a grace-period
125 * mechanism. The rcu_dereference_raw_notrace() calls are needed to handle
126 * concurrent insertions into the ftrace_global_list.
128 * Silly Alpha and silly pointer-speculation compiler optimizations!
130 #define do_for_each_ftrace_op(op, list) \
131 op = rcu_dereference_raw_notrace(list); \
135 * Optimized for just a single item in the list (as that is the normal case).
137 #define while_for_each_ftrace_op(op) \
138 while (likely(op = rcu_dereference_raw_notrace((op)->next)) && \
139 unlikely((op) != &ftrace_list_end))
141 static inline void ftrace_ops_init(struct ftrace_ops *ops)
143 #ifdef CONFIG_DYNAMIC_FTRACE
144 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
145 mutex_init(&ops->local_hash.regex_lock);
146 ops->func_hash = &ops->local_hash;
147 ops->flags |= FTRACE_OPS_FL_INITIALIZED;
153 * ftrace_nr_registered_ops - return number of ops registered
155 * Returns the number of ftrace_ops registered and tracing functions
157 int ftrace_nr_registered_ops(void)
159 struct ftrace_ops *ops;
162 mutex_lock(&ftrace_lock);
164 for (ops = ftrace_ops_list;
165 ops != &ftrace_list_end; ops = ops->next)
168 mutex_unlock(&ftrace_lock);
173 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
174 struct ftrace_ops *op, struct pt_regs *regs)
176 if (!test_tsk_trace_trace(current))
179 ftrace_pid_function(ip, parent_ip, op, regs);
182 static void set_ftrace_pid_function(ftrace_func_t func)
184 /* do not set ftrace_pid_function to itself! */
185 if (func != ftrace_pid_func)
186 ftrace_pid_function = func;
190 * clear_ftrace_function - reset the ftrace function
192 * This NULLs the ftrace function and in essence stops
193 * tracing. There may be lag
195 void clear_ftrace_function(void)
197 ftrace_trace_function = ftrace_stub;
198 ftrace_pid_function = ftrace_stub;
201 static void control_ops_disable_all(struct ftrace_ops *ops)
205 for_each_possible_cpu(cpu)
206 *per_cpu_ptr(ops->disabled, cpu) = 1;
209 static int control_ops_alloc(struct ftrace_ops *ops)
211 int __percpu *disabled;
213 disabled = alloc_percpu(int);
217 ops->disabled = disabled;
218 control_ops_disable_all(ops);
222 static void ftrace_sync(struct work_struct *work)
225 * This function is just a stub to implement a hard force
226 * of synchronize_sched(). This requires synchronizing
227 * tasks even in userspace and idle.
229 * Yes, function tracing is rude.
233 static void ftrace_sync_ipi(void *data)
235 /* Probably not needed, but do it anyway */
239 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
240 static void update_function_graph_func(void);
242 static inline void update_function_graph_func(void) { }
245 static void update_ftrace_function(void)
250 * If we are at the end of the list and this ops is
251 * recursion safe and not dynamic and the arch supports passing ops,
252 * then have the mcount trampoline call the function directly.
254 if (ftrace_ops_list == &ftrace_list_end ||
255 (ftrace_ops_list->next == &ftrace_list_end &&
256 !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC) &&
257 (ftrace_ops_list->flags & FTRACE_OPS_FL_RECURSION_SAFE) &&
258 !FTRACE_FORCE_LIST_FUNC)) {
259 /* Set the ftrace_ops that the arch callback uses */
260 set_function_trace_op = ftrace_ops_list;
261 func = ftrace_ops_list->func;
263 /* Just use the default ftrace_ops */
264 set_function_trace_op = &ftrace_list_end;
265 func = ftrace_ops_list_func;
268 update_function_graph_func();
270 /* If there's no change, then do nothing more here */
271 if (ftrace_trace_function == func)
275 * If we are using the list function, it doesn't care
276 * about the function_trace_ops.
278 if (func == ftrace_ops_list_func) {
279 ftrace_trace_function = func;
281 * Don't even bother setting function_trace_ops,
282 * it would be racy to do so anyway.
287 #ifndef CONFIG_DYNAMIC_FTRACE
289 * For static tracing, we need to be a bit more careful.
290 * The function change takes affect immediately. Thus,
291 * we need to coorditate the setting of the function_trace_ops
292 * with the setting of the ftrace_trace_function.
294 * Set the function to the list ops, which will call the
295 * function we want, albeit indirectly, but it handles the
296 * ftrace_ops and doesn't depend on function_trace_op.
298 ftrace_trace_function = ftrace_ops_list_func;
300 * Make sure all CPUs see this. Yes this is slow, but static
301 * tracing is slow and nasty to have enabled.
303 schedule_on_each_cpu(ftrace_sync);
304 /* Now all cpus are using the list ops. */
305 function_trace_op = set_function_trace_op;
306 /* Make sure the function_trace_op is visible on all CPUs */
308 /* Nasty way to force a rmb on all cpus */
309 smp_call_function(ftrace_sync_ipi, NULL, 1);
310 /* OK, we are all set to update the ftrace_trace_function now! */
311 #endif /* !CONFIG_DYNAMIC_FTRACE */
313 ftrace_trace_function = func;
316 int using_ftrace_ops_list_func(void)
318 return ftrace_trace_function == ftrace_ops_list_func;
321 static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
325 * We are entering ops into the list but another
326 * CPU might be walking that list. We need to make sure
327 * the ops->next pointer is valid before another CPU sees
328 * the ops pointer included into the list.
330 rcu_assign_pointer(*list, ops);
333 static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
335 struct ftrace_ops **p;
338 * If we are removing the last function, then simply point
339 * to the ftrace_stub.
341 if (*list == ops && ops->next == &ftrace_list_end) {
342 *list = &ftrace_list_end;
346 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
357 static void add_ftrace_list_ops(struct ftrace_ops **list,
358 struct ftrace_ops *main_ops,
359 struct ftrace_ops *ops)
361 int first = *list == &ftrace_list_end;
362 add_ftrace_ops(list, ops);
364 add_ftrace_ops(&ftrace_ops_list, main_ops);
367 static int remove_ftrace_list_ops(struct ftrace_ops **list,
368 struct ftrace_ops *main_ops,
369 struct ftrace_ops *ops)
371 int ret = remove_ftrace_ops(list, ops);
372 if (!ret && *list == &ftrace_list_end)
373 ret = remove_ftrace_ops(&ftrace_ops_list, main_ops);
377 static int __register_ftrace_function(struct ftrace_ops *ops)
379 if (ops->flags & FTRACE_OPS_FL_DELETED)
382 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
385 #ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
387 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
388 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
389 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
391 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
392 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
395 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
396 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
399 if (!core_kernel_data((unsigned long)ops))
400 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
402 if (ops->flags & FTRACE_OPS_FL_CONTROL) {
403 if (control_ops_alloc(ops))
405 add_ftrace_list_ops(&ftrace_control_list, &control_ops, ops);
407 add_ftrace_ops(&ftrace_ops_list, ops);
410 update_ftrace_function();
415 static int __unregister_ftrace_function(struct ftrace_ops *ops)
419 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
422 if (ops->flags & FTRACE_OPS_FL_CONTROL) {
423 ret = remove_ftrace_list_ops(&ftrace_control_list,
426 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
432 update_ftrace_function();
437 static void ftrace_update_pid_func(void)
439 /* Only do something if we are tracing something */
440 if (ftrace_trace_function == ftrace_stub)
443 update_ftrace_function();
446 #ifdef CONFIG_FUNCTION_PROFILER
447 struct ftrace_profile {
448 struct hlist_node node;
450 unsigned long counter;
451 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
452 unsigned long long time;
453 unsigned long long time_squared;
457 struct ftrace_profile_page {
458 struct ftrace_profile_page *next;
460 struct ftrace_profile records[];
463 struct ftrace_profile_stat {
465 struct hlist_head *hash;
466 struct ftrace_profile_page *pages;
467 struct ftrace_profile_page *start;
468 struct tracer_stat stat;
471 #define PROFILE_RECORDS_SIZE \
472 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
474 #define PROFILES_PER_PAGE \
475 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
477 static int ftrace_profile_enabled __read_mostly;
479 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
480 static DEFINE_MUTEX(ftrace_profile_lock);
482 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
484 #define FTRACE_PROFILE_HASH_BITS 10
485 #define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
488 function_stat_next(void *v, int idx)
490 struct ftrace_profile *rec = v;
491 struct ftrace_profile_page *pg;
493 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
499 if ((void *)rec >= (void *)&pg->records[pg->index]) {
503 rec = &pg->records[0];
511 static void *function_stat_start(struct tracer_stat *trace)
513 struct ftrace_profile_stat *stat =
514 container_of(trace, struct ftrace_profile_stat, stat);
516 if (!stat || !stat->start)
519 return function_stat_next(&stat->start->records[0], 0);
522 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
523 /* function graph compares on total time */
524 static int function_stat_cmp(void *p1, void *p2)
526 struct ftrace_profile *a = p1;
527 struct ftrace_profile *b = p2;
529 if (a->time < b->time)
531 if (a->time > b->time)
537 /* not function graph compares against hits */
538 static int function_stat_cmp(void *p1, void *p2)
540 struct ftrace_profile *a = p1;
541 struct ftrace_profile *b = p2;
543 if (a->counter < b->counter)
545 if (a->counter > b->counter)
552 static int function_stat_headers(struct seq_file *m)
554 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
555 seq_printf(m, " Function "
558 "--- ---- --- ---\n");
560 seq_printf(m, " Function Hit\n"
566 static int function_stat_show(struct seq_file *m, void *v)
568 struct ftrace_profile *rec = v;
569 char str[KSYM_SYMBOL_LEN];
571 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
572 static struct trace_seq s;
573 unsigned long long avg;
574 unsigned long long stddev;
576 mutex_lock(&ftrace_profile_lock);
578 /* we raced with function_profile_reset() */
579 if (unlikely(rec->counter == 0)) {
584 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
585 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
587 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
590 do_div(avg, rec->counter);
592 /* Sample standard deviation (s^2) */
593 if (rec->counter <= 1)
597 * Apply Welford's method:
598 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
600 stddev = rec->counter * rec->time_squared -
601 rec->time * rec->time;
604 * Divide only 1000 for ns^2 -> us^2 conversion.
605 * trace_print_graph_duration will divide 1000 again.
607 do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
611 trace_print_graph_duration(rec->time, &s);
612 trace_seq_puts(&s, " ");
613 trace_print_graph_duration(avg, &s);
614 trace_seq_puts(&s, " ");
615 trace_print_graph_duration(stddev, &s);
616 trace_print_seq(m, &s);
620 mutex_unlock(&ftrace_profile_lock);
625 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
627 struct ftrace_profile_page *pg;
629 pg = stat->pages = stat->start;
632 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
637 memset(stat->hash, 0,
638 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
641 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
643 struct ftrace_profile_page *pg;
648 /* If we already allocated, do nothing */
652 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
656 #ifdef CONFIG_DYNAMIC_FTRACE
657 functions = ftrace_update_tot_cnt;
660 * We do not know the number of functions that exist because
661 * dynamic tracing is what counts them. With past experience
662 * we have around 20K functions. That should be more than enough.
663 * It is highly unlikely we will execute every function in
669 pg = stat->start = stat->pages;
671 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
673 for (i = 1; i < pages; i++) {
674 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
685 unsigned long tmp = (unsigned long)pg;
697 static int ftrace_profile_init_cpu(int cpu)
699 struct ftrace_profile_stat *stat;
702 stat = &per_cpu(ftrace_profile_stats, cpu);
705 /* If the profile is already created, simply reset it */
706 ftrace_profile_reset(stat);
711 * We are profiling all functions, but usually only a few thousand
712 * functions are hit. We'll make a hash of 1024 items.
714 size = FTRACE_PROFILE_HASH_SIZE;
716 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
721 /* Preallocate the function profiling pages */
722 if (ftrace_profile_pages_init(stat) < 0) {
731 static int ftrace_profile_init(void)
736 for_each_possible_cpu(cpu) {
737 ret = ftrace_profile_init_cpu(cpu);
745 /* interrupts must be disabled */
746 static struct ftrace_profile *
747 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
749 struct ftrace_profile *rec;
750 struct hlist_head *hhd;
753 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
754 hhd = &stat->hash[key];
756 if (hlist_empty(hhd))
759 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
767 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
768 struct ftrace_profile *rec)
772 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
773 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
777 * The memory is already allocated, this simply finds a new record to use.
779 static struct ftrace_profile *
780 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
782 struct ftrace_profile *rec = NULL;
784 /* prevent recursion (from NMIs) */
785 if (atomic_inc_return(&stat->disabled) != 1)
789 * Try to find the function again since an NMI
790 * could have added it
792 rec = ftrace_find_profiled_func(stat, ip);
796 if (stat->pages->index == PROFILES_PER_PAGE) {
797 if (!stat->pages->next)
799 stat->pages = stat->pages->next;
802 rec = &stat->pages->records[stat->pages->index++];
804 ftrace_add_profile(stat, rec);
807 atomic_dec(&stat->disabled);
813 function_profile_call(unsigned long ip, unsigned long parent_ip,
814 struct ftrace_ops *ops, struct pt_regs *regs)
816 struct ftrace_profile_stat *stat;
817 struct ftrace_profile *rec;
820 if (!ftrace_profile_enabled)
823 local_irq_save(flags);
825 stat = this_cpu_ptr(&ftrace_profile_stats);
826 if (!stat->hash || !ftrace_profile_enabled)
829 rec = ftrace_find_profiled_func(stat, ip);
831 rec = ftrace_profile_alloc(stat, ip);
838 local_irq_restore(flags);
841 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
842 static int profile_graph_entry(struct ftrace_graph_ent *trace)
844 function_profile_call(trace->func, 0, NULL, NULL);
848 static void profile_graph_return(struct ftrace_graph_ret *trace)
850 struct ftrace_profile_stat *stat;
851 unsigned long long calltime;
852 struct ftrace_profile *rec;
855 local_irq_save(flags);
856 stat = this_cpu_ptr(&ftrace_profile_stats);
857 if (!stat->hash || !ftrace_profile_enabled)
860 /* If the calltime was zero'd ignore it */
861 if (!trace->calltime)
864 calltime = trace->rettime - trace->calltime;
866 if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
869 index = trace->depth;
871 /* Append this call time to the parent time to subtract */
873 current->ret_stack[index - 1].subtime += calltime;
875 if (current->ret_stack[index].subtime < calltime)
876 calltime -= current->ret_stack[index].subtime;
881 rec = ftrace_find_profiled_func(stat, trace->func);
883 rec->time += calltime;
884 rec->time_squared += calltime * calltime;
888 local_irq_restore(flags);
891 static int register_ftrace_profiler(void)
893 return register_ftrace_graph(&profile_graph_return,
894 &profile_graph_entry);
897 static void unregister_ftrace_profiler(void)
899 unregister_ftrace_graph();
902 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
903 .func = function_profile_call,
904 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
905 INIT_OPS_HASH(ftrace_profile_ops)
908 static int register_ftrace_profiler(void)
910 return register_ftrace_function(&ftrace_profile_ops);
913 static void unregister_ftrace_profiler(void)
915 unregister_ftrace_function(&ftrace_profile_ops);
917 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
920 ftrace_profile_write(struct file *filp, const char __user *ubuf,
921 size_t cnt, loff_t *ppos)
926 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
932 mutex_lock(&ftrace_profile_lock);
933 if (ftrace_profile_enabled ^ val) {
935 ret = ftrace_profile_init();
941 ret = register_ftrace_profiler();
946 ftrace_profile_enabled = 1;
948 ftrace_profile_enabled = 0;
950 * unregister_ftrace_profiler calls stop_machine
951 * so this acts like an synchronize_sched.
953 unregister_ftrace_profiler();
957 mutex_unlock(&ftrace_profile_lock);
965 ftrace_profile_read(struct file *filp, char __user *ubuf,
966 size_t cnt, loff_t *ppos)
968 char buf[64]; /* big enough to hold a number */
971 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
972 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
975 static const struct file_operations ftrace_profile_fops = {
976 .open = tracing_open_generic,
977 .read = ftrace_profile_read,
978 .write = ftrace_profile_write,
979 .llseek = default_llseek,
982 /* used to initialize the real stat files */
983 static struct tracer_stat function_stats __initdata = {
985 .stat_start = function_stat_start,
986 .stat_next = function_stat_next,
987 .stat_cmp = function_stat_cmp,
988 .stat_headers = function_stat_headers,
989 .stat_show = function_stat_show
992 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
994 struct ftrace_profile_stat *stat;
995 struct dentry *entry;
1000 for_each_possible_cpu(cpu) {
1001 stat = &per_cpu(ftrace_profile_stats, cpu);
1003 /* allocate enough for function name + cpu number */
1004 name = kmalloc(32, GFP_KERNEL);
1007 * The files created are permanent, if something happens
1008 * we still do not free memory.
1011 "Could not allocate stat file for cpu %d\n",
1015 stat->stat = function_stats;
1016 snprintf(name, 32, "function%d", cpu);
1017 stat->stat.name = name;
1018 ret = register_stat_tracer(&stat->stat);
1021 "Could not register function stat for cpu %d\n",
1028 entry = debugfs_create_file("function_profile_enabled", 0644,
1029 d_tracer, NULL, &ftrace_profile_fops);
1031 pr_warning("Could not create debugfs "
1032 "'function_profile_enabled' entry\n");
1035 #else /* CONFIG_FUNCTION_PROFILER */
1036 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1039 #endif /* CONFIG_FUNCTION_PROFILER */
1041 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1043 #ifdef CONFIG_DYNAMIC_FTRACE
1045 static struct ftrace_ops *removed_ops;
1047 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1048 # error Dynamic ftrace depends on MCOUNT_RECORD
1051 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1053 struct ftrace_func_probe {
1054 struct hlist_node node;
1055 struct ftrace_probe_ops *ops;
1056 unsigned long flags;
1059 struct list_head free_list;
1062 struct ftrace_func_entry {
1063 struct hlist_node hlist;
1067 struct ftrace_hash {
1068 unsigned long size_bits;
1069 struct hlist_head *buckets;
1070 unsigned long count;
1071 struct rcu_head rcu;
1075 * We make these constant because no one should touch them,
1076 * but they are used as the default "empty hash", to avoid allocating
1077 * it all the time. These are in a read only section such that if
1078 * anyone does try to modify it, it will cause an exception.
1080 static const struct hlist_head empty_buckets[1];
1081 static const struct ftrace_hash empty_hash = {
1082 .buckets = (struct hlist_head *)empty_buckets,
1084 #define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1086 static struct ftrace_ops global_ops = {
1087 .func = ftrace_stub,
1088 .local_hash.notrace_hash = EMPTY_HASH,
1089 .local_hash.filter_hash = EMPTY_HASH,
1090 INIT_OPS_HASH(global_ops)
1091 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
1092 FTRACE_OPS_FL_INITIALIZED,
1095 struct ftrace_page {
1096 struct ftrace_page *next;
1097 struct dyn_ftrace *records;
1102 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1103 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1105 /* estimate from running different kernels */
1106 #define NR_TO_INIT 10000
1108 static struct ftrace_page *ftrace_pages_start;
1109 static struct ftrace_page *ftrace_pages;
1111 static bool __always_inline ftrace_hash_empty(struct ftrace_hash *hash)
1113 return !hash || !hash->count;
1116 static struct ftrace_func_entry *
1117 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1120 struct ftrace_func_entry *entry;
1121 struct hlist_head *hhd;
1123 if (ftrace_hash_empty(hash))
1126 if (hash->size_bits > 0)
1127 key = hash_long(ip, hash->size_bits);
1131 hhd = &hash->buckets[key];
1133 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1134 if (entry->ip == ip)
1140 static void __add_hash_entry(struct ftrace_hash *hash,
1141 struct ftrace_func_entry *entry)
1143 struct hlist_head *hhd;
1146 if (hash->size_bits)
1147 key = hash_long(entry->ip, hash->size_bits);
1151 hhd = &hash->buckets[key];
1152 hlist_add_head(&entry->hlist, hhd);
1156 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1158 struct ftrace_func_entry *entry;
1160 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1165 __add_hash_entry(hash, entry);
1171 free_hash_entry(struct ftrace_hash *hash,
1172 struct ftrace_func_entry *entry)
1174 hlist_del(&entry->hlist);
1180 remove_hash_entry(struct ftrace_hash *hash,
1181 struct ftrace_func_entry *entry)
1183 hlist_del(&entry->hlist);
1187 static void ftrace_hash_clear(struct ftrace_hash *hash)
1189 struct hlist_head *hhd;
1190 struct hlist_node *tn;
1191 struct ftrace_func_entry *entry;
1192 int size = 1 << hash->size_bits;
1198 for (i = 0; i < size; i++) {
1199 hhd = &hash->buckets[i];
1200 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1201 free_hash_entry(hash, entry);
1203 FTRACE_WARN_ON(hash->count);
1206 static void free_ftrace_hash(struct ftrace_hash *hash)
1208 if (!hash || hash == EMPTY_HASH)
1210 ftrace_hash_clear(hash);
1211 kfree(hash->buckets);
1215 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1217 struct ftrace_hash *hash;
1219 hash = container_of(rcu, struct ftrace_hash, rcu);
1220 free_ftrace_hash(hash);
1223 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1225 if (!hash || hash == EMPTY_HASH)
1227 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1230 void ftrace_free_filter(struct ftrace_ops *ops)
1232 ftrace_ops_init(ops);
1233 free_ftrace_hash(ops->func_hash->filter_hash);
1234 free_ftrace_hash(ops->func_hash->notrace_hash);
1237 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1239 struct ftrace_hash *hash;
1242 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1246 size = 1 << size_bits;
1247 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1249 if (!hash->buckets) {
1254 hash->size_bits = size_bits;
1259 static struct ftrace_hash *
1260 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1262 struct ftrace_func_entry *entry;
1263 struct ftrace_hash *new_hash;
1268 new_hash = alloc_ftrace_hash(size_bits);
1273 if (ftrace_hash_empty(hash))
1276 size = 1 << hash->size_bits;
1277 for (i = 0; i < size; i++) {
1278 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1279 ret = add_hash_entry(new_hash, entry->ip);
1285 FTRACE_WARN_ON(new_hash->count != hash->count);
1290 free_ftrace_hash(new_hash);
1295 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1297 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1300 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1301 struct ftrace_hash **dst, struct ftrace_hash *src)
1303 struct ftrace_func_entry *entry;
1304 struct hlist_node *tn;
1305 struct hlist_head *hhd;
1306 struct ftrace_hash *old_hash;
1307 struct ftrace_hash *new_hash;
1308 int size = src->count;
1313 * If the new source is empty, just free dst and assign it
1317 new_hash = EMPTY_HASH;
1322 * Make the hash size about 1/2 the # found
1324 for (size /= 2; size; size >>= 1)
1327 /* Don't allocate too much */
1328 if (bits > FTRACE_HASH_MAX_BITS)
1329 bits = FTRACE_HASH_MAX_BITS;
1331 new_hash = alloc_ftrace_hash(bits);
1335 size = 1 << src->size_bits;
1336 for (i = 0; i < size; i++) {
1337 hhd = &src->buckets[i];
1338 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1339 remove_hash_entry(src, entry);
1340 __add_hash_entry(new_hash, entry);
1346 * Remove the current set, update the hash and add
1349 ftrace_hash_rec_disable_modify(ops, enable);
1352 rcu_assign_pointer(*dst, new_hash);
1353 free_ftrace_hash_rcu(old_hash);
1355 ftrace_hash_rec_enable_modify(ops, enable);
1361 * Test the hashes for this ops to see if we want to call
1362 * the ops->func or not.
1364 * It's a match if the ip is in the ops->filter_hash or
1365 * the filter_hash does not exist or is empty,
1367 * the ip is not in the ops->notrace_hash.
1369 * This needs to be called with preemption disabled as
1370 * the hashes are freed with call_rcu_sched().
1373 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1375 struct ftrace_hash *filter_hash;
1376 struct ftrace_hash *notrace_hash;
1379 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1381 * There's a small race when adding ops that the ftrace handler
1382 * that wants regs, may be called without them. We can not
1383 * allow that handler to be called if regs is NULL.
1385 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1389 filter_hash = rcu_dereference_raw_notrace(ops->func_hash->filter_hash);
1390 notrace_hash = rcu_dereference_raw_notrace(ops->func_hash->notrace_hash);
1392 if ((ftrace_hash_empty(filter_hash) ||
1393 ftrace_lookup_ip(filter_hash, ip)) &&
1394 (ftrace_hash_empty(notrace_hash) ||
1395 !ftrace_lookup_ip(notrace_hash, ip)))
1404 * This is a double for. Do not use 'break' to break out of the loop,
1405 * you must use a goto.
1407 #define do_for_each_ftrace_rec(pg, rec) \
1408 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1410 for (_____i = 0; _____i < pg->index; _____i++) { \
1411 rec = &pg->records[_____i];
1413 #define while_for_each_ftrace_rec() \
1418 static int ftrace_cmp_recs(const void *a, const void *b)
1420 const struct dyn_ftrace *key = a;
1421 const struct dyn_ftrace *rec = b;
1423 if (key->flags < rec->ip)
1425 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1430 static unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1432 struct ftrace_page *pg;
1433 struct dyn_ftrace *rec;
1434 struct dyn_ftrace key;
1437 key.flags = end; /* overload flags, as it is unsigned long */
1439 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1440 if (end < pg->records[0].ip ||
1441 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1443 rec = bsearch(&key, pg->records, pg->index,
1444 sizeof(struct dyn_ftrace),
1454 * ftrace_location - return true if the ip giving is a traced location
1455 * @ip: the instruction pointer to check
1457 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1458 * That is, the instruction that is either a NOP or call to
1459 * the function tracer. It checks the ftrace internal tables to
1460 * determine if the address belongs or not.
1462 unsigned long ftrace_location(unsigned long ip)
1464 return ftrace_location_range(ip, ip);
1468 * ftrace_text_reserved - return true if range contains an ftrace location
1469 * @start: start of range to search
1470 * @end: end of range to search (inclusive). @end points to the last byte to check.
1472 * Returns 1 if @start and @end contains a ftrace location.
1473 * That is, the instruction that is either a NOP or call to
1474 * the function tracer. It checks the ftrace internal tables to
1475 * determine if the address belongs or not.
1477 int ftrace_text_reserved(const void *start, const void *end)
1481 ret = ftrace_location_range((unsigned long)start,
1482 (unsigned long)end);
1487 /* Test if ops registered to this rec needs regs */
1488 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1490 struct ftrace_ops *ops;
1491 bool keep_regs = false;
1493 for (ops = ftrace_ops_list;
1494 ops != &ftrace_list_end; ops = ops->next) {
1495 /* pass rec in as regs to have non-NULL val */
1496 if (ftrace_ops_test(ops, rec->ip, rec)) {
1497 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1507 static void ftrace_remove_tramp(struct ftrace_ops *ops,
1508 struct dyn_ftrace *rec)
1510 /* If TRAMP is not set, no ops should have a trampoline for this */
1511 if (!(rec->flags & FTRACE_FL_TRAMP))
1514 rec->flags &= ~FTRACE_FL_TRAMP;
1516 if ((!ftrace_hash_empty(ops->func_hash->filter_hash) &&
1517 !ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip)) ||
1518 ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
1521 * The tramp_hash entry will be removed at time
1524 ops->nr_trampolines--;
1527 static void ftrace_clear_tramps(struct dyn_ftrace *rec, struct ftrace_ops *ops)
1529 struct ftrace_ops *op;
1531 /* If TRAMP is not set, no ops should have a trampoline for this */
1532 if (!(rec->flags & FTRACE_FL_TRAMP))
1535 do_for_each_ftrace_op(op, ftrace_ops_list) {
1537 * This function is called to clear other tramps
1538 * not the one that is being updated.
1542 if (op->nr_trampolines)
1543 ftrace_remove_tramp(op, rec);
1544 } while_for_each_ftrace_op(op);
1547 static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1551 struct ftrace_hash *hash;
1552 struct ftrace_hash *other_hash;
1553 struct ftrace_page *pg;
1554 struct dyn_ftrace *rec;
1558 /* Only update if the ops has been registered */
1559 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1563 * In the filter_hash case:
1564 * If the count is zero, we update all records.
1565 * Otherwise we just update the items in the hash.
1567 * In the notrace_hash case:
1568 * We enable the update in the hash.
1569 * As disabling notrace means enabling the tracing,
1570 * and enabling notrace means disabling, the inc variable
1574 hash = ops->func_hash->filter_hash;
1575 other_hash = ops->func_hash->notrace_hash;
1576 if (ftrace_hash_empty(hash))
1580 hash = ops->func_hash->notrace_hash;
1581 other_hash = ops->func_hash->filter_hash;
1583 * If the notrace hash has no items,
1584 * then there's nothing to do.
1586 if (ftrace_hash_empty(hash))
1590 do_for_each_ftrace_rec(pg, rec) {
1591 int in_other_hash = 0;
1597 * Only the filter_hash affects all records.
1598 * Update if the record is not in the notrace hash.
1600 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1603 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1604 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1607 * If filter_hash is set, we want to match all functions
1608 * that are in the hash but not in the other hash.
1610 * If filter_hash is not set, then we are decrementing.
1611 * That means we match anything that is in the hash
1612 * and also in the other_hash. That is, we need to turn
1613 * off functions in the other hash because they are disabled
1616 if (filter_hash && in_hash && !in_other_hash)
1618 else if (!filter_hash && in_hash &&
1619 (in_other_hash || ftrace_hash_empty(other_hash)))
1627 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1631 * If there's only a single callback registered to a
1632 * function, and the ops has a trampoline registered
1633 * for it, then we can call it directly.
1635 if (ftrace_rec_count(rec) == 1 && ops->trampoline) {
1636 rec->flags |= FTRACE_FL_TRAMP;
1637 ops->nr_trampolines++;
1640 * If we are adding another function callback
1641 * to this function, and the previous had a
1642 * custom trampoline in use, then we need to go
1643 * back to the default trampoline.
1645 ftrace_clear_tramps(rec, ops);
1649 * If any ops wants regs saved for this function
1650 * then all ops will get saved regs.
1652 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1653 rec->flags |= FTRACE_FL_REGS;
1655 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1659 if (ops->trampoline && !ftrace_rec_count(rec))
1660 ftrace_remove_tramp(ops, rec);
1663 * If the rec had REGS enabled and the ops that is
1664 * being removed had REGS set, then see if there is
1665 * still any ops for this record that wants regs.
1666 * If not, we can stop recording them.
1668 if (ftrace_rec_count(rec) > 0 &&
1669 rec->flags & FTRACE_FL_REGS &&
1670 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1671 if (!test_rec_ops_needs_regs(rec))
1672 rec->flags &= ~FTRACE_FL_REGS;
1676 * flags will be cleared in ftrace_check_record()
1677 * if rec count is zero.
1681 /* Shortcut, if we handled all records, we are done. */
1682 if (!all && count == hash->count)
1684 } while_for_each_ftrace_rec();
1687 static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1690 __ftrace_hash_rec_update(ops, filter_hash, 0);
1693 static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1696 __ftrace_hash_rec_update(ops, filter_hash, 1);
1699 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1700 int filter_hash, int inc)
1702 struct ftrace_ops *op;
1704 __ftrace_hash_rec_update(ops, filter_hash, inc);
1706 if (ops->func_hash != &global_ops.local_hash)
1710 * If the ops shares the global_ops hash, then we need to update
1711 * all ops that are enabled and use this hash.
1713 do_for_each_ftrace_op(op, ftrace_ops_list) {
1717 if (op->func_hash == &global_ops.local_hash)
1718 __ftrace_hash_rec_update(op, filter_hash, inc);
1719 } while_for_each_ftrace_op(op);
1722 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1725 ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1728 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1731 ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1734 static void print_ip_ins(const char *fmt, unsigned char *p)
1738 printk(KERN_CONT "%s", fmt);
1740 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1741 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1745 * ftrace_bug - report and shutdown function tracer
1746 * @failed: The failed type (EFAULT, EINVAL, EPERM)
1747 * @ip: The address that failed
1749 * The arch code that enables or disables the function tracing
1750 * can call ftrace_bug() when it has detected a problem in
1751 * modifying the code. @failed should be one of either:
1752 * EFAULT - if the problem happens on reading the @ip address
1753 * EINVAL - if what is read at @ip is not what was expected
1754 * EPERM - if the problem happens on writting to the @ip address
1756 void ftrace_bug(int failed, unsigned long ip)
1760 FTRACE_WARN_ON_ONCE(1);
1761 pr_info("ftrace faulted on modifying ");
1765 FTRACE_WARN_ON_ONCE(1);
1766 pr_info("ftrace failed to modify ");
1768 print_ip_ins(" actual: ", (unsigned char *)ip);
1769 printk(KERN_CONT "\n");
1772 FTRACE_WARN_ON_ONCE(1);
1773 pr_info("ftrace faulted on writing ");
1777 FTRACE_WARN_ON_ONCE(1);
1778 pr_info("ftrace faulted on unknown error ");
1783 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1785 unsigned long flag = 0UL;
1788 * If we are updating calls:
1790 * If the record has a ref count, then we need to enable it
1791 * because someone is using it.
1793 * Otherwise we make sure its disabled.
1795 * If we are disabling calls, then disable all records that
1798 if (enable && ftrace_rec_count(rec))
1799 flag = FTRACE_FL_ENABLED;
1802 * If enabling and the REGS flag does not match the REGS_EN, or
1803 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
1804 * this record. Set flags to fail the compare against ENABLED.
1807 if (!(rec->flags & FTRACE_FL_REGS) !=
1808 !(rec->flags & FTRACE_FL_REGS_EN))
1809 flag |= FTRACE_FL_REGS;
1811 if (!(rec->flags & FTRACE_FL_TRAMP) !=
1812 !(rec->flags & FTRACE_FL_TRAMP_EN))
1813 flag |= FTRACE_FL_TRAMP;
1816 /* If the state of this record hasn't changed, then do nothing */
1817 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1818 return FTRACE_UPDATE_IGNORE;
1821 /* Save off if rec is being enabled (for return value) */
1822 flag ^= rec->flags & FTRACE_FL_ENABLED;
1825 rec->flags |= FTRACE_FL_ENABLED;
1826 if (flag & FTRACE_FL_REGS) {
1827 if (rec->flags & FTRACE_FL_REGS)
1828 rec->flags |= FTRACE_FL_REGS_EN;
1830 rec->flags &= ~FTRACE_FL_REGS_EN;
1832 if (flag & FTRACE_FL_TRAMP) {
1833 if (rec->flags & FTRACE_FL_TRAMP)
1834 rec->flags |= FTRACE_FL_TRAMP_EN;
1836 rec->flags &= ~FTRACE_FL_TRAMP_EN;
1841 * If this record is being updated from a nop, then
1842 * return UPDATE_MAKE_CALL.
1844 * return UPDATE_MODIFY_CALL to tell the caller to convert
1845 * from the save regs, to a non-save regs function or
1846 * vice versa, or from a trampoline call.
1848 if (flag & FTRACE_FL_ENABLED)
1849 return FTRACE_UPDATE_MAKE_CALL;
1851 return FTRACE_UPDATE_MODIFY_CALL;
1855 /* If there's no more users, clear all flags */
1856 if (!ftrace_rec_count(rec))
1859 /* Just disable the record (keep REGS state) */
1860 rec->flags &= ~FTRACE_FL_ENABLED;
1863 return FTRACE_UPDATE_MAKE_NOP;
1867 * ftrace_update_record, set a record that now is tracing or not
1868 * @rec: the record to update
1869 * @enable: set to 1 if the record is tracing, zero to force disable
1871 * The records that represent all functions that can be traced need
1872 * to be updated when tracing has been enabled.
1874 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
1876 return ftrace_check_record(rec, enable, 1);
1880 * ftrace_test_record, check if the record has been enabled or not
1881 * @rec: the record to test
1882 * @enable: set to 1 to check if enabled, 0 if it is disabled
1884 * The arch code may need to test if a record is already set to
1885 * tracing to determine how to modify the function code that it
1888 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1890 return ftrace_check_record(rec, enable, 0);
1893 static struct ftrace_ops *
1894 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
1896 struct ftrace_ops *op;
1898 /* Removed ops need to be tested first */
1899 if (removed_ops && removed_ops->tramp_hash) {
1900 if (ftrace_lookup_ip(removed_ops->tramp_hash, rec->ip))
1904 do_for_each_ftrace_op(op, ftrace_ops_list) {
1905 if (!op->tramp_hash)
1908 if (ftrace_lookup_ip(op->tramp_hash, rec->ip))
1911 } while_for_each_ftrace_op(op);
1916 static struct ftrace_ops *
1917 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
1919 struct ftrace_ops *op;
1921 do_for_each_ftrace_op(op, ftrace_ops_list) {
1922 /* pass rec in as regs to have non-NULL val */
1923 if (ftrace_ops_test(op, rec->ip, rec))
1925 } while_for_each_ftrace_op(op);
1931 * ftrace_get_addr_new - Get the call address to set to
1932 * @rec: The ftrace record descriptor
1934 * If the record has the FTRACE_FL_REGS set, that means that it
1935 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
1936 * is not not set, then it wants to convert to the normal callback.
1938 * Returns the address of the trampoline to set to
1940 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
1942 struct ftrace_ops *ops;
1944 /* Trampolines take precedence over regs */
1945 if (rec->flags & FTRACE_FL_TRAMP) {
1946 ops = ftrace_find_tramp_ops_new(rec);
1947 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
1948 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
1949 (void *)rec->ip, (void *)rec->ip, rec->flags);
1950 /* Ftrace is shutting down, return anything */
1951 return (unsigned long)FTRACE_ADDR;
1953 return ops->trampoline;
1956 if (rec->flags & FTRACE_FL_REGS)
1957 return (unsigned long)FTRACE_REGS_ADDR;
1959 return (unsigned long)FTRACE_ADDR;
1963 * ftrace_get_addr_curr - Get the call address that is already there
1964 * @rec: The ftrace record descriptor
1966 * The FTRACE_FL_REGS_EN is set when the record already points to
1967 * a function that saves all the regs. Basically the '_EN' version
1968 * represents the current state of the function.
1970 * Returns the address of the trampoline that is currently being called
1972 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
1974 struct ftrace_ops *ops;
1976 /* Trampolines take precedence over regs */
1977 if (rec->flags & FTRACE_FL_TRAMP_EN) {
1978 ops = ftrace_find_tramp_ops_curr(rec);
1979 if (FTRACE_WARN_ON(!ops)) {
1980 pr_warning("Bad trampoline accounting at: %p (%pS)\n",
1981 (void *)rec->ip, (void *)rec->ip);
1982 /* Ftrace is shutting down, return anything */
1983 return (unsigned long)FTRACE_ADDR;
1985 return ops->trampoline;
1988 if (rec->flags & FTRACE_FL_REGS_EN)
1989 return (unsigned long)FTRACE_REGS_ADDR;
1991 return (unsigned long)FTRACE_ADDR;
1995 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1997 unsigned long ftrace_old_addr;
1998 unsigned long ftrace_addr;
2001 ftrace_addr = ftrace_get_addr_new(rec);
2003 /* This needs to be done before we call ftrace_update_record */
2004 ftrace_old_addr = ftrace_get_addr_curr(rec);
2006 ret = ftrace_update_record(rec, enable);
2009 case FTRACE_UPDATE_IGNORE:
2012 case FTRACE_UPDATE_MAKE_CALL:
2013 return ftrace_make_call(rec, ftrace_addr);
2015 case FTRACE_UPDATE_MAKE_NOP:
2016 return ftrace_make_nop(NULL, rec, ftrace_addr);
2018 case FTRACE_UPDATE_MODIFY_CALL:
2019 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2022 return -1; /* unknow ftrace bug */
2025 void __weak ftrace_replace_code(int enable)
2027 struct dyn_ftrace *rec;
2028 struct ftrace_page *pg;
2031 if (unlikely(ftrace_disabled))
2034 do_for_each_ftrace_rec(pg, rec) {
2035 failed = __ftrace_replace_code(rec, enable);
2037 ftrace_bug(failed, rec->ip);
2038 /* Stop processing */
2041 } while_for_each_ftrace_rec();
2044 struct ftrace_rec_iter {
2045 struct ftrace_page *pg;
2050 * ftrace_rec_iter_start, start up iterating over traced functions
2052 * Returns an iterator handle that is used to iterate over all
2053 * the records that represent address locations where functions
2056 * May return NULL if no records are available.
2058 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2061 * We only use a single iterator.
2062 * Protected by the ftrace_lock mutex.
2064 static struct ftrace_rec_iter ftrace_rec_iter;
2065 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2067 iter->pg = ftrace_pages_start;
2070 /* Could have empty pages */
2071 while (iter->pg && !iter->pg->index)
2072 iter->pg = iter->pg->next;
2081 * ftrace_rec_iter_next, get the next record to process.
2082 * @iter: The handle to the iterator.
2084 * Returns the next iterator after the given iterator @iter.
2086 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2090 if (iter->index >= iter->pg->index) {
2091 iter->pg = iter->pg->next;
2094 /* Could have empty pages */
2095 while (iter->pg && !iter->pg->index)
2096 iter->pg = iter->pg->next;
2106 * ftrace_rec_iter_record, get the record at the iterator location
2107 * @iter: The current iterator location
2109 * Returns the record that the current @iter is at.
2111 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2113 return &iter->pg->records[iter->index];
2117 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2124 if (unlikely(ftrace_disabled))
2127 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2129 ftrace_bug(ret, ip);
2136 * archs can override this function if they must do something
2137 * before the modifying code is performed.
2139 int __weak ftrace_arch_code_modify_prepare(void)
2145 * archs can override this function if they must do something
2146 * after the modifying code is performed.
2148 int __weak ftrace_arch_code_modify_post_process(void)
2153 void ftrace_modify_all_code(int command)
2155 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2159 * If the ftrace_caller calls a ftrace_ops func directly,
2160 * we need to make sure that it only traces functions it
2161 * expects to trace. When doing the switch of functions,
2162 * we need to update to the ftrace_ops_list_func first
2163 * before the transition between old and new calls are set,
2164 * as the ftrace_ops_list_func will check the ops hashes
2165 * to make sure the ops are having the right functions
2169 err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2170 if (FTRACE_WARN_ON(err))
2174 if (command & FTRACE_UPDATE_CALLS)
2175 ftrace_replace_code(1);
2176 else if (command & FTRACE_DISABLE_CALLS)
2177 ftrace_replace_code(0);
2179 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2180 function_trace_op = set_function_trace_op;
2182 /* If irqs are disabled, we are in stop machine */
2183 if (!irqs_disabled())
2184 smp_call_function(ftrace_sync_ipi, NULL, 1);
2185 err = ftrace_update_ftrace_func(ftrace_trace_function);
2186 if (FTRACE_WARN_ON(err))
2190 if (command & FTRACE_START_FUNC_RET)
2191 err = ftrace_enable_ftrace_graph_caller();
2192 else if (command & FTRACE_STOP_FUNC_RET)
2193 err = ftrace_disable_ftrace_graph_caller();
2194 FTRACE_WARN_ON(err);
2197 static int __ftrace_modify_code(void *data)
2199 int *command = data;
2201 ftrace_modify_all_code(*command);
2207 * ftrace_run_stop_machine, go back to the stop machine method
2208 * @command: The command to tell ftrace what to do
2210 * If an arch needs to fall back to the stop machine method, the
2211 * it can call this function.
2213 void ftrace_run_stop_machine(int command)
2215 stop_machine(__ftrace_modify_code, &command, NULL);
2219 * arch_ftrace_update_code, modify the code to trace or not trace
2220 * @command: The command that needs to be done
2222 * Archs can override this function if it does not need to
2223 * run stop_machine() to modify code.
2225 void __weak arch_ftrace_update_code(int command)
2227 ftrace_run_stop_machine(command);
2230 static int ftrace_save_ops_tramp_hash(struct ftrace_ops *ops)
2232 struct ftrace_page *pg;
2233 struct dyn_ftrace *rec;
2237 size = ops->nr_trampolines;
2240 * Make the hash size about 1/2 the # found
2242 for (size /= 2; size; size >>= 1)
2245 ops->tramp_hash = alloc_ftrace_hash(bits);
2247 * TODO: a failed allocation is going to screw up
2248 * the accounting of what needs to be modified
2249 * and not. For now, we kill ftrace if we fail
2250 * to allocate here. But there are ways around this,
2251 * but that will take a little more work.
2253 if (!ops->tramp_hash)
2256 do_for_each_ftrace_rec(pg, rec) {
2257 if (ftrace_rec_count(rec) == 1 &&
2258 ftrace_ops_test(ops, rec->ip, rec)) {
2261 * If another ops adds to a rec, the rec will
2262 * lose its trampoline and never get it back
2263 * until all ops are off of it.
2265 if (!(rec->flags & FTRACE_FL_TRAMP))
2268 /* This record had better have a trampoline */
2269 if (FTRACE_WARN_ON(!(rec->flags & FTRACE_FL_TRAMP_EN)))
2272 ret = add_hash_entry(ops->tramp_hash, rec->ip);
2276 } while_for_each_ftrace_rec();
2278 /* The number of recs in the hash must match nr_trampolines */
2279 if (FTRACE_WARN_ON(ops->tramp_hash->count != ops->nr_trampolines))
2280 pr_warn("count=%ld trampolines=%d\n",
2281 ops->tramp_hash->count,
2282 ops->nr_trampolines);
2287 static int ftrace_save_tramp_hashes(void)
2289 struct ftrace_ops *op;
2293 * Now that any trampoline is being used, we need to save the
2294 * hashes for the ops that have them. This allows the mapping
2295 * back from the record to the ops that has the trampoline to
2296 * know what code is being replaced. Modifying code must always
2297 * verify what it is changing.
2299 do_for_each_ftrace_op(op, ftrace_ops_list) {
2301 /* The tramp_hash is recreated each time. */
2302 free_ftrace_hash(op->tramp_hash);
2303 op->tramp_hash = NULL;
2305 if (op->nr_trampolines) {
2306 ret = ftrace_save_ops_tramp_hash(op);
2311 } while_for_each_ftrace_op(op);
2316 static void ftrace_run_update_code(int command)
2320 ret = ftrace_arch_code_modify_prepare();
2321 FTRACE_WARN_ON(ret);
2326 * By default we use stop_machine() to modify the code.
2327 * But archs can do what ever they want as long as it
2328 * is safe. The stop_machine() is the safest, but also
2329 * produces the most overhead.
2331 arch_ftrace_update_code(command);
2333 ret = ftrace_arch_code_modify_post_process();
2334 FTRACE_WARN_ON(ret);
2336 ret = ftrace_save_tramp_hashes();
2337 FTRACE_WARN_ON(ret);
2340 static ftrace_func_t saved_ftrace_func;
2341 static int ftrace_start_up;
2343 static void control_ops_free(struct ftrace_ops *ops)
2345 free_percpu(ops->disabled);
2348 static void ftrace_startup_enable(int command)
2350 if (saved_ftrace_func != ftrace_trace_function) {
2351 saved_ftrace_func = ftrace_trace_function;
2352 command |= FTRACE_UPDATE_TRACE_FUNC;
2355 if (!command || !ftrace_enabled)
2358 ftrace_run_update_code(command);
2361 static int ftrace_startup(struct ftrace_ops *ops, int command)
2365 if (unlikely(ftrace_disabled))
2368 ret = __register_ftrace_function(ops);
2373 command |= FTRACE_UPDATE_CALLS;
2375 ops->flags |= FTRACE_OPS_FL_ENABLED;
2377 ftrace_hash_rec_enable(ops, 1);
2379 ftrace_startup_enable(command);
2384 static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2388 if (unlikely(ftrace_disabled))
2391 ret = __unregister_ftrace_function(ops);
2397 * Just warn in case of unbalance, no need to kill ftrace, it's not
2398 * critical but the ftrace_call callers may be never nopped again after
2399 * further ftrace uses.
2401 WARN_ON_ONCE(ftrace_start_up < 0);
2403 ftrace_hash_rec_disable(ops, 1);
2405 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2407 command |= FTRACE_UPDATE_CALLS;
2409 if (saved_ftrace_func != ftrace_trace_function) {
2410 saved_ftrace_func = ftrace_trace_function;
2411 command |= FTRACE_UPDATE_TRACE_FUNC;
2414 if (!command || !ftrace_enabled) {
2416 * If these are control ops, they still need their
2417 * per_cpu field freed. Since, function tracing is
2418 * not currently active, we can just free them
2419 * without synchronizing all CPUs.
2421 if (ops->flags & FTRACE_OPS_FL_CONTROL)
2422 control_ops_free(ops);
2427 * If the ops uses a trampoline, then it needs to be
2428 * tested first on update.
2432 ftrace_run_update_code(command);
2437 * Dynamic ops may be freed, we must make sure that all
2438 * callers are done before leaving this function.
2439 * The same goes for freeing the per_cpu data of the control
2442 * Again, normal synchronize_sched() is not good enough.
2443 * We need to do a hard force of sched synchronization.
2444 * This is because we use preempt_disable() to do RCU, but
2445 * the function tracers can be called where RCU is not watching
2446 * (like before user_exit()). We can not rely on the RCU
2447 * infrastructure to do the synchronization, thus we must do it
2450 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_CONTROL)) {
2451 schedule_on_each_cpu(ftrace_sync);
2453 if (ops->flags & FTRACE_OPS_FL_CONTROL)
2454 control_ops_free(ops);
2460 static void ftrace_startup_sysctl(void)
2462 if (unlikely(ftrace_disabled))
2465 /* Force update next time */
2466 saved_ftrace_func = NULL;
2467 /* ftrace_start_up is true if we want ftrace running */
2468 if (ftrace_start_up)
2469 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
2472 static void ftrace_shutdown_sysctl(void)
2474 if (unlikely(ftrace_disabled))
2477 /* ftrace_start_up is true if ftrace is running */
2478 if (ftrace_start_up)
2479 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
2482 static cycle_t ftrace_update_time;
2483 unsigned long ftrace_update_tot_cnt;
2485 static inline int ops_traces_mod(struct ftrace_ops *ops)
2488 * Filter_hash being empty will default to trace module.
2489 * But notrace hash requires a test of individual module functions.
2491 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2492 ftrace_hash_empty(ops->func_hash->notrace_hash);
2496 * Check if the current ops references the record.
2498 * If the ops traces all functions, then it was already accounted for.
2499 * If the ops does not trace the current record function, skip it.
2500 * If the ops ignores the function via notrace filter, skip it.
2503 ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2505 /* If ops isn't enabled, ignore it */
2506 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2509 /* If ops traces all mods, we already accounted for it */
2510 if (ops_traces_mod(ops))
2513 /* The function must be in the filter */
2514 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2515 !ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2518 /* If in notrace hash, we ignore it too */
2519 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2525 static int referenced_filters(struct dyn_ftrace *rec)
2527 struct ftrace_ops *ops;
2530 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
2531 if (ops_references_rec(ops, rec))
2538 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2540 struct ftrace_page *pg;
2541 struct dyn_ftrace *p;
2542 cycle_t start, stop;
2543 unsigned long update_cnt = 0;
2544 unsigned long ref = 0;
2549 * When adding a module, we need to check if tracers are
2550 * currently enabled and if they are set to trace all functions.
2551 * If they are, we need to enable the module functions as well
2552 * as update the reference counts for those function records.
2555 struct ftrace_ops *ops;
2557 for (ops = ftrace_ops_list;
2558 ops != &ftrace_list_end; ops = ops->next) {
2559 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
2560 if (ops_traces_mod(ops))
2568 start = ftrace_now(raw_smp_processor_id());
2570 for (pg = new_pgs; pg; pg = pg->next) {
2572 for (i = 0; i < pg->index; i++) {
2575 /* If something went wrong, bail without enabling anything */
2576 if (unlikely(ftrace_disabled))
2579 p = &pg->records[i];
2581 cnt += referenced_filters(p);
2585 * Do the initial record conversion from mcount jump
2586 * to the NOP instructions.
2588 if (!ftrace_code_disable(mod, p))
2594 * If the tracing is enabled, go ahead and enable the record.
2596 * The reason not to enable the record immediatelly is the
2597 * inherent check of ftrace_make_nop/ftrace_make_call for
2598 * correct previous instructions. Making first the NOP
2599 * conversion puts the module to the correct state, thus
2600 * passing the ftrace_make_call check.
2602 if (ftrace_start_up && cnt) {
2603 int failed = __ftrace_replace_code(p, 1);
2605 ftrace_bug(failed, p->ip);
2610 stop = ftrace_now(raw_smp_processor_id());
2611 ftrace_update_time = stop - start;
2612 ftrace_update_tot_cnt += update_cnt;
2617 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2622 if (WARN_ON(!count))
2625 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2628 * We want to fill as much as possible. No more than a page
2631 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2635 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2638 /* if we can't allocate this size, try something smaller */
2645 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2654 static struct ftrace_page *
2655 ftrace_allocate_pages(unsigned long num_to_init)
2657 struct ftrace_page *start_pg;
2658 struct ftrace_page *pg;
2665 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2670 * Try to allocate as much as possible in one continues
2671 * location that fills in all of the space. We want to
2672 * waste as little space as possible.
2675 cnt = ftrace_allocate_records(pg, num_to_init);
2683 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2695 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2696 free_pages((unsigned long)pg->records, order);
2697 start_pg = pg->next;
2701 pr_info("ftrace: FAILED to allocate memory for functions\n");
2705 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
2707 struct ftrace_iterator {
2710 struct ftrace_page *pg;
2711 struct dyn_ftrace *func;
2712 struct ftrace_func_probe *probe;
2713 struct trace_parser parser;
2714 struct ftrace_hash *hash;
2715 struct ftrace_ops *ops;
2722 t_hash_next(struct seq_file *m, loff_t *pos)
2724 struct ftrace_iterator *iter = m->private;
2725 struct hlist_node *hnd = NULL;
2726 struct hlist_head *hhd;
2732 hnd = &iter->probe->node;
2734 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
2737 hhd = &ftrace_func_hash[iter->hidx];
2739 if (hlist_empty(hhd)) {
2755 if (WARN_ON_ONCE(!hnd))
2758 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
2763 static void *t_hash_start(struct seq_file *m, loff_t *pos)
2765 struct ftrace_iterator *iter = m->private;
2769 if (!(iter->flags & FTRACE_ITER_DO_HASH))
2772 if (iter->func_pos > *pos)
2776 for (l = 0; l <= (*pos - iter->func_pos); ) {
2777 p = t_hash_next(m, &l);
2784 /* Only set this if we have an item */
2785 iter->flags |= FTRACE_ITER_HASH;
2791 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2793 struct ftrace_func_probe *rec;
2796 if (WARN_ON_ONCE(!rec))
2799 if (rec->ops->print)
2800 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2802 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2805 seq_printf(m, ":%p", rec->data);
2812 t_next(struct seq_file *m, void *v, loff_t *pos)
2814 struct ftrace_iterator *iter = m->private;
2815 struct ftrace_ops *ops = iter->ops;
2816 struct dyn_ftrace *rec = NULL;
2818 if (unlikely(ftrace_disabled))
2821 if (iter->flags & FTRACE_ITER_HASH)
2822 return t_hash_next(m, pos);
2825 iter->pos = iter->func_pos = *pos;
2827 if (iter->flags & FTRACE_ITER_PRINTALL)
2828 return t_hash_start(m, pos);
2831 if (iter->idx >= iter->pg->index) {
2832 if (iter->pg->next) {
2833 iter->pg = iter->pg->next;
2838 rec = &iter->pg->records[iter->idx++];
2839 if (((iter->flags & FTRACE_ITER_FILTER) &&
2840 !(ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))) ||
2842 ((iter->flags & FTRACE_ITER_NOTRACE) &&
2843 !ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip)) ||
2845 ((iter->flags & FTRACE_ITER_ENABLED) &&
2846 !(rec->flags & FTRACE_FL_ENABLED))) {
2854 return t_hash_start(m, pos);
2861 static void reset_iter_read(struct ftrace_iterator *iter)
2865 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
2868 static void *t_start(struct seq_file *m, loff_t *pos)
2870 struct ftrace_iterator *iter = m->private;
2871 struct ftrace_ops *ops = iter->ops;
2875 mutex_lock(&ftrace_lock);
2877 if (unlikely(ftrace_disabled))
2881 * If an lseek was done, then reset and start from beginning.
2883 if (*pos < iter->pos)
2884 reset_iter_read(iter);
2887 * For set_ftrace_filter reading, if we have the filter
2888 * off, we can short cut and just print out that all
2889 * functions are enabled.
2891 if ((iter->flags & FTRACE_ITER_FILTER &&
2892 ftrace_hash_empty(ops->func_hash->filter_hash)) ||
2893 (iter->flags & FTRACE_ITER_NOTRACE &&
2894 ftrace_hash_empty(ops->func_hash->notrace_hash))) {
2896 return t_hash_start(m, pos);
2897 iter->flags |= FTRACE_ITER_PRINTALL;
2898 /* reset in case of seek/pread */
2899 iter->flags &= ~FTRACE_ITER_HASH;
2903 if (iter->flags & FTRACE_ITER_HASH)
2904 return t_hash_start(m, pos);
2907 * Unfortunately, we need to restart at ftrace_pages_start
2908 * every time we let go of the ftrace_mutex. This is because
2909 * those pointers can change without the lock.
2911 iter->pg = ftrace_pages_start;
2913 for (l = 0; l <= *pos; ) {
2914 p = t_next(m, p, &l);
2920 return t_hash_start(m, pos);
2925 static void t_stop(struct seq_file *m, void *p)
2927 mutex_unlock(&ftrace_lock);
2930 static int t_show(struct seq_file *m, void *v)
2932 struct ftrace_iterator *iter = m->private;
2933 struct dyn_ftrace *rec;
2935 if (iter->flags & FTRACE_ITER_HASH)
2936 return t_hash_show(m, iter);
2938 if (iter->flags & FTRACE_ITER_PRINTALL) {
2939 if (iter->flags & FTRACE_ITER_NOTRACE)
2940 seq_printf(m, "#### no functions disabled ####\n");
2942 seq_printf(m, "#### all functions enabled ####\n");
2951 seq_printf(m, "%ps", (void *)rec->ip);
2952 if (iter->flags & FTRACE_ITER_ENABLED) {
2953 seq_printf(m, " (%ld)%s",
2954 ftrace_rec_count(rec),
2955 rec->flags & FTRACE_FL_REGS ? " R" : " ");
2956 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2957 struct ftrace_ops *ops;
2959 ops = ftrace_find_tramp_ops_curr(rec);
2960 if (ops && ops->trampoline)
2961 seq_printf(m, "\ttramp: %pS",
2962 (void *)ops->trampoline);
2964 seq_printf(m, "\ttramp: ERROR!");
2968 seq_printf(m, "\n");
2973 static const struct seq_operations show_ftrace_seq_ops = {
2981 ftrace_avail_open(struct inode *inode, struct file *file)
2983 struct ftrace_iterator *iter;
2985 if (unlikely(ftrace_disabled))
2988 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2990 iter->pg = ftrace_pages_start;
2991 iter->ops = &global_ops;
2994 return iter ? 0 : -ENOMEM;
2998 ftrace_enabled_open(struct inode *inode, struct file *file)
3000 struct ftrace_iterator *iter;
3002 if (unlikely(ftrace_disabled))
3005 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3007 iter->pg = ftrace_pages_start;
3008 iter->flags = FTRACE_ITER_ENABLED;
3009 iter->ops = &global_ops;
3012 return iter ? 0 : -ENOMEM;
3016 * ftrace_regex_open - initialize function tracer filter files
3017 * @ops: The ftrace_ops that hold the hash filters
3018 * @flag: The type of filter to process
3019 * @inode: The inode, usually passed in to your open routine
3020 * @file: The file, usually passed in to your open routine
3022 * ftrace_regex_open() initializes the filter files for the
3023 * @ops. Depending on @flag it may process the filter hash or
3024 * the notrace hash of @ops. With this called from the open
3025 * routine, you can use ftrace_filter_write() for the write
3026 * routine if @flag has FTRACE_ITER_FILTER set, or
3027 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3028 * tracing_lseek() should be used as the lseek routine, and
3029 * release must call ftrace_regex_release().
3032 ftrace_regex_open(struct ftrace_ops *ops, int flag,
3033 struct inode *inode, struct file *file)
3035 struct ftrace_iterator *iter;
3036 struct ftrace_hash *hash;
3039 ftrace_ops_init(ops);
3041 if (unlikely(ftrace_disabled))
3044 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3048 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3056 mutex_lock(&ops->func_hash->regex_lock);
3058 if (flag & FTRACE_ITER_NOTRACE)
3059 hash = ops->func_hash->notrace_hash;
3061 hash = ops->func_hash->filter_hash;
3063 if (file->f_mode & FMODE_WRITE) {
3064 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3066 if (file->f_flags & O_TRUNC)
3067 iter->hash = alloc_ftrace_hash(size_bits);
3069 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3072 trace_parser_put(&iter->parser);
3079 if (file->f_mode & FMODE_READ) {
3080 iter->pg = ftrace_pages_start;
3082 ret = seq_open(file, &show_ftrace_seq_ops);
3084 struct seq_file *m = file->private_data;
3088 free_ftrace_hash(iter->hash);
3089 trace_parser_put(&iter->parser);
3093 file->private_data = iter;
3096 mutex_unlock(&ops->func_hash->regex_lock);
3102 ftrace_filter_open(struct inode *inode, struct file *file)
3104 struct ftrace_ops *ops = inode->i_private;
3106 return ftrace_regex_open(ops,
3107 FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
3112 ftrace_notrace_open(struct inode *inode, struct file *file)
3114 struct ftrace_ops *ops = inode->i_private;
3116 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3120 static int ftrace_match(char *str, char *regex, int len, int type)
3127 if (strcmp(str, regex) == 0)
3130 case MATCH_FRONT_ONLY:
3131 if (strncmp(str, regex, len) == 0)
3134 case MATCH_MIDDLE_ONLY:
3135 if (strstr(str, regex))
3138 case MATCH_END_ONLY:
3140 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
3149 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
3151 struct ftrace_func_entry *entry;
3154 entry = ftrace_lookup_ip(hash, rec->ip);
3156 /* Do nothing if it doesn't exist */
3160 free_hash_entry(hash, entry);
3162 /* Do nothing if it exists */
3166 ret = add_hash_entry(hash, rec->ip);
3172 ftrace_match_record(struct dyn_ftrace *rec, char *mod,
3173 char *regex, int len, int type)
3175 char str[KSYM_SYMBOL_LEN];
3178 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3181 /* module lookup requires matching the module */
3182 if (!modname || strcmp(modname, mod))
3185 /* blank search means to match all funcs in the mod */
3190 return ftrace_match(str, regex, len, type);
3194 match_records(struct ftrace_hash *hash, char *buff,
3195 int len, char *mod, int not)
3197 unsigned search_len = 0;
3198 struct ftrace_page *pg;
3199 struct dyn_ftrace *rec;
3200 int type = MATCH_FULL;
3201 char *search = buff;
3206 type = filter_parse_regex(buff, len, &search, ¬);
3207 search_len = strlen(search);
3210 mutex_lock(&ftrace_lock);
3212 if (unlikely(ftrace_disabled))
3215 do_for_each_ftrace_rec(pg, rec) {
3216 if (ftrace_match_record(rec, mod, search, search_len, type)) {
3217 ret = enter_record(hash, rec, not);
3224 } while_for_each_ftrace_rec();
3226 mutex_unlock(&ftrace_lock);
3232 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3234 return match_records(hash, buff, len, NULL, 0);
3238 ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
3242 /* blank or '*' mean the same */
3243 if (strcmp(buff, "*") == 0)
3246 /* handle the case of 'dont filter this module' */
3247 if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
3252 return match_records(hash, buff, strlen(buff), mod, not);
3256 * We register the module command as a template to show others how
3257 * to register the a command as well.
3261 ftrace_mod_callback(struct ftrace_hash *hash,
3262 char *func, char *cmd, char *param, int enable)
3268 * cmd == 'mod' because we only registered this func
3269 * for the 'mod' ftrace_func_command.
3270 * But if you register one func with multiple commands,
3271 * you can tell which command was used by the cmd
3275 /* we must have a module name */
3279 mod = strsep(¶m, ":");
3283 ret = ftrace_match_module_records(hash, func, mod);
3292 static struct ftrace_func_command ftrace_mod_cmd = {
3294 .func = ftrace_mod_callback,
3297 static int __init ftrace_mod_cmd_init(void)
3299 return register_ftrace_command(&ftrace_mod_cmd);
3301 core_initcall(ftrace_mod_cmd_init);
3303 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
3304 struct ftrace_ops *op, struct pt_regs *pt_regs)
3306 struct ftrace_func_probe *entry;
3307 struct hlist_head *hhd;
3310 key = hash_long(ip, FTRACE_HASH_BITS);
3312 hhd = &ftrace_func_hash[key];
3314 if (hlist_empty(hhd))
3318 * Disable preemption for these calls to prevent a RCU grace
3319 * period. This syncs the hash iteration and freeing of items
3320 * on the hash. rcu_read_lock is too dangerous here.
3322 preempt_disable_notrace();
3323 hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3324 if (entry->ip == ip)
3325 entry->ops->func(ip, parent_ip, &entry->data);
3327 preempt_enable_notrace();
3330 static struct ftrace_ops trace_probe_ops __read_mostly =
3332 .func = function_trace_probe_call,
3333 .flags = FTRACE_OPS_FL_INITIALIZED,
3334 INIT_OPS_HASH(trace_probe_ops)
3337 static int ftrace_probe_registered;
3339 static void __enable_ftrace_function_probe(void)
3344 if (ftrace_probe_registered) {
3345 /* still need to update the function call sites */
3347 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3351 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3352 struct hlist_head *hhd = &ftrace_func_hash[i];
3356 /* Nothing registered? */
3357 if (i == FTRACE_FUNC_HASHSIZE)
3360 ret = ftrace_startup(&trace_probe_ops, 0);
3362 ftrace_probe_registered = 1;
3365 static void __disable_ftrace_function_probe(void)
3369 if (!ftrace_probe_registered)
3372 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3373 struct hlist_head *hhd = &ftrace_func_hash[i];
3378 /* no more funcs left */
3379 ftrace_shutdown(&trace_probe_ops, 0);
3381 ftrace_probe_registered = 0;
3385 static void ftrace_free_entry(struct ftrace_func_probe *entry)
3387 if (entry->ops->free)
3388 entry->ops->free(entry->ops, entry->ip, &entry->data);
3393 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3396 struct ftrace_func_probe *entry;
3397 struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3398 struct ftrace_hash *hash;
3399 struct ftrace_page *pg;
3400 struct dyn_ftrace *rec;
3407 type = filter_parse_regex(glob, strlen(glob), &search, ¬);
3408 len = strlen(search);
3410 /* we do not support '!' for function probes */
3414 mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3416 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3422 if (unlikely(ftrace_disabled)) {
3427 mutex_lock(&ftrace_lock);
3429 do_for_each_ftrace_rec(pg, rec) {
3431 if (!ftrace_match_record(rec, NULL, search, len, type))
3434 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3436 /* If we did not process any, then return error */
3447 * The caller might want to do something special
3448 * for each function we find. We call the callback
3449 * to give the caller an opportunity to do so.
3452 if (ops->init(ops, rec->ip, &entry->data) < 0) {
3453 /* caller does not like this func */
3459 ret = enter_record(hash, rec, 0);
3467 entry->ip = rec->ip;
3469 key = hash_long(entry->ip, FTRACE_HASH_BITS);
3470 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3472 } while_for_each_ftrace_rec();
3474 ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3478 __enable_ftrace_function_probe();
3481 mutex_unlock(&ftrace_lock);
3483 mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3484 free_ftrace_hash(hash);
3490 PROBE_TEST_FUNC = 1,
3495 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3496 void *data, int flags)
3498 struct ftrace_func_entry *rec_entry;
3499 struct ftrace_func_probe *entry;
3500 struct ftrace_func_probe *p;
3501 struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3502 struct list_head free_list;
3503 struct ftrace_hash *hash;
3504 struct hlist_node *tmp;
3505 char str[KSYM_SYMBOL_LEN];
3506 int type = MATCH_FULL;
3510 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3515 type = filter_parse_regex(glob, strlen(glob), &search, ¬);
3516 len = strlen(search);
3518 /* we do not support '!' for function probes */
3523 mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3525 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3527 /* Hmm, should report this somehow */
3530 INIT_LIST_HEAD(&free_list);
3532 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3533 struct hlist_head *hhd = &ftrace_func_hash[i];
3535 hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3537 /* break up if statements for readability */
3538 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3541 if ((flags & PROBE_TEST_DATA) && entry->data != data)
3544 /* do this last, since it is the most expensive */
3546 kallsyms_lookup(entry->ip, NULL, NULL,
3548 if (!ftrace_match(str, glob, len, type))
3552 rec_entry = ftrace_lookup_ip(hash, entry->ip);
3553 /* It is possible more than one entry had this ip */
3555 free_hash_entry(hash, rec_entry);
3557 hlist_del_rcu(&entry->node);
3558 list_add(&entry->free_list, &free_list);
3561 mutex_lock(&ftrace_lock);
3562 __disable_ftrace_function_probe();
3564 * Remove after the disable is called. Otherwise, if the last
3565 * probe is removed, a null hash means *all enabled*.
3567 ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3568 synchronize_sched();
3569 list_for_each_entry_safe(entry, p, &free_list, free_list) {
3570 list_del(&entry->free_list);
3571 ftrace_free_entry(entry);
3573 mutex_unlock(&ftrace_lock);
3576 mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3577 free_ftrace_hash(hash);
3581 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3584 __unregister_ftrace_function_probe(glob, ops, data,
3585 PROBE_TEST_FUNC | PROBE_TEST_DATA);
3589 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3591 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3594 void unregister_ftrace_function_probe_all(char *glob)
3596 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3599 static LIST_HEAD(ftrace_commands);
3600 static DEFINE_MUTEX(ftrace_cmd_mutex);
3603 * Currently we only register ftrace commands from __init, so mark this
3606 __init int register_ftrace_command(struct ftrace_func_command *cmd)
3608 struct ftrace_func_command *p;
3611 mutex_lock(&ftrace_cmd_mutex);
3612 list_for_each_entry(p, &ftrace_commands, list) {
3613 if (strcmp(cmd->name, p->name) == 0) {
3618 list_add(&cmd->list, &ftrace_commands);
3620 mutex_unlock(&ftrace_cmd_mutex);
3626 * Currently we only unregister ftrace commands from __init, so mark
3629 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
3631 struct ftrace_func_command *p, *n;
3634 mutex_lock(&ftrace_cmd_mutex);
3635 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3636 if (strcmp(cmd->name, p->name) == 0) {
3638 list_del_init(&p->list);
3643 mutex_unlock(&ftrace_cmd_mutex);
3648 static int ftrace_process_regex(struct ftrace_hash *hash,
3649 char *buff, int len, int enable)
3651 char *func, *command, *next = buff;
3652 struct ftrace_func_command *p;
3655 func = strsep(&next, ":");
3658 ret = ftrace_match_records(hash, func, len);
3668 command = strsep(&next, ":");
3670 mutex_lock(&ftrace_cmd_mutex);
3671 list_for_each_entry(p, &ftrace_commands, list) {
3672 if (strcmp(p->name, command) == 0) {
3673 ret = p->func(hash, func, command, next, enable);
3678 mutex_unlock(&ftrace_cmd_mutex);
3684 ftrace_regex_write(struct file *file, const char __user *ubuf,
3685 size_t cnt, loff_t *ppos, int enable)
3687 struct ftrace_iterator *iter;
3688 struct trace_parser *parser;
3694 if (file->f_mode & FMODE_READ) {
3695 struct seq_file *m = file->private_data;
3698 iter = file->private_data;
3700 if (unlikely(ftrace_disabled))
3703 /* iter->hash is a local copy, so we don't need regex_lock */
3705 parser = &iter->parser;
3706 read = trace_get_user(parser, ubuf, cnt, ppos);
3708 if (read >= 0 && trace_parser_loaded(parser) &&
3709 !trace_parser_cont(parser)) {
3710 ret = ftrace_process_regex(iter->hash, parser->buffer,
3711 parser->idx, enable);
3712 trace_parser_clear(parser);
3723 ftrace_filter_write(struct file *file, const char __user *ubuf,
3724 size_t cnt, loff_t *ppos)
3726 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
3730 ftrace_notrace_write(struct file *file, const char __user *ubuf,
3731 size_t cnt, loff_t *ppos)
3733 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
3737 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
3739 struct ftrace_func_entry *entry;
3741 if (!ftrace_location(ip))
3745 entry = ftrace_lookup_ip(hash, ip);
3748 free_hash_entry(hash, entry);
3752 return add_hash_entry(hash, ip);
3755 static void ftrace_ops_update_code(struct ftrace_ops *ops)
3757 if (ops->flags & FTRACE_OPS_FL_ENABLED && ftrace_enabled)
3758 ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3762 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
3763 unsigned long ip, int remove, int reset, int enable)
3765 struct ftrace_hash **orig_hash;
3766 struct ftrace_hash *hash;
3769 if (unlikely(ftrace_disabled))
3772 mutex_lock(&ops->func_hash->regex_lock);
3775 orig_hash = &ops->func_hash->filter_hash;
3777 orig_hash = &ops->func_hash->notrace_hash;
3780 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
3782 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3786 goto out_regex_unlock;
3789 if (buf && !ftrace_match_records(hash, buf, len)) {
3791 goto out_regex_unlock;
3794 ret = ftrace_match_addr(hash, ip, remove);
3796 goto out_regex_unlock;
3799 mutex_lock(&ftrace_lock);
3800 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3802 ftrace_ops_update_code(ops);
3804 mutex_unlock(&ftrace_lock);
3807 mutex_unlock(&ops->func_hash->regex_lock);
3809 free_ftrace_hash(hash);
3814 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
3815 int reset, int enable)
3817 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
3821 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
3822 * @ops - the ops to set the filter with
3823 * @ip - the address to add to or remove from the filter.
3824 * @remove - non zero to remove the ip from the filter
3825 * @reset - non zero to reset all filters before applying this filter.
3827 * Filters denote which functions should be enabled when tracing is enabled
3828 * If @ip is NULL, it failes to update filter.
3830 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
3831 int remove, int reset)
3833 ftrace_ops_init(ops);
3834 return ftrace_set_addr(ops, ip, remove, reset, 1);
3836 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
3839 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3840 int reset, int enable)
3842 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
3846 * ftrace_set_filter - set a function to filter on in ftrace
3847 * @ops - the ops to set the filter with
3848 * @buf - the string that holds the function filter text.
3849 * @len - the length of the string.
3850 * @reset - non zero to reset all filters before applying this filter.
3852 * Filters denote which functions should be enabled when tracing is enabled.
3853 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3855 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
3858 ftrace_ops_init(ops);
3859 return ftrace_set_regex(ops, buf, len, reset, 1);
3861 EXPORT_SYMBOL_GPL(ftrace_set_filter);
3864 * ftrace_set_notrace - set a function to not trace in ftrace
3865 * @ops - the ops to set the notrace filter with
3866 * @buf - the string that holds the function notrace text.
3867 * @len - the length of the string.
3868 * @reset - non zero to reset all filters before applying this filter.
3870 * Notrace Filters denote which functions should not be enabled when tracing
3871 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3874 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
3877 ftrace_ops_init(ops);
3878 return ftrace_set_regex(ops, buf, len, reset, 0);
3880 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
3882 * ftrace_set_global_filter - set a function to filter on with global tracers
3883 * @buf - the string that holds the function filter text.
3884 * @len - the length of the string.
3885 * @reset - non zero to reset all filters before applying this filter.
3887 * Filters denote which functions should be enabled when tracing is enabled.
3888 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3890 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
3892 ftrace_set_regex(&global_ops, buf, len, reset, 1);
3894 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
3897 * ftrace_set_global_notrace - set a function to not trace with global tracers
3898 * @buf - the string that holds the function notrace text.
3899 * @len - the length of the string.
3900 * @reset - non zero to reset all filters before applying this filter.
3902 * Notrace Filters denote which functions should not be enabled when tracing
3903 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3906 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
3908 ftrace_set_regex(&global_ops, buf, len, reset, 0);
3910 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3913 * command line interface to allow users to set filters on boot up.
3915 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
3916 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3917 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3919 /* Used by function selftest to not test if filter is set */
3920 bool ftrace_filter_param __initdata;
3922 static int __init set_ftrace_notrace(char *str)
3924 ftrace_filter_param = true;
3925 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3928 __setup("ftrace_notrace=", set_ftrace_notrace);
3930 static int __init set_ftrace_filter(char *str)
3932 ftrace_filter_param = true;
3933 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3936 __setup("ftrace_filter=", set_ftrace_filter);
3938 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3939 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3940 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3941 static int ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer);
3943 static int __init set_graph_function(char *str)
3945 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3948 __setup("ftrace_graph_filter=", set_graph_function);
3950 static int __init set_graph_notrace_function(char *str)
3952 strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
3955 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
3957 static void __init set_ftrace_early_graph(char *buf, int enable)
3961 unsigned long *table = ftrace_graph_funcs;
3962 int *count = &ftrace_graph_count;
3965 table = ftrace_graph_notrace_funcs;
3966 count = &ftrace_graph_notrace_count;
3970 func = strsep(&buf, ",");
3971 /* we allow only one expression at a time */
3972 ret = ftrace_set_func(table, count, FTRACE_GRAPH_MAX_FUNCS, func);
3974 printk(KERN_DEBUG "ftrace: function %s not "
3975 "traceable\n", func);
3978 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3981 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3985 ftrace_ops_init(ops);
3988 func = strsep(&buf, ",");
3989 ftrace_set_regex(ops, func, strlen(func), 0, enable);
3993 static void __init set_ftrace_early_filters(void)
3995 if (ftrace_filter_buf[0])
3996 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
3997 if (ftrace_notrace_buf[0])
3998 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
3999 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4000 if (ftrace_graph_buf[0])
4001 set_ftrace_early_graph(ftrace_graph_buf, 1);
4002 if (ftrace_graph_notrace_buf[0])
4003 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
4004 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4007 int ftrace_regex_release(struct inode *inode, struct file *file)
4009 struct seq_file *m = (struct seq_file *)file->private_data;
4010 struct ftrace_iterator *iter;
4011 struct ftrace_hash **orig_hash;
4012 struct trace_parser *parser;
4016 if (file->f_mode & FMODE_READ) {
4018 seq_release(inode, file);
4020 iter = file->private_data;
4022 parser = &iter->parser;
4023 if (trace_parser_loaded(parser)) {
4024 parser->buffer[parser->idx] = 0;
4025 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
4028 trace_parser_put(parser);
4030 mutex_lock(&iter->ops->func_hash->regex_lock);
4032 if (file->f_mode & FMODE_WRITE) {
4033 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
4036 orig_hash = &iter->ops->func_hash->filter_hash;
4038 orig_hash = &iter->ops->func_hash->notrace_hash;
4040 mutex_lock(&ftrace_lock);
4041 ret = ftrace_hash_move(iter->ops, filter_hash,
4042 orig_hash, iter->hash);
4044 ftrace_ops_update_code(iter->ops);
4046 mutex_unlock(&ftrace_lock);
4049 mutex_unlock(&iter->ops->func_hash->regex_lock);
4050 free_ftrace_hash(iter->hash);
4056 static const struct file_operations ftrace_avail_fops = {
4057 .open = ftrace_avail_open,
4059 .llseek = seq_lseek,
4060 .release = seq_release_private,
4063 static const struct file_operations ftrace_enabled_fops = {
4064 .open = ftrace_enabled_open,
4066 .llseek = seq_lseek,
4067 .release = seq_release_private,
4070 static const struct file_operations ftrace_filter_fops = {
4071 .open = ftrace_filter_open,
4073 .write = ftrace_filter_write,
4074 .llseek = tracing_lseek,
4075 .release = ftrace_regex_release,
4078 static const struct file_operations ftrace_notrace_fops = {
4079 .open = ftrace_notrace_open,
4081 .write = ftrace_notrace_write,
4082 .llseek = tracing_lseek,
4083 .release = ftrace_regex_release,
4086 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4088 static DEFINE_MUTEX(graph_lock);
4090 int ftrace_graph_count;
4091 int ftrace_graph_notrace_count;
4092 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4093 unsigned long ftrace_graph_notrace_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4095 struct ftrace_graph_data {
4096 unsigned long *table;
4099 const struct seq_operations *seq_ops;
4103 __g_next(struct seq_file *m, loff_t *pos)
4105 struct ftrace_graph_data *fgd = m->private;
4107 if (*pos >= *fgd->count)
4109 return &fgd->table[*pos];
4113 g_next(struct seq_file *m, void *v, loff_t *pos)
4116 return __g_next(m, pos);
4119 static void *g_start(struct seq_file *m, loff_t *pos)
4121 struct ftrace_graph_data *fgd = m->private;
4123 mutex_lock(&graph_lock);
4125 /* Nothing, tell g_show to print all functions are enabled */
4126 if (!*fgd->count && !*pos)
4129 return __g_next(m, pos);
4132 static void g_stop(struct seq_file *m, void *p)
4134 mutex_unlock(&graph_lock);
4137 static int g_show(struct seq_file *m, void *v)
4139 unsigned long *ptr = v;
4144 if (ptr == (unsigned long *)1) {
4145 struct ftrace_graph_data *fgd = m->private;
4147 if (fgd->table == ftrace_graph_funcs)
4148 seq_printf(m, "#### all functions enabled ####\n");
4150 seq_printf(m, "#### no functions disabled ####\n");
4154 seq_printf(m, "%ps\n", (void *)*ptr);
4159 static const struct seq_operations ftrace_graph_seq_ops = {
4167 __ftrace_graph_open(struct inode *inode, struct file *file,
4168 struct ftrace_graph_data *fgd)
4172 mutex_lock(&graph_lock);
4173 if ((file->f_mode & FMODE_WRITE) &&
4174 (file->f_flags & O_TRUNC)) {
4176 memset(fgd->table, 0, fgd->size * sizeof(*fgd->table));
4178 mutex_unlock(&graph_lock);
4180 if (file->f_mode & FMODE_READ) {
4181 ret = seq_open(file, fgd->seq_ops);
4183 struct seq_file *m = file->private_data;
4187 file->private_data = fgd;
4193 ftrace_graph_open(struct inode *inode, struct file *file)
4195 struct ftrace_graph_data *fgd;
4197 if (unlikely(ftrace_disabled))
4200 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4204 fgd->table = ftrace_graph_funcs;
4205 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4206 fgd->count = &ftrace_graph_count;
4207 fgd->seq_ops = &ftrace_graph_seq_ops;
4209 return __ftrace_graph_open(inode, file, fgd);
4213 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
4215 struct ftrace_graph_data *fgd;
4217 if (unlikely(ftrace_disabled))
4220 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4224 fgd->table = ftrace_graph_notrace_funcs;
4225 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4226 fgd->count = &ftrace_graph_notrace_count;
4227 fgd->seq_ops = &ftrace_graph_seq_ops;
4229 return __ftrace_graph_open(inode, file, fgd);
4233 ftrace_graph_release(struct inode *inode, struct file *file)
4235 if (file->f_mode & FMODE_READ) {
4236 struct seq_file *m = file->private_data;
4239 seq_release(inode, file);
4241 kfree(file->private_data);
4248 ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer)
4250 struct dyn_ftrace *rec;
4251 struct ftrace_page *pg;
4260 type = filter_parse_regex(buffer, strlen(buffer), &search, ¬);
4261 if (!not && *idx >= size)
4264 search_len = strlen(search);
4266 mutex_lock(&ftrace_lock);
4268 if (unlikely(ftrace_disabled)) {
4269 mutex_unlock(&ftrace_lock);
4273 do_for_each_ftrace_rec(pg, rec) {
4275 if (ftrace_match_record(rec, NULL, search, search_len, type)) {
4276 /* if it is in the array */
4278 for (i = 0; i < *idx; i++) {
4279 if (array[i] == rec->ip) {
4288 array[(*idx)++] = rec->ip;
4294 array[i] = array[--(*idx)];
4300 } while_for_each_ftrace_rec();
4302 mutex_unlock(&ftrace_lock);
4311 ftrace_graph_write(struct file *file, const char __user *ubuf,
4312 size_t cnt, loff_t *ppos)
4314 struct trace_parser parser;
4315 ssize_t read, ret = 0;
4316 struct ftrace_graph_data *fgd = file->private_data;
4321 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX))
4324 read = trace_get_user(&parser, ubuf, cnt, ppos);
4326 if (read >= 0 && trace_parser_loaded((&parser))) {
4327 parser.buffer[parser.idx] = 0;
4329 mutex_lock(&graph_lock);
4331 /* we allow only one expression at a time */
4332 ret = ftrace_set_func(fgd->table, fgd->count, fgd->size,
4335 mutex_unlock(&graph_lock);
4341 trace_parser_put(&parser);
4346 static const struct file_operations ftrace_graph_fops = {
4347 .open = ftrace_graph_open,
4349 .write = ftrace_graph_write,
4350 .llseek = tracing_lseek,
4351 .release = ftrace_graph_release,
4354 static const struct file_operations ftrace_graph_notrace_fops = {
4355 .open = ftrace_graph_notrace_open,
4357 .write = ftrace_graph_write,
4358 .llseek = tracing_lseek,
4359 .release = ftrace_graph_release,
4361 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4363 void ftrace_create_filter_files(struct ftrace_ops *ops,
4364 struct dentry *parent)
4367 trace_create_file("set_ftrace_filter", 0644, parent,
4368 ops, &ftrace_filter_fops);
4370 trace_create_file("set_ftrace_notrace", 0644, parent,
4371 ops, &ftrace_notrace_fops);
4375 * The name "destroy_filter_files" is really a misnomer. Although
4376 * in the future, it may actualy delete the files, but this is
4377 * really intended to make sure the ops passed in are disabled
4378 * and that when this function returns, the caller is free to
4381 * The "destroy" name is only to match the "create" name that this
4382 * should be paired with.
4384 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
4386 mutex_lock(&ftrace_lock);
4387 if (ops->flags & FTRACE_OPS_FL_ENABLED)
4388 ftrace_shutdown(ops, 0);
4389 ops->flags |= FTRACE_OPS_FL_DELETED;
4390 mutex_unlock(&ftrace_lock);
4393 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
4396 trace_create_file("available_filter_functions", 0444,
4397 d_tracer, NULL, &ftrace_avail_fops);
4399 trace_create_file("enabled_functions", 0444,
4400 d_tracer, NULL, &ftrace_enabled_fops);
4402 ftrace_create_filter_files(&global_ops, d_tracer);
4404 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4405 trace_create_file("set_graph_function", 0444, d_tracer,
4407 &ftrace_graph_fops);
4408 trace_create_file("set_graph_notrace", 0444, d_tracer,
4410 &ftrace_graph_notrace_fops);
4411 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4416 static int ftrace_cmp_ips(const void *a, const void *b)
4418 const unsigned long *ipa = a;
4419 const unsigned long *ipb = b;
4428 static void ftrace_swap_ips(void *a, void *b, int size)
4430 unsigned long *ipa = a;
4431 unsigned long *ipb = b;
4439 static int ftrace_process_locs(struct module *mod,
4440 unsigned long *start,
4443 struct ftrace_page *start_pg;
4444 struct ftrace_page *pg;
4445 struct dyn_ftrace *rec;
4446 unsigned long count;
4449 unsigned long flags = 0; /* Shut up gcc */
4452 count = end - start;
4457 sort(start, count, sizeof(*start),
4458 ftrace_cmp_ips, ftrace_swap_ips);
4460 start_pg = ftrace_allocate_pages(count);
4464 mutex_lock(&ftrace_lock);
4467 * Core and each module needs their own pages, as
4468 * modules will free them when they are removed.
4469 * Force a new page to be allocated for modules.
4472 WARN_ON(ftrace_pages || ftrace_pages_start);
4473 /* First initialization */
4474 ftrace_pages = ftrace_pages_start = start_pg;
4479 if (WARN_ON(ftrace_pages->next)) {
4480 /* Hmm, we have free pages? */
4481 while (ftrace_pages->next)
4482 ftrace_pages = ftrace_pages->next;
4485 ftrace_pages->next = start_pg;
4491 addr = ftrace_call_adjust(*p++);
4493 * Some architecture linkers will pad between
4494 * the different mcount_loc sections of different
4495 * object files to satisfy alignments.
4496 * Skip any NULL pointers.
4501 if (pg->index == pg->size) {
4502 /* We should have allocated enough */
4503 if (WARN_ON(!pg->next))
4508 rec = &pg->records[pg->index++];
4512 /* We should have used all pages */
4515 /* Assign the last page to ftrace_pages */
4519 * We only need to disable interrupts on start up
4520 * because we are modifying code that an interrupt
4521 * may execute, and the modification is not atomic.
4522 * But for modules, nothing runs the code we modify
4523 * until we are finished with it, and there's no
4524 * reason to cause large interrupt latencies while we do it.
4527 local_irq_save(flags);
4528 ftrace_update_code(mod, start_pg);
4530 local_irq_restore(flags);
4533 mutex_unlock(&ftrace_lock);
4538 #ifdef CONFIG_MODULES
4540 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4542 void ftrace_release_mod(struct module *mod)
4544 struct dyn_ftrace *rec;
4545 struct ftrace_page **last_pg;
4546 struct ftrace_page *pg;
4549 mutex_lock(&ftrace_lock);
4551 if (ftrace_disabled)
4555 * Each module has its own ftrace_pages, remove
4556 * them from the list.
4558 last_pg = &ftrace_pages_start;
4559 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4560 rec = &pg->records[0];
4561 if (within_module_core(rec->ip, mod)) {
4563 * As core pages are first, the first
4564 * page should never be a module page.
4566 if (WARN_ON(pg == ftrace_pages_start))
4569 /* Check if we are deleting the last page */
4570 if (pg == ftrace_pages)
4571 ftrace_pages = next_to_ftrace_page(last_pg);
4573 *last_pg = pg->next;
4574 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4575 free_pages((unsigned long)pg->records, order);
4578 last_pg = &pg->next;
4581 mutex_unlock(&ftrace_lock);
4584 static void ftrace_init_module(struct module *mod,
4585 unsigned long *start, unsigned long *end)
4587 if (ftrace_disabled || start == end)
4589 ftrace_process_locs(mod, start, end);
4592 void ftrace_module_init(struct module *mod)
4594 ftrace_init_module(mod, mod->ftrace_callsites,
4595 mod->ftrace_callsites +
4596 mod->num_ftrace_callsites);
4599 static int ftrace_module_notify_exit(struct notifier_block *self,
4600 unsigned long val, void *data)
4602 struct module *mod = data;
4604 if (val == MODULE_STATE_GOING)
4605 ftrace_release_mod(mod);
4610 static int ftrace_module_notify_exit(struct notifier_block *self,
4611 unsigned long val, void *data)
4615 #endif /* CONFIG_MODULES */
4617 struct notifier_block ftrace_module_exit_nb = {
4618 .notifier_call = ftrace_module_notify_exit,
4619 .priority = INT_MIN, /* Run after anything that can remove kprobes */
4622 void __init ftrace_init(void)
4624 extern unsigned long __start_mcount_loc[];
4625 extern unsigned long __stop_mcount_loc[];
4626 unsigned long count, flags;
4629 local_irq_save(flags);
4630 ret = ftrace_dyn_arch_init();
4631 local_irq_restore(flags);
4635 count = __stop_mcount_loc - __start_mcount_loc;
4637 pr_info("ftrace: No functions to be traced?\n");
4641 pr_info("ftrace: allocating %ld entries in %ld pages\n",
4642 count, count / ENTRIES_PER_PAGE + 1);
4644 last_ftrace_enabled = ftrace_enabled = 1;
4646 ret = ftrace_process_locs(NULL,
4650 ret = register_module_notifier(&ftrace_module_exit_nb);
4652 pr_warning("Failed to register trace ftrace module exit notifier\n");
4654 set_ftrace_early_filters();
4658 ftrace_disabled = 1;
4663 static struct ftrace_ops global_ops = {
4664 .func = ftrace_stub,
4665 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4666 INIT_OPS_HASH(global_ops)
4669 static int __init ftrace_nodyn_init(void)
4674 core_initcall(ftrace_nodyn_init);
4676 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
4677 static inline void ftrace_startup_enable(int command) { }
4678 /* Keep as macros so we do not need to define the commands */
4679 # define ftrace_startup(ops, command) \
4681 int ___ret = __register_ftrace_function(ops); \
4683 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
4686 # define ftrace_shutdown(ops, command) \
4688 int ___ret = __unregister_ftrace_function(ops); \
4690 (ops)->flags &= ~FTRACE_OPS_FL_ENABLED; \
4694 # define ftrace_startup_sysctl() do { } while (0)
4695 # define ftrace_shutdown_sysctl() do { } while (0)
4698 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
4703 #endif /* CONFIG_DYNAMIC_FTRACE */
4705 __init void ftrace_init_global_array_ops(struct trace_array *tr)
4707 tr->ops = &global_ops;
4708 tr->ops->private = tr;
4711 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
4713 /* If we filter on pids, update to use the pid function */
4714 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
4715 if (WARN_ON(tr->ops->func != ftrace_stub))
4716 printk("ftrace ops had %pS for function\n",
4718 /* Only the top level instance does pid tracing */
4719 if (!list_empty(&ftrace_pids)) {
4720 set_ftrace_pid_function(func);
4721 func = ftrace_pid_func;
4724 tr->ops->func = func;
4725 tr->ops->private = tr;
4728 void ftrace_reset_array_ops(struct trace_array *tr)
4730 tr->ops->func = ftrace_stub;
4734 ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip,
4735 struct ftrace_ops *op, struct pt_regs *regs)
4737 if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
4741 * Some of the ops may be dynamically allocated,
4742 * they must be freed after a synchronize_sched().
4744 preempt_disable_notrace();
4745 trace_recursion_set(TRACE_CONTROL_BIT);
4748 * Control funcs (perf) uses RCU. Only trace if
4749 * RCU is currently active.
4751 if (!rcu_is_watching())
4754 do_for_each_ftrace_op(op, ftrace_control_list) {
4755 if (!(op->flags & FTRACE_OPS_FL_STUB) &&
4756 !ftrace_function_local_disabled(op) &&
4757 ftrace_ops_test(op, ip, regs))
4758 op->func(ip, parent_ip, op, regs);
4759 } while_for_each_ftrace_op(op);
4761 trace_recursion_clear(TRACE_CONTROL_BIT);
4762 preempt_enable_notrace();
4765 static struct ftrace_ops control_ops = {
4766 .func = ftrace_ops_control_func,
4767 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4768 INIT_OPS_HASH(control_ops)
4772 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4773 struct ftrace_ops *ignored, struct pt_regs *regs)
4775 struct ftrace_ops *op;
4778 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
4783 * Some of the ops may be dynamically allocated,
4784 * they must be freed after a synchronize_sched().
4786 preempt_disable_notrace();
4787 do_for_each_ftrace_op(op, ftrace_ops_list) {
4788 if (ftrace_ops_test(op, ip, regs)) {
4789 if (FTRACE_WARN_ON(!op->func)) {
4790 pr_warn("op=%p %pS\n", op, op);
4793 op->func(ip, parent_ip, op, regs);
4795 } while_for_each_ftrace_op(op);
4797 preempt_enable_notrace();
4798 trace_clear_recursion(bit);
4802 * Some archs only support passing ip and parent_ip. Even though
4803 * the list function ignores the op parameter, we do not want any
4804 * C side effects, where a function is called without the caller
4805 * sending a third parameter.
4806 * Archs are to support both the regs and ftrace_ops at the same time.
4807 * If they support ftrace_ops, it is assumed they support regs.
4808 * If call backs want to use regs, they must either check for regs
4809 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
4810 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
4811 * An architecture can pass partial regs with ftrace_ops and still
4812 * set the ARCH_SUPPORT_FTARCE_OPS.
4814 #if ARCH_SUPPORTS_FTRACE_OPS
4815 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4816 struct ftrace_ops *op, struct pt_regs *regs)
4818 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
4821 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
4823 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
4827 static void clear_ftrace_swapper(void)
4829 struct task_struct *p;
4833 for_each_online_cpu(cpu) {
4835 clear_tsk_trace_trace(p);
4840 static void set_ftrace_swapper(void)
4842 struct task_struct *p;
4846 for_each_online_cpu(cpu) {
4848 set_tsk_trace_trace(p);
4853 static void clear_ftrace_pid(struct pid *pid)
4855 struct task_struct *p;
4858 do_each_pid_task(pid, PIDTYPE_PID, p) {
4859 clear_tsk_trace_trace(p);
4860 } while_each_pid_task(pid, PIDTYPE_PID, p);
4866 static void set_ftrace_pid(struct pid *pid)
4868 struct task_struct *p;
4871 do_each_pid_task(pid, PIDTYPE_PID, p) {
4872 set_tsk_trace_trace(p);
4873 } while_each_pid_task(pid, PIDTYPE_PID, p);
4877 static void clear_ftrace_pid_task(struct pid *pid)
4879 if (pid == ftrace_swapper_pid)
4880 clear_ftrace_swapper();
4882 clear_ftrace_pid(pid);
4885 static void set_ftrace_pid_task(struct pid *pid)
4887 if (pid == ftrace_swapper_pid)
4888 set_ftrace_swapper();
4890 set_ftrace_pid(pid);
4893 static int ftrace_pid_add(int p)
4896 struct ftrace_pid *fpid;
4899 mutex_lock(&ftrace_lock);
4902 pid = ftrace_swapper_pid;
4904 pid = find_get_pid(p);
4911 list_for_each_entry(fpid, &ftrace_pids, list)
4912 if (fpid->pid == pid)
4917 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
4921 list_add(&fpid->list, &ftrace_pids);
4924 set_ftrace_pid_task(pid);
4926 ftrace_update_pid_func();
4927 ftrace_startup_enable(0);
4929 mutex_unlock(&ftrace_lock);
4933 if (pid != ftrace_swapper_pid)
4937 mutex_unlock(&ftrace_lock);
4941 static void ftrace_pid_reset(void)
4943 struct ftrace_pid *fpid, *safe;
4945 mutex_lock(&ftrace_lock);
4946 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
4947 struct pid *pid = fpid->pid;
4949 clear_ftrace_pid_task(pid);
4951 list_del(&fpid->list);
4955 ftrace_update_pid_func();
4956 ftrace_startup_enable(0);
4958 mutex_unlock(&ftrace_lock);
4961 static void *fpid_start(struct seq_file *m, loff_t *pos)
4963 mutex_lock(&ftrace_lock);
4965 if (list_empty(&ftrace_pids) && (!*pos))
4968 return seq_list_start(&ftrace_pids, *pos);
4971 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
4976 return seq_list_next(v, &ftrace_pids, pos);
4979 static void fpid_stop(struct seq_file *m, void *p)
4981 mutex_unlock(&ftrace_lock);
4984 static int fpid_show(struct seq_file *m, void *v)
4986 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
4988 if (v == (void *)1) {
4989 seq_printf(m, "no pid\n");
4993 if (fpid->pid == ftrace_swapper_pid)
4994 seq_printf(m, "swapper tasks\n");
4996 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
5001 static const struct seq_operations ftrace_pid_sops = {
5002 .start = fpid_start,
5009 ftrace_pid_open(struct inode *inode, struct file *file)
5013 if ((file->f_mode & FMODE_WRITE) &&
5014 (file->f_flags & O_TRUNC))
5017 if (file->f_mode & FMODE_READ)
5018 ret = seq_open(file, &ftrace_pid_sops);
5024 ftrace_pid_write(struct file *filp, const char __user *ubuf,
5025 size_t cnt, loff_t *ppos)
5031 if (cnt >= sizeof(buf))
5034 if (copy_from_user(&buf, ubuf, cnt))
5040 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
5041 * to clean the filter quietly.
5043 tmp = strstrip(buf);
5044 if (strlen(tmp) == 0)
5047 ret = kstrtol(tmp, 10, &val);
5051 ret = ftrace_pid_add(val);
5053 return ret ? ret : cnt;
5057 ftrace_pid_release(struct inode *inode, struct file *file)
5059 if (file->f_mode & FMODE_READ)
5060 seq_release(inode, file);
5065 static const struct file_operations ftrace_pid_fops = {
5066 .open = ftrace_pid_open,
5067 .write = ftrace_pid_write,
5069 .llseek = tracing_lseek,
5070 .release = ftrace_pid_release,
5073 static __init int ftrace_init_debugfs(void)
5075 struct dentry *d_tracer;
5077 d_tracer = tracing_init_dentry();
5081 ftrace_init_dyn_debugfs(d_tracer);
5083 trace_create_file("set_ftrace_pid", 0644, d_tracer,
5084 NULL, &ftrace_pid_fops);
5086 ftrace_profile_debugfs(d_tracer);
5090 fs_initcall(ftrace_init_debugfs);
5093 * ftrace_kill - kill ftrace
5095 * This function should be used by panic code. It stops ftrace
5096 * but in a not so nice way. If you need to simply kill ftrace
5097 * from a non-atomic section, use ftrace_kill.
5099 void ftrace_kill(void)
5101 ftrace_disabled = 1;
5103 clear_ftrace_function();
5107 * Test if ftrace is dead or not.
5109 int ftrace_is_dead(void)
5111 return ftrace_disabled;
5115 * register_ftrace_function - register a function for profiling
5116 * @ops - ops structure that holds the function for profiling.
5118 * Register a function to be called by all functions in the
5121 * Note: @ops->func and all the functions it calls must be labeled
5122 * with "notrace", otherwise it will go into a
5125 int register_ftrace_function(struct ftrace_ops *ops)
5129 ftrace_ops_init(ops);
5131 mutex_lock(&ftrace_lock);
5133 ret = ftrace_startup(ops, 0);
5135 mutex_unlock(&ftrace_lock);
5139 EXPORT_SYMBOL_GPL(register_ftrace_function);
5142 * unregister_ftrace_function - unregister a function for profiling.
5143 * @ops - ops structure that holds the function to unregister
5145 * Unregister a function that was added to be called by ftrace profiling.
5147 int unregister_ftrace_function(struct ftrace_ops *ops)
5151 mutex_lock(&ftrace_lock);
5152 ret = ftrace_shutdown(ops, 0);
5153 mutex_unlock(&ftrace_lock);
5157 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
5160 ftrace_enable_sysctl(struct ctl_table *table, int write,
5161 void __user *buffer, size_t *lenp,
5166 mutex_lock(&ftrace_lock);
5168 if (unlikely(ftrace_disabled))
5171 ret = proc_dointvec(table, write, buffer, lenp, ppos);
5173 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
5176 last_ftrace_enabled = !!ftrace_enabled;
5178 if (ftrace_enabled) {
5180 ftrace_startup_sysctl();
5182 /* we are starting ftrace again */
5183 if (ftrace_ops_list != &ftrace_list_end)
5184 update_ftrace_function();
5187 /* stopping ftrace calls (just send to ftrace_stub) */
5188 ftrace_trace_function = ftrace_stub;
5190 ftrace_shutdown_sysctl();
5194 mutex_unlock(&ftrace_lock);
5198 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5200 static int ftrace_graph_active;
5202 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
5207 /* The callbacks that hook a function */
5208 trace_func_graph_ret_t ftrace_graph_return =
5209 (trace_func_graph_ret_t)ftrace_stub;
5210 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
5211 static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
5213 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
5214 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
5218 unsigned long flags;
5219 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
5220 struct task_struct *g, *t;
5222 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
5223 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
5224 * sizeof(struct ftrace_ret_stack),
5226 if (!ret_stack_list[i]) {
5234 read_lock_irqsave(&tasklist_lock, flags);
5235 do_each_thread(g, t) {
5241 if (t->ret_stack == NULL) {
5242 atomic_set(&t->tracing_graph_pause, 0);
5243 atomic_set(&t->trace_overrun, 0);
5244 t->curr_ret_stack = -1;
5245 /* Make sure the tasks see the -1 first: */
5247 t->ret_stack = ret_stack_list[start++];
5249 } while_each_thread(g, t);
5252 read_unlock_irqrestore(&tasklist_lock, flags);
5254 for (i = start; i < end; i++)
5255 kfree(ret_stack_list[i]);
5260 ftrace_graph_probe_sched_switch(void *ignore,
5261 struct task_struct *prev, struct task_struct *next)
5263 unsigned long long timestamp;
5267 * Does the user want to count the time a function was asleep.
5268 * If so, do not update the time stamps.
5270 if (trace_flags & TRACE_ITER_SLEEP_TIME)
5273 timestamp = trace_clock_local();
5275 prev->ftrace_timestamp = timestamp;
5277 /* only process tasks that we timestamped */
5278 if (!next->ftrace_timestamp)
5282 * Update all the counters in next to make up for the
5283 * time next was sleeping.
5285 timestamp -= next->ftrace_timestamp;
5287 for (index = next->curr_ret_stack; index >= 0; index--)
5288 next->ret_stack[index].calltime += timestamp;
5291 /* Allocate a return stack for each task */
5292 static int start_graph_tracing(void)
5294 struct ftrace_ret_stack **ret_stack_list;
5297 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
5298 sizeof(struct ftrace_ret_stack *),
5301 if (!ret_stack_list)
5304 /* The cpu_boot init_task->ret_stack will never be freed */
5305 for_each_online_cpu(cpu) {
5306 if (!idle_task(cpu)->ret_stack)
5307 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
5311 ret = alloc_retstack_tasklist(ret_stack_list);
5312 } while (ret == -EAGAIN);
5315 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5317 pr_info("ftrace_graph: Couldn't activate tracepoint"
5318 " probe to kernel_sched_switch\n");
5321 kfree(ret_stack_list);
5326 * Hibernation protection.
5327 * The state of the current task is too much unstable during
5328 * suspend/restore to disk. We want to protect against that.
5331 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
5335 case PM_HIBERNATION_PREPARE:
5336 pause_graph_tracing();
5339 case PM_POST_HIBERNATION:
5340 unpause_graph_tracing();
5346 static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
5348 if (!ftrace_ops_test(&global_ops, trace->func, NULL))
5350 return __ftrace_graph_entry(trace);
5354 * The function graph tracer should only trace the functions defined
5355 * by set_ftrace_filter and set_ftrace_notrace. If another function
5356 * tracer ops is registered, the graph tracer requires testing the
5357 * function against the global ops, and not just trace any function
5358 * that any ftrace_ops registered.
5360 static void update_function_graph_func(void)
5362 if (ftrace_ops_list == &ftrace_list_end ||
5363 (ftrace_ops_list == &global_ops &&
5364 global_ops.next == &ftrace_list_end))
5365 ftrace_graph_entry = __ftrace_graph_entry;
5367 ftrace_graph_entry = ftrace_graph_entry_test;
5370 static struct notifier_block ftrace_suspend_notifier = {
5371 .notifier_call = ftrace_suspend_notifier_call,
5374 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
5375 trace_func_graph_ent_t entryfunc)
5379 mutex_lock(&ftrace_lock);
5381 /* we currently allow only one tracer registered at a time */
5382 if (ftrace_graph_active) {
5387 register_pm_notifier(&ftrace_suspend_notifier);
5389 ftrace_graph_active++;
5390 ret = start_graph_tracing();
5392 ftrace_graph_active--;
5396 ftrace_graph_return = retfunc;
5399 * Update the indirect function to the entryfunc, and the
5400 * function that gets called to the entry_test first. Then
5401 * call the update fgraph entry function to determine if
5402 * the entryfunc should be called directly or not.
5404 __ftrace_graph_entry = entryfunc;
5405 ftrace_graph_entry = ftrace_graph_entry_test;
5406 update_function_graph_func();
5408 /* Function graph doesn't use the .func field of global_ops */
5409 global_ops.flags |= FTRACE_OPS_FL_STUB;
5411 #ifdef CONFIG_DYNAMIC_FTRACE
5412 /* Optimize function graph calling (if implemented by arch) */
5413 if (FTRACE_GRAPH_TRAMP_ADDR != 0)
5414 global_ops.trampoline = FTRACE_GRAPH_TRAMP_ADDR;
5417 ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
5420 mutex_unlock(&ftrace_lock);
5424 void unregister_ftrace_graph(void)
5426 mutex_lock(&ftrace_lock);
5428 if (unlikely(!ftrace_graph_active))
5431 ftrace_graph_active--;
5432 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
5433 ftrace_graph_entry = ftrace_graph_entry_stub;
5434 __ftrace_graph_entry = ftrace_graph_entry_stub;
5435 ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
5436 global_ops.flags &= ~FTRACE_OPS_FL_STUB;
5437 #ifdef CONFIG_DYNAMIC_FTRACE
5438 if (FTRACE_GRAPH_TRAMP_ADDR != 0)
5439 global_ops.trampoline = 0;
5441 unregister_pm_notifier(&ftrace_suspend_notifier);
5442 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5445 mutex_unlock(&ftrace_lock);
5448 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
5451 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
5453 atomic_set(&t->tracing_graph_pause, 0);
5454 atomic_set(&t->trace_overrun, 0);
5455 t->ftrace_timestamp = 0;
5456 /* make curr_ret_stack visible before we add the ret_stack */
5458 t->ret_stack = ret_stack;
5462 * Allocate a return stack for the idle task. May be the first
5463 * time through, or it may be done by CPU hotplug online.
5465 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
5467 t->curr_ret_stack = -1;
5469 * The idle task has no parent, it either has its own
5470 * stack or no stack at all.
5473 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
5475 if (ftrace_graph_active) {
5476 struct ftrace_ret_stack *ret_stack;
5478 ret_stack = per_cpu(idle_ret_stack, cpu);
5480 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5481 * sizeof(struct ftrace_ret_stack),
5485 per_cpu(idle_ret_stack, cpu) = ret_stack;
5487 graph_init_task(t, ret_stack);
5491 /* Allocate a return stack for newly created task */
5492 void ftrace_graph_init_task(struct task_struct *t)
5494 /* Make sure we do not use the parent ret_stack */
5495 t->ret_stack = NULL;
5496 t->curr_ret_stack = -1;
5498 if (ftrace_graph_active) {
5499 struct ftrace_ret_stack *ret_stack;
5501 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5502 * sizeof(struct ftrace_ret_stack),
5506 graph_init_task(t, ret_stack);
5510 void ftrace_graph_exit_task(struct task_struct *t)
5512 struct ftrace_ret_stack *ret_stack = t->ret_stack;
5514 t->ret_stack = NULL;
5515 /* NULL must become visible to IRQs before we free it: */