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/tracefs.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 #ifdef CONFIG_DYNAMIC_FTRACE
66 #define INIT_OPS_HASH(opsname) \
67 .func_hash = &opsname.local_hash, \
68 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
69 #define ASSIGN_OPS_HASH(opsname, val) \
71 .local_hash.regex_lock = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
73 #define INIT_OPS_HASH(opsname)
74 #define ASSIGN_OPS_HASH(opsname, val)
77 static struct ftrace_ops ftrace_list_end __read_mostly = {
79 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
80 INIT_OPS_HASH(ftrace_list_end)
83 /* ftrace_enabled is a method to turn ftrace on or off */
84 int ftrace_enabled __read_mostly;
85 static int last_ftrace_enabled;
87 /* Current function tracing op */
88 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
89 /* What to set function_trace_op to */
90 static struct ftrace_ops *set_function_trace_op;
92 /* List for set_ftrace_pid's pids. */
93 LIST_HEAD(ftrace_pids);
95 struct list_head list;
99 static bool ftrace_pids_enabled(void)
101 return !list_empty(&ftrace_pids);
104 static void ftrace_update_trampoline(struct ftrace_ops *ops);
107 * ftrace_disabled is set when an anomaly is discovered.
108 * ftrace_disabled is much stronger than ftrace_enabled.
110 static int ftrace_disabled __read_mostly;
112 static DEFINE_MUTEX(ftrace_lock);
114 static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
115 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
116 static struct ftrace_ops global_ops;
118 #if ARCH_SUPPORTS_FTRACE_OPS
119 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
120 struct ftrace_ops *op, struct pt_regs *regs);
122 /* See comment below, where ftrace_ops_list_func is defined */
123 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
124 #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
128 * Traverse the ftrace_global_list, invoking all entries. The reason that we
129 * can use rcu_dereference_raw_notrace() is that elements removed from this list
130 * are simply leaked, so there is no need to interact with a grace-period
131 * mechanism. The rcu_dereference_raw_notrace() calls are needed to handle
132 * concurrent insertions into the ftrace_global_list.
134 * Silly Alpha and silly pointer-speculation compiler optimizations!
136 #define do_for_each_ftrace_op(op, list) \
137 op = rcu_dereference_raw_notrace(list); \
141 * Optimized for just a single item in the list (as that is the normal case).
143 #define while_for_each_ftrace_op(op) \
144 while (likely(op = rcu_dereference_raw_notrace((op)->next)) && \
145 unlikely((op) != &ftrace_list_end))
147 static inline void ftrace_ops_init(struct ftrace_ops *ops)
149 #ifdef CONFIG_DYNAMIC_FTRACE
150 if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
151 mutex_init(&ops->local_hash.regex_lock);
152 ops->func_hash = &ops->local_hash;
153 ops->flags |= FTRACE_OPS_FL_INITIALIZED;
159 * ftrace_nr_registered_ops - return number of ops registered
161 * Returns the number of ftrace_ops registered and tracing functions
163 int ftrace_nr_registered_ops(void)
165 struct ftrace_ops *ops;
168 mutex_lock(&ftrace_lock);
170 for (ops = ftrace_ops_list;
171 ops != &ftrace_list_end; ops = ops->next)
174 mutex_unlock(&ftrace_lock);
179 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
180 struct ftrace_ops *op, struct pt_regs *regs)
182 if (!test_tsk_trace_trace(current))
185 op->saved_func(ip, parent_ip, op, regs);
189 * clear_ftrace_function - reset the ftrace function
191 * This NULLs the ftrace function and in essence stops
192 * tracing. There may be lag
194 void clear_ftrace_function(void)
196 ftrace_trace_function = ftrace_stub;
199 static void per_cpu_ops_disable_all(struct ftrace_ops *ops)
203 for_each_possible_cpu(cpu)
204 *per_cpu_ptr(ops->disabled, cpu) = 1;
207 static int per_cpu_ops_alloc(struct ftrace_ops *ops)
209 int __percpu *disabled;
211 if (WARN_ON_ONCE(!(ops->flags & FTRACE_OPS_FL_PER_CPU)))
214 disabled = alloc_percpu(int);
218 ops->disabled = disabled;
219 per_cpu_ops_disable_all(ops);
223 static void ftrace_sync(struct work_struct *work)
226 * This function is just a stub to implement a hard force
227 * of synchronize_sched(). This requires synchronizing
228 * tasks even in userspace and idle.
230 * Yes, function tracing is rude.
234 static void ftrace_sync_ipi(void *data)
236 /* Probably not needed, but do it anyway */
240 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
241 static void update_function_graph_func(void);
243 /* Both enabled by default (can be cleared by function_graph tracer flags */
244 static bool fgraph_sleep_time = true;
245 static bool fgraph_graph_time = true;
248 static inline void update_function_graph_func(void) { }
252 static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
255 * If this is a dynamic, RCU, or per CPU ops, or we force list func,
256 * then it needs to call the list anyway.
258 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU |
259 FTRACE_OPS_FL_RCU) || FTRACE_FORCE_LIST_FUNC)
260 return ftrace_ops_list_func;
262 return ftrace_ops_get_func(ops);
265 static void update_ftrace_function(void)
270 * Prepare the ftrace_ops that the arch callback will use.
271 * If there's only one ftrace_ops registered, the ftrace_ops_list
272 * will point to the ops we want.
274 set_function_trace_op = ftrace_ops_list;
276 /* If there's no ftrace_ops registered, just call the stub function */
277 if (ftrace_ops_list == &ftrace_list_end) {
281 * If we are at the end of the list and this ops is
282 * recursion safe and not dynamic and the arch supports passing ops,
283 * then have the mcount trampoline call the function directly.
285 } else if (ftrace_ops_list->next == &ftrace_list_end) {
286 func = ftrace_ops_get_list_func(ftrace_ops_list);
289 /* Just use the default ftrace_ops */
290 set_function_trace_op = &ftrace_list_end;
291 func = ftrace_ops_list_func;
294 update_function_graph_func();
296 /* If there's no change, then do nothing more here */
297 if (ftrace_trace_function == func)
301 * If we are using the list function, it doesn't care
302 * about the function_trace_ops.
304 if (func == ftrace_ops_list_func) {
305 ftrace_trace_function = func;
307 * Don't even bother setting function_trace_ops,
308 * it would be racy to do so anyway.
313 #ifndef CONFIG_DYNAMIC_FTRACE
315 * For static tracing, we need to be a bit more careful.
316 * The function change takes affect immediately. Thus,
317 * we need to coorditate the setting of the function_trace_ops
318 * with the setting of the ftrace_trace_function.
320 * Set the function to the list ops, which will call the
321 * function we want, albeit indirectly, but it handles the
322 * ftrace_ops and doesn't depend on function_trace_op.
324 ftrace_trace_function = ftrace_ops_list_func;
326 * Make sure all CPUs see this. Yes this is slow, but static
327 * tracing is slow and nasty to have enabled.
329 schedule_on_each_cpu(ftrace_sync);
330 /* Now all cpus are using the list ops. */
331 function_trace_op = set_function_trace_op;
332 /* Make sure the function_trace_op is visible on all CPUs */
334 /* Nasty way to force a rmb on all cpus */
335 smp_call_function(ftrace_sync_ipi, NULL, 1);
336 /* OK, we are all set to update the ftrace_trace_function now! */
337 #endif /* !CONFIG_DYNAMIC_FTRACE */
339 ftrace_trace_function = func;
342 int using_ftrace_ops_list_func(void)
344 return ftrace_trace_function == ftrace_ops_list_func;
347 static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
351 * We are entering ops into the list but another
352 * CPU might be walking that list. We need to make sure
353 * the ops->next pointer is valid before another CPU sees
354 * the ops pointer included into the list.
356 rcu_assign_pointer(*list, ops);
359 static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
361 struct ftrace_ops **p;
364 * If we are removing the last function, then simply point
365 * to the ftrace_stub.
367 if (*list == ops && ops->next == &ftrace_list_end) {
368 *list = &ftrace_list_end;
372 for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
383 static void ftrace_update_trampoline(struct ftrace_ops *ops);
385 static int __register_ftrace_function(struct ftrace_ops *ops)
387 if (ops->flags & FTRACE_OPS_FL_DELETED)
390 if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
393 #ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
395 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
396 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
397 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
399 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
400 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
403 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
404 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
407 if (!core_kernel_data((unsigned long)ops))
408 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
410 if (ops->flags & FTRACE_OPS_FL_PER_CPU) {
411 if (per_cpu_ops_alloc(ops))
415 add_ftrace_ops(&ftrace_ops_list, ops);
417 /* Always save the function, and reset at unregistering */
418 ops->saved_func = ops->func;
420 if (ops->flags & FTRACE_OPS_FL_PID && ftrace_pids_enabled())
421 ops->func = ftrace_pid_func;
423 ftrace_update_trampoline(ops);
426 update_ftrace_function();
431 static int __unregister_ftrace_function(struct ftrace_ops *ops)
435 if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
438 ret = remove_ftrace_ops(&ftrace_ops_list, ops);
444 update_ftrace_function();
446 ops->func = ops->saved_func;
451 static void ftrace_update_pid_func(void)
453 bool enabled = ftrace_pids_enabled();
454 struct ftrace_ops *op;
456 /* Only do something if we are tracing something */
457 if (ftrace_trace_function == ftrace_stub)
460 do_for_each_ftrace_op(op, ftrace_ops_list) {
461 if (op->flags & FTRACE_OPS_FL_PID) {
462 op->func = enabled ? ftrace_pid_func :
464 ftrace_update_trampoline(op);
466 } while_for_each_ftrace_op(op);
468 update_ftrace_function();
471 #ifdef CONFIG_FUNCTION_PROFILER
472 struct ftrace_profile {
473 struct hlist_node node;
475 unsigned long counter;
476 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
477 unsigned long long time;
478 unsigned long long time_squared;
482 struct ftrace_profile_page {
483 struct ftrace_profile_page *next;
485 struct ftrace_profile records[];
488 struct ftrace_profile_stat {
490 struct hlist_head *hash;
491 struct ftrace_profile_page *pages;
492 struct ftrace_profile_page *start;
493 struct tracer_stat stat;
496 #define PROFILE_RECORDS_SIZE \
497 (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
499 #define PROFILES_PER_PAGE \
500 (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
502 static int ftrace_profile_enabled __read_mostly;
504 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
505 static DEFINE_MUTEX(ftrace_profile_lock);
507 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
509 #define FTRACE_PROFILE_HASH_BITS 10
510 #define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
513 function_stat_next(void *v, int idx)
515 struct ftrace_profile *rec = v;
516 struct ftrace_profile_page *pg;
518 pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
524 if ((void *)rec >= (void *)&pg->records[pg->index]) {
528 rec = &pg->records[0];
536 static void *function_stat_start(struct tracer_stat *trace)
538 struct ftrace_profile_stat *stat =
539 container_of(trace, struct ftrace_profile_stat, stat);
541 if (!stat || !stat->start)
544 return function_stat_next(&stat->start->records[0], 0);
547 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
548 /* function graph compares on total time */
549 static int function_stat_cmp(void *p1, void *p2)
551 struct ftrace_profile *a = p1;
552 struct ftrace_profile *b = p2;
554 if (a->time < b->time)
556 if (a->time > b->time)
562 /* not function graph compares against hits */
563 static int function_stat_cmp(void *p1, void *p2)
565 struct ftrace_profile *a = p1;
566 struct ftrace_profile *b = p2;
568 if (a->counter < b->counter)
570 if (a->counter > b->counter)
577 static int function_stat_headers(struct seq_file *m)
579 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
580 seq_puts(m, " Function "
583 "--- ---- --- ---\n");
585 seq_puts(m, " Function Hit\n"
591 static int function_stat_show(struct seq_file *m, void *v)
593 struct ftrace_profile *rec = v;
594 char str[KSYM_SYMBOL_LEN];
596 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
597 static struct trace_seq s;
598 unsigned long long avg;
599 unsigned long long stddev;
601 mutex_lock(&ftrace_profile_lock);
603 /* we raced with function_profile_reset() */
604 if (unlikely(rec->counter == 0)) {
609 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
611 do_div(avg, rec->counter);
612 if (tracing_thresh && (avg < tracing_thresh))
616 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
617 seq_printf(m, " %-30.30s %10lu", str, rec->counter);
619 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
622 /* Sample standard deviation (s^2) */
623 if (rec->counter <= 1)
627 * Apply Welford's method:
628 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
630 stddev = rec->counter * rec->time_squared -
631 rec->time * rec->time;
634 * Divide only 1000 for ns^2 -> us^2 conversion.
635 * trace_print_graph_duration will divide 1000 again.
637 do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
641 trace_print_graph_duration(rec->time, &s);
642 trace_seq_puts(&s, " ");
643 trace_print_graph_duration(avg, &s);
644 trace_seq_puts(&s, " ");
645 trace_print_graph_duration(stddev, &s);
646 trace_print_seq(m, &s);
650 mutex_unlock(&ftrace_profile_lock);
655 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
657 struct ftrace_profile_page *pg;
659 pg = stat->pages = stat->start;
662 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
667 memset(stat->hash, 0,
668 FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
671 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
673 struct ftrace_profile_page *pg;
678 /* If we already allocated, do nothing */
682 stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
686 #ifdef CONFIG_DYNAMIC_FTRACE
687 functions = ftrace_update_tot_cnt;
690 * We do not know the number of functions that exist because
691 * dynamic tracing is what counts them. With past experience
692 * we have around 20K functions. That should be more than enough.
693 * It is highly unlikely we will execute every function in
699 pg = stat->start = stat->pages;
701 pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
703 for (i = 1; i < pages; i++) {
704 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
715 unsigned long tmp = (unsigned long)pg;
727 static int ftrace_profile_init_cpu(int cpu)
729 struct ftrace_profile_stat *stat;
732 stat = &per_cpu(ftrace_profile_stats, cpu);
735 /* If the profile is already created, simply reset it */
736 ftrace_profile_reset(stat);
741 * We are profiling all functions, but usually only a few thousand
742 * functions are hit. We'll make a hash of 1024 items.
744 size = FTRACE_PROFILE_HASH_SIZE;
746 stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
751 /* Preallocate the function profiling pages */
752 if (ftrace_profile_pages_init(stat) < 0) {
761 static int ftrace_profile_init(void)
766 for_each_possible_cpu(cpu) {
767 ret = ftrace_profile_init_cpu(cpu);
775 /* interrupts must be disabled */
776 static struct ftrace_profile *
777 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
779 struct ftrace_profile *rec;
780 struct hlist_head *hhd;
783 key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
784 hhd = &stat->hash[key];
786 if (hlist_empty(hhd))
789 hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
797 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
798 struct ftrace_profile *rec)
802 key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
803 hlist_add_head_rcu(&rec->node, &stat->hash[key]);
807 * The memory is already allocated, this simply finds a new record to use.
809 static struct ftrace_profile *
810 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
812 struct ftrace_profile *rec = NULL;
814 /* prevent recursion (from NMIs) */
815 if (atomic_inc_return(&stat->disabled) != 1)
819 * Try to find the function again since an NMI
820 * could have added it
822 rec = ftrace_find_profiled_func(stat, ip);
826 if (stat->pages->index == PROFILES_PER_PAGE) {
827 if (!stat->pages->next)
829 stat->pages = stat->pages->next;
832 rec = &stat->pages->records[stat->pages->index++];
834 ftrace_add_profile(stat, rec);
837 atomic_dec(&stat->disabled);
843 function_profile_call(unsigned long ip, unsigned long parent_ip,
844 struct ftrace_ops *ops, struct pt_regs *regs)
846 struct ftrace_profile_stat *stat;
847 struct ftrace_profile *rec;
850 if (!ftrace_profile_enabled)
853 local_irq_save(flags);
855 stat = this_cpu_ptr(&ftrace_profile_stats);
856 if (!stat->hash || !ftrace_profile_enabled)
859 rec = ftrace_find_profiled_func(stat, ip);
861 rec = ftrace_profile_alloc(stat, ip);
868 local_irq_restore(flags);
871 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
872 static int profile_graph_entry(struct ftrace_graph_ent *trace)
874 function_profile_call(trace->func, 0, NULL, NULL);
878 static void profile_graph_return(struct ftrace_graph_ret *trace)
880 struct ftrace_profile_stat *stat;
881 unsigned long long calltime;
882 struct ftrace_profile *rec;
885 local_irq_save(flags);
886 stat = this_cpu_ptr(&ftrace_profile_stats);
887 if (!stat->hash || !ftrace_profile_enabled)
890 /* If the calltime was zero'd ignore it */
891 if (!trace->calltime)
894 calltime = trace->rettime - trace->calltime;
896 if (!fgraph_graph_time) {
899 index = trace->depth;
901 /* Append this call time to the parent time to subtract */
903 current->ret_stack[index - 1].subtime += calltime;
905 if (current->ret_stack[index].subtime < calltime)
906 calltime -= current->ret_stack[index].subtime;
911 rec = ftrace_find_profiled_func(stat, trace->func);
913 rec->time += calltime;
914 rec->time_squared += calltime * calltime;
918 local_irq_restore(flags);
921 static int register_ftrace_profiler(void)
923 return register_ftrace_graph(&profile_graph_return,
924 &profile_graph_entry);
927 static void unregister_ftrace_profiler(void)
929 unregister_ftrace_graph();
932 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
933 .func = function_profile_call,
934 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
935 INIT_OPS_HASH(ftrace_profile_ops)
938 static int register_ftrace_profiler(void)
940 return register_ftrace_function(&ftrace_profile_ops);
943 static void unregister_ftrace_profiler(void)
945 unregister_ftrace_function(&ftrace_profile_ops);
947 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
950 ftrace_profile_write(struct file *filp, const char __user *ubuf,
951 size_t cnt, loff_t *ppos)
956 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
962 mutex_lock(&ftrace_profile_lock);
963 if (ftrace_profile_enabled ^ val) {
965 ret = ftrace_profile_init();
971 ret = register_ftrace_profiler();
976 ftrace_profile_enabled = 1;
978 ftrace_profile_enabled = 0;
980 * unregister_ftrace_profiler calls stop_machine
981 * so this acts like an synchronize_sched.
983 unregister_ftrace_profiler();
987 mutex_unlock(&ftrace_profile_lock);
995 ftrace_profile_read(struct file *filp, char __user *ubuf,
996 size_t cnt, loff_t *ppos)
998 char buf[64]; /* big enough to hold a number */
1001 r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1002 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1005 static const struct file_operations ftrace_profile_fops = {
1006 .open = tracing_open_generic,
1007 .read = ftrace_profile_read,
1008 .write = ftrace_profile_write,
1009 .llseek = default_llseek,
1012 /* used to initialize the real stat files */
1013 static struct tracer_stat function_stats __initdata = {
1014 .name = "functions",
1015 .stat_start = function_stat_start,
1016 .stat_next = function_stat_next,
1017 .stat_cmp = function_stat_cmp,
1018 .stat_headers = function_stat_headers,
1019 .stat_show = function_stat_show
1022 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1024 struct ftrace_profile_stat *stat;
1025 struct dentry *entry;
1030 for_each_possible_cpu(cpu) {
1031 stat = &per_cpu(ftrace_profile_stats, cpu);
1033 name = kasprintf(GFP_KERNEL, "function%d", cpu);
1036 * The files created are permanent, if something happens
1037 * we still do not free memory.
1040 "Could not allocate stat file for cpu %d\n",
1044 stat->stat = function_stats;
1045 stat->stat.name = name;
1046 ret = register_stat_tracer(&stat->stat);
1049 "Could not register function stat for cpu %d\n",
1056 entry = tracefs_create_file("function_profile_enabled", 0644,
1057 d_tracer, NULL, &ftrace_profile_fops);
1059 pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1062 #else /* CONFIG_FUNCTION_PROFILER */
1063 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1066 #endif /* CONFIG_FUNCTION_PROFILER */
1068 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1070 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1071 static int ftrace_graph_active;
1073 # define ftrace_graph_active 0
1076 #ifdef CONFIG_DYNAMIC_FTRACE
1078 static struct ftrace_ops *removed_ops;
1081 * Set when doing a global update, like enabling all recs or disabling them.
1082 * It is not set when just updating a single ftrace_ops.
1084 static bool update_all_ops;
1086 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1087 # error Dynamic ftrace depends on MCOUNT_RECORD
1090 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1092 struct ftrace_func_probe {
1093 struct hlist_node node;
1094 struct ftrace_probe_ops *ops;
1095 unsigned long flags;
1098 struct list_head free_list;
1101 struct ftrace_func_entry {
1102 struct hlist_node hlist;
1106 struct ftrace_hash {
1107 unsigned long size_bits;
1108 struct hlist_head *buckets;
1109 unsigned long count;
1110 struct rcu_head rcu;
1114 * We make these constant because no one should touch them,
1115 * but they are used as the default "empty hash", to avoid allocating
1116 * it all the time. These are in a read only section such that if
1117 * anyone does try to modify it, it will cause an exception.
1119 static const struct hlist_head empty_buckets[1];
1120 static const struct ftrace_hash empty_hash = {
1121 .buckets = (struct hlist_head *)empty_buckets,
1123 #define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
1125 static struct ftrace_ops global_ops = {
1126 .func = ftrace_stub,
1127 .local_hash.notrace_hash = EMPTY_HASH,
1128 .local_hash.filter_hash = EMPTY_HASH,
1129 INIT_OPS_HASH(global_ops)
1130 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
1131 FTRACE_OPS_FL_INITIALIZED |
1136 * This is used by __kernel_text_address() to return true if the
1137 * address is on a dynamically allocated trampoline that would
1138 * not return true for either core_kernel_text() or
1139 * is_module_text_address().
1141 bool is_ftrace_trampoline(unsigned long addr)
1143 struct ftrace_ops *op;
1147 * Some of the ops may be dynamically allocated,
1148 * they are freed after a synchronize_sched().
1150 preempt_disable_notrace();
1152 do_for_each_ftrace_op(op, ftrace_ops_list) {
1154 * This is to check for dynamically allocated trampolines.
1155 * Trampolines that are in kernel text will have
1156 * core_kernel_text() return true.
1158 if (op->trampoline && op->trampoline_size)
1159 if (addr >= op->trampoline &&
1160 addr < op->trampoline + op->trampoline_size) {
1164 } while_for_each_ftrace_op(op);
1167 preempt_enable_notrace();
1172 struct ftrace_page {
1173 struct ftrace_page *next;
1174 struct dyn_ftrace *records;
1179 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1180 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1182 /* estimate from running different kernels */
1183 #define NR_TO_INIT 10000
1185 static struct ftrace_page *ftrace_pages_start;
1186 static struct ftrace_page *ftrace_pages;
1188 static bool __always_inline ftrace_hash_empty(struct ftrace_hash *hash)
1190 return !hash || !hash->count;
1193 static struct ftrace_func_entry *
1194 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1197 struct ftrace_func_entry *entry;
1198 struct hlist_head *hhd;
1200 if (ftrace_hash_empty(hash))
1203 if (hash->size_bits > 0)
1204 key = hash_long(ip, hash->size_bits);
1208 hhd = &hash->buckets[key];
1210 hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1211 if (entry->ip == ip)
1217 static void __add_hash_entry(struct ftrace_hash *hash,
1218 struct ftrace_func_entry *entry)
1220 struct hlist_head *hhd;
1223 if (hash->size_bits)
1224 key = hash_long(entry->ip, hash->size_bits);
1228 hhd = &hash->buckets[key];
1229 hlist_add_head(&entry->hlist, hhd);
1233 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1235 struct ftrace_func_entry *entry;
1237 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1242 __add_hash_entry(hash, entry);
1248 free_hash_entry(struct ftrace_hash *hash,
1249 struct ftrace_func_entry *entry)
1251 hlist_del(&entry->hlist);
1257 remove_hash_entry(struct ftrace_hash *hash,
1258 struct ftrace_func_entry *entry)
1260 hlist_del(&entry->hlist);
1264 static void ftrace_hash_clear(struct ftrace_hash *hash)
1266 struct hlist_head *hhd;
1267 struct hlist_node *tn;
1268 struct ftrace_func_entry *entry;
1269 int size = 1 << hash->size_bits;
1275 for (i = 0; i < size; i++) {
1276 hhd = &hash->buckets[i];
1277 hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1278 free_hash_entry(hash, entry);
1280 FTRACE_WARN_ON(hash->count);
1283 static void free_ftrace_hash(struct ftrace_hash *hash)
1285 if (!hash || hash == EMPTY_HASH)
1287 ftrace_hash_clear(hash);
1288 kfree(hash->buckets);
1292 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1294 struct ftrace_hash *hash;
1296 hash = container_of(rcu, struct ftrace_hash, rcu);
1297 free_ftrace_hash(hash);
1300 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1302 if (!hash || hash == EMPTY_HASH)
1304 call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1307 void ftrace_free_filter(struct ftrace_ops *ops)
1309 ftrace_ops_init(ops);
1310 free_ftrace_hash(ops->func_hash->filter_hash);
1311 free_ftrace_hash(ops->func_hash->notrace_hash);
1314 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1316 struct ftrace_hash *hash;
1319 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1323 size = 1 << size_bits;
1324 hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1326 if (!hash->buckets) {
1331 hash->size_bits = size_bits;
1336 static struct ftrace_hash *
1337 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1339 struct ftrace_func_entry *entry;
1340 struct ftrace_hash *new_hash;
1345 new_hash = alloc_ftrace_hash(size_bits);
1350 if (ftrace_hash_empty(hash))
1353 size = 1 << hash->size_bits;
1354 for (i = 0; i < size; i++) {
1355 hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1356 ret = add_hash_entry(new_hash, entry->ip);
1362 FTRACE_WARN_ON(new_hash->count != hash->count);
1367 free_ftrace_hash(new_hash);
1372 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1374 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1376 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1377 struct ftrace_hash *new_hash);
1380 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1381 struct ftrace_hash **dst, struct ftrace_hash *src)
1383 struct ftrace_func_entry *entry;
1384 struct hlist_node *tn;
1385 struct hlist_head *hhd;
1386 struct ftrace_hash *new_hash;
1387 int size = src->count;
1392 /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1393 if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1397 * If the new source is empty, just free dst and assign it
1401 new_hash = EMPTY_HASH;
1406 * Make the hash size about 1/2 the # found
1408 for (size /= 2; size; size >>= 1)
1411 /* Don't allocate too much */
1412 if (bits > FTRACE_HASH_MAX_BITS)
1413 bits = FTRACE_HASH_MAX_BITS;
1415 new_hash = alloc_ftrace_hash(bits);
1419 size = 1 << src->size_bits;
1420 for (i = 0; i < size; i++) {
1421 hhd = &src->buckets[i];
1422 hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1423 remove_hash_entry(src, entry);
1424 __add_hash_entry(new_hash, entry);
1429 /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1431 /* IPMODIFY should be updated only when filter_hash updating */
1432 ret = ftrace_hash_ipmodify_update(ops, new_hash);
1434 free_ftrace_hash(new_hash);
1440 * Remove the current set, update the hash and add
1443 ftrace_hash_rec_disable_modify(ops, enable);
1445 rcu_assign_pointer(*dst, new_hash);
1447 ftrace_hash_rec_enable_modify(ops, enable);
1452 static bool hash_contains_ip(unsigned long ip,
1453 struct ftrace_ops_hash *hash)
1456 * The function record is a match if it exists in the filter
1457 * hash and not in the notrace hash. Note, an emty hash is
1458 * considered a match for the filter hash, but an empty
1459 * notrace hash is considered not in the notrace hash.
1461 return (ftrace_hash_empty(hash->filter_hash) ||
1462 ftrace_lookup_ip(hash->filter_hash, ip)) &&
1463 (ftrace_hash_empty(hash->notrace_hash) ||
1464 !ftrace_lookup_ip(hash->notrace_hash, ip));
1468 * Test the hashes for this ops to see if we want to call
1469 * the ops->func or not.
1471 * It's a match if the ip is in the ops->filter_hash or
1472 * the filter_hash does not exist or is empty,
1474 * the ip is not in the ops->notrace_hash.
1476 * This needs to be called with preemption disabled as
1477 * the hashes are freed with call_rcu_sched().
1480 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1482 struct ftrace_ops_hash hash;
1485 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1487 * There's a small race when adding ops that the ftrace handler
1488 * that wants regs, may be called without them. We can not
1489 * allow that handler to be called if regs is NULL.
1491 if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1495 hash.filter_hash = rcu_dereference_raw_notrace(ops->func_hash->filter_hash);
1496 hash.notrace_hash = rcu_dereference_raw_notrace(ops->func_hash->notrace_hash);
1498 if (hash_contains_ip(ip, &hash))
1507 * This is a double for. Do not use 'break' to break out of the loop,
1508 * you must use a goto.
1510 #define do_for_each_ftrace_rec(pg, rec) \
1511 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1513 for (_____i = 0; _____i < pg->index; _____i++) { \
1514 rec = &pg->records[_____i];
1516 #define while_for_each_ftrace_rec() \
1521 static int ftrace_cmp_recs(const void *a, const void *b)
1523 const struct dyn_ftrace *key = a;
1524 const struct dyn_ftrace *rec = b;
1526 if (key->flags < rec->ip)
1528 if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1534 * ftrace_location_range - return the first address of a traced location
1535 * if it touches the given ip range
1536 * @start: start of range to search.
1537 * @end: end of range to search (inclusive). @end points to the last byte
1540 * Returns rec->ip if the related ftrace location is a least partly within
1541 * the given address range. That is, the first address of the instruction
1542 * that is either a NOP or call to the function tracer. It checks the ftrace
1543 * internal tables to determine if the address belongs or not.
1545 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1547 struct ftrace_page *pg;
1548 struct dyn_ftrace *rec;
1549 struct dyn_ftrace key;
1552 key.flags = end; /* overload flags, as it is unsigned long */
1554 for (pg = ftrace_pages_start; pg; pg = pg->next) {
1555 if (end < pg->records[0].ip ||
1556 start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1558 rec = bsearch(&key, pg->records, pg->index,
1559 sizeof(struct dyn_ftrace),
1569 * ftrace_location - return true if the ip giving is a traced location
1570 * @ip: the instruction pointer to check
1572 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1573 * That is, the instruction that is either a NOP or call to
1574 * the function tracer. It checks the ftrace internal tables to
1575 * determine if the address belongs or not.
1577 unsigned long ftrace_location(unsigned long ip)
1579 return ftrace_location_range(ip, ip);
1583 * ftrace_text_reserved - return true if range contains an ftrace location
1584 * @start: start of range to search
1585 * @end: end of range to search (inclusive). @end points to the last byte to check.
1587 * Returns 1 if @start and @end contains a ftrace location.
1588 * That is, the instruction that is either a NOP or call to
1589 * the function tracer. It checks the ftrace internal tables to
1590 * determine if the address belongs or not.
1592 int ftrace_text_reserved(const void *start, const void *end)
1596 ret = ftrace_location_range((unsigned long)start,
1597 (unsigned long)end);
1602 /* Test if ops registered to this rec needs regs */
1603 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1605 struct ftrace_ops *ops;
1606 bool keep_regs = false;
1608 for (ops = ftrace_ops_list;
1609 ops != &ftrace_list_end; ops = ops->next) {
1610 /* pass rec in as regs to have non-NULL val */
1611 if (ftrace_ops_test(ops, rec->ip, rec)) {
1612 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1622 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1626 struct ftrace_hash *hash;
1627 struct ftrace_hash *other_hash;
1628 struct ftrace_page *pg;
1629 struct dyn_ftrace *rec;
1630 bool update = false;
1634 /* Only update if the ops has been registered */
1635 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1639 * In the filter_hash case:
1640 * If the count is zero, we update all records.
1641 * Otherwise we just update the items in the hash.
1643 * In the notrace_hash case:
1644 * We enable the update in the hash.
1645 * As disabling notrace means enabling the tracing,
1646 * and enabling notrace means disabling, the inc variable
1650 hash = ops->func_hash->filter_hash;
1651 other_hash = ops->func_hash->notrace_hash;
1652 if (ftrace_hash_empty(hash))
1656 hash = ops->func_hash->notrace_hash;
1657 other_hash = ops->func_hash->filter_hash;
1659 * If the notrace hash has no items,
1660 * then there's nothing to do.
1662 if (ftrace_hash_empty(hash))
1666 do_for_each_ftrace_rec(pg, rec) {
1667 int in_other_hash = 0;
1671 if (rec->flags & FTRACE_FL_DISABLED)
1676 * Only the filter_hash affects all records.
1677 * Update if the record is not in the notrace hash.
1679 if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1682 in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1683 in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1686 * If filter_hash is set, we want to match all functions
1687 * that are in the hash but not in the other hash.
1689 * If filter_hash is not set, then we are decrementing.
1690 * That means we match anything that is in the hash
1691 * and also in the other_hash. That is, we need to turn
1692 * off functions in the other hash because they are disabled
1695 if (filter_hash && in_hash && !in_other_hash)
1697 else if (!filter_hash && in_hash &&
1698 (in_other_hash || ftrace_hash_empty(other_hash)))
1706 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1710 * If there's only a single callback registered to a
1711 * function, and the ops has a trampoline registered
1712 * for it, then we can call it directly.
1714 if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1715 rec->flags |= FTRACE_FL_TRAMP;
1718 * If we are adding another function callback
1719 * to this function, and the previous had a
1720 * custom trampoline in use, then we need to go
1721 * back to the default trampoline.
1723 rec->flags &= ~FTRACE_FL_TRAMP;
1726 * If any ops wants regs saved for this function
1727 * then all ops will get saved regs.
1729 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1730 rec->flags |= FTRACE_FL_REGS;
1732 if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1737 * If the rec had REGS enabled and the ops that is
1738 * being removed had REGS set, then see if there is
1739 * still any ops for this record that wants regs.
1740 * If not, we can stop recording them.
1742 if (ftrace_rec_count(rec) > 0 &&
1743 rec->flags & FTRACE_FL_REGS &&
1744 ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1745 if (!test_rec_ops_needs_regs(rec))
1746 rec->flags &= ~FTRACE_FL_REGS;
1750 * If the rec had TRAMP enabled, then it needs to
1751 * be cleared. As TRAMP can only be enabled iff
1752 * there is only a single ops attached to it.
1753 * In otherwords, always disable it on decrementing.
1754 * In the future, we may set it if rec count is
1755 * decremented to one, and the ops that is left
1758 rec->flags &= ~FTRACE_FL_TRAMP;
1761 * flags will be cleared in ftrace_check_record()
1762 * if rec count is zero.
1767 /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1768 update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1770 /* Shortcut, if we handled all records, we are done. */
1771 if (!all && count == hash->count)
1773 } while_for_each_ftrace_rec();
1778 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1781 return __ftrace_hash_rec_update(ops, filter_hash, 0);
1784 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1787 return __ftrace_hash_rec_update(ops, filter_hash, 1);
1790 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1791 int filter_hash, int inc)
1793 struct ftrace_ops *op;
1795 __ftrace_hash_rec_update(ops, filter_hash, inc);
1797 if (ops->func_hash != &global_ops.local_hash)
1801 * If the ops shares the global_ops hash, then we need to update
1802 * all ops that are enabled and use this hash.
1804 do_for_each_ftrace_op(op, ftrace_ops_list) {
1808 if (op->func_hash == &global_ops.local_hash)
1809 __ftrace_hash_rec_update(op, filter_hash, inc);
1810 } while_for_each_ftrace_op(op);
1813 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1816 ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1819 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1822 ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1826 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1827 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1828 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1829 * Note that old_hash and new_hash has below meanings
1830 * - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1831 * - If the hash is EMPTY_HASH, it hits nothing
1832 * - Anything else hits the recs which match the hash entries.
1834 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1835 struct ftrace_hash *old_hash,
1836 struct ftrace_hash *new_hash)
1838 struct ftrace_page *pg;
1839 struct dyn_ftrace *rec, *end = NULL;
1842 /* Only update if the ops has been registered */
1843 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1846 if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1850 * Since the IPMODIFY is a very address sensitive action, we do not
1851 * allow ftrace_ops to set all functions to new hash.
1853 if (!new_hash || !old_hash)
1856 /* Update rec->flags */
1857 do_for_each_ftrace_rec(pg, rec) {
1858 /* We need to update only differences of filter_hash */
1859 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1860 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1861 if (in_old == in_new)
1865 /* New entries must ensure no others are using it */
1866 if (rec->flags & FTRACE_FL_IPMODIFY)
1868 rec->flags |= FTRACE_FL_IPMODIFY;
1869 } else /* Removed entry */
1870 rec->flags &= ~FTRACE_FL_IPMODIFY;
1871 } while_for_each_ftrace_rec();
1878 /* Roll back what we did above */
1879 do_for_each_ftrace_rec(pg, rec) {
1883 in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1884 in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1885 if (in_old == in_new)
1889 rec->flags &= ~FTRACE_FL_IPMODIFY;
1891 rec->flags |= FTRACE_FL_IPMODIFY;
1892 } while_for_each_ftrace_rec();
1898 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1900 struct ftrace_hash *hash = ops->func_hash->filter_hash;
1902 if (ftrace_hash_empty(hash))
1905 return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1908 /* Disabling always succeeds */
1909 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1911 struct ftrace_hash *hash = ops->func_hash->filter_hash;
1913 if (ftrace_hash_empty(hash))
1916 __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1919 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1920 struct ftrace_hash *new_hash)
1922 struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1924 if (ftrace_hash_empty(old_hash))
1927 if (ftrace_hash_empty(new_hash))
1930 return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
1933 static void print_ip_ins(const char *fmt, const unsigned char *p)
1937 printk(KERN_CONT "%s", fmt);
1939 for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1940 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1943 static struct ftrace_ops *
1944 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1945 static struct ftrace_ops *
1946 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1948 enum ftrace_bug_type ftrace_bug_type;
1949 const void *ftrace_expected;
1951 static void print_bug_type(void)
1953 switch (ftrace_bug_type) {
1954 case FTRACE_BUG_UNKNOWN:
1956 case FTRACE_BUG_INIT:
1957 pr_info("Initializing ftrace call sites\n");
1959 case FTRACE_BUG_NOP:
1960 pr_info("Setting ftrace call site to NOP\n");
1962 case FTRACE_BUG_CALL:
1963 pr_info("Setting ftrace call site to call ftrace function\n");
1965 case FTRACE_BUG_UPDATE:
1966 pr_info("Updating ftrace call site to call a different ftrace function\n");
1972 * ftrace_bug - report and shutdown function tracer
1973 * @failed: The failed type (EFAULT, EINVAL, EPERM)
1974 * @rec: The record that failed
1976 * The arch code that enables or disables the function tracing
1977 * can call ftrace_bug() when it has detected a problem in
1978 * modifying the code. @failed should be one of either:
1979 * EFAULT - if the problem happens on reading the @ip address
1980 * EINVAL - if what is read at @ip is not what was expected
1981 * EPERM - if the problem happens on writting to the @ip address
1983 void ftrace_bug(int failed, struct dyn_ftrace *rec)
1985 unsigned long ip = rec ? rec->ip : 0;
1989 FTRACE_WARN_ON_ONCE(1);
1990 pr_info("ftrace faulted on modifying ");
1994 FTRACE_WARN_ON_ONCE(1);
1995 pr_info("ftrace failed to modify ");
1997 print_ip_ins(" actual: ", (unsigned char *)ip);
1999 if (ftrace_expected) {
2000 print_ip_ins(" expected: ", ftrace_expected);
2005 FTRACE_WARN_ON_ONCE(1);
2006 pr_info("ftrace faulted on writing ");
2010 FTRACE_WARN_ON_ONCE(1);
2011 pr_info("ftrace faulted on unknown error ");
2016 struct ftrace_ops *ops = NULL;
2018 pr_info("ftrace record flags: %lx\n", rec->flags);
2019 pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2020 rec->flags & FTRACE_FL_REGS ? " R" : " ");
2021 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2022 ops = ftrace_find_tramp_ops_any(rec);
2025 pr_cont("\ttramp: %pS (%pS)",
2026 (void *)ops->trampoline,
2028 ops = ftrace_find_tramp_ops_next(rec, ops);
2031 pr_cont("\ttramp: ERROR!");
2034 ip = ftrace_get_addr_curr(rec);
2035 pr_cont("\n expected tramp: %lx\n", ip);
2039 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2041 unsigned long flag = 0UL;
2043 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2045 if (rec->flags & FTRACE_FL_DISABLED)
2046 return FTRACE_UPDATE_IGNORE;
2049 * If we are updating calls:
2051 * If the record has a ref count, then we need to enable it
2052 * because someone is using it.
2054 * Otherwise we make sure its disabled.
2056 * If we are disabling calls, then disable all records that
2059 if (enable && ftrace_rec_count(rec))
2060 flag = FTRACE_FL_ENABLED;
2063 * If enabling and the REGS flag does not match the REGS_EN, or
2064 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2065 * this record. Set flags to fail the compare against ENABLED.
2068 if (!(rec->flags & FTRACE_FL_REGS) !=
2069 !(rec->flags & FTRACE_FL_REGS_EN))
2070 flag |= FTRACE_FL_REGS;
2072 if (!(rec->flags & FTRACE_FL_TRAMP) !=
2073 !(rec->flags & FTRACE_FL_TRAMP_EN))
2074 flag |= FTRACE_FL_TRAMP;
2077 /* If the state of this record hasn't changed, then do nothing */
2078 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2079 return FTRACE_UPDATE_IGNORE;
2082 /* Save off if rec is being enabled (for return value) */
2083 flag ^= rec->flags & FTRACE_FL_ENABLED;
2086 rec->flags |= FTRACE_FL_ENABLED;
2087 if (flag & FTRACE_FL_REGS) {
2088 if (rec->flags & FTRACE_FL_REGS)
2089 rec->flags |= FTRACE_FL_REGS_EN;
2091 rec->flags &= ~FTRACE_FL_REGS_EN;
2093 if (flag & FTRACE_FL_TRAMP) {
2094 if (rec->flags & FTRACE_FL_TRAMP)
2095 rec->flags |= FTRACE_FL_TRAMP_EN;
2097 rec->flags &= ~FTRACE_FL_TRAMP_EN;
2102 * If this record is being updated from a nop, then
2103 * return UPDATE_MAKE_CALL.
2105 * return UPDATE_MODIFY_CALL to tell the caller to convert
2106 * from the save regs, to a non-save regs function or
2107 * vice versa, or from a trampoline call.
2109 if (flag & FTRACE_FL_ENABLED) {
2110 ftrace_bug_type = FTRACE_BUG_CALL;
2111 return FTRACE_UPDATE_MAKE_CALL;
2114 ftrace_bug_type = FTRACE_BUG_UPDATE;
2115 return FTRACE_UPDATE_MODIFY_CALL;
2119 /* If there's no more users, clear all flags */
2120 if (!ftrace_rec_count(rec))
2124 * Just disable the record, but keep the ops TRAMP
2125 * and REGS states. The _EN flags must be disabled though.
2127 rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2131 ftrace_bug_type = FTRACE_BUG_NOP;
2132 return FTRACE_UPDATE_MAKE_NOP;
2136 * ftrace_update_record, set a record that now is tracing or not
2137 * @rec: the record to update
2138 * @enable: set to 1 if the record is tracing, zero to force disable
2140 * The records that represent all functions that can be traced need
2141 * to be updated when tracing has been enabled.
2143 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2145 return ftrace_check_record(rec, enable, 1);
2149 * ftrace_test_record, check if the record has been enabled or not
2150 * @rec: the record to test
2151 * @enable: set to 1 to check if enabled, 0 if it is disabled
2153 * The arch code may need to test if a record is already set to
2154 * tracing to determine how to modify the function code that it
2157 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2159 return ftrace_check_record(rec, enable, 0);
2162 static struct ftrace_ops *
2163 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2165 struct ftrace_ops *op;
2166 unsigned long ip = rec->ip;
2168 do_for_each_ftrace_op(op, ftrace_ops_list) {
2170 if (!op->trampoline)
2173 if (hash_contains_ip(ip, op->func_hash))
2175 } while_for_each_ftrace_op(op);
2180 static struct ftrace_ops *
2181 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2182 struct ftrace_ops *op)
2184 unsigned long ip = rec->ip;
2186 while_for_each_ftrace_op(op) {
2188 if (!op->trampoline)
2191 if (hash_contains_ip(ip, op->func_hash))
2198 static struct ftrace_ops *
2199 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2201 struct ftrace_ops *op;
2202 unsigned long ip = rec->ip;
2205 * Need to check removed ops first.
2206 * If they are being removed, and this rec has a tramp,
2207 * and this rec is in the ops list, then it would be the
2208 * one with the tramp.
2211 if (hash_contains_ip(ip, &removed_ops->old_hash))
2216 * Need to find the current trampoline for a rec.
2217 * Now, a trampoline is only attached to a rec if there
2218 * was a single 'ops' attached to it. But this can be called
2219 * when we are adding another op to the rec or removing the
2220 * current one. Thus, if the op is being added, we can
2221 * ignore it because it hasn't attached itself to the rec
2224 * If an ops is being modified (hooking to different functions)
2225 * then we don't care about the new functions that are being
2226 * added, just the old ones (that are probably being removed).
2228 * If we are adding an ops to a function that already is using
2229 * a trampoline, it needs to be removed (trampolines are only
2230 * for single ops connected), then an ops that is not being
2231 * modified also needs to be checked.
2233 do_for_each_ftrace_op(op, ftrace_ops_list) {
2235 if (!op->trampoline)
2239 * If the ops is being added, it hasn't gotten to
2240 * the point to be removed from this tree yet.
2242 if (op->flags & FTRACE_OPS_FL_ADDING)
2247 * If the ops is being modified and is in the old
2248 * hash, then it is probably being removed from this
2251 if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2252 hash_contains_ip(ip, &op->old_hash))
2255 * If the ops is not being added or modified, and it's
2256 * in its normal filter hash, then this must be the one
2259 if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2260 hash_contains_ip(ip, op->func_hash))
2263 } while_for_each_ftrace_op(op);
2268 static struct ftrace_ops *
2269 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2271 struct ftrace_ops *op;
2272 unsigned long ip = rec->ip;
2274 do_for_each_ftrace_op(op, ftrace_ops_list) {
2275 /* pass rec in as regs to have non-NULL val */
2276 if (hash_contains_ip(ip, op->func_hash))
2278 } while_for_each_ftrace_op(op);
2284 * ftrace_get_addr_new - Get the call address to set to
2285 * @rec: The ftrace record descriptor
2287 * If the record has the FTRACE_FL_REGS set, that means that it
2288 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2289 * is not not set, then it wants to convert to the normal callback.
2291 * Returns the address of the trampoline to set to
2293 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2295 struct ftrace_ops *ops;
2297 /* Trampolines take precedence over regs */
2298 if (rec->flags & FTRACE_FL_TRAMP) {
2299 ops = ftrace_find_tramp_ops_new(rec);
2300 if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2301 pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2302 (void *)rec->ip, (void *)rec->ip, rec->flags);
2303 /* Ftrace is shutting down, return anything */
2304 return (unsigned long)FTRACE_ADDR;
2306 return ops->trampoline;
2309 if (rec->flags & FTRACE_FL_REGS)
2310 return (unsigned long)FTRACE_REGS_ADDR;
2312 return (unsigned long)FTRACE_ADDR;
2316 * ftrace_get_addr_curr - Get the call address that is already there
2317 * @rec: The ftrace record descriptor
2319 * The FTRACE_FL_REGS_EN is set when the record already points to
2320 * a function that saves all the regs. Basically the '_EN' version
2321 * represents the current state of the function.
2323 * Returns the address of the trampoline that is currently being called
2325 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2327 struct ftrace_ops *ops;
2329 /* Trampolines take precedence over regs */
2330 if (rec->flags & FTRACE_FL_TRAMP_EN) {
2331 ops = ftrace_find_tramp_ops_curr(rec);
2332 if (FTRACE_WARN_ON(!ops)) {
2333 pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2334 (void *)rec->ip, (void *)rec->ip);
2335 /* Ftrace is shutting down, return anything */
2336 return (unsigned long)FTRACE_ADDR;
2338 return ops->trampoline;
2341 if (rec->flags & FTRACE_FL_REGS_EN)
2342 return (unsigned long)FTRACE_REGS_ADDR;
2344 return (unsigned long)FTRACE_ADDR;
2348 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2350 unsigned long ftrace_old_addr;
2351 unsigned long ftrace_addr;
2354 ftrace_addr = ftrace_get_addr_new(rec);
2356 /* This needs to be done before we call ftrace_update_record */
2357 ftrace_old_addr = ftrace_get_addr_curr(rec);
2359 ret = ftrace_update_record(rec, enable);
2361 ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2364 case FTRACE_UPDATE_IGNORE:
2367 case FTRACE_UPDATE_MAKE_CALL:
2368 ftrace_bug_type = FTRACE_BUG_CALL;
2369 return ftrace_make_call(rec, ftrace_addr);
2371 case FTRACE_UPDATE_MAKE_NOP:
2372 ftrace_bug_type = FTRACE_BUG_NOP;
2373 return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2375 case FTRACE_UPDATE_MODIFY_CALL:
2376 ftrace_bug_type = FTRACE_BUG_UPDATE;
2377 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2380 return -1; /* unknow ftrace bug */
2383 void __weak ftrace_replace_code(int enable)
2385 struct dyn_ftrace *rec;
2386 struct ftrace_page *pg;
2389 if (unlikely(ftrace_disabled))
2392 do_for_each_ftrace_rec(pg, rec) {
2393 failed = __ftrace_replace_code(rec, enable);
2395 ftrace_bug(failed, rec);
2396 /* Stop processing */
2399 } while_for_each_ftrace_rec();
2402 struct ftrace_rec_iter {
2403 struct ftrace_page *pg;
2408 * ftrace_rec_iter_start, start up iterating over traced functions
2410 * Returns an iterator handle that is used to iterate over all
2411 * the records that represent address locations where functions
2414 * May return NULL if no records are available.
2416 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2419 * We only use a single iterator.
2420 * Protected by the ftrace_lock mutex.
2422 static struct ftrace_rec_iter ftrace_rec_iter;
2423 struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2425 iter->pg = ftrace_pages_start;
2428 /* Could have empty pages */
2429 while (iter->pg && !iter->pg->index)
2430 iter->pg = iter->pg->next;
2439 * ftrace_rec_iter_next, get the next record to process.
2440 * @iter: The handle to the iterator.
2442 * Returns the next iterator after the given iterator @iter.
2444 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2448 if (iter->index >= iter->pg->index) {
2449 iter->pg = iter->pg->next;
2452 /* Could have empty pages */
2453 while (iter->pg && !iter->pg->index)
2454 iter->pg = iter->pg->next;
2464 * ftrace_rec_iter_record, get the record at the iterator location
2465 * @iter: The current iterator location
2467 * Returns the record that the current @iter is at.
2469 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2471 return &iter->pg->records[iter->index];
2475 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2479 if (unlikely(ftrace_disabled))
2482 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2484 ftrace_bug_type = FTRACE_BUG_INIT;
2485 ftrace_bug(ret, rec);
2492 * archs can override this function if they must do something
2493 * before the modifying code is performed.
2495 int __weak ftrace_arch_code_modify_prepare(void)
2501 * archs can override this function if they must do something
2502 * after the modifying code is performed.
2504 int __weak ftrace_arch_code_modify_post_process(void)
2509 void ftrace_modify_all_code(int command)
2511 int update = command & FTRACE_UPDATE_TRACE_FUNC;
2515 * If the ftrace_caller calls a ftrace_ops func directly,
2516 * we need to make sure that it only traces functions it
2517 * expects to trace. When doing the switch of functions,
2518 * we need to update to the ftrace_ops_list_func first
2519 * before the transition between old and new calls are set,
2520 * as the ftrace_ops_list_func will check the ops hashes
2521 * to make sure the ops are having the right functions
2525 err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2526 if (FTRACE_WARN_ON(err))
2530 if (command & FTRACE_UPDATE_CALLS)
2531 ftrace_replace_code(1);
2532 else if (command & FTRACE_DISABLE_CALLS)
2533 ftrace_replace_code(0);
2535 if (update && ftrace_trace_function != ftrace_ops_list_func) {
2536 function_trace_op = set_function_trace_op;
2538 /* If irqs are disabled, we are in stop machine */
2539 if (!irqs_disabled())
2540 smp_call_function(ftrace_sync_ipi, NULL, 1);
2541 err = ftrace_update_ftrace_func(ftrace_trace_function);
2542 if (FTRACE_WARN_ON(err))
2546 if (command & FTRACE_START_FUNC_RET)
2547 err = ftrace_enable_ftrace_graph_caller();
2548 else if (command & FTRACE_STOP_FUNC_RET)
2549 err = ftrace_disable_ftrace_graph_caller();
2550 FTRACE_WARN_ON(err);
2553 static int __ftrace_modify_code(void *data)
2555 int *command = data;
2557 ftrace_modify_all_code(*command);
2563 * ftrace_run_stop_machine, go back to the stop machine method
2564 * @command: The command to tell ftrace what to do
2566 * If an arch needs to fall back to the stop machine method, the
2567 * it can call this function.
2569 void ftrace_run_stop_machine(int command)
2571 stop_machine(__ftrace_modify_code, &command, NULL);
2575 * arch_ftrace_update_code, modify the code to trace or not trace
2576 * @command: The command that needs to be done
2578 * Archs can override this function if it does not need to
2579 * run stop_machine() to modify code.
2581 void __weak arch_ftrace_update_code(int command)
2583 ftrace_run_stop_machine(command);
2586 static void ftrace_run_update_code(int command)
2590 ret = ftrace_arch_code_modify_prepare();
2591 FTRACE_WARN_ON(ret);
2596 * By default we use stop_machine() to modify the code.
2597 * But archs can do what ever they want as long as it
2598 * is safe. The stop_machine() is the safest, but also
2599 * produces the most overhead.
2601 arch_ftrace_update_code(command);
2603 ret = ftrace_arch_code_modify_post_process();
2604 FTRACE_WARN_ON(ret);
2607 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2608 struct ftrace_ops_hash *old_hash)
2610 ops->flags |= FTRACE_OPS_FL_MODIFYING;
2611 ops->old_hash.filter_hash = old_hash->filter_hash;
2612 ops->old_hash.notrace_hash = old_hash->notrace_hash;
2613 ftrace_run_update_code(command);
2614 ops->old_hash.filter_hash = NULL;
2615 ops->old_hash.notrace_hash = NULL;
2616 ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2619 static ftrace_func_t saved_ftrace_func;
2620 static int ftrace_start_up;
2622 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2626 static void per_cpu_ops_free(struct ftrace_ops *ops)
2628 free_percpu(ops->disabled);
2631 static void ftrace_startup_enable(int command)
2633 if (saved_ftrace_func != ftrace_trace_function) {
2634 saved_ftrace_func = ftrace_trace_function;
2635 command |= FTRACE_UPDATE_TRACE_FUNC;
2638 if (!command || !ftrace_enabled)
2641 ftrace_run_update_code(command);
2644 static void ftrace_startup_all(int command)
2646 update_all_ops = true;
2647 ftrace_startup_enable(command);
2648 update_all_ops = false;
2651 static int ftrace_startup(struct ftrace_ops *ops, int command)
2655 if (unlikely(ftrace_disabled))
2658 ret = __register_ftrace_function(ops);
2665 * Note that ftrace probes uses this to start up
2666 * and modify functions it will probe. But we still
2667 * set the ADDING flag for modification, as probes
2668 * do not have trampolines. If they add them in the
2669 * future, then the probes will need to distinguish
2670 * between adding and updating probes.
2672 ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2674 ret = ftrace_hash_ipmodify_enable(ops);
2676 /* Rollback registration process */
2677 __unregister_ftrace_function(ops);
2679 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2683 if (ftrace_hash_rec_enable(ops, 1))
2684 command |= FTRACE_UPDATE_CALLS;
2686 ftrace_startup_enable(command);
2688 ops->flags &= ~FTRACE_OPS_FL_ADDING;
2693 static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2697 if (unlikely(ftrace_disabled))
2700 ret = __unregister_ftrace_function(ops);
2706 * Just warn in case of unbalance, no need to kill ftrace, it's not
2707 * critical but the ftrace_call callers may be never nopped again after
2708 * further ftrace uses.
2710 WARN_ON_ONCE(ftrace_start_up < 0);
2712 /* Disabling ipmodify never fails */
2713 ftrace_hash_ipmodify_disable(ops);
2715 if (ftrace_hash_rec_disable(ops, 1))
2716 command |= FTRACE_UPDATE_CALLS;
2718 ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2720 if (saved_ftrace_func != ftrace_trace_function) {
2721 saved_ftrace_func = ftrace_trace_function;
2722 command |= FTRACE_UPDATE_TRACE_FUNC;
2725 if (!command || !ftrace_enabled) {
2727 * If these are per_cpu ops, they still need their
2728 * per_cpu field freed. Since, function tracing is
2729 * not currently active, we can just free them
2730 * without synchronizing all CPUs.
2732 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2733 per_cpu_ops_free(ops);
2738 * If the ops uses a trampoline, then it needs to be
2739 * tested first on update.
2741 ops->flags |= FTRACE_OPS_FL_REMOVING;
2744 /* The trampoline logic checks the old hashes */
2745 ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2746 ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2748 ftrace_run_update_code(command);
2751 * If there's no more ops registered with ftrace, run a
2752 * sanity check to make sure all rec flags are cleared.
2754 if (ftrace_ops_list == &ftrace_list_end) {
2755 struct ftrace_page *pg;
2756 struct dyn_ftrace *rec;
2758 do_for_each_ftrace_rec(pg, rec) {
2759 if (FTRACE_WARN_ON_ONCE(rec->flags))
2760 pr_warn(" %pS flags:%lx\n",
2761 (void *)rec->ip, rec->flags);
2762 } while_for_each_ftrace_rec();
2765 ops->old_hash.filter_hash = NULL;
2766 ops->old_hash.notrace_hash = NULL;
2769 ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2772 * Dynamic ops may be freed, we must make sure that all
2773 * callers are done before leaving this function.
2774 * The same goes for freeing the per_cpu data of the per_cpu
2777 * Again, normal synchronize_sched() is not good enough.
2778 * We need to do a hard force of sched synchronization.
2779 * This is because we use preempt_disable() to do RCU, but
2780 * the function tracers can be called where RCU is not watching
2781 * (like before user_exit()). We can not rely on the RCU
2782 * infrastructure to do the synchronization, thus we must do it
2785 if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) {
2786 schedule_on_each_cpu(ftrace_sync);
2788 arch_ftrace_trampoline_free(ops);
2790 if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2791 per_cpu_ops_free(ops);
2797 static void ftrace_startup_sysctl(void)
2801 if (unlikely(ftrace_disabled))
2804 /* Force update next time */
2805 saved_ftrace_func = NULL;
2806 /* ftrace_start_up is true if we want ftrace running */
2807 if (ftrace_start_up) {
2808 command = FTRACE_UPDATE_CALLS;
2809 if (ftrace_graph_active)
2810 command |= FTRACE_START_FUNC_RET;
2811 ftrace_startup_enable(command);
2815 static void ftrace_shutdown_sysctl(void)
2819 if (unlikely(ftrace_disabled))
2822 /* ftrace_start_up is true if ftrace is running */
2823 if (ftrace_start_up) {
2824 command = FTRACE_DISABLE_CALLS;
2825 if (ftrace_graph_active)
2826 command |= FTRACE_STOP_FUNC_RET;
2827 ftrace_run_update_code(command);
2831 static cycle_t ftrace_update_time;
2832 unsigned long ftrace_update_tot_cnt;
2834 static inline int ops_traces_mod(struct ftrace_ops *ops)
2837 * Filter_hash being empty will default to trace module.
2838 * But notrace hash requires a test of individual module functions.
2840 return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2841 ftrace_hash_empty(ops->func_hash->notrace_hash);
2845 * Check if the current ops references the record.
2847 * If the ops traces all functions, then it was already accounted for.
2848 * If the ops does not trace the current record function, skip it.
2849 * If the ops ignores the function via notrace filter, skip it.
2852 ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2854 /* If ops isn't enabled, ignore it */
2855 if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2858 /* If ops traces all then it includes this function */
2859 if (ops_traces_mod(ops))
2862 /* The function must be in the filter */
2863 if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2864 !ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2867 /* If in notrace hash, we ignore it too */
2868 if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2874 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2876 struct ftrace_page *pg;
2877 struct dyn_ftrace *p;
2878 cycle_t start, stop;
2879 unsigned long update_cnt = 0;
2880 unsigned long rec_flags = 0;
2883 start = ftrace_now(raw_smp_processor_id());
2886 * When a module is loaded, this function is called to convert
2887 * the calls to mcount in its text to nops, and also to create
2888 * an entry in the ftrace data. Now, if ftrace is activated
2889 * after this call, but before the module sets its text to
2890 * read-only, the modification of enabling ftrace can fail if
2891 * the read-only is done while ftrace is converting the calls.
2892 * To prevent this, the module's records are set as disabled
2893 * and will be enabled after the call to set the module's text
2897 rec_flags |= FTRACE_FL_DISABLED;
2899 for (pg = new_pgs; pg; pg = pg->next) {
2901 for (i = 0; i < pg->index; i++) {
2903 /* If something went wrong, bail without enabling anything */
2904 if (unlikely(ftrace_disabled))
2907 p = &pg->records[i];
2908 p->flags = rec_flags;
2911 * Do the initial record conversion from mcount jump
2912 * to the NOP instructions.
2914 if (!ftrace_code_disable(mod, p))
2921 stop = ftrace_now(raw_smp_processor_id());
2922 ftrace_update_time = stop - start;
2923 ftrace_update_tot_cnt += update_cnt;
2928 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2933 if (WARN_ON(!count))
2936 order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2939 * We want to fill as much as possible. No more than a page
2942 while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2946 pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2949 /* if we can't allocate this size, try something smaller */
2956 cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2965 static struct ftrace_page *
2966 ftrace_allocate_pages(unsigned long num_to_init)
2968 struct ftrace_page *start_pg;
2969 struct ftrace_page *pg;
2976 start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2981 * Try to allocate as much as possible in one continues
2982 * location that fills in all of the space. We want to
2983 * waste as little space as possible.
2986 cnt = ftrace_allocate_records(pg, num_to_init);
2994 pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3006 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3007 free_pages((unsigned long)pg->records, order);
3008 start_pg = pg->next;
3012 pr_info("ftrace: FAILED to allocate memory for functions\n");
3016 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3018 struct ftrace_iterator {
3021 struct ftrace_page *pg;
3022 struct dyn_ftrace *func;
3023 struct ftrace_func_probe *probe;
3024 struct trace_parser parser;
3025 struct ftrace_hash *hash;
3026 struct ftrace_ops *ops;
3033 t_hash_next(struct seq_file *m, loff_t *pos)
3035 struct ftrace_iterator *iter = m->private;
3036 struct hlist_node *hnd = NULL;
3037 struct hlist_head *hhd;
3043 hnd = &iter->probe->node;
3045 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
3048 hhd = &ftrace_func_hash[iter->hidx];
3050 if (hlist_empty(hhd)) {
3066 if (WARN_ON_ONCE(!hnd))
3069 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
3074 static void *t_hash_start(struct seq_file *m, loff_t *pos)
3076 struct ftrace_iterator *iter = m->private;
3080 if (!(iter->flags & FTRACE_ITER_DO_HASH))
3083 if (iter->func_pos > *pos)
3087 for (l = 0; l <= (*pos - iter->func_pos); ) {
3088 p = t_hash_next(m, &l);
3095 /* Only set this if we have an item */
3096 iter->flags |= FTRACE_ITER_HASH;
3102 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
3104 struct ftrace_func_probe *rec;
3107 if (WARN_ON_ONCE(!rec))
3110 if (rec->ops->print)
3111 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
3113 seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
3116 seq_printf(m, ":%p", rec->data);
3123 t_next(struct seq_file *m, void *v, loff_t *pos)
3125 struct ftrace_iterator *iter = m->private;
3126 struct ftrace_ops *ops = iter->ops;
3127 struct dyn_ftrace *rec = NULL;
3129 if (unlikely(ftrace_disabled))
3132 if (iter->flags & FTRACE_ITER_HASH)
3133 return t_hash_next(m, pos);
3136 iter->pos = iter->func_pos = *pos;
3138 if (iter->flags & FTRACE_ITER_PRINTALL)
3139 return t_hash_start(m, pos);
3142 if (iter->idx >= iter->pg->index) {
3143 if (iter->pg->next) {
3144 iter->pg = iter->pg->next;
3149 rec = &iter->pg->records[iter->idx++];
3150 if (((iter->flags & FTRACE_ITER_FILTER) &&
3151 !(ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))) ||
3153 ((iter->flags & FTRACE_ITER_NOTRACE) &&
3154 !ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip)) ||
3156 ((iter->flags & FTRACE_ITER_ENABLED) &&
3157 !(rec->flags & FTRACE_FL_ENABLED))) {
3165 return t_hash_start(m, pos);
3172 static void reset_iter_read(struct ftrace_iterator *iter)
3176 iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
3179 static void *t_start(struct seq_file *m, loff_t *pos)
3181 struct ftrace_iterator *iter = m->private;
3182 struct ftrace_ops *ops = iter->ops;
3186 mutex_lock(&ftrace_lock);
3188 if (unlikely(ftrace_disabled))
3192 * If an lseek was done, then reset and start from beginning.
3194 if (*pos < iter->pos)
3195 reset_iter_read(iter);
3198 * For set_ftrace_filter reading, if we have the filter
3199 * off, we can short cut and just print out that all
3200 * functions are enabled.
3202 if ((iter->flags & FTRACE_ITER_FILTER &&
3203 ftrace_hash_empty(ops->func_hash->filter_hash)) ||
3204 (iter->flags & FTRACE_ITER_NOTRACE &&
3205 ftrace_hash_empty(ops->func_hash->notrace_hash))) {
3207 return t_hash_start(m, pos);
3208 iter->flags |= FTRACE_ITER_PRINTALL;
3209 /* reset in case of seek/pread */
3210 iter->flags &= ~FTRACE_ITER_HASH;
3214 if (iter->flags & FTRACE_ITER_HASH)
3215 return t_hash_start(m, pos);
3218 * Unfortunately, we need to restart at ftrace_pages_start
3219 * every time we let go of the ftrace_mutex. This is because
3220 * those pointers can change without the lock.
3222 iter->pg = ftrace_pages_start;
3224 for (l = 0; l <= *pos; ) {
3225 p = t_next(m, p, &l);
3231 return t_hash_start(m, pos);
3236 static void t_stop(struct seq_file *m, void *p)
3238 mutex_unlock(&ftrace_lock);
3242 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3247 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3248 struct dyn_ftrace *rec)
3252 ptr = arch_ftrace_trampoline_func(ops, rec);
3254 seq_printf(m, " ->%pS", ptr);
3257 static int t_show(struct seq_file *m, void *v)
3259 struct ftrace_iterator *iter = m->private;
3260 struct dyn_ftrace *rec;
3262 if (iter->flags & FTRACE_ITER_HASH)
3263 return t_hash_show(m, iter);
3265 if (iter->flags & FTRACE_ITER_PRINTALL) {
3266 if (iter->flags & FTRACE_ITER_NOTRACE)
3267 seq_puts(m, "#### no functions disabled ####\n");
3269 seq_puts(m, "#### all functions enabled ####\n");
3278 seq_printf(m, "%ps", (void *)rec->ip);
3279 if (iter->flags & FTRACE_ITER_ENABLED) {
3280 struct ftrace_ops *ops;
3282 seq_printf(m, " (%ld)%s%s",
3283 ftrace_rec_count(rec),
3284 rec->flags & FTRACE_FL_REGS ? " R" : " ",
3285 rec->flags & FTRACE_FL_IPMODIFY ? " I" : " ");
3286 if (rec->flags & FTRACE_FL_TRAMP_EN) {
3287 ops = ftrace_find_tramp_ops_any(rec);
3290 seq_printf(m, "\ttramp: %pS (%pS)",
3291 (void *)ops->trampoline,
3293 add_trampoline_func(m, ops, rec);
3294 ops = ftrace_find_tramp_ops_next(rec, ops);
3297 seq_puts(m, "\ttramp: ERROR!");
3299 add_trampoline_func(m, NULL, rec);
3308 static const struct seq_operations show_ftrace_seq_ops = {
3316 ftrace_avail_open(struct inode *inode, struct file *file)
3318 struct ftrace_iterator *iter;
3320 if (unlikely(ftrace_disabled))
3323 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3325 iter->pg = ftrace_pages_start;
3326 iter->ops = &global_ops;
3329 return iter ? 0 : -ENOMEM;
3333 ftrace_enabled_open(struct inode *inode, struct file *file)
3335 struct ftrace_iterator *iter;
3337 iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3339 iter->pg = ftrace_pages_start;
3340 iter->flags = FTRACE_ITER_ENABLED;
3341 iter->ops = &global_ops;
3344 return iter ? 0 : -ENOMEM;
3348 * ftrace_regex_open - initialize function tracer filter files
3349 * @ops: The ftrace_ops that hold the hash filters
3350 * @flag: The type of filter to process
3351 * @inode: The inode, usually passed in to your open routine
3352 * @file: The file, usually passed in to your open routine
3354 * ftrace_regex_open() initializes the filter files for the
3355 * @ops. Depending on @flag it may process the filter hash or
3356 * the notrace hash of @ops. With this called from the open
3357 * routine, you can use ftrace_filter_write() for the write
3358 * routine if @flag has FTRACE_ITER_FILTER set, or
3359 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3360 * tracing_lseek() should be used as the lseek routine, and
3361 * release must call ftrace_regex_release().
3364 ftrace_regex_open(struct ftrace_ops *ops, int flag,
3365 struct inode *inode, struct file *file)
3367 struct ftrace_iterator *iter;
3368 struct ftrace_hash *hash;
3371 ftrace_ops_init(ops);
3373 if (unlikely(ftrace_disabled))
3376 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3380 if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3388 mutex_lock(&ops->func_hash->regex_lock);
3390 if (flag & FTRACE_ITER_NOTRACE)
3391 hash = ops->func_hash->notrace_hash;
3393 hash = ops->func_hash->filter_hash;
3395 if (file->f_mode & FMODE_WRITE) {
3396 const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3398 if (file->f_flags & O_TRUNC)
3399 iter->hash = alloc_ftrace_hash(size_bits);
3401 iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3404 trace_parser_put(&iter->parser);
3411 if (file->f_mode & FMODE_READ) {
3412 iter->pg = ftrace_pages_start;
3414 ret = seq_open(file, &show_ftrace_seq_ops);
3416 struct seq_file *m = file->private_data;
3420 free_ftrace_hash(iter->hash);
3421 trace_parser_put(&iter->parser);
3425 file->private_data = iter;
3428 mutex_unlock(&ops->func_hash->regex_lock);
3434 ftrace_filter_open(struct inode *inode, struct file *file)
3436 struct ftrace_ops *ops = inode->i_private;
3438 return ftrace_regex_open(ops,
3439 FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
3444 ftrace_notrace_open(struct inode *inode, struct file *file)
3446 struct ftrace_ops *ops = inode->i_private;
3448 return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3452 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3453 struct ftrace_glob {
3460 * If symbols in an architecture don't correspond exactly to the user-visible
3461 * name of what they represent, it is possible to define this function to
3462 * perform the necessary adjustments.
3464 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3469 static int ftrace_match(char *str, struct ftrace_glob *g)
3474 str = arch_ftrace_match_adjust(str, g->search);
3478 if (strcmp(str, g->search) == 0)
3481 case MATCH_FRONT_ONLY:
3482 if (strncmp(str, g->search, g->len) == 0)
3485 case MATCH_MIDDLE_ONLY:
3486 if (strstr(str, g->search))
3489 case MATCH_END_ONLY:
3491 if (slen >= g->len &&
3492 memcmp(str + slen - g->len, g->search, g->len) == 0)
3501 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3503 struct ftrace_func_entry *entry;
3506 entry = ftrace_lookup_ip(hash, rec->ip);
3508 /* Do nothing if it doesn't exist */
3512 free_hash_entry(hash, entry);
3514 /* Do nothing if it exists */
3518 ret = add_hash_entry(hash, rec->ip);
3524 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3525 struct ftrace_glob *mod_g, int exclude_mod)
3527 char str[KSYM_SYMBOL_LEN];
3530 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3533 int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3535 /* blank module name to match all modules */
3537 /* blank module globbing: modname xor exclude_mod */
3538 if ((!exclude_mod) != (!modname))
3543 /* not matching the module */
3544 if (!modname || !mod_matches) {
3551 if (mod_matches && exclude_mod)
3555 /* blank search means to match all funcs in the mod */
3560 return ftrace_match(str, func_g);
3564 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3566 struct ftrace_page *pg;
3567 struct dyn_ftrace *rec;
3568 struct ftrace_glob func_g = { .type = MATCH_FULL };
3569 struct ftrace_glob mod_g = { .type = MATCH_FULL };
3570 struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3571 int exclude_mod = 0;
3577 func_g.type = filter_parse_regex(func, len, &func_g.search,
3579 func_g.len = strlen(func_g.search);
3583 mod_g.type = filter_parse_regex(mod, strlen(mod),
3584 &mod_g.search, &exclude_mod);
3585 mod_g.len = strlen(mod_g.search);
3588 mutex_lock(&ftrace_lock);
3590 if (unlikely(ftrace_disabled))
3593 do_for_each_ftrace_rec(pg, rec) {
3594 if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3595 ret = enter_record(hash, rec, clear_filter);
3602 } while_for_each_ftrace_rec();
3604 mutex_unlock(&ftrace_lock);
3610 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3612 return match_records(hash, buff, len, NULL);
3617 * We register the module command as a template to show others how
3618 * to register the a command as well.
3622 ftrace_mod_callback(struct ftrace_hash *hash,
3623 char *func, char *cmd, char *module, int enable)
3628 * cmd == 'mod' because we only registered this func
3629 * for the 'mod' ftrace_func_command.
3630 * But if you register one func with multiple commands,
3631 * you can tell which command was used by the cmd
3634 ret = match_records(hash, func, strlen(func), module);
3642 static struct ftrace_func_command ftrace_mod_cmd = {
3644 .func = ftrace_mod_callback,
3647 static int __init ftrace_mod_cmd_init(void)
3649 return register_ftrace_command(&ftrace_mod_cmd);
3651 core_initcall(ftrace_mod_cmd_init);
3653 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
3654 struct ftrace_ops *op, struct pt_regs *pt_regs)
3656 struct ftrace_func_probe *entry;
3657 struct hlist_head *hhd;
3660 key = hash_long(ip, FTRACE_HASH_BITS);
3662 hhd = &ftrace_func_hash[key];
3664 if (hlist_empty(hhd))
3668 * Disable preemption for these calls to prevent a RCU grace
3669 * period. This syncs the hash iteration and freeing of items
3670 * on the hash. rcu_read_lock is too dangerous here.
3672 preempt_disable_notrace();
3673 hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3674 if (entry->ip == ip)
3675 entry->ops->func(ip, parent_ip, &entry->data);
3677 preempt_enable_notrace();
3680 static struct ftrace_ops trace_probe_ops __read_mostly =
3682 .func = function_trace_probe_call,
3683 .flags = FTRACE_OPS_FL_INITIALIZED,
3684 INIT_OPS_HASH(trace_probe_ops)
3687 static int ftrace_probe_registered;
3689 static void __enable_ftrace_function_probe(struct ftrace_ops_hash *old_hash)
3694 if (ftrace_probe_registered) {
3695 /* still need to update the function call sites */
3697 ftrace_run_modify_code(&trace_probe_ops, FTRACE_UPDATE_CALLS,
3702 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3703 struct hlist_head *hhd = &ftrace_func_hash[i];
3707 /* Nothing registered? */
3708 if (i == FTRACE_FUNC_HASHSIZE)
3711 ret = ftrace_startup(&trace_probe_ops, 0);
3713 ftrace_probe_registered = 1;
3716 static void __disable_ftrace_function_probe(void)
3720 if (!ftrace_probe_registered)
3723 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3724 struct hlist_head *hhd = &ftrace_func_hash[i];
3729 /* no more funcs left */
3730 ftrace_shutdown(&trace_probe_ops, 0);
3732 ftrace_probe_registered = 0;
3736 static void ftrace_free_entry(struct ftrace_func_probe *entry)
3738 if (entry->ops->free)
3739 entry->ops->free(entry->ops, entry->ip, &entry->data);
3744 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3747 struct ftrace_ops_hash old_hash_ops;
3748 struct ftrace_func_probe *entry;
3749 struct ftrace_glob func_g;
3750 struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3751 struct ftrace_hash *old_hash = *orig_hash;
3752 struct ftrace_hash *hash;
3753 struct ftrace_page *pg;
3754 struct dyn_ftrace *rec;
3760 func_g.type = filter_parse_regex(glob, strlen(glob),
3761 &func_g.search, ¬);
3762 func_g.len = strlen(func_g.search);
3764 /* we do not support '!' for function probes */
3768 mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3770 old_hash_ops.filter_hash = old_hash;
3771 /* Probes only have filters */
3772 old_hash_ops.notrace_hash = NULL;
3774 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
3780 if (unlikely(ftrace_disabled)) {
3785 mutex_lock(&ftrace_lock);
3787 do_for_each_ftrace_rec(pg, rec) {
3789 if (!ftrace_match_record(rec, &func_g, NULL, 0))
3792 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3794 /* If we did not process any, then return error */
3805 * The caller might want to do something special
3806 * for each function we find. We call the callback
3807 * to give the caller an opportunity to do so.
3810 if (ops->init(ops, rec->ip, &entry->data) < 0) {
3811 /* caller does not like this func */
3817 ret = enter_record(hash, rec, 0);
3825 entry->ip = rec->ip;
3827 key = hash_long(entry->ip, FTRACE_HASH_BITS);
3828 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3830 } while_for_each_ftrace_rec();
3832 ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3834 __enable_ftrace_function_probe(&old_hash_ops);
3837 free_ftrace_hash_rcu(old_hash);
3842 mutex_unlock(&ftrace_lock);
3844 mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3845 free_ftrace_hash(hash);
3851 PROBE_TEST_FUNC = 1,
3856 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3857 void *data, int flags)
3859 struct ftrace_func_entry *rec_entry;
3860 struct ftrace_func_probe *entry;
3861 struct ftrace_func_probe *p;
3862 struct ftrace_glob func_g;
3863 struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3864 struct ftrace_hash *old_hash = *orig_hash;
3865 struct list_head free_list;
3866 struct ftrace_hash *hash;
3867 struct hlist_node *tmp;
3868 char str[KSYM_SYMBOL_LEN];
3871 if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3872 func_g.search = NULL;
3876 func_g.type = filter_parse_regex(glob, strlen(glob),
3877 &func_g.search, ¬);
3878 func_g.len = strlen(func_g.search);
3879 func_g.search = glob;
3881 /* we do not support '!' for function probes */
3886 mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3888 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3890 /* Hmm, should report this somehow */
3893 INIT_LIST_HEAD(&free_list);
3895 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3896 struct hlist_head *hhd = &ftrace_func_hash[i];
3898 hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3900 /* break up if statements for readability */
3901 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3904 if ((flags & PROBE_TEST_DATA) && entry->data != data)
3907 /* do this last, since it is the most expensive */
3908 if (func_g.search) {
3909 kallsyms_lookup(entry->ip, NULL, NULL,
3911 if (!ftrace_match(str, &func_g))
3915 rec_entry = ftrace_lookup_ip(hash, entry->ip);
3916 /* It is possible more than one entry had this ip */
3918 free_hash_entry(hash, rec_entry);
3920 hlist_del_rcu(&entry->node);
3921 list_add(&entry->free_list, &free_list);
3924 mutex_lock(&ftrace_lock);
3925 __disable_ftrace_function_probe();
3927 * Remove after the disable is called. Otherwise, if the last
3928 * probe is removed, a null hash means *all enabled*.
3930 ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3931 synchronize_sched();
3933 free_ftrace_hash_rcu(old_hash);
3935 list_for_each_entry_safe(entry, p, &free_list, free_list) {
3936 list_del(&entry->free_list);
3937 ftrace_free_entry(entry);
3939 mutex_unlock(&ftrace_lock);
3942 mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3943 free_ftrace_hash(hash);
3947 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3950 __unregister_ftrace_function_probe(glob, ops, data,
3951 PROBE_TEST_FUNC | PROBE_TEST_DATA);
3955 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3957 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3960 void unregister_ftrace_function_probe_all(char *glob)
3962 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3965 static LIST_HEAD(ftrace_commands);
3966 static DEFINE_MUTEX(ftrace_cmd_mutex);
3969 * Currently we only register ftrace commands from __init, so mark this
3972 __init int register_ftrace_command(struct ftrace_func_command *cmd)
3974 struct ftrace_func_command *p;
3977 mutex_lock(&ftrace_cmd_mutex);
3978 list_for_each_entry(p, &ftrace_commands, list) {
3979 if (strcmp(cmd->name, p->name) == 0) {
3984 list_add(&cmd->list, &ftrace_commands);
3986 mutex_unlock(&ftrace_cmd_mutex);
3992 * Currently we only unregister ftrace commands from __init, so mark
3995 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
3997 struct ftrace_func_command *p, *n;
4000 mutex_lock(&ftrace_cmd_mutex);
4001 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4002 if (strcmp(cmd->name, p->name) == 0) {
4004 list_del_init(&p->list);
4009 mutex_unlock(&ftrace_cmd_mutex);
4014 static int ftrace_process_regex(struct ftrace_hash *hash,
4015 char *buff, int len, int enable)
4017 char *func, *command, *next = buff;
4018 struct ftrace_func_command *p;
4021 func = strsep(&next, ":");
4024 ret = ftrace_match_records(hash, func, len);
4034 command = strsep(&next, ":");
4036 mutex_lock(&ftrace_cmd_mutex);
4037 list_for_each_entry(p, &ftrace_commands, list) {
4038 if (strcmp(p->name, command) == 0) {
4039 ret = p->func(hash, func, command, next, enable);
4044 mutex_unlock(&ftrace_cmd_mutex);
4050 ftrace_regex_write(struct file *file, const char __user *ubuf,
4051 size_t cnt, loff_t *ppos, int enable)
4053 struct ftrace_iterator *iter;
4054 struct trace_parser *parser;
4060 if (file->f_mode & FMODE_READ) {
4061 struct seq_file *m = file->private_data;
4064 iter = file->private_data;
4066 if (unlikely(ftrace_disabled))
4069 /* iter->hash is a local copy, so we don't need regex_lock */
4071 parser = &iter->parser;
4072 read = trace_get_user(parser, ubuf, cnt, ppos);
4074 if (read >= 0 && trace_parser_loaded(parser) &&
4075 !trace_parser_cont(parser)) {
4076 ret = ftrace_process_regex(iter->hash, parser->buffer,
4077 parser->idx, enable);
4078 trace_parser_clear(parser);
4089 ftrace_filter_write(struct file *file, const char __user *ubuf,
4090 size_t cnt, loff_t *ppos)
4092 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4096 ftrace_notrace_write(struct file *file, const char __user *ubuf,
4097 size_t cnt, loff_t *ppos)
4099 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4103 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4105 struct ftrace_func_entry *entry;
4107 if (!ftrace_location(ip))
4111 entry = ftrace_lookup_ip(hash, ip);
4114 free_hash_entry(hash, entry);
4118 return add_hash_entry(hash, ip);
4121 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4122 struct ftrace_ops_hash *old_hash)
4124 struct ftrace_ops *op;
4126 if (!ftrace_enabled)
4129 if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4130 ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4135 * If this is the shared global_ops filter, then we need to
4136 * check if there is another ops that shares it, is enabled.
4137 * If so, we still need to run the modify code.
4139 if (ops->func_hash != &global_ops.local_hash)
4142 do_for_each_ftrace_op(op, ftrace_ops_list) {
4143 if (op->func_hash == &global_ops.local_hash &&
4144 op->flags & FTRACE_OPS_FL_ENABLED) {
4145 ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4146 /* Only need to do this once */
4149 } while_for_each_ftrace_op(op);
4153 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4154 unsigned long ip, int remove, int reset, int enable)
4156 struct ftrace_hash **orig_hash;
4157 struct ftrace_ops_hash old_hash_ops;
4158 struct ftrace_hash *old_hash;
4159 struct ftrace_hash *hash;
4162 if (unlikely(ftrace_disabled))
4165 mutex_lock(&ops->func_hash->regex_lock);
4168 orig_hash = &ops->func_hash->filter_hash;
4170 orig_hash = &ops->func_hash->notrace_hash;
4173 hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4175 hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4179 goto out_regex_unlock;
4182 if (buf && !ftrace_match_records(hash, buf, len)) {
4184 goto out_regex_unlock;
4187 ret = ftrace_match_addr(hash, ip, remove);
4189 goto out_regex_unlock;
4192 mutex_lock(&ftrace_lock);
4193 old_hash = *orig_hash;
4194 old_hash_ops.filter_hash = ops->func_hash->filter_hash;
4195 old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
4196 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
4198 ftrace_ops_update_code(ops, &old_hash_ops);
4199 free_ftrace_hash_rcu(old_hash);
4201 mutex_unlock(&ftrace_lock);
4204 mutex_unlock(&ops->func_hash->regex_lock);
4206 free_ftrace_hash(hash);
4211 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4212 int reset, int enable)
4214 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4218 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4219 * @ops - the ops to set the filter with
4220 * @ip - the address to add to or remove from the filter.
4221 * @remove - non zero to remove the ip from the filter
4222 * @reset - non zero to reset all filters before applying this filter.
4224 * Filters denote which functions should be enabled when tracing is enabled
4225 * If @ip is NULL, it failes to update filter.
4227 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4228 int remove, int reset)
4230 ftrace_ops_init(ops);
4231 return ftrace_set_addr(ops, ip, remove, reset, 1);
4233 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4236 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4237 int reset, int enable)
4239 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4243 * ftrace_set_filter - set a function to filter on in ftrace
4244 * @ops - the ops to set the filter with
4245 * @buf - the string that holds the function filter text.
4246 * @len - the length of the string.
4247 * @reset - non zero to reset all filters before applying this filter.
4249 * Filters denote which functions should be enabled when tracing is enabled.
4250 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4252 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4255 ftrace_ops_init(ops);
4256 return ftrace_set_regex(ops, buf, len, reset, 1);
4258 EXPORT_SYMBOL_GPL(ftrace_set_filter);
4261 * ftrace_set_notrace - set a function to not trace in ftrace
4262 * @ops - the ops to set the notrace filter with
4263 * @buf - the string that holds the function notrace text.
4264 * @len - the length of the string.
4265 * @reset - non zero to reset all filters before applying this filter.
4267 * Notrace Filters denote which functions should not be enabled when tracing
4268 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4271 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4274 ftrace_ops_init(ops);
4275 return ftrace_set_regex(ops, buf, len, reset, 0);
4277 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4279 * ftrace_set_global_filter - set a function to filter on with global tracers
4280 * @buf - the string that holds the function filter text.
4281 * @len - the length of the string.
4282 * @reset - non zero to reset all filters before applying this filter.
4284 * Filters denote which functions should be enabled when tracing is enabled.
4285 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4287 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4289 ftrace_set_regex(&global_ops, buf, len, reset, 1);
4291 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4294 * ftrace_set_global_notrace - set a function to not trace with global tracers
4295 * @buf - the string that holds the function notrace text.
4296 * @len - the length of the string.
4297 * @reset - non zero to reset all filters before applying this filter.
4299 * Notrace Filters denote which functions should not be enabled when tracing
4300 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4303 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4305 ftrace_set_regex(&global_ops, buf, len, reset, 0);
4307 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4310 * command line interface to allow users to set filters on boot up.
4312 #define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
4313 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4314 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4316 /* Used by function selftest to not test if filter is set */
4317 bool ftrace_filter_param __initdata;
4319 static int __init set_ftrace_notrace(char *str)
4321 ftrace_filter_param = true;
4322 strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4325 __setup("ftrace_notrace=", set_ftrace_notrace);
4327 static int __init set_ftrace_filter(char *str)
4329 ftrace_filter_param = true;
4330 strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4333 __setup("ftrace_filter=", set_ftrace_filter);
4335 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4336 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4337 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4338 static int ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer);
4340 static unsigned long save_global_trampoline;
4341 static unsigned long save_global_flags;
4343 static int __init set_graph_function(char *str)
4345 strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4348 __setup("ftrace_graph_filter=", set_graph_function);
4350 static int __init set_graph_notrace_function(char *str)
4352 strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4355 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
4357 static void __init set_ftrace_early_graph(char *buf, int enable)
4361 unsigned long *table = ftrace_graph_funcs;
4362 int *count = &ftrace_graph_count;
4365 table = ftrace_graph_notrace_funcs;
4366 count = &ftrace_graph_notrace_count;
4370 func = strsep(&buf, ",");
4371 /* we allow only one expression at a time */
4372 ret = ftrace_set_func(table, count, FTRACE_GRAPH_MAX_FUNCS, func);
4374 printk(KERN_DEBUG "ftrace: function %s not "
4375 "traceable\n", func);
4378 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4381 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
4385 ftrace_ops_init(ops);
4388 func = strsep(&buf, ",");
4389 ftrace_set_regex(ops, func, strlen(func), 0, enable);
4393 static void __init set_ftrace_early_filters(void)
4395 if (ftrace_filter_buf[0])
4396 ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
4397 if (ftrace_notrace_buf[0])
4398 ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
4399 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4400 if (ftrace_graph_buf[0])
4401 set_ftrace_early_graph(ftrace_graph_buf, 1);
4402 if (ftrace_graph_notrace_buf[0])
4403 set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
4404 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4407 int ftrace_regex_release(struct inode *inode, struct file *file)
4409 struct seq_file *m = (struct seq_file *)file->private_data;
4410 struct ftrace_ops_hash old_hash_ops;
4411 struct ftrace_iterator *iter;
4412 struct ftrace_hash **orig_hash;
4413 struct ftrace_hash *old_hash;
4414 struct trace_parser *parser;
4418 if (file->f_mode & FMODE_READ) {
4420 seq_release(inode, file);
4422 iter = file->private_data;
4424 parser = &iter->parser;
4425 if (trace_parser_loaded(parser)) {
4426 parser->buffer[parser->idx] = 0;
4427 ftrace_match_records(iter->hash, parser->buffer, parser->idx);
4430 trace_parser_put(parser);
4432 mutex_lock(&iter->ops->func_hash->regex_lock);
4434 if (file->f_mode & FMODE_WRITE) {
4435 filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
4438 orig_hash = &iter->ops->func_hash->filter_hash;
4440 orig_hash = &iter->ops->func_hash->notrace_hash;
4442 mutex_lock(&ftrace_lock);
4443 old_hash = *orig_hash;
4444 old_hash_ops.filter_hash = iter->ops->func_hash->filter_hash;
4445 old_hash_ops.notrace_hash = iter->ops->func_hash->notrace_hash;
4446 ret = ftrace_hash_move(iter->ops, filter_hash,
4447 orig_hash, iter->hash);
4449 ftrace_ops_update_code(iter->ops, &old_hash_ops);
4450 free_ftrace_hash_rcu(old_hash);
4452 mutex_unlock(&ftrace_lock);
4455 mutex_unlock(&iter->ops->func_hash->regex_lock);
4456 free_ftrace_hash(iter->hash);
4462 static const struct file_operations ftrace_avail_fops = {
4463 .open = ftrace_avail_open,
4465 .llseek = seq_lseek,
4466 .release = seq_release_private,
4469 static const struct file_operations ftrace_enabled_fops = {
4470 .open = ftrace_enabled_open,
4472 .llseek = seq_lseek,
4473 .release = seq_release_private,
4476 static const struct file_operations ftrace_filter_fops = {
4477 .open = ftrace_filter_open,
4479 .write = ftrace_filter_write,
4480 .llseek = tracing_lseek,
4481 .release = ftrace_regex_release,
4484 static const struct file_operations ftrace_notrace_fops = {
4485 .open = ftrace_notrace_open,
4487 .write = ftrace_notrace_write,
4488 .llseek = tracing_lseek,
4489 .release = ftrace_regex_release,
4492 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4494 static DEFINE_MUTEX(graph_lock);
4496 int ftrace_graph_count;
4497 int ftrace_graph_notrace_count;
4498 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4499 unsigned long ftrace_graph_notrace_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4501 struct ftrace_graph_data {
4502 unsigned long *table;
4505 const struct seq_operations *seq_ops;
4509 __g_next(struct seq_file *m, loff_t *pos)
4511 struct ftrace_graph_data *fgd = m->private;
4513 if (*pos >= *fgd->count)
4515 return &fgd->table[*pos];
4519 g_next(struct seq_file *m, void *v, loff_t *pos)
4522 return __g_next(m, pos);
4525 static void *g_start(struct seq_file *m, loff_t *pos)
4527 struct ftrace_graph_data *fgd = m->private;
4529 mutex_lock(&graph_lock);
4531 /* Nothing, tell g_show to print all functions are enabled */
4532 if (!*fgd->count && !*pos)
4535 return __g_next(m, pos);
4538 static void g_stop(struct seq_file *m, void *p)
4540 mutex_unlock(&graph_lock);
4543 static int g_show(struct seq_file *m, void *v)
4545 unsigned long *ptr = v;
4550 if (ptr == (unsigned long *)1) {
4551 struct ftrace_graph_data *fgd = m->private;
4553 if (fgd->table == ftrace_graph_funcs)
4554 seq_puts(m, "#### all functions enabled ####\n");
4556 seq_puts(m, "#### no functions disabled ####\n");
4560 seq_printf(m, "%ps\n", (void *)*ptr);
4565 static const struct seq_operations ftrace_graph_seq_ops = {
4573 __ftrace_graph_open(struct inode *inode, struct file *file,
4574 struct ftrace_graph_data *fgd)
4578 mutex_lock(&graph_lock);
4579 if ((file->f_mode & FMODE_WRITE) &&
4580 (file->f_flags & O_TRUNC)) {
4582 memset(fgd->table, 0, fgd->size * sizeof(*fgd->table));
4584 mutex_unlock(&graph_lock);
4586 if (file->f_mode & FMODE_READ) {
4587 ret = seq_open(file, fgd->seq_ops);
4589 struct seq_file *m = file->private_data;
4593 file->private_data = fgd;
4599 ftrace_graph_open(struct inode *inode, struct file *file)
4601 struct ftrace_graph_data *fgd;
4603 if (unlikely(ftrace_disabled))
4606 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4610 fgd->table = ftrace_graph_funcs;
4611 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4612 fgd->count = &ftrace_graph_count;
4613 fgd->seq_ops = &ftrace_graph_seq_ops;
4615 return __ftrace_graph_open(inode, file, fgd);
4619 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
4621 struct ftrace_graph_data *fgd;
4623 if (unlikely(ftrace_disabled))
4626 fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4630 fgd->table = ftrace_graph_notrace_funcs;
4631 fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4632 fgd->count = &ftrace_graph_notrace_count;
4633 fgd->seq_ops = &ftrace_graph_seq_ops;
4635 return __ftrace_graph_open(inode, file, fgd);
4639 ftrace_graph_release(struct inode *inode, struct file *file)
4641 if (file->f_mode & FMODE_READ) {
4642 struct seq_file *m = file->private_data;
4645 seq_release(inode, file);
4647 kfree(file->private_data);
4654 ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer)
4656 struct ftrace_glob func_g;
4657 struct dyn_ftrace *rec;
4658 struct ftrace_page *pg;
4665 func_g.type = filter_parse_regex(buffer, strlen(buffer),
4666 &func_g.search, ¬);
4667 if (!not && *idx >= size)
4670 func_g.len = strlen(func_g.search);
4672 mutex_lock(&ftrace_lock);
4674 if (unlikely(ftrace_disabled)) {
4675 mutex_unlock(&ftrace_lock);
4679 do_for_each_ftrace_rec(pg, rec) {
4681 if (ftrace_match_record(rec, &func_g, NULL, 0)) {
4682 /* if it is in the array */
4684 for (i = 0; i < *idx; i++) {
4685 if (array[i] == rec->ip) {
4694 array[(*idx)++] = rec->ip;
4700 array[i] = array[--(*idx)];
4706 } while_for_each_ftrace_rec();
4708 mutex_unlock(&ftrace_lock);
4717 ftrace_graph_write(struct file *file, const char __user *ubuf,
4718 size_t cnt, loff_t *ppos)
4720 struct trace_parser parser;
4721 ssize_t read, ret = 0;
4722 struct ftrace_graph_data *fgd = file->private_data;
4727 if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX))
4730 read = trace_get_user(&parser, ubuf, cnt, ppos);
4732 if (read >= 0 && trace_parser_loaded((&parser))) {
4733 parser.buffer[parser.idx] = 0;
4735 mutex_lock(&graph_lock);
4737 /* we allow only one expression at a time */
4738 ret = ftrace_set_func(fgd->table, fgd->count, fgd->size,
4741 mutex_unlock(&graph_lock);
4747 trace_parser_put(&parser);
4752 static const struct file_operations ftrace_graph_fops = {
4753 .open = ftrace_graph_open,
4755 .write = ftrace_graph_write,
4756 .llseek = tracing_lseek,
4757 .release = ftrace_graph_release,
4760 static const struct file_operations ftrace_graph_notrace_fops = {
4761 .open = ftrace_graph_notrace_open,
4763 .write = ftrace_graph_write,
4764 .llseek = tracing_lseek,
4765 .release = ftrace_graph_release,
4767 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4769 void ftrace_create_filter_files(struct ftrace_ops *ops,
4770 struct dentry *parent)
4773 trace_create_file("set_ftrace_filter", 0644, parent,
4774 ops, &ftrace_filter_fops);
4776 trace_create_file("set_ftrace_notrace", 0644, parent,
4777 ops, &ftrace_notrace_fops);
4781 * The name "destroy_filter_files" is really a misnomer. Although
4782 * in the future, it may actualy delete the files, but this is
4783 * really intended to make sure the ops passed in are disabled
4784 * and that when this function returns, the caller is free to
4787 * The "destroy" name is only to match the "create" name that this
4788 * should be paired with.
4790 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
4792 mutex_lock(&ftrace_lock);
4793 if (ops->flags & FTRACE_OPS_FL_ENABLED)
4794 ftrace_shutdown(ops, 0);
4795 ops->flags |= FTRACE_OPS_FL_DELETED;
4796 mutex_unlock(&ftrace_lock);
4799 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
4802 trace_create_file("available_filter_functions", 0444,
4803 d_tracer, NULL, &ftrace_avail_fops);
4805 trace_create_file("enabled_functions", 0444,
4806 d_tracer, NULL, &ftrace_enabled_fops);
4808 ftrace_create_filter_files(&global_ops, d_tracer);
4810 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4811 trace_create_file("set_graph_function", 0444, d_tracer,
4813 &ftrace_graph_fops);
4814 trace_create_file("set_graph_notrace", 0444, d_tracer,
4816 &ftrace_graph_notrace_fops);
4817 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4822 static int ftrace_cmp_ips(const void *a, const void *b)
4824 const unsigned long *ipa = a;
4825 const unsigned long *ipb = b;
4834 static int ftrace_process_locs(struct module *mod,
4835 unsigned long *start,
4838 struct ftrace_page *start_pg;
4839 struct ftrace_page *pg;
4840 struct dyn_ftrace *rec;
4841 unsigned long count;
4844 unsigned long flags = 0; /* Shut up gcc */
4847 count = end - start;
4852 sort(start, count, sizeof(*start),
4853 ftrace_cmp_ips, NULL);
4855 start_pg = ftrace_allocate_pages(count);
4859 mutex_lock(&ftrace_lock);
4862 * Core and each module needs their own pages, as
4863 * modules will free them when they are removed.
4864 * Force a new page to be allocated for modules.
4867 WARN_ON(ftrace_pages || ftrace_pages_start);
4868 /* First initialization */
4869 ftrace_pages = ftrace_pages_start = start_pg;
4874 if (WARN_ON(ftrace_pages->next)) {
4875 /* Hmm, we have free pages? */
4876 while (ftrace_pages->next)
4877 ftrace_pages = ftrace_pages->next;
4880 ftrace_pages->next = start_pg;
4886 addr = ftrace_call_adjust(*p++);
4888 * Some architecture linkers will pad between
4889 * the different mcount_loc sections of different
4890 * object files to satisfy alignments.
4891 * Skip any NULL pointers.
4896 if (pg->index == pg->size) {
4897 /* We should have allocated enough */
4898 if (WARN_ON(!pg->next))
4903 rec = &pg->records[pg->index++];
4907 /* We should have used all pages */
4910 /* Assign the last page to ftrace_pages */
4914 * We only need to disable interrupts on start up
4915 * because we are modifying code that an interrupt
4916 * may execute, and the modification is not atomic.
4917 * But for modules, nothing runs the code we modify
4918 * until we are finished with it, and there's no
4919 * reason to cause large interrupt latencies while we do it.
4922 local_irq_save(flags);
4923 ftrace_update_code(mod, start_pg);
4925 local_irq_restore(flags);
4928 mutex_unlock(&ftrace_lock);
4933 #ifdef CONFIG_MODULES
4935 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4937 static int referenced_filters(struct dyn_ftrace *rec)
4939 struct ftrace_ops *ops;
4942 for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
4943 if (ops_references_rec(ops, rec))
4950 void ftrace_release_mod(struct module *mod)
4952 struct dyn_ftrace *rec;
4953 struct ftrace_page **last_pg;
4954 struct ftrace_page *pg;
4957 mutex_lock(&ftrace_lock);
4959 if (ftrace_disabled)
4963 * Each module has its own ftrace_pages, remove
4964 * them from the list.
4966 last_pg = &ftrace_pages_start;
4967 for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4968 rec = &pg->records[0];
4969 if (within_module_core(rec->ip, mod)) {
4971 * As core pages are first, the first
4972 * page should never be a module page.
4974 if (WARN_ON(pg == ftrace_pages_start))
4977 /* Check if we are deleting the last page */
4978 if (pg == ftrace_pages)
4979 ftrace_pages = next_to_ftrace_page(last_pg);
4981 *last_pg = pg->next;
4982 order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4983 free_pages((unsigned long)pg->records, order);
4986 last_pg = &pg->next;
4989 mutex_unlock(&ftrace_lock);
4992 void ftrace_module_enable(struct module *mod)
4994 struct dyn_ftrace *rec;
4995 struct ftrace_page *pg;
4997 mutex_lock(&ftrace_lock);
4999 if (ftrace_disabled)
5003 * If the tracing is enabled, go ahead and enable the record.
5005 * The reason not to enable the record immediatelly is the
5006 * inherent check of ftrace_make_nop/ftrace_make_call for
5007 * correct previous instructions. Making first the NOP
5008 * conversion puts the module to the correct state, thus
5009 * passing the ftrace_make_call check.
5011 * We also delay this to after the module code already set the
5012 * text to read-only, as we now need to set it back to read-write
5013 * so that we can modify the text.
5015 if (ftrace_start_up)
5016 ftrace_arch_code_modify_prepare();
5018 do_for_each_ftrace_rec(pg, rec) {
5021 * do_for_each_ftrace_rec() is a double loop.
5022 * module text shares the pg. If a record is
5023 * not part of this module, then skip this pg,
5024 * which the "break" will do.
5026 if (!within_module_core(rec->ip, mod))
5032 * When adding a module, we need to check if tracers are
5033 * currently enabled and if they are, and can trace this record,
5034 * we need to enable the module functions as well as update the
5035 * reference counts for those function records.
5037 if (ftrace_start_up)
5038 cnt += referenced_filters(rec);
5040 /* This clears FTRACE_FL_DISABLED */
5043 if (ftrace_start_up && cnt) {
5044 int failed = __ftrace_replace_code(rec, 1);
5046 ftrace_bug(failed, rec);
5051 } while_for_each_ftrace_rec();
5054 if (ftrace_start_up)
5055 ftrace_arch_code_modify_post_process();
5058 mutex_unlock(&ftrace_lock);
5061 void ftrace_module_init(struct module *mod)
5063 if (ftrace_disabled || !mod->num_ftrace_callsites)
5066 ftrace_process_locs(mod, mod->ftrace_callsites,
5067 mod->ftrace_callsites + mod->num_ftrace_callsites);
5069 #endif /* CONFIG_MODULES */
5071 void __init ftrace_init(void)
5073 extern unsigned long __start_mcount_loc[];
5074 extern unsigned long __stop_mcount_loc[];
5075 unsigned long count, flags;
5078 local_irq_save(flags);
5079 ret = ftrace_dyn_arch_init();
5080 local_irq_restore(flags);
5084 count = __stop_mcount_loc - __start_mcount_loc;
5086 pr_info("ftrace: No functions to be traced?\n");
5090 pr_info("ftrace: allocating %ld entries in %ld pages\n",
5091 count, count / ENTRIES_PER_PAGE + 1);
5093 last_ftrace_enabled = ftrace_enabled = 1;
5095 ret = ftrace_process_locs(NULL,
5099 set_ftrace_early_filters();
5103 ftrace_disabled = 1;
5106 /* Do nothing if arch does not support this */
5107 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
5111 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5115 * Currently there's no safe way to free a trampoline when the kernel
5116 * is configured with PREEMPT. That is because a task could be preempted
5117 * when it jumped to the trampoline, it may be preempted for a long time
5118 * depending on the system load, and currently there's no way to know
5119 * when it will be off the trampoline. If the trampoline is freed
5120 * too early, when the task runs again, it will be executing on freed
5123 #ifdef CONFIG_PREEMPT
5124 /* Currently, only non dynamic ops can have a trampoline */
5125 if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
5129 arch_ftrace_update_trampoline(ops);
5134 static struct ftrace_ops global_ops = {
5135 .func = ftrace_stub,
5136 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
5137 FTRACE_OPS_FL_INITIALIZED |
5141 static int __init ftrace_nodyn_init(void)
5146 core_initcall(ftrace_nodyn_init);
5148 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
5149 static inline void ftrace_startup_enable(int command) { }
5150 static inline void ftrace_startup_all(int command) { }
5151 /* Keep as macros so we do not need to define the commands */
5152 # define ftrace_startup(ops, command) \
5154 int ___ret = __register_ftrace_function(ops); \
5156 (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
5159 # define ftrace_shutdown(ops, command) \
5161 int ___ret = __unregister_ftrace_function(ops); \
5163 (ops)->flags &= ~FTRACE_OPS_FL_ENABLED; \
5167 # define ftrace_startup_sysctl() do { } while (0)
5168 # define ftrace_shutdown_sysctl() do { } while (0)
5171 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
5176 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5180 #endif /* CONFIG_DYNAMIC_FTRACE */
5182 __init void ftrace_init_global_array_ops(struct trace_array *tr)
5184 tr->ops = &global_ops;
5185 tr->ops->private = tr;
5188 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
5190 /* If we filter on pids, update to use the pid function */
5191 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
5192 if (WARN_ON(tr->ops->func != ftrace_stub))
5193 printk("ftrace ops had %pS for function\n",
5196 tr->ops->func = func;
5197 tr->ops->private = tr;
5200 void ftrace_reset_array_ops(struct trace_array *tr)
5202 tr->ops->func = ftrace_stub;
5206 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5207 struct ftrace_ops *ignored, struct pt_regs *regs)
5209 struct ftrace_ops *op;
5212 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5217 * Some of the ops may be dynamically allocated,
5218 * they must be freed after a synchronize_sched().
5220 preempt_disable_notrace();
5222 do_for_each_ftrace_op(op, ftrace_ops_list) {
5224 * Check the following for each ops before calling their func:
5225 * if RCU flag is set, then rcu_is_watching() must be true
5226 * if PER_CPU is set, then ftrace_function_local_disable()
5228 * Otherwise test if the ip matches the ops filter
5230 * If any of the above fails then the op->func() is not executed.
5232 if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
5233 (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5234 !ftrace_function_local_disabled(op)) &&
5235 ftrace_ops_test(op, ip, regs)) {
5237 if (FTRACE_WARN_ON(!op->func)) {
5238 pr_warn("op=%p %pS\n", op, op);
5241 op->func(ip, parent_ip, op, regs);
5243 } while_for_each_ftrace_op(op);
5245 preempt_enable_notrace();
5246 trace_clear_recursion(bit);
5250 * Some archs only support passing ip and parent_ip. Even though
5251 * the list function ignores the op parameter, we do not want any
5252 * C side effects, where a function is called without the caller
5253 * sending a third parameter.
5254 * Archs are to support both the regs and ftrace_ops at the same time.
5255 * If they support ftrace_ops, it is assumed they support regs.
5256 * If call backs want to use regs, they must either check for regs
5257 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
5258 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
5259 * An architecture can pass partial regs with ftrace_ops and still
5260 * set the ARCH_SUPPORTS_FTRACE_OPS.
5262 #if ARCH_SUPPORTS_FTRACE_OPS
5263 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5264 struct ftrace_ops *op, struct pt_regs *regs)
5266 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
5269 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
5271 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
5276 * If there's only one function registered but it does not support
5277 * recursion, needs RCU protection and/or requires per cpu handling, then
5278 * this function will be called by the mcount trampoline.
5280 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
5281 struct ftrace_ops *op, struct pt_regs *regs)
5285 if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
5288 bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5292 preempt_disable_notrace();
5294 if (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5295 !ftrace_function_local_disabled(op)) {
5296 op->func(ip, parent_ip, op, regs);
5299 preempt_enable_notrace();
5300 trace_clear_recursion(bit);
5304 * ftrace_ops_get_func - get the function a trampoline should call
5305 * @ops: the ops to get the function for
5307 * Normally the mcount trampoline will call the ops->func, but there
5308 * are times that it should not. For example, if the ops does not
5309 * have its own recursion protection, then it should call the
5310 * ftrace_ops_recurs_func() instead.
5312 * Returns the function that the trampoline should call for @ops.
5314 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
5317 * If the function does not handle recursion, needs to be RCU safe,
5318 * or does per cpu logic, then we need to call the assist handler.
5320 if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
5321 ops->flags & (FTRACE_OPS_FL_RCU | FTRACE_OPS_FL_PER_CPU))
5322 return ftrace_ops_assist_func;
5327 static void clear_ftrace_swapper(void)
5329 struct task_struct *p;
5333 for_each_online_cpu(cpu) {
5335 clear_tsk_trace_trace(p);
5340 static void set_ftrace_swapper(void)
5342 struct task_struct *p;
5346 for_each_online_cpu(cpu) {
5348 set_tsk_trace_trace(p);
5353 static void clear_ftrace_pid(struct pid *pid)
5355 struct task_struct *p;
5358 do_each_pid_task(pid, PIDTYPE_PID, p) {
5359 clear_tsk_trace_trace(p);
5360 } while_each_pid_task(pid, PIDTYPE_PID, p);
5366 static void set_ftrace_pid(struct pid *pid)
5368 struct task_struct *p;
5371 do_each_pid_task(pid, PIDTYPE_PID, p) {
5372 set_tsk_trace_trace(p);
5373 } while_each_pid_task(pid, PIDTYPE_PID, p);
5377 static void clear_ftrace_pid_task(struct pid *pid)
5379 if (pid == ftrace_swapper_pid)
5380 clear_ftrace_swapper();
5382 clear_ftrace_pid(pid);
5385 static void set_ftrace_pid_task(struct pid *pid)
5387 if (pid == ftrace_swapper_pid)
5388 set_ftrace_swapper();
5390 set_ftrace_pid(pid);
5393 static int ftrace_pid_add(int p)
5396 struct ftrace_pid *fpid;
5399 mutex_lock(&ftrace_lock);
5402 pid = ftrace_swapper_pid;
5404 pid = find_get_pid(p);
5411 list_for_each_entry(fpid, &ftrace_pids, list)
5412 if (fpid->pid == pid)
5417 fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
5421 list_add(&fpid->list, &ftrace_pids);
5424 set_ftrace_pid_task(pid);
5426 ftrace_update_pid_func();
5428 ftrace_startup_all(0);
5430 mutex_unlock(&ftrace_lock);
5434 if (pid != ftrace_swapper_pid)
5438 mutex_unlock(&ftrace_lock);
5442 static void ftrace_pid_reset(void)
5444 struct ftrace_pid *fpid, *safe;
5446 mutex_lock(&ftrace_lock);
5447 list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
5448 struct pid *pid = fpid->pid;
5450 clear_ftrace_pid_task(pid);
5452 list_del(&fpid->list);
5456 ftrace_update_pid_func();
5457 ftrace_startup_all(0);
5459 mutex_unlock(&ftrace_lock);
5462 static void *fpid_start(struct seq_file *m, loff_t *pos)
5464 mutex_lock(&ftrace_lock);
5466 if (!ftrace_pids_enabled() && (!*pos))
5469 return seq_list_start(&ftrace_pids, *pos);
5472 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
5477 return seq_list_next(v, &ftrace_pids, pos);
5480 static void fpid_stop(struct seq_file *m, void *p)
5482 mutex_unlock(&ftrace_lock);
5485 static int fpid_show(struct seq_file *m, void *v)
5487 const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
5489 if (v == (void *)1) {
5490 seq_puts(m, "no pid\n");
5494 if (fpid->pid == ftrace_swapper_pid)
5495 seq_puts(m, "swapper tasks\n");
5497 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
5502 static const struct seq_operations ftrace_pid_sops = {
5503 .start = fpid_start,
5510 ftrace_pid_open(struct inode *inode, struct file *file)
5514 if ((file->f_mode & FMODE_WRITE) &&
5515 (file->f_flags & O_TRUNC))
5518 if (file->f_mode & FMODE_READ)
5519 ret = seq_open(file, &ftrace_pid_sops);
5525 ftrace_pid_write(struct file *filp, const char __user *ubuf,
5526 size_t cnt, loff_t *ppos)
5532 if (cnt >= sizeof(buf))
5535 if (copy_from_user(&buf, ubuf, cnt))
5541 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
5542 * to clean the filter quietly.
5544 tmp = strstrip(buf);
5545 if (strlen(tmp) == 0)
5548 ret = kstrtol(tmp, 10, &val);
5552 ret = ftrace_pid_add(val);
5554 return ret ? ret : cnt;
5558 ftrace_pid_release(struct inode *inode, struct file *file)
5560 if (file->f_mode & FMODE_READ)
5561 seq_release(inode, file);
5566 static const struct file_operations ftrace_pid_fops = {
5567 .open = ftrace_pid_open,
5568 .write = ftrace_pid_write,
5570 .llseek = tracing_lseek,
5571 .release = ftrace_pid_release,
5574 static __init int ftrace_init_tracefs(void)
5576 struct dentry *d_tracer;
5578 d_tracer = tracing_init_dentry();
5579 if (IS_ERR(d_tracer))
5582 ftrace_init_dyn_tracefs(d_tracer);
5584 trace_create_file("set_ftrace_pid", 0644, d_tracer,
5585 NULL, &ftrace_pid_fops);
5587 ftrace_profile_tracefs(d_tracer);
5591 fs_initcall(ftrace_init_tracefs);
5594 * ftrace_kill - kill ftrace
5596 * This function should be used by panic code. It stops ftrace
5597 * but in a not so nice way. If you need to simply kill ftrace
5598 * from a non-atomic section, use ftrace_kill.
5600 void ftrace_kill(void)
5602 ftrace_disabled = 1;
5604 clear_ftrace_function();
5608 * Test if ftrace is dead or not.
5610 int ftrace_is_dead(void)
5612 return ftrace_disabled;
5616 * register_ftrace_function - register a function for profiling
5617 * @ops - ops structure that holds the function for profiling.
5619 * Register a function to be called by all functions in the
5622 * Note: @ops->func and all the functions it calls must be labeled
5623 * with "notrace", otherwise it will go into a
5626 int register_ftrace_function(struct ftrace_ops *ops)
5630 ftrace_ops_init(ops);
5632 mutex_lock(&ftrace_lock);
5634 ret = ftrace_startup(ops, 0);
5636 mutex_unlock(&ftrace_lock);
5640 EXPORT_SYMBOL_GPL(register_ftrace_function);
5643 * unregister_ftrace_function - unregister a function for profiling.
5644 * @ops - ops structure that holds the function to unregister
5646 * Unregister a function that was added to be called by ftrace profiling.
5648 int unregister_ftrace_function(struct ftrace_ops *ops)
5652 mutex_lock(&ftrace_lock);
5653 ret = ftrace_shutdown(ops, 0);
5654 mutex_unlock(&ftrace_lock);
5658 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
5661 ftrace_enable_sysctl(struct ctl_table *table, int write,
5662 void __user *buffer, size_t *lenp,
5667 mutex_lock(&ftrace_lock);
5669 if (unlikely(ftrace_disabled))
5672 ret = proc_dointvec(table, write, buffer, lenp, ppos);
5674 if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
5677 last_ftrace_enabled = !!ftrace_enabled;
5679 if (ftrace_enabled) {
5681 /* we are starting ftrace again */
5682 if (ftrace_ops_list != &ftrace_list_end)
5683 update_ftrace_function();
5685 ftrace_startup_sysctl();
5688 /* stopping ftrace calls (just send to ftrace_stub) */
5689 ftrace_trace_function = ftrace_stub;
5691 ftrace_shutdown_sysctl();
5695 mutex_unlock(&ftrace_lock);
5699 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5701 static struct ftrace_ops graph_ops = {
5702 .func = ftrace_stub,
5703 .flags = FTRACE_OPS_FL_RECURSION_SAFE |
5704 FTRACE_OPS_FL_INITIALIZED |
5707 #ifdef FTRACE_GRAPH_TRAMP_ADDR
5708 .trampoline = FTRACE_GRAPH_TRAMP_ADDR,
5709 /* trampoline_size is only needed for dynamically allocated tramps */
5711 ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
5714 void ftrace_graph_sleep_time_control(bool enable)
5716 fgraph_sleep_time = enable;
5719 void ftrace_graph_graph_time_control(bool enable)
5721 fgraph_graph_time = enable;
5724 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
5729 /* The callbacks that hook a function */
5730 trace_func_graph_ret_t ftrace_graph_return =
5731 (trace_func_graph_ret_t)ftrace_stub;
5732 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
5733 static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
5735 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
5736 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
5740 unsigned long flags;
5741 int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
5742 struct task_struct *g, *t;
5744 for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
5745 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
5746 * sizeof(struct ftrace_ret_stack),
5748 if (!ret_stack_list[i]) {
5756 read_lock_irqsave(&tasklist_lock, flags);
5757 do_each_thread(g, t) {
5763 if (t->ret_stack == NULL) {
5764 atomic_set(&t->tracing_graph_pause, 0);
5765 atomic_set(&t->trace_overrun, 0);
5766 t->curr_ret_stack = -1;
5767 /* Make sure the tasks see the -1 first: */
5769 t->ret_stack = ret_stack_list[start++];
5771 } while_each_thread(g, t);
5774 read_unlock_irqrestore(&tasklist_lock, flags);
5776 for (i = start; i < end; i++)
5777 kfree(ret_stack_list[i]);
5782 ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
5783 struct task_struct *prev, struct task_struct *next)
5785 unsigned long long timestamp;
5789 * Does the user want to count the time a function was asleep.
5790 * If so, do not update the time stamps.
5792 if (fgraph_sleep_time)
5795 timestamp = trace_clock_local();
5797 prev->ftrace_timestamp = timestamp;
5799 /* only process tasks that we timestamped */
5800 if (!next->ftrace_timestamp)
5804 * Update all the counters in next to make up for the
5805 * time next was sleeping.
5807 timestamp -= next->ftrace_timestamp;
5809 for (index = next->curr_ret_stack; index >= 0; index--)
5810 next->ret_stack[index].calltime += timestamp;
5813 /* Allocate a return stack for each task */
5814 static int start_graph_tracing(void)
5816 struct ftrace_ret_stack **ret_stack_list;
5819 ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
5820 sizeof(struct ftrace_ret_stack *),
5823 if (!ret_stack_list)
5826 /* The cpu_boot init_task->ret_stack will never be freed */
5827 for_each_online_cpu(cpu) {
5828 if (!idle_task(cpu)->ret_stack)
5829 ftrace_graph_init_idle_task(idle_task(cpu), cpu);
5833 ret = alloc_retstack_tasklist(ret_stack_list);
5834 } while (ret == -EAGAIN);
5837 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5839 pr_info("ftrace_graph: Couldn't activate tracepoint"
5840 " probe to kernel_sched_switch\n");
5843 kfree(ret_stack_list);
5848 * Hibernation protection.
5849 * The state of the current task is too much unstable during
5850 * suspend/restore to disk. We want to protect against that.
5853 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
5857 case PM_HIBERNATION_PREPARE:
5858 pause_graph_tracing();
5861 case PM_POST_HIBERNATION:
5862 unpause_graph_tracing();
5868 static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
5870 if (!ftrace_ops_test(&global_ops, trace->func, NULL))
5872 return __ftrace_graph_entry(trace);
5876 * The function graph tracer should only trace the functions defined
5877 * by set_ftrace_filter and set_ftrace_notrace. If another function
5878 * tracer ops is registered, the graph tracer requires testing the
5879 * function against the global ops, and not just trace any function
5880 * that any ftrace_ops registered.
5882 static void update_function_graph_func(void)
5884 struct ftrace_ops *op;
5885 bool do_test = false;
5888 * The graph and global ops share the same set of functions
5889 * to test. If any other ops is on the list, then
5890 * the graph tracing needs to test if its the function
5893 do_for_each_ftrace_op(op, ftrace_ops_list) {
5894 if (op != &global_ops && op != &graph_ops &&
5895 op != &ftrace_list_end) {
5897 /* in double loop, break out with goto */
5900 } while_for_each_ftrace_op(op);
5903 ftrace_graph_entry = ftrace_graph_entry_test;
5905 ftrace_graph_entry = __ftrace_graph_entry;
5908 static struct notifier_block ftrace_suspend_notifier = {
5909 .notifier_call = ftrace_suspend_notifier_call,
5912 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
5913 trace_func_graph_ent_t entryfunc)
5917 mutex_lock(&ftrace_lock);
5919 /* we currently allow only one tracer registered at a time */
5920 if (ftrace_graph_active) {
5925 register_pm_notifier(&ftrace_suspend_notifier);
5927 ftrace_graph_active++;
5928 ret = start_graph_tracing();
5930 ftrace_graph_active--;
5934 ftrace_graph_return = retfunc;
5937 * Update the indirect function to the entryfunc, and the
5938 * function that gets called to the entry_test first. Then
5939 * call the update fgraph entry function to determine if
5940 * the entryfunc should be called directly or not.
5942 __ftrace_graph_entry = entryfunc;
5943 ftrace_graph_entry = ftrace_graph_entry_test;
5944 update_function_graph_func();
5946 ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
5948 mutex_unlock(&ftrace_lock);
5952 void unregister_ftrace_graph(void)
5954 mutex_lock(&ftrace_lock);
5956 if (unlikely(!ftrace_graph_active))
5959 ftrace_graph_active--;
5960 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
5961 ftrace_graph_entry = ftrace_graph_entry_stub;
5962 __ftrace_graph_entry = ftrace_graph_entry_stub;
5963 ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
5964 unregister_pm_notifier(&ftrace_suspend_notifier);
5965 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5967 #ifdef CONFIG_DYNAMIC_FTRACE
5969 * Function graph does not allocate the trampoline, but
5970 * other global_ops do. We need to reset the ALLOC_TRAMP flag
5973 global_ops.trampoline = save_global_trampoline;
5974 if (save_global_flags & FTRACE_OPS_FL_ALLOC_TRAMP)
5975 global_ops.flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
5979 mutex_unlock(&ftrace_lock);
5982 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
5985 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
5987 atomic_set(&t->tracing_graph_pause, 0);
5988 atomic_set(&t->trace_overrun, 0);
5989 t->ftrace_timestamp = 0;
5990 /* make curr_ret_stack visible before we add the ret_stack */
5992 t->ret_stack = ret_stack;
5996 * Allocate a return stack for the idle task. May be the first
5997 * time through, or it may be done by CPU hotplug online.
5999 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
6001 t->curr_ret_stack = -1;
6003 * The idle task has no parent, it either has its own
6004 * stack or no stack at all.
6007 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
6009 if (ftrace_graph_active) {
6010 struct ftrace_ret_stack *ret_stack;
6012 ret_stack = per_cpu(idle_ret_stack, cpu);
6014 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6015 * sizeof(struct ftrace_ret_stack),
6019 per_cpu(idle_ret_stack, cpu) = ret_stack;
6021 graph_init_task(t, ret_stack);
6025 /* Allocate a return stack for newly created task */
6026 void ftrace_graph_init_task(struct task_struct *t)
6028 /* Make sure we do not use the parent ret_stack */
6029 t->ret_stack = NULL;
6030 t->curr_ret_stack = -1;
6032 if (ftrace_graph_active) {
6033 struct ftrace_ret_stack *ret_stack;
6035 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6036 * sizeof(struct ftrace_ret_stack),
6040 graph_init_task(t, ret_stack);
6044 void ftrace_graph_exit_task(struct task_struct *t)
6046 struct ftrace_ret_stack *ret_stack = t->ret_stack;
6048 t->ret_stack = NULL;
6049 /* NULL must become visible to IRQs before we free it: */