Root/kernel/trace/ftrace.c

1/*
2 * Infrastructure for profiling code inserted by 'gcc -pg'.
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Originally ported from the -rt patch by:
8 * Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
9 *
10 * Based on code in the latency_tracer, that is:
11 *
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 William Lee Irwin III
14 */
15
16#include <linux/stop_machine.h>
17#include <linux/clocksource.h>
18#include <linux/kallsyms.h>
19#include <linux/seq_file.h>
20#include <linux/suspend.h>
21#include <linux/debugfs.h>
22#include <linux/hardirq.h>
23#include <linux/kthread.h>
24#include <linux/uaccess.h>
25#include <linux/ftrace.h>
26#include <linux/sysctl.h>
27#include <linux/slab.h>
28#include <linux/ctype.h>
29#include <linux/list.h>
30#include <linux/hash.h>
31#include <linux/rcupdate.h>
32
33#include <trace/events/sched.h>
34
35#include <asm/ftrace.h>
36#include <asm/setup.h>
37
38#include "trace_output.h"
39#include "trace_stat.h"
40
41#define FTRACE_WARN_ON(cond) \
42    ({ \
43        int ___r = cond; \
44        if (WARN_ON(___r)) \
45            ftrace_kill(); \
46        ___r; \
47    })
48
49#define FTRACE_WARN_ON_ONCE(cond) \
50    ({ \
51        int ___r = cond; \
52        if (WARN_ON_ONCE(___r)) \
53            ftrace_kill(); \
54        ___r; \
55    })
56
57/* hash bits for specific function selection */
58#define FTRACE_HASH_BITS 7
59#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
60#define FTRACE_HASH_DEFAULT_BITS 10
61#define FTRACE_HASH_MAX_BITS 12
62
63/* ftrace_enabled is a method to turn ftrace on or off */
64int ftrace_enabled __read_mostly;
65static int last_ftrace_enabled;
66
67/* Quick disabling of function tracer. */
68int function_trace_stop;
69
70/* List for set_ftrace_pid's pids. */
71LIST_HEAD(ftrace_pids);
72struct ftrace_pid {
73    struct list_head list;
74    struct pid *pid;
75};
76
77/*
78 * ftrace_disabled is set when an anomaly is discovered.
79 * ftrace_disabled is much stronger than ftrace_enabled.
80 */
81static int ftrace_disabled __read_mostly;
82
83static DEFINE_MUTEX(ftrace_lock);
84
85static struct ftrace_ops ftrace_list_end __read_mostly =
86{
87    .func = ftrace_stub,
88};
89
90static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
91static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
92ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
93ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
94ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
95static struct ftrace_ops global_ops;
96
97static void
98ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip);
99
100/*
101 * Traverse the ftrace_global_list, invoking all entries. The reason that we
102 * can use rcu_dereference_raw() is that elements removed from this list
103 * are simply leaked, so there is no need to interact with a grace-period
104 * mechanism. The rcu_dereference_raw() calls are needed to handle
105 * concurrent insertions into the ftrace_global_list.
106 *
107 * Silly Alpha and silly pointer-speculation compiler optimizations!
108 */
109static void ftrace_global_list_func(unsigned long ip,
110                    unsigned long parent_ip)
111{
112    struct ftrace_ops *op;
113
114    if (unlikely(trace_recursion_test(TRACE_GLOBAL_BIT)))
115        return;
116
117    trace_recursion_set(TRACE_GLOBAL_BIT);
118    op = rcu_dereference_raw(ftrace_global_list); /*see above*/
119    while (op != &ftrace_list_end) {
120        op->func(ip, parent_ip);
121        op = rcu_dereference_raw(op->next); /*see above*/
122    };
123    trace_recursion_clear(TRACE_GLOBAL_BIT);
124}
125
126static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
127{
128    if (!test_tsk_trace_trace(current))
129        return;
130
131    ftrace_pid_function(ip, parent_ip);
132}
133
134static void set_ftrace_pid_function(ftrace_func_t func)
135{
136    /* do not set ftrace_pid_function to itself! */
137    if (func != ftrace_pid_func)
138        ftrace_pid_function = func;
139}
140
141/**
142 * clear_ftrace_function - reset the ftrace function
143 *
144 * This NULLs the ftrace function and in essence stops
145 * tracing. There may be lag
146 */
147void clear_ftrace_function(void)
148{
149    ftrace_trace_function = ftrace_stub;
150    __ftrace_trace_function = ftrace_stub;
151    ftrace_pid_function = ftrace_stub;
152}
153
154#ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
155/*
156 * For those archs that do not test ftrace_trace_stop in their
157 * mcount call site, we need to do it from C.
158 */
159static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
160{
161    if (function_trace_stop)
162        return;
163
164    __ftrace_trace_function(ip, parent_ip);
165}
166#endif
167
168static void update_global_ops(void)
169{
170    ftrace_func_t func;
171
172    /*
173     * If there's only one function registered, then call that
174     * function directly. Otherwise, we need to iterate over the
175     * registered callers.
176     */
177    if (ftrace_global_list == &ftrace_list_end ||
178        ftrace_global_list->next == &ftrace_list_end)
179        func = ftrace_global_list->func;
180    else
181        func = ftrace_global_list_func;
182
183    /* If we filter on pids, update to use the pid function */
184    if (!list_empty(&ftrace_pids)) {
185        set_ftrace_pid_function(func);
186        func = ftrace_pid_func;
187    }
188
189    global_ops.func = func;
190}
191
192static void update_ftrace_function(void)
193{
194    ftrace_func_t func;
195
196    update_global_ops();
197
198    /*
199     * If we are at the end of the list and this ops is
200     * not dynamic, then have the mcount trampoline call
201     * the function directly
202     */
203    if (ftrace_ops_list == &ftrace_list_end ||
204        (ftrace_ops_list->next == &ftrace_list_end &&
205         !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC)))
206        func = ftrace_ops_list->func;
207    else
208        func = ftrace_ops_list_func;
209
210#ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
211    ftrace_trace_function = func;
212#else
213    __ftrace_trace_function = func;
214    ftrace_trace_function = ftrace_test_stop_func;
215#endif
216}
217
218static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
219{
220    ops->next = *list;
221    /*
222     * We are entering ops into the list but another
223     * CPU might be walking that list. We need to make sure
224     * the ops->next pointer is valid before another CPU sees
225     * the ops pointer included into the list.
226     */
227    rcu_assign_pointer(*list, ops);
228}
229
230static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
231{
232    struct ftrace_ops **p;
233
234    /*
235     * If we are removing the last function, then simply point
236     * to the ftrace_stub.
237     */
238    if (*list == ops && ops->next == &ftrace_list_end) {
239        *list = &ftrace_list_end;
240        return 0;
241    }
242
243    for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
244        if (*p == ops)
245            break;
246
247    if (*p != ops)
248        return -1;
249
250    *p = (*p)->next;
251    return 0;
252}
253
254static int __register_ftrace_function(struct ftrace_ops *ops)
255{
256    if (ftrace_disabled)
257        return -ENODEV;
258
259    if (FTRACE_WARN_ON(ops == &global_ops))
260        return -EINVAL;
261
262    if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
263        return -EBUSY;
264
265    if (!core_kernel_data((unsigned long)ops))
266        ops->flags |= FTRACE_OPS_FL_DYNAMIC;
267
268    if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
269        int first = ftrace_global_list == &ftrace_list_end;
270        add_ftrace_ops(&ftrace_global_list, ops);
271        ops->flags |= FTRACE_OPS_FL_ENABLED;
272        if (first)
273            add_ftrace_ops(&ftrace_ops_list, &global_ops);
274    } else
275        add_ftrace_ops(&ftrace_ops_list, ops);
276
277    if (ftrace_enabled)
278        update_ftrace_function();
279
280    return 0;
281}
282
283static int __unregister_ftrace_function(struct ftrace_ops *ops)
284{
285    int ret;
286
287    if (ftrace_disabled)
288        return -ENODEV;
289
290    if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
291        return -EBUSY;
292
293    if (FTRACE_WARN_ON(ops == &global_ops))
294        return -EINVAL;
295
296    if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
297        ret = remove_ftrace_ops(&ftrace_global_list, ops);
298        if (!ret && ftrace_global_list == &ftrace_list_end)
299            ret = remove_ftrace_ops(&ftrace_ops_list, &global_ops);
300        if (!ret)
301            ops->flags &= ~FTRACE_OPS_FL_ENABLED;
302    } else
303        ret = remove_ftrace_ops(&ftrace_ops_list, ops);
304
305    if (ret < 0)
306        return ret;
307
308    if (ftrace_enabled)
309        update_ftrace_function();
310
311    /*
312     * Dynamic ops may be freed, we must make sure that all
313     * callers are done before leaving this function.
314     */
315    if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
316        synchronize_sched();
317
318    return 0;
319}
320
321static void ftrace_update_pid_func(void)
322{
323    /* Only do something if we are tracing something */
324    if (ftrace_trace_function == ftrace_stub)
325        return;
326
327    update_ftrace_function();
328}
329
330#ifdef CONFIG_FUNCTION_PROFILER
331struct ftrace_profile {
332    struct hlist_node node;
333    unsigned long ip;
334    unsigned long counter;
335#ifdef CONFIG_FUNCTION_GRAPH_TRACER
336    unsigned long long time;
337    unsigned long long time_squared;
338#endif
339};
340
341struct ftrace_profile_page {
342    struct ftrace_profile_page *next;
343    unsigned long index;
344    struct ftrace_profile records[];
345};
346
347struct ftrace_profile_stat {
348    atomic_t disabled;
349    struct hlist_head *hash;
350    struct ftrace_profile_page *pages;
351    struct ftrace_profile_page *start;
352    struct tracer_stat stat;
353};
354
355#define PROFILE_RECORDS_SIZE \
356    (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
357
358#define PROFILES_PER_PAGE \
359    (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
360
361static int ftrace_profile_bits __read_mostly;
362static int ftrace_profile_enabled __read_mostly;
363
364/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
365static DEFINE_MUTEX(ftrace_profile_lock);
366
367static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
368
369#define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
370
371static void *
372function_stat_next(void *v, int idx)
373{
374    struct ftrace_profile *rec = v;
375    struct ftrace_profile_page *pg;
376
377    pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
378
379 again:
380    if (idx != 0)
381        rec++;
382
383    if ((void *)rec >= (void *)&pg->records[pg->index]) {
384        pg = pg->next;
385        if (!pg)
386            return NULL;
387        rec = &pg->records[0];
388        if (!rec->counter)
389            goto again;
390    }
391
392    return rec;
393}
394
395static void *function_stat_start(struct tracer_stat *trace)
396{
397    struct ftrace_profile_stat *stat =
398        container_of(trace, struct ftrace_profile_stat, stat);
399
400    if (!stat || !stat->start)
401        return NULL;
402
403    return function_stat_next(&stat->start->records[0], 0);
404}
405
406#ifdef CONFIG_FUNCTION_GRAPH_TRACER
407/* function graph compares on total time */
408static int function_stat_cmp(void *p1, void *p2)
409{
410    struct ftrace_profile *a = p1;
411    struct ftrace_profile *b = p2;
412
413    if (a->time < b->time)
414        return -1;
415    if (a->time > b->time)
416        return 1;
417    else
418        return 0;
419}
420#else
421/* not function graph compares against hits */
422static int function_stat_cmp(void *p1, void *p2)
423{
424    struct ftrace_profile *a = p1;
425    struct ftrace_profile *b = p2;
426
427    if (a->counter < b->counter)
428        return -1;
429    if (a->counter > b->counter)
430        return 1;
431    else
432        return 0;
433}
434#endif
435
436static int function_stat_headers(struct seq_file *m)
437{
438#ifdef CONFIG_FUNCTION_GRAPH_TRACER
439    seq_printf(m, " Function "
440           "Hit Time Avg s^2\n"
441              " -------- "
442           "--- ---- --- ---\n");
443#else
444    seq_printf(m, " Function Hit\n"
445              " -------- ---\n");
446#endif
447    return 0;
448}
449
450static int function_stat_show(struct seq_file *m, void *v)
451{
452    struct ftrace_profile *rec = v;
453    char str[KSYM_SYMBOL_LEN];
454    int ret = 0;
455#ifdef CONFIG_FUNCTION_GRAPH_TRACER
456    static struct trace_seq s;
457    unsigned long long avg;
458    unsigned long long stddev;
459#endif
460    mutex_lock(&ftrace_profile_lock);
461
462    /* we raced with function_profile_reset() */
463    if (unlikely(rec->counter == 0)) {
464        ret = -EBUSY;
465        goto out;
466    }
467
468    kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
469    seq_printf(m, " %-30.30s %10lu", str, rec->counter);
470
471#ifdef CONFIG_FUNCTION_GRAPH_TRACER
472    seq_printf(m, " ");
473    avg = rec->time;
474    do_div(avg, rec->counter);
475
476    /* Sample standard deviation (s^2) */
477    if (rec->counter <= 1)
478        stddev = 0;
479    else {
480        stddev = rec->time_squared - rec->counter * avg * avg;
481        /*
482         * Divide only 1000 for ns^2 -> us^2 conversion.
483         * trace_print_graph_duration will divide 1000 again.
484         */
485        do_div(stddev, (rec->counter - 1) * 1000);
486    }
487
488    trace_seq_init(&s);
489    trace_print_graph_duration(rec->time, &s);
490    trace_seq_puts(&s, " ");
491    trace_print_graph_duration(avg, &s);
492    trace_seq_puts(&s, " ");
493    trace_print_graph_duration(stddev, &s);
494    trace_print_seq(m, &s);
495#endif
496    seq_putc(m, '\n');
497out:
498    mutex_unlock(&ftrace_profile_lock);
499
500    return ret;
501}
502
503static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
504{
505    struct ftrace_profile_page *pg;
506
507    pg = stat->pages = stat->start;
508
509    while (pg) {
510        memset(pg->records, 0, PROFILE_RECORDS_SIZE);
511        pg->index = 0;
512        pg = pg->next;
513    }
514
515    memset(stat->hash, 0,
516           FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
517}
518
519int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
520{
521    struct ftrace_profile_page *pg;
522    int functions;
523    int pages;
524    int i;
525
526    /* If we already allocated, do nothing */
527    if (stat->pages)
528        return 0;
529
530    stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
531    if (!stat->pages)
532        return -ENOMEM;
533
534#ifdef CONFIG_DYNAMIC_FTRACE
535    functions = ftrace_update_tot_cnt;
536#else
537    /*
538     * We do not know the number of functions that exist because
539     * dynamic tracing is what counts them. With past experience
540     * we have around 20K functions. That should be more than enough.
541     * It is highly unlikely we will execute every function in
542     * the kernel.
543     */
544    functions = 20000;
545#endif
546
547    pg = stat->start = stat->pages;
548
549    pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
550
551    for (i = 0; i < pages; i++) {
552        pg->next = (void *)get_zeroed_page(GFP_KERNEL);
553        if (!pg->next)
554            goto out_free;
555        pg = pg->next;
556    }
557
558    return 0;
559
560 out_free:
561    pg = stat->start;
562    while (pg) {
563        unsigned long tmp = (unsigned long)pg;
564
565        pg = pg->next;
566        free_page(tmp);
567    }
568
569    free_page((unsigned long)stat->pages);
570    stat->pages = NULL;
571    stat->start = NULL;
572
573    return -ENOMEM;
574}
575
576static int ftrace_profile_init_cpu(int cpu)
577{
578    struct ftrace_profile_stat *stat;
579    int size;
580
581    stat = &per_cpu(ftrace_profile_stats, cpu);
582
583    if (stat->hash) {
584        /* If the profile is already created, simply reset it */
585        ftrace_profile_reset(stat);
586        return 0;
587    }
588
589    /*
590     * We are profiling all functions, but usually only a few thousand
591     * functions are hit. We'll make a hash of 1024 items.
592     */
593    size = FTRACE_PROFILE_HASH_SIZE;
594
595    stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
596
597    if (!stat->hash)
598        return -ENOMEM;
599
600    if (!ftrace_profile_bits) {
601        size--;
602
603        for (; size; size >>= 1)
604            ftrace_profile_bits++;
605    }
606
607    /* Preallocate the function profiling pages */
608    if (ftrace_profile_pages_init(stat) < 0) {
609        kfree(stat->hash);
610        stat->hash = NULL;
611        return -ENOMEM;
612    }
613
614    return 0;
615}
616
617static int ftrace_profile_init(void)
618{
619    int cpu;
620    int ret = 0;
621
622    for_each_online_cpu(cpu) {
623        ret = ftrace_profile_init_cpu(cpu);
624        if (ret)
625            break;
626    }
627
628    return ret;
629}
630
631/* interrupts must be disabled */
632static struct ftrace_profile *
633ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
634{
635    struct ftrace_profile *rec;
636    struct hlist_head *hhd;
637    struct hlist_node *n;
638    unsigned long key;
639
640    key = hash_long(ip, ftrace_profile_bits);
641    hhd = &stat->hash[key];
642
643    if (hlist_empty(hhd))
644        return NULL;
645
646    hlist_for_each_entry_rcu(rec, n, hhd, node) {
647        if (rec->ip == ip)
648            return rec;
649    }
650
651    return NULL;
652}
653
654static void ftrace_add_profile(struct ftrace_profile_stat *stat,
655                   struct ftrace_profile *rec)
656{
657    unsigned long key;
658
659    key = hash_long(rec->ip, ftrace_profile_bits);
660    hlist_add_head_rcu(&rec->node, &stat->hash[key]);
661}
662
663/*
664 * The memory is already allocated, this simply finds a new record to use.
665 */
666static struct ftrace_profile *
667ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
668{
669    struct ftrace_profile *rec = NULL;
670
671    /* prevent recursion (from NMIs) */
672    if (atomic_inc_return(&stat->disabled) != 1)
673        goto out;
674
675    /*
676     * Try to find the function again since an NMI
677     * could have added it
678     */
679    rec = ftrace_find_profiled_func(stat, ip);
680    if (rec)
681        goto out;
682
683    if (stat->pages->index == PROFILES_PER_PAGE) {
684        if (!stat->pages->next)
685            goto out;
686        stat->pages = stat->pages->next;
687    }
688
689    rec = &stat->pages->records[stat->pages->index++];
690    rec->ip = ip;
691    ftrace_add_profile(stat, rec);
692
693 out:
694    atomic_dec(&stat->disabled);
695
696    return rec;
697}
698
699static void
700function_profile_call(unsigned long ip, unsigned long parent_ip)
701{
702    struct ftrace_profile_stat *stat;
703    struct ftrace_profile *rec;
704    unsigned long flags;
705
706    if (!ftrace_profile_enabled)
707        return;
708
709    local_irq_save(flags);
710
711    stat = &__get_cpu_var(ftrace_profile_stats);
712    if (!stat->hash || !ftrace_profile_enabled)
713        goto out;
714
715    rec = ftrace_find_profiled_func(stat, ip);
716    if (!rec) {
717        rec = ftrace_profile_alloc(stat, ip);
718        if (!rec)
719            goto out;
720    }
721
722    rec->counter++;
723 out:
724    local_irq_restore(flags);
725}
726
727#ifdef CONFIG_FUNCTION_GRAPH_TRACER
728static int profile_graph_entry(struct ftrace_graph_ent *trace)
729{
730    function_profile_call(trace->func, 0);
731    return 1;
732}
733
734static void profile_graph_return(struct ftrace_graph_ret *trace)
735{
736    struct ftrace_profile_stat *stat;
737    unsigned long long calltime;
738    struct ftrace_profile *rec;
739    unsigned long flags;
740
741    local_irq_save(flags);
742    stat = &__get_cpu_var(ftrace_profile_stats);
743    if (!stat->hash || !ftrace_profile_enabled)
744        goto out;
745
746    /* If the calltime was zero'd ignore it */
747    if (!trace->calltime)
748        goto out;
749
750    calltime = trace->rettime - trace->calltime;
751
752    if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
753        int index;
754
755        index = trace->depth;
756
757        /* Append this call time to the parent time to subtract */
758        if (index)
759            current->ret_stack[index - 1].subtime += calltime;
760
761        if (current->ret_stack[index].subtime < calltime)
762            calltime -= current->ret_stack[index].subtime;
763        else
764            calltime = 0;
765    }
766
767    rec = ftrace_find_profiled_func(stat, trace->func);
768    if (rec) {
769        rec->time += calltime;
770        rec->time_squared += calltime * calltime;
771    }
772
773 out:
774    local_irq_restore(flags);
775}
776
777static int register_ftrace_profiler(void)
778{
779    return register_ftrace_graph(&profile_graph_return,
780                     &profile_graph_entry);
781}
782
783static void unregister_ftrace_profiler(void)
784{
785    unregister_ftrace_graph();
786}
787#else
788static struct ftrace_ops ftrace_profile_ops __read_mostly =
789{
790    .func = function_profile_call,
791};
792
793static int register_ftrace_profiler(void)
794{
795    return register_ftrace_function(&ftrace_profile_ops);
796}
797
798static void unregister_ftrace_profiler(void)
799{
800    unregister_ftrace_function(&ftrace_profile_ops);
801}
802#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
803
804static ssize_t
805ftrace_profile_write(struct file *filp, const char __user *ubuf,
806             size_t cnt, loff_t *ppos)
807{
808    unsigned long val;
809    char buf[64]; /* big enough to hold a number */
810    int ret;
811
812    if (cnt >= sizeof(buf))
813        return -EINVAL;
814
815    if (copy_from_user(&buf, ubuf, cnt))
816        return -EFAULT;
817
818    buf[cnt] = 0;
819
820    ret = strict_strtoul(buf, 10, &val);
821    if (ret < 0)
822        return ret;
823
824    val = !!val;
825
826    mutex_lock(&ftrace_profile_lock);
827    if (ftrace_profile_enabled ^ val) {
828        if (val) {
829            ret = ftrace_profile_init();
830            if (ret < 0) {
831                cnt = ret;
832                goto out;
833            }
834
835            ret = register_ftrace_profiler();
836            if (ret < 0) {
837                cnt = ret;
838                goto out;
839            }
840            ftrace_profile_enabled = 1;
841        } else {
842            ftrace_profile_enabled = 0;
843            /*
844             * unregister_ftrace_profiler calls stop_machine
845             * so this acts like an synchronize_sched.
846             */
847            unregister_ftrace_profiler();
848        }
849    }
850 out:
851    mutex_unlock(&ftrace_profile_lock);
852
853    *ppos += cnt;
854
855    return cnt;
856}
857
858static ssize_t
859ftrace_profile_read(struct file *filp, char __user *ubuf,
860             size_t cnt, loff_t *ppos)
861{
862    char buf[64]; /* big enough to hold a number */
863    int r;
864
865    r = sprintf(buf, "%u\n", ftrace_profile_enabled);
866    return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
867}
868
869static const struct file_operations ftrace_profile_fops = {
870    .open = tracing_open_generic,
871    .read = ftrace_profile_read,
872    .write = ftrace_profile_write,
873    .llseek = default_llseek,
874};
875
876/* used to initialize the real stat files */
877static struct tracer_stat function_stats __initdata = {
878    .name = "functions",
879    .stat_start = function_stat_start,
880    .stat_next = function_stat_next,
881    .stat_cmp = function_stat_cmp,
882    .stat_headers = function_stat_headers,
883    .stat_show = function_stat_show
884};
885
886static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
887{
888    struct ftrace_profile_stat *stat;
889    struct dentry *entry;
890    char *name;
891    int ret;
892    int cpu;
893
894    for_each_possible_cpu(cpu) {
895        stat = &per_cpu(ftrace_profile_stats, cpu);
896
897        /* allocate enough for function name + cpu number */
898        name = kmalloc(32, GFP_KERNEL);
899        if (!name) {
900            /*
901             * The files created are permanent, if something happens
902             * we still do not free memory.
903             */
904            WARN(1,
905                 "Could not allocate stat file for cpu %d\n",
906                 cpu);
907            return;
908        }
909        stat->stat = function_stats;
910        snprintf(name, 32, "function%d", cpu);
911        stat->stat.name = name;
912        ret = register_stat_tracer(&stat->stat);
913        if (ret) {
914            WARN(1,
915                 "Could not register function stat for cpu %d\n",
916                 cpu);
917            kfree(name);
918            return;
919        }
920    }
921
922    entry = debugfs_create_file("function_profile_enabled", 0644,
923                    d_tracer, NULL, &ftrace_profile_fops);
924    if (!entry)
925        pr_warning("Could not create debugfs "
926               "'function_profile_enabled' entry\n");
927}
928
929#else /* CONFIG_FUNCTION_PROFILER */
930static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
931{
932}
933#endif /* CONFIG_FUNCTION_PROFILER */
934
935static struct pid * const ftrace_swapper_pid = &init_struct_pid;
936
937#ifdef CONFIG_DYNAMIC_FTRACE
938
939#ifndef CONFIG_FTRACE_MCOUNT_RECORD
940# error Dynamic ftrace depends on MCOUNT_RECORD
941#endif
942
943static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
944
945struct ftrace_func_probe {
946    struct hlist_node node;
947    struct ftrace_probe_ops *ops;
948    unsigned long flags;
949    unsigned long ip;
950    void *data;
951    struct rcu_head rcu;
952};
953
954enum {
955    FTRACE_ENABLE_CALLS = (1 << 0),
956    FTRACE_DISABLE_CALLS = (1 << 1),
957    FTRACE_UPDATE_TRACE_FUNC = (1 << 2),
958    FTRACE_START_FUNC_RET = (1 << 3),
959    FTRACE_STOP_FUNC_RET = (1 << 4),
960};
961struct ftrace_func_entry {
962    struct hlist_node hlist;
963    unsigned long ip;
964};
965
966struct ftrace_hash {
967    unsigned long size_bits;
968    struct hlist_head *buckets;
969    unsigned long count;
970    struct rcu_head rcu;
971};
972
973/*
974 * We make these constant because no one should touch them,
975 * but they are used as the default "empty hash", to avoid allocating
976 * it all the time. These are in a read only section such that if
977 * anyone does try to modify it, it will cause an exception.
978 */
979static const struct hlist_head empty_buckets[1];
980static const struct ftrace_hash empty_hash = {
981    .buckets = (struct hlist_head *)empty_buckets,
982};
983#define EMPTY_HASH ((struct ftrace_hash *)&empty_hash)
984
985static struct ftrace_ops global_ops = {
986    .func = ftrace_stub,
987    .notrace_hash = EMPTY_HASH,
988    .filter_hash = EMPTY_HASH,
989};
990
991static struct dyn_ftrace *ftrace_new_addrs;
992
993static DEFINE_MUTEX(ftrace_regex_lock);
994
995struct ftrace_page {
996    struct ftrace_page *next;
997    int index;
998    struct dyn_ftrace records[];
999};
1000
1001#define ENTRIES_PER_PAGE \
1002  ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
1003
1004/* estimate from running different kernels */
1005#define NR_TO_INIT 10000
1006
1007static struct ftrace_page *ftrace_pages_start;
1008static struct ftrace_page *ftrace_pages;
1009
1010static struct dyn_ftrace *ftrace_free_records;
1011
1012static struct ftrace_func_entry *
1013ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1014{
1015    unsigned long key;
1016    struct ftrace_func_entry *entry;
1017    struct hlist_head *hhd;
1018    struct hlist_node *n;
1019
1020    if (!hash->count)
1021        return NULL;
1022
1023    if (hash->size_bits > 0)
1024        key = hash_long(ip, hash->size_bits);
1025    else
1026        key = 0;
1027
1028    hhd = &hash->buckets[key];
1029
1030    hlist_for_each_entry_rcu(entry, n, hhd, hlist) {
1031        if (entry->ip == ip)
1032            return entry;
1033    }
1034    return NULL;
1035}
1036
1037static void __add_hash_entry(struct ftrace_hash *hash,
1038                 struct ftrace_func_entry *entry)
1039{
1040    struct hlist_head *hhd;
1041    unsigned long key;
1042
1043    if (hash->size_bits)
1044        key = hash_long(entry->ip, hash->size_bits);
1045    else
1046        key = 0;
1047
1048    hhd = &hash->buckets[key];
1049    hlist_add_head(&entry->hlist, hhd);
1050    hash->count++;
1051}
1052
1053static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1054{
1055    struct ftrace_func_entry *entry;
1056
1057    entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1058    if (!entry)
1059        return -ENOMEM;
1060
1061    entry->ip = ip;
1062    __add_hash_entry(hash, entry);
1063
1064    return 0;
1065}
1066
1067static void
1068free_hash_entry(struct ftrace_hash *hash,
1069          struct ftrace_func_entry *entry)
1070{
1071    hlist_del(&entry->hlist);
1072    kfree(entry);
1073    hash->count--;
1074}
1075
1076static void
1077remove_hash_entry(struct ftrace_hash *hash,
1078          struct ftrace_func_entry *entry)
1079{
1080    hlist_del(&entry->hlist);
1081    hash->count--;
1082}
1083
1084static void ftrace_hash_clear(struct ftrace_hash *hash)
1085{
1086    struct hlist_head *hhd;
1087    struct hlist_node *tp, *tn;
1088    struct ftrace_func_entry *entry;
1089    int size = 1 << hash->size_bits;
1090    int i;
1091
1092    if (!hash->count)
1093        return;
1094
1095    for (i = 0; i < size; i++) {
1096        hhd = &hash->buckets[i];
1097        hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist)
1098            free_hash_entry(hash, entry);
1099    }
1100    FTRACE_WARN_ON(hash->count);
1101}
1102
1103static void free_ftrace_hash(struct ftrace_hash *hash)
1104{
1105    if (!hash || hash == EMPTY_HASH)
1106        return;
1107    ftrace_hash_clear(hash);
1108    kfree(hash->buckets);
1109    kfree(hash);
1110}
1111
1112static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1113{
1114    struct ftrace_hash *hash;
1115
1116    hash = container_of(rcu, struct ftrace_hash, rcu);
1117    free_ftrace_hash(hash);
1118}
1119
1120static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1121{
1122    if (!hash || hash == EMPTY_HASH)
1123        return;
1124    call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1125}
1126
1127static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1128{
1129    struct ftrace_hash *hash;
1130    int size;
1131
1132    hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1133    if (!hash)
1134        return NULL;
1135
1136    size = 1 << size_bits;
1137    hash->buckets = kzalloc(sizeof(*hash->buckets) * size, GFP_KERNEL);
1138
1139    if (!hash->buckets) {
1140        kfree(hash);
1141        return NULL;
1142    }
1143
1144    hash->size_bits = size_bits;
1145
1146    return hash;
1147}
1148
1149static struct ftrace_hash *
1150alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1151{
1152    struct ftrace_func_entry *entry;
1153    struct ftrace_hash *new_hash;
1154    struct hlist_node *tp;
1155    int size;
1156    int ret;
1157    int i;
1158
1159    new_hash = alloc_ftrace_hash(size_bits);
1160    if (!new_hash)
1161        return NULL;
1162
1163    /* Empty hash? */
1164    if (!hash || !hash->count)
1165        return new_hash;
1166
1167    size = 1 << hash->size_bits;
1168    for (i = 0; i < size; i++) {
1169        hlist_for_each_entry(entry, tp, &hash->buckets[i], hlist) {
1170            ret = add_hash_entry(new_hash, entry->ip);
1171            if (ret < 0)
1172                goto free_hash;
1173        }
1174    }
1175
1176    FTRACE_WARN_ON(new_hash->count != hash->count);
1177
1178    return new_hash;
1179
1180 free_hash:
1181    free_ftrace_hash(new_hash);
1182    return NULL;
1183}
1184
1185static int
1186ftrace_hash_move(struct ftrace_hash **dst, struct ftrace_hash *src)
1187{
1188    struct ftrace_func_entry *entry;
1189    struct hlist_node *tp, *tn;
1190    struct hlist_head *hhd;
1191    struct ftrace_hash *old_hash;
1192    struct ftrace_hash *new_hash;
1193    unsigned long key;
1194    int size = src->count;
1195    int bits = 0;
1196    int i;
1197
1198    /*
1199     * If the new source is empty, just free dst and assign it
1200     * the empty_hash.
1201     */
1202    if (!src->count) {
1203        free_ftrace_hash_rcu(*dst);
1204        rcu_assign_pointer(*dst, EMPTY_HASH);
1205        return 0;
1206    }
1207
1208    /*
1209     * Make the hash size about 1/2 the # found
1210     */
1211    for (size /= 2; size; size >>= 1)
1212        bits++;
1213
1214    /* Don't allocate too much */
1215    if (bits > FTRACE_HASH_MAX_BITS)
1216        bits = FTRACE_HASH_MAX_BITS;
1217
1218    new_hash = alloc_ftrace_hash(bits);
1219    if (!new_hash)
1220        return -ENOMEM;
1221
1222    size = 1 << src->size_bits;
1223    for (i = 0; i < size; i++) {
1224        hhd = &src->buckets[i];
1225        hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist) {
1226            if (bits > 0)
1227                key = hash_long(entry->ip, bits);
1228            else
1229                key = 0;
1230            remove_hash_entry(src, entry);
1231            __add_hash_entry(new_hash, entry);
1232        }
1233    }
1234
1235    old_hash = *dst;
1236    rcu_assign_pointer(*dst, new_hash);
1237    free_ftrace_hash_rcu(old_hash);
1238
1239    return 0;
1240}
1241
1242/*
1243 * Test the hashes for this ops to see if we want to call
1244 * the ops->func or not.
1245 *
1246 * It's a match if the ip is in the ops->filter_hash or
1247 * the filter_hash does not exist or is empty,
1248 * AND
1249 * the ip is not in the ops->notrace_hash.
1250 *
1251 * This needs to be called with preemption disabled as
1252 * the hashes are freed with call_rcu_sched().
1253 */
1254static int
1255ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
1256{
1257    struct ftrace_hash *filter_hash;
1258    struct ftrace_hash *notrace_hash;
1259    int ret;
1260
1261    filter_hash = rcu_dereference_raw(ops->filter_hash);
1262    notrace_hash = rcu_dereference_raw(ops->notrace_hash);
1263
1264    if ((!filter_hash || !filter_hash->count ||
1265         ftrace_lookup_ip(filter_hash, ip)) &&
1266        (!notrace_hash || !notrace_hash->count ||
1267         !ftrace_lookup_ip(notrace_hash, ip)))
1268        ret = 1;
1269    else
1270        ret = 0;
1271
1272    return ret;
1273}
1274
1275/*
1276 * This is a double for. Do not use 'break' to break out of the loop,
1277 * you must use a goto.
1278 */
1279#define do_for_each_ftrace_rec(pg, rec) \
1280    for (pg = ftrace_pages_start; pg; pg = pg->next) { \
1281        int _____i; \
1282        for (_____i = 0; _____i < pg->index; _____i++) { \
1283            rec = &pg->records[_____i];
1284
1285#define while_for_each_ftrace_rec() \
1286        } \
1287    }
1288
1289static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1290                     int filter_hash,
1291                     bool inc)
1292{
1293    struct ftrace_hash *hash;
1294    struct ftrace_hash *other_hash;
1295    struct ftrace_page *pg;
1296    struct dyn_ftrace *rec;
1297    int count = 0;
1298    int all = 0;
1299
1300    /* Only update if the ops has been registered */
1301    if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1302        return;
1303
1304    /*
1305     * In the filter_hash case:
1306     * If the count is zero, we update all records.
1307     * Otherwise we just update the items in the hash.
1308     *
1309     * In the notrace_hash case:
1310     * We enable the update in the hash.
1311     * As disabling notrace means enabling the tracing,
1312     * and enabling notrace means disabling, the inc variable
1313     * gets inversed.
1314     */
1315    if (filter_hash) {
1316        hash = ops->filter_hash;
1317        other_hash = ops->notrace_hash;
1318        if (!hash || !hash->count)
1319            all = 1;
1320    } else {
1321        inc = !inc;
1322        hash = ops->notrace_hash;
1323        other_hash = ops->filter_hash;
1324        /*
1325         * If the notrace hash has no items,
1326         * then there's nothing to do.
1327         */
1328        if (hash && !hash->count)
1329            return;
1330    }
1331
1332    do_for_each_ftrace_rec(pg, rec) {
1333        int in_other_hash = 0;
1334        int in_hash = 0;
1335        int match = 0;
1336
1337        if (all) {
1338            /*
1339             * Only the filter_hash affects all records.
1340             * Update if the record is not in the notrace hash.
1341             */
1342            if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1343                match = 1;
1344        } else {
1345            in_hash = hash && !!ftrace_lookup_ip(hash, rec->ip);
1346            in_other_hash = other_hash && !!ftrace_lookup_ip(other_hash, rec->ip);
1347
1348            /*
1349             *
1350             */
1351            if (filter_hash && in_hash && !in_other_hash)
1352                match = 1;
1353            else if (!filter_hash && in_hash &&
1354                 (in_other_hash || !other_hash->count))
1355                match = 1;
1356        }
1357        if (!match)
1358            continue;
1359
1360        if (inc) {
1361            rec->flags++;
1362            if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1363                return;
1364        } else {
1365            if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1366                return;
1367            rec->flags--;
1368        }
1369        count++;
1370        /* Shortcut, if we handled all records, we are done. */
1371        if (!all && count == hash->count)
1372            return;
1373    } while_for_each_ftrace_rec();
1374}
1375
1376static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1377                    int filter_hash)
1378{
1379    __ftrace_hash_rec_update(ops, filter_hash, 0);
1380}
1381
1382static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1383                   int filter_hash)
1384{
1385    __ftrace_hash_rec_update(ops, filter_hash, 1);
1386}
1387
1388static void ftrace_free_rec(struct dyn_ftrace *rec)
1389{
1390    rec->freelist = ftrace_free_records;
1391    ftrace_free_records = rec;
1392    rec->flags |= FTRACE_FL_FREE;
1393}
1394
1395static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
1396{
1397    struct dyn_ftrace *rec;
1398
1399    /* First check for freed records */
1400    if (ftrace_free_records) {
1401        rec = ftrace_free_records;
1402
1403        if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
1404            FTRACE_WARN_ON_ONCE(1);
1405            ftrace_free_records = NULL;
1406            return NULL;
1407        }
1408
1409        ftrace_free_records = rec->freelist;
1410        memset(rec, 0, sizeof(*rec));
1411        return rec;
1412    }
1413
1414    if (ftrace_pages->index == ENTRIES_PER_PAGE) {
1415        if (!ftrace_pages->next) {
1416            /* allocate another page */
1417            ftrace_pages->next =
1418                (void *)get_zeroed_page(GFP_KERNEL);
1419            if (!ftrace_pages->next)
1420                return NULL;
1421        }
1422        ftrace_pages = ftrace_pages->next;
1423    }
1424
1425    return &ftrace_pages->records[ftrace_pages->index++];
1426}
1427
1428static struct dyn_ftrace *
1429ftrace_record_ip(unsigned long ip)
1430{
1431    struct dyn_ftrace *rec;
1432
1433    if (ftrace_disabled)
1434        return NULL;
1435
1436    rec = ftrace_alloc_dyn_node(ip);
1437    if (!rec)
1438        return NULL;
1439
1440    rec->ip = ip;
1441    rec->newlist = ftrace_new_addrs;
1442    ftrace_new_addrs = rec;
1443
1444    return rec;
1445}
1446
1447static void print_ip_ins(const char *fmt, unsigned char *p)
1448{
1449    int i;
1450
1451    printk(KERN_CONT "%s", fmt);
1452
1453    for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1454        printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1455}
1456
1457static void ftrace_bug(int failed, unsigned long ip)
1458{
1459    switch (failed) {
1460    case -EFAULT:
1461        FTRACE_WARN_ON_ONCE(1);
1462        pr_info("ftrace faulted on modifying ");
1463        print_ip_sym(ip);
1464        break;
1465    case -EINVAL:
1466        FTRACE_WARN_ON_ONCE(1);
1467        pr_info("ftrace failed to modify ");
1468        print_ip_sym(ip);
1469        print_ip_ins(" actual: ", (unsigned char *)ip);
1470        printk(KERN_CONT "\n");
1471        break;
1472    case -EPERM:
1473        FTRACE_WARN_ON_ONCE(1);
1474        pr_info("ftrace faulted on writing ");
1475        print_ip_sym(ip);
1476        break;
1477    default:
1478        FTRACE_WARN_ON_ONCE(1);
1479        pr_info("ftrace faulted on unknown error ");
1480        print_ip_sym(ip);
1481    }
1482}
1483
1484
1485/* Return 1 if the address range is reserved for ftrace */
1486int ftrace_text_reserved(void *start, void *end)
1487{
1488    struct dyn_ftrace *rec;
1489    struct ftrace_page *pg;
1490
1491    do_for_each_ftrace_rec(pg, rec) {
1492        if (rec->ip <= (unsigned long)end &&
1493            rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1494            return 1;
1495    } while_for_each_ftrace_rec();
1496    return 0;
1497}
1498
1499
1500static int
1501__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1502{
1503    unsigned long ftrace_addr;
1504    unsigned long flag = 0UL;
1505
1506    ftrace_addr = (unsigned long)FTRACE_ADDR;
1507
1508    /*
1509     * If we are enabling tracing:
1510     *
1511     * If the record has a ref count, then we need to enable it
1512     * because someone is using it.
1513     *
1514     * Otherwise we make sure its disabled.
1515     *
1516     * If we are disabling tracing, then disable all records that
1517     * are enabled.
1518     */
1519    if (enable && (rec->flags & ~FTRACE_FL_MASK))
1520        flag = FTRACE_FL_ENABLED;
1521
1522    /* If the state of this record hasn't changed, then do nothing */
1523    if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1524        return 0;
1525
1526    if (flag) {
1527        rec->flags |= FTRACE_FL_ENABLED;
1528        return ftrace_make_call(rec, ftrace_addr);
1529    }
1530
1531    rec->flags &= ~FTRACE_FL_ENABLED;
1532    return ftrace_make_nop(NULL, rec, ftrace_addr);
1533}
1534
1535static void ftrace_replace_code(int enable)
1536{
1537    struct dyn_ftrace *rec;
1538    struct ftrace_page *pg;
1539    int failed;
1540
1541    if (unlikely(ftrace_disabled))
1542        return;
1543
1544    do_for_each_ftrace_rec(pg, rec) {
1545        /* Skip over free records */
1546        if (rec->flags & FTRACE_FL_FREE)
1547            continue;
1548
1549        failed = __ftrace_replace_code(rec, enable);
1550        if (failed) {
1551            ftrace_bug(failed, rec->ip);
1552            /* Stop processing */
1553            return;
1554        }
1555    } while_for_each_ftrace_rec();
1556}
1557
1558static int
1559ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1560{
1561    unsigned long ip;
1562    int ret;
1563
1564    ip = rec->ip;
1565
1566    if (unlikely(ftrace_disabled))
1567        return 0;
1568
1569    ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1570    if (ret) {
1571        ftrace_bug(ret, ip);
1572        return 0;
1573    }
1574    return 1;
1575}
1576
1577/*
1578 * archs can override this function if they must do something
1579 * before the modifying code is performed.
1580 */
1581int __weak ftrace_arch_code_modify_prepare(void)
1582{
1583    return 0;
1584}
1585
1586/*
1587 * archs can override this function if they must do something
1588 * after the modifying code is performed.
1589 */
1590int __weak ftrace_arch_code_modify_post_process(void)
1591{
1592    return 0;
1593}
1594
1595static int __ftrace_modify_code(void *data)
1596{
1597    int *command = data;
1598
1599    if (*command & FTRACE_ENABLE_CALLS)
1600        ftrace_replace_code(1);
1601    else if (*command & FTRACE_DISABLE_CALLS)
1602        ftrace_replace_code(0);
1603
1604    if (*command & FTRACE_UPDATE_TRACE_FUNC)
1605        ftrace_update_ftrace_func(ftrace_trace_function);
1606
1607    if (*command & FTRACE_START_FUNC_RET)
1608        ftrace_enable_ftrace_graph_caller();
1609    else if (*command & FTRACE_STOP_FUNC_RET)
1610        ftrace_disable_ftrace_graph_caller();
1611
1612    return 0;
1613}
1614
1615static void ftrace_run_update_code(int command)
1616{
1617    int ret;
1618
1619    ret = ftrace_arch_code_modify_prepare();
1620    FTRACE_WARN_ON(ret);
1621    if (ret)
1622        return;
1623
1624    stop_machine(__ftrace_modify_code, &command, NULL);
1625
1626    ret = ftrace_arch_code_modify_post_process();
1627    FTRACE_WARN_ON(ret);
1628}
1629
1630static ftrace_func_t saved_ftrace_func;
1631static int ftrace_start_up;
1632static int global_start_up;
1633
1634static void ftrace_startup_enable(int command)
1635{
1636    if (saved_ftrace_func != ftrace_trace_function) {
1637        saved_ftrace_func = ftrace_trace_function;
1638        command |= FTRACE_UPDATE_TRACE_FUNC;
1639    }
1640
1641    if (!command || !ftrace_enabled)
1642        return;
1643
1644    ftrace_run_update_code(command);
1645}
1646
1647static int ftrace_startup(struct ftrace_ops *ops, int command)
1648{
1649    bool hash_enable = true;
1650
1651    if (unlikely(ftrace_disabled))
1652        return -ENODEV;
1653
1654    ftrace_start_up++;
1655    command |= FTRACE_ENABLE_CALLS;
1656
1657    /* ops marked global share the filter hashes */
1658    if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1659        ops = &global_ops;
1660        /* Don't update hash if global is already set */
1661        if (global_start_up)
1662            hash_enable = false;
1663        global_start_up++;
1664    }
1665
1666    ops->flags |= FTRACE_OPS_FL_ENABLED;
1667    if (hash_enable)
1668        ftrace_hash_rec_enable(ops, 1);
1669
1670    ftrace_startup_enable(command);
1671
1672    return 0;
1673}
1674
1675static void ftrace_shutdown(struct ftrace_ops *ops, int command)
1676{
1677    bool hash_disable = true;
1678
1679    if (unlikely(ftrace_disabled))
1680        return;
1681
1682    ftrace_start_up--;
1683    /*
1684     * Just warn in case of unbalance, no need to kill ftrace, it's not
1685     * critical but the ftrace_call callers may be never nopped again after
1686     * further ftrace uses.
1687     */
1688    WARN_ON_ONCE(ftrace_start_up < 0);
1689
1690    if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1691        ops = &global_ops;
1692        global_start_up--;
1693        WARN_ON_ONCE(global_start_up < 0);
1694        /* Don't update hash if global still has users */
1695        if (global_start_up) {
1696            WARN_ON_ONCE(!ftrace_start_up);
1697            hash_disable = false;
1698        }
1699    }
1700
1701    if (hash_disable)
1702        ftrace_hash_rec_disable(ops, 1);
1703
1704    if (ops != &global_ops || !global_start_up)
1705        ops->flags &= ~FTRACE_OPS_FL_ENABLED;
1706
1707    if (!ftrace_start_up)
1708        command |= FTRACE_DISABLE_CALLS;
1709
1710    if (saved_ftrace_func != ftrace_trace_function) {
1711        saved_ftrace_func = ftrace_trace_function;
1712        command |= FTRACE_UPDATE_TRACE_FUNC;
1713    }
1714
1715    if (!command || !ftrace_enabled)
1716        return;
1717
1718    ftrace_run_update_code(command);
1719}
1720
1721static void ftrace_startup_sysctl(void)
1722{
1723    if (unlikely(ftrace_disabled))
1724        return;
1725
1726    /* Force update next time */
1727    saved_ftrace_func = NULL;
1728    /* ftrace_start_up is true if we want ftrace running */
1729    if (ftrace_start_up)
1730        ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1731}
1732
1733static void ftrace_shutdown_sysctl(void)
1734{
1735    if (unlikely(ftrace_disabled))
1736        return;
1737
1738    /* ftrace_start_up is true if ftrace is running */
1739    if (ftrace_start_up)
1740        ftrace_run_update_code(FTRACE_DISABLE_CALLS);
1741}
1742
1743static cycle_t ftrace_update_time;
1744static unsigned long ftrace_update_cnt;
1745unsigned long ftrace_update_tot_cnt;
1746
1747static int ftrace_update_code(struct module *mod)
1748{
1749    struct dyn_ftrace *p;
1750    cycle_t start, stop;
1751
1752    start = ftrace_now(raw_smp_processor_id());
1753    ftrace_update_cnt = 0;
1754
1755    while (ftrace_new_addrs) {
1756
1757        /* If something went wrong, bail without enabling anything */
1758        if (unlikely(ftrace_disabled))
1759            return -1;
1760
1761        p = ftrace_new_addrs;
1762        ftrace_new_addrs = p->newlist;
1763        p->flags = 0L;
1764
1765        /*
1766         * Do the initial record conversion from mcount jump
1767         * to the NOP instructions.
1768         */
1769        if (!ftrace_code_disable(mod, p)) {
1770            ftrace_free_rec(p);
1771            /* Game over */
1772            break;
1773        }
1774
1775        ftrace_update_cnt++;
1776
1777        /*
1778         * If the tracing is enabled, go ahead and enable the record.
1779         *
1780         * The reason not to enable the record immediatelly is the
1781         * inherent check of ftrace_make_nop/ftrace_make_call for
1782         * correct previous instructions. Making first the NOP
1783         * conversion puts the module to the correct state, thus
1784         * passing the ftrace_make_call check.
1785         */
1786        if (ftrace_start_up) {
1787            int failed = __ftrace_replace_code(p, 1);
1788            if (failed) {
1789                ftrace_bug(failed, p->ip);
1790                ftrace_free_rec(p);
1791            }
1792        }
1793    }
1794
1795    stop = ftrace_now(raw_smp_processor_id());
1796    ftrace_update_time = stop - start;
1797    ftrace_update_tot_cnt += ftrace_update_cnt;
1798
1799    return 0;
1800}
1801
1802static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1803{
1804    struct ftrace_page *pg;
1805    int cnt;
1806    int i;
1807
1808    /* allocate a few pages */
1809    ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1810    if (!ftrace_pages_start)
1811        return -1;
1812
1813    /*
1814     * Allocate a few more pages.
1815     *
1816     * TODO: have some parser search vmlinux before
1817     * final linking to find all calls to ftrace.
1818     * Then we can:
1819     * a) know how many pages to allocate.
1820     * and/or
1821     * b) set up the table then.
1822     *
1823     * The dynamic code is still necessary for
1824     * modules.
1825     */
1826
1827    pg = ftrace_pages = ftrace_pages_start;
1828
1829    cnt = num_to_init / ENTRIES_PER_PAGE;
1830    pr_info("ftrace: allocating %ld entries in %d pages\n",
1831        num_to_init, cnt + 1);
1832
1833    for (i = 0; i < cnt; i++) {
1834        pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1835
1836        /* If we fail, we'll try later anyway */
1837        if (!pg->next)
1838            break;
1839
1840        pg = pg->next;
1841    }
1842
1843    return 0;
1844}
1845
1846enum {
1847    FTRACE_ITER_FILTER = (1 << 0),
1848    FTRACE_ITER_NOTRACE = (1 << 1),
1849    FTRACE_ITER_PRINTALL = (1 << 2),
1850    FTRACE_ITER_HASH = (1 << 3),
1851    FTRACE_ITER_ENABLED = (1 << 4),
1852};
1853
1854#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1855
1856struct ftrace_iterator {
1857    loff_t pos;
1858    loff_t func_pos;
1859    struct ftrace_page *pg;
1860    struct dyn_ftrace *func;
1861    struct ftrace_func_probe *probe;
1862    struct trace_parser parser;
1863    struct ftrace_hash *hash;
1864    struct ftrace_ops *ops;
1865    int hidx;
1866    int idx;
1867    unsigned flags;
1868};
1869
1870static void *
1871t_hash_next(struct seq_file *m, loff_t *pos)
1872{
1873    struct ftrace_iterator *iter = m->private;
1874    struct hlist_node *hnd = NULL;
1875    struct hlist_head *hhd;
1876
1877    (*pos)++;
1878    iter->pos = *pos;
1879
1880    if (iter->probe)
1881        hnd = &iter->probe->node;
1882 retry:
1883    if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1884        return NULL;
1885
1886    hhd = &ftrace_func_hash[iter->hidx];
1887
1888    if (hlist_empty(hhd)) {
1889        iter->hidx++;
1890        hnd = NULL;
1891        goto retry;
1892    }
1893
1894    if (!hnd)
1895        hnd = hhd->first;
1896    else {
1897        hnd = hnd->next;
1898        if (!hnd) {
1899            iter->hidx++;
1900            goto retry;
1901        }
1902    }
1903
1904    if (WARN_ON_ONCE(!hnd))
1905        return NULL;
1906
1907    iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
1908
1909    return iter;
1910}
1911
1912static void *t_hash_start(struct seq_file *m, loff_t *pos)
1913{
1914    struct ftrace_iterator *iter = m->private;
1915    void *p = NULL;
1916    loff_t l;
1917
1918    if (iter->func_pos > *pos)
1919        return NULL;
1920
1921    iter->hidx = 0;
1922    for (l = 0; l <= (*pos - iter->func_pos); ) {
1923        p = t_hash_next(m, &l);
1924        if (!p)
1925            break;
1926    }
1927    if (!p)
1928        return NULL;
1929
1930    /* Only set this if we have an item */
1931    iter->flags |= FTRACE_ITER_HASH;
1932
1933    return iter;
1934}
1935
1936static int
1937t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
1938{
1939    struct ftrace_func_probe *rec;
1940
1941    rec = iter->probe;
1942    if (WARN_ON_ONCE(!rec))
1943        return -EIO;
1944
1945    if (rec->ops->print)
1946        return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1947
1948    seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
1949
1950    if (rec->data)
1951        seq_printf(m, ":%p", rec->data);
1952    seq_putc(m, '\n');
1953
1954    return 0;
1955}
1956
1957static void *
1958t_next(struct seq_file *m, void *v, loff_t *pos)
1959{
1960    struct ftrace_iterator *iter = m->private;
1961    struct ftrace_ops *ops = &global_ops;
1962    struct dyn_ftrace *rec = NULL;
1963
1964    if (unlikely(ftrace_disabled))
1965        return NULL;
1966
1967    if (iter->flags & FTRACE_ITER_HASH)
1968        return t_hash_next(m, pos);
1969
1970    (*pos)++;
1971    iter->pos = iter->func_pos = *pos;
1972
1973    if (iter->flags & FTRACE_ITER_PRINTALL)
1974        return t_hash_start(m, pos);
1975
1976 retry:
1977    if (iter->idx >= iter->pg->index) {
1978        if (iter->pg->next) {
1979            iter->pg = iter->pg->next;
1980            iter->idx = 0;
1981            goto retry;
1982        }
1983    } else {
1984        rec = &iter->pg->records[iter->idx++];
1985        if ((rec->flags & FTRACE_FL_FREE) ||
1986
1987            ((iter->flags & FTRACE_ITER_FILTER) &&
1988             !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
1989
1990            ((iter->flags & FTRACE_ITER_NOTRACE) &&
1991             !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
1992
1993            ((iter->flags & FTRACE_ITER_ENABLED) &&
1994             !(rec->flags & ~FTRACE_FL_MASK))) {
1995
1996            rec = NULL;
1997            goto retry;
1998        }
1999    }
2000
2001    if (!rec)
2002        return t_hash_start(m, pos);
2003
2004    iter->func = rec;
2005
2006    return iter;
2007}
2008
2009static void reset_iter_read(struct ftrace_iterator *iter)
2010{
2011    iter->pos = 0;
2012    iter->func_pos = 0;
2013    iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
2014}
2015
2016static void *t_start(struct seq_file *m, loff_t *pos)
2017{
2018    struct ftrace_iterator *iter = m->private;
2019    struct ftrace_ops *ops = &global_ops;
2020    void *p = NULL;
2021    loff_t l;
2022
2023    mutex_lock(&ftrace_lock);
2024
2025    if (unlikely(ftrace_disabled))
2026        return NULL;
2027
2028    /*
2029     * If an lseek was done, then reset and start from beginning.
2030     */
2031    if (*pos < iter->pos)
2032        reset_iter_read(iter);
2033
2034    /*
2035     * For set_ftrace_filter reading, if we have the filter
2036     * off, we can short cut and just print out that all
2037     * functions are enabled.
2038     */
2039    if (iter->flags & FTRACE_ITER_FILTER && !ops->filter_hash->count) {
2040        if (*pos > 0)
2041            return t_hash_start(m, pos);
2042        iter->flags |= FTRACE_ITER_PRINTALL;
2043        /* reset in case of seek/pread */
2044        iter->flags &= ~FTRACE_ITER_HASH;
2045        return iter;
2046    }
2047
2048    if (iter->flags & FTRACE_ITER_HASH)
2049        return t_hash_start(m, pos);
2050
2051    /*
2052     * Unfortunately, we need to restart at ftrace_pages_start
2053     * every time we let go of the ftrace_mutex. This is because
2054     * those pointers can change without the lock.
2055     */
2056    iter->pg = ftrace_pages_start;
2057    iter->idx = 0;
2058    for (l = 0; l <= *pos; ) {
2059        p = t_next(m, p, &l);
2060        if (!p)
2061            break;
2062    }
2063
2064    if (!p) {
2065        if (iter->flags & FTRACE_ITER_FILTER)
2066            return t_hash_start(m, pos);
2067
2068        return NULL;
2069    }
2070
2071    return iter;
2072}
2073
2074static void t_stop(struct seq_file *m, void *p)
2075{
2076    mutex_unlock(&ftrace_lock);
2077}
2078
2079static int t_show(struct seq_file *m, void *v)
2080{
2081    struct ftrace_iterator *iter = m->private;
2082    struct dyn_ftrace *rec;
2083
2084    if (iter->flags & FTRACE_ITER_HASH)
2085        return t_hash_show(m, iter);
2086
2087    if (iter->flags & FTRACE_ITER_PRINTALL) {
2088        seq_printf(m, "#### all functions enabled ####\n");
2089        return 0;
2090    }
2091
2092    rec = iter->func;
2093
2094    if (!rec)
2095        return 0;
2096
2097    seq_printf(m, "%ps", (void *)rec->ip);
2098    if (iter->flags & FTRACE_ITER_ENABLED)
2099        seq_printf(m, " (%ld)",
2100               rec->flags & ~FTRACE_FL_MASK);
2101    seq_printf(m, "\n");
2102
2103    return 0;
2104}
2105
2106static const struct seq_operations show_ftrace_seq_ops = {
2107    .start = t_start,
2108    .next = t_next,
2109    .stop = t_stop,
2110    .show = t_show,
2111};
2112
2113static int
2114ftrace_avail_open(struct inode *inode, struct file *file)
2115{
2116    struct ftrace_iterator *iter;
2117    int ret;
2118
2119    if (unlikely(ftrace_disabled))
2120        return -ENODEV;
2121
2122    iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2123    if (!iter)
2124        return -ENOMEM;
2125
2126    iter->pg = ftrace_pages_start;
2127
2128    ret = seq_open(file, &show_ftrace_seq_ops);
2129    if (!ret) {
2130        struct seq_file *m = file->private_data;
2131
2132        m->private = iter;
2133    } else {
2134        kfree(iter);
2135    }
2136
2137    return ret;
2138}
2139
2140static int
2141ftrace_enabled_open(struct inode *inode, struct file *file)
2142{
2143    struct ftrace_iterator *iter;
2144    int ret;
2145
2146    if (unlikely(ftrace_disabled))
2147        return -ENODEV;
2148
2149    iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2150    if (!iter)
2151        return -ENOMEM;
2152
2153    iter->pg = ftrace_pages_start;
2154    iter->flags = FTRACE_ITER_ENABLED;
2155
2156    ret = seq_open(file, &show_ftrace_seq_ops);
2157    if (!ret) {
2158        struct seq_file *m = file->private_data;
2159
2160        m->private = iter;
2161    } else {
2162        kfree(iter);
2163    }
2164
2165    return ret;
2166}
2167
2168static void ftrace_filter_reset(struct ftrace_hash *hash)
2169{
2170    mutex_lock(&ftrace_lock);
2171    ftrace_hash_clear(hash);
2172    mutex_unlock(&ftrace_lock);
2173}
2174
2175static int
2176ftrace_regex_open(struct ftrace_ops *ops, int flag,
2177          struct inode *inode, struct file *file)
2178{
2179    struct ftrace_iterator *iter;
2180    struct ftrace_hash *hash;
2181    int ret = 0;
2182
2183    if (unlikely(ftrace_disabled))
2184        return -ENODEV;
2185
2186    iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2187    if (!iter)
2188        return -ENOMEM;
2189
2190    if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2191        kfree(iter);
2192        return -ENOMEM;
2193    }
2194
2195    if (flag & FTRACE_ITER_NOTRACE)
2196        hash = ops->notrace_hash;
2197    else
2198        hash = ops->filter_hash;
2199
2200    iter->ops = ops;
2201    iter->flags = flag;
2202
2203    if (file->f_mode & FMODE_WRITE) {
2204        mutex_lock(&ftrace_lock);
2205        iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2206        mutex_unlock(&ftrace_lock);
2207
2208        if (!iter->hash) {
2209            trace_parser_put(&iter->parser);
2210            kfree(iter);
2211            return -ENOMEM;
2212        }
2213    }
2214
2215    mutex_lock(&ftrace_regex_lock);
2216
2217    if ((file->f_mode & FMODE_WRITE) &&
2218        (file->f_flags & O_TRUNC))
2219        ftrace_filter_reset(iter->hash);
2220
2221    if (file->f_mode & FMODE_READ) {
2222        iter->pg = ftrace_pages_start;
2223
2224        ret = seq_open(file, &show_ftrace_seq_ops);
2225        if (!ret) {
2226            struct seq_file *m = file->private_data;
2227            m->private = iter;
2228        } else {
2229            /* Failed */
2230            free_ftrace_hash(iter->hash);
2231            trace_parser_put(&iter->parser);
2232            kfree(iter);
2233        }
2234    } else
2235        file->private_data = iter;
2236    mutex_unlock(&ftrace_regex_lock);
2237
2238    return ret;
2239}
2240
2241static int
2242ftrace_filter_open(struct inode *inode, struct file *file)
2243{
2244    return ftrace_regex_open(&global_ops, FTRACE_ITER_FILTER,
2245                 inode, file);
2246}
2247
2248static int
2249ftrace_notrace_open(struct inode *inode, struct file *file)
2250{
2251    return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2252                 inode, file);
2253}
2254
2255static loff_t
2256ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
2257{
2258    loff_t ret;
2259
2260    if (file->f_mode & FMODE_READ)
2261        ret = seq_lseek(file, offset, origin);
2262    else
2263        file->f_pos = ret = 1;
2264
2265    return ret;
2266}
2267
2268static int ftrace_match(char *str, char *regex, int len, int type)
2269{
2270    int matched = 0;
2271    int slen;
2272
2273    switch (type) {
2274    case MATCH_FULL:
2275        if (strcmp(str, regex) == 0)
2276            matched = 1;
2277        break;
2278    case MATCH_FRONT_ONLY:
2279        if (strncmp(str, regex, len) == 0)
2280            matched = 1;
2281        break;
2282    case MATCH_MIDDLE_ONLY:
2283        if (strstr(str, regex))
2284            matched = 1;
2285        break;
2286    case MATCH_END_ONLY:
2287        slen = strlen(str);
2288        if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2289            matched = 1;
2290        break;
2291    }
2292
2293    return matched;
2294}
2295
2296static int
2297enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2298{
2299    struct ftrace_func_entry *entry;
2300    int ret = 0;
2301
2302    entry = ftrace_lookup_ip(hash, rec->ip);
2303    if (not) {
2304        /* Do nothing if it doesn't exist */
2305        if (!entry)
2306            return 0;
2307
2308        free_hash_entry(hash, entry);
2309    } else {
2310        /* Do nothing if it exists */
2311        if (entry)
2312            return 0;
2313
2314        ret = add_hash_entry(hash, rec->ip);
2315    }
2316    return ret;
2317}
2318
2319static int
2320ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2321            char *regex, int len, int type)
2322{
2323    char str[KSYM_SYMBOL_LEN];
2324    char *modname;
2325
2326    kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2327
2328    if (mod) {
2329        /* module lookup requires matching the module */
2330        if (!modname || strcmp(modname, mod))
2331            return 0;
2332
2333        /* blank search means to match all funcs in the mod */
2334        if (!len)
2335            return 1;
2336    }
2337
2338    return ftrace_match(str, regex, len, type);
2339}
2340
2341static int
2342match_records(struct ftrace_hash *hash, char *buff,
2343          int len, char *mod, int not)
2344{
2345    unsigned search_len = 0;
2346    struct ftrace_page *pg;
2347    struct dyn_ftrace *rec;
2348    int type = MATCH_FULL;
2349    char *search = buff;
2350    int found = 0;
2351    int ret;
2352
2353    if (len) {
2354        type = filter_parse_regex(buff, len, &search, &not);
2355        search_len = strlen(search);
2356    }
2357
2358    mutex_lock(&ftrace_lock);
2359
2360    if (unlikely(ftrace_disabled))
2361        goto out_unlock;
2362
2363    do_for_each_ftrace_rec(pg, rec) {
2364
2365        if (ftrace_match_record(rec, mod, search, search_len, type)) {
2366            ret = enter_record(hash, rec, not);
2367            if (ret < 0) {
2368                found = ret;
2369                goto out_unlock;
2370            }
2371            found = 1;
2372        }
2373    } while_for_each_ftrace_rec();
2374 out_unlock:
2375    mutex_unlock(&ftrace_lock);
2376
2377    return found;
2378}
2379
2380static int
2381ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2382{
2383    return match_records(hash, buff, len, NULL, 0);
2384}
2385
2386static int
2387ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2388{
2389    int not = 0;
2390
2391    /* blank or '*' mean the same */
2392    if (strcmp(buff, "*") == 0)
2393        buff[0] = 0;
2394
2395    /* handle the case of 'dont filter this module' */
2396    if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2397        buff[0] = 0;
2398        not = 1;
2399    }
2400
2401    return match_records(hash, buff, strlen(buff), mod, not);
2402}
2403
2404/*
2405 * We register the module command as a template to show others how
2406 * to register the a command as well.
2407 */
2408
2409static int
2410ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
2411{
2412    struct ftrace_ops *ops = &global_ops;
2413    struct ftrace_hash *hash;
2414    char *mod;
2415    int ret = -EINVAL;
2416
2417    /*
2418     * cmd == 'mod' because we only registered this func
2419     * for the 'mod' ftrace_func_command.
2420     * But if you register one func with multiple commands,
2421     * you can tell which command was used by the cmd
2422     * parameter.
2423     */
2424
2425    /* we must have a module name */
2426    if (!param)
2427        return ret;
2428
2429    mod = strsep(&param, ":");
2430    if (!strlen(mod))
2431        return ret;
2432
2433    if (enable)
2434        hash = ops->filter_hash;
2435    else
2436        hash = ops->notrace_hash;
2437
2438    ret = ftrace_match_module_records(hash, func, mod);
2439    if (!ret)
2440        ret = -EINVAL;
2441    if (ret < 0)
2442        return ret;
2443
2444    return 0;
2445}
2446
2447static struct ftrace_func_command ftrace_mod_cmd = {
2448    .name = "mod",
2449    .func = ftrace_mod_callback,
2450};
2451
2452static int __init ftrace_mod_cmd_init(void)
2453{
2454    return register_ftrace_command(&ftrace_mod_cmd);
2455}
2456device_initcall(ftrace_mod_cmd_init);
2457
2458static void
2459function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
2460{
2461    struct ftrace_func_probe *entry;
2462    struct hlist_head *hhd;
2463    struct hlist_node *n;
2464    unsigned long key;
2465
2466    key = hash_long(ip, FTRACE_HASH_BITS);
2467
2468    hhd = &ftrace_func_hash[key];
2469
2470    if (hlist_empty(hhd))
2471        return;
2472
2473    /*
2474     * Disable preemption for these calls to prevent a RCU grace
2475     * period. This syncs the hash iteration and freeing of items
2476     * on the hash. rcu_read_lock is too dangerous here.
2477     */
2478    preempt_disable_notrace();
2479    hlist_for_each_entry_rcu(entry, n, hhd, node) {
2480        if (entry->ip == ip)
2481            entry->ops->func(ip, parent_ip, &entry->data);
2482    }
2483    preempt_enable_notrace();
2484}
2485
2486static struct ftrace_ops trace_probe_ops __read_mostly =
2487{
2488    .func = function_trace_probe_call,
2489};
2490
2491static int ftrace_probe_registered;
2492
2493static void __enable_ftrace_function_probe(void)
2494{
2495    int ret;
2496    int i;
2497
2498    if (ftrace_probe_registered)
2499        return;
2500
2501    for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2502        struct hlist_head *hhd = &ftrace_func_hash[i];
2503        if (hhd->first)
2504            break;
2505    }
2506    /* Nothing registered? */
2507    if (i == FTRACE_FUNC_HASHSIZE)
2508        return;
2509
2510    ret = __register_ftrace_function(&trace_probe_ops);
2511    if (!ret)
2512        ret = ftrace_startup(&trace_probe_ops, 0);
2513
2514    ftrace_probe_registered = 1;
2515}
2516
2517static void __disable_ftrace_function_probe(void)
2518{
2519    int ret;
2520    int i;
2521
2522    if (!ftrace_probe_registered)
2523        return;
2524
2525    for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2526        struct hlist_head *hhd = &ftrace_func_hash[i];
2527        if (hhd->first)
2528            return;
2529    }
2530
2531    /* no more funcs left */
2532    ret = __unregister_ftrace_function(&trace_probe_ops);
2533    if (!ret)
2534        ftrace_shutdown(&trace_probe_ops, 0);
2535
2536    ftrace_probe_registered = 0;
2537}
2538
2539
2540static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2541{
2542    struct ftrace_func_probe *entry =
2543        container_of(rhp, struct ftrace_func_probe, rcu);
2544
2545    if (entry->ops->free)
2546        entry->ops->free(&entry->data);
2547    kfree(entry);
2548}
2549
2550
2551int
2552register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2553                  void *data)
2554{
2555    struct ftrace_func_probe *entry;
2556    struct ftrace_page *pg;
2557    struct dyn_ftrace *rec;
2558    int type, len, not;
2559    unsigned long key;
2560    int count = 0;
2561    char *search;
2562
2563    type = filter_parse_regex(glob, strlen(glob), &search, &not);
2564    len = strlen(search);
2565
2566    /* we do not support '!' for function probes */
2567    if (WARN_ON(not))
2568        return -EINVAL;
2569
2570    mutex_lock(&ftrace_lock);
2571
2572    if (unlikely(ftrace_disabled))
2573        goto out_unlock;
2574
2575    do_for_each_ftrace_rec(pg, rec) {
2576
2577        if (!ftrace_match_record(rec, NULL, search, len, type))
2578            continue;
2579
2580        entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2581        if (!entry) {
2582            /* If we did not process any, then return error */
2583            if (!count)
2584                count = -ENOMEM;
2585            goto out_unlock;
2586        }
2587
2588        count++;
2589
2590        entry->data = data;
2591
2592        /*
2593         * The caller might want to do something special
2594         * for each function we find. We call the callback
2595         * to give the caller an opportunity to do so.
2596         */
2597        if (ops->callback) {
2598            if (ops->callback(rec->ip, &entry->data) < 0) {
2599                /* caller does not like this func */
2600                kfree(entry);
2601                continue;
2602            }
2603        }
2604
2605        entry->ops = ops;
2606        entry->ip = rec->ip;
2607
2608        key = hash_long(entry->ip, FTRACE_HASH_BITS);
2609        hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2610
2611    } while_for_each_ftrace_rec();
2612    __enable_ftrace_function_probe();
2613
2614 out_unlock:
2615    mutex_unlock(&ftrace_lock);
2616
2617    return count;
2618}
2619
2620enum {
2621    PROBE_TEST_FUNC = 1,
2622    PROBE_TEST_DATA = 2
2623};
2624
2625static void
2626__unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2627                  void *data, int flags)
2628{
2629    struct ftrace_func_probe *entry;
2630    struct hlist_node *n, *tmp;
2631    char str[KSYM_SYMBOL_LEN];
2632    int type = MATCH_FULL;
2633    int i, len = 0;
2634    char *search;
2635
2636    if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2637        glob = NULL;
2638    else if (glob) {
2639        int not;
2640
2641        type = filter_parse_regex(glob, strlen(glob), &search, &not);
2642        len = strlen(search);
2643
2644        /* we do not support '!' for function probes */
2645        if (WARN_ON(not))
2646            return;
2647    }
2648
2649    mutex_lock(&ftrace_lock);
2650    for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2651        struct hlist_head *hhd = &ftrace_func_hash[i];
2652
2653        hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2654
2655            /* break up if statements for readability */
2656            if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2657                continue;
2658
2659            if ((flags & PROBE_TEST_DATA) && entry->data != data)
2660                continue;
2661
2662            /* do this last, since it is the most expensive */
2663            if (glob) {
2664                kallsyms_lookup(entry->ip, NULL, NULL,
2665                        NULL, str);
2666                if (!ftrace_match(str, glob, len, type))
2667                    continue;
2668            }
2669
2670            hlist_del(&entry->node);
2671            call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2672        }
2673    }
2674    __disable_ftrace_function_probe();
2675    mutex_unlock(&ftrace_lock);
2676}
2677
2678void
2679unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2680                void *data)
2681{
2682    __unregister_ftrace_function_probe(glob, ops, data,
2683                      PROBE_TEST_FUNC | PROBE_TEST_DATA);
2684}
2685
2686void
2687unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2688{
2689    __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2690}
2691
2692void unregister_ftrace_function_probe_all(char *glob)
2693{
2694    __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2695}
2696
2697static LIST_HEAD(ftrace_commands);
2698static DEFINE_MUTEX(ftrace_cmd_mutex);
2699
2700int register_ftrace_command(struct ftrace_func_command *cmd)
2701{
2702    struct ftrace_func_command *p;
2703    int ret = 0;
2704
2705    mutex_lock(&ftrace_cmd_mutex);
2706    list_for_each_entry(p, &ftrace_commands, list) {
2707        if (strcmp(cmd->name, p->name) == 0) {
2708            ret = -EBUSY;
2709            goto out_unlock;
2710        }
2711    }
2712    list_add(&cmd->list, &ftrace_commands);
2713 out_unlock:
2714    mutex_unlock(&ftrace_cmd_mutex);
2715
2716    return ret;
2717}
2718
2719int unregister_ftrace_command(struct ftrace_func_command *cmd)
2720{
2721    struct ftrace_func_command *p, *n;
2722    int ret = -ENODEV;
2723
2724    mutex_lock(&ftrace_cmd_mutex);
2725    list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2726        if (strcmp(cmd->name, p->name) == 0) {
2727            ret = 0;
2728            list_del_init(&p->list);
2729            goto out_unlock;
2730        }
2731    }
2732 out_unlock:
2733    mutex_unlock(&ftrace_cmd_mutex);
2734
2735    return ret;
2736}
2737
2738static int ftrace_process_regex(struct ftrace_hash *hash,
2739                char *buff, int len, int enable)
2740{
2741    char *func, *command, *next = buff;
2742    struct ftrace_func_command *p;
2743    int ret = -EINVAL;
2744
2745    func = strsep(&next, ":");
2746
2747    if (!next) {
2748        ret = ftrace_match_records(hash, func, len);
2749        if (!ret)
2750            ret = -EINVAL;
2751        if (ret < 0)
2752            return ret;
2753        return 0;
2754    }
2755
2756    /* command found */
2757
2758    command = strsep(&next, ":");
2759
2760    mutex_lock(&ftrace_cmd_mutex);
2761    list_for_each_entry(p, &ftrace_commands, list) {
2762        if (strcmp(p->name, command) == 0) {
2763            ret = p->func(func, command, next, enable);
2764            goto out_unlock;
2765        }
2766    }
2767 out_unlock:
2768    mutex_unlock(&ftrace_cmd_mutex);
2769
2770    return ret;
2771}
2772
2773static ssize_t
2774ftrace_regex_write(struct file *file, const char __user *ubuf,
2775           size_t cnt, loff_t *ppos, int enable)
2776{
2777    struct ftrace_iterator *iter;
2778    struct trace_parser *parser;
2779    ssize_t ret, read;
2780
2781    if (!cnt)
2782        return 0;
2783
2784    mutex_lock(&ftrace_regex_lock);
2785
2786    ret = -ENODEV;
2787    if (unlikely(ftrace_disabled))
2788        goto out_unlock;
2789
2790    if (file->f_mode & FMODE_READ) {
2791        struct seq_file *m = file->private_data;
2792        iter = m->private;
2793    } else
2794        iter = file->private_data;
2795
2796    parser = &iter->parser;
2797    read = trace_get_user(parser, ubuf, cnt, ppos);
2798
2799    if (read >= 0 && trace_parser_loaded(parser) &&
2800        !trace_parser_cont(parser)) {
2801        ret = ftrace_process_regex(iter->hash, parser->buffer,
2802                       parser->idx, enable);
2803        trace_parser_clear(parser);
2804        if (ret)
2805            goto out_unlock;
2806    }
2807
2808    ret = read;
2809out_unlock:
2810    mutex_unlock(&ftrace_regex_lock);
2811
2812    return ret;
2813}
2814
2815static ssize_t
2816ftrace_filter_write(struct file *file, const char __user *ubuf,
2817            size_t cnt, loff_t *ppos)
2818{
2819    return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2820}
2821
2822static ssize_t
2823ftrace_notrace_write(struct file *file, const char __user *ubuf,
2824             size_t cnt, loff_t *ppos)
2825{
2826    return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2827}
2828
2829static int
2830ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
2831         int reset, int enable)
2832{
2833    struct ftrace_hash **orig_hash;
2834    struct ftrace_hash *hash;
2835    int ret;
2836
2837    /* All global ops uses the global ops filters */
2838    if (ops->flags & FTRACE_OPS_FL_GLOBAL)
2839        ops = &global_ops;
2840
2841    if (unlikely(ftrace_disabled))
2842        return -ENODEV;
2843
2844    if (enable)
2845        orig_hash = &ops->filter_hash;
2846    else
2847        orig_hash = &ops->notrace_hash;
2848
2849    hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
2850    if (!hash)
2851        return -ENOMEM;
2852
2853    mutex_lock(&ftrace_regex_lock);
2854    if (reset)
2855        ftrace_filter_reset(hash);
2856    if (buf)
2857        ftrace_match_records(hash, buf, len);
2858
2859    mutex_lock(&ftrace_lock);
2860    ret = ftrace_hash_move(orig_hash, hash);
2861    mutex_unlock(&ftrace_lock);
2862
2863    mutex_unlock(&ftrace_regex_lock);
2864
2865    free_ftrace_hash(hash);
2866    return ret;
2867}
2868
2869/**
2870 * ftrace_set_filter - set a function to filter on in ftrace
2871 * @ops - the ops to set the filter with
2872 * @buf - the string that holds the function filter text.
2873 * @len - the length of the string.
2874 * @reset - non zero to reset all filters before applying this filter.
2875 *
2876 * Filters denote which functions should be enabled when tracing is enabled.
2877 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2878 */
2879void ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
2880               int len, int reset)
2881{
2882    ftrace_set_regex(ops, buf, len, reset, 1);
2883}
2884EXPORT_SYMBOL_GPL(ftrace_set_filter);
2885
2886/**
2887 * ftrace_set_notrace - set a function to not trace in ftrace
2888 * @ops - the ops to set the notrace filter with
2889 * @buf - the string that holds the function notrace text.
2890 * @len - the length of the string.
2891 * @reset - non zero to reset all filters before applying this filter.
2892 *
2893 * Notrace Filters denote which functions should not be enabled when tracing
2894 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2895 * for tracing.
2896 */
2897void ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
2898            int len, int reset)
2899{
2900    ftrace_set_regex(ops, buf, len, reset, 0);
2901}
2902EXPORT_SYMBOL_GPL(ftrace_set_notrace);
2903/**
2904 * ftrace_set_filter - set a function to filter on in ftrace
2905 * @ops - the ops to set the filter with
2906 * @buf - the string that holds the function filter text.
2907 * @len - the length of the string.
2908 * @reset - non zero to reset all filters before applying this filter.
2909 *
2910 * Filters denote which functions should be enabled when tracing is enabled.
2911 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2912 */
2913void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
2914{
2915    ftrace_set_regex(&global_ops, buf, len, reset, 1);
2916}
2917EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
2918
2919/**
2920 * ftrace_set_notrace - set a function to not trace in ftrace
2921 * @ops - the ops to set the notrace filter with
2922 * @buf - the string that holds the function notrace text.
2923 * @len - the length of the string.
2924 * @reset - non zero to reset all filters before applying this filter.
2925 *
2926 * Notrace Filters denote which functions should not be enabled when tracing
2927 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2928 * for tracing.
2929 */
2930void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
2931{
2932    ftrace_set_regex(&global_ops, buf, len, reset, 0);
2933}
2934EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
2935
2936/*
2937 * command line interface to allow users to set filters on boot up.
2938 */
2939#define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE
2940static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2941static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2942
2943static int __init set_ftrace_notrace(char *str)
2944{
2945    strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2946    return 1;
2947}
2948__setup("ftrace_notrace=", set_ftrace_notrace);
2949
2950static int __init set_ftrace_filter(char *str)
2951{
2952    strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2953    return 1;
2954}
2955__setup("ftrace_filter=", set_ftrace_filter);
2956
2957#ifdef CONFIG_FUNCTION_GRAPH_TRACER
2958static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
2959static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
2960
2961static int __init set_graph_function(char *str)
2962{
2963    strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
2964    return 1;
2965}
2966__setup("ftrace_graph_filter=", set_graph_function);
2967
2968static void __init set_ftrace_early_graph(char *buf)
2969{
2970    int ret;
2971    char *func;
2972
2973    while (buf) {
2974        func = strsep(&buf, ",");
2975        /* we allow only one expression at a time */
2976        ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2977                      func);
2978        if (ret)
2979            printk(KERN_DEBUG "ftrace: function %s not "
2980                      "traceable\n", func);
2981    }
2982}
2983#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2984
2985static void __init
2986set_ftrace_early_filter(struct ftrace_ops *ops, char *buf, int enable)
2987{
2988    char *func;
2989
2990    while (buf) {
2991        func = strsep(&buf, ",");
2992        ftrace_set_regex(ops, func, strlen(func), 0, enable);
2993    }
2994}
2995
2996static void __init set_ftrace_early_filters(void)
2997{
2998    if (ftrace_filter_buf[0])
2999        set_ftrace_early_filter(&global_ops, ftrace_filter_buf, 1);
3000    if (ftrace_notrace_buf[0])
3001        set_ftrace_early_filter(&global_ops, ftrace_notrace_buf, 0);
3002#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3003    if (ftrace_graph_buf[0])
3004        set_ftrace_early_graph(ftrace_graph_buf);
3005#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3006}
3007
3008static int
3009ftrace_regex_release(struct inode *inode, struct file *file)
3010{
3011    struct seq_file *m = (struct seq_file *)file->private_data;
3012    struct ftrace_iterator *iter;
3013    struct ftrace_hash **orig_hash;
3014    struct trace_parser *parser;
3015    int filter_hash;
3016    int ret;
3017
3018    mutex_lock(&ftrace_regex_lock);
3019    if (file->f_mode & FMODE_READ) {
3020        iter = m->private;
3021
3022        seq_release(inode, file);
3023    } else
3024        iter = file->private_data;
3025
3026    parser = &iter->parser;
3027    if (trace_parser_loaded(parser)) {
3028        parser->buffer[parser->idx] = 0;
3029        ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3030    }
3031
3032    trace_parser_put(parser);
3033
3034    if (file->f_mode & FMODE_WRITE) {
3035        filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3036
3037        if (filter_hash)
3038            orig_hash = &iter->ops->filter_hash;
3039        else
3040            orig_hash = &iter->ops->notrace_hash;
3041
3042        mutex_lock(&ftrace_lock);
3043        /*
3044         * Remove the current set, update the hash and add
3045         * them back.
3046         */
3047        ftrace_hash_rec_disable(iter->ops, filter_hash);
3048        ret = ftrace_hash_move(orig_hash, iter->hash);
3049        if (!ret) {
3050            ftrace_hash_rec_enable(iter->ops, filter_hash);
3051            if (iter->ops->flags & FTRACE_OPS_FL_ENABLED
3052                && ftrace_enabled)
3053                ftrace_run_update_code(FTRACE_ENABLE_CALLS);
3054        }
3055        mutex_unlock(&ftrace_lock);
3056    }
3057    free_ftrace_hash(iter->hash);
3058    kfree(iter);
3059
3060    mutex_unlock(&ftrace_regex_lock);
3061    return 0;
3062}
3063
3064static const struct file_operations ftrace_avail_fops = {
3065    .open = ftrace_avail_open,
3066    .read = seq_read,
3067    .llseek = seq_lseek,
3068    .release = seq_release_private,
3069};
3070
3071static const struct file_operations ftrace_enabled_fops = {
3072    .open = ftrace_enabled_open,
3073    .read = seq_read,
3074    .llseek = seq_lseek,
3075    .release = seq_release_private,
3076};
3077
3078static const struct file_operations ftrace_filter_fops = {
3079    .open = ftrace_filter_open,
3080    .read = seq_read,
3081    .write = ftrace_filter_write,
3082    .llseek = ftrace_regex_lseek,
3083    .release = ftrace_regex_release,
3084};
3085
3086static const struct file_operations ftrace_notrace_fops = {
3087    .open = ftrace_notrace_open,
3088    .read = seq_read,
3089    .write = ftrace_notrace_write,
3090    .llseek = ftrace_regex_lseek,
3091    .release = ftrace_regex_release,
3092};
3093
3094#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3095
3096static DEFINE_MUTEX(graph_lock);
3097
3098int ftrace_graph_count;
3099int ftrace_graph_filter_enabled;
3100unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3101
3102static void *
3103__g_next(struct seq_file *m, loff_t *pos)
3104{
3105    if (*pos >= ftrace_graph_count)
3106        return NULL;
3107    return &ftrace_graph_funcs[*pos];
3108}
3109
3110static void *
3111g_next(struct seq_file *m, void *v, loff_t *pos)
3112{
3113    (*pos)++;
3114    return __g_next(m, pos);
3115}
3116
3117static void *g_start(struct seq_file *m, loff_t *pos)
3118{
3119    mutex_lock(&graph_lock);
3120
3121    /* Nothing, tell g_show to print all functions are enabled */
3122    if (!ftrace_graph_filter_enabled && !*pos)
3123        return (void *)1;
3124
3125    return __g_next(m, pos);
3126}
3127
3128static void g_stop(struct seq_file *m, void *p)
3129{
3130    mutex_unlock(&graph_lock);
3131}
3132
3133static int g_show(struct seq_file *m, void *v)
3134{
3135    unsigned long *ptr = v;
3136
3137    if (!ptr)
3138        return 0;
3139
3140    if (ptr == (unsigned long *)1) {
3141        seq_printf(m, "#### all functions enabled ####\n");
3142        return 0;
3143    }
3144
3145    seq_printf(m, "%ps\n", (void *)*ptr);
3146
3147    return 0;
3148}
3149
3150static const struct seq_operations ftrace_graph_seq_ops = {
3151    .start = g_start,
3152    .next = g_next,
3153    .stop = g_stop,
3154    .show = g_show,
3155};
3156
3157static int
3158ftrace_graph_open(struct inode *inode, struct file *file)
3159{
3160    int ret = 0;
3161
3162    if (unlikely(ftrace_disabled))
3163        return -ENODEV;
3164
3165    mutex_lock(&graph_lock);
3166    if ((file->f_mode & FMODE_WRITE) &&
3167        (file->f_flags & O_TRUNC)) {
3168        ftrace_graph_filter_enabled = 0;
3169        ftrace_graph_count = 0;
3170        memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
3171    }
3172    mutex_unlock(&graph_lock);
3173
3174    if (file->f_mode & FMODE_READ)
3175        ret = seq_open(file, &ftrace_graph_seq_ops);
3176
3177    return ret;
3178}
3179
3180static int
3181ftrace_graph_release(struct inode *inode, struct file *file)
3182{
3183    if (file->f_mode & FMODE_READ)
3184        seq_release(inode, file);
3185    return 0;
3186}
3187
3188static int
3189ftrace_set_func(unsigned long *array, int *idx, char *buffer)
3190{
3191    struct dyn_ftrace *rec;
3192    struct ftrace_page *pg;
3193    int search_len;
3194    int fail = 1;
3195    int type, not;
3196    char *search;
3197    bool exists;
3198    int i;
3199
3200    /* decode regex */
3201    type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
3202    if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
3203        return -EBUSY;
3204
3205    search_len = strlen(search);
3206
3207    mutex_lock(&ftrace_lock);
3208
3209    if (unlikely(ftrace_disabled)) {
3210        mutex_unlock(&ftrace_lock);
3211        return -ENODEV;
3212    }
3213
3214    do_for_each_ftrace_rec(pg, rec) {
3215
3216        if (rec->flags & FTRACE_FL_FREE)
3217            continue;
3218
3219        if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3220            /* if it is in the array */
3221            exists = false;
3222            for (i = 0; i < *idx; i++) {
3223                if (array[i] == rec->ip) {
3224                    exists = true;
3225                    break;
3226                }
3227            }
3228
3229            if (!not) {
3230                fail = 0;
3231                if (!exists) {
3232                    array[(*idx)++] = rec->ip;
3233                    if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
3234                        goto out;
3235                }
3236            } else {
3237                if (exists) {
3238                    array[i] = array[--(*idx)];
3239                    array[*idx] = 0;
3240                    fail = 0;
3241                }
3242            }
3243        }
3244    } while_for_each_ftrace_rec();
3245out:
3246    mutex_unlock(&ftrace_lock);
3247
3248    if (fail)
3249        return -EINVAL;
3250
3251    ftrace_graph_filter_enabled = 1;
3252    return 0;
3253}
3254
3255static ssize_t
3256ftrace_graph_write(struct file *file, const char __user *ubuf,
3257           size_t cnt, loff_t *ppos)
3258{
3259    struct trace_parser parser;
3260    ssize_t read, ret;
3261
3262    if (!cnt)
3263        return 0;
3264
3265    mutex_lock(&graph_lock);
3266
3267    if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
3268        ret = -ENOMEM;
3269        goto out_unlock;
3270    }
3271
3272    read = trace_get_user(&parser, ubuf, cnt, ppos);
3273
3274    if (read >= 0 && trace_parser_loaded((&parser))) {
3275        parser.buffer[parser.idx] = 0;
3276
3277        /* we allow only one expression at a time */
3278        ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3279                    parser.buffer);
3280        if (ret)
3281            goto out_free;
3282    }
3283
3284    ret = read;
3285
3286out_free:
3287    trace_parser_put(&parser);
3288out_unlock:
3289    mutex_unlock(&graph_lock);
3290
3291    return ret;
3292}
3293
3294static const struct file_operations ftrace_graph_fops = {
3295    .open = ftrace_graph_open,
3296    .read = seq_read,
3297    .write = ftrace_graph_write,
3298    .release = ftrace_graph_release,
3299    .llseek = seq_lseek,
3300};
3301#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3302
3303static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
3304{
3305
3306    trace_create_file("available_filter_functions", 0444,
3307            d_tracer, NULL, &ftrace_avail_fops);
3308
3309    trace_create_file("enabled_functions", 0444,
3310            d_tracer, NULL, &ftrace_enabled_fops);
3311
3312    trace_create_file("set_ftrace_filter", 0644, d_tracer,
3313            NULL, &ftrace_filter_fops);
3314
3315    trace_create_file("set_ftrace_notrace", 0644, d_tracer,
3316                    NULL, &ftrace_notrace_fops);
3317
3318#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3319    trace_create_file("set_graph_function", 0444, d_tracer,
3320                    NULL,
3321                    &ftrace_graph_fops);
3322#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3323
3324    return 0;
3325}
3326
3327static int ftrace_process_locs(struct module *mod,
3328                   unsigned long *start,
3329                   unsigned long *end)
3330{
3331    unsigned long *p;
3332    unsigned long addr;
3333    unsigned long flags;
3334
3335    mutex_lock(&ftrace_lock);
3336    p = start;
3337    while (p < end) {
3338        addr = ftrace_call_adjust(*p++);
3339        /*
3340         * Some architecture linkers will pad between
3341         * the different mcount_loc sections of different
3342         * object files to satisfy alignments.
3343         * Skip any NULL pointers.
3344         */
3345        if (!addr)
3346            continue;
3347        ftrace_record_ip(addr);
3348    }
3349
3350    /*
3351     * Disable interrupts to prevent interrupts from executing
3352     * code that is being modified.
3353     */
3354    local_irq_save(flags);
3355    ftrace_update_code(mod);
3356    local_irq_restore(flags);
3357    mutex_unlock(&ftrace_lock);
3358
3359    return 0;
3360}
3361
3362#ifdef CONFIG_MODULES
3363void ftrace_release_mod(struct module *mod)
3364{
3365    struct dyn_ftrace *rec;
3366    struct ftrace_page *pg;
3367
3368    mutex_lock(&ftrace_lock);
3369
3370    if (ftrace_disabled)
3371        goto out_unlock;
3372
3373    do_for_each_ftrace_rec(pg, rec) {
3374        if (within_module_core(rec->ip, mod)) {
3375            /*
3376             * rec->ip is changed in ftrace_free_rec()
3377             * It should not between s and e if record was freed.
3378             */
3379            FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
3380            ftrace_free_rec(rec);
3381        }
3382    } while_for_each_ftrace_rec();
3383 out_unlock:
3384    mutex_unlock(&ftrace_lock);
3385}
3386
3387static void ftrace_init_module(struct module *mod,
3388                   unsigned long *start, unsigned long *end)
3389{
3390    if (ftrace_disabled || start == end)
3391        return;
3392    ftrace_process_locs(mod, start, end);
3393}
3394
3395static int ftrace_module_notify(struct notifier_block *self,
3396                unsigned long val, void *data)
3397{
3398    struct module *mod = data;
3399
3400    switch (val) {
3401    case MODULE_STATE_COMING:
3402        ftrace_init_module(mod, mod->ftrace_callsites,
3403                   mod->ftrace_callsites +
3404                   mod->num_ftrace_callsites);
3405        break;
3406    case MODULE_STATE_GOING:
3407        ftrace_release_mod(mod);
3408        break;
3409    }
3410
3411    return 0;
3412}
3413#else
3414static int ftrace_module_notify(struct notifier_block *self,
3415                unsigned long val, void *data)
3416{
3417    return 0;
3418}
3419#endif /* CONFIG_MODULES */
3420
3421struct notifier_block ftrace_module_nb = {
3422    .notifier_call = ftrace_module_notify,
3423    .priority = 0,
3424};
3425
3426extern unsigned long __start_mcount_loc[];
3427extern unsigned long __stop_mcount_loc[];
3428
3429void __init ftrace_init(void)
3430{
3431    unsigned long count, addr, flags;
3432    int ret;
3433
3434    /* Keep the ftrace pointer to the stub */
3435    addr = (unsigned long)ftrace_stub;
3436
3437    local_irq_save(flags);
3438    ftrace_dyn_arch_init(&addr);
3439    local_irq_restore(flags);
3440
3441    /* ftrace_dyn_arch_init places the return code in addr */
3442    if (addr)
3443        goto failed;
3444
3445    count = __stop_mcount_loc - __start_mcount_loc;
3446
3447    ret = ftrace_dyn_table_alloc(count);
3448    if (ret)
3449        goto failed;
3450
3451    last_ftrace_enabled = ftrace_enabled = 1;
3452
3453    ret = ftrace_process_locs(NULL,
3454                  __start_mcount_loc,
3455                  __stop_mcount_loc);
3456
3457    ret = register_module_notifier(&ftrace_module_nb);
3458    if (ret)
3459        pr_warning("Failed to register trace ftrace module notifier\n");
3460
3461    set_ftrace_early_filters();
3462
3463    return;
3464 failed:
3465    ftrace_disabled = 1;
3466}
3467
3468#else
3469
3470static struct ftrace_ops global_ops = {
3471    .func = ftrace_stub,
3472};
3473
3474static int __init ftrace_nodyn_init(void)
3475{
3476    ftrace_enabled = 1;
3477    return 0;
3478}
3479device_initcall(ftrace_nodyn_init);
3480
3481static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
3482static inline void ftrace_startup_enable(int command) { }
3483/* Keep as macros so we do not need to define the commands */
3484# define ftrace_startup(ops, command) \
3485    ({ \
3486        (ops)->flags |= FTRACE_OPS_FL_ENABLED; \
3487        0; \
3488    })
3489# define ftrace_shutdown(ops, command) do { } while (0)
3490# define ftrace_startup_sysctl() do { } while (0)
3491# define ftrace_shutdown_sysctl() do { } while (0)
3492
3493static inline int
3494ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
3495{
3496    return 1;
3497}
3498
3499#endif /* CONFIG_DYNAMIC_FTRACE */
3500
3501static void
3502ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
3503{
3504    struct ftrace_ops *op;
3505
3506    if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT)))
3507        return;
3508
3509    trace_recursion_set(TRACE_INTERNAL_BIT);
3510    /*
3511     * Some of the ops may be dynamically allocated,
3512     * they must be freed after a synchronize_sched().
3513     */
3514    preempt_disable_notrace();
3515    op = rcu_dereference_raw(ftrace_ops_list);
3516    while (op != &ftrace_list_end) {
3517        if (ftrace_ops_test(op, ip))
3518            op->func(ip, parent_ip);
3519        op = rcu_dereference_raw(op->next);
3520    };
3521    preempt_enable_notrace();
3522    trace_recursion_clear(TRACE_INTERNAL_BIT);
3523}
3524
3525static void clear_ftrace_swapper(void)
3526{
3527    struct task_struct *p;
3528    int cpu;
3529
3530    get_online_cpus();
3531    for_each_online_cpu(cpu) {
3532        p = idle_task(cpu);
3533        clear_tsk_trace_trace(p);
3534    }
3535    put_online_cpus();
3536}
3537
3538static void set_ftrace_swapper(void)
3539{
3540    struct task_struct *p;
3541    int cpu;
3542
3543    get_online_cpus();
3544    for_each_online_cpu(cpu) {
3545        p = idle_task(cpu);
3546        set_tsk_trace_trace(p);
3547    }
3548    put_online_cpus();
3549}
3550
3551static void clear_ftrace_pid(struct pid *pid)
3552{
3553    struct task_struct *p;
3554
3555    rcu_read_lock();
3556    do_each_pid_task(pid, PIDTYPE_PID, p) {
3557        clear_tsk_trace_trace(p);
3558    } while_each_pid_task(pid, PIDTYPE_PID, p);
3559    rcu_read_unlock();
3560
3561    put_pid(pid);
3562}
3563
3564static void set_ftrace_pid(struct pid *pid)
3565{
3566    struct task_struct *p;
3567
3568    rcu_read_lock();
3569    do_each_pid_task(pid, PIDTYPE_PID, p) {
3570        set_tsk_trace_trace(p);
3571    } while_each_pid_task(pid, PIDTYPE_PID, p);
3572    rcu_read_unlock();
3573}
3574
3575static void clear_ftrace_pid_task(struct pid *pid)
3576{
3577    if (pid == ftrace_swapper_pid)
3578        clear_ftrace_swapper();
3579    else
3580        clear_ftrace_pid(pid);
3581}
3582
3583static void set_ftrace_pid_task(struct pid *pid)
3584{
3585    if (pid == ftrace_swapper_pid)
3586        set_ftrace_swapper();
3587    else
3588        set_ftrace_pid(pid);
3589}
3590
3591static int ftrace_pid_add(int p)
3592{
3593    struct pid *pid;
3594    struct ftrace_pid *fpid;
3595    int ret = -EINVAL;
3596
3597    mutex_lock(&ftrace_lock);
3598
3599    if (!p)
3600        pid = ftrace_swapper_pid;
3601    else
3602        pid = find_get_pid(p);
3603
3604    if (!pid)
3605        goto out;
3606
3607    ret = 0;
3608
3609    list_for_each_entry(fpid, &ftrace_pids, list)
3610        if (fpid->pid == pid)
3611            goto out_put;
3612
3613    ret = -ENOMEM;
3614
3615    fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
3616    if (!fpid)
3617        goto out_put;
3618
3619    list_add(&fpid->list, &ftrace_pids);
3620    fpid->pid = pid;
3621
3622    set_ftrace_pid_task(pid);
3623
3624    ftrace_update_pid_func();
3625    ftrace_startup_enable(0);
3626
3627    mutex_unlock(&ftrace_lock);
3628    return 0;
3629
3630out_put:
3631    if (pid != ftrace_swapper_pid)
3632        put_pid(pid);
3633
3634out:
3635    mutex_unlock(&ftrace_lock);
3636    return ret;
3637}
3638
3639static void ftrace_pid_reset(void)
3640{
3641    struct ftrace_pid *fpid, *safe;
3642
3643    mutex_lock(&ftrace_lock);
3644    list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
3645        struct pid *pid = fpid->pid;
3646
3647        clear_ftrace_pid_task(pid);
3648
3649        list_del(&fpid->list);
3650        kfree(fpid);
3651    }
3652
3653    ftrace_update_pid_func();
3654    ftrace_startup_enable(0);
3655
3656    mutex_unlock(&ftrace_lock);
3657}
3658
3659static void *fpid_start(struct seq_file *m, loff_t *pos)
3660{
3661    mutex_lock(&ftrace_lock);
3662
3663    if (list_empty(&ftrace_pids) && (!*pos))
3664        return (void *) 1;
3665
3666    return seq_list_start(&ftrace_pids, *pos);
3667}
3668
3669static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
3670{
3671    if (v == (void *)1)
3672        return NULL;
3673
3674    return seq_list_next(v, &ftrace_pids, pos);
3675}
3676
3677static void fpid_stop(struct seq_file *m, void *p)
3678{
3679    mutex_unlock(&ftrace_lock);
3680}
3681
3682static int fpid_show(struct seq_file *m, void *v)
3683{
3684    const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
3685
3686    if (v == (void *)1) {
3687        seq_printf(m, "no pid\n");
3688        return 0;
3689    }
3690
3691    if (fpid->pid == ftrace_swapper_pid)
3692        seq_printf(m, "swapper tasks\n");
3693    else
3694        seq_printf(m, "%u\n", pid_vnr(fpid->pid));
3695
3696    return 0;
3697}
3698
3699static const struct seq_operations ftrace_pid_sops = {
3700    .start = fpid_start,
3701    .next = fpid_next,
3702    .stop = fpid_stop,
3703    .show = fpid_show,
3704};
3705
3706static int
3707ftrace_pid_open(struct inode *inode, struct file *file)
3708{
3709    int ret = 0;
3710
3711    if ((file->f_mode & FMODE_WRITE) &&
3712        (file->f_flags & O_TRUNC))
3713        ftrace_pid_reset();
3714
3715    if (file->f_mode & FMODE_READ)
3716        ret = seq_open(file, &ftrace_pid_sops);
3717
3718    return ret;
3719}
3720
3721static ssize_t
3722ftrace_pid_write(struct file *filp, const char __user *ubuf,
3723           size_t cnt, loff_t *ppos)
3724{
3725    char buf[64], *tmp;
3726    long val;
3727    int ret;
3728
3729    if (cnt >= sizeof(buf))
3730        return -EINVAL;
3731
3732    if (copy_from_user(&buf, ubuf, cnt))
3733        return -EFAULT;
3734
3735    buf[cnt] = 0;
3736
3737    /*
3738     * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
3739     * to clean the filter quietly.
3740     */
3741    tmp = strstrip(buf);
3742    if (strlen(tmp) == 0)
3743        return 1;
3744
3745    ret = strict_strtol(tmp, 10, &val);
3746    if (ret < 0)
3747        return ret;
3748
3749    ret = ftrace_pid_add(val);
3750
3751    return ret ? ret : cnt;
3752}
3753
3754static int
3755ftrace_pid_release(struct inode *inode, struct file *file)
3756{
3757    if (file->f_mode & FMODE_READ)
3758        seq_release(inode, file);
3759
3760    return 0;
3761}
3762
3763static const struct file_operations ftrace_pid_fops = {
3764    .open = ftrace_pid_open,
3765    .write = ftrace_pid_write,
3766    .read = seq_read,
3767    .llseek = seq_lseek,
3768    .release = ftrace_pid_release,
3769};
3770
3771static __init int ftrace_init_debugfs(void)
3772{
3773    struct dentry *d_tracer;
3774
3775    d_tracer = tracing_init_dentry();
3776    if (!d_tracer)
3777        return 0;
3778
3779    ftrace_init_dyn_debugfs(d_tracer);
3780
3781    trace_create_file("set_ftrace_pid", 0644, d_tracer,
3782                NULL, &ftrace_pid_fops);
3783
3784    ftrace_profile_debugfs(d_tracer);
3785
3786    return 0;
3787}
3788fs_initcall(ftrace_init_debugfs);
3789
3790/**
3791 * ftrace_kill - kill ftrace
3792 *
3793 * This function should be used by panic code. It stops ftrace
3794 * but in a not so nice way. If you need to simply kill ftrace
3795 * from a non-atomic section, use ftrace_kill.
3796 */
3797void ftrace_kill(void)
3798{
3799    ftrace_disabled = 1;
3800    ftrace_enabled = 0;
3801    clear_ftrace_function();
3802}
3803
3804/**
3805 * register_ftrace_function - register a function for profiling
3806 * @ops - ops structure that holds the function for profiling.
3807 *
3808 * Register a function to be called by all functions in the
3809 * kernel.
3810 *
3811 * Note: @ops->func and all the functions it calls must be labeled
3812 * with "notrace", otherwise it will go into a
3813 * recursive loop.
3814 */
3815int register_ftrace_function(struct ftrace_ops *ops)
3816{
3817    int ret = -1;
3818
3819    mutex_lock(&ftrace_lock);
3820
3821    if (unlikely(ftrace_disabled))
3822        goto out_unlock;
3823
3824    ret = __register_ftrace_function(ops);
3825    if (!ret)
3826        ret = ftrace_startup(ops, 0);
3827
3828
3829 out_unlock:
3830    mutex_unlock(&ftrace_lock);
3831    return ret;
3832}
3833EXPORT_SYMBOL_GPL(register_ftrace_function);
3834
3835/**
3836 * unregister_ftrace_function - unregister a function for profiling.
3837 * @ops - ops structure that holds the function to unregister
3838 *
3839 * Unregister a function that was added to be called by ftrace profiling.
3840 */
3841int unregister_ftrace_function(struct ftrace_ops *ops)
3842{
3843    int ret;
3844
3845    mutex_lock(&ftrace_lock);
3846    ret = __unregister_ftrace_function(ops);
3847    if (!ret)
3848        ftrace_shutdown(ops, 0);
3849    mutex_unlock(&ftrace_lock);
3850
3851    return ret;
3852}
3853EXPORT_SYMBOL_GPL(unregister_ftrace_function);
3854
3855int
3856ftrace_enable_sysctl(struct ctl_table *table, int write,
3857             void __user *buffer, size_t *lenp,
3858             loff_t *ppos)
3859{
3860    int ret = -ENODEV;
3861
3862    mutex_lock(&ftrace_lock);
3863
3864    if (unlikely(ftrace_disabled))
3865        goto out;
3866
3867    ret = proc_dointvec(table, write, buffer, lenp, ppos);
3868
3869    if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3870        goto out;
3871
3872    last_ftrace_enabled = !!ftrace_enabled;
3873
3874    if (ftrace_enabled) {
3875
3876        ftrace_startup_sysctl();
3877
3878        /* we are starting ftrace again */
3879        if (ftrace_ops_list != &ftrace_list_end) {
3880            if (ftrace_ops_list->next == &ftrace_list_end)
3881                ftrace_trace_function = ftrace_ops_list->func;
3882            else
3883                ftrace_trace_function = ftrace_ops_list_func;
3884        }
3885
3886    } else {
3887        /* stopping ftrace calls (just send to ftrace_stub) */
3888        ftrace_trace_function = ftrace_stub;
3889
3890        ftrace_shutdown_sysctl();
3891    }
3892
3893 out:
3894    mutex_unlock(&ftrace_lock);
3895    return ret;
3896}
3897
3898#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3899
3900static int ftrace_graph_active;
3901static struct notifier_block ftrace_suspend_notifier;
3902
3903int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3904{
3905    return 0;
3906}
3907
3908/* The callbacks that hook a function */
3909trace_func_graph_ret_t ftrace_graph_return =
3910            (trace_func_graph_ret_t)ftrace_stub;
3911trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3912
3913/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3914static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3915{
3916    int i;
3917    int ret = 0;
3918    unsigned long flags;
3919    int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3920    struct task_struct *g, *t;
3921
3922    for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3923        ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3924                    * sizeof(struct ftrace_ret_stack),
3925                    GFP_KERNEL);
3926        if (!ret_stack_list[i]) {
3927            start = 0;
3928            end = i;
3929            ret = -ENOMEM;
3930            goto free;
3931        }
3932    }
3933
3934    read_lock_irqsave(&tasklist_lock, flags);
3935    do_each_thread(g, t) {
3936        if (start == end) {
3937            ret = -EAGAIN;
3938            goto unlock;
3939        }
3940
3941        if (t->ret_stack == NULL) {
3942            atomic_set(&t->tracing_graph_pause, 0);
3943            atomic_set(&t->trace_overrun, 0);
3944            t->curr_ret_stack = -1;
3945            /* Make sure the tasks see the -1 first: */
3946            smp_wmb();
3947            t->ret_stack = ret_stack_list[start++];
3948        }
3949    } while_each_thread(g, t);
3950
3951unlock:
3952    read_unlock_irqrestore(&tasklist_lock, flags);
3953free:
3954    for (i = start; i < end; i++)
3955        kfree(ret_stack_list[i]);
3956    return ret;
3957}
3958
3959static void
3960ftrace_graph_probe_sched_switch(void *ignore,
3961            struct task_struct *prev, struct task_struct *next)
3962{
3963    unsigned long long timestamp;
3964    int index;
3965
3966    /*
3967     * Does the user want to count the time a function was asleep.
3968     * If so, do not update the time stamps.
3969     */
3970    if (trace_flags & TRACE_ITER_SLEEP_TIME)
3971        return;
3972
3973    timestamp = trace_clock_local();
3974
3975    prev->ftrace_timestamp = timestamp;
3976
3977    /* only process tasks that we timestamped */
3978    if (!next->ftrace_timestamp)
3979        return;
3980
3981    /*
3982     * Update all the counters in next to make up for the
3983     * time next was sleeping.
3984     */
3985    timestamp -= next->ftrace_timestamp;
3986
3987    for (index = next->curr_ret_stack; index >= 0; index--)
3988        next->ret_stack[index].calltime += timestamp;
3989}
3990
3991/* Allocate a return stack for each task */
3992static int start_graph_tracing(void)
3993{
3994    struct ftrace_ret_stack **ret_stack_list;
3995    int ret, cpu;
3996
3997    ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3998                sizeof(struct ftrace_ret_stack *),
3999                GFP_KERNEL);
4000
4001    if (!ret_stack_list)
4002        return -ENOMEM;
4003
4004    /* The cpu_boot init_task->ret_stack will never be freed */
4005    for_each_online_cpu(cpu) {
4006        if (!idle_task(cpu)->ret_stack)
4007            ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4008    }
4009
4010    do {
4011        ret = alloc_retstack_tasklist(ret_stack_list);
4012    } while (ret == -EAGAIN);
4013
4014    if (!ret) {
4015        ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4016        if (ret)
4017            pr_info("ftrace_graph: Couldn't activate tracepoint"
4018                " probe to kernel_sched_switch\n");
4019    }
4020
4021    kfree(ret_stack_list);
4022    return ret;
4023}
4024
4025/*
4026 * Hibernation protection.
4027 * The state of the current task is too much unstable during
4028 * suspend/restore to disk. We want to protect against that.
4029 */
4030static int
4031ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4032                            void *unused)
4033{
4034    switch (state) {
4035    case PM_HIBERNATION_PREPARE:
4036        pause_graph_tracing();
4037        break;
4038
4039    case PM_POST_HIBERNATION:
4040        unpause_graph_tracing();
4041        break;
4042    }
4043    return NOTIFY_DONE;
4044}
4045
4046int register_ftrace_graph(trace_func_graph_ret_t retfunc,
4047            trace_func_graph_ent_t entryfunc)
4048{
4049    int ret = 0;
4050
4051    mutex_lock(&ftrace_lock);
4052
4053    /* we currently allow only one tracer registered at a time */
4054    if (ftrace_graph_active) {
4055        ret = -EBUSY;
4056        goto out;
4057    }
4058
4059    ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
4060    register_pm_notifier(&ftrace_suspend_notifier);
4061
4062    ftrace_graph_active++;
4063    ret = start_graph_tracing();
4064    if (ret) {
4065        ftrace_graph_active--;
4066        goto out;
4067    }
4068
4069    ftrace_graph_return = retfunc;
4070    ftrace_graph_entry = entryfunc;
4071
4072    ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
4073
4074out:
4075    mutex_unlock(&ftrace_lock);
4076    return ret;
4077}
4078
4079void unregister_ftrace_graph(void)
4080{
4081    mutex_lock(&ftrace_lock);
4082
4083    if (unlikely(!ftrace_graph_active))
4084        goto out;
4085
4086    ftrace_graph_active--;
4087    ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
4088    ftrace_graph_entry = ftrace_graph_entry_stub;
4089    ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
4090    unregister_pm_notifier(&ftrace_suspend_notifier);
4091    unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4092
4093 out:
4094    mutex_unlock(&ftrace_lock);
4095}
4096
4097static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
4098
4099static void
4100graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
4101{
4102    atomic_set(&t->tracing_graph_pause, 0);
4103    atomic_set(&t->trace_overrun, 0);
4104    t->ftrace_timestamp = 0;
4105    /* make curr_ret_stack visible before we add the ret_stack */
4106    smp_wmb();
4107    t->ret_stack = ret_stack;
4108}
4109
4110/*
4111 * Allocate a return stack for the idle task. May be the first
4112 * time through, or it may be done by CPU hotplug online.
4113 */
4114void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
4115{
4116    t->curr_ret_stack = -1;
4117    /*
4118     * The idle task has no parent, it either has its own
4119     * stack or no stack at all.
4120     */
4121    if (t->ret_stack)
4122        WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
4123
4124    if (ftrace_graph_active) {
4125        struct ftrace_ret_stack *ret_stack;
4126
4127        ret_stack = per_cpu(idle_ret_stack, cpu);
4128        if (!ret_stack) {
4129            ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4130                        * sizeof(struct ftrace_ret_stack),
4131                        GFP_KERNEL);
4132            if (!ret_stack)
4133                return;
4134            per_cpu(idle_ret_stack, cpu) = ret_stack;
4135        }
4136        graph_init_task(t, ret_stack);
4137    }
4138}
4139
4140/* Allocate a return stack for newly created task */
4141void ftrace_graph_init_task(struct task_struct *t)
4142{
4143    /* Make sure we do not use the parent ret_stack */
4144    t->ret_stack = NULL;
4145    t->curr_ret_stack = -1;
4146
4147    if (ftrace_graph_active) {
4148        struct ftrace_ret_stack *ret_stack;
4149
4150        ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4151                * sizeof(struct ftrace_ret_stack),
4152                GFP_KERNEL);
4153        if (!ret_stack)
4154            return;
4155        graph_init_task(t, ret_stack);
4156    }
4157}
4158
4159void ftrace_graph_exit_task(struct task_struct *t)
4160{
4161    struct ftrace_ret_stack *ret_stack = t->ret_stack;
4162
4163    t->ret_stack = NULL;
4164    /* NULL must become visible to IRQs before we free it: */
4165    barrier();
4166
4167    kfree(ret_stack);
4168}
4169
4170void ftrace_graph_stop(void)
4171{
4172    ftrace_stop();
4173}
4174#endif
4175

Archive Download this file



interactive