linux/kernel/trace/ftrace.c
<<
>>
Prefs
   1/*
   2 * Infrastructure for profiling code inserted by 'gcc -pg'.
   3 *
   4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5 * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
   6 *
   7 * Originally ported from the -rt patch by:
   8 *   Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
   9 *
  10 * Based on code in the latency_tracer, that is:
  11 *
  12 *  Copyright (C) 2004-2006 Ingo Molnar
  13 *  Copyright (C) 2004 Nadia Yvette Chambers
  14 */
  15
  16#include <linux/stop_machine.h>
  17#include <linux/clocksource.h>
  18#include <linux/sched/task.h>
  19#include <linux/kallsyms.h>
  20#include <linux/seq_file.h>
  21#include <linux/suspend.h>
  22#include <linux/tracefs.h>
  23#include <linux/hardirq.h>
  24#include <linux/kthread.h>
  25#include <linux/uaccess.h>
  26#include <linux/bsearch.h>
  27#include <linux/module.h>
  28#include <linux/ftrace.h>
  29#include <linux/sysctl.h>
  30#include <linux/slab.h>
  31#include <linux/ctype.h>
  32#include <linux/sort.h>
  33#include <linux/list.h>
  34#include <linux/hash.h>
  35#include <linux/rcupdate.h>
  36
  37#include <trace/events/sched.h>
  38
  39#include <asm/sections.h>
  40#include <asm/setup.h>
  41
  42#include "trace_output.h"
  43#include "trace_stat.h"
  44
  45#define FTRACE_WARN_ON(cond)                    \
  46        ({                                      \
  47                int ___r = cond;                \
  48                if (WARN_ON(___r))              \
  49                        ftrace_kill();          \
  50                ___r;                           \
  51        })
  52
  53#define FTRACE_WARN_ON_ONCE(cond)               \
  54        ({                                      \
  55                int ___r = cond;                \
  56                if (WARN_ON_ONCE(___r))         \
  57                        ftrace_kill();          \
  58                ___r;                           \
  59        })
  60
  61/* hash bits for specific function selection */
  62#define FTRACE_HASH_BITS 7
  63#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
  64#define FTRACE_HASH_DEFAULT_BITS 10
  65#define FTRACE_HASH_MAX_BITS 12
  66
  67#ifdef CONFIG_DYNAMIC_FTRACE
  68#define INIT_OPS_HASH(opsname)  \
  69        .func_hash              = &opsname.local_hash,                  \
  70        .local_hash.regex_lock  = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
  71#define ASSIGN_OPS_HASH(opsname, val) \
  72        .func_hash              = val, \
  73        .local_hash.regex_lock  = __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
  74#else
  75#define INIT_OPS_HASH(opsname)
  76#define ASSIGN_OPS_HASH(opsname, val)
  77#endif
  78
  79static struct ftrace_ops ftrace_list_end __read_mostly = {
  80        .func           = ftrace_stub,
  81        .flags          = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
  82        INIT_OPS_HASH(ftrace_list_end)
  83};
  84
  85/* ftrace_enabled is a method to turn ftrace on or off */
  86int ftrace_enabled __read_mostly;
  87static int last_ftrace_enabled;
  88
  89/* Current function tracing op */
  90struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
  91/* What to set function_trace_op to */
  92static struct ftrace_ops *set_function_trace_op;
  93
  94static bool ftrace_pids_enabled(struct ftrace_ops *ops)
  95{
  96        struct trace_array *tr;
  97
  98        if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
  99                return false;
 100
 101        tr = ops->private;
 102
 103        return tr->function_pids != NULL;
 104}
 105
 106static void ftrace_update_trampoline(struct ftrace_ops *ops);
 107
 108/*
 109 * ftrace_disabled is set when an anomaly is discovered.
 110 * ftrace_disabled is much stronger than ftrace_enabled.
 111 */
 112static int ftrace_disabled __read_mostly;
 113
 114static DEFINE_MUTEX(ftrace_lock);
 115
 116static struct ftrace_ops __rcu *ftrace_ops_list __read_mostly = &ftrace_list_end;
 117ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
 118static struct ftrace_ops global_ops;
 119
 120#if ARCH_SUPPORTS_FTRACE_OPS
 121static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
 122                                 struct ftrace_ops *op, struct pt_regs *regs);
 123#else
 124/* See comment below, where ftrace_ops_list_func is defined */
 125static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
 126#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
 127#endif
 128
 129/*
 130 * Traverse the ftrace_global_list, invoking all entries.  The reason that we
 131 * can use rcu_dereference_raw_notrace() is that elements removed from this list
 132 * are simply leaked, so there is no need to interact with a grace-period
 133 * mechanism.  The rcu_dereference_raw_notrace() calls are needed to handle
 134 * concurrent insertions into the ftrace_global_list.
 135 *
 136 * Silly Alpha and silly pointer-speculation compiler optimizations!
 137 */
 138#define do_for_each_ftrace_op(op, list)                 \
 139        op = rcu_dereference_raw_notrace(list);                 \
 140        do
 141
 142/*
 143 * Optimized for just a single item in the list (as that is the normal case).
 144 */
 145#define while_for_each_ftrace_op(op)                            \
 146        while (likely(op = rcu_dereference_raw_notrace((op)->next)) &&  \
 147               unlikely((op) != &ftrace_list_end))
 148
 149static inline void ftrace_ops_init(struct ftrace_ops *ops)
 150{
 151#ifdef CONFIG_DYNAMIC_FTRACE
 152        if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
 153                mutex_init(&ops->local_hash.regex_lock);
 154                ops->func_hash = &ops->local_hash;
 155                ops->flags |= FTRACE_OPS_FL_INITIALIZED;
 156        }
 157#endif
 158}
 159
 160/**
 161 * ftrace_nr_registered_ops - return number of ops registered
 162 *
 163 * Returns the number of ftrace_ops registered and tracing functions
 164 */
 165int ftrace_nr_registered_ops(void)
 166{
 167        struct ftrace_ops *ops;
 168        int cnt = 0;
 169
 170        mutex_lock(&ftrace_lock);
 171
 172        for (ops = rcu_dereference_protected(ftrace_ops_list,
 173                                             lockdep_is_held(&ftrace_lock));
 174             ops != &ftrace_list_end;
 175             ops = rcu_dereference_protected(ops->next,
 176                                             lockdep_is_held(&ftrace_lock)))
 177                cnt++;
 178
 179        mutex_unlock(&ftrace_lock);
 180
 181        return cnt;
 182}
 183
 184static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
 185                            struct ftrace_ops *op, struct pt_regs *regs)
 186{
 187        struct trace_array *tr = op->private;
 188
 189        if (tr && this_cpu_read(tr->trace_buffer.data->ftrace_ignore_pid))
 190                return;
 191
 192        op->saved_func(ip, parent_ip, op, regs);
 193}
 194
 195/**
 196 * clear_ftrace_function - reset the ftrace function
 197 *
 198 * This NULLs the ftrace function and in essence stops
 199 * tracing.  There may be lag
 200 */
 201void clear_ftrace_function(void)
 202{
 203        ftrace_trace_function = ftrace_stub;
 204}
 205
 206static void ftrace_sync(struct work_struct *work)
 207{
 208        /*
 209         * This function is just a stub to implement a hard force
 210         * of synchronize_sched(). This requires synchronizing
 211         * tasks even in userspace and idle.
 212         *
 213         * Yes, function tracing is rude.
 214         */
 215}
 216
 217static void ftrace_sync_ipi(void *data)
 218{
 219        /* Probably not needed, but do it anyway */
 220        smp_rmb();
 221}
 222
 223#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 224static void update_function_graph_func(void);
 225
 226/* Both enabled by default (can be cleared by function_graph tracer flags */
 227static bool fgraph_sleep_time = true;
 228static bool fgraph_graph_time = true;
 229
 230#else
 231static inline void update_function_graph_func(void) { }
 232#endif
 233
 234
 235static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
 236{
 237        /*
 238         * If this is a dynamic, RCU, or per CPU ops, or we force list func,
 239         * then it needs to call the list anyway.
 240         */
 241        if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_RCU) ||
 242            FTRACE_FORCE_LIST_FUNC)
 243                return ftrace_ops_list_func;
 244
 245        return ftrace_ops_get_func(ops);
 246}
 247
 248static void update_ftrace_function(void)
 249{
 250        ftrace_func_t func;
 251
 252        /*
 253         * Prepare the ftrace_ops that the arch callback will use.
 254         * If there's only one ftrace_ops registered, the ftrace_ops_list
 255         * will point to the ops we want.
 256         */
 257        set_function_trace_op = rcu_dereference_protected(ftrace_ops_list,
 258                                                lockdep_is_held(&ftrace_lock));
 259
 260        /* If there's no ftrace_ops registered, just call the stub function */
 261        if (set_function_trace_op == &ftrace_list_end) {
 262                func = ftrace_stub;
 263
 264        /*
 265         * If we are at the end of the list and this ops is
 266         * recursion safe and not dynamic and the arch supports passing ops,
 267         * then have the mcount trampoline call the function directly.
 268         */
 269        } else if (rcu_dereference_protected(ftrace_ops_list->next,
 270                        lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
 271                func = ftrace_ops_get_list_func(ftrace_ops_list);
 272
 273        } else {
 274                /* Just use the default ftrace_ops */
 275                set_function_trace_op = &ftrace_list_end;
 276                func = ftrace_ops_list_func;
 277        }
 278
 279        update_function_graph_func();
 280
 281        /* If there's no change, then do nothing more here */
 282        if (ftrace_trace_function == func)
 283                return;
 284
 285        /*
 286         * If we are using the list function, it doesn't care
 287         * about the function_trace_ops.
 288         */
 289        if (func == ftrace_ops_list_func) {
 290                ftrace_trace_function = func;
 291                /*
 292                 * Don't even bother setting function_trace_ops,
 293                 * it would be racy to do so anyway.
 294                 */
 295                return;
 296        }
 297
 298#ifndef CONFIG_DYNAMIC_FTRACE
 299        /*
 300         * For static tracing, we need to be a bit more careful.
 301         * The function change takes affect immediately. Thus,
 302         * we need to coorditate the setting of the function_trace_ops
 303         * with the setting of the ftrace_trace_function.
 304         *
 305         * Set the function to the list ops, which will call the
 306         * function we want, albeit indirectly, but it handles the
 307         * ftrace_ops and doesn't depend on function_trace_op.
 308         */
 309        ftrace_trace_function = ftrace_ops_list_func;
 310        /*
 311         * Make sure all CPUs see this. Yes this is slow, but static
 312         * tracing is slow and nasty to have enabled.
 313         */
 314        schedule_on_each_cpu(ftrace_sync);
 315        /* Now all cpus are using the list ops. */
 316        function_trace_op = set_function_trace_op;
 317        /* Make sure the function_trace_op is visible on all CPUs */
 318        smp_wmb();
 319        /* Nasty way to force a rmb on all cpus */
 320        smp_call_function(ftrace_sync_ipi, NULL, 1);
 321        /* OK, we are all set to update the ftrace_trace_function now! */
 322#endif /* !CONFIG_DYNAMIC_FTRACE */
 323
 324        ftrace_trace_function = func;
 325}
 326
 327int using_ftrace_ops_list_func(void)
 328{
 329        return ftrace_trace_function == ftrace_ops_list_func;
 330}
 331
 332static void add_ftrace_ops(struct ftrace_ops __rcu **list,
 333                           struct ftrace_ops *ops)
 334{
 335        rcu_assign_pointer(ops->next, *list);
 336
 337        /*
 338         * We are entering ops into the list but another
 339         * CPU might be walking that list. We need to make sure
 340         * the ops->next pointer is valid before another CPU sees
 341         * the ops pointer included into the list.
 342         */
 343        rcu_assign_pointer(*list, ops);
 344}
 345
 346static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
 347                             struct ftrace_ops *ops)
 348{
 349        struct ftrace_ops **p;
 350
 351        /*
 352         * If we are removing the last function, then simply point
 353         * to the ftrace_stub.
 354         */
 355        if (rcu_dereference_protected(*list,
 356                        lockdep_is_held(&ftrace_lock)) == ops &&
 357            rcu_dereference_protected(ops->next,
 358                        lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
 359                *list = &ftrace_list_end;
 360                return 0;
 361        }
 362
 363        for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
 364                if (*p == ops)
 365                        break;
 366
 367        if (*p != ops)
 368                return -1;
 369
 370        *p = (*p)->next;
 371        return 0;
 372}
 373
 374static void ftrace_update_trampoline(struct ftrace_ops *ops);
 375
 376static int __register_ftrace_function(struct ftrace_ops *ops)
 377{
 378        if (ops->flags & FTRACE_OPS_FL_DELETED)
 379                return -EINVAL;
 380
 381        if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
 382                return -EBUSY;
 383
 384#ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
 385        /*
 386         * If the ftrace_ops specifies SAVE_REGS, then it only can be used
 387         * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
 388         * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
 389         */
 390        if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
 391            !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
 392                return -EINVAL;
 393
 394        if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
 395                ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
 396#endif
 397
 398        if (!core_kernel_data((unsigned long)ops))
 399                ops->flags |= FTRACE_OPS_FL_DYNAMIC;
 400
 401        add_ftrace_ops(&ftrace_ops_list, ops);
 402
 403        /* Always save the function, and reset at unregistering */
 404        ops->saved_func = ops->func;
 405
 406        if (ftrace_pids_enabled(ops))
 407                ops->func = ftrace_pid_func;
 408
 409        ftrace_update_trampoline(ops);
 410
 411        if (ftrace_enabled)
 412                update_ftrace_function();
 413
 414        return 0;
 415}
 416
 417static int __unregister_ftrace_function(struct ftrace_ops *ops)
 418{
 419        int ret;
 420
 421        if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
 422                return -EBUSY;
 423
 424        ret = remove_ftrace_ops(&ftrace_ops_list, ops);
 425
 426        if (ret < 0)
 427                return ret;
 428
 429        if (ftrace_enabled)
 430                update_ftrace_function();
 431
 432        ops->func = ops->saved_func;
 433
 434        return 0;
 435}
 436
 437static void ftrace_update_pid_func(void)
 438{
 439        struct ftrace_ops *op;
 440
 441        /* Only do something if we are tracing something */
 442        if (ftrace_trace_function == ftrace_stub)
 443                return;
 444
 445        do_for_each_ftrace_op(op, ftrace_ops_list) {
 446                if (op->flags & FTRACE_OPS_FL_PID) {
 447                        op->func = ftrace_pids_enabled(op) ?
 448                                ftrace_pid_func : op->saved_func;
 449                        ftrace_update_trampoline(op);
 450                }
 451        } while_for_each_ftrace_op(op);
 452
 453        update_ftrace_function();
 454}
 455
 456#ifdef CONFIG_FUNCTION_PROFILER
 457struct ftrace_profile {
 458        struct hlist_node               node;
 459        unsigned long                   ip;
 460        unsigned long                   counter;
 461#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 462        unsigned long long              time;
 463        unsigned long long              time_squared;
 464#endif
 465};
 466
 467struct ftrace_profile_page {
 468        struct ftrace_profile_page      *next;
 469        unsigned long                   index;
 470        struct ftrace_profile           records[];
 471};
 472
 473struct ftrace_profile_stat {
 474        atomic_t                        disabled;
 475        struct hlist_head               *hash;
 476        struct ftrace_profile_page      *pages;
 477        struct ftrace_profile_page      *start;
 478        struct tracer_stat              stat;
 479};
 480
 481#define PROFILE_RECORDS_SIZE                                            \
 482        (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
 483
 484#define PROFILES_PER_PAGE                                       \
 485        (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
 486
 487static int ftrace_profile_enabled __read_mostly;
 488
 489/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
 490static DEFINE_MUTEX(ftrace_profile_lock);
 491
 492static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
 493
 494#define FTRACE_PROFILE_HASH_BITS 10
 495#define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
 496
 497static void *
 498function_stat_next(void *v, int idx)
 499{
 500        struct ftrace_profile *rec = v;
 501        struct ftrace_profile_page *pg;
 502
 503        pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
 504
 505 again:
 506        if (idx != 0)
 507                rec++;
 508
 509        if ((void *)rec >= (void *)&pg->records[pg->index]) {
 510                pg = pg->next;
 511                if (!pg)
 512                        return NULL;
 513                rec = &pg->records[0];
 514                if (!rec->counter)
 515                        goto again;
 516        }
 517
 518        return rec;
 519}
 520
 521static void *function_stat_start(struct tracer_stat *trace)
 522{
 523        struct ftrace_profile_stat *stat =
 524                container_of(trace, struct ftrace_profile_stat, stat);
 525
 526        if (!stat || !stat->start)
 527                return NULL;
 528
 529        return function_stat_next(&stat->start->records[0], 0);
 530}
 531
 532#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 533/* function graph compares on total time */
 534static int function_stat_cmp(void *p1, void *p2)
 535{
 536        struct ftrace_profile *a = p1;
 537        struct ftrace_profile *b = p2;
 538
 539        if (a->time < b->time)
 540                return -1;
 541        if (a->time > b->time)
 542                return 1;
 543        else
 544                return 0;
 545}
 546#else
 547/* not function graph compares against hits */
 548static int function_stat_cmp(void *p1, void *p2)
 549{
 550        struct ftrace_profile *a = p1;
 551        struct ftrace_profile *b = p2;
 552
 553        if (a->counter < b->counter)
 554                return -1;
 555        if (a->counter > b->counter)
 556                return 1;
 557        else
 558                return 0;
 559}
 560#endif
 561
 562static int function_stat_headers(struct seq_file *m)
 563{
 564#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 565        seq_puts(m, "  Function                               "
 566                 "Hit    Time            Avg             s^2\n"
 567                    "  --------                               "
 568                 "---    ----            ---             ---\n");
 569#else
 570        seq_puts(m, "  Function                               Hit\n"
 571                    "  --------                               ---\n");
 572#endif
 573        return 0;
 574}
 575
 576static int function_stat_show(struct seq_file *m, void *v)
 577{
 578        struct ftrace_profile *rec = v;
 579        char str[KSYM_SYMBOL_LEN];
 580        int ret = 0;
 581#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 582        static struct trace_seq s;
 583        unsigned long long avg;
 584        unsigned long long stddev;
 585#endif
 586        mutex_lock(&ftrace_profile_lock);
 587
 588        /* we raced with function_profile_reset() */
 589        if (unlikely(rec->counter == 0)) {
 590                ret = -EBUSY;
 591                goto out;
 592        }
 593
 594#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 595        avg = rec->time;
 596        do_div(avg, rec->counter);
 597        if (tracing_thresh && (avg < tracing_thresh))
 598                goto out;
 599#endif
 600
 601        kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 602        seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
 603
 604#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 605        seq_puts(m, "    ");
 606
 607        /* Sample standard deviation (s^2) */
 608        if (rec->counter <= 1)
 609                stddev = 0;
 610        else {
 611                /*
 612                 * Apply Welford's method:
 613                 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
 614                 */
 615                stddev = rec->counter * rec->time_squared -
 616                         rec->time * rec->time;
 617
 618                /*
 619                 * Divide only 1000 for ns^2 -> us^2 conversion.
 620                 * trace_print_graph_duration will divide 1000 again.
 621                 */
 622                do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
 623        }
 624
 625        trace_seq_init(&s);
 626        trace_print_graph_duration(rec->time, &s);
 627        trace_seq_puts(&s, "    ");
 628        trace_print_graph_duration(avg, &s);
 629        trace_seq_puts(&s, "    ");
 630        trace_print_graph_duration(stddev, &s);
 631        trace_print_seq(m, &s);
 632#endif
 633        seq_putc(m, '\n');
 634out:
 635        mutex_unlock(&ftrace_profile_lock);
 636
 637        return ret;
 638}
 639
 640static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
 641{
 642        struct ftrace_profile_page *pg;
 643
 644        pg = stat->pages = stat->start;
 645
 646        while (pg) {
 647                memset(pg->records, 0, PROFILE_RECORDS_SIZE);
 648                pg->index = 0;
 649                pg = pg->next;
 650        }
 651
 652        memset(stat->hash, 0,
 653               FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
 654}
 655
 656int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
 657{
 658        struct ftrace_profile_page *pg;
 659        int functions;
 660        int pages;
 661        int i;
 662
 663        /* If we already allocated, do nothing */
 664        if (stat->pages)
 665                return 0;
 666
 667        stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
 668        if (!stat->pages)
 669                return -ENOMEM;
 670
 671#ifdef CONFIG_DYNAMIC_FTRACE
 672        functions = ftrace_update_tot_cnt;
 673#else
 674        /*
 675         * We do not know the number of functions that exist because
 676         * dynamic tracing is what counts them. With past experience
 677         * we have around 20K functions. That should be more than enough.
 678         * It is highly unlikely we will execute every function in
 679         * the kernel.
 680         */
 681        functions = 20000;
 682#endif
 683
 684        pg = stat->start = stat->pages;
 685
 686        pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
 687
 688        for (i = 1; i < pages; i++) {
 689                pg->next = (void *)get_zeroed_page(GFP_KERNEL);
 690                if (!pg->next)
 691                        goto out_free;
 692                pg = pg->next;
 693        }
 694
 695        return 0;
 696
 697 out_free:
 698        pg = stat->start;
 699        while (pg) {
 700                unsigned long tmp = (unsigned long)pg;
 701
 702                pg = pg->next;
 703                free_page(tmp);
 704        }
 705
 706        stat->pages = NULL;
 707        stat->start = NULL;
 708
 709        return -ENOMEM;
 710}
 711
 712static int ftrace_profile_init_cpu(int cpu)
 713{
 714        struct ftrace_profile_stat *stat;
 715        int size;
 716
 717        stat = &per_cpu(ftrace_profile_stats, cpu);
 718
 719        if (stat->hash) {
 720                /* If the profile is already created, simply reset it */
 721                ftrace_profile_reset(stat);
 722                return 0;
 723        }
 724
 725        /*
 726         * We are profiling all functions, but usually only a few thousand
 727         * functions are hit. We'll make a hash of 1024 items.
 728         */
 729        size = FTRACE_PROFILE_HASH_SIZE;
 730
 731        stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
 732
 733        if (!stat->hash)
 734                return -ENOMEM;
 735
 736        /* Preallocate the function profiling pages */
 737        if (ftrace_profile_pages_init(stat) < 0) {
 738                kfree(stat->hash);
 739                stat->hash = NULL;
 740                return -ENOMEM;
 741        }
 742
 743        return 0;
 744}
 745
 746static int ftrace_profile_init(void)
 747{
 748        int cpu;
 749        int ret = 0;
 750
 751        for_each_possible_cpu(cpu) {
 752                ret = ftrace_profile_init_cpu(cpu);
 753                if (ret)
 754                        break;
 755        }
 756
 757        return ret;
 758}
 759
 760/* interrupts must be disabled */
 761static struct ftrace_profile *
 762ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
 763{
 764        struct ftrace_profile *rec;
 765        struct hlist_head *hhd;
 766        unsigned long key;
 767
 768        key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
 769        hhd = &stat->hash[key];
 770
 771        if (hlist_empty(hhd))
 772                return NULL;
 773
 774        hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
 775                if (rec->ip == ip)
 776                        return rec;
 777        }
 778
 779        return NULL;
 780}
 781
 782static void ftrace_add_profile(struct ftrace_profile_stat *stat,
 783                               struct ftrace_profile *rec)
 784{
 785        unsigned long key;
 786
 787        key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
 788        hlist_add_head_rcu(&rec->node, &stat->hash[key]);
 789}
 790
 791/*
 792 * The memory is already allocated, this simply finds a new record to use.
 793 */
 794static struct ftrace_profile *
 795ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
 796{
 797        struct ftrace_profile *rec = NULL;
 798
 799        /* prevent recursion (from NMIs) */
 800        if (atomic_inc_return(&stat->disabled) != 1)
 801                goto out;
 802
 803        /*
 804         * Try to find the function again since an NMI
 805         * could have added it
 806         */
 807        rec = ftrace_find_profiled_func(stat, ip);
 808        if (rec)
 809                goto out;
 810
 811        if (stat->pages->index == PROFILES_PER_PAGE) {
 812                if (!stat->pages->next)
 813                        goto out;
 814                stat->pages = stat->pages->next;
 815        }
 816
 817        rec = &stat->pages->records[stat->pages->index++];
 818        rec->ip = ip;
 819        ftrace_add_profile(stat, rec);
 820
 821 out:
 822        atomic_dec(&stat->disabled);
 823
 824        return rec;
 825}
 826
 827static void
 828function_profile_call(unsigned long ip, unsigned long parent_ip,
 829                      struct ftrace_ops *ops, struct pt_regs *regs)
 830{
 831        struct ftrace_profile_stat *stat;
 832        struct ftrace_profile *rec;
 833        unsigned long flags;
 834
 835        if (!ftrace_profile_enabled)
 836                return;
 837
 838        local_irq_save(flags);
 839
 840        stat = this_cpu_ptr(&ftrace_profile_stats);
 841        if (!stat->hash || !ftrace_profile_enabled)
 842                goto out;
 843
 844        rec = ftrace_find_profiled_func(stat, ip);
 845        if (!rec) {
 846                rec = ftrace_profile_alloc(stat, ip);
 847                if (!rec)
 848                        goto out;
 849        }
 850
 851        rec->counter++;
 852 out:
 853        local_irq_restore(flags);
 854}
 855
 856#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 857static int profile_graph_entry(struct ftrace_graph_ent *trace)
 858{
 859        int index = trace->depth;
 860
 861        function_profile_call(trace->func, 0, NULL, NULL);
 862
 863        /* If function graph is shutting down, ret_stack can be NULL */
 864        if (!current->ret_stack)
 865                return 0;
 866
 867        if (index >= 0 && index < FTRACE_RETFUNC_DEPTH)
 868                current->ret_stack[index].subtime = 0;
 869
 870        return 1;
 871}
 872
 873static void profile_graph_return(struct ftrace_graph_ret *trace)
 874{
 875        struct ftrace_profile_stat *stat;
 876        unsigned long long calltime;
 877        struct ftrace_profile *rec;
 878        unsigned long flags;
 879
 880        local_irq_save(flags);
 881        stat = this_cpu_ptr(&ftrace_profile_stats);
 882        if (!stat->hash || !ftrace_profile_enabled)
 883                goto out;
 884
 885        /* If the calltime was zero'd ignore it */
 886        if (!trace->calltime)
 887                goto out;
 888
 889        calltime = trace->rettime - trace->calltime;
 890
 891        if (!fgraph_graph_time) {
 892                int index;
 893
 894                index = trace->depth;
 895
 896                /* Append this call time to the parent time to subtract */
 897                if (index)
 898                        current->ret_stack[index - 1].subtime += calltime;
 899
 900                if (current->ret_stack[index].subtime < calltime)
 901                        calltime -= current->ret_stack[index].subtime;
 902                else
 903                        calltime = 0;
 904        }
 905
 906        rec = ftrace_find_profiled_func(stat, trace->func);
 907        if (rec) {
 908                rec->time += calltime;
 909                rec->time_squared += calltime * calltime;
 910        }
 911
 912 out:
 913        local_irq_restore(flags);
 914}
 915
 916static int register_ftrace_profiler(void)
 917{
 918        return register_ftrace_graph(&profile_graph_return,
 919                                     &profile_graph_entry);
 920}
 921
 922static void unregister_ftrace_profiler(void)
 923{
 924        unregister_ftrace_graph();
 925}
 926#else
 927static struct ftrace_ops ftrace_profile_ops __read_mostly = {
 928        .func           = function_profile_call,
 929        .flags          = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
 930        INIT_OPS_HASH(ftrace_profile_ops)
 931};
 932
 933static int register_ftrace_profiler(void)
 934{
 935        return register_ftrace_function(&ftrace_profile_ops);
 936}
 937
 938static void unregister_ftrace_profiler(void)
 939{
 940        unregister_ftrace_function(&ftrace_profile_ops);
 941}
 942#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 943
 944static ssize_t
 945ftrace_profile_write(struct file *filp, const char __user *ubuf,
 946                     size_t cnt, loff_t *ppos)
 947{
 948        unsigned long val;
 949        int ret;
 950
 951        ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 952        if (ret)
 953                return ret;
 954
 955        val = !!val;
 956
 957        mutex_lock(&ftrace_profile_lock);
 958        if (ftrace_profile_enabled ^ val) {
 959                if (val) {
 960                        ret = ftrace_profile_init();
 961                        if (ret < 0) {
 962                                cnt = ret;
 963                                goto out;
 964                        }
 965
 966                        ret = register_ftrace_profiler();
 967                        if (ret < 0) {
 968                                cnt = ret;
 969                                goto out;
 970                        }
 971                        ftrace_profile_enabled = 1;
 972                } else {
 973                        ftrace_profile_enabled = 0;
 974                        /*
 975                         * unregister_ftrace_profiler calls stop_machine
 976                         * so this acts like an synchronize_sched.
 977                         */
 978                        unregister_ftrace_profiler();
 979                }
 980        }
 981 out:
 982        mutex_unlock(&ftrace_profile_lock);
 983
 984        *ppos += cnt;
 985
 986        return cnt;
 987}
 988
 989static ssize_t
 990ftrace_profile_read(struct file *filp, char __user *ubuf,
 991                     size_t cnt, loff_t *ppos)
 992{
 993        char buf[64];           /* big enough to hold a number */
 994        int r;
 995
 996        r = sprintf(buf, "%u\n", ftrace_profile_enabled);
 997        return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 998}
 999
1000static const struct file_operations ftrace_profile_fops = {
1001        .open           = tracing_open_generic,
1002        .read           = ftrace_profile_read,
1003        .write          = ftrace_profile_write,
1004        .llseek         = default_llseek,
1005};
1006
1007/* used to initialize the real stat files */
1008static struct tracer_stat function_stats __initdata = {
1009        .name           = "functions",
1010        .stat_start     = function_stat_start,
1011        .stat_next      = function_stat_next,
1012        .stat_cmp       = function_stat_cmp,
1013        .stat_headers   = function_stat_headers,
1014        .stat_show      = function_stat_show
1015};
1016
1017static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1018{
1019        struct ftrace_profile_stat *stat;
1020        struct dentry *entry;
1021        char *name;
1022        int ret;
1023        int cpu;
1024
1025        for_each_possible_cpu(cpu) {
1026                stat = &per_cpu(ftrace_profile_stats, cpu);
1027
1028                name = kasprintf(GFP_KERNEL, "function%d", cpu);
1029                if (!name) {
1030                        /*
1031                         * The files created are permanent, if something happens
1032                         * we still do not free memory.
1033                         */
1034                        WARN(1,
1035                             "Could not allocate stat file for cpu %d\n",
1036                             cpu);
1037                        return;
1038                }
1039                stat->stat = function_stats;
1040                stat->stat.name = name;
1041                ret = register_stat_tracer(&stat->stat);
1042                if (ret) {
1043                        WARN(1,
1044                             "Could not register function stat for cpu %d\n",
1045                             cpu);
1046                        kfree(name);
1047                        return;
1048                }
1049        }
1050
1051        entry = tracefs_create_file("function_profile_enabled", 0644,
1052                                    d_tracer, NULL, &ftrace_profile_fops);
1053        if (!entry)
1054                pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1055}
1056
1057#else /* CONFIG_FUNCTION_PROFILER */
1058static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1059{
1060}
1061#endif /* CONFIG_FUNCTION_PROFILER */
1062
1063static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1064
1065#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1066static int ftrace_graph_active;
1067#else
1068# define ftrace_graph_active 0
1069#endif
1070
1071#ifdef CONFIG_DYNAMIC_FTRACE
1072
1073static struct ftrace_ops *removed_ops;
1074
1075/*
1076 * Set when doing a global update, like enabling all recs or disabling them.
1077 * It is not set when just updating a single ftrace_ops.
1078 */
1079static bool update_all_ops;
1080
1081#ifndef CONFIG_FTRACE_MCOUNT_RECORD
1082# error Dynamic ftrace depends on MCOUNT_RECORD
1083#endif
1084
1085struct ftrace_func_entry {
1086        struct hlist_node hlist;
1087        unsigned long ip;
1088};
1089
1090struct ftrace_func_probe {
1091        struct ftrace_probe_ops *probe_ops;
1092        struct ftrace_ops       ops;
1093        struct trace_array      *tr;
1094        struct list_head        list;
1095        void                    *data;
1096        int                     ref;
1097};
1098
1099/*
1100 * We make these constant because no one should touch them,
1101 * but they are used as the default "empty hash", to avoid allocating
1102 * it all the time. These are in a read only section such that if
1103 * anyone does try to modify it, it will cause an exception.
1104 */
1105static const struct hlist_head empty_buckets[1];
1106static const struct ftrace_hash empty_hash = {
1107        .buckets = (struct hlist_head *)empty_buckets,
1108};
1109#define EMPTY_HASH      ((struct ftrace_hash *)&empty_hash)
1110
1111static struct ftrace_ops global_ops = {
1112        .func                           = ftrace_stub,
1113        .local_hash.notrace_hash        = EMPTY_HASH,
1114        .local_hash.filter_hash         = EMPTY_HASH,
1115        INIT_OPS_HASH(global_ops)
1116        .flags                          = FTRACE_OPS_FL_RECURSION_SAFE |
1117                                          FTRACE_OPS_FL_INITIALIZED |
1118                                          FTRACE_OPS_FL_PID,
1119};
1120
1121/*
1122 * Used by the stack undwinder to know about dynamic ftrace trampolines.
1123 */
1124struct ftrace_ops *ftrace_ops_trampoline(unsigned long addr)
1125{
1126        struct ftrace_ops *op = NULL;
1127
1128        /*
1129         * Some of the ops may be dynamically allocated,
1130         * they are freed after a synchronize_sched().
1131         */
1132        preempt_disable_notrace();
1133
1134        do_for_each_ftrace_op(op, ftrace_ops_list) {
1135                /*
1136                 * This is to check for dynamically allocated trampolines.
1137                 * Trampolines that are in kernel text will have
1138                 * core_kernel_text() return true.
1139                 */
1140                if (op->trampoline && op->trampoline_size)
1141                        if (addr >= op->trampoline &&
1142                            addr < op->trampoline + op->trampoline_size) {
1143                                preempt_enable_notrace();
1144                                return op;
1145                        }
1146        } while_for_each_ftrace_op(op);
1147        preempt_enable_notrace();
1148
1149        return NULL;
1150}
1151
1152/*
1153 * This is used by __kernel_text_address() to return true if the
1154 * address is on a dynamically allocated trampoline that would
1155 * not return true for either core_kernel_text() or
1156 * is_module_text_address().
1157 */
1158bool is_ftrace_trampoline(unsigned long addr)
1159{
1160        return ftrace_ops_trampoline(addr) != NULL;
1161}
1162
1163struct ftrace_page {
1164        struct ftrace_page      *next;
1165        struct dyn_ftrace       *records;
1166        int                     index;
1167        int                     size;
1168};
1169
1170#define ENTRY_SIZE sizeof(struct dyn_ftrace)
1171#define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1172
1173/* estimate from running different kernels */
1174#define NR_TO_INIT              10000
1175
1176static struct ftrace_page       *ftrace_pages_start;
1177static struct ftrace_page       *ftrace_pages;
1178
1179static __always_inline unsigned long
1180ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1181{
1182        if (hash->size_bits > 0)
1183                return hash_long(ip, hash->size_bits);
1184
1185        return 0;
1186}
1187
1188/* Only use this function if ftrace_hash_empty() has already been tested */
1189static __always_inline struct ftrace_func_entry *
1190__ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1191{
1192        unsigned long key;
1193        struct ftrace_func_entry *entry;
1194        struct hlist_head *hhd;
1195
1196        key = ftrace_hash_key(hash, ip);
1197        hhd = &hash->buckets[key];
1198
1199        hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1200                if (entry->ip == ip)
1201                        return entry;
1202        }
1203        return NULL;
1204}
1205
1206/**
1207 * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1208 * @hash: The hash to look at
1209 * @ip: The instruction pointer to test
1210 *
1211 * Search a given @hash to see if a given instruction pointer (@ip)
1212 * exists in it.
1213 *
1214 * Returns the entry that holds the @ip if found. NULL otherwise.
1215 */
1216struct ftrace_func_entry *
1217ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1218{
1219        if (ftrace_hash_empty(hash))
1220                return NULL;
1221
1222        return __ftrace_lookup_ip(hash, ip);
1223}
1224
1225static void __add_hash_entry(struct ftrace_hash *hash,
1226                             struct ftrace_func_entry *entry)
1227{
1228        struct hlist_head *hhd;
1229        unsigned long key;
1230
1231        key = ftrace_hash_key(hash, entry->ip);
1232        hhd = &hash->buckets[key];
1233        hlist_add_head(&entry->hlist, hhd);
1234        hash->count++;
1235}
1236
1237static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1238{
1239        struct ftrace_func_entry *entry;
1240
1241        entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1242        if (!entry)
1243                return -ENOMEM;
1244
1245        entry->ip = ip;
1246        __add_hash_entry(hash, entry);
1247
1248        return 0;
1249}
1250
1251static void
1252free_hash_entry(struct ftrace_hash *hash,
1253                  struct ftrace_func_entry *entry)
1254{
1255        hlist_del(&entry->hlist);
1256        kfree(entry);
1257        hash->count--;
1258}
1259
1260static void
1261remove_hash_entry(struct ftrace_hash *hash,
1262                  struct ftrace_func_entry *entry)
1263{
1264        hlist_del_rcu(&entry->hlist);
1265        hash->count--;
1266}
1267
1268static void ftrace_hash_clear(struct ftrace_hash *hash)
1269{
1270        struct hlist_head *hhd;
1271        struct hlist_node *tn;
1272        struct ftrace_func_entry *entry;
1273        int size = 1 << hash->size_bits;
1274        int i;
1275
1276        if (!hash->count)
1277                return;
1278
1279        for (i = 0; i < size; i++) {
1280                hhd = &hash->buckets[i];
1281                hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1282                        free_hash_entry(hash, entry);
1283        }
1284        FTRACE_WARN_ON(hash->count);
1285}
1286
1287static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1288{
1289        list_del(&ftrace_mod->list);
1290        kfree(ftrace_mod->module);
1291        kfree(ftrace_mod->func);
1292        kfree(ftrace_mod);
1293}
1294
1295static void clear_ftrace_mod_list(struct list_head *head)
1296{
1297        struct ftrace_mod_load *p, *n;
1298
1299        /* stack tracer isn't supported yet */
1300        if (!head)
1301                return;
1302
1303        mutex_lock(&ftrace_lock);
1304        list_for_each_entry_safe(p, n, head, list)
1305                free_ftrace_mod(p);
1306        mutex_unlock(&ftrace_lock);
1307}
1308
1309static void free_ftrace_hash(struct ftrace_hash *hash)
1310{
1311        if (!hash || hash == EMPTY_HASH)
1312                return;
1313        ftrace_hash_clear(hash);
1314        kfree(hash->buckets);
1315        kfree(hash);
1316}
1317
1318static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1319{
1320        struct ftrace_hash *hash;
1321
1322        hash = container_of(rcu, struct ftrace_hash, rcu);
1323        free_ftrace_hash(hash);
1324}
1325
1326static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1327{
1328        if (!hash || hash == EMPTY_HASH)
1329                return;
1330        call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1331}
1332
1333void ftrace_free_filter(struct ftrace_ops *ops)
1334{
1335        ftrace_ops_init(ops);
1336        free_ftrace_hash(ops->func_hash->filter_hash);
1337        free_ftrace_hash(ops->func_hash->notrace_hash);
1338}
1339
1340static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1341{
1342        struct ftrace_hash *hash;
1343        int size;
1344
1345        hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1346        if (!hash)
1347                return NULL;
1348
1349        size = 1 << size_bits;
1350        hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1351
1352        if (!hash->buckets) {
1353                kfree(hash);
1354                return NULL;
1355        }
1356
1357        hash->size_bits = size_bits;
1358
1359        return hash;
1360}
1361
1362
1363static int ftrace_add_mod(struct trace_array *tr,
1364                          const char *func, const char *module,
1365                          int enable)
1366{
1367        struct ftrace_mod_load *ftrace_mod;
1368        struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1369
1370        ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1371        if (!ftrace_mod)
1372                return -ENOMEM;
1373
1374        ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1375        ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1376        ftrace_mod->enable = enable;
1377
1378        if (!ftrace_mod->func || !ftrace_mod->module)
1379                goto out_free;
1380
1381        list_add(&ftrace_mod->list, mod_head);
1382
1383        return 0;
1384
1385 out_free:
1386        free_ftrace_mod(ftrace_mod);
1387
1388        return -ENOMEM;
1389}
1390
1391static struct ftrace_hash *
1392alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1393{
1394        struct ftrace_func_entry *entry;
1395        struct ftrace_hash *new_hash;
1396        int size;
1397        int ret;
1398        int i;
1399
1400        new_hash = alloc_ftrace_hash(size_bits);
1401        if (!new_hash)
1402                return NULL;
1403
1404        if (hash)
1405                new_hash->flags = hash->flags;
1406
1407        /* Empty hash? */
1408        if (ftrace_hash_empty(hash))
1409                return new_hash;
1410
1411        size = 1 << hash->size_bits;
1412        for (i = 0; i < size; i++) {
1413                hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1414                        ret = add_hash_entry(new_hash, entry->ip);
1415                        if (ret < 0)
1416                                goto free_hash;
1417                }
1418        }
1419
1420        FTRACE_WARN_ON(new_hash->count != hash->count);
1421
1422        return new_hash;
1423
1424 free_hash:
1425        free_ftrace_hash(new_hash);
1426        return NULL;
1427}
1428
1429static void
1430ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1431static void
1432ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1433
1434static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1435                                       struct ftrace_hash *new_hash);
1436
1437static struct ftrace_hash *
1438__ftrace_hash_move(struct ftrace_hash *src)
1439{
1440        struct ftrace_func_entry *entry;
1441        struct hlist_node *tn;
1442        struct hlist_head *hhd;
1443        struct ftrace_hash *new_hash;
1444        int size = src->count;
1445        int bits = 0;
1446        int i;
1447
1448        /*
1449         * If the new source is empty, just return the empty_hash.
1450         */
1451        if (ftrace_hash_empty(src))
1452                return EMPTY_HASH;
1453
1454        /*
1455         * Make the hash size about 1/2 the # found
1456         */
1457        for (size /= 2; size; size >>= 1)
1458                bits++;
1459
1460        /* Don't allocate too much */
1461        if (bits > FTRACE_HASH_MAX_BITS)
1462                bits = FTRACE_HASH_MAX_BITS;
1463
1464        new_hash = alloc_ftrace_hash(bits);
1465        if (!new_hash)
1466                return NULL;
1467
1468        new_hash->flags = src->flags;
1469
1470        size = 1 << src->size_bits;
1471        for (i = 0; i < size; i++) {
1472                hhd = &src->buckets[i];
1473                hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1474                        remove_hash_entry(src, entry);
1475                        __add_hash_entry(new_hash, entry);
1476                }
1477        }
1478
1479        return new_hash;
1480}
1481
1482static int
1483ftrace_hash_move(struct ftrace_ops *ops, int enable,
1484                 struct ftrace_hash **dst, struct ftrace_hash *src)
1485{
1486        struct ftrace_hash *new_hash;
1487        int ret;
1488
1489        /* Reject setting notrace hash on IPMODIFY ftrace_ops */
1490        if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1491                return -EINVAL;
1492
1493        new_hash = __ftrace_hash_move(src);
1494        if (!new_hash)
1495                return -ENOMEM;
1496
1497        /* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1498        if (enable) {
1499                /* IPMODIFY should be updated only when filter_hash updating */
1500                ret = ftrace_hash_ipmodify_update(ops, new_hash);
1501                if (ret < 0) {
1502                        free_ftrace_hash(new_hash);
1503                        return ret;
1504                }
1505        }
1506
1507        /*
1508         * Remove the current set, update the hash and add
1509         * them back.
1510         */
1511        ftrace_hash_rec_disable_modify(ops, enable);
1512
1513        rcu_assign_pointer(*dst, new_hash);
1514
1515        ftrace_hash_rec_enable_modify(ops, enable);
1516
1517        return 0;
1518}
1519
1520static bool hash_contains_ip(unsigned long ip,
1521                             struct ftrace_ops_hash *hash)
1522{
1523        /*
1524         * The function record is a match if it exists in the filter
1525         * hash and not in the notrace hash. Note, an emty hash is
1526         * considered a match for the filter hash, but an empty
1527         * notrace hash is considered not in the notrace hash.
1528         */
1529        return (ftrace_hash_empty(hash->filter_hash) ||
1530                __ftrace_lookup_ip(hash->filter_hash, ip)) &&
1531                (ftrace_hash_empty(hash->notrace_hash) ||
1532                 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1533}
1534
1535/*
1536 * Test the hashes for this ops to see if we want to call
1537 * the ops->func or not.
1538 *
1539 * It's a match if the ip is in the ops->filter_hash or
1540 * the filter_hash does not exist or is empty,
1541 *  AND
1542 * the ip is not in the ops->notrace_hash.
1543 *
1544 * This needs to be called with preemption disabled as
1545 * the hashes are freed with call_rcu_sched().
1546 */
1547static int
1548ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1549{
1550        struct ftrace_ops_hash hash;
1551        int ret;
1552
1553#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1554        /*
1555         * There's a small race when adding ops that the ftrace handler
1556         * that wants regs, may be called without them. We can not
1557         * allow that handler to be called if regs is NULL.
1558         */
1559        if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1560                return 0;
1561#endif
1562
1563        rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1564        rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1565
1566        if (hash_contains_ip(ip, &hash))
1567                ret = 1;
1568        else
1569                ret = 0;
1570
1571        return ret;
1572}
1573
1574/*
1575 * This is a double for. Do not use 'break' to break out of the loop,
1576 * you must use a goto.
1577 */
1578#define do_for_each_ftrace_rec(pg, rec)                                 \
1579        for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
1580                int _____i;                                             \
1581                for (_____i = 0; _____i < pg->index; _____i++) {        \
1582                        rec = &pg->records[_____i];
1583
1584#define while_for_each_ftrace_rec()             \
1585                }                               \
1586        }
1587
1588
1589static int ftrace_cmp_recs(const void *a, const void *b)
1590{
1591        const struct dyn_ftrace *key = a;
1592        const struct dyn_ftrace *rec = b;
1593
1594        if (key->flags < rec->ip)
1595                return -1;
1596        if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1597                return 1;
1598        return 0;
1599}
1600
1601/**
1602 * ftrace_location_range - return the first address of a traced location
1603 *      if it touches the given ip range
1604 * @start: start of range to search.
1605 * @end: end of range to search (inclusive). @end points to the last byte
1606 *      to check.
1607 *
1608 * Returns rec->ip if the related ftrace location is a least partly within
1609 * the given address range. That is, the first address of the instruction
1610 * that is either a NOP or call to the function tracer. It checks the ftrace
1611 * internal tables to determine if the address belongs or not.
1612 */
1613unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1614{
1615        struct ftrace_page *pg;
1616        struct dyn_ftrace *rec;
1617        struct dyn_ftrace key;
1618
1619        key.ip = start;
1620        key.flags = end;        /* overload flags, as it is unsigned long */
1621
1622        for (pg = ftrace_pages_start; pg; pg = pg->next) {
1623                if (end < pg->records[0].ip ||
1624                    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1625                        continue;
1626                rec = bsearch(&key, pg->records, pg->index,
1627                              sizeof(struct dyn_ftrace),
1628                              ftrace_cmp_recs);
1629                if (rec)
1630                        return rec->ip;
1631        }
1632
1633        return 0;
1634}
1635
1636/**
1637 * ftrace_location - return true if the ip giving is a traced location
1638 * @ip: the instruction pointer to check
1639 *
1640 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1641 * That is, the instruction that is either a NOP or call to
1642 * the function tracer. It checks the ftrace internal tables to
1643 * determine if the address belongs or not.
1644 */
1645unsigned long ftrace_location(unsigned long ip)
1646{
1647        return ftrace_location_range(ip, ip);
1648}
1649
1650/**
1651 * ftrace_text_reserved - return true if range contains an ftrace location
1652 * @start: start of range to search
1653 * @end: end of range to search (inclusive). @end points to the last byte to check.
1654 *
1655 * Returns 1 if @start and @end contains a ftrace location.
1656 * That is, the instruction that is either a NOP or call to
1657 * the function tracer. It checks the ftrace internal tables to
1658 * determine if the address belongs or not.
1659 */
1660int ftrace_text_reserved(const void *start, const void *end)
1661{
1662        unsigned long ret;
1663
1664        ret = ftrace_location_range((unsigned long)start,
1665                                    (unsigned long)end);
1666
1667        return (int)!!ret;
1668}
1669
1670/* Test if ops registered to this rec needs regs */
1671static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1672{
1673        struct ftrace_ops *ops;
1674        bool keep_regs = false;
1675
1676        for (ops = ftrace_ops_list;
1677             ops != &ftrace_list_end; ops = ops->next) {
1678                /* pass rec in as regs to have non-NULL val */
1679                if (ftrace_ops_test(ops, rec->ip, rec)) {
1680                        if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1681                                keep_regs = true;
1682                                break;
1683                        }
1684                }
1685        }
1686
1687        return  keep_regs;
1688}
1689
1690static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1691                                     int filter_hash,
1692                                     bool inc)
1693{
1694        struct ftrace_hash *hash;
1695        struct ftrace_hash *other_hash;
1696        struct ftrace_page *pg;
1697        struct dyn_ftrace *rec;
1698        bool update = false;
1699        int count = 0;
1700        int all = false;
1701
1702        /* Only update if the ops has been registered */
1703        if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1704                return false;
1705
1706        /*
1707         * In the filter_hash case:
1708         *   If the count is zero, we update all records.
1709         *   Otherwise we just update the items in the hash.
1710         *
1711         * In the notrace_hash case:
1712         *   We enable the update in the hash.
1713         *   As disabling notrace means enabling the tracing,
1714         *   and enabling notrace means disabling, the inc variable
1715         *   gets inversed.
1716         */
1717        if (filter_hash) {
1718                hash = ops->func_hash->filter_hash;
1719                other_hash = ops->func_hash->notrace_hash;
1720                if (ftrace_hash_empty(hash))
1721                        all = true;
1722        } else {
1723                inc = !inc;
1724                hash = ops->func_hash->notrace_hash;
1725                other_hash = ops->func_hash->filter_hash;
1726                /*
1727                 * If the notrace hash has no items,
1728                 * then there's nothing to do.
1729                 */
1730                if (ftrace_hash_empty(hash))
1731                        return false;
1732        }
1733
1734        do_for_each_ftrace_rec(pg, rec) {
1735                int in_other_hash = 0;
1736                int in_hash = 0;
1737                int match = 0;
1738
1739                if (rec->flags & FTRACE_FL_DISABLED)
1740                        continue;
1741
1742                if (all) {
1743                        /*
1744                         * Only the filter_hash affects all records.
1745                         * Update if the record is not in the notrace hash.
1746                         */
1747                        if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1748                                match = 1;
1749                } else {
1750                        in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1751                        in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1752
1753                        /*
1754                         * If filter_hash is set, we want to match all functions
1755                         * that are in the hash but not in the other hash.
1756                         *
1757                         * If filter_hash is not set, then we are decrementing.
1758                         * That means we match anything that is in the hash
1759                         * and also in the other_hash. That is, we need to turn
1760                         * off functions in the other hash because they are disabled
1761                         * by this hash.
1762                         */
1763                        if (filter_hash && in_hash && !in_other_hash)
1764                                match = 1;
1765                        else if (!filter_hash && in_hash &&
1766                                 (in_other_hash || ftrace_hash_empty(other_hash)))
1767                                match = 1;
1768                }
1769                if (!match)
1770                        continue;
1771
1772                if (inc) {
1773                        rec->flags++;
1774                        if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1775                                return false;
1776
1777                        /*
1778                         * If there's only a single callback registered to a
1779                         * function, and the ops has a trampoline registered
1780                         * for it, then we can call it directly.
1781                         */
1782                        if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1783                                rec->flags |= FTRACE_FL_TRAMP;
1784                        else
1785                                /*
1786                                 * If we are adding another function callback
1787                                 * to this function, and the previous had a
1788                                 * custom trampoline in use, then we need to go
1789                                 * back to the default trampoline.
1790                                 */
1791                                rec->flags &= ~FTRACE_FL_TRAMP;
1792
1793                        /*
1794                         * If any ops wants regs saved for this function
1795                         * then all ops will get saved regs.
1796                         */
1797                        if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1798                                rec->flags |= FTRACE_FL_REGS;
1799                } else {
1800                        if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1801                                return false;
1802                        rec->flags--;
1803
1804                        /*
1805                         * If the rec had REGS enabled and the ops that is
1806                         * being removed had REGS set, then see if there is
1807                         * still any ops for this record that wants regs.
1808                         * If not, we can stop recording them.
1809                         */
1810                        if (ftrace_rec_count(rec) > 0 &&
1811                            rec->flags & FTRACE_FL_REGS &&
1812                            ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1813                                if (!test_rec_ops_needs_regs(rec))
1814                                        rec->flags &= ~FTRACE_FL_REGS;
1815                        }
1816
1817                        /*
1818                         * If the rec had TRAMP enabled, then it needs to
1819                         * be cleared. As TRAMP can only be enabled iff
1820                         * there is only a single ops attached to it.
1821                         * In otherwords, always disable it on decrementing.
1822                         * In the future, we may set it if rec count is
1823                         * decremented to one, and the ops that is left
1824                         * has a trampoline.
1825                         */
1826                        rec->flags &= ~FTRACE_FL_TRAMP;
1827
1828                        /*
1829                         * flags will be cleared in ftrace_check_record()
1830                         * if rec count is zero.
1831                         */
1832                }
1833                count++;
1834
1835                /* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1836                update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1837
1838                /* Shortcut, if we handled all records, we are done. */
1839                if (!all && count == hash->count)
1840                        return update;
1841        } while_for_each_ftrace_rec();
1842
1843        return update;
1844}
1845
1846static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1847                                    int filter_hash)
1848{
1849        return __ftrace_hash_rec_update(ops, filter_hash, 0);
1850}
1851
1852static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1853                                   int filter_hash)
1854{
1855        return __ftrace_hash_rec_update(ops, filter_hash, 1);
1856}
1857
1858static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1859                                          int filter_hash, int inc)
1860{
1861        struct ftrace_ops *op;
1862
1863        __ftrace_hash_rec_update(ops, filter_hash, inc);
1864
1865        if (ops->func_hash != &global_ops.local_hash)
1866                return;
1867
1868        /*
1869         * If the ops shares the global_ops hash, then we need to update
1870         * all ops that are enabled and use this hash.
1871         */
1872        do_for_each_ftrace_op(op, ftrace_ops_list) {
1873                /* Already done */
1874                if (op == ops)
1875                        continue;
1876                if (op->func_hash == &global_ops.local_hash)
1877                        __ftrace_hash_rec_update(op, filter_hash, inc);
1878        } while_for_each_ftrace_op(op);
1879}
1880
1881static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1882                                           int filter_hash)
1883{
1884        ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1885}
1886
1887static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1888                                          int filter_hash)
1889{
1890        ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1891}
1892
1893/*
1894 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1895 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1896 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1897 * Note that old_hash and new_hash has below meanings
1898 *  - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1899 *  - If the hash is EMPTY_HASH, it hits nothing
1900 *  - Anything else hits the recs which match the hash entries.
1901 */
1902static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1903                                         struct ftrace_hash *old_hash,
1904                                         struct ftrace_hash *new_hash)
1905{
1906        struct ftrace_page *pg;
1907        struct dyn_ftrace *rec, *end = NULL;
1908        int in_old, in_new;
1909
1910        /* Only update if the ops has been registered */
1911        if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1912                return 0;
1913
1914        if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1915                return 0;
1916
1917        /*
1918         * Since the IPMODIFY is a very address sensitive action, we do not
1919         * allow ftrace_ops to set all functions to new hash.
1920         */
1921        if (!new_hash || !old_hash)
1922                return -EINVAL;
1923
1924        /* Update rec->flags */
1925        do_for_each_ftrace_rec(pg, rec) {
1926
1927                if (rec->flags & FTRACE_FL_DISABLED)
1928                        continue;
1929
1930                /* We need to update only differences of filter_hash */
1931                in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1932                in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1933                if (in_old == in_new)
1934                        continue;
1935
1936                if (in_new) {
1937                        /* New entries must ensure no others are using it */
1938                        if (rec->flags & FTRACE_FL_IPMODIFY)
1939                                goto rollback;
1940                        rec->flags |= FTRACE_FL_IPMODIFY;
1941                } else /* Removed entry */
1942                        rec->flags &= ~FTRACE_FL_IPMODIFY;
1943        } while_for_each_ftrace_rec();
1944
1945        return 0;
1946
1947rollback:
1948        end = rec;
1949
1950        /* Roll back what we did above */
1951        do_for_each_ftrace_rec(pg, rec) {
1952
1953                if (rec->flags & FTRACE_FL_DISABLED)
1954                        continue;
1955
1956                if (rec == end)
1957                        goto err_out;
1958
1959                in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1960                in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1961                if (in_old == in_new)
1962                        continue;
1963
1964                if (in_new)
1965                        rec->flags &= ~FTRACE_FL_IPMODIFY;
1966                else
1967                        rec->flags |= FTRACE_FL_IPMODIFY;
1968        } while_for_each_ftrace_rec();
1969
1970err_out:
1971        return -EBUSY;
1972}
1973
1974static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1975{
1976        struct ftrace_hash *hash = ops->func_hash->filter_hash;
1977
1978        if (ftrace_hash_empty(hash))
1979                hash = NULL;
1980
1981        return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1982}
1983
1984/* Disabling always succeeds */
1985static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1986{
1987        struct ftrace_hash *hash = ops->func_hash->filter_hash;
1988
1989        if (ftrace_hash_empty(hash))
1990                hash = NULL;
1991
1992        __ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1993}
1994
1995static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1996                                       struct ftrace_hash *new_hash)
1997{
1998        struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1999
2000        if (ftrace_hash_empty(old_hash))
2001                old_hash = NULL;
2002
2003        if (ftrace_hash_empty(new_hash))
2004                new_hash = NULL;
2005
2006        return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
2007}
2008
2009static void print_ip_ins(const char *fmt, const unsigned char *p)
2010{
2011        int i;
2012
2013        printk(KERN_CONT "%s", fmt);
2014
2015        for (i = 0; i < MCOUNT_INSN_SIZE; i++)
2016                printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
2017}
2018
2019static struct ftrace_ops *
2020ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
2021static struct ftrace_ops *
2022ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
2023
2024enum ftrace_bug_type ftrace_bug_type;
2025const void *ftrace_expected;
2026
2027static void print_bug_type(void)
2028{
2029        switch (ftrace_bug_type) {
2030        case FTRACE_BUG_UNKNOWN:
2031                break;
2032        case FTRACE_BUG_INIT:
2033                pr_info("Initializing ftrace call sites\n");
2034                break;
2035        case FTRACE_BUG_NOP:
2036                pr_info("Setting ftrace call site to NOP\n");
2037                break;
2038        case FTRACE_BUG_CALL:
2039                pr_info("Setting ftrace call site to call ftrace function\n");
2040                break;
2041        case FTRACE_BUG_UPDATE:
2042                pr_info("Updating ftrace call site to call a different ftrace function\n");
2043                break;
2044        }
2045}
2046
2047/**
2048 * ftrace_bug - report and shutdown function tracer
2049 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2050 * @rec: The record that failed
2051 *
2052 * The arch code that enables or disables the function tracing
2053 * can call ftrace_bug() when it has detected a problem in
2054 * modifying the code. @failed should be one of either:
2055 * EFAULT - if the problem happens on reading the @ip address
2056 * EINVAL - if what is read at @ip is not what was expected
2057 * EPERM - if the problem happens on writting to the @ip address
2058 */
2059void ftrace_bug(int failed, struct dyn_ftrace *rec)
2060{
2061        unsigned long ip = rec ? rec->ip : 0;
2062
2063        switch (failed) {
2064        case -EFAULT:
2065                FTRACE_WARN_ON_ONCE(1);
2066                pr_info("ftrace faulted on modifying ");
2067                print_ip_sym(ip);
2068                break;
2069        case -EINVAL:
2070                FTRACE_WARN_ON_ONCE(1);
2071                pr_info("ftrace failed to modify ");
2072                print_ip_sym(ip);
2073                print_ip_ins(" actual:   ", (unsigned char *)ip);
2074                pr_cont("\n");
2075                if (ftrace_expected) {
2076                        print_ip_ins(" expected: ", ftrace_expected);
2077                        pr_cont("\n");
2078                }
2079                break;
2080        case -EPERM:
2081                FTRACE_WARN_ON_ONCE(1);
2082                pr_info("ftrace faulted on writing ");
2083                print_ip_sym(ip);
2084                break;
2085        default:
2086                FTRACE_WARN_ON_ONCE(1);
2087                pr_info("ftrace faulted on unknown error ");
2088                print_ip_sym(ip);
2089        }
2090        print_bug_type();
2091        if (rec) {
2092                struct ftrace_ops *ops = NULL;
2093
2094                pr_info("ftrace record flags: %lx\n", rec->flags);
2095                pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2096                        rec->flags & FTRACE_FL_REGS ? " R" : "  ");
2097                if (rec->flags & FTRACE_FL_TRAMP_EN) {
2098                        ops = ftrace_find_tramp_ops_any(rec);
2099                        if (ops) {
2100                                do {
2101                                        pr_cont("\ttramp: %pS (%pS)",
2102                                                (void *)ops->trampoline,
2103                                                (void *)ops->func);
2104                                        ops = ftrace_find_tramp_ops_next(rec, ops);
2105                                } while (ops);
2106                        } else
2107                                pr_cont("\ttramp: ERROR!");
2108
2109                }
2110                ip = ftrace_get_addr_curr(rec);
2111                pr_cont("\n expected tramp: %lx\n", ip);
2112        }
2113}
2114
2115static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2116{
2117        unsigned long flag = 0UL;
2118
2119        ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2120
2121        if (rec->flags & FTRACE_FL_DISABLED)
2122                return FTRACE_UPDATE_IGNORE;
2123
2124        /*
2125         * If we are updating calls:
2126         *
2127         *   If the record has a ref count, then we need to enable it
2128         *   because someone is using it.
2129         *
2130         *   Otherwise we make sure its disabled.
2131         *
2132         * If we are disabling calls, then disable all records that
2133         * are enabled.
2134         */
2135        if (enable && ftrace_rec_count(rec))
2136                flag = FTRACE_FL_ENABLED;
2137
2138        /*
2139         * If enabling and the REGS flag does not match the REGS_EN, or
2140         * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2141         * this record. Set flags to fail the compare against ENABLED.
2142         */
2143        if (flag) {
2144                if (!(rec->flags & FTRACE_FL_REGS) != 
2145                    !(rec->flags & FTRACE_FL_REGS_EN))
2146                        flag |= FTRACE_FL_REGS;
2147
2148                if (!(rec->flags & FTRACE_FL_TRAMP) != 
2149                    !(rec->flags & FTRACE_FL_TRAMP_EN))
2150                        flag |= FTRACE_FL_TRAMP;
2151        }
2152
2153        /* If the state of this record hasn't changed, then do nothing */
2154        if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2155                return FTRACE_UPDATE_IGNORE;
2156
2157        if (flag) {
2158                /* Save off if rec is being enabled (for return value) */
2159                flag ^= rec->flags & FTRACE_FL_ENABLED;
2160
2161                if (update) {
2162                        rec->flags |= FTRACE_FL_ENABLED;
2163                        if (flag & FTRACE_FL_REGS) {
2164                                if (rec->flags & FTRACE_FL_REGS)
2165                                        rec->flags |= FTRACE_FL_REGS_EN;
2166                                else
2167                                        rec->flags &= ~FTRACE_FL_REGS_EN;
2168                        }
2169                        if (flag & FTRACE_FL_TRAMP) {
2170                                if (rec->flags & FTRACE_FL_TRAMP)
2171                                        rec->flags |= FTRACE_FL_TRAMP_EN;
2172                                else
2173                                        rec->flags &= ~FTRACE_FL_TRAMP_EN;
2174                        }
2175                }
2176
2177                /*
2178                 * If this record is being updated from a nop, then
2179                 *   return UPDATE_MAKE_CALL.
2180                 * Otherwise,
2181                 *   return UPDATE_MODIFY_CALL to tell the caller to convert
2182                 *   from the save regs, to a non-save regs function or
2183                 *   vice versa, or from a trampoline call.
2184                 */
2185                if (flag & FTRACE_FL_ENABLED) {
2186                        ftrace_bug_type = FTRACE_BUG_CALL;
2187                        return FTRACE_UPDATE_MAKE_CALL;
2188                }
2189
2190                ftrace_bug_type = FTRACE_BUG_UPDATE;
2191                return FTRACE_UPDATE_MODIFY_CALL;
2192        }
2193
2194        if (update) {
2195                /* If there's no more users, clear all flags */
2196                if (!ftrace_rec_count(rec))
2197                        rec->flags = 0;
2198                else
2199                        /*
2200                         * Just disable the record, but keep the ops TRAMP
2201                         * and REGS states. The _EN flags must be disabled though.
2202                         */
2203                        rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2204                                        FTRACE_FL_REGS_EN);
2205        }
2206
2207        ftrace_bug_type = FTRACE_BUG_NOP;
2208        return FTRACE_UPDATE_MAKE_NOP;
2209}
2210
2211/**
2212 * ftrace_update_record, set a record that now is tracing or not
2213 * @rec: the record to update
2214 * @enable: set to 1 if the record is tracing, zero to force disable
2215 *
2216 * The records that represent all functions that can be traced need
2217 * to be updated when tracing has been enabled.
2218 */
2219int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2220{
2221        return ftrace_check_record(rec, enable, 1);
2222}
2223
2224/**
2225 * ftrace_test_record, check if the record has been enabled or not
2226 * @rec: the record to test
2227 * @enable: set to 1 to check if enabled, 0 if it is disabled
2228 *
2229 * The arch code may need to test if a record is already set to
2230 * tracing to determine how to modify the function code that it
2231 * represents.
2232 */
2233int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2234{
2235        return ftrace_check_record(rec, enable, 0);
2236}
2237
2238static struct ftrace_ops *
2239ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2240{
2241        struct ftrace_ops *op;
2242        unsigned long ip = rec->ip;
2243
2244        do_for_each_ftrace_op(op, ftrace_ops_list) {
2245
2246                if (!op->trampoline)
2247                        continue;
2248
2249                if (hash_contains_ip(ip, op->func_hash))
2250                        return op;
2251        } while_for_each_ftrace_op(op);
2252
2253        return NULL;
2254}
2255
2256static struct ftrace_ops *
2257ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2258                           struct ftrace_ops *op)
2259{
2260        unsigned long ip = rec->ip;
2261
2262        while_for_each_ftrace_op(op) {
2263
2264                if (!op->trampoline)
2265                        continue;
2266
2267                if (hash_contains_ip(ip, op->func_hash))
2268                        return op;
2269        } 
2270
2271        return NULL;
2272}
2273
2274static struct ftrace_ops *
2275ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2276{
2277        struct ftrace_ops *op;
2278        unsigned long ip = rec->ip;
2279
2280        /*
2281         * Need to check removed ops first.
2282         * If they are being removed, and this rec has a tramp,
2283         * and this rec is in the ops list, then it would be the
2284         * one with the tramp.
2285         */
2286        if (removed_ops) {
2287                if (hash_contains_ip(ip, &removed_ops->old_hash))
2288                        return removed_ops;
2289        }
2290
2291        /*
2292         * Need to find the current trampoline for a rec.
2293         * Now, a trampoline is only attached to a rec if there
2294         * was a single 'ops' attached to it. But this can be called
2295         * when we are adding another op to the rec or removing the
2296         * current one. Thus, if the op is being added, we can
2297         * ignore it because it hasn't attached itself to the rec
2298         * yet.
2299         *
2300         * If an ops is being modified (hooking to different functions)
2301         * then we don't care about the new functions that are being
2302         * added, just the old ones (that are probably being removed).
2303         *
2304         * If we are adding an ops to a function that already is using
2305         * a trampoline, it needs to be removed (trampolines are only
2306         * for single ops connected), then an ops that is not being
2307         * modified also needs to be checked.
2308         */
2309        do_for_each_ftrace_op(op, ftrace_ops_list) {
2310
2311                if (!op->trampoline)
2312                        continue;
2313
2314                /*
2315                 * If the ops is being added, it hasn't gotten to
2316                 * the point to be removed from this tree yet.
2317                 */
2318                if (op->flags & FTRACE_OPS_FL_ADDING)
2319                        continue;
2320
2321
2322                /*
2323                 * If the ops is being modified and is in the old
2324                 * hash, then it is probably being removed from this
2325                 * function.
2326                 */
2327                if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2328                    hash_contains_ip(ip, &op->old_hash))
2329                        return op;
2330                /*
2331                 * If the ops is not being added or modified, and it's
2332                 * in its normal filter hash, then this must be the one
2333                 * we want!
2334                 */
2335                if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2336                    hash_contains_ip(ip, op->func_hash))
2337                        return op;
2338
2339        } while_for_each_ftrace_op(op);
2340
2341        return NULL;
2342}
2343
2344static struct ftrace_ops *
2345ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2346{
2347        struct ftrace_ops *op;
2348        unsigned long ip = rec->ip;
2349
2350        do_for_each_ftrace_op(op, ftrace_ops_list) {
2351                /* pass rec in as regs to have non-NULL val */
2352                if (hash_contains_ip(ip, op->func_hash))
2353                        return op;
2354        } while_for_each_ftrace_op(op);
2355
2356        return NULL;
2357}
2358
2359/**
2360 * ftrace_get_addr_new - Get the call address to set to
2361 * @rec:  The ftrace record descriptor
2362 *
2363 * If the record has the FTRACE_FL_REGS set, that means that it
2364 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2365 * is not not set, then it wants to convert to the normal callback.
2366 *
2367 * Returns the address of the trampoline to set to
2368 */
2369unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2370{
2371        struct ftrace_ops *ops;
2372
2373        /* Trampolines take precedence over regs */
2374        if (rec->flags & FTRACE_FL_TRAMP) {
2375                ops = ftrace_find_tramp_ops_new(rec);
2376                if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2377                        pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2378                                (void *)rec->ip, (void *)rec->ip, rec->flags);
2379                        /* Ftrace is shutting down, return anything */
2380                        return (unsigned long)FTRACE_ADDR;
2381                }
2382                return ops->trampoline;
2383        }
2384
2385        if (rec->flags & FTRACE_FL_REGS)
2386                return (unsigned long)FTRACE_REGS_ADDR;
2387        else
2388                return (unsigned long)FTRACE_ADDR;
2389}
2390
2391/**
2392 * ftrace_get_addr_curr - Get the call address that is already there
2393 * @rec:  The ftrace record descriptor
2394 *
2395 * The FTRACE_FL_REGS_EN is set when the record already points to
2396 * a function that saves all the regs. Basically the '_EN' version
2397 * represents the current state of the function.
2398 *
2399 * Returns the address of the trampoline that is currently being called
2400 */
2401unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2402{
2403        struct ftrace_ops *ops;
2404
2405        /* Trampolines take precedence over regs */
2406        if (rec->flags & FTRACE_FL_TRAMP_EN) {
2407                ops = ftrace_find_tramp_ops_curr(rec);
2408                if (FTRACE_WARN_ON(!ops)) {
2409                        pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2410                                (void *)rec->ip, (void *)rec->ip);
2411                        /* Ftrace is shutting down, return anything */
2412                        return (unsigned long)FTRACE_ADDR;
2413                }
2414                return ops->trampoline;
2415        }
2416
2417        if (rec->flags & FTRACE_FL_REGS_EN)
2418                return (unsigned long)FTRACE_REGS_ADDR;
2419        else
2420                return (unsigned long)FTRACE_ADDR;
2421}
2422
2423static int
2424__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2425{
2426        unsigned long ftrace_old_addr;
2427        unsigned long ftrace_addr;
2428        int ret;
2429
2430        ftrace_addr = ftrace_get_addr_new(rec);
2431
2432        /* This needs to be done before we call ftrace_update_record */
2433        ftrace_old_addr = ftrace_get_addr_curr(rec);
2434
2435        ret = ftrace_update_record(rec, enable);
2436
2437        ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2438
2439        switch (ret) {
2440        case FTRACE_UPDATE_IGNORE:
2441                return 0;
2442
2443        case FTRACE_UPDATE_MAKE_CALL:
2444                ftrace_bug_type = FTRACE_BUG_CALL;
2445                return ftrace_make_call(rec, ftrace_addr);
2446
2447        case FTRACE_UPDATE_MAKE_NOP:
2448                ftrace_bug_type = FTRACE_BUG_NOP;
2449                return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2450
2451        case FTRACE_UPDATE_MODIFY_CALL:
2452                ftrace_bug_type = FTRACE_BUG_UPDATE;
2453                return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2454        }
2455
2456        return -1; /* unknow ftrace bug */
2457}
2458
2459void __weak ftrace_replace_code(int enable)
2460{
2461        struct dyn_ftrace *rec;
2462        struct ftrace_page *pg;
2463        int failed;
2464
2465        if (unlikely(ftrace_disabled))
2466                return;
2467
2468        do_for_each_ftrace_rec(pg, rec) {
2469
2470                if (rec->flags & FTRACE_FL_DISABLED)
2471                        continue;
2472
2473                failed = __ftrace_replace_code(rec, enable);
2474                if (failed) {
2475                        ftrace_bug(failed, rec);
2476                        /* Stop processing */
2477                        return;
2478                }
2479        } while_for_each_ftrace_rec();
2480}
2481
2482struct ftrace_rec_iter {
2483        struct ftrace_page      *pg;
2484        int                     index;
2485};
2486
2487/**
2488 * ftrace_rec_iter_start, start up iterating over traced functions
2489 *
2490 * Returns an iterator handle that is used to iterate over all
2491 * the records that represent address locations where functions
2492 * are traced.
2493 *
2494 * May return NULL if no records are available.
2495 */
2496struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2497{
2498        /*
2499         * We only use a single iterator.
2500         * Protected by the ftrace_lock mutex.
2501         */
2502        static struct ftrace_rec_iter ftrace_rec_iter;
2503        struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2504
2505        iter->pg = ftrace_pages_start;
2506        iter->index = 0;
2507
2508        /* Could have empty pages */
2509        while (iter->pg && !iter->pg->index)
2510                iter->pg = iter->pg->next;
2511
2512        if (!iter->pg)
2513                return NULL;
2514
2515        return iter;
2516}
2517
2518/**
2519 * ftrace_rec_iter_next, get the next record to process.
2520 * @iter: The handle to the iterator.
2521 *
2522 * Returns the next iterator after the given iterator @iter.
2523 */
2524struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2525{
2526        iter->index++;
2527
2528        if (iter->index >= iter->pg->index) {
2529                iter->pg = iter->pg->next;
2530                iter->index = 0;
2531
2532                /* Could have empty pages */
2533                while (iter->pg && !iter->pg->index)
2534                        iter->pg = iter->pg->next;
2535        }
2536
2537        if (!iter->pg)
2538                return NULL;
2539
2540        return iter;
2541}
2542
2543/**
2544 * ftrace_rec_iter_record, get the record at the iterator location
2545 * @iter: The current iterator location
2546 *
2547 * Returns the record that the current @iter is at.
2548 */
2549struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2550{
2551        return &iter->pg->records[iter->index];
2552}
2553
2554static int
2555ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2556{
2557        int ret;
2558
2559        if (unlikely(ftrace_disabled))
2560                return 0;
2561
2562        ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2563        if (ret) {
2564                ftrace_bug_type = FTRACE_BUG_INIT;
2565                ftrace_bug(ret, rec);
2566                return 0;
2567        }
2568        return 1;
2569}
2570
2571/*
2572 * archs can override this function if they must do something
2573 * before the modifying code is performed.
2574 */
2575int __weak ftrace_arch_code_modify_prepare(void)
2576{
2577        return 0;
2578}
2579
2580/*
2581 * archs can override this function if they must do something
2582 * after the modifying code is performed.
2583 */
2584int __weak ftrace_arch_code_modify_post_process(void)
2585{
2586        return 0;
2587}
2588
2589void ftrace_modify_all_code(int command)
2590{
2591        int update = command & FTRACE_UPDATE_TRACE_FUNC;
2592        int err = 0;
2593
2594        /*
2595         * If the ftrace_caller calls a ftrace_ops func directly,
2596         * we need to make sure that it only traces functions it
2597         * expects to trace. When doing the switch of functions,
2598         * we need to update to the ftrace_ops_list_func first
2599         * before the transition between old and new calls are set,
2600         * as the ftrace_ops_list_func will check the ops hashes
2601         * to make sure the ops are having the right functions
2602         * traced.
2603         */
2604        if (update) {
2605                err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2606                if (FTRACE_WARN_ON(err))
2607                        return;
2608        }
2609
2610        if (command & FTRACE_UPDATE_CALLS)
2611                ftrace_replace_code(1);
2612        else if (command & FTRACE_DISABLE_CALLS)
2613                ftrace_replace_code(0);
2614
2615        if (update && ftrace_trace_function != ftrace_ops_list_func) {
2616                function_trace_op = set_function_trace_op;
2617                smp_wmb();
2618                /* If irqs are disabled, we are in stop machine */
2619                if (!irqs_disabled())
2620                        smp_call_function(ftrace_sync_ipi, NULL, 1);
2621                err = ftrace_update_ftrace_func(ftrace_trace_function);
2622                if (FTRACE_WARN_ON(err))
2623                        return;
2624        }
2625
2626        if (command & FTRACE_START_FUNC_RET)
2627                err = ftrace_enable_ftrace_graph_caller();
2628        else if (command & FTRACE_STOP_FUNC_RET)
2629                err = ftrace_disable_ftrace_graph_caller();
2630        FTRACE_WARN_ON(err);
2631}
2632
2633static int __ftrace_modify_code(void *data)
2634{
2635        int *command = data;
2636
2637        ftrace_modify_all_code(*command);
2638
2639        return 0;
2640}
2641
2642/**
2643 * ftrace_run_stop_machine, go back to the stop machine method
2644 * @command: The command to tell ftrace what to do
2645 *
2646 * If an arch needs to fall back to the stop machine method, the
2647 * it can call this function.
2648 */
2649void ftrace_run_stop_machine(int command)
2650{
2651        stop_machine(__ftrace_modify_code, &command, NULL);
2652}
2653
2654/**
2655 * arch_ftrace_update_code, modify the code to trace or not trace
2656 * @command: The command that needs to be done
2657 *
2658 * Archs can override this function if it does not need to
2659 * run stop_machine() to modify code.
2660 */
2661void __weak arch_ftrace_update_code(int command)
2662{
2663        ftrace_run_stop_machine(command);
2664}
2665
2666static void ftrace_run_update_code(int command)
2667{
2668        int ret;
2669
2670        ret = ftrace_arch_code_modify_prepare();
2671        FTRACE_WARN_ON(ret);
2672        if (ret)
2673                return;
2674
2675        /*
2676         * By default we use stop_machine() to modify the code.
2677         * But archs can do what ever they want as long as it
2678         * is safe. The stop_machine() is the safest, but also
2679         * produces the most overhead.
2680         */
2681        arch_ftrace_update_code(command);
2682
2683        ret = ftrace_arch_code_modify_post_process();
2684        FTRACE_WARN_ON(ret);
2685}
2686
2687static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2688                                   struct ftrace_ops_hash *old_hash)
2689{
2690        ops->flags |= FTRACE_OPS_FL_MODIFYING;
2691        ops->old_hash.filter_hash = old_hash->filter_hash;
2692        ops->old_hash.notrace_hash = old_hash->notrace_hash;
2693        ftrace_run_update_code(command);
2694        ops->old_hash.filter_hash = NULL;
2695        ops->old_hash.notrace_hash = NULL;
2696        ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2697}
2698
2699static ftrace_func_t saved_ftrace_func;
2700static int ftrace_start_up;
2701
2702void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2703{
2704}
2705
2706static void ftrace_startup_enable(int command)
2707{
2708        if (saved_ftrace_func != ftrace_trace_function) {
2709                saved_ftrace_func = ftrace_trace_function;
2710                command |= FTRACE_UPDATE_TRACE_FUNC;
2711        }
2712
2713        if (!command || !ftrace_enabled)
2714                return;
2715
2716        ftrace_run_update_code(command);
2717}
2718
2719static void ftrace_startup_all(int command)
2720{
2721        update_all_ops = true;
2722        ftrace_startup_enable(command);
2723        update_all_ops = false;
2724}
2725
2726static int ftrace_startup(struct ftrace_ops *ops, int command)
2727{
2728        int ret;
2729
2730        if (unlikely(ftrace_disabled))
2731                return -ENODEV;
2732
2733        ret = __register_ftrace_function(ops);
2734        if (ret)
2735                return ret;
2736
2737        ftrace_start_up++;
2738
2739        /*
2740         * Note that ftrace probes uses this to start up
2741         * and modify functions it will probe. But we still
2742         * set the ADDING flag for modification, as probes
2743         * do not have trampolines. If they add them in the
2744         * future, then the probes will need to distinguish
2745         * between adding and updating probes.
2746         */
2747        ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2748
2749        ret = ftrace_hash_ipmodify_enable(ops);
2750        if (ret < 0) {
2751                /* Rollback registration process */
2752                __unregister_ftrace_function(ops);
2753                ftrace_start_up--;
2754                ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2755                return ret;
2756        }
2757
2758        if (ftrace_hash_rec_enable(ops, 1))
2759                command |= FTRACE_UPDATE_CALLS;
2760
2761        ftrace_startup_enable(command);
2762
2763        ops->flags &= ~FTRACE_OPS_FL_ADDING;
2764
2765        return 0;
2766}
2767
2768static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2769{
2770        int ret;
2771
2772        if (unlikely(ftrace_disabled))
2773                return -ENODEV;
2774
2775        ret = __unregister_ftrace_function(ops);
2776        if (ret)
2777                return ret;
2778
2779        ftrace_start_up--;
2780        /*
2781         * Just warn in case of unbalance, no need to kill ftrace, it's not
2782         * critical but the ftrace_call callers may be never nopped again after
2783         * further ftrace uses.
2784         */
2785        WARN_ON_ONCE(ftrace_start_up < 0);
2786
2787        /* Disabling ipmodify never fails */
2788        ftrace_hash_ipmodify_disable(ops);
2789
2790        if (ftrace_hash_rec_disable(ops, 1))
2791                command |= FTRACE_UPDATE_CALLS;
2792
2793        ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2794
2795        if (saved_ftrace_func != ftrace_trace_function) {
2796                saved_ftrace_func = ftrace_trace_function;
2797                command |= FTRACE_UPDATE_TRACE_FUNC;
2798        }
2799
2800        if (!command || !ftrace_enabled) {
2801                /*
2802                 * If these are dynamic or per_cpu ops, they still
2803                 * need their data freed. Since, function tracing is
2804                 * not currently active, we can just free them
2805                 * without synchronizing all CPUs.
2806                 */
2807                if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
2808                        goto free_ops;
2809
2810                return 0;
2811        }
2812
2813        /*
2814         * If the ops uses a trampoline, then it needs to be
2815         * tested first on update.
2816         */
2817        ops->flags |= FTRACE_OPS_FL_REMOVING;
2818        removed_ops = ops;
2819
2820        /* The trampoline logic checks the old hashes */
2821        ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2822        ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2823
2824        ftrace_run_update_code(command);
2825
2826        /*
2827         * If there's no more ops registered with ftrace, run a
2828         * sanity check to make sure all rec flags are cleared.
2829         */
2830        if (rcu_dereference_protected(ftrace_ops_list,
2831                        lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
2832                struct ftrace_page *pg;
2833                struct dyn_ftrace *rec;
2834
2835                do_for_each_ftrace_rec(pg, rec) {
2836                        if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_FL_DISABLED))
2837                                pr_warn("  %pS flags:%lx\n",
2838                                        (void *)rec->ip, rec->flags);
2839                } while_for_each_ftrace_rec();
2840        }
2841
2842        ops->old_hash.filter_hash = NULL;
2843        ops->old_hash.notrace_hash = NULL;
2844
2845        removed_ops = NULL;
2846        ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2847
2848        /*
2849         * Dynamic ops may be freed, we must make sure that all
2850         * callers are done before leaving this function.
2851         * The same goes for freeing the per_cpu data of the per_cpu
2852         * ops.
2853         */
2854        if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
2855                /*
2856                 * We need to do a hard force of sched synchronization.
2857                 * This is because we use preempt_disable() to do RCU, but
2858                 * the function tracers can be called where RCU is not watching
2859                 * (like before user_exit()). We can not rely on the RCU
2860                 * infrastructure to do the synchronization, thus we must do it
2861                 * ourselves.
2862                 */
2863                schedule_on_each_cpu(ftrace_sync);
2864
2865                /*
2866                 * When the kernel is preeptive, tasks can be preempted
2867                 * while on a ftrace trampoline. Just scheduling a task on
2868                 * a CPU is not good enough to flush them. Calling
2869                 * synchornize_rcu_tasks() will wait for those tasks to
2870                 * execute and either schedule voluntarily or enter user space.
2871                 */
2872                if (IS_ENABLED(CONFIG_PREEMPT))
2873                        synchronize_rcu_tasks();
2874
2875 free_ops:
2876                arch_ftrace_trampoline_free(ops);
2877        }
2878
2879        return 0;
2880}
2881
2882static void ftrace_startup_sysctl(void)
2883{
2884        int command;
2885
2886        if (unlikely(ftrace_disabled))
2887                return;
2888
2889        /* Force update next time */
2890        saved_ftrace_func = NULL;
2891        /* ftrace_start_up is true if we want ftrace running */
2892        if (ftrace_start_up) {
2893                command = FTRACE_UPDATE_CALLS;
2894                if (ftrace_graph_active)
2895                        command |= FTRACE_START_FUNC_RET;
2896                ftrace_startup_enable(command);
2897        }
2898}
2899
2900static void ftrace_shutdown_sysctl(void)
2901{
2902        int command;
2903
2904        if (unlikely(ftrace_disabled))
2905                return;
2906
2907        /* ftrace_start_up is true if ftrace is running */
2908        if (ftrace_start_up) {
2909                command = FTRACE_DISABLE_CALLS;
2910                if (ftrace_graph_active)
2911                        command |= FTRACE_STOP_FUNC_RET;
2912                ftrace_run_update_code(command);
2913        }
2914}
2915
2916static u64              ftrace_update_time;
2917unsigned long           ftrace_update_tot_cnt;
2918
2919static inline int ops_traces_mod(struct ftrace_ops *ops)
2920{
2921        /*
2922         * Filter_hash being empty will default to trace module.
2923         * But notrace hash requires a test of individual module functions.
2924         */
2925        return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2926                ftrace_hash_empty(ops->func_hash->notrace_hash);
2927}
2928
2929/*
2930 * Check if the current ops references the record.
2931 *
2932 * If the ops traces all functions, then it was already accounted for.
2933 * If the ops does not trace the current record function, skip it.
2934 * If the ops ignores the function via notrace filter, skip it.
2935 */
2936static inline bool
2937ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2938{
2939        /* If ops isn't enabled, ignore it */
2940        if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2941                return 0;
2942
2943        /* If ops traces all then it includes this function */
2944        if (ops_traces_mod(ops))
2945                return 1;
2946
2947        /* The function must be in the filter */
2948        if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2949            !__ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2950                return 0;
2951
2952        /* If in notrace hash, we ignore it too */
2953        if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2954                return 0;
2955
2956        return 1;
2957}
2958
2959static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2960{
2961        struct ftrace_page *pg;
2962        struct dyn_ftrace *p;
2963        u64 start, stop;
2964        unsigned long update_cnt = 0;
2965        unsigned long rec_flags = 0;
2966        int i;
2967
2968        start = ftrace_now(raw_smp_processor_id());
2969
2970        /*
2971         * When a module is loaded, this function is called to convert
2972         * the calls to mcount in its text to nops, and also to create
2973         * an entry in the ftrace data. Now, if ftrace is activated
2974         * after this call, but before the module sets its text to
2975         * read-only, the modification of enabling ftrace can fail if
2976         * the read-only is done while ftrace is converting the calls.
2977         * To prevent this, the module's records are set as disabled
2978         * and will be enabled after the call to set the module's text
2979         * to read-only.
2980         */
2981        if (mod)
2982                rec_flags |= FTRACE_FL_DISABLED;
2983
2984        for (pg = new_pgs; pg; pg = pg->next) {
2985
2986                for (i = 0; i < pg->index; i++) {
2987
2988                        /* If something went wrong, bail without enabling anything */
2989                        if (unlikely(ftrace_disabled))
2990                                return -1;
2991
2992                        p = &pg->records[i];
2993                        p->flags = rec_flags;
2994
2995                        /*
2996                         * Do the initial record conversion from mcount jump
2997                         * to the NOP instructions.
2998                         */
2999                        if (!ftrace_code_disable(mod, p))
3000                                break;
3001
3002                        update_cnt++;
3003                }
3004        }
3005
3006        stop = ftrace_now(raw_smp_processor_id());
3007        ftrace_update_time = stop - start;
3008        ftrace_update_tot_cnt += update_cnt;
3009
3010        return 0;
3011}
3012
3013static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3014{
3015        int order;
3016        int cnt;
3017
3018        if (WARN_ON(!count))
3019                return -EINVAL;
3020
3021        order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
3022
3023        /*
3024         * We want to fill as much as possible. No more than a page
3025         * may be empty.
3026         */
3027        while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
3028                order--;
3029
3030 again:
3031        pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3032
3033        if (!pg->records) {
3034                /* if we can't allocate this size, try something smaller */
3035                if (!order)
3036                        return -ENOMEM;
3037                order >>= 1;
3038                goto again;
3039        }
3040
3041        cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3042        pg->size = cnt;
3043
3044        if (cnt > count)
3045                cnt = count;
3046
3047        return cnt;
3048}
3049
3050static struct ftrace_page *
3051ftrace_allocate_pages(unsigned long num_to_init)
3052{
3053        struct ftrace_page *start_pg;
3054        struct ftrace_page *pg;
3055        int order;
3056        int cnt;
3057
3058        if (!num_to_init)
3059                return 0;
3060
3061        start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3062        if (!pg)
3063                return NULL;
3064
3065        /*
3066         * Try to allocate as much as possible in one continues
3067         * location that fills in all of the space. We want to
3068         * waste as little space as possible.
3069         */
3070        for (;;) {
3071                cnt = ftrace_allocate_records(pg, num_to_init);
3072                if (cnt < 0)
3073                        goto free_pages;
3074
3075                num_to_init -= cnt;
3076                if (!num_to_init)
3077                        break;
3078
3079                pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3080                if (!pg->next)
3081                        goto free_pages;
3082
3083                pg = pg->next;
3084        }
3085
3086        return start_pg;
3087
3088 free_pages:
3089        pg = start_pg;
3090        while (pg) {
3091                order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3092                free_pages((unsigned long)pg->records, order);
3093                start_pg = pg->next;
3094                kfree(pg);
3095                pg = start_pg;
3096        }
3097        pr_info("ftrace: FAILED to allocate memory for functions\n");
3098        return NULL;
3099}
3100
3101#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3102
3103struct ftrace_iterator {
3104        loff_t                          pos;
3105        loff_t                          func_pos;
3106        loff_t                          mod_pos;
3107        struct ftrace_page              *pg;
3108        struct dyn_ftrace               *func;
3109        struct ftrace_func_probe        *probe;
3110        struct ftrace_func_entry        *probe_entry;
3111        struct trace_parser             parser;
3112        struct ftrace_hash              *hash;
3113        struct ftrace_ops               *ops;
3114        struct trace_array              *tr;
3115        struct list_head                *mod_list;
3116        int                             pidx;
3117        int                             idx;
3118        unsigned                        flags;
3119};
3120
3121static void *
3122t_probe_next(struct seq_file *m, loff_t *pos)
3123{
3124        struct ftrace_iterator *iter = m->private;
3125        struct trace_array *tr = iter->ops->private;
3126        struct list_head *func_probes;
3127        struct ftrace_hash *hash;
3128        struct list_head *next;
3129        struct hlist_node *hnd = NULL;
3130        struct hlist_head *hhd;
3131        int size;
3132
3133        (*pos)++;
3134        iter->pos = *pos;
3135
3136        if (!tr)
3137                return NULL;
3138
3139        func_probes = &tr->func_probes;
3140        if (list_empty(func_probes))
3141                return NULL;
3142
3143        if (!iter->probe) {
3144                next = func_probes->next;
3145                iter->probe = list_entry(next, struct ftrace_func_probe, list);
3146        }
3147
3148        if (iter->probe_entry)
3149                hnd = &iter->probe_entry->hlist;
3150
3151        hash = iter->probe->ops.func_hash->filter_hash;
3152        size = 1 << hash->size_bits;
3153
3154 retry:
3155        if (iter->pidx >= size) {
3156                if (iter->probe->list.next == func_probes)
3157                        return NULL;
3158                next = iter->probe->list.next;
3159                iter->probe = list_entry(next, struct ftrace_func_probe, list);
3160                hash = iter->probe->ops.func_hash->filter_hash;
3161                size = 1 << hash->size_bits;
3162                iter->pidx = 0;
3163        }
3164
3165        hhd = &hash->buckets[iter->pidx];
3166
3167        if (hlist_empty(hhd)) {
3168                iter->pidx++;
3169                hnd = NULL;
3170                goto retry;
3171        }
3172
3173        if (!hnd)
3174                hnd = hhd->first;
3175        else {
3176                hnd = hnd->next;
3177                if (!hnd) {
3178                        iter->pidx++;
3179                        goto retry;
3180                }
3181        }
3182
3183        if (WARN_ON_ONCE(!hnd))
3184                return NULL;
3185
3186        iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3187
3188        return iter;
3189}
3190
3191static void *t_probe_start(struct seq_file *m, loff_t *pos)
3192{
3193        struct ftrace_iterator *iter = m->private;
3194        void *p = NULL;
3195        loff_t l;
3196
3197        if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3198                return NULL;
3199
3200        if (iter->mod_pos > *pos)
3201                return NULL;
3202
3203        iter->probe = NULL;
3204        iter->probe_entry = NULL;
3205        iter->pidx = 0;
3206        for (l = 0; l <= (*pos - iter->mod_pos); ) {
3207                p = t_probe_next(m, &l);
3208                if (!p)
3209                        break;
3210        }
3211        if (!p)
3212                return NULL;
3213
3214        /* Only set this if we have an item */
3215        iter->flags |= FTRACE_ITER_PROBE;
3216
3217        return iter;
3218}
3219
3220static int
3221t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3222{
3223        struct ftrace_func_entry *probe_entry;
3224        struct ftrace_probe_ops *probe_ops;
3225        struct ftrace_func_probe *probe;
3226
3227        probe = iter->probe;
3228        probe_entry = iter->probe_entry;
3229
3230        if (WARN_ON_ONCE(!probe || !probe_entry))
3231                return -EIO;
3232
3233        probe_ops = probe->probe_ops;
3234
3235        if (probe_ops->print)
3236                return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3237
3238        seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3239                   (void *)probe_ops->func);
3240
3241        return 0;
3242}
3243
3244static void *
3245t_mod_next(struct seq_file *m, loff_t *pos)
3246{
3247        struct ftrace_iterator *iter = m->private;
3248        struct trace_array *tr = iter->tr;
3249
3250        (*pos)++;
3251        iter->pos = *pos;
3252
3253        iter->mod_list = iter->mod_list->next;
3254
3255        if (iter->mod_list == &tr->mod_trace ||
3256            iter->mod_list == &tr->mod_notrace) {
3257                iter->flags &= ~FTRACE_ITER_MOD;
3258                return NULL;
3259        }
3260
3261        iter->mod_pos = *pos;
3262
3263        return iter;
3264}
3265
3266static void *t_mod_start(struct seq_file *m, loff_t *pos)
3267{
3268        struct ftrace_iterator *iter = m->private;
3269        void *p = NULL;
3270        loff_t l;
3271
3272        if (iter->func_pos > *pos)
3273                return NULL;
3274
3275        iter->mod_pos = iter->func_pos;
3276
3277        /* probes are only available if tr is set */
3278        if (!iter->tr)
3279                return NULL;
3280
3281        for (l = 0; l <= (*pos - iter->func_pos); ) {
3282                p = t_mod_next(m, &l);
3283                if (!p)
3284                        break;
3285        }
3286        if (!p) {
3287                iter->flags &= ~FTRACE_ITER_MOD;
3288                return t_probe_start(m, pos);
3289        }
3290
3291        /* Only set this if we have an item */
3292        iter->flags |= FTRACE_ITER_MOD;
3293
3294        return iter;
3295}
3296
3297static int
3298t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
3299{
3300        struct ftrace_mod_load *ftrace_mod;
3301        struct trace_array *tr = iter->tr;
3302
3303        if (WARN_ON_ONCE(!iter->mod_list) ||
3304                         iter->mod_list == &tr->mod_trace ||
3305                         iter->mod_list == &tr->mod_notrace)
3306                return -EIO;
3307
3308        ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
3309
3310        if (ftrace_mod->func)
3311                seq_printf(m, "%s", ftrace_mod->func);
3312        else
3313                seq_putc(m, '*');
3314
3315        seq_printf(m, ":mod:%s\n", ftrace_mod->module);
3316
3317        return 0;
3318}
3319
3320static void *
3321t_func_next(struct seq_file *m, loff_t *pos)
3322{
3323        struct ftrace_iterator *iter = m->private;
3324        struct dyn_ftrace *rec = NULL;
3325
3326        (*pos)++;
3327
3328 retry:
3329        if (iter->idx >= iter->pg->index) {
3330                if (iter->pg->next) {
3331                        iter->pg = iter->pg->next;
3332                        iter->idx = 0;
3333                        goto retry;
3334                }
3335        } else {
3336                rec = &iter->pg->records[iter->idx++];
3337                if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3338                     !ftrace_lookup_ip(iter->hash, rec->ip)) ||
3339
3340                    ((iter->flags & FTRACE_ITER_ENABLED) &&
3341                     !(rec->flags & FTRACE_FL_ENABLED))) {
3342
3343                        rec = NULL;
3344                        goto retry;
3345                }
3346        }
3347
3348        if (!rec)
3349                return NULL;
3350
3351        iter->pos = iter->func_pos = *pos;
3352        iter->func = rec;
3353
3354        return iter;
3355}
3356
3357static void *
3358t_next(struct seq_file *m, void *v, loff_t *pos)
3359{
3360        struct ftrace_iterator *iter = m->private;
3361        loff_t l = *pos; /* t_probe_start() must use original pos */
3362        void *ret;
3363
3364        if (unlikely(ftrace_disabled))
3365                return NULL;
3366
3367        if (iter->flags & FTRACE_ITER_PROBE)
3368                return t_probe_next(m, pos);
3369
3370        if (iter->flags & FTRACE_ITER_MOD)
3371                return t_mod_next(m, pos);
3372
3373        if (iter->flags & FTRACE_ITER_PRINTALL) {
3374                /* next must increment pos, and t_probe_start does not */
3375                (*pos)++;
3376                return t_mod_start(m, &l);
3377        }
3378
3379        ret = t_func_next(m, pos);
3380
3381        if (!ret)
3382                return t_mod_start(m, &l);
3383
3384        return ret;
3385}
3386
3387static void reset_iter_read(struct ftrace_iterator *iter)
3388{
3389        iter->pos = 0;
3390        iter->func_pos = 0;
3391        iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
3392}
3393
3394static void *t_start(struct seq_file *m, loff_t *pos)
3395{
3396        struct ftrace_iterator *iter = m->private;
3397        void *p = NULL;
3398        loff_t l;
3399
3400        mutex_lock(&ftrace_lock);
3401
3402        if (unlikely(ftrace_disabled))
3403                return NULL;
3404
3405        /*
3406         * If an lseek was done, then reset and start from beginning.
3407         */
3408        if (*pos < iter->pos)
3409                reset_iter_read(iter);
3410
3411        /*
3412         * For set_ftrace_filter reading, if we have the filter
3413         * off, we can short cut and just print out that all
3414         * functions are enabled.
3415         */
3416        if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3417            ftrace_hash_empty(iter->hash)) {
3418                iter->func_pos = 1; /* Account for the message */
3419                if (*pos > 0)
3420                        return t_mod_start(m, pos);
3421                iter->flags |= FTRACE_ITER_PRINTALL;
3422                /* reset in case of seek/pread */
3423                iter->flags &= ~FTRACE_ITER_PROBE;
3424                return iter;
3425        }
3426
3427        if (iter->flags & FTRACE_ITER_MOD)
3428                return t_mod_start(m, pos);
3429
3430        /*
3431         * Unfortunately, we need to restart at ftrace_pages_start
3432         * every time we let go of the ftrace_mutex. This is because
3433         * those pointers can change without the lock.
3434         */
3435        iter->pg = ftrace_pages_start;
3436        iter->idx = 0;
3437        for (l = 0; l <= *pos; ) {
3438                p = t_func_next(m, &l);
3439                if (!p)
3440                        break;
3441        }
3442
3443        if (!p)
3444                return t_mod_start(m, pos);
3445
3446        return iter;
3447}
3448
3449static void t_stop(struct seq_file *m, void *p)
3450{
3451        mutex_unlock(&ftrace_lock);
3452}
3453
3454void * __weak
3455arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3456{
3457        return NULL;
3458}
3459
3460static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3461                                struct dyn_ftrace *rec)
3462{
3463        void *ptr;
3464
3465        ptr = arch_ftrace_trampoline_func(ops, rec);
3466        if (ptr)
3467                seq_printf(m, " ->%pS", ptr);
3468}
3469
3470static int t_show(struct seq_file *m, void *v)
3471{
3472        struct ftrace_iterator *iter = m->private;
3473        struct dyn_ftrace *rec;
3474
3475        if (iter->flags & FTRACE_ITER_PROBE)
3476                return t_probe_show(m, iter);
3477
3478        if (iter->flags & FTRACE_ITER_MOD)
3479                return t_mod_show(m, iter);
3480
3481        if (iter->flags & FTRACE_ITER_PRINTALL) {
3482                if (iter->flags & FTRACE_ITER_NOTRACE)
3483                        seq_puts(m, "#### no functions disabled ####\n");
3484                else
3485                        seq_puts(m, "#### all functions enabled ####\n");
3486                return 0;
3487        }
3488
3489        rec = iter->func;
3490
3491        if (!rec)
3492                return 0;
3493
3494        seq_printf(m, "%ps", (void *)rec->ip);
3495        if (iter->flags & FTRACE_ITER_ENABLED) {
3496                struct ftrace_ops *ops;
3497
3498                seq_printf(m, " (%ld)%s%s",
3499                           ftrace_rec_count(rec),
3500                           rec->flags & FTRACE_FL_REGS ? " R" : "  ",
3501                           rec->flags & FTRACE_FL_IPMODIFY ? " I" : "  ");
3502                if (rec->flags & FTRACE_FL_TRAMP_EN) {
3503                        ops = ftrace_find_tramp_ops_any(rec);
3504                        if (ops) {
3505                                do {
3506                                        seq_printf(m, "\ttramp: %pS (%pS)",
3507                                                   (void *)ops->trampoline,
3508                                                   (void *)ops->func);
3509                                        add_trampoline_func(m, ops, rec);
3510                                        ops = ftrace_find_tramp_ops_next(rec, ops);
3511                                } while (ops);
3512                        } else
3513                                seq_puts(m, "\ttramp: ERROR!");
3514                } else {
3515                        add_trampoline_func(m, NULL, rec);
3516                }
3517        }       
3518
3519        seq_putc(m, '\n');
3520
3521        return 0;
3522}
3523
3524static const struct seq_operations show_ftrace_seq_ops = {
3525        .start = t_start,
3526        .next = t_next,
3527        .stop = t_stop,
3528        .show = t_show,
3529};
3530
3531static int
3532ftrace_avail_open(struct inode *inode, struct file *file)
3533{
3534        struct ftrace_iterator *iter;
3535
3536        if (unlikely(ftrace_disabled))
3537                return -ENODEV;
3538
3539        iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3540        if (!iter)
3541                return -ENOMEM;
3542
3543        iter->pg = ftrace_pages_start;
3544        iter->ops = &global_ops;
3545
3546        return 0;
3547}
3548
3549static int
3550ftrace_enabled_open(struct inode *inode, struct file *file)
3551{
3552        struct ftrace_iterator *iter;
3553
3554        iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3555        if (!iter)
3556                return -ENOMEM;
3557
3558        iter->pg = ftrace_pages_start;
3559        iter->flags = FTRACE_ITER_ENABLED;
3560        iter->ops = &global_ops;
3561
3562        return 0;
3563}
3564
3565/**
3566 * ftrace_regex_open - initialize function tracer filter files
3567 * @ops: The ftrace_ops that hold the hash filters
3568 * @flag: The type of filter to process
3569 * @inode: The inode, usually passed in to your open routine
3570 * @file: The file, usually passed in to your open routine
3571 *
3572 * ftrace_regex_open() initializes the filter files for the
3573 * @ops. Depending on @flag it may process the filter hash or
3574 * the notrace hash of @ops. With this called from the open
3575 * routine, you can use ftrace_filter_write() for the write
3576 * routine if @flag has FTRACE_ITER_FILTER set, or
3577 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3578 * tracing_lseek() should be used as the lseek routine, and
3579 * release must call ftrace_regex_release().
3580 */
3581int
3582ftrace_regex_open(struct ftrace_ops *ops, int flag,
3583                  struct inode *inode, struct file *file)
3584{
3585        struct ftrace_iterator *iter;
3586        struct ftrace_hash *hash;
3587        struct list_head *mod_head;
3588        struct trace_array *tr = ops->private;
3589        int ret = 0;
3590
3591        ftrace_ops_init(ops);
3592
3593        if (unlikely(ftrace_disabled))
3594                return -ENODEV;
3595
3596        iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3597        if (!iter)
3598                return -ENOMEM;
3599
3600        if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3601                kfree(iter);
3602                return -ENOMEM;
3603        }
3604
3605        iter->ops = ops;
3606        iter->flags = flag;
3607        iter->tr = tr;
3608
3609        mutex_lock(&ops->func_hash->regex_lock);
3610
3611        if (flag & FTRACE_ITER_NOTRACE) {
3612                hash = ops->func_hash->notrace_hash;
3613                mod_head = tr ? &tr->mod_notrace : NULL;
3614        } else {
3615                hash = ops->func_hash->filter_hash;
3616                mod_head = tr ? &tr->mod_trace : NULL;
3617        }
3618
3619        iter->mod_list = mod_head;
3620
3621        if (file->f_mode & FMODE_WRITE) {
3622                const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3623
3624                if (file->f_flags & O_TRUNC) {
3625                        iter->hash = alloc_ftrace_hash(size_bits);
3626                        clear_ftrace_mod_list(mod_head);
3627                } else {
3628                        iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3629                }
3630
3631                if (!iter->hash) {
3632                        trace_parser_put(&iter->parser);
3633                        kfree(iter);
3634                        ret = -ENOMEM;
3635                        goto out_unlock;
3636                }
3637        } else
3638                iter->hash = hash;
3639
3640        if (file->f_mode & FMODE_READ) {
3641                iter->pg = ftrace_pages_start;
3642
3643                ret = seq_open(file, &show_ftrace_seq_ops);
3644                if (!ret) {
3645                        struct seq_file *m = file->private_data;
3646                        m->private = iter;
3647                } else {
3648                        /* Failed */
3649                        free_ftrace_hash(iter->hash);
3650                        trace_parser_put(&iter->parser);
3651                        kfree(iter);
3652                }
3653        } else
3654                file->private_data = iter;
3655
3656 out_unlock:
3657        mutex_unlock(&ops->func_hash->regex_lock);
3658
3659        return ret;
3660}
3661
3662static int
3663ftrace_filter_open(struct inode *inode, struct file *file)
3664{
3665        struct ftrace_ops *ops = inode->i_private;
3666
3667        return ftrace_regex_open(ops,
3668                        FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
3669                        inode, file);
3670}
3671
3672static int
3673ftrace_notrace_open(struct inode *inode, struct file *file)
3674{
3675        struct ftrace_ops *ops = inode->i_private;
3676
3677        return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3678                                 inode, file);
3679}
3680
3681/* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3682struct ftrace_glob {
3683        char *search;
3684        unsigned len;
3685        int type;
3686};
3687
3688/*
3689 * If symbols in an architecture don't correspond exactly to the user-visible
3690 * name of what they represent, it is possible to define this function to
3691 * perform the necessary adjustments.
3692*/
3693char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3694{
3695        return str;
3696}
3697
3698static int ftrace_match(char *str, struct ftrace_glob *g)
3699{
3700        int matched = 0;
3701        int slen;
3702
3703        str = arch_ftrace_match_adjust(str, g->search);
3704
3705        switch (g->type) {
3706        case MATCH_FULL:
3707                if (strcmp(str, g->search) == 0)
3708                        matched = 1;
3709                break;
3710        case MATCH_FRONT_ONLY:
3711                if (strncmp(str, g->search, g->len) == 0)
3712                        matched = 1;
3713                break;
3714        case MATCH_MIDDLE_ONLY:
3715                if (strstr(str, g->search))
3716                        matched = 1;
3717                break;
3718        case MATCH_END_ONLY:
3719                slen = strlen(str);
3720                if (slen >= g->len &&
3721                    memcmp(str + slen - g->len, g->search, g->len) == 0)
3722                        matched = 1;
3723                break;
3724        case MATCH_GLOB:
3725                if (glob_match(g->search, str))
3726                        matched = 1;
3727                break;
3728        }
3729
3730        return matched;
3731}
3732
3733static int
3734enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3735{
3736        struct ftrace_func_entry *entry;
3737        int ret = 0;
3738
3739        entry = ftrace_lookup_ip(hash, rec->ip);
3740        if (clear_filter) {
3741                /* Do nothing if it doesn't exist */
3742                if (!entry)
3743                        return 0;
3744
3745                free_hash_entry(hash, entry);
3746        } else {
3747                /* Do nothing if it exists */
3748                if (entry)
3749                        return 0;
3750
3751                ret = add_hash_entry(hash, rec->ip);
3752        }
3753        return ret;
3754}
3755
3756static int
3757ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3758                struct ftrace_glob *mod_g, int exclude_mod)
3759{
3760        char str[KSYM_SYMBOL_LEN];
3761        char *modname;
3762
3763        kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3764
3765        if (mod_g) {
3766                int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3767
3768                /* blank module name to match all modules */
3769                if (!mod_g->len) {
3770                        /* blank module globbing: modname xor exclude_mod */
3771                        if (!exclude_mod != !modname)
3772                                goto func_match;
3773                        return 0;
3774                }
3775
3776                /*
3777                 * exclude_mod is set to trace everything but the given
3778                 * module. If it is set and the module matches, then
3779                 * return 0. If it is not set, and the module doesn't match
3780                 * also return 0. Otherwise, check the function to see if
3781                 * that matches.
3782                 */
3783                if (!mod_matches == !exclude_mod)
3784                        return 0;
3785func_match:
3786                /* blank search means to match all funcs in the mod */
3787                if (!func_g->len)
3788                        return 1;
3789        }
3790
3791        return ftrace_match(str, func_g);
3792}
3793
3794static int
3795match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3796{
3797        struct ftrace_page *pg;
3798        struct dyn_ftrace *rec;
3799        struct ftrace_glob func_g = { .type = MATCH_FULL };
3800        struct ftrace_glob mod_g = { .type = MATCH_FULL };
3801        struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3802        int exclude_mod = 0;
3803        int found = 0;
3804        int ret;
3805        int clear_filter = 0;
3806
3807        if (func) {
3808                func_g.type = filter_parse_regex(func, len, &func_g.search,
3809                                                 &clear_filter);
3810                func_g.len = strlen(func_g.search);
3811        }
3812
3813        if (mod) {
3814                mod_g.type = filter_parse_regex(mod, strlen(mod),
3815                                &mod_g.search, &exclude_mod);
3816                mod_g.len = strlen(mod_g.search);
3817        }
3818
3819        mutex_lock(&ftrace_lock);
3820
3821        if (unlikely(ftrace_disabled))
3822                goto out_unlock;
3823
3824        do_for_each_ftrace_rec(pg, rec) {
3825
3826                if (rec->flags & FTRACE_FL_DISABLED)
3827                        continue;
3828
3829                if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3830                        ret = enter_record(hash, rec, clear_filter);
3831                        if (ret < 0) {
3832                                found = ret;
3833                                goto out_unlock;
3834                        }
3835                        found = 1;
3836                }
3837        } while_for_each_ftrace_rec();
3838 out_unlock:
3839        mutex_unlock(&ftrace_lock);
3840
3841        return found;
3842}
3843
3844static int
3845ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3846{
3847        return match_records(hash, buff, len, NULL);
3848}
3849
3850static void ftrace_ops_update_code(struct ftrace_ops *ops,
3851                                   struct ftrace_ops_hash *old_hash)
3852{
3853        struct ftrace_ops *op;
3854
3855        if (!ftrace_enabled)
3856                return;
3857
3858        if (ops->flags & FTRACE_OPS_FL_ENABLED) {
3859                ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
3860                return;
3861        }
3862
3863        /*
3864         * If this is the shared global_ops filter, then we need to
3865         * check if there is another ops that shares it, is enabled.
3866         * If so, we still need to run the modify code.
3867         */
3868        if (ops->func_hash != &global_ops.local_hash)
3869                return;
3870
3871        do_for_each_ftrace_op(op, ftrace_ops_list) {
3872                if (op->func_hash == &global_ops.local_hash &&
3873                    op->flags & FTRACE_OPS_FL_ENABLED) {
3874                        ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
3875                        /* Only need to do this once */
3876                        return;
3877                }
3878        } while_for_each_ftrace_op(op);
3879}
3880
3881static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
3882                                           struct ftrace_hash **orig_hash,
3883                                           struct ftrace_hash *hash,
3884                                           int enable)
3885{
3886        struct ftrace_ops_hash old_hash_ops;
3887        struct ftrace_hash *old_hash;
3888        int ret;
3889
3890        old_hash = *orig_hash;
3891        old_hash_ops.filter_hash = ops->func_hash->filter_hash;
3892        old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
3893        ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3894        if (!ret) {
3895                ftrace_ops_update_code(ops, &old_hash_ops);
3896                free_ftrace_hash_rcu(old_hash);
3897        }
3898        return ret;
3899}
3900
3901static bool module_exists(const char *module)
3902{
3903        /* All modules have the symbol __this_module */
3904        const char this_mod[] = "__this_module";
3905        char modname[MAX_PARAM_PREFIX_LEN + sizeof(this_mod) + 2];
3906        unsigned long val;
3907        int n;
3908
3909        n = snprintf(modname, sizeof(modname), "%s:%s", module, this_mod);
3910
3911        if (n > sizeof(modname) - 1)
3912                return false;
3913
3914        val = module_kallsyms_lookup_name(modname);
3915        return val != 0;
3916}
3917
3918static int cache_mod(struct trace_array *tr,
3919                     const char *func, char *module, int enable)
3920{
3921        struct ftrace_mod_load *ftrace_mod, *n;
3922        struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
3923        int ret;
3924
3925        mutex_lock(&ftrace_lock);
3926
3927        /* We do not cache inverse filters */
3928        if (func[0] == '!') {
3929                func++;
3930                ret = -EINVAL;
3931
3932                /* Look to remove this hash */
3933                list_for_each_entry_safe(ftrace_mod, n, head, list) {
3934                        if (strcmp(ftrace_mod->module, module) != 0)
3935                                continue;
3936
3937                        /* no func matches all */
3938                        if (strcmp(func, "*") == 0 ||
3939                            (ftrace_mod->func &&
3940                             strcmp(ftrace_mod->func, func) == 0)) {
3941                                ret = 0;
3942                                free_ftrace_mod(ftrace_mod);
3943                                continue;
3944                        }
3945                }
3946                goto out;
3947        }
3948
3949        ret = -EINVAL;
3950        /* We only care about modules that have not been loaded yet */
3951        if (module_exists(module))
3952                goto out;
3953
3954        /* Save this string off, and execute it when the module is loaded */
3955        ret = ftrace_add_mod(tr, func, module, enable);
3956 out:
3957        mutex_unlock(&ftrace_lock);
3958
3959        return ret;
3960}
3961
3962static int
3963ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3964                 int reset, int enable);
3965
3966#ifdef CONFIG_MODULES
3967static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
3968                             char *mod, bool enable)
3969{
3970        struct ftrace_mod_load *ftrace_mod, *n;
3971        struct ftrace_hash **orig_hash, *new_hash;
3972        LIST_HEAD(process_mods);
3973        char *func;
3974        int ret;
3975
3976        mutex_lock(&ops->func_hash->regex_lock);
3977
3978        if (enable)
3979                orig_hash = &ops->func_hash->filter_hash;
3980        else
3981                orig_hash = &ops->func_hash->notrace_hash;
3982
3983        new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
3984                                              *orig_hash);
3985        if (!new_hash)
3986                goto out; /* warn? */
3987
3988        mutex_lock(&ftrace_lock);
3989
3990        list_for_each_entry_safe(ftrace_mod, n, head, list) {
3991
3992                if (strcmp(ftrace_mod->module, mod) != 0)
3993                        continue;
3994
3995                if (ftrace_mod->func)
3996                        func = kstrdup(ftrace_mod->func, GFP_KERNEL);
3997                else
3998                        func = kstrdup("*", GFP_KERNEL);
3999
4000                if (!func) /* warn? */
4001                        continue;
4002
4003                list_del(&ftrace_mod->list);
4004                list_add(&ftrace_mod->list, &process_mods);
4005
4006                /* Use the newly allocated func, as it may be "*" */
4007                kfree(ftrace_mod->func);
4008                ftrace_mod->func = func;
4009        }
4010
4011        mutex_unlock(&ftrace_lock);
4012
4013        list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
4014
4015                func = ftrace_mod->func;
4016
4017                /* Grabs ftrace_lock, which is why we have this extra step */
4018                match_records(new_hash, func, strlen(func), mod);
4019                free_ftrace_mod(ftrace_mod);
4020        }
4021
4022        if (enable && list_empty(head))
4023                new_hash->flags &= ~FTRACE_HASH_FL_MOD;
4024
4025        mutex_lock(&ftrace_lock);
4026
4027        ret = ftrace_hash_move_and_update_ops(ops, orig_hash,
4028                                              new_hash, enable);
4029        mutex_unlock(&ftrace_lock);
4030
4031 out:
4032        mutex_unlock(&ops->func_hash->regex_lock);
4033
4034        free_ftrace_hash(new_hash);
4035}
4036
4037static void process_cached_mods(const char *mod_name)
4038{
4039        struct trace_array *tr;
4040        char *mod;
4041
4042        mod = kstrdup(mod_name, GFP_KERNEL);
4043        if (!mod)
4044                return;
4045
4046        mutex_lock(&trace_types_lock);
4047        list_for_each_entry(tr, &ftrace_trace_arrays, list) {
4048                if (!list_empty(&tr->mod_trace))
4049                        process_mod_list(&tr->mod_trace, tr->ops, mod, true);
4050                if (!list_empty(&tr->mod_notrace))
4051                        process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
4052        }
4053        mutex_unlock(&trace_types_lock);
4054
4055        kfree(mod);
4056}
4057#endif
4058
4059/*
4060 * We register the module command as a template to show others how
4061 * to register the a command as well.
4062 */
4063
4064static int
4065ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
4066                    char *func_orig, char *cmd, char *module, int enable)
4067{
4068        char *func;
4069        int ret;
4070
4071        /* match_records() modifies func, and we need the original */
4072        func = kstrdup(func_orig, GFP_KERNEL);
4073        if (!func)
4074                return -ENOMEM;
4075
4076        /*
4077         * cmd == 'mod' because we only registered this func
4078         * for the 'mod' ftrace_func_command.
4079         * But if you register one func with multiple commands,
4080         * you can tell which command was used by the cmd
4081         * parameter.
4082         */
4083        ret = match_records(hash, func, strlen(func), module);
4084        kfree(func);
4085
4086        if (!ret)
4087                return cache_mod(tr, func_orig, module, enable);
4088        if (ret < 0)
4089                return ret;
4090        return 0;
4091}
4092
4093static struct ftrace_func_command ftrace_mod_cmd = {
4094        .name                   = "mod",
4095        .func                   = ftrace_mod_callback,
4096};
4097
4098static int __init ftrace_mod_cmd_init(void)
4099{
4100        return register_ftrace_command(&ftrace_mod_cmd);
4101}
4102core_initcall(ftrace_mod_cmd_init);
4103
4104static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
4105                                      struct ftrace_ops *op, struct pt_regs *pt_regs)
4106{
4107        struct ftrace_probe_ops *probe_ops;
4108        struct ftrace_func_probe *probe;
4109
4110        probe = container_of(op, struct ftrace_func_probe, ops);
4111        probe_ops = probe->probe_ops;
4112
4113        /*
4114         * Disable preemption for these calls to prevent a RCU grace
4115         * period. This syncs the hash iteration and freeing of items
4116         * on the hash. rcu_read_lock is too dangerous here.
4117         */
4118        preempt_disable_notrace();
4119        probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
4120        preempt_enable_notrace();
4121}
4122
4123struct ftrace_func_map {
4124        struct ftrace_func_entry        entry;
4125        void                            *data;
4126};
4127
4128struct ftrace_func_mapper {
4129        struct ftrace_hash              hash;
4130};
4131
4132/**
4133 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
4134 *
4135 * Returns a ftrace_func_mapper descriptor that can be used to map ips to data.
4136 */
4137struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
4138{
4139        struct ftrace_hash *hash;
4140
4141        /*
4142         * The mapper is simply a ftrace_hash, but since the entries
4143         * in the hash are not ftrace_func_entry type, we define it
4144         * as a separate structure.
4145         */
4146        hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4147        return (struct ftrace_func_mapper *)hash;
4148}
4149
4150/**
4151 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
4152 * @mapper: The mapper that has the ip maps
4153 * @ip: the instruction pointer to find the data for
4154 *
4155 * Returns the data mapped to @ip if found otherwise NULL. The return
4156 * is actually the address of the mapper data pointer. The address is
4157 * returned for use cases where the data is no bigger than a long, and
4158 * the user can use the data pointer as its data instead of having to
4159 * allocate more memory for the reference.
4160 */
4161void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
4162                                  unsigned long ip)
4163{
4164        struct ftrace_func_entry *entry;
4165        struct ftrace_func_map *map;
4166
4167        entry = ftrace_lookup_ip(&mapper->hash, ip);
4168        if (!entry)
4169                return NULL;
4170
4171        map = (struct ftrace_func_map *)entry;
4172        return &map->data;
4173}
4174
4175/**
4176 * ftrace_func_mapper_add_ip - Map some data to an ip
4177 * @mapper: The mapper that has the ip maps
4178 * @ip: The instruction pointer address to map @data to
4179 * @data: The data to map to @ip
4180 *
4181 * Returns 0 on succes otherwise an error.
4182 */
4183int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
4184                              unsigned long ip, void *data)
4185{
4186        struct ftrace_func_entry *entry;
4187        struct ftrace_func_map *map;
4188
4189        entry = ftrace_lookup_ip(&mapper->hash, ip);
4190        if (entry)
4191                return -EBUSY;
4192
4193        map = kmalloc(sizeof(*map), GFP_KERNEL);
4194        if (!map)
4195                return -ENOMEM;
4196
4197        map->entry.ip = ip;
4198        map->data = data;
4199
4200        __add_hash_entry(&mapper->hash, &map->entry);
4201
4202        return 0;
4203}
4204
4205/**
4206 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
4207 * @mapper: The mapper that has the ip maps
4208 * @ip: The instruction pointer address to remove the data from
4209 *
4210 * Returns the data if it is found, otherwise NULL.
4211 * Note, if the data pointer is used as the data itself, (see 
4212 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
4213 * if the data pointer was set to zero.
4214 */
4215void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
4216                                   unsigned long ip)
4217{
4218        struct ftrace_func_entry *entry;
4219        struct ftrace_func_map *map;
4220        void *data;
4221
4222        entry = ftrace_lookup_ip(&mapper->hash, ip);
4223        if (!entry)
4224                return NULL;
4225
4226        map = (struct ftrace_func_map *)entry;
4227        data = map->data;
4228
4229        remove_hash_entry(&mapper->hash, entry);
4230        kfree(entry);
4231
4232        return data;
4233}
4234
4235/**
4236 * free_ftrace_func_mapper - free a mapping of ips and data
4237 * @mapper: The mapper that has the ip maps
4238 * @free_func: A function to be called on each data item.
4239 *
4240 * This is used to free the function mapper. The @free_func is optional
4241 * and can be used if the data needs to be freed as well.
4242 */
4243void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
4244                             ftrace_mapper_func free_func)
4245{
4246        struct ftrace_func_entry *entry;
4247        struct ftrace_func_map *map;
4248        struct hlist_head *hhd;
4249        int size = 1 << mapper->hash.size_bits;
4250        int i;
4251
4252        if (free_func && mapper->hash.count) {
4253                for (i = 0; i < size; i++) {
4254                        hhd = &mapper->hash.buckets[i];
4255                        hlist_for_each_entry(entry, hhd, hlist) {
4256                                map = (struct ftrace_func_map *)entry;
4257                                free_func(map);
4258                        }
4259                }
4260        }
4261        free_ftrace_hash(&mapper->hash);
4262}
4263
4264static void release_probe(struct ftrace_func_probe *probe)
4265{
4266        struct ftrace_probe_ops *probe_ops;
4267
4268        mutex_lock(&ftrace_lock);
4269
4270        WARN_ON(probe->ref <= 0);
4271
4272        /* Subtract the ref that was used to protect this instance */
4273        probe->ref--;
4274
4275        if (!probe->ref) {
4276                probe_ops = probe->probe_ops;
4277                /*
4278                 * Sending zero as ip tells probe_ops to free
4279                 * the probe->data itself
4280                 */
4281                if (probe_ops->free)
4282                        probe_ops->free(probe_ops, probe->tr, 0, probe->data);
4283                list_del(&probe->list);
4284                kfree(probe);
4285        }
4286        mutex_unlock(&ftrace_lock);
4287}
4288
4289static void acquire_probe_locked(struct ftrace_func_probe *probe)
4290{
4291        /*
4292         * Add one ref to keep it from being freed when releasing the
4293         * ftrace_lock mutex.
4294         */
4295        probe->ref++;
4296}
4297
4298int
4299register_ftrace_function_probe(char *glob, struct trace_array *tr,
4300                               struct ftrace_probe_ops *probe_ops,
4301                               void *data)
4302{
4303        struct ftrace_func_entry *entry;
4304        struct ftrace_func_probe *probe;
4305        struct ftrace_hash **orig_hash;
4306        struct ftrace_hash *old_hash;
4307        struct ftrace_hash *hash;
4308        int count = 0;
4309        int size;
4310        int ret;
4311        int i;
4312
4313        if (WARN_ON(!tr))
4314                return -EINVAL;
4315
4316        /* We do not support '!' for function probes */
4317        if (WARN_ON(glob[0] == '!'))
4318                return -EINVAL;
4319
4320
4321        mutex_lock(&ftrace_lock);
4322        /* Check if the probe_ops is already registered */
4323        list_for_each_entry(probe, &tr->func_probes, list) {
4324                if (probe->probe_ops == probe_ops)
4325                        break;
4326        }
4327        if (&probe->list == &tr->func_probes) {
4328                probe = kzalloc(sizeof(*probe), GFP_KERNEL);
4329                if (!probe) {
4330                        mutex_unlock(&ftrace_lock);
4331                        return -ENOMEM;
4332                }
4333                probe->probe_ops = probe_ops;
4334                probe->ops.func = function_trace_probe_call;
4335                probe->tr = tr;
4336                ftrace_ops_init(&probe->ops);
4337                list_add(&probe->list, &tr->func_probes);
4338        }
4339
4340        acquire_probe_locked(probe);
4341
4342        mutex_unlock(&ftrace_lock);
4343
4344        mutex_lock(&probe->ops.func_hash->regex_lock);
4345
4346        orig_hash = &probe->ops.func_hash->filter_hash;
4347        old_hash = *orig_hash;
4348        hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4349
4350        ret = ftrace_match_records(hash, glob, strlen(glob));
4351
4352        /* Nothing found? */
4353        if (!ret)
4354                ret = -EINVAL;
4355
4356        if (ret < 0)
4357                goto out;
4358
4359        size = 1 << hash->size_bits;
4360        for (i = 0; i < size; i++) {
4361                hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4362                        if (ftrace_lookup_ip(old_hash, entry->ip))
4363                                continue;
4364                        /*
4365                         * The caller might want to do something special
4366                         * for each function we find. We call the callback
4367                         * to give the caller an opportunity to do so.
4368                         */
4369                        if (probe_ops->init) {
4370                                ret = probe_ops->init(probe_ops, tr,
4371                                                      entry->ip, data,
4372                                                      &probe->data);
4373                                if (ret < 0) {
4374                                        if (probe_ops->free && count)
4375                                                probe_ops->free(probe_ops, tr,
4376                                                                0, probe->data);
4377                                        probe->data = NULL;
4378                                        goto out;
4379                                }
4380                        }
4381                        count++;
4382                }
4383        }
4384
4385        mutex_lock(&ftrace_lock);
4386
4387        if (!count) {
4388                /* Nothing was added? */
4389                ret = -EINVAL;
4390                goto out_unlock;
4391        }
4392
4393        ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4394                                              hash, 1);
4395        if (ret < 0)
4396                goto err_unlock;
4397
4398        /* One ref for each new function traced */
4399        probe->ref += count;
4400
4401        if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
4402                ret = ftrace_startup(&probe->ops, 0);
4403
4404 out_unlock:
4405        mutex_unlock(&ftrace_lock);
4406
4407        if (!ret)
4408                ret = count;
4409 out:
4410        mutex_unlock(&probe->ops.func_hash->regex_lock);
4411        free_ftrace_hash(hash);
4412
4413        release_probe(probe);
4414
4415        return ret;
4416
4417 err_unlock:
4418        if (!probe_ops->free || !count)
4419                goto out_unlock;
4420
4421        /* Failed to do the move, need to call the free functions */
4422        for (i = 0; i < size; i++) {
4423                hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4424                        if (ftrace_lookup_ip(old_hash, entry->ip))
4425                                continue;
4426                        probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4427                }
4428        }
4429        goto out_unlock;
4430}
4431
4432int
4433unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
4434                                      struct ftrace_probe_ops *probe_ops)
4435{
4436        struct ftrace_ops_hash old_hash_ops;
4437        struct ftrace_func_entry *entry;
4438        struct ftrace_func_probe *probe;
4439        struct ftrace_glob func_g;
4440        struct ftrace_hash **orig_hash;
4441        struct ftrace_hash *old_hash;
4442        struct ftrace_hash *hash = NULL;
4443        struct hlist_node *tmp;
4444        struct hlist_head hhd;
4445        char str[KSYM_SYMBOL_LEN];
4446        int count = 0;
4447        int i, ret = -ENODEV;
4448        int size;
4449
4450        if (!glob || !strlen(glob) || !strcmp(glob, "*"))
4451                func_g.search = NULL;
4452        else {
4453                int not;
4454
4455                func_g.type = filter_parse_regex(glob, strlen(glob),
4456                                                 &func_g.search, &not);
4457                func_g.len = strlen(func_g.search);
4458
4459                /* we do not support '!' for function probes */
4460                if (WARN_ON(not))
4461                        return -EINVAL;
4462        }
4463
4464        mutex_lock(&ftrace_lock);
4465        /* Check if the probe_ops is already registered */
4466        list_for_each_entry(probe, &tr->func_probes, list) {
4467                if (probe->probe_ops == probe_ops)
4468                        break;
4469        }
4470        if (&probe->list == &tr->func_probes)
4471                goto err_unlock_ftrace;
4472
4473        ret = -EINVAL;
4474        if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
4475                goto err_unlock_ftrace;
4476
4477        acquire_probe_locked(probe);
4478
4479        mutex_unlock(&ftrace_lock);
4480
4481        mutex_lock(&probe->ops.func_hash->regex_lock);
4482
4483        orig_hash = &probe->ops.func_hash->filter_hash;
4484        old_hash = *orig_hash;
4485
4486        if (ftrace_hash_empty(old_hash))
4487                goto out_unlock;
4488
4489        old_hash_ops.filter_hash = old_hash;
4490        /* Probes only have filters */
4491        old_hash_ops.notrace_hash = NULL;
4492
4493        ret = -ENOMEM;
4494        hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4495        if (!hash)
4496                goto out_unlock;
4497
4498        INIT_HLIST_HEAD(&hhd);
4499
4500        size = 1 << hash->size_bits;
4501        for (i = 0; i < size; i++) {
4502                hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
4503
4504                        if (func_g.search) {
4505                                kallsyms_lookup(entry->ip, NULL, NULL,
4506                                                NULL, str);
4507                                if (!ftrace_match(str, &func_g))
4508                                        continue;
4509                        }
4510                        count++;
4511                        remove_hash_entry(hash, entry);
4512                        hlist_add_head(&entry->hlist, &hhd);
4513                }
4514        }
4515
4516        /* Nothing found? */
4517        if (!count) {
4518                ret = -EINVAL;
4519                goto out_unlock;
4520        }
4521
4522        mutex_lock(&ftrace_lock);
4523
4524        WARN_ON(probe->ref < count);
4525
4526        probe->ref -= count;
4527
4528        if (ftrace_hash_empty(hash))
4529                ftrace_shutdown(&probe->ops, 0);
4530
4531        ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4532                                              hash, 1);
4533
4534        /* still need to update the function call sites */
4535        if (ftrace_enabled && !ftrace_hash_empty(hash))
4536                ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
4537                                       &old_hash_ops);
4538        synchronize_sched();
4539
4540        hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
4541                hlist_del(&entry->hlist);
4542                if (probe_ops->free)
4543                        probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4544                kfree(entry);
4545        }
4546        mutex_unlock(&ftrace_lock);
4547
4548 out_unlock:
4549        mutex_unlock(&probe->ops.func_hash->regex_lock);
4550        free_ftrace_hash(hash);
4551
4552        release_probe(probe);
4553
4554        return ret;
4555
4556 err_unlock_ftrace:
4557        mutex_unlock(&ftrace_lock);
4558        return ret;
4559}
4560
4561void clear_ftrace_function_probes(struct trace_array *tr)
4562{
4563        struct ftrace_func_probe *probe, *n;
4564
4565        list_for_each_entry_safe(probe, n, &tr->func_probes, list)
4566                unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
4567}
4568
4569static LIST_HEAD(ftrace_commands);
4570static DEFINE_MUTEX(ftrace_cmd_mutex);
4571
4572/*
4573 * Currently we only register ftrace commands from __init, so mark this
4574 * __init too.
4575 */
4576__init int register_ftrace_command(struct ftrace_func_command *cmd)
4577{
4578        struct ftrace_func_command *p;
4579        int ret = 0;
4580
4581        mutex_lock(&ftrace_cmd_mutex);
4582        list_for_each_entry(p, &ftrace_commands, list) {
4583                if (strcmp(cmd->name, p->name) == 0) {
4584                        ret = -EBUSY;
4585                        goto out_unlock;
4586                }
4587        }
4588        list_add(&cmd->list, &ftrace_commands);
4589 out_unlock:
4590        mutex_unlock(&ftrace_cmd_mutex);
4591
4592        return ret;
4593}
4594
4595/*
4596 * Currently we only unregister ftrace commands from __init, so mark
4597 * this __init too.
4598 */
4599__init int unregister_ftrace_command(struct ftrace_func_command *cmd)
4600{
4601        struct ftrace_func_command *p, *n;
4602        int ret = -ENODEV;
4603
4604        mutex_lock(&ftrace_cmd_mutex);
4605        list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4606                if (strcmp(cmd->name, p->name) == 0) {
4607                        ret = 0;
4608                        list_del_init(&p->list);
4609                        goto out_unlock;
4610                }
4611        }
4612 out_unlock:
4613        mutex_unlock(&ftrace_cmd_mutex);
4614
4615        return ret;
4616}
4617
4618static int ftrace_process_regex(struct ftrace_iterator *iter,
4619                                char *buff, int len, int enable)
4620{
4621        struct ftrace_hash *hash = iter->hash;
4622        struct trace_array *tr = iter->ops->private;
4623        char *func, *command, *next = buff;
4624        struct ftrace_func_command *p;
4625        int ret = -EINVAL;
4626
4627        func = strsep(&next, ":");
4628
4629        if (!next) {
4630                ret = ftrace_match_records(hash, func, len);
4631                if (!ret)
4632                        ret = -EINVAL;
4633                if (ret < 0)
4634                        return ret;
4635                return 0;
4636        }
4637
4638        /* command found */
4639
4640        command = strsep(&next, ":");
4641
4642        mutex_lock(&ftrace_cmd_mutex);
4643        list_for_each_entry(p, &ftrace_commands, list) {
4644                if (strcmp(p->name, command) == 0) {
4645                        ret = p->func(tr, hash, func, command, next, enable);
4646                        goto out_unlock;
4647                }
4648        }
4649 out_unlock:
4650        mutex_unlock(&ftrace_cmd_mutex);
4651
4652        return ret;
4653}
4654
4655static ssize_t
4656ftrace_regex_write(struct file *file, const char __user *ubuf,
4657                   size_t cnt, loff_t *ppos, int enable)
4658{
4659        struct ftrace_iterator *iter;
4660        struct trace_parser *parser;
4661        ssize_t ret, read;
4662
4663        if (!cnt)
4664                return 0;
4665
4666        if (file->f_mode & FMODE_READ) {
4667                struct seq_file *m = file->private_data;
4668                iter = m->private;
4669        } else
4670                iter = file->private_data;
4671
4672        if (unlikely(ftrace_disabled))
4673                return -ENODEV;
4674
4675        /* iter->hash is a local copy, so we don't need regex_lock */
4676
4677        parser = &iter->parser;
4678        read = trace_get_user(parser, ubuf, cnt, ppos);
4679
4680        if (read >= 0 && trace_parser_loaded(parser) &&
4681            !trace_parser_cont(parser)) {
4682                ret = ftrace_process_regex(iter, parser->buffer,
4683                                           parser->idx, enable);
4684                trace_parser_clear(parser);
4685                if (ret < 0)
4686                        goto out;
4687        }
4688
4689        ret = read;
4690 out:
4691        return ret;
4692}
4693
4694ssize_t
4695ftrace_filter_write(struct file *file, const char __user *ubuf,
4696                    size_t cnt, loff_t *ppos)
4697{
4698        return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4699}
4700
4701ssize_t
4702ftrace_notrace_write(struct file *file, const char __user *ubuf,
4703                     size_t cnt, loff_t *ppos)
4704{
4705        return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4706}
4707
4708static int
4709ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4710{
4711        struct ftrace_func_entry *entry;
4712
4713        if (!ftrace_location(ip))
4714                return -EINVAL;
4715
4716        if (remove) {
4717                entry = ftrace_lookup_ip(hash, ip);
4718                if (!entry)
4719                        return -ENOENT;
4720                free_hash_entry(hash, entry);
4721                return 0;
4722        }
4723
4724        return add_hash_entry(hash, ip);
4725}
4726
4727static int
4728ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4729                unsigned long ip, int remove, int reset, int enable)
4730{
4731        struct ftrace_hash **orig_hash;
4732        struct ftrace_hash *hash;
4733        int ret;
4734
4735        if (unlikely(ftrace_disabled))
4736                return -ENODEV;
4737
4738        mutex_lock(&ops->func_hash->regex_lock);
4739
4740        if (enable)
4741                orig_hash = &ops->func_hash->filter_hash;
4742        else
4743                orig_hash = &ops->func_hash->notrace_hash;
4744
4745        if (reset)
4746                hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4747        else
4748                hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4749
4750        if (!hash) {
4751                ret = -ENOMEM;
4752                goto out_regex_unlock;
4753        }
4754
4755        if (buf && !ftrace_match_records(hash, buf, len)) {
4756                ret = -EINVAL;
4757                goto out_regex_unlock;
4758        }
4759        if (ip) {
4760                ret = ftrace_match_addr(hash, ip, remove);
4761                if (ret < 0)
4762                        goto out_regex_unlock;
4763        }
4764
4765        mutex_lock(&ftrace_lock);
4766        ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4767        mutex_unlock(&ftrace_lock);
4768
4769 out_regex_unlock:
4770        mutex_unlock(&ops->func_hash->regex_lock);
4771
4772        free_ftrace_hash(hash);
4773        return ret;
4774}
4775
4776static int
4777ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4778                int reset, int enable)
4779{
4780        return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4781}
4782
4783/**
4784 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4785 * @ops - the ops to set the filter with
4786 * @ip - the address to add to or remove from the filter.
4787 * @remove - non zero to remove the ip from the filter
4788 * @reset - non zero to reset all filters before applying this filter.
4789 *
4790 * Filters denote which functions should be enabled when tracing is enabled
4791 * If @ip is NULL, it failes to update filter.
4792 */
4793int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4794                         int remove, int reset)
4795{
4796        ftrace_ops_init(ops);
4797        return ftrace_set_addr(ops, ip, remove, reset, 1);
4798}
4799EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4800
4801/**
4802 * ftrace_ops_set_global_filter - setup ops to use global filters
4803 * @ops - the ops which will use the global filters
4804 *
4805 * ftrace users who need global function trace filtering should call this.
4806 * It can set the global filter only if ops were not initialized before.
4807 */
4808void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
4809{
4810        if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
4811                return;
4812
4813        ftrace_ops_init(ops);
4814        ops->func_hash = &global_ops.local_hash;
4815}
4816EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
4817
4818static int
4819ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4820                 int reset, int enable)
4821{
4822        return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4823}
4824
4825/**
4826 * ftrace_set_filter - set a function to filter on in ftrace
4827 * @ops - the ops to set the filter with
4828 * @buf - the string that holds the function filter text.
4829 * @len - the length of the string.
4830 * @reset - non zero to reset all filters before applying this filter.
4831 *
4832 * Filters denote which functions should be enabled when tracing is enabled.
4833 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4834 */
4835int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4836                       int len, int reset)
4837{
4838        ftrace_ops_init(ops);
4839        return ftrace_set_regex(ops, buf, len, reset, 1);
4840}
4841EXPORT_SYMBOL_GPL(ftrace_set_filter);
4842
4843/**
4844 * ftrace_set_notrace - set a function to not trace in ftrace
4845 * @ops - the ops to set the notrace filter with
4846 * @buf - the string that holds the function notrace text.
4847 * @len - the length of the string.
4848 * @reset - non zero to reset all filters before applying this filter.
4849 *
4850 * Notrace Filters denote which functions should not be enabled when tracing
4851 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4852 * for tracing.
4853 */
4854int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4855                        int len, int reset)
4856{
4857        ftrace_ops_init(ops);
4858        return ftrace_set_regex(ops, buf, len, reset, 0);
4859}
4860EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4861/**
4862 * ftrace_set_global_filter - set a function to filter on with global tracers
4863 * @buf - the string that holds the function filter text.
4864 * @len - the length of the string.
4865 * @reset - non zero to reset all filters before applying this filter.
4866 *
4867 * Filters denote which functions should be enabled when tracing is enabled.
4868 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4869 */
4870void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4871{
4872        ftrace_set_regex(&global_ops, buf, len, reset, 1);
4873}
4874EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4875
4876/**
4877 * ftrace_set_global_notrace - set a function to not trace with global tracers
4878 * @buf - the string that holds the function notrace text.
4879 * @len - the length of the string.
4880 * @reset - non zero to reset all filters before applying this filter.
4881 *
4882 * Notrace Filters denote which functions should not be enabled when tracing
4883 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4884 * for tracing.
4885 */
4886void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4887{
4888        ftrace_set_regex(&global_ops, buf, len, reset, 0);
4889}
4890EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4891
4892/*
4893 * command line interface to allow users to set filters on boot up.
4894 */
4895#define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
4896static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4897static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4898
4899/* Used by function selftest to not test if filter is set */
4900bool ftrace_filter_param __initdata;
4901
4902static int __init set_ftrace_notrace(char *str)
4903{
4904        ftrace_filter_param = true;
4905        strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4906        return 1;
4907}
4908__setup("ftrace_notrace=", set_ftrace_notrace);
4909
4910static int __init set_ftrace_filter(char *str)
4911{
4912        ftrace_filter_param = true;
4913        strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4914        return 1;
4915}
4916__setup("ftrace_filter=", set_ftrace_filter);
4917
4918#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4919static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4920static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4921static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
4922
4923static int __init set_graph_function(char *str)
4924{
4925        strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4926        return 1;
4927}
4928__setup("ftrace_graph_filter=", set_graph_function);
4929
4930static int __init set_graph_notrace_function(char *str)
4931{
4932        strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4933        return 1;
4934}
4935__setup("ftrace_graph_notrace=", set_graph_notrace_function);
4936
4937static int __init set_graph_max_depth_function(char *str)
4938{
4939        if (!str)
4940                return 0;
4941        fgraph_max_depth = simple_strtoul(str, NULL, 0);
4942        return 1;
4943}
4944__setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
4945
4946static void __init set_ftrace_early_graph(char *buf, int enable)
4947{
4948        int ret;
4949        char *func;
4950        struct ftrace_hash *hash;
4951
4952        hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4953        if (WARN_ON(!hash))
4954                return;
4955
4956        while (buf) {
4957                func = strsep(&buf, ",");
4958                /* we allow only one expression at a time */
4959                ret = ftrace_graph_set_hash(hash, func);
4960                if (ret)
4961                        printk(KERN_DEBUG "ftrace: function %s not "
4962                                          "traceable\n", func);
4963        }
4964
4965        if (enable)
4966                ftrace_graph_hash = hash;
4967        else
4968                ftrace_graph_notrace_hash = hash;
4969}
4970#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4971
4972void __init
4973ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
4974{
4975        char *func;
4976
4977        ftrace_ops_init(ops);
4978
4979        while (buf) {
4980                func = strsep(&buf, ",");
4981                ftrace_set_regex(ops, func, strlen(func), 0, enable);
4982        }
4983}
4984
4985static void __init set_ftrace_early_filters(void)
4986{
4987        if (ftrace_filter_buf[0])
4988                ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
4989        if (ftrace_notrace_buf[0])
4990                ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
4991#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4992        if (ftrace_graph_buf[0])
4993                set_ftrace_early_graph(ftrace_graph_buf, 1);
4994        if (ftrace_graph_notrace_buf[0])
4995                set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
4996#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4997}
4998
4999int ftrace_regex_release(struct inode *inode, struct file *file)
5000{
5001        struct seq_file *m = (struct seq_file *)file->private_data;
5002        struct ftrace_iterator *iter;
5003        struct ftrace_hash **orig_hash;
5004        struct trace_parser *parser;
5005        int filter_hash;
5006        int ret;
5007
5008        if (file->f_mode & FMODE_READ) {
5009                iter = m->private;
5010                seq_release(inode, file);
5011        } else
5012                iter = file->private_data;
5013
5014        parser = &iter->parser;
5015        if (trace_parser_loaded(parser)) {
5016                ftrace_match_records(iter->hash, parser->buffer, parser->idx);
5017        }
5018
5019        trace_parser_put(parser);
5020
5021        mutex_lock(&iter->ops->func_hash->regex_lock);
5022
5023        if (file->f_mode & FMODE_WRITE) {
5024                filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
5025
5026                if (filter_hash) {
5027                        orig_hash = &iter->ops->func_hash->filter_hash;
5028                        if (iter->tr && !list_empty(&iter->tr->mod_trace))
5029                                iter->hash->flags |= FTRACE_HASH_FL_MOD;
5030                } else
5031                        orig_hash = &iter->ops->func_hash->notrace_hash;
5032
5033                mutex_lock(&ftrace_lock);
5034                ret = ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
5035                                                      iter->hash, filter_hash);
5036                mutex_unlock(&ftrace_lock);
5037        } else {
5038                /* For read only, the hash is the ops hash */
5039                iter->hash = NULL;
5040        }
5041
5042        mutex_unlock(&iter->ops->func_hash->regex_lock);
5043        free_ftrace_hash(iter->hash);
5044        kfree(iter);
5045
5046        return 0;
5047}
5048
5049static const struct file_operations ftrace_avail_fops = {
5050        .open = ftrace_avail_open,
5051        .read = seq_read,
5052        .llseek = seq_lseek,
5053        .release = seq_release_private,
5054};
5055
5056static const struct file_operations ftrace_enabled_fops = {
5057        .open = ftrace_enabled_open,
5058        .read = seq_read,
5059        .llseek = seq_lseek,
5060        .release = seq_release_private,
5061};
5062
5063static const struct file_operations ftrace_filter_fops = {
5064        .open = ftrace_filter_open,
5065        .read = seq_read,
5066        .write = ftrace_filter_write,
5067        .llseek = tracing_lseek,
5068        .release = ftrace_regex_release,
5069};
5070
5071static const struct file_operations ftrace_notrace_fops = {
5072        .open = ftrace_notrace_open,
5073        .read = seq_read,
5074        .write = ftrace_notrace_write,
5075        .llseek = tracing_lseek,
5076        .release = ftrace_regex_release,
5077};
5078
5079#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5080
5081static DEFINE_MUTEX(graph_lock);
5082
5083struct ftrace_hash *ftrace_graph_hash = EMPTY_HASH;
5084struct ftrace_hash *ftrace_graph_notrace_hash = EMPTY_HASH;
5085
5086enum graph_filter_type {
5087        GRAPH_FILTER_NOTRACE    = 0,
5088        GRAPH_FILTER_FUNCTION,
5089};
5090
5091#define FTRACE_GRAPH_EMPTY      ((void *)1)
5092
5093struct ftrace_graph_data {
5094        struct ftrace_hash              *hash;
5095        struct ftrace_func_entry        *entry;
5096        int                             idx;   /* for hash table iteration */
5097        enum graph_filter_type          type;
5098        struct ftrace_hash              *new_hash;
5099        const struct seq_operations     *seq_ops;
5100        struct trace_parser             parser;
5101};
5102
5103static void *
5104__g_next(struct seq_file *m, loff_t *pos)
5105{
5106        struct ftrace_graph_data *fgd = m->private;
5107        struct ftrace_func_entry *entry = fgd->entry;
5108        struct hlist_head *head;
5109        int i, idx = fgd->idx;
5110
5111        if (*pos >= fgd->hash->count)
5112                return NULL;
5113
5114        if (entry) {
5115                hlist_for_each_entry_continue(entry, hlist) {
5116                        fgd->entry = entry;
5117                        return entry;
5118                }
5119
5120                idx++;
5121        }
5122
5123        for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
5124                head = &fgd->hash->buckets[i];
5125                hlist_for_each_entry(entry, head, hlist) {
5126                        fgd->entry = entry;
5127                        fgd->idx = i;
5128                        return entry;
5129                }
5130        }
5131        return NULL;
5132}
5133
5134static void *
5135g_next(struct seq_file *m, void *v, loff_t *pos)
5136{
5137        (*pos)++;
5138        return __g_next(m, pos);
5139}
5140
5141static void *g_start(struct seq_file *m, loff_t *pos)
5142{
5143        struct ftrace_graph_data *fgd = m->private;
5144
5145        mutex_lock(&graph_lock);
5146
5147        if (fgd->type == GRAPH_FILTER_FUNCTION)
5148                fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5149                                        lockdep_is_held(&graph_lock));
5150        else
5151                fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5152                                        lockdep_is_held(&graph_lock));
5153
5154        /* Nothing, tell g_show to print all functions are enabled */
5155        if (ftrace_hash_empty(fgd->hash) && !*pos)
5156                return FTRACE_GRAPH_EMPTY;
5157
5158        fgd->idx = 0;
5159        fgd->entry = NULL;
5160        return __g_next(m, pos);
5161}
5162
5163static void g_stop(struct seq_file *m, void *p)
5164{
5165        mutex_unlock(&graph_lock);
5166}
5167
5168static int g_show(struct seq_file *m, void *v)
5169{
5170        struct ftrace_func_entry *entry = v;
5171
5172        if (!entry)
5173                return 0;
5174
5175        if (entry == FTRACE_GRAPH_EMPTY) {
5176                struct ftrace_graph_data *fgd = m->private;
5177
5178                if (fgd->type == GRAPH_FILTER_FUNCTION)
5179                        seq_puts(m, "#### all functions enabled ####\n");
5180                else
5181                        seq_puts(m, "#### no functions disabled ####\n");
5182                return 0;
5183        }
5184
5185        seq_printf(m, "%ps\n", (void *)entry->ip);
5186
5187        return 0;
5188}
5189
5190static const struct seq_operations ftrace_graph_seq_ops = {
5191        .start = g_start,
5192        .next = g_next,
5193        .stop = g_stop,
5194        .show = g_show,
5195};
5196
5197static int
5198__ftrace_graph_open(struct inode *inode, struct file *file,
5199                    struct ftrace_graph_data *fgd)
5200{
5201        int ret = 0;
5202        struct ftrace_hash *new_hash = NULL;
5203
5204        if (file->f_mode & FMODE_WRITE) {
5205                const int size_bits = FTRACE_HASH_DEFAULT_BITS;
5206
5207                if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
5208                        return -ENOMEM;
5209
5210                if (file->f_flags & O_TRUNC)
5211                        new_hash = alloc_ftrace_hash(size_bits);
5212                else
5213                        new_hash = alloc_and_copy_ftrace_hash(size_bits,
5214                                                              fgd->hash);
5215                if (!new_hash) {
5216                        ret = -ENOMEM;
5217                        goto out;
5218                }
5219        }
5220
5221        if (file->f_mode & FMODE_READ) {
5222                ret = seq_open(file, &ftrace_graph_seq_ops);
5223                if (!ret) {
5224                        struct seq_file *m = file->private_data;
5225                        m->private = fgd;
5226                } else {
5227                        /* Failed */
5228                        free_ftrace_hash(new_hash);
5229                        new_hash = NULL;
5230                }
5231        } else
5232                file->private_data = fgd;
5233
5234out:
5235        if (ret < 0 && file->f_mode & FMODE_WRITE)
5236                trace_parser_put(&fgd->parser);
5237
5238        fgd->new_hash = new_hash;
5239
5240        /*
5241         * All uses of fgd->hash must be taken with the graph_lock
5242         * held. The graph_lock is going to be released, so force
5243         * fgd->hash to be reinitialized when it is taken again.
5244         */
5245        fgd->hash = NULL;
5246
5247        return ret;
5248}
5249
5250static int
5251ftrace_graph_open(struct inode *inode, struct file *file)
5252{
5253        struct ftrace_graph_data *fgd;
5254        int ret;
5255
5256        if (unlikely(ftrace_disabled))
5257                return -ENODEV;
5258
5259        fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5260        if (fgd == NULL)
5261                return -ENOMEM;
5262
5263        mutex_lock(&graph_lock);
5264
5265        fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5266                                        lockdep_is_held(&graph_lock));
5267        fgd->type = GRAPH_FILTER_FUNCTION;
5268        fgd->seq_ops = &ftrace_graph_seq_ops;
5269
5270        ret = __ftrace_graph_open(inode, file, fgd);
5271        if (ret < 0)
5272                kfree(fgd);
5273
5274        mutex_unlock(&graph_lock);
5275        return ret;
5276}
5277
5278static int
5279ftrace_graph_notrace_open(struct inode *inode, struct file *file)
5280{
5281        struct ftrace_graph_data *fgd;
5282        int ret;
5283
5284        if (unlikely(ftrace_disabled))
5285                return -ENODEV;
5286
5287        fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5288        if (fgd == NULL)
5289                return -ENOMEM;
5290
5291        mutex_lock(&graph_lock);
5292
5293        fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5294                                        lockdep_is_held(&graph_lock));
5295        fgd->type = GRAPH_FILTER_NOTRACE;
5296        fgd->seq_ops = &ftrace_graph_seq_ops;
5297
5298        ret = __ftrace_graph_open(inode, file, fgd);
5299        if (ret < 0)
5300                kfree(fgd);
5301
5302        mutex_unlock(&graph_lock);
5303        return ret;
5304}
5305
5306static int
5307ftrace_graph_release(struct inode *inode, struct file *file)
5308{
5309        struct ftrace_graph_data *fgd;
5310        struct ftrace_hash *old_hash, *new_hash;
5311        struct trace_parser *parser;
5312        int ret = 0;
5313
5314        if (file->f_mode & FMODE_READ) {
5315                struct seq_file *m = file->private_data;
5316
5317                fgd = m->private;
5318                seq_release(inode, file);
5319        } else {
5320                fgd = file->private_data;
5321        }
5322
5323
5324        if (file->f_mode & FMODE_WRITE) {
5325
5326                parser = &fgd->parser;
5327
5328                if (trace_parser_loaded((parser))) {
5329                        ret = ftrace_graph_set_hash(fgd->new_hash,
5330                                                    parser->buffer);
5331                }
5332
5333                trace_parser_put(parser);
5334
5335                new_hash = __ftrace_hash_move(fgd->new_hash);
5336                if (!new_hash) {
5337                        ret = -ENOMEM;
5338                        goto out;
5339                }
5340
5341                mutex_lock(&graph_lock);
5342
5343                if (fgd->type == GRAPH_FILTER_FUNCTION) {
5344                        old_hash = rcu_dereference_protected(ftrace_graph_hash,
5345                                        lockdep_is_held(&graph_lock));
5346                        rcu_assign_pointer(ftrace_graph_hash, new_hash);
5347                } else {
5348                        old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5349                                        lockdep_is_held(&graph_lock));
5350                        rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
5351                }
5352
5353                mutex_unlock(&graph_lock);
5354
5355                /* Wait till all users are no longer using the old hash */
5356                synchronize_sched();
5357
5358                free_ftrace_hash(old_hash);
5359        }
5360
5361 out:
5362        free_ftrace_hash(fgd->new_hash);
5363        kfree(fgd);
5364
5365        return ret;
5366}
5367
5368static int
5369ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
5370{
5371        struct ftrace_glob func_g;
5372        struct dyn_ftrace *rec;
5373        struct ftrace_page *pg;
5374        struct ftrace_func_entry *entry;
5375        int fail = 1;
5376        int not;
5377
5378        /* decode regex */
5379        func_g.type = filter_parse_regex(buffer, strlen(buffer),
5380                                         &func_g.search, &not);
5381
5382        func_g.len = strlen(func_g.search);
5383
5384        mutex_lock(&ftrace_lock);
5385
5386        if (unlikely(ftrace_disabled)) {
5387                mutex_unlock(&ftrace_lock);
5388                return -ENODEV;
5389        }
5390
5391        do_for_each_ftrace_rec(pg, rec) {
5392
5393                if (rec->flags & FTRACE_FL_DISABLED)
5394                        continue;
5395
5396                if (ftrace_match_record(rec, &func_g, NULL, 0)) {
5397                        entry = ftrace_lookup_ip(hash, rec->ip);
5398
5399                        if (!not) {
5400                                fail = 0;
5401
5402                                if (entry)
5403                                        continue;
5404                                if (add_hash_entry(hash, rec->ip) < 0)
5405                                        goto out;
5406                        } else {
5407                                if (entry) {
5408                                        free_hash_entry(hash, entry);
5409                                        fail = 0;
5410                                }
5411                        }
5412                }
5413        } while_for_each_ftrace_rec();
5414out:
5415        mutex_unlock(&ftrace_lock);
5416
5417        if (fail)
5418                return -EINVAL;
5419
5420        return 0;
5421}
5422
5423static ssize_t
5424ftrace_graph_write(struct file *file, const char __user *ubuf,
5425                   size_t cnt, loff_t *ppos)
5426{
5427        ssize_t read, ret = 0;
5428        struct ftrace_graph_data *fgd = file->private_data;
5429        struct trace_parser *parser;
5430
5431        if (!cnt)
5432                return 0;
5433
5434        /* Read mode uses seq functions */
5435        if (file->f_mode & FMODE_READ) {
5436                struct seq_file *m = file->private_data;
5437                fgd = m->private;
5438        }
5439
5440        parser = &fgd->parser;
5441
5442        read = trace_get_user(parser, ubuf, cnt, ppos);
5443
5444        if (read >= 0 && trace_parser_loaded(parser) &&
5445            !trace_parser_cont(parser)) {
5446
5447                ret = ftrace_graph_set_hash(fgd->new_hash,
5448                                            parser->buffer);
5449                trace_parser_clear(parser);
5450        }
5451
5452        if (!ret)
5453                ret = read;
5454
5455        return ret;
5456}
5457
5458static const struct file_operations ftrace_graph_fops = {
5459        .open           = ftrace_graph_open,
5460        .read           = seq_read,
5461        .write          = ftrace_graph_write,
5462        .llseek         = tracing_lseek,
5463        .release        = ftrace_graph_release,
5464};
5465
5466static const struct file_operations ftrace_graph_notrace_fops = {
5467        .open           = ftrace_graph_notrace_open,
5468        .read           = seq_read,
5469        .write          = ftrace_graph_write,
5470        .llseek         = tracing_lseek,
5471        .release        = ftrace_graph_release,
5472};
5473#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5474
5475void ftrace_create_filter_files(struct ftrace_ops *ops,
5476                                struct dentry *parent)
5477{
5478
5479        trace_create_file("set_ftrace_filter", 0644, parent,
5480                          ops, &ftrace_filter_fops);
5481
5482        trace_create_file("set_ftrace_notrace", 0644, parent,
5483                          ops, &ftrace_notrace_fops);
5484}
5485
5486/*
5487 * The name "destroy_filter_files" is really a misnomer. Although
5488 * in the future, it may actualy delete the files, but this is
5489 * really intended to make sure the ops passed in are disabled
5490 * and that when this function returns, the caller is free to
5491 * free the ops.
5492 *
5493 * The "destroy" name is only to match the "create" name that this
5494 * should be paired with.
5495 */
5496void ftrace_destroy_filter_files(struct ftrace_ops *ops)
5497{
5498        mutex_lock(&ftrace_lock);
5499        if (ops->flags & FTRACE_OPS_FL_ENABLED)
5500                ftrace_shutdown(ops, 0);
5501        ops->flags |= FTRACE_OPS_FL_DELETED;
5502        mutex_unlock(&ftrace_lock);
5503}
5504
5505static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
5506{
5507
5508        trace_create_file("available_filter_functions", 0444,
5509                        d_tracer, NULL, &ftrace_avail_fops);
5510
5511        trace_create_file("enabled_functions", 0444,
5512                        d_tracer, NULL, &ftrace_enabled_fops);
5513
5514        ftrace_create_filter_files(&global_ops, d_tracer);
5515
5516#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5517        trace_create_file("set_graph_function", 0644, d_tracer,
5518                                    NULL,
5519                                    &ftrace_graph_fops);
5520        trace_create_file("set_graph_notrace", 0644, d_tracer,
5521                                    NULL,
5522                                    &ftrace_graph_notrace_fops);
5523#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5524
5525        return 0;
5526}
5527
5528static int ftrace_cmp_ips(const void *a, const void *b)
5529{
5530        const unsigned long *ipa = a;
5531        const unsigned long *ipb = b;
5532
5533        if (*ipa > *ipb)
5534                return 1;
5535        if (*ipa < *ipb)
5536                return -1;
5537        return 0;
5538}
5539
5540static int ftrace_process_locs(struct module *mod,
5541                               unsigned long *start,
5542                               unsigned long *end)
5543{
5544        struct ftrace_page *start_pg;
5545        struct ftrace_page *pg;
5546        struct dyn_ftrace *rec;
5547        unsigned long count;
5548        unsigned long *p;
5549        unsigned long addr;
5550        unsigned long flags = 0; /* Shut up gcc */
5551        int ret = -ENOMEM;
5552
5553        count = end - start;
5554
5555        if (!count)
5556                return 0;
5557
5558        sort(start, count, sizeof(*start),
5559             ftrace_cmp_ips, NULL);
5560
5561        start_pg = ftrace_allocate_pages(count);
5562        if (!start_pg)
5563                return -ENOMEM;
5564
5565        mutex_lock(&ftrace_lock);
5566
5567        /*
5568         * Core and each module needs their own pages, as
5569         * modules will free them when they are removed.
5570         * Force a new page to be allocated for modules.
5571         */
5572        if (!mod) {
5573                WARN_ON(ftrace_pages || ftrace_pages_start);
5574                /* First initialization */
5575                ftrace_pages = ftrace_pages_start = start_pg;
5576        } else {
5577                if (!ftrace_pages)
5578                        goto out;
5579
5580                if (WARN_ON(ftrace_pages->next)) {
5581                        /* Hmm, we have free pages? */
5582                        while (ftrace_pages->next)
5583                                ftrace_pages = ftrace_pages->next;
5584                }
5585
5586                ftrace_pages->next = start_pg;
5587        }
5588
5589        p = start;
5590        pg = start_pg;
5591        while (p < end) {
5592                addr = ftrace_call_adjust(*p++);
5593                /*
5594                 * Some architecture linkers will pad between
5595                 * the different mcount_loc sections of different
5596                 * object files to satisfy alignments.
5597                 * Skip any NULL pointers.
5598                 */
5599                if (!addr)
5600                        continue;
5601
5602                if (pg->index == pg->size) {
5603                        /* We should have allocated enough */
5604                        if (WARN_ON(!pg->next))
5605                                break;
5606                        pg = pg->next;
5607                }
5608
5609                rec = &pg->records[pg->index++];
5610                rec->ip = addr;
5611        }
5612
5613        /* We should have used all pages */
5614        WARN_ON(pg->next);
5615
5616        /* Assign the last page to ftrace_pages */
5617        ftrace_pages = pg;
5618
5619        /*
5620         * We only need to disable interrupts on start up
5621         * because we are modifying code that an interrupt
5622         * may execute, and the modification is not atomic.
5623         * But for modules, nothing runs the code we modify
5624         * until we are finished with it, and there's no
5625         * reason to cause large interrupt latencies while we do it.
5626         */
5627        if (!mod)
5628                local_irq_save(flags);
5629        ftrace_update_code(mod, start_pg);
5630        if (!mod)
5631                local_irq_restore(flags);
5632        ret = 0;
5633 out:
5634        mutex_unlock(&ftrace_lock);
5635
5636        return ret;
5637}
5638
5639struct ftrace_mod_func {
5640        struct list_head        list;
5641        char                    *name;
5642        unsigned long           ip;
5643        unsigned int            size;
5644};
5645
5646struct ftrace_mod_map {
5647        struct rcu_head         rcu;
5648        struct list_head        list;
5649        struct module           *mod;
5650        unsigned long           start_addr;
5651        unsigned long           end_addr;
5652        struct list_head        funcs;
5653        unsigned int            num_funcs;
5654};
5655
5656#ifdef CONFIG_MODULES
5657
5658#define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
5659
5660static LIST_HEAD(ftrace_mod_maps);
5661
5662static int referenced_filters(struct dyn_ftrace *rec)
5663{
5664        struct ftrace_ops *ops;
5665        int cnt = 0;
5666
5667        for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
5668                if (ops_references_rec(ops, rec))
5669                    cnt++;
5670        }
5671
5672        return cnt;
5673}
5674
5675static void
5676clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
5677{
5678        struct ftrace_func_entry *entry;
5679        struct dyn_ftrace *rec;
5680        int i;
5681
5682        if (ftrace_hash_empty(hash))
5683                return;
5684
5685        for (i = 0; i < pg->index; i++) {
5686                rec = &pg->records[i];
5687                entry = __ftrace_lookup_ip(hash, rec->ip);
5688                /*
5689                 * Do not allow this rec to match again.
5690                 * Yeah, it may waste some memory, but will be removed
5691                 * if/when the hash is modified again.
5692                 */
5693                if (entry)
5694                        entry->ip = 0;
5695        }
5696}
5697
5698/* Clear any records from hashs */
5699static void clear_mod_from_hashes(struct ftrace_page *pg)
5700{
5701        struct trace_array *tr;
5702
5703        mutex_lock(&trace_types_lock);
5704        list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5705                if (!tr->ops || !tr->ops->func_hash)
5706                        continue;
5707                mutex_lock(&tr->ops->func_hash->regex_lock);
5708                clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
5709                clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
5710                mutex_unlock(&tr->ops->func_hash->regex_lock);
5711        }
5712        mutex_unlock(&trace_types_lock);
5713}
5714
5715static void ftrace_free_mod_map(struct rcu_head *rcu)
5716{
5717        struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
5718        struct ftrace_mod_func *mod_func;
5719        struct ftrace_mod_func *n;
5720
5721        /* All the contents of mod_map are now not visible to readers */
5722        list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
5723                kfree(mod_func->name);
5724                list_del(&mod_func->list);
5725                kfree(mod_func);
5726        }
5727
5728        kfree(mod_map);
5729}
5730
5731void ftrace_release_mod(struct module *mod)
5732{
5733        struct ftrace_mod_map *mod_map;
5734        struct ftrace_mod_map *n;
5735        struct dyn_ftrace *rec;
5736        struct ftrace_page **last_pg;
5737        struct ftrace_page *tmp_page = NULL;
5738        struct ftrace_page *pg;
5739        int order;
5740
5741        mutex_lock(&ftrace_lock);
5742
5743        if (ftrace_disabled)
5744                goto out_unlock;
5745
5746        list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
5747                if (mod_map->mod == mod) {
5748                        list_del_rcu(&mod_map->list);
5749                        call_rcu_sched(&mod_map->rcu, ftrace_free_mod_map);
5750                        break;
5751                }
5752        }
5753
5754        /*
5755         * Each module has its own ftrace_pages, remove
5756         * them from the list.
5757         */
5758        last_pg = &ftrace_pages_start;
5759        for (pg = ftrace_pages_start; pg; pg = *last_pg) {
5760                rec = &pg->records[0];
5761                if (within_module_core(rec->ip, mod) ||
5762                    within_module_init(rec->ip, mod)) {
5763                        /*
5764                         * As core pages are first, the first
5765                         * page should never be a module page.
5766                         */
5767                        if (WARN_ON(pg == ftrace_pages_start))
5768                                goto out_unlock;
5769
5770                        /* Check if we are deleting the last page */
5771                        if (pg == ftrace_pages)
5772                                ftrace_pages = next_to_ftrace_page(last_pg);
5773
5774                        ftrace_update_tot_cnt -= pg->index;
5775                        *last_pg = pg->next;
5776
5777                        pg->next = tmp_page;
5778                        tmp_page = pg;
5779                } else
5780                        last_pg = &pg->next;
5781        }
5782 out_unlock:
5783        mutex_unlock(&ftrace_lock);
5784
5785        for (pg = tmp_page; pg; pg = tmp_page) {
5786
5787                /* Needs to be called outside of ftrace_lock */
5788                clear_mod_from_hashes(pg);
5789
5790                order = get_count_order(pg->size / ENTRIES_PER_PAGE);
5791                free_pages((unsigned long)pg->records, order);
5792                tmp_page = pg->next;
5793                kfree(pg);
5794        }
5795}
5796
5797void ftrace_module_enable(struct module *mod)
5798{
5799        struct dyn_ftrace *rec;
5800        struct ftrace_page *pg;
5801
5802        mutex_lock(&ftrace_lock);
5803
5804        if (ftrace_disabled)
5805                goto out_unlock;
5806
5807        /*
5808         * If the tracing is enabled, go ahead and enable the record.
5809         *
5810         * The reason not to enable the record immediatelly is the
5811         * inherent check of ftrace_make_nop/ftrace_make_call for
5812         * correct previous instructions.  Making first the NOP
5813         * conversion puts the module to the correct state, thus
5814         * passing the ftrace_make_call check.
5815         *
5816         * We also delay this to after the module code already set the
5817         * text to read-only, as we now need to set it back to read-write
5818         * so that we can modify the text.
5819         */
5820        if (ftrace_start_up)
5821                ftrace_arch_code_modify_prepare();
5822
5823        do_for_each_ftrace_rec(pg, rec) {
5824                int cnt;
5825                /*
5826                 * do_for_each_ftrace_rec() is a double loop.
5827                 * module text shares the pg. If a record is
5828                 * not part of this module, then skip this pg,
5829                 * which the "break" will do.
5830                 */
5831                if (!within_module_core(rec->ip, mod) &&
5832                    !within_module_init(rec->ip, mod))
5833                        break;
5834
5835                cnt = 0;
5836
5837                /*
5838                 * When adding a module, we need to check if tracers are
5839                 * currently enabled and if they are, and can trace this record,
5840                 * we need to enable the module functions as well as update the
5841                 * reference counts for those function records.
5842                 */
5843                if (ftrace_start_up)
5844                        cnt += referenced_filters(rec);
5845
5846                /* This clears FTRACE_FL_DISABLED */
5847                rec->flags = cnt;
5848
5849                if (ftrace_start_up && cnt) {
5850                        int failed = __ftrace_replace_code(rec, 1);
5851                        if (failed) {
5852                                ftrace_bug(failed, rec);
5853                                goto out_loop;
5854                        }
5855                }
5856
5857        } while_for_each_ftrace_rec();
5858
5859 out_loop:
5860        if (ftrace_start_up)
5861                ftrace_arch_code_modify_post_process();
5862
5863 out_unlock:
5864        mutex_unlock(&ftrace_lock);
5865
5866        process_cached_mods(mod->name);
5867}
5868
5869void ftrace_module_init(struct module *mod)
5870{
5871        if (ftrace_disabled || !mod->num_ftrace_callsites)
5872                return;
5873
5874        ftrace_process_locs(mod, mod->ftrace_callsites,
5875                            mod->ftrace_callsites + mod->num_ftrace_callsites);
5876}
5877
5878static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
5879                                struct dyn_ftrace *rec)
5880{
5881        struct ftrace_mod_func *mod_func;
5882        unsigned long symsize;
5883        unsigned long offset;
5884        char str[KSYM_SYMBOL_LEN];
5885        char *modname;
5886        const char *ret;
5887
5888        ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
5889        if (!ret)
5890                return;
5891
5892        mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
5893        if (!mod_func)
5894                return;
5895
5896        mod_func->name = kstrdup(str, GFP_KERNEL);
5897        if (!mod_func->name) {
5898                kfree(mod_func);
5899                return;
5900        }
5901
5902        mod_func->ip = rec->ip - offset;
5903        mod_func->size = symsize;
5904
5905        mod_map->num_funcs++;
5906
5907        list_add_rcu(&mod_func->list, &mod_map->funcs);
5908}
5909
5910static struct ftrace_mod_map *
5911allocate_ftrace_mod_map(struct module *mod,
5912                        unsigned long start, unsigned long end)
5913{
5914        struct ftrace_mod_map *mod_map;
5915
5916        mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
5917        if (!mod_map)
5918                return NULL;
5919
5920        mod_map->mod = mod;
5921        mod_map->start_addr = start;
5922        mod_map->end_addr = end;
5923        mod_map->num_funcs = 0;
5924
5925        INIT_LIST_HEAD_RCU(&mod_map->funcs);
5926
5927        list_add_rcu(&mod_map->list, &ftrace_mod_maps);
5928
5929        return mod_map;
5930}
5931
5932static const char *
5933ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
5934                           unsigned long addr, unsigned long *size,
5935                           unsigned long *off, char *sym)
5936{
5937        struct ftrace_mod_func *found_func =  NULL;
5938        struct ftrace_mod_func *mod_func;
5939
5940        list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
5941                if (addr >= mod_func->ip &&
5942                    addr < mod_func->ip + mod_func->size) {
5943                        found_func = mod_func;
5944                        break;
5945                }
5946        }
5947
5948        if (found_func) {
5949                if (size)
5950                        *size = found_func->size;
5951                if (off)
5952                        *off = addr - found_func->ip;
5953                if (sym)
5954                        strlcpy(sym, found_func->name, KSYM_NAME_LEN);
5955
5956                return found_func->name;
5957        }
5958
5959        return NULL;
5960}
5961
5962const char *
5963ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
5964                   unsigned long *off, char **modname, char *sym)
5965{
5966        struct ftrace_mod_map *mod_map;
5967        const char *ret = NULL;
5968
5969        /* mod_map is freed via call_rcu_sched() */
5970        preempt_disable();
5971        list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
5972                ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
5973                if (ret) {
5974                        if (modname)
5975                                *modname = mod_map->mod->name;
5976                        break;
5977                }
5978        }
5979        preempt_enable();
5980
5981        return ret;
5982}
5983
5984int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
5985                           char *type, char *name,
5986                           char *module_name, int *exported)
5987{
5988        struct ftrace_mod_map *mod_map;
5989        struct ftrace_mod_func *mod_func;
5990
5991        preempt_disable();
5992        list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
5993
5994                if (symnum >= mod_map->num_funcs) {
5995                        symnum -= mod_map->num_funcs;
5996                        continue;
5997                }
5998
5999                list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
6000                        if (symnum > 1) {
6001                                symnum--;
6002                                continue;
6003                        }
6004
6005                        *value = mod_func->ip;
6006                        *type = 'T';
6007                        strlcpy(name, mod_func->name, KSYM_NAME_LEN);
6008                        strlcpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
6009                        *exported = 1;
6010                        preempt_enable();
6011                        return 0;
6012                }
6013                WARN_ON(1);
6014                break;
6015        }
6016        preempt_enable();
6017        return -ERANGE;
6018}
6019
6020#else
6021static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
6022                                struct dyn_ftrace *rec) { }
6023static inline struct ftrace_mod_map *
6024allocate_ftrace_mod_map(struct module *mod,
6025                        unsigned long start, unsigned long end)
6026{
6027        return NULL;
6028}
6029#endif /* CONFIG_MODULES */
6030
6031struct ftrace_init_func {
6032        struct list_head list;
6033        unsigned long ip;
6034};
6035
6036/* Clear any init ips from hashes */
6037static void
6038clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
6039{
6040        struct ftrace_func_entry *entry;
6041
6042        if (ftrace_hash_empty(hash))
6043                return;
6044
6045        entry = __ftrace_lookup_ip(hash, func->ip);
6046
6047        /*
6048         * Do not allow this rec to match again.
6049         * Yeah, it may waste some memory, but will be removed
6050         * if/when the hash is modified again.
6051         */
6052        if (entry)
6053                entry->ip = 0;
6054}
6055
6056static void
6057clear_func_from_hashes(struct ftrace_init_func *func)
6058{
6059        struct trace_array *tr;
6060
6061        mutex_lock(&trace_types_lock);
6062        list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6063                if (!tr->ops || !tr->ops->func_hash)
6064                        continue;
6065                mutex_lock(&tr->ops->func_hash->regex_lock);
6066                clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
6067                clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
6068                mutex_unlock(&tr->ops->func_hash->regex_lock);
6069        }
6070        mutex_unlock(&trace_types_lock);
6071}
6072
6073static void add_to_clear_hash_list(struct list_head *clear_list,
6074                                   struct dyn_ftrace *rec)
6075{
6076        struct ftrace_init_func *func;
6077
6078        func = kmalloc(sizeof(*func), GFP_KERNEL);
6079        if (!func) {
6080                WARN_ONCE(1, "alloc failure, ftrace filter could be stale\n");
6081                return;
6082        }
6083
6084        func->ip = rec->ip;
6085        list_add(&func->list, clear_list);
6086}
6087
6088void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
6089{
6090        unsigned long start = (unsigned long)(start_ptr);
6091        unsigned long end = (unsigned long)(end_ptr);
6092        struct ftrace_page **last_pg = &ftrace_pages_start;
6093        struct ftrace_page *pg;
6094        struct dyn_ftrace *rec;
6095        struct dyn_ftrace key;
6096        struct ftrace_mod_map *mod_map = NULL;
6097        struct ftrace_init_func *func, *func_next;
6098        struct list_head clear_hash;
6099        int order;
6100
6101        INIT_LIST_HEAD(&clear_hash);
6102
6103        key.ip = start;
6104        key.flags = end;        /* overload flags, as it is unsigned long */
6105
6106        mutex_lock(&ftrace_lock);
6107
6108        /*
6109         * If we are freeing module init memory, then check if
6110         * any tracer is active. If so, we need to save a mapping of
6111         * the module functions being freed with the address.
6112         */
6113        if (mod && ftrace_ops_list != &ftrace_list_end)
6114                mod_map = allocate_ftrace_mod_map(mod, start, end);
6115
6116        for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
6117                if (end < pg->records[0].ip ||
6118                    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
6119                        continue;
6120 again:
6121                rec = bsearch(&key, pg->records, pg->index,
6122                              sizeof(struct dyn_ftrace),
6123                              ftrace_cmp_recs);
6124                if (!rec)
6125                        continue;
6126
6127                /* rec will be cleared from hashes after ftrace_lock unlock */
6128                add_to_clear_hash_list(&clear_hash, rec);
6129
6130                if (mod_map)
6131                        save_ftrace_mod_rec(mod_map, rec);
6132
6133                pg->index--;
6134                ftrace_update_tot_cnt--;
6135                if (!pg->index) {
6136                        *last_pg = pg->next;
6137                        order = get_count_order(pg->size / ENTRIES_PER_PAGE);
6138                        free_pages((unsigned long)pg->records, order);
6139                        kfree(pg);
6140                        pg = container_of(last_pg, struct ftrace_page, next);
6141                        if (!(*last_pg))
6142                                ftrace_pages = pg;
6143                        continue;
6144                }
6145                memmove(rec, rec + 1,
6146                        (pg->index - (rec - pg->records)) * sizeof(*rec));
6147                /* More than one function may be in this block */
6148                goto again;
6149        }
6150        mutex_unlock(&ftrace_lock);
6151
6152        list_for_each_entry_safe(func, func_next, &clear_hash, list) {
6153                clear_func_from_hashes(func);
6154                kfree(func);
6155        }
6156}
6157
6158void __init ftrace_free_init_mem(void)
6159{
6160        void *start = (void *)(&__init_begin);
6161        void *end = (void *)(&__init_end);
6162
6163        ftrace_free_mem(NULL, start, end);
6164}
6165
6166void __init ftrace_init(void)
6167{
6168        extern unsigned long __start_mcount_loc[];
6169        extern unsigned long __stop_mcount_loc[];
6170        unsigned long count, flags;
6171        int ret;
6172
6173        local_irq_save(flags);
6174        ret = ftrace_dyn_arch_init();
6175        local_irq_restore(flags);
6176        if (ret)
6177                goto failed;
6178
6179        count = __stop_mcount_loc - __start_mcount_loc;
6180        if (!count) {
6181                pr_info("ftrace: No functions to be traced?\n");
6182                goto failed;
6183        }
6184
6185        pr_info("ftrace: allocating %ld entries in %ld pages\n",
6186                count, count / ENTRIES_PER_PAGE + 1);
6187
6188        last_ftrace_enabled = ftrace_enabled = 1;
6189
6190        ret = ftrace_process_locs(NULL,
6191                                  __start_mcount_loc,
6192                                  __stop_mcount_loc);
6193
6194        set_ftrace_early_filters();
6195
6196        return;
6197 failed:
6198        ftrace_disabled = 1;
6199}
6200
6201/* Do nothing if arch does not support this */
6202void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
6203{
6204}
6205
6206static void ftrace_update_trampoline(struct ftrace_ops *ops)
6207{
6208        arch_ftrace_update_trampoline(ops);
6209}
6210
6211void ftrace_init_trace_array(struct trace_array *tr)
6212{
6213        INIT_LIST_HEAD(&tr->func_probes);
6214        INIT_LIST_HEAD(&tr->mod_trace);
6215        INIT_LIST_HEAD(&tr->mod_notrace);
6216}
6217#else
6218
6219static struct ftrace_ops global_ops = {
6220        .func                   = ftrace_stub,
6221        .flags                  = FTRACE_OPS_FL_RECURSION_SAFE |
6222                                  FTRACE_OPS_FL_INITIALIZED |
6223                                  FTRACE_OPS_FL_PID,
6224};
6225
6226static int __init ftrace_nodyn_init(void)
6227{
6228        ftrace_enabled = 1;
6229        return 0;
6230}
6231core_initcall(ftrace_nodyn_init);
6232
6233static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
6234static inline void ftrace_startup_enable(int command) { }
6235static inline void ftrace_startup_all(int command) { }
6236/* Keep as macros so we do not need to define the commands */
6237# define ftrace_startup(ops, command)                                   \
6238        ({                                                              \
6239                int ___ret = __register_ftrace_function(ops);           \
6240                if (!___ret)                                            \
6241                        (ops)->flags |= FTRACE_OPS_FL_ENABLED;          \
6242                ___ret;                                                 \
6243        })
6244# define ftrace_shutdown(ops, command)                                  \
6245        ({                                                              \
6246                int ___ret = __unregister_ftrace_function(ops);         \
6247                if (!___ret)                                            \
6248                        (ops)->flags &= ~FTRACE_OPS_FL_ENABLED;         \
6249                ___ret;                                                 \
6250        })
6251
6252# define ftrace_startup_sysctl()        do { } while (0)
6253# define ftrace_shutdown_sysctl()       do { } while (0)
6254
6255static inline int
6256ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
6257{
6258        return 1;
6259}
6260
6261static void ftrace_update_trampoline(struct ftrace_ops *ops)
6262{
6263}
6264
6265#endif /* CONFIG_DYNAMIC_FTRACE */
6266
6267__init void ftrace_init_global_array_ops(struct trace_array *tr)
6268{
6269        tr->ops = &global_ops;
6270        tr->ops->private = tr;
6271        ftrace_init_trace_array(tr);
6272}
6273
6274void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
6275{
6276        /* If we filter on pids, update to use the pid function */
6277        if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
6278                if (WARN_ON(tr->ops->func != ftrace_stub))
6279                        printk("ftrace ops had %pS for function\n",
6280                               tr->ops->func);
6281        }
6282        tr->ops->func = func;
6283        tr->ops->private = tr;
6284}
6285
6286void ftrace_reset_array_ops(struct trace_array *tr)
6287{
6288        tr->ops->func = ftrace_stub;
6289}
6290
6291static inline void
6292__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6293                       struct ftrace_ops *ignored, struct pt_regs *regs)
6294{
6295        struct ftrace_ops *op;
6296        int bit;
6297
6298        bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6299        if (bit < 0)
6300                return;
6301
6302        /*
6303         * Some of the ops may be dynamically allocated,
6304         * they must be freed after a synchronize_sched().
6305         */
6306        preempt_disable_notrace();
6307
6308        do_for_each_ftrace_op(op, ftrace_ops_list) {
6309                /*
6310                 * Check the following for each ops before calling their func:
6311                 *  if RCU flag is set, then rcu_is_watching() must be true
6312                 *  if PER_CPU is set, then ftrace_function_local_disable()
6313                 *                          must be false
6314                 *  Otherwise test if the ip matches the ops filter
6315                 *
6316                 * If any of the above fails then the op->func() is not executed.
6317                 */
6318                if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
6319                    ftrace_ops_test(op, ip, regs)) {
6320                        if (FTRACE_WARN_ON(!op->func)) {
6321                                pr_warn("op=%p %pS\n", op, op);
6322                                goto out;
6323                        }
6324                        op->func(ip, parent_ip, op, regs);
6325                }
6326        } while_for_each_ftrace_op(op);
6327out:
6328        preempt_enable_notrace();
6329        trace_clear_recursion(bit);
6330}
6331
6332/*
6333 * Some archs only support passing ip and parent_ip. Even though
6334 * the list function ignores the op parameter, we do not want any
6335 * C side effects, where a function is called without the caller
6336 * sending a third parameter.
6337 * Archs are to support both the regs and ftrace_ops at the same time.
6338 * If they support ftrace_ops, it is assumed they support regs.
6339 * If call backs want to use regs, they must either check for regs
6340 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
6341 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
6342 * An architecture can pass partial regs with ftrace_ops and still
6343 * set the ARCH_SUPPORTS_FTRACE_OPS.
6344 */
6345#if ARCH_SUPPORTS_FTRACE_OPS
6346static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6347                                 struct ftrace_ops *op, struct pt_regs *regs)
6348{
6349        __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
6350}
6351#else
6352static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
6353{
6354        __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
6355}
6356#endif
6357
6358/*
6359 * If there's only one function registered but it does not support
6360 * recursion, needs RCU protection and/or requires per cpu handling, then
6361 * this function will be called by the mcount trampoline.
6362 */
6363static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
6364                                   struct ftrace_ops *op, struct pt_regs *regs)
6365{
6366        int bit;
6367
6368        if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
6369                return;
6370
6371        bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
6372        if (bit < 0)
6373                return;
6374
6375        preempt_disable_notrace();
6376
6377        op->func(ip, parent_ip, op, regs);
6378
6379        preempt_enable_notrace();
6380        trace_clear_recursion(bit);
6381}
6382
6383/**
6384 * ftrace_ops_get_func - get the function a trampoline should call
6385 * @ops: the ops to get the function for
6386 *
6387 * Normally the mcount trampoline will call the ops->func, but there
6388 * are times that it should not. For example, if the ops does not
6389 * have its own recursion protection, then it should call the
6390 * ftrace_ops_assist_func() instead.
6391 *
6392 * Returns the function that the trampoline should call for @ops.
6393 */
6394ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
6395{
6396        /*
6397         * If the function does not handle recursion, needs to be RCU safe,
6398         * or does per cpu logic, then we need to call the assist handler.
6399         */
6400        if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
6401            ops->flags & FTRACE_OPS_FL_RCU)
6402                return ftrace_ops_assist_func;
6403
6404        return ops->func;
6405}
6406
6407static void
6408ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
6409                    struct task_struct *prev, struct task_struct *next)
6410{
6411        struct trace_array *tr = data;
6412        struct trace_pid_list *pid_list;
6413
6414        pid_list = rcu_dereference_sched(tr->function_pids);
6415
6416        this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6417                       trace_ignore_this_task(pid_list, next));
6418}
6419
6420static void
6421ftrace_pid_follow_sched_process_fork(void *data,
6422                                     struct task_struct *self,
6423                                     struct task_struct *task)
6424{
6425        struct trace_pid_list *pid_list;
6426        struct trace_array *tr = data;
6427
6428        pid_list = rcu_dereference_sched(tr->function_pids);
6429        trace_filter_add_remove_task(pid_list, self, task);
6430}
6431
6432static void
6433ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
6434{
6435        struct trace_pid_list *pid_list;
6436        struct trace_array *tr = data;
6437
6438        pid_list = rcu_dereference_sched(tr->function_pids);
6439        trace_filter_add_remove_task(pid_list, NULL, task);
6440}
6441
6442void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
6443{
6444        if (enable) {
6445                register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6446                                                  tr);
6447                register_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6448                                                  tr);
6449        } else {
6450                unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
6451                                                    tr);
6452                unregister_trace_sched_process_exit(ftrace_pid_follow_sched_process_exit,
6453                                                    tr);
6454        }
6455}
6456
6457static void clear_ftrace_pids(struct trace_array *tr)
6458{
6459        struct trace_pid_list *pid_list;
6460        int cpu;
6461
6462        pid_list = rcu_dereference_protected(tr->function_pids,
6463                                             lockdep_is_held(&ftrace_lock));
6464        if (!pid_list)
6465                return;
6466
6467        unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6468
6469        for_each_possible_cpu(cpu)
6470                per_cpu_ptr(tr->trace_buffer.data, cpu)->ftrace_ignore_pid = false;
6471
6472        rcu_assign_pointer(tr->function_pids, NULL);
6473
6474        /* Wait till all users are no longer using pid filtering */
6475        synchronize_sched();
6476
6477        trace_free_pid_list(pid_list);
6478}
6479
6480void ftrace_clear_pids(struct trace_array *tr)
6481{
6482        mutex_lock(&ftrace_lock);
6483
6484        clear_ftrace_pids(tr);
6485
6486        mutex_unlock(&ftrace_lock);
6487}
6488
6489static void ftrace_pid_reset(struct trace_array *tr)
6490{
6491        mutex_lock(&ftrace_lock);
6492        clear_ftrace_pids(tr);
6493
6494        ftrace_update_pid_func();
6495        ftrace_startup_all(0);
6496
6497        mutex_unlock(&ftrace_lock);
6498}
6499
6500/* Greater than any max PID */
6501#define FTRACE_NO_PIDS          (void *)(PID_MAX_LIMIT + 1)
6502
6503static void *fpid_start(struct seq_file *m, loff_t *pos)
6504        __acquires(RCU)
6505{
6506        struct trace_pid_list *pid_list;
6507        struct trace_array *tr = m->private;
6508
6509        mutex_lock(&ftrace_lock);
6510        rcu_read_lock_sched();
6511
6512        pid_list = rcu_dereference_sched(tr->function_pids);
6513
6514        if (!pid_list)
6515                return !(*pos) ? FTRACE_NO_PIDS : NULL;
6516
6517        return trace_pid_start(pid_list, pos);
6518}
6519
6520static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
6521{
6522        struct trace_array *tr = m->private;
6523        struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
6524
6525        if (v == FTRACE_NO_PIDS)
6526                return NULL;
6527
6528        return trace_pid_next(pid_list, v, pos);
6529}
6530
6531static void fpid_stop(struct seq_file *m, void *p)
6532        __releases(RCU)
6533{
6534        rcu_read_unlock_sched();
6535        mutex_unlock(&ftrace_lock);
6536}
6537
6538static int fpid_show(struct seq_file *m, void *v)
6539{
6540        if (v == FTRACE_NO_PIDS) {
6541                seq_puts(m, "no pid\n");
6542                return 0;
6543        }
6544
6545        return trace_pid_show(m, v);
6546}
6547
6548static const struct seq_operations ftrace_pid_sops = {
6549        .start = fpid_start,
6550        .next = fpid_next,
6551        .stop = fpid_stop,
6552        .show = fpid_show,
6553};
6554
6555static int
6556ftrace_pid_open(struct inode *inode, struct file *file)
6557{
6558        struct trace_array *tr = inode->i_private;
6559        struct seq_file *m;
6560        int ret = 0;
6561
6562        if (trace_array_get(tr) < 0)
6563                return -ENODEV;
6564
6565        if ((file->f_mode & FMODE_WRITE) &&
6566            (file->f_flags & O_TRUNC))
6567                ftrace_pid_reset(tr);
6568
6569        ret = seq_open(file, &ftrace_pid_sops);
6570        if (ret < 0) {
6571                trace_array_put(tr);
6572        } else {
6573                m = file->private_data;
6574                /* copy tr over to seq ops */
6575                m->private = tr;
6576        }
6577
6578        return ret;
6579}
6580
6581static void ignore_task_cpu(void *data)
6582{
6583        struct trace_array *tr = data;
6584        struct trace_pid_list *pid_list;
6585
6586        /*
6587         * This function is called by on_each_cpu() while the
6588         * event_mutex is held.
6589         */
6590        pid_list = rcu_dereference_protected(tr->function_pids,
6591                                             mutex_is_locked(&ftrace_lock));
6592
6593        this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
6594                       trace_ignore_this_task(pid_list, current));
6595}
6596
6597static ssize_t
6598ftrace_pid_write(struct file *filp, const char __user *ubuf,
6599                   size_t cnt, loff_t *ppos)
6600{
6601        struct seq_file *m = filp->private_data;
6602        struct trace_array *tr = m->private;
6603        struct trace_pid_list *filtered_pids = NULL;
6604        struct trace_pid_list *pid_list;
6605        ssize_t ret;
6606
6607        if (!cnt)
6608                return 0;
6609
6610        mutex_lock(&ftrace_lock);
6611
6612        filtered_pids = rcu_dereference_protected(tr->function_pids,
6613                                             lockdep_is_held(&ftrace_lock));
6614
6615        ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
6616        if (ret < 0)
6617                goto out;
6618
6619        rcu_assign_pointer(tr->function_pids, pid_list);
6620
6621        if (filtered_pids) {
6622                synchronize_sched();
6623                trace_free_pid_list(filtered_pids);
6624        } else if (pid_list) {
6625                /* Register a probe to set whether to ignore the tracing of a task */
6626                register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
6627        }
6628
6629        /*
6630         * Ignoring of pids is done at task switch. But we have to
6631         * check for those tasks that are currently running.
6632         * Always do this in case a pid was appended or removed.
6633         */
6634        on_each_cpu(ignore_task_cpu, tr, 1);
6635
6636        ftrace_update_pid_func();
6637        ftrace_startup_all(0);
6638 out:
6639        mutex_unlock(&ftrace_lock);
6640
6641        if (ret > 0)
6642                *ppos += ret;
6643
6644        return ret;
6645}
6646
6647static int
6648ftrace_pid_release(struct inode *inode, struct file *file)
6649{
6650        struct trace_array *tr = inode->i_private;
6651
6652        trace_array_put(tr);
6653
6654        return seq_release(inode, file);
6655}
6656
6657static const struct file_operations ftrace_pid_fops = {
6658        .open           = ftrace_pid_open,
6659        .write          = ftrace_pid_write,
6660        .read           = seq_read,
6661        .llseek         = tracing_lseek,
6662        .release        = ftrace_pid_release,
6663};
6664
6665void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
6666{
6667        trace_create_file("set_ftrace_pid", 0644, d_tracer,
6668                            tr, &ftrace_pid_fops);
6669}
6670
6671void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
6672                                         struct dentry *d_tracer)
6673{
6674        /* Only the top level directory has the dyn_tracefs and profile */
6675        WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
6676
6677        ftrace_init_dyn_tracefs(d_tracer);
6678        ftrace_profile_tracefs(d_tracer);
6679}
6680
6681/**
6682 * ftrace_kill - kill ftrace
6683 *
6684 * This function should be used by panic code. It stops ftrace
6685 * but in a not so nice way. If you need to simply kill ftrace
6686 * from a non-atomic section, use ftrace_kill.
6687 */
6688void ftrace_kill(void)
6689{
6690        ftrace_disabled = 1;
6691        ftrace_enabled = 0;
6692        clear_ftrace_function();
6693}
6694
6695/**
6696 * Test if ftrace is dead or not.
6697 */
6698int ftrace_is_dead(void)
6699{
6700        return ftrace_disabled;
6701}
6702
6703/**
6704 * register_ftrace_function - register a function for profiling
6705 * @ops - ops structure that holds the function for profiling.
6706 *
6707 * Register a function to be called by all functions in the
6708 * kernel.
6709 *
6710 * Note: @ops->func and all the functions it calls must be labeled
6711 *       with "notrace", otherwise it will go into a
6712 *       recursive loop.
6713 */
6714int register_ftrace_function(struct ftrace_ops *ops)
6715{
6716        int ret = -1;
6717
6718        ftrace_ops_init(ops);
6719
6720        mutex_lock(&ftrace_lock);
6721
6722        ret = ftrace_startup(ops, 0);
6723
6724        mutex_unlock(&ftrace_lock);
6725
6726        return ret;
6727}
6728EXPORT_SYMBOL_GPL(register_ftrace_function);
6729
6730/**
6731 * unregister_ftrace_function - unregister a function for profiling.
6732 * @ops - ops structure that holds the function to unregister
6733 *
6734 * Unregister a function that was added to be called by ftrace profiling.
6735 */
6736int unregister_ftrace_function(struct ftrace_ops *ops)
6737{
6738        int ret;
6739
6740        mutex_lock(&ftrace_lock);
6741        ret = ftrace_shutdown(ops, 0);
6742        mutex_unlock(&ftrace_lock);
6743
6744        return ret;
6745}
6746EXPORT_SYMBOL_GPL(unregister_ftrace_function);
6747
6748int
6749ftrace_enable_sysctl(struct ctl_table *table, int write,
6750                     void __user *buffer, size_t *lenp,
6751                     loff_t *ppos)
6752{
6753        int ret = -ENODEV;
6754
6755        mutex_lock(&ftrace_lock);
6756
6757        if (unlikely(ftrace_disabled))
6758                goto out;
6759
6760        ret = proc_dointvec(table, write, buffer, lenp, ppos);
6761
6762        if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
6763                goto out;
6764
6765        last_ftrace_enabled = !!ftrace_enabled;
6766
6767        if (ftrace_enabled) {
6768
6769                /* we are starting ftrace again */
6770                if (rcu_dereference_protected(ftrace_ops_list,
6771                        lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
6772                        update_ftrace_function();
6773
6774                ftrace_startup_sysctl();
6775
6776        } else {
6777                /* stopping ftrace calls (just send to ftrace_stub) */
6778                ftrace_trace_function = ftrace_stub;
6779
6780                ftrace_shutdown_sysctl();
6781        }
6782
6783 out:
6784        mutex_unlock(&ftrace_lock);
6785        return ret;
6786}
6787
6788#ifdef CONFIG_FUNCTION_GRAPH_TRACER
6789
6790static struct ftrace_ops graph_ops = {
6791        .func                   = ftrace_stub,
6792        .flags                  = FTRACE_OPS_FL_RECURSION_SAFE |
6793                                   FTRACE_OPS_FL_INITIALIZED |
6794                                   FTRACE_OPS_FL_PID |
6795                                   FTRACE_OPS_FL_STUB,
6796#ifdef FTRACE_GRAPH_TRAMP_ADDR
6797        .trampoline             = FTRACE_GRAPH_TRAMP_ADDR,
6798        /* trampoline_size is only needed for dynamically allocated tramps */
6799#endif
6800        ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
6801};
6802
6803void ftrace_graph_sleep_time_control(bool enable)
6804{
6805        fgraph_sleep_time = enable;
6806}
6807
6808void ftrace_graph_graph_time_control(bool enable)
6809{
6810        fgraph_graph_time = enable;
6811}
6812
6813int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
6814{
6815        return 0;
6816}
6817
6818/* The callbacks that hook a function */
6819trace_func_graph_ret_t ftrace_graph_return =
6820                        (trace_func_graph_ret_t)ftrace_stub;
6821trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
6822static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
6823
6824/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
6825static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
6826{
6827        int i;
6828        int ret = 0;
6829        int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
6830        struct task_struct *g, *t;
6831
6832        for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
6833                ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
6834                                        * sizeof(struct ftrace_ret_stack),
6835                                        GFP_KERNEL);
6836                if (!ret_stack_list[i]) {
6837                        start = 0;
6838                        end = i;
6839                        ret = -ENOMEM;
6840                        goto free;
6841                }
6842        }
6843
6844        read_lock(&tasklist_lock);
6845        do_each_thread(g, t) {
6846                if (start == end) {
6847                        ret = -EAGAIN;
6848                        goto unlock;
6849                }
6850
6851                if (t->ret_stack == NULL) {
6852                        atomic_set(&t->tracing_graph_pause, 0);
6853                        atomic_set(&t->trace_overrun, 0);
6854                        t->curr_ret_stack = -1;
6855                        /* Make sure the tasks see the -1 first: */
6856                        smp_wmb();
6857                        t->ret_stack = ret_stack_list[start++];
6858                }
6859        } while_each_thread(g, t);
6860
6861unlock:
6862        read_unlock(&tasklist_lock);
6863free:
6864        for (i = start; i < end; i++)
6865                kfree(ret_stack_list[i]);
6866        return ret;
6867}
6868
6869static void
6870ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
6871                        struct task_struct *prev, struct task_struct *next)
6872{
6873        unsigned long long timestamp;
6874        int index;
6875
6876        /*
6877         * Does the user want to count the time a function was asleep.
6878         * If so, do not update the time stamps.
6879         */
6880        if (fgraph_sleep_time)
6881                return;
6882
6883        timestamp = trace_clock_local();
6884
6885        prev->ftrace_timestamp = timestamp;
6886
6887        /* only process tasks that we timestamped */
6888        if (!next->ftrace_timestamp)
6889                return;
6890
6891        /*
6892         * Update all the counters in next to make up for the
6893         * time next was sleeping.
6894         */
6895        timestamp -= next->ftrace_timestamp;
6896
6897        for (index = next->curr_ret_stack; index >= 0; index--)
6898                next->ret_stack[index].calltime += timestamp;
6899}
6900
6901/* Allocate a return stack for each task */
6902static int start_graph_tracing(void)
6903{
6904        struct ftrace_ret_stack **ret_stack_list;
6905        int ret, cpu;
6906
6907        ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
6908                                sizeof(struct ftrace_ret_stack *),
6909                                GFP_KERNEL);
6910
6911        if (!ret_stack_list)
6912                return -ENOMEM;
6913
6914        /* The cpu_boot init_task->ret_stack will never be freed */
6915        for_each_online_cpu(cpu) {
6916                if (!idle_task(cpu)->ret_stack)
6917                        ftrace_graph_init_idle_task(idle_task(cpu), cpu);
6918        }
6919
6920        do {
6921                ret = alloc_retstack_tasklist(ret_stack_list);
6922        } while (ret == -EAGAIN);
6923
6924        if (!ret) {
6925                ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
6926                if (ret)
6927                        pr_info("ftrace_graph: Couldn't activate tracepoint"
6928                                " probe to kernel_sched_switch\n");
6929        }
6930
6931        kfree(ret_stack_list);
6932        return ret;
6933}
6934
6935/*
6936 * Hibernation protection.
6937 * The state of the current task is too much unstable during
6938 * suspend/restore to disk. We want to protect against that.
6939 */
6940static int
6941ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
6942                                                        void *unused)
6943{
6944        switch (state) {
6945        case PM_HIBERNATION_PREPARE:
6946                pause_graph_tracing();
6947                break;
6948
6949        case PM_POST_HIBERNATION:
6950                unpause_graph_tracing();
6951                break;
6952        }
6953        return NOTIFY_DONE;
6954}
6955
6956static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
6957{
6958        if (!ftrace_ops_test(&global_ops, trace->func, NULL))
6959                return 0;
6960        return __ftrace_graph_entry(trace);
6961}
6962
6963/*
6964 * The function graph tracer should only trace the functions defined
6965 * by set_ftrace_filter and set_ftrace_notrace. If another function
6966 * tracer ops is registered, the graph tracer requires testing the
6967 * function against the global ops, and not just trace any function
6968 * that any ftrace_ops registered.
6969 */
6970static void update_function_graph_func(void)
6971{
6972        struct ftrace_ops *op;
6973        bool do_test = false;
6974
6975        /*
6976         * The graph and global ops share the same set of functions
6977         * to test. If any other ops is on the list, then
6978         * the graph tracing needs to test if its the function
6979         * it should call.
6980         */
6981        do_for_each_ftrace_op(op, ftrace_ops_list) {
6982                if (op != &global_ops && op != &graph_ops &&
6983                    op != &ftrace_list_end) {
6984                        do_test = true;
6985                        /* in double loop, break out with goto */
6986                        goto out;
6987                }
6988        } while_for_each_ftrace_op(op);
6989 out:
6990        if (do_test)
6991                ftrace_graph_entry = ftrace_graph_entry_test;
6992        else
6993                ftrace_graph_entry = __ftrace_graph_entry;
6994}
6995
6996static struct notifier_block ftrace_suspend_notifier = {
6997        .notifier_call = ftrace_suspend_notifier_call,
6998};
6999
7000int register_ftrace_graph(trace_func_graph_ret_t retfunc,
7001                        trace_func_graph_ent_t entryfunc)
7002{
7003        int ret = 0;
7004
7005        mutex_lock(&ftrace_lock);
7006
7007        /* we currently allow only one tracer registered at a time */
7008        if (ftrace_graph_active) {
7009                ret = -EBUSY;
7010                goto out;
7011        }
7012
7013        register_pm_notifier(&ftrace_suspend_notifier);
7014
7015        ftrace_graph_active++;
7016        ret = start_graph_tracing();
7017        if (ret) {
7018                ftrace_graph_active--;
7019                goto out;
7020        }
7021
7022        ftrace_graph_return = retfunc;
7023
7024        /*
7025         * Update the indirect function to the entryfunc, and the
7026         * function that gets called to the entry_test first. Then
7027         * call the update fgraph entry function to determine if
7028         * the entryfunc should be called directly or not.
7029         */
7030        __ftrace_graph_entry = entryfunc;
7031        ftrace_graph_entry = ftrace_graph_entry_test;
7032        update_function_graph_func();
7033
7034        ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
7035out:
7036        mutex_unlock(&ftrace_lock);
7037        return ret;
7038}
7039
7040void unregister_ftrace_graph(void)
7041{
7042        mutex_lock(&ftrace_lock);
7043
7044        if (unlikely(!ftrace_graph_active))
7045                goto out;
7046
7047        ftrace_graph_active--;
7048        ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
7049        ftrace_graph_entry = ftrace_graph_entry_stub;
7050        __ftrace_graph_entry = ftrace_graph_entry_stub;
7051        ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
7052        unregister_pm_notifier(&ftrace_suspend_notifier);
7053        unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
7054
7055 out:
7056        mutex_unlock(&ftrace_lock);
7057}
7058
7059static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
7060
7061static void
7062graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
7063{
7064        atomic_set(&t->tracing_graph_pause, 0);
7065        atomic_set(&t->trace_overrun, 0);
7066        t->ftrace_timestamp = 0;
7067        /* make curr_ret_stack visible before we add the ret_stack */
7068        smp_wmb();
7069        t->ret_stack = ret_stack;
7070}
7071
7072/*
7073 * Allocate a return stack for the idle task. May be the first
7074 * time through, or it may be done by CPU hotplug online.
7075 */
7076void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
7077{
7078        t->curr_ret_stack = -1;
7079        /*
7080         * The idle task has no parent, it either has its own
7081         * stack or no stack at all.
7082         */
7083        if (t->ret_stack)
7084                WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
7085
7086        if (ftrace_graph_active) {
7087                struct ftrace_ret_stack *ret_stack;
7088
7089                ret_stack = per_cpu(idle_ret_stack, cpu);
7090                if (!ret_stack) {
7091                        ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
7092                                            * sizeof(struct ftrace_ret_stack),
7093                                            GFP_KERNEL);
7094                        if (!ret_stack)
7095                                return;
7096                        per_cpu(idle_ret_stack, cpu) = ret_stack;
7097                }
7098                graph_init_task(t, ret_stack);
7099        }
7100}
7101
7102/* Allocate a return stack for newly created task */
7103void ftrace_graph_init_task(struct task_struct *t)
7104{
7105        /* Make sure we do not use the parent ret_stack */
7106        t->ret_stack = NULL;
7107        t->curr_ret_stack = -1;
7108
7109        if (ftrace_graph_active) {
7110                struct ftrace_ret_stack *ret_stack;
7111
7112                ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
7113                                * sizeof(struct ftrace_ret_stack),
7114                                GFP_KERNEL);
7115                if (!ret_stack)
7116                        return;
7117                graph_init_task(t, ret_stack);
7118        }
7119}
7120
7121void ftrace_graph_exit_task(struct task_struct *t)
7122{
7123        struct ftrace_ret_stack *ret_stack = t->ret_stack;
7124
7125        t->ret_stack = NULL;
7126        /* NULL must become visible to IRQs before we free it: */
7127        barrier();
7128
7129        kfree(ret_stack);
7130}
7131#endif
7132