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/kallsyms.h>
  19#include <linux/seq_file.h>
  20#include <linux/suspend.h>
  21#include <linux/debugfs.h>
  22#include <linux/hardirq.h>
  23#include <linux/kthread.h>
  24#include <linux/uaccess.h>
  25#include <linux/bsearch.h>
  26#include <linux/module.h>
  27#include <linux/ftrace.h>
  28#include <linux/sysctl.h>
  29#include <linux/slab.h>
  30#include <linux/ctype.h>
  31#include <linux/sort.h>
  32#include <linux/list.h>
  33#include <linux/hash.h>
  34#include <linux/rcupdate.h>
  35
  36#include <trace/events/sched.h>
  37
  38#include <asm/setup.h>
  39
  40#include "trace_output.h"
  41#include "trace_stat.h"
  42
  43#define FTRACE_WARN_ON(cond)                    \
  44        ({                                      \
  45                int ___r = cond;                \
  46                if (WARN_ON(___r))              \
  47                        ftrace_kill();          \
  48                ___r;                           \
  49        })
  50
  51#define FTRACE_WARN_ON_ONCE(cond)               \
  52        ({                                      \
  53                int ___r = cond;                \
  54                if (WARN_ON_ONCE(___r))         \
  55                        ftrace_kill();          \
  56                ___r;                           \
  57        })
  58
  59/* hash bits for specific function selection */
  60#define FTRACE_HASH_BITS 7
  61#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
  62#define FTRACE_HASH_DEFAULT_BITS 10
  63#define FTRACE_HASH_MAX_BITS 12
  64
  65#define FL_GLOBAL_CONTROL_MASK (FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_CONTROL)
  66
  67#ifdef CONFIG_DYNAMIC_FTRACE
  68#define INIT_REGEX_LOCK(opsname)        \
  69        .regex_lock     = __MUTEX_INITIALIZER(opsname.regex_lock),
  70#else
  71#define INIT_REGEX_LOCK(opsname)
  72#endif
  73
  74static struct ftrace_ops ftrace_list_end __read_mostly = {
  75        .func           = ftrace_stub,
  76        .flags          = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
  77};
  78
  79/* ftrace_enabled is a method to turn ftrace on or off */
  80int ftrace_enabled __read_mostly;
  81static int last_ftrace_enabled;
  82
  83/* Quick disabling of function tracer. */
  84int function_trace_stop __read_mostly;
  85
  86/* Current function tracing op */
  87struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
  88
  89/* List for set_ftrace_pid's pids. */
  90LIST_HEAD(ftrace_pids);
  91struct ftrace_pid {
  92        struct list_head list;
  93        struct pid *pid;
  94};
  95
  96/*
  97 * ftrace_disabled is set when an anomaly is discovered.
  98 * ftrace_disabled is much stronger than ftrace_enabled.
  99 */
 100static int ftrace_disabled __read_mostly;
 101
 102static DEFINE_MUTEX(ftrace_lock);
 103
 104static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
 105static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end;
 106static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
 107ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
 108ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
 109static struct ftrace_ops global_ops;
 110static struct ftrace_ops control_ops;
 111
 112#if ARCH_SUPPORTS_FTRACE_OPS
 113static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
 114                                 struct ftrace_ops *op, struct pt_regs *regs);
 115#else
 116/* See comment below, where ftrace_ops_list_func is defined */
 117static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
 118#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
 119#endif
 120
 121/*
 122 * Traverse the ftrace_global_list, invoking all entries.  The reason that we
 123 * can use rcu_dereference_raw_notrace() is that elements removed from this list
 124 * are simply leaked, so there is no need to interact with a grace-period
 125 * mechanism.  The rcu_dereference_raw_notrace() calls are needed to handle
 126 * concurrent insertions into the ftrace_global_list.
 127 *
 128 * Silly Alpha and silly pointer-speculation compiler optimizations!
 129 */
 130#define do_for_each_ftrace_op(op, list)                 \
 131        op = rcu_dereference_raw_notrace(list);                 \
 132        do
 133
 134/*
 135 * Optimized for just a single item in the list (as that is the normal case).
 136 */
 137#define while_for_each_ftrace_op(op)                            \
 138        while (likely(op = rcu_dereference_raw_notrace((op)->next)) &&  \
 139               unlikely((op) != &ftrace_list_end))
 140
 141static inline void ftrace_ops_init(struct ftrace_ops *ops)
 142{
 143#ifdef CONFIG_DYNAMIC_FTRACE
 144        if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
 145                mutex_init(&ops->regex_lock);
 146                ops->flags |= FTRACE_OPS_FL_INITIALIZED;
 147        }
 148#endif
 149}
 150
 151/**
 152 * ftrace_nr_registered_ops - return number of ops registered
 153 *
 154 * Returns the number of ftrace_ops registered and tracing functions
 155 */
 156int ftrace_nr_registered_ops(void)
 157{
 158        struct ftrace_ops *ops;
 159        int cnt = 0;
 160
 161        mutex_lock(&ftrace_lock);
 162
 163        for (ops = ftrace_ops_list;
 164             ops != &ftrace_list_end; ops = ops->next)
 165                cnt++;
 166
 167        mutex_unlock(&ftrace_lock);
 168
 169        return cnt;
 170}
 171
 172static void
 173ftrace_global_list_func(unsigned long ip, unsigned long parent_ip,
 174                        struct ftrace_ops *op, struct pt_regs *regs)
 175{
 176        int bit;
 177
 178        bit = trace_test_and_set_recursion(TRACE_GLOBAL_START, TRACE_GLOBAL_MAX);
 179        if (bit < 0)
 180                return;
 181
 182        do_for_each_ftrace_op(op, ftrace_global_list) {
 183                op->func(ip, parent_ip, op, regs);
 184        } while_for_each_ftrace_op(op);
 185
 186        trace_clear_recursion(bit);
 187}
 188
 189static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
 190                            struct ftrace_ops *op, struct pt_regs *regs)
 191{
 192        if (!test_tsk_trace_trace(current))
 193                return;
 194
 195        ftrace_pid_function(ip, parent_ip, op, regs);
 196}
 197
 198static void set_ftrace_pid_function(ftrace_func_t func)
 199{
 200        /* do not set ftrace_pid_function to itself! */
 201        if (func != ftrace_pid_func)
 202                ftrace_pid_function = func;
 203}
 204
 205/**
 206 * clear_ftrace_function - reset the ftrace function
 207 *
 208 * This NULLs the ftrace function and in essence stops
 209 * tracing.  There may be lag
 210 */
 211void clear_ftrace_function(void)
 212{
 213        ftrace_trace_function = ftrace_stub;
 214        ftrace_pid_function = ftrace_stub;
 215}
 216
 217static void control_ops_disable_all(struct ftrace_ops *ops)
 218{
 219        int cpu;
 220
 221        for_each_possible_cpu(cpu)
 222                *per_cpu_ptr(ops->disabled, cpu) = 1;
 223}
 224
 225static int control_ops_alloc(struct ftrace_ops *ops)
 226{
 227        int __percpu *disabled;
 228
 229        disabled = alloc_percpu(int);
 230        if (!disabled)
 231                return -ENOMEM;
 232
 233        ops->disabled = disabled;
 234        control_ops_disable_all(ops);
 235        return 0;
 236}
 237
 238static void control_ops_free(struct ftrace_ops *ops)
 239{
 240        free_percpu(ops->disabled);
 241}
 242
 243static void update_global_ops(void)
 244{
 245        ftrace_func_t func;
 246
 247        /*
 248         * If there's only one function registered, then call that
 249         * function directly. Otherwise, we need to iterate over the
 250         * registered callers.
 251         */
 252        if (ftrace_global_list == &ftrace_list_end ||
 253            ftrace_global_list->next == &ftrace_list_end) {
 254                func = ftrace_global_list->func;
 255                /*
 256                 * As we are calling the function directly.
 257                 * If it does not have recursion protection,
 258                 * the function_trace_op needs to be updated
 259                 * accordingly.
 260                 */
 261                if (ftrace_global_list->flags & FTRACE_OPS_FL_RECURSION_SAFE)
 262                        global_ops.flags |= FTRACE_OPS_FL_RECURSION_SAFE;
 263                else
 264                        global_ops.flags &= ~FTRACE_OPS_FL_RECURSION_SAFE;
 265        } else {
 266                func = ftrace_global_list_func;
 267                /* The list has its own recursion protection. */
 268                global_ops.flags |= FTRACE_OPS_FL_RECURSION_SAFE;
 269        }
 270
 271
 272        /* If we filter on pids, update to use the pid function */
 273        if (!list_empty(&ftrace_pids)) {
 274                set_ftrace_pid_function(func);
 275                func = ftrace_pid_func;
 276        }
 277
 278        global_ops.func = func;
 279}
 280
 281static void update_ftrace_function(void)
 282{
 283        ftrace_func_t func;
 284
 285        update_global_ops();
 286
 287        /*
 288         * If we are at the end of the list and this ops is
 289         * recursion safe and not dynamic and the arch supports passing ops,
 290         * then have the mcount trampoline call the function directly.
 291         */
 292        if (ftrace_ops_list == &ftrace_list_end ||
 293            (ftrace_ops_list->next == &ftrace_list_end &&
 294             !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC) &&
 295             (ftrace_ops_list->flags & FTRACE_OPS_FL_RECURSION_SAFE) &&
 296             !FTRACE_FORCE_LIST_FUNC)) {
 297                /* Set the ftrace_ops that the arch callback uses */
 298                if (ftrace_ops_list == &global_ops)
 299                        function_trace_op = ftrace_global_list;
 300                else
 301                        function_trace_op = ftrace_ops_list;
 302                func = ftrace_ops_list->func;
 303        } else {
 304                /* Just use the default ftrace_ops */
 305                function_trace_op = &ftrace_list_end;
 306                func = ftrace_ops_list_func;
 307        }
 308
 309        ftrace_trace_function = func;
 310}
 311
 312static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
 313{
 314        ops->next = *list;
 315        /*
 316         * We are entering ops into the list but another
 317         * CPU might be walking that list. We need to make sure
 318         * the ops->next pointer is valid before another CPU sees
 319         * the ops pointer included into the list.
 320         */
 321        rcu_assign_pointer(*list, ops);
 322}
 323
 324static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
 325{
 326        struct ftrace_ops **p;
 327
 328        /*
 329         * If we are removing the last function, then simply point
 330         * to the ftrace_stub.
 331         */
 332        if (*list == ops && ops->next == &ftrace_list_end) {
 333                *list = &ftrace_list_end;
 334                return 0;
 335        }
 336
 337        for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
 338                if (*p == ops)
 339                        break;
 340
 341        if (*p != ops)
 342                return -1;
 343
 344        *p = (*p)->next;
 345        return 0;
 346}
 347
 348static void add_ftrace_list_ops(struct ftrace_ops **list,
 349                                struct ftrace_ops *main_ops,
 350                                struct ftrace_ops *ops)
 351{
 352        int first = *list == &ftrace_list_end;
 353        add_ftrace_ops(list, ops);
 354        if (first)
 355                add_ftrace_ops(&ftrace_ops_list, main_ops);
 356}
 357
 358static int remove_ftrace_list_ops(struct ftrace_ops **list,
 359                                  struct ftrace_ops *main_ops,
 360                                  struct ftrace_ops *ops)
 361{
 362        int ret = remove_ftrace_ops(list, ops);
 363        if (!ret && *list == &ftrace_list_end)
 364                ret = remove_ftrace_ops(&ftrace_ops_list, main_ops);
 365        return ret;
 366}
 367
 368static int __register_ftrace_function(struct ftrace_ops *ops)
 369{
 370        if (unlikely(ftrace_disabled))
 371                return -ENODEV;
 372
 373        if (FTRACE_WARN_ON(ops == &global_ops))
 374                return -EINVAL;
 375
 376        if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
 377                return -EBUSY;
 378
 379        /* We don't support both control and global flags set. */
 380        if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK)
 381                return -EINVAL;
 382
 383#ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
 384        /*
 385         * If the ftrace_ops specifies SAVE_REGS, then it only can be used
 386         * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
 387         * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
 388         */
 389        if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
 390            !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
 391                return -EINVAL;
 392
 393        if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
 394                ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
 395#endif
 396
 397        if (!core_kernel_data((unsigned long)ops))
 398                ops->flags |= FTRACE_OPS_FL_DYNAMIC;
 399
 400        if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
 401                add_ftrace_list_ops(&ftrace_global_list, &global_ops, ops);
 402                ops->flags |= FTRACE_OPS_FL_ENABLED;
 403        } else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
 404                if (control_ops_alloc(ops))
 405                        return -ENOMEM;
 406                add_ftrace_list_ops(&ftrace_control_list, &control_ops, ops);
 407        } else
 408                add_ftrace_ops(&ftrace_ops_list, ops);
 409
 410        if (ftrace_enabled)
 411                update_ftrace_function();
 412
 413        return 0;
 414}
 415
 416static void ftrace_sync(struct work_struct *work)
 417{
 418        /*
 419         * This function is just a stub to implement a hard force
 420         * of synchronize_sched(). This requires synchronizing
 421         * tasks even in userspace and idle.
 422         *
 423         * Yes, function tracing is rude.
 424         */
 425}
 426
 427static int __unregister_ftrace_function(struct ftrace_ops *ops)
 428{
 429        int ret;
 430
 431        if (ftrace_disabled)
 432                return -ENODEV;
 433
 434        if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
 435                return -EBUSY;
 436
 437        if (FTRACE_WARN_ON(ops == &global_ops))
 438                return -EINVAL;
 439
 440        if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
 441                ret = remove_ftrace_list_ops(&ftrace_global_list,
 442                                             &global_ops, ops);
 443                if (!ret)
 444                        ops->flags &= ~FTRACE_OPS_FL_ENABLED;
 445        } else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
 446                ret = remove_ftrace_list_ops(&ftrace_control_list,
 447                                             &control_ops, ops);
 448                if (!ret) {
 449                        /*
 450                         * The ftrace_ops is now removed from the list,
 451                         * so there'll be no new users. We must ensure
 452                         * all current users are done before we free
 453                         * the control data.
 454                         * Note synchronize_sched() is not enough, as we
 455                         * use preempt_disable() to do RCU, but the function
 456                         * tracer can be called where RCU is not active
 457                         * (before user_exit()).
 458                         */
 459                        schedule_on_each_cpu(ftrace_sync);
 460                        control_ops_free(ops);
 461                }
 462        } else
 463                ret = remove_ftrace_ops(&ftrace_ops_list, ops);
 464
 465        if (ret < 0)
 466                return ret;
 467
 468        if (ftrace_enabled)
 469                update_ftrace_function();
 470
 471        /*
 472         * Dynamic ops may be freed, we must make sure that all
 473         * callers are done before leaving this function.
 474         *
 475         * Again, normal synchronize_sched() is not good enough.
 476         * We need to do a hard force of sched synchronization.
 477         */
 478        if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
 479                schedule_on_each_cpu(ftrace_sync);
 480
 481
 482        return 0;
 483}
 484
 485static void ftrace_update_pid_func(void)
 486{
 487        /* Only do something if we are tracing something */
 488        if (ftrace_trace_function == ftrace_stub)
 489                return;
 490
 491        update_ftrace_function();
 492}
 493
 494#ifdef CONFIG_FUNCTION_PROFILER
 495struct ftrace_profile {
 496        struct hlist_node               node;
 497        unsigned long                   ip;
 498        unsigned long                   counter;
 499#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 500        unsigned long long              time;
 501        unsigned long long              time_squared;
 502#endif
 503};
 504
 505struct ftrace_profile_page {
 506        struct ftrace_profile_page      *next;
 507        unsigned long                   index;
 508        struct ftrace_profile           records[];
 509};
 510
 511struct ftrace_profile_stat {
 512        atomic_t                        disabled;
 513        struct hlist_head               *hash;
 514        struct ftrace_profile_page      *pages;
 515        struct ftrace_profile_page      *start;
 516        struct tracer_stat              stat;
 517};
 518
 519#define PROFILE_RECORDS_SIZE                                            \
 520        (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
 521
 522#define PROFILES_PER_PAGE                                       \
 523        (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
 524
 525static int ftrace_profile_enabled __read_mostly;
 526
 527/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
 528static DEFINE_MUTEX(ftrace_profile_lock);
 529
 530static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
 531
 532#define FTRACE_PROFILE_HASH_BITS 10
 533#define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
 534
 535static void *
 536function_stat_next(void *v, int idx)
 537{
 538        struct ftrace_profile *rec = v;
 539        struct ftrace_profile_page *pg;
 540
 541        pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
 542
 543 again:
 544        if (idx != 0)
 545                rec++;
 546
 547        if ((void *)rec >= (void *)&pg->records[pg->index]) {
 548                pg = pg->next;
 549                if (!pg)
 550                        return NULL;
 551                rec = &pg->records[0];
 552                if (!rec->counter)
 553                        goto again;
 554        }
 555
 556        return rec;
 557}
 558
 559static void *function_stat_start(struct tracer_stat *trace)
 560{
 561        struct ftrace_profile_stat *stat =
 562                container_of(trace, struct ftrace_profile_stat, stat);
 563
 564        if (!stat || !stat->start)
 565                return NULL;
 566
 567        return function_stat_next(&stat->start->records[0], 0);
 568}
 569
 570#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 571/* function graph compares on total time */
 572static int function_stat_cmp(void *p1, void *p2)
 573{
 574        struct ftrace_profile *a = p1;
 575        struct ftrace_profile *b = p2;
 576
 577        if (a->time < b->time)
 578                return -1;
 579        if (a->time > b->time)
 580                return 1;
 581        else
 582                return 0;
 583}
 584#else
 585/* not function graph compares against hits */
 586static int function_stat_cmp(void *p1, void *p2)
 587{
 588        struct ftrace_profile *a = p1;
 589        struct ftrace_profile *b = p2;
 590
 591        if (a->counter < b->counter)
 592                return -1;
 593        if (a->counter > b->counter)
 594                return 1;
 595        else
 596                return 0;
 597}
 598#endif
 599
 600static int function_stat_headers(struct seq_file *m)
 601{
 602#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 603        seq_printf(m, "  Function                               "
 604                   "Hit    Time            Avg             s^2\n"
 605                      "  --------                               "
 606                   "---    ----            ---             ---\n");
 607#else
 608        seq_printf(m, "  Function                               Hit\n"
 609                      "  --------                               ---\n");
 610#endif
 611        return 0;
 612}
 613
 614static int function_stat_show(struct seq_file *m, void *v)
 615{
 616        struct ftrace_profile *rec = v;
 617        char str[KSYM_SYMBOL_LEN];
 618        int ret = 0;
 619#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 620        static struct trace_seq s;
 621        unsigned long long avg;
 622        unsigned long long stddev;
 623#endif
 624        mutex_lock(&ftrace_profile_lock);
 625
 626        /* we raced with function_profile_reset() */
 627        if (unlikely(rec->counter == 0)) {
 628                ret = -EBUSY;
 629                goto out;
 630        }
 631
 632        kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 633        seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
 634
 635#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 636        seq_printf(m, "    ");
 637        avg = rec->time;
 638        do_div(avg, rec->counter);
 639
 640        /* Sample standard deviation (s^2) */
 641        if (rec->counter <= 1)
 642                stddev = 0;
 643        else {
 644                /*
 645                 * Apply Welford's method:
 646                 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
 647                 */
 648                stddev = rec->counter * rec->time_squared -
 649                         rec->time * rec->time;
 650
 651                /*
 652                 * Divide only 1000 for ns^2 -> us^2 conversion.
 653                 * trace_print_graph_duration will divide 1000 again.
 654                 */
 655                do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
 656        }
 657
 658        trace_seq_init(&s);
 659        trace_print_graph_duration(rec->time, &s);
 660        trace_seq_puts(&s, "    ");
 661        trace_print_graph_duration(avg, &s);
 662        trace_seq_puts(&s, "    ");
 663        trace_print_graph_duration(stddev, &s);
 664        trace_print_seq(m, &s);
 665#endif
 666        seq_putc(m, '\n');
 667out:
 668        mutex_unlock(&ftrace_profile_lock);
 669
 670        return ret;
 671}
 672
 673static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
 674{
 675        struct ftrace_profile_page *pg;
 676
 677        pg = stat->pages = stat->start;
 678
 679        while (pg) {
 680                memset(pg->records, 0, PROFILE_RECORDS_SIZE);
 681                pg->index = 0;
 682                pg = pg->next;
 683        }
 684
 685        memset(stat->hash, 0,
 686               FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
 687}
 688
 689int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
 690{
 691        struct ftrace_profile_page *pg;
 692        int functions;
 693        int pages;
 694        int i;
 695
 696        /* If we already allocated, do nothing */
 697        if (stat->pages)
 698                return 0;
 699
 700        stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
 701        if (!stat->pages)
 702                return -ENOMEM;
 703
 704#ifdef CONFIG_DYNAMIC_FTRACE
 705        functions = ftrace_update_tot_cnt;
 706#else
 707        /*
 708         * We do not know the number of functions that exist because
 709         * dynamic tracing is what counts them. With past experience
 710         * we have around 20K functions. That should be more than enough.
 711         * It is highly unlikely we will execute every function in
 712         * the kernel.
 713         */
 714        functions = 20000;
 715#endif
 716
 717        pg = stat->start = stat->pages;
 718
 719        pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
 720
 721        for (i = 1; i < pages; i++) {
 722                pg->next = (void *)get_zeroed_page(GFP_KERNEL);
 723                if (!pg->next)
 724                        goto out_free;
 725                pg = pg->next;
 726        }
 727
 728        return 0;
 729
 730 out_free:
 731        pg = stat->start;
 732        while (pg) {
 733                unsigned long tmp = (unsigned long)pg;
 734
 735                pg = pg->next;
 736                free_page(tmp);
 737        }
 738
 739        stat->pages = NULL;
 740        stat->start = NULL;
 741
 742        return -ENOMEM;
 743}
 744
 745static int ftrace_profile_init_cpu(int cpu)
 746{
 747        struct ftrace_profile_stat *stat;
 748        int size;
 749
 750        stat = &per_cpu(ftrace_profile_stats, cpu);
 751
 752        if (stat->hash) {
 753                /* If the profile is already created, simply reset it */
 754                ftrace_profile_reset(stat);
 755                return 0;
 756        }
 757
 758        /*
 759         * We are profiling all functions, but usually only a few thousand
 760         * functions are hit. We'll make a hash of 1024 items.
 761         */
 762        size = FTRACE_PROFILE_HASH_SIZE;
 763
 764        stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
 765
 766        if (!stat->hash)
 767                return -ENOMEM;
 768
 769        /* Preallocate the function profiling pages */
 770        if (ftrace_profile_pages_init(stat) < 0) {
 771                kfree(stat->hash);
 772                stat->hash = NULL;
 773                return -ENOMEM;
 774        }
 775
 776        return 0;
 777}
 778
 779static int ftrace_profile_init(void)
 780{
 781        int cpu;
 782        int ret = 0;
 783
 784        for_each_online_cpu(cpu) {
 785                ret = ftrace_profile_init_cpu(cpu);
 786                if (ret)
 787                        break;
 788        }
 789
 790        return ret;
 791}
 792
 793/* interrupts must be disabled */
 794static struct ftrace_profile *
 795ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
 796{
 797        struct ftrace_profile *rec;
 798        struct hlist_head *hhd;
 799        unsigned long key;
 800
 801        key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
 802        hhd = &stat->hash[key];
 803
 804        if (hlist_empty(hhd))
 805                return NULL;
 806
 807        hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
 808                if (rec->ip == ip)
 809                        return rec;
 810        }
 811
 812        return NULL;
 813}
 814
 815static void ftrace_add_profile(struct ftrace_profile_stat *stat,
 816                               struct ftrace_profile *rec)
 817{
 818        unsigned long key;
 819
 820        key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
 821        hlist_add_head_rcu(&rec->node, &stat->hash[key]);
 822}
 823
 824/*
 825 * The memory is already allocated, this simply finds a new record to use.
 826 */
 827static struct ftrace_profile *
 828ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
 829{
 830        struct ftrace_profile *rec = NULL;
 831
 832        /* prevent recursion (from NMIs) */
 833        if (atomic_inc_return(&stat->disabled) != 1)
 834                goto out;
 835
 836        /*
 837         * Try to find the function again since an NMI
 838         * could have added it
 839         */
 840        rec = ftrace_find_profiled_func(stat, ip);
 841        if (rec)
 842                goto out;
 843
 844        if (stat->pages->index == PROFILES_PER_PAGE) {
 845                if (!stat->pages->next)
 846                        goto out;
 847                stat->pages = stat->pages->next;
 848        }
 849
 850        rec = &stat->pages->records[stat->pages->index++];
 851        rec->ip = ip;
 852        ftrace_add_profile(stat, rec);
 853
 854 out:
 855        atomic_dec(&stat->disabled);
 856
 857        return rec;
 858}
 859
 860static void
 861function_profile_call(unsigned long ip, unsigned long parent_ip,
 862                      struct ftrace_ops *ops, struct pt_regs *regs)
 863{
 864        struct ftrace_profile_stat *stat;
 865        struct ftrace_profile *rec;
 866        unsigned long flags;
 867
 868        if (!ftrace_profile_enabled)
 869                return;
 870
 871        local_irq_save(flags);
 872
 873        stat = &__get_cpu_var(ftrace_profile_stats);
 874        if (!stat->hash || !ftrace_profile_enabled)
 875                goto out;
 876
 877        rec = ftrace_find_profiled_func(stat, ip);
 878        if (!rec) {
 879                rec = ftrace_profile_alloc(stat, ip);
 880                if (!rec)
 881                        goto out;
 882        }
 883
 884        rec->counter++;
 885 out:
 886        local_irq_restore(flags);
 887}
 888
 889#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 890static int profile_graph_entry(struct ftrace_graph_ent *trace)
 891{
 892        function_profile_call(trace->func, 0, NULL, NULL);
 893        return 1;
 894}
 895
 896static void profile_graph_return(struct ftrace_graph_ret *trace)
 897{
 898        struct ftrace_profile_stat *stat;
 899        unsigned long long calltime;
 900        struct ftrace_profile *rec;
 901        unsigned long flags;
 902
 903        local_irq_save(flags);
 904        stat = &__get_cpu_var(ftrace_profile_stats);
 905        if (!stat->hash || !ftrace_profile_enabled)
 906                goto out;
 907
 908        /* If the calltime was zero'd ignore it */
 909        if (!trace->calltime)
 910                goto out;
 911
 912        calltime = trace->rettime - trace->calltime;
 913
 914        if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
 915                int index;
 916
 917                index = trace->depth;
 918
 919                /* Append this call time to the parent time to subtract */
 920                if (index)
 921                        current->ret_stack[index - 1].subtime += calltime;
 922
 923                if (current->ret_stack[index].subtime < calltime)
 924                        calltime -= current->ret_stack[index].subtime;
 925                else
 926                        calltime = 0;
 927        }
 928
 929        rec = ftrace_find_profiled_func(stat, trace->func);
 930        if (rec) {
 931                rec->time += calltime;
 932                rec->time_squared += calltime * calltime;
 933        }
 934
 935 out:
 936        local_irq_restore(flags);
 937}
 938
 939static int register_ftrace_profiler(void)
 940{
 941        return register_ftrace_graph(&profile_graph_return,
 942                                     &profile_graph_entry);
 943}
 944
 945static void unregister_ftrace_profiler(void)
 946{
 947        unregister_ftrace_graph();
 948}
 949#else
 950static struct ftrace_ops ftrace_profile_ops __read_mostly = {
 951        .func           = function_profile_call,
 952        .flags          = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
 953        INIT_REGEX_LOCK(ftrace_profile_ops)
 954};
 955
 956static int register_ftrace_profiler(void)
 957{
 958        return register_ftrace_function(&ftrace_profile_ops);
 959}
 960
 961static void unregister_ftrace_profiler(void)
 962{
 963        unregister_ftrace_function(&ftrace_profile_ops);
 964}
 965#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 966
 967static ssize_t
 968ftrace_profile_write(struct file *filp, const char __user *ubuf,
 969                     size_t cnt, loff_t *ppos)
 970{
 971        unsigned long val;
 972        int ret;
 973
 974        ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 975        if (ret)
 976                return ret;
 977
 978        val = !!val;
 979
 980        mutex_lock(&ftrace_profile_lock);
 981        if (ftrace_profile_enabled ^ val) {
 982                if (val) {
 983                        ret = ftrace_profile_init();
 984                        if (ret < 0) {
 985                                cnt = ret;
 986                                goto out;
 987                        }
 988
 989                        ret = register_ftrace_profiler();
 990                        if (ret < 0) {
 991                                cnt = ret;
 992                                goto out;
 993                        }
 994                        ftrace_profile_enabled = 1;
 995                } else {
 996                        ftrace_profile_enabled = 0;
 997                        /*
 998                         * unregister_ftrace_profiler calls stop_machine
 999                         * so this acts like an synchronize_sched.
1000                         */
1001                        unregister_ftrace_profiler();
1002                }
1003        }
1004 out:
1005        mutex_unlock(&ftrace_profile_lock);
1006
1007        *ppos += cnt;
1008
1009        return cnt;
1010}
1011
1012static ssize_t
1013ftrace_profile_read(struct file *filp, char __user *ubuf,
1014                     size_t cnt, loff_t *ppos)
1015{
1016        char buf[64];           /* big enough to hold a number */
1017        int r;
1018
1019        r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1020        return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1021}
1022
1023static const struct file_operations ftrace_profile_fops = {
1024        .open           = tracing_open_generic,
1025        .read           = ftrace_profile_read,
1026        .write          = ftrace_profile_write,
1027        .llseek         = default_llseek,
1028};
1029
1030/* used to initialize the real stat files */
1031static struct tracer_stat function_stats __initdata = {
1032        .name           = "functions",
1033        .stat_start     = function_stat_start,
1034        .stat_next      = function_stat_next,
1035        .stat_cmp       = function_stat_cmp,
1036        .stat_headers   = function_stat_headers,
1037        .stat_show      = function_stat_show
1038};
1039
1040static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1041{
1042        struct ftrace_profile_stat *stat;
1043        struct dentry *entry;
1044        char *name;
1045        int ret;
1046        int cpu;
1047
1048        for_each_possible_cpu(cpu) {
1049                stat = &per_cpu(ftrace_profile_stats, cpu);
1050
1051                /* allocate enough for function name + cpu number */
1052                name = kmalloc(32, GFP_KERNEL);
1053                if (!name) {
1054                        /*
1055                         * The files created are permanent, if something happens
1056                         * we still do not free memory.
1057                         */
1058                        WARN(1,
1059                             "Could not allocate stat file for cpu %d\n",
1060                             cpu);
1061                        return;
1062                }
1063                stat->stat = function_stats;
1064                snprintf(name, 32, "function%d", cpu);
1065                stat->stat.name = name;
1066                ret = register_stat_tracer(&stat->stat);
1067                if (ret) {
1068                        WARN(1,
1069                             "Could not register function stat for cpu %d\n",
1070                             cpu);
1071                        kfree(name);
1072                        return;
1073                }
1074        }
1075
1076        entry = debugfs_create_file("function_profile_enabled", 0644,
1077                                    d_tracer, NULL, &ftrace_profile_fops);
1078        if (!entry)
1079                pr_warning("Could not create debugfs "
1080                           "'function_profile_enabled' entry\n");
1081}
1082
1083#else /* CONFIG_FUNCTION_PROFILER */
1084static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1085{
1086}
1087#endif /* CONFIG_FUNCTION_PROFILER */
1088
1089static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1090
1091loff_t
1092ftrace_filter_lseek(struct file *file, loff_t offset, int whence)
1093{
1094        loff_t ret;
1095
1096        if (file->f_mode & FMODE_READ)
1097                ret = seq_lseek(file, offset, whence);
1098        else
1099                file->f_pos = ret = 1;
1100
1101        return ret;
1102}
1103
1104#ifdef CONFIG_DYNAMIC_FTRACE
1105
1106#ifndef CONFIG_FTRACE_MCOUNT_RECORD
1107# error Dynamic ftrace depends on MCOUNT_RECORD
1108#endif
1109
1110static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1111
1112struct ftrace_func_probe {
1113        struct hlist_node       node;
1114        struct ftrace_probe_ops *ops;
1115        unsigned long           flags;
1116        unsigned long           ip;
1117        void                    *data;
1118        struct list_head        free_list;
1119};
1120
1121struct ftrace_func_entry {
1122        struct hlist_node hlist;
1123        unsigned long ip;
1124};
1125
1126struct ftrace_hash {
1127        unsigned long           size_bits;
1128        struct hlist_head       *buckets;
1129        unsigned long           count;
1130        struct rcu_head         rcu;
1131};
1132
1133/*
1134 * We make these constant because no one should touch them,
1135 * but they are used as the default "empty hash", to avoid allocating
1136 * it all the time. These are in a read only section such that if
1137 * anyone does try to modify it, it will cause an exception.
1138 */
1139static const struct hlist_head empty_buckets[1];
1140static const struct ftrace_hash empty_hash = {
1141        .buckets = (struct hlist_head *)empty_buckets,
1142};
1143#define EMPTY_HASH      ((struct ftrace_hash *)&empty_hash)
1144
1145static struct ftrace_ops global_ops = {
1146        .func                   = ftrace_stub,
1147        .notrace_hash           = EMPTY_HASH,
1148        .filter_hash            = EMPTY_HASH,
1149        .flags                  = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
1150        INIT_REGEX_LOCK(global_ops)
1151};
1152
1153struct ftrace_page {
1154        struct ftrace_page      *next;
1155        struct dyn_ftrace       *records;
1156        int                     index;
1157        int                     size;
1158};
1159
1160static struct ftrace_page *ftrace_new_pgs;
1161
1162#define ENTRY_SIZE sizeof(struct dyn_ftrace)
1163#define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1164
1165/* estimate from running different kernels */
1166#define NR_TO_INIT              10000
1167
1168static struct ftrace_page       *ftrace_pages_start;
1169static struct ftrace_page       *ftrace_pages;
1170
1171static bool ftrace_hash_empty(struct ftrace_hash *hash)
1172{
1173        return !hash || !hash->count;
1174}
1175
1176static struct ftrace_func_entry *
1177ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1178{
1179        unsigned long key;
1180        struct ftrace_func_entry *entry;
1181        struct hlist_head *hhd;
1182
1183        if (ftrace_hash_empty(hash))
1184                return NULL;
1185
1186        if (hash->size_bits > 0)
1187                key = hash_long(ip, hash->size_bits);
1188        else
1189                key = 0;
1190
1191        hhd = &hash->buckets[key];
1192
1193        hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1194                if (entry->ip == ip)
1195                        return entry;
1196        }
1197        return NULL;
1198}
1199
1200static void __add_hash_entry(struct ftrace_hash *hash,
1201                             struct ftrace_func_entry *entry)
1202{
1203        struct hlist_head *hhd;
1204        unsigned long key;
1205
1206        if (hash->size_bits)
1207                key = hash_long(entry->ip, hash->size_bits);
1208        else
1209                key = 0;
1210
1211        hhd = &hash->buckets[key];
1212        hlist_add_head(&entry->hlist, hhd);
1213        hash->count++;
1214}
1215
1216static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1217{
1218        struct ftrace_func_entry *entry;
1219
1220        entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1221        if (!entry)
1222                return -ENOMEM;
1223
1224        entry->ip = ip;
1225        __add_hash_entry(hash, entry);
1226
1227        return 0;
1228}
1229
1230static void
1231free_hash_entry(struct ftrace_hash *hash,
1232                  struct ftrace_func_entry *entry)
1233{
1234        hlist_del(&entry->hlist);
1235        kfree(entry);
1236        hash->count--;
1237}
1238
1239static void
1240remove_hash_entry(struct ftrace_hash *hash,
1241                  struct ftrace_func_entry *entry)
1242{
1243        hlist_del(&entry->hlist);
1244        hash->count--;
1245}
1246
1247static void ftrace_hash_clear(struct ftrace_hash *hash)
1248{
1249        struct hlist_head *hhd;
1250        struct hlist_node *tn;
1251        struct ftrace_func_entry *entry;
1252        int size = 1 << hash->size_bits;
1253        int i;
1254
1255        if (!hash->count)
1256                return;
1257
1258        for (i = 0; i < size; i++) {
1259                hhd = &hash->buckets[i];
1260                hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1261                        free_hash_entry(hash, entry);
1262        }
1263        FTRACE_WARN_ON(hash->count);
1264}
1265
1266static void free_ftrace_hash(struct ftrace_hash *hash)
1267{
1268        if (!hash || hash == EMPTY_HASH)
1269                return;
1270        ftrace_hash_clear(hash);
1271        kfree(hash->buckets);
1272        kfree(hash);
1273}
1274
1275static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1276{
1277        struct ftrace_hash *hash;
1278
1279        hash = container_of(rcu, struct ftrace_hash, rcu);
1280        free_ftrace_hash(hash);
1281}
1282
1283static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1284{
1285        if (!hash || hash == EMPTY_HASH)
1286                return;
1287        call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1288}
1289
1290void ftrace_free_filter(struct ftrace_ops *ops)
1291{
1292        ftrace_ops_init(ops);
1293        free_ftrace_hash(ops->filter_hash);
1294        free_ftrace_hash(ops->notrace_hash);
1295}
1296
1297static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1298{
1299        struct ftrace_hash *hash;
1300        int size;
1301
1302        hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1303        if (!hash)
1304                return NULL;
1305
1306        size = 1 << size_bits;
1307        hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1308
1309        if (!hash->buckets) {
1310                kfree(hash);
1311                return NULL;
1312        }
1313
1314        hash->size_bits = size_bits;
1315
1316        return hash;
1317}
1318
1319static struct ftrace_hash *
1320alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1321{
1322        struct ftrace_func_entry *entry;
1323        struct ftrace_hash *new_hash;
1324        int size;
1325        int ret;
1326        int i;
1327
1328        new_hash = alloc_ftrace_hash(size_bits);
1329        if (!new_hash)
1330                return NULL;
1331
1332        /* Empty hash? */
1333        if (ftrace_hash_empty(hash))
1334                return new_hash;
1335
1336        size = 1 << hash->size_bits;
1337        for (i = 0; i < size; i++) {
1338                hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1339                        ret = add_hash_entry(new_hash, entry->ip);
1340                        if (ret < 0)
1341                                goto free_hash;
1342                }
1343        }
1344
1345        FTRACE_WARN_ON(new_hash->count != hash->count);
1346
1347        return new_hash;
1348
1349 free_hash:
1350        free_ftrace_hash(new_hash);
1351        return NULL;
1352}
1353
1354static void
1355ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1356static void
1357ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1358
1359static int
1360ftrace_hash_move(struct ftrace_ops *ops, int enable,
1361                 struct ftrace_hash **dst, struct ftrace_hash *src)
1362{
1363        struct ftrace_func_entry *entry;
1364        struct hlist_node *tn;
1365        struct hlist_head *hhd;
1366        struct ftrace_hash *old_hash;
1367        struct ftrace_hash *new_hash;
1368        int size = src->count;
1369        int bits = 0;
1370        int ret;
1371        int i;
1372
1373        /*
1374         * Remove the current set, update the hash and add
1375         * them back.
1376         */
1377        ftrace_hash_rec_disable(ops, enable);
1378
1379        /*
1380         * If the new source is empty, just free dst and assign it
1381         * the empty_hash.
1382         */
1383        if (!src->count) {
1384                free_ftrace_hash_rcu(*dst);
1385                rcu_assign_pointer(*dst, EMPTY_HASH);
1386                /* still need to update the function records */
1387                ret = 0;
1388                goto out;
1389        }
1390
1391        /*
1392         * Make the hash size about 1/2 the # found
1393         */
1394        for (size /= 2; size; size >>= 1)
1395                bits++;
1396
1397        /* Don't allocate too much */
1398        if (bits > FTRACE_HASH_MAX_BITS)
1399                bits = FTRACE_HASH_MAX_BITS;
1400
1401        ret = -ENOMEM;
1402        new_hash = alloc_ftrace_hash(bits);
1403        if (!new_hash)
1404                goto out;
1405
1406        size = 1 << src->size_bits;
1407        for (i = 0; i < size; i++) {
1408                hhd = &src->buckets[i];
1409                hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1410                        remove_hash_entry(src, entry);
1411                        __add_hash_entry(new_hash, entry);
1412                }
1413        }
1414
1415        old_hash = *dst;
1416        rcu_assign_pointer(*dst, new_hash);
1417        free_ftrace_hash_rcu(old_hash);
1418
1419        ret = 0;
1420 out:
1421        /*
1422         * Enable regardless of ret:
1423         *  On success, we enable the new hash.
1424         *  On failure, we re-enable the original hash.
1425         */
1426        ftrace_hash_rec_enable(ops, enable);
1427
1428        return ret;
1429}
1430
1431/*
1432 * Test the hashes for this ops to see if we want to call
1433 * the ops->func or not.
1434 *
1435 * It's a match if the ip is in the ops->filter_hash or
1436 * the filter_hash does not exist or is empty,
1437 *  AND
1438 * the ip is not in the ops->notrace_hash.
1439 *
1440 * This needs to be called with preemption disabled as
1441 * the hashes are freed with call_rcu_sched().
1442 */
1443static int
1444ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1445{
1446        struct ftrace_hash *filter_hash;
1447        struct ftrace_hash *notrace_hash;
1448        int ret;
1449
1450#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1451        /*
1452         * There's a small race when adding ops that the ftrace handler
1453         * that wants regs, may be called without them. We can not
1454         * allow that handler to be called if regs is NULL.
1455         */
1456        if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1457                return 0;
1458#endif
1459
1460        filter_hash = rcu_dereference_raw_notrace(ops->filter_hash);
1461        notrace_hash = rcu_dereference_raw_notrace(ops->notrace_hash);
1462
1463        if ((ftrace_hash_empty(filter_hash) ||
1464             ftrace_lookup_ip(filter_hash, ip)) &&
1465            (ftrace_hash_empty(notrace_hash) ||
1466             !ftrace_lookup_ip(notrace_hash, ip)))
1467                ret = 1;
1468        else
1469                ret = 0;
1470
1471        return ret;
1472}
1473
1474/*
1475 * This is a double for. Do not use 'break' to break out of the loop,
1476 * you must use a goto.
1477 */
1478#define do_for_each_ftrace_rec(pg, rec)                                 \
1479        for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
1480                int _____i;                                             \
1481                for (_____i = 0; _____i < pg->index; _____i++) {        \
1482                        rec = &pg->records[_____i];
1483
1484#define while_for_each_ftrace_rec()             \
1485                }                               \
1486        }
1487
1488
1489static int ftrace_cmp_recs(const void *a, const void *b)
1490{
1491        const struct dyn_ftrace *key = a;
1492        const struct dyn_ftrace *rec = b;
1493
1494        if (key->flags < rec->ip)
1495                return -1;
1496        if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1497                return 1;
1498        return 0;
1499}
1500
1501static unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1502{
1503        struct ftrace_page *pg;
1504        struct dyn_ftrace *rec;
1505        struct dyn_ftrace key;
1506
1507        key.ip = start;
1508        key.flags = end;        /* overload flags, as it is unsigned long */
1509
1510        for (pg = ftrace_pages_start; pg; pg = pg->next) {
1511                if (end < pg->records[0].ip ||
1512                    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1513                        continue;
1514                rec = bsearch(&key, pg->records, pg->index,
1515                              sizeof(struct dyn_ftrace),
1516                              ftrace_cmp_recs);
1517                if (rec)
1518                        return rec->ip;
1519        }
1520
1521        return 0;
1522}
1523
1524/**
1525 * ftrace_location - return true if the ip giving is a traced location
1526 * @ip: the instruction pointer to check
1527 *
1528 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1529 * That is, the instruction that is either a NOP or call to
1530 * the function tracer. It checks the ftrace internal tables to
1531 * determine if the address belongs or not.
1532 */
1533unsigned long ftrace_location(unsigned long ip)
1534{
1535        return ftrace_location_range(ip, ip);
1536}
1537
1538/**
1539 * ftrace_text_reserved - return true if range contains an ftrace location
1540 * @start: start of range to search
1541 * @end: end of range to search (inclusive). @end points to the last byte to check.
1542 *
1543 * Returns 1 if @start and @end contains a ftrace location.
1544 * That is, the instruction that is either a NOP or call to
1545 * the function tracer. It checks the ftrace internal tables to
1546 * determine if the address belongs or not.
1547 */
1548int ftrace_text_reserved(void *start, void *end)
1549{
1550        unsigned long ret;
1551
1552        ret = ftrace_location_range((unsigned long)start,
1553                                    (unsigned long)end);
1554
1555        return (int)!!ret;
1556}
1557
1558static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1559                                     int filter_hash,
1560                                     bool inc)
1561{
1562        struct ftrace_hash *hash;
1563        struct ftrace_hash *other_hash;
1564        struct ftrace_page *pg;
1565        struct dyn_ftrace *rec;
1566        int count = 0;
1567        int all = 0;
1568
1569        /* Only update if the ops has been registered */
1570        if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1571                return;
1572
1573        /*
1574         * In the filter_hash case:
1575         *   If the count is zero, we update all records.
1576         *   Otherwise we just update the items in the hash.
1577         *
1578         * In the notrace_hash case:
1579         *   We enable the update in the hash.
1580         *   As disabling notrace means enabling the tracing,
1581         *   and enabling notrace means disabling, the inc variable
1582         *   gets inversed.
1583         */
1584        if (filter_hash) {
1585                hash = ops->filter_hash;
1586                other_hash = ops->notrace_hash;
1587                if (ftrace_hash_empty(hash))
1588                        all = 1;
1589        } else {
1590                inc = !inc;
1591                hash = ops->notrace_hash;
1592                other_hash = ops->filter_hash;
1593                /*
1594                 * If the notrace hash has no items,
1595                 * then there's nothing to do.
1596                 */
1597                if (ftrace_hash_empty(hash))
1598                        return;
1599        }
1600
1601        do_for_each_ftrace_rec(pg, rec) {
1602                int in_other_hash = 0;
1603                int in_hash = 0;
1604                int match = 0;
1605
1606                if (all) {
1607                        /*
1608                         * Only the filter_hash affects all records.
1609                         * Update if the record is not in the notrace hash.
1610                         */
1611                        if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1612                                match = 1;
1613                } else {
1614                        in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1615                        in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1616
1617                        /*
1618                         *
1619                         */
1620                        if (filter_hash && in_hash && !in_other_hash)
1621                                match = 1;
1622                        else if (!filter_hash && in_hash &&
1623                                 (in_other_hash || ftrace_hash_empty(other_hash)))
1624                                match = 1;
1625                }
1626                if (!match)
1627                        continue;
1628
1629                if (inc) {
1630                        rec->flags++;
1631                        if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1632                                return;
1633                        /*
1634                         * If any ops wants regs saved for this function
1635                         * then all ops will get saved regs.
1636                         */
1637                        if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1638                                rec->flags |= FTRACE_FL_REGS;
1639                } else {
1640                        if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1641                                return;
1642                        rec->flags--;
1643                }
1644                count++;
1645                /* Shortcut, if we handled all records, we are done. */
1646                if (!all && count == hash->count)
1647                        return;
1648        } while_for_each_ftrace_rec();
1649}
1650
1651static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1652                                    int filter_hash)
1653{
1654        __ftrace_hash_rec_update(ops, filter_hash, 0);
1655}
1656
1657static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1658                                   int filter_hash)
1659{
1660        __ftrace_hash_rec_update(ops, filter_hash, 1);
1661}
1662
1663static void print_ip_ins(const char *fmt, unsigned char *p)
1664{
1665        int i;
1666
1667        printk(KERN_CONT "%s", fmt);
1668
1669        for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1670                printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1671}
1672
1673/**
1674 * ftrace_bug - report and shutdown function tracer
1675 * @failed: The failed type (EFAULT, EINVAL, EPERM)
1676 * @ip: The address that failed
1677 *
1678 * The arch code that enables or disables the function tracing
1679 * can call ftrace_bug() when it has detected a problem in
1680 * modifying the code. @failed should be one of either:
1681 * EFAULT - if the problem happens on reading the @ip address
1682 * EINVAL - if what is read at @ip is not what was expected
1683 * EPERM - if the problem happens on writting to the @ip address
1684 */
1685void ftrace_bug(int failed, unsigned long ip)
1686{
1687        switch (failed) {
1688        case -EFAULT:
1689                FTRACE_WARN_ON_ONCE(1);
1690                pr_info("ftrace faulted on modifying ");
1691                print_ip_sym(ip);
1692                break;
1693        case -EINVAL:
1694                FTRACE_WARN_ON_ONCE(1);
1695                pr_info("ftrace failed to modify ");
1696                print_ip_sym(ip);
1697                print_ip_ins(" actual: ", (unsigned char *)ip);
1698                printk(KERN_CONT "\n");
1699                break;
1700        case -EPERM:
1701                FTRACE_WARN_ON_ONCE(1);
1702                pr_info("ftrace faulted on writing ");
1703                print_ip_sym(ip);
1704                break;
1705        default:
1706                FTRACE_WARN_ON_ONCE(1);
1707                pr_info("ftrace faulted on unknown error ");
1708                print_ip_sym(ip);
1709        }
1710}
1711
1712static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1713{
1714        unsigned long flag = 0UL;
1715
1716        /*
1717         * If we are updating calls:
1718         *
1719         *   If the record has a ref count, then we need to enable it
1720         *   because someone is using it.
1721         *
1722         *   Otherwise we make sure its disabled.
1723         *
1724         * If we are disabling calls, then disable all records that
1725         * are enabled.
1726         */
1727        if (enable && (rec->flags & ~FTRACE_FL_MASK))
1728                flag = FTRACE_FL_ENABLED;
1729
1730        /*
1731         * If enabling and the REGS flag does not match the REGS_EN, then
1732         * do not ignore this record. Set flags to fail the compare against
1733         * ENABLED.
1734         */
1735        if (flag &&
1736            (!(rec->flags & FTRACE_FL_REGS) != !(rec->flags & FTRACE_FL_REGS_EN)))
1737                flag |= FTRACE_FL_REGS;
1738
1739        /* If the state of this record hasn't changed, then do nothing */
1740        if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1741                return FTRACE_UPDATE_IGNORE;
1742
1743        if (flag) {
1744                /* Save off if rec is being enabled (for return value) */
1745                flag ^= rec->flags & FTRACE_FL_ENABLED;
1746
1747                if (update) {
1748                        rec->flags |= FTRACE_FL_ENABLED;
1749                        if (flag & FTRACE_FL_REGS) {
1750                                if (rec->flags & FTRACE_FL_REGS)
1751                                        rec->flags |= FTRACE_FL_REGS_EN;
1752                                else
1753                                        rec->flags &= ~FTRACE_FL_REGS_EN;
1754                        }
1755                }
1756
1757                /*
1758                 * If this record is being updated from a nop, then
1759                 *   return UPDATE_MAKE_CALL.
1760                 * Otherwise, if the EN flag is set, then return
1761                 *   UPDATE_MODIFY_CALL_REGS to tell the caller to convert
1762                 *   from the non-save regs, to a save regs function.
1763                 * Otherwise,
1764                 *   return UPDATE_MODIFY_CALL to tell the caller to convert
1765                 *   from the save regs, to a non-save regs function.
1766                 */
1767                if (flag & FTRACE_FL_ENABLED)
1768                        return FTRACE_UPDATE_MAKE_CALL;
1769                else if (rec->flags & FTRACE_FL_REGS_EN)
1770                        return FTRACE_UPDATE_MODIFY_CALL_REGS;
1771                else
1772                        return FTRACE_UPDATE_MODIFY_CALL;
1773        }
1774
1775        if (update) {
1776                /* If there's no more users, clear all flags */
1777                if (!(rec->flags & ~FTRACE_FL_MASK))
1778                        rec->flags = 0;
1779                else
1780                        /* Just disable the record (keep REGS state) */
1781                        rec->flags &= ~FTRACE_FL_ENABLED;
1782        }
1783
1784        return FTRACE_UPDATE_MAKE_NOP;
1785}
1786
1787/**
1788 * ftrace_update_record, set a record that now is tracing or not
1789 * @rec: the record to update
1790 * @enable: set to 1 if the record is tracing, zero to force disable
1791 *
1792 * The records that represent all functions that can be traced need
1793 * to be updated when tracing has been enabled.
1794 */
1795int ftrace_update_record(struct dyn_ftrace *rec, int enable)
1796{
1797        return ftrace_check_record(rec, enable, 1);
1798}
1799
1800/**
1801 * ftrace_test_record, check if the record has been enabled or not
1802 * @rec: the record to test
1803 * @enable: set to 1 to check if enabled, 0 if it is disabled
1804 *
1805 * The arch code may need to test if a record is already set to
1806 * tracing to determine how to modify the function code that it
1807 * represents.
1808 */
1809int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1810{
1811        return ftrace_check_record(rec, enable, 0);
1812}
1813
1814static int
1815__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1816{
1817        unsigned long ftrace_old_addr;
1818        unsigned long ftrace_addr;
1819        int ret;
1820
1821        ret = ftrace_update_record(rec, enable);
1822
1823        if (rec->flags & FTRACE_FL_REGS)
1824                ftrace_addr = (unsigned long)FTRACE_REGS_ADDR;
1825        else
1826                ftrace_addr = (unsigned long)FTRACE_ADDR;
1827
1828        switch (ret) {
1829        case FTRACE_UPDATE_IGNORE:
1830                return 0;
1831
1832        case FTRACE_UPDATE_MAKE_CALL:
1833                return ftrace_make_call(rec, ftrace_addr);
1834
1835        case FTRACE_UPDATE_MAKE_NOP:
1836                return ftrace_make_nop(NULL, rec, ftrace_addr);
1837
1838        case FTRACE_UPDATE_MODIFY_CALL_REGS:
1839        case FTRACE_UPDATE_MODIFY_CALL:
1840                if (rec->flags & FTRACE_FL_REGS)
1841                        ftrace_old_addr = (unsigned long)FTRACE_ADDR;
1842                else
1843                        ftrace_old_addr = (unsigned long)FTRACE_REGS_ADDR;
1844
1845                return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
1846        }
1847
1848        return -1; /* unknow ftrace bug */
1849}
1850
1851void __weak ftrace_replace_code(int enable)
1852{
1853        struct dyn_ftrace *rec;
1854        struct ftrace_page *pg;
1855        int failed;
1856
1857        if (unlikely(ftrace_disabled))
1858                return;
1859
1860        do_for_each_ftrace_rec(pg, rec) {
1861                failed = __ftrace_replace_code(rec, enable);
1862                if (failed) {
1863                        ftrace_bug(failed, rec->ip);
1864                        /* Stop processing */
1865                        return;
1866                }
1867        } while_for_each_ftrace_rec();
1868}
1869
1870struct ftrace_rec_iter {
1871        struct ftrace_page      *pg;
1872        int                     index;
1873};
1874
1875/**
1876 * ftrace_rec_iter_start, start up iterating over traced functions
1877 *
1878 * Returns an iterator handle that is used to iterate over all
1879 * the records that represent address locations where functions
1880 * are traced.
1881 *
1882 * May return NULL if no records are available.
1883 */
1884struct ftrace_rec_iter *ftrace_rec_iter_start(void)
1885{
1886        /*
1887         * We only use a single iterator.
1888         * Protected by the ftrace_lock mutex.
1889         */
1890        static struct ftrace_rec_iter ftrace_rec_iter;
1891        struct ftrace_rec_iter *iter = &ftrace_rec_iter;
1892
1893        iter->pg = ftrace_pages_start;
1894        iter->index = 0;
1895
1896        /* Could have empty pages */
1897        while (iter->pg && !iter->pg->index)
1898                iter->pg = iter->pg->next;
1899
1900        if (!iter->pg)
1901                return NULL;
1902
1903        return iter;
1904}
1905
1906/**
1907 * ftrace_rec_iter_next, get the next record to process.
1908 * @iter: The handle to the iterator.
1909 *
1910 * Returns the next iterator after the given iterator @iter.
1911 */
1912struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
1913{
1914        iter->index++;
1915
1916        if (iter->index >= iter->pg->index) {
1917                iter->pg = iter->pg->next;
1918                iter->index = 0;
1919
1920                /* Could have empty pages */
1921                while (iter->pg && !iter->pg->index)
1922                        iter->pg = iter->pg->next;
1923        }
1924
1925        if (!iter->pg)
1926                return NULL;
1927
1928        return iter;
1929}
1930
1931/**
1932 * ftrace_rec_iter_record, get the record at the iterator location
1933 * @iter: The current iterator location
1934 *
1935 * Returns the record that the current @iter is at.
1936 */
1937struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
1938{
1939        return &iter->pg->records[iter->index];
1940}
1941
1942static int
1943ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1944{
1945        unsigned long ip;
1946        int ret;
1947
1948        ip = rec->ip;
1949
1950        if (unlikely(ftrace_disabled))
1951                return 0;
1952
1953        ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1954        if (ret) {
1955                ftrace_bug(ret, ip);
1956                return 0;
1957        }
1958        return 1;
1959}
1960
1961/*
1962 * archs can override this function if they must do something
1963 * before the modifying code is performed.
1964 */
1965int __weak ftrace_arch_code_modify_prepare(void)
1966{
1967        return 0;
1968}
1969
1970/*
1971 * archs can override this function if they must do something
1972 * after the modifying code is performed.
1973 */
1974int __weak ftrace_arch_code_modify_post_process(void)
1975{
1976        return 0;
1977}
1978
1979void ftrace_modify_all_code(int command)
1980{
1981        if (command & FTRACE_UPDATE_CALLS)
1982                ftrace_replace_code(1);
1983        else if (command & FTRACE_DISABLE_CALLS)
1984                ftrace_replace_code(0);
1985
1986        if (command & FTRACE_UPDATE_TRACE_FUNC)
1987                ftrace_update_ftrace_func(ftrace_trace_function);
1988
1989        if (command & FTRACE_START_FUNC_RET)
1990                ftrace_enable_ftrace_graph_caller();
1991        else if (command & FTRACE_STOP_FUNC_RET)
1992                ftrace_disable_ftrace_graph_caller();
1993}
1994
1995static int __ftrace_modify_code(void *data)
1996{
1997        int *command = data;
1998
1999        ftrace_modify_all_code(*command);
2000
2001        return 0;
2002}
2003
2004/**
2005 * ftrace_run_stop_machine, go back to the stop machine method
2006 * @command: The command to tell ftrace what to do
2007 *
2008 * If an arch needs to fall back to the stop machine method, the
2009 * it can call this function.
2010 */
2011void ftrace_run_stop_machine(int command)
2012{
2013        stop_machine(__ftrace_modify_code, &command, NULL);
2014}
2015
2016/**
2017 * arch_ftrace_update_code, modify the code to trace or not trace
2018 * @command: The command that needs to be done
2019 *
2020 * Archs can override this function if it does not need to
2021 * run stop_machine() to modify code.
2022 */
2023void __weak arch_ftrace_update_code(int command)
2024{
2025        ftrace_run_stop_machine(command);
2026}
2027
2028static void ftrace_run_update_code(int command)
2029{
2030        int ret;
2031
2032        ret = ftrace_arch_code_modify_prepare();
2033        FTRACE_WARN_ON(ret);
2034        if (ret)
2035                return;
2036        /*
2037         * Do not call function tracer while we update the code.
2038         * We are in stop machine.
2039         */
2040        function_trace_stop++;
2041
2042        /*
2043         * By default we use stop_machine() to modify the code.
2044         * But archs can do what ever they want as long as it
2045         * is safe. The stop_machine() is the safest, but also
2046         * produces the most overhead.
2047         */
2048        arch_ftrace_update_code(command);
2049
2050        function_trace_stop--;
2051
2052        ret = ftrace_arch_code_modify_post_process();
2053        FTRACE_WARN_ON(ret);
2054}
2055
2056static ftrace_func_t saved_ftrace_func;
2057static int ftrace_start_up;
2058static int global_start_up;
2059
2060static void ftrace_startup_enable(int command)
2061{
2062        if (saved_ftrace_func != ftrace_trace_function) {
2063                saved_ftrace_func = ftrace_trace_function;
2064                command |= FTRACE_UPDATE_TRACE_FUNC;
2065        }
2066
2067        if (!command || !ftrace_enabled)
2068                return;
2069
2070        ftrace_run_update_code(command);
2071}
2072
2073static int ftrace_startup(struct ftrace_ops *ops, int command)
2074{
2075        bool hash_enable = true;
2076
2077        if (unlikely(ftrace_disabled))
2078                return -ENODEV;
2079
2080        ftrace_start_up++;
2081        command |= FTRACE_UPDATE_CALLS;
2082
2083        /* ops marked global share the filter hashes */
2084        if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2085                ops = &global_ops;
2086                /* Don't update hash if global is already set */
2087                if (global_start_up)
2088                        hash_enable = false;
2089                global_start_up++;
2090        }
2091
2092        ops->flags |= FTRACE_OPS_FL_ENABLED;
2093        if (hash_enable)
2094                ftrace_hash_rec_enable(ops, 1);
2095
2096        ftrace_startup_enable(command);
2097
2098        return 0;
2099}
2100
2101static void ftrace_shutdown(struct ftrace_ops *ops, int command)
2102{
2103        bool hash_disable = true;
2104
2105        if (unlikely(ftrace_disabled))
2106                return;
2107
2108        ftrace_start_up--;
2109        /*
2110         * Just warn in case of unbalance, no need to kill ftrace, it's not
2111         * critical but the ftrace_call callers may be never nopped again after
2112         * further ftrace uses.
2113         */
2114        WARN_ON_ONCE(ftrace_start_up < 0);
2115
2116        if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2117                ops = &global_ops;
2118                global_start_up--;
2119                WARN_ON_ONCE(global_start_up < 0);
2120                /* Don't update hash if global still has users */
2121                if (global_start_up) {
2122                        WARN_ON_ONCE(!ftrace_start_up);
2123                        hash_disable = false;
2124                }
2125        }
2126
2127        if (hash_disable)
2128                ftrace_hash_rec_disable(ops, 1);
2129
2130        if (ops != &global_ops || !global_start_up)
2131                ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2132
2133        command |= FTRACE_UPDATE_CALLS;
2134
2135        if (saved_ftrace_func != ftrace_trace_function) {
2136                saved_ftrace_func = ftrace_trace_function;
2137                command |= FTRACE_UPDATE_TRACE_FUNC;
2138        }
2139
2140        if (!command || !ftrace_enabled)
2141                return;
2142
2143        ftrace_run_update_code(command);
2144}
2145
2146static void ftrace_startup_sysctl(void)
2147{
2148        if (unlikely(ftrace_disabled))
2149                return;
2150
2151        /* Force update next time */
2152        saved_ftrace_func = NULL;
2153        /* ftrace_start_up is true if we want ftrace running */
2154        if (ftrace_start_up)
2155                ftrace_run_update_code(FTRACE_UPDATE_CALLS);
2156}
2157
2158static void ftrace_shutdown_sysctl(void)
2159{
2160        if (unlikely(ftrace_disabled))
2161                return;
2162
2163        /* ftrace_start_up is true if ftrace is running */
2164        if (ftrace_start_up)
2165                ftrace_run_update_code(FTRACE_DISABLE_CALLS);
2166}
2167
2168static cycle_t          ftrace_update_time;
2169static unsigned long    ftrace_update_cnt;
2170unsigned long           ftrace_update_tot_cnt;
2171
2172static inline int ops_traces_mod(struct ftrace_ops *ops)
2173{
2174        /*
2175         * Filter_hash being empty will default to trace module.
2176         * But notrace hash requires a test of individual module functions.
2177         */
2178        return ftrace_hash_empty(ops->filter_hash) &&
2179                ftrace_hash_empty(ops->notrace_hash);
2180}
2181
2182/*
2183 * Check if the current ops references the record.
2184 *
2185 * If the ops traces all functions, then it was already accounted for.
2186 * If the ops does not trace the current record function, skip it.
2187 * If the ops ignores the function via notrace filter, skip it.
2188 */
2189static inline bool
2190ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2191{
2192        /* If ops isn't enabled, ignore it */
2193        if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2194                return 0;
2195
2196        /* If ops traces all mods, we already accounted for it */
2197        if (ops_traces_mod(ops))
2198                return 0;
2199
2200        /* The function must be in the filter */
2201        if (!ftrace_hash_empty(ops->filter_hash) &&
2202            !ftrace_lookup_ip(ops->filter_hash, rec->ip))
2203                return 0;
2204
2205        /* If in notrace hash, we ignore it too */
2206        if (ftrace_lookup_ip(ops->notrace_hash, rec->ip))
2207                return 0;
2208
2209        return 1;
2210}
2211
2212static int referenced_filters(struct dyn_ftrace *rec)
2213{
2214        struct ftrace_ops *ops;
2215        int cnt = 0;
2216
2217        for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
2218                if (ops_references_rec(ops, rec))
2219                    cnt++;
2220        }
2221
2222        return cnt;
2223}
2224
2225static int ftrace_update_code(struct module *mod)
2226{
2227        struct ftrace_page *pg;
2228        struct dyn_ftrace *p;
2229        cycle_t start, stop;
2230        unsigned long ref = 0;
2231        bool test = false;
2232        int i;
2233
2234        /*
2235         * When adding a module, we need to check if tracers are
2236         * currently enabled and if they are set to trace all functions.
2237         * If they are, we need to enable the module functions as well
2238         * as update the reference counts for those function records.
2239         */
2240        if (mod) {
2241                struct ftrace_ops *ops;
2242
2243                for (ops = ftrace_ops_list;
2244                     ops != &ftrace_list_end; ops = ops->next) {
2245                        if (ops->flags & FTRACE_OPS_FL_ENABLED) {
2246                                if (ops_traces_mod(ops))
2247                                        ref++;
2248                                else
2249                                        test = true;
2250                        }
2251                }
2252        }
2253
2254        start = ftrace_now(raw_smp_processor_id());
2255        ftrace_update_cnt = 0;
2256
2257        for (pg = ftrace_new_pgs; pg; pg = pg->next) {
2258
2259                for (i = 0; i < pg->index; i++) {
2260                        int cnt = ref;
2261
2262                        /* If something went wrong, bail without enabling anything */
2263                        if (unlikely(ftrace_disabled))
2264                                return -1;
2265
2266                        p = &pg->records[i];
2267                        if (test)
2268                                cnt += referenced_filters(p);
2269                        p->flags = cnt;
2270
2271                        /*
2272                         * Do the initial record conversion from mcount jump
2273                         * to the NOP instructions.
2274                         */
2275                        if (!ftrace_code_disable(mod, p))
2276                                break;
2277
2278                        ftrace_update_cnt++;
2279
2280                        /*
2281                         * If the tracing is enabled, go ahead and enable the record.
2282                         *
2283                         * The reason not to enable the record immediatelly is the
2284                         * inherent check of ftrace_make_nop/ftrace_make_call for
2285                         * correct previous instructions.  Making first the NOP
2286                         * conversion puts the module to the correct state, thus
2287                         * passing the ftrace_make_call check.
2288                         */
2289                        if (ftrace_start_up && cnt) {
2290                                int failed = __ftrace_replace_code(p, 1);
2291                                if (failed)
2292                                        ftrace_bug(failed, p->ip);
2293                        }
2294                }
2295        }
2296
2297        ftrace_new_pgs = NULL;
2298
2299        stop = ftrace_now(raw_smp_processor_id());
2300        ftrace_update_time = stop - start;
2301        ftrace_update_tot_cnt += ftrace_update_cnt;
2302
2303        return 0;
2304}
2305
2306static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2307{
2308        int order;
2309        int cnt;
2310
2311        if (WARN_ON(!count))
2312                return -EINVAL;
2313
2314        order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2315
2316        /*
2317         * We want to fill as much as possible. No more than a page
2318         * may be empty.
2319         */
2320        while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2321                order--;
2322
2323 again:
2324        pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2325
2326        if (!pg->records) {
2327                /* if we can't allocate this size, try something smaller */
2328                if (!order)
2329                        return -ENOMEM;
2330                order >>= 1;
2331                goto again;
2332        }
2333
2334        cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2335        pg->size = cnt;
2336
2337        if (cnt > count)
2338                cnt = count;
2339
2340        return cnt;
2341}
2342
2343static struct ftrace_page *
2344ftrace_allocate_pages(unsigned long num_to_init)
2345{
2346        struct ftrace_page *start_pg;
2347        struct ftrace_page *pg;
2348        int order;
2349        int cnt;
2350
2351        if (!num_to_init)
2352                return 0;
2353
2354        start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2355        if (!pg)
2356                return NULL;
2357
2358        /*
2359         * Try to allocate as much as possible in one continues
2360         * location that fills in all of the space. We want to
2361         * waste as little space as possible.
2362         */
2363        for (;;) {
2364                cnt = ftrace_allocate_records(pg, num_to_init);
2365                if (cnt < 0)
2366                        goto free_pages;
2367
2368                num_to_init -= cnt;
2369                if (!num_to_init)
2370                        break;
2371
2372                pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2373                if (!pg->next)
2374                        goto free_pages;
2375
2376                pg = pg->next;
2377        }
2378
2379        return start_pg;
2380
2381 free_pages:
2382        while (start_pg) {
2383                order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2384                free_pages((unsigned long)pg->records, order);
2385                start_pg = pg->next;
2386                kfree(pg);
2387                pg = start_pg;
2388        }
2389        pr_info("ftrace: FAILED to allocate memory for functions\n");
2390        return NULL;
2391}
2392
2393static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
2394{
2395        int cnt;
2396
2397        if (!num_to_init) {
2398                pr_info("ftrace: No functions to be traced?\n");
2399                return -1;
2400        }
2401
2402        cnt = num_to_init / ENTRIES_PER_PAGE;
2403        pr_info("ftrace: allocating %ld entries in %d pages\n",
2404                num_to_init, cnt + 1);
2405
2406        return 0;
2407}
2408
2409#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
2410
2411struct ftrace_iterator {
2412        loff_t                          pos;
2413        loff_t                          func_pos;
2414        struct ftrace_page              *pg;
2415        struct dyn_ftrace               *func;
2416        struct ftrace_func_probe        *probe;
2417        struct trace_parser             parser;
2418        struct ftrace_hash              *hash;
2419        struct ftrace_ops               *ops;
2420        int                             hidx;
2421        int                             idx;
2422        unsigned                        flags;
2423};
2424
2425static void *
2426t_hash_next(struct seq_file *m, loff_t *pos)
2427{
2428        struct ftrace_iterator *iter = m->private;
2429        struct hlist_node *hnd = NULL;
2430        struct hlist_head *hhd;
2431
2432        (*pos)++;
2433        iter->pos = *pos;
2434
2435        if (iter->probe)
2436                hnd = &iter->probe->node;
2437 retry:
2438        if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
2439                return NULL;
2440
2441        hhd = &ftrace_func_hash[iter->hidx];
2442
2443        if (hlist_empty(hhd)) {
2444                iter->hidx++;
2445                hnd = NULL;
2446                goto retry;
2447        }
2448
2449        if (!hnd)
2450                hnd = hhd->first;
2451        else {
2452                hnd = hnd->next;
2453                if (!hnd) {
2454                        iter->hidx++;
2455                        goto retry;
2456                }
2457        }
2458
2459        if (WARN_ON_ONCE(!hnd))
2460                return NULL;
2461
2462        iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
2463
2464        return iter;
2465}
2466
2467static void *t_hash_start(struct seq_file *m, loff_t *pos)
2468{
2469        struct ftrace_iterator *iter = m->private;
2470        void *p = NULL;
2471        loff_t l;
2472
2473        if (!(iter->flags & FTRACE_ITER_DO_HASH))
2474                return NULL;
2475
2476        if (iter->func_pos > *pos)
2477                return NULL;
2478
2479        iter->hidx = 0;
2480        for (l = 0; l <= (*pos - iter->func_pos); ) {
2481                p = t_hash_next(m, &l);
2482                if (!p)
2483                        break;
2484        }
2485        if (!p)
2486                return NULL;
2487
2488        /* Only set this if we have an item */
2489        iter->flags |= FTRACE_ITER_HASH;
2490
2491        return iter;
2492}
2493
2494static int
2495t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2496{
2497        struct ftrace_func_probe *rec;
2498
2499        rec = iter->probe;
2500        if (WARN_ON_ONCE(!rec))
2501                return -EIO;
2502
2503        if (rec->ops->print)
2504                return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2505
2506        seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2507
2508        if (rec->data)
2509                seq_printf(m, ":%p", rec->data);
2510        seq_putc(m, '\n');
2511
2512        return 0;
2513}
2514
2515static void *
2516t_next(struct seq_file *m, void *v, loff_t *pos)
2517{
2518        struct ftrace_iterator *iter = m->private;
2519        struct ftrace_ops *ops = iter->ops;
2520        struct dyn_ftrace *rec = NULL;
2521
2522        if (unlikely(ftrace_disabled))
2523                return NULL;
2524
2525        if (iter->flags & FTRACE_ITER_HASH)
2526                return t_hash_next(m, pos);
2527
2528        (*pos)++;
2529        iter->pos = iter->func_pos = *pos;
2530
2531        if (iter->flags & FTRACE_ITER_PRINTALL)
2532                return t_hash_start(m, pos);
2533
2534 retry:
2535        if (iter->idx >= iter->pg->index) {
2536                if (iter->pg->next) {
2537                        iter->pg = iter->pg->next;
2538                        iter->idx = 0;
2539                        goto retry;
2540                }
2541        } else {
2542                rec = &iter->pg->records[iter->idx++];
2543                if (((iter->flags & FTRACE_ITER_FILTER) &&
2544                     !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2545
2546                    ((iter->flags & FTRACE_ITER_NOTRACE) &&
2547                     !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2548
2549                    ((iter->flags & FTRACE_ITER_ENABLED) &&
2550                     !(rec->flags & FTRACE_FL_ENABLED))) {
2551
2552                        rec = NULL;
2553                        goto retry;
2554                }
2555        }
2556
2557        if (!rec)
2558                return t_hash_start(m, pos);
2559
2560        iter->func = rec;
2561
2562        return iter;
2563}
2564
2565static void reset_iter_read(struct ftrace_iterator *iter)
2566{
2567        iter->pos = 0;
2568        iter->func_pos = 0;
2569        iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
2570}
2571
2572static void *t_start(struct seq_file *m, loff_t *pos)
2573{
2574        struct ftrace_iterator *iter = m->private;
2575        struct ftrace_ops *ops = iter->ops;
2576        void *p = NULL;
2577        loff_t l;
2578
2579        mutex_lock(&ftrace_lock);
2580
2581        if (unlikely(ftrace_disabled))
2582                return NULL;
2583
2584        /*
2585         * If an lseek was done, then reset and start from beginning.
2586         */
2587        if (*pos < iter->pos)
2588                reset_iter_read(iter);
2589
2590        /*
2591         * For set_ftrace_filter reading, if we have the filter
2592         * off, we can short cut and just print out that all
2593         * functions are enabled.
2594         */
2595        if (iter->flags & FTRACE_ITER_FILTER &&
2596            ftrace_hash_empty(ops->filter_hash)) {
2597                if (*pos > 0)
2598                        return t_hash_start(m, pos);
2599                iter->flags |= FTRACE_ITER_PRINTALL;
2600                /* reset in case of seek/pread */
2601                iter->flags &= ~FTRACE_ITER_HASH;
2602                return iter;
2603        }
2604
2605        if (iter->flags & FTRACE_ITER_HASH)
2606                return t_hash_start(m, pos);
2607
2608        /*
2609         * Unfortunately, we need to restart at ftrace_pages_start
2610         * every time we let go of the ftrace_mutex. This is because
2611         * those pointers can change without the lock.
2612         */
2613        iter->pg = ftrace_pages_start;
2614        iter->idx = 0;
2615        for (l = 0; l <= *pos; ) {
2616                p = t_next(m, p, &l);
2617                if (!p)
2618                        break;
2619        }
2620
2621        if (!p)
2622                return t_hash_start(m, pos);
2623
2624        return iter;
2625}
2626
2627static void t_stop(struct seq_file *m, void *p)
2628{
2629        mutex_unlock(&ftrace_lock);
2630}
2631
2632static int t_show(struct seq_file *m, void *v)
2633{
2634        struct ftrace_iterator *iter = m->private;
2635        struct dyn_ftrace *rec;
2636
2637        if (iter->flags & FTRACE_ITER_HASH)
2638                return t_hash_show(m, iter);
2639
2640        if (iter->flags & FTRACE_ITER_PRINTALL) {
2641                seq_printf(m, "#### all functions enabled ####\n");
2642                return 0;
2643        }
2644
2645        rec = iter->func;
2646
2647        if (!rec)
2648                return 0;
2649
2650        seq_printf(m, "%ps", (void *)rec->ip);
2651        if (iter->flags & FTRACE_ITER_ENABLED)
2652                seq_printf(m, " (%ld)%s",
2653                           rec->flags & ~FTRACE_FL_MASK,
2654                           rec->flags & FTRACE_FL_REGS ? " R" : "");
2655        seq_printf(m, "\n");
2656
2657        return 0;
2658}
2659
2660static const struct seq_operations show_ftrace_seq_ops = {
2661        .start = t_start,
2662        .next = t_next,
2663        .stop = t_stop,
2664        .show = t_show,
2665};
2666
2667static int
2668ftrace_avail_open(struct inode *inode, struct file *file)
2669{
2670        struct ftrace_iterator *iter;
2671
2672        if (unlikely(ftrace_disabled))
2673                return -ENODEV;
2674
2675        iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2676        if (iter) {
2677                iter->pg = ftrace_pages_start;
2678                iter->ops = &global_ops;
2679        }
2680
2681        return iter ? 0 : -ENOMEM;
2682}
2683
2684static int
2685ftrace_enabled_open(struct inode *inode, struct file *file)
2686{
2687        struct ftrace_iterator *iter;
2688
2689        if (unlikely(ftrace_disabled))
2690                return -ENODEV;
2691
2692        iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2693        if (iter) {
2694                iter->pg = ftrace_pages_start;
2695                iter->flags = FTRACE_ITER_ENABLED;
2696                iter->ops = &global_ops;
2697        }
2698
2699        return iter ? 0 : -ENOMEM;
2700}
2701
2702static void ftrace_filter_reset(struct ftrace_hash *hash)
2703{
2704        mutex_lock(&ftrace_lock);
2705        ftrace_hash_clear(hash);
2706        mutex_unlock(&ftrace_lock);
2707}
2708
2709/**
2710 * ftrace_regex_open - initialize function tracer filter files
2711 * @ops: The ftrace_ops that hold the hash filters
2712 * @flag: The type of filter to process
2713 * @inode: The inode, usually passed in to your open routine
2714 * @file: The file, usually passed in to your open routine
2715 *
2716 * ftrace_regex_open() initializes the filter files for the
2717 * @ops. Depending on @flag it may process the filter hash or
2718 * the notrace hash of @ops. With this called from the open
2719 * routine, you can use ftrace_filter_write() for the write
2720 * routine if @flag has FTRACE_ITER_FILTER set, or
2721 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
2722 * ftrace_filter_lseek() should be used as the lseek routine, and
2723 * release must call ftrace_regex_release().
2724 */
2725int
2726ftrace_regex_open(struct ftrace_ops *ops, int flag,
2727                  struct inode *inode, struct file *file)
2728{
2729        struct ftrace_iterator *iter;
2730        struct ftrace_hash *hash;
2731        int ret = 0;
2732
2733        ftrace_ops_init(ops);
2734
2735        if (unlikely(ftrace_disabled))
2736                return -ENODEV;
2737
2738        iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2739        if (!iter)
2740                return -ENOMEM;
2741
2742        if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2743                kfree(iter);
2744                return -ENOMEM;
2745        }
2746
2747        iter->ops = ops;
2748        iter->flags = flag;
2749
2750        mutex_lock(&ops->regex_lock);
2751
2752        if (flag & FTRACE_ITER_NOTRACE)
2753                hash = ops->notrace_hash;
2754        else
2755                hash = ops->filter_hash;
2756
2757        if (file->f_mode & FMODE_WRITE) {
2758                iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2759                if (!iter->hash) {
2760                        trace_parser_put(&iter->parser);
2761                        kfree(iter);
2762                        ret = -ENOMEM;
2763                        goto out_unlock;
2764                }
2765        }
2766
2767        if ((file->f_mode & FMODE_WRITE) &&
2768            (file->f_flags & O_TRUNC))
2769                ftrace_filter_reset(iter->hash);
2770
2771        if (file->f_mode & FMODE_READ) {
2772                iter->pg = ftrace_pages_start;
2773
2774                ret = seq_open(file, &show_ftrace_seq_ops);
2775                if (!ret) {
2776                        struct seq_file *m = file->private_data;
2777                        m->private = iter;
2778                } else {
2779                        /* Failed */
2780                        free_ftrace_hash(iter->hash);
2781                        trace_parser_put(&iter->parser);
2782                        kfree(iter);
2783                }
2784        } else
2785                file->private_data = iter;
2786
2787 out_unlock:
2788        mutex_unlock(&ops->regex_lock);
2789
2790        return ret;
2791}
2792
2793static int
2794ftrace_filter_open(struct inode *inode, struct file *file)
2795{
2796        return ftrace_regex_open(&global_ops,
2797                        FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
2798                        inode, file);
2799}
2800
2801static int
2802ftrace_notrace_open(struct inode *inode, struct file *file)
2803{
2804        return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2805                                 inode, file);
2806}
2807
2808static int ftrace_match(char *str, char *regex, int len, int type)
2809{
2810        int matched = 0;
2811        int slen;
2812
2813        switch (type) {
2814        case MATCH_FULL:
2815                if (strcmp(str, regex) == 0)
2816                        matched = 1;
2817                break;
2818        case MATCH_FRONT_ONLY:
2819                if (strncmp(str, regex, len) == 0)
2820                        matched = 1;
2821                break;
2822        case MATCH_MIDDLE_ONLY:
2823                if (strstr(str, regex))
2824                        matched = 1;
2825                break;
2826        case MATCH_END_ONLY:
2827                slen = strlen(str);
2828                if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2829                        matched = 1;
2830                break;
2831        }
2832
2833        return matched;
2834}
2835
2836static int
2837enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2838{
2839        struct ftrace_func_entry *entry;
2840        int ret = 0;
2841
2842        entry = ftrace_lookup_ip(hash, rec->ip);
2843        if (not) {
2844                /* Do nothing if it doesn't exist */
2845                if (!entry)
2846                        return 0;
2847
2848                free_hash_entry(hash, entry);
2849        } else {
2850                /* Do nothing if it exists */
2851                if (entry)
2852                        return 0;
2853
2854                ret = add_hash_entry(hash, rec->ip);
2855        }
2856        return ret;
2857}
2858
2859static int
2860ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2861                    char *regex, int len, int type)
2862{
2863        char str[KSYM_SYMBOL_LEN];
2864        char *modname;
2865
2866        kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2867
2868        if (mod) {
2869                /* module lookup requires matching the module */
2870                if (!modname || strcmp(modname, mod))
2871                        return 0;
2872
2873                /* blank search means to match all funcs in the mod */
2874                if (!len)
2875                        return 1;
2876        }
2877
2878        return ftrace_match(str, regex, len, type);
2879}
2880
2881static int
2882match_records(struct ftrace_hash *hash, char *buff,
2883              int len, char *mod, int not)
2884{
2885        unsigned search_len = 0;
2886        struct ftrace_page *pg;
2887        struct dyn_ftrace *rec;
2888        int type = MATCH_FULL;
2889        char *search = buff;
2890        int found = 0;
2891        int ret;
2892
2893        if (len) {
2894                type = filter_parse_regex(buff, len, &search, &not);
2895                search_len = strlen(search);
2896        }
2897
2898        mutex_lock(&ftrace_lock);
2899
2900        if (unlikely(ftrace_disabled))
2901                goto out_unlock;
2902
2903        do_for_each_ftrace_rec(pg, rec) {
2904                if (ftrace_match_record(rec, mod, search, search_len, type)) {
2905                        ret = enter_record(hash, rec, not);
2906                        if (ret < 0) {
2907                                found = ret;
2908                                goto out_unlock;
2909                        }
2910                        found = 1;
2911                }
2912        } while_for_each_ftrace_rec();
2913 out_unlock:
2914        mutex_unlock(&ftrace_lock);
2915
2916        return found;
2917}
2918
2919static int
2920ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2921{
2922        return match_records(hash, buff, len, NULL, 0);
2923}
2924
2925static int
2926ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2927{
2928        int not = 0;
2929
2930        /* blank or '*' mean the same */
2931        if (strcmp(buff, "*") == 0)
2932                buff[0] = 0;
2933
2934        /* handle the case of 'dont filter this module' */
2935        if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2936                buff[0] = 0;
2937                not = 1;
2938        }
2939
2940        return match_records(hash, buff, strlen(buff), mod, not);
2941}
2942
2943/*
2944 * We register the module command as a template to show others how
2945 * to register the a command as well.
2946 */
2947
2948static int
2949ftrace_mod_callback(struct ftrace_hash *hash,
2950                    char *func, char *cmd, char *param, int enable)
2951{
2952        char *mod;
2953        int ret = -EINVAL;
2954
2955        /*
2956         * cmd == 'mod' because we only registered this func
2957         * for the 'mod' ftrace_func_command.
2958         * But if you register one func with multiple commands,
2959         * you can tell which command was used by the cmd
2960         * parameter.
2961         */
2962
2963        /* we must have a module name */
2964        if (!param)
2965                return ret;
2966
2967        mod = strsep(&param, ":");
2968        if (!strlen(mod))
2969                return ret;
2970
2971        ret = ftrace_match_module_records(hash, func, mod);
2972        if (!ret)
2973                ret = -EINVAL;
2974        if (ret < 0)
2975                return ret;
2976
2977        return 0;
2978}
2979
2980static struct ftrace_func_command ftrace_mod_cmd = {
2981        .name                   = "mod",
2982        .func                   = ftrace_mod_callback,
2983};
2984
2985static int __init ftrace_mod_cmd_init(void)
2986{
2987        return register_ftrace_command(&ftrace_mod_cmd);
2988}
2989core_initcall(ftrace_mod_cmd_init);
2990
2991static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
2992                                      struct ftrace_ops *op, struct pt_regs *pt_regs)
2993{
2994        struct ftrace_func_probe *entry;
2995        struct hlist_head *hhd;
2996        unsigned long key;
2997
2998        key = hash_long(ip, FTRACE_HASH_BITS);
2999
3000        hhd = &ftrace_func_hash[key];
3001
3002        if (hlist_empty(hhd))
3003                return;
3004
3005        /*
3006         * Disable preemption for these calls to prevent a RCU grace
3007         * period. This syncs the hash iteration and freeing of items
3008         * on the hash. rcu_read_lock is too dangerous here.
3009         */
3010        preempt_disable_notrace();
3011        hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3012                if (entry->ip == ip)
3013                        entry->ops->func(ip, parent_ip, &entry->data);
3014        }
3015        preempt_enable_notrace();
3016}
3017
3018static struct ftrace_ops trace_probe_ops __read_mostly =
3019{
3020        .func           = function_trace_probe_call,
3021        .flags          = FTRACE_OPS_FL_INITIALIZED,
3022        INIT_REGEX_LOCK(trace_probe_ops)
3023};
3024
3025static int ftrace_probe_registered;
3026
3027static void __enable_ftrace_function_probe(void)
3028{
3029        int ret;
3030        int i;
3031
3032        if (ftrace_probe_registered) {
3033                /* still need to update the function call sites */
3034                if (ftrace_enabled)
3035                        ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3036                return;
3037        }
3038
3039        for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3040                struct hlist_head *hhd = &ftrace_func_hash[i];
3041                if (hhd->first)
3042                        break;
3043        }
3044        /* Nothing registered? */
3045        if (i == FTRACE_FUNC_HASHSIZE)
3046                return;
3047
3048        ret = __register_ftrace_function(&trace_probe_ops);
3049        if (!ret)
3050                ret = ftrace_startup(&trace_probe_ops, 0);
3051
3052        ftrace_probe_registered = 1;
3053}
3054
3055static void __disable_ftrace_function_probe(void)
3056{
3057        int ret;
3058        int i;
3059
3060        if (!ftrace_probe_registered)
3061                return;
3062
3063        for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3064                struct hlist_head *hhd = &ftrace_func_hash[i];
3065                if (hhd->first)
3066                        return;
3067        }
3068
3069        /* no more funcs left */
3070        ret = __unregister_ftrace_function(&trace_probe_ops);
3071        if (!ret)
3072                ftrace_shutdown(&trace_probe_ops, 0);
3073
3074        ftrace_probe_registered = 0;
3075}
3076
3077
3078static void ftrace_free_entry(struct ftrace_func_probe *entry)
3079{
3080        if (entry->ops->free)
3081                entry->ops->free(entry->ops, entry->ip, &entry->data);
3082        kfree(entry);
3083}
3084
3085int
3086register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3087                              void *data)
3088{
3089        struct ftrace_func_probe *entry;
3090        struct ftrace_hash **orig_hash = &trace_probe_ops.filter_hash;
3091        struct ftrace_hash *hash;
3092        struct ftrace_page *pg;
3093        struct dyn_ftrace *rec;
3094        int type, len, not;
3095        unsigned long key;
3096        int count = 0;
3097        char *search;
3098        int ret;
3099
3100        type = filter_parse_regex(glob, strlen(glob), &search, &not);
3101        len = strlen(search);
3102
3103        /* we do not support '!' for function probes */
3104        if (WARN_ON(not))
3105                return -EINVAL;
3106
3107        mutex_lock(&trace_probe_ops.regex_lock);
3108
3109        hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3110        if (!hash) {
3111                count = -ENOMEM;
3112                goto out;
3113        }
3114
3115        if (unlikely(ftrace_disabled)) {
3116                count = -ENODEV;
3117                goto out;
3118        }
3119
3120        mutex_lock(&ftrace_lock);
3121
3122        do_for_each_ftrace_rec(pg, rec) {
3123
3124                if (!ftrace_match_record(rec, NULL, search, len, type))
3125                        continue;
3126
3127                entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3128                if (!entry) {
3129                        /* If we did not process any, then return error */
3130                        if (!count)
3131                                count = -ENOMEM;
3132                        goto out_unlock;
3133                }
3134
3135                count++;
3136
3137                entry->data = data;
3138
3139                /*
3140                 * The caller might want to do something special
3141                 * for each function we find. We call the callback
3142                 * to give the caller an opportunity to do so.
3143                 */
3144                if (ops->init) {
3145                        if (ops->init(ops, rec->ip, &entry->data) < 0) {
3146                                /* caller does not like this func */
3147                                kfree(entry);
3148                                continue;
3149                        }
3150                }
3151
3152                ret = enter_record(hash, rec, 0);
3153                if (ret < 0) {
3154                        kfree(entry);
3155                        count = ret;
3156                        goto out_unlock;
3157                }
3158
3159                entry->ops = ops;
3160                entry->ip = rec->ip;
3161
3162                key = hash_long(entry->ip, FTRACE_HASH_BITS);
3163                hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3164
3165        } while_for_each_ftrace_rec();
3166
3167        ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3168        if (ret < 0)
3169                count = ret;
3170
3171        __enable_ftrace_function_probe();
3172
3173 out_unlock:
3174        mutex_unlock(&ftrace_lock);
3175 out:
3176        mutex_unlock(&trace_probe_ops.regex_lock);
3177        free_ftrace_hash(hash);
3178
3179        return count;
3180}
3181
3182enum {
3183        PROBE_TEST_FUNC         = 1,
3184        PROBE_TEST_DATA         = 2
3185};
3186
3187static void
3188__unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3189                                  void *data, int flags)
3190{
3191        struct ftrace_func_entry *rec_entry;
3192        struct ftrace_func_probe *entry;
3193        struct ftrace_func_probe *p;
3194        struct ftrace_hash **orig_hash = &trace_probe_ops.filter_hash;
3195        struct list_head free_list;
3196        struct ftrace_hash *hash;
3197        struct hlist_node *tmp;
3198        char str[KSYM_SYMBOL_LEN];
3199        int type = MATCH_FULL;
3200        int i, len = 0;
3201        char *search;
3202
3203        if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3204                glob = NULL;
3205        else if (glob) {
3206                int not;
3207
3208                type = filter_parse_regex(glob, strlen(glob), &search, &not);
3209                len = strlen(search);
3210
3211                /* we do not support '!' for function probes */
3212                if (WARN_ON(not))
3213                        return;
3214        }
3215
3216        mutex_lock(&trace_probe_ops.regex_lock);
3217
3218        hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3219        if (!hash)
3220                /* Hmm, should report this somehow */
3221                goto out_unlock;
3222
3223        INIT_LIST_HEAD(&free_list);
3224
3225        for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3226                struct hlist_head *hhd = &ftrace_func_hash[i];
3227
3228                hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3229
3230                        /* break up if statements for readability */
3231                        if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3232                                continue;
3233
3234                        if ((flags & PROBE_TEST_DATA) && entry->data != data)
3235                                continue;
3236
3237                        /* do this last, since it is the most expensive */
3238                        if (glob) {
3239                                kallsyms_lookup(entry->ip, NULL, NULL,
3240                                                NULL, str);
3241                                if (!ftrace_match(str, glob, len, type))
3242                                        continue;
3243                        }
3244
3245                        rec_entry = ftrace_lookup_ip(hash, entry->ip);
3246                        /* It is possible more than one entry had this ip */
3247                        if (rec_entry)
3248                                free_hash_entry(hash, rec_entry);
3249
3250                        hlist_del_rcu(&entry->node);
3251                        list_add(&entry->free_list, &free_list);
3252                }
3253        }
3254        mutex_lock(&ftrace_lock);
3255        __disable_ftrace_function_probe();
3256        /*
3257         * Remove after the disable is called. Otherwise, if the last
3258         * probe is removed, a null hash means *all enabled*.
3259         */
3260        ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3261        synchronize_sched();
3262        list_for_each_entry_safe(entry, p, &free_list, free_list) {
3263                list_del(&entry->free_list);
3264                ftrace_free_entry(entry);
3265        }
3266        mutex_unlock(&ftrace_lock);
3267                
3268 out_unlock:
3269        mutex_unlock(&trace_probe_ops.regex_lock);
3270        free_ftrace_hash(hash);
3271}
3272
3273void
3274unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3275                                void *data)
3276{
3277        __unregister_ftrace_function_probe(glob, ops, data,
3278                                          PROBE_TEST_FUNC | PROBE_TEST_DATA);
3279}
3280
3281void
3282unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3283{
3284        __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3285}
3286
3287void unregister_ftrace_function_probe_all(char *glob)
3288{
3289        __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3290}
3291
3292static LIST_HEAD(ftrace_commands);
3293static DEFINE_MUTEX(ftrace_cmd_mutex);
3294
3295int register_ftrace_command(struct ftrace_func_command *cmd)
3296{
3297        struct ftrace_func_command *p;
3298        int ret = 0;
3299
3300        mutex_lock(&ftrace_cmd_mutex);
3301        list_for_each_entry(p, &ftrace_commands, list) {
3302                if (strcmp(cmd->name, p->name) == 0) {
3303                        ret = -EBUSY;
3304                        goto out_unlock;
3305                }
3306        }
3307        list_add(&cmd->list, &ftrace_commands);
3308 out_unlock:
3309        mutex_unlock(&ftrace_cmd_mutex);
3310
3311        return ret;
3312}
3313
3314int unregister_ftrace_command(struct ftrace_func_command *cmd)
3315{
3316        struct ftrace_func_command *p, *n;
3317        int ret = -ENODEV;
3318
3319        mutex_lock(&ftrace_cmd_mutex);
3320        list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3321                if (strcmp(cmd->name, p->name) == 0) {
3322                        ret = 0;
3323                        list_del_init(&p->list);
3324                        goto out_unlock;
3325                }
3326        }
3327 out_unlock:
3328        mutex_unlock(&ftrace_cmd_mutex);
3329
3330        return ret;
3331}
3332
3333static int ftrace_process_regex(struct ftrace_hash *hash,
3334                                char *buff, int len, int enable)
3335{
3336        char *func, *command, *next = buff;
3337        struct ftrace_func_command *p;
3338        int ret = -EINVAL;
3339
3340        func = strsep(&next, ":");
3341
3342        if (!next) {
3343                ret = ftrace_match_records(hash, func, len);
3344                if (!ret)
3345                        ret = -EINVAL;
3346                if (ret < 0)
3347                        return ret;
3348                return 0;
3349        }
3350
3351        /* command found */
3352
3353        command = strsep(&next, ":");
3354
3355        mutex_lock(&ftrace_cmd_mutex);
3356        list_for_each_entry(p, &ftrace_commands, list) {
3357                if (strcmp(p->name, command) == 0) {
3358                        ret = p->func(hash, func, command, next, enable);
3359                        goto out_unlock;
3360                }
3361        }
3362 out_unlock:
3363        mutex_unlock(&ftrace_cmd_mutex);
3364
3365        return ret;
3366}
3367
3368static ssize_t
3369ftrace_regex_write(struct file *file, const char __user *ubuf,
3370                   size_t cnt, loff_t *ppos, int enable)
3371{
3372        struct ftrace_iterator *iter;
3373        struct trace_parser *parser;
3374        ssize_t ret, read;
3375
3376        if (!cnt)
3377                return 0;
3378
3379        if (file->f_mode & FMODE_READ) {
3380                struct seq_file *m = file->private_data;
3381                iter = m->private;
3382        } else
3383                iter = file->private_data;
3384
3385        if (unlikely(ftrace_disabled))
3386                return -ENODEV;
3387
3388        /* iter->hash is a local copy, so we don't need regex_lock */
3389
3390        parser = &iter->parser;
3391        read = trace_get_user(parser, ubuf, cnt, ppos);
3392
3393        if (read >= 0 && trace_parser_loaded(parser) &&
3394            !trace_parser_cont(parser)) {
3395                ret = ftrace_process_regex(iter->hash, parser->buffer,
3396                                           parser->idx, enable);
3397                trace_parser_clear(parser);
3398                if (ret < 0)
3399                        goto out;
3400        }
3401
3402        ret = read;
3403 out:
3404        return ret;
3405}
3406
3407ssize_t
3408ftrace_filter_write(struct file *file, const char __user *ubuf,
3409                    size_t cnt, loff_t *ppos)
3410{
3411        return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
3412}
3413
3414ssize_t
3415ftrace_notrace_write(struct file *file, const char __user *ubuf,
3416                     size_t cnt, loff_t *ppos)
3417{
3418        return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
3419}
3420
3421static int
3422ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
3423{
3424        struct ftrace_func_entry *entry;
3425
3426        if (!ftrace_location(ip))
3427                return -EINVAL;
3428
3429        if (remove) {
3430                entry = ftrace_lookup_ip(hash, ip);
3431                if (!entry)
3432                        return -ENOENT;
3433                free_hash_entry(hash, entry);
3434                return 0;
3435        }
3436
3437        return add_hash_entry(hash, ip);
3438}
3439
3440static void ftrace_ops_update_code(struct ftrace_ops *ops)
3441{
3442        if (ops->flags & FTRACE_OPS_FL_ENABLED && ftrace_enabled)
3443                ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3444}
3445
3446static int
3447ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
3448                unsigned long ip, int remove, int reset, int enable)
3449{
3450        struct ftrace_hash **orig_hash;
3451        struct ftrace_hash *hash;
3452        int ret;
3453
3454        /* All global ops uses the global ops filters */
3455        if (ops->flags & FTRACE_OPS_FL_GLOBAL)
3456                ops = &global_ops;
3457
3458        if (unlikely(ftrace_disabled))
3459                return -ENODEV;
3460
3461        mutex_lock(&ops->regex_lock);
3462
3463        if (enable)
3464                orig_hash = &ops->filter_hash;
3465        else
3466                orig_hash = &ops->notrace_hash;
3467
3468        hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3469        if (!hash) {
3470                ret = -ENOMEM;
3471                goto out_regex_unlock;
3472        }
3473
3474        if (reset)
3475                ftrace_filter_reset(hash);
3476        if (buf && !ftrace_match_records(hash, buf, len)) {
3477                ret = -EINVAL;
3478                goto out_regex_unlock;
3479        }
3480        if (ip) {
3481                ret = ftrace_match_addr(hash, ip, remove);
3482                if (ret < 0)
3483                        goto out_regex_unlock;
3484        }
3485
3486        mutex_lock(&ftrace_lock);
3487        ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3488        if (!ret)
3489                ftrace_ops_update_code(ops);
3490
3491        mutex_unlock(&ftrace_lock);
3492
3493 out_regex_unlock:
3494        mutex_unlock(&ops->regex_lock);
3495
3496        free_ftrace_hash(hash);
3497        return ret;
3498}
3499
3500static int
3501ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
3502                int reset, int enable)
3503{
3504        return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
3505}
3506
3507/**
3508 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
3509 * @ops - the ops to set the filter with
3510 * @ip - the address to add to or remove from the filter.
3511 * @remove - non zero to remove the ip from the filter
3512 * @reset - non zero to reset all filters before applying this filter.
3513 *
3514 * Filters denote which functions should be enabled when tracing is enabled
3515 * If @ip is NULL, it failes to update filter.
3516 */
3517int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
3518                         int remove, int reset)
3519{
3520        ftrace_ops_init(ops);
3521        return ftrace_set_addr(ops, ip, remove, reset, 1);
3522}
3523EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
3524
3525static int
3526ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3527                 int reset, int enable)
3528{
3529        return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
3530}
3531
3532/**
3533 * ftrace_set_filter - set a function to filter on in ftrace
3534 * @ops - the ops to set the filter with
3535 * @buf - the string that holds the function filter text.
3536 * @len - the length of the string.
3537 * @reset - non zero to reset all filters before applying this filter.
3538 *
3539 * Filters denote which functions should be enabled when tracing is enabled.
3540 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3541 */
3542int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
3543                       int len, int reset)
3544{
3545        ftrace_ops_init(ops);
3546        return ftrace_set_regex(ops, buf, len, reset, 1);
3547}
3548EXPORT_SYMBOL_GPL(ftrace_set_filter);
3549
3550/**
3551 * ftrace_set_notrace - set a function to not trace in ftrace
3552 * @ops - the ops to set the notrace filter with
3553 * @buf - the string that holds the function notrace text.
3554 * @len - the length of the string.
3555 * @reset - non zero to reset all filters before applying this filter.
3556 *
3557 * Notrace Filters denote which functions should not be enabled when tracing
3558 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3559 * for tracing.
3560 */
3561int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
3562                        int len, int reset)
3563{
3564        ftrace_ops_init(ops);
3565        return ftrace_set_regex(ops, buf, len, reset, 0);
3566}
3567EXPORT_SYMBOL_GPL(ftrace_set_notrace);
3568/**
3569 * ftrace_set_filter - set a function to filter on in ftrace
3570 * @ops - the ops to set the filter with
3571 * @buf - the string that holds the function filter text.
3572 * @len - the length of the string.
3573 * @reset - non zero to reset all filters before applying this filter.
3574 *
3575 * Filters denote which functions should be enabled when tracing is enabled.
3576 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3577 */
3578void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
3579{
3580        ftrace_set_regex(&global_ops, buf, len, reset, 1);
3581}
3582EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
3583
3584/**
3585 * ftrace_set_notrace - set a function to not trace in ftrace
3586 * @ops - the ops to set the notrace filter with
3587 * @buf - the string that holds the function notrace text.
3588 * @len - the length of the string.
3589 * @reset - non zero to reset all filters before applying this filter.
3590 *
3591 * Notrace Filters denote which functions should not be enabled when tracing
3592 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3593 * for tracing.
3594 */
3595void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
3596{
3597        ftrace_set_regex(&global_ops, buf, len, reset, 0);
3598}
3599EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3600
3601/*
3602 * command line interface to allow users to set filters on boot up.
3603 */
3604#define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
3605static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3606static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3607
3608/* Used by function selftest to not test if filter is set */
3609bool ftrace_filter_param __initdata;
3610
3611static int __init set_ftrace_notrace(char *str)
3612{
3613        ftrace_filter_param = true;
3614        strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3615        return 1;
3616}
3617__setup("ftrace_notrace=", set_ftrace_notrace);
3618
3619static int __init set_ftrace_filter(char *str)
3620{
3621        ftrace_filter_param = true;
3622        strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3623        return 1;
3624}
3625__setup("ftrace_filter=", set_ftrace_filter);
3626
3627#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3628static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3629static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
3630
3631static int __init set_graph_function(char *str)
3632{
3633        strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3634        return 1;
3635}
3636__setup("ftrace_graph_filter=", set_graph_function);
3637
3638static void __init set_ftrace_early_graph(char *buf)
3639{
3640        int ret;
3641        char *func;
3642
3643        while (buf) {
3644                func = strsep(&buf, ",");
3645                /* we allow only one expression at a time */
3646                ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3647                                      func);
3648                if (ret)
3649                        printk(KERN_DEBUG "ftrace: function %s not "
3650                                          "traceable\n", func);
3651        }
3652}
3653#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3654
3655void __init
3656ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3657{
3658        char *func;
3659
3660        ftrace_ops_init(ops);
3661
3662        while (buf) {
3663                func = strsep(&buf, ",");
3664                ftrace_set_regex(ops, func, strlen(func), 0, enable);
3665        }
3666}
3667
3668static void __init set_ftrace_early_filters(void)
3669{
3670        if (ftrace_filter_buf[0])
3671                ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
3672        if (ftrace_notrace_buf[0])
3673                ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
3674#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3675        if (ftrace_graph_buf[0])
3676                set_ftrace_early_graph(ftrace_graph_buf);
3677#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3678}
3679
3680int ftrace_regex_release(struct inode *inode, struct file *file)
3681{
3682        struct seq_file *m = (struct seq_file *)file->private_data;
3683        struct ftrace_iterator *iter;
3684        struct ftrace_hash **orig_hash;
3685        struct trace_parser *parser;
3686        int filter_hash;
3687        int ret;
3688
3689        if (file->f_mode & FMODE_READ) {
3690                iter = m->private;
3691                seq_release(inode, file);
3692        } else
3693                iter = file->private_data;
3694
3695        parser = &iter->parser;
3696        if (trace_parser_loaded(parser)) {
3697                parser->buffer[parser->idx] = 0;
3698                ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3699        }
3700
3701        trace_parser_put(parser);
3702
3703        mutex_lock(&iter->ops->regex_lock);
3704
3705        if (file->f_mode & FMODE_WRITE) {
3706                filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3707
3708                if (filter_hash)
3709                        orig_hash = &iter->ops->filter_hash;
3710                else
3711                        orig_hash = &iter->ops->notrace_hash;
3712
3713                mutex_lock(&ftrace_lock);
3714                ret = ftrace_hash_move(iter->ops, filter_hash,
3715                                       orig_hash, iter->hash);
3716                if (!ret)
3717                        ftrace_ops_update_code(iter->ops);
3718
3719                mutex_unlock(&ftrace_lock);
3720        }
3721
3722        mutex_unlock(&iter->ops->regex_lock);
3723        free_ftrace_hash(iter->hash);
3724        kfree(iter);
3725
3726        return 0;
3727}
3728
3729static const struct file_operations ftrace_avail_fops = {
3730        .open = ftrace_avail_open,
3731        .read = seq_read,
3732        .llseek = seq_lseek,
3733        .release = seq_release_private,
3734};
3735
3736static const struct file_operations ftrace_enabled_fops = {
3737        .open = ftrace_enabled_open,
3738        .read = seq_read,
3739        .llseek = seq_lseek,
3740        .release = seq_release_private,
3741};
3742
3743static const struct file_operations ftrace_filter_fops = {
3744        .open = ftrace_filter_open,
3745        .read = seq_read,
3746        .write = ftrace_filter_write,
3747        .llseek = ftrace_filter_lseek,
3748        .release = ftrace_regex_release,
3749};
3750
3751static const struct file_operations ftrace_notrace_fops = {
3752        .open = ftrace_notrace_open,
3753        .read = seq_read,
3754        .write = ftrace_notrace_write,
3755        .llseek = ftrace_filter_lseek,
3756        .release = ftrace_regex_release,
3757};
3758
3759#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3760
3761static DEFINE_MUTEX(graph_lock);
3762
3763int ftrace_graph_count;
3764int ftrace_graph_filter_enabled;
3765unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3766
3767static void *
3768__g_next(struct seq_file *m, loff_t *pos)
3769{
3770        if (*pos >= ftrace_graph_count)
3771                return NULL;
3772        return &ftrace_graph_funcs[*pos];
3773}
3774
3775static void *
3776g_next(struct seq_file *m, void *v, loff_t *pos)
3777{
3778        (*pos)++;
3779        return __g_next(m, pos);
3780}
3781
3782static void *g_start(struct seq_file *m, loff_t *pos)
3783{
3784        mutex_lock(&graph_lock);
3785
3786        /* Nothing, tell g_show to print all functions are enabled */
3787        if (!ftrace_graph_filter_enabled && !*pos)
3788                return (void *)1;
3789
3790        return __g_next(m, pos);
3791}
3792
3793static void g_stop(struct seq_file *m, void *p)
3794{
3795        mutex_unlock(&graph_lock);
3796}
3797
3798static int g_show(struct seq_file *m, void *v)
3799{
3800        unsigned long *ptr = v;
3801
3802        if (!ptr)
3803                return 0;
3804
3805        if (ptr == (unsigned long *)1) {
3806                seq_printf(m, "#### all functions enabled ####\n");
3807                return 0;
3808        }
3809
3810        seq_printf(m, "%ps\n", (void *)*ptr);
3811
3812        return 0;
3813}
3814
3815static const struct seq_operations ftrace_graph_seq_ops = {
3816        .start = g_start,
3817        .next = g_next,
3818        .stop = g_stop,
3819        .show = g_show,
3820};
3821
3822static int
3823ftrace_graph_open(struct inode *inode, struct file *file)
3824{
3825        int ret = 0;
3826
3827        if (unlikely(ftrace_disabled))
3828                return -ENODEV;
3829
3830        mutex_lock(&graph_lock);
3831        if ((file->f_mode & FMODE_WRITE) &&
3832            (file->f_flags & O_TRUNC)) {
3833                ftrace_graph_filter_enabled = 0;
3834                ftrace_graph_count = 0;
3835                memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
3836        }
3837        mutex_unlock(&graph_lock);
3838
3839        if (file->f_mode & FMODE_READ)
3840                ret = seq_open(file, &ftrace_graph_seq_ops);
3841
3842        return ret;
3843}
3844
3845static int
3846ftrace_graph_release(struct inode *inode, struct file *file)
3847{
3848        if (file->f_mode & FMODE_READ)
3849                seq_release(inode, file);
3850        return 0;
3851}
3852
3853static int
3854ftrace_set_func(unsigned long *array, int *idx, char *buffer)
3855{
3856        struct dyn_ftrace *rec;
3857        struct ftrace_page *pg;
3858        int search_len;
3859        int fail = 1;
3860        int type, not;
3861        char *search;
3862        bool exists;
3863        int i;
3864
3865        /* decode regex */
3866        type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
3867        if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
3868                return -EBUSY;
3869
3870        search_len = strlen(search);
3871
3872        mutex_lock(&ftrace_lock);
3873
3874        if (unlikely(ftrace_disabled)) {
3875                mutex_unlock(&ftrace_lock);
3876                return -ENODEV;
3877        }
3878
3879        do_for_each_ftrace_rec(pg, rec) {
3880
3881                if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3882                        /* if it is in the array */
3883                        exists = false;
3884                        for (i = 0; i < *idx; i++) {
3885                                if (array[i] == rec->ip) {
3886                                        exists = true;
3887                                        break;
3888                                }
3889                        }
3890
3891                        if (!not) {
3892                                fail = 0;
3893                                if (!exists) {
3894                                        array[(*idx)++] = rec->ip;
3895                                        if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
3896                                                goto out;
3897                                }
3898                        } else {
3899                                if (exists) {
3900                                        array[i] = array[--(*idx)];
3901                                        array[*idx] = 0;
3902                                        fail = 0;
3903                                }
3904                        }
3905                }
3906        } while_for_each_ftrace_rec();
3907out:
3908        mutex_unlock(&ftrace_lock);
3909
3910        if (fail)
3911                return -EINVAL;
3912
3913        ftrace_graph_filter_enabled = !!(*idx);
3914
3915        return 0;
3916}
3917
3918static ssize_t
3919ftrace_graph_write(struct file *file, const char __user *ubuf,
3920                   size_t cnt, loff_t *ppos)
3921{
3922        struct trace_parser parser;
3923        ssize_t read, ret;
3924
3925        if (!cnt)
3926                return 0;
3927
3928        mutex_lock(&graph_lock);
3929
3930        if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
3931                ret = -ENOMEM;
3932                goto out_unlock;
3933        }
3934
3935        read = trace_get_user(&parser, ubuf, cnt, ppos);
3936
3937        if (read >= 0 && trace_parser_loaded((&parser))) {
3938                parser.buffer[parser.idx] = 0;
3939
3940                /* we allow only one expression at a time */
3941                ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3942                                        parser.buffer);
3943                if (ret)
3944                        goto out_free;
3945        }
3946
3947        ret = read;
3948
3949out_free:
3950        trace_parser_put(&parser);
3951out_unlock:
3952        mutex_unlock(&graph_lock);
3953
3954        return ret;
3955}
3956
3957static const struct file_operations ftrace_graph_fops = {
3958        .open           = ftrace_graph_open,
3959        .read           = seq_read,
3960        .write          = ftrace_graph_write,
3961        .llseek         = ftrace_filter_lseek,
3962        .release        = ftrace_graph_release,
3963};
3964#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3965
3966static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
3967{
3968
3969        trace_create_file("available_filter_functions", 0444,
3970                        d_tracer, NULL, &ftrace_avail_fops);
3971
3972        trace_create_file("enabled_functions", 0444,
3973                        d_tracer, NULL, &ftrace_enabled_fops);
3974
3975        trace_create_file("set_ftrace_filter", 0644, d_tracer,
3976                        NULL, &ftrace_filter_fops);
3977
3978        trace_create_file("set_ftrace_notrace", 0644, d_tracer,
3979                                    NULL, &ftrace_notrace_fops);
3980
3981#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3982        trace_create_file("set_graph_function", 0444, d_tracer,
3983                                    NULL,
3984                                    &ftrace_graph_fops);
3985#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3986
3987        return 0;
3988}
3989
3990static int ftrace_cmp_ips(const void *a, const void *b)
3991{
3992        const unsigned long *ipa = a;
3993        const unsigned long *ipb = b;
3994
3995        if (*ipa > *ipb)
3996                return 1;
3997        if (*ipa < *ipb)
3998                return -1;
3999        return 0;
4000}
4001
4002static void ftrace_swap_ips(void *a, void *b, int size)
4003{
4004        unsigned long *ipa = a;
4005        unsigned long *ipb = b;
4006        unsigned long t;
4007
4008        t = *ipa;
4009        *ipa = *ipb;
4010        *ipb = t;
4011}
4012
4013static int ftrace_process_locs(struct module *mod,
4014                               unsigned long *start,
4015                               unsigned long *end)
4016{
4017        struct ftrace_page *start_pg;
4018        struct ftrace_page *pg;
4019        struct dyn_ftrace *rec;
4020        unsigned long count;
4021        unsigned long *p;
4022        unsigned long addr;
4023        unsigned long flags = 0; /* Shut up gcc */
4024        int ret = -ENOMEM;
4025
4026        count = end - start;
4027
4028        if (!count)
4029                return 0;
4030
4031        sort(start, count, sizeof(*start),
4032             ftrace_cmp_ips, ftrace_swap_ips);
4033
4034        start_pg = ftrace_allocate_pages(count);
4035        if (!start_pg)
4036                return -ENOMEM;
4037
4038        mutex_lock(&ftrace_lock);
4039
4040        /*
4041         * Core and each module needs their own pages, as
4042         * modules will free them when they are removed.
4043         * Force a new page to be allocated for modules.
4044         */
4045        if (!mod) {
4046                WARN_ON(ftrace_pages || ftrace_pages_start);
4047                /* First initialization */
4048                ftrace_pages = ftrace_pages_start = start_pg;
4049        } else {
4050                if (!ftrace_pages)
4051                        goto out;
4052
4053                if (WARN_ON(ftrace_pages->next)) {
4054                        /* Hmm, we have free pages? */
4055                        while (ftrace_pages->next)
4056                                ftrace_pages = ftrace_pages->next;
4057                }
4058
4059                ftrace_pages->next = start_pg;
4060        }
4061
4062        p = start;
4063        pg = start_pg;
4064        while (p < end) {
4065                addr = ftrace_call_adjust(*p++);
4066                /*
4067                 * Some architecture linkers will pad between
4068                 * the different mcount_loc sections of different
4069                 * object files to satisfy alignments.
4070                 * Skip any NULL pointers.
4071                 */
4072                if (!addr)
4073                        continue;
4074
4075                if (pg->index == pg->size) {
4076                        /* We should have allocated enough */
4077                        if (WARN_ON(!pg->next))
4078                                break;
4079                        pg = pg->next;
4080                }
4081
4082                rec = &pg->records[pg->index++];
4083                rec->ip = addr;
4084        }
4085
4086        /* We should have used all pages */
4087        WARN_ON(pg->next);
4088
4089        /* Assign the last page to ftrace_pages */
4090        ftrace_pages = pg;
4091
4092        /* These new locations need to be initialized */
4093        ftrace_new_pgs = start_pg;
4094
4095        /*
4096         * We only need to disable interrupts on start up
4097         * because we are modifying code that an interrupt
4098         * may execute, and the modification is not atomic.
4099         * But for modules, nothing runs the code we modify
4100         * until we are finished with it, and there's no
4101         * reason to cause large interrupt latencies while we do it.
4102         */
4103        if (!mod)
4104                local_irq_save(flags);
4105        ftrace_update_code(mod);
4106        if (!mod)
4107                local_irq_restore(flags);
4108        ret = 0;
4109 out:
4110        mutex_unlock(&ftrace_lock);
4111
4112        return ret;
4113}
4114
4115#ifdef CONFIG_MODULES
4116
4117#define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4118
4119void ftrace_release_mod(struct module *mod)
4120{
4121        struct dyn_ftrace *rec;
4122        struct ftrace_page **last_pg;
4123        struct ftrace_page *pg;
4124        int order;
4125
4126        mutex_lock(&ftrace_lock);
4127
4128        if (ftrace_disabled)
4129                goto out_unlock;
4130
4131        /*
4132         * Each module has its own ftrace_pages, remove
4133         * them from the list.
4134         */
4135        last_pg = &ftrace_pages_start;
4136        for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4137                rec = &pg->records[0];
4138                if (within_module_core(rec->ip, mod)) {
4139                        /*
4140                         * As core pages are first, the first
4141                         * page should never be a module page.
4142                         */
4143                        if (WARN_ON(pg == ftrace_pages_start))
4144                                goto out_unlock;
4145
4146                        /* Check if we are deleting the last page */
4147                        if (pg == ftrace_pages)
4148                                ftrace_pages = next_to_ftrace_page(last_pg);
4149
4150                        *last_pg = pg->next;
4151                        order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4152                        free_pages((unsigned long)pg->records, order);
4153                        kfree(pg);
4154                } else
4155                        last_pg = &pg->next;
4156        }
4157 out_unlock:
4158        mutex_unlock(&ftrace_lock);
4159}
4160
4161static void ftrace_init_module(struct module *mod,
4162                               unsigned long *start, unsigned long *end)
4163{
4164        if (ftrace_disabled || start == end)
4165                return;
4166        ftrace_process_locs(mod, start, end);
4167}
4168
4169static int ftrace_module_notify_enter(struct notifier_block *self,
4170                                      unsigned long val, void *data)
4171{
4172        struct module *mod = data;
4173
4174        if (val == MODULE_STATE_COMING)
4175                ftrace_init_module(mod, mod->ftrace_callsites,
4176                                   mod->ftrace_callsites +
4177                                   mod->num_ftrace_callsites);
4178        return 0;
4179}
4180
4181static int ftrace_module_notify_exit(struct notifier_block *self,
4182                                     unsigned long val, void *data)
4183{
4184        struct module *mod = data;
4185
4186        if (val == MODULE_STATE_GOING)
4187                ftrace_release_mod(mod);
4188
4189        return 0;
4190}
4191#else
4192static int ftrace_module_notify_enter(struct notifier_block *self,
4193                                      unsigned long val, void *data)
4194{
4195        return 0;
4196}
4197static int ftrace_module_notify_exit(struct notifier_block *self,
4198                                     unsigned long val, void *data)
4199{
4200        return 0;
4201}
4202#endif /* CONFIG_MODULES */
4203
4204struct notifier_block ftrace_module_enter_nb = {
4205        .notifier_call = ftrace_module_notify_enter,
4206        .priority = INT_MAX,    /* Run before anything that can use kprobes */
4207};
4208
4209struct notifier_block ftrace_module_exit_nb = {
4210        .notifier_call = ftrace_module_notify_exit,
4211        .priority = INT_MIN,    /* Run after anything that can remove kprobes */
4212};
4213
4214extern unsigned long __start_mcount_loc[];
4215extern unsigned long __stop_mcount_loc[];
4216
4217void __init ftrace_init(void)
4218{
4219        unsigned long count, addr, flags;
4220        int ret;
4221
4222        /* Keep the ftrace pointer to the stub */
4223        addr = (unsigned long)ftrace_stub;
4224
4225        local_irq_save(flags);
4226        ftrace_dyn_arch_init(&addr);
4227        local_irq_restore(flags);
4228
4229        /* ftrace_dyn_arch_init places the return code in addr */
4230        if (addr)
4231                goto failed;
4232
4233        count = __stop_mcount_loc - __start_mcount_loc;
4234
4235        ret = ftrace_dyn_table_alloc(count);
4236        if (ret)
4237                goto failed;
4238
4239        last_ftrace_enabled = ftrace_enabled = 1;
4240
4241        ret = ftrace_process_locs(NULL,
4242                                  __start_mcount_loc,
4243                                  __stop_mcount_loc);
4244
4245        ret = register_module_notifier(&ftrace_module_enter_nb);
4246        if (ret)
4247                pr_warning("Failed to register trace ftrace module enter notifier\n");
4248
4249        ret = register_module_notifier(&ftrace_module_exit_nb);
4250        if (ret)
4251                pr_warning("Failed to register trace ftrace module exit notifier\n");
4252
4253        set_ftrace_early_filters();
4254
4255        return;
4256 failed:
4257        ftrace_disabled = 1;
4258}
4259
4260#else
4261
4262static struct ftrace_ops global_ops = {
4263        .func                   = ftrace_stub,
4264        .flags                  = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4265        INIT_REGEX_LOCK(global_ops)
4266};
4267
4268static int __init ftrace_nodyn_init(void)
4269{
4270        ftrace_enabled = 1;
4271        return 0;
4272}
4273core_initcall(ftrace_nodyn_init);
4274
4275static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
4276static inline void ftrace_startup_enable(int command) { }
4277/* Keep as macros so we do not need to define the commands */
4278# define ftrace_startup(ops, command)                   \
4279        ({                                              \
4280                (ops)->flags |= FTRACE_OPS_FL_ENABLED;  \
4281                0;                                      \
4282        })
4283# define ftrace_shutdown(ops, command)  do { } while (0)
4284# define ftrace_startup_sysctl()        do { } while (0)
4285# define ftrace_shutdown_sysctl()       do { } while (0)
4286
4287static inline int
4288ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
4289{
4290        return 1;
4291}
4292
4293#endif /* CONFIG_DYNAMIC_FTRACE */
4294
4295static void
4296ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip,
4297                        struct ftrace_ops *op, struct pt_regs *regs)
4298{
4299        if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
4300                return;
4301
4302        /*
4303         * Some of the ops may be dynamically allocated,
4304         * they must be freed after a synchronize_sched().
4305         */
4306        preempt_disable_notrace();
4307        trace_recursion_set(TRACE_CONTROL_BIT);
4308        do_for_each_ftrace_op(op, ftrace_control_list) {
4309                if (!(op->flags & FTRACE_OPS_FL_STUB) &&
4310                    !ftrace_function_local_disabled(op) &&
4311                    ftrace_ops_test(op, ip, regs))
4312                        op->func(ip, parent_ip, op, regs);
4313        } while_for_each_ftrace_op(op);
4314        trace_recursion_clear(TRACE_CONTROL_BIT);
4315        preempt_enable_notrace();
4316}
4317
4318static struct ftrace_ops control_ops = {
4319        .func   = ftrace_ops_control_func,
4320        .flags  = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
4321        INIT_REGEX_LOCK(control_ops)
4322};
4323
4324static inline void
4325__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4326                       struct ftrace_ops *ignored, struct pt_regs *regs)
4327{
4328        struct ftrace_ops *op;
4329        int bit;
4330
4331        if (function_trace_stop)
4332                return;
4333
4334        bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
4335        if (bit < 0)
4336                return;
4337
4338        /*
4339         * Some of the ops may be dynamically allocated,
4340         * they must be freed after a synchronize_sched().
4341         */
4342        preempt_disable_notrace();
4343        do_for_each_ftrace_op(op, ftrace_ops_list) {
4344                if (ftrace_ops_test(op, ip, regs))
4345                        op->func(ip, parent_ip, op, regs);
4346        } while_for_each_ftrace_op(op);
4347        preempt_enable_notrace();
4348        trace_clear_recursion(bit);
4349}
4350
4351/*
4352 * Some archs only support passing ip and parent_ip. Even though
4353 * the list function ignores the op parameter, we do not want any
4354 * C side effects, where a function is called without the caller
4355 * sending a third parameter.
4356 * Archs are to support both the regs and ftrace_ops at the same time.
4357 * If they support ftrace_ops, it is assumed they support regs.
4358 * If call backs want to use regs, they must either check for regs
4359 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
4360 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
4361 * An architecture can pass partial regs with ftrace_ops and still
4362 * set the ARCH_SUPPORT_FTARCE_OPS.
4363 */
4364#if ARCH_SUPPORTS_FTRACE_OPS
4365static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4366                                 struct ftrace_ops *op, struct pt_regs *regs)
4367{
4368        __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
4369}
4370#else
4371static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
4372{
4373        __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
4374}
4375#endif
4376
4377static void clear_ftrace_swapper(void)
4378{
4379        struct task_struct *p;
4380        int cpu;
4381
4382        get_online_cpus();
4383        for_each_online_cpu(cpu) {
4384                p = idle_task(cpu);
4385                clear_tsk_trace_trace(p);
4386        }
4387        put_online_cpus();
4388}
4389
4390static void set_ftrace_swapper(void)
4391{
4392        struct task_struct *p;
4393        int cpu;
4394
4395        get_online_cpus();
4396        for_each_online_cpu(cpu) {
4397                p = idle_task(cpu);
4398                set_tsk_trace_trace(p);
4399        }
4400        put_online_cpus();
4401}
4402
4403static void clear_ftrace_pid(struct pid *pid)
4404{
4405        struct task_struct *p;
4406
4407        rcu_read_lock();
4408        do_each_pid_task(pid, PIDTYPE_PID, p) {
4409                clear_tsk_trace_trace(p);
4410        } while_each_pid_task(pid, PIDTYPE_PID, p);
4411        rcu_read_unlock();
4412
4413        put_pid(pid);
4414}
4415
4416static void set_ftrace_pid(struct pid *pid)
4417{
4418        struct task_struct *p;
4419
4420        rcu_read_lock();
4421        do_each_pid_task(pid, PIDTYPE_PID, p) {
4422                set_tsk_trace_trace(p);
4423        } while_each_pid_task(pid, PIDTYPE_PID, p);
4424        rcu_read_unlock();
4425}
4426
4427static void clear_ftrace_pid_task(struct pid *pid)
4428{
4429        if (pid == ftrace_swapper_pid)
4430                clear_ftrace_swapper();
4431        else
4432                clear_ftrace_pid(pid);
4433}
4434
4435static void set_ftrace_pid_task(struct pid *pid)
4436{
4437        if (pid == ftrace_swapper_pid)
4438                set_ftrace_swapper();
4439        else
4440                set_ftrace_pid(pid);
4441}
4442
4443static int ftrace_pid_add(int p)
4444{
4445        struct pid *pid;
4446        struct ftrace_pid *fpid;
4447        int ret = -EINVAL;
4448
4449        mutex_lock(&ftrace_lock);
4450
4451        if (!p)
4452                pid = ftrace_swapper_pid;
4453        else
4454                pid = find_get_pid(p);
4455
4456        if (!pid)
4457                goto out;
4458
4459        ret = 0;
4460
4461        list_for_each_entry(fpid, &ftrace_pids, list)
4462                if (fpid->pid == pid)
4463                        goto out_put;
4464
4465        ret = -ENOMEM;
4466
4467        fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
4468        if (!fpid)
4469                goto out_put;
4470
4471        list_add(&fpid->list, &ftrace_pids);
4472        fpid->pid = pid;
4473
4474        set_ftrace_pid_task(pid);
4475
4476        ftrace_update_pid_func();
4477        ftrace_startup_enable(0);
4478
4479        mutex_unlock(&ftrace_lock);
4480        return 0;
4481
4482out_put:
4483        if (pid != ftrace_swapper_pid)
4484                put_pid(pid);
4485
4486out:
4487        mutex_unlock(&ftrace_lock);
4488        return ret;
4489}
4490
4491static void ftrace_pid_reset(void)
4492{
4493        struct ftrace_pid *fpid, *safe;
4494
4495        mutex_lock(&ftrace_lock);
4496        list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
4497                struct pid *pid = fpid->pid;
4498
4499                clear_ftrace_pid_task(pid);
4500
4501                list_del(&fpid->list);
4502                kfree(fpid);
4503        }
4504
4505        ftrace_update_pid_func();
4506        ftrace_startup_enable(0);
4507
4508        mutex_unlock(&ftrace_lock);
4509}
4510
4511static void *fpid_start(struct seq_file *m, loff_t *pos)
4512{
4513        mutex_lock(&ftrace_lock);
4514
4515        if (list_empty(&ftrace_pids) && (!*pos))
4516                return (void *) 1;
4517
4518        return seq_list_start(&ftrace_pids, *pos);
4519}
4520
4521static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
4522{
4523        if (v == (void *)1)
4524                return NULL;
4525
4526        return seq_list_next(v, &ftrace_pids, pos);
4527}
4528
4529static void fpid_stop(struct seq_file *m, void *p)
4530{
4531        mutex_unlock(&ftrace_lock);
4532}
4533
4534static int fpid_show(struct seq_file *m, void *v)
4535{
4536        const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
4537
4538        if (v == (void *)1) {
4539                seq_printf(m, "no pid\n");
4540                return 0;
4541        }
4542
4543        if (fpid->pid == ftrace_swapper_pid)
4544                seq_printf(m, "swapper tasks\n");
4545        else
4546                seq_printf(m, "%u\n", pid_vnr(fpid->pid));
4547
4548        return 0;
4549}
4550
4551static const struct seq_operations ftrace_pid_sops = {
4552        .start = fpid_start,
4553        .next = fpid_next,
4554        .stop = fpid_stop,
4555        .show = fpid_show,
4556};
4557
4558static int
4559ftrace_pid_open(struct inode *inode, struct file *file)
4560{
4561        int ret = 0;
4562
4563        if ((file->f_mode & FMODE_WRITE) &&
4564            (file->f_flags & O_TRUNC))
4565                ftrace_pid_reset();
4566
4567        if (file->f_mode & FMODE_READ)
4568                ret = seq_open(file, &ftrace_pid_sops);
4569
4570        return ret;
4571}
4572
4573static ssize_t
4574ftrace_pid_write(struct file *filp, const char __user *ubuf,
4575                   size_t cnt, loff_t *ppos)
4576{
4577        char buf[64], *tmp;
4578        long val;
4579        int ret;
4580
4581        if (cnt >= sizeof(buf))
4582                return -EINVAL;
4583
4584        if (copy_from_user(&buf, ubuf, cnt))
4585                return -EFAULT;
4586
4587        buf[cnt] = 0;
4588
4589        /*
4590         * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
4591         * to clean the filter quietly.
4592         */
4593        tmp = strstrip(buf);
4594        if (strlen(tmp) == 0)
4595                return 1;
4596
4597        ret = kstrtol(tmp, 10, &val);
4598        if (ret < 0)
4599                return ret;
4600
4601        ret = ftrace_pid_add(val);
4602
4603        return ret ? ret : cnt;
4604}
4605
4606static int
4607ftrace_pid_release(struct inode *inode, struct file *file)
4608{
4609        if (file->f_mode & FMODE_READ)
4610                seq_release(inode, file);
4611
4612        return 0;
4613}
4614
4615static const struct file_operations ftrace_pid_fops = {
4616        .open           = ftrace_pid_open,
4617        .write          = ftrace_pid_write,
4618        .read           = seq_read,
4619        .llseek         = ftrace_filter_lseek,
4620        .release        = ftrace_pid_release,
4621};
4622
4623static __init int ftrace_init_debugfs(void)
4624{
4625        struct dentry *d_tracer;
4626
4627        d_tracer = tracing_init_dentry();
4628        if (!d_tracer)
4629                return 0;
4630
4631        ftrace_init_dyn_debugfs(d_tracer);
4632
4633        trace_create_file("set_ftrace_pid", 0644, d_tracer,
4634                            NULL, &ftrace_pid_fops);
4635
4636        ftrace_profile_debugfs(d_tracer);
4637
4638        return 0;
4639}
4640fs_initcall(ftrace_init_debugfs);
4641
4642/**
4643 * ftrace_kill - kill ftrace
4644 *
4645 * This function should be used by panic code. It stops ftrace
4646 * but in a not so nice way. If you need to simply kill ftrace
4647 * from a non-atomic section, use ftrace_kill.
4648 */
4649void ftrace_kill(void)
4650{
4651        ftrace_disabled = 1;
4652        ftrace_enabled = 0;
4653        clear_ftrace_function();
4654}
4655
4656/**
4657 * Test if ftrace is dead or not.
4658 */
4659int ftrace_is_dead(void)
4660{
4661        return ftrace_disabled;
4662}
4663
4664/**
4665 * register_ftrace_function - register a function for profiling
4666 * @ops - ops structure that holds the function for profiling.
4667 *
4668 * Register a function to be called by all functions in the
4669 * kernel.
4670 *
4671 * Note: @ops->func and all the functions it calls must be labeled
4672 *       with "notrace", otherwise it will go into a
4673 *       recursive loop.
4674 */
4675int register_ftrace_function(struct ftrace_ops *ops)
4676{
4677        int ret = -1;
4678
4679        ftrace_ops_init(ops);
4680
4681        mutex_lock(&ftrace_lock);
4682
4683        ret = __register_ftrace_function(ops);
4684        if (!ret)
4685                ret = ftrace_startup(ops, 0);
4686
4687        mutex_unlock(&ftrace_lock);
4688
4689        return ret;
4690}
4691EXPORT_SYMBOL_GPL(register_ftrace_function);
4692
4693/**
4694 * unregister_ftrace_function - unregister a function for profiling.
4695 * @ops - ops structure that holds the function to unregister
4696 *
4697 * Unregister a function that was added to be called by ftrace profiling.
4698 */
4699int unregister_ftrace_function(struct ftrace_ops *ops)
4700{
4701        int ret;
4702
4703        mutex_lock(&ftrace_lock);
4704        ret = __unregister_ftrace_function(ops);
4705        if (!ret)
4706                ftrace_shutdown(ops, 0);
4707        mutex_unlock(&ftrace_lock);
4708
4709        return ret;
4710}
4711EXPORT_SYMBOL_GPL(unregister_ftrace_function);
4712
4713int
4714ftrace_enable_sysctl(struct ctl_table *table, int write,
4715                     void __user *buffer, size_t *lenp,
4716                     loff_t *ppos)
4717{
4718        int ret = -ENODEV;
4719
4720        mutex_lock(&ftrace_lock);
4721
4722        if (unlikely(ftrace_disabled))
4723                goto out;
4724
4725        ret = proc_dointvec(table, write, buffer, lenp, ppos);
4726
4727        if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
4728                goto out;
4729
4730        last_ftrace_enabled = !!ftrace_enabled;
4731
4732        if (ftrace_enabled) {
4733
4734                ftrace_startup_sysctl();
4735
4736                /* we are starting ftrace again */
4737                if (ftrace_ops_list != &ftrace_list_end)
4738                        update_ftrace_function();
4739
4740        } else {
4741                /* stopping ftrace calls (just send to ftrace_stub) */
4742                ftrace_trace_function = ftrace_stub;
4743
4744                ftrace_shutdown_sysctl();
4745        }
4746
4747 out:
4748        mutex_unlock(&ftrace_lock);
4749        return ret;
4750}
4751
4752#ifdef CONFIG_FUNCTION_GRAPH_TRACER
4753
4754static int ftrace_graph_active;
4755static struct notifier_block ftrace_suspend_notifier;
4756
4757int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
4758{
4759        return 0;
4760}
4761
4762/* The callbacks that hook a function */
4763trace_func_graph_ret_t ftrace_graph_return =
4764                        (trace_func_graph_ret_t)ftrace_stub;
4765trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
4766
4767/* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
4768static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
4769{
4770        int i;
4771        int ret = 0;
4772        unsigned long flags;
4773        int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
4774        struct task_struct *g, *t;
4775
4776        for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
4777                ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
4778                                        * sizeof(struct ftrace_ret_stack),
4779                                        GFP_KERNEL);
4780                if (!ret_stack_list[i]) {
4781                        start = 0;
4782                        end = i;
4783                        ret = -ENOMEM;
4784                        goto free;
4785                }
4786        }
4787
4788        read_lock_irqsave(&tasklist_lock, flags);
4789        do_each_thread(g, t) {
4790                if (start == end) {
4791                        ret = -EAGAIN;
4792                        goto unlock;
4793                }
4794
4795                if (t->ret_stack == NULL) {
4796                        atomic_set(&t->tracing_graph_pause, 0);
4797                        atomic_set(&t->trace_overrun, 0);
4798                        t->curr_ret_stack = -1;
4799                        /* Make sure the tasks see the -1 first: */
4800                        smp_wmb();
4801                        t->ret_stack = ret_stack_list[start++];
4802                }
4803        } while_each_thread(g, t);
4804
4805unlock:
4806        read_unlock_irqrestore(&tasklist_lock, flags);
4807free:
4808        for (i = start; i < end; i++)
4809                kfree(ret_stack_list[i]);
4810        return ret;
4811}
4812
4813static void
4814ftrace_graph_probe_sched_switch(void *ignore,
4815                        struct task_struct *prev, struct task_struct *next)
4816{
4817        unsigned long long timestamp;
4818        int index;
4819
4820        /*
4821         * Does the user want to count the time a function was asleep.
4822         * If so, do not update the time stamps.
4823         */
4824        if (trace_flags & TRACE_ITER_SLEEP_TIME)
4825                return;
4826
4827        timestamp = trace_clock_local();
4828
4829        prev->ftrace_timestamp = timestamp;
4830
4831        /* only process tasks that we timestamped */
4832        if (!next->ftrace_timestamp)
4833                return;
4834
4835        /*
4836         * Update all the counters in next to make up for the
4837         * time next was sleeping.
4838         */
4839        timestamp -= next->ftrace_timestamp;
4840
4841        for (index = next->curr_ret_stack; index >= 0; index--)
4842                next->ret_stack[index].calltime += timestamp;
4843}
4844
4845/* Allocate a return stack for each task */
4846static int start_graph_tracing(void)
4847{
4848        struct ftrace_ret_stack **ret_stack_list;
4849        int ret, cpu;
4850
4851        ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
4852                                sizeof(struct ftrace_ret_stack *),
4853                                GFP_KERNEL);
4854
4855        if (!ret_stack_list)
4856                return -ENOMEM;
4857
4858        /* The cpu_boot init_task->ret_stack will never be freed */
4859        for_each_online_cpu(cpu) {
4860                if (!idle_task(cpu)->ret_stack)
4861                        ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4862        }
4863
4864        do {
4865                ret = alloc_retstack_tasklist(ret_stack_list);
4866        } while (ret == -EAGAIN);
4867
4868        if (!ret) {
4869                ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4870                if (ret)
4871                        pr_info("ftrace_graph: Couldn't activate tracepoint"
4872                                " probe to kernel_sched_switch\n");
4873        }
4874
4875        kfree(ret_stack_list);
4876        return ret;
4877}
4878
4879/*
4880 * Hibernation protection.
4881 * The state of the current task is too much unstable during
4882 * suspend/restore to disk. We want to protect against that.
4883 */
4884static int
4885ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4886                                                        void *unused)
4887{
4888        switch (state) {
4889        case PM_HIBERNATION_PREPARE:
4890                pause_graph_tracing();
4891                break;
4892
4893        case PM_POST_HIBERNATION:
4894                unpause_graph_tracing();
4895                break;
4896        }
4897        return NOTIFY_DONE;
4898}
4899
4900int register_ftrace_graph(trace_func_graph_ret_t retfunc,
4901                        trace_func_graph_ent_t entryfunc)
4902{
4903        int ret = 0;
4904
4905        mutex_lock(&ftrace_lock);
4906
4907        /* we currently allow only one tracer registered at a time */
4908        if (ftrace_graph_active) {
4909                ret = -EBUSY;
4910                goto out;
4911        }
4912
4913        ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
4914        register_pm_notifier(&ftrace_suspend_notifier);
4915
4916        ftrace_graph_active++;
4917        ret = start_graph_tracing();
4918        if (ret) {
4919                ftrace_graph_active--;
4920                goto out;
4921        }
4922
4923        ftrace_graph_return = retfunc;
4924        ftrace_graph_entry = entryfunc;
4925
4926        ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
4927
4928out:
4929        mutex_unlock(&ftrace_lock);
4930        return ret;
4931}
4932
4933void unregister_ftrace_graph(void)
4934{
4935        mutex_lock(&ftrace_lock);
4936
4937        if (unlikely(!ftrace_graph_active))
4938                goto out;
4939
4940        ftrace_graph_active--;
4941        ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
4942        ftrace_graph_entry = ftrace_graph_entry_stub;
4943        ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
4944        unregister_pm_notifier(&ftrace_suspend_notifier);
4945        unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4946
4947 out:
4948        mutex_unlock(&ftrace_lock);
4949}
4950
4951static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
4952
4953static void
4954graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
4955{
4956        atomic_set(&t->tracing_graph_pause, 0);
4957        atomic_set(&t->trace_overrun, 0);
4958        t->ftrace_timestamp = 0;
4959        /* make curr_ret_stack visible before we add the ret_stack */
4960        smp_wmb();
4961        t->ret_stack = ret_stack;
4962}
4963
4964/*
4965 * Allocate a return stack for the idle task. May be the first
4966 * time through, or it may be done by CPU hotplug online.
4967 */
4968void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
4969{
4970        t->curr_ret_stack = -1;
4971        /*
4972         * The idle task has no parent, it either has its own
4973         * stack or no stack at all.
4974         */
4975        if (t->ret_stack)
4976                WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
4977
4978        if (ftrace_graph_active) {
4979                struct ftrace_ret_stack *ret_stack;
4980
4981                ret_stack = per_cpu(idle_ret_stack, cpu);
4982                if (!ret_stack) {
4983                        ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4984                                            * sizeof(struct ftrace_ret_stack),
4985                                            GFP_KERNEL);
4986                        if (!ret_stack)
4987                                return;
4988                        per_cpu(idle_ret_stack, cpu) = ret_stack;
4989                }
4990                graph_init_task(t, ret_stack);
4991        }
4992}
4993
4994/* Allocate a return stack for newly created task */
4995void ftrace_graph_init_task(struct task_struct *t)
4996{
4997        /* Make sure we do not use the parent ret_stack */
4998        t->ret_stack = NULL;
4999        t->curr_ret_stack = -1;
5000
5001        if (ftrace_graph_active) {
5002                struct ftrace_ret_stack *ret_stack;
5003
5004                ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5005                                * sizeof(struct ftrace_ret_stack),
5006                                GFP_KERNEL);
5007                if (!ret_stack)
5008                        return;
5009                graph_init_task(t, ret_stack);
5010        }
5011}
5012
5013void ftrace_graph_exit_task(struct task_struct *t)
5014{
5015        struct ftrace_ret_stack *ret_stack = t->ret_stack;
5016
5017        t->ret_stack = NULL;
5018        /* NULL must become visible to IRQs before we free it: */
5019        barrier();
5020
5021        kfree(ret_stack);
5022}
5023
5024void ftrace_graph_stop(void)
5025{
5026        ftrace_stop();
5027}
5028#endif
5029