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