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