kernel/trace/ftrace.c

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