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         int ret = 0;
 385 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 386         static struct trace_seq s;
 387         unsigned long long avg;
 388         unsigned long long stddev;
 389 #endif
 390         mutex_lock(&ftrace_profile_lock);
 391
 392         /* we raced with function_profile_reset() */
 393         if (unlikely(rec->counter == 0)) {
 394                 ret = -EBUSY;
 395                 goto out;
 396         }
 397
 398         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 399         seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
 400
 401 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 402         seq_printf(m, "    ");
 403         avg = rec->time;
 404         do_div(avg, rec->counter);
 405
 406         /* Sample standard deviation (s^2) */
 407         if (rec->counter <= 1)
 408                 stddev = 0;
 409         else {
 410                 stddev = rec->time_squared - rec->counter * avg * avg;
 411                 /*
 412                  * Divide only 1000 for ns^2 -> us^2 conversion.
 413                  * trace_print_graph_duration will divide 1000 again.
 414                  */
 415                 do_div(stddev, (rec->counter - 1) * 1000);
 416         }
 417
 418         trace_seq_init(&s);
 419         trace_print_graph_duration(rec->time, &s);
 420         trace_seq_puts(&s, "    ");
 421         trace_print_graph_duration(avg, &s);
 422         trace_seq_puts(&s, "    ");
 423         trace_print_graph_duration(stddev, &s);
 424         trace_print_seq(m, &s);
 425 #endif
 426         seq_putc(m, '\n');
 427 out:
 428         mutex_unlock(&ftrace_profile_lock);
 429
 430         return ret;
 431 }
 432
 433 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
 434 {
 435         struct ftrace_profile_page *pg;
 436
 437         pg = stat->pages = stat->start;
 438
 439         while (pg) {
 440                 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
 441                 pg->index = 0;
 442                 pg = pg->next;
 443         }
 444
 445         memset(stat->hash, 0,
 446                FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
 447 }
 448
 449 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
 450 {
 451         struct ftrace_profile_page *pg;
 452         int functions;
 453         int pages;
 454         int i;
 455
 456         /* If we already allocated, do nothing */
 457         if (stat->pages)
 458                 return 0;
 459
 460         stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
 461         if (!stat->pages)
 462                 return -ENOMEM;
 463
 464 #ifdef CONFIG_DYNAMIC_FTRACE
 465         functions = ftrace_update_tot_cnt;
 466 #else
 467         /*
 468          * We do not know the number of functions that exist because
 469          * dynamic tracing is what counts them. With past experience
 470          * we have around 20K functions. That should be more than enough.
 471          * It is highly unlikely we will execute every function in
 472          * the kernel.
 473          */
 474         functions = 20000;
 475 #endif
 476
 477         pg = stat->start = stat->pages;
 478
 479         pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
 480
 481         for (i = 0; i < pages; i++) {
 482                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
 483                 if (!pg->next)
 484                         goto out_free;
 485                 pg = pg->next;
 486         }
 487
 488         return 0;
 489
 490  out_free:
 491         pg = stat->start;
 492         while (pg) {
 493                 unsigned long tmp = (unsigned long)pg;
 494
 495                 pg = pg->next;
 496                 free_page(tmp);
 497         }
 498
 499         free_page((unsigned long)stat->pages);
 500         stat->pages = NULL;
 501         stat->start = NULL;
 502
 503         return -ENOMEM;
 504 }
 505
 506 static int ftrace_profile_init_cpu(int cpu)
 507 {
 508         struct ftrace_profile_stat *stat;
 509         int size;
 510
 511         stat = &per_cpu(ftrace_profile_stats, cpu);
 512
 513         if (stat->hash) {
 514                 /* If the profile is already created, simply reset it */
 515                 ftrace_profile_reset(stat);
 516                 return 0;
 517         }
 518
 519         /*
 520          * We are profiling all functions, but usually only a few thousand
 521          * functions are hit. We'll make a hash of 1024 items.
 522          */
 523         size = FTRACE_PROFILE_HASH_SIZE;
 524
 525         stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
 526
 527         if (!stat->hash)
 528                 return -ENOMEM;
 529
 530         if (!ftrace_profile_bits) {
 531                 size--;
 532
 533                 for (; size; size >>= 1)
 534                         ftrace_profile_bits++;
 535         }
 536
 537         /* Preallocate the function profiling pages */
 538         if (ftrace_profile_pages_init(stat) < 0) {
 539                 kfree(stat->hash);
 540                 stat->hash = NULL;
 541                 return -ENOMEM;
 542         }
 543
 544         return 0;
 545 }
 546
 547 static int ftrace_profile_init(void)
 548 {
 549         int cpu;
 550         int ret = 0;
 551
 552         for_each_online_cpu(cpu) {
 553                 ret = ftrace_profile_init_cpu(cpu);
 554                 if (ret)
 555                         break;
 556         }
 557
 558         return ret;
 559 }
 560
 561 /* interrupts must be disabled */
 562 static struct ftrace_profile *
 563 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
 564 {
 565         struct ftrace_profile *rec;
 566         struct hlist_head *hhd;
 567         struct hlist_node *n;
 568         unsigned long key;
 569
 570         key = hash_long(ip, ftrace_profile_bits);
 571         hhd = &stat->hash[key];
 572
 573         if (hlist_empty(hhd))
 574                 return NULL;
 575
 576         hlist_for_each_entry_rcu(rec, n, hhd, node) {
 577                 if (rec->ip == ip)
 578                         return rec;
 579         }
 580
 581         return NULL;
 582 }
 583
 584 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
 585                                struct ftrace_profile *rec)
 586 {
 587         unsigned long key;
 588
 589         key = hash_long(rec->ip, ftrace_profile_bits);
 590         hlist_add_head_rcu(&rec->node, &stat->hash[key]);
 591 }
 592
 593 /*
 594  * The memory is already allocated, this simply finds a new record to use.
 595  */
 596 static struct ftrace_profile *
 597 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
 598 {
 599         struct ftrace_profile *rec = NULL;
 600
 601         /* prevent recursion (from NMIs) */
 602         if (atomic_inc_return(&stat->disabled) != 1)
 603                 goto out;
 604
 605         /*
 606          * Try to find the function again since an NMI
 607          * could have added it
 608          */
 609         rec = ftrace_find_profiled_func(stat, ip);
 610         if (rec)
 611                 goto out;
 612
 613         if (stat->pages->index == PROFILES_PER_PAGE) {
 614                 if (!stat->pages->next)
 615                         goto out;
 616                 stat->pages = stat->pages->next;
 617         }
 618
 619         rec = &stat->pages->records[stat->pages->index++];
 620         rec->ip = ip;
 621         ftrace_add_profile(stat, rec);
 622
 623  out:
 624         atomic_dec(&stat->disabled);
 625
 626         return rec;
 627 }
 628
 629 static void
 630 function_profile_call(unsigned long ip, unsigned long parent_ip)
 631 {
 632         struct ftrace_profile_stat *stat;
 633         struct ftrace_profile *rec;
 634         unsigned long flags;
 635
 636         if (!ftrace_profile_enabled)
 637                 return;
 638
 639         local_irq_save(flags);
 640
 641         stat = &__get_cpu_var(ftrace_profile_stats);
 642         if (!stat->hash || !ftrace_profile_enabled)
 643                 goto out;
 644
 645         rec = ftrace_find_profiled_func(stat, ip);
 646         if (!rec) {
 647                 rec = ftrace_profile_alloc(stat, ip);
 648                 if (!rec)
 649                         goto out;
 650         }
 651
 652         rec->counter++;
 653  out:
 654         local_irq_restore(flags);
 655 }
 656
 657 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 658 static int profile_graph_entry(struct ftrace_graph_ent *trace)
 659 {
 660         function_profile_call(trace->func, 0);
 661         return 1;
 662 }
 663
 664 static void profile_graph_return(struct ftrace_graph_ret *trace)
 665 {
 666         struct ftrace_profile_stat *stat;
 667         unsigned long long calltime;
 668         struct ftrace_profile *rec;
 669         unsigned long flags;
 670
 671         local_irq_save(flags);
 672         stat = &__get_cpu_var(ftrace_profile_stats);
 673         if (!stat->hash || !ftrace_profile_enabled)
 674                 goto out;
 675
 676         /* If the calltime was zero'd ignore it */
 677         if (!trace->calltime)
 678                 goto out;
 679
 680         calltime = trace->rettime - trace->calltime;
 681
 682         if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
 683                 int index;
 684
 685                 index = trace->depth;
 686
 687                 /* Append this call time to the parent time to subtract */
 688                 if (index)
 689                         current->ret_stack[index - 1].subtime += calltime;
 690
 691                 if (current->ret_stack[index].subtime < calltime)
 692                         calltime -= current->ret_stack[index].subtime;
 693                 else
 694                         calltime = 0;
 695         }
 696
 697         rec = ftrace_find_profiled_func(stat, trace->func);
 698         if (rec) {
 699                 rec->time += calltime;
 700                 rec->time_squared += calltime * calltime;
 701         }
 702
 703  out:
 704         local_irq_restore(flags);
 705 }
 706
 707 static int register_ftrace_profiler(void)
 708 {
 709         return register_ftrace_graph(&profile_graph_return,
 710                                      &profile_graph_entry);
 711 }
 712
 713 static void unregister_ftrace_profiler(void)
 714 {
 715         unregister_ftrace_graph();
 716 }
 717 #else
 718 static struct ftrace_ops ftrace_profile_ops __read_mostly =
 719 {
 720         .func           = function_profile_call,
 721 };
 722
 723 static int register_ftrace_profiler(void)
 724 {
 725         return register_ftrace_function(&ftrace_profile_ops);
 726 }
 727
 728 static void unregister_ftrace_profiler(void)
 729 {
 730         unregister_ftrace_function(&ftrace_profile_ops);
 731 }
 732 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 733
 734 static ssize_t
 735 ftrace_profile_write(struct file *filp, const char __user *ubuf,
 736                      size_t cnt, loff_t *ppos)
 737 {
 738         unsigned long val;
 739         char buf[64];           /* big enough to hold a number */
 740         int ret;
 741
 742         if (cnt >= sizeof(buf))
 743                 return -EINVAL;
 744
 745         if (copy_from_user(&buf, ubuf, cnt))
 746                 return -EFAULT;
 747
 748         buf[cnt] = 0;
 749
 750         ret = strict_strtoul(buf, 10, &val);
 751         if (ret < 0)
 752                 return ret;
 753
 754         val = !!val;
 755
 756         mutex_lock(&ftrace_profile_lock);
 757         if (ftrace_profile_enabled ^ val) {
 758                 if (val) {
 759                         ret = ftrace_profile_init();
 760                         if (ret < 0) {
 761                                 cnt = ret;
 762                                 goto out;
 763                         }
 764
 765                         ret = register_ftrace_profiler();
 766                         if (ret < 0) {
 767                                 cnt = ret;
 768                                 goto out;
 769                         }
 770                         ftrace_profile_enabled = 1;
 771                 } else {
 772                         ftrace_profile_enabled = 0;
 773                         /*
 774                          * unregister_ftrace_profiler calls stop_machine
 775                          * so this acts like an synchronize_sched.
 776                          */
 777                         unregister_ftrace_profiler();
 778                 }
 779         }
 780  out:
 781         mutex_unlock(&ftrace_profile_lock);
 782
 783         *ppos += cnt;
 784
 785         return cnt;
 786 }
 787
 788 static ssize_t
 789 ftrace_profile_read(struct file *filp, char __user *ubuf,
 790                      size_t cnt, loff_t *ppos)
 791 {
 792         char buf[64];           /* big enough to hold a number */
 793         int r;
 794
 795         r = sprintf(buf, "%u\n", ftrace_profile_enabled);
 796         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 797 }
 798
 799 static const struct file_operations ftrace_profile_fops = {
 800         .open           = tracing_open_generic,
 801         .read           = ftrace_profile_read,
 802         .write          = ftrace_profile_write,
 803         .llseek         = default_llseek,
 804 };
 805
 806 /* used to initialize the real stat files */
 807 static struct tracer_stat function_stats __initdata = {
 808         .name           = "functions",
 809         .stat_start     = function_stat_start,
 810         .stat_next      = function_stat_next,
 811         .stat_cmp       = function_stat_cmp,
 812         .stat_headers   = function_stat_headers,
 813         .stat_show      = function_stat_show
 814 };
 815
 816 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 817 {
 818         struct ftrace_profile_stat *stat;
 819         struct dentry *entry;
 820         char *name;
 821         int ret;
 822         int cpu;
 823
 824         for_each_possible_cpu(cpu) {
 825                 stat = &per_cpu(ftrace_profile_stats, cpu);
 826
 827                 /* allocate enough for function name + cpu number */
 828                 name = kmalloc(32, GFP_KERNEL);
 829                 if (!name) {
 830                         /*
 831                          * The files created are permanent, if something happens
 832                          * we still do not free memory.
 833                          */
 834                         WARN(1,
 835                              "Could not allocate stat file for cpu %d\n",
 836                              cpu);
 837                         return;
 838                 }
 839                 stat->stat = function_stats;
 840                 snprintf(name, 32, "function%d", cpu);
 841                 stat->stat.name = name;
 842                 ret = register_stat_tracer(&stat->stat);
 843                 if (ret) {
 844                         WARN(1,
 845                              "Could not register function stat for cpu %d\n",
 846                              cpu);
 847                         kfree(name);
 848                         return;
 849                 }
 850         }
 851
 852         entry = debugfs_create_file("function_profile_enabled", 0644,
 853                                     d_tracer, NULL, &ftrace_profile_fops);
 854         if (!entry)
 855                 pr_warning("Could not create debugfs "
 856                            "'function_profile_enabled' entry\n");
 857 }
 858
 859 #else /* CONFIG_FUNCTION_PROFILER */
 860 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 861 {
 862 }
 863 #endif /* CONFIG_FUNCTION_PROFILER */
 864
 865 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
 866
 867 #ifdef CONFIG_DYNAMIC_FTRACE
 868
 869 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
 870 # error Dynamic ftrace depends on MCOUNT_RECORD
 871 #endif
 872
 873 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
 874
 875 struct ftrace_func_probe {
 876         struct hlist_node       node;
 877         struct ftrace_probe_ops *ops;
 878         unsigned long           flags;
 879         unsigned long           ip;
 880         void                    *data;
 881         struct rcu_head         rcu;
 882 };
 883
 884 enum {
 885         FTRACE_ENABLE_CALLS             = (1 << 0),
 886         FTRACE_DISABLE_CALLS            = (1 << 1),
 887         FTRACE_UPDATE_TRACE_FUNC        = (1 << 2),
 888         FTRACE_START_FUNC_RET           = (1 << 3),
 889         FTRACE_STOP_FUNC_RET            = (1 << 4),
 890 };
 891
 892 static int ftrace_filtered;
 893
 894 static struct dyn_ftrace *ftrace_new_addrs;
 895
 896 static DEFINE_MUTEX(ftrace_regex_lock);
 897
 898 struct ftrace_page {
 899         struct ftrace_page      *next;
 900         int                     index;
 901         struct dyn_ftrace       records[];
 902 };
 903
 904 #define ENTRIES_PER_PAGE \
 905   ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
 906
 907 /* estimate from running different kernels */
 908 #define NR_TO_INIT              10000
 909
 910 static struct ftrace_page       *ftrace_pages_start;
 911 static struct ftrace_page       *ftrace_pages;
 912
 913 static struct dyn_ftrace *ftrace_free_records;
 914
 915 /*
 916  * This is a double for. Do not use 'break' to break out of the loop,
 917  * you must use a goto.
 918  */
 919 #define do_for_each_ftrace_rec(pg, rec)                                 \
 920         for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
 921                 int _____i;                                             \
 922                 for (_____i = 0; _____i < pg->index; _____i++) {        \
 923                         rec = &pg->records[_____i];
 924
 925 #define while_for_each_ftrace_rec()             \
 926                 }                               \
 927         }
 928
 929 static void ftrace_free_rec(struct dyn_ftrace *rec)
 930 {
 931         rec->freelist = ftrace_free_records;
 932         ftrace_free_records = rec;
 933         rec->flags |= FTRACE_FL_FREE;
 934 }
 935
 936 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
 937 {
 938         struct dyn_ftrace *rec;
 939
 940         /* First check for freed records */
 941         if (ftrace_free_records) {
 942                 rec = ftrace_free_records;
 943
 944                 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
 945                         FTRACE_WARN_ON_ONCE(1);
 946                         ftrace_free_records = NULL;
 947                         return NULL;
 948                 }
 949
 950                 ftrace_free_records = rec->freelist;
 951                 memset(rec, 0, sizeof(*rec));
 952                 return rec;
 953         }
 954
 955         if (ftrace_pages->index == ENTRIES_PER_PAGE) {
 956                 if (!ftrace_pages->next) {
 957                         /* allocate another page */
 958                         ftrace_pages->next =
 959                                 (void *)get_zeroed_page(GFP_KERNEL);
 960                         if (!ftrace_pages->next)
 961                                 return NULL;
 962                 }
 963                 ftrace_pages = ftrace_pages->next;
 964         }
 965
 966         return &ftrace_pages->records[ftrace_pages->index++];
 967 }
 968
 969 static struct dyn_ftrace *
 970 ftrace_record_ip(unsigned long ip)
 971 {
 972         struct dyn_ftrace *rec;
 973
 974         if (ftrace_disabled)
 975                 return NULL;
 976
 977         rec = ftrace_alloc_dyn_node(ip);
 978         if (!rec)
 979                 return NULL;
 980
 981         rec->ip = ip;
 982         rec->newlist = ftrace_new_addrs;
 983         ftrace_new_addrs = rec;
 984
 985         return rec;
 986 }
 987
 988 static void print_ip_ins(const char *fmt, unsigned char *p)
 989 {
 990         int i;
 991
 992         printk(KERN_CONT "%s", fmt);
 993
 994         for (i = 0; i < MCOUNT_INSN_SIZE; i++)
 995                 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
 996 }
 997
 998 static void ftrace_bug(int failed, unsigned long ip)
 999 {
1000         switch (failed) {
1001         case -EFAULT:
1002                 FTRACE_WARN_ON_ONCE(1);
1003                 pr_info("ftrace faulted on modifying ");
1004                 print_ip_sym(ip);
1005                 break;
1006         case -EINVAL:
1007                 FTRACE_WARN_ON_ONCE(1);
1008                 pr_info("ftrace failed to modify ");
1009                 print_ip_sym(ip);
1010                 print_ip_ins(" actual: ", (unsigned char *)ip);
1011                 printk(KERN_CONT "\n");
1012                 break;
1013         case -EPERM:
1014                 FTRACE_WARN_ON_ONCE(1);
1015                 pr_info("ftrace faulted on writing ");
1016                 print_ip_sym(ip);
1017                 break;
1018         default:
1019                 FTRACE_WARN_ON_ONCE(1);
1020                 pr_info("ftrace faulted on unknown error ");
1021                 print_ip_sym(ip);
1022         }
1023 }
1024
1025
1026 /* Return 1 if the address range is reserved for ftrace */
1027 int ftrace_text_reserved(void *start, void *end)
1028 {
1029         struct dyn_ftrace *rec;
1030         struct ftrace_page *pg;
1031
1032         do_for_each_ftrace_rec(pg, rec) {
1033                 if (rec->ip <= (unsigned long)end &&
1034                     rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1035                         return 1;
1036         } while_for_each_ftrace_rec();
1037         return 0;
1038 }
1039
1040
1041 static int
1042 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1043 {
1044         unsigned long ftrace_addr;
1045         unsigned long flag = 0UL;
1046
1047         ftrace_addr = (unsigned long)FTRACE_ADDR;
1048
1049         /*
1050          * If this record is not to be traced or we want to disable it,
1051          * then disable it.
1052          *
1053          * If we want to enable it and filtering is off, then enable it.
1054          *
1055          * If we want to enable it and filtering is on, enable it only if
1056          * it's filtered
1057          */
1058         if (enable && !(rec->flags & FTRACE_FL_NOTRACE)) {
1059                 if (!ftrace_filtered || (rec->flags & FTRACE_FL_FILTER))
1060                         flag = FTRACE_FL_ENABLED;
1061         }
1062
1063         /* If the state of this record hasn't changed, then do nothing */
1064         if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1065                 return 0;
1066
1067         if (flag) {
1068                 rec->flags |= FTRACE_FL_ENABLED;
1069                 return ftrace_make_call(rec, ftrace_addr);
1070         }
1071
1072         rec->flags &= ~FTRACE_FL_ENABLED;
1073         return ftrace_make_nop(NULL, rec, ftrace_addr);
1074 }
1075
1076 static void ftrace_replace_code(int enable)
1077 {
1078         struct dyn_ftrace *rec;
1079         struct ftrace_page *pg;
1080         int failed;
1081
1082         do_for_each_ftrace_rec(pg, rec) {
1083                 /*
1084                  * Skip over free records, records that have
1085                  * failed and not converted.
1086                  */
1087                 if (rec->flags & FTRACE_FL_FREE ||
1088                     rec->flags & FTRACE_FL_FAILED ||
1089                     !(rec->flags & FTRACE_FL_CONVERTED))
1090                         continue;
1091
1092                 failed = __ftrace_replace_code(rec, enable);
1093                 if (failed) {
1094                         rec->flags |= FTRACE_FL_FAILED;
1095                         ftrace_bug(failed, rec->ip);
1096                         /* Stop processing */
1097                         return;
1098                 }
1099         } while_for_each_ftrace_rec();
1100 }
1101
1102 static int
1103 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1104 {
1105         unsigned long ip;
1106         int ret;
1107
1108         ip = rec->ip;
1109
1110         ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1111         if (ret) {
1112                 ftrace_bug(ret, ip);
1113                 rec->flags |= FTRACE_FL_FAILED;
1114                 return 0;
1115         }
1116         return 1;
1117 }
1118
1119 /*
1120  * archs can override this function if they must do something
1121  * before the modifying code is performed.
1122  */
1123 int __weak ftrace_arch_code_modify_prepare(void)
1124 {
1125         return 0;
1126 }
1127
1128 /*
1129  * archs can override this function if they must do something
1130  * after the modifying code is performed.
1131  */
1132 int __weak ftrace_arch_code_modify_post_process(void)
1133 {
1134         return 0;
1135 }
1136
1137 static int __ftrace_modify_code(void *data)
1138 {
1139         int *command = data;
1140
1141         if (*command & FTRACE_ENABLE_CALLS)
1142                 ftrace_replace_code(1);
1143         else if (*command & FTRACE_DISABLE_CALLS)
1144                 ftrace_replace_code(0);
1145
1146         if (*command & FTRACE_UPDATE_TRACE_FUNC)
1147                 ftrace_update_ftrace_func(ftrace_trace_function);
1148
1149         if (*command & FTRACE_START_FUNC_RET)
1150                 ftrace_enable_ftrace_graph_caller();
1151         else if (*command & FTRACE_STOP_FUNC_RET)
1152                 ftrace_disable_ftrace_graph_caller();
1153
1154         return 0;
1155 }
1156
1157 static void ftrace_run_update_code(int command)
1158 {
1159         int ret;
1160
1161         ret = ftrace_arch_code_modify_prepare();
1162         FTRACE_WARN_ON(ret);
1163         if (ret)
1164                 return;
1165
1166         stop_machine(__ftrace_modify_code, &command, NULL);
1167
1168         ret = ftrace_arch_code_modify_post_process();
1169         FTRACE_WARN_ON(ret);
1170 }
1171
1172 static ftrace_func_t saved_ftrace_func;
1173 static int ftrace_start_up;
1174
1175 static void ftrace_startup_enable(int command)
1176 {
1177         if (saved_ftrace_func != ftrace_trace_function) {
1178                 saved_ftrace_func = ftrace_trace_function;
1179                 command |= FTRACE_UPDATE_TRACE_FUNC;
1180         }
1181
1182         if (!command || !ftrace_enabled)
1183                 return;
1184
1185         ftrace_run_update_code(command);
1186 }
1187
1188 static void ftrace_startup(int command)
1189 {
1190         if (unlikely(ftrace_disabled))
1191                 return;
1192
1193         ftrace_start_up++;
1194         command |= FTRACE_ENABLE_CALLS;
1195
1196         ftrace_startup_enable(command);
1197 }
1198
1199 static void ftrace_shutdown(int command)
1200 {
1201         if (unlikely(ftrace_disabled))
1202                 return;
1203
1204         ftrace_start_up--;
1205         /*
1206          * Just warn in case of unbalance, no need to kill ftrace, it's not
1207          * critical but the ftrace_call callers may be never nopped again after
1208          * further ftrace uses.
1209          */
1210         WARN_ON_ONCE(ftrace_start_up < 0);
1211
1212         if (!ftrace_start_up)
1213                 command |= FTRACE_DISABLE_CALLS;
1214
1215         if (saved_ftrace_func != ftrace_trace_function) {
1216                 saved_ftrace_func = ftrace_trace_function;
1217                 command |= FTRACE_UPDATE_TRACE_FUNC;
1218         }
1219
1220         if (!command || !ftrace_enabled)
1221                 return;
1222
1223         ftrace_run_update_code(command);
1224 }
1225
1226 static void ftrace_startup_sysctl(void)
1227 {
1228         if (unlikely(ftrace_disabled))
1229                 return;
1230
1231         /* Force update next time */
1232         saved_ftrace_func = NULL;
1233         /* ftrace_start_up is true if we want ftrace running */
1234         if (ftrace_start_up)
1235                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1236 }
1237
1238 static void ftrace_shutdown_sysctl(void)
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                 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
1246 }
1247
1248 static cycle_t          ftrace_update_time;
1249 static unsigned long    ftrace_update_cnt;
1250 unsigned long           ftrace_update_tot_cnt;
1251
1252 static int ftrace_update_code(struct module *mod)
1253 {
1254         struct dyn_ftrace *p;
1255         cycle_t start, stop;
1256
1257         start = ftrace_now(raw_smp_processor_id());
1258         ftrace_update_cnt = 0;
1259
1260         while (ftrace_new_addrs) {
1261
1262                 /* If something went wrong, bail without enabling anything */
1263                 if (unlikely(ftrace_disabled))
1264                         return -1;
1265
1266                 p = ftrace_new_addrs;
1267                 ftrace_new_addrs = p->newlist;
1268                 p->flags = 0L;
1269
1270                 /*
1271                  * Do the initial record conversion from mcount jump
1272                  * to the NOP instructions.
1273                  */
1274                 if (!ftrace_code_disable(mod, p)) {
1275                         ftrace_free_rec(p);
1276                         continue;
1277                 }
1278
1279                 p->flags |= FTRACE_FL_CONVERTED;
1280                 ftrace_update_cnt++;
1281
1282                 /*
1283                  * If the tracing is enabled, go ahead and enable the record.
1284                  *
1285                  * The reason not to enable the record immediatelly is the
1286                  * inherent check of ftrace_make_nop/ftrace_make_call for
1287                  * correct previous instructions.  Making first the NOP
1288                  * conversion puts the module to the correct state, thus
1289                  * passing the ftrace_make_call check.
1290                  */
1291                 if (ftrace_start_up) {
1292                         int failed = __ftrace_replace_code(p, 1);
1293                         if (failed) {
1294                                 ftrace_bug(failed, p->ip);
1295                                 ftrace_free_rec(p);
1296                         }
1297                 }
1298         }
1299
1300         stop = ftrace_now(raw_smp_processor_id());
1301         ftrace_update_time = stop - start;
1302         ftrace_update_tot_cnt += ftrace_update_cnt;
1303
1304         return 0;
1305 }
1306
1307 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1308 {
1309         struct ftrace_page *pg;
1310         int cnt;
1311         int i;
1312
1313         /* allocate a few pages */
1314         ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1315         if (!ftrace_pages_start)
1316                 return -1;
1317
1318         /*
1319          * Allocate a few more pages.
1320          *
1321          * TODO: have some parser search vmlinux before
1322          *   final linking to find all calls to ftrace.
1323          *   Then we can:
1324          *    a) know how many pages to allocate.
1325          *     and/or
1326          *    b) set up the table then.
1327          *
1328          *  The dynamic code is still necessary for
1329          *  modules.
1330          */
1331
1332         pg = ftrace_pages = ftrace_pages_start;
1333
1334         cnt = num_to_init / ENTRIES_PER_PAGE;
1335         pr_info("ftrace: allocating %ld entries in %d pages\n",
1336                 num_to_init, cnt + 1);
1337
1338         for (i = 0; i < cnt; i++) {
1339                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1340
1341                 /* If we fail, we'll try later anyway */
1342                 if (!pg->next)
1343                         break;
1344
1345                 pg = pg->next;
1346         }
1347
1348         return 0;
1349 }
1350
1351 enum {
1352         FTRACE_ITER_FILTER      = (1 << 0),
1353         FTRACE_ITER_NOTRACE     = (1 << 1),
1354         FTRACE_ITER_FAILURES    = (1 << 2),
1355         FTRACE_ITER_PRINTALL    = (1 << 3),
1356         FTRACE_ITER_HASH        = (1 << 4),
1357 };
1358
1359 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1360
1361 struct ftrace_iterator {
1362         loff_t                          pos;
1363         loff_t                          func_pos;
1364         struct ftrace_page              *pg;
1365         struct dyn_ftrace               *func;
1366         struct ftrace_func_probe        *probe;
1367         struct trace_parser             parser;
1368         int                             hidx;
1369         int                             idx;
1370         unsigned                        flags;
1371 };
1372
1373 static void *
1374 t_hash_next(struct seq_file *m, loff_t *pos)
1375 {
1376         struct ftrace_iterator *iter = m->private;
1377         struct hlist_node *hnd = NULL;
1378         struct hlist_head *hhd;
1379
1380         (*pos)++;
1381         iter->pos = *pos;
1382
1383         if (iter->probe)
1384                 hnd = &iter->probe->node;
1385  retry:
1386         if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1387                 return NULL;
1388
1389         hhd = &ftrace_func_hash[iter->hidx];
1390
1391         if (hlist_empty(hhd)) {
1392                 iter->hidx++;
1393                 hnd = NULL;
1394                 goto retry;
1395         }
1396
1397         if (!hnd)
1398                 hnd = hhd->first;
1399         else {
1400                 hnd = hnd->next;
1401                 if (!hnd) {
1402                         iter->hidx++;
1403                         goto retry;
1404                 }
1405         }
1406
1407         if (WARN_ON_ONCE(!hnd))
1408                 return NULL;
1409
1410         iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
1411
1412         return iter;
1413 }
1414
1415 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1416 {
1417         struct ftrace_iterator *iter = m->private;
1418         void *p = NULL;
1419         loff_t l;
1420
1421         if (iter->func_pos > *pos)
1422                 return NULL;
1423
1424         iter->hidx = 0;
1425         for (l = 0; l <= (*pos - iter->func_pos); ) {
1426                 p = t_hash_next(m, &l);
1427                 if (!p)
1428                         break;
1429         }
1430         if (!p)
1431                 return NULL;
1432
1433         /* Only set this if we have an item */
1434         iter->flags |= FTRACE_ITER_HASH;
1435
1436         return iter;
1437 }
1438
1439 static int
1440 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
1441 {
1442         struct ftrace_func_probe *rec;
1443
1444         rec = iter->probe;
1445         if (WARN_ON_ONCE(!rec))
1446                 return -EIO;
1447
1448         if (rec->ops->print)
1449                 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1450
1451         seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
1452
1453         if (rec->data)
1454                 seq_printf(m, ":%p", rec->data);
1455         seq_putc(m, '\n');
1456
1457         return 0;
1458 }
1459
1460 static void *
1461 t_next(struct seq_file *m, void *v, loff_t *pos)
1462 {
1463         struct ftrace_iterator *iter = m->private;
1464         struct dyn_ftrace *rec = NULL;
1465
1466         if (iter->flags & FTRACE_ITER_HASH)
1467                 return t_hash_next(m, pos);
1468
1469         (*pos)++;
1470         iter->pos = iter->func_pos = *pos;
1471
1472         if (iter->flags & FTRACE_ITER_PRINTALL)
1473                 return t_hash_start(m, pos);
1474
1475  retry:
1476         if (iter->idx >= iter->pg->index) {
1477                 if (iter->pg->next) {
1478                         iter->pg = iter->pg->next;
1479                         iter->idx = 0;
1480                         goto retry;
1481                 }
1482         } else {
1483                 rec = &iter->pg->records[iter->idx++];
1484                 if ((rec->flags & FTRACE_FL_FREE) ||
1485
1486                     (!(iter->flags & FTRACE_ITER_FAILURES) &&
1487                      (rec->flags & FTRACE_FL_FAILED)) ||
1488
1489                     ((iter->flags & FTRACE_ITER_FAILURES) &&
1490                      !(rec->flags & FTRACE_FL_FAILED)) ||
1491
1492                     ((iter->flags & FTRACE_ITER_FILTER) &&
1493                      !(rec->flags & FTRACE_FL_FILTER)) ||
1494
1495                     ((iter->flags & FTRACE_ITER_NOTRACE) &&
1496                      !(rec->flags & FTRACE_FL_NOTRACE))) {
1497                         rec = NULL;
1498                         goto retry;
1499                 }
1500         }
1501
1502         if (!rec)
1503                 return t_hash_start(m, pos);
1504
1505         iter->func = rec;
1506
1507         return iter;
1508 }
1509
1510 static void reset_iter_read(struct ftrace_iterator *iter)
1511 {
1512         iter->pos = 0;
1513         iter->func_pos = 0;
1514         iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
1515 }
1516
1517 static void *t_start(struct seq_file *m, loff_t *pos)
1518 {
1519         struct ftrace_iterator *iter = m->private;
1520         void *p = NULL;
1521         loff_t l;
1522
1523         mutex_lock(&ftrace_lock);
1524         /*
1525          * If an lseek was done, then reset and start from beginning.
1526          */
1527         if (*pos < iter->pos)
1528                 reset_iter_read(iter);
1529
1530         /*
1531          * For set_ftrace_filter reading, if we have the filter
1532          * off, we can short cut and just print out that all
1533          * functions are enabled.
1534          */
1535         if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1536                 if (*pos > 0)
1537                         return t_hash_start(m, pos);
1538                 iter->flags |= FTRACE_ITER_PRINTALL;
1539                 /* reset in case of seek/pread */
1540                 iter->flags &= ~FTRACE_ITER_HASH;
1541                 return iter;
1542         }
1543
1544         if (iter->flags & FTRACE_ITER_HASH)
1545                 return t_hash_start(m, pos);
1546
1547         /*
1548          * Unfortunately, we need to restart at ftrace_pages_start
1549          * every time we let go of the ftrace_mutex. This is because
1550          * those pointers can change without the lock.
1551          */
1552         iter->pg = ftrace_pages_start;
1553         iter->idx = 0;
1554         for (l = 0; l <= *pos; ) {
1555                 p = t_next(m, p, &l);
1556                 if (!p)
1557                         break;
1558         }
1559
1560         if (!p) {
1561                 if (iter->flags & FTRACE_ITER_FILTER)
1562                         return t_hash_start(m, pos);
1563
1564                 return NULL;
1565         }
1566
1567         return iter;
1568 }
1569
1570 static void t_stop(struct seq_file *m, void *p)
1571 {
1572         mutex_unlock(&ftrace_lock);
1573 }
1574
1575 static int t_show(struct seq_file *m, void *v)
1576 {
1577         struct ftrace_iterator *iter = m->private;
1578         struct dyn_ftrace *rec;
1579
1580         if (iter->flags & FTRACE_ITER_HASH)
1581                 return t_hash_show(m, iter);
1582
1583         if (iter->flags & FTRACE_ITER_PRINTALL) {
1584                 seq_printf(m, "#### all functions enabled ####\n");
1585                 return 0;
1586         }
1587
1588         rec = iter->func;
1589
1590         if (!rec)
1591                 return 0;
1592
1593         seq_printf(m, "%ps\n", (void *)rec->ip);
1594
1595         return 0;
1596 }
1597
1598 static const struct seq_operations show_ftrace_seq_ops = {
1599         .start = t_start,
1600         .next = t_next,
1601         .stop = t_stop,
1602         .show = t_show,
1603 };
1604
1605 static int
1606 ftrace_avail_open(struct inode *inode, struct file *file)
1607 {
1608         struct ftrace_iterator *iter;
1609         int ret;
1610
1611         if (unlikely(ftrace_disabled))
1612                 return -ENODEV;
1613
1614         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1615         if (!iter)
1616                 return -ENOMEM;
1617
1618         iter->pg = ftrace_pages_start;
1619
1620         ret = seq_open(file, &show_ftrace_seq_ops);
1621         if (!ret) {
1622                 struct seq_file *m = file->private_data;
1623
1624                 m->private = iter;
1625         } else {
1626                 kfree(iter);
1627         }
1628
1629         return ret;
1630 }
1631
1632 static int
1633 ftrace_failures_open(struct inode *inode, struct file *file)
1634 {
1635         int ret;
1636         struct seq_file *m;
1637         struct ftrace_iterator *iter;
1638
1639         ret = ftrace_avail_open(inode, file);
1640         if (!ret) {
1641                 m = file->private_data;
1642                 iter = m->private;
1643                 iter->flags = FTRACE_ITER_FAILURES;
1644         }
1645
1646         return ret;
1647 }
1648
1649
1650 static void ftrace_filter_reset(int enable)
1651 {
1652         struct ftrace_page *pg;
1653         struct dyn_ftrace *rec;
1654         unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1655
1656         mutex_lock(&ftrace_lock);
1657         if (enable)
1658                 ftrace_filtered = 0;
1659         do_for_each_ftrace_rec(pg, rec) {
1660                 if (rec->flags & FTRACE_FL_FAILED)
1661                         continue;
1662                 rec->flags &= ~type;
1663         } while_for_each_ftrace_rec();
1664         mutex_unlock(&ftrace_lock);
1665 }
1666
1667 static int
1668 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1669 {
1670         struct ftrace_iterator *iter;
1671         int ret = 0;
1672
1673         if (unlikely(ftrace_disabled))
1674                 return -ENODEV;
1675
1676         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1677         if (!iter)
1678                 return -ENOMEM;
1679
1680         if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
1681                 kfree(iter);
1682                 return -ENOMEM;
1683         }
1684
1685         mutex_lock(&ftrace_regex_lock);
1686         if ((file->f_mode & FMODE_WRITE) &&
1687             (file->f_flags & O_TRUNC))
1688                 ftrace_filter_reset(enable);
1689
1690         if (file->f_mode & FMODE_READ) {
1691                 iter->pg = ftrace_pages_start;
1692                 iter->flags = enable ? FTRACE_ITER_FILTER :
1693                         FTRACE_ITER_NOTRACE;
1694
1695                 ret = seq_open(file, &show_ftrace_seq_ops);
1696                 if (!ret) {
1697                         struct seq_file *m = file->private_data;
1698                         m->private = iter;
1699                 } else {
1700                         trace_parser_put(&iter->parser);
1701                         kfree(iter);
1702                 }
1703         } else
1704                 file->private_data = iter;
1705         mutex_unlock(&ftrace_regex_lock);
1706
1707         return ret;
1708 }
1709
1710 static int
1711 ftrace_filter_open(struct inode *inode, struct file *file)
1712 {
1713         return ftrace_regex_open(inode, file, 1);
1714 }
1715
1716 static int
1717 ftrace_notrace_open(struct inode *inode, struct file *file)
1718 {
1719         return ftrace_regex_open(inode, file, 0);
1720 }
1721
1722 static loff_t
1723 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1724 {
1725         loff_t ret;
1726
1727         if (file->f_mode & FMODE_READ)
1728                 ret = seq_lseek(file, offset, origin);
1729         else
1730                 file->f_pos = ret = 1;
1731
1732         return ret;
1733 }
1734
1735 static int ftrace_match(char *str, char *regex, int len, int type)
1736 {
1737         int matched = 0;
1738         int slen;
1739
1740         switch (type) {
1741         case MATCH_FULL:
1742                 if (strcmp(str, regex) == 0)
1743                         matched = 1;
1744                 break;
1745         case MATCH_FRONT_ONLY:
1746                 if (strncmp(str, regex, len) == 0)
1747                         matched = 1;
1748                 break;
1749         case MATCH_MIDDLE_ONLY:
1750                 if (strstr(str, regex))
1751                         matched = 1;
1752                 break;
1753         case MATCH_END_ONLY:
1754                 slen = strlen(str);
1755                 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
1756                         matched = 1;
1757                 break;
1758         }
1759
1760         return matched;
1761 }
1762
1763 static int
1764 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1765 {
1766         char str[KSYM_SYMBOL_LEN];
1767
1768         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1769         return ftrace_match(str, regex, len, type);
1770 }
1771
1772 static int ftrace_match_records(char *buff, int len, int enable)
1773 {
1774         unsigned int search_len;
1775         struct ftrace_page *pg;
1776         struct dyn_ftrace *rec;
1777         unsigned long flag;
1778         char *search;
1779         int type;
1780         int not;
1781         int found = 0;
1782
1783         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1784         type = filter_parse_regex(buff, len, &search, &not);
1785
1786         search_len = strlen(search);
1787
1788         mutex_lock(&ftrace_lock);
1789         do_for_each_ftrace_rec(pg, rec) {
1790
1791                 if (rec->flags & FTRACE_FL_FAILED)
1792                         continue;
1793
1794                 if (ftrace_match_record(rec, search, search_len, type)) {
1795                         if (not)
1796                                 rec->flags &= ~flag;
1797                         else
1798                                 rec->flags |= flag;
1799                         found = 1;
1800                 }
1801                 /*
1802                  * Only enable filtering if we have a function that
1803                  * is filtered on.
1804                  */
1805                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1806                         ftrace_filtered = 1;
1807         } while_for_each_ftrace_rec();
1808         mutex_unlock(&ftrace_lock);
1809
1810         return found;
1811 }
1812
1813 static int
1814 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1815                            char *regex, int len, int type)
1816 {
1817         char str[KSYM_SYMBOL_LEN];
1818         char *modname;
1819
1820         kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1821
1822         if (!modname || strcmp(modname, mod))
1823                 return 0;
1824
1825         /* blank search means to match all funcs in the mod */
1826         if (len)
1827                 return ftrace_match(str, regex, len, type);
1828         else
1829                 return 1;
1830 }
1831
1832 static int ftrace_match_module_records(char *buff, char *mod, int enable)
1833 {
1834         unsigned search_len = 0;
1835         struct ftrace_page *pg;
1836         struct dyn_ftrace *rec;
1837         int type = MATCH_FULL;
1838         char *search = buff;
1839         unsigned long flag;
1840         int not = 0;
1841         int found = 0;
1842
1843         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1844
1845         /* blank or '*' mean the same */
1846         if (strcmp(buff, "*") == 0)
1847                 buff[0] = 0;
1848
1849         /* handle the case of 'dont filter this module' */
1850         if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1851                 buff[0] = 0;
1852                 not = 1;
1853         }
1854
1855         if (strlen(buff)) {
1856                 type = filter_parse_regex(buff, strlen(buff), &search, &not);
1857                 search_len = strlen(search);
1858         }
1859
1860         mutex_lock(&ftrace_lock);
1861         do_for_each_ftrace_rec(pg, rec) {
1862
1863                 if (rec->flags & FTRACE_FL_FAILED)
1864                         continue;
1865
1866                 if (ftrace_match_module_record(rec, mod,
1867                                                search, search_len, type)) {
1868                         if (not)
1869                                 rec->flags &= ~flag;
1870                         else
1871                                 rec->flags |= flag;
1872                         found = 1;
1873                 }
1874                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1875                         ftrace_filtered = 1;
1876
1877         } while_for_each_ftrace_rec();
1878         mutex_unlock(&ftrace_lock);
1879
1880         return found;
1881 }
1882
1883 /*
1884  * We register the module command as a template to show others how
1885  * to register the a command as well.
1886  */
1887
1888 static int
1889 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1890 {
1891         char *mod;
1892
1893         /*
1894          * cmd == 'mod' because we only registered this func
1895          * for the 'mod' ftrace_func_command.
1896          * But if you register one func with multiple commands,
1897          * you can tell which command was used by the cmd
1898          * parameter.
1899          */
1900
1901         /* we must have a module name */
1902         if (!param)
1903                 return -EINVAL;
1904
1905         mod = strsep(&param, ":");
1906         if (!strlen(mod))
1907                 return -EINVAL;
1908
1909         if (ftrace_match_module_records(func, mod, enable))
1910                 return 0;
1911         return -EINVAL;
1912 }
1913
1914 static struct ftrace_func_command ftrace_mod_cmd = {
1915         .name                   = "mod",
1916         .func                   = ftrace_mod_callback,
1917 };
1918
1919 static int __init ftrace_mod_cmd_init(void)
1920 {
1921         return register_ftrace_command(&ftrace_mod_cmd);
1922 }
1923 device_initcall(ftrace_mod_cmd_init);
1924
1925 static void
1926 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1927 {
1928         struct ftrace_func_probe *entry;
1929         struct hlist_head *hhd;
1930         struct hlist_node *n;
1931         unsigned long key;
1932
1933         key = hash_long(ip, FTRACE_HASH_BITS);
1934
1935         hhd = &ftrace_func_hash[key];
1936
1937         if (hlist_empty(hhd))
1938                 return;
1939
1940         /*
1941          * Disable preemption for these calls to prevent a RCU grace
1942          * period. This syncs the hash iteration and freeing of items
1943          * on the hash. rcu_read_lock is too dangerous here.
1944          */
1945         preempt_disable_notrace();
1946         hlist_for_each_entry_rcu(entry, n, hhd, node) {
1947                 if (entry->ip == ip)
1948                         entry->ops->func(ip, parent_ip, &entry->data);
1949         }
1950         preempt_enable_notrace();
1951 }
1952
1953 static struct ftrace_ops trace_probe_ops __read_mostly =
1954 {
1955         .func           = function_trace_probe_call,
1956 };
1957
1958 static int ftrace_probe_registered;
1959
1960 static void __enable_ftrace_function_probe(void)
1961 {
1962         int i;
1963
1964         if (ftrace_probe_registered)
1965                 return;
1966
1967         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1968                 struct hlist_head *hhd = &ftrace_func_hash[i];
1969                 if (hhd->first)
1970                         break;
1971         }
1972         /* Nothing registered? */
1973         if (i == FTRACE_FUNC_HASHSIZE)
1974                 return;
1975
1976         __register_ftrace_function(&trace_probe_ops);
1977         ftrace_startup(0);
1978         ftrace_probe_registered = 1;
1979 }
1980
1981 static void __disable_ftrace_function_probe(void)
1982 {
1983         int i;
1984
1985         if (!ftrace_probe_registered)
1986                 return;
1987
1988         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1989                 struct hlist_head *hhd = &ftrace_func_hash[i];
1990                 if (hhd->first)
1991                         return;
1992         }
1993
1994         /* no more funcs left */
1995         __unregister_ftrace_function(&trace_probe_ops);
1996         ftrace_shutdown(0);
1997         ftrace_probe_registered = 0;
1998 }
1999
2000
2001 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2002 {
2003         struct ftrace_func_probe *entry =
2004                 container_of(rhp, struct ftrace_func_probe, rcu);
2005
2006         if (entry->ops->free)
2007                 entry->ops->free(&entry->data);
2008         kfree(entry);
2009 }
2010
2011
2012 int
2013 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2014                               void *data)
2015 {
2016         struct ftrace_func_probe *entry;
2017         struct ftrace_page *pg;
2018         struct dyn_ftrace *rec;
2019         int type, len, not;
2020         unsigned long key;
2021         int count = 0;
2022         char *search;
2023
2024         type = filter_parse_regex(glob, strlen(glob), &search, &not);
2025         len = strlen(search);
2026
2027         /* we do not support '!' for function probes */
2028         if (WARN_ON(not))
2029                 return -EINVAL;
2030
2031         mutex_lock(&ftrace_lock);
2032         do_for_each_ftrace_rec(pg, rec) {
2033
2034                 if (rec->flags & FTRACE_FL_FAILED)
2035                         continue;
2036
2037                 if (!ftrace_match_record(rec, search, len, type))
2038                         continue;
2039
2040                 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2041                 if (!entry) {
2042                         /* If we did not process any, then return error */
2043                         if (!count)
2044                                 count = -ENOMEM;
2045                         goto out_unlock;
2046                 }
2047
2048                 count++;
2049
2050                 entry->data = data;
2051
2052                 /*
2053                  * The caller might want to do something special
2054                  * for each function we find. We call the callback
2055                  * to give the caller an opportunity to do so.
2056                  */
2057                 if (ops->callback) {
2058                         if (ops->callback(rec->ip, &entry->data) < 0) {
2059                                 /* caller does not like this func */
2060                                 kfree(entry);
2061                                 continue;
2062                         }
2063                 }
2064
2065                 entry->ops = ops;
2066                 entry->ip = rec->ip;
2067
2068                 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2069                 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2070
2071         } while_for_each_ftrace_rec();
2072         __enable_ftrace_function_probe();
2073
2074  out_unlock:
2075         mutex_unlock(&ftrace_lock);
2076
2077         return count;
2078 }
2079
2080 enum {
2081         PROBE_TEST_FUNC         = 1,
2082         PROBE_TEST_DATA         = 2
2083 };
2084
2085 static void
2086 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2087                                   void *data, int flags)
2088 {
2089         struct ftrace_func_probe *entry;
2090         struct hlist_node *n, *tmp;
2091         char str[KSYM_SYMBOL_LEN];
2092         int type = MATCH_FULL;
2093         int i, len = 0;
2094         char *search;
2095
2096         if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2097                 glob = NULL;
2098         else if (glob) {
2099                 int not;
2100
2101                 type = filter_parse_regex(glob, strlen(glob), &search, &not);
2102                 len = strlen(search);
2103
2104                 /* we do not support '!' for function probes */
2105                 if (WARN_ON(not))
2106                         return;
2107         }
2108
2109         mutex_lock(&ftrace_lock);
2110         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2111                 struct hlist_head *hhd = &ftrace_func_hash[i];
2112
2113                 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2114
2115                         /* break up if statements for readability */
2116                         if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2117                                 continue;
2118
2119                         if ((flags & PROBE_TEST_DATA) && entry->data != data)
2120                                 continue;
2121
2122                         /* do this last, since it is the most expensive */
2123                         if (glob) {
2124                                 kallsyms_lookup(entry->ip, NULL, NULL,
2125                                                 NULL, str);
2126                                 if (!ftrace_match(str, glob, len, type))
2127                                         continue;
2128                         }
2129
2130                         hlist_del(&entry->node);
2131                         call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2132                 }
2133         }
2134         __disable_ftrace_function_probe();
2135         mutex_unlock(&ftrace_lock);
2136 }
2137
2138 void
2139 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2140                                 void *data)
2141 {
2142         __unregister_ftrace_function_probe(glob, ops, data,
2143                                           PROBE_TEST_FUNC | PROBE_TEST_DATA);
2144 }
2145
2146 void
2147 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2148 {
2149         __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2150 }
2151
2152 void unregister_ftrace_function_probe_all(char *glob)
2153 {
2154         __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2155 }
2156
2157 static LIST_HEAD(ftrace_commands);
2158 static DEFINE_MUTEX(ftrace_cmd_mutex);
2159
2160 int register_ftrace_command(struct ftrace_func_command *cmd)
2161 {
2162         struct ftrace_func_command *p;
2163         int ret = 0;
2164
2165         mutex_lock(&ftrace_cmd_mutex);
2166         list_for_each_entry(p, &ftrace_commands, list) {
2167                 if (strcmp(cmd->name, p->name) == 0) {
2168                         ret = -EBUSY;
2169                         goto out_unlock;
2170                 }
2171         }
2172         list_add(&cmd->list, &ftrace_commands);
2173  out_unlock:
2174         mutex_unlock(&ftrace_cmd_mutex);
2175
2176         return ret;
2177 }
2178
2179 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2180 {
2181         struct ftrace_func_command *p, *n;
2182         int ret = -ENODEV;
2183
2184         mutex_lock(&ftrace_cmd_mutex);
2185         list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2186                 if (strcmp(cmd->name, p->name) == 0) {
2187                         ret = 0;
2188                         list_del_init(&p->list);
2189                         goto out_unlock;
2190                 }
2191         }
2192  out_unlock:
2193         mutex_unlock(&ftrace_cmd_mutex);
2194
2195         return ret;
2196 }
2197
2198 static int ftrace_process_regex(char *buff, int len, int enable)
2199 {
2200         char *func, *command, *next = buff;
2201         struct ftrace_func_command *p;
2202         int ret = -EINVAL;
2203
2204         func = strsep(&next, ":");
2205
2206         if (!next) {
2207                 if (ftrace_match_records(func, len, enable))
2208                         return 0;
2209                 return ret;
2210         }
2211
2212         /* command found */
2213
2214         command = strsep(&next, ":");
2215
2216         mutex_lock(&ftrace_cmd_mutex);
2217         list_for_each_entry(p, &ftrace_commands, list) {
2218                 if (strcmp(p->name, command) == 0) {
2219                         ret = p->func(func, command, next, enable);
2220                         goto out_unlock;
2221                 }
2222         }
2223  out_unlock:
2224         mutex_unlock(&ftrace_cmd_mutex);
2225
2226         return ret;
2227 }
2228
2229 static ssize_t
2230 ftrace_regex_write(struct file *file, const char __user *ubuf,
2231                    size_t cnt, loff_t *ppos, int enable)
2232 {
2233         struct ftrace_iterator *iter;
2234         struct trace_parser *parser;
2235         ssize_t ret, read;
2236
2237         if (!cnt)
2238                 return 0;
2239
2240         mutex_lock(&ftrace_regex_lock);
2241
2242         if (file->f_mode & FMODE_READ) {
2243                 struct seq_file *m = file->private_data;
2244                 iter = m->private;
2245         } else
2246                 iter = file->private_data;
2247
2248         parser = &iter->parser;
2249         read = trace_get_user(parser, ubuf, cnt, ppos);
2250
2251         if (read >= 0 && trace_parser_loaded(parser) &&
2252             !trace_parser_cont(parser)) {
2253                 ret = ftrace_process_regex(parser->buffer,
2254                                            parser->idx, enable);
2255                 trace_parser_clear(parser);
2256                 if (ret)
2257                         goto out_unlock;
2258         }
2259
2260         ret = read;
2261 out_unlock:
2262         mutex_unlock(&ftrace_regex_lock);
2263
2264         return ret;
2265 }
2266
2267 static ssize_t
2268 ftrace_filter_write(struct file *file, const char __user *ubuf,
2269                     size_t cnt, loff_t *ppos)
2270 {
2271         return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2272 }
2273
2274 static ssize_t
2275 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2276                      size_t cnt, loff_t *ppos)
2277 {
2278         return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2279 }
2280
2281 static void
2282 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2283 {
2284         if (unlikely(ftrace_disabled))
2285                 return;
2286
2287         mutex_lock(&ftrace_regex_lock);
2288         if (reset)
2289                 ftrace_filter_reset(enable);
2290         if (buf)
2291                 ftrace_match_records(buf, len, enable);
2292         mutex_unlock(&ftrace_regex_lock);
2293 }
2294
2295 /**
2296  * ftrace_set_filter - set a function to filter on in ftrace
2297  * @buf - the string that holds the function filter text.
2298  * @len - the length of the string.
2299  * @reset - non zero to reset all filters before applying this filter.
2300  *
2301  * Filters denote which functions should be enabled when tracing is enabled.
2302  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2303  */
2304 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2305 {
2306         ftrace_set_regex(buf, len, reset, 1);
2307 }
2308
2309 /**
2310  * ftrace_set_notrace - set a function to not trace in ftrace
2311  * @buf - the string that holds the function notrace text.
2312  * @len - the length of the string.
2313  * @reset - non zero to reset all filters before applying this filter.
2314  *
2315  * Notrace Filters denote which functions should not be enabled when tracing
2316  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2317  * for tracing.
2318  */
2319 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2320 {
2321         ftrace_set_regex(buf, len, reset, 0);
2322 }
2323
2324 /*
2325  * command line interface to allow users to set filters on boot up.
2326  */
2327 #define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
2328 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2329 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2330
2331 static int __init set_ftrace_notrace(char *str)
2332 {
2333         strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2334         return 1;
2335 }
2336 __setup("ftrace_notrace=", set_ftrace_notrace);
2337
2338 static int __init set_ftrace_filter(char *str)
2339 {
2340         strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2341         return 1;
2342 }
2343 __setup("ftrace_filter=", set_ftrace_filter);
2344
2345 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2346 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
2347 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
2348
2349 static int __init set_graph_function(char *str)
2350 {
2351         strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
2352         return 1;
2353 }
2354 __setup("ftrace_graph_filter=", set_graph_function);
2355
2356 static void __init set_ftrace_early_graph(char *buf)
2357 {
2358         int ret;
2359         char *func;
2360
2361         while (buf) {
2362                 func = strsep(&buf, ",");
2363                 /* we allow only one expression at a time */
2364                 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2365                                       func);
2366                 if (ret)
2367                         printk(KERN_DEBUG "ftrace: function %s not "
2368                                           "traceable\n", func);
2369         }
2370 }
2371 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2372
2373 static void __init set_ftrace_early_filter(char *buf, int enable)
2374 {
2375         char *func;
2376
2377         while (buf) {
2378                 func = strsep(&buf, ",");
2379                 ftrace_set_regex(func, strlen(func), 0, enable);
2380         }
2381 }
2382
2383 static void __init set_ftrace_early_filters(void)
2384 {
2385         if (ftrace_filter_buf[0])
2386                 set_ftrace_early_filter(ftrace_filter_buf, 1);
2387         if (ftrace_notrace_buf[0])
2388                 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2389 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2390         if (ftrace_graph_buf[0])
2391                 set_ftrace_early_graph(ftrace_graph_buf);
2392 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2393 }
2394
2395 static int
2396 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2397 {
2398         struct seq_file *m = (struct seq_file *)file->private_data;
2399         struct ftrace_iterator *iter;
2400         struct trace_parser *parser;
2401
2402         mutex_lock(&ftrace_regex_lock);
2403         if (file->f_mode & FMODE_READ) {
2404                 iter = m->private;
2405
2406                 seq_release(inode, file);
2407         } else
2408                 iter = file->private_data;
2409
2410         parser = &iter->parser;
2411         if (trace_parser_loaded(parser)) {
2412                 parser->buffer[parser->idx] = 0;
2413                 ftrace_match_records(parser->buffer, parser->idx, enable);
2414         }
2415
2416         mutex_lock(&ftrace_lock);
2417         if (ftrace_start_up && ftrace_enabled)
2418                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2419         mutex_unlock(&ftrace_lock);
2420
2421         trace_parser_put(parser);
2422         kfree(iter);
2423
2424         mutex_unlock(&ftrace_regex_lock);
2425         return 0;
2426 }
2427
2428 static int
2429 ftrace_filter_release(struct inode *inode, struct file *file)
2430 {
2431         return ftrace_regex_release(inode, file, 1);
2432 }
2433
2434 static int
2435 ftrace_notrace_release(struct inode *inode, struct file *file)
2436 {
2437         return ftrace_regex_release(inode, file, 0);
2438 }
2439
2440 static const struct file_operations ftrace_avail_fops = {
2441         .open = ftrace_avail_open,
2442         .read = seq_read,
2443         .llseek = seq_lseek,
2444         .release = seq_release_private,
2445 };
2446
2447 static const struct file_operations ftrace_failures_fops = {
2448         .open = ftrace_failures_open,
2449         .read = seq_read,
2450         .llseek = seq_lseek,
2451         .release = seq_release_private,
2452 };
2453
2454 static const struct file_operations ftrace_filter_fops = {
2455         .open = ftrace_filter_open,
2456         .read = seq_read,
2457         .write = ftrace_filter_write,
2458         .llseek = ftrace_regex_lseek,
2459         .release = ftrace_filter_release,
2460 };
2461
2462 static const struct file_operations ftrace_notrace_fops = {
2463         .open = ftrace_notrace_open,
2464         .read = seq_read,
2465         .write = ftrace_notrace_write,
2466         .llseek = ftrace_regex_lseek,
2467         .release = ftrace_notrace_release,
2468 };
2469
2470 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2471
2472 static DEFINE_MUTEX(graph_lock);
2473
2474 int ftrace_graph_count;
2475 int ftrace_graph_filter_enabled;
2476 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2477
2478 static void *
2479 __g_next(struct seq_file *m, loff_t *pos)
2480 {
2481         if (*pos >= ftrace_graph_count)
2482                 return NULL;
2483         return &ftrace_graph_funcs[*pos];
2484 }
2485
2486 static void *
2487 g_next(struct seq_file *m, void *v, loff_t *pos)
2488 {
2489         (*pos)++;
2490         return __g_next(m, pos);
2491 }
2492
2493 static void *g_start(struct seq_file *m, loff_t *pos)
2494 {
2495         mutex_lock(&graph_lock);
2496
2497         /* Nothing, tell g_show to print all functions are enabled */
2498         if (!ftrace_graph_filter_enabled && !*pos)
2499                 return (void *)1;
2500
2501         return __g_next(m, pos);
2502 }
2503
2504 static void g_stop(struct seq_file *m, void *p)
2505 {
2506         mutex_unlock(&graph_lock);
2507 }
2508
2509 static int g_show(struct seq_file *m, void *v)
2510 {
2511         unsigned long *ptr = v;
2512
2513         if (!ptr)
2514                 return 0;
2515
2516         if (ptr == (unsigned long *)1) {
2517                 seq_printf(m, "#### all functions enabled ####\n");
2518                 return 0;
2519         }
2520
2521         seq_printf(m, "%ps\n", (void *)*ptr);
2522
2523         return 0;
2524 }
2525
2526 static const struct seq_operations ftrace_graph_seq_ops = {
2527         .start = g_start,
2528         .next = g_next,
2529         .stop = g_stop,
2530         .show = g_show,
2531 };
2532
2533 static int
2534 ftrace_graph_open(struct inode *inode, struct file *file)
2535 {
2536         int ret = 0;
2537
2538         if (unlikely(ftrace_disabled))
2539                 return -ENODEV;
2540
2541         mutex_lock(&graph_lock);
2542         if ((file->f_mode & FMODE_WRITE) &&
2543             (file->f_flags & O_TRUNC)) {
2544                 ftrace_graph_filter_enabled = 0;
2545                 ftrace_graph_count = 0;
2546                 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2547         }
2548         mutex_unlock(&graph_lock);
2549
2550         if (file->f_mode & FMODE_READ)
2551                 ret = seq_open(file, &ftrace_graph_seq_ops);
2552
2553         return ret;
2554 }
2555
2556 static int
2557 ftrace_graph_release(struct inode *inode, struct file *file)
2558 {
2559         if (file->f_mode & FMODE_READ)
2560                 seq_release(inode, file);
2561         return 0;
2562 }
2563
2564 static int
2565 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2566 {
2567         struct dyn_ftrace *rec;
2568         struct ftrace_page *pg;
2569         int search_len;
2570         int fail = 1;
2571         int type, not;
2572         char *search;
2573         bool exists;
2574         int i;
2575
2576         if (ftrace_disabled)
2577                 return -ENODEV;
2578
2579         /* decode regex */
2580         type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
2581         if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
2582                 return -EBUSY;
2583
2584         search_len = strlen(search);
2585
2586         mutex_lock(&ftrace_lock);
2587         do_for_each_ftrace_rec(pg, rec) {
2588
2589                 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2590                         continue;
2591
2592                 if (ftrace_match_record(rec, search, search_len, type)) {
2593                         /* if it is in the array */
2594                         exists = false;
2595                         for (i = 0; i < *idx; i++) {
2596                                 if (array[i] == rec->ip) {
2597                                         exists = true;
2598                                         break;
2599                                 }
2600                         }
2601
2602                         if (!not) {
2603                                 fail = 0;
2604                                 if (!exists) {
2605                                         array[(*idx)++] = rec->ip;
2606                                         if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2607                                                 goto out;
2608                                 }
2609                         } else {
2610                                 if (exists) {
2611                                         array[i] = array[--(*idx)];
2612                                         array[*idx] = 0;
2613                                         fail = 0;
2614                                 }
2615                         }
2616                 }
2617         } while_for_each_ftrace_rec();
2618 out:
2619         mutex_unlock(&ftrace_lock);
2620
2621         if (fail)
2622                 return -EINVAL;
2623
2624         ftrace_graph_filter_enabled = 1;
2625         return 0;
2626 }
2627
2628 static ssize_t
2629 ftrace_graph_write(struct file *file, const char __user *ubuf,
2630                    size_t cnt, loff_t *ppos)
2631 {
2632         struct trace_parser parser;
2633         ssize_t read, ret;
2634
2635         if (!cnt)
2636                 return 0;
2637
2638         mutex_lock(&graph_lock);
2639
2640         if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
2641                 ret = -ENOMEM;
2642                 goto out_unlock;
2643         }
2644
2645         read = trace_get_user(&parser, ubuf, cnt, ppos);
2646
2647         if (read >= 0 && trace_parser_loaded((&parser))) {
2648                 parser.buffer[parser.idx] = 0;
2649
2650                 /* we allow only one expression at a time */
2651                 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2652                                         parser.buffer);
2653                 if (ret)
2654                         goto out_free;
2655         }
2656
2657         ret = read;
2658
2659 out_free:
2660         trace_parser_put(&parser);
2661 out_unlock:
2662         mutex_unlock(&graph_lock);
2663
2664         return ret;
2665 }
2666
2667 static const struct file_operations ftrace_graph_fops = {
2668         .open           = ftrace_graph_open,
2669         .read           = seq_read,
2670         .write          = ftrace_graph_write,
2671         .release        = ftrace_graph_release,
2672         .llseek         = seq_lseek,
2673 };
2674 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2675
2676 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2677 {
2678
2679         trace_create_file("available_filter_functions", 0444,
2680                         d_tracer, NULL, &ftrace_avail_fops);
2681
2682         trace_create_file("failures", 0444,
2683                         d_tracer, NULL, &ftrace_failures_fops);
2684
2685         trace_create_file("set_ftrace_filter", 0644, d_tracer,
2686                         NULL, &ftrace_filter_fops);
2687
2688         trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2689                                     NULL, &ftrace_notrace_fops);
2690
2691 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2692         trace_create_file("set_graph_function", 0444, d_tracer,
2693                                     NULL,
2694                                     &ftrace_graph_fops);
2695 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2696
2697         return 0;
2698 }
2699
2700 static int ftrace_process_locs(struct module *mod,
2701                                unsigned long *start,
2702                                unsigned long *end)
2703 {
2704         unsigned long *p;
2705         unsigned long addr;
2706         unsigned long flags;
2707
2708         mutex_lock(&ftrace_lock);
2709         p = start;
2710         while (p < end) {
2711                 addr = ftrace_call_adjust(*p++);
2712                 /*
2713                  * Some architecture linkers will pad between
2714                  * the different mcount_loc sections of different
2715                  * object files to satisfy alignments.
2716                  * Skip any NULL pointers.
2717                  */
2718                 if (!addr)
2719                         continue;
2720                 ftrace_record_ip(addr);
2721         }
2722
2723         /* disable interrupts to prevent kstop machine */
2724         local_irq_save(flags);
2725         ftrace_update_code(mod);
2726         local_irq_restore(flags);
2727         mutex_unlock(&ftrace_lock);
2728
2729         return 0;
2730 }
2731
2732 #ifdef CONFIG_MODULES
2733 void ftrace_release_mod(struct module *mod)
2734 {
2735         struct dyn_ftrace *rec;
2736         struct ftrace_page *pg;
2737
2738         if (ftrace_disabled)
2739                 return;
2740
2741         mutex_lock(&ftrace_lock);
2742         do_for_each_ftrace_rec(pg, rec) {
2743                 if (within_module_core(rec->ip, mod)) {
2744                         /*
2745                          * rec->ip is changed in ftrace_free_rec()
2746                          * It should not between s and e if record was freed.
2747                          */
2748                         FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2749                         ftrace_free_rec(rec);
2750                 }
2751         } while_for_each_ftrace_rec();
2752         mutex_unlock(&ftrace_lock);
2753 }
2754
2755 static void ftrace_init_module(struct module *mod,
2756                                unsigned long *start, unsigned long *end)
2757 {
2758         if (ftrace_disabled || start == end)
2759                 return;
2760         ftrace_process_locs(mod, start, end);
2761 }
2762
2763 static int ftrace_module_notify(struct notifier_block *self,
2764                                 unsigned long val, void *data)
2765 {
2766         struct module *mod = data;
2767
2768         switch (val) {
2769         case MODULE_STATE_COMING:
2770                 ftrace_init_module(mod, mod->ftrace_callsites,
2771                                    mod->ftrace_callsites +
2772                                    mod->num_ftrace_callsites);
2773                 break;
2774         case MODULE_STATE_GOING:
2775                 ftrace_release_mod(mod);
2776                 break;
2777         }
2778
2779         return 0;
2780 }
2781 #else
2782 static int ftrace_module_notify(struct notifier_block *self,
2783                                 unsigned long val, void *data)
2784 {
2785         return 0;
2786 }
2787 #endif /* CONFIG_MODULES */
2788
2789 struct notifier_block ftrace_module_nb = {
2790         .notifier_call = ftrace_module_notify,
2791         .priority = 0,
2792 };
2793
2794 extern unsigned long __start_mcount_loc[];
2795 extern unsigned long __stop_mcount_loc[];
2796
2797 void __init ftrace_init(void)
2798 {
2799         unsigned long count, addr, flags;
2800         int ret;
2801
2802         /* Keep the ftrace pointer to the stub */
2803         addr = (unsigned long)ftrace_stub;
2804
2805         local_irq_save(flags);
2806         ftrace_dyn_arch_init(&addr);
2807         local_irq_restore(flags);
2808
2809         /* ftrace_dyn_arch_init places the return code in addr */
2810         if (addr)
2811                 goto failed;
2812
2813         count = __stop_mcount_loc - __start_mcount_loc;
2814
2815         ret = ftrace_dyn_table_alloc(count);
2816         if (ret)
2817                 goto failed;
2818
2819         last_ftrace_enabled = ftrace_enabled = 1;
2820
2821         ret = ftrace_process_locs(NULL,
2822                                   __start_mcount_loc,
2823                                   __stop_mcount_loc);
2824
2825         ret = register_module_notifier(&ftrace_module_nb);
2826         if (ret)
2827                 pr_warning("Failed to register trace ftrace module notifier\n");
2828
2829         set_ftrace_early_filters();
2830
2831         return;
2832  failed:
2833         ftrace_disabled = 1;
2834 }
2835
2836 #else
2837
2838 static int __init ftrace_nodyn_init(void)
2839 {
2840         ftrace_enabled = 1;
2841         return 0;
2842 }
2843 device_initcall(ftrace_nodyn_init);
2844
2845 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2846 static inline void ftrace_startup_enable(int command) { }
2847 /* Keep as macros so we do not need to define the commands */
2848 # define ftrace_startup(command)        do { } while (0)
2849 # define ftrace_shutdown(command)       do { } while (0)
2850 # define ftrace_startup_sysctl()        do { } while (0)
2851 # define ftrace_shutdown_sysctl()       do { } while (0)
2852 #endif /* CONFIG_DYNAMIC_FTRACE */
2853
2854 static void clear_ftrace_swapper(void)
2855 {
2856         struct task_struct *p;
2857         int cpu;
2858
2859         get_online_cpus();
2860         for_each_online_cpu(cpu) {
2861                 p = idle_task(cpu);
2862                 clear_tsk_trace_trace(p);
2863         }
2864         put_online_cpus();
2865 }
2866
2867 static void set_ftrace_swapper(void)
2868 {
2869         struct task_struct *p;
2870         int cpu;
2871
2872         get_online_cpus();
2873         for_each_online_cpu(cpu) {
2874                 p = idle_task(cpu);
2875                 set_tsk_trace_trace(p);
2876         }
2877         put_online_cpus();
2878 }
2879
2880 static void clear_ftrace_pid(struct pid *pid)
2881 {
2882         struct task_struct *p;
2883
2884         rcu_read_lock();
2885         do_each_pid_task(pid, PIDTYPE_PID, p) {
2886                 clear_tsk_trace_trace(p);
2887         } while_each_pid_task(pid, PIDTYPE_PID, p);
2888         rcu_read_unlock();
2889
2890         put_pid(pid);
2891 }
2892
2893 static void set_ftrace_pid(struct pid *pid)
2894 {
2895         struct task_struct *p;
2896
2897         rcu_read_lock();
2898         do_each_pid_task(pid, PIDTYPE_PID, p) {
2899                 set_tsk_trace_trace(p);
2900         } while_each_pid_task(pid, PIDTYPE_PID, p);
2901         rcu_read_unlock();
2902 }
2903
2904 static void clear_ftrace_pid_task(struct pid *pid)
2905 {
2906         if (pid == ftrace_swapper_pid)
2907                 clear_ftrace_swapper();
2908         else
2909                 clear_ftrace_pid(pid);
2910 }
2911
2912 static void set_ftrace_pid_task(struct pid *pid)
2913 {
2914         if (pid == ftrace_swapper_pid)
2915                 set_ftrace_swapper();
2916         else
2917                 set_ftrace_pid(pid);
2918 }
2919
2920 static int ftrace_pid_add(int p)
2921 {
2922         struct pid *pid;
2923         struct ftrace_pid *fpid;
2924         int ret = -EINVAL;
2925
2926         mutex_lock(&ftrace_lock);
2927
2928         if (!p)
2929                 pid = ftrace_swapper_pid;
2930         else
2931                 pid = find_get_pid(p);
2932
2933         if (!pid)
2934                 goto out;
2935
2936         ret = 0;
2937
2938         list_for_each_entry(fpid, &ftrace_pids, list)
2939                 if (fpid->pid == pid)
2940                         goto out_put;
2941
2942         ret = -ENOMEM;
2943
2944         fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
2945         if (!fpid)
2946                 goto out_put;
2947
2948         list_add(&fpid->list, &ftrace_pids);
2949         fpid->pid = pid;
2950
2951         set_ftrace_pid_task(pid);
2952
2953         ftrace_update_pid_func();
2954         ftrace_startup_enable(0);
2955
2956         mutex_unlock(&ftrace_lock);
2957         return 0;
2958
2959 out_put:
2960         if (pid != ftrace_swapper_pid)
2961                 put_pid(pid);
2962
2963 out:
2964         mutex_unlock(&ftrace_lock);
2965         return ret;
2966 }
2967
2968 static void ftrace_pid_reset(void)
2969 {
2970         struct ftrace_pid *fpid, *safe;
2971
2972         mutex_lock(&ftrace_lock);
2973         list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
2974                 struct pid *pid = fpid->pid;
2975
2976                 clear_ftrace_pid_task(pid);
2977
2978                 list_del(&fpid->list);
2979                 kfree(fpid);
2980         }
2981
2982         ftrace_update_pid_func();
2983         ftrace_startup_enable(0);
2984
2985         mutex_unlock(&ftrace_lock);
2986 }
2987
2988 static void *fpid_start(struct seq_file *m, loff_t *pos)
2989 {
2990         mutex_lock(&ftrace_lock);
2991
2992         if (list_empty(&ftrace_pids) && (!*pos))
2993                 return (void *) 1;
2994
2995         return seq_list_start(&ftrace_pids, *pos);
2996 }
2997
2998 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
2999 {
3000         if (v == (void *)1)
3001                 return NULL;
3002
3003         return seq_list_next(v, &ftrace_pids, pos);
3004 }
3005
3006 static void fpid_stop(struct seq_file *m, void *p)
3007 {
3008         mutex_unlock(&ftrace_lock);
3009 }
3010
3011 static int fpid_show(struct seq_file *m, void *v)
3012 {
3013         const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
3014
3015         if (v == (void *)1) {
3016                 seq_printf(m, "no pid\n");
3017                 return 0;
3018         }
3019
3020         if (fpid->pid == ftrace_swapper_pid)
3021                 seq_printf(m, "swapper tasks\n");
3022         else
3023                 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
3024
3025         return 0;
3026 }
3027
3028 static const struct seq_operations ftrace_pid_sops = {
3029         .start = fpid_start,
3030         .next = fpid_next,
3031         .stop = fpid_stop,
3032         .show = fpid_show,
3033 };
3034
3035 static int
3036 ftrace_pid_open(struct inode *inode, struct file *file)
3037 {
3038         int ret = 0;
3039
3040         if ((file->f_mode & FMODE_WRITE) &&
3041             (file->f_flags & O_TRUNC))
3042                 ftrace_pid_reset();
3043
3044         if (file->f_mode & FMODE_READ)
3045                 ret = seq_open(file, &ftrace_pid_sops);
3046
3047         return ret;
3048 }
3049
3050 static ssize_t
3051 ftrace_pid_write(struct file *filp, const char __user *ubuf,
3052                    size_t cnt, loff_t *ppos)
3053 {
3054         char buf[64], *tmp;
3055         long val;
3056         int ret;
3057
3058         if (cnt >= sizeof(buf))
3059                 return -EINVAL;
3060
3061         if (copy_from_user(&buf, ubuf, cnt))
3062                 return -EFAULT;
3063
3064         buf[cnt] = 0;
3065
3066         /*
3067          * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
3068          * to clean the filter quietly.
3069          */
3070         tmp = strstrip(buf);
3071         if (strlen(tmp) == 0)
3072                 return 1;
3073
3074         ret = strict_strtol(tmp, 10, &val);
3075         if (ret < 0)
3076                 return ret;
3077
3078         ret = ftrace_pid_add(val);
3079
3080         return ret ? ret : cnt;
3081 }
3082
3083 static int
3084 ftrace_pid_release(struct inode *inode, struct file *file)
3085 {
3086         if (file->f_mode & FMODE_READ)
3087                 seq_release(inode, file);
3088
3089         return 0;
3090 }
3091
3092 static const struct file_operations ftrace_pid_fops = {
3093         .open           = ftrace_pid_open,
3094         .write          = ftrace_pid_write,
3095         .read           = seq_read,
3096         .llseek         = seq_lseek,
3097         .release        = ftrace_pid_release,
3098 };
3099
3100 static __init int ftrace_init_debugfs(void)
3101 {
3102         struct dentry *d_tracer;
3103
3104         d_tracer = tracing_init_dentry();
3105         if (!d_tracer)
3106                 return 0;
3107
3108         ftrace_init_dyn_debugfs(d_tracer);
3109
3110         trace_create_file("set_ftrace_pid", 0644, d_tracer,
3111                             NULL, &ftrace_pid_fops);
3112
3113         ftrace_profile_debugfs(d_tracer);
3114
3115         return 0;
3116 }
3117 fs_initcall(ftrace_init_debugfs);
3118
3119 /**
3120  * ftrace_kill - kill ftrace
3121  *
3122  * This function should be used by panic code. It stops ftrace
3123  * but in a not so nice way. If you need to simply kill ftrace
3124  * from a non-atomic section, use ftrace_kill.
3125  */
3126 void ftrace_kill(void)
3127 {
3128         ftrace_disabled = 1;
3129         ftrace_enabled = 0;
3130         clear_ftrace_function();
3131 }
3132
3133 /**
3134  * register_ftrace_function - register a function for profiling
3135  * @ops - ops structure that holds the function for profiling.
3136  *
3137  * Register a function to be called by all functions in the
3138  * kernel.
3139  *
3140  * Note: @ops->func and all the functions it calls must be labeled
3141  *       with "notrace", otherwise it will go into a
3142  *       recursive loop.
3143  */
3144 int register_ftrace_function(struct ftrace_ops *ops)
3145 {
3146         int ret;
3147
3148         if (unlikely(ftrace_disabled))
3149                 return -1;
3150
3151         mutex_lock(&ftrace_lock);
3152
3153         ret = __register_ftrace_function(ops);
3154         ftrace_startup(0);
3155
3156         mutex_unlock(&ftrace_lock);
3157         return ret;
3158 }
3159
3160 /**
3161  * unregister_ftrace_function - unregister a function for profiling.
3162  * @ops - ops structure that holds the function to unregister
3163  *
3164  * Unregister a function that was added to be called by ftrace profiling.
3165  */
3166 int unregister_ftrace_function(struct ftrace_ops *ops)
3167 {
3168         int ret;
3169
3170         mutex_lock(&ftrace_lock);
3171         ret = __unregister_ftrace_function(ops);
3172         ftrace_shutdown(0);
3173         mutex_unlock(&ftrace_lock);
3174
3175         return ret;
3176 }
3177
3178 int
3179 ftrace_enable_sysctl(struct ctl_table *table, int write,
3180                      void __user *buffer, size_t *lenp,
3181                      loff_t *ppos)
3182 {
3183         int ret;
3184
3185         if (unlikely(ftrace_disabled))
3186                 return -ENODEV;
3187
3188         mutex_lock(&ftrace_lock);
3189
3190         ret  = proc_dointvec(table, write, buffer, lenp, ppos);
3191
3192         if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3193                 goto out;
3194
3195         last_ftrace_enabled = !!ftrace_enabled;
3196
3197         if (ftrace_enabled) {
3198
3199                 ftrace_startup_sysctl();
3200
3201                 /* we are starting ftrace again */
3202                 if (ftrace_list != &ftrace_list_end) {
3203                         if (ftrace_list->next == &ftrace_list_end)
3204                                 ftrace_trace_function = ftrace_list->func;
3205                         else
3206                                 ftrace_trace_function = ftrace_list_func;
3207                 }
3208
3209         } else {
3210                 /* stopping ftrace calls (just send to ftrace_stub) */
3211                 ftrace_trace_function = ftrace_stub;
3212
3213                 ftrace_shutdown_sysctl();
3214         }
3215
3216  out:
3217         mutex_unlock(&ftrace_lock);
3218         return ret;
3219 }
3220
3221 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3222
3223 static int ftrace_graph_active;
3224 static struct notifier_block ftrace_suspend_notifier;
3225
3226 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3227 {
3228         return 0;
3229 }
3230
3231 /* The callbacks that hook a function */
3232 trace_func_graph_ret_t ftrace_graph_return =
3233                         (trace_func_graph_ret_t)ftrace_stub;
3234 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3235
3236 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3237 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3238 {
3239         int i;
3240         int ret = 0;
3241         unsigned long flags;
3242         int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3243         struct task_struct *g, *t;
3244
3245         for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3246                 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3247                                         * sizeof(struct ftrace_ret_stack),
3248                                         GFP_KERNEL);
3249                 if (!ret_stack_list[i]) {
3250                         start = 0;
3251                         end = i;
3252                         ret = -ENOMEM;
3253                         goto free;
3254                 }
3255         }
3256
3257         read_lock_irqsave(&tasklist_lock, flags);
3258         do_each_thread(g, t) {
3259                 if (start == end) {
3260                         ret = -EAGAIN;
3261                         goto unlock;
3262                 }
3263
3264                 if (t->ret_stack == NULL) {
3265                         atomic_set(&t->tracing_graph_pause, 0);
3266                         atomic_set(&t->trace_overrun, 0);
3267                         t->curr_ret_stack = -1;
3268                         /* Make sure the tasks see the -1 first: */
3269                         smp_wmb();
3270                         t->ret_stack = ret_stack_list[start++];
3271                 }
3272         } while_each_thread(g, t);
3273
3274 unlock:
3275         read_unlock_irqrestore(&tasklist_lock, flags);
3276 free:
3277         for (i = start; i < end; i++)
3278                 kfree(ret_stack_list[i]);
3279         return ret;
3280 }
3281
3282 static void
3283 ftrace_graph_probe_sched_switch(void *ignore,
3284                         struct task_struct *prev, struct task_struct *next)
3285 {
3286         unsigned long long timestamp;
3287         int index;
3288
3289         /*
3290          * Does the user want to count the time a function was asleep.
3291          * If so, do not update the time stamps.
3292          */
3293         if (trace_flags & TRACE_ITER_SLEEP_TIME)
3294                 return;
3295
3296         timestamp = trace_clock_local();
3297
3298         prev->ftrace_timestamp = timestamp;
3299
3300         /* only process tasks that we timestamped */
3301         if (!next->ftrace_timestamp)
3302                 return;
3303
3304         /*
3305          * Update all the counters in next to make up for the
3306          * time next was sleeping.
3307          */
3308         timestamp -= next->ftrace_timestamp;
3309
3310         for (index = next->curr_ret_stack; index >= 0; index--)
3311                 next->ret_stack[index].calltime += timestamp;
3312 }
3313
3314 /* Allocate a return stack for each task */
3315 static int start_graph_tracing(void)
3316 {
3317         struct ftrace_ret_stack **ret_stack_list;
3318         int ret, cpu;
3319
3320         ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3321                                 sizeof(struct ftrace_ret_stack *),
3322                                 GFP_KERNEL);
3323
3324         if (!ret_stack_list)
3325                 return -ENOMEM;
3326
3327         /* The cpu_boot init_task->ret_stack will never be freed */
3328         for_each_online_cpu(cpu) {
3329                 if (!idle_task(cpu)->ret_stack)
3330                         ftrace_graph_init_idle_task(idle_task(cpu), cpu);
3331         }
3332
3333         do {
3334                 ret = alloc_retstack_tasklist(ret_stack_list);
3335         } while (ret == -EAGAIN);
3336
3337         if (!ret) {
3338                 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3339                 if (ret)
3340                         pr_info("ftrace_graph: Couldn't activate tracepoint"
3341                                 " probe to kernel_sched_switch\n");
3342         }
3343
3344         kfree(ret_stack_list);
3345         return ret;
3346 }
3347
3348 /*
3349  * Hibernation protection.
3350  * The state of the current task is too much unstable during
3351  * suspend/restore to disk. We want to protect against that.
3352  */
3353 static int
3354 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3355                                                         void *unused)
3356 {
3357         switch (state) {
3358         case PM_HIBERNATION_PREPARE:
3359                 pause_graph_tracing();
3360                 break;
3361
3362         case PM_POST_HIBERNATION:
3363                 unpause_graph_tracing();
3364                 break;
3365         }
3366         return NOTIFY_DONE;
3367 }
3368
3369 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3370                         trace_func_graph_ent_t entryfunc)
3371 {
3372         int ret = 0;
3373
3374         mutex_lock(&ftrace_lock);
3375
3376         /* we currently allow only one tracer registered at a time */
3377         if (ftrace_graph_active) {
3378                 ret = -EBUSY;
3379                 goto out;
3380         }
3381
3382         ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3383         register_pm_notifier(&ftrace_suspend_notifier);
3384
3385         ftrace_graph_active++;
3386         ret = start_graph_tracing();
3387         if (ret) {
3388                 ftrace_graph_active--;
3389                 goto out;
3390         }
3391
3392         ftrace_graph_return = retfunc;
3393         ftrace_graph_entry = entryfunc;
3394
3395         ftrace_startup(FTRACE_START_FUNC_RET);
3396
3397 out:
3398         mutex_unlock(&ftrace_lock);
3399         return ret;
3400 }
3401
3402 void unregister_ftrace_graph(void)
3403 {
3404         mutex_lock(&ftrace_lock);
3405
3406         if (unlikely(!ftrace_graph_active))
3407                 goto out;
3408
3409         ftrace_graph_active--;
3410         ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3411         ftrace_graph_entry = ftrace_graph_entry_stub;
3412         ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3413         unregister_pm_notifier(&ftrace_suspend_notifier);
3414         unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3415
3416  out:
3417         mutex_unlock(&ftrace_lock);
3418 }
3419
3420 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
3421
3422 static void
3423 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
3424 {
3425         atomic_set(&t->tracing_graph_pause, 0);
3426         atomic_set(&t->trace_overrun, 0);
3427         t->ftrace_timestamp = 0;
3428         /* make curr_ret_stack visible before we add the ret_stack */
3429         smp_wmb();
3430         t->ret_stack = ret_stack;
3431 }
3432
3433 /*
3434  * Allocate a return stack for the idle task. May be the first
3435  * time through, or it may be done by CPU hotplug online.
3436  */
3437 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
3438 {
3439         t->curr_ret_stack = -1;
3440         /*
3441          * The idle task has no parent, it either has its own
3442          * stack or no stack at all.
3443          */
3444         if (t->ret_stack)
3445                 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
3446
3447         if (ftrace_graph_active) {
3448                 struct ftrace_ret_stack *ret_stack;
3449
3450                 ret_stack = per_cpu(idle_ret_stack, cpu);
3451                 if (!ret_stack) {
3452                         ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3453                                             * sizeof(struct ftrace_ret_stack),
3454                                             GFP_KERNEL);
3455                         if (!ret_stack)
3456                                 return;
3457                         per_cpu(idle_ret_stack, cpu) = ret_stack;
3458                 }
3459                 graph_init_task(t, ret_stack);
3460         }
3461 }
3462
3463 /* Allocate a return stack for newly created task */
3464 void ftrace_graph_init_task(struct task_struct *t)
3465 {
3466         /* Make sure we do not use the parent ret_stack */
3467         t->ret_stack = NULL;
3468         t->curr_ret_stack = -1;
3469
3470         if (ftrace_graph_active) {
3471                 struct ftrace_ret_stack *ret_stack;
3472
3473                 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3474                                 * sizeof(struct ftrace_ret_stack),
3475                                 GFP_KERNEL);
3476                 if (!ret_stack)
3477                         return;
3478                 graph_init_task(t, ret_stack);
3479         }
3480 }
3481
3482 void ftrace_graph_exit_task(struct task_struct *t)
3483 {
3484         struct ftrace_ret_stack *ret_stack = t->ret_stack;
3485
3486         t->ret_stack = NULL;
3487         /* NULL must become visible to IRQs before we free it: */
3488         barrier();
3489
3490         kfree(ret_stack);
3491 }
3492
3493 void ftrace_graph_stop(void)
3494 {
3495         ftrace_stop();
3496 }
3497 #endif