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 };
 804
 805 /* used to initialize the real stat files */
 806 static struct tracer_stat function_stats __initdata = {
 807         .name           = "functions",
 808         .stat_start     = function_stat_start,
 809         .stat_next      = function_stat_next,
 810         .stat_cmp       = function_stat_cmp,
 811         .stat_headers   = function_stat_headers,
 812         .stat_show      = function_stat_show
 813 };
 814
 815 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 816 {
 817         struct ftrace_profile_stat *stat;
 818         struct dentry *entry;
 819         char *name;
 820         int ret;
 821         int cpu;
 822
 823         for_each_possible_cpu(cpu) {
 824                 stat = &per_cpu(ftrace_profile_stats, cpu);
 825
 826                 /* allocate enough for function name + cpu number */
 827                 name = kmalloc(32, GFP_KERNEL);
 828                 if (!name) {
 829                         /*
 830                          * The files created are permanent, if something happens
 831                          * we still do not free memory.
 832                          */
 833                         WARN(1,
 834                              "Could not allocate stat file for cpu %d\n",
 835                              cpu);
 836                         return;
 837                 }
 838                 stat->stat = function_stats;
 839                 snprintf(name, 32, "function%d", cpu);
 840                 stat->stat.name = name;
 841                 ret = register_stat_tracer(&stat->stat);
 842                 if (ret) {
 843                         WARN(1,
 844                              "Could not register function stat for cpu %d\n",
 845                              cpu);
 846                         kfree(name);
 847                         return;
 848                 }
 849         }
 850
 851         entry = debugfs_create_file("function_profile_enabled", 0644,
 852                                     d_tracer, NULL, &ftrace_profile_fops);
 853         if (!entry)
 854                 pr_warning("Could not create debugfs "
 855                            "'function_profile_enabled' entry\n");
 856 }
 857
 858 #else /* CONFIG_FUNCTION_PROFILER */
 859 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 860 {
 861 }
 862 #endif /* CONFIG_FUNCTION_PROFILER */
 863
 864 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
 865
 866 #ifdef CONFIG_DYNAMIC_FTRACE
 867
 868 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
 869 # error Dynamic ftrace depends on MCOUNT_RECORD
 870 #endif
 871
 872 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
 873
 874 struct ftrace_func_probe {
 875         struct hlist_node       node;
 876         struct ftrace_probe_ops *ops;
 877         unsigned long           flags;
 878         unsigned long           ip;
 879         void                    *data;
 880         struct rcu_head         rcu;
 881 };
 882
 883 enum {
 884         FTRACE_ENABLE_CALLS             = (1 << 0),
 885         FTRACE_DISABLE_CALLS            = (1 << 1),
 886         FTRACE_UPDATE_TRACE_FUNC        = (1 << 2),
 887         FTRACE_ENABLE_MCOUNT            = (1 << 3),
 888         FTRACE_DISABLE_MCOUNT           = (1 << 4),
 889         FTRACE_START_FUNC_RET           = (1 << 5),
 890         FTRACE_STOP_FUNC_RET            = (1 << 6),
 891 };
 892
 893 static int ftrace_filtered;
 894
 895 static struct dyn_ftrace *ftrace_new_addrs;
 896
 897 static DEFINE_MUTEX(ftrace_regex_lock);
 898
 899 struct ftrace_page {
 900         struct ftrace_page      *next;
 901         int                     index;
 902         struct dyn_ftrace       records[];
 903 };
 904
 905 #define ENTRIES_PER_PAGE \
 906   ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
 907
 908 /* estimate from running different kernels */
 909 #define NR_TO_INIT              10000
 910
 911 static struct ftrace_page       *ftrace_pages_start;
 912 static struct ftrace_page       *ftrace_pages;
 913
 914 static struct dyn_ftrace *ftrace_free_records;
 915
 916 /*
 917  * This is a double for. Do not use 'break' to break out of the loop,
 918  * you must use a goto.
 919  */
 920 #define do_for_each_ftrace_rec(pg, rec)                                 \
 921         for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
 922                 int _____i;                                             \
 923                 for (_____i = 0; _____i < pg->index; _____i++) {        \
 924                         rec = &pg->records[_____i];
 925
 926 #define while_for_each_ftrace_rec()             \
 927                 }                               \
 928         }
 929
 930 static void ftrace_free_rec(struct dyn_ftrace *rec)
 931 {
 932         rec->freelist = ftrace_free_records;
 933         ftrace_free_records = rec;
 934         rec->flags |= FTRACE_FL_FREE;
 935 }
 936
 937 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
 938 {
 939         struct dyn_ftrace *rec;
 940
 941         /* First check for freed records */
 942         if (ftrace_free_records) {
 943                 rec = ftrace_free_records;
 944
 945                 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
 946                         FTRACE_WARN_ON_ONCE(1);
 947                         ftrace_free_records = NULL;
 948                         return NULL;
 949                 }
 950
 951                 ftrace_free_records = rec->freelist;
 952                 memset(rec, 0, sizeof(*rec));
 953                 return rec;
 954         }
 955
 956         if (ftrace_pages->index == ENTRIES_PER_PAGE) {
 957                 if (!ftrace_pages->next) {
 958                         /* allocate another page */
 959                         ftrace_pages->next =
 960                                 (void *)get_zeroed_page(GFP_KERNEL);
 961                         if (!ftrace_pages->next)
 962                                 return NULL;
 963                 }
 964                 ftrace_pages = ftrace_pages->next;
 965         }
 966
 967         return &ftrace_pages->records[ftrace_pages->index++];
 968 }
 969
 970 static struct dyn_ftrace *
 971 ftrace_record_ip(unsigned long ip)
 972 {
 973         struct dyn_ftrace *rec;
 974
 975         if (ftrace_disabled)
 976                 return NULL;
 977
 978         rec = ftrace_alloc_dyn_node(ip);
 979         if (!rec)
 980                 return NULL;
 981
 982         rec->ip = ip;
 983         rec->newlist = ftrace_new_addrs;
 984         ftrace_new_addrs = rec;
 985
 986         return rec;
 987 }
 988
 989 static void print_ip_ins(const char *fmt, unsigned char *p)
 990 {
 991         int i;
 992
 993         printk(KERN_CONT "%s", fmt);
 994
 995         for (i = 0; i < MCOUNT_INSN_SIZE; i++)
 996                 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
 997 }
 998
 999 static void ftrace_bug(int failed, unsigned long ip)
1000 {
1001         switch (failed) {
1002         case -EFAULT:
1003                 FTRACE_WARN_ON_ONCE(1);
1004                 pr_info("ftrace faulted on modifying ");
1005                 print_ip_sym(ip);
1006                 break;
1007         case -EINVAL:
1008                 FTRACE_WARN_ON_ONCE(1);
1009                 pr_info("ftrace failed to modify ");
1010                 print_ip_sym(ip);
1011                 print_ip_ins(" actual: ", (unsigned char *)ip);
1012                 printk(KERN_CONT "\n");
1013                 break;
1014         case -EPERM:
1015                 FTRACE_WARN_ON_ONCE(1);
1016                 pr_info("ftrace faulted on writing ");
1017                 print_ip_sym(ip);
1018                 break;
1019         default:
1020                 FTRACE_WARN_ON_ONCE(1);
1021                 pr_info("ftrace faulted on unknown error ");
1022                 print_ip_sym(ip);
1023         }
1024 }
1025
1026
1027 /* Return 1 if the address range is reserved for ftrace */
1028 int ftrace_text_reserved(void *start, void *end)
1029 {
1030         struct dyn_ftrace *rec;
1031         struct ftrace_page *pg;
1032
1033         do_for_each_ftrace_rec(pg, rec) {
1034                 if (rec->ip <= (unsigned long)end &&
1035                     rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1036                         return 1;
1037         } while_for_each_ftrace_rec();
1038         return 0;
1039 }
1040
1041
1042 static int
1043 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1044 {
1045         unsigned long ftrace_addr;
1046         unsigned long flag = 0UL;
1047
1048         ftrace_addr = (unsigned long)FTRACE_ADDR;
1049
1050         /*
1051          * If this record is not to be traced or we want to disable it,
1052          * then disable it.
1053          *
1054          * If we want to enable it and filtering is off, then enable it.
1055          *
1056          * If we want to enable it and filtering is on, enable it only if
1057          * it's filtered
1058          */
1059         if (enable && !(rec->flags & FTRACE_FL_NOTRACE)) {
1060                 if (!ftrace_filtered || (rec->flags & FTRACE_FL_FILTER))
1061                         flag = FTRACE_FL_ENABLED;
1062         }
1063
1064         /* If the state of this record hasn't changed, then do nothing */
1065         if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1066                 return 0;
1067
1068         if (flag) {
1069                 rec->flags |= FTRACE_FL_ENABLED;
1070                 return ftrace_make_call(rec, ftrace_addr);
1071         }
1072
1073         rec->flags &= ~FTRACE_FL_ENABLED;
1074         return ftrace_make_nop(NULL, rec, ftrace_addr);
1075 }
1076
1077 static void ftrace_replace_code(int enable)
1078 {
1079         struct dyn_ftrace *rec;
1080         struct ftrace_page *pg;
1081         int failed;
1082
1083         do_for_each_ftrace_rec(pg, rec) {
1084                 /*
1085                  * Skip over free records, records that have
1086                  * failed and not converted.
1087                  */
1088                 if (rec->flags & FTRACE_FL_FREE ||
1089                     rec->flags & FTRACE_FL_FAILED ||
1090                     !(rec->flags & FTRACE_FL_CONVERTED))
1091                         continue;
1092
1093                 failed = __ftrace_replace_code(rec, enable);
1094                 if (failed) {
1095                         rec->flags |= FTRACE_FL_FAILED;
1096                         ftrace_bug(failed, rec->ip);
1097                         /* Stop processing */
1098                         return;
1099                 }
1100         } while_for_each_ftrace_rec();
1101 }
1102
1103 static int
1104 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1105 {
1106         unsigned long ip;
1107         int ret;
1108
1109         ip = rec->ip;
1110
1111         ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1112         if (ret) {
1113                 ftrace_bug(ret, ip);
1114                 rec->flags |= FTRACE_FL_FAILED;
1115                 return 0;
1116         }
1117         return 1;
1118 }
1119
1120 /*
1121  * archs can override this function if they must do something
1122  * before the modifying code is performed.
1123  */
1124 int __weak ftrace_arch_code_modify_prepare(void)
1125 {
1126         return 0;
1127 }
1128
1129 /*
1130  * archs can override this function if they must do something
1131  * after the modifying code is performed.
1132  */
1133 int __weak ftrace_arch_code_modify_post_process(void)
1134 {
1135         return 0;
1136 }
1137
1138 static int __ftrace_modify_code(void *data)
1139 {
1140         int *command = data;
1141
1142         if (*command & FTRACE_ENABLE_CALLS)
1143                 ftrace_replace_code(1);
1144         else if (*command & FTRACE_DISABLE_CALLS)
1145                 ftrace_replace_code(0);
1146
1147         if (*command & FTRACE_UPDATE_TRACE_FUNC)
1148                 ftrace_update_ftrace_func(ftrace_trace_function);
1149
1150         if (*command & FTRACE_START_FUNC_RET)
1151                 ftrace_enable_ftrace_graph_caller();
1152         else if (*command & FTRACE_STOP_FUNC_RET)
1153                 ftrace_disable_ftrace_graph_caller();
1154
1155         return 0;
1156 }
1157
1158 static void ftrace_run_update_code(int command)
1159 {
1160         int ret;
1161
1162         ret = ftrace_arch_code_modify_prepare();
1163         FTRACE_WARN_ON(ret);
1164         if (ret)
1165                 return;
1166
1167         stop_machine(__ftrace_modify_code, &command, NULL);
1168
1169         ret = ftrace_arch_code_modify_post_process();
1170         FTRACE_WARN_ON(ret);
1171 }
1172
1173 static ftrace_func_t saved_ftrace_func;
1174 static int ftrace_start_up;
1175
1176 static void ftrace_startup_enable(int command)
1177 {
1178         if (saved_ftrace_func != ftrace_trace_function) {
1179                 saved_ftrace_func = ftrace_trace_function;
1180                 command |= FTRACE_UPDATE_TRACE_FUNC;
1181         }
1182
1183         if (!command || !ftrace_enabled)
1184                 return;
1185
1186         ftrace_run_update_code(command);
1187 }
1188
1189 static void ftrace_startup(int command)
1190 {
1191         if (unlikely(ftrace_disabled))
1192                 return;
1193
1194         ftrace_start_up++;
1195         command |= FTRACE_ENABLE_CALLS;
1196
1197         ftrace_startup_enable(command);
1198 }
1199
1200 static void ftrace_shutdown(int command)
1201 {
1202         if (unlikely(ftrace_disabled))
1203                 return;
1204
1205         ftrace_start_up--;
1206         /*
1207          * Just warn in case of unbalance, no need to kill ftrace, it's not
1208          * critical but the ftrace_call callers may be never nopped again after
1209          * further ftrace uses.
1210          */
1211         WARN_ON_ONCE(ftrace_start_up < 0);
1212
1213         if (!ftrace_start_up)
1214                 command |= FTRACE_DISABLE_CALLS;
1215
1216         if (saved_ftrace_func != ftrace_trace_function) {
1217                 saved_ftrace_func = ftrace_trace_function;
1218                 command |= FTRACE_UPDATE_TRACE_FUNC;
1219         }
1220
1221         if (!command || !ftrace_enabled)
1222                 return;
1223
1224         ftrace_run_update_code(command);
1225 }
1226
1227 static void ftrace_startup_sysctl(void)
1228 {
1229         int command = FTRACE_ENABLE_MCOUNT;
1230
1231         if (unlikely(ftrace_disabled))
1232                 return;
1233
1234         /* Force update next time */
1235         saved_ftrace_func = NULL;
1236         /* ftrace_start_up is true if we want ftrace running */
1237         if (ftrace_start_up)
1238                 command |= FTRACE_ENABLE_CALLS;
1239
1240         ftrace_run_update_code(command);
1241 }
1242
1243 static void ftrace_shutdown_sysctl(void)
1244 {
1245         int command = FTRACE_DISABLE_MCOUNT;
1246
1247         if (unlikely(ftrace_disabled))
1248                 return;
1249
1250         /* ftrace_start_up is true if ftrace is running */
1251         if (ftrace_start_up)
1252                 command |= FTRACE_DISABLE_CALLS;
1253
1254         ftrace_run_update_code(command);
1255 }
1256
1257 static cycle_t          ftrace_update_time;
1258 static unsigned long    ftrace_update_cnt;
1259 unsigned long           ftrace_update_tot_cnt;
1260
1261 static int ftrace_update_code(struct module *mod)
1262 {
1263         struct dyn_ftrace *p;
1264         cycle_t start, stop;
1265
1266         start = ftrace_now(raw_smp_processor_id());
1267         ftrace_update_cnt = 0;
1268
1269         while (ftrace_new_addrs) {
1270
1271                 /* If something went wrong, bail without enabling anything */
1272                 if (unlikely(ftrace_disabled))
1273                         return -1;
1274
1275                 p = ftrace_new_addrs;
1276                 ftrace_new_addrs = p->newlist;
1277                 p->flags = 0L;
1278
1279                 /*
1280                  * Do the initial record convertion from mcount jump
1281                  * to the NOP instructions.
1282                  */
1283                 if (!ftrace_code_disable(mod, p)) {
1284                         ftrace_free_rec(p);
1285                         continue;
1286                 }
1287
1288                 p->flags |= FTRACE_FL_CONVERTED;
1289                 ftrace_update_cnt++;
1290
1291                 /*
1292                  * If the tracing is enabled, go ahead and enable the record.
1293                  *
1294                  * The reason not to enable the record immediatelly is the
1295                  * inherent check of ftrace_make_nop/ftrace_make_call for
1296                  * correct previous instructions.  Making first the NOP
1297                  * conversion puts the module to the correct state, thus
1298                  * passing the ftrace_make_call check.
1299                  */
1300                 if (ftrace_start_up) {
1301                         int failed = __ftrace_replace_code(p, 1);
1302                         if (failed) {
1303                                 ftrace_bug(failed, p->ip);
1304                                 ftrace_free_rec(p);
1305                         }
1306                 }
1307         }
1308
1309         stop = ftrace_now(raw_smp_processor_id());
1310         ftrace_update_time = stop - start;
1311         ftrace_update_tot_cnt += ftrace_update_cnt;
1312
1313         return 0;
1314 }
1315
1316 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1317 {
1318         struct ftrace_page *pg;
1319         int cnt;
1320         int i;
1321
1322         /* allocate a few pages */
1323         ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1324         if (!ftrace_pages_start)
1325                 return -1;
1326
1327         /*
1328          * Allocate a few more pages.
1329          *
1330          * TODO: have some parser search vmlinux before
1331          *   final linking to find all calls to ftrace.
1332          *   Then we can:
1333          *    a) know how many pages to allocate.
1334          *     and/or
1335          *    b) set up the table then.
1336          *
1337          *  The dynamic code is still necessary for
1338          *  modules.
1339          */
1340
1341         pg = ftrace_pages = ftrace_pages_start;
1342
1343         cnt = num_to_init / ENTRIES_PER_PAGE;
1344         pr_info("ftrace: allocating %ld entries in %d pages\n",
1345                 num_to_init, cnt + 1);
1346
1347         for (i = 0; i < cnt; i++) {
1348                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1349
1350                 /* If we fail, we'll try later anyway */
1351                 if (!pg->next)
1352                         break;
1353
1354                 pg = pg->next;
1355         }
1356
1357         return 0;
1358 }
1359
1360 enum {
1361         FTRACE_ITER_FILTER      = (1 << 0),
1362         FTRACE_ITER_NOTRACE     = (1 << 1),
1363         FTRACE_ITER_FAILURES    = (1 << 2),
1364         FTRACE_ITER_PRINTALL    = (1 << 3),
1365         FTRACE_ITER_HASH        = (1 << 4),
1366 };
1367
1368 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1369
1370 struct ftrace_iterator {
1371         struct ftrace_page      *pg;
1372         int                     hidx;
1373         int                     idx;
1374         unsigned                flags;
1375         struct trace_parser     parser;
1376 };
1377
1378 static void *
1379 t_hash_next(struct seq_file *m, void *v, loff_t *pos)
1380 {
1381         struct ftrace_iterator *iter = m->private;
1382         struct hlist_node *hnd = v;
1383         struct hlist_head *hhd;
1384
1385         WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
1386
1387         (*pos)++;
1388
1389  retry:
1390         if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1391                 return NULL;
1392
1393         hhd = &ftrace_func_hash[iter->hidx];
1394
1395         if (hlist_empty(hhd)) {
1396                 iter->hidx++;
1397                 hnd = NULL;
1398                 goto retry;
1399         }
1400
1401         if (!hnd)
1402                 hnd = hhd->first;
1403         else {
1404                 hnd = hnd->next;
1405                 if (!hnd) {
1406                         iter->hidx++;
1407                         goto retry;
1408                 }
1409         }
1410
1411         return hnd;
1412 }
1413
1414 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1415 {
1416         struct ftrace_iterator *iter = m->private;
1417         void *p = NULL;
1418         loff_t l;
1419
1420         if (!(iter->flags & FTRACE_ITER_HASH))
1421                 *pos = 0;
1422
1423         iter->flags |= FTRACE_ITER_HASH;
1424
1425         iter->hidx = 0;
1426         for (l = 0; l <= *pos; ) {
1427                 p = t_hash_next(m, p, &l);
1428                 if (!p)
1429                         break;
1430         }
1431         return p;
1432 }
1433
1434 static int t_hash_show(struct seq_file *m, void *v)
1435 {
1436         struct ftrace_func_probe *rec;
1437         struct hlist_node *hnd = v;
1438
1439         rec = hlist_entry(hnd, struct ftrace_func_probe, node);
1440
1441         if (rec->ops->print)
1442                 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1443
1444         seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
1445
1446         if (rec->data)
1447                 seq_printf(m, ":%p", rec->data);
1448         seq_putc(m, '\n');
1449
1450         return 0;
1451 }
1452
1453 static void *
1454 t_next(struct seq_file *m, void *v, loff_t *pos)
1455 {
1456         struct ftrace_iterator *iter = m->private;
1457         struct dyn_ftrace *rec = NULL;
1458
1459         if (iter->flags & FTRACE_ITER_HASH)
1460                 return t_hash_next(m, v, pos);
1461
1462         (*pos)++;
1463
1464         if (iter->flags & FTRACE_ITER_PRINTALL)
1465                 return NULL;
1466
1467  retry:
1468         if (iter->idx >= iter->pg->index) {
1469                 if (iter->pg->next) {
1470                         iter->pg = iter->pg->next;
1471                         iter->idx = 0;
1472                         goto retry;
1473                 }
1474         } else {
1475                 rec = &iter->pg->records[iter->idx++];
1476                 if ((rec->flags & FTRACE_FL_FREE) ||
1477
1478                     (!(iter->flags & FTRACE_ITER_FAILURES) &&
1479                      (rec->flags & FTRACE_FL_FAILED)) ||
1480
1481                     ((iter->flags & FTRACE_ITER_FAILURES) &&
1482                      !(rec->flags & FTRACE_FL_FAILED)) ||
1483
1484                     ((iter->flags & FTRACE_ITER_FILTER) &&
1485                      !(rec->flags & FTRACE_FL_FILTER)) ||
1486
1487                     ((iter->flags & FTRACE_ITER_NOTRACE) &&
1488                      !(rec->flags & FTRACE_FL_NOTRACE))) {
1489                         rec = NULL;
1490                         goto retry;
1491                 }
1492         }
1493
1494         return rec;
1495 }
1496
1497 static void *t_start(struct seq_file *m, loff_t *pos)
1498 {
1499         struct ftrace_iterator *iter = m->private;
1500         void *p = NULL;
1501         loff_t l;
1502
1503         mutex_lock(&ftrace_lock);
1504         /*
1505          * For set_ftrace_filter reading, if we have the filter
1506          * off, we can short cut and just print out that all
1507          * functions are enabled.
1508          */
1509         if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1510                 if (*pos > 0)
1511                         return t_hash_start(m, pos);
1512                 iter->flags |= FTRACE_ITER_PRINTALL;
1513                 return iter;
1514         }
1515
1516         if (iter->flags & FTRACE_ITER_HASH)
1517                 return t_hash_start(m, pos);
1518
1519         iter->pg = ftrace_pages_start;
1520         iter->idx = 0;
1521         for (l = 0; l <= *pos; ) {
1522                 p = t_next(m, p, &l);
1523                 if (!p)
1524                         break;
1525         }
1526
1527         if (!p && iter->flags & FTRACE_ITER_FILTER)
1528                 return t_hash_start(m, pos);
1529
1530         return p;
1531 }
1532
1533 static void t_stop(struct seq_file *m, void *p)
1534 {
1535         mutex_unlock(&ftrace_lock);
1536 }
1537
1538 static int t_show(struct seq_file *m, void *v)
1539 {
1540         struct ftrace_iterator *iter = m->private;
1541         struct dyn_ftrace *rec = v;
1542
1543         if (iter->flags & FTRACE_ITER_HASH)
1544                 return t_hash_show(m, v);
1545
1546         if (iter->flags & FTRACE_ITER_PRINTALL) {
1547                 seq_printf(m, "#### all functions enabled ####\n");
1548                 return 0;
1549         }
1550
1551         if (!rec)
1552                 return 0;
1553
1554         seq_printf(m, "%ps\n", (void *)rec->ip);
1555
1556         return 0;
1557 }
1558
1559 static const struct seq_operations show_ftrace_seq_ops = {
1560         .start = t_start,
1561         .next = t_next,
1562         .stop = t_stop,
1563         .show = t_show,
1564 };
1565
1566 static int
1567 ftrace_avail_open(struct inode *inode, struct file *file)
1568 {
1569         struct ftrace_iterator *iter;
1570         int ret;
1571
1572         if (unlikely(ftrace_disabled))
1573                 return -ENODEV;
1574
1575         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1576         if (!iter)
1577                 return -ENOMEM;
1578
1579         iter->pg = ftrace_pages_start;
1580
1581         ret = seq_open(file, &show_ftrace_seq_ops);
1582         if (!ret) {
1583                 struct seq_file *m = file->private_data;
1584
1585                 m->private = iter;
1586         } else {
1587                 kfree(iter);
1588         }
1589
1590         return ret;
1591 }
1592
1593 static int
1594 ftrace_failures_open(struct inode *inode, struct file *file)
1595 {
1596         int ret;
1597         struct seq_file *m;
1598         struct ftrace_iterator *iter;
1599
1600         ret = ftrace_avail_open(inode, file);
1601         if (!ret) {
1602                 m = (struct seq_file *)file->private_data;
1603                 iter = (struct ftrace_iterator *)m->private;
1604                 iter->flags = FTRACE_ITER_FAILURES;
1605         }
1606
1607         return ret;
1608 }
1609
1610
1611 static void ftrace_filter_reset(int enable)
1612 {
1613         struct ftrace_page *pg;
1614         struct dyn_ftrace *rec;
1615         unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1616
1617         mutex_lock(&ftrace_lock);
1618         if (enable)
1619                 ftrace_filtered = 0;
1620         do_for_each_ftrace_rec(pg, rec) {
1621                 if (rec->flags & FTRACE_FL_FAILED)
1622                         continue;
1623                 rec->flags &= ~type;
1624         } while_for_each_ftrace_rec();
1625         mutex_unlock(&ftrace_lock);
1626 }
1627
1628 static int
1629 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1630 {
1631         struct ftrace_iterator *iter;
1632         int ret = 0;
1633
1634         if (unlikely(ftrace_disabled))
1635                 return -ENODEV;
1636
1637         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1638         if (!iter)
1639                 return -ENOMEM;
1640
1641         if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
1642                 kfree(iter);
1643                 return -ENOMEM;
1644         }
1645
1646         mutex_lock(&ftrace_regex_lock);
1647         if ((file->f_mode & FMODE_WRITE) &&
1648             (file->f_flags & O_TRUNC))
1649                 ftrace_filter_reset(enable);
1650
1651         if (file->f_mode & FMODE_READ) {
1652                 iter->pg = ftrace_pages_start;
1653                 iter->flags = enable ? FTRACE_ITER_FILTER :
1654                         FTRACE_ITER_NOTRACE;
1655
1656                 ret = seq_open(file, &show_ftrace_seq_ops);
1657                 if (!ret) {
1658                         struct seq_file *m = file->private_data;
1659                         m->private = iter;
1660                 } else {
1661                         trace_parser_put(&iter->parser);
1662                         kfree(iter);
1663                 }
1664         } else
1665                 file->private_data = iter;
1666         mutex_unlock(&ftrace_regex_lock);
1667
1668         return ret;
1669 }
1670
1671 static int
1672 ftrace_filter_open(struct inode *inode, struct file *file)
1673 {
1674         return ftrace_regex_open(inode, file, 1);
1675 }
1676
1677 static int
1678 ftrace_notrace_open(struct inode *inode, struct file *file)
1679 {
1680         return ftrace_regex_open(inode, file, 0);
1681 }
1682
1683 static loff_t
1684 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1685 {
1686         loff_t ret;
1687
1688         if (file->f_mode & FMODE_READ)
1689                 ret = seq_lseek(file, offset, origin);
1690         else
1691                 file->f_pos = ret = 1;
1692
1693         return ret;
1694 }
1695
1696 static int ftrace_match(char *str, char *regex, int len, int type)
1697 {
1698         int matched = 0;
1699         int slen;
1700
1701         switch (type) {
1702         case MATCH_FULL:
1703                 if (strcmp(str, regex) == 0)
1704                         matched = 1;
1705                 break;
1706         case MATCH_FRONT_ONLY:
1707                 if (strncmp(str, regex, len) == 0)
1708                         matched = 1;
1709                 break;
1710         case MATCH_MIDDLE_ONLY:
1711                 if (strstr(str, regex))
1712                         matched = 1;
1713                 break;
1714         case MATCH_END_ONLY:
1715                 slen = strlen(str);
1716                 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
1717                         matched = 1;
1718                 break;
1719         }
1720
1721         return matched;
1722 }
1723
1724 static int
1725 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1726 {
1727         char str[KSYM_SYMBOL_LEN];
1728
1729         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1730         return ftrace_match(str, regex, len, type);
1731 }
1732
1733 static int ftrace_match_records(char *buff, int len, int enable)
1734 {
1735         unsigned int search_len;
1736         struct ftrace_page *pg;
1737         struct dyn_ftrace *rec;
1738         unsigned long flag;
1739         char *search;
1740         int type;
1741         int not;
1742         int found = 0;
1743
1744         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1745         type = filter_parse_regex(buff, len, &search, &not);
1746
1747         search_len = strlen(search);
1748
1749         mutex_lock(&ftrace_lock);
1750         do_for_each_ftrace_rec(pg, rec) {
1751
1752                 if (rec->flags & FTRACE_FL_FAILED)
1753                         continue;
1754
1755                 if (ftrace_match_record(rec, search, search_len, type)) {
1756                         if (not)
1757                                 rec->flags &= ~flag;
1758                         else
1759                                 rec->flags |= flag;
1760                         found = 1;
1761                 }
1762                 /*
1763                  * Only enable filtering if we have a function that
1764                  * is filtered on.
1765                  */
1766                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1767                         ftrace_filtered = 1;
1768         } while_for_each_ftrace_rec();
1769         mutex_unlock(&ftrace_lock);
1770
1771         return found;
1772 }
1773
1774 static int
1775 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1776                            char *regex, int len, int type)
1777 {
1778         char str[KSYM_SYMBOL_LEN];
1779         char *modname;
1780
1781         kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1782
1783         if (!modname || strcmp(modname, mod))
1784                 return 0;
1785
1786         /* blank search means to match all funcs in the mod */
1787         if (len)
1788                 return ftrace_match(str, regex, len, type);
1789         else
1790                 return 1;
1791 }
1792
1793 static int ftrace_match_module_records(char *buff, char *mod, int enable)
1794 {
1795         unsigned search_len = 0;
1796         struct ftrace_page *pg;
1797         struct dyn_ftrace *rec;
1798         int type = MATCH_FULL;
1799         char *search = buff;
1800         unsigned long flag;
1801         int not = 0;
1802         int found = 0;
1803
1804         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1805
1806         /* blank or '*' mean the same */
1807         if (strcmp(buff, "*") == 0)
1808                 buff[0] = 0;
1809
1810         /* handle the case of 'dont filter this module' */
1811         if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1812                 buff[0] = 0;
1813                 not = 1;
1814         }
1815
1816         if (strlen(buff)) {
1817                 type = filter_parse_regex(buff, strlen(buff), &search, &not);
1818                 search_len = strlen(search);
1819         }
1820
1821         mutex_lock(&ftrace_lock);
1822         do_for_each_ftrace_rec(pg, rec) {
1823
1824                 if (rec->flags & FTRACE_FL_FAILED)
1825                         continue;
1826
1827                 if (ftrace_match_module_record(rec, mod,
1828                                                search, search_len, type)) {
1829                         if (not)
1830                                 rec->flags &= ~flag;
1831                         else
1832                                 rec->flags |= flag;
1833                         found = 1;
1834                 }
1835                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1836                         ftrace_filtered = 1;
1837
1838         } while_for_each_ftrace_rec();
1839         mutex_unlock(&ftrace_lock);
1840
1841         return found;
1842 }
1843
1844 /*
1845  * We register the module command as a template to show others how
1846  * to register the a command as well.
1847  */
1848
1849 static int
1850 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1851 {
1852         char *mod;
1853
1854         /*
1855          * cmd == 'mod' because we only registered this func
1856          * for the 'mod' ftrace_func_command.
1857          * But if you register one func with multiple commands,
1858          * you can tell which command was used by the cmd
1859          * parameter.
1860          */
1861
1862         /* we must have a module name */
1863         if (!param)
1864                 return -EINVAL;
1865
1866         mod = strsep(&param, ":");
1867         if (!strlen(mod))
1868                 return -EINVAL;
1869
1870         if (ftrace_match_module_records(func, mod, enable))
1871                 return 0;
1872         return -EINVAL;
1873 }
1874
1875 static struct ftrace_func_command ftrace_mod_cmd = {
1876         .name                   = "mod",
1877         .func                   = ftrace_mod_callback,
1878 };
1879
1880 static int __init ftrace_mod_cmd_init(void)
1881 {
1882         return register_ftrace_command(&ftrace_mod_cmd);
1883 }
1884 device_initcall(ftrace_mod_cmd_init);
1885
1886 static void
1887 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1888 {
1889         struct ftrace_func_probe *entry;
1890         struct hlist_head *hhd;
1891         struct hlist_node *n;
1892         unsigned long key;
1893
1894         key = hash_long(ip, FTRACE_HASH_BITS);
1895
1896         hhd = &ftrace_func_hash[key];
1897
1898         if (hlist_empty(hhd))
1899                 return;
1900
1901         /*
1902          * Disable preemption for these calls to prevent a RCU grace
1903          * period. This syncs the hash iteration and freeing of items
1904          * on the hash. rcu_read_lock is too dangerous here.
1905          */
1906         preempt_disable_notrace();
1907         hlist_for_each_entry_rcu(entry, n, hhd, node) {
1908                 if (entry->ip == ip)
1909                         entry->ops->func(ip, parent_ip, &entry->data);
1910         }
1911         preempt_enable_notrace();
1912 }
1913
1914 static struct ftrace_ops trace_probe_ops __read_mostly =
1915 {
1916         .func           = function_trace_probe_call,
1917 };
1918
1919 static int ftrace_probe_registered;
1920
1921 static void __enable_ftrace_function_probe(void)
1922 {
1923         int i;
1924
1925         if (ftrace_probe_registered)
1926                 return;
1927
1928         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1929                 struct hlist_head *hhd = &ftrace_func_hash[i];
1930                 if (hhd->first)
1931                         break;
1932         }
1933         /* Nothing registered? */
1934         if (i == FTRACE_FUNC_HASHSIZE)
1935                 return;
1936
1937         __register_ftrace_function(&trace_probe_ops);
1938         ftrace_startup(0);
1939         ftrace_probe_registered = 1;
1940 }
1941
1942 static void __disable_ftrace_function_probe(void)
1943 {
1944         int i;
1945
1946         if (!ftrace_probe_registered)
1947                 return;
1948
1949         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1950                 struct hlist_head *hhd = &ftrace_func_hash[i];
1951                 if (hhd->first)
1952                         return;
1953         }
1954
1955         /* no more funcs left */
1956         __unregister_ftrace_function(&trace_probe_ops);
1957         ftrace_shutdown(0);
1958         ftrace_probe_registered = 0;
1959 }
1960
1961
1962 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
1963 {
1964         struct ftrace_func_probe *entry =
1965                 container_of(rhp, struct ftrace_func_probe, rcu);
1966
1967         if (entry->ops->free)
1968                 entry->ops->free(&entry->data);
1969         kfree(entry);
1970 }
1971
1972
1973 int
1974 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
1975                               void *data)
1976 {
1977         struct ftrace_func_probe *entry;
1978         struct ftrace_page *pg;
1979         struct dyn_ftrace *rec;
1980         int type, len, not;
1981         unsigned long key;
1982         int count = 0;
1983         char *search;
1984
1985         type = filter_parse_regex(glob, strlen(glob), &search, &not);
1986         len = strlen(search);
1987
1988         /* we do not support '!' for function probes */
1989         if (WARN_ON(not))
1990                 return -EINVAL;
1991
1992         mutex_lock(&ftrace_lock);
1993         do_for_each_ftrace_rec(pg, rec) {
1994
1995                 if (rec->flags & FTRACE_FL_FAILED)
1996                         continue;
1997
1998                 if (!ftrace_match_record(rec, search, len, type))
1999                         continue;
2000
2001                 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2002                 if (!entry) {
2003                         /* If we did not process any, then return error */
2004                         if (!count)
2005                                 count = -ENOMEM;
2006                         goto out_unlock;
2007                 }
2008
2009                 count++;
2010
2011                 entry->data = data;
2012
2013                 /*
2014                  * The caller might want to do something special
2015                  * for each function we find. We call the callback
2016                  * to give the caller an opportunity to do so.
2017                  */
2018                 if (ops->callback) {
2019                         if (ops->callback(rec->ip, &entry->data) < 0) {
2020                                 /* caller does not like this func */
2021                                 kfree(entry);
2022                                 continue;
2023                         }
2024                 }
2025
2026                 entry->ops = ops;
2027                 entry->ip = rec->ip;
2028
2029                 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2030                 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2031
2032         } while_for_each_ftrace_rec();
2033         __enable_ftrace_function_probe();
2034
2035  out_unlock:
2036         mutex_unlock(&ftrace_lock);
2037
2038         return count;
2039 }
2040
2041 enum {
2042         PROBE_TEST_FUNC         = 1,
2043         PROBE_TEST_DATA         = 2
2044 };
2045
2046 static void
2047 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2048                                   void *data, int flags)
2049 {
2050         struct ftrace_func_probe *entry;
2051         struct hlist_node *n, *tmp;
2052         char str[KSYM_SYMBOL_LEN];
2053         int type = MATCH_FULL;
2054         int i, len = 0;
2055         char *search;
2056
2057         if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2058                 glob = NULL;
2059         else if (glob) {
2060                 int not;
2061
2062                 type = filter_parse_regex(glob, strlen(glob), &search, &not);
2063                 len = strlen(search);
2064
2065                 /* we do not support '!' for function probes */
2066                 if (WARN_ON(not))
2067                         return;
2068         }
2069
2070         mutex_lock(&ftrace_lock);
2071         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2072                 struct hlist_head *hhd = &ftrace_func_hash[i];
2073
2074                 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2075
2076                         /* break up if statements for readability */
2077                         if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2078                                 continue;
2079
2080                         if ((flags & PROBE_TEST_DATA) && entry->data != data)
2081                                 continue;
2082
2083                         /* do this last, since it is the most expensive */
2084                         if (glob) {
2085                                 kallsyms_lookup(entry->ip, NULL, NULL,
2086                                                 NULL, str);
2087                                 if (!ftrace_match(str, glob, len, type))
2088                                         continue;
2089                         }
2090
2091                         hlist_del(&entry->node);
2092                         call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2093                 }
2094         }
2095         __disable_ftrace_function_probe();
2096         mutex_unlock(&ftrace_lock);
2097 }
2098
2099 void
2100 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2101                                 void *data)
2102 {
2103         __unregister_ftrace_function_probe(glob, ops, data,
2104                                           PROBE_TEST_FUNC | PROBE_TEST_DATA);
2105 }
2106
2107 void
2108 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2109 {
2110         __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2111 }
2112
2113 void unregister_ftrace_function_probe_all(char *glob)
2114 {
2115         __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2116 }
2117
2118 static LIST_HEAD(ftrace_commands);
2119 static DEFINE_MUTEX(ftrace_cmd_mutex);
2120
2121 int register_ftrace_command(struct ftrace_func_command *cmd)
2122 {
2123         struct ftrace_func_command *p;
2124         int ret = 0;
2125
2126         mutex_lock(&ftrace_cmd_mutex);
2127         list_for_each_entry(p, &ftrace_commands, list) {
2128                 if (strcmp(cmd->name, p->name) == 0) {
2129                         ret = -EBUSY;
2130                         goto out_unlock;
2131                 }
2132         }
2133         list_add(&cmd->list, &ftrace_commands);
2134  out_unlock:
2135         mutex_unlock(&ftrace_cmd_mutex);
2136
2137         return ret;
2138 }
2139
2140 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2141 {
2142         struct ftrace_func_command *p, *n;
2143         int ret = -ENODEV;
2144
2145         mutex_lock(&ftrace_cmd_mutex);
2146         list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2147                 if (strcmp(cmd->name, p->name) == 0) {
2148                         ret = 0;
2149                         list_del_init(&p->list);
2150                         goto out_unlock;
2151                 }
2152         }
2153  out_unlock:
2154         mutex_unlock(&ftrace_cmd_mutex);
2155
2156         return ret;
2157 }
2158
2159 static int ftrace_process_regex(char *buff, int len, int enable)
2160 {
2161         char *func, *command, *next = buff;
2162         struct ftrace_func_command *p;
2163         int ret = -EINVAL;
2164
2165         func = strsep(&next, ":");
2166
2167         if (!next) {
2168                 if (ftrace_match_records(func, len, enable))
2169                         return 0;
2170                 return ret;
2171         }
2172
2173         /* command found */
2174
2175         command = strsep(&next, ":");
2176
2177         mutex_lock(&ftrace_cmd_mutex);
2178         list_for_each_entry(p, &ftrace_commands, list) {
2179                 if (strcmp(p->name, command) == 0) {
2180                         ret = p->func(func, command, next, enable);
2181                         goto out_unlock;
2182                 }
2183         }
2184  out_unlock:
2185         mutex_unlock(&ftrace_cmd_mutex);
2186
2187         return ret;
2188 }
2189
2190 static ssize_t
2191 ftrace_regex_write(struct file *file, const char __user *ubuf,
2192                    size_t cnt, loff_t *ppos, int enable)
2193 {
2194         struct ftrace_iterator *iter;
2195         struct trace_parser *parser;
2196         ssize_t ret, read;
2197
2198         if (!cnt)
2199                 return 0;
2200
2201         mutex_lock(&ftrace_regex_lock);
2202
2203         if (file->f_mode & FMODE_READ) {
2204                 struct seq_file *m = file->private_data;
2205                 iter = m->private;
2206         } else
2207                 iter = file->private_data;
2208
2209         parser = &iter->parser;
2210         read = trace_get_user(parser, ubuf, cnt, ppos);
2211
2212         if (read >= 0 && trace_parser_loaded(parser) &&
2213             !trace_parser_cont(parser)) {
2214                 ret = ftrace_process_regex(parser->buffer,
2215                                            parser->idx, enable);
2216                 trace_parser_clear(parser);
2217                 if (ret)
2218                         goto out_unlock;
2219         }
2220
2221         ret = read;
2222 out_unlock:
2223         mutex_unlock(&ftrace_regex_lock);
2224
2225         return ret;
2226 }
2227
2228 static ssize_t
2229 ftrace_filter_write(struct file *file, const char __user *ubuf,
2230                     size_t cnt, loff_t *ppos)
2231 {
2232         return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2233 }
2234
2235 static ssize_t
2236 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2237                      size_t cnt, loff_t *ppos)
2238 {
2239         return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2240 }
2241
2242 static void
2243 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2244 {
2245         if (unlikely(ftrace_disabled))
2246                 return;
2247
2248         mutex_lock(&ftrace_regex_lock);
2249         if (reset)
2250                 ftrace_filter_reset(enable);
2251         if (buf)
2252                 ftrace_match_records(buf, len, enable);
2253         mutex_unlock(&ftrace_regex_lock);
2254 }
2255
2256 /**
2257  * ftrace_set_filter - set a function to filter on in ftrace
2258  * @buf - the string that holds the function filter text.
2259  * @len - the length of the string.
2260  * @reset - non zero to reset all filters before applying this filter.
2261  *
2262  * Filters denote which functions should be enabled when tracing is enabled.
2263  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2264  */
2265 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2266 {
2267         ftrace_set_regex(buf, len, reset, 1);
2268 }
2269
2270 /**
2271  * ftrace_set_notrace - set a function to not trace in ftrace
2272  * @buf - the string that holds the function notrace text.
2273  * @len - the length of the string.
2274  * @reset - non zero to reset all filters before applying this filter.
2275  *
2276  * Notrace Filters denote which functions should not be enabled when tracing
2277  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2278  * for tracing.
2279  */
2280 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2281 {
2282         ftrace_set_regex(buf, len, reset, 0);
2283 }
2284
2285 /*
2286  * command line interface to allow users to set filters on boot up.
2287  */
2288 #define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
2289 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2290 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2291
2292 static int __init set_ftrace_notrace(char *str)
2293 {
2294         strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2295         return 1;
2296 }
2297 __setup("ftrace_notrace=", set_ftrace_notrace);
2298
2299 static int __init set_ftrace_filter(char *str)
2300 {
2301         strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2302         return 1;
2303 }
2304 __setup("ftrace_filter=", set_ftrace_filter);
2305
2306 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2307 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
2308 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
2309
2310 static int __init set_graph_function(char *str)
2311 {
2312         strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
2313         return 1;
2314 }
2315 __setup("ftrace_graph_filter=", set_graph_function);
2316
2317 static void __init set_ftrace_early_graph(char *buf)
2318 {
2319         int ret;
2320         char *func;
2321
2322         while (buf) {
2323                 func = strsep(&buf, ",");
2324                 /* we allow only one expression at a time */
2325                 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2326                                       func);
2327                 if (ret)
2328                         printk(KERN_DEBUG "ftrace: function %s not "
2329                                           "traceable\n", func);
2330         }
2331 }
2332 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2333
2334 static void __init set_ftrace_early_filter(char *buf, int enable)
2335 {
2336         char *func;
2337
2338         while (buf) {
2339                 func = strsep(&buf, ",");
2340                 ftrace_set_regex(func, strlen(func), 0, enable);
2341         }
2342 }
2343
2344 static void __init set_ftrace_early_filters(void)
2345 {
2346         if (ftrace_filter_buf[0])
2347                 set_ftrace_early_filter(ftrace_filter_buf, 1);
2348         if (ftrace_notrace_buf[0])
2349                 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2350 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2351         if (ftrace_graph_buf[0])
2352                 set_ftrace_early_graph(ftrace_graph_buf);
2353 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2354 }
2355
2356 static int
2357 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2358 {
2359         struct seq_file *m = (struct seq_file *)file->private_data;
2360         struct ftrace_iterator *iter;
2361         struct trace_parser *parser;
2362
2363         mutex_lock(&ftrace_regex_lock);
2364         if (file->f_mode & FMODE_READ) {
2365                 iter = m->private;
2366
2367                 seq_release(inode, file);
2368         } else
2369                 iter = file->private_data;
2370
2371         parser = &iter->parser;
2372         if (trace_parser_loaded(parser)) {
2373                 parser->buffer[parser->idx] = 0;
2374                 ftrace_match_records(parser->buffer, parser->idx, enable);
2375         }
2376
2377         mutex_lock(&ftrace_lock);
2378         if (ftrace_start_up && ftrace_enabled)
2379                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2380         mutex_unlock(&ftrace_lock);
2381
2382         trace_parser_put(parser);
2383         kfree(iter);
2384
2385         mutex_unlock(&ftrace_regex_lock);
2386         return 0;
2387 }
2388
2389 static int
2390 ftrace_filter_release(struct inode *inode, struct file *file)
2391 {
2392         return ftrace_regex_release(inode, file, 1);
2393 }
2394
2395 static int
2396 ftrace_notrace_release(struct inode *inode, struct file *file)
2397 {
2398         return ftrace_regex_release(inode, file, 0);
2399 }
2400
2401 static const struct file_operations ftrace_avail_fops = {
2402         .open = ftrace_avail_open,
2403         .read = seq_read,
2404         .llseek = seq_lseek,
2405         .release = seq_release_private,
2406 };
2407
2408 static const struct file_operations ftrace_failures_fops = {
2409         .open = ftrace_failures_open,
2410         .read = seq_read,
2411         .llseek = seq_lseek,
2412         .release = seq_release_private,
2413 };
2414
2415 static const struct file_operations ftrace_filter_fops = {
2416         .open = ftrace_filter_open,
2417         .read = seq_read,
2418         .write = ftrace_filter_write,
2419         .llseek = no_llseek,
2420         .release = ftrace_filter_release,
2421 };
2422
2423 static const struct file_operations ftrace_notrace_fops = {
2424         .open = ftrace_notrace_open,
2425         .read = seq_read,
2426         .write = ftrace_notrace_write,
2427         .llseek = ftrace_regex_lseek,
2428         .release = ftrace_notrace_release,
2429 };
2430
2431 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2432
2433 static DEFINE_MUTEX(graph_lock);
2434
2435 int ftrace_graph_count;
2436 int ftrace_graph_filter_enabled;
2437 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2438
2439 static void *
2440 __g_next(struct seq_file *m, loff_t *pos)
2441 {
2442         if (*pos >= ftrace_graph_count)
2443                 return NULL;
2444         return &ftrace_graph_funcs[*pos];
2445 }
2446
2447 static void *
2448 g_next(struct seq_file *m, void *v, loff_t *pos)
2449 {
2450         (*pos)++;
2451         return __g_next(m, pos);
2452 }
2453
2454 static void *g_start(struct seq_file *m, loff_t *pos)
2455 {
2456         mutex_lock(&graph_lock);
2457
2458         /* Nothing, tell g_show to print all functions are enabled */
2459         if (!ftrace_graph_filter_enabled && !*pos)
2460                 return (void *)1;
2461
2462         return __g_next(m, pos);
2463 }
2464
2465 static void g_stop(struct seq_file *m, void *p)
2466 {
2467         mutex_unlock(&graph_lock);
2468 }
2469
2470 static int g_show(struct seq_file *m, void *v)
2471 {
2472         unsigned long *ptr = v;
2473
2474         if (!ptr)
2475                 return 0;
2476
2477         if (ptr == (unsigned long *)1) {
2478                 seq_printf(m, "#### all functions enabled ####\n");
2479                 return 0;
2480         }
2481
2482         seq_printf(m, "%ps\n", (void *)*ptr);
2483
2484         return 0;
2485 }
2486
2487 static const struct seq_operations ftrace_graph_seq_ops = {
2488         .start = g_start,
2489         .next = g_next,
2490         .stop = g_stop,
2491         .show = g_show,
2492 };
2493
2494 static int
2495 ftrace_graph_open(struct inode *inode, struct file *file)
2496 {
2497         int ret = 0;
2498
2499         if (unlikely(ftrace_disabled))
2500                 return -ENODEV;
2501
2502         mutex_lock(&graph_lock);
2503         if ((file->f_mode & FMODE_WRITE) &&
2504             (file->f_flags & O_TRUNC)) {
2505                 ftrace_graph_filter_enabled = 0;
2506                 ftrace_graph_count = 0;
2507                 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2508         }
2509         mutex_unlock(&graph_lock);
2510
2511         if (file->f_mode & FMODE_READ)
2512                 ret = seq_open(file, &ftrace_graph_seq_ops);
2513
2514         return ret;
2515 }
2516
2517 static int
2518 ftrace_graph_release(struct inode *inode, struct file *file)
2519 {
2520         if (file->f_mode & FMODE_READ)
2521                 seq_release(inode, file);
2522         return 0;
2523 }
2524
2525 static int
2526 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2527 {
2528         struct dyn_ftrace *rec;
2529         struct ftrace_page *pg;
2530         int search_len;
2531         int fail = 1;
2532         int type, not;
2533         char *search;
2534         bool exists;
2535         int i;
2536
2537         if (ftrace_disabled)
2538                 return -ENODEV;
2539
2540         /* decode regex */
2541         type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
2542         if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
2543                 return -EBUSY;
2544
2545         search_len = strlen(search);
2546
2547         mutex_lock(&ftrace_lock);
2548         do_for_each_ftrace_rec(pg, rec) {
2549
2550                 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2551                         continue;
2552
2553                 if (ftrace_match_record(rec, search, search_len, type)) {
2554                         /* if it is in the array */
2555                         exists = false;
2556                         for (i = 0; i < *idx; i++) {
2557                                 if (array[i] == rec->ip) {
2558                                         exists = true;
2559                                         break;
2560                                 }
2561                         }
2562
2563                         if (!not) {
2564                                 fail = 0;
2565                                 if (!exists) {
2566                                         array[(*idx)++] = rec->ip;
2567                                         if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2568                                                 goto out;
2569                                 }
2570                         } else {
2571                                 if (exists) {
2572                                         array[i] = array[--(*idx)];
2573                                         array[*idx] = 0;
2574                                         fail = 0;
2575                                 }
2576                         }
2577                 }
2578         } while_for_each_ftrace_rec();
2579 out:
2580         mutex_unlock(&ftrace_lock);
2581
2582         if (fail)
2583                 return -EINVAL;
2584
2585         ftrace_graph_filter_enabled = 1;
2586         return 0;
2587 }
2588
2589 static ssize_t
2590 ftrace_graph_write(struct file *file, const char __user *ubuf,
2591                    size_t cnt, loff_t *ppos)
2592 {
2593         struct trace_parser parser;
2594         ssize_t read, ret;
2595
2596         if (!cnt)
2597                 return 0;
2598
2599         mutex_lock(&graph_lock);
2600
2601         if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
2602                 ret = -ENOMEM;
2603                 goto out_unlock;
2604         }
2605
2606         read = trace_get_user(&parser, ubuf, cnt, ppos);
2607
2608         if (read >= 0 && trace_parser_loaded((&parser))) {
2609                 parser.buffer[parser.idx] = 0;
2610
2611                 /* we allow only one expression at a time */
2612                 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2613                                         parser.buffer);
2614                 if (ret)
2615                         goto out_free;
2616         }
2617
2618         ret = read;
2619
2620 out_free:
2621         trace_parser_put(&parser);
2622 out_unlock:
2623         mutex_unlock(&graph_lock);
2624
2625         return ret;
2626 }
2627
2628 static const struct file_operations ftrace_graph_fops = {
2629         .open           = ftrace_graph_open,
2630         .read           = seq_read,
2631         .write          = ftrace_graph_write,
2632         .release        = ftrace_graph_release,
2633 };
2634 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2635
2636 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2637 {
2638
2639         trace_create_file("available_filter_functions", 0444,
2640                         d_tracer, NULL, &ftrace_avail_fops);
2641
2642         trace_create_file("failures", 0444,
2643                         d_tracer, NULL, &ftrace_failures_fops);
2644
2645         trace_create_file("set_ftrace_filter", 0644, d_tracer,
2646                         NULL, &ftrace_filter_fops);
2647
2648         trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2649                                     NULL, &ftrace_notrace_fops);
2650
2651 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2652         trace_create_file("set_graph_function", 0444, d_tracer,
2653                                     NULL,
2654                                     &ftrace_graph_fops);
2655 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2656
2657         return 0;
2658 }
2659
2660 static int ftrace_process_locs(struct module *mod,
2661                                unsigned long *start,
2662                                unsigned long *end)
2663 {
2664         unsigned long *p;
2665         unsigned long addr;
2666         unsigned long flags;
2667
2668         mutex_lock(&ftrace_lock);
2669         p = start;
2670         while (p < end) {
2671                 addr = ftrace_call_adjust(*p++);
2672                 /*
2673                  * Some architecture linkers will pad between
2674                  * the different mcount_loc sections of different
2675                  * object files to satisfy alignments.
2676                  * Skip any NULL pointers.
2677                  */
2678                 if (!addr)
2679                         continue;
2680                 ftrace_record_ip(addr);
2681         }
2682
2683         /* disable interrupts to prevent kstop machine */
2684         local_irq_save(flags);
2685         ftrace_update_code(mod);
2686         local_irq_restore(flags);
2687         mutex_unlock(&ftrace_lock);
2688
2689         return 0;
2690 }
2691
2692 #ifdef CONFIG_MODULES
2693 void ftrace_release_mod(struct module *mod)
2694 {
2695         struct dyn_ftrace *rec;
2696         struct ftrace_page *pg;
2697
2698         if (ftrace_disabled)
2699                 return;
2700
2701         mutex_lock(&ftrace_lock);
2702         do_for_each_ftrace_rec(pg, rec) {
2703                 if (within_module_core(rec->ip, mod)) {
2704                         /*
2705                          * rec->ip is changed in ftrace_free_rec()
2706                          * It should not between s and e if record was freed.
2707                          */
2708                         FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2709                         ftrace_free_rec(rec);
2710                 }
2711         } while_for_each_ftrace_rec();
2712         mutex_unlock(&ftrace_lock);
2713 }
2714
2715 static void ftrace_init_module(struct module *mod,
2716                                unsigned long *start, unsigned long *end)
2717 {
2718         if (ftrace_disabled || start == end)
2719                 return;
2720         ftrace_process_locs(mod, start, end);
2721 }
2722
2723 static int ftrace_module_notify(struct notifier_block *self,
2724                                 unsigned long val, void *data)
2725 {
2726         struct module *mod = data;
2727
2728         switch (val) {
2729         case MODULE_STATE_COMING:
2730                 ftrace_init_module(mod, mod->ftrace_callsites,
2731                                    mod->ftrace_callsites +
2732                                    mod->num_ftrace_callsites);
2733                 break;
2734         case MODULE_STATE_GOING:
2735                 ftrace_release_mod(mod);
2736                 break;
2737         }
2738
2739         return 0;
2740 }
2741 #else
2742 static int ftrace_module_notify(struct notifier_block *self,
2743                                 unsigned long val, void *data)
2744 {
2745         return 0;
2746 }
2747 #endif /* CONFIG_MODULES */
2748
2749 struct notifier_block ftrace_module_nb = {
2750         .notifier_call = ftrace_module_notify,
2751         .priority = 0,
2752 };
2753
2754 extern unsigned long __start_mcount_loc[];
2755 extern unsigned long __stop_mcount_loc[];
2756
2757 void __init ftrace_init(void)
2758 {
2759         unsigned long count, addr, flags;
2760         int ret;
2761
2762         /* Keep the ftrace pointer to the stub */
2763         addr = (unsigned long)ftrace_stub;
2764
2765         local_irq_save(flags);
2766         ftrace_dyn_arch_init(&addr);
2767         local_irq_restore(flags);
2768
2769         /* ftrace_dyn_arch_init places the return code in addr */
2770         if (addr)
2771                 goto failed;
2772
2773         count = __stop_mcount_loc - __start_mcount_loc;
2774
2775         ret = ftrace_dyn_table_alloc(count);
2776         if (ret)
2777                 goto failed;
2778
2779         last_ftrace_enabled = ftrace_enabled = 1;
2780
2781         ret = ftrace_process_locs(NULL,
2782                                   __start_mcount_loc,
2783                                   __stop_mcount_loc);
2784
2785         ret = register_module_notifier(&ftrace_module_nb);
2786         if (ret)
2787                 pr_warning("Failed to register trace ftrace module notifier\n");
2788
2789         set_ftrace_early_filters();
2790
2791         return;
2792  failed:
2793         ftrace_disabled = 1;
2794 }
2795
2796 #else
2797
2798 static int __init ftrace_nodyn_init(void)
2799 {
2800         ftrace_enabled = 1;
2801         return 0;
2802 }
2803 device_initcall(ftrace_nodyn_init);
2804
2805 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2806 static inline void ftrace_startup_enable(int command) { }
2807 /* Keep as macros so we do not need to define the commands */
2808 # define ftrace_startup(command)        do { } while (0)
2809 # define ftrace_shutdown(command)       do { } while (0)
2810 # define ftrace_startup_sysctl()        do { } while (0)
2811 # define ftrace_shutdown_sysctl()       do { } while (0)
2812 #endif /* CONFIG_DYNAMIC_FTRACE */
2813
2814 static void clear_ftrace_swapper(void)
2815 {
2816         struct task_struct *p;
2817         int cpu;
2818
2819         get_online_cpus();
2820         for_each_online_cpu(cpu) {
2821                 p = idle_task(cpu);
2822                 clear_tsk_trace_trace(p);
2823         }
2824         put_online_cpus();
2825 }
2826
2827 static void set_ftrace_swapper(void)
2828 {
2829         struct task_struct *p;
2830         int cpu;
2831
2832         get_online_cpus();
2833         for_each_online_cpu(cpu) {
2834                 p = idle_task(cpu);
2835                 set_tsk_trace_trace(p);
2836         }
2837         put_online_cpus();
2838 }
2839
2840 static void clear_ftrace_pid(struct pid *pid)
2841 {
2842         struct task_struct *p;
2843
2844         rcu_read_lock();
2845         do_each_pid_task(pid, PIDTYPE_PID, p) {
2846                 clear_tsk_trace_trace(p);
2847         } while_each_pid_task(pid, PIDTYPE_PID, p);
2848         rcu_read_unlock();
2849
2850         put_pid(pid);
2851 }
2852
2853 static void set_ftrace_pid(struct pid *pid)
2854 {
2855         struct task_struct *p;
2856
2857         rcu_read_lock();
2858         do_each_pid_task(pid, PIDTYPE_PID, p) {
2859                 set_tsk_trace_trace(p);
2860         } while_each_pid_task(pid, PIDTYPE_PID, p);
2861         rcu_read_unlock();
2862 }
2863
2864 static void clear_ftrace_pid_task(struct pid *pid)
2865 {
2866         if (pid == ftrace_swapper_pid)
2867                 clear_ftrace_swapper();
2868         else
2869                 clear_ftrace_pid(pid);
2870 }
2871
2872 static void set_ftrace_pid_task(struct pid *pid)
2873 {
2874         if (pid == ftrace_swapper_pid)
2875                 set_ftrace_swapper();
2876         else
2877                 set_ftrace_pid(pid);
2878 }
2879
2880 static int ftrace_pid_add(int p)
2881 {
2882         struct pid *pid;
2883         struct ftrace_pid *fpid;
2884         int ret = -EINVAL;
2885
2886         mutex_lock(&ftrace_lock);
2887
2888         if (!p)
2889                 pid = ftrace_swapper_pid;
2890         else
2891                 pid = find_get_pid(p);
2892
2893         if (!pid)
2894                 goto out;
2895
2896         ret = 0;
2897
2898         list_for_each_entry(fpid, &ftrace_pids, list)
2899                 if (fpid->pid == pid)
2900                         goto out_put;
2901
2902         ret = -ENOMEM;
2903
2904         fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
2905         if (!fpid)
2906                 goto out_put;
2907
2908         list_add(&fpid->list, &ftrace_pids);
2909         fpid->pid = pid;
2910
2911         set_ftrace_pid_task(pid);
2912
2913         ftrace_update_pid_func();
2914         ftrace_startup_enable(0);
2915
2916         mutex_unlock(&ftrace_lock);
2917         return 0;
2918
2919 out_put:
2920         if (pid != ftrace_swapper_pid)
2921                 put_pid(pid);
2922
2923 out:
2924         mutex_unlock(&ftrace_lock);
2925         return ret;
2926 }
2927
2928 static void ftrace_pid_reset(void)
2929 {
2930         struct ftrace_pid *fpid, *safe;
2931
2932         mutex_lock(&ftrace_lock);
2933         list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
2934                 struct pid *pid = fpid->pid;
2935
2936                 clear_ftrace_pid_task(pid);
2937
2938                 list_del(&fpid->list);
2939                 kfree(fpid);
2940         }
2941
2942         ftrace_update_pid_func();
2943         ftrace_startup_enable(0);
2944
2945         mutex_unlock(&ftrace_lock);
2946 }
2947
2948 static void *fpid_start(struct seq_file *m, loff_t *pos)
2949 {
2950         mutex_lock(&ftrace_lock);
2951
2952         if (list_empty(&ftrace_pids) && (!*pos))
2953                 return (void *) 1;
2954
2955         return seq_list_start(&ftrace_pids, *pos);
2956 }
2957
2958 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
2959 {
2960         if (v == (void *)1)
2961                 return NULL;
2962
2963         return seq_list_next(v, &ftrace_pids, pos);
2964 }
2965
2966 static void fpid_stop(struct seq_file *m, void *p)
2967 {
2968         mutex_unlock(&ftrace_lock);
2969 }
2970
2971 static int fpid_show(struct seq_file *m, void *v)
2972 {
2973         const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
2974
2975         if (v == (void *)1) {
2976                 seq_printf(m, "no pid\n");
2977                 return 0;
2978         }
2979
2980         if (fpid->pid == ftrace_swapper_pid)
2981                 seq_printf(m, "swapper tasks\n");
2982         else
2983                 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
2984
2985         return 0;
2986 }
2987
2988 static const struct seq_operations ftrace_pid_sops = {
2989         .start = fpid_start,
2990         .next = fpid_next,
2991         .stop = fpid_stop,
2992         .show = fpid_show,
2993 };
2994
2995 static int
2996 ftrace_pid_open(struct inode *inode, struct file *file)
2997 {
2998         int ret = 0;
2999
3000         if ((file->f_mode & FMODE_WRITE) &&
3001             (file->f_flags & O_TRUNC))
3002                 ftrace_pid_reset();
3003
3004         if (file->f_mode & FMODE_READ)
3005                 ret = seq_open(file, &ftrace_pid_sops);
3006
3007         return ret;
3008 }
3009
3010 static ssize_t
3011 ftrace_pid_write(struct file *filp, const char __user *ubuf,
3012                    size_t cnt, loff_t *ppos)
3013 {
3014         char buf[64], *tmp;
3015         long val;
3016         int ret;
3017
3018         if (cnt >= sizeof(buf))
3019                 return -EINVAL;
3020
3021         if (copy_from_user(&buf, ubuf, cnt))
3022                 return -EFAULT;
3023
3024         buf[cnt] = 0;
3025
3026         /*
3027          * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
3028          * to clean the filter quietly.
3029          */
3030         tmp = strstrip(buf);
3031         if (strlen(tmp) == 0)
3032                 return 1;
3033
3034         ret = strict_strtol(tmp, 10, &val);
3035         if (ret < 0)
3036                 return ret;
3037
3038         ret = ftrace_pid_add(val);
3039
3040         return ret ? ret : cnt;
3041 }
3042
3043 static int
3044 ftrace_pid_release(struct inode *inode, struct file *file)
3045 {
3046         if (file->f_mode & FMODE_READ)
3047                 seq_release(inode, file);
3048
3049         return 0;
3050 }
3051
3052 static const struct file_operations ftrace_pid_fops = {
3053         .open           = ftrace_pid_open,
3054         .write          = ftrace_pid_write,
3055         .read           = seq_read,
3056         .llseek         = seq_lseek,
3057         .release        = ftrace_pid_release,
3058 };
3059
3060 static __init int ftrace_init_debugfs(void)
3061 {
3062         struct dentry *d_tracer;
3063
3064         d_tracer = tracing_init_dentry();
3065         if (!d_tracer)
3066                 return 0;
3067
3068         ftrace_init_dyn_debugfs(d_tracer);
3069
3070         trace_create_file("set_ftrace_pid", 0644, d_tracer,
3071                             NULL, &ftrace_pid_fops);
3072
3073         ftrace_profile_debugfs(d_tracer);
3074
3075         return 0;
3076 }
3077 fs_initcall(ftrace_init_debugfs);
3078
3079 /**
3080  * ftrace_kill - kill ftrace
3081  *
3082  * This function should be used by panic code. It stops ftrace
3083  * but in a not so nice way. If you need to simply kill ftrace
3084  * from a non-atomic section, use ftrace_kill.
3085  */
3086 void ftrace_kill(void)
3087 {
3088         ftrace_disabled = 1;
3089         ftrace_enabled = 0;
3090         clear_ftrace_function();
3091 }
3092
3093 /**
3094  * register_ftrace_function - register a function for profiling
3095  * @ops - ops structure that holds the function for profiling.
3096  *
3097  * Register a function to be called by all functions in the
3098  * kernel.
3099  *
3100  * Note: @ops->func and all the functions it calls must be labeled
3101  *       with "notrace", otherwise it will go into a
3102  *       recursive loop.
3103  */
3104 int register_ftrace_function(struct ftrace_ops *ops)
3105 {
3106         int ret;
3107
3108         if (unlikely(ftrace_disabled))
3109                 return -1;
3110
3111         mutex_lock(&ftrace_lock);
3112
3113         ret = __register_ftrace_function(ops);
3114         ftrace_startup(0);
3115
3116         mutex_unlock(&ftrace_lock);
3117         return ret;
3118 }
3119
3120 /**
3121  * unregister_ftrace_function - unregister a function for profiling.
3122  * @ops - ops structure that holds the function to unregister
3123  *
3124  * Unregister a function that was added to be called by ftrace profiling.
3125  */
3126 int unregister_ftrace_function(struct ftrace_ops *ops)
3127 {
3128         int ret;
3129
3130         mutex_lock(&ftrace_lock);
3131         ret = __unregister_ftrace_function(ops);
3132         ftrace_shutdown(0);
3133         mutex_unlock(&ftrace_lock);
3134
3135         return ret;
3136 }
3137
3138 int
3139 ftrace_enable_sysctl(struct ctl_table *table, int write,
3140                      void __user *buffer, size_t *lenp,
3141                      loff_t *ppos)
3142 {
3143         int ret;
3144
3145         if (unlikely(ftrace_disabled))
3146                 return -ENODEV;
3147
3148         mutex_lock(&ftrace_lock);
3149
3150         ret  = proc_dointvec(table, write, buffer, lenp, ppos);
3151
3152         if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3153                 goto out;
3154
3155         last_ftrace_enabled = !!ftrace_enabled;
3156
3157         if (ftrace_enabled) {
3158
3159                 ftrace_startup_sysctl();
3160
3161                 /* we are starting ftrace again */
3162                 if (ftrace_list != &ftrace_list_end) {
3163                         if (ftrace_list->next == &ftrace_list_end)
3164                                 ftrace_trace_function = ftrace_list->func;
3165                         else
3166                                 ftrace_trace_function = ftrace_list_func;
3167                 }
3168
3169         } else {
3170                 /* stopping ftrace calls (just send to ftrace_stub) */
3171                 ftrace_trace_function = ftrace_stub;
3172
3173                 ftrace_shutdown_sysctl();
3174         }
3175
3176  out:
3177         mutex_unlock(&ftrace_lock);
3178         return ret;
3179 }
3180
3181 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3182
3183 static int ftrace_graph_active;
3184 static struct notifier_block ftrace_suspend_notifier;
3185
3186 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3187 {
3188         return 0;
3189 }
3190
3191 /* The callbacks that hook a function */
3192 trace_func_graph_ret_t ftrace_graph_return =
3193                         (trace_func_graph_ret_t)ftrace_stub;
3194 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3195
3196 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3197 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3198 {
3199         int i;
3200         int ret = 0;
3201         unsigned long flags;
3202         int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3203         struct task_struct *g, *t;
3204
3205         for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3206                 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3207                                         * sizeof(struct ftrace_ret_stack),
3208                                         GFP_KERNEL);
3209                 if (!ret_stack_list[i]) {
3210                         start = 0;
3211                         end = i;
3212                         ret = -ENOMEM;
3213                         goto free;
3214                 }
3215         }
3216
3217         read_lock_irqsave(&tasklist_lock, flags);
3218         do_each_thread(g, t) {
3219                 if (start == end) {
3220                         ret = -EAGAIN;
3221                         goto unlock;
3222                 }
3223
3224                 if (t->ret_stack == NULL) {
3225                         atomic_set(&t->tracing_graph_pause, 0);
3226                         atomic_set(&t->trace_overrun, 0);
3227                         t->curr_ret_stack = -1;
3228                         /* Make sure the tasks see the -1 first: */
3229                         smp_wmb();
3230                         t->ret_stack = ret_stack_list[start++];
3231                 }
3232         } while_each_thread(g, t);
3233
3234 unlock:
3235         read_unlock_irqrestore(&tasklist_lock, flags);
3236 free:
3237         for (i = start; i < end; i++)
3238                 kfree(ret_stack_list[i]);
3239         return ret;
3240 }
3241
3242 static void
3243 ftrace_graph_probe_sched_switch(void *ignore,
3244                         struct task_struct *prev, struct task_struct *next)
3245 {
3246         unsigned long long timestamp;
3247         int index;
3248
3249         /*
3250          * Does the user want to count the time a function was asleep.
3251          * If so, do not update the time stamps.
3252          */
3253         if (trace_flags & TRACE_ITER_SLEEP_TIME)
3254                 return;
3255
3256         timestamp = trace_clock_local();
3257
3258         prev->ftrace_timestamp = timestamp;
3259
3260         /* only process tasks that we timestamped */
3261         if (!next->ftrace_timestamp)
3262                 return;
3263
3264         /*
3265          * Update all the counters in next to make up for the
3266          * time next was sleeping.
3267          */
3268         timestamp -= next->ftrace_timestamp;
3269
3270         for (index = next->curr_ret_stack; index >= 0; index--)
3271                 next->ret_stack[index].calltime += timestamp;
3272 }
3273
3274 /* Allocate a return stack for each task */
3275 static int start_graph_tracing(void)
3276 {
3277         struct ftrace_ret_stack **ret_stack_list;
3278         int ret, cpu;
3279
3280         ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3281                                 sizeof(struct ftrace_ret_stack *),
3282                                 GFP_KERNEL);
3283
3284         if (!ret_stack_list)
3285                 return -ENOMEM;
3286
3287         /* The cpu_boot init_task->ret_stack will never be freed */
3288         for_each_online_cpu(cpu) {
3289                 if (!idle_task(cpu)->ret_stack)
3290                         ftrace_graph_init_task(idle_task(cpu));
3291         }
3292
3293         do {
3294                 ret = alloc_retstack_tasklist(ret_stack_list);
3295         } while (ret == -EAGAIN);
3296
3297         if (!ret) {
3298                 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3299                 if (ret)
3300                         pr_info("ftrace_graph: Couldn't activate tracepoint"
3301                                 " probe to kernel_sched_switch\n");
3302         }
3303
3304         kfree(ret_stack_list);
3305         return ret;
3306 }
3307
3308 /*
3309  * Hibernation protection.
3310  * The state of the current task is too much unstable during
3311  * suspend/restore to disk. We want to protect against that.
3312  */
3313 static int
3314 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3315                                                         void *unused)
3316 {
3317         switch (state) {
3318         case PM_HIBERNATION_PREPARE:
3319                 pause_graph_tracing();
3320                 break;
3321
3322         case PM_POST_HIBERNATION:
3323                 unpause_graph_tracing();
3324                 break;
3325         }
3326         return NOTIFY_DONE;
3327 }
3328
3329 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3330                         trace_func_graph_ent_t entryfunc)
3331 {
3332         int ret = 0;
3333
3334         mutex_lock(&ftrace_lock);
3335
3336         /* we currently allow only one tracer registered at a time */
3337         if (ftrace_graph_active) {
3338                 ret = -EBUSY;
3339                 goto out;
3340         }
3341
3342         ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3343         register_pm_notifier(&ftrace_suspend_notifier);
3344
3345         ftrace_graph_active++;
3346         ret = start_graph_tracing();
3347         if (ret) {
3348                 ftrace_graph_active--;
3349                 goto out;
3350         }
3351
3352         ftrace_graph_return = retfunc;
3353         ftrace_graph_entry = entryfunc;
3354
3355         ftrace_startup(FTRACE_START_FUNC_RET);
3356
3357 out:
3358         mutex_unlock(&ftrace_lock);
3359         return ret;
3360 }
3361
3362 void unregister_ftrace_graph(void)
3363 {
3364         mutex_lock(&ftrace_lock);
3365
3366         if (unlikely(!ftrace_graph_active))
3367                 goto out;
3368
3369         ftrace_graph_active--;
3370         ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3371         ftrace_graph_entry = ftrace_graph_entry_stub;
3372         ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3373         unregister_pm_notifier(&ftrace_suspend_notifier);
3374         unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3375
3376  out:
3377         mutex_unlock(&ftrace_lock);
3378 }
3379
3380 /* Allocate a return stack for newly created task */
3381 void ftrace_graph_init_task(struct task_struct *t)
3382 {
3383         /* Make sure we do not use the parent ret_stack */
3384         t->ret_stack = NULL;
3385         t->curr_ret_stack = -1;
3386
3387         if (ftrace_graph_active) {
3388                 struct ftrace_ret_stack *ret_stack;
3389
3390                 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3391                                 * sizeof(struct ftrace_ret_stack),
3392                                 GFP_KERNEL);
3393                 if (!ret_stack)
3394                         return;
3395                 atomic_set(&t->tracing_graph_pause, 0);
3396                 atomic_set(&t->trace_overrun, 0);
3397                 t->ftrace_timestamp = 0;
3398                 /* make curr_ret_stack visable before we add the ret_stack */
3399                 smp_wmb();
3400                 t->ret_stack = ret_stack;
3401         }
3402 }
3403
3404 void ftrace_graph_exit_task(struct task_struct *t)
3405 {
3406         struct ftrace_ret_stack *ret_stack = t->ret_stack;
3407
3408         t->ret_stack = NULL;
3409         /* NULL must become visible to IRQs before we free it: */
3410         barrier();
3411
3412         kfree(ret_stack);
3413 }
3414
3415 void ftrace_graph_stop(void)
3416 {
3417         ftrace_stop();
3418 }
3419 #endif