kernel/trace/ftrace.c

   1 /*
   2  * Infrastructure for profiling code inserted by 'gcc -pg'.
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally ported from the -rt patch by:
   8  *   Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code in the latency_tracer, that is:
  11  *
  12  *  Copyright (C) 2004-2006 Ingo Molnar
  13  *  Copyright (C) 2004 William Lee Irwin III
  14  */
  15
  16 #include <linux/stop_machine.h>
  17 #include <linux/clocksource.h>
  18 #include <linux/kallsyms.h>
  19 #include <linux/seq_file.h>
  20 #include <linux/suspend.h>
  21 #include <linux/debugfs.h>
  22 #include <linux/hardirq.h>
  23 #include <linux/kthread.h>
  24 #include <linux/uaccess.h>
  25 #include <linux/ftrace.h>
  26 #include <linux/sysctl.h>
  27 #include <linux/slab.h>
  28 #include <linux/ctype.h>
  29 #include <linux/list.h>
  30 #include <linux/hash.h>
  31 #include <linux/rcupdate.h>
  32
  33 #include <trace/events/sched.h>
  34
  35 #include <asm/ftrace.h>
  36 #include <asm/setup.h>
  37
  38 #include "trace_output.h"
  39 #include "trace_stat.h"
  40
  41 #define FTRACE_WARN_ON(cond)                    \
  42         do {                                    \
  43                 if (WARN_ON(cond))              \
  44                         ftrace_kill();          \
  45         } while (0)
  46
  47 #define FTRACE_WARN_ON_ONCE(cond)               \
  48         do {                                    \
  49                 if (WARN_ON_ONCE(cond))         \
  50                         ftrace_kill();          \
  51         } while (0)
  52
  53 /* hash bits for specific function selection */
  54 #define FTRACE_HASH_BITS 7
  55 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
  56
  57 /* ftrace_enabled is a method to turn ftrace on or off */
  58 int ftrace_enabled __read_mostly;
  59 static int last_ftrace_enabled;
  60
  61 /* Quick disabling of function tracer. */
  62 int function_trace_stop;
  63
  64 /* List for set_ftrace_pid's pids. */
  65 LIST_HEAD(ftrace_pids);
  66 struct ftrace_pid {
  67         struct list_head list;
  68         struct pid *pid;
  69 };
  70
  71 /*
  72  * ftrace_disabled is set when an anomaly is discovered.
  73  * ftrace_disabled is much stronger than ftrace_enabled.
  74  */
  75 static int ftrace_disabled __read_mostly;
  76
  77 static DEFINE_MUTEX(ftrace_lock);
  78
  79 static struct ftrace_ops ftrace_list_end __read_mostly =
  80 {
  81         .func           = ftrace_stub,
  82 };
  83
  84 static struct ftrace_ops *ftrace_list __read_mostly = &ftrace_list_end;
  85 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
  86 ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
  87 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
  88
  89 /*
  90  * Traverse the ftrace_list, invoking all entries.  The reason that we
  91  * can use rcu_dereference_raw() is that elements removed from this list
  92  * are simply leaked, so there is no need to interact with a grace-period
  93  * mechanism.  The rcu_dereference_raw() calls are needed to handle
  94  * concurrent insertions into the ftrace_list.
  95  *
  96  * Silly Alpha and silly pointer-speculation compiler optimizations!
  97  */
  98 static void ftrace_list_func(unsigned long ip, unsigned long parent_ip)
  99 {
 100         struct ftrace_ops *op = rcu_dereference_raw(ftrace_list); /*see above*/
 101
 102         while (op != &ftrace_list_end) {
 103                 op->func(ip, parent_ip);
 104                 op = rcu_dereference_raw(op->next); /*see above*/
 105         };
 106 }
 107
 108 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
 109 {
 110         if (!test_tsk_trace_trace(current))
 111                 return;
 112
 113         ftrace_pid_function(ip, parent_ip);
 114 }
 115
 116 static void set_ftrace_pid_function(ftrace_func_t func)
 117 {
 118         /* do not set ftrace_pid_function to itself! */
 119         if (func != ftrace_pid_func)
 120                 ftrace_pid_function = func;
 121 }
 122
 123 /**
 124  * clear_ftrace_function - reset the ftrace function
 125  *
 126  * This NULLs the ftrace function and in essence stops
 127  * tracing.  There may be lag
 128  */
 129 void clear_ftrace_function(void)
 130 {
 131         ftrace_trace_function = ftrace_stub;
 132         __ftrace_trace_function = ftrace_stub;
 133         ftrace_pid_function = ftrace_stub;
 134 }
 135
 136 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
 137 /*
 138  * For those archs that do not test ftrace_trace_stop in their
 139  * mcount call site, we need to do it from C.
 140  */
 141 static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
 142 {
 143         if (function_trace_stop)
 144                 return;
 145
 146         __ftrace_trace_function(ip, parent_ip);
 147 }
 148 #endif
 149
 150 static int __register_ftrace_function(struct ftrace_ops *ops)
 151 {
 152         ops->next = ftrace_list;
 153         /*
 154          * We are entering ops into the ftrace_list but another
 155          * CPU might be walking that list. We need to make sure
 156          * the ops->next pointer is valid before another CPU sees
 157          * the ops pointer included into the ftrace_list.
 158          */
 159         rcu_assign_pointer(ftrace_list, ops);
 160
 161         if (ftrace_enabled) {
 162                 ftrace_func_t func;
 163
 164                 if (ops->next == &ftrace_list_end)
 165                         func = ops->func;
 166                 else
 167                         func = ftrace_list_func;
 168
 169                 if (!list_empty(&ftrace_pids)) {
 170                         set_ftrace_pid_function(func);
 171                         func = ftrace_pid_func;
 172                 }
 173
 174                 /*
 175                  * For one func, simply call it directly.
 176                  * For more than one func, call the chain.
 177                  */
 178 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
 179                 ftrace_trace_function = func;
 180 #else
 181                 __ftrace_trace_function = func;
 182                 ftrace_trace_function = ftrace_test_stop_func;
 183 #endif
 184         }
 185
 186         return 0;
 187 }
 188
 189 static int __unregister_ftrace_function(struct ftrace_ops *ops)
 190 {
 191         struct ftrace_ops **p;
 192
 193         /*
 194          * If we are removing the last function, then simply point
 195          * to the ftrace_stub.
 196          */
 197         if (ftrace_list == ops && ops->next == &ftrace_list_end) {
 198                 ftrace_trace_function = ftrace_stub;
 199                 ftrace_list = &ftrace_list_end;
 200                 return 0;
 201         }
 202
 203         for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next)
 204                 if (*p == ops)
 205                         break;
 206
 207         if (*p != ops)
 208                 return -1;
 209
 210         *p = (*p)->next;
 211
 212         if (ftrace_enabled) {
 213                 /* If we only have one func left, then call that directly */
 214                 if (ftrace_list->next == &ftrace_list_end) {
 215                         ftrace_func_t func = ftrace_list->func;
 216
 217                         if (!list_empty(&ftrace_pids)) {
 218                                 set_ftrace_pid_function(func);
 219                                 func = ftrace_pid_func;
 220                         }
 221 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
 222                         ftrace_trace_function = func;
 223 #else
 224                         __ftrace_trace_function = func;
 225 #endif
 226                 }
 227         }
 228
 229         return 0;
 230 }
 231
 232 static void ftrace_update_pid_func(void)
 233 {
 234         ftrace_func_t func;
 235
 236         if (ftrace_trace_function == ftrace_stub)
 237                 return;
 238
 239 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
 240         func = ftrace_trace_function;
 241 #else
 242         func = __ftrace_trace_function;
 243 #endif
 244
 245         if (!list_empty(&ftrace_pids)) {
 246                 set_ftrace_pid_function(func);
 247                 func = ftrace_pid_func;
 248         } else {
 249                 if (func == ftrace_pid_func)
 250                         func = ftrace_pid_function;
 251         }
 252
 253 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
 254         ftrace_trace_function = func;
 255 #else
 256         __ftrace_trace_function = func;
 257 #endif
 258 }
 259
 260 #ifdef CONFIG_FUNCTION_PROFILER
 261 struct ftrace_profile {
 262         struct hlist_node               node;
 263         unsigned long                   ip;
 264         unsigned long                   counter;
 265 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 266         unsigned long long              time;
 267         unsigned long long              time_squared;
 268 #endif
 269 };
 270
 271 struct ftrace_profile_page {
 272         struct ftrace_profile_page      *next;
 273         unsigned long                   index;
 274         struct ftrace_profile           records[];
 275 };
 276
 277 struct ftrace_profile_stat {
 278         atomic_t                        disabled;
 279         struct hlist_head               *hash;
 280         struct ftrace_profile_page      *pages;
 281         struct ftrace_profile_page      *start;
 282         struct tracer_stat              stat;
 283 };
 284
 285 #define PROFILE_RECORDS_SIZE                                            \
 286         (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
 287
 288 #define PROFILES_PER_PAGE                                       \
 289         (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
 290
 291 static int ftrace_profile_bits __read_mostly;
 292 static int ftrace_profile_enabled __read_mostly;
 293
 294 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
 295 static DEFINE_MUTEX(ftrace_profile_lock);
 296
 297 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
 298
 299 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
 300
 301 static void *
 302 function_stat_next(void *v, int idx)
 303 {
 304         struct ftrace_profile *rec = v;
 305         struct ftrace_profile_page *pg;
 306
 307         pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
 308
 309  again:
 310         if (idx != 0)
 311                 rec++;
 312
 313         if ((void *)rec >= (void *)&pg->records[pg->index]) {
 314                 pg = pg->next;
 315                 if (!pg)
 316                         return NULL;
 317                 rec = &pg->records[0];
 318                 if (!rec->counter)
 319                         goto again;
 320         }
 321
 322         return rec;
 323 }
 324
 325 static void *function_stat_start(struct tracer_stat *trace)
 326 {
 327         struct ftrace_profile_stat *stat =
 328                 container_of(trace, struct ftrace_profile_stat, stat);
 329
 330         if (!stat || !stat->start)
 331                 return NULL;
 332
 333         return function_stat_next(&stat->start->records[0], 0);
 334 }
 335
 336 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 337 /* function graph compares on total time */
 338 static int function_stat_cmp(void *p1, void *p2)
 339 {
 340         struct ftrace_profile *a = p1;
 341         struct ftrace_profile *b = p2;
 342
 343         if (a->time < b->time)
 344                 return -1;
 345         if (a->time > b->time)
 346                 return 1;
 347         else
 348                 return 0;
 349 }
 350 #else
 351 /* not function graph compares against hits */
 352 static int function_stat_cmp(void *p1, void *p2)
 353 {
 354         struct ftrace_profile *a = p1;
 355         struct ftrace_profile *b = p2;
 356
 357         if (a->counter < b->counter)
 358                 return -1;
 359         if (a->counter > b->counter)
 360                 return 1;
 361         else
 362                 return 0;
 363 }
 364 #endif
 365
 366 static int function_stat_headers(struct seq_file *m)
 367 {
 368 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 369         seq_printf(m, "  Function                               "
 370                    "Hit    Time            Avg             s^2\n"
 371                       "  --------                               "
 372                    "---    ----            ---             ---\n");
 373 #else
 374         seq_printf(m, "  Function                               Hit\n"
 375                       "  --------                               ---\n");
 376 #endif
 377         return 0;
 378 }
 379
 380 static int function_stat_show(struct seq_file *m, void *v)
 381 {
 382         struct ftrace_profile *rec = v;
 383         char str[KSYM_SYMBOL_LEN];
 384         int ret = 0;
 385 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 386         static struct trace_seq s;
 387         unsigned long long avg;
 388         unsigned long long stddev;
 389 #endif
 390         mutex_lock(&ftrace_profile_lock);
 391
 392         /* we raced with function_profile_reset() */
 393         if (unlikely(rec->counter == 0)) {
 394                 ret = -EBUSY;
 395                 goto out;
 396         }
 397
 398         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 399         seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
 400
 401 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 402         seq_printf(m, "    ");
 403         avg = rec->time;
 404         do_div(avg, rec->counter);
 405
 406         /* Sample standard deviation (s^2) */
 407         if (rec->counter <= 1)
 408                 stddev = 0;
 409         else {
 410                 stddev = rec->time_squared - rec->counter * avg * avg;
 411                 /*
 412                  * Divide only 1000 for ns^2 -> us^2 conversion.
 413                  * trace_print_graph_duration will divide 1000 again.
 414                  */
 415                 do_div(stddev, (rec->counter - 1) * 1000);
 416         }
 417
 418         trace_seq_init(&s);
 419         trace_print_graph_duration(rec->time, &s);
 420         trace_seq_puts(&s, "    ");
 421         trace_print_graph_duration(avg, &s);
 422         trace_seq_puts(&s, "    ");
 423         trace_print_graph_duration(stddev, &s);
 424         trace_print_seq(m, &s);
 425 #endif
 426         seq_putc(m, '\n');
 427 out:
 428         mutex_unlock(&ftrace_profile_lock);
 429
 430         return ret;
 431 }
 432
 433 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
 434 {
 435         struct ftrace_profile_page *pg;
 436
 437         pg = stat->pages = stat->start;
 438
 439         while (pg) {
 440                 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
 441                 pg->index = 0;
 442                 pg = pg->next;
 443         }
 444
 445         memset(stat->hash, 0,
 446                FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
 447 }
 448
 449 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
 450 {
 451         struct ftrace_profile_page *pg;
 452         int functions;
 453         int pages;
 454         int i;
 455
 456         /* If we already allocated, do nothing */
 457         if (stat->pages)
 458                 return 0;
 459
 460         stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
 461         if (!stat->pages)
 462                 return -ENOMEM;
 463
 464 #ifdef CONFIG_DYNAMIC_FTRACE
 465         functions = ftrace_update_tot_cnt;
 466 #else
 467         /*
 468          * We do not know the number of functions that exist because
 469          * dynamic tracing is what counts them. With past experience
 470          * we have around 20K functions. That should be more than enough.
 471          * It is highly unlikely we will execute every function in
 472          * the kernel.
 473          */
 474         functions = 20000;
 475 #endif
 476
 477         pg = stat->start = stat->pages;
 478
 479         pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
 480
 481         for (i = 0; i < pages; i++) {
 482                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
 483                 if (!pg->next)
 484                         goto out_free;
 485                 pg = pg->next;
 486         }
 487
 488         return 0;
 489
 490  out_free:
 491         pg = stat->start;
 492         while (pg) {
 493                 unsigned long tmp = (unsigned long)pg;
 494
 495                 pg = pg->next;
 496                 free_page(tmp);
 497         }
 498
 499         free_page((unsigned long)stat->pages);
 500         stat->pages = NULL;
 501         stat->start = NULL;
 502
 503         return -ENOMEM;
 504 }
 505
 506 static int ftrace_profile_init_cpu(int cpu)
 507 {
 508         struct ftrace_profile_stat *stat;
 509         int size;
 510
 511         stat = &per_cpu(ftrace_profile_stats, cpu);
 512
 513         if (stat->hash) {
 514                 /* If the profile is already created, simply reset it */
 515                 ftrace_profile_reset(stat);
 516                 return 0;
 517         }
 518
 519         /*
 520          * We are profiling all functions, but usually only a few thousand
 521          * functions are hit. We'll make a hash of 1024 items.
 522          */
 523         size = FTRACE_PROFILE_HASH_SIZE;
 524
 525         stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
 526
 527         if (!stat->hash)
 528                 return -ENOMEM;
 529
 530         if (!ftrace_profile_bits) {
 531                 size--;
 532
 533                 for (; size; size >>= 1)
 534                         ftrace_profile_bits++;
 535         }
 536
 537         /* Preallocate the function profiling pages */
 538         if (ftrace_profile_pages_init(stat) < 0) {
 539                 kfree(stat->hash);
 540                 stat->hash = NULL;
 541                 return -ENOMEM;
 542         }
 543
 544         return 0;
 545 }
 546
 547 static int ftrace_profile_init(void)
 548 {
 549         int cpu;
 550         int ret = 0;
 551
 552         for_each_online_cpu(cpu) {
 553                 ret = ftrace_profile_init_cpu(cpu);
 554                 if (ret)
 555                         break;
 556         }
 557
 558         return ret;
 559 }
 560
 561 /* interrupts must be disabled */
 562 static struct ftrace_profile *
 563 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
 564 {
 565         struct ftrace_profile *rec;
 566         struct hlist_head *hhd;
 567         struct hlist_node *n;
 568         unsigned long key;
 569
 570         key = hash_long(ip, ftrace_profile_bits);
 571         hhd = &stat->hash[key];
 572
 573         if (hlist_empty(hhd))
 574                 return NULL;
 575
 576         hlist_for_each_entry_rcu(rec, n, hhd, node) {
 577                 if (rec->ip == ip)
 578                         return rec;
 579         }
 580
 581         return NULL;
 582 }
 583
 584 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
 585                                struct ftrace_profile *rec)
 586 {
 587         unsigned long key;
 588
 589         key = hash_long(rec->ip, ftrace_profile_bits);
 590         hlist_add_head_rcu(&rec->node, &stat->hash[key]);
 591 }
 592
 593 /*
 594  * The memory is already allocated, this simply finds a new record to use.
 595  */
 596 static struct ftrace_profile *
 597 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
 598 {
 599         struct ftrace_profile *rec = NULL;
 600
 601         /* prevent recursion (from NMIs) */
 602         if (atomic_inc_return(&stat->disabled) != 1)
 603                 goto out;
 604
 605         /*
 606          * Try to find the function again since an NMI
 607          * could have added it
 608          */
 609         rec = ftrace_find_profiled_func(stat, ip);
 610         if (rec)
 611                 goto out;
 612
 613         if (stat->pages->index == PROFILES_PER_PAGE) {
 614                 if (!stat->pages->next)
 615                         goto out;
 616                 stat->pages = stat->pages->next;
 617         }
 618
 619         rec = &stat->pages->records[stat->pages->index++];
 620         rec->ip = ip;
 621         ftrace_add_profile(stat, rec);
 622
 623  out:
 624         atomic_dec(&stat->disabled);
 625
 626         return rec;
 627 }
 628
 629 static void
 630 function_profile_call(unsigned long ip, unsigned long parent_ip)
 631 {
 632         struct ftrace_profile_stat *stat;
 633         struct ftrace_profile *rec;
 634         unsigned long flags;
 635
 636         if (!ftrace_profile_enabled)
 637                 return;
 638
 639         local_irq_save(flags);
 640
 641         stat = &__get_cpu_var(ftrace_profile_stats);
 642         if (!stat->hash || !ftrace_profile_enabled)
 643                 goto out;
 644
 645         rec = ftrace_find_profiled_func(stat, ip);
 646         if (!rec) {
 647                 rec = ftrace_profile_alloc(stat, ip);
 648                 if (!rec)
 649                         goto out;
 650         }
 651
 652         rec->counter++;
 653  out:
 654         local_irq_restore(flags);
 655 }
 656
 657 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 658 static int profile_graph_entry(struct ftrace_graph_ent *trace)
 659 {
 660         function_profile_call(trace->func, 0);
 661         return 1;
 662 }
 663
 664 static void profile_graph_return(struct ftrace_graph_ret *trace)
 665 {
 666         struct ftrace_profile_stat *stat;
 667         unsigned long long calltime;
 668         struct ftrace_profile *rec;
 669         unsigned long flags;
 670
 671         local_irq_save(flags);
 672         stat = &__get_cpu_var(ftrace_profile_stats);
 673         if (!stat->hash || !ftrace_profile_enabled)
 674                 goto out;
 675
 676         /* If the calltime was zero'd ignore it */
 677         if (!trace->calltime)
 678                 goto out;
 679
 680         calltime = trace->rettime - trace->calltime;
 681
 682         if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
 683                 int index;
 684
 685                 index = trace->depth;
 686
 687                 /* Append this call time to the parent time to subtract */
 688                 if (index)
 689                         current->ret_stack[index - 1].subtime += calltime;
 690
 691                 if (current->ret_stack[index].subtime < calltime)
 692                         calltime -= current->ret_stack[index].subtime;
 693                 else
 694                         calltime = 0;
 695         }
 696
 697         rec = ftrace_find_profiled_func(stat, trace->func);
 698         if (rec) {
 699                 rec->time += calltime;
 700                 rec->time_squared += calltime * calltime;
 701         }
 702
 703  out:
 704         local_irq_restore(flags);
 705 }
 706
 707 static int register_ftrace_profiler(void)
 708 {
 709         return register_ftrace_graph(&profile_graph_return,
 710                                      &profile_graph_entry);
 711 }
 712
 713 static void unregister_ftrace_profiler(void)
 714 {
 715         unregister_ftrace_graph();
 716 }
 717 #else
 718 static struct ftrace_ops ftrace_profile_ops __read_mostly =
 719 {
 720         .func           = function_profile_call,
 721 };
 722
 723 static int register_ftrace_profiler(void)
 724 {
 725         return register_ftrace_function(&ftrace_profile_ops);
 726 }
 727
 728 static void unregister_ftrace_profiler(void)
 729 {
 730         unregister_ftrace_function(&ftrace_profile_ops);
 731 }
 732 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 733
 734 static ssize_t
 735 ftrace_profile_write(struct file *filp, const char __user *ubuf,
 736                      size_t cnt, loff_t *ppos)
 737 {
 738         unsigned long val;
 739         char buf[64];           /* big enough to hold a number */
 740         int ret;
 741
 742         if (cnt >= sizeof(buf))
 743                 return -EINVAL;
 744
 745         if (copy_from_user(&buf, ubuf, cnt))
 746                 return -EFAULT;
 747
 748         buf[cnt] = 0;
 749
 750         ret = strict_strtoul(buf, 10, &val);
 751         if (ret < 0)
 752                 return ret;
 753
 754         val = !!val;
 755
 756         mutex_lock(&ftrace_profile_lock);
 757         if (ftrace_profile_enabled ^ val) {
 758                 if (val) {
 759                         ret = ftrace_profile_init();
 760                         if (ret < 0) {
 761                                 cnt = ret;
 762                                 goto out;
 763                         }
 764
 765                         ret = register_ftrace_profiler();
 766                         if (ret < 0) {
 767                                 cnt = ret;
 768                                 goto out;
 769                         }
 770                         ftrace_profile_enabled = 1;
 771                 } else {
 772                         ftrace_profile_enabled = 0;
 773                         /*
 774                          * unregister_ftrace_profiler calls stop_machine
 775                          * so this acts like an synchronize_sched.
 776                          */
 777                         unregister_ftrace_profiler();
 778                 }
 779         }
 780  out:
 781         mutex_unlock(&ftrace_profile_lock);
 782
 783         *ppos += cnt;
 784
 785         return cnt;
 786 }
 787
 788 static ssize_t
 789 ftrace_profile_read(struct file *filp, char __user *ubuf,
 790                      size_t cnt, loff_t *ppos)
 791 {
 792         char buf[64];           /* big enough to hold a number */
 793         int r;
 794
 795         r = sprintf(buf, "%u\n", ftrace_profile_enabled);
 796         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 797 }
 798
 799 static const struct file_operations ftrace_profile_fops = {
 800         .open           = tracing_open_generic,
 801         .read           = ftrace_profile_read,
 802         .write          = ftrace_profile_write,
 803         .llseek         = default_llseek,
 804 };
 805
 806 /* used to initialize the real stat files */
 807 static struct tracer_stat function_stats __initdata = {
 808         .name           = "functions",
 809         .stat_start     = function_stat_start,
 810         .stat_next      = function_stat_next,
 811         .stat_cmp       = function_stat_cmp,
 812         .stat_headers   = function_stat_headers,
 813         .stat_show      = function_stat_show
 814 };
 815
 816 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 817 {
 818         struct ftrace_profile_stat *stat;
 819         struct dentry *entry;
 820         char *name;
 821         int ret;
 822         int cpu;
 823
 824         for_each_possible_cpu(cpu) {
 825                 stat = &per_cpu(ftrace_profile_stats, cpu);
 826
 827                 /* allocate enough for function name + cpu number */
 828                 name = kmalloc(32, GFP_KERNEL);
 829                 if (!name) {
 830                         /*
 831                          * The files created are permanent, if something happens
 832                          * we still do not free memory.
 833                          */
 834                         WARN(1,
 835                              "Could not allocate stat file for cpu %d\n",
 836                              cpu);
 837                         return;
 838                 }
 839                 stat->stat = function_stats;
 840                 snprintf(name, 32, "function%d", cpu);
 841                 stat->stat.name = name;
 842                 ret = register_stat_tracer(&stat->stat);
 843                 if (ret) {
 844                         WARN(1,
 845                              "Could not register function stat for cpu %d\n",
 846                              cpu);
 847                         kfree(name);
 848                         return;
 849                 }
 850         }
 851
 852         entry = debugfs_create_file("function_profile_enabled", 0644,
 853                                     d_tracer, NULL, &ftrace_profile_fops);
 854         if (!entry)
 855                 pr_warning("Could not create debugfs "
 856                            "'function_profile_enabled' entry\n");
 857 }
 858
 859 #else /* CONFIG_FUNCTION_PROFILER */
 860 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 861 {
 862 }
 863 #endif /* CONFIG_FUNCTION_PROFILER */
 864
 865 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
 866
 867 #ifdef CONFIG_DYNAMIC_FTRACE
 868
 869 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
 870 # error Dynamic ftrace depends on MCOUNT_RECORD
 871 #endif
 872
 873 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
 874
 875 struct ftrace_func_probe {
 876         struct hlist_node       node;
 877         struct ftrace_probe_ops *ops;
 878         unsigned long           flags;
 879         unsigned long           ip;
 880         void                    *data;
 881         struct rcu_head         rcu;
 882 };
 883
 884 enum {
 885         FTRACE_ENABLE_CALLS             = (1 << 0),
 886         FTRACE_DISABLE_CALLS            = (1 << 1),
 887         FTRACE_UPDATE_TRACE_FUNC        = (1 << 2),
 888         FTRACE_START_FUNC_RET           = (1 << 3),
 889         FTRACE_STOP_FUNC_RET            = (1 << 4),
 890 };
 891
 892 static int ftrace_filtered;
 893
 894 static struct dyn_ftrace *ftrace_new_addrs;
 895
 896 static DEFINE_MUTEX(ftrace_regex_lock);
 897
 898 struct ftrace_page {
 899         struct ftrace_page      *next;
 900         int                     index;
 901         struct dyn_ftrace       records[];
 902 };
 903
 904 #define ENTRIES_PER_PAGE \
 905   ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
 906
 907 /* estimate from running different kernels */
 908 #define NR_TO_INIT              10000
 909
 910 static struct ftrace_page       *ftrace_pages_start;
 911 static struct ftrace_page       *ftrace_pages;
 912
 913 static struct dyn_ftrace *ftrace_free_records;
 914
 915 /*
 916  * This is a double for. Do not use 'break' to break out of the loop,
 917  * you must use a goto.
 918  */
 919 #define do_for_each_ftrace_rec(pg, rec)                                 \
 920         for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
 921                 int _____i;                                             \
 922                 for (_____i = 0; _____i < pg->index; _____i++) {        \
 923                         rec = &pg->records[_____i];
 924
 925 #define while_for_each_ftrace_rec()             \
 926                 }                               \
 927         }
 928
 929 static void ftrace_free_rec(struct dyn_ftrace *rec)
 930 {
 931         rec->freelist = ftrace_free_records;
 932         ftrace_free_records = rec;
 933         rec->flags |= FTRACE_FL_FREE;
 934 }
 935
 936 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
 937 {
 938         struct dyn_ftrace *rec;
 939
 940         /* First check for freed records */
 941         if (ftrace_free_records) {
 942                 rec = ftrace_free_records;
 943
 944                 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
 945                         FTRACE_WARN_ON_ONCE(1);
 946                         ftrace_free_records = NULL;
 947                         return NULL;
 948                 }
 949
 950                 ftrace_free_records = rec->freelist;
 951                 memset(rec, 0, sizeof(*rec));
 952                 return rec;
 953         }
 954
 955         if (ftrace_pages->index == ENTRIES_PER_PAGE) {
 956                 if (!ftrace_pages->next) {
 957                         /* allocate another page */
 958                         ftrace_pages->next =
 959                                 (void *)get_zeroed_page(GFP_KERNEL);
 960                         if (!ftrace_pages->next)
 961                                 return NULL;
 962                 }
 963                 ftrace_pages = ftrace_pages->next;
 964         }
 965
 966         return &ftrace_pages->records[ftrace_pages->index++];
 967 }
 968
 969 static struct dyn_ftrace *
 970 ftrace_record_ip(unsigned long ip)
 971 {
 972         struct dyn_ftrace *rec;
 973
 974         if (ftrace_disabled)
 975                 return NULL;
 976
 977         rec = ftrace_alloc_dyn_node(ip);
 978         if (!rec)
 979                 return NULL;
 980
 981         rec->ip = ip;
 982         rec->newlist = ftrace_new_addrs;
 983         ftrace_new_addrs = rec;
 984
 985         return rec;
 986 }
 987
 988 static void print_ip_ins(const char *fmt, unsigned char *p)
 989 {
 990         int i;
 991
 992         printk(KERN_CONT "%s", fmt);
 993
 994         for (i = 0; i < MCOUNT_INSN_SIZE; i++)
 995                 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
 996 }
 997
 998 static void ftrace_bug(int failed, unsigned long ip)
 999 {
1000         switch (failed) {
1001         case -EFAULT:
1002                 FTRACE_WARN_ON_ONCE(1);
1003                 pr_info("ftrace faulted on modifying ");
1004                 print_ip_sym(ip);
1005                 break;
1006         case -EINVAL:
1007                 FTRACE_WARN_ON_ONCE(1);
1008                 pr_info("ftrace failed to modify ");
1009                 print_ip_sym(ip);
1010                 print_ip_ins(" actual: ", (unsigned char *)ip);
1011                 printk(KERN_CONT "\n");
1012                 break;
1013         case -EPERM:
1014                 FTRACE_WARN_ON_ONCE(1);
1015                 pr_info("ftrace faulted on writing ");
1016                 print_ip_sym(ip);
1017                 break;
1018         default:
1019                 FTRACE_WARN_ON_ONCE(1);
1020                 pr_info("ftrace faulted on unknown error ");
1021                 print_ip_sym(ip);
1022         }
1023 }
1024
1025
1026 /* Return 1 if the address range is reserved for ftrace */
1027 int ftrace_text_reserved(void *start, void *end)
1028 {
1029         struct dyn_ftrace *rec;
1030         struct ftrace_page *pg;
1031
1032         do_for_each_ftrace_rec(pg, rec) {
1033                 if (rec->ip <= (unsigned long)end &&
1034                     rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1035                         return 1;
1036         } while_for_each_ftrace_rec();
1037         return 0;
1038 }
1039
1040
1041 static int
1042 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1043 {
1044         unsigned long ftrace_addr;
1045         unsigned long flag = 0UL;
1046
1047         ftrace_addr = (unsigned long)FTRACE_ADDR;
1048
1049         /*
1050          * If this record is not to be traced or we want to disable it,
1051          * then disable it.
1052          *
1053          * If we want to enable it and filtering is off, then enable it.
1054          *
1055          * If we want to enable it and filtering is on, enable it only if
1056          * it's filtered
1057          */
1058         if (enable && !(rec->flags & FTRACE_FL_NOTRACE)) {
1059                 if (!ftrace_filtered || (rec->flags & FTRACE_FL_FILTER))
1060                         flag = FTRACE_FL_ENABLED;
1061         }
1062
1063         /* If the state of this record hasn't changed, then do nothing */
1064         if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1065                 return 0;
1066
1067         if (flag) {
1068                 rec->flags |= FTRACE_FL_ENABLED;
1069                 return ftrace_make_call(rec, ftrace_addr);
1070         }
1071
1072         rec->flags &= ~FTRACE_FL_ENABLED;
1073         return ftrace_make_nop(NULL, rec, ftrace_addr);
1074 }
1075
1076 static void ftrace_replace_code(int enable)
1077 {
1078         struct dyn_ftrace *rec;
1079         struct ftrace_page *pg;
1080         int failed;
1081
1082         do_for_each_ftrace_rec(pg, rec) {
1083                 /*
1084                  * Skip over free records, records that have
1085                  * failed and not converted.
1086                  */
1087                 if (rec->flags & FTRACE_FL_FREE ||
1088                     rec->flags & FTRACE_FL_FAILED ||
1089                     !(rec->flags & FTRACE_FL_CONVERTED))
1090                         continue;
1091
1092                 failed = __ftrace_replace_code(rec, enable);
1093                 if (failed) {
1094                         rec->flags |= FTRACE_FL_FAILED;
1095                         ftrace_bug(failed, rec->ip);
1096                         /* Stop processing */
1097                         return;
1098                 }
1099         } while_for_each_ftrace_rec();
1100 }
1101
1102 static int
1103 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1104 {
1105         unsigned long ip;
1106         int ret;
1107
1108         ip = rec->ip;
1109
1110         ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1111         if (ret) {
1112                 ftrace_bug(ret, ip);
1113                 rec->flags |= FTRACE_FL_FAILED;
1114                 return 0;
1115         }
1116         return 1;
1117 }
1118
1119 /*
1120  * archs can override this function if they must do something
1121  * before the modifying code is performed.
1122  */
1123 int __weak ftrace_arch_code_modify_prepare(void)
1124 {
1125         return 0;
1126 }
1127
1128 /*
1129  * archs can override this function if they must do something
1130  * after the modifying code is performed.
1131  */
1132 int __weak ftrace_arch_code_modify_post_process(void)
1133 {
1134         return 0;
1135 }
1136
1137 static int __ftrace_modify_code(void *data)
1138 {
1139         int *command = data;
1140
1141         if (*command & FTRACE_ENABLE_CALLS)
1142                 ftrace_replace_code(1);
1143         else if (*command & FTRACE_DISABLE_CALLS)
1144                 ftrace_replace_code(0);
1145
1146         if (*command & FTRACE_UPDATE_TRACE_FUNC)
1147                 ftrace_update_ftrace_func(ftrace_trace_function);
1148
1149         if (*command & FTRACE_START_FUNC_RET)
1150                 ftrace_enable_ftrace_graph_caller();
1151         else if (*command & FTRACE_STOP_FUNC_RET)
1152                 ftrace_disable_ftrace_graph_caller();
1153
1154         return 0;
1155 }
1156
1157 static void ftrace_run_update_code(int command)
1158 {
1159         int ret;
1160
1161         ret = ftrace_arch_code_modify_prepare();
1162         FTRACE_WARN_ON(ret);
1163         if (ret)
1164                 return;
1165
1166         stop_machine(__ftrace_modify_code, &command, NULL);
1167
1168         ret = ftrace_arch_code_modify_post_process();
1169         FTRACE_WARN_ON(ret);
1170 }
1171
1172 static ftrace_func_t saved_ftrace_func;
1173 static int ftrace_start_up;
1174
1175 static void ftrace_startup_enable(int command)
1176 {
1177         if (saved_ftrace_func != ftrace_trace_function) {
1178                 saved_ftrace_func = ftrace_trace_function;
1179                 command |= FTRACE_UPDATE_TRACE_FUNC;
1180         }
1181
1182         if (!command || !ftrace_enabled)
1183                 return;
1184
1185         ftrace_run_update_code(command);
1186 }
1187
1188 static void ftrace_startup(int command)
1189 {
1190         if (unlikely(ftrace_disabled))
1191                 return;
1192
1193         ftrace_start_up++;
1194         command |= FTRACE_ENABLE_CALLS;
1195
1196         ftrace_startup_enable(command);
1197 }
1198
1199 static void ftrace_shutdown(int command)
1200 {
1201         if (unlikely(ftrace_disabled))
1202                 return;
1203
1204         ftrace_start_up--;
1205         /*
1206          * Just warn in case of unbalance, no need to kill ftrace, it's not
1207          * critical but the ftrace_call callers may be never nopped again after
1208          * further ftrace uses.
1209          */
1210         WARN_ON_ONCE(ftrace_start_up < 0);
1211
1212         if (!ftrace_start_up)
1213                 command |= FTRACE_DISABLE_CALLS;
1214
1215         if (saved_ftrace_func != ftrace_trace_function) {
1216                 saved_ftrace_func = ftrace_trace_function;
1217                 command |= FTRACE_UPDATE_TRACE_FUNC;
1218         }
1219
1220         if (!command || !ftrace_enabled)
1221                 return;
1222
1223         ftrace_run_update_code(command);
1224 }
1225
1226 static void ftrace_startup_sysctl(void)
1227 {
1228         if (unlikely(ftrace_disabled))
1229                 return;
1230
1231         /* Force update next time */
1232         saved_ftrace_func = NULL;
1233         /* ftrace_start_up is true if we want ftrace running */
1234         if (ftrace_start_up)
1235                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1236 }
1237
1238 static void ftrace_shutdown_sysctl(void)
1239 {
1240         if (unlikely(ftrace_disabled))
1241                 return;
1242
1243         /* ftrace_start_up is true if ftrace is running */
1244         if (ftrace_start_up)
1245                 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
1246 }
1247
1248 static cycle_t          ftrace_update_time;
1249 static unsigned long    ftrace_update_cnt;
1250 unsigned long           ftrace_update_tot_cnt;
1251
1252 static int ftrace_update_code(struct module *mod)
1253 {
1254         struct dyn_ftrace *p;
1255         cycle_t start, stop;
1256
1257         start = ftrace_now(raw_smp_processor_id());
1258         ftrace_update_cnt = 0;
1259
1260         while (ftrace_new_addrs) {
1261
1262                 /* If something went wrong, bail without enabling anything */
1263                 if (unlikely(ftrace_disabled))
1264                         return -1;
1265
1266                 p = ftrace_new_addrs;
1267                 ftrace_new_addrs = p->newlist;
1268                 p->flags = 0L;
1269
1270                 /*
1271                  * Do the initial record convertion from mcount jump
1272                  * to the NOP instructions.
1273                  */
1274                 if (!ftrace_code_disable(mod, p)) {
1275                         ftrace_free_rec(p);
1276                         continue;
1277                 }
1278
1279                 p->flags |= FTRACE_FL_CONVERTED;
1280                 ftrace_update_cnt++;
1281
1282                 /*
1283                  * If the tracing is enabled, go ahead and enable the record.
1284                  *
1285                  * The reason not to enable the record immediatelly is the
1286                  * inherent check of ftrace_make_nop/ftrace_make_call for
1287                  * correct previous instructions.  Making first the NOP
1288                  * conversion puts the module to the correct state, thus
1289                  * passing the ftrace_make_call check.
1290                  */
1291                 if (ftrace_start_up) {
1292                         int failed = __ftrace_replace_code(p, 1);
1293                         if (failed) {
1294                                 ftrace_bug(failed, p->ip);
1295                                 ftrace_free_rec(p);
1296                         }
1297                 }
1298         }
1299
1300         stop = ftrace_now(raw_smp_processor_id());
1301         ftrace_update_time = stop - start;
1302         ftrace_update_tot_cnt += ftrace_update_cnt;
1303
1304         return 0;
1305 }
1306
1307 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1308 {
1309         struct ftrace_page *pg;
1310         int cnt;
1311         int i;
1312
1313         /* allocate a few pages */
1314         ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1315         if (!ftrace_pages_start)
1316                 return -1;
1317
1318         /*
1319          * Allocate a few more pages.
1320          *
1321          * TODO: have some parser search vmlinux before
1322          *   final linking to find all calls to ftrace.
1323          *   Then we can:
1324          *    a) know how many pages to allocate.
1325          *     and/or
1326          *    b) set up the table then.
1327          *
1328          *  The dynamic code is still necessary for
1329          *  modules.
1330          */
1331
1332         pg = ftrace_pages = ftrace_pages_start;
1333
1334         cnt = num_to_init / ENTRIES_PER_PAGE;
1335         pr_info("ftrace: allocating %ld entries in %d pages\n",
1336                 num_to_init, cnt + 1);
1337
1338         for (i = 0; i < cnt; i++) {
1339                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1340
1341                 /* If we fail, we'll try later anyway */
1342                 if (!pg->next)
1343                         break;
1344
1345                 pg = pg->next;
1346         }
1347
1348         return 0;
1349 }
1350
1351 enum {
1352         FTRACE_ITER_FILTER      = (1 << 0),
1353         FTRACE_ITER_NOTRACE     = (1 << 1),
1354         FTRACE_ITER_FAILURES    = (1 << 2),
1355         FTRACE_ITER_PRINTALL    = (1 << 3),
1356         FTRACE_ITER_HASH        = (1 << 4),
1357 };
1358
1359 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1360
1361 struct ftrace_iterator {
1362         loff_t                          pos;
1363         loff_t                          func_pos;
1364         struct ftrace_page              *pg;
1365         struct dyn_ftrace               *func;
1366         struct ftrace_func_probe        *probe;
1367         struct trace_parser             parser;
1368         int                             hidx;
1369         int                             idx;
1370         unsigned                        flags;
1371 };
1372
1373 static void *
1374 t_hash_next(struct seq_file *m, loff_t *pos)
1375 {
1376         struct ftrace_iterator *iter = m->private;
1377         struct hlist_node *hnd = NULL;
1378         struct hlist_head *hhd;
1379
1380         (*pos)++;
1381         iter->pos = *pos;
1382
1383         if (iter->probe)
1384                 hnd = &iter->probe->node;
1385  retry:
1386         if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1387                 return NULL;
1388
1389         hhd = &ftrace_func_hash[iter->hidx];
1390
1391         if (hlist_empty(hhd)) {
1392                 iter->hidx++;
1393                 hnd = NULL;
1394                 goto retry;
1395         }
1396
1397         if (!hnd)
1398                 hnd = hhd->first;
1399         else {
1400                 hnd = hnd->next;
1401                 if (!hnd) {
1402                         iter->hidx++;
1403                         goto retry;
1404                 }
1405         }
1406
1407         if (WARN_ON_ONCE(!hnd))
1408                 return NULL;
1409
1410         iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
1411
1412         return iter;
1413 }
1414
1415 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1416 {
1417         struct ftrace_iterator *iter = m->private;
1418         void *p = NULL;
1419         loff_t l;
1420
1421         if (iter->func_pos > *pos)
1422                 return NULL;
1423
1424         iter->hidx = 0;
1425         for (l = 0; l <= (*pos - iter->func_pos); ) {
1426                 p = t_hash_next(m, &l);
1427                 if (!p)
1428                         break;
1429         }
1430         if (!p)
1431                 return NULL;
1432
1433         /* Only set this if we have an item */
1434         iter->flags |= FTRACE_ITER_HASH;
1435
1436         return iter;
1437 }
1438
1439 static int
1440 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
1441 {
1442         struct ftrace_func_probe *rec;
1443
1444         rec = iter->probe;
1445         if (WARN_ON_ONCE(!rec))
1446                 return -EIO;
1447
1448         if (rec->ops->print)
1449                 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1450
1451         seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
1452
1453         if (rec->data)
1454                 seq_printf(m, ":%p", rec->data);
1455         seq_putc(m, '\n');
1456
1457         return 0;
1458 }
1459
1460 static void *
1461 t_next(struct seq_file *m, void *v, loff_t *pos)
1462 {
1463         struct ftrace_iterator *iter = m->private;
1464         struct dyn_ftrace *rec = NULL;
1465
1466         if (iter->flags & FTRACE_ITER_HASH)
1467                 return t_hash_next(m, pos);
1468
1469         (*pos)++;
1470         iter->pos = *pos;
1471
1472         if (iter->flags & FTRACE_ITER_PRINTALL)
1473                 return t_hash_start(m, pos);
1474
1475  retry:
1476         if (iter->idx >= iter->pg->index) {
1477                 if (iter->pg->next) {
1478                         iter->pg = iter->pg->next;
1479                         iter->idx = 0;
1480                         goto retry;
1481                 }
1482         } else {
1483                 rec = &iter->pg->records[iter->idx++];
1484                 if ((rec->flags & FTRACE_FL_FREE) ||
1485
1486                     (!(iter->flags & FTRACE_ITER_FAILURES) &&
1487                      (rec->flags & FTRACE_FL_FAILED)) ||
1488
1489                     ((iter->flags & FTRACE_ITER_FAILURES) &&
1490                      !(rec->flags & FTRACE_FL_FAILED)) ||
1491
1492                     ((iter->flags & FTRACE_ITER_FILTER) &&
1493                      !(rec->flags & FTRACE_FL_FILTER)) ||
1494
1495                     ((iter->flags & FTRACE_ITER_NOTRACE) &&
1496                      !(rec->flags & FTRACE_FL_NOTRACE))) {
1497                         rec = NULL;
1498                         goto retry;
1499                 }
1500         }
1501
1502         if (!rec)
1503                 return t_hash_start(m, pos);
1504
1505         iter->func_pos = *pos;
1506         iter->func = rec;
1507
1508         return iter;
1509 }
1510
1511 static void reset_iter_read(struct ftrace_iterator *iter)
1512 {
1513         iter->pos = 0;
1514         iter->func_pos = 0;
1515         iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
1516 }
1517
1518 static void *t_start(struct seq_file *m, loff_t *pos)
1519 {
1520         struct ftrace_iterator *iter = m->private;
1521         void *p = NULL;
1522         loff_t l;
1523
1524         mutex_lock(&ftrace_lock);
1525         /*
1526          * If an lseek was done, then reset and start from beginning.
1527          */
1528         if (*pos < iter->pos)
1529                 reset_iter_read(iter);
1530
1531         /*
1532          * For set_ftrace_filter reading, if we have the filter
1533          * off, we can short cut and just print out that all
1534          * functions are enabled.
1535          */
1536         if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1537                 if (*pos > 0)
1538                         return t_hash_start(m, pos);
1539                 iter->flags |= FTRACE_ITER_PRINTALL;
1540                 /* reset in case of seek/pread */
1541                 iter->flags &= ~FTRACE_ITER_HASH;
1542                 return iter;
1543         }
1544
1545         if (iter->flags & FTRACE_ITER_HASH)
1546                 return t_hash_start(m, pos);
1547
1548         /*
1549          * Unfortunately, we need to restart at ftrace_pages_start
1550          * every time we let go of the ftrace_mutex. This is because
1551          * those pointers can change without the lock.
1552          */
1553         iter->pg = ftrace_pages_start;
1554         iter->idx = 0;
1555         for (l = 0; l <= *pos; ) {
1556                 p = t_next(m, p, &l);
1557                 if (!p)
1558                         break;
1559         }
1560
1561         if (!p) {
1562                 if (iter->flags & FTRACE_ITER_FILTER)
1563                         return t_hash_start(m, pos);
1564
1565                 return NULL;
1566         }
1567
1568         return iter;
1569 }
1570
1571 static void t_stop(struct seq_file *m, void *p)
1572 {
1573         mutex_unlock(&ftrace_lock);
1574 }
1575
1576 static int t_show(struct seq_file *m, void *v)
1577 {
1578         struct ftrace_iterator *iter = m->private;
1579         struct dyn_ftrace *rec;
1580
1581         if (iter->flags & FTRACE_ITER_HASH)
1582                 return t_hash_show(m, iter);
1583
1584         if (iter->flags & FTRACE_ITER_PRINTALL) {
1585                 seq_printf(m, "#### all functions enabled ####\n");
1586                 return 0;
1587         }
1588
1589         rec = iter->func;
1590
1591         if (!rec)
1592                 return 0;
1593
1594         seq_printf(m, "%ps\n", (void *)rec->ip);
1595
1596         return 0;
1597 }
1598
1599 static const struct seq_operations show_ftrace_seq_ops = {
1600         .start = t_start,
1601         .next = t_next,
1602         .stop = t_stop,
1603         .show = t_show,
1604 };
1605
1606 static int
1607 ftrace_avail_open(struct inode *inode, struct file *file)
1608 {
1609         struct ftrace_iterator *iter;
1610         int ret;
1611
1612         if (unlikely(ftrace_disabled))
1613                 return -ENODEV;
1614
1615         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1616         if (!iter)
1617                 return -ENOMEM;
1618
1619         iter->pg = ftrace_pages_start;
1620
1621         ret = seq_open(file, &show_ftrace_seq_ops);
1622         if (!ret) {
1623                 struct seq_file *m = file->private_data;
1624
1625                 m->private = iter;
1626         } else {
1627                 kfree(iter);
1628         }
1629
1630         return ret;
1631 }
1632
1633 static int
1634 ftrace_failures_open(struct inode *inode, struct file *file)
1635 {
1636         int ret;
1637         struct seq_file *m;
1638         struct ftrace_iterator *iter;
1639
1640         ret = ftrace_avail_open(inode, file);
1641         if (!ret) {
1642                 m = file->private_data;
1643                 iter = m->private;
1644                 iter->flags = FTRACE_ITER_FAILURES;
1645         }
1646
1647         return ret;
1648 }
1649
1650
1651 static void ftrace_filter_reset(int enable)
1652 {
1653         struct ftrace_page *pg;
1654         struct dyn_ftrace *rec;
1655         unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1656
1657         mutex_lock(&ftrace_lock);
1658         if (enable)
1659                 ftrace_filtered = 0;
1660         do_for_each_ftrace_rec(pg, rec) {
1661                 if (rec->flags & FTRACE_FL_FAILED)
1662                         continue;
1663                 rec->flags &= ~type;
1664         } while_for_each_ftrace_rec();
1665         mutex_unlock(&ftrace_lock);
1666 }
1667
1668 static int
1669 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1670 {
1671         struct ftrace_iterator *iter;
1672         int ret = 0;
1673
1674         if (unlikely(ftrace_disabled))
1675                 return -ENODEV;
1676
1677         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1678         if (!iter)
1679                 return -ENOMEM;
1680
1681         if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
1682                 kfree(iter);
1683                 return -ENOMEM;
1684         }
1685
1686         mutex_lock(&ftrace_regex_lock);
1687         if ((file->f_mode & FMODE_WRITE) &&
1688             (file->f_flags & O_TRUNC))
1689                 ftrace_filter_reset(enable);
1690
1691         if (file->f_mode & FMODE_READ) {
1692                 iter->pg = ftrace_pages_start;
1693                 iter->flags = enable ? FTRACE_ITER_FILTER :
1694                         FTRACE_ITER_NOTRACE;
1695
1696                 ret = seq_open(file, &show_ftrace_seq_ops);
1697                 if (!ret) {
1698                         struct seq_file *m = file->private_data;
1699                         m->private = iter;
1700                 } else {
1701                         trace_parser_put(&iter->parser);
1702                         kfree(iter);
1703                 }
1704         } else
1705                 file->private_data = iter;
1706         mutex_unlock(&ftrace_regex_lock);
1707
1708         return ret;
1709 }
1710
1711 static int
1712 ftrace_filter_open(struct inode *inode, struct file *file)
1713 {
1714         return ftrace_regex_open(inode, file, 1);
1715 }
1716
1717 static int
1718 ftrace_notrace_open(struct inode *inode, struct file *file)
1719 {
1720         return ftrace_regex_open(inode, file, 0);
1721 }
1722
1723 static loff_t
1724 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1725 {
1726         loff_t ret;
1727
1728         if (file->f_mode & FMODE_READ)
1729                 ret = seq_lseek(file, offset, origin);
1730         else
1731                 file->f_pos = ret = 1;
1732
1733         return ret;
1734 }
1735
1736 static int ftrace_match(char *str, char *regex, int len, int type)
1737 {
1738         int matched = 0;
1739         int slen;
1740
1741         switch (type) {
1742         case MATCH_FULL:
1743                 if (strcmp(str, regex) == 0)
1744                         matched = 1;
1745                 break;
1746         case MATCH_FRONT_ONLY:
1747                 if (strncmp(str, regex, len) == 0)
1748                         matched = 1;
1749                 break;
1750         case MATCH_MIDDLE_ONLY:
1751                 if (strstr(str, regex))
1752                         matched = 1;
1753                 break;
1754         case MATCH_END_ONLY:
1755                 slen = strlen(str);
1756                 if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
1757                         matched = 1;
1758                 break;
1759         }
1760
1761         return matched;
1762 }
1763
1764 static int
1765 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1766 {
1767         char str[KSYM_SYMBOL_LEN];
1768
1769         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1770         return ftrace_match(str, regex, len, type);
1771 }
1772
1773 static int ftrace_match_records(char *buff, int len, int enable)
1774 {
1775         unsigned int search_len;
1776         struct ftrace_page *pg;
1777         struct dyn_ftrace *rec;
1778         unsigned long flag;
1779         char *search;
1780         int type;
1781         int not;
1782         int found = 0;
1783
1784         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1785         type = filter_parse_regex(buff, len, &search, &not);
1786
1787         search_len = strlen(search);
1788
1789         mutex_lock(&ftrace_lock);
1790         do_for_each_ftrace_rec(pg, rec) {
1791
1792                 if (rec->flags & FTRACE_FL_FAILED)
1793                         continue;
1794
1795                 if (ftrace_match_record(rec, search, search_len, type)) {
1796                         if (not)
1797                                 rec->flags &= ~flag;
1798                         else
1799                                 rec->flags |= flag;
1800                         found = 1;
1801                 }
1802                 /*
1803                  * Only enable filtering if we have a function that
1804                  * is filtered on.
1805                  */
1806                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1807                         ftrace_filtered = 1;
1808         } while_for_each_ftrace_rec();
1809         mutex_unlock(&ftrace_lock);
1810
1811         return found;
1812 }
1813
1814 static int
1815 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1816                            char *regex, int len, int type)
1817 {
1818         char str[KSYM_SYMBOL_LEN];
1819         char *modname;
1820
1821         kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1822
1823         if (!modname || strcmp(modname, mod))
1824                 return 0;
1825
1826         /* blank search means to match all funcs in the mod */
1827         if (len)
1828                 return ftrace_match(str, regex, len, type);
1829         else
1830                 return 1;
1831 }
1832
1833 static int ftrace_match_module_records(char *buff, char *mod, int enable)
1834 {
1835         unsigned search_len = 0;
1836         struct ftrace_page *pg;
1837         struct dyn_ftrace *rec;
1838         int type = MATCH_FULL;
1839         char *search = buff;
1840         unsigned long flag;
1841         int not = 0;
1842         int found = 0;
1843
1844         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1845
1846         /* blank or '*' mean the same */
1847         if (strcmp(buff, "*") == 0)
1848                 buff[0] = 0;
1849
1850         /* handle the case of 'dont filter this module' */
1851         if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1852                 buff[0] = 0;
1853                 not = 1;
1854         }
1855
1856         if (strlen(buff)) {
1857                 type = filter_parse_regex(buff, strlen(buff), &search, &not);
1858                 search_len = strlen(search);
1859         }
1860
1861         mutex_lock(&ftrace_lock);
1862         do_for_each_ftrace_rec(pg, rec) {
1863
1864                 if (rec->flags & FTRACE_FL_FAILED)
1865                         continue;
1866
1867                 if (ftrace_match_module_record(rec, mod,
1868                                                search, search_len, type)) {
1869                         if (not)
1870                                 rec->flags &= ~flag;
1871                         else
1872                                 rec->flags |= flag;
1873                         found = 1;
1874                 }
1875                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1876                         ftrace_filtered = 1;
1877
1878         } while_for_each_ftrace_rec();
1879         mutex_unlock(&ftrace_lock);
1880
1881         return found;
1882 }
1883
1884 /*
1885  * We register the module command as a template to show others how
1886  * to register the a command as well.
1887  */
1888
1889 static int
1890 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1891 {
1892         char *mod;
1893
1894         /*
1895          * cmd == 'mod' because we only registered this func
1896          * for the 'mod' ftrace_func_command.
1897          * But if you register one func with multiple commands,
1898          * you can tell which command was used by the cmd
1899          * parameter.
1900          */
1901
1902         /* we must have a module name */
1903         if (!param)
1904                 return -EINVAL;
1905
1906         mod = strsep(&param, ":");
1907         if (!strlen(mod))
1908                 return -EINVAL;
1909
1910         if (ftrace_match_module_records(func, mod, enable))
1911                 return 0;
1912         return -EINVAL;
1913 }
1914
1915 static struct ftrace_func_command ftrace_mod_cmd = {
1916         .name                   = "mod",
1917         .func                   = ftrace_mod_callback,
1918 };
1919
1920 static int __init ftrace_mod_cmd_init(void)
1921 {
1922         return register_ftrace_command(&ftrace_mod_cmd);
1923 }
1924 device_initcall(ftrace_mod_cmd_init);
1925
1926 static void
1927 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1928 {
1929         struct ftrace_func_probe *entry;
1930         struct hlist_head *hhd;
1931         struct hlist_node *n;
1932         unsigned long key;
1933
1934         key = hash_long(ip, FTRACE_HASH_BITS);
1935
1936         hhd = &ftrace_func_hash[key];
1937
1938         if (hlist_empty(hhd))
1939                 return;
1940
1941         /*
1942          * Disable preemption for these calls to prevent a RCU grace
1943          * period. This syncs the hash iteration and freeing of items
1944          * on the hash. rcu_read_lock is too dangerous here.
1945          */
1946         preempt_disable_notrace();
1947         hlist_for_each_entry_rcu(entry, n, hhd, node) {
1948                 if (entry->ip == ip)
1949                         entry->ops->func(ip, parent_ip, &entry->data);
1950         }
1951         preempt_enable_notrace();
1952 }
1953
1954 static struct ftrace_ops trace_probe_ops __read_mostly =
1955 {
1956         .func           = function_trace_probe_call,
1957 };
1958
1959 static int ftrace_probe_registered;
1960
1961 static void __enable_ftrace_function_probe(void)
1962 {
1963         int i;
1964
1965         if (ftrace_probe_registered)
1966                 return;
1967
1968         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1969                 struct hlist_head *hhd = &ftrace_func_hash[i];
1970                 if (hhd->first)
1971                         break;
1972         }
1973         /* Nothing registered? */
1974         if (i == FTRACE_FUNC_HASHSIZE)
1975                 return;
1976
1977         __register_ftrace_function(&trace_probe_ops);
1978         ftrace_startup(0);
1979         ftrace_probe_registered = 1;
1980 }
1981
1982 static void __disable_ftrace_function_probe(void)
1983 {
1984         int i;
1985
1986         if (!ftrace_probe_registered)
1987                 return;
1988
1989         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1990                 struct hlist_head *hhd = &ftrace_func_hash[i];
1991                 if (hhd->first)
1992                         return;
1993         }
1994
1995         /* no more funcs left */
1996         __unregister_ftrace_function(&trace_probe_ops);
1997         ftrace_shutdown(0);
1998         ftrace_probe_registered = 0;
1999 }
2000
2001
2002 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2003 {
2004         struct ftrace_func_probe *entry =
2005                 container_of(rhp, struct ftrace_func_probe, rcu);
2006
2007         if (entry->ops->free)
2008                 entry->ops->free(&entry->data);
2009         kfree(entry);
2010 }
2011
2012
2013 int
2014 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2015                               void *data)
2016 {
2017         struct ftrace_func_probe *entry;
2018         struct ftrace_page *pg;
2019         struct dyn_ftrace *rec;
2020         int type, len, not;
2021         unsigned long key;
2022         int count = 0;
2023         char *search;
2024
2025         type = filter_parse_regex(glob, strlen(glob), &search, &not);
2026         len = strlen(search);
2027
2028         /* we do not support '!' for function probes */
2029         if (WARN_ON(not))
2030                 return -EINVAL;
2031
2032         mutex_lock(&ftrace_lock);
2033         do_for_each_ftrace_rec(pg, rec) {
2034
2035                 if (rec->flags & FTRACE_FL_FAILED)
2036                         continue;
2037
2038                 if (!ftrace_match_record(rec, search, len, type))
2039                         continue;
2040
2041                 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2042                 if (!entry) {
2043                         /* If we did not process any, then return error */
2044                         if (!count)
2045                                 count = -ENOMEM;
2046                         goto out_unlock;
2047                 }
2048
2049                 count++;
2050
2051                 entry->data = data;
2052
2053                 /*
2054                  * The caller might want to do something special
2055                  * for each function we find. We call the callback
2056                  * to give the caller an opportunity to do so.
2057                  */
2058                 if (ops->callback) {
2059                         if (ops->callback(rec->ip, &entry->data) < 0) {
2060                                 /* caller does not like this func */
2061                                 kfree(entry);
2062                                 continue;
2063                         }
2064                 }
2065
2066                 entry->ops = ops;
2067                 entry->ip = rec->ip;
2068
2069                 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2070                 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2071
2072         } while_for_each_ftrace_rec();
2073         __enable_ftrace_function_probe();
2074
2075  out_unlock:
2076         mutex_unlock(&ftrace_lock);
2077
2078         return count;
2079 }
2080
2081 enum {
2082         PROBE_TEST_FUNC         = 1,
2083         PROBE_TEST_DATA         = 2
2084 };
2085
2086 static void
2087 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2088                                   void *data, int flags)
2089 {
2090         struct ftrace_func_probe *entry;
2091         struct hlist_node *n, *tmp;
2092         char str[KSYM_SYMBOL_LEN];
2093         int type = MATCH_FULL;
2094         int i, len = 0;
2095         char *search;
2096
2097         if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2098                 glob = NULL;
2099         else if (glob) {
2100                 int not;
2101
2102                 type = filter_parse_regex(glob, strlen(glob), &search, &not);
2103                 len = strlen(search);
2104
2105                 /* we do not support '!' for function probes */
2106                 if (WARN_ON(not))
2107                         return;
2108         }
2109
2110         mutex_lock(&ftrace_lock);
2111         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2112                 struct hlist_head *hhd = &ftrace_func_hash[i];
2113
2114                 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2115
2116                         /* break up if statements for readability */
2117                         if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2118                                 continue;
2119
2120                         if ((flags & PROBE_TEST_DATA) && entry->data != data)
2121                                 continue;
2122
2123                         /* do this last, since it is the most expensive */
2124                         if (glob) {
2125                                 kallsyms_lookup(entry->ip, NULL, NULL,
2126                                                 NULL, str);
2127                                 if (!ftrace_match(str, glob, len, type))
2128                                         continue;
2129                         }
2130
2131                         hlist_del(&entry->node);
2132                         call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2133                 }
2134         }
2135         __disable_ftrace_function_probe();
2136         mutex_unlock(&ftrace_lock);
2137 }
2138
2139 void
2140 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2141                                 void *data)
2142 {
2143         __unregister_ftrace_function_probe(glob, ops, data,
2144                                           PROBE_TEST_FUNC | PROBE_TEST_DATA);
2145 }
2146
2147 void
2148 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2149 {
2150         __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2151 }
2152
2153 void unregister_ftrace_function_probe_all(char *glob)
2154 {
2155         __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2156 }
2157
2158 static LIST_HEAD(ftrace_commands);
2159 static DEFINE_MUTEX(ftrace_cmd_mutex);
2160
2161 int register_ftrace_command(struct ftrace_func_command *cmd)
2162 {
2163         struct ftrace_func_command *p;
2164         int ret = 0;
2165
2166         mutex_lock(&ftrace_cmd_mutex);
2167         list_for_each_entry(p, &ftrace_commands, list) {
2168                 if (strcmp(cmd->name, p->name) == 0) {
2169                         ret = -EBUSY;
2170                         goto out_unlock;
2171                 }
2172         }
2173         list_add(&cmd->list, &ftrace_commands);
2174  out_unlock:
2175         mutex_unlock(&ftrace_cmd_mutex);
2176
2177         return ret;
2178 }
2179
2180 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2181 {
2182         struct ftrace_func_command *p, *n;
2183         int ret = -ENODEV;
2184
2185         mutex_lock(&ftrace_cmd_mutex);
2186         list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2187                 if (strcmp(cmd->name, p->name) == 0) {
2188                         ret = 0;
2189                         list_del_init(&p->list);
2190                         goto out_unlock;
2191                 }
2192         }
2193  out_unlock:
2194         mutex_unlock(&ftrace_cmd_mutex);
2195
2196         return ret;
2197 }
2198
2199 static int ftrace_process_regex(char *buff, int len, int enable)
2200 {
2201         char *func, *command, *next = buff;
2202         struct ftrace_func_command *p;
2203         int ret = -EINVAL;
2204
2205         func = strsep(&next, ":");
2206
2207         if (!next) {
2208                 if (ftrace_match_records(func, len, enable))
2209                         return 0;
2210                 return ret;
2211         }
2212
2213         /* command found */
2214
2215         command = strsep(&next, ":");
2216
2217         mutex_lock(&ftrace_cmd_mutex);
2218         list_for_each_entry(p, &ftrace_commands, list) {
2219                 if (strcmp(p->name, command) == 0) {
2220                         ret = p->func(func, command, next, enable);
2221                         goto out_unlock;
2222                 }
2223         }
2224  out_unlock:
2225         mutex_unlock(&ftrace_cmd_mutex);
2226
2227         return ret;
2228 }
2229
2230 static ssize_t
2231 ftrace_regex_write(struct file *file, const char __user *ubuf,
2232                    size_t cnt, loff_t *ppos, int enable)
2233 {
2234         struct ftrace_iterator *iter;
2235         struct trace_parser *parser;
2236         ssize_t ret, read;
2237
2238         if (!cnt)
2239                 return 0;
2240
2241         mutex_lock(&ftrace_regex_lock);
2242
2243         if (file->f_mode & FMODE_READ) {
2244                 struct seq_file *m = file->private_data;
2245                 iter = m->private;
2246         } else
2247                 iter = file->private_data;
2248
2249         parser = &iter->parser;
2250         read = trace_get_user(parser, ubuf, cnt, ppos);
2251
2252         if (read >= 0 && trace_parser_loaded(parser) &&
2253             !trace_parser_cont(parser)) {
2254                 ret = ftrace_process_regex(parser->buffer,
2255                                            parser->idx, enable);
2256                 trace_parser_clear(parser);
2257                 if (ret)
2258                         goto out_unlock;
2259         }
2260
2261         ret = read;
2262 out_unlock:
2263         mutex_unlock(&ftrace_regex_lock);
2264
2265         return ret;
2266 }
2267
2268 static ssize_t
2269 ftrace_filter_write(struct file *file, const char __user *ubuf,
2270                     size_t cnt, loff_t *ppos)
2271 {
2272         return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2273 }
2274
2275 static ssize_t
2276 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2277                      size_t cnt, loff_t *ppos)
2278 {
2279         return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2280 }
2281
2282 static void
2283 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2284 {
2285         if (unlikely(ftrace_disabled))
2286                 return;
2287
2288         mutex_lock(&ftrace_regex_lock);
2289         if (reset)
2290                 ftrace_filter_reset(enable);
2291         if (buf)
2292                 ftrace_match_records(buf, len, enable);
2293         mutex_unlock(&ftrace_regex_lock);
2294 }
2295
2296 /**
2297  * ftrace_set_filter - set a function to filter on in ftrace
2298  * @buf - the string that holds the function filter text.
2299  * @len - the length of the string.
2300  * @reset - non zero to reset all filters before applying this filter.
2301  *
2302  * Filters denote which functions should be enabled when tracing is enabled.
2303  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2304  */
2305 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2306 {
2307         ftrace_set_regex(buf, len, reset, 1);
2308 }
2309
2310 /**
2311  * ftrace_set_notrace - set a function to not trace in ftrace
2312  * @buf - the string that holds the function notrace text.
2313  * @len - the length of the string.
2314  * @reset - non zero to reset all filters before applying this filter.
2315  *
2316  * Notrace Filters denote which functions should not be enabled when tracing
2317  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2318  * for tracing.
2319  */
2320 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2321 {
2322         ftrace_set_regex(buf, len, reset, 0);
2323 }
2324
2325 /*
2326  * command line interface to allow users to set filters on boot up.
2327  */
2328 #define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
2329 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2330 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2331
2332 static int __init set_ftrace_notrace(char *str)
2333 {
2334         strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2335         return 1;
2336 }
2337 __setup("ftrace_notrace=", set_ftrace_notrace);
2338
2339 static int __init set_ftrace_filter(char *str)
2340 {
2341         strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2342         return 1;
2343 }
2344 __setup("ftrace_filter=", set_ftrace_filter);
2345
2346 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2347 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
2348 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
2349
2350 static int __init set_graph_function(char *str)
2351 {
2352         strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
2353         return 1;
2354 }
2355 __setup("ftrace_graph_filter=", set_graph_function);
2356
2357 static void __init set_ftrace_early_graph(char *buf)
2358 {
2359         int ret;
2360         char *func;
2361
2362         while (buf) {
2363                 func = strsep(&buf, ",");
2364                 /* we allow only one expression at a time */
2365                 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2366                                       func);
2367                 if (ret)
2368                         printk(KERN_DEBUG "ftrace: function %s not "
2369                                           "traceable\n", func);
2370         }
2371 }
2372 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2373
2374 static void __init set_ftrace_early_filter(char *buf, int enable)
2375 {
2376         char *func;
2377
2378         while (buf) {
2379                 func = strsep(&buf, ",");
2380                 ftrace_set_regex(func, strlen(func), 0, enable);
2381         }
2382 }
2383
2384 static void __init set_ftrace_early_filters(void)
2385 {
2386         if (ftrace_filter_buf[0])
2387                 set_ftrace_early_filter(ftrace_filter_buf, 1);
2388         if (ftrace_notrace_buf[0])
2389                 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2390 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2391         if (ftrace_graph_buf[0])
2392                 set_ftrace_early_graph(ftrace_graph_buf);
2393 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2394 }
2395
2396 static int
2397 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2398 {
2399         struct seq_file *m = (struct seq_file *)file->private_data;
2400         struct ftrace_iterator *iter;
2401         struct trace_parser *parser;
2402
2403         mutex_lock(&ftrace_regex_lock);
2404         if (file->f_mode & FMODE_READ) {
2405                 iter = m->private;
2406
2407                 seq_release(inode, file);
2408         } else
2409                 iter = file->private_data;
2410
2411         parser = &iter->parser;
2412         if (trace_parser_loaded(parser)) {
2413                 parser->buffer[parser->idx] = 0;
2414                 ftrace_match_records(parser->buffer, parser->idx, enable);
2415         }
2416
2417         mutex_lock(&ftrace_lock);
2418         if (ftrace_start_up && ftrace_enabled)
2419                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2420         mutex_unlock(&ftrace_lock);
2421
2422         trace_parser_put(parser);
2423         kfree(iter);
2424
2425         mutex_unlock(&ftrace_regex_lock);
2426         return 0;
2427 }
2428
2429 static int
2430 ftrace_filter_release(struct inode *inode, struct file *file)
2431 {
2432         return ftrace_regex_release(inode, file, 1);
2433 }
2434
2435 static int
2436 ftrace_notrace_release(struct inode *inode, struct file *file)
2437 {
2438         return ftrace_regex_release(inode, file, 0);
2439 }
2440
2441 static const struct file_operations ftrace_avail_fops = {
2442         .open = ftrace_avail_open,
2443         .read = seq_read,
2444         .llseek = seq_lseek,
2445         .release = seq_release_private,
2446 };
2447
2448 static const struct file_operations ftrace_failures_fops = {
2449         .open = ftrace_failures_open,
2450         .read = seq_read,
2451         .llseek = seq_lseek,
2452         .release = seq_release_private,
2453 };
2454
2455 static const struct file_operations ftrace_filter_fops = {
2456         .open = ftrace_filter_open,
2457         .read = seq_read,
2458         .write = ftrace_filter_write,
2459         .llseek = ftrace_regex_lseek,
2460         .release = ftrace_filter_release,
2461 };
2462
2463 static const struct file_operations ftrace_notrace_fops = {
2464         .open = ftrace_notrace_open,
2465         .read = seq_read,
2466         .write = ftrace_notrace_write,
2467         .llseek = ftrace_regex_lseek,
2468         .release = ftrace_notrace_release,
2469 };
2470
2471 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2472
2473 static DEFINE_MUTEX(graph_lock);
2474
2475 int ftrace_graph_count;
2476 int ftrace_graph_filter_enabled;
2477 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2478
2479 static void *
2480 __g_next(struct seq_file *m, loff_t *pos)
2481 {
2482         if (*pos >= ftrace_graph_count)
2483                 return NULL;
2484         return &ftrace_graph_funcs[*pos];
2485 }
2486
2487 static void *
2488 g_next(struct seq_file *m, void *v, loff_t *pos)
2489 {
2490         (*pos)++;
2491         return __g_next(m, pos);
2492 }
2493
2494 static void *g_start(struct seq_file *m, loff_t *pos)
2495 {
2496         mutex_lock(&graph_lock);
2497
2498         /* Nothing, tell g_show to print all functions are enabled */
2499         if (!ftrace_graph_filter_enabled && !*pos)
2500                 return (void *)1;
2501
2502         return __g_next(m, pos);
2503 }
2504
2505 static void g_stop(struct seq_file *m, void *p)
2506 {
2507         mutex_unlock(&graph_lock);
2508 }
2509
2510 static int g_show(struct seq_file *m, void *v)
2511 {
2512         unsigned long *ptr = v;
2513
2514         if (!ptr)
2515                 return 0;
2516
2517         if (ptr == (unsigned long *)1) {
2518                 seq_printf(m, "#### all functions enabled ####\n");
2519                 return 0;
2520         }
2521
2522         seq_printf(m, "%ps\n", (void *)*ptr);
2523
2524         return 0;
2525 }
2526
2527 static const struct seq_operations ftrace_graph_seq_ops = {
2528         .start = g_start,
2529         .next = g_next,
2530         .stop = g_stop,
2531         .show = g_show,
2532 };
2533
2534 static int
2535 ftrace_graph_open(struct inode *inode, struct file *file)
2536 {
2537         int ret = 0;
2538
2539         if (unlikely(ftrace_disabled))
2540                 return -ENODEV;
2541
2542         mutex_lock(&graph_lock);
2543         if ((file->f_mode & FMODE_WRITE) &&
2544             (file->f_flags & O_TRUNC)) {
2545                 ftrace_graph_filter_enabled = 0;
2546                 ftrace_graph_count = 0;
2547                 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2548         }
2549         mutex_unlock(&graph_lock);
2550
2551         if (file->f_mode & FMODE_READ)
2552                 ret = seq_open(file, &ftrace_graph_seq_ops);
2553
2554         return ret;
2555 }
2556
2557 static int
2558 ftrace_graph_release(struct inode *inode, struct file *file)
2559 {
2560         if (file->f_mode & FMODE_READ)
2561                 seq_release(inode, file);
2562         return 0;
2563 }
2564
2565 static int
2566 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2567 {
2568         struct dyn_ftrace *rec;
2569         struct ftrace_page *pg;
2570         int search_len;
2571         int fail = 1;
2572         int type, not;
2573         char *search;
2574         bool exists;
2575         int i;
2576
2577         if (ftrace_disabled)
2578                 return -ENODEV;
2579
2580         /* decode regex */
2581         type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
2582         if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
2583                 return -EBUSY;
2584
2585         search_len = strlen(search);
2586
2587         mutex_lock(&ftrace_lock);
2588         do_for_each_ftrace_rec(pg, rec) {
2589
2590                 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2591                         continue;
2592
2593                 if (ftrace_match_record(rec, search, search_len, type)) {
2594                         /* if it is in the array */
2595                         exists = false;
2596                         for (i = 0; i < *idx; i++) {
2597                                 if (array[i] == rec->ip) {
2598                                         exists = true;
2599                                         break;
2600                                 }
2601                         }
2602
2603                         if (!not) {
2604                                 fail = 0;
2605                                 if (!exists) {
2606                                         array[(*idx)++] = rec->ip;
2607                                         if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2608                                                 goto out;
2609                                 }
2610                         } else {
2611                                 if (exists) {
2612                                         array[i] = array[--(*idx)];
2613                                         array[*idx] = 0;
2614                                         fail = 0;
2615                                 }
2616                         }
2617                 }
2618         } while_for_each_ftrace_rec();
2619 out:
2620         mutex_unlock(&ftrace_lock);
2621
2622         if (fail)
2623                 return -EINVAL;
2624
2625         ftrace_graph_filter_enabled = 1;
2626         return 0;
2627 }
2628
2629 static ssize_t
2630 ftrace_graph_write(struct file *file, const char __user *ubuf,
2631                    size_t cnt, loff_t *ppos)
2632 {
2633         struct trace_parser parser;
2634         ssize_t read, ret;
2635
2636         if (!cnt)
2637                 return 0;
2638
2639         mutex_lock(&graph_lock);
2640
2641         if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
2642                 ret = -ENOMEM;
2643                 goto out_unlock;
2644         }
2645
2646         read = trace_get_user(&parser, ubuf, cnt, ppos);
2647
2648         if (read >= 0 && trace_parser_loaded((&parser))) {
2649                 parser.buffer[parser.idx] = 0;
2650
2651                 /* we allow only one expression at a time */
2652                 ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
2653                                         parser.buffer);
2654                 if (ret)
2655                         goto out_free;
2656         }
2657
2658         ret = read;
2659
2660 out_free:
2661         trace_parser_put(&parser);
2662 out_unlock:
2663         mutex_unlock(&graph_lock);
2664
2665         return ret;
2666 }
2667
2668 static const struct file_operations ftrace_graph_fops = {
2669         .open           = ftrace_graph_open,
2670         .read           = seq_read,
2671         .write          = ftrace_graph_write,
2672         .release        = ftrace_graph_release,
2673         .llseek         = seq_lseek,
2674 };
2675 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2676
2677 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2678 {
2679
2680         trace_create_file("available_filter_functions", 0444,
2681                         d_tracer, NULL, &ftrace_avail_fops);
2682
2683         trace_create_file("failures", 0444,
2684                         d_tracer, NULL, &ftrace_failures_fops);
2685
2686         trace_create_file("set_ftrace_filter", 0644, d_tracer,
2687                         NULL, &ftrace_filter_fops);
2688
2689         trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2690                                     NULL, &ftrace_notrace_fops);
2691
2692 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2693         trace_create_file("set_graph_function", 0444, d_tracer,
2694                                     NULL,
2695                                     &ftrace_graph_fops);
2696 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2697
2698         return 0;
2699 }
2700
2701 static int ftrace_process_locs(struct module *mod,
2702                                unsigned long *start,
2703                                unsigned long *end)
2704 {
2705         unsigned long *p;
2706         unsigned long addr;
2707         unsigned long flags;
2708
2709         mutex_lock(&ftrace_lock);
2710         p = start;
2711         while (p < end) {
2712                 addr = ftrace_call_adjust(*p++);
2713                 /*
2714                  * Some architecture linkers will pad between
2715                  * the different mcount_loc sections of different
2716                  * object files to satisfy alignments.
2717                  * Skip any NULL pointers.
2718                  */
2719                 if (!addr)
2720                         continue;
2721                 ftrace_record_ip(addr);
2722         }
2723
2724         /* disable interrupts to prevent kstop machine */
2725         local_irq_save(flags);
2726         ftrace_update_code(mod);
2727         local_irq_restore(flags);
2728         mutex_unlock(&ftrace_lock);
2729
2730         return 0;
2731 }
2732
2733 #ifdef CONFIG_MODULES
2734 void ftrace_release_mod(struct module *mod)
2735 {
2736         struct dyn_ftrace *rec;
2737         struct ftrace_page *pg;
2738
2739         if (ftrace_disabled)
2740                 return;
2741
2742         mutex_lock(&ftrace_lock);
2743         do_for_each_ftrace_rec(pg, rec) {
2744                 if (within_module_core(rec->ip, mod)) {
2745                         /*
2746                          * rec->ip is changed in ftrace_free_rec()
2747                          * It should not between s and e if record was freed.
2748                          */
2749                         FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2750                         ftrace_free_rec(rec);
2751                 }
2752         } while_for_each_ftrace_rec();
2753         mutex_unlock(&ftrace_lock);
2754 }
2755
2756 static void ftrace_init_module(struct module *mod,
2757                                unsigned long *start, unsigned long *end)
2758 {
2759         if (ftrace_disabled || start == end)
2760                 return;
2761         ftrace_process_locs(mod, start, end);
2762 }
2763
2764 static int ftrace_module_notify(struct notifier_block *self,
2765                                 unsigned long val, void *data)
2766 {
2767         struct module *mod = data;
2768
2769         switch (val) {
2770         case MODULE_STATE_COMING:
2771                 ftrace_init_module(mod, mod->ftrace_callsites,
2772                                    mod->ftrace_callsites +
2773                                    mod->num_ftrace_callsites);
2774                 break;
2775         case MODULE_STATE_GOING:
2776                 ftrace_release_mod(mod);
2777                 break;
2778         }
2779
2780         return 0;
2781 }
2782 #else
2783 static int ftrace_module_notify(struct notifier_block *self,
2784                                 unsigned long val, void *data)
2785 {
2786         return 0;
2787 }
2788 #endif /* CONFIG_MODULES */
2789
2790 struct notifier_block ftrace_module_nb = {
2791         .notifier_call = ftrace_module_notify,
2792         .priority = 0,
2793 };
2794
2795 extern unsigned long __start_mcount_loc[];
2796 extern unsigned long __stop_mcount_loc[];
2797
2798 void __init ftrace_init(void)
2799 {
2800         unsigned long count, addr, flags;
2801         int ret;
2802
2803         /* Keep the ftrace pointer to the stub */
2804         addr = (unsigned long)ftrace_stub;
2805
2806         local_irq_save(flags);
2807         ftrace_dyn_arch_init(&addr);
2808         local_irq_restore(flags);
2809
2810         /* ftrace_dyn_arch_init places the return code in addr */
2811         if (addr)
2812                 goto failed;
2813
2814         count = __stop_mcount_loc - __start_mcount_loc;
2815
2816         ret = ftrace_dyn_table_alloc(count);
2817         if (ret)
2818                 goto failed;
2819
2820         last_ftrace_enabled = ftrace_enabled = 1;
2821
2822         ret = ftrace_process_locs(NULL,
2823                                   __start_mcount_loc,
2824                                   __stop_mcount_loc);
2825
2826         ret = register_module_notifier(&ftrace_module_nb);
2827         if (ret)
2828                 pr_warning("Failed to register trace ftrace module notifier\n");
2829
2830         set_ftrace_early_filters();
2831
2832         return;
2833  failed:
2834         ftrace_disabled = 1;
2835 }
2836
2837 #else
2838
2839 static int __init ftrace_nodyn_init(void)
2840 {
2841         ftrace_enabled = 1;
2842         return 0;
2843 }
2844 device_initcall(ftrace_nodyn_init);
2845
2846 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2847 static inline void ftrace_startup_enable(int command) { }
2848 /* Keep as macros so we do not need to define the commands */
2849 # define ftrace_startup(command)        do { } while (0)
2850 # define ftrace_shutdown(command)       do { } while (0)
2851 # define ftrace_startup_sysctl()        do { } while (0)
2852 # define ftrace_shutdown_sysctl()       do { } while (0)
2853 #endif /* CONFIG_DYNAMIC_FTRACE */
2854
2855 static void clear_ftrace_swapper(void)
2856 {
2857         struct task_struct *p;
2858         int cpu;
2859
2860         get_online_cpus();
2861         for_each_online_cpu(cpu) {
2862                 p = idle_task(cpu);
2863                 clear_tsk_trace_trace(p);
2864         }
2865         put_online_cpus();
2866 }
2867
2868 static void set_ftrace_swapper(void)
2869 {
2870         struct task_struct *p;
2871         int cpu;
2872
2873         get_online_cpus();
2874         for_each_online_cpu(cpu) {
2875                 p = idle_task(cpu);
2876                 set_tsk_trace_trace(p);
2877         }
2878         put_online_cpus();
2879 }
2880
2881 static void clear_ftrace_pid(struct pid *pid)
2882 {
2883         struct task_struct *p;
2884
2885         rcu_read_lock();
2886         do_each_pid_task(pid, PIDTYPE_PID, p) {
2887                 clear_tsk_trace_trace(p);
2888         } while_each_pid_task(pid, PIDTYPE_PID, p);
2889         rcu_read_unlock();
2890
2891         put_pid(pid);
2892 }
2893
2894 static void set_ftrace_pid(struct pid *pid)
2895 {
2896         struct task_struct *p;
2897
2898         rcu_read_lock();
2899         do_each_pid_task(pid, PIDTYPE_PID, p) {
2900                 set_tsk_trace_trace(p);
2901         } while_each_pid_task(pid, PIDTYPE_PID, p);
2902         rcu_read_unlock();
2903 }
2904
2905 static void clear_ftrace_pid_task(struct pid *pid)
2906 {
2907         if (pid == ftrace_swapper_pid)
2908                 clear_ftrace_swapper();
2909         else
2910                 clear_ftrace_pid(pid);
2911 }
2912
2913 static void set_ftrace_pid_task(struct pid *pid)
2914 {
2915         if (pid == ftrace_swapper_pid)
2916                 set_ftrace_swapper();
2917         else
2918                 set_ftrace_pid(pid);
2919 }
2920
2921 static int ftrace_pid_add(int p)
2922 {
2923         struct pid *pid;
2924         struct ftrace_pid *fpid;
2925         int ret = -EINVAL;
2926
2927         mutex_lock(&ftrace_lock);
2928
2929         if (!p)
2930                 pid = ftrace_swapper_pid;
2931         else
2932                 pid = find_get_pid(p);
2933
2934         if (!pid)
2935                 goto out;
2936
2937         ret = 0;
2938
2939         list_for_each_entry(fpid, &ftrace_pids, list)
2940                 if (fpid->pid == pid)
2941                         goto out_put;
2942
2943         ret = -ENOMEM;
2944
2945         fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
2946         if (!fpid)
2947                 goto out_put;
2948
2949         list_add(&fpid->list, &ftrace_pids);
2950         fpid->pid = pid;
2951
2952         set_ftrace_pid_task(pid);
2953
2954         ftrace_update_pid_func();
2955         ftrace_startup_enable(0);
2956
2957         mutex_unlock(&ftrace_lock);
2958         return 0;
2959
2960 out_put:
2961         if (pid != ftrace_swapper_pid)
2962                 put_pid(pid);
2963
2964 out:
2965         mutex_unlock(&ftrace_lock);
2966         return ret;
2967 }
2968
2969 static void ftrace_pid_reset(void)
2970 {
2971         struct ftrace_pid *fpid, *safe;
2972
2973         mutex_lock(&ftrace_lock);
2974         list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
2975                 struct pid *pid = fpid->pid;
2976
2977                 clear_ftrace_pid_task(pid);
2978
2979                 list_del(&fpid->list);
2980                 kfree(fpid);
2981         }
2982
2983         ftrace_update_pid_func();
2984         ftrace_startup_enable(0);
2985
2986         mutex_unlock(&ftrace_lock);
2987 }
2988
2989 static void *fpid_start(struct seq_file *m, loff_t *pos)
2990 {
2991         mutex_lock(&ftrace_lock);
2992
2993         if (list_empty(&ftrace_pids) && (!*pos))
2994                 return (void *) 1;
2995
2996         return seq_list_start(&ftrace_pids, *pos);
2997 }
2998
2999 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
3000 {
3001         if (v == (void *)1)
3002                 return NULL;
3003
3004         return seq_list_next(v, &ftrace_pids, pos);
3005 }
3006
3007 static void fpid_stop(struct seq_file *m, void *p)
3008 {
3009         mutex_unlock(&ftrace_lock);
3010 }
3011
3012 static int fpid_show(struct seq_file *m, void *v)
3013 {
3014         const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
3015
3016         if (v == (void *)1) {
3017                 seq_printf(m, "no pid\n");
3018                 return 0;
3019         }
3020
3021         if (fpid->pid == ftrace_swapper_pid)
3022                 seq_printf(m, "swapper tasks\n");
3023         else
3024                 seq_printf(m, "%u\n", pid_vnr(fpid->pid));
3025
3026         return 0;
3027 }
3028
3029 static const struct seq_operations ftrace_pid_sops = {
3030         .start = fpid_start,
3031         .next = fpid_next,
3032         .stop = fpid_stop,
3033         .show = fpid_show,
3034 };
3035
3036 static int
3037 ftrace_pid_open(struct inode *inode, struct file *file)
3038 {
3039         int ret = 0;
3040
3041         if ((file->f_mode & FMODE_WRITE) &&
3042             (file->f_flags & O_TRUNC))
3043                 ftrace_pid_reset();
3044
3045         if (file->f_mode & FMODE_READ)
3046                 ret = seq_open(file, &ftrace_pid_sops);
3047
3048         return ret;
3049 }
3050
3051 static ssize_t
3052 ftrace_pid_write(struct file *filp, const char __user *ubuf,
3053                    size_t cnt, loff_t *ppos)
3054 {
3055         char buf[64], *tmp;
3056         long val;
3057         int ret;
3058
3059         if (cnt >= sizeof(buf))
3060                 return -EINVAL;
3061
3062         if (copy_from_user(&buf, ubuf, cnt))
3063                 return -EFAULT;
3064
3065         buf[cnt] = 0;
3066
3067         /*
3068          * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
3069          * to clean the filter quietly.
3070          */
3071         tmp = strstrip(buf);
3072         if (strlen(tmp) == 0)
3073                 return 1;
3074
3075         ret = strict_strtol(tmp, 10, &val);
3076         if (ret < 0)
3077                 return ret;
3078
3079         ret = ftrace_pid_add(val);
3080
3081         return ret ? ret : cnt;
3082 }
3083
3084 static int
3085 ftrace_pid_release(struct inode *inode, struct file *file)
3086 {
3087         if (file->f_mode & FMODE_READ)
3088                 seq_release(inode, file);
3089
3090         return 0;
3091 }
3092
3093 static const struct file_operations ftrace_pid_fops = {
3094         .open           = ftrace_pid_open,
3095         .write          = ftrace_pid_write,
3096         .read           = seq_read,
3097         .llseek         = seq_lseek,
3098         .release        = ftrace_pid_release,
3099 };
3100
3101 static __init int ftrace_init_debugfs(void)
3102 {
3103         struct dentry *d_tracer;
3104
3105         d_tracer = tracing_init_dentry();
3106         if (!d_tracer)
3107                 return 0;
3108
3109         ftrace_init_dyn_debugfs(d_tracer);
3110
3111         trace_create_file("set_ftrace_pid", 0644, d_tracer,
3112                             NULL, &ftrace_pid_fops);
3113
3114         ftrace_profile_debugfs(d_tracer);
3115
3116         return 0;
3117 }
3118 fs_initcall(ftrace_init_debugfs);
3119
3120 /**
3121  * ftrace_kill - kill ftrace
3122  *
3123  * This function should be used by panic code. It stops ftrace
3124  * but in a not so nice way. If you need to simply kill ftrace
3125  * from a non-atomic section, use ftrace_kill.
3126  */
3127 void ftrace_kill(void)
3128 {
3129         ftrace_disabled = 1;
3130         ftrace_enabled = 0;
3131         clear_ftrace_function();
3132 }
3133
3134 /**
3135  * register_ftrace_function - register a function for profiling
3136  * @ops - ops structure that holds the function for profiling.
3137  *
3138  * Register a function to be called by all functions in the
3139  * kernel.
3140  *
3141  * Note: @ops->func and all the functions it calls must be labeled
3142  *       with "notrace", otherwise it will go into a
3143  *       recursive loop.
3144  */
3145 int register_ftrace_function(struct ftrace_ops *ops)
3146 {
3147         int ret;
3148
3149         if (unlikely(ftrace_disabled))
3150                 return -1;
3151
3152         mutex_lock(&ftrace_lock);
3153
3154         ret = __register_ftrace_function(ops);
3155         ftrace_startup(0);
3156
3157         mutex_unlock(&ftrace_lock);
3158         return ret;
3159 }
3160
3161 /**
3162  * unregister_ftrace_function - unregister a function for profiling.
3163  * @ops - ops structure that holds the function to unregister
3164  *
3165  * Unregister a function that was added to be called by ftrace profiling.
3166  */
3167 int unregister_ftrace_function(struct ftrace_ops *ops)
3168 {
3169         int ret;
3170
3171         mutex_lock(&ftrace_lock);
3172         ret = __unregister_ftrace_function(ops);
3173         ftrace_shutdown(0);
3174         mutex_unlock(&ftrace_lock);
3175
3176         return ret;
3177 }
3178
3179 int
3180 ftrace_enable_sysctl(struct ctl_table *table, int write,
3181                      void __user *buffer, size_t *lenp,
3182                      loff_t *ppos)
3183 {
3184         int ret;
3185
3186         if (unlikely(ftrace_disabled))
3187                 return -ENODEV;
3188
3189         mutex_lock(&ftrace_lock);
3190
3191         ret  = proc_dointvec(table, write, buffer, lenp, ppos);
3192
3193         if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3194                 goto out;
3195
3196         last_ftrace_enabled = !!ftrace_enabled;
3197
3198         if (ftrace_enabled) {
3199
3200                 ftrace_startup_sysctl();
3201
3202                 /* we are starting ftrace again */
3203                 if (ftrace_list != &ftrace_list_end) {
3204                         if (ftrace_list->next == &ftrace_list_end)
3205                                 ftrace_trace_function = ftrace_list->func;
3206                         else
3207                                 ftrace_trace_function = ftrace_list_func;
3208                 }
3209
3210         } else {
3211                 /* stopping ftrace calls (just send to ftrace_stub) */
3212                 ftrace_trace_function = ftrace_stub;
3213
3214                 ftrace_shutdown_sysctl();
3215         }
3216
3217  out:
3218         mutex_unlock(&ftrace_lock);
3219         return ret;
3220 }
3221
3222 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3223
3224 static int ftrace_graph_active;
3225 static struct notifier_block ftrace_suspend_notifier;
3226
3227 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3228 {
3229         return 0;
3230 }
3231
3232 /* The callbacks that hook a function */
3233 trace_func_graph_ret_t ftrace_graph_return =
3234                         (trace_func_graph_ret_t)ftrace_stub;
3235 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3236
3237 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3238 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3239 {
3240         int i;
3241         int ret = 0;
3242         unsigned long flags;
3243         int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3244         struct task_struct *g, *t;
3245
3246         for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3247                 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3248                                         * sizeof(struct ftrace_ret_stack),
3249                                         GFP_KERNEL);
3250                 if (!ret_stack_list[i]) {
3251                         start = 0;
3252                         end = i;
3253                         ret = -ENOMEM;
3254                         goto free;
3255                 }
3256         }
3257
3258         read_lock_irqsave(&tasklist_lock, flags);
3259         do_each_thread(g, t) {
3260                 if (start == end) {
3261                         ret = -EAGAIN;
3262                         goto unlock;
3263                 }
3264
3265                 if (t->ret_stack == NULL) {
3266                         atomic_set(&t->tracing_graph_pause, 0);
3267                         atomic_set(&t->trace_overrun, 0);
3268                         t->curr_ret_stack = -1;
3269                         /* Make sure the tasks see the -1 first: */
3270                         smp_wmb();
3271                         t->ret_stack = ret_stack_list[start++];
3272                 }
3273         } while_each_thread(g, t);
3274
3275 unlock:
3276         read_unlock_irqrestore(&tasklist_lock, flags);
3277 free:
3278         for (i = start; i < end; i++)
3279                 kfree(ret_stack_list[i]);
3280         return ret;
3281 }
3282
3283 static void
3284 ftrace_graph_probe_sched_switch(void *ignore,
3285                         struct task_struct *prev, struct task_struct *next)
3286 {
3287         unsigned long long timestamp;
3288         int index;
3289
3290         /*
3291          * Does the user want to count the time a function was asleep.
3292          * If so, do not update the time stamps.
3293          */
3294         if (trace_flags & TRACE_ITER_SLEEP_TIME)
3295                 return;
3296
3297         timestamp = trace_clock_local();
3298
3299         prev->ftrace_timestamp = timestamp;
3300
3301         /* only process tasks that we timestamped */
3302         if (!next->ftrace_timestamp)
3303                 return;
3304
3305         /*
3306          * Update all the counters in next to make up for the
3307          * time next was sleeping.
3308          */
3309         timestamp -= next->ftrace_timestamp;
3310
3311         for (index = next->curr_ret_stack; index >= 0; index--)
3312                 next->ret_stack[index].calltime += timestamp;
3313 }
3314
3315 /* Allocate a return stack for each task */
3316 static int start_graph_tracing(void)
3317 {
3318         struct ftrace_ret_stack **ret_stack_list;
3319         int ret, cpu;
3320
3321         ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3322                                 sizeof(struct ftrace_ret_stack *),
3323                                 GFP_KERNEL);
3324
3325         if (!ret_stack_list)
3326                 return -ENOMEM;
3327
3328         /* The cpu_boot init_task->ret_stack will never be freed */
3329         for_each_online_cpu(cpu) {
3330                 if (!idle_task(cpu)->ret_stack)
3331                         ftrace_graph_init_idle_task(idle_task(cpu), cpu);
3332         }
3333
3334         do {
3335                 ret = alloc_retstack_tasklist(ret_stack_list);
3336         } while (ret == -EAGAIN);
3337
3338         if (!ret) {
3339                 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3340                 if (ret)
3341                         pr_info("ftrace_graph: Couldn't activate tracepoint"
3342                                 " probe to kernel_sched_switch\n");
3343         }
3344
3345         kfree(ret_stack_list);
3346         return ret;
3347 }
3348
3349 /*
3350  * Hibernation protection.
3351  * The state of the current task is too much unstable during
3352  * suspend/restore to disk. We want to protect against that.
3353  */
3354 static int
3355 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3356                                                         void *unused)
3357 {
3358         switch (state) {
3359         case PM_HIBERNATION_PREPARE:
3360                 pause_graph_tracing();
3361                 break;
3362
3363         case PM_POST_HIBERNATION:
3364                 unpause_graph_tracing();
3365                 break;
3366         }
3367         return NOTIFY_DONE;
3368 }
3369
3370 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3371                         trace_func_graph_ent_t entryfunc)
3372 {
3373         int ret = 0;
3374
3375         mutex_lock(&ftrace_lock);
3376
3377         /* we currently allow only one tracer registered at a time */
3378         if (ftrace_graph_active) {
3379                 ret = -EBUSY;
3380                 goto out;
3381         }
3382
3383         ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3384         register_pm_notifier(&ftrace_suspend_notifier);
3385
3386         ftrace_graph_active++;
3387         ret = start_graph_tracing();
3388         if (ret) {
3389                 ftrace_graph_active--;
3390                 goto out;
3391         }
3392
3393         ftrace_graph_return = retfunc;
3394         ftrace_graph_entry = entryfunc;
3395
3396         ftrace_startup(FTRACE_START_FUNC_RET);
3397
3398 out:
3399         mutex_unlock(&ftrace_lock);
3400         return ret;
3401 }
3402
3403 void unregister_ftrace_graph(void)
3404 {
3405         mutex_lock(&ftrace_lock);
3406
3407         if (unlikely(!ftrace_graph_active))
3408                 goto out;
3409
3410         ftrace_graph_active--;
3411         ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3412         ftrace_graph_entry = ftrace_graph_entry_stub;
3413         ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3414         unregister_pm_notifier(&ftrace_suspend_notifier);
3415         unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
3416
3417  out:
3418         mutex_unlock(&ftrace_lock);
3419 }
3420
3421 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
3422
3423 static void
3424 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
3425 {
3426         atomic_set(&t->tracing_graph_pause, 0);
3427         atomic_set(&t->trace_overrun, 0);
3428         t->ftrace_timestamp = 0;
3429         /* make curr_ret_stack visable before we add the ret_stack */
3430         smp_wmb();
3431         t->ret_stack = ret_stack;
3432 }
3433
3434 /*
3435  * Allocate a return stack for the idle task. May be the first
3436  * time through, or it may be done by CPU hotplug online.
3437  */
3438 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
3439 {
3440         t->curr_ret_stack = -1;
3441         /*
3442          * The idle task has no parent, it either has its own
3443          * stack or no stack at all.
3444          */
3445         if (t->ret_stack)
3446                 WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
3447
3448         if (ftrace_graph_active) {
3449                 struct ftrace_ret_stack *ret_stack;
3450
3451                 ret_stack = per_cpu(idle_ret_stack, cpu);
3452                 if (!ret_stack) {
3453                         ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3454                                             * sizeof(struct ftrace_ret_stack),
3455                                             GFP_KERNEL);
3456                         if (!ret_stack)
3457                                 return;
3458                         per_cpu(idle_ret_stack, cpu) = ret_stack;
3459                 }
3460                 graph_init_task(t, ret_stack);
3461         }
3462 }
3463
3464 /* Allocate a return stack for newly created task */
3465 void ftrace_graph_init_task(struct task_struct *t)
3466 {
3467         /* Make sure we do not use the parent ret_stack */
3468         t->ret_stack = NULL;
3469         t->curr_ret_stack = -1;
3470
3471         if (ftrace_graph_active) {
3472                 struct ftrace_ret_stack *ret_stack;
3473
3474                 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3475                                 * sizeof(struct ftrace_ret_stack),
3476                                 GFP_KERNEL);
3477                 if (!ret_stack)
3478                         return;
3479                 graph_init_task(t, ret_stack);
3480         }
3481 }
3482
3483 void ftrace_graph_exit_task(struct task_struct *t)
3484 {
3485         struct ftrace_ret_stack *ret_stack = t->ret_stack;
3486
3487         t->ret_stack = NULL;
3488         /* NULL must become visible to IRQs before we free it: */
3489         barrier();
3490
3491         kfree(ret_stack);
3492 }
3493
3494 void ftrace_graph_stop(void)
3495 {
3496         ftrace_stop();
3497 }
3498 #endif