kernel/trace/ftrace.c

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