kernel/trace/ftrace.c

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