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