SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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         func = ftrace_trace_function;
 229
 230         if (ftrace_pid_trace) {
 231                 set_ftrace_pid_function(func);
 232                 func = ftrace_pid_func;
 233         } else {
 234                 if (func == ftrace_pid_func)
 235                         func = ftrace_pid_function;
 236         }
 237
 238 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
 239         ftrace_trace_function = func;
 240 #else
 241         __ftrace_trace_function = func;
 242 #endif
 243 }
 244
 245 #ifdef CONFIG_FUNCTION_PROFILER
 246 struct ftrace_profile {
 247         struct hlist_node               node;
 248         unsigned long                   ip;
 249         unsigned long                   counter;
 250 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 251         unsigned long long              time;
 252 #endif
 253 };
 254
 255 struct ftrace_profile_page {
 256         struct ftrace_profile_page      *next;
 257         unsigned long                   index;
 258         struct ftrace_profile           records[];
 259 };
 260
 261 struct ftrace_profile_stat {
 262         atomic_t                        disabled;
 263         struct hlist_head               *hash;
 264         struct ftrace_profile_page      *pages;
 265         struct ftrace_profile_page      *start;
 266         struct tracer_stat              stat;
 267 };
 268
 269 #define PROFILE_RECORDS_SIZE                                            \
 270         (PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
 271
 272 #define PROFILES_PER_PAGE                                       \
 273         (PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
 274
 275 static int ftrace_profile_bits __read_mostly;
 276 static int ftrace_profile_enabled __read_mostly;
 277
 278 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
 279 static DEFINE_MUTEX(ftrace_profile_lock);
 280
 281 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
 282
 283 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
 284
 285 static void *
 286 function_stat_next(void *v, int idx)
 287 {
 288         struct ftrace_profile *rec = v;
 289         struct ftrace_profile_page *pg;
 290
 291         pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
 292
 293  again:
 294         if (idx != 0)
 295                 rec++;
 296
 297         if ((void *)rec >= (void *)&pg->records[pg->index]) {
 298                 pg = pg->next;
 299                 if (!pg)
 300                         return NULL;
 301                 rec = &pg->records[0];
 302                 if (!rec->counter)
 303                         goto again;
 304         }
 305
 306         return rec;
 307 }
 308
 309 static void *function_stat_start(struct tracer_stat *trace)
 310 {
 311         struct ftrace_profile_stat *stat =
 312                 container_of(trace, struct ftrace_profile_stat, stat);
 313
 314         if (!stat || !stat->start)
 315                 return NULL;
 316
 317         return function_stat_next(&stat->start->records[0], 0);
 318 }
 319
 320 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 321 /* function graph compares on total time */
 322 static int function_stat_cmp(void *p1, void *p2)
 323 {
 324         struct ftrace_profile *a = p1;
 325         struct ftrace_profile *b = p2;
 326
 327         if (a->time < b->time)
 328                 return -1;
 329         if (a->time > b->time)
 330                 return 1;
 331         else
 332                 return 0;
 333 }
 334 #else
 335 /* not function graph compares against hits */
 336 static int function_stat_cmp(void *p1, void *p2)
 337 {
 338         struct ftrace_profile *a = p1;
 339         struct ftrace_profile *b = p2;
 340
 341         if (a->counter < b->counter)
 342                 return -1;
 343         if (a->counter > b->counter)
 344                 return 1;
 345         else
 346                 return 0;
 347 }
 348 #endif
 349
 350 static int function_stat_headers(struct seq_file *m)
 351 {
 352 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 353         seq_printf(m, "  Function                               "
 354                    "Hit    Time            Avg\n"
 355                       "  --------                               "
 356                    "---    ----            ---\n");
 357 #else
 358         seq_printf(m, "  Function                               Hit\n"
 359                       "  --------                               ---\n");
 360 #endif
 361         return 0;
 362 }
 363
 364 static int function_stat_show(struct seq_file *m, void *v)
 365 {
 366         struct ftrace_profile *rec = v;
 367         char str[KSYM_SYMBOL_LEN];
 368 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 369         static DEFINE_MUTEX(mutex);
 370         static struct trace_seq s;
 371         unsigned long long avg;
 372 #endif
 373
 374         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 375         seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
 376
 377 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 378         seq_printf(m, "    ");
 379         avg = rec->time;
 380         do_div(avg, rec->counter);
 381
 382         mutex_lock(&mutex);
 383         trace_seq_init(&s);
 384         trace_print_graph_duration(rec->time, &s);
 385         trace_seq_puts(&s, "    ");
 386         trace_print_graph_duration(avg, &s);
 387         trace_print_seq(m, &s);
 388         mutex_unlock(&mutex);
 389 #endif
 390         seq_putc(m, '\n');
 391
 392         return 0;
 393 }
 394
 395 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
 396 {
 397         struct ftrace_profile_page *pg;
 398
 399         pg = stat->pages = stat->start;
 400
 401         while (pg) {
 402                 memset(pg->records, 0, PROFILE_RECORDS_SIZE);
 403                 pg->index = 0;
 404                 pg = pg->next;
 405         }
 406
 407         memset(stat->hash, 0,
 408                FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
 409 }
 410
 411 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
 412 {
 413         struct ftrace_profile_page *pg;
 414         int functions;
 415         int pages;
 416         int i;
 417
 418         /* If we already allocated, do nothing */
 419         if (stat->pages)
 420                 return 0;
 421
 422         stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
 423         if (!stat->pages)
 424                 return -ENOMEM;
 425
 426 #ifdef CONFIG_DYNAMIC_FTRACE
 427         functions = ftrace_update_tot_cnt;
 428 #else
 429         /*
 430          * We do not know the number of functions that exist because
 431          * dynamic tracing is what counts them. With past experience
 432          * we have around 20K functions. That should be more than enough.
 433          * It is highly unlikely we will execute every function in
 434          * the kernel.
 435          */
 436         functions = 20000;
 437 #endif
 438
 439         pg = stat->start = stat->pages;
 440
 441         pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
 442
 443         for (i = 0; i < pages; i++) {
 444                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
 445                 if (!pg->next)
 446                         goto out_free;
 447                 pg = pg->next;
 448         }
 449
 450         return 0;
 451
 452  out_free:
 453         pg = stat->start;
 454         while (pg) {
 455                 unsigned long tmp = (unsigned long)pg;
 456
 457                 pg = pg->next;
 458                 free_page(tmp);
 459         }
 460
 461         free_page((unsigned long)stat->pages);
 462         stat->pages = NULL;
 463         stat->start = NULL;
 464
 465         return -ENOMEM;
 466 }
 467
 468 static int ftrace_profile_init_cpu(int cpu)
 469 {
 470         struct ftrace_profile_stat *stat;
 471         int size;
 472
 473         stat = &per_cpu(ftrace_profile_stats, cpu);
 474
 475         if (stat->hash) {
 476                 /* If the profile is already created, simply reset it */
 477                 ftrace_profile_reset(stat);
 478                 return 0;
 479         }
 480
 481         /*
 482          * We are profiling all functions, but usually only a few thousand
 483          * functions are hit. We'll make a hash of 1024 items.
 484          */
 485         size = FTRACE_PROFILE_HASH_SIZE;
 486
 487         stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
 488
 489         if (!stat->hash)
 490                 return -ENOMEM;
 491
 492         if (!ftrace_profile_bits) {
 493                 size--;
 494
 495                 for (; size; size >>= 1)
 496                         ftrace_profile_bits++;
 497         }
 498
 499         /* Preallocate the function profiling pages */
 500         if (ftrace_profile_pages_init(stat) < 0) {
 501                 kfree(stat->hash);
 502                 stat->hash = NULL;
 503                 return -ENOMEM;
 504         }
 505
 506         return 0;
 507 }
 508
 509 static int ftrace_profile_init(void)
 510 {
 511         int cpu;
 512         int ret = 0;
 513
 514         for_each_online_cpu(cpu) {
 515                 ret = ftrace_profile_init_cpu(cpu);
 516                 if (ret)
 517                         break;
 518         }
 519
 520         return ret;
 521 }
 522
 523 /* interrupts must be disabled */
 524 static struct ftrace_profile *
 525 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
 526 {
 527         struct ftrace_profile *rec;
 528         struct hlist_head *hhd;
 529         struct hlist_node *n;
 530         unsigned long key;
 531
 532         key = hash_long(ip, ftrace_profile_bits);
 533         hhd = &stat->hash[key];
 534
 535         if (hlist_empty(hhd))
 536                 return NULL;
 537
 538         hlist_for_each_entry_rcu(rec, n, hhd, node) {
 539                 if (rec->ip == ip)
 540                         return rec;
 541         }
 542
 543         return NULL;
 544 }
 545
 546 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
 547                                struct ftrace_profile *rec)
 548 {
 549         unsigned long key;
 550
 551         key = hash_long(rec->ip, ftrace_profile_bits);
 552         hlist_add_head_rcu(&rec->node, &stat->hash[key]);
 553 }
 554
 555 /*
 556  * The memory is already allocated, this simply finds a new record to use.
 557  */
 558 static struct ftrace_profile *
 559 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
 560 {
 561         struct ftrace_profile *rec = NULL;
 562
 563         /* prevent recursion (from NMIs) */
 564         if (atomic_inc_return(&stat->disabled) != 1)
 565                 goto out;
 566
 567         /*
 568          * Try to find the function again since an NMI
 569          * could have added it
 570          */
 571         rec = ftrace_find_profiled_func(stat, ip);
 572         if (rec)
 573                 goto out;
 574
 575         if (stat->pages->index == PROFILES_PER_PAGE) {
 576                 if (!stat->pages->next)
 577                         goto out;
 578                 stat->pages = stat->pages->next;
 579         }
 580
 581         rec = &stat->pages->records[stat->pages->index++];
 582         rec->ip = ip;
 583         ftrace_add_profile(stat, rec);
 584
 585  out:
 586         atomic_dec(&stat->disabled);
 587
 588         return rec;
 589 }
 590
 591 static void
 592 function_profile_call(unsigned long ip, unsigned long parent_ip)
 593 {
 594         struct ftrace_profile_stat *stat;
 595         struct ftrace_profile *rec;
 596         unsigned long flags;
 597
 598         if (!ftrace_profile_enabled)
 599                 return;
 600
 601         local_irq_save(flags);
 602
 603         stat = &__get_cpu_var(ftrace_profile_stats);
 604         if (!stat->hash || !ftrace_profile_enabled)
 605                 goto out;
 606
 607         rec = ftrace_find_profiled_func(stat, ip);
 608         if (!rec) {
 609                 rec = ftrace_profile_alloc(stat, ip);
 610                 if (!rec)
 611                         goto out;
 612         }
 613
 614         rec->counter++;
 615  out:
 616         local_irq_restore(flags);
 617 }
 618
 619 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 620 static int profile_graph_entry(struct ftrace_graph_ent *trace)
 621 {
 622         function_profile_call(trace->func, 0);
 623         return 1;
 624 }
 625
 626 static void profile_graph_return(struct ftrace_graph_ret *trace)
 627 {
 628         struct ftrace_profile_stat *stat;
 629         unsigned long long calltime;
 630         struct ftrace_profile *rec;
 631         unsigned long flags;
 632
 633         local_irq_save(flags);
 634         stat = &__get_cpu_var(ftrace_profile_stats);
 635         if (!stat->hash || !ftrace_profile_enabled)
 636                 goto out;
 637
 638         calltime = trace->rettime - trace->calltime;
 639
 640         if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
 641                 int index;
 642
 643                 index = trace->depth;
 644
 645                 /* Append this call time to the parent time to subtract */
 646                 if (index)
 647                         current->ret_stack[index - 1].subtime += calltime;
 648
 649                 if (current->ret_stack[index].subtime < calltime)
 650                         calltime -= current->ret_stack[index].subtime;
 651                 else
 652                         calltime = 0;
 653         }
 654
 655         rec = ftrace_find_profiled_func(stat, trace->func);
 656         if (rec)
 657                 rec->time += calltime;
 658
 659  out:
 660         local_irq_restore(flags);
 661 }
 662
 663 static int register_ftrace_profiler(void)
 664 {
 665         return register_ftrace_graph(&profile_graph_return,
 666                                      &profile_graph_entry);
 667 }
 668
 669 static void unregister_ftrace_profiler(void)
 670 {
 671         unregister_ftrace_graph();
 672 }
 673 #else
 674 static struct ftrace_ops ftrace_profile_ops __read_mostly =
 675 {
 676         .func           = function_profile_call,
 677 };
 678
 679 static int register_ftrace_profiler(void)
 680 {
 681         return register_ftrace_function(&ftrace_profile_ops);
 682 }
 683
 684 static void unregister_ftrace_profiler(void)
 685 {
 686         unregister_ftrace_function(&ftrace_profile_ops);
 687 }
 688 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 689
 690 static ssize_t
 691 ftrace_profile_write(struct file *filp, const char __user *ubuf,
 692                      size_t cnt, loff_t *ppos)
 693 {
 694         unsigned long val;
 695         char buf[64];           /* big enough to hold a number */
 696         int ret;
 697
 698         if (cnt >= sizeof(buf))
 699                 return -EINVAL;
 700
 701         if (copy_from_user(&buf, ubuf, cnt))
 702                 return -EFAULT;
 703
 704         buf[cnt] = 0;
 705
 706         ret = strict_strtoul(buf, 10, &val);
 707         if (ret < 0)
 708                 return ret;
 709
 710         val = !!val;
 711
 712         mutex_lock(&ftrace_profile_lock);
 713         if (ftrace_profile_enabled ^ val) {
 714                 if (val) {
 715                         ret = ftrace_profile_init();
 716                         if (ret < 0) {
 717                                 cnt = ret;
 718                                 goto out;
 719                         }
 720
 721                         ret = register_ftrace_profiler();
 722                         if (ret < 0) {
 723                                 cnt = ret;
 724                                 goto out;
 725                         }
 726                         ftrace_profile_enabled = 1;
 727                 } else {
 728                         ftrace_profile_enabled = 0;
 729                         /*
 730                          * unregister_ftrace_profiler calls stop_machine
 731                          * so this acts like an synchronize_sched.
 732                          */
 733                         unregister_ftrace_profiler();
 734                 }
 735         }
 736  out:
 737         mutex_unlock(&ftrace_profile_lock);
 738
 739         filp->f_pos += cnt;
 740
 741         return cnt;
 742 }
 743
 744 static ssize_t
 745 ftrace_profile_read(struct file *filp, char __user *ubuf,
 746                      size_t cnt, loff_t *ppos)
 747 {
 748         char buf[64];           /* big enough to hold a number */
 749         int r;
 750
 751         r = sprintf(buf, "%u\n", ftrace_profile_enabled);
 752         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 753 }
 754
 755 static const struct file_operations ftrace_profile_fops = {
 756         .open           = tracing_open_generic,
 757         .read           = ftrace_profile_read,
 758         .write          = ftrace_profile_write,
 759 };
 760
 761 /* used to initialize the real stat files */
 762 static struct tracer_stat function_stats __initdata = {
 763         .name           = "functions",
 764         .stat_start     = function_stat_start,
 765         .stat_next      = function_stat_next,
 766         .stat_cmp       = function_stat_cmp,
 767         .stat_headers   = function_stat_headers,
 768         .stat_show      = function_stat_show
 769 };
 770
 771 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 772 {
 773         struct ftrace_profile_stat *stat;
 774         struct dentry *entry;
 775         char *name;
 776         int ret;
 777         int cpu;
 778
 779         for_each_possible_cpu(cpu) {
 780                 stat = &per_cpu(ftrace_profile_stats, cpu);
 781
 782                 /* allocate enough for function name + cpu number */
 783                 name = kmalloc(32, GFP_KERNEL);
 784                 if (!name) {
 785                         /*
 786                          * The files created are permanent, if something happens
 787                          * we still do not free memory.
 788                          */
 789                         WARN(1,
 790                              "Could not allocate stat file for cpu %d\n",
 791                              cpu);
 792                         return;
 793                 }
 794                 stat->stat = function_stats;
 795                 snprintf(name, 32, "function%d", cpu);
 796                 stat->stat.name = name;
 797                 ret = register_stat_tracer(&stat->stat);
 798                 if (ret) {
 799                         WARN(1,
 800                              "Could not register function stat for cpu %d\n",
 801                              cpu);
 802                         kfree(name);
 803                         return;
 804                 }
 805         }
 806
 807         entry = debugfs_create_file("function_profile_enabled", 0644,
 808                                     d_tracer, NULL, &ftrace_profile_fops);
 809         if (!entry)
 810                 pr_warning("Could not create debugfs "
 811                            "'function_profile_enabled' entry\n");
 812 }
 813
 814 #else /* CONFIG_FUNCTION_PROFILER */
 815 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
 816 {
 817 }
 818 #endif /* CONFIG_FUNCTION_PROFILER */
 819
 820 /* set when tracing only a pid */
 821 struct pid *ftrace_pid_trace;
 822 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
 823
 824 #ifdef CONFIG_DYNAMIC_FTRACE
 825
 826 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
 827 # error Dynamic ftrace depends on MCOUNT_RECORD
 828 #endif
 829
 830 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
 831
 832 struct ftrace_func_probe {
 833         struct hlist_node       node;
 834         struct ftrace_probe_ops *ops;
 835         unsigned long           flags;
 836         unsigned long           ip;
 837         void                    *data;
 838         struct rcu_head         rcu;
 839 };
 840
 841 enum {
 842         FTRACE_ENABLE_CALLS             = (1 << 0),
 843         FTRACE_DISABLE_CALLS            = (1 << 1),
 844         FTRACE_UPDATE_TRACE_FUNC        = (1 << 2),
 845         FTRACE_ENABLE_MCOUNT            = (1 << 3),
 846         FTRACE_DISABLE_MCOUNT           = (1 << 4),
 847         FTRACE_START_FUNC_RET           = (1 << 5),
 848         FTRACE_STOP_FUNC_RET            = (1 << 6),
 849 };
 850
 851 static int ftrace_filtered;
 852
 853 static struct dyn_ftrace *ftrace_new_addrs;
 854
 855 static DEFINE_MUTEX(ftrace_regex_lock);
 856
 857 struct ftrace_page {
 858         struct ftrace_page      *next;
 859         int                     index;
 860         struct dyn_ftrace       records[];
 861 };
 862
 863 #define ENTRIES_PER_PAGE \
 864   ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
 865
 866 /* estimate from running different kernels */
 867 #define NR_TO_INIT              10000
 868
 869 static struct ftrace_page       *ftrace_pages_start;
 870 static struct ftrace_page       *ftrace_pages;
 871
 872 static struct dyn_ftrace *ftrace_free_records;
 873
 874 /*
 875  * This is a double for. Do not use 'break' to break out of the loop,
 876  * you must use a goto.
 877  */
 878 #define do_for_each_ftrace_rec(pg, rec)                                 \
 879         for (pg = ftrace_pages_start; pg; pg = pg->next) {              \
 880                 int _____i;                                             \
 881                 for (_____i = 0; _____i < pg->index; _____i++) {        \
 882                         rec = &pg->records[_____i];
 883
 884 #define while_for_each_ftrace_rec()             \
 885                 }                               \
 886         }
 887
 888 #ifdef CONFIG_KPROBES
 889
 890 static int frozen_record_count;
 891
 892 static inline void freeze_record(struct dyn_ftrace *rec)
 893 {
 894         if (!(rec->flags & FTRACE_FL_FROZEN)) {
 895                 rec->flags |= FTRACE_FL_FROZEN;
 896                 frozen_record_count++;
 897         }
 898 }
 899
 900 static inline void unfreeze_record(struct dyn_ftrace *rec)
 901 {
 902         if (rec->flags & FTRACE_FL_FROZEN) {
 903                 rec->flags &= ~FTRACE_FL_FROZEN;
 904                 frozen_record_count--;
 905         }
 906 }
 907
 908 static inline int record_frozen(struct dyn_ftrace *rec)
 909 {
 910         return rec->flags & FTRACE_FL_FROZEN;
 911 }
 912 #else
 913 # define freeze_record(rec)                     ({ 0; })
 914 # define unfreeze_record(rec)                   ({ 0; })
 915 # define record_frozen(rec)                     ({ 0; })
 916 #endif /* CONFIG_KPROBES */
 917
 918 static void ftrace_free_rec(struct dyn_ftrace *rec)
 919 {
 920         rec->freelist = ftrace_free_records;
 921         ftrace_free_records = rec;
 922         rec->flags |= FTRACE_FL_FREE;
 923 }
 924
 925 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
 926 {
 927         struct dyn_ftrace *rec;
 928
 929         /* First check for freed records */
 930         if (ftrace_free_records) {
 931                 rec = ftrace_free_records;
 932
 933                 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
 934                         FTRACE_WARN_ON_ONCE(1);
 935                         ftrace_free_records = NULL;
 936                         return NULL;
 937                 }
 938
 939                 ftrace_free_records = rec->freelist;
 940                 memset(rec, 0, sizeof(*rec));
 941                 return rec;
 942         }
 943
 944         if (ftrace_pages->index == ENTRIES_PER_PAGE) {
 945                 if (!ftrace_pages->next) {
 946                         /* allocate another page */
 947                         ftrace_pages->next =
 948                                 (void *)get_zeroed_page(GFP_KERNEL);
 949                         if (!ftrace_pages->next)
 950                                 return NULL;
 951                 }
 952                 ftrace_pages = ftrace_pages->next;
 953         }
 954
 955         return &ftrace_pages->records[ftrace_pages->index++];
 956 }
 957
 958 static struct dyn_ftrace *
 959 ftrace_record_ip(unsigned long ip)
 960 {
 961         struct dyn_ftrace *rec;
 962
 963         if (ftrace_disabled)
 964                 return NULL;
 965
 966         rec = ftrace_alloc_dyn_node(ip);
 967         if (!rec)
 968                 return NULL;
 969
 970         rec->ip = ip;
 971         rec->newlist = ftrace_new_addrs;
 972         ftrace_new_addrs = rec;
 973
 974         return rec;
 975 }
 976
 977 static void print_ip_ins(const char *fmt, unsigned char *p)
 978 {
 979         int i;
 980
 981         printk(KERN_CONT "%s", fmt);
 982
 983         for (i = 0; i < MCOUNT_INSN_SIZE; i++)
 984                 printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
 985 }
 986
 987 static void ftrace_bug(int failed, unsigned long ip)
 988 {
 989         switch (failed) {
 990         case -EFAULT:
 991                 FTRACE_WARN_ON_ONCE(1);
 992                 pr_info("ftrace faulted on modifying ");
 993                 print_ip_sym(ip);
 994                 break;
 995         case -EINVAL:
 996                 FTRACE_WARN_ON_ONCE(1);
 997                 pr_info("ftrace failed to modify ");
 998                 print_ip_sym(ip);
 999                 print_ip_ins(" actual: ", (unsigned char *)ip);
1000                 printk(KERN_CONT "\n");
1001                 break;
1002         case -EPERM:
1003                 FTRACE_WARN_ON_ONCE(1);
1004                 pr_info("ftrace faulted on writing ");
1005                 print_ip_sym(ip);
1006                 break;
1007         default:
1008                 FTRACE_WARN_ON_ONCE(1);
1009                 pr_info("ftrace faulted on unknown error ");
1010                 print_ip_sym(ip);
1011         }
1012 }
1013
1014
1015 static int
1016 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1017 {
1018         unsigned long ftrace_addr;
1019         unsigned long flag = 0UL;
1020
1021         ftrace_addr = (unsigned long)FTRACE_ADDR;
1022
1023         /*
1024          * If this record is not to be traced or we want to disable it,
1025          * then disable it.
1026          *
1027          * If we want to enable it and filtering is off, then enable it.
1028          *
1029          * If we want to enable it and filtering is on, enable it only if
1030          * it's filtered
1031          */
1032         if (enable && !(rec->flags & FTRACE_FL_NOTRACE)) {
1033                 if (!ftrace_filtered || (rec->flags & FTRACE_FL_FILTER))
1034                         flag = FTRACE_FL_ENABLED;
1035         }
1036
1037         /* If the state of this record hasn't changed, then do nothing */
1038         if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1039                 return 0;
1040
1041         if (flag) {
1042                 rec->flags |= FTRACE_FL_ENABLED;
1043                 return ftrace_make_call(rec, ftrace_addr);
1044         }
1045
1046         rec->flags &= ~FTRACE_FL_ENABLED;
1047         return ftrace_make_nop(NULL, rec, ftrace_addr);
1048 }
1049
1050 static void ftrace_replace_code(int enable)
1051 {
1052         struct dyn_ftrace *rec;
1053         struct ftrace_page *pg;
1054         int failed;
1055
1056         do_for_each_ftrace_rec(pg, rec) {
1057                 /*
1058                  * Skip over free records, records that have
1059                  * failed and not converted.
1060                  */
1061                 if (rec->flags & FTRACE_FL_FREE ||
1062                     rec->flags & FTRACE_FL_FAILED ||
1063                     !(rec->flags & FTRACE_FL_CONVERTED))
1064                         continue;
1065
1066                 /* ignore updates to this record's mcount site */
1067                 if (get_kprobe((void *)rec->ip)) {
1068                         freeze_record(rec);
1069                         continue;
1070                 } else {
1071                         unfreeze_record(rec);
1072                 }
1073
1074                 failed = __ftrace_replace_code(rec, enable);
1075                 if (failed) {
1076                         rec->flags |= FTRACE_FL_FAILED;
1077                         if ((system_state == SYSTEM_BOOTING) ||
1078                             !core_kernel_text(rec->ip)) {
1079                                 ftrace_free_rec(rec);
1080                                 } else {
1081                                 ftrace_bug(failed, rec->ip);
1082                                         /* Stop processing */
1083                                         return;
1084                                 }
1085                 }
1086         } while_for_each_ftrace_rec();
1087 }
1088
1089 static int
1090 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1091 {
1092         unsigned long ip;
1093         int ret;
1094
1095         ip = rec->ip;
1096
1097         ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1098         if (ret) {
1099                 ftrace_bug(ret, ip);
1100                 rec->flags |= FTRACE_FL_FAILED;
1101                 return 0;
1102         }
1103         return 1;
1104 }
1105
1106 /*
1107  * archs can override this function if they must do something
1108  * before the modifying code is performed.
1109  */
1110 int __weak ftrace_arch_code_modify_prepare(void)
1111 {
1112         return 0;
1113 }
1114
1115 /*
1116  * archs can override this function if they must do something
1117  * after the modifying code is performed.
1118  */
1119 int __weak ftrace_arch_code_modify_post_process(void)
1120 {
1121         return 0;
1122 }
1123
1124 static int __ftrace_modify_code(void *data)
1125 {
1126         int *command = data;
1127
1128         if (*command & FTRACE_ENABLE_CALLS)
1129                 ftrace_replace_code(1);
1130         else if (*command & FTRACE_DISABLE_CALLS)
1131                 ftrace_replace_code(0);
1132
1133         if (*command & FTRACE_UPDATE_TRACE_FUNC)
1134                 ftrace_update_ftrace_func(ftrace_trace_function);
1135
1136         if (*command & FTRACE_START_FUNC_RET)
1137                 ftrace_enable_ftrace_graph_caller();
1138         else if (*command & FTRACE_STOP_FUNC_RET)
1139                 ftrace_disable_ftrace_graph_caller();
1140
1141         return 0;
1142 }
1143
1144 static void ftrace_run_update_code(int command)
1145 {
1146         int ret;
1147
1148         ret = ftrace_arch_code_modify_prepare();
1149         FTRACE_WARN_ON(ret);
1150         if (ret)
1151                 return;
1152
1153         stop_machine(__ftrace_modify_code, &command, NULL);
1154
1155         ret = ftrace_arch_code_modify_post_process();
1156         FTRACE_WARN_ON(ret);
1157 }
1158
1159 static ftrace_func_t saved_ftrace_func;
1160 static int ftrace_start_up;
1161
1162 static void ftrace_startup_enable(int command)
1163 {
1164         if (saved_ftrace_func != ftrace_trace_function) {
1165                 saved_ftrace_func = ftrace_trace_function;
1166                 command |= FTRACE_UPDATE_TRACE_FUNC;
1167         }
1168
1169         if (!command || !ftrace_enabled)
1170                 return;
1171
1172         ftrace_run_update_code(command);
1173 }
1174
1175 static void ftrace_startup(int command)
1176 {
1177         if (unlikely(ftrace_disabled))
1178                 return;
1179
1180         ftrace_start_up++;
1181         command |= FTRACE_ENABLE_CALLS;
1182
1183         ftrace_startup_enable(command);
1184 }
1185
1186 static void ftrace_shutdown(int command)
1187 {
1188         if (unlikely(ftrace_disabled))
1189                 return;
1190
1191         ftrace_start_up--;
1192         /*
1193          * Just warn in case of unbalance, no need to kill ftrace, it's not
1194          * critical but the ftrace_call callers may be never nopped again after
1195          * further ftrace uses.
1196          */
1197         WARN_ON_ONCE(ftrace_start_up < 0);
1198
1199         if (!ftrace_start_up)
1200                 command |= FTRACE_DISABLE_CALLS;
1201
1202         if (saved_ftrace_func != ftrace_trace_function) {
1203                 saved_ftrace_func = ftrace_trace_function;
1204                 command |= FTRACE_UPDATE_TRACE_FUNC;
1205         }
1206
1207         if (!command || !ftrace_enabled)
1208                 return;
1209
1210         ftrace_run_update_code(command);
1211 }
1212
1213 static void ftrace_startup_sysctl(void)
1214 {
1215         int command = FTRACE_ENABLE_MCOUNT;
1216
1217         if (unlikely(ftrace_disabled))
1218                 return;
1219
1220         /* Force update next time */
1221         saved_ftrace_func = NULL;
1222         /* ftrace_start_up is true if we want ftrace running */
1223         if (ftrace_start_up)
1224                 command |= FTRACE_ENABLE_CALLS;
1225
1226         ftrace_run_update_code(command);
1227 }
1228
1229 static void ftrace_shutdown_sysctl(void)
1230 {
1231         int command = FTRACE_DISABLE_MCOUNT;
1232
1233         if (unlikely(ftrace_disabled))
1234                 return;
1235
1236         /* ftrace_start_up is true if ftrace is running */
1237         if (ftrace_start_up)
1238                 command |= FTRACE_DISABLE_CALLS;
1239
1240         ftrace_run_update_code(command);
1241 }
1242
1243 static cycle_t          ftrace_update_time;
1244 static unsigned long    ftrace_update_cnt;
1245 unsigned long           ftrace_update_tot_cnt;
1246
1247 static int ftrace_update_code(struct module *mod)
1248 {
1249         struct dyn_ftrace *p;
1250         cycle_t start, stop;
1251
1252         start = ftrace_now(raw_smp_processor_id());
1253         ftrace_update_cnt = 0;
1254
1255         while (ftrace_new_addrs) {
1256
1257                 /* If something went wrong, bail without enabling anything */
1258                 if (unlikely(ftrace_disabled))
1259                         return -1;
1260
1261                 p = ftrace_new_addrs;
1262                 ftrace_new_addrs = p->newlist;
1263                 p->flags = 0L;
1264
1265                 /* convert record (i.e, patch mcount-call with NOP) */
1266                 if (ftrace_code_disable(mod, p)) {
1267                         p->flags |= FTRACE_FL_CONVERTED;
1268                         ftrace_update_cnt++;
1269                 } else
1270                         ftrace_free_rec(p);
1271         }
1272
1273         stop = ftrace_now(raw_smp_processor_id());
1274         ftrace_update_time = stop - start;
1275         ftrace_update_tot_cnt += ftrace_update_cnt;
1276
1277         return 0;
1278 }
1279
1280 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
1281 {
1282         struct ftrace_page *pg;
1283         int cnt;
1284         int i;
1285
1286         /* allocate a few pages */
1287         ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
1288         if (!ftrace_pages_start)
1289                 return -1;
1290
1291         /*
1292          * Allocate a few more pages.
1293          *
1294          * TODO: have some parser search vmlinux before
1295          *   final linking to find all calls to ftrace.
1296          *   Then we can:
1297          *    a) know how many pages to allocate.
1298          *     and/or
1299          *    b) set up the table then.
1300          *
1301          *  The dynamic code is still necessary for
1302          *  modules.
1303          */
1304
1305         pg = ftrace_pages = ftrace_pages_start;
1306
1307         cnt = num_to_init / ENTRIES_PER_PAGE;
1308         pr_info("ftrace: allocating %ld entries in %d pages\n",
1309                 num_to_init, cnt + 1);
1310
1311         for (i = 0; i < cnt; i++) {
1312                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
1313
1314                 /* If we fail, we'll try later anyway */
1315                 if (!pg->next)
1316                         break;
1317
1318                 pg = pg->next;
1319         }
1320
1321         return 0;
1322 }
1323
1324 enum {
1325         FTRACE_ITER_FILTER      = (1 << 0),
1326         FTRACE_ITER_NOTRACE     = (1 << 1),
1327         FTRACE_ITER_FAILURES    = (1 << 2),
1328         FTRACE_ITER_PRINTALL    = (1 << 3),
1329         FTRACE_ITER_HASH        = (1 << 4),
1330 };
1331
1332 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1333
1334 struct ftrace_iterator {
1335         struct ftrace_page      *pg;
1336         int                     hidx;
1337         int                     idx;
1338         unsigned                flags;
1339         struct trace_parser     parser;
1340 };
1341
1342 static void *
1343 t_hash_next(struct seq_file *m, void *v, loff_t *pos)
1344 {
1345         struct ftrace_iterator *iter = m->private;
1346         struct hlist_node *hnd = v;
1347         struct hlist_head *hhd;
1348
1349         WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
1350
1351         (*pos)++;
1352
1353  retry:
1354         if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1355                 return NULL;
1356
1357         hhd = &ftrace_func_hash[iter->hidx];
1358
1359         if (hlist_empty(hhd)) {
1360                 iter->hidx++;
1361                 hnd = NULL;
1362                 goto retry;
1363         }
1364
1365         if (!hnd)
1366                 hnd = hhd->first;
1367         else {
1368                 hnd = hnd->next;
1369                 if (!hnd) {
1370                         iter->hidx++;
1371                         goto retry;
1372                 }
1373         }
1374
1375         return hnd;
1376 }
1377
1378 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1379 {
1380         struct ftrace_iterator *iter = m->private;
1381         void *p = NULL;
1382         loff_t l;
1383
1384         if (!(iter->flags & FTRACE_ITER_HASH))
1385                 *pos = 0;
1386
1387         iter->flags |= FTRACE_ITER_HASH;
1388
1389         iter->hidx = 0;
1390         for (l = 0; l <= *pos; ) {
1391                 p = t_hash_next(m, p, &l);
1392                 if (!p)
1393                         break;
1394         }
1395         return p;
1396 }
1397
1398 static int t_hash_show(struct seq_file *m, void *v)
1399 {
1400         struct ftrace_func_probe *rec;
1401         struct hlist_node *hnd = v;
1402
1403         rec = hlist_entry(hnd, struct ftrace_func_probe, node);
1404
1405         if (rec->ops->print)
1406                 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1407
1408         seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
1409
1410         if (rec->data)
1411                 seq_printf(m, ":%p", rec->data);
1412         seq_putc(m, '\n');
1413
1414         return 0;
1415 }
1416
1417 static void *
1418 t_next(struct seq_file *m, void *v, loff_t *pos)
1419 {
1420         struct ftrace_iterator *iter = m->private;
1421         struct dyn_ftrace *rec = NULL;
1422
1423         if (iter->flags & FTRACE_ITER_HASH)
1424                 return t_hash_next(m, v, pos);
1425
1426         (*pos)++;
1427
1428         if (iter->flags & FTRACE_ITER_PRINTALL)
1429                 return NULL;
1430
1431  retry:
1432         if (iter->idx >= iter->pg->index) {
1433                 if (iter->pg->next) {
1434                         iter->pg = iter->pg->next;
1435                         iter->idx = 0;
1436                         goto retry;
1437                 }
1438         } else {
1439                 rec = &iter->pg->records[iter->idx++];
1440                 if ((rec->flags & FTRACE_FL_FREE) ||
1441
1442                     (!(iter->flags & FTRACE_ITER_FAILURES) &&
1443                      (rec->flags & FTRACE_FL_FAILED)) ||
1444
1445                     ((iter->flags & FTRACE_ITER_FAILURES) &&
1446                      !(rec->flags & FTRACE_FL_FAILED)) ||
1447
1448                     ((iter->flags & FTRACE_ITER_FILTER) &&
1449                      !(rec->flags & FTRACE_FL_FILTER)) ||
1450
1451                     ((iter->flags & FTRACE_ITER_NOTRACE) &&
1452                      !(rec->flags & FTRACE_FL_NOTRACE))) {
1453                         rec = NULL;
1454                         goto retry;
1455                 }
1456         }
1457
1458         return rec;
1459 }
1460
1461 static void *t_start(struct seq_file *m, loff_t *pos)
1462 {
1463         struct ftrace_iterator *iter = m->private;
1464         void *p = NULL;
1465         loff_t l;
1466
1467         mutex_lock(&ftrace_lock);
1468         /*
1469          * For set_ftrace_filter reading, if we have the filter
1470          * off, we can short cut and just print out that all
1471          * functions are enabled.
1472          */
1473         if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1474                 if (*pos > 0)
1475                         return t_hash_start(m, pos);
1476                 iter->flags |= FTRACE_ITER_PRINTALL;
1477                 return iter;
1478         }
1479
1480         if (iter->flags & FTRACE_ITER_HASH)
1481                 return t_hash_start(m, pos);
1482
1483         iter->pg = ftrace_pages_start;
1484         iter->idx = 0;
1485         for (l = 0; l <= *pos; ) {
1486                 p = t_next(m, p, &l);
1487                 if (!p)
1488                         break;
1489         }
1490
1491         if (!p && iter->flags & FTRACE_ITER_FILTER)
1492                 return t_hash_start(m, pos);
1493
1494         return p;
1495 }
1496
1497 static void t_stop(struct seq_file *m, void *p)
1498 {
1499         mutex_unlock(&ftrace_lock);
1500 }
1501
1502 static int t_show(struct seq_file *m, void *v)
1503 {
1504         struct ftrace_iterator *iter = m->private;
1505         struct dyn_ftrace *rec = v;
1506
1507         if (iter->flags & FTRACE_ITER_HASH)
1508                 return t_hash_show(m, v);
1509
1510         if (iter->flags & FTRACE_ITER_PRINTALL) {
1511                 seq_printf(m, "#### all functions enabled ####\n");
1512                 return 0;
1513         }
1514
1515         if (!rec)
1516                 return 0;
1517
1518         seq_printf(m, "%ps\n", (void *)rec->ip);
1519
1520         return 0;
1521 }
1522
1523 static struct seq_operations show_ftrace_seq_ops = {
1524         .start = t_start,
1525         .next = t_next,
1526         .stop = t_stop,
1527         .show = t_show,
1528 };
1529
1530 static int
1531 ftrace_avail_open(struct inode *inode, struct file *file)
1532 {
1533         struct ftrace_iterator *iter;
1534         int ret;
1535
1536         if (unlikely(ftrace_disabled))
1537                 return -ENODEV;
1538
1539         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1540         if (!iter)
1541                 return -ENOMEM;
1542
1543         iter->pg = ftrace_pages_start;
1544
1545         ret = seq_open(file, &show_ftrace_seq_ops);
1546         if (!ret) {
1547                 struct seq_file *m = file->private_data;
1548
1549                 m->private = iter;
1550         } else {
1551                 kfree(iter);
1552         }
1553
1554         return ret;
1555 }
1556
1557 static int
1558 ftrace_failures_open(struct inode *inode, struct file *file)
1559 {
1560         int ret;
1561         struct seq_file *m;
1562         struct ftrace_iterator *iter;
1563
1564         ret = ftrace_avail_open(inode, file);
1565         if (!ret) {
1566                 m = (struct seq_file *)file->private_data;
1567                 iter = (struct ftrace_iterator *)m->private;
1568                 iter->flags = FTRACE_ITER_FAILURES;
1569         }
1570
1571         return ret;
1572 }
1573
1574
1575 static void ftrace_filter_reset(int enable)
1576 {
1577         struct ftrace_page *pg;
1578         struct dyn_ftrace *rec;
1579         unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1580
1581         mutex_lock(&ftrace_lock);
1582         if (enable)
1583                 ftrace_filtered = 0;
1584         do_for_each_ftrace_rec(pg, rec) {
1585                 if (rec->flags & FTRACE_FL_FAILED)
1586                         continue;
1587                 rec->flags &= ~type;
1588         } while_for_each_ftrace_rec();
1589         mutex_unlock(&ftrace_lock);
1590 }
1591
1592 static int
1593 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1594 {
1595         struct ftrace_iterator *iter;
1596         int ret = 0;
1597
1598         if (unlikely(ftrace_disabled))
1599                 return -ENODEV;
1600
1601         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1602         if (!iter)
1603                 return -ENOMEM;
1604
1605         if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
1606                 kfree(iter);
1607                 return -ENOMEM;
1608         }
1609
1610         mutex_lock(&ftrace_regex_lock);
1611         if ((file->f_mode & FMODE_WRITE) &&
1612             (file->f_flags & O_TRUNC))
1613                 ftrace_filter_reset(enable);
1614
1615         if (file->f_mode & FMODE_READ) {
1616                 iter->pg = ftrace_pages_start;
1617                 iter->flags = enable ? FTRACE_ITER_FILTER :
1618                         FTRACE_ITER_NOTRACE;
1619
1620                 ret = seq_open(file, &show_ftrace_seq_ops);
1621                 if (!ret) {
1622                         struct seq_file *m = file->private_data;
1623                         m->private = iter;
1624                 } else
1625                         kfree(iter);
1626         } else
1627                 file->private_data = iter;
1628         mutex_unlock(&ftrace_regex_lock);
1629
1630         return ret;
1631 }
1632
1633 static int
1634 ftrace_filter_open(struct inode *inode, struct file *file)
1635 {
1636         return ftrace_regex_open(inode, file, 1);
1637 }
1638
1639 static int
1640 ftrace_notrace_open(struct inode *inode, struct file *file)
1641 {
1642         return ftrace_regex_open(inode, file, 0);
1643 }
1644
1645 static loff_t
1646 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1647 {
1648         loff_t ret;
1649
1650         if (file->f_mode & FMODE_READ)
1651                 ret = seq_lseek(file, offset, origin);
1652         else
1653                 file->f_pos = ret = 1;
1654
1655         return ret;
1656 }
1657
1658 static int ftrace_match(char *str, char *regex, int len, int type)
1659 {
1660         int matched = 0;
1661         char *ptr;
1662
1663         switch (type) {
1664         case MATCH_FULL:
1665                 if (strcmp(str, regex) == 0)
1666                         matched = 1;
1667                 break;
1668         case MATCH_FRONT_ONLY:
1669                 if (strncmp(str, regex, len) == 0)
1670                         matched = 1;
1671                 break;
1672         case MATCH_MIDDLE_ONLY:
1673                 if (strstr(str, regex))
1674                         matched = 1;
1675                 break;
1676         case MATCH_END_ONLY:
1677                 ptr = strstr(str, regex);
1678                 if (ptr && (ptr[len] == 0))
1679                         matched = 1;
1680                 break;
1681         }
1682
1683         return matched;
1684 }
1685
1686 static int
1687 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1688 {
1689         char str[KSYM_SYMBOL_LEN];
1690
1691         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1692         return ftrace_match(str, regex, len, type);
1693 }
1694
1695 static void ftrace_match_records(char *buff, int len, int enable)
1696 {
1697         unsigned int search_len;
1698         struct ftrace_page *pg;
1699         struct dyn_ftrace *rec;
1700         unsigned long flag;
1701         char *search;
1702         int type;
1703         int not;
1704
1705         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1706         type = filter_parse_regex(buff, len, &search, &not);
1707
1708         search_len = strlen(search);
1709
1710         mutex_lock(&ftrace_lock);
1711         do_for_each_ftrace_rec(pg, rec) {
1712
1713                 if (rec->flags & FTRACE_FL_FAILED)
1714                         continue;
1715
1716                 if (ftrace_match_record(rec, search, search_len, type)) {
1717                         if (not)
1718                                 rec->flags &= ~flag;
1719                         else
1720                                 rec->flags |= flag;
1721                 }
1722                 /*
1723                  * Only enable filtering if we have a function that
1724                  * is filtered on.
1725                  */
1726                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1727                         ftrace_filtered = 1;
1728         } while_for_each_ftrace_rec();
1729         mutex_unlock(&ftrace_lock);
1730 }
1731
1732 static int
1733 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1734                            char *regex, int len, int type)
1735 {
1736         char str[KSYM_SYMBOL_LEN];
1737         char *modname;
1738
1739         kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1740
1741         if (!modname || strcmp(modname, mod))
1742                 return 0;
1743
1744         /* blank search means to match all funcs in the mod */
1745         if (len)
1746                 return ftrace_match(str, regex, len, type);
1747         else
1748                 return 1;
1749 }
1750
1751 static void ftrace_match_module_records(char *buff, char *mod, int enable)
1752 {
1753         unsigned search_len = 0;
1754         struct ftrace_page *pg;
1755         struct dyn_ftrace *rec;
1756         int type = MATCH_FULL;
1757         char *search = buff;
1758         unsigned long flag;
1759         int not = 0;
1760
1761         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1762
1763         /* blank or '*' mean the same */
1764         if (strcmp(buff, "*") == 0)
1765                 buff[0] = 0;
1766
1767         /* handle the case of 'dont filter this module' */
1768         if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1769                 buff[0] = 0;
1770                 not = 1;
1771         }
1772
1773         if (strlen(buff)) {
1774                 type = filter_parse_regex(buff, strlen(buff), &search, &not);
1775                 search_len = strlen(search);
1776         }
1777
1778         mutex_lock(&ftrace_lock);
1779         do_for_each_ftrace_rec(pg, rec) {
1780
1781                 if (rec->flags & FTRACE_FL_FAILED)
1782                         continue;
1783
1784                 if (ftrace_match_module_record(rec, mod,
1785                                                search, search_len, type)) {
1786                         if (not)
1787                                 rec->flags &= ~flag;
1788                         else
1789                                 rec->flags |= flag;
1790                 }
1791                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1792                         ftrace_filtered = 1;
1793
1794         } while_for_each_ftrace_rec();
1795         mutex_unlock(&ftrace_lock);
1796 }
1797
1798 /*
1799  * We register the module command as a template to show others how
1800  * to register the a command as well.
1801  */
1802
1803 static int
1804 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1805 {
1806         char *mod;
1807
1808         /*
1809          * cmd == 'mod' because we only registered this func
1810          * for the 'mod' ftrace_func_command.
1811          * But if you register one func with multiple commands,
1812          * you can tell which command was used by the cmd
1813          * parameter.
1814          */
1815
1816         /* we must have a module name */
1817         if (!param)
1818                 return -EINVAL;
1819
1820         mod = strsep(&param, ":");
1821         if (!strlen(mod))
1822                 return -EINVAL;
1823
1824         ftrace_match_module_records(func, mod, enable);
1825         return 0;
1826 }
1827
1828 static struct ftrace_func_command ftrace_mod_cmd = {
1829         .name                   = "mod",
1830         .func                   = ftrace_mod_callback,
1831 };
1832
1833 static int __init ftrace_mod_cmd_init(void)
1834 {
1835         return register_ftrace_command(&ftrace_mod_cmd);
1836 }
1837 device_initcall(ftrace_mod_cmd_init);
1838
1839 static void
1840 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1841 {
1842         struct ftrace_func_probe *entry;
1843         struct hlist_head *hhd;
1844         struct hlist_node *n;
1845         unsigned long key;
1846         int resched;
1847
1848         key = hash_long(ip, FTRACE_HASH_BITS);
1849
1850         hhd = &ftrace_func_hash[key];
1851
1852         if (hlist_empty(hhd))
1853                 return;
1854
1855         /*
1856          * Disable preemption for these calls to prevent a RCU grace
1857          * period. This syncs the hash iteration and freeing of items
1858          * on the hash. rcu_read_lock is too dangerous here.
1859          */
1860         resched = ftrace_preempt_disable();
1861         hlist_for_each_entry_rcu(entry, n, hhd, node) {
1862                 if (entry->ip == ip)
1863                         entry->ops->func(ip, parent_ip, &entry->data);
1864         }
1865         ftrace_preempt_enable(resched);
1866 }
1867
1868 static struct ftrace_ops trace_probe_ops __read_mostly =
1869 {
1870         .func           = function_trace_probe_call,
1871 };
1872
1873 static int ftrace_probe_registered;
1874
1875 static void __enable_ftrace_function_probe(void)
1876 {
1877         int i;
1878
1879         if (ftrace_probe_registered)
1880                 return;
1881
1882         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1883                 struct hlist_head *hhd = &ftrace_func_hash[i];
1884                 if (hhd->first)
1885                         break;
1886         }
1887         /* Nothing registered? */
1888         if (i == FTRACE_FUNC_HASHSIZE)
1889                 return;
1890
1891         __register_ftrace_function(&trace_probe_ops);
1892         ftrace_startup(0);
1893         ftrace_probe_registered = 1;
1894 }
1895
1896 static void __disable_ftrace_function_probe(void)
1897 {
1898         int i;
1899
1900         if (!ftrace_probe_registered)
1901                 return;
1902
1903         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1904                 struct hlist_head *hhd = &ftrace_func_hash[i];
1905                 if (hhd->first)
1906                         return;
1907         }
1908
1909         /* no more funcs left */
1910         __unregister_ftrace_function(&trace_probe_ops);
1911         ftrace_shutdown(0);
1912         ftrace_probe_registered = 0;
1913 }
1914
1915
1916 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
1917 {
1918         struct ftrace_func_probe *entry =
1919                 container_of(rhp, struct ftrace_func_probe, rcu);
1920
1921         if (entry->ops->free)
1922                 entry->ops->free(&entry->data);
1923         kfree(entry);
1924 }
1925
1926
1927 int
1928 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
1929                               void *data)
1930 {
1931         struct ftrace_func_probe *entry;
1932         struct ftrace_page *pg;
1933         struct dyn_ftrace *rec;
1934         int type, len, not;
1935         unsigned long key;
1936         int count = 0;
1937         char *search;
1938
1939         type = filter_parse_regex(glob, strlen(glob), &search, &not);
1940         len = strlen(search);
1941
1942         /* we do not support '!' for function probes */
1943         if (WARN_ON(not))
1944                 return -EINVAL;
1945
1946         mutex_lock(&ftrace_lock);
1947         do_for_each_ftrace_rec(pg, rec) {
1948
1949                 if (rec->flags & FTRACE_FL_FAILED)
1950                         continue;
1951
1952                 if (!ftrace_match_record(rec, search, len, type))
1953                         continue;
1954
1955                 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1956                 if (!entry) {
1957                         /* If we did not process any, then return error */
1958                         if (!count)
1959                                 count = -ENOMEM;
1960                         goto out_unlock;
1961                 }
1962
1963                 count++;
1964
1965                 entry->data = data;
1966
1967                 /*
1968                  * The caller might want to do something special
1969                  * for each function we find. We call the callback
1970                  * to give the caller an opportunity to do so.
1971                  */
1972                 if (ops->callback) {
1973                         if (ops->callback(rec->ip, &entry->data) < 0) {
1974                                 /* caller does not like this func */
1975                                 kfree(entry);
1976                                 continue;
1977                         }
1978                 }
1979
1980                 entry->ops = ops;
1981                 entry->ip = rec->ip;
1982
1983                 key = hash_long(entry->ip, FTRACE_HASH_BITS);
1984                 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
1985
1986         } while_for_each_ftrace_rec();
1987         __enable_ftrace_function_probe();
1988
1989  out_unlock:
1990         mutex_unlock(&ftrace_lock);
1991
1992         return count;
1993 }
1994
1995 enum {
1996         PROBE_TEST_FUNC         = 1,
1997         PROBE_TEST_DATA         = 2
1998 };
1999
2000 static void
2001 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2002                                   void *data, int flags)
2003 {
2004         struct ftrace_func_probe *entry;
2005         struct hlist_node *n, *tmp;
2006         char str[KSYM_SYMBOL_LEN];
2007         int type = MATCH_FULL;
2008         int i, len = 0;
2009         char *search;
2010
2011         if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
2012                 glob = NULL;
2013         else if (glob) {
2014                 int not;
2015
2016                 type = filter_parse_regex(glob, strlen(glob), &search, &not);
2017                 len = strlen(search);
2018
2019                 /* we do not support '!' for function probes */
2020                 if (WARN_ON(not))
2021                         return;
2022         }
2023
2024         mutex_lock(&ftrace_lock);
2025         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2026                 struct hlist_head *hhd = &ftrace_func_hash[i];
2027
2028                 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2029
2030                         /* break up if statements for readability */
2031                         if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2032                                 continue;
2033
2034                         if ((flags & PROBE_TEST_DATA) && entry->data != data)
2035                                 continue;
2036
2037                         /* do this last, since it is the most expensive */
2038                         if (glob) {
2039                                 kallsyms_lookup(entry->ip, NULL, NULL,
2040                                                 NULL, str);
2041                                 if (!ftrace_match(str, glob, len, type))
2042                                         continue;
2043                         }
2044
2045                         hlist_del(&entry->node);
2046                         call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2047                 }
2048         }
2049         __disable_ftrace_function_probe();
2050         mutex_unlock(&ftrace_lock);
2051 }
2052
2053 void
2054 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2055                                 void *data)
2056 {
2057         __unregister_ftrace_function_probe(glob, ops, data,
2058                                           PROBE_TEST_FUNC | PROBE_TEST_DATA);
2059 }
2060
2061 void
2062 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2063 {
2064         __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2065 }
2066
2067 void unregister_ftrace_function_probe_all(char *glob)
2068 {
2069         __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2070 }
2071
2072 static LIST_HEAD(ftrace_commands);
2073 static DEFINE_MUTEX(ftrace_cmd_mutex);
2074
2075 int register_ftrace_command(struct ftrace_func_command *cmd)
2076 {
2077         struct ftrace_func_command *p;
2078         int ret = 0;
2079
2080         mutex_lock(&ftrace_cmd_mutex);
2081         list_for_each_entry(p, &ftrace_commands, list) {
2082                 if (strcmp(cmd->name, p->name) == 0) {
2083                         ret = -EBUSY;
2084                         goto out_unlock;
2085                 }
2086         }
2087         list_add(&cmd->list, &ftrace_commands);
2088  out_unlock:
2089         mutex_unlock(&ftrace_cmd_mutex);
2090
2091         return ret;
2092 }
2093
2094 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2095 {
2096         struct ftrace_func_command *p, *n;
2097         int ret = -ENODEV;
2098
2099         mutex_lock(&ftrace_cmd_mutex);
2100         list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2101                 if (strcmp(cmd->name, p->name) == 0) {
2102                         ret = 0;
2103                         list_del_init(&p->list);
2104                         goto out_unlock;
2105                 }
2106         }
2107  out_unlock:
2108         mutex_unlock(&ftrace_cmd_mutex);
2109
2110         return ret;
2111 }
2112
2113 static int ftrace_process_regex(char *buff, int len, int enable)
2114 {
2115         char *func, *command, *next = buff;
2116         struct ftrace_func_command *p;
2117         int ret = -EINVAL;
2118
2119         func = strsep(&next, ":");
2120
2121         if (!next) {
2122                 ftrace_match_records(func, len, enable);
2123                 return 0;
2124         }
2125
2126         /* command found */
2127
2128         command = strsep(&next, ":");
2129
2130         mutex_lock(&ftrace_cmd_mutex);
2131         list_for_each_entry(p, &ftrace_commands, list) {
2132                 if (strcmp(p->name, command) == 0) {
2133                         ret = p->func(func, command, next, enable);
2134                         goto out_unlock;
2135                 }
2136         }
2137  out_unlock:
2138         mutex_unlock(&ftrace_cmd_mutex);
2139
2140         return ret;
2141 }
2142
2143 static ssize_t
2144 ftrace_regex_write(struct file *file, const char __user *ubuf,
2145                    size_t cnt, loff_t *ppos, int enable)
2146 {
2147         struct ftrace_iterator *iter;
2148         struct trace_parser *parser;
2149         ssize_t ret, read;
2150
2151         if (!cnt || cnt < 0)
2152                 return 0;
2153
2154         mutex_lock(&ftrace_regex_lock);
2155
2156         if (file->f_mode & FMODE_READ) {
2157                 struct seq_file *m = file->private_data;
2158                 iter = m->private;
2159         } else
2160                 iter = file->private_data;
2161
2162         parser = &iter->parser;
2163         read = trace_get_user(parser, ubuf, cnt, ppos);
2164
2165         if (trace_parser_loaded(parser) &&
2166             !trace_parser_cont(parser)) {
2167                 ret = ftrace_process_regex(parser->buffer,
2168                                            parser->idx, enable);
2169                 if (ret)
2170                         goto out;
2171
2172                 trace_parser_clear(parser);
2173         }
2174
2175         ret = read;
2176
2177         mutex_unlock(&ftrace_regex_lock);
2178 out:
2179         return ret;
2180 }
2181
2182 static ssize_t
2183 ftrace_filter_write(struct file *file, const char __user *ubuf,
2184                     size_t cnt, loff_t *ppos)
2185 {
2186         return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2187 }
2188
2189 static ssize_t
2190 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2191                      size_t cnt, loff_t *ppos)
2192 {
2193         return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2194 }
2195
2196 static void
2197 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2198 {
2199         if (unlikely(ftrace_disabled))
2200                 return;
2201
2202         mutex_lock(&ftrace_regex_lock);
2203         if (reset)
2204                 ftrace_filter_reset(enable);
2205         if (buf)
2206                 ftrace_match_records(buf, len, enable);
2207         mutex_unlock(&ftrace_regex_lock);
2208 }
2209
2210 /**
2211  * ftrace_set_filter - set a function to filter on in ftrace
2212  * @buf - the string that holds the function filter text.
2213  * @len - the length of the string.
2214  * @reset - non zero to reset all filters before applying this filter.
2215  *
2216  * Filters denote which functions should be enabled when tracing is enabled.
2217  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2218  */
2219 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2220 {
2221         ftrace_set_regex(buf, len, reset, 1);
2222 }
2223
2224 /**
2225  * ftrace_set_notrace - set a function to not trace in ftrace
2226  * @buf - the string that holds the function notrace text.
2227  * @len - the length of the string.
2228  * @reset - non zero to reset all filters before applying this filter.
2229  *
2230  * Notrace Filters denote which functions should not be enabled when tracing
2231  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2232  * for tracing.
2233  */
2234 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2235 {
2236         ftrace_set_regex(buf, len, reset, 0);
2237 }
2238
2239 /*
2240  * command line interface to allow users to set filters on boot up.
2241  */
2242 #define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
2243 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2244 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2245
2246 static int __init set_ftrace_notrace(char *str)
2247 {
2248         strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2249         return 1;
2250 }
2251 __setup("ftrace_notrace=", set_ftrace_notrace);
2252
2253 static int __init set_ftrace_filter(char *str)
2254 {
2255         strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2256         return 1;
2257 }
2258 __setup("ftrace_filter=", set_ftrace_filter);
2259
2260 static void __init set_ftrace_early_filter(char *buf, int enable)
2261 {
2262         char *func;
2263
2264         while (buf) {
2265                 func = strsep(&buf, ",");
2266                 ftrace_set_regex(func, strlen(func), 0, enable);
2267         }
2268 }
2269
2270 static void __init set_ftrace_early_filters(void)
2271 {
2272         if (ftrace_filter_buf[0])
2273                 set_ftrace_early_filter(ftrace_filter_buf, 1);
2274         if (ftrace_notrace_buf[0])
2275                 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2276 }
2277
2278 static int
2279 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2280 {
2281         struct seq_file *m = (struct seq_file *)file->private_data;
2282         struct ftrace_iterator *iter;
2283         struct trace_parser *parser;
2284
2285         mutex_lock(&ftrace_regex_lock);
2286         if (file->f_mode & FMODE_READ) {
2287                 iter = m->private;
2288
2289                 seq_release(inode, file);
2290         } else
2291                 iter = file->private_data;
2292
2293         parser = &iter->parser;
2294         if (trace_parser_loaded(parser)) {
2295                 parser->buffer[parser->idx] = 0;
2296                 ftrace_match_records(parser->buffer, parser->idx, enable);
2297         }
2298
2299         mutex_lock(&ftrace_lock);
2300         if (ftrace_start_up && ftrace_enabled)
2301                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2302         mutex_unlock(&ftrace_lock);
2303
2304         trace_parser_put(parser);
2305         kfree(iter);
2306
2307         mutex_unlock(&ftrace_regex_lock);
2308         return 0;
2309 }
2310
2311 static int
2312 ftrace_filter_release(struct inode *inode, struct file *file)
2313 {
2314         return ftrace_regex_release(inode, file, 1);
2315 }
2316
2317 static int
2318 ftrace_notrace_release(struct inode *inode, struct file *file)
2319 {
2320         return ftrace_regex_release(inode, file, 0);
2321 }
2322
2323 static const struct file_operations ftrace_avail_fops = {
2324         .open = ftrace_avail_open,
2325         .read = seq_read,
2326         .llseek = seq_lseek,
2327         .release = seq_release_private,
2328 };
2329
2330 static const struct file_operations ftrace_failures_fops = {
2331         .open = ftrace_failures_open,
2332         .read = seq_read,
2333         .llseek = seq_lseek,
2334         .release = seq_release_private,
2335 };
2336
2337 static const struct file_operations ftrace_filter_fops = {
2338         .open = ftrace_filter_open,
2339         .read = seq_read,
2340         .write = ftrace_filter_write,
2341         .llseek = ftrace_regex_lseek,
2342         .release = ftrace_filter_release,
2343 };
2344
2345 static const struct file_operations ftrace_notrace_fops = {
2346         .open = ftrace_notrace_open,
2347         .read = seq_read,
2348         .write = ftrace_notrace_write,
2349         .llseek = ftrace_regex_lseek,
2350         .release = ftrace_notrace_release,
2351 };
2352
2353 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2354
2355 static DEFINE_MUTEX(graph_lock);
2356
2357 int ftrace_graph_count;
2358 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2359
2360 static void *
2361 __g_next(struct seq_file *m, loff_t *pos)
2362 {
2363         unsigned long *array = m->private;
2364
2365         if (*pos >= ftrace_graph_count)
2366                 return NULL;
2367         return &array[*pos];
2368 }
2369
2370 static void *
2371 g_next(struct seq_file *m, void *v, loff_t *pos)
2372 {
2373         (*pos)++;
2374         return __g_next(m, pos);
2375 }
2376
2377 static void *g_start(struct seq_file *m, loff_t *pos)
2378 {
2379         mutex_lock(&graph_lock);
2380
2381         /* Nothing, tell g_show to print all functions are enabled */
2382         if (!ftrace_graph_count && !*pos)
2383                 return (void *)1;
2384
2385         return __g_next(m, pos);
2386 }
2387
2388 static void g_stop(struct seq_file *m, void *p)
2389 {
2390         mutex_unlock(&graph_lock);
2391 }
2392
2393 static int g_show(struct seq_file *m, void *v)
2394 {
2395         unsigned long *ptr = v;
2396
2397         if (!ptr)
2398                 return 0;
2399
2400         if (ptr == (unsigned long *)1) {
2401                 seq_printf(m, "#### all functions enabled ####\n");
2402                 return 0;
2403         }
2404
2405         seq_printf(m, "%ps\n", (void *)*ptr);
2406
2407         return 0;
2408 }
2409
2410 static struct seq_operations ftrace_graph_seq_ops = {
2411         .start = g_start,
2412         .next = g_next,
2413         .stop = g_stop,
2414         .show = g_show,
2415 };
2416
2417 static int
2418 ftrace_graph_open(struct inode *inode, struct file *file)
2419 {
2420         int ret = 0;
2421
2422         if (unlikely(ftrace_disabled))
2423                 return -ENODEV;
2424
2425         mutex_lock(&graph_lock);
2426         if ((file->f_mode & FMODE_WRITE) &&
2427             (file->f_flags & O_TRUNC)) {
2428                 ftrace_graph_count = 0;
2429                 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2430         }
2431
2432         if (file->f_mode & FMODE_READ) {
2433                 ret = seq_open(file, &ftrace_graph_seq_ops);
2434                 if (!ret) {
2435                         struct seq_file *m = file->private_data;
2436                         m->private = ftrace_graph_funcs;
2437                 }
2438         } else
2439                 file->private_data = ftrace_graph_funcs;
2440         mutex_unlock(&graph_lock);
2441
2442         return ret;
2443 }
2444
2445 static int
2446 ftrace_graph_release(struct inode *inode, struct file *file)
2447 {
2448         if (file->f_mode & FMODE_READ)
2449                 seq_release(inode, file);
2450         return 0;
2451 }
2452
2453 static int
2454 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2455 {
2456         struct dyn_ftrace *rec;
2457         struct ftrace_page *pg;
2458         int search_len;
2459         int found = 0;
2460         int type, not;
2461         char *search;
2462         bool exists;
2463         int i;
2464
2465         if (ftrace_disabled)
2466                 return -ENODEV;
2467
2468         /* decode regex */
2469         type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
2470         if (not)
2471                 return -EINVAL;
2472
2473         search_len = strlen(search);
2474
2475         mutex_lock(&ftrace_lock);
2476         do_for_each_ftrace_rec(pg, rec) {
2477
2478                 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2479                         break;
2480
2481                 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2482                         continue;
2483
2484                 if (ftrace_match_record(rec, search, search_len, type)) {
2485                         /* ensure it is not already in the array */
2486                         exists = false;
2487                         for (i = 0; i < *idx; i++)
2488                                 if (array[i] == rec->ip) {
2489                                         exists = true;
2490                                         break;
2491                                 }
2492                         if (!exists) {
2493                                 array[(*idx)++] = rec->ip;
2494                                 found = 1;
2495                         }
2496                 }
2497         } while_for_each_ftrace_rec();
2498
2499         mutex_unlock(&ftrace_lock);
2500
2501         return found ? 0 : -EINVAL;
2502 }
2503
2504 static ssize_t
2505 ftrace_graph_write(struct file *file, const char __user *ubuf,
2506                    size_t cnt, loff_t *ppos)
2507 {
2508         struct trace_parser parser;
2509         unsigned long *array;
2510         size_t read = 0;
2511         ssize_t ret;
2512
2513         if (!cnt || cnt < 0)
2514                 return 0;
2515
2516         mutex_lock(&graph_lock);
2517
2518         if (ftrace_graph_count >= FTRACE_GRAPH_MAX_FUNCS) {
2519                 ret = -EBUSY;
2520                 goto out;
2521         }
2522
2523         if (file->f_mode & FMODE_READ) {
2524                 struct seq_file *m = file->private_data;
2525                 array = m->private;
2526         } else
2527                 array = file->private_data;
2528
2529         if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
2530                 ret = -ENOMEM;
2531                 goto out;
2532         }
2533
2534         read = trace_get_user(&parser, ubuf, cnt, ppos);
2535
2536         if (trace_parser_loaded((&parser))) {
2537                 parser.buffer[parser.idx] = 0;
2538
2539                 /* we allow only one expression at a time */
2540                 ret = ftrace_set_func(array, &ftrace_graph_count,
2541                                         parser.buffer);
2542                 if (ret)
2543                         goto out;
2544         }
2545
2546         ret = read;
2547  out:
2548         trace_parser_put(&parser);
2549         mutex_unlock(&graph_lock);
2550
2551         return ret;
2552 }
2553
2554 static const struct file_operations ftrace_graph_fops = {
2555         .open           = ftrace_graph_open,
2556         .read           = seq_read,
2557         .write          = ftrace_graph_write,
2558         .release        = ftrace_graph_release,
2559 };
2560 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2561
2562 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2563 {
2564
2565         trace_create_file("available_filter_functions", 0444,
2566                         d_tracer, NULL, &ftrace_avail_fops);
2567
2568         trace_create_file("failures", 0444,
2569                         d_tracer, NULL, &ftrace_failures_fops);
2570
2571         trace_create_file("set_ftrace_filter", 0644, d_tracer,
2572                         NULL, &ftrace_filter_fops);
2573
2574         trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2575                                     NULL, &ftrace_notrace_fops);
2576
2577 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2578         trace_create_file("set_graph_function", 0444, d_tracer,
2579                                     NULL,
2580                                     &ftrace_graph_fops);
2581 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2582
2583         return 0;
2584 }
2585
2586 static int ftrace_convert_nops(struct module *mod,
2587                                unsigned long *start,
2588                                unsigned long *end)
2589 {
2590         unsigned long *p;
2591         unsigned long addr;
2592         unsigned long flags;
2593
2594         mutex_lock(&ftrace_lock);
2595         p = start;
2596         while (p < end) {
2597                 addr = ftrace_call_adjust(*p++);
2598                 /*
2599                  * Some architecture linkers will pad between
2600                  * the different mcount_loc sections of different
2601                  * object files to satisfy alignments.
2602                  * Skip any NULL pointers.
2603                  */
2604                 if (!addr)
2605                         continue;
2606                 ftrace_record_ip(addr);
2607         }
2608
2609         /* disable interrupts to prevent kstop machine */
2610         local_irq_save(flags);
2611         ftrace_update_code(mod);
2612         local_irq_restore(flags);
2613         mutex_unlock(&ftrace_lock);
2614
2615         return 0;
2616 }
2617
2618 #ifdef CONFIG_MODULES
2619 void ftrace_release(void *start, void *end)
2620 {
2621         struct dyn_ftrace *rec;
2622         struct ftrace_page *pg;
2623         unsigned long s = (unsigned long)start;
2624         unsigned long e = (unsigned long)end;
2625
2626         if (ftrace_disabled || !start || start == end)
2627                 return;
2628
2629         mutex_lock(&ftrace_lock);
2630         do_for_each_ftrace_rec(pg, rec) {
2631                 if ((rec->ip >= s) && (rec->ip < e)) {
2632                         /*
2633                          * rec->ip is changed in ftrace_free_rec()
2634                          * It should not between s and e if record was freed.
2635                          */
2636                         FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2637                         ftrace_free_rec(rec);
2638                 }
2639         } while_for_each_ftrace_rec();
2640         mutex_unlock(&ftrace_lock);
2641 }
2642
2643 static void ftrace_init_module(struct module *mod,
2644                                unsigned long *start, unsigned long *end)
2645 {
2646         if (ftrace_disabled || start == end)
2647                 return;
2648         ftrace_convert_nops(mod, start, end);
2649 }
2650
2651 static int ftrace_module_notify(struct notifier_block *self,
2652                                 unsigned long val, void *data)
2653 {
2654         struct module *mod = data;
2655
2656         switch (val) {
2657         case MODULE_STATE_COMING:
2658                 ftrace_init_module(mod, mod->ftrace_callsites,
2659                                    mod->ftrace_callsites +
2660                                    mod->num_ftrace_callsites);
2661                 break;
2662         case MODULE_STATE_GOING:
2663                 ftrace_release(mod->ftrace_callsites,
2664                                mod->ftrace_callsites +
2665                                mod->num_ftrace_callsites);
2666                 break;
2667         }
2668
2669         return 0;
2670 }
2671 #else
2672 static int ftrace_module_notify(struct notifier_block *self,
2673                                 unsigned long val, void *data)
2674 {
2675         return 0;
2676 }
2677 #endif /* CONFIG_MODULES */
2678
2679 struct notifier_block ftrace_module_nb = {
2680         .notifier_call = ftrace_module_notify,
2681         .priority = 0,
2682 };
2683
2684 extern unsigned long __start_mcount_loc[];
2685 extern unsigned long __stop_mcount_loc[];
2686
2687 void __init ftrace_init(void)
2688 {
2689         unsigned long count, addr, flags;
2690         int ret;
2691
2692         /* Keep the ftrace pointer to the stub */
2693         addr = (unsigned long)ftrace_stub;
2694
2695         local_irq_save(flags);
2696         ftrace_dyn_arch_init(&addr);
2697         local_irq_restore(flags);
2698
2699         /* ftrace_dyn_arch_init places the return code in addr */
2700         if (addr)
2701                 goto failed;
2702
2703         count = __stop_mcount_loc - __start_mcount_loc;
2704
2705         ret = ftrace_dyn_table_alloc(count);
2706         if (ret)
2707                 goto failed;
2708
2709         last_ftrace_enabled = ftrace_enabled = 1;
2710
2711         ret = ftrace_convert_nops(NULL,
2712                                   __start_mcount_loc,
2713                                   __stop_mcount_loc);
2714
2715         ret = register_module_notifier(&ftrace_module_nb);
2716         if (ret)
2717                 pr_warning("Failed to register trace ftrace module notifier\n");
2718
2719         set_ftrace_early_filters();
2720
2721         return;
2722  failed:
2723         ftrace_disabled = 1;
2724 }
2725
2726 #else
2727
2728 static int __init ftrace_nodyn_init(void)
2729 {
2730         ftrace_enabled = 1;
2731         return 0;
2732 }
2733 device_initcall(ftrace_nodyn_init);
2734
2735 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2736 static inline void ftrace_startup_enable(int command) { }
2737 /* Keep as macros so we do not need to define the commands */
2738 # define ftrace_startup(command)        do { } while (0)
2739 # define ftrace_shutdown(command)       do { } while (0)
2740 # define ftrace_startup_sysctl()        do { } while (0)
2741 # define ftrace_shutdown_sysctl()       do { } while (0)
2742 #endif /* CONFIG_DYNAMIC_FTRACE */
2743
2744 static ssize_t
2745 ftrace_pid_read(struct file *file, char __user *ubuf,
2746                        size_t cnt, loff_t *ppos)
2747 {
2748         char buf[64];
2749         int r;
2750
2751         if (ftrace_pid_trace == ftrace_swapper_pid)
2752                 r = sprintf(buf, "swapper tasks\n");
2753         else if (ftrace_pid_trace)
2754                 r = sprintf(buf, "%u\n", pid_vnr(ftrace_pid_trace));
2755         else
2756                 r = sprintf(buf, "no pid\n");
2757
2758         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2759 }
2760
2761 static void clear_ftrace_swapper(void)
2762 {
2763         struct task_struct *p;
2764         int cpu;
2765
2766         get_online_cpus();
2767         for_each_online_cpu(cpu) {
2768                 p = idle_task(cpu);
2769                 clear_tsk_trace_trace(p);
2770         }
2771         put_online_cpus();
2772 }
2773
2774 static void set_ftrace_swapper(void)
2775 {
2776         struct task_struct *p;
2777         int cpu;
2778
2779         get_online_cpus();
2780         for_each_online_cpu(cpu) {
2781                 p = idle_task(cpu);
2782                 set_tsk_trace_trace(p);
2783         }
2784         put_online_cpus();
2785 }
2786
2787 static void clear_ftrace_pid(struct pid *pid)
2788 {
2789         struct task_struct *p;
2790
2791         rcu_read_lock();
2792         do_each_pid_task(pid, PIDTYPE_PID, p) {
2793                 clear_tsk_trace_trace(p);
2794         } while_each_pid_task(pid, PIDTYPE_PID, p);
2795         rcu_read_unlock();
2796
2797         put_pid(pid);
2798 }
2799
2800 static void set_ftrace_pid(struct pid *pid)
2801 {
2802         struct task_struct *p;
2803
2804         rcu_read_lock();
2805         do_each_pid_task(pid, PIDTYPE_PID, p) {
2806                 set_tsk_trace_trace(p);
2807         } while_each_pid_task(pid, PIDTYPE_PID, p);
2808         rcu_read_unlock();
2809 }
2810
2811 static void clear_ftrace_pid_task(struct pid **pid)
2812 {
2813         if (*pid == ftrace_swapper_pid)
2814                 clear_ftrace_swapper();
2815         else
2816                 clear_ftrace_pid(*pid);
2817
2818         *pid = NULL;
2819 }
2820
2821 static void set_ftrace_pid_task(struct pid *pid)
2822 {
2823         if (pid == ftrace_swapper_pid)
2824                 set_ftrace_swapper();
2825         else
2826                 set_ftrace_pid(pid);
2827 }
2828
2829 static ssize_t
2830 ftrace_pid_write(struct file *filp, const char __user *ubuf,
2831                    size_t cnt, loff_t *ppos)
2832 {
2833         struct pid *pid;
2834         char buf[64];
2835         long val;
2836         int ret;
2837
2838         if (cnt >= sizeof(buf))
2839                 return -EINVAL;
2840
2841         if (copy_from_user(&buf, ubuf, cnt))
2842                 return -EFAULT;
2843
2844         buf[cnt] = 0;
2845
2846         ret = strict_strtol(buf, 10, &val);
2847         if (ret < 0)
2848                 return ret;
2849
2850         mutex_lock(&ftrace_lock);
2851         if (val < 0) {
2852                 /* disable pid tracing */
2853                 if (!ftrace_pid_trace)
2854                         goto out;
2855
2856                 clear_ftrace_pid_task(&ftrace_pid_trace);
2857
2858         } else {
2859                 /* swapper task is special */
2860                 if (!val) {
2861                         pid = ftrace_swapper_pid;
2862                         if (pid == ftrace_pid_trace)
2863                                 goto out;
2864                 } else {
2865                         pid = find_get_pid(val);
2866
2867                         if (pid == ftrace_pid_trace) {
2868                                 put_pid(pid);
2869                                 goto out;
2870                         }
2871                 }
2872
2873                 if (ftrace_pid_trace)
2874                         clear_ftrace_pid_task(&ftrace_pid_trace);
2875
2876                 if (!pid)
2877                         goto out;
2878
2879                 ftrace_pid_trace = pid;
2880
2881                 set_ftrace_pid_task(ftrace_pid_trace);
2882         }
2883
2884         /* update the function call */
2885         ftrace_update_pid_func();
2886         ftrace_startup_enable(0);
2887
2888  out:
2889         mutex_unlock(&ftrace_lock);
2890
2891         return cnt;
2892 }
2893
2894 static const struct file_operations ftrace_pid_fops = {
2895         .read = ftrace_pid_read,
2896         .write = ftrace_pid_write,
2897 };
2898
2899 static __init int ftrace_init_debugfs(void)
2900 {
2901         struct dentry *d_tracer;
2902
2903         d_tracer = tracing_init_dentry();
2904         if (!d_tracer)
2905                 return 0;
2906
2907         ftrace_init_dyn_debugfs(d_tracer);
2908
2909         trace_create_file("set_ftrace_pid", 0644, d_tracer,
2910                             NULL, &ftrace_pid_fops);
2911
2912         ftrace_profile_debugfs(d_tracer);
2913
2914         return 0;
2915 }
2916 fs_initcall(ftrace_init_debugfs);
2917
2918 /**
2919  * ftrace_kill - kill ftrace
2920  *
2921  * This function should be used by panic code. It stops ftrace
2922  * but in a not so nice way. If you need to simply kill ftrace
2923  * from a non-atomic section, use ftrace_kill.
2924  */
2925 void ftrace_kill(void)
2926 {
2927         ftrace_disabled = 1;
2928         ftrace_enabled = 0;
2929         clear_ftrace_function();
2930 }
2931
2932 /**
2933  * register_ftrace_function - register a function for profiling
2934  * @ops - ops structure that holds the function for profiling.
2935  *
2936  * Register a function to be called by all functions in the
2937  * kernel.
2938  *
2939  * Note: @ops->func and all the functions it calls must be labeled
2940  *       with "notrace", otherwise it will go into a
2941  *       recursive loop.
2942  */
2943 int register_ftrace_function(struct ftrace_ops *ops)
2944 {
2945         int ret;
2946
2947         if (unlikely(ftrace_disabled))
2948                 return -1;
2949
2950         mutex_lock(&ftrace_lock);
2951
2952         ret = __register_ftrace_function(ops);
2953         ftrace_startup(0);
2954
2955         mutex_unlock(&ftrace_lock);
2956         return ret;
2957 }
2958
2959 /**
2960  * unregister_ftrace_function - unregister a function for profiling.
2961  * @ops - ops structure that holds the function to unregister
2962  *
2963  * Unregister a function that was added to be called by ftrace profiling.
2964  */
2965 int unregister_ftrace_function(struct ftrace_ops *ops)
2966 {
2967         int ret;
2968
2969         mutex_lock(&ftrace_lock);
2970         ret = __unregister_ftrace_function(ops);
2971         ftrace_shutdown(0);
2972         mutex_unlock(&ftrace_lock);
2973
2974         return ret;
2975 }
2976
2977 int
2978 ftrace_enable_sysctl(struct ctl_table *table, int write,
2979                      struct file *file, void __user *buffer, size_t *lenp,
2980                      loff_t *ppos)
2981 {
2982         int ret;
2983
2984         if (unlikely(ftrace_disabled))
2985                 return -ENODEV;
2986
2987         mutex_lock(&ftrace_lock);
2988
2989         ret  = proc_dointvec(table, write, file, buffer, lenp, ppos);
2990
2991         if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
2992                 goto out;
2993
2994         last_ftrace_enabled = !!ftrace_enabled;
2995
2996         if (ftrace_enabled) {
2997
2998                 ftrace_startup_sysctl();
2999
3000                 /* we are starting ftrace again */
3001                 if (ftrace_list != &ftrace_list_end) {
3002                         if (ftrace_list->next == &ftrace_list_end)
3003                                 ftrace_trace_function = ftrace_list->func;
3004                         else
3005                                 ftrace_trace_function = ftrace_list_func;
3006                 }
3007
3008         } else {
3009                 /* stopping ftrace calls (just send to ftrace_stub) */
3010                 ftrace_trace_function = ftrace_stub;
3011
3012                 ftrace_shutdown_sysctl();
3013         }
3014
3015  out:
3016         mutex_unlock(&ftrace_lock);
3017         return ret;
3018 }
3019
3020 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3021
3022 static int ftrace_graph_active;
3023 static struct notifier_block ftrace_suspend_notifier;
3024
3025 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3026 {
3027         return 0;
3028 }
3029
3030 /* The callbacks that hook a function */
3031 trace_func_graph_ret_t ftrace_graph_return =
3032                         (trace_func_graph_ret_t)ftrace_stub;
3033 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3034
3035 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3036 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3037 {
3038         int i;
3039         int ret = 0;
3040         unsigned long flags;
3041         int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3042         struct task_struct *g, *t;
3043
3044         for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3045                 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3046                                         * sizeof(struct ftrace_ret_stack),
3047                                         GFP_KERNEL);
3048                 if (!ret_stack_list[i]) {
3049                         start = 0;
3050                         end = i;
3051                         ret = -ENOMEM;
3052                         goto free;
3053                 }
3054         }
3055
3056         read_lock_irqsave(&tasklist_lock, flags);
3057         do_each_thread(g, t) {
3058                 if (start == end) {
3059                         ret = -EAGAIN;
3060                         goto unlock;
3061                 }
3062
3063                 if (t->ret_stack == NULL) {
3064                         atomic_set(&t->tracing_graph_pause, 0);
3065                         atomic_set(&t->trace_overrun, 0);
3066                         t->curr_ret_stack = -1;
3067                         /* Make sure the tasks see the -1 first: */
3068                         smp_wmb();
3069                         t->ret_stack = ret_stack_list[start++];
3070                 }
3071         } while_each_thread(g, t);
3072
3073 unlock:
3074         read_unlock_irqrestore(&tasklist_lock, flags);
3075 free:
3076         for (i = start; i < end; i++)
3077                 kfree(ret_stack_list[i]);
3078         return ret;
3079 }
3080
3081 static void
3082 ftrace_graph_probe_sched_switch(struct rq *__rq, struct task_struct *prev,
3083                                 struct task_struct *next)
3084 {
3085         unsigned long long timestamp;
3086         int index;
3087
3088         /*
3089          * Does the user want to count the time a function was asleep.
3090          * If so, do not update the time stamps.
3091          */
3092         if (trace_flags & TRACE_ITER_SLEEP_TIME)
3093                 return;
3094
3095         timestamp = trace_clock_local();
3096
3097         prev->ftrace_timestamp = timestamp;
3098
3099         /* only process tasks that we timestamped */
3100         if (!next->ftrace_timestamp)
3101                 return;
3102
3103         /*
3104          * Update all the counters in next to make up for the
3105          * time next was sleeping.
3106          */
3107         timestamp -= next->ftrace_timestamp;
3108
3109         for (index = next->curr_ret_stack; index >= 0; index--)
3110                 next->ret_stack[index].calltime += timestamp;
3111 }
3112
3113 /* Allocate a return stack for each task */
3114 static int start_graph_tracing(void)
3115 {
3116         struct ftrace_ret_stack **ret_stack_list;
3117         int ret, cpu;
3118
3119         ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3120                                 sizeof(struct ftrace_ret_stack *),
3121                                 GFP_KERNEL);
3122
3123         if (!ret_stack_list)
3124                 return -ENOMEM;
3125
3126         /* The cpu_boot init_task->ret_stack will never be freed */
3127         for_each_online_cpu(cpu) {
3128                 if (!idle_task(cpu)->ret_stack)
3129                         ftrace_graph_init_task(idle_task(cpu));
3130         }
3131
3132         do {
3133                 ret = alloc_retstack_tasklist(ret_stack_list);
3134         } while (ret == -EAGAIN);
3135
3136         if (!ret) {
3137                 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch);
3138                 if (ret)
3139                         pr_info("ftrace_graph: Couldn't activate tracepoint"
3140                                 " probe to kernel_sched_switch\n");
3141         }
3142
3143         kfree(ret_stack_list);
3144         return ret;
3145 }
3146
3147 /*
3148  * Hibernation protection.
3149  * The state of the current task is too much unstable during
3150  * suspend/restore to disk. We want to protect against that.
3151  */
3152 static int
3153 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3154                                                         void *unused)
3155 {
3156         switch (state) {
3157         case PM_HIBERNATION_PREPARE:
3158                 pause_graph_tracing();
3159                 break;
3160
3161         case PM_POST_HIBERNATION:
3162                 unpause_graph_tracing();
3163                 break;
3164         }
3165         return NOTIFY_DONE;
3166 }
3167
3168 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3169                         trace_func_graph_ent_t entryfunc)
3170 {
3171         int ret = 0;
3172
3173         mutex_lock(&ftrace_lock);
3174
3175         /* we currently allow only one tracer registered at a time */
3176         if (ftrace_graph_active) {
3177                 ret = -EBUSY;
3178                 goto out;
3179         }
3180
3181         ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3182         register_pm_notifier(&ftrace_suspend_notifier);
3183
3184         ftrace_graph_active++;
3185         ret = start_graph_tracing();
3186         if (ret) {
3187                 ftrace_graph_active--;
3188                 goto out;
3189         }
3190
3191         ftrace_graph_return = retfunc;
3192         ftrace_graph_entry = entryfunc;
3193
3194         ftrace_startup(FTRACE_START_FUNC_RET);
3195
3196 out:
3197         mutex_unlock(&ftrace_lock);
3198         return ret;
3199 }
3200
3201 void unregister_ftrace_graph(void)
3202 {
3203         mutex_lock(&ftrace_lock);
3204
3205         if (unlikely(!ftrace_graph_active))
3206                 goto out;
3207
3208         ftrace_graph_active--;
3209         unregister_trace_sched_switch(ftrace_graph_probe_sched_switch);
3210         ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3211         ftrace_graph_entry = ftrace_graph_entry_stub;
3212         ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3213         unregister_pm_notifier(&ftrace_suspend_notifier);
3214
3215  out:
3216         mutex_unlock(&ftrace_lock);
3217 }
3218
3219 /* Allocate a return stack for newly created task */
3220 void ftrace_graph_init_task(struct task_struct *t)
3221 {
3222         /* Make sure we do not use the parent ret_stack */
3223         t->ret_stack = NULL;
3224
3225         if (ftrace_graph_active) {
3226                 struct ftrace_ret_stack *ret_stack;
3227
3228                 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3229                                 * sizeof(struct ftrace_ret_stack),
3230                                 GFP_KERNEL);
3231                 if (!ret_stack)
3232                         return;
3233                 t->curr_ret_stack = -1;
3234                 atomic_set(&t->tracing_graph_pause, 0);
3235                 atomic_set(&t->trace_overrun, 0);
3236                 t->ftrace_timestamp = 0;
3237                 /* make curr_ret_stack visable before we add the ret_stack */
3238                 smp_wmb();
3239                 t->ret_stack = ret_stack;
3240         }
3241 }
3242
3243 void ftrace_graph_exit_task(struct task_struct *t)
3244 {
3245         struct ftrace_ret_stack *ret_stack = t->ret_stack;
3246
3247         t->ret_stack = NULL;
3248         /* NULL must become visible to IRQs before we free it: */
3249         barrier();
3250
3251         kfree(ret_stack);
3252 }
3253
3254 void ftrace_graph_stop(void)
3255 {
3256         ftrace_stop();
3257 }
3258 #endif
3259