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_CONT        = (1 << 1),
1327         FTRACE_ITER_NOTRACE     = (1 << 2),
1328         FTRACE_ITER_FAILURES    = (1 << 3),
1329         FTRACE_ITER_PRINTALL    = (1 << 4),
1330         FTRACE_ITER_HASH        = (1 << 5),
1331 };
1332
1333 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1334
1335 struct ftrace_iterator {
1336         struct ftrace_page      *pg;
1337         int                     hidx;
1338         int                     idx;
1339         unsigned                flags;
1340         unsigned char           buffer[FTRACE_BUFF_MAX+1];
1341         unsigned                buffer_idx;
1342 };
1343
1344 static void *
1345 t_hash_next(struct seq_file *m, void *v, loff_t *pos)
1346 {
1347         struct ftrace_iterator *iter = m->private;
1348         struct hlist_node *hnd = v;
1349         struct hlist_head *hhd;
1350
1351         WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
1352
1353         (*pos)++;
1354
1355  retry:
1356         if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1357                 return NULL;
1358
1359         hhd = &ftrace_func_hash[iter->hidx];
1360
1361         if (hlist_empty(hhd)) {
1362                 iter->hidx++;
1363                 hnd = NULL;
1364                 goto retry;
1365         }
1366
1367         if (!hnd)
1368                 hnd = hhd->first;
1369         else {
1370                 hnd = hnd->next;
1371                 if (!hnd) {
1372                         iter->hidx++;
1373                         goto retry;
1374                 }
1375         }
1376
1377         return hnd;
1378 }
1379
1380 static void *t_hash_start(struct seq_file *m, loff_t *pos)
1381 {
1382         struct ftrace_iterator *iter = m->private;
1383         void *p = NULL;
1384         loff_t l;
1385
1386         if (!(iter->flags & FTRACE_ITER_HASH))
1387                 *pos = 0;
1388
1389         iter->flags |= FTRACE_ITER_HASH;
1390
1391         iter->hidx = 0;
1392         for (l = 0; l <= *pos; ) {
1393                 p = t_hash_next(m, p, &l);
1394                 if (!p)
1395                         break;
1396         }
1397         return p;
1398 }
1399
1400 static int t_hash_show(struct seq_file *m, void *v)
1401 {
1402         struct ftrace_func_probe *rec;
1403         struct hlist_node *hnd = v;
1404
1405         rec = hlist_entry(hnd, struct ftrace_func_probe, node);
1406
1407         if (rec->ops->print)
1408                 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
1409
1410         seq_printf(m, "%pf:%pf", (void *)rec->ip, (void *)rec->ops->func);
1411
1412         if (rec->data)
1413                 seq_printf(m, ":%p", rec->data);
1414         seq_putc(m, '\n');
1415
1416         return 0;
1417 }
1418
1419 static void *
1420 t_next(struct seq_file *m, void *v, loff_t *pos)
1421 {
1422         struct ftrace_iterator *iter = m->private;
1423         struct dyn_ftrace *rec = NULL;
1424
1425         if (iter->flags & FTRACE_ITER_HASH)
1426                 return t_hash_next(m, v, pos);
1427
1428         (*pos)++;
1429
1430         if (iter->flags & FTRACE_ITER_PRINTALL)
1431                 return NULL;
1432
1433  retry:
1434         if (iter->idx >= iter->pg->index) {
1435                 if (iter->pg->next) {
1436                         iter->pg = iter->pg->next;
1437                         iter->idx = 0;
1438                         goto retry;
1439                 }
1440         } else {
1441                 rec = &iter->pg->records[iter->idx++];
1442                 if ((rec->flags & FTRACE_FL_FREE) ||
1443
1444                     (!(iter->flags & FTRACE_ITER_FAILURES) &&
1445                      (rec->flags & FTRACE_FL_FAILED)) ||
1446
1447                     ((iter->flags & FTRACE_ITER_FAILURES) &&
1448                      !(rec->flags & FTRACE_FL_FAILED)) ||
1449
1450                     ((iter->flags & FTRACE_ITER_FILTER) &&
1451                      !(rec->flags & FTRACE_FL_FILTER)) ||
1452
1453                     ((iter->flags & FTRACE_ITER_NOTRACE) &&
1454                      !(rec->flags & FTRACE_FL_NOTRACE))) {
1455                         rec = NULL;
1456                         goto retry;
1457                 }
1458         }
1459
1460         return rec;
1461 }
1462
1463 static void *t_start(struct seq_file *m, loff_t *pos)
1464 {
1465         struct ftrace_iterator *iter = m->private;
1466         void *p = NULL;
1467         loff_t l;
1468
1469         mutex_lock(&ftrace_lock);
1470         /*
1471          * For set_ftrace_filter reading, if we have the filter
1472          * off, we can short cut and just print out that all
1473          * functions are enabled.
1474          */
1475         if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
1476                 if (*pos > 0)
1477                         return t_hash_start(m, pos);
1478                 iter->flags |= FTRACE_ITER_PRINTALL;
1479                 return iter;
1480         }
1481
1482         if (iter->flags & FTRACE_ITER_HASH)
1483                 return t_hash_start(m, pos);
1484
1485         iter->pg = ftrace_pages_start;
1486         iter->idx = 0;
1487         for (l = 0; l <= *pos; ) {
1488                 p = t_next(m, p, &l);
1489                 if (!p)
1490                         break;
1491         }
1492
1493         if (!p && iter->flags & FTRACE_ITER_FILTER)
1494                 return t_hash_start(m, pos);
1495
1496         return p;
1497 }
1498
1499 static void t_stop(struct seq_file *m, void *p)
1500 {
1501         mutex_unlock(&ftrace_lock);
1502 }
1503
1504 static int t_show(struct seq_file *m, void *v)
1505 {
1506         struct ftrace_iterator *iter = m->private;
1507         struct dyn_ftrace *rec = v;
1508
1509         if (iter->flags & FTRACE_ITER_HASH)
1510                 return t_hash_show(m, v);
1511
1512         if (iter->flags & FTRACE_ITER_PRINTALL) {
1513                 seq_printf(m, "#### all functions enabled ####\n");
1514                 return 0;
1515         }
1516
1517         if (!rec)
1518                 return 0;
1519
1520         seq_printf(m, "%pf\n", (void *)rec->ip);
1521
1522         return 0;
1523 }
1524
1525 static struct seq_operations show_ftrace_seq_ops = {
1526         .start = t_start,
1527         .next = t_next,
1528         .stop = t_stop,
1529         .show = t_show,
1530 };
1531
1532 static int
1533 ftrace_avail_open(struct inode *inode, struct file *file)
1534 {
1535         struct ftrace_iterator *iter;
1536         int ret;
1537
1538         if (unlikely(ftrace_disabled))
1539                 return -ENODEV;
1540
1541         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1542         if (!iter)
1543                 return -ENOMEM;
1544
1545         iter->pg = ftrace_pages_start;
1546
1547         ret = seq_open(file, &show_ftrace_seq_ops);
1548         if (!ret) {
1549                 struct seq_file *m = file->private_data;
1550
1551                 m->private = iter;
1552         } else {
1553                 kfree(iter);
1554         }
1555
1556         return ret;
1557 }
1558
1559 static int
1560 ftrace_failures_open(struct inode *inode, struct file *file)
1561 {
1562         int ret;
1563         struct seq_file *m;
1564         struct ftrace_iterator *iter;
1565
1566         ret = ftrace_avail_open(inode, file);
1567         if (!ret) {
1568                 m = (struct seq_file *)file->private_data;
1569                 iter = (struct ftrace_iterator *)m->private;
1570                 iter->flags = FTRACE_ITER_FAILURES;
1571         }
1572
1573         return ret;
1574 }
1575
1576
1577 static void ftrace_filter_reset(int enable)
1578 {
1579         struct ftrace_page *pg;
1580         struct dyn_ftrace *rec;
1581         unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1582
1583         mutex_lock(&ftrace_lock);
1584         if (enable)
1585                 ftrace_filtered = 0;
1586         do_for_each_ftrace_rec(pg, rec) {
1587                 if (rec->flags & FTRACE_FL_FAILED)
1588                         continue;
1589                 rec->flags &= ~type;
1590         } while_for_each_ftrace_rec();
1591         mutex_unlock(&ftrace_lock);
1592 }
1593
1594 static int
1595 ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1596 {
1597         struct ftrace_iterator *iter;
1598         int ret = 0;
1599
1600         if (unlikely(ftrace_disabled))
1601                 return -ENODEV;
1602
1603         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1604         if (!iter)
1605                 return -ENOMEM;
1606
1607         mutex_lock(&ftrace_regex_lock);
1608         if ((file->f_mode & FMODE_WRITE) &&
1609             (file->f_flags & O_TRUNC))
1610                 ftrace_filter_reset(enable);
1611
1612         if (file->f_mode & FMODE_READ) {
1613                 iter->pg = ftrace_pages_start;
1614                 iter->flags = enable ? FTRACE_ITER_FILTER :
1615                         FTRACE_ITER_NOTRACE;
1616
1617                 ret = seq_open(file, &show_ftrace_seq_ops);
1618                 if (!ret) {
1619                         struct seq_file *m = file->private_data;
1620                         m->private = iter;
1621                 } else
1622                         kfree(iter);
1623         } else
1624                 file->private_data = iter;
1625         mutex_unlock(&ftrace_regex_lock);
1626
1627         return ret;
1628 }
1629
1630 static int
1631 ftrace_filter_open(struct inode *inode, struct file *file)
1632 {
1633         return ftrace_regex_open(inode, file, 1);
1634 }
1635
1636 static int
1637 ftrace_notrace_open(struct inode *inode, struct file *file)
1638 {
1639         return ftrace_regex_open(inode, file, 0);
1640 }
1641
1642 static loff_t
1643 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1644 {
1645         loff_t ret;
1646
1647         if (file->f_mode & FMODE_READ)
1648                 ret = seq_lseek(file, offset, origin);
1649         else
1650                 file->f_pos = ret = 1;
1651
1652         return ret;
1653 }
1654
1655 enum {
1656         MATCH_FULL,
1657         MATCH_FRONT_ONLY,
1658         MATCH_MIDDLE_ONLY,
1659         MATCH_END_ONLY,
1660 };
1661
1662 /*
1663  * (static function - no need for kernel doc)
1664  *
1665  * Pass in a buffer containing a glob and this function will
1666  * set search to point to the search part of the buffer and
1667  * return the type of search it is (see enum above).
1668  * This does modify buff.
1669  *
1670  * Returns enum type.
1671  *  search returns the pointer to use for comparison.
1672  *  not returns 1 if buff started with a '!'
1673  *     0 otherwise.
1674  */
1675 static int
1676 ftrace_setup_glob(char *buff, int len, char **search, int *not)
1677 {
1678         int type = MATCH_FULL;
1679         int i;
1680
1681         if (buff[0] == '!') {
1682                 *not = 1;
1683                 buff++;
1684                 len--;
1685         } else
1686                 *not = 0;
1687
1688         *search = buff;
1689
1690         for (i = 0; i < len; i++) {
1691                 if (buff[i] == '*') {
1692                         if (!i) {
1693                                 *search = buff + 1;
1694                                 type = MATCH_END_ONLY;
1695                         } else {
1696                                 if (type == MATCH_END_ONLY)
1697                                         type = MATCH_MIDDLE_ONLY;
1698                                 else
1699                                         type = MATCH_FRONT_ONLY;
1700                                 buff[i] = 0;
1701                                 break;
1702                         }
1703                 }
1704         }
1705
1706         return type;
1707 }
1708
1709 static int ftrace_match(char *str, char *regex, int len, int type)
1710 {
1711         int matched = 0;
1712         char *ptr;
1713
1714         switch (type) {
1715         case MATCH_FULL:
1716                 if (strcmp(str, regex) == 0)
1717                         matched = 1;
1718                 break;
1719         case MATCH_FRONT_ONLY:
1720                 if (strncmp(str, regex, len) == 0)
1721                         matched = 1;
1722                 break;
1723         case MATCH_MIDDLE_ONLY:
1724                 if (strstr(str, regex))
1725                         matched = 1;
1726                 break;
1727         case MATCH_END_ONLY:
1728                 ptr = strstr(str, regex);
1729                 if (ptr && (ptr[len] == 0))
1730                         matched = 1;
1731                 break;
1732         }
1733
1734         return matched;
1735 }
1736
1737 static int
1738 ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1739 {
1740         char str[KSYM_SYMBOL_LEN];
1741
1742         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1743         return ftrace_match(str, regex, len, type);
1744 }
1745
1746 static void ftrace_match_records(char *buff, int len, int enable)
1747 {
1748         unsigned int search_len;
1749         struct ftrace_page *pg;
1750         struct dyn_ftrace *rec;
1751         unsigned long flag;
1752         char *search;
1753         int type;
1754         int not;
1755
1756         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1757         type = ftrace_setup_glob(buff, len, &search, &not);
1758
1759         search_len = strlen(search);
1760
1761         mutex_lock(&ftrace_lock);
1762         do_for_each_ftrace_rec(pg, rec) {
1763
1764                 if (rec->flags & FTRACE_FL_FAILED)
1765                         continue;
1766
1767                 if (ftrace_match_record(rec, search, search_len, type)) {
1768                         if (not)
1769                                 rec->flags &= ~flag;
1770                         else
1771                                 rec->flags |= flag;
1772                 }
1773                 /*
1774                  * Only enable filtering if we have a function that
1775                  * is filtered on.
1776                  */
1777                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1778                         ftrace_filtered = 1;
1779         } while_for_each_ftrace_rec();
1780         mutex_unlock(&ftrace_lock);
1781 }
1782
1783 static int
1784 ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1785                            char *regex, int len, int type)
1786 {
1787         char str[KSYM_SYMBOL_LEN];
1788         char *modname;
1789
1790         kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1791
1792         if (!modname || strcmp(modname, mod))
1793                 return 0;
1794
1795         /* blank search means to match all funcs in the mod */
1796         if (len)
1797                 return ftrace_match(str, regex, len, type);
1798         else
1799                 return 1;
1800 }
1801
1802 static void ftrace_match_module_records(char *buff, char *mod, int enable)
1803 {
1804         unsigned search_len = 0;
1805         struct ftrace_page *pg;
1806         struct dyn_ftrace *rec;
1807         int type = MATCH_FULL;
1808         char *search = buff;
1809         unsigned long flag;
1810         int not = 0;
1811
1812         flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1813
1814         /* blank or '*' mean the same */
1815         if (strcmp(buff, "*") == 0)
1816                 buff[0] = 0;
1817
1818         /* handle the case of 'dont filter this module' */
1819         if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1820                 buff[0] = 0;
1821                 not = 1;
1822         }
1823
1824         if (strlen(buff)) {
1825                 type = ftrace_setup_glob(buff, strlen(buff), &search, &not);
1826                 search_len = strlen(search);
1827         }
1828
1829         mutex_lock(&ftrace_lock);
1830         do_for_each_ftrace_rec(pg, rec) {
1831
1832                 if (rec->flags & FTRACE_FL_FAILED)
1833                         continue;
1834
1835                 if (ftrace_match_module_record(rec, mod,
1836                                                search, search_len, type)) {
1837                         if (not)
1838                                 rec->flags &= ~flag;
1839                         else
1840                                 rec->flags |= flag;
1841                 }
1842                 if (enable && (rec->flags & FTRACE_FL_FILTER))
1843                         ftrace_filtered = 1;
1844
1845         } while_for_each_ftrace_rec();
1846         mutex_unlock(&ftrace_lock);
1847 }
1848
1849 /*
1850  * We register the module command as a template to show others how
1851  * to register the a command as well.
1852  */
1853
1854 static int
1855 ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1856 {
1857         char *mod;
1858
1859         /*
1860          * cmd == 'mod' because we only registered this func
1861          * for the 'mod' ftrace_func_command.
1862          * But if you register one func with multiple commands,
1863          * you can tell which command was used by the cmd
1864          * parameter.
1865          */
1866
1867         /* we must have a module name */
1868         if (!param)
1869                 return -EINVAL;
1870
1871         mod = strsep(&param, ":");
1872         if (!strlen(mod))
1873                 return -EINVAL;
1874
1875         ftrace_match_module_records(func, mod, enable);
1876         return 0;
1877 }
1878
1879 static struct ftrace_func_command ftrace_mod_cmd = {
1880         .name                   = "mod",
1881         .func                   = ftrace_mod_callback,
1882 };
1883
1884 static int __init ftrace_mod_cmd_init(void)
1885 {
1886         return register_ftrace_command(&ftrace_mod_cmd);
1887 }
1888 device_initcall(ftrace_mod_cmd_init);
1889
1890 static void
1891 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1892 {
1893         struct ftrace_func_probe *entry;
1894         struct hlist_head *hhd;
1895         struct hlist_node *n;
1896         unsigned long key;
1897         int resched;
1898
1899         key = hash_long(ip, FTRACE_HASH_BITS);
1900
1901         hhd = &ftrace_func_hash[key];
1902
1903         if (hlist_empty(hhd))
1904                 return;
1905
1906         /*
1907          * Disable preemption for these calls to prevent a RCU grace
1908          * period. This syncs the hash iteration and freeing of items
1909          * on the hash. rcu_read_lock is too dangerous here.
1910          */
1911         resched = ftrace_preempt_disable();
1912         hlist_for_each_entry_rcu(entry, n, hhd, node) {
1913                 if (entry->ip == ip)
1914                         entry->ops->func(ip, parent_ip, &entry->data);
1915         }
1916         ftrace_preempt_enable(resched);
1917 }
1918
1919 static struct ftrace_ops trace_probe_ops __read_mostly =
1920 {
1921         .func           = function_trace_probe_call,
1922 };
1923
1924 static int ftrace_probe_registered;
1925
1926 static void __enable_ftrace_function_probe(void)
1927 {
1928         int i;
1929
1930         if (ftrace_probe_registered)
1931                 return;
1932
1933         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1934                 struct hlist_head *hhd = &ftrace_func_hash[i];
1935                 if (hhd->first)
1936                         break;
1937         }
1938         /* Nothing registered? */
1939         if (i == FTRACE_FUNC_HASHSIZE)
1940                 return;
1941
1942         __register_ftrace_function(&trace_probe_ops);
1943         ftrace_startup(0);
1944         ftrace_probe_registered = 1;
1945 }
1946
1947 static void __disable_ftrace_function_probe(void)
1948 {
1949         int i;
1950
1951         if (!ftrace_probe_registered)
1952                 return;
1953
1954         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1955                 struct hlist_head *hhd = &ftrace_func_hash[i];
1956                 if (hhd->first)
1957                         return;
1958         }
1959
1960         /* no more funcs left */
1961         __unregister_ftrace_function(&trace_probe_ops);
1962         ftrace_shutdown(0);
1963         ftrace_probe_registered = 0;
1964 }
1965
1966
1967 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
1968 {
1969         struct ftrace_func_probe *entry =
1970                 container_of(rhp, struct ftrace_func_probe, rcu);
1971
1972         if (entry->ops->free)
1973                 entry->ops->free(&entry->data);
1974         kfree(entry);
1975 }
1976
1977
1978 int
1979 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
1980                               void *data)
1981 {
1982         struct ftrace_func_probe *entry;
1983         struct ftrace_page *pg;
1984         struct dyn_ftrace *rec;
1985         int type, len, not;
1986         unsigned long key;
1987         int count = 0;
1988         char *search;
1989
1990         type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
1991         len = strlen(search);
1992
1993         /* we do not support '!' for function probes */
1994         if (WARN_ON(not))
1995                 return -EINVAL;
1996
1997         mutex_lock(&ftrace_lock);
1998         do_for_each_ftrace_rec(pg, rec) {
1999
2000                 if (rec->flags & FTRACE_FL_FAILED)
2001                         continue;
2002
2003                 if (!ftrace_match_record(rec, search, len, type))
2004                         continue;
2005
2006                 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2007                 if (!entry) {
2008                         /* If we did not process any, then return error */
2009                         if (!count)
2010                                 count = -ENOMEM;
2011                         goto out_unlock;
2012                 }
2013
2014                 count++;
2015
2016                 entry->data = data;
2017
2018                 /*
2019                  * The caller might want to do something special
2020                  * for each function we find. We call the callback
2021                  * to give the caller an opportunity to do so.
2022                  */
2023                 if (ops->callback) {
2024                         if (ops->callback(rec->ip, &entry->data) < 0) {
2025                                 /* caller does not like this func */
2026                                 kfree(entry);
2027                                 continue;
2028                         }
2029                 }
2030
2031                 entry->ops = ops;
2032                 entry->ip = rec->ip;
2033
2034                 key = hash_long(entry->ip, FTRACE_HASH_BITS);
2035                 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2036
2037         } while_for_each_ftrace_rec();
2038         __enable_ftrace_function_probe();
2039
2040  out_unlock:
2041         mutex_unlock(&ftrace_lock);
2042
2043         return count;
2044 }
2045
2046 enum {
2047         PROBE_TEST_FUNC         = 1,
2048         PROBE_TEST_DATA         = 2
2049 };
2050
2051 static void
2052 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2053                                   void *data, int flags)
2054 {
2055         struct ftrace_func_probe *entry;
2056         struct hlist_node *n, *tmp;
2057         char str[KSYM_SYMBOL_LEN];
2058         int type = MATCH_FULL;
2059         int i, len = 0;
2060         char *search;
2061
2062         if (glob && (strcmp(glob, "*") || !strlen(glob)))
2063                 glob = NULL;
2064         else {
2065                 int not;
2066
2067                 type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
2068                 len = strlen(search);
2069
2070                 /* we do not support '!' for function probes */
2071                 if (WARN_ON(not))
2072                         return;
2073         }
2074
2075         mutex_lock(&ftrace_lock);
2076         for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2077                 struct hlist_head *hhd = &ftrace_func_hash[i];
2078
2079                 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
2080
2081                         /* break up if statements for readability */
2082                         if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
2083                                 continue;
2084
2085                         if ((flags & PROBE_TEST_DATA) && entry->data != data)
2086                                 continue;
2087
2088                         /* do this last, since it is the most expensive */
2089                         if (glob) {
2090                                 kallsyms_lookup(entry->ip, NULL, NULL,
2091                                                 NULL, str);
2092                                 if (!ftrace_match(str, glob, len, type))
2093                                         continue;
2094                         }
2095
2096                         hlist_del(&entry->node);
2097                         call_rcu(&entry->rcu, ftrace_free_entry_rcu);
2098                 }
2099         }
2100         __disable_ftrace_function_probe();
2101         mutex_unlock(&ftrace_lock);
2102 }
2103
2104 void
2105 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2106                                 void *data)
2107 {
2108         __unregister_ftrace_function_probe(glob, ops, data,
2109                                           PROBE_TEST_FUNC | PROBE_TEST_DATA);
2110 }
2111
2112 void
2113 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
2114 {
2115         __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
2116 }
2117
2118 void unregister_ftrace_function_probe_all(char *glob)
2119 {
2120         __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
2121 }
2122
2123 static LIST_HEAD(ftrace_commands);
2124 static DEFINE_MUTEX(ftrace_cmd_mutex);
2125
2126 int register_ftrace_command(struct ftrace_func_command *cmd)
2127 {
2128         struct ftrace_func_command *p;
2129         int ret = 0;
2130
2131         mutex_lock(&ftrace_cmd_mutex);
2132         list_for_each_entry(p, &ftrace_commands, list) {
2133                 if (strcmp(cmd->name, p->name) == 0) {
2134                         ret = -EBUSY;
2135                         goto out_unlock;
2136                 }
2137         }
2138         list_add(&cmd->list, &ftrace_commands);
2139  out_unlock:
2140         mutex_unlock(&ftrace_cmd_mutex);
2141
2142         return ret;
2143 }
2144
2145 int unregister_ftrace_command(struct ftrace_func_command *cmd)
2146 {
2147         struct ftrace_func_command *p, *n;
2148         int ret = -ENODEV;
2149
2150         mutex_lock(&ftrace_cmd_mutex);
2151         list_for_each_entry_safe(p, n, &ftrace_commands, list) {
2152                 if (strcmp(cmd->name, p->name) == 0) {
2153                         ret = 0;
2154                         list_del_init(&p->list);
2155                         goto out_unlock;
2156                 }
2157         }
2158  out_unlock:
2159         mutex_unlock(&ftrace_cmd_mutex);
2160
2161         return ret;
2162 }
2163
2164 static int ftrace_process_regex(char *buff, int len, int enable)
2165 {
2166         char *func, *command, *next = buff;
2167         struct ftrace_func_command *p;
2168         int ret = -EINVAL;
2169
2170         func = strsep(&next, ":");
2171
2172         if (!next) {
2173                 ftrace_match_records(func, len, enable);
2174                 return 0;
2175         }
2176
2177         /* command found */
2178
2179         command = strsep(&next, ":");
2180
2181         mutex_lock(&ftrace_cmd_mutex);
2182         list_for_each_entry(p, &ftrace_commands, list) {
2183                 if (strcmp(p->name, command) == 0) {
2184                         ret = p->func(func, command, next, enable);
2185                         goto out_unlock;
2186                 }
2187         }
2188  out_unlock:
2189         mutex_unlock(&ftrace_cmd_mutex);
2190
2191         return ret;
2192 }
2193
2194 static ssize_t
2195 ftrace_regex_write(struct file *file, const char __user *ubuf,
2196                    size_t cnt, loff_t *ppos, int enable)
2197 {
2198         struct ftrace_iterator *iter;
2199         char ch;
2200         size_t read = 0;
2201         ssize_t ret;
2202
2203         if (!cnt || cnt < 0)
2204                 return 0;
2205
2206         mutex_lock(&ftrace_regex_lock);
2207
2208         if (file->f_mode & FMODE_READ) {
2209                 struct seq_file *m = file->private_data;
2210                 iter = m->private;
2211         } else
2212                 iter = file->private_data;
2213
2214         if (!*ppos) {
2215                 iter->flags &= ~FTRACE_ITER_CONT;
2216                 iter->buffer_idx = 0;
2217         }
2218
2219         ret = get_user(ch, ubuf++);
2220         if (ret)
2221                 goto out;
2222         read++;
2223         cnt--;
2224
2225         /*
2226          * If the parser haven't finished with the last write,
2227          * continue reading the user input without skipping spaces.
2228          */
2229         if (!(iter->flags & FTRACE_ITER_CONT)) {
2230                 /* skip white space */
2231                 while (cnt && isspace(ch)) {
2232                         ret = get_user(ch, ubuf++);
2233                         if (ret)
2234                                 goto out;
2235                         read++;
2236                         cnt--;
2237                 }
2238
2239                 /* only spaces were written */
2240                 if (isspace(ch)) {
2241                         *ppos += read;
2242                         ret = read;
2243                         goto out;
2244                 }
2245
2246                 iter->buffer_idx = 0;
2247         }
2248
2249         while (cnt && !isspace(ch)) {
2250                 if (iter->buffer_idx < FTRACE_BUFF_MAX)
2251                         iter->buffer[iter->buffer_idx++] = ch;
2252                 else {
2253                         ret = -EINVAL;
2254                         goto out;
2255                 }
2256                 ret = get_user(ch, ubuf++);
2257                 if (ret)
2258                         goto out;
2259                 read++;
2260                 cnt--;
2261         }
2262
2263         if (isspace(ch)) {
2264                 iter->buffer[iter->buffer_idx] = 0;
2265                 ret = ftrace_process_regex(iter->buffer,
2266                                            iter->buffer_idx, enable);
2267                 if (ret)
2268                         goto out;
2269                 iter->buffer_idx = 0;
2270         } else {
2271                 iter->flags |= FTRACE_ITER_CONT;
2272                 iter->buffer[iter->buffer_idx++] = ch;
2273         }
2274
2275         *ppos += read;
2276         ret = read;
2277  out:
2278         mutex_unlock(&ftrace_regex_lock);
2279
2280         return ret;
2281 }
2282
2283 static ssize_t
2284 ftrace_filter_write(struct file *file, const char __user *ubuf,
2285                     size_t cnt, loff_t *ppos)
2286 {
2287         return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
2288 }
2289
2290 static ssize_t
2291 ftrace_notrace_write(struct file *file, const char __user *ubuf,
2292                      size_t cnt, loff_t *ppos)
2293 {
2294         return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
2295 }
2296
2297 static void
2298 ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
2299 {
2300         if (unlikely(ftrace_disabled))
2301                 return;
2302
2303         mutex_lock(&ftrace_regex_lock);
2304         if (reset)
2305                 ftrace_filter_reset(enable);
2306         if (buf)
2307                 ftrace_match_records(buf, len, enable);
2308         mutex_unlock(&ftrace_regex_lock);
2309 }
2310
2311 /**
2312  * ftrace_set_filter - set a function to filter on in ftrace
2313  * @buf - the string that holds the function filter text.
2314  * @len - the length of the string.
2315  * @reset - non zero to reset all filters before applying this filter.
2316  *
2317  * Filters denote which functions should be enabled when tracing is enabled.
2318  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
2319  */
2320 void ftrace_set_filter(unsigned char *buf, int len, int reset)
2321 {
2322         ftrace_set_regex(buf, len, reset, 1);
2323 }
2324
2325 /**
2326  * ftrace_set_notrace - set a function to not trace in ftrace
2327  * @buf - the string that holds the function notrace text.
2328  * @len - the length of the string.
2329  * @reset - non zero to reset all filters before applying this filter.
2330  *
2331  * Notrace Filters denote which functions should not be enabled when tracing
2332  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
2333  * for tracing.
2334  */
2335 void ftrace_set_notrace(unsigned char *buf, int len, int reset)
2336 {
2337         ftrace_set_regex(buf, len, reset, 0);
2338 }
2339
2340 /*
2341  * command line interface to allow users to set filters on boot up.
2342  */
2343 #define FTRACE_FILTER_SIZE              COMMAND_LINE_SIZE
2344 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
2345 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
2346
2347 static int __init set_ftrace_notrace(char *str)
2348 {
2349         strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
2350         return 1;
2351 }
2352 __setup("ftrace_notrace=", set_ftrace_notrace);
2353
2354 static int __init set_ftrace_filter(char *str)
2355 {
2356         strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
2357         return 1;
2358 }
2359 __setup("ftrace_filter=", set_ftrace_filter);
2360
2361 static void __init set_ftrace_early_filter(char *buf, int enable)
2362 {
2363         char *func;
2364
2365         while (buf) {
2366                 func = strsep(&buf, ",");
2367                 ftrace_set_regex(func, strlen(func), 0, enable);
2368         }
2369 }
2370
2371 static void __init set_ftrace_early_filters(void)
2372 {
2373         if (ftrace_filter_buf[0])
2374                 set_ftrace_early_filter(ftrace_filter_buf, 1);
2375         if (ftrace_notrace_buf[0])
2376                 set_ftrace_early_filter(ftrace_notrace_buf, 0);
2377 }
2378
2379 static int
2380 ftrace_regex_release(struct inode *inode, struct file *file, int enable)
2381 {
2382         struct seq_file *m = (struct seq_file *)file->private_data;
2383         struct ftrace_iterator *iter;
2384
2385         mutex_lock(&ftrace_regex_lock);
2386         if (file->f_mode & FMODE_READ) {
2387                 iter = m->private;
2388
2389                 seq_release(inode, file);
2390         } else
2391                 iter = file->private_data;
2392
2393         if (iter->buffer_idx) {
2394                 iter->buffer[iter->buffer_idx] = 0;
2395                 ftrace_match_records(iter->buffer, iter->buffer_idx, enable);
2396         }
2397
2398         mutex_lock(&ftrace_lock);
2399         if (ftrace_start_up && ftrace_enabled)
2400                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
2401         mutex_unlock(&ftrace_lock);
2402
2403         kfree(iter);
2404         mutex_unlock(&ftrace_regex_lock);
2405         return 0;
2406 }
2407
2408 static int
2409 ftrace_filter_release(struct inode *inode, struct file *file)
2410 {
2411         return ftrace_regex_release(inode, file, 1);
2412 }
2413
2414 static int
2415 ftrace_notrace_release(struct inode *inode, struct file *file)
2416 {
2417         return ftrace_regex_release(inode, file, 0);
2418 }
2419
2420 static const struct file_operations ftrace_avail_fops = {
2421         .open = ftrace_avail_open,
2422         .read = seq_read,
2423         .llseek = seq_lseek,
2424         .release = seq_release_private,
2425 };
2426
2427 static const struct file_operations ftrace_failures_fops = {
2428         .open = ftrace_failures_open,
2429         .read = seq_read,
2430         .llseek = seq_lseek,
2431         .release = seq_release_private,
2432 };
2433
2434 static const struct file_operations ftrace_filter_fops = {
2435         .open = ftrace_filter_open,
2436         .read = seq_read,
2437         .write = ftrace_filter_write,
2438         .llseek = ftrace_regex_lseek,
2439         .release = ftrace_filter_release,
2440 };
2441
2442 static const struct file_operations ftrace_notrace_fops = {
2443         .open = ftrace_notrace_open,
2444         .read = seq_read,
2445         .write = ftrace_notrace_write,
2446         .llseek = ftrace_regex_lseek,
2447         .release = ftrace_notrace_release,
2448 };
2449
2450 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2451
2452 static DEFINE_MUTEX(graph_lock);
2453
2454 int ftrace_graph_count;
2455 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
2456
2457 static void *
2458 __g_next(struct seq_file *m, loff_t *pos)
2459 {
2460         unsigned long *array = m->private;
2461
2462         if (*pos >= ftrace_graph_count)
2463                 return NULL;
2464         return &array[*pos];
2465 }
2466
2467 static void *
2468 g_next(struct seq_file *m, void *v, loff_t *pos)
2469 {
2470         (*pos)++;
2471         return __g_next(m, pos);
2472 }
2473
2474 static void *g_start(struct seq_file *m, loff_t *pos)
2475 {
2476         mutex_lock(&graph_lock);
2477
2478         /* Nothing, tell g_show to print all functions are enabled */
2479         if (!ftrace_graph_count && !*pos)
2480                 return (void *)1;
2481
2482         return __g_next(m, pos);
2483 }
2484
2485 static void g_stop(struct seq_file *m, void *p)
2486 {
2487         mutex_unlock(&graph_lock);
2488 }
2489
2490 static int g_show(struct seq_file *m, void *v)
2491 {
2492         unsigned long *ptr = v;
2493
2494         if (!ptr)
2495                 return 0;
2496
2497         if (ptr == (unsigned long *)1) {
2498                 seq_printf(m, "#### all functions enabled ####\n");
2499                 return 0;
2500         }
2501
2502         seq_printf(m, "%pf\n", v);
2503
2504         return 0;
2505 }
2506
2507 static struct seq_operations ftrace_graph_seq_ops = {
2508         .start = g_start,
2509         .next = g_next,
2510         .stop = g_stop,
2511         .show = g_show,
2512 };
2513
2514 static int
2515 ftrace_graph_open(struct inode *inode, struct file *file)
2516 {
2517         int ret = 0;
2518
2519         if (unlikely(ftrace_disabled))
2520                 return -ENODEV;
2521
2522         mutex_lock(&graph_lock);
2523         if ((file->f_mode & FMODE_WRITE) &&
2524             (file->f_flags & O_TRUNC)) {
2525                 ftrace_graph_count = 0;
2526                 memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
2527         }
2528
2529         if (file->f_mode & FMODE_READ) {
2530                 ret = seq_open(file, &ftrace_graph_seq_ops);
2531                 if (!ret) {
2532                         struct seq_file *m = file->private_data;
2533                         m->private = ftrace_graph_funcs;
2534                 }
2535         } else
2536                 file->private_data = ftrace_graph_funcs;
2537         mutex_unlock(&graph_lock);
2538
2539         return ret;
2540 }
2541
2542 static int
2543 ftrace_graph_release(struct inode *inode, struct file *file)
2544 {
2545         if (file->f_mode & FMODE_READ)
2546                 seq_release(inode, file);
2547         return 0;
2548 }
2549
2550 static int
2551 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
2552 {
2553         struct dyn_ftrace *rec;
2554         struct ftrace_page *pg;
2555         int search_len;
2556         int found = 0;
2557         int type, not;
2558         char *search;
2559         bool exists;
2560         int i;
2561
2562         if (ftrace_disabled)
2563                 return -ENODEV;
2564
2565         /* decode regex */
2566         type = ftrace_setup_glob(buffer, strlen(buffer), &search, &not);
2567         if (not)
2568                 return -EINVAL;
2569
2570         search_len = strlen(search);
2571
2572         mutex_lock(&ftrace_lock);
2573         do_for_each_ftrace_rec(pg, rec) {
2574
2575                 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
2576                         break;
2577
2578                 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2579                         continue;
2580
2581                 if (ftrace_match_record(rec, search, search_len, type)) {
2582                         /* ensure it is not already in the array */
2583                         exists = false;
2584                         for (i = 0; i < *idx; i++)
2585                                 if (array[i] == rec->ip) {
2586                                         exists = true;
2587                                         break;
2588                                 }
2589                         if (!exists) {
2590                                 array[(*idx)++] = rec->ip;
2591                                 found = 1;
2592                         }
2593                 }
2594         } while_for_each_ftrace_rec();
2595
2596         mutex_unlock(&ftrace_lock);
2597
2598         return found ? 0 : -EINVAL;
2599 }
2600
2601 static ssize_t
2602 ftrace_graph_write(struct file *file, const char __user *ubuf,
2603                    size_t cnt, loff_t *ppos)
2604 {
2605         unsigned char buffer[FTRACE_BUFF_MAX+1];
2606         unsigned long *array;
2607         size_t read = 0;
2608         ssize_t ret;
2609         int index = 0;
2610         char ch;
2611
2612         if (!cnt || cnt < 0)
2613                 return 0;
2614
2615         mutex_lock(&graph_lock);
2616
2617         if (ftrace_graph_count >= FTRACE_GRAPH_MAX_FUNCS) {
2618                 ret = -EBUSY;
2619                 goto out;
2620         }
2621
2622         if (file->f_mode & FMODE_READ) {
2623                 struct seq_file *m = file->private_data;
2624                 array = m->private;
2625         } else
2626                 array = file->private_data;
2627
2628         ret = get_user(ch, ubuf++);
2629         if (ret)
2630                 goto out;
2631         read++;
2632         cnt--;
2633
2634         /* skip white space */
2635         while (cnt && isspace(ch)) {
2636                 ret = get_user(ch, ubuf++);
2637                 if (ret)
2638                         goto out;
2639                 read++;
2640                 cnt--;
2641         }
2642
2643         if (isspace(ch)) {
2644                 *ppos += read;
2645                 ret = read;
2646                 goto out;
2647         }
2648
2649         while (cnt && !isspace(ch)) {
2650                 if (index < FTRACE_BUFF_MAX)
2651                         buffer[index++] = ch;
2652                 else {
2653                         ret = -EINVAL;
2654                         goto out;
2655                 }
2656                 ret = get_user(ch, ubuf++);
2657                 if (ret)
2658                         goto out;
2659                 read++;
2660                 cnt--;
2661         }
2662         buffer[index] = 0;
2663
2664         /* we allow only one expression at a time */
2665         ret = ftrace_set_func(array, &ftrace_graph_count, buffer);
2666         if (ret)
2667                 goto out;
2668
2669         file->f_pos += read;
2670
2671         ret = read;
2672  out:
2673         mutex_unlock(&graph_lock);
2674
2675         return ret;
2676 }
2677
2678 static const struct file_operations ftrace_graph_fops = {
2679         .open           = ftrace_graph_open,
2680         .read           = seq_read,
2681         .write          = ftrace_graph_write,
2682         .release        = ftrace_graph_release,
2683 };
2684 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2685
2686 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
2687 {
2688
2689         trace_create_file("available_filter_functions", 0444,
2690                         d_tracer, NULL, &ftrace_avail_fops);
2691
2692         trace_create_file("failures", 0444,
2693                         d_tracer, NULL, &ftrace_failures_fops);
2694
2695         trace_create_file("set_ftrace_filter", 0644, d_tracer,
2696                         NULL, &ftrace_filter_fops);
2697
2698         trace_create_file("set_ftrace_notrace", 0644, d_tracer,
2699                                     NULL, &ftrace_notrace_fops);
2700
2701 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2702         trace_create_file("set_graph_function", 0444, d_tracer,
2703                                     NULL,
2704                                     &ftrace_graph_fops);
2705 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2706
2707         return 0;
2708 }
2709
2710 static int ftrace_convert_nops(struct module *mod,
2711                                unsigned long *start,
2712                                unsigned long *end)
2713 {
2714         unsigned long *p;
2715         unsigned long addr;
2716         unsigned long flags;
2717
2718         mutex_lock(&ftrace_lock);
2719         p = start;
2720         while (p < end) {
2721                 addr = ftrace_call_adjust(*p++);
2722                 /*
2723                  * Some architecture linkers will pad between
2724                  * the different mcount_loc sections of different
2725                  * object files to satisfy alignments.
2726                  * Skip any NULL pointers.
2727                  */
2728                 if (!addr)
2729                         continue;
2730                 ftrace_record_ip(addr);
2731         }
2732
2733         /* disable interrupts to prevent kstop machine */
2734         local_irq_save(flags);
2735         ftrace_update_code(mod);
2736         local_irq_restore(flags);
2737         mutex_unlock(&ftrace_lock);
2738
2739         return 0;
2740 }
2741
2742 #ifdef CONFIG_MODULES
2743 void ftrace_release(void *start, void *end)
2744 {
2745         struct dyn_ftrace *rec;
2746         struct ftrace_page *pg;
2747         unsigned long s = (unsigned long)start;
2748         unsigned long e = (unsigned long)end;
2749
2750         if (ftrace_disabled || !start || start == end)
2751                 return;
2752
2753         mutex_lock(&ftrace_lock);
2754         do_for_each_ftrace_rec(pg, rec) {
2755                 if ((rec->ip >= s) && (rec->ip < e)) {
2756                         /*
2757                          * rec->ip is changed in ftrace_free_rec()
2758                          * It should not between s and e if record was freed.
2759                          */
2760                         FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
2761                         ftrace_free_rec(rec);
2762                 }
2763         } while_for_each_ftrace_rec();
2764         mutex_unlock(&ftrace_lock);
2765 }
2766
2767 static void ftrace_init_module(struct module *mod,
2768                                unsigned long *start, unsigned long *end)
2769 {
2770         if (ftrace_disabled || start == end)
2771                 return;
2772         ftrace_convert_nops(mod, start, end);
2773 }
2774
2775 static int ftrace_module_notify(struct notifier_block *self,
2776                                 unsigned long val, void *data)
2777 {
2778         struct module *mod = data;
2779
2780         switch (val) {
2781         case MODULE_STATE_COMING:
2782                 ftrace_init_module(mod, mod->ftrace_callsites,
2783                                    mod->ftrace_callsites +
2784                                    mod->num_ftrace_callsites);
2785                 break;
2786         case MODULE_STATE_GOING:
2787                 ftrace_release(mod->ftrace_callsites,
2788                                mod->ftrace_callsites +
2789                                mod->num_ftrace_callsites);
2790                 break;
2791         }
2792
2793         return 0;
2794 }
2795 #else
2796 static int ftrace_module_notify(struct notifier_block *self,
2797                                 unsigned long val, void *data)
2798 {
2799         return 0;
2800 }
2801 #endif /* CONFIG_MODULES */
2802
2803 struct notifier_block ftrace_module_nb = {
2804         .notifier_call = ftrace_module_notify,
2805         .priority = 0,
2806 };
2807
2808 extern unsigned long __start_mcount_loc[];
2809 extern unsigned long __stop_mcount_loc[];
2810
2811 void __init ftrace_init(void)
2812 {
2813         unsigned long count, addr, flags;
2814         int ret;
2815
2816         /* Keep the ftrace pointer to the stub */
2817         addr = (unsigned long)ftrace_stub;
2818
2819         local_irq_save(flags);
2820         ftrace_dyn_arch_init(&addr);
2821         local_irq_restore(flags);
2822
2823         /* ftrace_dyn_arch_init places the return code in addr */
2824         if (addr)
2825                 goto failed;
2826
2827         count = __stop_mcount_loc - __start_mcount_loc;
2828
2829         ret = ftrace_dyn_table_alloc(count);
2830         if (ret)
2831                 goto failed;
2832
2833         last_ftrace_enabled = ftrace_enabled = 1;
2834
2835         ret = ftrace_convert_nops(NULL,
2836                                   __start_mcount_loc,
2837                                   __stop_mcount_loc);
2838
2839         ret = register_module_notifier(&ftrace_module_nb);
2840         if (ret)
2841                 pr_warning("Failed to register trace ftrace module notifier\n");
2842
2843         set_ftrace_early_filters();
2844
2845         return;
2846  failed:
2847         ftrace_disabled = 1;
2848 }
2849
2850 #else
2851
2852 static int __init ftrace_nodyn_init(void)
2853 {
2854         ftrace_enabled = 1;
2855         return 0;
2856 }
2857 device_initcall(ftrace_nodyn_init);
2858
2859 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
2860 static inline void ftrace_startup_enable(int command) { }
2861 /* Keep as macros so we do not need to define the commands */
2862 # define ftrace_startup(command)        do { } while (0)
2863 # define ftrace_shutdown(command)       do { } while (0)
2864 # define ftrace_startup_sysctl()        do { } while (0)
2865 # define ftrace_shutdown_sysctl()       do { } while (0)
2866 #endif /* CONFIG_DYNAMIC_FTRACE */
2867
2868 static ssize_t
2869 ftrace_pid_read(struct file *file, char __user *ubuf,
2870                        size_t cnt, loff_t *ppos)
2871 {
2872         char buf[64];
2873         int r;
2874
2875         if (ftrace_pid_trace == ftrace_swapper_pid)
2876                 r = sprintf(buf, "swapper tasks\n");
2877         else if (ftrace_pid_trace)
2878                 r = sprintf(buf, "%u\n", pid_vnr(ftrace_pid_trace));
2879         else
2880                 r = sprintf(buf, "no pid\n");
2881
2882         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2883 }
2884
2885 static void clear_ftrace_swapper(void)
2886 {
2887         struct task_struct *p;
2888         int cpu;
2889
2890         get_online_cpus();
2891         for_each_online_cpu(cpu) {
2892                 p = idle_task(cpu);
2893                 clear_tsk_trace_trace(p);
2894         }
2895         put_online_cpus();
2896 }
2897
2898 static void set_ftrace_swapper(void)
2899 {
2900         struct task_struct *p;
2901         int cpu;
2902
2903         get_online_cpus();
2904         for_each_online_cpu(cpu) {
2905                 p = idle_task(cpu);
2906                 set_tsk_trace_trace(p);
2907         }
2908         put_online_cpus();
2909 }
2910
2911 static void clear_ftrace_pid(struct pid *pid)
2912 {
2913         struct task_struct *p;
2914
2915         rcu_read_lock();
2916         do_each_pid_task(pid, PIDTYPE_PID, p) {
2917                 clear_tsk_trace_trace(p);
2918         } while_each_pid_task(pid, PIDTYPE_PID, p);
2919         rcu_read_unlock();
2920
2921         put_pid(pid);
2922 }
2923
2924 static void set_ftrace_pid(struct pid *pid)
2925 {
2926         struct task_struct *p;
2927
2928         rcu_read_lock();
2929         do_each_pid_task(pid, PIDTYPE_PID, p) {
2930                 set_tsk_trace_trace(p);
2931         } while_each_pid_task(pid, PIDTYPE_PID, p);
2932         rcu_read_unlock();
2933 }
2934
2935 static void clear_ftrace_pid_task(struct pid **pid)
2936 {
2937         if (*pid == ftrace_swapper_pid)
2938                 clear_ftrace_swapper();
2939         else
2940                 clear_ftrace_pid(*pid);
2941
2942         *pid = NULL;
2943 }
2944
2945 static void set_ftrace_pid_task(struct pid *pid)
2946 {
2947         if (pid == ftrace_swapper_pid)
2948                 set_ftrace_swapper();
2949         else
2950                 set_ftrace_pid(pid);
2951 }
2952
2953 static ssize_t
2954 ftrace_pid_write(struct file *filp, const char __user *ubuf,
2955                    size_t cnt, loff_t *ppos)
2956 {
2957         struct pid *pid;
2958         char buf[64];
2959         long val;
2960         int ret;
2961
2962         if (cnt >= sizeof(buf))
2963                 return -EINVAL;
2964
2965         if (copy_from_user(&buf, ubuf, cnt))
2966                 return -EFAULT;
2967
2968         buf[cnt] = 0;
2969
2970         ret = strict_strtol(buf, 10, &val);
2971         if (ret < 0)
2972                 return ret;
2973
2974         mutex_lock(&ftrace_lock);
2975         if (val < 0) {
2976                 /* disable pid tracing */
2977                 if (!ftrace_pid_trace)
2978                         goto out;
2979
2980                 clear_ftrace_pid_task(&ftrace_pid_trace);
2981
2982         } else {
2983                 /* swapper task is special */
2984                 if (!val) {
2985                         pid = ftrace_swapper_pid;
2986                         if (pid == ftrace_pid_trace)
2987                                 goto out;
2988                 } else {
2989                         pid = find_get_pid(val);
2990
2991                         if (pid == ftrace_pid_trace) {
2992                                 put_pid(pid);
2993                                 goto out;
2994                         }
2995                 }
2996
2997                 if (ftrace_pid_trace)
2998                         clear_ftrace_pid_task(&ftrace_pid_trace);
2999
3000                 if (!pid)
3001                         goto out;
3002
3003                 ftrace_pid_trace = pid;
3004
3005                 set_ftrace_pid_task(ftrace_pid_trace);
3006         }
3007
3008         /* update the function call */
3009         ftrace_update_pid_func();
3010         ftrace_startup_enable(0);
3011
3012  out:
3013         mutex_unlock(&ftrace_lock);
3014
3015         return cnt;
3016 }
3017
3018 static const struct file_operations ftrace_pid_fops = {
3019         .read = ftrace_pid_read,
3020         .write = ftrace_pid_write,
3021 };
3022
3023 static __init int ftrace_init_debugfs(void)
3024 {
3025         struct dentry *d_tracer;
3026
3027         d_tracer = tracing_init_dentry();
3028         if (!d_tracer)
3029                 return 0;
3030
3031         ftrace_init_dyn_debugfs(d_tracer);
3032
3033         trace_create_file("set_ftrace_pid", 0644, d_tracer,
3034                             NULL, &ftrace_pid_fops);
3035
3036         ftrace_profile_debugfs(d_tracer);
3037
3038         return 0;
3039 }
3040 fs_initcall(ftrace_init_debugfs);
3041
3042 /**
3043  * ftrace_kill - kill ftrace
3044  *
3045  * This function should be used by panic code. It stops ftrace
3046  * but in a not so nice way. If you need to simply kill ftrace
3047  * from a non-atomic section, use ftrace_kill.
3048  */
3049 void ftrace_kill(void)
3050 {
3051         ftrace_disabled = 1;
3052         ftrace_enabled = 0;
3053         clear_ftrace_function();
3054 }
3055
3056 /**
3057  * register_ftrace_function - register a function for profiling
3058  * @ops - ops structure that holds the function for profiling.
3059  *
3060  * Register a function to be called by all functions in the
3061  * kernel.
3062  *
3063  * Note: @ops->func and all the functions it calls must be labeled
3064  *       with "notrace", otherwise it will go into a
3065  *       recursive loop.
3066  */
3067 int register_ftrace_function(struct ftrace_ops *ops)
3068 {
3069         int ret;
3070
3071         if (unlikely(ftrace_disabled))
3072                 return -1;
3073
3074         mutex_lock(&ftrace_lock);
3075
3076         ret = __register_ftrace_function(ops);
3077         ftrace_startup(0);
3078
3079         mutex_unlock(&ftrace_lock);
3080         return ret;
3081 }
3082
3083 /**
3084  * unregister_ftrace_function - unregister a function for profiling.
3085  * @ops - ops structure that holds the function to unregister
3086  *
3087  * Unregister a function that was added to be called by ftrace profiling.
3088  */
3089 int unregister_ftrace_function(struct ftrace_ops *ops)
3090 {
3091         int ret;
3092
3093         mutex_lock(&ftrace_lock);
3094         ret = __unregister_ftrace_function(ops);
3095         ftrace_shutdown(0);
3096         mutex_unlock(&ftrace_lock);
3097
3098         return ret;
3099 }
3100
3101 int
3102 ftrace_enable_sysctl(struct ctl_table *table, int write,
3103                      struct file *file, void __user *buffer, size_t *lenp,
3104                      loff_t *ppos)
3105 {
3106         int ret;
3107
3108         if (unlikely(ftrace_disabled))
3109                 return -ENODEV;
3110
3111         mutex_lock(&ftrace_lock);
3112
3113         ret  = proc_dointvec(table, write, file, buffer, lenp, ppos);
3114
3115         if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
3116                 goto out;
3117
3118         last_ftrace_enabled = !!ftrace_enabled;
3119
3120         if (ftrace_enabled) {
3121
3122                 ftrace_startup_sysctl();
3123
3124                 /* we are starting ftrace again */
3125                 if (ftrace_list != &ftrace_list_end) {
3126                         if (ftrace_list->next == &ftrace_list_end)
3127                                 ftrace_trace_function = ftrace_list->func;
3128                         else
3129                                 ftrace_trace_function = ftrace_list_func;
3130                 }
3131
3132         } else {
3133                 /* stopping ftrace calls (just send to ftrace_stub) */
3134                 ftrace_trace_function = ftrace_stub;
3135
3136                 ftrace_shutdown_sysctl();
3137         }
3138
3139  out:
3140         mutex_unlock(&ftrace_lock);
3141         return ret;
3142 }
3143
3144 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3145
3146 static int ftrace_graph_active;
3147 static struct notifier_block ftrace_suspend_notifier;
3148
3149 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
3150 {
3151         return 0;
3152 }
3153
3154 /* The callbacks that hook a function */
3155 trace_func_graph_ret_t ftrace_graph_return =
3156                         (trace_func_graph_ret_t)ftrace_stub;
3157 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
3158
3159 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
3160 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
3161 {
3162         int i;
3163         int ret = 0;
3164         unsigned long flags;
3165         int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
3166         struct task_struct *g, *t;
3167
3168         for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
3169                 ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
3170                                         * sizeof(struct ftrace_ret_stack),
3171                                         GFP_KERNEL);
3172                 if (!ret_stack_list[i]) {
3173                         start = 0;
3174                         end = i;
3175                         ret = -ENOMEM;
3176                         goto free;
3177                 }
3178         }
3179
3180         read_lock_irqsave(&tasklist_lock, flags);
3181         do_each_thread(g, t) {
3182                 if (start == end) {
3183                         ret = -EAGAIN;
3184                         goto unlock;
3185                 }
3186
3187                 if (t->ret_stack == NULL) {
3188                         atomic_set(&t->tracing_graph_pause, 0);
3189                         atomic_set(&t->trace_overrun, 0);
3190                         t->curr_ret_stack = -1;
3191                         /* Make sure the tasks see the -1 first: */
3192                         smp_wmb();
3193                         t->ret_stack = ret_stack_list[start++];
3194                 }
3195         } while_each_thread(g, t);
3196
3197 unlock:
3198         read_unlock_irqrestore(&tasklist_lock, flags);
3199 free:
3200         for (i = start; i < end; i++)
3201                 kfree(ret_stack_list[i]);
3202         return ret;
3203 }
3204
3205 static void
3206 ftrace_graph_probe_sched_switch(struct rq *__rq, struct task_struct *prev,
3207                                 struct task_struct *next)
3208 {
3209         unsigned long long timestamp;
3210         int index;
3211
3212         /*
3213          * Does the user want to count the time a function was asleep.
3214          * If so, do not update the time stamps.
3215          */
3216         if (trace_flags & TRACE_ITER_SLEEP_TIME)
3217                 return;
3218
3219         timestamp = trace_clock_local();
3220
3221         prev->ftrace_timestamp = timestamp;
3222
3223         /* only process tasks that we timestamped */
3224         if (!next->ftrace_timestamp)
3225                 return;
3226
3227         /*
3228          * Update all the counters in next to make up for the
3229          * time next was sleeping.
3230          */
3231         timestamp -= next->ftrace_timestamp;
3232
3233         for (index = next->curr_ret_stack; index >= 0; index--)
3234                 next->ret_stack[index].calltime += timestamp;
3235 }
3236
3237 /* Allocate a return stack for each task */
3238 static int start_graph_tracing(void)
3239 {
3240         struct ftrace_ret_stack **ret_stack_list;
3241         int ret, cpu;
3242
3243         ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
3244                                 sizeof(struct ftrace_ret_stack *),
3245                                 GFP_KERNEL);
3246
3247         if (!ret_stack_list)
3248                 return -ENOMEM;
3249
3250         /* The cpu_boot init_task->ret_stack will never be freed */
3251         for_each_online_cpu(cpu) {
3252                 if (!idle_task(cpu)->ret_stack)
3253                         ftrace_graph_init_task(idle_task(cpu));
3254         }
3255
3256         do {
3257                 ret = alloc_retstack_tasklist(ret_stack_list);
3258         } while (ret == -EAGAIN);
3259
3260         if (!ret) {
3261                 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch);
3262                 if (ret)
3263                         pr_info("ftrace_graph: Couldn't activate tracepoint"
3264                                 " probe to kernel_sched_switch\n");
3265         }
3266
3267         kfree(ret_stack_list);
3268         return ret;
3269 }
3270
3271 /*
3272  * Hibernation protection.
3273  * The state of the current task is too much unstable during
3274  * suspend/restore to disk. We want to protect against that.
3275  */
3276 static int
3277 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
3278                                                         void *unused)
3279 {
3280         switch (state) {
3281         case PM_HIBERNATION_PREPARE:
3282                 pause_graph_tracing();
3283                 break;
3284
3285         case PM_POST_HIBERNATION:
3286                 unpause_graph_tracing();
3287                 break;
3288         }
3289         return NOTIFY_DONE;
3290 }
3291
3292 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
3293                         trace_func_graph_ent_t entryfunc)
3294 {
3295         int ret = 0;
3296
3297         mutex_lock(&ftrace_lock);
3298
3299         /* we currently allow only one tracer registered at a time */
3300         if (ftrace_graph_active) {
3301                 ret = -EBUSY;
3302                 goto out;
3303         }
3304
3305         ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
3306         register_pm_notifier(&ftrace_suspend_notifier);
3307
3308         ftrace_graph_active++;
3309         ret = start_graph_tracing();
3310         if (ret) {
3311                 ftrace_graph_active--;
3312                 goto out;
3313         }
3314
3315         ftrace_graph_return = retfunc;
3316         ftrace_graph_entry = entryfunc;
3317
3318         ftrace_startup(FTRACE_START_FUNC_RET);
3319
3320 out:
3321         mutex_unlock(&ftrace_lock);
3322         return ret;
3323 }
3324
3325 void unregister_ftrace_graph(void)
3326 {
3327         mutex_lock(&ftrace_lock);
3328
3329         if (unlikely(!ftrace_graph_active))
3330                 goto out;
3331
3332         ftrace_graph_active--;
3333         unregister_trace_sched_switch(ftrace_graph_probe_sched_switch);
3334         ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
3335         ftrace_graph_entry = ftrace_graph_entry_stub;
3336         ftrace_shutdown(FTRACE_STOP_FUNC_RET);
3337         unregister_pm_notifier(&ftrace_suspend_notifier);
3338
3339  out:
3340         mutex_unlock(&ftrace_lock);
3341 }
3342
3343 /* Allocate a return stack for newly created task */
3344 void ftrace_graph_init_task(struct task_struct *t)
3345 {
3346         /* Make sure we do not use the parent ret_stack */
3347         t->ret_stack = NULL;
3348
3349         if (ftrace_graph_active) {
3350                 struct ftrace_ret_stack *ret_stack;
3351
3352                 ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
3353                                 * sizeof(struct ftrace_ret_stack),
3354                                 GFP_KERNEL);
3355                 if (!ret_stack)
3356                         return;
3357                 t->curr_ret_stack = -1;
3358                 atomic_set(&t->tracing_graph_pause, 0);
3359                 atomic_set(&t->trace_overrun, 0);
3360                 t->ftrace_timestamp = 0;
3361                 /* make curr_ret_stack visable before we add the ret_stack */
3362                 smp_wmb();
3363                 t->ret_stack = ret_stack;
3364         }
3365 }
3366
3367 void ftrace_graph_exit_task(struct task_struct *t)
3368 {
3369         struct ftrace_ret_stack *ret_stack = t->ret_stack;
3370
3371         t->ret_stack = NULL;
3372         /* NULL must become visible to IRQs before we free it: */
3373         barrier();
3374
3375         kfree(ret_stack);
3376 }
3377
3378 void ftrace_graph_stop(void)
3379 {
3380         ftrace_stop();
3381 }
3382 #endif
3383