kernel/trace/ftrace.c

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