kernel/trace/ftrace.c

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