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/debugfs.h>
  21 #include <linux/kthread.h>
  22 #include <linux/hardirq.h>
  23 #include <linux/ftrace.h>
  24 #include <linux/uaccess.h>
  25 #include <linux/sysctl.h>
  26 #include <linux/hash.h>
  27 #include <linux/ctype.h>
  28 #include <linux/list.h>
  29
  30 #include "trace.h"
  31
  32 /* ftrace_enabled is a method to turn ftrace on or off */
  33 int ftrace_enabled __read_mostly;
  34 static int last_ftrace_enabled;
  35
  36 /*
  37  * ftrace_disabled is set when an anomaly is discovered.
  38  * ftrace_disabled is much stronger than ftrace_enabled.
  39  */
  40 static int ftrace_disabled __read_mostly;
  41
  42 static DEFINE_SPINLOCK(ftrace_lock);
  43 static DEFINE_MUTEX(ftrace_sysctl_lock);
  44
  45 static struct ftrace_ops ftrace_list_end __read_mostly =
  46 {
  47         .func = ftrace_stub,
  48 };
  49
  50 static struct ftrace_ops *ftrace_list __read_mostly = &ftrace_list_end;
  51 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
  52
  53 /* mcount is defined per arch in assembly */
  54 EXPORT_SYMBOL(mcount);
  55
  56 void ftrace_list_func(unsigned long ip, unsigned long parent_ip)
  57 {
  58         struct ftrace_ops *op = ftrace_list;
  59
  60         /* in case someone actually ports this to alpha! */
  61         read_barrier_depends();
  62
  63         while (op != &ftrace_list_end) {
  64                 /* silly alpha */
  65                 read_barrier_depends();
  66                 op->func(ip, parent_ip);
  67                 op = op->next;
  68         };
  69 }
  70
  71 /**
  72  * clear_ftrace_function - reset the ftrace function
  73  *
  74  * This NULLs the ftrace function and in essence stops
  75  * tracing.  There may be lag
  76  */
  77 void clear_ftrace_function(void)
  78 {
  79         ftrace_trace_function = ftrace_stub;
  80 }
  81
  82 static int __register_ftrace_function(struct ftrace_ops *ops)
  83 {
  84         /* Should never be called by interrupts */
  85         spin_lock(&ftrace_lock);
  86
  87         ops->next = ftrace_list;
  88         /*
  89          * We are entering ops into the ftrace_list but another
  90          * CPU might be walking that list. We need to make sure
  91          * the ops->next pointer is valid before another CPU sees
  92          * the ops pointer included into the ftrace_list.
  93          */
  94         smp_wmb();
  95         ftrace_list = ops;
  96
  97         if (ftrace_enabled) {
  98                 /*
  99                  * For one func, simply call it directly.
 100                  * For more than one func, call the chain.
 101                  */
 102                 if (ops->next == &ftrace_list_end)
 103                         ftrace_trace_function = ops->func;
 104                 else
 105                         ftrace_trace_function = ftrace_list_func;
 106         }
 107
 108         spin_unlock(&ftrace_lock);
 109
 110         return 0;
 111 }
 112
 113 static int __unregister_ftrace_function(struct ftrace_ops *ops)
 114 {
 115         struct ftrace_ops **p;
 116         int ret = 0;
 117
 118         spin_lock(&ftrace_lock);
 119
 120         /*
 121          * If we are removing the last function, then simply point
 122          * to the ftrace_stub.
 123          */
 124         if (ftrace_list == ops && ops->next == &ftrace_list_end) {
 125                 ftrace_trace_function = ftrace_stub;
 126                 ftrace_list = &ftrace_list_end;
 127                 goto out;
 128         }
 129
 130         for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next)
 131                 if (*p == ops)
 132                         break;
 133
 134         if (*p != ops) {
 135                 ret = -1;
 136                 goto out;
 137         }
 138
 139         *p = (*p)->next;
 140
 141         if (ftrace_enabled) {
 142                 /* If we only have one func left, then call that directly */
 143                 if (ftrace_list == &ftrace_list_end ||
 144                     ftrace_list->next == &ftrace_list_end)
 145                         ftrace_trace_function = ftrace_list->func;
 146         }
 147
 148  out:
 149         spin_unlock(&ftrace_lock);
 150
 151         return ret;
 152 }
 153
 154 #ifdef CONFIG_DYNAMIC_FTRACE
 155
 156 static struct task_struct *ftraced_task;
 157 static DECLARE_WAIT_QUEUE_HEAD(ftraced_waiters);
 158 static unsigned long ftraced_iteration_counter;
 159
 160 enum {
 161         FTRACE_ENABLE_CALLS             = (1 << 0),
 162         FTRACE_DISABLE_CALLS            = (1 << 1),
 163         FTRACE_UPDATE_TRACE_FUNC        = (1 << 2),
 164         FTRACE_ENABLE_MCOUNT            = (1 << 3),
 165         FTRACE_DISABLE_MCOUNT           = (1 << 4),
 166 };
 167
 168 static int ftrace_filtered;
 169
 170 static struct hlist_head ftrace_hash[FTRACE_HASHSIZE];
 171
 172 static DEFINE_PER_CPU(int, ftrace_shutdown_disable_cpu);
 173
 174 static DEFINE_SPINLOCK(ftrace_shutdown_lock);
 175 static DEFINE_MUTEX(ftraced_lock);
 176 static DEFINE_MUTEX(ftrace_filter_lock);
 177
 178 struct ftrace_page {
 179         struct ftrace_page      *next;
 180         int                     index;
 181         struct dyn_ftrace       records[];
 182 } __attribute__((packed));
 183
 184 #define ENTRIES_PER_PAGE \
 185   ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
 186
 187 /* estimate from running different kernels */
 188 #define NR_TO_INIT              10000
 189
 190 static struct ftrace_page       *ftrace_pages_start;
 191 static struct ftrace_page       *ftrace_pages;
 192
 193 static int ftraced_trigger;
 194 static int ftraced_suspend;
 195
 196 static int ftrace_record_suspend;
 197
 198 static struct dyn_ftrace *ftrace_free_records;
 199
 200 static inline int
 201 ftrace_ip_in_hash(unsigned long ip, unsigned long key)
 202 {
 203         struct dyn_ftrace *p;
 204         struct hlist_node *t;
 205         int found = 0;
 206
 207         hlist_for_each_entry(p, t, &ftrace_hash[key], node) {
 208                 if (p->ip == ip) {
 209                         found = 1;
 210                         break;
 211                 }
 212         }
 213
 214         return found;
 215 }
 216
 217 static inline void
 218 ftrace_add_hash(struct dyn_ftrace *node, unsigned long key)
 219 {
 220         hlist_add_head(&node->node, &ftrace_hash[key]);
 221 }
 222
 223 static void ftrace_free_rec(struct dyn_ftrace *rec)
 224 {
 225         /* no locking, only called from kstop_machine */
 226
 227         rec->ip = (unsigned long)ftrace_free_records;
 228         ftrace_free_records = rec;
 229         rec->flags |= FTRACE_FL_FREE;
 230 }
 231
 232 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
 233 {
 234         struct dyn_ftrace *rec;
 235
 236         /* First check for freed records */
 237         if (ftrace_free_records) {
 238                 rec = ftrace_free_records;
 239
 240                 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
 241                         WARN_ON_ONCE(1);
 242                         ftrace_free_records = NULL;
 243                         ftrace_disabled = 1;
 244                         ftrace_enabled = 0;
 245                         return NULL;
 246                 }
 247
 248                 ftrace_free_records = (void *)rec->ip;
 249                 memset(rec, 0, sizeof(*rec));
 250                 return rec;
 251         }
 252
 253         if (ftrace_pages->index == ENTRIES_PER_PAGE) {
 254                 if (!ftrace_pages->next)
 255                         return NULL;
 256                 ftrace_pages = ftrace_pages->next;
 257         }
 258
 259         return &ftrace_pages->records[ftrace_pages->index++];
 260 }
 261
 262 static void
 263 ftrace_record_ip(unsigned long ip)
 264 {
 265         struct dyn_ftrace *node;
 266         unsigned long flags;
 267         unsigned long key;
 268         int resched;
 269         int atomic;
 270
 271         if (!ftrace_enabled || ftrace_disabled)
 272                 return;
 273
 274         resched = need_resched();
 275         preempt_disable_notrace();
 276
 277         /* We simply need to protect against recursion */
 278         __get_cpu_var(ftrace_shutdown_disable_cpu)++;
 279         if (__get_cpu_var(ftrace_shutdown_disable_cpu) != 1)
 280                 goto out;
 281
 282         if (unlikely(ftrace_record_suspend))
 283                 goto out;
 284
 285         key = hash_long(ip, FTRACE_HASHBITS);
 286
 287         WARN_ON_ONCE(key >= FTRACE_HASHSIZE);
 288
 289         if (ftrace_ip_in_hash(ip, key))
 290                 goto out;
 291
 292         atomic = irqs_disabled();
 293
 294         spin_lock_irqsave(&ftrace_shutdown_lock, flags);
 295
 296         /* This ip may have hit the hash before the lock */
 297         if (ftrace_ip_in_hash(ip, key))
 298                 goto out_unlock;
 299
 300         /*
 301          * There's a slight race that the ftraced will update the
 302          * hash and reset here. If it is already converted, skip it.
 303          */
 304         if (ftrace_ip_converted(ip))
 305                 goto out_unlock;
 306
 307         node = ftrace_alloc_dyn_node(ip);
 308         if (!node)
 309                 goto out_unlock;
 310
 311         node->ip = ip;
 312
 313         ftrace_add_hash(node, key);
 314
 315         ftraced_trigger = 1;
 316
 317  out_unlock:
 318         spin_unlock_irqrestore(&ftrace_shutdown_lock, flags);
 319  out:
 320         __get_cpu_var(ftrace_shutdown_disable_cpu)--;
 321
 322         /* prevent recursion with scheduler */
 323         if (resched)
 324                 preempt_enable_no_resched_notrace();
 325         else
 326                 preempt_enable_notrace();
 327 }
 328
 329 #define FTRACE_ADDR ((long)(ftrace_caller))
 330 #define MCOUNT_ADDR ((long)(mcount))
 331
 332 static void
 333 __ftrace_replace_code(struct dyn_ftrace *rec,
 334                       unsigned char *old, unsigned char *new, int enable)
 335 {
 336         unsigned long ip;
 337         int failed;
 338
 339         ip = rec->ip;
 340
 341         if (ftrace_filtered && enable) {
 342                 unsigned long fl;
 343                 /*
 344                  * If filtering is on:
 345                  *
 346                  * If this record is set to be filtered and
 347                  * is enabled then do nothing.
 348                  *
 349                  * If this record is set to be filtered and
 350                  * it is not enabled, enable it.
 351                  *
 352                  * If this record is not set to be filtered
 353                  * and it is not enabled do nothing.
 354                  *
 355                  * If this record is not set to be filtered and
 356                  * it is enabled, disable it.
 357                  */
 358                 fl = rec->flags & (FTRACE_FL_FILTER | FTRACE_FL_ENABLED);
 359
 360                 if ((fl ==  (FTRACE_FL_FILTER | FTRACE_FL_ENABLED)) ||
 361                     (fl == 0))
 362                         return;
 363
 364                 /*
 365                  * If it is enabled disable it,
 366                  * otherwise enable it!
 367                  */
 368                 if (fl == FTRACE_FL_ENABLED) {
 369                         /* swap new and old */
 370                         new = old;
 371                         old = ftrace_call_replace(ip, FTRACE_ADDR);
 372                         rec->flags &= ~FTRACE_FL_ENABLED;
 373                 } else {
 374                         new = ftrace_call_replace(ip, FTRACE_ADDR);
 375                         rec->flags |= FTRACE_FL_ENABLED;
 376                 }
 377         } else {
 378
 379                 if (enable)
 380                         new = ftrace_call_replace(ip, FTRACE_ADDR);
 381                 else
 382                         old = ftrace_call_replace(ip, FTRACE_ADDR);
 383
 384                 if (enable) {
 385                         if (rec->flags & FTRACE_FL_ENABLED)
 386                                 return;
 387                         rec->flags |= FTRACE_FL_ENABLED;
 388                 } else {
 389                         if (!(rec->flags & FTRACE_FL_ENABLED))
 390                                 return;
 391                         rec->flags &= ~FTRACE_FL_ENABLED;
 392                 }
 393         }
 394
 395         failed = ftrace_modify_code(ip, old, new);
 396         if (failed) {
 397                 unsigned long key;
 398                 /* It is possible that the function hasn't been converted yet */
 399                 key = hash_long(ip, FTRACE_HASHBITS);
 400                 if (!ftrace_ip_in_hash(ip, key)) {
 401                         rec->flags |= FTRACE_FL_FAILED;
 402                         ftrace_free_rec(rec);
 403                 }
 404
 405         }
 406 }
 407
 408 static void ftrace_replace_code(int enable)
 409 {
 410         unsigned char *new = NULL, *old = NULL;
 411         struct dyn_ftrace *rec;
 412         struct ftrace_page *pg;
 413         int i;
 414
 415         if (enable)
 416                 old = ftrace_nop_replace();
 417         else
 418                 new = ftrace_nop_replace();
 419
 420         for (pg = ftrace_pages_start; pg; pg = pg->next) {
 421                 for (i = 0; i < pg->index; i++) {
 422                         rec = &pg->records[i];
 423
 424                         /* don't modify code that has already faulted */
 425                         if (rec->flags & FTRACE_FL_FAILED)
 426                                 continue;
 427
 428                         __ftrace_replace_code(rec, old, new, enable);
 429                 }
 430         }
 431 }
 432
 433 static void ftrace_shutdown_replenish(void)
 434 {
 435         if (ftrace_pages->next)
 436                 return;
 437
 438         /* allocate another page */
 439         ftrace_pages->next = (void *)get_zeroed_page(GFP_KERNEL);
 440 }
 441
 442 static void
 443 ftrace_code_disable(struct dyn_ftrace *rec)
 444 {
 445         unsigned long ip;
 446         unsigned char *nop, *call;
 447         int failed;
 448
 449         ip = rec->ip;
 450
 451         nop = ftrace_nop_replace();
 452         call = ftrace_call_replace(ip, MCOUNT_ADDR);
 453
 454         failed = ftrace_modify_code(ip, call, nop);
 455         if (failed) {
 456                 rec->flags |= FTRACE_FL_FAILED;
 457                 ftrace_free_rec(rec);
 458         }
 459 }
 460
 461 static int __ftrace_modify_code(void *data)
 462 {
 463         unsigned long addr;
 464         int *command = data;
 465
 466         if (*command & FTRACE_ENABLE_CALLS)
 467                 ftrace_replace_code(1);
 468         else if (*command & FTRACE_DISABLE_CALLS)
 469                 ftrace_replace_code(0);
 470
 471         if (*command & FTRACE_UPDATE_TRACE_FUNC)
 472                 ftrace_update_ftrace_func(ftrace_trace_function);
 473
 474         if (*command & FTRACE_ENABLE_MCOUNT) {
 475                 addr = (unsigned long)ftrace_record_ip;
 476                 ftrace_mcount_set(&addr);
 477         } else if (*command & FTRACE_DISABLE_MCOUNT) {
 478                 addr = (unsigned long)ftrace_stub;
 479                 ftrace_mcount_set(&addr);
 480         }
 481
 482         return 0;
 483 }
 484
 485 static void ftrace_run_update_code(int command)
 486 {
 487         stop_machine_run(__ftrace_modify_code, &command, NR_CPUS);
 488 }
 489
 490 static ftrace_func_t saved_ftrace_func;
 491
 492 static void ftrace_startup(void)
 493 {
 494         int command = 0;
 495
 496         if (unlikely(ftrace_disabled))
 497                 return;
 498
 499         mutex_lock(&ftraced_lock);
 500         ftraced_suspend++;
 501         if (ftraced_suspend == 1)
 502                 command |= FTRACE_ENABLE_CALLS;
 503
 504         if (saved_ftrace_func != ftrace_trace_function) {
 505                 saved_ftrace_func = ftrace_trace_function;
 506                 command |= FTRACE_UPDATE_TRACE_FUNC;
 507         }
 508
 509         if (!command || !ftrace_enabled)
 510                 goto out;
 511
 512         ftrace_run_update_code(command);
 513  out:
 514         mutex_unlock(&ftraced_lock);
 515 }
 516
 517 static void ftrace_shutdown(void)
 518 {
 519         int command = 0;
 520
 521         if (unlikely(ftrace_disabled))
 522                 return;
 523
 524         mutex_lock(&ftraced_lock);
 525         ftraced_suspend--;
 526         if (!ftraced_suspend)
 527                 command |= FTRACE_DISABLE_CALLS;
 528
 529         if (saved_ftrace_func != ftrace_trace_function) {
 530                 saved_ftrace_func = ftrace_trace_function;
 531                 command |= FTRACE_UPDATE_TRACE_FUNC;
 532         }
 533
 534         if (!command || !ftrace_enabled)
 535                 goto out;
 536
 537         ftrace_run_update_code(command);
 538  out:
 539         mutex_unlock(&ftraced_lock);
 540 }
 541
 542 static void ftrace_startup_sysctl(void)
 543 {
 544         int command = FTRACE_ENABLE_MCOUNT;
 545
 546         if (unlikely(ftrace_disabled))
 547                 return;
 548
 549         mutex_lock(&ftraced_lock);
 550         /* Force update next time */
 551         saved_ftrace_func = NULL;
 552         /* ftraced_suspend is true if we want ftrace running */
 553         if (ftraced_suspend)
 554                 command |= FTRACE_ENABLE_CALLS;
 555
 556         ftrace_run_update_code(command);
 557         mutex_unlock(&ftraced_lock);
 558 }
 559
 560 static void ftrace_shutdown_sysctl(void)
 561 {
 562         int command = FTRACE_DISABLE_MCOUNT;
 563
 564         if (unlikely(ftrace_disabled))
 565                 return;
 566
 567         mutex_lock(&ftraced_lock);
 568         /* ftraced_suspend is true if ftrace is running */
 569         if (ftraced_suspend)
 570                 command |= FTRACE_DISABLE_CALLS;
 571
 572         ftrace_run_update_code(command);
 573         mutex_unlock(&ftraced_lock);
 574 }
 575
 576 static cycle_t          ftrace_update_time;
 577 static unsigned long    ftrace_update_cnt;
 578 unsigned long           ftrace_update_tot_cnt;
 579
 580 static int __ftrace_update_code(void *ignore)
 581 {
 582         struct dyn_ftrace *p;
 583         struct hlist_head head;
 584         struct hlist_node *t;
 585         int save_ftrace_enabled;
 586         cycle_t start, stop;
 587         int i;
 588
 589         /* Don't be recording funcs now */
 590         save_ftrace_enabled = ftrace_enabled;
 591         ftrace_enabled = 0;
 592
 593         start = ftrace_now(raw_smp_processor_id());
 594         ftrace_update_cnt = 0;
 595
 596         /* No locks needed, the machine is stopped! */
 597         for (i = 0; i < FTRACE_HASHSIZE; i++) {
 598                 if (hlist_empty(&ftrace_hash[i]))
 599                         continue;
 600
 601                 head = ftrace_hash[i];
 602                 INIT_HLIST_HEAD(&ftrace_hash[i]);
 603
 604                 /* all CPUS are stopped, we are safe to modify code */
 605                 hlist_for_each_entry(p, t, &head, node) {
 606                         ftrace_code_disable(p);
 607                         ftrace_update_cnt++;
 608                 }
 609
 610         }
 611
 612         stop = ftrace_now(raw_smp_processor_id());
 613         ftrace_update_time = stop - start;
 614         ftrace_update_tot_cnt += ftrace_update_cnt;
 615
 616         ftrace_enabled = save_ftrace_enabled;
 617
 618         return 0;
 619 }
 620
 621 static void ftrace_update_code(void)
 622 {
 623         if (unlikely(ftrace_disabled))
 624                 return;
 625
 626         stop_machine_run(__ftrace_update_code, NULL, NR_CPUS);
 627 }
 628
 629 static int ftraced(void *ignore)
 630 {
 631         unsigned long usecs;
 632
 633         while (!kthread_should_stop()) {
 634
 635                 set_current_state(TASK_INTERRUPTIBLE);
 636
 637                 /* check once a second */
 638                 schedule_timeout(HZ);
 639
 640                 if (unlikely(ftrace_disabled))
 641                         continue;
 642
 643                 mutex_lock(&ftrace_sysctl_lock);
 644                 mutex_lock(&ftraced_lock);
 645                 if (ftrace_enabled && ftraced_trigger && !ftraced_suspend) {
 646                         ftrace_record_suspend++;
 647                         ftrace_update_code();
 648                         usecs = nsecs_to_usecs(ftrace_update_time);
 649                         if (ftrace_update_tot_cnt > 100000) {
 650                                 ftrace_update_tot_cnt = 0;
 651                                 pr_info("hm, dftrace overflow: %lu change%s"
 652                                          " (%lu total) in %lu usec%s\n",
 653                                         ftrace_update_cnt,
 654                                         ftrace_update_cnt != 1 ? "s" : "",
 655                                         ftrace_update_tot_cnt,
 656                                         usecs, usecs != 1 ? "s" : "");
 657                                 ftrace_disabled = 1;
 658                                 WARN_ON_ONCE(1);
 659                         }
 660                         ftraced_trigger = 0;
 661                         ftrace_record_suspend--;
 662                 }
 663                 ftraced_iteration_counter++;
 664                 mutex_unlock(&ftraced_lock);
 665                 mutex_unlock(&ftrace_sysctl_lock);
 666
 667                 wake_up_interruptible(&ftraced_waiters);
 668
 669                 ftrace_shutdown_replenish();
 670         }
 671         __set_current_state(TASK_RUNNING);
 672         return 0;
 673 }
 674
 675 static int __init ftrace_dyn_table_alloc(void)
 676 {
 677         struct ftrace_page *pg;
 678         int cnt;
 679         int i;
 680
 681         /* allocate a few pages */
 682         ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
 683         if (!ftrace_pages_start)
 684                 return -1;
 685
 686         /*
 687          * Allocate a few more pages.
 688          *
 689          * TODO: have some parser search vmlinux before
 690          *   final linking to find all calls to ftrace.
 691          *   Then we can:
 692          *    a) know how many pages to allocate.
 693          *     and/or
 694          *    b) set up the table then.
 695          *
 696          *  The dynamic code is still necessary for
 697          *  modules.
 698          */
 699
 700         pg = ftrace_pages = ftrace_pages_start;
 701
 702         cnt = NR_TO_INIT / ENTRIES_PER_PAGE;
 703
 704         for (i = 0; i < cnt; i++) {
 705                 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
 706
 707                 /* If we fail, we'll try later anyway */
 708                 if (!pg->next)
 709                         break;
 710
 711                 pg = pg->next;
 712         }
 713
 714         return 0;
 715 }
 716
 717 enum {
 718         FTRACE_ITER_FILTER      = (1 << 0),
 719         FTRACE_ITER_CONT        = (1 << 1),
 720 };
 721
 722 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
 723
 724 struct ftrace_iterator {
 725         loff_t                  pos;
 726         struct ftrace_page      *pg;
 727         unsigned                idx;
 728         unsigned                flags;
 729         unsigned char           buffer[FTRACE_BUFF_MAX+1];
 730         unsigned                buffer_idx;
 731         unsigned                filtered;
 732 };
 733
 734 static void *
 735 t_next(struct seq_file *m, void *v, loff_t *pos)
 736 {
 737         struct ftrace_iterator *iter = m->private;
 738         struct dyn_ftrace *rec = NULL;
 739
 740         (*pos)++;
 741
 742  retry:
 743         if (iter->idx >= iter->pg->index) {
 744                 if (iter->pg->next) {
 745                         iter->pg = iter->pg->next;
 746                         iter->idx = 0;
 747                         goto retry;
 748                 }
 749         } else {
 750                 rec = &iter->pg->records[iter->idx++];
 751                 if ((rec->flags & FTRACE_FL_FAILED) ||
 752                     ((iter->flags & FTRACE_ITER_FILTER) &&
 753                      !(rec->flags & FTRACE_FL_FILTER))) {
 754                         rec = NULL;
 755                         goto retry;
 756                 }
 757         }
 758
 759         iter->pos = *pos;
 760
 761         return rec;
 762 }
 763
 764 static void *t_start(struct seq_file *m, loff_t *pos)
 765 {
 766         struct ftrace_iterator *iter = m->private;
 767         void *p = NULL;
 768         loff_t l = -1;
 769
 770         if (*pos != iter->pos) {
 771                 for (p = t_next(m, p, &l); p && l < *pos; p = t_next(m, p, &l))
 772                         ;
 773         } else {
 774                 l = *pos;
 775                 p = t_next(m, p, &l);
 776         }
 777
 778         return p;
 779 }
 780
 781 static void t_stop(struct seq_file *m, void *p)
 782 {
 783 }
 784
 785 static int t_show(struct seq_file *m, void *v)
 786 {
 787         struct dyn_ftrace *rec = v;
 788         char str[KSYM_SYMBOL_LEN];
 789
 790         if (!rec)
 791                 return 0;
 792
 793         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 794
 795         seq_printf(m, "%s\n", str);
 796
 797         return 0;
 798 }
 799
 800 static struct seq_operations show_ftrace_seq_ops = {
 801         .start = t_start,
 802         .next = t_next,
 803         .stop = t_stop,
 804         .show = t_show,
 805 };
 806
 807 static int
 808 ftrace_avail_open(struct inode *inode, struct file *file)
 809 {
 810         struct ftrace_iterator *iter;
 811         int ret;
 812
 813         if (unlikely(ftrace_disabled))
 814                 return -ENODEV;
 815
 816         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
 817         if (!iter)
 818                 return -ENOMEM;
 819
 820         iter->pg = ftrace_pages_start;
 821         iter->pos = -1;
 822
 823         ret = seq_open(file, &show_ftrace_seq_ops);
 824         if (!ret) {
 825                 struct seq_file *m = file->private_data;
 826
 827                 m->private = iter;
 828         } else {
 829                 kfree(iter);
 830         }
 831
 832         return ret;
 833 }
 834
 835 int ftrace_avail_release(struct inode *inode, struct file *file)
 836 {
 837         struct seq_file *m = (struct seq_file *)file->private_data;
 838         struct ftrace_iterator *iter = m->private;
 839
 840         seq_release(inode, file);
 841         kfree(iter);
 842
 843         return 0;
 844 }
 845
 846 static void ftrace_filter_reset(void)
 847 {
 848         struct ftrace_page *pg;
 849         struct dyn_ftrace *rec;
 850         unsigned i;
 851
 852         /* keep kstop machine from running */
 853         preempt_disable();
 854         ftrace_filtered = 0;
 855         pg = ftrace_pages_start;
 856         while (pg) {
 857                 for (i = 0; i < pg->index; i++) {
 858                         rec = &pg->records[i];
 859                         if (rec->flags & FTRACE_FL_FAILED)
 860                                 continue;
 861                         rec->flags &= ~FTRACE_FL_FILTER;
 862                 }
 863                 pg = pg->next;
 864         }
 865         preempt_enable();
 866 }
 867
 868 static int
 869 ftrace_filter_open(struct inode *inode, struct file *file)
 870 {
 871         struct ftrace_iterator *iter;
 872         int ret = 0;
 873
 874         if (unlikely(ftrace_disabled))
 875                 return -ENODEV;
 876
 877         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
 878         if (!iter)
 879                 return -ENOMEM;
 880
 881         mutex_lock(&ftrace_filter_lock);
 882         if ((file->f_mode & FMODE_WRITE) &&
 883             !(file->f_flags & O_APPEND))
 884                 ftrace_filter_reset();
 885
 886         if (file->f_mode & FMODE_READ) {
 887                 iter->pg = ftrace_pages_start;
 888                 iter->pos = -1;
 889                 iter->flags = FTRACE_ITER_FILTER;
 890
 891                 ret = seq_open(file, &show_ftrace_seq_ops);
 892                 if (!ret) {
 893                         struct seq_file *m = file->private_data;
 894                         m->private = iter;
 895                 } else
 896                         kfree(iter);
 897         } else
 898                 file->private_data = iter;
 899         mutex_unlock(&ftrace_filter_lock);
 900
 901         return ret;
 902 }
 903
 904 static ssize_t
 905 ftrace_filter_read(struct file *file, char __user *ubuf,
 906                        size_t cnt, loff_t *ppos)
 907 {
 908         if (file->f_mode & FMODE_READ)
 909                 return seq_read(file, ubuf, cnt, ppos);
 910         else
 911                 return -EPERM;
 912 }
 913
 914 static loff_t
 915 ftrace_filter_lseek(struct file *file, loff_t offset, int origin)
 916 {
 917         loff_t ret;
 918
 919         if (file->f_mode & FMODE_READ)
 920                 ret = seq_lseek(file, offset, origin);
 921         else
 922                 file->f_pos = ret = 1;
 923
 924         return ret;
 925 }
 926
 927 enum {
 928         MATCH_FULL,
 929         MATCH_FRONT_ONLY,
 930         MATCH_MIDDLE_ONLY,
 931         MATCH_END_ONLY,
 932 };
 933
 934 static void
 935 ftrace_match(unsigned char *buff, int len)
 936 {
 937         char str[KSYM_SYMBOL_LEN];
 938         char *search = NULL;
 939         struct ftrace_page *pg;
 940         struct dyn_ftrace *rec;
 941         int type = MATCH_FULL;
 942         unsigned i, match = 0, search_len = 0;
 943
 944         for (i = 0; i < len; i++) {
 945                 if (buff[i] == '*') {
 946                         if (!i) {
 947                                 search = buff + i + 1;
 948                                 type = MATCH_END_ONLY;
 949                                 search_len = len - (i + 1);
 950                         } else {
 951                                 if (type == MATCH_END_ONLY) {
 952                                         type = MATCH_MIDDLE_ONLY;
 953                                 } else {
 954                                         match = i;
 955                                         type = MATCH_FRONT_ONLY;
 956                                 }
 957                                 buff[i] = 0;
 958                                 break;
 959                         }
 960                 }
 961         }
 962
 963         /* keep kstop machine from running */
 964         preempt_disable();
 965         ftrace_filtered = 1;
 966         pg = ftrace_pages_start;
 967         while (pg) {
 968                 for (i = 0; i < pg->index; i++) {
 969                         int matched = 0;
 970                         char *ptr;
 971
 972                         rec = &pg->records[i];
 973                         if (rec->flags & FTRACE_FL_FAILED)
 974                                 continue;
 975                         kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 976                         switch (type) {
 977                         case MATCH_FULL:
 978                                 if (strcmp(str, buff) == 0)
 979                                         matched = 1;
 980                                 break;
 981                         case MATCH_FRONT_ONLY:
 982                                 if (memcmp(str, buff, match) == 0)
 983                                         matched = 1;
 984                                 break;
 985                         case MATCH_MIDDLE_ONLY:
 986                                 if (strstr(str, search))
 987                                         matched = 1;
 988                                 break;
 989                         case MATCH_END_ONLY:
 990                                 ptr = strstr(str, search);
 991                                 if (ptr && (ptr[search_len] == 0))
 992                                         matched = 1;
 993                                 break;
 994                         }
 995                         if (matched)
 996                                 rec->flags |= FTRACE_FL_FILTER;
 997                 }
 998                 pg = pg->next;
 999         }
1000         preempt_enable();
1001 }
1002
1003 static ssize_t
1004 ftrace_filter_write(struct file *file, const char __user *ubuf,
1005                     size_t cnt, loff_t *ppos)
1006 {
1007         struct ftrace_iterator *iter;
1008         char ch;
1009         size_t read = 0;
1010         ssize_t ret;
1011
1012         if (!cnt || cnt < 0)
1013                 return 0;
1014
1015         mutex_lock(&ftrace_filter_lock);
1016
1017         if (file->f_mode & FMODE_READ) {
1018                 struct seq_file *m = file->private_data;
1019                 iter = m->private;
1020         } else
1021                 iter = file->private_data;
1022
1023         if (!*ppos) {
1024                 iter->flags &= ~FTRACE_ITER_CONT;
1025                 iter->buffer_idx = 0;
1026         }
1027
1028         ret = get_user(ch, ubuf++);
1029         if (ret)
1030                 goto out;
1031         read++;
1032         cnt--;
1033
1034         if (!(iter->flags & ~FTRACE_ITER_CONT)) {
1035                 /* skip white space */
1036                 while (cnt && isspace(ch)) {
1037                         ret = get_user(ch, ubuf++);
1038                         if (ret)
1039                                 goto out;
1040                         read++;
1041                         cnt--;
1042                 }
1043
1044
1045                 if (isspace(ch)) {
1046                         file->f_pos += read;
1047                         ret = read;
1048                         goto out;
1049                 }
1050
1051                 iter->buffer_idx = 0;
1052         }
1053
1054         while (cnt && !isspace(ch)) {
1055                 if (iter->buffer_idx < FTRACE_BUFF_MAX)
1056                         iter->buffer[iter->buffer_idx++] = ch;
1057                 else {
1058                         ret = -EINVAL;
1059                         goto out;
1060                 }
1061                 ret = get_user(ch, ubuf++);
1062                 if (ret)
1063                         goto out;
1064                 read++;
1065                 cnt--;
1066         }
1067
1068         if (isspace(ch)) {
1069                 iter->filtered++;
1070                 iter->buffer[iter->buffer_idx] = 0;
1071                 ftrace_match(iter->buffer, iter->buffer_idx);
1072                 iter->buffer_idx = 0;
1073         } else
1074                 iter->flags |= FTRACE_ITER_CONT;
1075
1076
1077         file->f_pos += read;
1078
1079         ret = read;
1080  out:
1081         mutex_unlock(&ftrace_filter_lock);
1082
1083         return ret;
1084 }
1085
1086 /**
1087  * ftrace_set_filter - set a function to filter on in ftrace
1088  * @buf - the string that holds the function filter text.
1089  * @len - the length of the string.
1090  * @reset - non zero to reset all filters before applying this filter.
1091  *
1092  * Filters denote which functions should be enabled when tracing is enabled.
1093  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
1094  */
1095 void ftrace_set_filter(unsigned char *buf, int len, int reset)
1096 {
1097         if (unlikely(ftrace_disabled))
1098                 return;
1099
1100         mutex_lock(&ftrace_filter_lock);
1101         if (reset)
1102                 ftrace_filter_reset();
1103         if (buf)
1104                 ftrace_match(buf, len);
1105         mutex_unlock(&ftrace_filter_lock);
1106 }
1107
1108 static int
1109 ftrace_filter_release(struct inode *inode, struct file *file)
1110 {
1111         struct seq_file *m = (struct seq_file *)file->private_data;
1112         struct ftrace_iterator *iter;
1113
1114         mutex_lock(&ftrace_filter_lock);
1115         if (file->f_mode & FMODE_READ) {
1116                 iter = m->private;
1117
1118                 seq_release(inode, file);
1119         } else
1120                 iter = file->private_data;
1121
1122         if (iter->buffer_idx) {
1123                 iter->filtered++;
1124                 iter->buffer[iter->buffer_idx] = 0;
1125                 ftrace_match(iter->buffer, iter->buffer_idx);
1126         }
1127
1128         mutex_lock(&ftrace_sysctl_lock);
1129         mutex_lock(&ftraced_lock);
1130         if (iter->filtered && ftraced_suspend && ftrace_enabled)
1131                 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1132         mutex_unlock(&ftraced_lock);
1133         mutex_unlock(&ftrace_sysctl_lock);
1134
1135         kfree(iter);
1136         mutex_unlock(&ftrace_filter_lock);
1137         return 0;
1138 }
1139
1140 static struct file_operations ftrace_avail_fops = {
1141         .open = ftrace_avail_open,
1142         .read = seq_read,
1143         .llseek = seq_lseek,
1144         .release = ftrace_avail_release,
1145 };
1146
1147 static struct file_operations ftrace_filter_fops = {
1148         .open = ftrace_filter_open,
1149         .read = ftrace_filter_read,
1150         .write = ftrace_filter_write,
1151         .llseek = ftrace_filter_lseek,
1152         .release = ftrace_filter_release,
1153 };
1154
1155 /**
1156  * ftrace_force_update - force an update to all recording ftrace functions
1157  *
1158  * The ftrace dynamic update daemon only wakes up once a second.
1159  * There may be cases where an update needs to be done immediately
1160  * for tests or internal kernel tracing to begin. This function
1161  * wakes the daemon to do an update and will not return until the
1162  * update is complete.
1163  */
1164 int ftrace_force_update(void)
1165 {
1166         unsigned long last_counter;
1167         DECLARE_WAITQUEUE(wait, current);
1168         int ret = 0;
1169
1170         if (unlikely(ftrace_disabled))
1171                 return -ENODEV;
1172
1173         mutex_lock(&ftraced_lock);
1174         last_counter = ftraced_iteration_counter;
1175
1176         set_current_state(TASK_INTERRUPTIBLE);
1177         add_wait_queue(&ftraced_waiters, &wait);
1178
1179         if (unlikely(!ftraced_task)) {
1180                 ret = -ENODEV;
1181                 goto out;
1182         }
1183
1184         do {
1185                 mutex_unlock(&ftraced_lock);
1186                 wake_up_process(ftraced_task);
1187                 schedule();
1188                 mutex_lock(&ftraced_lock);
1189                 if (signal_pending(current)) {
1190                         ret = -EINTR;
1191                         break;
1192                 }
1193                 set_current_state(TASK_INTERRUPTIBLE);
1194         } while (last_counter == ftraced_iteration_counter);
1195
1196  out:
1197         mutex_unlock(&ftraced_lock);
1198         remove_wait_queue(&ftraced_waiters, &wait);
1199         set_current_state(TASK_RUNNING);
1200
1201         return ret;
1202 }
1203
1204 static void ftrace_force_shutdown(void)
1205 {
1206         struct task_struct *task;
1207         int command = FTRACE_DISABLE_CALLS | FTRACE_UPDATE_TRACE_FUNC;
1208
1209         mutex_lock(&ftraced_lock);
1210         task = ftraced_task;
1211         ftraced_task = NULL;
1212         ftraced_suspend = -1;
1213         ftrace_run_update_code(command);
1214         mutex_unlock(&ftraced_lock);
1215
1216         if (task)
1217                 kthread_stop(task);
1218 }
1219
1220 static __init int ftrace_init_debugfs(void)
1221 {
1222         struct dentry *d_tracer;
1223         struct dentry *entry;
1224
1225         d_tracer = tracing_init_dentry();
1226
1227         entry = debugfs_create_file("available_filter_functions", 0444,
1228                                     d_tracer, NULL, &ftrace_avail_fops);
1229         if (!entry)
1230                 pr_warning("Could not create debugfs "
1231                            "'available_filter_functions' entry\n");
1232
1233         entry = debugfs_create_file("set_ftrace_filter", 0644, d_tracer,
1234                                     NULL, &ftrace_filter_fops);
1235         if (!entry)
1236                 pr_warning("Could not create debugfs "
1237                            "'set_ftrace_filter' entry\n");
1238         return 0;
1239 }
1240
1241 fs_initcall(ftrace_init_debugfs);
1242
1243 static int __init ftrace_dynamic_init(void)
1244 {
1245         struct task_struct *p;
1246         unsigned long addr;
1247         int ret;
1248
1249         addr = (unsigned long)ftrace_record_ip;
1250
1251         stop_machine_run(ftrace_dyn_arch_init, &addr, NR_CPUS);
1252
1253         /* ftrace_dyn_arch_init places the return code in addr */
1254         if (addr) {
1255                 ret = (int)addr;
1256                 goto failed;
1257         }
1258
1259         ret = ftrace_dyn_table_alloc();
1260         if (ret)
1261                 goto failed;
1262
1263         p = kthread_run(ftraced, NULL, "ftraced");
1264         if (IS_ERR(p)) {
1265                 ret = -1;
1266                 goto failed;
1267         }
1268
1269         last_ftrace_enabled = ftrace_enabled = 1;
1270         ftraced_task = p;
1271
1272         return 0;
1273
1274  failed:
1275         ftrace_disabled = 1;
1276         return ret;
1277 }
1278
1279 core_initcall(ftrace_dynamic_init);
1280 #else
1281 # define ftrace_startup()               do { } while (0)
1282 # define ftrace_shutdown()              do { } while (0)
1283 # define ftrace_startup_sysctl()        do { } while (0)
1284 # define ftrace_shutdown_sysctl()       do { } while (0)
1285 # define ftrace_force_shutdown()        do { } while (0)
1286 #endif /* CONFIG_DYNAMIC_FTRACE */
1287
1288 /**
1289  * ftrace_kill - totally shutdown ftrace
1290  *
1291  * This is a safety measure. If something was detected that seems
1292  * wrong, calling this function will keep ftrace from doing
1293  * any more modifications, and updates.
1294  * used when something went wrong.
1295  */
1296 void ftrace_kill(void)
1297 {
1298         mutex_lock(&ftrace_sysctl_lock);
1299         ftrace_disabled = 1;
1300         ftrace_enabled = 0;
1301
1302         clear_ftrace_function();
1303         mutex_unlock(&ftrace_sysctl_lock);
1304
1305         /* Try to totally disable ftrace */
1306         ftrace_force_shutdown();
1307 }
1308
1309 /**
1310  * register_ftrace_function - register a function for profiling
1311  * @ops - ops structure that holds the function for profiling.
1312  *
1313  * Register a function to be called by all functions in the
1314  * kernel.
1315  *
1316  * Note: @ops->func and all the functions it calls must be labeled
1317  *       with "notrace", otherwise it will go into a
1318  *       recursive loop.
1319  */
1320 int register_ftrace_function(struct ftrace_ops *ops)
1321 {
1322         int ret;
1323
1324         if (unlikely(ftrace_disabled))
1325                 return -1;
1326
1327         mutex_lock(&ftrace_sysctl_lock);
1328         ret = __register_ftrace_function(ops);
1329         ftrace_startup();
1330         mutex_unlock(&ftrace_sysctl_lock);
1331
1332         return ret;
1333 }
1334
1335 /**
1336  * unregister_ftrace_function - unresgister a function for profiling.
1337  * @ops - ops structure that holds the function to unregister
1338  *
1339  * Unregister a function that was added to be called by ftrace profiling.
1340  */
1341 int unregister_ftrace_function(struct ftrace_ops *ops)
1342 {
1343         int ret;
1344
1345         mutex_lock(&ftrace_sysctl_lock);
1346         ret = __unregister_ftrace_function(ops);
1347         ftrace_shutdown();
1348         mutex_unlock(&ftrace_sysctl_lock);
1349
1350         return ret;
1351 }
1352
1353 int
1354 ftrace_enable_sysctl(struct ctl_table *table, int write,
1355                      struct file *file, void __user *buffer, size_t *lenp,
1356                      loff_t *ppos)
1357 {
1358         int ret;
1359
1360         if (unlikely(ftrace_disabled))
1361                 return -ENODEV;
1362
1363         mutex_lock(&ftrace_sysctl_lock);
1364
1365         ret  = proc_dointvec(table, write, file, buffer, lenp, ppos);
1366
1367         if (ret || !write || (last_ftrace_enabled == ftrace_enabled))
1368                 goto out;
1369
1370         last_ftrace_enabled = ftrace_enabled;
1371
1372         if (ftrace_enabled) {
1373
1374                 ftrace_startup_sysctl();
1375
1376                 /* we are starting ftrace again */
1377                 if (ftrace_list != &ftrace_list_end) {
1378                         if (ftrace_list->next == &ftrace_list_end)
1379                                 ftrace_trace_function = ftrace_list->func;
1380                         else
1381                                 ftrace_trace_function = ftrace_list_func;
1382                 }
1383
1384         } else {
1385                 /* stopping ftrace calls (just send to ftrace_stub) */
1386                 ftrace_trace_function = ftrace_stub;
1387
1388                 ftrace_shutdown_sysctl();
1389         }
1390
1391  out:
1392         mutex_unlock(&ftrace_sysctl_lock);
1393         return ret;
1394 }