kernel/trace/trace.c

   1 /*
   2  * ring buffer based function tracer
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally taken from the RT patch by:
   8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code from the latency_tracer, that is:
  11  *  Copyright (C) 2004-2006 Ingo Molnar
  12  *  Copyright (C) 2004 William Lee Irwin III
  13  */
  14 #include <linux/utsrelease.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/debugfs.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/linkage.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/ftrace.h>
  23 #include <linux/module.h>
  24 #include <linux/percpu.h>
  25 #include <linux/ctype.h>
  26 #include <linux/init.h>
  27 #include <linux/gfp.h>
  28 #include <linux/fs.h>
  29
  30 #include "trace.h"
  31
  32 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  33 unsigned long __read_mostly     tracing_thresh;
  34
  35 static int tracing_disabled = 1;
  36
  37 static long notrace
  38 ns2usecs(cycle_t nsec)
  39 {
  40         nsec += 500;
  41         do_div(nsec, 1000);
  42         return nsec;
  43 }
  44
  45 notrace cycle_t ftrace_now(int cpu)
  46 {
  47         return cpu_clock(cpu);
  48 }
  49
  50 static atomic_t                 tracer_counter;
  51 static struct trace_array       global_trace;
  52
  53 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  54
  55 static struct trace_array       max_tr;
  56
  57 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  58
  59 static int                      tracer_enabled;
  60 static unsigned long            trace_nr_entries = 16384UL;
  61
  62 static struct tracer            *trace_types __read_mostly;
  63 static struct tracer            *current_trace __read_mostly;
  64 static int                      max_tracer_type_len;
  65
  66 static DEFINE_MUTEX(trace_types_lock);
  67
  68 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  69
  70 static int __init set_nr_entries(char *str)
  71 {
  72         if (!str)
  73                 return 0;
  74         trace_nr_entries = simple_strtoul(str, &str, 0);
  75         return 1;
  76 }
  77 __setup("trace_entries=", set_nr_entries);
  78
  79 unsigned long nsecs_to_usecs(unsigned long nsecs)
  80 {
  81         return nsecs / 1000;
  82 }
  83
  84 enum trace_type {
  85         __TRACE_FIRST_TYPE = 0,
  86
  87         TRACE_FN,
  88         TRACE_CTX,
  89
  90         __TRACE_LAST_TYPE
  91 };
  92
  93 enum trace_flag_type {
  94         TRACE_FLAG_IRQS_OFF             = 0x01,
  95         TRACE_FLAG_NEED_RESCHED         = 0x02,
  96         TRACE_FLAG_HARDIRQ              = 0x04,
  97         TRACE_FLAG_SOFTIRQ              = 0x08,
  98 };
  99
 100 enum trace_iterator_flags {
 101         TRACE_ITER_PRINT_PARENT         = 0x01,
 102         TRACE_ITER_SYM_OFFSET           = 0x02,
 103         TRACE_ITER_SYM_ADDR             = 0x04,
 104         TRACE_ITER_VERBOSE              = 0x08,
 105 };
 106
 107 #define TRACE_ITER_SYM_MASK \
 108         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 109
 110 /* These must match the bit postions above */
 111 static const char *trace_options[] = {
 112         "print-parent",
 113         "sym-offset",
 114         "sym-addr",
 115         "verbose",
 116         NULL
 117 };
 118
 119 static unsigned trace_flags;
 120
 121 static DEFINE_SPINLOCK(ftrace_max_lock);
 122
 123 /*
 124  * Copy the new maximum trace into the separate maximum-trace
 125  * structure. (this way the maximum trace is permanently saved,
 126  * for later retrieval via /debugfs/tracing/latency_trace)
 127  */
 128 static notrace void
 129 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 130 {
 131         struct trace_array_cpu *data = tr->data[cpu];
 132
 133         max_tr.cpu = cpu;
 134         max_tr.time_start = data->preempt_timestamp;
 135
 136         data = max_tr.data[cpu];
 137         data->saved_latency = tracing_max_latency;
 138
 139         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 140         data->pid = tsk->pid;
 141         data->uid = tsk->uid;
 142         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 143         data->policy = tsk->policy;
 144         data->rt_priority = tsk->rt_priority;
 145
 146         /* record this tasks comm */
 147         tracing_record_cmdline(current);
 148 }
 149
 150 void check_pages(struct trace_array_cpu *data)
 151 {
 152         struct page *page, *tmp;
 153
 154         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 155         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 156
 157         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 158                 BUG_ON(page->lru.next->prev != &page->lru);
 159                 BUG_ON(page->lru.prev->next != &page->lru);
 160         }
 161 }
 162
 163 void *head_page(struct trace_array_cpu *data)
 164 {
 165         struct page *page;
 166
 167         check_pages(data);
 168         if (list_empty(&data->trace_pages))
 169                 return NULL;
 170
 171         page = list_entry(data->trace_pages.next, struct page, lru);
 172         BUG_ON(&page->lru == &data->trace_pages);
 173
 174         return page_address(page);
 175 }
 176
 177 static notrace int
 178 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 179 {
 180         int len = (PAGE_SIZE - 1) - s->len;
 181         va_list ap;
 182         int ret;
 183
 184         if (!len)
 185                 return 0;
 186
 187         va_start(ap, fmt);
 188         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 189         va_end(ap);
 190
 191         /* If we can't write it all, don't bother writing anything */
 192         if (ret > len)
 193                 return 0;
 194
 195         s->len += ret;
 196
 197         return len;
 198 }
 199
 200 static notrace int
 201 trace_seq_puts(struct trace_seq *s, const char *str)
 202 {
 203         int len = strlen(str);
 204
 205         if (len > ((PAGE_SIZE - 1) - s->len))
 206                 return 0;
 207
 208         memcpy(s->buffer + s->len, str, len);
 209         s->len += len;
 210
 211         return len;
 212 }
 213
 214 static notrace int
 215 trace_seq_putc(struct trace_seq *s, unsigned char c)
 216 {
 217         if (s->len >= (PAGE_SIZE - 1))
 218                 return 0;
 219
 220         s->buffer[s->len++] = c;
 221
 222         return 1;
 223 }
 224
 225 static notrace void
 226 trace_seq_reset(struct trace_seq *s)
 227 {
 228         s->len = 0;
 229 }
 230
 231 static notrace void
 232 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 233 {
 234         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 235
 236         s->buffer[len] = 0;
 237         seq_puts(m, s->buffer);
 238
 239         trace_seq_reset(s);
 240 }
 241
 242 notrace static void
 243 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 244 {
 245         struct list_head flip_pages;
 246
 247         INIT_LIST_HEAD(&flip_pages);
 248
 249         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 250                 sizeof(struct trace_array_cpu) -
 251                 offsetof(struct trace_array_cpu, trace_head_idx));
 252
 253         check_pages(tr1);
 254         check_pages(tr2);
 255         list_splice_init(&tr1->trace_pages, &flip_pages);
 256         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 257         list_splice_init(&flip_pages, &tr2->trace_pages);
 258         BUG_ON(!list_empty(&flip_pages));
 259         check_pages(tr1);
 260         check_pages(tr2);
 261 }
 262
 263 notrace void
 264 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 265 {
 266         struct trace_array_cpu *data;
 267         int i;
 268
 269         WARN_ON_ONCE(!irqs_disabled());
 270         spin_lock(&ftrace_max_lock);
 271         /* clear out all the previous traces */
 272         for_each_possible_cpu(i) {
 273                 data = tr->data[i];
 274                 flip_trace(max_tr.data[i], data);
 275                 tracing_reset(data);
 276         }
 277
 278         __update_max_tr(tr, tsk, cpu);
 279         spin_unlock(&ftrace_max_lock);
 280 }
 281
 282 /**
 283  * update_max_tr_single - only copy one trace over, and reset the rest
 284  * @tr - tracer
 285  * @tsk - task with the latency
 286  * @cpu - the cpu of the buffer to copy.
 287  */
 288 notrace void
 289 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 290 {
 291         struct trace_array_cpu *data = tr->data[cpu];
 292         int i;
 293
 294         WARN_ON_ONCE(!irqs_disabled());
 295         spin_lock(&ftrace_max_lock);
 296         for_each_possible_cpu(i)
 297                 tracing_reset(max_tr.data[i]);
 298
 299         flip_trace(max_tr.data[cpu], data);
 300         tracing_reset(data);
 301
 302         __update_max_tr(tr, tsk, cpu);
 303         spin_unlock(&ftrace_max_lock);
 304 }
 305
 306 int register_tracer(struct tracer *type)
 307 {
 308         struct tracer *t;
 309         int len;
 310         int ret = 0;
 311
 312         if (!type->name) {
 313                 pr_info("Tracer must have a name\n");
 314                 return -1;
 315         }
 316
 317         mutex_lock(&trace_types_lock);
 318         for (t = trace_types; t; t = t->next) {
 319                 if (strcmp(type->name, t->name) == 0) {
 320                         /* already found */
 321                         pr_info("Trace %s already registered\n",
 322                                 type->name);
 323                         ret = -1;
 324                         goto out;
 325                 }
 326         }
 327
 328 #ifdef CONFIG_FTRACE_STARTUP_TEST
 329         if (type->selftest) {
 330                 struct tracer *saved_tracer = current_trace;
 331                 struct trace_array_cpu *data;
 332                 struct trace_array *tr = &global_trace;
 333                 int saved_ctrl = tr->ctrl;
 334                 int i;
 335                 /*
 336                  * Run a selftest on this tracer.
 337                  * Here we reset the trace buffer, and set the current
 338                  * tracer to be this tracer. The tracer can then run some
 339                  * internal tracing to verify that everything is in order.
 340                  * If we fail, we do not register this tracer.
 341                  */
 342                 for_each_possible_cpu(i) {
 343                         data = tr->data[i];
 344                         if (!head_page(data))
 345                                 continue;
 346                         tracing_reset(data);
 347                 }
 348                 current_trace = type;
 349                 tr->ctrl = 0;
 350                 /* the test is responsible for initializing and enabling */
 351                 pr_info("Testing tracer %s: ", type->name);
 352                 ret = type->selftest(type, tr);
 353                 /* the test is responsible for resetting too */
 354                 current_trace = saved_tracer;
 355                 tr->ctrl = saved_ctrl;
 356                 if (ret) {
 357                         printk(KERN_CONT "FAILED!\n");
 358                         goto out;
 359                 }
 360                 /* Only reset on passing, to avoid touching corrupted buffers */
 361                 for_each_possible_cpu(i) {
 362                         data = tr->data[i];
 363                         if (!head_page(data))
 364                                 continue;
 365                         tracing_reset(data);
 366                 }
 367                 printk(KERN_CONT "PASSED\n");
 368         }
 369 #endif
 370
 371         type->next = trace_types;
 372         trace_types = type;
 373         len = strlen(type->name);
 374         if (len > max_tracer_type_len)
 375                 max_tracer_type_len = len;
 376
 377  out:
 378         mutex_unlock(&trace_types_lock);
 379
 380         return ret;
 381 }
 382
 383 void unregister_tracer(struct tracer *type)
 384 {
 385         struct tracer **t;
 386         int len;
 387
 388         mutex_lock(&trace_types_lock);
 389         for (t = &trace_types; *t; t = &(*t)->next) {
 390                 if (*t == type)
 391                         goto found;
 392         }
 393         pr_info("Trace %s not registered\n", type->name);
 394         goto out;
 395
 396  found:
 397         *t = (*t)->next;
 398         if (strlen(type->name) != max_tracer_type_len)
 399                 goto out;
 400
 401         max_tracer_type_len = 0;
 402         for (t = &trace_types; *t; t = &(*t)->next) {
 403                 len = strlen((*t)->name);
 404                 if (len > max_tracer_type_len)
 405                         max_tracer_type_len = len;
 406         }
 407  out:
 408         mutex_unlock(&trace_types_lock);
 409 }
 410
 411 notrace void tracing_reset(struct trace_array_cpu *data)
 412 {
 413         data->trace_idx = 0;
 414         data->trace_head = data->trace_tail = head_page(data);
 415         data->trace_head_idx = 0;
 416         data->trace_tail_idx = 0;
 417 }
 418
 419 #ifdef CONFIG_FTRACE
 420 static notrace void
 421 function_trace_call(unsigned long ip, unsigned long parent_ip)
 422 {
 423         struct trace_array *tr = &global_trace;
 424         struct trace_array_cpu *data;
 425         unsigned long flags;
 426         long disabled;
 427         int cpu;
 428
 429         if (unlikely(!tracer_enabled))
 430                 return;
 431
 432         local_irq_save(flags);
 433         cpu = raw_smp_processor_id();
 434         data = tr->data[cpu];
 435         disabled = atomic_inc_return(&data->disabled);
 436
 437         if (likely(disabled == 1))
 438                 ftrace(tr, data, ip, parent_ip, flags);
 439
 440         atomic_dec(&data->disabled);
 441         local_irq_restore(flags);
 442 }
 443
 444 static struct ftrace_ops trace_ops __read_mostly =
 445 {
 446         .func = function_trace_call,
 447 };
 448 #endif
 449
 450 notrace void tracing_start_function_trace(void)
 451 {
 452         register_ftrace_function(&trace_ops);
 453 }
 454
 455 notrace void tracing_stop_function_trace(void)
 456 {
 457         unregister_ftrace_function(&trace_ops);
 458 }
 459
 460 #define SAVED_CMDLINES 128
 461 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 462 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 463 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 464 static int cmdline_idx;
 465 static DEFINE_SPINLOCK(trace_cmdline_lock);
 466 atomic_t trace_record_cmdline_disabled;
 467
 468 static void trace_init_cmdlines(void)
 469 {
 470         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 471         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 472         cmdline_idx = 0;
 473 }
 474
 475 notrace void trace_stop_cmdline_recording(void);
 476
 477 static notrace void trace_save_cmdline(struct task_struct *tsk)
 478 {
 479         unsigned map;
 480         unsigned idx;
 481
 482         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 483                 return;
 484
 485         /*
 486          * It's not the end of the world if we don't get
 487          * the lock, but we also don't want to spin
 488          * nor do we want to disable interrupts,
 489          * so if we miss here, then better luck next time.
 490          */
 491         if (!spin_trylock(&trace_cmdline_lock))
 492                 return;
 493
 494         idx = map_pid_to_cmdline[tsk->pid];
 495         if (idx >= SAVED_CMDLINES) {
 496                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 497
 498                 map = map_cmdline_to_pid[idx];
 499                 if (map <= PID_MAX_DEFAULT)
 500                         map_pid_to_cmdline[map] = (unsigned)-1;
 501
 502                 map_pid_to_cmdline[tsk->pid] = idx;
 503
 504                 cmdline_idx = idx;
 505         }
 506
 507         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 508
 509         spin_unlock(&trace_cmdline_lock);
 510 }
 511
 512 static notrace char *trace_find_cmdline(int pid)
 513 {
 514         char *cmdline = "<...>";
 515         unsigned map;
 516
 517         if (!pid)
 518                 return "<idle>";
 519
 520         if (pid > PID_MAX_DEFAULT)
 521                 goto out;
 522
 523         map = map_pid_to_cmdline[pid];
 524         if (map >= SAVED_CMDLINES)
 525                 goto out;
 526
 527         cmdline = saved_cmdlines[map];
 528
 529  out:
 530         return cmdline;
 531 }
 532
 533 notrace void tracing_record_cmdline(struct task_struct *tsk)
 534 {
 535         if (atomic_read(&trace_record_cmdline_disabled))
 536                 return;
 537
 538         trace_save_cmdline(tsk);
 539 }
 540
 541 static inline notrace struct list_head *
 542 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 543 {
 544         /*
 545          * Roundrobin - but skip the head (which is not a real page):
 546          */
 547         next = next->next;
 548         if (unlikely(next == &data->trace_pages))
 549                 next = next->next;
 550         BUG_ON(next == &data->trace_pages);
 551
 552         return next;
 553 }
 554
 555 static inline notrace void *
 556 trace_next_page(struct trace_array_cpu *data, void *addr)
 557 {
 558         struct list_head *next;
 559         struct page *page;
 560
 561         page = virt_to_page(addr);
 562
 563         next = trace_next_list(data, &page->lru);
 564         page = list_entry(next, struct page, lru);
 565
 566         return page_address(page);
 567 }
 568
 569 static inline notrace struct trace_entry *
 570 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 571 {
 572         unsigned long idx, idx_next;
 573         struct trace_entry *entry;
 574
 575         data->trace_idx++;
 576         idx = data->trace_head_idx;
 577         idx_next = idx + 1;
 578
 579         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 580
 581         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 582
 583         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 584                 data->trace_head = trace_next_page(data, data->trace_head);
 585                 idx_next = 0;
 586         }
 587
 588         if (data->trace_head == data->trace_tail &&
 589             idx_next == data->trace_tail_idx) {
 590                 /* overrun */
 591                 data->trace_tail_idx++;
 592                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 593                         data->trace_tail =
 594                                 trace_next_page(data, data->trace_tail);
 595                         data->trace_tail_idx = 0;
 596                 }
 597         }
 598
 599         data->trace_head_idx = idx_next;
 600
 601         return entry;
 602 }
 603
 604 static inline notrace void
 605 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 606 {
 607         struct task_struct *tsk = current;
 608         unsigned long pc;
 609
 610         pc = preempt_count();
 611
 612         entry->idx              = atomic_inc_return(&tracer_counter);
 613         entry->preempt_count    = pc & 0xff;
 614         entry->pid              = tsk->pid;
 615         entry->t                = ftrace_now(raw_smp_processor_id());
 616         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 617                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 618                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 619                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 620 }
 621
 622 notrace void
 623 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 624        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 625 {
 626         struct trace_entry *entry;
 627
 628         spin_lock(&data->lock);
 629         entry                   = tracing_get_trace_entry(tr, data);
 630         tracing_generic_entry_update(entry, flags);
 631         entry->type             = TRACE_FN;
 632         entry->fn.ip            = ip;
 633         entry->fn.parent_ip     = parent_ip;
 634         spin_unlock(&data->lock);
 635 }
 636
 637 notrace void
 638 tracing_sched_switch_trace(struct trace_array *tr,
 639                            struct trace_array_cpu *data,
 640                            struct task_struct *prev, struct task_struct *next,
 641                            unsigned long flags)
 642 {
 643         struct trace_entry *entry;
 644
 645         spin_lock(&data->lock);
 646         entry                   = tracing_get_trace_entry(tr, data);
 647         tracing_generic_entry_update(entry, flags);
 648         entry->type             = TRACE_CTX;
 649         entry->ctx.prev_pid     = prev->pid;
 650         entry->ctx.prev_prio    = prev->prio;
 651         entry->ctx.prev_state   = prev->state;
 652         entry->ctx.next_pid     = next->pid;
 653         entry->ctx.next_prio    = next->prio;
 654         spin_unlock(&data->lock);
 655 }
 656
 657 enum trace_file_type {
 658         TRACE_FILE_LAT_FMT      = 1,
 659 };
 660
 661 static struct trace_entry *
 662 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 663                 struct trace_iterator *iter, int cpu)
 664 {
 665         struct page *page;
 666         struct trace_entry *array;
 667
 668         if (iter->next_idx[cpu] >= tr->entries ||
 669             iter->next_idx[cpu] >= data->trace_idx ||
 670             (data->trace_head == data->trace_tail &&
 671              data->trace_head_idx == data->trace_tail_idx))
 672                 return NULL;
 673
 674         if (!iter->next_page[cpu]) {
 675                 /* Initialize the iterator for this cpu trace buffer */
 676                 WARN_ON(!data->trace_tail);
 677                 page = virt_to_page(data->trace_tail);
 678                 iter->next_page[cpu] = &page->lru;
 679                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 680         }
 681
 682         page = list_entry(iter->next_page[cpu], struct page, lru);
 683         BUG_ON(&data->trace_pages == &page->lru);
 684
 685         array = page_address(page);
 686
 687         /* Still possible to catch up to the tail */
 688         if (iter->next_idx[cpu] && array == data->trace_tail &&
 689             iter->next_page_idx[cpu] == data->trace_tail_idx)
 690                 return NULL;
 691
 692         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 693         return &array[iter->next_page_idx[cpu]];
 694 }
 695
 696 static struct notrace trace_entry *
 697 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 698 {
 699         struct trace_array *tr = iter->tr;
 700         struct trace_entry *ent, *next = NULL;
 701         int next_cpu = -1;
 702         int cpu;
 703
 704         for_each_possible_cpu(cpu) {
 705                 if (!head_page(tr->data[cpu]))
 706                         continue;
 707                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 708                 if (ent &&
 709                     (!next || (long)(next->idx - ent->idx) > 0)) {
 710                         next = ent;
 711                         next_cpu = cpu;
 712                 }
 713         }
 714
 715         if (ent_cpu)
 716                 *ent_cpu = next_cpu;
 717
 718         return next;
 719 }
 720
 721 static notrace void
 722 trace_iterator_increment(struct trace_iterator *iter)
 723 {
 724         iter->idx++;
 725         iter->next_idx[iter->cpu]++;
 726         iter->next_page_idx[iter->cpu]++;
 727         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 728                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 729
 730                 iter->next_page_idx[iter->cpu] = 0;
 731                 iter->next_page[iter->cpu] =
 732                         trace_next_list(data, iter->next_page[iter->cpu]);
 733         }
 734 }
 735
 736 static notrace void
 737 trace_consume(struct trace_iterator *iter)
 738 {
 739         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 740
 741         data->trace_tail_idx++;
 742         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 743                 data->trace_tail = trace_next_page(data, data->trace_tail);
 744                 data->trace_tail_idx = 0;
 745         }
 746
 747         /* Check if we empty it, then reset the index */
 748         if (data->trace_head == data->trace_tail &&
 749             data->trace_head_idx == data->trace_tail_idx)
 750                 data->trace_idx = 0;
 751
 752         trace_iterator_increment(iter);
 753 }
 754
 755 static notrace void *
 756 find_next_entry_inc(struct trace_iterator *iter)
 757 {
 758         struct trace_entry *next;
 759         int next_cpu = -1;
 760
 761         next = find_next_entry(iter, &next_cpu);
 762
 763         iter->prev_ent = iter->ent;
 764         iter->prev_cpu = iter->cpu;
 765
 766         iter->ent = next;
 767         iter->cpu = next_cpu;
 768
 769         if (next)
 770                 trace_iterator_increment(iter);
 771
 772         return next ? iter : NULL;
 773 }
 774
 775 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 776 {
 777         struct trace_iterator *iter = m->private;
 778         void *last_ent = iter->ent;
 779         int i = (int)*pos;
 780         void *ent;
 781
 782         (*pos)++;
 783
 784         /* can't go backwards */
 785         if (iter->idx > i)
 786                 return NULL;
 787
 788         if (iter->idx < 0)
 789                 ent = find_next_entry_inc(iter);
 790         else
 791                 ent = iter;
 792
 793         while (ent && iter->idx < i)
 794                 ent = find_next_entry_inc(iter);
 795
 796         iter->pos = *pos;
 797
 798         if (last_ent && !ent)
 799                 seq_puts(m, "\n\nvim:ft=help\n");
 800
 801         return ent;
 802 }
 803
 804 static void *s_start(struct seq_file *m, loff_t *pos)
 805 {
 806         struct trace_iterator *iter = m->private;
 807         void *p = NULL;
 808         loff_t l = 0;
 809         int i;
 810
 811         mutex_lock(&trace_types_lock);
 812
 813         if (!current_trace || current_trace != iter->trace)
 814                 return NULL;
 815
 816         atomic_inc(&trace_record_cmdline_disabled);
 817
 818         /* let the tracer grab locks here if needed */
 819         if (current_trace->start)
 820                 current_trace->start(iter);
 821
 822         if (*pos != iter->pos) {
 823                 iter->ent = NULL;
 824                 iter->cpu = 0;
 825                 iter->idx = -1;
 826                 iter->prev_ent = NULL;
 827                 iter->prev_cpu = -1;
 828
 829                 for_each_possible_cpu(i) {
 830                         iter->next_idx[i] = 0;
 831                         iter->next_page[i] = NULL;
 832                 }
 833
 834                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 835                         ;
 836
 837         } else {
 838                 l = *pos - 1;
 839                 p = s_next(m, p, &l);
 840         }
 841
 842         return p;
 843 }
 844
 845 static void s_stop(struct seq_file *m, void *p)
 846 {
 847         struct trace_iterator *iter = m->private;
 848
 849         atomic_dec(&trace_record_cmdline_disabled);
 850
 851         /* let the tracer release locks here if needed */
 852         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 853                 iter->trace->stop(iter);
 854
 855         mutex_unlock(&trace_types_lock);
 856 }
 857
 858 static int
 859 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 860 {
 861 #ifdef CONFIG_KALLSYMS
 862         char str[KSYM_SYMBOL_LEN];
 863
 864         kallsyms_lookup(address, NULL, NULL, NULL, str);
 865
 866         return trace_seq_printf(s, fmt, str);
 867 #endif
 868         return 1;
 869 }
 870
 871 static int
 872 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 873                      unsigned long address)
 874 {
 875 #ifdef CONFIG_KALLSYMS
 876         char str[KSYM_SYMBOL_LEN];
 877
 878         sprint_symbol(str, address);
 879         return trace_seq_printf(s, fmt, str);
 880 #endif
 881         return 1;
 882 }
 883
 884 #ifndef CONFIG_64BIT
 885 # define IP_FMT "%08lx"
 886 #else
 887 # define IP_FMT "%016lx"
 888 #endif
 889
 890 static notrace int
 891 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 892 {
 893         int ret;
 894
 895         if (!ip)
 896                 return trace_seq_printf(s, "0");
 897
 898         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 899                 ret = seq_print_sym_offset(s, "%s", ip);
 900         else
 901                 ret = seq_print_sym_short(s, "%s", ip);
 902
 903         if (!ret)
 904                 return 0;
 905
 906         if (sym_flags & TRACE_ITER_SYM_ADDR)
 907                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 908         return ret;
 909 }
 910
 911 static notrace void print_lat_help_header(struct seq_file *m)
 912 {
 913         seq_puts(m, "#                _------=> CPU#            \n");
 914         seq_puts(m, "#               / _-----=> irqs-off        \n");
 915         seq_puts(m, "#              | / _----=> need-resched    \n");
 916         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 917         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 918         seq_puts(m, "#              |||| /                      \n");
 919         seq_puts(m, "#              |||||     delay             \n");
 920         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 921         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 922 }
 923
 924 static notrace void print_func_help_header(struct seq_file *m)
 925 {
 926         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 927         seq_puts(m, "#              | |      |          |         |\n");
 928 }
 929
 930
 931 static notrace void
 932 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 933 {
 934         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 935         struct trace_array *tr = iter->tr;
 936         struct trace_array_cpu *data = tr->data[tr->cpu];
 937         struct tracer *type = current_trace;
 938         unsigned long total   = 0;
 939         unsigned long entries = 0;
 940         int cpu;
 941         const char *name = "preemption";
 942
 943         if (type)
 944                 name = type->name;
 945
 946         for_each_possible_cpu(cpu) {
 947                 if (head_page(tr->data[cpu])) {
 948                         total += tr->data[cpu]->trace_idx;
 949                         if (tr->data[cpu]->trace_idx > tr->entries)
 950                                 entries += tr->entries;
 951                         else
 952                                 entries += tr->data[cpu]->trace_idx;
 953                 }
 954         }
 955
 956         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
 957                    name, UTS_RELEASE);
 958         seq_puts(m, "-----------------------------------"
 959                  "---------------------------------\n");
 960         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
 961                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
 962                    nsecs_to_usecs(data->saved_latency),
 963                    entries,
 964                    total,
 965                    tr->cpu,
 966 #if defined(CONFIG_PREEMPT_NONE)
 967                    "server",
 968 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
 969                    "desktop",
 970 #elif defined(CONFIG_PREEMPT_DESKTOP)
 971                    "preempt",
 972 #else
 973                    "unknown",
 974 #endif
 975                    /* These are reserved for later use */
 976                    0, 0, 0, 0);
 977 #ifdef CONFIG_SMP
 978         seq_printf(m, " #P:%d)\n", num_online_cpus());
 979 #else
 980         seq_puts(m, ")\n");
 981 #endif
 982         seq_puts(m, "    -----------------\n");
 983         seq_printf(m, "    | task: %.16s-%d "
 984                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
 985                    data->comm, data->pid, data->uid, data->nice,
 986                    data->policy, data->rt_priority);
 987         seq_puts(m, "    -----------------\n");
 988
 989         if (data->critical_start) {
 990                 seq_puts(m, " => started at: ");
 991                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
 992                 trace_print_seq(m, &iter->seq);
 993                 seq_puts(m, "\n => ended at:   ");
 994                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
 995                 trace_print_seq(m, &iter->seq);
 996                 seq_puts(m, "\n");
 997         }
 998
 999         seq_puts(m, "\n");
1000 }
1001
1002 static notrace void
1003 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1004 {
1005         int hardirq, softirq;
1006         char *comm;
1007
1008         comm = trace_find_cmdline(entry->pid);
1009
1010         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1011         trace_seq_printf(s, "%d", cpu);
1012         trace_seq_printf(s, "%c%c",
1013                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1014                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1015
1016         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1017         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1018         if (hardirq && softirq)
1019                 trace_seq_putc(s, 'H');
1020         else {
1021                 if (hardirq)
1022                         trace_seq_putc(s, 'h');
1023                 else {
1024                         if (softirq)
1025                                 trace_seq_putc(s, 's');
1026                         else
1027                                 trace_seq_putc(s, '.');
1028                 }
1029         }
1030
1031         if (entry->preempt_count)
1032                 trace_seq_printf(s, "%x", entry->preempt_count);
1033         else
1034                 trace_seq_puts(s, ".");
1035 }
1036
1037 unsigned long preempt_mark_thresh = 100;
1038
1039 static notrace void
1040 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1041                     unsigned long rel_usecs)
1042 {
1043         trace_seq_printf(s, " %4lldus", abs_usecs);
1044         if (rel_usecs > preempt_mark_thresh)
1045                 trace_seq_puts(s, "!: ");
1046         else if (rel_usecs > 1)
1047                 trace_seq_puts(s, "+: ");
1048         else
1049                 trace_seq_puts(s, " : ");
1050 }
1051
1052 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1053
1054 static notrace void
1055 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1056 {
1057         struct trace_seq *s = &iter->seq;
1058         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1059         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1060         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1061         struct trace_entry *entry = iter->ent;
1062         unsigned long abs_usecs;
1063         unsigned long rel_usecs;
1064         char *comm;
1065         int S;
1066
1067         if (!next_entry)
1068                 next_entry = entry;
1069         rel_usecs = ns2usecs(next_entry->t - entry->t);
1070         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1071
1072         if (verbose) {
1073                 comm = trace_find_cmdline(entry->pid);
1074                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1075                                  " %ld.%03ldms (+%ld.%03ldms): ",
1076                                  comm,
1077                                  entry->pid, cpu, entry->flags,
1078                                  entry->preempt_count, trace_idx,
1079                                  ns2usecs(entry->t),
1080                                  abs_usecs/1000,
1081                                  abs_usecs % 1000, rel_usecs/1000,
1082                                  rel_usecs % 1000);
1083         } else {
1084                 lat_print_generic(s, entry, cpu);
1085                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1086         }
1087         switch (entry->type) {
1088         case TRACE_FN:
1089                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1090                 trace_seq_puts(s, " (");
1091                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1092                 trace_seq_puts(s, ")\n");
1093                 break;
1094         case TRACE_CTX:
1095                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1096                         state_to_char[entry->ctx.prev_state] : 'X';
1097                 comm = trace_find_cmdline(entry->ctx.next_pid);
1098                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1099                                  entry->ctx.prev_pid,
1100                                  entry->ctx.prev_prio,
1101                                  S,
1102                                  entry->ctx.next_pid,
1103                                  entry->ctx.next_prio,
1104                                  comm);
1105                 break;
1106         default:
1107                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1108         }
1109 }
1110
1111 static notrace void sync_time_offset(struct trace_iterator *iter)
1112 {
1113         struct trace_array_cpu *prev_array, *array;
1114         struct trace_entry *prev_entry, *entry;
1115         cycle_t prev_t, t;
1116
1117         entry = iter->ent;
1118         prev_entry = iter->prev_ent;
1119         if (!prev_entry)
1120                 return;
1121
1122         prev_array = iter->tr->data[iter->prev_cpu];
1123         array = iter->tr->data[iter->cpu];
1124
1125         prev_t = prev_entry->t + prev_array->time_offset;
1126         t = entry->t + array->time_offset;
1127
1128         /*
1129          * If time goes backwards we increase the offset of
1130          * the current array, to not have observable time warps.
1131          * This will quickly synchronize the time offsets of
1132          * multiple CPUs:
1133          */
1134         if (t < prev_t)
1135                 array->time_offset += prev_t - t;
1136 }
1137
1138 static notrace int
1139 print_trace_fmt(struct trace_iterator *iter)
1140 {
1141         struct trace_seq *s = &iter->seq;
1142         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1143         struct trace_entry *entry;
1144         unsigned long usec_rem;
1145         unsigned long long t;
1146         unsigned long secs;
1147         char *comm;
1148         int S;
1149         int ret;
1150
1151         sync_time_offset(iter);
1152         entry = iter->ent;
1153
1154         comm = trace_find_cmdline(iter->ent->pid);
1155
1156         t = ns2usecs(entry->t + iter->tr->data[iter->cpu]->time_offset);
1157         usec_rem = do_div(t, 1000000ULL);
1158         secs = (unsigned long)t;
1159
1160         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1161         if (!ret)
1162                 return 0;
1163         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1164         if (!ret)
1165                 return 0;
1166         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1167         if (!ret)
1168                 return 0;
1169
1170         switch (entry->type) {
1171         case TRACE_FN:
1172                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1173                 if (!ret)
1174                         return 0;
1175                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1176                                                 entry->fn.parent_ip) {
1177                         ret = trace_seq_printf(s, " <-");
1178                         if (!ret)
1179                                 return 0;
1180                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1181                                                sym_flags);
1182                         if (!ret)
1183                                 return 0;
1184                 }
1185                 ret = trace_seq_printf(s, "\n");
1186                 if (!ret)
1187                         return 0;
1188                 break;
1189         case TRACE_CTX:
1190                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1191                         state_to_char[entry->ctx.prev_state] : 'X';
1192                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1193                                        entry->ctx.prev_pid,
1194                                        entry->ctx.prev_prio,
1195                                        S,
1196                                        entry->ctx.next_pid,
1197                                        entry->ctx.next_prio);
1198                 if (!ret)
1199                         return 0;
1200                 break;
1201         }
1202         return 1;
1203 }
1204
1205 static int trace_empty(struct trace_iterator *iter)
1206 {
1207         struct trace_array_cpu *data;
1208         int cpu;
1209
1210         for_each_possible_cpu(cpu) {
1211                 data = iter->tr->data[cpu];
1212
1213                 if (head_page(data) && data->trace_idx &&
1214                     (data->trace_tail != data->trace_head ||
1215                      data->trace_tail_idx != data->trace_head_idx))
1216                         return 0;
1217         }
1218         return 1;
1219 }
1220
1221 static int s_show(struct seq_file *m, void *v)
1222 {
1223         struct trace_iterator *iter = v;
1224
1225         if (iter->ent == NULL) {
1226                 if (iter->tr) {
1227                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1228                         seq_puts(m, "#\n");
1229                 }
1230                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1231                         /* print nothing if the buffers are empty */
1232                         if (trace_empty(iter))
1233                                 return 0;
1234                         print_trace_header(m, iter);
1235                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1236                                 print_lat_help_header(m);
1237                 } else {
1238                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1239                                 print_func_help_header(m);
1240                 }
1241         } else {
1242                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1243                         print_lat_fmt(iter, iter->idx, iter->cpu);
1244                 else
1245                         print_trace_fmt(iter);
1246                 trace_print_seq(m, &iter->seq);
1247         }
1248
1249         return 0;
1250 }
1251
1252 static struct seq_operations tracer_seq_ops = {
1253         .start          = s_start,
1254         .next           = s_next,
1255         .stop           = s_stop,
1256         .show           = s_show,
1257 };
1258
1259 static struct trace_iterator notrace *
1260 __tracing_open(struct inode *inode, struct file *file, int *ret)
1261 {
1262         struct trace_iterator *iter;
1263
1264         if (tracing_disabled) {
1265                 *ret = -ENODEV;
1266                 return NULL;
1267         }
1268
1269         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1270         if (!iter) {
1271                 *ret = -ENOMEM;
1272                 goto out;
1273         }
1274
1275         mutex_lock(&trace_types_lock);
1276         if (current_trace && current_trace->print_max)
1277                 iter->tr = &max_tr;
1278         else
1279                 iter->tr = inode->i_private;
1280         iter->trace = current_trace;
1281         iter->pos = -1;
1282
1283         /* TODO stop tracer */
1284         *ret = seq_open(file, &tracer_seq_ops);
1285         if (!*ret) {
1286                 struct seq_file *m = file->private_data;
1287                 m->private = iter;
1288
1289                 /* stop the trace while dumping */
1290                 if (iter->tr->ctrl)
1291                         tracer_enabled = 0;
1292
1293                 if (iter->trace && iter->trace->open)
1294                         iter->trace->open(iter);
1295         } else {
1296                 kfree(iter);
1297                 iter = NULL;
1298         }
1299         mutex_unlock(&trace_types_lock);
1300
1301  out:
1302         return iter;
1303 }
1304
1305 int tracing_open_generic(struct inode *inode, struct file *filp)
1306 {
1307         if (tracing_disabled)
1308                 return -ENODEV;
1309
1310         filp->private_data = inode->i_private;
1311         return 0;
1312 }
1313
1314 int tracing_release(struct inode *inode, struct file *file)
1315 {
1316         struct seq_file *m = (struct seq_file *)file->private_data;
1317         struct trace_iterator *iter = m->private;
1318
1319         mutex_lock(&trace_types_lock);
1320         if (iter->trace && iter->trace->close)
1321                 iter->trace->close(iter);
1322
1323         /* reenable tracing if it was previously enabled */
1324         if (iter->tr->ctrl)
1325                 tracer_enabled = 1;
1326         mutex_unlock(&trace_types_lock);
1327
1328         seq_release(inode, file);
1329         kfree(iter);
1330         return 0;
1331 }
1332
1333 static int tracing_open(struct inode *inode, struct file *file)
1334 {
1335         int ret;
1336
1337         __tracing_open(inode, file, &ret);
1338
1339         return ret;
1340 }
1341
1342 static int tracing_lt_open(struct inode *inode, struct file *file)
1343 {
1344         struct trace_iterator *iter;
1345         int ret;
1346
1347         iter = __tracing_open(inode, file, &ret);
1348
1349         if (!ret)
1350                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1351
1352         return ret;
1353 }
1354
1355
1356 static notrace void *
1357 t_next(struct seq_file *m, void *v, loff_t *pos)
1358 {
1359         struct tracer *t = m->private;
1360
1361         (*pos)++;
1362
1363         if (t)
1364                 t = t->next;
1365
1366         m->private = t;
1367
1368         return t;
1369 }
1370
1371 static void *t_start(struct seq_file *m, loff_t *pos)
1372 {
1373         struct tracer *t = m->private;
1374         loff_t l = 0;
1375
1376         mutex_lock(&trace_types_lock);
1377         for (; t && l < *pos; t = t_next(m, t, &l))
1378                 ;
1379
1380         return t;
1381 }
1382
1383 static void t_stop(struct seq_file *m, void *p)
1384 {
1385         mutex_unlock(&trace_types_lock);
1386 }
1387
1388 static int t_show(struct seq_file *m, void *v)
1389 {
1390         struct tracer *t = v;
1391
1392         if (!t)
1393                 return 0;
1394
1395         seq_printf(m, "%s", t->name);
1396         if (t->next)
1397                 seq_putc(m, ' ');
1398         else
1399                 seq_putc(m, '\n');
1400
1401         return 0;
1402 }
1403
1404 static struct seq_operations show_traces_seq_ops = {
1405         .start          = t_start,
1406         .next           = t_next,
1407         .stop           = t_stop,
1408         .show           = t_show,
1409 };
1410
1411 static int show_traces_open(struct inode *inode, struct file *file)
1412 {
1413         int ret;
1414
1415         if (tracing_disabled)
1416                 return -ENODEV;
1417
1418         ret = seq_open(file, &show_traces_seq_ops);
1419         if (!ret) {
1420                 struct seq_file *m = file->private_data;
1421                 m->private = trace_types;
1422         }
1423
1424         return ret;
1425 }
1426
1427 static struct file_operations tracing_fops = {
1428         .open           = tracing_open,
1429         .read           = seq_read,
1430         .llseek         = seq_lseek,
1431         .release        = tracing_release,
1432 };
1433
1434 static struct file_operations tracing_lt_fops = {
1435         .open           = tracing_lt_open,
1436         .read           = seq_read,
1437         .llseek         = seq_lseek,
1438         .release        = tracing_release,
1439 };
1440
1441 static struct file_operations show_traces_fops = {
1442         .open = show_traces_open,
1443         .read = seq_read,
1444         .release = seq_release,
1445 };
1446
1447 static ssize_t
1448 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1449                        size_t cnt, loff_t *ppos)
1450 {
1451         char *buf;
1452         int r = 0;
1453         int len = 0;
1454         int i;
1455
1456         /* calulate max size */
1457         for (i = 0; trace_options[i]; i++) {
1458                 len += strlen(trace_options[i]);
1459                 len += 3; /* "no" and space */
1460         }
1461
1462         /* +2 for \n and \0 */
1463         buf = kmalloc(len + 2, GFP_KERNEL);
1464         if (!buf)
1465                 return -ENOMEM;
1466
1467         for (i = 0; trace_options[i]; i++) {
1468                 if (trace_flags & (1 << i))
1469                         r += sprintf(buf + r, "%s ", trace_options[i]);
1470                 else
1471                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1472         }
1473
1474         r += sprintf(buf + r, "\n");
1475         WARN_ON(r >= len + 2);
1476
1477         r = simple_read_from_buffer(ubuf, cnt, ppos,
1478                                     buf, r);
1479
1480         kfree(buf);
1481
1482         return r;
1483 }
1484
1485 static ssize_t
1486 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1487                         size_t cnt, loff_t *ppos)
1488 {
1489         char buf[64];
1490         char *cmp = buf;
1491         int neg = 0;
1492         int i;
1493
1494         if (cnt > 63)
1495                 cnt = 63;
1496
1497         if (copy_from_user(&buf, ubuf, cnt))
1498                 return -EFAULT;
1499
1500         buf[cnt] = 0;
1501
1502         if (strncmp(buf, "no", 2) == 0) {
1503                 neg = 1;
1504                 cmp += 2;
1505         }
1506
1507         for (i = 0; trace_options[i]; i++) {
1508                 int len = strlen(trace_options[i]);
1509
1510                 if (strncmp(cmp, trace_options[i], len) == 0) {
1511                         if (neg)
1512                                 trace_flags &= ~(1 << i);
1513                         else
1514                                 trace_flags |= (1 << i);
1515                         break;
1516                 }
1517         }
1518
1519         filp->f_pos += cnt;
1520
1521         return cnt;
1522 }
1523
1524 static struct file_operations tracing_iter_fops = {
1525         .open = tracing_open_generic,
1526         .read = tracing_iter_ctrl_read,
1527         .write = tracing_iter_ctrl_write,
1528 };
1529
1530 static const char readme_msg[] =
1531         "tracing mini-HOWTO:\n\n"
1532         "# mkdir /debug\n"
1533         "# mount -t debugfs nodev /debug\n\n"
1534         "# cat /debug/tracing/available_tracers\n"
1535         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1536         "# cat /debug/tracing/current_tracer\n"
1537         "none\n"
1538         "# echo sched_switch > /debug/tracing/current_tracer\n"
1539         "# cat /debug/tracing/current_tracer\n"
1540         "sched_switch\n"
1541         "# cat /debug/tracing/iter_ctrl\n"
1542         "noprint-parent nosym-offset nosym-addr noverbose\n"
1543         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1544         "# echo 1 > /debug/tracing/tracing_enabled\n"
1545         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1546         "echo 0 > /debug/tracing/tracing_enabled\n"
1547 ;
1548
1549 static ssize_t
1550 tracing_readme_read(struct file *filp, char __user *ubuf,
1551                        size_t cnt, loff_t *ppos)
1552 {
1553         return simple_read_from_buffer(ubuf, cnt, ppos,
1554                                         readme_msg, strlen(readme_msg));
1555 }
1556
1557 static struct file_operations tracing_readme_fops = {
1558         .open = tracing_open_generic,
1559         .read = tracing_readme_read,
1560 };
1561
1562
1563 static ssize_t
1564 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1565                   size_t cnt, loff_t *ppos)
1566 {
1567         struct trace_array *tr = filp->private_data;
1568         char buf[64];
1569         int r;
1570
1571         r = sprintf(buf, "%ld\n", tr->ctrl);
1572         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1573 }
1574
1575 static ssize_t
1576 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1577                    size_t cnt, loff_t *ppos)
1578 {
1579         struct trace_array *tr = filp->private_data;
1580         long val;
1581         char buf[64];
1582
1583         if (cnt > 63)
1584                 cnt = 63;
1585
1586         if (copy_from_user(&buf, ubuf, cnt))
1587                 return -EFAULT;
1588
1589         buf[cnt] = 0;
1590
1591         val = simple_strtoul(buf, NULL, 10);
1592
1593         val = !!val;
1594
1595         mutex_lock(&trace_types_lock);
1596         if (tr->ctrl ^ val) {
1597                 if (val)
1598                         tracer_enabled = 1;
1599                 else
1600                         tracer_enabled = 0;
1601
1602                 tr->ctrl = val;
1603
1604                 if (current_trace && current_trace->ctrl_update)
1605                         current_trace->ctrl_update(tr);
1606         }
1607         mutex_unlock(&trace_types_lock);
1608
1609         filp->f_pos += cnt;
1610
1611         return cnt;
1612 }
1613
1614 static ssize_t
1615 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1616                        size_t cnt, loff_t *ppos)
1617 {
1618         char buf[max_tracer_type_len+2];
1619         int r;
1620
1621         mutex_lock(&trace_types_lock);
1622         if (current_trace)
1623                 r = sprintf(buf, "%s\n", current_trace->name);
1624         else
1625                 r = sprintf(buf, "\n");
1626         mutex_unlock(&trace_types_lock);
1627
1628         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1629 }
1630
1631 static ssize_t
1632 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1633                         size_t cnt, loff_t *ppos)
1634 {
1635         struct trace_array *tr = &global_trace;
1636         struct tracer *t;
1637         char buf[max_tracer_type_len+1];
1638         int i;
1639
1640         if (cnt > max_tracer_type_len)
1641                 cnt = max_tracer_type_len;
1642
1643         if (copy_from_user(&buf, ubuf, cnt))
1644                 return -EFAULT;
1645
1646         buf[cnt] = 0;
1647
1648         /* strip ending whitespace. */
1649         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1650                 buf[i] = 0;
1651
1652         mutex_lock(&trace_types_lock);
1653         for (t = trace_types; t; t = t->next) {
1654                 if (strcmp(t->name, buf) == 0)
1655                         break;
1656         }
1657         if (!t || t == current_trace)
1658                 goto out;
1659
1660         if (current_trace && current_trace->reset)
1661                 current_trace->reset(tr);
1662
1663         current_trace = t;
1664         if (t->init)
1665                 t->init(tr);
1666
1667  out:
1668         mutex_unlock(&trace_types_lock);
1669
1670         filp->f_pos += cnt;
1671
1672         return cnt;
1673 }
1674
1675 static ssize_t
1676 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1677                      size_t cnt, loff_t *ppos)
1678 {
1679         unsigned long *ptr = filp->private_data;
1680         char buf[64];
1681         int r;
1682
1683         r = snprintf(buf, 64, "%ld\n",
1684                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1685         if (r > 64)
1686                 r = 64;
1687         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1688 }
1689
1690 static ssize_t
1691 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1692                       size_t cnt, loff_t *ppos)
1693 {
1694         long *ptr = filp->private_data;
1695         long val;
1696         char buf[64];
1697
1698         if (cnt > 63)
1699                 cnt = 63;
1700
1701         if (copy_from_user(&buf, ubuf, cnt))
1702                 return -EFAULT;
1703
1704         buf[cnt] = 0;
1705
1706         val = simple_strtoul(buf, NULL, 10);
1707
1708         *ptr = val * 1000;
1709
1710         return cnt;
1711 }
1712
1713 static atomic_t tracing_reader;
1714
1715 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1716 {
1717         struct trace_iterator *iter;
1718
1719         if (tracing_disabled)
1720                 return -ENODEV;
1721
1722         /* We only allow for reader of the pipe */
1723         if (atomic_inc_return(&tracing_reader) != 1) {
1724                 atomic_dec(&tracing_reader);
1725                 return -EBUSY;
1726         }
1727
1728         /* create a buffer to store the information to pass to userspace */
1729         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1730         if (!iter)
1731                 return -ENOMEM;
1732
1733         iter->tr = &global_trace;
1734
1735         filp->private_data = iter;
1736
1737         return 0;
1738 }
1739
1740 static int tracing_release_pipe(struct inode *inode, struct file *file)
1741 {
1742         struct trace_iterator *iter = file->private_data;
1743
1744         kfree(iter);
1745         atomic_dec(&tracing_reader);
1746
1747         return 0;
1748 }
1749
1750 /*
1751  * Consumer reader.
1752  */
1753 static ssize_t
1754 tracing_read_pipe(struct file *filp, char __user *ubuf,
1755                   size_t cnt, loff_t *ppos)
1756 {
1757         struct trace_iterator *iter = filp->private_data;
1758         struct trace_array_cpu *data;
1759         static cpumask_t mask;
1760         struct trace_entry *entry;
1761         static int start;
1762         unsigned long flags;
1763         int read = 0;
1764         int cpu;
1765         int len;
1766         int ret;
1767
1768         /* return any leftover data */
1769         if (iter->seq.len > start) {
1770                 len = iter->seq.len - start;
1771                 if (cnt > len)
1772                         cnt = len;
1773                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1774                 if (ret)
1775                         cnt = -EFAULT;
1776
1777                 start += len;
1778
1779                 return cnt;
1780         }
1781
1782         trace_seq_reset(&iter->seq);
1783         start = 0;
1784
1785         while (trace_empty(iter)) {
1786                 /*
1787                  * This is a make-shift waitqueue. The reason we don't use
1788                  * an actual wait queue is because:
1789                  *  1) we only ever have one waiter
1790                  *  2) the tracing, traces all functions, we don't want
1791                  *     the overhead of calling wake_up and friends
1792                  *     (and tracing them too)
1793                  *     Anyway, this is really very primitive wakeup.
1794                  */
1795                 set_current_state(TASK_INTERRUPTIBLE);
1796                 iter->tr->waiter = current;
1797
1798                 /* sleep for one second, and try again. */
1799                 schedule_timeout(HZ);
1800
1801                 iter->tr->waiter = NULL;
1802
1803                 if (signal_pending(current))
1804                         return -EINTR;
1805
1806                 /*
1807                  * We block until we read something and tracing is disabled.
1808                  * We still block if tracing is disabled, but we have never
1809                  * read anything. This allows a user to cat this file, and
1810                  * then enable tracing. But after we have read something,
1811                  * we give an EOF when tracing is again disabled.
1812                  *
1813                  * iter->pos will be 0 if we haven't read anything.
1814                  */
1815                 if (!tracer_enabled && iter->pos)
1816                         break;
1817
1818                 continue;
1819         }
1820
1821         /* stop when tracing is finished */
1822         if (trace_empty(iter))
1823                 return 0;
1824
1825         if (cnt >= PAGE_SIZE)
1826                 cnt = PAGE_SIZE - 1;
1827
1828         memset(iter, 0, sizeof(*iter));
1829         iter->tr = &global_trace;
1830         iter->pos = -1;
1831
1832         /*
1833          * We need to stop all tracing on all CPUS to read the
1834          * the next buffer. This is a bit expensive, but is
1835          * not done often. We fill all what we can read,
1836          * and then release the locks again.
1837          */
1838
1839         cpus_clear(mask);
1840         local_irq_save(flags);
1841         for_each_possible_cpu(cpu) {
1842                 data = iter->tr->data[cpu];
1843
1844                 if (!head_page(data) || !data->trace_idx)
1845                         continue;
1846
1847                 atomic_inc(&data->disabled);
1848                 spin_lock(&data->lock);
1849                 cpu_set(cpu, mask);
1850         }
1851
1852         while ((entry = find_next_entry(iter, &cpu))) {
1853
1854                 if (!entry)
1855                         break;
1856
1857                 iter->ent = entry;
1858                 iter->cpu = cpu;
1859
1860                 ret = print_trace_fmt(iter);
1861                 if (!ret)
1862                         break;
1863
1864                 trace_consume(iter);
1865
1866                 if (iter->seq.len >= cnt)
1867                         break;
1868
1869         }
1870
1871         for_each_cpu_mask(cpu, mask) {
1872                 data = iter->tr->data[cpu];
1873                 spin_unlock(&data->lock);
1874                 atomic_dec(&data->disabled);
1875         }
1876         local_irq_restore(flags);
1877
1878         /* Now copy what we have to the user */
1879         read = iter->seq.len;
1880         if (read > cnt)
1881                 read = cnt;
1882
1883         ret = copy_to_user(ubuf, iter->seq.buffer, read);
1884
1885         if (read < iter->seq.len)
1886                 start = read;
1887         else
1888                 trace_seq_reset(&iter->seq);
1889
1890         if (ret)
1891                 read = -EFAULT;
1892
1893         return read;
1894 }
1895
1896 static struct file_operations tracing_max_lat_fops = {
1897         .open           = tracing_open_generic,
1898         .read           = tracing_max_lat_read,
1899         .write          = tracing_max_lat_write,
1900 };
1901
1902 static struct file_operations tracing_ctrl_fops = {
1903         .open           = tracing_open_generic,
1904         .read           = tracing_ctrl_read,
1905         .write          = tracing_ctrl_write,
1906 };
1907
1908 static struct file_operations set_tracer_fops = {
1909         .open           = tracing_open_generic,
1910         .read           = tracing_set_trace_read,
1911         .write          = tracing_set_trace_write,
1912 };
1913
1914 static struct file_operations tracing_pipe_fops = {
1915         .open           = tracing_open_pipe,
1916         .read           = tracing_read_pipe,
1917         .release        = tracing_release_pipe,
1918 };
1919
1920 #ifdef CONFIG_DYNAMIC_FTRACE
1921
1922 static ssize_t
1923 tracing_read_long(struct file *filp, char __user *ubuf,
1924                   size_t cnt, loff_t *ppos)
1925 {
1926         unsigned long *p = filp->private_data;
1927         char buf[64];
1928         int r;
1929
1930         r = sprintf(buf, "%ld\n", *p);
1931
1932         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1933 }
1934
1935 static struct file_operations tracing_read_long_fops = {
1936         .open           = tracing_open_generic,
1937         .read           = tracing_read_long,
1938 };
1939 #endif
1940
1941 static struct dentry *d_tracer;
1942
1943 struct dentry *tracing_init_dentry(void)
1944 {
1945         static int once;
1946
1947         if (d_tracer)
1948                 return d_tracer;
1949
1950         d_tracer = debugfs_create_dir("tracing", NULL);
1951
1952         if (!d_tracer && !once) {
1953                 once = 1;
1954                 pr_warning("Could not create debugfs directory 'tracing'\n");
1955                 return NULL;
1956         }
1957
1958         return d_tracer;
1959 }
1960
1961 #ifdef CONFIG_FTRACE_SELFTEST
1962 /* Let selftest have access to static functions in this file */
1963 #include "trace_selftest.c"
1964 #endif
1965
1966 static __init void tracer_init_debugfs(void)
1967 {
1968         struct dentry *d_tracer;
1969         struct dentry *entry;
1970
1971         d_tracer = tracing_init_dentry();
1972
1973         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
1974                                     &global_trace, &tracing_ctrl_fops);
1975         if (!entry)
1976                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
1977
1978         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
1979                                     NULL, &tracing_iter_fops);
1980         if (!entry)
1981                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
1982
1983         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
1984                                     &global_trace, &tracing_lt_fops);
1985         if (!entry)
1986                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
1987
1988         entry = debugfs_create_file("trace", 0444, d_tracer,
1989                                     &global_trace, &tracing_fops);
1990         if (!entry)
1991                 pr_warning("Could not create debugfs 'trace' entry\n");
1992
1993         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
1994                                     &global_trace, &show_traces_fops);
1995         if (!entry)
1996                 pr_warning("Could not create debugfs 'trace' entry\n");
1997
1998         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
1999                                     &global_trace, &set_tracer_fops);
2000         if (!entry)
2001                 pr_warning("Could not create debugfs 'trace' entry\n");
2002
2003         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2004                                     &tracing_max_latency,
2005                                     &tracing_max_lat_fops);
2006         if (!entry)
2007                 pr_warning("Could not create debugfs "
2008                            "'tracing_max_latency' entry\n");
2009
2010         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2011                                     &tracing_thresh, &tracing_max_lat_fops);
2012         if (!entry)
2013                 pr_warning("Could not create debugfs "
2014                            "'tracing_threash' entry\n");
2015         entry = debugfs_create_file("README", 0644, d_tracer,
2016                                     NULL, &tracing_readme_fops);
2017         if (!entry)
2018                 pr_warning("Could not create debugfs 'README' entry\n");
2019
2020         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2021                                     NULL, &tracing_pipe_fops);
2022         if (!entry)
2023                 pr_warning("Could not create debugfs "
2024                            "'tracing_threash' entry\n");
2025
2026 #ifdef CONFIG_DYNAMIC_FTRACE
2027         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2028                                     &ftrace_update_tot_cnt,
2029                                     &tracing_read_long_fops);
2030         if (!entry)
2031                 pr_warning("Could not create debugfs "
2032                            "'dyn_ftrace_total_info' entry\n");
2033 #endif
2034 }
2035
2036 /* dummy trace to disable tracing */
2037 static struct tracer no_tracer __read_mostly =
2038 {
2039         .name           = "none",
2040 };
2041
2042 static int trace_alloc_page(void)
2043 {
2044         struct trace_array_cpu *data;
2045         struct page *page, *tmp;
2046         LIST_HEAD(pages);
2047         void *array;
2048         int i;
2049
2050         /* first allocate a page for each CPU */
2051         for_each_possible_cpu(i) {
2052                 array = (void *)__get_free_page(GFP_KERNEL);
2053                 if (array == NULL) {
2054                         printk(KERN_ERR "tracer: failed to allocate page"
2055                                "for trace buffer!\n");
2056                         goto free_pages;
2057                 }
2058
2059                 page = virt_to_page(array);
2060                 list_add(&page->lru, &pages);
2061
2062 /* Only allocate if we are actually using the max trace */
2063 #ifdef CONFIG_TRACER_MAX_TRACE
2064                 array = (void *)__get_free_page(GFP_KERNEL);
2065                 if (array == NULL) {
2066                         printk(KERN_ERR "tracer: failed to allocate page"
2067                                "for trace buffer!\n");
2068                         goto free_pages;
2069                 }
2070                 page = virt_to_page(array);
2071                 list_add(&page->lru, &pages);
2072 #endif
2073         }
2074
2075         /* Now that we successfully allocate a page per CPU, add them */
2076         for_each_possible_cpu(i) {
2077                 data = global_trace.data[i];
2078                 spin_lock_init(&data->lock);
2079                 lockdep_set_class(&data->lock, &data->lock_key);
2080                 page = list_entry(pages.next, struct page, lru);
2081                 list_del_init(&page->lru);
2082                 list_add_tail(&page->lru, &data->trace_pages);
2083                 ClearPageLRU(page);
2084
2085 #ifdef CONFIG_TRACER_MAX_TRACE
2086                 data = max_tr.data[i];
2087                 spin_lock_init(&data->lock);
2088                 lockdep_set_class(&data->lock, &data->lock_key);
2089                 page = list_entry(pages.next, struct page, lru);
2090                 list_del_init(&page->lru);
2091                 list_add_tail(&page->lru, &data->trace_pages);
2092                 SetPageLRU(page);
2093 #endif
2094         }
2095         global_trace.entries += ENTRIES_PER_PAGE;
2096
2097         return 0;
2098
2099  free_pages:
2100         list_for_each_entry_safe(page, tmp, &pages, lru) {
2101                 list_del_init(&page->lru);
2102                 __free_page(page);
2103         }
2104         return -ENOMEM;
2105 }
2106
2107 __init static int tracer_alloc_buffers(void)
2108 {
2109         struct trace_array_cpu *data;
2110         void *array;
2111         struct page *page;
2112         int pages = 0;
2113         int ret = -ENOMEM;
2114         int i;
2115
2116         /* Allocate the first page for all buffers */
2117         for_each_possible_cpu(i) {
2118                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2119                 max_tr.data[i] = &per_cpu(max_data, i);
2120
2121                 array = (void *)__get_free_page(GFP_KERNEL);
2122                 if (array == NULL) {
2123                         printk(KERN_ERR "tracer: failed to allocate page"
2124                                "for trace buffer!\n");
2125                         goto free_buffers;
2126                 }
2127
2128                 /* set the array to the list */
2129                 INIT_LIST_HEAD(&data->trace_pages);
2130                 page = virt_to_page(array);
2131                 list_add(&page->lru, &data->trace_pages);
2132                 /* use the LRU flag to differentiate the two buffers */
2133                 ClearPageLRU(page);
2134
2135 /* Only allocate if we are actually using the max trace */
2136 #ifdef CONFIG_TRACER_MAX_TRACE
2137                 array = (void *)__get_free_page(GFP_KERNEL);
2138                 if (array == NULL) {
2139                         printk(KERN_ERR "tracer: failed to allocate page"
2140                                "for trace buffer!\n");
2141                         goto free_buffers;
2142                 }
2143
2144                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2145                 page = virt_to_page(array);
2146                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2147                 SetPageLRU(page);
2148 #endif
2149         }
2150
2151         /*
2152          * Since we allocate by orders of pages, we may be able to
2153          * round up a bit.
2154          */
2155         global_trace.entries = ENTRIES_PER_PAGE;
2156         pages++;
2157
2158         while (global_trace.entries < trace_nr_entries) {
2159                 if (trace_alloc_page())
2160                         break;
2161                 pages++;
2162         }
2163         max_tr.entries = global_trace.entries;
2164
2165         pr_info("tracer: %d pages allocated for %ld",
2166                 pages, trace_nr_entries);
2167         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2168         pr_info("   actual entries %ld\n", global_trace.entries);
2169
2170         tracer_init_debugfs();
2171
2172         trace_init_cmdlines();
2173
2174         register_tracer(&no_tracer);
2175         current_trace = &no_tracer;
2176
2177         /* All seems OK, enable tracing */
2178         tracing_disabled = 0;
2179
2180         return 0;
2181
2182  free_buffers:
2183         for (i-- ; i >= 0; i--) {
2184                 struct page *page, *tmp;
2185                 struct trace_array_cpu *data = global_trace.data[i];
2186
2187                 if (data) {
2188                         list_for_each_entry_safe(page, tmp,
2189                                                  &data->trace_pages, lru) {
2190                                 list_del_init(&page->lru);
2191                                 __free_page(page);
2192                         }
2193                 }
2194
2195 #ifdef CONFIG_TRACER_MAX_TRACE
2196                 data = max_tr.data[i];
2197                 if (data) {
2198                         list_for_each_entry_safe(page, tmp,
2199                                                  &data->trace_pages, lru) {
2200                                 list_del_init(&page->lru);
2201                                 __free_page(page);
2202                         }
2203                 }
2204 #endif
2205         }
2206         return ret;
2207 }
2208 fs_initcall(tracer_alloc_buffers);