git.oblomov.eu Git - linux-2.6/blob - 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 static atomic_t                 tracer_counter;
  46 static struct trace_array       global_trace;
  47
  48 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  49
  50 static struct trace_array       max_tr;
  51
  52 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  53
  54 static int                      tracer_enabled;
  55 static unsigned long            trace_nr_entries = 16384UL;
  56
  57 static struct tracer            *trace_types __read_mostly;
  58 static struct tracer            *current_trace __read_mostly;
  59 static int                      max_tracer_type_len;
  60
  61 static DEFINE_MUTEX(trace_types_lock);
  62
  63 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  64
  65 static int __init set_nr_entries(char *str)
  66 {
  67         if (!str)
  68                 return 0;
  69         trace_nr_entries = simple_strtoul(str, &str, 0);
  70         return 1;
  71 }
  72 __setup("trace_entries=", set_nr_entries);
  73
  74 unsigned long nsecs_to_usecs(unsigned long nsecs)
  75 {
  76         return nsecs / 1000;
  77 }
  78
  79 enum trace_type {
  80         __TRACE_FIRST_TYPE = 0,
  81
  82         TRACE_FN,
  83         TRACE_CTX,
  84
  85         __TRACE_LAST_TYPE
  86 };
  87
  88 enum trace_flag_type {
  89         TRACE_FLAG_IRQS_OFF             = 0x01,
  90         TRACE_FLAG_NEED_RESCHED         = 0x02,
  91         TRACE_FLAG_HARDIRQ              = 0x04,
  92         TRACE_FLAG_SOFTIRQ              = 0x08,
  93 };
  94
  95 enum trace_iterator_flags {
  96         TRACE_ITER_PRINT_PARENT         = 0x01,
  97         TRACE_ITER_SYM_OFFSET           = 0x02,
  98         TRACE_ITER_SYM_ADDR             = 0x04,
  99         TRACE_ITER_VERBOSE              = 0x08,
 100 };
 101
 102 #define TRACE_ITER_SYM_MASK \
 103         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 104
 105 /* These must match the bit postions above */
 106 static const char *trace_options[] = {
 107         "print-parent",
 108         "sym-offset",
 109         "sym-addr",
 110         "verbose",
 111         NULL
 112 };
 113
 114 static unsigned trace_flags;
 115
 116 static DEFINE_SPINLOCK(ftrace_max_lock);
 117
 118 /*
 119  * Copy the new maximum trace into the separate maximum-trace
 120  * structure. (this way the maximum trace is permanently saved,
 121  * for later retrieval via /debugfs/tracing/latency_trace)
 122  */
 123 static notrace void
 124 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 125 {
 126         struct trace_array_cpu *data = tr->data[cpu];
 127
 128         max_tr.cpu = cpu;
 129         max_tr.time_start = data->preempt_timestamp;
 130
 131         data = max_tr.data[cpu];
 132         data->saved_latency = tracing_max_latency;
 133
 134         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 135         data->pid = tsk->pid;
 136         data->uid = tsk->uid;
 137         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 138         data->policy = tsk->policy;
 139         data->rt_priority = tsk->rt_priority;
 140
 141         /* record this tasks comm */
 142         tracing_record_cmdline(current);
 143 }
 144
 145 void check_pages(struct trace_array_cpu *data)
 146 {
 147         struct page *page, *tmp;
 148
 149         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 150         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 151
 152         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 153                 BUG_ON(page->lru.next->prev != &page->lru);
 154                 BUG_ON(page->lru.prev->next != &page->lru);
 155         }
 156 }
 157
 158 void *head_page(struct trace_array_cpu *data)
 159 {
 160         struct page *page;
 161
 162         check_pages(data);
 163         if (list_empty(&data->trace_pages))
 164                 return NULL;
 165
 166         page = list_entry(data->trace_pages.next, struct page, lru);
 167         BUG_ON(&page->lru == &data->trace_pages);
 168
 169         return page_address(page);
 170 }
 171
 172 static notrace int
 173 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 174 {
 175         int len = (PAGE_SIZE - 1) - s->len;
 176         va_list ap;
 177         int ret;
 178
 179         if (!len)
 180                 return 0;
 181
 182         va_start(ap, fmt);
 183         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 184         va_end(ap);
 185
 186         /* If we can't write it all, don't bother writing anything */
 187         if (ret > len)
 188                 return 0;
 189
 190         s->len += ret;
 191
 192         return len;
 193 }
 194
 195 static notrace int
 196 trace_seq_puts(struct trace_seq *s, const char *str)
 197 {
 198         int len = strlen(str);
 199
 200         if (len > ((PAGE_SIZE - 1) - s->len))
 201                 return 0;
 202
 203         memcpy(s->buffer + s->len, str, len);
 204         s->len += len;
 205
 206         return len;
 207 }
 208
 209 static notrace int
 210 trace_seq_putc(struct trace_seq *s, unsigned char c)
 211 {
 212         if (s->len >= (PAGE_SIZE - 1))
 213                 return 0;
 214
 215         s->buffer[s->len++] = c;
 216
 217         return 1;
 218 }
 219
 220 static notrace void
 221 trace_seq_reset(struct trace_seq *s)
 222 {
 223         s->len = 0;
 224 }
 225
 226 static notrace void
 227 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 228 {
 229         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 230
 231         s->buffer[len] = 0;
 232         seq_puts(m, s->buffer);
 233
 234         trace_seq_reset(s);
 235 }
 236
 237 notrace static void
 238 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 239 {
 240         struct list_head flip_pages;
 241
 242         INIT_LIST_HEAD(&flip_pages);
 243
 244         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 245                 sizeof(struct trace_array_cpu) -
 246                 offsetof(struct trace_array_cpu, trace_head_idx));
 247
 248         check_pages(tr1);
 249         check_pages(tr2);
 250         list_splice_init(&tr1->trace_pages, &flip_pages);
 251         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 252         list_splice_init(&flip_pages, &tr2->trace_pages);
 253         BUG_ON(!list_empty(&flip_pages));
 254         check_pages(tr1);
 255         check_pages(tr2);
 256 }
 257
 258 notrace void
 259 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 260 {
 261         struct trace_array_cpu *data;
 262         int i;
 263
 264         WARN_ON_ONCE(!irqs_disabled());
 265         spin_lock(&ftrace_max_lock);
 266         /* clear out all the previous traces */
 267         for_each_possible_cpu(i) {
 268                 data = tr->data[i];
 269                 flip_trace(max_tr.data[i], data);
 270                 tracing_reset(data);
 271         }
 272
 273         __update_max_tr(tr, tsk, cpu);
 274         spin_unlock(&ftrace_max_lock);
 275 }
 276
 277 /**
 278  * update_max_tr_single - only copy one trace over, and reset the rest
 279  * @tr - tracer
 280  * @tsk - task with the latency
 281  * @cpu - the cpu of the buffer to copy.
 282  */
 283 notrace void
 284 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 285 {
 286         struct trace_array_cpu *data = tr->data[cpu];
 287         int i;
 288
 289         WARN_ON_ONCE(!irqs_disabled());
 290         spin_lock(&ftrace_max_lock);
 291         for_each_possible_cpu(i)
 292                 tracing_reset(max_tr.data[i]);
 293
 294         flip_trace(max_tr.data[cpu], data);
 295         tracing_reset(data);
 296
 297         __update_max_tr(tr, tsk, cpu);
 298         spin_unlock(&ftrace_max_lock);
 299 }
 300
 301 int register_tracer(struct tracer *type)
 302 {
 303         struct tracer *t;
 304         int len;
 305         int ret = 0;
 306
 307         if (!type->name) {
 308                 pr_info("Tracer must have a name\n");
 309                 return -1;
 310         }
 311
 312         mutex_lock(&trace_types_lock);
 313         for (t = trace_types; t; t = t->next) {
 314                 if (strcmp(type->name, t->name) == 0) {
 315                         /* already found */
 316                         pr_info("Trace %s already registered\n",
 317                                 type->name);
 318                         ret = -1;
 319                         goto out;
 320                 }
 321         }
 322
 323 #ifdef CONFIG_FTRACE_STARTUP_TEST
 324         if (type->selftest) {
 325                 struct tracer *saved_tracer = current_trace;
 326                 struct trace_array_cpu *data;
 327                 struct trace_array *tr = &global_trace;
 328                 int saved_ctrl = tr->ctrl;
 329                 int i;
 330                 /*
 331                  * Run a selftest on this tracer.
 332                  * Here we reset the trace buffer, and set the current
 333                  * tracer to be this tracer. The tracer can then run some
 334                  * internal tracing to verify that everything is in order.
 335                  * If we fail, we do not register this tracer.
 336                  */
 337                 for_each_possible_cpu(i) {
 338                         data = tr->data[i];
 339                         if (!head_page(data))
 340                                 continue;
 341                         tracing_reset(data);
 342                 }
 343                 current_trace = type;
 344                 tr->ctrl = 0;
 345                 /* the test is responsible for initializing and enabling */
 346                 pr_info("Testing tracer %s: ", type->name);
 347                 ret = type->selftest(type, tr);
 348                 /* the test is responsible for resetting too */
 349                 current_trace = saved_tracer;
 350                 tr->ctrl = saved_ctrl;
 351                 if (ret) {
 352                         printk(KERN_CONT "FAILED!\n");
 353                         goto out;
 354                 }
 355                 /* Only reset on passing, to avoid touching corrupted buffers */
 356                 for_each_possible_cpu(i) {
 357                         data = tr->data[i];
 358                         if (!head_page(data))
 359                                 continue;
 360                         tracing_reset(data);
 361                 }
 362                 printk(KERN_CONT "PASSED\n");
 363         }
 364 #endif
 365
 366         type->next = trace_types;
 367         trace_types = type;
 368         len = strlen(type->name);
 369         if (len > max_tracer_type_len)
 370                 max_tracer_type_len = len;
 371
 372  out:
 373         mutex_unlock(&trace_types_lock);
 374
 375         return ret;
 376 }
 377
 378 void unregister_tracer(struct tracer *type)
 379 {
 380         struct tracer **t;
 381         int len;
 382
 383         mutex_lock(&trace_types_lock);
 384         for (t = &trace_types; *t; t = &(*t)->next) {
 385                 if (*t == type)
 386                         goto found;
 387         }
 388         pr_info("Trace %s not registered\n", type->name);
 389         goto out;
 390
 391  found:
 392         *t = (*t)->next;
 393         if (strlen(type->name) != max_tracer_type_len)
 394                 goto out;
 395
 396         max_tracer_type_len = 0;
 397         for (t = &trace_types; *t; t = &(*t)->next) {
 398                 len = strlen((*t)->name);
 399                 if (len > max_tracer_type_len)
 400                         max_tracer_type_len = len;
 401         }
 402  out:
 403         mutex_unlock(&trace_types_lock);
 404 }
 405
 406 notrace void tracing_reset(struct trace_array_cpu *data)
 407 {
 408         data->trace_idx = 0;
 409         data->trace_head = data->trace_tail = head_page(data);
 410         data->trace_head_idx = 0;
 411         data->trace_tail_idx = 0;
 412 }
 413
 414 #ifdef CONFIG_FTRACE
 415 static notrace void
 416 function_trace_call(unsigned long ip, unsigned long parent_ip)
 417 {
 418         struct trace_array *tr = &global_trace;
 419         struct trace_array_cpu *data;
 420         unsigned long flags;
 421         long disabled;
 422         int cpu;
 423
 424         if (unlikely(!tracer_enabled))
 425                 return;
 426
 427         local_irq_save(flags);
 428         cpu = raw_smp_processor_id();
 429         data = tr->data[cpu];
 430         disabled = atomic_inc_return(&data->disabled);
 431
 432         if (likely(disabled == 1))
 433                 ftrace(tr, data, ip, parent_ip, flags);
 434
 435         atomic_dec(&data->disabled);
 436         local_irq_restore(flags);
 437 }
 438
 439 static struct ftrace_ops trace_ops __read_mostly =
 440 {
 441         .func = function_trace_call,
 442 };
 443 #endif
 444
 445 notrace void tracing_start_function_trace(void)
 446 {
 447         register_ftrace_function(&trace_ops);
 448 }
 449
 450 notrace void tracing_stop_function_trace(void)
 451 {
 452         unregister_ftrace_function(&trace_ops);
 453 }
 454
 455 #define SAVED_CMDLINES 128
 456 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 457 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 458 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 459 static int cmdline_idx;
 460 static DEFINE_SPINLOCK(trace_cmdline_lock);
 461 atomic_t trace_record_cmdline_disabled;
 462
 463 static void trace_init_cmdlines(void)
 464 {
 465         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 466         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 467         cmdline_idx = 0;
 468 }
 469
 470 notrace void trace_stop_cmdline_recording(void);
 471
 472 static notrace void trace_save_cmdline(struct task_struct *tsk)
 473 {
 474         unsigned map;
 475         unsigned idx;
 476
 477         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 478                 return;
 479
 480         /*
 481          * It's not the end of the world if we don't get
 482          * the lock, but we also don't want to spin
 483          * nor do we want to disable interrupts,
 484          * so if we miss here, then better luck next time.
 485          */
 486         if (!spin_trylock(&trace_cmdline_lock))
 487                 return;
 488
 489         idx = map_pid_to_cmdline[tsk->pid];
 490         if (idx >= SAVED_CMDLINES) {
 491                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 492
 493                 map = map_cmdline_to_pid[idx];
 494                 if (map <= PID_MAX_DEFAULT)
 495                         map_pid_to_cmdline[map] = (unsigned)-1;
 496
 497                 map_pid_to_cmdline[tsk->pid] = idx;
 498
 499                 cmdline_idx = idx;
 500         }
 501
 502         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 503
 504         spin_unlock(&trace_cmdline_lock);
 505 }
 506
 507 static notrace char *trace_find_cmdline(int pid)
 508 {
 509         char *cmdline = "<...>";
 510         unsigned map;
 511
 512         if (!pid)
 513                 return "<idle>";
 514
 515         if (pid > PID_MAX_DEFAULT)
 516                 goto out;
 517
 518         map = map_pid_to_cmdline[pid];
 519         if (map >= SAVED_CMDLINES)
 520                 goto out;
 521
 522         cmdline = saved_cmdlines[map];
 523
 524  out:
 525         return cmdline;
 526 }
 527
 528 notrace void tracing_record_cmdline(struct task_struct *tsk)
 529 {
 530         if (atomic_read(&trace_record_cmdline_disabled))
 531                 return;
 532
 533         trace_save_cmdline(tsk);
 534 }
 535
 536 static inline notrace struct list_head *
 537 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 538 {
 539         /*
 540          * Roundrobin - but skip the head (which is not a real page):
 541          */
 542         next = next->next;
 543         if (unlikely(next == &data->trace_pages))
 544                 next = next->next;
 545         BUG_ON(next == &data->trace_pages);
 546
 547         return next;
 548 }
 549
 550 static inline notrace void *
 551 trace_next_page(struct trace_array_cpu *data, void *addr)
 552 {
 553         struct list_head *next;
 554         struct page *page;
 555
 556         page = virt_to_page(addr);
 557
 558         next = trace_next_list(data, &page->lru);
 559         page = list_entry(next, struct page, lru);
 560
 561         return page_address(page);
 562 }
 563
 564 static inline notrace struct trace_entry *
 565 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 566 {
 567         unsigned long idx, idx_next;
 568         struct trace_entry *entry;
 569
 570         data->trace_idx++;
 571         idx = data->trace_head_idx;
 572         idx_next = idx + 1;
 573
 574         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 575
 576         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 577
 578         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 579                 data->trace_head = trace_next_page(data, data->trace_head);
 580                 idx_next = 0;
 581         }
 582
 583         if (data->trace_head == data->trace_tail &&
 584             idx_next == data->trace_tail_idx) {
 585                 /* overrun */
 586                 data->trace_tail_idx++;
 587                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 588                         data->trace_tail =
 589                                 trace_next_page(data, data->trace_tail);
 590                         data->trace_tail_idx = 0;
 591                 }
 592         }
 593
 594         data->trace_head_idx = idx_next;
 595
 596         return entry;
 597 }
 598
 599 static inline notrace void
 600 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 601 {
 602         struct task_struct *tsk = current;
 603         unsigned long pc;
 604
 605         pc = preempt_count();
 606
 607         entry->idx              = atomic_inc_return(&tracer_counter);
 608         entry->preempt_count    = pc & 0xff;
 609         entry->pid              = tsk->pid;
 610         entry->t                = now(raw_smp_processor_id());
 611         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 612                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 613                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 614                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 615 }
 616
 617 notrace void
 618 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 619        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 620 {
 621         struct trace_entry *entry;
 622
 623         spin_lock(&data->lock);
 624         entry                   = tracing_get_trace_entry(tr, data);
 625         tracing_generic_entry_update(entry, flags);
 626         entry->type             = TRACE_FN;
 627         entry->fn.ip            = ip;
 628         entry->fn.parent_ip     = parent_ip;
 629         spin_unlock(&data->lock);
 630 }
 631
 632 notrace void
 633 tracing_sched_switch_trace(struct trace_array *tr,
 634                            struct trace_array_cpu *data,
 635                            struct task_struct *prev, struct task_struct *next,
 636                            unsigned long flags)
 637 {
 638         struct trace_entry *entry;
 639
 640         spin_lock(&data->lock);
 641         entry                   = tracing_get_trace_entry(tr, data);
 642         tracing_generic_entry_update(entry, flags);
 643         entry->type             = TRACE_CTX;
 644         entry->ctx.prev_pid     = prev->pid;
 645         entry->ctx.prev_prio    = prev->prio;
 646         entry->ctx.prev_state   = prev->state;
 647         entry->ctx.next_pid     = next->pid;
 648         entry->ctx.next_prio    = next->prio;
 649         spin_unlock(&data->lock);
 650 }
 651
 652 enum trace_file_type {
 653         TRACE_FILE_LAT_FMT      = 1,
 654 };
 655
 656 static struct trace_entry *
 657 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 658                 struct trace_iterator *iter, int cpu)
 659 {
 660         struct page *page;
 661         struct trace_entry *array;
 662
 663         if (iter->next_idx[cpu] >= tr->entries ||
 664             iter->next_idx[cpu] >= data->trace_idx ||
 665             (data->trace_head == data->trace_tail &&
 666              data->trace_head_idx == data->trace_tail_idx))
 667                 return NULL;
 668
 669         if (!iter->next_page[cpu]) {
 670                 /* Initialize the iterator for this cpu trace buffer */
 671                 WARN_ON(!data->trace_tail);
 672                 page = virt_to_page(data->trace_tail);
 673                 iter->next_page[cpu] = &page->lru;
 674                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 675         }
 676
 677         page = list_entry(iter->next_page[cpu], struct page, lru);
 678         BUG_ON(&data->trace_pages == &page->lru);
 679
 680         array = page_address(page);
 681
 682         /* Still possible to catch up to the tail */
 683         if (iter->next_idx[cpu] && array == data->trace_tail &&
 684             iter->next_page_idx[cpu] == data->trace_tail_idx)
 685                 return NULL;
 686
 687         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 688         return &array[iter->next_page_idx[cpu]];
 689 }
 690
 691 static struct notrace trace_entry *
 692 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 693 {
 694         struct trace_array *tr = iter->tr;
 695         struct trace_entry *ent, *next = NULL;
 696         int next_cpu = -1;
 697         int cpu;
 698
 699         for_each_possible_cpu(cpu) {
 700                 if (!head_page(tr->data[cpu]))
 701                         continue;
 702                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 703                 if (ent &&
 704                     (!next || (long)(next->idx - ent->idx) > 0)) {
 705                         next = ent;
 706                         next_cpu = cpu;
 707                 }
 708         }
 709
 710         if (ent_cpu)
 711                 *ent_cpu = next_cpu;
 712
 713         return next;
 714 }
 715
 716 static notrace void
 717 trace_iterator_increment(struct trace_iterator *iter)
 718 {
 719         iter->idx++;
 720         iter->next_idx[iter->cpu]++;
 721         iter->next_page_idx[iter->cpu]++;
 722         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 723                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 724
 725                 iter->next_page_idx[iter->cpu] = 0;
 726                 iter->next_page[iter->cpu] =
 727                         trace_next_list(data, iter->next_page[iter->cpu]);
 728         }
 729 }
 730
 731 static notrace void
 732 trace_consume(struct trace_iterator *iter)
 733 {
 734         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 735
 736         data->trace_tail_idx++;
 737         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 738                 data->trace_tail = trace_next_page(data, data->trace_tail);
 739                 data->trace_tail_idx = 0;
 740         }
 741
 742         /* Check if we empty it, then reset the index */
 743         if (data->trace_head == data->trace_tail &&
 744             data->trace_head_idx == data->trace_tail_idx)
 745                 data->trace_idx = 0;
 746
 747         trace_iterator_increment(iter);
 748 }
 749
 750 static notrace void *
 751 find_next_entry_inc(struct trace_iterator *iter)
 752 {
 753         struct trace_entry *next;
 754         int next_cpu = -1;
 755
 756         next = find_next_entry(iter, &next_cpu);
 757
 758         iter->prev_ent = iter->ent;
 759         iter->prev_cpu = iter->cpu;
 760
 761         iter->ent = next;
 762         iter->cpu = next_cpu;
 763
 764         if (next)
 765                 trace_iterator_increment(iter);
 766
 767         return next ? iter : NULL;
 768 }
 769
 770 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 771 {
 772         struct trace_iterator *iter = m->private;
 773         void *last_ent = iter->ent;
 774         int i = (int)*pos;
 775         void *ent;
 776
 777         (*pos)++;
 778
 779         /* can't go backwards */
 780         if (iter->idx > i)
 781                 return NULL;
 782
 783         if (iter->idx < 0)
 784                 ent = find_next_entry_inc(iter);
 785         else
 786                 ent = iter;
 787
 788         while (ent && iter->idx < i)
 789                 ent = find_next_entry_inc(iter);
 790
 791         iter->pos = *pos;
 792
 793         if (last_ent && !ent)
 794                 seq_puts(m, "\n\nvim:ft=help\n");
 795
 796         return ent;
 797 }
 798
 799 static void *s_start(struct seq_file *m, loff_t *pos)
 800 {
 801         struct trace_iterator *iter = m->private;
 802         void *p = NULL;
 803         loff_t l = 0;
 804         int i;
 805
 806         mutex_lock(&trace_types_lock);
 807
 808         if (!current_trace || current_trace != iter->trace)
 809                 return NULL;
 810
 811         atomic_inc(&trace_record_cmdline_disabled);
 812
 813         /* let the tracer grab locks here if needed */
 814         if (current_trace->start)
 815                 current_trace->start(iter);
 816
 817         if (*pos != iter->pos) {
 818                 iter->ent = NULL;
 819                 iter->cpu = 0;
 820                 iter->idx = -1;
 821                 iter->prev_ent = NULL;
 822                 iter->prev_cpu = -1;
 823
 824                 for_each_possible_cpu(i) {
 825                         iter->next_idx[i] = 0;
 826                         iter->next_page[i] = NULL;
 827                 }
 828
 829                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 830                         ;
 831
 832         } else {
 833                 l = *pos - 1;
 834                 p = s_next(m, p, &l);
 835         }
 836
 837         return p;
 838 }
 839
 840 static void s_stop(struct seq_file *m, void *p)
 841 {
 842         struct trace_iterator *iter = m->private;
 843
 844         atomic_dec(&trace_record_cmdline_disabled);
 845
 846         /* let the tracer release locks here if needed */
 847         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 848                 iter->trace->stop(iter);
 849
 850         mutex_unlock(&trace_types_lock);
 851 }
 852
 853 static int
 854 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 855 {
 856 #ifdef CONFIG_KALLSYMS
 857         char str[KSYM_SYMBOL_LEN];
 858
 859         kallsyms_lookup(address, NULL, NULL, NULL, str);
 860
 861         return trace_seq_printf(s, fmt, str);
 862 #endif
 863         return 1;
 864 }
 865
 866 static int
 867 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 868                      unsigned long address)
 869 {
 870 #ifdef CONFIG_KALLSYMS
 871         char str[KSYM_SYMBOL_LEN];
 872
 873         sprint_symbol(str, address);
 874         return trace_seq_printf(s, fmt, str);
 875 #endif
 876         return 1;
 877 }
 878
 879 #ifndef CONFIG_64BIT
 880 # define IP_FMT "%08lx"
 881 #else
 882 # define IP_FMT "%016lx"
 883 #endif
 884
 885 static notrace int
 886 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 887 {
 888         int ret;
 889
 890         if (!ip)
 891                 return trace_seq_printf(s, "0");
 892
 893         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 894                 ret = seq_print_sym_offset(s, "%s", ip);
 895         else
 896                 ret = seq_print_sym_short(s, "%s", ip);
 897
 898         if (!ret)
 899                 return 0;
 900
 901         if (sym_flags & TRACE_ITER_SYM_ADDR)
 902                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 903         return ret;
 904 }
 905
 906 static notrace void print_lat_help_header(struct seq_file *m)
 907 {
 908         seq_puts(m, "#                _------=> CPU#            \n");
 909         seq_puts(m, "#               / _-----=> irqs-off        \n");
 910         seq_puts(m, "#              | / _----=> need-resched    \n");
 911         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 912         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 913         seq_puts(m, "#              |||| /                      \n");
 914         seq_puts(m, "#              |||||     delay             \n");
 915         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 916         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 917 }
 918
 919 static notrace void print_func_help_header(struct seq_file *m)
 920 {
 921         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 922         seq_puts(m, "#              | |      |          |         |\n");
 923 }
 924
 925
 926 static notrace void
 927 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 928 {
 929         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 930         struct trace_array *tr = iter->tr;
 931         struct trace_array_cpu *data = tr->data[tr->cpu];
 932         struct tracer *type = current_trace;
 933         unsigned long total   = 0;
 934         unsigned long entries = 0;
 935         int cpu;
 936         const char *name = "preemption";
 937
 938         if (type)
 939                 name = type->name;
 940
 941         for_each_possible_cpu(cpu) {
 942                 if (head_page(tr->data[cpu])) {
 943                         total += tr->data[cpu]->trace_idx;
 944                         if (tr->data[cpu]->trace_idx > tr->entries)
 945                                 entries += tr->entries;
 946                         else
 947                                 entries += tr->data[cpu]->trace_idx;
 948                 }
 949         }
 950
 951         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
 952                    name, UTS_RELEASE);
 953         seq_puts(m, "-----------------------------------"
 954                  "---------------------------------\n");
 955         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
 956                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
 957                    nsecs_to_usecs(data->saved_latency),
 958                    entries,
 959                    total,
 960                    tr->cpu,
 961 #if defined(CONFIG_PREEMPT_NONE)
 962                    "server",
 963 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
 964                    "desktop",
 965 #elif defined(CONFIG_PREEMPT_DESKTOP)
 966                    "preempt",
 967 #else
 968                    "unknown",
 969 #endif
 970                    /* These are reserved for later use */
 971                    0, 0, 0, 0);
 972 #ifdef CONFIG_SMP
 973         seq_printf(m, " #P:%d)\n", num_online_cpus());
 974 #else
 975         seq_puts(m, ")\n");
 976 #endif
 977         seq_puts(m, "    -----------------\n");
 978         seq_printf(m, "    | task: %.16s-%d "
 979                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
 980                    data->comm, data->pid, data->uid, data->nice,
 981                    data->policy, data->rt_priority);
 982         seq_puts(m, "    -----------------\n");
 983
 984         if (data->critical_start) {
 985                 seq_puts(m, " => started at: ");
 986                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
 987                 trace_print_seq(m, &iter->seq);
 988                 seq_puts(m, "\n => ended at:   ");
 989                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
 990                 trace_print_seq(m, &iter->seq);
 991                 seq_puts(m, "\n");
 992         }
 993
 994         seq_puts(m, "\n");
 995 }
 996
 997 static notrace void
 998 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
 999 {
1000         int hardirq, softirq;
1001         char *comm;
1002
1003         comm = trace_find_cmdline(entry->pid);
1004
1005         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1006         trace_seq_printf(s, "%d", cpu);
1007         trace_seq_printf(s, "%c%c",
1008                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1009                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1010
1011         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1012         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1013         if (hardirq && softirq)
1014                 trace_seq_putc(s, 'H');
1015         else {
1016                 if (hardirq)
1017                         trace_seq_putc(s, 'h');
1018                 else {
1019                         if (softirq)
1020                                 trace_seq_putc(s, 's');
1021                         else
1022                                 trace_seq_putc(s, '.');
1023                 }
1024         }
1025
1026         if (entry->preempt_count)
1027                 trace_seq_printf(s, "%x", entry->preempt_count);
1028         else
1029                 trace_seq_puts(s, ".");
1030 }
1031
1032 unsigned long preempt_mark_thresh = 100;
1033
1034 static notrace void
1035 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1036                     unsigned long rel_usecs)
1037 {
1038         trace_seq_printf(s, " %4lldus", abs_usecs);
1039         if (rel_usecs > preempt_mark_thresh)
1040                 trace_seq_puts(s, "!: ");
1041         else if (rel_usecs > 1)
1042                 trace_seq_puts(s, "+: ");
1043         else
1044                 trace_seq_puts(s, " : ");
1045 }
1046
1047 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1048
1049 static notrace void
1050 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1051 {
1052         struct trace_seq *s = &iter->seq;
1053         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1054         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1055         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1056         struct trace_entry *entry = iter->ent;
1057         unsigned long abs_usecs;
1058         unsigned long rel_usecs;
1059         char *comm;
1060         int S;
1061
1062         if (!next_entry)
1063                 next_entry = entry;
1064         rel_usecs = ns2usecs(next_entry->t - entry->t);
1065         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1066
1067         if (verbose) {
1068                 comm = trace_find_cmdline(entry->pid);
1069                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1070                                  " %ld.%03ldms (+%ld.%03ldms): ",
1071                                  comm,
1072                                  entry->pid, cpu, entry->flags,
1073                                  entry->preempt_count, trace_idx,
1074                                  ns2usecs(entry->t),
1075                                  abs_usecs/1000,
1076                                  abs_usecs % 1000, rel_usecs/1000,
1077                                  rel_usecs % 1000);
1078         } else {
1079                 lat_print_generic(s, entry, cpu);
1080                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1081         }
1082         switch (entry->type) {
1083         case TRACE_FN:
1084                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1085                 trace_seq_puts(s, " (");
1086                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1087                 trace_seq_puts(s, ")\n");
1088                 break;
1089         case TRACE_CTX:
1090                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1091                         state_to_char[entry->ctx.prev_state] : 'X';
1092                 comm = trace_find_cmdline(entry->ctx.next_pid);
1093                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1094                                  entry->ctx.prev_pid,
1095                                  entry->ctx.prev_prio,
1096                                  S,
1097                                  entry->ctx.next_pid,
1098                                  entry->ctx.next_prio,
1099                                  comm);
1100                 break;
1101         default:
1102                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1103         }
1104 }
1105
1106 static notrace void sync_time_offset(struct trace_iterator *iter)
1107 {
1108         struct trace_array_cpu *prev_array, *array;
1109         struct trace_entry *prev_entry, *entry;
1110         cycle_t prev_t, t;
1111
1112         entry = iter->ent;
1113         prev_entry = iter->prev_ent;
1114         if (!prev_entry)
1115                 return;
1116
1117         prev_array = iter->tr->data[iter->prev_cpu];
1118         array = iter->tr->data[iter->cpu];
1119
1120         prev_t = prev_entry->t + prev_array->time_offset;
1121         t = entry->t + array->time_offset;
1122
1123         /*
1124          * If time goes backwards we increase the offset of
1125          * the current array, to not have observable time warps.
1126          * This will quickly synchronize the time offsets of
1127          * multiple CPUs:
1128          */
1129         if (t < prev_t)
1130                 array->time_offset += prev_t - t;
1131 }
1132
1133 static notrace int
1134 print_trace_fmt(struct trace_iterator *iter)
1135 {
1136         struct trace_seq *s = &iter->seq;
1137         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1138         struct trace_entry *entry;
1139         unsigned long usec_rem;
1140         unsigned long long t;
1141         unsigned long secs;
1142         char *comm;
1143         int S;
1144         int ret;
1145
1146         sync_time_offset(iter);
1147         entry = iter->ent;
1148
1149         comm = trace_find_cmdline(iter->ent->pid);
1150
1151         t = ns2usecs(entry->t + iter->tr->data[iter->cpu]->time_offset);
1152         usec_rem = do_div(t, 1000000ULL);
1153         secs = (unsigned long)t;
1154
1155         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1156         if (!ret)
1157                 return 0;
1158         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1159         if (!ret)
1160                 return 0;
1161         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1162         if (!ret)
1163                 return 0;
1164
1165         switch (entry->type) {
1166         case TRACE_FN:
1167                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1168                 if (!ret)
1169                         return 0;
1170                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1171                                                 entry->fn.parent_ip) {
1172                         ret = trace_seq_printf(s, " <-");
1173                         if (!ret)
1174                                 return 0;
1175                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1176                                                sym_flags);
1177                         if (!ret)
1178                                 return 0;
1179                 }
1180                 ret = trace_seq_printf(s, "\n");
1181                 if (!ret)
1182                         return 0;
1183                 break;
1184         case TRACE_CTX:
1185                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1186                         state_to_char[entry->ctx.prev_state] : 'X';
1187                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1188                                        entry->ctx.prev_pid,
1189                                        entry->ctx.prev_prio,
1190                                        S,
1191                                        entry->ctx.next_pid,
1192                                        entry->ctx.next_prio);
1193                 if (!ret)
1194                         return 0;
1195                 break;
1196         }
1197         return 1;
1198 }
1199
1200 static int trace_empty(struct trace_iterator *iter)
1201 {
1202         struct trace_array_cpu *data;
1203         int cpu;
1204
1205         for_each_possible_cpu(cpu) {
1206                 data = iter->tr->data[cpu];
1207
1208                 if (head_page(data) && data->trace_idx &&
1209                     (data->trace_tail != data->trace_head ||
1210                      data->trace_tail_idx != data->trace_head_idx))
1211                         return 0;
1212         }
1213         return 1;
1214 }
1215
1216 static int s_show(struct seq_file *m, void *v)
1217 {
1218         struct trace_iterator *iter = v;
1219
1220         if (iter->ent == NULL) {
1221                 if (iter->tr) {
1222                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1223                         seq_puts(m, "#\n");
1224                 }
1225                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1226                         /* print nothing if the buffers are empty */
1227                         if (trace_empty(iter))
1228                                 return 0;
1229                         print_trace_header(m, iter);
1230                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1231                                 print_lat_help_header(m);
1232                 } else {
1233                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1234                                 print_func_help_header(m);
1235                 }
1236         } else {
1237                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1238                         print_lat_fmt(iter, iter->idx, iter->cpu);
1239                 else
1240                         print_trace_fmt(iter);
1241                 trace_print_seq(m, &iter->seq);
1242         }
1243
1244         return 0;
1245 }
1246
1247 static struct seq_operations tracer_seq_ops = {
1248         .start = s_start,
1249         .next = s_next,
1250         .stop = s_stop,
1251         .show = s_show,
1252 };
1253
1254 static struct trace_iterator notrace *
1255 __tracing_open(struct inode *inode, struct file *file, int *ret)
1256 {
1257         struct trace_iterator *iter;
1258
1259         if (tracing_disabled) {
1260                 *ret = -ENODEV;
1261                 return NULL;
1262         }
1263
1264         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1265         if (!iter) {
1266                 *ret = -ENOMEM;
1267                 goto out;
1268         }
1269
1270         mutex_lock(&trace_types_lock);
1271         if (current_trace && current_trace->print_max)
1272                 iter->tr = &max_tr;
1273         else
1274                 iter->tr = inode->i_private;
1275         iter->trace = current_trace;
1276         iter->pos = -1;
1277
1278         /* TODO stop tracer */
1279         *ret = seq_open(file, &tracer_seq_ops);
1280         if (!*ret) {
1281                 struct seq_file *m = file->private_data;
1282                 m->private = iter;
1283
1284                 /* stop the trace while dumping */
1285                 if (iter->tr->ctrl)
1286                         tracer_enabled = 0;
1287
1288                 if (iter->trace && iter->trace->open)
1289                         iter->trace->open(iter);
1290         } else {
1291                 kfree(iter);
1292                 iter = NULL;
1293         }
1294         mutex_unlock(&trace_types_lock);
1295
1296  out:
1297         return iter;
1298 }
1299
1300 int tracing_open_generic(struct inode *inode, struct file *filp)
1301 {
1302         if (tracing_disabled)
1303                 return -ENODEV;
1304
1305         filp->private_data = inode->i_private;
1306         return 0;
1307 }
1308
1309 int tracing_release(struct inode *inode, struct file *file)
1310 {
1311         struct seq_file *m = (struct seq_file *)file->private_data;
1312         struct trace_iterator *iter = m->private;
1313
1314         mutex_lock(&trace_types_lock);
1315         if (iter->trace && iter->trace->close)
1316                 iter->trace->close(iter);
1317
1318         /* reenable tracing if it was previously enabled */
1319         if (iter->tr->ctrl)
1320                 tracer_enabled = 1;
1321         mutex_unlock(&trace_types_lock);
1322
1323         seq_release(inode, file);
1324         kfree(iter);
1325         return 0;
1326 }
1327
1328 static int tracing_open(struct inode *inode, struct file *file)
1329 {
1330         int ret;
1331
1332         __tracing_open(inode, file, &ret);
1333
1334         return ret;
1335 }
1336
1337 static int tracing_lt_open(struct inode *inode, struct file *file)
1338 {
1339         struct trace_iterator *iter;
1340         int ret;
1341
1342         iter = __tracing_open(inode, file, &ret);
1343
1344         if (!ret)
1345                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1346
1347         return ret;
1348 }
1349
1350
1351 static notrace void *
1352 t_next(struct seq_file *m, void *v, loff_t *pos)
1353 {
1354         struct tracer *t = m->private;
1355
1356         (*pos)++;
1357
1358         if (t)
1359                 t = t->next;
1360
1361         m->private = t;
1362
1363         return t;
1364 }
1365
1366 static void *t_start(struct seq_file *m, loff_t *pos)
1367 {
1368         struct tracer *t = m->private;
1369         loff_t l = 0;
1370
1371         mutex_lock(&trace_types_lock);
1372         for (; t && l < *pos; t = t_next(m, t, &l))
1373                 ;
1374
1375         return t;
1376 }
1377
1378 static void t_stop(struct seq_file *m, void *p)
1379 {
1380         mutex_unlock(&trace_types_lock);
1381 }
1382
1383 static int t_show(struct seq_file *m, void *v)
1384 {
1385         struct tracer *t = v;
1386
1387         if (!t)
1388                 return 0;
1389
1390         seq_printf(m, "%s", t->name);
1391         if (t->next)
1392                 seq_putc(m, ' ');
1393         else
1394                 seq_putc(m, '\n');
1395
1396         return 0;
1397 }
1398
1399 static struct seq_operations show_traces_seq_ops = {
1400         .start = t_start,
1401         .next = t_next,
1402         .stop = t_stop,
1403         .show = t_show,
1404 };
1405
1406 static int show_traces_open(struct inode *inode, struct file *file)
1407 {
1408         int ret;
1409
1410         if (tracing_disabled)
1411                 return -ENODEV;
1412
1413         ret = seq_open(file, &show_traces_seq_ops);
1414         if (!ret) {
1415                 struct seq_file *m = file->private_data;
1416                 m->private = trace_types;
1417         }
1418
1419         return ret;
1420 }
1421
1422 static struct file_operations tracing_fops = {
1423         .open = tracing_open,
1424         .read = seq_read,
1425         .llseek = seq_lseek,
1426         .release = tracing_release,
1427 };
1428
1429 static struct file_operations tracing_lt_fops = {
1430         .open = tracing_lt_open,
1431         .read = seq_read,
1432         .llseek = seq_lseek,
1433         .release = tracing_release,
1434 };
1435
1436 static struct file_operations show_traces_fops = {
1437         .open = show_traces_open,
1438         .read = seq_read,
1439         .release = seq_release,
1440 };
1441
1442 static ssize_t
1443 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1444                        size_t cnt, loff_t *ppos)
1445 {
1446         char *buf;
1447         int r = 0;
1448         int len = 0;
1449         int i;
1450
1451         /* calulate max size */
1452         for (i = 0; trace_options[i]; i++) {
1453                 len += strlen(trace_options[i]);
1454                 len += 3; /* "no" and space */
1455         }
1456
1457         /* +2 for \n and \0 */
1458         buf = kmalloc(len + 2, GFP_KERNEL);
1459         if (!buf)
1460                 return -ENOMEM;
1461
1462         for (i = 0; trace_options[i]; i++) {
1463                 if (trace_flags & (1 << i))
1464                         r += sprintf(buf + r, "%s ", trace_options[i]);
1465                 else
1466                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1467         }
1468
1469         r += sprintf(buf + r, "\n");
1470         WARN_ON(r >= len + 2);
1471
1472         r = simple_read_from_buffer(ubuf, cnt, ppos,
1473                                     buf, r);
1474
1475         kfree(buf);
1476
1477         return r;
1478 }
1479
1480 static ssize_t
1481 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1482                         size_t cnt, loff_t *ppos)
1483 {
1484         char buf[64];
1485         char *cmp = buf;
1486         int neg = 0;
1487         int i;
1488
1489         if (cnt > 63)
1490                 cnt = 63;
1491
1492         if (copy_from_user(&buf, ubuf, cnt))
1493                 return -EFAULT;
1494
1495         buf[cnt] = 0;
1496
1497         if (strncmp(buf, "no", 2) == 0) {
1498                 neg = 1;
1499                 cmp += 2;
1500         }
1501
1502         for (i = 0; trace_options[i]; i++) {
1503                 int len = strlen(trace_options[i]);
1504
1505                 if (strncmp(cmp, trace_options[i], len) == 0) {
1506                         if (neg)
1507                                 trace_flags &= ~(1 << i);
1508                         else
1509                                 trace_flags |= (1 << i);
1510                         break;
1511                 }
1512         }
1513
1514         filp->f_pos += cnt;
1515
1516         return cnt;
1517 }
1518
1519 static struct file_operations tracing_iter_fops = {
1520         .open = tracing_open_generic,
1521         .read = tracing_iter_ctrl_read,
1522         .write = tracing_iter_ctrl_write,
1523 };
1524
1525 static const char readme_msg[] =
1526         "tracing mini-HOWTO:\n\n"
1527         "# mkdir /debug\n"
1528         "# mount -t debugfs nodev /debug\n\n"
1529         "# cat /debug/tracing/available_tracers\n"
1530         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1531         "# cat /debug/tracing/current_tracer\n"
1532         "none\n"
1533         "# echo sched_switch > /debug/tracing/current_tracer\n"
1534         "# cat /debug/tracing/current_tracer\n"
1535         "sched_switch\n"
1536         "# cat /debug/tracing/iter_ctrl\n"
1537         "noprint-parent nosym-offset nosym-addr noverbose\n"
1538         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1539         "# echo 1 > /debug/tracing/tracing_enabled\n"
1540         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1541         "echo 0 > /debug/tracing/tracing_enabled\n"
1542 ;
1543
1544 static ssize_t
1545 tracing_readme_read(struct file *filp, char __user *ubuf,
1546                        size_t cnt, loff_t *ppos)
1547 {
1548         return simple_read_from_buffer(ubuf, cnt, ppos,
1549                                         readme_msg, strlen(readme_msg));
1550 }
1551
1552 static struct file_operations tracing_readme_fops = {
1553         .open = tracing_open_generic,
1554         .read = tracing_readme_read,
1555 };
1556
1557
1558 static ssize_t
1559 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1560                   size_t cnt, loff_t *ppos)
1561 {
1562         struct trace_array *tr = filp->private_data;
1563         char buf[64];
1564         int r;
1565
1566         r = sprintf(buf, "%ld\n", tr->ctrl);
1567         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1568 }
1569
1570 static ssize_t
1571 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1572                    size_t cnt, loff_t *ppos)
1573 {
1574         struct trace_array *tr = filp->private_data;
1575         long val;
1576         char buf[64];
1577
1578         if (cnt > 63)
1579                 cnt = 63;
1580
1581         if (copy_from_user(&buf, ubuf, cnt))
1582                 return -EFAULT;
1583
1584         buf[cnt] = 0;
1585
1586         val = simple_strtoul(buf, NULL, 10);
1587
1588         val = !!val;
1589
1590         mutex_lock(&trace_types_lock);
1591         if (tr->ctrl ^ val) {
1592                 if (val)
1593                         tracer_enabled = 1;
1594                 else
1595                         tracer_enabled = 0;
1596
1597                 tr->ctrl = val;
1598
1599                 if (current_trace && current_trace->ctrl_update)
1600                         current_trace->ctrl_update(tr);
1601         }
1602         mutex_unlock(&trace_types_lock);
1603
1604         filp->f_pos += cnt;
1605
1606         return cnt;
1607 }
1608
1609 static ssize_t
1610 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1611                        size_t cnt, loff_t *ppos)
1612 {
1613         char buf[max_tracer_type_len+2];
1614         int r;
1615
1616         mutex_lock(&trace_types_lock);
1617         if (current_trace)
1618                 r = sprintf(buf, "%s\n", current_trace->name);
1619         else
1620                 r = sprintf(buf, "\n");
1621         mutex_unlock(&trace_types_lock);
1622
1623         return simple_read_from_buffer(ubuf, cnt, ppos,
1624                                        buf, r);
1625 }
1626
1627 static ssize_t
1628 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1629                         size_t cnt, loff_t *ppos)
1630 {
1631         struct trace_array *tr = &global_trace;
1632         struct tracer *t;
1633         char buf[max_tracer_type_len+1];
1634         int i;
1635
1636         if (cnt > max_tracer_type_len)
1637                 cnt = max_tracer_type_len;
1638
1639         if (copy_from_user(&buf, ubuf, cnt))
1640                 return -EFAULT;
1641
1642         buf[cnt] = 0;
1643
1644         /* strip ending whitespace. */
1645         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1646                 buf[i] = 0;
1647
1648         mutex_lock(&trace_types_lock);
1649         for (t = trace_types; t; t = t->next) {
1650                 if (strcmp(t->name, buf) == 0)
1651                         break;
1652         }
1653         if (!t || t == current_trace)
1654                 goto out;
1655
1656         if (current_trace && current_trace->reset)
1657                 current_trace->reset(tr);
1658
1659         current_trace = t;
1660         if (t->init)
1661                 t->init(tr);
1662
1663  out:
1664         mutex_unlock(&trace_types_lock);
1665
1666         filp->f_pos += cnt;
1667
1668         return cnt;
1669 }
1670
1671 static ssize_t
1672 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1673                      size_t cnt, loff_t *ppos)
1674 {
1675         unsigned long *ptr = filp->private_data;
1676         char buf[64];
1677         int r;
1678
1679         r = snprintf(buf, 64, "%ld\n",
1680                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1681         if (r > 64)
1682                 r = 64;
1683         return simple_read_from_buffer(ubuf, cnt, ppos,
1684                                        buf, r);
1685 }
1686
1687 static ssize_t
1688 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1689                       size_t cnt, loff_t *ppos)
1690 {
1691         long *ptr = filp->private_data;
1692         long val;
1693         char buf[64];
1694
1695         if (cnt > 63)
1696                 cnt = 63;
1697
1698         if (copy_from_user(&buf, ubuf, cnt))
1699                 return -EFAULT;
1700
1701         buf[cnt] = 0;
1702
1703         val = simple_strtoul(buf, NULL, 10);
1704
1705         *ptr = val * 1000;
1706
1707         return cnt;
1708 }
1709
1710 static atomic_t tracing_reader;
1711
1712 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1713 {
1714         struct trace_iterator *iter;
1715
1716         if (tracing_disabled)
1717                 return -ENODEV;
1718
1719         /* We only allow for reader of the pipe */
1720         if (atomic_inc_return(&tracing_reader) != 1) {
1721                 atomic_dec(&tracing_reader);
1722                 return -EBUSY;
1723         }
1724
1725         /* create a buffer to store the information to pass to userspace */
1726         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1727         if (!iter)
1728                 return -ENOMEM;
1729
1730         iter->tr = &global_trace;
1731
1732         filp->private_data = iter;
1733
1734         return 0;
1735 }
1736
1737 static int tracing_release_pipe(struct inode *inode, struct file *file)
1738 {
1739         struct trace_iterator *iter = file->private_data;
1740
1741         kfree(iter);
1742         atomic_dec(&tracing_reader);
1743
1744         return 0;
1745 }
1746
1747 /*
1748  * Consumer reader.
1749  */
1750 static ssize_t
1751 tracing_read_pipe(struct file *filp, char __user *ubuf,
1752                   size_t cnt, loff_t *ppos)
1753 {
1754         struct trace_iterator *iter = filp->private_data;
1755         struct trace_array_cpu *data;
1756         static cpumask_t mask;
1757         struct trace_entry *entry;
1758         static int start;
1759         unsigned long flags;
1760         int read = 0;
1761         int cpu;
1762         int len;
1763         int ret;
1764
1765         /* return any leftover data */
1766         if (iter->seq.len > start) {
1767                 len = iter->seq.len - start;
1768                 if (cnt > len)
1769                         cnt = len;
1770                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1771                 if (ret)
1772                         cnt = -EFAULT;
1773
1774                 start += len;
1775
1776                 return cnt;
1777         }
1778
1779         trace_seq_reset(&iter->seq);
1780         start = 0;
1781
1782         while (trace_empty(iter)) {
1783                 /*
1784                  * This is a make-shift waitqueue. The reason we don't use
1785                  * an actual wait queue is because:
1786                  *  1) we only ever have one waiter
1787                  *  2) the tracing, traces all functions, we don't want
1788                  *     the overhead of calling wake_up and friends
1789                  *     (and tracing them too)
1790                  *     Anyway, this is really very primitive wakeup.
1791                  */
1792                 set_current_state(TASK_INTERRUPTIBLE);
1793                 iter->tr->waiter = current;
1794
1795                 /* sleep for one second, and try again. */
1796                 schedule_timeout(HZ);
1797
1798                 iter->tr->waiter = NULL;
1799
1800                 if (signal_pending(current))
1801                         return -EINTR;
1802
1803                 /*
1804                  * We block until we read something and tracing is disabled.
1805                  * We still block if tracing is disabled, but we have never
1806                  * read anything. This allows a user to cat this file, and
1807                  * then enable tracing. But after we have read something,
1808                  * we give an EOF when tracing is again disabled.
1809                  *
1810                  * iter->pos will be 0 if we haven't read anything.
1811                  */
1812                 if (!tracer_enabled && iter->pos)
1813                         break;
1814
1815                 continue;
1816         }
1817
1818         /* stop when tracing is finished */
1819         if (trace_empty(iter))
1820                 return 0;
1821
1822         if (cnt >= PAGE_SIZE)
1823                 cnt = PAGE_SIZE - 1;
1824
1825         memset(iter, 0, sizeof(*iter));
1826         iter->tr = &global_trace;
1827         iter->pos = -1;
1828
1829         /*
1830          * We need to stop all tracing on all CPUS to read the
1831          * the next buffer. This is a bit expensive, but is
1832          * not done often. We fill all what we can read,
1833          * and then release the locks again.
1834          */
1835
1836         cpus_clear(mask);
1837         local_irq_save(flags);
1838         for_each_possible_cpu(cpu) {
1839                 data = iter->tr->data[cpu];
1840
1841                 if (!head_page(data) || !data->trace_idx)
1842                         continue;
1843
1844                 atomic_inc(&data->disabled);
1845                 spin_lock(&data->lock);
1846                 cpu_set(cpu, mask);
1847         }
1848
1849         while ((entry = find_next_entry(iter, &cpu))) {
1850
1851                 if (!entry)
1852                         break;
1853
1854                 iter->ent = entry;
1855                 iter->cpu = cpu;
1856
1857                 ret = print_trace_fmt(iter);
1858                 if (!ret)
1859                         break;
1860
1861                 trace_consume(iter);
1862
1863                 if (iter->seq.len >= cnt)
1864                         break;
1865
1866         }
1867
1868         for_each_possible_cpu(cpu) {
1869                 data = iter->tr->data[cpu];
1870
1871                 if (!cpu_isset(cpu, mask))
1872                         continue;
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         return simple_read_from_buffer(ubuf, cnt, ppos,
1932                                        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                 page = list_entry(pages.next, struct page, lru);
2080                 list_del_init(&page->lru);
2081                 list_add_tail(&page->lru, &data->trace_pages);
2082                 ClearPageLRU(page);
2083
2084 #ifdef CONFIG_TRACER_MAX_TRACE
2085                 data = max_tr.data[i];
2086                 spin_lock_init(&data->lock);
2087                 page = list_entry(pages.next, struct page, lru);
2088                 list_del_init(&page->lru);
2089                 list_add_tail(&page->lru, &data->trace_pages);
2090                 SetPageLRU(page);
2091 #endif
2092         }
2093         global_trace.entries += ENTRIES_PER_PAGE;
2094
2095         return 0;
2096
2097  free_pages:
2098         list_for_each_entry_safe(page, tmp, &pages, lru) {
2099                 list_del_init(&page->lru);
2100                 __free_page(page);
2101         }
2102         return -ENOMEM;
2103 }
2104
2105 __init static int tracer_alloc_buffers(void)
2106 {
2107         struct trace_array_cpu *data;
2108         void *array;
2109         struct page *page;
2110         int pages = 0;
2111         int ret = -ENOMEM;
2112         int i;
2113
2114         /* Allocate the first page for all buffers */
2115         for_each_possible_cpu(i) {
2116                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2117                 max_tr.data[i] = &per_cpu(max_data, i);
2118
2119                 array = (void *)__get_free_page(GFP_KERNEL);
2120                 if (array == NULL) {
2121                         printk(KERN_ERR "tracer: failed to allocate page"
2122                                "for trace buffer!\n");
2123                         goto free_buffers;
2124                 }
2125
2126                 /* set the array to the list */
2127                 INIT_LIST_HEAD(&data->trace_pages);
2128                 page = virt_to_page(array);
2129                 list_add(&page->lru, &data->trace_pages);
2130                 /* use the LRU flag to differentiate the two buffers */
2131                 ClearPageLRU(page);
2132
2133 /* Only allocate if we are actually using the max trace */
2134 #ifdef CONFIG_TRACER_MAX_TRACE
2135                 array = (void *)__get_free_page(GFP_KERNEL);
2136                 if (array == NULL) {
2137                         printk(KERN_ERR "tracer: failed to allocate page"
2138                                "for trace buffer!\n");
2139                         goto free_buffers;
2140                 }
2141
2142                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2143                 page = virt_to_page(array);
2144                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2145                 SetPageLRU(page);
2146 #endif
2147         }
2148
2149         /*
2150          * Since we allocate by orders of pages, we may be able to
2151          * round up a bit.
2152          */
2153         global_trace.entries = ENTRIES_PER_PAGE;
2154         pages++;
2155
2156         while (global_trace.entries < trace_nr_entries) {
2157                 if (trace_alloc_page())
2158                         break;
2159                 pages++;
2160         }
2161         max_tr.entries = global_trace.entries;
2162
2163         pr_info("tracer: %d pages allocated for %ld",
2164                 pages, trace_nr_entries);
2165         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2166         pr_info("   actual entries %ld\n", global_trace.entries);
2167
2168         tracer_init_debugfs();
2169
2170         trace_init_cmdlines();
2171
2172         register_tracer(&no_tracer);
2173         current_trace = &no_tracer;
2174
2175         /* All seems OK, enable tracing */
2176         tracing_disabled = 0;
2177
2178         return 0;
2179
2180  free_buffers:
2181         for (i-- ; i >= 0; i--) {
2182                 struct page *page, *tmp;
2183                 struct trace_array_cpu *data = global_trace.data[i];
2184
2185                 if (data) {
2186                         list_for_each_entry_safe(page, tmp,
2187                                                  &data->trace_pages, lru) {
2188                                 list_del_init(&page->lru);
2189                                 __free_page(page);
2190                         }
2191                 }
2192
2193 #ifdef CONFIG_TRACER_MAX_TRACE
2194                 data = max_tr.data[i];
2195                 if (data) {
2196                         list_for_each_entry_safe(page, tmp,
2197                                                  &data->trace_pages, lru) {
2198                                 list_del_init(&page->lru);
2199                                 __free_page(page);
2200                         }
2201                 }
2202 #endif
2203         }
2204         return ret;
2205 }
2206
2207 fs_initcall(tracer_alloc_buffers);