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 const int time_sync_freq_max = 128;
  46 static const cycle_t time_sync_thresh = 100000;
  47
  48 static DEFINE_PER_CPU(cycle_t, time_offset);
  49 static DEFINE_PER_CPU(cycle_t, prev_cpu_time);
  50 static DEFINE_PER_CPU(int, time_sync_count);
  51 static DEFINE_PER_CPU(int, time_sync_freq);
  52
  53 /*
  54  * Global lock which we take every now and then to synchronize
  55  * the CPUs time. This method is not warp-safe, but it's good
  56  * enough to synchronize slowly diverging time sources and thus
  57  * it's good enough for tracing:
  58  */
  59 static DEFINE_SPINLOCK(time_sync_lock);
  60 static cycle_t prev_global_time;
  61
  62 static notrace cycle_t __ftrace_now_sync(cycles_t time, int cpu)
  63 {
  64         unsigned long flags;
  65
  66         spin_lock_irqsave(&time_sync_lock, flags);
  67
  68         /*
  69          * Update the synchronization frequency:
  70          */
  71         if (per_cpu(time_sync_freq, cpu) < time_sync_freq_max)
  72                 per_cpu(time_sync_freq, cpu) *= 2;
  73         per_cpu(time_sync_count, cpu) = per_cpu(time_sync_freq, cpu);
  74
  75         if (time < prev_global_time) {
  76                 per_cpu(time_offset, cpu) += prev_global_time - time;
  77                 time = prev_global_time;
  78         } else {
  79                 prev_global_time = time;
  80         }
  81
  82         spin_unlock_irqrestore(&time_sync_lock, flags);
  83
  84         return time;
  85 }
  86
  87 notrace cycle_t ftrace_now(int cpu)
  88 {
  89         cycle_t prev_cpu_time, time, delta_time;
  90
  91         prev_cpu_time = per_cpu(prev_cpu_time, cpu);
  92         time = sched_clock() + per_cpu(time_offset, cpu);
  93         delta_time = time-prev_cpu_time;
  94
  95         if (unlikely(delta_time > time_sync_thresh ||
  96                                 --per_cpu(time_sync_count, cpu) <= 0))
  97                 time = __ftrace_now_sync(time, cpu);
  98
  99         return time;
 100 }
 101
 102 static struct trace_array       global_trace;
 103
 104 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
 105
 106 static struct trace_array       max_tr;
 107
 108 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 109
 110 static int                      tracer_enabled;
 111 static unsigned long            trace_nr_entries = 16384UL;
 112
 113 static struct tracer            *trace_types __read_mostly;
 114 static struct tracer            *current_trace __read_mostly;
 115 static int                      max_tracer_type_len;
 116
 117 static DEFINE_MUTEX(trace_types_lock);
 118
 119 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
 120
 121 static int __init set_nr_entries(char *str)
 122 {
 123         if (!str)
 124                 return 0;
 125         trace_nr_entries = simple_strtoul(str, &str, 0);
 126         return 1;
 127 }
 128 __setup("trace_entries=", set_nr_entries);
 129
 130 unsigned long nsecs_to_usecs(unsigned long nsecs)
 131 {
 132         return nsecs / 1000;
 133 }
 134
 135 enum trace_type {
 136         __TRACE_FIRST_TYPE = 0,
 137
 138         TRACE_FN,
 139         TRACE_CTX,
 140
 141         __TRACE_LAST_TYPE
 142 };
 143
 144 enum trace_flag_type {
 145         TRACE_FLAG_IRQS_OFF             = 0x01,
 146         TRACE_FLAG_NEED_RESCHED         = 0x02,
 147         TRACE_FLAG_HARDIRQ              = 0x04,
 148         TRACE_FLAG_SOFTIRQ              = 0x08,
 149 };
 150
 151 enum trace_iterator_flags {
 152         TRACE_ITER_PRINT_PARENT         = 0x01,
 153         TRACE_ITER_SYM_OFFSET           = 0x02,
 154         TRACE_ITER_SYM_ADDR             = 0x04,
 155         TRACE_ITER_VERBOSE              = 0x08,
 156 };
 157
 158 #define TRACE_ITER_SYM_MASK \
 159         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 160
 161 /* These must match the bit postions above */
 162 static const char *trace_options[] = {
 163         "print-parent",
 164         "sym-offset",
 165         "sym-addr",
 166         "verbose",
 167         NULL
 168 };
 169
 170 static unsigned trace_flags;
 171
 172 static DEFINE_SPINLOCK(ftrace_max_lock);
 173
 174 /*
 175  * Copy the new maximum trace into the separate maximum-trace
 176  * structure. (this way the maximum trace is permanently saved,
 177  * for later retrieval via /debugfs/tracing/latency_trace)
 178  */
 179 static notrace void
 180 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 181 {
 182         struct trace_array_cpu *data = tr->data[cpu];
 183
 184         max_tr.cpu = cpu;
 185         max_tr.time_start = data->preempt_timestamp;
 186
 187         data = max_tr.data[cpu];
 188         data->saved_latency = tracing_max_latency;
 189
 190         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 191         data->pid = tsk->pid;
 192         data->uid = tsk->uid;
 193         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 194         data->policy = tsk->policy;
 195         data->rt_priority = tsk->rt_priority;
 196
 197         /* record this tasks comm */
 198         tracing_record_cmdline(current);
 199 }
 200
 201 void check_pages(struct trace_array_cpu *data)
 202 {
 203         struct page *page, *tmp;
 204
 205         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 206         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 207
 208         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 209                 BUG_ON(page->lru.next->prev != &page->lru);
 210                 BUG_ON(page->lru.prev->next != &page->lru);
 211         }
 212 }
 213
 214 void *head_page(struct trace_array_cpu *data)
 215 {
 216         struct page *page;
 217
 218         check_pages(data);
 219         if (list_empty(&data->trace_pages))
 220                 return NULL;
 221
 222         page = list_entry(data->trace_pages.next, struct page, lru);
 223         BUG_ON(&page->lru == &data->trace_pages);
 224
 225         return page_address(page);
 226 }
 227
 228 static notrace int
 229 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 230 {
 231         int len = (PAGE_SIZE - 1) - s->len;
 232         va_list ap;
 233         int ret;
 234
 235         if (!len)
 236                 return 0;
 237
 238         va_start(ap, fmt);
 239         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 240         va_end(ap);
 241
 242         /* If we can't write it all, don't bother writing anything */
 243         if (ret > len)
 244                 return 0;
 245
 246         s->len += ret;
 247
 248         return len;
 249 }
 250
 251 static notrace int
 252 trace_seq_puts(struct trace_seq *s, const char *str)
 253 {
 254         int len = strlen(str);
 255
 256         if (len > ((PAGE_SIZE - 1) - s->len))
 257                 return 0;
 258
 259         memcpy(s->buffer + s->len, str, len);
 260         s->len += len;
 261
 262         return len;
 263 }
 264
 265 static notrace int
 266 trace_seq_putc(struct trace_seq *s, unsigned char c)
 267 {
 268         if (s->len >= (PAGE_SIZE - 1))
 269                 return 0;
 270
 271         s->buffer[s->len++] = c;
 272
 273         return 1;
 274 }
 275
 276 static notrace void
 277 trace_seq_reset(struct trace_seq *s)
 278 {
 279         s->len = 0;
 280 }
 281
 282 static notrace void
 283 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 284 {
 285         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 286
 287         s->buffer[len] = 0;
 288         seq_puts(m, s->buffer);
 289
 290         trace_seq_reset(s);
 291 }
 292
 293 notrace static void
 294 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 295 {
 296         struct list_head flip_pages;
 297
 298         INIT_LIST_HEAD(&flip_pages);
 299
 300         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 301                 sizeof(struct trace_array_cpu) -
 302                 offsetof(struct trace_array_cpu, trace_head_idx));
 303
 304         check_pages(tr1);
 305         check_pages(tr2);
 306         list_splice_init(&tr1->trace_pages, &flip_pages);
 307         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 308         list_splice_init(&flip_pages, &tr2->trace_pages);
 309         BUG_ON(!list_empty(&flip_pages));
 310         check_pages(tr1);
 311         check_pages(tr2);
 312 }
 313
 314 notrace void
 315 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 316 {
 317         struct trace_array_cpu *data;
 318         int i;
 319
 320         WARN_ON_ONCE(!irqs_disabled());
 321         spin_lock(&ftrace_max_lock);
 322         /* clear out all the previous traces */
 323         for_each_possible_cpu(i) {
 324                 data = tr->data[i];
 325                 flip_trace(max_tr.data[i], data);
 326                 tracing_reset(data);
 327         }
 328
 329         __update_max_tr(tr, tsk, cpu);
 330         spin_unlock(&ftrace_max_lock);
 331 }
 332
 333 /**
 334  * update_max_tr_single - only copy one trace over, and reset the rest
 335  * @tr - tracer
 336  * @tsk - task with the latency
 337  * @cpu - the cpu of the buffer to copy.
 338  */
 339 notrace void
 340 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 341 {
 342         struct trace_array_cpu *data = tr->data[cpu];
 343         int i;
 344
 345         WARN_ON_ONCE(!irqs_disabled());
 346         spin_lock(&ftrace_max_lock);
 347         for_each_possible_cpu(i)
 348                 tracing_reset(max_tr.data[i]);
 349
 350         flip_trace(max_tr.data[cpu], data);
 351         tracing_reset(data);
 352
 353         __update_max_tr(tr, tsk, cpu);
 354         spin_unlock(&ftrace_max_lock);
 355 }
 356
 357 int register_tracer(struct tracer *type)
 358 {
 359         struct tracer *t;
 360         int len;
 361         int ret = 0;
 362
 363         if (!type->name) {
 364                 pr_info("Tracer must have a name\n");
 365                 return -1;
 366         }
 367
 368         mutex_lock(&trace_types_lock);
 369         for (t = trace_types; t; t = t->next) {
 370                 if (strcmp(type->name, t->name) == 0) {
 371                         /* already found */
 372                         pr_info("Trace %s already registered\n",
 373                                 type->name);
 374                         ret = -1;
 375                         goto out;
 376                 }
 377         }
 378
 379 #ifdef CONFIG_FTRACE_STARTUP_TEST
 380         if (type->selftest) {
 381                 struct tracer *saved_tracer = current_trace;
 382                 struct trace_array_cpu *data;
 383                 struct trace_array *tr = &global_trace;
 384                 int saved_ctrl = tr->ctrl;
 385                 int i;
 386                 /*
 387                  * Run a selftest on this tracer.
 388                  * Here we reset the trace buffer, and set the current
 389                  * tracer to be this tracer. The tracer can then run some
 390                  * internal tracing to verify that everything is in order.
 391                  * If we fail, we do not register this tracer.
 392                  */
 393                 for_each_possible_cpu(i) {
 394                         data = tr->data[i];
 395                         if (!head_page(data))
 396                                 continue;
 397                         tracing_reset(data);
 398                 }
 399                 current_trace = type;
 400                 tr->ctrl = 0;
 401                 /* the test is responsible for initializing and enabling */
 402                 pr_info("Testing tracer %s: ", type->name);
 403                 ret = type->selftest(type, tr);
 404                 /* the test is responsible for resetting too */
 405                 current_trace = saved_tracer;
 406                 tr->ctrl = saved_ctrl;
 407                 if (ret) {
 408                         printk(KERN_CONT "FAILED!\n");
 409                         goto out;
 410                 }
 411                 /* Only reset on passing, to avoid touching corrupted buffers */
 412                 for_each_possible_cpu(i) {
 413                         data = tr->data[i];
 414                         if (!head_page(data))
 415                                 continue;
 416                         tracing_reset(data);
 417                 }
 418                 printk(KERN_CONT "PASSED\n");
 419         }
 420 #endif
 421
 422         type->next = trace_types;
 423         trace_types = type;
 424         len = strlen(type->name);
 425         if (len > max_tracer_type_len)
 426                 max_tracer_type_len = len;
 427
 428  out:
 429         mutex_unlock(&trace_types_lock);
 430
 431         return ret;
 432 }
 433
 434 void unregister_tracer(struct tracer *type)
 435 {
 436         struct tracer **t;
 437         int len;
 438
 439         mutex_lock(&trace_types_lock);
 440         for (t = &trace_types; *t; t = &(*t)->next) {
 441                 if (*t == type)
 442                         goto found;
 443         }
 444         pr_info("Trace %s not registered\n", type->name);
 445         goto out;
 446
 447  found:
 448         *t = (*t)->next;
 449         if (strlen(type->name) != max_tracer_type_len)
 450                 goto out;
 451
 452         max_tracer_type_len = 0;
 453         for (t = &trace_types; *t; t = &(*t)->next) {
 454                 len = strlen((*t)->name);
 455                 if (len > max_tracer_type_len)
 456                         max_tracer_type_len = len;
 457         }
 458  out:
 459         mutex_unlock(&trace_types_lock);
 460 }
 461
 462 notrace void tracing_reset(struct trace_array_cpu *data)
 463 {
 464         data->trace_idx = 0;
 465         data->trace_head = data->trace_tail = head_page(data);
 466         data->trace_head_idx = 0;
 467         data->trace_tail_idx = 0;
 468 }
 469
 470 #ifdef CONFIG_FTRACE
 471 static notrace void
 472 function_trace_call(unsigned long ip, unsigned long parent_ip)
 473 {
 474         struct trace_array *tr = &global_trace;
 475         struct trace_array_cpu *data;
 476         unsigned long flags;
 477         long disabled;
 478         int cpu;
 479
 480         if (unlikely(!tracer_enabled))
 481                 return;
 482
 483         local_irq_save(flags);
 484         cpu = raw_smp_processor_id();
 485         data = tr->data[cpu];
 486         disabled = atomic_inc_return(&data->disabled);
 487
 488         if (likely(disabled == 1))
 489                 ftrace(tr, data, ip, parent_ip, flags);
 490
 491         atomic_dec(&data->disabled);
 492         local_irq_restore(flags);
 493 }
 494
 495 static struct ftrace_ops trace_ops __read_mostly =
 496 {
 497         .func = function_trace_call,
 498 };
 499 #endif
 500
 501 notrace void tracing_start_function_trace(void)
 502 {
 503         register_ftrace_function(&trace_ops);
 504 }
 505
 506 notrace void tracing_stop_function_trace(void)
 507 {
 508         unregister_ftrace_function(&trace_ops);
 509 }
 510
 511 #define SAVED_CMDLINES 128
 512 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 513 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 514 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 515 static int cmdline_idx;
 516 static DEFINE_SPINLOCK(trace_cmdline_lock);
 517 atomic_t trace_record_cmdline_disabled;
 518
 519 static void trace_init_cmdlines(void)
 520 {
 521         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 522         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 523         cmdline_idx = 0;
 524 }
 525
 526 notrace void trace_stop_cmdline_recording(void);
 527
 528 static notrace void trace_save_cmdline(struct task_struct *tsk)
 529 {
 530         unsigned map;
 531         unsigned idx;
 532
 533         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 534                 return;
 535
 536         /*
 537          * It's not the end of the world if we don't get
 538          * the lock, but we also don't want to spin
 539          * nor do we want to disable interrupts,
 540          * so if we miss here, then better luck next time.
 541          */
 542         if (!spin_trylock(&trace_cmdline_lock))
 543                 return;
 544
 545         idx = map_pid_to_cmdline[tsk->pid];
 546         if (idx >= SAVED_CMDLINES) {
 547                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 548
 549                 map = map_cmdline_to_pid[idx];
 550                 if (map <= PID_MAX_DEFAULT)
 551                         map_pid_to_cmdline[map] = (unsigned)-1;
 552
 553                 map_pid_to_cmdline[tsk->pid] = idx;
 554
 555                 cmdline_idx = idx;
 556         }
 557
 558         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 559
 560         spin_unlock(&trace_cmdline_lock);
 561 }
 562
 563 static notrace char *trace_find_cmdline(int pid)
 564 {
 565         char *cmdline = "<...>";
 566         unsigned map;
 567
 568         if (!pid)
 569                 return "<idle>";
 570
 571         if (pid > PID_MAX_DEFAULT)
 572                 goto out;
 573
 574         map = map_pid_to_cmdline[pid];
 575         if (map >= SAVED_CMDLINES)
 576                 goto out;
 577
 578         cmdline = saved_cmdlines[map];
 579
 580  out:
 581         return cmdline;
 582 }
 583
 584 notrace void tracing_record_cmdline(struct task_struct *tsk)
 585 {
 586         if (atomic_read(&trace_record_cmdline_disabled))
 587                 return;
 588
 589         trace_save_cmdline(tsk);
 590 }
 591
 592 static inline notrace struct list_head *
 593 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 594 {
 595         /*
 596          * Roundrobin - but skip the head (which is not a real page):
 597          */
 598         next = next->next;
 599         if (unlikely(next == &data->trace_pages))
 600                 next = next->next;
 601         BUG_ON(next == &data->trace_pages);
 602
 603         return next;
 604 }
 605
 606 static inline notrace void *
 607 trace_next_page(struct trace_array_cpu *data, void *addr)
 608 {
 609         struct list_head *next;
 610         struct page *page;
 611
 612         page = virt_to_page(addr);
 613
 614         next = trace_next_list(data, &page->lru);
 615         page = list_entry(next, struct page, lru);
 616
 617         return page_address(page);
 618 }
 619
 620 static inline notrace struct trace_entry *
 621 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 622 {
 623         unsigned long idx, idx_next;
 624         struct trace_entry *entry;
 625
 626         data->trace_idx++;
 627         idx = data->trace_head_idx;
 628         idx_next = idx + 1;
 629
 630         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 631
 632         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 633
 634         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 635                 data->trace_head = trace_next_page(data, data->trace_head);
 636                 idx_next = 0;
 637         }
 638
 639         if (data->trace_head == data->trace_tail &&
 640             idx_next == data->trace_tail_idx) {
 641                 /* overrun */
 642                 data->trace_tail_idx++;
 643                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 644                         data->trace_tail =
 645                                 trace_next_page(data, data->trace_tail);
 646                         data->trace_tail_idx = 0;
 647                 }
 648         }
 649
 650         data->trace_head_idx = idx_next;
 651
 652         return entry;
 653 }
 654
 655 static inline notrace void
 656 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 657 {
 658         struct task_struct *tsk = current;
 659         unsigned long pc;
 660
 661         pc = preempt_count();
 662
 663         entry->preempt_count    = pc & 0xff;
 664         entry->pid              = tsk->pid;
 665         entry->t                = ftrace_now(raw_smp_processor_id());
 666         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 667                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 668                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 669                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 670 }
 671
 672 notrace void
 673 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 674        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 675 {
 676         struct trace_entry *entry;
 677
 678         spin_lock(&data->lock);
 679         entry                   = tracing_get_trace_entry(tr, data);
 680         tracing_generic_entry_update(entry, flags);
 681         entry->type             = TRACE_FN;
 682         entry->fn.ip            = ip;
 683         entry->fn.parent_ip     = parent_ip;
 684         spin_unlock(&data->lock);
 685 }
 686
 687 notrace void
 688 tracing_sched_switch_trace(struct trace_array *tr,
 689                            struct trace_array_cpu *data,
 690                            struct task_struct *prev, struct task_struct *next,
 691                            unsigned long flags)
 692 {
 693         struct trace_entry *entry;
 694
 695         spin_lock(&data->lock);
 696         entry                   = tracing_get_trace_entry(tr, data);
 697         tracing_generic_entry_update(entry, flags);
 698         entry->type             = TRACE_CTX;
 699         entry->ctx.prev_pid     = prev->pid;
 700         entry->ctx.prev_prio    = prev->prio;
 701         entry->ctx.prev_state   = prev->state;
 702         entry->ctx.next_pid     = next->pid;
 703         entry->ctx.next_prio    = next->prio;
 704         spin_unlock(&data->lock);
 705 }
 706
 707 enum trace_file_type {
 708         TRACE_FILE_LAT_FMT      = 1,
 709 };
 710
 711 static struct trace_entry *
 712 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 713                 struct trace_iterator *iter, int cpu)
 714 {
 715         struct page *page;
 716         struct trace_entry *array;
 717
 718         if (iter->next_idx[cpu] >= tr->entries ||
 719             iter->next_idx[cpu] >= data->trace_idx ||
 720             (data->trace_head == data->trace_tail &&
 721              data->trace_head_idx == data->trace_tail_idx))
 722                 return NULL;
 723
 724         if (!iter->next_page[cpu]) {
 725                 /* Initialize the iterator for this cpu trace buffer */
 726                 WARN_ON(!data->trace_tail);
 727                 page = virt_to_page(data->trace_tail);
 728                 iter->next_page[cpu] = &page->lru;
 729                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 730         }
 731
 732         page = list_entry(iter->next_page[cpu], struct page, lru);
 733         BUG_ON(&data->trace_pages == &page->lru);
 734
 735         array = page_address(page);
 736
 737         /* Still possible to catch up to the tail */
 738         if (iter->next_idx[cpu] && array == data->trace_tail &&
 739             iter->next_page_idx[cpu] == data->trace_tail_idx)
 740                 return NULL;
 741
 742         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 743         return &array[iter->next_page_idx[cpu]];
 744 }
 745
 746 static struct notrace trace_entry *
 747 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 748 {
 749         struct trace_array *tr = iter->tr;
 750         struct trace_entry *ent, *next = NULL;
 751         int next_cpu = -1;
 752         int cpu;
 753
 754         for_each_possible_cpu(cpu) {
 755                 if (!head_page(tr->data[cpu]))
 756                         continue;
 757                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 758                 /*
 759                  * Pick the entry with the smallest timestamp:
 760                  */
 761                 if (ent && (!next || ent->t < next->t)) {
 762                         next = ent;
 763                         next_cpu = cpu;
 764                 }
 765         }
 766
 767         if (ent_cpu)
 768                 *ent_cpu = next_cpu;
 769
 770         return next;
 771 }
 772
 773 static notrace void
 774 trace_iterator_increment(struct trace_iterator *iter)
 775 {
 776         iter->idx++;
 777         iter->next_idx[iter->cpu]++;
 778         iter->next_page_idx[iter->cpu]++;
 779         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 780                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 781
 782                 iter->next_page_idx[iter->cpu] = 0;
 783                 iter->next_page[iter->cpu] =
 784                         trace_next_list(data, iter->next_page[iter->cpu]);
 785         }
 786 }
 787
 788 static notrace void
 789 trace_consume(struct trace_iterator *iter)
 790 {
 791         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 792
 793         data->trace_tail_idx++;
 794         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 795                 data->trace_tail = trace_next_page(data, data->trace_tail);
 796                 data->trace_tail_idx = 0;
 797         }
 798
 799         /* Check if we empty it, then reset the index */
 800         if (data->trace_head == data->trace_tail &&
 801             data->trace_head_idx == data->trace_tail_idx)
 802                 data->trace_idx = 0;
 803 }
 804
 805 static notrace void *
 806 find_next_entry_inc(struct trace_iterator *iter)
 807 {
 808         struct trace_entry *next;
 809         int next_cpu = -1;
 810
 811         next = find_next_entry(iter, &next_cpu);
 812
 813         iter->prev_ent = iter->ent;
 814         iter->prev_cpu = iter->cpu;
 815
 816         iter->ent = next;
 817         iter->cpu = next_cpu;
 818
 819         if (next)
 820                 trace_iterator_increment(iter);
 821
 822         return next ? iter : NULL;
 823 }
 824
 825 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 826 {
 827         struct trace_iterator *iter = m->private;
 828         void *last_ent = iter->ent;
 829         int i = (int)*pos;
 830         void *ent;
 831
 832         (*pos)++;
 833
 834         /* can't go backwards */
 835         if (iter->idx > i)
 836                 return NULL;
 837
 838         if (iter->idx < 0)
 839                 ent = find_next_entry_inc(iter);
 840         else
 841                 ent = iter;
 842
 843         while (ent && iter->idx < i)
 844                 ent = find_next_entry_inc(iter);
 845
 846         iter->pos = *pos;
 847
 848         if (last_ent && !ent)
 849                 seq_puts(m, "\n\nvim:ft=help\n");
 850
 851         return ent;
 852 }
 853
 854 static void *s_start(struct seq_file *m, loff_t *pos)
 855 {
 856         struct trace_iterator *iter = m->private;
 857         void *p = NULL;
 858         loff_t l = 0;
 859         int i;
 860
 861         mutex_lock(&trace_types_lock);
 862
 863         if (!current_trace || current_trace != iter->trace)
 864                 return NULL;
 865
 866         atomic_inc(&trace_record_cmdline_disabled);
 867
 868         /* let the tracer grab locks here if needed */
 869         if (current_trace->start)
 870                 current_trace->start(iter);
 871
 872         if (*pos != iter->pos) {
 873                 iter->ent = NULL;
 874                 iter->cpu = 0;
 875                 iter->idx = -1;
 876                 iter->prev_ent = NULL;
 877                 iter->prev_cpu = -1;
 878
 879                 for_each_possible_cpu(i) {
 880                         iter->next_idx[i] = 0;
 881                         iter->next_page[i] = NULL;
 882                 }
 883
 884                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 885                         ;
 886
 887         } else {
 888                 l = *pos - 1;
 889                 p = s_next(m, p, &l);
 890         }
 891
 892         return p;
 893 }
 894
 895 static void s_stop(struct seq_file *m, void *p)
 896 {
 897         struct trace_iterator *iter = m->private;
 898
 899         atomic_dec(&trace_record_cmdline_disabled);
 900
 901         /* let the tracer release locks here if needed */
 902         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 903                 iter->trace->stop(iter);
 904
 905         mutex_unlock(&trace_types_lock);
 906 }
 907
 908 static int
 909 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 910 {
 911 #ifdef CONFIG_KALLSYMS
 912         char str[KSYM_SYMBOL_LEN];
 913
 914         kallsyms_lookup(address, NULL, NULL, NULL, str);
 915
 916         return trace_seq_printf(s, fmt, str);
 917 #endif
 918         return 1;
 919 }
 920
 921 static int
 922 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 923                      unsigned long address)
 924 {
 925 #ifdef CONFIG_KALLSYMS
 926         char str[KSYM_SYMBOL_LEN];
 927
 928         sprint_symbol(str, address);
 929         return trace_seq_printf(s, fmt, str);
 930 #endif
 931         return 1;
 932 }
 933
 934 #ifndef CONFIG_64BIT
 935 # define IP_FMT "%08lx"
 936 #else
 937 # define IP_FMT "%016lx"
 938 #endif
 939
 940 static notrace int
 941 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 942 {
 943         int ret;
 944
 945         if (!ip)
 946                 return trace_seq_printf(s, "0");
 947
 948         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 949                 ret = seq_print_sym_offset(s, "%s", ip);
 950         else
 951                 ret = seq_print_sym_short(s, "%s", ip);
 952
 953         if (!ret)
 954                 return 0;
 955
 956         if (sym_flags & TRACE_ITER_SYM_ADDR)
 957                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 958         return ret;
 959 }
 960
 961 static notrace void print_lat_help_header(struct seq_file *m)
 962 {
 963         seq_puts(m, "#                _------=> CPU#            \n");
 964         seq_puts(m, "#               / _-----=> irqs-off        \n");
 965         seq_puts(m, "#              | / _----=> need-resched    \n");
 966         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 967         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 968         seq_puts(m, "#              |||| /                      \n");
 969         seq_puts(m, "#              |||||     delay             \n");
 970         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 971         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 972 }
 973
 974 static notrace void print_func_help_header(struct seq_file *m)
 975 {
 976         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 977         seq_puts(m, "#              | |      |          |         |\n");
 978 }
 979
 980
 981 static notrace void
 982 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 983 {
 984         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 985         struct trace_array *tr = iter->tr;
 986         struct trace_array_cpu *data = tr->data[tr->cpu];
 987         struct tracer *type = current_trace;
 988         unsigned long total   = 0;
 989         unsigned long entries = 0;
 990         int cpu;
 991         const char *name = "preemption";
 992
 993         if (type)
 994                 name = type->name;
 995
 996         for_each_possible_cpu(cpu) {
 997                 if (head_page(tr->data[cpu])) {
 998                         total += tr->data[cpu]->trace_idx;
 999                         if (tr->data[cpu]->trace_idx > tr->entries)
1000                                 entries += tr->entries;
1001                         else
1002                                 entries += tr->data[cpu]->trace_idx;
1003                 }
1004         }
1005
1006         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1007                    name, UTS_RELEASE);
1008         seq_puts(m, "-----------------------------------"
1009                  "---------------------------------\n");
1010         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1011                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1012                    nsecs_to_usecs(data->saved_latency),
1013                    entries,
1014                    total,
1015                    tr->cpu,
1016 #if defined(CONFIG_PREEMPT_NONE)
1017                    "server",
1018 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1019                    "desktop",
1020 #elif defined(CONFIG_PREEMPT_DESKTOP)
1021                    "preempt",
1022 #else
1023                    "unknown",
1024 #endif
1025                    /* These are reserved for later use */
1026                    0, 0, 0, 0);
1027 #ifdef CONFIG_SMP
1028         seq_printf(m, " #P:%d)\n", num_online_cpus());
1029 #else
1030         seq_puts(m, ")\n");
1031 #endif
1032         seq_puts(m, "    -----------------\n");
1033         seq_printf(m, "    | task: %.16s-%d "
1034                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1035                    data->comm, data->pid, data->uid, data->nice,
1036                    data->policy, data->rt_priority);
1037         seq_puts(m, "    -----------------\n");
1038
1039         if (data->critical_start) {
1040                 seq_puts(m, " => started at: ");
1041                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1042                 trace_print_seq(m, &iter->seq);
1043                 seq_puts(m, "\n => ended at:   ");
1044                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1045                 trace_print_seq(m, &iter->seq);
1046                 seq_puts(m, "\n");
1047         }
1048
1049         seq_puts(m, "\n");
1050 }
1051
1052 static notrace void
1053 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1054 {
1055         int hardirq, softirq;
1056         char *comm;
1057
1058         comm = trace_find_cmdline(entry->pid);
1059
1060         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1061         trace_seq_printf(s, "%d", cpu);
1062         trace_seq_printf(s, "%c%c",
1063                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1064                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1065
1066         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1067         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1068         if (hardirq && softirq)
1069                 trace_seq_putc(s, 'H');
1070         else {
1071                 if (hardirq)
1072                         trace_seq_putc(s, 'h');
1073                 else {
1074                         if (softirq)
1075                                 trace_seq_putc(s, 's');
1076                         else
1077                                 trace_seq_putc(s, '.');
1078                 }
1079         }
1080
1081         if (entry->preempt_count)
1082                 trace_seq_printf(s, "%x", entry->preempt_count);
1083         else
1084                 trace_seq_puts(s, ".");
1085 }
1086
1087 unsigned long preempt_mark_thresh = 100;
1088
1089 static notrace void
1090 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1091                     unsigned long rel_usecs)
1092 {
1093         trace_seq_printf(s, " %4lldus", abs_usecs);
1094         if (rel_usecs > preempt_mark_thresh)
1095                 trace_seq_puts(s, "!: ");
1096         else if (rel_usecs > 1)
1097                 trace_seq_puts(s, "+: ");
1098         else
1099                 trace_seq_puts(s, " : ");
1100 }
1101
1102 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1103
1104 static notrace void
1105 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1106 {
1107         struct trace_seq *s = &iter->seq;
1108         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1109         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1110         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1111         struct trace_entry *entry = iter->ent;
1112         unsigned long abs_usecs;
1113         unsigned long rel_usecs;
1114         char *comm;
1115         int S;
1116
1117         if (!next_entry)
1118                 next_entry = entry;
1119         rel_usecs = ns2usecs(next_entry->t - entry->t);
1120         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1121
1122         if (verbose) {
1123                 comm = trace_find_cmdline(entry->pid);
1124                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1125                                  " %ld.%03ldms (+%ld.%03ldms): ",
1126                                  comm,
1127                                  entry->pid, cpu, entry->flags,
1128                                  entry->preempt_count, trace_idx,
1129                                  ns2usecs(entry->t),
1130                                  abs_usecs/1000,
1131                                  abs_usecs % 1000, rel_usecs/1000,
1132                                  rel_usecs % 1000);
1133         } else {
1134                 lat_print_generic(s, entry, cpu);
1135                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1136         }
1137         switch (entry->type) {
1138         case TRACE_FN:
1139                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1140                 trace_seq_puts(s, " (");
1141                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1142                 trace_seq_puts(s, ")\n");
1143                 break;
1144         case TRACE_CTX:
1145                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1146                         state_to_char[entry->ctx.prev_state] : 'X';
1147                 comm = trace_find_cmdline(entry->ctx.next_pid);
1148                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1149                                  entry->ctx.prev_pid,
1150                                  entry->ctx.prev_prio,
1151                                  S,
1152                                  entry->ctx.next_pid,
1153                                  entry->ctx.next_prio,
1154                                  comm);
1155                 break;
1156         default:
1157                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1158         }
1159 }
1160
1161 static notrace int
1162 print_trace_fmt(struct trace_iterator *iter)
1163 {
1164         struct trace_seq *s = &iter->seq;
1165         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1166         struct trace_entry *entry;
1167         unsigned long usec_rem;
1168         unsigned long long t;
1169         unsigned long secs;
1170         char *comm;
1171         int S;
1172         int ret;
1173
1174         entry = iter->ent;
1175
1176         comm = trace_find_cmdline(iter->ent->pid);
1177
1178         t = ns2usecs(entry->t);
1179         usec_rem = do_div(t, 1000000ULL);
1180         secs = (unsigned long)t;
1181
1182         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1183         if (!ret)
1184                 return 0;
1185         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1186         if (!ret)
1187                 return 0;
1188         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1189         if (!ret)
1190                 return 0;
1191
1192         switch (entry->type) {
1193         case TRACE_FN:
1194                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1195                 if (!ret)
1196                         return 0;
1197                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1198                                                 entry->fn.parent_ip) {
1199                         ret = trace_seq_printf(s, " <-");
1200                         if (!ret)
1201                                 return 0;
1202                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1203                                                sym_flags);
1204                         if (!ret)
1205                                 return 0;
1206                 }
1207                 ret = trace_seq_printf(s, "\n");
1208                 if (!ret)
1209                         return 0;
1210                 break;
1211         case TRACE_CTX:
1212                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1213                         state_to_char[entry->ctx.prev_state] : 'X';
1214                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1215                                        entry->ctx.prev_pid,
1216                                        entry->ctx.prev_prio,
1217                                        S,
1218                                        entry->ctx.next_pid,
1219                                        entry->ctx.next_prio);
1220                 if (!ret)
1221                         return 0;
1222                 break;
1223         }
1224         return 1;
1225 }
1226
1227 static int trace_empty(struct trace_iterator *iter)
1228 {
1229         struct trace_array_cpu *data;
1230         int cpu;
1231
1232         for_each_possible_cpu(cpu) {
1233                 data = iter->tr->data[cpu];
1234
1235                 if (head_page(data) && data->trace_idx &&
1236                     (data->trace_tail != data->trace_head ||
1237                      data->trace_tail_idx != data->trace_head_idx))
1238                         return 0;
1239         }
1240         return 1;
1241 }
1242
1243 static int s_show(struct seq_file *m, void *v)
1244 {
1245         struct trace_iterator *iter = v;
1246
1247         if (iter->ent == NULL) {
1248                 if (iter->tr) {
1249                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1250                         seq_puts(m, "#\n");
1251                 }
1252                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1253                         /* print nothing if the buffers are empty */
1254                         if (trace_empty(iter))
1255                                 return 0;
1256                         print_trace_header(m, iter);
1257                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1258                                 print_lat_help_header(m);
1259                 } else {
1260                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1261                                 print_func_help_header(m);
1262                 }
1263         } else {
1264                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1265                         print_lat_fmt(iter, iter->idx, iter->cpu);
1266                 else
1267                         print_trace_fmt(iter);
1268                 trace_print_seq(m, &iter->seq);
1269         }
1270
1271         return 0;
1272 }
1273
1274 static struct seq_operations tracer_seq_ops = {
1275         .start          = s_start,
1276         .next           = s_next,
1277         .stop           = s_stop,
1278         .show           = s_show,
1279 };
1280
1281 static struct trace_iterator notrace *
1282 __tracing_open(struct inode *inode, struct file *file, int *ret)
1283 {
1284         struct trace_iterator *iter;
1285
1286         if (tracing_disabled) {
1287                 *ret = -ENODEV;
1288                 return NULL;
1289         }
1290
1291         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1292         if (!iter) {
1293                 *ret = -ENOMEM;
1294                 goto out;
1295         }
1296
1297         mutex_lock(&trace_types_lock);
1298         if (current_trace && current_trace->print_max)
1299                 iter->tr = &max_tr;
1300         else
1301                 iter->tr = inode->i_private;
1302         iter->trace = current_trace;
1303         iter->pos = -1;
1304
1305         /* TODO stop tracer */
1306         *ret = seq_open(file, &tracer_seq_ops);
1307         if (!*ret) {
1308                 struct seq_file *m = file->private_data;
1309                 m->private = iter;
1310
1311                 /* stop the trace while dumping */
1312                 if (iter->tr->ctrl)
1313                         tracer_enabled = 0;
1314
1315                 if (iter->trace && iter->trace->open)
1316                         iter->trace->open(iter);
1317         } else {
1318                 kfree(iter);
1319                 iter = NULL;
1320         }
1321         mutex_unlock(&trace_types_lock);
1322
1323  out:
1324         return iter;
1325 }
1326
1327 int tracing_open_generic(struct inode *inode, struct file *filp)
1328 {
1329         if (tracing_disabled)
1330                 return -ENODEV;
1331
1332         filp->private_data = inode->i_private;
1333         return 0;
1334 }
1335
1336 int tracing_release(struct inode *inode, struct file *file)
1337 {
1338         struct seq_file *m = (struct seq_file *)file->private_data;
1339         struct trace_iterator *iter = m->private;
1340
1341         mutex_lock(&trace_types_lock);
1342         if (iter->trace && iter->trace->close)
1343                 iter->trace->close(iter);
1344
1345         /* reenable tracing if it was previously enabled */
1346         if (iter->tr->ctrl)
1347                 tracer_enabled = 1;
1348         mutex_unlock(&trace_types_lock);
1349
1350         seq_release(inode, file);
1351         kfree(iter);
1352         return 0;
1353 }
1354
1355 static int tracing_open(struct inode *inode, struct file *file)
1356 {
1357         int ret;
1358
1359         __tracing_open(inode, file, &ret);
1360
1361         return ret;
1362 }
1363
1364 static int tracing_lt_open(struct inode *inode, struct file *file)
1365 {
1366         struct trace_iterator *iter;
1367         int ret;
1368
1369         iter = __tracing_open(inode, file, &ret);
1370
1371         if (!ret)
1372                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1373
1374         return ret;
1375 }
1376
1377
1378 static notrace void *
1379 t_next(struct seq_file *m, void *v, loff_t *pos)
1380 {
1381         struct tracer *t = m->private;
1382
1383         (*pos)++;
1384
1385         if (t)
1386                 t = t->next;
1387
1388         m->private = t;
1389
1390         return t;
1391 }
1392
1393 static void *t_start(struct seq_file *m, loff_t *pos)
1394 {
1395         struct tracer *t = m->private;
1396         loff_t l = 0;
1397
1398         mutex_lock(&trace_types_lock);
1399         for (; t && l < *pos; t = t_next(m, t, &l))
1400                 ;
1401
1402         return t;
1403 }
1404
1405 static void t_stop(struct seq_file *m, void *p)
1406 {
1407         mutex_unlock(&trace_types_lock);
1408 }
1409
1410 static int t_show(struct seq_file *m, void *v)
1411 {
1412         struct tracer *t = v;
1413
1414         if (!t)
1415                 return 0;
1416
1417         seq_printf(m, "%s", t->name);
1418         if (t->next)
1419                 seq_putc(m, ' ');
1420         else
1421                 seq_putc(m, '\n');
1422
1423         return 0;
1424 }
1425
1426 static struct seq_operations show_traces_seq_ops = {
1427         .start          = t_start,
1428         .next           = t_next,
1429         .stop           = t_stop,
1430         .show           = t_show,
1431 };
1432
1433 static int show_traces_open(struct inode *inode, struct file *file)
1434 {
1435         int ret;
1436
1437         if (tracing_disabled)
1438                 return -ENODEV;
1439
1440         ret = seq_open(file, &show_traces_seq_ops);
1441         if (!ret) {
1442                 struct seq_file *m = file->private_data;
1443                 m->private = trace_types;
1444         }
1445
1446         return ret;
1447 }
1448
1449 static struct file_operations tracing_fops = {
1450         .open           = tracing_open,
1451         .read           = seq_read,
1452         .llseek         = seq_lseek,
1453         .release        = tracing_release,
1454 };
1455
1456 static struct file_operations tracing_lt_fops = {
1457         .open           = tracing_lt_open,
1458         .read           = seq_read,
1459         .llseek         = seq_lseek,
1460         .release        = tracing_release,
1461 };
1462
1463 static struct file_operations show_traces_fops = {
1464         .open = show_traces_open,
1465         .read = seq_read,
1466         .release = seq_release,
1467 };
1468
1469 static ssize_t
1470 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1471                        size_t cnt, loff_t *ppos)
1472 {
1473         char *buf;
1474         int r = 0;
1475         int len = 0;
1476         int i;
1477
1478         /* calulate max size */
1479         for (i = 0; trace_options[i]; i++) {
1480                 len += strlen(trace_options[i]);
1481                 len += 3; /* "no" and space */
1482         }
1483
1484         /* +2 for \n and \0 */
1485         buf = kmalloc(len + 2, GFP_KERNEL);
1486         if (!buf)
1487                 return -ENOMEM;
1488
1489         for (i = 0; trace_options[i]; i++) {
1490                 if (trace_flags & (1 << i))
1491                         r += sprintf(buf + r, "%s ", trace_options[i]);
1492                 else
1493                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1494         }
1495
1496         r += sprintf(buf + r, "\n");
1497         WARN_ON(r >= len + 2);
1498
1499         r = simple_read_from_buffer(ubuf, cnt, ppos,
1500                                     buf, r);
1501
1502         kfree(buf);
1503
1504         return r;
1505 }
1506
1507 static ssize_t
1508 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1509                         size_t cnt, loff_t *ppos)
1510 {
1511         char buf[64];
1512         char *cmp = buf;
1513         int neg = 0;
1514         int i;
1515
1516         if (cnt > 63)
1517                 cnt = 63;
1518
1519         if (copy_from_user(&buf, ubuf, cnt))
1520                 return -EFAULT;
1521
1522         buf[cnt] = 0;
1523
1524         if (strncmp(buf, "no", 2) == 0) {
1525                 neg = 1;
1526                 cmp += 2;
1527         }
1528
1529         for (i = 0; trace_options[i]; i++) {
1530                 int len = strlen(trace_options[i]);
1531
1532                 if (strncmp(cmp, trace_options[i], len) == 0) {
1533                         if (neg)
1534                                 trace_flags &= ~(1 << i);
1535                         else
1536                                 trace_flags |= (1 << i);
1537                         break;
1538                 }
1539         }
1540
1541         filp->f_pos += cnt;
1542
1543         return cnt;
1544 }
1545
1546 static struct file_operations tracing_iter_fops = {
1547         .open = tracing_open_generic,
1548         .read = tracing_iter_ctrl_read,
1549         .write = tracing_iter_ctrl_write,
1550 };
1551
1552 static const char readme_msg[] =
1553         "tracing mini-HOWTO:\n\n"
1554         "# mkdir /debug\n"
1555         "# mount -t debugfs nodev /debug\n\n"
1556         "# cat /debug/tracing/available_tracers\n"
1557         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1558         "# cat /debug/tracing/current_tracer\n"
1559         "none\n"
1560         "# echo sched_switch > /debug/tracing/current_tracer\n"
1561         "# cat /debug/tracing/current_tracer\n"
1562         "sched_switch\n"
1563         "# cat /debug/tracing/iter_ctrl\n"
1564         "noprint-parent nosym-offset nosym-addr noverbose\n"
1565         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1566         "# echo 1 > /debug/tracing/tracing_enabled\n"
1567         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1568         "echo 0 > /debug/tracing/tracing_enabled\n"
1569 ;
1570
1571 static ssize_t
1572 tracing_readme_read(struct file *filp, char __user *ubuf,
1573                        size_t cnt, loff_t *ppos)
1574 {
1575         return simple_read_from_buffer(ubuf, cnt, ppos,
1576                                         readme_msg, strlen(readme_msg));
1577 }
1578
1579 static struct file_operations tracing_readme_fops = {
1580         .open = tracing_open_generic,
1581         .read = tracing_readme_read,
1582 };
1583
1584
1585 static ssize_t
1586 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1587                   size_t cnt, loff_t *ppos)
1588 {
1589         struct trace_array *tr = filp->private_data;
1590         char buf[64];
1591         int r;
1592
1593         r = sprintf(buf, "%ld\n", tr->ctrl);
1594         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1595 }
1596
1597 static ssize_t
1598 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1599                    size_t cnt, loff_t *ppos)
1600 {
1601         struct trace_array *tr = filp->private_data;
1602         long val;
1603         char buf[64];
1604
1605         if (cnt > 63)
1606                 cnt = 63;
1607
1608         if (copy_from_user(&buf, ubuf, cnt))
1609                 return -EFAULT;
1610
1611         buf[cnt] = 0;
1612
1613         val = simple_strtoul(buf, NULL, 10);
1614
1615         val = !!val;
1616
1617         mutex_lock(&trace_types_lock);
1618         if (tr->ctrl ^ val) {
1619                 if (val)
1620                         tracer_enabled = 1;
1621                 else
1622                         tracer_enabled = 0;
1623
1624                 tr->ctrl = val;
1625
1626                 if (current_trace && current_trace->ctrl_update)
1627                         current_trace->ctrl_update(tr);
1628         }
1629         mutex_unlock(&trace_types_lock);
1630
1631         filp->f_pos += cnt;
1632
1633         return cnt;
1634 }
1635
1636 static ssize_t
1637 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1638                        size_t cnt, loff_t *ppos)
1639 {
1640         char buf[max_tracer_type_len+2];
1641         int r;
1642
1643         mutex_lock(&trace_types_lock);
1644         if (current_trace)
1645                 r = sprintf(buf, "%s\n", current_trace->name);
1646         else
1647                 r = sprintf(buf, "\n");
1648         mutex_unlock(&trace_types_lock);
1649
1650         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1651 }
1652
1653 static ssize_t
1654 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1655                         size_t cnt, loff_t *ppos)
1656 {
1657         struct trace_array *tr = &global_trace;
1658         struct tracer *t;
1659         char buf[max_tracer_type_len+1];
1660         int i;
1661
1662         if (cnt > max_tracer_type_len)
1663                 cnt = max_tracer_type_len;
1664
1665         if (copy_from_user(&buf, ubuf, cnt))
1666                 return -EFAULT;
1667
1668         buf[cnt] = 0;
1669
1670         /* strip ending whitespace. */
1671         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1672                 buf[i] = 0;
1673
1674         mutex_lock(&trace_types_lock);
1675         for (t = trace_types; t; t = t->next) {
1676                 if (strcmp(t->name, buf) == 0)
1677                         break;
1678         }
1679         if (!t || t == current_trace)
1680                 goto out;
1681
1682         if (current_trace && current_trace->reset)
1683                 current_trace->reset(tr);
1684
1685         current_trace = t;
1686         if (t->init)
1687                 t->init(tr);
1688
1689  out:
1690         mutex_unlock(&trace_types_lock);
1691
1692         filp->f_pos += cnt;
1693
1694         return cnt;
1695 }
1696
1697 static ssize_t
1698 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1699                      size_t cnt, loff_t *ppos)
1700 {
1701         unsigned long *ptr = filp->private_data;
1702         char buf[64];
1703         int r;
1704
1705         r = snprintf(buf, 64, "%ld\n",
1706                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1707         if (r > 64)
1708                 r = 64;
1709         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1710 }
1711
1712 static ssize_t
1713 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1714                       size_t cnt, loff_t *ppos)
1715 {
1716         long *ptr = filp->private_data;
1717         long val;
1718         char buf[64];
1719
1720         if (cnt > 63)
1721                 cnt = 63;
1722
1723         if (copy_from_user(&buf, ubuf, cnt))
1724                 return -EFAULT;
1725
1726         buf[cnt] = 0;
1727
1728         val = simple_strtoul(buf, NULL, 10);
1729
1730         *ptr = val * 1000;
1731
1732         return cnt;
1733 }
1734
1735 static atomic_t tracing_reader;
1736
1737 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1738 {
1739         struct trace_iterator *iter;
1740
1741         if (tracing_disabled)
1742                 return -ENODEV;
1743
1744         /* We only allow for reader of the pipe */
1745         if (atomic_inc_return(&tracing_reader) != 1) {
1746                 atomic_dec(&tracing_reader);
1747                 return -EBUSY;
1748         }
1749
1750         /* create a buffer to store the information to pass to userspace */
1751         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1752         if (!iter)
1753                 return -ENOMEM;
1754
1755         iter->tr = &global_trace;
1756
1757         filp->private_data = iter;
1758
1759         return 0;
1760 }
1761
1762 static int tracing_release_pipe(struct inode *inode, struct file *file)
1763 {
1764         struct trace_iterator *iter = file->private_data;
1765
1766         kfree(iter);
1767         atomic_dec(&tracing_reader);
1768
1769         return 0;
1770 }
1771
1772 /*
1773  * Consumer reader.
1774  */
1775 static ssize_t
1776 tracing_read_pipe(struct file *filp, char __user *ubuf,
1777                   size_t cnt, loff_t *ppos)
1778 {
1779         struct trace_iterator *iter = filp->private_data;
1780         struct trace_array_cpu *data;
1781         static cpumask_t mask;
1782         struct trace_entry *entry;
1783         static int start;
1784         unsigned long flags;
1785         int read = 0;
1786         int cpu;
1787         int len;
1788         int ret;
1789
1790         /* return any leftover data */
1791         if (iter->seq.len > start) {
1792                 len = iter->seq.len - start;
1793                 if (cnt > len)
1794                         cnt = len;
1795                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1796                 if (ret)
1797                         cnt = -EFAULT;
1798
1799                 start += len;
1800
1801                 return cnt;
1802         }
1803
1804         trace_seq_reset(&iter->seq);
1805         start = 0;
1806
1807         while (trace_empty(iter)) {
1808                 /*
1809                  * This is a make-shift waitqueue. The reason we don't use
1810                  * an actual wait queue is because:
1811                  *  1) we only ever have one waiter
1812                  *  2) the tracing, traces all functions, we don't want
1813                  *     the overhead of calling wake_up and friends
1814                  *     (and tracing them too)
1815                  *     Anyway, this is really very primitive wakeup.
1816                  */
1817                 set_current_state(TASK_INTERRUPTIBLE);
1818                 iter->tr->waiter = current;
1819
1820                 /* sleep for one second, and try again. */
1821                 schedule_timeout(HZ);
1822
1823                 iter->tr->waiter = NULL;
1824
1825                 if (signal_pending(current))
1826                         return -EINTR;
1827
1828                 /*
1829                  * We block until we read something and tracing is disabled.
1830                  * We still block if tracing is disabled, but we have never
1831                  * read anything. This allows a user to cat this file, and
1832                  * then enable tracing. But after we have read something,
1833                  * we give an EOF when tracing is again disabled.
1834                  *
1835                  * iter->pos will be 0 if we haven't read anything.
1836                  */
1837                 if (!tracer_enabled && iter->pos)
1838                         break;
1839
1840                 continue;
1841         }
1842
1843         /* stop when tracing is finished */
1844         if (trace_empty(iter))
1845                 return 0;
1846
1847         if (cnt >= PAGE_SIZE)
1848                 cnt = PAGE_SIZE - 1;
1849
1850         memset(iter, 0, sizeof(*iter));
1851         iter->tr = &global_trace;
1852         iter->pos = -1;
1853
1854         /*
1855          * We need to stop all tracing on all CPUS to read the
1856          * the next buffer. This is a bit expensive, but is
1857          * not done often. We fill all what we can read,
1858          * and then release the locks again.
1859          */
1860
1861         cpus_clear(mask);
1862         local_irq_save(flags);
1863         for_each_possible_cpu(cpu) {
1864                 data = iter->tr->data[cpu];
1865
1866                 if (!head_page(data) || !data->trace_idx)
1867                         continue;
1868
1869                 atomic_inc(&data->disabled);
1870                 spin_lock(&data->lock);
1871                 cpu_set(cpu, mask);
1872         }
1873
1874         while ((entry = find_next_entry(iter, &cpu))) {
1875
1876                 if (!entry)
1877                         break;
1878
1879                 iter->ent = entry;
1880                 iter->cpu = cpu;
1881
1882                 ret = print_trace_fmt(iter);
1883                 if (!ret)
1884                         break;
1885
1886                 trace_consume(iter);
1887
1888                 if (iter->seq.len >= cnt)
1889                         break;
1890
1891         }
1892
1893         for_each_cpu_mask(cpu, mask) {
1894                 data = iter->tr->data[cpu];
1895                 spin_unlock(&data->lock);
1896                 atomic_dec(&data->disabled);
1897         }
1898         local_irq_restore(flags);
1899
1900         /* Now copy what we have to the user */
1901         read = iter->seq.len;
1902         if (read > cnt)
1903                 read = cnt;
1904
1905         ret = copy_to_user(ubuf, iter->seq.buffer, read);
1906
1907         if (read < iter->seq.len)
1908                 start = read;
1909         else
1910                 trace_seq_reset(&iter->seq);
1911
1912         if (ret)
1913                 read = -EFAULT;
1914
1915         return read;
1916 }
1917
1918 static struct file_operations tracing_max_lat_fops = {
1919         .open           = tracing_open_generic,
1920         .read           = tracing_max_lat_read,
1921         .write          = tracing_max_lat_write,
1922 };
1923
1924 static struct file_operations tracing_ctrl_fops = {
1925         .open           = tracing_open_generic,
1926         .read           = tracing_ctrl_read,
1927         .write          = tracing_ctrl_write,
1928 };
1929
1930 static struct file_operations set_tracer_fops = {
1931         .open           = tracing_open_generic,
1932         .read           = tracing_set_trace_read,
1933         .write          = tracing_set_trace_write,
1934 };
1935
1936 static struct file_operations tracing_pipe_fops = {
1937         .open           = tracing_open_pipe,
1938         .read           = tracing_read_pipe,
1939         .release        = tracing_release_pipe,
1940 };
1941
1942 #ifdef CONFIG_DYNAMIC_FTRACE
1943
1944 static ssize_t
1945 tracing_read_long(struct file *filp, char __user *ubuf,
1946                   size_t cnt, loff_t *ppos)
1947 {
1948         unsigned long *p = filp->private_data;
1949         char buf[64];
1950         int r;
1951
1952         r = sprintf(buf, "%ld\n", *p);
1953
1954         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1955 }
1956
1957 static struct file_operations tracing_read_long_fops = {
1958         .open           = tracing_open_generic,
1959         .read           = tracing_read_long,
1960 };
1961 #endif
1962
1963 static struct dentry *d_tracer;
1964
1965 struct dentry *tracing_init_dentry(void)
1966 {
1967         static int once;
1968
1969         if (d_tracer)
1970                 return d_tracer;
1971
1972         d_tracer = debugfs_create_dir("tracing", NULL);
1973
1974         if (!d_tracer && !once) {
1975                 once = 1;
1976                 pr_warning("Could not create debugfs directory 'tracing'\n");
1977                 return NULL;
1978         }
1979
1980         return d_tracer;
1981 }
1982
1983 #ifdef CONFIG_FTRACE_SELFTEST
1984 /* Let selftest have access to static functions in this file */
1985 #include "trace_selftest.c"
1986 #endif
1987
1988 static __init void tracer_init_debugfs(void)
1989 {
1990         struct dentry *d_tracer;
1991         struct dentry *entry;
1992
1993         d_tracer = tracing_init_dentry();
1994
1995         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
1996                                     &global_trace, &tracing_ctrl_fops);
1997         if (!entry)
1998                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
1999
2000         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2001                                     NULL, &tracing_iter_fops);
2002         if (!entry)
2003                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2004
2005         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2006                                     &global_trace, &tracing_lt_fops);
2007         if (!entry)
2008                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2009
2010         entry = debugfs_create_file("trace", 0444, d_tracer,
2011                                     &global_trace, &tracing_fops);
2012         if (!entry)
2013                 pr_warning("Could not create debugfs 'trace' entry\n");
2014
2015         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2016                                     &global_trace, &show_traces_fops);
2017         if (!entry)
2018                 pr_warning("Could not create debugfs 'trace' entry\n");
2019
2020         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2021                                     &global_trace, &set_tracer_fops);
2022         if (!entry)
2023                 pr_warning("Could not create debugfs 'trace' entry\n");
2024
2025         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2026                                     &tracing_max_latency,
2027                                     &tracing_max_lat_fops);
2028         if (!entry)
2029                 pr_warning("Could not create debugfs "
2030                            "'tracing_max_latency' entry\n");
2031
2032         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2033                                     &tracing_thresh, &tracing_max_lat_fops);
2034         if (!entry)
2035                 pr_warning("Could not create debugfs "
2036                            "'tracing_threash' entry\n");
2037         entry = debugfs_create_file("README", 0644, d_tracer,
2038                                     NULL, &tracing_readme_fops);
2039         if (!entry)
2040                 pr_warning("Could not create debugfs 'README' entry\n");
2041
2042         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2043                                     NULL, &tracing_pipe_fops);
2044         if (!entry)
2045                 pr_warning("Could not create debugfs "
2046                            "'tracing_threash' entry\n");
2047
2048 #ifdef CONFIG_DYNAMIC_FTRACE
2049         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2050                                     &ftrace_update_tot_cnt,
2051                                     &tracing_read_long_fops);
2052         if (!entry)
2053                 pr_warning("Could not create debugfs "
2054                            "'dyn_ftrace_total_info' entry\n");
2055 #endif
2056 }
2057
2058 /* dummy trace to disable tracing */
2059 static struct tracer no_tracer __read_mostly =
2060 {
2061         .name           = "none",
2062 };
2063
2064 static int trace_alloc_page(void)
2065 {
2066         struct trace_array_cpu *data;
2067         struct page *page, *tmp;
2068         LIST_HEAD(pages);
2069         void *array;
2070         int i;
2071
2072         /* first allocate a page for each CPU */
2073         for_each_possible_cpu(i) {
2074                 array = (void *)__get_free_page(GFP_KERNEL);
2075                 if (array == NULL) {
2076                         printk(KERN_ERR "tracer: failed to allocate page"
2077                                "for trace buffer!\n");
2078                         goto free_pages;
2079                 }
2080
2081                 page = virt_to_page(array);
2082                 list_add(&page->lru, &pages);
2083
2084 /* Only allocate if we are actually using the max trace */
2085 #ifdef CONFIG_TRACER_MAX_TRACE
2086                 array = (void *)__get_free_page(GFP_KERNEL);
2087                 if (array == NULL) {
2088                         printk(KERN_ERR "tracer: failed to allocate page"
2089                                "for trace buffer!\n");
2090                         goto free_pages;
2091                 }
2092                 page = virt_to_page(array);
2093                 list_add(&page->lru, &pages);
2094 #endif
2095         }
2096
2097         /* Now that we successfully allocate a page per CPU, add them */
2098         for_each_possible_cpu(i) {
2099                 data = global_trace.data[i];
2100                 spin_lock_init(&data->lock);
2101                 lockdep_set_class(&data->lock, &data->lock_key);
2102                 page = list_entry(pages.next, struct page, lru);
2103                 list_del_init(&page->lru);
2104                 list_add_tail(&page->lru, &data->trace_pages);
2105                 ClearPageLRU(page);
2106
2107 #ifdef CONFIG_TRACER_MAX_TRACE
2108                 data = max_tr.data[i];
2109                 spin_lock_init(&data->lock);
2110                 lockdep_set_class(&data->lock, &data->lock_key);
2111                 page = list_entry(pages.next, struct page, lru);
2112                 list_del_init(&page->lru);
2113                 list_add_tail(&page->lru, &data->trace_pages);
2114                 SetPageLRU(page);
2115 #endif
2116         }
2117         global_trace.entries += ENTRIES_PER_PAGE;
2118
2119         return 0;
2120
2121  free_pages:
2122         list_for_each_entry_safe(page, tmp, &pages, lru) {
2123                 list_del_init(&page->lru);
2124                 __free_page(page);
2125         }
2126         return -ENOMEM;
2127 }
2128
2129 __init static int tracer_alloc_buffers(void)
2130 {
2131         struct trace_array_cpu *data;
2132         void *array;
2133         struct page *page;
2134         int pages = 0;
2135         int ret = -ENOMEM;
2136         int i;
2137
2138         /* Allocate the first page for all buffers */
2139         for_each_possible_cpu(i) {
2140                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2141                 max_tr.data[i] = &per_cpu(max_data, i);
2142
2143                 array = (void *)__get_free_page(GFP_KERNEL);
2144                 if (array == NULL) {
2145                         printk(KERN_ERR "tracer: failed to allocate page"
2146                                "for trace buffer!\n");
2147                         goto free_buffers;
2148                 }
2149
2150                 /* set the array to the list */
2151                 INIT_LIST_HEAD(&data->trace_pages);
2152                 page = virt_to_page(array);
2153                 list_add(&page->lru, &data->trace_pages);
2154                 /* use the LRU flag to differentiate the two buffers */
2155                 ClearPageLRU(page);
2156
2157 /* Only allocate if we are actually using the max trace */
2158 #ifdef CONFIG_TRACER_MAX_TRACE
2159                 array = (void *)__get_free_page(GFP_KERNEL);
2160                 if (array == NULL) {
2161                         printk(KERN_ERR "tracer: failed to allocate page"
2162                                "for trace buffer!\n");
2163                         goto free_buffers;
2164                 }
2165
2166                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2167                 page = virt_to_page(array);
2168                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2169                 SetPageLRU(page);
2170 #endif
2171         }
2172
2173         /*
2174          * Since we allocate by orders of pages, we may be able to
2175          * round up a bit.
2176          */
2177         global_trace.entries = ENTRIES_PER_PAGE;
2178         pages++;
2179
2180         while (global_trace.entries < trace_nr_entries) {
2181                 if (trace_alloc_page())
2182                         break;
2183                 pages++;
2184         }
2185         max_tr.entries = global_trace.entries;
2186
2187         pr_info("tracer: %d pages allocated for %ld",
2188                 pages, trace_nr_entries);
2189         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2190         pr_info("   actual entries %ld\n", global_trace.entries);
2191
2192         tracer_init_debugfs();
2193
2194         trace_init_cmdlines();
2195
2196         register_tracer(&no_tracer);
2197         current_trace = &no_tracer;
2198
2199         /* All seems OK, enable tracing */
2200         tracing_disabled = 0;
2201
2202         return 0;
2203
2204  free_buffers:
2205         for (i-- ; i >= 0; i--) {
2206                 struct page *page, *tmp;
2207                 struct trace_array_cpu *data = global_trace.data[i];
2208
2209                 if (data) {
2210                         list_for_each_entry_safe(page, tmp,
2211                                                  &data->trace_pages, lru) {
2212                                 list_del_init(&page->lru);
2213                                 __free_page(page);
2214                         }
2215                 }
2216
2217 #ifdef CONFIG_TRACER_MAX_TRACE
2218                 data = max_tr.data[i];
2219                 if (data) {
2220                         list_for_each_entry_safe(page, tmp,
2221                                                  &data->trace_pages, lru) {
2222                                 list_del_init(&page->lru);
2223                                 __free_page(page);
2224                         }
2225                 }
2226 #endif
2227         }
2228         return ret;
2229 }
2230 fs_initcall(tracer_alloc_buffers);