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 trace_iterator_increment(struct trace_iterator *iter)
 774 {
 775         iter->idx++;
 776         iter->next_idx[iter->cpu]++;
 777         iter->next_page_idx[iter->cpu]++;
 778
 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 trace_consume(struct trace_iterator *iter)
 789 {
 790         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 791
 792         data->trace_tail_idx++;
 793         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 794                 data->trace_tail = trace_next_page(data, data->trace_tail);
 795                 data->trace_tail_idx = 0;
 796         }
 797
 798         /* Check if we empty it, then reset the index */
 799         if (data->trace_head == data->trace_tail &&
 800             data->trace_head_idx == data->trace_tail_idx)
 801                 data->trace_idx = 0;
 802 }
 803
 804 static notrace void *find_next_entry_inc(struct trace_iterator *iter)
 805 {
 806         struct trace_entry *next;
 807         int next_cpu = -1;
 808
 809         next = find_next_entry(iter, &next_cpu);
 810
 811         iter->prev_ent = iter->ent;
 812         iter->prev_cpu = iter->cpu;
 813
 814         iter->ent = next;
 815         iter->cpu = next_cpu;
 816
 817         if (next)
 818                 trace_iterator_increment(iter);
 819
 820         return next ? iter : NULL;
 821 }
 822
 823 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 824 {
 825         struct trace_iterator *iter = m->private;
 826         void *last_ent = iter->ent;
 827         int i = (int)*pos;
 828         void *ent;
 829
 830         (*pos)++;
 831
 832         /* can't go backwards */
 833         if (iter->idx > i)
 834                 return NULL;
 835
 836         if (iter->idx < 0)
 837                 ent = find_next_entry_inc(iter);
 838         else
 839                 ent = iter;
 840
 841         while (ent && iter->idx < i)
 842                 ent = find_next_entry_inc(iter);
 843
 844         iter->pos = *pos;
 845
 846         if (last_ent && !ent)
 847                 seq_puts(m, "\n\nvim:ft=help\n");
 848
 849         return ent;
 850 }
 851
 852 static void *s_start(struct seq_file *m, loff_t *pos)
 853 {
 854         struct trace_iterator *iter = m->private;
 855         void *p = NULL;
 856         loff_t l = 0;
 857         int i;
 858
 859         mutex_lock(&trace_types_lock);
 860
 861         if (!current_trace || current_trace != iter->trace)
 862                 return NULL;
 863
 864         atomic_inc(&trace_record_cmdline_disabled);
 865
 866         /* let the tracer grab locks here if needed */
 867         if (current_trace->start)
 868                 current_trace->start(iter);
 869
 870         if (*pos != iter->pos) {
 871                 iter->ent = NULL;
 872                 iter->cpu = 0;
 873                 iter->idx = -1;
 874                 iter->prev_ent = NULL;
 875                 iter->prev_cpu = -1;
 876
 877                 for_each_possible_cpu(i) {
 878                         iter->next_idx[i] = 0;
 879                         iter->next_page[i] = NULL;
 880                 }
 881
 882                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 883                         ;
 884
 885         } else {
 886                 l = *pos - 1;
 887                 p = s_next(m, p, &l);
 888         }
 889
 890         return p;
 891 }
 892
 893 static void s_stop(struct seq_file *m, void *p)
 894 {
 895         struct trace_iterator *iter = m->private;
 896
 897         atomic_dec(&trace_record_cmdline_disabled);
 898
 899         /* let the tracer release locks here if needed */
 900         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 901                 iter->trace->stop(iter);
 902
 903         mutex_unlock(&trace_types_lock);
 904 }
 905
 906 static int
 907 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 908 {
 909 #ifdef CONFIG_KALLSYMS
 910         char str[KSYM_SYMBOL_LEN];
 911
 912         kallsyms_lookup(address, NULL, NULL, NULL, str);
 913
 914         return trace_seq_printf(s, fmt, str);
 915 #endif
 916         return 1;
 917 }
 918
 919 static int
 920 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 921                      unsigned long address)
 922 {
 923 #ifdef CONFIG_KALLSYMS
 924         char str[KSYM_SYMBOL_LEN];
 925
 926         sprint_symbol(str, address);
 927         return trace_seq_printf(s, fmt, str);
 928 #endif
 929         return 1;
 930 }
 931
 932 #ifndef CONFIG_64BIT
 933 # define IP_FMT "%08lx"
 934 #else
 935 # define IP_FMT "%016lx"
 936 #endif
 937
 938 static notrace int
 939 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 940 {
 941         int ret;
 942
 943         if (!ip)
 944                 return trace_seq_printf(s, "0");
 945
 946         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 947                 ret = seq_print_sym_offset(s, "%s", ip);
 948         else
 949                 ret = seq_print_sym_short(s, "%s", ip);
 950
 951         if (!ret)
 952                 return 0;
 953
 954         if (sym_flags & TRACE_ITER_SYM_ADDR)
 955                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 956         return ret;
 957 }
 958
 959 static notrace void print_lat_help_header(struct seq_file *m)
 960 {
 961         seq_puts(m, "#                _------=> CPU#            \n");
 962         seq_puts(m, "#               / _-----=> irqs-off        \n");
 963         seq_puts(m, "#              | / _----=> need-resched    \n");
 964         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 965         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 966         seq_puts(m, "#              |||| /                      \n");
 967         seq_puts(m, "#              |||||     delay             \n");
 968         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 969         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 970 }
 971
 972 static notrace void print_func_help_header(struct seq_file *m)
 973 {
 974         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 975         seq_puts(m, "#              | |      |          |         |\n");
 976 }
 977
 978
 979 static notrace void
 980 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 981 {
 982         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 983         struct trace_array *tr = iter->tr;
 984         struct trace_array_cpu *data = tr->data[tr->cpu];
 985         struct tracer *type = current_trace;
 986         unsigned long total   = 0;
 987         unsigned long entries = 0;
 988         int cpu;
 989         const char *name = "preemption";
 990
 991         if (type)
 992                 name = type->name;
 993
 994         for_each_possible_cpu(cpu) {
 995                 if (head_page(tr->data[cpu])) {
 996                         total += tr->data[cpu]->trace_idx;
 997                         if (tr->data[cpu]->trace_idx > tr->entries)
 998                                 entries += tr->entries;
 999                         else
1000                                 entries += tr->data[cpu]->trace_idx;
1001                 }
1002         }
1003
1004         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1005                    name, UTS_RELEASE);
1006         seq_puts(m, "-----------------------------------"
1007                  "---------------------------------\n");
1008         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1009                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1010                    nsecs_to_usecs(data->saved_latency),
1011                    entries,
1012                    total,
1013                    tr->cpu,
1014 #if defined(CONFIG_PREEMPT_NONE)
1015                    "server",
1016 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1017                    "desktop",
1018 #elif defined(CONFIG_PREEMPT_DESKTOP)
1019                    "preempt",
1020 #else
1021                    "unknown",
1022 #endif
1023                    /* These are reserved for later use */
1024                    0, 0, 0, 0);
1025 #ifdef CONFIG_SMP
1026         seq_printf(m, " #P:%d)\n", num_online_cpus());
1027 #else
1028         seq_puts(m, ")\n");
1029 #endif
1030         seq_puts(m, "    -----------------\n");
1031         seq_printf(m, "    | task: %.16s-%d "
1032                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1033                    data->comm, data->pid, data->uid, data->nice,
1034                    data->policy, data->rt_priority);
1035         seq_puts(m, "    -----------------\n");
1036
1037         if (data->critical_start) {
1038                 seq_puts(m, " => started at: ");
1039                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1040                 trace_print_seq(m, &iter->seq);
1041                 seq_puts(m, "\n => ended at:   ");
1042                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1043                 trace_print_seq(m, &iter->seq);
1044                 seq_puts(m, "\n");
1045         }
1046
1047         seq_puts(m, "\n");
1048 }
1049
1050 static notrace void
1051 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1052 {
1053         int hardirq, softirq;
1054         char *comm;
1055
1056         comm = trace_find_cmdline(entry->pid);
1057
1058         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1059         trace_seq_printf(s, "%d", cpu);
1060         trace_seq_printf(s, "%c%c",
1061                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1062                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1063
1064         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1065         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1066         if (hardirq && softirq)
1067                 trace_seq_putc(s, 'H');
1068         else {
1069                 if (hardirq)
1070                         trace_seq_putc(s, 'h');
1071                 else {
1072                         if (softirq)
1073                                 trace_seq_putc(s, 's');
1074                         else
1075                                 trace_seq_putc(s, '.');
1076                 }
1077         }
1078
1079         if (entry->preempt_count)
1080                 trace_seq_printf(s, "%x", entry->preempt_count);
1081         else
1082                 trace_seq_puts(s, ".");
1083 }
1084
1085 unsigned long preempt_mark_thresh = 100;
1086
1087 static notrace void
1088 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1089                     unsigned long rel_usecs)
1090 {
1091         trace_seq_printf(s, " %4lldus", abs_usecs);
1092         if (rel_usecs > preempt_mark_thresh)
1093                 trace_seq_puts(s, "!: ");
1094         else if (rel_usecs > 1)
1095                 trace_seq_puts(s, "+: ");
1096         else
1097                 trace_seq_puts(s, " : ");
1098 }
1099
1100 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1101
1102 static notrace void
1103 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1104 {
1105         struct trace_seq *s = &iter->seq;
1106         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1107         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1108         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1109         struct trace_entry *entry = iter->ent;
1110         unsigned long abs_usecs;
1111         unsigned long rel_usecs;
1112         char *comm;
1113         int S;
1114
1115         if (!next_entry)
1116                 next_entry = entry;
1117         rel_usecs = ns2usecs(next_entry->t - entry->t);
1118         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1119
1120         if (verbose) {
1121                 comm = trace_find_cmdline(entry->pid);
1122                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1123                                  " %ld.%03ldms (+%ld.%03ldms): ",
1124                                  comm,
1125                                  entry->pid, cpu, entry->flags,
1126                                  entry->preempt_count, trace_idx,
1127                                  ns2usecs(entry->t),
1128                                  abs_usecs/1000,
1129                                  abs_usecs % 1000, rel_usecs/1000,
1130                                  rel_usecs % 1000);
1131         } else {
1132                 lat_print_generic(s, entry, cpu);
1133                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1134         }
1135         switch (entry->type) {
1136         case TRACE_FN:
1137                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1138                 trace_seq_puts(s, " (");
1139                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1140                 trace_seq_puts(s, ")\n");
1141                 break;
1142         case TRACE_CTX:
1143                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1144                         state_to_char[entry->ctx.prev_state] : 'X';
1145                 comm = trace_find_cmdline(entry->ctx.next_pid);
1146                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1147                                  entry->ctx.prev_pid,
1148                                  entry->ctx.prev_prio,
1149                                  S,
1150                                  entry->ctx.next_pid,
1151                                  entry->ctx.next_prio,
1152                                  comm);
1153                 break;
1154         default:
1155                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1156         }
1157 }
1158
1159 static notrace int
1160 print_trace_fmt(struct trace_iterator *iter)
1161 {
1162         struct trace_seq *s = &iter->seq;
1163         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1164         struct trace_entry *entry;
1165         unsigned long usec_rem;
1166         unsigned long long t;
1167         unsigned long secs;
1168         char *comm;
1169         int S;
1170         int ret;
1171
1172         entry = iter->ent;
1173
1174         comm = trace_find_cmdline(iter->ent->pid);
1175
1176         t = ns2usecs(entry->t);
1177         usec_rem = do_div(t, 1000000ULL);
1178         secs = (unsigned long)t;
1179
1180         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1181         if (!ret)
1182                 return 0;
1183         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1184         if (!ret)
1185                 return 0;
1186         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1187         if (!ret)
1188                 return 0;
1189
1190         switch (entry->type) {
1191         case TRACE_FN:
1192                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1193                 if (!ret)
1194                         return 0;
1195                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1196                                                 entry->fn.parent_ip) {
1197                         ret = trace_seq_printf(s, " <-");
1198                         if (!ret)
1199                                 return 0;
1200                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1201                                                sym_flags);
1202                         if (!ret)
1203                                 return 0;
1204                 }
1205                 ret = trace_seq_printf(s, "\n");
1206                 if (!ret)
1207                         return 0;
1208                 break;
1209         case TRACE_CTX:
1210                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1211                         state_to_char[entry->ctx.prev_state] : 'X';
1212                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1213                                        entry->ctx.prev_pid,
1214                                        entry->ctx.prev_prio,
1215                                        S,
1216                                        entry->ctx.next_pid,
1217                                        entry->ctx.next_prio);
1218                 if (!ret)
1219                         return 0;
1220                 break;
1221         }
1222         return 1;
1223 }
1224
1225 static int trace_empty(struct trace_iterator *iter)
1226 {
1227         struct trace_array_cpu *data;
1228         int cpu;
1229
1230         for_each_possible_cpu(cpu) {
1231                 data = iter->tr->data[cpu];
1232
1233                 if (head_page(data) && data->trace_idx &&
1234                     (data->trace_tail != data->trace_head ||
1235                      data->trace_tail_idx != data->trace_head_idx))
1236                         return 0;
1237         }
1238         return 1;
1239 }
1240
1241 static int s_show(struct seq_file *m, void *v)
1242 {
1243         struct trace_iterator *iter = v;
1244
1245         if (iter->ent == NULL) {
1246                 if (iter->tr) {
1247                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1248                         seq_puts(m, "#\n");
1249                 }
1250                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1251                         /* print nothing if the buffers are empty */
1252                         if (trace_empty(iter))
1253                                 return 0;
1254                         print_trace_header(m, iter);
1255                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1256                                 print_lat_help_header(m);
1257                 } else {
1258                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1259                                 print_func_help_header(m);
1260                 }
1261         } else {
1262                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1263                         print_lat_fmt(iter, iter->idx, iter->cpu);
1264                 else
1265                         print_trace_fmt(iter);
1266                 trace_print_seq(m, &iter->seq);
1267         }
1268
1269         return 0;
1270 }
1271
1272 static struct seq_operations tracer_seq_ops = {
1273         .start          = s_start,
1274         .next           = s_next,
1275         .stop           = s_stop,
1276         .show           = s_show,
1277 };
1278
1279 static struct trace_iterator notrace *
1280 __tracing_open(struct inode *inode, struct file *file, int *ret)
1281 {
1282         struct trace_iterator *iter;
1283
1284         if (tracing_disabled) {
1285                 *ret = -ENODEV;
1286                 return NULL;
1287         }
1288
1289         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1290         if (!iter) {
1291                 *ret = -ENOMEM;
1292                 goto out;
1293         }
1294
1295         mutex_lock(&trace_types_lock);
1296         if (current_trace && current_trace->print_max)
1297                 iter->tr = &max_tr;
1298         else
1299                 iter->tr = inode->i_private;
1300         iter->trace = current_trace;
1301         iter->pos = -1;
1302
1303         /* TODO stop tracer */
1304         *ret = seq_open(file, &tracer_seq_ops);
1305         if (!*ret) {
1306                 struct seq_file *m = file->private_data;
1307                 m->private = iter;
1308
1309                 /* stop the trace while dumping */
1310                 if (iter->tr->ctrl)
1311                         tracer_enabled = 0;
1312
1313                 if (iter->trace && iter->trace->open)
1314                         iter->trace->open(iter);
1315         } else {
1316                 kfree(iter);
1317                 iter = NULL;
1318         }
1319         mutex_unlock(&trace_types_lock);
1320
1321  out:
1322         return iter;
1323 }
1324
1325 int tracing_open_generic(struct inode *inode, struct file *filp)
1326 {
1327         if (tracing_disabled)
1328                 return -ENODEV;
1329
1330         filp->private_data = inode->i_private;
1331         return 0;
1332 }
1333
1334 int tracing_release(struct inode *inode, struct file *file)
1335 {
1336         struct seq_file *m = (struct seq_file *)file->private_data;
1337         struct trace_iterator *iter = m->private;
1338
1339         mutex_lock(&trace_types_lock);
1340         if (iter->trace && iter->trace->close)
1341                 iter->trace->close(iter);
1342
1343         /* reenable tracing if it was previously enabled */
1344         if (iter->tr->ctrl)
1345                 tracer_enabled = 1;
1346         mutex_unlock(&trace_types_lock);
1347
1348         seq_release(inode, file);
1349         kfree(iter);
1350         return 0;
1351 }
1352
1353 static int tracing_open(struct inode *inode, struct file *file)
1354 {
1355         int ret;
1356
1357         __tracing_open(inode, file, &ret);
1358
1359         return ret;
1360 }
1361
1362 static int tracing_lt_open(struct inode *inode, struct file *file)
1363 {
1364         struct trace_iterator *iter;
1365         int ret;
1366
1367         iter = __tracing_open(inode, file, &ret);
1368
1369         if (!ret)
1370                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1371
1372         return ret;
1373 }
1374
1375
1376 static notrace void *
1377 t_next(struct seq_file *m, void *v, loff_t *pos)
1378 {
1379         struct tracer *t = m->private;
1380
1381         (*pos)++;
1382
1383         if (t)
1384                 t = t->next;
1385
1386         m->private = t;
1387
1388         return t;
1389 }
1390
1391 static void *t_start(struct seq_file *m, loff_t *pos)
1392 {
1393         struct tracer *t = m->private;
1394         loff_t l = 0;
1395
1396         mutex_lock(&trace_types_lock);
1397         for (; t && l < *pos; t = t_next(m, t, &l))
1398                 ;
1399
1400         return t;
1401 }
1402
1403 static void t_stop(struct seq_file *m, void *p)
1404 {
1405         mutex_unlock(&trace_types_lock);
1406 }
1407
1408 static int t_show(struct seq_file *m, void *v)
1409 {
1410         struct tracer *t = v;
1411
1412         if (!t)
1413                 return 0;
1414
1415         seq_printf(m, "%s", t->name);
1416         if (t->next)
1417                 seq_putc(m, ' ');
1418         else
1419                 seq_putc(m, '\n');
1420
1421         return 0;
1422 }
1423
1424 static struct seq_operations show_traces_seq_ops = {
1425         .start          = t_start,
1426         .next           = t_next,
1427         .stop           = t_stop,
1428         .show           = t_show,
1429 };
1430
1431 static int show_traces_open(struct inode *inode, struct file *file)
1432 {
1433         int ret;
1434
1435         if (tracing_disabled)
1436                 return -ENODEV;
1437
1438         ret = seq_open(file, &show_traces_seq_ops);
1439         if (!ret) {
1440                 struct seq_file *m = file->private_data;
1441                 m->private = trace_types;
1442         }
1443
1444         return ret;
1445 }
1446
1447 static struct file_operations tracing_fops = {
1448         .open           = tracing_open,
1449         .read           = seq_read,
1450         .llseek         = seq_lseek,
1451         .release        = tracing_release,
1452 };
1453
1454 static struct file_operations tracing_lt_fops = {
1455         .open           = tracing_lt_open,
1456         .read           = seq_read,
1457         .llseek         = seq_lseek,
1458         .release        = tracing_release,
1459 };
1460
1461 static struct file_operations show_traces_fops = {
1462         .open = show_traces_open,
1463         .read = seq_read,
1464         .release = seq_release,
1465 };
1466
1467 static ssize_t
1468 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1469                        size_t cnt, loff_t *ppos)
1470 {
1471         char *buf;
1472         int r = 0;
1473         int len = 0;
1474         int i;
1475
1476         /* calulate max size */
1477         for (i = 0; trace_options[i]; i++) {
1478                 len += strlen(trace_options[i]);
1479                 len += 3; /* "no" and space */
1480         }
1481
1482         /* +2 for \n and \0 */
1483         buf = kmalloc(len + 2, GFP_KERNEL);
1484         if (!buf)
1485                 return -ENOMEM;
1486
1487         for (i = 0; trace_options[i]; i++) {
1488                 if (trace_flags & (1 << i))
1489                         r += sprintf(buf + r, "%s ", trace_options[i]);
1490                 else
1491                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1492         }
1493
1494         r += sprintf(buf + r, "\n");
1495         WARN_ON(r >= len + 2);
1496
1497         r = simple_read_from_buffer(ubuf, cnt, ppos,
1498                                     buf, r);
1499
1500         kfree(buf);
1501
1502         return r;
1503 }
1504
1505 static ssize_t
1506 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1507                         size_t cnt, loff_t *ppos)
1508 {
1509         char buf[64];
1510         char *cmp = buf;
1511         int neg = 0;
1512         int i;
1513
1514         if (cnt > 63)
1515                 cnt = 63;
1516
1517         if (copy_from_user(&buf, ubuf, cnt))
1518                 return -EFAULT;
1519
1520         buf[cnt] = 0;
1521
1522         if (strncmp(buf, "no", 2) == 0) {
1523                 neg = 1;
1524                 cmp += 2;
1525         }
1526
1527         for (i = 0; trace_options[i]; i++) {
1528                 int len = strlen(trace_options[i]);
1529
1530                 if (strncmp(cmp, trace_options[i], len) == 0) {
1531                         if (neg)
1532                                 trace_flags &= ~(1 << i);
1533                         else
1534                                 trace_flags |= (1 << i);
1535                         break;
1536                 }
1537         }
1538
1539         filp->f_pos += cnt;
1540
1541         return cnt;
1542 }
1543
1544 static struct file_operations tracing_iter_fops = {
1545         .open = tracing_open_generic,
1546         .read = tracing_iter_ctrl_read,
1547         .write = tracing_iter_ctrl_write,
1548 };
1549
1550 static const char readme_msg[] =
1551         "tracing mini-HOWTO:\n\n"
1552         "# mkdir /debug\n"
1553         "# mount -t debugfs nodev /debug\n\n"
1554         "# cat /debug/tracing/available_tracers\n"
1555         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1556         "# cat /debug/tracing/current_tracer\n"
1557         "none\n"
1558         "# echo sched_switch > /debug/tracing/current_tracer\n"
1559         "# cat /debug/tracing/current_tracer\n"
1560         "sched_switch\n"
1561         "# cat /debug/tracing/iter_ctrl\n"
1562         "noprint-parent nosym-offset nosym-addr noverbose\n"
1563         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1564         "# echo 1 > /debug/tracing/tracing_enabled\n"
1565         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1566         "echo 0 > /debug/tracing/tracing_enabled\n"
1567 ;
1568
1569 static ssize_t
1570 tracing_readme_read(struct file *filp, char __user *ubuf,
1571                        size_t cnt, loff_t *ppos)
1572 {
1573         return simple_read_from_buffer(ubuf, cnt, ppos,
1574                                         readme_msg, strlen(readme_msg));
1575 }
1576
1577 static struct file_operations tracing_readme_fops = {
1578         .open = tracing_open_generic,
1579         .read = tracing_readme_read,
1580 };
1581
1582
1583 static ssize_t
1584 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1585                   size_t cnt, loff_t *ppos)
1586 {
1587         struct trace_array *tr = filp->private_data;
1588         char buf[64];
1589         int r;
1590
1591         r = sprintf(buf, "%ld\n", tr->ctrl);
1592         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1593 }
1594
1595 static ssize_t
1596 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1597                    size_t cnt, loff_t *ppos)
1598 {
1599         struct trace_array *tr = filp->private_data;
1600         long val;
1601         char buf[64];
1602
1603         if (cnt > 63)
1604                 cnt = 63;
1605
1606         if (copy_from_user(&buf, ubuf, cnt))
1607                 return -EFAULT;
1608
1609         buf[cnt] = 0;
1610
1611         val = simple_strtoul(buf, NULL, 10);
1612
1613         val = !!val;
1614
1615         mutex_lock(&trace_types_lock);
1616         if (tr->ctrl ^ val) {
1617                 if (val)
1618                         tracer_enabled = 1;
1619                 else
1620                         tracer_enabled = 0;
1621
1622                 tr->ctrl = val;
1623
1624                 if (current_trace && current_trace->ctrl_update)
1625                         current_trace->ctrl_update(tr);
1626         }
1627         mutex_unlock(&trace_types_lock);
1628
1629         filp->f_pos += cnt;
1630
1631         return cnt;
1632 }
1633
1634 static ssize_t
1635 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1636                        size_t cnt, loff_t *ppos)
1637 {
1638         char buf[max_tracer_type_len+2];
1639         int r;
1640
1641         mutex_lock(&trace_types_lock);
1642         if (current_trace)
1643                 r = sprintf(buf, "%s\n", current_trace->name);
1644         else
1645                 r = sprintf(buf, "\n");
1646         mutex_unlock(&trace_types_lock);
1647
1648         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1649 }
1650
1651 static ssize_t
1652 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1653                         size_t cnt, loff_t *ppos)
1654 {
1655         struct trace_array *tr = &global_trace;
1656         struct tracer *t;
1657         char buf[max_tracer_type_len+1];
1658         int i;
1659
1660         if (cnt > max_tracer_type_len)
1661                 cnt = max_tracer_type_len;
1662
1663         if (copy_from_user(&buf, ubuf, cnt))
1664                 return -EFAULT;
1665
1666         buf[cnt] = 0;
1667
1668         /* strip ending whitespace. */
1669         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1670                 buf[i] = 0;
1671
1672         mutex_lock(&trace_types_lock);
1673         for (t = trace_types; t; t = t->next) {
1674                 if (strcmp(t->name, buf) == 0)
1675                         break;
1676         }
1677         if (!t || t == current_trace)
1678                 goto out;
1679
1680         if (current_trace && current_trace->reset)
1681                 current_trace->reset(tr);
1682
1683         current_trace = t;
1684         if (t->init)
1685                 t->init(tr);
1686
1687  out:
1688         mutex_unlock(&trace_types_lock);
1689
1690         filp->f_pos += cnt;
1691
1692         return cnt;
1693 }
1694
1695 static ssize_t
1696 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1697                      size_t cnt, loff_t *ppos)
1698 {
1699         unsigned long *ptr = filp->private_data;
1700         char buf[64];
1701         int r;
1702
1703         r = snprintf(buf, 64, "%ld\n",
1704                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1705         if (r > 64)
1706                 r = 64;
1707         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1708 }
1709
1710 static ssize_t
1711 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1712                       size_t cnt, loff_t *ppos)
1713 {
1714         long *ptr = filp->private_data;
1715         long val;
1716         char buf[64];
1717
1718         if (cnt > 63)
1719                 cnt = 63;
1720
1721         if (copy_from_user(&buf, ubuf, cnt))
1722                 return -EFAULT;
1723
1724         buf[cnt] = 0;
1725
1726         val = simple_strtoul(buf, NULL, 10);
1727
1728         *ptr = val * 1000;
1729
1730         return cnt;
1731 }
1732
1733 static atomic_t tracing_reader;
1734
1735 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1736 {
1737         struct trace_iterator *iter;
1738
1739         if (tracing_disabled)
1740                 return -ENODEV;
1741
1742         /* We only allow for reader of the pipe */
1743         if (atomic_inc_return(&tracing_reader) != 1) {
1744                 atomic_dec(&tracing_reader);
1745                 return -EBUSY;
1746         }
1747
1748         /* create a buffer to store the information to pass to userspace */
1749         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1750         if (!iter)
1751                 return -ENOMEM;
1752
1753         iter->tr = &global_trace;
1754
1755         filp->private_data = iter;
1756
1757         return 0;
1758 }
1759
1760 static int tracing_release_pipe(struct inode *inode, struct file *file)
1761 {
1762         struct trace_iterator *iter = file->private_data;
1763
1764         kfree(iter);
1765         atomic_dec(&tracing_reader);
1766
1767         return 0;
1768 }
1769
1770 /*
1771  * Consumer reader.
1772  */
1773 static ssize_t
1774 tracing_read_pipe(struct file *filp, char __user *ubuf,
1775                   size_t cnt, loff_t *ppos)
1776 {
1777         struct trace_iterator *iter = filp->private_data;
1778         struct trace_array_cpu *data;
1779         static cpumask_t mask;
1780         struct trace_entry *entry;
1781         static int start;
1782         unsigned long flags;
1783         int read = 0;
1784         int cpu;
1785         int len;
1786         int ret;
1787
1788         /* return any leftover data */
1789         if (iter->seq.len > start) {
1790                 len = iter->seq.len - start;
1791                 if (cnt > len)
1792                         cnt = len;
1793                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1794                 if (ret)
1795                         cnt = -EFAULT;
1796
1797                 start += len;
1798
1799                 return cnt;
1800         }
1801
1802         trace_seq_reset(&iter->seq);
1803         start = 0;
1804
1805         while (trace_empty(iter)) {
1806                 /*
1807                  * This is a make-shift waitqueue. The reason we don't use
1808                  * an actual wait queue is because:
1809                  *  1) we only ever have one waiter
1810                  *  2) the tracing, traces all functions, we don't want
1811                  *     the overhead of calling wake_up and friends
1812                  *     (and tracing them too)
1813                  *     Anyway, this is really very primitive wakeup.
1814                  */
1815                 set_current_state(TASK_INTERRUPTIBLE);
1816                 iter->tr->waiter = current;
1817
1818                 /* sleep for one second, and try again. */
1819                 schedule_timeout(HZ);
1820
1821                 iter->tr->waiter = NULL;
1822
1823                 if (signal_pending(current))
1824                         return -EINTR;
1825
1826                 /*
1827                  * We block until we read something and tracing is disabled.
1828                  * We still block if tracing is disabled, but we have never
1829                  * read anything. This allows a user to cat this file, and
1830                  * then enable tracing. But after we have read something,
1831                  * we give an EOF when tracing is again disabled.
1832                  *
1833                  * iter->pos will be 0 if we haven't read anything.
1834                  */
1835                 if (!tracer_enabled && iter->pos)
1836                         break;
1837
1838                 continue;
1839         }
1840
1841         /* stop when tracing is finished */
1842         if (trace_empty(iter))
1843                 return 0;
1844
1845         if (cnt >= PAGE_SIZE)
1846                 cnt = PAGE_SIZE - 1;
1847
1848         memset(iter, 0, sizeof(*iter));
1849         iter->tr = &global_trace;
1850         iter->pos = -1;
1851
1852         /*
1853          * We need to stop all tracing on all CPUS to read the
1854          * the next buffer. This is a bit expensive, but is
1855          * not done often. We fill all what we can read,
1856          * and then release the locks again.
1857          */
1858
1859         cpus_clear(mask);
1860         local_irq_save(flags);
1861         for_each_possible_cpu(cpu) {
1862                 data = iter->tr->data[cpu];
1863
1864                 if (!head_page(data) || !data->trace_idx)
1865                         continue;
1866
1867                 atomic_inc(&data->disabled);
1868                 spin_lock(&data->lock);
1869                 cpu_set(cpu, mask);
1870         }
1871
1872         while ((entry = find_next_entry_inc(iter)) != NULL) {
1873                 ret = print_trace_fmt(iter);
1874                 if (!ret)
1875                         break;
1876
1877                 trace_consume(iter);
1878
1879                 if (iter->seq.len >= cnt)
1880                         break;
1881         }
1882
1883         for_each_cpu_mask(cpu, mask) {
1884                 data = iter->tr->data[cpu];
1885                 spin_unlock(&data->lock);
1886                 atomic_dec(&data->disabled);
1887         }
1888         local_irq_restore(flags);
1889
1890         /* Now copy what we have to the user */
1891         read = iter->seq.len;
1892         if (read > cnt)
1893                 read = cnt;
1894
1895         ret = copy_to_user(ubuf, iter->seq.buffer, read);
1896
1897         if (read < iter->seq.len)
1898                 start = read;
1899         else
1900                 trace_seq_reset(&iter->seq);
1901
1902         if (ret)
1903                 read = -EFAULT;
1904
1905         return read;
1906 }
1907
1908 static struct file_operations tracing_max_lat_fops = {
1909         .open           = tracing_open_generic,
1910         .read           = tracing_max_lat_read,
1911         .write          = tracing_max_lat_write,
1912 };
1913
1914 static struct file_operations tracing_ctrl_fops = {
1915         .open           = tracing_open_generic,
1916         .read           = tracing_ctrl_read,
1917         .write          = tracing_ctrl_write,
1918 };
1919
1920 static struct file_operations set_tracer_fops = {
1921         .open           = tracing_open_generic,
1922         .read           = tracing_set_trace_read,
1923         .write          = tracing_set_trace_write,
1924 };
1925
1926 static struct file_operations tracing_pipe_fops = {
1927         .open           = tracing_open_pipe,
1928         .read           = tracing_read_pipe,
1929         .release        = tracing_release_pipe,
1930 };
1931
1932 #ifdef CONFIG_DYNAMIC_FTRACE
1933
1934 static ssize_t
1935 tracing_read_long(struct file *filp, char __user *ubuf,
1936                   size_t cnt, loff_t *ppos)
1937 {
1938         unsigned long *p = filp->private_data;
1939         char buf[64];
1940         int r;
1941
1942         r = sprintf(buf, "%ld\n", *p);
1943
1944         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1945 }
1946
1947 static struct file_operations tracing_read_long_fops = {
1948         .open           = tracing_open_generic,
1949         .read           = tracing_read_long,
1950 };
1951 #endif
1952
1953 static struct dentry *d_tracer;
1954
1955 struct dentry *tracing_init_dentry(void)
1956 {
1957         static int once;
1958
1959         if (d_tracer)
1960                 return d_tracer;
1961
1962         d_tracer = debugfs_create_dir("tracing", NULL);
1963
1964         if (!d_tracer && !once) {
1965                 once = 1;
1966                 pr_warning("Could not create debugfs directory 'tracing'\n");
1967                 return NULL;
1968         }
1969
1970         return d_tracer;
1971 }
1972
1973 #ifdef CONFIG_FTRACE_SELFTEST
1974 /* Let selftest have access to static functions in this file */
1975 #include "trace_selftest.c"
1976 #endif
1977
1978 static __init void tracer_init_debugfs(void)
1979 {
1980         struct dentry *d_tracer;
1981         struct dentry *entry;
1982
1983         d_tracer = tracing_init_dentry();
1984
1985         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
1986                                     &global_trace, &tracing_ctrl_fops);
1987         if (!entry)
1988                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
1989
1990         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
1991                                     NULL, &tracing_iter_fops);
1992         if (!entry)
1993                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
1994
1995         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
1996                                     &global_trace, &tracing_lt_fops);
1997         if (!entry)
1998                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
1999
2000         entry = debugfs_create_file("trace", 0444, d_tracer,
2001                                     &global_trace, &tracing_fops);
2002         if (!entry)
2003                 pr_warning("Could not create debugfs 'trace' entry\n");
2004
2005         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2006                                     &global_trace, &show_traces_fops);
2007         if (!entry)
2008                 pr_warning("Could not create debugfs 'trace' entry\n");
2009
2010         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2011                                     &global_trace, &set_tracer_fops);
2012         if (!entry)
2013                 pr_warning("Could not create debugfs 'trace' entry\n");
2014
2015         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2016                                     &tracing_max_latency,
2017                                     &tracing_max_lat_fops);
2018         if (!entry)
2019                 pr_warning("Could not create debugfs "
2020                            "'tracing_max_latency' entry\n");
2021
2022         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2023                                     &tracing_thresh, &tracing_max_lat_fops);
2024         if (!entry)
2025                 pr_warning("Could not create debugfs "
2026                            "'tracing_threash' entry\n");
2027         entry = debugfs_create_file("README", 0644, d_tracer,
2028                                     NULL, &tracing_readme_fops);
2029         if (!entry)
2030                 pr_warning("Could not create debugfs 'README' entry\n");
2031
2032         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2033                                     NULL, &tracing_pipe_fops);
2034         if (!entry)
2035                 pr_warning("Could not create debugfs "
2036                            "'tracing_threash' entry\n");
2037
2038 #ifdef CONFIG_DYNAMIC_FTRACE
2039         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2040                                     &ftrace_update_tot_cnt,
2041                                     &tracing_read_long_fops);
2042         if (!entry)
2043                 pr_warning("Could not create debugfs "
2044                            "'dyn_ftrace_total_info' entry\n");
2045 #endif
2046 }
2047
2048 /* dummy trace to disable tracing */
2049 static struct tracer no_tracer __read_mostly =
2050 {
2051         .name           = "none",
2052 };
2053
2054 static int trace_alloc_page(void)
2055 {
2056         struct trace_array_cpu *data;
2057         struct page *page, *tmp;
2058         LIST_HEAD(pages);
2059         void *array;
2060         int i;
2061
2062         /* first allocate a page for each CPU */
2063         for_each_possible_cpu(i) {
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
2071                 page = virt_to_page(array);
2072                 list_add(&page->lru, &pages);
2073
2074 /* Only allocate if we are actually using the max trace */
2075 #ifdef CONFIG_TRACER_MAX_TRACE
2076                 array = (void *)__get_free_page(GFP_KERNEL);
2077                 if (array == NULL) {
2078                         printk(KERN_ERR "tracer: failed to allocate page"
2079                                "for trace buffer!\n");
2080                         goto free_pages;
2081                 }
2082                 page = virt_to_page(array);
2083                 list_add(&page->lru, &pages);
2084 #endif
2085         }
2086
2087         /* Now that we successfully allocate a page per CPU, add them */
2088         for_each_possible_cpu(i) {
2089                 data = global_trace.data[i];
2090                 spin_lock_init(&data->lock);
2091                 lockdep_set_class(&data->lock, &data->lock_key);
2092                 page = list_entry(pages.next, struct page, lru);
2093                 list_del_init(&page->lru);
2094                 list_add_tail(&page->lru, &data->trace_pages);
2095                 ClearPageLRU(page);
2096
2097 #ifdef CONFIG_TRACER_MAX_TRACE
2098                 data = max_tr.data[i];
2099                 spin_lock_init(&data->lock);
2100                 lockdep_set_class(&data->lock, &data->lock_key);
2101                 page = list_entry(pages.next, struct page, lru);
2102                 list_del_init(&page->lru);
2103                 list_add_tail(&page->lru, &data->trace_pages);
2104                 SetPageLRU(page);
2105 #endif
2106         }
2107         global_trace.entries += ENTRIES_PER_PAGE;
2108
2109         return 0;
2110
2111  free_pages:
2112         list_for_each_entry_safe(page, tmp, &pages, lru) {
2113                 list_del_init(&page->lru);
2114                 __free_page(page);
2115         }
2116         return -ENOMEM;
2117 }
2118
2119 __init static int tracer_alloc_buffers(void)
2120 {
2121         struct trace_array_cpu *data;
2122         void *array;
2123         struct page *page;
2124         int pages = 0;
2125         int ret = -ENOMEM;
2126         int i;
2127
2128         /* Allocate the first page for all buffers */
2129         for_each_possible_cpu(i) {
2130                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2131                 max_tr.data[i] = &per_cpu(max_data, i);
2132
2133                 array = (void *)__get_free_page(GFP_KERNEL);
2134                 if (array == NULL) {
2135                         printk(KERN_ERR "tracer: failed to allocate page"
2136                                "for trace buffer!\n");
2137                         goto free_buffers;
2138                 }
2139
2140                 /* set the array to the list */
2141                 INIT_LIST_HEAD(&data->trace_pages);
2142                 page = virt_to_page(array);
2143                 list_add(&page->lru, &data->trace_pages);
2144                 /* use the LRU flag to differentiate the two buffers */
2145                 ClearPageLRU(page);
2146
2147 /* Only allocate if we are actually using the max trace */
2148 #ifdef CONFIG_TRACER_MAX_TRACE
2149                 array = (void *)__get_free_page(GFP_KERNEL);
2150                 if (array == NULL) {
2151                         printk(KERN_ERR "tracer: failed to allocate page"
2152                                "for trace buffer!\n");
2153                         goto free_buffers;
2154                 }
2155
2156                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2157                 page = virt_to_page(array);
2158                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2159                 SetPageLRU(page);
2160 #endif
2161         }
2162
2163         /*
2164          * Since we allocate by orders of pages, we may be able to
2165          * round up a bit.
2166          */
2167         global_trace.entries = ENTRIES_PER_PAGE;
2168         pages++;
2169
2170         while (global_trace.entries < trace_nr_entries) {
2171                 if (trace_alloc_page())
2172                         break;
2173                 pages++;
2174         }
2175         max_tr.entries = global_trace.entries;
2176
2177         pr_info("tracer: %d pages allocated for %ld",
2178                 pages, trace_nr_entries);
2179         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2180         pr_info("   actual entries %ld\n", global_trace.entries);
2181
2182         tracer_init_debugfs();
2183
2184         trace_init_cmdlines();
2185
2186         register_tracer(&no_tracer);
2187         current_trace = &no_tracer;
2188
2189         /* All seems OK, enable tracing */
2190         tracing_disabled = 0;
2191
2192         return 0;
2193
2194  free_buffers:
2195         for (i-- ; i >= 0; i--) {
2196                 struct page *page, *tmp;
2197                 struct trace_array_cpu *data = global_trace.data[i];
2198
2199                 if (data) {
2200                         list_for_each_entry_safe(page, tmp,
2201                                                  &data->trace_pages, lru) {
2202                                 list_del_init(&page->lru);
2203                                 __free_page(page);
2204                         }
2205                 }
2206
2207 #ifdef CONFIG_TRACER_MAX_TRACE
2208                 data = max_tr.data[i];
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 #endif
2217         }
2218         return ret;
2219 }
2220 fs_initcall(tracer_alloc_buffers);