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         TRACE_SPECIAL,
 141
 142         __TRACE_LAST_TYPE
 143 };
 144
 145 enum trace_flag_type {
 146         TRACE_FLAG_IRQS_OFF             = 0x01,
 147         TRACE_FLAG_NEED_RESCHED         = 0x02,
 148         TRACE_FLAG_HARDIRQ              = 0x04,
 149         TRACE_FLAG_SOFTIRQ              = 0x08,
 150 };
 151
 152 enum trace_iterator_flags {
 153         TRACE_ITER_PRINT_PARENT         = 0x01,
 154         TRACE_ITER_SYM_OFFSET           = 0x02,
 155         TRACE_ITER_SYM_ADDR             = 0x04,
 156         TRACE_ITER_VERBOSE              = 0x08,
 157         TRACE_ITER_RAW                  = 0x10,
 158         TRACE_ITER_BIN                  = 0x20,
 159 };
 160
 161 #define TRACE_ITER_SYM_MASK \
 162         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 163
 164 /* These must match the bit postions above */
 165 static const char *trace_options[] = {
 166         "print-parent",
 167         "sym-offset",
 168         "sym-addr",
 169         "verbose",
 170         "raw",
 171         "bin",
 172         NULL
 173 };
 174
 175 static unsigned trace_flags;
 176
 177 static DEFINE_SPINLOCK(ftrace_max_lock);
 178
 179 /*
 180  * Copy the new maximum trace into the separate maximum-trace
 181  * structure. (this way the maximum trace is permanently saved,
 182  * for later retrieval via /debugfs/tracing/latency_trace)
 183  */
 184 static notrace void
 185 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 186 {
 187         struct trace_array_cpu *data = tr->data[cpu];
 188
 189         max_tr.cpu = cpu;
 190         max_tr.time_start = data->preempt_timestamp;
 191
 192         data = max_tr.data[cpu];
 193         data->saved_latency = tracing_max_latency;
 194
 195         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 196         data->pid = tsk->pid;
 197         data->uid = tsk->uid;
 198         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 199         data->policy = tsk->policy;
 200         data->rt_priority = tsk->rt_priority;
 201
 202         /* record this tasks comm */
 203         tracing_record_cmdline(current);
 204 }
 205
 206 void check_pages(struct trace_array_cpu *data)
 207 {
 208         struct page *page, *tmp;
 209
 210         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 211         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 212
 213         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 214                 BUG_ON(page->lru.next->prev != &page->lru);
 215                 BUG_ON(page->lru.prev->next != &page->lru);
 216         }
 217 }
 218
 219 void *head_page(struct trace_array_cpu *data)
 220 {
 221         struct page *page;
 222
 223         check_pages(data);
 224         if (list_empty(&data->trace_pages))
 225                 return NULL;
 226
 227         page = list_entry(data->trace_pages.next, struct page, lru);
 228         BUG_ON(&page->lru == &data->trace_pages);
 229
 230         return page_address(page);
 231 }
 232
 233 static notrace int
 234 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 235 {
 236         int len = (PAGE_SIZE - 1) - s->len;
 237         va_list ap;
 238         int ret;
 239
 240         if (!len)
 241                 return 0;
 242
 243         va_start(ap, fmt);
 244         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 245         va_end(ap);
 246
 247         /* If we can't write it all, don't bother writing anything */
 248         if (ret > len)
 249                 return 0;
 250
 251         s->len += ret;
 252
 253         return len;
 254 }
 255
 256 static notrace int
 257 trace_seq_puts(struct trace_seq *s, const char *str)
 258 {
 259         int len = strlen(str);
 260
 261         if (len > ((PAGE_SIZE - 1) - s->len))
 262                 return 0;
 263
 264         memcpy(s->buffer + s->len, str, len);
 265         s->len += len;
 266
 267         return len;
 268 }
 269
 270 static notrace int
 271 trace_seq_putc(struct trace_seq *s, unsigned char c)
 272 {
 273         if (s->len >= (PAGE_SIZE - 1))
 274                 return 0;
 275
 276         s->buffer[s->len++] = c;
 277
 278         return 1;
 279 }
 280
 281 static notrace int
 282 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 283 {
 284         if (len > ((PAGE_SIZE - 1) - s->len))
 285                 return 0;
 286
 287         memcpy(s->buffer + s->len, mem, len);
 288         s->len += len;
 289
 290         return len;
 291 }
 292
 293 static notrace void
 294 trace_seq_reset(struct trace_seq *s)
 295 {
 296         s->len = 0;
 297 }
 298
 299 static notrace void
 300 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 301 {
 302         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 303
 304         s->buffer[len] = 0;
 305         seq_puts(m, s->buffer);
 306
 307         trace_seq_reset(s);
 308 }
 309
 310 notrace static void
 311 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 312 {
 313         struct list_head flip_pages;
 314
 315         INIT_LIST_HEAD(&flip_pages);
 316
 317         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 318                 sizeof(struct trace_array_cpu) -
 319                 offsetof(struct trace_array_cpu, trace_head_idx));
 320
 321         check_pages(tr1);
 322         check_pages(tr2);
 323         list_splice_init(&tr1->trace_pages, &flip_pages);
 324         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 325         list_splice_init(&flip_pages, &tr2->trace_pages);
 326         BUG_ON(!list_empty(&flip_pages));
 327         check_pages(tr1);
 328         check_pages(tr2);
 329 }
 330
 331 notrace void
 332 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 333 {
 334         struct trace_array_cpu *data;
 335         int i;
 336
 337         WARN_ON_ONCE(!irqs_disabled());
 338         spin_lock(&ftrace_max_lock);
 339         /* clear out all the previous traces */
 340         for_each_possible_cpu(i) {
 341                 data = tr->data[i];
 342                 flip_trace(max_tr.data[i], data);
 343                 tracing_reset(data);
 344         }
 345
 346         __update_max_tr(tr, tsk, cpu);
 347         spin_unlock(&ftrace_max_lock);
 348 }
 349
 350 /**
 351  * update_max_tr_single - only copy one trace over, and reset the rest
 352  * @tr - tracer
 353  * @tsk - task with the latency
 354  * @cpu - the cpu of the buffer to copy.
 355  */
 356 notrace void
 357 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 358 {
 359         struct trace_array_cpu *data = tr->data[cpu];
 360         int i;
 361
 362         WARN_ON_ONCE(!irqs_disabled());
 363         spin_lock(&ftrace_max_lock);
 364         for_each_possible_cpu(i)
 365                 tracing_reset(max_tr.data[i]);
 366
 367         flip_trace(max_tr.data[cpu], data);
 368         tracing_reset(data);
 369
 370         __update_max_tr(tr, tsk, cpu);
 371         spin_unlock(&ftrace_max_lock);
 372 }
 373
 374 int register_tracer(struct tracer *type)
 375 {
 376         struct tracer *t;
 377         int len;
 378         int ret = 0;
 379
 380         if (!type->name) {
 381                 pr_info("Tracer must have a name\n");
 382                 return -1;
 383         }
 384
 385         mutex_lock(&trace_types_lock);
 386         for (t = trace_types; t; t = t->next) {
 387                 if (strcmp(type->name, t->name) == 0) {
 388                         /* already found */
 389                         pr_info("Trace %s already registered\n",
 390                                 type->name);
 391                         ret = -1;
 392                         goto out;
 393                 }
 394         }
 395
 396 #ifdef CONFIG_FTRACE_STARTUP_TEST
 397         if (type->selftest) {
 398                 struct tracer *saved_tracer = current_trace;
 399                 struct trace_array_cpu *data;
 400                 struct trace_array *tr = &global_trace;
 401                 int saved_ctrl = tr->ctrl;
 402                 int i;
 403                 /*
 404                  * Run a selftest on this tracer.
 405                  * Here we reset the trace buffer, and set the current
 406                  * tracer to be this tracer. The tracer can then run some
 407                  * internal tracing to verify that everything is in order.
 408                  * If we fail, we do not register this tracer.
 409                  */
 410                 for_each_possible_cpu(i) {
 411                         data = tr->data[i];
 412                         if (!head_page(data))
 413                                 continue;
 414                         tracing_reset(data);
 415                 }
 416                 current_trace = type;
 417                 tr->ctrl = 0;
 418                 /* the test is responsible for initializing and enabling */
 419                 pr_info("Testing tracer %s: ", type->name);
 420                 ret = type->selftest(type, tr);
 421                 /* the test is responsible for resetting too */
 422                 current_trace = saved_tracer;
 423                 tr->ctrl = saved_ctrl;
 424                 if (ret) {
 425                         printk(KERN_CONT "FAILED!\n");
 426                         goto out;
 427                 }
 428                 /* Only reset on passing, to avoid touching corrupted buffers */
 429                 for_each_possible_cpu(i) {
 430                         data = tr->data[i];
 431                         if (!head_page(data))
 432                                 continue;
 433                         tracing_reset(data);
 434                 }
 435                 printk(KERN_CONT "PASSED\n");
 436         }
 437 #endif
 438
 439         type->next = trace_types;
 440         trace_types = type;
 441         len = strlen(type->name);
 442         if (len > max_tracer_type_len)
 443                 max_tracer_type_len = len;
 444
 445  out:
 446         mutex_unlock(&trace_types_lock);
 447
 448         return ret;
 449 }
 450
 451 void unregister_tracer(struct tracer *type)
 452 {
 453         struct tracer **t;
 454         int len;
 455
 456         mutex_lock(&trace_types_lock);
 457         for (t = &trace_types; *t; t = &(*t)->next) {
 458                 if (*t == type)
 459                         goto found;
 460         }
 461         pr_info("Trace %s not registered\n", type->name);
 462         goto out;
 463
 464  found:
 465         *t = (*t)->next;
 466         if (strlen(type->name) != max_tracer_type_len)
 467                 goto out;
 468
 469         max_tracer_type_len = 0;
 470         for (t = &trace_types; *t; t = &(*t)->next) {
 471                 len = strlen((*t)->name);
 472                 if (len > max_tracer_type_len)
 473                         max_tracer_type_len = len;
 474         }
 475  out:
 476         mutex_unlock(&trace_types_lock);
 477 }
 478
 479 notrace void tracing_reset(struct trace_array_cpu *data)
 480 {
 481         data->trace_idx = 0;
 482         data->trace_head = data->trace_tail = head_page(data);
 483         data->trace_head_idx = 0;
 484         data->trace_tail_idx = 0;
 485 }
 486
 487 #ifdef CONFIG_FTRACE
 488 static notrace void
 489 function_trace_call(unsigned long ip, unsigned long parent_ip)
 490 {
 491         struct trace_array *tr = &global_trace;
 492         struct trace_array_cpu *data;
 493         unsigned long flags;
 494         long disabled;
 495         int cpu;
 496
 497         if (unlikely(!tracer_enabled))
 498                 return;
 499
 500         local_irq_save(flags);
 501         cpu = raw_smp_processor_id();
 502         data = tr->data[cpu];
 503         disabled = atomic_inc_return(&data->disabled);
 504
 505         if (likely(disabled == 1))
 506                 ftrace(tr, data, ip, parent_ip, flags);
 507
 508         atomic_dec(&data->disabled);
 509         local_irq_restore(flags);
 510 }
 511
 512 static struct ftrace_ops trace_ops __read_mostly =
 513 {
 514         .func = function_trace_call,
 515 };
 516 #endif
 517
 518 notrace void tracing_start_function_trace(void)
 519 {
 520         register_ftrace_function(&trace_ops);
 521 }
 522
 523 notrace void tracing_stop_function_trace(void)
 524 {
 525         unregister_ftrace_function(&trace_ops);
 526 }
 527
 528 #define SAVED_CMDLINES 128
 529 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 530 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 531 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 532 static int cmdline_idx;
 533 static DEFINE_SPINLOCK(trace_cmdline_lock);
 534 atomic_t trace_record_cmdline_disabled;
 535
 536 static void trace_init_cmdlines(void)
 537 {
 538         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 539         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 540         cmdline_idx = 0;
 541 }
 542
 543 notrace void trace_stop_cmdline_recording(void);
 544
 545 static notrace void trace_save_cmdline(struct task_struct *tsk)
 546 {
 547         unsigned map;
 548         unsigned idx;
 549
 550         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 551                 return;
 552
 553         /*
 554          * It's not the end of the world if we don't get
 555          * the lock, but we also don't want to spin
 556          * nor do we want to disable interrupts,
 557          * so if we miss here, then better luck next time.
 558          */
 559         if (!spin_trylock(&trace_cmdline_lock))
 560                 return;
 561
 562         idx = map_pid_to_cmdline[tsk->pid];
 563         if (idx >= SAVED_CMDLINES) {
 564                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 565
 566                 map = map_cmdline_to_pid[idx];
 567                 if (map <= PID_MAX_DEFAULT)
 568                         map_pid_to_cmdline[map] = (unsigned)-1;
 569
 570                 map_pid_to_cmdline[tsk->pid] = idx;
 571
 572                 cmdline_idx = idx;
 573         }
 574
 575         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 576
 577         spin_unlock(&trace_cmdline_lock);
 578 }
 579
 580 static notrace char *trace_find_cmdline(int pid)
 581 {
 582         char *cmdline = "<...>";
 583         unsigned map;
 584
 585         if (!pid)
 586                 return "<idle>";
 587
 588         if (pid > PID_MAX_DEFAULT)
 589                 goto out;
 590
 591         map = map_pid_to_cmdline[pid];
 592         if (map >= SAVED_CMDLINES)
 593                 goto out;
 594
 595         cmdline = saved_cmdlines[map];
 596
 597  out:
 598         return cmdline;
 599 }
 600
 601 notrace void tracing_record_cmdline(struct task_struct *tsk)
 602 {
 603         if (atomic_read(&trace_record_cmdline_disabled))
 604                 return;
 605
 606         trace_save_cmdline(tsk);
 607 }
 608
 609 static inline notrace struct list_head *
 610 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 611 {
 612         /*
 613          * Roundrobin - but skip the head (which is not a real page):
 614          */
 615         next = next->next;
 616         if (unlikely(next == &data->trace_pages))
 617                 next = next->next;
 618         BUG_ON(next == &data->trace_pages);
 619
 620         return next;
 621 }
 622
 623 static inline notrace void *
 624 trace_next_page(struct trace_array_cpu *data, void *addr)
 625 {
 626         struct list_head *next;
 627         struct page *page;
 628
 629         page = virt_to_page(addr);
 630
 631         next = trace_next_list(data, &page->lru);
 632         page = list_entry(next, struct page, lru);
 633
 634         return page_address(page);
 635 }
 636
 637 static inline notrace struct trace_entry *
 638 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 639 {
 640         unsigned long idx, idx_next;
 641         struct trace_entry *entry;
 642
 643         data->trace_idx++;
 644         idx = data->trace_head_idx;
 645         idx_next = idx + 1;
 646
 647         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 648
 649         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 650
 651         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 652                 data->trace_head = trace_next_page(data, data->trace_head);
 653                 idx_next = 0;
 654         }
 655
 656         if (data->trace_head == data->trace_tail &&
 657             idx_next == data->trace_tail_idx) {
 658                 /* overrun */
 659                 data->trace_tail_idx++;
 660                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 661                         data->trace_tail =
 662                                 trace_next_page(data, data->trace_tail);
 663                         data->trace_tail_idx = 0;
 664                 }
 665         }
 666
 667         data->trace_head_idx = idx_next;
 668
 669         return entry;
 670 }
 671
 672 static inline notrace void
 673 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 674 {
 675         struct task_struct *tsk = current;
 676         unsigned long pc;
 677
 678         pc = preempt_count();
 679
 680         entry->preempt_count    = pc & 0xff;
 681         entry->pid              = tsk->pid;
 682         entry->t                = ftrace_now(raw_smp_processor_id());
 683         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 684                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 685                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 686                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 687 }
 688
 689 notrace void
 690 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 691        unsigned long ip, unsigned long parent_ip, 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_FN;
 699         entry->fn.ip            = ip;
 700         entry->fn.parent_ip     = parent_ip;
 701         spin_unlock(&data->lock);
 702 }
 703
 704 notrace void
 705 trace_special(struct trace_array *tr, struct trace_array_cpu *data,
 706               unsigned long arg1, unsigned long arg2, unsigned long arg3)
 707 {
 708         struct trace_entry *entry;
 709
 710         spin_lock(&data->lock);
 711         entry                   = tracing_get_trace_entry(tr, data);
 712         tracing_generic_entry_update(entry, 0);
 713         entry->type             = TRACE_SPECIAL;
 714         entry->special.arg1     = arg1;
 715         entry->special.arg2     = arg2;
 716         entry->special.arg3     = arg3;
 717         spin_unlock(&data->lock);
 718 }
 719
 720 notrace void
 721 tracing_sched_switch_trace(struct trace_array *tr,
 722                            struct trace_array_cpu *data,
 723                            struct task_struct *prev, struct task_struct *next,
 724                            unsigned long flags)
 725 {
 726         struct trace_entry *entry;
 727
 728         spin_lock(&data->lock);
 729         entry                   = tracing_get_trace_entry(tr, data);
 730         tracing_generic_entry_update(entry, flags);
 731         entry->type             = TRACE_CTX;
 732         entry->ctx.prev_pid     = prev->pid;
 733         entry->ctx.prev_prio    = prev->prio;
 734         entry->ctx.prev_state   = prev->state;
 735         entry->ctx.next_pid     = next->pid;
 736         entry->ctx.next_prio    = next->prio;
 737         spin_unlock(&data->lock);
 738 }
 739
 740 enum trace_file_type {
 741         TRACE_FILE_LAT_FMT      = 1,
 742 };
 743
 744 static struct trace_entry *
 745 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 746                 struct trace_iterator *iter, int cpu)
 747 {
 748         struct page *page;
 749         struct trace_entry *array;
 750
 751         if (iter->next_idx[cpu] >= tr->entries ||
 752             iter->next_idx[cpu] >= data->trace_idx ||
 753             (data->trace_head == data->trace_tail &&
 754              data->trace_head_idx == data->trace_tail_idx))
 755                 return NULL;
 756
 757         if (!iter->next_page[cpu]) {
 758                 /* Initialize the iterator for this cpu trace buffer */
 759                 WARN_ON(!data->trace_tail);
 760                 page = virt_to_page(data->trace_tail);
 761                 iter->next_page[cpu] = &page->lru;
 762                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 763         }
 764
 765         page = list_entry(iter->next_page[cpu], struct page, lru);
 766         BUG_ON(&data->trace_pages == &page->lru);
 767
 768         array = page_address(page);
 769
 770         /* Still possible to catch up to the tail */
 771         if (iter->next_idx[cpu] && array == data->trace_tail &&
 772             iter->next_page_idx[cpu] == data->trace_tail_idx)
 773                 return NULL;
 774
 775         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 776         return &array[iter->next_page_idx[cpu]];
 777 }
 778
 779 static struct notrace trace_entry *
 780 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 781 {
 782         struct trace_array *tr = iter->tr;
 783         struct trace_entry *ent, *next = NULL;
 784         int next_cpu = -1;
 785         int cpu;
 786
 787         for_each_possible_cpu(cpu) {
 788                 if (!head_page(tr->data[cpu]))
 789                         continue;
 790                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 791                 /*
 792                  * Pick the entry with the smallest timestamp:
 793                  */
 794                 if (ent && (!next || ent->t < next->t)) {
 795                         next = ent;
 796                         next_cpu = cpu;
 797                 }
 798         }
 799
 800         if (ent_cpu)
 801                 *ent_cpu = next_cpu;
 802
 803         return next;
 804 }
 805
 806 static notrace void trace_iterator_increment(struct trace_iterator *iter)
 807 {
 808         iter->idx++;
 809         iter->next_idx[iter->cpu]++;
 810         iter->next_page_idx[iter->cpu]++;
 811
 812         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 813                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 814
 815                 iter->next_page_idx[iter->cpu] = 0;
 816                 iter->next_page[iter->cpu] =
 817                         trace_next_list(data, iter->next_page[iter->cpu]);
 818         }
 819 }
 820
 821 static notrace void trace_consume(struct trace_iterator *iter)
 822 {
 823         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 824
 825         data->trace_tail_idx++;
 826         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 827                 data->trace_tail = trace_next_page(data, data->trace_tail);
 828                 data->trace_tail_idx = 0;
 829         }
 830
 831         /* Check if we empty it, then reset the index */
 832         if (data->trace_head == data->trace_tail &&
 833             data->trace_head_idx == data->trace_tail_idx)
 834                 data->trace_idx = 0;
 835 }
 836
 837 static notrace void *find_next_entry_inc(struct trace_iterator *iter)
 838 {
 839         struct trace_entry *next;
 840         int next_cpu = -1;
 841
 842         next = find_next_entry(iter, &next_cpu);
 843
 844         iter->prev_ent = iter->ent;
 845         iter->prev_cpu = iter->cpu;
 846
 847         iter->ent = next;
 848         iter->cpu = next_cpu;
 849
 850         if (next)
 851                 trace_iterator_increment(iter);
 852
 853         return next ? iter : NULL;
 854 }
 855
 856 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 857 {
 858         struct trace_iterator *iter = m->private;
 859         void *last_ent = iter->ent;
 860         int i = (int)*pos;
 861         void *ent;
 862
 863         (*pos)++;
 864
 865         /* can't go backwards */
 866         if (iter->idx > i)
 867                 return NULL;
 868
 869         if (iter->idx < 0)
 870                 ent = find_next_entry_inc(iter);
 871         else
 872                 ent = iter;
 873
 874         while (ent && iter->idx < i)
 875                 ent = find_next_entry_inc(iter);
 876
 877         iter->pos = *pos;
 878
 879         if (last_ent && !ent)
 880                 seq_puts(m, "\n\nvim:ft=help\n");
 881
 882         return ent;
 883 }
 884
 885 static void *s_start(struct seq_file *m, loff_t *pos)
 886 {
 887         struct trace_iterator *iter = m->private;
 888         void *p = NULL;
 889         loff_t l = 0;
 890         int i;
 891
 892         mutex_lock(&trace_types_lock);
 893
 894         if (!current_trace || current_trace != iter->trace)
 895                 return NULL;
 896
 897         atomic_inc(&trace_record_cmdline_disabled);
 898
 899         /* let the tracer grab locks here if needed */
 900         if (current_trace->start)
 901                 current_trace->start(iter);
 902
 903         if (*pos != iter->pos) {
 904                 iter->ent = NULL;
 905                 iter->cpu = 0;
 906                 iter->idx = -1;
 907                 iter->prev_ent = NULL;
 908                 iter->prev_cpu = -1;
 909
 910                 for_each_possible_cpu(i) {
 911                         iter->next_idx[i] = 0;
 912                         iter->next_page[i] = NULL;
 913                 }
 914
 915                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 916                         ;
 917
 918         } else {
 919                 l = *pos - 1;
 920                 p = s_next(m, p, &l);
 921         }
 922
 923         return p;
 924 }
 925
 926 static void s_stop(struct seq_file *m, void *p)
 927 {
 928         struct trace_iterator *iter = m->private;
 929
 930         atomic_dec(&trace_record_cmdline_disabled);
 931
 932         /* let the tracer release locks here if needed */
 933         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 934                 iter->trace->stop(iter);
 935
 936         mutex_unlock(&trace_types_lock);
 937 }
 938
 939 static int
 940 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 941 {
 942 #ifdef CONFIG_KALLSYMS
 943         char str[KSYM_SYMBOL_LEN];
 944
 945         kallsyms_lookup(address, NULL, NULL, NULL, str);
 946
 947         return trace_seq_printf(s, fmt, str);
 948 #endif
 949         return 1;
 950 }
 951
 952 static int
 953 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 954                      unsigned long address)
 955 {
 956 #ifdef CONFIG_KALLSYMS
 957         char str[KSYM_SYMBOL_LEN];
 958
 959         sprint_symbol(str, address);
 960         return trace_seq_printf(s, fmt, str);
 961 #endif
 962         return 1;
 963 }
 964
 965 #ifndef CONFIG_64BIT
 966 # define IP_FMT "%08lx"
 967 #else
 968 # define IP_FMT "%016lx"
 969 #endif
 970
 971 static notrace int
 972 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 973 {
 974         int ret;
 975
 976         if (!ip)
 977                 return trace_seq_printf(s, "0");
 978
 979         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 980                 ret = seq_print_sym_offset(s, "%s", ip);
 981         else
 982                 ret = seq_print_sym_short(s, "%s", ip);
 983
 984         if (!ret)
 985                 return 0;
 986
 987         if (sym_flags & TRACE_ITER_SYM_ADDR)
 988                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 989         return ret;
 990 }
 991
 992 static notrace void print_lat_help_header(struct seq_file *m)
 993 {
 994         seq_puts(m, "#                _------=> CPU#            \n");
 995         seq_puts(m, "#               / _-----=> irqs-off        \n");
 996         seq_puts(m, "#              | / _----=> need-resched    \n");
 997         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 998         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 999         seq_puts(m, "#              |||| /                      \n");
1000         seq_puts(m, "#              |||||     delay             \n");
1001         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1002         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1003 }
1004
1005 static notrace void print_func_help_header(struct seq_file *m)
1006 {
1007         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1008         seq_puts(m, "#              | |      |          |         |\n");
1009 }
1010
1011
1012 static notrace void
1013 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1014 {
1015         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1016         struct trace_array *tr = iter->tr;
1017         struct trace_array_cpu *data = tr->data[tr->cpu];
1018         struct tracer *type = current_trace;
1019         unsigned long total   = 0;
1020         unsigned long entries = 0;
1021         int cpu;
1022         const char *name = "preemption";
1023
1024         if (type)
1025                 name = type->name;
1026
1027         for_each_possible_cpu(cpu) {
1028                 if (head_page(tr->data[cpu])) {
1029                         total += tr->data[cpu]->trace_idx;
1030                         if (tr->data[cpu]->trace_idx > tr->entries)
1031                                 entries += tr->entries;
1032                         else
1033                                 entries += tr->data[cpu]->trace_idx;
1034                 }
1035         }
1036
1037         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1038                    name, UTS_RELEASE);
1039         seq_puts(m, "-----------------------------------"
1040                  "---------------------------------\n");
1041         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1042                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1043                    nsecs_to_usecs(data->saved_latency),
1044                    entries,
1045                    total,
1046                    tr->cpu,
1047 #if defined(CONFIG_PREEMPT_NONE)
1048                    "server",
1049 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1050                    "desktop",
1051 #elif defined(CONFIG_PREEMPT_DESKTOP)
1052                    "preempt",
1053 #else
1054                    "unknown",
1055 #endif
1056                    /* These are reserved for later use */
1057                    0, 0, 0, 0);
1058 #ifdef CONFIG_SMP
1059         seq_printf(m, " #P:%d)\n", num_online_cpus());
1060 #else
1061         seq_puts(m, ")\n");
1062 #endif
1063         seq_puts(m, "    -----------------\n");
1064         seq_printf(m, "    | task: %.16s-%d "
1065                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1066                    data->comm, data->pid, data->uid, data->nice,
1067                    data->policy, data->rt_priority);
1068         seq_puts(m, "    -----------------\n");
1069
1070         if (data->critical_start) {
1071                 seq_puts(m, " => started at: ");
1072                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1073                 trace_print_seq(m, &iter->seq);
1074                 seq_puts(m, "\n => ended at:   ");
1075                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1076                 trace_print_seq(m, &iter->seq);
1077                 seq_puts(m, "\n");
1078         }
1079
1080         seq_puts(m, "\n");
1081 }
1082
1083 static notrace void
1084 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1085 {
1086         int hardirq, softirq;
1087         char *comm;
1088
1089         comm = trace_find_cmdline(entry->pid);
1090
1091         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1092         trace_seq_printf(s, "%d", cpu);
1093         trace_seq_printf(s, "%c%c",
1094                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1095                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1096
1097         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1098         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1099         if (hardirq && softirq)
1100                 trace_seq_putc(s, 'H');
1101         else {
1102                 if (hardirq)
1103                         trace_seq_putc(s, 'h');
1104                 else {
1105                         if (softirq)
1106                                 trace_seq_putc(s, 's');
1107                         else
1108                                 trace_seq_putc(s, '.');
1109                 }
1110         }
1111
1112         if (entry->preempt_count)
1113                 trace_seq_printf(s, "%x", entry->preempt_count);
1114         else
1115                 trace_seq_puts(s, ".");
1116 }
1117
1118 unsigned long preempt_mark_thresh = 100;
1119
1120 static notrace void
1121 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1122                     unsigned long rel_usecs)
1123 {
1124         trace_seq_printf(s, " %4lldus", abs_usecs);
1125         if (rel_usecs > preempt_mark_thresh)
1126                 trace_seq_puts(s, "!: ");
1127         else if (rel_usecs > 1)
1128                 trace_seq_puts(s, "+: ");
1129         else
1130                 trace_seq_puts(s, " : ");
1131 }
1132
1133 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1134
1135 static notrace int
1136 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1137 {
1138         struct trace_seq *s = &iter->seq;
1139         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1140         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1141         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1142         struct trace_entry *entry = iter->ent;
1143         unsigned long abs_usecs;
1144         unsigned long rel_usecs;
1145         char *comm;
1146         int S;
1147
1148         if (!next_entry)
1149                 next_entry = entry;
1150         rel_usecs = ns2usecs(next_entry->t - entry->t);
1151         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1152
1153         if (verbose) {
1154                 comm = trace_find_cmdline(entry->pid);
1155                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1156                                  " %ld.%03ldms (+%ld.%03ldms): ",
1157                                  comm,
1158                                  entry->pid, cpu, entry->flags,
1159                                  entry->preempt_count, trace_idx,
1160                                  ns2usecs(entry->t),
1161                                  abs_usecs/1000,
1162                                  abs_usecs % 1000, rel_usecs/1000,
1163                                  rel_usecs % 1000);
1164         } else {
1165                 lat_print_generic(s, entry, cpu);
1166                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1167         }
1168         switch (entry->type) {
1169         case TRACE_FN:
1170                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1171                 trace_seq_puts(s, " (");
1172                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1173                 trace_seq_puts(s, ")\n");
1174                 break;
1175         case TRACE_CTX:
1176                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1177                         state_to_char[entry->ctx.prev_state] : 'X';
1178                 comm = trace_find_cmdline(entry->ctx.next_pid);
1179                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1180                                  entry->ctx.prev_pid,
1181                                  entry->ctx.prev_prio,
1182                                  S,
1183                                  entry->ctx.next_pid,
1184                                  entry->ctx.next_prio,
1185                                  comm);
1186                 break;
1187         case TRACE_SPECIAL:
1188                 trace_seq_printf(s, " %lx %lx %lx\n",
1189                                  entry->special.arg1,
1190                                  entry->special.arg2,
1191                                  entry->special.arg3);
1192                 break;
1193         default:
1194                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1195         }
1196         return 1;
1197 }
1198
1199 static notrace int print_trace_fmt(struct trace_iterator *iter)
1200 {
1201         struct trace_seq *s = &iter->seq;
1202         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1203         struct trace_entry *entry;
1204         unsigned long usec_rem;
1205         unsigned long long t;
1206         unsigned long secs;
1207         char *comm;
1208         int S;
1209         int ret;
1210
1211         entry = iter->ent;
1212
1213         comm = trace_find_cmdline(iter->ent->pid);
1214
1215         t = ns2usecs(entry->t);
1216         usec_rem = do_div(t, 1000000ULL);
1217         secs = (unsigned long)t;
1218
1219         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1220         if (!ret)
1221                 return 0;
1222         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1223         if (!ret)
1224                 return 0;
1225         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1226         if (!ret)
1227                 return 0;
1228
1229         switch (entry->type) {
1230         case TRACE_FN:
1231                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1232                 if (!ret)
1233                         return 0;
1234                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1235                                                 entry->fn.parent_ip) {
1236                         ret = trace_seq_printf(s, " <-");
1237                         if (!ret)
1238                                 return 0;
1239                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1240                                                sym_flags);
1241                         if (!ret)
1242                                 return 0;
1243                 }
1244                 ret = trace_seq_printf(s, "\n");
1245                 if (!ret)
1246                         return 0;
1247                 break;
1248         case TRACE_CTX:
1249                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1250                         state_to_char[entry->ctx.prev_state] : 'X';
1251                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1252                                        entry->ctx.prev_pid,
1253                                        entry->ctx.prev_prio,
1254                                        S,
1255                                        entry->ctx.next_pid,
1256                                        entry->ctx.next_prio);
1257                 if (!ret)
1258                         return 0;
1259                 break;
1260         case TRACE_SPECIAL:
1261                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1262                                  entry->special.arg1,
1263                                  entry->special.arg2,
1264                                  entry->special.arg3);
1265                 if (!ret)
1266                         return 0;
1267                 break;
1268         }
1269         return 1;
1270 }
1271
1272 static notrace int print_raw_fmt(struct trace_iterator *iter)
1273 {
1274         struct trace_seq *s = &iter->seq;
1275         struct trace_entry *entry;
1276         int ret;
1277         int S;
1278
1279         entry = iter->ent;
1280
1281         ret = trace_seq_printf(s, "%d %d %llu ",
1282                 entry->pid, iter->cpu, entry->t);
1283         if (!ret)
1284                 return 0;
1285
1286         switch (entry->type) {
1287         case TRACE_FN:
1288                 ret = trace_seq_printf(s, "%x %x\n",
1289                                         entry->fn.ip, entry->fn.parent_ip);
1290                 if (!ret)
1291                         return 0;
1292                 break;
1293         case TRACE_CTX:
1294                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1295                         state_to_char[entry->ctx.prev_state] : 'X';
1296                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1297                                        entry->ctx.prev_pid,
1298                                        entry->ctx.prev_prio,
1299                                        S,
1300                                        entry->ctx.next_pid,
1301                                        entry->ctx.next_prio);
1302                 if (!ret)
1303                         return 0;
1304                 break;
1305         case TRACE_SPECIAL:
1306                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1307                                  entry->special.arg1,
1308                                  entry->special.arg2,
1309                                  entry->special.arg3);
1310                 if (!ret)
1311                         return 0;
1312                 break;
1313         }
1314         return 1;
1315 }
1316
1317 #define SEQ_PUT_FIELD_RET(s, x)                         \
1318 do {                                                    \
1319         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1320                 return 0;                               \
1321 } while (0)
1322
1323 static notrace int print_bin_fmt(struct trace_iterator *iter)
1324 {
1325         struct trace_seq *s = &iter->seq;
1326         struct trace_entry *entry;
1327
1328         entry = iter->ent;
1329
1330         SEQ_PUT_FIELD_RET(s, entry->pid);
1331         SEQ_PUT_FIELD_RET(s, entry->cpu);
1332         SEQ_PUT_FIELD_RET(s, entry->t);
1333
1334         switch (entry->type) {
1335         case TRACE_FN:
1336                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1337                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1338                 break;
1339         case TRACE_CTX:
1340                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1341                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1342                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1343                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1344                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1345                 break;
1346         case TRACE_SPECIAL:
1347                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1348                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1349                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1350                 break;
1351         }
1352         return 1;
1353 }
1354
1355 static int trace_empty(struct trace_iterator *iter)
1356 {
1357         struct trace_array_cpu *data;
1358         int cpu;
1359
1360         for_each_possible_cpu(cpu) {
1361                 data = iter->tr->data[cpu];
1362
1363                 if (head_page(data) && data->trace_idx &&
1364                     (data->trace_tail != data->trace_head ||
1365                      data->trace_tail_idx != data->trace_head_idx))
1366                         return 0;
1367         }
1368         return 1;
1369 }
1370
1371 static int print_trace_line(struct trace_iterator *iter)
1372 {
1373         if (trace_flags & TRACE_ITER_BIN)
1374                 return print_bin_fmt(iter);
1375
1376         if (trace_flags & TRACE_ITER_RAW)
1377                 return print_raw_fmt(iter);
1378
1379         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1380                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1381
1382         return print_trace_fmt(iter);
1383 }
1384
1385 static int s_show(struct seq_file *m, void *v)
1386 {
1387         struct trace_iterator *iter = v;
1388
1389         if (iter->ent == NULL) {
1390                 if (iter->tr) {
1391                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1392                         seq_puts(m, "#\n");
1393                 }
1394                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1395                         /* print nothing if the buffers are empty */
1396                         if (trace_empty(iter))
1397                                 return 0;
1398                         print_trace_header(m, iter);
1399                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1400                                 print_lat_help_header(m);
1401                 } else {
1402                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1403                                 print_func_help_header(m);
1404                 }
1405         } else {
1406                 print_trace_line(iter);
1407                 trace_print_seq(m, &iter->seq);
1408         }
1409
1410         return 0;
1411 }
1412
1413 static struct seq_operations tracer_seq_ops = {
1414         .start          = s_start,
1415         .next           = s_next,
1416         .stop           = s_stop,
1417         .show           = s_show,
1418 };
1419
1420 static struct trace_iterator notrace *
1421 __tracing_open(struct inode *inode, struct file *file, int *ret)
1422 {
1423         struct trace_iterator *iter;
1424
1425         if (tracing_disabled) {
1426                 *ret = -ENODEV;
1427                 return NULL;
1428         }
1429
1430         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1431         if (!iter) {
1432                 *ret = -ENOMEM;
1433                 goto out;
1434         }
1435
1436         mutex_lock(&trace_types_lock);
1437         if (current_trace && current_trace->print_max)
1438                 iter->tr = &max_tr;
1439         else
1440                 iter->tr = inode->i_private;
1441         iter->trace = current_trace;
1442         iter->pos = -1;
1443
1444         /* TODO stop tracer */
1445         *ret = seq_open(file, &tracer_seq_ops);
1446         if (!*ret) {
1447                 struct seq_file *m = file->private_data;
1448                 m->private = iter;
1449
1450                 /* stop the trace while dumping */
1451                 if (iter->tr->ctrl)
1452                         tracer_enabled = 0;
1453
1454                 if (iter->trace && iter->trace->open)
1455                         iter->trace->open(iter);
1456         } else {
1457                 kfree(iter);
1458                 iter = NULL;
1459         }
1460         mutex_unlock(&trace_types_lock);
1461
1462  out:
1463         return iter;
1464 }
1465
1466 int tracing_open_generic(struct inode *inode, struct file *filp)
1467 {
1468         if (tracing_disabled)
1469                 return -ENODEV;
1470
1471         filp->private_data = inode->i_private;
1472         return 0;
1473 }
1474
1475 int tracing_release(struct inode *inode, struct file *file)
1476 {
1477         struct seq_file *m = (struct seq_file *)file->private_data;
1478         struct trace_iterator *iter = m->private;
1479
1480         mutex_lock(&trace_types_lock);
1481         if (iter->trace && iter->trace->close)
1482                 iter->trace->close(iter);
1483
1484         /* reenable tracing if it was previously enabled */
1485         if (iter->tr->ctrl)
1486                 tracer_enabled = 1;
1487         mutex_unlock(&trace_types_lock);
1488
1489         seq_release(inode, file);
1490         kfree(iter);
1491         return 0;
1492 }
1493
1494 static int tracing_open(struct inode *inode, struct file *file)
1495 {
1496         int ret;
1497
1498         __tracing_open(inode, file, &ret);
1499
1500         return ret;
1501 }
1502
1503 static int tracing_lt_open(struct inode *inode, struct file *file)
1504 {
1505         struct trace_iterator *iter;
1506         int ret;
1507
1508         iter = __tracing_open(inode, file, &ret);
1509
1510         if (!ret)
1511                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1512
1513         return ret;
1514 }
1515
1516
1517 static notrace void *
1518 t_next(struct seq_file *m, void *v, loff_t *pos)
1519 {
1520         struct tracer *t = m->private;
1521
1522         (*pos)++;
1523
1524         if (t)
1525                 t = t->next;
1526
1527         m->private = t;
1528
1529         return t;
1530 }
1531
1532 static void *t_start(struct seq_file *m, loff_t *pos)
1533 {
1534         struct tracer *t = m->private;
1535         loff_t l = 0;
1536
1537         mutex_lock(&trace_types_lock);
1538         for (; t && l < *pos; t = t_next(m, t, &l))
1539                 ;
1540
1541         return t;
1542 }
1543
1544 static void t_stop(struct seq_file *m, void *p)
1545 {
1546         mutex_unlock(&trace_types_lock);
1547 }
1548
1549 static int t_show(struct seq_file *m, void *v)
1550 {
1551         struct tracer *t = v;
1552
1553         if (!t)
1554                 return 0;
1555
1556         seq_printf(m, "%s", t->name);
1557         if (t->next)
1558                 seq_putc(m, ' ');
1559         else
1560                 seq_putc(m, '\n');
1561
1562         return 0;
1563 }
1564
1565 static struct seq_operations show_traces_seq_ops = {
1566         .start          = t_start,
1567         .next           = t_next,
1568         .stop           = t_stop,
1569         .show           = t_show,
1570 };
1571
1572 static int show_traces_open(struct inode *inode, struct file *file)
1573 {
1574         int ret;
1575
1576         if (tracing_disabled)
1577                 return -ENODEV;
1578
1579         ret = seq_open(file, &show_traces_seq_ops);
1580         if (!ret) {
1581                 struct seq_file *m = file->private_data;
1582                 m->private = trace_types;
1583         }
1584
1585         return ret;
1586 }
1587
1588 static struct file_operations tracing_fops = {
1589         .open           = tracing_open,
1590         .read           = seq_read,
1591         .llseek         = seq_lseek,
1592         .release        = tracing_release,
1593 };
1594
1595 static struct file_operations tracing_lt_fops = {
1596         .open           = tracing_lt_open,
1597         .read           = seq_read,
1598         .llseek         = seq_lseek,
1599         .release        = tracing_release,
1600 };
1601
1602 static struct file_operations show_traces_fops = {
1603         .open = show_traces_open,
1604         .read = seq_read,
1605         .release = seq_release,
1606 };
1607
1608 static ssize_t
1609 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1610                        size_t cnt, loff_t *ppos)
1611 {
1612         char *buf;
1613         int r = 0;
1614         int len = 0;
1615         int i;
1616
1617         /* calulate max size */
1618         for (i = 0; trace_options[i]; i++) {
1619                 len += strlen(trace_options[i]);
1620                 len += 3; /* "no" and space */
1621         }
1622
1623         /* +2 for \n and \0 */
1624         buf = kmalloc(len + 2, GFP_KERNEL);
1625         if (!buf)
1626                 return -ENOMEM;
1627
1628         for (i = 0; trace_options[i]; i++) {
1629                 if (trace_flags & (1 << i))
1630                         r += sprintf(buf + r, "%s ", trace_options[i]);
1631                 else
1632                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1633         }
1634
1635         r += sprintf(buf + r, "\n");
1636         WARN_ON(r >= len + 2);
1637
1638         r = simple_read_from_buffer(ubuf, cnt, ppos,
1639                                     buf, r);
1640
1641         kfree(buf);
1642
1643         return r;
1644 }
1645
1646 static ssize_t
1647 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1648                         size_t cnt, loff_t *ppos)
1649 {
1650         char buf[64];
1651         char *cmp = buf;
1652         int neg = 0;
1653         int i;
1654
1655         if (cnt > 63)
1656                 cnt = 63;
1657
1658         if (copy_from_user(&buf, ubuf, cnt))
1659                 return -EFAULT;
1660
1661         buf[cnt] = 0;
1662
1663         if (strncmp(buf, "no", 2) == 0) {
1664                 neg = 1;
1665                 cmp += 2;
1666         }
1667
1668         for (i = 0; trace_options[i]; i++) {
1669                 int len = strlen(trace_options[i]);
1670
1671                 if (strncmp(cmp, trace_options[i], len) == 0) {
1672                         if (neg)
1673                                 trace_flags &= ~(1 << i);
1674                         else
1675                                 trace_flags |= (1 << i);
1676                         break;
1677                 }
1678         }
1679
1680         filp->f_pos += cnt;
1681
1682         return cnt;
1683 }
1684
1685 static struct file_operations tracing_iter_fops = {
1686         .open = tracing_open_generic,
1687         .read = tracing_iter_ctrl_read,
1688         .write = tracing_iter_ctrl_write,
1689 };
1690
1691 static const char readme_msg[] =
1692         "tracing mini-HOWTO:\n\n"
1693         "# mkdir /debug\n"
1694         "# mount -t debugfs nodev /debug\n\n"
1695         "# cat /debug/tracing/available_tracers\n"
1696         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1697         "# cat /debug/tracing/current_tracer\n"
1698         "none\n"
1699         "# echo sched_switch > /debug/tracing/current_tracer\n"
1700         "# cat /debug/tracing/current_tracer\n"
1701         "sched_switch\n"
1702         "# cat /debug/tracing/iter_ctrl\n"
1703         "noprint-parent nosym-offset nosym-addr noverbose\n"
1704         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1705         "# echo 1 > /debug/tracing/tracing_enabled\n"
1706         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1707         "echo 0 > /debug/tracing/tracing_enabled\n"
1708 ;
1709
1710 static ssize_t
1711 tracing_readme_read(struct file *filp, char __user *ubuf,
1712                        size_t cnt, loff_t *ppos)
1713 {
1714         return simple_read_from_buffer(ubuf, cnt, ppos,
1715                                         readme_msg, strlen(readme_msg));
1716 }
1717
1718 static struct file_operations tracing_readme_fops = {
1719         .open = tracing_open_generic,
1720         .read = tracing_readme_read,
1721 };
1722
1723
1724 static ssize_t
1725 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1726                   size_t cnt, loff_t *ppos)
1727 {
1728         struct trace_array *tr = filp->private_data;
1729         char buf[64];
1730         int r;
1731
1732         r = sprintf(buf, "%ld\n", tr->ctrl);
1733         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1734 }
1735
1736 static ssize_t
1737 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1738                    size_t cnt, loff_t *ppos)
1739 {
1740         struct trace_array *tr = filp->private_data;
1741         long val;
1742         char buf[64];
1743
1744         if (cnt > 63)
1745                 cnt = 63;
1746
1747         if (copy_from_user(&buf, ubuf, cnt))
1748                 return -EFAULT;
1749
1750         buf[cnt] = 0;
1751
1752         val = simple_strtoul(buf, NULL, 10);
1753
1754         val = !!val;
1755
1756         mutex_lock(&trace_types_lock);
1757         if (tr->ctrl ^ val) {
1758                 if (val)
1759                         tracer_enabled = 1;
1760                 else
1761                         tracer_enabled = 0;
1762
1763                 tr->ctrl = val;
1764
1765                 if (current_trace && current_trace->ctrl_update)
1766                         current_trace->ctrl_update(tr);
1767         }
1768         mutex_unlock(&trace_types_lock);
1769
1770         filp->f_pos += cnt;
1771
1772         return cnt;
1773 }
1774
1775 static ssize_t
1776 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1777                        size_t cnt, loff_t *ppos)
1778 {
1779         char buf[max_tracer_type_len+2];
1780         int r;
1781
1782         mutex_lock(&trace_types_lock);
1783         if (current_trace)
1784                 r = sprintf(buf, "%s\n", current_trace->name);
1785         else
1786                 r = sprintf(buf, "\n");
1787         mutex_unlock(&trace_types_lock);
1788
1789         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1790 }
1791
1792 static ssize_t
1793 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1794                         size_t cnt, loff_t *ppos)
1795 {
1796         struct trace_array *tr = &global_trace;
1797         struct tracer *t;
1798         char buf[max_tracer_type_len+1];
1799         int i;
1800
1801         if (cnt > max_tracer_type_len)
1802                 cnt = max_tracer_type_len;
1803
1804         if (copy_from_user(&buf, ubuf, cnt))
1805                 return -EFAULT;
1806
1807         buf[cnt] = 0;
1808
1809         /* strip ending whitespace. */
1810         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1811                 buf[i] = 0;
1812
1813         mutex_lock(&trace_types_lock);
1814         for (t = trace_types; t; t = t->next) {
1815                 if (strcmp(t->name, buf) == 0)
1816                         break;
1817         }
1818         if (!t || t == current_trace)
1819                 goto out;
1820
1821         if (current_trace && current_trace->reset)
1822                 current_trace->reset(tr);
1823
1824         current_trace = t;
1825         if (t->init)
1826                 t->init(tr);
1827
1828  out:
1829         mutex_unlock(&trace_types_lock);
1830
1831         filp->f_pos += cnt;
1832
1833         return cnt;
1834 }
1835
1836 static ssize_t
1837 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1838                      size_t cnt, loff_t *ppos)
1839 {
1840         unsigned long *ptr = filp->private_data;
1841         char buf[64];
1842         int r;
1843
1844         r = snprintf(buf, 64, "%ld\n",
1845                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1846         if (r > 64)
1847                 r = 64;
1848         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1849 }
1850
1851 static ssize_t
1852 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1853                       size_t cnt, loff_t *ppos)
1854 {
1855         long *ptr = filp->private_data;
1856         long val;
1857         char buf[64];
1858
1859         if (cnt > 63)
1860                 cnt = 63;
1861
1862         if (copy_from_user(&buf, ubuf, cnt))
1863                 return -EFAULT;
1864
1865         buf[cnt] = 0;
1866
1867         val = simple_strtoul(buf, NULL, 10);
1868
1869         *ptr = val * 1000;
1870
1871         return cnt;
1872 }
1873
1874 static atomic_t tracing_reader;
1875
1876 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1877 {
1878         struct trace_iterator *iter;
1879
1880         if (tracing_disabled)
1881                 return -ENODEV;
1882
1883         /* We only allow for reader of the pipe */
1884         if (atomic_inc_return(&tracing_reader) != 1) {
1885                 atomic_dec(&tracing_reader);
1886                 return -EBUSY;
1887         }
1888
1889         /* create a buffer to store the information to pass to userspace */
1890         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1891         if (!iter)
1892                 return -ENOMEM;
1893
1894         iter->tr = &global_trace;
1895
1896         filp->private_data = iter;
1897
1898         return 0;
1899 }
1900
1901 static int tracing_release_pipe(struct inode *inode, struct file *file)
1902 {
1903         struct trace_iterator *iter = file->private_data;
1904
1905         kfree(iter);
1906         atomic_dec(&tracing_reader);
1907
1908         return 0;
1909 }
1910
1911 /*
1912  * Consumer reader.
1913  */
1914 static ssize_t
1915 tracing_read_pipe(struct file *filp, char __user *ubuf,
1916                   size_t cnt, loff_t *ppos)
1917 {
1918         struct trace_iterator *iter = filp->private_data;
1919         struct trace_array_cpu *data;
1920         static cpumask_t mask;
1921         struct trace_entry *entry;
1922         static int start;
1923         unsigned long flags;
1924         int read = 0;
1925         int cpu;
1926         int len;
1927         int ret;
1928
1929         /* return any leftover data */
1930         if (iter->seq.len > start) {
1931                 len = iter->seq.len - start;
1932                 if (cnt > len)
1933                         cnt = len;
1934                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1935                 if (ret)
1936                         cnt = -EFAULT;
1937
1938                 start += len;
1939
1940                 return cnt;
1941         }
1942
1943         trace_seq_reset(&iter->seq);
1944         start = 0;
1945
1946         while (trace_empty(iter)) {
1947                 /*
1948                  * This is a make-shift waitqueue. The reason we don't use
1949                  * an actual wait queue is because:
1950                  *  1) we only ever have one waiter
1951                  *  2) the tracing, traces all functions, we don't want
1952                  *     the overhead of calling wake_up and friends
1953                  *     (and tracing them too)
1954                  *     Anyway, this is really very primitive wakeup.
1955                  */
1956                 set_current_state(TASK_INTERRUPTIBLE);
1957                 iter->tr->waiter = current;
1958
1959                 /* sleep for one second, and try again. */
1960                 schedule_timeout(HZ);
1961
1962                 iter->tr->waiter = NULL;
1963
1964                 if (signal_pending(current))
1965                         return -EINTR;
1966
1967                 /*
1968                  * We block until we read something and tracing is disabled.
1969                  * We still block if tracing is disabled, but we have never
1970                  * read anything. This allows a user to cat this file, and
1971                  * then enable tracing. But after we have read something,
1972                  * we give an EOF when tracing is again disabled.
1973                  *
1974                  * iter->pos will be 0 if we haven't read anything.
1975                  */
1976                 if (!tracer_enabled && iter->pos)
1977                         break;
1978
1979                 continue;
1980         }
1981
1982         /* stop when tracing is finished */
1983         if (trace_empty(iter))
1984                 return 0;
1985
1986         if (cnt >= PAGE_SIZE)
1987                 cnt = PAGE_SIZE - 1;
1988
1989         memset(iter, 0, sizeof(*iter));
1990         iter->tr = &global_trace;
1991         iter->pos = -1;
1992
1993         /*
1994          * We need to stop all tracing on all CPUS to read the
1995          * the next buffer. This is a bit expensive, but is
1996          * not done often. We fill all what we can read,
1997          * and then release the locks again.
1998          */
1999
2000         cpus_clear(mask);
2001         local_irq_save(flags);
2002         for_each_possible_cpu(cpu) {
2003                 data = iter->tr->data[cpu];
2004
2005                 if (!head_page(data) || !data->trace_idx)
2006                         continue;
2007
2008                 atomic_inc(&data->disabled);
2009                 spin_lock(&data->lock);
2010                 cpu_set(cpu, mask);
2011         }
2012
2013         while ((entry = find_next_entry_inc(iter)) != NULL) {
2014                 ret = print_trace_line(iter);
2015                 if (!ret)
2016                         break;
2017
2018                 trace_consume(iter);
2019
2020                 if (iter->seq.len >= cnt)
2021                         break;
2022         }
2023
2024         for_each_cpu_mask(cpu, mask) {
2025                 data = iter->tr->data[cpu];
2026                 spin_unlock(&data->lock);
2027                 atomic_dec(&data->disabled);
2028         }
2029         local_irq_restore(flags);
2030
2031         /* Now copy what we have to the user */
2032         read = iter->seq.len;
2033         if (read > cnt)
2034                 read = cnt;
2035
2036         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2037
2038         if (read < iter->seq.len)
2039                 start = read;
2040         else
2041                 trace_seq_reset(&iter->seq);
2042
2043         if (ret)
2044                 read = -EFAULT;
2045
2046         return read;
2047 }
2048
2049 static struct file_operations tracing_max_lat_fops = {
2050         .open           = tracing_open_generic,
2051         .read           = tracing_max_lat_read,
2052         .write          = tracing_max_lat_write,
2053 };
2054
2055 static struct file_operations tracing_ctrl_fops = {
2056         .open           = tracing_open_generic,
2057         .read           = tracing_ctrl_read,
2058         .write          = tracing_ctrl_write,
2059 };
2060
2061 static struct file_operations set_tracer_fops = {
2062         .open           = tracing_open_generic,
2063         .read           = tracing_set_trace_read,
2064         .write          = tracing_set_trace_write,
2065 };
2066
2067 static struct file_operations tracing_pipe_fops = {
2068         .open           = tracing_open_pipe,
2069         .read           = tracing_read_pipe,
2070         .release        = tracing_release_pipe,
2071 };
2072
2073 #ifdef CONFIG_DYNAMIC_FTRACE
2074
2075 static ssize_t
2076 tracing_read_long(struct file *filp, char __user *ubuf,
2077                   size_t cnt, loff_t *ppos)
2078 {
2079         unsigned long *p = filp->private_data;
2080         char buf[64];
2081         int r;
2082
2083         r = sprintf(buf, "%ld\n", *p);
2084
2085         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2086 }
2087
2088 static struct file_operations tracing_read_long_fops = {
2089         .open           = tracing_open_generic,
2090         .read           = tracing_read_long,
2091 };
2092 #endif
2093
2094 static struct dentry *d_tracer;
2095
2096 struct dentry *tracing_init_dentry(void)
2097 {
2098         static int once;
2099
2100         if (d_tracer)
2101                 return d_tracer;
2102
2103         d_tracer = debugfs_create_dir("tracing", NULL);
2104
2105         if (!d_tracer && !once) {
2106                 once = 1;
2107                 pr_warning("Could not create debugfs directory 'tracing'\n");
2108                 return NULL;
2109         }
2110
2111         return d_tracer;
2112 }
2113
2114 #ifdef CONFIG_FTRACE_SELFTEST
2115 /* Let selftest have access to static functions in this file */
2116 #include "trace_selftest.c"
2117 #endif
2118
2119 static __init void tracer_init_debugfs(void)
2120 {
2121         struct dentry *d_tracer;
2122         struct dentry *entry;
2123
2124         d_tracer = tracing_init_dentry();
2125
2126         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2127                                     &global_trace, &tracing_ctrl_fops);
2128         if (!entry)
2129                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2130
2131         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2132                                     NULL, &tracing_iter_fops);
2133         if (!entry)
2134                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2135
2136         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2137                                     &global_trace, &tracing_lt_fops);
2138         if (!entry)
2139                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2140
2141         entry = debugfs_create_file("trace", 0444, d_tracer,
2142                                     &global_trace, &tracing_fops);
2143         if (!entry)
2144                 pr_warning("Could not create debugfs 'trace' entry\n");
2145
2146         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2147                                     &global_trace, &show_traces_fops);
2148         if (!entry)
2149                 pr_warning("Could not create debugfs 'trace' entry\n");
2150
2151         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2152                                     &global_trace, &set_tracer_fops);
2153         if (!entry)
2154                 pr_warning("Could not create debugfs 'trace' entry\n");
2155
2156         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2157                                     &tracing_max_latency,
2158                                     &tracing_max_lat_fops);
2159         if (!entry)
2160                 pr_warning("Could not create debugfs "
2161                            "'tracing_max_latency' entry\n");
2162
2163         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2164                                     &tracing_thresh, &tracing_max_lat_fops);
2165         if (!entry)
2166                 pr_warning("Could not create debugfs "
2167                            "'tracing_threash' entry\n");
2168         entry = debugfs_create_file("README", 0644, d_tracer,
2169                                     NULL, &tracing_readme_fops);
2170         if (!entry)
2171                 pr_warning("Could not create debugfs 'README' entry\n");
2172
2173         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2174                                     NULL, &tracing_pipe_fops);
2175         if (!entry)
2176                 pr_warning("Could not create debugfs "
2177                            "'tracing_threash' entry\n");
2178
2179 #ifdef CONFIG_DYNAMIC_FTRACE
2180         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2181                                     &ftrace_update_tot_cnt,
2182                                     &tracing_read_long_fops);
2183         if (!entry)
2184                 pr_warning("Could not create debugfs "
2185                            "'dyn_ftrace_total_info' entry\n");
2186 #endif
2187 }
2188
2189 /* dummy trace to disable tracing */
2190 static struct tracer no_tracer __read_mostly =
2191 {
2192         .name           = "none",
2193 };
2194
2195 static int trace_alloc_page(void)
2196 {
2197         struct trace_array_cpu *data;
2198         struct page *page, *tmp;
2199         LIST_HEAD(pages);
2200         void *array;
2201         int i;
2202
2203         /* first allocate a page for each CPU */
2204         for_each_possible_cpu(i) {
2205                 array = (void *)__get_free_page(GFP_KERNEL);
2206                 if (array == NULL) {
2207                         printk(KERN_ERR "tracer: failed to allocate page"
2208                                "for trace buffer!\n");
2209                         goto free_pages;
2210                 }
2211
2212                 page = virt_to_page(array);
2213                 list_add(&page->lru, &pages);
2214
2215 /* Only allocate if we are actually using the max trace */
2216 #ifdef CONFIG_TRACER_MAX_TRACE
2217                 array = (void *)__get_free_page(GFP_KERNEL);
2218                 if (array == NULL) {
2219                         printk(KERN_ERR "tracer: failed to allocate page"
2220                                "for trace buffer!\n");
2221                         goto free_pages;
2222                 }
2223                 page = virt_to_page(array);
2224                 list_add(&page->lru, &pages);
2225 #endif
2226         }
2227
2228         /* Now that we successfully allocate a page per CPU, add them */
2229         for_each_possible_cpu(i) {
2230                 data = global_trace.data[i];
2231                 spin_lock_init(&data->lock);
2232                 lockdep_set_class(&data->lock, &data->lock_key);
2233                 page = list_entry(pages.next, struct page, lru);
2234                 list_del_init(&page->lru);
2235                 list_add_tail(&page->lru, &data->trace_pages);
2236                 ClearPageLRU(page);
2237
2238 #ifdef CONFIG_TRACER_MAX_TRACE
2239                 data = max_tr.data[i];
2240                 spin_lock_init(&data->lock);
2241                 lockdep_set_class(&data->lock, &data->lock_key);
2242                 page = list_entry(pages.next, struct page, lru);
2243                 list_del_init(&page->lru);
2244                 list_add_tail(&page->lru, &data->trace_pages);
2245                 SetPageLRU(page);
2246 #endif
2247         }
2248         global_trace.entries += ENTRIES_PER_PAGE;
2249
2250         return 0;
2251
2252  free_pages:
2253         list_for_each_entry_safe(page, tmp, &pages, lru) {
2254                 list_del_init(&page->lru);
2255                 __free_page(page);
2256         }
2257         return -ENOMEM;
2258 }
2259
2260 __init static int tracer_alloc_buffers(void)
2261 {
2262         struct trace_array_cpu *data;
2263         void *array;
2264         struct page *page;
2265         int pages = 0;
2266         int ret = -ENOMEM;
2267         int i;
2268
2269         /* Allocate the first page for all buffers */
2270         for_each_possible_cpu(i) {
2271                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2272                 max_tr.data[i] = &per_cpu(max_data, i);
2273
2274                 array = (void *)__get_free_page(GFP_KERNEL);
2275                 if (array == NULL) {
2276                         printk(KERN_ERR "tracer: failed to allocate page"
2277                                "for trace buffer!\n");
2278                         goto free_buffers;
2279                 }
2280
2281                 /* set the array to the list */
2282                 INIT_LIST_HEAD(&data->trace_pages);
2283                 page = virt_to_page(array);
2284                 list_add(&page->lru, &data->trace_pages);
2285                 /* use the LRU flag to differentiate the two buffers */
2286                 ClearPageLRU(page);
2287
2288 /* Only allocate if we are actually using the max trace */
2289 #ifdef CONFIG_TRACER_MAX_TRACE
2290                 array = (void *)__get_free_page(GFP_KERNEL);
2291                 if (array == NULL) {
2292                         printk(KERN_ERR "tracer: failed to allocate page"
2293                                "for trace buffer!\n");
2294                         goto free_buffers;
2295                 }
2296
2297                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2298                 page = virt_to_page(array);
2299                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2300                 SetPageLRU(page);
2301 #endif
2302         }
2303
2304         /*
2305          * Since we allocate by orders of pages, we may be able to
2306          * round up a bit.
2307          */
2308         global_trace.entries = ENTRIES_PER_PAGE;
2309         pages++;
2310
2311         while (global_trace.entries < trace_nr_entries) {
2312                 if (trace_alloc_page())
2313                         break;
2314                 pages++;
2315         }
2316         max_tr.entries = global_trace.entries;
2317
2318         pr_info("tracer: %d pages allocated for %ld",
2319                 pages, trace_nr_entries);
2320         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2321         pr_info("   actual entries %ld\n", global_trace.entries);
2322
2323         tracer_init_debugfs();
2324
2325         trace_init_cmdlines();
2326
2327         register_tracer(&no_tracer);
2328         current_trace = &no_tracer;
2329
2330         /* All seems OK, enable tracing */
2331         tracing_disabled = 0;
2332
2333         return 0;
2334
2335  free_buffers:
2336         for (i-- ; i >= 0; i--) {
2337                 struct page *page, *tmp;
2338                 struct trace_array_cpu *data = global_trace.data[i];
2339
2340                 if (data) {
2341                         list_for_each_entry_safe(page, tmp,
2342                                                  &data->trace_pages, lru) {
2343                                 list_del_init(&page->lru);
2344                                 __free_page(page);
2345                         }
2346                 }
2347
2348 #ifdef CONFIG_TRACER_MAX_TRACE
2349                 data = max_tr.data[i];
2350                 if (data) {
2351                         list_for_each_entry_safe(page, tmp,
2352                                                  &data->trace_pages, lru) {
2353                                 list_del_init(&page->lru);
2354                                 __free_page(page);
2355                         }
2356                 }
2357 #endif
2358         }
2359         return ret;
2360 }
2361 fs_initcall(tracer_alloc_buffers);