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         unsigned long irq_flags;
 695
 696         spin_lock_irqsave(&data->lock, irq_flags);
 697         entry                   = tracing_get_trace_entry(tr, data);
 698         tracing_generic_entry_update(entry, flags);
 699         entry->type             = TRACE_FN;
 700         entry->fn.ip            = ip;
 701         entry->fn.parent_ip     = parent_ip;
 702         spin_unlock_irqrestore(&data->lock, irq_flags);
 703 }
 704
 705 notrace void
 706 trace_special(struct trace_array *tr, struct trace_array_cpu *data,
 707               unsigned long arg1, unsigned long arg2, unsigned long arg3)
 708 {
 709         struct trace_entry *entry;
 710         unsigned long irq_flags;
 711
 712         spin_lock_irqsave(&data->lock, irq_flags);
 713         entry                   = tracing_get_trace_entry(tr, data);
 714         tracing_generic_entry_update(entry, 0);
 715         entry->type             = TRACE_SPECIAL;
 716         entry->special.arg1     = arg1;
 717         entry->special.arg2     = arg2;
 718         entry->special.arg3     = arg3;
 719         spin_unlock_irqrestore(&data->lock, irq_flags);
 720 }
 721
 722 notrace void
 723 tracing_sched_switch_trace(struct trace_array *tr,
 724                            struct trace_array_cpu *data,
 725                            struct task_struct *prev, struct task_struct *next,
 726                            unsigned long flags)
 727 {
 728         struct trace_entry *entry;
 729         unsigned long irq_flags;
 730
 731         spin_lock_irqsave(&data->lock, irq_flags);
 732         entry                   = tracing_get_trace_entry(tr, data);
 733         tracing_generic_entry_update(entry, flags);
 734         entry->type             = TRACE_CTX;
 735         entry->ctx.prev_pid     = prev->pid;
 736         entry->ctx.prev_prio    = prev->prio;
 737         entry->ctx.prev_state   = prev->state;
 738         entry->ctx.next_pid     = next->pid;
 739         entry->ctx.next_prio    = next->prio;
 740         spin_unlock_irqrestore(&data->lock, irq_flags);
 741 }
 742
 743 enum trace_file_type {
 744         TRACE_FILE_LAT_FMT      = 1,
 745 };
 746
 747 static struct trace_entry *
 748 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 749                 struct trace_iterator *iter, int cpu)
 750 {
 751         struct page *page;
 752         struct trace_entry *array;
 753
 754         if (iter->next_idx[cpu] >= tr->entries ||
 755             iter->next_idx[cpu] >= data->trace_idx ||
 756             (data->trace_head == data->trace_tail &&
 757              data->trace_head_idx == data->trace_tail_idx))
 758                 return NULL;
 759
 760         if (!iter->next_page[cpu]) {
 761                 /* Initialize the iterator for this cpu trace buffer */
 762                 WARN_ON(!data->trace_tail);
 763                 page = virt_to_page(data->trace_tail);
 764                 iter->next_page[cpu] = &page->lru;
 765                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 766         }
 767
 768         page = list_entry(iter->next_page[cpu], struct page, lru);
 769         BUG_ON(&data->trace_pages == &page->lru);
 770
 771         array = page_address(page);
 772
 773         /* Still possible to catch up to the tail */
 774         if (iter->next_idx[cpu] && array == data->trace_tail &&
 775             iter->next_page_idx[cpu] == data->trace_tail_idx)
 776                 return NULL;
 777
 778         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 779         return &array[iter->next_page_idx[cpu]];
 780 }
 781
 782 static struct notrace trace_entry *
 783 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 784 {
 785         struct trace_array *tr = iter->tr;
 786         struct trace_entry *ent, *next = NULL;
 787         int next_cpu = -1;
 788         int cpu;
 789
 790         for_each_possible_cpu(cpu) {
 791                 if (!head_page(tr->data[cpu]))
 792                         continue;
 793                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 794                 /*
 795                  * Pick the entry with the smallest timestamp:
 796                  */
 797                 if (ent && (!next || ent->t < next->t)) {
 798                         next = ent;
 799                         next_cpu = cpu;
 800                 }
 801         }
 802
 803         if (ent_cpu)
 804                 *ent_cpu = next_cpu;
 805
 806         return next;
 807 }
 808
 809 static notrace void trace_iterator_increment(struct trace_iterator *iter)
 810 {
 811         iter->idx++;
 812         iter->next_idx[iter->cpu]++;
 813         iter->next_page_idx[iter->cpu]++;
 814
 815         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 816                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 817
 818                 iter->next_page_idx[iter->cpu] = 0;
 819                 iter->next_page[iter->cpu] =
 820                         trace_next_list(data, iter->next_page[iter->cpu]);
 821         }
 822 }
 823
 824 static notrace void trace_consume(struct trace_iterator *iter)
 825 {
 826         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 827
 828         data->trace_tail_idx++;
 829         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 830                 data->trace_tail = trace_next_page(data, data->trace_tail);
 831                 data->trace_tail_idx = 0;
 832         }
 833
 834         /* Check if we empty it, then reset the index */
 835         if (data->trace_head == data->trace_tail &&
 836             data->trace_head_idx == data->trace_tail_idx)
 837                 data->trace_idx = 0;
 838 }
 839
 840 static notrace void *find_next_entry_inc(struct trace_iterator *iter)
 841 {
 842         struct trace_entry *next;
 843         int next_cpu = -1;
 844
 845         next = find_next_entry(iter, &next_cpu);
 846
 847         iter->prev_ent = iter->ent;
 848         iter->prev_cpu = iter->cpu;
 849
 850         iter->ent = next;
 851         iter->cpu = next_cpu;
 852
 853         if (next)
 854                 trace_iterator_increment(iter);
 855
 856         return next ? iter : NULL;
 857 }
 858
 859 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 860 {
 861         struct trace_iterator *iter = m->private;
 862         void *last_ent = iter->ent;
 863         int i = (int)*pos;
 864         void *ent;
 865
 866         (*pos)++;
 867
 868         /* can't go backwards */
 869         if (iter->idx > i)
 870                 return NULL;
 871
 872         if (iter->idx < 0)
 873                 ent = find_next_entry_inc(iter);
 874         else
 875                 ent = iter;
 876
 877         while (ent && iter->idx < i)
 878                 ent = find_next_entry_inc(iter);
 879
 880         iter->pos = *pos;
 881
 882         if (last_ent && !ent)
 883                 seq_puts(m, "\n\nvim:ft=help\n");
 884
 885         return ent;
 886 }
 887
 888 static void *s_start(struct seq_file *m, loff_t *pos)
 889 {
 890         struct trace_iterator *iter = m->private;
 891         void *p = NULL;
 892         loff_t l = 0;
 893         int i;
 894
 895         mutex_lock(&trace_types_lock);
 896
 897         if (!current_trace || current_trace != iter->trace)
 898                 return NULL;
 899
 900         atomic_inc(&trace_record_cmdline_disabled);
 901
 902         /* let the tracer grab locks here if needed */
 903         if (current_trace->start)
 904                 current_trace->start(iter);
 905
 906         if (*pos != iter->pos) {
 907                 iter->ent = NULL;
 908                 iter->cpu = 0;
 909                 iter->idx = -1;
 910                 iter->prev_ent = NULL;
 911                 iter->prev_cpu = -1;
 912
 913                 for_each_possible_cpu(i) {
 914                         iter->next_idx[i] = 0;
 915                         iter->next_page[i] = NULL;
 916                 }
 917
 918                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 919                         ;
 920
 921         } else {
 922                 l = *pos - 1;
 923                 p = s_next(m, p, &l);
 924         }
 925
 926         return p;
 927 }
 928
 929 static void s_stop(struct seq_file *m, void *p)
 930 {
 931         struct trace_iterator *iter = m->private;
 932
 933         atomic_dec(&trace_record_cmdline_disabled);
 934
 935         /* let the tracer release locks here if needed */
 936         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 937                 iter->trace->stop(iter);
 938
 939         mutex_unlock(&trace_types_lock);
 940 }
 941
 942 static int
 943 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 944 {
 945 #ifdef CONFIG_KALLSYMS
 946         char str[KSYM_SYMBOL_LEN];
 947
 948         kallsyms_lookup(address, NULL, NULL, NULL, str);
 949
 950         return trace_seq_printf(s, fmt, str);
 951 #endif
 952         return 1;
 953 }
 954
 955 static int
 956 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 957                      unsigned long address)
 958 {
 959 #ifdef CONFIG_KALLSYMS
 960         char str[KSYM_SYMBOL_LEN];
 961
 962         sprint_symbol(str, address);
 963         return trace_seq_printf(s, fmt, str);
 964 #endif
 965         return 1;
 966 }
 967
 968 #ifndef CONFIG_64BIT
 969 # define IP_FMT "%08lx"
 970 #else
 971 # define IP_FMT "%016lx"
 972 #endif
 973
 974 static notrace int
 975 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 976 {
 977         int ret;
 978
 979         if (!ip)
 980                 return trace_seq_printf(s, "0");
 981
 982         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 983                 ret = seq_print_sym_offset(s, "%s", ip);
 984         else
 985                 ret = seq_print_sym_short(s, "%s", ip);
 986
 987         if (!ret)
 988                 return 0;
 989
 990         if (sym_flags & TRACE_ITER_SYM_ADDR)
 991                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 992         return ret;
 993 }
 994
 995 static notrace void print_lat_help_header(struct seq_file *m)
 996 {
 997         seq_puts(m, "#                _------=> CPU#            \n");
 998         seq_puts(m, "#               / _-----=> irqs-off        \n");
 999         seq_puts(m, "#              | / _----=> need-resched    \n");
1000         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1001         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1002         seq_puts(m, "#              |||| /                      \n");
1003         seq_puts(m, "#              |||||     delay             \n");
1004         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1005         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1006 }
1007
1008 static notrace void print_func_help_header(struct seq_file *m)
1009 {
1010         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1011         seq_puts(m, "#              | |      |          |         |\n");
1012 }
1013
1014
1015 static notrace void
1016 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1017 {
1018         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1019         struct trace_array *tr = iter->tr;
1020         struct trace_array_cpu *data = tr->data[tr->cpu];
1021         struct tracer *type = current_trace;
1022         unsigned long total   = 0;
1023         unsigned long entries = 0;
1024         int cpu;
1025         const char *name = "preemption";
1026
1027         if (type)
1028                 name = type->name;
1029
1030         for_each_possible_cpu(cpu) {
1031                 if (head_page(tr->data[cpu])) {
1032                         total += tr->data[cpu]->trace_idx;
1033                         if (tr->data[cpu]->trace_idx > tr->entries)
1034                                 entries += tr->entries;
1035                         else
1036                                 entries += tr->data[cpu]->trace_idx;
1037                 }
1038         }
1039
1040         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1041                    name, UTS_RELEASE);
1042         seq_puts(m, "-----------------------------------"
1043                  "---------------------------------\n");
1044         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1045                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1046                    nsecs_to_usecs(data->saved_latency),
1047                    entries,
1048                    total,
1049                    tr->cpu,
1050 #if defined(CONFIG_PREEMPT_NONE)
1051                    "server",
1052 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1053                    "desktop",
1054 #elif defined(CONFIG_PREEMPT_DESKTOP)
1055                    "preempt",
1056 #else
1057                    "unknown",
1058 #endif
1059                    /* These are reserved for later use */
1060                    0, 0, 0, 0);
1061 #ifdef CONFIG_SMP
1062         seq_printf(m, " #P:%d)\n", num_online_cpus());
1063 #else
1064         seq_puts(m, ")\n");
1065 #endif
1066         seq_puts(m, "    -----------------\n");
1067         seq_printf(m, "    | task: %.16s-%d "
1068                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1069                    data->comm, data->pid, data->uid, data->nice,
1070                    data->policy, data->rt_priority);
1071         seq_puts(m, "    -----------------\n");
1072
1073         if (data->critical_start) {
1074                 seq_puts(m, " => started at: ");
1075                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1076                 trace_print_seq(m, &iter->seq);
1077                 seq_puts(m, "\n => ended at:   ");
1078                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1079                 trace_print_seq(m, &iter->seq);
1080                 seq_puts(m, "\n");
1081         }
1082
1083         seq_puts(m, "\n");
1084 }
1085
1086 static notrace void
1087 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1088 {
1089         int hardirq, softirq;
1090         char *comm;
1091
1092         comm = trace_find_cmdline(entry->pid);
1093
1094         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1095         trace_seq_printf(s, "%d", cpu);
1096         trace_seq_printf(s, "%c%c",
1097                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1098                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1099
1100         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1101         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1102         if (hardirq && softirq)
1103                 trace_seq_putc(s, 'H');
1104         else {
1105                 if (hardirq)
1106                         trace_seq_putc(s, 'h');
1107                 else {
1108                         if (softirq)
1109                                 trace_seq_putc(s, 's');
1110                         else
1111                                 trace_seq_putc(s, '.');
1112                 }
1113         }
1114
1115         if (entry->preempt_count)
1116                 trace_seq_printf(s, "%x", entry->preempt_count);
1117         else
1118                 trace_seq_puts(s, ".");
1119 }
1120
1121 unsigned long preempt_mark_thresh = 100;
1122
1123 static notrace void
1124 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1125                     unsigned long rel_usecs)
1126 {
1127         trace_seq_printf(s, " %4lldus", abs_usecs);
1128         if (rel_usecs > preempt_mark_thresh)
1129                 trace_seq_puts(s, "!: ");
1130         else if (rel_usecs > 1)
1131                 trace_seq_puts(s, "+: ");
1132         else
1133                 trace_seq_puts(s, " : ");
1134 }
1135
1136 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1137
1138 static notrace int
1139 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1140 {
1141         struct trace_seq *s = &iter->seq;
1142         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1143         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1144         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1145         struct trace_entry *entry = iter->ent;
1146         unsigned long abs_usecs;
1147         unsigned long rel_usecs;
1148         char *comm;
1149         int S;
1150
1151         if (!next_entry)
1152                 next_entry = entry;
1153         rel_usecs = ns2usecs(next_entry->t - entry->t);
1154         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1155
1156         if (verbose) {
1157                 comm = trace_find_cmdline(entry->pid);
1158                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1159                                  " %ld.%03ldms (+%ld.%03ldms): ",
1160                                  comm,
1161                                  entry->pid, cpu, entry->flags,
1162                                  entry->preempt_count, trace_idx,
1163                                  ns2usecs(entry->t),
1164                                  abs_usecs/1000,
1165                                  abs_usecs % 1000, rel_usecs/1000,
1166                                  rel_usecs % 1000);
1167         } else {
1168                 lat_print_generic(s, entry, cpu);
1169                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1170         }
1171         switch (entry->type) {
1172         case TRACE_FN:
1173                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1174                 trace_seq_puts(s, " (");
1175                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1176                 trace_seq_puts(s, ")\n");
1177                 break;
1178         case TRACE_CTX:
1179                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1180                         state_to_char[entry->ctx.prev_state] : 'X';
1181                 comm = trace_find_cmdline(entry->ctx.next_pid);
1182                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1183                                  entry->ctx.prev_pid,
1184                                  entry->ctx.prev_prio,
1185                                  S,
1186                                  entry->ctx.next_pid,
1187                                  entry->ctx.next_prio,
1188                                  comm);
1189                 break;
1190         case TRACE_SPECIAL:
1191                 trace_seq_printf(s, " %lx %lx %lx\n",
1192                                  entry->special.arg1,
1193                                  entry->special.arg2,
1194                                  entry->special.arg3);
1195                 break;
1196         default:
1197                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1198         }
1199         return 1;
1200 }
1201
1202 static notrace int print_trace_fmt(struct trace_iterator *iter)
1203 {
1204         struct trace_seq *s = &iter->seq;
1205         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1206         struct trace_entry *entry;
1207         unsigned long usec_rem;
1208         unsigned long long t;
1209         unsigned long secs;
1210         char *comm;
1211         int S;
1212         int ret;
1213
1214         entry = iter->ent;
1215
1216         comm = trace_find_cmdline(iter->ent->pid);
1217
1218         t = ns2usecs(entry->t);
1219         usec_rem = do_div(t, 1000000ULL);
1220         secs = (unsigned long)t;
1221
1222         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1223         if (!ret)
1224                 return 0;
1225         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1226         if (!ret)
1227                 return 0;
1228         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1229         if (!ret)
1230                 return 0;
1231
1232         switch (entry->type) {
1233         case TRACE_FN:
1234                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1235                 if (!ret)
1236                         return 0;
1237                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1238                                                 entry->fn.parent_ip) {
1239                         ret = trace_seq_printf(s, " <-");
1240                         if (!ret)
1241                                 return 0;
1242                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1243                                                sym_flags);
1244                         if (!ret)
1245                                 return 0;
1246                 }
1247                 ret = trace_seq_printf(s, "\n");
1248                 if (!ret)
1249                         return 0;
1250                 break;
1251         case TRACE_CTX:
1252                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1253                         state_to_char[entry->ctx.prev_state] : 'X';
1254                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1255                                        entry->ctx.prev_pid,
1256                                        entry->ctx.prev_prio,
1257                                        S,
1258                                        entry->ctx.next_pid,
1259                                        entry->ctx.next_prio);
1260                 if (!ret)
1261                         return 0;
1262                 break;
1263         case TRACE_SPECIAL:
1264                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1265                                  entry->special.arg1,
1266                                  entry->special.arg2,
1267                                  entry->special.arg3);
1268                 if (!ret)
1269                         return 0;
1270                 break;
1271         }
1272         return 1;
1273 }
1274
1275 static notrace int print_raw_fmt(struct trace_iterator *iter)
1276 {
1277         struct trace_seq *s = &iter->seq;
1278         struct trace_entry *entry;
1279         int ret;
1280         int S;
1281
1282         entry = iter->ent;
1283
1284         ret = trace_seq_printf(s, "%d %d %llu ",
1285                 entry->pid, iter->cpu, entry->t);
1286         if (!ret)
1287                 return 0;
1288
1289         switch (entry->type) {
1290         case TRACE_FN:
1291                 ret = trace_seq_printf(s, "%x %x\n",
1292                                         entry->fn.ip, entry->fn.parent_ip);
1293                 if (!ret)
1294                         return 0;
1295                 break;
1296         case TRACE_CTX:
1297                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1298                         state_to_char[entry->ctx.prev_state] : 'X';
1299                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1300                                        entry->ctx.prev_pid,
1301                                        entry->ctx.prev_prio,
1302                                        S,
1303                                        entry->ctx.next_pid,
1304                                        entry->ctx.next_prio);
1305                 if (!ret)
1306                         return 0;
1307                 break;
1308         case TRACE_SPECIAL:
1309                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1310                                  entry->special.arg1,
1311                                  entry->special.arg2,
1312                                  entry->special.arg3);
1313                 if (!ret)
1314                         return 0;
1315                 break;
1316         }
1317         return 1;
1318 }
1319
1320 #define SEQ_PUT_FIELD_RET(s, x)                         \
1321 do {                                                    \
1322         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1323                 return 0;                               \
1324 } while (0)
1325
1326 static notrace int print_bin_fmt(struct trace_iterator *iter)
1327 {
1328         struct trace_seq *s = &iter->seq;
1329         struct trace_entry *entry;
1330
1331         entry = iter->ent;
1332
1333         SEQ_PUT_FIELD_RET(s, entry->pid);
1334         SEQ_PUT_FIELD_RET(s, entry->cpu);
1335         SEQ_PUT_FIELD_RET(s, entry->t);
1336
1337         switch (entry->type) {
1338         case TRACE_FN:
1339                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1340                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1341                 break;
1342         case TRACE_CTX:
1343                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1344                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1345                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1346                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1347                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1348                 break;
1349         case TRACE_SPECIAL:
1350                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1351                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1352                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1353                 break;
1354         }
1355         return 1;
1356 }
1357
1358 static int trace_empty(struct trace_iterator *iter)
1359 {
1360         struct trace_array_cpu *data;
1361         int cpu;
1362
1363         for_each_possible_cpu(cpu) {
1364                 data = iter->tr->data[cpu];
1365
1366                 if (head_page(data) && data->trace_idx &&
1367                     (data->trace_tail != data->trace_head ||
1368                      data->trace_tail_idx != data->trace_head_idx))
1369                         return 0;
1370         }
1371         return 1;
1372 }
1373
1374 static int print_trace_line(struct trace_iterator *iter)
1375 {
1376         if (trace_flags & TRACE_ITER_BIN)
1377                 return print_bin_fmt(iter);
1378
1379         if (trace_flags & TRACE_ITER_RAW)
1380                 return print_raw_fmt(iter);
1381
1382         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1383                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1384
1385         return print_trace_fmt(iter);
1386 }
1387
1388 static int s_show(struct seq_file *m, void *v)
1389 {
1390         struct trace_iterator *iter = v;
1391
1392         if (iter->ent == NULL) {
1393                 if (iter->tr) {
1394                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1395                         seq_puts(m, "#\n");
1396                 }
1397                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1398                         /* print nothing if the buffers are empty */
1399                         if (trace_empty(iter))
1400                                 return 0;
1401                         print_trace_header(m, iter);
1402                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1403                                 print_lat_help_header(m);
1404                 } else {
1405                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1406                                 print_func_help_header(m);
1407                 }
1408         } else {
1409                 print_trace_line(iter);
1410                 trace_print_seq(m, &iter->seq);
1411         }
1412
1413         return 0;
1414 }
1415
1416 static struct seq_operations tracer_seq_ops = {
1417         .start          = s_start,
1418         .next           = s_next,
1419         .stop           = s_stop,
1420         .show           = s_show,
1421 };
1422
1423 static struct trace_iterator notrace *
1424 __tracing_open(struct inode *inode, struct file *file, int *ret)
1425 {
1426         struct trace_iterator *iter;
1427
1428         if (tracing_disabled) {
1429                 *ret = -ENODEV;
1430                 return NULL;
1431         }
1432
1433         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1434         if (!iter) {
1435                 *ret = -ENOMEM;
1436                 goto out;
1437         }
1438
1439         mutex_lock(&trace_types_lock);
1440         if (current_trace && current_trace->print_max)
1441                 iter->tr = &max_tr;
1442         else
1443                 iter->tr = inode->i_private;
1444         iter->trace = current_trace;
1445         iter->pos = -1;
1446
1447         /* TODO stop tracer */
1448         *ret = seq_open(file, &tracer_seq_ops);
1449         if (!*ret) {
1450                 struct seq_file *m = file->private_data;
1451                 m->private = iter;
1452
1453                 /* stop the trace while dumping */
1454                 if (iter->tr->ctrl)
1455                         tracer_enabled = 0;
1456
1457                 if (iter->trace && iter->trace->open)
1458                         iter->trace->open(iter);
1459         } else {
1460                 kfree(iter);
1461                 iter = NULL;
1462         }
1463         mutex_unlock(&trace_types_lock);
1464
1465  out:
1466         return iter;
1467 }
1468
1469 int tracing_open_generic(struct inode *inode, struct file *filp)
1470 {
1471         if (tracing_disabled)
1472                 return -ENODEV;
1473
1474         filp->private_data = inode->i_private;
1475         return 0;
1476 }
1477
1478 int tracing_release(struct inode *inode, struct file *file)
1479 {
1480         struct seq_file *m = (struct seq_file *)file->private_data;
1481         struct trace_iterator *iter = m->private;
1482
1483         mutex_lock(&trace_types_lock);
1484         if (iter->trace && iter->trace->close)
1485                 iter->trace->close(iter);
1486
1487         /* reenable tracing if it was previously enabled */
1488         if (iter->tr->ctrl)
1489                 tracer_enabled = 1;
1490         mutex_unlock(&trace_types_lock);
1491
1492         seq_release(inode, file);
1493         kfree(iter);
1494         return 0;
1495 }
1496
1497 static int tracing_open(struct inode *inode, struct file *file)
1498 {
1499         int ret;
1500
1501         __tracing_open(inode, file, &ret);
1502
1503         return ret;
1504 }
1505
1506 static int tracing_lt_open(struct inode *inode, struct file *file)
1507 {
1508         struct trace_iterator *iter;
1509         int ret;
1510
1511         iter = __tracing_open(inode, file, &ret);
1512
1513         if (!ret)
1514                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1515
1516         return ret;
1517 }
1518
1519
1520 static notrace void *
1521 t_next(struct seq_file *m, void *v, loff_t *pos)
1522 {
1523         struct tracer *t = m->private;
1524
1525         (*pos)++;
1526
1527         if (t)
1528                 t = t->next;
1529
1530         m->private = t;
1531
1532         return t;
1533 }
1534
1535 static void *t_start(struct seq_file *m, loff_t *pos)
1536 {
1537         struct tracer *t = m->private;
1538         loff_t l = 0;
1539
1540         mutex_lock(&trace_types_lock);
1541         for (; t && l < *pos; t = t_next(m, t, &l))
1542                 ;
1543
1544         return t;
1545 }
1546
1547 static void t_stop(struct seq_file *m, void *p)
1548 {
1549         mutex_unlock(&trace_types_lock);
1550 }
1551
1552 static int t_show(struct seq_file *m, void *v)
1553 {
1554         struct tracer *t = v;
1555
1556         if (!t)
1557                 return 0;
1558
1559         seq_printf(m, "%s", t->name);
1560         if (t->next)
1561                 seq_putc(m, ' ');
1562         else
1563                 seq_putc(m, '\n');
1564
1565         return 0;
1566 }
1567
1568 static struct seq_operations show_traces_seq_ops = {
1569         .start          = t_start,
1570         .next           = t_next,
1571         .stop           = t_stop,
1572         .show           = t_show,
1573 };
1574
1575 static int show_traces_open(struct inode *inode, struct file *file)
1576 {
1577         int ret;
1578
1579         if (tracing_disabled)
1580                 return -ENODEV;
1581
1582         ret = seq_open(file, &show_traces_seq_ops);
1583         if (!ret) {
1584                 struct seq_file *m = file->private_data;
1585                 m->private = trace_types;
1586         }
1587
1588         return ret;
1589 }
1590
1591 static struct file_operations tracing_fops = {
1592         .open           = tracing_open,
1593         .read           = seq_read,
1594         .llseek         = seq_lseek,
1595         .release        = tracing_release,
1596 };
1597
1598 static struct file_operations tracing_lt_fops = {
1599         .open           = tracing_lt_open,
1600         .read           = seq_read,
1601         .llseek         = seq_lseek,
1602         .release        = tracing_release,
1603 };
1604
1605 static struct file_operations show_traces_fops = {
1606         .open = show_traces_open,
1607         .read = seq_read,
1608         .release = seq_release,
1609 };
1610
1611 static ssize_t
1612 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1613                        size_t cnt, loff_t *ppos)
1614 {
1615         char *buf;
1616         int r = 0;
1617         int len = 0;
1618         int i;
1619
1620         /* calulate max size */
1621         for (i = 0; trace_options[i]; i++) {
1622                 len += strlen(trace_options[i]);
1623                 len += 3; /* "no" and space */
1624         }
1625
1626         /* +2 for \n and \0 */
1627         buf = kmalloc(len + 2, GFP_KERNEL);
1628         if (!buf)
1629                 return -ENOMEM;
1630
1631         for (i = 0; trace_options[i]; i++) {
1632                 if (trace_flags & (1 << i))
1633                         r += sprintf(buf + r, "%s ", trace_options[i]);
1634                 else
1635                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1636         }
1637
1638         r += sprintf(buf + r, "\n");
1639         WARN_ON(r >= len + 2);
1640
1641         r = simple_read_from_buffer(ubuf, cnt, ppos,
1642                                     buf, r);
1643
1644         kfree(buf);
1645
1646         return r;
1647 }
1648
1649 static ssize_t
1650 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1651                         size_t cnt, loff_t *ppos)
1652 {
1653         char buf[64];
1654         char *cmp = buf;
1655         int neg = 0;
1656         int i;
1657
1658         if (cnt > 63)
1659                 cnt = 63;
1660
1661         if (copy_from_user(&buf, ubuf, cnt))
1662                 return -EFAULT;
1663
1664         buf[cnt] = 0;
1665
1666         if (strncmp(buf, "no", 2) == 0) {
1667                 neg = 1;
1668                 cmp += 2;
1669         }
1670
1671         for (i = 0; trace_options[i]; i++) {
1672                 int len = strlen(trace_options[i]);
1673
1674                 if (strncmp(cmp, trace_options[i], len) == 0) {
1675                         if (neg)
1676                                 trace_flags &= ~(1 << i);
1677                         else
1678                                 trace_flags |= (1 << i);
1679                         break;
1680                 }
1681         }
1682
1683         filp->f_pos += cnt;
1684
1685         return cnt;
1686 }
1687
1688 static struct file_operations tracing_iter_fops = {
1689         .open = tracing_open_generic,
1690         .read = tracing_iter_ctrl_read,
1691         .write = tracing_iter_ctrl_write,
1692 };
1693
1694 static const char readme_msg[] =
1695         "tracing mini-HOWTO:\n\n"
1696         "# mkdir /debug\n"
1697         "# mount -t debugfs nodev /debug\n\n"
1698         "# cat /debug/tracing/available_tracers\n"
1699         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1700         "# cat /debug/tracing/current_tracer\n"
1701         "none\n"
1702         "# echo sched_switch > /debug/tracing/current_tracer\n"
1703         "# cat /debug/tracing/current_tracer\n"
1704         "sched_switch\n"
1705         "# cat /debug/tracing/iter_ctrl\n"
1706         "noprint-parent nosym-offset nosym-addr noverbose\n"
1707         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1708         "# echo 1 > /debug/tracing/tracing_enabled\n"
1709         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1710         "echo 0 > /debug/tracing/tracing_enabled\n"
1711 ;
1712
1713 static ssize_t
1714 tracing_readme_read(struct file *filp, char __user *ubuf,
1715                        size_t cnt, loff_t *ppos)
1716 {
1717         return simple_read_from_buffer(ubuf, cnt, ppos,
1718                                         readme_msg, strlen(readme_msg));
1719 }
1720
1721 static struct file_operations tracing_readme_fops = {
1722         .open = tracing_open_generic,
1723         .read = tracing_readme_read,
1724 };
1725
1726
1727 static ssize_t
1728 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1729                   size_t cnt, loff_t *ppos)
1730 {
1731         struct trace_array *tr = filp->private_data;
1732         char buf[64];
1733         int r;
1734
1735         r = sprintf(buf, "%ld\n", tr->ctrl);
1736         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1737 }
1738
1739 static ssize_t
1740 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1741                    size_t cnt, loff_t *ppos)
1742 {
1743         struct trace_array *tr = filp->private_data;
1744         long val;
1745         char buf[64];
1746
1747         if (cnt > 63)
1748                 cnt = 63;
1749
1750         if (copy_from_user(&buf, ubuf, cnt))
1751                 return -EFAULT;
1752
1753         buf[cnt] = 0;
1754
1755         val = simple_strtoul(buf, NULL, 10);
1756
1757         val = !!val;
1758
1759         mutex_lock(&trace_types_lock);
1760         if (tr->ctrl ^ val) {
1761                 if (val)
1762                         tracer_enabled = 1;
1763                 else
1764                         tracer_enabled = 0;
1765
1766                 tr->ctrl = val;
1767
1768                 if (current_trace && current_trace->ctrl_update)
1769                         current_trace->ctrl_update(tr);
1770         }
1771         mutex_unlock(&trace_types_lock);
1772
1773         filp->f_pos += cnt;
1774
1775         return cnt;
1776 }
1777
1778 static ssize_t
1779 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1780                        size_t cnt, loff_t *ppos)
1781 {
1782         char buf[max_tracer_type_len+2];
1783         int r;
1784
1785         mutex_lock(&trace_types_lock);
1786         if (current_trace)
1787                 r = sprintf(buf, "%s\n", current_trace->name);
1788         else
1789                 r = sprintf(buf, "\n");
1790         mutex_unlock(&trace_types_lock);
1791
1792         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1793 }
1794
1795 static ssize_t
1796 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1797                         size_t cnt, loff_t *ppos)
1798 {
1799         struct trace_array *tr = &global_trace;
1800         struct tracer *t;
1801         char buf[max_tracer_type_len+1];
1802         int i;
1803
1804         if (cnt > max_tracer_type_len)
1805                 cnt = max_tracer_type_len;
1806
1807         if (copy_from_user(&buf, ubuf, cnt))
1808                 return -EFAULT;
1809
1810         buf[cnt] = 0;
1811
1812         /* strip ending whitespace. */
1813         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1814                 buf[i] = 0;
1815
1816         mutex_lock(&trace_types_lock);
1817         for (t = trace_types; t; t = t->next) {
1818                 if (strcmp(t->name, buf) == 0)
1819                         break;
1820         }
1821         if (!t || t == current_trace)
1822                 goto out;
1823
1824         if (current_trace && current_trace->reset)
1825                 current_trace->reset(tr);
1826
1827         current_trace = t;
1828         if (t->init)
1829                 t->init(tr);
1830
1831  out:
1832         mutex_unlock(&trace_types_lock);
1833
1834         filp->f_pos += cnt;
1835
1836         return cnt;
1837 }
1838
1839 static ssize_t
1840 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1841                      size_t cnt, loff_t *ppos)
1842 {
1843         unsigned long *ptr = filp->private_data;
1844         char buf[64];
1845         int r;
1846
1847         r = snprintf(buf, 64, "%ld\n",
1848                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1849         if (r > 64)
1850                 r = 64;
1851         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1852 }
1853
1854 static ssize_t
1855 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1856                       size_t cnt, loff_t *ppos)
1857 {
1858         long *ptr = filp->private_data;
1859         long val;
1860         char buf[64];
1861
1862         if (cnt > 63)
1863                 cnt = 63;
1864
1865         if (copy_from_user(&buf, ubuf, cnt))
1866                 return -EFAULT;
1867
1868         buf[cnt] = 0;
1869
1870         val = simple_strtoul(buf, NULL, 10);
1871
1872         *ptr = val * 1000;
1873
1874         return cnt;
1875 }
1876
1877 static atomic_t tracing_reader;
1878
1879 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1880 {
1881         struct trace_iterator *iter;
1882
1883         if (tracing_disabled)
1884                 return -ENODEV;
1885
1886         /* We only allow for reader of the pipe */
1887         if (atomic_inc_return(&tracing_reader) != 1) {
1888                 atomic_dec(&tracing_reader);
1889                 return -EBUSY;
1890         }
1891
1892         /* create a buffer to store the information to pass to userspace */
1893         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1894         if (!iter)
1895                 return -ENOMEM;
1896
1897         iter->tr = &global_trace;
1898
1899         filp->private_data = iter;
1900
1901         return 0;
1902 }
1903
1904 static int tracing_release_pipe(struct inode *inode, struct file *file)
1905 {
1906         struct trace_iterator *iter = file->private_data;
1907
1908         kfree(iter);
1909         atomic_dec(&tracing_reader);
1910
1911         return 0;
1912 }
1913
1914 /*
1915  * Consumer reader.
1916  */
1917 static ssize_t
1918 tracing_read_pipe(struct file *filp, char __user *ubuf,
1919                   size_t cnt, loff_t *ppos)
1920 {
1921         struct trace_iterator *iter = filp->private_data;
1922         struct trace_array_cpu *data;
1923         static cpumask_t mask;
1924         struct trace_entry *entry;
1925         static int start;
1926         unsigned long flags;
1927         int read = 0;
1928         int cpu;
1929         int len;
1930         int ret;
1931
1932         /* return any leftover data */
1933         if (iter->seq.len > start) {
1934                 len = iter->seq.len - start;
1935                 if (cnt > len)
1936                         cnt = len;
1937                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1938                 if (ret)
1939                         cnt = -EFAULT;
1940
1941                 start += len;
1942
1943                 return cnt;
1944         }
1945
1946         trace_seq_reset(&iter->seq);
1947         start = 0;
1948
1949         while (trace_empty(iter)) {
1950                 /*
1951                  * This is a make-shift waitqueue. The reason we don't use
1952                  * an actual wait queue is because:
1953                  *  1) we only ever have one waiter
1954                  *  2) the tracing, traces all functions, we don't want
1955                  *     the overhead of calling wake_up and friends
1956                  *     (and tracing them too)
1957                  *     Anyway, this is really very primitive wakeup.
1958                  */
1959                 set_current_state(TASK_INTERRUPTIBLE);
1960                 iter->tr->waiter = current;
1961
1962                 /* sleep for one second, and try again. */
1963                 schedule_timeout(HZ);
1964
1965                 iter->tr->waiter = NULL;
1966
1967                 if (signal_pending(current))
1968                         return -EINTR;
1969
1970                 /*
1971                  * We block until we read something and tracing is disabled.
1972                  * We still block if tracing is disabled, but we have never
1973                  * read anything. This allows a user to cat this file, and
1974                  * then enable tracing. But after we have read something,
1975                  * we give an EOF when tracing is again disabled.
1976                  *
1977                  * iter->pos will be 0 if we haven't read anything.
1978                  */
1979                 if (!tracer_enabled && iter->pos)
1980                         break;
1981
1982                 continue;
1983         }
1984
1985         /* stop when tracing is finished */
1986         if (trace_empty(iter))
1987                 return 0;
1988
1989         if (cnt >= PAGE_SIZE)
1990                 cnt = PAGE_SIZE - 1;
1991
1992         memset(iter, 0, sizeof(*iter));
1993         iter->tr = &global_trace;
1994         iter->pos = -1;
1995
1996         /*
1997          * We need to stop all tracing on all CPUS to read the
1998          * the next buffer. This is a bit expensive, but is
1999          * not done often. We fill all what we can read,
2000          * and then release the locks again.
2001          */
2002
2003         cpus_clear(mask);
2004         local_irq_save(flags);
2005         for_each_possible_cpu(cpu) {
2006                 data = iter->tr->data[cpu];
2007
2008                 if (!head_page(data) || !data->trace_idx)
2009                         continue;
2010
2011                 atomic_inc(&data->disabled);
2012                 spin_lock(&data->lock);
2013                 cpu_set(cpu, mask);
2014         }
2015
2016         while ((entry = find_next_entry_inc(iter)) != NULL) {
2017                 ret = print_trace_line(iter);
2018                 if (!ret)
2019                         break;
2020
2021                 trace_consume(iter);
2022
2023                 if (iter->seq.len >= cnt)
2024                         break;
2025         }
2026
2027         for_each_cpu_mask(cpu, mask) {
2028                 data = iter->tr->data[cpu];
2029                 spin_unlock(&data->lock);
2030                 atomic_dec(&data->disabled);
2031         }
2032         local_irq_restore(flags);
2033
2034         /* Now copy what we have to the user */
2035         read = iter->seq.len;
2036         if (read > cnt)
2037                 read = cnt;
2038
2039         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2040
2041         if (read < iter->seq.len)
2042                 start = read;
2043         else
2044                 trace_seq_reset(&iter->seq);
2045
2046         if (ret)
2047                 read = -EFAULT;
2048
2049         return read;
2050 }
2051
2052 static struct file_operations tracing_max_lat_fops = {
2053         .open           = tracing_open_generic,
2054         .read           = tracing_max_lat_read,
2055         .write          = tracing_max_lat_write,
2056 };
2057
2058 static struct file_operations tracing_ctrl_fops = {
2059         .open           = tracing_open_generic,
2060         .read           = tracing_ctrl_read,
2061         .write          = tracing_ctrl_write,
2062 };
2063
2064 static struct file_operations set_tracer_fops = {
2065         .open           = tracing_open_generic,
2066         .read           = tracing_set_trace_read,
2067         .write          = tracing_set_trace_write,
2068 };
2069
2070 static struct file_operations tracing_pipe_fops = {
2071         .open           = tracing_open_pipe,
2072         .read           = tracing_read_pipe,
2073         .release        = tracing_release_pipe,
2074 };
2075
2076 #ifdef CONFIG_DYNAMIC_FTRACE
2077
2078 static ssize_t
2079 tracing_read_long(struct file *filp, char __user *ubuf,
2080                   size_t cnt, loff_t *ppos)
2081 {
2082         unsigned long *p = filp->private_data;
2083         char buf[64];
2084         int r;
2085
2086         r = sprintf(buf, "%ld\n", *p);
2087
2088         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2089 }
2090
2091 static struct file_operations tracing_read_long_fops = {
2092         .open           = tracing_open_generic,
2093         .read           = tracing_read_long,
2094 };
2095 #endif
2096
2097 static struct dentry *d_tracer;
2098
2099 struct dentry *tracing_init_dentry(void)
2100 {
2101         static int once;
2102
2103         if (d_tracer)
2104                 return d_tracer;
2105
2106         d_tracer = debugfs_create_dir("tracing", NULL);
2107
2108         if (!d_tracer && !once) {
2109                 once = 1;
2110                 pr_warning("Could not create debugfs directory 'tracing'\n");
2111                 return NULL;
2112         }
2113
2114         return d_tracer;
2115 }
2116
2117 #ifdef CONFIG_FTRACE_SELFTEST
2118 /* Let selftest have access to static functions in this file */
2119 #include "trace_selftest.c"
2120 #endif
2121
2122 static __init void tracer_init_debugfs(void)
2123 {
2124         struct dentry *d_tracer;
2125         struct dentry *entry;
2126
2127         d_tracer = tracing_init_dentry();
2128
2129         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2130                                     &global_trace, &tracing_ctrl_fops);
2131         if (!entry)
2132                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2133
2134         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2135                                     NULL, &tracing_iter_fops);
2136         if (!entry)
2137                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2138
2139         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2140                                     &global_trace, &tracing_lt_fops);
2141         if (!entry)
2142                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2143
2144         entry = debugfs_create_file("trace", 0444, d_tracer,
2145                                     &global_trace, &tracing_fops);
2146         if (!entry)
2147                 pr_warning("Could not create debugfs 'trace' entry\n");
2148
2149         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2150                                     &global_trace, &show_traces_fops);
2151         if (!entry)
2152                 pr_warning("Could not create debugfs 'trace' entry\n");
2153
2154         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2155                                     &global_trace, &set_tracer_fops);
2156         if (!entry)
2157                 pr_warning("Could not create debugfs 'trace' entry\n");
2158
2159         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2160                                     &tracing_max_latency,
2161                                     &tracing_max_lat_fops);
2162         if (!entry)
2163                 pr_warning("Could not create debugfs "
2164                            "'tracing_max_latency' entry\n");
2165
2166         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2167                                     &tracing_thresh, &tracing_max_lat_fops);
2168         if (!entry)
2169                 pr_warning("Could not create debugfs "
2170                            "'tracing_threash' entry\n");
2171         entry = debugfs_create_file("README", 0644, d_tracer,
2172                                     NULL, &tracing_readme_fops);
2173         if (!entry)
2174                 pr_warning("Could not create debugfs 'README' entry\n");
2175
2176         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2177                                     NULL, &tracing_pipe_fops);
2178         if (!entry)
2179                 pr_warning("Could not create debugfs "
2180                            "'tracing_threash' entry\n");
2181
2182 #ifdef CONFIG_DYNAMIC_FTRACE
2183         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2184                                     &ftrace_update_tot_cnt,
2185                                     &tracing_read_long_fops);
2186         if (!entry)
2187                 pr_warning("Could not create debugfs "
2188                            "'dyn_ftrace_total_info' entry\n");
2189 #endif
2190 }
2191
2192 /* dummy trace to disable tracing */
2193 static struct tracer no_tracer __read_mostly =
2194 {
2195         .name           = "none",
2196 };
2197
2198 static int trace_alloc_page(void)
2199 {
2200         struct trace_array_cpu *data;
2201         struct page *page, *tmp;
2202         LIST_HEAD(pages);
2203         void *array;
2204         int i;
2205
2206         /* first allocate a page for each CPU */
2207         for_each_possible_cpu(i) {
2208                 array = (void *)__get_free_page(GFP_KERNEL);
2209                 if (array == NULL) {
2210                         printk(KERN_ERR "tracer: failed to allocate page"
2211                                "for trace buffer!\n");
2212                         goto free_pages;
2213                 }
2214
2215                 page = virt_to_page(array);
2216                 list_add(&page->lru, &pages);
2217
2218 /* Only allocate if we are actually using the max trace */
2219 #ifdef CONFIG_TRACER_MAX_TRACE
2220                 array = (void *)__get_free_page(GFP_KERNEL);
2221                 if (array == NULL) {
2222                         printk(KERN_ERR "tracer: failed to allocate page"
2223                                "for trace buffer!\n");
2224                         goto free_pages;
2225                 }
2226                 page = virt_to_page(array);
2227                 list_add(&page->lru, &pages);
2228 #endif
2229         }
2230
2231         /* Now that we successfully allocate a page per CPU, add them */
2232         for_each_possible_cpu(i) {
2233                 data = global_trace.data[i];
2234                 spin_lock_init(&data->lock);
2235                 lockdep_set_class(&data->lock, &data->lock_key);
2236                 page = list_entry(pages.next, struct page, lru);
2237                 list_del_init(&page->lru);
2238                 list_add_tail(&page->lru, &data->trace_pages);
2239                 ClearPageLRU(page);
2240
2241 #ifdef CONFIG_TRACER_MAX_TRACE
2242                 data = max_tr.data[i];
2243                 spin_lock_init(&data->lock);
2244                 lockdep_set_class(&data->lock, &data->lock_key);
2245                 page = list_entry(pages.next, struct page, lru);
2246                 list_del_init(&page->lru);
2247                 list_add_tail(&page->lru, &data->trace_pages);
2248                 SetPageLRU(page);
2249 #endif
2250         }
2251         global_trace.entries += ENTRIES_PER_PAGE;
2252
2253         return 0;
2254
2255  free_pages:
2256         list_for_each_entry_safe(page, tmp, &pages, lru) {
2257                 list_del_init(&page->lru);
2258                 __free_page(page);
2259         }
2260         return -ENOMEM;
2261 }
2262
2263 __init static int tracer_alloc_buffers(void)
2264 {
2265         struct trace_array_cpu *data;
2266         void *array;
2267         struct page *page;
2268         int pages = 0;
2269         int ret = -ENOMEM;
2270         int i;
2271
2272         /* Allocate the first page for all buffers */
2273         for_each_possible_cpu(i) {
2274                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2275                 max_tr.data[i] = &per_cpu(max_data, i);
2276
2277                 array = (void *)__get_free_page(GFP_KERNEL);
2278                 if (array == NULL) {
2279                         printk(KERN_ERR "tracer: failed to allocate page"
2280                                "for trace buffer!\n");
2281                         goto free_buffers;
2282                 }
2283
2284                 /* set the array to the list */
2285                 INIT_LIST_HEAD(&data->trace_pages);
2286                 page = virt_to_page(array);
2287                 list_add(&page->lru, &data->trace_pages);
2288                 /* use the LRU flag to differentiate the two buffers */
2289                 ClearPageLRU(page);
2290
2291 /* Only allocate if we are actually using the max trace */
2292 #ifdef CONFIG_TRACER_MAX_TRACE
2293                 array = (void *)__get_free_page(GFP_KERNEL);
2294                 if (array == NULL) {
2295                         printk(KERN_ERR "tracer: failed to allocate page"
2296                                "for trace buffer!\n");
2297                         goto free_buffers;
2298                 }
2299
2300                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2301                 page = virt_to_page(array);
2302                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2303                 SetPageLRU(page);
2304 #endif
2305         }
2306
2307         /*
2308          * Since we allocate by orders of pages, we may be able to
2309          * round up a bit.
2310          */
2311         global_trace.entries = ENTRIES_PER_PAGE;
2312         pages++;
2313
2314         while (global_trace.entries < trace_nr_entries) {
2315                 if (trace_alloc_page())
2316                         break;
2317                 pages++;
2318         }
2319         max_tr.entries = global_trace.entries;
2320
2321         pr_info("tracer: %d pages allocated for %ld",
2322                 pages, trace_nr_entries);
2323         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2324         pr_info("   actual entries %ld\n", global_trace.entries);
2325
2326         tracer_init_debugfs();
2327
2328         trace_init_cmdlines();
2329
2330         register_tracer(&no_tracer);
2331         current_trace = &no_tracer;
2332
2333         /* All seems OK, enable tracing */
2334         tracing_disabled = 0;
2335
2336         return 0;
2337
2338  free_buffers:
2339         for (i-- ; i >= 0; i--) {
2340                 struct page *page, *tmp;
2341                 struct trace_array_cpu *data = global_trace.data[i];
2342
2343                 if (data) {
2344                         list_for_each_entry_safe(page, tmp,
2345                                                  &data->trace_pages, lru) {
2346                                 list_del_init(&page->lru);
2347                                 __free_page(page);
2348                         }
2349                 }
2350
2351 #ifdef CONFIG_TRACER_MAX_TRACE
2352                 data = max_tr.data[i];
2353                 if (data) {
2354                         list_for_each_entry_safe(page, tmp,
2355                                                  &data->trace_pages, lru) {
2356                                 list_del_init(&page->lru);
2357                                 __free_page(page);
2358                         }
2359                 }
2360 #endif
2361         }
2362         return ret;
2363 }
2364 fs_initcall(tracer_alloc_buffers);