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/poll.h>
  28 #include <linux/gfp.h>
  29 #include <linux/fs.h>
  30
  31 #include "trace.h"
  32
  33 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  34 unsigned long __read_mostly     tracing_thresh;
  35
  36 static int tracing_disabled = 1;
  37
  38 static long notrace
  39 ns2usecs(cycle_t nsec)
  40 {
  41         nsec += 500;
  42         do_div(nsec, 1000);
  43         return nsec;
  44 }
  45
  46 notrace cycle_t ftrace_now(int cpu)
  47 {
  48         return cpu_clock(cpu);
  49 }
  50
  51 static struct trace_array       global_trace;
  52
  53 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  54
  55 static struct trace_array       max_tr;
  56
  57 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  58
  59 static int                      tracer_enabled = 1;
  60 static unsigned long            trace_nr_entries = 16384UL;
  61
  62 static struct tracer            *trace_types __read_mostly;
  63 static struct tracer            *current_trace __read_mostly;
  64 static int                      max_tracer_type_len;
  65
  66 static DEFINE_MUTEX(trace_types_lock);
  67 static DECLARE_WAIT_QUEUE_HEAD (trace_wait);
  68
  69 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  70
  71 static int __init set_nr_entries(char *str)
  72 {
  73         if (!str)
  74                 return 0;
  75         trace_nr_entries = simple_strtoul(str, &str, 0);
  76         return 1;
  77 }
  78 __setup("trace_entries=", set_nr_entries);
  79
  80 unsigned long nsecs_to_usecs(unsigned long nsecs)
  81 {
  82         return nsecs / 1000;
  83 }
  84
  85 enum trace_type {
  86         __TRACE_FIRST_TYPE = 0,
  87
  88         TRACE_FN,
  89         TRACE_CTX,
  90         TRACE_SPECIAL,
  91
  92         __TRACE_LAST_TYPE
  93 };
  94
  95 enum trace_flag_type {
  96         TRACE_FLAG_IRQS_OFF             = 0x01,
  97         TRACE_FLAG_NEED_RESCHED         = 0x02,
  98         TRACE_FLAG_HARDIRQ              = 0x04,
  99         TRACE_FLAG_SOFTIRQ              = 0x08,
 100 };
 101
 102 enum trace_iterator_flags {
 103         TRACE_ITER_PRINT_PARENT         = 0x01,
 104         TRACE_ITER_SYM_OFFSET           = 0x02,
 105         TRACE_ITER_SYM_ADDR             = 0x04,
 106         TRACE_ITER_VERBOSE              = 0x08,
 107         TRACE_ITER_RAW                  = 0x10,
 108         TRACE_ITER_HEX                  = 0x20,
 109         TRACE_ITER_BIN                  = 0x40,
 110         TRACE_ITER_BLOCK                = 0x80,
 111 };
 112
 113 #define TRACE_ITER_SYM_MASK \
 114         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 115
 116 /* These must match the bit postions above */
 117 static const char *trace_options[] = {
 118         "print-parent",
 119         "sym-offset",
 120         "sym-addr",
 121         "verbose",
 122         "raw",
 123         "hex",
 124         "bin",
 125         "block",
 126         NULL
 127 };
 128
 129 static unsigned trace_flags;
 130
 131 static DEFINE_SPINLOCK(ftrace_max_lock);
 132
 133 /*
 134  * Copy the new maximum trace into the separate maximum-trace
 135  * structure. (this way the maximum trace is permanently saved,
 136  * for later retrieval via /debugfs/tracing/latency_trace)
 137  */
 138 static notrace void
 139 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 140 {
 141         struct trace_array_cpu *data = tr->data[cpu];
 142
 143         max_tr.cpu = cpu;
 144         max_tr.time_start = data->preempt_timestamp;
 145
 146         data = max_tr.data[cpu];
 147         data->saved_latency = tracing_max_latency;
 148
 149         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 150         data->pid = tsk->pid;
 151         data->uid = tsk->uid;
 152         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 153         data->policy = tsk->policy;
 154         data->rt_priority = tsk->rt_priority;
 155
 156         /* record this tasks comm */
 157         tracing_record_cmdline(current);
 158 }
 159
 160 void check_pages(struct trace_array_cpu *data)
 161 {
 162         struct page *page, *tmp;
 163
 164         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 165         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 166
 167         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 168                 BUG_ON(page->lru.next->prev != &page->lru);
 169                 BUG_ON(page->lru.prev->next != &page->lru);
 170         }
 171 }
 172
 173 void *head_page(struct trace_array_cpu *data)
 174 {
 175         struct page *page;
 176
 177         check_pages(data);
 178         if (list_empty(&data->trace_pages))
 179                 return NULL;
 180
 181         page = list_entry(data->trace_pages.next, struct page, lru);
 182         BUG_ON(&page->lru == &data->trace_pages);
 183
 184         return page_address(page);
 185 }
 186
 187 static notrace int
 188 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 189 {
 190         int len = (PAGE_SIZE - 1) - s->len;
 191         va_list ap;
 192         int ret;
 193
 194         if (!len)
 195                 return 0;
 196
 197         va_start(ap, fmt);
 198         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 199         va_end(ap);
 200
 201         /* If we can't write it all, don't bother writing anything */
 202         if (ret > len)
 203                 return 0;
 204
 205         s->len += ret;
 206
 207         return len;
 208 }
 209
 210 static notrace int
 211 trace_seq_puts(struct trace_seq *s, const char *str)
 212 {
 213         int len = strlen(str);
 214
 215         if (len > ((PAGE_SIZE - 1) - s->len))
 216                 return 0;
 217
 218         memcpy(s->buffer + s->len, str, len);
 219         s->len += len;
 220
 221         return len;
 222 }
 223
 224 static notrace int
 225 trace_seq_putc(struct trace_seq *s, unsigned char c)
 226 {
 227         if (s->len >= (PAGE_SIZE - 1))
 228                 return 0;
 229
 230         s->buffer[s->len++] = c;
 231
 232         return 1;
 233 }
 234
 235 static notrace int
 236 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 237 {
 238         if (len > ((PAGE_SIZE - 1) - s->len))
 239                 return 0;
 240
 241         memcpy(s->buffer + s->len, mem, len);
 242         s->len += len;
 243
 244         return len;
 245 }
 246
 247 #define HEX_CHARS 17
 248
 249 static notrace int
 250 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 251 {
 252         unsigned char hex[HEX_CHARS];
 253         unsigned char *data;
 254         unsigned char byte;
 255         int i, j;
 256
 257         BUG_ON(len >= HEX_CHARS);
 258
 259         data = mem;
 260
 261 #ifdef __BIG_ENDIAN
 262         for (i = 0, j = 0; i < len; i++) {
 263 #else
 264         for (i = len-1, j = 0; i >= 0; i--) {
 265 #endif
 266                 byte = data[i];
 267
 268                 hex[j]   = byte & 0x0f;
 269                 if (hex[j] >= 10)
 270                         hex[j] += 'a' - 10;
 271                 else
 272                         hex[j] += '0';
 273                 j++;
 274
 275                 hex[j] = byte >> 4;
 276                 if (hex[j] >= 10)
 277                         hex[j] += 'a' - 10;
 278                 else
 279                         hex[j] += '0';
 280                 j++;
 281         }
 282         hex[j] = ' ';
 283         j++;
 284
 285         return trace_seq_putmem(s, hex, j);
 286 }
 287
 288 static notrace void
 289 trace_seq_reset(struct trace_seq *s)
 290 {
 291         s->len = 0;
 292 }
 293
 294 static notrace void
 295 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 296 {
 297         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 298
 299         s->buffer[len] = 0;
 300         seq_puts(m, s->buffer);
 301
 302         trace_seq_reset(s);
 303 }
 304
 305 notrace static void
 306 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 307 {
 308         struct list_head flip_pages;
 309
 310         INIT_LIST_HEAD(&flip_pages);
 311
 312         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 313                 sizeof(struct trace_array_cpu) -
 314                 offsetof(struct trace_array_cpu, trace_head_idx));
 315
 316         check_pages(tr1);
 317         check_pages(tr2);
 318         list_splice_init(&tr1->trace_pages, &flip_pages);
 319         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 320         list_splice_init(&flip_pages, &tr2->trace_pages);
 321         BUG_ON(!list_empty(&flip_pages));
 322         check_pages(tr1);
 323         check_pages(tr2);
 324 }
 325
 326 notrace void
 327 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 328 {
 329         struct trace_array_cpu *data;
 330         int i;
 331
 332         WARN_ON_ONCE(!irqs_disabled());
 333         spin_lock(&ftrace_max_lock);
 334         /* clear out all the previous traces */
 335         for_each_possible_cpu(i) {
 336                 data = tr->data[i];
 337                 flip_trace(max_tr.data[i], data);
 338                 tracing_reset(data);
 339         }
 340
 341         __update_max_tr(tr, tsk, cpu);
 342         spin_unlock(&ftrace_max_lock);
 343 }
 344
 345 /**
 346  * update_max_tr_single - only copy one trace over, and reset the rest
 347  * @tr - tracer
 348  * @tsk - task with the latency
 349  * @cpu - the cpu of the buffer to copy.
 350  */
 351 notrace void
 352 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 353 {
 354         struct trace_array_cpu *data = tr->data[cpu];
 355         int i;
 356
 357         WARN_ON_ONCE(!irqs_disabled());
 358         spin_lock(&ftrace_max_lock);
 359         for_each_possible_cpu(i)
 360                 tracing_reset(max_tr.data[i]);
 361
 362         flip_trace(max_tr.data[cpu], data);
 363         tracing_reset(data);
 364
 365         __update_max_tr(tr, tsk, cpu);
 366         spin_unlock(&ftrace_max_lock);
 367 }
 368
 369 int register_tracer(struct tracer *type)
 370 {
 371         struct tracer *t;
 372         int len;
 373         int ret = 0;
 374
 375         if (!type->name) {
 376                 pr_info("Tracer must have a name\n");
 377                 return -1;
 378         }
 379
 380         mutex_lock(&trace_types_lock);
 381         for (t = trace_types; t; t = t->next) {
 382                 if (strcmp(type->name, t->name) == 0) {
 383                         /* already found */
 384                         pr_info("Trace %s already registered\n",
 385                                 type->name);
 386                         ret = -1;
 387                         goto out;
 388                 }
 389         }
 390
 391 #ifdef CONFIG_FTRACE_STARTUP_TEST
 392         if (type->selftest) {
 393                 struct tracer *saved_tracer = current_trace;
 394                 struct trace_array_cpu *data;
 395                 struct trace_array *tr = &global_trace;
 396                 int saved_ctrl = tr->ctrl;
 397                 int i;
 398                 /*
 399                  * Run a selftest on this tracer.
 400                  * Here we reset the trace buffer, and set the current
 401                  * tracer to be this tracer. The tracer can then run some
 402                  * internal tracing to verify that everything is in order.
 403                  * If we fail, we do not register this tracer.
 404                  */
 405                 for_each_possible_cpu(i) {
 406                         data = tr->data[i];
 407                         if (!head_page(data))
 408                                 continue;
 409                         tracing_reset(data);
 410                 }
 411                 current_trace = type;
 412                 tr->ctrl = 0;
 413                 /* the test is responsible for initializing and enabling */
 414                 pr_info("Testing tracer %s: ", type->name);
 415                 ret = type->selftest(type, tr);
 416                 /* the test is responsible for resetting too */
 417                 current_trace = saved_tracer;
 418                 tr->ctrl = saved_ctrl;
 419                 if (ret) {
 420                         printk(KERN_CONT "FAILED!\n");
 421                         goto out;
 422                 }
 423                 /* Only reset on passing, to avoid touching corrupted buffers */
 424                 for_each_possible_cpu(i) {
 425                         data = tr->data[i];
 426                         if (!head_page(data))
 427                                 continue;
 428                         tracing_reset(data);
 429                 }
 430                 printk(KERN_CONT "PASSED\n");
 431         }
 432 #endif
 433
 434         type->next = trace_types;
 435         trace_types = type;
 436         len = strlen(type->name);
 437         if (len > max_tracer_type_len)
 438                 max_tracer_type_len = len;
 439
 440  out:
 441         mutex_unlock(&trace_types_lock);
 442
 443         return ret;
 444 }
 445
 446 void unregister_tracer(struct tracer *type)
 447 {
 448         struct tracer **t;
 449         int len;
 450
 451         mutex_lock(&trace_types_lock);
 452         for (t = &trace_types; *t; t = &(*t)->next) {
 453                 if (*t == type)
 454                         goto found;
 455         }
 456         pr_info("Trace %s not registered\n", type->name);
 457         goto out;
 458
 459  found:
 460         *t = (*t)->next;
 461         if (strlen(type->name) != max_tracer_type_len)
 462                 goto out;
 463
 464         max_tracer_type_len = 0;
 465         for (t = &trace_types; *t; t = &(*t)->next) {
 466                 len = strlen((*t)->name);
 467                 if (len > max_tracer_type_len)
 468                         max_tracer_type_len = len;
 469         }
 470  out:
 471         mutex_unlock(&trace_types_lock);
 472 }
 473
 474 notrace void tracing_reset(struct trace_array_cpu *data)
 475 {
 476         data->trace_idx = 0;
 477         data->trace_head = data->trace_tail = head_page(data);
 478         data->trace_head_idx = 0;
 479         data->trace_tail_idx = 0;
 480 }
 481
 482 #define SAVED_CMDLINES 128
 483 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 484 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 485 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 486 static int cmdline_idx;
 487 static DEFINE_SPINLOCK(trace_cmdline_lock);
 488 atomic_t trace_record_cmdline_disabled;
 489
 490 static void trace_init_cmdlines(void)
 491 {
 492         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 493         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 494         cmdline_idx = 0;
 495 }
 496
 497 notrace void trace_stop_cmdline_recording(void);
 498
 499 static notrace void trace_save_cmdline(struct task_struct *tsk)
 500 {
 501         unsigned map;
 502         unsigned idx;
 503
 504         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 505                 return;
 506
 507         /*
 508          * It's not the end of the world if we don't get
 509          * the lock, but we also don't want to spin
 510          * nor do we want to disable interrupts,
 511          * so if we miss here, then better luck next time.
 512          */
 513         if (!spin_trylock(&trace_cmdline_lock))
 514                 return;
 515
 516         idx = map_pid_to_cmdline[tsk->pid];
 517         if (idx >= SAVED_CMDLINES) {
 518                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 519
 520                 map = map_cmdline_to_pid[idx];
 521                 if (map <= PID_MAX_DEFAULT)
 522                         map_pid_to_cmdline[map] = (unsigned)-1;
 523
 524                 map_pid_to_cmdline[tsk->pid] = idx;
 525
 526                 cmdline_idx = idx;
 527         }
 528
 529         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 530
 531         spin_unlock(&trace_cmdline_lock);
 532 }
 533
 534 static notrace char *trace_find_cmdline(int pid)
 535 {
 536         char *cmdline = "<...>";
 537         unsigned map;
 538
 539         if (!pid)
 540                 return "<idle>";
 541
 542         if (pid > PID_MAX_DEFAULT)
 543                 goto out;
 544
 545         map = map_pid_to_cmdline[pid];
 546         if (map >= SAVED_CMDLINES)
 547                 goto out;
 548
 549         cmdline = saved_cmdlines[map];
 550
 551  out:
 552         return cmdline;
 553 }
 554
 555 notrace void tracing_record_cmdline(struct task_struct *tsk)
 556 {
 557         if (atomic_read(&trace_record_cmdline_disabled))
 558                 return;
 559
 560         trace_save_cmdline(tsk);
 561 }
 562
 563 static inline notrace struct list_head *
 564 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 565 {
 566         /*
 567          * Roundrobin - but skip the head (which is not a real page):
 568          */
 569         next = next->next;
 570         if (unlikely(next == &data->trace_pages))
 571                 next = next->next;
 572         BUG_ON(next == &data->trace_pages);
 573
 574         return next;
 575 }
 576
 577 static inline notrace void *
 578 trace_next_page(struct trace_array_cpu *data, void *addr)
 579 {
 580         struct list_head *next;
 581         struct page *page;
 582
 583         page = virt_to_page(addr);
 584
 585         next = trace_next_list(data, &page->lru);
 586         page = list_entry(next, struct page, lru);
 587
 588         return page_address(page);
 589 }
 590
 591 static inline notrace struct trace_entry *
 592 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 593 {
 594         unsigned long idx, idx_next;
 595         struct trace_entry *entry;
 596
 597         data->trace_idx++;
 598         idx = data->trace_head_idx;
 599         idx_next = idx + 1;
 600
 601         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 602
 603         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 604
 605         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 606                 data->trace_head = trace_next_page(data, data->trace_head);
 607                 idx_next = 0;
 608         }
 609
 610         if (data->trace_head == data->trace_tail &&
 611             idx_next == data->trace_tail_idx) {
 612                 /* overrun */
 613                 data->trace_tail_idx++;
 614                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 615                         data->trace_tail =
 616                                 trace_next_page(data, data->trace_tail);
 617                         data->trace_tail_idx = 0;
 618                 }
 619         }
 620
 621         data->trace_head_idx = idx_next;
 622
 623         return entry;
 624 }
 625
 626 static inline notrace void
 627 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 628 {
 629         struct task_struct *tsk = current;
 630         unsigned long pc;
 631
 632         pc = preempt_count();
 633
 634         entry->preempt_count    = pc & 0xff;
 635         entry->pid              = tsk->pid;
 636         entry->t                = ftrace_now(raw_smp_processor_id());
 637         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 638                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 639                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 640                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 641 }
 642
 643 notrace void
 644 __ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 645          unsigned long ip, unsigned long parent_ip, unsigned long flags)
 646 {
 647         struct trace_entry *entry;
 648         unsigned long irq_flags;
 649
 650         spin_lock_irqsave(&data->lock, irq_flags);
 651         entry                   = tracing_get_trace_entry(tr, data);
 652         tracing_generic_entry_update(entry, flags);
 653         entry->type             = TRACE_FN;
 654         entry->fn.ip            = ip;
 655         entry->fn.parent_ip     = parent_ip;
 656         spin_unlock_irqrestore(&data->lock, irq_flags);
 657
 658         if (!(trace_flags & TRACE_ITER_BLOCK))
 659                 wake_up (&trace_wait);
 660 }
 661
 662 notrace void
 663 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 664        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 665 {
 666         if (likely(!atomic_read(&data->disabled)))
 667                 __ftrace(tr, data, ip, parent_ip, flags);
 668 }
 669
 670 notrace void
 671 trace_special(struct trace_array *tr, struct trace_array_cpu *data,
 672               unsigned long arg1, unsigned long arg2, unsigned long arg3)
 673 {
 674         struct trace_entry *entry;
 675         unsigned long irq_flags;
 676
 677         spin_lock_irqsave(&data->lock, irq_flags);
 678         entry                   = tracing_get_trace_entry(tr, data);
 679         tracing_generic_entry_update(entry, 0);
 680         entry->type             = TRACE_SPECIAL;
 681         entry->special.arg1     = arg1;
 682         entry->special.arg2     = arg2;
 683         entry->special.arg3     = arg3;
 684         spin_unlock_irqrestore(&data->lock, irq_flags);
 685
 686         if (!(trace_flags & TRACE_ITER_BLOCK))
 687                 wake_up (&trace_wait);
 688 }
 689
 690 notrace void
 691 tracing_sched_switch_trace(struct trace_array *tr,
 692                            struct trace_array_cpu *data,
 693                            struct task_struct *prev, struct task_struct *next,
 694                            unsigned long flags)
 695 {
 696         struct trace_entry *entry;
 697         unsigned long irq_flags;
 698
 699         spin_lock_irqsave(&data->lock, irq_flags);
 700         entry                   = tracing_get_trace_entry(tr, data);
 701         tracing_generic_entry_update(entry, flags);
 702         entry->type             = TRACE_CTX;
 703         entry->ctx.prev_pid     = prev->pid;
 704         entry->ctx.prev_prio    = prev->prio;
 705         entry->ctx.prev_state   = prev->state;
 706         entry->ctx.next_pid     = next->pid;
 707         entry->ctx.next_prio    = next->prio;
 708         spin_unlock_irqrestore(&data->lock, irq_flags);
 709
 710         if (!(trace_flags & TRACE_ITER_BLOCK))
 711                 wake_up (&trace_wait);
 712 }
 713
 714 #ifdef CONFIG_FTRACE
 715 static notrace void
 716 function_trace_call(unsigned long ip, unsigned long parent_ip)
 717 {
 718         struct trace_array *tr = &global_trace;
 719         struct trace_array_cpu *data;
 720         unsigned long flags;
 721         long disabled;
 722         int cpu;
 723
 724         if (unlikely(!tracer_enabled))
 725                 return;
 726
 727         local_irq_save(flags);
 728         cpu = raw_smp_processor_id();
 729         data = tr->data[cpu];
 730         disabled = atomic_inc_return(&data->disabled);
 731
 732         if (likely(disabled == 1))
 733                 __ftrace(tr, data, ip, parent_ip, flags);
 734
 735         atomic_dec(&data->disabled);
 736         local_irq_restore(flags);
 737 }
 738
 739 static struct ftrace_ops trace_ops __read_mostly =
 740 {
 741         .func = function_trace_call,
 742 };
 743
 744 notrace void tracing_start_function_trace(void)
 745 {
 746         register_ftrace_function(&trace_ops);
 747 }
 748
 749 notrace void tracing_stop_function_trace(void)
 750 {
 751         unregister_ftrace_function(&trace_ops);
 752 }
 753 #endif
 754
 755 enum trace_file_type {
 756         TRACE_FILE_LAT_FMT      = 1,
 757 };
 758
 759 static struct trace_entry *
 760 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 761                 struct trace_iterator *iter, int cpu)
 762 {
 763         struct page *page;
 764         struct trace_entry *array;
 765
 766         if (iter->next_idx[cpu] >= tr->entries ||
 767             iter->next_idx[cpu] >= data->trace_idx ||
 768             (data->trace_head == data->trace_tail &&
 769              data->trace_head_idx == data->trace_tail_idx))
 770                 return NULL;
 771
 772         if (!iter->next_page[cpu]) {
 773                 /* Initialize the iterator for this cpu trace buffer */
 774                 WARN_ON(!data->trace_tail);
 775                 page = virt_to_page(data->trace_tail);
 776                 iter->next_page[cpu] = &page->lru;
 777                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 778         }
 779
 780         page = list_entry(iter->next_page[cpu], struct page, lru);
 781         BUG_ON(&data->trace_pages == &page->lru);
 782
 783         array = page_address(page);
 784
 785         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 786         return &array[iter->next_page_idx[cpu]];
 787 }
 788
 789 static struct trace_entry * notrace
 790 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 791 {
 792         struct trace_array *tr = iter->tr;
 793         struct trace_entry *ent, *next = NULL;
 794         int next_cpu = -1;
 795         int cpu;
 796
 797         for_each_possible_cpu(cpu) {
 798                 if (!head_page(tr->data[cpu]))
 799                         continue;
 800                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 801                 /*
 802                  * Pick the entry with the smallest timestamp:
 803                  */
 804                 if (ent && (!next || ent->t < next->t)) {
 805                         next = ent;
 806                         next_cpu = cpu;
 807                 }
 808         }
 809
 810         if (ent_cpu)
 811                 *ent_cpu = next_cpu;
 812
 813         return next;
 814 }
 815
 816 static notrace void trace_iterator_increment(struct trace_iterator *iter)
 817 {
 818         iter->idx++;
 819         iter->next_idx[iter->cpu]++;
 820         iter->next_page_idx[iter->cpu]++;
 821
 822         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 823                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 824
 825                 iter->next_page_idx[iter->cpu] = 0;
 826                 iter->next_page[iter->cpu] =
 827                         trace_next_list(data, iter->next_page[iter->cpu]);
 828         }
 829 }
 830
 831 static notrace void trace_consume(struct trace_iterator *iter)
 832 {
 833         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 834
 835         data->trace_tail_idx++;
 836         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 837                 data->trace_tail = trace_next_page(data, data->trace_tail);
 838                 data->trace_tail_idx = 0;
 839         }
 840
 841         /* Check if we empty it, then reset the index */
 842         if (data->trace_head == data->trace_tail &&
 843             data->trace_head_idx == data->trace_tail_idx)
 844                 data->trace_idx = 0;
 845 }
 846
 847 static notrace void *find_next_entry_inc(struct trace_iterator *iter)
 848 {
 849         struct trace_entry *next;
 850         int next_cpu = -1;
 851
 852         next = find_next_entry(iter, &next_cpu);
 853
 854         iter->prev_ent = iter->ent;
 855         iter->prev_cpu = iter->cpu;
 856
 857         iter->ent = next;
 858         iter->cpu = next_cpu;
 859
 860         if (next)
 861                 trace_iterator_increment(iter);
 862
 863         return next ? iter : NULL;
 864 }
 865
 866 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 867 {
 868         struct trace_iterator *iter = m->private;
 869         void *last_ent = iter->ent;
 870         int i = (int)*pos;
 871         void *ent;
 872
 873         (*pos)++;
 874
 875         /* can't go backwards */
 876         if (iter->idx > i)
 877                 return NULL;
 878
 879         if (iter->idx < 0)
 880                 ent = find_next_entry_inc(iter);
 881         else
 882                 ent = iter;
 883
 884         while (ent && iter->idx < i)
 885                 ent = find_next_entry_inc(iter);
 886
 887         iter->pos = *pos;
 888
 889         if (last_ent && !ent)
 890                 seq_puts(m, "\n\nvim:ft=help\n");
 891
 892         return ent;
 893 }
 894
 895 static void *s_start(struct seq_file *m, loff_t *pos)
 896 {
 897         struct trace_iterator *iter = m->private;
 898         void *p = NULL;
 899         loff_t l = 0;
 900         int i;
 901
 902         mutex_lock(&trace_types_lock);
 903
 904         if (!current_trace || current_trace != iter->trace)
 905                 return NULL;
 906
 907         atomic_inc(&trace_record_cmdline_disabled);
 908
 909         /* let the tracer grab locks here if needed */
 910         if (current_trace->start)
 911                 current_trace->start(iter);
 912
 913         if (*pos != iter->pos) {
 914                 iter->ent = NULL;
 915                 iter->cpu = 0;
 916                 iter->idx = -1;
 917                 iter->prev_ent = NULL;
 918                 iter->prev_cpu = -1;
 919
 920                 for_each_possible_cpu(i) {
 921                         iter->next_idx[i] = 0;
 922                         iter->next_page[i] = NULL;
 923                 }
 924
 925                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 926                         ;
 927
 928         } else {
 929                 l = *pos - 1;
 930                 p = s_next(m, p, &l);
 931         }
 932
 933         return p;
 934 }
 935
 936 static void s_stop(struct seq_file *m, void *p)
 937 {
 938         struct trace_iterator *iter = m->private;
 939
 940         atomic_dec(&trace_record_cmdline_disabled);
 941
 942         /* let the tracer release locks here if needed */
 943         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 944                 iter->trace->stop(iter);
 945
 946         mutex_unlock(&trace_types_lock);
 947 }
 948
 949 static int
 950 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 951 {
 952 #ifdef CONFIG_KALLSYMS
 953         char str[KSYM_SYMBOL_LEN];
 954
 955         kallsyms_lookup(address, NULL, NULL, NULL, str);
 956
 957         return trace_seq_printf(s, fmt, str);
 958 #endif
 959         return 1;
 960 }
 961
 962 static int
 963 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 964                      unsigned long address)
 965 {
 966 #ifdef CONFIG_KALLSYMS
 967         char str[KSYM_SYMBOL_LEN];
 968
 969         sprint_symbol(str, address);
 970         return trace_seq_printf(s, fmt, str);
 971 #endif
 972         return 1;
 973 }
 974
 975 #ifndef CONFIG_64BIT
 976 # define IP_FMT "%08lx"
 977 #else
 978 # define IP_FMT "%016lx"
 979 #endif
 980
 981 static notrace int
 982 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 983 {
 984         int ret;
 985
 986         if (!ip)
 987                 return trace_seq_printf(s, "0");
 988
 989         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 990                 ret = seq_print_sym_offset(s, "%s", ip);
 991         else
 992                 ret = seq_print_sym_short(s, "%s", ip);
 993
 994         if (!ret)
 995                 return 0;
 996
 997         if (sym_flags & TRACE_ITER_SYM_ADDR)
 998                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 999         return ret;
1000 }
1001
1002 static notrace void print_lat_help_header(struct seq_file *m)
1003 {
1004         seq_puts(m, "#                _------=> CPU#            \n");
1005         seq_puts(m, "#               / _-----=> irqs-off        \n");
1006         seq_puts(m, "#              | / _----=> need-resched    \n");
1007         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1008         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1009         seq_puts(m, "#              |||| /                      \n");
1010         seq_puts(m, "#              |||||     delay             \n");
1011         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1012         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1013 }
1014
1015 static notrace void print_func_help_header(struct seq_file *m)
1016 {
1017         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1018         seq_puts(m, "#              | |      |          |         |\n");
1019 }
1020
1021
1022 static notrace void
1023 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1024 {
1025         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1026         struct trace_array *tr = iter->tr;
1027         struct trace_array_cpu *data = tr->data[tr->cpu];
1028         struct tracer *type = current_trace;
1029         unsigned long total   = 0;
1030         unsigned long entries = 0;
1031         int cpu;
1032         const char *name = "preemption";
1033
1034         if (type)
1035                 name = type->name;
1036
1037         for_each_possible_cpu(cpu) {
1038                 if (head_page(tr->data[cpu])) {
1039                         total += tr->data[cpu]->trace_idx;
1040                         if (tr->data[cpu]->trace_idx > tr->entries)
1041                                 entries += tr->entries;
1042                         else
1043                                 entries += tr->data[cpu]->trace_idx;
1044                 }
1045         }
1046
1047         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1048                    name, UTS_RELEASE);
1049         seq_puts(m, "-----------------------------------"
1050                  "---------------------------------\n");
1051         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1052                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1053                    nsecs_to_usecs(data->saved_latency),
1054                    entries,
1055                    total,
1056                    tr->cpu,
1057 #if defined(CONFIG_PREEMPT_NONE)
1058                    "server",
1059 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1060                    "desktop",
1061 #elif defined(CONFIG_PREEMPT_DESKTOP)
1062                    "preempt",
1063 #else
1064                    "unknown",
1065 #endif
1066                    /* These are reserved for later use */
1067                    0, 0, 0, 0);
1068 #ifdef CONFIG_SMP
1069         seq_printf(m, " #P:%d)\n", num_online_cpus());
1070 #else
1071         seq_puts(m, ")\n");
1072 #endif
1073         seq_puts(m, "    -----------------\n");
1074         seq_printf(m, "    | task: %.16s-%d "
1075                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1076                    data->comm, data->pid, data->uid, data->nice,
1077                    data->policy, data->rt_priority);
1078         seq_puts(m, "    -----------------\n");
1079
1080         if (data->critical_start) {
1081                 seq_puts(m, " => started at: ");
1082                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1083                 trace_print_seq(m, &iter->seq);
1084                 seq_puts(m, "\n => ended at:   ");
1085                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1086                 trace_print_seq(m, &iter->seq);
1087                 seq_puts(m, "\n");
1088         }
1089
1090         seq_puts(m, "\n");
1091 }
1092
1093 static notrace void
1094 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1095 {
1096         int hardirq, softirq;
1097         char *comm;
1098
1099         comm = trace_find_cmdline(entry->pid);
1100
1101         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1102         trace_seq_printf(s, "%d", cpu);
1103         trace_seq_printf(s, "%c%c",
1104                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1105                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1106
1107         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1108         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1109         if (hardirq && softirq)
1110                 trace_seq_putc(s, 'H');
1111         else {
1112                 if (hardirq)
1113                         trace_seq_putc(s, 'h');
1114                 else {
1115                         if (softirq)
1116                                 trace_seq_putc(s, 's');
1117                         else
1118                                 trace_seq_putc(s, '.');
1119                 }
1120         }
1121
1122         if (entry->preempt_count)
1123                 trace_seq_printf(s, "%x", entry->preempt_count);
1124         else
1125                 trace_seq_puts(s, ".");
1126 }
1127
1128 unsigned long preempt_mark_thresh = 100;
1129
1130 static notrace void
1131 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1132                     unsigned long rel_usecs)
1133 {
1134         trace_seq_printf(s, " %4lldus", abs_usecs);
1135         if (rel_usecs > preempt_mark_thresh)
1136                 trace_seq_puts(s, "!: ");
1137         else if (rel_usecs > 1)
1138                 trace_seq_puts(s, "+: ");
1139         else
1140                 trace_seq_puts(s, " : ");
1141 }
1142
1143 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1144
1145 static notrace int
1146 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1147 {
1148         struct trace_seq *s = &iter->seq;
1149         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1150         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1151         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1152         struct trace_entry *entry = iter->ent;
1153         unsigned long abs_usecs;
1154         unsigned long rel_usecs;
1155         char *comm;
1156         int S;
1157
1158         if (!next_entry)
1159                 next_entry = entry;
1160         rel_usecs = ns2usecs(next_entry->t - entry->t);
1161         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1162
1163         if (verbose) {
1164                 comm = trace_find_cmdline(entry->pid);
1165                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1166                                  " %ld.%03ldms (+%ld.%03ldms): ",
1167                                  comm,
1168                                  entry->pid, cpu, entry->flags,
1169                                  entry->preempt_count, trace_idx,
1170                                  ns2usecs(entry->t),
1171                                  abs_usecs/1000,
1172                                  abs_usecs % 1000, rel_usecs/1000,
1173                                  rel_usecs % 1000);
1174         } else {
1175                 lat_print_generic(s, entry, cpu);
1176                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1177         }
1178         switch (entry->type) {
1179         case TRACE_FN:
1180                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1181                 trace_seq_puts(s, " (");
1182                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1183                 trace_seq_puts(s, ")\n");
1184                 break;
1185         case TRACE_CTX:
1186                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1187                         state_to_char[entry->ctx.prev_state] : 'X';
1188                 comm = trace_find_cmdline(entry->ctx.next_pid);
1189                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1190                                  entry->ctx.prev_pid,
1191                                  entry->ctx.prev_prio,
1192                                  S,
1193                                  entry->ctx.next_pid,
1194                                  entry->ctx.next_prio,
1195                                  comm);
1196                 break;
1197         case TRACE_SPECIAL:
1198                 trace_seq_printf(s, " %lx %lx %lx\n",
1199                                  entry->special.arg1,
1200                                  entry->special.arg2,
1201                                  entry->special.arg3);
1202                 break;
1203         default:
1204                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1205         }
1206         return 1;
1207 }
1208
1209 static notrace int print_trace_fmt(struct trace_iterator *iter)
1210 {
1211         struct trace_seq *s = &iter->seq;
1212         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1213         struct trace_entry *entry;
1214         unsigned long usec_rem;
1215         unsigned long long t;
1216         unsigned long secs;
1217         char *comm;
1218         int S;
1219         int ret;
1220
1221         entry = iter->ent;
1222
1223         comm = trace_find_cmdline(iter->ent->pid);
1224
1225         t = ns2usecs(entry->t);
1226         usec_rem = do_div(t, 1000000ULL);
1227         secs = (unsigned long)t;
1228
1229         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1230         if (!ret)
1231                 return 0;
1232         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1233         if (!ret)
1234                 return 0;
1235         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1236         if (!ret)
1237                 return 0;
1238
1239         switch (entry->type) {
1240         case TRACE_FN:
1241                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1242                 if (!ret)
1243                         return 0;
1244                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1245                                                 entry->fn.parent_ip) {
1246                         ret = trace_seq_printf(s, " <-");
1247                         if (!ret)
1248                                 return 0;
1249                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1250                                                sym_flags);
1251                         if (!ret)
1252                                 return 0;
1253                 }
1254                 ret = trace_seq_printf(s, "\n");
1255                 if (!ret)
1256                         return 0;
1257                 break;
1258         case TRACE_CTX:
1259                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1260                         state_to_char[entry->ctx.prev_state] : 'X';
1261                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1262                                        entry->ctx.prev_pid,
1263                                        entry->ctx.prev_prio,
1264                                        S,
1265                                        entry->ctx.next_pid,
1266                                        entry->ctx.next_prio);
1267                 if (!ret)
1268                         return 0;
1269                 break;
1270         case TRACE_SPECIAL:
1271                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1272                                  entry->special.arg1,
1273                                  entry->special.arg2,
1274                                  entry->special.arg3);
1275                 if (!ret)
1276                         return 0;
1277                 break;
1278         }
1279         return 1;
1280 }
1281
1282 static notrace int print_raw_fmt(struct trace_iterator *iter)
1283 {
1284         struct trace_seq *s = &iter->seq;
1285         struct trace_entry *entry;
1286         int ret;
1287         int S;
1288
1289         entry = iter->ent;
1290
1291         ret = trace_seq_printf(s, "%d %d %llu ",
1292                 entry->pid, iter->cpu, entry->t);
1293         if (!ret)
1294                 return 0;
1295
1296         switch (entry->type) {
1297         case TRACE_FN:
1298                 ret = trace_seq_printf(s, "%x %x\n",
1299                                         entry->fn.ip, entry->fn.parent_ip);
1300                 if (!ret)
1301                         return 0;
1302                 break;
1303         case TRACE_CTX:
1304                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1305                         state_to_char[entry->ctx.prev_state] : 'X';
1306                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1307                                        entry->ctx.prev_pid,
1308                                        entry->ctx.prev_prio,
1309                                        S,
1310                                        entry->ctx.next_pid,
1311                                        entry->ctx.next_prio);
1312                 if (!ret)
1313                         return 0;
1314                 break;
1315         case TRACE_SPECIAL:
1316                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1317                                  entry->special.arg1,
1318                                  entry->special.arg2,
1319                                  entry->special.arg3);
1320                 if (!ret)
1321                         return 0;
1322                 break;
1323         }
1324         return 1;
1325 }
1326
1327 #define SEQ_PUT_FIELD_RET(s, x)                         \
1328 do {                                                    \
1329         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1330                 return 0;                               \
1331 } while (0)
1332
1333 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1334 do {                                                    \
1335         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1336                 return 0;                               \
1337 } while (0)
1338
1339 static notrace int print_hex_fmt(struct trace_iterator *iter)
1340 {
1341         struct trace_seq *s = &iter->seq;
1342         unsigned char newline = '\n';
1343         struct trace_entry *entry;
1344         int S;
1345
1346         entry = iter->ent;
1347
1348         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1349         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1350         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1351
1352         switch (entry->type) {
1353         case TRACE_FN:
1354                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1355                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1356                 break;
1357         case TRACE_CTX:
1358                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1359                         state_to_char[entry->ctx.prev_state] : 'X';
1360                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1361                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1362                 SEQ_PUT_HEX_FIELD_RET(s, S);
1363                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1364                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1365                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1366                 break;
1367         case TRACE_SPECIAL:
1368                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1369                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1370                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1371                 break;
1372         }
1373         SEQ_PUT_FIELD_RET(s, newline);
1374
1375         return 1;
1376 }
1377
1378 static notrace int print_bin_fmt(struct trace_iterator *iter)
1379 {
1380         struct trace_seq *s = &iter->seq;
1381         struct trace_entry *entry;
1382
1383         entry = iter->ent;
1384
1385         SEQ_PUT_FIELD_RET(s, entry->pid);
1386         SEQ_PUT_FIELD_RET(s, entry->cpu);
1387         SEQ_PUT_FIELD_RET(s, entry->t);
1388
1389         switch (entry->type) {
1390         case TRACE_FN:
1391                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1392                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1393                 break;
1394         case TRACE_CTX:
1395                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1396                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1397                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1398                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1399                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1400                 break;
1401         case TRACE_SPECIAL:
1402                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1403                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1404                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1405                 break;
1406         }
1407         return 1;
1408 }
1409
1410 static int trace_empty(struct trace_iterator *iter)
1411 {
1412         struct trace_array_cpu *data;
1413         int cpu;
1414
1415         for_each_possible_cpu(cpu) {
1416                 data = iter->tr->data[cpu];
1417
1418                 if (head_page(data) && data->trace_idx &&
1419                     (data->trace_tail != data->trace_head ||
1420                      data->trace_tail_idx != data->trace_head_idx))
1421                         return 0;
1422         }
1423         return 1;
1424 }
1425
1426 static int print_trace_line(struct trace_iterator *iter)
1427 {
1428         if (trace_flags & TRACE_ITER_BIN)
1429                 return print_bin_fmt(iter);
1430
1431         if (trace_flags & TRACE_ITER_HEX)
1432                 return print_hex_fmt(iter);
1433
1434         if (trace_flags & TRACE_ITER_RAW)
1435                 return print_raw_fmt(iter);
1436
1437         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1438                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1439
1440         return print_trace_fmt(iter);
1441 }
1442
1443 static int s_show(struct seq_file *m, void *v)
1444 {
1445         struct trace_iterator *iter = v;
1446
1447         if (iter->ent == NULL) {
1448                 if (iter->tr) {
1449                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1450                         seq_puts(m, "#\n");
1451                 }
1452                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1453                         /* print nothing if the buffers are empty */
1454                         if (trace_empty(iter))
1455                                 return 0;
1456                         print_trace_header(m, iter);
1457                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1458                                 print_lat_help_header(m);
1459                 } else {
1460                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1461                                 print_func_help_header(m);
1462                 }
1463         } else {
1464                 print_trace_line(iter);
1465                 trace_print_seq(m, &iter->seq);
1466         }
1467
1468         return 0;
1469 }
1470
1471 static struct seq_operations tracer_seq_ops = {
1472         .start          = s_start,
1473         .next           = s_next,
1474         .stop           = s_stop,
1475         .show           = s_show,
1476 };
1477
1478 static struct trace_iterator notrace *
1479 __tracing_open(struct inode *inode, struct file *file, int *ret)
1480 {
1481         struct trace_iterator *iter;
1482
1483         if (tracing_disabled) {
1484                 *ret = -ENODEV;
1485                 return NULL;
1486         }
1487
1488         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1489         if (!iter) {
1490                 *ret = -ENOMEM;
1491                 goto out;
1492         }
1493
1494         mutex_lock(&trace_types_lock);
1495         if (current_trace && current_trace->print_max)
1496                 iter->tr = &max_tr;
1497         else
1498                 iter->tr = inode->i_private;
1499         iter->trace = current_trace;
1500         iter->pos = -1;
1501
1502         /* TODO stop tracer */
1503         *ret = seq_open(file, &tracer_seq_ops);
1504         if (!*ret) {
1505                 struct seq_file *m = file->private_data;
1506                 m->private = iter;
1507
1508                 /* stop the trace while dumping */
1509                 if (iter->tr->ctrl)
1510                         tracer_enabled = 0;
1511
1512                 if (iter->trace && iter->trace->open)
1513                         iter->trace->open(iter);
1514         } else {
1515                 kfree(iter);
1516                 iter = NULL;
1517         }
1518         mutex_unlock(&trace_types_lock);
1519
1520  out:
1521         return iter;
1522 }
1523
1524 int tracing_open_generic(struct inode *inode, struct file *filp)
1525 {
1526         if (tracing_disabled)
1527                 return -ENODEV;
1528
1529         filp->private_data = inode->i_private;
1530         return 0;
1531 }
1532
1533 int tracing_release(struct inode *inode, struct file *file)
1534 {
1535         struct seq_file *m = (struct seq_file *)file->private_data;
1536         struct trace_iterator *iter = m->private;
1537
1538         mutex_lock(&trace_types_lock);
1539         if (iter->trace && iter->trace->close)
1540                 iter->trace->close(iter);
1541
1542         /* reenable tracing if it was previously enabled */
1543         if (iter->tr->ctrl)
1544                 tracer_enabled = 1;
1545         mutex_unlock(&trace_types_lock);
1546
1547         seq_release(inode, file);
1548         kfree(iter);
1549         return 0;
1550 }
1551
1552 static int tracing_open(struct inode *inode, struct file *file)
1553 {
1554         int ret;
1555
1556         __tracing_open(inode, file, &ret);
1557
1558         return ret;
1559 }
1560
1561 static int tracing_lt_open(struct inode *inode, struct file *file)
1562 {
1563         struct trace_iterator *iter;
1564         int ret;
1565
1566         iter = __tracing_open(inode, file, &ret);
1567
1568         if (!ret)
1569                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1570
1571         return ret;
1572 }
1573
1574
1575 static notrace void *
1576 t_next(struct seq_file *m, void *v, loff_t *pos)
1577 {
1578         struct tracer *t = m->private;
1579
1580         (*pos)++;
1581
1582         if (t)
1583                 t = t->next;
1584
1585         m->private = t;
1586
1587         return t;
1588 }
1589
1590 static void *t_start(struct seq_file *m, loff_t *pos)
1591 {
1592         struct tracer *t = m->private;
1593         loff_t l = 0;
1594
1595         mutex_lock(&trace_types_lock);
1596         for (; t && l < *pos; t = t_next(m, t, &l))
1597                 ;
1598
1599         return t;
1600 }
1601
1602 static void t_stop(struct seq_file *m, void *p)
1603 {
1604         mutex_unlock(&trace_types_lock);
1605 }
1606
1607 static int t_show(struct seq_file *m, void *v)
1608 {
1609         struct tracer *t = v;
1610
1611         if (!t)
1612                 return 0;
1613
1614         seq_printf(m, "%s", t->name);
1615         if (t->next)
1616                 seq_putc(m, ' ');
1617         else
1618                 seq_putc(m, '\n');
1619
1620         return 0;
1621 }
1622
1623 static struct seq_operations show_traces_seq_ops = {
1624         .start          = t_start,
1625         .next           = t_next,
1626         .stop           = t_stop,
1627         .show           = t_show,
1628 };
1629
1630 static int show_traces_open(struct inode *inode, struct file *file)
1631 {
1632         int ret;
1633
1634         if (tracing_disabled)
1635                 return -ENODEV;
1636
1637         ret = seq_open(file, &show_traces_seq_ops);
1638         if (!ret) {
1639                 struct seq_file *m = file->private_data;
1640                 m->private = trace_types;
1641         }
1642
1643         return ret;
1644 }
1645
1646 static struct file_operations tracing_fops = {
1647         .open           = tracing_open,
1648         .read           = seq_read,
1649         .llseek         = seq_lseek,
1650         .release        = tracing_release,
1651 };
1652
1653 static struct file_operations tracing_lt_fops = {
1654         .open           = tracing_lt_open,
1655         .read           = seq_read,
1656         .llseek         = seq_lseek,
1657         .release        = tracing_release,
1658 };
1659
1660 static struct file_operations show_traces_fops = {
1661         .open = show_traces_open,
1662         .read = seq_read,
1663         .release = seq_release,
1664 };
1665
1666 static ssize_t
1667 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1668                        size_t cnt, loff_t *ppos)
1669 {
1670         char *buf;
1671         int r = 0;
1672         int len = 0;
1673         int i;
1674
1675         /* calulate max size */
1676         for (i = 0; trace_options[i]; i++) {
1677                 len += strlen(trace_options[i]);
1678                 len += 3; /* "no" and space */
1679         }
1680
1681         /* +2 for \n and \0 */
1682         buf = kmalloc(len + 2, GFP_KERNEL);
1683         if (!buf)
1684                 return -ENOMEM;
1685
1686         for (i = 0; trace_options[i]; i++) {
1687                 if (trace_flags & (1 << i))
1688                         r += sprintf(buf + r, "%s ", trace_options[i]);
1689                 else
1690                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1691         }
1692
1693         r += sprintf(buf + r, "\n");
1694         WARN_ON(r >= len + 2);
1695
1696         r = simple_read_from_buffer(ubuf, cnt, ppos,
1697                                     buf, r);
1698
1699         kfree(buf);
1700
1701         return r;
1702 }
1703
1704 static ssize_t
1705 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1706                         size_t cnt, loff_t *ppos)
1707 {
1708         char buf[64];
1709         char *cmp = buf;
1710         int neg = 0;
1711         int i;
1712
1713         if (cnt > 63)
1714                 cnt = 63;
1715
1716         if (copy_from_user(&buf, ubuf, cnt))
1717                 return -EFAULT;
1718
1719         buf[cnt] = 0;
1720
1721         if (strncmp(buf, "no", 2) == 0) {
1722                 neg = 1;
1723                 cmp += 2;
1724         }
1725
1726         for (i = 0; trace_options[i]; i++) {
1727                 int len = strlen(trace_options[i]);
1728
1729                 if (strncmp(cmp, trace_options[i], len) == 0) {
1730                         if (neg)
1731                                 trace_flags &= ~(1 << i);
1732                         else
1733                                 trace_flags |= (1 << i);
1734                         break;
1735                 }
1736         }
1737
1738         filp->f_pos += cnt;
1739
1740         return cnt;
1741 }
1742
1743 static struct file_operations tracing_iter_fops = {
1744         .open = tracing_open_generic,
1745         .read = tracing_iter_ctrl_read,
1746         .write = tracing_iter_ctrl_write,
1747 };
1748
1749 static const char readme_msg[] =
1750         "tracing mini-HOWTO:\n\n"
1751         "# mkdir /debug\n"
1752         "# mount -t debugfs nodev /debug\n\n"
1753         "# cat /debug/tracing/available_tracers\n"
1754         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1755         "# cat /debug/tracing/current_tracer\n"
1756         "none\n"
1757         "# echo sched_switch > /debug/tracing/current_tracer\n"
1758         "# cat /debug/tracing/current_tracer\n"
1759         "sched_switch\n"
1760         "# cat /debug/tracing/iter_ctrl\n"
1761         "noprint-parent nosym-offset nosym-addr noverbose\n"
1762         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1763         "# echo 1 > /debug/tracing/tracing_enabled\n"
1764         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1765         "echo 0 > /debug/tracing/tracing_enabled\n"
1766 ;
1767
1768 static ssize_t
1769 tracing_readme_read(struct file *filp, char __user *ubuf,
1770                        size_t cnt, loff_t *ppos)
1771 {
1772         return simple_read_from_buffer(ubuf, cnt, ppos,
1773                                         readme_msg, strlen(readme_msg));
1774 }
1775
1776 static struct file_operations tracing_readme_fops = {
1777         .open = tracing_open_generic,
1778         .read = tracing_readme_read,
1779 };
1780
1781 static ssize_t
1782 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1783                   size_t cnt, loff_t *ppos)
1784 {
1785         struct trace_array *tr = filp->private_data;
1786         char buf[64];
1787         int r;
1788
1789         r = sprintf(buf, "%ld\n", tr->ctrl);
1790         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1791 }
1792
1793 static ssize_t
1794 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1795                    size_t cnt, loff_t *ppos)
1796 {
1797         struct trace_array *tr = filp->private_data;
1798         long val;
1799         char buf[64];
1800
1801         if (cnt > 63)
1802                 cnt = 63;
1803
1804         if (copy_from_user(&buf, ubuf, cnt))
1805                 return -EFAULT;
1806
1807         buf[cnt] = 0;
1808
1809         val = simple_strtoul(buf, NULL, 10);
1810
1811         val = !!val;
1812
1813         mutex_lock(&trace_types_lock);
1814         if (tr->ctrl ^ val) {
1815                 if (val)
1816                         tracer_enabled = 1;
1817                 else
1818                         tracer_enabled = 0;
1819
1820                 tr->ctrl = val;
1821
1822                 if (current_trace && current_trace->ctrl_update)
1823                         current_trace->ctrl_update(tr);
1824         }
1825         mutex_unlock(&trace_types_lock);
1826
1827         filp->f_pos += cnt;
1828
1829         return cnt;
1830 }
1831
1832 static ssize_t
1833 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1834                        size_t cnt, loff_t *ppos)
1835 {
1836         char buf[max_tracer_type_len+2];
1837         int r;
1838
1839         mutex_lock(&trace_types_lock);
1840         if (current_trace)
1841                 r = sprintf(buf, "%s\n", current_trace->name);
1842         else
1843                 r = sprintf(buf, "\n");
1844         mutex_unlock(&trace_types_lock);
1845
1846         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1847 }
1848
1849 static ssize_t
1850 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1851                         size_t cnt, loff_t *ppos)
1852 {
1853         struct trace_array *tr = &global_trace;
1854         struct tracer *t;
1855         char buf[max_tracer_type_len+1];
1856         int i;
1857
1858         if (cnt > max_tracer_type_len)
1859                 cnt = max_tracer_type_len;
1860
1861         if (copy_from_user(&buf, ubuf, cnt))
1862                 return -EFAULT;
1863
1864         buf[cnt] = 0;
1865
1866         /* strip ending whitespace. */
1867         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1868                 buf[i] = 0;
1869
1870         mutex_lock(&trace_types_lock);
1871         for (t = trace_types; t; t = t->next) {
1872                 if (strcmp(t->name, buf) == 0)
1873                         break;
1874         }
1875         if (!t || t == current_trace)
1876                 goto out;
1877
1878         if (current_trace && current_trace->reset)
1879                 current_trace->reset(tr);
1880
1881         current_trace = t;
1882         if (t->init)
1883                 t->init(tr);
1884
1885  out:
1886         mutex_unlock(&trace_types_lock);
1887
1888         filp->f_pos += cnt;
1889
1890         return cnt;
1891 }
1892
1893 static ssize_t
1894 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1895                      size_t cnt, loff_t *ppos)
1896 {
1897         unsigned long *ptr = filp->private_data;
1898         char buf[64];
1899         int r;
1900
1901         r = snprintf(buf, 64, "%ld\n",
1902                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1903         if (r > 64)
1904                 r = 64;
1905         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1906 }
1907
1908 static ssize_t
1909 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1910                       size_t cnt, loff_t *ppos)
1911 {
1912         long *ptr = filp->private_data;
1913         long val;
1914         char buf[64];
1915
1916         if (cnt > 63)
1917                 cnt = 63;
1918
1919         if (copy_from_user(&buf, ubuf, cnt))
1920                 return -EFAULT;
1921
1922         buf[cnt] = 0;
1923
1924         val = simple_strtoul(buf, NULL, 10);
1925
1926         *ptr = val * 1000;
1927
1928         return cnt;
1929 }
1930
1931 static atomic_t tracing_reader;
1932
1933 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1934 {
1935         struct trace_iterator *iter;
1936
1937         if (tracing_disabled)
1938                 return -ENODEV;
1939
1940         /* We only allow for reader of the pipe */
1941         if (atomic_inc_return(&tracing_reader) != 1) {
1942                 atomic_dec(&tracing_reader);
1943                 return -EBUSY;
1944         }
1945
1946         /* create a buffer to store the information to pass to userspace */
1947         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1948         if (!iter)
1949                 return -ENOMEM;
1950
1951         iter->tr = &global_trace;
1952
1953         filp->private_data = iter;
1954
1955         return 0;
1956 }
1957
1958 static int tracing_release_pipe(struct inode *inode, struct file *file)
1959 {
1960         struct trace_iterator *iter = file->private_data;
1961
1962         kfree(iter);
1963         atomic_dec(&tracing_reader);
1964
1965         return 0;
1966 }
1967
1968 static unsigned int
1969 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
1970 {
1971         struct trace_iterator *iter = filp->private_data;
1972
1973         if (trace_flags & TRACE_ITER_BLOCK) {
1974                 /*
1975                  * Always select as readable when in blocking mode
1976                  */
1977                 return POLLIN | POLLRDNORM;
1978         }
1979         else {
1980                 if (!trace_empty(iter))
1981                         return POLLIN | POLLRDNORM;
1982                 poll_wait(filp, &trace_wait, poll_table);
1983                 if (!trace_empty(iter))
1984                         return POLLIN | POLLRDNORM;
1985
1986                 return 0;
1987         }
1988 }
1989
1990 /*
1991  * Consumer reader.
1992  */
1993 static ssize_t
1994 tracing_read_pipe(struct file *filp, char __user *ubuf,
1995                   size_t cnt, loff_t *ppos)
1996 {
1997         struct trace_iterator *iter = filp->private_data;
1998         struct trace_array_cpu *data;
1999         static cpumask_t mask;
2000         static int start;
2001         unsigned long flags;
2002 #ifdef CONFIG_FTRACE
2003         int ftrace_save;
2004 #endif
2005         int read = 0;
2006         int cpu;
2007         int len;
2008         int ret;
2009
2010         /* return any leftover data */
2011         if (iter->seq.len > start) {
2012                 len = iter->seq.len - start;
2013                 if (cnt > len)
2014                         cnt = len;
2015                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2016                 if (ret)
2017                         cnt = -EFAULT;
2018
2019                 start += len;
2020
2021                 return cnt;
2022         }
2023
2024         trace_seq_reset(&iter->seq);
2025         start = 0;
2026
2027         while (trace_empty(iter)) {
2028                 if (!(trace_flags & TRACE_ITER_BLOCK))
2029                         return -EWOULDBLOCK;
2030                 /*
2031                  * This is a make-shift waitqueue. The reason we don't use
2032                  * an actual wait queue is because:
2033                  *  1) we only ever have one waiter
2034                  *  2) the tracing, traces all functions, we don't want
2035                  *     the overhead of calling wake_up and friends
2036                  *     (and tracing them too)
2037                  *     Anyway, this is really very primitive wakeup.
2038                  */
2039                 set_current_state(TASK_INTERRUPTIBLE);
2040                 iter->tr->waiter = current;
2041
2042                 /* sleep for one second, and try again. */
2043                 schedule_timeout(HZ);
2044
2045                 iter->tr->waiter = NULL;
2046
2047                 if (signal_pending(current))
2048                         return -EINTR;
2049
2050                 /*
2051                  * We block until we read something and tracing is disabled.
2052                  * We still block if tracing is disabled, but we have never
2053                  * read anything. This allows a user to cat this file, and
2054                  * then enable tracing. But after we have read something,
2055                  * we give an EOF when tracing is again disabled.
2056                  *
2057                  * iter->pos will be 0 if we haven't read anything.
2058                  */
2059                 if (!tracer_enabled && iter->pos)
2060                         break;
2061
2062                 continue;
2063         }
2064
2065         /* stop when tracing is finished */
2066         if (trace_empty(iter))
2067                 return 0;
2068
2069         if (cnt >= PAGE_SIZE)
2070                 cnt = PAGE_SIZE - 1;
2071
2072         memset(iter, 0, sizeof(*iter));
2073         iter->tr = &global_trace;
2074         iter->pos = -1;
2075
2076         /*
2077          * We need to stop all tracing on all CPUS to read the
2078          * the next buffer. This is a bit expensive, but is
2079          * not done often. We fill all what we can read,
2080          * and then release the locks again.
2081          */
2082
2083         cpus_clear(mask);
2084         local_irq_save(flags);
2085 #ifdef CONFIG_FTRACE
2086         ftrace_save = ftrace_enabled;
2087         ftrace_enabled = 0;
2088 #endif
2089         smp_wmb();
2090         for_each_possible_cpu(cpu) {
2091                 data = iter->tr->data[cpu];
2092
2093                 if (!head_page(data) || !data->trace_idx)
2094                         continue;
2095
2096                 atomic_inc(&data->disabled);
2097                 cpu_set(cpu, mask);
2098         }
2099
2100         for_each_cpu_mask(cpu, mask) {
2101                 data = iter->tr->data[cpu];
2102                 spin_lock(&data->lock);
2103         }
2104
2105         while (find_next_entry_inc(iter) != NULL) {
2106                 int len = iter->seq.len;
2107
2108                 ret = print_trace_line(iter);
2109                 if (!ret) {
2110                         /* don't print partial lines */
2111                         iter->seq.len = len;
2112                         break;
2113                 }
2114
2115                 trace_consume(iter);
2116
2117                 if (iter->seq.len >= cnt)
2118                         break;
2119         }
2120
2121         for_each_cpu_mask(cpu, mask) {
2122                 data = iter->tr->data[cpu];
2123                 spin_unlock(&data->lock);
2124         }
2125
2126         for_each_cpu_mask(cpu, mask) {
2127                 data = iter->tr->data[cpu];
2128                 atomic_dec(&data->disabled);
2129         }
2130 #ifdef CONFIG_FTRACE
2131         ftrace_enabled = ftrace_save;
2132 #endif
2133         local_irq_restore(flags);
2134
2135         /* Now copy what we have to the user */
2136         read = iter->seq.len;
2137         if (read > cnt)
2138                 read = cnt;
2139
2140         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2141
2142         if (read < iter->seq.len)
2143                 start = read;
2144         else
2145                 trace_seq_reset(&iter->seq);
2146
2147         if (ret)
2148                 read = -EFAULT;
2149
2150         return read;
2151 }
2152
2153 static struct file_operations tracing_max_lat_fops = {
2154         .open           = tracing_open_generic,
2155         .read           = tracing_max_lat_read,
2156         .write          = tracing_max_lat_write,
2157 };
2158
2159 static struct file_operations tracing_ctrl_fops = {
2160         .open           = tracing_open_generic,
2161         .read           = tracing_ctrl_read,
2162         .write          = tracing_ctrl_write,
2163 };
2164
2165 static struct file_operations set_tracer_fops = {
2166         .open           = tracing_open_generic,
2167         .read           = tracing_set_trace_read,
2168         .write          = tracing_set_trace_write,
2169 };
2170
2171 static struct file_operations tracing_pipe_fops = {
2172         .open           = tracing_open_pipe,
2173         .poll           = tracing_poll_pipe,
2174         .read           = tracing_read_pipe,
2175         .release        = tracing_release_pipe,
2176 };
2177
2178 #ifdef CONFIG_DYNAMIC_FTRACE
2179
2180 static ssize_t
2181 tracing_read_long(struct file *filp, char __user *ubuf,
2182                   size_t cnt, loff_t *ppos)
2183 {
2184         unsigned long *p = filp->private_data;
2185         char buf[64];
2186         int r;
2187
2188         r = sprintf(buf, "%ld\n", *p);
2189
2190         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2191 }
2192
2193 static struct file_operations tracing_read_long_fops = {
2194         .open           = tracing_open_generic,
2195         .read           = tracing_read_long,
2196 };
2197 #endif
2198
2199 static struct dentry *d_tracer;
2200
2201 struct dentry *tracing_init_dentry(void)
2202 {
2203         static int once;
2204
2205         if (d_tracer)
2206                 return d_tracer;
2207
2208         d_tracer = debugfs_create_dir("tracing", NULL);
2209
2210         if (!d_tracer && !once) {
2211                 once = 1;
2212                 pr_warning("Could not create debugfs directory 'tracing'\n");
2213                 return NULL;
2214         }
2215
2216         return d_tracer;
2217 }
2218
2219 #ifdef CONFIG_FTRACE_SELFTEST
2220 /* Let selftest have access to static functions in this file */
2221 #include "trace_selftest.c"
2222 #endif
2223
2224 static __init void tracer_init_debugfs(void)
2225 {
2226         struct dentry *d_tracer;
2227         struct dentry *entry;
2228
2229         d_tracer = tracing_init_dentry();
2230
2231         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2232                                     &global_trace, &tracing_ctrl_fops);
2233         if (!entry)
2234                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2235
2236         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2237                                     NULL, &tracing_iter_fops);
2238         if (!entry)
2239                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2240
2241         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2242                                     &global_trace, &tracing_lt_fops);
2243         if (!entry)
2244                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2245
2246         entry = debugfs_create_file("trace", 0444, d_tracer,
2247                                     &global_trace, &tracing_fops);
2248         if (!entry)
2249                 pr_warning("Could not create debugfs 'trace' entry\n");
2250
2251         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2252                                     &global_trace, &show_traces_fops);
2253         if (!entry)
2254                 pr_warning("Could not create debugfs 'trace' entry\n");
2255
2256         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2257                                     &global_trace, &set_tracer_fops);
2258         if (!entry)
2259                 pr_warning("Could not create debugfs 'trace' entry\n");
2260
2261         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2262                                     &tracing_max_latency,
2263                                     &tracing_max_lat_fops);
2264         if (!entry)
2265                 pr_warning("Could not create debugfs "
2266                            "'tracing_max_latency' entry\n");
2267
2268         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2269                                     &tracing_thresh, &tracing_max_lat_fops);
2270         if (!entry)
2271                 pr_warning("Could not create debugfs "
2272                            "'tracing_threash' entry\n");
2273         entry = debugfs_create_file("README", 0644, d_tracer,
2274                                     NULL, &tracing_readme_fops);
2275         if (!entry)
2276                 pr_warning("Could not create debugfs 'README' entry\n");
2277
2278         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2279                                     NULL, &tracing_pipe_fops);
2280         if (!entry)
2281                 pr_warning("Could not create debugfs "
2282                            "'tracing_threash' entry\n");
2283
2284 #ifdef CONFIG_DYNAMIC_FTRACE
2285         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2286                                     &ftrace_update_tot_cnt,
2287                                     &tracing_read_long_fops);
2288         if (!entry)
2289                 pr_warning("Could not create debugfs "
2290                            "'dyn_ftrace_total_info' entry\n");
2291 #endif
2292 }
2293
2294 /* dummy trace to disable tracing */
2295 static struct tracer no_tracer __read_mostly =
2296 {
2297         .name           = "none",
2298 };
2299
2300 static int trace_alloc_page(void)
2301 {
2302         struct trace_array_cpu *data;
2303         struct page *page, *tmp;
2304         LIST_HEAD(pages);
2305         void *array;
2306         int i;
2307
2308         /* first allocate a page for each CPU */
2309         for_each_possible_cpu(i) {
2310                 array = (void *)__get_free_page(GFP_KERNEL);
2311                 if (array == NULL) {
2312                         printk(KERN_ERR "tracer: failed to allocate page"
2313                                "for trace buffer!\n");
2314                         goto free_pages;
2315                 }
2316
2317                 page = virt_to_page(array);
2318                 list_add(&page->lru, &pages);
2319
2320 /* Only allocate if we are actually using the max trace */
2321 #ifdef CONFIG_TRACER_MAX_TRACE
2322                 array = (void *)__get_free_page(GFP_KERNEL);
2323                 if (array == NULL) {
2324                         printk(KERN_ERR "tracer: failed to allocate page"
2325                                "for trace buffer!\n");
2326                         goto free_pages;
2327                 }
2328                 page = virt_to_page(array);
2329                 list_add(&page->lru, &pages);
2330 #endif
2331         }
2332
2333         /* Now that we successfully allocate a page per CPU, add them */
2334         for_each_possible_cpu(i) {
2335                 data = global_trace.data[i];
2336                 spin_lock_init(&data->lock);
2337                 lockdep_set_class(&data->lock, &data->lock_key);
2338                 page = list_entry(pages.next, struct page, lru);
2339                 list_del_init(&page->lru);
2340                 list_add_tail(&page->lru, &data->trace_pages);
2341                 ClearPageLRU(page);
2342
2343 #ifdef CONFIG_TRACER_MAX_TRACE
2344                 data = max_tr.data[i];
2345                 spin_lock_init(&data->lock);
2346                 lockdep_set_class(&data->lock, &data->lock_key);
2347                 page = list_entry(pages.next, struct page, lru);
2348                 list_del_init(&page->lru);
2349                 list_add_tail(&page->lru, &data->trace_pages);
2350                 SetPageLRU(page);
2351 #endif
2352         }
2353         global_trace.entries += ENTRIES_PER_PAGE;
2354
2355         return 0;
2356
2357  free_pages:
2358         list_for_each_entry_safe(page, tmp, &pages, lru) {
2359                 list_del_init(&page->lru);
2360                 __free_page(page);
2361         }
2362         return -ENOMEM;
2363 }
2364
2365 __init static int tracer_alloc_buffers(void)
2366 {
2367         struct trace_array_cpu *data;
2368         void *array;
2369         struct page *page;
2370         int pages = 0;
2371         int ret = -ENOMEM;
2372         int i;
2373
2374         global_trace.ctrl = tracer_enabled;
2375
2376         /* Allocate the first page for all buffers */
2377         for_each_possible_cpu(i) {
2378                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2379                 max_tr.data[i] = &per_cpu(max_data, i);
2380
2381                 array = (void *)__get_free_page(GFP_KERNEL);
2382                 if (array == NULL) {
2383                         printk(KERN_ERR "tracer: failed to allocate page"
2384                                "for trace buffer!\n");
2385                         goto free_buffers;
2386                 }
2387
2388                 /* set the array to the list */
2389                 INIT_LIST_HEAD(&data->trace_pages);
2390                 page = virt_to_page(array);
2391                 list_add(&page->lru, &data->trace_pages);
2392                 /* use the LRU flag to differentiate the two buffers */
2393                 ClearPageLRU(page);
2394
2395 /* Only allocate if we are actually using the max trace */
2396 #ifdef CONFIG_TRACER_MAX_TRACE
2397                 array = (void *)__get_free_page(GFP_KERNEL);
2398                 if (array == NULL) {
2399                         printk(KERN_ERR "tracer: failed to allocate page"
2400                                "for trace buffer!\n");
2401                         goto free_buffers;
2402                 }
2403
2404                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2405                 page = virt_to_page(array);
2406                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2407                 SetPageLRU(page);
2408 #endif
2409         }
2410
2411         /*
2412          * Since we allocate by orders of pages, we may be able to
2413          * round up a bit.
2414          */
2415         global_trace.entries = ENTRIES_PER_PAGE;
2416         pages++;
2417
2418         while (global_trace.entries < trace_nr_entries) {
2419                 if (trace_alloc_page())
2420                         break;
2421                 pages++;
2422         }
2423         max_tr.entries = global_trace.entries;
2424
2425         pr_info("tracer: %d pages allocated for %ld",
2426                 pages, trace_nr_entries);
2427         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2428         pr_info("   actual entries %ld\n", global_trace.entries);
2429
2430         tracer_init_debugfs();
2431
2432         trace_init_cmdlines();
2433
2434         register_tracer(&no_tracer);
2435         current_trace = &no_tracer;
2436
2437         /* All seems OK, enable tracing */
2438         tracing_disabled = 0;
2439
2440         return 0;
2441
2442  free_buffers:
2443         for (i-- ; i >= 0; i--) {
2444                 struct page *page, *tmp;
2445                 struct trace_array_cpu *data = global_trace.data[i];
2446
2447                 if (data) {
2448                         list_for_each_entry_safe(page, tmp,
2449                                                  &data->trace_pages, lru) {
2450                                 list_del_init(&page->lru);
2451                                 __free_page(page);
2452                         }
2453                 }
2454
2455 #ifdef CONFIG_TRACER_MAX_TRACE
2456                 data = max_tr.data[i];
2457                 if (data) {
2458                         list_for_each_entry_safe(page, tmp,
2459                                                  &data->trace_pages, lru) {
2460                                 list_del_init(&page->lru);
2461                                 __free_page(page);
2462                         }
2463                 }
2464 #endif
2465         }
2466         return ret;
2467 }
2468 fs_initcall(tracer_alloc_buffers);