git.oblomov.eu Git - linux-2.6/blob - kernel/trace/trace.c

   1 /*
   2  * ring buffer based function tracer
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally taken from the RT patch by:
   8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code from the latency_tracer, that is:
  11  *  Copyright (C) 2004-2006 Ingo Molnar
  12  *  Copyright (C) 2004 William Lee Irwin III
  13  */
  14 #include <linux/utsrelease.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/debugfs.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/linkage.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/ftrace.h>
  23 #include <linux/module.h>
  24 #include <linux/percpu.h>
  25 #include <linux/ctype.h>
  26 #include <linux/init.h>
  27 #include <linux/poll.h>
  28 #include <linux/gfp.h>
  29 #include <linux/fs.h>
  30
  31 #include <linux/stacktrace.h>
  32
  33 #include "trace.h"
  34
  35 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  36 unsigned long __read_mostly     tracing_thresh;
  37
  38 static int tracing_disabled = 1;
  39
  40 static long
  41 ns2usecs(cycle_t nsec)
  42 {
  43         nsec += 500;
  44         do_div(nsec, 1000);
  45         return nsec;
  46 }
  47
  48 cycle_t ftrace_now(int cpu)
  49 {
  50         return cpu_clock(cpu);
  51 }
  52
  53 static struct trace_array       global_trace;
  54
  55 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  56
  57 static struct trace_array       max_tr;
  58
  59 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  60
  61 static int                      tracer_enabled = 1;
  62 static unsigned long            trace_nr_entries = 65536UL;
  63
  64 static struct tracer            *trace_types __read_mostly;
  65 static struct tracer            *current_trace __read_mostly;
  66 static int                      max_tracer_type_len;
  67
  68 static DEFINE_MUTEX(trace_types_lock);
  69 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
  70
  71 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
  72
  73 void trace_wake_up(void)
  74 {
  75         /*
  76          * The runqueue_is_locked() can fail, but this is the best we
  77          * have for now:
  78          */
  79         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
  80                 wake_up(&trace_wait);
  81 }
  82
  83 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  84
  85 static int __init set_nr_entries(char *str)
  86 {
  87         if (!str)
  88                 return 0;
  89         trace_nr_entries = simple_strtoul(str, &str, 0);
  90         return 1;
  91 }
  92 __setup("trace_entries=", set_nr_entries);
  93
  94 unsigned long nsecs_to_usecs(unsigned long nsecs)
  95 {
  96         return nsecs / 1000;
  97 }
  98
  99 enum trace_type {
 100         __TRACE_FIRST_TYPE = 0,
 101
 102         TRACE_FN,
 103         TRACE_CTX,
 104         TRACE_WAKE,
 105         TRACE_STACK,
 106         TRACE_SPECIAL,
 107
 108         __TRACE_LAST_TYPE
 109 };
 110
 111 enum trace_flag_type {
 112         TRACE_FLAG_IRQS_OFF             = 0x01,
 113         TRACE_FLAG_NEED_RESCHED         = 0x02,
 114         TRACE_FLAG_HARDIRQ              = 0x04,
 115         TRACE_FLAG_SOFTIRQ              = 0x08,
 116 };
 117
 118 #define TRACE_ITER_SYM_MASK \
 119         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 120
 121 /* These must match the bit postions above */
 122 static const char *trace_options[] = {
 123         "print-parent",
 124         "sym-offset",
 125         "sym-addr",
 126         "verbose",
 127         "raw",
 128         "hex",
 129         "bin",
 130         "block",
 131         "stacktrace",
 132         NULL
 133 };
 134
 135 static DEFINE_SPINLOCK(ftrace_max_lock);
 136
 137 /*
 138  * Copy the new maximum trace into the separate maximum-trace
 139  * structure. (this way the maximum trace is permanently saved,
 140  * for later retrieval via /debugfs/tracing/latency_trace)
 141  */
 142 static void
 143 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 144 {
 145         struct trace_array_cpu *data = tr->data[cpu];
 146
 147         max_tr.cpu = cpu;
 148         max_tr.time_start = data->preempt_timestamp;
 149
 150         data = max_tr.data[cpu];
 151         data->saved_latency = tracing_max_latency;
 152
 153         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 154         data->pid = tsk->pid;
 155         data->uid = tsk->uid;
 156         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 157         data->policy = tsk->policy;
 158         data->rt_priority = tsk->rt_priority;
 159
 160         /* record this tasks comm */
 161         tracing_record_cmdline(current);
 162 }
 163
 164 void check_pages(struct trace_array_cpu *data)
 165 {
 166         struct page *page, *tmp;
 167
 168         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 169         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 170
 171         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 172                 BUG_ON(page->lru.next->prev != &page->lru);
 173                 BUG_ON(page->lru.prev->next != &page->lru);
 174         }
 175 }
 176
 177 void *head_page(struct trace_array_cpu *data)
 178 {
 179         struct page *page;
 180
 181         check_pages(data);
 182         if (list_empty(&data->trace_pages))
 183                 return NULL;
 184
 185         page = list_entry(data->trace_pages.next, struct page, lru);
 186         BUG_ON(&page->lru == &data->trace_pages);
 187
 188         return page_address(page);
 189 }
 190
 191 static int
 192 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 193 {
 194         int len = (PAGE_SIZE - 1) - s->len;
 195         va_list ap;
 196         int ret;
 197
 198         if (!len)
 199                 return 0;
 200
 201         va_start(ap, fmt);
 202         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 203         va_end(ap);
 204
 205         /* If we can't write it all, don't bother writing anything */
 206         if (ret > len)
 207                 return 0;
 208
 209         s->len += ret;
 210
 211         return len;
 212 }
 213
 214 static int
 215 trace_seq_puts(struct trace_seq *s, const char *str)
 216 {
 217         int len = strlen(str);
 218
 219         if (len > ((PAGE_SIZE - 1) - s->len))
 220                 return 0;
 221
 222         memcpy(s->buffer + s->len, str, len);
 223         s->len += len;
 224
 225         return len;
 226 }
 227
 228 static int
 229 trace_seq_putc(struct trace_seq *s, unsigned char c)
 230 {
 231         if (s->len >= (PAGE_SIZE - 1))
 232                 return 0;
 233
 234         s->buffer[s->len++] = c;
 235
 236         return 1;
 237 }
 238
 239 static int
 240 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 241 {
 242         if (len > ((PAGE_SIZE - 1) - s->len))
 243                 return 0;
 244
 245         memcpy(s->buffer + s->len, mem, len);
 246         s->len += len;
 247
 248         return len;
 249 }
 250
 251 #define HEX_CHARS 17
 252
 253 static int
 254 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 255 {
 256         unsigned char hex[HEX_CHARS];
 257         unsigned char *data;
 258         unsigned char byte;
 259         int i, j;
 260
 261         BUG_ON(len >= HEX_CHARS);
 262
 263         data = mem;
 264
 265 #ifdef __BIG_ENDIAN
 266         for (i = 0, j = 0; i < len; i++) {
 267 #else
 268         for (i = len-1, j = 0; i >= 0; i--) {
 269 #endif
 270                 byte = data[i];
 271
 272                 hex[j]   = byte & 0x0f;
 273                 if (hex[j] >= 10)
 274                         hex[j] += 'a' - 10;
 275                 else
 276                         hex[j] += '0';
 277                 j++;
 278
 279                 hex[j] = byte >> 4;
 280                 if (hex[j] >= 10)
 281                         hex[j] += 'a' - 10;
 282                 else
 283                         hex[j] += '0';
 284                 j++;
 285         }
 286         hex[j] = ' ';
 287         j++;
 288
 289         return trace_seq_putmem(s, hex, j);
 290 }
 291
 292 static void
 293 trace_seq_reset(struct trace_seq *s)
 294 {
 295         s->len = 0;
 296 }
 297
 298 static void
 299 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 300 {
 301         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 302
 303         s->buffer[len] = 0;
 304         seq_puts(m, s->buffer);
 305
 306         trace_seq_reset(s);
 307 }
 308
 309 static void
 310 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 311 {
 312         struct list_head flip_pages;
 313
 314         INIT_LIST_HEAD(&flip_pages);
 315
 316         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 317                 sizeof(struct trace_array_cpu) -
 318                 offsetof(struct trace_array_cpu, trace_head_idx));
 319
 320         check_pages(tr1);
 321         check_pages(tr2);
 322         list_splice_init(&tr1->trace_pages, &flip_pages);
 323         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 324         list_splice_init(&flip_pages, &tr2->trace_pages);
 325         BUG_ON(!list_empty(&flip_pages));
 326         check_pages(tr1);
 327         check_pages(tr2);
 328 }
 329
 330 void
 331 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 332 {
 333         struct trace_array_cpu *data;
 334         int i;
 335
 336         WARN_ON_ONCE(!irqs_disabled());
 337         spin_lock(&ftrace_max_lock);
 338         /* clear out all the previous traces */
 339         for_each_possible_cpu(i) {
 340                 data = tr->data[i];
 341                 flip_trace(max_tr.data[i], data);
 342                 tracing_reset(data);
 343         }
 344
 345         __update_max_tr(tr, tsk, cpu);
 346         spin_unlock(&ftrace_max_lock);
 347 }
 348
 349 /**
 350  * update_max_tr_single - only copy one trace over, and reset the rest
 351  * @tr - tracer
 352  * @tsk - task with the latency
 353  * @cpu - the cpu of the buffer to copy.
 354  */
 355 void
 356 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 357 {
 358         struct trace_array_cpu *data = tr->data[cpu];
 359         int i;
 360
 361         WARN_ON_ONCE(!irqs_disabled());
 362         spin_lock(&ftrace_max_lock);
 363         for_each_possible_cpu(i)
 364                 tracing_reset(max_tr.data[i]);
 365
 366         flip_trace(max_tr.data[cpu], data);
 367         tracing_reset(data);
 368
 369         __update_max_tr(tr, tsk, cpu);
 370         spin_unlock(&ftrace_max_lock);
 371 }
 372
 373 int register_tracer(struct tracer *type)
 374 {
 375         struct tracer *t;
 376         int len;
 377         int ret = 0;
 378
 379         if (!type->name) {
 380                 pr_info("Tracer must have a name\n");
 381                 return -1;
 382         }
 383
 384         mutex_lock(&trace_types_lock);
 385         for (t = trace_types; t; t = t->next) {
 386                 if (strcmp(type->name, t->name) == 0) {
 387                         /* already found */
 388                         pr_info("Trace %s already registered\n",
 389                                 type->name);
 390                         ret = -1;
 391                         goto out;
 392                 }
 393         }
 394
 395 #ifdef CONFIG_FTRACE_STARTUP_TEST
 396         if (type->selftest) {
 397                 struct tracer *saved_tracer = current_trace;
 398                 struct trace_array_cpu *data;
 399                 struct trace_array *tr = &global_trace;
 400                 int saved_ctrl = tr->ctrl;
 401                 int i;
 402                 /*
 403                  * Run a selftest on this tracer.
 404                  * Here we reset the trace buffer, and set the current
 405                  * tracer to be this tracer. The tracer can then run some
 406                  * internal tracing to verify that everything is in order.
 407                  * If we fail, we do not register this tracer.
 408                  */
 409                 for_each_possible_cpu(i) {
 410                         data = tr->data[i];
 411                         if (!head_page(data))
 412                                 continue;
 413                         tracing_reset(data);
 414                 }
 415                 current_trace = type;
 416                 tr->ctrl = 0;
 417                 /* the test is responsible for initializing and enabling */
 418                 pr_info("Testing tracer %s: ", type->name);
 419                 ret = type->selftest(type, tr);
 420                 /* the test is responsible for resetting too */
 421                 current_trace = saved_tracer;
 422                 tr->ctrl = saved_ctrl;
 423                 if (ret) {
 424                         printk(KERN_CONT "FAILED!\n");
 425                         goto out;
 426                 }
 427                 /* Only reset on passing, to avoid touching corrupted buffers */
 428                 for_each_possible_cpu(i) {
 429                         data = tr->data[i];
 430                         if (!head_page(data))
 431                                 continue;
 432                         tracing_reset(data);
 433                 }
 434                 printk(KERN_CONT "PASSED\n");
 435         }
 436 #endif
 437
 438         type->next = trace_types;
 439         trace_types = type;
 440         len = strlen(type->name);
 441         if (len > max_tracer_type_len)
 442                 max_tracer_type_len = len;
 443
 444  out:
 445         mutex_unlock(&trace_types_lock);
 446
 447         return ret;
 448 }
 449
 450 void unregister_tracer(struct tracer *type)
 451 {
 452         struct tracer **t;
 453         int len;
 454
 455         mutex_lock(&trace_types_lock);
 456         for (t = &trace_types; *t; t = &(*t)->next) {
 457                 if (*t == type)
 458                         goto found;
 459         }
 460         pr_info("Trace %s not registered\n", type->name);
 461         goto out;
 462
 463  found:
 464         *t = (*t)->next;
 465         if (strlen(type->name) != max_tracer_type_len)
 466                 goto out;
 467
 468         max_tracer_type_len = 0;
 469         for (t = &trace_types; *t; t = &(*t)->next) {
 470                 len = strlen((*t)->name);
 471                 if (len > max_tracer_type_len)
 472                         max_tracer_type_len = len;
 473         }
 474  out:
 475         mutex_unlock(&trace_types_lock);
 476 }
 477
 478 void tracing_reset(struct trace_array_cpu *data)
 479 {
 480         data->trace_idx = 0;
 481         data->trace_head = data->trace_tail = head_page(data);
 482         data->trace_head_idx = 0;
 483         data->trace_tail_idx = 0;
 484 }
 485
 486 #define SAVED_CMDLINES 128
 487 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 488 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 489 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 490 static int cmdline_idx;
 491 static DEFINE_SPINLOCK(trace_cmdline_lock);
 492 atomic_t trace_record_cmdline_disabled;
 493
 494 static void trace_init_cmdlines(void)
 495 {
 496         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 497         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 498         cmdline_idx = 0;
 499 }
 500
 501 void trace_stop_cmdline_recording(void);
 502
 503 static void trace_save_cmdline(struct task_struct *tsk)
 504 {
 505         unsigned map;
 506         unsigned idx;
 507
 508         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 509                 return;
 510
 511         /*
 512          * It's not the end of the world if we don't get
 513          * the lock, but we also don't want to spin
 514          * nor do we want to disable interrupts,
 515          * so if we miss here, then better luck next time.
 516          */
 517         if (!spin_trylock(&trace_cmdline_lock))
 518                 return;
 519
 520         idx = map_pid_to_cmdline[tsk->pid];
 521         if (idx >= SAVED_CMDLINES) {
 522                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 523
 524                 map = map_cmdline_to_pid[idx];
 525                 if (map <= PID_MAX_DEFAULT)
 526                         map_pid_to_cmdline[map] = (unsigned)-1;
 527
 528                 map_pid_to_cmdline[tsk->pid] = idx;
 529
 530                 cmdline_idx = idx;
 531         }
 532
 533         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 534
 535         spin_unlock(&trace_cmdline_lock);
 536 }
 537
 538 static char *trace_find_cmdline(int pid)
 539 {
 540         char *cmdline = "<...>";
 541         unsigned map;
 542
 543         if (!pid)
 544                 return "<idle>";
 545
 546         if (pid > PID_MAX_DEFAULT)
 547                 goto out;
 548
 549         map = map_pid_to_cmdline[pid];
 550         if (map >= SAVED_CMDLINES)
 551                 goto out;
 552
 553         cmdline = saved_cmdlines[map];
 554
 555  out:
 556         return cmdline;
 557 }
 558
 559 void tracing_record_cmdline(struct task_struct *tsk)
 560 {
 561         if (atomic_read(&trace_record_cmdline_disabled))
 562                 return;
 563
 564         trace_save_cmdline(tsk);
 565 }
 566
 567 static inline struct list_head *
 568 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 569 {
 570         /*
 571          * Roundrobin - but skip the head (which is not a real page):
 572          */
 573         next = next->next;
 574         if (unlikely(next == &data->trace_pages))
 575                 next = next->next;
 576         BUG_ON(next == &data->trace_pages);
 577
 578         return next;
 579 }
 580
 581 static inline void *
 582 trace_next_page(struct trace_array_cpu *data, void *addr)
 583 {
 584         struct list_head *next;
 585         struct page *page;
 586
 587         page = virt_to_page(addr);
 588
 589         next = trace_next_list(data, &page->lru);
 590         page = list_entry(next, struct page, lru);
 591
 592         return page_address(page);
 593 }
 594
 595 static inline struct trace_entry *
 596 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 597 {
 598         unsigned long idx, idx_next;
 599         struct trace_entry *entry;
 600
 601         data->trace_idx++;
 602         idx = data->trace_head_idx;
 603         idx_next = idx + 1;
 604
 605         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 606
 607         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 608
 609         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 610                 data->trace_head = trace_next_page(data, data->trace_head);
 611                 idx_next = 0;
 612         }
 613
 614         if (data->trace_head == data->trace_tail &&
 615             idx_next == data->trace_tail_idx) {
 616                 /* overrun */
 617                 data->trace_tail_idx++;
 618                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 619                         data->trace_tail =
 620                                 trace_next_page(data, data->trace_tail);
 621                         data->trace_tail_idx = 0;
 622                 }
 623         }
 624
 625         data->trace_head_idx = idx_next;
 626
 627         return entry;
 628 }
 629
 630 static inline void
 631 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 632 {
 633         struct task_struct *tsk = current;
 634         unsigned long pc;
 635
 636         pc = preempt_count();
 637
 638         entry->preempt_count    = pc & 0xff;
 639         entry->pid              = tsk->pid;
 640         entry->t                = ftrace_now(raw_smp_processor_id());
 641         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 642                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 643                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 644                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 645 }
 646
 647 void
 648 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 649                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 650 {
 651         struct trace_entry *entry;
 652         unsigned long irq_flags;
 653
 654         spin_lock_irqsave(&data->lock, irq_flags);
 655         entry                   = tracing_get_trace_entry(tr, data);
 656         tracing_generic_entry_update(entry, flags);
 657         entry->type             = TRACE_FN;
 658         entry->fn.ip            = ip;
 659         entry->fn.parent_ip     = parent_ip;
 660         spin_unlock_irqrestore(&data->lock, irq_flags);
 661
 662         trace_wake_up();
 663 }
 664
 665 void
 666 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 667        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 668 {
 669         if (likely(!atomic_read(&data->disabled)))
 670                 trace_function(tr, data, ip, parent_ip, flags);
 671 }
 672
 673 void
 674 __trace_special(void *__tr, void *__data,
 675                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 676 {
 677         struct trace_array_cpu *data = __data;
 678         struct trace_array *tr = __tr;
 679         struct trace_entry *entry;
 680         unsigned long irq_flags;
 681
 682         spin_lock_irqsave(&data->lock, irq_flags);
 683         entry                   = tracing_get_trace_entry(tr, data);
 684         tracing_generic_entry_update(entry, 0);
 685         entry->type             = TRACE_SPECIAL;
 686         entry->special.arg1     = arg1;
 687         entry->special.arg2     = arg2;
 688         entry->special.arg3     = arg3;
 689         spin_unlock_irqrestore(&data->lock, irq_flags);
 690
 691         trace_wake_up();
 692 }
 693
 694 void __trace_stack(struct trace_array *tr,
 695                    struct trace_array_cpu *data,
 696                    unsigned long flags,
 697                    int skip)
 698 {
 699         struct trace_entry *entry;
 700         struct stack_trace trace;
 701
 702         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 703                 return;
 704
 705         entry                   = tracing_get_trace_entry(tr, data);
 706         tracing_generic_entry_update(entry, flags);
 707         entry->type             = TRACE_STACK;
 708
 709         memset(&entry->stack, 0, sizeof(entry->stack));
 710
 711         trace.nr_entries        = 0;
 712         trace.max_entries       = FTRACE_STACK_ENTRIES;
 713         trace.skip              = skip;
 714         trace.entries           = entry->stack.caller;
 715
 716         save_stack_trace(&trace);
 717 }
 718
 719 void
 720 tracing_sched_switch_trace(struct trace_array *tr,
 721                            struct trace_array_cpu *data,
 722                            struct task_struct *prev,
 723                            struct task_struct *next,
 724                            unsigned long flags)
 725 {
 726         struct trace_entry *entry;
 727         unsigned long irq_flags;
 728
 729         spin_lock_irqsave(&data->lock, irq_flags);
 730         entry                   = tracing_get_trace_entry(tr, data);
 731         tracing_generic_entry_update(entry, flags);
 732         entry->type             = TRACE_CTX;
 733         entry->ctx.prev_pid     = prev->pid;
 734         entry->ctx.prev_prio    = prev->prio;
 735         entry->ctx.prev_state   = prev->state;
 736         entry->ctx.next_pid     = next->pid;
 737         entry->ctx.next_prio    = next->prio;
 738         __trace_stack(tr, data, flags, 4);
 739         spin_unlock_irqrestore(&data->lock, irq_flags);
 740 }
 741
 742 void
 743 tracing_sched_wakeup_trace(struct trace_array *tr,
 744                            struct trace_array_cpu *data,
 745                            struct task_struct *wakee,
 746                            struct task_struct *curr,
 747                            unsigned long flags)
 748 {
 749         struct trace_entry *entry;
 750         unsigned long irq_flags;
 751
 752         spin_lock_irqsave(&data->lock, irq_flags);
 753         entry                   = tracing_get_trace_entry(tr, data);
 754         tracing_generic_entry_update(entry, flags);
 755         entry->type             = TRACE_WAKE;
 756         entry->ctx.prev_pid     = curr->pid;
 757         entry->ctx.prev_prio    = curr->prio;
 758         entry->ctx.prev_state   = curr->state;
 759         entry->ctx.next_pid     = wakee->pid;
 760         entry->ctx.next_prio    = wakee->prio;
 761         __trace_stack(tr, data, flags, 5);
 762         spin_unlock_irqrestore(&data->lock, irq_flags);
 763
 764         trace_wake_up();
 765 }
 766
 767 #ifdef CONFIG_FTRACE
 768 static void
 769 function_trace_call(unsigned long ip, unsigned long parent_ip)
 770 {
 771         struct trace_array *tr = &global_trace;
 772         struct trace_array_cpu *data;
 773         unsigned long flags;
 774         long disabled;
 775         int cpu;
 776
 777         if (unlikely(!tracer_enabled))
 778                 return;
 779
 780         local_irq_save(flags);
 781         cpu = raw_smp_processor_id();
 782         data = tr->data[cpu];
 783         disabled = atomic_inc_return(&data->disabled);
 784
 785         if (likely(disabled == 1))
 786                 trace_function(tr, data, ip, parent_ip, flags);
 787
 788         atomic_dec(&data->disabled);
 789         local_irq_restore(flags);
 790 }
 791
 792 static struct ftrace_ops trace_ops __read_mostly =
 793 {
 794         .func = function_trace_call,
 795 };
 796
 797 void tracing_start_function_trace(void)
 798 {
 799         register_ftrace_function(&trace_ops);
 800 }
 801
 802 void tracing_stop_function_trace(void)
 803 {
 804         unregister_ftrace_function(&trace_ops);
 805 }
 806 #endif
 807
 808 enum trace_file_type {
 809         TRACE_FILE_LAT_FMT      = 1,
 810 };
 811
 812 static struct trace_entry *
 813 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 814                 struct trace_iterator *iter, int cpu)
 815 {
 816         struct page *page;
 817         struct trace_entry *array;
 818
 819         if (iter->next_idx[cpu] >= tr->entries ||
 820             iter->next_idx[cpu] >= data->trace_idx ||
 821             (data->trace_head == data->trace_tail &&
 822              data->trace_head_idx == data->trace_tail_idx))
 823                 return NULL;
 824
 825         if (!iter->next_page[cpu]) {
 826                 /* Initialize the iterator for this cpu trace buffer */
 827                 WARN_ON(!data->trace_tail);
 828                 page = virt_to_page(data->trace_tail);
 829                 iter->next_page[cpu] = &page->lru;
 830                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 831         }
 832
 833         page = list_entry(iter->next_page[cpu], struct page, lru);
 834         BUG_ON(&data->trace_pages == &page->lru);
 835
 836         array = page_address(page);
 837
 838         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 839         return &array[iter->next_page_idx[cpu]];
 840 }
 841
 842 static struct trace_entry *
 843 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 844 {
 845         struct trace_array *tr = iter->tr;
 846         struct trace_entry *ent, *next = NULL;
 847         int next_cpu = -1;
 848         int cpu;
 849
 850         for_each_possible_cpu(cpu) {
 851                 if (!head_page(tr->data[cpu]))
 852                         continue;
 853                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 854                 /*
 855                  * Pick the entry with the smallest timestamp:
 856                  */
 857                 if (ent && (!next || ent->t < next->t)) {
 858                         next = ent;
 859                         next_cpu = cpu;
 860                 }
 861         }
 862
 863         if (ent_cpu)
 864                 *ent_cpu = next_cpu;
 865
 866         return next;
 867 }
 868
 869 static void trace_iterator_increment(struct trace_iterator *iter)
 870 {
 871         iter->idx++;
 872         iter->next_idx[iter->cpu]++;
 873         iter->next_page_idx[iter->cpu]++;
 874
 875         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 876                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 877
 878                 iter->next_page_idx[iter->cpu] = 0;
 879                 iter->next_page[iter->cpu] =
 880                         trace_next_list(data, iter->next_page[iter->cpu]);
 881         }
 882 }
 883
 884 static void trace_consume(struct trace_iterator *iter)
 885 {
 886         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 887
 888         data->trace_tail_idx++;
 889         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 890                 data->trace_tail = trace_next_page(data, data->trace_tail);
 891                 data->trace_tail_idx = 0;
 892         }
 893
 894         /* Check if we empty it, then reset the index */
 895         if (data->trace_head == data->trace_tail &&
 896             data->trace_head_idx == data->trace_tail_idx)
 897                 data->trace_idx = 0;
 898 }
 899
 900 static void *find_next_entry_inc(struct trace_iterator *iter)
 901 {
 902         struct trace_entry *next;
 903         int next_cpu = -1;
 904
 905         next = find_next_entry(iter, &next_cpu);
 906
 907         iter->prev_ent = iter->ent;
 908         iter->prev_cpu = iter->cpu;
 909
 910         iter->ent = next;
 911         iter->cpu = next_cpu;
 912
 913         if (next)
 914                 trace_iterator_increment(iter);
 915
 916         return next ? iter : NULL;
 917 }
 918
 919 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 920 {
 921         struct trace_iterator *iter = m->private;
 922         void *last_ent = iter->ent;
 923         int i = (int)*pos;
 924         void *ent;
 925
 926         (*pos)++;
 927
 928         /* can't go backwards */
 929         if (iter->idx > i)
 930                 return NULL;
 931
 932         if (iter->idx < 0)
 933                 ent = find_next_entry_inc(iter);
 934         else
 935                 ent = iter;
 936
 937         while (ent && iter->idx < i)
 938                 ent = find_next_entry_inc(iter);
 939
 940         iter->pos = *pos;
 941
 942         if (last_ent && !ent)
 943                 seq_puts(m, "\n\nvim:ft=help\n");
 944
 945         return ent;
 946 }
 947
 948 static void *s_start(struct seq_file *m, loff_t *pos)
 949 {
 950         struct trace_iterator *iter = m->private;
 951         void *p = NULL;
 952         loff_t l = 0;
 953         int i;
 954
 955         mutex_lock(&trace_types_lock);
 956
 957         if (!current_trace || current_trace != iter->trace)
 958                 return NULL;
 959
 960         atomic_inc(&trace_record_cmdline_disabled);
 961
 962         /* let the tracer grab locks here if needed */
 963         if (current_trace->start)
 964                 current_trace->start(iter);
 965
 966         if (*pos != iter->pos) {
 967                 iter->ent = NULL;
 968                 iter->cpu = 0;
 969                 iter->idx = -1;
 970                 iter->prev_ent = NULL;
 971                 iter->prev_cpu = -1;
 972
 973                 for_each_possible_cpu(i) {
 974                         iter->next_idx[i] = 0;
 975                         iter->next_page[i] = NULL;
 976                 }
 977
 978                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 979                         ;
 980
 981         } else {
 982                 l = *pos - 1;
 983                 p = s_next(m, p, &l);
 984         }
 985
 986         return p;
 987 }
 988
 989 static void s_stop(struct seq_file *m, void *p)
 990 {
 991         struct trace_iterator *iter = m->private;
 992
 993         atomic_dec(&trace_record_cmdline_disabled);
 994
 995         /* let the tracer release locks here if needed */
 996         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 997                 iter->trace->stop(iter);
 998
 999         mutex_unlock(&trace_types_lock);
1000 }
1001
1002 static int
1003 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1004 {
1005 #ifdef CONFIG_KALLSYMS
1006         char str[KSYM_SYMBOL_LEN];
1007
1008         kallsyms_lookup(address, NULL, NULL, NULL, str);
1009
1010         return trace_seq_printf(s, fmt, str);
1011 #endif
1012         return 1;
1013 }
1014
1015 static int
1016 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1017                      unsigned long address)
1018 {
1019 #ifdef CONFIG_KALLSYMS
1020         char str[KSYM_SYMBOL_LEN];
1021
1022         sprint_symbol(str, address);
1023         return trace_seq_printf(s, fmt, str);
1024 #endif
1025         return 1;
1026 }
1027
1028 #ifndef CONFIG_64BIT
1029 # define IP_FMT "%08lx"
1030 #else
1031 # define IP_FMT "%016lx"
1032 #endif
1033
1034 static int
1035 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1036 {
1037         int ret;
1038
1039         if (!ip)
1040                 return trace_seq_printf(s, "0");
1041
1042         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1043                 ret = seq_print_sym_offset(s, "%s", ip);
1044         else
1045                 ret = seq_print_sym_short(s, "%s", ip);
1046
1047         if (!ret)
1048                 return 0;
1049
1050         if (sym_flags & TRACE_ITER_SYM_ADDR)
1051                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1052         return ret;
1053 }
1054
1055 static void print_lat_help_header(struct seq_file *m)
1056 {
1057         seq_puts(m, "#                _------=> CPU#            \n");
1058         seq_puts(m, "#               / _-----=> irqs-off        \n");
1059         seq_puts(m, "#              | / _----=> need-resched    \n");
1060         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1061         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1062         seq_puts(m, "#              |||| /                      \n");
1063         seq_puts(m, "#              |||||     delay             \n");
1064         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1065         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1066 }
1067
1068 static void print_func_help_header(struct seq_file *m)
1069 {
1070         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1071         seq_puts(m, "#              | |      |          |         |\n");
1072 }
1073
1074
1075 static void
1076 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1077 {
1078         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1079         struct trace_array *tr = iter->tr;
1080         struct trace_array_cpu *data = tr->data[tr->cpu];
1081         struct tracer *type = current_trace;
1082         unsigned long total   = 0;
1083         unsigned long entries = 0;
1084         int cpu;
1085         const char *name = "preemption";
1086
1087         if (type)
1088                 name = type->name;
1089
1090         for_each_possible_cpu(cpu) {
1091                 if (head_page(tr->data[cpu])) {
1092                         total += tr->data[cpu]->trace_idx;
1093                         if (tr->data[cpu]->trace_idx > tr->entries)
1094                                 entries += tr->entries;
1095                         else
1096                                 entries += tr->data[cpu]->trace_idx;
1097                 }
1098         }
1099
1100         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1101                    name, UTS_RELEASE);
1102         seq_puts(m, "-----------------------------------"
1103                  "---------------------------------\n");
1104         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1105                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1106                    nsecs_to_usecs(data->saved_latency),
1107                    entries,
1108                    total,
1109                    tr->cpu,
1110 #if defined(CONFIG_PREEMPT_NONE)
1111                    "server",
1112 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1113                    "desktop",
1114 #elif defined(CONFIG_PREEMPT_DESKTOP)
1115                    "preempt",
1116 #else
1117                    "unknown",
1118 #endif
1119                    /* These are reserved for later use */
1120                    0, 0, 0, 0);
1121 #ifdef CONFIG_SMP
1122         seq_printf(m, " #P:%d)\n", num_online_cpus());
1123 #else
1124         seq_puts(m, ")\n");
1125 #endif
1126         seq_puts(m, "    -----------------\n");
1127         seq_printf(m, "    | task: %.16s-%d "
1128                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1129                    data->comm, data->pid, data->uid, data->nice,
1130                    data->policy, data->rt_priority);
1131         seq_puts(m, "    -----------------\n");
1132
1133         if (data->critical_start) {
1134                 seq_puts(m, " => started at: ");
1135                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1136                 trace_print_seq(m, &iter->seq);
1137                 seq_puts(m, "\n => ended at:   ");
1138                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1139                 trace_print_seq(m, &iter->seq);
1140                 seq_puts(m, "\n");
1141         }
1142
1143         seq_puts(m, "\n");
1144 }
1145
1146 static void
1147 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1148 {
1149         int hardirq, softirq;
1150         char *comm;
1151
1152         comm = trace_find_cmdline(entry->pid);
1153
1154         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1155         trace_seq_printf(s, "%d", cpu);
1156         trace_seq_printf(s, "%c%c",
1157                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1158                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1159
1160         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1161         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1162         if (hardirq && softirq)
1163                 trace_seq_putc(s, 'H');
1164         else {
1165                 if (hardirq)
1166                         trace_seq_putc(s, 'h');
1167                 else {
1168                         if (softirq)
1169                                 trace_seq_putc(s, 's');
1170                         else
1171                                 trace_seq_putc(s, '.');
1172                 }
1173         }
1174
1175         if (entry->preempt_count)
1176                 trace_seq_printf(s, "%x", entry->preempt_count);
1177         else
1178                 trace_seq_puts(s, ".");
1179 }
1180
1181 unsigned long preempt_mark_thresh = 100;
1182
1183 static void
1184 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1185                     unsigned long rel_usecs)
1186 {
1187         trace_seq_printf(s, " %4lldus", abs_usecs);
1188         if (rel_usecs > preempt_mark_thresh)
1189                 trace_seq_puts(s, "!: ");
1190         else if (rel_usecs > 1)
1191                 trace_seq_puts(s, "+: ");
1192         else
1193                 trace_seq_puts(s, " : ");
1194 }
1195
1196 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1197
1198 static int
1199 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1200 {
1201         struct trace_seq *s = &iter->seq;
1202         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1203         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1204         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1205         struct trace_entry *entry = iter->ent;
1206         unsigned long abs_usecs;
1207         unsigned long rel_usecs;
1208         char *comm;
1209         int S;
1210         int i;
1211
1212         if (!next_entry)
1213                 next_entry = entry;
1214         rel_usecs = ns2usecs(next_entry->t - entry->t);
1215         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1216
1217         if (verbose) {
1218                 comm = trace_find_cmdline(entry->pid);
1219                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1220                                  " %ld.%03ldms (+%ld.%03ldms): ",
1221                                  comm,
1222                                  entry->pid, cpu, entry->flags,
1223                                  entry->preempt_count, trace_idx,
1224                                  ns2usecs(entry->t),
1225                                  abs_usecs/1000,
1226                                  abs_usecs % 1000, rel_usecs/1000,
1227                                  rel_usecs % 1000);
1228         } else {
1229                 if (entry->type != TRACE_STACK) {
1230                         lat_print_generic(s, entry, cpu);
1231                         lat_print_timestamp(s, abs_usecs, rel_usecs);
1232                 }
1233         }
1234         switch (entry->type) {
1235         case TRACE_FN:
1236                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1237                 trace_seq_puts(s, " (");
1238                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1239                 trace_seq_puts(s, ")\n");
1240                 break;
1241         case TRACE_CTX:
1242         case TRACE_WAKE:
1243                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1244                         state_to_char[entry->ctx.prev_state] : 'X';
1245                 comm = trace_find_cmdline(entry->ctx.next_pid);
1246                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d %s\n",
1247                                  entry->ctx.prev_pid,
1248                                  entry->ctx.prev_prio,
1249                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1250                                  entry->ctx.next_pid,
1251                                  entry->ctx.next_prio,
1252                                  comm);
1253                 break;
1254         case TRACE_SPECIAL:
1255                 trace_seq_printf(s, " %ld %ld %ld\n",
1256                                  entry->special.arg1,
1257                                  entry->special.arg2,
1258                                  entry->special.arg3);
1259                 break;
1260         case TRACE_STACK:
1261                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1262                         if (i)
1263                                 trace_seq_puts(s, " <= ");
1264                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1265                 }
1266                 trace_seq_puts(s, "\n");
1267                 break;
1268         default:
1269                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1270         }
1271         return 1;
1272 }
1273
1274 static int print_trace_fmt(struct trace_iterator *iter)
1275 {
1276         struct trace_seq *s = &iter->seq;
1277         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1278         struct trace_entry *entry;
1279         unsigned long usec_rem;
1280         unsigned long long t;
1281         unsigned long secs;
1282         char *comm;
1283         int ret;
1284         int S;
1285         int i;
1286
1287         entry = iter->ent;
1288
1289         comm = trace_find_cmdline(iter->ent->pid);
1290
1291         t = ns2usecs(entry->t);
1292         usec_rem = do_div(t, 1000000ULL);
1293         secs = (unsigned long)t;
1294
1295         if (entry->type != TRACE_STACK) {
1296                 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1297                 if (!ret)
1298                         return 0;
1299                 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1300                 if (!ret)
1301                         return 0;
1302                 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1303                 if (!ret)
1304                         return 0;
1305         }
1306
1307         switch (entry->type) {
1308         case TRACE_FN:
1309                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1310                 if (!ret)
1311                         return 0;
1312                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1313                                                 entry->fn.parent_ip) {
1314                         ret = trace_seq_printf(s, " <-");
1315                         if (!ret)
1316                                 return 0;
1317                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1318                                                sym_flags);
1319                         if (!ret)
1320                                 return 0;
1321                 }
1322                 ret = trace_seq_printf(s, "\n");
1323                 if (!ret)
1324                         return 0;
1325                 break;
1326         case TRACE_CTX:
1327         case TRACE_WAKE:
1328                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1329                         state_to_char[entry->ctx.prev_state] : 'X';
1330                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d\n",
1331                                        entry->ctx.prev_pid,
1332                                        entry->ctx.prev_prio,
1333                                        S,
1334                                        entry->type == TRACE_CTX ? "==>" : "  +",
1335                                        entry->ctx.next_pid,
1336                                        entry->ctx.next_prio);
1337                 if (!ret)
1338                         return 0;
1339                 break;
1340         case TRACE_SPECIAL:
1341                 ret = trace_seq_printf(s, " %ld %ld %ld\n",
1342                                  entry->special.arg1,
1343                                  entry->special.arg2,
1344                                  entry->special.arg3);
1345                 if (!ret)
1346                         return 0;
1347                 break;
1348         case TRACE_STACK:
1349                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1350                         if (i) {
1351                                 ret = trace_seq_puts(s, " <= ");
1352                                 if (!ret)
1353                                         return 0;
1354                         }
1355                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1356                                                sym_flags);
1357                         if (!ret)
1358                                 return 0;
1359                 }
1360                 ret = trace_seq_puts(s, "\n");
1361                 if (!ret)
1362                         return 0;
1363                 break;
1364         }
1365         return 1;
1366 }
1367
1368 static int print_raw_fmt(struct trace_iterator *iter)
1369 {
1370         struct trace_seq *s = &iter->seq;
1371         struct trace_entry *entry;
1372         int ret;
1373         int S;
1374
1375         entry = iter->ent;
1376
1377         ret = trace_seq_printf(s, "%d %d %llu ",
1378                 entry->pid, iter->cpu, entry->t);
1379         if (!ret)
1380                 return 0;
1381
1382         switch (entry->type) {
1383         case TRACE_FN:
1384                 ret = trace_seq_printf(s, "%x %x\n",
1385                                         entry->fn.ip, entry->fn.parent_ip);
1386                 if (!ret)
1387                         return 0;
1388                 break;
1389         case TRACE_CTX:
1390         case TRACE_WAKE:
1391                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1392                         state_to_char[entry->ctx.prev_state] : 'X';
1393                 if (entry->type == TRACE_WAKE)
1394                         S = '+';
1395                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1396                                        entry->ctx.prev_pid,
1397                                        entry->ctx.prev_prio,
1398                                        S,
1399                                        entry->ctx.next_pid,
1400                                        entry->ctx.next_prio);
1401                 if (!ret)
1402                         return 0;
1403                 break;
1404         case TRACE_SPECIAL:
1405         case TRACE_STACK:
1406                 ret = trace_seq_printf(s, " %ld %ld %ld\n",
1407                                  entry->special.arg1,
1408                                  entry->special.arg2,
1409                                  entry->special.arg3);
1410                 if (!ret)
1411                         return 0;
1412                 break;
1413         }
1414         return 1;
1415 }
1416
1417 #define SEQ_PUT_FIELD_RET(s, x)                         \
1418 do {                                                    \
1419         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1420                 return 0;                               \
1421 } while (0)
1422
1423 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1424 do {                                                    \
1425         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1426                 return 0;                               \
1427 } while (0)
1428
1429 static int print_hex_fmt(struct trace_iterator *iter)
1430 {
1431         struct trace_seq *s = &iter->seq;
1432         unsigned char newline = '\n';
1433         struct trace_entry *entry;
1434         int S;
1435
1436         entry = iter->ent;
1437
1438         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1439         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1440         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1441
1442         switch (entry->type) {
1443         case TRACE_FN:
1444                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1445                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1446                 break;
1447         case TRACE_CTX:
1448         case TRACE_WAKE:
1449                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1450                         state_to_char[entry->ctx.prev_state] : 'X';
1451                 if (entry->type == TRACE_WAKE)
1452                         S = '+';
1453                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1454                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1455                 SEQ_PUT_HEX_FIELD_RET(s, S);
1456                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1457                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1458                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1459                 break;
1460         case TRACE_SPECIAL:
1461         case TRACE_STACK:
1462                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1463                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1464                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1465                 break;
1466         }
1467         SEQ_PUT_FIELD_RET(s, newline);
1468
1469         return 1;
1470 }
1471
1472 static int print_bin_fmt(struct trace_iterator *iter)
1473 {
1474         struct trace_seq *s = &iter->seq;
1475         struct trace_entry *entry;
1476
1477         entry = iter->ent;
1478
1479         SEQ_PUT_FIELD_RET(s, entry->pid);
1480         SEQ_PUT_FIELD_RET(s, entry->cpu);
1481         SEQ_PUT_FIELD_RET(s, entry->t);
1482
1483         switch (entry->type) {
1484         case TRACE_FN:
1485                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1486                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1487                 break;
1488         case TRACE_CTX:
1489                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1490                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1491                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1492                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1493                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1494                 break;
1495         case TRACE_SPECIAL:
1496         case TRACE_STACK:
1497                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1498                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1499                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1500                 break;
1501         }
1502         return 1;
1503 }
1504
1505 static int trace_empty(struct trace_iterator *iter)
1506 {
1507         struct trace_array_cpu *data;
1508         int cpu;
1509
1510         for_each_possible_cpu(cpu) {
1511                 data = iter->tr->data[cpu];
1512
1513                 if (head_page(data) && data->trace_idx &&
1514                     (data->trace_tail != data->trace_head ||
1515                      data->trace_tail_idx != data->trace_head_idx))
1516                         return 0;
1517         }
1518         return 1;
1519 }
1520
1521 static int print_trace_line(struct trace_iterator *iter)
1522 {
1523         if (trace_flags & TRACE_ITER_BIN)
1524                 return print_bin_fmt(iter);
1525
1526         if (trace_flags & TRACE_ITER_HEX)
1527                 return print_hex_fmt(iter);
1528
1529         if (trace_flags & TRACE_ITER_RAW)
1530                 return print_raw_fmt(iter);
1531
1532         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1533                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1534
1535         return print_trace_fmt(iter);
1536 }
1537
1538 static int s_show(struct seq_file *m, void *v)
1539 {
1540         struct trace_iterator *iter = v;
1541
1542         if (iter->ent == NULL) {
1543                 if (iter->tr) {
1544                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1545                         seq_puts(m, "#\n");
1546                 }
1547                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1548                         /* print nothing if the buffers are empty */
1549                         if (trace_empty(iter))
1550                                 return 0;
1551                         print_trace_header(m, iter);
1552                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1553                                 print_lat_help_header(m);
1554                 } else {
1555                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1556                                 print_func_help_header(m);
1557                 }
1558         } else {
1559                 print_trace_line(iter);
1560                 trace_print_seq(m, &iter->seq);
1561         }
1562
1563         return 0;
1564 }
1565
1566 static struct seq_operations tracer_seq_ops = {
1567         .start          = s_start,
1568         .next           = s_next,
1569         .stop           = s_stop,
1570         .show           = s_show,
1571 };
1572
1573 static struct trace_iterator *
1574 __tracing_open(struct inode *inode, struct file *file, int *ret)
1575 {
1576         struct trace_iterator *iter;
1577
1578         if (tracing_disabled) {
1579                 *ret = -ENODEV;
1580                 return NULL;
1581         }
1582
1583         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1584         if (!iter) {
1585                 *ret = -ENOMEM;
1586                 goto out;
1587         }
1588
1589         mutex_lock(&trace_types_lock);
1590         if (current_trace && current_trace->print_max)
1591                 iter->tr = &max_tr;
1592         else
1593                 iter->tr = inode->i_private;
1594         iter->trace = current_trace;
1595         iter->pos = -1;
1596
1597         /* TODO stop tracer */
1598         *ret = seq_open(file, &tracer_seq_ops);
1599         if (!*ret) {
1600                 struct seq_file *m = file->private_data;
1601                 m->private = iter;
1602
1603                 /* stop the trace while dumping */
1604                 if (iter->tr->ctrl)
1605                         tracer_enabled = 0;
1606
1607                 if (iter->trace && iter->trace->open)
1608                         iter->trace->open(iter);
1609         } else {
1610                 kfree(iter);
1611                 iter = NULL;
1612         }
1613         mutex_unlock(&trace_types_lock);
1614
1615  out:
1616         return iter;
1617 }
1618
1619 int tracing_open_generic(struct inode *inode, struct file *filp)
1620 {
1621         if (tracing_disabled)
1622                 return -ENODEV;
1623
1624         filp->private_data = inode->i_private;
1625         return 0;
1626 }
1627
1628 int tracing_release(struct inode *inode, struct file *file)
1629 {
1630         struct seq_file *m = (struct seq_file *)file->private_data;
1631         struct trace_iterator *iter = m->private;
1632
1633         mutex_lock(&trace_types_lock);
1634         if (iter->trace && iter->trace->close)
1635                 iter->trace->close(iter);
1636
1637         /* reenable tracing if it was previously enabled */
1638         if (iter->tr->ctrl)
1639                 tracer_enabled = 1;
1640         mutex_unlock(&trace_types_lock);
1641
1642         seq_release(inode, file);
1643         kfree(iter);
1644         return 0;
1645 }
1646
1647 static int tracing_open(struct inode *inode, struct file *file)
1648 {
1649         int ret;
1650
1651         __tracing_open(inode, file, &ret);
1652
1653         return ret;
1654 }
1655
1656 static int tracing_lt_open(struct inode *inode, struct file *file)
1657 {
1658         struct trace_iterator *iter;
1659         int ret;
1660
1661         iter = __tracing_open(inode, file, &ret);
1662
1663         if (!ret)
1664                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1665
1666         return ret;
1667 }
1668
1669
1670 static void *
1671 t_next(struct seq_file *m, void *v, loff_t *pos)
1672 {
1673         struct tracer *t = m->private;
1674
1675         (*pos)++;
1676
1677         if (t)
1678                 t = t->next;
1679
1680         m->private = t;
1681
1682         return t;
1683 }
1684
1685 static void *t_start(struct seq_file *m, loff_t *pos)
1686 {
1687         struct tracer *t = m->private;
1688         loff_t l = 0;
1689
1690         mutex_lock(&trace_types_lock);
1691         for (; t && l < *pos; t = t_next(m, t, &l))
1692                 ;
1693
1694         return t;
1695 }
1696
1697 static void t_stop(struct seq_file *m, void *p)
1698 {
1699         mutex_unlock(&trace_types_lock);
1700 }
1701
1702 static int t_show(struct seq_file *m, void *v)
1703 {
1704         struct tracer *t = v;
1705
1706         if (!t)
1707                 return 0;
1708
1709         seq_printf(m, "%s", t->name);
1710         if (t->next)
1711                 seq_putc(m, ' ');
1712         else
1713                 seq_putc(m, '\n');
1714
1715         return 0;
1716 }
1717
1718 static struct seq_operations show_traces_seq_ops = {
1719         .start          = t_start,
1720         .next           = t_next,
1721         .stop           = t_stop,
1722         .show           = t_show,
1723 };
1724
1725 static int show_traces_open(struct inode *inode, struct file *file)
1726 {
1727         int ret;
1728
1729         if (tracing_disabled)
1730                 return -ENODEV;
1731
1732         ret = seq_open(file, &show_traces_seq_ops);
1733         if (!ret) {
1734                 struct seq_file *m = file->private_data;
1735                 m->private = trace_types;
1736         }
1737
1738         return ret;
1739 }
1740
1741 static struct file_operations tracing_fops = {
1742         .open           = tracing_open,
1743         .read           = seq_read,
1744         .llseek         = seq_lseek,
1745         .release        = tracing_release,
1746 };
1747
1748 static struct file_operations tracing_lt_fops = {
1749         .open           = tracing_lt_open,
1750         .read           = seq_read,
1751         .llseek         = seq_lseek,
1752         .release        = tracing_release,
1753 };
1754
1755 static struct file_operations show_traces_fops = {
1756         .open = show_traces_open,
1757         .read = seq_read,
1758         .release = seq_release,
1759 };
1760
1761 static ssize_t
1762 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1763                        size_t cnt, loff_t *ppos)
1764 {
1765         char *buf;
1766         int r = 0;
1767         int len = 0;
1768         int i;
1769
1770         /* calulate max size */
1771         for (i = 0; trace_options[i]; i++) {
1772                 len += strlen(trace_options[i]);
1773                 len += 3; /* "no" and space */
1774         }
1775
1776         /* +2 for \n and \0 */
1777         buf = kmalloc(len + 2, GFP_KERNEL);
1778         if (!buf)
1779                 return -ENOMEM;
1780
1781         for (i = 0; trace_options[i]; i++) {
1782                 if (trace_flags & (1 << i))
1783                         r += sprintf(buf + r, "%s ", trace_options[i]);
1784                 else
1785                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1786         }
1787
1788         r += sprintf(buf + r, "\n");
1789         WARN_ON(r >= len + 2);
1790
1791         r = simple_read_from_buffer(ubuf, cnt, ppos,
1792                                     buf, r);
1793
1794         kfree(buf);
1795
1796         return r;
1797 }
1798
1799 static ssize_t
1800 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1801                         size_t cnt, loff_t *ppos)
1802 {
1803         char buf[64];
1804         char *cmp = buf;
1805         int neg = 0;
1806         int i;
1807
1808         if (cnt > 63)
1809                 cnt = 63;
1810
1811         if (copy_from_user(&buf, ubuf, cnt))
1812                 return -EFAULT;
1813
1814         buf[cnt] = 0;
1815
1816         if (strncmp(buf, "no", 2) == 0) {
1817                 neg = 1;
1818                 cmp += 2;
1819         }
1820
1821         for (i = 0; trace_options[i]; i++) {
1822                 int len = strlen(trace_options[i]);
1823
1824                 if (strncmp(cmp, trace_options[i], len) == 0) {
1825                         if (neg)
1826                                 trace_flags &= ~(1 << i);
1827                         else
1828                                 trace_flags |= (1 << i);
1829                         break;
1830                 }
1831         }
1832
1833         filp->f_pos += cnt;
1834
1835         return cnt;
1836 }
1837
1838 static struct file_operations tracing_iter_fops = {
1839         .open = tracing_open_generic,
1840         .read = tracing_iter_ctrl_read,
1841         .write = tracing_iter_ctrl_write,
1842 };
1843
1844 static const char readme_msg[] =
1845         "tracing mini-HOWTO:\n\n"
1846         "# mkdir /debug\n"
1847         "# mount -t debugfs nodev /debug\n\n"
1848         "# cat /debug/tracing/available_tracers\n"
1849         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1850         "# cat /debug/tracing/current_tracer\n"
1851         "none\n"
1852         "# echo sched_switch > /debug/tracing/current_tracer\n"
1853         "# cat /debug/tracing/current_tracer\n"
1854         "sched_switch\n"
1855         "# cat /debug/tracing/iter_ctrl\n"
1856         "noprint-parent nosym-offset nosym-addr noverbose\n"
1857         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1858         "# echo 1 > /debug/tracing/tracing_enabled\n"
1859         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1860         "echo 0 > /debug/tracing/tracing_enabled\n"
1861 ;
1862
1863 static ssize_t
1864 tracing_readme_read(struct file *filp, char __user *ubuf,
1865                        size_t cnt, loff_t *ppos)
1866 {
1867         return simple_read_from_buffer(ubuf, cnt, ppos,
1868                                         readme_msg, strlen(readme_msg));
1869 }
1870
1871 static struct file_operations tracing_readme_fops = {
1872         .open = tracing_open_generic,
1873         .read = tracing_readme_read,
1874 };
1875
1876 static ssize_t
1877 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1878                   size_t cnt, loff_t *ppos)
1879 {
1880         struct trace_array *tr = filp->private_data;
1881         char buf[64];
1882         int r;
1883
1884         r = sprintf(buf, "%ld\n", tr->ctrl);
1885         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1886 }
1887
1888 static ssize_t
1889 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1890                    size_t cnt, loff_t *ppos)
1891 {
1892         struct trace_array *tr = filp->private_data;
1893         long val;
1894         char buf[64];
1895
1896         if (cnt > 63)
1897                 cnt = 63;
1898
1899         if (copy_from_user(&buf, ubuf, cnt))
1900                 return -EFAULT;
1901
1902         buf[cnt] = 0;
1903
1904         val = simple_strtoul(buf, NULL, 10);
1905
1906         val = !!val;
1907
1908         mutex_lock(&trace_types_lock);
1909         if (tr->ctrl ^ val) {
1910                 if (val)
1911                         tracer_enabled = 1;
1912                 else
1913                         tracer_enabled = 0;
1914
1915                 tr->ctrl = val;
1916
1917                 if (current_trace && current_trace->ctrl_update)
1918                         current_trace->ctrl_update(tr);
1919         }
1920         mutex_unlock(&trace_types_lock);
1921
1922         filp->f_pos += cnt;
1923
1924         return cnt;
1925 }
1926
1927 static ssize_t
1928 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1929                        size_t cnt, loff_t *ppos)
1930 {
1931         char buf[max_tracer_type_len+2];
1932         int r;
1933
1934         mutex_lock(&trace_types_lock);
1935         if (current_trace)
1936                 r = sprintf(buf, "%s\n", current_trace->name);
1937         else
1938                 r = sprintf(buf, "\n");
1939         mutex_unlock(&trace_types_lock);
1940
1941         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1942 }
1943
1944 static ssize_t
1945 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1946                         size_t cnt, loff_t *ppos)
1947 {
1948         struct trace_array *tr = &global_trace;
1949         struct tracer *t;
1950         char buf[max_tracer_type_len+1];
1951         int i;
1952
1953         if (cnt > max_tracer_type_len)
1954                 cnt = max_tracer_type_len;
1955
1956         if (copy_from_user(&buf, ubuf, cnt))
1957                 return -EFAULT;
1958
1959         buf[cnt] = 0;
1960
1961         /* strip ending whitespace. */
1962         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1963                 buf[i] = 0;
1964
1965         mutex_lock(&trace_types_lock);
1966         for (t = trace_types; t; t = t->next) {
1967                 if (strcmp(t->name, buf) == 0)
1968                         break;
1969         }
1970         if (!t || t == current_trace)
1971                 goto out;
1972
1973         if (current_trace && current_trace->reset)
1974                 current_trace->reset(tr);
1975
1976         current_trace = t;
1977         if (t->init)
1978                 t->init(tr);
1979
1980  out:
1981         mutex_unlock(&trace_types_lock);
1982
1983         filp->f_pos += cnt;
1984
1985         return cnt;
1986 }
1987
1988 static ssize_t
1989 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1990                      size_t cnt, loff_t *ppos)
1991 {
1992         unsigned long *ptr = filp->private_data;
1993         char buf[64];
1994         int r;
1995
1996         r = snprintf(buf, 64, "%ld\n",
1997                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1998         if (r > 64)
1999                 r = 64;
2000         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2001 }
2002
2003 static ssize_t
2004 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2005                       size_t cnt, loff_t *ppos)
2006 {
2007         long *ptr = filp->private_data;
2008         long val;
2009         char buf[64];
2010
2011         if (cnt > 63)
2012                 cnt = 63;
2013
2014         if (copy_from_user(&buf, ubuf, cnt))
2015                 return -EFAULT;
2016
2017         buf[cnt] = 0;
2018
2019         val = simple_strtoul(buf, NULL, 10);
2020
2021         *ptr = val * 1000;
2022
2023         return cnt;
2024 }
2025
2026 static atomic_t tracing_reader;
2027
2028 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2029 {
2030         struct trace_iterator *iter;
2031
2032         if (tracing_disabled)
2033                 return -ENODEV;
2034
2035         /* We only allow for reader of the pipe */
2036         if (atomic_inc_return(&tracing_reader) != 1) {
2037                 atomic_dec(&tracing_reader);
2038                 return -EBUSY;
2039         }
2040
2041         /* create a buffer to store the information to pass to userspace */
2042         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2043         if (!iter)
2044                 return -ENOMEM;
2045
2046         iter->tr = &global_trace;
2047
2048         filp->private_data = iter;
2049
2050         return 0;
2051 }
2052
2053 static int tracing_release_pipe(struct inode *inode, struct file *file)
2054 {
2055         struct trace_iterator *iter = file->private_data;
2056
2057         kfree(iter);
2058         atomic_dec(&tracing_reader);
2059
2060         return 0;
2061 }
2062
2063 static unsigned int
2064 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2065 {
2066         struct trace_iterator *iter = filp->private_data;
2067
2068         if (trace_flags & TRACE_ITER_BLOCK) {
2069                 /*
2070                  * Always select as readable when in blocking mode
2071                  */
2072                 return POLLIN | POLLRDNORM;
2073         }
2074         else {
2075                 if (!trace_empty(iter))
2076                         return POLLIN | POLLRDNORM;
2077                 poll_wait(filp, &trace_wait, poll_table);
2078                 if (!trace_empty(iter))
2079                         return POLLIN | POLLRDNORM;
2080
2081                 return 0;
2082         }
2083 }
2084
2085 /*
2086  * Consumer reader.
2087  */
2088 static ssize_t
2089 tracing_read_pipe(struct file *filp, char __user *ubuf,
2090                   size_t cnt, loff_t *ppos)
2091 {
2092         struct trace_iterator *iter = filp->private_data;
2093         struct trace_array_cpu *data;
2094         static cpumask_t mask;
2095         static int start;
2096         unsigned long flags;
2097 #ifdef CONFIG_FTRACE
2098         int ftrace_save;
2099 #endif
2100         int read = 0;
2101         int cpu;
2102         int len;
2103         int ret;
2104
2105         /* return any leftover data */
2106         if (iter->seq.len > start) {
2107                 len = iter->seq.len - start;
2108                 if (cnt > len)
2109                         cnt = len;
2110                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2111                 if (ret)
2112                         cnt = -EFAULT;
2113
2114                 start += len;
2115
2116                 return cnt;
2117         }
2118
2119         trace_seq_reset(&iter->seq);
2120         start = 0;
2121
2122         while (trace_empty(iter)) {
2123                 if (!(trace_flags & TRACE_ITER_BLOCK))
2124                         return -EWOULDBLOCK;
2125                 /*
2126                  * This is a make-shift waitqueue. The reason we don't use
2127                  * an actual wait queue is because:
2128                  *  1) we only ever have one waiter
2129                  *  2) the tracing, traces all functions, we don't want
2130                  *     the overhead of calling wake_up and friends
2131                  *     (and tracing them too)
2132                  *     Anyway, this is really very primitive wakeup.
2133                  */
2134                 set_current_state(TASK_INTERRUPTIBLE);
2135                 iter->tr->waiter = current;
2136
2137                 /* sleep for one second, and try again. */
2138                 schedule_timeout(HZ);
2139
2140                 iter->tr->waiter = NULL;
2141
2142                 if (signal_pending(current))
2143                         return -EINTR;
2144
2145                 /*
2146                  * We block until we read something and tracing is disabled.
2147                  * We still block if tracing is disabled, but we have never
2148                  * read anything. This allows a user to cat this file, and
2149                  * then enable tracing. But after we have read something,
2150                  * we give an EOF when tracing is again disabled.
2151                  *
2152                  * iter->pos will be 0 if we haven't read anything.
2153                  */
2154                 if (!tracer_enabled && iter->pos)
2155                         break;
2156
2157                 continue;
2158         }
2159
2160         /* stop when tracing is finished */
2161         if (trace_empty(iter))
2162                 return 0;
2163
2164         if (cnt >= PAGE_SIZE)
2165                 cnt = PAGE_SIZE - 1;
2166
2167         memset(iter, 0, sizeof(*iter));
2168         iter->tr = &global_trace;
2169         iter->pos = -1;
2170
2171         /*
2172          * We need to stop all tracing on all CPUS to read the
2173          * the next buffer. This is a bit expensive, but is
2174          * not done often. We fill all what we can read,
2175          * and then release the locks again.
2176          */
2177
2178         cpus_clear(mask);
2179         local_irq_save(flags);
2180 #ifdef CONFIG_FTRACE
2181         ftrace_save = ftrace_enabled;
2182         ftrace_enabled = 0;
2183 #endif
2184         smp_wmb();
2185         for_each_possible_cpu(cpu) {
2186                 data = iter->tr->data[cpu];
2187
2188                 if (!head_page(data) || !data->trace_idx)
2189                         continue;
2190
2191                 atomic_inc(&data->disabled);
2192                 cpu_set(cpu, mask);
2193         }
2194
2195         for_each_cpu_mask(cpu, mask) {
2196                 data = iter->tr->data[cpu];
2197                 spin_lock(&data->lock);
2198         }
2199
2200         while (find_next_entry_inc(iter) != NULL) {
2201                 int len = iter->seq.len;
2202
2203                 ret = print_trace_line(iter);
2204                 if (!ret) {
2205                         /* don't print partial lines */
2206                         iter->seq.len = len;
2207                         break;
2208                 }
2209
2210                 trace_consume(iter);
2211
2212                 if (iter->seq.len >= cnt)
2213                         break;
2214         }
2215
2216         for_each_cpu_mask(cpu, mask) {
2217                 data = iter->tr->data[cpu];
2218                 spin_unlock(&data->lock);
2219         }
2220
2221         for_each_cpu_mask(cpu, mask) {
2222                 data = iter->tr->data[cpu];
2223                 atomic_dec(&data->disabled);
2224         }
2225 #ifdef CONFIG_FTRACE
2226         ftrace_enabled = ftrace_save;
2227 #endif
2228         local_irq_restore(flags);
2229
2230         /* Now copy what we have to the user */
2231         read = iter->seq.len;
2232         if (read > cnt)
2233                 read = cnt;
2234
2235         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2236
2237         if (read < iter->seq.len)
2238                 start = read;
2239         else
2240                 trace_seq_reset(&iter->seq);
2241
2242         if (ret)
2243                 read = -EFAULT;
2244
2245         return read;
2246 }
2247
2248 static struct file_operations tracing_max_lat_fops = {
2249         .open           = tracing_open_generic,
2250         .read           = tracing_max_lat_read,
2251         .write          = tracing_max_lat_write,
2252 };
2253
2254 static struct file_operations tracing_ctrl_fops = {
2255         .open           = tracing_open_generic,
2256         .read           = tracing_ctrl_read,
2257         .write          = tracing_ctrl_write,
2258 };
2259
2260 static struct file_operations set_tracer_fops = {
2261         .open           = tracing_open_generic,
2262         .read           = tracing_set_trace_read,
2263         .write          = tracing_set_trace_write,
2264 };
2265
2266 static struct file_operations tracing_pipe_fops = {
2267         .open           = tracing_open_pipe,
2268         .poll           = tracing_poll_pipe,
2269         .read           = tracing_read_pipe,
2270         .release        = tracing_release_pipe,
2271 };
2272
2273 #ifdef CONFIG_DYNAMIC_FTRACE
2274
2275 static ssize_t
2276 tracing_read_long(struct file *filp, char __user *ubuf,
2277                   size_t cnt, loff_t *ppos)
2278 {
2279         unsigned long *p = filp->private_data;
2280         char buf[64];
2281         int r;
2282
2283         r = sprintf(buf, "%ld\n", *p);
2284
2285         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2286 }
2287
2288 static struct file_operations tracing_read_long_fops = {
2289         .open           = tracing_open_generic,
2290         .read           = tracing_read_long,
2291 };
2292 #endif
2293
2294 static struct dentry *d_tracer;
2295
2296 struct dentry *tracing_init_dentry(void)
2297 {
2298         static int once;
2299
2300         if (d_tracer)
2301                 return d_tracer;
2302
2303         d_tracer = debugfs_create_dir("tracing", NULL);
2304
2305         if (!d_tracer && !once) {
2306                 once = 1;
2307                 pr_warning("Could not create debugfs directory 'tracing'\n");
2308                 return NULL;
2309         }
2310
2311         return d_tracer;
2312 }
2313
2314 #ifdef CONFIG_FTRACE_SELFTEST
2315 /* Let selftest have access to static functions in this file */
2316 #include "trace_selftest.c"
2317 #endif
2318
2319 static __init void tracer_init_debugfs(void)
2320 {
2321         struct dentry *d_tracer;
2322         struct dentry *entry;
2323
2324         d_tracer = tracing_init_dentry();
2325
2326         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2327                                     &global_trace, &tracing_ctrl_fops);
2328         if (!entry)
2329                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2330
2331         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2332                                     NULL, &tracing_iter_fops);
2333         if (!entry)
2334                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2335
2336         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2337                                     &global_trace, &tracing_lt_fops);
2338         if (!entry)
2339                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2340
2341         entry = debugfs_create_file("trace", 0444, d_tracer,
2342                                     &global_trace, &tracing_fops);
2343         if (!entry)
2344                 pr_warning("Could not create debugfs 'trace' entry\n");
2345
2346         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2347                                     &global_trace, &show_traces_fops);
2348         if (!entry)
2349                 pr_warning("Could not create debugfs 'trace' entry\n");
2350
2351         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2352                                     &global_trace, &set_tracer_fops);
2353         if (!entry)
2354                 pr_warning("Could not create debugfs 'trace' entry\n");
2355
2356         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2357                                     &tracing_max_latency,
2358                                     &tracing_max_lat_fops);
2359         if (!entry)
2360                 pr_warning("Could not create debugfs "
2361                            "'tracing_max_latency' entry\n");
2362
2363         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2364                                     &tracing_thresh, &tracing_max_lat_fops);
2365         if (!entry)
2366                 pr_warning("Could not create debugfs "
2367                            "'tracing_threash' entry\n");
2368         entry = debugfs_create_file("README", 0644, d_tracer,
2369                                     NULL, &tracing_readme_fops);
2370         if (!entry)
2371                 pr_warning("Could not create debugfs 'README' entry\n");
2372
2373         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2374                                     NULL, &tracing_pipe_fops);
2375         if (!entry)
2376                 pr_warning("Could not create debugfs "
2377                            "'tracing_threash' entry\n");
2378
2379 #ifdef CONFIG_DYNAMIC_FTRACE
2380         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2381                                     &ftrace_update_tot_cnt,
2382                                     &tracing_read_long_fops);
2383         if (!entry)
2384                 pr_warning("Could not create debugfs "
2385                            "'dyn_ftrace_total_info' entry\n");
2386 #endif
2387 }
2388
2389 /* dummy trace to disable tracing */
2390 static struct tracer no_tracer __read_mostly =
2391 {
2392         .name           = "none",
2393 };
2394
2395 static int trace_alloc_page(void)
2396 {
2397         struct trace_array_cpu *data;
2398         struct page *page, *tmp;
2399         LIST_HEAD(pages);
2400         void *array;
2401         int i;
2402
2403         /* first allocate a page for each CPU */
2404         for_each_possible_cpu(i) {
2405                 array = (void *)__get_free_page(GFP_KERNEL);
2406                 if (array == NULL) {
2407                         printk(KERN_ERR "tracer: failed to allocate page"
2408                                "for trace buffer!\n");
2409                         goto free_pages;
2410                 }
2411
2412                 page = virt_to_page(array);
2413                 list_add(&page->lru, &pages);
2414
2415 /* Only allocate if we are actually using the max trace */
2416 #ifdef CONFIG_TRACER_MAX_TRACE
2417                 array = (void *)__get_free_page(GFP_KERNEL);
2418                 if (array == NULL) {
2419                         printk(KERN_ERR "tracer: failed to allocate page"
2420                                "for trace buffer!\n");
2421                         goto free_pages;
2422                 }
2423                 page = virt_to_page(array);
2424                 list_add(&page->lru, &pages);
2425 #endif
2426         }
2427
2428         /* Now that we successfully allocate a page per CPU, add them */
2429         for_each_possible_cpu(i) {
2430                 data = global_trace.data[i];
2431                 spin_lock_init(&data->lock);
2432                 lockdep_set_class(&data->lock, &data->lock_key);
2433                 page = list_entry(pages.next, struct page, lru);
2434                 list_del_init(&page->lru);
2435                 list_add_tail(&page->lru, &data->trace_pages);
2436                 ClearPageLRU(page);
2437
2438 #ifdef CONFIG_TRACER_MAX_TRACE
2439                 data = max_tr.data[i];
2440                 spin_lock_init(&data->lock);
2441                 lockdep_set_class(&data->lock, &data->lock_key);
2442                 page = list_entry(pages.next, struct page, lru);
2443                 list_del_init(&page->lru);
2444                 list_add_tail(&page->lru, &data->trace_pages);
2445                 SetPageLRU(page);
2446 #endif
2447         }
2448         global_trace.entries += ENTRIES_PER_PAGE;
2449
2450         return 0;
2451
2452  free_pages:
2453         list_for_each_entry_safe(page, tmp, &pages, lru) {
2454                 list_del_init(&page->lru);
2455                 __free_page(page);
2456         }
2457         return -ENOMEM;
2458 }
2459
2460 __init static int tracer_alloc_buffers(void)
2461 {
2462         struct trace_array_cpu *data;
2463         void *array;
2464         struct page *page;
2465         int pages = 0;
2466         int ret = -ENOMEM;
2467         int i;
2468
2469         global_trace.ctrl = tracer_enabled;
2470
2471         /* Allocate the first page for all buffers */
2472         for_each_possible_cpu(i) {
2473                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2474                 max_tr.data[i] = &per_cpu(max_data, i);
2475
2476                 array = (void *)__get_free_page(GFP_KERNEL);
2477                 if (array == NULL) {
2478                         printk(KERN_ERR "tracer: failed to allocate page"
2479                                "for trace buffer!\n");
2480                         goto free_buffers;
2481                 }
2482
2483                 /* set the array to the list */
2484                 INIT_LIST_HEAD(&data->trace_pages);
2485                 page = virt_to_page(array);
2486                 list_add(&page->lru, &data->trace_pages);
2487                 /* use the LRU flag to differentiate the two buffers */
2488                 ClearPageLRU(page);
2489
2490 /* Only allocate if we are actually using the max trace */
2491 #ifdef CONFIG_TRACER_MAX_TRACE
2492                 array = (void *)__get_free_page(GFP_KERNEL);
2493                 if (array == NULL) {
2494                         printk(KERN_ERR "tracer: failed to allocate page"
2495                                "for trace buffer!\n");
2496                         goto free_buffers;
2497                 }
2498
2499                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2500                 page = virt_to_page(array);
2501                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2502                 SetPageLRU(page);
2503 #endif
2504         }
2505
2506         /*
2507          * Since we allocate by orders of pages, we may be able to
2508          * round up a bit.
2509          */
2510         global_trace.entries = ENTRIES_PER_PAGE;
2511         pages++;
2512
2513         while (global_trace.entries < trace_nr_entries) {
2514                 if (trace_alloc_page())
2515                         break;
2516                 pages++;
2517         }
2518         max_tr.entries = global_trace.entries;
2519
2520         pr_info("tracer: %d pages allocated for %ld",
2521                 pages, trace_nr_entries);
2522         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2523         pr_info("   actual entries %ld\n", global_trace.entries);
2524
2525         tracer_init_debugfs();
2526
2527         trace_init_cmdlines();
2528
2529         register_tracer(&no_tracer);
2530         current_trace = &no_tracer;
2531
2532         /* All seems OK, enable tracing */
2533         tracing_disabled = 0;
2534
2535         return 0;
2536
2537  free_buffers:
2538         for (i-- ; i >= 0; i--) {
2539                 struct page *page, *tmp;
2540                 struct trace_array_cpu *data = global_trace.data[i];
2541
2542                 if (data) {
2543                         list_for_each_entry_safe(page, tmp,
2544                                                  &data->trace_pages, lru) {
2545                                 list_del_init(&page->lru);
2546                                 __free_page(page);
2547                         }
2548                 }
2549
2550 #ifdef CONFIG_TRACER_MAX_TRACE
2551                 data = max_tr.data[i];
2552                 if (data) {
2553                         list_for_each_entry_safe(page, tmp,
2554                                                  &data->trace_pages, lru) {
2555                                 list_del_init(&page->lru);
2556                                 __free_page(page);
2557                         }
2558                 }
2559 #endif
2560         }
2561         return ret;
2562 }
2563 fs_initcall(tracer_alloc_buffers);