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 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_flag_type {
 100         TRACE_FLAG_IRQS_OFF             = 0x01,
 101         TRACE_FLAG_NEED_RESCHED         = 0x02,
 102         TRACE_FLAG_HARDIRQ              = 0x04,
 103         TRACE_FLAG_SOFTIRQ              = 0x08,
 104 };
 105
 106 #define TRACE_ITER_SYM_MASK \
 107         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 108
 109 /* These must match the bit postions above */
 110 static const char *trace_options[] = {
 111         "print-parent",
 112         "sym-offset",
 113         "sym-addr",
 114         "verbose",
 115         "raw",
 116         "hex",
 117         "bin",
 118         "block",
 119         "stacktrace",
 120         "sched-tree",
 121         NULL
 122 };
 123
 124 static raw_spinlock_t ftrace_max_lock =
 125         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 126
 127 /*
 128  * Copy the new maximum trace into the separate maximum-trace
 129  * structure. (this way the maximum trace is permanently saved,
 130  * for later retrieval via /debugfs/tracing/latency_trace)
 131  */
 132 static void
 133 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 134 {
 135         struct trace_array_cpu *data = tr->data[cpu];
 136
 137         max_tr.cpu = cpu;
 138         max_tr.time_start = data->preempt_timestamp;
 139
 140         data = max_tr.data[cpu];
 141         data->saved_latency = tracing_max_latency;
 142
 143         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 144         data->pid = tsk->pid;
 145         data->uid = tsk->uid;
 146         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 147         data->policy = tsk->policy;
 148         data->rt_priority = tsk->rt_priority;
 149
 150         /* record this tasks comm */
 151         tracing_record_cmdline(current);
 152 }
 153
 154 void check_pages(struct trace_array_cpu *data)
 155 {
 156         struct page *page, *tmp;
 157
 158         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 159         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 160
 161         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 162                 BUG_ON(page->lru.next->prev != &page->lru);
 163                 BUG_ON(page->lru.prev->next != &page->lru);
 164         }
 165 }
 166
 167 void *head_page(struct trace_array_cpu *data)
 168 {
 169         struct page *page;
 170
 171         check_pages(data);
 172         if (list_empty(&data->trace_pages))
 173                 return NULL;
 174
 175         page = list_entry(data->trace_pages.next, struct page, lru);
 176         BUG_ON(&page->lru == &data->trace_pages);
 177
 178         return page_address(page);
 179 }
 180
 181 int
 182 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 183 {
 184         int len = (PAGE_SIZE - 1) - s->len;
 185         va_list ap;
 186         int ret;
 187
 188         if (!len)
 189                 return 0;
 190
 191         va_start(ap, fmt);
 192         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 193         va_end(ap);
 194
 195         /* If we can't write it all, don't bother writing anything */
 196         if (ret >= len)
 197                 return 0;
 198
 199         s->len += ret;
 200
 201         return len;
 202 }
 203
 204 static int
 205 trace_seq_puts(struct trace_seq *s, const char *str)
 206 {
 207         int len = strlen(str);
 208
 209         if (len > ((PAGE_SIZE - 1) - s->len))
 210                 return 0;
 211
 212         memcpy(s->buffer + s->len, str, len);
 213         s->len += len;
 214
 215         return len;
 216 }
 217
 218 static int
 219 trace_seq_putc(struct trace_seq *s, unsigned char c)
 220 {
 221         if (s->len >= (PAGE_SIZE - 1))
 222                 return 0;
 223
 224         s->buffer[s->len++] = c;
 225
 226         return 1;
 227 }
 228
 229 static int
 230 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 231 {
 232         if (len > ((PAGE_SIZE - 1) - s->len))
 233                 return 0;
 234
 235         memcpy(s->buffer + s->len, mem, len);
 236         s->len += len;
 237
 238         return len;
 239 }
 240
 241 #define HEX_CHARS 17
 242
 243 static int
 244 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 245 {
 246         unsigned char hex[HEX_CHARS];
 247         unsigned char *data;
 248         unsigned char byte;
 249         int i, j;
 250
 251         BUG_ON(len >= HEX_CHARS);
 252
 253         data = mem;
 254
 255 #ifdef __BIG_ENDIAN
 256         for (i = 0, j = 0; i < len; i++) {
 257 #else
 258         for (i = len-1, j = 0; i >= 0; i--) {
 259 #endif
 260                 byte = data[i];
 261
 262                 hex[j]   = byte & 0x0f;
 263                 if (hex[j] >= 10)
 264                         hex[j] += 'a' - 10;
 265                 else
 266                         hex[j] += '0';
 267                 j++;
 268
 269                 hex[j] = byte >> 4;
 270                 if (hex[j] >= 10)
 271                         hex[j] += 'a' - 10;
 272                 else
 273                         hex[j] += '0';
 274                 j++;
 275         }
 276         hex[j] = ' ';
 277         j++;
 278
 279         return trace_seq_putmem(s, hex, j);
 280 }
 281
 282 static void
 283 trace_seq_reset(struct trace_seq *s)
 284 {
 285         s->len = 0;
 286 }
 287
 288 static void
 289 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 290 {
 291         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 292
 293         s->buffer[len] = 0;
 294         seq_puts(m, s->buffer);
 295
 296         trace_seq_reset(s);
 297 }
 298
 299 static void
 300 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 301 {
 302         struct list_head flip_pages;
 303
 304         INIT_LIST_HEAD(&flip_pages);
 305
 306         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 307                 sizeof(struct trace_array_cpu) -
 308                 offsetof(struct trace_array_cpu, trace_head_idx));
 309
 310         check_pages(tr1);
 311         check_pages(tr2);
 312         list_splice_init(&tr1->trace_pages, &flip_pages);
 313         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 314         list_splice_init(&flip_pages, &tr2->trace_pages);
 315         BUG_ON(!list_empty(&flip_pages));
 316         check_pages(tr1);
 317         check_pages(tr2);
 318 }
 319
 320 void
 321 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 322 {
 323         struct trace_array_cpu *data;
 324         int i;
 325
 326         WARN_ON_ONCE(!irqs_disabled());
 327         __raw_spin_lock(&ftrace_max_lock);
 328         /* clear out all the previous traces */
 329         for_each_possible_cpu(i) {
 330                 data = tr->data[i];
 331                 flip_trace(max_tr.data[i], data);
 332                 tracing_reset(data);
 333         }
 334
 335         __update_max_tr(tr, tsk, cpu);
 336         __raw_spin_unlock(&ftrace_max_lock);
 337 }
 338
 339 /**
 340  * update_max_tr_single - only copy one trace over, and reset the rest
 341  * @tr - tracer
 342  * @tsk - task with the latency
 343  * @cpu - the cpu of the buffer to copy.
 344  */
 345 void
 346 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 347 {
 348         struct trace_array_cpu *data = tr->data[cpu];
 349         int i;
 350
 351         WARN_ON_ONCE(!irqs_disabled());
 352         __raw_spin_lock(&ftrace_max_lock);
 353         for_each_possible_cpu(i)
 354                 tracing_reset(max_tr.data[i]);
 355
 356         flip_trace(max_tr.data[cpu], data);
 357         tracing_reset(data);
 358
 359         __update_max_tr(tr, tsk, cpu);
 360         __raw_spin_unlock(&ftrace_max_lock);
 361 }
 362
 363 int register_tracer(struct tracer *type)
 364 {
 365         struct tracer *t;
 366         int len;
 367         int ret = 0;
 368
 369         if (!type->name) {
 370                 pr_info("Tracer must have a name\n");
 371                 return -1;
 372         }
 373
 374         mutex_lock(&trace_types_lock);
 375         for (t = trace_types; t; t = t->next) {
 376                 if (strcmp(type->name, t->name) == 0) {
 377                         /* already found */
 378                         pr_info("Trace %s already registered\n",
 379                                 type->name);
 380                         ret = -1;
 381                         goto out;
 382                 }
 383         }
 384
 385 #ifdef CONFIG_FTRACE_STARTUP_TEST
 386         if (type->selftest) {
 387                 struct tracer *saved_tracer = current_trace;
 388                 struct trace_array_cpu *data;
 389                 struct trace_array *tr = &global_trace;
 390                 int saved_ctrl = tr->ctrl;
 391                 int i;
 392                 /*
 393                  * Run a selftest on this tracer.
 394                  * Here we reset the trace buffer, and set the current
 395                  * tracer to be this tracer. The tracer can then run some
 396                  * internal tracing to verify that everything is in order.
 397                  * If we fail, we do not register this tracer.
 398                  */
 399                 for_each_possible_cpu(i) {
 400                         data = tr->data[i];
 401                         if (!head_page(data))
 402                                 continue;
 403                         tracing_reset(data);
 404                 }
 405                 current_trace = type;
 406                 tr->ctrl = 0;
 407                 /* the test is responsible for initializing and enabling */
 408                 pr_info("Testing tracer %s: ", type->name);
 409                 ret = type->selftest(type, tr);
 410                 /* the test is responsible for resetting too */
 411                 current_trace = saved_tracer;
 412                 tr->ctrl = saved_ctrl;
 413                 if (ret) {
 414                         printk(KERN_CONT "FAILED!\n");
 415                         goto out;
 416                 }
 417                 /* Only reset on passing, to avoid touching corrupted buffers */
 418                 for_each_possible_cpu(i) {
 419                         data = tr->data[i];
 420                         if (!head_page(data))
 421                                 continue;
 422                         tracing_reset(data);
 423                 }
 424                 printk(KERN_CONT "PASSED\n");
 425         }
 426 #endif
 427
 428         type->next = trace_types;
 429         trace_types = type;
 430         len = strlen(type->name);
 431         if (len > max_tracer_type_len)
 432                 max_tracer_type_len = len;
 433
 434  out:
 435         mutex_unlock(&trace_types_lock);
 436
 437         return ret;
 438 }
 439
 440 void unregister_tracer(struct tracer *type)
 441 {
 442         struct tracer **t;
 443         int len;
 444
 445         mutex_lock(&trace_types_lock);
 446         for (t = &trace_types; *t; t = &(*t)->next) {
 447                 if (*t == type)
 448                         goto found;
 449         }
 450         pr_info("Trace %s not registered\n", type->name);
 451         goto out;
 452
 453  found:
 454         *t = (*t)->next;
 455         if (strlen(type->name) != max_tracer_type_len)
 456                 goto out;
 457
 458         max_tracer_type_len = 0;
 459         for (t = &trace_types; *t; t = &(*t)->next) {
 460                 len = strlen((*t)->name);
 461                 if (len > max_tracer_type_len)
 462                         max_tracer_type_len = len;
 463         }
 464  out:
 465         mutex_unlock(&trace_types_lock);
 466 }
 467
 468 void tracing_reset(struct trace_array_cpu *data)
 469 {
 470         data->trace_idx = 0;
 471         data->trace_head = data->trace_tail = head_page(data);
 472         data->trace_head_idx = 0;
 473         data->trace_tail_idx = 0;
 474 }
 475
 476 #define SAVED_CMDLINES 128
 477 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 478 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 479 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 480 static int cmdline_idx;
 481 static DEFINE_SPINLOCK(trace_cmdline_lock);
 482 atomic_t trace_record_cmdline_disabled;
 483
 484 static void trace_init_cmdlines(void)
 485 {
 486         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 487         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 488         cmdline_idx = 0;
 489 }
 490
 491 void trace_stop_cmdline_recording(void);
 492
 493 static void trace_save_cmdline(struct task_struct *tsk)
 494 {
 495         unsigned map;
 496         unsigned idx;
 497
 498         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 499                 return;
 500
 501         /*
 502          * It's not the end of the world if we don't get
 503          * the lock, but we also don't want to spin
 504          * nor do we want to disable interrupts,
 505          * so if we miss here, then better luck next time.
 506          */
 507         if (!spin_trylock(&trace_cmdline_lock))
 508                 return;
 509
 510         idx = map_pid_to_cmdline[tsk->pid];
 511         if (idx >= SAVED_CMDLINES) {
 512                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 513
 514                 map = map_cmdline_to_pid[idx];
 515                 if (map <= PID_MAX_DEFAULT)
 516                         map_pid_to_cmdline[map] = (unsigned)-1;
 517
 518                 map_pid_to_cmdline[tsk->pid] = idx;
 519
 520                 cmdline_idx = idx;
 521         }
 522
 523         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 524
 525         spin_unlock(&trace_cmdline_lock);
 526 }
 527
 528 static char *trace_find_cmdline(int pid)
 529 {
 530         char *cmdline = "<...>";
 531         unsigned map;
 532
 533         if (!pid)
 534                 return "<idle>";
 535
 536         if (pid > PID_MAX_DEFAULT)
 537                 goto out;
 538
 539         map = map_pid_to_cmdline[pid];
 540         if (map >= SAVED_CMDLINES)
 541                 goto out;
 542
 543         cmdline = saved_cmdlines[map];
 544
 545  out:
 546         return cmdline;
 547 }
 548
 549 void tracing_record_cmdline(struct task_struct *tsk)
 550 {
 551         if (atomic_read(&trace_record_cmdline_disabled))
 552                 return;
 553
 554         trace_save_cmdline(tsk);
 555 }
 556
 557 static inline struct list_head *
 558 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 559 {
 560         /*
 561          * Roundrobin - but skip the head (which is not a real page):
 562          */
 563         next = next->next;
 564         if (unlikely(next == &data->trace_pages))
 565                 next = next->next;
 566         BUG_ON(next == &data->trace_pages);
 567
 568         return next;
 569 }
 570
 571 static inline void *
 572 trace_next_page(struct trace_array_cpu *data, void *addr)
 573 {
 574         struct list_head *next;
 575         struct page *page;
 576
 577         page = virt_to_page(addr);
 578
 579         next = trace_next_list(data, &page->lru);
 580         page = list_entry(next, struct page, lru);
 581
 582         return page_address(page);
 583 }
 584
 585 static inline struct trace_entry *
 586 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 587 {
 588         unsigned long idx, idx_next;
 589         struct trace_entry *entry;
 590
 591         data->trace_idx++;
 592         idx = data->trace_head_idx;
 593         idx_next = idx + 1;
 594
 595         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 596
 597         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 598
 599         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 600                 data->trace_head = trace_next_page(data, data->trace_head);
 601                 idx_next = 0;
 602         }
 603
 604         if (data->trace_head == data->trace_tail &&
 605             idx_next == data->trace_tail_idx) {
 606                 /* overrun */
 607                 data->trace_tail_idx++;
 608                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 609                         data->trace_tail =
 610                                 trace_next_page(data, data->trace_tail);
 611                         data->trace_tail_idx = 0;
 612                 }
 613         }
 614
 615         data->trace_head_idx = idx_next;
 616
 617         return entry;
 618 }
 619
 620 static inline void
 621 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 622 {
 623         struct task_struct *tsk = current;
 624         unsigned long pc;
 625
 626         pc = preempt_count();
 627
 628         entry->preempt_count    = pc & 0xff;
 629         entry->pid              = (tsk) ? tsk->pid : 0;
 630         entry->t                = ftrace_now(raw_smp_processor_id());
 631         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 632                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 633                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 634                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 635 }
 636
 637 void
 638 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 639                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 640 {
 641         struct trace_entry *entry;
 642         unsigned long irq_flags;
 643
 644         raw_local_irq_save(irq_flags);
 645         __raw_spin_lock(&data->lock);
 646         entry                   = tracing_get_trace_entry(tr, data);
 647         tracing_generic_entry_update(entry, flags);
 648         entry->type             = TRACE_FN;
 649         entry->fn.ip            = ip;
 650         entry->fn.parent_ip     = parent_ip;
 651         __raw_spin_unlock(&data->lock);
 652         raw_local_irq_restore(irq_flags);
 653 }
 654
 655 void
 656 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 657        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 658 {
 659         if (likely(!atomic_read(&data->disabled)))
 660                 trace_function(tr, data, ip, parent_ip, flags);
 661 }
 662
 663 void
 664 __trace_special(void *__tr, void *__data,
 665                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 666 {
 667         struct trace_array_cpu *data = __data;
 668         struct trace_array *tr = __tr;
 669         struct trace_entry *entry;
 670         unsigned long irq_flags;
 671
 672         raw_local_irq_save(irq_flags);
 673         __raw_spin_lock(&data->lock);
 674         entry                   = tracing_get_trace_entry(tr, data);
 675         tracing_generic_entry_update(entry, 0);
 676         entry->type             = TRACE_SPECIAL;
 677         entry->special.arg1     = arg1;
 678         entry->special.arg2     = arg2;
 679         entry->special.arg3     = arg3;
 680         __raw_spin_unlock(&data->lock);
 681         raw_local_irq_restore(irq_flags);
 682
 683         trace_wake_up();
 684 }
 685
 686 void __trace_stack(struct trace_array *tr,
 687                    struct trace_array_cpu *data,
 688                    unsigned long flags,
 689                    int skip)
 690 {
 691         struct trace_entry *entry;
 692         struct stack_trace trace;
 693
 694         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 695                 return;
 696
 697         entry                   = tracing_get_trace_entry(tr, data);
 698         tracing_generic_entry_update(entry, flags);
 699         entry->type             = TRACE_STACK;
 700
 701         memset(&entry->stack, 0, sizeof(entry->stack));
 702
 703         trace.nr_entries        = 0;
 704         trace.max_entries       = FTRACE_STACK_ENTRIES;
 705         trace.skip              = skip;
 706         trace.entries           = entry->stack.caller;
 707
 708         save_stack_trace(&trace);
 709 }
 710
 711 void
 712 tracing_sched_switch_trace(struct trace_array *tr,
 713                            struct trace_array_cpu *data,
 714                            struct task_struct *prev,
 715                            struct task_struct *next,
 716                            unsigned long flags)
 717 {
 718         struct trace_entry *entry;
 719         unsigned long irq_flags;
 720
 721         raw_local_irq_save(irq_flags);
 722         __raw_spin_lock(&data->lock);
 723         entry                   = tracing_get_trace_entry(tr, data);
 724         tracing_generic_entry_update(entry, flags);
 725         entry->type             = TRACE_CTX;
 726         entry->ctx.prev_pid     = prev->pid;
 727         entry->ctx.prev_prio    = prev->prio;
 728         entry->ctx.prev_state   = prev->state;
 729         entry->ctx.next_pid     = next->pid;
 730         entry->ctx.next_prio    = next->prio;
 731         entry->ctx.next_state   = next->state;
 732         __trace_stack(tr, data, flags, 4);
 733         __raw_spin_unlock(&data->lock);
 734         raw_local_irq_restore(irq_flags);
 735 }
 736
 737 void
 738 tracing_sched_wakeup_trace(struct trace_array *tr,
 739                            struct trace_array_cpu *data,
 740                            struct task_struct *wakee,
 741                            struct task_struct *curr,
 742                            unsigned long flags)
 743 {
 744         struct trace_entry *entry;
 745         unsigned long irq_flags;
 746
 747         raw_local_irq_save(irq_flags);
 748         __raw_spin_lock(&data->lock);
 749         entry                   = tracing_get_trace_entry(tr, data);
 750         tracing_generic_entry_update(entry, flags);
 751         entry->type             = TRACE_WAKE;
 752         entry->ctx.prev_pid     = curr->pid;
 753         entry->ctx.prev_prio    = curr->prio;
 754         entry->ctx.prev_state   = curr->state;
 755         entry->ctx.next_pid     = wakee->pid;
 756         entry->ctx.next_prio    = wakee->prio;
 757         entry->ctx.next_state   = wakee->state;
 758         __trace_stack(tr, data, flags, 5);
 759         __raw_spin_unlock(&data->lock);
 760         raw_local_irq_restore(irq_flags);
 761
 762         trace_wake_up();
 763 }
 764
 765 #ifdef CONFIG_FTRACE
 766 static void
 767 function_trace_call(unsigned long ip, unsigned long parent_ip)
 768 {
 769         struct trace_array *tr = &global_trace;
 770         struct trace_array_cpu *data;
 771         unsigned long flags;
 772         long disabled;
 773         int cpu;
 774
 775         if (unlikely(!tracer_enabled))
 776                 return;
 777
 778         local_irq_save(flags);
 779         cpu = raw_smp_processor_id();
 780         data = tr->data[cpu];
 781         disabled = atomic_inc_return(&data->disabled);
 782
 783         if (likely(disabled == 1))
 784                 trace_function(tr, data, ip, parent_ip, flags);
 785
 786         atomic_dec(&data->disabled);
 787         local_irq_restore(flags);
 788 }
 789
 790 static struct ftrace_ops trace_ops __read_mostly =
 791 {
 792         .func = function_trace_call,
 793 };
 794
 795 void tracing_start_function_trace(void)
 796 {
 797         register_ftrace_function(&trace_ops);
 798 }
 799
 800 void tracing_stop_function_trace(void)
 801 {
 802         unregister_ftrace_function(&trace_ops);
 803 }
 804 #endif
 805
 806 enum trace_file_type {
 807         TRACE_FILE_LAT_FMT      = 1,
 808 };
 809
 810 static struct trace_entry *
 811 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 812                 struct trace_iterator *iter, int cpu)
 813 {
 814         struct page *page;
 815         struct trace_entry *array;
 816
 817         if (iter->next_idx[cpu] >= tr->entries ||
 818             iter->next_idx[cpu] >= data->trace_idx ||
 819             (data->trace_head == data->trace_tail &&
 820              data->trace_head_idx == data->trace_tail_idx))
 821                 return NULL;
 822
 823         if (!iter->next_page[cpu]) {
 824                 /* Initialize the iterator for this cpu trace buffer */
 825                 WARN_ON(!data->trace_tail);
 826                 page = virt_to_page(data->trace_tail);
 827                 iter->next_page[cpu] = &page->lru;
 828                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 829         }
 830
 831         page = list_entry(iter->next_page[cpu], struct page, lru);
 832         BUG_ON(&data->trace_pages == &page->lru);
 833
 834         array = page_address(page);
 835
 836         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 837         return &array[iter->next_page_idx[cpu]];
 838 }
 839
 840 static struct trace_entry *
 841 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 842 {
 843         struct trace_array *tr = iter->tr;
 844         struct trace_entry *ent, *next = NULL;
 845         int next_cpu = -1;
 846         int cpu;
 847
 848         for_each_possible_cpu(cpu) {
 849                 if (!head_page(tr->data[cpu]))
 850                         continue;
 851                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 852                 /*
 853                  * Pick the entry with the smallest timestamp:
 854                  */
 855                 if (ent && (!next || ent->t < next->t)) {
 856                         next = ent;
 857                         next_cpu = cpu;
 858                 }
 859         }
 860
 861         if (ent_cpu)
 862                 *ent_cpu = next_cpu;
 863
 864         return next;
 865 }
 866
 867 static void trace_iterator_increment(struct trace_iterator *iter)
 868 {
 869         iter->idx++;
 870         iter->next_idx[iter->cpu]++;
 871         iter->next_page_idx[iter->cpu]++;
 872
 873         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 874                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 875
 876                 iter->next_page_idx[iter->cpu] = 0;
 877                 iter->next_page[iter->cpu] =
 878                         trace_next_list(data, iter->next_page[iter->cpu]);
 879         }
 880 }
 881
 882 static void trace_consume(struct trace_iterator *iter)
 883 {
 884         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 885
 886         data->trace_tail_idx++;
 887         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 888                 data->trace_tail = trace_next_page(data, data->trace_tail);
 889                 data->trace_tail_idx = 0;
 890         }
 891
 892         /* Check if we empty it, then reset the index */
 893         if (data->trace_head == data->trace_tail &&
 894             data->trace_head_idx == data->trace_tail_idx)
 895                 data->trace_idx = 0;
 896 }
 897
 898 static void *find_next_entry_inc(struct trace_iterator *iter)
 899 {
 900         struct trace_entry *next;
 901         int next_cpu = -1;
 902
 903         next = find_next_entry(iter, &next_cpu);
 904
 905         iter->prev_ent = iter->ent;
 906         iter->prev_cpu = iter->cpu;
 907
 908         iter->ent = next;
 909         iter->cpu = next_cpu;
 910
 911         if (next)
 912                 trace_iterator_increment(iter);
 913
 914         return next ? iter : NULL;
 915 }
 916
 917 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 918 {
 919         struct trace_iterator *iter = m->private;
 920         void *last_ent = iter->ent;
 921         int i = (int)*pos;
 922         void *ent;
 923
 924         (*pos)++;
 925
 926         /* can't go backwards */
 927         if (iter->idx > i)
 928                 return NULL;
 929
 930         if (iter->idx < 0)
 931                 ent = find_next_entry_inc(iter);
 932         else
 933                 ent = iter;
 934
 935         while (ent && iter->idx < i)
 936                 ent = find_next_entry_inc(iter);
 937
 938         iter->pos = *pos;
 939
 940         if (last_ent && !ent)
 941                 seq_puts(m, "\n\nvim:ft=help\n");
 942
 943         return ent;
 944 }
 945
 946 static void *s_start(struct seq_file *m, loff_t *pos)
 947 {
 948         struct trace_iterator *iter = m->private;
 949         void *p = NULL;
 950         loff_t l = 0;
 951         int i;
 952
 953         mutex_lock(&trace_types_lock);
 954
 955         if (!current_trace || current_trace != iter->trace) {
 956                 mutex_unlock(&trace_types_lock);
 957                 return NULL;
 958         }
 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, T;
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                 lat_print_generic(s, entry, cpu);
1230                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1231         }
1232         switch (entry->type) {
1233         case TRACE_FN:
1234                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1235                 trace_seq_puts(s, " (");
1236                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1237                 trace_seq_puts(s, ")\n");
1238                 break;
1239         case TRACE_CTX:
1240         case TRACE_WAKE:
1241                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1242                         state_to_char[entry->ctx.prev_state] : 'X';
1243                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1244                         state_to_char[entry->ctx.next_state] : 'X';
1245
1246                 comm = trace_find_cmdline(entry->ctx.next_pid);
1247                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1248                                  entry->ctx.prev_pid,
1249                                  entry->ctx.prev_prio,
1250                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1251                                  entry->ctx.next_pid,
1252                                  entry->ctx.next_prio,
1253                                  T, comm);
1254                 break;
1255         case TRACE_SPECIAL:
1256                 trace_seq_printf(s, "# %ld %ld %ld\n",
1257                                  entry->special.arg1,
1258                                  entry->special.arg2,
1259                                  entry->special.arg3);
1260                 break;
1261         case TRACE_STACK:
1262                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1263                         if (i)
1264                                 trace_seq_puts(s, " <= ");
1265                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1266                 }
1267                 trace_seq_puts(s, "\n");
1268                 break;
1269         default:
1270                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1271         }
1272         return 1;
1273 }
1274
1275 static int print_trace_fmt(struct trace_iterator *iter)
1276 {
1277         struct trace_seq *s = &iter->seq;
1278         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1279         struct trace_entry *entry;
1280         unsigned long usec_rem;
1281         unsigned long long t;
1282         unsigned long secs;
1283         char *comm;
1284         int ret;
1285         int S, T;
1286         int i;
1287
1288         entry = iter->ent;
1289
1290         comm = trace_find_cmdline(iter->ent->pid);
1291
1292         t = ns2usecs(entry->t);
1293         usec_rem = do_div(t, 1000000ULL);
1294         secs = (unsigned long)t;
1295
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         switch (entry->type) {
1307         case TRACE_FN:
1308                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1309                 if (!ret)
1310                         return 0;
1311                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1312                                                 entry->fn.parent_ip) {
1313                         ret = trace_seq_printf(s, " <-");
1314                         if (!ret)
1315                                 return 0;
1316                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1317                                                sym_flags);
1318                         if (!ret)
1319                                 return 0;
1320                 }
1321                 ret = trace_seq_printf(s, "\n");
1322                 if (!ret)
1323                         return 0;
1324                 break;
1325         case TRACE_CTX:
1326         case TRACE_WAKE:
1327                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1328                         state_to_char[entry->ctx.prev_state] : 'X';
1329                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1330                         state_to_char[entry->ctx.next_state] : 'X';
1331                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1332                                        entry->ctx.prev_pid,
1333                                        entry->ctx.prev_prio,
1334                                        S,
1335                                        entry->type == TRACE_CTX ? "==>" : "  +",
1336                                        entry->ctx.next_pid,
1337                                        entry->ctx.next_prio,
1338                                        T);
1339                 if (!ret)
1340                         return 0;
1341                 break;
1342         case TRACE_SPECIAL:
1343                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1344                                  entry->special.arg1,
1345                                  entry->special.arg2,
1346                                  entry->special.arg3);
1347                 if (!ret)
1348                         return 0;
1349                 break;
1350         case TRACE_STACK:
1351                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1352                         if (i) {
1353                                 ret = trace_seq_puts(s, " <= ");
1354                                 if (!ret)
1355                                         return 0;
1356                         }
1357                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1358                                                sym_flags);
1359                         if (!ret)
1360                                 return 0;
1361                 }
1362                 ret = trace_seq_puts(s, "\n");
1363                 if (!ret)
1364                         return 0;
1365                 break;
1366         }
1367         return 1;
1368 }
1369
1370 static int print_raw_fmt(struct trace_iterator *iter)
1371 {
1372         struct trace_seq *s = &iter->seq;
1373         struct trace_entry *entry;
1374         int ret;
1375         int S, T;
1376
1377         entry = iter->ent;
1378
1379         ret = trace_seq_printf(s, "%d %d %llu ",
1380                 entry->pid, iter->cpu, entry->t);
1381         if (!ret)
1382                 return 0;
1383
1384         switch (entry->type) {
1385         case TRACE_FN:
1386                 ret = trace_seq_printf(s, "%x %x\n",
1387                                         entry->fn.ip, entry->fn.parent_ip);
1388                 if (!ret)
1389                         return 0;
1390                 break;
1391         case TRACE_CTX:
1392         case TRACE_WAKE:
1393                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1394                         state_to_char[entry->ctx.prev_state] : 'X';
1395                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1396                         state_to_char[entry->ctx.next_state] : 'X';
1397                 if (entry->type == TRACE_WAKE)
1398                         S = '+';
1399                 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1400                                        entry->ctx.prev_pid,
1401                                        entry->ctx.prev_prio,
1402                                        S,
1403                                        entry->ctx.next_pid,
1404                                        entry->ctx.next_prio,
1405                                        T);
1406                 if (!ret)
1407                         return 0;
1408                 break;
1409         case TRACE_SPECIAL:
1410         case TRACE_STACK:
1411                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1412                                  entry->special.arg1,
1413                                  entry->special.arg2,
1414                                  entry->special.arg3);
1415                 if (!ret)
1416                         return 0;
1417                 break;
1418         }
1419         return 1;
1420 }
1421
1422 #define SEQ_PUT_FIELD_RET(s, x)                         \
1423 do {                                                    \
1424         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1425                 return 0;                               \
1426 } while (0)
1427
1428 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1429 do {                                                    \
1430         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1431                 return 0;                               \
1432 } while (0)
1433
1434 static int print_hex_fmt(struct trace_iterator *iter)
1435 {
1436         struct trace_seq *s = &iter->seq;
1437         unsigned char newline = '\n';
1438         struct trace_entry *entry;
1439         int S, T;
1440
1441         entry = iter->ent;
1442
1443         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1444         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1445         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1446
1447         switch (entry->type) {
1448         case TRACE_FN:
1449                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1450                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1451                 break;
1452         case TRACE_CTX:
1453         case TRACE_WAKE:
1454                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1455                         state_to_char[entry->ctx.prev_state] : 'X';
1456                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1457                         state_to_char[entry->ctx.next_state] : 'X';
1458                 if (entry->type == TRACE_WAKE)
1459                         S = '+';
1460                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1461                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1462                 SEQ_PUT_HEX_FIELD_RET(s, S);
1463                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1464                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1465                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1466                 SEQ_PUT_HEX_FIELD_RET(s, T);
1467                 break;
1468         case TRACE_SPECIAL:
1469         case TRACE_STACK:
1470                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1471                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1472                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1473                 break;
1474         }
1475         SEQ_PUT_FIELD_RET(s, newline);
1476
1477         return 1;
1478 }
1479
1480 static int print_bin_fmt(struct trace_iterator *iter)
1481 {
1482         struct trace_seq *s = &iter->seq;
1483         struct trace_entry *entry;
1484
1485         entry = iter->ent;
1486
1487         SEQ_PUT_FIELD_RET(s, entry->pid);
1488         SEQ_PUT_FIELD_RET(s, entry->cpu);
1489         SEQ_PUT_FIELD_RET(s, entry->t);
1490
1491         switch (entry->type) {
1492         case TRACE_FN:
1493                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1494                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1495                 break;
1496         case TRACE_CTX:
1497                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1498                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1499                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1500                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1501                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1502                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1503                 break;
1504         case TRACE_SPECIAL:
1505         case TRACE_STACK:
1506                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1507                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1508                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1509                 break;
1510         }
1511         return 1;
1512 }
1513
1514 static int trace_empty(struct trace_iterator *iter)
1515 {
1516         struct trace_array_cpu *data;
1517         int cpu;
1518
1519         for_each_possible_cpu(cpu) {
1520                 data = iter->tr->data[cpu];
1521
1522                 if (head_page(data) && data->trace_idx &&
1523                     (data->trace_tail != data->trace_head ||
1524                      data->trace_tail_idx != data->trace_head_idx))
1525                         return 0;
1526         }
1527         return 1;
1528 }
1529
1530 static int print_trace_line(struct trace_iterator *iter)
1531 {
1532         if (iter->trace && iter->trace->print_line)
1533                 return iter->trace->print_line(iter);
1534
1535         if (trace_flags & TRACE_ITER_BIN)
1536                 return print_bin_fmt(iter);
1537
1538         if (trace_flags & TRACE_ITER_HEX)
1539                 return print_hex_fmt(iter);
1540
1541         if (trace_flags & TRACE_ITER_RAW)
1542                 return print_raw_fmt(iter);
1543
1544         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1545                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1546
1547         return print_trace_fmt(iter);
1548 }
1549
1550 static int s_show(struct seq_file *m, void *v)
1551 {
1552         struct trace_iterator *iter = v;
1553
1554         if (iter->ent == NULL) {
1555                 if (iter->tr) {
1556                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1557                         seq_puts(m, "#\n");
1558                 }
1559                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1560                         /* print nothing if the buffers are empty */
1561                         if (trace_empty(iter))
1562                                 return 0;
1563                         print_trace_header(m, iter);
1564                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1565                                 print_lat_help_header(m);
1566                 } else {
1567                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1568                                 print_func_help_header(m);
1569                 }
1570         } else {
1571                 print_trace_line(iter);
1572                 trace_print_seq(m, &iter->seq);
1573         }
1574
1575         return 0;
1576 }
1577
1578 static struct seq_operations tracer_seq_ops = {
1579         .start          = s_start,
1580         .next           = s_next,
1581         .stop           = s_stop,
1582         .show           = s_show,
1583 };
1584
1585 static struct trace_iterator *
1586 __tracing_open(struct inode *inode, struct file *file, int *ret)
1587 {
1588         struct trace_iterator *iter;
1589
1590         if (tracing_disabled) {
1591                 *ret = -ENODEV;
1592                 return NULL;
1593         }
1594
1595         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1596         if (!iter) {
1597                 *ret = -ENOMEM;
1598                 goto out;
1599         }
1600
1601         mutex_lock(&trace_types_lock);
1602         if (current_trace && current_trace->print_max)
1603                 iter->tr = &max_tr;
1604         else
1605                 iter->tr = inode->i_private;
1606         iter->trace = current_trace;
1607         iter->pos = -1;
1608
1609         /* TODO stop tracer */
1610         *ret = seq_open(file, &tracer_seq_ops);
1611         if (!*ret) {
1612                 struct seq_file *m = file->private_data;
1613                 m->private = iter;
1614
1615                 /* stop the trace while dumping */
1616                 if (iter->tr->ctrl)
1617                         tracer_enabled = 0;
1618
1619                 if (iter->trace && iter->trace->open)
1620                         iter->trace->open(iter);
1621         } else {
1622                 kfree(iter);
1623                 iter = NULL;
1624         }
1625         mutex_unlock(&trace_types_lock);
1626
1627  out:
1628         return iter;
1629 }
1630
1631 int tracing_open_generic(struct inode *inode, struct file *filp)
1632 {
1633         if (tracing_disabled)
1634                 return -ENODEV;
1635
1636         filp->private_data = inode->i_private;
1637         return 0;
1638 }
1639
1640 int tracing_release(struct inode *inode, struct file *file)
1641 {
1642         struct seq_file *m = (struct seq_file *)file->private_data;
1643         struct trace_iterator *iter = m->private;
1644
1645         mutex_lock(&trace_types_lock);
1646         if (iter->trace && iter->trace->close)
1647                 iter->trace->close(iter);
1648
1649         /* reenable tracing if it was previously enabled */
1650         if (iter->tr->ctrl)
1651                 tracer_enabled = 1;
1652         mutex_unlock(&trace_types_lock);
1653
1654         seq_release(inode, file);
1655         kfree(iter);
1656         return 0;
1657 }
1658
1659 static int tracing_open(struct inode *inode, struct file *file)
1660 {
1661         int ret;
1662
1663         __tracing_open(inode, file, &ret);
1664
1665         return ret;
1666 }
1667
1668 static int tracing_lt_open(struct inode *inode, struct file *file)
1669 {
1670         struct trace_iterator *iter;
1671         int ret;
1672
1673         iter = __tracing_open(inode, file, &ret);
1674
1675         if (!ret)
1676                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1677
1678         return ret;
1679 }
1680
1681
1682 static void *
1683 t_next(struct seq_file *m, void *v, loff_t *pos)
1684 {
1685         struct tracer *t = m->private;
1686
1687         (*pos)++;
1688
1689         if (t)
1690                 t = t->next;
1691
1692         m->private = t;
1693
1694         return t;
1695 }
1696
1697 static void *t_start(struct seq_file *m, loff_t *pos)
1698 {
1699         struct tracer *t = m->private;
1700         loff_t l = 0;
1701
1702         mutex_lock(&trace_types_lock);
1703         for (; t && l < *pos; t = t_next(m, t, &l))
1704                 ;
1705
1706         return t;
1707 }
1708
1709 static void t_stop(struct seq_file *m, void *p)
1710 {
1711         mutex_unlock(&trace_types_lock);
1712 }
1713
1714 static int t_show(struct seq_file *m, void *v)
1715 {
1716         struct tracer *t = v;
1717
1718         if (!t)
1719                 return 0;
1720
1721         seq_printf(m, "%s", t->name);
1722         if (t->next)
1723                 seq_putc(m, ' ');
1724         else
1725                 seq_putc(m, '\n');
1726
1727         return 0;
1728 }
1729
1730 static struct seq_operations show_traces_seq_ops = {
1731         .start          = t_start,
1732         .next           = t_next,
1733         .stop           = t_stop,
1734         .show           = t_show,
1735 };
1736
1737 static int show_traces_open(struct inode *inode, struct file *file)
1738 {
1739         int ret;
1740
1741         if (tracing_disabled)
1742                 return -ENODEV;
1743
1744         ret = seq_open(file, &show_traces_seq_ops);
1745         if (!ret) {
1746                 struct seq_file *m = file->private_data;
1747                 m->private = trace_types;
1748         }
1749
1750         return ret;
1751 }
1752
1753 static struct file_operations tracing_fops = {
1754         .open           = tracing_open,
1755         .read           = seq_read,
1756         .llseek         = seq_lseek,
1757         .release        = tracing_release,
1758 };
1759
1760 static struct file_operations tracing_lt_fops = {
1761         .open           = tracing_lt_open,
1762         .read           = seq_read,
1763         .llseek         = seq_lseek,
1764         .release        = tracing_release,
1765 };
1766
1767 static struct file_operations show_traces_fops = {
1768         .open           = show_traces_open,
1769         .read           = seq_read,
1770         .release        = seq_release,
1771 };
1772
1773 /*
1774  * Only trace on a CPU if the bitmask is set:
1775  */
1776 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1777
1778 /*
1779  * When tracing/tracing_cpu_mask is modified then this holds
1780  * the new bitmask we are about to install:
1781  */
1782 static cpumask_t tracing_cpumask_new;
1783
1784 /*
1785  * The tracer itself will not take this lock, but still we want
1786  * to provide a consistent cpumask to user-space:
1787  */
1788 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1789
1790 /*
1791  * Temporary storage for the character representation of the
1792  * CPU bitmask (and one more byte for the newline):
1793  */
1794 static char mask_str[NR_CPUS + 1];
1795
1796 static ssize_t
1797 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1798                      size_t count, loff_t *ppos)
1799 {
1800         int len;
1801
1802         mutex_lock(&tracing_cpumask_update_lock);
1803
1804         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1805         if (count - len < 2) {
1806                 count = -EINVAL;
1807                 goto out_err;
1808         }
1809         len += sprintf(mask_str + len, "\n");
1810         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1811
1812 out_err:
1813         mutex_unlock(&tracing_cpumask_update_lock);
1814
1815         return count;
1816 }
1817
1818 static ssize_t
1819 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1820                       size_t count, loff_t *ppos)
1821 {
1822         int err, cpu;
1823
1824         mutex_lock(&tracing_cpumask_update_lock);
1825         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1826         if (err)
1827                 goto err_unlock;
1828
1829         raw_local_irq_disable();
1830         __raw_spin_lock(&ftrace_max_lock);
1831         for_each_possible_cpu(cpu) {
1832                 /*
1833                  * Increase/decrease the disabled counter if we are
1834                  * about to flip a bit in the cpumask:
1835                  */
1836                 if (cpu_isset(cpu, tracing_cpumask) &&
1837                                 !cpu_isset(cpu, tracing_cpumask_new)) {
1838                         atomic_inc(&global_trace.data[cpu]->disabled);
1839                 }
1840                 if (!cpu_isset(cpu, tracing_cpumask) &&
1841                                 cpu_isset(cpu, tracing_cpumask_new)) {
1842                         atomic_dec(&global_trace.data[cpu]->disabled);
1843                 }
1844         }
1845         __raw_spin_unlock(&ftrace_max_lock);
1846         raw_local_irq_enable();
1847
1848         tracing_cpumask = tracing_cpumask_new;
1849
1850         mutex_unlock(&tracing_cpumask_update_lock);
1851
1852         return count;
1853
1854 err_unlock:
1855         mutex_unlock(&tracing_cpumask_update_lock);
1856
1857         return err;
1858 }
1859
1860 static struct file_operations tracing_cpumask_fops = {
1861         .open           = tracing_open_generic,
1862         .read           = tracing_cpumask_read,
1863         .write          = tracing_cpumask_write,
1864 };
1865
1866 static ssize_t
1867 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1868                        size_t cnt, loff_t *ppos)
1869 {
1870         char *buf;
1871         int r = 0;
1872         int len = 0;
1873         int i;
1874
1875         /* calulate max size */
1876         for (i = 0; trace_options[i]; i++) {
1877                 len += strlen(trace_options[i]);
1878                 len += 3; /* "no" and space */
1879         }
1880
1881         /* +2 for \n and \0 */
1882         buf = kmalloc(len + 2, GFP_KERNEL);
1883         if (!buf)
1884                 return -ENOMEM;
1885
1886         for (i = 0; trace_options[i]; i++) {
1887                 if (trace_flags & (1 << i))
1888                         r += sprintf(buf + r, "%s ", trace_options[i]);
1889                 else
1890                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1891         }
1892
1893         r += sprintf(buf + r, "\n");
1894         WARN_ON(r >= len + 2);
1895
1896         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1897
1898         kfree(buf);
1899
1900         return r;
1901 }
1902
1903 static ssize_t
1904 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1905                         size_t cnt, loff_t *ppos)
1906 {
1907         char buf[64];
1908         char *cmp = buf;
1909         int neg = 0;
1910         int i;
1911
1912         if (cnt > 63)
1913                 cnt = 63;
1914
1915         if (copy_from_user(&buf, ubuf, cnt))
1916                 return -EFAULT;
1917
1918         buf[cnt] = 0;
1919
1920         if (strncmp(buf, "no", 2) == 0) {
1921                 neg = 1;
1922                 cmp += 2;
1923         }
1924
1925         for (i = 0; trace_options[i]; i++) {
1926                 int len = strlen(trace_options[i]);
1927
1928                 if (strncmp(cmp, trace_options[i], len) == 0) {
1929                         if (neg)
1930                                 trace_flags &= ~(1 << i);
1931                         else
1932                                 trace_flags |= (1 << i);
1933                         break;
1934                 }
1935         }
1936         /*
1937          * If no option could be set, return an error:
1938          */
1939         if (!trace_options[i])
1940                 return -EINVAL;
1941
1942         filp->f_pos += cnt;
1943
1944         return cnt;
1945 }
1946
1947 static struct file_operations tracing_iter_fops = {
1948         .open           = tracing_open_generic,
1949         .read           = tracing_iter_ctrl_read,
1950         .write          = tracing_iter_ctrl_write,
1951 };
1952
1953 static const char readme_msg[] =
1954         "tracing mini-HOWTO:\n\n"
1955         "# mkdir /debug\n"
1956         "# mount -t debugfs nodev /debug\n\n"
1957         "# cat /debug/tracing/available_tracers\n"
1958         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1959         "# cat /debug/tracing/current_tracer\n"
1960         "none\n"
1961         "# echo sched_switch > /debug/tracing/current_tracer\n"
1962         "# cat /debug/tracing/current_tracer\n"
1963         "sched_switch\n"
1964         "# cat /debug/tracing/iter_ctrl\n"
1965         "noprint-parent nosym-offset nosym-addr noverbose\n"
1966         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1967         "# echo 1 > /debug/tracing/tracing_enabled\n"
1968         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1969         "echo 0 > /debug/tracing/tracing_enabled\n"
1970 ;
1971
1972 static ssize_t
1973 tracing_readme_read(struct file *filp, char __user *ubuf,
1974                        size_t cnt, loff_t *ppos)
1975 {
1976         return simple_read_from_buffer(ubuf, cnt, ppos,
1977                                         readme_msg, strlen(readme_msg));
1978 }
1979
1980 static struct file_operations tracing_readme_fops = {
1981         .open           = tracing_open_generic,
1982         .read           = tracing_readme_read,
1983 };
1984
1985 static ssize_t
1986 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1987                   size_t cnt, loff_t *ppos)
1988 {
1989         struct trace_array *tr = filp->private_data;
1990         char buf[64];
1991         int r;
1992
1993         r = sprintf(buf, "%ld\n", tr->ctrl);
1994         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1995 }
1996
1997 static ssize_t
1998 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1999                    size_t cnt, loff_t *ppos)
2000 {
2001         struct trace_array *tr = filp->private_data;
2002         long val;
2003         char buf[64];
2004
2005         if (cnt > 63)
2006                 cnt = 63;
2007
2008         if (copy_from_user(&buf, ubuf, cnt))
2009                 return -EFAULT;
2010
2011         buf[cnt] = 0;
2012
2013         val = simple_strtoul(buf, NULL, 10);
2014
2015         val = !!val;
2016
2017         mutex_lock(&trace_types_lock);
2018         if (tr->ctrl ^ val) {
2019                 if (val)
2020                         tracer_enabled = 1;
2021                 else
2022                         tracer_enabled = 0;
2023
2024                 tr->ctrl = val;
2025
2026                 if (current_trace && current_trace->ctrl_update)
2027                         current_trace->ctrl_update(tr);
2028         }
2029         mutex_unlock(&trace_types_lock);
2030
2031         filp->f_pos += cnt;
2032
2033         return cnt;
2034 }
2035
2036 static ssize_t
2037 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2038                        size_t cnt, loff_t *ppos)
2039 {
2040         char buf[max_tracer_type_len+2];
2041         int r;
2042
2043         mutex_lock(&trace_types_lock);
2044         if (current_trace)
2045                 r = sprintf(buf, "%s\n", current_trace->name);
2046         else
2047                 r = sprintf(buf, "\n");
2048         mutex_unlock(&trace_types_lock);
2049
2050         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2051 }
2052
2053 static ssize_t
2054 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2055                         size_t cnt, loff_t *ppos)
2056 {
2057         struct trace_array *tr = &global_trace;
2058         struct tracer *t;
2059         char buf[max_tracer_type_len+1];
2060         int i;
2061
2062         if (cnt > max_tracer_type_len)
2063                 cnt = max_tracer_type_len;
2064
2065         if (copy_from_user(&buf, ubuf, cnt))
2066                 return -EFAULT;
2067
2068         buf[cnt] = 0;
2069
2070         /* strip ending whitespace. */
2071         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2072                 buf[i] = 0;
2073
2074         mutex_lock(&trace_types_lock);
2075         for (t = trace_types; t; t = t->next) {
2076                 if (strcmp(t->name, buf) == 0)
2077                         break;
2078         }
2079         if (!t || t == current_trace)
2080                 goto out;
2081
2082         if (current_trace && current_trace->reset)
2083                 current_trace->reset(tr);
2084
2085         current_trace = t;
2086         if (t->init)
2087                 t->init(tr);
2088
2089  out:
2090         mutex_unlock(&trace_types_lock);
2091
2092         filp->f_pos += cnt;
2093
2094         return cnt;
2095 }
2096
2097 static ssize_t
2098 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2099                      size_t cnt, loff_t *ppos)
2100 {
2101         unsigned long *ptr = filp->private_data;
2102         char buf[64];
2103         int r;
2104
2105         r = snprintf(buf, 64, "%ld\n",
2106                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2107         if (r > 64)
2108                 r = 64;
2109         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2110 }
2111
2112 static ssize_t
2113 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2114                       size_t cnt, loff_t *ppos)
2115 {
2116         long *ptr = filp->private_data;
2117         long val;
2118         char buf[64];
2119
2120         if (cnt > 63)
2121                 cnt = 63;
2122
2123         if (copy_from_user(&buf, ubuf, cnt))
2124                 return -EFAULT;
2125
2126         buf[cnt] = 0;
2127
2128         val = simple_strtoul(buf, NULL, 10);
2129
2130         *ptr = val * 1000;
2131
2132         return cnt;
2133 }
2134
2135 static atomic_t tracing_reader;
2136
2137 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2138 {
2139         struct trace_iterator *iter;
2140
2141         if (tracing_disabled)
2142                 return -ENODEV;
2143
2144         /* We only allow for reader of the pipe */
2145         if (atomic_inc_return(&tracing_reader) != 1) {
2146                 atomic_dec(&tracing_reader);
2147                 return -EBUSY;
2148         }
2149
2150         /* create a buffer to store the information to pass to userspace */
2151         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2152         if (!iter)
2153                 return -ENOMEM;
2154
2155         iter->tr = &global_trace;
2156         iter->trace = current_trace;
2157
2158         filp->private_data = iter;
2159
2160         return 0;
2161 }
2162
2163 static int tracing_release_pipe(struct inode *inode, struct file *file)
2164 {
2165         struct trace_iterator *iter = file->private_data;
2166
2167         kfree(iter);
2168         atomic_dec(&tracing_reader);
2169
2170         return 0;
2171 }
2172
2173 static unsigned int
2174 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2175 {
2176         struct trace_iterator *iter = filp->private_data;
2177
2178         if (trace_flags & TRACE_ITER_BLOCK) {
2179                 /*
2180                  * Always select as readable when in blocking mode
2181                  */
2182                 return POLLIN | POLLRDNORM;
2183         }
2184         else {
2185                 if (!trace_empty(iter))
2186                         return POLLIN | POLLRDNORM;
2187                 poll_wait(filp, &trace_wait, poll_table);
2188                 if (!trace_empty(iter))
2189                         return POLLIN | POLLRDNORM;
2190
2191                 return 0;
2192         }
2193 }
2194
2195 /*
2196  * Consumer reader.
2197  */
2198 static ssize_t
2199 tracing_read_pipe(struct file *filp, char __user *ubuf,
2200                   size_t cnt, loff_t *ppos)
2201 {
2202         struct trace_iterator *iter = filp->private_data;
2203         struct trace_array_cpu *data;
2204         static cpumask_t mask;
2205         static int start;
2206         unsigned long flags;
2207 #ifdef CONFIG_FTRACE
2208         int ftrace_save;
2209 #endif
2210         int read = 0;
2211         int cpu;
2212         int len;
2213         int ret;
2214
2215         /* return any leftover data */
2216         if (iter->seq.len > start) {
2217                 len = iter->seq.len - start;
2218                 if (cnt > len)
2219                         cnt = len;
2220                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2221                 if (ret)
2222                         cnt = -EFAULT;
2223
2224                 start += len;
2225
2226                 return cnt;
2227         }
2228
2229         trace_seq_reset(&iter->seq);
2230         start = 0;
2231
2232         while (trace_empty(iter)) {
2233                 if (!(trace_flags & TRACE_ITER_BLOCK))
2234                         return -EWOULDBLOCK;
2235                 /*
2236                  * This is a make-shift waitqueue. The reason we don't use
2237                  * an actual wait queue is because:
2238                  *  1) we only ever have one waiter
2239                  *  2) the tracing, traces all functions, we don't want
2240                  *     the overhead of calling wake_up and friends
2241                  *     (and tracing them too)
2242                  *     Anyway, this is really very primitive wakeup.
2243                  */
2244                 set_current_state(TASK_INTERRUPTIBLE);
2245                 iter->tr->waiter = current;
2246
2247                 /* sleep for one second, and try again. */
2248                 schedule_timeout(HZ);
2249
2250                 iter->tr->waiter = NULL;
2251
2252                 if (signal_pending(current))
2253                         return -EINTR;
2254
2255                 /*
2256                  * We block until we read something and tracing is disabled.
2257                  * We still block if tracing is disabled, but we have never
2258                  * read anything. This allows a user to cat this file, and
2259                  * then enable tracing. But after we have read something,
2260                  * we give an EOF when tracing is again disabled.
2261                  *
2262                  * iter->pos will be 0 if we haven't read anything.
2263                  */
2264                 if (!tracer_enabled && iter->pos)
2265                         break;
2266
2267                 continue;
2268         }
2269
2270         /* stop when tracing is finished */
2271         if (trace_empty(iter))
2272                 return 0;
2273
2274         if (cnt >= PAGE_SIZE)
2275                 cnt = PAGE_SIZE - 1;
2276
2277         memset(iter, 0, sizeof(*iter));
2278         iter->tr = &global_trace;
2279         iter->pos = -1;
2280
2281         /*
2282          * We need to stop all tracing on all CPUS to read the
2283          * the next buffer. This is a bit expensive, but is
2284          * not done often. We fill all what we can read,
2285          * and then release the locks again.
2286          */
2287
2288         cpus_clear(mask);
2289         local_irq_save(flags);
2290 #ifdef CONFIG_FTRACE
2291         ftrace_save = ftrace_enabled;
2292         ftrace_enabled = 0;
2293 #endif
2294         smp_wmb();
2295         for_each_possible_cpu(cpu) {
2296                 data = iter->tr->data[cpu];
2297
2298                 if (!head_page(data) || !data->trace_idx)
2299                         continue;
2300
2301                 atomic_inc(&data->disabled);
2302                 cpu_set(cpu, mask);
2303         }
2304
2305         for_each_cpu_mask(cpu, mask) {
2306                 data = iter->tr->data[cpu];
2307                 __raw_spin_lock(&data->lock);
2308         }
2309
2310         while (find_next_entry_inc(iter) != NULL) {
2311                 int len = iter->seq.len;
2312
2313                 ret = print_trace_line(iter);
2314                 if (!ret) {
2315                         /* don't print partial lines */
2316                         iter->seq.len = len;
2317                         break;
2318                 }
2319
2320                 trace_consume(iter);
2321
2322                 if (iter->seq.len >= cnt)
2323                         break;
2324         }
2325
2326         for_each_cpu_mask(cpu, mask) {
2327                 data = iter->tr->data[cpu];
2328                 __raw_spin_unlock(&data->lock);
2329         }
2330
2331         for_each_cpu_mask(cpu, mask) {
2332                 data = iter->tr->data[cpu];
2333                 atomic_dec(&data->disabled);
2334         }
2335 #ifdef CONFIG_FTRACE
2336         ftrace_enabled = ftrace_save;
2337 #endif
2338         local_irq_restore(flags);
2339
2340         /* Now copy what we have to the user */
2341         read = iter->seq.len;
2342         if (read > cnt)
2343                 read = cnt;
2344
2345         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2346
2347         if (read < iter->seq.len)
2348                 start = read;
2349         else
2350                 trace_seq_reset(&iter->seq);
2351
2352         if (ret)
2353                 read = -EFAULT;
2354
2355         return read;
2356 }
2357
2358 static struct file_operations tracing_max_lat_fops = {
2359         .open           = tracing_open_generic,
2360         .read           = tracing_max_lat_read,
2361         .write          = tracing_max_lat_write,
2362 };
2363
2364 static struct file_operations tracing_ctrl_fops = {
2365         .open           = tracing_open_generic,
2366         .read           = tracing_ctrl_read,
2367         .write          = tracing_ctrl_write,
2368 };
2369
2370 static struct file_operations set_tracer_fops = {
2371         .open           = tracing_open_generic,
2372         .read           = tracing_set_trace_read,
2373         .write          = tracing_set_trace_write,
2374 };
2375
2376 static struct file_operations tracing_pipe_fops = {
2377         .open           = tracing_open_pipe,
2378         .poll           = tracing_poll_pipe,
2379         .read           = tracing_read_pipe,
2380         .release        = tracing_release_pipe,
2381 };
2382
2383 #ifdef CONFIG_DYNAMIC_FTRACE
2384
2385 static ssize_t
2386 tracing_read_long(struct file *filp, char __user *ubuf,
2387                   size_t cnt, loff_t *ppos)
2388 {
2389         unsigned long *p = filp->private_data;
2390         char buf[64];
2391         int r;
2392
2393         r = sprintf(buf, "%ld\n", *p);
2394
2395         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2396 }
2397
2398 static struct file_operations tracing_read_long_fops = {
2399         .open           = tracing_open_generic,
2400         .read           = tracing_read_long,
2401 };
2402 #endif
2403
2404 static struct dentry *d_tracer;
2405
2406 struct dentry *tracing_init_dentry(void)
2407 {
2408         static int once;
2409
2410         if (d_tracer)
2411                 return d_tracer;
2412
2413         d_tracer = debugfs_create_dir("tracing", NULL);
2414
2415         if (!d_tracer && !once) {
2416                 once = 1;
2417                 pr_warning("Could not create debugfs directory 'tracing'\n");
2418                 return NULL;
2419         }
2420
2421         return d_tracer;
2422 }
2423
2424 #ifdef CONFIG_FTRACE_SELFTEST
2425 /* Let selftest have access to static functions in this file */
2426 #include "trace_selftest.c"
2427 #endif
2428
2429 static __init void tracer_init_debugfs(void)
2430 {
2431         struct dentry *d_tracer;
2432         struct dentry *entry;
2433
2434         d_tracer = tracing_init_dentry();
2435
2436         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2437                                     &global_trace, &tracing_ctrl_fops);
2438         if (!entry)
2439                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2440
2441         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2442                                     NULL, &tracing_iter_fops);
2443         if (!entry)
2444                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2445
2446         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2447                                     NULL, &tracing_cpumask_fops);
2448         if (!entry)
2449                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2450
2451         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2452                                     &global_trace, &tracing_lt_fops);
2453         if (!entry)
2454                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2455
2456         entry = debugfs_create_file("trace", 0444, d_tracer,
2457                                     &global_trace, &tracing_fops);
2458         if (!entry)
2459                 pr_warning("Could not create debugfs 'trace' entry\n");
2460
2461         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2462                                     &global_trace, &show_traces_fops);
2463         if (!entry)
2464                 pr_warning("Could not create debugfs 'trace' entry\n");
2465
2466         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2467                                     &global_trace, &set_tracer_fops);
2468         if (!entry)
2469                 pr_warning("Could not create debugfs 'trace' entry\n");
2470
2471         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2472                                     &tracing_max_latency,
2473                                     &tracing_max_lat_fops);
2474         if (!entry)
2475                 pr_warning("Could not create debugfs "
2476                            "'tracing_max_latency' entry\n");
2477
2478         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2479                                     &tracing_thresh, &tracing_max_lat_fops);
2480         if (!entry)
2481                 pr_warning("Could not create debugfs "
2482                            "'tracing_threash' entry\n");
2483         entry = debugfs_create_file("README", 0644, d_tracer,
2484                                     NULL, &tracing_readme_fops);
2485         if (!entry)
2486                 pr_warning("Could not create debugfs 'README' entry\n");
2487
2488         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2489                                     NULL, &tracing_pipe_fops);
2490         if (!entry)
2491                 pr_warning("Could not create debugfs "
2492                            "'tracing_threash' entry\n");
2493
2494 #ifdef CONFIG_DYNAMIC_FTRACE
2495         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2496                                     &ftrace_update_tot_cnt,
2497                                     &tracing_read_long_fops);
2498         if (!entry)
2499                 pr_warning("Could not create debugfs "
2500                            "'dyn_ftrace_total_info' entry\n");
2501 #endif
2502 }
2503
2504 /* dummy trace to disable tracing */
2505 static struct tracer no_tracer __read_mostly =
2506 {
2507         .name           = "none",
2508 };
2509
2510 static int trace_alloc_page(void)
2511 {
2512         struct trace_array_cpu *data;
2513         struct page *page, *tmp;
2514         LIST_HEAD(pages);
2515         void *array;
2516         int i;
2517
2518         /* first allocate a page for each CPU */
2519         for_each_possible_cpu(i) {
2520                 array = (void *)__get_free_page(GFP_KERNEL);
2521                 if (array == NULL) {
2522                         printk(KERN_ERR "tracer: failed to allocate page"
2523                                "for trace buffer!\n");
2524                         goto free_pages;
2525                 }
2526
2527                 page = virt_to_page(array);
2528                 list_add(&page->lru, &pages);
2529
2530 /* Only allocate if we are actually using the max trace */
2531 #ifdef CONFIG_TRACER_MAX_TRACE
2532                 array = (void *)__get_free_page(GFP_KERNEL);
2533                 if (array == NULL) {
2534                         printk(KERN_ERR "tracer: failed to allocate page"
2535                                "for trace buffer!\n");
2536                         goto free_pages;
2537                 }
2538                 page = virt_to_page(array);
2539                 list_add(&page->lru, &pages);
2540 #endif
2541         }
2542
2543         /* Now that we successfully allocate a page per CPU, add them */
2544         for_each_possible_cpu(i) {
2545                 data = global_trace.data[i];
2546                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2547                 page = list_entry(pages.next, struct page, lru);
2548                 list_del_init(&page->lru);
2549                 list_add_tail(&page->lru, &data->trace_pages);
2550                 ClearPageLRU(page);
2551
2552 #ifdef CONFIG_TRACER_MAX_TRACE
2553                 data = max_tr.data[i];
2554                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2555                 page = list_entry(pages.next, struct page, lru);
2556                 list_del_init(&page->lru);
2557                 list_add_tail(&page->lru, &data->trace_pages);
2558                 SetPageLRU(page);
2559 #endif
2560         }
2561         global_trace.entries += ENTRIES_PER_PAGE;
2562
2563         return 0;
2564
2565  free_pages:
2566         list_for_each_entry_safe(page, tmp, &pages, lru) {
2567                 list_del_init(&page->lru);
2568                 __free_page(page);
2569         }
2570         return -ENOMEM;
2571 }
2572
2573 __init static int tracer_alloc_buffers(void)
2574 {
2575         struct trace_array_cpu *data;
2576         void *array;
2577         struct page *page;
2578         int pages = 0;
2579         int ret = -ENOMEM;
2580         int i;
2581
2582         global_trace.ctrl = tracer_enabled;
2583
2584         /* Allocate the first page for all buffers */
2585         for_each_possible_cpu(i) {
2586                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2587                 max_tr.data[i] = &per_cpu(max_data, i);
2588
2589                 array = (void *)__get_free_page(GFP_KERNEL);
2590                 if (array == NULL) {
2591                         printk(KERN_ERR "tracer: failed to allocate page"
2592                                "for trace buffer!\n");
2593                         goto free_buffers;
2594                 }
2595
2596                 /* set the array to the list */
2597                 INIT_LIST_HEAD(&data->trace_pages);
2598                 page = virt_to_page(array);
2599                 list_add(&page->lru, &data->trace_pages);
2600                 /* use the LRU flag to differentiate the two buffers */
2601                 ClearPageLRU(page);
2602
2603 /* Only allocate if we are actually using the max trace */
2604 #ifdef CONFIG_TRACER_MAX_TRACE
2605                 array = (void *)__get_free_page(GFP_KERNEL);
2606                 if (array == NULL) {
2607                         printk(KERN_ERR "tracer: failed to allocate page"
2608                                "for trace buffer!\n");
2609                         goto free_buffers;
2610                 }
2611
2612                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2613                 page = virt_to_page(array);
2614                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2615                 SetPageLRU(page);
2616 #endif
2617         }
2618
2619         /*
2620          * Since we allocate by orders of pages, we may be able to
2621          * round up a bit.
2622          */
2623         global_trace.entries = ENTRIES_PER_PAGE;
2624         pages++;
2625
2626         while (global_trace.entries < trace_nr_entries) {
2627                 if (trace_alloc_page())
2628                         break;
2629                 pages++;
2630         }
2631         max_tr.entries = global_trace.entries;
2632
2633         pr_info("tracer: %d pages allocated for %ld",
2634                 pages, trace_nr_entries);
2635         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2636         pr_info("   actual entries %ld\n", global_trace.entries);
2637
2638         tracer_init_debugfs();
2639
2640         trace_init_cmdlines();
2641
2642         register_tracer(&no_tracer);
2643         current_trace = &no_tracer;
2644
2645         /* All seems OK, enable tracing */
2646         tracing_disabled = 0;
2647
2648         return 0;
2649
2650  free_buffers:
2651         for (i-- ; i >= 0; i--) {
2652                 struct page *page, *tmp;
2653                 struct trace_array_cpu *data = global_trace.data[i];
2654
2655                 if (data) {
2656                         list_for_each_entry_safe(page, tmp,
2657                                                  &data->trace_pages, lru) {
2658                                 list_del_init(&page->lru);
2659                                 __free_page(page);
2660                         }
2661                 }
2662
2663 #ifdef CONFIG_TRACER_MAX_TRACE
2664                 data = max_tr.data[i];
2665                 if (data) {
2666                         list_for_each_entry_safe(page, tmp,
2667                                                  &data->trace_pages, lru) {
2668                                 list_del_init(&page->lru);
2669                                 __free_page(page);
2670                         }
2671                 }
2672 #endif
2673         }
2674         return ret;
2675 }
2676 fs_initcall(tracer_alloc_buffers);