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

   1 /*
   2  * ring buffer based function tracer
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally taken from the RT patch by:
   8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code from the latency_tracer, that is:
  11  *  Copyright (C) 2004-2006 Ingo Molnar
  12  *  Copyright (C) 2004 William Lee Irwin III
  13  */
  14 #include <linux/utsrelease.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/debugfs.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/linkage.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/ftrace.h>
  23 #include <linux/module.h>
  24 #include <linux/percpu.h>
  25 #include <linux/ctype.h>
  26 #include <linux/init.h>
  27 #include <linux/poll.h>
  28 #include <linux/gfp.h>
  29 #include <linux/fs.h>
  30
  31 #include <linux/stacktrace.h>
  32
  33 #include "trace.h"
  34
  35 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  36 unsigned long __read_mostly     tracing_thresh;
  37
  38 static int tracing_disabled = 1;
  39
  40 static long
  41 ns2usecs(cycle_t nsec)
  42 {
  43         nsec += 500;
  44         do_div(nsec, 1000);
  45         return nsec;
  46 }
  47
  48 cycle_t ftrace_now(int cpu)
  49 {
  50         return cpu_clock(cpu);
  51 }
  52
  53 static struct trace_array       global_trace;
  54
  55 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  56
  57 static struct trace_array       max_tr;
  58
  59 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  60
  61 static int                      tracer_enabled = 1;
  62 static unsigned long            trace_nr_entries = 65536UL;
  63
  64 static struct tracer            *trace_types __read_mostly;
  65 static struct tracer            *current_trace __read_mostly;
  66 static int                      max_tracer_type_len;
  67
  68 static DEFINE_MUTEX(trace_types_lock);
  69 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
  70
  71 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
  72
  73 void trace_wake_up(void)
  74 {
  75         /*
  76          * The runqueue_is_locked() can fail, but this is the best we
  77          * have for now:
  78          */
  79         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
  80                 wake_up(&trace_wait);
  81 }
  82
  83 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  84
  85 static int __init set_nr_entries(char *str)
  86 {
  87         if (!str)
  88                 return 0;
  89         trace_nr_entries = simple_strtoul(str, &str, 0);
  90         return 1;
  91 }
  92 __setup("trace_entries=", set_nr_entries);
  93
  94 unsigned long nsecs_to_usecs(unsigned long nsecs)
  95 {
  96         return nsecs / 1000;
  97 }
  98
  99 enum trace_type {
 100         __TRACE_FIRST_TYPE = 0,
 101
 102         TRACE_FN,
 103         TRACE_CTX,
 104         TRACE_WAKE,
 105         TRACE_STACK,
 106         TRACE_SPECIAL,
 107
 108         __TRACE_LAST_TYPE
 109 };
 110
 111 enum trace_flag_type {
 112         TRACE_FLAG_IRQS_OFF             = 0x01,
 113         TRACE_FLAG_NEED_RESCHED         = 0x02,
 114         TRACE_FLAG_HARDIRQ              = 0x04,
 115         TRACE_FLAG_SOFTIRQ              = 0x08,
 116 };
 117
 118 #define TRACE_ITER_SYM_MASK \
 119         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 120
 121 /* These must match the bit postions above */
 122 static const char *trace_options[] = {
 123         "print-parent",
 124         "sym-offset",
 125         "sym-addr",
 126         "verbose",
 127         "raw",
 128         "hex",
 129         "bin",
 130         "block",
 131         "stacktrace",
 132         "sched-tree",
 133         NULL
 134 };
 135
 136 static DEFINE_SPINLOCK(ftrace_max_lock);
 137
 138 /*
 139  * Copy the new maximum trace into the separate maximum-trace
 140  * structure. (this way the maximum trace is permanently saved,
 141  * for later retrieval via /debugfs/tracing/latency_trace)
 142  */
 143 static void
 144 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 145 {
 146         struct trace_array_cpu *data = tr->data[cpu];
 147
 148         max_tr.cpu = cpu;
 149         max_tr.time_start = data->preempt_timestamp;
 150
 151         data = max_tr.data[cpu];
 152         data->saved_latency = tracing_max_latency;
 153
 154         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 155         data->pid = tsk->pid;
 156         data->uid = tsk->uid;
 157         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 158         data->policy = tsk->policy;
 159         data->rt_priority = tsk->rt_priority;
 160
 161         /* record this tasks comm */
 162         tracing_record_cmdline(current);
 163 }
 164
 165 void check_pages(struct trace_array_cpu *data)
 166 {
 167         struct page *page, *tmp;
 168
 169         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 170         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 171
 172         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 173                 BUG_ON(page->lru.next->prev != &page->lru);
 174                 BUG_ON(page->lru.prev->next != &page->lru);
 175         }
 176 }
 177
 178 void *head_page(struct trace_array_cpu *data)
 179 {
 180         struct page *page;
 181
 182         check_pages(data);
 183         if (list_empty(&data->trace_pages))
 184                 return NULL;
 185
 186         page = list_entry(data->trace_pages.next, struct page, lru);
 187         BUG_ON(&page->lru == &data->trace_pages);
 188
 189         return page_address(page);
 190 }
 191
 192 static int
 193 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 194 {
 195         int len = (PAGE_SIZE - 1) - s->len;
 196         va_list ap;
 197         int ret;
 198
 199         if (!len)
 200                 return 0;
 201
 202         va_start(ap, fmt);
 203         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 204         va_end(ap);
 205
 206         /* If we can't write it all, don't bother writing anything */
 207         if (ret > len)
 208                 return 0;
 209
 210         s->len += ret;
 211
 212         return len;
 213 }
 214
 215 static int
 216 trace_seq_puts(struct trace_seq *s, const char *str)
 217 {
 218         int len = strlen(str);
 219
 220         if (len > ((PAGE_SIZE - 1) - s->len))
 221                 return 0;
 222
 223         memcpy(s->buffer + s->len, str, len);
 224         s->len += len;
 225
 226         return len;
 227 }
 228
 229 static int
 230 trace_seq_putc(struct trace_seq *s, unsigned char c)
 231 {
 232         if (s->len >= (PAGE_SIZE - 1))
 233                 return 0;
 234
 235         s->buffer[s->len++] = c;
 236
 237         return 1;
 238 }
 239
 240 static int
 241 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 242 {
 243         if (len > ((PAGE_SIZE - 1) - s->len))
 244                 return 0;
 245
 246         memcpy(s->buffer + s->len, mem, len);
 247         s->len += len;
 248
 249         return len;
 250 }
 251
 252 #define HEX_CHARS 17
 253
 254 static int
 255 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 256 {
 257         unsigned char hex[HEX_CHARS];
 258         unsigned char *data;
 259         unsigned char byte;
 260         int i, j;
 261
 262         BUG_ON(len >= HEX_CHARS);
 263
 264         data = mem;
 265
 266 #ifdef __BIG_ENDIAN
 267         for (i = 0, j = 0; i < len; i++) {
 268 #else
 269         for (i = len-1, j = 0; i >= 0; i--) {
 270 #endif
 271                 byte = data[i];
 272
 273                 hex[j]   = byte & 0x0f;
 274                 if (hex[j] >= 10)
 275                         hex[j] += 'a' - 10;
 276                 else
 277                         hex[j] += '0';
 278                 j++;
 279
 280                 hex[j] = byte >> 4;
 281                 if (hex[j] >= 10)
 282                         hex[j] += 'a' - 10;
 283                 else
 284                         hex[j] += '0';
 285                 j++;
 286         }
 287         hex[j] = ' ';
 288         j++;
 289
 290         return trace_seq_putmem(s, hex, j);
 291 }
 292
 293 static void
 294 trace_seq_reset(struct trace_seq *s)
 295 {
 296         s->len = 0;
 297 }
 298
 299 static void
 300 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 301 {
 302         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 303
 304         s->buffer[len] = 0;
 305         seq_puts(m, s->buffer);
 306
 307         trace_seq_reset(s);
 308 }
 309
 310 static void
 311 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 312 {
 313         struct list_head flip_pages;
 314
 315         INIT_LIST_HEAD(&flip_pages);
 316
 317         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 318                 sizeof(struct trace_array_cpu) -
 319                 offsetof(struct trace_array_cpu, trace_head_idx));
 320
 321         check_pages(tr1);
 322         check_pages(tr2);
 323         list_splice_init(&tr1->trace_pages, &flip_pages);
 324         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 325         list_splice_init(&flip_pages, &tr2->trace_pages);
 326         BUG_ON(!list_empty(&flip_pages));
 327         check_pages(tr1);
 328         check_pages(tr2);
 329 }
 330
 331 void
 332 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 333 {
 334         struct trace_array_cpu *data;
 335         int i;
 336
 337         WARN_ON_ONCE(!irqs_disabled());
 338         spin_lock(&ftrace_max_lock);
 339         /* clear out all the previous traces */
 340         for_each_possible_cpu(i) {
 341                 data = tr->data[i];
 342                 flip_trace(max_tr.data[i], data);
 343                 tracing_reset(data);
 344         }
 345
 346         __update_max_tr(tr, tsk, cpu);
 347         spin_unlock(&ftrace_max_lock);
 348 }
 349
 350 /**
 351  * update_max_tr_single - only copy one trace over, and reset the rest
 352  * @tr - tracer
 353  * @tsk - task with the latency
 354  * @cpu - the cpu of the buffer to copy.
 355  */
 356 void
 357 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 358 {
 359         struct trace_array_cpu *data = tr->data[cpu];
 360         int i;
 361
 362         WARN_ON_ONCE(!irqs_disabled());
 363         spin_lock(&ftrace_max_lock);
 364         for_each_possible_cpu(i)
 365                 tracing_reset(max_tr.data[i]);
 366
 367         flip_trace(max_tr.data[cpu], data);
 368         tracing_reset(data);
 369
 370         __update_max_tr(tr, tsk, cpu);
 371         spin_unlock(&ftrace_max_lock);
 372 }
 373
 374 int register_tracer(struct tracer *type)
 375 {
 376         struct tracer *t;
 377         int len;
 378         int ret = 0;
 379
 380         if (!type->name) {
 381                 pr_info("Tracer must have a name\n");
 382                 return -1;
 383         }
 384
 385         mutex_lock(&trace_types_lock);
 386         for (t = trace_types; t; t = t->next) {
 387                 if (strcmp(type->name, t->name) == 0) {
 388                         /* already found */
 389                         pr_info("Trace %s already registered\n",
 390                                 type->name);
 391                         ret = -1;
 392                         goto out;
 393                 }
 394         }
 395
 396 #ifdef CONFIG_FTRACE_STARTUP_TEST
 397         if (type->selftest) {
 398                 struct tracer *saved_tracer = current_trace;
 399                 struct trace_array_cpu *data;
 400                 struct trace_array *tr = &global_trace;
 401                 int saved_ctrl = tr->ctrl;
 402                 int i;
 403                 /*
 404                  * Run a selftest on this tracer.
 405                  * Here we reset the trace buffer, and set the current
 406                  * tracer to be this tracer. The tracer can then run some
 407                  * internal tracing to verify that everything is in order.
 408                  * If we fail, we do not register this tracer.
 409                  */
 410                 for_each_possible_cpu(i) {
 411                         data = tr->data[i];
 412                         if (!head_page(data))
 413                                 continue;
 414                         tracing_reset(data);
 415                 }
 416                 current_trace = type;
 417                 tr->ctrl = 0;
 418                 /* the test is responsible for initializing and enabling */
 419                 pr_info("Testing tracer %s: ", type->name);
 420                 ret = type->selftest(type, tr);
 421                 /* the test is responsible for resetting too */
 422                 current_trace = saved_tracer;
 423                 tr->ctrl = saved_ctrl;
 424                 if (ret) {
 425                         printk(KERN_CONT "FAILED!\n");
 426                         goto out;
 427                 }
 428                 /* Only reset on passing, to avoid touching corrupted buffers */
 429                 for_each_possible_cpu(i) {
 430                         data = tr->data[i];
 431                         if (!head_page(data))
 432                                 continue;
 433                         tracing_reset(data);
 434                 }
 435                 printk(KERN_CONT "PASSED\n");
 436         }
 437 #endif
 438
 439         type->next = trace_types;
 440         trace_types = type;
 441         len = strlen(type->name);
 442         if (len > max_tracer_type_len)
 443                 max_tracer_type_len = len;
 444
 445  out:
 446         mutex_unlock(&trace_types_lock);
 447
 448         return ret;
 449 }
 450
 451 void unregister_tracer(struct tracer *type)
 452 {
 453         struct tracer **t;
 454         int len;
 455
 456         mutex_lock(&trace_types_lock);
 457         for (t = &trace_types; *t; t = &(*t)->next) {
 458                 if (*t == type)
 459                         goto found;
 460         }
 461         pr_info("Trace %s not registered\n", type->name);
 462         goto out;
 463
 464  found:
 465         *t = (*t)->next;
 466         if (strlen(type->name) != max_tracer_type_len)
 467                 goto out;
 468
 469         max_tracer_type_len = 0;
 470         for (t = &trace_types; *t; t = &(*t)->next) {
 471                 len = strlen((*t)->name);
 472                 if (len > max_tracer_type_len)
 473                         max_tracer_type_len = len;
 474         }
 475  out:
 476         mutex_unlock(&trace_types_lock);
 477 }
 478
 479 void tracing_reset(struct trace_array_cpu *data)
 480 {
 481         data->trace_idx = 0;
 482         data->trace_head = data->trace_tail = head_page(data);
 483         data->trace_head_idx = 0;
 484         data->trace_tail_idx = 0;
 485 }
 486
 487 #define SAVED_CMDLINES 128
 488 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 489 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 490 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 491 static int cmdline_idx;
 492 static DEFINE_SPINLOCK(trace_cmdline_lock);
 493 atomic_t trace_record_cmdline_disabled;
 494
 495 static void trace_init_cmdlines(void)
 496 {
 497         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 498         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 499         cmdline_idx = 0;
 500 }
 501
 502 void trace_stop_cmdline_recording(void);
 503
 504 static void trace_save_cmdline(struct task_struct *tsk)
 505 {
 506         unsigned map;
 507         unsigned idx;
 508
 509         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 510                 return;
 511
 512         /*
 513          * It's not the end of the world if we don't get
 514          * the lock, but we also don't want to spin
 515          * nor do we want to disable interrupts,
 516          * so if we miss here, then better luck next time.
 517          */
 518         if (!spin_trylock(&trace_cmdline_lock))
 519                 return;
 520
 521         idx = map_pid_to_cmdline[tsk->pid];
 522         if (idx >= SAVED_CMDLINES) {
 523                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 524
 525                 map = map_cmdline_to_pid[idx];
 526                 if (map <= PID_MAX_DEFAULT)
 527                         map_pid_to_cmdline[map] = (unsigned)-1;
 528
 529                 map_pid_to_cmdline[tsk->pid] = idx;
 530
 531                 cmdline_idx = idx;
 532         }
 533
 534         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 535
 536         spin_unlock(&trace_cmdline_lock);
 537 }
 538
 539 static char *trace_find_cmdline(int pid)
 540 {
 541         char *cmdline = "<...>";
 542         unsigned map;
 543
 544         if (!pid)
 545                 return "<idle>";
 546
 547         if (pid > PID_MAX_DEFAULT)
 548                 goto out;
 549
 550         map = map_pid_to_cmdline[pid];
 551         if (map >= SAVED_CMDLINES)
 552                 goto out;
 553
 554         cmdline = saved_cmdlines[map];
 555
 556  out:
 557         return cmdline;
 558 }
 559
 560 void tracing_record_cmdline(struct task_struct *tsk)
 561 {
 562         if (atomic_read(&trace_record_cmdline_disabled))
 563                 return;
 564
 565         trace_save_cmdline(tsk);
 566 }
 567
 568 static inline struct list_head *
 569 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 570 {
 571         /*
 572          * Roundrobin - but skip the head (which is not a real page):
 573          */
 574         next = next->next;
 575         if (unlikely(next == &data->trace_pages))
 576                 next = next->next;
 577         BUG_ON(next == &data->trace_pages);
 578
 579         return next;
 580 }
 581
 582 static inline void *
 583 trace_next_page(struct trace_array_cpu *data, void *addr)
 584 {
 585         struct list_head *next;
 586         struct page *page;
 587
 588         page = virt_to_page(addr);
 589
 590         next = trace_next_list(data, &page->lru);
 591         page = list_entry(next, struct page, lru);
 592
 593         return page_address(page);
 594 }
 595
 596 static inline struct trace_entry *
 597 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 598 {
 599         unsigned long idx, idx_next;
 600         struct trace_entry *entry;
 601
 602         data->trace_idx++;
 603         idx = data->trace_head_idx;
 604         idx_next = idx + 1;
 605
 606         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 607
 608         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 609
 610         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 611                 data->trace_head = trace_next_page(data, data->trace_head);
 612                 idx_next = 0;
 613         }
 614
 615         if (data->trace_head == data->trace_tail &&
 616             idx_next == data->trace_tail_idx) {
 617                 /* overrun */
 618                 data->trace_tail_idx++;
 619                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 620                         data->trace_tail =
 621                                 trace_next_page(data, data->trace_tail);
 622                         data->trace_tail_idx = 0;
 623                 }
 624         }
 625
 626         data->trace_head_idx = idx_next;
 627
 628         return entry;
 629 }
 630
 631 static inline void
 632 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 633 {
 634         struct task_struct *tsk = current;
 635         unsigned long pc;
 636
 637         pc = preempt_count();
 638
 639         entry->preempt_count    = pc & 0xff;
 640         entry->pid              = tsk->pid;
 641         entry->t                = ftrace_now(raw_smp_processor_id());
 642         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 643                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 644                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 645                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 646 }
 647
 648 void
 649 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 650                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 651 {
 652         struct trace_entry *entry;
 653         unsigned long irq_flags;
 654
 655         spin_lock_irqsave(&data->lock, irq_flags);
 656         entry                   = tracing_get_trace_entry(tr, data);
 657         tracing_generic_entry_update(entry, flags);
 658         entry->type             = TRACE_FN;
 659         entry->fn.ip            = ip;
 660         entry->fn.parent_ip     = parent_ip;
 661         spin_unlock_irqrestore(&data->lock, irq_flags);
 662
 663         trace_wake_up();
 664 }
 665
 666 void
 667 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 668        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 669 {
 670         if (likely(!atomic_read(&data->disabled)))
 671                 trace_function(tr, data, ip, parent_ip, flags);
 672 }
 673
 674 void
 675 __trace_special(void *__tr, void *__data,
 676                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 677 {
 678         struct trace_array_cpu *data = __data;
 679         struct trace_array *tr = __tr;
 680         struct trace_entry *entry;
 681         unsigned long irq_flags;
 682
 683         spin_lock_irqsave(&data->lock, irq_flags);
 684         entry                   = tracing_get_trace_entry(tr, data);
 685         tracing_generic_entry_update(entry, 0);
 686         entry->type             = TRACE_SPECIAL;
 687         entry->special.arg1     = arg1;
 688         entry->special.arg2     = arg2;
 689         entry->special.arg3     = arg3;
 690         spin_unlock_irqrestore(&data->lock, irq_flags);
 691
 692         trace_wake_up();
 693 }
 694
 695 void __trace_stack(struct trace_array *tr,
 696                    struct trace_array_cpu *data,
 697                    unsigned long flags,
 698                    int skip)
 699 {
 700         struct trace_entry *entry;
 701         struct stack_trace trace;
 702
 703         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 704                 return;
 705
 706         entry                   = tracing_get_trace_entry(tr, data);
 707         tracing_generic_entry_update(entry, flags);
 708         entry->type             = TRACE_STACK;
 709
 710         memset(&entry->stack, 0, sizeof(entry->stack));
 711
 712         trace.nr_entries        = 0;
 713         trace.max_entries       = FTRACE_STACK_ENTRIES;
 714         trace.skip              = skip;
 715         trace.entries           = entry->stack.caller;
 716
 717         save_stack_trace(&trace);
 718 }
 719
 720 void
 721 tracing_sched_switch_trace(struct trace_array *tr,
 722                            struct trace_array_cpu *data,
 723                            struct task_struct *prev,
 724                            struct task_struct *next,
 725                            unsigned long flags)
 726 {
 727         struct trace_entry *entry;
 728         unsigned long irq_flags;
 729
 730         spin_lock_irqsave(&data->lock, irq_flags);
 731         entry                   = tracing_get_trace_entry(tr, data);
 732         tracing_generic_entry_update(entry, flags);
 733         entry->type             = TRACE_CTX;
 734         entry->ctx.prev_pid     = prev->pid;
 735         entry->ctx.prev_prio    = prev->prio;
 736         entry->ctx.prev_state   = prev->state;
 737         entry->ctx.next_pid     = next->pid;
 738         entry->ctx.next_prio    = next->prio;
 739         __trace_stack(tr, data, flags, 4);
 740         spin_unlock_irqrestore(&data->lock, irq_flags);
 741 }
 742
 743 void
 744 tracing_sched_wakeup_trace(struct trace_array *tr,
 745                            struct trace_array_cpu *data,
 746                            struct task_struct *wakee,
 747                            struct task_struct *curr,
 748                            unsigned long flags)
 749 {
 750         struct trace_entry *entry;
 751         unsigned long irq_flags;
 752
 753         spin_lock_irqsave(&data->lock, irq_flags);
 754         entry                   = tracing_get_trace_entry(tr, data);
 755         tracing_generic_entry_update(entry, flags);
 756         entry->type             = TRACE_WAKE;
 757         entry->ctx.prev_pid     = curr->pid;
 758         entry->ctx.prev_prio    = curr->prio;
 759         entry->ctx.prev_state   = curr->state;
 760         entry->ctx.next_pid     = wakee->pid;
 761         entry->ctx.next_prio    = wakee->prio;
 762         __trace_stack(tr, data, flags, 5);
 763         spin_unlock_irqrestore(&data->lock, irq_flags);
 764
 765         trace_wake_up();
 766 }
 767
 768 #ifdef CONFIG_FTRACE
 769 static void
 770 function_trace_call(unsigned long ip, unsigned long parent_ip)
 771 {
 772         struct trace_array *tr = &global_trace;
 773         struct trace_array_cpu *data;
 774         unsigned long flags;
 775         long disabled;
 776         int cpu;
 777
 778         if (unlikely(!tracer_enabled))
 779                 return;
 780
 781         local_irq_save(flags);
 782         cpu = raw_smp_processor_id();
 783         data = tr->data[cpu];
 784         disabled = atomic_inc_return(&data->disabled);
 785
 786         if (likely(disabled == 1))
 787                 trace_function(tr, data, ip, parent_ip, flags);
 788
 789         atomic_dec(&data->disabled);
 790         local_irq_restore(flags);
 791 }
 792
 793 static struct ftrace_ops trace_ops __read_mostly =
 794 {
 795         .func = function_trace_call,
 796 };
 797
 798 void tracing_start_function_trace(void)
 799 {
 800         register_ftrace_function(&trace_ops);
 801 }
 802
 803 void tracing_stop_function_trace(void)
 804 {
 805         unregister_ftrace_function(&trace_ops);
 806 }
 807 #endif
 808
 809 enum trace_file_type {
 810         TRACE_FILE_LAT_FMT      = 1,
 811 };
 812
 813 static struct trace_entry *
 814 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 815                 struct trace_iterator *iter, int cpu)
 816 {
 817         struct page *page;
 818         struct trace_entry *array;
 819
 820         if (iter->next_idx[cpu] >= tr->entries ||
 821             iter->next_idx[cpu] >= data->trace_idx ||
 822             (data->trace_head == data->trace_tail &&
 823              data->trace_head_idx == data->trace_tail_idx))
 824                 return NULL;
 825
 826         if (!iter->next_page[cpu]) {
 827                 /* Initialize the iterator for this cpu trace buffer */
 828                 WARN_ON(!data->trace_tail);
 829                 page = virt_to_page(data->trace_tail);
 830                 iter->next_page[cpu] = &page->lru;
 831                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 832         }
 833
 834         page = list_entry(iter->next_page[cpu], struct page, lru);
 835         BUG_ON(&data->trace_pages == &page->lru);
 836
 837         array = page_address(page);
 838
 839         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 840         return &array[iter->next_page_idx[cpu]];
 841 }
 842
 843 static struct trace_entry *
 844 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 845 {
 846         struct trace_array *tr = iter->tr;
 847         struct trace_entry *ent, *next = NULL;
 848         int next_cpu = -1;
 849         int cpu;
 850
 851         for_each_possible_cpu(cpu) {
 852                 if (!head_page(tr->data[cpu]))
 853                         continue;
 854                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 855                 /*
 856                  * Pick the entry with the smallest timestamp:
 857                  */
 858                 if (ent && (!next || ent->t < next->t)) {
 859                         next = ent;
 860                         next_cpu = cpu;
 861                 }
 862         }
 863
 864         if (ent_cpu)
 865                 *ent_cpu = next_cpu;
 866
 867         return next;
 868 }
 869
 870 static void trace_iterator_increment(struct trace_iterator *iter)
 871 {
 872         iter->idx++;
 873         iter->next_idx[iter->cpu]++;
 874         iter->next_page_idx[iter->cpu]++;
 875
 876         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 877                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 878
 879                 iter->next_page_idx[iter->cpu] = 0;
 880                 iter->next_page[iter->cpu] =
 881                         trace_next_list(data, iter->next_page[iter->cpu]);
 882         }
 883 }
 884
 885 static void trace_consume(struct trace_iterator *iter)
 886 {
 887         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 888
 889         data->trace_tail_idx++;
 890         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 891                 data->trace_tail = trace_next_page(data, data->trace_tail);
 892                 data->trace_tail_idx = 0;
 893         }
 894
 895         /* Check if we empty it, then reset the index */
 896         if (data->trace_head == data->trace_tail &&
 897             data->trace_head_idx == data->trace_tail_idx)
 898                 data->trace_idx = 0;
 899 }
 900
 901 static void *find_next_entry_inc(struct trace_iterator *iter)
 902 {
 903         struct trace_entry *next;
 904         int next_cpu = -1;
 905
 906         next = find_next_entry(iter, &next_cpu);
 907
 908         iter->prev_ent = iter->ent;
 909         iter->prev_cpu = iter->cpu;
 910
 911         iter->ent = next;
 912         iter->cpu = next_cpu;
 913
 914         if (next)
 915                 trace_iterator_increment(iter);
 916
 917         return next ? iter : NULL;
 918 }
 919
 920 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 921 {
 922         struct trace_iterator *iter = m->private;
 923         void *last_ent = iter->ent;
 924         int i = (int)*pos;
 925         void *ent;
 926
 927         (*pos)++;
 928
 929         /* can't go backwards */
 930         if (iter->idx > i)
 931                 return NULL;
 932
 933         if (iter->idx < 0)
 934                 ent = find_next_entry_inc(iter);
 935         else
 936                 ent = iter;
 937
 938         while (ent && iter->idx < i)
 939                 ent = find_next_entry_inc(iter);
 940
 941         iter->pos = *pos;
 942
 943         if (last_ent && !ent)
 944                 seq_puts(m, "\n\nvim:ft=help\n");
 945
 946         return ent;
 947 }
 948
 949 static void *s_start(struct seq_file *m, loff_t *pos)
 950 {
 951         struct trace_iterator *iter = m->private;
 952         void *p = NULL;
 953         loff_t l = 0;
 954         int i;
 955
 956         mutex_lock(&trace_types_lock);
 957
 958         if (!current_trace || current_trace != iter->trace)
 959                 return NULL;
 960
 961         atomic_inc(&trace_record_cmdline_disabled);
 962
 963         /* let the tracer grab locks here if needed */
 964         if (current_trace->start)
 965                 current_trace->start(iter);
 966
 967         if (*pos != iter->pos) {
 968                 iter->ent = NULL;
 969                 iter->cpu = 0;
 970                 iter->idx = -1;
 971                 iter->prev_ent = NULL;
 972                 iter->prev_cpu = -1;
 973
 974                 for_each_possible_cpu(i) {
 975                         iter->next_idx[i] = 0;
 976                         iter->next_page[i] = NULL;
 977                 }
 978
 979                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 980                         ;
 981
 982         } else {
 983                 l = *pos - 1;
 984                 p = s_next(m, p, &l);
 985         }
 986
 987         return p;
 988 }
 989
 990 static void s_stop(struct seq_file *m, void *p)
 991 {
 992         struct trace_iterator *iter = m->private;
 993
 994         atomic_dec(&trace_record_cmdline_disabled);
 995
 996         /* let the tracer release locks here if needed */
 997         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 998                 iter->trace->stop(iter);
 999
1000         mutex_unlock(&trace_types_lock);
1001 }
1002
1003 static int
1004 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1005 {
1006 #ifdef CONFIG_KALLSYMS
1007         char str[KSYM_SYMBOL_LEN];
1008
1009         kallsyms_lookup(address, NULL, NULL, NULL, str);
1010
1011         return trace_seq_printf(s, fmt, str);
1012 #endif
1013         return 1;
1014 }
1015
1016 static int
1017 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1018                      unsigned long address)
1019 {
1020 #ifdef CONFIG_KALLSYMS
1021         char str[KSYM_SYMBOL_LEN];
1022
1023         sprint_symbol(str, address);
1024         return trace_seq_printf(s, fmt, str);
1025 #endif
1026         return 1;
1027 }
1028
1029 #ifndef CONFIG_64BIT
1030 # define IP_FMT "%08lx"
1031 #else
1032 # define IP_FMT "%016lx"
1033 #endif
1034
1035 static int
1036 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1037 {
1038         int ret;
1039
1040         if (!ip)
1041                 return trace_seq_printf(s, "0");
1042
1043         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1044                 ret = seq_print_sym_offset(s, "%s", ip);
1045         else
1046                 ret = seq_print_sym_short(s, "%s", ip);
1047
1048         if (!ret)
1049                 return 0;
1050
1051         if (sym_flags & TRACE_ITER_SYM_ADDR)
1052                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1053         return ret;
1054 }
1055
1056 static void print_lat_help_header(struct seq_file *m)
1057 {
1058         seq_puts(m, "#                _------=> CPU#            \n");
1059         seq_puts(m, "#               / _-----=> irqs-off        \n");
1060         seq_puts(m, "#              | / _----=> need-resched    \n");
1061         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1062         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1063         seq_puts(m, "#              |||| /                      \n");
1064         seq_puts(m, "#              |||||     delay             \n");
1065         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1066         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1067 }
1068
1069 static void print_func_help_header(struct seq_file *m)
1070 {
1071         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1072         seq_puts(m, "#              | |      |          |         |\n");
1073 }
1074
1075
1076 static void
1077 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1078 {
1079         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1080         struct trace_array *tr = iter->tr;
1081         struct trace_array_cpu *data = tr->data[tr->cpu];
1082         struct tracer *type = current_trace;
1083         unsigned long total   = 0;
1084         unsigned long entries = 0;
1085         int cpu;
1086         const char *name = "preemption";
1087
1088         if (type)
1089                 name = type->name;
1090
1091         for_each_possible_cpu(cpu) {
1092                 if (head_page(tr->data[cpu])) {
1093                         total += tr->data[cpu]->trace_idx;
1094                         if (tr->data[cpu]->trace_idx > tr->entries)
1095                                 entries += tr->entries;
1096                         else
1097                                 entries += tr->data[cpu]->trace_idx;
1098                 }
1099         }
1100
1101         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1102                    name, UTS_RELEASE);
1103         seq_puts(m, "-----------------------------------"
1104                  "---------------------------------\n");
1105         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1106                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1107                    nsecs_to_usecs(data->saved_latency),
1108                    entries,
1109                    total,
1110                    tr->cpu,
1111 #if defined(CONFIG_PREEMPT_NONE)
1112                    "server",
1113 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1114                    "desktop",
1115 #elif defined(CONFIG_PREEMPT_DESKTOP)
1116                    "preempt",
1117 #else
1118                    "unknown",
1119 #endif
1120                    /* These are reserved for later use */
1121                    0, 0, 0, 0);
1122 #ifdef CONFIG_SMP
1123         seq_printf(m, " #P:%d)\n", num_online_cpus());
1124 #else
1125         seq_puts(m, ")\n");
1126 #endif
1127         seq_puts(m, "    -----------------\n");
1128         seq_printf(m, "    | task: %.16s-%d "
1129                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1130                    data->comm, data->pid, data->uid, data->nice,
1131                    data->policy, data->rt_priority);
1132         seq_puts(m, "    -----------------\n");
1133
1134         if (data->critical_start) {
1135                 seq_puts(m, " => started at: ");
1136                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1137                 trace_print_seq(m, &iter->seq);
1138                 seq_puts(m, "\n => ended at:   ");
1139                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1140                 trace_print_seq(m, &iter->seq);
1141                 seq_puts(m, "\n");
1142         }
1143
1144         seq_puts(m, "\n");
1145 }
1146
1147 static void
1148 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1149 {
1150         int hardirq, softirq;
1151         char *comm;
1152
1153         comm = trace_find_cmdline(entry->pid);
1154
1155         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1156         trace_seq_printf(s, "%d", cpu);
1157         trace_seq_printf(s, "%c%c",
1158                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1159                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1160
1161         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1162         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1163         if (hardirq && softirq)
1164                 trace_seq_putc(s, 'H');
1165         else {
1166                 if (hardirq)
1167                         trace_seq_putc(s, 'h');
1168                 else {
1169                         if (softirq)
1170                                 trace_seq_putc(s, 's');
1171                         else
1172                                 trace_seq_putc(s, '.');
1173                 }
1174         }
1175
1176         if (entry->preempt_count)
1177                 trace_seq_printf(s, "%x", entry->preempt_count);
1178         else
1179                 trace_seq_puts(s, ".");
1180 }
1181
1182 unsigned long preempt_mark_thresh = 100;
1183
1184 static void
1185 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1186                     unsigned long rel_usecs)
1187 {
1188         trace_seq_printf(s, " %4lldus", abs_usecs);
1189         if (rel_usecs > preempt_mark_thresh)
1190                 trace_seq_puts(s, "!: ");
1191         else if (rel_usecs > 1)
1192                 trace_seq_puts(s, "+: ");
1193         else
1194                 trace_seq_puts(s, " : ");
1195 }
1196
1197 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1198
1199 static int
1200 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1201 {
1202         struct trace_seq *s = &iter->seq;
1203         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1204         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1205         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1206         struct trace_entry *entry = iter->ent;
1207         unsigned long abs_usecs;
1208         unsigned long rel_usecs;
1209         char *comm;
1210         int S;
1211         int i;
1212
1213         if (!next_entry)
1214                 next_entry = entry;
1215         rel_usecs = ns2usecs(next_entry->t - entry->t);
1216         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1217
1218         if (verbose) {
1219                 comm = trace_find_cmdline(entry->pid);
1220                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1221                                  " %ld.%03ldms (+%ld.%03ldms): ",
1222                                  comm,
1223                                  entry->pid, cpu, entry->flags,
1224                                  entry->preempt_count, trace_idx,
1225                                  ns2usecs(entry->t),
1226                                  abs_usecs/1000,
1227                                  abs_usecs % 1000, rel_usecs/1000,
1228                                  rel_usecs % 1000);
1229         } else {
1230                 if (entry->type != TRACE_STACK) {
1231                         lat_print_generic(s, entry, cpu);
1232                         lat_print_timestamp(s, abs_usecs, rel_usecs);
1233                 }
1234         }
1235         switch (entry->type) {
1236         case TRACE_FN:
1237                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1238                 trace_seq_puts(s, " (");
1239                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1240                 trace_seq_puts(s, ")\n");
1241                 break;
1242         case TRACE_CTX:
1243         case TRACE_WAKE:
1244                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1245                         state_to_char[entry->ctx.prev_state] : 'X';
1246                 comm = trace_find_cmdline(entry->ctx.next_pid);
1247                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d %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                                  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;
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         if (entry->type != TRACE_STACK) {
1297                 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1298                 if (!ret)
1299                         return 0;
1300                 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1301                 if (!ret)
1302                         return 0;
1303                 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1304                 if (!ret)
1305                         return 0;
1306         }
1307
1308         switch (entry->type) {
1309         case TRACE_FN:
1310                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1311                 if (!ret)
1312                         return 0;
1313                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1314                                                 entry->fn.parent_ip) {
1315                         ret = trace_seq_printf(s, " <-");
1316                         if (!ret)
1317                                 return 0;
1318                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1319                                                sym_flags);
1320                         if (!ret)
1321                                 return 0;
1322                 }
1323                 ret = trace_seq_printf(s, "\n");
1324                 if (!ret)
1325                         return 0;
1326                 break;
1327         case TRACE_CTX:
1328         case TRACE_WAKE:
1329                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1330                         state_to_char[entry->ctx.prev_state] : 'X';
1331                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d\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                 if (!ret)
1339                         return 0;
1340                 break;
1341         case TRACE_SPECIAL:
1342                 ret = trace_seq_printf(s, " %ld %ld %ld\n",
1343                                  entry->special.arg1,
1344                                  entry->special.arg2,
1345                                  entry->special.arg3);
1346                 if (!ret)
1347                         return 0;
1348                 break;
1349         case TRACE_STACK:
1350                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1351                         if (i) {
1352                                 ret = trace_seq_puts(s, " <= ");
1353                                 if (!ret)
1354                                         return 0;
1355                         }
1356                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1357                                                sym_flags);
1358                         if (!ret)
1359                                 return 0;
1360                 }
1361                 ret = trace_seq_puts(s, "\n");
1362                 if (!ret)
1363                         return 0;
1364                 break;
1365         }
1366         return 1;
1367 }
1368
1369 static int print_raw_fmt(struct trace_iterator *iter)
1370 {
1371         struct trace_seq *s = &iter->seq;
1372         struct trace_entry *entry;
1373         int ret;
1374         int S;
1375
1376         entry = iter->ent;
1377
1378         ret = trace_seq_printf(s, "%d %d %llu ",
1379                 entry->pid, iter->cpu, entry->t);
1380         if (!ret)
1381                 return 0;
1382
1383         switch (entry->type) {
1384         case TRACE_FN:
1385                 ret = trace_seq_printf(s, "%x %x\n",
1386                                         entry->fn.ip, entry->fn.parent_ip);
1387                 if (!ret)
1388                         return 0;
1389                 break;
1390         case TRACE_CTX:
1391         case TRACE_WAKE:
1392                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1393                         state_to_char[entry->ctx.prev_state] : 'X';
1394                 if (entry->type == TRACE_WAKE)
1395                         S = '+';
1396                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1397                                        entry->ctx.prev_pid,
1398                                        entry->ctx.prev_prio,
1399                                        S,
1400                                        entry->ctx.next_pid,
1401                                        entry->ctx.next_prio);
1402                 if (!ret)
1403                         return 0;
1404                 break;
1405         case TRACE_SPECIAL:
1406         case TRACE_STACK:
1407                 ret = trace_seq_printf(s, " %ld %ld %ld\n",
1408                                  entry->special.arg1,
1409                                  entry->special.arg2,
1410                                  entry->special.arg3);
1411                 if (!ret)
1412                         return 0;
1413                 break;
1414         }
1415         return 1;
1416 }
1417
1418 #define SEQ_PUT_FIELD_RET(s, x)                         \
1419 do {                                                    \
1420         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1421                 return 0;                               \
1422 } while (0)
1423
1424 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1425 do {                                                    \
1426         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1427                 return 0;                               \
1428 } while (0)
1429
1430 static int print_hex_fmt(struct trace_iterator *iter)
1431 {
1432         struct trace_seq *s = &iter->seq;
1433         unsigned char newline = '\n';
1434         struct trace_entry *entry;
1435         int S;
1436
1437         entry = iter->ent;
1438
1439         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1440         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1441         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1442
1443         switch (entry->type) {
1444         case TRACE_FN:
1445                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1446                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1447                 break;
1448         case TRACE_CTX:
1449         case TRACE_WAKE:
1450                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1451                         state_to_char[entry->ctx.prev_state] : 'X';
1452                 if (entry->type == TRACE_WAKE)
1453                         S = '+';
1454                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1455                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1456                 SEQ_PUT_HEX_FIELD_RET(s, S);
1457                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1458                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1459                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1460                 break;
1461         case TRACE_SPECIAL:
1462         case TRACE_STACK:
1463                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1464                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1465                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1466                 break;
1467         }
1468         SEQ_PUT_FIELD_RET(s, newline);
1469
1470         return 1;
1471 }
1472
1473 static int print_bin_fmt(struct trace_iterator *iter)
1474 {
1475         struct trace_seq *s = &iter->seq;
1476         struct trace_entry *entry;
1477
1478         entry = iter->ent;
1479
1480         SEQ_PUT_FIELD_RET(s, entry->pid);
1481         SEQ_PUT_FIELD_RET(s, entry->cpu);
1482         SEQ_PUT_FIELD_RET(s, entry->t);
1483
1484         switch (entry->type) {
1485         case TRACE_FN:
1486                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1487                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1488                 break;
1489         case TRACE_CTX:
1490                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1491                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1492                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1493                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1494                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1495                 break;
1496         case TRACE_SPECIAL:
1497         case TRACE_STACK:
1498                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1499                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1500                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1501                 break;
1502         }
1503         return 1;
1504 }
1505
1506 static int trace_empty(struct trace_iterator *iter)
1507 {
1508         struct trace_array_cpu *data;
1509         int cpu;
1510
1511         for_each_possible_cpu(cpu) {
1512                 data = iter->tr->data[cpu];
1513
1514                 if (head_page(data) && data->trace_idx &&
1515                     (data->trace_tail != data->trace_head ||
1516                      data->trace_tail_idx != data->trace_head_idx))
1517                         return 0;
1518         }
1519         return 1;
1520 }
1521
1522 static int print_trace_line(struct trace_iterator *iter)
1523 {
1524         if (trace_flags & TRACE_ITER_BIN)
1525                 return print_bin_fmt(iter);
1526
1527         if (trace_flags & TRACE_ITER_HEX)
1528                 return print_hex_fmt(iter);
1529
1530         if (trace_flags & TRACE_ITER_RAW)
1531                 return print_raw_fmt(iter);
1532
1533         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1534                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1535
1536         return print_trace_fmt(iter);
1537 }
1538
1539 static int s_show(struct seq_file *m, void *v)
1540 {
1541         struct trace_iterator *iter = v;
1542
1543         if (iter->ent == NULL) {
1544                 if (iter->tr) {
1545                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1546                         seq_puts(m, "#\n");
1547                 }
1548                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1549                         /* print nothing if the buffers are empty */
1550                         if (trace_empty(iter))
1551                                 return 0;
1552                         print_trace_header(m, iter);
1553                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1554                                 print_lat_help_header(m);
1555                 } else {
1556                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1557                                 print_func_help_header(m);
1558                 }
1559         } else {
1560                 print_trace_line(iter);
1561                 trace_print_seq(m, &iter->seq);
1562         }
1563
1564         return 0;
1565 }
1566
1567 static struct seq_operations tracer_seq_ops = {
1568         .start          = s_start,
1569         .next           = s_next,
1570         .stop           = s_stop,
1571         .show           = s_show,
1572 };
1573
1574 static struct trace_iterator *
1575 __tracing_open(struct inode *inode, struct file *file, int *ret)
1576 {
1577         struct trace_iterator *iter;
1578
1579         if (tracing_disabled) {
1580                 *ret = -ENODEV;
1581                 return NULL;
1582         }
1583
1584         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1585         if (!iter) {
1586                 *ret = -ENOMEM;
1587                 goto out;
1588         }
1589
1590         mutex_lock(&trace_types_lock);
1591         if (current_trace && current_trace->print_max)
1592                 iter->tr = &max_tr;
1593         else
1594                 iter->tr = inode->i_private;
1595         iter->trace = current_trace;
1596         iter->pos = -1;
1597
1598         /* TODO stop tracer */
1599         *ret = seq_open(file, &tracer_seq_ops);
1600         if (!*ret) {
1601                 struct seq_file *m = file->private_data;
1602                 m->private = iter;
1603
1604                 /* stop the trace while dumping */
1605                 if (iter->tr->ctrl)
1606                         tracer_enabled = 0;
1607
1608                 if (iter->trace && iter->trace->open)
1609                         iter->trace->open(iter);
1610         } else {
1611                 kfree(iter);
1612                 iter = NULL;
1613         }
1614         mutex_unlock(&trace_types_lock);
1615
1616  out:
1617         return iter;
1618 }
1619
1620 int tracing_open_generic(struct inode *inode, struct file *filp)
1621 {
1622         if (tracing_disabled)
1623                 return -ENODEV;
1624
1625         filp->private_data = inode->i_private;
1626         return 0;
1627 }
1628
1629 int tracing_release(struct inode *inode, struct file *file)
1630 {
1631         struct seq_file *m = (struct seq_file *)file->private_data;
1632         struct trace_iterator *iter = m->private;
1633
1634         mutex_lock(&trace_types_lock);
1635         if (iter->trace && iter->trace->close)
1636                 iter->trace->close(iter);
1637
1638         /* reenable tracing if it was previously enabled */
1639         if (iter->tr->ctrl)
1640                 tracer_enabled = 1;
1641         mutex_unlock(&trace_types_lock);
1642
1643         seq_release(inode, file);
1644         kfree(iter);
1645         return 0;
1646 }
1647
1648 static int tracing_open(struct inode *inode, struct file *file)
1649 {
1650         int ret;
1651
1652         __tracing_open(inode, file, &ret);
1653
1654         return ret;
1655 }
1656
1657 static int tracing_lt_open(struct inode *inode, struct file *file)
1658 {
1659         struct trace_iterator *iter;
1660         int ret;
1661
1662         iter = __tracing_open(inode, file, &ret);
1663
1664         if (!ret)
1665                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1666
1667         return ret;
1668 }
1669
1670
1671 static void *
1672 t_next(struct seq_file *m, void *v, loff_t *pos)
1673 {
1674         struct tracer *t = m->private;
1675
1676         (*pos)++;
1677
1678         if (t)
1679                 t = t->next;
1680
1681         m->private = t;
1682
1683         return t;
1684 }
1685
1686 static void *t_start(struct seq_file *m, loff_t *pos)
1687 {
1688         struct tracer *t = m->private;
1689         loff_t l = 0;
1690
1691         mutex_lock(&trace_types_lock);
1692         for (; t && l < *pos; t = t_next(m, t, &l))
1693                 ;
1694
1695         return t;
1696 }
1697
1698 static void t_stop(struct seq_file *m, void *p)
1699 {
1700         mutex_unlock(&trace_types_lock);
1701 }
1702
1703 static int t_show(struct seq_file *m, void *v)
1704 {
1705         struct tracer *t = v;
1706
1707         if (!t)
1708                 return 0;
1709
1710         seq_printf(m, "%s", t->name);
1711         if (t->next)
1712                 seq_putc(m, ' ');
1713         else
1714                 seq_putc(m, '\n');
1715
1716         return 0;
1717 }
1718
1719 static struct seq_operations show_traces_seq_ops = {
1720         .start          = t_start,
1721         .next           = t_next,
1722         .stop           = t_stop,
1723         .show           = t_show,
1724 };
1725
1726 static int show_traces_open(struct inode *inode, struct file *file)
1727 {
1728         int ret;
1729
1730         if (tracing_disabled)
1731                 return -ENODEV;
1732
1733         ret = seq_open(file, &show_traces_seq_ops);
1734         if (!ret) {
1735                 struct seq_file *m = file->private_data;
1736                 m->private = trace_types;
1737         }
1738
1739         return ret;
1740 }
1741
1742 static struct file_operations tracing_fops = {
1743         .open           = tracing_open,
1744         .read           = seq_read,
1745         .llseek         = seq_lseek,
1746         .release        = tracing_release,
1747 };
1748
1749 static struct file_operations tracing_lt_fops = {
1750         .open           = tracing_lt_open,
1751         .read           = seq_read,
1752         .llseek         = seq_lseek,
1753         .release        = tracing_release,
1754 };
1755
1756 static struct file_operations show_traces_fops = {
1757         .open           = show_traces_open,
1758         .read           = seq_read,
1759         .release        = seq_release,
1760 };
1761
1762 /*
1763  * Only trace on a CPU if the bitmask is set:
1764  */
1765 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1766
1767 /*
1768  * When tracing/tracing_cpu_mask is modified then this holds
1769  * the new bitmask we are about to install:
1770  */
1771 static cpumask_t tracing_cpumask_new;
1772
1773 /*
1774  * The tracer itself will not take this lock, but still we want
1775  * to provide a consistent cpumask to user-space:
1776  */
1777 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1778
1779 /*
1780  * Temporary storage for the character representation of the
1781  * CPU bitmask (and one more byte for the newline):
1782  */
1783 static char mask_str[NR_CPUS + 1];
1784
1785 static ssize_t
1786 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1787                      size_t count, loff_t *ppos)
1788 {
1789         int len;
1790
1791         mutex_lock(&tracing_cpumask_update_lock);
1792
1793         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1794         if (count - len < 2) {
1795                 count = -EINVAL;
1796                 goto out_err;
1797         }
1798         len += sprintf(mask_str + len, "\n");
1799         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1800
1801 out_err:
1802         mutex_unlock(&tracing_cpumask_update_lock);
1803
1804         return count;
1805 }
1806
1807 static ssize_t
1808 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1809                       size_t count, loff_t *ppos)
1810 {
1811         int err, cpu;
1812
1813         mutex_lock(&tracing_cpumask_update_lock);
1814         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1815         if (err)
1816                 goto err_unlock;
1817
1818         spin_lock_irq(&ftrace_max_lock);
1819         for_each_possible_cpu(cpu) {
1820                 /*
1821                  * Increase/decrease the disabled counter if we are
1822                  * about to flip a bit in the cpumask:
1823                  */
1824                 if (cpu_isset(cpu, tracing_cpumask) &&
1825                                 !cpu_isset(cpu, tracing_cpumask_new)) {
1826                         atomic_inc(&global_trace.data[cpu]->disabled);
1827                 }
1828                 if (!cpu_isset(cpu, tracing_cpumask) &&
1829                                 cpu_isset(cpu, tracing_cpumask_new)) {
1830                         atomic_dec(&global_trace.data[cpu]->disabled);
1831                 }
1832         }
1833         spin_unlock_irq(&ftrace_max_lock);
1834
1835         tracing_cpumask = tracing_cpumask_new;
1836
1837         mutex_unlock(&tracing_cpumask_update_lock);
1838
1839         return count;
1840
1841 err_unlock:
1842         mutex_unlock(&tracing_cpumask_update_lock);
1843
1844         return err;
1845 }
1846
1847 static struct file_operations tracing_cpumask_fops = {
1848         .open           = tracing_open_generic,
1849         .read           = tracing_cpumask_read,
1850         .write          = tracing_cpumask_write,
1851 };
1852
1853 static ssize_t
1854 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1855                        size_t cnt, loff_t *ppos)
1856 {
1857         char *buf;
1858         int r = 0;
1859         int len = 0;
1860         int i;
1861
1862         /* calulate max size */
1863         for (i = 0; trace_options[i]; i++) {
1864                 len += strlen(trace_options[i]);
1865                 len += 3; /* "no" and space */
1866         }
1867
1868         /* +2 for \n and \0 */
1869         buf = kmalloc(len + 2, GFP_KERNEL);
1870         if (!buf)
1871                 return -ENOMEM;
1872
1873         for (i = 0; trace_options[i]; i++) {
1874                 if (trace_flags & (1 << i))
1875                         r += sprintf(buf + r, "%s ", trace_options[i]);
1876                 else
1877                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1878         }
1879
1880         r += sprintf(buf + r, "\n");
1881         WARN_ON(r >= len + 2);
1882
1883         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1884
1885         kfree(buf);
1886
1887         return r;
1888 }
1889
1890 static ssize_t
1891 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1892                         size_t cnt, loff_t *ppos)
1893 {
1894         char buf[64];
1895         char *cmp = buf;
1896         int neg = 0;
1897         int i;
1898
1899         if (cnt > 63)
1900                 cnt = 63;
1901
1902         if (copy_from_user(&buf, ubuf, cnt))
1903                 return -EFAULT;
1904
1905         buf[cnt] = 0;
1906
1907         if (strncmp(buf, "no", 2) == 0) {
1908                 neg = 1;
1909                 cmp += 2;
1910         }
1911
1912         for (i = 0; trace_options[i]; i++) {
1913                 int len = strlen(trace_options[i]);
1914
1915                 if (strncmp(cmp, trace_options[i], len) == 0) {
1916                         if (neg)
1917                                 trace_flags &= ~(1 << i);
1918                         else
1919                                 trace_flags |= (1 << i);
1920                         break;
1921                 }
1922         }
1923         /*
1924          * If no option could be set, return an error:
1925          */
1926         if (!trace_options[i])
1927                 return -EINVAL;
1928
1929         filp->f_pos += cnt;
1930
1931         return cnt;
1932 }
1933
1934 static struct file_operations tracing_iter_fops = {
1935         .open           = tracing_open_generic,
1936         .read           = tracing_iter_ctrl_read,
1937         .write          = tracing_iter_ctrl_write,
1938 };
1939
1940 static const char readme_msg[] =
1941         "tracing mini-HOWTO:\n\n"
1942         "# mkdir /debug\n"
1943         "# mount -t debugfs nodev /debug\n\n"
1944         "# cat /debug/tracing/available_tracers\n"
1945         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1946         "# cat /debug/tracing/current_tracer\n"
1947         "none\n"
1948         "# echo sched_switch > /debug/tracing/current_tracer\n"
1949         "# cat /debug/tracing/current_tracer\n"
1950         "sched_switch\n"
1951         "# cat /debug/tracing/iter_ctrl\n"
1952         "noprint-parent nosym-offset nosym-addr noverbose\n"
1953         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1954         "# echo 1 > /debug/tracing/tracing_enabled\n"
1955         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1956         "echo 0 > /debug/tracing/tracing_enabled\n"
1957 ;
1958
1959 static ssize_t
1960 tracing_readme_read(struct file *filp, char __user *ubuf,
1961                        size_t cnt, loff_t *ppos)
1962 {
1963         return simple_read_from_buffer(ubuf, cnt, ppos,
1964                                         readme_msg, strlen(readme_msg));
1965 }
1966
1967 static struct file_operations tracing_readme_fops = {
1968         .open           = tracing_open_generic,
1969         .read           = tracing_readme_read,
1970 };
1971
1972 static ssize_t
1973 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1974                   size_t cnt, loff_t *ppos)
1975 {
1976         struct trace_array *tr = filp->private_data;
1977         char buf[64];
1978         int r;
1979
1980         r = sprintf(buf, "%ld\n", tr->ctrl);
1981         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1982 }
1983
1984 static ssize_t
1985 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1986                    size_t cnt, loff_t *ppos)
1987 {
1988         struct trace_array *tr = filp->private_data;
1989         long val;
1990         char buf[64];
1991
1992         if (cnt > 63)
1993                 cnt = 63;
1994
1995         if (copy_from_user(&buf, ubuf, cnt))
1996                 return -EFAULT;
1997
1998         buf[cnt] = 0;
1999
2000         val = simple_strtoul(buf, NULL, 10);
2001
2002         val = !!val;
2003
2004         mutex_lock(&trace_types_lock);
2005         if (tr->ctrl ^ val) {
2006                 if (val)
2007                         tracer_enabled = 1;
2008                 else
2009                         tracer_enabled = 0;
2010
2011                 tr->ctrl = val;
2012
2013                 if (current_trace && current_trace->ctrl_update)
2014                         current_trace->ctrl_update(tr);
2015         }
2016         mutex_unlock(&trace_types_lock);
2017
2018         filp->f_pos += cnt;
2019
2020         return cnt;
2021 }
2022
2023 static ssize_t
2024 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2025                        size_t cnt, loff_t *ppos)
2026 {
2027         char buf[max_tracer_type_len+2];
2028         int r;
2029
2030         mutex_lock(&trace_types_lock);
2031         if (current_trace)
2032                 r = sprintf(buf, "%s\n", current_trace->name);
2033         else
2034                 r = sprintf(buf, "\n");
2035         mutex_unlock(&trace_types_lock);
2036
2037         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2038 }
2039
2040 static ssize_t
2041 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2042                         size_t cnt, loff_t *ppos)
2043 {
2044         struct trace_array *tr = &global_trace;
2045         struct tracer *t;
2046         char buf[max_tracer_type_len+1];
2047         int i;
2048
2049         if (cnt > max_tracer_type_len)
2050                 cnt = max_tracer_type_len;
2051
2052         if (copy_from_user(&buf, ubuf, cnt))
2053                 return -EFAULT;
2054
2055         buf[cnt] = 0;
2056
2057         /* strip ending whitespace. */
2058         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2059                 buf[i] = 0;
2060
2061         mutex_lock(&trace_types_lock);
2062         for (t = trace_types; t; t = t->next) {
2063                 if (strcmp(t->name, buf) == 0)
2064                         break;
2065         }
2066         if (!t || t == current_trace)
2067                 goto out;
2068
2069         if (current_trace && current_trace->reset)
2070                 current_trace->reset(tr);
2071
2072         current_trace = t;
2073         if (t->init)
2074                 t->init(tr);
2075
2076  out:
2077         mutex_unlock(&trace_types_lock);
2078
2079         filp->f_pos += cnt;
2080
2081         return cnt;
2082 }
2083
2084 static ssize_t
2085 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2086                      size_t cnt, loff_t *ppos)
2087 {
2088         unsigned long *ptr = filp->private_data;
2089         char buf[64];
2090         int r;
2091
2092         r = snprintf(buf, 64, "%ld\n",
2093                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2094         if (r > 64)
2095                 r = 64;
2096         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2097 }
2098
2099 static ssize_t
2100 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2101                       size_t cnt, loff_t *ppos)
2102 {
2103         long *ptr = filp->private_data;
2104         long val;
2105         char buf[64];
2106
2107         if (cnt > 63)
2108                 cnt = 63;
2109
2110         if (copy_from_user(&buf, ubuf, cnt))
2111                 return -EFAULT;
2112
2113         buf[cnt] = 0;
2114
2115         val = simple_strtoul(buf, NULL, 10);
2116
2117         *ptr = val * 1000;
2118
2119         return cnt;
2120 }
2121
2122 static atomic_t tracing_reader;
2123
2124 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2125 {
2126         struct trace_iterator *iter;
2127
2128         if (tracing_disabled)
2129                 return -ENODEV;
2130
2131         /* We only allow for reader of the pipe */
2132         if (atomic_inc_return(&tracing_reader) != 1) {
2133                 atomic_dec(&tracing_reader);
2134                 return -EBUSY;
2135         }
2136
2137         /* create a buffer to store the information to pass to userspace */
2138         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2139         if (!iter)
2140                 return -ENOMEM;
2141
2142         iter->tr = &global_trace;
2143
2144         filp->private_data = iter;
2145
2146         return 0;
2147 }
2148
2149 static int tracing_release_pipe(struct inode *inode, struct file *file)
2150 {
2151         struct trace_iterator *iter = file->private_data;
2152
2153         kfree(iter);
2154         atomic_dec(&tracing_reader);
2155
2156         return 0;
2157 }
2158
2159 static unsigned int
2160 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2161 {
2162         struct trace_iterator *iter = filp->private_data;
2163
2164         if (trace_flags & TRACE_ITER_BLOCK) {
2165                 /*
2166                  * Always select as readable when in blocking mode
2167                  */
2168                 return POLLIN | POLLRDNORM;
2169         }
2170         else {
2171                 if (!trace_empty(iter))
2172                         return POLLIN | POLLRDNORM;
2173                 poll_wait(filp, &trace_wait, poll_table);
2174                 if (!trace_empty(iter))
2175                         return POLLIN | POLLRDNORM;
2176
2177                 return 0;
2178         }
2179 }
2180
2181 /*
2182  * Consumer reader.
2183  */
2184 static ssize_t
2185 tracing_read_pipe(struct file *filp, char __user *ubuf,
2186                   size_t cnt, loff_t *ppos)
2187 {
2188         struct trace_iterator *iter = filp->private_data;
2189         struct trace_array_cpu *data;
2190         static cpumask_t mask;
2191         static int start;
2192         unsigned long flags;
2193 #ifdef CONFIG_FTRACE
2194         int ftrace_save;
2195 #endif
2196         int read = 0;
2197         int cpu;
2198         int len;
2199         int ret;
2200
2201         /* return any leftover data */
2202         if (iter->seq.len > start) {
2203                 len = iter->seq.len - start;
2204                 if (cnt > len)
2205                         cnt = len;
2206                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2207                 if (ret)
2208                         cnt = -EFAULT;
2209
2210                 start += len;
2211
2212                 return cnt;
2213         }
2214
2215         trace_seq_reset(&iter->seq);
2216         start = 0;
2217
2218         while (trace_empty(iter)) {
2219                 if (!(trace_flags & TRACE_ITER_BLOCK))
2220                         return -EWOULDBLOCK;
2221                 /*
2222                  * This is a make-shift waitqueue. The reason we don't use
2223                  * an actual wait queue is because:
2224                  *  1) we only ever have one waiter
2225                  *  2) the tracing, traces all functions, we don't want
2226                  *     the overhead of calling wake_up and friends
2227                  *     (and tracing them too)
2228                  *     Anyway, this is really very primitive wakeup.
2229                  */
2230                 set_current_state(TASK_INTERRUPTIBLE);
2231                 iter->tr->waiter = current;
2232
2233                 /* sleep for one second, and try again. */
2234                 schedule_timeout(HZ);
2235
2236                 iter->tr->waiter = NULL;
2237
2238                 if (signal_pending(current))
2239                         return -EINTR;
2240
2241                 /*
2242                  * We block until we read something and tracing is disabled.
2243                  * We still block if tracing is disabled, but we have never
2244                  * read anything. This allows a user to cat this file, and
2245                  * then enable tracing. But after we have read something,
2246                  * we give an EOF when tracing is again disabled.
2247                  *
2248                  * iter->pos will be 0 if we haven't read anything.
2249                  */
2250                 if (!tracer_enabled && iter->pos)
2251                         break;
2252
2253                 continue;
2254         }
2255
2256         /* stop when tracing is finished */
2257         if (trace_empty(iter))
2258                 return 0;
2259
2260         if (cnt >= PAGE_SIZE)
2261                 cnt = PAGE_SIZE - 1;
2262
2263         memset(iter, 0, sizeof(*iter));
2264         iter->tr = &global_trace;
2265         iter->pos = -1;
2266
2267         /*
2268          * We need to stop all tracing on all CPUS to read the
2269          * the next buffer. This is a bit expensive, but is
2270          * not done often. We fill all what we can read,
2271          * and then release the locks again.
2272          */
2273
2274         cpus_clear(mask);
2275         local_irq_save(flags);
2276 #ifdef CONFIG_FTRACE
2277         ftrace_save = ftrace_enabled;
2278         ftrace_enabled = 0;
2279 #endif
2280         smp_wmb();
2281         for_each_possible_cpu(cpu) {
2282                 data = iter->tr->data[cpu];
2283
2284                 if (!head_page(data) || !data->trace_idx)
2285                         continue;
2286
2287                 atomic_inc(&data->disabled);
2288                 cpu_set(cpu, mask);
2289         }
2290
2291         for_each_cpu_mask(cpu, mask) {
2292                 data = iter->tr->data[cpu];
2293                 spin_lock(&data->lock);
2294         }
2295
2296         while (find_next_entry_inc(iter) != NULL) {
2297                 int len = iter->seq.len;
2298
2299                 ret = print_trace_line(iter);
2300                 if (!ret) {
2301                         /* don't print partial lines */
2302                         iter->seq.len = len;
2303                         break;
2304                 }
2305
2306                 trace_consume(iter);
2307
2308                 if (iter->seq.len >= cnt)
2309                         break;
2310         }
2311
2312         for_each_cpu_mask(cpu, mask) {
2313                 data = iter->tr->data[cpu];
2314                 spin_unlock(&data->lock);
2315         }
2316
2317         for_each_cpu_mask(cpu, mask) {
2318                 data = iter->tr->data[cpu];
2319                 atomic_dec(&data->disabled);
2320         }
2321 #ifdef CONFIG_FTRACE
2322         ftrace_enabled = ftrace_save;
2323 #endif
2324         local_irq_restore(flags);
2325
2326         /* Now copy what we have to the user */
2327         read = iter->seq.len;
2328         if (read > cnt)
2329                 read = cnt;
2330
2331         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2332
2333         if (read < iter->seq.len)
2334                 start = read;
2335         else
2336                 trace_seq_reset(&iter->seq);
2337
2338         if (ret)
2339                 read = -EFAULT;
2340
2341         return read;
2342 }
2343
2344 static struct file_operations tracing_max_lat_fops = {
2345         .open           = tracing_open_generic,
2346         .read           = tracing_max_lat_read,
2347         .write          = tracing_max_lat_write,
2348 };
2349
2350 static struct file_operations tracing_ctrl_fops = {
2351         .open           = tracing_open_generic,
2352         .read           = tracing_ctrl_read,
2353         .write          = tracing_ctrl_write,
2354 };
2355
2356 static struct file_operations set_tracer_fops = {
2357         .open           = tracing_open_generic,
2358         .read           = tracing_set_trace_read,
2359         .write          = tracing_set_trace_write,
2360 };
2361
2362 static struct file_operations tracing_pipe_fops = {
2363         .open           = tracing_open_pipe,
2364         .poll           = tracing_poll_pipe,
2365         .read           = tracing_read_pipe,
2366         .release        = tracing_release_pipe,
2367 };
2368
2369 #ifdef CONFIG_DYNAMIC_FTRACE
2370
2371 static ssize_t
2372 tracing_read_long(struct file *filp, char __user *ubuf,
2373                   size_t cnt, loff_t *ppos)
2374 {
2375         unsigned long *p = filp->private_data;
2376         char buf[64];
2377         int r;
2378
2379         r = sprintf(buf, "%ld\n", *p);
2380
2381         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2382 }
2383
2384 static struct file_operations tracing_read_long_fops = {
2385         .open           = tracing_open_generic,
2386         .read           = tracing_read_long,
2387 };
2388 #endif
2389
2390 static struct dentry *d_tracer;
2391
2392 struct dentry *tracing_init_dentry(void)
2393 {
2394         static int once;
2395
2396         if (d_tracer)
2397                 return d_tracer;
2398
2399         d_tracer = debugfs_create_dir("tracing", NULL);
2400
2401         if (!d_tracer && !once) {
2402                 once = 1;
2403                 pr_warning("Could not create debugfs directory 'tracing'\n");
2404                 return NULL;
2405         }
2406
2407         return d_tracer;
2408 }
2409
2410 #ifdef CONFIG_FTRACE_SELFTEST
2411 /* Let selftest have access to static functions in this file */
2412 #include "trace_selftest.c"
2413 #endif
2414
2415 static __init void tracer_init_debugfs(void)
2416 {
2417         struct dentry *d_tracer;
2418         struct dentry *entry;
2419
2420         d_tracer = tracing_init_dentry();
2421
2422         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2423                                     &global_trace, &tracing_ctrl_fops);
2424         if (!entry)
2425                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2426
2427         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2428                                     NULL, &tracing_iter_fops);
2429         if (!entry)
2430                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2431
2432         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2433                                     NULL, &tracing_cpumask_fops);
2434         if (!entry)
2435                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2436
2437         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2438                                     &global_trace, &tracing_lt_fops);
2439         if (!entry)
2440                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2441
2442         entry = debugfs_create_file("trace", 0444, d_tracer,
2443                                     &global_trace, &tracing_fops);
2444         if (!entry)
2445                 pr_warning("Could not create debugfs 'trace' entry\n");
2446
2447         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2448                                     &global_trace, &show_traces_fops);
2449         if (!entry)
2450                 pr_warning("Could not create debugfs 'trace' entry\n");
2451
2452         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2453                                     &global_trace, &set_tracer_fops);
2454         if (!entry)
2455                 pr_warning("Could not create debugfs 'trace' entry\n");
2456
2457         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2458                                     &tracing_max_latency,
2459                                     &tracing_max_lat_fops);
2460         if (!entry)
2461                 pr_warning("Could not create debugfs "
2462                            "'tracing_max_latency' entry\n");
2463
2464         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2465                                     &tracing_thresh, &tracing_max_lat_fops);
2466         if (!entry)
2467                 pr_warning("Could not create debugfs "
2468                            "'tracing_threash' entry\n");
2469         entry = debugfs_create_file("README", 0644, d_tracer,
2470                                     NULL, &tracing_readme_fops);
2471         if (!entry)
2472                 pr_warning("Could not create debugfs 'README' entry\n");
2473
2474         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2475                                     NULL, &tracing_pipe_fops);
2476         if (!entry)
2477                 pr_warning("Could not create debugfs "
2478                            "'tracing_threash' entry\n");
2479
2480 #ifdef CONFIG_DYNAMIC_FTRACE
2481         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2482                                     &ftrace_update_tot_cnt,
2483                                     &tracing_read_long_fops);
2484         if (!entry)
2485                 pr_warning("Could not create debugfs "
2486                            "'dyn_ftrace_total_info' entry\n");
2487 #endif
2488 }
2489
2490 /* dummy trace to disable tracing */
2491 static struct tracer no_tracer __read_mostly =
2492 {
2493         .name           = "none",
2494 };
2495
2496 static int trace_alloc_page(void)
2497 {
2498         struct trace_array_cpu *data;
2499         struct page *page, *tmp;
2500         LIST_HEAD(pages);
2501         void *array;
2502         int i;
2503
2504         /* first allocate a page for each CPU */
2505         for_each_possible_cpu(i) {
2506                 array = (void *)__get_free_page(GFP_KERNEL);
2507                 if (array == NULL) {
2508                         printk(KERN_ERR "tracer: failed to allocate page"
2509                                "for trace buffer!\n");
2510                         goto free_pages;
2511                 }
2512
2513                 page = virt_to_page(array);
2514                 list_add(&page->lru, &pages);
2515
2516 /* Only allocate if we are actually using the max trace */
2517 #ifdef CONFIG_TRACER_MAX_TRACE
2518                 array = (void *)__get_free_page(GFP_KERNEL);
2519                 if (array == NULL) {
2520                         printk(KERN_ERR "tracer: failed to allocate page"
2521                                "for trace buffer!\n");
2522                         goto free_pages;
2523                 }
2524                 page = virt_to_page(array);
2525                 list_add(&page->lru, &pages);
2526 #endif
2527         }
2528
2529         /* Now that we successfully allocate a page per CPU, add them */
2530         for_each_possible_cpu(i) {
2531                 data = global_trace.data[i];
2532                 spin_lock_init(&data->lock);
2533                 lockdep_set_class(&data->lock, &data->lock_key);
2534                 page = list_entry(pages.next, struct page, lru);
2535                 list_del_init(&page->lru);
2536                 list_add_tail(&page->lru, &data->trace_pages);
2537                 ClearPageLRU(page);
2538
2539 #ifdef CONFIG_TRACER_MAX_TRACE
2540                 data = max_tr.data[i];
2541                 spin_lock_init(&data->lock);
2542                 lockdep_set_class(&data->lock, &data->lock_key);
2543                 page = list_entry(pages.next, struct page, lru);
2544                 list_del_init(&page->lru);
2545                 list_add_tail(&page->lru, &data->trace_pages);
2546                 SetPageLRU(page);
2547 #endif
2548         }
2549         global_trace.entries += ENTRIES_PER_PAGE;
2550
2551         return 0;
2552
2553  free_pages:
2554         list_for_each_entry_safe(page, tmp, &pages, lru) {
2555                 list_del_init(&page->lru);
2556                 __free_page(page);
2557         }
2558         return -ENOMEM;
2559 }
2560
2561 __init static int tracer_alloc_buffers(void)
2562 {
2563         struct trace_array_cpu *data;
2564         void *array;
2565         struct page *page;
2566         int pages = 0;
2567         int ret = -ENOMEM;
2568         int i;
2569
2570         global_trace.ctrl = tracer_enabled;
2571
2572         /* Allocate the first page for all buffers */
2573         for_each_possible_cpu(i) {
2574                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2575                 max_tr.data[i] = &per_cpu(max_data, i);
2576
2577                 array = (void *)__get_free_page(GFP_KERNEL);
2578                 if (array == NULL) {
2579                         printk(KERN_ERR "tracer: failed to allocate page"
2580                                "for trace buffer!\n");
2581                         goto free_buffers;
2582                 }
2583
2584                 /* set the array to the list */
2585                 INIT_LIST_HEAD(&data->trace_pages);
2586                 page = virt_to_page(array);
2587                 list_add(&page->lru, &data->trace_pages);
2588                 /* use the LRU flag to differentiate the two buffers */
2589                 ClearPageLRU(page);
2590
2591 /* Only allocate if we are actually using the max trace */
2592 #ifdef CONFIG_TRACER_MAX_TRACE
2593                 array = (void *)__get_free_page(GFP_KERNEL);
2594                 if (array == NULL) {
2595                         printk(KERN_ERR "tracer: failed to allocate page"
2596                                "for trace buffer!\n");
2597                         goto free_buffers;
2598                 }
2599
2600                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2601                 page = virt_to_page(array);
2602                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2603                 SetPageLRU(page);
2604 #endif
2605         }
2606
2607         /*
2608          * Since we allocate by orders of pages, we may be able to
2609          * round up a bit.
2610          */
2611         global_trace.entries = ENTRIES_PER_PAGE;
2612         pages++;
2613
2614         while (global_trace.entries < trace_nr_entries) {
2615                 if (trace_alloc_page())
2616                         break;
2617                 pages++;
2618         }
2619         max_tr.entries = global_trace.entries;
2620
2621         pr_info("tracer: %d pages allocated for %ld",
2622                 pages, trace_nr_entries);
2623         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2624         pr_info("   actual entries %ld\n", global_trace.entries);
2625
2626         tracer_init_debugfs();
2627
2628         trace_init_cmdlines();
2629
2630         register_tracer(&no_tracer);
2631         current_trace = &no_tracer;
2632
2633         /* All seems OK, enable tracing */
2634         tracing_disabled = 0;
2635
2636         return 0;
2637
2638  free_buffers:
2639         for (i-- ; i >= 0; i--) {
2640                 struct page *page, *tmp;
2641                 struct trace_array_cpu *data = global_trace.data[i];
2642
2643                 if (data) {
2644                         list_for_each_entry_safe(page, tmp,
2645                                                  &data->trace_pages, lru) {
2646                                 list_del_init(&page->lru);
2647                                 __free_page(page);
2648                         }
2649                 }
2650
2651 #ifdef CONFIG_TRACER_MAX_TRACE
2652                 data = max_tr.data[i];
2653                 if (data) {
2654                         list_for_each_entry_safe(page, tmp,
2655                                                  &data->trace_pages, lru) {
2656                                 list_del_init(&page->lru);
2657                                 __free_page(page);
2658                         }
2659                 }
2660 #endif
2661         }
2662         return ret;
2663 }
2664 fs_initcall(tracer_alloc_buffers);