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/notifier.h>
  18 #include <linux/debugfs.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/hardirq.h>
  21 #include <linux/linkage.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/ftrace.h>
  24 #include <linux/module.h>
  25 #include <linux/percpu.h>
  26 #include <linux/kdebug.h>
  27 #include <linux/ctype.h>
  28 #include <linux/init.h>
  29 #include <linux/poll.h>
  30 #include <linux/gfp.h>
  31 #include <linux/fs.h>
  32 #include <linux/kprobes.h>
  33 #include <linux/writeback.h>
  34
  35 #include <linux/stacktrace.h>
  36 #include <linux/ring_buffer.h>
  37 #include <linux/irqflags.h>
  38
  39 #include "trace.h"
  40
  41 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
  42
  43 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  44 unsigned long __read_mostly     tracing_thresh;
  45
  46 /* For tracers that don't implement custom flags */
  47 static struct tracer_opt dummy_tracer_opt[] = {
  48         { }
  49 };
  50
  51 static struct tracer_flags dummy_tracer_flags = {
  52         .val = 0,
  53         .opts = dummy_tracer_opt
  54 };
  55
  56 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
  57 {
  58         return 0;
  59 }
  60
  61 /*
  62  * Kill all tracing for good (never come back).
  63  * It is initialized to 1 but will turn to zero if the initialization
  64  * of the tracer is successful. But that is the only place that sets
  65  * this back to zero.
  66  */
  67 int tracing_disabled = 1;
  68
  69 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
  70
  71 static inline void ftrace_disable_cpu(void)
  72 {
  73         preempt_disable();
  74         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
  75 }
  76
  77 static inline void ftrace_enable_cpu(void)
  78 {
  79         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
  80         preempt_enable();
  81 }
  82
  83 static cpumask_t __read_mostly          tracing_buffer_mask;
  84
  85 #define for_each_tracing_cpu(cpu)       \
  86         for_each_cpu_mask(cpu, tracing_buffer_mask)
  87
  88 /*
  89  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
  90  *
  91  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
  92  * is set, then ftrace_dump is called. This will output the contents
  93  * of the ftrace buffers to the console.  This is very useful for
  94  * capturing traces that lead to crashes and outputing it to a
  95  * serial console.
  96  *
  97  * It is default off, but you can enable it with either specifying
  98  * "ftrace_dump_on_oops" in the kernel command line, or setting
  99  * /proc/sys/kernel/ftrace_dump_on_oops to true.
 100  */
 101 int ftrace_dump_on_oops;
 102
 103 static int tracing_set_tracer(char *buf);
 104
 105 static int __init set_ftrace(char *str)
 106 {
 107         tracing_set_tracer(str);
 108         return 1;
 109 }
 110 __setup("ftrace", set_ftrace);
 111
 112 static int __init set_ftrace_dump_on_oops(char *str)
 113 {
 114         ftrace_dump_on_oops = 1;
 115         return 1;
 116 }
 117 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
 118
 119 long
 120 ns2usecs(cycle_t nsec)
 121 {
 122         nsec += 500;
 123         do_div(nsec, 1000);
 124         return nsec;
 125 }
 126
 127 cycle_t ftrace_now(int cpu)
 128 {
 129         u64 ts = ring_buffer_time_stamp(cpu);
 130         ring_buffer_normalize_time_stamp(cpu, &ts);
 131         return ts;
 132 }
 133
 134 /*
 135  * The global_trace is the descriptor that holds the tracing
 136  * buffers for the live tracing. For each CPU, it contains
 137  * a link list of pages that will store trace entries. The
 138  * page descriptor of the pages in the memory is used to hold
 139  * the link list by linking the lru item in the page descriptor
 140  * to each of the pages in the buffer per CPU.
 141  *
 142  * For each active CPU there is a data field that holds the
 143  * pages for the buffer for that CPU. Each CPU has the same number
 144  * of pages allocated for its buffer.
 145  */
 146 static struct trace_array       global_trace;
 147
 148 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
 149
 150 /*
 151  * The max_tr is used to snapshot the global_trace when a maximum
 152  * latency is reached. Some tracers will use this to store a maximum
 153  * trace while it continues examining live traces.
 154  *
 155  * The buffers for the max_tr are set up the same as the global_trace.
 156  * When a snapshot is taken, the link list of the max_tr is swapped
 157  * with the link list of the global_trace and the buffers are reset for
 158  * the global_trace so the tracing can continue.
 159  */
 160 static struct trace_array       max_tr;
 161
 162 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 163
 164 /* tracer_enabled is used to toggle activation of a tracer */
 165 static int                      tracer_enabled = 1;
 166
 167 /**
 168  * tracing_is_enabled - return tracer_enabled status
 169  *
 170  * This function is used by other tracers to know the status
 171  * of the tracer_enabled flag.  Tracers may use this function
 172  * to know if it should enable their features when starting
 173  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
 174  */
 175 int tracing_is_enabled(void)
 176 {
 177         return tracer_enabled;
 178 }
 179
 180 /* function tracing enabled */
 181 int                             ftrace_function_enabled;
 182
 183 /*
 184  * trace_buf_size is the size in bytes that is allocated
 185  * for a buffer. Note, the number of bytes is always rounded
 186  * to page size.
 187  *
 188  * This number is purposely set to a low number of 16384.
 189  * If the dump on oops happens, it will be much appreciated
 190  * to not have to wait for all that output. Anyway this can be
 191  * boot time and run time configurable.
 192  */
 193 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
 194
 195 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
 196
 197 /* trace_types holds a link list of available tracers. */
 198 static struct tracer            *trace_types __read_mostly;
 199
 200 /* current_trace points to the tracer that is currently active */
 201 static struct tracer            *current_trace __read_mostly;
 202
 203 /*
 204  * max_tracer_type_len is used to simplify the allocating of
 205  * buffers to read userspace tracer names. We keep track of
 206  * the longest tracer name registered.
 207  */
 208 static int                      max_tracer_type_len;
 209
 210 /*
 211  * trace_types_lock is used to protect the trace_types list.
 212  * This lock is also used to keep user access serialized.
 213  * Accesses from userspace will grab this lock while userspace
 214  * activities happen inside the kernel.
 215  */
 216 static DEFINE_MUTEX(trace_types_lock);
 217
 218 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
 219 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 220
 221 /* trace_flags holds trace_options default values */
 222 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
 223         TRACE_ITER_ANNOTATE;
 224
 225 /**
 226  * trace_wake_up - wake up tasks waiting for trace input
 227  *
 228  * Simply wakes up any task that is blocked on the trace_wait
 229  * queue. These is used with trace_poll for tasks polling the trace.
 230  */
 231 void trace_wake_up(void)
 232 {
 233         /*
 234          * The runqueue_is_locked() can fail, but this is the best we
 235          * have for now:
 236          */
 237         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
 238                 wake_up(&trace_wait);
 239 }
 240
 241 static int __init set_buf_size(char *str)
 242 {
 243         unsigned long buf_size;
 244         int ret;
 245
 246         if (!str)
 247                 return 0;
 248         ret = strict_strtoul(str, 0, &buf_size);
 249         /* nr_entries can not be zero */
 250         if (ret < 0 || buf_size == 0)
 251                 return 0;
 252         trace_buf_size = buf_size;
 253         return 1;
 254 }
 255 __setup("trace_buf_size=", set_buf_size);
 256
 257 unsigned long nsecs_to_usecs(unsigned long nsecs)
 258 {
 259         return nsecs / 1000;
 260 }
 261
 262 /* These must match the bit postions in trace_iterator_flags */
 263 static const char *trace_options[] = {
 264         "print-parent",
 265         "sym-offset",
 266         "sym-addr",
 267         "verbose",
 268         "raw",
 269         "hex",
 270         "bin",
 271         "block",
 272         "stacktrace",
 273         "sched-tree",
 274         "ftrace_printk",
 275         "ftrace_preempt",
 276         "branch",
 277         "annotate",
 278         NULL
 279 };
 280
 281 /*
 282  * ftrace_max_lock is used to protect the swapping of buffers
 283  * when taking a max snapshot. The buffers themselves are
 284  * protected by per_cpu spinlocks. But the action of the swap
 285  * needs its own lock.
 286  *
 287  * This is defined as a raw_spinlock_t in order to help
 288  * with performance when lockdep debugging is enabled.
 289  */
 290 static raw_spinlock_t ftrace_max_lock =
 291         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 292
 293 /*
 294  * Copy the new maximum trace into the separate maximum-trace
 295  * structure. (this way the maximum trace is permanently saved,
 296  * for later retrieval via /debugfs/tracing/latency_trace)
 297  */
 298 static void
 299 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 300 {
 301         struct trace_array_cpu *data = tr->data[cpu];
 302
 303         max_tr.cpu = cpu;
 304         max_tr.time_start = data->preempt_timestamp;
 305
 306         data = max_tr.data[cpu];
 307         data->saved_latency = tracing_max_latency;
 308
 309         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 310         data->pid = tsk->pid;
 311         data->uid = tsk->uid;
 312         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 313         data->policy = tsk->policy;
 314         data->rt_priority = tsk->rt_priority;
 315
 316         /* record this tasks comm */
 317         tracing_record_cmdline(current);
 318 }
 319
 320 /**
 321  * trace_seq_printf - sequence printing of trace information
 322  * @s: trace sequence descriptor
 323  * @fmt: printf format string
 324  *
 325  * The tracer may use either sequence operations or its own
 326  * copy to user routines. To simplify formating of a trace
 327  * trace_seq_printf is used to store strings into a special
 328  * buffer (@s). Then the output may be either used by
 329  * the sequencer or pulled into another buffer.
 330  */
 331 int
 332 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 333 {
 334         int len = (PAGE_SIZE - 1) - s->len;
 335         va_list ap;
 336         int ret;
 337
 338         if (!len)
 339                 return 0;
 340
 341         va_start(ap, fmt);
 342         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 343         va_end(ap);
 344
 345         /* If we can't write it all, don't bother writing anything */
 346         if (ret >= len)
 347                 return 0;
 348
 349         s->len += ret;
 350
 351         return len;
 352 }
 353
 354 /**
 355  * trace_seq_puts - trace sequence printing of simple string
 356  * @s: trace sequence descriptor
 357  * @str: simple string to record
 358  *
 359  * The tracer may use either the sequence operations or its own
 360  * copy to user routines. This function records a simple string
 361  * into a special buffer (@s) for later retrieval by a sequencer
 362  * or other mechanism.
 363  */
 364 static int
 365 trace_seq_puts(struct trace_seq *s, const char *str)
 366 {
 367         int len = strlen(str);
 368
 369         if (len > ((PAGE_SIZE - 1) - s->len))
 370                 return 0;
 371
 372         memcpy(s->buffer + s->len, str, len);
 373         s->len += len;
 374
 375         return len;
 376 }
 377
 378 static int
 379 trace_seq_putc(struct trace_seq *s, unsigned char c)
 380 {
 381         if (s->len >= (PAGE_SIZE - 1))
 382                 return 0;
 383
 384         s->buffer[s->len++] = c;
 385
 386         return 1;
 387 }
 388
 389 static int
 390 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 391 {
 392         if (len > ((PAGE_SIZE - 1) - s->len))
 393                 return 0;
 394
 395         memcpy(s->buffer + s->len, mem, len);
 396         s->len += len;
 397
 398         return len;
 399 }
 400
 401 #define MAX_MEMHEX_BYTES        8
 402 #define HEX_CHARS               (MAX_MEMHEX_BYTES*2 + 1)
 403
 404 static int
 405 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 406 {
 407         unsigned char hex[HEX_CHARS];
 408         unsigned char *data = mem;
 409         int i, j;
 410
 411 #ifdef __BIG_ENDIAN
 412         for (i = 0, j = 0; i < len; i++) {
 413 #else
 414         for (i = len-1, j = 0; i >= 0; i--) {
 415 #endif
 416                 hex[j++] = hex_asc_hi(data[i]);
 417                 hex[j++] = hex_asc_lo(data[i]);
 418         }
 419         hex[j++] = ' ';
 420
 421         return trace_seq_putmem(s, hex, j);
 422 }
 423
 424 static void
 425 trace_seq_reset(struct trace_seq *s)
 426 {
 427         s->len = 0;
 428         s->readpos = 0;
 429 }
 430
 431 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
 432 {
 433         int len;
 434         int ret;
 435
 436         if (s->len <= s->readpos)
 437                 return -EBUSY;
 438
 439         len = s->len - s->readpos;
 440         if (cnt > len)
 441                 cnt = len;
 442         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
 443         if (ret)
 444                 return -EFAULT;
 445
 446         s->readpos += len;
 447         return cnt;
 448 }
 449
 450 static void
 451 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 452 {
 453         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 454
 455         s->buffer[len] = 0;
 456         seq_puts(m, s->buffer);
 457
 458         trace_seq_reset(s);
 459 }
 460
 461 /**
 462  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
 463  * @tr: tracer
 464  * @tsk: the task with the latency
 465  * @cpu: The cpu that initiated the trace.
 466  *
 467  * Flip the buffers between the @tr and the max_tr and record information
 468  * about which task was the cause of this latency.
 469  */
 470 void
 471 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 472 {
 473         struct ring_buffer *buf = tr->buffer;
 474
 475         WARN_ON_ONCE(!irqs_disabled());
 476         __raw_spin_lock(&ftrace_max_lock);
 477
 478         tr->buffer = max_tr.buffer;
 479         max_tr.buffer = buf;
 480
 481         ftrace_disable_cpu();
 482         ring_buffer_reset(tr->buffer);
 483         ftrace_enable_cpu();
 484
 485         __update_max_tr(tr, tsk, cpu);
 486         __raw_spin_unlock(&ftrace_max_lock);
 487 }
 488
 489 /**
 490  * update_max_tr_single - only copy one trace over, and reset the rest
 491  * @tr - tracer
 492  * @tsk - task with the latency
 493  * @cpu - the cpu of the buffer to copy.
 494  *
 495  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
 496  */
 497 void
 498 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 499 {
 500         int ret;
 501
 502         WARN_ON_ONCE(!irqs_disabled());
 503         __raw_spin_lock(&ftrace_max_lock);
 504
 505         ftrace_disable_cpu();
 506
 507         ring_buffer_reset(max_tr.buffer);
 508         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
 509
 510         ftrace_enable_cpu();
 511
 512         WARN_ON_ONCE(ret);
 513
 514         __update_max_tr(tr, tsk, cpu);
 515         __raw_spin_unlock(&ftrace_max_lock);
 516 }
 517
 518 /**
 519  * register_tracer - register a tracer with the ftrace system.
 520  * @type - the plugin for the tracer
 521  *
 522  * Register a new plugin tracer.
 523  */
 524 int register_tracer(struct tracer *type)
 525 {
 526         struct tracer *t;
 527         int len;
 528         int ret = 0;
 529
 530         if (!type->name) {
 531                 pr_info("Tracer must have a name\n");
 532                 return -1;
 533         }
 534
 535         /*
 536          * When this gets called we hold the BKL which means that
 537          * preemption is disabled. Various trace selftests however
 538          * need to disable and enable preemption for successful tests.
 539          * So we drop the BKL here and grab it after the tests again.
 540          */
 541         unlock_kernel();
 542         mutex_lock(&trace_types_lock);
 543
 544         for (t = trace_types; t; t = t->next) {
 545                 if (strcmp(type->name, t->name) == 0) {
 546                         /* already found */
 547                         pr_info("Trace %s already registered\n",
 548                                 type->name);
 549                         ret = -1;
 550                         goto out;
 551                 }
 552         }
 553
 554         if (!type->set_flag)
 555                 type->set_flag = &dummy_set_flag;
 556         if (!type->flags)
 557                 type->flags = &dummy_tracer_flags;
 558         else
 559                 if (!type->flags->opts)
 560                         type->flags->opts = dummy_tracer_opt;
 561
 562 #ifdef CONFIG_FTRACE_STARTUP_TEST
 563         if (type->selftest) {
 564                 struct tracer *saved_tracer = current_trace;
 565                 struct trace_array *tr = &global_trace;
 566                 int i;
 567                 /*
 568                  * Run a selftest on this tracer.
 569                  * Here we reset the trace buffer, and set the current
 570                  * tracer to be this tracer. The tracer can then run some
 571                  * internal tracing to verify that everything is in order.
 572                  * If we fail, we do not register this tracer.
 573                  */
 574                 for_each_tracing_cpu(i)
 575                         tracing_reset(tr, i);
 576
 577                 current_trace = type;
 578                 /* the test is responsible for initializing and enabling */
 579                 pr_info("Testing tracer %s: ", type->name);
 580                 ret = type->selftest(type, tr);
 581                 /* the test is responsible for resetting too */
 582                 current_trace = saved_tracer;
 583                 if (ret) {
 584                         printk(KERN_CONT "FAILED!\n");
 585                         goto out;
 586                 }
 587                 /* Only reset on passing, to avoid touching corrupted buffers */
 588                 for_each_tracing_cpu(i)
 589                         tracing_reset(tr, i);
 590
 591                 printk(KERN_CONT "PASSED\n");
 592         }
 593 #endif
 594
 595         type->next = trace_types;
 596         trace_types = type;
 597         len = strlen(type->name);
 598         if (len > max_tracer_type_len)
 599                 max_tracer_type_len = len;
 600
 601  out:
 602         mutex_unlock(&trace_types_lock);
 603         lock_kernel();
 604
 605         return ret;
 606 }
 607
 608 void unregister_tracer(struct tracer *type)
 609 {
 610         struct tracer **t;
 611         int len;
 612
 613         mutex_lock(&trace_types_lock);
 614         for (t = &trace_types; *t; t = &(*t)->next) {
 615                 if (*t == type)
 616                         goto found;
 617         }
 618         pr_info("Trace %s not registered\n", type->name);
 619         goto out;
 620
 621  found:
 622         *t = (*t)->next;
 623         if (strlen(type->name) != max_tracer_type_len)
 624                 goto out;
 625
 626         max_tracer_type_len = 0;
 627         for (t = &trace_types; *t; t = &(*t)->next) {
 628                 len = strlen((*t)->name);
 629                 if (len > max_tracer_type_len)
 630                         max_tracer_type_len = len;
 631         }
 632  out:
 633         mutex_unlock(&trace_types_lock);
 634 }
 635
 636 void tracing_reset(struct trace_array *tr, int cpu)
 637 {
 638         ftrace_disable_cpu();
 639         ring_buffer_reset_cpu(tr->buffer, cpu);
 640         ftrace_enable_cpu();
 641 }
 642
 643 #define SAVED_CMDLINES 128
 644 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 645 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 646 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 647 static int cmdline_idx;
 648 static DEFINE_SPINLOCK(trace_cmdline_lock);
 649
 650 /* temporary disable recording */
 651 atomic_t trace_record_cmdline_disabled __read_mostly;
 652
 653 static void trace_init_cmdlines(void)
 654 {
 655         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 656         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 657         cmdline_idx = 0;
 658 }
 659
 660 static int trace_stop_count;
 661 static DEFINE_SPINLOCK(tracing_start_lock);
 662
 663 /**
 664  * ftrace_off_permanent - disable all ftrace code permanently
 665  *
 666  * This should only be called when a serious anomally has
 667  * been detected.  This will turn off the function tracing,
 668  * ring buffers, and other tracing utilites. It takes no
 669  * locks and can be called from any context.
 670  */
 671 void ftrace_off_permanent(void)
 672 {
 673         tracing_disabled = 1;
 674         ftrace_stop();
 675         tracing_off_permanent();
 676 }
 677
 678 /**
 679  * tracing_start - quick start of the tracer
 680  *
 681  * If tracing is enabled but was stopped by tracing_stop,
 682  * this will start the tracer back up.
 683  */
 684 void tracing_start(void)
 685 {
 686         struct ring_buffer *buffer;
 687         unsigned long flags;
 688
 689         if (tracing_disabled)
 690                 return;
 691
 692         spin_lock_irqsave(&tracing_start_lock, flags);
 693         if (--trace_stop_count)
 694                 goto out;
 695
 696         if (trace_stop_count < 0) {
 697                 /* Someone screwed up their debugging */
 698                 WARN_ON_ONCE(1);
 699                 trace_stop_count = 0;
 700                 goto out;
 701         }
 702
 703
 704         buffer = global_trace.buffer;
 705         if (buffer)
 706                 ring_buffer_record_enable(buffer);
 707
 708         buffer = max_tr.buffer;
 709         if (buffer)
 710                 ring_buffer_record_enable(buffer);
 711
 712         ftrace_start();
 713  out:
 714         spin_unlock_irqrestore(&tracing_start_lock, flags);
 715 }
 716
 717 /**
 718  * tracing_stop - quick stop of the tracer
 719  *
 720  * Light weight way to stop tracing. Use in conjunction with
 721  * tracing_start.
 722  */
 723 void tracing_stop(void)
 724 {
 725         struct ring_buffer *buffer;
 726         unsigned long flags;
 727
 728         ftrace_stop();
 729         spin_lock_irqsave(&tracing_start_lock, flags);
 730         if (trace_stop_count++)
 731                 goto out;
 732
 733         buffer = global_trace.buffer;
 734         if (buffer)
 735                 ring_buffer_record_disable(buffer);
 736
 737         buffer = max_tr.buffer;
 738         if (buffer)
 739                 ring_buffer_record_disable(buffer);
 740
 741  out:
 742         spin_unlock_irqrestore(&tracing_start_lock, flags);
 743 }
 744
 745 void trace_stop_cmdline_recording(void);
 746
 747 static void trace_save_cmdline(struct task_struct *tsk)
 748 {
 749         unsigned map;
 750         unsigned idx;
 751
 752         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 753                 return;
 754
 755         /*
 756          * It's not the end of the world if we don't get
 757          * the lock, but we also don't want to spin
 758          * nor do we want to disable interrupts,
 759          * so if we miss here, then better luck next time.
 760          */
 761         if (!spin_trylock(&trace_cmdline_lock))
 762                 return;
 763
 764         idx = map_pid_to_cmdline[tsk->pid];
 765         if (idx >= SAVED_CMDLINES) {
 766                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 767
 768                 map = map_cmdline_to_pid[idx];
 769                 if (map <= PID_MAX_DEFAULT)
 770                         map_pid_to_cmdline[map] = (unsigned)-1;
 771
 772                 map_pid_to_cmdline[tsk->pid] = idx;
 773
 774                 cmdline_idx = idx;
 775         }
 776
 777         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 778
 779         spin_unlock(&trace_cmdline_lock);
 780 }
 781
 782 static char *trace_find_cmdline(int pid)
 783 {
 784         char *cmdline = "<...>";
 785         unsigned map;
 786
 787         if (!pid)
 788                 return "<idle>";
 789
 790         if (pid > PID_MAX_DEFAULT)
 791                 goto out;
 792
 793         map = map_pid_to_cmdline[pid];
 794         if (map >= SAVED_CMDLINES)
 795                 goto out;
 796
 797         cmdline = saved_cmdlines[map];
 798
 799  out:
 800         return cmdline;
 801 }
 802
 803 void tracing_record_cmdline(struct task_struct *tsk)
 804 {
 805         if (atomic_read(&trace_record_cmdline_disabled))
 806                 return;
 807
 808         trace_save_cmdline(tsk);
 809 }
 810
 811 void
 812 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
 813                              int pc)
 814 {
 815         struct task_struct *tsk = current;
 816
 817         entry->preempt_count            = pc & 0xff;
 818         entry->pid                      = (tsk) ? tsk->pid : 0;
 819         entry->flags =
 820 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
 821                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 822 #else
 823                 TRACE_FLAG_IRQS_NOSUPPORT |
 824 #endif
 825                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 826                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 827                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 828 }
 829
 830 void
 831 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 832                unsigned long ip, unsigned long parent_ip, unsigned long flags,
 833                int pc)
 834 {
 835         struct ring_buffer_event *event;
 836         struct ftrace_entry *entry;
 837         unsigned long irq_flags;
 838
 839         /* If we are reading the ring buffer, don't trace */
 840         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 841                 return;
 842
 843         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 844                                          &irq_flags);
 845         if (!event)
 846                 return;
 847         entry   = ring_buffer_event_data(event);
 848         tracing_generic_entry_update(&entry->ent, flags, pc);
 849         entry->ent.type                 = TRACE_FN;
 850         entry->ip                       = ip;
 851         entry->parent_ip                = parent_ip;
 852         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 853 }
 854
 855 #ifdef CONFIG_FUNCTION_RET_TRACER
 856 static void __trace_function_return(struct trace_array *tr,
 857                                 struct trace_array_cpu *data,
 858                                 struct ftrace_retfunc *trace,
 859                                 unsigned long flags,
 860                                 int pc)
 861 {
 862         struct ring_buffer_event *event;
 863         struct ftrace_ret_entry *entry;
 864         unsigned long irq_flags;
 865
 866         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 867                 return;
 868
 869         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
 870                                          &irq_flags);
 871         if (!event)
 872                 return;
 873         entry   = ring_buffer_event_data(event);
 874         tracing_generic_entry_update(&entry->ent, flags, pc);
 875         entry->ent.type                 = TRACE_FN_RET;
 876         entry->ip                       = trace->func;
 877         entry->parent_ip        = trace->ret;
 878         entry->rettime          = trace->rettime;
 879         entry->calltime         = trace->calltime;
 880         entry->overrun          = trace->overrun;
 881         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
 882 }
 883 #endif
 884
 885 void
 886 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 887        unsigned long ip, unsigned long parent_ip, unsigned long flags,
 888        int pc)
 889 {
 890         if (likely(!atomic_read(&data->disabled)))
 891                 trace_function(tr, data, ip, parent_ip, flags, pc);
 892 }
 893
 894 static void ftrace_trace_stack(struct trace_array *tr,
 895                                struct trace_array_cpu *data,
 896                                unsigned long flags,
 897                                int skip, int pc)
 898 {
 899 #ifdef CONFIG_STACKTRACE
 900         struct ring_buffer_event *event;
 901         struct stack_entry *entry;
 902         struct stack_trace trace;
 903         unsigned long irq_flags;
 904
 905         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 906                 return;
 907
 908         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 909                                          &irq_flags);
 910         if (!event)
 911                 return;
 912         entry   = ring_buffer_event_data(event);
 913         tracing_generic_entry_update(&entry->ent, flags, pc);
 914         entry->ent.type         = TRACE_STACK;
 915
 916         memset(&entry->caller, 0, sizeof(entry->caller));
 917
 918         trace.nr_entries        = 0;
 919         trace.max_entries       = FTRACE_STACK_ENTRIES;
 920         trace.skip              = skip;
 921         trace.entries           = entry->caller;
 922
 923         save_stack_trace(&trace);
 924         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 925 #endif
 926 }
 927
 928 void __trace_stack(struct trace_array *tr,
 929                    struct trace_array_cpu *data,
 930                    unsigned long flags,
 931                    int skip)
 932 {
 933         ftrace_trace_stack(tr, data, flags, skip, preempt_count());
 934 }
 935
 936 static void
 937 ftrace_trace_special(void *__tr, void *__data,
 938                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
 939                      int pc)
 940 {
 941         struct ring_buffer_event *event;
 942         struct trace_array_cpu *data = __data;
 943         struct trace_array *tr = __tr;
 944         struct special_entry *entry;
 945         unsigned long irq_flags;
 946
 947         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 948                                          &irq_flags);
 949         if (!event)
 950                 return;
 951         entry   = ring_buffer_event_data(event);
 952         tracing_generic_entry_update(&entry->ent, 0, pc);
 953         entry->ent.type                 = TRACE_SPECIAL;
 954         entry->arg1                     = arg1;
 955         entry->arg2                     = arg2;
 956         entry->arg3                     = arg3;
 957         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 958         ftrace_trace_stack(tr, data, irq_flags, 4, pc);
 959
 960         trace_wake_up();
 961 }
 962
 963 void
 964 __trace_special(void *__tr, void *__data,
 965                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 966 {
 967         ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
 968 }
 969
 970 void
 971 tracing_sched_switch_trace(struct trace_array *tr,
 972                            struct trace_array_cpu *data,
 973                            struct task_struct *prev,
 974                            struct task_struct *next,
 975                            unsigned long flags, int pc)
 976 {
 977         struct ring_buffer_event *event;
 978         struct ctx_switch_entry *entry;
 979         unsigned long irq_flags;
 980
 981         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 982                                            &irq_flags);
 983         if (!event)
 984                 return;
 985         entry   = ring_buffer_event_data(event);
 986         tracing_generic_entry_update(&entry->ent, flags, pc);
 987         entry->ent.type                 = TRACE_CTX;
 988         entry->prev_pid                 = prev->pid;
 989         entry->prev_prio                = prev->prio;
 990         entry->prev_state               = prev->state;
 991         entry->next_pid                 = next->pid;
 992         entry->next_prio                = next->prio;
 993         entry->next_state               = next->state;
 994         entry->next_cpu = task_cpu(next);
 995         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 996         ftrace_trace_stack(tr, data, flags, 5, pc);
 997 }
 998
 999 void
1000 tracing_sched_wakeup_trace(struct trace_array *tr,
1001                            struct trace_array_cpu *data,
1002                            struct task_struct *wakee,
1003                            struct task_struct *curr,
1004                            unsigned long flags, int pc)
1005 {
1006         struct ring_buffer_event *event;
1007         struct ctx_switch_entry *entry;
1008         unsigned long irq_flags;
1009
1010         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1011                                            &irq_flags);
1012         if (!event)
1013                 return;
1014         entry   = ring_buffer_event_data(event);
1015         tracing_generic_entry_update(&entry->ent, flags, pc);
1016         entry->ent.type                 = TRACE_WAKE;
1017         entry->prev_pid                 = curr->pid;
1018         entry->prev_prio                = curr->prio;
1019         entry->prev_state               = curr->state;
1020         entry->next_pid                 = wakee->pid;
1021         entry->next_prio                = wakee->prio;
1022         entry->next_state               = wakee->state;
1023         entry->next_cpu                 = task_cpu(wakee);
1024         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1025         ftrace_trace_stack(tr, data, flags, 6, pc);
1026
1027         trace_wake_up();
1028 }
1029
1030 void
1031 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1032 {
1033         struct trace_array *tr = &global_trace;
1034         struct trace_array_cpu *data;
1035         unsigned long flags;
1036         int cpu;
1037         int pc;
1038
1039         if (tracing_disabled)
1040                 return;
1041
1042         pc = preempt_count();
1043         local_irq_save(flags);
1044         cpu = raw_smp_processor_id();
1045         data = tr->data[cpu];
1046
1047         if (likely(atomic_inc_return(&data->disabled) == 1))
1048                 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1049
1050         atomic_dec(&data->disabled);
1051         local_irq_restore(flags);
1052 }
1053
1054 #ifdef CONFIG_FUNCTION_TRACER
1055 static void
1056 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1057 {
1058         struct trace_array *tr = &global_trace;
1059         struct trace_array_cpu *data;
1060         unsigned long flags;
1061         long disabled;
1062         int cpu, resched;
1063         int pc;
1064
1065         if (unlikely(!ftrace_function_enabled))
1066                 return;
1067
1068         pc = preempt_count();
1069         resched = ftrace_preempt_disable();
1070         local_save_flags(flags);
1071         cpu = raw_smp_processor_id();
1072         data = tr->data[cpu];
1073         disabled = atomic_inc_return(&data->disabled);
1074
1075         if (likely(disabled == 1))
1076                 trace_function(tr, data, ip, parent_ip, flags, pc);
1077
1078         atomic_dec(&data->disabled);
1079         ftrace_preempt_enable(resched);
1080 }
1081
1082 static void
1083 function_trace_call(unsigned long ip, unsigned long parent_ip)
1084 {
1085         struct trace_array *tr = &global_trace;
1086         struct trace_array_cpu *data;
1087         unsigned long flags;
1088         long disabled;
1089         int cpu;
1090         int pc;
1091
1092         if (unlikely(!ftrace_function_enabled))
1093                 return;
1094
1095         /*
1096          * Need to use raw, since this must be called before the
1097          * recursive protection is performed.
1098          */
1099         local_irq_save(flags);
1100         cpu = raw_smp_processor_id();
1101         data = tr->data[cpu];
1102         disabled = atomic_inc_return(&data->disabled);
1103
1104         if (likely(disabled == 1)) {
1105                 pc = preempt_count();
1106                 trace_function(tr, data, ip, parent_ip, flags, pc);
1107         }
1108
1109         atomic_dec(&data->disabled);
1110         local_irq_restore(flags);
1111 }
1112
1113 #ifdef CONFIG_FUNCTION_RET_TRACER
1114 void trace_function_return(struct ftrace_retfunc *trace)
1115 {
1116         struct trace_array *tr = &global_trace;
1117         struct trace_array_cpu *data;
1118         unsigned long flags;
1119         long disabled;
1120         int cpu;
1121         int pc;
1122
1123         raw_local_irq_save(flags);
1124         cpu = raw_smp_processor_id();
1125         data = tr->data[cpu];
1126         disabled = atomic_inc_return(&data->disabled);
1127         if (likely(disabled == 1)) {
1128                 pc = preempt_count();
1129                 __trace_function_return(tr, data, trace, flags, pc);
1130         }
1131         atomic_dec(&data->disabled);
1132         raw_local_irq_restore(flags);
1133 }
1134 #endif /* CONFIG_FUNCTION_RET_TRACER */
1135
1136 static struct ftrace_ops trace_ops __read_mostly =
1137 {
1138         .func = function_trace_call,
1139 };
1140
1141 void tracing_start_function_trace(void)
1142 {
1143         ftrace_function_enabled = 0;
1144
1145         if (trace_flags & TRACE_ITER_PREEMPTONLY)
1146                 trace_ops.func = function_trace_call_preempt_only;
1147         else
1148                 trace_ops.func = function_trace_call;
1149
1150         register_ftrace_function(&trace_ops);
1151         ftrace_function_enabled = 1;
1152 }
1153
1154 void tracing_stop_function_trace(void)
1155 {
1156         ftrace_function_enabled = 0;
1157         unregister_ftrace_function(&trace_ops);
1158 }
1159 #endif
1160
1161 enum trace_file_type {
1162         TRACE_FILE_LAT_FMT      = 1,
1163         TRACE_FILE_ANNOTATE     = 2,
1164 };
1165
1166 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1167 {
1168         /* Don't allow ftrace to trace into the ring buffers */
1169         ftrace_disable_cpu();
1170
1171         iter->idx++;
1172         if (iter->buffer_iter[iter->cpu])
1173                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1174
1175         ftrace_enable_cpu();
1176 }
1177
1178 static struct trace_entry *
1179 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1180 {
1181         struct ring_buffer_event *event;
1182         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1183
1184         /* Don't allow ftrace to trace into the ring buffers */
1185         ftrace_disable_cpu();
1186
1187         if (buf_iter)
1188                 event = ring_buffer_iter_peek(buf_iter, ts);
1189         else
1190                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1191
1192         ftrace_enable_cpu();
1193
1194         return event ? ring_buffer_event_data(event) : NULL;
1195 }
1196
1197 static struct trace_entry *
1198 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1199 {
1200         struct ring_buffer *buffer = iter->tr->buffer;
1201         struct trace_entry *ent, *next = NULL;
1202         u64 next_ts = 0, ts;
1203         int next_cpu = -1;
1204         int cpu;
1205
1206         for_each_tracing_cpu(cpu) {
1207
1208                 if (ring_buffer_empty_cpu(buffer, cpu))
1209                         continue;
1210
1211                 ent = peek_next_entry(iter, cpu, &ts);
1212
1213                 /*
1214                  * Pick the entry with the smallest timestamp:
1215                  */
1216                 if (ent && (!next || ts < next_ts)) {
1217                         next = ent;
1218                         next_cpu = cpu;
1219                         next_ts = ts;
1220                 }
1221         }
1222
1223         if (ent_cpu)
1224                 *ent_cpu = next_cpu;
1225
1226         if (ent_ts)
1227                 *ent_ts = next_ts;
1228
1229         return next;
1230 }
1231
1232 /* Find the next real entry, without updating the iterator itself */
1233 static struct trace_entry *
1234 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1235 {
1236         return __find_next_entry(iter, ent_cpu, ent_ts);
1237 }
1238
1239 /* Find the next real entry, and increment the iterator to the next entry */
1240 static void *find_next_entry_inc(struct trace_iterator *iter)
1241 {
1242         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1243
1244         if (iter->ent)
1245                 trace_iterator_increment(iter, iter->cpu);
1246
1247         return iter->ent ? iter : NULL;
1248 }
1249
1250 static void trace_consume(struct trace_iterator *iter)
1251 {
1252         /* Don't allow ftrace to trace into the ring buffers */
1253         ftrace_disable_cpu();
1254         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1255         ftrace_enable_cpu();
1256 }
1257
1258 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1259 {
1260         struct trace_iterator *iter = m->private;
1261         int i = (int)*pos;
1262         void *ent;
1263
1264         (*pos)++;
1265
1266         /* can't go backwards */
1267         if (iter->idx > i)
1268                 return NULL;
1269
1270         if (iter->idx < 0)
1271                 ent = find_next_entry_inc(iter);
1272         else
1273                 ent = iter;
1274
1275         while (ent && iter->idx < i)
1276                 ent = find_next_entry_inc(iter);
1277
1278         iter->pos = *pos;
1279
1280         return ent;
1281 }
1282
1283 static void *s_start(struct seq_file *m, loff_t *pos)
1284 {
1285         struct trace_iterator *iter = m->private;
1286         void *p = NULL;
1287         loff_t l = 0;
1288         int cpu;
1289
1290         mutex_lock(&trace_types_lock);
1291
1292         if (!current_trace || current_trace != iter->trace) {
1293                 mutex_unlock(&trace_types_lock);
1294                 return NULL;
1295         }
1296
1297         atomic_inc(&trace_record_cmdline_disabled);
1298
1299         if (*pos != iter->pos) {
1300                 iter->ent = NULL;
1301                 iter->cpu = 0;
1302                 iter->idx = -1;
1303
1304                 ftrace_disable_cpu();
1305
1306                 for_each_tracing_cpu(cpu) {
1307                         ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1308                 }
1309
1310                 ftrace_enable_cpu();
1311
1312                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1313                         ;
1314
1315         } else {
1316                 l = *pos - 1;
1317                 p = s_next(m, p, &l);
1318         }
1319
1320         return p;
1321 }
1322
1323 static void s_stop(struct seq_file *m, void *p)
1324 {
1325         atomic_dec(&trace_record_cmdline_disabled);
1326         mutex_unlock(&trace_types_lock);
1327 }
1328
1329 #ifdef CONFIG_KRETPROBES
1330 static inline const char *kretprobed(const char *name)
1331 {
1332         static const char tramp_name[] = "kretprobe_trampoline";
1333         int size = sizeof(tramp_name);
1334
1335         if (strncmp(tramp_name, name, size) == 0)
1336                 return "[unknown/kretprobe'd]";
1337         return name;
1338 }
1339 #else
1340 static inline const char *kretprobed(const char *name)
1341 {
1342         return name;
1343 }
1344 #endif /* CONFIG_KRETPROBES */
1345
1346 static int
1347 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1348 {
1349 #ifdef CONFIG_KALLSYMS
1350         char str[KSYM_SYMBOL_LEN];
1351         const char *name;
1352
1353         kallsyms_lookup(address, NULL, NULL, NULL, str);
1354
1355         name = kretprobed(str);
1356
1357         return trace_seq_printf(s, fmt, name);
1358 #endif
1359         return 1;
1360 }
1361
1362 static int
1363 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1364                      unsigned long address)
1365 {
1366 #ifdef CONFIG_KALLSYMS
1367         char str[KSYM_SYMBOL_LEN];
1368         const char *name;
1369
1370         sprint_symbol(str, address);
1371         name = kretprobed(str);
1372
1373         return trace_seq_printf(s, fmt, name);
1374 #endif
1375         return 1;
1376 }
1377
1378 #ifndef CONFIG_64BIT
1379 # define IP_FMT "%08lx"
1380 #else
1381 # define IP_FMT "%016lx"
1382 #endif
1383
1384 int
1385 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1386 {
1387         int ret;
1388
1389         if (!ip)
1390                 return trace_seq_printf(s, "0");
1391
1392         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1393                 ret = seq_print_sym_offset(s, "%s", ip);
1394         else
1395                 ret = seq_print_sym_short(s, "%s", ip);
1396
1397         if (!ret)
1398                 return 0;
1399
1400         if (sym_flags & TRACE_ITER_SYM_ADDR)
1401                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1402         return ret;
1403 }
1404
1405 static void print_lat_help_header(struct seq_file *m)
1406 {
1407         seq_puts(m, "#                  _------=> CPU#            \n");
1408         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1409         seq_puts(m, "#                | / _----=> need-resched    \n");
1410         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1411         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1412         seq_puts(m, "#                |||| /                      \n");
1413         seq_puts(m, "#                |||||     delay             \n");
1414         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1415         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1416 }
1417
1418 static void print_func_help_header(struct seq_file *m)
1419 {
1420         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1421         seq_puts(m, "#              | |       |          |         |\n");
1422 }
1423
1424
1425 static void
1426 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1427 {
1428         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1429         struct trace_array *tr = iter->tr;
1430         struct trace_array_cpu *data = tr->data[tr->cpu];
1431         struct tracer *type = current_trace;
1432         unsigned long total;
1433         unsigned long entries;
1434         const char *name = "preemption";
1435
1436         if (type)
1437                 name = type->name;
1438
1439         entries = ring_buffer_entries(iter->tr->buffer);
1440         total = entries +
1441                 ring_buffer_overruns(iter->tr->buffer);
1442
1443         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1444                    name, UTS_RELEASE);
1445         seq_puts(m, "-----------------------------------"
1446                  "---------------------------------\n");
1447         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1448                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1449                    nsecs_to_usecs(data->saved_latency),
1450                    entries,
1451                    total,
1452                    tr->cpu,
1453 #if defined(CONFIG_PREEMPT_NONE)
1454                    "server",
1455 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1456                    "desktop",
1457 #elif defined(CONFIG_PREEMPT)
1458                    "preempt",
1459 #else
1460                    "unknown",
1461 #endif
1462                    /* These are reserved for later use */
1463                    0, 0, 0, 0);
1464 #ifdef CONFIG_SMP
1465         seq_printf(m, " #P:%d)\n", num_online_cpus());
1466 #else
1467         seq_puts(m, ")\n");
1468 #endif
1469         seq_puts(m, "    -----------------\n");
1470         seq_printf(m, "    | task: %.16s-%d "
1471                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1472                    data->comm, data->pid, data->uid, data->nice,
1473                    data->policy, data->rt_priority);
1474         seq_puts(m, "    -----------------\n");
1475
1476         if (data->critical_start) {
1477                 seq_puts(m, " => started at: ");
1478                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1479                 trace_print_seq(m, &iter->seq);
1480                 seq_puts(m, "\n => ended at:   ");
1481                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1482                 trace_print_seq(m, &iter->seq);
1483                 seq_puts(m, "\n");
1484         }
1485
1486         seq_puts(m, "\n");
1487 }
1488
1489 static void
1490 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1491 {
1492         int hardirq, softirq;
1493         char *comm;
1494
1495         comm = trace_find_cmdline(entry->pid);
1496
1497         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1498         trace_seq_printf(s, "%3d", cpu);
1499         trace_seq_printf(s, "%c%c",
1500                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1501                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1502                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1503
1504         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1505         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1506         if (hardirq && softirq) {
1507                 trace_seq_putc(s, 'H');
1508         } else {
1509                 if (hardirq) {
1510                         trace_seq_putc(s, 'h');
1511                 } else {
1512                         if (softirq)
1513                                 trace_seq_putc(s, 's');
1514                         else
1515                                 trace_seq_putc(s, '.');
1516                 }
1517         }
1518
1519         if (entry->preempt_count)
1520                 trace_seq_printf(s, "%x", entry->preempt_count);
1521         else
1522                 trace_seq_puts(s, ".");
1523 }
1524
1525 unsigned long preempt_mark_thresh = 100;
1526
1527 static void
1528 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1529                     unsigned long rel_usecs)
1530 {
1531         trace_seq_printf(s, " %4lldus", abs_usecs);
1532         if (rel_usecs > preempt_mark_thresh)
1533                 trace_seq_puts(s, "!: ");
1534         else if (rel_usecs > 1)
1535                 trace_seq_puts(s, "+: ");
1536         else
1537                 trace_seq_puts(s, " : ");
1538 }
1539
1540 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1541
1542 /*
1543  * The message is supposed to contain an ending newline.
1544  * If the printing stops prematurely, try to add a newline of our own.
1545  */
1546 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1547 {
1548         struct trace_entry *ent;
1549         struct trace_field_cont *cont;
1550         bool ok = true;
1551
1552         ent = peek_next_entry(iter, iter->cpu, NULL);
1553         if (!ent || ent->type != TRACE_CONT) {
1554                 trace_seq_putc(s, '\n');
1555                 return;
1556         }
1557
1558         do {
1559                 cont = (struct trace_field_cont *)ent;
1560                 if (ok)
1561                         ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1562
1563                 ftrace_disable_cpu();
1564
1565                 if (iter->buffer_iter[iter->cpu])
1566                         ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1567                 else
1568                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1569
1570                 ftrace_enable_cpu();
1571
1572                 ent = peek_next_entry(iter, iter->cpu, NULL);
1573         } while (ent && ent->type == TRACE_CONT);
1574
1575         if (!ok)
1576                 trace_seq_putc(s, '\n');
1577 }
1578
1579 static void test_cpu_buff_start(struct trace_iterator *iter)
1580 {
1581         struct trace_seq *s = &iter->seq;
1582
1583         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1584                 return;
1585
1586         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1587                 return;
1588
1589         if (cpu_isset(iter->cpu, iter->started))
1590                 return;
1591
1592         cpu_set(iter->cpu, iter->started);
1593         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1594 }
1595
1596 static enum print_line_t
1597 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1598 {
1599         struct trace_seq *s = &iter->seq;
1600         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1601         struct trace_entry *next_entry;
1602         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1603         struct trace_entry *entry = iter->ent;
1604         unsigned long abs_usecs;
1605         unsigned long rel_usecs;
1606         u64 next_ts;
1607         char *comm;
1608         int S, T;
1609         int i;
1610         unsigned state;
1611
1612         if (entry->type == TRACE_CONT)
1613                 return TRACE_TYPE_HANDLED;
1614
1615         test_cpu_buff_start(iter);
1616
1617         next_entry = find_next_entry(iter, NULL, &next_ts);
1618         if (!next_entry)
1619                 next_ts = iter->ts;
1620         rel_usecs = ns2usecs(next_ts - iter->ts);
1621         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1622
1623         if (verbose) {
1624                 comm = trace_find_cmdline(entry->pid);
1625                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1626                                  " %ld.%03ldms (+%ld.%03ldms): ",
1627                                  comm,
1628                                  entry->pid, cpu, entry->flags,
1629                                  entry->preempt_count, trace_idx,
1630                                  ns2usecs(iter->ts),
1631                                  abs_usecs/1000,
1632                                  abs_usecs % 1000, rel_usecs/1000,
1633                                  rel_usecs % 1000);
1634         } else {
1635                 lat_print_generic(s, entry, cpu);
1636                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1637         }
1638         switch (entry->type) {
1639         case TRACE_FN: {
1640                 struct ftrace_entry *field;
1641
1642                 trace_assign_type(field, entry);
1643
1644                 seq_print_ip_sym(s, field->ip, sym_flags);
1645                 trace_seq_puts(s, " (");
1646                 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1647                 trace_seq_puts(s, ")\n");
1648                 break;
1649         }
1650         case TRACE_CTX:
1651         case TRACE_WAKE: {
1652                 struct ctx_switch_entry *field;
1653
1654                 trace_assign_type(field, entry);
1655
1656                 T = field->next_state < sizeof(state_to_char) ?
1657                         state_to_char[field->next_state] : 'X';
1658
1659                 state = field->prev_state ?
1660                         __ffs(field->prev_state) + 1 : 0;
1661                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1662                 comm = trace_find_cmdline(field->next_pid);
1663                 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1664                                  field->prev_pid,
1665                                  field->prev_prio,
1666                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1667                                  field->next_cpu,
1668                                  field->next_pid,
1669                                  field->next_prio,
1670                                  T, comm);
1671                 break;
1672         }
1673         case TRACE_SPECIAL: {
1674                 struct special_entry *field;
1675
1676                 trace_assign_type(field, entry);
1677
1678                 trace_seq_printf(s, "# %ld %ld %ld\n",
1679                                  field->arg1,
1680                                  field->arg2,
1681                                  field->arg3);
1682                 break;
1683         }
1684         case TRACE_STACK: {
1685                 struct stack_entry *field;
1686
1687                 trace_assign_type(field, entry);
1688
1689                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1690                         if (i)
1691                                 trace_seq_puts(s, " <= ");
1692                         seq_print_ip_sym(s, field->caller[i], sym_flags);
1693                 }
1694                 trace_seq_puts(s, "\n");
1695                 break;
1696         }
1697         case TRACE_PRINT: {
1698                 struct print_entry *field;
1699
1700                 trace_assign_type(field, entry);
1701
1702                 seq_print_ip_sym(s, field->ip, sym_flags);
1703                 trace_seq_printf(s, ": %s", field->buf);
1704                 if (entry->flags & TRACE_FLAG_CONT)
1705                         trace_seq_print_cont(s, iter);
1706                 break;
1707         }
1708         case TRACE_BRANCH: {
1709                 struct trace_branch *field;
1710
1711                 trace_assign_type(field, entry);
1712
1713                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1714                                  field->correct ? "  ok  " : " MISS ",
1715                                  field->func,
1716                                  field->file,
1717                                  field->line);
1718                 break;
1719         }
1720         default:
1721                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1722         }
1723         return TRACE_TYPE_HANDLED;
1724 }
1725
1726 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1727 {
1728         struct trace_seq *s = &iter->seq;
1729         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1730         struct trace_entry *entry;
1731         unsigned long usec_rem;
1732         unsigned long long t;
1733         unsigned long secs;
1734         char *comm;
1735         int ret;
1736         int S, T;
1737         int i;
1738
1739         entry = iter->ent;
1740
1741         if (entry->type == TRACE_CONT)
1742                 return TRACE_TYPE_HANDLED;
1743
1744         test_cpu_buff_start(iter);
1745
1746         comm = trace_find_cmdline(iter->ent->pid);
1747
1748         t = ns2usecs(iter->ts);
1749         usec_rem = do_div(t, 1000000ULL);
1750         secs = (unsigned long)t;
1751
1752         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1753         if (!ret)
1754                 return TRACE_TYPE_PARTIAL_LINE;
1755         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1756         if (!ret)
1757                 return TRACE_TYPE_PARTIAL_LINE;
1758         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1759         if (!ret)
1760                 return TRACE_TYPE_PARTIAL_LINE;
1761
1762         switch (entry->type) {
1763         case TRACE_FN: {
1764                 struct ftrace_entry *field;
1765
1766                 trace_assign_type(field, entry);
1767
1768                 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1769                 if (!ret)
1770                         return TRACE_TYPE_PARTIAL_LINE;
1771                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1772                                                 field->parent_ip) {
1773                         ret = trace_seq_printf(s, " <-");
1774                         if (!ret)
1775                                 return TRACE_TYPE_PARTIAL_LINE;
1776                         ret = seq_print_ip_sym(s,
1777                                                field->parent_ip,
1778                                                sym_flags);
1779                         if (!ret)
1780                                 return TRACE_TYPE_PARTIAL_LINE;
1781                 }
1782                 ret = trace_seq_printf(s, "\n");
1783                 if (!ret)
1784                         return TRACE_TYPE_PARTIAL_LINE;
1785                 break;
1786         }
1787         case TRACE_CTX:
1788         case TRACE_WAKE: {
1789                 struct ctx_switch_entry *field;
1790
1791                 trace_assign_type(field, entry);
1792
1793                 S = field->prev_state < sizeof(state_to_char) ?
1794                         state_to_char[field->prev_state] : 'X';
1795                 T = field->next_state < sizeof(state_to_char) ?
1796                         state_to_char[field->next_state] : 'X';
1797                 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1798                                        field->prev_pid,
1799                                        field->prev_prio,
1800                                        S,
1801                                        entry->type == TRACE_CTX ? "==>" : "  +",
1802                                        field->next_cpu,
1803                                        field->next_pid,
1804                                        field->next_prio,
1805                                        T);
1806                 if (!ret)
1807                         return TRACE_TYPE_PARTIAL_LINE;
1808                 break;
1809         }
1810         case TRACE_SPECIAL: {
1811                 struct special_entry *field;
1812
1813                 trace_assign_type(field, entry);
1814
1815                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1816                                  field->arg1,
1817                                  field->arg2,
1818                                  field->arg3);
1819                 if (!ret)
1820                         return TRACE_TYPE_PARTIAL_LINE;
1821                 break;
1822         }
1823         case TRACE_STACK: {
1824                 struct stack_entry *field;
1825
1826                 trace_assign_type(field, entry);
1827
1828                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1829                         if (i) {
1830                                 ret = trace_seq_puts(s, " <= ");
1831                                 if (!ret)
1832                                         return TRACE_TYPE_PARTIAL_LINE;
1833                         }
1834                         ret = seq_print_ip_sym(s, field->caller[i],
1835                                                sym_flags);
1836                         if (!ret)
1837                                 return TRACE_TYPE_PARTIAL_LINE;
1838                 }
1839                 ret = trace_seq_puts(s, "\n");
1840                 if (!ret)
1841                         return TRACE_TYPE_PARTIAL_LINE;
1842                 break;
1843         }
1844         case TRACE_PRINT: {
1845                 struct print_entry *field;
1846
1847                 trace_assign_type(field, entry);
1848
1849                 seq_print_ip_sym(s, field->ip, sym_flags);
1850                 trace_seq_printf(s, ": %s", field->buf);
1851                 if (entry->flags & TRACE_FLAG_CONT)
1852                         trace_seq_print_cont(s, iter);
1853                 break;
1854         }
1855         case TRACE_FN_RET: {
1856                 return print_return_function(iter);
1857                 break;
1858         }
1859         case TRACE_BRANCH: {
1860                 struct trace_branch *field;
1861
1862                 trace_assign_type(field, entry);
1863
1864                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1865                                  field->correct ? "  ok  " : " MISS ",
1866                                  field->func,
1867                                  field->file,
1868                                  field->line);
1869                 break;
1870         }
1871         }
1872         return TRACE_TYPE_HANDLED;
1873 }
1874
1875 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1876 {
1877         struct trace_seq *s = &iter->seq;
1878         struct trace_entry *entry;
1879         int ret;
1880         int S, T;
1881
1882         entry = iter->ent;
1883
1884         if (entry->type == TRACE_CONT)
1885                 return TRACE_TYPE_HANDLED;
1886
1887         ret = trace_seq_printf(s, "%d %d %llu ",
1888                 entry->pid, iter->cpu, iter->ts);
1889         if (!ret)
1890                 return TRACE_TYPE_PARTIAL_LINE;
1891
1892         switch (entry->type) {
1893         case TRACE_FN: {
1894                 struct ftrace_entry *field;
1895
1896                 trace_assign_type(field, entry);
1897
1898                 ret = trace_seq_printf(s, "%x %x\n",
1899                                         field->ip,
1900                                         field->parent_ip);
1901                 if (!ret)
1902                         return TRACE_TYPE_PARTIAL_LINE;
1903                 break;
1904         }
1905         case TRACE_CTX:
1906         case TRACE_WAKE: {
1907                 struct ctx_switch_entry *field;
1908
1909                 trace_assign_type(field, entry);
1910
1911                 S = field->prev_state < sizeof(state_to_char) ?
1912                         state_to_char[field->prev_state] : 'X';
1913                 T = field->next_state < sizeof(state_to_char) ?
1914                         state_to_char[field->next_state] : 'X';
1915                 if (entry->type == TRACE_WAKE)
1916                         S = '+';
1917                 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
1918                                        field->prev_pid,
1919                                        field->prev_prio,
1920                                        S,
1921                                        field->next_cpu,
1922                                        field->next_pid,
1923                                        field->next_prio,
1924                                        T);
1925                 if (!ret)
1926                         return TRACE_TYPE_PARTIAL_LINE;
1927                 break;
1928         }
1929         case TRACE_SPECIAL:
1930         case TRACE_STACK: {
1931                 struct special_entry *field;
1932
1933                 trace_assign_type(field, entry);
1934
1935                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1936                                  field->arg1,
1937                                  field->arg2,
1938                                  field->arg3);
1939                 if (!ret)
1940                         return TRACE_TYPE_PARTIAL_LINE;
1941                 break;
1942         }
1943         case TRACE_PRINT: {
1944                 struct print_entry *field;
1945
1946                 trace_assign_type(field, entry);
1947
1948                 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
1949                 if (entry->flags & TRACE_FLAG_CONT)
1950                         trace_seq_print_cont(s, iter);
1951                 break;
1952         }
1953         }
1954         return TRACE_TYPE_HANDLED;
1955 }
1956
1957 #define SEQ_PUT_FIELD_RET(s, x)                         \
1958 do {                                                    \
1959         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1960                 return 0;                               \
1961 } while (0)
1962
1963 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1964 do {                                                    \
1965         BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES);     \
1966         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1967                 return 0;                               \
1968 } while (0)
1969
1970 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1971 {
1972         struct trace_seq *s = &iter->seq;
1973         unsigned char newline = '\n';
1974         struct trace_entry *entry;
1975         int S, T;
1976
1977         entry = iter->ent;
1978
1979         if (entry->type == TRACE_CONT)
1980                 return TRACE_TYPE_HANDLED;
1981
1982         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1983         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1984         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1985
1986         switch (entry->type) {
1987         case TRACE_FN: {
1988                 struct ftrace_entry *field;
1989
1990                 trace_assign_type(field, entry);
1991
1992                 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
1993                 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
1994                 break;
1995         }
1996         case TRACE_CTX:
1997         case TRACE_WAKE: {
1998                 struct ctx_switch_entry *field;
1999
2000                 trace_assign_type(field, entry);
2001
2002                 S = field->prev_state < sizeof(state_to_char) ?
2003                         state_to_char[field->prev_state] : 'X';
2004                 T = field->next_state < sizeof(state_to_char) ?
2005                         state_to_char[field->next_state] : 'X';
2006                 if (entry->type == TRACE_WAKE)
2007                         S = '+';
2008                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2009                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2010                 SEQ_PUT_HEX_FIELD_RET(s, S);
2011                 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2012                 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2013                 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2014                 SEQ_PUT_HEX_FIELD_RET(s, T);
2015                 break;
2016         }
2017         case TRACE_SPECIAL:
2018         case TRACE_STACK: {
2019                 struct special_entry *field;
2020
2021                 trace_assign_type(field, entry);
2022
2023                 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2024                 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2025                 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2026                 break;
2027         }
2028         }
2029         SEQ_PUT_FIELD_RET(s, newline);
2030
2031         return TRACE_TYPE_HANDLED;
2032 }
2033
2034 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2035 {
2036         struct trace_seq *s = &iter->seq;
2037         struct trace_entry *entry;
2038
2039         entry = iter->ent;
2040
2041         if (entry->type == TRACE_CONT)
2042                 return TRACE_TYPE_HANDLED;
2043
2044         SEQ_PUT_FIELD_RET(s, entry->pid);
2045         SEQ_PUT_FIELD_RET(s, entry->cpu);
2046         SEQ_PUT_FIELD_RET(s, iter->ts);
2047
2048         switch (entry->type) {
2049         case TRACE_FN: {
2050                 struct ftrace_entry *field;
2051
2052                 trace_assign_type(field, entry);
2053
2054                 SEQ_PUT_FIELD_RET(s, field->ip);
2055                 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2056                 break;
2057         }
2058         case TRACE_CTX: {
2059                 struct ctx_switch_entry *field;
2060
2061                 trace_assign_type(field, entry);
2062
2063                 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2064                 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2065                 SEQ_PUT_FIELD_RET(s, field->prev_state);
2066                 SEQ_PUT_FIELD_RET(s, field->next_pid);
2067                 SEQ_PUT_FIELD_RET(s, field->next_prio);
2068                 SEQ_PUT_FIELD_RET(s, field->next_state);
2069                 break;
2070         }
2071         case TRACE_SPECIAL:
2072         case TRACE_STACK: {
2073                 struct special_entry *field;
2074
2075                 trace_assign_type(field, entry);
2076
2077                 SEQ_PUT_FIELD_RET(s, field->arg1);
2078                 SEQ_PUT_FIELD_RET(s, field->arg2);
2079                 SEQ_PUT_FIELD_RET(s, field->arg3);
2080                 break;
2081         }
2082         }
2083         return 1;
2084 }
2085
2086 static int trace_empty(struct trace_iterator *iter)
2087 {
2088         int cpu;
2089
2090         for_each_tracing_cpu(cpu) {
2091                 if (iter->buffer_iter[cpu]) {
2092                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2093                                 return 0;
2094                 } else {
2095                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2096                                 return 0;
2097                 }
2098         }
2099
2100         return 1;
2101 }
2102
2103 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2104 {
2105         enum print_line_t ret;
2106
2107         if (iter->trace && iter->trace->print_line) {
2108                 ret = iter->trace->print_line(iter);
2109                 if (ret != TRACE_TYPE_UNHANDLED)
2110                         return ret;
2111         }
2112
2113         if (trace_flags & TRACE_ITER_BIN)
2114                 return print_bin_fmt(iter);
2115
2116         if (trace_flags & TRACE_ITER_HEX)
2117                 return print_hex_fmt(iter);
2118
2119         if (trace_flags & TRACE_ITER_RAW)
2120                 return print_raw_fmt(iter);
2121
2122         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2123                 return print_lat_fmt(iter, iter->idx, iter->cpu);
2124
2125         return print_trace_fmt(iter);
2126 }
2127
2128 static int s_show(struct seq_file *m, void *v)
2129 {
2130         struct trace_iterator *iter = v;
2131
2132         if (iter->ent == NULL) {
2133                 if (iter->tr) {
2134                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2135                         seq_puts(m, "#\n");
2136                 }
2137                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2138                         /* print nothing if the buffers are empty */
2139                         if (trace_empty(iter))
2140                                 return 0;
2141                         print_trace_header(m, iter);
2142                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2143                                 print_lat_help_header(m);
2144                 } else {
2145                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2146                                 print_func_help_header(m);
2147                 }
2148         } else {
2149                 print_trace_line(iter);
2150                 trace_print_seq(m, &iter->seq);
2151         }
2152
2153         return 0;
2154 }
2155
2156 static struct seq_operations tracer_seq_ops = {
2157         .start          = s_start,
2158         .next           = s_next,
2159         .stop           = s_stop,
2160         .show           = s_show,
2161 };
2162
2163 static struct trace_iterator *
2164 __tracing_open(struct inode *inode, struct file *file, int *ret)
2165 {
2166         struct trace_iterator *iter;
2167         struct seq_file *m;
2168         int cpu;
2169
2170         if (tracing_disabled) {
2171                 *ret = -ENODEV;
2172                 return NULL;
2173         }
2174
2175         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2176         if (!iter) {
2177                 *ret = -ENOMEM;
2178                 goto out;
2179         }
2180
2181         mutex_lock(&trace_types_lock);
2182         if (current_trace && current_trace->print_max)
2183                 iter->tr = &max_tr;
2184         else
2185                 iter->tr = inode->i_private;
2186         iter->trace = current_trace;
2187         iter->pos = -1;
2188
2189         /* Annotate start of buffers if we had overruns */
2190         if (ring_buffer_overruns(iter->tr->buffer))
2191                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2192
2193
2194         for_each_tracing_cpu(cpu) {
2195
2196                 iter->buffer_iter[cpu] =
2197                         ring_buffer_read_start(iter->tr->buffer, cpu);
2198
2199                 if (!iter->buffer_iter[cpu])
2200                         goto fail_buffer;
2201         }
2202
2203         /* TODO stop tracer */
2204         *ret = seq_open(file, &tracer_seq_ops);
2205         if (*ret)
2206                 goto fail_buffer;
2207
2208         m = file->private_data;
2209         m->private = iter;
2210
2211         /* stop the trace while dumping */
2212         tracing_stop();
2213
2214         if (iter->trace && iter->trace->open)
2215                         iter->trace->open(iter);
2216
2217         mutex_unlock(&trace_types_lock);
2218
2219  out:
2220         return iter;
2221
2222  fail_buffer:
2223         for_each_tracing_cpu(cpu) {
2224                 if (iter->buffer_iter[cpu])
2225                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2226         }
2227         mutex_unlock(&trace_types_lock);
2228         kfree(iter);
2229
2230         return ERR_PTR(-ENOMEM);
2231 }
2232
2233 int tracing_open_generic(struct inode *inode, struct file *filp)
2234 {
2235         if (tracing_disabled)
2236                 return -ENODEV;
2237
2238         filp->private_data = inode->i_private;
2239         return 0;
2240 }
2241
2242 int tracing_release(struct inode *inode, struct file *file)
2243 {
2244         struct seq_file *m = (struct seq_file *)file->private_data;
2245         struct trace_iterator *iter = m->private;
2246         int cpu;
2247
2248         mutex_lock(&trace_types_lock);
2249         for_each_tracing_cpu(cpu) {
2250                 if (iter->buffer_iter[cpu])
2251                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2252         }
2253
2254         if (iter->trace && iter->trace->close)
2255                 iter->trace->close(iter);
2256
2257         /* reenable tracing if it was previously enabled */
2258         tracing_start();
2259         mutex_unlock(&trace_types_lock);
2260
2261         seq_release(inode, file);
2262         kfree(iter);
2263         return 0;
2264 }
2265
2266 static int tracing_open(struct inode *inode, struct file *file)
2267 {
2268         int ret;
2269
2270         __tracing_open(inode, file, &ret);
2271
2272         return ret;
2273 }
2274
2275 static int tracing_lt_open(struct inode *inode, struct file *file)
2276 {
2277         struct trace_iterator *iter;
2278         int ret;
2279
2280         iter = __tracing_open(inode, file, &ret);
2281
2282         if (!ret)
2283                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2284
2285         return ret;
2286 }
2287
2288
2289 static void *
2290 t_next(struct seq_file *m, void *v, loff_t *pos)
2291 {
2292         struct tracer *t = m->private;
2293
2294         (*pos)++;
2295
2296         if (t)
2297                 t = t->next;
2298
2299         m->private = t;
2300
2301         return t;
2302 }
2303
2304 static void *t_start(struct seq_file *m, loff_t *pos)
2305 {
2306         struct tracer *t = m->private;
2307         loff_t l = 0;
2308
2309         mutex_lock(&trace_types_lock);
2310         for (; t && l < *pos; t = t_next(m, t, &l))
2311                 ;
2312
2313         return t;
2314 }
2315
2316 static void t_stop(struct seq_file *m, void *p)
2317 {
2318         mutex_unlock(&trace_types_lock);
2319 }
2320
2321 static int t_show(struct seq_file *m, void *v)
2322 {
2323         struct tracer *t = v;
2324
2325         if (!t)
2326                 return 0;
2327
2328         seq_printf(m, "%s", t->name);
2329         if (t->next)
2330                 seq_putc(m, ' ');
2331         else
2332                 seq_putc(m, '\n');
2333
2334         return 0;
2335 }
2336
2337 static struct seq_operations show_traces_seq_ops = {
2338         .start          = t_start,
2339         .next           = t_next,
2340         .stop           = t_stop,
2341         .show           = t_show,
2342 };
2343
2344 static int show_traces_open(struct inode *inode, struct file *file)
2345 {
2346         int ret;
2347
2348         if (tracing_disabled)
2349                 return -ENODEV;
2350
2351         ret = seq_open(file, &show_traces_seq_ops);
2352         if (!ret) {
2353                 struct seq_file *m = file->private_data;
2354                 m->private = trace_types;
2355         }
2356
2357         return ret;
2358 }
2359
2360 static struct file_operations tracing_fops = {
2361         .open           = tracing_open,
2362         .read           = seq_read,
2363         .llseek         = seq_lseek,
2364         .release        = tracing_release,
2365 };
2366
2367 static struct file_operations tracing_lt_fops = {
2368         .open           = tracing_lt_open,
2369         .read           = seq_read,
2370         .llseek         = seq_lseek,
2371         .release        = tracing_release,
2372 };
2373
2374 static struct file_operations show_traces_fops = {
2375         .open           = show_traces_open,
2376         .read           = seq_read,
2377         .release        = seq_release,
2378 };
2379
2380 /*
2381  * Only trace on a CPU if the bitmask is set:
2382  */
2383 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2384
2385 /*
2386  * When tracing/tracing_cpu_mask is modified then this holds
2387  * the new bitmask we are about to install:
2388  */
2389 static cpumask_t tracing_cpumask_new;
2390
2391 /*
2392  * The tracer itself will not take this lock, but still we want
2393  * to provide a consistent cpumask to user-space:
2394  */
2395 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2396
2397 /*
2398  * Temporary storage for the character representation of the
2399  * CPU bitmask (and one more byte for the newline):
2400  */
2401 static char mask_str[NR_CPUS + 1];
2402
2403 static ssize_t
2404 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2405                      size_t count, loff_t *ppos)
2406 {
2407         int len;
2408
2409         mutex_lock(&tracing_cpumask_update_lock);
2410
2411         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2412         if (count - len < 2) {
2413                 count = -EINVAL;
2414                 goto out_err;
2415         }
2416         len += sprintf(mask_str + len, "\n");
2417         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2418
2419 out_err:
2420         mutex_unlock(&tracing_cpumask_update_lock);
2421
2422         return count;
2423 }
2424
2425 static ssize_t
2426 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2427                       size_t count, loff_t *ppos)
2428 {
2429         int err, cpu;
2430
2431         mutex_lock(&tracing_cpumask_update_lock);
2432         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2433         if (err)
2434                 goto err_unlock;
2435
2436         raw_local_irq_disable();
2437         __raw_spin_lock(&ftrace_max_lock);
2438         for_each_tracing_cpu(cpu) {
2439                 /*
2440                  * Increase/decrease the disabled counter if we are
2441                  * about to flip a bit in the cpumask:
2442                  */
2443                 if (cpu_isset(cpu, tracing_cpumask) &&
2444                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2445                         atomic_inc(&global_trace.data[cpu]->disabled);
2446                 }
2447                 if (!cpu_isset(cpu, tracing_cpumask) &&
2448                                 cpu_isset(cpu, tracing_cpumask_new)) {
2449                         atomic_dec(&global_trace.data[cpu]->disabled);
2450                 }
2451         }
2452         __raw_spin_unlock(&ftrace_max_lock);
2453         raw_local_irq_enable();
2454
2455         tracing_cpumask = tracing_cpumask_new;
2456
2457         mutex_unlock(&tracing_cpumask_update_lock);
2458
2459         return count;
2460
2461 err_unlock:
2462         mutex_unlock(&tracing_cpumask_update_lock);
2463
2464         return err;
2465 }
2466
2467 static struct file_operations tracing_cpumask_fops = {
2468         .open           = tracing_open_generic,
2469         .read           = tracing_cpumask_read,
2470         .write          = tracing_cpumask_write,
2471 };
2472
2473 static ssize_t
2474 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2475                        size_t cnt, loff_t *ppos)
2476 {
2477         int i;
2478         char *buf;
2479         int r = 0;
2480         int len = 0;
2481         u32 tracer_flags = current_trace->flags->val;
2482         struct tracer_opt *trace_opts = current_trace->flags->opts;
2483
2484
2485         /* calulate max size */
2486         for (i = 0; trace_options[i]; i++) {
2487                 len += strlen(trace_options[i]);
2488                 len += 3; /* "no" and space */
2489         }
2490
2491         /*
2492          * Increase the size with names of options specific
2493          * of the current tracer.
2494          */
2495         for (i = 0; trace_opts[i].name; i++) {
2496                 len += strlen(trace_opts[i].name);
2497                 len += 3; /* "no" and space */
2498         }
2499
2500         /* +2 for \n and \0 */
2501         buf = kmalloc(len + 2, GFP_KERNEL);
2502         if (!buf)
2503                 return -ENOMEM;
2504
2505         for (i = 0; trace_options[i]; i++) {
2506                 if (trace_flags & (1 << i))
2507                         r += sprintf(buf + r, "%s ", trace_options[i]);
2508                 else
2509                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2510         }
2511
2512         for (i = 0; trace_opts[i].name; i++) {
2513                 if (tracer_flags & trace_opts[i].bit)
2514                         r += sprintf(buf + r, "%s ",
2515                                 trace_opts[i].name);
2516                 else
2517                         r += sprintf(buf + r, "no%s ",
2518                                 trace_opts[i].name);
2519         }
2520
2521         r += sprintf(buf + r, "\n");
2522         WARN_ON(r >= len + 2);
2523
2524         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2525
2526         kfree(buf);
2527
2528         return r;
2529 }
2530
2531 /* Try to assign a tracer specific option */
2532 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2533 {
2534         struct tracer_flags *trace_flags = trace->flags;
2535         struct tracer_opt *opts = NULL;
2536         int ret = 0, i = 0;
2537         int len;
2538
2539         for (i = 0; trace_flags->opts[i].name; i++) {
2540                 opts = &trace_flags->opts[i];
2541                 len = strlen(opts->name);
2542
2543                 if (strncmp(cmp, opts->name, len) == 0) {
2544                         ret = trace->set_flag(trace_flags->val,
2545                                 opts->bit, !neg);
2546                         break;
2547                 }
2548         }
2549         /* Not found */
2550         if (!trace_flags->opts[i].name)
2551                 return -EINVAL;
2552
2553         /* Refused to handle */
2554         if (ret)
2555                 return ret;
2556
2557         if (neg)
2558                 trace_flags->val &= ~opts->bit;
2559         else
2560                 trace_flags->val |= opts->bit;
2561
2562         return 0;
2563 }
2564
2565 static ssize_t
2566 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2567                         size_t cnt, loff_t *ppos)
2568 {
2569         char buf[64];
2570         char *cmp = buf;
2571         int neg = 0;
2572         int ret;
2573         int i;
2574
2575         if (cnt >= sizeof(buf))
2576                 return -EINVAL;
2577
2578         if (copy_from_user(&buf, ubuf, cnt))
2579                 return -EFAULT;
2580
2581         buf[cnt] = 0;
2582
2583         if (strncmp(buf, "no", 2) == 0) {
2584                 neg = 1;
2585                 cmp += 2;
2586         }
2587
2588         for (i = 0; trace_options[i]; i++) {
2589                 int len = strlen(trace_options[i]);
2590
2591                 if (strncmp(cmp, trace_options[i], len) == 0) {
2592                         if (neg)
2593                                 trace_flags &= ~(1 << i);
2594                         else
2595                                 trace_flags |= (1 << i);
2596                         break;
2597                 }
2598         }
2599
2600         /* If no option could be set, test the specific tracer options */
2601         if (!trace_options[i]) {
2602                 ret = set_tracer_option(current_trace, cmp, neg);
2603                 if (ret)
2604                         return ret;
2605         }
2606
2607         filp->f_pos += cnt;
2608
2609         return cnt;
2610 }
2611
2612 static struct file_operations tracing_iter_fops = {
2613         .open           = tracing_open_generic,
2614         .read           = tracing_trace_options_read,
2615         .write          = tracing_trace_options_write,
2616 };
2617
2618 static const char readme_msg[] =
2619         "tracing mini-HOWTO:\n\n"
2620         "# mkdir /debug\n"
2621         "# mount -t debugfs nodev /debug\n\n"
2622         "# cat /debug/tracing/available_tracers\n"
2623         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2624         "# cat /debug/tracing/current_tracer\n"
2625         "none\n"
2626         "# echo sched_switch > /debug/tracing/current_tracer\n"
2627         "# cat /debug/tracing/current_tracer\n"
2628         "sched_switch\n"
2629         "# cat /debug/tracing/trace_options\n"
2630         "noprint-parent nosym-offset nosym-addr noverbose\n"
2631         "# echo print-parent > /debug/tracing/trace_options\n"
2632         "# echo 1 > /debug/tracing/tracing_enabled\n"
2633         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2634         "echo 0 > /debug/tracing/tracing_enabled\n"
2635 ;
2636
2637 static ssize_t
2638 tracing_readme_read(struct file *filp, char __user *ubuf,
2639                        size_t cnt, loff_t *ppos)
2640 {
2641         return simple_read_from_buffer(ubuf, cnt, ppos,
2642                                         readme_msg, strlen(readme_msg));
2643 }
2644
2645 static struct file_operations tracing_readme_fops = {
2646         .open           = tracing_open_generic,
2647         .read           = tracing_readme_read,
2648 };
2649
2650 static ssize_t
2651 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2652                   size_t cnt, loff_t *ppos)
2653 {
2654         char buf[64];
2655         int r;
2656
2657         r = sprintf(buf, "%u\n", tracer_enabled);
2658         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2659 }
2660
2661 static ssize_t
2662 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2663                    size_t cnt, loff_t *ppos)
2664 {
2665         struct trace_array *tr = filp->private_data;
2666         char buf[64];
2667         long val;
2668         int ret;
2669
2670         if (cnt >= sizeof(buf))
2671                 return -EINVAL;
2672
2673         if (copy_from_user(&buf, ubuf, cnt))
2674                 return -EFAULT;
2675
2676         buf[cnt] = 0;
2677
2678         ret = strict_strtoul(buf, 10, &val);
2679         if (ret < 0)
2680                 return ret;
2681
2682         val = !!val;
2683
2684         mutex_lock(&trace_types_lock);
2685         if (tracer_enabled ^ val) {
2686                 if (val) {
2687                         tracer_enabled = 1;
2688                         if (current_trace->start)
2689                                 current_trace->start(tr);
2690                         tracing_start();
2691                 } else {
2692                         tracer_enabled = 0;
2693                         tracing_stop();
2694                         if (current_trace->stop)
2695                                 current_trace->stop(tr);
2696                 }
2697         }
2698         mutex_unlock(&trace_types_lock);
2699
2700         filp->f_pos += cnt;
2701
2702         return cnt;
2703 }
2704
2705 static ssize_t
2706 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2707                        size_t cnt, loff_t *ppos)
2708 {
2709         char buf[max_tracer_type_len+2];
2710         int r;
2711
2712         mutex_lock(&trace_types_lock);
2713         if (current_trace)
2714                 r = sprintf(buf, "%s\n", current_trace->name);
2715         else
2716                 r = sprintf(buf, "\n");
2717         mutex_unlock(&trace_types_lock);
2718
2719         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2720 }
2721
2722 static int tracing_set_tracer(char *buf)
2723 {
2724         struct trace_array *tr = &global_trace;
2725         struct tracer *t;
2726         int ret = 0;
2727
2728         mutex_lock(&trace_types_lock);
2729         for (t = trace_types; t; t = t->next) {
2730                 if (strcmp(t->name, buf) == 0)
2731                         break;
2732         }
2733         if (!t) {
2734                 ret = -EINVAL;
2735                 goto out;
2736         }
2737         if (t == current_trace)
2738                 goto out;
2739
2740         trace_branch_disable();
2741         if (current_trace && current_trace->reset)
2742                 current_trace->reset(tr);
2743
2744         current_trace = t;
2745         if (t->init) {
2746                 ret = t->init(tr);
2747                 if (ret)
2748                         goto out;
2749         }
2750
2751         trace_branch_enable(tr);
2752  out:
2753         mutex_unlock(&trace_types_lock);
2754
2755         return ret;
2756 }
2757
2758 static ssize_t
2759 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2760                         size_t cnt, loff_t *ppos)
2761 {
2762         char buf[max_tracer_type_len+1];
2763         int i;
2764         size_t ret;
2765         int err;
2766
2767         ret = cnt;
2768
2769         if (cnt > max_tracer_type_len)
2770                 cnt = max_tracer_type_len;
2771
2772         if (copy_from_user(&buf, ubuf, cnt))
2773                 return -EFAULT;
2774
2775         buf[cnt] = 0;
2776
2777         /* strip ending whitespace. */
2778         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2779                 buf[i] = 0;
2780
2781         err = tracing_set_tracer(buf);
2782         if (err)
2783                 return err;
2784
2785         filp->f_pos += ret;
2786
2787         return ret;
2788 }
2789
2790 static ssize_t
2791 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2792                      size_t cnt, loff_t *ppos)
2793 {
2794         unsigned long *ptr = filp->private_data;
2795         char buf[64];
2796         int r;
2797
2798         r = snprintf(buf, sizeof(buf), "%ld\n",
2799                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2800         if (r > sizeof(buf))
2801                 r = sizeof(buf);
2802         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2803 }
2804
2805 static ssize_t
2806 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2807                       size_t cnt, loff_t *ppos)
2808 {
2809         long *ptr = filp->private_data;
2810         char buf[64];
2811         long val;
2812         int ret;
2813
2814         if (cnt >= sizeof(buf))
2815                 return -EINVAL;
2816
2817         if (copy_from_user(&buf, ubuf, cnt))
2818                 return -EFAULT;
2819
2820         buf[cnt] = 0;
2821
2822         ret = strict_strtoul(buf, 10, &val);
2823         if (ret < 0)
2824                 return ret;
2825
2826         *ptr = val * 1000;
2827
2828         return cnt;
2829 }
2830
2831 static atomic_t tracing_reader;
2832
2833 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2834 {
2835         struct trace_iterator *iter;
2836
2837         if (tracing_disabled)
2838                 return -ENODEV;
2839
2840         /* We only allow for reader of the pipe */
2841         if (atomic_inc_return(&tracing_reader) != 1) {
2842                 atomic_dec(&tracing_reader);
2843                 return -EBUSY;
2844         }
2845
2846         /* create a buffer to store the information to pass to userspace */
2847         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2848         if (!iter)
2849                 return -ENOMEM;
2850
2851         mutex_lock(&trace_types_lock);
2852
2853         /* trace pipe does not show start of buffer */
2854         cpus_setall(iter->started);
2855
2856         iter->tr = &global_trace;
2857         iter->trace = current_trace;
2858         filp->private_data = iter;
2859
2860         if (iter->trace->pipe_open)
2861                 iter->trace->pipe_open(iter);
2862         mutex_unlock(&trace_types_lock);
2863
2864         return 0;
2865 }
2866
2867 static int tracing_release_pipe(struct inode *inode, struct file *file)
2868 {
2869         struct trace_iterator *iter = file->private_data;
2870
2871         kfree(iter);
2872         atomic_dec(&tracing_reader);
2873
2874         return 0;
2875 }
2876
2877 static unsigned int
2878 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2879 {
2880         struct trace_iterator *iter = filp->private_data;
2881
2882         if (trace_flags & TRACE_ITER_BLOCK) {
2883                 /*
2884                  * Always select as readable when in blocking mode
2885                  */
2886                 return POLLIN | POLLRDNORM;
2887         } else {
2888                 if (!trace_empty(iter))
2889                         return POLLIN | POLLRDNORM;
2890                 poll_wait(filp, &trace_wait, poll_table);
2891                 if (!trace_empty(iter))
2892                         return POLLIN | POLLRDNORM;
2893
2894                 return 0;
2895         }
2896 }
2897
2898 /*
2899  * Consumer reader.
2900  */
2901 static ssize_t
2902 tracing_read_pipe(struct file *filp, char __user *ubuf,
2903                   size_t cnt, loff_t *ppos)
2904 {
2905         struct trace_iterator *iter = filp->private_data;
2906         ssize_t sret;
2907
2908         /* return any leftover data */
2909         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2910         if (sret != -EBUSY)
2911                 return sret;
2912
2913         trace_seq_reset(&iter->seq);
2914
2915         mutex_lock(&trace_types_lock);
2916         if (iter->trace->read) {
2917                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2918                 if (sret)
2919                         goto out;
2920         }
2921
2922 waitagain:
2923         sret = 0;
2924         while (trace_empty(iter)) {
2925
2926                 if ((filp->f_flags & O_NONBLOCK)) {
2927                         sret = -EAGAIN;
2928                         goto out;
2929                 }
2930
2931                 /*
2932                  * This is a make-shift waitqueue. The reason we don't use
2933                  * an actual wait queue is because:
2934                  *  1) we only ever have one waiter
2935                  *  2) the tracing, traces all functions, we don't want
2936                  *     the overhead of calling wake_up and friends
2937                  *     (and tracing them too)
2938                  *     Anyway, this is really very primitive wakeup.
2939                  */
2940                 set_current_state(TASK_INTERRUPTIBLE);
2941                 iter->tr->waiter = current;
2942
2943                 mutex_unlock(&trace_types_lock);
2944
2945                 /* sleep for 100 msecs, and try again. */
2946                 schedule_timeout(HZ/10);
2947
2948                 mutex_lock(&trace_types_lock);
2949
2950                 iter->tr->waiter = NULL;
2951
2952                 if (signal_pending(current)) {
2953                         sret = -EINTR;
2954                         goto out;
2955                 }
2956
2957                 if (iter->trace != current_trace)
2958                         goto out;
2959
2960                 /*
2961                  * We block until we read something and tracing is disabled.
2962                  * We still block if tracing is disabled, but we have never
2963                  * read anything. This allows a user to cat this file, and
2964                  * then enable tracing. But after we have read something,
2965                  * we give an EOF when tracing is again disabled.
2966                  *
2967                  * iter->pos will be 0 if we haven't read anything.
2968                  */
2969                 if (!tracer_enabled && iter->pos)
2970                         break;
2971
2972                 continue;
2973         }
2974
2975         /* stop when tracing is finished */
2976         if (trace_empty(iter))
2977                 goto out;
2978
2979         if (cnt >= PAGE_SIZE)
2980                 cnt = PAGE_SIZE - 1;
2981
2982         /* reset all but tr, trace, and overruns */
2983         memset(&iter->seq, 0,
2984                sizeof(struct trace_iterator) -
2985                offsetof(struct trace_iterator, seq));
2986         iter->pos = -1;
2987
2988         while (find_next_entry_inc(iter) != NULL) {
2989                 enum print_line_t ret;
2990                 int len = iter->seq.len;
2991
2992                 ret = print_trace_line(iter);
2993                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2994                         /* don't print partial lines */
2995                         iter->seq.len = len;
2996                         break;
2997                 }
2998
2999                 trace_consume(iter);
3000
3001                 if (iter->seq.len >= cnt)
3002                         break;
3003         }
3004
3005         /* Now copy what we have to the user */
3006         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3007         if (iter->seq.readpos >= iter->seq.len)
3008                 trace_seq_reset(&iter->seq);
3009
3010         /*
3011          * If there was nothing to send to user, inspite of consuming trace
3012          * entries, go back to wait for more entries.
3013          */
3014         if (sret == -EBUSY)
3015                 goto waitagain;
3016
3017 out:
3018         mutex_unlock(&trace_types_lock);
3019
3020         return sret;
3021 }
3022
3023 static ssize_t
3024 tracing_entries_read(struct file *filp, char __user *ubuf,
3025                      size_t cnt, loff_t *ppos)
3026 {
3027         struct trace_array *tr = filp->private_data;
3028         char buf[64];
3029         int r;
3030
3031         r = sprintf(buf, "%lu\n", tr->entries >> 10);
3032         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3033 }
3034
3035 static ssize_t
3036 tracing_entries_write(struct file *filp, const char __user *ubuf,
3037                       size_t cnt, loff_t *ppos)
3038 {
3039         unsigned long val;
3040         char buf[64];
3041         int ret, cpu;
3042
3043         if (cnt >= sizeof(buf))
3044                 return -EINVAL;
3045
3046         if (copy_from_user(&buf, ubuf, cnt))
3047                 return -EFAULT;
3048
3049         buf[cnt] = 0;
3050
3051         ret = strict_strtoul(buf, 10, &val);
3052         if (ret < 0)
3053                 return ret;
3054
3055         /* must have at least 1 entry */
3056         if (!val)
3057                 return -EINVAL;
3058
3059         mutex_lock(&trace_types_lock);
3060
3061         tracing_stop();
3062
3063         /* disable all cpu buffers */
3064         for_each_tracing_cpu(cpu) {
3065                 if (global_trace.data[cpu])
3066                         atomic_inc(&global_trace.data[cpu]->disabled);
3067                 if (max_tr.data[cpu])
3068                         atomic_inc(&max_tr.data[cpu]->disabled);
3069         }
3070
3071         /* value is in KB */
3072         val <<= 10;
3073
3074         if (val != global_trace.entries) {
3075                 ret = ring_buffer_resize(global_trace.buffer, val);
3076                 if (ret < 0) {
3077                         cnt = ret;
3078                         goto out;
3079                 }
3080
3081                 ret = ring_buffer_resize(max_tr.buffer, val);
3082                 if (ret < 0) {
3083                         int r;
3084                         cnt = ret;
3085                         r = ring_buffer_resize(global_trace.buffer,
3086                                                global_trace.entries);
3087                         if (r < 0) {
3088                                 /* AARGH! We are left with different
3089                                  * size max buffer!!!! */
3090                                 WARN_ON(1);
3091                                 tracing_disabled = 1;
3092                         }
3093                         goto out;
3094                 }
3095
3096                 global_trace.entries = val;
3097         }
3098
3099         filp->f_pos += cnt;
3100
3101         /* If check pages failed, return ENOMEM */
3102         if (tracing_disabled)
3103                 cnt = -ENOMEM;
3104  out:
3105         for_each_tracing_cpu(cpu) {
3106                 if (global_trace.data[cpu])
3107                         atomic_dec(&global_trace.data[cpu]->disabled);
3108                 if (max_tr.data[cpu])
3109                         atomic_dec(&max_tr.data[cpu]->disabled);
3110         }
3111
3112         tracing_start();
3113         max_tr.entries = global_trace.entries;
3114         mutex_unlock(&trace_types_lock);
3115
3116         return cnt;
3117 }
3118
3119 static int mark_printk(const char *fmt, ...)
3120 {
3121         int ret;
3122         va_list args;
3123         va_start(args, fmt);
3124         ret = trace_vprintk(0, fmt, args);
3125         va_end(args);
3126         return ret;
3127 }
3128
3129 static ssize_t
3130 tracing_mark_write(struct file *filp, const char __user *ubuf,
3131                                         size_t cnt, loff_t *fpos)
3132 {
3133         char *buf;
3134         char *end;
3135
3136         if (tracing_disabled)
3137                 return -EINVAL;
3138
3139         if (cnt > TRACE_BUF_SIZE)
3140                 cnt = TRACE_BUF_SIZE;
3141
3142         buf = kmalloc(cnt + 1, GFP_KERNEL);
3143         if (buf == NULL)
3144                 return -ENOMEM;
3145
3146         if (copy_from_user(buf, ubuf, cnt)) {
3147                 kfree(buf);
3148                 return -EFAULT;
3149         }
3150
3151         /* Cut from the first nil or newline. */
3152         buf[cnt] = '\0';
3153         end = strchr(buf, '\n');
3154         if (end)
3155                 *end = '\0';
3156
3157         cnt = mark_printk("%s\n", buf);
3158         kfree(buf);
3159         *fpos += cnt;
3160
3161         return cnt;
3162 }
3163
3164 static struct file_operations tracing_max_lat_fops = {
3165         .open           = tracing_open_generic,
3166         .read           = tracing_max_lat_read,
3167         .write          = tracing_max_lat_write,
3168 };
3169
3170 static struct file_operations tracing_ctrl_fops = {
3171         .open           = tracing_open_generic,
3172         .read           = tracing_ctrl_read,
3173         .write          = tracing_ctrl_write,
3174 };
3175
3176 static struct file_operations set_tracer_fops = {
3177         .open           = tracing_open_generic,
3178         .read           = tracing_set_trace_read,
3179         .write          = tracing_set_trace_write,
3180 };
3181
3182 static struct file_operations tracing_pipe_fops = {
3183         .open           = tracing_open_pipe,
3184         .poll           = tracing_poll_pipe,
3185         .read           = tracing_read_pipe,
3186         .release        = tracing_release_pipe,
3187 };
3188
3189 static struct file_operations tracing_entries_fops = {
3190         .open           = tracing_open_generic,
3191         .read           = tracing_entries_read,
3192         .write          = tracing_entries_write,
3193 };
3194
3195 static struct file_operations tracing_mark_fops = {
3196         .open           = tracing_open_generic,
3197         .write          = tracing_mark_write,
3198 };
3199
3200 #ifdef CONFIG_DYNAMIC_FTRACE
3201
3202 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3203 {
3204         return 0;
3205 }
3206
3207 static ssize_t
3208 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3209                   size_t cnt, loff_t *ppos)
3210 {
3211         static char ftrace_dyn_info_buffer[1024];
3212         static DEFINE_MUTEX(dyn_info_mutex);
3213         unsigned long *p = filp->private_data;
3214         char *buf = ftrace_dyn_info_buffer;
3215         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3216         int r;
3217
3218         mutex_lock(&dyn_info_mutex);
3219         r = sprintf(buf, "%ld ", *p);
3220
3221         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3222         buf[r++] = '\n';
3223
3224         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3225
3226         mutex_unlock(&dyn_info_mutex);
3227
3228         return r;
3229 }
3230
3231 static struct file_operations tracing_dyn_info_fops = {
3232         .open           = tracing_open_generic,
3233         .read           = tracing_read_dyn_info,
3234 };
3235 #endif
3236
3237 static struct dentry *d_tracer;
3238
3239 struct dentry *tracing_init_dentry(void)
3240 {
3241         static int once;
3242
3243         if (d_tracer)
3244                 return d_tracer;
3245
3246         d_tracer = debugfs_create_dir("tracing", NULL);
3247
3248         if (!d_tracer && !once) {
3249                 once = 1;
3250                 pr_warning("Could not create debugfs directory 'tracing'\n");
3251                 return NULL;
3252         }
3253
3254         return d_tracer;
3255 }
3256
3257 #ifdef CONFIG_FTRACE_SELFTEST
3258 /* Let selftest have access to static functions in this file */
3259 #include "trace_selftest.c"
3260 #endif
3261
3262 static __init int tracer_init_debugfs(void)
3263 {
3264         struct dentry *d_tracer;
3265         struct dentry *entry;
3266
3267         d_tracer = tracing_init_dentry();
3268
3269         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3270                                     &global_trace, &tracing_ctrl_fops);
3271         if (!entry)
3272                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3273
3274         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3275                                     NULL, &tracing_iter_fops);
3276         if (!entry)
3277                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3278
3279         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3280                                     NULL, &tracing_cpumask_fops);
3281         if (!entry)
3282                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3283
3284         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3285                                     &global_trace, &tracing_lt_fops);
3286         if (!entry)
3287                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3288
3289         entry = debugfs_create_file("trace", 0444, d_tracer,
3290                                     &global_trace, &tracing_fops);
3291         if (!entry)
3292                 pr_warning("Could not create debugfs 'trace' entry\n");
3293
3294         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3295                                     &global_trace, &show_traces_fops);
3296         if (!entry)
3297                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3298
3299         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3300                                     &global_trace, &set_tracer_fops);
3301         if (!entry)
3302                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3303
3304         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3305                                     &tracing_max_latency,
3306                                     &tracing_max_lat_fops);
3307         if (!entry)
3308                 pr_warning("Could not create debugfs "
3309                            "'tracing_max_latency' entry\n");
3310
3311         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3312                                     &tracing_thresh, &tracing_max_lat_fops);
3313         if (!entry)
3314                 pr_warning("Could not create debugfs "
3315                            "'tracing_thresh' entry\n");
3316         entry = debugfs_create_file("README", 0644, d_tracer,
3317                                     NULL, &tracing_readme_fops);
3318         if (!entry)
3319                 pr_warning("Could not create debugfs 'README' entry\n");
3320
3321         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3322                                     NULL, &tracing_pipe_fops);
3323         if (!entry)
3324                 pr_warning("Could not create debugfs "
3325                            "'trace_pipe' entry\n");
3326
3327         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3328                                     &global_trace, &tracing_entries_fops);
3329         if (!entry)
3330                 pr_warning("Could not create debugfs "
3331                            "'buffer_size_kb' entry\n");
3332
3333         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3334                                     NULL, &tracing_mark_fops);
3335         if (!entry)
3336                 pr_warning("Could not create debugfs "
3337                            "'trace_marker' entry\n");
3338
3339 #ifdef CONFIG_DYNAMIC_FTRACE
3340         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3341                                     &ftrace_update_tot_cnt,
3342                                     &tracing_dyn_info_fops);
3343         if (!entry)
3344                 pr_warning("Could not create debugfs "
3345                            "'dyn_ftrace_total_info' entry\n");
3346 #endif
3347 #ifdef CONFIG_SYSPROF_TRACER
3348         init_tracer_sysprof_debugfs(d_tracer);
3349 #endif
3350         return 0;
3351 }
3352
3353 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3354 {
3355         static DEFINE_SPINLOCK(trace_buf_lock);
3356         static char trace_buf[TRACE_BUF_SIZE];
3357
3358         struct ring_buffer_event *event;
3359         struct trace_array *tr = &global_trace;
3360         struct trace_array_cpu *data;
3361         struct print_entry *entry;
3362         unsigned long flags, irq_flags;
3363         int cpu, len = 0, size, pc;
3364
3365         if (tracing_disabled)
3366                 return 0;
3367
3368         pc = preempt_count();
3369         preempt_disable_notrace();
3370         cpu = raw_smp_processor_id();
3371         data = tr->data[cpu];
3372
3373         if (unlikely(atomic_read(&data->disabled)))
3374                 goto out;
3375
3376         spin_lock_irqsave(&trace_buf_lock, flags);
3377         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3378
3379         len = min(len, TRACE_BUF_SIZE-1);
3380         trace_buf[len] = 0;
3381
3382         size = sizeof(*entry) + len + 1;
3383         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3384         if (!event)
3385                 goto out_unlock;
3386         entry = ring_buffer_event_data(event);
3387         tracing_generic_entry_update(&entry->ent, flags, pc);
3388         entry->ent.type                 = TRACE_PRINT;
3389         entry->ip                       = ip;
3390
3391         memcpy(&entry->buf, trace_buf, len);
3392         entry->buf[len] = 0;
3393         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3394
3395  out_unlock:
3396         spin_unlock_irqrestore(&trace_buf_lock, flags);
3397
3398  out:
3399         preempt_enable_notrace();
3400
3401         return len;
3402 }
3403 EXPORT_SYMBOL_GPL(trace_vprintk);
3404
3405 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3406 {
3407         int ret;
3408         va_list ap;
3409
3410         if (!(trace_flags & TRACE_ITER_PRINTK))
3411                 return 0;
3412
3413         va_start(ap, fmt);
3414         ret = trace_vprintk(ip, fmt, ap);
3415         va_end(ap);
3416         return ret;
3417 }
3418 EXPORT_SYMBOL_GPL(__ftrace_printk);
3419
3420 static int trace_panic_handler(struct notifier_block *this,
3421                                unsigned long event, void *unused)
3422 {
3423         if (ftrace_dump_on_oops)
3424                 ftrace_dump();
3425         return NOTIFY_OK;
3426 }
3427
3428 static struct notifier_block trace_panic_notifier = {
3429         .notifier_call  = trace_panic_handler,
3430         .next           = NULL,
3431         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3432 };
3433
3434 static int trace_die_handler(struct notifier_block *self,
3435                              unsigned long val,
3436                              void *data)
3437 {
3438         switch (val) {
3439         case DIE_OOPS:
3440                 if (ftrace_dump_on_oops)
3441                         ftrace_dump();
3442                 break;
3443         default:
3444                 break;
3445         }
3446         return NOTIFY_OK;
3447 }
3448
3449 static struct notifier_block trace_die_notifier = {
3450         .notifier_call = trace_die_handler,
3451         .priority = 200
3452 };
3453
3454 /*
3455  * printk is set to max of 1024, we really don't need it that big.
3456  * Nothing should be printing 1000 characters anyway.
3457  */
3458 #define TRACE_MAX_PRINT         1000
3459
3460 /*
3461  * Define here KERN_TRACE so that we have one place to modify
3462  * it if we decide to change what log level the ftrace dump
3463  * should be at.
3464  */
3465 #define KERN_TRACE              KERN_INFO
3466
3467 static void
3468 trace_printk_seq(struct trace_seq *s)
3469 {
3470         /* Probably should print a warning here. */
3471         if (s->len >= 1000)
3472                 s->len = 1000;
3473
3474         /* should be zero ended, but we are paranoid. */
3475         s->buffer[s->len] = 0;
3476
3477         printk(KERN_TRACE "%s", s->buffer);
3478
3479         trace_seq_reset(s);
3480 }
3481
3482 void ftrace_dump(void)
3483 {
3484         static DEFINE_SPINLOCK(ftrace_dump_lock);
3485         /* use static because iter can be a bit big for the stack */
3486         static struct trace_iterator iter;
3487         static cpumask_t mask;
3488         static int dump_ran;
3489         unsigned long flags;
3490         int cnt = 0, cpu;
3491
3492         /* only one dump */
3493         spin_lock_irqsave(&ftrace_dump_lock, flags);
3494         if (dump_ran)
3495                 goto out;
3496
3497         dump_ran = 1;
3498
3499         /* No turning back! */
3500         ftrace_kill();
3501
3502         for_each_tracing_cpu(cpu) {
3503                 atomic_inc(&global_trace.data[cpu]->disabled);
3504         }
3505
3506         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3507
3508         iter.tr = &global_trace;
3509         iter.trace = current_trace;
3510
3511         /*
3512          * We need to stop all tracing on all CPUS to read the
3513          * the next buffer. This is a bit expensive, but is
3514          * not done often. We fill all what we can read,
3515          * and then release the locks again.
3516          */
3517
3518         cpus_clear(mask);
3519
3520         while (!trace_empty(&iter)) {
3521
3522                 if (!cnt)
3523                         printk(KERN_TRACE "---------------------------------\n");
3524
3525                 cnt++;
3526
3527                 /* reset all but tr, trace, and overruns */
3528                 memset(&iter.seq, 0,
3529                        sizeof(struct trace_iterator) -
3530                        offsetof(struct trace_iterator, seq));
3531                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3532                 iter.pos = -1;
3533
3534                 if (find_next_entry_inc(&iter) != NULL) {
3535                         print_trace_line(&iter);
3536                         trace_consume(&iter);
3537                 }
3538
3539                 trace_printk_seq(&iter.seq);
3540         }
3541
3542         if (!cnt)
3543                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3544         else
3545                 printk(KERN_TRACE "---------------------------------\n");
3546
3547  out:
3548         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3549 }
3550
3551 __init static int tracer_alloc_buffers(void)
3552 {
3553         struct trace_array_cpu *data;
3554         int i;
3555
3556         /* TODO: make the number of buffers hot pluggable with CPUS */
3557         tracing_buffer_mask = cpu_possible_map;
3558
3559         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3560                                                    TRACE_BUFFER_FLAGS);
3561         if (!global_trace.buffer) {
3562                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3563                 WARN_ON(1);
3564                 return 0;
3565         }
3566         global_trace.entries = ring_buffer_size(global_trace.buffer);
3567
3568 #ifdef CONFIG_TRACER_MAX_TRACE
3569         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3570                                              TRACE_BUFFER_FLAGS);
3571         if (!max_tr.buffer) {
3572                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3573                 WARN_ON(1);
3574                 ring_buffer_free(global_trace.buffer);
3575                 return 0;
3576         }
3577         max_tr.entries = ring_buffer_size(max_tr.buffer);
3578         WARN_ON(max_tr.entries != global_trace.entries);
3579 #endif
3580
3581         /* Allocate the first page for all buffers */
3582         for_each_tracing_cpu(i) {
3583                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3584                 max_tr.data[i] = &per_cpu(max_data, i);
3585         }
3586
3587         trace_init_cmdlines();
3588
3589         register_tracer(&nop_trace);
3590 #ifdef CONFIG_BOOT_TRACER
3591         register_tracer(&boot_tracer);
3592         current_trace = &boot_tracer;
3593         current_trace->init(&global_trace);
3594 #else
3595         current_trace = &nop_trace;
3596 #endif
3597
3598         /* All seems OK, enable tracing */
3599         tracing_disabled = 0;
3600
3601         atomic_notifier_chain_register(&panic_notifier_list,
3602                                        &trace_panic_notifier);
3603
3604         register_die_notifier(&trace_die_notifier);
3605
3606         return 0;
3607 }
3608 early_initcall(tracer_alloc_buffers);
3609 fs_initcall(tracer_init_debugfs);