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/seq_file.h>
  34 #include <linux/writeback.h>
  35
  36 #include <linux/stacktrace.h>
  37 #include <linux/ring_buffer.h>
  38 #include <linux/irqflags.h>
  39
  40 #include "trace.h"
  41
  42 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
  43
  44 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  45 unsigned long __read_mostly     tracing_thresh;
  46
  47 /* For tracers that don't implement custom flags */
  48 static struct tracer_opt dummy_tracer_opt[] = {
  49         { }
  50 };
  51
  52 static struct tracer_flags dummy_tracer_flags = {
  53         .val = 0,
  54         .opts = dummy_tracer_opt
  55 };
  56
  57 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
  58 {
  59         return 0;
  60 }
  61
  62 /*
  63  * Kill all tracing for good (never come back).
  64  * It is initialized to 1 but will turn to zero if the initialization
  65  * of the tracer is successful. But that is the only place that sets
  66  * this back to zero.
  67  */
  68 int tracing_disabled = 1;
  69
  70 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
  71
  72 static inline void ftrace_disable_cpu(void)
  73 {
  74         preempt_disable();
  75         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
  76 }
  77
  78 static inline void ftrace_enable_cpu(void)
  79 {
  80         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
  81         preempt_enable();
  82 }
  83
  84 static cpumask_t __read_mostly          tracing_buffer_mask;
  85
  86 #define for_each_tracing_cpu(cpu)       \
  87         for_each_cpu_mask(cpu, tracing_buffer_mask)
  88
  89 /*
  90  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
  91  *
  92  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
  93  * is set, then ftrace_dump is called. This will output the contents
  94  * of the ftrace buffers to the console.  This is very useful for
  95  * capturing traces that lead to crashes and outputing it to a
  96  * serial console.
  97  *
  98  * It is default off, but you can enable it with either specifying
  99  * "ftrace_dump_on_oops" in the kernel command line, or setting
 100  * /proc/sys/kernel/ftrace_dump_on_oops to true.
 101  */
 102 int ftrace_dump_on_oops;
 103
 104 static int tracing_set_tracer(char *buf);
 105
 106 static int __init set_ftrace(char *str)
 107 {
 108         tracing_set_tracer(str);
 109         return 1;
 110 }
 111 __setup("ftrace", set_ftrace);
 112
 113 static int __init set_ftrace_dump_on_oops(char *str)
 114 {
 115         ftrace_dump_on_oops = 1;
 116         return 1;
 117 }
 118 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
 119
 120 long
 121 ns2usecs(cycle_t nsec)
 122 {
 123         nsec += 500;
 124         do_div(nsec, 1000);
 125         return nsec;
 126 }
 127
 128 cycle_t ftrace_now(int cpu)
 129 {
 130         u64 ts = ring_buffer_time_stamp(cpu);
 131         ring_buffer_normalize_time_stamp(cpu, &ts);
 132         return ts;
 133 }
 134
 135 /*
 136  * The global_trace is the descriptor that holds the tracing
 137  * buffers for the live tracing. For each CPU, it contains
 138  * a link list of pages that will store trace entries. The
 139  * page descriptor of the pages in the memory is used to hold
 140  * the link list by linking the lru item in the page descriptor
 141  * to each of the pages in the buffer per CPU.
 142  *
 143  * For each active CPU there is a data field that holds the
 144  * pages for the buffer for that CPU. Each CPU has the same number
 145  * of pages allocated for its buffer.
 146  */
 147 static struct trace_array       global_trace;
 148
 149 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
 150
 151 /*
 152  * The max_tr is used to snapshot the global_trace when a maximum
 153  * latency is reached. Some tracers will use this to store a maximum
 154  * trace while it continues examining live traces.
 155  *
 156  * The buffers for the max_tr are set up the same as the global_trace.
 157  * When a snapshot is taken, the link list of the max_tr is swapped
 158  * with the link list of the global_trace and the buffers are reset for
 159  * the global_trace so the tracing can continue.
 160  */
 161 static struct trace_array       max_tr;
 162
 163 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 164
 165 /* tracer_enabled is used to toggle activation of a tracer */
 166 static int                      tracer_enabled = 1;
 167
 168 /**
 169  * tracing_is_enabled - return tracer_enabled status
 170  *
 171  * This function is used by other tracers to know the status
 172  * of the tracer_enabled flag.  Tracers may use this function
 173  * to know if it should enable their features when starting
 174  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
 175  */
 176 int tracing_is_enabled(void)
 177 {
 178         return tracer_enabled;
 179 }
 180
 181 /* function tracing enabled */
 182 int                             ftrace_function_enabled;
 183
 184 /*
 185  * trace_buf_size is the size in bytes that is allocated
 186  * for a buffer. Note, the number of bytes is always rounded
 187  * to page size.
 188  *
 189  * This number is purposely set to a low number of 16384.
 190  * If the dump on oops happens, it will be much appreciated
 191  * to not have to wait for all that output. Anyway this can be
 192  * boot time and run time configurable.
 193  */
 194 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
 195
 196 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
 197
 198 /* trace_types holds a link list of available tracers. */
 199 static struct tracer            *trace_types __read_mostly;
 200
 201 /* current_trace points to the tracer that is currently active */
 202 static struct tracer            *current_trace __read_mostly;
 203
 204 /*
 205  * max_tracer_type_len is used to simplify the allocating of
 206  * buffers to read userspace tracer names. We keep track of
 207  * the longest tracer name registered.
 208  */
 209 static int                      max_tracer_type_len;
 210
 211 /*
 212  * trace_types_lock is used to protect the trace_types list.
 213  * This lock is also used to keep user access serialized.
 214  * Accesses from userspace will grab this lock while userspace
 215  * activities happen inside the kernel.
 216  */
 217 static DEFINE_MUTEX(trace_types_lock);
 218
 219 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
 220 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 221
 222 /* trace_flags holds trace_options default values */
 223 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
 224         TRACE_ITER_ANNOTATE;
 225
 226 /**
 227  * trace_wake_up - wake up tasks waiting for trace input
 228  *
 229  * Simply wakes up any task that is blocked on the trace_wait
 230  * queue. These is used with trace_poll for tasks polling the trace.
 231  */
 232 void trace_wake_up(void)
 233 {
 234         /*
 235          * The runqueue_is_locked() can fail, but this is the best we
 236          * have for now:
 237          */
 238         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
 239                 wake_up(&trace_wait);
 240 }
 241
 242 static int __init set_buf_size(char *str)
 243 {
 244         unsigned long buf_size;
 245         int ret;
 246
 247         if (!str)
 248                 return 0;
 249         ret = strict_strtoul(str, 0, &buf_size);
 250         /* nr_entries can not be zero */
 251         if (ret < 0 || buf_size == 0)
 252                 return 0;
 253         trace_buf_size = buf_size;
 254         return 1;
 255 }
 256 __setup("trace_buf_size=", set_buf_size);
 257
 258 unsigned long nsecs_to_usecs(unsigned long nsecs)
 259 {
 260         return nsecs / 1000;
 261 }
 262
 263 /* These must match the bit postions in trace_iterator_flags */
 264 static const char *trace_options[] = {
 265         "print-parent",
 266         "sym-offset",
 267         "sym-addr",
 268         "verbose",
 269         "raw",
 270         "hex",
 271         "bin",
 272         "block",
 273         "stacktrace",
 274         "sched-tree",
 275         "ftrace_printk",
 276         "ftrace_preempt",
 277         "branch",
 278         "annotate",
 279         "userstacktrace",
 280         "sym-userobj",
 281         NULL
 282 };
 283
 284 /*
 285  * ftrace_max_lock is used to protect the swapping of buffers
 286  * when taking a max snapshot. The buffers themselves are
 287  * protected by per_cpu spinlocks. But the action of the swap
 288  * needs its own lock.
 289  *
 290  * This is defined as a raw_spinlock_t in order to help
 291  * with performance when lockdep debugging is enabled.
 292  */
 293 static raw_spinlock_t ftrace_max_lock =
 294         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 295
 296 /*
 297  * Copy the new maximum trace into the separate maximum-trace
 298  * structure. (this way the maximum trace is permanently saved,
 299  * for later retrieval via /debugfs/tracing/latency_trace)
 300  */
 301 static void
 302 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 303 {
 304         struct trace_array_cpu *data = tr->data[cpu];
 305
 306         max_tr.cpu = cpu;
 307         max_tr.time_start = data->preempt_timestamp;
 308
 309         data = max_tr.data[cpu];
 310         data->saved_latency = tracing_max_latency;
 311
 312         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 313         data->pid = tsk->pid;
 314         data->uid = tsk->uid;
 315         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 316         data->policy = tsk->policy;
 317         data->rt_priority = tsk->rt_priority;
 318
 319         /* record this tasks comm */
 320         tracing_record_cmdline(current);
 321 }
 322
 323 /**
 324  * trace_seq_printf - sequence printing of trace information
 325  * @s: trace sequence descriptor
 326  * @fmt: printf format string
 327  *
 328  * The tracer may use either sequence operations or its own
 329  * copy to user routines. To simplify formating of a trace
 330  * trace_seq_printf is used to store strings into a special
 331  * buffer (@s). Then the output may be either used by
 332  * the sequencer or pulled into another buffer.
 333  */
 334 int
 335 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 336 {
 337         int len = (PAGE_SIZE - 1) - s->len;
 338         va_list ap;
 339         int ret;
 340
 341         if (!len)
 342                 return 0;
 343
 344         va_start(ap, fmt);
 345         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 346         va_end(ap);
 347
 348         /* If we can't write it all, don't bother writing anything */
 349         if (ret >= len)
 350                 return 0;
 351
 352         s->len += ret;
 353
 354         return len;
 355 }
 356
 357 /**
 358  * trace_seq_puts - trace sequence printing of simple string
 359  * @s: trace sequence descriptor
 360  * @str: simple string to record
 361  *
 362  * The tracer may use either the sequence operations or its own
 363  * copy to user routines. This function records a simple string
 364  * into a special buffer (@s) for later retrieval by a sequencer
 365  * or other mechanism.
 366  */
 367 static int
 368 trace_seq_puts(struct trace_seq *s, const char *str)
 369 {
 370         int len = strlen(str);
 371
 372         if (len > ((PAGE_SIZE - 1) - s->len))
 373                 return 0;
 374
 375         memcpy(s->buffer + s->len, str, len);
 376         s->len += len;
 377
 378         return len;
 379 }
 380
 381 static int
 382 trace_seq_putc(struct trace_seq *s, unsigned char c)
 383 {
 384         if (s->len >= (PAGE_SIZE - 1))
 385                 return 0;
 386
 387         s->buffer[s->len++] = c;
 388
 389         return 1;
 390 }
 391
 392 static int
 393 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 394 {
 395         if (len > ((PAGE_SIZE - 1) - s->len))
 396                 return 0;
 397
 398         memcpy(s->buffer + s->len, mem, len);
 399         s->len += len;
 400
 401         return len;
 402 }
 403
 404 #define MAX_MEMHEX_BYTES        8
 405 #define HEX_CHARS               (MAX_MEMHEX_BYTES*2 + 1)
 406
 407 static int
 408 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 409 {
 410         unsigned char hex[HEX_CHARS];
 411         unsigned char *data = mem;
 412         int i, j;
 413
 414 #ifdef __BIG_ENDIAN
 415         for (i = 0, j = 0; i < len; i++) {
 416 #else
 417         for (i = len-1, j = 0; i >= 0; i--) {
 418 #endif
 419                 hex[j++] = hex_asc_hi(data[i]);
 420                 hex[j++] = hex_asc_lo(data[i]);
 421         }
 422         hex[j++] = ' ';
 423
 424         return trace_seq_putmem(s, hex, j);
 425 }
 426
 427 static int
 428 trace_seq_path(struct trace_seq *s, struct path *path)
 429 {
 430         unsigned char *p;
 431
 432         if (s->len >= (PAGE_SIZE - 1))
 433                 return 0;
 434         p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
 435         if (!IS_ERR(p)) {
 436                 p = mangle_path(s->buffer + s->len, p, "\n");
 437                 if (p) {
 438                         s->len = p - s->buffer;
 439                         return 1;
 440                 }
 441         } else {
 442                 s->buffer[s->len++] = '?';
 443                 return 1;
 444         }
 445
 446         return 0;
 447 }
 448
 449 static void
 450 trace_seq_reset(struct trace_seq *s)
 451 {
 452         s->len = 0;
 453         s->readpos = 0;
 454 }
 455
 456 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
 457 {
 458         int len;
 459         int ret;
 460
 461         if (s->len <= s->readpos)
 462                 return -EBUSY;
 463
 464         len = s->len - s->readpos;
 465         if (cnt > len)
 466                 cnt = len;
 467         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
 468         if (ret)
 469                 return -EFAULT;
 470
 471         s->readpos += len;
 472         return cnt;
 473 }
 474
 475 static void
 476 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 477 {
 478         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 479
 480         s->buffer[len] = 0;
 481         seq_puts(m, s->buffer);
 482
 483         trace_seq_reset(s);
 484 }
 485
 486 /**
 487  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
 488  * @tr: tracer
 489  * @tsk: the task with the latency
 490  * @cpu: The cpu that initiated the trace.
 491  *
 492  * Flip the buffers between the @tr and the max_tr and record information
 493  * about which task was the cause of this latency.
 494  */
 495 void
 496 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 497 {
 498         struct ring_buffer *buf = tr->buffer;
 499
 500         WARN_ON_ONCE(!irqs_disabled());
 501         __raw_spin_lock(&ftrace_max_lock);
 502
 503         tr->buffer = max_tr.buffer;
 504         max_tr.buffer = buf;
 505
 506         ftrace_disable_cpu();
 507         ring_buffer_reset(tr->buffer);
 508         ftrace_enable_cpu();
 509
 510         __update_max_tr(tr, tsk, cpu);
 511         __raw_spin_unlock(&ftrace_max_lock);
 512 }
 513
 514 /**
 515  * update_max_tr_single - only copy one trace over, and reset the rest
 516  * @tr - tracer
 517  * @tsk - task with the latency
 518  * @cpu - the cpu of the buffer to copy.
 519  *
 520  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
 521  */
 522 void
 523 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 524 {
 525         int ret;
 526
 527         WARN_ON_ONCE(!irqs_disabled());
 528         __raw_spin_lock(&ftrace_max_lock);
 529
 530         ftrace_disable_cpu();
 531
 532         ring_buffer_reset(max_tr.buffer);
 533         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
 534
 535         ftrace_enable_cpu();
 536
 537         WARN_ON_ONCE(ret);
 538
 539         __update_max_tr(tr, tsk, cpu);
 540         __raw_spin_unlock(&ftrace_max_lock);
 541 }
 542
 543 /**
 544  * register_tracer - register a tracer with the ftrace system.
 545  * @type - the plugin for the tracer
 546  *
 547  * Register a new plugin tracer.
 548  */
 549 int register_tracer(struct tracer *type)
 550 {
 551         struct tracer *t;
 552         int len;
 553         int ret = 0;
 554
 555         if (!type->name) {
 556                 pr_info("Tracer must have a name\n");
 557                 return -1;
 558         }
 559
 560         /*
 561          * When this gets called we hold the BKL which means that
 562          * preemption is disabled. Various trace selftests however
 563          * need to disable and enable preemption for successful tests.
 564          * So we drop the BKL here and grab it after the tests again.
 565          */
 566         unlock_kernel();
 567         mutex_lock(&trace_types_lock);
 568
 569         for (t = trace_types; t; t = t->next) {
 570                 if (strcmp(type->name, t->name) == 0) {
 571                         /* already found */
 572                         pr_info("Trace %s already registered\n",
 573                                 type->name);
 574                         ret = -1;
 575                         goto out;
 576                 }
 577         }
 578
 579         if (!type->set_flag)
 580                 type->set_flag = &dummy_set_flag;
 581         if (!type->flags)
 582                 type->flags = &dummy_tracer_flags;
 583         else
 584                 if (!type->flags->opts)
 585                         type->flags->opts = dummy_tracer_opt;
 586
 587 #ifdef CONFIG_FTRACE_STARTUP_TEST
 588         if (type->selftest) {
 589                 struct tracer *saved_tracer = current_trace;
 590                 struct trace_array *tr = &global_trace;
 591                 int i;
 592                 /*
 593                  * Run a selftest on this tracer.
 594                  * Here we reset the trace buffer, and set the current
 595                  * tracer to be this tracer. The tracer can then run some
 596                  * internal tracing to verify that everything is in order.
 597                  * If we fail, we do not register this tracer.
 598                  */
 599                 for_each_tracing_cpu(i)
 600                         tracing_reset(tr, i);
 601
 602                 current_trace = type;
 603                 /* the test is responsible for initializing and enabling */
 604                 pr_info("Testing tracer %s: ", type->name);
 605                 ret = type->selftest(type, tr);
 606                 /* the test is responsible for resetting too */
 607                 current_trace = saved_tracer;
 608                 if (ret) {
 609                         printk(KERN_CONT "FAILED!\n");
 610                         goto out;
 611                 }
 612                 /* Only reset on passing, to avoid touching corrupted buffers */
 613                 for_each_tracing_cpu(i)
 614                         tracing_reset(tr, i);
 615
 616                 printk(KERN_CONT "PASSED\n");
 617         }
 618 #endif
 619
 620         type->next = trace_types;
 621         trace_types = type;
 622         len = strlen(type->name);
 623         if (len > max_tracer_type_len)
 624                 max_tracer_type_len = len;
 625
 626  out:
 627         mutex_unlock(&trace_types_lock);
 628         lock_kernel();
 629
 630         return ret;
 631 }
 632
 633 void unregister_tracer(struct tracer *type)
 634 {
 635         struct tracer **t;
 636         int len;
 637
 638         mutex_lock(&trace_types_lock);
 639         for (t = &trace_types; *t; t = &(*t)->next) {
 640                 if (*t == type)
 641                         goto found;
 642         }
 643         pr_info("Trace %s not registered\n", type->name);
 644         goto out;
 645
 646  found:
 647         *t = (*t)->next;
 648         if (strlen(type->name) != max_tracer_type_len)
 649                 goto out;
 650
 651         max_tracer_type_len = 0;
 652         for (t = &trace_types; *t; t = &(*t)->next) {
 653                 len = strlen((*t)->name);
 654                 if (len > max_tracer_type_len)
 655                         max_tracer_type_len = len;
 656         }
 657  out:
 658         mutex_unlock(&trace_types_lock);
 659 }
 660
 661 void tracing_reset(struct trace_array *tr, int cpu)
 662 {
 663         ftrace_disable_cpu();
 664         ring_buffer_reset_cpu(tr->buffer, cpu);
 665         ftrace_enable_cpu();
 666 }
 667
 668 #define SAVED_CMDLINES 128
 669 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 670 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 671 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 672 static int cmdline_idx;
 673 static DEFINE_SPINLOCK(trace_cmdline_lock);
 674
 675 /* temporary disable recording */
 676 atomic_t trace_record_cmdline_disabled __read_mostly;
 677
 678 static void trace_init_cmdlines(void)
 679 {
 680         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 681         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 682         cmdline_idx = 0;
 683 }
 684
 685 static int trace_stop_count;
 686 static DEFINE_SPINLOCK(tracing_start_lock);
 687
 688 /**
 689  * ftrace_off_permanent - disable all ftrace code permanently
 690  *
 691  * This should only be called when a serious anomally has
 692  * been detected.  This will turn off the function tracing,
 693  * ring buffers, and other tracing utilites. It takes no
 694  * locks and can be called from any context.
 695  */
 696 void ftrace_off_permanent(void)
 697 {
 698         tracing_disabled = 1;
 699         ftrace_stop();
 700         tracing_off_permanent();
 701 }
 702
 703 /**
 704  * tracing_start - quick start of the tracer
 705  *
 706  * If tracing is enabled but was stopped by tracing_stop,
 707  * this will start the tracer back up.
 708  */
 709 void tracing_start(void)
 710 {
 711         struct ring_buffer *buffer;
 712         unsigned long flags;
 713
 714         if (tracing_disabled)
 715                 return;
 716
 717         spin_lock_irqsave(&tracing_start_lock, flags);
 718         if (--trace_stop_count)
 719                 goto out;
 720
 721         if (trace_stop_count < 0) {
 722                 /* Someone screwed up their debugging */
 723                 WARN_ON_ONCE(1);
 724                 trace_stop_count = 0;
 725                 goto out;
 726         }
 727
 728
 729         buffer = global_trace.buffer;
 730         if (buffer)
 731                 ring_buffer_record_enable(buffer);
 732
 733         buffer = max_tr.buffer;
 734         if (buffer)
 735                 ring_buffer_record_enable(buffer);
 736
 737         ftrace_start();
 738  out:
 739         spin_unlock_irqrestore(&tracing_start_lock, flags);
 740 }
 741
 742 /**
 743  * tracing_stop - quick stop of the tracer
 744  *
 745  * Light weight way to stop tracing. Use in conjunction with
 746  * tracing_start.
 747  */
 748 void tracing_stop(void)
 749 {
 750         struct ring_buffer *buffer;
 751         unsigned long flags;
 752
 753         ftrace_stop();
 754         spin_lock_irqsave(&tracing_start_lock, flags);
 755         if (trace_stop_count++)
 756                 goto out;
 757
 758         buffer = global_trace.buffer;
 759         if (buffer)
 760                 ring_buffer_record_disable(buffer);
 761
 762         buffer = max_tr.buffer;
 763         if (buffer)
 764                 ring_buffer_record_disable(buffer);
 765
 766  out:
 767         spin_unlock_irqrestore(&tracing_start_lock, flags);
 768 }
 769
 770 void trace_stop_cmdline_recording(void);
 771
 772 static void trace_save_cmdline(struct task_struct *tsk)
 773 {
 774         unsigned map;
 775         unsigned idx;
 776
 777         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 778                 return;
 779
 780         /*
 781          * It's not the end of the world if we don't get
 782          * the lock, but we also don't want to spin
 783          * nor do we want to disable interrupts,
 784          * so if we miss here, then better luck next time.
 785          */
 786         if (!spin_trylock(&trace_cmdline_lock))
 787                 return;
 788
 789         idx = map_pid_to_cmdline[tsk->pid];
 790         if (idx >= SAVED_CMDLINES) {
 791                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 792
 793                 map = map_cmdline_to_pid[idx];
 794                 if (map <= PID_MAX_DEFAULT)
 795                         map_pid_to_cmdline[map] = (unsigned)-1;
 796
 797                 map_pid_to_cmdline[tsk->pid] = idx;
 798
 799                 cmdline_idx = idx;
 800         }
 801
 802         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 803
 804         spin_unlock(&trace_cmdline_lock);
 805 }
 806
 807 char *trace_find_cmdline(int pid)
 808 {
 809         char *cmdline = "<...>";
 810         unsigned map;
 811
 812         if (!pid)
 813                 return "<idle>";
 814
 815         if (pid > PID_MAX_DEFAULT)
 816                 goto out;
 817
 818         map = map_pid_to_cmdline[pid];
 819         if (map >= SAVED_CMDLINES)
 820                 goto out;
 821
 822         cmdline = saved_cmdlines[map];
 823
 824  out:
 825         return cmdline;
 826 }
 827
 828 void tracing_record_cmdline(struct task_struct *tsk)
 829 {
 830         if (atomic_read(&trace_record_cmdline_disabled))
 831                 return;
 832
 833         trace_save_cmdline(tsk);
 834 }
 835
 836 void
 837 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
 838                              int pc)
 839 {
 840         struct task_struct *tsk = current;
 841
 842         entry->preempt_count            = pc & 0xff;
 843         entry->pid                      = (tsk) ? tsk->pid : 0;
 844         entry->tgid                     = (tsk) ? tsk->tgid : 0;
 845         entry->flags =
 846 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
 847                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 848 #else
 849                 TRACE_FLAG_IRQS_NOSUPPORT |
 850 #endif
 851                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 852                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 853                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 854 }
 855
 856 void
 857 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 858                unsigned long ip, unsigned long parent_ip, unsigned long flags,
 859                int pc)
 860 {
 861         struct ring_buffer_event *event;
 862         struct ftrace_entry *entry;
 863         unsigned long irq_flags;
 864
 865         /* If we are reading the ring buffer, don't trace */
 866         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 867                 return;
 868
 869         event = ring_buffer_lock_reserve(tr->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;
 876         entry->ip                       = ip;
 877         entry->parent_ip                = parent_ip;
 878         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 879 }
 880
 881 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 882 static void __trace_graph_entry(struct trace_array *tr,
 883                                 struct trace_array_cpu *data,
 884                                 struct ftrace_graph_ent *trace,
 885                                 unsigned long flags,
 886                                 int pc)
 887 {
 888         struct ring_buffer_event *event;
 889         struct ftrace_graph_ent_entry *entry;
 890         unsigned long irq_flags;
 891
 892         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 893                 return;
 894
 895         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
 896                                          &irq_flags);
 897         if (!event)
 898                 return;
 899         entry   = ring_buffer_event_data(event);
 900         tracing_generic_entry_update(&entry->ent, flags, pc);
 901         entry->ent.type                 = TRACE_GRAPH_ENT;
 902         entry->graph_ent                        = *trace;
 903         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
 904 }
 905
 906 static void __trace_graph_return(struct trace_array *tr,
 907                                 struct trace_array_cpu *data,
 908                                 struct ftrace_graph_ret *trace,
 909                                 unsigned long flags,
 910                                 int pc)
 911 {
 912         struct ring_buffer_event *event;
 913         struct ftrace_graph_ret_entry *entry;
 914         unsigned long irq_flags;
 915
 916         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 917                 return;
 918
 919         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
 920                                          &irq_flags);
 921         if (!event)
 922                 return;
 923         entry   = ring_buffer_event_data(event);
 924         tracing_generic_entry_update(&entry->ent, flags, pc);
 925         entry->ent.type                 = TRACE_GRAPH_RET;
 926         entry->ret                              = *trace;
 927         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
 928 }
 929 #endif
 930
 931 void
 932 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 933        unsigned long ip, unsigned long parent_ip, unsigned long flags,
 934        int pc)
 935 {
 936         if (likely(!atomic_read(&data->disabled)))
 937                 trace_function(tr, data, ip, parent_ip, flags, pc);
 938 }
 939
 940 static void ftrace_trace_stack(struct trace_array *tr,
 941                                struct trace_array_cpu *data,
 942                                unsigned long flags,
 943                                int skip, int pc)
 944 {
 945 #ifdef CONFIG_STACKTRACE
 946         struct ring_buffer_event *event;
 947         struct stack_entry *entry;
 948         struct stack_trace trace;
 949         unsigned long irq_flags;
 950
 951         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 952                 return;
 953
 954         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 955                                          &irq_flags);
 956         if (!event)
 957                 return;
 958         entry   = ring_buffer_event_data(event);
 959         tracing_generic_entry_update(&entry->ent, flags, pc);
 960         entry->ent.type         = TRACE_STACK;
 961
 962         memset(&entry->caller, 0, sizeof(entry->caller));
 963
 964         trace.nr_entries        = 0;
 965         trace.max_entries       = FTRACE_STACK_ENTRIES;
 966         trace.skip              = skip;
 967         trace.entries           = entry->caller;
 968
 969         save_stack_trace(&trace);
 970         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 971 #endif
 972 }
 973
 974 void __trace_stack(struct trace_array *tr,
 975                    struct trace_array_cpu *data,
 976                    unsigned long flags,
 977                    int skip)
 978 {
 979         ftrace_trace_stack(tr, data, flags, skip, preempt_count());
 980 }
 981
 982 static void ftrace_trace_userstack(struct trace_array *tr,
 983                    struct trace_array_cpu *data,
 984                    unsigned long flags, int pc)
 985 {
 986         struct ring_buffer_event *event;
 987         struct userstack_entry *entry;
 988         struct stack_trace trace;
 989         unsigned long irq_flags;
 990
 991         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
 992                 return;
 993
 994         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 995                                          &irq_flags);
 996         if (!event)
 997                 return;
 998         entry   = ring_buffer_event_data(event);
 999         tracing_generic_entry_update(&entry->ent, flags, pc);
1000         entry->ent.type         = TRACE_USER_STACK;
1001
1002         memset(&entry->caller, 0, sizeof(entry->caller));
1003
1004         trace.nr_entries        = 0;
1005         trace.max_entries       = FTRACE_STACK_ENTRIES;
1006         trace.skip              = 0;
1007         trace.entries           = entry->caller;
1008
1009         save_stack_trace_user(&trace);
1010         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1011 }
1012
1013 void __trace_userstack(struct trace_array *tr,
1014                    struct trace_array_cpu *data,
1015                    unsigned long flags)
1016 {
1017         ftrace_trace_userstack(tr, data, flags, preempt_count());
1018 }
1019
1020 static void
1021 ftrace_trace_special(void *__tr, void *__data,
1022                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1023                      int pc)
1024 {
1025         struct ring_buffer_event *event;
1026         struct trace_array_cpu *data = __data;
1027         struct trace_array *tr = __tr;
1028         struct special_entry *entry;
1029         unsigned long irq_flags;
1030
1031         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1032                                          &irq_flags);
1033         if (!event)
1034                 return;
1035         entry   = ring_buffer_event_data(event);
1036         tracing_generic_entry_update(&entry->ent, 0, pc);
1037         entry->ent.type                 = TRACE_SPECIAL;
1038         entry->arg1                     = arg1;
1039         entry->arg2                     = arg2;
1040         entry->arg3                     = arg3;
1041         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1042         ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1043         ftrace_trace_userstack(tr, data, irq_flags, pc);
1044
1045         trace_wake_up();
1046 }
1047
1048 void
1049 __trace_special(void *__tr, void *__data,
1050                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1051 {
1052         ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1053 }
1054
1055 void
1056 tracing_sched_switch_trace(struct trace_array *tr,
1057                            struct trace_array_cpu *data,
1058                            struct task_struct *prev,
1059                            struct task_struct *next,
1060                            unsigned long flags, int pc)
1061 {
1062         struct ring_buffer_event *event;
1063         struct ctx_switch_entry *entry;
1064         unsigned long irq_flags;
1065
1066         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1067                                            &irq_flags);
1068         if (!event)
1069                 return;
1070         entry   = ring_buffer_event_data(event);
1071         tracing_generic_entry_update(&entry->ent, flags, pc);
1072         entry->ent.type                 = TRACE_CTX;
1073         entry->prev_pid                 = prev->pid;
1074         entry->prev_prio                = prev->prio;
1075         entry->prev_state               = prev->state;
1076         entry->next_pid                 = next->pid;
1077         entry->next_prio                = next->prio;
1078         entry->next_state               = next->state;
1079         entry->next_cpu = task_cpu(next);
1080         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1081         ftrace_trace_stack(tr, data, flags, 5, pc);
1082         ftrace_trace_userstack(tr, data, flags, pc);
1083 }
1084
1085 void
1086 tracing_sched_wakeup_trace(struct trace_array *tr,
1087                            struct trace_array_cpu *data,
1088                            struct task_struct *wakee,
1089                            struct task_struct *curr,
1090                            unsigned long flags, int pc)
1091 {
1092         struct ring_buffer_event *event;
1093         struct ctx_switch_entry *entry;
1094         unsigned long irq_flags;
1095
1096         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1097                                            &irq_flags);
1098         if (!event)
1099                 return;
1100         entry   = ring_buffer_event_data(event);
1101         tracing_generic_entry_update(&entry->ent, flags, pc);
1102         entry->ent.type                 = TRACE_WAKE;
1103         entry->prev_pid                 = curr->pid;
1104         entry->prev_prio                = curr->prio;
1105         entry->prev_state               = curr->state;
1106         entry->next_pid                 = wakee->pid;
1107         entry->next_prio                = wakee->prio;
1108         entry->next_state               = wakee->state;
1109         entry->next_cpu                 = task_cpu(wakee);
1110         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1111         ftrace_trace_stack(tr, data, flags, 6, pc);
1112         ftrace_trace_userstack(tr, data, flags, pc);
1113
1114         trace_wake_up();
1115 }
1116
1117 void
1118 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1119 {
1120         struct trace_array *tr = &global_trace;
1121         struct trace_array_cpu *data;
1122         unsigned long flags;
1123         int cpu;
1124         int pc;
1125
1126         if (tracing_disabled)
1127                 return;
1128
1129         pc = preempt_count();
1130         local_irq_save(flags);
1131         cpu = raw_smp_processor_id();
1132         data = tr->data[cpu];
1133
1134         if (likely(atomic_inc_return(&data->disabled) == 1))
1135                 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1136
1137         atomic_dec(&data->disabled);
1138         local_irq_restore(flags);
1139 }
1140
1141 #ifdef CONFIG_FUNCTION_TRACER
1142 static void
1143 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1144 {
1145         struct trace_array *tr = &global_trace;
1146         struct trace_array_cpu *data;
1147         unsigned long flags;
1148         long disabled;
1149         int cpu, resched;
1150         int pc;
1151
1152         if (unlikely(!ftrace_function_enabled))
1153                 return;
1154
1155         pc = preempt_count();
1156         resched = ftrace_preempt_disable();
1157         local_save_flags(flags);
1158         cpu = raw_smp_processor_id();
1159         data = tr->data[cpu];
1160         disabled = atomic_inc_return(&data->disabled);
1161
1162         if (likely(disabled == 1))
1163                 trace_function(tr, data, ip, parent_ip, flags, pc);
1164
1165         atomic_dec(&data->disabled);
1166         ftrace_preempt_enable(resched);
1167 }
1168
1169 static void
1170 function_trace_call(unsigned long ip, unsigned long parent_ip)
1171 {
1172         struct trace_array *tr = &global_trace;
1173         struct trace_array_cpu *data;
1174         unsigned long flags;
1175         long disabled;
1176         int cpu;
1177         int pc;
1178
1179         if (unlikely(!ftrace_function_enabled))
1180                 return;
1181
1182         /*
1183          * Need to use raw, since this must be called before the
1184          * recursive protection is performed.
1185          */
1186         local_irq_save(flags);
1187         cpu = raw_smp_processor_id();
1188         data = tr->data[cpu];
1189         disabled = atomic_inc_return(&data->disabled);
1190
1191         if (likely(disabled == 1)) {
1192                 pc = preempt_count();
1193                 trace_function(tr, data, ip, parent_ip, flags, pc);
1194         }
1195
1196         atomic_dec(&data->disabled);
1197         local_irq_restore(flags);
1198 }
1199
1200 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1201 void trace_graph_entry(struct ftrace_graph_ent *trace)
1202 {
1203         struct trace_array *tr = &global_trace;
1204         struct trace_array_cpu *data;
1205         unsigned long flags;
1206         long disabled;
1207         int cpu;
1208         int pc;
1209
1210         raw_local_irq_save(flags);
1211         cpu = raw_smp_processor_id();
1212         data = tr->data[cpu];
1213         disabled = atomic_inc_return(&data->disabled);
1214         if (likely(disabled == 1)) {
1215                 pc = preempt_count();
1216                 __trace_graph_entry(tr, data, trace, flags, pc);
1217         }
1218         atomic_dec(&data->disabled);
1219         raw_local_irq_restore(flags);
1220 }
1221
1222 void trace_graph_return(struct ftrace_graph_ret *trace)
1223 {
1224         struct trace_array *tr = &global_trace;
1225         struct trace_array_cpu *data;
1226         unsigned long flags;
1227         long disabled;
1228         int cpu;
1229         int pc;
1230
1231         raw_local_irq_save(flags);
1232         cpu = raw_smp_processor_id();
1233         data = tr->data[cpu];
1234         disabled = atomic_inc_return(&data->disabled);
1235         if (likely(disabled == 1)) {
1236                 pc = preempt_count();
1237                 __trace_graph_return(tr, data, trace, flags, pc);
1238         }
1239         atomic_dec(&data->disabled);
1240         raw_local_irq_restore(flags);
1241 }
1242 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1243
1244 static struct ftrace_ops trace_ops __read_mostly =
1245 {
1246         .func = function_trace_call,
1247 };
1248
1249 void tracing_start_function_trace(void)
1250 {
1251         ftrace_function_enabled = 0;
1252
1253         if (trace_flags & TRACE_ITER_PREEMPTONLY)
1254                 trace_ops.func = function_trace_call_preempt_only;
1255         else
1256                 trace_ops.func = function_trace_call;
1257
1258         register_ftrace_function(&trace_ops);
1259         ftrace_function_enabled = 1;
1260 }
1261
1262 void tracing_stop_function_trace(void)
1263 {
1264         ftrace_function_enabled = 0;
1265         unregister_ftrace_function(&trace_ops);
1266 }
1267 #endif
1268
1269 enum trace_file_type {
1270         TRACE_FILE_LAT_FMT      = 1,
1271         TRACE_FILE_ANNOTATE     = 2,
1272 };
1273
1274 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1275 {
1276         /* Don't allow ftrace to trace into the ring buffers */
1277         ftrace_disable_cpu();
1278
1279         iter->idx++;
1280         if (iter->buffer_iter[iter->cpu])
1281                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1282
1283         ftrace_enable_cpu();
1284 }
1285
1286 static struct trace_entry *
1287 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1288 {
1289         struct ring_buffer_event *event;
1290         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1291
1292         /* Don't allow ftrace to trace into the ring buffers */
1293         ftrace_disable_cpu();
1294
1295         if (buf_iter)
1296                 event = ring_buffer_iter_peek(buf_iter, ts);
1297         else
1298                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1299
1300         ftrace_enable_cpu();
1301
1302         return event ? ring_buffer_event_data(event) : NULL;
1303 }
1304
1305 static struct trace_entry *
1306 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1307 {
1308         struct ring_buffer *buffer = iter->tr->buffer;
1309         struct trace_entry *ent, *next = NULL;
1310         u64 next_ts = 0, ts;
1311         int next_cpu = -1;
1312         int cpu;
1313
1314         for_each_tracing_cpu(cpu) {
1315
1316                 if (ring_buffer_empty_cpu(buffer, cpu))
1317                         continue;
1318
1319                 ent = peek_next_entry(iter, cpu, &ts);
1320
1321                 /*
1322                  * Pick the entry with the smallest timestamp:
1323                  */
1324                 if (ent && (!next || ts < next_ts)) {
1325                         next = ent;
1326                         next_cpu = cpu;
1327                         next_ts = ts;
1328                 }
1329         }
1330
1331         if (ent_cpu)
1332                 *ent_cpu = next_cpu;
1333
1334         if (ent_ts)
1335                 *ent_ts = next_ts;
1336
1337         return next;
1338 }
1339
1340 /* Find the next real entry, without updating the iterator itself */
1341 static struct trace_entry *
1342 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1343 {
1344         return __find_next_entry(iter, ent_cpu, ent_ts);
1345 }
1346
1347 /* Find the next real entry, and increment the iterator to the next entry */
1348 static void *find_next_entry_inc(struct trace_iterator *iter)
1349 {
1350         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1351
1352         if (iter->ent)
1353                 trace_iterator_increment(iter, iter->cpu);
1354
1355         return iter->ent ? iter : NULL;
1356 }
1357
1358 static void trace_consume(struct trace_iterator *iter)
1359 {
1360         /* Don't allow ftrace to trace into the ring buffers */
1361         ftrace_disable_cpu();
1362         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1363         ftrace_enable_cpu();
1364 }
1365
1366 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1367 {
1368         struct trace_iterator *iter = m->private;
1369         int i = (int)*pos;
1370         void *ent;
1371
1372         (*pos)++;
1373
1374         /* can't go backwards */
1375         if (iter->idx > i)
1376                 return NULL;
1377
1378         if (iter->idx < 0)
1379                 ent = find_next_entry_inc(iter);
1380         else
1381                 ent = iter;
1382
1383         while (ent && iter->idx < i)
1384                 ent = find_next_entry_inc(iter);
1385
1386         iter->pos = *pos;
1387
1388         return ent;
1389 }
1390
1391 static void *s_start(struct seq_file *m, loff_t *pos)
1392 {
1393         struct trace_iterator *iter = m->private;
1394         void *p = NULL;
1395         loff_t l = 0;
1396         int cpu;
1397
1398         mutex_lock(&trace_types_lock);
1399
1400         if (!current_trace || current_trace != iter->trace) {
1401                 mutex_unlock(&trace_types_lock);
1402                 return NULL;
1403         }
1404
1405         atomic_inc(&trace_record_cmdline_disabled);
1406
1407         if (*pos != iter->pos) {
1408                 iter->ent = NULL;
1409                 iter->cpu = 0;
1410                 iter->idx = -1;
1411
1412                 ftrace_disable_cpu();
1413
1414                 for_each_tracing_cpu(cpu) {
1415                         ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1416                 }
1417
1418                 ftrace_enable_cpu();
1419
1420                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1421                         ;
1422
1423         } else {
1424                 l = *pos - 1;
1425                 p = s_next(m, p, &l);
1426         }
1427
1428         return p;
1429 }
1430
1431 static void s_stop(struct seq_file *m, void *p)
1432 {
1433         atomic_dec(&trace_record_cmdline_disabled);
1434         mutex_unlock(&trace_types_lock);
1435 }
1436
1437 #ifdef CONFIG_KRETPROBES
1438 static inline const char *kretprobed(const char *name)
1439 {
1440         static const char tramp_name[] = "kretprobe_trampoline";
1441         int size = sizeof(tramp_name);
1442
1443         if (strncmp(tramp_name, name, size) == 0)
1444                 return "[unknown/kretprobe'd]";
1445         return name;
1446 }
1447 #else
1448 static inline const char *kretprobed(const char *name)
1449 {
1450         return name;
1451 }
1452 #endif /* CONFIG_KRETPROBES */
1453
1454 static int
1455 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1456 {
1457 #ifdef CONFIG_KALLSYMS
1458         char str[KSYM_SYMBOL_LEN];
1459         const char *name;
1460
1461         kallsyms_lookup(address, NULL, NULL, NULL, str);
1462
1463         name = kretprobed(str);
1464
1465         return trace_seq_printf(s, fmt, name);
1466 #endif
1467         return 1;
1468 }
1469
1470 static int
1471 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1472                      unsigned long address)
1473 {
1474 #ifdef CONFIG_KALLSYMS
1475         char str[KSYM_SYMBOL_LEN];
1476         const char *name;
1477
1478         sprint_symbol(str, address);
1479         name = kretprobed(str);
1480
1481         return trace_seq_printf(s, fmt, name);
1482 #endif
1483         return 1;
1484 }
1485
1486 #ifndef CONFIG_64BIT
1487 # define IP_FMT "%08lx"
1488 #else
1489 # define IP_FMT "%016lx"
1490 #endif
1491
1492 int
1493 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1494 {
1495         int ret;
1496
1497         if (!ip)
1498                 return trace_seq_printf(s, "0");
1499
1500         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1501                 ret = seq_print_sym_offset(s, "%s", ip);
1502         else
1503                 ret = seq_print_sym_short(s, "%s", ip);
1504
1505         if (!ret)
1506                 return 0;
1507
1508         if (sym_flags & TRACE_ITER_SYM_ADDR)
1509                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1510         return ret;
1511 }
1512
1513 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1514                                     unsigned long ip, unsigned long sym_flags)
1515 {
1516         struct file *file = NULL;
1517         unsigned long vmstart = 0;
1518         int ret = 1;
1519
1520         if (mm) {
1521                 const struct vm_area_struct *vma;
1522
1523                 down_read(&mm->mmap_sem);
1524                 vma = find_vma(mm, ip);
1525                 if (vma) {
1526                         file = vma->vm_file;
1527                         vmstart = vma->vm_start;
1528                 }
1529                 if (file) {
1530                         ret = trace_seq_path(s, &file->f_path);
1531                         if (ret)
1532                                 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1533                 }
1534                 up_read(&mm->mmap_sem);
1535         }
1536         if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1537                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1538         return ret;
1539 }
1540
1541 static int
1542 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1543                       unsigned long sym_flags)
1544 {
1545         struct mm_struct *mm = NULL;
1546         int ret = 1;
1547         unsigned int i;
1548
1549         if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1550                 struct task_struct *task;
1551                 /*
1552                  * we do the lookup on the thread group leader,
1553                  * since individual threads might have already quit!
1554                  */
1555                 rcu_read_lock();
1556                 task = find_task_by_vpid(entry->ent.tgid);
1557                 if (task)
1558                         mm = get_task_mm(task);
1559                 rcu_read_unlock();
1560         }
1561
1562         for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1563                 unsigned long ip = entry->caller[i];
1564
1565                 if (ip == ULONG_MAX || !ret)
1566                         break;
1567                 if (i && ret)
1568                         ret = trace_seq_puts(s, " <- ");
1569                 if (!ip) {
1570                         if (ret)
1571                                 ret = trace_seq_puts(s, "??");
1572                         continue;
1573                 }
1574                 if (!ret)
1575                         break;
1576                 if (ret)
1577                         ret = seq_print_user_ip(s, mm, ip, sym_flags);
1578         }
1579
1580         if (mm)
1581                 mmput(mm);
1582         return ret;
1583 }
1584
1585 static void print_lat_help_header(struct seq_file *m)
1586 {
1587         seq_puts(m, "#                  _------=> CPU#            \n");
1588         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1589         seq_puts(m, "#                | / _----=> need-resched    \n");
1590         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1591         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1592         seq_puts(m, "#                |||| /                      \n");
1593         seq_puts(m, "#                |||||     delay             \n");
1594         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1595         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1596 }
1597
1598 static void print_func_help_header(struct seq_file *m)
1599 {
1600         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1601         seq_puts(m, "#              | |       |          |         |\n");
1602 }
1603
1604
1605 static void
1606 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1607 {
1608         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1609         struct trace_array *tr = iter->tr;
1610         struct trace_array_cpu *data = tr->data[tr->cpu];
1611         struct tracer *type = current_trace;
1612         unsigned long total;
1613         unsigned long entries;
1614         const char *name = "preemption";
1615
1616         if (type)
1617                 name = type->name;
1618
1619         entries = ring_buffer_entries(iter->tr->buffer);
1620         total = entries +
1621                 ring_buffer_overruns(iter->tr->buffer);
1622
1623         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1624                    name, UTS_RELEASE);
1625         seq_puts(m, "-----------------------------------"
1626                  "---------------------------------\n");
1627         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1628                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1629                    nsecs_to_usecs(data->saved_latency),
1630                    entries,
1631                    total,
1632                    tr->cpu,
1633 #if defined(CONFIG_PREEMPT_NONE)
1634                    "server",
1635 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1636                    "desktop",
1637 #elif defined(CONFIG_PREEMPT)
1638                    "preempt",
1639 #else
1640                    "unknown",
1641 #endif
1642                    /* These are reserved for later use */
1643                    0, 0, 0, 0);
1644 #ifdef CONFIG_SMP
1645         seq_printf(m, " #P:%d)\n", num_online_cpus());
1646 #else
1647         seq_puts(m, ")\n");
1648 #endif
1649         seq_puts(m, "    -----------------\n");
1650         seq_printf(m, "    | task: %.16s-%d "
1651                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1652                    data->comm, data->pid, data->uid, data->nice,
1653                    data->policy, data->rt_priority);
1654         seq_puts(m, "    -----------------\n");
1655
1656         if (data->critical_start) {
1657                 seq_puts(m, " => started at: ");
1658                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1659                 trace_print_seq(m, &iter->seq);
1660                 seq_puts(m, "\n => ended at:   ");
1661                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1662                 trace_print_seq(m, &iter->seq);
1663                 seq_puts(m, "\n");
1664         }
1665
1666         seq_puts(m, "\n");
1667 }
1668
1669 static void
1670 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1671 {
1672         int hardirq, softirq;
1673         char *comm;
1674
1675         comm = trace_find_cmdline(entry->pid);
1676
1677         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1678         trace_seq_printf(s, "%3d", cpu);
1679         trace_seq_printf(s, "%c%c",
1680                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1681                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1682                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1683
1684         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1685         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1686         if (hardirq && softirq) {
1687                 trace_seq_putc(s, 'H');
1688         } else {
1689                 if (hardirq) {
1690                         trace_seq_putc(s, 'h');
1691                 } else {
1692                         if (softirq)
1693                                 trace_seq_putc(s, 's');
1694                         else
1695                                 trace_seq_putc(s, '.');
1696                 }
1697         }
1698
1699         if (entry->preempt_count)
1700                 trace_seq_printf(s, "%x", entry->preempt_count);
1701         else
1702                 trace_seq_puts(s, ".");
1703 }
1704
1705 unsigned long preempt_mark_thresh = 100;
1706
1707 static void
1708 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1709                     unsigned long rel_usecs)
1710 {
1711         trace_seq_printf(s, " %4lldus", abs_usecs);
1712         if (rel_usecs > preempt_mark_thresh)
1713                 trace_seq_puts(s, "!: ");
1714         else if (rel_usecs > 1)
1715                 trace_seq_puts(s, "+: ");
1716         else
1717                 trace_seq_puts(s, " : ");
1718 }
1719
1720 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1721
1722 /*
1723  * The message is supposed to contain an ending newline.
1724  * If the printing stops prematurely, try to add a newline of our own.
1725  */
1726 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1727 {
1728         struct trace_entry *ent;
1729         struct trace_field_cont *cont;
1730         bool ok = true;
1731
1732         ent = peek_next_entry(iter, iter->cpu, NULL);
1733         if (!ent || ent->type != TRACE_CONT) {
1734                 trace_seq_putc(s, '\n');
1735                 return;
1736         }
1737
1738         do {
1739                 cont = (struct trace_field_cont *)ent;
1740                 if (ok)
1741                         ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1742
1743                 ftrace_disable_cpu();
1744
1745                 if (iter->buffer_iter[iter->cpu])
1746                         ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1747                 else
1748                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1749
1750                 ftrace_enable_cpu();
1751
1752                 ent = peek_next_entry(iter, iter->cpu, NULL);
1753         } while (ent && ent->type == TRACE_CONT);
1754
1755         if (!ok)
1756                 trace_seq_putc(s, '\n');
1757 }
1758
1759 static void test_cpu_buff_start(struct trace_iterator *iter)
1760 {
1761         struct trace_seq *s = &iter->seq;
1762
1763         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1764                 return;
1765
1766         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1767                 return;
1768
1769         if (cpu_isset(iter->cpu, iter->started))
1770                 return;
1771
1772         cpu_set(iter->cpu, iter->started);
1773         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1774 }
1775
1776 static enum print_line_t
1777 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1778 {
1779         struct trace_seq *s = &iter->seq;
1780         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1781         struct trace_entry *next_entry;
1782         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1783         struct trace_entry *entry = iter->ent;
1784         unsigned long abs_usecs;
1785         unsigned long rel_usecs;
1786         u64 next_ts;
1787         char *comm;
1788         int S, T;
1789         int i;
1790         unsigned state;
1791
1792         if (entry->type == TRACE_CONT)
1793                 return TRACE_TYPE_HANDLED;
1794
1795         test_cpu_buff_start(iter);
1796
1797         next_entry = find_next_entry(iter, NULL, &next_ts);
1798         if (!next_entry)
1799                 next_ts = iter->ts;
1800         rel_usecs = ns2usecs(next_ts - iter->ts);
1801         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1802
1803         if (verbose) {
1804                 comm = trace_find_cmdline(entry->pid);
1805                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1806                                  " %ld.%03ldms (+%ld.%03ldms): ",
1807                                  comm,
1808                                  entry->pid, cpu, entry->flags,
1809                                  entry->preempt_count, trace_idx,
1810                                  ns2usecs(iter->ts),
1811                                  abs_usecs/1000,
1812                                  abs_usecs % 1000, rel_usecs/1000,
1813                                  rel_usecs % 1000);
1814         } else {
1815                 lat_print_generic(s, entry, cpu);
1816                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1817         }
1818         switch (entry->type) {
1819         case TRACE_FN: {
1820                 struct ftrace_entry *field;
1821
1822                 trace_assign_type(field, entry);
1823
1824                 seq_print_ip_sym(s, field->ip, sym_flags);
1825                 trace_seq_puts(s, " (");
1826                 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1827                 trace_seq_puts(s, ")\n");
1828                 break;
1829         }
1830         case TRACE_CTX:
1831         case TRACE_WAKE: {
1832                 struct ctx_switch_entry *field;
1833
1834                 trace_assign_type(field, entry);
1835
1836                 T = field->next_state < sizeof(state_to_char) ?
1837                         state_to_char[field->next_state] : 'X';
1838
1839                 state = field->prev_state ?
1840                         __ffs(field->prev_state) + 1 : 0;
1841                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1842                 comm = trace_find_cmdline(field->next_pid);
1843                 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1844                                  field->prev_pid,
1845                                  field->prev_prio,
1846                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1847                                  field->next_cpu,
1848                                  field->next_pid,
1849                                  field->next_prio,
1850                                  T, comm);
1851                 break;
1852         }
1853         case TRACE_SPECIAL: {
1854                 struct special_entry *field;
1855
1856                 trace_assign_type(field, entry);
1857
1858                 trace_seq_printf(s, "# %ld %ld %ld\n",
1859                                  field->arg1,
1860                                  field->arg2,
1861                                  field->arg3);
1862                 break;
1863         }
1864         case TRACE_STACK: {
1865                 struct stack_entry *field;
1866
1867                 trace_assign_type(field, entry);
1868
1869                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1870                         if (i)
1871                                 trace_seq_puts(s, " <= ");
1872                         seq_print_ip_sym(s, field->caller[i], sym_flags);
1873                 }
1874                 trace_seq_puts(s, "\n");
1875                 break;
1876         }
1877         case TRACE_PRINT: {
1878                 struct print_entry *field;
1879
1880                 trace_assign_type(field, entry);
1881
1882                 seq_print_ip_sym(s, field->ip, sym_flags);
1883                 trace_seq_printf(s, ": %s", field->buf);
1884                 if (entry->flags & TRACE_FLAG_CONT)
1885                         trace_seq_print_cont(s, iter);
1886                 break;
1887         }
1888         case TRACE_BRANCH: {
1889                 struct trace_branch *field;
1890
1891                 trace_assign_type(field, entry);
1892
1893                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1894                                  field->correct ? "  ok  " : " MISS ",
1895                                  field->func,
1896                                  field->file,
1897                                  field->line);
1898                 break;
1899         }
1900         case TRACE_USER_STACK: {
1901                 struct userstack_entry *field;
1902
1903                 trace_assign_type(field, entry);
1904
1905                 seq_print_userip_objs(field, s, sym_flags);
1906                 trace_seq_putc(s, '\n');
1907                 break;
1908         }
1909         default:
1910                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1911         }
1912         return TRACE_TYPE_HANDLED;
1913 }
1914
1915 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1916 {
1917         struct trace_seq *s = &iter->seq;
1918         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1919         struct trace_entry *entry;
1920         unsigned long usec_rem;
1921         unsigned long long t;
1922         unsigned long secs;
1923         char *comm;
1924         int ret;
1925         int S, T;
1926         int i;
1927
1928         entry = iter->ent;
1929
1930         if (entry->type == TRACE_CONT)
1931                 return TRACE_TYPE_HANDLED;
1932
1933         test_cpu_buff_start(iter);
1934
1935         comm = trace_find_cmdline(iter->ent->pid);
1936
1937         t = ns2usecs(iter->ts);
1938         usec_rem = do_div(t, 1000000ULL);
1939         secs = (unsigned long)t;
1940
1941         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1942         if (!ret)
1943                 return TRACE_TYPE_PARTIAL_LINE;
1944         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1945         if (!ret)
1946                 return TRACE_TYPE_PARTIAL_LINE;
1947         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1948         if (!ret)
1949                 return TRACE_TYPE_PARTIAL_LINE;
1950
1951         switch (entry->type) {
1952         case TRACE_FN: {
1953                 struct ftrace_entry *field;
1954
1955                 trace_assign_type(field, entry);
1956
1957                 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1958                 if (!ret)
1959                         return TRACE_TYPE_PARTIAL_LINE;
1960                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1961                                                 field->parent_ip) {
1962                         ret = trace_seq_printf(s, " <-");
1963                         if (!ret)
1964                                 return TRACE_TYPE_PARTIAL_LINE;
1965                         ret = seq_print_ip_sym(s,
1966                                                field->parent_ip,
1967                                                sym_flags);
1968                         if (!ret)
1969                                 return TRACE_TYPE_PARTIAL_LINE;
1970                 }
1971                 ret = trace_seq_printf(s, "\n");
1972                 if (!ret)
1973                         return TRACE_TYPE_PARTIAL_LINE;
1974                 break;
1975         }
1976         case TRACE_CTX:
1977         case TRACE_WAKE: {
1978                 struct ctx_switch_entry *field;
1979
1980                 trace_assign_type(field, entry);
1981
1982                 S = field->prev_state < sizeof(state_to_char) ?
1983                         state_to_char[field->prev_state] : 'X';
1984                 T = field->next_state < sizeof(state_to_char) ?
1985                         state_to_char[field->next_state] : 'X';
1986                 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1987                                        field->prev_pid,
1988                                        field->prev_prio,
1989                                        S,
1990                                        entry->type == TRACE_CTX ? "==>" : "  +",
1991                                        field->next_cpu,
1992                                        field->next_pid,
1993                                        field->next_prio,
1994                                        T);
1995                 if (!ret)
1996                         return TRACE_TYPE_PARTIAL_LINE;
1997                 break;
1998         }
1999         case TRACE_SPECIAL: {
2000                 struct special_entry *field;
2001
2002                 trace_assign_type(field, entry);
2003
2004                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2005                                  field->arg1,
2006                                  field->arg2,
2007                                  field->arg3);
2008                 if (!ret)
2009                         return TRACE_TYPE_PARTIAL_LINE;
2010                 break;
2011         }
2012         case TRACE_STACK: {
2013                 struct stack_entry *field;
2014
2015                 trace_assign_type(field, entry);
2016
2017                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2018                         if (i) {
2019                                 ret = trace_seq_puts(s, " <= ");
2020                                 if (!ret)
2021                                         return TRACE_TYPE_PARTIAL_LINE;
2022                         }
2023                         ret = seq_print_ip_sym(s, field->caller[i],
2024                                                sym_flags);
2025                         if (!ret)
2026                                 return TRACE_TYPE_PARTIAL_LINE;
2027                 }
2028                 ret = trace_seq_puts(s, "\n");
2029                 if (!ret)
2030                         return TRACE_TYPE_PARTIAL_LINE;
2031                 break;
2032         }
2033         case TRACE_PRINT: {
2034                 struct print_entry *field;
2035
2036                 trace_assign_type(field, entry);
2037
2038                 seq_print_ip_sym(s, field->ip, sym_flags);
2039                 trace_seq_printf(s, ": %s", field->buf);
2040                 if (entry->flags & TRACE_FLAG_CONT)
2041                         trace_seq_print_cont(s, iter);
2042                 break;
2043         }
2044         case TRACE_GRAPH_RET: {
2045                 return print_graph_function(iter);
2046         }
2047         case TRACE_GRAPH_ENT: {
2048                 return print_graph_function(iter);
2049         }
2050         case TRACE_BRANCH: {
2051                 struct trace_branch *field;
2052
2053                 trace_assign_type(field, entry);
2054
2055                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2056                                  field->correct ? "  ok  " : " MISS ",
2057                                  field->func,
2058                                  field->file,
2059                                  field->line);
2060                 break;
2061         }
2062         case TRACE_USER_STACK: {
2063                 struct userstack_entry *field;
2064
2065                 trace_assign_type(field, entry);
2066
2067                 ret = seq_print_userip_objs(field, s, sym_flags);
2068                 if (!ret)
2069                         return TRACE_TYPE_PARTIAL_LINE;
2070                 ret = trace_seq_putc(s, '\n');
2071                 if (!ret)
2072                         return TRACE_TYPE_PARTIAL_LINE;
2073                 break;
2074         }
2075         }
2076         return TRACE_TYPE_HANDLED;
2077 }
2078
2079 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2080 {
2081         struct trace_seq *s = &iter->seq;
2082         struct trace_entry *entry;
2083         int ret;
2084         int S, T;
2085
2086         entry = iter->ent;
2087
2088         if (entry->type == TRACE_CONT)
2089                 return TRACE_TYPE_HANDLED;
2090
2091         ret = trace_seq_printf(s, "%d %d %llu ",
2092                 entry->pid, iter->cpu, iter->ts);
2093         if (!ret)
2094                 return TRACE_TYPE_PARTIAL_LINE;
2095
2096         switch (entry->type) {
2097         case TRACE_FN: {
2098                 struct ftrace_entry *field;
2099
2100                 trace_assign_type(field, entry);
2101
2102                 ret = trace_seq_printf(s, "%x %x\n",
2103                                         field->ip,
2104                                         field->parent_ip);
2105                 if (!ret)
2106                         return TRACE_TYPE_PARTIAL_LINE;
2107                 break;
2108         }
2109         case TRACE_CTX:
2110         case TRACE_WAKE: {
2111                 struct ctx_switch_entry *field;
2112
2113                 trace_assign_type(field, entry);
2114
2115                 S = field->prev_state < sizeof(state_to_char) ?
2116                         state_to_char[field->prev_state] : 'X';
2117                 T = field->next_state < sizeof(state_to_char) ?
2118                         state_to_char[field->next_state] : 'X';
2119                 if (entry->type == TRACE_WAKE)
2120                         S = '+';
2121                 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2122                                        field->prev_pid,
2123                                        field->prev_prio,
2124                                        S,
2125                                        field->next_cpu,
2126                                        field->next_pid,
2127                                        field->next_prio,
2128                                        T);
2129                 if (!ret)
2130                         return TRACE_TYPE_PARTIAL_LINE;
2131                 break;
2132         }
2133         case TRACE_SPECIAL:
2134         case TRACE_USER_STACK:
2135         case TRACE_STACK: {
2136                 struct special_entry *field;
2137
2138                 trace_assign_type(field, entry);
2139
2140                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2141                                  field->arg1,
2142                                  field->arg2,
2143                                  field->arg3);
2144                 if (!ret)
2145                         return TRACE_TYPE_PARTIAL_LINE;
2146                 break;
2147         }
2148         case TRACE_PRINT: {
2149                 struct print_entry *field;
2150
2151                 trace_assign_type(field, entry);
2152
2153                 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2154                 if (entry->flags & TRACE_FLAG_CONT)
2155                         trace_seq_print_cont(s, iter);
2156                 break;
2157         }
2158         }
2159         return TRACE_TYPE_HANDLED;
2160 }
2161
2162 #define SEQ_PUT_FIELD_RET(s, x)                         \
2163 do {                                                    \
2164         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
2165                 return 0;                               \
2166 } while (0)
2167
2168 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
2169 do {                                                    \
2170         BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES);     \
2171         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
2172                 return 0;                               \
2173 } while (0)
2174
2175 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2176 {
2177         struct trace_seq *s = &iter->seq;
2178         unsigned char newline = '\n';
2179         struct trace_entry *entry;
2180         int S, T;
2181
2182         entry = iter->ent;
2183
2184         if (entry->type == TRACE_CONT)
2185                 return TRACE_TYPE_HANDLED;
2186
2187         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2188         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2189         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2190
2191         switch (entry->type) {
2192         case TRACE_FN: {
2193                 struct ftrace_entry *field;
2194
2195                 trace_assign_type(field, entry);
2196
2197                 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2198                 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2199                 break;
2200         }
2201         case TRACE_CTX:
2202         case TRACE_WAKE: {
2203                 struct ctx_switch_entry *field;
2204
2205                 trace_assign_type(field, entry);
2206
2207                 S = field->prev_state < sizeof(state_to_char) ?
2208                         state_to_char[field->prev_state] : 'X';
2209                 T = field->next_state < sizeof(state_to_char) ?
2210                         state_to_char[field->next_state] : 'X';
2211                 if (entry->type == TRACE_WAKE)
2212                         S = '+';
2213                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2214                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2215                 SEQ_PUT_HEX_FIELD_RET(s, S);
2216                 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2217                 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2218                 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2219                 SEQ_PUT_HEX_FIELD_RET(s, T);
2220                 break;
2221         }
2222         case TRACE_SPECIAL:
2223         case TRACE_USER_STACK:
2224         case TRACE_STACK: {
2225                 struct special_entry *field;
2226
2227                 trace_assign_type(field, entry);
2228
2229                 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2230                 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2231                 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2232                 break;
2233         }
2234         }
2235         SEQ_PUT_FIELD_RET(s, newline);
2236
2237         return TRACE_TYPE_HANDLED;
2238 }
2239
2240 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2241 {
2242         struct trace_seq *s = &iter->seq;
2243         struct trace_entry *entry;
2244
2245         entry = iter->ent;
2246
2247         if (entry->type == TRACE_CONT)
2248                 return TRACE_TYPE_HANDLED;
2249
2250         SEQ_PUT_FIELD_RET(s, entry->pid);
2251         SEQ_PUT_FIELD_RET(s, entry->cpu);
2252         SEQ_PUT_FIELD_RET(s, iter->ts);
2253
2254         switch (entry->type) {
2255         case TRACE_FN: {
2256                 struct ftrace_entry *field;
2257
2258                 trace_assign_type(field, entry);
2259
2260                 SEQ_PUT_FIELD_RET(s, field->ip);
2261                 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2262                 break;
2263         }
2264         case TRACE_CTX: {
2265                 struct ctx_switch_entry *field;
2266
2267                 trace_assign_type(field, entry);
2268
2269                 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2270                 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2271                 SEQ_PUT_FIELD_RET(s, field->prev_state);
2272                 SEQ_PUT_FIELD_RET(s, field->next_pid);
2273                 SEQ_PUT_FIELD_RET(s, field->next_prio);
2274                 SEQ_PUT_FIELD_RET(s, field->next_state);
2275                 break;
2276         }
2277         case TRACE_SPECIAL:
2278         case TRACE_USER_STACK:
2279         case TRACE_STACK: {
2280                 struct special_entry *field;
2281
2282                 trace_assign_type(field, entry);
2283
2284                 SEQ_PUT_FIELD_RET(s, field->arg1);
2285                 SEQ_PUT_FIELD_RET(s, field->arg2);
2286                 SEQ_PUT_FIELD_RET(s, field->arg3);
2287                 break;
2288         }
2289         }
2290         return 1;
2291 }
2292
2293 static int trace_empty(struct trace_iterator *iter)
2294 {
2295         int cpu;
2296
2297         for_each_tracing_cpu(cpu) {
2298                 if (iter->buffer_iter[cpu]) {
2299                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2300                                 return 0;
2301                 } else {
2302                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2303                                 return 0;
2304                 }
2305         }
2306
2307         return 1;
2308 }
2309
2310 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2311 {
2312         enum print_line_t ret;
2313
2314         if (iter->trace && iter->trace->print_line) {
2315                 ret = iter->trace->print_line(iter);
2316                 if (ret != TRACE_TYPE_UNHANDLED)
2317                         return ret;
2318         }
2319
2320         if (trace_flags & TRACE_ITER_BIN)
2321                 return print_bin_fmt(iter);
2322
2323         if (trace_flags & TRACE_ITER_HEX)
2324                 return print_hex_fmt(iter);
2325
2326         if (trace_flags & TRACE_ITER_RAW)
2327                 return print_raw_fmt(iter);
2328
2329         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2330                 return print_lat_fmt(iter, iter->idx, iter->cpu);
2331
2332         return print_trace_fmt(iter);
2333 }
2334
2335 static int s_show(struct seq_file *m, void *v)
2336 {
2337         struct trace_iterator *iter = v;
2338
2339         if (iter->ent == NULL) {
2340                 if (iter->tr) {
2341                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2342                         seq_puts(m, "#\n");
2343                 }
2344                 if (iter->trace && iter->trace->print_header)
2345                         iter->trace->print_header(m);
2346                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2347                         /* print nothing if the buffers are empty */
2348                         if (trace_empty(iter))
2349                                 return 0;
2350                         print_trace_header(m, iter);
2351                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2352                                 print_lat_help_header(m);
2353                 } else {
2354                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2355                                 print_func_help_header(m);
2356                 }
2357         } else {
2358                 print_trace_line(iter);
2359                 trace_print_seq(m, &iter->seq);
2360         }
2361
2362         return 0;
2363 }
2364
2365 static struct seq_operations tracer_seq_ops = {
2366         .start          = s_start,
2367         .next           = s_next,
2368         .stop           = s_stop,
2369         .show           = s_show,
2370 };
2371
2372 static struct trace_iterator *
2373 __tracing_open(struct inode *inode, struct file *file, int *ret)
2374 {
2375         struct trace_iterator *iter;
2376         struct seq_file *m;
2377         int cpu;
2378
2379         if (tracing_disabled) {
2380                 *ret = -ENODEV;
2381                 return NULL;
2382         }
2383
2384         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2385         if (!iter) {
2386                 *ret = -ENOMEM;
2387                 goto out;
2388         }
2389
2390         mutex_lock(&trace_types_lock);
2391         if (current_trace && current_trace->print_max)
2392                 iter->tr = &max_tr;
2393         else
2394                 iter->tr = inode->i_private;
2395         iter->trace = current_trace;
2396         iter->pos = -1;
2397
2398         /* Notify the tracer early; before we stop tracing. */
2399         if (iter->trace && iter->trace->open)
2400                         iter->trace->open(iter);
2401
2402         /* Annotate start of buffers if we had overruns */
2403         if (ring_buffer_overruns(iter->tr->buffer))
2404                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2405
2406
2407         for_each_tracing_cpu(cpu) {
2408
2409                 iter->buffer_iter[cpu] =
2410                         ring_buffer_read_start(iter->tr->buffer, cpu);
2411
2412                 if (!iter->buffer_iter[cpu])
2413                         goto fail_buffer;
2414         }
2415
2416         /* TODO stop tracer */
2417         *ret = seq_open(file, &tracer_seq_ops);
2418         if (*ret)
2419                 goto fail_buffer;
2420
2421         m = file->private_data;
2422         m->private = iter;
2423
2424         /* stop the trace while dumping */
2425         tracing_stop();
2426
2427         mutex_unlock(&trace_types_lock);
2428
2429  out:
2430         return iter;
2431
2432  fail_buffer:
2433         for_each_tracing_cpu(cpu) {
2434                 if (iter->buffer_iter[cpu])
2435                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2436         }
2437         mutex_unlock(&trace_types_lock);
2438         kfree(iter);
2439
2440         return ERR_PTR(-ENOMEM);
2441 }
2442
2443 int tracing_open_generic(struct inode *inode, struct file *filp)
2444 {
2445         if (tracing_disabled)
2446                 return -ENODEV;
2447
2448         filp->private_data = inode->i_private;
2449         return 0;
2450 }
2451
2452 int tracing_release(struct inode *inode, struct file *file)
2453 {
2454         struct seq_file *m = (struct seq_file *)file->private_data;
2455         struct trace_iterator *iter = m->private;
2456         int cpu;
2457
2458         mutex_lock(&trace_types_lock);
2459         for_each_tracing_cpu(cpu) {
2460                 if (iter->buffer_iter[cpu])
2461                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2462         }
2463
2464         if (iter->trace && iter->trace->close)
2465                 iter->trace->close(iter);
2466
2467         /* reenable tracing if it was previously enabled */
2468         tracing_start();
2469         mutex_unlock(&trace_types_lock);
2470
2471         seq_release(inode, file);
2472         kfree(iter);
2473         return 0;
2474 }
2475
2476 static int tracing_open(struct inode *inode, struct file *file)
2477 {
2478         int ret;
2479
2480         __tracing_open(inode, file, &ret);
2481
2482         return ret;
2483 }
2484
2485 static int tracing_lt_open(struct inode *inode, struct file *file)
2486 {
2487         struct trace_iterator *iter;
2488         int ret;
2489
2490         iter = __tracing_open(inode, file, &ret);
2491
2492         if (!ret)
2493                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2494
2495         return ret;
2496 }
2497
2498
2499 static void *
2500 t_next(struct seq_file *m, void *v, loff_t *pos)
2501 {
2502         struct tracer *t = m->private;
2503
2504         (*pos)++;
2505
2506         if (t)
2507                 t = t->next;
2508
2509         m->private = t;
2510
2511         return t;
2512 }
2513
2514 static void *t_start(struct seq_file *m, loff_t *pos)
2515 {
2516         struct tracer *t = m->private;
2517         loff_t l = 0;
2518
2519         mutex_lock(&trace_types_lock);
2520         for (; t && l < *pos; t = t_next(m, t, &l))
2521                 ;
2522
2523         return t;
2524 }
2525
2526 static void t_stop(struct seq_file *m, void *p)
2527 {
2528         mutex_unlock(&trace_types_lock);
2529 }
2530
2531 static int t_show(struct seq_file *m, void *v)
2532 {
2533         struct tracer *t = v;
2534
2535         if (!t)
2536                 return 0;
2537
2538         seq_printf(m, "%s", t->name);
2539         if (t->next)
2540                 seq_putc(m, ' ');
2541         else
2542                 seq_putc(m, '\n');
2543
2544         return 0;
2545 }
2546
2547 static struct seq_operations show_traces_seq_ops = {
2548         .start          = t_start,
2549         .next           = t_next,
2550         .stop           = t_stop,
2551         .show           = t_show,
2552 };
2553
2554 static int show_traces_open(struct inode *inode, struct file *file)
2555 {
2556         int ret;
2557
2558         if (tracing_disabled)
2559                 return -ENODEV;
2560
2561         ret = seq_open(file, &show_traces_seq_ops);
2562         if (!ret) {
2563                 struct seq_file *m = file->private_data;
2564                 m->private = trace_types;
2565         }
2566
2567         return ret;
2568 }
2569
2570 static struct file_operations tracing_fops = {
2571         .open           = tracing_open,
2572         .read           = seq_read,
2573         .llseek         = seq_lseek,
2574         .release        = tracing_release,
2575 };
2576
2577 static struct file_operations tracing_lt_fops = {
2578         .open           = tracing_lt_open,
2579         .read           = seq_read,
2580         .llseek         = seq_lseek,
2581         .release        = tracing_release,
2582 };
2583
2584 static struct file_operations show_traces_fops = {
2585         .open           = show_traces_open,
2586         .read           = seq_read,
2587         .release        = seq_release,
2588 };
2589
2590 /*
2591  * Only trace on a CPU if the bitmask is set:
2592  */
2593 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2594
2595 /*
2596  * When tracing/tracing_cpu_mask is modified then this holds
2597  * the new bitmask we are about to install:
2598  */
2599 static cpumask_t tracing_cpumask_new;
2600
2601 /*
2602  * The tracer itself will not take this lock, but still we want
2603  * to provide a consistent cpumask to user-space:
2604  */
2605 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2606
2607 /*
2608  * Temporary storage for the character representation of the
2609  * CPU bitmask (and one more byte for the newline):
2610  */
2611 static char mask_str[NR_CPUS + 1];
2612
2613 static ssize_t
2614 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2615                      size_t count, loff_t *ppos)
2616 {
2617         int len;
2618
2619         mutex_lock(&tracing_cpumask_update_lock);
2620
2621         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2622         if (count - len < 2) {
2623                 count = -EINVAL;
2624                 goto out_err;
2625         }
2626         len += sprintf(mask_str + len, "\n");
2627         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2628
2629 out_err:
2630         mutex_unlock(&tracing_cpumask_update_lock);
2631
2632         return count;
2633 }
2634
2635 static ssize_t
2636 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2637                       size_t count, loff_t *ppos)
2638 {
2639         int err, cpu;
2640
2641         mutex_lock(&tracing_cpumask_update_lock);
2642         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2643         if (err)
2644                 goto err_unlock;
2645
2646         raw_local_irq_disable();
2647         __raw_spin_lock(&ftrace_max_lock);
2648         for_each_tracing_cpu(cpu) {
2649                 /*
2650                  * Increase/decrease the disabled counter if we are
2651                  * about to flip a bit in the cpumask:
2652                  */
2653                 if (cpu_isset(cpu, tracing_cpumask) &&
2654                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2655                         atomic_inc(&global_trace.data[cpu]->disabled);
2656                 }
2657                 if (!cpu_isset(cpu, tracing_cpumask) &&
2658                                 cpu_isset(cpu, tracing_cpumask_new)) {
2659                         atomic_dec(&global_trace.data[cpu]->disabled);
2660                 }
2661         }
2662         __raw_spin_unlock(&ftrace_max_lock);
2663         raw_local_irq_enable();
2664
2665         tracing_cpumask = tracing_cpumask_new;
2666
2667         mutex_unlock(&tracing_cpumask_update_lock);
2668
2669         return count;
2670
2671 err_unlock:
2672         mutex_unlock(&tracing_cpumask_update_lock);
2673
2674         return err;
2675 }
2676
2677 static struct file_operations tracing_cpumask_fops = {
2678         .open           = tracing_open_generic,
2679         .read           = tracing_cpumask_read,
2680         .write          = tracing_cpumask_write,
2681 };
2682
2683 static ssize_t
2684 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2685                        size_t cnt, loff_t *ppos)
2686 {
2687         int i;
2688         char *buf;
2689         int r = 0;
2690         int len = 0;
2691         u32 tracer_flags = current_trace->flags->val;
2692         struct tracer_opt *trace_opts = current_trace->flags->opts;
2693
2694
2695         /* calulate max size */
2696         for (i = 0; trace_options[i]; i++) {
2697                 len += strlen(trace_options[i]);
2698                 len += 3; /* "no" and space */
2699         }
2700
2701         /*
2702          * Increase the size with names of options specific
2703          * of the current tracer.
2704          */
2705         for (i = 0; trace_opts[i].name; i++) {
2706                 len += strlen(trace_opts[i].name);
2707                 len += 3; /* "no" and space */
2708         }
2709
2710         /* +2 for \n and \0 */
2711         buf = kmalloc(len + 2, GFP_KERNEL);
2712         if (!buf)
2713                 return -ENOMEM;
2714
2715         for (i = 0; trace_options[i]; i++) {
2716                 if (trace_flags & (1 << i))
2717                         r += sprintf(buf + r, "%s ", trace_options[i]);
2718                 else
2719                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2720         }
2721
2722         for (i = 0; trace_opts[i].name; i++) {
2723                 if (tracer_flags & trace_opts[i].bit)
2724                         r += sprintf(buf + r, "%s ",
2725                                 trace_opts[i].name);
2726                 else
2727                         r += sprintf(buf + r, "no%s ",
2728                                 trace_opts[i].name);
2729         }
2730
2731         r += sprintf(buf + r, "\n");
2732         WARN_ON(r >= len + 2);
2733
2734         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2735
2736         kfree(buf);
2737
2738         return r;
2739 }
2740
2741 /* Try to assign a tracer specific option */
2742 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2743 {
2744         struct tracer_flags *trace_flags = trace->flags;
2745         struct tracer_opt *opts = NULL;
2746         int ret = 0, i = 0;
2747         int len;
2748
2749         for (i = 0; trace_flags->opts[i].name; i++) {
2750                 opts = &trace_flags->opts[i];
2751                 len = strlen(opts->name);
2752
2753                 if (strncmp(cmp, opts->name, len) == 0) {
2754                         ret = trace->set_flag(trace_flags->val,
2755                                 opts->bit, !neg);
2756                         break;
2757                 }
2758         }
2759         /* Not found */
2760         if (!trace_flags->opts[i].name)
2761                 return -EINVAL;
2762
2763         /* Refused to handle */
2764         if (ret)
2765                 return ret;
2766
2767         if (neg)
2768                 trace_flags->val &= ~opts->bit;
2769         else
2770                 trace_flags->val |= opts->bit;
2771
2772         return 0;
2773 }
2774
2775 static ssize_t
2776 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2777                         size_t cnt, loff_t *ppos)
2778 {
2779         char buf[64];
2780         char *cmp = buf;
2781         int neg = 0;
2782         int ret;
2783         int i;
2784
2785         if (cnt >= sizeof(buf))
2786                 return -EINVAL;
2787
2788         if (copy_from_user(&buf, ubuf, cnt))
2789                 return -EFAULT;
2790
2791         buf[cnt] = 0;
2792
2793         if (strncmp(buf, "no", 2) == 0) {
2794                 neg = 1;
2795                 cmp += 2;
2796         }
2797
2798         for (i = 0; trace_options[i]; i++) {
2799                 int len = strlen(trace_options[i]);
2800
2801                 if (strncmp(cmp, trace_options[i], len) == 0) {
2802                         if (neg)
2803                                 trace_flags &= ~(1 << i);
2804                         else
2805                                 trace_flags |= (1 << i);
2806                         break;
2807                 }
2808         }
2809
2810         /* If no option could be set, test the specific tracer options */
2811         if (!trace_options[i]) {
2812                 ret = set_tracer_option(current_trace, cmp, neg);
2813                 if (ret)
2814                         return ret;
2815         }
2816
2817         filp->f_pos += cnt;
2818
2819         return cnt;
2820 }
2821
2822 static struct file_operations tracing_iter_fops = {
2823         .open           = tracing_open_generic,
2824         .read           = tracing_trace_options_read,
2825         .write          = tracing_trace_options_write,
2826 };
2827
2828 static const char readme_msg[] =
2829         "tracing mini-HOWTO:\n\n"
2830         "# mkdir /debug\n"
2831         "# mount -t debugfs nodev /debug\n\n"
2832         "# cat /debug/tracing/available_tracers\n"
2833         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2834         "# cat /debug/tracing/current_tracer\n"
2835         "none\n"
2836         "# echo sched_switch > /debug/tracing/current_tracer\n"
2837         "# cat /debug/tracing/current_tracer\n"
2838         "sched_switch\n"
2839         "# cat /debug/tracing/trace_options\n"
2840         "noprint-parent nosym-offset nosym-addr noverbose\n"
2841         "# echo print-parent > /debug/tracing/trace_options\n"
2842         "# echo 1 > /debug/tracing/tracing_enabled\n"
2843         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2844         "echo 0 > /debug/tracing/tracing_enabled\n"
2845 ;
2846
2847 static ssize_t
2848 tracing_readme_read(struct file *filp, char __user *ubuf,
2849                        size_t cnt, loff_t *ppos)
2850 {
2851         return simple_read_from_buffer(ubuf, cnt, ppos,
2852                                         readme_msg, strlen(readme_msg));
2853 }
2854
2855 static struct file_operations tracing_readme_fops = {
2856         .open           = tracing_open_generic,
2857         .read           = tracing_readme_read,
2858 };
2859
2860 static ssize_t
2861 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2862                   size_t cnt, loff_t *ppos)
2863 {
2864         char buf[64];
2865         int r;
2866
2867         r = sprintf(buf, "%u\n", tracer_enabled);
2868         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2869 }
2870
2871 static ssize_t
2872 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2873                    size_t cnt, loff_t *ppos)
2874 {
2875         struct trace_array *tr = filp->private_data;
2876         char buf[64];
2877         long val;
2878         int ret;
2879
2880         if (cnt >= sizeof(buf))
2881                 return -EINVAL;
2882
2883         if (copy_from_user(&buf, ubuf, cnt))
2884                 return -EFAULT;
2885
2886         buf[cnt] = 0;
2887
2888         ret = strict_strtoul(buf, 10, &val);
2889         if (ret < 0)
2890                 return ret;
2891
2892         val = !!val;
2893
2894         mutex_lock(&trace_types_lock);
2895         if (tracer_enabled ^ val) {
2896                 if (val) {
2897                         tracer_enabled = 1;
2898                         if (current_trace->start)
2899                                 current_trace->start(tr);
2900                         tracing_start();
2901                 } else {
2902                         tracer_enabled = 0;
2903                         tracing_stop();
2904                         if (current_trace->stop)
2905                                 current_trace->stop(tr);
2906                 }
2907         }
2908         mutex_unlock(&trace_types_lock);
2909
2910         filp->f_pos += cnt;
2911
2912         return cnt;
2913 }
2914
2915 static ssize_t
2916 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2917                        size_t cnt, loff_t *ppos)
2918 {
2919         char buf[max_tracer_type_len+2];
2920         int r;
2921
2922         mutex_lock(&trace_types_lock);
2923         if (current_trace)
2924                 r = sprintf(buf, "%s\n", current_trace->name);
2925         else
2926                 r = sprintf(buf, "\n");
2927         mutex_unlock(&trace_types_lock);
2928
2929         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2930 }
2931
2932 static int tracing_set_tracer(char *buf)
2933 {
2934         struct trace_array *tr = &global_trace;
2935         struct tracer *t;
2936         int ret = 0;
2937
2938         mutex_lock(&trace_types_lock);
2939         for (t = trace_types; t; t = t->next) {
2940                 if (strcmp(t->name, buf) == 0)
2941                         break;
2942         }
2943         if (!t) {
2944                 ret = -EINVAL;
2945                 goto out;
2946         }
2947         if (t == current_trace)
2948                 goto out;
2949
2950         trace_branch_disable();
2951         if (current_trace && current_trace->reset)
2952                 current_trace->reset(tr);
2953
2954         current_trace = t;
2955         if (t->init) {
2956                 ret = t->init(tr);
2957                 if (ret)
2958                         goto out;
2959         }
2960
2961         trace_branch_enable(tr);
2962  out:
2963         mutex_unlock(&trace_types_lock);
2964
2965         return ret;
2966 }
2967
2968 static ssize_t
2969 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2970                         size_t cnt, loff_t *ppos)
2971 {
2972         char buf[max_tracer_type_len+1];
2973         int i;
2974         size_t ret;
2975         int err;
2976
2977         ret = cnt;
2978
2979         if (cnt > max_tracer_type_len)
2980                 cnt = max_tracer_type_len;
2981
2982         if (copy_from_user(&buf, ubuf, cnt))
2983                 return -EFAULT;
2984
2985         buf[cnt] = 0;
2986
2987         /* strip ending whitespace. */
2988         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2989                 buf[i] = 0;
2990
2991         err = tracing_set_tracer(buf);
2992         if (err)
2993                 return err;
2994
2995         filp->f_pos += ret;
2996
2997         return ret;
2998 }
2999
3000 static ssize_t
3001 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3002                      size_t cnt, loff_t *ppos)
3003 {
3004         unsigned long *ptr = filp->private_data;
3005         char buf[64];
3006         int r;
3007
3008         r = snprintf(buf, sizeof(buf), "%ld\n",
3009                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3010         if (r > sizeof(buf))
3011                 r = sizeof(buf);
3012         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3013 }
3014
3015 static ssize_t
3016 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3017                       size_t cnt, loff_t *ppos)
3018 {
3019         long *ptr = filp->private_data;
3020         char buf[64];
3021         long val;
3022         int ret;
3023
3024         if (cnt >= sizeof(buf))
3025                 return -EINVAL;
3026
3027         if (copy_from_user(&buf, ubuf, cnt))
3028                 return -EFAULT;
3029
3030         buf[cnt] = 0;
3031
3032         ret = strict_strtoul(buf, 10, &val);
3033         if (ret < 0)
3034                 return ret;
3035
3036         *ptr = val * 1000;
3037
3038         return cnt;
3039 }
3040
3041 static atomic_t tracing_reader;
3042
3043 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3044 {
3045         struct trace_iterator *iter;
3046
3047         if (tracing_disabled)
3048                 return -ENODEV;
3049
3050         /* We only allow for reader of the pipe */
3051         if (atomic_inc_return(&tracing_reader) != 1) {
3052                 atomic_dec(&tracing_reader);
3053                 return -EBUSY;
3054         }
3055
3056         /* create a buffer to store the information to pass to userspace */
3057         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3058         if (!iter)
3059                 return -ENOMEM;
3060
3061         mutex_lock(&trace_types_lock);
3062
3063         /* trace pipe does not show start of buffer */
3064         cpus_setall(iter->started);
3065
3066         iter->tr = &global_trace;
3067         iter->trace = current_trace;
3068         filp->private_data = iter;
3069
3070         if (iter->trace->pipe_open)
3071                 iter->trace->pipe_open(iter);
3072         mutex_unlock(&trace_types_lock);
3073
3074         return 0;
3075 }
3076
3077 static int tracing_release_pipe(struct inode *inode, struct file *file)
3078 {
3079         struct trace_iterator *iter = file->private_data;
3080
3081         kfree(iter);
3082         atomic_dec(&tracing_reader);
3083
3084         return 0;
3085 }
3086
3087 static unsigned int
3088 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3089 {
3090         struct trace_iterator *iter = filp->private_data;
3091
3092         if (trace_flags & TRACE_ITER_BLOCK) {
3093                 /*
3094                  * Always select as readable when in blocking mode
3095                  */
3096                 return POLLIN | POLLRDNORM;
3097         } else {
3098                 if (!trace_empty(iter))
3099                         return POLLIN | POLLRDNORM;
3100                 poll_wait(filp, &trace_wait, poll_table);
3101                 if (!trace_empty(iter))
3102                         return POLLIN | POLLRDNORM;
3103
3104                 return 0;
3105         }
3106 }
3107
3108 /*
3109  * Consumer reader.
3110  */
3111 static ssize_t
3112 tracing_read_pipe(struct file *filp, char __user *ubuf,
3113                   size_t cnt, loff_t *ppos)
3114 {
3115         struct trace_iterator *iter = filp->private_data;
3116         ssize_t sret;
3117
3118         /* return any leftover data */
3119         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3120         if (sret != -EBUSY)
3121                 return sret;
3122
3123         trace_seq_reset(&iter->seq);
3124
3125         mutex_lock(&trace_types_lock);
3126         if (iter->trace->read) {
3127                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3128                 if (sret)
3129                         goto out;
3130         }
3131
3132 waitagain:
3133         sret = 0;
3134         while (trace_empty(iter)) {
3135
3136                 if ((filp->f_flags & O_NONBLOCK)) {
3137                         sret = -EAGAIN;
3138                         goto out;
3139                 }
3140
3141                 /*
3142                  * This is a make-shift waitqueue. The reason we don't use
3143                  * an actual wait queue is because:
3144                  *  1) we only ever have one waiter
3145                  *  2) the tracing, traces all functions, we don't want
3146                  *     the overhead of calling wake_up and friends
3147                  *     (and tracing them too)
3148                  *     Anyway, this is really very primitive wakeup.
3149                  */
3150                 set_current_state(TASK_INTERRUPTIBLE);
3151                 iter->tr->waiter = current;
3152
3153                 mutex_unlock(&trace_types_lock);
3154
3155                 /* sleep for 100 msecs, and try again. */
3156                 schedule_timeout(HZ/10);
3157
3158                 mutex_lock(&trace_types_lock);
3159
3160                 iter->tr->waiter = NULL;
3161
3162                 if (signal_pending(current)) {
3163                         sret = -EINTR;
3164                         goto out;
3165                 }
3166
3167                 if (iter->trace != current_trace)
3168                         goto out;
3169
3170                 /*
3171                  * We block until we read something and tracing is disabled.
3172                  * We still block if tracing is disabled, but we have never
3173                  * read anything. This allows a user to cat this file, and
3174                  * then enable tracing. But after we have read something,
3175                  * we give an EOF when tracing is again disabled.
3176                  *
3177                  * iter->pos will be 0 if we haven't read anything.
3178                  */
3179                 if (!tracer_enabled && iter->pos)
3180                         break;
3181
3182                 continue;
3183         }
3184
3185         /* stop when tracing is finished */
3186         if (trace_empty(iter))
3187                 goto out;
3188
3189         if (cnt >= PAGE_SIZE)
3190                 cnt = PAGE_SIZE - 1;
3191
3192         /* reset all but tr, trace, and overruns */
3193         memset(&iter->seq, 0,
3194                sizeof(struct trace_iterator) -
3195                offsetof(struct trace_iterator, seq));
3196         iter->pos = -1;
3197
3198         while (find_next_entry_inc(iter) != NULL) {
3199                 enum print_line_t ret;
3200                 int len = iter->seq.len;
3201
3202                 ret = print_trace_line(iter);
3203                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3204                         /* don't print partial lines */
3205                         iter->seq.len = len;
3206                         break;
3207                 }
3208
3209                 trace_consume(iter);
3210
3211                 if (iter->seq.len >= cnt)
3212                         break;
3213         }
3214
3215         /* Now copy what we have to the user */
3216         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3217         if (iter->seq.readpos >= iter->seq.len)
3218                 trace_seq_reset(&iter->seq);
3219
3220         /*
3221          * If there was nothing to send to user, inspite of consuming trace
3222          * entries, go back to wait for more entries.
3223          */
3224         if (sret == -EBUSY)
3225                 goto waitagain;
3226
3227 out:
3228         mutex_unlock(&trace_types_lock);
3229
3230         return sret;
3231 }
3232
3233 static ssize_t
3234 tracing_entries_read(struct file *filp, char __user *ubuf,
3235                      size_t cnt, loff_t *ppos)
3236 {
3237         struct trace_array *tr = filp->private_data;
3238         char buf[64];
3239         int r;
3240
3241         r = sprintf(buf, "%lu\n", tr->entries >> 10);
3242         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3243 }
3244
3245 static ssize_t
3246 tracing_entries_write(struct file *filp, const char __user *ubuf,
3247                       size_t cnt, loff_t *ppos)
3248 {
3249         unsigned long val;
3250         char buf[64];
3251         int ret, cpu;
3252
3253         if (cnt >= sizeof(buf))
3254                 return -EINVAL;
3255
3256         if (copy_from_user(&buf, ubuf, cnt))
3257                 return -EFAULT;
3258
3259         buf[cnt] = 0;
3260
3261         ret = strict_strtoul(buf, 10, &val);
3262         if (ret < 0)
3263                 return ret;
3264
3265         /* must have at least 1 entry */
3266         if (!val)
3267                 return -EINVAL;
3268
3269         mutex_lock(&trace_types_lock);
3270
3271         tracing_stop();
3272
3273         /* disable all cpu buffers */
3274         for_each_tracing_cpu(cpu) {
3275                 if (global_trace.data[cpu])
3276                         atomic_inc(&global_trace.data[cpu]->disabled);
3277                 if (max_tr.data[cpu])
3278                         atomic_inc(&max_tr.data[cpu]->disabled);
3279         }
3280
3281         /* value is in KB */
3282         val <<= 10;
3283
3284         if (val != global_trace.entries) {
3285                 ret = ring_buffer_resize(global_trace.buffer, val);
3286                 if (ret < 0) {
3287                         cnt = ret;
3288                         goto out;
3289                 }
3290
3291                 ret = ring_buffer_resize(max_tr.buffer, val);
3292                 if (ret < 0) {
3293                         int r;
3294                         cnt = ret;
3295                         r = ring_buffer_resize(global_trace.buffer,
3296                                                global_trace.entries);
3297                         if (r < 0) {
3298                                 /* AARGH! We are left with different
3299                                  * size max buffer!!!! */
3300                                 WARN_ON(1);
3301                                 tracing_disabled = 1;
3302                         }
3303                         goto out;
3304                 }
3305
3306                 global_trace.entries = val;
3307         }
3308
3309         filp->f_pos += cnt;
3310
3311         /* If check pages failed, return ENOMEM */
3312         if (tracing_disabled)
3313                 cnt = -ENOMEM;
3314  out:
3315         for_each_tracing_cpu(cpu) {
3316                 if (global_trace.data[cpu])
3317                         atomic_dec(&global_trace.data[cpu]->disabled);
3318                 if (max_tr.data[cpu])
3319                         atomic_dec(&max_tr.data[cpu]->disabled);
3320         }
3321
3322         tracing_start();
3323         max_tr.entries = global_trace.entries;
3324         mutex_unlock(&trace_types_lock);
3325
3326         return cnt;
3327 }
3328
3329 static int mark_printk(const char *fmt, ...)
3330 {
3331         int ret;
3332         va_list args;
3333         va_start(args, fmt);
3334         ret = trace_vprintk(0, fmt, args);
3335         va_end(args);
3336         return ret;
3337 }
3338
3339 static ssize_t
3340 tracing_mark_write(struct file *filp, const char __user *ubuf,
3341                                         size_t cnt, loff_t *fpos)
3342 {
3343         char *buf;
3344         char *end;
3345
3346         if (tracing_disabled)
3347                 return -EINVAL;
3348
3349         if (cnt > TRACE_BUF_SIZE)
3350                 cnt = TRACE_BUF_SIZE;
3351
3352         buf = kmalloc(cnt + 1, GFP_KERNEL);
3353         if (buf == NULL)
3354                 return -ENOMEM;
3355
3356         if (copy_from_user(buf, ubuf, cnt)) {
3357                 kfree(buf);
3358                 return -EFAULT;
3359         }
3360
3361         /* Cut from the first nil or newline. */
3362         buf[cnt] = '\0';
3363         end = strchr(buf, '\n');
3364         if (end)
3365                 *end = '\0';
3366
3367         cnt = mark_printk("%s\n", buf);
3368         kfree(buf);
3369         *fpos += cnt;
3370
3371         return cnt;
3372 }
3373
3374 static struct file_operations tracing_max_lat_fops = {
3375         .open           = tracing_open_generic,
3376         .read           = tracing_max_lat_read,
3377         .write          = tracing_max_lat_write,
3378 };
3379
3380 static struct file_operations tracing_ctrl_fops = {
3381         .open           = tracing_open_generic,
3382         .read           = tracing_ctrl_read,
3383         .write          = tracing_ctrl_write,
3384 };
3385
3386 static struct file_operations set_tracer_fops = {
3387         .open           = tracing_open_generic,
3388         .read           = tracing_set_trace_read,
3389         .write          = tracing_set_trace_write,
3390 };
3391
3392 static struct file_operations tracing_pipe_fops = {
3393         .open           = tracing_open_pipe,
3394         .poll           = tracing_poll_pipe,
3395         .read           = tracing_read_pipe,
3396         .release        = tracing_release_pipe,
3397 };
3398
3399 static struct file_operations tracing_entries_fops = {
3400         .open           = tracing_open_generic,
3401         .read           = tracing_entries_read,
3402         .write          = tracing_entries_write,
3403 };
3404
3405 static struct file_operations tracing_mark_fops = {
3406         .open           = tracing_open_generic,
3407         .write          = tracing_mark_write,
3408 };
3409
3410 #ifdef CONFIG_DYNAMIC_FTRACE
3411
3412 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3413 {
3414         return 0;
3415 }
3416
3417 static ssize_t
3418 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3419                   size_t cnt, loff_t *ppos)
3420 {
3421         static char ftrace_dyn_info_buffer[1024];
3422         static DEFINE_MUTEX(dyn_info_mutex);
3423         unsigned long *p = filp->private_data;
3424         char *buf = ftrace_dyn_info_buffer;
3425         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3426         int r;
3427
3428         mutex_lock(&dyn_info_mutex);
3429         r = sprintf(buf, "%ld ", *p);
3430
3431         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3432         buf[r++] = '\n';
3433
3434         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3435
3436         mutex_unlock(&dyn_info_mutex);
3437
3438         return r;
3439 }
3440
3441 static struct file_operations tracing_dyn_info_fops = {
3442         .open           = tracing_open_generic,
3443         .read           = tracing_read_dyn_info,
3444 };
3445 #endif
3446
3447 static struct dentry *d_tracer;
3448
3449 struct dentry *tracing_init_dentry(void)
3450 {
3451         static int once;
3452
3453         if (d_tracer)
3454                 return d_tracer;
3455
3456         d_tracer = debugfs_create_dir("tracing", NULL);
3457
3458         if (!d_tracer && !once) {
3459                 once = 1;
3460                 pr_warning("Could not create debugfs directory 'tracing'\n");
3461                 return NULL;
3462         }
3463
3464         return d_tracer;
3465 }
3466
3467 #ifdef CONFIG_FTRACE_SELFTEST
3468 /* Let selftest have access to static functions in this file */
3469 #include "trace_selftest.c"
3470 #endif
3471
3472 static __init int tracer_init_debugfs(void)
3473 {
3474         struct dentry *d_tracer;
3475         struct dentry *entry;
3476
3477         d_tracer = tracing_init_dentry();
3478
3479         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3480                                     &global_trace, &tracing_ctrl_fops);
3481         if (!entry)
3482                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3483
3484         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3485                                     NULL, &tracing_iter_fops);
3486         if (!entry)
3487                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3488
3489         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3490                                     NULL, &tracing_cpumask_fops);
3491         if (!entry)
3492                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3493
3494         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3495                                     &global_trace, &tracing_lt_fops);
3496         if (!entry)
3497                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3498
3499         entry = debugfs_create_file("trace", 0444, d_tracer,
3500                                     &global_trace, &tracing_fops);
3501         if (!entry)
3502                 pr_warning("Could not create debugfs 'trace' entry\n");
3503
3504         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3505                                     &global_trace, &show_traces_fops);
3506         if (!entry)
3507                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3508
3509         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3510                                     &global_trace, &set_tracer_fops);
3511         if (!entry)
3512                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3513
3514         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3515                                     &tracing_max_latency,
3516                                     &tracing_max_lat_fops);
3517         if (!entry)
3518                 pr_warning("Could not create debugfs "
3519                            "'tracing_max_latency' entry\n");
3520
3521         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3522                                     &tracing_thresh, &tracing_max_lat_fops);
3523         if (!entry)
3524                 pr_warning("Could not create debugfs "
3525                            "'tracing_thresh' entry\n");
3526         entry = debugfs_create_file("README", 0644, d_tracer,
3527                                     NULL, &tracing_readme_fops);
3528         if (!entry)
3529                 pr_warning("Could not create debugfs 'README' entry\n");
3530
3531         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3532                                     NULL, &tracing_pipe_fops);
3533         if (!entry)
3534                 pr_warning("Could not create debugfs "
3535                            "'trace_pipe' entry\n");
3536
3537         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3538                                     &global_trace, &tracing_entries_fops);
3539         if (!entry)
3540                 pr_warning("Could not create debugfs "
3541                            "'buffer_size_kb' entry\n");
3542
3543         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3544                                     NULL, &tracing_mark_fops);
3545         if (!entry)
3546                 pr_warning("Could not create debugfs "
3547                            "'trace_marker' entry\n");
3548
3549 #ifdef CONFIG_DYNAMIC_FTRACE
3550         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3551                                     &ftrace_update_tot_cnt,
3552                                     &tracing_dyn_info_fops);
3553         if (!entry)
3554                 pr_warning("Could not create debugfs "
3555                            "'dyn_ftrace_total_info' entry\n");
3556 #endif
3557 #ifdef CONFIG_SYSPROF_TRACER
3558         init_tracer_sysprof_debugfs(d_tracer);
3559 #endif
3560         return 0;
3561 }
3562
3563 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3564 {
3565         static DEFINE_SPINLOCK(trace_buf_lock);
3566         static char trace_buf[TRACE_BUF_SIZE];
3567
3568         struct ring_buffer_event *event;
3569         struct trace_array *tr = &global_trace;
3570         struct trace_array_cpu *data;
3571         struct print_entry *entry;
3572         unsigned long flags, irq_flags;
3573         int cpu, len = 0, size, pc;
3574
3575         if (tracing_disabled)
3576                 return 0;
3577
3578         pc = preempt_count();
3579         preempt_disable_notrace();
3580         cpu = raw_smp_processor_id();
3581         data = tr->data[cpu];
3582
3583         if (unlikely(atomic_read(&data->disabled)))
3584                 goto out;
3585
3586         spin_lock_irqsave(&trace_buf_lock, flags);
3587         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3588
3589         len = min(len, TRACE_BUF_SIZE-1);
3590         trace_buf[len] = 0;
3591
3592         size = sizeof(*entry) + len + 1;
3593         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3594         if (!event)
3595                 goto out_unlock;
3596         entry = ring_buffer_event_data(event);
3597         tracing_generic_entry_update(&entry->ent, flags, pc);
3598         entry->ent.type                 = TRACE_PRINT;
3599         entry->ip                       = ip;
3600
3601         memcpy(&entry->buf, trace_buf, len);
3602         entry->buf[len] = 0;
3603         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3604
3605  out_unlock:
3606         spin_unlock_irqrestore(&trace_buf_lock, flags);
3607
3608  out:
3609         preempt_enable_notrace();
3610
3611         return len;
3612 }
3613 EXPORT_SYMBOL_GPL(trace_vprintk);
3614
3615 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3616 {
3617         int ret;
3618         va_list ap;
3619
3620         if (!(trace_flags & TRACE_ITER_PRINTK))
3621                 return 0;
3622
3623         va_start(ap, fmt);
3624         ret = trace_vprintk(ip, fmt, ap);
3625         va_end(ap);
3626         return ret;
3627 }
3628 EXPORT_SYMBOL_GPL(__ftrace_printk);
3629
3630 static int trace_panic_handler(struct notifier_block *this,
3631                                unsigned long event, void *unused)
3632 {
3633         if (ftrace_dump_on_oops)
3634                 ftrace_dump();
3635         return NOTIFY_OK;
3636 }
3637
3638 static struct notifier_block trace_panic_notifier = {
3639         .notifier_call  = trace_panic_handler,
3640         .next           = NULL,
3641         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3642 };
3643
3644 static int trace_die_handler(struct notifier_block *self,
3645                              unsigned long val,
3646                              void *data)
3647 {
3648         switch (val) {
3649         case DIE_OOPS:
3650                 if (ftrace_dump_on_oops)
3651                         ftrace_dump();
3652                 break;
3653         default:
3654                 break;
3655         }
3656         return NOTIFY_OK;
3657 }
3658
3659 static struct notifier_block trace_die_notifier = {
3660         .notifier_call = trace_die_handler,
3661         .priority = 200
3662 };
3663
3664 /*
3665  * printk is set to max of 1024, we really don't need it that big.
3666  * Nothing should be printing 1000 characters anyway.
3667  */
3668 #define TRACE_MAX_PRINT         1000
3669
3670 /*
3671  * Define here KERN_TRACE so that we have one place to modify
3672  * it if we decide to change what log level the ftrace dump
3673  * should be at.
3674  */
3675 #define KERN_TRACE              KERN_INFO
3676
3677 static void
3678 trace_printk_seq(struct trace_seq *s)
3679 {
3680         /* Probably should print a warning here. */
3681         if (s->len >= 1000)
3682                 s->len = 1000;
3683
3684         /* should be zero ended, but we are paranoid. */
3685         s->buffer[s->len] = 0;
3686
3687         printk(KERN_TRACE "%s", s->buffer);
3688
3689         trace_seq_reset(s);
3690 }
3691
3692 void ftrace_dump(void)
3693 {
3694         static DEFINE_SPINLOCK(ftrace_dump_lock);
3695         /* use static because iter can be a bit big for the stack */
3696         static struct trace_iterator iter;
3697         static cpumask_t mask;
3698         static int dump_ran;
3699         unsigned long flags;
3700         int cnt = 0, cpu;
3701
3702         /* only one dump */
3703         spin_lock_irqsave(&ftrace_dump_lock, flags);
3704         if (dump_ran)
3705                 goto out;
3706
3707         dump_ran = 1;
3708
3709         /* No turning back! */
3710         ftrace_kill();
3711
3712         for_each_tracing_cpu(cpu) {
3713                 atomic_inc(&global_trace.data[cpu]->disabled);
3714         }
3715
3716         /* don't look at user memory in panic mode */
3717         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3718
3719         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3720
3721         iter.tr = &global_trace;
3722         iter.trace = current_trace;
3723
3724         /*
3725          * We need to stop all tracing on all CPUS to read the
3726          * the next buffer. This is a bit expensive, but is
3727          * not done often. We fill all what we can read,
3728          * and then release the locks again.
3729          */
3730
3731         cpus_clear(mask);
3732
3733         while (!trace_empty(&iter)) {
3734
3735                 if (!cnt)
3736                         printk(KERN_TRACE "---------------------------------\n");
3737
3738                 cnt++;
3739
3740                 /* reset all but tr, trace, and overruns */
3741                 memset(&iter.seq, 0,
3742                        sizeof(struct trace_iterator) -
3743                        offsetof(struct trace_iterator, seq));
3744                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3745                 iter.pos = -1;
3746
3747                 if (find_next_entry_inc(&iter) != NULL) {
3748                         print_trace_line(&iter);
3749                         trace_consume(&iter);
3750                 }
3751
3752                 trace_printk_seq(&iter.seq);
3753         }
3754
3755         if (!cnt)
3756                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3757         else
3758                 printk(KERN_TRACE "---------------------------------\n");
3759
3760  out:
3761         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3762 }
3763
3764 __init static int tracer_alloc_buffers(void)
3765 {
3766         struct trace_array_cpu *data;
3767         int i;
3768
3769         /* TODO: make the number of buffers hot pluggable with CPUS */
3770         tracing_buffer_mask = cpu_possible_map;
3771
3772         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3773                                                    TRACE_BUFFER_FLAGS);
3774         if (!global_trace.buffer) {
3775                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3776                 WARN_ON(1);
3777                 return 0;
3778         }
3779         global_trace.entries = ring_buffer_size(global_trace.buffer);
3780
3781 #ifdef CONFIG_TRACER_MAX_TRACE
3782         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3783                                              TRACE_BUFFER_FLAGS);
3784         if (!max_tr.buffer) {
3785                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3786                 WARN_ON(1);
3787                 ring_buffer_free(global_trace.buffer);
3788                 return 0;
3789         }
3790         max_tr.entries = ring_buffer_size(max_tr.buffer);
3791         WARN_ON(max_tr.entries != global_trace.entries);
3792 #endif
3793
3794         /* Allocate the first page for all buffers */
3795         for_each_tracing_cpu(i) {
3796                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3797                 max_tr.data[i] = &per_cpu(max_data, i);
3798         }
3799
3800         trace_init_cmdlines();
3801
3802         register_tracer(&nop_trace);
3803 #ifdef CONFIG_BOOT_TRACER
3804         register_tracer(&boot_tracer);
3805         current_trace = &boot_tracer;
3806         current_trace->init(&global_trace);
3807 #else
3808         current_trace = &nop_trace;
3809 #endif
3810
3811         /* All seems OK, enable tracing */
3812         tracing_disabled = 0;
3813
3814         atomic_notifier_chain_register(&panic_notifier_list,
3815                                        &trace_panic_notifier);
3816
3817         register_die_notifier(&trace_die_notifier);
3818
3819         return 0;
3820 }
3821 early_initcall(tracer_alloc_buffers);
3822 fs_initcall(tracer_init_debugfs);