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 #ifdef CONFIG_STACKTRACE
 987         struct ring_buffer_event *event;
 988         struct userstack_entry *entry;
 989         struct stack_trace trace;
 990         unsigned long irq_flags;
 991
 992         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
 993                 return;
 994
 995         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 996                                          &irq_flags);
 997         if (!event)
 998                 return;
 999         entry   = ring_buffer_event_data(event);
1000         tracing_generic_entry_update(&entry->ent, flags, pc);
1001         entry->ent.type         = TRACE_USER_STACK;
1002
1003         memset(&entry->caller, 0, sizeof(entry->caller));
1004
1005         trace.nr_entries        = 0;
1006         trace.max_entries       = FTRACE_STACK_ENTRIES;
1007         trace.skip              = 0;
1008         trace.entries           = entry->caller;
1009
1010         save_stack_trace_user(&trace);
1011         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1012 #endif
1013 }
1014
1015 void __trace_userstack(struct trace_array *tr,
1016                    struct trace_array_cpu *data,
1017                    unsigned long flags)
1018 {
1019         ftrace_trace_userstack(tr, data, flags, preempt_count());
1020 }
1021
1022 static void
1023 ftrace_trace_special(void *__tr, void *__data,
1024                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1025                      int pc)
1026 {
1027         struct ring_buffer_event *event;
1028         struct trace_array_cpu *data = __data;
1029         struct trace_array *tr = __tr;
1030         struct special_entry *entry;
1031         unsigned long irq_flags;
1032
1033         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1034                                          &irq_flags);
1035         if (!event)
1036                 return;
1037         entry   = ring_buffer_event_data(event);
1038         tracing_generic_entry_update(&entry->ent, 0, pc);
1039         entry->ent.type                 = TRACE_SPECIAL;
1040         entry->arg1                     = arg1;
1041         entry->arg2                     = arg2;
1042         entry->arg3                     = arg3;
1043         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1044         ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1045         ftrace_trace_userstack(tr, data, irq_flags, pc);
1046
1047         trace_wake_up();
1048 }
1049
1050 void
1051 __trace_special(void *__tr, void *__data,
1052                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1053 {
1054         ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1055 }
1056
1057 void
1058 tracing_sched_switch_trace(struct trace_array *tr,
1059                            struct trace_array_cpu *data,
1060                            struct task_struct *prev,
1061                            struct task_struct *next,
1062                            unsigned long flags, int pc)
1063 {
1064         struct ring_buffer_event *event;
1065         struct ctx_switch_entry *entry;
1066         unsigned long irq_flags;
1067
1068         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1069                                            &irq_flags);
1070         if (!event)
1071                 return;
1072         entry   = ring_buffer_event_data(event);
1073         tracing_generic_entry_update(&entry->ent, flags, pc);
1074         entry->ent.type                 = TRACE_CTX;
1075         entry->prev_pid                 = prev->pid;
1076         entry->prev_prio                = prev->prio;
1077         entry->prev_state               = prev->state;
1078         entry->next_pid                 = next->pid;
1079         entry->next_prio                = next->prio;
1080         entry->next_state               = next->state;
1081         entry->next_cpu = task_cpu(next);
1082         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1083         ftrace_trace_stack(tr, data, flags, 5, pc);
1084         ftrace_trace_userstack(tr, data, flags, pc);
1085 }
1086
1087 void
1088 tracing_sched_wakeup_trace(struct trace_array *tr,
1089                            struct trace_array_cpu *data,
1090                            struct task_struct *wakee,
1091                            struct task_struct *curr,
1092                            unsigned long flags, int pc)
1093 {
1094         struct ring_buffer_event *event;
1095         struct ctx_switch_entry *entry;
1096         unsigned long irq_flags;
1097
1098         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1099                                            &irq_flags);
1100         if (!event)
1101                 return;
1102         entry   = ring_buffer_event_data(event);
1103         tracing_generic_entry_update(&entry->ent, flags, pc);
1104         entry->ent.type                 = TRACE_WAKE;
1105         entry->prev_pid                 = curr->pid;
1106         entry->prev_prio                = curr->prio;
1107         entry->prev_state               = curr->state;
1108         entry->next_pid                 = wakee->pid;
1109         entry->next_prio                = wakee->prio;
1110         entry->next_state               = wakee->state;
1111         entry->next_cpu                 = task_cpu(wakee);
1112         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1113         ftrace_trace_stack(tr, data, flags, 6, pc);
1114         ftrace_trace_userstack(tr, data, flags, pc);
1115
1116         trace_wake_up();
1117 }
1118
1119 void
1120 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1121 {
1122         struct trace_array *tr = &global_trace;
1123         struct trace_array_cpu *data;
1124         unsigned long flags;
1125         int cpu;
1126         int pc;
1127
1128         if (tracing_disabled)
1129                 return;
1130
1131         pc = preempt_count();
1132         local_irq_save(flags);
1133         cpu = raw_smp_processor_id();
1134         data = tr->data[cpu];
1135
1136         if (likely(atomic_inc_return(&data->disabled) == 1))
1137                 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1138
1139         atomic_dec(&data->disabled);
1140         local_irq_restore(flags);
1141 }
1142
1143 #ifdef CONFIG_FUNCTION_TRACER
1144 static void
1145 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1146 {
1147         struct trace_array *tr = &global_trace;
1148         struct trace_array_cpu *data;
1149         unsigned long flags;
1150         long disabled;
1151         int cpu, resched;
1152         int pc;
1153
1154         if (unlikely(!ftrace_function_enabled))
1155                 return;
1156
1157         pc = preempt_count();
1158         resched = ftrace_preempt_disable();
1159         local_save_flags(flags);
1160         cpu = raw_smp_processor_id();
1161         data = tr->data[cpu];
1162         disabled = atomic_inc_return(&data->disabled);
1163
1164         if (likely(disabled == 1))
1165                 trace_function(tr, data, ip, parent_ip, flags, pc);
1166
1167         atomic_dec(&data->disabled);
1168         ftrace_preempt_enable(resched);
1169 }
1170
1171 static void
1172 function_trace_call(unsigned long ip, unsigned long parent_ip)
1173 {
1174         struct trace_array *tr = &global_trace;
1175         struct trace_array_cpu *data;
1176         unsigned long flags;
1177         long disabled;
1178         int cpu;
1179         int pc;
1180
1181         if (unlikely(!ftrace_function_enabled))
1182                 return;
1183
1184         /*
1185          * Need to use raw, since this must be called before the
1186          * recursive protection is performed.
1187          */
1188         local_irq_save(flags);
1189         cpu = raw_smp_processor_id();
1190         data = tr->data[cpu];
1191         disabled = atomic_inc_return(&data->disabled);
1192
1193         if (likely(disabled == 1)) {
1194                 pc = preempt_count();
1195                 trace_function(tr, data, ip, parent_ip, flags, pc);
1196         }
1197
1198         atomic_dec(&data->disabled);
1199         local_irq_restore(flags);
1200 }
1201
1202 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1203 int trace_graph_entry(struct ftrace_graph_ent *trace)
1204 {
1205         struct trace_array *tr = &global_trace;
1206         struct trace_array_cpu *data;
1207         unsigned long flags;
1208         long disabled;
1209         int cpu;
1210         int pc;
1211
1212         if (!ftrace_graph_addr(trace->func))
1213                 return 0;
1214
1215         local_irq_save(flags);
1216         cpu = raw_smp_processor_id();
1217         data = tr->data[cpu];
1218         disabled = atomic_inc_return(&data->disabled);
1219         if (likely(disabled == 1)) {
1220                 pc = preempt_count();
1221                 __trace_graph_entry(tr, data, trace, flags, pc);
1222         }
1223         /* Only do the atomic if it is not already set */
1224         if (!test_tsk_trace_graph(current))
1225                 set_tsk_trace_graph(current);
1226         atomic_dec(&data->disabled);
1227         local_irq_restore(flags);
1228
1229         return 1;
1230 }
1231
1232 void trace_graph_return(struct ftrace_graph_ret *trace)
1233 {
1234         struct trace_array *tr = &global_trace;
1235         struct trace_array_cpu *data;
1236         unsigned long flags;
1237         long disabled;
1238         int cpu;
1239         int pc;
1240
1241         local_irq_save(flags);
1242         cpu = raw_smp_processor_id();
1243         data = tr->data[cpu];
1244         disabled = atomic_inc_return(&data->disabled);
1245         if (likely(disabled == 1)) {
1246                 pc = preempt_count();
1247                 __trace_graph_return(tr, data, trace, flags, pc);
1248         }
1249         if (!trace->depth)
1250                 clear_tsk_trace_graph(current);
1251         atomic_dec(&data->disabled);
1252         local_irq_restore(flags);
1253 }
1254 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1255
1256 static struct ftrace_ops trace_ops __read_mostly =
1257 {
1258         .func = function_trace_call,
1259 };
1260
1261 void tracing_start_function_trace(void)
1262 {
1263         ftrace_function_enabled = 0;
1264
1265         if (trace_flags & TRACE_ITER_PREEMPTONLY)
1266                 trace_ops.func = function_trace_call_preempt_only;
1267         else
1268                 trace_ops.func = function_trace_call;
1269
1270         register_ftrace_function(&trace_ops);
1271         ftrace_function_enabled = 1;
1272 }
1273
1274 void tracing_stop_function_trace(void)
1275 {
1276         ftrace_function_enabled = 0;
1277         unregister_ftrace_function(&trace_ops);
1278 }
1279 #endif
1280
1281 enum trace_file_type {
1282         TRACE_FILE_LAT_FMT      = 1,
1283         TRACE_FILE_ANNOTATE     = 2,
1284 };
1285
1286 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1287 {
1288         /* Don't allow ftrace to trace into the ring buffers */
1289         ftrace_disable_cpu();
1290
1291         iter->idx++;
1292         if (iter->buffer_iter[iter->cpu])
1293                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1294
1295         ftrace_enable_cpu();
1296 }
1297
1298 static struct trace_entry *
1299 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1300 {
1301         struct ring_buffer_event *event;
1302         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1303
1304         /* Don't allow ftrace to trace into the ring buffers */
1305         ftrace_disable_cpu();
1306
1307         if (buf_iter)
1308                 event = ring_buffer_iter_peek(buf_iter, ts);
1309         else
1310                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1311
1312         ftrace_enable_cpu();
1313
1314         return event ? ring_buffer_event_data(event) : NULL;
1315 }
1316
1317 static struct trace_entry *
1318 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1319 {
1320         struct ring_buffer *buffer = iter->tr->buffer;
1321         struct trace_entry *ent, *next = NULL;
1322         u64 next_ts = 0, ts;
1323         int next_cpu = -1;
1324         int cpu;
1325
1326         for_each_tracing_cpu(cpu) {
1327
1328                 if (ring_buffer_empty_cpu(buffer, cpu))
1329                         continue;
1330
1331                 ent = peek_next_entry(iter, cpu, &ts);
1332
1333                 /*
1334                  * Pick the entry with the smallest timestamp:
1335                  */
1336                 if (ent && (!next || ts < next_ts)) {
1337                         next = ent;
1338                         next_cpu = cpu;
1339                         next_ts = ts;
1340                 }
1341         }
1342
1343         if (ent_cpu)
1344                 *ent_cpu = next_cpu;
1345
1346         if (ent_ts)
1347                 *ent_ts = next_ts;
1348
1349         return next;
1350 }
1351
1352 /* Find the next real entry, without updating the iterator itself */
1353 static struct trace_entry *
1354 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1355 {
1356         return __find_next_entry(iter, ent_cpu, ent_ts);
1357 }
1358
1359 /* Find the next real entry, and increment the iterator to the next entry */
1360 static void *find_next_entry_inc(struct trace_iterator *iter)
1361 {
1362         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1363
1364         if (iter->ent)
1365                 trace_iterator_increment(iter, iter->cpu);
1366
1367         return iter->ent ? iter : NULL;
1368 }
1369
1370 static void trace_consume(struct trace_iterator *iter)
1371 {
1372         /* Don't allow ftrace to trace into the ring buffers */
1373         ftrace_disable_cpu();
1374         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1375         ftrace_enable_cpu();
1376 }
1377
1378 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1379 {
1380         struct trace_iterator *iter = m->private;
1381         int i = (int)*pos;
1382         void *ent;
1383
1384         (*pos)++;
1385
1386         /* can't go backwards */
1387         if (iter->idx > i)
1388                 return NULL;
1389
1390         if (iter->idx < 0)
1391                 ent = find_next_entry_inc(iter);
1392         else
1393                 ent = iter;
1394
1395         while (ent && iter->idx < i)
1396                 ent = find_next_entry_inc(iter);
1397
1398         iter->pos = *pos;
1399
1400         return ent;
1401 }
1402
1403 static void *s_start(struct seq_file *m, loff_t *pos)
1404 {
1405         struct trace_iterator *iter = m->private;
1406         void *p = NULL;
1407         loff_t l = 0;
1408         int cpu;
1409
1410         mutex_lock(&trace_types_lock);
1411
1412         if (!current_trace || current_trace != iter->trace) {
1413                 mutex_unlock(&trace_types_lock);
1414                 return NULL;
1415         }
1416
1417         atomic_inc(&trace_record_cmdline_disabled);
1418
1419         if (*pos != iter->pos) {
1420                 iter->ent = NULL;
1421                 iter->cpu = 0;
1422                 iter->idx = -1;
1423
1424                 ftrace_disable_cpu();
1425
1426                 for_each_tracing_cpu(cpu) {
1427                         ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1428                 }
1429
1430                 ftrace_enable_cpu();
1431
1432                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1433                         ;
1434
1435         } else {
1436                 l = *pos - 1;
1437                 p = s_next(m, p, &l);
1438         }
1439
1440         return p;
1441 }
1442
1443 static void s_stop(struct seq_file *m, void *p)
1444 {
1445         atomic_dec(&trace_record_cmdline_disabled);
1446         mutex_unlock(&trace_types_lock);
1447 }
1448
1449 #ifdef CONFIG_KRETPROBES
1450 static inline const char *kretprobed(const char *name)
1451 {
1452         static const char tramp_name[] = "kretprobe_trampoline";
1453         int size = sizeof(tramp_name);
1454
1455         if (strncmp(tramp_name, name, size) == 0)
1456                 return "[unknown/kretprobe'd]";
1457         return name;
1458 }
1459 #else
1460 static inline const char *kretprobed(const char *name)
1461 {
1462         return name;
1463 }
1464 #endif /* CONFIG_KRETPROBES */
1465
1466 static int
1467 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1468 {
1469 #ifdef CONFIG_KALLSYMS
1470         char str[KSYM_SYMBOL_LEN];
1471         const char *name;
1472
1473         kallsyms_lookup(address, NULL, NULL, NULL, str);
1474
1475         name = kretprobed(str);
1476
1477         return trace_seq_printf(s, fmt, name);
1478 #endif
1479         return 1;
1480 }
1481
1482 static int
1483 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1484                      unsigned long address)
1485 {
1486 #ifdef CONFIG_KALLSYMS
1487         char str[KSYM_SYMBOL_LEN];
1488         const char *name;
1489
1490         sprint_symbol(str, address);
1491         name = kretprobed(str);
1492
1493         return trace_seq_printf(s, fmt, name);
1494 #endif
1495         return 1;
1496 }
1497
1498 #ifndef CONFIG_64BIT
1499 # define IP_FMT "%08lx"
1500 #else
1501 # define IP_FMT "%016lx"
1502 #endif
1503
1504 int
1505 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1506 {
1507         int ret;
1508
1509         if (!ip)
1510                 return trace_seq_printf(s, "0");
1511
1512         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1513                 ret = seq_print_sym_offset(s, "%s", ip);
1514         else
1515                 ret = seq_print_sym_short(s, "%s", ip);
1516
1517         if (!ret)
1518                 return 0;
1519
1520         if (sym_flags & TRACE_ITER_SYM_ADDR)
1521                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1522         return ret;
1523 }
1524
1525 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1526                                     unsigned long ip, unsigned long sym_flags)
1527 {
1528         struct file *file = NULL;
1529         unsigned long vmstart = 0;
1530         int ret = 1;
1531
1532         if (mm) {
1533                 const struct vm_area_struct *vma;
1534
1535                 down_read(&mm->mmap_sem);
1536                 vma = find_vma(mm, ip);
1537                 if (vma) {
1538                         file = vma->vm_file;
1539                         vmstart = vma->vm_start;
1540                 }
1541                 if (file) {
1542                         ret = trace_seq_path(s, &file->f_path);
1543                         if (ret)
1544                                 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1545                 }
1546                 up_read(&mm->mmap_sem);
1547         }
1548         if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1549                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1550         return ret;
1551 }
1552
1553 static int
1554 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1555                       unsigned long sym_flags)
1556 {
1557         struct mm_struct *mm = NULL;
1558         int ret = 1;
1559         unsigned int i;
1560
1561         if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1562                 struct task_struct *task;
1563                 /*
1564                  * we do the lookup on the thread group leader,
1565                  * since individual threads might have already quit!
1566                  */
1567                 rcu_read_lock();
1568                 task = find_task_by_vpid(entry->ent.tgid);
1569                 if (task)
1570                         mm = get_task_mm(task);
1571                 rcu_read_unlock();
1572         }
1573
1574         for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1575                 unsigned long ip = entry->caller[i];
1576
1577                 if (ip == ULONG_MAX || !ret)
1578                         break;
1579                 if (i && ret)
1580                         ret = trace_seq_puts(s, " <- ");
1581                 if (!ip) {
1582                         if (ret)
1583                                 ret = trace_seq_puts(s, "??");
1584                         continue;
1585                 }
1586                 if (!ret)
1587                         break;
1588                 if (ret)
1589                         ret = seq_print_user_ip(s, mm, ip, sym_flags);
1590         }
1591
1592         if (mm)
1593                 mmput(mm);
1594         return ret;
1595 }
1596
1597 static void print_lat_help_header(struct seq_file *m)
1598 {
1599         seq_puts(m, "#                  _------=> CPU#            \n");
1600         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1601         seq_puts(m, "#                | / _----=> need-resched    \n");
1602         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1603         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1604         seq_puts(m, "#                |||| /                      \n");
1605         seq_puts(m, "#                |||||     delay             \n");
1606         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1607         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1608 }
1609
1610 static void print_func_help_header(struct seq_file *m)
1611 {
1612         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1613         seq_puts(m, "#              | |       |          |         |\n");
1614 }
1615
1616
1617 static void
1618 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1619 {
1620         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1621         struct trace_array *tr = iter->tr;
1622         struct trace_array_cpu *data = tr->data[tr->cpu];
1623         struct tracer *type = current_trace;
1624         unsigned long total;
1625         unsigned long entries;
1626         const char *name = "preemption";
1627
1628         if (type)
1629                 name = type->name;
1630
1631         entries = ring_buffer_entries(iter->tr->buffer);
1632         total = entries +
1633                 ring_buffer_overruns(iter->tr->buffer);
1634
1635         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1636                    name, UTS_RELEASE);
1637         seq_puts(m, "-----------------------------------"
1638                  "---------------------------------\n");
1639         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1640                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1641                    nsecs_to_usecs(data->saved_latency),
1642                    entries,
1643                    total,
1644                    tr->cpu,
1645 #if defined(CONFIG_PREEMPT_NONE)
1646                    "server",
1647 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1648                    "desktop",
1649 #elif defined(CONFIG_PREEMPT)
1650                    "preempt",
1651 #else
1652                    "unknown",
1653 #endif
1654                    /* These are reserved for later use */
1655                    0, 0, 0, 0);
1656 #ifdef CONFIG_SMP
1657         seq_printf(m, " #P:%d)\n", num_online_cpus());
1658 #else
1659         seq_puts(m, ")\n");
1660 #endif
1661         seq_puts(m, "    -----------------\n");
1662         seq_printf(m, "    | task: %.16s-%d "
1663                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1664                    data->comm, data->pid, data->uid, data->nice,
1665                    data->policy, data->rt_priority);
1666         seq_puts(m, "    -----------------\n");
1667
1668         if (data->critical_start) {
1669                 seq_puts(m, " => started at: ");
1670                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1671                 trace_print_seq(m, &iter->seq);
1672                 seq_puts(m, "\n => ended at:   ");
1673                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1674                 trace_print_seq(m, &iter->seq);
1675                 seq_puts(m, "\n");
1676         }
1677
1678         seq_puts(m, "\n");
1679 }
1680
1681 static void
1682 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1683 {
1684         int hardirq, softirq;
1685         char *comm;
1686
1687         comm = trace_find_cmdline(entry->pid);
1688
1689         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1690         trace_seq_printf(s, "%3d", cpu);
1691         trace_seq_printf(s, "%c%c",
1692                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1693                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1694                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1695
1696         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1697         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1698         if (hardirq && softirq) {
1699                 trace_seq_putc(s, 'H');
1700         } else {
1701                 if (hardirq) {
1702                         trace_seq_putc(s, 'h');
1703                 } else {
1704                         if (softirq)
1705                                 trace_seq_putc(s, 's');
1706                         else
1707                                 trace_seq_putc(s, '.');
1708                 }
1709         }
1710
1711         if (entry->preempt_count)
1712                 trace_seq_printf(s, "%x", entry->preempt_count);
1713         else
1714                 trace_seq_puts(s, ".");
1715 }
1716
1717 unsigned long preempt_mark_thresh = 100;
1718
1719 static void
1720 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1721                     unsigned long rel_usecs)
1722 {
1723         trace_seq_printf(s, " %4lldus", abs_usecs);
1724         if (rel_usecs > preempt_mark_thresh)
1725                 trace_seq_puts(s, "!: ");
1726         else if (rel_usecs > 1)
1727                 trace_seq_puts(s, "+: ");
1728         else
1729                 trace_seq_puts(s, " : ");
1730 }
1731
1732 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1733
1734 /*
1735  * The message is supposed to contain an ending newline.
1736  * If the printing stops prematurely, try to add a newline of our own.
1737  */
1738 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1739 {
1740         struct trace_entry *ent;
1741         struct trace_field_cont *cont;
1742         bool ok = true;
1743
1744         ent = peek_next_entry(iter, iter->cpu, NULL);
1745         if (!ent || ent->type != TRACE_CONT) {
1746                 trace_seq_putc(s, '\n');
1747                 return;
1748         }
1749
1750         do {
1751                 cont = (struct trace_field_cont *)ent;
1752                 if (ok)
1753                         ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1754
1755                 ftrace_disable_cpu();
1756
1757                 if (iter->buffer_iter[iter->cpu])
1758                         ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1759                 else
1760                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1761
1762                 ftrace_enable_cpu();
1763
1764                 ent = peek_next_entry(iter, iter->cpu, NULL);
1765         } while (ent && ent->type == TRACE_CONT);
1766
1767         if (!ok)
1768                 trace_seq_putc(s, '\n');
1769 }
1770
1771 static void test_cpu_buff_start(struct trace_iterator *iter)
1772 {
1773         struct trace_seq *s = &iter->seq;
1774
1775         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1776                 return;
1777
1778         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1779                 return;
1780
1781         if (cpu_isset(iter->cpu, iter->started))
1782                 return;
1783
1784         cpu_set(iter->cpu, iter->started);
1785         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1786 }
1787
1788 static enum print_line_t
1789 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1790 {
1791         struct trace_seq *s = &iter->seq;
1792         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1793         struct trace_entry *next_entry;
1794         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1795         struct trace_entry *entry = iter->ent;
1796         unsigned long abs_usecs;
1797         unsigned long rel_usecs;
1798         u64 next_ts;
1799         char *comm;
1800         int S, T;
1801         int i;
1802         unsigned state;
1803
1804         if (entry->type == TRACE_CONT)
1805                 return TRACE_TYPE_HANDLED;
1806
1807         test_cpu_buff_start(iter);
1808
1809         next_entry = find_next_entry(iter, NULL, &next_ts);
1810         if (!next_entry)
1811                 next_ts = iter->ts;
1812         rel_usecs = ns2usecs(next_ts - iter->ts);
1813         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1814
1815         if (verbose) {
1816                 comm = trace_find_cmdline(entry->pid);
1817                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1818                                  " %ld.%03ldms (+%ld.%03ldms): ",
1819                                  comm,
1820                                  entry->pid, cpu, entry->flags,
1821                                  entry->preempt_count, trace_idx,
1822                                  ns2usecs(iter->ts),
1823                                  abs_usecs/1000,
1824                                  abs_usecs % 1000, rel_usecs/1000,
1825                                  rel_usecs % 1000);
1826         } else {
1827                 lat_print_generic(s, entry, cpu);
1828                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1829         }
1830         switch (entry->type) {
1831         case TRACE_FN: {
1832                 struct ftrace_entry *field;
1833
1834                 trace_assign_type(field, entry);
1835
1836                 seq_print_ip_sym(s, field->ip, sym_flags);
1837                 trace_seq_puts(s, " (");
1838                 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1839                 trace_seq_puts(s, ")\n");
1840                 break;
1841         }
1842         case TRACE_CTX:
1843         case TRACE_WAKE: {
1844                 struct ctx_switch_entry *field;
1845
1846                 trace_assign_type(field, entry);
1847
1848                 T = field->next_state < sizeof(state_to_char) ?
1849                         state_to_char[field->next_state] : 'X';
1850
1851                 state = field->prev_state ?
1852                         __ffs(field->prev_state) + 1 : 0;
1853                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1854                 comm = trace_find_cmdline(field->next_pid);
1855                 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1856                                  field->prev_pid,
1857                                  field->prev_prio,
1858                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1859                                  field->next_cpu,
1860                                  field->next_pid,
1861                                  field->next_prio,
1862                                  T, comm);
1863                 break;
1864         }
1865         case TRACE_SPECIAL: {
1866                 struct special_entry *field;
1867
1868                 trace_assign_type(field, entry);
1869
1870                 trace_seq_printf(s, "# %ld %ld %ld\n",
1871                                  field->arg1,
1872                                  field->arg2,
1873                                  field->arg3);
1874                 break;
1875         }
1876         case TRACE_STACK: {
1877                 struct stack_entry *field;
1878
1879                 trace_assign_type(field, entry);
1880
1881                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1882                         if (i)
1883                                 trace_seq_puts(s, " <= ");
1884                         seq_print_ip_sym(s, field->caller[i], sym_flags);
1885                 }
1886                 trace_seq_puts(s, "\n");
1887                 break;
1888         }
1889         case TRACE_PRINT: {
1890                 struct print_entry *field;
1891
1892                 trace_assign_type(field, entry);
1893
1894                 seq_print_ip_sym(s, field->ip, sym_flags);
1895                 trace_seq_printf(s, ": %s", field->buf);
1896                 if (entry->flags & TRACE_FLAG_CONT)
1897                         trace_seq_print_cont(s, iter);
1898                 break;
1899         }
1900         case TRACE_BRANCH: {
1901                 struct trace_branch *field;
1902
1903                 trace_assign_type(field, entry);
1904
1905                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1906                                  field->correct ? "  ok  " : " MISS ",
1907                                  field->func,
1908                                  field->file,
1909                                  field->line);
1910                 break;
1911         }
1912         case TRACE_USER_STACK: {
1913                 struct userstack_entry *field;
1914
1915                 trace_assign_type(field, entry);
1916
1917                 seq_print_userip_objs(field, s, sym_flags);
1918                 trace_seq_putc(s, '\n');
1919                 break;
1920         }
1921         default:
1922                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1923         }
1924         return TRACE_TYPE_HANDLED;
1925 }
1926
1927 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1928 {
1929         struct trace_seq *s = &iter->seq;
1930         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1931         struct trace_entry *entry;
1932         unsigned long usec_rem;
1933         unsigned long long t;
1934         unsigned long secs;
1935         char *comm;
1936         int ret;
1937         int S, T;
1938         int i;
1939
1940         entry = iter->ent;
1941
1942         if (entry->type == TRACE_CONT)
1943                 return TRACE_TYPE_HANDLED;
1944
1945         test_cpu_buff_start(iter);
1946
1947         comm = trace_find_cmdline(iter->ent->pid);
1948
1949         t = ns2usecs(iter->ts);
1950         usec_rem = do_div(t, 1000000ULL);
1951         secs = (unsigned long)t;
1952
1953         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1954         if (!ret)
1955                 return TRACE_TYPE_PARTIAL_LINE;
1956         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1957         if (!ret)
1958                 return TRACE_TYPE_PARTIAL_LINE;
1959         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1960         if (!ret)
1961                 return TRACE_TYPE_PARTIAL_LINE;
1962
1963         switch (entry->type) {
1964         case TRACE_FN: {
1965                 struct ftrace_entry *field;
1966
1967                 trace_assign_type(field, entry);
1968
1969                 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1970                 if (!ret)
1971                         return TRACE_TYPE_PARTIAL_LINE;
1972                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1973                                                 field->parent_ip) {
1974                         ret = trace_seq_printf(s, " <-");
1975                         if (!ret)
1976                                 return TRACE_TYPE_PARTIAL_LINE;
1977                         ret = seq_print_ip_sym(s,
1978                                                field->parent_ip,
1979                                                sym_flags);
1980                         if (!ret)
1981                                 return TRACE_TYPE_PARTIAL_LINE;
1982                 }
1983                 ret = trace_seq_printf(s, "\n");
1984                 if (!ret)
1985                         return TRACE_TYPE_PARTIAL_LINE;
1986                 break;
1987         }
1988         case TRACE_CTX:
1989         case TRACE_WAKE: {
1990                 struct ctx_switch_entry *field;
1991
1992                 trace_assign_type(field, entry);
1993
1994                 S = field->prev_state < sizeof(state_to_char) ?
1995                         state_to_char[field->prev_state] : 'X';
1996                 T = field->next_state < sizeof(state_to_char) ?
1997                         state_to_char[field->next_state] : 'X';
1998                 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1999                                        field->prev_pid,
2000                                        field->prev_prio,
2001                                        S,
2002                                        entry->type == TRACE_CTX ? "==>" : "  +",
2003                                        field->next_cpu,
2004                                        field->next_pid,
2005                                        field->next_prio,
2006                                        T);
2007                 if (!ret)
2008                         return TRACE_TYPE_PARTIAL_LINE;
2009                 break;
2010         }
2011         case TRACE_SPECIAL: {
2012                 struct special_entry *field;
2013
2014                 trace_assign_type(field, entry);
2015
2016                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2017                                  field->arg1,
2018                                  field->arg2,
2019                                  field->arg3);
2020                 if (!ret)
2021                         return TRACE_TYPE_PARTIAL_LINE;
2022                 break;
2023         }
2024         case TRACE_STACK: {
2025                 struct stack_entry *field;
2026
2027                 trace_assign_type(field, entry);
2028
2029                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2030                         if (i) {
2031                                 ret = trace_seq_puts(s, " <= ");
2032                                 if (!ret)
2033                                         return TRACE_TYPE_PARTIAL_LINE;
2034                         }
2035                         ret = seq_print_ip_sym(s, field->caller[i],
2036                                                sym_flags);
2037                         if (!ret)
2038                                 return TRACE_TYPE_PARTIAL_LINE;
2039                 }
2040                 ret = trace_seq_puts(s, "\n");
2041                 if (!ret)
2042                         return TRACE_TYPE_PARTIAL_LINE;
2043                 break;
2044         }
2045         case TRACE_PRINT: {
2046                 struct print_entry *field;
2047
2048                 trace_assign_type(field, entry);
2049
2050                 seq_print_ip_sym(s, field->ip, sym_flags);
2051                 trace_seq_printf(s, ": %s", field->buf);
2052                 if (entry->flags & TRACE_FLAG_CONT)
2053                         trace_seq_print_cont(s, iter);
2054                 break;
2055         }
2056         case TRACE_GRAPH_RET: {
2057                 return print_graph_function(iter);
2058         }
2059         case TRACE_GRAPH_ENT: {
2060                 return print_graph_function(iter);
2061         }
2062         case TRACE_BRANCH: {
2063                 struct trace_branch *field;
2064
2065                 trace_assign_type(field, entry);
2066
2067                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2068                                  field->correct ? "  ok  " : " MISS ",
2069                                  field->func,
2070                                  field->file,
2071                                  field->line);
2072                 break;
2073         }
2074         case TRACE_USER_STACK: {
2075                 struct userstack_entry *field;
2076
2077                 trace_assign_type(field, entry);
2078
2079                 ret = seq_print_userip_objs(field, s, sym_flags);
2080                 if (!ret)
2081                         return TRACE_TYPE_PARTIAL_LINE;
2082                 ret = trace_seq_putc(s, '\n');
2083                 if (!ret)
2084                         return TRACE_TYPE_PARTIAL_LINE;
2085                 break;
2086         }
2087         }
2088         return TRACE_TYPE_HANDLED;
2089 }
2090
2091 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2092 {
2093         struct trace_seq *s = &iter->seq;
2094         struct trace_entry *entry;
2095         int ret;
2096         int S, T;
2097
2098         entry = iter->ent;
2099
2100         if (entry->type == TRACE_CONT)
2101                 return TRACE_TYPE_HANDLED;
2102
2103         ret = trace_seq_printf(s, "%d %d %llu ",
2104                 entry->pid, iter->cpu, iter->ts);
2105         if (!ret)
2106                 return TRACE_TYPE_PARTIAL_LINE;
2107
2108         switch (entry->type) {
2109         case TRACE_FN: {
2110                 struct ftrace_entry *field;
2111
2112                 trace_assign_type(field, entry);
2113
2114                 ret = trace_seq_printf(s, "%x %x\n",
2115                                         field->ip,
2116                                         field->parent_ip);
2117                 if (!ret)
2118                         return TRACE_TYPE_PARTIAL_LINE;
2119                 break;
2120         }
2121         case TRACE_CTX:
2122         case TRACE_WAKE: {
2123                 struct ctx_switch_entry *field;
2124
2125                 trace_assign_type(field, entry);
2126
2127                 S = field->prev_state < sizeof(state_to_char) ?
2128                         state_to_char[field->prev_state] : 'X';
2129                 T = field->next_state < sizeof(state_to_char) ?
2130                         state_to_char[field->next_state] : 'X';
2131                 if (entry->type == TRACE_WAKE)
2132                         S = '+';
2133                 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2134                                        field->prev_pid,
2135                                        field->prev_prio,
2136                                        S,
2137                                        field->next_cpu,
2138                                        field->next_pid,
2139                                        field->next_prio,
2140                                        T);
2141                 if (!ret)
2142                         return TRACE_TYPE_PARTIAL_LINE;
2143                 break;
2144         }
2145         case TRACE_SPECIAL:
2146         case TRACE_USER_STACK:
2147         case TRACE_STACK: {
2148                 struct special_entry *field;
2149
2150                 trace_assign_type(field, entry);
2151
2152                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2153                                  field->arg1,
2154                                  field->arg2,
2155                                  field->arg3);
2156                 if (!ret)
2157                         return TRACE_TYPE_PARTIAL_LINE;
2158                 break;
2159         }
2160         case TRACE_PRINT: {
2161                 struct print_entry *field;
2162
2163                 trace_assign_type(field, entry);
2164
2165                 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2166                 if (entry->flags & TRACE_FLAG_CONT)
2167                         trace_seq_print_cont(s, iter);
2168                 break;
2169         }
2170         }
2171         return TRACE_TYPE_HANDLED;
2172 }
2173
2174 #define SEQ_PUT_FIELD_RET(s, x)                         \
2175 do {                                                    \
2176         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
2177                 return 0;                               \
2178 } while (0)
2179
2180 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
2181 do {                                                    \
2182         BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES);     \
2183         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
2184                 return 0;                               \
2185 } while (0)
2186
2187 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2188 {
2189         struct trace_seq *s = &iter->seq;
2190         unsigned char newline = '\n';
2191         struct trace_entry *entry;
2192         int S, T;
2193
2194         entry = iter->ent;
2195
2196         if (entry->type == TRACE_CONT)
2197                 return TRACE_TYPE_HANDLED;
2198
2199         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2200         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2201         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2202
2203         switch (entry->type) {
2204         case TRACE_FN: {
2205                 struct ftrace_entry *field;
2206
2207                 trace_assign_type(field, entry);
2208
2209                 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2210                 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2211                 break;
2212         }
2213         case TRACE_CTX:
2214         case TRACE_WAKE: {
2215                 struct ctx_switch_entry *field;
2216
2217                 trace_assign_type(field, entry);
2218
2219                 S = field->prev_state < sizeof(state_to_char) ?
2220                         state_to_char[field->prev_state] : 'X';
2221                 T = field->next_state < sizeof(state_to_char) ?
2222                         state_to_char[field->next_state] : 'X';
2223                 if (entry->type == TRACE_WAKE)
2224                         S = '+';
2225                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2226                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2227                 SEQ_PUT_HEX_FIELD_RET(s, S);
2228                 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2229                 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2230                 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2231                 SEQ_PUT_HEX_FIELD_RET(s, T);
2232                 break;
2233         }
2234         case TRACE_SPECIAL:
2235         case TRACE_USER_STACK:
2236         case TRACE_STACK: {
2237                 struct special_entry *field;
2238
2239                 trace_assign_type(field, entry);
2240
2241                 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2242                 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2243                 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2244                 break;
2245         }
2246         }
2247         SEQ_PUT_FIELD_RET(s, newline);
2248
2249         return TRACE_TYPE_HANDLED;
2250 }
2251
2252 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2253 {
2254         struct trace_seq *s = &iter->seq;
2255         struct trace_entry *entry;
2256
2257         entry = iter->ent;
2258
2259         if (entry->type == TRACE_CONT)
2260                 return TRACE_TYPE_HANDLED;
2261
2262         SEQ_PUT_FIELD_RET(s, entry->pid);
2263         SEQ_PUT_FIELD_RET(s, entry->cpu);
2264         SEQ_PUT_FIELD_RET(s, iter->ts);
2265
2266         switch (entry->type) {
2267         case TRACE_FN: {
2268                 struct ftrace_entry *field;
2269
2270                 trace_assign_type(field, entry);
2271
2272                 SEQ_PUT_FIELD_RET(s, field->ip);
2273                 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2274                 break;
2275         }
2276         case TRACE_CTX: {
2277                 struct ctx_switch_entry *field;
2278
2279                 trace_assign_type(field, entry);
2280
2281                 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2282                 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2283                 SEQ_PUT_FIELD_RET(s, field->prev_state);
2284                 SEQ_PUT_FIELD_RET(s, field->next_pid);
2285                 SEQ_PUT_FIELD_RET(s, field->next_prio);
2286                 SEQ_PUT_FIELD_RET(s, field->next_state);
2287                 break;
2288         }
2289         case TRACE_SPECIAL:
2290         case TRACE_USER_STACK:
2291         case TRACE_STACK: {
2292                 struct special_entry *field;
2293
2294                 trace_assign_type(field, entry);
2295
2296                 SEQ_PUT_FIELD_RET(s, field->arg1);
2297                 SEQ_PUT_FIELD_RET(s, field->arg2);
2298                 SEQ_PUT_FIELD_RET(s, field->arg3);
2299                 break;
2300         }
2301         }
2302         return 1;
2303 }
2304
2305 static int trace_empty(struct trace_iterator *iter)
2306 {
2307         int cpu;
2308
2309         for_each_tracing_cpu(cpu) {
2310                 if (iter->buffer_iter[cpu]) {
2311                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2312                                 return 0;
2313                 } else {
2314                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2315                                 return 0;
2316                 }
2317         }
2318
2319         return 1;
2320 }
2321
2322 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2323 {
2324         enum print_line_t ret;
2325
2326         if (iter->trace && iter->trace->print_line) {
2327                 ret = iter->trace->print_line(iter);
2328                 if (ret != TRACE_TYPE_UNHANDLED)
2329                         return ret;
2330         }
2331
2332         if (trace_flags & TRACE_ITER_BIN)
2333                 return print_bin_fmt(iter);
2334
2335         if (trace_flags & TRACE_ITER_HEX)
2336                 return print_hex_fmt(iter);
2337
2338         if (trace_flags & TRACE_ITER_RAW)
2339                 return print_raw_fmt(iter);
2340
2341         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2342                 return print_lat_fmt(iter, iter->idx, iter->cpu);
2343
2344         return print_trace_fmt(iter);
2345 }
2346
2347 static int s_show(struct seq_file *m, void *v)
2348 {
2349         struct trace_iterator *iter = v;
2350
2351         if (iter->ent == NULL) {
2352                 if (iter->tr) {
2353                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2354                         seq_puts(m, "#\n");
2355                 }
2356                 if (iter->trace && iter->trace->print_header)
2357                         iter->trace->print_header(m);
2358                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2359                         /* print nothing if the buffers are empty */
2360                         if (trace_empty(iter))
2361                                 return 0;
2362                         print_trace_header(m, iter);
2363                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2364                                 print_lat_help_header(m);
2365                 } else {
2366                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2367                                 print_func_help_header(m);
2368                 }
2369         } else {
2370                 print_trace_line(iter);
2371                 trace_print_seq(m, &iter->seq);
2372         }
2373
2374         return 0;
2375 }
2376
2377 static struct seq_operations tracer_seq_ops = {
2378         .start          = s_start,
2379         .next           = s_next,
2380         .stop           = s_stop,
2381         .show           = s_show,
2382 };
2383
2384 static struct trace_iterator *
2385 __tracing_open(struct inode *inode, struct file *file, int *ret)
2386 {
2387         struct trace_iterator *iter;
2388         struct seq_file *m;
2389         int cpu;
2390
2391         if (tracing_disabled) {
2392                 *ret = -ENODEV;
2393                 return NULL;
2394         }
2395
2396         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2397         if (!iter) {
2398                 *ret = -ENOMEM;
2399                 goto out;
2400         }
2401
2402         mutex_lock(&trace_types_lock);
2403         if (current_trace && current_trace->print_max)
2404                 iter->tr = &max_tr;
2405         else
2406                 iter->tr = inode->i_private;
2407         iter->trace = current_trace;
2408         iter->pos = -1;
2409
2410         /* Notify the tracer early; before we stop tracing. */
2411         if (iter->trace && iter->trace->open)
2412                         iter->trace->open(iter);
2413
2414         /* Annotate start of buffers if we had overruns */
2415         if (ring_buffer_overruns(iter->tr->buffer))
2416                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2417
2418
2419         for_each_tracing_cpu(cpu) {
2420
2421                 iter->buffer_iter[cpu] =
2422                         ring_buffer_read_start(iter->tr->buffer, cpu);
2423
2424                 if (!iter->buffer_iter[cpu])
2425                         goto fail_buffer;
2426         }
2427
2428         /* TODO stop tracer */
2429         *ret = seq_open(file, &tracer_seq_ops);
2430         if (*ret)
2431                 goto fail_buffer;
2432
2433         m = file->private_data;
2434         m->private = iter;
2435
2436         /* stop the trace while dumping */
2437         tracing_stop();
2438
2439         mutex_unlock(&trace_types_lock);
2440
2441  out:
2442         return iter;
2443
2444  fail_buffer:
2445         for_each_tracing_cpu(cpu) {
2446                 if (iter->buffer_iter[cpu])
2447                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2448         }
2449         mutex_unlock(&trace_types_lock);
2450         kfree(iter);
2451
2452         return ERR_PTR(-ENOMEM);
2453 }
2454
2455 int tracing_open_generic(struct inode *inode, struct file *filp)
2456 {
2457         if (tracing_disabled)
2458                 return -ENODEV;
2459
2460         filp->private_data = inode->i_private;
2461         return 0;
2462 }
2463
2464 int tracing_release(struct inode *inode, struct file *file)
2465 {
2466         struct seq_file *m = (struct seq_file *)file->private_data;
2467         struct trace_iterator *iter = m->private;
2468         int cpu;
2469
2470         mutex_lock(&trace_types_lock);
2471         for_each_tracing_cpu(cpu) {
2472                 if (iter->buffer_iter[cpu])
2473                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2474         }
2475
2476         if (iter->trace && iter->trace->close)
2477                 iter->trace->close(iter);
2478
2479         /* reenable tracing if it was previously enabled */
2480         tracing_start();
2481         mutex_unlock(&trace_types_lock);
2482
2483         seq_release(inode, file);
2484         kfree(iter);
2485         return 0;
2486 }
2487
2488 static int tracing_open(struct inode *inode, struct file *file)
2489 {
2490         int ret;
2491
2492         __tracing_open(inode, file, &ret);
2493
2494         return ret;
2495 }
2496
2497 static int tracing_lt_open(struct inode *inode, struct file *file)
2498 {
2499         struct trace_iterator *iter;
2500         int ret;
2501
2502         iter = __tracing_open(inode, file, &ret);
2503
2504         if (!ret)
2505                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2506
2507         return ret;
2508 }
2509
2510
2511 static void *
2512 t_next(struct seq_file *m, void *v, loff_t *pos)
2513 {
2514         struct tracer *t = m->private;
2515
2516         (*pos)++;
2517
2518         if (t)
2519                 t = t->next;
2520
2521         m->private = t;
2522
2523         return t;
2524 }
2525
2526 static void *t_start(struct seq_file *m, loff_t *pos)
2527 {
2528         struct tracer *t = m->private;
2529         loff_t l = 0;
2530
2531         mutex_lock(&trace_types_lock);
2532         for (; t && l < *pos; t = t_next(m, t, &l))
2533                 ;
2534
2535         return t;
2536 }
2537
2538 static void t_stop(struct seq_file *m, void *p)
2539 {
2540         mutex_unlock(&trace_types_lock);
2541 }
2542
2543 static int t_show(struct seq_file *m, void *v)
2544 {
2545         struct tracer *t = v;
2546
2547         if (!t)
2548                 return 0;
2549
2550         seq_printf(m, "%s", t->name);
2551         if (t->next)
2552                 seq_putc(m, ' ');
2553         else
2554                 seq_putc(m, '\n');
2555
2556         return 0;
2557 }
2558
2559 static struct seq_operations show_traces_seq_ops = {
2560         .start          = t_start,
2561         .next           = t_next,
2562         .stop           = t_stop,
2563         .show           = t_show,
2564 };
2565
2566 static int show_traces_open(struct inode *inode, struct file *file)
2567 {
2568         int ret;
2569
2570         if (tracing_disabled)
2571                 return -ENODEV;
2572
2573         ret = seq_open(file, &show_traces_seq_ops);
2574         if (!ret) {
2575                 struct seq_file *m = file->private_data;
2576                 m->private = trace_types;
2577         }
2578
2579         return ret;
2580 }
2581
2582 static struct file_operations tracing_fops = {
2583         .open           = tracing_open,
2584         .read           = seq_read,
2585         .llseek         = seq_lseek,
2586         .release        = tracing_release,
2587 };
2588
2589 static struct file_operations tracing_lt_fops = {
2590         .open           = tracing_lt_open,
2591         .read           = seq_read,
2592         .llseek         = seq_lseek,
2593         .release        = tracing_release,
2594 };
2595
2596 static struct file_operations show_traces_fops = {
2597         .open           = show_traces_open,
2598         .read           = seq_read,
2599         .release        = seq_release,
2600 };
2601
2602 /*
2603  * Only trace on a CPU if the bitmask is set:
2604  */
2605 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2606
2607 /*
2608  * When tracing/tracing_cpu_mask is modified then this holds
2609  * the new bitmask we are about to install:
2610  */
2611 static cpumask_t tracing_cpumask_new;
2612
2613 /*
2614  * The tracer itself will not take this lock, but still we want
2615  * to provide a consistent cpumask to user-space:
2616  */
2617 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2618
2619 /*
2620  * Temporary storage for the character representation of the
2621  * CPU bitmask (and one more byte for the newline):
2622  */
2623 static char mask_str[NR_CPUS + 1];
2624
2625 static ssize_t
2626 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2627                      size_t count, loff_t *ppos)
2628 {
2629         int len;
2630
2631         mutex_lock(&tracing_cpumask_update_lock);
2632
2633         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2634         if (count - len < 2) {
2635                 count = -EINVAL;
2636                 goto out_err;
2637         }
2638         len += sprintf(mask_str + len, "\n");
2639         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2640
2641 out_err:
2642         mutex_unlock(&tracing_cpumask_update_lock);
2643
2644         return count;
2645 }
2646
2647 static ssize_t
2648 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2649                       size_t count, loff_t *ppos)
2650 {
2651         int err, cpu;
2652
2653         mutex_lock(&tracing_cpumask_update_lock);
2654         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2655         if (err)
2656                 goto err_unlock;
2657
2658         local_irq_disable();
2659         __raw_spin_lock(&ftrace_max_lock);
2660         for_each_tracing_cpu(cpu) {
2661                 /*
2662                  * Increase/decrease the disabled counter if we are
2663                  * about to flip a bit in the cpumask:
2664                  */
2665                 if (cpu_isset(cpu, tracing_cpumask) &&
2666                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2667                         atomic_inc(&global_trace.data[cpu]->disabled);
2668                 }
2669                 if (!cpu_isset(cpu, tracing_cpumask) &&
2670                                 cpu_isset(cpu, tracing_cpumask_new)) {
2671                         atomic_dec(&global_trace.data[cpu]->disabled);
2672                 }
2673         }
2674         __raw_spin_unlock(&ftrace_max_lock);
2675         local_irq_enable();
2676
2677         tracing_cpumask = tracing_cpumask_new;
2678
2679         mutex_unlock(&tracing_cpumask_update_lock);
2680
2681         return count;
2682
2683 err_unlock:
2684         mutex_unlock(&tracing_cpumask_update_lock);
2685
2686         return err;
2687 }
2688
2689 static struct file_operations tracing_cpumask_fops = {
2690         .open           = tracing_open_generic,
2691         .read           = tracing_cpumask_read,
2692         .write          = tracing_cpumask_write,
2693 };
2694
2695 static ssize_t
2696 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2697                        size_t cnt, loff_t *ppos)
2698 {
2699         int i;
2700         char *buf;
2701         int r = 0;
2702         int len = 0;
2703         u32 tracer_flags = current_trace->flags->val;
2704         struct tracer_opt *trace_opts = current_trace->flags->opts;
2705
2706
2707         /* calulate max size */
2708         for (i = 0; trace_options[i]; i++) {
2709                 len += strlen(trace_options[i]);
2710                 len += 3; /* "no" and space */
2711         }
2712
2713         /*
2714          * Increase the size with names of options specific
2715          * of the current tracer.
2716          */
2717         for (i = 0; trace_opts[i].name; i++) {
2718                 len += strlen(trace_opts[i].name);
2719                 len += 3; /* "no" and space */
2720         }
2721
2722         /* +2 for \n and \0 */
2723         buf = kmalloc(len + 2, GFP_KERNEL);
2724         if (!buf)
2725                 return -ENOMEM;
2726
2727         for (i = 0; trace_options[i]; i++) {
2728                 if (trace_flags & (1 << i))
2729                         r += sprintf(buf + r, "%s ", trace_options[i]);
2730                 else
2731                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2732         }
2733
2734         for (i = 0; trace_opts[i].name; i++) {
2735                 if (tracer_flags & trace_opts[i].bit)
2736                         r += sprintf(buf + r, "%s ",
2737                                 trace_opts[i].name);
2738                 else
2739                         r += sprintf(buf + r, "no%s ",
2740                                 trace_opts[i].name);
2741         }
2742
2743         r += sprintf(buf + r, "\n");
2744         WARN_ON(r >= len + 2);
2745
2746         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2747
2748         kfree(buf);
2749
2750         return r;
2751 }
2752
2753 /* Try to assign a tracer specific option */
2754 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2755 {
2756         struct tracer_flags *trace_flags = trace->flags;
2757         struct tracer_opt *opts = NULL;
2758         int ret = 0, i = 0;
2759         int len;
2760
2761         for (i = 0; trace_flags->opts[i].name; i++) {
2762                 opts = &trace_flags->opts[i];
2763                 len = strlen(opts->name);
2764
2765                 if (strncmp(cmp, opts->name, len) == 0) {
2766                         ret = trace->set_flag(trace_flags->val,
2767                                 opts->bit, !neg);
2768                         break;
2769                 }
2770         }
2771         /* Not found */
2772         if (!trace_flags->opts[i].name)
2773                 return -EINVAL;
2774
2775         /* Refused to handle */
2776         if (ret)
2777                 return ret;
2778
2779         if (neg)
2780                 trace_flags->val &= ~opts->bit;
2781         else
2782                 trace_flags->val |= opts->bit;
2783
2784         return 0;
2785 }
2786
2787 static ssize_t
2788 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2789                         size_t cnt, loff_t *ppos)
2790 {
2791         char buf[64];
2792         char *cmp = buf;
2793         int neg = 0;
2794         int ret;
2795         int i;
2796
2797         if (cnt >= sizeof(buf))
2798                 return -EINVAL;
2799
2800         if (copy_from_user(&buf, ubuf, cnt))
2801                 return -EFAULT;
2802
2803         buf[cnt] = 0;
2804
2805         if (strncmp(buf, "no", 2) == 0) {
2806                 neg = 1;
2807                 cmp += 2;
2808         }
2809
2810         for (i = 0; trace_options[i]; i++) {
2811                 int len = strlen(trace_options[i]);
2812
2813                 if (strncmp(cmp, trace_options[i], len) == 0) {
2814                         if (neg)
2815                                 trace_flags &= ~(1 << i);
2816                         else
2817                                 trace_flags |= (1 << i);
2818                         break;
2819                 }
2820         }
2821
2822         /* If no option could be set, test the specific tracer options */
2823         if (!trace_options[i]) {
2824                 ret = set_tracer_option(current_trace, cmp, neg);
2825                 if (ret)
2826                         return ret;
2827         }
2828
2829         filp->f_pos += cnt;
2830
2831         return cnt;
2832 }
2833
2834 static struct file_operations tracing_iter_fops = {
2835         .open           = tracing_open_generic,
2836         .read           = tracing_trace_options_read,
2837         .write          = tracing_trace_options_write,
2838 };
2839
2840 static const char readme_msg[] =
2841         "tracing mini-HOWTO:\n\n"
2842         "# mkdir /debug\n"
2843         "# mount -t debugfs nodev /debug\n\n"
2844         "# cat /debug/tracing/available_tracers\n"
2845         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2846         "# cat /debug/tracing/current_tracer\n"
2847         "none\n"
2848         "# echo sched_switch > /debug/tracing/current_tracer\n"
2849         "# cat /debug/tracing/current_tracer\n"
2850         "sched_switch\n"
2851         "# cat /debug/tracing/trace_options\n"
2852         "noprint-parent nosym-offset nosym-addr noverbose\n"
2853         "# echo print-parent > /debug/tracing/trace_options\n"
2854         "# echo 1 > /debug/tracing/tracing_enabled\n"
2855         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2856         "echo 0 > /debug/tracing/tracing_enabled\n"
2857 ;
2858
2859 static ssize_t
2860 tracing_readme_read(struct file *filp, char __user *ubuf,
2861                        size_t cnt, loff_t *ppos)
2862 {
2863         return simple_read_from_buffer(ubuf, cnt, ppos,
2864                                         readme_msg, strlen(readme_msg));
2865 }
2866
2867 static struct file_operations tracing_readme_fops = {
2868         .open           = tracing_open_generic,
2869         .read           = tracing_readme_read,
2870 };
2871
2872 static ssize_t
2873 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2874                   size_t cnt, loff_t *ppos)
2875 {
2876         char buf[64];
2877         int r;
2878
2879         r = sprintf(buf, "%u\n", tracer_enabled);
2880         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2881 }
2882
2883 static ssize_t
2884 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2885                    size_t cnt, loff_t *ppos)
2886 {
2887         struct trace_array *tr = filp->private_data;
2888         char buf[64];
2889         long val;
2890         int ret;
2891
2892         if (cnt >= sizeof(buf))
2893                 return -EINVAL;
2894
2895         if (copy_from_user(&buf, ubuf, cnt))
2896                 return -EFAULT;
2897
2898         buf[cnt] = 0;
2899
2900         ret = strict_strtoul(buf, 10, &val);
2901         if (ret < 0)
2902                 return ret;
2903
2904         val = !!val;
2905
2906         mutex_lock(&trace_types_lock);
2907         if (tracer_enabled ^ val) {
2908                 if (val) {
2909                         tracer_enabled = 1;
2910                         if (current_trace->start)
2911                                 current_trace->start(tr);
2912                         tracing_start();
2913                 } else {
2914                         tracer_enabled = 0;
2915                         tracing_stop();
2916                         if (current_trace->stop)
2917                                 current_trace->stop(tr);
2918                 }
2919         }
2920         mutex_unlock(&trace_types_lock);
2921
2922         filp->f_pos += cnt;
2923
2924         return cnt;
2925 }
2926
2927 static ssize_t
2928 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2929                        size_t cnt, loff_t *ppos)
2930 {
2931         char buf[max_tracer_type_len+2];
2932         int r;
2933
2934         mutex_lock(&trace_types_lock);
2935         if (current_trace)
2936                 r = sprintf(buf, "%s\n", current_trace->name);
2937         else
2938                 r = sprintf(buf, "\n");
2939         mutex_unlock(&trace_types_lock);
2940
2941         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2942 }
2943
2944 static int tracing_set_tracer(char *buf)
2945 {
2946         struct trace_array *tr = &global_trace;
2947         struct tracer *t;
2948         int ret = 0;
2949
2950         mutex_lock(&trace_types_lock);
2951         for (t = trace_types; t; t = t->next) {
2952                 if (strcmp(t->name, buf) == 0)
2953                         break;
2954         }
2955         if (!t) {
2956                 ret = -EINVAL;
2957                 goto out;
2958         }
2959         if (t == current_trace)
2960                 goto out;
2961
2962         trace_branch_disable();
2963         if (current_trace && current_trace->reset)
2964                 current_trace->reset(tr);
2965
2966         current_trace = t;
2967         if (t->init) {
2968                 ret = t->init(tr);
2969                 if (ret)
2970                         goto out;
2971         }
2972
2973         trace_branch_enable(tr);
2974  out:
2975         mutex_unlock(&trace_types_lock);
2976
2977         return ret;
2978 }
2979
2980 static ssize_t
2981 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2982                         size_t cnt, loff_t *ppos)
2983 {
2984         char buf[max_tracer_type_len+1];
2985         int i;
2986         size_t ret;
2987         int err;
2988
2989         ret = cnt;
2990
2991         if (cnt > max_tracer_type_len)
2992                 cnt = max_tracer_type_len;
2993
2994         if (copy_from_user(&buf, ubuf, cnt))
2995                 return -EFAULT;
2996
2997         buf[cnt] = 0;
2998
2999         /* strip ending whitespace. */
3000         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3001                 buf[i] = 0;
3002
3003         err = tracing_set_tracer(buf);
3004         if (err)
3005                 return err;
3006
3007         filp->f_pos += ret;
3008
3009         return ret;
3010 }
3011
3012 static ssize_t
3013 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3014                      size_t cnt, loff_t *ppos)
3015 {
3016         unsigned long *ptr = filp->private_data;
3017         char buf[64];
3018         int r;
3019
3020         r = snprintf(buf, sizeof(buf), "%ld\n",
3021                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3022         if (r > sizeof(buf))
3023                 r = sizeof(buf);
3024         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3025 }
3026
3027 static ssize_t
3028 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3029                       size_t cnt, loff_t *ppos)
3030 {
3031         long *ptr = filp->private_data;
3032         char buf[64];
3033         long val;
3034         int ret;
3035
3036         if (cnt >= sizeof(buf))
3037                 return -EINVAL;
3038
3039         if (copy_from_user(&buf, ubuf, cnt))
3040                 return -EFAULT;
3041
3042         buf[cnt] = 0;
3043
3044         ret = strict_strtoul(buf, 10, &val);
3045         if (ret < 0)
3046                 return ret;
3047
3048         *ptr = val * 1000;
3049
3050         return cnt;
3051 }
3052
3053 static atomic_t tracing_reader;
3054
3055 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3056 {
3057         struct trace_iterator *iter;
3058
3059         if (tracing_disabled)
3060                 return -ENODEV;
3061
3062         /* We only allow for reader of the pipe */
3063         if (atomic_inc_return(&tracing_reader) != 1) {
3064                 atomic_dec(&tracing_reader);
3065                 return -EBUSY;
3066         }
3067
3068         /* create a buffer to store the information to pass to userspace */
3069         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3070         if (!iter)
3071                 return -ENOMEM;
3072
3073         mutex_lock(&trace_types_lock);
3074
3075         /* trace pipe does not show start of buffer */
3076         cpus_setall(iter->started);
3077
3078         iter->tr = &global_trace;
3079         iter->trace = current_trace;
3080         filp->private_data = iter;
3081
3082         if (iter->trace->pipe_open)
3083                 iter->trace->pipe_open(iter);
3084         mutex_unlock(&trace_types_lock);
3085
3086         return 0;
3087 }
3088
3089 static int tracing_release_pipe(struct inode *inode, struct file *file)
3090 {
3091         struct trace_iterator *iter = file->private_data;
3092
3093         kfree(iter);
3094         atomic_dec(&tracing_reader);
3095
3096         return 0;
3097 }
3098
3099 static unsigned int
3100 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3101 {
3102         struct trace_iterator *iter = filp->private_data;
3103
3104         if (trace_flags & TRACE_ITER_BLOCK) {
3105                 /*
3106                  * Always select as readable when in blocking mode
3107                  */
3108                 return POLLIN | POLLRDNORM;
3109         } else {
3110                 if (!trace_empty(iter))
3111                         return POLLIN | POLLRDNORM;
3112                 poll_wait(filp, &trace_wait, poll_table);
3113                 if (!trace_empty(iter))
3114                         return POLLIN | POLLRDNORM;
3115
3116                 return 0;
3117         }
3118 }
3119
3120 /*
3121  * Consumer reader.
3122  */
3123 static ssize_t
3124 tracing_read_pipe(struct file *filp, char __user *ubuf,
3125                   size_t cnt, loff_t *ppos)
3126 {
3127         struct trace_iterator *iter = filp->private_data;
3128         ssize_t sret;
3129
3130         /* return any leftover data */
3131         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3132         if (sret != -EBUSY)
3133                 return sret;
3134
3135         trace_seq_reset(&iter->seq);
3136
3137         mutex_lock(&trace_types_lock);
3138         if (iter->trace->read) {
3139                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3140                 if (sret)
3141                         goto out;
3142         }
3143
3144 waitagain:
3145         sret = 0;
3146         while (trace_empty(iter)) {
3147
3148                 if ((filp->f_flags & O_NONBLOCK)) {
3149                         sret = -EAGAIN;
3150                         goto out;
3151                 }
3152
3153                 /*
3154                  * This is a make-shift waitqueue. The reason we don't use
3155                  * an actual wait queue is because:
3156                  *  1) we only ever have one waiter
3157                  *  2) the tracing, traces all functions, we don't want
3158                  *     the overhead of calling wake_up and friends
3159                  *     (and tracing them too)
3160                  *     Anyway, this is really very primitive wakeup.
3161                  */
3162                 set_current_state(TASK_INTERRUPTIBLE);
3163                 iter->tr->waiter = current;
3164
3165                 mutex_unlock(&trace_types_lock);
3166
3167                 /* sleep for 100 msecs, and try again. */
3168                 schedule_timeout(HZ/10);
3169
3170                 mutex_lock(&trace_types_lock);
3171
3172                 iter->tr->waiter = NULL;
3173
3174                 if (signal_pending(current)) {
3175                         sret = -EINTR;
3176                         goto out;
3177                 }
3178
3179                 if (iter->trace != current_trace)
3180                         goto out;
3181
3182                 /*
3183                  * We block until we read something and tracing is disabled.
3184                  * We still block if tracing is disabled, but we have never
3185                  * read anything. This allows a user to cat this file, and
3186                  * then enable tracing. But after we have read something,
3187                  * we give an EOF when tracing is again disabled.
3188                  *
3189                  * iter->pos will be 0 if we haven't read anything.
3190                  */
3191                 if (!tracer_enabled && iter->pos)
3192                         break;
3193
3194                 continue;
3195         }
3196
3197         /* stop when tracing is finished */
3198         if (trace_empty(iter))
3199                 goto out;
3200
3201         if (cnt >= PAGE_SIZE)
3202                 cnt = PAGE_SIZE - 1;
3203
3204         /* reset all but tr, trace, and overruns */
3205         memset(&iter->seq, 0,
3206                sizeof(struct trace_iterator) -
3207                offsetof(struct trace_iterator, seq));
3208         iter->pos = -1;
3209
3210         while (find_next_entry_inc(iter) != NULL) {
3211                 enum print_line_t ret;
3212                 int len = iter->seq.len;
3213
3214                 ret = print_trace_line(iter);
3215                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3216                         /* don't print partial lines */
3217                         iter->seq.len = len;
3218                         break;
3219                 }
3220
3221                 trace_consume(iter);
3222
3223                 if (iter->seq.len >= cnt)
3224                         break;
3225         }
3226
3227         /* Now copy what we have to the user */
3228         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3229         if (iter->seq.readpos >= iter->seq.len)
3230                 trace_seq_reset(&iter->seq);
3231
3232         /*
3233          * If there was nothing to send to user, inspite of consuming trace
3234          * entries, go back to wait for more entries.
3235          */
3236         if (sret == -EBUSY)
3237                 goto waitagain;
3238
3239 out:
3240         mutex_unlock(&trace_types_lock);
3241
3242         return sret;
3243 }
3244
3245 static ssize_t
3246 tracing_entries_read(struct file *filp, char __user *ubuf,
3247                      size_t cnt, loff_t *ppos)
3248 {
3249         struct trace_array *tr = filp->private_data;
3250         char buf[64];
3251         int r;
3252
3253         r = sprintf(buf, "%lu\n", tr->entries >> 10);
3254         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3255 }
3256
3257 static ssize_t
3258 tracing_entries_write(struct file *filp, const char __user *ubuf,
3259                       size_t cnt, loff_t *ppos)
3260 {
3261         unsigned long val;
3262         char buf[64];
3263         int ret, cpu;
3264
3265         if (cnt >= sizeof(buf))
3266                 return -EINVAL;
3267
3268         if (copy_from_user(&buf, ubuf, cnt))
3269                 return -EFAULT;
3270
3271         buf[cnt] = 0;
3272
3273         ret = strict_strtoul(buf, 10, &val);
3274         if (ret < 0)
3275                 return ret;
3276
3277         /* must have at least 1 entry */
3278         if (!val)
3279                 return -EINVAL;
3280
3281         mutex_lock(&trace_types_lock);
3282
3283         tracing_stop();
3284
3285         /* disable all cpu buffers */
3286         for_each_tracing_cpu(cpu) {
3287                 if (global_trace.data[cpu])
3288                         atomic_inc(&global_trace.data[cpu]->disabled);
3289                 if (max_tr.data[cpu])
3290                         atomic_inc(&max_tr.data[cpu]->disabled);
3291         }
3292
3293         /* value is in KB */
3294         val <<= 10;
3295
3296         if (val != global_trace.entries) {
3297                 ret = ring_buffer_resize(global_trace.buffer, val);
3298                 if (ret < 0) {
3299                         cnt = ret;
3300                         goto out;
3301                 }
3302
3303                 ret = ring_buffer_resize(max_tr.buffer, val);
3304                 if (ret < 0) {
3305                         int r;
3306                         cnt = ret;
3307                         r = ring_buffer_resize(global_trace.buffer,
3308                                                global_trace.entries);
3309                         if (r < 0) {
3310                                 /* AARGH! We are left with different
3311                                  * size max buffer!!!! */
3312                                 WARN_ON(1);
3313                                 tracing_disabled = 1;
3314                         }
3315                         goto out;
3316                 }
3317
3318                 global_trace.entries = val;
3319         }
3320
3321         filp->f_pos += cnt;
3322
3323         /* If check pages failed, return ENOMEM */
3324         if (tracing_disabled)
3325                 cnt = -ENOMEM;
3326  out:
3327         for_each_tracing_cpu(cpu) {
3328                 if (global_trace.data[cpu])
3329                         atomic_dec(&global_trace.data[cpu]->disabled);
3330                 if (max_tr.data[cpu])
3331                         atomic_dec(&max_tr.data[cpu]->disabled);
3332         }
3333
3334         tracing_start();
3335         max_tr.entries = global_trace.entries;
3336         mutex_unlock(&trace_types_lock);
3337
3338         return cnt;
3339 }
3340
3341 static int mark_printk(const char *fmt, ...)
3342 {
3343         int ret;
3344         va_list args;
3345         va_start(args, fmt);
3346         ret = trace_vprintk(0, fmt, args);
3347         va_end(args);
3348         return ret;
3349 }
3350
3351 static ssize_t
3352 tracing_mark_write(struct file *filp, const char __user *ubuf,
3353                                         size_t cnt, loff_t *fpos)
3354 {
3355         char *buf;
3356         char *end;
3357
3358         if (tracing_disabled)
3359                 return -EINVAL;
3360
3361         if (cnt > TRACE_BUF_SIZE)
3362                 cnt = TRACE_BUF_SIZE;
3363
3364         buf = kmalloc(cnt + 1, GFP_KERNEL);
3365         if (buf == NULL)
3366                 return -ENOMEM;
3367
3368         if (copy_from_user(buf, ubuf, cnt)) {
3369                 kfree(buf);
3370                 return -EFAULT;
3371         }
3372
3373         /* Cut from the first nil or newline. */
3374         buf[cnt] = '\0';
3375         end = strchr(buf, '\n');
3376         if (end)
3377                 *end = '\0';
3378
3379         cnt = mark_printk("%s\n", buf);
3380         kfree(buf);
3381         *fpos += cnt;
3382
3383         return cnt;
3384 }
3385
3386 static struct file_operations tracing_max_lat_fops = {
3387         .open           = tracing_open_generic,
3388         .read           = tracing_max_lat_read,
3389         .write          = tracing_max_lat_write,
3390 };
3391
3392 static struct file_operations tracing_ctrl_fops = {
3393         .open           = tracing_open_generic,
3394         .read           = tracing_ctrl_read,
3395         .write          = tracing_ctrl_write,
3396 };
3397
3398 static struct file_operations set_tracer_fops = {
3399         .open           = tracing_open_generic,
3400         .read           = tracing_set_trace_read,
3401         .write          = tracing_set_trace_write,
3402 };
3403
3404 static struct file_operations tracing_pipe_fops = {
3405         .open           = tracing_open_pipe,
3406         .poll           = tracing_poll_pipe,
3407         .read           = tracing_read_pipe,
3408         .release        = tracing_release_pipe,
3409 };
3410
3411 static struct file_operations tracing_entries_fops = {
3412         .open           = tracing_open_generic,
3413         .read           = tracing_entries_read,
3414         .write          = tracing_entries_write,
3415 };
3416
3417 static struct file_operations tracing_mark_fops = {
3418         .open           = tracing_open_generic,
3419         .write          = tracing_mark_write,
3420 };
3421
3422 #ifdef CONFIG_DYNAMIC_FTRACE
3423
3424 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3425 {
3426         return 0;
3427 }
3428
3429 static ssize_t
3430 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3431                   size_t cnt, loff_t *ppos)
3432 {
3433         static char ftrace_dyn_info_buffer[1024];
3434         static DEFINE_MUTEX(dyn_info_mutex);
3435         unsigned long *p = filp->private_data;
3436         char *buf = ftrace_dyn_info_buffer;
3437         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3438         int r;
3439
3440         mutex_lock(&dyn_info_mutex);
3441         r = sprintf(buf, "%ld ", *p);
3442
3443         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3444         buf[r++] = '\n';
3445
3446         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3447
3448         mutex_unlock(&dyn_info_mutex);
3449
3450         return r;
3451 }
3452
3453 static struct file_operations tracing_dyn_info_fops = {
3454         .open           = tracing_open_generic,
3455         .read           = tracing_read_dyn_info,
3456 };
3457 #endif
3458
3459 static struct dentry *d_tracer;
3460
3461 struct dentry *tracing_init_dentry(void)
3462 {
3463         static int once;
3464
3465         if (d_tracer)
3466                 return d_tracer;
3467
3468         d_tracer = debugfs_create_dir("tracing", NULL);
3469
3470         if (!d_tracer && !once) {
3471                 once = 1;
3472                 pr_warning("Could not create debugfs directory 'tracing'\n");
3473                 return NULL;
3474         }
3475
3476         return d_tracer;
3477 }
3478
3479 #ifdef CONFIG_FTRACE_SELFTEST
3480 /* Let selftest have access to static functions in this file */
3481 #include "trace_selftest.c"
3482 #endif
3483
3484 static __init int tracer_init_debugfs(void)
3485 {
3486         struct dentry *d_tracer;
3487         struct dentry *entry;
3488
3489         d_tracer = tracing_init_dentry();
3490
3491         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3492                                     &global_trace, &tracing_ctrl_fops);
3493         if (!entry)
3494                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3495
3496         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3497                                     NULL, &tracing_iter_fops);
3498         if (!entry)
3499                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3500
3501         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3502                                     NULL, &tracing_cpumask_fops);
3503         if (!entry)
3504                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3505
3506         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3507                                     &global_trace, &tracing_lt_fops);
3508         if (!entry)
3509                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3510
3511         entry = debugfs_create_file("trace", 0444, d_tracer,
3512                                     &global_trace, &tracing_fops);
3513         if (!entry)
3514                 pr_warning("Could not create debugfs 'trace' entry\n");
3515
3516         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3517                                     &global_trace, &show_traces_fops);
3518         if (!entry)
3519                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3520
3521         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3522                                     &global_trace, &set_tracer_fops);
3523         if (!entry)
3524                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3525
3526         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3527                                     &tracing_max_latency,
3528                                     &tracing_max_lat_fops);
3529         if (!entry)
3530                 pr_warning("Could not create debugfs "
3531                            "'tracing_max_latency' entry\n");
3532
3533         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3534                                     &tracing_thresh, &tracing_max_lat_fops);
3535         if (!entry)
3536                 pr_warning("Could not create debugfs "
3537                            "'tracing_thresh' entry\n");
3538         entry = debugfs_create_file("README", 0644, d_tracer,
3539                                     NULL, &tracing_readme_fops);
3540         if (!entry)
3541                 pr_warning("Could not create debugfs 'README' entry\n");
3542
3543         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3544                                     NULL, &tracing_pipe_fops);
3545         if (!entry)
3546                 pr_warning("Could not create debugfs "
3547                            "'trace_pipe' entry\n");
3548
3549         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3550                                     &global_trace, &tracing_entries_fops);
3551         if (!entry)
3552                 pr_warning("Could not create debugfs "
3553                            "'buffer_size_kb' entry\n");
3554
3555         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3556                                     NULL, &tracing_mark_fops);
3557         if (!entry)
3558                 pr_warning("Could not create debugfs "
3559                            "'trace_marker' entry\n");
3560
3561 #ifdef CONFIG_DYNAMIC_FTRACE
3562         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3563                                     &ftrace_update_tot_cnt,
3564                                     &tracing_dyn_info_fops);
3565         if (!entry)
3566                 pr_warning("Could not create debugfs "
3567                            "'dyn_ftrace_total_info' entry\n");
3568 #endif
3569 #ifdef CONFIG_SYSPROF_TRACER
3570         init_tracer_sysprof_debugfs(d_tracer);
3571 #endif
3572         return 0;
3573 }
3574
3575 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3576 {
3577         static DEFINE_SPINLOCK(trace_buf_lock);
3578         static char trace_buf[TRACE_BUF_SIZE];
3579
3580         struct ring_buffer_event *event;
3581         struct trace_array *tr = &global_trace;
3582         struct trace_array_cpu *data;
3583         struct print_entry *entry;
3584         unsigned long flags, irq_flags;
3585         int cpu, len = 0, size, pc;
3586
3587         if (tracing_disabled)
3588                 return 0;
3589
3590         pc = preempt_count();
3591         preempt_disable_notrace();
3592         cpu = raw_smp_processor_id();
3593         data = tr->data[cpu];
3594
3595         if (unlikely(atomic_read(&data->disabled)))
3596                 goto out;
3597
3598         spin_lock_irqsave(&trace_buf_lock, flags);
3599         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3600
3601         len = min(len, TRACE_BUF_SIZE-1);
3602         trace_buf[len] = 0;
3603
3604         size = sizeof(*entry) + len + 1;
3605         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3606         if (!event)
3607                 goto out_unlock;
3608         entry = ring_buffer_event_data(event);
3609         tracing_generic_entry_update(&entry->ent, flags, pc);
3610         entry->ent.type                 = TRACE_PRINT;
3611         entry->ip                       = ip;
3612
3613         memcpy(&entry->buf, trace_buf, len);
3614         entry->buf[len] = 0;
3615         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3616
3617  out_unlock:
3618         spin_unlock_irqrestore(&trace_buf_lock, flags);
3619
3620  out:
3621         preempt_enable_notrace();
3622
3623         return len;
3624 }
3625 EXPORT_SYMBOL_GPL(trace_vprintk);
3626
3627 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3628 {
3629         int ret;
3630         va_list ap;
3631
3632         if (!(trace_flags & TRACE_ITER_PRINTK))
3633                 return 0;
3634
3635         va_start(ap, fmt);
3636         ret = trace_vprintk(ip, fmt, ap);
3637         va_end(ap);
3638         return ret;
3639 }
3640 EXPORT_SYMBOL_GPL(__ftrace_printk);
3641
3642 static int trace_panic_handler(struct notifier_block *this,
3643                                unsigned long event, void *unused)
3644 {
3645         if (ftrace_dump_on_oops)
3646                 ftrace_dump();
3647         return NOTIFY_OK;
3648 }
3649
3650 static struct notifier_block trace_panic_notifier = {
3651         .notifier_call  = trace_panic_handler,
3652         .next           = NULL,
3653         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3654 };
3655
3656 static int trace_die_handler(struct notifier_block *self,
3657                              unsigned long val,
3658                              void *data)
3659 {
3660         switch (val) {
3661         case DIE_OOPS:
3662                 if (ftrace_dump_on_oops)
3663                         ftrace_dump();
3664                 break;
3665         default:
3666                 break;
3667         }
3668         return NOTIFY_OK;
3669 }
3670
3671 static struct notifier_block trace_die_notifier = {
3672         .notifier_call = trace_die_handler,
3673         .priority = 200
3674 };
3675
3676 /*
3677  * printk is set to max of 1024, we really don't need it that big.
3678  * Nothing should be printing 1000 characters anyway.
3679  */
3680 #define TRACE_MAX_PRINT         1000
3681
3682 /*
3683  * Define here KERN_TRACE so that we have one place to modify
3684  * it if we decide to change what log level the ftrace dump
3685  * should be at.
3686  */
3687 #define KERN_TRACE              KERN_INFO
3688
3689 static void
3690 trace_printk_seq(struct trace_seq *s)
3691 {
3692         /* Probably should print a warning here. */
3693         if (s->len >= 1000)
3694                 s->len = 1000;
3695
3696         /* should be zero ended, but we are paranoid. */
3697         s->buffer[s->len] = 0;
3698
3699         printk(KERN_TRACE "%s", s->buffer);
3700
3701         trace_seq_reset(s);
3702 }
3703
3704 void ftrace_dump(void)
3705 {
3706         static DEFINE_SPINLOCK(ftrace_dump_lock);
3707         /* use static because iter can be a bit big for the stack */
3708         static struct trace_iterator iter;
3709         static cpumask_t mask;
3710         static int dump_ran;
3711         unsigned long flags;
3712         int cnt = 0, cpu;
3713
3714         /* only one dump */
3715         spin_lock_irqsave(&ftrace_dump_lock, flags);
3716         if (dump_ran)
3717                 goto out;
3718
3719         dump_ran = 1;
3720
3721         /* No turning back! */
3722         ftrace_kill();
3723
3724         for_each_tracing_cpu(cpu) {
3725                 atomic_inc(&global_trace.data[cpu]->disabled);
3726         }
3727
3728         /* don't look at user memory in panic mode */
3729         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3730
3731         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3732
3733         iter.tr = &global_trace;
3734         iter.trace = current_trace;
3735
3736         /*
3737          * We need to stop all tracing on all CPUS to read the
3738          * the next buffer. This is a bit expensive, but is
3739          * not done often. We fill all what we can read,
3740          * and then release the locks again.
3741          */
3742
3743         cpus_clear(mask);
3744
3745         while (!trace_empty(&iter)) {
3746
3747                 if (!cnt)
3748                         printk(KERN_TRACE "---------------------------------\n");
3749
3750                 cnt++;
3751
3752                 /* reset all but tr, trace, and overruns */
3753                 memset(&iter.seq, 0,
3754                        sizeof(struct trace_iterator) -
3755                        offsetof(struct trace_iterator, seq));
3756                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3757                 iter.pos = -1;
3758
3759                 if (find_next_entry_inc(&iter) != NULL) {
3760                         print_trace_line(&iter);
3761                         trace_consume(&iter);
3762                 }
3763
3764                 trace_printk_seq(&iter.seq);
3765         }
3766
3767         if (!cnt)
3768                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3769         else
3770                 printk(KERN_TRACE "---------------------------------\n");
3771
3772  out:
3773         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3774 }
3775
3776 __init static int tracer_alloc_buffers(void)
3777 {
3778         struct trace_array_cpu *data;
3779         int i;
3780
3781         /* TODO: make the number of buffers hot pluggable with CPUS */
3782         tracing_buffer_mask = cpu_possible_map;
3783
3784         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3785                                                    TRACE_BUFFER_FLAGS);
3786         if (!global_trace.buffer) {
3787                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3788                 WARN_ON(1);
3789                 return 0;
3790         }
3791         global_trace.entries = ring_buffer_size(global_trace.buffer);
3792
3793 #ifdef CONFIG_TRACER_MAX_TRACE
3794         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3795                                              TRACE_BUFFER_FLAGS);
3796         if (!max_tr.buffer) {
3797                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3798                 WARN_ON(1);
3799                 ring_buffer_free(global_trace.buffer);
3800                 return 0;
3801         }
3802         max_tr.entries = ring_buffer_size(max_tr.buffer);
3803         WARN_ON(max_tr.entries != global_trace.entries);
3804 #endif
3805
3806         /* Allocate the first page for all buffers */
3807         for_each_tracing_cpu(i) {
3808                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3809                 max_tr.data[i] = &per_cpu(max_data, i);
3810         }
3811
3812         trace_init_cmdlines();
3813
3814         register_tracer(&nop_trace);
3815 #ifdef CONFIG_BOOT_TRACER
3816         register_tracer(&boot_tracer);
3817         current_trace = &boot_tracer;
3818         current_trace->init(&global_trace);
3819 #else
3820         current_trace = &nop_trace;
3821 #endif
3822
3823         /* All seems OK, enable tracing */
3824         tracing_disabled = 0;
3825
3826         atomic_notifier_chain_register(&panic_notifier_list,
3827                                        &trace_panic_notifier);
3828
3829         register_die_notifier(&trace_die_notifier);
3830
3831         return 0;
3832 }
3833 early_initcall(tracer_alloc_buffers);
3834 fs_initcall(tracer_init_debugfs);