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 /*
  48  * We need to change this state when a selftest is running.
  49  * A selftest will lurk into the ring-buffer to count the
  50  * entries inserted during the selftest although some concurrent
  51  * insertions into the ring-buffer such as ftrace_printk could occurred
  52  * at the same time, giving false positive or negative results.
  53  */
  54 static bool __read_mostly tracing_selftest_running;
  55
  56 /* For tracers that don't implement custom flags */
  57 static struct tracer_opt dummy_tracer_opt[] = {
  58         { }
  59 };
  60
  61 static struct tracer_flags dummy_tracer_flags = {
  62         .val = 0,
  63         .opts = dummy_tracer_opt
  64 };
  65
  66 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
  67 {
  68         return 0;
  69 }
  70
  71 /*
  72  * Kill all tracing for good (never come back).
  73  * It is initialized to 1 but will turn to zero if the initialization
  74  * of the tracer is successful. But that is the only place that sets
  75  * this back to zero.
  76  */
  77 int tracing_disabled = 1;
  78
  79 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
  80
  81 static inline void ftrace_disable_cpu(void)
  82 {
  83         preempt_disable();
  84         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
  85 }
  86
  87 static inline void ftrace_enable_cpu(void)
  88 {
  89         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
  90         preempt_enable();
  91 }
  92
  93 static cpumask_t __read_mostly          tracing_buffer_mask;
  94
  95 #define for_each_tracing_cpu(cpu)       \
  96         for_each_cpu_mask(cpu, tracing_buffer_mask)
  97
  98 /*
  99  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
 100  *
 101  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
 102  * is set, then ftrace_dump is called. This will output the contents
 103  * of the ftrace buffers to the console.  This is very useful for
 104  * capturing traces that lead to crashes and outputing it to a
 105  * serial console.
 106  *
 107  * It is default off, but you can enable it with either specifying
 108  * "ftrace_dump_on_oops" in the kernel command line, or setting
 109  * /proc/sys/kernel/ftrace_dump_on_oops to true.
 110  */
 111 int ftrace_dump_on_oops;
 112
 113 static int tracing_set_tracer(char *buf);
 114
 115 static int __init set_ftrace(char *str)
 116 {
 117         tracing_set_tracer(str);
 118         return 1;
 119 }
 120 __setup("ftrace", set_ftrace);
 121
 122 static int __init set_ftrace_dump_on_oops(char *str)
 123 {
 124         ftrace_dump_on_oops = 1;
 125         return 1;
 126 }
 127 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
 128
 129 long
 130 ns2usecs(cycle_t nsec)
 131 {
 132         nsec += 500;
 133         do_div(nsec, 1000);
 134         return nsec;
 135 }
 136
 137 cycle_t ftrace_now(int cpu)
 138 {
 139         u64 ts = ring_buffer_time_stamp(cpu);
 140         ring_buffer_normalize_time_stamp(cpu, &ts);
 141         return ts;
 142 }
 143
 144 /*
 145  * The global_trace is the descriptor that holds the tracing
 146  * buffers for the live tracing. For each CPU, it contains
 147  * a link list of pages that will store trace entries. The
 148  * page descriptor of the pages in the memory is used to hold
 149  * the link list by linking the lru item in the page descriptor
 150  * to each of the pages in the buffer per CPU.
 151  *
 152  * For each active CPU there is a data field that holds the
 153  * pages for the buffer for that CPU. Each CPU has the same number
 154  * of pages allocated for its buffer.
 155  */
 156 static struct trace_array       global_trace;
 157
 158 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
 159
 160 /*
 161  * The max_tr is used to snapshot the global_trace when a maximum
 162  * latency is reached. Some tracers will use this to store a maximum
 163  * trace while it continues examining live traces.
 164  *
 165  * The buffers for the max_tr are set up the same as the global_trace.
 166  * When a snapshot is taken, the link list of the max_tr is swapped
 167  * with the link list of the global_trace and the buffers are reset for
 168  * the global_trace so the tracing can continue.
 169  */
 170 static struct trace_array       max_tr;
 171
 172 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 173
 174 /* tracer_enabled is used to toggle activation of a tracer */
 175 static int                      tracer_enabled = 1;
 176
 177 /**
 178  * tracing_is_enabled - return tracer_enabled status
 179  *
 180  * This function is used by other tracers to know the status
 181  * of the tracer_enabled flag.  Tracers may use this function
 182  * to know if it should enable their features when starting
 183  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
 184  */
 185 int tracing_is_enabled(void)
 186 {
 187         return tracer_enabled;
 188 }
 189
 190 /* function tracing enabled */
 191 int                             ftrace_function_enabled;
 192
 193 /*
 194  * trace_buf_size is the size in bytes that is allocated
 195  * for a buffer. Note, the number of bytes is always rounded
 196  * to page size.
 197  *
 198  * This number is purposely set to a low number of 16384.
 199  * If the dump on oops happens, it will be much appreciated
 200  * to not have to wait for all that output. Anyway this can be
 201  * boot time and run time configurable.
 202  */
 203 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
 204
 205 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
 206
 207 /* trace_types holds a link list of available tracers. */
 208 static struct tracer            *trace_types __read_mostly;
 209
 210 /* current_trace points to the tracer that is currently active */
 211 static struct tracer            *current_trace __read_mostly;
 212
 213 /*
 214  * max_tracer_type_len is used to simplify the allocating of
 215  * buffers to read userspace tracer names. We keep track of
 216  * the longest tracer name registered.
 217  */
 218 static int                      max_tracer_type_len;
 219
 220 /*
 221  * trace_types_lock is used to protect the trace_types list.
 222  * This lock is also used to keep user access serialized.
 223  * Accesses from userspace will grab this lock while userspace
 224  * activities happen inside the kernel.
 225  */
 226 static DEFINE_MUTEX(trace_types_lock);
 227
 228 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
 229 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 230
 231 /* trace_flags holds trace_options default values */
 232 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
 233         TRACE_ITER_ANNOTATE;
 234
 235 /**
 236  * trace_wake_up - wake up tasks waiting for trace input
 237  *
 238  * Simply wakes up any task that is blocked on the trace_wait
 239  * queue. These is used with trace_poll for tasks polling the trace.
 240  */
 241 void trace_wake_up(void)
 242 {
 243         /*
 244          * The runqueue_is_locked() can fail, but this is the best we
 245          * have for now:
 246          */
 247         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
 248                 wake_up(&trace_wait);
 249 }
 250
 251 static int __init set_buf_size(char *str)
 252 {
 253         unsigned long buf_size;
 254         int ret;
 255
 256         if (!str)
 257                 return 0;
 258         ret = strict_strtoul(str, 0, &buf_size);
 259         /* nr_entries can not be zero */
 260         if (ret < 0 || buf_size == 0)
 261                 return 0;
 262         trace_buf_size = buf_size;
 263         return 1;
 264 }
 265 __setup("trace_buf_size=", set_buf_size);
 266
 267 unsigned long nsecs_to_usecs(unsigned long nsecs)
 268 {
 269         return nsecs / 1000;
 270 }
 271
 272 /* These must match the bit postions in trace_iterator_flags */
 273 static const char *trace_options[] = {
 274         "print-parent",
 275         "sym-offset",
 276         "sym-addr",
 277         "verbose",
 278         "raw",
 279         "hex",
 280         "bin",
 281         "block",
 282         "stacktrace",
 283         "sched-tree",
 284         "ftrace_printk",
 285         "ftrace_preempt",
 286         "branch",
 287         "annotate",
 288         "userstacktrace",
 289         "sym-userobj",
 290         "printk-msg-only",
 291         NULL
 292 };
 293
 294 /*
 295  * ftrace_max_lock is used to protect the swapping of buffers
 296  * when taking a max snapshot. The buffers themselves are
 297  * protected by per_cpu spinlocks. But the action of the swap
 298  * needs its own lock.
 299  *
 300  * This is defined as a raw_spinlock_t in order to help
 301  * with performance when lockdep debugging is enabled.
 302  */
 303 static raw_spinlock_t ftrace_max_lock =
 304         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 305
 306 /*
 307  * Copy the new maximum trace into the separate maximum-trace
 308  * structure. (this way the maximum trace is permanently saved,
 309  * for later retrieval via /debugfs/tracing/latency_trace)
 310  */
 311 static void
 312 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 313 {
 314         struct trace_array_cpu *data = tr->data[cpu];
 315
 316         max_tr.cpu = cpu;
 317         max_tr.time_start = data->preempt_timestamp;
 318
 319         data = max_tr.data[cpu];
 320         data->saved_latency = tracing_max_latency;
 321
 322         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 323         data->pid = tsk->pid;
 324         data->uid = task_uid(tsk);
 325         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 326         data->policy = tsk->policy;
 327         data->rt_priority = tsk->rt_priority;
 328
 329         /* record this tasks comm */
 330         tracing_record_cmdline(current);
 331 }
 332
 333 /**
 334  * trace_seq_printf - sequence printing of trace information
 335  * @s: trace sequence descriptor
 336  * @fmt: printf format string
 337  *
 338  * The tracer may use either sequence operations or its own
 339  * copy to user routines. To simplify formating of a trace
 340  * trace_seq_printf is used to store strings into a special
 341  * buffer (@s). Then the output may be either used by
 342  * the sequencer or pulled into another buffer.
 343  */
 344 int
 345 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 346 {
 347         int len = (PAGE_SIZE - 1) - s->len;
 348         va_list ap;
 349         int ret;
 350
 351         if (!len)
 352                 return 0;
 353
 354         va_start(ap, fmt);
 355         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 356         va_end(ap);
 357
 358         /* If we can't write it all, don't bother writing anything */
 359         if (ret >= len)
 360                 return 0;
 361
 362         s->len += ret;
 363
 364         return len;
 365 }
 366
 367 /**
 368  * trace_seq_puts - trace sequence printing of simple string
 369  * @s: trace sequence descriptor
 370  * @str: simple string to record
 371  *
 372  * The tracer may use either the sequence operations or its own
 373  * copy to user routines. This function records a simple string
 374  * into a special buffer (@s) for later retrieval by a sequencer
 375  * or other mechanism.
 376  */
 377 static int
 378 trace_seq_puts(struct trace_seq *s, const char *str)
 379 {
 380         int len = strlen(str);
 381
 382         if (len > ((PAGE_SIZE - 1) - s->len))
 383                 return 0;
 384
 385         memcpy(s->buffer + s->len, str, len);
 386         s->len += len;
 387
 388         return len;
 389 }
 390
 391 static int
 392 trace_seq_putc(struct trace_seq *s, unsigned char c)
 393 {
 394         if (s->len >= (PAGE_SIZE - 1))
 395                 return 0;
 396
 397         s->buffer[s->len++] = c;
 398
 399         return 1;
 400 }
 401
 402 static int
 403 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 404 {
 405         if (len > ((PAGE_SIZE - 1) - s->len))
 406                 return 0;
 407
 408         memcpy(s->buffer + s->len, mem, len);
 409         s->len += len;
 410
 411         return len;
 412 }
 413
 414 #define MAX_MEMHEX_BYTES        8
 415 #define HEX_CHARS               (MAX_MEMHEX_BYTES*2 + 1)
 416
 417 static int
 418 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 419 {
 420         unsigned char hex[HEX_CHARS];
 421         unsigned char *data = mem;
 422         int i, j;
 423
 424 #ifdef __BIG_ENDIAN
 425         for (i = 0, j = 0; i < len; i++) {
 426 #else
 427         for (i = len-1, j = 0; i >= 0; i--) {
 428 #endif
 429                 hex[j++] = hex_asc_hi(data[i]);
 430                 hex[j++] = hex_asc_lo(data[i]);
 431         }
 432         hex[j++] = ' ';
 433
 434         return trace_seq_putmem(s, hex, j);
 435 }
 436
 437 static int
 438 trace_seq_path(struct trace_seq *s, struct path *path)
 439 {
 440         unsigned char *p;
 441
 442         if (s->len >= (PAGE_SIZE - 1))
 443                 return 0;
 444         p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
 445         if (!IS_ERR(p)) {
 446                 p = mangle_path(s->buffer + s->len, p, "\n");
 447                 if (p) {
 448                         s->len = p - s->buffer;
 449                         return 1;
 450                 }
 451         } else {
 452                 s->buffer[s->len++] = '?';
 453                 return 1;
 454         }
 455
 456         return 0;
 457 }
 458
 459 static void
 460 trace_seq_reset(struct trace_seq *s)
 461 {
 462         s->len = 0;
 463         s->readpos = 0;
 464 }
 465
 466 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
 467 {
 468         int len;
 469         int ret;
 470
 471         if (s->len <= s->readpos)
 472                 return -EBUSY;
 473
 474         len = s->len - s->readpos;
 475         if (cnt > len)
 476                 cnt = len;
 477         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
 478         if (ret)
 479                 return -EFAULT;
 480
 481         s->readpos += len;
 482         return cnt;
 483 }
 484
 485 static void
 486 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 487 {
 488         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 489
 490         s->buffer[len] = 0;
 491         seq_puts(m, s->buffer);
 492
 493         trace_seq_reset(s);
 494 }
 495
 496 /**
 497  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
 498  * @tr: tracer
 499  * @tsk: the task with the latency
 500  * @cpu: The cpu that initiated the trace.
 501  *
 502  * Flip the buffers between the @tr and the max_tr and record information
 503  * about which task was the cause of this latency.
 504  */
 505 void
 506 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 507 {
 508         struct ring_buffer *buf = tr->buffer;
 509
 510         WARN_ON_ONCE(!irqs_disabled());
 511         __raw_spin_lock(&ftrace_max_lock);
 512
 513         tr->buffer = max_tr.buffer;
 514         max_tr.buffer = buf;
 515
 516         ftrace_disable_cpu();
 517         ring_buffer_reset(tr->buffer);
 518         ftrace_enable_cpu();
 519
 520         __update_max_tr(tr, tsk, cpu);
 521         __raw_spin_unlock(&ftrace_max_lock);
 522 }
 523
 524 /**
 525  * update_max_tr_single - only copy one trace over, and reset the rest
 526  * @tr - tracer
 527  * @tsk - task with the latency
 528  * @cpu - the cpu of the buffer to copy.
 529  *
 530  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
 531  */
 532 void
 533 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 534 {
 535         int ret;
 536
 537         WARN_ON_ONCE(!irqs_disabled());
 538         __raw_spin_lock(&ftrace_max_lock);
 539
 540         ftrace_disable_cpu();
 541
 542         ring_buffer_reset(max_tr.buffer);
 543         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
 544
 545         ftrace_enable_cpu();
 546
 547         WARN_ON_ONCE(ret);
 548
 549         __update_max_tr(tr, tsk, cpu);
 550         __raw_spin_unlock(&ftrace_max_lock);
 551 }
 552
 553 /**
 554  * register_tracer - register a tracer with the ftrace system.
 555  * @type - the plugin for the tracer
 556  *
 557  * Register a new plugin tracer.
 558  */
 559 int register_tracer(struct tracer *type)
 560 {
 561         struct tracer *t;
 562         int len;
 563         int ret = 0;
 564
 565         if (!type->name) {
 566                 pr_info("Tracer must have a name\n");
 567                 return -1;
 568         }
 569
 570         /*
 571          * When this gets called we hold the BKL which means that
 572          * preemption is disabled. Various trace selftests however
 573          * need to disable and enable preemption for successful tests.
 574          * So we drop the BKL here and grab it after the tests again.
 575          */
 576         unlock_kernel();
 577         mutex_lock(&trace_types_lock);
 578
 579         tracing_selftest_running = true;
 580
 581         for (t = trace_types; t; t = t->next) {
 582                 if (strcmp(type->name, t->name) == 0) {
 583                         /* already found */
 584                         pr_info("Trace %s already registered\n",
 585                                 type->name);
 586                         ret = -1;
 587                         goto out;
 588                 }
 589         }
 590
 591         if (!type->set_flag)
 592                 type->set_flag = &dummy_set_flag;
 593         if (!type->flags)
 594                 type->flags = &dummy_tracer_flags;
 595         else
 596                 if (!type->flags->opts)
 597                         type->flags->opts = dummy_tracer_opt;
 598
 599 #ifdef CONFIG_FTRACE_STARTUP_TEST
 600         if (type->selftest) {
 601                 struct tracer *saved_tracer = current_trace;
 602                 struct trace_array *tr = &global_trace;
 603                 int i;
 604
 605                 /*
 606                  * Run a selftest on this tracer.
 607                  * Here we reset the trace buffer, and set the current
 608                  * tracer to be this tracer. The tracer can then run some
 609                  * internal tracing to verify that everything is in order.
 610                  * If we fail, we do not register this tracer.
 611                  */
 612                 for_each_tracing_cpu(i)
 613                         tracing_reset(tr, i);
 614
 615                 current_trace = type;
 616                 /* the test is responsible for initializing and enabling */
 617                 pr_info("Testing tracer %s: ", type->name);
 618                 ret = type->selftest(type, tr);
 619                 /* the test is responsible for resetting too */
 620                 current_trace = saved_tracer;
 621                 if (ret) {
 622                         printk(KERN_CONT "FAILED!\n");
 623                         goto out;
 624                 }
 625                 /* Only reset on passing, to avoid touching corrupted buffers */
 626                 for_each_tracing_cpu(i)
 627                         tracing_reset(tr, i);
 628
 629                 printk(KERN_CONT "PASSED\n");
 630         }
 631 #endif
 632
 633         type->next = trace_types;
 634         trace_types = type;
 635         len = strlen(type->name);
 636         if (len > max_tracer_type_len)
 637                 max_tracer_type_len = len;
 638
 639  out:
 640         tracing_selftest_running = false;
 641         mutex_unlock(&trace_types_lock);
 642         lock_kernel();
 643
 644         return ret;
 645 }
 646
 647 void unregister_tracer(struct tracer *type)
 648 {
 649         struct tracer **t;
 650         int len;
 651
 652         mutex_lock(&trace_types_lock);
 653         for (t = &trace_types; *t; t = &(*t)->next) {
 654                 if (*t == type)
 655                         goto found;
 656         }
 657         pr_info("Trace %s not registered\n", type->name);
 658         goto out;
 659
 660  found:
 661         *t = (*t)->next;
 662         if (strlen(type->name) != max_tracer_type_len)
 663                 goto out;
 664
 665         max_tracer_type_len = 0;
 666         for (t = &trace_types; *t; t = &(*t)->next) {
 667                 len = strlen((*t)->name);
 668                 if (len > max_tracer_type_len)
 669                         max_tracer_type_len = len;
 670         }
 671  out:
 672         mutex_unlock(&trace_types_lock);
 673 }
 674
 675 void tracing_reset(struct trace_array *tr, int cpu)
 676 {
 677         ftrace_disable_cpu();
 678         ring_buffer_reset_cpu(tr->buffer, cpu);
 679         ftrace_enable_cpu();
 680 }
 681
 682 void tracing_reset_online_cpus(struct trace_array *tr)
 683 {
 684         int cpu;
 685
 686         tr->time_start = ftrace_now(tr->cpu);
 687
 688         for_each_online_cpu(cpu)
 689                 tracing_reset(tr, cpu);
 690 }
 691
 692 #define SAVED_CMDLINES 128
 693 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 694 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 695 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 696 static int cmdline_idx;
 697 static DEFINE_SPINLOCK(trace_cmdline_lock);
 698
 699 /* temporary disable recording */
 700 atomic_t trace_record_cmdline_disabled __read_mostly;
 701
 702 static void trace_init_cmdlines(void)
 703 {
 704         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 705         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 706         cmdline_idx = 0;
 707 }
 708
 709 static int trace_stop_count;
 710 static DEFINE_SPINLOCK(tracing_start_lock);
 711
 712 /**
 713  * ftrace_off_permanent - disable all ftrace code permanently
 714  *
 715  * This should only be called when a serious anomally has
 716  * been detected.  This will turn off the function tracing,
 717  * ring buffers, and other tracing utilites. It takes no
 718  * locks and can be called from any context.
 719  */
 720 void ftrace_off_permanent(void)
 721 {
 722         tracing_disabled = 1;
 723         ftrace_stop();
 724         tracing_off_permanent();
 725 }
 726
 727 /**
 728  * tracing_start - quick start of the tracer
 729  *
 730  * If tracing is enabled but was stopped by tracing_stop,
 731  * this will start the tracer back up.
 732  */
 733 void tracing_start(void)
 734 {
 735         struct ring_buffer *buffer;
 736         unsigned long flags;
 737
 738         if (tracing_disabled)
 739                 return;
 740
 741         spin_lock_irqsave(&tracing_start_lock, flags);
 742         if (--trace_stop_count)
 743                 goto out;
 744
 745         if (trace_stop_count < 0) {
 746                 /* Someone screwed up their debugging */
 747                 WARN_ON_ONCE(1);
 748                 trace_stop_count = 0;
 749                 goto out;
 750         }
 751
 752
 753         buffer = global_trace.buffer;
 754         if (buffer)
 755                 ring_buffer_record_enable(buffer);
 756
 757         buffer = max_tr.buffer;
 758         if (buffer)
 759                 ring_buffer_record_enable(buffer);
 760
 761         ftrace_start();
 762  out:
 763         spin_unlock_irqrestore(&tracing_start_lock, flags);
 764 }
 765
 766 /**
 767  * tracing_stop - quick stop of the tracer
 768  *
 769  * Light weight way to stop tracing. Use in conjunction with
 770  * tracing_start.
 771  */
 772 void tracing_stop(void)
 773 {
 774         struct ring_buffer *buffer;
 775         unsigned long flags;
 776
 777         ftrace_stop();
 778         spin_lock_irqsave(&tracing_start_lock, flags);
 779         if (trace_stop_count++)
 780                 goto out;
 781
 782         buffer = global_trace.buffer;
 783         if (buffer)
 784                 ring_buffer_record_disable(buffer);
 785
 786         buffer = max_tr.buffer;
 787         if (buffer)
 788                 ring_buffer_record_disable(buffer);
 789
 790  out:
 791         spin_unlock_irqrestore(&tracing_start_lock, flags);
 792 }
 793
 794 void trace_stop_cmdline_recording(void);
 795
 796 static void trace_save_cmdline(struct task_struct *tsk)
 797 {
 798         unsigned map;
 799         unsigned idx;
 800
 801         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 802                 return;
 803
 804         /*
 805          * It's not the end of the world if we don't get
 806          * the lock, but we also don't want to spin
 807          * nor do we want to disable interrupts,
 808          * so if we miss here, then better luck next time.
 809          */
 810         if (!spin_trylock(&trace_cmdline_lock))
 811                 return;
 812
 813         idx = map_pid_to_cmdline[tsk->pid];
 814         if (idx >= SAVED_CMDLINES) {
 815                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 816
 817                 map = map_cmdline_to_pid[idx];
 818                 if (map <= PID_MAX_DEFAULT)
 819                         map_pid_to_cmdline[map] = (unsigned)-1;
 820
 821                 map_pid_to_cmdline[tsk->pid] = idx;
 822
 823                 cmdline_idx = idx;
 824         }
 825
 826         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 827
 828         spin_unlock(&trace_cmdline_lock);
 829 }
 830
 831 char *trace_find_cmdline(int pid)
 832 {
 833         char *cmdline = "<...>";
 834         unsigned map;
 835
 836         if (!pid)
 837                 return "<idle>";
 838
 839         if (pid > PID_MAX_DEFAULT)
 840                 goto out;
 841
 842         map = map_pid_to_cmdline[pid];
 843         if (map >= SAVED_CMDLINES)
 844                 goto out;
 845
 846         cmdline = saved_cmdlines[map];
 847
 848  out:
 849         return cmdline;
 850 }
 851
 852 void tracing_record_cmdline(struct task_struct *tsk)
 853 {
 854         if (atomic_read(&trace_record_cmdline_disabled))
 855                 return;
 856
 857         trace_save_cmdline(tsk);
 858 }
 859
 860 void
 861 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
 862                              int pc)
 863 {
 864         struct task_struct *tsk = current;
 865
 866         entry->preempt_count            = pc & 0xff;
 867         entry->pid                      = (tsk) ? tsk->pid : 0;
 868         entry->tgid                     = (tsk) ? tsk->tgid : 0;
 869         entry->flags =
 870 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
 871                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 872 #else
 873                 TRACE_FLAG_IRQS_NOSUPPORT |
 874 #endif
 875                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 876                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 877                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 878 }
 879
 880 void
 881 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 882                unsigned long ip, unsigned long parent_ip, unsigned long flags,
 883                int pc)
 884 {
 885         struct ring_buffer_event *event;
 886         struct ftrace_entry *entry;
 887         unsigned long irq_flags;
 888
 889         /* If we are reading the ring buffer, don't trace */
 890         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 891                 return;
 892
 893         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 894                                          &irq_flags);
 895         if (!event)
 896                 return;
 897         entry   = ring_buffer_event_data(event);
 898         tracing_generic_entry_update(&entry->ent, flags, pc);
 899         entry->ent.type                 = TRACE_FN;
 900         entry->ip                       = ip;
 901         entry->parent_ip                = parent_ip;
 902         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 903 }
 904
 905 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 906 static void __trace_graph_entry(struct trace_array *tr,
 907                                 struct trace_array_cpu *data,
 908                                 struct ftrace_graph_ent *trace,
 909                                 unsigned long flags,
 910                                 int pc)
 911 {
 912         struct ring_buffer_event *event;
 913         struct ftrace_graph_ent_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_ENT;
 926         entry->graph_ent                        = *trace;
 927         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
 928 }
 929
 930 static void __trace_graph_return(struct trace_array *tr,
 931                                 struct trace_array_cpu *data,
 932                                 struct ftrace_graph_ret *trace,
 933                                 unsigned long flags,
 934                                 int pc)
 935 {
 936         struct ring_buffer_event *event;
 937         struct ftrace_graph_ret_entry *entry;
 938         unsigned long irq_flags;
 939
 940         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 941                 return;
 942
 943         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
 944                                          &irq_flags);
 945         if (!event)
 946                 return;
 947         entry   = ring_buffer_event_data(event);
 948         tracing_generic_entry_update(&entry->ent, flags, pc);
 949         entry->ent.type                 = TRACE_GRAPH_RET;
 950         entry->ret                              = *trace;
 951         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
 952 }
 953 #endif
 954
 955 void
 956 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 957        unsigned long ip, unsigned long parent_ip, unsigned long flags,
 958        int pc)
 959 {
 960         if (likely(!atomic_read(&data->disabled)))
 961                 trace_function(tr, data, ip, parent_ip, flags, pc);
 962 }
 963
 964 static void ftrace_trace_stack(struct trace_array *tr,
 965                                struct trace_array_cpu *data,
 966                                unsigned long flags,
 967                                int skip, int pc)
 968 {
 969 #ifdef CONFIG_STACKTRACE
 970         struct ring_buffer_event *event;
 971         struct stack_entry *entry;
 972         struct stack_trace trace;
 973         unsigned long irq_flags;
 974
 975         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 976                 return;
 977
 978         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 979                                          &irq_flags);
 980         if (!event)
 981                 return;
 982         entry   = ring_buffer_event_data(event);
 983         tracing_generic_entry_update(&entry->ent, flags, pc);
 984         entry->ent.type         = TRACE_STACK;
 985
 986         memset(&entry->caller, 0, sizeof(entry->caller));
 987
 988         trace.nr_entries        = 0;
 989         trace.max_entries       = FTRACE_STACK_ENTRIES;
 990         trace.skip              = skip;
 991         trace.entries           = entry->caller;
 992
 993         save_stack_trace(&trace);
 994         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 995 #endif
 996 }
 997
 998 void __trace_stack(struct trace_array *tr,
 999                    struct trace_array_cpu *data,
1000                    unsigned long flags,
1001                    int skip)
1002 {
1003         ftrace_trace_stack(tr, data, flags, skip, preempt_count());
1004 }
1005
1006 static void ftrace_trace_userstack(struct trace_array *tr,
1007                    struct trace_array_cpu *data,
1008                    unsigned long flags, int pc)
1009 {
1010 #ifdef CONFIG_STACKTRACE
1011         struct ring_buffer_event *event;
1012         struct userstack_entry *entry;
1013         struct stack_trace trace;
1014         unsigned long irq_flags;
1015
1016         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1017                 return;
1018
1019         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1020                                          &irq_flags);
1021         if (!event)
1022                 return;
1023         entry   = ring_buffer_event_data(event);
1024         tracing_generic_entry_update(&entry->ent, flags, pc);
1025         entry->ent.type         = TRACE_USER_STACK;
1026
1027         memset(&entry->caller, 0, sizeof(entry->caller));
1028
1029         trace.nr_entries        = 0;
1030         trace.max_entries       = FTRACE_STACK_ENTRIES;
1031         trace.skip              = 0;
1032         trace.entries           = entry->caller;
1033
1034         save_stack_trace_user(&trace);
1035         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1036 #endif
1037 }
1038
1039 void __trace_userstack(struct trace_array *tr,
1040                    struct trace_array_cpu *data,
1041                    unsigned long flags)
1042 {
1043         ftrace_trace_userstack(tr, data, flags, preempt_count());
1044 }
1045
1046 static void
1047 ftrace_trace_special(void *__tr, void *__data,
1048                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1049                      int pc)
1050 {
1051         struct ring_buffer_event *event;
1052         struct trace_array_cpu *data = __data;
1053         struct trace_array *tr = __tr;
1054         struct special_entry *entry;
1055         unsigned long irq_flags;
1056
1057         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1058                                          &irq_flags);
1059         if (!event)
1060                 return;
1061         entry   = ring_buffer_event_data(event);
1062         tracing_generic_entry_update(&entry->ent, 0, pc);
1063         entry->ent.type                 = TRACE_SPECIAL;
1064         entry->arg1                     = arg1;
1065         entry->arg2                     = arg2;
1066         entry->arg3                     = arg3;
1067         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1068         ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1069         ftrace_trace_userstack(tr, data, irq_flags, pc);
1070
1071         trace_wake_up();
1072 }
1073
1074 void
1075 __trace_special(void *__tr, void *__data,
1076                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1077 {
1078         ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1079 }
1080
1081 void
1082 tracing_sched_switch_trace(struct trace_array *tr,
1083                            struct trace_array_cpu *data,
1084                            struct task_struct *prev,
1085                            struct task_struct *next,
1086                            unsigned long flags, int pc)
1087 {
1088         struct ring_buffer_event *event;
1089         struct ctx_switch_entry *entry;
1090         unsigned long irq_flags;
1091
1092         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1093                                            &irq_flags);
1094         if (!event)
1095                 return;
1096         entry   = ring_buffer_event_data(event);
1097         tracing_generic_entry_update(&entry->ent, flags, pc);
1098         entry->ent.type                 = TRACE_CTX;
1099         entry->prev_pid                 = prev->pid;
1100         entry->prev_prio                = prev->prio;
1101         entry->prev_state               = prev->state;
1102         entry->next_pid                 = next->pid;
1103         entry->next_prio                = next->prio;
1104         entry->next_state               = next->state;
1105         entry->next_cpu = task_cpu(next);
1106         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1107         ftrace_trace_stack(tr, data, flags, 5, pc);
1108         ftrace_trace_userstack(tr, data, flags, pc);
1109 }
1110
1111 void
1112 tracing_sched_wakeup_trace(struct trace_array *tr,
1113                            struct trace_array_cpu *data,
1114                            struct task_struct *wakee,
1115                            struct task_struct *curr,
1116                            unsigned long flags, int pc)
1117 {
1118         struct ring_buffer_event *event;
1119         struct ctx_switch_entry *entry;
1120         unsigned long irq_flags;
1121
1122         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1123                                            &irq_flags);
1124         if (!event)
1125                 return;
1126         entry   = ring_buffer_event_data(event);
1127         tracing_generic_entry_update(&entry->ent, flags, pc);
1128         entry->ent.type                 = TRACE_WAKE;
1129         entry->prev_pid                 = curr->pid;
1130         entry->prev_prio                = curr->prio;
1131         entry->prev_state               = curr->state;
1132         entry->next_pid                 = wakee->pid;
1133         entry->next_prio                = wakee->prio;
1134         entry->next_state               = wakee->state;
1135         entry->next_cpu                 = task_cpu(wakee);
1136         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1137         ftrace_trace_stack(tr, data, flags, 6, pc);
1138         ftrace_trace_userstack(tr, data, flags, pc);
1139
1140         trace_wake_up();
1141 }
1142
1143 void
1144 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1145 {
1146         struct trace_array *tr = &global_trace;
1147         struct trace_array_cpu *data;
1148         unsigned long flags;
1149         int cpu;
1150         int pc;
1151
1152         if (tracing_disabled)
1153                 return;
1154
1155         pc = preempt_count();
1156         local_irq_save(flags);
1157         cpu = raw_smp_processor_id();
1158         data = tr->data[cpu];
1159
1160         if (likely(atomic_inc_return(&data->disabled) == 1))
1161                 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1162
1163         atomic_dec(&data->disabled);
1164         local_irq_restore(flags);
1165 }
1166
1167 #ifdef CONFIG_FUNCTION_TRACER
1168 static void
1169 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1170 {
1171         struct trace_array *tr = &global_trace;
1172         struct trace_array_cpu *data;
1173         unsigned long flags;
1174         long disabled;
1175         int cpu, resched;
1176         int pc;
1177
1178         if (unlikely(!ftrace_function_enabled))
1179                 return;
1180
1181         pc = preempt_count();
1182         resched = ftrace_preempt_disable();
1183         local_save_flags(flags);
1184         cpu = raw_smp_processor_id();
1185         data = tr->data[cpu];
1186         disabled = atomic_inc_return(&data->disabled);
1187
1188         if (likely(disabled == 1))
1189                 trace_function(tr, data, ip, parent_ip, flags, pc);
1190
1191         atomic_dec(&data->disabled);
1192         ftrace_preempt_enable(resched);
1193 }
1194
1195 static void
1196 function_trace_call(unsigned long ip, unsigned long parent_ip)
1197 {
1198         struct trace_array *tr = &global_trace;
1199         struct trace_array_cpu *data;
1200         unsigned long flags;
1201         long disabled;
1202         int cpu;
1203         int pc;
1204
1205         if (unlikely(!ftrace_function_enabled))
1206                 return;
1207
1208         /*
1209          * Need to use raw, since this must be called before the
1210          * recursive protection is performed.
1211          */
1212         local_irq_save(flags);
1213         cpu = raw_smp_processor_id();
1214         data = tr->data[cpu];
1215         disabled = atomic_inc_return(&data->disabled);
1216
1217         if (likely(disabled == 1)) {
1218                 pc = preempt_count();
1219                 trace_function(tr, data, ip, parent_ip, flags, pc);
1220         }
1221
1222         atomic_dec(&data->disabled);
1223         local_irq_restore(flags);
1224 }
1225
1226 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1227 int trace_graph_entry(struct ftrace_graph_ent *trace)
1228 {
1229         struct trace_array *tr = &global_trace;
1230         struct trace_array_cpu *data;
1231         unsigned long flags;
1232         long disabled;
1233         int cpu;
1234         int pc;
1235
1236         if (!ftrace_trace_task(current))
1237                 return 0;
1238
1239         if (!ftrace_graph_addr(trace->func))
1240                 return 0;
1241
1242         local_irq_save(flags);
1243         cpu = raw_smp_processor_id();
1244         data = tr->data[cpu];
1245         disabled = atomic_inc_return(&data->disabled);
1246         if (likely(disabled == 1)) {
1247                 pc = preempt_count();
1248                 __trace_graph_entry(tr, data, trace, flags, pc);
1249         }
1250         /* Only do the atomic if it is not already set */
1251         if (!test_tsk_trace_graph(current))
1252                 set_tsk_trace_graph(current);
1253         atomic_dec(&data->disabled);
1254         local_irq_restore(flags);
1255
1256         return 1;
1257 }
1258
1259 void trace_graph_return(struct ftrace_graph_ret *trace)
1260 {
1261         struct trace_array *tr = &global_trace;
1262         struct trace_array_cpu *data;
1263         unsigned long flags;
1264         long disabled;
1265         int cpu;
1266         int pc;
1267
1268         local_irq_save(flags);
1269         cpu = raw_smp_processor_id();
1270         data = tr->data[cpu];
1271         disabled = atomic_inc_return(&data->disabled);
1272         if (likely(disabled == 1)) {
1273                 pc = preempt_count();
1274                 __trace_graph_return(tr, data, trace, flags, pc);
1275         }
1276         if (!trace->depth)
1277                 clear_tsk_trace_graph(current);
1278         atomic_dec(&data->disabled);
1279         local_irq_restore(flags);
1280 }
1281 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1282
1283 static struct ftrace_ops trace_ops __read_mostly =
1284 {
1285         .func = function_trace_call,
1286 };
1287
1288 void tracing_start_function_trace(void)
1289 {
1290         ftrace_function_enabled = 0;
1291
1292         if (trace_flags & TRACE_ITER_PREEMPTONLY)
1293                 trace_ops.func = function_trace_call_preempt_only;
1294         else
1295                 trace_ops.func = function_trace_call;
1296
1297         register_ftrace_function(&trace_ops);
1298         ftrace_function_enabled = 1;
1299 }
1300
1301 void tracing_stop_function_trace(void)
1302 {
1303         ftrace_function_enabled = 0;
1304         unregister_ftrace_function(&trace_ops);
1305 }
1306 #endif
1307
1308 enum trace_file_type {
1309         TRACE_FILE_LAT_FMT      = 1,
1310         TRACE_FILE_ANNOTATE     = 2,
1311 };
1312
1313 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1314 {
1315         /* Don't allow ftrace to trace into the ring buffers */
1316         ftrace_disable_cpu();
1317
1318         iter->idx++;
1319         if (iter->buffer_iter[iter->cpu])
1320                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1321
1322         ftrace_enable_cpu();
1323 }
1324
1325 static struct trace_entry *
1326 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1327 {
1328         struct ring_buffer_event *event;
1329         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1330
1331         /* Don't allow ftrace to trace into the ring buffers */
1332         ftrace_disable_cpu();
1333
1334         if (buf_iter)
1335                 event = ring_buffer_iter_peek(buf_iter, ts);
1336         else
1337                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1338
1339         ftrace_enable_cpu();
1340
1341         return event ? ring_buffer_event_data(event) : NULL;
1342 }
1343
1344 static struct trace_entry *
1345 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1346 {
1347         struct ring_buffer *buffer = iter->tr->buffer;
1348         struct trace_entry *ent, *next = NULL;
1349         u64 next_ts = 0, ts;
1350         int next_cpu = -1;
1351         int cpu;
1352
1353         for_each_tracing_cpu(cpu) {
1354
1355                 if (ring_buffer_empty_cpu(buffer, cpu))
1356                         continue;
1357
1358                 ent = peek_next_entry(iter, cpu, &ts);
1359
1360                 /*
1361                  * Pick the entry with the smallest timestamp:
1362                  */
1363                 if (ent && (!next || ts < next_ts)) {
1364                         next = ent;
1365                         next_cpu = cpu;
1366                         next_ts = ts;
1367                 }
1368         }
1369
1370         if (ent_cpu)
1371                 *ent_cpu = next_cpu;
1372
1373         if (ent_ts)
1374                 *ent_ts = next_ts;
1375
1376         return next;
1377 }
1378
1379 /* Find the next real entry, without updating the iterator itself */
1380 static struct trace_entry *
1381 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1382 {
1383         return __find_next_entry(iter, ent_cpu, ent_ts);
1384 }
1385
1386 /* Find the next real entry, and increment the iterator to the next entry */
1387 static void *find_next_entry_inc(struct trace_iterator *iter)
1388 {
1389         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1390
1391         if (iter->ent)
1392                 trace_iterator_increment(iter, iter->cpu);
1393
1394         return iter->ent ? iter : NULL;
1395 }
1396
1397 static void trace_consume(struct trace_iterator *iter)
1398 {
1399         /* Don't allow ftrace to trace into the ring buffers */
1400         ftrace_disable_cpu();
1401         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1402         ftrace_enable_cpu();
1403 }
1404
1405 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1406 {
1407         struct trace_iterator *iter = m->private;
1408         int i = (int)*pos;
1409         void *ent;
1410
1411         (*pos)++;
1412
1413         /* can't go backwards */
1414         if (iter->idx > i)
1415                 return NULL;
1416
1417         if (iter->idx < 0)
1418                 ent = find_next_entry_inc(iter);
1419         else
1420                 ent = iter;
1421
1422         while (ent && iter->idx < i)
1423                 ent = find_next_entry_inc(iter);
1424
1425         iter->pos = *pos;
1426
1427         return ent;
1428 }
1429
1430 static void *s_start(struct seq_file *m, loff_t *pos)
1431 {
1432         struct trace_iterator *iter = m->private;
1433         void *p = NULL;
1434         loff_t l = 0;
1435         int cpu;
1436
1437         mutex_lock(&trace_types_lock);
1438
1439         if (!current_trace || current_trace != iter->trace) {
1440                 mutex_unlock(&trace_types_lock);
1441                 return NULL;
1442         }
1443
1444         atomic_inc(&trace_record_cmdline_disabled);
1445
1446         if (*pos != iter->pos) {
1447                 iter->ent = NULL;
1448                 iter->cpu = 0;
1449                 iter->idx = -1;
1450
1451                 ftrace_disable_cpu();
1452
1453                 for_each_tracing_cpu(cpu) {
1454                         ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1455                 }
1456
1457                 ftrace_enable_cpu();
1458
1459                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1460                         ;
1461
1462         } else {
1463                 l = *pos - 1;
1464                 p = s_next(m, p, &l);
1465         }
1466
1467         return p;
1468 }
1469
1470 static void s_stop(struct seq_file *m, void *p)
1471 {
1472         atomic_dec(&trace_record_cmdline_disabled);
1473         mutex_unlock(&trace_types_lock);
1474 }
1475
1476 #ifdef CONFIG_KRETPROBES
1477 static inline const char *kretprobed(const char *name)
1478 {
1479         static const char tramp_name[] = "kretprobe_trampoline";
1480         int size = sizeof(tramp_name);
1481
1482         if (strncmp(tramp_name, name, size) == 0)
1483                 return "[unknown/kretprobe'd]";
1484         return name;
1485 }
1486 #else
1487 static inline const char *kretprobed(const char *name)
1488 {
1489         return name;
1490 }
1491 #endif /* CONFIG_KRETPROBES */
1492
1493 static int
1494 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1495 {
1496 #ifdef CONFIG_KALLSYMS
1497         char str[KSYM_SYMBOL_LEN];
1498         const char *name;
1499
1500         kallsyms_lookup(address, NULL, NULL, NULL, str);
1501
1502         name = kretprobed(str);
1503
1504         return trace_seq_printf(s, fmt, name);
1505 #endif
1506         return 1;
1507 }
1508
1509 static int
1510 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1511                      unsigned long address)
1512 {
1513 #ifdef CONFIG_KALLSYMS
1514         char str[KSYM_SYMBOL_LEN];
1515         const char *name;
1516
1517         sprint_symbol(str, address);
1518         name = kretprobed(str);
1519
1520         return trace_seq_printf(s, fmt, name);
1521 #endif
1522         return 1;
1523 }
1524
1525 #ifndef CONFIG_64BIT
1526 # define IP_FMT "%08lx"
1527 #else
1528 # define IP_FMT "%016lx"
1529 #endif
1530
1531 int
1532 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1533 {
1534         int ret;
1535
1536         if (!ip)
1537                 return trace_seq_printf(s, "0");
1538
1539         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1540                 ret = seq_print_sym_offset(s, "%s", ip);
1541         else
1542                 ret = seq_print_sym_short(s, "%s", ip);
1543
1544         if (!ret)
1545                 return 0;
1546
1547         if (sym_flags & TRACE_ITER_SYM_ADDR)
1548                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1549         return ret;
1550 }
1551
1552 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1553                                     unsigned long ip, unsigned long sym_flags)
1554 {
1555         struct file *file = NULL;
1556         unsigned long vmstart = 0;
1557         int ret = 1;
1558
1559         if (mm) {
1560                 const struct vm_area_struct *vma;
1561
1562                 down_read(&mm->mmap_sem);
1563                 vma = find_vma(mm, ip);
1564                 if (vma) {
1565                         file = vma->vm_file;
1566                         vmstart = vma->vm_start;
1567                 }
1568                 if (file) {
1569                         ret = trace_seq_path(s, &file->f_path);
1570                         if (ret)
1571                                 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1572                 }
1573                 up_read(&mm->mmap_sem);
1574         }
1575         if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1576                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1577         return ret;
1578 }
1579
1580 static int
1581 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1582                       unsigned long sym_flags)
1583 {
1584         struct mm_struct *mm = NULL;
1585         int ret = 1;
1586         unsigned int i;
1587
1588         if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1589                 struct task_struct *task;
1590                 /*
1591                  * we do the lookup on the thread group leader,
1592                  * since individual threads might have already quit!
1593                  */
1594                 rcu_read_lock();
1595                 task = find_task_by_vpid(entry->ent.tgid);
1596                 if (task)
1597                         mm = get_task_mm(task);
1598                 rcu_read_unlock();
1599         }
1600
1601         for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1602                 unsigned long ip = entry->caller[i];
1603
1604                 if (ip == ULONG_MAX || !ret)
1605                         break;
1606                 if (i && ret)
1607                         ret = trace_seq_puts(s, " <- ");
1608                 if (!ip) {
1609                         if (ret)
1610                                 ret = trace_seq_puts(s, "??");
1611                         continue;
1612                 }
1613                 if (!ret)
1614                         break;
1615                 if (ret)
1616                         ret = seq_print_user_ip(s, mm, ip, sym_flags);
1617         }
1618
1619         if (mm)
1620                 mmput(mm);
1621         return ret;
1622 }
1623
1624 static void print_lat_help_header(struct seq_file *m)
1625 {
1626         seq_puts(m, "#                  _------=> CPU#            \n");
1627         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1628         seq_puts(m, "#                | / _----=> need-resched    \n");
1629         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1630         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1631         seq_puts(m, "#                |||| /                      \n");
1632         seq_puts(m, "#                |||||     delay             \n");
1633         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1634         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1635 }
1636
1637 static void print_func_help_header(struct seq_file *m)
1638 {
1639         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1640         seq_puts(m, "#              | |       |          |         |\n");
1641 }
1642
1643
1644 static void
1645 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1646 {
1647         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1648         struct trace_array *tr = iter->tr;
1649         struct trace_array_cpu *data = tr->data[tr->cpu];
1650         struct tracer *type = current_trace;
1651         unsigned long total;
1652         unsigned long entries;
1653         const char *name = "preemption";
1654
1655         if (type)
1656                 name = type->name;
1657
1658         entries = ring_buffer_entries(iter->tr->buffer);
1659         total = entries +
1660                 ring_buffer_overruns(iter->tr->buffer);
1661
1662         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1663                    name, UTS_RELEASE);
1664         seq_puts(m, "-----------------------------------"
1665                  "---------------------------------\n");
1666         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1667                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1668                    nsecs_to_usecs(data->saved_latency),
1669                    entries,
1670                    total,
1671                    tr->cpu,
1672 #if defined(CONFIG_PREEMPT_NONE)
1673                    "server",
1674 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1675                    "desktop",
1676 #elif defined(CONFIG_PREEMPT)
1677                    "preempt",
1678 #else
1679                    "unknown",
1680 #endif
1681                    /* These are reserved for later use */
1682                    0, 0, 0, 0);
1683 #ifdef CONFIG_SMP
1684         seq_printf(m, " #P:%d)\n", num_online_cpus());
1685 #else
1686         seq_puts(m, ")\n");
1687 #endif
1688         seq_puts(m, "    -----------------\n");
1689         seq_printf(m, "    | task: %.16s-%d "
1690                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1691                    data->comm, data->pid, data->uid, data->nice,
1692                    data->policy, data->rt_priority);
1693         seq_puts(m, "    -----------------\n");
1694
1695         if (data->critical_start) {
1696                 seq_puts(m, " => started at: ");
1697                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1698                 trace_print_seq(m, &iter->seq);
1699                 seq_puts(m, "\n => ended at:   ");
1700                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1701                 trace_print_seq(m, &iter->seq);
1702                 seq_puts(m, "\n");
1703         }
1704
1705         seq_puts(m, "\n");
1706 }
1707
1708 static void
1709 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1710 {
1711         int hardirq, softirq;
1712         char *comm;
1713
1714         comm = trace_find_cmdline(entry->pid);
1715
1716         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1717         trace_seq_printf(s, "%3d", cpu);
1718         trace_seq_printf(s, "%c%c",
1719                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1720                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1721                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1722
1723         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1724         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1725         if (hardirq && softirq) {
1726                 trace_seq_putc(s, 'H');
1727         } else {
1728                 if (hardirq) {
1729                         trace_seq_putc(s, 'h');
1730                 } else {
1731                         if (softirq)
1732                                 trace_seq_putc(s, 's');
1733                         else
1734                                 trace_seq_putc(s, '.');
1735                 }
1736         }
1737
1738         if (entry->preempt_count)
1739                 trace_seq_printf(s, "%x", entry->preempt_count);
1740         else
1741                 trace_seq_puts(s, ".");
1742 }
1743
1744 unsigned long preempt_mark_thresh = 100;
1745
1746 static void
1747 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1748                     unsigned long rel_usecs)
1749 {
1750         trace_seq_printf(s, " %4lldus", abs_usecs);
1751         if (rel_usecs > preempt_mark_thresh)
1752                 trace_seq_puts(s, "!: ");
1753         else if (rel_usecs > 1)
1754                 trace_seq_puts(s, "+: ");
1755         else
1756                 trace_seq_puts(s, " : ");
1757 }
1758
1759 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1760
1761 static int task_state_char(unsigned long state)
1762 {
1763         int bit = state ? __ffs(state) + 1 : 0;
1764
1765         return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?';
1766 }
1767
1768 /*
1769  * The message is supposed to contain an ending newline.
1770  * If the printing stops prematurely, try to add a newline of our own.
1771  */
1772 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1773 {
1774         struct trace_entry *ent;
1775         struct trace_field_cont *cont;
1776         bool ok = true;
1777
1778         ent = peek_next_entry(iter, iter->cpu, NULL);
1779         if (!ent || ent->type != TRACE_CONT) {
1780                 trace_seq_putc(s, '\n');
1781                 return;
1782         }
1783
1784         do {
1785                 cont = (struct trace_field_cont *)ent;
1786                 if (ok)
1787                         ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1788
1789                 ftrace_disable_cpu();
1790
1791                 if (iter->buffer_iter[iter->cpu])
1792                         ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1793                 else
1794                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1795
1796                 ftrace_enable_cpu();
1797
1798                 ent = peek_next_entry(iter, iter->cpu, NULL);
1799         } while (ent && ent->type == TRACE_CONT);
1800
1801         if (!ok)
1802                 trace_seq_putc(s, '\n');
1803 }
1804
1805 static void test_cpu_buff_start(struct trace_iterator *iter)
1806 {
1807         struct trace_seq *s = &iter->seq;
1808
1809         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1810                 return;
1811
1812         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1813                 return;
1814
1815         if (cpu_isset(iter->cpu, iter->started))
1816                 return;
1817
1818         cpu_set(iter->cpu, iter->started);
1819         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1820 }
1821
1822 static enum print_line_t
1823 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1824 {
1825         struct trace_seq *s = &iter->seq;
1826         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1827         struct trace_entry *next_entry;
1828         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1829         struct trace_entry *entry = iter->ent;
1830         unsigned long abs_usecs;
1831         unsigned long rel_usecs;
1832         u64 next_ts;
1833         char *comm;
1834         int S, T;
1835         int i;
1836
1837         if (entry->type == TRACE_CONT)
1838                 return TRACE_TYPE_HANDLED;
1839
1840         test_cpu_buff_start(iter);
1841
1842         next_entry = find_next_entry(iter, NULL, &next_ts);
1843         if (!next_entry)
1844                 next_ts = iter->ts;
1845         rel_usecs = ns2usecs(next_ts - iter->ts);
1846         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1847
1848         if (verbose) {
1849                 comm = trace_find_cmdline(entry->pid);
1850                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1851                                  " %ld.%03ldms (+%ld.%03ldms): ",
1852                                  comm,
1853                                  entry->pid, cpu, entry->flags,
1854                                  entry->preempt_count, trace_idx,
1855                                  ns2usecs(iter->ts),
1856                                  abs_usecs/1000,
1857                                  abs_usecs % 1000, rel_usecs/1000,
1858                                  rel_usecs % 1000);
1859         } else {
1860                 lat_print_generic(s, entry, cpu);
1861                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1862         }
1863         switch (entry->type) {
1864         case TRACE_FN: {
1865                 struct ftrace_entry *field;
1866
1867                 trace_assign_type(field, entry);
1868
1869                 seq_print_ip_sym(s, field->ip, sym_flags);
1870                 trace_seq_puts(s, " (");
1871                 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1872                 trace_seq_puts(s, ")\n");
1873                 break;
1874         }
1875         case TRACE_CTX:
1876         case TRACE_WAKE: {
1877                 struct ctx_switch_entry *field;
1878
1879                 trace_assign_type(field, entry);
1880
1881                 T = task_state_char(field->next_state);
1882                 S = task_state_char(field->prev_state);
1883                 comm = trace_find_cmdline(field->next_pid);
1884                 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1885                                  field->prev_pid,
1886                                  field->prev_prio,
1887                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1888                                  field->next_cpu,
1889                                  field->next_pid,
1890                                  field->next_prio,
1891                                  T, comm);
1892                 break;
1893         }
1894         case TRACE_SPECIAL: {
1895                 struct special_entry *field;
1896
1897                 trace_assign_type(field, entry);
1898
1899                 trace_seq_printf(s, "# %ld %ld %ld\n",
1900                                  field->arg1,
1901                                  field->arg2,
1902                                  field->arg3);
1903                 break;
1904         }
1905         case TRACE_STACK: {
1906                 struct stack_entry *field;
1907
1908                 trace_assign_type(field, entry);
1909
1910                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1911                         if (i)
1912                                 trace_seq_puts(s, " <= ");
1913                         seq_print_ip_sym(s, field->caller[i], sym_flags);
1914                 }
1915                 trace_seq_puts(s, "\n");
1916                 break;
1917         }
1918         case TRACE_PRINT: {
1919                 struct print_entry *field;
1920
1921                 trace_assign_type(field, entry);
1922
1923                 seq_print_ip_sym(s, field->ip, sym_flags);
1924                 trace_seq_printf(s, ": %s", field->buf);
1925                 if (entry->flags & TRACE_FLAG_CONT)
1926                         trace_seq_print_cont(s, iter);
1927                 break;
1928         }
1929         case TRACE_BRANCH: {
1930                 struct trace_branch *field;
1931
1932                 trace_assign_type(field, entry);
1933
1934                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1935                                  field->correct ? "  ok  " : " MISS ",
1936                                  field->func,
1937                                  field->file,
1938                                  field->line);
1939                 break;
1940         }
1941         case TRACE_USER_STACK: {
1942                 struct userstack_entry *field;
1943
1944                 trace_assign_type(field, entry);
1945
1946                 seq_print_userip_objs(field, s, sym_flags);
1947                 trace_seq_putc(s, '\n');
1948                 break;
1949         }
1950         default:
1951                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1952         }
1953         return TRACE_TYPE_HANDLED;
1954 }
1955
1956 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1957 {
1958         struct trace_seq *s = &iter->seq;
1959         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1960         struct trace_entry *entry;
1961         unsigned long usec_rem;
1962         unsigned long long t;
1963         unsigned long secs;
1964         char *comm;
1965         int ret;
1966         int S, T;
1967         int i;
1968
1969         entry = iter->ent;
1970
1971         if (entry->type == TRACE_CONT)
1972                 return TRACE_TYPE_HANDLED;
1973
1974         test_cpu_buff_start(iter);
1975
1976         comm = trace_find_cmdline(iter->ent->pid);
1977
1978         t = ns2usecs(iter->ts);
1979         usec_rem = do_div(t, 1000000ULL);
1980         secs = (unsigned long)t;
1981
1982         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1983         if (!ret)
1984                 return TRACE_TYPE_PARTIAL_LINE;
1985         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1986         if (!ret)
1987                 return TRACE_TYPE_PARTIAL_LINE;
1988         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1989         if (!ret)
1990                 return TRACE_TYPE_PARTIAL_LINE;
1991
1992         switch (entry->type) {
1993         case TRACE_FN: {
1994                 struct ftrace_entry *field;
1995
1996                 trace_assign_type(field, entry);
1997
1998                 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1999                 if (!ret)
2000                         return TRACE_TYPE_PARTIAL_LINE;
2001                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
2002                                                 field->parent_ip) {
2003                         ret = trace_seq_printf(s, " <-");
2004                         if (!ret)
2005                                 return TRACE_TYPE_PARTIAL_LINE;
2006                         ret = seq_print_ip_sym(s,
2007                                                field->parent_ip,
2008                                                sym_flags);
2009                         if (!ret)
2010                                 return TRACE_TYPE_PARTIAL_LINE;
2011                 }
2012                 ret = trace_seq_printf(s, "\n");
2013                 if (!ret)
2014                         return TRACE_TYPE_PARTIAL_LINE;
2015                 break;
2016         }
2017         case TRACE_CTX:
2018         case TRACE_WAKE: {
2019                 struct ctx_switch_entry *field;
2020
2021                 trace_assign_type(field, entry);
2022
2023                 T = task_state_char(field->next_state);
2024                 S = task_state_char(field->prev_state);
2025                 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
2026                                        field->prev_pid,
2027                                        field->prev_prio,
2028                                        S,
2029                                        entry->type == TRACE_CTX ? "==>" : "  +",
2030                                        field->next_cpu,
2031                                        field->next_pid,
2032                                        field->next_prio,
2033                                        T);
2034                 if (!ret)
2035                         return TRACE_TYPE_PARTIAL_LINE;
2036                 break;
2037         }
2038         case TRACE_SPECIAL: {
2039                 struct special_entry *field;
2040
2041                 trace_assign_type(field, entry);
2042
2043                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2044                                  field->arg1,
2045                                  field->arg2,
2046                                  field->arg3);
2047                 if (!ret)
2048                         return TRACE_TYPE_PARTIAL_LINE;
2049                 break;
2050         }
2051         case TRACE_STACK: {
2052                 struct stack_entry *field;
2053
2054                 trace_assign_type(field, entry);
2055
2056                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2057                         if (i) {
2058                                 ret = trace_seq_puts(s, " <= ");
2059                                 if (!ret)
2060                                         return TRACE_TYPE_PARTIAL_LINE;
2061                         }
2062                         ret = seq_print_ip_sym(s, field->caller[i],
2063                                                sym_flags);
2064                         if (!ret)
2065                                 return TRACE_TYPE_PARTIAL_LINE;
2066                 }
2067                 ret = trace_seq_puts(s, "\n");
2068                 if (!ret)
2069                         return TRACE_TYPE_PARTIAL_LINE;
2070                 break;
2071         }
2072         case TRACE_PRINT: {
2073                 struct print_entry *field;
2074
2075                 trace_assign_type(field, entry);
2076
2077                 seq_print_ip_sym(s, field->ip, sym_flags);
2078                 trace_seq_printf(s, ": %s", field->buf);
2079                 if (entry->flags & TRACE_FLAG_CONT)
2080                         trace_seq_print_cont(s, iter);
2081                 break;
2082         }
2083         case TRACE_GRAPH_RET: {
2084                 return print_graph_function(iter);
2085         }
2086         case TRACE_GRAPH_ENT: {
2087                 return print_graph_function(iter);
2088         }
2089         case TRACE_BRANCH: {
2090                 struct trace_branch *field;
2091
2092                 trace_assign_type(field, entry);
2093
2094                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2095                                  field->correct ? "  ok  " : " MISS ",
2096                                  field->func,
2097                                  field->file,
2098                                  field->line);
2099                 break;
2100         }
2101         case TRACE_USER_STACK: {
2102                 struct userstack_entry *field;
2103
2104                 trace_assign_type(field, entry);
2105
2106                 ret = seq_print_userip_objs(field, s, sym_flags);
2107                 if (!ret)
2108                         return TRACE_TYPE_PARTIAL_LINE;
2109                 ret = trace_seq_putc(s, '\n');
2110                 if (!ret)
2111                         return TRACE_TYPE_PARTIAL_LINE;
2112                 break;
2113         }
2114         }
2115         return TRACE_TYPE_HANDLED;
2116 }
2117
2118 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2119 {
2120         struct trace_seq *s = &iter->seq;
2121         struct trace_entry *entry;
2122         int ret;
2123         int S, T;
2124
2125         entry = iter->ent;
2126
2127         if (entry->type == TRACE_CONT)
2128                 return TRACE_TYPE_HANDLED;
2129
2130         ret = trace_seq_printf(s, "%d %d %llu ",
2131                 entry->pid, iter->cpu, iter->ts);
2132         if (!ret)
2133                 return TRACE_TYPE_PARTIAL_LINE;
2134
2135         switch (entry->type) {
2136         case TRACE_FN: {
2137                 struct ftrace_entry *field;
2138
2139                 trace_assign_type(field, entry);
2140
2141                 ret = trace_seq_printf(s, "%x %x\n",
2142                                         field->ip,
2143                                         field->parent_ip);
2144                 if (!ret)
2145                         return TRACE_TYPE_PARTIAL_LINE;
2146                 break;
2147         }
2148         case TRACE_CTX:
2149         case TRACE_WAKE: {
2150                 struct ctx_switch_entry *field;
2151
2152                 trace_assign_type(field, entry);
2153
2154                 T = task_state_char(field->next_state);
2155                 S = entry->type == TRACE_WAKE ? '+' :
2156                         task_state_char(field->prev_state);
2157                 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2158                                        field->prev_pid,
2159                                        field->prev_prio,
2160                                        S,
2161                                        field->next_cpu,
2162                                        field->next_pid,
2163                                        field->next_prio,
2164                                        T);
2165                 if (!ret)
2166                         return TRACE_TYPE_PARTIAL_LINE;
2167                 break;
2168         }
2169         case TRACE_SPECIAL:
2170         case TRACE_USER_STACK:
2171         case TRACE_STACK: {
2172                 struct special_entry *field;
2173
2174                 trace_assign_type(field, entry);
2175
2176                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2177                                  field->arg1,
2178                                  field->arg2,
2179                                  field->arg3);
2180                 if (!ret)
2181                         return TRACE_TYPE_PARTIAL_LINE;
2182                 break;
2183         }
2184         case TRACE_PRINT: {
2185                 struct print_entry *field;
2186
2187                 trace_assign_type(field, entry);
2188
2189                 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2190                 if (entry->flags & TRACE_FLAG_CONT)
2191                         trace_seq_print_cont(s, iter);
2192                 break;
2193         }
2194         }
2195         return TRACE_TYPE_HANDLED;
2196 }
2197
2198 #define SEQ_PUT_FIELD_RET(s, x)                         \
2199 do {                                                    \
2200         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
2201                 return 0;                               \
2202 } while (0)
2203
2204 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
2205 do {                                                    \
2206         BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES);     \
2207         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
2208                 return 0;                               \
2209 } while (0)
2210
2211 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2212 {
2213         struct trace_seq *s = &iter->seq;
2214         unsigned char newline = '\n';
2215         struct trace_entry *entry;
2216         int S, T;
2217
2218         entry = iter->ent;
2219
2220         if (entry->type == TRACE_CONT)
2221                 return TRACE_TYPE_HANDLED;
2222
2223         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2224         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2225         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2226
2227         switch (entry->type) {
2228         case TRACE_FN: {
2229                 struct ftrace_entry *field;
2230
2231                 trace_assign_type(field, entry);
2232
2233                 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2234                 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2235                 break;
2236         }
2237         case TRACE_CTX:
2238         case TRACE_WAKE: {
2239                 struct ctx_switch_entry *field;
2240
2241                 trace_assign_type(field, entry);
2242
2243                 T = task_state_char(field->next_state);
2244                 S = entry->type == TRACE_WAKE ? '+' :
2245                         task_state_char(field->prev_state);
2246                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2247                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2248                 SEQ_PUT_HEX_FIELD_RET(s, S);
2249                 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2250                 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2251                 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2252                 SEQ_PUT_HEX_FIELD_RET(s, T);
2253                 break;
2254         }
2255         case TRACE_SPECIAL:
2256         case TRACE_USER_STACK:
2257         case TRACE_STACK: {
2258                 struct special_entry *field;
2259
2260                 trace_assign_type(field, entry);
2261
2262                 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2263                 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2264                 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2265                 break;
2266         }
2267         }
2268         SEQ_PUT_FIELD_RET(s, newline);
2269
2270         return TRACE_TYPE_HANDLED;
2271 }
2272
2273 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
2274 {
2275         struct trace_seq *s = &iter->seq;
2276         struct trace_entry *entry = iter->ent;
2277         struct print_entry *field;
2278         int ret;
2279
2280         trace_assign_type(field, entry);
2281
2282         ret = trace_seq_printf(s, field->buf);
2283         if (!ret)
2284                 return TRACE_TYPE_PARTIAL_LINE;
2285
2286         if (entry->flags & TRACE_FLAG_CONT)
2287                 trace_seq_print_cont(s, iter);
2288
2289         return TRACE_TYPE_HANDLED;
2290 }
2291
2292 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2293 {
2294         struct trace_seq *s = &iter->seq;
2295         struct trace_entry *entry;
2296
2297         entry = iter->ent;
2298
2299         if (entry->type == TRACE_CONT)
2300                 return TRACE_TYPE_HANDLED;
2301
2302         SEQ_PUT_FIELD_RET(s, entry->pid);
2303         SEQ_PUT_FIELD_RET(s, entry->cpu);
2304         SEQ_PUT_FIELD_RET(s, iter->ts);
2305
2306         switch (entry->type) {
2307         case TRACE_FN: {
2308                 struct ftrace_entry *field;
2309
2310                 trace_assign_type(field, entry);
2311
2312                 SEQ_PUT_FIELD_RET(s, field->ip);
2313                 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2314                 break;
2315         }
2316         case TRACE_CTX: {
2317                 struct ctx_switch_entry *field;
2318
2319                 trace_assign_type(field, entry);
2320
2321                 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2322                 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2323                 SEQ_PUT_FIELD_RET(s, field->prev_state);
2324                 SEQ_PUT_FIELD_RET(s, field->next_pid);
2325                 SEQ_PUT_FIELD_RET(s, field->next_prio);
2326                 SEQ_PUT_FIELD_RET(s, field->next_state);
2327                 break;
2328         }
2329         case TRACE_SPECIAL:
2330         case TRACE_USER_STACK:
2331         case TRACE_STACK: {
2332                 struct special_entry *field;
2333
2334                 trace_assign_type(field, entry);
2335
2336                 SEQ_PUT_FIELD_RET(s, field->arg1);
2337                 SEQ_PUT_FIELD_RET(s, field->arg2);
2338                 SEQ_PUT_FIELD_RET(s, field->arg3);
2339                 break;
2340         }
2341         }
2342         return 1;
2343 }
2344
2345 static int trace_empty(struct trace_iterator *iter)
2346 {
2347         int cpu;
2348
2349         for_each_tracing_cpu(cpu) {
2350                 if (iter->buffer_iter[cpu]) {
2351                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2352                                 return 0;
2353                 } else {
2354                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2355                                 return 0;
2356                 }
2357         }
2358
2359         return 1;
2360 }
2361
2362 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2363 {
2364         enum print_line_t ret;
2365
2366         if (iter->trace && iter->trace->print_line) {
2367                 ret = iter->trace->print_line(iter);
2368                 if (ret != TRACE_TYPE_UNHANDLED)
2369                         return ret;
2370         }
2371
2372         if (iter->ent->type == TRACE_PRINT &&
2373                         trace_flags & TRACE_ITER_PRINTK &&
2374                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2375                 return print_printk_msg_only(iter);
2376
2377         if (trace_flags & TRACE_ITER_BIN)
2378                 return print_bin_fmt(iter);
2379
2380         if (trace_flags & TRACE_ITER_HEX)
2381                 return print_hex_fmt(iter);
2382
2383         if (trace_flags & TRACE_ITER_RAW)
2384                 return print_raw_fmt(iter);
2385
2386         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2387                 return print_lat_fmt(iter, iter->idx, iter->cpu);
2388
2389         return print_trace_fmt(iter);
2390 }
2391
2392 static int s_show(struct seq_file *m, void *v)
2393 {
2394         struct trace_iterator *iter = v;
2395
2396         if (iter->ent == NULL) {
2397                 if (iter->tr) {
2398                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2399                         seq_puts(m, "#\n");
2400                 }
2401                 if (iter->trace && iter->trace->print_header)
2402                         iter->trace->print_header(m);
2403                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2404                         /* print nothing if the buffers are empty */
2405                         if (trace_empty(iter))
2406                                 return 0;
2407                         print_trace_header(m, iter);
2408                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2409                                 print_lat_help_header(m);
2410                 } else {
2411                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2412                                 print_func_help_header(m);
2413                 }
2414         } else {
2415                 print_trace_line(iter);
2416                 trace_print_seq(m, &iter->seq);
2417         }
2418
2419         return 0;
2420 }
2421
2422 static struct seq_operations tracer_seq_ops = {
2423         .start          = s_start,
2424         .next           = s_next,
2425         .stop           = s_stop,
2426         .show           = s_show,
2427 };
2428
2429 static struct trace_iterator *
2430 __tracing_open(struct inode *inode, struct file *file, int *ret)
2431 {
2432         struct trace_iterator *iter;
2433         struct seq_file *m;
2434         int cpu;
2435
2436         if (tracing_disabled) {
2437                 *ret = -ENODEV;
2438                 return NULL;
2439         }
2440
2441         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2442         if (!iter) {
2443                 *ret = -ENOMEM;
2444                 goto out;
2445         }
2446
2447         mutex_lock(&trace_types_lock);
2448         if (current_trace && current_trace->print_max)
2449                 iter->tr = &max_tr;
2450         else
2451                 iter->tr = inode->i_private;
2452         iter->trace = current_trace;
2453         iter->pos = -1;
2454
2455         /* Notify the tracer early; before we stop tracing. */
2456         if (iter->trace && iter->trace->open)
2457                 iter->trace->open(iter);
2458
2459         /* Annotate start of buffers if we had overruns */
2460         if (ring_buffer_overruns(iter->tr->buffer))
2461                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2462
2463
2464         for_each_tracing_cpu(cpu) {
2465
2466                 iter->buffer_iter[cpu] =
2467                         ring_buffer_read_start(iter->tr->buffer, cpu);
2468
2469                 if (!iter->buffer_iter[cpu])
2470                         goto fail_buffer;
2471         }
2472
2473         /* TODO stop tracer */
2474         *ret = seq_open(file, &tracer_seq_ops);
2475         if (*ret)
2476                 goto fail_buffer;
2477
2478         m = file->private_data;
2479         m->private = iter;
2480
2481         /* stop the trace while dumping */
2482         tracing_stop();
2483
2484         mutex_unlock(&trace_types_lock);
2485
2486  out:
2487         return iter;
2488
2489  fail_buffer:
2490         for_each_tracing_cpu(cpu) {
2491                 if (iter->buffer_iter[cpu])
2492                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2493         }
2494         mutex_unlock(&trace_types_lock);
2495         kfree(iter);
2496
2497         return ERR_PTR(-ENOMEM);
2498 }
2499
2500 int tracing_open_generic(struct inode *inode, struct file *filp)
2501 {
2502         if (tracing_disabled)
2503                 return -ENODEV;
2504
2505         filp->private_data = inode->i_private;
2506         return 0;
2507 }
2508
2509 int tracing_release(struct inode *inode, struct file *file)
2510 {
2511         struct seq_file *m = (struct seq_file *)file->private_data;
2512         struct trace_iterator *iter = m->private;
2513         int cpu;
2514
2515         mutex_lock(&trace_types_lock);
2516         for_each_tracing_cpu(cpu) {
2517                 if (iter->buffer_iter[cpu])
2518                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2519         }
2520
2521         if (iter->trace && iter->trace->close)
2522                 iter->trace->close(iter);
2523
2524         /* reenable tracing if it was previously enabled */
2525         tracing_start();
2526         mutex_unlock(&trace_types_lock);
2527
2528         seq_release(inode, file);
2529         kfree(iter);
2530         return 0;
2531 }
2532
2533 static int tracing_open(struct inode *inode, struct file *file)
2534 {
2535         int ret;
2536
2537         __tracing_open(inode, file, &ret);
2538
2539         return ret;
2540 }
2541
2542 static int tracing_lt_open(struct inode *inode, struct file *file)
2543 {
2544         struct trace_iterator *iter;
2545         int ret;
2546
2547         iter = __tracing_open(inode, file, &ret);
2548
2549         if (!ret)
2550                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2551
2552         return ret;
2553 }
2554
2555
2556 static void *
2557 t_next(struct seq_file *m, void *v, loff_t *pos)
2558 {
2559         struct tracer *t = m->private;
2560
2561         (*pos)++;
2562
2563         if (t)
2564                 t = t->next;
2565
2566         m->private = t;
2567
2568         return t;
2569 }
2570
2571 static void *t_start(struct seq_file *m, loff_t *pos)
2572 {
2573         struct tracer *t = m->private;
2574         loff_t l = 0;
2575
2576         mutex_lock(&trace_types_lock);
2577         for (; t && l < *pos; t = t_next(m, t, &l))
2578                 ;
2579
2580         return t;
2581 }
2582
2583 static void t_stop(struct seq_file *m, void *p)
2584 {
2585         mutex_unlock(&trace_types_lock);
2586 }
2587
2588 static int t_show(struct seq_file *m, void *v)
2589 {
2590         struct tracer *t = v;
2591
2592         if (!t)
2593                 return 0;
2594
2595         seq_printf(m, "%s", t->name);
2596         if (t->next)
2597                 seq_putc(m, ' ');
2598         else
2599                 seq_putc(m, '\n');
2600
2601         return 0;
2602 }
2603
2604 static struct seq_operations show_traces_seq_ops = {
2605         .start          = t_start,
2606         .next           = t_next,
2607         .stop           = t_stop,
2608         .show           = t_show,
2609 };
2610
2611 static int show_traces_open(struct inode *inode, struct file *file)
2612 {
2613         int ret;
2614
2615         if (tracing_disabled)
2616                 return -ENODEV;
2617
2618         ret = seq_open(file, &show_traces_seq_ops);
2619         if (!ret) {
2620                 struct seq_file *m = file->private_data;
2621                 m->private = trace_types;
2622         }
2623
2624         return ret;
2625 }
2626
2627 static struct file_operations tracing_fops = {
2628         .open           = tracing_open,
2629         .read           = seq_read,
2630         .llseek         = seq_lseek,
2631         .release        = tracing_release,
2632 };
2633
2634 static struct file_operations tracing_lt_fops = {
2635         .open           = tracing_lt_open,
2636         .read           = seq_read,
2637         .llseek         = seq_lseek,
2638         .release        = tracing_release,
2639 };
2640
2641 static struct file_operations show_traces_fops = {
2642         .open           = show_traces_open,
2643         .read           = seq_read,
2644         .release        = seq_release,
2645 };
2646
2647 /*
2648  * Only trace on a CPU if the bitmask is set:
2649  */
2650 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2651
2652 /*
2653  * When tracing/tracing_cpu_mask is modified then this holds
2654  * the new bitmask we are about to install:
2655  */
2656 static cpumask_t tracing_cpumask_new;
2657
2658 /*
2659  * The tracer itself will not take this lock, but still we want
2660  * to provide a consistent cpumask to user-space:
2661  */
2662 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2663
2664 /*
2665  * Temporary storage for the character representation of the
2666  * CPU bitmask (and one more byte for the newline):
2667  */
2668 static char mask_str[NR_CPUS + 1];
2669
2670 static ssize_t
2671 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2672                      size_t count, loff_t *ppos)
2673 {
2674         int len;
2675
2676         mutex_lock(&tracing_cpumask_update_lock);
2677
2678         len = cpumask_scnprintf(mask_str, count, &tracing_cpumask);
2679         if (count - len < 2) {
2680                 count = -EINVAL;
2681                 goto out_err;
2682         }
2683         len += sprintf(mask_str + len, "\n");
2684         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2685
2686 out_err:
2687         mutex_unlock(&tracing_cpumask_update_lock);
2688
2689         return count;
2690 }
2691
2692 static ssize_t
2693 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2694                       size_t count, loff_t *ppos)
2695 {
2696         int err, cpu;
2697
2698         mutex_lock(&tracing_cpumask_update_lock);
2699         err = cpumask_parse_user(ubuf, count, &tracing_cpumask_new);
2700         if (err)
2701                 goto err_unlock;
2702
2703         local_irq_disable();
2704         __raw_spin_lock(&ftrace_max_lock);
2705         for_each_tracing_cpu(cpu) {
2706                 /*
2707                  * Increase/decrease the disabled counter if we are
2708                  * about to flip a bit in the cpumask:
2709                  */
2710                 if (cpu_isset(cpu, tracing_cpumask) &&
2711                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2712                         atomic_inc(&global_trace.data[cpu]->disabled);
2713                 }
2714                 if (!cpu_isset(cpu, tracing_cpumask) &&
2715                                 cpu_isset(cpu, tracing_cpumask_new)) {
2716                         atomic_dec(&global_trace.data[cpu]->disabled);
2717                 }
2718         }
2719         __raw_spin_unlock(&ftrace_max_lock);
2720         local_irq_enable();
2721
2722         tracing_cpumask = tracing_cpumask_new;
2723
2724         mutex_unlock(&tracing_cpumask_update_lock);
2725
2726         return count;
2727
2728 err_unlock:
2729         mutex_unlock(&tracing_cpumask_update_lock);
2730
2731         return err;
2732 }
2733
2734 static struct file_operations tracing_cpumask_fops = {
2735         .open           = tracing_open_generic,
2736         .read           = tracing_cpumask_read,
2737         .write          = tracing_cpumask_write,
2738 };
2739
2740 static ssize_t
2741 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2742                        size_t cnt, loff_t *ppos)
2743 {
2744         int i;
2745         char *buf;
2746         int r = 0;
2747         int len = 0;
2748         u32 tracer_flags = current_trace->flags->val;
2749         struct tracer_opt *trace_opts = current_trace->flags->opts;
2750
2751
2752         /* calulate max size */
2753         for (i = 0; trace_options[i]; i++) {
2754                 len += strlen(trace_options[i]);
2755                 len += 3; /* "no" and space */
2756         }
2757
2758         /*
2759          * Increase the size with names of options specific
2760          * of the current tracer.
2761          */
2762         for (i = 0; trace_opts[i].name; i++) {
2763                 len += strlen(trace_opts[i].name);
2764                 len += 3; /* "no" and space */
2765         }
2766
2767         /* +2 for \n and \0 */
2768         buf = kmalloc(len + 2, GFP_KERNEL);
2769         if (!buf)
2770                 return -ENOMEM;
2771
2772         for (i = 0; trace_options[i]; i++) {
2773                 if (trace_flags & (1 << i))
2774                         r += sprintf(buf + r, "%s ", trace_options[i]);
2775                 else
2776                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2777         }
2778
2779         for (i = 0; trace_opts[i].name; i++) {
2780                 if (tracer_flags & trace_opts[i].bit)
2781                         r += sprintf(buf + r, "%s ",
2782                                 trace_opts[i].name);
2783                 else
2784                         r += sprintf(buf + r, "no%s ",
2785                                 trace_opts[i].name);
2786         }
2787
2788         r += sprintf(buf + r, "\n");
2789         WARN_ON(r >= len + 2);
2790
2791         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2792
2793         kfree(buf);
2794
2795         return r;
2796 }
2797
2798 /* Try to assign a tracer specific option */
2799 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2800 {
2801         struct tracer_flags *trace_flags = trace->flags;
2802         struct tracer_opt *opts = NULL;
2803         int ret = 0, i = 0;
2804         int len;
2805
2806         for (i = 0; trace_flags->opts[i].name; i++) {
2807                 opts = &trace_flags->opts[i];
2808                 len = strlen(opts->name);
2809
2810                 if (strncmp(cmp, opts->name, len) == 0) {
2811                         ret = trace->set_flag(trace_flags->val,
2812                                 opts->bit, !neg);
2813                         break;
2814                 }
2815         }
2816         /* Not found */
2817         if (!trace_flags->opts[i].name)
2818                 return -EINVAL;
2819
2820         /* Refused to handle */
2821         if (ret)
2822                 return ret;
2823
2824         if (neg)
2825                 trace_flags->val &= ~opts->bit;
2826         else
2827                 trace_flags->val |= opts->bit;
2828
2829         return 0;
2830 }
2831
2832 static ssize_t
2833 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2834                         size_t cnt, loff_t *ppos)
2835 {
2836         char buf[64];
2837         char *cmp = buf;
2838         int neg = 0;
2839         int ret;
2840         int i;
2841
2842         if (cnt >= sizeof(buf))
2843                 return -EINVAL;
2844
2845         if (copy_from_user(&buf, ubuf, cnt))
2846                 return -EFAULT;
2847
2848         buf[cnt] = 0;
2849
2850         if (strncmp(buf, "no", 2) == 0) {
2851                 neg = 1;
2852                 cmp += 2;
2853         }
2854
2855         for (i = 0; trace_options[i]; i++) {
2856                 int len = strlen(trace_options[i]);
2857
2858                 if (strncmp(cmp, trace_options[i], len) == 0) {
2859                         if (neg)
2860                                 trace_flags &= ~(1 << i);
2861                         else
2862                                 trace_flags |= (1 << i);
2863                         break;
2864                 }
2865         }
2866
2867         /* If no option could be set, test the specific tracer options */
2868         if (!trace_options[i]) {
2869                 ret = set_tracer_option(current_trace, cmp, neg);
2870                 if (ret)
2871                         return ret;
2872         }
2873
2874         filp->f_pos += cnt;
2875
2876         return cnt;
2877 }
2878
2879 static struct file_operations tracing_iter_fops = {
2880         .open           = tracing_open_generic,
2881         .read           = tracing_trace_options_read,
2882         .write          = tracing_trace_options_write,
2883 };
2884
2885 static const char readme_msg[] =
2886         "tracing mini-HOWTO:\n\n"
2887         "# mkdir /debug\n"
2888         "# mount -t debugfs nodev /debug\n\n"
2889         "# cat /debug/tracing/available_tracers\n"
2890         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2891         "# cat /debug/tracing/current_tracer\n"
2892         "none\n"
2893         "# echo sched_switch > /debug/tracing/current_tracer\n"
2894         "# cat /debug/tracing/current_tracer\n"
2895         "sched_switch\n"
2896         "# cat /debug/tracing/trace_options\n"
2897         "noprint-parent nosym-offset nosym-addr noverbose\n"
2898         "# echo print-parent > /debug/tracing/trace_options\n"
2899         "# echo 1 > /debug/tracing/tracing_enabled\n"
2900         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2901         "echo 0 > /debug/tracing/tracing_enabled\n"
2902 ;
2903
2904 static ssize_t
2905 tracing_readme_read(struct file *filp, char __user *ubuf,
2906                        size_t cnt, loff_t *ppos)
2907 {
2908         return simple_read_from_buffer(ubuf, cnt, ppos,
2909                                         readme_msg, strlen(readme_msg));
2910 }
2911
2912 static struct file_operations tracing_readme_fops = {
2913         .open           = tracing_open_generic,
2914         .read           = tracing_readme_read,
2915 };
2916
2917 static ssize_t
2918 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2919                   size_t cnt, loff_t *ppos)
2920 {
2921         char buf[64];
2922         int r;
2923
2924         r = sprintf(buf, "%u\n", tracer_enabled);
2925         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2926 }
2927
2928 static ssize_t
2929 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2930                    size_t cnt, loff_t *ppos)
2931 {
2932         struct trace_array *tr = filp->private_data;
2933         char buf[64];
2934         long val;
2935         int ret;
2936
2937         if (cnt >= sizeof(buf))
2938                 return -EINVAL;
2939
2940         if (copy_from_user(&buf, ubuf, cnt))
2941                 return -EFAULT;
2942
2943         buf[cnt] = 0;
2944
2945         ret = strict_strtoul(buf, 10, &val);
2946         if (ret < 0)
2947                 return ret;
2948
2949         val = !!val;
2950
2951         mutex_lock(&trace_types_lock);
2952         if (tracer_enabled ^ val) {
2953                 if (val) {
2954                         tracer_enabled = 1;
2955                         if (current_trace->start)
2956                                 current_trace->start(tr);
2957                         tracing_start();
2958                 } else {
2959                         tracer_enabled = 0;
2960                         tracing_stop();
2961                         if (current_trace->stop)
2962                                 current_trace->stop(tr);
2963                 }
2964         }
2965         mutex_unlock(&trace_types_lock);
2966
2967         filp->f_pos += cnt;
2968
2969         return cnt;
2970 }
2971
2972 static ssize_t
2973 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2974                        size_t cnt, loff_t *ppos)
2975 {
2976         char buf[max_tracer_type_len+2];
2977         int r;
2978
2979         mutex_lock(&trace_types_lock);
2980         if (current_trace)
2981                 r = sprintf(buf, "%s\n", current_trace->name);
2982         else
2983                 r = sprintf(buf, "\n");
2984         mutex_unlock(&trace_types_lock);
2985
2986         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2987 }
2988
2989 static int tracing_set_tracer(char *buf)
2990 {
2991         struct trace_array *tr = &global_trace;
2992         struct tracer *t;
2993         int ret = 0;
2994
2995         mutex_lock(&trace_types_lock);
2996         for (t = trace_types; t; t = t->next) {
2997                 if (strcmp(t->name, buf) == 0)
2998                         break;
2999         }
3000         if (!t) {
3001                 ret = -EINVAL;
3002                 goto out;
3003         }
3004         if (t == current_trace)
3005                 goto out;
3006
3007         trace_branch_disable();
3008         if (current_trace && current_trace->reset)
3009                 current_trace->reset(tr);
3010
3011         current_trace = t;
3012         if (t->init) {
3013                 ret = t->init(tr);
3014                 if (ret)
3015                         goto out;
3016         }
3017
3018         trace_branch_enable(tr);
3019  out:
3020         mutex_unlock(&trace_types_lock);
3021
3022         return ret;
3023 }
3024
3025 static ssize_t
3026 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3027                         size_t cnt, loff_t *ppos)
3028 {
3029         char buf[max_tracer_type_len+1];
3030         int i;
3031         size_t ret;
3032         int err;
3033
3034         ret = cnt;
3035
3036         if (cnt > max_tracer_type_len)
3037                 cnt = max_tracer_type_len;
3038
3039         if (copy_from_user(&buf, ubuf, cnt))
3040                 return -EFAULT;
3041
3042         buf[cnt] = 0;
3043
3044         /* strip ending whitespace. */
3045         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3046                 buf[i] = 0;
3047
3048         err = tracing_set_tracer(buf);
3049         if (err)
3050                 return err;
3051
3052         filp->f_pos += ret;
3053
3054         return ret;
3055 }
3056
3057 static ssize_t
3058 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3059                      size_t cnt, loff_t *ppos)
3060 {
3061         unsigned long *ptr = filp->private_data;
3062         char buf[64];
3063         int r;
3064
3065         r = snprintf(buf, sizeof(buf), "%ld\n",
3066                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3067         if (r > sizeof(buf))
3068                 r = sizeof(buf);
3069         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3070 }
3071
3072 static ssize_t
3073 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3074                       size_t cnt, loff_t *ppos)
3075 {
3076         long *ptr = filp->private_data;
3077         char buf[64];
3078         long val;
3079         int ret;
3080
3081         if (cnt >= sizeof(buf))
3082                 return -EINVAL;
3083
3084         if (copy_from_user(&buf, ubuf, cnt))
3085                 return -EFAULT;
3086
3087         buf[cnt] = 0;
3088
3089         ret = strict_strtoul(buf, 10, &val);
3090         if (ret < 0)
3091                 return ret;
3092
3093         *ptr = val * 1000;
3094
3095         return cnt;
3096 }
3097
3098 static atomic_t tracing_reader;
3099
3100 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3101 {
3102         struct trace_iterator *iter;
3103
3104         if (tracing_disabled)
3105                 return -ENODEV;
3106
3107         /* We only allow for reader of the pipe */
3108         if (atomic_inc_return(&tracing_reader) != 1) {
3109                 atomic_dec(&tracing_reader);
3110                 return -EBUSY;
3111         }
3112
3113         /* create a buffer to store the information to pass to userspace */
3114         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3115         if (!iter)
3116                 return -ENOMEM;
3117
3118         mutex_lock(&trace_types_lock);
3119
3120         /* trace pipe does not show start of buffer */
3121         cpus_setall(iter->started);
3122
3123         iter->tr = &global_trace;
3124         iter->trace = current_trace;
3125         filp->private_data = iter;
3126
3127         if (iter->trace->pipe_open)
3128                 iter->trace->pipe_open(iter);
3129         mutex_unlock(&trace_types_lock);
3130
3131         return 0;
3132 }
3133
3134 static int tracing_release_pipe(struct inode *inode, struct file *file)
3135 {
3136         struct trace_iterator *iter = file->private_data;
3137
3138         kfree(iter);
3139         atomic_dec(&tracing_reader);
3140
3141         return 0;
3142 }
3143
3144 static unsigned int
3145 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3146 {
3147         struct trace_iterator *iter = filp->private_data;
3148
3149         if (trace_flags & TRACE_ITER_BLOCK) {
3150                 /*
3151                  * Always select as readable when in blocking mode
3152                  */
3153                 return POLLIN | POLLRDNORM;
3154         } else {
3155                 if (!trace_empty(iter))
3156                         return POLLIN | POLLRDNORM;
3157                 poll_wait(filp, &trace_wait, poll_table);
3158                 if (!trace_empty(iter))
3159                         return POLLIN | POLLRDNORM;
3160
3161                 return 0;
3162         }
3163 }
3164
3165 /*
3166  * Consumer reader.
3167  */
3168 static ssize_t
3169 tracing_read_pipe(struct file *filp, char __user *ubuf,
3170                   size_t cnt, loff_t *ppos)
3171 {
3172         struct trace_iterator *iter = filp->private_data;
3173         ssize_t sret;
3174
3175         /* return any leftover data */
3176         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3177         if (sret != -EBUSY)
3178                 return sret;
3179
3180         trace_seq_reset(&iter->seq);
3181
3182         mutex_lock(&trace_types_lock);
3183         if (iter->trace->read) {
3184                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3185                 if (sret)
3186                         goto out;
3187         }
3188
3189 waitagain:
3190         sret = 0;
3191         while (trace_empty(iter)) {
3192
3193                 if ((filp->f_flags & O_NONBLOCK)) {
3194                         sret = -EAGAIN;
3195                         goto out;
3196                 }
3197
3198                 /*
3199                  * This is a make-shift waitqueue. The reason we don't use
3200                  * an actual wait queue is because:
3201                  *  1) we only ever have one waiter
3202                  *  2) the tracing, traces all functions, we don't want
3203                  *     the overhead of calling wake_up and friends
3204                  *     (and tracing them too)
3205                  *     Anyway, this is really very primitive wakeup.
3206                  */
3207                 set_current_state(TASK_INTERRUPTIBLE);
3208                 iter->tr->waiter = current;
3209
3210                 mutex_unlock(&trace_types_lock);
3211
3212                 /* sleep for 100 msecs, and try again. */
3213                 schedule_timeout(HZ/10);
3214
3215                 mutex_lock(&trace_types_lock);
3216
3217                 iter->tr->waiter = NULL;
3218
3219                 if (signal_pending(current)) {
3220                         sret = -EINTR;
3221                         goto out;
3222                 }
3223
3224                 if (iter->trace != current_trace)
3225                         goto out;
3226
3227                 /*
3228                  * We block until we read something and tracing is disabled.
3229                  * We still block if tracing is disabled, but we have never
3230                  * read anything. This allows a user to cat this file, and
3231                  * then enable tracing. But after we have read something,
3232                  * we give an EOF when tracing is again disabled.
3233                  *
3234                  * iter->pos will be 0 if we haven't read anything.
3235                  */
3236                 if (!tracer_enabled && iter->pos)
3237                         break;
3238
3239                 continue;
3240         }
3241
3242         /* stop when tracing is finished */
3243         if (trace_empty(iter))
3244                 goto out;
3245
3246         if (cnt >= PAGE_SIZE)
3247                 cnt = PAGE_SIZE - 1;
3248
3249         /* reset all but tr, trace, and overruns */
3250         memset(&iter->seq, 0,
3251                sizeof(struct trace_iterator) -
3252                offsetof(struct trace_iterator, seq));
3253         iter->pos = -1;
3254
3255         while (find_next_entry_inc(iter) != NULL) {
3256                 enum print_line_t ret;
3257                 int len = iter->seq.len;
3258
3259                 ret = print_trace_line(iter);
3260                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3261                         /* don't print partial lines */
3262                         iter->seq.len = len;
3263                         break;
3264                 }
3265
3266                 trace_consume(iter);
3267
3268                 if (iter->seq.len >= cnt)
3269                         break;
3270         }
3271
3272         /* Now copy what we have to the user */
3273         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3274         if (iter->seq.readpos >= iter->seq.len)
3275                 trace_seq_reset(&iter->seq);
3276
3277         /*
3278          * If there was nothing to send to user, inspite of consuming trace
3279          * entries, go back to wait for more entries.
3280          */
3281         if (sret == -EBUSY)
3282                 goto waitagain;
3283
3284 out:
3285         mutex_unlock(&trace_types_lock);
3286
3287         return sret;
3288 }
3289
3290 static ssize_t
3291 tracing_entries_read(struct file *filp, char __user *ubuf,
3292                      size_t cnt, loff_t *ppos)
3293 {
3294         struct trace_array *tr = filp->private_data;
3295         char buf[64];
3296         int r;
3297
3298         r = sprintf(buf, "%lu\n", tr->entries >> 10);
3299         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3300 }
3301
3302 static ssize_t
3303 tracing_entries_write(struct file *filp, const char __user *ubuf,
3304                       size_t cnt, loff_t *ppos)
3305 {
3306         unsigned long val;
3307         char buf[64];
3308         int ret, cpu;
3309
3310         if (cnt >= sizeof(buf))
3311                 return -EINVAL;
3312
3313         if (copy_from_user(&buf, ubuf, cnt))
3314                 return -EFAULT;
3315
3316         buf[cnt] = 0;
3317
3318         ret = strict_strtoul(buf, 10, &val);
3319         if (ret < 0)
3320                 return ret;
3321
3322         /* must have at least 1 entry */
3323         if (!val)
3324                 return -EINVAL;
3325
3326         mutex_lock(&trace_types_lock);
3327
3328         tracing_stop();
3329
3330         /* disable all cpu buffers */
3331         for_each_tracing_cpu(cpu) {
3332                 if (global_trace.data[cpu])
3333                         atomic_inc(&global_trace.data[cpu]->disabled);
3334                 if (max_tr.data[cpu])
3335                         atomic_inc(&max_tr.data[cpu]->disabled);
3336         }
3337
3338         /* value is in KB */
3339         val <<= 10;
3340
3341         if (val != global_trace.entries) {
3342                 ret = ring_buffer_resize(global_trace.buffer, val);
3343                 if (ret < 0) {
3344                         cnt = ret;
3345                         goto out;
3346                 }
3347
3348                 ret = ring_buffer_resize(max_tr.buffer, val);
3349                 if (ret < 0) {
3350                         int r;
3351                         cnt = ret;
3352                         r = ring_buffer_resize(global_trace.buffer,
3353                                                global_trace.entries);
3354                         if (r < 0) {
3355                                 /* AARGH! We are left with different
3356                                  * size max buffer!!!! */
3357                                 WARN_ON(1);
3358                                 tracing_disabled = 1;
3359                         }
3360                         goto out;
3361                 }
3362
3363                 global_trace.entries = val;
3364         }
3365
3366         filp->f_pos += cnt;
3367
3368         /* If check pages failed, return ENOMEM */
3369         if (tracing_disabled)
3370                 cnt = -ENOMEM;
3371  out:
3372         for_each_tracing_cpu(cpu) {
3373                 if (global_trace.data[cpu])
3374                         atomic_dec(&global_trace.data[cpu]->disabled);
3375                 if (max_tr.data[cpu])
3376                         atomic_dec(&max_tr.data[cpu]->disabled);
3377         }
3378
3379         tracing_start();
3380         max_tr.entries = global_trace.entries;
3381         mutex_unlock(&trace_types_lock);
3382
3383         return cnt;
3384 }
3385
3386 static int mark_printk(const char *fmt, ...)
3387 {
3388         int ret;
3389         va_list args;
3390         va_start(args, fmt);
3391         ret = trace_vprintk(0, -1, fmt, args);
3392         va_end(args);
3393         return ret;
3394 }
3395
3396 static ssize_t
3397 tracing_mark_write(struct file *filp, const char __user *ubuf,
3398                                         size_t cnt, loff_t *fpos)
3399 {
3400         char *buf;
3401         char *end;
3402
3403         if (tracing_disabled)
3404                 return -EINVAL;
3405
3406         if (cnt > TRACE_BUF_SIZE)
3407                 cnt = TRACE_BUF_SIZE;
3408
3409         buf = kmalloc(cnt + 1, GFP_KERNEL);
3410         if (buf == NULL)
3411                 return -ENOMEM;
3412
3413         if (copy_from_user(buf, ubuf, cnt)) {
3414                 kfree(buf);
3415                 return -EFAULT;
3416         }
3417
3418         /* Cut from the first nil or newline. */
3419         buf[cnt] = '\0';
3420         end = strchr(buf, '\n');
3421         if (end)
3422                 *end = '\0';
3423
3424         cnt = mark_printk("%s\n", buf);
3425         kfree(buf);
3426         *fpos += cnt;
3427
3428         return cnt;
3429 }
3430
3431 static struct file_operations tracing_max_lat_fops = {
3432         .open           = tracing_open_generic,
3433         .read           = tracing_max_lat_read,
3434         .write          = tracing_max_lat_write,
3435 };
3436
3437 static struct file_operations tracing_ctrl_fops = {
3438         .open           = tracing_open_generic,
3439         .read           = tracing_ctrl_read,
3440         .write          = tracing_ctrl_write,
3441 };
3442
3443 static struct file_operations set_tracer_fops = {
3444         .open           = tracing_open_generic,
3445         .read           = tracing_set_trace_read,
3446         .write          = tracing_set_trace_write,
3447 };
3448
3449 static struct file_operations tracing_pipe_fops = {
3450         .open           = tracing_open_pipe,
3451         .poll           = tracing_poll_pipe,
3452         .read           = tracing_read_pipe,
3453         .release        = tracing_release_pipe,
3454 };
3455
3456 static struct file_operations tracing_entries_fops = {
3457         .open           = tracing_open_generic,
3458         .read           = tracing_entries_read,
3459         .write          = tracing_entries_write,
3460 };
3461
3462 static struct file_operations tracing_mark_fops = {
3463         .open           = tracing_open_generic,
3464         .write          = tracing_mark_write,
3465 };
3466
3467 #ifdef CONFIG_DYNAMIC_FTRACE
3468
3469 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3470 {
3471         return 0;
3472 }
3473
3474 static ssize_t
3475 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3476                   size_t cnt, loff_t *ppos)
3477 {
3478         static char ftrace_dyn_info_buffer[1024];
3479         static DEFINE_MUTEX(dyn_info_mutex);
3480         unsigned long *p = filp->private_data;
3481         char *buf = ftrace_dyn_info_buffer;
3482         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3483         int r;
3484
3485         mutex_lock(&dyn_info_mutex);
3486         r = sprintf(buf, "%ld ", *p);
3487
3488         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3489         buf[r++] = '\n';
3490
3491         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3492
3493         mutex_unlock(&dyn_info_mutex);
3494
3495         return r;
3496 }
3497
3498 static struct file_operations tracing_dyn_info_fops = {
3499         .open           = tracing_open_generic,
3500         .read           = tracing_read_dyn_info,
3501 };
3502 #endif
3503
3504 static struct dentry *d_tracer;
3505
3506 struct dentry *tracing_init_dentry(void)
3507 {
3508         static int once;
3509
3510         if (d_tracer)
3511                 return d_tracer;
3512
3513         d_tracer = debugfs_create_dir("tracing", NULL);
3514
3515         if (!d_tracer && !once) {
3516                 once = 1;
3517                 pr_warning("Could not create debugfs directory 'tracing'\n");
3518                 return NULL;
3519         }
3520
3521         return d_tracer;
3522 }
3523
3524 #ifdef CONFIG_FTRACE_SELFTEST
3525 /* Let selftest have access to static functions in this file */
3526 #include "trace_selftest.c"
3527 #endif
3528
3529 static __init int tracer_init_debugfs(void)
3530 {
3531         struct dentry *d_tracer;
3532         struct dentry *entry;
3533
3534         d_tracer = tracing_init_dentry();
3535
3536         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3537                                     &global_trace, &tracing_ctrl_fops);
3538         if (!entry)
3539                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3540
3541         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3542                                     NULL, &tracing_iter_fops);
3543         if (!entry)
3544                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3545
3546         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3547                                     NULL, &tracing_cpumask_fops);
3548         if (!entry)
3549                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3550
3551         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3552                                     &global_trace, &tracing_lt_fops);
3553         if (!entry)
3554                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3555
3556         entry = debugfs_create_file("trace", 0444, d_tracer,
3557                                     &global_trace, &tracing_fops);
3558         if (!entry)
3559                 pr_warning("Could not create debugfs 'trace' entry\n");
3560
3561         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3562                                     &global_trace, &show_traces_fops);
3563         if (!entry)
3564                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3565
3566         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3567                                     &global_trace, &set_tracer_fops);
3568         if (!entry)
3569                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3570
3571         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3572                                     &tracing_max_latency,
3573                                     &tracing_max_lat_fops);
3574         if (!entry)
3575                 pr_warning("Could not create debugfs "
3576                            "'tracing_max_latency' entry\n");
3577
3578         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3579                                     &tracing_thresh, &tracing_max_lat_fops);
3580         if (!entry)
3581                 pr_warning("Could not create debugfs "
3582                            "'tracing_thresh' entry\n");
3583         entry = debugfs_create_file("README", 0644, d_tracer,
3584                                     NULL, &tracing_readme_fops);
3585         if (!entry)
3586                 pr_warning("Could not create debugfs 'README' entry\n");
3587
3588         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3589                                     NULL, &tracing_pipe_fops);
3590         if (!entry)
3591                 pr_warning("Could not create debugfs "
3592                            "'trace_pipe' entry\n");
3593
3594         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3595                                     &global_trace, &tracing_entries_fops);
3596         if (!entry)
3597                 pr_warning("Could not create debugfs "
3598                            "'buffer_size_kb' entry\n");
3599
3600         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3601                                     NULL, &tracing_mark_fops);
3602         if (!entry)
3603                 pr_warning("Could not create debugfs "
3604                            "'trace_marker' entry\n");
3605
3606 #ifdef CONFIG_DYNAMIC_FTRACE
3607         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3608                                     &ftrace_update_tot_cnt,
3609                                     &tracing_dyn_info_fops);
3610         if (!entry)
3611                 pr_warning("Could not create debugfs "
3612                            "'dyn_ftrace_total_info' entry\n");
3613 #endif
3614 #ifdef CONFIG_SYSPROF_TRACER
3615         init_tracer_sysprof_debugfs(d_tracer);
3616 #endif
3617         return 0;
3618 }
3619
3620 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
3621 {
3622         static DEFINE_SPINLOCK(trace_buf_lock);
3623         static char trace_buf[TRACE_BUF_SIZE];
3624
3625         struct ring_buffer_event *event;
3626         struct trace_array *tr = &global_trace;
3627         struct trace_array_cpu *data;
3628         int cpu, len = 0, size, pc;
3629         struct print_entry *entry;
3630         unsigned long irq_flags;
3631
3632         if (tracing_disabled || tracing_selftest_running)
3633                 return 0;
3634
3635         pc = preempt_count();
3636         preempt_disable_notrace();
3637         cpu = raw_smp_processor_id();
3638         data = tr->data[cpu];
3639
3640         if (unlikely(atomic_read(&data->disabled)))
3641                 goto out;
3642
3643         pause_graph_tracing();
3644         spin_lock_irqsave(&trace_buf_lock, irq_flags);
3645         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3646
3647         len = min(len, TRACE_BUF_SIZE-1);
3648         trace_buf[len] = 0;
3649
3650         size = sizeof(*entry) + len + 1;
3651         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3652         if (!event)
3653                 goto out_unlock;
3654         entry = ring_buffer_event_data(event);
3655         tracing_generic_entry_update(&entry->ent, irq_flags, pc);
3656         entry->ent.type                 = TRACE_PRINT;
3657         entry->ip                       = ip;
3658         entry->depth                    = depth;
3659
3660         memcpy(&entry->buf, trace_buf, len);
3661         entry->buf[len] = 0;
3662         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3663
3664  out_unlock:
3665         spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
3666         unpause_graph_tracing();
3667  out:
3668         preempt_enable_notrace();
3669
3670         return len;
3671 }
3672 EXPORT_SYMBOL_GPL(trace_vprintk);
3673
3674 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3675 {
3676         int ret;
3677         va_list ap;
3678
3679         if (!(trace_flags & TRACE_ITER_PRINTK))
3680                 return 0;
3681
3682         va_start(ap, fmt);
3683         ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
3684         va_end(ap);
3685         return ret;
3686 }
3687 EXPORT_SYMBOL_GPL(__ftrace_printk);
3688
3689 static int trace_panic_handler(struct notifier_block *this,
3690                                unsigned long event, void *unused)
3691 {
3692         if (ftrace_dump_on_oops)
3693                 ftrace_dump();
3694         return NOTIFY_OK;
3695 }
3696
3697 static struct notifier_block trace_panic_notifier = {
3698         .notifier_call  = trace_panic_handler,
3699         .next           = NULL,
3700         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3701 };
3702
3703 static int trace_die_handler(struct notifier_block *self,
3704                              unsigned long val,
3705                              void *data)
3706 {
3707         switch (val) {
3708         case DIE_OOPS:
3709                 if (ftrace_dump_on_oops)
3710                         ftrace_dump();
3711                 break;
3712         default:
3713                 break;
3714         }
3715         return NOTIFY_OK;
3716 }
3717
3718 static struct notifier_block trace_die_notifier = {
3719         .notifier_call = trace_die_handler,
3720         .priority = 200
3721 };
3722
3723 /*
3724  * printk is set to max of 1024, we really don't need it that big.
3725  * Nothing should be printing 1000 characters anyway.
3726  */
3727 #define TRACE_MAX_PRINT         1000
3728
3729 /*
3730  * Define here KERN_TRACE so that we have one place to modify
3731  * it if we decide to change what log level the ftrace dump
3732  * should be at.
3733  */
3734 #define KERN_TRACE              KERN_INFO
3735
3736 static void
3737 trace_printk_seq(struct trace_seq *s)
3738 {
3739         /* Probably should print a warning here. */
3740         if (s->len >= 1000)
3741                 s->len = 1000;
3742
3743         /* should be zero ended, but we are paranoid. */
3744         s->buffer[s->len] = 0;
3745
3746         printk(KERN_TRACE "%s", s->buffer);
3747
3748         trace_seq_reset(s);
3749 }
3750
3751 void ftrace_dump(void)
3752 {
3753         static DEFINE_SPINLOCK(ftrace_dump_lock);
3754         /* use static because iter can be a bit big for the stack */
3755         static struct trace_iterator iter;
3756         static cpumask_t mask;
3757         static int dump_ran;
3758         unsigned long flags;
3759         int cnt = 0, cpu;
3760
3761         /* only one dump */
3762         spin_lock_irqsave(&ftrace_dump_lock, flags);
3763         if (dump_ran)
3764                 goto out;
3765
3766         dump_ran = 1;
3767
3768         /* No turning back! */
3769         ftrace_kill();
3770
3771         for_each_tracing_cpu(cpu) {
3772                 atomic_inc(&global_trace.data[cpu]->disabled);
3773         }
3774
3775         /* don't look at user memory in panic mode */
3776         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3777
3778         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3779
3780         iter.tr = &global_trace;
3781         iter.trace = current_trace;
3782
3783         /*
3784          * We need to stop all tracing on all CPUS to read the
3785          * the next buffer. This is a bit expensive, but is
3786          * not done often. We fill all what we can read,
3787          * and then release the locks again.
3788          */
3789
3790         cpus_clear(mask);
3791
3792         while (!trace_empty(&iter)) {
3793
3794                 if (!cnt)
3795                         printk(KERN_TRACE "---------------------------------\n");
3796
3797                 cnt++;
3798
3799                 /* reset all but tr, trace, and overruns */
3800                 memset(&iter.seq, 0,
3801                        sizeof(struct trace_iterator) -
3802                        offsetof(struct trace_iterator, seq));
3803                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3804                 iter.pos = -1;
3805
3806                 if (find_next_entry_inc(&iter) != NULL) {
3807                         print_trace_line(&iter);
3808                         trace_consume(&iter);
3809                 }
3810
3811                 trace_printk_seq(&iter.seq);
3812         }
3813
3814         if (!cnt)
3815                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3816         else
3817                 printk(KERN_TRACE "---------------------------------\n");
3818
3819  out:
3820         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3821 }
3822
3823 __init static int tracer_alloc_buffers(void)
3824 {
3825         struct trace_array_cpu *data;
3826         int i;
3827
3828         /* TODO: make the number of buffers hot pluggable with CPUS */
3829         tracing_buffer_mask = cpu_possible_map;
3830
3831         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3832                                                    TRACE_BUFFER_FLAGS);
3833         if (!global_trace.buffer) {
3834                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3835                 WARN_ON(1);
3836                 return 0;
3837         }
3838         global_trace.entries = ring_buffer_size(global_trace.buffer);
3839
3840 #ifdef CONFIG_TRACER_MAX_TRACE
3841         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3842                                              TRACE_BUFFER_FLAGS);
3843         if (!max_tr.buffer) {
3844                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3845                 WARN_ON(1);
3846                 ring_buffer_free(global_trace.buffer);
3847                 return 0;
3848         }
3849         max_tr.entries = ring_buffer_size(max_tr.buffer);
3850         WARN_ON(max_tr.entries != global_trace.entries);
3851 #endif
3852
3853         /* Allocate the first page for all buffers */
3854         for_each_tracing_cpu(i) {
3855                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3856                 max_tr.data[i] = &per_cpu(max_data, i);
3857         }
3858
3859         trace_init_cmdlines();
3860
3861         register_tracer(&nop_trace);
3862 #ifdef CONFIG_BOOT_TRACER
3863         register_tracer(&boot_tracer);
3864         current_trace = &boot_tracer;
3865         current_trace->init(&global_trace);
3866 #else
3867         current_trace = &nop_trace;
3868 #endif
3869
3870         /* All seems OK, enable tracing */
3871         tracing_disabled = 0;
3872
3873         atomic_notifier_chain_register(&panic_notifier_list,
3874                                        &trace_panic_notifier);
3875
3876         register_die_notifier(&trace_die_notifier);
3877
3878         return 0;
3879 }
3880 early_initcall(tracer_alloc_buffers);
3881 fs_initcall(tracer_init_debugfs);