kernel/trace/trace.c

   1 /*
   2  * ring buffer based function tracer
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally taken from the RT patch by:
   8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code from the latency_tracer, that is:
  11  *  Copyright (C) 2004-2006 Ingo Molnar
  12  *  Copyright (C) 2004 William Lee Irwin III
  13  */
  14 #include <linux/utsrelease.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/notifier.h>
  18 #include <linux/debugfs.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/hardirq.h>
  21 #include <linux/linkage.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/ftrace.h>
  24 #include <linux/module.h>
  25 #include <linux/percpu.h>
  26 #include <linux/kdebug.h>
  27 #include <linux/ctype.h>
  28 #include <linux/init.h>
  29 #include <linux/poll.h>
  30 #include <linux/gfp.h>
  31 #include <linux/fs.h>
  32 #include <linux/kprobes.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/writeback.h>
  35
  36 #include <linux/stacktrace.h>
  37 #include <linux/ring_buffer.h>
  38 #include <linux/irqflags.h>
  39
  40 #include "trace.h"
  41
  42 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
  43
  44 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  45 unsigned long __read_mostly     tracing_thresh;
  46
  47 /* For tracers that don't implement custom flags */
  48 static struct tracer_opt dummy_tracer_opt[] = {
  49         { }
  50 };
  51
  52 static struct tracer_flags dummy_tracer_flags = {
  53         .val = 0,
  54         .opts = dummy_tracer_opt
  55 };
  56
  57 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
  58 {
  59         return 0;
  60 }
  61
  62 /*
  63  * Kill all tracing for good (never come back).
  64  * It is initialized to 1 but will turn to zero if the initialization
  65  * of the tracer is successful. But that is the only place that sets
  66  * this back to zero.
  67  */
  68 int tracing_disabled = 1;
  69
  70 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
  71
  72 static inline void ftrace_disable_cpu(void)
  73 {
  74         preempt_disable();
  75         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
  76 }
  77
  78 static inline void ftrace_enable_cpu(void)
  79 {
  80         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
  81         preempt_enable();
  82 }
  83
  84 static cpumask_t __read_mostly          tracing_buffer_mask;
  85
  86 #define for_each_tracing_cpu(cpu)       \
  87         for_each_cpu_mask(cpu, tracing_buffer_mask)
  88
  89 /*
  90  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
  91  *
  92  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
  93  * is set, then ftrace_dump is called. This will output the contents
  94  * of the ftrace buffers to the console.  This is very useful for
  95  * capturing traces that lead to crashes and outputing it to a
  96  * serial console.
  97  *
  98  * It is default off, but you can enable it with either specifying
  99  * "ftrace_dump_on_oops" in the kernel command line, or setting
 100  * /proc/sys/kernel/ftrace_dump_on_oops to true.
 101  */
 102 int ftrace_dump_on_oops;
 103
 104 static int tracing_set_tracer(char *buf);
 105
 106 static int __init set_ftrace(char *str)
 107 {
 108         tracing_set_tracer(str);
 109         return 1;
 110 }
 111 __setup("ftrace", set_ftrace);
 112
 113 static int __init set_ftrace_dump_on_oops(char *str)
 114 {
 115         ftrace_dump_on_oops = 1;
 116         return 1;
 117 }
 118 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
 119
 120 long
 121 ns2usecs(cycle_t nsec)
 122 {
 123         nsec += 500;
 124         do_div(nsec, 1000);
 125         return nsec;
 126 }
 127
 128 cycle_t ftrace_now(int cpu)
 129 {
 130         u64 ts = ring_buffer_time_stamp(cpu);
 131         ring_buffer_normalize_time_stamp(cpu, &ts);
 132         return ts;
 133 }
 134
 135 /*
 136  * The global_trace is the descriptor that holds the tracing
 137  * buffers for the live tracing. For each CPU, it contains
 138  * a link list of pages that will store trace entries. The
 139  * page descriptor of the pages in the memory is used to hold
 140  * the link list by linking the lru item in the page descriptor
 141  * to each of the pages in the buffer per CPU.
 142  *
 143  * For each active CPU there is a data field that holds the
 144  * pages for the buffer for that CPU. Each CPU has the same number
 145  * of pages allocated for its buffer.
 146  */
 147 static struct trace_array       global_trace;
 148
 149 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
 150
 151 /*
 152  * The max_tr is used to snapshot the global_trace when a maximum
 153  * latency is reached. Some tracers will use this to store a maximum
 154  * trace while it continues examining live traces.
 155  *
 156  * The buffers for the max_tr are set up the same as the global_trace.
 157  * When a snapshot is taken, the link list of the max_tr is swapped
 158  * with the link list of the global_trace and the buffers are reset for
 159  * the global_trace so the tracing can continue.
 160  */
 161 static struct trace_array       max_tr;
 162
 163 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 164
 165 /* tracer_enabled is used to toggle activation of a tracer */
 166 static int                      tracer_enabled = 1;
 167
 168 /**
 169  * tracing_is_enabled - return tracer_enabled status
 170  *
 171  * This function is used by other tracers to know the status
 172  * of the tracer_enabled flag.  Tracers may use this function
 173  * to know if it should enable their features when starting
 174  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
 175  */
 176 int tracing_is_enabled(void)
 177 {
 178         return tracer_enabled;
 179 }
 180
 181 /* function tracing enabled */
 182 int                             ftrace_function_enabled;
 183
 184 /*
 185  * trace_buf_size is the size in bytes that is allocated
 186  * for a buffer. Note, the number of bytes is always rounded
 187  * to page size.
 188  *
 189  * This number is purposely set to a low number of 16384.
 190  * If the dump on oops happens, it will be much appreciated
 191  * to not have to wait for all that output. Anyway this can be
 192  * boot time and run time configurable.
 193  */
 194 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
 195
 196 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
 197
 198 /* trace_types holds a link list of available tracers. */
 199 static struct tracer            *trace_types __read_mostly;
 200
 201 /* current_trace points to the tracer that is currently active */
 202 static struct tracer            *current_trace __read_mostly;
 203
 204 /*
 205  * max_tracer_type_len is used to simplify the allocating of
 206  * buffers to read userspace tracer names. We keep track of
 207  * the longest tracer name registered.
 208  */
 209 static int                      max_tracer_type_len;
 210
 211 /*
 212  * trace_types_lock is used to protect the trace_types list.
 213  * This lock is also used to keep user access serialized.
 214  * Accesses from userspace will grab this lock while userspace
 215  * activities happen inside the kernel.
 216  */
 217 static DEFINE_MUTEX(trace_types_lock);
 218
 219 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
 220 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 221
 222 /* trace_flags holds trace_options default values */
 223 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
 224         TRACE_ITER_ANNOTATE;
 225
 226 /**
 227  * trace_wake_up - wake up tasks waiting for trace input
 228  *
 229  * Simply wakes up any task that is blocked on the trace_wait
 230  * queue. These is used with trace_poll for tasks polling the trace.
 231  */
 232 void trace_wake_up(void)
 233 {
 234         /*
 235          * The runqueue_is_locked() can fail, but this is the best we
 236          * have for now:
 237          */
 238         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
 239                 wake_up(&trace_wait);
 240 }
 241
 242 static int __init set_buf_size(char *str)
 243 {
 244         unsigned long buf_size;
 245         int ret;
 246
 247         if (!str)
 248                 return 0;
 249         ret = strict_strtoul(str, 0, &buf_size);
 250         /* nr_entries can not be zero */
 251         if (ret < 0 || buf_size == 0)
 252                 return 0;
 253         trace_buf_size = buf_size;
 254         return 1;
 255 }
 256 __setup("trace_buf_size=", set_buf_size);
 257
 258 unsigned long nsecs_to_usecs(unsigned long nsecs)
 259 {
 260         return nsecs / 1000;
 261 }
 262
 263 /* These must match the bit postions in trace_iterator_flags */
 264 static const char *trace_options[] = {
 265         "print-parent",
 266         "sym-offset",
 267         "sym-addr",
 268         "verbose",
 269         "raw",
 270         "hex",
 271         "bin",
 272         "block",
 273         "stacktrace",
 274         "sched-tree",
 275         "ftrace_printk",
 276         "ftrace_preempt",
 277         "branch",
 278         "annotate",
 279         "userstacktrace",
 280         "sym-userobj",
 281         NULL
 282 };
 283
 284 /*
 285  * ftrace_max_lock is used to protect the swapping of buffers
 286  * when taking a max snapshot. The buffers themselves are
 287  * protected by per_cpu spinlocks. But the action of the swap
 288  * needs its own lock.
 289  *
 290  * This is defined as a raw_spinlock_t in order to help
 291  * with performance when lockdep debugging is enabled.
 292  */
 293 static raw_spinlock_t ftrace_max_lock =
 294         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 295
 296 /*
 297  * Copy the new maximum trace into the separate maximum-trace
 298  * structure. (this way the maximum trace is permanently saved,
 299  * for later retrieval via /debugfs/tracing/latency_trace)
 300  */
 301 static void
 302 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 303 {
 304         struct trace_array_cpu *data = tr->data[cpu];
 305
 306         max_tr.cpu = cpu;
 307         max_tr.time_start = data->preempt_timestamp;
 308
 309         data = max_tr.data[cpu];
 310         data->saved_latency = tracing_max_latency;
 311
 312         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 313         data->pid = tsk->pid;
 314         data->uid = tsk->uid;
 315         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 316         data->policy = tsk->policy;
 317         data->rt_priority = tsk->rt_priority;
 318
 319         /* record this tasks comm */
 320         tracing_record_cmdline(current);
 321 }
 322
 323 /**
 324  * trace_seq_printf - sequence printing of trace information
 325  * @s: trace sequence descriptor
 326  * @fmt: printf format string
 327  *
 328  * The tracer may use either sequence operations or its own
 329  * copy to user routines. To simplify formating of a trace
 330  * trace_seq_printf is used to store strings into a special
 331  * buffer (@s). Then the output may be either used by
 332  * the sequencer or pulled into another buffer.
 333  */
 334 int
 335 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 336 {
 337         int len = (PAGE_SIZE - 1) - s->len;
 338         va_list ap;
 339         int ret;
 340
 341         if (!len)
 342                 return 0;
 343
 344         va_start(ap, fmt);
 345         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 346         va_end(ap);
 347
 348         /* If we can't write it all, don't bother writing anything */
 349         if (ret >= len)
 350                 return 0;
 351
 352         s->len += ret;
 353
 354         return len;
 355 }
 356
 357 /**
 358  * trace_seq_puts - trace sequence printing of simple string
 359  * @s: trace sequence descriptor
 360  * @str: simple string to record
 361  *
 362  * The tracer may use either the sequence operations or its own
 363  * copy to user routines. This function records a simple string
 364  * into a special buffer (@s) for later retrieval by a sequencer
 365  * or other mechanism.
 366  */
 367 static int
 368 trace_seq_puts(struct trace_seq *s, const char *str)
 369 {
 370         int len = strlen(str);
 371
 372         if (len > ((PAGE_SIZE - 1) - s->len))
 373                 return 0;
 374
 375         memcpy(s->buffer + s->len, str, len);
 376         s->len += len;
 377
 378         return len;
 379 }
 380
 381 static int
 382 trace_seq_putc(struct trace_seq *s, unsigned char c)
 383 {
 384         if (s->len >= (PAGE_SIZE - 1))
 385                 return 0;
 386
 387         s->buffer[s->len++] = c;
 388
 389         return 1;
 390 }
 391
 392 static int
 393 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 394 {
 395         if (len > ((PAGE_SIZE - 1) - s->len))
 396                 return 0;
 397
 398         memcpy(s->buffer + s->len, mem, len);
 399         s->len += len;
 400
 401         return len;
 402 }
 403
 404 #define MAX_MEMHEX_BYTES        8
 405 #define HEX_CHARS               (MAX_MEMHEX_BYTES*2 + 1)
 406
 407 static int
 408 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 409 {
 410         unsigned char hex[HEX_CHARS];
 411         unsigned char *data = mem;
 412         int i, j;
 413
 414 #ifdef __BIG_ENDIAN
 415         for (i = 0, j = 0; i < len; i++) {
 416 #else
 417         for (i = len-1, j = 0; i >= 0; i--) {
 418 #endif
 419                 hex[j++] = hex_asc_hi(data[i]);
 420                 hex[j++] = hex_asc_lo(data[i]);
 421         }
 422         hex[j++] = ' ';
 423
 424         return trace_seq_putmem(s, hex, j);
 425 }
 426
 427 static int
 428 trace_seq_path(struct trace_seq *s, struct path *path)
 429 {
 430         unsigned char *p;
 431
 432         if (s->len >= (PAGE_SIZE - 1))
 433                 return 0;
 434         p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
 435         if (!IS_ERR(p)) {
 436                 p = mangle_path(s->buffer + s->len, p, "\n");
 437                 if (p) {
 438                         s->len = p - s->buffer;
 439                         return 1;
 440                 }
 441         } else {
 442                 s->buffer[s->len++] = '?';
 443                 return 1;
 444         }
 445
 446         return 0;
 447 }
 448
 449 static void
 450 trace_seq_reset(struct trace_seq *s)
 451 {
 452         s->len = 0;
 453         s->readpos = 0;
 454 }
 455
 456 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
 457 {
 458         int len;
 459         int ret;
 460
 461         if (s->len <= s->readpos)
 462                 return -EBUSY;
 463
 464         len = s->len - s->readpos;
 465         if (cnt > len)
 466                 cnt = len;
 467         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
 468         if (ret)
 469                 return -EFAULT;
 470
 471         s->readpos += len;
 472         return cnt;
 473 }
 474
 475 static void
 476 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 477 {
 478         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 479
 480         s->buffer[len] = 0;
 481         seq_puts(m, s->buffer);
 482
 483         trace_seq_reset(s);
 484 }
 485
 486 /**
 487  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
 488  * @tr: tracer
 489  * @tsk: the task with the latency
 490  * @cpu: The cpu that initiated the trace.
 491  *
 492  * Flip the buffers between the @tr and the max_tr and record information
 493  * about which task was the cause of this latency.
 494  */
 495 void
 496 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 497 {
 498         struct ring_buffer *buf = tr->buffer;
 499
 500         WARN_ON_ONCE(!irqs_disabled());
 501         __raw_spin_lock(&ftrace_max_lock);
 502
 503         tr->buffer = max_tr.buffer;
 504         max_tr.buffer = buf;
 505
 506         ftrace_disable_cpu();
 507         ring_buffer_reset(tr->buffer);
 508         ftrace_enable_cpu();
 509
 510         __update_max_tr(tr, tsk, cpu);
 511         __raw_spin_unlock(&ftrace_max_lock);
 512 }
 513
 514 /**
 515  * update_max_tr_single - only copy one trace over, and reset the rest
 516  * @tr - tracer
 517  * @tsk - task with the latency
 518  * @cpu - the cpu of the buffer to copy.
 519  *
 520  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
 521  */
 522 void
 523 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 524 {
 525         int ret;
 526
 527         WARN_ON_ONCE(!irqs_disabled());
 528         __raw_spin_lock(&ftrace_max_lock);
 529
 530         ftrace_disable_cpu();
 531
 532         ring_buffer_reset(max_tr.buffer);
 533         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
 534
 535         ftrace_enable_cpu();
 536
 537         WARN_ON_ONCE(ret);
 538
 539         __update_max_tr(tr, tsk, cpu);
 540         __raw_spin_unlock(&ftrace_max_lock);
 541 }
 542
 543 /**
 544  * register_tracer - register a tracer with the ftrace system.
 545  * @type - the plugin for the tracer
 546  *
 547  * Register a new plugin tracer.
 548  */
 549 int register_tracer(struct tracer *type)
 550 {
 551         struct tracer *t;
 552         int len;
 553         int ret = 0;
 554
 555         if (!type->name) {
 556                 pr_info("Tracer must have a name\n");
 557                 return -1;
 558         }
 559
 560         /*
 561          * When this gets called we hold the BKL which means that
 562          * preemption is disabled. Various trace selftests however
 563          * need to disable and enable preemption for successful tests.
 564          * So we drop the BKL here and grab it after the tests again.
 565          */
 566         unlock_kernel();
 567         mutex_lock(&trace_types_lock);
 568
 569         for (t = trace_types; t; t = t->next) {
 570                 if (strcmp(type->name, t->name) == 0) {
 571                         /* already found */
 572                         pr_info("Trace %s already registered\n",
 573                                 type->name);
 574                         ret = -1;
 575                         goto out;
 576                 }
 577         }
 578
 579         if (!type->set_flag)
 580                 type->set_flag = &dummy_set_flag;
 581         if (!type->flags)
 582                 type->flags = &dummy_tracer_flags;
 583         else
 584                 if (!type->flags->opts)
 585                         type->flags->opts = dummy_tracer_opt;
 586
 587 #ifdef CONFIG_FTRACE_STARTUP_TEST
 588         if (type->selftest) {
 589                 struct tracer *saved_tracer = current_trace;
 590                 struct trace_array *tr = &global_trace;
 591                 int i;
 592                 /*
 593                  * Run a selftest on this tracer.
 594                  * Here we reset the trace buffer, and set the current
 595                  * tracer to be this tracer. The tracer can then run some
 596                  * internal tracing to verify that everything is in order.
 597                  * If we fail, we do not register this tracer.
 598                  */
 599                 for_each_tracing_cpu(i)
 600                         tracing_reset(tr, i);
 601
 602                 current_trace = type;
 603                 /* the test is responsible for initializing and enabling */
 604                 pr_info("Testing tracer %s: ", type->name);
 605                 ret = type->selftest(type, tr);
 606                 /* the test is responsible for resetting too */
 607                 current_trace = saved_tracer;
 608                 if (ret) {
 609                         printk(KERN_CONT "FAILED!\n");
 610                         goto out;
 611                 }
 612                 /* Only reset on passing, to avoid touching corrupted buffers */
 613                 for_each_tracing_cpu(i)
 614                         tracing_reset(tr, i);
 615
 616                 printk(KERN_CONT "PASSED\n");
 617         }
 618 #endif
 619
 620         type->next = trace_types;
 621         trace_types = type;
 622         len = strlen(type->name);
 623         if (len > max_tracer_type_len)
 624                 max_tracer_type_len = len;
 625
 626  out:
 627         mutex_unlock(&trace_types_lock);
 628         lock_kernel();
 629
 630         return ret;
 631 }
 632
 633 void unregister_tracer(struct tracer *type)
 634 {
 635         struct tracer **t;
 636         int len;
 637
 638         mutex_lock(&trace_types_lock);
 639         for (t = &trace_types; *t; t = &(*t)->next) {
 640                 if (*t == type)
 641                         goto found;
 642         }
 643         pr_info("Trace %s not registered\n", type->name);
 644         goto out;
 645
 646  found:
 647         *t = (*t)->next;
 648         if (strlen(type->name) != max_tracer_type_len)
 649                 goto out;
 650
 651         max_tracer_type_len = 0;
 652         for (t = &trace_types; *t; t = &(*t)->next) {
 653                 len = strlen((*t)->name);
 654                 if (len > max_tracer_type_len)
 655                         max_tracer_type_len = len;
 656         }
 657  out:
 658         mutex_unlock(&trace_types_lock);
 659 }
 660
 661 void tracing_reset(struct trace_array *tr, int cpu)
 662 {
 663         ftrace_disable_cpu();
 664         ring_buffer_reset_cpu(tr->buffer, cpu);
 665         ftrace_enable_cpu();
 666 }
 667
 668 #define SAVED_CMDLINES 128
 669 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 670 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 671 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 672 static int cmdline_idx;
 673 static DEFINE_SPINLOCK(trace_cmdline_lock);
 674
 675 /* temporary disable recording */
 676 atomic_t trace_record_cmdline_disabled __read_mostly;
 677
 678 static void trace_init_cmdlines(void)
 679 {
 680         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 681         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 682         cmdline_idx = 0;
 683 }
 684
 685 static int trace_stop_count;
 686 static DEFINE_SPINLOCK(tracing_start_lock);
 687
 688 /**
 689  * ftrace_off_permanent - disable all ftrace code permanently
 690  *
 691  * This should only be called when a serious anomally has
 692  * been detected.  This will turn off the function tracing,
 693  * ring buffers, and other tracing utilites. It takes no
 694  * locks and can be called from any context.
 695  */
 696 void ftrace_off_permanent(void)
 697 {
 698         tracing_disabled = 1;
 699         ftrace_stop();
 700         tracing_off_permanent();
 701 }
 702
 703 /**
 704  * tracing_start - quick start of the tracer
 705  *
 706  * If tracing is enabled but was stopped by tracing_stop,
 707  * this will start the tracer back up.
 708  */
 709 void tracing_start(void)
 710 {
 711         struct ring_buffer *buffer;
 712         unsigned long flags;
 713
 714         if (tracing_disabled)
 715                 return;
 716
 717         spin_lock_irqsave(&tracing_start_lock, flags);
 718         if (--trace_stop_count)
 719                 goto out;
 720
 721         if (trace_stop_count < 0) {
 722                 /* Someone screwed up their debugging */
 723                 WARN_ON_ONCE(1);
 724                 trace_stop_count = 0;
 725                 goto out;
 726         }
 727
 728
 729         buffer = global_trace.buffer;
 730         if (buffer)
 731                 ring_buffer_record_enable(buffer);
 732
 733         buffer = max_tr.buffer;
 734         if (buffer)
 735                 ring_buffer_record_enable(buffer);
 736
 737         ftrace_start();
 738  out:
 739         spin_unlock_irqrestore(&tracing_start_lock, flags);
 740 }
 741
 742 /**
 743  * tracing_stop - quick stop of the tracer
 744  *
 745  * Light weight way to stop tracing. Use in conjunction with
 746  * tracing_start.
 747  */
 748 void tracing_stop(void)
 749 {
 750         struct ring_buffer *buffer;
 751         unsigned long flags;
 752
 753         ftrace_stop();
 754         spin_lock_irqsave(&tracing_start_lock, flags);
 755         if (trace_stop_count++)
 756                 goto out;
 757
 758         buffer = global_trace.buffer;
 759         if (buffer)
 760                 ring_buffer_record_disable(buffer);
 761
 762         buffer = max_tr.buffer;
 763         if (buffer)
 764                 ring_buffer_record_disable(buffer);
 765
 766  out:
 767         spin_unlock_irqrestore(&tracing_start_lock, flags);
 768 }
 769
 770 void trace_stop_cmdline_recording(void);
 771
 772 static void trace_save_cmdline(struct task_struct *tsk)
 773 {
 774         unsigned map;
 775         unsigned idx;
 776
 777         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 778                 return;
 779
 780         /*
 781          * It's not the end of the world if we don't get
 782          * the lock, but we also don't want to spin
 783          * nor do we want to disable interrupts,
 784          * so if we miss here, then better luck next time.
 785          */
 786         if (!spin_trylock(&trace_cmdline_lock))
 787                 return;
 788
 789         idx = map_pid_to_cmdline[tsk->pid];
 790         if (idx >= SAVED_CMDLINES) {
 791                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 792
 793                 map = map_cmdline_to_pid[idx];
 794                 if (map <= PID_MAX_DEFAULT)
 795                         map_pid_to_cmdline[map] = (unsigned)-1;
 796
 797                 map_pid_to_cmdline[tsk->pid] = idx;
 798
 799                 cmdline_idx = idx;
 800         }
 801
 802         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 803
 804         spin_unlock(&trace_cmdline_lock);
 805 }
 806
 807 char *trace_find_cmdline(int pid)
 808 {
 809         char *cmdline = "<...>";
 810         unsigned map;
 811
 812         if (!pid)
 813                 return "<idle>";
 814
 815         if (pid > PID_MAX_DEFAULT)
 816                 goto out;
 817
 818         map = map_pid_to_cmdline[pid];
 819         if (map >= SAVED_CMDLINES)
 820                 goto out;
 821
 822         cmdline = saved_cmdlines[map];
 823
 824  out:
 825         return cmdline;
 826 }
 827
 828 void tracing_record_cmdline(struct task_struct *tsk)
 829 {
 830         if (atomic_read(&trace_record_cmdline_disabled))
 831                 return;
 832
 833         trace_save_cmdline(tsk);
 834 }
 835
 836 void
 837 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
 838                              int pc)
 839 {
 840         struct task_struct *tsk = current;
 841
 842         entry->preempt_count            = pc & 0xff;
 843         entry->pid                      = (tsk) ? tsk->pid : 0;
 844         entry->tgid                     = (tsk) ? tsk->tgid : 0;
 845         entry->flags =
 846 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
 847                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 848 #else
 849                 TRACE_FLAG_IRQS_NOSUPPORT |
 850 #endif
 851                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 852                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 853                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 854 }
 855
 856 void
 857 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 858                unsigned long ip, unsigned long parent_ip, unsigned long flags,
 859                int pc)
 860 {
 861         struct ring_buffer_event *event;
 862         struct ftrace_entry *entry;
 863         unsigned long irq_flags;
 864
 865         /* If we are reading the ring buffer, don't trace */
 866         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 867                 return;
 868
 869         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 870                                          &irq_flags);
 871         if (!event)
 872                 return;
 873         entry   = ring_buffer_event_data(event);
 874         tracing_generic_entry_update(&entry->ent, flags, pc);
 875         entry->ent.type                 = TRACE_FN;
 876         entry->ip                       = ip;
 877         entry->parent_ip                = parent_ip;
 878         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 879 }
 880
 881 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 882 static void __trace_graph_entry(struct trace_array *tr,
 883                                 struct trace_array_cpu *data,
 884                                 struct ftrace_graph_ent *trace,
 885                                 unsigned long flags,
 886                                 int pc)
 887 {
 888         struct ring_buffer_event *event;
 889         struct ftrace_graph_ent_entry *entry;
 890         unsigned long irq_flags;
 891
 892         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 893                 return;
 894
 895         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
 896                                          &irq_flags);
 897         if (!event)
 898                 return;
 899         entry   = ring_buffer_event_data(event);
 900         tracing_generic_entry_update(&entry->ent, flags, pc);
 901         entry->ent.type                 = TRACE_GRAPH_ENT;
 902         entry->graph_ent                        = *trace;
 903         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
 904 }
 905
 906 static void __trace_graph_return(struct trace_array *tr,
 907                                 struct trace_array_cpu *data,
 908                                 struct ftrace_graph_ret *trace,
 909                                 unsigned long flags,
 910                                 int pc)
 911 {
 912         struct ring_buffer_event *event;
 913         struct ftrace_graph_ret_entry *entry;
 914         unsigned long irq_flags;
 915
 916         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 917                 return;
 918
 919         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
 920                                          &irq_flags);
 921         if (!event)
 922                 return;
 923         entry   = ring_buffer_event_data(event);
 924         tracing_generic_entry_update(&entry->ent, flags, pc);
 925         entry->ent.type                 = TRACE_GRAPH_RET;
 926         entry->ret                              = *trace;
 927         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
 928 }
 929 #endif
 930
 931 void
 932 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 933        unsigned long ip, unsigned long parent_ip, unsigned long flags,
 934        int pc)
 935 {
 936         if (likely(!atomic_read(&data->disabled)))
 937                 trace_function(tr, data, ip, parent_ip, flags, pc);
 938 }
 939
 940 static void ftrace_trace_stack(struct trace_array *tr,
 941                                struct trace_array_cpu *data,
 942                                unsigned long flags,
 943                                int skip, int pc)
 944 {
 945 #ifdef CONFIG_STACKTRACE
 946         struct ring_buffer_event *event;
 947         struct stack_entry *entry;
 948         struct stack_trace trace;
 949         unsigned long irq_flags;
 950
 951         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 952                 return;
 953
 954         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 955                                          &irq_flags);
 956         if (!event)
 957                 return;
 958         entry   = ring_buffer_event_data(event);
 959         tracing_generic_entry_update(&entry->ent, flags, pc);
 960         entry->ent.type         = TRACE_STACK;
 961
 962         memset(&entry->caller, 0, sizeof(entry->caller));
 963
 964         trace.nr_entries        = 0;
 965         trace.max_entries       = FTRACE_STACK_ENTRIES;
 966         trace.skip              = skip;
 967         trace.entries           = entry->caller;
 968
 969         save_stack_trace(&trace);
 970         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
 971 #endif
 972 }
 973
 974 void __trace_stack(struct trace_array *tr,
 975                    struct trace_array_cpu *data,
 976                    unsigned long flags,
 977                    int skip)
 978 {
 979         ftrace_trace_stack(tr, data, flags, skip, preempt_count());
 980 }
 981
 982 static void ftrace_trace_userstack(struct trace_array *tr,
 983                    struct trace_array_cpu *data,
 984                    unsigned long flags, int pc)
 985 {
 986 #ifdef CONFIG_STACKTRACE
 987         struct ring_buffer_event *event;
 988         struct userstack_entry *entry;
 989         struct stack_trace trace;
 990         unsigned long irq_flags;
 991
 992         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
 993                 return;
 994
 995         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
 996                                          &irq_flags);
 997         if (!event)
 998                 return;
 999         entry   = ring_buffer_event_data(event);
1000         tracing_generic_entry_update(&entry->ent, flags, pc);
1001         entry->ent.type         = TRACE_USER_STACK;
1002
1003         memset(&entry->caller, 0, sizeof(entry->caller));
1004
1005         trace.nr_entries        = 0;
1006         trace.max_entries       = FTRACE_STACK_ENTRIES;
1007         trace.skip              = 0;
1008         trace.entries           = entry->caller;
1009
1010         save_stack_trace_user(&trace);
1011         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1012 #endif
1013 }
1014
1015 void __trace_userstack(struct trace_array *tr,
1016                    struct trace_array_cpu *data,
1017                    unsigned long flags)
1018 {
1019         ftrace_trace_userstack(tr, data, flags, preempt_count());
1020 }
1021
1022 static void
1023 ftrace_trace_special(void *__tr, void *__data,
1024                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1025                      int pc)
1026 {
1027         struct ring_buffer_event *event;
1028         struct trace_array_cpu *data = __data;
1029         struct trace_array *tr = __tr;
1030         struct special_entry *entry;
1031         unsigned long irq_flags;
1032
1033         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1034                                          &irq_flags);
1035         if (!event)
1036                 return;
1037         entry   = ring_buffer_event_data(event);
1038         tracing_generic_entry_update(&entry->ent, 0, pc);
1039         entry->ent.type                 = TRACE_SPECIAL;
1040         entry->arg1                     = arg1;
1041         entry->arg2                     = arg2;
1042         entry->arg3                     = arg3;
1043         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1044         ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1045         ftrace_trace_userstack(tr, data, irq_flags, pc);
1046
1047         trace_wake_up();
1048 }
1049
1050 void
1051 __trace_special(void *__tr, void *__data,
1052                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1053 {
1054         ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1055 }
1056
1057 void
1058 tracing_sched_switch_trace(struct trace_array *tr,
1059                            struct trace_array_cpu *data,
1060                            struct task_struct *prev,
1061                            struct task_struct *next,
1062                            unsigned long flags, int pc)
1063 {
1064         struct ring_buffer_event *event;
1065         struct ctx_switch_entry *entry;
1066         unsigned long irq_flags;
1067
1068         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1069                                            &irq_flags);
1070         if (!event)
1071                 return;
1072         entry   = ring_buffer_event_data(event);
1073         tracing_generic_entry_update(&entry->ent, flags, pc);
1074         entry->ent.type                 = TRACE_CTX;
1075         entry->prev_pid                 = prev->pid;
1076         entry->prev_prio                = prev->prio;
1077         entry->prev_state               = prev->state;
1078         entry->next_pid                 = next->pid;
1079         entry->next_prio                = next->prio;
1080         entry->next_state               = next->state;
1081         entry->next_cpu = task_cpu(next);
1082         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1083         ftrace_trace_stack(tr, data, flags, 5, pc);
1084         ftrace_trace_userstack(tr, data, flags, pc);
1085 }
1086
1087 void
1088 tracing_sched_wakeup_trace(struct trace_array *tr,
1089                            struct trace_array_cpu *data,
1090                            struct task_struct *wakee,
1091                            struct task_struct *curr,
1092                            unsigned long flags, int pc)
1093 {
1094         struct ring_buffer_event *event;
1095         struct ctx_switch_entry *entry;
1096         unsigned long irq_flags;
1097
1098         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1099                                            &irq_flags);
1100         if (!event)
1101                 return;
1102         entry   = ring_buffer_event_data(event);
1103         tracing_generic_entry_update(&entry->ent, flags, pc);
1104         entry->ent.type                 = TRACE_WAKE;
1105         entry->prev_pid                 = curr->pid;
1106         entry->prev_prio                = curr->prio;
1107         entry->prev_state               = curr->state;
1108         entry->next_pid                 = wakee->pid;
1109         entry->next_prio                = wakee->prio;
1110         entry->next_state               = wakee->state;
1111         entry->next_cpu                 = task_cpu(wakee);
1112         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1113         ftrace_trace_stack(tr, data, flags, 6, pc);
1114         ftrace_trace_userstack(tr, data, flags, pc);
1115
1116         trace_wake_up();
1117 }
1118
1119 void
1120 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1121 {
1122         struct trace_array *tr = &global_trace;
1123         struct trace_array_cpu *data;
1124         unsigned long flags;
1125         int cpu;
1126         int pc;
1127
1128         if (tracing_disabled)
1129                 return;
1130
1131         pc = preempt_count();
1132         local_irq_save(flags);
1133         cpu = raw_smp_processor_id();
1134         data = tr->data[cpu];
1135
1136         if (likely(atomic_inc_return(&data->disabled) == 1))
1137                 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1138
1139         atomic_dec(&data->disabled);
1140         local_irq_restore(flags);
1141 }
1142
1143 #ifdef CONFIG_FUNCTION_TRACER
1144 static void
1145 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1146 {
1147         struct trace_array *tr = &global_trace;
1148         struct trace_array_cpu *data;
1149         unsigned long flags;
1150         long disabled;
1151         int cpu, resched;
1152         int pc;
1153
1154         if (unlikely(!ftrace_function_enabled))
1155                 return;
1156
1157         pc = preempt_count();
1158         resched = ftrace_preempt_disable();
1159         local_save_flags(flags);
1160         cpu = raw_smp_processor_id();
1161         data = tr->data[cpu];
1162         disabled = atomic_inc_return(&data->disabled);
1163
1164         if (likely(disabled == 1))
1165                 trace_function(tr, data, ip, parent_ip, flags, pc);
1166
1167         atomic_dec(&data->disabled);
1168         ftrace_preempt_enable(resched);
1169 }
1170
1171 static void
1172 function_trace_call(unsigned long ip, unsigned long parent_ip)
1173 {
1174         struct trace_array *tr = &global_trace;
1175         struct trace_array_cpu *data;
1176         unsigned long flags;
1177         long disabled;
1178         int cpu;
1179         int pc;
1180
1181         if (unlikely(!ftrace_function_enabled))
1182                 return;
1183
1184         /*
1185          * Need to use raw, since this must be called before the
1186          * recursive protection is performed.
1187          */
1188         local_irq_save(flags);
1189         cpu = raw_smp_processor_id();
1190         data = tr->data[cpu];
1191         disabled = atomic_inc_return(&data->disabled);
1192
1193         if (likely(disabled == 1)) {
1194                 pc = preempt_count();
1195                 trace_function(tr, data, ip, parent_ip, flags, pc);
1196         }
1197
1198         atomic_dec(&data->disabled);
1199         local_irq_restore(flags);
1200 }
1201
1202 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1203 int trace_graph_entry(struct ftrace_graph_ent *trace)
1204 {
1205         struct trace_array *tr = &global_trace;
1206         struct trace_array_cpu *data;
1207         unsigned long flags;
1208         long disabled;
1209         int cpu;
1210         int pc;
1211
1212         local_irq_save(flags);
1213         cpu = raw_smp_processor_id();
1214         data = tr->data[cpu];
1215         disabled = atomic_inc_return(&data->disabled);
1216         if (likely(disabled == 1)) {
1217                 pc = preempt_count();
1218                 __trace_graph_entry(tr, data, trace, flags, pc);
1219         }
1220         atomic_dec(&data->disabled);
1221         local_irq_restore(flags);
1222
1223         return 1;
1224 }
1225
1226 void trace_graph_return(struct ftrace_graph_ret *trace)
1227 {
1228         struct trace_array *tr = &global_trace;
1229         struct trace_array_cpu *data;
1230         unsigned long flags;
1231         long disabled;
1232         int cpu;
1233         int pc;
1234
1235         local_irq_save(flags);
1236         cpu = raw_smp_processor_id();
1237         data = tr->data[cpu];
1238         disabled = atomic_inc_return(&data->disabled);
1239         if (likely(disabled == 1)) {
1240                 pc = preempt_count();
1241                 __trace_graph_return(tr, data, trace, flags, pc);
1242         }
1243         atomic_dec(&data->disabled);
1244         local_irq_restore(flags);
1245 }
1246 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1247
1248 static struct ftrace_ops trace_ops __read_mostly =
1249 {
1250         .func = function_trace_call,
1251 };
1252
1253 void tracing_start_function_trace(void)
1254 {
1255         ftrace_function_enabled = 0;
1256
1257         if (trace_flags & TRACE_ITER_PREEMPTONLY)
1258                 trace_ops.func = function_trace_call_preempt_only;
1259         else
1260                 trace_ops.func = function_trace_call;
1261
1262         register_ftrace_function(&trace_ops);
1263         ftrace_function_enabled = 1;
1264 }
1265
1266 void tracing_stop_function_trace(void)
1267 {
1268         ftrace_function_enabled = 0;
1269         unregister_ftrace_function(&trace_ops);
1270 }
1271 #endif
1272
1273 enum trace_file_type {
1274         TRACE_FILE_LAT_FMT      = 1,
1275         TRACE_FILE_ANNOTATE     = 2,
1276 };
1277
1278 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1279 {
1280         /* Don't allow ftrace to trace into the ring buffers */
1281         ftrace_disable_cpu();
1282
1283         iter->idx++;
1284         if (iter->buffer_iter[iter->cpu])
1285                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1286
1287         ftrace_enable_cpu();
1288 }
1289
1290 static struct trace_entry *
1291 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1292 {
1293         struct ring_buffer_event *event;
1294         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1295
1296         /* Don't allow ftrace to trace into the ring buffers */
1297         ftrace_disable_cpu();
1298
1299         if (buf_iter)
1300                 event = ring_buffer_iter_peek(buf_iter, ts);
1301         else
1302                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1303
1304         ftrace_enable_cpu();
1305
1306         return event ? ring_buffer_event_data(event) : NULL;
1307 }
1308
1309 static struct trace_entry *
1310 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1311 {
1312         struct ring_buffer *buffer = iter->tr->buffer;
1313         struct trace_entry *ent, *next = NULL;
1314         u64 next_ts = 0, ts;
1315         int next_cpu = -1;
1316         int cpu;
1317
1318         for_each_tracing_cpu(cpu) {
1319
1320                 if (ring_buffer_empty_cpu(buffer, cpu))
1321                         continue;
1322
1323                 ent = peek_next_entry(iter, cpu, &ts);
1324
1325                 /*
1326                  * Pick the entry with the smallest timestamp:
1327                  */
1328                 if (ent && (!next || ts < next_ts)) {
1329                         next = ent;
1330                         next_cpu = cpu;
1331                         next_ts = ts;
1332                 }
1333         }
1334
1335         if (ent_cpu)
1336                 *ent_cpu = next_cpu;
1337
1338         if (ent_ts)
1339                 *ent_ts = next_ts;
1340
1341         return next;
1342 }
1343
1344 /* Find the next real entry, without updating the iterator itself */
1345 static struct trace_entry *
1346 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1347 {
1348         return __find_next_entry(iter, ent_cpu, ent_ts);
1349 }
1350
1351 /* Find the next real entry, and increment the iterator to the next entry */
1352 static void *find_next_entry_inc(struct trace_iterator *iter)
1353 {
1354         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1355
1356         if (iter->ent)
1357                 trace_iterator_increment(iter, iter->cpu);
1358
1359         return iter->ent ? iter : NULL;
1360 }
1361
1362 static void trace_consume(struct trace_iterator *iter)
1363 {
1364         /* Don't allow ftrace to trace into the ring buffers */
1365         ftrace_disable_cpu();
1366         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1367         ftrace_enable_cpu();
1368 }
1369
1370 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1371 {
1372         struct trace_iterator *iter = m->private;
1373         int i = (int)*pos;
1374         void *ent;
1375
1376         (*pos)++;
1377
1378         /* can't go backwards */
1379         if (iter->idx > i)
1380                 return NULL;
1381
1382         if (iter->idx < 0)
1383                 ent = find_next_entry_inc(iter);
1384         else
1385                 ent = iter;
1386
1387         while (ent && iter->idx < i)
1388                 ent = find_next_entry_inc(iter);
1389
1390         iter->pos = *pos;
1391
1392         return ent;
1393 }
1394
1395 static void *s_start(struct seq_file *m, loff_t *pos)
1396 {
1397         struct trace_iterator *iter = m->private;
1398         void *p = NULL;
1399         loff_t l = 0;
1400         int cpu;
1401
1402         mutex_lock(&trace_types_lock);
1403
1404         if (!current_trace || current_trace != iter->trace) {
1405                 mutex_unlock(&trace_types_lock);
1406                 return NULL;
1407         }
1408
1409         atomic_inc(&trace_record_cmdline_disabled);
1410
1411         if (*pos != iter->pos) {
1412                 iter->ent = NULL;
1413                 iter->cpu = 0;
1414                 iter->idx = -1;
1415
1416                 ftrace_disable_cpu();
1417
1418                 for_each_tracing_cpu(cpu) {
1419                         ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1420                 }
1421
1422                 ftrace_enable_cpu();
1423
1424                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1425                         ;
1426
1427         } else {
1428                 l = *pos - 1;
1429                 p = s_next(m, p, &l);
1430         }
1431
1432         return p;
1433 }
1434
1435 static void s_stop(struct seq_file *m, void *p)
1436 {
1437         atomic_dec(&trace_record_cmdline_disabled);
1438         mutex_unlock(&trace_types_lock);
1439 }
1440
1441 #ifdef CONFIG_KRETPROBES
1442 static inline const char *kretprobed(const char *name)
1443 {
1444         static const char tramp_name[] = "kretprobe_trampoline";
1445         int size = sizeof(tramp_name);
1446
1447         if (strncmp(tramp_name, name, size) == 0)
1448                 return "[unknown/kretprobe'd]";
1449         return name;
1450 }
1451 #else
1452 static inline const char *kretprobed(const char *name)
1453 {
1454         return name;
1455 }
1456 #endif /* CONFIG_KRETPROBES */
1457
1458 static int
1459 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1460 {
1461 #ifdef CONFIG_KALLSYMS
1462         char str[KSYM_SYMBOL_LEN];
1463         const char *name;
1464
1465         kallsyms_lookup(address, NULL, NULL, NULL, str);
1466
1467         name = kretprobed(str);
1468
1469         return trace_seq_printf(s, fmt, name);
1470 #endif
1471         return 1;
1472 }
1473
1474 static int
1475 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1476                      unsigned long address)
1477 {
1478 #ifdef CONFIG_KALLSYMS
1479         char str[KSYM_SYMBOL_LEN];
1480         const char *name;
1481
1482         sprint_symbol(str, address);
1483         name = kretprobed(str);
1484
1485         return trace_seq_printf(s, fmt, name);
1486 #endif
1487         return 1;
1488 }
1489
1490 #ifndef CONFIG_64BIT
1491 # define IP_FMT "%08lx"
1492 #else
1493 # define IP_FMT "%016lx"
1494 #endif
1495
1496 int
1497 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1498 {
1499         int ret;
1500
1501         if (!ip)
1502                 return trace_seq_printf(s, "0");
1503
1504         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1505                 ret = seq_print_sym_offset(s, "%s", ip);
1506         else
1507                 ret = seq_print_sym_short(s, "%s", ip);
1508
1509         if (!ret)
1510                 return 0;
1511
1512         if (sym_flags & TRACE_ITER_SYM_ADDR)
1513                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1514         return ret;
1515 }
1516
1517 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1518                                     unsigned long ip, unsigned long sym_flags)
1519 {
1520         struct file *file = NULL;
1521         unsigned long vmstart = 0;
1522         int ret = 1;
1523
1524         if (mm) {
1525                 const struct vm_area_struct *vma;
1526
1527                 down_read(&mm->mmap_sem);
1528                 vma = find_vma(mm, ip);
1529                 if (vma) {
1530                         file = vma->vm_file;
1531                         vmstart = vma->vm_start;
1532                 }
1533                 if (file) {
1534                         ret = trace_seq_path(s, &file->f_path);
1535                         if (ret)
1536                                 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1537                 }
1538                 up_read(&mm->mmap_sem);
1539         }
1540         if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1541                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1542         return ret;
1543 }
1544
1545 static int
1546 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1547                       unsigned long sym_flags)
1548 {
1549         struct mm_struct *mm = NULL;
1550         int ret = 1;
1551         unsigned int i;
1552
1553         if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1554                 struct task_struct *task;
1555                 /*
1556                  * we do the lookup on the thread group leader,
1557                  * since individual threads might have already quit!
1558                  */
1559                 rcu_read_lock();
1560                 task = find_task_by_vpid(entry->ent.tgid);
1561                 if (task)
1562                         mm = get_task_mm(task);
1563                 rcu_read_unlock();
1564         }
1565
1566         for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1567                 unsigned long ip = entry->caller[i];
1568
1569                 if (ip == ULONG_MAX || !ret)
1570                         break;
1571                 if (i && ret)
1572                         ret = trace_seq_puts(s, " <- ");
1573                 if (!ip) {
1574                         if (ret)
1575                                 ret = trace_seq_puts(s, "??");
1576                         continue;
1577                 }
1578                 if (!ret)
1579                         break;
1580                 if (ret)
1581                         ret = seq_print_user_ip(s, mm, ip, sym_flags);
1582         }
1583
1584         if (mm)
1585                 mmput(mm);
1586         return ret;
1587 }
1588
1589 static void print_lat_help_header(struct seq_file *m)
1590 {
1591         seq_puts(m, "#                  _------=> CPU#            \n");
1592         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1593         seq_puts(m, "#                | / _----=> need-resched    \n");
1594         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1595         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1596         seq_puts(m, "#                |||| /                      \n");
1597         seq_puts(m, "#                |||||     delay             \n");
1598         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1599         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1600 }
1601
1602 static void print_func_help_header(struct seq_file *m)
1603 {
1604         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1605         seq_puts(m, "#              | |       |          |         |\n");
1606 }
1607
1608
1609 static void
1610 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1611 {
1612         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1613         struct trace_array *tr = iter->tr;
1614         struct trace_array_cpu *data = tr->data[tr->cpu];
1615         struct tracer *type = current_trace;
1616         unsigned long total;
1617         unsigned long entries;
1618         const char *name = "preemption";
1619
1620         if (type)
1621                 name = type->name;
1622
1623         entries = ring_buffer_entries(iter->tr->buffer);
1624         total = entries +
1625                 ring_buffer_overruns(iter->tr->buffer);
1626
1627         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1628                    name, UTS_RELEASE);
1629         seq_puts(m, "-----------------------------------"
1630                  "---------------------------------\n");
1631         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1632                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1633                    nsecs_to_usecs(data->saved_latency),
1634                    entries,
1635                    total,
1636                    tr->cpu,
1637 #if defined(CONFIG_PREEMPT_NONE)
1638                    "server",
1639 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1640                    "desktop",
1641 #elif defined(CONFIG_PREEMPT)
1642                    "preempt",
1643 #else
1644                    "unknown",
1645 #endif
1646                    /* These are reserved for later use */
1647                    0, 0, 0, 0);
1648 #ifdef CONFIG_SMP
1649         seq_printf(m, " #P:%d)\n", num_online_cpus());
1650 #else
1651         seq_puts(m, ")\n");
1652 #endif
1653         seq_puts(m, "    -----------------\n");
1654         seq_printf(m, "    | task: %.16s-%d "
1655                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1656                    data->comm, data->pid, data->uid, data->nice,
1657                    data->policy, data->rt_priority);
1658         seq_puts(m, "    -----------------\n");
1659
1660         if (data->critical_start) {
1661                 seq_puts(m, " => started at: ");
1662                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1663                 trace_print_seq(m, &iter->seq);
1664                 seq_puts(m, "\n => ended at:   ");
1665                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1666                 trace_print_seq(m, &iter->seq);
1667                 seq_puts(m, "\n");
1668         }
1669
1670         seq_puts(m, "\n");
1671 }
1672
1673 static void
1674 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1675 {
1676         int hardirq, softirq;
1677         char *comm;
1678
1679         comm = trace_find_cmdline(entry->pid);
1680
1681         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1682         trace_seq_printf(s, "%3d", cpu);
1683         trace_seq_printf(s, "%c%c",
1684                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1685                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1686                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1687
1688         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1689         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1690         if (hardirq && softirq) {
1691                 trace_seq_putc(s, 'H');
1692         } else {
1693                 if (hardirq) {
1694                         trace_seq_putc(s, 'h');
1695                 } else {
1696                         if (softirq)
1697                                 trace_seq_putc(s, 's');
1698                         else
1699                                 trace_seq_putc(s, '.');
1700                 }
1701         }
1702
1703         if (entry->preempt_count)
1704                 trace_seq_printf(s, "%x", entry->preempt_count);
1705         else
1706                 trace_seq_puts(s, ".");
1707 }
1708
1709 unsigned long preempt_mark_thresh = 100;
1710
1711 static void
1712 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1713                     unsigned long rel_usecs)
1714 {
1715         trace_seq_printf(s, " %4lldus", abs_usecs);
1716         if (rel_usecs > preempt_mark_thresh)
1717                 trace_seq_puts(s, "!: ");
1718         else if (rel_usecs > 1)
1719                 trace_seq_puts(s, "+: ");
1720         else
1721                 trace_seq_puts(s, " : ");
1722 }
1723
1724 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1725
1726 /*
1727  * The message is supposed to contain an ending newline.
1728  * If the printing stops prematurely, try to add a newline of our own.
1729  */
1730 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1731 {
1732         struct trace_entry *ent;
1733         struct trace_field_cont *cont;
1734         bool ok = true;
1735
1736         ent = peek_next_entry(iter, iter->cpu, NULL);
1737         if (!ent || ent->type != TRACE_CONT) {
1738                 trace_seq_putc(s, '\n');
1739                 return;
1740         }
1741
1742         do {
1743                 cont = (struct trace_field_cont *)ent;
1744                 if (ok)
1745                         ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1746
1747                 ftrace_disable_cpu();
1748
1749                 if (iter->buffer_iter[iter->cpu])
1750                         ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1751                 else
1752                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1753
1754                 ftrace_enable_cpu();
1755
1756                 ent = peek_next_entry(iter, iter->cpu, NULL);
1757         } while (ent && ent->type == TRACE_CONT);
1758
1759         if (!ok)
1760                 trace_seq_putc(s, '\n');
1761 }
1762
1763 static void test_cpu_buff_start(struct trace_iterator *iter)
1764 {
1765         struct trace_seq *s = &iter->seq;
1766
1767         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1768                 return;
1769
1770         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1771                 return;
1772
1773         if (cpu_isset(iter->cpu, iter->started))
1774                 return;
1775
1776         cpu_set(iter->cpu, iter->started);
1777         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1778 }
1779
1780 static enum print_line_t
1781 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1782 {
1783         struct trace_seq *s = &iter->seq;
1784         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1785         struct trace_entry *next_entry;
1786         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1787         struct trace_entry *entry = iter->ent;
1788         unsigned long abs_usecs;
1789         unsigned long rel_usecs;
1790         u64 next_ts;
1791         char *comm;
1792         int S, T;
1793         int i;
1794         unsigned state;
1795
1796         if (entry->type == TRACE_CONT)
1797                 return TRACE_TYPE_HANDLED;
1798
1799         test_cpu_buff_start(iter);
1800
1801         next_entry = find_next_entry(iter, NULL, &next_ts);
1802         if (!next_entry)
1803                 next_ts = iter->ts;
1804         rel_usecs = ns2usecs(next_ts - iter->ts);
1805         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1806
1807         if (verbose) {
1808                 comm = trace_find_cmdline(entry->pid);
1809                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1810                                  " %ld.%03ldms (+%ld.%03ldms): ",
1811                                  comm,
1812                                  entry->pid, cpu, entry->flags,
1813                                  entry->preempt_count, trace_idx,
1814                                  ns2usecs(iter->ts),
1815                                  abs_usecs/1000,
1816                                  abs_usecs % 1000, rel_usecs/1000,
1817                                  rel_usecs % 1000);
1818         } else {
1819                 lat_print_generic(s, entry, cpu);
1820                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1821         }
1822         switch (entry->type) {
1823         case TRACE_FN: {
1824                 struct ftrace_entry *field;
1825
1826                 trace_assign_type(field, entry);
1827
1828                 seq_print_ip_sym(s, field->ip, sym_flags);
1829                 trace_seq_puts(s, " (");
1830                 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1831                 trace_seq_puts(s, ")\n");
1832                 break;
1833         }
1834         case TRACE_CTX:
1835         case TRACE_WAKE: {
1836                 struct ctx_switch_entry *field;
1837
1838                 trace_assign_type(field, entry);
1839
1840                 T = field->next_state < sizeof(state_to_char) ?
1841                         state_to_char[field->next_state] : 'X';
1842
1843                 state = field->prev_state ?
1844                         __ffs(field->prev_state) + 1 : 0;
1845                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1846                 comm = trace_find_cmdline(field->next_pid);
1847                 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1848                                  field->prev_pid,
1849                                  field->prev_prio,
1850                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1851                                  field->next_cpu,
1852                                  field->next_pid,
1853                                  field->next_prio,
1854                                  T, comm);
1855                 break;
1856         }
1857         case TRACE_SPECIAL: {
1858                 struct special_entry *field;
1859
1860                 trace_assign_type(field, entry);
1861
1862                 trace_seq_printf(s, "# %ld %ld %ld\n",
1863                                  field->arg1,
1864                                  field->arg2,
1865                                  field->arg3);
1866                 break;
1867         }
1868         case TRACE_STACK: {
1869                 struct stack_entry *field;
1870
1871                 trace_assign_type(field, entry);
1872
1873                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1874                         if (i)
1875                                 trace_seq_puts(s, " <= ");
1876                         seq_print_ip_sym(s, field->caller[i], sym_flags);
1877                 }
1878                 trace_seq_puts(s, "\n");
1879                 break;
1880         }
1881         case TRACE_PRINT: {
1882                 struct print_entry *field;
1883
1884                 trace_assign_type(field, entry);
1885
1886                 seq_print_ip_sym(s, field->ip, sym_flags);
1887                 trace_seq_printf(s, ": %s", field->buf);
1888                 if (entry->flags & TRACE_FLAG_CONT)
1889                         trace_seq_print_cont(s, iter);
1890                 break;
1891         }
1892         case TRACE_BRANCH: {
1893                 struct trace_branch *field;
1894
1895                 trace_assign_type(field, entry);
1896
1897                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1898                                  field->correct ? "  ok  " : " MISS ",
1899                                  field->func,
1900                                  field->file,
1901                                  field->line);
1902                 break;
1903         }
1904         case TRACE_USER_STACK: {
1905                 struct userstack_entry *field;
1906
1907                 trace_assign_type(field, entry);
1908
1909                 seq_print_userip_objs(field, s, sym_flags);
1910                 trace_seq_putc(s, '\n');
1911                 break;
1912         }
1913         default:
1914                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1915         }
1916         return TRACE_TYPE_HANDLED;
1917 }
1918
1919 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1920 {
1921         struct trace_seq *s = &iter->seq;
1922         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1923         struct trace_entry *entry;
1924         unsigned long usec_rem;
1925         unsigned long long t;
1926         unsigned long secs;
1927         char *comm;
1928         int ret;
1929         int S, T;
1930         int i;
1931
1932         entry = iter->ent;
1933
1934         if (entry->type == TRACE_CONT)
1935                 return TRACE_TYPE_HANDLED;
1936
1937         test_cpu_buff_start(iter);
1938
1939         comm = trace_find_cmdline(iter->ent->pid);
1940
1941         t = ns2usecs(iter->ts);
1942         usec_rem = do_div(t, 1000000ULL);
1943         secs = (unsigned long)t;
1944
1945         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1946         if (!ret)
1947                 return TRACE_TYPE_PARTIAL_LINE;
1948         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1949         if (!ret)
1950                 return TRACE_TYPE_PARTIAL_LINE;
1951         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1952         if (!ret)
1953                 return TRACE_TYPE_PARTIAL_LINE;
1954
1955         switch (entry->type) {
1956         case TRACE_FN: {
1957                 struct ftrace_entry *field;
1958
1959                 trace_assign_type(field, entry);
1960
1961                 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1962                 if (!ret)
1963                         return TRACE_TYPE_PARTIAL_LINE;
1964                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1965                                                 field->parent_ip) {
1966                         ret = trace_seq_printf(s, " <-");
1967                         if (!ret)
1968                                 return TRACE_TYPE_PARTIAL_LINE;
1969                         ret = seq_print_ip_sym(s,
1970                                                field->parent_ip,
1971                                                sym_flags);
1972                         if (!ret)
1973                                 return TRACE_TYPE_PARTIAL_LINE;
1974                 }
1975                 ret = trace_seq_printf(s, "\n");
1976                 if (!ret)
1977                         return TRACE_TYPE_PARTIAL_LINE;
1978                 break;
1979         }
1980         case TRACE_CTX:
1981         case TRACE_WAKE: {
1982                 struct ctx_switch_entry *field;
1983
1984                 trace_assign_type(field, entry);
1985
1986                 S = field->prev_state < sizeof(state_to_char) ?
1987                         state_to_char[field->prev_state] : 'X';
1988                 T = field->next_state < sizeof(state_to_char) ?
1989                         state_to_char[field->next_state] : 'X';
1990                 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1991                                        field->prev_pid,
1992                                        field->prev_prio,
1993                                        S,
1994                                        entry->type == TRACE_CTX ? "==>" : "  +",
1995                                        field->next_cpu,
1996                                        field->next_pid,
1997                                        field->next_prio,
1998                                        T);
1999                 if (!ret)
2000                         return TRACE_TYPE_PARTIAL_LINE;
2001                 break;
2002         }
2003         case TRACE_SPECIAL: {
2004                 struct special_entry *field;
2005
2006                 trace_assign_type(field, entry);
2007
2008                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2009                                  field->arg1,
2010                                  field->arg2,
2011                                  field->arg3);
2012                 if (!ret)
2013                         return TRACE_TYPE_PARTIAL_LINE;
2014                 break;
2015         }
2016         case TRACE_STACK: {
2017                 struct stack_entry *field;
2018
2019                 trace_assign_type(field, entry);
2020
2021                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2022                         if (i) {
2023                                 ret = trace_seq_puts(s, " <= ");
2024                                 if (!ret)
2025                                         return TRACE_TYPE_PARTIAL_LINE;
2026                         }
2027                         ret = seq_print_ip_sym(s, field->caller[i],
2028                                                sym_flags);
2029                         if (!ret)
2030                                 return TRACE_TYPE_PARTIAL_LINE;
2031                 }
2032                 ret = trace_seq_puts(s, "\n");
2033                 if (!ret)
2034                         return TRACE_TYPE_PARTIAL_LINE;
2035                 break;
2036         }
2037         case TRACE_PRINT: {
2038                 struct print_entry *field;
2039
2040                 trace_assign_type(field, entry);
2041
2042                 seq_print_ip_sym(s, field->ip, sym_flags);
2043                 trace_seq_printf(s, ": %s", field->buf);
2044                 if (entry->flags & TRACE_FLAG_CONT)
2045                         trace_seq_print_cont(s, iter);
2046                 break;
2047         }
2048         case TRACE_GRAPH_RET: {
2049                 return print_graph_function(iter);
2050         }
2051         case TRACE_GRAPH_ENT: {
2052                 return print_graph_function(iter);
2053         }
2054         case TRACE_BRANCH: {
2055                 struct trace_branch *field;
2056
2057                 trace_assign_type(field, entry);
2058
2059                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2060                                  field->correct ? "  ok  " : " MISS ",
2061                                  field->func,
2062                                  field->file,
2063                                  field->line);
2064                 break;
2065         }
2066         case TRACE_USER_STACK: {
2067                 struct userstack_entry *field;
2068
2069                 trace_assign_type(field, entry);
2070
2071                 ret = seq_print_userip_objs(field, s, sym_flags);
2072                 if (!ret)
2073                         return TRACE_TYPE_PARTIAL_LINE;
2074                 ret = trace_seq_putc(s, '\n');
2075                 if (!ret)
2076                         return TRACE_TYPE_PARTIAL_LINE;
2077                 break;
2078         }
2079         }
2080         return TRACE_TYPE_HANDLED;
2081 }
2082
2083 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2084 {
2085         struct trace_seq *s = &iter->seq;
2086         struct trace_entry *entry;
2087         int ret;
2088         int S, T;
2089
2090         entry = iter->ent;
2091
2092         if (entry->type == TRACE_CONT)
2093                 return TRACE_TYPE_HANDLED;
2094
2095         ret = trace_seq_printf(s, "%d %d %llu ",
2096                 entry->pid, iter->cpu, iter->ts);
2097         if (!ret)
2098                 return TRACE_TYPE_PARTIAL_LINE;
2099
2100         switch (entry->type) {
2101         case TRACE_FN: {
2102                 struct ftrace_entry *field;
2103
2104                 trace_assign_type(field, entry);
2105
2106                 ret = trace_seq_printf(s, "%x %x\n",
2107                                         field->ip,
2108                                         field->parent_ip);
2109                 if (!ret)
2110                         return TRACE_TYPE_PARTIAL_LINE;
2111                 break;
2112         }
2113         case TRACE_CTX:
2114         case TRACE_WAKE: {
2115                 struct ctx_switch_entry *field;
2116
2117                 trace_assign_type(field, entry);
2118
2119                 S = field->prev_state < sizeof(state_to_char) ?
2120                         state_to_char[field->prev_state] : 'X';
2121                 T = field->next_state < sizeof(state_to_char) ?
2122                         state_to_char[field->next_state] : 'X';
2123                 if (entry->type == TRACE_WAKE)
2124                         S = '+';
2125                 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2126                                        field->prev_pid,
2127                                        field->prev_prio,
2128                                        S,
2129                                        field->next_cpu,
2130                                        field->next_pid,
2131                                        field->next_prio,
2132                                        T);
2133                 if (!ret)
2134                         return TRACE_TYPE_PARTIAL_LINE;
2135                 break;
2136         }
2137         case TRACE_SPECIAL:
2138         case TRACE_USER_STACK:
2139         case TRACE_STACK: {
2140                 struct special_entry *field;
2141
2142                 trace_assign_type(field, entry);
2143
2144                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2145                                  field->arg1,
2146                                  field->arg2,
2147                                  field->arg3);
2148                 if (!ret)
2149                         return TRACE_TYPE_PARTIAL_LINE;
2150                 break;
2151         }
2152         case TRACE_PRINT: {
2153                 struct print_entry *field;
2154
2155                 trace_assign_type(field, entry);
2156
2157                 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2158                 if (entry->flags & TRACE_FLAG_CONT)
2159                         trace_seq_print_cont(s, iter);
2160                 break;
2161         }
2162         }
2163         return TRACE_TYPE_HANDLED;
2164 }
2165
2166 #define SEQ_PUT_FIELD_RET(s, x)                         \
2167 do {                                                    \
2168         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
2169                 return 0;                               \
2170 } while (0)
2171
2172 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
2173 do {                                                    \
2174         BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES);     \
2175         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
2176                 return 0;                               \
2177 } while (0)
2178
2179 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2180 {
2181         struct trace_seq *s = &iter->seq;
2182         unsigned char newline = '\n';
2183         struct trace_entry *entry;
2184         int S, T;
2185
2186         entry = iter->ent;
2187
2188         if (entry->type == TRACE_CONT)
2189                 return TRACE_TYPE_HANDLED;
2190
2191         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2192         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2193         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2194
2195         switch (entry->type) {
2196         case TRACE_FN: {
2197                 struct ftrace_entry *field;
2198
2199                 trace_assign_type(field, entry);
2200
2201                 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2202                 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2203                 break;
2204         }
2205         case TRACE_CTX:
2206         case TRACE_WAKE: {
2207                 struct ctx_switch_entry *field;
2208
2209                 trace_assign_type(field, entry);
2210
2211                 S = field->prev_state < sizeof(state_to_char) ?
2212                         state_to_char[field->prev_state] : 'X';
2213                 T = field->next_state < sizeof(state_to_char) ?
2214                         state_to_char[field->next_state] : 'X';
2215                 if (entry->type == TRACE_WAKE)
2216                         S = '+';
2217                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2218                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2219                 SEQ_PUT_HEX_FIELD_RET(s, S);
2220                 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2221                 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2222                 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2223                 SEQ_PUT_HEX_FIELD_RET(s, T);
2224                 break;
2225         }
2226         case TRACE_SPECIAL:
2227         case TRACE_USER_STACK:
2228         case TRACE_STACK: {
2229                 struct special_entry *field;
2230
2231                 trace_assign_type(field, entry);
2232
2233                 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2234                 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2235                 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2236                 break;
2237         }
2238         }
2239         SEQ_PUT_FIELD_RET(s, newline);
2240
2241         return TRACE_TYPE_HANDLED;
2242 }
2243
2244 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2245 {
2246         struct trace_seq *s = &iter->seq;
2247         struct trace_entry *entry;
2248
2249         entry = iter->ent;
2250
2251         if (entry->type == TRACE_CONT)
2252                 return TRACE_TYPE_HANDLED;
2253
2254         SEQ_PUT_FIELD_RET(s, entry->pid);
2255         SEQ_PUT_FIELD_RET(s, entry->cpu);
2256         SEQ_PUT_FIELD_RET(s, iter->ts);
2257
2258         switch (entry->type) {
2259         case TRACE_FN: {
2260                 struct ftrace_entry *field;
2261
2262                 trace_assign_type(field, entry);
2263
2264                 SEQ_PUT_FIELD_RET(s, field->ip);
2265                 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2266                 break;
2267         }
2268         case TRACE_CTX: {
2269                 struct ctx_switch_entry *field;
2270
2271                 trace_assign_type(field, entry);
2272
2273                 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2274                 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2275                 SEQ_PUT_FIELD_RET(s, field->prev_state);
2276                 SEQ_PUT_FIELD_RET(s, field->next_pid);
2277                 SEQ_PUT_FIELD_RET(s, field->next_prio);
2278                 SEQ_PUT_FIELD_RET(s, field->next_state);
2279                 break;
2280         }
2281         case TRACE_SPECIAL:
2282         case TRACE_USER_STACK:
2283         case TRACE_STACK: {
2284                 struct special_entry *field;
2285
2286                 trace_assign_type(field, entry);
2287
2288                 SEQ_PUT_FIELD_RET(s, field->arg1);
2289                 SEQ_PUT_FIELD_RET(s, field->arg2);
2290                 SEQ_PUT_FIELD_RET(s, field->arg3);
2291                 break;
2292         }
2293         }
2294         return 1;
2295 }
2296
2297 static int trace_empty(struct trace_iterator *iter)
2298 {
2299         int cpu;
2300
2301         for_each_tracing_cpu(cpu) {
2302                 if (iter->buffer_iter[cpu]) {
2303                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2304                                 return 0;
2305                 } else {
2306                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2307                                 return 0;
2308                 }
2309         }
2310
2311         return 1;
2312 }
2313
2314 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2315 {
2316         enum print_line_t ret;
2317
2318         if (iter->trace && iter->trace->print_line) {
2319                 ret = iter->trace->print_line(iter);
2320                 if (ret != TRACE_TYPE_UNHANDLED)
2321                         return ret;
2322         }
2323
2324         if (trace_flags & TRACE_ITER_BIN)
2325                 return print_bin_fmt(iter);
2326
2327         if (trace_flags & TRACE_ITER_HEX)
2328                 return print_hex_fmt(iter);
2329
2330         if (trace_flags & TRACE_ITER_RAW)
2331                 return print_raw_fmt(iter);
2332
2333         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2334                 return print_lat_fmt(iter, iter->idx, iter->cpu);
2335
2336         return print_trace_fmt(iter);
2337 }
2338
2339 static int s_show(struct seq_file *m, void *v)
2340 {
2341         struct trace_iterator *iter = v;
2342
2343         if (iter->ent == NULL) {
2344                 if (iter->tr) {
2345                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2346                         seq_puts(m, "#\n");
2347                 }
2348                 if (iter->trace && iter->trace->print_header)
2349                         iter->trace->print_header(m);
2350                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2351                         /* print nothing if the buffers are empty */
2352                         if (trace_empty(iter))
2353                                 return 0;
2354                         print_trace_header(m, iter);
2355                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2356                                 print_lat_help_header(m);
2357                 } else {
2358                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2359                                 print_func_help_header(m);
2360                 }
2361         } else {
2362                 print_trace_line(iter);
2363                 trace_print_seq(m, &iter->seq);
2364         }
2365
2366         return 0;
2367 }
2368
2369 static struct seq_operations tracer_seq_ops = {
2370         .start          = s_start,
2371         .next           = s_next,
2372         .stop           = s_stop,
2373         .show           = s_show,
2374 };
2375
2376 static struct trace_iterator *
2377 __tracing_open(struct inode *inode, struct file *file, int *ret)
2378 {
2379         struct trace_iterator *iter;
2380         struct seq_file *m;
2381         int cpu;
2382
2383         if (tracing_disabled) {
2384                 *ret = -ENODEV;
2385                 return NULL;
2386         }
2387
2388         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2389         if (!iter) {
2390                 *ret = -ENOMEM;
2391                 goto out;
2392         }
2393
2394         mutex_lock(&trace_types_lock);
2395         if (current_trace && current_trace->print_max)
2396                 iter->tr = &max_tr;
2397         else
2398                 iter->tr = inode->i_private;
2399         iter->trace = current_trace;
2400         iter->pos = -1;
2401
2402         /* Notify the tracer early; before we stop tracing. */
2403         if (iter->trace && iter->trace->open)
2404                         iter->trace->open(iter);
2405
2406         /* Annotate start of buffers if we had overruns */
2407         if (ring_buffer_overruns(iter->tr->buffer))
2408                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2409
2410
2411         for_each_tracing_cpu(cpu) {
2412
2413                 iter->buffer_iter[cpu] =
2414                         ring_buffer_read_start(iter->tr->buffer, cpu);
2415
2416                 if (!iter->buffer_iter[cpu])
2417                         goto fail_buffer;
2418         }
2419
2420         /* TODO stop tracer */
2421         *ret = seq_open(file, &tracer_seq_ops);
2422         if (*ret)
2423                 goto fail_buffer;
2424
2425         m = file->private_data;
2426         m->private = iter;
2427
2428         /* stop the trace while dumping */
2429         tracing_stop();
2430
2431         mutex_unlock(&trace_types_lock);
2432
2433  out:
2434         return iter;
2435
2436  fail_buffer:
2437         for_each_tracing_cpu(cpu) {
2438                 if (iter->buffer_iter[cpu])
2439                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2440         }
2441         mutex_unlock(&trace_types_lock);
2442         kfree(iter);
2443
2444         return ERR_PTR(-ENOMEM);
2445 }
2446
2447 int tracing_open_generic(struct inode *inode, struct file *filp)
2448 {
2449         if (tracing_disabled)
2450                 return -ENODEV;
2451
2452         filp->private_data = inode->i_private;
2453         return 0;
2454 }
2455
2456 int tracing_release(struct inode *inode, struct file *file)
2457 {
2458         struct seq_file *m = (struct seq_file *)file->private_data;
2459         struct trace_iterator *iter = m->private;
2460         int cpu;
2461
2462         mutex_lock(&trace_types_lock);
2463         for_each_tracing_cpu(cpu) {
2464                 if (iter->buffer_iter[cpu])
2465                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2466         }
2467
2468         if (iter->trace && iter->trace->close)
2469                 iter->trace->close(iter);
2470
2471         /* reenable tracing if it was previously enabled */
2472         tracing_start();
2473         mutex_unlock(&trace_types_lock);
2474
2475         seq_release(inode, file);
2476         kfree(iter);
2477         return 0;
2478 }
2479
2480 static int tracing_open(struct inode *inode, struct file *file)
2481 {
2482         int ret;
2483
2484         __tracing_open(inode, file, &ret);
2485
2486         return ret;
2487 }
2488
2489 static int tracing_lt_open(struct inode *inode, struct file *file)
2490 {
2491         struct trace_iterator *iter;
2492         int ret;
2493
2494         iter = __tracing_open(inode, file, &ret);
2495
2496         if (!ret)
2497                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2498
2499         return ret;
2500 }
2501
2502
2503 static void *
2504 t_next(struct seq_file *m, void *v, loff_t *pos)
2505 {
2506         struct tracer *t = m->private;
2507
2508         (*pos)++;
2509
2510         if (t)
2511                 t = t->next;
2512
2513         m->private = t;
2514
2515         return t;
2516 }
2517
2518 static void *t_start(struct seq_file *m, loff_t *pos)
2519 {
2520         struct tracer *t = m->private;
2521         loff_t l = 0;
2522
2523         mutex_lock(&trace_types_lock);
2524         for (; t && l < *pos; t = t_next(m, t, &l))
2525                 ;
2526
2527         return t;
2528 }
2529
2530 static void t_stop(struct seq_file *m, void *p)
2531 {
2532         mutex_unlock(&trace_types_lock);
2533 }
2534
2535 static int t_show(struct seq_file *m, void *v)
2536 {
2537         struct tracer *t = v;
2538
2539         if (!t)
2540                 return 0;
2541
2542         seq_printf(m, "%s", t->name);
2543         if (t->next)
2544                 seq_putc(m, ' ');
2545         else
2546                 seq_putc(m, '\n');
2547
2548         return 0;
2549 }
2550
2551 static struct seq_operations show_traces_seq_ops = {
2552         .start          = t_start,
2553         .next           = t_next,
2554         .stop           = t_stop,
2555         .show           = t_show,
2556 };
2557
2558 static int show_traces_open(struct inode *inode, struct file *file)
2559 {
2560         int ret;
2561
2562         if (tracing_disabled)
2563                 return -ENODEV;
2564
2565         ret = seq_open(file, &show_traces_seq_ops);
2566         if (!ret) {
2567                 struct seq_file *m = file->private_data;
2568                 m->private = trace_types;
2569         }
2570
2571         return ret;
2572 }
2573
2574 static struct file_operations tracing_fops = {
2575         .open           = tracing_open,
2576         .read           = seq_read,
2577         .llseek         = seq_lseek,
2578         .release        = tracing_release,
2579 };
2580
2581 static struct file_operations tracing_lt_fops = {
2582         .open           = tracing_lt_open,
2583         .read           = seq_read,
2584         .llseek         = seq_lseek,
2585         .release        = tracing_release,
2586 };
2587
2588 static struct file_operations show_traces_fops = {
2589         .open           = show_traces_open,
2590         .read           = seq_read,
2591         .release        = seq_release,
2592 };
2593
2594 /*
2595  * Only trace on a CPU if the bitmask is set:
2596  */
2597 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2598
2599 /*
2600  * When tracing/tracing_cpu_mask is modified then this holds
2601  * the new bitmask we are about to install:
2602  */
2603 static cpumask_t tracing_cpumask_new;
2604
2605 /*
2606  * The tracer itself will not take this lock, but still we want
2607  * to provide a consistent cpumask to user-space:
2608  */
2609 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2610
2611 /*
2612  * Temporary storage for the character representation of the
2613  * CPU bitmask (and one more byte for the newline):
2614  */
2615 static char mask_str[NR_CPUS + 1];
2616
2617 static ssize_t
2618 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2619                      size_t count, loff_t *ppos)
2620 {
2621         int len;
2622
2623         mutex_lock(&tracing_cpumask_update_lock);
2624
2625         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2626         if (count - len < 2) {
2627                 count = -EINVAL;
2628                 goto out_err;
2629         }
2630         len += sprintf(mask_str + len, "\n");
2631         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2632
2633 out_err:
2634         mutex_unlock(&tracing_cpumask_update_lock);
2635
2636         return count;
2637 }
2638
2639 static ssize_t
2640 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2641                       size_t count, loff_t *ppos)
2642 {
2643         int err, cpu;
2644
2645         mutex_lock(&tracing_cpumask_update_lock);
2646         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2647         if (err)
2648                 goto err_unlock;
2649
2650         local_irq_disable();
2651         __raw_spin_lock(&ftrace_max_lock);
2652         for_each_tracing_cpu(cpu) {
2653                 /*
2654                  * Increase/decrease the disabled counter if we are
2655                  * about to flip a bit in the cpumask:
2656                  */
2657                 if (cpu_isset(cpu, tracing_cpumask) &&
2658                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2659                         atomic_inc(&global_trace.data[cpu]->disabled);
2660                 }
2661                 if (!cpu_isset(cpu, tracing_cpumask) &&
2662                                 cpu_isset(cpu, tracing_cpumask_new)) {
2663                         atomic_dec(&global_trace.data[cpu]->disabled);
2664                 }
2665         }
2666         __raw_spin_unlock(&ftrace_max_lock);
2667         local_irq_enable();
2668
2669         tracing_cpumask = tracing_cpumask_new;
2670
2671         mutex_unlock(&tracing_cpumask_update_lock);
2672
2673         return count;
2674
2675 err_unlock:
2676         mutex_unlock(&tracing_cpumask_update_lock);
2677
2678         return err;
2679 }
2680
2681 static struct file_operations tracing_cpumask_fops = {
2682         .open           = tracing_open_generic,
2683         .read           = tracing_cpumask_read,
2684         .write          = tracing_cpumask_write,
2685 };
2686
2687 static ssize_t
2688 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2689                        size_t cnt, loff_t *ppos)
2690 {
2691         int i;
2692         char *buf;
2693         int r = 0;
2694         int len = 0;
2695         u32 tracer_flags = current_trace->flags->val;
2696         struct tracer_opt *trace_opts = current_trace->flags->opts;
2697
2698
2699         /* calulate max size */
2700         for (i = 0; trace_options[i]; i++) {
2701                 len += strlen(trace_options[i]);
2702                 len += 3; /* "no" and space */
2703         }
2704
2705         /*
2706          * Increase the size with names of options specific
2707          * of the current tracer.
2708          */
2709         for (i = 0; trace_opts[i].name; i++) {
2710                 len += strlen(trace_opts[i].name);
2711                 len += 3; /* "no" and space */
2712         }
2713
2714         /* +2 for \n and \0 */
2715         buf = kmalloc(len + 2, GFP_KERNEL);
2716         if (!buf)
2717                 return -ENOMEM;
2718
2719         for (i = 0; trace_options[i]; i++) {
2720                 if (trace_flags & (1 << i))
2721                         r += sprintf(buf + r, "%s ", trace_options[i]);
2722                 else
2723                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2724         }
2725
2726         for (i = 0; trace_opts[i].name; i++) {
2727                 if (tracer_flags & trace_opts[i].bit)
2728                         r += sprintf(buf + r, "%s ",
2729                                 trace_opts[i].name);
2730                 else
2731                         r += sprintf(buf + r, "no%s ",
2732                                 trace_opts[i].name);
2733         }
2734
2735         r += sprintf(buf + r, "\n");
2736         WARN_ON(r >= len + 2);
2737
2738         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2739
2740         kfree(buf);
2741
2742         return r;
2743 }
2744
2745 /* Try to assign a tracer specific option */
2746 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2747 {
2748         struct tracer_flags *trace_flags = trace->flags;
2749         struct tracer_opt *opts = NULL;
2750         int ret = 0, i = 0;
2751         int len;
2752
2753         for (i = 0; trace_flags->opts[i].name; i++) {
2754                 opts = &trace_flags->opts[i];
2755                 len = strlen(opts->name);
2756
2757                 if (strncmp(cmp, opts->name, len) == 0) {
2758                         ret = trace->set_flag(trace_flags->val,
2759                                 opts->bit, !neg);
2760                         break;
2761                 }
2762         }
2763         /* Not found */
2764         if (!trace_flags->opts[i].name)
2765                 return -EINVAL;
2766
2767         /* Refused to handle */
2768         if (ret)
2769                 return ret;
2770
2771         if (neg)
2772                 trace_flags->val &= ~opts->bit;
2773         else
2774                 trace_flags->val |= opts->bit;
2775
2776         return 0;
2777 }
2778
2779 static ssize_t
2780 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2781                         size_t cnt, loff_t *ppos)
2782 {
2783         char buf[64];
2784         char *cmp = buf;
2785         int neg = 0;
2786         int ret;
2787         int i;
2788
2789         if (cnt >= sizeof(buf))
2790                 return -EINVAL;
2791
2792         if (copy_from_user(&buf, ubuf, cnt))
2793                 return -EFAULT;
2794
2795         buf[cnt] = 0;
2796
2797         if (strncmp(buf, "no", 2) == 0) {
2798                 neg = 1;
2799                 cmp += 2;
2800         }
2801
2802         for (i = 0; trace_options[i]; i++) {
2803                 int len = strlen(trace_options[i]);
2804
2805                 if (strncmp(cmp, trace_options[i], len) == 0) {
2806                         if (neg)
2807                                 trace_flags &= ~(1 << i);
2808                         else
2809                                 trace_flags |= (1 << i);
2810                         break;
2811                 }
2812         }
2813
2814         /* If no option could be set, test the specific tracer options */
2815         if (!trace_options[i]) {
2816                 ret = set_tracer_option(current_trace, cmp, neg);
2817                 if (ret)
2818                         return ret;
2819         }
2820
2821         filp->f_pos += cnt;
2822
2823         return cnt;
2824 }
2825
2826 static struct file_operations tracing_iter_fops = {
2827         .open           = tracing_open_generic,
2828         .read           = tracing_trace_options_read,
2829         .write          = tracing_trace_options_write,
2830 };
2831
2832 static const char readme_msg[] =
2833         "tracing mini-HOWTO:\n\n"
2834         "# mkdir /debug\n"
2835         "# mount -t debugfs nodev /debug\n\n"
2836         "# cat /debug/tracing/available_tracers\n"
2837         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2838         "# cat /debug/tracing/current_tracer\n"
2839         "none\n"
2840         "# echo sched_switch > /debug/tracing/current_tracer\n"
2841         "# cat /debug/tracing/current_tracer\n"
2842         "sched_switch\n"
2843         "# cat /debug/tracing/trace_options\n"
2844         "noprint-parent nosym-offset nosym-addr noverbose\n"
2845         "# echo print-parent > /debug/tracing/trace_options\n"
2846         "# echo 1 > /debug/tracing/tracing_enabled\n"
2847         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2848         "echo 0 > /debug/tracing/tracing_enabled\n"
2849 ;
2850
2851 static ssize_t
2852 tracing_readme_read(struct file *filp, char __user *ubuf,
2853                        size_t cnt, loff_t *ppos)
2854 {
2855         return simple_read_from_buffer(ubuf, cnt, ppos,
2856                                         readme_msg, strlen(readme_msg));
2857 }
2858
2859 static struct file_operations tracing_readme_fops = {
2860         .open           = tracing_open_generic,
2861         .read           = tracing_readme_read,
2862 };
2863
2864 static ssize_t
2865 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2866                   size_t cnt, loff_t *ppos)
2867 {
2868         char buf[64];
2869         int r;
2870
2871         r = sprintf(buf, "%u\n", tracer_enabled);
2872         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2873 }
2874
2875 static ssize_t
2876 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2877                    size_t cnt, loff_t *ppos)
2878 {
2879         struct trace_array *tr = filp->private_data;
2880         char buf[64];
2881         long val;
2882         int ret;
2883
2884         if (cnt >= sizeof(buf))
2885                 return -EINVAL;
2886
2887         if (copy_from_user(&buf, ubuf, cnt))
2888                 return -EFAULT;
2889
2890         buf[cnt] = 0;
2891
2892         ret = strict_strtoul(buf, 10, &val);
2893         if (ret < 0)
2894                 return ret;
2895
2896         val = !!val;
2897
2898         mutex_lock(&trace_types_lock);
2899         if (tracer_enabled ^ val) {
2900                 if (val) {
2901                         tracer_enabled = 1;
2902                         if (current_trace->start)
2903                                 current_trace->start(tr);
2904                         tracing_start();
2905                 } else {
2906                         tracer_enabled = 0;
2907                         tracing_stop();
2908                         if (current_trace->stop)
2909                                 current_trace->stop(tr);
2910                 }
2911         }
2912         mutex_unlock(&trace_types_lock);
2913
2914         filp->f_pos += cnt;
2915
2916         return cnt;
2917 }
2918
2919 static ssize_t
2920 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2921                        size_t cnt, loff_t *ppos)
2922 {
2923         char buf[max_tracer_type_len+2];
2924         int r;
2925
2926         mutex_lock(&trace_types_lock);
2927         if (current_trace)
2928                 r = sprintf(buf, "%s\n", current_trace->name);
2929         else
2930                 r = sprintf(buf, "\n");
2931         mutex_unlock(&trace_types_lock);
2932
2933         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2934 }
2935
2936 static int tracing_set_tracer(char *buf)
2937 {
2938         struct trace_array *tr = &global_trace;
2939         struct tracer *t;
2940         int ret = 0;
2941
2942         mutex_lock(&trace_types_lock);
2943         for (t = trace_types; t; t = t->next) {
2944                 if (strcmp(t->name, buf) == 0)
2945                         break;
2946         }
2947         if (!t) {
2948                 ret = -EINVAL;
2949                 goto out;
2950         }
2951         if (t == current_trace)
2952                 goto out;
2953
2954         trace_branch_disable();
2955         if (current_trace && current_trace->reset)
2956                 current_trace->reset(tr);
2957
2958         current_trace = t;
2959         if (t->init) {
2960                 ret = t->init(tr);
2961                 if (ret)
2962                         goto out;
2963         }
2964
2965         trace_branch_enable(tr);
2966  out:
2967         mutex_unlock(&trace_types_lock);
2968
2969         return ret;
2970 }
2971
2972 static ssize_t
2973 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2974                         size_t cnt, loff_t *ppos)
2975 {
2976         char buf[max_tracer_type_len+1];
2977         int i;
2978         size_t ret;
2979         int err;
2980
2981         ret = cnt;
2982
2983         if (cnt > max_tracer_type_len)
2984                 cnt = max_tracer_type_len;
2985
2986         if (copy_from_user(&buf, ubuf, cnt))
2987                 return -EFAULT;
2988
2989         buf[cnt] = 0;
2990
2991         /* strip ending whitespace. */
2992         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2993                 buf[i] = 0;
2994
2995         err = tracing_set_tracer(buf);
2996         if (err)
2997                 return err;
2998
2999         filp->f_pos += ret;
3000
3001         return ret;
3002 }
3003
3004 static ssize_t
3005 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3006                      size_t cnt, loff_t *ppos)
3007 {
3008         unsigned long *ptr = filp->private_data;
3009         char buf[64];
3010         int r;
3011
3012         r = snprintf(buf, sizeof(buf), "%ld\n",
3013                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3014         if (r > sizeof(buf))
3015                 r = sizeof(buf);
3016         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3017 }
3018
3019 static ssize_t
3020 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3021                       size_t cnt, loff_t *ppos)
3022 {
3023         long *ptr = filp->private_data;
3024         char buf[64];
3025         long val;
3026         int ret;
3027
3028         if (cnt >= sizeof(buf))
3029                 return -EINVAL;
3030
3031         if (copy_from_user(&buf, ubuf, cnt))
3032                 return -EFAULT;
3033
3034         buf[cnt] = 0;
3035
3036         ret = strict_strtoul(buf, 10, &val);
3037         if (ret < 0)
3038                 return ret;
3039
3040         *ptr = val * 1000;
3041
3042         return cnt;
3043 }
3044
3045 static atomic_t tracing_reader;
3046
3047 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3048 {
3049         struct trace_iterator *iter;
3050
3051         if (tracing_disabled)
3052                 return -ENODEV;
3053
3054         /* We only allow for reader of the pipe */
3055         if (atomic_inc_return(&tracing_reader) != 1) {
3056                 atomic_dec(&tracing_reader);
3057                 return -EBUSY;
3058         }
3059
3060         /* create a buffer to store the information to pass to userspace */
3061         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3062         if (!iter)
3063                 return -ENOMEM;
3064
3065         mutex_lock(&trace_types_lock);
3066
3067         /* trace pipe does not show start of buffer */
3068         cpus_setall(iter->started);
3069
3070         iter->tr = &global_trace;
3071         iter->trace = current_trace;
3072         filp->private_data = iter;
3073
3074         if (iter->trace->pipe_open)
3075                 iter->trace->pipe_open(iter);
3076         mutex_unlock(&trace_types_lock);
3077
3078         return 0;
3079 }
3080
3081 static int tracing_release_pipe(struct inode *inode, struct file *file)
3082 {
3083         struct trace_iterator *iter = file->private_data;
3084
3085         kfree(iter);
3086         atomic_dec(&tracing_reader);
3087
3088         return 0;
3089 }
3090
3091 static unsigned int
3092 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3093 {
3094         struct trace_iterator *iter = filp->private_data;
3095
3096         if (trace_flags & TRACE_ITER_BLOCK) {
3097                 /*
3098                  * Always select as readable when in blocking mode
3099                  */
3100                 return POLLIN | POLLRDNORM;
3101         } else {
3102                 if (!trace_empty(iter))
3103                         return POLLIN | POLLRDNORM;
3104                 poll_wait(filp, &trace_wait, poll_table);
3105                 if (!trace_empty(iter))
3106                         return POLLIN | POLLRDNORM;
3107
3108                 return 0;
3109         }
3110 }
3111
3112 /*
3113  * Consumer reader.
3114  */
3115 static ssize_t
3116 tracing_read_pipe(struct file *filp, char __user *ubuf,
3117                   size_t cnt, loff_t *ppos)
3118 {
3119         struct trace_iterator *iter = filp->private_data;
3120         ssize_t sret;
3121
3122         /* return any leftover data */
3123         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3124         if (sret != -EBUSY)
3125                 return sret;
3126
3127         trace_seq_reset(&iter->seq);
3128
3129         mutex_lock(&trace_types_lock);
3130         if (iter->trace->read) {
3131                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3132                 if (sret)
3133                         goto out;
3134         }
3135
3136 waitagain:
3137         sret = 0;
3138         while (trace_empty(iter)) {
3139
3140                 if ((filp->f_flags & O_NONBLOCK)) {
3141                         sret = -EAGAIN;
3142                         goto out;
3143                 }
3144
3145                 /*
3146                  * This is a make-shift waitqueue. The reason we don't use
3147                  * an actual wait queue is because:
3148                  *  1) we only ever have one waiter
3149                  *  2) the tracing, traces all functions, we don't want
3150                  *     the overhead of calling wake_up and friends
3151                  *     (and tracing them too)
3152                  *     Anyway, this is really very primitive wakeup.
3153                  */
3154                 set_current_state(TASK_INTERRUPTIBLE);
3155                 iter->tr->waiter = current;
3156
3157                 mutex_unlock(&trace_types_lock);
3158
3159                 /* sleep for 100 msecs, and try again. */
3160                 schedule_timeout(HZ/10);
3161
3162                 mutex_lock(&trace_types_lock);
3163
3164                 iter->tr->waiter = NULL;
3165
3166                 if (signal_pending(current)) {
3167                         sret = -EINTR;
3168                         goto out;
3169                 }
3170
3171                 if (iter->trace != current_trace)
3172                         goto out;
3173
3174                 /*
3175                  * We block until we read something and tracing is disabled.
3176                  * We still block if tracing is disabled, but we have never
3177                  * read anything. This allows a user to cat this file, and
3178                  * then enable tracing. But after we have read something,
3179                  * we give an EOF when tracing is again disabled.
3180                  *
3181                  * iter->pos will be 0 if we haven't read anything.
3182                  */
3183                 if (!tracer_enabled && iter->pos)
3184                         break;
3185
3186                 continue;
3187         }
3188
3189         /* stop when tracing is finished */
3190         if (trace_empty(iter))
3191                 goto out;
3192
3193         if (cnt >= PAGE_SIZE)
3194                 cnt = PAGE_SIZE - 1;
3195
3196         /* reset all but tr, trace, and overruns */
3197         memset(&iter->seq, 0,
3198                sizeof(struct trace_iterator) -
3199                offsetof(struct trace_iterator, seq));
3200         iter->pos = -1;
3201
3202         while (find_next_entry_inc(iter) != NULL) {
3203                 enum print_line_t ret;
3204                 int len = iter->seq.len;
3205
3206                 ret = print_trace_line(iter);
3207                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3208                         /* don't print partial lines */
3209                         iter->seq.len = len;
3210                         break;
3211                 }
3212
3213                 trace_consume(iter);
3214
3215                 if (iter->seq.len >= cnt)
3216                         break;
3217         }
3218
3219         /* Now copy what we have to the user */
3220         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3221         if (iter->seq.readpos >= iter->seq.len)
3222                 trace_seq_reset(&iter->seq);
3223
3224         /*
3225          * If there was nothing to send to user, inspite of consuming trace
3226          * entries, go back to wait for more entries.
3227          */
3228         if (sret == -EBUSY)
3229                 goto waitagain;
3230
3231 out:
3232         mutex_unlock(&trace_types_lock);
3233
3234         return sret;
3235 }
3236
3237 static ssize_t
3238 tracing_entries_read(struct file *filp, char __user *ubuf,
3239                      size_t cnt, loff_t *ppos)
3240 {
3241         struct trace_array *tr = filp->private_data;
3242         char buf[64];
3243         int r;
3244
3245         r = sprintf(buf, "%lu\n", tr->entries >> 10);
3246         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3247 }
3248
3249 static ssize_t
3250 tracing_entries_write(struct file *filp, const char __user *ubuf,
3251                       size_t cnt, loff_t *ppos)
3252 {
3253         unsigned long val;
3254         char buf[64];
3255         int ret, cpu;
3256
3257         if (cnt >= sizeof(buf))
3258                 return -EINVAL;
3259
3260         if (copy_from_user(&buf, ubuf, cnt))
3261                 return -EFAULT;
3262
3263         buf[cnt] = 0;
3264
3265         ret = strict_strtoul(buf, 10, &val);
3266         if (ret < 0)
3267                 return ret;
3268
3269         /* must have at least 1 entry */
3270         if (!val)
3271                 return -EINVAL;
3272
3273         mutex_lock(&trace_types_lock);
3274
3275         tracing_stop();
3276
3277         /* disable all cpu buffers */
3278         for_each_tracing_cpu(cpu) {
3279                 if (global_trace.data[cpu])
3280                         atomic_inc(&global_trace.data[cpu]->disabled);
3281                 if (max_tr.data[cpu])
3282                         atomic_inc(&max_tr.data[cpu]->disabled);
3283         }
3284
3285         /* value is in KB */
3286         val <<= 10;
3287
3288         if (val != global_trace.entries) {
3289                 ret = ring_buffer_resize(global_trace.buffer, val);
3290                 if (ret < 0) {
3291                         cnt = ret;
3292                         goto out;
3293                 }
3294
3295                 ret = ring_buffer_resize(max_tr.buffer, val);
3296                 if (ret < 0) {
3297                         int r;
3298                         cnt = ret;
3299                         r = ring_buffer_resize(global_trace.buffer,
3300                                                global_trace.entries);
3301                         if (r < 0) {
3302                                 /* AARGH! We are left with different
3303                                  * size max buffer!!!! */
3304                                 WARN_ON(1);
3305                                 tracing_disabled = 1;
3306                         }
3307                         goto out;
3308                 }
3309
3310                 global_trace.entries = val;
3311         }
3312
3313         filp->f_pos += cnt;
3314
3315         /* If check pages failed, return ENOMEM */
3316         if (tracing_disabled)
3317                 cnt = -ENOMEM;
3318  out:
3319         for_each_tracing_cpu(cpu) {
3320                 if (global_trace.data[cpu])
3321                         atomic_dec(&global_trace.data[cpu]->disabled);
3322                 if (max_tr.data[cpu])
3323                         atomic_dec(&max_tr.data[cpu]->disabled);
3324         }
3325
3326         tracing_start();
3327         max_tr.entries = global_trace.entries;
3328         mutex_unlock(&trace_types_lock);
3329
3330         return cnt;
3331 }
3332
3333 static int mark_printk(const char *fmt, ...)
3334 {
3335         int ret;
3336         va_list args;
3337         va_start(args, fmt);
3338         ret = trace_vprintk(0, fmt, args);
3339         va_end(args);
3340         return ret;
3341 }
3342
3343 static ssize_t
3344 tracing_mark_write(struct file *filp, const char __user *ubuf,
3345                                         size_t cnt, loff_t *fpos)
3346 {
3347         char *buf;
3348         char *end;
3349
3350         if (tracing_disabled)
3351                 return -EINVAL;
3352
3353         if (cnt > TRACE_BUF_SIZE)
3354                 cnt = TRACE_BUF_SIZE;
3355
3356         buf = kmalloc(cnt + 1, GFP_KERNEL);
3357         if (buf == NULL)
3358                 return -ENOMEM;
3359
3360         if (copy_from_user(buf, ubuf, cnt)) {
3361                 kfree(buf);
3362                 return -EFAULT;
3363         }
3364
3365         /* Cut from the first nil or newline. */
3366         buf[cnt] = '\0';
3367         end = strchr(buf, '\n');
3368         if (end)
3369                 *end = '\0';
3370
3371         cnt = mark_printk("%s\n", buf);
3372         kfree(buf);
3373         *fpos += cnt;
3374
3375         return cnt;
3376 }
3377
3378 static struct file_operations tracing_max_lat_fops = {
3379         .open           = tracing_open_generic,
3380         .read           = tracing_max_lat_read,
3381         .write          = tracing_max_lat_write,
3382 };
3383
3384 static struct file_operations tracing_ctrl_fops = {
3385         .open           = tracing_open_generic,
3386         .read           = tracing_ctrl_read,
3387         .write          = tracing_ctrl_write,
3388 };
3389
3390 static struct file_operations set_tracer_fops = {
3391         .open           = tracing_open_generic,
3392         .read           = tracing_set_trace_read,
3393         .write          = tracing_set_trace_write,
3394 };
3395
3396 static struct file_operations tracing_pipe_fops = {
3397         .open           = tracing_open_pipe,
3398         .poll           = tracing_poll_pipe,
3399         .read           = tracing_read_pipe,
3400         .release        = tracing_release_pipe,
3401 };
3402
3403 static struct file_operations tracing_entries_fops = {
3404         .open           = tracing_open_generic,
3405         .read           = tracing_entries_read,
3406         .write          = tracing_entries_write,
3407 };
3408
3409 static struct file_operations tracing_mark_fops = {
3410         .open           = tracing_open_generic,
3411         .write          = tracing_mark_write,
3412 };
3413
3414 #ifdef CONFIG_DYNAMIC_FTRACE
3415
3416 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3417 {
3418         return 0;
3419 }
3420
3421 static ssize_t
3422 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3423                   size_t cnt, loff_t *ppos)
3424 {
3425         static char ftrace_dyn_info_buffer[1024];
3426         static DEFINE_MUTEX(dyn_info_mutex);
3427         unsigned long *p = filp->private_data;
3428         char *buf = ftrace_dyn_info_buffer;
3429         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3430         int r;
3431
3432         mutex_lock(&dyn_info_mutex);
3433         r = sprintf(buf, "%ld ", *p);
3434
3435         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3436         buf[r++] = '\n';
3437
3438         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3439
3440         mutex_unlock(&dyn_info_mutex);
3441
3442         return r;
3443 }
3444
3445 static struct file_operations tracing_dyn_info_fops = {
3446         .open           = tracing_open_generic,
3447         .read           = tracing_read_dyn_info,
3448 };
3449 #endif
3450
3451 static struct dentry *d_tracer;
3452
3453 struct dentry *tracing_init_dentry(void)
3454 {
3455         static int once;
3456
3457         if (d_tracer)
3458                 return d_tracer;
3459
3460         d_tracer = debugfs_create_dir("tracing", NULL);
3461
3462         if (!d_tracer && !once) {
3463                 once = 1;
3464                 pr_warning("Could not create debugfs directory 'tracing'\n");
3465                 return NULL;
3466         }
3467
3468         return d_tracer;
3469 }
3470
3471 #ifdef CONFIG_FTRACE_SELFTEST
3472 /* Let selftest have access to static functions in this file */
3473 #include "trace_selftest.c"
3474 #endif
3475
3476 static __init int tracer_init_debugfs(void)
3477 {
3478         struct dentry *d_tracer;
3479         struct dentry *entry;
3480
3481         d_tracer = tracing_init_dentry();
3482
3483         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3484                                     &global_trace, &tracing_ctrl_fops);
3485         if (!entry)
3486                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3487
3488         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3489                                     NULL, &tracing_iter_fops);
3490         if (!entry)
3491                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3492
3493         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3494                                     NULL, &tracing_cpumask_fops);
3495         if (!entry)
3496                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3497
3498         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3499                                     &global_trace, &tracing_lt_fops);
3500         if (!entry)
3501                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3502
3503         entry = debugfs_create_file("trace", 0444, d_tracer,
3504                                     &global_trace, &tracing_fops);
3505         if (!entry)
3506                 pr_warning("Could not create debugfs 'trace' entry\n");
3507
3508         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3509                                     &global_trace, &show_traces_fops);
3510         if (!entry)
3511                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3512
3513         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3514                                     &global_trace, &set_tracer_fops);
3515         if (!entry)
3516                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3517
3518         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3519                                     &tracing_max_latency,
3520                                     &tracing_max_lat_fops);
3521         if (!entry)
3522                 pr_warning("Could not create debugfs "
3523                            "'tracing_max_latency' entry\n");
3524
3525         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3526                                     &tracing_thresh, &tracing_max_lat_fops);
3527         if (!entry)
3528                 pr_warning("Could not create debugfs "
3529                            "'tracing_thresh' entry\n");
3530         entry = debugfs_create_file("README", 0644, d_tracer,
3531                                     NULL, &tracing_readme_fops);
3532         if (!entry)
3533                 pr_warning("Could not create debugfs 'README' entry\n");
3534
3535         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3536                                     NULL, &tracing_pipe_fops);
3537         if (!entry)
3538                 pr_warning("Could not create debugfs "
3539                            "'trace_pipe' entry\n");
3540
3541         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3542                                     &global_trace, &tracing_entries_fops);
3543         if (!entry)
3544                 pr_warning("Could not create debugfs "
3545                            "'buffer_size_kb' entry\n");
3546
3547         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3548                                     NULL, &tracing_mark_fops);
3549         if (!entry)
3550                 pr_warning("Could not create debugfs "
3551                            "'trace_marker' entry\n");
3552
3553 #ifdef CONFIG_DYNAMIC_FTRACE
3554         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3555                                     &ftrace_update_tot_cnt,
3556                                     &tracing_dyn_info_fops);
3557         if (!entry)
3558                 pr_warning("Could not create debugfs "
3559                            "'dyn_ftrace_total_info' entry\n");
3560 #endif
3561 #ifdef CONFIG_SYSPROF_TRACER
3562         init_tracer_sysprof_debugfs(d_tracer);
3563 #endif
3564         return 0;
3565 }
3566
3567 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3568 {
3569         static DEFINE_SPINLOCK(trace_buf_lock);
3570         static char trace_buf[TRACE_BUF_SIZE];
3571
3572         struct ring_buffer_event *event;
3573         struct trace_array *tr = &global_trace;
3574         struct trace_array_cpu *data;
3575         struct print_entry *entry;
3576         unsigned long flags, irq_flags;
3577         int cpu, len = 0, size, pc;
3578
3579         if (tracing_disabled)
3580                 return 0;
3581
3582         pc = preempt_count();
3583         preempt_disable_notrace();
3584         cpu = raw_smp_processor_id();
3585         data = tr->data[cpu];
3586
3587         if (unlikely(atomic_read(&data->disabled)))
3588                 goto out;
3589
3590         spin_lock_irqsave(&trace_buf_lock, flags);
3591         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3592
3593         len = min(len, TRACE_BUF_SIZE-1);
3594         trace_buf[len] = 0;
3595
3596         size = sizeof(*entry) + len + 1;
3597         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3598         if (!event)
3599                 goto out_unlock;
3600         entry = ring_buffer_event_data(event);
3601         tracing_generic_entry_update(&entry->ent, flags, pc);
3602         entry->ent.type                 = TRACE_PRINT;
3603         entry->ip                       = ip;
3604
3605         memcpy(&entry->buf, trace_buf, len);
3606         entry->buf[len] = 0;
3607         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3608
3609  out_unlock:
3610         spin_unlock_irqrestore(&trace_buf_lock, flags);
3611
3612  out:
3613         preempt_enable_notrace();
3614
3615         return len;
3616 }
3617 EXPORT_SYMBOL_GPL(trace_vprintk);
3618
3619 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3620 {
3621         int ret;
3622         va_list ap;
3623
3624         if (!(trace_flags & TRACE_ITER_PRINTK))
3625                 return 0;
3626
3627         va_start(ap, fmt);
3628         ret = trace_vprintk(ip, fmt, ap);
3629         va_end(ap);
3630         return ret;
3631 }
3632 EXPORT_SYMBOL_GPL(__ftrace_printk);
3633
3634 static int trace_panic_handler(struct notifier_block *this,
3635                                unsigned long event, void *unused)
3636 {
3637         if (ftrace_dump_on_oops)
3638                 ftrace_dump();
3639         return NOTIFY_OK;
3640 }
3641
3642 static struct notifier_block trace_panic_notifier = {
3643         .notifier_call  = trace_panic_handler,
3644         .next           = NULL,
3645         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3646 };
3647
3648 static int trace_die_handler(struct notifier_block *self,
3649                              unsigned long val,
3650                              void *data)
3651 {
3652         switch (val) {
3653         case DIE_OOPS:
3654                 if (ftrace_dump_on_oops)
3655                         ftrace_dump();
3656                 break;
3657         default:
3658                 break;
3659         }
3660         return NOTIFY_OK;
3661 }
3662
3663 static struct notifier_block trace_die_notifier = {
3664         .notifier_call = trace_die_handler,
3665         .priority = 200
3666 };
3667
3668 /*
3669  * printk is set to max of 1024, we really don't need it that big.
3670  * Nothing should be printing 1000 characters anyway.
3671  */
3672 #define TRACE_MAX_PRINT         1000
3673
3674 /*
3675  * Define here KERN_TRACE so that we have one place to modify
3676  * it if we decide to change what log level the ftrace dump
3677  * should be at.
3678  */
3679 #define KERN_TRACE              KERN_INFO
3680
3681 static void
3682 trace_printk_seq(struct trace_seq *s)
3683 {
3684         /* Probably should print a warning here. */
3685         if (s->len >= 1000)
3686                 s->len = 1000;
3687
3688         /* should be zero ended, but we are paranoid. */
3689         s->buffer[s->len] = 0;
3690
3691         printk(KERN_TRACE "%s", s->buffer);
3692
3693         trace_seq_reset(s);
3694 }
3695
3696 void ftrace_dump(void)
3697 {
3698         static DEFINE_SPINLOCK(ftrace_dump_lock);
3699         /* use static because iter can be a bit big for the stack */
3700         static struct trace_iterator iter;
3701         static cpumask_t mask;
3702         static int dump_ran;
3703         unsigned long flags;
3704         int cnt = 0, cpu;
3705
3706         /* only one dump */
3707         spin_lock_irqsave(&ftrace_dump_lock, flags);
3708         if (dump_ran)
3709                 goto out;
3710
3711         dump_ran = 1;
3712
3713         /* No turning back! */
3714         ftrace_kill();
3715
3716         for_each_tracing_cpu(cpu) {
3717                 atomic_inc(&global_trace.data[cpu]->disabled);
3718         }
3719
3720         /* don't look at user memory in panic mode */
3721         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3722
3723         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3724
3725         iter.tr = &global_trace;
3726         iter.trace = current_trace;
3727
3728         /*
3729          * We need to stop all tracing on all CPUS to read the
3730          * the next buffer. This is a bit expensive, but is
3731          * not done often. We fill all what we can read,
3732          * and then release the locks again.
3733          */
3734
3735         cpus_clear(mask);
3736
3737         while (!trace_empty(&iter)) {
3738
3739                 if (!cnt)
3740                         printk(KERN_TRACE "---------------------------------\n");
3741
3742                 cnt++;
3743
3744                 /* reset all but tr, trace, and overruns */
3745                 memset(&iter.seq, 0,
3746                        sizeof(struct trace_iterator) -
3747                        offsetof(struct trace_iterator, seq));
3748                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3749                 iter.pos = -1;
3750
3751                 if (find_next_entry_inc(&iter) != NULL) {
3752                         print_trace_line(&iter);
3753                         trace_consume(&iter);
3754                 }
3755
3756                 trace_printk_seq(&iter.seq);
3757         }
3758
3759         if (!cnt)
3760                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3761         else
3762                 printk(KERN_TRACE "---------------------------------\n");
3763
3764  out:
3765         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3766 }
3767
3768 __init static int tracer_alloc_buffers(void)
3769 {
3770         struct trace_array_cpu *data;
3771         int i;
3772
3773         /* TODO: make the number of buffers hot pluggable with CPUS */
3774         tracing_buffer_mask = cpu_possible_map;
3775
3776         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3777                                                    TRACE_BUFFER_FLAGS);
3778         if (!global_trace.buffer) {
3779                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3780                 WARN_ON(1);
3781                 return 0;
3782         }
3783         global_trace.entries = ring_buffer_size(global_trace.buffer);
3784
3785 #ifdef CONFIG_TRACER_MAX_TRACE
3786         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3787                                              TRACE_BUFFER_FLAGS);
3788         if (!max_tr.buffer) {
3789                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3790                 WARN_ON(1);
3791                 ring_buffer_free(global_trace.buffer);
3792                 return 0;
3793         }
3794         max_tr.entries = ring_buffer_size(max_tr.buffer);
3795         WARN_ON(max_tr.entries != global_trace.entries);
3796 #endif
3797
3798         /* Allocate the first page for all buffers */
3799         for_each_tracing_cpu(i) {
3800                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3801                 max_tr.data[i] = &per_cpu(max_data, i);
3802         }
3803
3804         trace_init_cmdlines();
3805
3806         register_tracer(&nop_trace);
3807 #ifdef CONFIG_BOOT_TRACER
3808         register_tracer(&boot_tracer);
3809         current_trace = &boot_tracer;
3810         current_trace->init(&global_trace);
3811 #else
3812         current_trace = &nop_trace;
3813 #endif
3814
3815         /* All seems OK, enable tracing */
3816         tracing_disabled = 0;
3817
3818         atomic_notifier_chain_register(&panic_notifier_list,
3819                                        &trace_panic_notifier);
3820
3821         register_die_notifier(&trace_die_notifier);
3822
3823         return 0;
3824 }
3825 early_initcall(tracer_alloc_buffers);
3826 fs_initcall(tracer_init_debugfs);