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