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