git.oblomov.eu Git - linux-2.6/blob - 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 #ifdef CONFIG_KRETPROBES
1092 static inline const char *kretprobed(const char *name)
1093 {
1094         static const char tramp_name[] = "kretprobe_trampoline";
1095         int size = sizeof(tramp_name);
1096
1097         if (strncmp(tramp_name, name, size) == 0)
1098                 return "[unknown/kretprobe'd]";
1099         return name;
1100 }
1101 #else
1102 static inline const char *kretprobed(const char *name)
1103 {
1104         return name;
1105 }
1106 #endif /* CONFIG_KRETPROBES */
1107
1108 static int
1109 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1110 {
1111 #ifdef CONFIG_KALLSYMS
1112         char str[KSYM_SYMBOL_LEN];
1113         const char *name;
1114
1115         kallsyms_lookup(address, NULL, NULL, NULL, str);
1116
1117         name = kretprobed(str);
1118
1119         return trace_seq_printf(s, fmt, name);
1120 #endif
1121         return 1;
1122 }
1123
1124 static int
1125 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1126                      unsigned long address)
1127 {
1128 #ifdef CONFIG_KALLSYMS
1129         char str[KSYM_SYMBOL_LEN];
1130         const char *name;
1131
1132         sprint_symbol(str, address);
1133         name = kretprobed(str);
1134
1135         return trace_seq_printf(s, fmt, name);
1136 #endif
1137         return 1;
1138 }
1139
1140 #ifndef CONFIG_64BIT
1141 # define IP_FMT "%08lx"
1142 #else
1143 # define IP_FMT "%016lx"
1144 #endif
1145
1146 static int
1147 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1148 {
1149         int ret;
1150
1151         if (!ip)
1152                 return trace_seq_printf(s, "0");
1153
1154         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1155                 ret = seq_print_sym_offset(s, "%s", ip);
1156         else
1157                 ret = seq_print_sym_short(s, "%s", ip);
1158
1159         if (!ret)
1160                 return 0;
1161
1162         if (sym_flags & TRACE_ITER_SYM_ADDR)
1163                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1164         return ret;
1165 }
1166
1167 static void print_lat_help_header(struct seq_file *m)
1168 {
1169         seq_puts(m, "#                  _------=> CPU#            \n");
1170         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1171         seq_puts(m, "#                | / _----=> need-resched    \n");
1172         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1173         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1174         seq_puts(m, "#                |||| /                      \n");
1175         seq_puts(m, "#                |||||     delay             \n");
1176         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1177         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1178 }
1179
1180 static void print_func_help_header(struct seq_file *m)
1181 {
1182         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1183         seq_puts(m, "#              | |       |          |         |\n");
1184 }
1185
1186
1187 static void
1188 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1189 {
1190         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1191         struct trace_array *tr = iter->tr;
1192         struct trace_array_cpu *data = tr->data[tr->cpu];
1193         struct tracer *type = current_trace;
1194         unsigned long total;
1195         unsigned long entries;
1196         const char *name = "preemption";
1197
1198         if (type)
1199                 name = type->name;
1200
1201         entries = ring_buffer_entries(iter->tr->buffer);
1202         total = entries +
1203                 ring_buffer_overruns(iter->tr->buffer);
1204
1205         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1206                    name, UTS_RELEASE);
1207         seq_puts(m, "-----------------------------------"
1208                  "---------------------------------\n");
1209         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1210                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1211                    nsecs_to_usecs(data->saved_latency),
1212                    entries,
1213                    total,
1214                    tr->cpu,
1215 #if defined(CONFIG_PREEMPT_NONE)
1216                    "server",
1217 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1218                    "desktop",
1219 #elif defined(CONFIG_PREEMPT)
1220                    "preempt",
1221 #else
1222                    "unknown",
1223 #endif
1224                    /* These are reserved for later use */
1225                    0, 0, 0, 0);
1226 #ifdef CONFIG_SMP
1227         seq_printf(m, " #P:%d)\n", num_online_cpus());
1228 #else
1229         seq_puts(m, ")\n");
1230 #endif
1231         seq_puts(m, "    -----------------\n");
1232         seq_printf(m, "    | task: %.16s-%d "
1233                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1234                    data->comm, data->pid, data->uid, data->nice,
1235                    data->policy, data->rt_priority);
1236         seq_puts(m, "    -----------------\n");
1237
1238         if (data->critical_start) {
1239                 seq_puts(m, " => started at: ");
1240                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1241                 trace_print_seq(m, &iter->seq);
1242                 seq_puts(m, "\n => ended at:   ");
1243                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1244                 trace_print_seq(m, &iter->seq);
1245                 seq_puts(m, "\n");
1246         }
1247
1248         seq_puts(m, "\n");
1249 }
1250
1251 static void
1252 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1253 {
1254         int hardirq, softirq;
1255         char *comm;
1256
1257         comm = trace_find_cmdline(entry->pid);
1258
1259         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1260         trace_seq_printf(s, "%3d", cpu);
1261         trace_seq_printf(s, "%c%c",
1262                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1263                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1264                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1265
1266         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1267         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1268         if (hardirq && softirq) {
1269                 trace_seq_putc(s, 'H');
1270         } else {
1271                 if (hardirq) {
1272                         trace_seq_putc(s, 'h');
1273                 } else {
1274                         if (softirq)
1275                                 trace_seq_putc(s, 's');
1276                         else
1277                                 trace_seq_putc(s, '.');
1278                 }
1279         }
1280
1281         if (entry->preempt_count)
1282                 trace_seq_printf(s, "%x", entry->preempt_count);
1283         else
1284                 trace_seq_puts(s, ".");
1285 }
1286
1287 unsigned long preempt_mark_thresh = 100;
1288
1289 static void
1290 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1291                     unsigned long rel_usecs)
1292 {
1293         trace_seq_printf(s, " %4lldus", abs_usecs);
1294         if (rel_usecs > preempt_mark_thresh)
1295                 trace_seq_puts(s, "!: ");
1296         else if (rel_usecs > 1)
1297                 trace_seq_puts(s, "+: ");
1298         else
1299                 trace_seq_puts(s, " : ");
1300 }
1301
1302 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1303
1304 /*
1305  * The message is supposed to contain an ending newline.
1306  * If the printing stops prematurely, try to add a newline of our own.
1307  */
1308 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1309 {
1310         struct trace_entry *ent;
1311         struct trace_field_cont *cont;
1312         bool ok = true;
1313
1314         ent = peek_next_entry(iter, iter->cpu, NULL);
1315         if (!ent || ent->type != TRACE_CONT) {
1316                 trace_seq_putc(s, '\n');
1317                 return;
1318         }
1319
1320         do {
1321                 cont = (struct trace_field_cont *)ent;
1322                 if (ok)
1323                         ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1324
1325                 ftrace_disable_cpu();
1326
1327                 if (iter->buffer_iter[iter->cpu])
1328                         ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1329                 else
1330                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1331
1332                 ftrace_enable_cpu();
1333
1334                 ent = peek_next_entry(iter, iter->cpu, NULL);
1335         } while (ent && ent->type == TRACE_CONT);
1336
1337         if (!ok)
1338                 trace_seq_putc(s, '\n');
1339 }
1340
1341 static enum print_line_t
1342 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1343 {
1344         struct trace_seq *s = &iter->seq;
1345         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1346         struct trace_entry *next_entry;
1347         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1348         struct trace_entry *entry = iter->ent;
1349         unsigned long abs_usecs;
1350         unsigned long rel_usecs;
1351         u64 next_ts;
1352         char *comm;
1353         int S, T;
1354         int i;
1355         unsigned state;
1356
1357         if (entry->type == TRACE_CONT)
1358                 return TRACE_TYPE_HANDLED;
1359
1360         next_entry = find_next_entry(iter, NULL, &next_ts);
1361         if (!next_entry)
1362                 next_ts = iter->ts;
1363         rel_usecs = ns2usecs(next_ts - iter->ts);
1364         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1365
1366         if (verbose) {
1367                 comm = trace_find_cmdline(entry->pid);
1368                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1369                                  " %ld.%03ldms (+%ld.%03ldms): ",
1370                                  comm,
1371                                  entry->pid, cpu, entry->flags,
1372                                  entry->preempt_count, trace_idx,
1373                                  ns2usecs(iter->ts),
1374                                  abs_usecs/1000,
1375                                  abs_usecs % 1000, rel_usecs/1000,
1376                                  rel_usecs % 1000);
1377         } else {
1378                 lat_print_generic(s, entry, cpu);
1379                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1380         }
1381         switch (entry->type) {
1382         case TRACE_FN: {
1383                 struct ftrace_entry *field;
1384
1385                 trace_assign_type(field, entry);
1386
1387                 seq_print_ip_sym(s, field->ip, sym_flags);
1388                 trace_seq_puts(s, " (");
1389                 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1390                 trace_seq_puts(s, ")\n");
1391                 break;
1392         }
1393         case TRACE_CTX:
1394         case TRACE_WAKE: {
1395                 struct ctx_switch_entry *field;
1396
1397                 trace_assign_type(field, entry);
1398
1399                 T = field->next_state < sizeof(state_to_char) ?
1400                         state_to_char[field->next_state] : 'X';
1401
1402                 state = field->prev_state ?
1403                         __ffs(field->prev_state) + 1 : 0;
1404                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1405                 comm = trace_find_cmdline(field->next_pid);
1406                 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1407                                  field->prev_pid,
1408                                  field->prev_prio,
1409                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1410                                  field->next_cpu,
1411                                  field->next_pid,
1412                                  field->next_prio,
1413                                  T, comm);
1414                 break;
1415         }
1416         case TRACE_SPECIAL: {
1417                 struct special_entry *field;
1418
1419                 trace_assign_type(field, entry);
1420
1421                 trace_seq_printf(s, "# %ld %ld %ld\n",
1422                                  field->arg1,
1423                                  field->arg2,
1424                                  field->arg3);
1425                 break;
1426         }
1427         case TRACE_STACK: {
1428                 struct stack_entry *field;
1429
1430                 trace_assign_type(field, entry);
1431
1432                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1433                         if (i)
1434                                 trace_seq_puts(s, " <= ");
1435                         seq_print_ip_sym(s, field->caller[i], sym_flags);
1436                 }
1437                 trace_seq_puts(s, "\n");
1438                 break;
1439         }
1440         case TRACE_PRINT: {
1441                 struct print_entry *field;
1442
1443                 trace_assign_type(field, entry);
1444
1445                 seq_print_ip_sym(s, field->ip, sym_flags);
1446                 trace_seq_printf(s, ": %s", field->buf);
1447                 if (entry->flags & TRACE_FLAG_CONT)
1448                         trace_seq_print_cont(s, iter);
1449                 break;
1450         }
1451         default:
1452                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1453         }
1454         return TRACE_TYPE_HANDLED;
1455 }
1456
1457 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1458 {
1459         struct trace_seq *s = &iter->seq;
1460         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1461         struct trace_entry *entry;
1462         unsigned long usec_rem;
1463         unsigned long long t;
1464         unsigned long secs;
1465         char *comm;
1466         int ret;
1467         int S, T;
1468         int i;
1469
1470         entry = iter->ent;
1471
1472         if (entry->type == TRACE_CONT)
1473                 return TRACE_TYPE_HANDLED;
1474
1475         comm = trace_find_cmdline(iter->ent->pid);
1476
1477         t = ns2usecs(iter->ts);
1478         usec_rem = do_div(t, 1000000ULL);
1479         secs = (unsigned long)t;
1480
1481         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1482         if (!ret)
1483                 return TRACE_TYPE_PARTIAL_LINE;
1484         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1485         if (!ret)
1486                 return TRACE_TYPE_PARTIAL_LINE;
1487         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1488         if (!ret)
1489                 return TRACE_TYPE_PARTIAL_LINE;
1490
1491         switch (entry->type) {
1492         case TRACE_FN: {
1493                 struct ftrace_entry *field;
1494
1495                 trace_assign_type(field, entry);
1496
1497                 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1498                 if (!ret)
1499                         return TRACE_TYPE_PARTIAL_LINE;
1500                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1501                                                 field->parent_ip) {
1502                         ret = trace_seq_printf(s, " <-");
1503                         if (!ret)
1504                                 return TRACE_TYPE_PARTIAL_LINE;
1505                         ret = seq_print_ip_sym(s,
1506                                                field->parent_ip,
1507                                                sym_flags);
1508                         if (!ret)
1509                                 return TRACE_TYPE_PARTIAL_LINE;
1510                 }
1511                 ret = trace_seq_printf(s, "\n");
1512                 if (!ret)
1513                         return TRACE_TYPE_PARTIAL_LINE;
1514                 break;
1515         }
1516         case TRACE_CTX:
1517         case TRACE_WAKE: {
1518                 struct ctx_switch_entry *field;
1519
1520                 trace_assign_type(field, entry);
1521
1522                 S = field->prev_state < sizeof(state_to_char) ?
1523                         state_to_char[field->prev_state] : 'X';
1524                 T = field->next_state < sizeof(state_to_char) ?
1525                         state_to_char[field->next_state] : 'X';
1526                 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1527                                        field->prev_pid,
1528                                        field->prev_prio,
1529                                        S,
1530                                        entry->type == TRACE_CTX ? "==>" : "  +",
1531                                        field->next_cpu,
1532                                        field->next_pid,
1533                                        field->next_prio,
1534                                        T);
1535                 if (!ret)
1536                         return TRACE_TYPE_PARTIAL_LINE;
1537                 break;
1538         }
1539         case TRACE_SPECIAL: {
1540                 struct special_entry *field;
1541
1542                 trace_assign_type(field, entry);
1543
1544                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1545                                  field->arg1,
1546                                  field->arg2,
1547                                  field->arg3);
1548                 if (!ret)
1549                         return TRACE_TYPE_PARTIAL_LINE;
1550                 break;
1551         }
1552         case TRACE_STACK: {
1553                 struct stack_entry *field;
1554
1555                 trace_assign_type(field, entry);
1556
1557                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1558                         if (i) {
1559                                 ret = trace_seq_puts(s, " <= ");
1560                                 if (!ret)
1561                                         return TRACE_TYPE_PARTIAL_LINE;
1562                         }
1563                         ret = seq_print_ip_sym(s, field->caller[i],
1564                                                sym_flags);
1565                         if (!ret)
1566                                 return TRACE_TYPE_PARTIAL_LINE;
1567                 }
1568                 ret = trace_seq_puts(s, "\n");
1569                 if (!ret)
1570                         return TRACE_TYPE_PARTIAL_LINE;
1571                 break;
1572         }
1573         case TRACE_PRINT: {
1574                 struct print_entry *field;
1575
1576                 trace_assign_type(field, entry);
1577
1578                 seq_print_ip_sym(s, field->ip, sym_flags);
1579                 trace_seq_printf(s, ": %s", field->buf);
1580                 if (entry->flags & TRACE_FLAG_CONT)
1581                         trace_seq_print_cont(s, iter);
1582                 break;
1583         }
1584         }
1585         return TRACE_TYPE_HANDLED;
1586 }
1587
1588 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1589 {
1590         struct trace_seq *s = &iter->seq;
1591         struct trace_entry *entry;
1592         int ret;
1593         int S, T;
1594
1595         entry = iter->ent;
1596
1597         if (entry->type == TRACE_CONT)
1598                 return TRACE_TYPE_HANDLED;
1599
1600         ret = trace_seq_printf(s, "%d %d %llu ",
1601                 entry->pid, iter->cpu, iter->ts);
1602         if (!ret)
1603                 return TRACE_TYPE_PARTIAL_LINE;
1604
1605         switch (entry->type) {
1606         case TRACE_FN: {
1607                 struct ftrace_entry *field;
1608
1609                 trace_assign_type(field, entry);
1610
1611                 ret = trace_seq_printf(s, "%x %x\n",
1612                                         field->ip,
1613                                         field->parent_ip);
1614                 if (!ret)
1615                         return TRACE_TYPE_PARTIAL_LINE;
1616                 break;
1617         }
1618         case TRACE_CTX:
1619         case TRACE_WAKE: {
1620                 struct ctx_switch_entry *field;
1621
1622                 trace_assign_type(field, entry);
1623
1624                 S = field->prev_state < sizeof(state_to_char) ?
1625                         state_to_char[field->prev_state] : 'X';
1626                 T = field->next_state < sizeof(state_to_char) ?
1627                         state_to_char[field->next_state] : 'X';
1628                 if (entry->type == TRACE_WAKE)
1629                         S = '+';
1630                 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
1631                                        field->prev_pid,
1632                                        field->prev_prio,
1633                                        S,
1634                                        field->next_cpu,
1635                                        field->next_pid,
1636                                        field->next_prio,
1637                                        T);
1638                 if (!ret)
1639                         return TRACE_TYPE_PARTIAL_LINE;
1640                 break;
1641         }
1642         case TRACE_SPECIAL:
1643         case TRACE_STACK: {
1644                 struct special_entry *field;
1645
1646                 trace_assign_type(field, entry);
1647
1648                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1649                                  field->arg1,
1650                                  field->arg2,
1651                                  field->arg3);
1652                 if (!ret)
1653                         return TRACE_TYPE_PARTIAL_LINE;
1654                 break;
1655         }
1656         case TRACE_PRINT: {
1657                 struct print_entry *field;
1658
1659                 trace_assign_type(field, entry);
1660
1661                 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
1662                 if (entry->flags & TRACE_FLAG_CONT)
1663                         trace_seq_print_cont(s, iter);
1664                 break;
1665         }
1666         }
1667         return TRACE_TYPE_HANDLED;
1668 }
1669
1670 #define SEQ_PUT_FIELD_RET(s, x)                         \
1671 do {                                                    \
1672         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1673                 return 0;                               \
1674 } while (0)
1675
1676 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1677 do {                                                    \
1678         BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES);     \
1679         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1680                 return 0;                               \
1681 } while (0)
1682
1683 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1684 {
1685         struct trace_seq *s = &iter->seq;
1686         unsigned char newline = '\n';
1687         struct trace_entry *entry;
1688         int S, T;
1689
1690         entry = iter->ent;
1691
1692         if (entry->type == TRACE_CONT)
1693                 return TRACE_TYPE_HANDLED;
1694
1695         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1696         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1697         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1698
1699         switch (entry->type) {
1700         case TRACE_FN: {
1701                 struct ftrace_entry *field;
1702
1703                 trace_assign_type(field, entry);
1704
1705                 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
1706                 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
1707                 break;
1708         }
1709         case TRACE_CTX:
1710         case TRACE_WAKE: {
1711                 struct ctx_switch_entry *field;
1712
1713                 trace_assign_type(field, entry);
1714
1715                 S = field->prev_state < sizeof(state_to_char) ?
1716                         state_to_char[field->prev_state] : 'X';
1717                 T = field->next_state < sizeof(state_to_char) ?
1718                         state_to_char[field->next_state] : 'X';
1719                 if (entry->type == TRACE_WAKE)
1720                         S = '+';
1721                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
1722                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
1723                 SEQ_PUT_HEX_FIELD_RET(s, S);
1724                 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
1725                 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
1726                 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
1727                 SEQ_PUT_HEX_FIELD_RET(s, T);
1728                 break;
1729         }
1730         case TRACE_SPECIAL:
1731         case TRACE_STACK: {
1732                 struct special_entry *field;
1733
1734                 trace_assign_type(field, entry);
1735
1736                 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
1737                 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
1738                 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
1739                 break;
1740         }
1741         }
1742         SEQ_PUT_FIELD_RET(s, newline);
1743
1744         return TRACE_TYPE_HANDLED;
1745 }
1746
1747 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1748 {
1749         struct trace_seq *s = &iter->seq;
1750         struct trace_entry *entry;
1751
1752         entry = iter->ent;
1753
1754         if (entry->type == TRACE_CONT)
1755                 return TRACE_TYPE_HANDLED;
1756
1757         SEQ_PUT_FIELD_RET(s, entry->pid);
1758         SEQ_PUT_FIELD_RET(s, entry->cpu);
1759         SEQ_PUT_FIELD_RET(s, iter->ts);
1760
1761         switch (entry->type) {
1762         case TRACE_FN: {
1763                 struct ftrace_entry *field;
1764
1765                 trace_assign_type(field, entry);
1766
1767                 SEQ_PUT_FIELD_RET(s, field->ip);
1768                 SEQ_PUT_FIELD_RET(s, field->parent_ip);
1769                 break;
1770         }
1771         case TRACE_CTX: {
1772                 struct ctx_switch_entry *field;
1773
1774                 trace_assign_type(field, entry);
1775
1776                 SEQ_PUT_FIELD_RET(s, field->prev_pid);
1777                 SEQ_PUT_FIELD_RET(s, field->prev_prio);
1778                 SEQ_PUT_FIELD_RET(s, field->prev_state);
1779                 SEQ_PUT_FIELD_RET(s, field->next_pid);
1780                 SEQ_PUT_FIELD_RET(s, field->next_prio);
1781                 SEQ_PUT_FIELD_RET(s, field->next_state);
1782                 break;
1783         }
1784         case TRACE_SPECIAL:
1785         case TRACE_STACK: {
1786                 struct special_entry *field;
1787
1788                 trace_assign_type(field, entry);
1789
1790                 SEQ_PUT_FIELD_RET(s, field->arg1);
1791                 SEQ_PUT_FIELD_RET(s, field->arg2);
1792                 SEQ_PUT_FIELD_RET(s, field->arg3);
1793                 break;
1794         }
1795         }
1796         return 1;
1797 }
1798
1799 static int trace_empty(struct trace_iterator *iter)
1800 {
1801         int cpu;
1802
1803         for_each_tracing_cpu(cpu) {
1804                 if (iter->buffer_iter[cpu]) {
1805                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1806                                 return 0;
1807                 } else {
1808                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1809                                 return 0;
1810                 }
1811         }
1812
1813         return 1;
1814 }
1815
1816 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1817 {
1818         enum print_line_t ret;
1819
1820         if (iter->trace && iter->trace->print_line) {
1821                 ret = iter->trace->print_line(iter);
1822                 if (ret != TRACE_TYPE_UNHANDLED)
1823                         return ret;
1824         }
1825
1826         if (trace_flags & TRACE_ITER_BIN)
1827                 return print_bin_fmt(iter);
1828
1829         if (trace_flags & TRACE_ITER_HEX)
1830                 return print_hex_fmt(iter);
1831
1832         if (trace_flags & TRACE_ITER_RAW)
1833                 return print_raw_fmt(iter);
1834
1835         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1836                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1837
1838         return print_trace_fmt(iter);
1839 }
1840
1841 static int s_show(struct seq_file *m, void *v)
1842 {
1843         struct trace_iterator *iter = v;
1844
1845         if (iter->ent == NULL) {
1846                 if (iter->tr) {
1847                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1848                         seq_puts(m, "#\n");
1849                 }
1850                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1851                         /* print nothing if the buffers are empty */
1852                         if (trace_empty(iter))
1853                                 return 0;
1854                         print_trace_header(m, iter);
1855                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1856                                 print_lat_help_header(m);
1857                 } else {
1858                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1859                                 print_func_help_header(m);
1860                 }
1861         } else {
1862                 print_trace_line(iter);
1863                 trace_print_seq(m, &iter->seq);
1864         }
1865
1866         return 0;
1867 }
1868
1869 static struct seq_operations tracer_seq_ops = {
1870         .start          = s_start,
1871         .next           = s_next,
1872         .stop           = s_stop,
1873         .show           = s_show,
1874 };
1875
1876 static struct trace_iterator *
1877 __tracing_open(struct inode *inode, struct file *file, int *ret)
1878 {
1879         struct trace_iterator *iter;
1880         struct seq_file *m;
1881         int cpu;
1882
1883         if (tracing_disabled) {
1884                 *ret = -ENODEV;
1885                 return NULL;
1886         }
1887
1888         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1889         if (!iter) {
1890                 *ret = -ENOMEM;
1891                 goto out;
1892         }
1893
1894         mutex_lock(&trace_types_lock);
1895         if (current_trace && current_trace->print_max)
1896                 iter->tr = &max_tr;
1897         else
1898                 iter->tr = inode->i_private;
1899         iter->trace = current_trace;
1900         iter->pos = -1;
1901
1902         for_each_tracing_cpu(cpu) {
1903
1904                 iter->buffer_iter[cpu] =
1905                         ring_buffer_read_start(iter->tr->buffer, cpu);
1906
1907                 if (!iter->buffer_iter[cpu])
1908                         goto fail_buffer;
1909         }
1910
1911         /* TODO stop tracer */
1912         *ret = seq_open(file, &tracer_seq_ops);
1913         if (*ret)
1914                 goto fail_buffer;
1915
1916         m = file->private_data;
1917         m->private = iter;
1918
1919         /* stop the trace while dumping */
1920         if (iter->tr->ctrl) {
1921                 tracer_enabled = 0;
1922                 ftrace_function_enabled = 0;
1923         }
1924
1925         if (iter->trace && iter->trace->open)
1926                         iter->trace->open(iter);
1927
1928         mutex_unlock(&trace_types_lock);
1929
1930  out:
1931         return iter;
1932
1933  fail_buffer:
1934         for_each_tracing_cpu(cpu) {
1935                 if (iter->buffer_iter[cpu])
1936                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1937         }
1938         mutex_unlock(&trace_types_lock);
1939         kfree(iter);
1940
1941         return ERR_PTR(-ENOMEM);
1942 }
1943
1944 int tracing_open_generic(struct inode *inode, struct file *filp)
1945 {
1946         if (tracing_disabled)
1947                 return -ENODEV;
1948
1949         filp->private_data = inode->i_private;
1950         return 0;
1951 }
1952
1953 int tracing_release(struct inode *inode, struct file *file)
1954 {
1955         struct seq_file *m = (struct seq_file *)file->private_data;
1956         struct trace_iterator *iter = m->private;
1957         int cpu;
1958
1959         mutex_lock(&trace_types_lock);
1960         for_each_tracing_cpu(cpu) {
1961                 if (iter->buffer_iter[cpu])
1962                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1963         }
1964
1965         if (iter->trace && iter->trace->close)
1966                 iter->trace->close(iter);
1967
1968         /* reenable tracing if it was previously enabled */
1969         if (iter->tr->ctrl) {
1970                 tracer_enabled = 1;
1971                 /*
1972                  * It is safe to enable function tracing even if it
1973                  * isn't used
1974                  */
1975                 ftrace_function_enabled = 1;
1976         }
1977         mutex_unlock(&trace_types_lock);
1978
1979         seq_release(inode, file);
1980         kfree(iter);
1981         return 0;
1982 }
1983
1984 static int tracing_open(struct inode *inode, struct file *file)
1985 {
1986         int ret;
1987
1988         __tracing_open(inode, file, &ret);
1989
1990         return ret;
1991 }
1992
1993 static int tracing_lt_open(struct inode *inode, struct file *file)
1994 {
1995         struct trace_iterator *iter;
1996         int ret;
1997
1998         iter = __tracing_open(inode, file, &ret);
1999
2000         if (!ret)
2001                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2002
2003         return ret;
2004 }
2005
2006
2007 static void *
2008 t_next(struct seq_file *m, void *v, loff_t *pos)
2009 {
2010         struct tracer *t = m->private;
2011
2012         (*pos)++;
2013
2014         if (t)
2015                 t = t->next;
2016
2017         m->private = t;
2018
2019         return t;
2020 }
2021
2022 static void *t_start(struct seq_file *m, loff_t *pos)
2023 {
2024         struct tracer *t = m->private;
2025         loff_t l = 0;
2026
2027         mutex_lock(&trace_types_lock);
2028         for (; t && l < *pos; t = t_next(m, t, &l))
2029                 ;
2030
2031         return t;
2032 }
2033
2034 static void t_stop(struct seq_file *m, void *p)
2035 {
2036         mutex_unlock(&trace_types_lock);
2037 }
2038
2039 static int t_show(struct seq_file *m, void *v)
2040 {
2041         struct tracer *t = v;
2042
2043         if (!t)
2044                 return 0;
2045
2046         seq_printf(m, "%s", t->name);
2047         if (t->next)
2048                 seq_putc(m, ' ');
2049         else
2050                 seq_putc(m, '\n');
2051
2052         return 0;
2053 }
2054
2055 static struct seq_operations show_traces_seq_ops = {
2056         .start          = t_start,
2057         .next           = t_next,
2058         .stop           = t_stop,
2059         .show           = t_show,
2060 };
2061
2062 static int show_traces_open(struct inode *inode, struct file *file)
2063 {
2064         int ret;
2065
2066         if (tracing_disabled)
2067                 return -ENODEV;
2068
2069         ret = seq_open(file, &show_traces_seq_ops);
2070         if (!ret) {
2071                 struct seq_file *m = file->private_data;
2072                 m->private = trace_types;
2073         }
2074
2075         return ret;
2076 }
2077
2078 static struct file_operations tracing_fops = {
2079         .open           = tracing_open,
2080         .read           = seq_read,
2081         .llseek         = seq_lseek,
2082         .release        = tracing_release,
2083 };
2084
2085 static struct file_operations tracing_lt_fops = {
2086         .open           = tracing_lt_open,
2087         .read           = seq_read,
2088         .llseek         = seq_lseek,
2089         .release        = tracing_release,
2090 };
2091
2092 static struct file_operations show_traces_fops = {
2093         .open           = show_traces_open,
2094         .read           = seq_read,
2095         .release        = seq_release,
2096 };
2097
2098 /*
2099  * Only trace on a CPU if the bitmask is set:
2100  */
2101 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2102
2103 /*
2104  * When tracing/tracing_cpu_mask is modified then this holds
2105  * the new bitmask we are about to install:
2106  */
2107 static cpumask_t tracing_cpumask_new;
2108
2109 /*
2110  * The tracer itself will not take this lock, but still we want
2111  * to provide a consistent cpumask to user-space:
2112  */
2113 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2114
2115 /*
2116  * Temporary storage for the character representation of the
2117  * CPU bitmask (and one more byte for the newline):
2118  */
2119 static char mask_str[NR_CPUS + 1];
2120
2121 static ssize_t
2122 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2123                      size_t count, loff_t *ppos)
2124 {
2125         int len;
2126
2127         mutex_lock(&tracing_cpumask_update_lock);
2128
2129         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2130         if (count - len < 2) {
2131                 count = -EINVAL;
2132                 goto out_err;
2133         }
2134         len += sprintf(mask_str + len, "\n");
2135         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2136
2137 out_err:
2138         mutex_unlock(&tracing_cpumask_update_lock);
2139
2140         return count;
2141 }
2142
2143 static ssize_t
2144 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2145                       size_t count, loff_t *ppos)
2146 {
2147         int err, cpu;
2148
2149         mutex_lock(&tracing_cpumask_update_lock);
2150         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2151         if (err)
2152                 goto err_unlock;
2153
2154         raw_local_irq_disable();
2155         __raw_spin_lock(&ftrace_max_lock);
2156         for_each_tracing_cpu(cpu) {
2157                 /*
2158                  * Increase/decrease the disabled counter if we are
2159                  * about to flip a bit in the cpumask:
2160                  */
2161                 if (cpu_isset(cpu, tracing_cpumask) &&
2162                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2163                         atomic_inc(&global_trace.data[cpu]->disabled);
2164                 }
2165                 if (!cpu_isset(cpu, tracing_cpumask) &&
2166                                 cpu_isset(cpu, tracing_cpumask_new)) {
2167                         atomic_dec(&global_trace.data[cpu]->disabled);
2168                 }
2169         }
2170         __raw_spin_unlock(&ftrace_max_lock);
2171         raw_local_irq_enable();
2172
2173         tracing_cpumask = tracing_cpumask_new;
2174
2175         mutex_unlock(&tracing_cpumask_update_lock);
2176
2177         return count;
2178
2179 err_unlock:
2180         mutex_unlock(&tracing_cpumask_update_lock);
2181
2182         return err;
2183 }
2184
2185 static struct file_operations tracing_cpumask_fops = {
2186         .open           = tracing_open_generic,
2187         .read           = tracing_cpumask_read,
2188         .write          = tracing_cpumask_write,
2189 };
2190
2191 static ssize_t
2192 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2193                        size_t cnt, loff_t *ppos)
2194 {
2195         char *buf;
2196         int r = 0;
2197         int len = 0;
2198         int i;
2199
2200         /* calulate max size */
2201         for (i = 0; trace_options[i]; i++) {
2202                 len += strlen(trace_options[i]);
2203                 len += 3; /* "no" and space */
2204         }
2205
2206         /* +2 for \n and \0 */
2207         buf = kmalloc(len + 2, GFP_KERNEL);
2208         if (!buf)
2209                 return -ENOMEM;
2210
2211         for (i = 0; trace_options[i]; i++) {
2212                 if (trace_flags & (1 << i))
2213                         r += sprintf(buf + r, "%s ", trace_options[i]);
2214                 else
2215                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2216         }
2217
2218         r += sprintf(buf + r, "\n");
2219         WARN_ON(r >= len + 2);
2220
2221         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2222
2223         kfree(buf);
2224
2225         return r;
2226 }
2227
2228 static ssize_t
2229 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2230                         size_t cnt, loff_t *ppos)
2231 {
2232         char buf[64];
2233         char *cmp = buf;
2234         int neg = 0;
2235         int i;
2236
2237         if (cnt >= sizeof(buf))
2238                 return -EINVAL;
2239
2240         if (copy_from_user(&buf, ubuf, cnt))
2241                 return -EFAULT;
2242
2243         buf[cnt] = 0;
2244
2245         if (strncmp(buf, "no", 2) == 0) {
2246                 neg = 1;
2247                 cmp += 2;
2248         }
2249
2250         for (i = 0; trace_options[i]; i++) {
2251                 int len = strlen(trace_options[i]);
2252
2253                 if (strncmp(cmp, trace_options[i], len) == 0) {
2254                         if (neg)
2255                                 trace_flags &= ~(1 << i);
2256                         else
2257                                 trace_flags |= (1 << i);
2258                         break;
2259                 }
2260         }
2261         /*
2262          * If no option could be set, return an error:
2263          */
2264         if (!trace_options[i])
2265                 return -EINVAL;
2266
2267         filp->f_pos += cnt;
2268
2269         return cnt;
2270 }
2271
2272 static struct file_operations tracing_iter_fops = {
2273         .open           = tracing_open_generic,
2274         .read           = tracing_iter_ctrl_read,
2275         .write          = tracing_iter_ctrl_write,
2276 };
2277
2278 static const char readme_msg[] =
2279         "tracing mini-HOWTO:\n\n"
2280         "# mkdir /debug\n"
2281         "# mount -t debugfs nodev /debug\n\n"
2282         "# cat /debug/tracing/available_tracers\n"
2283         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2284         "# cat /debug/tracing/current_tracer\n"
2285         "none\n"
2286         "# echo sched_switch > /debug/tracing/current_tracer\n"
2287         "# cat /debug/tracing/current_tracer\n"
2288         "sched_switch\n"
2289         "# cat /debug/tracing/iter_ctrl\n"
2290         "noprint-parent nosym-offset nosym-addr noverbose\n"
2291         "# echo print-parent > /debug/tracing/iter_ctrl\n"
2292         "# echo 1 > /debug/tracing/tracing_enabled\n"
2293         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2294         "echo 0 > /debug/tracing/tracing_enabled\n"
2295 ;
2296
2297 static ssize_t
2298 tracing_readme_read(struct file *filp, char __user *ubuf,
2299                        size_t cnt, loff_t *ppos)
2300 {
2301         return simple_read_from_buffer(ubuf, cnt, ppos,
2302                                         readme_msg, strlen(readme_msg));
2303 }
2304
2305 static struct file_operations tracing_readme_fops = {
2306         .open           = tracing_open_generic,
2307         .read           = tracing_readme_read,
2308 };
2309
2310 static ssize_t
2311 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2312                   size_t cnt, loff_t *ppos)
2313 {
2314         struct trace_array *tr = filp->private_data;
2315         char buf[64];
2316         int r;
2317
2318         r = sprintf(buf, "%ld\n", tr->ctrl);
2319         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2320 }
2321
2322 static ssize_t
2323 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2324                    size_t cnt, loff_t *ppos)
2325 {
2326         struct trace_array *tr = filp->private_data;
2327         char buf[64];
2328         long val;
2329         int ret;
2330
2331         if (cnt >= sizeof(buf))
2332                 return -EINVAL;
2333
2334         if (copy_from_user(&buf, ubuf, cnt))
2335                 return -EFAULT;
2336
2337         buf[cnt] = 0;
2338
2339         ret = strict_strtoul(buf, 10, &val);
2340         if (ret < 0)
2341                 return ret;
2342
2343         val = !!val;
2344
2345         mutex_lock(&trace_types_lock);
2346         if (tr->ctrl ^ val) {
2347                 if (val)
2348                         tracer_enabled = 1;
2349                 else
2350                         tracer_enabled = 0;
2351
2352                 tr->ctrl = val;
2353
2354                 if (current_trace && current_trace->ctrl_update)
2355                         current_trace->ctrl_update(tr);
2356         }
2357         mutex_unlock(&trace_types_lock);
2358
2359         filp->f_pos += cnt;
2360
2361         return cnt;
2362 }
2363
2364 static ssize_t
2365 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2366                        size_t cnt, loff_t *ppos)
2367 {
2368         char buf[max_tracer_type_len+2];
2369         int r;
2370
2371         mutex_lock(&trace_types_lock);
2372         if (current_trace)
2373                 r = sprintf(buf, "%s\n", current_trace->name);
2374         else
2375                 r = sprintf(buf, "\n");
2376         mutex_unlock(&trace_types_lock);
2377
2378         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2379 }
2380
2381 static ssize_t
2382 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2383                         size_t cnt, loff_t *ppos)
2384 {
2385         struct trace_array *tr = &global_trace;
2386         struct tracer *t;
2387         char buf[max_tracer_type_len+1];
2388         int i;
2389         size_t ret;
2390
2391         ret = cnt;
2392
2393         if (cnt > max_tracer_type_len)
2394                 cnt = max_tracer_type_len;
2395
2396         if (copy_from_user(&buf, ubuf, cnt))
2397                 return -EFAULT;
2398
2399         buf[cnt] = 0;
2400
2401         /* strip ending whitespace. */
2402         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2403                 buf[i] = 0;
2404
2405         mutex_lock(&trace_types_lock);
2406         for (t = trace_types; t; t = t->next) {
2407                 if (strcmp(t->name, buf) == 0)
2408                         break;
2409         }
2410         if (!t) {
2411                 ret = -EINVAL;
2412                 goto out;
2413         }
2414         if (t == current_trace)
2415                 goto out;
2416
2417         if (current_trace && current_trace->reset)
2418                 current_trace->reset(tr);
2419
2420         current_trace = t;
2421         if (t->init)
2422                 t->init(tr);
2423
2424  out:
2425         mutex_unlock(&trace_types_lock);
2426
2427         if (ret > 0)
2428                 filp->f_pos += ret;
2429
2430         return ret;
2431 }
2432
2433 static ssize_t
2434 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2435                      size_t cnt, loff_t *ppos)
2436 {
2437         unsigned long *ptr = filp->private_data;
2438         char buf[64];
2439         int r;
2440
2441         r = snprintf(buf, sizeof(buf), "%ld\n",
2442                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2443         if (r > sizeof(buf))
2444                 r = sizeof(buf);
2445         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2446 }
2447
2448 static ssize_t
2449 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2450                       size_t cnt, loff_t *ppos)
2451 {
2452         long *ptr = filp->private_data;
2453         char buf[64];
2454         long val;
2455         int ret;
2456
2457         if (cnt >= sizeof(buf))
2458                 return -EINVAL;
2459
2460         if (copy_from_user(&buf, ubuf, cnt))
2461                 return -EFAULT;
2462
2463         buf[cnt] = 0;
2464
2465         ret = strict_strtoul(buf, 10, &val);
2466         if (ret < 0)
2467                 return ret;
2468
2469         *ptr = val * 1000;
2470
2471         return cnt;
2472 }
2473
2474 static atomic_t tracing_reader;
2475
2476 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2477 {
2478         struct trace_iterator *iter;
2479
2480         if (tracing_disabled)
2481                 return -ENODEV;
2482
2483         /* We only allow for reader of the pipe */
2484         if (atomic_inc_return(&tracing_reader) != 1) {
2485                 atomic_dec(&tracing_reader);
2486                 return -EBUSY;
2487         }
2488
2489         /* create a buffer to store the information to pass to userspace */
2490         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2491         if (!iter)
2492                 return -ENOMEM;
2493
2494         mutex_lock(&trace_types_lock);
2495         iter->tr = &global_trace;
2496         iter->trace = current_trace;
2497         filp->private_data = iter;
2498
2499         if (iter->trace->pipe_open)
2500                 iter->trace->pipe_open(iter);
2501         mutex_unlock(&trace_types_lock);
2502
2503         return 0;
2504 }
2505
2506 static int tracing_release_pipe(struct inode *inode, struct file *file)
2507 {
2508         struct trace_iterator *iter = file->private_data;
2509
2510         kfree(iter);
2511         atomic_dec(&tracing_reader);
2512
2513         return 0;
2514 }
2515
2516 static unsigned int
2517 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2518 {
2519         struct trace_iterator *iter = filp->private_data;
2520
2521         if (trace_flags & TRACE_ITER_BLOCK) {
2522                 /*
2523                  * Always select as readable when in blocking mode
2524                  */
2525                 return POLLIN | POLLRDNORM;
2526         } else {
2527                 if (!trace_empty(iter))
2528                         return POLLIN | POLLRDNORM;
2529                 poll_wait(filp, &trace_wait, poll_table);
2530                 if (!trace_empty(iter))
2531                         return POLLIN | POLLRDNORM;
2532
2533                 return 0;
2534         }
2535 }
2536
2537 /*
2538  * Consumer reader.
2539  */
2540 static ssize_t
2541 tracing_read_pipe(struct file *filp, char __user *ubuf,
2542                   size_t cnt, loff_t *ppos)
2543 {
2544         struct trace_iterator *iter = filp->private_data;
2545         ssize_t sret;
2546
2547         /* return any leftover data */
2548         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2549         if (sret != -EBUSY)
2550                 return sret;
2551
2552         trace_seq_reset(&iter->seq);
2553
2554         mutex_lock(&trace_types_lock);
2555         if (iter->trace->read) {
2556                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2557                 if (sret)
2558                         goto out;
2559         }
2560
2561 waitagain:
2562         sret = 0;
2563         while (trace_empty(iter)) {
2564
2565                 if ((filp->f_flags & O_NONBLOCK)) {
2566                         sret = -EAGAIN;
2567                         goto out;
2568                 }
2569
2570                 /*
2571                  * This is a make-shift waitqueue. The reason we don't use
2572                  * an actual wait queue is because:
2573                  *  1) we only ever have one waiter
2574                  *  2) the tracing, traces all functions, we don't want
2575                  *     the overhead of calling wake_up and friends
2576                  *     (and tracing them too)
2577                  *     Anyway, this is really very primitive wakeup.
2578                  */
2579                 set_current_state(TASK_INTERRUPTIBLE);
2580                 iter->tr->waiter = current;
2581
2582                 mutex_unlock(&trace_types_lock);
2583
2584                 /* sleep for 100 msecs, and try again. */
2585                 schedule_timeout(HZ/10);
2586
2587                 mutex_lock(&trace_types_lock);
2588
2589                 iter->tr->waiter = NULL;
2590
2591                 if (signal_pending(current)) {
2592                         sret = -EINTR;
2593                         goto out;
2594                 }
2595
2596                 if (iter->trace != current_trace)
2597                         goto out;
2598
2599                 /*
2600                  * We block until we read something and tracing is disabled.
2601                  * We still block if tracing is disabled, but we have never
2602                  * read anything. This allows a user to cat this file, and
2603                  * then enable tracing. But after we have read something,
2604                  * we give an EOF when tracing is again disabled.
2605                  *
2606                  * iter->pos will be 0 if we haven't read anything.
2607                  */
2608                 if (!tracer_enabled && iter->pos)
2609                         break;
2610
2611                 continue;
2612         }
2613
2614         /* stop when tracing is finished */
2615         if (trace_empty(iter))
2616                 goto out;
2617
2618         if (cnt >= PAGE_SIZE)
2619                 cnt = PAGE_SIZE - 1;
2620
2621         /* reset all but tr, trace, and overruns */
2622         memset(&iter->seq, 0,
2623                sizeof(struct trace_iterator) -
2624                offsetof(struct trace_iterator, seq));
2625         iter->pos = -1;
2626
2627         while (find_next_entry_inc(iter) != NULL) {
2628                 enum print_line_t ret;
2629                 int len = iter->seq.len;
2630
2631                 ret = print_trace_line(iter);
2632                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2633                         /* don't print partial lines */
2634                         iter->seq.len = len;
2635                         break;
2636                 }
2637
2638                 trace_consume(iter);
2639
2640                 if (iter->seq.len >= cnt)
2641                         break;
2642         }
2643
2644         /* Now copy what we have to the user */
2645         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2646         if (iter->seq.readpos >= iter->seq.len)
2647                 trace_seq_reset(&iter->seq);
2648
2649         /*
2650          * If there was nothing to send to user, inspite of consuming trace
2651          * entries, go back to wait for more entries.
2652          */
2653         if (sret == -EBUSY)
2654                 goto waitagain;
2655
2656 out:
2657         mutex_unlock(&trace_types_lock);
2658
2659         return sret;
2660 }
2661
2662 static ssize_t
2663 tracing_entries_read(struct file *filp, char __user *ubuf,
2664                      size_t cnt, loff_t *ppos)
2665 {
2666         struct trace_array *tr = filp->private_data;
2667         char buf[64];
2668         int r;
2669
2670         r = sprintf(buf, "%lu\n", tr->entries);
2671         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2672 }
2673
2674 static ssize_t
2675 tracing_entries_write(struct file *filp, const char __user *ubuf,
2676                       size_t cnt, loff_t *ppos)
2677 {
2678         unsigned long val;
2679         char buf[64];
2680         int ret, cpu;
2681         struct trace_array *tr = filp->private_data;
2682
2683         if (cnt >= sizeof(buf))
2684                 return -EINVAL;
2685
2686         if (copy_from_user(&buf, ubuf, cnt))
2687                 return -EFAULT;
2688
2689         buf[cnt] = 0;
2690
2691         ret = strict_strtoul(buf, 10, &val);
2692         if (ret < 0)
2693                 return ret;
2694
2695         /* must have at least 1 entry */
2696         if (!val)
2697                 return -EINVAL;
2698
2699         mutex_lock(&trace_types_lock);
2700
2701         if (tr->ctrl) {
2702                 cnt = -EBUSY;
2703                 pr_info("ftrace: please disable tracing"
2704                         " before modifying buffer size\n");
2705                 goto out;
2706         }
2707
2708         /* disable all cpu buffers */
2709         for_each_tracing_cpu(cpu) {
2710                 if (global_trace.data[cpu])
2711                         atomic_inc(&global_trace.data[cpu]->disabled);
2712                 if (max_tr.data[cpu])
2713                         atomic_inc(&max_tr.data[cpu]->disabled);
2714         }
2715
2716         if (val != global_trace.entries) {
2717                 ret = ring_buffer_resize(global_trace.buffer, val);
2718                 if (ret < 0) {
2719                         cnt = ret;
2720                         goto out;
2721                 }
2722
2723                 ret = ring_buffer_resize(max_tr.buffer, val);
2724                 if (ret < 0) {
2725                         int r;
2726                         cnt = ret;
2727                         r = ring_buffer_resize(global_trace.buffer,
2728                                                global_trace.entries);
2729                         if (r < 0) {
2730                                 /* AARGH! We are left with different
2731                                  * size max buffer!!!! */
2732                                 WARN_ON(1);
2733                                 tracing_disabled = 1;
2734                         }
2735                         goto out;
2736                 }
2737
2738                 global_trace.entries = val;
2739         }
2740
2741         filp->f_pos += cnt;
2742
2743         /* If check pages failed, return ENOMEM */
2744         if (tracing_disabled)
2745                 cnt = -ENOMEM;
2746  out:
2747         for_each_tracing_cpu(cpu) {
2748                 if (global_trace.data[cpu])
2749                         atomic_dec(&global_trace.data[cpu]->disabled);
2750                 if (max_tr.data[cpu])
2751                         atomic_dec(&max_tr.data[cpu]->disabled);
2752         }
2753
2754         max_tr.entries = global_trace.entries;
2755         mutex_unlock(&trace_types_lock);
2756
2757         return cnt;
2758 }
2759
2760 static int mark_printk(const char *fmt, ...)
2761 {
2762         int ret;
2763         va_list args;
2764         va_start(args, fmt);
2765         ret = trace_vprintk(0, fmt, args);
2766         va_end(args);
2767         return ret;
2768 }
2769
2770 static ssize_t
2771 tracing_mark_write(struct file *filp, const char __user *ubuf,
2772                                         size_t cnt, loff_t *fpos)
2773 {
2774         char *buf;
2775         char *end;
2776         struct trace_array *tr = &global_trace;
2777
2778         if (!tr->ctrl || tracing_disabled)
2779                 return -EINVAL;
2780
2781         if (cnt > TRACE_BUF_SIZE)
2782                 cnt = TRACE_BUF_SIZE;
2783
2784         buf = kmalloc(cnt + 1, GFP_KERNEL);
2785         if (buf == NULL)
2786                 return -ENOMEM;
2787
2788         if (copy_from_user(buf, ubuf, cnt)) {
2789                 kfree(buf);
2790                 return -EFAULT;
2791         }
2792
2793         /* Cut from the first nil or newline. */
2794         buf[cnt] = '\0';
2795         end = strchr(buf, '\n');
2796         if (end)
2797                 *end = '\0';
2798
2799         cnt = mark_printk("%s\n", buf);
2800         kfree(buf);
2801         *fpos += cnt;
2802
2803         return cnt;
2804 }
2805
2806 static struct file_operations tracing_max_lat_fops = {
2807         .open           = tracing_open_generic,
2808         .read           = tracing_max_lat_read,
2809         .write          = tracing_max_lat_write,
2810 };
2811
2812 static struct file_operations tracing_ctrl_fops = {
2813         .open           = tracing_open_generic,
2814         .read           = tracing_ctrl_read,
2815         .write          = tracing_ctrl_write,
2816 };
2817
2818 static struct file_operations set_tracer_fops = {
2819         .open           = tracing_open_generic,
2820         .read           = tracing_set_trace_read,
2821         .write          = tracing_set_trace_write,
2822 };
2823
2824 static struct file_operations tracing_pipe_fops = {
2825         .open           = tracing_open_pipe,
2826         .poll           = tracing_poll_pipe,
2827         .read           = tracing_read_pipe,
2828         .release        = tracing_release_pipe,
2829 };
2830
2831 static struct file_operations tracing_entries_fops = {
2832         .open           = tracing_open_generic,
2833         .read           = tracing_entries_read,
2834         .write          = tracing_entries_write,
2835 };
2836
2837 static struct file_operations tracing_mark_fops = {
2838         .open           = tracing_open_generic,
2839         .write          = tracing_mark_write,
2840 };
2841
2842 #ifdef CONFIG_DYNAMIC_FTRACE
2843
2844 static ssize_t
2845 tracing_read_long(struct file *filp, char __user *ubuf,
2846                   size_t cnt, loff_t *ppos)
2847 {
2848         unsigned long *p = filp->private_data;
2849         char buf[64];
2850         int r;
2851
2852         r = sprintf(buf, "%ld\n", *p);
2853
2854         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2855 }
2856
2857 static struct file_operations tracing_read_long_fops = {
2858         .open           = tracing_open_generic,
2859         .read           = tracing_read_long,
2860 };
2861 #endif
2862
2863 static struct dentry *d_tracer;
2864
2865 struct dentry *tracing_init_dentry(void)
2866 {
2867         static int once;
2868
2869         if (d_tracer)
2870                 return d_tracer;
2871
2872         d_tracer = debugfs_create_dir("tracing", NULL);
2873
2874         if (!d_tracer && !once) {
2875                 once = 1;
2876                 pr_warning("Could not create debugfs directory 'tracing'\n");
2877                 return NULL;
2878         }
2879
2880         return d_tracer;
2881 }
2882
2883 #ifdef CONFIG_FTRACE_SELFTEST
2884 /* Let selftest have access to static functions in this file */
2885 #include "trace_selftest.c"
2886 #endif
2887
2888 static __init int tracer_init_debugfs(void)
2889 {
2890         struct dentry *d_tracer;
2891         struct dentry *entry;
2892
2893         d_tracer = tracing_init_dentry();
2894
2895         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2896                                     &global_trace, &tracing_ctrl_fops);
2897         if (!entry)
2898                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2899
2900         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2901                                     NULL, &tracing_iter_fops);
2902         if (!entry)
2903                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2904
2905         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2906                                     NULL, &tracing_cpumask_fops);
2907         if (!entry)
2908                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2909
2910         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2911                                     &global_trace, &tracing_lt_fops);
2912         if (!entry)
2913                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2914
2915         entry = debugfs_create_file("trace", 0444, d_tracer,
2916                                     &global_trace, &tracing_fops);
2917         if (!entry)
2918                 pr_warning("Could not create debugfs 'trace' entry\n");
2919
2920         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2921                                     &global_trace, &show_traces_fops);
2922         if (!entry)
2923                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
2924
2925         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2926                                     &global_trace, &set_tracer_fops);
2927         if (!entry)
2928                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
2929
2930         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2931                                     &tracing_max_latency,
2932                                     &tracing_max_lat_fops);
2933         if (!entry)
2934                 pr_warning("Could not create debugfs "
2935                            "'tracing_max_latency' entry\n");
2936
2937         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2938                                     &tracing_thresh, &tracing_max_lat_fops);
2939         if (!entry)
2940                 pr_warning("Could not create debugfs "
2941                            "'tracing_thresh' entry\n");
2942         entry = debugfs_create_file("README", 0644, d_tracer,
2943                                     NULL, &tracing_readme_fops);
2944         if (!entry)
2945                 pr_warning("Could not create debugfs 'README' entry\n");
2946
2947         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2948                                     NULL, &tracing_pipe_fops);
2949         if (!entry)
2950                 pr_warning("Could not create debugfs "
2951                            "'trace_pipe' entry\n");
2952
2953         entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2954                                     &global_trace, &tracing_entries_fops);
2955         if (!entry)
2956                 pr_warning("Could not create debugfs "
2957                            "'trace_entries' entry\n");
2958
2959         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
2960                                     NULL, &tracing_mark_fops);
2961         if (!entry)
2962                 pr_warning("Could not create debugfs "
2963                            "'trace_marker' entry\n");
2964
2965 #ifdef CONFIG_DYNAMIC_FTRACE
2966         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2967                                     &ftrace_update_tot_cnt,
2968                                     &tracing_read_long_fops);
2969         if (!entry)
2970                 pr_warning("Could not create debugfs "
2971                            "'dyn_ftrace_total_info' entry\n");
2972 #endif
2973 #ifdef CONFIG_SYSPROF_TRACER
2974         init_tracer_sysprof_debugfs(d_tracer);
2975 #endif
2976         return 0;
2977 }
2978
2979 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2980 {
2981         static DEFINE_SPINLOCK(trace_buf_lock);
2982         static char trace_buf[TRACE_BUF_SIZE];
2983
2984         struct ring_buffer_event *event;
2985         struct trace_array *tr = &global_trace;
2986         struct trace_array_cpu *data;
2987         struct print_entry *entry;
2988         unsigned long flags, irq_flags;
2989         int cpu, len = 0, size, pc;
2990
2991         if (!tr->ctrl || tracing_disabled)
2992                 return 0;
2993
2994         pc = preempt_count();
2995         preempt_disable_notrace();
2996         cpu = raw_smp_processor_id();
2997         data = tr->data[cpu];
2998
2999         if (unlikely(atomic_read(&data->disabled)))
3000                 goto out;
3001
3002         spin_lock_irqsave(&trace_buf_lock, flags);
3003         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3004
3005         len = min(len, TRACE_BUF_SIZE-1);
3006         trace_buf[len] = 0;
3007
3008         size = sizeof(*entry) + len + 1;
3009         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3010         if (!event)
3011                 goto out_unlock;
3012         entry = ring_buffer_event_data(event);
3013         tracing_generic_entry_update(&entry->ent, flags, pc);
3014         entry->ent.type                 = TRACE_PRINT;
3015         entry->ip                       = ip;
3016
3017         memcpy(&entry->buf, trace_buf, len);
3018         entry->buf[len] = 0;
3019         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3020
3021  out_unlock:
3022         spin_unlock_irqrestore(&trace_buf_lock, flags);
3023
3024  out:
3025         preempt_enable_notrace();
3026
3027         return len;
3028 }
3029 EXPORT_SYMBOL_GPL(trace_vprintk);
3030
3031 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3032 {
3033         int ret;
3034         va_list ap;
3035
3036         if (!(trace_flags & TRACE_ITER_PRINTK))
3037                 return 0;
3038
3039         va_start(ap, fmt);
3040         ret = trace_vprintk(ip, fmt, ap);
3041         va_end(ap);
3042         return ret;
3043 }
3044 EXPORT_SYMBOL_GPL(__ftrace_printk);
3045
3046 static int trace_panic_handler(struct notifier_block *this,
3047                                unsigned long event, void *unused)
3048 {
3049         ftrace_dump();
3050         return NOTIFY_OK;
3051 }
3052
3053 static struct notifier_block trace_panic_notifier = {
3054         .notifier_call  = trace_panic_handler,
3055         .next           = NULL,
3056         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3057 };
3058
3059 static int trace_die_handler(struct notifier_block *self,
3060                              unsigned long val,
3061                              void *data)
3062 {
3063         switch (val) {
3064         case DIE_OOPS:
3065                 ftrace_dump();
3066                 break;
3067         default:
3068                 break;
3069         }
3070         return NOTIFY_OK;
3071 }
3072
3073 static struct notifier_block trace_die_notifier = {
3074         .notifier_call = trace_die_handler,
3075         .priority = 200
3076 };
3077
3078 /*
3079  * printk is set to max of 1024, we really don't need it that big.
3080  * Nothing should be printing 1000 characters anyway.
3081  */
3082 #define TRACE_MAX_PRINT         1000
3083
3084 /*
3085  * Define here KERN_TRACE so that we have one place to modify
3086  * it if we decide to change what log level the ftrace dump
3087  * should be at.
3088  */
3089 #define KERN_TRACE              KERN_INFO
3090
3091 static void
3092 trace_printk_seq(struct trace_seq *s)
3093 {
3094         /* Probably should print a warning here. */
3095         if (s->len >= 1000)
3096                 s->len = 1000;
3097
3098         /* should be zero ended, but we are paranoid. */
3099         s->buffer[s->len] = 0;
3100
3101         printk(KERN_TRACE "%s", s->buffer);
3102
3103         trace_seq_reset(s);
3104 }
3105
3106
3107 void ftrace_dump(void)
3108 {
3109         static DEFINE_SPINLOCK(ftrace_dump_lock);
3110         /* use static because iter can be a bit big for the stack */
3111         static struct trace_iterator iter;
3112         static cpumask_t mask;
3113         static int dump_ran;
3114         unsigned long flags;
3115         int cnt = 0, cpu;
3116
3117         /* only one dump */
3118         spin_lock_irqsave(&ftrace_dump_lock, flags);
3119         if (dump_ran)
3120                 goto out;
3121
3122         dump_ran = 1;
3123
3124         /* No turning back! */
3125         ftrace_kill();
3126
3127         for_each_tracing_cpu(cpu) {
3128                 atomic_inc(&global_trace.data[cpu]->disabled);
3129         }
3130
3131         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3132
3133         iter.tr = &global_trace;
3134         iter.trace = current_trace;
3135
3136         /*
3137          * We need to stop all tracing on all CPUS to read the
3138          * the next buffer. This is a bit expensive, but is
3139          * not done often. We fill all what we can read,
3140          * and then release the locks again.
3141          */
3142
3143         cpus_clear(mask);
3144
3145         while (!trace_empty(&iter)) {
3146
3147                 if (!cnt)
3148                         printk(KERN_TRACE "---------------------------------\n");
3149
3150                 cnt++;
3151
3152                 /* reset all but tr, trace, and overruns */
3153                 memset(&iter.seq, 0,
3154                        sizeof(struct trace_iterator) -
3155                        offsetof(struct trace_iterator, seq));
3156                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3157                 iter.pos = -1;
3158
3159                 if (find_next_entry_inc(&iter) != NULL) {
3160                         print_trace_line(&iter);
3161                         trace_consume(&iter);
3162                 }
3163
3164                 trace_printk_seq(&iter.seq);
3165         }
3166
3167         if (!cnt)
3168                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3169         else
3170                 printk(KERN_TRACE "---------------------------------\n");
3171
3172  out:
3173         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3174 }
3175
3176 __init static int tracer_alloc_buffers(void)
3177 {
3178         struct trace_array_cpu *data;
3179         int i;
3180
3181         /* TODO: make the number of buffers hot pluggable with CPUS */
3182         tracing_buffer_mask = cpu_possible_map;
3183
3184         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3185                                                    TRACE_BUFFER_FLAGS);
3186         if (!global_trace.buffer) {
3187                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3188                 WARN_ON(1);
3189                 return 0;
3190         }
3191         global_trace.entries = ring_buffer_size(global_trace.buffer);
3192
3193 #ifdef CONFIG_TRACER_MAX_TRACE
3194         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3195                                              TRACE_BUFFER_FLAGS);
3196         if (!max_tr.buffer) {
3197                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3198                 WARN_ON(1);
3199                 ring_buffer_free(global_trace.buffer);
3200                 return 0;
3201         }
3202         max_tr.entries = ring_buffer_size(max_tr.buffer);
3203         WARN_ON(max_tr.entries != global_trace.entries);
3204 #endif
3205
3206         /* Allocate the first page for all buffers */
3207         for_each_tracing_cpu(i) {
3208                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3209                 max_tr.data[i] = &per_cpu(max_data, i);
3210         }
3211
3212         trace_init_cmdlines();
3213
3214         register_tracer(&nop_trace);
3215 #ifdef CONFIG_BOOT_TRACER
3216         register_tracer(&boot_tracer);
3217         current_trace = &boot_tracer;
3218         current_trace->init(&global_trace);
3219 #else
3220         current_trace = &nop_trace;
3221 #endif
3222
3223         /* All seems OK, enable tracing */
3224         global_trace.ctrl = tracer_enabled;
3225         tracing_disabled = 0;
3226
3227         atomic_notifier_chain_register(&panic_notifier_list,
3228                                        &trace_panic_notifier);
3229
3230         register_die_notifier(&trace_die_notifier);
3231
3232         return 0;
3233 }
3234 early_initcall(tracer_alloc_buffers);
3235 fs_initcall(tracer_init_debugfs);