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