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/debugfs.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/linkage.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/ftrace.h>
  23 #include <linux/module.h>
  24 #include <linux/percpu.h>
  25 #include <linux/ctype.h>
  26 #include <linux/init.h>
  27 #include <linux/poll.h>
  28 #include <linux/gfp.h>
  29 #include <linux/fs.h>
  30 #include <linux/kprobes.h>
  31 #include <linux/writeback.h>
  32
  33 #include <linux/stacktrace.h>
  34
  35 #include "trace.h"
  36
  37 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  38 unsigned long __read_mostly     tracing_thresh;
  39
  40 static unsigned long __read_mostly      tracing_nr_buffers;
  41 static cpumask_t __read_mostly          tracing_buffer_mask;
  42
  43 #define for_each_tracing_cpu(cpu)       \
  44         for_each_cpu_mask(cpu, tracing_buffer_mask)
  45
  46 static int trace_alloc_page(void);
  47 static int trace_free_page(void);
  48
  49 static int tracing_disabled = 1;
  50
  51 static unsigned long tracing_pages_allocated;
  52
  53 long
  54 ns2usecs(cycle_t nsec)
  55 {
  56         nsec += 500;
  57         do_div(nsec, 1000);
  58         return nsec;
  59 }
  60
  61 cycle_t ftrace_now(int cpu)
  62 {
  63         return cpu_clock(cpu);
  64 }
  65
  66 /*
  67  * The global_trace is the descriptor that holds the tracing
  68  * buffers for the live tracing. For each CPU, it contains
  69  * a link list of pages that will store trace entries. The
  70  * page descriptor of the pages in the memory is used to hold
  71  * the link list by linking the lru item in the page descriptor
  72  * to each of the pages in the buffer per CPU.
  73  *
  74  * For each active CPU there is a data field that holds the
  75  * pages for the buffer for that CPU. Each CPU has the same number
  76  * of pages allocated for its buffer.
  77  */
  78 static struct trace_array       global_trace;
  79
  80 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  81
  82 /*
  83  * The max_tr is used to snapshot the global_trace when a maximum
  84  * latency is reached. Some tracers will use this to store a maximum
  85  * trace while it continues examining live traces.
  86  *
  87  * The buffers for the max_tr are set up the same as the global_trace.
  88  * When a snapshot is taken, the link list of the max_tr is swapped
  89  * with the link list of the global_trace and the buffers are reset for
  90  * the global_trace so the tracing can continue.
  91  */
  92 static struct trace_array       max_tr;
  93
  94 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  95
  96 /* tracer_enabled is used to toggle activation of a tracer */
  97 static int                      tracer_enabled = 1;
  98
  99 /* function tracing enabled */
 100 int                             ftrace_function_enabled;
 101
 102 /*
 103  * trace_nr_entries is the number of entries that is allocated
 104  * for a buffer. Note, the number of entries is always rounded
 105  * to ENTRIES_PER_PAGE.
 106  */
 107 static unsigned long            trace_nr_entries = 65536UL;
 108
 109 /* trace_types holds a link list of available tracers. */
 110 static struct tracer            *trace_types __read_mostly;
 111
 112 /* current_trace points to the tracer that is currently active */
 113 static struct tracer            *current_trace __read_mostly;
 114
 115 /*
 116  * max_tracer_type_len is used to simplify the allocating of
 117  * buffers to read userspace tracer names. We keep track of
 118  * the longest tracer name registered.
 119  */
 120 static int                      max_tracer_type_len;
 121
 122 /*
 123  * trace_types_lock is used to protect the trace_types list.
 124  * This lock is also used to keep user access serialized.
 125  * Accesses from userspace will grab this lock while userspace
 126  * activities happen inside the kernel.
 127  */
 128 static DEFINE_MUTEX(trace_types_lock);
 129
 130 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
 131 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
 132
 133 /* trace_flags holds iter_ctrl options */
 134 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
 135
 136 static notrace void no_trace_init(struct trace_array *tr)
 137 {
 138         int cpu;
 139
 140         ftrace_function_enabled = 0;
 141         if(tr->ctrl)
 142                 for_each_online_cpu(cpu)
 143                         tracing_reset(tr->data[cpu]);
 144         tracer_enabled = 0;
 145 }
 146
 147 /* dummy trace to disable tracing */
 148 static struct tracer no_tracer __read_mostly = {
 149         .name           = "none",
 150         .init           = no_trace_init
 151 };
 152
 153
 154 /**
 155  * trace_wake_up - wake up tasks waiting for trace input
 156  *
 157  * Simply wakes up any task that is blocked on the trace_wait
 158  * queue. These is used with trace_poll for tasks polling the trace.
 159  */
 160 void trace_wake_up(void)
 161 {
 162         /*
 163          * The runqueue_is_locked() can fail, but this is the best we
 164          * have for now:
 165          */
 166         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
 167                 wake_up(&trace_wait);
 168 }
 169
 170 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
 171
 172 static int __init set_nr_entries(char *str)
 173 {
 174         unsigned long nr_entries;
 175         int ret;
 176
 177         if (!str)
 178                 return 0;
 179         ret = strict_strtoul(str, 0, &nr_entries);
 180         /* nr_entries can not be zero */
 181         if (ret < 0 || nr_entries == 0)
 182                 return 0;
 183         trace_nr_entries = nr_entries;
 184         return 1;
 185 }
 186 __setup("trace_entries=", set_nr_entries);
 187
 188 unsigned long nsecs_to_usecs(unsigned long nsecs)
 189 {
 190         return nsecs / 1000;
 191 }
 192
 193 /*
 194  * trace_flag_type is an enumeration that holds different
 195  * states when a trace occurs. These are:
 196  *  IRQS_OFF    - interrupts were disabled
 197  *  NEED_RESCED - reschedule is requested
 198  *  HARDIRQ     - inside an interrupt handler
 199  *  SOFTIRQ     - inside a softirq handler
 200  */
 201 enum trace_flag_type {
 202         TRACE_FLAG_IRQS_OFF             = 0x01,
 203         TRACE_FLAG_NEED_RESCHED         = 0x02,
 204         TRACE_FLAG_HARDIRQ              = 0x04,
 205         TRACE_FLAG_SOFTIRQ              = 0x08,
 206 };
 207
 208 /*
 209  * TRACE_ITER_SYM_MASK masks the options in trace_flags that
 210  * control the output of kernel symbols.
 211  */
 212 #define TRACE_ITER_SYM_MASK \
 213         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 214
 215 /* These must match the bit postions in trace_iterator_flags */
 216 static const char *trace_options[] = {
 217         "print-parent",
 218         "sym-offset",
 219         "sym-addr",
 220         "verbose",
 221         "raw",
 222         "hex",
 223         "bin",
 224         "block",
 225         "stacktrace",
 226         "sched-tree",
 227         NULL
 228 };
 229
 230 /*
 231  * ftrace_max_lock is used to protect the swapping of buffers
 232  * when taking a max snapshot. The buffers themselves are
 233  * protected by per_cpu spinlocks. But the action of the swap
 234  * needs its own lock.
 235  *
 236  * This is defined as a raw_spinlock_t in order to help
 237  * with performance when lockdep debugging is enabled.
 238  */
 239 static raw_spinlock_t ftrace_max_lock =
 240         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 241
 242 /*
 243  * Copy the new maximum trace into the separate maximum-trace
 244  * structure. (this way the maximum trace is permanently saved,
 245  * for later retrieval via /debugfs/tracing/latency_trace)
 246  */
 247 static void
 248 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 249 {
 250         struct trace_array_cpu *data = tr->data[cpu];
 251
 252         max_tr.cpu = cpu;
 253         max_tr.time_start = data->preempt_timestamp;
 254
 255         data = max_tr.data[cpu];
 256         data->saved_latency = tracing_max_latency;
 257
 258         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 259         data->pid = tsk->pid;
 260         data->uid = tsk->uid;
 261         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 262         data->policy = tsk->policy;
 263         data->rt_priority = tsk->rt_priority;
 264
 265         /* record this tasks comm */
 266         tracing_record_cmdline(current);
 267 }
 268
 269 #define CHECK_COND(cond)                        \
 270         if (unlikely(cond)) {                   \
 271                 tracing_disabled = 1;           \
 272                 WARN_ON(1);                     \
 273                 return -1;                      \
 274         }
 275
 276 /**
 277  * check_pages - integrity check of trace buffers
 278  *
 279  * As a safty measure we check to make sure the data pages have not
 280  * been corrupted.
 281  */
 282 int check_pages(struct trace_array_cpu *data)
 283 {
 284         struct page *page, *tmp;
 285
 286         CHECK_COND(data->trace_pages.next->prev != &data->trace_pages);
 287         CHECK_COND(data->trace_pages.prev->next != &data->trace_pages);
 288
 289         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 290                 CHECK_COND(page->lru.next->prev != &page->lru);
 291                 CHECK_COND(page->lru.prev->next != &page->lru);
 292         }
 293
 294         return 0;
 295 }
 296
 297 /**
 298  * head_page - page address of the first page in per_cpu buffer.
 299  *
 300  * head_page returns the page address of the first page in
 301  * a per_cpu buffer. This also preforms various consistency
 302  * checks to make sure the buffer has not been corrupted.
 303  */
 304 void *head_page(struct trace_array_cpu *data)
 305 {
 306         struct page *page;
 307
 308         if (list_empty(&data->trace_pages))
 309                 return NULL;
 310
 311         page = list_entry(data->trace_pages.next, struct page, lru);
 312         BUG_ON(&page->lru == &data->trace_pages);
 313
 314         return page_address(page);
 315 }
 316
 317 /**
 318  * trace_seq_printf - sequence printing of trace information
 319  * @s: trace sequence descriptor
 320  * @fmt: printf format string
 321  *
 322  * The tracer may use either sequence operations or its own
 323  * copy to user routines. To simplify formating of a trace
 324  * trace_seq_printf is used to store strings into a special
 325  * buffer (@s). Then the output may be either used by
 326  * the sequencer or pulled into another buffer.
 327  */
 328 int
 329 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 330 {
 331         int len = (PAGE_SIZE - 1) - s->len;
 332         va_list ap;
 333         int ret;
 334
 335         if (!len)
 336                 return 0;
 337
 338         va_start(ap, fmt);
 339         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 340         va_end(ap);
 341
 342         /* If we can't write it all, don't bother writing anything */
 343         if (ret >= len)
 344                 return 0;
 345
 346         s->len += ret;
 347
 348         return len;
 349 }
 350
 351 /**
 352  * trace_seq_puts - trace sequence printing of simple string
 353  * @s: trace sequence descriptor
 354  * @str: simple string to record
 355  *
 356  * The tracer may use either the sequence operations or its own
 357  * copy to user routines. This function records a simple string
 358  * into a special buffer (@s) for later retrieval by a sequencer
 359  * or other mechanism.
 360  */
 361 static int
 362 trace_seq_puts(struct trace_seq *s, const char *str)
 363 {
 364         int len = strlen(str);
 365
 366         if (len > ((PAGE_SIZE - 1) - s->len))
 367                 return 0;
 368
 369         memcpy(s->buffer + s->len, str, len);
 370         s->len += len;
 371
 372         return len;
 373 }
 374
 375 static int
 376 trace_seq_putc(struct trace_seq *s, unsigned char c)
 377 {
 378         if (s->len >= (PAGE_SIZE - 1))
 379                 return 0;
 380
 381         s->buffer[s->len++] = c;
 382
 383         return 1;
 384 }
 385
 386 static int
 387 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 388 {
 389         if (len > ((PAGE_SIZE - 1) - s->len))
 390                 return 0;
 391
 392         memcpy(s->buffer + s->len, mem, len);
 393         s->len += len;
 394
 395         return len;
 396 }
 397
 398 #define HEX_CHARS 17
 399 static const char hex2asc[] = "0123456789abcdef";
 400
 401 static int
 402 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 403 {
 404         unsigned char hex[HEX_CHARS];
 405         unsigned char *data = mem;
 406         unsigned char byte;
 407         int i, j;
 408
 409         BUG_ON(len >= HEX_CHARS);
 410
 411 #ifdef __BIG_ENDIAN
 412         for (i = 0, j = 0; i < len; i++) {
 413 #else
 414         for (i = len-1, j = 0; i >= 0; i--) {
 415 #endif
 416                 byte = data[i];
 417
 418                 hex[j++] = hex2asc[byte & 0x0f];
 419                 hex[j++] = hex2asc[byte >> 4];
 420         }
 421         hex[j++] = ' ';
 422
 423         return trace_seq_putmem(s, hex, j);
 424 }
 425
 426 static void
 427 trace_seq_reset(struct trace_seq *s)
 428 {
 429         s->len = 0;
 430         s->readpos = 0;
 431 }
 432
 433 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
 434 {
 435         int len;
 436         int ret;
 437
 438         if (s->len <= s->readpos)
 439                 return -EBUSY;
 440
 441         len = s->len - s->readpos;
 442         if (cnt > len)
 443                 cnt = len;
 444         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
 445         if (ret)
 446                 return -EFAULT;
 447
 448         s->readpos += len;
 449         return cnt;
 450 }
 451
 452 static void
 453 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 454 {
 455         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 456
 457         s->buffer[len] = 0;
 458         seq_puts(m, s->buffer);
 459
 460         trace_seq_reset(s);
 461 }
 462
 463 /*
 464  * flip the trace buffers between two trace descriptors.
 465  * This usually is the buffers between the global_trace and
 466  * the max_tr to record a snapshot of a current trace.
 467  *
 468  * The ftrace_max_lock must be held.
 469  */
 470 static void
 471 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 472 {
 473         struct list_head flip_pages;
 474
 475         INIT_LIST_HEAD(&flip_pages);
 476
 477         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 478                 sizeof(struct trace_array_cpu) -
 479                 offsetof(struct trace_array_cpu, trace_head_idx));
 480
 481         check_pages(tr1);
 482         check_pages(tr2);
 483         list_splice_init(&tr1->trace_pages, &flip_pages);
 484         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 485         list_splice_init(&flip_pages, &tr2->trace_pages);
 486         BUG_ON(!list_empty(&flip_pages));
 487         check_pages(tr1);
 488         check_pages(tr2);
 489 }
 490
 491 /**
 492  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
 493  * @tr: tracer
 494  * @tsk: the task with the latency
 495  * @cpu: The cpu that initiated the trace.
 496  *
 497  * Flip the buffers between the @tr and the max_tr and record information
 498  * about which task was the cause of this latency.
 499  */
 500 void
 501 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 502 {
 503         struct trace_array_cpu *data;
 504         int i;
 505
 506         WARN_ON_ONCE(!irqs_disabled());
 507         __raw_spin_lock(&ftrace_max_lock);
 508         /* clear out all the previous traces */
 509         for_each_tracing_cpu(i) {
 510                 data = tr->data[i];
 511                 flip_trace(max_tr.data[i], data);
 512                 tracing_reset(data);
 513         }
 514
 515         __update_max_tr(tr, tsk, cpu);
 516         __raw_spin_unlock(&ftrace_max_lock);
 517 }
 518
 519 /**
 520  * update_max_tr_single - only copy one trace over, and reset the rest
 521  * @tr - tracer
 522  * @tsk - task with the latency
 523  * @cpu - the cpu of the buffer to copy.
 524  *
 525  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
 526  */
 527 void
 528 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 529 {
 530         struct trace_array_cpu *data = tr->data[cpu];
 531         int i;
 532
 533         WARN_ON_ONCE(!irqs_disabled());
 534         __raw_spin_lock(&ftrace_max_lock);
 535         for_each_tracing_cpu(i)
 536                 tracing_reset(max_tr.data[i]);
 537
 538         flip_trace(max_tr.data[cpu], data);
 539         tracing_reset(data);
 540
 541         __update_max_tr(tr, tsk, cpu);
 542         __raw_spin_unlock(&ftrace_max_lock);
 543 }
 544
 545 /**
 546  * register_tracer - register a tracer with the ftrace system.
 547  * @type - the plugin for the tracer
 548  *
 549  * Register a new plugin tracer.
 550  */
 551 int register_tracer(struct tracer *type)
 552 {
 553         struct tracer *t;
 554         int len;
 555         int ret = 0;
 556
 557         if (!type->name) {
 558                 pr_info("Tracer must have a name\n");
 559                 return -1;
 560         }
 561
 562         mutex_lock(&trace_types_lock);
 563         for (t = trace_types; t; t = t->next) {
 564                 if (strcmp(type->name, t->name) == 0) {
 565                         /* already found */
 566                         pr_info("Trace %s already registered\n",
 567                                 type->name);
 568                         ret = -1;
 569                         goto out;
 570                 }
 571         }
 572
 573 #ifdef CONFIG_FTRACE_STARTUP_TEST
 574         if (type->selftest) {
 575                 struct tracer *saved_tracer = current_trace;
 576                 struct trace_array_cpu *data;
 577                 struct trace_array *tr = &global_trace;
 578                 int saved_ctrl = tr->ctrl;
 579                 int i;
 580                 /*
 581                  * Run a selftest on this tracer.
 582                  * Here we reset the trace buffer, and set the current
 583                  * tracer to be this tracer. The tracer can then run some
 584                  * internal tracing to verify that everything is in order.
 585                  * If we fail, we do not register this tracer.
 586                  */
 587                 for_each_tracing_cpu(i) {
 588                         data = tr->data[i];
 589                         if (!head_page(data))
 590                                 continue;
 591                         tracing_reset(data);
 592                 }
 593                 current_trace = type;
 594                 tr->ctrl = 0;
 595                 /* the test is responsible for initializing and enabling */
 596                 pr_info("Testing tracer %s: ", type->name);
 597                 ret = type->selftest(type, tr);
 598                 /* the test is responsible for resetting too */
 599                 current_trace = saved_tracer;
 600                 tr->ctrl = saved_ctrl;
 601                 if (ret) {
 602                         printk(KERN_CONT "FAILED!\n");
 603                         goto out;
 604                 }
 605                 /* Only reset on passing, to avoid touching corrupted buffers */
 606                 for_each_tracing_cpu(i) {
 607                         data = tr->data[i];
 608                         if (!head_page(data))
 609                                 continue;
 610                         tracing_reset(data);
 611                 }
 612                 printk(KERN_CONT "PASSED\n");
 613         }
 614 #endif
 615
 616         type->next = trace_types;
 617         trace_types = type;
 618         len = strlen(type->name);
 619         if (len > max_tracer_type_len)
 620                 max_tracer_type_len = len;
 621
 622  out:
 623         mutex_unlock(&trace_types_lock);
 624
 625         return ret;
 626 }
 627
 628 void unregister_tracer(struct tracer *type)
 629 {
 630         struct tracer **t;
 631         int len;
 632
 633         mutex_lock(&trace_types_lock);
 634         for (t = &trace_types; *t; t = &(*t)->next) {
 635                 if (*t == type)
 636                         goto found;
 637         }
 638         pr_info("Trace %s not registered\n", type->name);
 639         goto out;
 640
 641  found:
 642         *t = (*t)->next;
 643         if (strlen(type->name) != max_tracer_type_len)
 644                 goto out;
 645
 646         max_tracer_type_len = 0;
 647         for (t = &trace_types; *t; t = &(*t)->next) {
 648                 len = strlen((*t)->name);
 649                 if (len > max_tracer_type_len)
 650                         max_tracer_type_len = len;
 651         }
 652  out:
 653         mutex_unlock(&trace_types_lock);
 654 }
 655
 656 void tracing_reset(struct trace_array_cpu *data)
 657 {
 658         data->trace_idx = 0;
 659         data->overrun = 0;
 660         data->trace_head = data->trace_tail = head_page(data);
 661         data->trace_head_idx = 0;
 662         data->trace_tail_idx = 0;
 663 }
 664
 665 #define SAVED_CMDLINES 128
 666 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 667 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 668 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 669 static int cmdline_idx;
 670 static DEFINE_SPINLOCK(trace_cmdline_lock);
 671
 672 /* temporary disable recording */
 673 atomic_t trace_record_cmdline_disabled __read_mostly;
 674
 675 static void trace_init_cmdlines(void)
 676 {
 677         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 678         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 679         cmdline_idx = 0;
 680 }
 681
 682 void trace_stop_cmdline_recording(void);
 683
 684 static void trace_save_cmdline(struct task_struct *tsk)
 685 {
 686         unsigned map;
 687         unsigned idx;
 688
 689         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 690                 return;
 691
 692         /*
 693          * It's not the end of the world if we don't get
 694          * the lock, but we also don't want to spin
 695          * nor do we want to disable interrupts,
 696          * so if we miss here, then better luck next time.
 697          */
 698         if (!spin_trylock(&trace_cmdline_lock))
 699                 return;
 700
 701         idx = map_pid_to_cmdline[tsk->pid];
 702         if (idx >= SAVED_CMDLINES) {
 703                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 704
 705                 map = map_cmdline_to_pid[idx];
 706                 if (map <= PID_MAX_DEFAULT)
 707                         map_pid_to_cmdline[map] = (unsigned)-1;
 708
 709                 map_pid_to_cmdline[tsk->pid] = idx;
 710
 711                 cmdline_idx = idx;
 712         }
 713
 714         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 715
 716         spin_unlock(&trace_cmdline_lock);
 717 }
 718
 719 static char *trace_find_cmdline(int pid)
 720 {
 721         char *cmdline = "<...>";
 722         unsigned map;
 723
 724         if (!pid)
 725                 return "<idle>";
 726
 727         if (pid > PID_MAX_DEFAULT)
 728                 goto out;
 729
 730         map = map_pid_to_cmdline[pid];
 731         if (map >= SAVED_CMDLINES)
 732                 goto out;
 733
 734         cmdline = saved_cmdlines[map];
 735
 736  out:
 737         return cmdline;
 738 }
 739
 740 void tracing_record_cmdline(struct task_struct *tsk)
 741 {
 742         if (atomic_read(&trace_record_cmdline_disabled))
 743                 return;
 744
 745         trace_save_cmdline(tsk);
 746 }
 747
 748 static inline struct list_head *
 749 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 750 {
 751         /*
 752          * Roundrobin - but skip the head (which is not a real page):
 753          */
 754         next = next->next;
 755         if (unlikely(next == &data->trace_pages))
 756                 next = next->next;
 757         BUG_ON(next == &data->trace_pages);
 758
 759         return next;
 760 }
 761
 762 static inline void *
 763 trace_next_page(struct trace_array_cpu *data, void *addr)
 764 {
 765         struct list_head *next;
 766         struct page *page;
 767
 768         page = virt_to_page(addr);
 769
 770         next = trace_next_list(data, &page->lru);
 771         page = list_entry(next, struct page, lru);
 772
 773         return page_address(page);
 774 }
 775
 776 static inline struct trace_entry *
 777 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 778 {
 779         unsigned long idx, idx_next;
 780         struct trace_entry *entry;
 781
 782         data->trace_idx++;
 783         idx = data->trace_head_idx;
 784         idx_next = idx + 1;
 785
 786         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 787
 788         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 789
 790         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 791                 data->trace_head = trace_next_page(data, data->trace_head);
 792                 idx_next = 0;
 793         }
 794
 795         if (data->trace_head == data->trace_tail &&
 796             idx_next == data->trace_tail_idx) {
 797                 /* overrun */
 798                 data->overrun++;
 799                 data->trace_tail_idx++;
 800                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 801                         data->trace_tail =
 802                                 trace_next_page(data, data->trace_tail);
 803                         data->trace_tail_idx = 0;
 804                 }
 805         }
 806
 807         data->trace_head_idx = idx_next;
 808
 809         return entry;
 810 }
 811
 812 static inline void
 813 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 814 {
 815         struct task_struct *tsk = current;
 816         unsigned long pc;
 817
 818         pc = preempt_count();
 819
 820         entry->preempt_count    = pc & 0xff;
 821         entry->pid              = (tsk) ? tsk->pid : 0;
 822         entry->t                = ftrace_now(raw_smp_processor_id());
 823         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 824                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 825                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 826                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 827 }
 828
 829 void
 830 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 831                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 832 {
 833         struct trace_entry *entry;
 834         unsigned long irq_flags;
 835
 836         raw_local_irq_save(irq_flags);
 837         __raw_spin_lock(&data->lock);
 838         entry                   = tracing_get_trace_entry(tr, data);
 839         tracing_generic_entry_update(entry, flags);
 840         entry->type             = TRACE_FN;
 841         entry->fn.ip            = ip;
 842         entry->fn.parent_ip     = parent_ip;
 843         __raw_spin_unlock(&data->lock);
 844         raw_local_irq_restore(irq_flags);
 845 }
 846
 847 void
 848 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 849        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 850 {
 851         if (likely(!atomic_read(&data->disabled)))
 852                 trace_function(tr, data, ip, parent_ip, flags);
 853 }
 854
 855 #ifdef CONFIG_MMIOTRACE
 856 void __trace_mmiotrace_rw(struct trace_array *tr, struct trace_array_cpu *data,
 857                                                 struct mmiotrace_rw *rw)
 858 {
 859         struct trace_entry *entry;
 860         unsigned long irq_flags;
 861
 862         raw_local_irq_save(irq_flags);
 863         __raw_spin_lock(&data->lock);
 864
 865         entry                   = tracing_get_trace_entry(tr, data);
 866         tracing_generic_entry_update(entry, 0);
 867         entry->type             = TRACE_MMIO_RW;
 868         entry->mmiorw           = *rw;
 869
 870         __raw_spin_unlock(&data->lock);
 871         raw_local_irq_restore(irq_flags);
 872
 873         trace_wake_up();
 874 }
 875
 876 void __trace_mmiotrace_map(struct trace_array *tr, struct trace_array_cpu *data,
 877                                                 struct mmiotrace_map *map)
 878 {
 879         struct trace_entry *entry;
 880         unsigned long irq_flags;
 881
 882         raw_local_irq_save(irq_flags);
 883         __raw_spin_lock(&data->lock);
 884
 885         entry                   = tracing_get_trace_entry(tr, data);
 886         tracing_generic_entry_update(entry, 0);
 887         entry->type             = TRACE_MMIO_MAP;
 888         entry->mmiomap          = *map;
 889
 890         __raw_spin_unlock(&data->lock);
 891         raw_local_irq_restore(irq_flags);
 892
 893         trace_wake_up();
 894 }
 895 #endif
 896
 897 void __trace_stack(struct trace_array *tr,
 898                    struct trace_array_cpu *data,
 899                    unsigned long flags,
 900                    int skip)
 901 {
 902         struct trace_entry *entry;
 903         struct stack_trace trace;
 904
 905         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 906                 return;
 907
 908         entry                   = tracing_get_trace_entry(tr, data);
 909         tracing_generic_entry_update(entry, flags);
 910         entry->type             = TRACE_STACK;
 911
 912         memset(&entry->stack, 0, sizeof(entry->stack));
 913
 914         trace.nr_entries        = 0;
 915         trace.max_entries       = FTRACE_STACK_ENTRIES;
 916         trace.skip              = skip;
 917         trace.entries           = entry->stack.caller;
 918
 919         save_stack_trace(&trace);
 920 }
 921
 922 void
 923 __trace_special(void *__tr, void *__data,
 924                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 925 {
 926         struct trace_array_cpu *data = __data;
 927         struct trace_array *tr = __tr;
 928         struct trace_entry *entry;
 929         unsigned long irq_flags;
 930
 931         raw_local_irq_save(irq_flags);
 932         __raw_spin_lock(&data->lock);
 933         entry                   = tracing_get_trace_entry(tr, data);
 934         tracing_generic_entry_update(entry, 0);
 935         entry->type             = TRACE_SPECIAL;
 936         entry->special.arg1     = arg1;
 937         entry->special.arg2     = arg2;
 938         entry->special.arg3     = arg3;
 939         __trace_stack(tr, data, irq_flags, 4);
 940         __raw_spin_unlock(&data->lock);
 941         raw_local_irq_restore(irq_flags);
 942
 943         trace_wake_up();
 944 }
 945
 946 void
 947 tracing_sched_switch_trace(struct trace_array *tr,
 948                            struct trace_array_cpu *data,
 949                            struct task_struct *prev,
 950                            struct task_struct *next,
 951                            unsigned long flags)
 952 {
 953         struct trace_entry *entry;
 954         unsigned long irq_flags;
 955
 956         raw_local_irq_save(irq_flags);
 957         __raw_spin_lock(&data->lock);
 958         entry                   = tracing_get_trace_entry(tr, data);
 959         tracing_generic_entry_update(entry, flags);
 960         entry->type             = TRACE_CTX;
 961         entry->ctx.prev_pid     = prev->pid;
 962         entry->ctx.prev_prio    = prev->prio;
 963         entry->ctx.prev_state   = prev->state;
 964         entry->ctx.next_pid     = next->pid;
 965         entry->ctx.next_prio    = next->prio;
 966         entry->ctx.next_state   = next->state;
 967         __trace_stack(tr, data, flags, 5);
 968         __raw_spin_unlock(&data->lock);
 969         raw_local_irq_restore(irq_flags);
 970 }
 971
 972 void
 973 tracing_sched_wakeup_trace(struct trace_array *tr,
 974                            struct trace_array_cpu *data,
 975                            struct task_struct *wakee,
 976                            struct task_struct *curr,
 977                            unsigned long flags)
 978 {
 979         struct trace_entry *entry;
 980         unsigned long irq_flags;
 981
 982         raw_local_irq_save(irq_flags);
 983         __raw_spin_lock(&data->lock);
 984         entry                   = tracing_get_trace_entry(tr, data);
 985         tracing_generic_entry_update(entry, flags);
 986         entry->type             = TRACE_WAKE;
 987         entry->ctx.prev_pid     = curr->pid;
 988         entry->ctx.prev_prio    = curr->prio;
 989         entry->ctx.prev_state   = curr->state;
 990         entry->ctx.next_pid     = wakee->pid;
 991         entry->ctx.next_prio    = wakee->prio;
 992         entry->ctx.next_state   = wakee->state;
 993         __trace_stack(tr, data, flags, 6);
 994         __raw_spin_unlock(&data->lock);
 995         raw_local_irq_restore(irq_flags);
 996
 997         trace_wake_up();
 998 }
 999
1000 void
1001 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1002 {
1003         struct trace_array *tr = &global_trace;
1004         struct trace_array_cpu *data;
1005         unsigned long flags;
1006         long disabled;
1007         int cpu;
1008
1009         if (tracing_disabled || current_trace == &no_tracer || !tr->ctrl)
1010                 return;
1011
1012         local_irq_save(flags);
1013         cpu = raw_smp_processor_id();
1014         data = tr->data[cpu];
1015         disabled = atomic_inc_return(&data->disabled);
1016
1017         if (likely(disabled == 1))
1018                 __trace_special(tr, data, arg1, arg2, arg3);
1019
1020         atomic_dec(&data->disabled);
1021         local_irq_restore(flags);
1022 }
1023
1024 #ifdef CONFIG_FTRACE
1025 static void
1026 function_trace_call(unsigned long ip, unsigned long parent_ip)
1027 {
1028         struct trace_array *tr = &global_trace;
1029         struct trace_array_cpu *data;
1030         unsigned long flags;
1031         long disabled;
1032         int cpu;
1033
1034         if (unlikely(!ftrace_function_enabled))
1035                 return;
1036
1037         if (skip_trace(ip))
1038                 return;
1039
1040         local_irq_save(flags);
1041         cpu = raw_smp_processor_id();
1042         data = tr->data[cpu];
1043         disabled = atomic_inc_return(&data->disabled);
1044
1045         if (likely(disabled == 1))
1046                 trace_function(tr, data, ip, parent_ip, flags);
1047
1048         atomic_dec(&data->disabled);
1049         local_irq_restore(flags);
1050 }
1051
1052 static struct ftrace_ops trace_ops __read_mostly =
1053 {
1054         .func = function_trace_call,
1055 };
1056
1057 void tracing_start_function_trace(void)
1058 {
1059         ftrace_function_enabled = 0;
1060         register_ftrace_function(&trace_ops);
1061         if (tracer_enabled)
1062                 ftrace_function_enabled = 1;
1063 }
1064
1065 void tracing_stop_function_trace(void)
1066 {
1067         ftrace_function_enabled = 0;
1068         unregister_ftrace_function(&trace_ops);
1069 }
1070 #endif
1071
1072 enum trace_file_type {
1073         TRACE_FILE_LAT_FMT      = 1,
1074 };
1075
1076 static struct trace_entry *
1077 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
1078                 struct trace_iterator *iter, int cpu)
1079 {
1080         struct page *page;
1081         struct trace_entry *array;
1082
1083         if (iter->next_idx[cpu] >= tr->entries ||
1084             iter->next_idx[cpu] >= data->trace_idx ||
1085             (data->trace_head == data->trace_tail &&
1086              data->trace_head_idx == data->trace_tail_idx))
1087                 return NULL;
1088
1089         if (!iter->next_page[cpu]) {
1090                 /* Initialize the iterator for this cpu trace buffer */
1091                 WARN_ON(!data->trace_tail);
1092                 page = virt_to_page(data->trace_tail);
1093                 iter->next_page[cpu] = &page->lru;
1094                 iter->next_page_idx[cpu] = data->trace_tail_idx;
1095         }
1096
1097         page = list_entry(iter->next_page[cpu], struct page, lru);
1098         BUG_ON(&data->trace_pages == &page->lru);
1099
1100         array = page_address(page);
1101
1102         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1103         return &array[iter->next_page_idx[cpu]];
1104 }
1105
1106 static struct trace_entry *
1107 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1108 {
1109         struct trace_array *tr = iter->tr;
1110         struct trace_entry *ent, *next = NULL;
1111         int next_cpu = -1;
1112         int cpu;
1113
1114         for_each_tracing_cpu(cpu) {
1115                 if (!head_page(tr->data[cpu]))
1116                         continue;
1117                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1118                 /*
1119                  * Pick the entry with the smallest timestamp:
1120                  */
1121                 if (ent && (!next || ent->t < next->t)) {
1122                         next = ent;
1123                         next_cpu = cpu;
1124                 }
1125         }
1126
1127         if (ent_cpu)
1128                 *ent_cpu = next_cpu;
1129
1130         return next;
1131 }
1132
1133 static void trace_iterator_increment(struct trace_iterator *iter)
1134 {
1135         iter->idx++;
1136         iter->next_idx[iter->cpu]++;
1137         iter->next_page_idx[iter->cpu]++;
1138
1139         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1140                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1141
1142                 iter->next_page_idx[iter->cpu] = 0;
1143                 iter->next_page[iter->cpu] =
1144                         trace_next_list(data, iter->next_page[iter->cpu]);
1145         }
1146 }
1147
1148 static void trace_consume(struct trace_iterator *iter)
1149 {
1150         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1151
1152         data->trace_tail_idx++;
1153         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1154                 data->trace_tail = trace_next_page(data, data->trace_tail);
1155                 data->trace_tail_idx = 0;
1156         }
1157
1158         /* Check if we empty it, then reset the index */
1159         if (data->trace_head == data->trace_tail &&
1160             data->trace_head_idx == data->trace_tail_idx)
1161                 data->trace_idx = 0;
1162 }
1163
1164 static void *find_next_entry_inc(struct trace_iterator *iter)
1165 {
1166         struct trace_entry *next;
1167         int next_cpu = -1;
1168
1169         next = find_next_entry(iter, &next_cpu);
1170
1171         iter->prev_ent = iter->ent;
1172         iter->prev_cpu = iter->cpu;
1173
1174         iter->ent = next;
1175         iter->cpu = next_cpu;
1176
1177         if (next)
1178                 trace_iterator_increment(iter);
1179
1180         return next ? iter : NULL;
1181 }
1182
1183 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1184 {
1185         struct trace_iterator *iter = m->private;
1186         void *last_ent = iter->ent;
1187         int i = (int)*pos;
1188         void *ent;
1189
1190         (*pos)++;
1191
1192         /* can't go backwards */
1193         if (iter->idx > i)
1194                 return NULL;
1195
1196         if (iter->idx < 0)
1197                 ent = find_next_entry_inc(iter);
1198         else
1199                 ent = iter;
1200
1201         while (ent && iter->idx < i)
1202                 ent = find_next_entry_inc(iter);
1203
1204         iter->pos = *pos;
1205
1206         return ent;
1207 }
1208
1209 static void *s_start(struct seq_file *m, loff_t *pos)
1210 {
1211         struct trace_iterator *iter = m->private;
1212         void *p = NULL;
1213         loff_t l = 0;
1214         int i;
1215
1216         mutex_lock(&trace_types_lock);
1217
1218         if (!current_trace || current_trace != iter->trace) {
1219                 mutex_unlock(&trace_types_lock);
1220                 return NULL;
1221         }
1222
1223         atomic_inc(&trace_record_cmdline_disabled);
1224
1225         /* let the tracer grab locks here if needed */
1226         if (current_trace->start)
1227                 current_trace->start(iter);
1228
1229         if (*pos != iter->pos) {
1230                 iter->ent = NULL;
1231                 iter->cpu = 0;
1232                 iter->idx = -1;
1233                 iter->prev_ent = NULL;
1234                 iter->prev_cpu = -1;
1235
1236                 for_each_tracing_cpu(i) {
1237                         iter->next_idx[i] = 0;
1238                         iter->next_page[i] = NULL;
1239                 }
1240
1241                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1242                         ;
1243
1244         } else {
1245                 l = *pos - 1;
1246                 p = s_next(m, p, &l);
1247         }
1248
1249         return p;
1250 }
1251
1252 static void s_stop(struct seq_file *m, void *p)
1253 {
1254         struct trace_iterator *iter = m->private;
1255
1256         atomic_dec(&trace_record_cmdline_disabled);
1257
1258         /* let the tracer release locks here if needed */
1259         if (current_trace && current_trace == iter->trace && iter->trace->stop)
1260                 iter->trace->stop(iter);
1261
1262         mutex_unlock(&trace_types_lock);
1263 }
1264
1265 #define KRETPROBE_MSG "[unknown/kretprobe'd]"
1266
1267 #ifdef CONFIG_KRETPROBES
1268 static inline int kretprobed(unsigned long addr)
1269 {
1270         return addr == (unsigned long)kretprobe_trampoline;
1271 }
1272 #else
1273 static inline int kretprobed(unsigned long addr)
1274 {
1275         return 0;
1276 }
1277 #endif /* CONFIG_KRETPROBES */
1278
1279 static int
1280 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1281 {
1282 #ifdef CONFIG_KALLSYMS
1283         char str[KSYM_SYMBOL_LEN];
1284
1285         kallsyms_lookup(address, NULL, NULL, NULL, str);
1286
1287         return trace_seq_printf(s, fmt, str);
1288 #endif
1289         return 1;
1290 }
1291
1292 static int
1293 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1294                      unsigned long address)
1295 {
1296 #ifdef CONFIG_KALLSYMS
1297         char str[KSYM_SYMBOL_LEN];
1298
1299         sprint_symbol(str, address);
1300         return trace_seq_printf(s, fmt, str);
1301 #endif
1302         return 1;
1303 }
1304
1305 #ifndef CONFIG_64BIT
1306 # define IP_FMT "%08lx"
1307 #else
1308 # define IP_FMT "%016lx"
1309 #endif
1310
1311 static int
1312 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1313 {
1314         int ret;
1315
1316         if (!ip)
1317                 return trace_seq_printf(s, "0");
1318
1319         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1320                 ret = seq_print_sym_offset(s, "%s", ip);
1321         else
1322                 ret = seq_print_sym_short(s, "%s", ip);
1323
1324         if (!ret)
1325                 return 0;
1326
1327         if (sym_flags & TRACE_ITER_SYM_ADDR)
1328                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1329         return ret;
1330 }
1331
1332 static void print_lat_help_header(struct seq_file *m)
1333 {
1334         seq_puts(m, "#                _------=> CPU#            \n");
1335         seq_puts(m, "#               / _-----=> irqs-off        \n");
1336         seq_puts(m, "#              | / _----=> need-resched    \n");
1337         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1338         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1339         seq_puts(m, "#              |||| /                      \n");
1340         seq_puts(m, "#              |||||     delay             \n");
1341         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1342         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1343 }
1344
1345 static void print_func_help_header(struct seq_file *m)
1346 {
1347         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1348         seq_puts(m, "#              | |      |          |         |\n");
1349 }
1350
1351
1352 static void
1353 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1354 {
1355         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1356         struct trace_array *tr = iter->tr;
1357         struct trace_array_cpu *data = tr->data[tr->cpu];
1358         struct tracer *type = current_trace;
1359         unsigned long total   = 0;
1360         unsigned long entries = 0;
1361         int cpu;
1362         const char *name = "preemption";
1363
1364         if (type)
1365                 name = type->name;
1366
1367         for_each_tracing_cpu(cpu) {
1368                 if (head_page(tr->data[cpu])) {
1369                         total += tr->data[cpu]->trace_idx;
1370                         if (tr->data[cpu]->trace_idx > tr->entries)
1371                                 entries += tr->entries;
1372                         else
1373                                 entries += tr->data[cpu]->trace_idx;
1374                 }
1375         }
1376
1377         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1378                    name, UTS_RELEASE);
1379         seq_puts(m, "-----------------------------------"
1380                  "---------------------------------\n");
1381         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1382                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1383                    nsecs_to_usecs(data->saved_latency),
1384                    entries,
1385                    total,
1386                    tr->cpu,
1387 #if defined(CONFIG_PREEMPT_NONE)
1388                    "server",
1389 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1390                    "desktop",
1391 #elif defined(CONFIG_PREEMPT)
1392                    "preempt",
1393 #else
1394                    "unknown",
1395 #endif
1396                    /* These are reserved for later use */
1397                    0, 0, 0, 0);
1398 #ifdef CONFIG_SMP
1399         seq_printf(m, " #P:%d)\n", num_online_cpus());
1400 #else
1401         seq_puts(m, ")\n");
1402 #endif
1403         seq_puts(m, "    -----------------\n");
1404         seq_printf(m, "    | task: %.16s-%d "
1405                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1406                    data->comm, data->pid, data->uid, data->nice,
1407                    data->policy, data->rt_priority);
1408         seq_puts(m, "    -----------------\n");
1409
1410         if (data->critical_start) {
1411                 seq_puts(m, " => started at: ");
1412                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1413                 trace_print_seq(m, &iter->seq);
1414                 seq_puts(m, "\n => ended at:   ");
1415                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1416                 trace_print_seq(m, &iter->seq);
1417                 seq_puts(m, "\n");
1418         }
1419
1420         seq_puts(m, "\n");
1421 }
1422
1423 static void
1424 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1425 {
1426         int hardirq, softirq;
1427         char *comm;
1428
1429         comm = trace_find_cmdline(entry->pid);
1430
1431         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1432         trace_seq_printf(s, "%d", cpu);
1433         trace_seq_printf(s, "%c%c",
1434                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1435                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1436
1437         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1438         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1439         if (hardirq && softirq) {
1440                 trace_seq_putc(s, 'H');
1441         } else {
1442                 if (hardirq) {
1443                         trace_seq_putc(s, 'h');
1444                 } else {
1445                         if (softirq)
1446                                 trace_seq_putc(s, 's');
1447                         else
1448                                 trace_seq_putc(s, '.');
1449                 }
1450         }
1451
1452         if (entry->preempt_count)
1453                 trace_seq_printf(s, "%x", entry->preempt_count);
1454         else
1455                 trace_seq_puts(s, ".");
1456 }
1457
1458 unsigned long preempt_mark_thresh = 100;
1459
1460 static void
1461 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1462                     unsigned long rel_usecs)
1463 {
1464         trace_seq_printf(s, " %4lldus", abs_usecs);
1465         if (rel_usecs > preempt_mark_thresh)
1466                 trace_seq_puts(s, "!: ");
1467         else if (rel_usecs > 1)
1468                 trace_seq_puts(s, "+: ");
1469         else
1470                 trace_seq_puts(s, " : ");
1471 }
1472
1473 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1474
1475 static int
1476 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1477 {
1478         struct trace_seq *s = &iter->seq;
1479         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1480         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1481         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1482         struct trace_entry *entry = iter->ent;
1483         unsigned long abs_usecs;
1484         unsigned long rel_usecs;
1485         char *comm;
1486         int S, T;
1487         int i;
1488         unsigned state;
1489
1490         if (!next_entry)
1491                 next_entry = entry;
1492         rel_usecs = ns2usecs(next_entry->t - entry->t);
1493         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1494
1495         if (verbose) {
1496                 comm = trace_find_cmdline(entry->pid);
1497                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1498                                  " %ld.%03ldms (+%ld.%03ldms): ",
1499                                  comm,
1500                                  entry->pid, cpu, entry->flags,
1501                                  entry->preempt_count, trace_idx,
1502                                  ns2usecs(entry->t),
1503                                  abs_usecs/1000,
1504                                  abs_usecs % 1000, rel_usecs/1000,
1505                                  rel_usecs % 1000);
1506         } else {
1507                 lat_print_generic(s, entry, cpu);
1508                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1509         }
1510         switch (entry->type) {
1511         case TRACE_FN:
1512                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1513                 trace_seq_puts(s, " (");
1514                 if (kretprobed(entry->fn.parent_ip))
1515                         trace_seq_puts(s, KRETPROBE_MSG);
1516                 else
1517                         seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1518                 trace_seq_puts(s, ")\n");
1519                 break;
1520         case TRACE_CTX:
1521         case TRACE_WAKE:
1522                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1523                         state_to_char[entry->ctx.next_state] : 'X';
1524
1525                 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1526                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1527                 comm = trace_find_cmdline(entry->ctx.next_pid);
1528                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1529                                  entry->ctx.prev_pid,
1530                                  entry->ctx.prev_prio,
1531                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1532                                  entry->ctx.next_pid,
1533                                  entry->ctx.next_prio,
1534                                  T, comm);
1535                 break;
1536         case TRACE_SPECIAL:
1537                 trace_seq_printf(s, "# %ld %ld %ld\n",
1538                                  entry->special.arg1,
1539                                  entry->special.arg2,
1540                                  entry->special.arg3);
1541                 break;
1542         case TRACE_STACK:
1543                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1544                         if (i)
1545                                 trace_seq_puts(s, " <= ");
1546                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1547                 }
1548                 trace_seq_puts(s, "\n");
1549                 break;
1550         default:
1551                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1552         }
1553         return 1;
1554 }
1555
1556 static int print_trace_fmt(struct trace_iterator *iter)
1557 {
1558         struct trace_seq *s = &iter->seq;
1559         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1560         struct trace_entry *entry;
1561         unsigned long usec_rem;
1562         unsigned long long t;
1563         unsigned long secs;
1564         char *comm;
1565         int ret;
1566         int S, T;
1567         int i;
1568
1569         entry = iter->ent;
1570
1571         comm = trace_find_cmdline(iter->ent->pid);
1572
1573         t = ns2usecs(entry->t);
1574         usec_rem = do_div(t, 1000000ULL);
1575         secs = (unsigned long)t;
1576
1577         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1578         if (!ret)
1579                 return 0;
1580         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1581         if (!ret)
1582                 return 0;
1583         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1584         if (!ret)
1585                 return 0;
1586
1587         switch (entry->type) {
1588         case TRACE_FN:
1589                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1590                 if (!ret)
1591                         return 0;
1592                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1593                                                 entry->fn.parent_ip) {
1594                         ret = trace_seq_printf(s, " <-");
1595                         if (!ret)
1596                                 return 0;
1597                         if (kretprobed(entry->fn.parent_ip))
1598                                 ret = trace_seq_puts(s, KRETPROBE_MSG);
1599                         else
1600                                 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1601                                                        sym_flags);
1602                         if (!ret)
1603                                 return 0;
1604                 }
1605                 ret = trace_seq_printf(s, "\n");
1606                 if (!ret)
1607                         return 0;
1608                 break;
1609         case TRACE_CTX:
1610         case TRACE_WAKE:
1611                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1612                         state_to_char[entry->ctx.prev_state] : 'X';
1613                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1614                         state_to_char[entry->ctx.next_state] : 'X';
1615                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1616                                        entry->ctx.prev_pid,
1617                                        entry->ctx.prev_prio,
1618                                        S,
1619                                        entry->type == TRACE_CTX ? "==>" : "  +",
1620                                        entry->ctx.next_pid,
1621                                        entry->ctx.next_prio,
1622                                        T);
1623                 if (!ret)
1624                         return 0;
1625                 break;
1626         case TRACE_SPECIAL:
1627                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1628                                  entry->special.arg1,
1629                                  entry->special.arg2,
1630                                  entry->special.arg3);
1631                 if (!ret)
1632                         return 0;
1633                 break;
1634         case TRACE_STACK:
1635                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1636                         if (i) {
1637                                 ret = trace_seq_puts(s, " <= ");
1638                                 if (!ret)
1639                                         return 0;
1640                         }
1641                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1642                                                sym_flags);
1643                         if (!ret)
1644                                 return 0;
1645                 }
1646                 ret = trace_seq_puts(s, "\n");
1647                 if (!ret)
1648                         return 0;
1649                 break;
1650         }
1651         return 1;
1652 }
1653
1654 static int print_raw_fmt(struct trace_iterator *iter)
1655 {
1656         struct trace_seq *s = &iter->seq;
1657         struct trace_entry *entry;
1658         int ret;
1659         int S, T;
1660
1661         entry = iter->ent;
1662
1663         ret = trace_seq_printf(s, "%d %d %llu ",
1664                 entry->pid, iter->cpu, entry->t);
1665         if (!ret)
1666                 return 0;
1667
1668         switch (entry->type) {
1669         case TRACE_FN:
1670                 ret = trace_seq_printf(s, "%x %x\n",
1671                                         entry->fn.ip, entry->fn.parent_ip);
1672                 if (!ret)
1673                         return 0;
1674                 break;
1675         case TRACE_CTX:
1676         case TRACE_WAKE:
1677                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1678                         state_to_char[entry->ctx.prev_state] : 'X';
1679                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1680                         state_to_char[entry->ctx.next_state] : 'X';
1681                 if (entry->type == TRACE_WAKE)
1682                         S = '+';
1683                 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1684                                        entry->ctx.prev_pid,
1685                                        entry->ctx.prev_prio,
1686                                        S,
1687                                        entry->ctx.next_pid,
1688                                        entry->ctx.next_prio,
1689                                        T);
1690                 if (!ret)
1691                         return 0;
1692                 break;
1693         case TRACE_SPECIAL:
1694         case TRACE_STACK:
1695                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1696                                  entry->special.arg1,
1697                                  entry->special.arg2,
1698                                  entry->special.arg3);
1699                 if (!ret)
1700                         return 0;
1701                 break;
1702         }
1703         return 1;
1704 }
1705
1706 #define SEQ_PUT_FIELD_RET(s, x)                         \
1707 do {                                                    \
1708         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1709                 return 0;                               \
1710 } while (0)
1711
1712 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1713 do {                                                    \
1714         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1715                 return 0;                               \
1716 } while (0)
1717
1718 static int print_hex_fmt(struct trace_iterator *iter)
1719 {
1720         struct trace_seq *s = &iter->seq;
1721         unsigned char newline = '\n';
1722         struct trace_entry *entry;
1723         int S, T;
1724
1725         entry = iter->ent;
1726
1727         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1728         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1729         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1730
1731         switch (entry->type) {
1732         case TRACE_FN:
1733                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1734                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1735                 break;
1736         case TRACE_CTX:
1737         case TRACE_WAKE:
1738                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1739                         state_to_char[entry->ctx.prev_state] : 'X';
1740                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1741                         state_to_char[entry->ctx.next_state] : 'X';
1742                 if (entry->type == TRACE_WAKE)
1743                         S = '+';
1744                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1745                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1746                 SEQ_PUT_HEX_FIELD_RET(s, S);
1747                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1748                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1749                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1750                 SEQ_PUT_HEX_FIELD_RET(s, T);
1751                 break;
1752         case TRACE_SPECIAL:
1753         case TRACE_STACK:
1754                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1755                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1756                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1757                 break;
1758         }
1759         SEQ_PUT_FIELD_RET(s, newline);
1760
1761         return 1;
1762 }
1763
1764 static int print_bin_fmt(struct trace_iterator *iter)
1765 {
1766         struct trace_seq *s = &iter->seq;
1767         struct trace_entry *entry;
1768
1769         entry = iter->ent;
1770
1771         SEQ_PUT_FIELD_RET(s, entry->pid);
1772         SEQ_PUT_FIELD_RET(s, entry->cpu);
1773         SEQ_PUT_FIELD_RET(s, entry->t);
1774
1775         switch (entry->type) {
1776         case TRACE_FN:
1777                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1778                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1779                 break;
1780         case TRACE_CTX:
1781                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1782                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1783                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1784                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1785                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1786                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1787                 break;
1788         case TRACE_SPECIAL:
1789         case TRACE_STACK:
1790                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1791                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1792                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1793                 break;
1794         }
1795         return 1;
1796 }
1797
1798 static int trace_empty(struct trace_iterator *iter)
1799 {
1800         struct trace_array_cpu *data;
1801         int cpu;
1802
1803         for_each_tracing_cpu(cpu) {
1804                 data = iter->tr->data[cpu];
1805
1806                 if (head_page(data) && data->trace_idx &&
1807                     (data->trace_tail != data->trace_head ||
1808                      data->trace_tail_idx != data->trace_head_idx))
1809                         return 0;
1810         }
1811         return 1;
1812 }
1813
1814 static int print_trace_line(struct trace_iterator *iter)
1815 {
1816         if (iter->trace && iter->trace->print_line)
1817                 return iter->trace->print_line(iter);
1818
1819         if (trace_flags & TRACE_ITER_BIN)
1820                 return print_bin_fmt(iter);
1821
1822         if (trace_flags & TRACE_ITER_HEX)
1823                 return print_hex_fmt(iter);
1824
1825         if (trace_flags & TRACE_ITER_RAW)
1826                 return print_raw_fmt(iter);
1827
1828         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1829                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1830
1831         return print_trace_fmt(iter);
1832 }
1833
1834 static int s_show(struct seq_file *m, void *v)
1835 {
1836         struct trace_iterator *iter = v;
1837
1838         if (iter->ent == NULL) {
1839                 if (iter->tr) {
1840                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1841                         seq_puts(m, "#\n");
1842                 }
1843                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1844                         /* print nothing if the buffers are empty */
1845                         if (trace_empty(iter))
1846                                 return 0;
1847                         print_trace_header(m, iter);
1848                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1849                                 print_lat_help_header(m);
1850                 } else {
1851                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1852                                 print_func_help_header(m);
1853                 }
1854         } else {
1855                 print_trace_line(iter);
1856                 trace_print_seq(m, &iter->seq);
1857         }
1858
1859         return 0;
1860 }
1861
1862 static struct seq_operations tracer_seq_ops = {
1863         .start          = s_start,
1864         .next           = s_next,
1865         .stop           = s_stop,
1866         .show           = s_show,
1867 };
1868
1869 static struct trace_iterator *
1870 __tracing_open(struct inode *inode, struct file *file, int *ret)
1871 {
1872         struct trace_iterator *iter;
1873
1874         if (tracing_disabled) {
1875                 *ret = -ENODEV;
1876                 return NULL;
1877         }
1878
1879         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1880         if (!iter) {
1881                 *ret = -ENOMEM;
1882                 goto out;
1883         }
1884
1885         mutex_lock(&trace_types_lock);
1886         if (current_trace && current_trace->print_max)
1887                 iter->tr = &max_tr;
1888         else
1889                 iter->tr = inode->i_private;
1890         iter->trace = current_trace;
1891         iter->pos = -1;
1892
1893         /* TODO stop tracer */
1894         *ret = seq_open(file, &tracer_seq_ops);
1895         if (!*ret) {
1896                 struct seq_file *m = file->private_data;
1897                 m->private = iter;
1898
1899                 /* stop the trace while dumping */
1900                 if (iter->tr->ctrl) {
1901                         tracer_enabled = 0;
1902                         ftrace_function_enabled = 0;
1903                 }
1904
1905                 if (iter->trace && iter->trace->open)
1906                         iter->trace->open(iter);
1907         } else {
1908                 kfree(iter);
1909                 iter = NULL;
1910         }
1911         mutex_unlock(&trace_types_lock);
1912
1913  out:
1914         return iter;
1915 }
1916
1917 int tracing_open_generic(struct inode *inode, struct file *filp)
1918 {
1919         if (tracing_disabled)
1920                 return -ENODEV;
1921
1922         filp->private_data = inode->i_private;
1923         return 0;
1924 }
1925
1926 int tracing_release(struct inode *inode, struct file *file)
1927 {
1928         struct seq_file *m = (struct seq_file *)file->private_data;
1929         struct trace_iterator *iter = m->private;
1930
1931         mutex_lock(&trace_types_lock);
1932         if (iter->trace && iter->trace->close)
1933                 iter->trace->close(iter);
1934
1935         /* reenable tracing if it was previously enabled */
1936         if (iter->tr->ctrl) {
1937                 tracer_enabled = 1;
1938                 /*
1939                  * It is safe to enable function tracing even if it
1940                  * isn't used
1941                  */
1942                 ftrace_function_enabled = 1;
1943         }
1944         mutex_unlock(&trace_types_lock);
1945
1946         seq_release(inode, file);
1947         kfree(iter);
1948         return 0;
1949 }
1950
1951 static int tracing_open(struct inode *inode, struct file *file)
1952 {
1953         int ret;
1954
1955         __tracing_open(inode, file, &ret);
1956
1957         return ret;
1958 }
1959
1960 static int tracing_lt_open(struct inode *inode, struct file *file)
1961 {
1962         struct trace_iterator *iter;
1963         int ret;
1964
1965         iter = __tracing_open(inode, file, &ret);
1966
1967         if (!ret)
1968                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1969
1970         return ret;
1971 }
1972
1973
1974 static void *
1975 t_next(struct seq_file *m, void *v, loff_t *pos)
1976 {
1977         struct tracer *t = m->private;
1978
1979         (*pos)++;
1980
1981         if (t)
1982                 t = t->next;
1983
1984         m->private = t;
1985
1986         return t;
1987 }
1988
1989 static void *t_start(struct seq_file *m, loff_t *pos)
1990 {
1991         struct tracer *t = m->private;
1992         loff_t l = 0;
1993
1994         mutex_lock(&trace_types_lock);
1995         for (; t && l < *pos; t = t_next(m, t, &l))
1996                 ;
1997
1998         return t;
1999 }
2000
2001 static void t_stop(struct seq_file *m, void *p)
2002 {
2003         mutex_unlock(&trace_types_lock);
2004 }
2005
2006 static int t_show(struct seq_file *m, void *v)
2007 {
2008         struct tracer *t = v;
2009
2010         if (!t)
2011                 return 0;
2012
2013         seq_printf(m, "%s", t->name);
2014         if (t->next)
2015                 seq_putc(m, ' ');
2016         else
2017                 seq_putc(m, '\n');
2018
2019         return 0;
2020 }
2021
2022 static struct seq_operations show_traces_seq_ops = {
2023         .start          = t_start,
2024         .next           = t_next,
2025         .stop           = t_stop,
2026         .show           = t_show,
2027 };
2028
2029 static int show_traces_open(struct inode *inode, struct file *file)
2030 {
2031         int ret;
2032
2033         if (tracing_disabled)
2034                 return -ENODEV;
2035
2036         ret = seq_open(file, &show_traces_seq_ops);
2037         if (!ret) {
2038                 struct seq_file *m = file->private_data;
2039                 m->private = trace_types;
2040         }
2041
2042         return ret;
2043 }
2044
2045 static struct file_operations tracing_fops = {
2046         .open           = tracing_open,
2047         .read           = seq_read,
2048         .llseek         = seq_lseek,
2049         .release        = tracing_release,
2050 };
2051
2052 static struct file_operations tracing_lt_fops = {
2053         .open           = tracing_lt_open,
2054         .read           = seq_read,
2055         .llseek         = seq_lseek,
2056         .release        = tracing_release,
2057 };
2058
2059 static struct file_operations show_traces_fops = {
2060         .open           = show_traces_open,
2061         .read           = seq_read,
2062         .release        = seq_release,
2063 };
2064
2065 /*
2066  * Only trace on a CPU if the bitmask is set:
2067  */
2068 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2069
2070 /*
2071  * When tracing/tracing_cpu_mask is modified then this holds
2072  * the new bitmask we are about to install:
2073  */
2074 static cpumask_t tracing_cpumask_new;
2075
2076 /*
2077  * The tracer itself will not take this lock, but still we want
2078  * to provide a consistent cpumask to user-space:
2079  */
2080 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2081
2082 /*
2083  * Temporary storage for the character representation of the
2084  * CPU bitmask (and one more byte for the newline):
2085  */
2086 static char mask_str[NR_CPUS + 1];
2087
2088 static ssize_t
2089 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2090                      size_t count, loff_t *ppos)
2091 {
2092         int len;
2093
2094         mutex_lock(&tracing_cpumask_update_lock);
2095
2096         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2097         if (count - len < 2) {
2098                 count = -EINVAL;
2099                 goto out_err;
2100         }
2101         len += sprintf(mask_str + len, "\n");
2102         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2103
2104 out_err:
2105         mutex_unlock(&tracing_cpumask_update_lock);
2106
2107         return count;
2108 }
2109
2110 static ssize_t
2111 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2112                       size_t count, loff_t *ppos)
2113 {
2114         int err, cpu;
2115
2116         mutex_lock(&tracing_cpumask_update_lock);
2117         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2118         if (err)
2119                 goto err_unlock;
2120
2121         raw_local_irq_disable();
2122         __raw_spin_lock(&ftrace_max_lock);
2123         for_each_tracing_cpu(cpu) {
2124                 /*
2125                  * Increase/decrease the disabled counter if we are
2126                  * about to flip a bit in the cpumask:
2127                  */
2128                 if (cpu_isset(cpu, tracing_cpumask) &&
2129                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2130                         atomic_inc(&global_trace.data[cpu]->disabled);
2131                 }
2132                 if (!cpu_isset(cpu, tracing_cpumask) &&
2133                                 cpu_isset(cpu, tracing_cpumask_new)) {
2134                         atomic_dec(&global_trace.data[cpu]->disabled);
2135                 }
2136         }
2137         __raw_spin_unlock(&ftrace_max_lock);
2138         raw_local_irq_enable();
2139
2140         tracing_cpumask = tracing_cpumask_new;
2141
2142         mutex_unlock(&tracing_cpumask_update_lock);
2143
2144         return count;
2145
2146 err_unlock:
2147         mutex_unlock(&tracing_cpumask_update_lock);
2148
2149         return err;
2150 }
2151
2152 static struct file_operations tracing_cpumask_fops = {
2153         .open           = tracing_open_generic,
2154         .read           = tracing_cpumask_read,
2155         .write          = tracing_cpumask_write,
2156 };
2157
2158 static ssize_t
2159 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2160                        size_t cnt, loff_t *ppos)
2161 {
2162         char *buf;
2163         int r = 0;
2164         int len = 0;
2165         int i;
2166
2167         /* calulate max size */
2168         for (i = 0; trace_options[i]; i++) {
2169                 len += strlen(trace_options[i]);
2170                 len += 3; /* "no" and space */
2171         }
2172
2173         /* +2 for \n and \0 */
2174         buf = kmalloc(len + 2, GFP_KERNEL);
2175         if (!buf)
2176                 return -ENOMEM;
2177
2178         for (i = 0; trace_options[i]; i++) {
2179                 if (trace_flags & (1 << i))
2180                         r += sprintf(buf + r, "%s ", trace_options[i]);
2181                 else
2182                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2183         }
2184
2185         r += sprintf(buf + r, "\n");
2186         WARN_ON(r >= len + 2);
2187
2188         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2189
2190         kfree(buf);
2191
2192         return r;
2193 }
2194
2195 static ssize_t
2196 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2197                         size_t cnt, loff_t *ppos)
2198 {
2199         char buf[64];
2200         char *cmp = buf;
2201         int neg = 0;
2202         int i;
2203
2204         if (cnt >= sizeof(buf))
2205                 return -EINVAL;
2206
2207         if (copy_from_user(&buf, ubuf, cnt))
2208                 return -EFAULT;
2209
2210         buf[cnt] = 0;
2211
2212         if (strncmp(buf, "no", 2) == 0) {
2213                 neg = 1;
2214                 cmp += 2;
2215         }
2216
2217         for (i = 0; trace_options[i]; i++) {
2218                 int len = strlen(trace_options[i]);
2219
2220                 if (strncmp(cmp, trace_options[i], len) == 0) {
2221                         if (neg)
2222                                 trace_flags &= ~(1 << i);
2223                         else
2224                                 trace_flags |= (1 << i);
2225                         break;
2226                 }
2227         }
2228         /*
2229          * If no option could be set, return an error:
2230          */
2231         if (!trace_options[i])
2232                 return -EINVAL;
2233
2234         filp->f_pos += cnt;
2235
2236         return cnt;
2237 }
2238
2239 static struct file_operations tracing_iter_fops = {
2240         .open           = tracing_open_generic,
2241         .read           = tracing_iter_ctrl_read,
2242         .write          = tracing_iter_ctrl_write,
2243 };
2244
2245 static const char readme_msg[] =
2246         "tracing mini-HOWTO:\n\n"
2247         "# mkdir /debug\n"
2248         "# mount -t debugfs nodev /debug\n\n"
2249         "# cat /debug/tracing/available_tracers\n"
2250         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2251         "# cat /debug/tracing/current_tracer\n"
2252         "none\n"
2253         "# echo sched_switch > /debug/tracing/current_tracer\n"
2254         "# cat /debug/tracing/current_tracer\n"
2255         "sched_switch\n"
2256         "# cat /debug/tracing/iter_ctrl\n"
2257         "noprint-parent nosym-offset nosym-addr noverbose\n"
2258         "# echo print-parent > /debug/tracing/iter_ctrl\n"
2259         "# echo 1 > /debug/tracing/tracing_enabled\n"
2260         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2261         "echo 0 > /debug/tracing/tracing_enabled\n"
2262 ;
2263
2264 static ssize_t
2265 tracing_readme_read(struct file *filp, char __user *ubuf,
2266                        size_t cnt, loff_t *ppos)
2267 {
2268         return simple_read_from_buffer(ubuf, cnt, ppos,
2269                                         readme_msg, strlen(readme_msg));
2270 }
2271
2272 static struct file_operations tracing_readme_fops = {
2273         .open           = tracing_open_generic,
2274         .read           = tracing_readme_read,
2275 };
2276
2277 static ssize_t
2278 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2279                   size_t cnt, loff_t *ppos)
2280 {
2281         struct trace_array *tr = filp->private_data;
2282         char buf[64];
2283         int r;
2284
2285         r = sprintf(buf, "%ld\n", tr->ctrl);
2286         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2287 }
2288
2289 static ssize_t
2290 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2291                    size_t cnt, loff_t *ppos)
2292 {
2293         struct trace_array *tr = filp->private_data;
2294         char buf[64];
2295         long val;
2296         int ret;
2297
2298         if (cnt >= sizeof(buf))
2299                 return -EINVAL;
2300
2301         if (copy_from_user(&buf, ubuf, cnt))
2302                 return -EFAULT;
2303
2304         buf[cnt] = 0;
2305
2306         ret = strict_strtoul(buf, 10, &val);
2307         if (ret < 0)
2308                 return ret;
2309
2310         val = !!val;
2311
2312         mutex_lock(&trace_types_lock);
2313         if (tr->ctrl ^ val) {
2314                 if (val)
2315                         tracer_enabled = 1;
2316                 else
2317                         tracer_enabled = 0;
2318
2319                 tr->ctrl = val;
2320
2321                 if (current_trace && current_trace->ctrl_update)
2322                         current_trace->ctrl_update(tr);
2323         }
2324         mutex_unlock(&trace_types_lock);
2325
2326         filp->f_pos += cnt;
2327
2328         return cnt;
2329 }
2330
2331 static ssize_t
2332 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2333                        size_t cnt, loff_t *ppos)
2334 {
2335         char buf[max_tracer_type_len+2];
2336         int r;
2337
2338         mutex_lock(&trace_types_lock);
2339         if (current_trace)
2340                 r = sprintf(buf, "%s\n", current_trace->name);
2341         else
2342                 r = sprintf(buf, "\n");
2343         mutex_unlock(&trace_types_lock);
2344
2345         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2346 }
2347
2348 static ssize_t
2349 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2350                         size_t cnt, loff_t *ppos)
2351 {
2352         struct trace_array *tr = &global_trace;
2353         struct tracer *t;
2354         char buf[max_tracer_type_len+1];
2355         int i;
2356
2357         if (cnt > max_tracer_type_len)
2358                 cnt = max_tracer_type_len;
2359
2360         if (copy_from_user(&buf, ubuf, cnt))
2361                 return -EFAULT;
2362
2363         buf[cnt] = 0;
2364
2365         /* strip ending whitespace. */
2366         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2367                 buf[i] = 0;
2368
2369         mutex_lock(&trace_types_lock);
2370         for (t = trace_types; t; t = t->next) {
2371                 if (strcmp(t->name, buf) == 0)
2372                         break;
2373         }
2374         if (!t || t == current_trace)
2375                 goto out;
2376
2377         if (current_trace && current_trace->reset)
2378                 current_trace->reset(tr);
2379
2380         current_trace = t;
2381         if (t->init)
2382                 t->init(tr);
2383
2384  out:
2385         mutex_unlock(&trace_types_lock);
2386
2387         filp->f_pos += cnt;
2388
2389         return cnt;
2390 }
2391
2392 static ssize_t
2393 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2394                      size_t cnt, loff_t *ppos)
2395 {
2396         unsigned long *ptr = filp->private_data;
2397         char buf[64];
2398         int r;
2399
2400         r = snprintf(buf, sizeof(buf), "%ld\n",
2401                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2402         if (r > sizeof(buf))
2403                 r = sizeof(buf);
2404         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2405 }
2406
2407 static ssize_t
2408 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2409                       size_t cnt, loff_t *ppos)
2410 {
2411         long *ptr = filp->private_data;
2412         char buf[64];
2413         long val;
2414         int ret;
2415
2416         if (cnt >= sizeof(buf))
2417                 return -EINVAL;
2418
2419         if (copy_from_user(&buf, ubuf, cnt))
2420                 return -EFAULT;
2421
2422         buf[cnt] = 0;
2423
2424         ret = strict_strtoul(buf, 10, &val);
2425         if (ret < 0)
2426                 return ret;
2427
2428         *ptr = val * 1000;
2429
2430         return cnt;
2431 }
2432
2433 static atomic_t tracing_reader;
2434
2435 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2436 {
2437         struct trace_iterator *iter;
2438
2439         if (tracing_disabled)
2440                 return -ENODEV;
2441
2442         /* We only allow for reader of the pipe */
2443         if (atomic_inc_return(&tracing_reader) != 1) {
2444                 atomic_dec(&tracing_reader);
2445                 return -EBUSY;
2446         }
2447
2448         /* create a buffer to store the information to pass to userspace */
2449         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2450         if (!iter)
2451                 return -ENOMEM;
2452
2453         mutex_lock(&trace_types_lock);
2454         iter->tr = &global_trace;
2455         iter->trace = current_trace;
2456         filp->private_data = iter;
2457
2458         if (iter->trace->pipe_open)
2459                 iter->trace->pipe_open(iter);
2460         mutex_unlock(&trace_types_lock);
2461
2462         return 0;
2463 }
2464
2465 static int tracing_release_pipe(struct inode *inode, struct file *file)
2466 {
2467         struct trace_iterator *iter = file->private_data;
2468
2469         kfree(iter);
2470         atomic_dec(&tracing_reader);
2471
2472         return 0;
2473 }
2474
2475 static unsigned int
2476 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2477 {
2478         struct trace_iterator *iter = filp->private_data;
2479
2480         if (trace_flags & TRACE_ITER_BLOCK) {
2481                 /*
2482                  * Always select as readable when in blocking mode
2483                  */
2484                 return POLLIN | POLLRDNORM;
2485         } else {
2486                 if (!trace_empty(iter))
2487                         return POLLIN | POLLRDNORM;
2488                 poll_wait(filp, &trace_wait, poll_table);
2489                 if (!trace_empty(iter))
2490                         return POLLIN | POLLRDNORM;
2491
2492                 return 0;
2493         }
2494 }
2495
2496 /*
2497  * Consumer reader.
2498  */
2499 static ssize_t
2500 tracing_read_pipe(struct file *filp, char __user *ubuf,
2501                   size_t cnt, loff_t *ppos)
2502 {
2503         struct trace_iterator *iter = filp->private_data;
2504         struct trace_array_cpu *data;
2505         static cpumask_t mask;
2506         unsigned long flags;
2507 #ifdef CONFIG_FTRACE
2508         int ftrace_save;
2509 #endif
2510         int cpu;
2511         ssize_t sret;
2512
2513         /* return any leftover data */
2514         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2515         if (sret != -EBUSY)
2516                 return sret;
2517         sret = 0;
2518
2519         trace_seq_reset(&iter->seq);
2520
2521         mutex_lock(&trace_types_lock);
2522         if (iter->trace->read) {
2523                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2524                 if (sret)
2525                         goto out;
2526         }
2527
2528         while (trace_empty(iter)) {
2529
2530                 if ((filp->f_flags & O_NONBLOCK)) {
2531                         sret = -EAGAIN;
2532                         goto out;
2533                 }
2534
2535                 /*
2536                  * This is a make-shift waitqueue. The reason we don't use
2537                  * an actual wait queue is because:
2538                  *  1) we only ever have one waiter
2539                  *  2) the tracing, traces all functions, we don't want
2540                  *     the overhead of calling wake_up and friends
2541                  *     (and tracing them too)
2542                  *     Anyway, this is really very primitive wakeup.
2543                  */
2544                 set_current_state(TASK_INTERRUPTIBLE);
2545                 iter->tr->waiter = current;
2546
2547                 mutex_unlock(&trace_types_lock);
2548
2549                 /* sleep for 100 msecs, and try again. */
2550                 schedule_timeout(HZ/10);
2551
2552                 mutex_lock(&trace_types_lock);
2553
2554                 iter->tr->waiter = NULL;
2555
2556                 if (signal_pending(current)) {
2557                         sret = -EINTR;
2558                         goto out;
2559                 }
2560
2561                 if (iter->trace != current_trace)
2562                         goto out;
2563
2564                 /*
2565                  * We block until we read something and tracing is disabled.
2566                  * We still block if tracing is disabled, but we have never
2567                  * read anything. This allows a user to cat this file, and
2568                  * then enable tracing. But after we have read something,
2569                  * we give an EOF when tracing is again disabled.
2570                  *
2571                  * iter->pos will be 0 if we haven't read anything.
2572                  */
2573                 if (!tracer_enabled && iter->pos)
2574                         break;
2575
2576                 continue;
2577         }
2578
2579         /* stop when tracing is finished */
2580         if (trace_empty(iter))
2581                 goto out;
2582
2583         if (cnt >= PAGE_SIZE)
2584                 cnt = PAGE_SIZE - 1;
2585
2586         /* reset all but tr, trace, and overruns */
2587         memset(&iter->seq, 0,
2588                sizeof(struct trace_iterator) -
2589                offsetof(struct trace_iterator, seq));
2590         iter->pos = -1;
2591
2592         /*
2593          * We need to stop all tracing on all CPUS to read the
2594          * the next buffer. This is a bit expensive, but is
2595          * not done often. We fill all what we can read,
2596          * and then release the locks again.
2597          */
2598
2599         cpus_clear(mask);
2600         local_irq_save(flags);
2601 #ifdef CONFIG_FTRACE
2602         ftrace_save = ftrace_enabled;
2603         ftrace_enabled = 0;
2604 #endif
2605         smp_wmb();
2606         for_each_tracing_cpu(cpu) {
2607                 data = iter->tr->data[cpu];
2608
2609                 if (!head_page(data) || !data->trace_idx)
2610                         continue;
2611
2612                 atomic_inc(&data->disabled);
2613                 cpu_set(cpu, mask);
2614         }
2615
2616         for_each_cpu_mask(cpu, mask) {
2617                 data = iter->tr->data[cpu];
2618                 __raw_spin_lock(&data->lock);
2619
2620                 if (data->overrun > iter->last_overrun[cpu])
2621                         iter->overrun[cpu] +=
2622                                 data->overrun - iter->last_overrun[cpu];
2623                 iter->last_overrun[cpu] = data->overrun;
2624         }
2625
2626         while (find_next_entry_inc(iter) != NULL) {
2627                 int ret;
2628                 int len = iter->seq.len;
2629
2630                 ret = print_trace_line(iter);
2631                 if (!ret) {
2632                         /* don't print partial lines */
2633                         iter->seq.len = len;
2634                         break;
2635                 }
2636
2637                 trace_consume(iter);
2638
2639                 if (iter->seq.len >= cnt)
2640                         break;
2641         }
2642
2643         for_each_cpu_mask(cpu, mask) {
2644                 data = iter->tr->data[cpu];
2645                 __raw_spin_unlock(&data->lock);
2646         }
2647
2648         for_each_cpu_mask(cpu, mask) {
2649                 data = iter->tr->data[cpu];
2650                 atomic_dec(&data->disabled);
2651         }
2652 #ifdef CONFIG_FTRACE
2653         ftrace_enabled = ftrace_save;
2654 #endif
2655         local_irq_restore(flags);
2656
2657         /* Now copy what we have to the user */
2658         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2659         if (iter->seq.readpos >= iter->seq.len)
2660                 trace_seq_reset(&iter->seq);
2661         if (sret == -EBUSY)
2662                 sret = 0;
2663
2664 out:
2665         mutex_unlock(&trace_types_lock);
2666
2667         return sret;
2668 }
2669
2670 static ssize_t
2671 tracing_entries_read(struct file *filp, char __user *ubuf,
2672                      size_t cnt, loff_t *ppos)
2673 {
2674         struct trace_array *tr = filp->private_data;
2675         char buf[64];
2676         int r;
2677
2678         r = sprintf(buf, "%lu\n", tr->entries);
2679         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2680 }
2681
2682 static ssize_t
2683 tracing_entries_write(struct file *filp, const char __user *ubuf,
2684                       size_t cnt, loff_t *ppos)
2685 {
2686         unsigned long val;
2687         char buf[64];
2688         int i, ret;
2689
2690         if (cnt >= sizeof(buf))
2691                 return -EINVAL;
2692
2693         if (copy_from_user(&buf, ubuf, cnt))
2694                 return -EFAULT;
2695
2696         buf[cnt] = 0;
2697
2698         ret = strict_strtoul(buf, 10, &val);
2699         if (ret < 0)
2700                 return ret;
2701
2702         /* must have at least 1 entry */
2703         if (!val)
2704                 return -EINVAL;
2705
2706         mutex_lock(&trace_types_lock);
2707
2708         if (current_trace != &no_tracer) {
2709                 cnt = -EBUSY;
2710                 pr_info("ftrace: set current_tracer to none"
2711                         " before modifying buffer size\n");
2712                 goto out;
2713         }
2714
2715         if (val > global_trace.entries) {
2716                 long pages_requested;
2717                 unsigned long freeable_pages;
2718
2719                 /* make sure we have enough memory before mapping */
2720                 pages_requested =
2721                         (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2722
2723                 /* account for each buffer (and max_tr) */
2724                 pages_requested *= tracing_nr_buffers * 2;
2725
2726                 /* Check for overflow */
2727                 if (pages_requested < 0) {
2728                         cnt = -ENOMEM;
2729                         goto out;
2730                 }
2731
2732                 freeable_pages = determine_dirtyable_memory();
2733
2734                 /* we only allow to request 1/4 of useable memory */
2735                 if (pages_requested >
2736                     ((freeable_pages + tracing_pages_allocated) / 4)) {
2737                         cnt = -ENOMEM;
2738                         goto out;
2739                 }
2740
2741                 while (global_trace.entries < val) {
2742                         if (trace_alloc_page()) {
2743                                 cnt = -ENOMEM;
2744                                 goto out;
2745                         }
2746                         /* double check that we don't go over the known pages */
2747                         if (tracing_pages_allocated > pages_requested)
2748                                 break;
2749                 }
2750
2751         } else {
2752                 /* include the number of entries in val (inc of page entries) */
2753                 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2754                         trace_free_page();
2755         }
2756
2757         /* check integrity */
2758         for_each_tracing_cpu(i)
2759                 check_pages(global_trace.data[i]);
2760
2761         filp->f_pos += cnt;
2762
2763         /* If check pages failed, return ENOMEM */
2764         if (tracing_disabled)
2765                 cnt = -ENOMEM;
2766  out:
2767         max_tr.entries = global_trace.entries;
2768         mutex_unlock(&trace_types_lock);
2769
2770         return cnt;
2771 }
2772
2773 static struct file_operations tracing_max_lat_fops = {
2774         .open           = tracing_open_generic,
2775         .read           = tracing_max_lat_read,
2776         .write          = tracing_max_lat_write,
2777 };
2778
2779 static struct file_operations tracing_ctrl_fops = {
2780         .open           = tracing_open_generic,
2781         .read           = tracing_ctrl_read,
2782         .write          = tracing_ctrl_write,
2783 };
2784
2785 static struct file_operations set_tracer_fops = {
2786         .open           = tracing_open_generic,
2787         .read           = tracing_set_trace_read,
2788         .write          = tracing_set_trace_write,
2789 };
2790
2791 static struct file_operations tracing_pipe_fops = {
2792         .open           = tracing_open_pipe,
2793         .poll           = tracing_poll_pipe,
2794         .read           = tracing_read_pipe,
2795         .release        = tracing_release_pipe,
2796 };
2797
2798 static struct file_operations tracing_entries_fops = {
2799         .open           = tracing_open_generic,
2800         .read           = tracing_entries_read,
2801         .write          = tracing_entries_write,
2802 };
2803
2804 #ifdef CONFIG_DYNAMIC_FTRACE
2805
2806 static ssize_t
2807 tracing_read_long(struct file *filp, char __user *ubuf,
2808                   size_t cnt, loff_t *ppos)
2809 {
2810         unsigned long *p = filp->private_data;
2811         char buf[64];
2812         int r;
2813
2814         r = sprintf(buf, "%ld\n", *p);
2815
2816         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2817 }
2818
2819 static struct file_operations tracing_read_long_fops = {
2820         .open           = tracing_open_generic,
2821         .read           = tracing_read_long,
2822 };
2823 #endif
2824
2825 static struct dentry *d_tracer;
2826
2827 struct dentry *tracing_init_dentry(void)
2828 {
2829         static int once;
2830
2831         if (d_tracer)
2832                 return d_tracer;
2833
2834         d_tracer = debugfs_create_dir("tracing", NULL);
2835
2836         if (!d_tracer && !once) {
2837                 once = 1;
2838                 pr_warning("Could not create debugfs directory 'tracing'\n");
2839                 return NULL;
2840         }
2841
2842         return d_tracer;
2843 }
2844
2845 #ifdef CONFIG_FTRACE_SELFTEST
2846 /* Let selftest have access to static functions in this file */
2847 #include "trace_selftest.c"
2848 #endif
2849
2850 static __init void tracer_init_debugfs(void)
2851 {
2852         struct dentry *d_tracer;
2853         struct dentry *entry;
2854
2855         d_tracer = tracing_init_dentry();
2856
2857         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2858                                     &global_trace, &tracing_ctrl_fops);
2859         if (!entry)
2860                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2861
2862         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2863                                     NULL, &tracing_iter_fops);
2864         if (!entry)
2865                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2866
2867         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2868                                     NULL, &tracing_cpumask_fops);
2869         if (!entry)
2870                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2871
2872         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2873                                     &global_trace, &tracing_lt_fops);
2874         if (!entry)
2875                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2876
2877         entry = debugfs_create_file("trace", 0444, d_tracer,
2878                                     &global_trace, &tracing_fops);
2879         if (!entry)
2880                 pr_warning("Could not create debugfs 'trace' entry\n");
2881
2882         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2883                                     &global_trace, &show_traces_fops);
2884         if (!entry)
2885                 pr_warning("Could not create debugfs 'trace' entry\n");
2886
2887         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2888                                     &global_trace, &set_tracer_fops);
2889         if (!entry)
2890                 pr_warning("Could not create debugfs 'trace' entry\n");
2891
2892         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2893                                     &tracing_max_latency,
2894                                     &tracing_max_lat_fops);
2895         if (!entry)
2896                 pr_warning("Could not create debugfs "
2897                            "'tracing_max_latency' entry\n");
2898
2899         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2900                                     &tracing_thresh, &tracing_max_lat_fops);
2901         if (!entry)
2902                 pr_warning("Could not create debugfs "
2903                            "'tracing_threash' entry\n");
2904         entry = debugfs_create_file("README", 0644, d_tracer,
2905                                     NULL, &tracing_readme_fops);
2906         if (!entry)
2907                 pr_warning("Could not create debugfs 'README' entry\n");
2908
2909         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2910                                     NULL, &tracing_pipe_fops);
2911         if (!entry)
2912                 pr_warning("Could not create debugfs "
2913                            "'tracing_threash' entry\n");
2914
2915         entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2916                                     &global_trace, &tracing_entries_fops);
2917         if (!entry)
2918                 pr_warning("Could not create debugfs "
2919                            "'tracing_threash' entry\n");
2920
2921 #ifdef CONFIG_DYNAMIC_FTRACE
2922         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2923                                     &ftrace_update_tot_cnt,
2924                                     &tracing_read_long_fops);
2925         if (!entry)
2926                 pr_warning("Could not create debugfs "
2927                            "'dyn_ftrace_total_info' entry\n");
2928 #endif
2929 #ifdef CONFIG_SYSPROF_TRACER
2930         init_tracer_sysprof_debugfs(d_tracer);
2931 #endif
2932 }
2933
2934 static int trace_alloc_page(void)
2935 {
2936         struct trace_array_cpu *data;
2937         struct page *page, *tmp;
2938         LIST_HEAD(pages);
2939         void *array;
2940         unsigned pages_allocated = 0;
2941         int i;
2942
2943         /* first allocate a page for each CPU */
2944         for_each_tracing_cpu(i) {
2945                 array = (void *)__get_free_page(GFP_KERNEL);
2946                 if (array == NULL) {
2947                         printk(KERN_ERR "tracer: failed to allocate page"
2948                                "for trace buffer!\n");
2949                         goto free_pages;
2950                 }
2951
2952                 pages_allocated++;
2953                 page = virt_to_page(array);
2954                 list_add(&page->lru, &pages);
2955
2956 /* Only allocate if we are actually using the max trace */
2957 #ifdef CONFIG_TRACER_MAX_TRACE
2958                 array = (void *)__get_free_page(GFP_KERNEL);
2959                 if (array == NULL) {
2960                         printk(KERN_ERR "tracer: failed to allocate page"
2961                                "for trace buffer!\n");
2962                         goto free_pages;
2963                 }
2964                 pages_allocated++;
2965                 page = virt_to_page(array);
2966                 list_add(&page->lru, &pages);
2967 #endif
2968         }
2969
2970         /* Now that we successfully allocate a page per CPU, add them */
2971         for_each_tracing_cpu(i) {
2972                 data = global_trace.data[i];
2973                 page = list_entry(pages.next, struct page, lru);
2974                 list_del_init(&page->lru);
2975                 list_add_tail(&page->lru, &data->trace_pages);
2976                 ClearPageLRU(page);
2977
2978 #ifdef CONFIG_TRACER_MAX_TRACE
2979                 data = max_tr.data[i];
2980                 page = list_entry(pages.next, struct page, lru);
2981                 list_del_init(&page->lru);
2982                 list_add_tail(&page->lru, &data->trace_pages);
2983                 SetPageLRU(page);
2984 #endif
2985         }
2986         tracing_pages_allocated += pages_allocated;
2987         global_trace.entries += ENTRIES_PER_PAGE;
2988
2989         return 0;
2990
2991  free_pages:
2992         list_for_each_entry_safe(page, tmp, &pages, lru) {
2993                 list_del_init(&page->lru);
2994                 __free_page(page);
2995         }
2996         return -ENOMEM;
2997 }
2998
2999 static int trace_free_page(void)
3000 {
3001         struct trace_array_cpu *data;
3002         struct page *page;
3003         struct list_head *p;
3004         int i;
3005         int ret = 0;
3006
3007         /* free one page from each buffer */
3008         for_each_tracing_cpu(i) {
3009                 data = global_trace.data[i];
3010                 p = data->trace_pages.next;
3011                 if (p == &data->trace_pages) {
3012                         /* should never happen */
3013                         WARN_ON(1);
3014                         tracing_disabled = 1;
3015                         ret = -1;
3016                         break;
3017                 }
3018                 page = list_entry(p, struct page, lru);
3019                 ClearPageLRU(page);
3020                 list_del(&page->lru);
3021                 tracing_pages_allocated--;
3022                 tracing_pages_allocated--;
3023                 __free_page(page);
3024
3025                 tracing_reset(data);
3026
3027 #ifdef CONFIG_TRACER_MAX_TRACE
3028                 data = max_tr.data[i];
3029                 p = data->trace_pages.next;
3030                 if (p == &data->trace_pages) {
3031                         /* should never happen */
3032                         WARN_ON(1);
3033                         tracing_disabled = 1;
3034                         ret = -1;
3035                         break;
3036                 }
3037                 page = list_entry(p, struct page, lru);
3038                 ClearPageLRU(page);
3039                 list_del(&page->lru);
3040                 __free_page(page);
3041
3042                 tracing_reset(data);
3043 #endif
3044         }
3045         global_trace.entries -= ENTRIES_PER_PAGE;
3046
3047         return ret;
3048 }
3049
3050 __init static int tracer_alloc_buffers(void)
3051 {
3052         struct trace_array_cpu *data;
3053         void *array;
3054         struct page *page;
3055         int pages = 0;
3056         int ret = -ENOMEM;
3057         int i;
3058
3059         /* TODO: make the number of buffers hot pluggable with CPUS */
3060         tracing_nr_buffers = num_possible_cpus();
3061         tracing_buffer_mask = cpu_possible_map;
3062
3063         /* Allocate the first page for all buffers */
3064         for_each_tracing_cpu(i) {
3065                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3066                 max_tr.data[i] = &per_cpu(max_data, i);
3067
3068                 array = (void *)__get_free_page(GFP_KERNEL);
3069                 if (array == NULL) {
3070                         printk(KERN_ERR "tracer: failed to allocate page"
3071                                "for trace buffer!\n");
3072                         goto free_buffers;
3073                 }
3074
3075                 /* set the array to the list */
3076                 INIT_LIST_HEAD(&data->trace_pages);
3077                 page = virt_to_page(array);
3078                 list_add(&page->lru, &data->trace_pages);
3079                 /* use the LRU flag to differentiate the two buffers */
3080                 ClearPageLRU(page);
3081
3082                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
3083                 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
3084
3085 /* Only allocate if we are actually using the max trace */
3086 #ifdef CONFIG_TRACER_MAX_TRACE
3087                 array = (void *)__get_free_page(GFP_KERNEL);
3088                 if (array == NULL) {
3089                         printk(KERN_ERR "tracer: failed to allocate page"
3090                                "for trace buffer!\n");
3091                         goto free_buffers;
3092                 }
3093
3094                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
3095                 page = virt_to_page(array);
3096                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
3097                 SetPageLRU(page);
3098 #endif
3099         }
3100
3101         /*
3102          * Since we allocate by orders of pages, we may be able to
3103          * round up a bit.
3104          */
3105         global_trace.entries = ENTRIES_PER_PAGE;
3106         pages++;
3107
3108         while (global_trace.entries < trace_nr_entries) {
3109                 if (trace_alloc_page())
3110                         break;
3111                 pages++;
3112         }
3113         max_tr.entries = global_trace.entries;
3114
3115         pr_info("tracer: %d pages allocated for %ld entries of %ld bytes\n",
3116                 pages, trace_nr_entries, (long)TRACE_ENTRY_SIZE);
3117         pr_info("   actual entries %ld\n", global_trace.entries);
3118
3119         tracer_init_debugfs();
3120
3121         trace_init_cmdlines();
3122
3123         register_tracer(&no_tracer);
3124         current_trace = &no_tracer;
3125
3126         /* All seems OK, enable tracing */
3127         global_trace.ctrl = tracer_enabled;
3128         tracing_disabled = 0;
3129
3130         return 0;
3131
3132  free_buffers:
3133         for (i-- ; i >= 0; i--) {
3134                 struct page *page, *tmp;
3135                 struct trace_array_cpu *data = global_trace.data[i];
3136
3137                 if (data) {
3138                         list_for_each_entry_safe(page, tmp,
3139                                                  &data->trace_pages, lru) {
3140                                 list_del_init(&page->lru);
3141                                 __free_page(page);
3142                         }
3143                 }
3144
3145 #ifdef CONFIG_TRACER_MAX_TRACE
3146                 data = max_tr.data[i];
3147                 if (data) {
3148                         list_for_each_entry_safe(page, tmp,
3149                                                  &data->trace_pages, lru) {
3150                                 list_del_init(&page->lru);
3151                                 __free_page(page);
3152                         }
3153                 }
3154 #endif
3155         }
3156         return ret;
3157 }
3158 fs_initcall(tracer_alloc_buffers);