git.oblomov.eu Git - linux-2.6/blob - kernel/trace/trace.c

   1 /*
   2  * ring buffer based function tracer
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally taken from the RT patch by:
   8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code from the latency_tracer, that is:
  11  *  Copyright (C) 2004-2006 Ingo Molnar
  12  *  Copyright (C) 2004 William Lee Irwin III
  13  */
  14 #include <linux/utsrelease.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/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/gfp.h>
  28 #include <linux/fs.h>
  29
  30 #include "trace.h"
  31
  32 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  33 unsigned long __read_mostly     tracing_thresh;
  34
  35 static int tracing_disabled = 1;
  36
  37 static long notrace
  38 ns2usecs(cycle_t nsec)
  39 {
  40         nsec += 500;
  41         do_div(nsec, 1000);
  42         return nsec;
  43 }
  44
  45 static atomic_t                 tracer_counter;
  46 static struct trace_array       global_trace;
  47
  48 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  49
  50 static struct trace_array       max_tr;
  51
  52 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  53
  54 static int                      tracer_enabled;
  55 static unsigned long            trace_nr_entries = 16384UL;
  56
  57 static struct tracer            *trace_types __read_mostly;
  58 static struct tracer            *current_trace __read_mostly;
  59 static int                      max_tracer_type_len;
  60
  61 static DEFINE_MUTEX(trace_types_lock);
  62
  63 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  64
  65 static int __init set_nr_entries(char *str)
  66 {
  67         if (!str)
  68                 return 0;
  69         trace_nr_entries = simple_strtoul(str, &str, 0);
  70         return 1;
  71 }
  72 __setup("trace_entries=", set_nr_entries);
  73
  74 unsigned long nsecs_to_usecs(unsigned long nsecs)
  75 {
  76         return nsecs / 1000;
  77 }
  78
  79 enum trace_type {
  80         __TRACE_FIRST_TYPE = 0,
  81
  82         TRACE_FN,
  83         TRACE_CTX,
  84
  85         __TRACE_LAST_TYPE
  86 };
  87
  88 enum trace_flag_type {
  89         TRACE_FLAG_IRQS_OFF             = 0x01,
  90         TRACE_FLAG_NEED_RESCHED         = 0x02,
  91         TRACE_FLAG_HARDIRQ              = 0x04,
  92         TRACE_FLAG_SOFTIRQ              = 0x08,
  93 };
  94
  95 enum trace_iterator_flags {
  96         TRACE_ITER_PRINT_PARENT         = 0x01,
  97         TRACE_ITER_SYM_OFFSET           = 0x02,
  98         TRACE_ITER_SYM_ADDR             = 0x04,
  99         TRACE_ITER_VERBOSE              = 0x08,
 100 };
 101
 102 #define TRACE_ITER_SYM_MASK \
 103         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 104
 105 /* These must match the bit postions above */
 106 static const char *trace_options[] = {
 107         "print-parent",
 108         "sym-offset",
 109         "sym-addr",
 110         "verbose",
 111         NULL
 112 };
 113
 114 static unsigned trace_flags;
 115
 116 static DEFINE_SPINLOCK(ftrace_max_lock);
 117
 118 /*
 119  * Copy the new maximum trace into the separate maximum-trace
 120  * structure. (this way the maximum trace is permanently saved,
 121  * for later retrieval via /debugfs/tracing/latency_trace)
 122  */
 123 static void notrace
 124 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 125 {
 126         struct trace_array_cpu *data = tr->data[cpu];
 127
 128         max_tr.cpu = cpu;
 129         max_tr.time_start = data->preempt_timestamp;
 130
 131         data = max_tr.data[cpu];
 132         data->saved_latency = tracing_max_latency;
 133
 134         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 135         data->pid = tsk->pid;
 136         data->uid = tsk->uid;
 137         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 138         data->policy = tsk->policy;
 139         data->rt_priority = tsk->rt_priority;
 140
 141         /* record this tasks comm */
 142         tracing_record_cmdline(current);
 143 }
 144
 145 notrace void
 146 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 147 {
 148         struct trace_array_cpu *data;
 149         void *save_trace;
 150         struct list_head save_pages;
 151         int i;
 152
 153         WARN_ON_ONCE(!irqs_disabled());
 154         spin_lock(&ftrace_max_lock);
 155         /* clear out all the previous traces */
 156         for_each_possible_cpu(i) {
 157                 data = tr->data[i];
 158                 save_trace = max_tr.data[i]->trace;
 159                 save_pages = max_tr.data[i]->trace_pages;
 160                 memcpy(max_tr.data[i], data, sizeof(*data));
 161                 data->trace = save_trace;
 162                 data->trace_pages = save_pages;
 163                 tracing_reset(data);
 164         }
 165
 166         __update_max_tr(tr, tsk, cpu);
 167         spin_unlock(&ftrace_max_lock);
 168 }
 169
 170 /**
 171  * update_max_tr_single - only copy one trace over, and reset the rest
 172  * @tr - tracer
 173  * @tsk - task with the latency
 174  * @cpu - the cpu of the buffer to copy.
 175  */
 176 notrace void
 177 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 178 {
 179         struct trace_array_cpu *data = tr->data[cpu];
 180         void *save_trace;
 181         struct list_head save_pages;
 182         int i;
 183
 184         WARN_ON_ONCE(!irqs_disabled());
 185         spin_lock(&ftrace_max_lock);
 186         for_each_possible_cpu(i)
 187                 tracing_reset(max_tr.data[i]);
 188
 189         save_trace = max_tr.data[cpu]->trace;
 190         save_pages = max_tr.data[cpu]->trace_pages;
 191         memcpy(max_tr.data[cpu], data, sizeof(*data));
 192         data->trace = save_trace;
 193         data->trace_pages = save_pages;
 194         tracing_reset(data);
 195
 196         __update_max_tr(tr, tsk, cpu);
 197         spin_unlock(&ftrace_max_lock);
 198 }
 199
 200 int register_tracer(struct tracer *type)
 201 {
 202         struct tracer *t;
 203         int len;
 204         int ret = 0;
 205
 206         if (!type->name) {
 207                 pr_info("Tracer must have a name\n");
 208                 return -1;
 209         }
 210
 211         mutex_lock(&trace_types_lock);
 212         for (t = trace_types; t; t = t->next) {
 213                 if (strcmp(type->name, t->name) == 0) {
 214                         /* already found */
 215                         pr_info("Trace %s already registered\n",
 216                                 type->name);
 217                         ret = -1;
 218                         goto out;
 219                 }
 220         }
 221
 222 #ifdef CONFIG_FTRACE_STARTUP_TEST
 223         if (type->selftest) {
 224                 struct tracer *saved_tracer = current_trace;
 225                 struct trace_array_cpu *data;
 226                 struct trace_array *tr = &global_trace;
 227                 int saved_ctrl = tr->ctrl;
 228                 int i;
 229                 /*
 230                  * Run a selftest on this tracer.
 231                  * Here we reset the trace buffer, and set the current
 232                  * tracer to be this tracer. The tracer can then run some
 233                  * internal tracing to verify that everything is in order.
 234                  * If we fail, we do not register this tracer.
 235                  */
 236                 for_each_possible_cpu(i) {
 237                         if (!data->trace)
 238                                 continue;
 239                         data = tr->data[i];
 240                         tracing_reset(data);
 241                 }
 242                 current_trace = type;
 243                 tr->ctrl = 0;
 244                 /* the test is responsible for initializing and enabling */
 245                 pr_info("Testing tracer %s: ", type->name);
 246                 ret = type->selftest(type, tr);
 247                 /* the test is responsible for resetting too */
 248                 current_trace = saved_tracer;
 249                 tr->ctrl = saved_ctrl;
 250                 if (ret) {
 251                         printk(KERN_CONT "FAILED!\n");
 252                         goto out;
 253                 }
 254                 printk(KERN_CONT "PASSED\n");
 255         }
 256 #endif
 257
 258         type->next = trace_types;
 259         trace_types = type;
 260         len = strlen(type->name);
 261         if (len > max_tracer_type_len)
 262                 max_tracer_type_len = len;
 263
 264  out:
 265         mutex_unlock(&trace_types_lock);
 266
 267         return ret;
 268 }
 269
 270 void unregister_tracer(struct tracer *type)
 271 {
 272         struct tracer **t;
 273         int len;
 274
 275         mutex_lock(&trace_types_lock);
 276         for (t = &trace_types; *t; t = &(*t)->next) {
 277                 if (*t == type)
 278                         goto found;
 279         }
 280         pr_info("Trace %s not registered\n", type->name);
 281         goto out;
 282
 283  found:
 284         *t = (*t)->next;
 285         if (strlen(type->name) != max_tracer_type_len)
 286                 goto out;
 287
 288         max_tracer_type_len = 0;
 289         for (t = &trace_types; *t; t = &(*t)->next) {
 290                 len = strlen((*t)->name);
 291                 if (len > max_tracer_type_len)
 292                         max_tracer_type_len = len;
 293         }
 294  out:
 295         mutex_unlock(&trace_types_lock);
 296 }
 297
 298 void notrace tracing_reset(struct trace_array_cpu *data)
 299 {
 300         data->trace_idx = 0;
 301         data->trace_current = data->trace;
 302         data->trace_current_idx = 0;
 303 }
 304
 305 #ifdef CONFIG_FTRACE
 306 static void notrace
 307 function_trace_call(unsigned long ip, unsigned long parent_ip)
 308 {
 309         struct trace_array *tr = &global_trace;
 310         struct trace_array_cpu *data;
 311         unsigned long flags;
 312         long disabled;
 313         int cpu;
 314
 315         if (unlikely(!tracer_enabled))
 316                 return;
 317
 318         local_irq_save(flags);
 319         cpu = raw_smp_processor_id();
 320         data = tr->data[cpu];
 321         disabled = atomic_inc_return(&data->disabled);
 322
 323         if (likely(disabled == 1))
 324                 ftrace(tr, data, ip, parent_ip, flags);
 325
 326         atomic_dec(&data->disabled);
 327         local_irq_restore(flags);
 328 }
 329
 330 static struct ftrace_ops trace_ops __read_mostly =
 331 {
 332         .func = function_trace_call,
 333 };
 334 #endif
 335
 336 notrace void tracing_start_function_trace(void)
 337 {
 338         register_ftrace_function(&trace_ops);
 339 }
 340
 341 notrace void tracing_stop_function_trace(void)
 342 {
 343         unregister_ftrace_function(&trace_ops);
 344 }
 345
 346 #define SAVED_CMDLINES 128
 347 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 348 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 349 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 350 static int cmdline_idx;
 351 static DEFINE_SPINLOCK(trace_cmdline_lock);
 352 atomic_t trace_record_cmdline_disabled;
 353
 354 static void trace_init_cmdlines(void)
 355 {
 356         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 357         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 358         cmdline_idx = 0;
 359 }
 360
 361 notrace void trace_stop_cmdline_recording(void);
 362
 363 static void notrace trace_save_cmdline(struct task_struct *tsk)
 364 {
 365         unsigned map;
 366         unsigned idx;
 367
 368         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 369                 return;
 370
 371         /*
 372          * It's not the end of the world if we don't get
 373          * the lock, but we also don't want to spin
 374          * nor do we want to disable interrupts,
 375          * so if we miss here, then better luck next time.
 376          */
 377         if (!spin_trylock(&trace_cmdline_lock))
 378                 return;
 379
 380         idx = map_pid_to_cmdline[tsk->pid];
 381         if (idx >= SAVED_CMDLINES) {
 382                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 383
 384                 map = map_cmdline_to_pid[idx];
 385                 if (map <= PID_MAX_DEFAULT)
 386                         map_pid_to_cmdline[map] = (unsigned)-1;
 387
 388                 map_pid_to_cmdline[tsk->pid] = idx;
 389
 390                 cmdline_idx = idx;
 391         }
 392
 393         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 394
 395         spin_unlock(&trace_cmdline_lock);
 396 }
 397
 398 static notrace char *trace_find_cmdline(int pid)
 399 {
 400         char *cmdline = "<...>";
 401         unsigned map;
 402
 403         if (!pid)
 404                 return "<idle>";
 405
 406         if (pid > PID_MAX_DEFAULT)
 407                 goto out;
 408
 409         map = map_pid_to_cmdline[pid];
 410         if (map >= SAVED_CMDLINES)
 411                 goto out;
 412
 413         cmdline = saved_cmdlines[map];
 414
 415  out:
 416         return cmdline;
 417 }
 418
 419 notrace void tracing_record_cmdline(struct task_struct *tsk)
 420 {
 421         if (atomic_read(&trace_record_cmdline_disabled))
 422                 return;
 423
 424         trace_save_cmdline(tsk);
 425 }
 426
 427 static inline notrace struct trace_entry *
 428 tracing_get_trace_entry(struct trace_array *tr,
 429                         struct trace_array_cpu *data)
 430 {
 431         unsigned long idx, idx_next;
 432         struct trace_entry *entry;
 433         struct page *page;
 434         struct list_head *next;
 435
 436         data->trace_idx++;
 437         idx = data->trace_current_idx;
 438         idx_next = idx + 1;
 439
 440         entry = data->trace_current + idx * TRACE_ENTRY_SIZE;
 441
 442         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 443                 page = virt_to_page(data->trace_current);
 444                 if (unlikely(&page->lru == data->trace_pages.prev))
 445                         next = data->trace_pages.next;
 446                 else
 447                         next = page->lru.next;
 448                 page = list_entry(next, struct page, lru);
 449                 data->trace_current = page_address(page);
 450                 idx_next = 0;
 451         }
 452
 453         data->trace_current_idx = idx_next;
 454
 455         return entry;
 456 }
 457
 458 static inline notrace void
 459 tracing_generic_entry_update(struct trace_entry *entry,
 460                              unsigned long flags)
 461 {
 462         struct task_struct *tsk = current;
 463         unsigned long pc;
 464
 465         pc = preempt_count();
 466
 467         entry->idx      = atomic_inc_return(&tracer_counter);
 468         entry->preempt_count = pc & 0xff;
 469         entry->pid       = tsk->pid;
 470         entry->t         = now(raw_smp_processor_id());
 471         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 472                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 473                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 474                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 475 }
 476
 477 notrace void
 478 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 479                        unsigned long ip, unsigned long parent_ip,
 480                        unsigned long flags)
 481 {
 482         struct trace_entry *entry;
 483
 484         entry = tracing_get_trace_entry(tr, data);
 485         tracing_generic_entry_update(entry, flags);
 486         entry->type         = TRACE_FN;
 487         entry->fn.ip        = ip;
 488         entry->fn.parent_ip = parent_ip;
 489 }
 490
 491 notrace void
 492 tracing_sched_switch_trace(struct trace_array *tr,
 493                            struct trace_array_cpu *data,
 494                            struct task_struct *prev, struct task_struct *next,
 495                            unsigned long flags)
 496 {
 497         struct trace_entry *entry;
 498
 499         entry = tracing_get_trace_entry(tr, data);
 500         tracing_generic_entry_update(entry, flags);
 501         entry->type             = TRACE_CTX;
 502         entry->ctx.prev_pid     = prev->pid;
 503         entry->ctx.prev_prio    = prev->prio;
 504         entry->ctx.prev_state   = prev->state;
 505         entry->ctx.next_pid     = next->pid;
 506         entry->ctx.next_prio    = next->prio;
 507 }
 508
 509 enum trace_file_type {
 510         TRACE_FILE_LAT_FMT      = 1,
 511 };
 512
 513 static struct trace_entry *
 514 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 515                 struct trace_iterator *iter, int cpu)
 516 {
 517         struct page *page;
 518         struct trace_entry *array;
 519
 520         if (iter->next_idx[cpu] >= tr->entries ||
 521             iter->next_idx[cpu] >= data->trace_idx)
 522                 return NULL;
 523
 524         if (!iter->next_page[cpu]) {
 525                 /*
 526                  * Initialize. If the count of elements in
 527                  * this buffer is greater than the max entries
 528                  * we had an underrun. Which means we looped around.
 529                  * We can simply use the current pointer as our
 530                  * starting point.
 531                  */
 532                 if (data->trace_idx >= tr->entries) {
 533                         page = virt_to_page(data->trace_current);
 534                         iter->next_page[cpu] = &page->lru;
 535                         iter->next_page_idx[cpu] = data->trace_current_idx;
 536                 } else {
 537                         iter->next_page[cpu] = data->trace_pages.next;
 538                         iter->next_page_idx[cpu] = 0;
 539                 }
 540         }
 541
 542         page = list_entry(iter->next_page[cpu], struct page, lru);
 543         array = page_address(page);
 544
 545         return &array[iter->next_page_idx[cpu]];
 546 }
 547
 548 static struct notrace trace_entry *
 549 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 550 {
 551         struct trace_array *tr = iter->tr;
 552         struct trace_entry *ent, *next = NULL;
 553         int next_cpu = -1;
 554         int cpu;
 555
 556         for_each_possible_cpu(cpu) {
 557                 if (!tr->data[cpu]->trace)
 558                         continue;
 559                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 560                 if (ent &&
 561                     (!next || (long)(next->idx - ent->idx) > 0)) {
 562                         next = ent;
 563                         next_cpu = cpu;
 564                 }
 565         }
 566
 567         if (ent_cpu)
 568                 *ent_cpu = next_cpu;
 569
 570         return next;
 571 }
 572
 573 static void *find_next_entry_inc(struct trace_iterator *iter)
 574 {
 575         struct trace_entry *next;
 576         int next_cpu = -1;
 577
 578         next = find_next_entry(iter, &next_cpu);
 579
 580         if (next) {
 581                 iter->idx++;
 582                 iter->next_idx[next_cpu]++;
 583                 iter->next_page_idx[next_cpu]++;
 584                 if (iter->next_page_idx[next_cpu] >= ENTRIES_PER_PAGE) {
 585                         struct trace_array_cpu *data = iter->tr->data[next_cpu];
 586
 587                         iter->next_page_idx[next_cpu] = 0;
 588                         iter->next_page[next_cpu] =
 589                                 iter->next_page[next_cpu]->next;
 590                         if (iter->next_page[next_cpu] == &data->trace_pages)
 591                                 iter->next_page[next_cpu] =
 592                                         data->trace_pages.next;
 593                 }
 594         }
 595         iter->ent = next;
 596         iter->cpu = next_cpu;
 597
 598         return next ? iter : NULL;
 599 }
 600
 601 static void notrace *
 602 s_next(struct seq_file *m, void *v, loff_t *pos)
 603 {
 604         struct trace_iterator *iter = m->private;
 605         void *ent;
 606         void *last_ent = iter->ent;
 607         int i = (int)*pos;
 608
 609         (*pos)++;
 610
 611         /* can't go backwards */
 612         if (iter->idx > i)
 613                 return NULL;
 614
 615         if (iter->idx < 0)
 616                 ent = find_next_entry_inc(iter);
 617         else
 618                 ent = iter;
 619
 620         while (ent && iter->idx < i)
 621                 ent = find_next_entry_inc(iter);
 622
 623         iter->pos = *pos;
 624
 625         if (last_ent && !ent)
 626                 seq_puts(m, "\n\nvim:ft=help\n");
 627
 628         return ent;
 629 }
 630
 631 static void *s_start(struct seq_file *m, loff_t *pos)
 632 {
 633         struct trace_iterator *iter = m->private;
 634         void *p = NULL;
 635         loff_t l = 0;
 636         int i;
 637
 638         mutex_lock(&trace_types_lock);
 639
 640         if (!current_trace || current_trace != iter->trace)
 641                 return NULL;
 642
 643         atomic_inc(&trace_record_cmdline_disabled);
 644
 645         /* let the tracer grab locks here if needed */
 646         if (current_trace->start)
 647                 current_trace->start(iter);
 648
 649         if (*pos != iter->pos) {
 650                 iter->ent = NULL;
 651                 iter->cpu = 0;
 652                 iter->idx = -1;
 653
 654                 for_each_possible_cpu(i) {
 655                         iter->next_idx[i] = 0;
 656                         iter->next_page[i] = NULL;
 657                 }
 658
 659                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 660                         ;
 661
 662         } else {
 663                 l = *pos - 1;
 664                 p = s_next(m, p, &l);
 665         }
 666
 667         return p;
 668 }
 669
 670 static void s_stop(struct seq_file *m, void *p)
 671 {
 672         struct trace_iterator *iter = m->private;
 673
 674         atomic_dec(&trace_record_cmdline_disabled);
 675
 676         /* let the tracer release locks here if needed */
 677         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 678                 iter->trace->stop(iter);
 679
 680         mutex_unlock(&trace_types_lock);
 681 }
 682
 683 static void
 684 seq_print_sym_short(struct seq_file *m, const char *fmt, unsigned long address)
 685 {
 686 #ifdef CONFIG_KALLSYMS
 687         char str[KSYM_SYMBOL_LEN];
 688
 689         kallsyms_lookup(address, NULL, NULL, NULL, str);
 690
 691         seq_printf(m, fmt, str);
 692 #endif
 693 }
 694
 695 static void
 696 seq_print_sym_offset(struct seq_file *m, const char *fmt, unsigned long address)
 697 {
 698 #ifdef CONFIG_KALLSYMS
 699         char str[KSYM_SYMBOL_LEN];
 700
 701         sprint_symbol(str, address);
 702         seq_printf(m, fmt, str);
 703 #endif
 704 }
 705
 706 #ifndef CONFIG_64BIT
 707 # define IP_FMT "%08lx"
 708 #else
 709 # define IP_FMT "%016lx"
 710 #endif
 711
 712 static void notrace
 713 seq_print_ip_sym(struct seq_file *m, unsigned long ip, unsigned long sym_flags)
 714 {
 715         if (!ip) {
 716                 seq_printf(m, "0");
 717                 return;
 718         }
 719
 720         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 721                 seq_print_sym_offset(m, "%s", ip);
 722         else
 723                 seq_print_sym_short(m, "%s", ip);
 724
 725         if (sym_flags & TRACE_ITER_SYM_ADDR)
 726                 seq_printf(m, " <" IP_FMT ">", ip);
 727 }
 728
 729 static void notrace print_lat_help_header(struct seq_file *m)
 730 {
 731         seq_puts(m, "#                _------=> CPU#            \n");
 732         seq_puts(m, "#               / _-----=> irqs-off        \n");
 733         seq_puts(m, "#              | / _----=> need-resched    \n");
 734         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 735         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 736         seq_puts(m, "#              |||| /                      \n");
 737         seq_puts(m, "#              |||||     delay             \n");
 738         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 739         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 740 }
 741
 742 static void notrace print_func_help_header(struct seq_file *m)
 743 {
 744         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 745         seq_puts(m, "#              | |      |          |         |\n");
 746 }
 747
 748
 749 static void notrace
 750 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 751 {
 752         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 753         struct trace_array *tr = iter->tr;
 754         struct trace_array_cpu *data = tr->data[tr->cpu];
 755         struct tracer *type = current_trace;
 756         unsigned long total   = 0;
 757         unsigned long entries = 0;
 758         int cpu;
 759         const char *name = "preemption";
 760
 761         if (type)
 762                 name = type->name;
 763
 764         for_each_possible_cpu(cpu) {
 765                 if (tr->data[cpu]->trace) {
 766                         total += tr->data[cpu]->trace_idx;
 767                         if (tr->data[cpu]->trace_idx > tr->entries)
 768                                 entries += tr->entries;
 769                         else
 770                                 entries += tr->data[cpu]->trace_idx;
 771                 }
 772         }
 773
 774         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
 775                    name, UTS_RELEASE);
 776         seq_puts(m, "-----------------------------------"
 777                  "---------------------------------\n");
 778         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
 779                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
 780                    nsecs_to_usecs(data->saved_latency),
 781                    entries,
 782                    total,
 783                    tr->cpu,
 784 #if defined(CONFIG_PREEMPT_NONE)
 785                    "server",
 786 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
 787                    "desktop",
 788 #elif defined(CONFIG_PREEMPT_DESKTOP)
 789                    "preempt",
 790 #else
 791                    "unknown",
 792 #endif
 793                    /* These are reserved for later use */
 794                    0, 0, 0, 0);
 795 #ifdef CONFIG_SMP
 796         seq_printf(m, " #P:%d)\n", num_online_cpus());
 797 #else
 798         seq_puts(m, ")\n");
 799 #endif
 800         seq_puts(m, "    -----------------\n");
 801         seq_printf(m, "    | task: %.16s-%d "
 802                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
 803                    data->comm, data->pid, data->uid, data->nice,
 804                    data->policy, data->rt_priority);
 805         seq_puts(m, "    -----------------\n");
 806
 807         if (data->critical_start) {
 808                 seq_puts(m, " => started at: ");
 809                 seq_print_ip_sym(m, data->critical_start, sym_flags);
 810                 seq_puts(m, "\n => ended at:   ");
 811                 seq_print_ip_sym(m, data->critical_end, sym_flags);
 812                 seq_puts(m, "\n");
 813         }
 814
 815         seq_puts(m, "\n");
 816 }
 817
 818 static void notrace
 819 lat_print_generic(struct seq_file *m, struct trace_entry *entry, int cpu)
 820 {
 821         int hardirq, softirq;
 822         char *comm;
 823
 824         comm = trace_find_cmdline(entry->pid);
 825
 826         seq_printf(m, "%8.8s-%-5d ", comm, entry->pid);
 827         seq_printf(m, "%d", cpu);
 828         seq_printf(m, "%c%c",
 829                    (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
 830                    ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
 831
 832         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
 833         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
 834         if (hardirq && softirq)
 835                 seq_putc(m, 'H');
 836         else {
 837                 if (hardirq)
 838                         seq_putc(m, 'h');
 839                 else {
 840                         if (softirq)
 841                                 seq_putc(m, 's');
 842                         else
 843                                 seq_putc(m, '.');
 844                 }
 845         }
 846
 847         if (entry->preempt_count)
 848                 seq_printf(m, "%x", entry->preempt_count);
 849         else
 850                 seq_puts(m, ".");
 851 }
 852
 853 unsigned long preempt_mark_thresh = 100;
 854
 855 static void notrace
 856 lat_print_timestamp(struct seq_file *m, unsigned long long abs_usecs,
 857                     unsigned long rel_usecs)
 858 {
 859         seq_printf(m, " %4lldus", abs_usecs);
 860         if (rel_usecs > preempt_mark_thresh)
 861                 seq_puts(m, "!: ");
 862         else if (rel_usecs > 1)
 863                 seq_puts(m, "+: ");
 864         else
 865                 seq_puts(m, " : ");
 866 }
 867
 868 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
 869
 870 static void notrace
 871 print_lat_fmt(struct seq_file *m, struct trace_iterator *iter,
 872               unsigned int trace_idx, int cpu)
 873 {
 874         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 875         struct trace_entry *next_entry = find_next_entry(iter, NULL);
 876         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
 877         struct trace_entry *entry = iter->ent;
 878         unsigned long abs_usecs;
 879         unsigned long rel_usecs;
 880         char *comm;
 881         int S;
 882
 883         if (!next_entry)
 884                 next_entry = entry;
 885         rel_usecs = ns2usecs(next_entry->t - entry->t);
 886         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
 887
 888         if (verbose) {
 889                 comm = trace_find_cmdline(entry->pid);
 890                 seq_printf(m, "%16s %5d %d %d %08x %08x [%08lx]"
 891                            " %ld.%03ldms (+%ld.%03ldms): ",
 892                            comm,
 893                            entry->pid, cpu, entry->flags,
 894                            entry->preempt_count, trace_idx,
 895                            ns2usecs(entry->t),
 896                            abs_usecs/1000,
 897                            abs_usecs % 1000, rel_usecs/1000, rel_usecs % 1000);
 898         } else {
 899                 lat_print_generic(m, entry, cpu);
 900                 lat_print_timestamp(m, abs_usecs, rel_usecs);
 901         }
 902         switch (entry->type) {
 903         case TRACE_FN:
 904                 seq_print_ip_sym(m, entry->fn.ip, sym_flags);
 905                 seq_puts(m, " (");
 906                 seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
 907                 seq_puts(m, ")\n");
 908                 break;
 909         case TRACE_CTX:
 910                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
 911                         state_to_char[entry->ctx.prev_state] : 'X';
 912                 comm = trace_find_cmdline(entry->ctx.next_pid);
 913                 seq_printf(m, " %d:%d:%c --> %d:%d %s\n",
 914                            entry->ctx.prev_pid,
 915                            entry->ctx.prev_prio,
 916                            S,
 917                            entry->ctx.next_pid,
 918                            entry->ctx.next_prio,
 919                            comm);
 920                 break;
 921         default:
 922                 seq_printf(m, "Unknown type %d\n", entry->type);
 923         }
 924 }
 925
 926 static void notrace
 927 print_trace_fmt(struct seq_file *m, struct trace_iterator *iter)
 928 {
 929         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 930         struct trace_entry *entry = iter->ent;
 931         unsigned long usec_rem;
 932         unsigned long long t;
 933         unsigned long secs;
 934         char *comm;
 935         int S;
 936
 937         comm = trace_find_cmdline(iter->ent->pid);
 938
 939         t = ns2usecs(entry->t);
 940         usec_rem = do_div(t, 1000000ULL);
 941         secs = (unsigned long)t;
 942
 943         seq_printf(m, "%16s-%-5d ", comm, entry->pid);
 944         seq_printf(m, "[%02d] ", iter->cpu);
 945         seq_printf(m, "%5lu.%06lu: ", secs, usec_rem);
 946
 947         switch (entry->type) {
 948         case TRACE_FN:
 949                 seq_print_ip_sym(m, entry->fn.ip, sym_flags);
 950                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
 951                                                 entry->fn.parent_ip) {
 952                         seq_printf(m, " <-");
 953                         seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
 954                 }
 955                 break;
 956         case TRACE_CTX:
 957                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
 958                         state_to_char[entry->ctx.prev_state] : 'X';
 959                 seq_printf(m, " %d:%d:%c ==> %d:%d\n",
 960                            entry->ctx.prev_pid,
 961                            entry->ctx.prev_prio,
 962                            S,
 963                            entry->ctx.next_pid,
 964                            entry->ctx.next_prio);
 965                 break;
 966         }
 967         seq_printf(m, "\n");
 968 }
 969
 970 static int trace_empty(struct trace_iterator *iter)
 971 {
 972         struct trace_array_cpu *data;
 973         int cpu;
 974
 975         for_each_possible_cpu(cpu) {
 976                 data = iter->tr->data[cpu];
 977
 978                 if (data->trace &&
 979                     data->trace_idx)
 980                         return 0;
 981         }
 982         return 1;
 983 }
 984
 985 static int s_show(struct seq_file *m, void *v)
 986 {
 987         struct trace_iterator *iter = v;
 988
 989         if (iter->ent == NULL) {
 990                 if (iter->tr) {
 991                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
 992                         seq_puts(m, "#\n");
 993                 }
 994                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
 995                         /* print nothing if the buffers are empty */
 996                         if (trace_empty(iter))
 997                                 return 0;
 998                         print_trace_header(m, iter);
 999                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1000                                 print_lat_help_header(m);
1001                 } else {
1002                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1003                                 print_func_help_header(m);
1004                 }
1005         } else {
1006                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1007                         print_lat_fmt(m, iter, iter->idx, iter->cpu);
1008                 else
1009                         print_trace_fmt(m, iter);
1010         }
1011
1012         return 0;
1013 }
1014
1015 static struct seq_operations tracer_seq_ops = {
1016         .start = s_start,
1017         .next = s_next,
1018         .stop = s_stop,
1019         .show = s_show,
1020 };
1021
1022 static struct trace_iterator notrace *
1023 __tracing_open(struct inode *inode, struct file *file, int *ret)
1024 {
1025         struct trace_iterator *iter;
1026
1027         if (tracing_disabled) {
1028                 *ret = -ENODEV;
1029                 return NULL;
1030         }
1031
1032         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1033         if (!iter) {
1034                 *ret = -ENOMEM;
1035                 goto out;
1036         }
1037
1038         mutex_lock(&trace_types_lock);
1039         if (current_trace && current_trace->print_max)
1040                 iter->tr = &max_tr;
1041         else
1042                 iter->tr = inode->i_private;
1043         iter->trace = current_trace;
1044         iter->pos = -1;
1045
1046         /* TODO stop tracer */
1047         *ret = seq_open(file, &tracer_seq_ops);
1048         if (!*ret) {
1049                 struct seq_file *m = file->private_data;
1050                 m->private = iter;
1051
1052                 /* stop the trace while dumping */
1053                 if (iter->tr->ctrl)
1054                         tracer_enabled = 0;
1055
1056                 if (iter->trace && iter->trace->open)
1057                         iter->trace->open(iter);
1058         } else {
1059                 kfree(iter);
1060                 iter = NULL;
1061         }
1062         mutex_unlock(&trace_types_lock);
1063
1064  out:
1065         return iter;
1066 }
1067
1068 int tracing_open_generic(struct inode *inode, struct file *filp)
1069 {
1070         if (tracing_disabled)
1071                 return -ENODEV;
1072
1073         filp->private_data = inode->i_private;
1074         return 0;
1075 }
1076
1077 int tracing_release(struct inode *inode, struct file *file)
1078 {
1079         struct seq_file *m = (struct seq_file *)file->private_data;
1080         struct trace_iterator *iter = m->private;
1081
1082         mutex_lock(&trace_types_lock);
1083         if (iter->trace && iter->trace->close)
1084                 iter->trace->close(iter);
1085
1086         /* reenable tracing if it was previously enabled */
1087         if (iter->tr->ctrl)
1088                 tracer_enabled = 1;
1089         mutex_unlock(&trace_types_lock);
1090
1091         seq_release(inode, file);
1092         kfree(iter);
1093         return 0;
1094 }
1095
1096 static int tracing_open(struct inode *inode, struct file *file)
1097 {
1098         int ret;
1099
1100         __tracing_open(inode, file, &ret);
1101
1102         return ret;
1103 }
1104
1105 static int tracing_lt_open(struct inode *inode, struct file *file)
1106 {
1107         struct trace_iterator *iter;
1108         int ret;
1109
1110         iter = __tracing_open(inode, file, &ret);
1111
1112         if (!ret)
1113                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1114
1115         return ret;
1116 }
1117
1118
1119 static void notrace *
1120 t_next(struct seq_file *m, void *v, loff_t *pos)
1121 {
1122         struct tracer *t = m->private;
1123
1124         (*pos)++;
1125
1126         if (t)
1127                 t = t->next;
1128
1129         m->private = t;
1130
1131         return t;
1132 }
1133
1134 static void *t_start(struct seq_file *m, loff_t *pos)
1135 {
1136         struct tracer *t = m->private;
1137         loff_t l = 0;
1138
1139         mutex_lock(&trace_types_lock);
1140         for (; t && l < *pos; t = t_next(m, t, &l))
1141                 ;
1142
1143         return t;
1144 }
1145
1146 static void t_stop(struct seq_file *m, void *p)
1147 {
1148         mutex_unlock(&trace_types_lock);
1149 }
1150
1151 static int t_show(struct seq_file *m, void *v)
1152 {
1153         struct tracer *t = v;
1154
1155         if (!t)
1156                 return 0;
1157
1158         seq_printf(m, "%s", t->name);
1159         if (t->next)
1160                 seq_putc(m, ' ');
1161         else
1162                 seq_putc(m, '\n');
1163
1164         return 0;
1165 }
1166
1167 static struct seq_operations show_traces_seq_ops = {
1168         .start = t_start,
1169         .next = t_next,
1170         .stop = t_stop,
1171         .show = t_show,
1172 };
1173
1174 static int show_traces_open(struct inode *inode, struct file *file)
1175 {
1176         int ret;
1177
1178         if (tracing_disabled)
1179                 return -ENODEV;
1180
1181         ret = seq_open(file, &show_traces_seq_ops);
1182         if (!ret) {
1183                 struct seq_file *m = file->private_data;
1184                 m->private = trace_types;
1185         }
1186
1187         return ret;
1188 }
1189
1190 static struct file_operations tracing_fops = {
1191         .open = tracing_open,
1192         .read = seq_read,
1193         .llseek = seq_lseek,
1194         .release = tracing_release,
1195 };
1196
1197 static struct file_operations tracing_lt_fops = {
1198         .open = tracing_lt_open,
1199         .read = seq_read,
1200         .llseek = seq_lseek,
1201         .release = tracing_release,
1202 };
1203
1204 static struct file_operations show_traces_fops = {
1205         .open = show_traces_open,
1206         .read = seq_read,
1207         .release = seq_release,
1208 };
1209
1210 static ssize_t
1211 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1212                        size_t cnt, loff_t *ppos)
1213 {
1214         char *buf;
1215         int r = 0;
1216         int len = 0;
1217         int i;
1218
1219         /* calulate max size */
1220         for (i = 0; trace_options[i]; i++) {
1221                 len += strlen(trace_options[i]);
1222                 len += 3; /* "no" and space */
1223         }
1224
1225         /* +2 for \n and \0 */
1226         buf = kmalloc(len + 2, GFP_KERNEL);
1227         if (!buf)
1228                 return -ENOMEM;
1229
1230         for (i = 0; trace_options[i]; i++) {
1231                 if (trace_flags & (1 << i))
1232                         r += sprintf(buf + r, "%s ", trace_options[i]);
1233                 else
1234                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1235         }
1236
1237         r += sprintf(buf + r, "\n");
1238         WARN_ON(r >= len + 2);
1239
1240         r = simple_read_from_buffer(ubuf, cnt, ppos,
1241                                     buf, r);
1242
1243         kfree(buf);
1244
1245         return r;
1246 }
1247
1248 static ssize_t
1249 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1250                         size_t cnt, loff_t *ppos)
1251 {
1252         char buf[64];
1253         char *cmp = buf;
1254         int neg = 0;
1255         int i;
1256
1257         if (cnt > 63)
1258                 cnt = 63;
1259
1260         if (copy_from_user(&buf, ubuf, cnt))
1261                 return -EFAULT;
1262
1263         buf[cnt] = 0;
1264
1265         if (strncmp(buf, "no", 2) == 0) {
1266                 neg = 1;
1267                 cmp += 2;
1268         }
1269
1270         for (i = 0; trace_options[i]; i++) {
1271                 int len = strlen(trace_options[i]);
1272
1273                 if (strncmp(cmp, trace_options[i], len) == 0) {
1274                         if (neg)
1275                                 trace_flags &= ~(1 << i);
1276                         else
1277                                 trace_flags |= (1 << i);
1278                         break;
1279                 }
1280         }
1281
1282         filp->f_pos += cnt;
1283
1284         return cnt;
1285 }
1286
1287 static struct file_operations tracing_iter_fops = {
1288         .open = tracing_open_generic,
1289         .read = tracing_iter_ctrl_read,
1290         .write = tracing_iter_ctrl_write,
1291 };
1292
1293 static ssize_t
1294 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1295                   size_t cnt, loff_t *ppos)
1296 {
1297         struct trace_array *tr = filp->private_data;
1298         char buf[64];
1299         int r;
1300
1301         r = sprintf(buf, "%ld\n", tr->ctrl);
1302         return simple_read_from_buffer(ubuf, cnt, ppos,
1303                                        buf, r);
1304 }
1305
1306 static ssize_t
1307 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1308                    size_t cnt, loff_t *ppos)
1309 {
1310         struct trace_array *tr = filp->private_data;
1311         long val;
1312         char buf[64];
1313
1314         if (cnt > 63)
1315                 cnt = 63;
1316
1317         if (copy_from_user(&buf, ubuf, cnt))
1318                 return -EFAULT;
1319
1320         buf[cnt] = 0;
1321
1322         val = simple_strtoul(buf, NULL, 10);
1323
1324         val = !!val;
1325
1326         mutex_lock(&trace_types_lock);
1327         if (tr->ctrl ^ val) {
1328                 if (val)
1329                         tracer_enabled = 1;
1330                 else
1331                         tracer_enabled = 0;
1332
1333                 tr->ctrl = val;
1334
1335                 if (current_trace && current_trace->ctrl_update)
1336                         current_trace->ctrl_update(tr);
1337         }
1338         mutex_unlock(&trace_types_lock);
1339
1340         filp->f_pos += cnt;
1341
1342         return cnt;
1343 }
1344
1345 static ssize_t
1346 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1347                        size_t cnt, loff_t *ppos)
1348 {
1349         char buf[max_tracer_type_len+2];
1350         int r;
1351
1352         mutex_lock(&trace_types_lock);
1353         if (current_trace)
1354                 r = sprintf(buf, "%s\n", current_trace->name);
1355         else
1356                 r = sprintf(buf, "\n");
1357         mutex_unlock(&trace_types_lock);
1358
1359         return simple_read_from_buffer(ubuf, cnt, ppos,
1360                                        buf, r);
1361 }
1362
1363 static ssize_t
1364 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1365                         size_t cnt, loff_t *ppos)
1366 {
1367         struct trace_array *tr = &global_trace;
1368         struct tracer *t;
1369         char buf[max_tracer_type_len+1];
1370         int i;
1371
1372         if (cnt > max_tracer_type_len)
1373                 cnt = max_tracer_type_len;
1374
1375         if (copy_from_user(&buf, ubuf, cnt))
1376                 return -EFAULT;
1377
1378         buf[cnt] = 0;
1379
1380         /* strip ending whitespace. */
1381         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1382                 buf[i] = 0;
1383
1384         mutex_lock(&trace_types_lock);
1385         for (t = trace_types; t; t = t->next) {
1386                 if (strcmp(t->name, buf) == 0)
1387                         break;
1388         }
1389         if (!t || t == current_trace)
1390                 goto out;
1391
1392         if (current_trace && current_trace->reset)
1393                 current_trace->reset(tr);
1394
1395         current_trace = t;
1396         if (t->init)
1397                 t->init(tr);
1398
1399  out:
1400         mutex_unlock(&trace_types_lock);
1401
1402         filp->f_pos += cnt;
1403
1404         return cnt;
1405 }
1406
1407 static ssize_t
1408 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1409                      size_t cnt, loff_t *ppos)
1410 {
1411         unsigned long *ptr = filp->private_data;
1412         char buf[64];
1413         int r;
1414
1415         r = snprintf(buf, 64, "%ld\n",
1416                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1417         if (r > 64)
1418                 r = 64;
1419         return simple_read_from_buffer(ubuf, cnt, ppos,
1420                                        buf, r);
1421 }
1422
1423 static ssize_t
1424 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1425                       size_t cnt, loff_t *ppos)
1426 {
1427         long *ptr = filp->private_data;
1428         long val;
1429         char buf[64];
1430
1431         if (cnt > 63)
1432                 cnt = 63;
1433
1434         if (copy_from_user(&buf, ubuf, cnt))
1435                 return -EFAULT;
1436
1437         buf[cnt] = 0;
1438
1439         val = simple_strtoul(buf, NULL, 10);
1440
1441         *ptr = val * 1000;
1442
1443         return cnt;
1444 }
1445
1446 static struct file_operations tracing_max_lat_fops = {
1447         .open = tracing_open_generic,
1448         .read = tracing_max_lat_read,
1449         .write = tracing_max_lat_write,
1450 };
1451
1452 static struct file_operations tracing_ctrl_fops = {
1453         .open = tracing_open_generic,
1454         .read = tracing_ctrl_read,
1455         .write = tracing_ctrl_write,
1456 };
1457
1458 static struct file_operations set_tracer_fops = {
1459         .open = tracing_open_generic,
1460         .read = tracing_set_trace_read,
1461         .write = tracing_set_trace_write,
1462 };
1463
1464 #ifdef CONFIG_DYNAMIC_FTRACE
1465
1466 static ssize_t
1467 tracing_read_long(struct file *filp, char __user *ubuf,
1468                   size_t cnt, loff_t *ppos)
1469 {
1470         unsigned long *p = filp->private_data;
1471         char buf[64];
1472         int r;
1473
1474         r = sprintf(buf, "%ld\n", *p);
1475         return simple_read_from_buffer(ubuf, cnt, ppos,
1476                                        buf, r);
1477 }
1478
1479 static struct file_operations tracing_read_long_fops = {
1480         .open = tracing_open_generic,
1481         .read = tracing_read_long,
1482 };
1483 #endif
1484
1485 static struct dentry *d_tracer;
1486
1487 struct dentry *tracing_init_dentry(void)
1488 {
1489         static int once;
1490
1491         if (d_tracer)
1492                 return d_tracer;
1493
1494         d_tracer = debugfs_create_dir("tracing", NULL);
1495
1496         if (!d_tracer && !once) {
1497                 once = 1;
1498                 pr_warning("Could not create debugfs directory 'tracing'\n");
1499                 return NULL;
1500         }
1501
1502         return d_tracer;
1503 }
1504
1505 #ifdef CONFIG_FTRACE_SELFTEST
1506 /* Let selftest have access to static functions in this file */
1507 #include "trace_selftest.c"
1508 #endif
1509
1510 static __init void tracer_init_debugfs(void)
1511 {
1512         struct dentry *d_tracer;
1513         struct dentry *entry;
1514
1515         d_tracer = tracing_init_dentry();
1516
1517         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
1518                                     &global_trace, &tracing_ctrl_fops);
1519         if (!entry)
1520                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
1521
1522         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
1523                                     NULL, &tracing_iter_fops);
1524         if (!entry)
1525                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
1526
1527         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
1528                                     &global_trace, &tracing_lt_fops);
1529         if (!entry)
1530                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
1531
1532         entry = debugfs_create_file("trace", 0444, d_tracer,
1533                                     &global_trace, &tracing_fops);
1534         if (!entry)
1535                 pr_warning("Could not create debugfs 'trace' entry\n");
1536
1537         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
1538                                     &global_trace, &show_traces_fops);
1539         if (!entry)
1540                 pr_warning("Could not create debugfs 'trace' entry\n");
1541
1542         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
1543                                     &global_trace, &set_tracer_fops);
1544         if (!entry)
1545                 pr_warning("Could not create debugfs 'trace' entry\n");
1546
1547         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
1548                                     &tracing_max_latency,
1549                                     &tracing_max_lat_fops);
1550         if (!entry)
1551                 pr_warning("Could not create debugfs "
1552                            "'tracing_max_latency' entry\n");
1553
1554         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
1555                                     &tracing_thresh, &tracing_max_lat_fops);
1556         if (!entry)
1557                 pr_warning("Could not create debugfs "
1558                            "'tracing_threash' entry\n");
1559
1560 #ifdef CONFIG_DYNAMIC_FTRACE
1561         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
1562                                     &ftrace_update_tot_cnt,
1563                                     &tracing_read_long_fops);
1564         if (!entry)
1565                 pr_warning("Could not create debugfs "
1566                            "'dyn_ftrace_total_info' entry\n");
1567 #endif
1568 }
1569
1570 /* dummy trace to disable tracing */
1571 static struct tracer no_tracer __read_mostly =
1572 {
1573         .name = "none",
1574 };
1575
1576 static int trace_alloc_page(void)
1577 {
1578         struct trace_array_cpu *data;
1579         void *array;
1580         struct page *page, *tmp;
1581         LIST_HEAD(pages);
1582         int i;
1583
1584         /* first allocate a page for each CPU */
1585         for_each_possible_cpu(i) {
1586                 array = (void *)__get_free_page(GFP_KERNEL);
1587                 if (array == NULL) {
1588                         printk(KERN_ERR "tracer: failed to allocate page"
1589                                "for trace buffer!\n");
1590                         goto free_pages;
1591                 }
1592
1593                 page = virt_to_page(array);
1594                 list_add(&page->lru, &pages);
1595
1596 /* Only allocate if we are actually using the max trace */
1597 #ifdef CONFIG_TRACER_MAX_TRACE
1598                 array = (void *)__get_free_page(GFP_KERNEL);
1599                 if (array == NULL) {
1600                         printk(KERN_ERR "tracer: failed to allocate page"
1601                                "for trace buffer!\n");
1602                         goto free_pages;
1603                 }
1604                 page = virt_to_page(array);
1605                 list_add(&page->lru, &pages);
1606 #endif
1607         }
1608
1609         /* Now that we successfully allocate a page per CPU, add them */
1610         for_each_possible_cpu(i) {
1611                 data = global_trace.data[i];
1612                 page = list_entry(pages.next, struct page, lru);
1613                 list_del(&page->lru);
1614                 list_add_tail(&page->lru, &data->trace_pages);
1615                 ClearPageLRU(page);
1616
1617 #ifdef CONFIG_TRACER_MAX_TRACE
1618                 data = max_tr.data[i];
1619                 page = list_entry(pages.next, struct page, lru);
1620                 list_del(&page->lru);
1621                 list_add_tail(&page->lru, &data->trace_pages);
1622                 SetPageLRU(page);
1623 #endif
1624         }
1625         global_trace.entries += ENTRIES_PER_PAGE;
1626
1627         return 0;
1628
1629  free_pages:
1630         list_for_each_entry_safe(page, tmp, &pages, lru) {
1631                 list_del(&page->lru);
1632                 __free_page(page);
1633         }
1634         return -ENOMEM;
1635 }
1636
1637 __init static int tracer_alloc_buffers(void)
1638 {
1639         struct trace_array_cpu *data;
1640         void *array;
1641         struct page *page;
1642         int pages = 0;
1643         int ret = -ENOMEM;
1644         int i;
1645
1646         /* Allocate the first page for all buffers */
1647         for_each_possible_cpu(i) {
1648                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
1649                 max_tr.data[i] = &per_cpu(max_data, i);
1650
1651                 array = (void *)__get_free_page(GFP_KERNEL);
1652                 if (array == NULL) {
1653                         printk(KERN_ERR "tracer: failed to allocate page"
1654                                "for trace buffer!\n");
1655                         goto free_buffers;
1656                 }
1657                 data->trace = array;
1658
1659                 /* set the array to the list */
1660                 INIT_LIST_HEAD(&data->trace_pages);
1661                 page = virt_to_page(array);
1662                 list_add(&page->lru, &data->trace_pages);
1663                 /* use the LRU flag to differentiate the two buffers */
1664                 ClearPageLRU(page);
1665
1666 /* Only allocate if we are actually using the max trace */
1667 #ifdef CONFIG_TRACER_MAX_TRACE
1668                 array = (void *)__get_free_page(GFP_KERNEL);
1669                 if (array == NULL) {
1670                         printk(KERN_ERR "tracer: failed to allocate page"
1671                                "for trace buffer!\n");
1672                         goto free_buffers;
1673                 }
1674                 max_tr.data[i]->trace = array;
1675
1676                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
1677                 page = virt_to_page(array);
1678                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
1679                 SetPageLRU(page);
1680 #endif
1681         }
1682
1683         /*
1684          * Since we allocate by orders of pages, we may be able to
1685          * round up a bit.
1686          */
1687         global_trace.entries = ENTRIES_PER_PAGE;
1688         pages++;
1689
1690         while (global_trace.entries < trace_nr_entries) {
1691                 if (trace_alloc_page())
1692                         break;
1693                 pages++;
1694         }
1695         max_tr.entries = global_trace.entries;
1696
1697         pr_info("tracer: %d pages allocated for %ld",
1698                 pages, trace_nr_entries);
1699         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
1700         pr_info("   actual entries %ld\n", global_trace.entries);
1701
1702         tracer_init_debugfs();
1703
1704         trace_init_cmdlines();
1705
1706         register_tracer(&no_tracer);
1707         current_trace = &no_tracer;
1708
1709         /* All seems OK, enable tracing */
1710         tracing_disabled = 0;
1711
1712         return 0;
1713
1714  free_buffers:
1715         for (i-- ; i >= 0; i--) {
1716                 struct page *page, *tmp;
1717                 struct trace_array_cpu *data = global_trace.data[i];
1718
1719                 if (data && data->trace) {
1720                         list_for_each_entry_safe(page, tmp,
1721                                                  &data->trace_pages, lru) {
1722                                 list_del(&page->lru);
1723                                 __free_page(page);
1724                         }
1725                         data->trace = NULL;
1726                 }
1727
1728 #ifdef CONFIG_TRACER_MAX_TRACE
1729                 data = max_tr.data[i];
1730                 if (data && data->trace) {
1731                         list_for_each_entry_safe(page, tmp,
1732                                                  &data->trace_pages, lru) {
1733                                 list_del(&page->lru);
1734                                 __free_page(page);
1735                         }
1736                         data->trace = NULL;
1737                 }
1738 #endif
1739         }
1740         return ret;
1741 }
1742
1743 fs_initcall(tracer_alloc_buffers);