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 void check_pages(struct trace_array_cpu *data)
 146 {
 147         struct page *page, *tmp;
 148
 149         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 150         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 151
 152         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 153                 BUG_ON(page->lru.next->prev != &page->lru);
 154                 BUG_ON(page->lru.prev->next != &page->lru);
 155         }
 156 }
 157
 158 void *head_page(struct trace_array_cpu *data)
 159 {
 160         struct page *page;
 161
 162         check_pages(data);
 163         if (list_empty(&data->trace_pages))
 164                 return NULL;
 165
 166         page = list_entry(data->trace_pages.next, struct page, lru);
 167         BUG_ON(&page->lru == &data->trace_pages);
 168
 169         return page_address(page);
 170 }
 171
 172 notrace static void
 173 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 174 {
 175         struct list_head flip_pages;
 176
 177         INIT_LIST_HEAD(&flip_pages);
 178
 179         tr1->trace_current = NULL;
 180         memcpy(&tr1->trace_current_idx, &tr2->trace_current_idx,
 181                 sizeof(struct trace_array_cpu) -
 182                 offsetof(struct trace_array_cpu, trace_current_idx));
 183
 184         check_pages(tr1);
 185         check_pages(tr2);
 186         list_splice_init(&tr1->trace_pages, &flip_pages);
 187         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 188         list_splice_init(&flip_pages, &tr2->trace_pages);
 189         BUG_ON(!list_empty(&flip_pages));
 190         check_pages(tr1);
 191         check_pages(tr2);
 192 }
 193
 194 notrace void
 195 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 196 {
 197         struct trace_array_cpu *data;
 198         int i;
 199
 200         WARN_ON_ONCE(!irqs_disabled());
 201         spin_lock(&ftrace_max_lock);
 202         /* clear out all the previous traces */
 203         for_each_possible_cpu(i) {
 204                 data = tr->data[i];
 205                 flip_trace(max_tr.data[i], data);
 206                 tracing_reset(data);
 207         }
 208
 209         __update_max_tr(tr, tsk, cpu);
 210         spin_unlock(&ftrace_max_lock);
 211 }
 212
 213 /**
 214  * update_max_tr_single - only copy one trace over, and reset the rest
 215  * @tr - tracer
 216  * @tsk - task with the latency
 217  * @cpu - the cpu of the buffer to copy.
 218  */
 219 notrace void
 220 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 221 {
 222         struct trace_array_cpu *data = tr->data[cpu];
 223         int i;
 224
 225         WARN_ON_ONCE(!irqs_disabled());
 226         spin_lock(&ftrace_max_lock);
 227         for_each_possible_cpu(i)
 228                 tracing_reset(max_tr.data[i]);
 229
 230         flip_trace(max_tr.data[cpu], data);
 231
 232         tracing_reset(data);
 233
 234         __update_max_tr(tr, tsk, cpu);
 235         spin_unlock(&ftrace_max_lock);
 236 }
 237
 238 int register_tracer(struct tracer *type)
 239 {
 240         struct tracer *t;
 241         int len;
 242         int ret = 0;
 243
 244         if (!type->name) {
 245                 pr_info("Tracer must have a name\n");
 246                 return -1;
 247         }
 248
 249         mutex_lock(&trace_types_lock);
 250         for (t = trace_types; t; t = t->next) {
 251                 if (strcmp(type->name, t->name) == 0) {
 252                         /* already found */
 253                         pr_info("Trace %s already registered\n",
 254                                 type->name);
 255                         ret = -1;
 256                         goto out;
 257                 }
 258         }
 259
 260 #ifdef CONFIG_FTRACE_STARTUP_TEST
 261         if (type->selftest) {
 262                 struct tracer *saved_tracer = current_trace;
 263                 struct trace_array_cpu *data;
 264                 struct trace_array *tr = &global_trace;
 265                 int saved_ctrl = tr->ctrl;
 266                 int i;
 267                 /*
 268                  * Run a selftest on this tracer.
 269                  * Here we reset the trace buffer, and set the current
 270                  * tracer to be this tracer. The tracer can then run some
 271                  * internal tracing to verify that everything is in order.
 272                  * If we fail, we do not register this tracer.
 273                  */
 274                 for_each_possible_cpu(i) {
 275                         data = tr->data[i];
 276                         if (!head_page(data))
 277                                 continue;
 278                         tracing_reset(data);
 279                 }
 280                 current_trace = type;
 281                 tr->ctrl = 0;
 282                 /* the test is responsible for initializing and enabling */
 283                 pr_info("Testing tracer %s: ", type->name);
 284                 ret = type->selftest(type, tr);
 285                 /* the test is responsible for resetting too */
 286                 current_trace = saved_tracer;
 287                 tr->ctrl = saved_ctrl;
 288                 if (ret) {
 289                         printk(KERN_CONT "FAILED!\n");
 290                         goto out;
 291                 }
 292                 printk(KERN_CONT "PASSED\n");
 293         }
 294 #endif
 295
 296         type->next = trace_types;
 297         trace_types = type;
 298         len = strlen(type->name);
 299         if (len > max_tracer_type_len)
 300                 max_tracer_type_len = len;
 301
 302  out:
 303         mutex_unlock(&trace_types_lock);
 304
 305         return ret;
 306 }
 307
 308 void unregister_tracer(struct tracer *type)
 309 {
 310         struct tracer **t;
 311         int len;
 312
 313         mutex_lock(&trace_types_lock);
 314         for (t = &trace_types; *t; t = &(*t)->next) {
 315                 if (*t == type)
 316                         goto found;
 317         }
 318         pr_info("Trace %s not registered\n", type->name);
 319         goto out;
 320
 321  found:
 322         *t = (*t)->next;
 323         if (strlen(type->name) != max_tracer_type_len)
 324                 goto out;
 325
 326         max_tracer_type_len = 0;
 327         for (t = &trace_types; *t; t = &(*t)->next) {
 328                 len = strlen((*t)->name);
 329                 if (len > max_tracer_type_len)
 330                         max_tracer_type_len = len;
 331         }
 332  out:
 333         mutex_unlock(&trace_types_lock);
 334 }
 335
 336 void notrace tracing_reset(struct trace_array_cpu *data)
 337 {
 338         data->trace_idx = 0;
 339         data->trace_current = head_page(data);
 340         data->trace_current_idx = 0;
 341 }
 342
 343 #ifdef CONFIG_FTRACE
 344 static void notrace
 345 function_trace_call(unsigned long ip, unsigned long parent_ip)
 346 {
 347         struct trace_array *tr = &global_trace;
 348         struct trace_array_cpu *data;
 349         unsigned long flags;
 350         long disabled;
 351         int cpu;
 352
 353         if (unlikely(!tracer_enabled))
 354                 return;
 355
 356         local_irq_save(flags);
 357         cpu = raw_smp_processor_id();
 358         data = tr->data[cpu];
 359         disabled = atomic_inc_return(&data->disabled);
 360
 361         if (likely(disabled == 1))
 362                 ftrace(tr, data, ip, parent_ip, flags);
 363
 364         atomic_dec(&data->disabled);
 365         local_irq_restore(flags);
 366 }
 367
 368 static struct ftrace_ops trace_ops __read_mostly =
 369 {
 370         .func = function_trace_call,
 371 };
 372 #endif
 373
 374 notrace void tracing_start_function_trace(void)
 375 {
 376         register_ftrace_function(&trace_ops);
 377 }
 378
 379 notrace void tracing_stop_function_trace(void)
 380 {
 381         unregister_ftrace_function(&trace_ops);
 382 }
 383
 384 #define SAVED_CMDLINES 128
 385 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 386 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 387 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 388 static int cmdline_idx;
 389 static DEFINE_SPINLOCK(trace_cmdline_lock);
 390 atomic_t trace_record_cmdline_disabled;
 391
 392 static void trace_init_cmdlines(void)
 393 {
 394         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 395         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 396         cmdline_idx = 0;
 397 }
 398
 399 notrace void trace_stop_cmdline_recording(void);
 400
 401 static void notrace trace_save_cmdline(struct task_struct *tsk)
 402 {
 403         unsigned map;
 404         unsigned idx;
 405
 406         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 407                 return;
 408
 409         /*
 410          * It's not the end of the world if we don't get
 411          * the lock, but we also don't want to spin
 412          * nor do we want to disable interrupts,
 413          * so if we miss here, then better luck next time.
 414          */
 415         if (!spin_trylock(&trace_cmdline_lock))
 416                 return;
 417
 418         idx = map_pid_to_cmdline[tsk->pid];
 419         if (idx >= SAVED_CMDLINES) {
 420                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 421
 422                 map = map_cmdline_to_pid[idx];
 423                 if (map <= PID_MAX_DEFAULT)
 424                         map_pid_to_cmdline[map] = (unsigned)-1;
 425
 426                 map_pid_to_cmdline[tsk->pid] = idx;
 427
 428                 cmdline_idx = idx;
 429         }
 430
 431         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 432
 433         spin_unlock(&trace_cmdline_lock);
 434 }
 435
 436 static notrace char *trace_find_cmdline(int pid)
 437 {
 438         char *cmdline = "<...>";
 439         unsigned map;
 440
 441         if (!pid)
 442                 return "<idle>";
 443
 444         if (pid > PID_MAX_DEFAULT)
 445                 goto out;
 446
 447         map = map_pid_to_cmdline[pid];
 448         if (map >= SAVED_CMDLINES)
 449                 goto out;
 450
 451         cmdline = saved_cmdlines[map];
 452
 453  out:
 454         return cmdline;
 455 }
 456
 457 notrace void tracing_record_cmdline(struct task_struct *tsk)
 458 {
 459         if (atomic_read(&trace_record_cmdline_disabled))
 460                 return;
 461
 462         trace_save_cmdline(tsk);
 463 }
 464
 465 static inline notrace struct trace_entry *
 466 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 467 {
 468         unsigned long idx, idx_next;
 469         struct trace_entry *entry;
 470         struct list_head *next;
 471         struct page *page;
 472
 473         data->trace_idx++;
 474         idx = data->trace_current_idx;
 475         idx_next = idx + 1;
 476
 477         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 478
 479         entry = data->trace_current + idx * TRACE_ENTRY_SIZE;
 480
 481         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 482                 page = virt_to_page(data->trace_current);
 483                 /*
 484                  * Roundrobin - but skip the head (which is not a real page):
 485                  */
 486                 next = page->lru.next;
 487                 if (unlikely(next == &data->trace_pages))
 488                         next = next->next;
 489                 BUG_ON(next == &data->trace_pages);
 490
 491                 page = list_entry(next, struct page, lru);
 492                 data->trace_current = page_address(page);
 493                 idx_next = 0;
 494         }
 495
 496         data->trace_current_idx = idx_next;
 497
 498         return entry;
 499 }
 500
 501 static inline notrace void
 502 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 503 {
 504         struct task_struct *tsk = current;
 505         unsigned long pc;
 506
 507         pc = preempt_count();
 508
 509         entry->idx              = atomic_inc_return(&tracer_counter);
 510         entry->preempt_count    = pc & 0xff;
 511         entry->pid              = tsk->pid;
 512         entry->t                = now(raw_smp_processor_id());
 513         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 514                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 515                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 516                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 517 }
 518
 519 notrace void
 520 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 521        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 522 {
 523         struct trace_entry *entry;
 524
 525         entry                   = tracing_get_trace_entry(tr, data);
 526         tracing_generic_entry_update(entry, flags);
 527         entry->type             = TRACE_FN;
 528         entry->fn.ip            = ip;
 529         entry->fn.parent_ip     = parent_ip;
 530 }
 531
 532 notrace void
 533 tracing_sched_switch_trace(struct trace_array *tr,
 534                            struct trace_array_cpu *data,
 535                            struct task_struct *prev, struct task_struct *next,
 536                            unsigned long flags)
 537 {
 538         struct trace_entry *entry;
 539
 540         entry                   = tracing_get_trace_entry(tr, data);
 541         tracing_generic_entry_update(entry, flags);
 542         entry->type             = TRACE_CTX;
 543         entry->ctx.prev_pid     = prev->pid;
 544         entry->ctx.prev_prio    = prev->prio;
 545         entry->ctx.prev_state   = prev->state;
 546         entry->ctx.next_pid     = next->pid;
 547         entry->ctx.next_prio    = next->prio;
 548 }
 549
 550 enum trace_file_type {
 551         TRACE_FILE_LAT_FMT      = 1,
 552 };
 553
 554 static struct trace_entry *
 555 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 556                 struct trace_iterator *iter, int cpu)
 557 {
 558         struct page *page;
 559         struct trace_entry *array;
 560
 561         if (iter->next_idx[cpu] >= tr->entries ||
 562             iter->next_idx[cpu] >= data->trace_idx)
 563                 return NULL;
 564
 565         if (!iter->next_page[cpu]) {
 566                 /*
 567                  * Initialize. If the count of elements in
 568                  * this buffer is greater than the max entries
 569                  * we had an underrun. Which means we looped around.
 570                  * We can simply use the current pointer as our
 571                  * starting point.
 572                  */
 573                 if (data->trace_idx >= tr->entries) {
 574                         page = virt_to_page(data->trace_current);
 575                         iter->next_page[cpu] = &page->lru;
 576                         iter->next_page_idx[cpu] = data->trace_current_idx;
 577                 } else {
 578                         iter->next_page[cpu] = data->trace_pages.next;
 579                         iter->next_page_idx[cpu] = 0;
 580                 }
 581         }
 582
 583         page = list_entry(iter->next_page[cpu], struct page, lru);
 584         BUG_ON(&data->trace_pages == &page->lru);
 585
 586         array = page_address(page);
 587
 588         return &array[iter->next_page_idx[cpu]];
 589 }
 590
 591 static struct notrace trace_entry *
 592 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 593 {
 594         struct trace_array *tr = iter->tr;
 595         struct trace_entry *ent, *next = NULL;
 596         int next_cpu = -1;
 597         int cpu;
 598
 599         for_each_possible_cpu(cpu) {
 600                 if (!head_page(tr->data[cpu]))
 601                         continue;
 602                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 603                 if (ent &&
 604                     (!next || (long)(next->idx - ent->idx) > 0)) {
 605                         next = ent;
 606                         next_cpu = cpu;
 607                 }
 608         }
 609
 610         if (ent_cpu)
 611                 *ent_cpu = next_cpu;
 612
 613         return next;
 614 }
 615
 616 static void *find_next_entry_inc(struct trace_iterator *iter)
 617 {
 618         struct trace_entry *next;
 619         int next_cpu = -1;
 620
 621         next = find_next_entry(iter, &next_cpu);
 622
 623         if (next) {
 624                 iter->idx++;
 625                 iter->next_idx[next_cpu]++;
 626                 iter->next_page_idx[next_cpu]++;
 627                 if (iter->next_page_idx[next_cpu] >= ENTRIES_PER_PAGE) {
 628                         struct trace_array_cpu *data = iter->tr->data[next_cpu];
 629
 630                         iter->next_page_idx[next_cpu] = 0;
 631                         iter->next_page[next_cpu] =
 632                                 iter->next_page[next_cpu]->next;
 633                         if (iter->next_page[next_cpu] == &data->trace_pages)
 634                                 iter->next_page[next_cpu] =
 635                                         data->trace_pages.next;
 636                 }
 637         }
 638         iter->ent = next;
 639         iter->cpu = next_cpu;
 640
 641         return next ? iter : NULL;
 642 }
 643
 644 static void notrace *
 645 s_next(struct seq_file *m, void *v, loff_t *pos)
 646 {
 647         struct trace_iterator *iter = m->private;
 648         void *ent;
 649         void *last_ent = iter->ent;
 650         int i = (int)*pos;
 651
 652         (*pos)++;
 653
 654         /* can't go backwards */
 655         if (iter->idx > i)
 656                 return NULL;
 657
 658         if (iter->idx < 0)
 659                 ent = find_next_entry_inc(iter);
 660         else
 661                 ent = iter;
 662
 663         while (ent && iter->idx < i)
 664                 ent = find_next_entry_inc(iter);
 665
 666         iter->pos = *pos;
 667
 668         if (last_ent && !ent)
 669                 seq_puts(m, "\n\nvim:ft=help\n");
 670
 671         return ent;
 672 }
 673
 674 static void *s_start(struct seq_file *m, loff_t *pos)
 675 {
 676         struct trace_iterator *iter = m->private;
 677         void *p = NULL;
 678         loff_t l = 0;
 679         int i;
 680
 681         mutex_lock(&trace_types_lock);
 682
 683         if (!current_trace || current_trace != iter->trace)
 684                 return NULL;
 685
 686         atomic_inc(&trace_record_cmdline_disabled);
 687
 688         /* let the tracer grab locks here if needed */
 689         if (current_trace->start)
 690                 current_trace->start(iter);
 691
 692         if (*pos != iter->pos) {
 693                 iter->ent = NULL;
 694                 iter->cpu = 0;
 695                 iter->idx = -1;
 696
 697                 for_each_possible_cpu(i) {
 698                         iter->next_idx[i] = 0;
 699                         iter->next_page[i] = NULL;
 700                 }
 701
 702                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 703                         ;
 704
 705         } else {
 706                 l = *pos - 1;
 707                 p = s_next(m, p, &l);
 708         }
 709
 710         return p;
 711 }
 712
 713 static void s_stop(struct seq_file *m, void *p)
 714 {
 715         struct trace_iterator *iter = m->private;
 716
 717         atomic_dec(&trace_record_cmdline_disabled);
 718
 719         /* let the tracer release locks here if needed */
 720         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 721                 iter->trace->stop(iter);
 722
 723         mutex_unlock(&trace_types_lock);
 724 }
 725
 726 static void
 727 seq_print_sym_short(struct seq_file *m, const char *fmt, unsigned long address)
 728 {
 729 #ifdef CONFIG_KALLSYMS
 730         char str[KSYM_SYMBOL_LEN];
 731
 732         kallsyms_lookup(address, NULL, NULL, NULL, str);
 733
 734         seq_printf(m, fmt, str);
 735 #endif
 736 }
 737
 738 static void
 739 seq_print_sym_offset(struct seq_file *m, const char *fmt, unsigned long address)
 740 {
 741 #ifdef CONFIG_KALLSYMS
 742         char str[KSYM_SYMBOL_LEN];
 743
 744         sprint_symbol(str, address);
 745         seq_printf(m, fmt, str);
 746 #endif
 747 }
 748
 749 #ifndef CONFIG_64BIT
 750 # define IP_FMT "%08lx"
 751 #else
 752 # define IP_FMT "%016lx"
 753 #endif
 754
 755 static void notrace
 756 seq_print_ip_sym(struct seq_file *m, unsigned long ip, unsigned long sym_flags)
 757 {
 758         if (!ip) {
 759                 seq_printf(m, "0");
 760                 return;
 761         }
 762
 763         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 764                 seq_print_sym_offset(m, "%s", ip);
 765         else
 766                 seq_print_sym_short(m, "%s", ip);
 767
 768         if (sym_flags & TRACE_ITER_SYM_ADDR)
 769                 seq_printf(m, " <" IP_FMT ">", ip);
 770 }
 771
 772 static void notrace print_lat_help_header(struct seq_file *m)
 773 {
 774         seq_puts(m, "#                _------=> CPU#            \n");
 775         seq_puts(m, "#               / _-----=> irqs-off        \n");
 776         seq_puts(m, "#              | / _----=> need-resched    \n");
 777         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 778         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 779         seq_puts(m, "#              |||| /                      \n");
 780         seq_puts(m, "#              |||||     delay             \n");
 781         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 782         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 783 }
 784
 785 static void notrace print_func_help_header(struct seq_file *m)
 786 {
 787         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 788         seq_puts(m, "#              | |      |          |         |\n");
 789 }
 790
 791
 792 static void notrace
 793 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 794 {
 795         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 796         struct trace_array *tr = iter->tr;
 797         struct trace_array_cpu *data = tr->data[tr->cpu];
 798         struct tracer *type = current_trace;
 799         unsigned long total   = 0;
 800         unsigned long entries = 0;
 801         int cpu;
 802         const char *name = "preemption";
 803
 804         if (type)
 805                 name = type->name;
 806
 807         for_each_possible_cpu(cpu) {
 808                 if (head_page(tr->data[cpu])) {
 809                         total += tr->data[cpu]->trace_idx;
 810                         if (tr->data[cpu]->trace_idx > tr->entries)
 811                                 entries += tr->entries;
 812                         else
 813                                 entries += tr->data[cpu]->trace_idx;
 814                 }
 815         }
 816
 817         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
 818                    name, UTS_RELEASE);
 819         seq_puts(m, "-----------------------------------"
 820                  "---------------------------------\n");
 821         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
 822                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
 823                    nsecs_to_usecs(data->saved_latency),
 824                    entries,
 825                    total,
 826                    tr->cpu,
 827 #if defined(CONFIG_PREEMPT_NONE)
 828                    "server",
 829 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
 830                    "desktop",
 831 #elif defined(CONFIG_PREEMPT_DESKTOP)
 832                    "preempt",
 833 #else
 834                    "unknown",
 835 #endif
 836                    /* These are reserved for later use */
 837                    0, 0, 0, 0);
 838 #ifdef CONFIG_SMP
 839         seq_printf(m, " #P:%d)\n", num_online_cpus());
 840 #else
 841         seq_puts(m, ")\n");
 842 #endif
 843         seq_puts(m, "    -----------------\n");
 844         seq_printf(m, "    | task: %.16s-%d "
 845                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
 846                    data->comm, data->pid, data->uid, data->nice,
 847                    data->policy, data->rt_priority);
 848         seq_puts(m, "    -----------------\n");
 849
 850         if (data->critical_start) {
 851                 seq_puts(m, " => started at: ");
 852                 seq_print_ip_sym(m, data->critical_start, sym_flags);
 853                 seq_puts(m, "\n => ended at:   ");
 854                 seq_print_ip_sym(m, data->critical_end, sym_flags);
 855                 seq_puts(m, "\n");
 856         }
 857
 858         seq_puts(m, "\n");
 859 }
 860
 861 static void notrace
 862 lat_print_generic(struct seq_file *m, struct trace_entry *entry, int cpu)
 863 {
 864         int hardirq, softirq;
 865         char *comm;
 866
 867         comm = trace_find_cmdline(entry->pid);
 868
 869         seq_printf(m, "%8.8s-%-5d ", comm, entry->pid);
 870         seq_printf(m, "%d", cpu);
 871         seq_printf(m, "%c%c",
 872                    (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
 873                    ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
 874
 875         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
 876         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
 877         if (hardirq && softirq)
 878                 seq_putc(m, 'H');
 879         else {
 880                 if (hardirq)
 881                         seq_putc(m, 'h');
 882                 else {
 883                         if (softirq)
 884                                 seq_putc(m, 's');
 885                         else
 886                                 seq_putc(m, '.');
 887                 }
 888         }
 889
 890         if (entry->preempt_count)
 891                 seq_printf(m, "%x", entry->preempt_count);
 892         else
 893                 seq_puts(m, ".");
 894 }
 895
 896 unsigned long preempt_mark_thresh = 100;
 897
 898 static void notrace
 899 lat_print_timestamp(struct seq_file *m, unsigned long long abs_usecs,
 900                     unsigned long rel_usecs)
 901 {
 902         seq_printf(m, " %4lldus", abs_usecs);
 903         if (rel_usecs > preempt_mark_thresh)
 904                 seq_puts(m, "!: ");
 905         else if (rel_usecs > 1)
 906                 seq_puts(m, "+: ");
 907         else
 908                 seq_puts(m, " : ");
 909 }
 910
 911 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
 912
 913 static void notrace
 914 print_lat_fmt(struct seq_file *m, struct trace_iterator *iter,
 915               unsigned int trace_idx, int cpu)
 916 {
 917         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 918         struct trace_entry *next_entry = find_next_entry(iter, NULL);
 919         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
 920         struct trace_entry *entry = iter->ent;
 921         unsigned long abs_usecs;
 922         unsigned long rel_usecs;
 923         char *comm;
 924         int S;
 925
 926         if (!next_entry)
 927                 next_entry = entry;
 928         rel_usecs = ns2usecs(next_entry->t - entry->t);
 929         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
 930
 931         if (verbose) {
 932                 comm = trace_find_cmdline(entry->pid);
 933                 seq_printf(m, "%16s %5d %d %d %08x %08x [%08lx]"
 934                            " %ld.%03ldms (+%ld.%03ldms): ",
 935                            comm,
 936                            entry->pid, cpu, entry->flags,
 937                            entry->preempt_count, trace_idx,
 938                            ns2usecs(entry->t),
 939                            abs_usecs/1000,
 940                            abs_usecs % 1000, rel_usecs/1000, rel_usecs % 1000);
 941         } else {
 942                 lat_print_generic(m, entry, cpu);
 943                 lat_print_timestamp(m, abs_usecs, rel_usecs);
 944         }
 945         switch (entry->type) {
 946         case TRACE_FN:
 947                 seq_print_ip_sym(m, entry->fn.ip, sym_flags);
 948                 seq_puts(m, " (");
 949                 seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
 950                 seq_puts(m, ")\n");
 951                 break;
 952         case TRACE_CTX:
 953                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
 954                         state_to_char[entry->ctx.prev_state] : 'X';
 955                 comm = trace_find_cmdline(entry->ctx.next_pid);
 956                 seq_printf(m, " %d:%d:%c --> %d:%d %s\n",
 957                            entry->ctx.prev_pid,
 958                            entry->ctx.prev_prio,
 959                            S,
 960                            entry->ctx.next_pid,
 961                            entry->ctx.next_prio,
 962                            comm);
 963                 break;
 964         default:
 965                 seq_printf(m, "Unknown type %d\n", entry->type);
 966         }
 967 }
 968
 969 static void notrace
 970 print_trace_fmt(struct seq_file *m, struct trace_iterator *iter)
 971 {
 972         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 973         struct trace_entry *entry = iter->ent;
 974         unsigned long usec_rem;
 975         unsigned long long t;
 976         unsigned long secs;
 977         char *comm;
 978         int S;
 979
 980         comm = trace_find_cmdline(iter->ent->pid);
 981
 982         t = ns2usecs(entry->t);
 983         usec_rem = do_div(t, 1000000ULL);
 984         secs = (unsigned long)t;
 985
 986         seq_printf(m, "%16s-%-5d ", comm, entry->pid);
 987         seq_printf(m, "[%02d] ", iter->cpu);
 988         seq_printf(m, "%5lu.%06lu: ", secs, usec_rem);
 989
 990         switch (entry->type) {
 991         case TRACE_FN:
 992                 seq_print_ip_sym(m, entry->fn.ip, sym_flags);
 993                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
 994                                                 entry->fn.parent_ip) {
 995                         seq_printf(m, " <-");
 996                         seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
 997                 }
 998                 seq_printf(m, "\n");
 999                 break;
1000         case TRACE_CTX:
1001                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1002                         state_to_char[entry->ctx.prev_state] : 'X';
1003                 seq_printf(m, " %d:%d:%c ==> %d:%d\n",
1004                            entry->ctx.prev_pid,
1005                            entry->ctx.prev_prio,
1006                            S,
1007                            entry->ctx.next_pid,
1008                            entry->ctx.next_prio);
1009                 break;
1010         }
1011 }
1012
1013 static int trace_empty(struct trace_iterator *iter)
1014 {
1015         struct trace_array_cpu *data;
1016         int cpu;
1017
1018         for_each_possible_cpu(cpu) {
1019                 data = iter->tr->data[cpu];
1020
1021                 if (head_page(data) && data->trace_idx)
1022                         return 0;
1023         }
1024         return 1;
1025 }
1026
1027 static int s_show(struct seq_file *m, void *v)
1028 {
1029         struct trace_iterator *iter = v;
1030
1031         if (iter->ent == NULL) {
1032                 if (iter->tr) {
1033                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1034                         seq_puts(m, "#\n");
1035                 }
1036                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1037                         /* print nothing if the buffers are empty */
1038                         if (trace_empty(iter))
1039                                 return 0;
1040                         print_trace_header(m, iter);
1041                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1042                                 print_lat_help_header(m);
1043                 } else {
1044                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1045                                 print_func_help_header(m);
1046                 }
1047         } else {
1048                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1049                         print_lat_fmt(m, iter, iter->idx, iter->cpu);
1050                 else
1051                         print_trace_fmt(m, iter);
1052         }
1053
1054         return 0;
1055 }
1056
1057 static struct seq_operations tracer_seq_ops = {
1058         .start = s_start,
1059         .next = s_next,
1060         .stop = s_stop,
1061         .show = s_show,
1062 };
1063
1064 static struct trace_iterator notrace *
1065 __tracing_open(struct inode *inode, struct file *file, int *ret)
1066 {
1067         struct trace_iterator *iter;
1068
1069         if (tracing_disabled) {
1070                 *ret = -ENODEV;
1071                 return NULL;
1072         }
1073
1074         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1075         if (!iter) {
1076                 *ret = -ENOMEM;
1077                 goto out;
1078         }
1079
1080         mutex_lock(&trace_types_lock);
1081         if (current_trace && current_trace->print_max)
1082                 iter->tr = &max_tr;
1083         else
1084                 iter->tr = inode->i_private;
1085         iter->trace = current_trace;
1086         iter->pos = -1;
1087
1088         /* TODO stop tracer */
1089         *ret = seq_open(file, &tracer_seq_ops);
1090         if (!*ret) {
1091                 struct seq_file *m = file->private_data;
1092                 m->private = iter;
1093
1094                 /* stop the trace while dumping */
1095                 if (iter->tr->ctrl)
1096                         tracer_enabled = 0;
1097
1098                 if (iter->trace && iter->trace->open)
1099                         iter->trace->open(iter);
1100         } else {
1101                 kfree(iter);
1102                 iter = NULL;
1103         }
1104         mutex_unlock(&trace_types_lock);
1105
1106  out:
1107         return iter;
1108 }
1109
1110 int tracing_open_generic(struct inode *inode, struct file *filp)
1111 {
1112         if (tracing_disabled)
1113                 return -ENODEV;
1114
1115         filp->private_data = inode->i_private;
1116         return 0;
1117 }
1118
1119 int tracing_release(struct inode *inode, struct file *file)
1120 {
1121         struct seq_file *m = (struct seq_file *)file->private_data;
1122         struct trace_iterator *iter = m->private;
1123
1124         mutex_lock(&trace_types_lock);
1125         if (iter->trace && iter->trace->close)
1126                 iter->trace->close(iter);
1127
1128         /* reenable tracing if it was previously enabled */
1129         if (iter->tr->ctrl)
1130                 tracer_enabled = 1;
1131         mutex_unlock(&trace_types_lock);
1132
1133         seq_release(inode, file);
1134         kfree(iter);
1135         return 0;
1136 }
1137
1138 static int tracing_open(struct inode *inode, struct file *file)
1139 {
1140         int ret;
1141
1142         __tracing_open(inode, file, &ret);
1143
1144         return ret;
1145 }
1146
1147 static int tracing_lt_open(struct inode *inode, struct file *file)
1148 {
1149         struct trace_iterator *iter;
1150         int ret;
1151
1152         iter = __tracing_open(inode, file, &ret);
1153
1154         if (!ret)
1155                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1156
1157         return ret;
1158 }
1159
1160
1161 static void notrace *
1162 t_next(struct seq_file *m, void *v, loff_t *pos)
1163 {
1164         struct tracer *t = m->private;
1165
1166         (*pos)++;
1167
1168         if (t)
1169                 t = t->next;
1170
1171         m->private = t;
1172
1173         return t;
1174 }
1175
1176 static void *t_start(struct seq_file *m, loff_t *pos)
1177 {
1178         struct tracer *t = m->private;
1179         loff_t l = 0;
1180
1181         mutex_lock(&trace_types_lock);
1182         for (; t && l < *pos; t = t_next(m, t, &l))
1183                 ;
1184
1185         return t;
1186 }
1187
1188 static void t_stop(struct seq_file *m, void *p)
1189 {
1190         mutex_unlock(&trace_types_lock);
1191 }
1192
1193 static int t_show(struct seq_file *m, void *v)
1194 {
1195         struct tracer *t = v;
1196
1197         if (!t)
1198                 return 0;
1199
1200         seq_printf(m, "%s", t->name);
1201         if (t->next)
1202                 seq_putc(m, ' ');
1203         else
1204                 seq_putc(m, '\n');
1205
1206         return 0;
1207 }
1208
1209 static struct seq_operations show_traces_seq_ops = {
1210         .start = t_start,
1211         .next = t_next,
1212         .stop = t_stop,
1213         .show = t_show,
1214 };
1215
1216 static int show_traces_open(struct inode *inode, struct file *file)
1217 {
1218         int ret;
1219
1220         if (tracing_disabled)
1221                 return -ENODEV;
1222
1223         ret = seq_open(file, &show_traces_seq_ops);
1224         if (!ret) {
1225                 struct seq_file *m = file->private_data;
1226                 m->private = trace_types;
1227         }
1228
1229         return ret;
1230 }
1231
1232 static struct file_operations tracing_fops = {
1233         .open = tracing_open,
1234         .read = seq_read,
1235         .llseek = seq_lseek,
1236         .release = tracing_release,
1237 };
1238
1239 static struct file_operations tracing_lt_fops = {
1240         .open = tracing_lt_open,
1241         .read = seq_read,
1242         .llseek = seq_lseek,
1243         .release = tracing_release,
1244 };
1245
1246 static struct file_operations show_traces_fops = {
1247         .open = show_traces_open,
1248         .read = seq_read,
1249         .release = seq_release,
1250 };
1251
1252 static ssize_t
1253 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1254                        size_t cnt, loff_t *ppos)
1255 {
1256         char *buf;
1257         int r = 0;
1258         int len = 0;
1259         int i;
1260
1261         /* calulate max size */
1262         for (i = 0; trace_options[i]; i++) {
1263                 len += strlen(trace_options[i]);
1264                 len += 3; /* "no" and space */
1265         }
1266
1267         /* +2 for \n and \0 */
1268         buf = kmalloc(len + 2, GFP_KERNEL);
1269         if (!buf)
1270                 return -ENOMEM;
1271
1272         for (i = 0; trace_options[i]; i++) {
1273                 if (trace_flags & (1 << i))
1274                         r += sprintf(buf + r, "%s ", trace_options[i]);
1275                 else
1276                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1277         }
1278
1279         r += sprintf(buf + r, "\n");
1280         WARN_ON(r >= len + 2);
1281
1282         r = simple_read_from_buffer(ubuf, cnt, ppos,
1283                                     buf, r);
1284
1285         kfree(buf);
1286
1287         return r;
1288 }
1289
1290 static ssize_t
1291 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1292                         size_t cnt, loff_t *ppos)
1293 {
1294         char buf[64];
1295         char *cmp = buf;
1296         int neg = 0;
1297         int i;
1298
1299         if (cnt > 63)
1300                 cnt = 63;
1301
1302         if (copy_from_user(&buf, ubuf, cnt))
1303                 return -EFAULT;
1304
1305         buf[cnt] = 0;
1306
1307         if (strncmp(buf, "no", 2) == 0) {
1308                 neg = 1;
1309                 cmp += 2;
1310         }
1311
1312         for (i = 0; trace_options[i]; i++) {
1313                 int len = strlen(trace_options[i]);
1314
1315                 if (strncmp(cmp, trace_options[i], len) == 0) {
1316                         if (neg)
1317                                 trace_flags &= ~(1 << i);
1318                         else
1319                                 trace_flags |= (1 << i);
1320                         break;
1321                 }
1322         }
1323
1324         filp->f_pos += cnt;
1325
1326         return cnt;
1327 }
1328
1329 static struct file_operations tracing_iter_fops = {
1330         .open = tracing_open_generic,
1331         .read = tracing_iter_ctrl_read,
1332         .write = tracing_iter_ctrl_write,
1333 };
1334
1335 static const char readme_msg[] =
1336         "tracing mini-HOWTO:\n\n"
1337         "# mkdir /debug\n"
1338         "# mount -t debugfs nodev /debug\n\n"
1339         "# cat /debug/tracing/available_tracers\n"
1340         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1341         "# cat /debug/tracing/current_tracer\n"
1342         "none\n"
1343         "# echo sched_switch > /debug/tracing/current_tracer\n"
1344         "# cat /debug/tracing/current_tracer\n"
1345         "sched_switch\n"
1346         "# cat /debug/tracing/iter_ctrl\n"
1347         "noprint-parent nosym-offset nosym-addr noverbose\n"
1348         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1349         "# echo 1 > /debug/tracing/tracing_enabled\n"
1350         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1351         "echo 0 > /debug/tracing/tracing_enabled\n"
1352 ;
1353
1354 static ssize_t
1355 tracing_readme_read(struct file *filp, char __user *ubuf,
1356                        size_t cnt, loff_t *ppos)
1357 {
1358         return simple_read_from_buffer(ubuf, cnt, ppos,
1359                                         readme_msg, strlen(readme_msg));
1360 }
1361
1362 static struct file_operations tracing_readme_fops = {
1363         .open = tracing_open_generic,
1364         .read = tracing_readme_read,
1365 };
1366
1367
1368 static ssize_t
1369 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1370                   size_t cnt, loff_t *ppos)
1371 {
1372         struct trace_array *tr = filp->private_data;
1373         char buf[64];
1374         int r;
1375
1376         r = sprintf(buf, "%ld\n", tr->ctrl);
1377         return simple_read_from_buffer(ubuf, cnt, ppos,
1378                                        buf, r);
1379 }
1380
1381 static ssize_t
1382 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1383                    size_t cnt, loff_t *ppos)
1384 {
1385         struct trace_array *tr = filp->private_data;
1386         long val;
1387         char buf[64];
1388
1389         if (cnt > 63)
1390                 cnt = 63;
1391
1392         if (copy_from_user(&buf, ubuf, cnt))
1393                 return -EFAULT;
1394
1395         buf[cnt] = 0;
1396
1397         val = simple_strtoul(buf, NULL, 10);
1398
1399         val = !!val;
1400
1401         mutex_lock(&trace_types_lock);
1402         if (tr->ctrl ^ val) {
1403                 if (val)
1404                         tracer_enabled = 1;
1405                 else
1406                         tracer_enabled = 0;
1407
1408                 tr->ctrl = val;
1409
1410                 if (current_trace && current_trace->ctrl_update)
1411                         current_trace->ctrl_update(tr);
1412         }
1413         mutex_unlock(&trace_types_lock);
1414
1415         filp->f_pos += cnt;
1416
1417         return cnt;
1418 }
1419
1420 static ssize_t
1421 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1422                        size_t cnt, loff_t *ppos)
1423 {
1424         char buf[max_tracer_type_len+2];
1425         int r;
1426
1427         mutex_lock(&trace_types_lock);
1428         if (current_trace)
1429                 r = sprintf(buf, "%s\n", current_trace->name);
1430         else
1431                 r = sprintf(buf, "\n");
1432         mutex_unlock(&trace_types_lock);
1433
1434         return simple_read_from_buffer(ubuf, cnt, ppos,
1435                                        buf, r);
1436 }
1437
1438 static ssize_t
1439 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1440                         size_t cnt, loff_t *ppos)
1441 {
1442         struct trace_array *tr = &global_trace;
1443         struct tracer *t;
1444         char buf[max_tracer_type_len+1];
1445         int i;
1446
1447         if (cnt > max_tracer_type_len)
1448                 cnt = max_tracer_type_len;
1449
1450         if (copy_from_user(&buf, ubuf, cnt))
1451                 return -EFAULT;
1452
1453         buf[cnt] = 0;
1454
1455         /* strip ending whitespace. */
1456         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1457                 buf[i] = 0;
1458
1459         mutex_lock(&trace_types_lock);
1460         for (t = trace_types; t; t = t->next) {
1461                 if (strcmp(t->name, buf) == 0)
1462                         break;
1463         }
1464         if (!t || t == current_trace)
1465                 goto out;
1466
1467         if (current_trace && current_trace->reset)
1468                 current_trace->reset(tr);
1469
1470         current_trace = t;
1471         if (t->init)
1472                 t->init(tr);
1473
1474  out:
1475         mutex_unlock(&trace_types_lock);
1476
1477         filp->f_pos += cnt;
1478
1479         return cnt;
1480 }
1481
1482 static ssize_t
1483 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1484                      size_t cnt, loff_t *ppos)
1485 {
1486         unsigned long *ptr = filp->private_data;
1487         char buf[64];
1488         int r;
1489
1490         r = snprintf(buf, 64, "%ld\n",
1491                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1492         if (r > 64)
1493                 r = 64;
1494         return simple_read_from_buffer(ubuf, cnt, ppos,
1495                                        buf, r);
1496 }
1497
1498 static ssize_t
1499 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1500                       size_t cnt, loff_t *ppos)
1501 {
1502         long *ptr = filp->private_data;
1503         long val;
1504         char buf[64];
1505
1506         if (cnt > 63)
1507                 cnt = 63;
1508
1509         if (copy_from_user(&buf, ubuf, cnt))
1510                 return -EFAULT;
1511
1512         buf[cnt] = 0;
1513
1514         val = simple_strtoul(buf, NULL, 10);
1515
1516         *ptr = val * 1000;
1517
1518         return cnt;
1519 }
1520
1521 static struct file_operations tracing_max_lat_fops = {
1522         .open = tracing_open_generic,
1523         .read = tracing_max_lat_read,
1524         .write = tracing_max_lat_write,
1525 };
1526
1527 static struct file_operations tracing_ctrl_fops = {
1528         .open = tracing_open_generic,
1529         .read = tracing_ctrl_read,
1530         .write = tracing_ctrl_write,
1531 };
1532
1533 static struct file_operations set_tracer_fops = {
1534         .open = tracing_open_generic,
1535         .read = tracing_set_trace_read,
1536         .write = tracing_set_trace_write,
1537 };
1538
1539 #ifdef CONFIG_DYNAMIC_FTRACE
1540
1541 static ssize_t
1542 tracing_read_long(struct file *filp, char __user *ubuf,
1543                   size_t cnt, loff_t *ppos)
1544 {
1545         unsigned long *p = filp->private_data;
1546         char buf[64];
1547         int r;
1548
1549         r = sprintf(buf, "%ld\n", *p);
1550         return simple_read_from_buffer(ubuf, cnt, ppos,
1551                                        buf, r);
1552 }
1553
1554 static struct file_operations tracing_read_long_fops = {
1555         .open = tracing_open_generic,
1556         .read = tracing_read_long,
1557 };
1558 #endif
1559
1560 static struct dentry *d_tracer;
1561
1562 struct dentry *tracing_init_dentry(void)
1563 {
1564         static int once;
1565
1566         if (d_tracer)
1567                 return d_tracer;
1568
1569         d_tracer = debugfs_create_dir("tracing", NULL);
1570
1571         if (!d_tracer && !once) {
1572                 once = 1;
1573                 pr_warning("Could not create debugfs directory 'tracing'\n");
1574                 return NULL;
1575         }
1576
1577         return d_tracer;
1578 }
1579
1580 #ifdef CONFIG_FTRACE_SELFTEST
1581 /* Let selftest have access to static functions in this file */
1582 #include "trace_selftest.c"
1583 #endif
1584
1585 static __init void tracer_init_debugfs(void)
1586 {
1587         struct dentry *d_tracer;
1588         struct dentry *entry;
1589
1590         d_tracer = tracing_init_dentry();
1591
1592         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
1593                                     &global_trace, &tracing_ctrl_fops);
1594         if (!entry)
1595                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
1596
1597         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
1598                                     NULL, &tracing_iter_fops);
1599         if (!entry)
1600                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
1601
1602         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
1603                                     &global_trace, &tracing_lt_fops);
1604         if (!entry)
1605                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
1606
1607         entry = debugfs_create_file("trace", 0444, d_tracer,
1608                                     &global_trace, &tracing_fops);
1609         if (!entry)
1610                 pr_warning("Could not create debugfs 'trace' entry\n");
1611
1612         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
1613                                     &global_trace, &show_traces_fops);
1614         if (!entry)
1615                 pr_warning("Could not create debugfs 'trace' entry\n");
1616
1617         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
1618                                     &global_trace, &set_tracer_fops);
1619         if (!entry)
1620                 pr_warning("Could not create debugfs 'trace' entry\n");
1621
1622         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
1623                                     &tracing_max_latency,
1624                                     &tracing_max_lat_fops);
1625         if (!entry)
1626                 pr_warning("Could not create debugfs "
1627                            "'tracing_max_latency' entry\n");
1628
1629         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
1630                                     &tracing_thresh, &tracing_max_lat_fops);
1631         if (!entry)
1632                 pr_warning("Could not create debugfs "
1633                            "'tracing_threash' entry\n");
1634         entry = debugfs_create_file("README", 0644, d_tracer,
1635                                     NULL, &tracing_readme_fops);
1636         if (!entry)
1637                 pr_warning("Could not create debugfs 'README' entry\n");
1638
1639
1640 #ifdef CONFIG_DYNAMIC_FTRACE
1641         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
1642                                     &ftrace_update_tot_cnt,
1643                                     &tracing_read_long_fops);
1644         if (!entry)
1645                 pr_warning("Could not create debugfs "
1646                            "'dyn_ftrace_total_info' entry\n");
1647 #endif
1648 }
1649
1650 /* dummy trace to disable tracing */
1651 static struct tracer no_tracer __read_mostly =
1652 {
1653         .name = "none",
1654 };
1655
1656 static int trace_alloc_page(void)
1657 {
1658         struct trace_array_cpu *data;
1659         struct page *page, *tmp;
1660         LIST_HEAD(pages);
1661         void *array;
1662         int i;
1663
1664         /* first allocate a page for each CPU */
1665         for_each_possible_cpu(i) {
1666                 array = (void *)__get_free_page(GFP_KERNEL);
1667                 if (array == NULL) {
1668                         printk(KERN_ERR "tracer: failed to allocate page"
1669                                "for trace buffer!\n");
1670                         goto free_pages;
1671                 }
1672
1673                 page = virt_to_page(array);
1674                 list_add(&page->lru, &pages);
1675
1676 /* Only allocate if we are actually using the max trace */
1677 #ifdef CONFIG_TRACER_MAX_TRACE
1678                 array = (void *)__get_free_page(GFP_KERNEL);
1679                 if (array == NULL) {
1680                         printk(KERN_ERR "tracer: failed to allocate page"
1681                                "for trace buffer!\n");
1682                         goto free_pages;
1683                 }
1684                 page = virt_to_page(array);
1685                 list_add(&page->lru, &pages);
1686 #endif
1687         }
1688
1689         /* Now that we successfully allocate a page per CPU, add them */
1690         for_each_possible_cpu(i) {
1691                 data = global_trace.data[i];
1692                 page = list_entry(pages.next, struct page, lru);
1693                 list_del_init(&page->lru);
1694                 list_add_tail(&page->lru, &data->trace_pages);
1695                 ClearPageLRU(page);
1696
1697 #ifdef CONFIG_TRACER_MAX_TRACE
1698                 data = max_tr.data[i];
1699                 page = list_entry(pages.next, struct page, lru);
1700                 list_del_init(&page->lru);
1701                 list_add_tail(&page->lru, &data->trace_pages);
1702                 SetPageLRU(page);
1703 #endif
1704         }
1705         global_trace.entries += ENTRIES_PER_PAGE;
1706
1707         return 0;
1708
1709  free_pages:
1710         list_for_each_entry_safe(page, tmp, &pages, lru) {
1711                 list_del_init(&page->lru);
1712                 __free_page(page);
1713         }
1714         return -ENOMEM;
1715 }
1716
1717 __init static int tracer_alloc_buffers(void)
1718 {
1719         struct trace_array_cpu *data;
1720         void *array;
1721         struct page *page;
1722         int pages = 0;
1723         int ret = -ENOMEM;
1724         int i;
1725
1726         /* Allocate the first page for all buffers */
1727         for_each_possible_cpu(i) {
1728                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
1729                 max_tr.data[i] = &per_cpu(max_data, i);
1730
1731                 array = (void *)__get_free_page(GFP_KERNEL);
1732                 if (array == NULL) {
1733                         printk(KERN_ERR "tracer: failed to allocate page"
1734                                "for trace buffer!\n");
1735                         goto free_buffers;
1736                 }
1737
1738                 /* set the array to the list */
1739                 INIT_LIST_HEAD(&data->trace_pages);
1740                 page = virt_to_page(array);
1741                 list_add(&page->lru, &data->trace_pages);
1742                 /* use the LRU flag to differentiate the two buffers */
1743                 ClearPageLRU(page);
1744
1745 /* Only allocate if we are actually using the max trace */
1746 #ifdef CONFIG_TRACER_MAX_TRACE
1747                 array = (void *)__get_free_page(GFP_KERNEL);
1748                 if (array == NULL) {
1749                         printk(KERN_ERR "tracer: failed to allocate page"
1750                                "for trace buffer!\n");
1751                         goto free_buffers;
1752                 }
1753
1754                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
1755                 page = virt_to_page(array);
1756                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
1757                 SetPageLRU(page);
1758 #endif
1759         }
1760
1761         /*
1762          * Since we allocate by orders of pages, we may be able to
1763          * round up a bit.
1764          */
1765         global_trace.entries = ENTRIES_PER_PAGE;
1766         pages++;
1767
1768         while (global_trace.entries < trace_nr_entries) {
1769                 if (trace_alloc_page())
1770                         break;
1771                 pages++;
1772         }
1773         max_tr.entries = global_trace.entries;
1774
1775         pr_info("tracer: %d pages allocated for %ld",
1776                 pages, trace_nr_entries);
1777         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
1778         pr_info("   actual entries %ld\n", global_trace.entries);
1779
1780         tracer_init_debugfs();
1781
1782         trace_init_cmdlines();
1783
1784         register_tracer(&no_tracer);
1785         current_trace = &no_tracer;
1786
1787         /* All seems OK, enable tracing */
1788         tracing_disabled = 0;
1789
1790         return 0;
1791
1792  free_buffers:
1793         for (i-- ; i >= 0; i--) {
1794                 struct page *page, *tmp;
1795                 struct trace_array_cpu *data = global_trace.data[i];
1796
1797                 if (data) {
1798                         list_for_each_entry_safe(page, tmp,
1799                                                  &data->trace_pages, lru) {
1800                                 list_del_init(&page->lru);
1801                                 __free_page(page);
1802                         }
1803                 }
1804
1805 #ifdef CONFIG_TRACER_MAX_TRACE
1806                 data = max_tr.data[i];
1807                 if (data) {
1808                         list_for_each_entry_safe(page, tmp,
1809                                                  &data->trace_pages, lru) {
1810                                 list_del_init(&page->lru);
1811                                 __free_page(page);
1812                         }
1813                 }
1814 #endif
1815         }
1816         return ret;
1817 }
1818
1819 fs_initcall(tracer_alloc_buffers);