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 notrace void
 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 notrace void 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         data->time_offset = 0;
 342 }
 343
 344 #ifdef CONFIG_FTRACE
 345 static notrace void
 346 function_trace_call(unsigned long ip, unsigned long parent_ip)
 347 {
 348         struct trace_array *tr = &global_trace;
 349         struct trace_array_cpu *data;
 350         unsigned long flags;
 351         long disabled;
 352         int cpu;
 353
 354         if (unlikely(!tracer_enabled))
 355                 return;
 356
 357         local_irq_save(flags);
 358         cpu = raw_smp_processor_id();
 359         data = tr->data[cpu];
 360         disabled = atomic_inc_return(&data->disabled);
 361
 362         if (likely(disabled == 1))
 363                 ftrace(tr, data, ip, parent_ip, flags);
 364
 365         atomic_dec(&data->disabled);
 366         local_irq_restore(flags);
 367 }
 368
 369 static struct ftrace_ops trace_ops __read_mostly =
 370 {
 371         .func = function_trace_call,
 372 };
 373 #endif
 374
 375 notrace void tracing_start_function_trace(void)
 376 {
 377         register_ftrace_function(&trace_ops);
 378 }
 379
 380 notrace void tracing_stop_function_trace(void)
 381 {
 382         unregister_ftrace_function(&trace_ops);
 383 }
 384
 385 #define SAVED_CMDLINES 128
 386 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 387 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 388 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 389 static int cmdline_idx;
 390 static DEFINE_SPINLOCK(trace_cmdline_lock);
 391 atomic_t trace_record_cmdline_disabled;
 392
 393 static void trace_init_cmdlines(void)
 394 {
 395         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 396         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 397         cmdline_idx = 0;
 398 }
 399
 400 notrace void trace_stop_cmdline_recording(void);
 401
 402 static notrace void trace_save_cmdline(struct task_struct *tsk)
 403 {
 404         unsigned map;
 405         unsigned idx;
 406
 407         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 408                 return;
 409
 410         /*
 411          * It's not the end of the world if we don't get
 412          * the lock, but we also don't want to spin
 413          * nor do we want to disable interrupts,
 414          * so if we miss here, then better luck next time.
 415          */
 416         if (!spin_trylock(&trace_cmdline_lock))
 417                 return;
 418
 419         idx = map_pid_to_cmdline[tsk->pid];
 420         if (idx >= SAVED_CMDLINES) {
 421                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 422
 423                 map = map_cmdline_to_pid[idx];
 424                 if (map <= PID_MAX_DEFAULT)
 425                         map_pid_to_cmdline[map] = (unsigned)-1;
 426
 427                 map_pid_to_cmdline[tsk->pid] = idx;
 428
 429                 cmdline_idx = idx;
 430         }
 431
 432         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 433
 434         spin_unlock(&trace_cmdline_lock);
 435 }
 436
 437 static notrace char *trace_find_cmdline(int pid)
 438 {
 439         char *cmdline = "<...>";
 440         unsigned map;
 441
 442         if (!pid)
 443                 return "<idle>";
 444
 445         if (pid > PID_MAX_DEFAULT)
 446                 goto out;
 447
 448         map = map_pid_to_cmdline[pid];
 449         if (map >= SAVED_CMDLINES)
 450                 goto out;
 451
 452         cmdline = saved_cmdlines[map];
 453
 454  out:
 455         return cmdline;
 456 }
 457
 458 notrace void tracing_record_cmdline(struct task_struct *tsk)
 459 {
 460         if (atomic_read(&trace_record_cmdline_disabled))
 461                 return;
 462
 463         trace_save_cmdline(tsk);
 464 }
 465
 466 static inline notrace struct trace_entry *
 467 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 468 {
 469         unsigned long idx, idx_next;
 470         struct trace_entry *entry;
 471         struct list_head *next;
 472         struct page *page;
 473
 474         data->trace_idx++;
 475         idx = data->trace_current_idx;
 476         idx_next = idx + 1;
 477
 478         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 479
 480         entry = data->trace_current + idx * TRACE_ENTRY_SIZE;
 481
 482         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 483                 page = virt_to_page(data->trace_current);
 484                 /*
 485                  * Roundrobin - but skip the head (which is not a real page):
 486                  */
 487                 next = page->lru.next;
 488                 if (unlikely(next == &data->trace_pages))
 489                         next = next->next;
 490                 BUG_ON(next == &data->trace_pages);
 491
 492                 page = list_entry(next, struct page, lru);
 493                 data->trace_current = page_address(page);
 494                 idx_next = 0;
 495         }
 496
 497         data->trace_current_idx = idx_next;
 498
 499         return entry;
 500 }
 501
 502 static inline notrace void
 503 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 504 {
 505         struct task_struct *tsk = current;
 506         unsigned long pc;
 507
 508         pc = preempt_count();
 509
 510         entry->idx              = atomic_inc_return(&tracer_counter);
 511         entry->preempt_count    = pc & 0xff;
 512         entry->pid              = tsk->pid;
 513         entry->t                = now(raw_smp_processor_id());
 514         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 515                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 516                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 517                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 518 }
 519
 520 notrace void
 521 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 522        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 523 {
 524         struct trace_entry *entry;
 525
 526         entry                   = tracing_get_trace_entry(tr, data);
 527         tracing_generic_entry_update(entry, flags);
 528         entry->type             = TRACE_FN;
 529         entry->fn.ip            = ip;
 530         entry->fn.parent_ip     = parent_ip;
 531 }
 532
 533 notrace void
 534 tracing_sched_switch_trace(struct trace_array *tr,
 535                            struct trace_array_cpu *data,
 536                            struct task_struct *prev, struct task_struct *next,
 537                            unsigned long flags)
 538 {
 539         struct trace_entry *entry;
 540
 541         entry                   = tracing_get_trace_entry(tr, data);
 542         tracing_generic_entry_update(entry, flags);
 543         entry->type             = TRACE_CTX;
 544         entry->ctx.prev_pid     = prev->pid;
 545         entry->ctx.prev_prio    = prev->prio;
 546         entry->ctx.prev_state   = prev->state;
 547         entry->ctx.next_pid     = next->pid;
 548         entry->ctx.next_prio    = next->prio;
 549 }
 550
 551 enum trace_file_type {
 552         TRACE_FILE_LAT_FMT      = 1,
 553 };
 554
 555 static struct trace_entry *
 556 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 557                 struct trace_iterator *iter, int cpu)
 558 {
 559         struct page *page;
 560         struct trace_entry *array;
 561
 562         if (iter->next_idx[cpu] >= tr->entries ||
 563             iter->next_idx[cpu] >= data->trace_idx)
 564                 return NULL;
 565
 566         if (!iter->next_page[cpu]) {
 567                 /*
 568                  * Initialize. If the count of elements in
 569                  * this buffer is greater than the max entries
 570                  * we had an underrun. Which means we looped around.
 571                  * We can simply use the current pointer as our
 572                  * starting point.
 573                  */
 574                 if (data->trace_idx >= tr->entries) {
 575                         page = virt_to_page(data->trace_current);
 576                         iter->next_page[cpu] = &page->lru;
 577                         iter->next_page_idx[cpu] = data->trace_current_idx;
 578                 } else {
 579                         iter->next_page[cpu] = data->trace_pages.next;
 580                         iter->next_page_idx[cpu] = 0;
 581                 }
 582         }
 583
 584         page = list_entry(iter->next_page[cpu], struct page, lru);
 585         BUG_ON(&data->trace_pages == &page->lru);
 586
 587         array = page_address(page);
 588
 589         return &array[iter->next_page_idx[cpu]];
 590 }
 591
 592 static struct notrace trace_entry *
 593 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 594 {
 595         struct trace_array *tr = iter->tr;
 596         struct trace_entry *ent, *next = NULL;
 597         int next_cpu = -1;
 598         int cpu;
 599
 600         for_each_possible_cpu(cpu) {
 601                 if (!head_page(tr->data[cpu]))
 602                         continue;
 603                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 604                 if (ent &&
 605                     (!next || (long)(next->idx - ent->idx) > 0)) {
 606                         next = ent;
 607                         next_cpu = cpu;
 608                 }
 609         }
 610
 611         if (ent_cpu)
 612                 *ent_cpu = next_cpu;
 613
 614         return next;
 615 }
 616
 617 static void *find_next_entry_inc(struct trace_iterator *iter)
 618 {
 619         struct trace_entry *next;
 620         int next_cpu = -1;
 621
 622         next = find_next_entry(iter, &next_cpu);
 623
 624         if (next) {
 625                 iter->idx++;
 626                 iter->next_idx[next_cpu]++;
 627                 iter->next_page_idx[next_cpu]++;
 628
 629                 if (iter->next_page_idx[next_cpu] >= ENTRIES_PER_PAGE) {
 630                         struct trace_array_cpu *data = iter->tr->data[next_cpu];
 631
 632                         iter->next_page_idx[next_cpu] = 0;
 633                         iter->next_page[next_cpu] =
 634                                 iter->next_page[next_cpu]->next;
 635                         if (iter->next_page[next_cpu] == &data->trace_pages)
 636                                 iter->next_page[next_cpu] =
 637                                         data->trace_pages.next;
 638                 }
 639         }
 640         iter->prev_ent = iter->ent;
 641         iter->prev_cpu = iter->cpu;
 642
 643         iter->ent = next;
 644         iter->cpu = next_cpu;
 645
 646         return next ? iter : NULL;
 647 }
 648
 649 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 650 {
 651         struct trace_iterator *iter = m->private;
 652         void *last_ent = iter->ent;
 653         int i = (int)*pos;
 654         void *ent;
 655
 656         (*pos)++;
 657
 658         /* can't go backwards */
 659         if (iter->idx > i)
 660                 return NULL;
 661
 662         if (iter->idx < 0)
 663                 ent = find_next_entry_inc(iter);
 664         else
 665                 ent = iter;
 666
 667         while (ent && iter->idx < i)
 668                 ent = find_next_entry_inc(iter);
 669
 670         iter->pos = *pos;
 671
 672         if (last_ent && !ent)
 673                 seq_puts(m, "\n\nvim:ft=help\n");
 674
 675         return ent;
 676 }
 677
 678 static void *s_start(struct seq_file *m, loff_t *pos)
 679 {
 680         struct trace_iterator *iter = m->private;
 681         void *p = NULL;
 682         loff_t l = 0;
 683         int i;
 684
 685         mutex_lock(&trace_types_lock);
 686
 687         if (!current_trace || current_trace != iter->trace)
 688                 return NULL;
 689
 690         atomic_inc(&trace_record_cmdline_disabled);
 691
 692         /* let the tracer grab locks here if needed */
 693         if (current_trace->start)
 694                 current_trace->start(iter);
 695
 696         if (*pos != iter->pos) {
 697                 iter->ent = NULL;
 698                 iter->cpu = 0;
 699                 iter->idx = -1;
 700                 iter->prev_ent = NULL;
 701                 iter->prev_cpu = -1;
 702
 703                 for_each_possible_cpu(i) {
 704                         iter->next_idx[i] = 0;
 705                         iter->next_page[i] = NULL;
 706                 }
 707
 708                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 709                         ;
 710
 711         } else {
 712                 l = *pos - 1;
 713                 p = s_next(m, p, &l);
 714         }
 715
 716         return p;
 717 }
 718
 719 static void s_stop(struct seq_file *m, void *p)
 720 {
 721         struct trace_iterator *iter = m->private;
 722
 723         atomic_dec(&trace_record_cmdline_disabled);
 724
 725         /* let the tracer release locks here if needed */
 726         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 727                 iter->trace->stop(iter);
 728
 729         mutex_unlock(&trace_types_lock);
 730 }
 731
 732 static void
 733 seq_print_sym_short(struct seq_file *m, const char *fmt, unsigned long address)
 734 {
 735 #ifdef CONFIG_KALLSYMS
 736         char str[KSYM_SYMBOL_LEN];
 737
 738         kallsyms_lookup(address, NULL, NULL, NULL, str);
 739
 740         seq_printf(m, fmt, str);
 741 #endif
 742 }
 743
 744 static void
 745 seq_print_sym_offset(struct seq_file *m, const char *fmt, unsigned long address)
 746 {
 747 #ifdef CONFIG_KALLSYMS
 748         char str[KSYM_SYMBOL_LEN];
 749
 750         sprint_symbol(str, address);
 751         seq_printf(m, fmt, str);
 752 #endif
 753 }
 754
 755 #ifndef CONFIG_64BIT
 756 # define IP_FMT "%08lx"
 757 #else
 758 # define IP_FMT "%016lx"
 759 #endif
 760
 761 static notrace void
 762 seq_print_ip_sym(struct seq_file *m, unsigned long ip, unsigned long sym_flags)
 763 {
 764         if (!ip) {
 765                 seq_printf(m, "0");
 766                 return;
 767         }
 768
 769         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 770                 seq_print_sym_offset(m, "%s", ip);
 771         else
 772                 seq_print_sym_short(m, "%s", ip);
 773
 774         if (sym_flags & TRACE_ITER_SYM_ADDR)
 775                 seq_printf(m, " <" IP_FMT ">", ip);
 776 }
 777
 778 static notrace void print_lat_help_header(struct seq_file *m)
 779 {
 780         seq_puts(m, "#                _------=> CPU#            \n");
 781         seq_puts(m, "#               / _-----=> irqs-off        \n");
 782         seq_puts(m, "#              | / _----=> need-resched    \n");
 783         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 784         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 785         seq_puts(m, "#              |||| /                      \n");
 786         seq_puts(m, "#              |||||     delay             \n");
 787         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 788         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 789 }
 790
 791 static notrace void print_func_help_header(struct seq_file *m)
 792 {
 793         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 794         seq_puts(m, "#              | |      |          |         |\n");
 795 }
 796
 797
 798 static notrace void
 799 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 800 {
 801         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 802         struct trace_array *tr = iter->tr;
 803         struct trace_array_cpu *data = tr->data[tr->cpu];
 804         struct tracer *type = current_trace;
 805         unsigned long total   = 0;
 806         unsigned long entries = 0;
 807         int cpu;
 808         const char *name = "preemption";
 809
 810         if (type)
 811                 name = type->name;
 812
 813         for_each_possible_cpu(cpu) {
 814                 if (head_page(tr->data[cpu])) {
 815                         total += tr->data[cpu]->trace_idx;
 816                         if (tr->data[cpu]->trace_idx > tr->entries)
 817                                 entries += tr->entries;
 818                         else
 819                                 entries += tr->data[cpu]->trace_idx;
 820                 }
 821         }
 822
 823         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
 824                    name, UTS_RELEASE);
 825         seq_puts(m, "-----------------------------------"
 826                  "---------------------------------\n");
 827         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
 828                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
 829                    nsecs_to_usecs(data->saved_latency),
 830                    entries,
 831                    total,
 832                    tr->cpu,
 833 #if defined(CONFIG_PREEMPT_NONE)
 834                    "server",
 835 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
 836                    "desktop",
 837 #elif defined(CONFIG_PREEMPT_DESKTOP)
 838                    "preempt",
 839 #else
 840                    "unknown",
 841 #endif
 842                    /* These are reserved for later use */
 843                    0, 0, 0, 0);
 844 #ifdef CONFIG_SMP
 845         seq_printf(m, " #P:%d)\n", num_online_cpus());
 846 #else
 847         seq_puts(m, ")\n");
 848 #endif
 849         seq_puts(m, "    -----------------\n");
 850         seq_printf(m, "    | task: %.16s-%d "
 851                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
 852                    data->comm, data->pid, data->uid, data->nice,
 853                    data->policy, data->rt_priority);
 854         seq_puts(m, "    -----------------\n");
 855
 856         if (data->critical_start) {
 857                 seq_puts(m, " => started at: ");
 858                 seq_print_ip_sym(m, data->critical_start, sym_flags);
 859                 seq_puts(m, "\n => ended at:   ");
 860                 seq_print_ip_sym(m, data->critical_end, sym_flags);
 861                 seq_puts(m, "\n");
 862         }
 863
 864         seq_puts(m, "\n");
 865 }
 866
 867 static notrace void
 868 lat_print_generic(struct seq_file *m, struct trace_entry *entry, int cpu)
 869 {
 870         int hardirq, softirq;
 871         char *comm;
 872
 873         comm = trace_find_cmdline(entry->pid);
 874
 875         seq_printf(m, "%8.8s-%-5d ", comm, entry->pid);
 876         seq_printf(m, "%d", cpu);
 877         seq_printf(m, "%c%c",
 878                    (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
 879                    ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
 880
 881         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
 882         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
 883         if (hardirq && softirq)
 884                 seq_putc(m, 'H');
 885         else {
 886                 if (hardirq)
 887                         seq_putc(m, 'h');
 888                 else {
 889                         if (softirq)
 890                                 seq_putc(m, 's');
 891                         else
 892                                 seq_putc(m, '.');
 893                 }
 894         }
 895
 896         if (entry->preempt_count)
 897                 seq_printf(m, "%x", entry->preempt_count);
 898         else
 899                 seq_puts(m, ".");
 900 }
 901
 902 unsigned long preempt_mark_thresh = 100;
 903
 904 static notrace void
 905 lat_print_timestamp(struct seq_file *m, unsigned long long abs_usecs,
 906                     unsigned long rel_usecs)
 907 {
 908         seq_printf(m, " %4lldus", abs_usecs);
 909         if (rel_usecs > preempt_mark_thresh)
 910                 seq_puts(m, "!: ");
 911         else if (rel_usecs > 1)
 912                 seq_puts(m, "+: ");
 913         else
 914                 seq_puts(m, " : ");
 915 }
 916
 917 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
 918
 919 static notrace void
 920 print_lat_fmt(struct seq_file *m, struct trace_iterator *iter,
 921               unsigned int trace_idx, int cpu)
 922 {
 923         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 924         struct trace_entry *next_entry = find_next_entry(iter, NULL);
 925         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
 926         struct trace_entry *entry = iter->ent;
 927         unsigned long abs_usecs;
 928         unsigned long rel_usecs;
 929         char *comm;
 930         int S;
 931
 932         if (!next_entry)
 933                 next_entry = entry;
 934         rel_usecs = ns2usecs(next_entry->t - entry->t);
 935         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
 936
 937         if (verbose) {
 938                 comm = trace_find_cmdline(entry->pid);
 939                 seq_printf(m, "%16s %5d %d %d %08x %08x [%08lx]"
 940                            " %ld.%03ldms (+%ld.%03ldms): ",
 941                            comm,
 942                            entry->pid, cpu, entry->flags,
 943                            entry->preempt_count, trace_idx,
 944                            ns2usecs(entry->t),
 945                            abs_usecs/1000,
 946                            abs_usecs % 1000, rel_usecs/1000, rel_usecs % 1000);
 947         } else {
 948                 lat_print_generic(m, entry, cpu);
 949                 lat_print_timestamp(m, abs_usecs, rel_usecs);
 950         }
 951         switch (entry->type) {
 952         case TRACE_FN:
 953                 seq_print_ip_sym(m, entry->fn.ip, sym_flags);
 954                 seq_puts(m, " (");
 955                 seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
 956                 seq_puts(m, ")\n");
 957                 break;
 958         case TRACE_CTX:
 959                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
 960                         state_to_char[entry->ctx.prev_state] : 'X';
 961                 comm = trace_find_cmdline(entry->ctx.next_pid);
 962                 seq_printf(m, " %d:%d:%c --> %d:%d %s\n",
 963                            entry->ctx.prev_pid,
 964                            entry->ctx.prev_prio,
 965                            S,
 966                            entry->ctx.next_pid,
 967                            entry->ctx.next_prio,
 968                            comm);
 969                 break;
 970         default:
 971                 seq_printf(m, "Unknown type %d\n", entry->type);
 972         }
 973 }
 974
 975 static notrace void sync_time_offset(struct trace_iterator *iter)
 976 {
 977         struct trace_array_cpu *prev_array, *array;
 978         struct trace_entry *prev_entry, *entry;
 979         cycle_t prev_t, t;
 980
 981         entry = iter->ent;
 982         prev_entry = iter->prev_ent;
 983         if (!prev_entry)
 984                 return;
 985
 986         prev_array = iter->tr->data[iter->prev_cpu];
 987         array = iter->tr->data[iter->cpu];
 988
 989         prev_t = prev_entry->t + prev_array->time_offset;
 990         t = entry->t + array->time_offset;
 991
 992         /*
 993          * If time goes backwards we increase the offset of
 994          * the current array, to not have observable time warps.
 995          * This will quickly synchronize the time offsets of
 996          * multiple CPUs:
 997          */
 998         if (t < prev_t)
 999                 array->time_offset += prev_t - t;
1000 }
1001
1002 static notrace void
1003 print_trace_fmt(struct seq_file *m, struct trace_iterator *iter)
1004 {
1005         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1006         struct trace_entry *entry;
1007         unsigned long usec_rem;
1008         unsigned long long t;
1009         unsigned long secs;
1010         char *comm;
1011         int S;
1012
1013         sync_time_offset(iter);
1014         entry = iter->ent;
1015
1016         comm = trace_find_cmdline(iter->ent->pid);
1017
1018         t = ns2usecs(entry->t + iter->tr->data[iter->cpu]->time_offset);
1019         usec_rem = do_div(t, 1000000ULL);
1020         secs = (unsigned long)t;
1021
1022         seq_printf(m, "%16s-%-5d ", comm, entry->pid);
1023         seq_printf(m, "[%02d] ", iter->cpu);
1024         seq_printf(m, "%5lu.%06lu: ", secs, usec_rem);
1025
1026         switch (entry->type) {
1027         case TRACE_FN:
1028                 seq_print_ip_sym(m, entry->fn.ip, sym_flags);
1029                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1030                                                 entry->fn.parent_ip) {
1031                         seq_printf(m, " <-");
1032                         seq_print_ip_sym(m, entry->fn.parent_ip, sym_flags);
1033                 }
1034                 seq_printf(m, "\n");
1035                 break;
1036         case TRACE_CTX:
1037                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1038                         state_to_char[entry->ctx.prev_state] : 'X';
1039                 seq_printf(m, " %d:%d:%c ==> %d:%d\n",
1040                            entry->ctx.prev_pid,
1041                            entry->ctx.prev_prio,
1042                            S,
1043                            entry->ctx.next_pid,
1044                            entry->ctx.next_prio);
1045                 break;
1046         }
1047 }
1048
1049 static int trace_empty(struct trace_iterator *iter)
1050 {
1051         struct trace_array_cpu *data;
1052         int cpu;
1053
1054         for_each_possible_cpu(cpu) {
1055                 data = iter->tr->data[cpu];
1056
1057                 if (head_page(data) && data->trace_idx)
1058                         return 0;
1059         }
1060         return 1;
1061 }
1062
1063 static int s_show(struct seq_file *m, void *v)
1064 {
1065         struct trace_iterator *iter = v;
1066
1067         if (iter->ent == NULL) {
1068                 if (iter->tr) {
1069                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1070                         seq_puts(m, "#\n");
1071                 }
1072                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1073                         /* print nothing if the buffers are empty */
1074                         if (trace_empty(iter))
1075                                 return 0;
1076                         print_trace_header(m, iter);
1077                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1078                                 print_lat_help_header(m);
1079                 } else {
1080                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1081                                 print_func_help_header(m);
1082                 }
1083         } else {
1084                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1085                         print_lat_fmt(m, iter, iter->idx, iter->cpu);
1086                 else
1087                         print_trace_fmt(m, iter);
1088         }
1089
1090         return 0;
1091 }
1092
1093 static struct seq_operations tracer_seq_ops = {
1094         .start = s_start,
1095         .next = s_next,
1096         .stop = s_stop,
1097         .show = s_show,
1098 };
1099
1100 static struct trace_iterator notrace *
1101 __tracing_open(struct inode *inode, struct file *file, int *ret)
1102 {
1103         struct trace_iterator *iter;
1104
1105         if (tracing_disabled) {
1106                 *ret = -ENODEV;
1107                 return NULL;
1108         }
1109
1110         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1111         if (!iter) {
1112                 *ret = -ENOMEM;
1113                 goto out;
1114         }
1115
1116         mutex_lock(&trace_types_lock);
1117         if (current_trace && current_trace->print_max)
1118                 iter->tr = &max_tr;
1119         else
1120                 iter->tr = inode->i_private;
1121         iter->trace = current_trace;
1122         iter->pos = -1;
1123
1124         /* TODO stop tracer */
1125         *ret = seq_open(file, &tracer_seq_ops);
1126         if (!*ret) {
1127                 struct seq_file *m = file->private_data;
1128                 m->private = iter;
1129
1130                 /* stop the trace while dumping */
1131                 if (iter->tr->ctrl)
1132                         tracer_enabled = 0;
1133
1134                 if (iter->trace && iter->trace->open)
1135                         iter->trace->open(iter);
1136         } else {
1137                 kfree(iter);
1138                 iter = NULL;
1139         }
1140         mutex_unlock(&trace_types_lock);
1141
1142  out:
1143         return iter;
1144 }
1145
1146 int tracing_open_generic(struct inode *inode, struct file *filp)
1147 {
1148         if (tracing_disabled)
1149                 return -ENODEV;
1150
1151         filp->private_data = inode->i_private;
1152         return 0;
1153 }
1154
1155 int tracing_release(struct inode *inode, struct file *file)
1156 {
1157         struct seq_file *m = (struct seq_file *)file->private_data;
1158         struct trace_iterator *iter = m->private;
1159
1160         mutex_lock(&trace_types_lock);
1161         if (iter->trace && iter->trace->close)
1162                 iter->trace->close(iter);
1163
1164         /* reenable tracing if it was previously enabled */
1165         if (iter->tr->ctrl)
1166                 tracer_enabled = 1;
1167         mutex_unlock(&trace_types_lock);
1168
1169         seq_release(inode, file);
1170         kfree(iter);
1171         return 0;
1172 }
1173
1174 static int tracing_open(struct inode *inode, struct file *file)
1175 {
1176         int ret;
1177
1178         __tracing_open(inode, file, &ret);
1179
1180         return ret;
1181 }
1182
1183 static int tracing_lt_open(struct inode *inode, struct file *file)
1184 {
1185         struct trace_iterator *iter;
1186         int ret;
1187
1188         iter = __tracing_open(inode, file, &ret);
1189
1190         if (!ret)
1191                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1192
1193         return ret;
1194 }
1195
1196
1197 static notrace void *
1198 t_next(struct seq_file *m, void *v, loff_t *pos)
1199 {
1200         struct tracer *t = m->private;
1201
1202         (*pos)++;
1203
1204         if (t)
1205                 t = t->next;
1206
1207         m->private = t;
1208
1209         return t;
1210 }
1211
1212 static void *t_start(struct seq_file *m, loff_t *pos)
1213 {
1214         struct tracer *t = m->private;
1215         loff_t l = 0;
1216
1217         mutex_lock(&trace_types_lock);
1218         for (; t && l < *pos; t = t_next(m, t, &l))
1219                 ;
1220
1221         return t;
1222 }
1223
1224 static void t_stop(struct seq_file *m, void *p)
1225 {
1226         mutex_unlock(&trace_types_lock);
1227 }
1228
1229 static int t_show(struct seq_file *m, void *v)
1230 {
1231         struct tracer *t = v;
1232
1233         if (!t)
1234                 return 0;
1235
1236         seq_printf(m, "%s", t->name);
1237         if (t->next)
1238                 seq_putc(m, ' ');
1239         else
1240                 seq_putc(m, '\n');
1241
1242         return 0;
1243 }
1244
1245 static struct seq_operations show_traces_seq_ops = {
1246         .start = t_start,
1247         .next = t_next,
1248         .stop = t_stop,
1249         .show = t_show,
1250 };
1251
1252 static int show_traces_open(struct inode *inode, struct file *file)
1253 {
1254         int ret;
1255
1256         if (tracing_disabled)
1257                 return -ENODEV;
1258
1259         ret = seq_open(file, &show_traces_seq_ops);
1260         if (!ret) {
1261                 struct seq_file *m = file->private_data;
1262                 m->private = trace_types;
1263         }
1264
1265         return ret;
1266 }
1267
1268 static struct file_operations tracing_fops = {
1269         .open = tracing_open,
1270         .read = seq_read,
1271         .llseek = seq_lseek,
1272         .release = tracing_release,
1273 };
1274
1275 static struct file_operations tracing_lt_fops = {
1276         .open = tracing_lt_open,
1277         .read = seq_read,
1278         .llseek = seq_lseek,
1279         .release = tracing_release,
1280 };
1281
1282 static struct file_operations show_traces_fops = {
1283         .open = show_traces_open,
1284         .read = seq_read,
1285         .release = seq_release,
1286 };
1287
1288 static ssize_t
1289 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1290                        size_t cnt, loff_t *ppos)
1291 {
1292         char *buf;
1293         int r = 0;
1294         int len = 0;
1295         int i;
1296
1297         /* calulate max size */
1298         for (i = 0; trace_options[i]; i++) {
1299                 len += strlen(trace_options[i]);
1300                 len += 3; /* "no" and space */
1301         }
1302
1303         /* +2 for \n and \0 */
1304         buf = kmalloc(len + 2, GFP_KERNEL);
1305         if (!buf)
1306                 return -ENOMEM;
1307
1308         for (i = 0; trace_options[i]; i++) {
1309                 if (trace_flags & (1 << i))
1310                         r += sprintf(buf + r, "%s ", trace_options[i]);
1311                 else
1312                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1313         }
1314
1315         r += sprintf(buf + r, "\n");
1316         WARN_ON(r >= len + 2);
1317
1318         r = simple_read_from_buffer(ubuf, cnt, ppos,
1319                                     buf, r);
1320
1321         kfree(buf);
1322
1323         return r;
1324 }
1325
1326 static ssize_t
1327 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1328                         size_t cnt, loff_t *ppos)
1329 {
1330         char buf[64];
1331         char *cmp = buf;
1332         int neg = 0;
1333         int i;
1334
1335         if (cnt > 63)
1336                 cnt = 63;
1337
1338         if (copy_from_user(&buf, ubuf, cnt))
1339                 return -EFAULT;
1340
1341         buf[cnt] = 0;
1342
1343         if (strncmp(buf, "no", 2) == 0) {
1344                 neg = 1;
1345                 cmp += 2;
1346         }
1347
1348         for (i = 0; trace_options[i]; i++) {
1349                 int len = strlen(trace_options[i]);
1350
1351                 if (strncmp(cmp, trace_options[i], len) == 0) {
1352                         if (neg)
1353                                 trace_flags &= ~(1 << i);
1354                         else
1355                                 trace_flags |= (1 << i);
1356                         break;
1357                 }
1358         }
1359
1360         filp->f_pos += cnt;
1361
1362         return cnt;
1363 }
1364
1365 static struct file_operations tracing_iter_fops = {
1366         .open = tracing_open_generic,
1367         .read = tracing_iter_ctrl_read,
1368         .write = tracing_iter_ctrl_write,
1369 };
1370
1371 static const char readme_msg[] =
1372         "tracing mini-HOWTO:\n\n"
1373         "# mkdir /debug\n"
1374         "# mount -t debugfs nodev /debug\n\n"
1375         "# cat /debug/tracing/available_tracers\n"
1376         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1377         "# cat /debug/tracing/current_tracer\n"
1378         "none\n"
1379         "# echo sched_switch > /debug/tracing/current_tracer\n"
1380         "# cat /debug/tracing/current_tracer\n"
1381         "sched_switch\n"
1382         "# cat /debug/tracing/iter_ctrl\n"
1383         "noprint-parent nosym-offset nosym-addr noverbose\n"
1384         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1385         "# echo 1 > /debug/tracing/tracing_enabled\n"
1386         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1387         "echo 0 > /debug/tracing/tracing_enabled\n"
1388 ;
1389
1390 static ssize_t
1391 tracing_readme_read(struct file *filp, char __user *ubuf,
1392                        size_t cnt, loff_t *ppos)
1393 {
1394         return simple_read_from_buffer(ubuf, cnt, ppos,
1395                                         readme_msg, strlen(readme_msg));
1396 }
1397
1398 static struct file_operations tracing_readme_fops = {
1399         .open = tracing_open_generic,
1400         .read = tracing_readme_read,
1401 };
1402
1403
1404 static ssize_t
1405 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1406                   size_t cnt, loff_t *ppos)
1407 {
1408         struct trace_array *tr = filp->private_data;
1409         char buf[64];
1410         int r;
1411
1412         r = sprintf(buf, "%ld\n", tr->ctrl);
1413         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1414 }
1415
1416 static ssize_t
1417 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1418                    size_t cnt, loff_t *ppos)
1419 {
1420         struct trace_array *tr = filp->private_data;
1421         long val;
1422         char buf[64];
1423
1424         if (cnt > 63)
1425                 cnt = 63;
1426
1427         if (copy_from_user(&buf, ubuf, cnt))
1428                 return -EFAULT;
1429
1430         buf[cnt] = 0;
1431
1432         val = simple_strtoul(buf, NULL, 10);
1433
1434         val = !!val;
1435
1436         mutex_lock(&trace_types_lock);
1437         if (tr->ctrl ^ val) {
1438                 if (val)
1439                         tracer_enabled = 1;
1440                 else
1441                         tracer_enabled = 0;
1442
1443                 tr->ctrl = val;
1444
1445                 if (current_trace && current_trace->ctrl_update)
1446                         current_trace->ctrl_update(tr);
1447         }
1448         mutex_unlock(&trace_types_lock);
1449
1450         filp->f_pos += cnt;
1451
1452         return cnt;
1453 }
1454
1455 static ssize_t
1456 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1457                        size_t cnt, loff_t *ppos)
1458 {
1459         char buf[max_tracer_type_len+2];
1460         int r;
1461
1462         mutex_lock(&trace_types_lock);
1463         if (current_trace)
1464                 r = sprintf(buf, "%s\n", current_trace->name);
1465         else
1466                 r = sprintf(buf, "\n");
1467         mutex_unlock(&trace_types_lock);
1468
1469         return simple_read_from_buffer(ubuf, cnt, ppos,
1470                                        buf, r);
1471 }
1472
1473 static ssize_t
1474 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1475                         size_t cnt, loff_t *ppos)
1476 {
1477         struct trace_array *tr = &global_trace;
1478         struct tracer *t;
1479         char buf[max_tracer_type_len+1];
1480         int i;
1481
1482         if (cnt > max_tracer_type_len)
1483                 cnt = max_tracer_type_len;
1484
1485         if (copy_from_user(&buf, ubuf, cnt))
1486                 return -EFAULT;
1487
1488         buf[cnt] = 0;
1489
1490         /* strip ending whitespace. */
1491         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1492                 buf[i] = 0;
1493
1494         mutex_lock(&trace_types_lock);
1495         for (t = trace_types; t; t = t->next) {
1496                 if (strcmp(t->name, buf) == 0)
1497                         break;
1498         }
1499         if (!t || t == current_trace)
1500                 goto out;
1501
1502         if (current_trace && current_trace->reset)
1503                 current_trace->reset(tr);
1504
1505         current_trace = t;
1506         if (t->init)
1507                 t->init(tr);
1508
1509  out:
1510         mutex_unlock(&trace_types_lock);
1511
1512         filp->f_pos += cnt;
1513
1514         return cnt;
1515 }
1516
1517 static ssize_t
1518 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1519                      size_t cnt, loff_t *ppos)
1520 {
1521         unsigned long *ptr = filp->private_data;
1522         char buf[64];
1523         int r;
1524
1525         r = snprintf(buf, 64, "%ld\n",
1526                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1527         if (r > 64)
1528                 r = 64;
1529         return simple_read_from_buffer(ubuf, cnt, ppos,
1530                                        buf, r);
1531 }
1532
1533 static ssize_t
1534 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1535                       size_t cnt, loff_t *ppos)
1536 {
1537         long *ptr = filp->private_data;
1538         long val;
1539         char buf[64];
1540
1541         if (cnt > 63)
1542                 cnt = 63;
1543
1544         if (copy_from_user(&buf, ubuf, cnt))
1545                 return -EFAULT;
1546
1547         buf[cnt] = 0;
1548
1549         val = simple_strtoul(buf, NULL, 10);
1550
1551         *ptr = val * 1000;
1552
1553         return cnt;
1554 }
1555
1556 static struct file_operations tracing_max_lat_fops = {
1557         .open = tracing_open_generic,
1558         .read = tracing_max_lat_read,
1559         .write = tracing_max_lat_write,
1560 };
1561
1562 static struct file_operations tracing_ctrl_fops = {
1563         .open = tracing_open_generic,
1564         .read = tracing_ctrl_read,
1565         .write = tracing_ctrl_write,
1566 };
1567
1568 static struct file_operations set_tracer_fops = {
1569         .open = tracing_open_generic,
1570         .read = tracing_set_trace_read,
1571         .write = tracing_set_trace_write,
1572 };
1573
1574 #ifdef CONFIG_DYNAMIC_FTRACE
1575
1576 static ssize_t
1577 tracing_read_long(struct file *filp, char __user *ubuf,
1578                   size_t cnt, loff_t *ppos)
1579 {
1580         unsigned long *p = filp->private_data;
1581         char buf[64];
1582         int r;
1583
1584         r = sprintf(buf, "%ld\n", *p);
1585         return simple_read_from_buffer(ubuf, cnt, ppos,
1586                                        buf, r);
1587 }
1588
1589 static struct file_operations tracing_read_long_fops = {
1590         .open = tracing_open_generic,
1591         .read = tracing_read_long,
1592 };
1593 #endif
1594
1595 static struct dentry *d_tracer;
1596
1597 struct dentry *tracing_init_dentry(void)
1598 {
1599         static int once;
1600
1601         if (d_tracer)
1602                 return d_tracer;
1603
1604         d_tracer = debugfs_create_dir("tracing", NULL);
1605
1606         if (!d_tracer && !once) {
1607                 once = 1;
1608                 pr_warning("Could not create debugfs directory 'tracing'\n");
1609                 return NULL;
1610         }
1611
1612         return d_tracer;
1613 }
1614
1615 #ifdef CONFIG_FTRACE_SELFTEST
1616 /* Let selftest have access to static functions in this file */
1617 #include "trace_selftest.c"
1618 #endif
1619
1620 static __init void tracer_init_debugfs(void)
1621 {
1622         struct dentry *d_tracer;
1623         struct dentry *entry;
1624
1625         d_tracer = tracing_init_dentry();
1626
1627         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
1628                                     &global_trace, &tracing_ctrl_fops);
1629         if (!entry)
1630                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
1631
1632         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
1633                                     NULL, &tracing_iter_fops);
1634         if (!entry)
1635                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
1636
1637         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
1638                                     &global_trace, &tracing_lt_fops);
1639         if (!entry)
1640                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
1641
1642         entry = debugfs_create_file("trace", 0444, d_tracer,
1643                                     &global_trace, &tracing_fops);
1644         if (!entry)
1645                 pr_warning("Could not create debugfs 'trace' entry\n");
1646
1647         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
1648                                     &global_trace, &show_traces_fops);
1649         if (!entry)
1650                 pr_warning("Could not create debugfs 'trace' entry\n");
1651
1652         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
1653                                     &global_trace, &set_tracer_fops);
1654         if (!entry)
1655                 pr_warning("Could not create debugfs 'trace' entry\n");
1656
1657         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
1658                                     &tracing_max_latency,
1659                                     &tracing_max_lat_fops);
1660         if (!entry)
1661                 pr_warning("Could not create debugfs "
1662                            "'tracing_max_latency' entry\n");
1663
1664         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
1665                                     &tracing_thresh, &tracing_max_lat_fops);
1666         if (!entry)
1667                 pr_warning("Could not create debugfs "
1668                            "'tracing_threash' entry\n");
1669         entry = debugfs_create_file("README", 0644, d_tracer,
1670                                     NULL, &tracing_readme_fops);
1671         if (!entry)
1672                 pr_warning("Could not create debugfs 'README' entry\n");
1673
1674
1675 #ifdef CONFIG_DYNAMIC_FTRACE
1676         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
1677                                     &ftrace_update_tot_cnt,
1678                                     &tracing_read_long_fops);
1679         if (!entry)
1680                 pr_warning("Could not create debugfs "
1681                            "'dyn_ftrace_total_info' entry\n");
1682 #endif
1683 }
1684
1685 /* dummy trace to disable tracing */
1686 static struct tracer no_tracer __read_mostly =
1687 {
1688         .name = "none",
1689 };
1690
1691 static int trace_alloc_page(void)
1692 {
1693         struct trace_array_cpu *data;
1694         struct page *page, *tmp;
1695         LIST_HEAD(pages);
1696         void *array;
1697         int i;
1698
1699         /* first allocate a page for each CPU */
1700         for_each_possible_cpu(i) {
1701                 array = (void *)__get_free_page(GFP_KERNEL);
1702                 if (array == NULL) {
1703                         printk(KERN_ERR "tracer: failed to allocate page"
1704                                "for trace buffer!\n");
1705                         goto free_pages;
1706                 }
1707
1708                 page = virt_to_page(array);
1709                 list_add(&page->lru, &pages);
1710
1711 /* Only allocate if we are actually using the max trace */
1712 #ifdef CONFIG_TRACER_MAX_TRACE
1713                 array = (void *)__get_free_page(GFP_KERNEL);
1714                 if (array == NULL) {
1715                         printk(KERN_ERR "tracer: failed to allocate page"
1716                                "for trace buffer!\n");
1717                         goto free_pages;
1718                 }
1719                 page = virt_to_page(array);
1720                 list_add(&page->lru, &pages);
1721 #endif
1722         }
1723
1724         /* Now that we successfully allocate a page per CPU, add them */
1725         for_each_possible_cpu(i) {
1726                 data = global_trace.data[i];
1727                 page = list_entry(pages.next, struct page, lru);
1728                 list_del_init(&page->lru);
1729                 list_add_tail(&page->lru, &data->trace_pages);
1730                 ClearPageLRU(page);
1731
1732 #ifdef CONFIG_TRACER_MAX_TRACE
1733                 data = max_tr.data[i];
1734                 page = list_entry(pages.next, struct page, lru);
1735                 list_del_init(&page->lru);
1736                 list_add_tail(&page->lru, &data->trace_pages);
1737                 SetPageLRU(page);
1738 #endif
1739         }
1740         global_trace.entries += ENTRIES_PER_PAGE;
1741
1742         return 0;
1743
1744  free_pages:
1745         list_for_each_entry_safe(page, tmp, &pages, lru) {
1746                 list_del_init(&page->lru);
1747                 __free_page(page);
1748         }
1749         return -ENOMEM;
1750 }
1751
1752 __init static int tracer_alloc_buffers(void)
1753 {
1754         struct trace_array_cpu *data;
1755         void *array;
1756         struct page *page;
1757         int pages = 0;
1758         int ret = -ENOMEM;
1759         int i;
1760
1761         /* Allocate the first page for all buffers */
1762         for_each_possible_cpu(i) {
1763                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
1764                 max_tr.data[i] = &per_cpu(max_data, i);
1765
1766                 array = (void *)__get_free_page(GFP_KERNEL);
1767                 if (array == NULL) {
1768                         printk(KERN_ERR "tracer: failed to allocate page"
1769                                "for trace buffer!\n");
1770                         goto free_buffers;
1771                 }
1772
1773                 /* set the array to the list */
1774                 INIT_LIST_HEAD(&data->trace_pages);
1775                 page = virt_to_page(array);
1776                 list_add(&page->lru, &data->trace_pages);
1777                 /* use the LRU flag to differentiate the two buffers */
1778                 ClearPageLRU(page);
1779
1780 /* Only allocate if we are actually using the max trace */
1781 #ifdef CONFIG_TRACER_MAX_TRACE
1782                 array = (void *)__get_free_page(GFP_KERNEL);
1783                 if (array == NULL) {
1784                         printk(KERN_ERR "tracer: failed to allocate page"
1785                                "for trace buffer!\n");
1786                         goto free_buffers;
1787                 }
1788
1789                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
1790                 page = virt_to_page(array);
1791                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
1792                 SetPageLRU(page);
1793 #endif
1794         }
1795
1796         /*
1797          * Since we allocate by orders of pages, we may be able to
1798          * round up a bit.
1799          */
1800         global_trace.entries = ENTRIES_PER_PAGE;
1801         pages++;
1802
1803         while (global_trace.entries < trace_nr_entries) {
1804                 if (trace_alloc_page())
1805                         break;
1806                 pages++;
1807         }
1808         max_tr.entries = global_trace.entries;
1809
1810         pr_info("tracer: %d pages allocated for %ld",
1811                 pages, trace_nr_entries);
1812         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
1813         pr_info("   actual entries %ld\n", global_trace.entries);
1814
1815         tracer_init_debugfs();
1816
1817         trace_init_cmdlines();
1818
1819         register_tracer(&no_tracer);
1820         current_trace = &no_tracer;
1821
1822         /* All seems OK, enable tracing */
1823         tracing_disabled = 0;
1824
1825         return 0;
1826
1827  free_buffers:
1828         for (i-- ; i >= 0; i--) {
1829                 struct page *page, *tmp;
1830                 struct trace_array_cpu *data = global_trace.data[i];
1831
1832                 if (data) {
1833                         list_for_each_entry_safe(page, tmp,
1834                                                  &data->trace_pages, lru) {
1835                                 list_del_init(&page->lru);
1836                                 __free_page(page);
1837                         }
1838                 }
1839
1840 #ifdef CONFIG_TRACER_MAX_TRACE
1841                 data = max_tr.data[i];
1842                 if (data) {
1843                         list_for_each_entry_safe(page, tmp,
1844                                                  &data->trace_pages, lru) {
1845                                 list_del_init(&page->lru);
1846                                 __free_page(page);
1847                         }
1848                 }
1849 #endif
1850         }
1851         return ret;
1852 }
1853
1854 fs_initcall(tracer_alloc_buffers);