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 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 static notrace int
 173 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 174 {
 175         int len = (PAGE_SIZE - 1) - s->len;
 176         va_list ap;
 177
 178         if (!len)
 179                 return 0;
 180
 181         va_start(ap, fmt);
 182         len = vsnprintf(s->buffer + s->len, len, fmt, ap);
 183         va_end(ap);
 184
 185         s->len += len;
 186
 187         return len;
 188 }
 189
 190 static notrace int
 191 trace_seq_puts(struct trace_seq *s, const char *str)
 192 {
 193         int len = strlen(str);
 194
 195         if (len > ((PAGE_SIZE - 1) - s->len))
 196                 len = (PAGE_SIZE - 1) - s->len;
 197
 198         memcpy(s->buffer + s->len, str, len);
 199         s->len += len;
 200
 201         return len;
 202 }
 203
 204 static notrace int
 205 trace_seq_putc(struct trace_seq *s, unsigned char c)
 206 {
 207         if (s->len >= (PAGE_SIZE - 1))
 208                 return 0;
 209
 210         s->buffer[s->len++] = c;
 211
 212         return 1;
 213 }
 214
 215 static notrace void
 216 trace_seq_reset(struct trace_seq *s)
 217 {
 218         s->len = 0;
 219 }
 220
 221 static notrace void
 222 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 223 {
 224         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 225
 226         s->buffer[len] = 0;
 227         seq_puts(m, s->buffer);
 228
 229         trace_seq_reset(s);
 230 }
 231
 232 notrace static void
 233 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 234 {
 235         struct list_head flip_pages;
 236
 237         INIT_LIST_HEAD(&flip_pages);
 238
 239         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 240                 sizeof(struct trace_array_cpu) -
 241                 offsetof(struct trace_array_cpu, trace_head_idx));
 242
 243         check_pages(tr1);
 244         check_pages(tr2);
 245         list_splice_init(&tr1->trace_pages, &flip_pages);
 246         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 247         list_splice_init(&flip_pages, &tr2->trace_pages);
 248         BUG_ON(!list_empty(&flip_pages));
 249         check_pages(tr1);
 250         check_pages(tr2);
 251 }
 252
 253 notrace void
 254 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 255 {
 256         struct trace_array_cpu *data;
 257         int i;
 258
 259         WARN_ON_ONCE(!irqs_disabled());
 260         spin_lock(&ftrace_max_lock);
 261         /* clear out all the previous traces */
 262         for_each_possible_cpu(i) {
 263                 data = tr->data[i];
 264                 flip_trace(max_tr.data[i], data);
 265                 tracing_reset(data);
 266         }
 267
 268         __update_max_tr(tr, tsk, cpu);
 269         spin_unlock(&ftrace_max_lock);
 270 }
 271
 272 /**
 273  * update_max_tr_single - only copy one trace over, and reset the rest
 274  * @tr - tracer
 275  * @tsk - task with the latency
 276  * @cpu - the cpu of the buffer to copy.
 277  */
 278 notrace void
 279 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 280 {
 281         struct trace_array_cpu *data = tr->data[cpu];
 282         int i;
 283
 284         WARN_ON_ONCE(!irqs_disabled());
 285         spin_lock(&ftrace_max_lock);
 286         for_each_possible_cpu(i)
 287                 tracing_reset(max_tr.data[i]);
 288
 289         flip_trace(max_tr.data[cpu], data);
 290         tracing_reset(data);
 291
 292         __update_max_tr(tr, tsk, cpu);
 293         spin_unlock(&ftrace_max_lock);
 294 }
 295
 296 int register_tracer(struct tracer *type)
 297 {
 298         struct tracer *t;
 299         int len;
 300         int ret = 0;
 301
 302         if (!type->name) {
 303                 pr_info("Tracer must have a name\n");
 304                 return -1;
 305         }
 306
 307         mutex_lock(&trace_types_lock);
 308         for (t = trace_types; t; t = t->next) {
 309                 if (strcmp(type->name, t->name) == 0) {
 310                         /* already found */
 311                         pr_info("Trace %s already registered\n",
 312                                 type->name);
 313                         ret = -1;
 314                         goto out;
 315                 }
 316         }
 317
 318 #ifdef CONFIG_FTRACE_STARTUP_TEST
 319         if (type->selftest) {
 320                 struct tracer *saved_tracer = current_trace;
 321                 struct trace_array_cpu *data;
 322                 struct trace_array *tr = &global_trace;
 323                 int saved_ctrl = tr->ctrl;
 324                 int i;
 325                 /*
 326                  * Run a selftest on this tracer.
 327                  * Here we reset the trace buffer, and set the current
 328                  * tracer to be this tracer. The tracer can then run some
 329                  * internal tracing to verify that everything is in order.
 330                  * If we fail, we do not register this tracer.
 331                  */
 332                 for_each_possible_cpu(i) {
 333                         data = tr->data[i];
 334                         if (!head_page(data))
 335                                 continue;
 336                         tracing_reset(data);
 337                 }
 338                 current_trace = type;
 339                 tr->ctrl = 0;
 340                 /* the test is responsible for initializing and enabling */
 341                 pr_info("Testing tracer %s: ", type->name);
 342                 ret = type->selftest(type, tr);
 343                 /* the test is responsible for resetting too */
 344                 current_trace = saved_tracer;
 345                 tr->ctrl = saved_ctrl;
 346                 if (ret) {
 347                         printk(KERN_CONT "FAILED!\n");
 348                         goto out;
 349                 }
 350                 /* Only reset on passing, to avoid touching corrupted buffers */
 351                 for_each_possible_cpu(i) {
 352                         data = tr->data[i];
 353                         if (!head_page(data))
 354                                 continue;
 355                         tracing_reset(data);
 356                 }
 357                 printk(KERN_CONT "PASSED\n");
 358         }
 359 #endif
 360
 361         type->next = trace_types;
 362         trace_types = type;
 363         len = strlen(type->name);
 364         if (len > max_tracer_type_len)
 365                 max_tracer_type_len = len;
 366
 367  out:
 368         mutex_unlock(&trace_types_lock);
 369
 370         return ret;
 371 }
 372
 373 void unregister_tracer(struct tracer *type)
 374 {
 375         struct tracer **t;
 376         int len;
 377
 378         mutex_lock(&trace_types_lock);
 379         for (t = &trace_types; *t; t = &(*t)->next) {
 380                 if (*t == type)
 381                         goto found;
 382         }
 383         pr_info("Trace %s not registered\n", type->name);
 384         goto out;
 385
 386  found:
 387         *t = (*t)->next;
 388         if (strlen(type->name) != max_tracer_type_len)
 389                 goto out;
 390
 391         max_tracer_type_len = 0;
 392         for (t = &trace_types; *t; t = &(*t)->next) {
 393                 len = strlen((*t)->name);
 394                 if (len > max_tracer_type_len)
 395                         max_tracer_type_len = len;
 396         }
 397  out:
 398         mutex_unlock(&trace_types_lock);
 399 }
 400
 401 notrace void tracing_reset(struct trace_array_cpu *data)
 402 {
 403         data->trace_idx = 0;
 404         data->trace_head = data->trace_tail = head_page(data);
 405         data->trace_head_idx = 0;
 406         data->trace_tail_idx = 0;
 407 }
 408
 409 #ifdef CONFIG_FTRACE
 410 static notrace void
 411 function_trace_call(unsigned long ip, unsigned long parent_ip)
 412 {
 413         struct trace_array *tr = &global_trace;
 414         struct trace_array_cpu *data;
 415         unsigned long flags;
 416         long disabled;
 417         int cpu;
 418
 419         if (unlikely(!tracer_enabled))
 420                 return;
 421
 422         local_irq_save(flags);
 423         cpu = raw_smp_processor_id();
 424         data = tr->data[cpu];
 425         disabled = atomic_inc_return(&data->disabled);
 426
 427         if (likely(disabled == 1))
 428                 ftrace(tr, data, ip, parent_ip, flags);
 429
 430         atomic_dec(&data->disabled);
 431         local_irq_restore(flags);
 432 }
 433
 434 static struct ftrace_ops trace_ops __read_mostly =
 435 {
 436         .func = function_trace_call,
 437 };
 438 #endif
 439
 440 notrace void tracing_start_function_trace(void)
 441 {
 442         register_ftrace_function(&trace_ops);
 443 }
 444
 445 notrace void tracing_stop_function_trace(void)
 446 {
 447         unregister_ftrace_function(&trace_ops);
 448 }
 449
 450 #define SAVED_CMDLINES 128
 451 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 452 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 453 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 454 static int cmdline_idx;
 455 static DEFINE_SPINLOCK(trace_cmdline_lock);
 456 atomic_t trace_record_cmdline_disabled;
 457
 458 static void trace_init_cmdlines(void)
 459 {
 460         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 461         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 462         cmdline_idx = 0;
 463 }
 464
 465 notrace void trace_stop_cmdline_recording(void);
 466
 467 static notrace void trace_save_cmdline(struct task_struct *tsk)
 468 {
 469         unsigned map;
 470         unsigned idx;
 471
 472         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 473                 return;
 474
 475         /*
 476          * It's not the end of the world if we don't get
 477          * the lock, but we also don't want to spin
 478          * nor do we want to disable interrupts,
 479          * so if we miss here, then better luck next time.
 480          */
 481         if (!spin_trylock(&trace_cmdline_lock))
 482                 return;
 483
 484         idx = map_pid_to_cmdline[tsk->pid];
 485         if (idx >= SAVED_CMDLINES) {
 486                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 487
 488                 map = map_cmdline_to_pid[idx];
 489                 if (map <= PID_MAX_DEFAULT)
 490                         map_pid_to_cmdline[map] = (unsigned)-1;
 491
 492                 map_pid_to_cmdline[tsk->pid] = idx;
 493
 494                 cmdline_idx = idx;
 495         }
 496
 497         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 498
 499         spin_unlock(&trace_cmdline_lock);
 500 }
 501
 502 static notrace char *trace_find_cmdline(int pid)
 503 {
 504         char *cmdline = "<...>";
 505         unsigned map;
 506
 507         if (!pid)
 508                 return "<idle>";
 509
 510         if (pid > PID_MAX_DEFAULT)
 511                 goto out;
 512
 513         map = map_pid_to_cmdline[pid];
 514         if (map >= SAVED_CMDLINES)
 515                 goto out;
 516
 517         cmdline = saved_cmdlines[map];
 518
 519  out:
 520         return cmdline;
 521 }
 522
 523 notrace void tracing_record_cmdline(struct task_struct *tsk)
 524 {
 525         if (atomic_read(&trace_record_cmdline_disabled))
 526                 return;
 527
 528         trace_save_cmdline(tsk);
 529 }
 530
 531 static inline notrace struct list_head *
 532 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 533 {
 534         /*
 535          * Roundrobin - but skip the head (which is not a real page):
 536          */
 537         next = next->next;
 538         if (unlikely(next == &data->trace_pages))
 539                 next = next->next;
 540         BUG_ON(next == &data->trace_pages);
 541
 542         return next;
 543 }
 544
 545 static inline notrace void *
 546 trace_next_page(struct trace_array_cpu *data, void *addr)
 547 {
 548         struct list_head *next;
 549         struct page *page;
 550
 551         page = virt_to_page(addr);
 552
 553         next = trace_next_list(data, &page->lru);
 554         page = list_entry(next, struct page, lru);
 555
 556         return page_address(page);
 557 }
 558
 559 static inline notrace struct trace_entry *
 560 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 561 {
 562         unsigned long idx, idx_next;
 563         struct trace_entry *entry;
 564
 565         data->trace_idx++;
 566         idx = data->trace_head_idx;
 567         idx_next = idx + 1;
 568
 569         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 570
 571         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 572
 573         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 574                 data->trace_head = trace_next_page(data, data->trace_head);
 575                 idx_next = 0;
 576         }
 577
 578         if (data->trace_head == data->trace_tail &&
 579             idx_next == data->trace_tail_idx) {
 580                 /* overrun */
 581                 data->trace_tail_idx++;
 582                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 583                         data->trace_tail =
 584                                 trace_next_page(data, data->trace_tail);
 585                         data->trace_tail_idx = 0;
 586                 }
 587         }
 588
 589         data->trace_head_idx = idx_next;
 590
 591         return entry;
 592 }
 593
 594 static inline notrace void
 595 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 596 {
 597         struct task_struct *tsk = current;
 598         unsigned long pc;
 599
 600         pc = preempt_count();
 601
 602         entry->idx              = atomic_inc_return(&tracer_counter);
 603         entry->preempt_count    = pc & 0xff;
 604         entry->pid              = tsk->pid;
 605         entry->t                = now(raw_smp_processor_id());
 606         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 607                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 608                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 609                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 610 }
 611
 612 notrace void
 613 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 614        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 615 {
 616         struct trace_entry *entry;
 617
 618         entry                   = tracing_get_trace_entry(tr, data);
 619         tracing_generic_entry_update(entry, flags);
 620         entry->type             = TRACE_FN;
 621         entry->fn.ip            = ip;
 622         entry->fn.parent_ip     = parent_ip;
 623 }
 624
 625 notrace void
 626 tracing_sched_switch_trace(struct trace_array *tr,
 627                            struct trace_array_cpu *data,
 628                            struct task_struct *prev, struct task_struct *next,
 629                            unsigned long flags)
 630 {
 631         struct trace_entry *entry;
 632
 633         entry                   = tracing_get_trace_entry(tr, data);
 634         tracing_generic_entry_update(entry, flags);
 635         entry->type             = TRACE_CTX;
 636         entry->ctx.prev_pid     = prev->pid;
 637         entry->ctx.prev_prio    = prev->prio;
 638         entry->ctx.prev_state   = prev->state;
 639         entry->ctx.next_pid     = next->pid;
 640         entry->ctx.next_prio    = next->prio;
 641 }
 642
 643 enum trace_file_type {
 644         TRACE_FILE_LAT_FMT      = 1,
 645 };
 646
 647 static struct trace_entry *
 648 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 649                 struct trace_iterator *iter, int cpu)
 650 {
 651         struct page *page;
 652         struct trace_entry *array;
 653
 654         if (iter->next_idx[cpu] >= tr->entries ||
 655             iter->next_idx[cpu] >= data->trace_idx)
 656                 return NULL;
 657
 658         if (!iter->next_page[cpu]) {
 659                 /* Initialize the iterator for this cpu trace buffer */
 660                 WARN_ON(!data->trace_tail);
 661                 page = virt_to_page(data->trace_tail);
 662                 iter->next_page[cpu] = &page->lru;
 663                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 664         }
 665
 666         page = list_entry(iter->next_page[cpu], struct page, lru);
 667         BUG_ON(&data->trace_pages == &page->lru);
 668
 669         array = page_address(page);
 670
 671         /* Still possible to catch up to the tail */
 672         if (iter->next_idx[cpu] && array == data->trace_tail &&
 673             iter->next_page_idx[cpu] == data->trace_tail_idx)
 674                 return NULL;
 675
 676         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 677         return &array[iter->next_page_idx[cpu]];
 678 }
 679
 680 static struct notrace trace_entry *
 681 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 682 {
 683         struct trace_array *tr = iter->tr;
 684         struct trace_entry *ent, *next = NULL;
 685         int next_cpu = -1;
 686         int cpu;
 687
 688         for_each_possible_cpu(cpu) {
 689                 if (!head_page(tr->data[cpu]))
 690                         continue;
 691                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 692                 if (ent &&
 693                     (!next || (long)(next->idx - ent->idx) > 0)) {
 694                         next = ent;
 695                         next_cpu = cpu;
 696                 }
 697         }
 698
 699         if (ent_cpu)
 700                 *ent_cpu = next_cpu;
 701
 702         return next;
 703 }
 704
 705 static void *find_next_entry_inc(struct trace_iterator *iter)
 706 {
 707         struct trace_entry *next;
 708         int next_cpu = -1;
 709
 710         next = find_next_entry(iter, &next_cpu);
 711
 712         if (next) {
 713                 iter->idx++;
 714                 iter->next_idx[next_cpu]++;
 715                 iter->next_page_idx[next_cpu]++;
 716
 717                 if (iter->next_page_idx[next_cpu] >= ENTRIES_PER_PAGE) {
 718                         struct trace_array_cpu *data = iter->tr->data[next_cpu];
 719
 720                         iter->next_page_idx[next_cpu] = 0;
 721                         iter->next_page[next_cpu] =
 722                              trace_next_list(data, iter->next_page[next_cpu]);
 723
 724                 }
 725         }
 726         iter->prev_ent = iter->ent;
 727         iter->prev_cpu = iter->cpu;
 728
 729         iter->ent = next;
 730         iter->cpu = next_cpu;
 731
 732         return next ? iter : NULL;
 733 }
 734
 735 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 736 {
 737         struct trace_iterator *iter = m->private;
 738         void *last_ent = iter->ent;
 739         int i = (int)*pos;
 740         void *ent;
 741
 742         (*pos)++;
 743
 744         /* can't go backwards */
 745         if (iter->idx > i)
 746                 return NULL;
 747
 748         if (iter->idx < 0)
 749                 ent = find_next_entry_inc(iter);
 750         else
 751                 ent = iter;
 752
 753         while (ent && iter->idx < i)
 754                 ent = find_next_entry_inc(iter);
 755
 756         iter->pos = *pos;
 757
 758         if (last_ent && !ent)
 759                 seq_puts(m, "\n\nvim:ft=help\n");
 760
 761         return ent;
 762 }
 763
 764 static void *s_start(struct seq_file *m, loff_t *pos)
 765 {
 766         struct trace_iterator *iter = m->private;
 767         void *p = NULL;
 768         loff_t l = 0;
 769         int i;
 770
 771         mutex_lock(&trace_types_lock);
 772
 773         if (!current_trace || current_trace != iter->trace)
 774                 return NULL;
 775
 776         atomic_inc(&trace_record_cmdline_disabled);
 777
 778         /* let the tracer grab locks here if needed */
 779         if (current_trace->start)
 780                 current_trace->start(iter);
 781
 782         if (*pos != iter->pos) {
 783                 iter->ent = NULL;
 784                 iter->cpu = 0;
 785                 iter->idx = -1;
 786                 iter->prev_ent = NULL;
 787                 iter->prev_cpu = -1;
 788
 789                 for_each_possible_cpu(i) {
 790                         iter->next_idx[i] = 0;
 791                         iter->next_page[i] = NULL;
 792                 }
 793
 794                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 795                         ;
 796
 797         } else {
 798                 l = *pos - 1;
 799                 p = s_next(m, p, &l);
 800         }
 801
 802         return p;
 803 }
 804
 805 static void s_stop(struct seq_file *m, void *p)
 806 {
 807         struct trace_iterator *iter = m->private;
 808
 809         atomic_dec(&trace_record_cmdline_disabled);
 810
 811         /* let the tracer release locks here if needed */
 812         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 813                 iter->trace->stop(iter);
 814
 815         mutex_unlock(&trace_types_lock);
 816 }
 817
 818 static void
 819 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 820 {
 821 #ifdef CONFIG_KALLSYMS
 822         char str[KSYM_SYMBOL_LEN];
 823
 824         kallsyms_lookup(address, NULL, NULL, NULL, str);
 825
 826         trace_seq_printf(s, fmt, str);
 827 #endif
 828 }
 829
 830 static void
 831 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 832                      unsigned long address)
 833 {
 834 #ifdef CONFIG_KALLSYMS
 835         char str[KSYM_SYMBOL_LEN];
 836
 837         sprint_symbol(str, address);
 838         trace_seq_printf(s, fmt, str);
 839 #endif
 840 }
 841
 842 #ifndef CONFIG_64BIT
 843 # define IP_FMT "%08lx"
 844 #else
 845 # define IP_FMT "%016lx"
 846 #endif
 847
 848 static notrace void
 849 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 850 {
 851         if (!ip) {
 852                 trace_seq_printf(s, "0");
 853                 return;
 854         }
 855
 856         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 857                 seq_print_sym_offset(s, "%s", ip);
 858         else
 859                 seq_print_sym_short(s, "%s", ip);
 860
 861         if (sym_flags & TRACE_ITER_SYM_ADDR)
 862                 trace_seq_printf(s, " <" IP_FMT ">", ip);
 863 }
 864
 865 static notrace void print_lat_help_header(struct seq_file *m)
 866 {
 867         seq_puts(m, "#                _------=> CPU#            \n");
 868         seq_puts(m, "#               / _-----=> irqs-off        \n");
 869         seq_puts(m, "#              | / _----=> need-resched    \n");
 870         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 871         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 872         seq_puts(m, "#              |||| /                      \n");
 873         seq_puts(m, "#              |||||     delay             \n");
 874         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 875         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
 876 }
 877
 878 static notrace void print_func_help_header(struct seq_file *m)
 879 {
 880         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
 881         seq_puts(m, "#              | |      |          |         |\n");
 882 }
 883
 884
 885 static notrace void
 886 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
 887 {
 888         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 889         struct trace_array *tr = iter->tr;
 890         struct trace_array_cpu *data = tr->data[tr->cpu];
 891         struct tracer *type = current_trace;
 892         unsigned long total   = 0;
 893         unsigned long entries = 0;
 894         int cpu;
 895         const char *name = "preemption";
 896
 897         if (type)
 898                 name = type->name;
 899
 900         for_each_possible_cpu(cpu) {
 901                 if (head_page(tr->data[cpu])) {
 902                         total += tr->data[cpu]->trace_idx;
 903                         if (tr->data[cpu]->trace_idx > tr->entries)
 904                                 entries += tr->entries;
 905                         else
 906                                 entries += tr->data[cpu]->trace_idx;
 907                 }
 908         }
 909
 910         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
 911                    name, UTS_RELEASE);
 912         seq_puts(m, "-----------------------------------"
 913                  "---------------------------------\n");
 914         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
 915                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
 916                    nsecs_to_usecs(data->saved_latency),
 917                    entries,
 918                    total,
 919                    tr->cpu,
 920 #if defined(CONFIG_PREEMPT_NONE)
 921                    "server",
 922 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
 923                    "desktop",
 924 #elif defined(CONFIG_PREEMPT_DESKTOP)
 925                    "preempt",
 926 #else
 927                    "unknown",
 928 #endif
 929                    /* These are reserved for later use */
 930                    0, 0, 0, 0);
 931 #ifdef CONFIG_SMP
 932         seq_printf(m, " #P:%d)\n", num_online_cpus());
 933 #else
 934         seq_puts(m, ")\n");
 935 #endif
 936         seq_puts(m, "    -----------------\n");
 937         seq_printf(m, "    | task: %.16s-%d "
 938                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
 939                    data->comm, data->pid, data->uid, data->nice,
 940                    data->policy, data->rt_priority);
 941         seq_puts(m, "    -----------------\n");
 942
 943         if (data->critical_start) {
 944                 seq_puts(m, " => started at: ");
 945                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
 946                 trace_print_seq(m, &iter->seq);
 947                 seq_puts(m, "\n => ended at:   ");
 948                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
 949                 trace_print_seq(m, &iter->seq);
 950                 seq_puts(m, "\n");
 951         }
 952
 953         seq_puts(m, "\n");
 954 }
 955
 956 static notrace void
 957 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
 958 {
 959         int hardirq, softirq;
 960         char *comm;
 961
 962         comm = trace_find_cmdline(entry->pid);
 963
 964         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
 965         trace_seq_printf(s, "%d", cpu);
 966         trace_seq_printf(s, "%c%c",
 967                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
 968                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
 969
 970         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
 971         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
 972         if (hardirq && softirq)
 973                 trace_seq_putc(s, 'H');
 974         else {
 975                 if (hardirq)
 976                         trace_seq_putc(s, 'h');
 977                 else {
 978                         if (softirq)
 979                                 trace_seq_putc(s, 's');
 980                         else
 981                                 trace_seq_putc(s, '.');
 982                 }
 983         }
 984
 985         if (entry->preempt_count)
 986                 trace_seq_printf(s, "%x", entry->preempt_count);
 987         else
 988                 trace_seq_puts(s, ".");
 989 }
 990
 991 unsigned long preempt_mark_thresh = 100;
 992
 993 static notrace void
 994 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
 995                     unsigned long rel_usecs)
 996 {
 997         trace_seq_printf(s, " %4lldus", abs_usecs);
 998         if (rel_usecs > preempt_mark_thresh)
 999                 trace_seq_puts(s, "!: ");
1000         else if (rel_usecs > 1)
1001                 trace_seq_puts(s, "+: ");
1002         else
1003                 trace_seq_puts(s, " : ");
1004 }
1005
1006 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1007
1008 static notrace void
1009 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1010 {
1011         struct trace_seq *s = &iter->seq;
1012         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1013         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1014         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1015         struct trace_entry *entry = iter->ent;
1016         unsigned long abs_usecs;
1017         unsigned long rel_usecs;
1018         char *comm;
1019         int S;
1020
1021         if (!next_entry)
1022                 next_entry = entry;
1023         rel_usecs = ns2usecs(next_entry->t - entry->t);
1024         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1025
1026         if (verbose) {
1027                 comm = trace_find_cmdline(entry->pid);
1028                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1029                                  " %ld.%03ldms (+%ld.%03ldms): ",
1030                                  comm,
1031                                  entry->pid, cpu, entry->flags,
1032                                  entry->preempt_count, trace_idx,
1033                                  ns2usecs(entry->t),
1034                                  abs_usecs/1000,
1035                                  abs_usecs % 1000, rel_usecs/1000,
1036                                  rel_usecs % 1000);
1037         } else {
1038                 lat_print_generic(s, entry, cpu);
1039                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1040         }
1041         switch (entry->type) {
1042         case TRACE_FN:
1043                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1044                 trace_seq_puts(s, " (");
1045                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1046                 trace_seq_puts(s, ")\n");
1047                 break;
1048         case TRACE_CTX:
1049                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1050                         state_to_char[entry->ctx.prev_state] : 'X';
1051                 comm = trace_find_cmdline(entry->ctx.next_pid);
1052                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1053                                  entry->ctx.prev_pid,
1054                                  entry->ctx.prev_prio,
1055                                  S,
1056                                  entry->ctx.next_pid,
1057                                  entry->ctx.next_prio,
1058                                  comm);
1059                 break;
1060         default:
1061                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1062         }
1063 }
1064
1065 static notrace void sync_time_offset(struct trace_iterator *iter)
1066 {
1067         struct trace_array_cpu *prev_array, *array;
1068         struct trace_entry *prev_entry, *entry;
1069         cycle_t prev_t, t;
1070
1071         entry = iter->ent;
1072         prev_entry = iter->prev_ent;
1073         if (!prev_entry)
1074                 return;
1075
1076         prev_array = iter->tr->data[iter->prev_cpu];
1077         array = iter->tr->data[iter->cpu];
1078
1079         prev_t = prev_entry->t + prev_array->time_offset;
1080         t = entry->t + array->time_offset;
1081
1082         /*
1083          * If time goes backwards we increase the offset of
1084          * the current array, to not have observable time warps.
1085          * This will quickly synchronize the time offsets of
1086          * multiple CPUs:
1087          */
1088         if (t < prev_t)
1089                 array->time_offset += prev_t - t;
1090 }
1091
1092 static notrace void
1093 print_trace_fmt(struct trace_iterator *iter)
1094 {
1095         struct trace_seq *s = &iter->seq;
1096         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1097         struct trace_entry *entry;
1098         unsigned long usec_rem;
1099         unsigned long long t;
1100         unsigned long secs;
1101         char *comm;
1102         int S;
1103
1104         sync_time_offset(iter);
1105         entry = iter->ent;
1106
1107         comm = trace_find_cmdline(iter->ent->pid);
1108
1109         t = ns2usecs(entry->t + iter->tr->data[iter->cpu]->time_offset);
1110         usec_rem = do_div(t, 1000000ULL);
1111         secs = (unsigned long)t;
1112
1113         trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1114         trace_seq_printf(s, "[%02d] ", iter->cpu);
1115         trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1116
1117         switch (entry->type) {
1118         case TRACE_FN:
1119                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1120                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1121                                                 entry->fn.parent_ip) {
1122                         trace_seq_printf(s, " <-");
1123                         seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1124                 }
1125                 trace_seq_printf(s, "\n");
1126                 break;
1127         case TRACE_CTX:
1128                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1129                         state_to_char[entry->ctx.prev_state] : 'X';
1130                 trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1131                                  entry->ctx.prev_pid,
1132                                  entry->ctx.prev_prio,
1133                                  S,
1134                                  entry->ctx.next_pid,
1135                                  entry->ctx.next_prio);
1136                 break;
1137         }
1138 }
1139
1140 static int trace_empty(struct trace_iterator *iter)
1141 {
1142         struct trace_array_cpu *data;
1143         int cpu;
1144
1145         for_each_possible_cpu(cpu) {
1146                 data = iter->tr->data[cpu];
1147
1148                 if (head_page(data) && data->trace_idx)
1149                         return 0;
1150         }
1151         return 1;
1152 }
1153
1154 static int s_show(struct seq_file *m, void *v)
1155 {
1156         struct trace_iterator *iter = v;
1157
1158         if (iter->ent == NULL) {
1159                 if (iter->tr) {
1160                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1161                         seq_puts(m, "#\n");
1162                 }
1163                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1164                         /* print nothing if the buffers are empty */
1165                         if (trace_empty(iter))
1166                                 return 0;
1167                         print_trace_header(m, iter);
1168                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1169                                 print_lat_help_header(m);
1170                 } else {
1171                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1172                                 print_func_help_header(m);
1173                 }
1174         } else {
1175                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1176                         print_lat_fmt(iter, iter->idx, iter->cpu);
1177                 else
1178                         print_trace_fmt(iter);
1179                 trace_print_seq(m, &iter->seq);
1180         }
1181
1182         return 0;
1183 }
1184
1185 static struct seq_operations tracer_seq_ops = {
1186         .start = s_start,
1187         .next = s_next,
1188         .stop = s_stop,
1189         .show = s_show,
1190 };
1191
1192 static struct trace_iterator notrace *
1193 __tracing_open(struct inode *inode, struct file *file, int *ret)
1194 {
1195         struct trace_iterator *iter;
1196
1197         if (tracing_disabled) {
1198                 *ret = -ENODEV;
1199                 return NULL;
1200         }
1201
1202         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1203         if (!iter) {
1204                 *ret = -ENOMEM;
1205                 goto out;
1206         }
1207
1208         mutex_lock(&trace_types_lock);
1209         if (current_trace && current_trace->print_max)
1210                 iter->tr = &max_tr;
1211         else
1212                 iter->tr = inode->i_private;
1213         iter->trace = current_trace;
1214         iter->pos = -1;
1215
1216         /* TODO stop tracer */
1217         *ret = seq_open(file, &tracer_seq_ops);
1218         if (!*ret) {
1219                 struct seq_file *m = file->private_data;
1220                 m->private = iter;
1221
1222                 /* stop the trace while dumping */
1223                 if (iter->tr->ctrl)
1224                         tracer_enabled = 0;
1225
1226                 if (iter->trace && iter->trace->open)
1227                         iter->trace->open(iter);
1228         } else {
1229                 kfree(iter);
1230                 iter = NULL;
1231         }
1232         mutex_unlock(&trace_types_lock);
1233
1234  out:
1235         return iter;
1236 }
1237
1238 int tracing_open_generic(struct inode *inode, struct file *filp)
1239 {
1240         if (tracing_disabled)
1241                 return -ENODEV;
1242
1243         filp->private_data = inode->i_private;
1244         return 0;
1245 }
1246
1247 int tracing_release(struct inode *inode, struct file *file)
1248 {
1249         struct seq_file *m = (struct seq_file *)file->private_data;
1250         struct trace_iterator *iter = m->private;
1251
1252         mutex_lock(&trace_types_lock);
1253         if (iter->trace && iter->trace->close)
1254                 iter->trace->close(iter);
1255
1256         /* reenable tracing if it was previously enabled */
1257         if (iter->tr->ctrl)
1258                 tracer_enabled = 1;
1259         mutex_unlock(&trace_types_lock);
1260
1261         seq_release(inode, file);
1262         kfree(iter);
1263         return 0;
1264 }
1265
1266 static int tracing_open(struct inode *inode, struct file *file)
1267 {
1268         int ret;
1269
1270         __tracing_open(inode, file, &ret);
1271
1272         return ret;
1273 }
1274
1275 static int tracing_lt_open(struct inode *inode, struct file *file)
1276 {
1277         struct trace_iterator *iter;
1278         int ret;
1279
1280         iter = __tracing_open(inode, file, &ret);
1281
1282         if (!ret)
1283                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1284
1285         return ret;
1286 }
1287
1288
1289 static notrace void *
1290 t_next(struct seq_file *m, void *v, loff_t *pos)
1291 {
1292         struct tracer *t = m->private;
1293
1294         (*pos)++;
1295
1296         if (t)
1297                 t = t->next;
1298
1299         m->private = t;
1300
1301         return t;
1302 }
1303
1304 static void *t_start(struct seq_file *m, loff_t *pos)
1305 {
1306         struct tracer *t = m->private;
1307         loff_t l = 0;
1308
1309         mutex_lock(&trace_types_lock);
1310         for (; t && l < *pos; t = t_next(m, t, &l))
1311                 ;
1312
1313         return t;
1314 }
1315
1316 static void t_stop(struct seq_file *m, void *p)
1317 {
1318         mutex_unlock(&trace_types_lock);
1319 }
1320
1321 static int t_show(struct seq_file *m, void *v)
1322 {
1323         struct tracer *t = v;
1324
1325         if (!t)
1326                 return 0;
1327
1328         seq_printf(m, "%s", t->name);
1329         if (t->next)
1330                 seq_putc(m, ' ');
1331         else
1332                 seq_putc(m, '\n');
1333
1334         return 0;
1335 }
1336
1337 static struct seq_operations show_traces_seq_ops = {
1338         .start = t_start,
1339         .next = t_next,
1340         .stop = t_stop,
1341         .show = t_show,
1342 };
1343
1344 static int show_traces_open(struct inode *inode, struct file *file)
1345 {
1346         int ret;
1347
1348         if (tracing_disabled)
1349                 return -ENODEV;
1350
1351         ret = seq_open(file, &show_traces_seq_ops);
1352         if (!ret) {
1353                 struct seq_file *m = file->private_data;
1354                 m->private = trace_types;
1355         }
1356
1357         return ret;
1358 }
1359
1360 static struct file_operations tracing_fops = {
1361         .open = tracing_open,
1362         .read = seq_read,
1363         .llseek = seq_lseek,
1364         .release = tracing_release,
1365 };
1366
1367 static struct file_operations tracing_lt_fops = {
1368         .open = tracing_lt_open,
1369         .read = seq_read,
1370         .llseek = seq_lseek,
1371         .release = tracing_release,
1372 };
1373
1374 static struct file_operations show_traces_fops = {
1375         .open = show_traces_open,
1376         .read = seq_read,
1377         .release = seq_release,
1378 };
1379
1380 static ssize_t
1381 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1382                        size_t cnt, loff_t *ppos)
1383 {
1384         char *buf;
1385         int r = 0;
1386         int len = 0;
1387         int i;
1388
1389         /* calulate max size */
1390         for (i = 0; trace_options[i]; i++) {
1391                 len += strlen(trace_options[i]);
1392                 len += 3; /* "no" and space */
1393         }
1394
1395         /* +2 for \n and \0 */
1396         buf = kmalloc(len + 2, GFP_KERNEL);
1397         if (!buf)
1398                 return -ENOMEM;
1399
1400         for (i = 0; trace_options[i]; i++) {
1401                 if (trace_flags & (1 << i))
1402                         r += sprintf(buf + r, "%s ", trace_options[i]);
1403                 else
1404                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1405         }
1406
1407         r += sprintf(buf + r, "\n");
1408         WARN_ON(r >= len + 2);
1409
1410         r = simple_read_from_buffer(ubuf, cnt, ppos,
1411                                     buf, r);
1412
1413         kfree(buf);
1414
1415         return r;
1416 }
1417
1418 static ssize_t
1419 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1420                         size_t cnt, loff_t *ppos)
1421 {
1422         char buf[64];
1423         char *cmp = buf;
1424         int neg = 0;
1425         int i;
1426
1427         if (cnt > 63)
1428                 cnt = 63;
1429
1430         if (copy_from_user(&buf, ubuf, cnt))
1431                 return -EFAULT;
1432
1433         buf[cnt] = 0;
1434
1435         if (strncmp(buf, "no", 2) == 0) {
1436                 neg = 1;
1437                 cmp += 2;
1438         }
1439
1440         for (i = 0; trace_options[i]; i++) {
1441                 int len = strlen(trace_options[i]);
1442
1443                 if (strncmp(cmp, trace_options[i], len) == 0) {
1444                         if (neg)
1445                                 trace_flags &= ~(1 << i);
1446                         else
1447                                 trace_flags |= (1 << i);
1448                         break;
1449                 }
1450         }
1451
1452         filp->f_pos += cnt;
1453
1454         return cnt;
1455 }
1456
1457 static struct file_operations tracing_iter_fops = {
1458         .open = tracing_open_generic,
1459         .read = tracing_iter_ctrl_read,
1460         .write = tracing_iter_ctrl_write,
1461 };
1462
1463 static const char readme_msg[] =
1464         "tracing mini-HOWTO:\n\n"
1465         "# mkdir /debug\n"
1466         "# mount -t debugfs nodev /debug\n\n"
1467         "# cat /debug/tracing/available_tracers\n"
1468         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1469         "# cat /debug/tracing/current_tracer\n"
1470         "none\n"
1471         "# echo sched_switch > /debug/tracing/current_tracer\n"
1472         "# cat /debug/tracing/current_tracer\n"
1473         "sched_switch\n"
1474         "# cat /debug/tracing/iter_ctrl\n"
1475         "noprint-parent nosym-offset nosym-addr noverbose\n"
1476         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1477         "# echo 1 > /debug/tracing/tracing_enabled\n"
1478         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1479         "echo 0 > /debug/tracing/tracing_enabled\n"
1480 ;
1481
1482 static ssize_t
1483 tracing_readme_read(struct file *filp, char __user *ubuf,
1484                        size_t cnt, loff_t *ppos)
1485 {
1486         return simple_read_from_buffer(ubuf, cnt, ppos,
1487                                         readme_msg, strlen(readme_msg));
1488 }
1489
1490 static struct file_operations tracing_readme_fops = {
1491         .open = tracing_open_generic,
1492         .read = tracing_readme_read,
1493 };
1494
1495
1496 static ssize_t
1497 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1498                   size_t cnt, loff_t *ppos)
1499 {
1500         struct trace_array *tr = filp->private_data;
1501         char buf[64];
1502         int r;
1503
1504         r = sprintf(buf, "%ld\n", tr->ctrl);
1505         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1506 }
1507
1508 static ssize_t
1509 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1510                    size_t cnt, loff_t *ppos)
1511 {
1512         struct trace_array *tr = filp->private_data;
1513         long val;
1514         char buf[64];
1515
1516         if (cnt > 63)
1517                 cnt = 63;
1518
1519         if (copy_from_user(&buf, ubuf, cnt))
1520                 return -EFAULT;
1521
1522         buf[cnt] = 0;
1523
1524         val = simple_strtoul(buf, NULL, 10);
1525
1526         val = !!val;
1527
1528         mutex_lock(&trace_types_lock);
1529         if (tr->ctrl ^ val) {
1530                 if (val)
1531                         tracer_enabled = 1;
1532                 else
1533                         tracer_enabled = 0;
1534
1535                 tr->ctrl = val;
1536
1537                 if (current_trace && current_trace->ctrl_update)
1538                         current_trace->ctrl_update(tr);
1539         }
1540         mutex_unlock(&trace_types_lock);
1541
1542         filp->f_pos += cnt;
1543
1544         return cnt;
1545 }
1546
1547 static ssize_t
1548 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1549                        size_t cnt, loff_t *ppos)
1550 {
1551         char buf[max_tracer_type_len+2];
1552         int r;
1553
1554         mutex_lock(&trace_types_lock);
1555         if (current_trace)
1556                 r = sprintf(buf, "%s\n", current_trace->name);
1557         else
1558                 r = sprintf(buf, "\n");
1559         mutex_unlock(&trace_types_lock);
1560
1561         return simple_read_from_buffer(ubuf, cnt, ppos,
1562                                        buf, r);
1563 }
1564
1565 static ssize_t
1566 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1567                         size_t cnt, loff_t *ppos)
1568 {
1569         struct trace_array *tr = &global_trace;
1570         struct tracer *t;
1571         char buf[max_tracer_type_len+1];
1572         int i;
1573
1574         if (cnt > max_tracer_type_len)
1575                 cnt = max_tracer_type_len;
1576
1577         if (copy_from_user(&buf, ubuf, cnt))
1578                 return -EFAULT;
1579
1580         buf[cnt] = 0;
1581
1582         /* strip ending whitespace. */
1583         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1584                 buf[i] = 0;
1585
1586         mutex_lock(&trace_types_lock);
1587         for (t = trace_types; t; t = t->next) {
1588                 if (strcmp(t->name, buf) == 0)
1589                         break;
1590         }
1591         if (!t || t == current_trace)
1592                 goto out;
1593
1594         if (current_trace && current_trace->reset)
1595                 current_trace->reset(tr);
1596
1597         current_trace = t;
1598         if (t->init)
1599                 t->init(tr);
1600
1601  out:
1602         mutex_unlock(&trace_types_lock);
1603
1604         filp->f_pos += cnt;
1605
1606         return cnt;
1607 }
1608
1609 static ssize_t
1610 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1611                      size_t cnt, loff_t *ppos)
1612 {
1613         unsigned long *ptr = filp->private_data;
1614         char buf[64];
1615         int r;
1616
1617         r = snprintf(buf, 64, "%ld\n",
1618                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1619         if (r > 64)
1620                 r = 64;
1621         return simple_read_from_buffer(ubuf, cnt, ppos,
1622                                        buf, r);
1623 }
1624
1625 static ssize_t
1626 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1627                       size_t cnt, loff_t *ppos)
1628 {
1629         long *ptr = filp->private_data;
1630         long val;
1631         char buf[64];
1632
1633         if (cnt > 63)
1634                 cnt = 63;
1635
1636         if (copy_from_user(&buf, ubuf, cnt))
1637                 return -EFAULT;
1638
1639         buf[cnt] = 0;
1640
1641         val = simple_strtoul(buf, NULL, 10);
1642
1643         *ptr = val * 1000;
1644
1645         return cnt;
1646 }
1647
1648 static struct file_operations tracing_max_lat_fops = {
1649         .open = tracing_open_generic,
1650         .read = tracing_max_lat_read,
1651         .write = tracing_max_lat_write,
1652 };
1653
1654 static struct file_operations tracing_ctrl_fops = {
1655         .open = tracing_open_generic,
1656         .read = tracing_ctrl_read,
1657         .write = tracing_ctrl_write,
1658 };
1659
1660 static struct file_operations set_tracer_fops = {
1661         .open = tracing_open_generic,
1662         .read = tracing_set_trace_read,
1663         .write = tracing_set_trace_write,
1664 };
1665
1666 #ifdef CONFIG_DYNAMIC_FTRACE
1667
1668 static ssize_t
1669 tracing_read_long(struct file *filp, char __user *ubuf,
1670                   size_t cnt, loff_t *ppos)
1671 {
1672         unsigned long *p = filp->private_data;
1673         char buf[64];
1674         int r;
1675
1676         r = sprintf(buf, "%ld\n", *p);
1677         return simple_read_from_buffer(ubuf, cnt, ppos,
1678                                        buf, r);
1679 }
1680
1681 static struct file_operations tracing_read_long_fops = {
1682         .open = tracing_open_generic,
1683         .read = tracing_read_long,
1684 };
1685 #endif
1686
1687 static struct dentry *d_tracer;
1688
1689 struct dentry *tracing_init_dentry(void)
1690 {
1691         static int once;
1692
1693         if (d_tracer)
1694                 return d_tracer;
1695
1696         d_tracer = debugfs_create_dir("tracing", NULL);
1697
1698         if (!d_tracer && !once) {
1699                 once = 1;
1700                 pr_warning("Could not create debugfs directory 'tracing'\n");
1701                 return NULL;
1702         }
1703
1704         return d_tracer;
1705 }
1706
1707 #ifdef CONFIG_FTRACE_SELFTEST
1708 /* Let selftest have access to static functions in this file */
1709 #include "trace_selftest.c"
1710 #endif
1711
1712 static __init void tracer_init_debugfs(void)
1713 {
1714         struct dentry *d_tracer;
1715         struct dentry *entry;
1716
1717         d_tracer = tracing_init_dentry();
1718
1719         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
1720                                     &global_trace, &tracing_ctrl_fops);
1721         if (!entry)
1722                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
1723
1724         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
1725                                     NULL, &tracing_iter_fops);
1726         if (!entry)
1727                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
1728
1729         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
1730                                     &global_trace, &tracing_lt_fops);
1731         if (!entry)
1732                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
1733
1734         entry = debugfs_create_file("trace", 0444, d_tracer,
1735                                     &global_trace, &tracing_fops);
1736         if (!entry)
1737                 pr_warning("Could not create debugfs 'trace' entry\n");
1738
1739         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
1740                                     &global_trace, &show_traces_fops);
1741         if (!entry)
1742                 pr_warning("Could not create debugfs 'trace' entry\n");
1743
1744         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
1745                                     &global_trace, &set_tracer_fops);
1746         if (!entry)
1747                 pr_warning("Could not create debugfs 'trace' entry\n");
1748
1749         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
1750                                     &tracing_max_latency,
1751                                     &tracing_max_lat_fops);
1752         if (!entry)
1753                 pr_warning("Could not create debugfs "
1754                            "'tracing_max_latency' entry\n");
1755
1756         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
1757                                     &tracing_thresh, &tracing_max_lat_fops);
1758         if (!entry)
1759                 pr_warning("Could not create debugfs "
1760                            "'tracing_threash' entry\n");
1761         entry = debugfs_create_file("README", 0644, d_tracer,
1762                                     NULL, &tracing_readme_fops);
1763         if (!entry)
1764                 pr_warning("Could not create debugfs 'README' entry\n");
1765
1766
1767 #ifdef CONFIG_DYNAMIC_FTRACE
1768         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
1769                                     &ftrace_update_tot_cnt,
1770                                     &tracing_read_long_fops);
1771         if (!entry)
1772                 pr_warning("Could not create debugfs "
1773                            "'dyn_ftrace_total_info' entry\n");
1774 #endif
1775 }
1776
1777 /* dummy trace to disable tracing */
1778 static struct tracer no_tracer __read_mostly =
1779 {
1780         .name = "none",
1781 };
1782
1783 static int trace_alloc_page(void)
1784 {
1785         struct trace_array_cpu *data;
1786         struct page *page, *tmp;
1787         LIST_HEAD(pages);
1788         void *array;
1789         int i;
1790
1791         /* first allocate a page for each CPU */
1792         for_each_possible_cpu(i) {
1793                 array = (void *)__get_free_page(GFP_KERNEL);
1794                 if (array == NULL) {
1795                         printk(KERN_ERR "tracer: failed to allocate page"
1796                                "for trace buffer!\n");
1797                         goto free_pages;
1798                 }
1799
1800                 page = virt_to_page(array);
1801                 list_add(&page->lru, &pages);
1802
1803 /* Only allocate if we are actually using the max trace */
1804 #ifdef CONFIG_TRACER_MAX_TRACE
1805                 array = (void *)__get_free_page(GFP_KERNEL);
1806                 if (array == NULL) {
1807                         printk(KERN_ERR "tracer: failed to allocate page"
1808                                "for trace buffer!\n");
1809                         goto free_pages;
1810                 }
1811                 page = virt_to_page(array);
1812                 list_add(&page->lru, &pages);
1813 #endif
1814         }
1815
1816         /* Now that we successfully allocate a page per CPU, add them */
1817         for_each_possible_cpu(i) {
1818                 data = global_trace.data[i];
1819                 page = list_entry(pages.next, struct page, lru);
1820                 list_del_init(&page->lru);
1821                 list_add_tail(&page->lru, &data->trace_pages);
1822                 ClearPageLRU(page);
1823
1824 #ifdef CONFIG_TRACER_MAX_TRACE
1825                 data = max_tr.data[i];
1826                 page = list_entry(pages.next, struct page, lru);
1827                 list_del_init(&page->lru);
1828                 list_add_tail(&page->lru, &data->trace_pages);
1829                 SetPageLRU(page);
1830 #endif
1831         }
1832         global_trace.entries += ENTRIES_PER_PAGE;
1833
1834         return 0;
1835
1836  free_pages:
1837         list_for_each_entry_safe(page, tmp, &pages, lru) {
1838                 list_del_init(&page->lru);
1839                 __free_page(page);
1840         }
1841         return -ENOMEM;
1842 }
1843
1844 __init static int tracer_alloc_buffers(void)
1845 {
1846         struct trace_array_cpu *data;
1847         void *array;
1848         struct page *page;
1849         int pages = 0;
1850         int ret = -ENOMEM;
1851         int i;
1852
1853         /* Allocate the first page for all buffers */
1854         for_each_possible_cpu(i) {
1855                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
1856                 max_tr.data[i] = &per_cpu(max_data, i);
1857
1858                 array = (void *)__get_free_page(GFP_KERNEL);
1859                 if (array == NULL) {
1860                         printk(KERN_ERR "tracer: failed to allocate page"
1861                                "for trace buffer!\n");
1862                         goto free_buffers;
1863                 }
1864
1865                 /* set the array to the list */
1866                 INIT_LIST_HEAD(&data->trace_pages);
1867                 page = virt_to_page(array);
1868                 list_add(&page->lru, &data->trace_pages);
1869                 /* use the LRU flag to differentiate the two buffers */
1870                 ClearPageLRU(page);
1871
1872 /* Only allocate if we are actually using the max trace */
1873 #ifdef CONFIG_TRACER_MAX_TRACE
1874                 array = (void *)__get_free_page(GFP_KERNEL);
1875                 if (array == NULL) {
1876                         printk(KERN_ERR "tracer: failed to allocate page"
1877                                "for trace buffer!\n");
1878                         goto free_buffers;
1879                 }
1880
1881                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
1882                 page = virt_to_page(array);
1883                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
1884                 SetPageLRU(page);
1885 #endif
1886         }
1887
1888         /*
1889          * Since we allocate by orders of pages, we may be able to
1890          * round up a bit.
1891          */
1892         global_trace.entries = ENTRIES_PER_PAGE;
1893         pages++;
1894
1895         while (global_trace.entries < trace_nr_entries) {
1896                 if (trace_alloc_page())
1897                         break;
1898                 pages++;
1899         }
1900         max_tr.entries = global_trace.entries;
1901
1902         pr_info("tracer: %d pages allocated for %ld",
1903                 pages, trace_nr_entries);
1904         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
1905         pr_info("   actual entries %ld\n", global_trace.entries);
1906
1907         tracer_init_debugfs();
1908
1909         trace_init_cmdlines();
1910
1911         register_tracer(&no_tracer);
1912         current_trace = &no_tracer;
1913
1914         /* All seems OK, enable tracing */
1915         tracing_disabled = 0;
1916
1917         return 0;
1918
1919  free_buffers:
1920         for (i-- ; i >= 0; i--) {
1921                 struct page *page, *tmp;
1922                 struct trace_array_cpu *data = global_trace.data[i];
1923
1924                 if (data) {
1925                         list_for_each_entry_safe(page, tmp,
1926                                                  &data->trace_pages, lru) {
1927                                 list_del_init(&page->lru);
1928                                 __free_page(page);
1929                         }
1930                 }
1931
1932 #ifdef CONFIG_TRACER_MAX_TRACE
1933                 data = max_tr.data[i];
1934                 if (data) {
1935                         list_for_each_entry_safe(page, tmp,
1936                                                  &data->trace_pages, lru) {
1937                                 list_del_init(&page->lru);
1938                                 __free_page(page);
1939                         }
1940                 }
1941 #endif
1942         }
1943         return ret;
1944 }
1945
1946 fs_initcall(tracer_alloc_buffers);