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 const int time_sync_freq_max = 128;
  46 static const cycle_t time_sync_thresh = 100000;
  47
  48 static DEFINE_PER_CPU(cycle_t, time_offset);
  49 static DEFINE_PER_CPU(cycle_t, prev_cpu_time);
  50 static DEFINE_PER_CPU(int, time_sync_count);
  51 static DEFINE_PER_CPU(int, time_sync_freq);
  52
  53 /*
  54  * Global lock which we take every now and then to synchronize
  55  * the CPUs time. This method is not warp-safe, but it's good
  56  * enough to synchronize slowly diverging time sources and thus
  57  * it's good enough for tracing:
  58  */
  59 static DEFINE_SPINLOCK(time_sync_lock);
  60 static cycle_t prev_global_time;
  61
  62 static notrace cycle_t __ftrace_now_sync(cycles_t time, int cpu)
  63 {
  64         unsigned long flags;
  65
  66         spin_lock_irqsave(&time_sync_lock, flags);
  67
  68         /*
  69          * Update the synchronization frequency:
  70          */
  71         if (per_cpu(time_sync_freq, cpu) < time_sync_freq_max)
  72                 per_cpu(time_sync_freq, cpu) *= 2;
  73         per_cpu(time_sync_count, cpu) = per_cpu(time_sync_freq, cpu);
  74
  75         if (time < prev_global_time) {
  76                 per_cpu(time_offset, cpu) += prev_global_time - time;
  77                 time = prev_global_time;
  78         } else {
  79                 prev_global_time = time;
  80         }
  81
  82         spin_unlock_irqrestore(&time_sync_lock, flags);
  83
  84         return time;
  85 }
  86
  87 notrace cycle_t ftrace_now(int cpu)
  88 {
  89         cycle_t prev_cpu_time, time, delta_time;
  90
  91         prev_cpu_time = per_cpu(prev_cpu_time, cpu);
  92         time = sched_clock() + per_cpu(time_offset, cpu);
  93         delta_time = time-prev_cpu_time;
  94
  95         if (unlikely(delta_time > time_sync_thresh ||
  96                                 --per_cpu(time_sync_count, cpu) <= 0))
  97                 time = __ftrace_now_sync(time, cpu);
  98
  99         return time;
 100 }
 101
 102 static struct trace_array       global_trace;
 103
 104 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
 105
 106 static struct trace_array       max_tr;
 107
 108 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
 109
 110 static int                      tracer_enabled = 1;
 111 static unsigned long            trace_nr_entries = 16384UL;
 112
 113 static struct tracer            *trace_types __read_mostly;
 114 static struct tracer            *current_trace __read_mostly;
 115 static int                      max_tracer_type_len;
 116
 117 static DEFINE_MUTEX(trace_types_lock);
 118
 119 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
 120
 121 static int __init set_nr_entries(char *str)
 122 {
 123         if (!str)
 124                 return 0;
 125         trace_nr_entries = simple_strtoul(str, &str, 0);
 126         return 1;
 127 }
 128 __setup("trace_entries=", set_nr_entries);
 129
 130 unsigned long nsecs_to_usecs(unsigned long nsecs)
 131 {
 132         return nsecs / 1000;
 133 }
 134
 135 enum trace_type {
 136         __TRACE_FIRST_TYPE = 0,
 137
 138         TRACE_FN,
 139         TRACE_CTX,
 140         TRACE_SPECIAL,
 141
 142         __TRACE_LAST_TYPE
 143 };
 144
 145 enum trace_flag_type {
 146         TRACE_FLAG_IRQS_OFF             = 0x01,
 147         TRACE_FLAG_NEED_RESCHED         = 0x02,
 148         TRACE_FLAG_HARDIRQ              = 0x04,
 149         TRACE_FLAG_SOFTIRQ              = 0x08,
 150 };
 151
 152 enum trace_iterator_flags {
 153         TRACE_ITER_PRINT_PARENT         = 0x01,
 154         TRACE_ITER_SYM_OFFSET           = 0x02,
 155         TRACE_ITER_SYM_ADDR             = 0x04,
 156         TRACE_ITER_VERBOSE              = 0x08,
 157         TRACE_ITER_RAW                  = 0x10,
 158         TRACE_ITER_BIN                  = 0x20,
 159 };
 160
 161 #define TRACE_ITER_SYM_MASK \
 162         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 163
 164 /* These must match the bit postions above */
 165 static const char *trace_options[] = {
 166         "print-parent",
 167         "sym-offset",
 168         "sym-addr",
 169         "verbose",
 170         "raw",
 171         "bin",
 172         NULL
 173 };
 174
 175 static unsigned trace_flags;
 176
 177 static DEFINE_SPINLOCK(ftrace_max_lock);
 178
 179 /*
 180  * Copy the new maximum trace into the separate maximum-trace
 181  * structure. (this way the maximum trace is permanently saved,
 182  * for later retrieval via /debugfs/tracing/latency_trace)
 183  */
 184 static notrace void
 185 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 186 {
 187         struct trace_array_cpu *data = tr->data[cpu];
 188
 189         max_tr.cpu = cpu;
 190         max_tr.time_start = data->preempt_timestamp;
 191
 192         data = max_tr.data[cpu];
 193         data->saved_latency = tracing_max_latency;
 194
 195         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 196         data->pid = tsk->pid;
 197         data->uid = tsk->uid;
 198         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 199         data->policy = tsk->policy;
 200         data->rt_priority = tsk->rt_priority;
 201
 202         /* record this tasks comm */
 203         tracing_record_cmdline(current);
 204 }
 205
 206 void check_pages(struct trace_array_cpu *data)
 207 {
 208         struct page *page, *tmp;
 209
 210         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 211         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 212
 213         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 214                 BUG_ON(page->lru.next->prev != &page->lru);
 215                 BUG_ON(page->lru.prev->next != &page->lru);
 216         }
 217 }
 218
 219 void *head_page(struct trace_array_cpu *data)
 220 {
 221         struct page *page;
 222
 223         check_pages(data);
 224         if (list_empty(&data->trace_pages))
 225                 return NULL;
 226
 227         page = list_entry(data->trace_pages.next, struct page, lru);
 228         BUG_ON(&page->lru == &data->trace_pages);
 229
 230         return page_address(page);
 231 }
 232
 233 static notrace int
 234 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 235 {
 236         int len = (PAGE_SIZE - 1) - s->len;
 237         va_list ap;
 238         int ret;
 239
 240         if (!len)
 241                 return 0;
 242
 243         va_start(ap, fmt);
 244         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 245         va_end(ap);
 246
 247         /* If we can't write it all, don't bother writing anything */
 248         if (ret > len)
 249                 return 0;
 250
 251         s->len += ret;
 252
 253         return len;
 254 }
 255
 256 static notrace int
 257 trace_seq_puts(struct trace_seq *s, const char *str)
 258 {
 259         int len = strlen(str);
 260
 261         if (len > ((PAGE_SIZE - 1) - s->len))
 262                 return 0;
 263
 264         memcpy(s->buffer + s->len, str, len);
 265         s->len += len;
 266
 267         return len;
 268 }
 269
 270 static notrace int
 271 trace_seq_putc(struct trace_seq *s, unsigned char c)
 272 {
 273         if (s->len >= (PAGE_SIZE - 1))
 274                 return 0;
 275
 276         s->buffer[s->len++] = c;
 277
 278         return 1;
 279 }
 280
 281 static notrace int
 282 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 283 {
 284         if (len > ((PAGE_SIZE - 1) - s->len))
 285                 return 0;
 286
 287         memcpy(s->buffer + s->len, mem, len);
 288         s->len += len;
 289
 290         return len;
 291 }
 292
 293 static notrace void
 294 trace_seq_reset(struct trace_seq *s)
 295 {
 296         s->len = 0;
 297 }
 298
 299 static notrace void
 300 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 301 {
 302         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 303
 304         s->buffer[len] = 0;
 305         seq_puts(m, s->buffer);
 306
 307         trace_seq_reset(s);
 308 }
 309
 310 notrace static void
 311 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 312 {
 313         struct list_head flip_pages;
 314
 315         INIT_LIST_HEAD(&flip_pages);
 316
 317         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 318                 sizeof(struct trace_array_cpu) -
 319                 offsetof(struct trace_array_cpu, trace_head_idx));
 320
 321         check_pages(tr1);
 322         check_pages(tr2);
 323         list_splice_init(&tr1->trace_pages, &flip_pages);
 324         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 325         list_splice_init(&flip_pages, &tr2->trace_pages);
 326         BUG_ON(!list_empty(&flip_pages));
 327         check_pages(tr1);
 328         check_pages(tr2);
 329 }
 330
 331 notrace void
 332 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 333 {
 334         struct trace_array_cpu *data;
 335         int i;
 336
 337         WARN_ON_ONCE(!irqs_disabled());
 338         spin_lock(&ftrace_max_lock);
 339         /* clear out all the previous traces */
 340         for_each_possible_cpu(i) {
 341                 data = tr->data[i];
 342                 flip_trace(max_tr.data[i], data);
 343                 tracing_reset(data);
 344         }
 345
 346         __update_max_tr(tr, tsk, cpu);
 347         spin_unlock(&ftrace_max_lock);
 348 }
 349
 350 /**
 351  * update_max_tr_single - only copy one trace over, and reset the rest
 352  * @tr - tracer
 353  * @tsk - task with the latency
 354  * @cpu - the cpu of the buffer to copy.
 355  */
 356 notrace void
 357 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 358 {
 359         struct trace_array_cpu *data = tr->data[cpu];
 360         int i;
 361
 362         WARN_ON_ONCE(!irqs_disabled());
 363         spin_lock(&ftrace_max_lock);
 364         for_each_possible_cpu(i)
 365                 tracing_reset(max_tr.data[i]);
 366
 367         flip_trace(max_tr.data[cpu], data);
 368         tracing_reset(data);
 369
 370         __update_max_tr(tr, tsk, cpu);
 371         spin_unlock(&ftrace_max_lock);
 372 }
 373
 374 int register_tracer(struct tracer *type)
 375 {
 376         struct tracer *t;
 377         int len;
 378         int ret = 0;
 379
 380         if (!type->name) {
 381                 pr_info("Tracer must have a name\n");
 382                 return -1;
 383         }
 384
 385         mutex_lock(&trace_types_lock);
 386         for (t = trace_types; t; t = t->next) {
 387                 if (strcmp(type->name, t->name) == 0) {
 388                         /* already found */
 389                         pr_info("Trace %s already registered\n",
 390                                 type->name);
 391                         ret = -1;
 392                         goto out;
 393                 }
 394         }
 395
 396 #ifdef CONFIG_FTRACE_STARTUP_TEST
 397         if (type->selftest) {
 398                 struct tracer *saved_tracer = current_trace;
 399                 struct trace_array_cpu *data;
 400                 struct trace_array *tr = &global_trace;
 401                 int saved_ctrl = tr->ctrl;
 402                 int i;
 403                 /*
 404                  * Run a selftest on this tracer.
 405                  * Here we reset the trace buffer, and set the current
 406                  * tracer to be this tracer. The tracer can then run some
 407                  * internal tracing to verify that everything is in order.
 408                  * If we fail, we do not register this tracer.
 409                  */
 410                 for_each_possible_cpu(i) {
 411                         data = tr->data[i];
 412                         if (!head_page(data))
 413                                 continue;
 414                         tracing_reset(data);
 415                 }
 416                 current_trace = type;
 417                 tr->ctrl = 0;
 418                 /* the test is responsible for initializing and enabling */
 419                 pr_info("Testing tracer %s: ", type->name);
 420                 ret = type->selftest(type, tr);
 421                 /* the test is responsible for resetting too */
 422                 current_trace = saved_tracer;
 423                 tr->ctrl = saved_ctrl;
 424                 if (ret) {
 425                         printk(KERN_CONT "FAILED!\n");
 426                         goto out;
 427                 }
 428                 /* Only reset on passing, to avoid touching corrupted buffers */
 429                 for_each_possible_cpu(i) {
 430                         data = tr->data[i];
 431                         if (!head_page(data))
 432                                 continue;
 433                         tracing_reset(data);
 434                 }
 435                 printk(KERN_CONT "PASSED\n");
 436         }
 437 #endif
 438
 439         type->next = trace_types;
 440         trace_types = type;
 441         len = strlen(type->name);
 442         if (len > max_tracer_type_len)
 443                 max_tracer_type_len = len;
 444
 445  out:
 446         mutex_unlock(&trace_types_lock);
 447
 448         return ret;
 449 }
 450
 451 void unregister_tracer(struct tracer *type)
 452 {
 453         struct tracer **t;
 454         int len;
 455
 456         mutex_lock(&trace_types_lock);
 457         for (t = &trace_types; *t; t = &(*t)->next) {
 458                 if (*t == type)
 459                         goto found;
 460         }
 461         pr_info("Trace %s not registered\n", type->name);
 462         goto out;
 463
 464  found:
 465         *t = (*t)->next;
 466         if (strlen(type->name) != max_tracer_type_len)
 467                 goto out;
 468
 469         max_tracer_type_len = 0;
 470         for (t = &trace_types; *t; t = &(*t)->next) {
 471                 len = strlen((*t)->name);
 472                 if (len > max_tracer_type_len)
 473                         max_tracer_type_len = len;
 474         }
 475  out:
 476         mutex_unlock(&trace_types_lock);
 477 }
 478
 479 notrace void tracing_reset(struct trace_array_cpu *data)
 480 {
 481         data->trace_idx = 0;
 482         data->trace_head = data->trace_tail = head_page(data);
 483         data->trace_head_idx = 0;
 484         data->trace_tail_idx = 0;
 485 }
 486
 487 #ifdef CONFIG_FTRACE
 488 static notrace void
 489 function_trace_call(unsigned long ip, unsigned long parent_ip)
 490 {
 491         struct trace_array *tr = &global_trace;
 492         struct trace_array_cpu *data;
 493         unsigned long flags;
 494         long disabled;
 495         int cpu;
 496
 497         if (unlikely(!tracer_enabled))
 498                 return;
 499
 500         local_irq_save(flags);
 501         cpu = raw_smp_processor_id();
 502         data = tr->data[cpu];
 503         disabled = atomic_inc_return(&data->disabled);
 504
 505         if (likely(disabled == 1))
 506                 ftrace(tr, data, ip, parent_ip, flags);
 507
 508         atomic_dec(&data->disabled);
 509         local_irq_restore(flags);
 510 }
 511
 512 static struct ftrace_ops trace_ops __read_mostly =
 513 {
 514         .func = function_trace_call,
 515 };
 516 #endif
 517
 518 notrace void tracing_start_function_trace(void)
 519 {
 520         register_ftrace_function(&trace_ops);
 521 }
 522
 523 notrace void tracing_stop_function_trace(void)
 524 {
 525         unregister_ftrace_function(&trace_ops);
 526 }
 527
 528 #define SAVED_CMDLINES 128
 529 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 530 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 531 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 532 static int cmdline_idx;
 533 static DEFINE_SPINLOCK(trace_cmdline_lock);
 534 atomic_t trace_record_cmdline_disabled;
 535
 536 static void trace_init_cmdlines(void)
 537 {
 538         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 539         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 540         cmdline_idx = 0;
 541 }
 542
 543 notrace void trace_stop_cmdline_recording(void);
 544
 545 static notrace void trace_save_cmdline(struct task_struct *tsk)
 546 {
 547         unsigned map;
 548         unsigned idx;
 549
 550         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 551                 return;
 552
 553         /*
 554          * It's not the end of the world if we don't get
 555          * the lock, but we also don't want to spin
 556          * nor do we want to disable interrupts,
 557          * so if we miss here, then better luck next time.
 558          */
 559         if (!spin_trylock(&trace_cmdline_lock))
 560                 return;
 561
 562         idx = map_pid_to_cmdline[tsk->pid];
 563         if (idx >= SAVED_CMDLINES) {
 564                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 565
 566                 map = map_cmdline_to_pid[idx];
 567                 if (map <= PID_MAX_DEFAULT)
 568                         map_pid_to_cmdline[map] = (unsigned)-1;
 569
 570                 map_pid_to_cmdline[tsk->pid] = idx;
 571
 572                 cmdline_idx = idx;
 573         }
 574
 575         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 576
 577         spin_unlock(&trace_cmdline_lock);
 578 }
 579
 580 static notrace char *trace_find_cmdline(int pid)
 581 {
 582         char *cmdline = "<...>";
 583         unsigned map;
 584
 585         if (!pid)
 586                 return "<idle>";
 587
 588         if (pid > PID_MAX_DEFAULT)
 589                 goto out;
 590
 591         map = map_pid_to_cmdline[pid];
 592         if (map >= SAVED_CMDLINES)
 593                 goto out;
 594
 595         cmdline = saved_cmdlines[map];
 596
 597  out:
 598         return cmdline;
 599 }
 600
 601 notrace void tracing_record_cmdline(struct task_struct *tsk)
 602 {
 603         if (atomic_read(&trace_record_cmdline_disabled))
 604                 return;
 605
 606         trace_save_cmdline(tsk);
 607 }
 608
 609 static inline notrace struct list_head *
 610 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 611 {
 612         /*
 613          * Roundrobin - but skip the head (which is not a real page):
 614          */
 615         next = next->next;
 616         if (unlikely(next == &data->trace_pages))
 617                 next = next->next;
 618         BUG_ON(next == &data->trace_pages);
 619
 620         return next;
 621 }
 622
 623 static inline notrace void *
 624 trace_next_page(struct trace_array_cpu *data, void *addr)
 625 {
 626         struct list_head *next;
 627         struct page *page;
 628
 629         page = virt_to_page(addr);
 630
 631         next = trace_next_list(data, &page->lru);
 632         page = list_entry(next, struct page, lru);
 633
 634         return page_address(page);
 635 }
 636
 637 static inline notrace struct trace_entry *
 638 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 639 {
 640         unsigned long idx, idx_next;
 641         struct trace_entry *entry;
 642
 643         data->trace_idx++;
 644         idx = data->trace_head_idx;
 645         idx_next = idx + 1;
 646
 647         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 648
 649         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 650
 651         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 652                 data->trace_head = trace_next_page(data, data->trace_head);
 653                 idx_next = 0;
 654         }
 655
 656         if (data->trace_head == data->trace_tail &&
 657             idx_next == data->trace_tail_idx) {
 658                 /* overrun */
 659                 data->trace_tail_idx++;
 660                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 661                         data->trace_tail =
 662                                 trace_next_page(data, data->trace_tail);
 663                         data->trace_tail_idx = 0;
 664                 }
 665         }
 666
 667         data->trace_head_idx = idx_next;
 668
 669         return entry;
 670 }
 671
 672 static inline notrace void
 673 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 674 {
 675         struct task_struct *tsk = current;
 676         unsigned long pc;
 677
 678         pc = preempt_count();
 679
 680         entry->preempt_count    = pc & 0xff;
 681         entry->pid              = tsk->pid;
 682         entry->t                = ftrace_now(raw_smp_processor_id());
 683         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 684                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 685                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 686                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 687 }
 688
 689 notrace void
 690 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 691        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 692 {
 693         struct trace_entry *entry;
 694         unsigned long irq_flags;
 695
 696         spin_lock_irqsave(&data->lock, irq_flags);
 697         entry                   = tracing_get_trace_entry(tr, data);
 698         tracing_generic_entry_update(entry, flags);
 699         entry->type             = TRACE_FN;
 700         entry->fn.ip            = ip;
 701         entry->fn.parent_ip     = parent_ip;
 702         spin_unlock_irqrestore(&data->lock, irq_flags);
 703 }
 704
 705 notrace void
 706 trace_special(struct trace_array *tr, struct trace_array_cpu *data,
 707               unsigned long arg1, unsigned long arg2, unsigned long arg3)
 708 {
 709         struct trace_entry *entry;
 710         unsigned long irq_flags;
 711
 712         spin_lock_irqsave(&data->lock, irq_flags);
 713         entry                   = tracing_get_trace_entry(tr, data);
 714         tracing_generic_entry_update(entry, 0);
 715         entry->type             = TRACE_SPECIAL;
 716         entry->special.arg1     = arg1;
 717         entry->special.arg2     = arg2;
 718         entry->special.arg3     = arg3;
 719         spin_unlock_irqrestore(&data->lock, irq_flags);
 720 }
 721
 722 notrace void
 723 tracing_sched_switch_trace(struct trace_array *tr,
 724                            struct trace_array_cpu *data,
 725                            struct task_struct *prev, struct task_struct *next,
 726                            unsigned long flags)
 727 {
 728         struct trace_entry *entry;
 729         unsigned long irq_flags;
 730
 731         spin_lock_irqsave(&data->lock, irq_flags);
 732         entry                   = tracing_get_trace_entry(tr, data);
 733         tracing_generic_entry_update(entry, flags);
 734         entry->type             = TRACE_CTX;
 735         entry->ctx.prev_pid     = prev->pid;
 736         entry->ctx.prev_prio    = prev->prio;
 737         entry->ctx.prev_state   = prev->state;
 738         entry->ctx.next_pid     = next->pid;
 739         entry->ctx.next_prio    = next->prio;
 740         spin_unlock_irqrestore(&data->lock, irq_flags);
 741 }
 742
 743 enum trace_file_type {
 744         TRACE_FILE_LAT_FMT      = 1,
 745 };
 746
 747 static struct trace_entry *
 748 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 749                 struct trace_iterator *iter, int cpu)
 750 {
 751         struct page *page;
 752         struct trace_entry *array;
 753
 754         if (iter->next_idx[cpu] >= tr->entries ||
 755             iter->next_idx[cpu] >= data->trace_idx ||
 756             (data->trace_head == data->trace_tail &&
 757              data->trace_head_idx == data->trace_tail_idx))
 758                 return NULL;
 759
 760         if (!iter->next_page[cpu]) {
 761                 /* Initialize the iterator for this cpu trace buffer */
 762                 WARN_ON(!data->trace_tail);
 763                 page = virt_to_page(data->trace_tail);
 764                 iter->next_page[cpu] = &page->lru;
 765                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 766         }
 767
 768         page = list_entry(iter->next_page[cpu], struct page, lru);
 769         BUG_ON(&data->trace_pages == &page->lru);
 770
 771         array = page_address(page);
 772
 773         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 774         return &array[iter->next_page_idx[cpu]];
 775 }
 776
 777 static struct notrace trace_entry *
 778 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 779 {
 780         struct trace_array *tr = iter->tr;
 781         struct trace_entry *ent, *next = NULL;
 782         int next_cpu = -1;
 783         int cpu;
 784
 785         for_each_possible_cpu(cpu) {
 786                 if (!head_page(tr->data[cpu]))
 787                         continue;
 788                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 789                 /*
 790                  * Pick the entry with the smallest timestamp:
 791                  */
 792                 if (ent && (!next || ent->t < next->t)) {
 793                         next = ent;
 794                         next_cpu = cpu;
 795                 }
 796         }
 797
 798         if (ent_cpu)
 799                 *ent_cpu = next_cpu;
 800
 801         return next;
 802 }
 803
 804 static notrace void trace_iterator_increment(struct trace_iterator *iter)
 805 {
 806         iter->idx++;
 807         iter->next_idx[iter->cpu]++;
 808         iter->next_page_idx[iter->cpu]++;
 809
 810         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 811                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 812
 813                 iter->next_page_idx[iter->cpu] = 0;
 814                 iter->next_page[iter->cpu] =
 815                         trace_next_list(data, iter->next_page[iter->cpu]);
 816         }
 817 }
 818
 819 static notrace void trace_consume(struct trace_iterator *iter)
 820 {
 821         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 822
 823         data->trace_tail_idx++;
 824         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 825                 data->trace_tail = trace_next_page(data, data->trace_tail);
 826                 data->trace_tail_idx = 0;
 827         }
 828
 829         /* Check if we empty it, then reset the index */
 830         if (data->trace_head == data->trace_tail &&
 831             data->trace_head_idx == data->trace_tail_idx)
 832                 data->trace_idx = 0;
 833 }
 834
 835 static notrace void *find_next_entry_inc(struct trace_iterator *iter)
 836 {
 837         struct trace_entry *next;
 838         int next_cpu = -1;
 839
 840         next = find_next_entry(iter, &next_cpu);
 841
 842         iter->prev_ent = iter->ent;
 843         iter->prev_cpu = iter->cpu;
 844
 845         iter->ent = next;
 846         iter->cpu = next_cpu;
 847
 848         if (next)
 849                 trace_iterator_increment(iter);
 850
 851         return next ? iter : NULL;
 852 }
 853
 854 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 855 {
 856         struct trace_iterator *iter = m->private;
 857         void *last_ent = iter->ent;
 858         int i = (int)*pos;
 859         void *ent;
 860
 861         (*pos)++;
 862
 863         /* can't go backwards */
 864         if (iter->idx > i)
 865                 return NULL;
 866
 867         if (iter->idx < 0)
 868                 ent = find_next_entry_inc(iter);
 869         else
 870                 ent = iter;
 871
 872         while (ent && iter->idx < i)
 873                 ent = find_next_entry_inc(iter);
 874
 875         iter->pos = *pos;
 876
 877         if (last_ent && !ent)
 878                 seq_puts(m, "\n\nvim:ft=help\n");
 879
 880         return ent;
 881 }
 882
 883 static void *s_start(struct seq_file *m, loff_t *pos)
 884 {
 885         struct trace_iterator *iter = m->private;
 886         void *p = NULL;
 887         loff_t l = 0;
 888         int i;
 889
 890         mutex_lock(&trace_types_lock);
 891
 892         if (!current_trace || current_trace != iter->trace)
 893                 return NULL;
 894
 895         atomic_inc(&trace_record_cmdline_disabled);
 896
 897         /* let the tracer grab locks here if needed */
 898         if (current_trace->start)
 899                 current_trace->start(iter);
 900
 901         if (*pos != iter->pos) {
 902                 iter->ent = NULL;
 903                 iter->cpu = 0;
 904                 iter->idx = -1;
 905                 iter->prev_ent = NULL;
 906                 iter->prev_cpu = -1;
 907
 908                 for_each_possible_cpu(i) {
 909                         iter->next_idx[i] = 0;
 910                         iter->next_page[i] = NULL;
 911                 }
 912
 913                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 914                         ;
 915
 916         } else {
 917                 l = *pos - 1;
 918                 p = s_next(m, p, &l);
 919         }
 920
 921         return p;
 922 }
 923
 924 static void s_stop(struct seq_file *m, void *p)
 925 {
 926         struct trace_iterator *iter = m->private;
 927
 928         atomic_dec(&trace_record_cmdline_disabled);
 929
 930         /* let the tracer release locks here if needed */
 931         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 932                 iter->trace->stop(iter);
 933
 934         mutex_unlock(&trace_types_lock);
 935 }
 936
 937 static int
 938 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 939 {
 940 #ifdef CONFIG_KALLSYMS
 941         char str[KSYM_SYMBOL_LEN];
 942
 943         kallsyms_lookup(address, NULL, NULL, NULL, str);
 944
 945         return trace_seq_printf(s, fmt, str);
 946 #endif
 947         return 1;
 948 }
 949
 950 static int
 951 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 952                      unsigned long address)
 953 {
 954 #ifdef CONFIG_KALLSYMS
 955         char str[KSYM_SYMBOL_LEN];
 956
 957         sprint_symbol(str, address);
 958         return trace_seq_printf(s, fmt, str);
 959 #endif
 960         return 1;
 961 }
 962
 963 #ifndef CONFIG_64BIT
 964 # define IP_FMT "%08lx"
 965 #else
 966 # define IP_FMT "%016lx"
 967 #endif
 968
 969 static notrace int
 970 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 971 {
 972         int ret;
 973
 974         if (!ip)
 975                 return trace_seq_printf(s, "0");
 976
 977         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 978                 ret = seq_print_sym_offset(s, "%s", ip);
 979         else
 980                 ret = seq_print_sym_short(s, "%s", ip);
 981
 982         if (!ret)
 983                 return 0;
 984
 985         if (sym_flags & TRACE_ITER_SYM_ADDR)
 986                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 987         return ret;
 988 }
 989
 990 static notrace void print_lat_help_header(struct seq_file *m)
 991 {
 992         seq_puts(m, "#                _------=> CPU#            \n");
 993         seq_puts(m, "#               / _-----=> irqs-off        \n");
 994         seq_puts(m, "#              | / _----=> need-resched    \n");
 995         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 996         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 997         seq_puts(m, "#              |||| /                      \n");
 998         seq_puts(m, "#              |||||     delay             \n");
 999         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1000         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1001 }
1002
1003 static notrace void print_func_help_header(struct seq_file *m)
1004 {
1005         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1006         seq_puts(m, "#              | |      |          |         |\n");
1007 }
1008
1009
1010 static notrace void
1011 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1012 {
1013         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1014         struct trace_array *tr = iter->tr;
1015         struct trace_array_cpu *data = tr->data[tr->cpu];
1016         struct tracer *type = current_trace;
1017         unsigned long total   = 0;
1018         unsigned long entries = 0;
1019         int cpu;
1020         const char *name = "preemption";
1021
1022         if (type)
1023                 name = type->name;
1024
1025         for_each_possible_cpu(cpu) {
1026                 if (head_page(tr->data[cpu])) {
1027                         total += tr->data[cpu]->trace_idx;
1028                         if (tr->data[cpu]->trace_idx > tr->entries)
1029                                 entries += tr->entries;
1030                         else
1031                                 entries += tr->data[cpu]->trace_idx;
1032                 }
1033         }
1034
1035         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1036                    name, UTS_RELEASE);
1037         seq_puts(m, "-----------------------------------"
1038                  "---------------------------------\n");
1039         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1040                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1041                    nsecs_to_usecs(data->saved_latency),
1042                    entries,
1043                    total,
1044                    tr->cpu,
1045 #if defined(CONFIG_PREEMPT_NONE)
1046                    "server",
1047 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1048                    "desktop",
1049 #elif defined(CONFIG_PREEMPT_DESKTOP)
1050                    "preempt",
1051 #else
1052                    "unknown",
1053 #endif
1054                    /* These are reserved for later use */
1055                    0, 0, 0, 0);
1056 #ifdef CONFIG_SMP
1057         seq_printf(m, " #P:%d)\n", num_online_cpus());
1058 #else
1059         seq_puts(m, ")\n");
1060 #endif
1061         seq_puts(m, "    -----------------\n");
1062         seq_printf(m, "    | task: %.16s-%d "
1063                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1064                    data->comm, data->pid, data->uid, data->nice,
1065                    data->policy, data->rt_priority);
1066         seq_puts(m, "    -----------------\n");
1067
1068         if (data->critical_start) {
1069                 seq_puts(m, " => started at: ");
1070                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1071                 trace_print_seq(m, &iter->seq);
1072                 seq_puts(m, "\n => ended at:   ");
1073                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1074                 trace_print_seq(m, &iter->seq);
1075                 seq_puts(m, "\n");
1076         }
1077
1078         seq_puts(m, "\n");
1079 }
1080
1081 static notrace void
1082 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1083 {
1084         int hardirq, softirq;
1085         char *comm;
1086
1087         comm = trace_find_cmdline(entry->pid);
1088
1089         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1090         trace_seq_printf(s, "%d", cpu);
1091         trace_seq_printf(s, "%c%c",
1092                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1093                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1094
1095         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1096         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1097         if (hardirq && softirq)
1098                 trace_seq_putc(s, 'H');
1099         else {
1100                 if (hardirq)
1101                         trace_seq_putc(s, 'h');
1102                 else {
1103                         if (softirq)
1104                                 trace_seq_putc(s, 's');
1105                         else
1106                                 trace_seq_putc(s, '.');
1107                 }
1108         }
1109
1110         if (entry->preempt_count)
1111                 trace_seq_printf(s, "%x", entry->preempt_count);
1112         else
1113                 trace_seq_puts(s, ".");
1114 }
1115
1116 unsigned long preempt_mark_thresh = 100;
1117
1118 static notrace void
1119 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1120                     unsigned long rel_usecs)
1121 {
1122         trace_seq_printf(s, " %4lldus", abs_usecs);
1123         if (rel_usecs > preempt_mark_thresh)
1124                 trace_seq_puts(s, "!: ");
1125         else if (rel_usecs > 1)
1126                 trace_seq_puts(s, "+: ");
1127         else
1128                 trace_seq_puts(s, " : ");
1129 }
1130
1131 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1132
1133 static notrace int
1134 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1135 {
1136         struct trace_seq *s = &iter->seq;
1137         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1138         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1139         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1140         struct trace_entry *entry = iter->ent;
1141         unsigned long abs_usecs;
1142         unsigned long rel_usecs;
1143         char *comm;
1144         int S;
1145
1146         if (!next_entry)
1147                 next_entry = entry;
1148         rel_usecs = ns2usecs(next_entry->t - entry->t);
1149         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1150
1151         if (verbose) {
1152                 comm = trace_find_cmdline(entry->pid);
1153                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1154                                  " %ld.%03ldms (+%ld.%03ldms): ",
1155                                  comm,
1156                                  entry->pid, cpu, entry->flags,
1157                                  entry->preempt_count, trace_idx,
1158                                  ns2usecs(entry->t),
1159                                  abs_usecs/1000,
1160                                  abs_usecs % 1000, rel_usecs/1000,
1161                                  rel_usecs % 1000);
1162         } else {
1163                 lat_print_generic(s, entry, cpu);
1164                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1165         }
1166         switch (entry->type) {
1167         case TRACE_FN:
1168                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1169                 trace_seq_puts(s, " (");
1170                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1171                 trace_seq_puts(s, ")\n");
1172                 break;
1173         case TRACE_CTX:
1174                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1175                         state_to_char[entry->ctx.prev_state] : 'X';
1176                 comm = trace_find_cmdline(entry->ctx.next_pid);
1177                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1178                                  entry->ctx.prev_pid,
1179                                  entry->ctx.prev_prio,
1180                                  S,
1181                                  entry->ctx.next_pid,
1182                                  entry->ctx.next_prio,
1183                                  comm);
1184                 break;
1185         case TRACE_SPECIAL:
1186                 trace_seq_printf(s, " %lx %lx %lx\n",
1187                                  entry->special.arg1,
1188                                  entry->special.arg2,
1189                                  entry->special.arg3);
1190                 break;
1191         default:
1192                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1193         }
1194         return 1;
1195 }
1196
1197 static notrace int print_trace_fmt(struct trace_iterator *iter)
1198 {
1199         struct trace_seq *s = &iter->seq;
1200         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1201         struct trace_entry *entry;
1202         unsigned long usec_rem;
1203         unsigned long long t;
1204         unsigned long secs;
1205         char *comm;
1206         int S;
1207         int ret;
1208
1209         entry = iter->ent;
1210
1211         comm = trace_find_cmdline(iter->ent->pid);
1212
1213         t = ns2usecs(entry->t);
1214         usec_rem = do_div(t, 1000000ULL);
1215         secs = (unsigned long)t;
1216
1217         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1218         if (!ret)
1219                 return 0;
1220         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1221         if (!ret)
1222                 return 0;
1223         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1224         if (!ret)
1225                 return 0;
1226
1227         switch (entry->type) {
1228         case TRACE_FN:
1229                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1230                 if (!ret)
1231                         return 0;
1232                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1233                                                 entry->fn.parent_ip) {
1234                         ret = trace_seq_printf(s, " <-");
1235                         if (!ret)
1236                                 return 0;
1237                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1238                                                sym_flags);
1239                         if (!ret)
1240                                 return 0;
1241                 }
1242                 ret = trace_seq_printf(s, "\n");
1243                 if (!ret)
1244                         return 0;
1245                 break;
1246         case TRACE_CTX:
1247                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1248                         state_to_char[entry->ctx.prev_state] : 'X';
1249                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1250                                        entry->ctx.prev_pid,
1251                                        entry->ctx.prev_prio,
1252                                        S,
1253                                        entry->ctx.next_pid,
1254                                        entry->ctx.next_prio);
1255                 if (!ret)
1256                         return 0;
1257                 break;
1258         case TRACE_SPECIAL:
1259                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1260                                  entry->special.arg1,
1261                                  entry->special.arg2,
1262                                  entry->special.arg3);
1263                 if (!ret)
1264                         return 0;
1265                 break;
1266         }
1267         return 1;
1268 }
1269
1270 static notrace int print_raw_fmt(struct trace_iterator *iter)
1271 {
1272         struct trace_seq *s = &iter->seq;
1273         struct trace_entry *entry;
1274         int ret;
1275         int S;
1276
1277         entry = iter->ent;
1278
1279         ret = trace_seq_printf(s, "%d %d %llu ",
1280                 entry->pid, iter->cpu, entry->t);
1281         if (!ret)
1282                 return 0;
1283
1284         switch (entry->type) {
1285         case TRACE_FN:
1286                 ret = trace_seq_printf(s, "%x %x\n",
1287                                         entry->fn.ip, entry->fn.parent_ip);
1288                 if (!ret)
1289                         return 0;
1290                 break;
1291         case TRACE_CTX:
1292                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1293                         state_to_char[entry->ctx.prev_state] : 'X';
1294                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1295                                        entry->ctx.prev_pid,
1296                                        entry->ctx.prev_prio,
1297                                        S,
1298                                        entry->ctx.next_pid,
1299                                        entry->ctx.next_prio);
1300                 if (!ret)
1301                         return 0;
1302                 break;
1303         case TRACE_SPECIAL:
1304                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1305                                  entry->special.arg1,
1306                                  entry->special.arg2,
1307                                  entry->special.arg3);
1308                 if (!ret)
1309                         return 0;
1310                 break;
1311         }
1312         return 1;
1313 }
1314
1315 #define SEQ_PUT_FIELD_RET(s, x)                         \
1316 do {                                                    \
1317         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1318                 return 0;                               \
1319 } while (0)
1320
1321 static notrace int print_bin_fmt(struct trace_iterator *iter)
1322 {
1323         struct trace_seq *s = &iter->seq;
1324         struct trace_entry *entry;
1325
1326         entry = iter->ent;
1327
1328         SEQ_PUT_FIELD_RET(s, entry->pid);
1329         SEQ_PUT_FIELD_RET(s, entry->cpu);
1330         SEQ_PUT_FIELD_RET(s, entry->t);
1331
1332         switch (entry->type) {
1333         case TRACE_FN:
1334                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1335                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1336                 break;
1337         case TRACE_CTX:
1338                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1339                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1340                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1341                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1342                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1343                 break;
1344         case TRACE_SPECIAL:
1345                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1346                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1347                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1348                 break;
1349         }
1350         return 1;
1351 }
1352
1353 static int trace_empty(struct trace_iterator *iter)
1354 {
1355         struct trace_array_cpu *data;
1356         int cpu;
1357
1358         for_each_possible_cpu(cpu) {
1359                 data = iter->tr->data[cpu];
1360
1361                 if (head_page(data) && data->trace_idx &&
1362                     (data->trace_tail != data->trace_head ||
1363                      data->trace_tail_idx != data->trace_head_idx))
1364                         return 0;
1365         }
1366         return 1;
1367 }
1368
1369 static int print_trace_line(struct trace_iterator *iter)
1370 {
1371         if (trace_flags & TRACE_ITER_BIN)
1372                 return print_bin_fmt(iter);
1373
1374         if (trace_flags & TRACE_ITER_RAW)
1375                 return print_raw_fmt(iter);
1376
1377         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1378                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1379
1380         return print_trace_fmt(iter);
1381 }
1382
1383 static int s_show(struct seq_file *m, void *v)
1384 {
1385         struct trace_iterator *iter = v;
1386
1387         if (iter->ent == NULL) {
1388                 if (iter->tr) {
1389                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1390                         seq_puts(m, "#\n");
1391                 }
1392                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1393                         /* print nothing if the buffers are empty */
1394                         if (trace_empty(iter))
1395                                 return 0;
1396                         print_trace_header(m, iter);
1397                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1398                                 print_lat_help_header(m);
1399                 } else {
1400                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1401                                 print_func_help_header(m);
1402                 }
1403         } else {
1404                 print_trace_line(iter);
1405                 trace_print_seq(m, &iter->seq);
1406         }
1407
1408         return 0;
1409 }
1410
1411 static struct seq_operations tracer_seq_ops = {
1412         .start          = s_start,
1413         .next           = s_next,
1414         .stop           = s_stop,
1415         .show           = s_show,
1416 };
1417
1418 static struct trace_iterator notrace *
1419 __tracing_open(struct inode *inode, struct file *file, int *ret)
1420 {
1421         struct trace_iterator *iter;
1422
1423         if (tracing_disabled) {
1424                 *ret = -ENODEV;
1425                 return NULL;
1426         }
1427
1428         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1429         if (!iter) {
1430                 *ret = -ENOMEM;
1431                 goto out;
1432         }
1433
1434         mutex_lock(&trace_types_lock);
1435         if (current_trace && current_trace->print_max)
1436                 iter->tr = &max_tr;
1437         else
1438                 iter->tr = inode->i_private;
1439         iter->trace = current_trace;
1440         iter->pos = -1;
1441
1442         /* TODO stop tracer */
1443         *ret = seq_open(file, &tracer_seq_ops);
1444         if (!*ret) {
1445                 struct seq_file *m = file->private_data;
1446                 m->private = iter;
1447
1448                 /* stop the trace while dumping */
1449                 if (iter->tr->ctrl)
1450                         tracer_enabled = 0;
1451
1452                 if (iter->trace && iter->trace->open)
1453                         iter->trace->open(iter);
1454         } else {
1455                 kfree(iter);
1456                 iter = NULL;
1457         }
1458         mutex_unlock(&trace_types_lock);
1459
1460  out:
1461         return iter;
1462 }
1463
1464 int tracing_open_generic(struct inode *inode, struct file *filp)
1465 {
1466         if (tracing_disabled)
1467                 return -ENODEV;
1468
1469         filp->private_data = inode->i_private;
1470         return 0;
1471 }
1472
1473 int tracing_release(struct inode *inode, struct file *file)
1474 {
1475         struct seq_file *m = (struct seq_file *)file->private_data;
1476         struct trace_iterator *iter = m->private;
1477
1478         mutex_lock(&trace_types_lock);
1479         if (iter->trace && iter->trace->close)
1480                 iter->trace->close(iter);
1481
1482         /* reenable tracing if it was previously enabled */
1483         if (iter->tr->ctrl)
1484                 tracer_enabled = 1;
1485         mutex_unlock(&trace_types_lock);
1486
1487         seq_release(inode, file);
1488         kfree(iter);
1489         return 0;
1490 }
1491
1492 static int tracing_open(struct inode *inode, struct file *file)
1493 {
1494         int ret;
1495
1496         __tracing_open(inode, file, &ret);
1497
1498         return ret;
1499 }
1500
1501 static int tracing_lt_open(struct inode *inode, struct file *file)
1502 {
1503         struct trace_iterator *iter;
1504         int ret;
1505
1506         iter = __tracing_open(inode, file, &ret);
1507
1508         if (!ret)
1509                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1510
1511         return ret;
1512 }
1513
1514
1515 static notrace void *
1516 t_next(struct seq_file *m, void *v, loff_t *pos)
1517 {
1518         struct tracer *t = m->private;
1519
1520         (*pos)++;
1521
1522         if (t)
1523                 t = t->next;
1524
1525         m->private = t;
1526
1527         return t;
1528 }
1529
1530 static void *t_start(struct seq_file *m, loff_t *pos)
1531 {
1532         struct tracer *t = m->private;
1533         loff_t l = 0;
1534
1535         mutex_lock(&trace_types_lock);
1536         for (; t && l < *pos; t = t_next(m, t, &l))
1537                 ;
1538
1539         return t;
1540 }
1541
1542 static void t_stop(struct seq_file *m, void *p)
1543 {
1544         mutex_unlock(&trace_types_lock);
1545 }
1546
1547 static int t_show(struct seq_file *m, void *v)
1548 {
1549         struct tracer *t = v;
1550
1551         if (!t)
1552                 return 0;
1553
1554         seq_printf(m, "%s", t->name);
1555         if (t->next)
1556                 seq_putc(m, ' ');
1557         else
1558                 seq_putc(m, '\n');
1559
1560         return 0;
1561 }
1562
1563 static struct seq_operations show_traces_seq_ops = {
1564         .start          = t_start,
1565         .next           = t_next,
1566         .stop           = t_stop,
1567         .show           = t_show,
1568 };
1569
1570 static int show_traces_open(struct inode *inode, struct file *file)
1571 {
1572         int ret;
1573
1574         if (tracing_disabled)
1575                 return -ENODEV;
1576
1577         ret = seq_open(file, &show_traces_seq_ops);
1578         if (!ret) {
1579                 struct seq_file *m = file->private_data;
1580                 m->private = trace_types;
1581         }
1582
1583         return ret;
1584 }
1585
1586 static struct file_operations tracing_fops = {
1587         .open           = tracing_open,
1588         .read           = seq_read,
1589         .llseek         = seq_lseek,
1590         .release        = tracing_release,
1591 };
1592
1593 static struct file_operations tracing_lt_fops = {
1594         .open           = tracing_lt_open,
1595         .read           = seq_read,
1596         .llseek         = seq_lseek,
1597         .release        = tracing_release,
1598 };
1599
1600 static struct file_operations show_traces_fops = {
1601         .open = show_traces_open,
1602         .read = seq_read,
1603         .release = seq_release,
1604 };
1605
1606 static ssize_t
1607 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1608                        size_t cnt, loff_t *ppos)
1609 {
1610         char *buf;
1611         int r = 0;
1612         int len = 0;
1613         int i;
1614
1615         /* calulate max size */
1616         for (i = 0; trace_options[i]; i++) {
1617                 len += strlen(trace_options[i]);
1618                 len += 3; /* "no" and space */
1619         }
1620
1621         /* +2 for \n and \0 */
1622         buf = kmalloc(len + 2, GFP_KERNEL);
1623         if (!buf)
1624                 return -ENOMEM;
1625
1626         for (i = 0; trace_options[i]; i++) {
1627                 if (trace_flags & (1 << i))
1628                         r += sprintf(buf + r, "%s ", trace_options[i]);
1629                 else
1630                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1631         }
1632
1633         r += sprintf(buf + r, "\n");
1634         WARN_ON(r >= len + 2);
1635
1636         r = simple_read_from_buffer(ubuf, cnt, ppos,
1637                                     buf, r);
1638
1639         kfree(buf);
1640
1641         return r;
1642 }
1643
1644 static ssize_t
1645 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1646                         size_t cnt, loff_t *ppos)
1647 {
1648         char buf[64];
1649         char *cmp = buf;
1650         int neg = 0;
1651         int i;
1652
1653         if (cnt > 63)
1654                 cnt = 63;
1655
1656         if (copy_from_user(&buf, ubuf, cnt))
1657                 return -EFAULT;
1658
1659         buf[cnt] = 0;
1660
1661         if (strncmp(buf, "no", 2) == 0) {
1662                 neg = 1;
1663                 cmp += 2;
1664         }
1665
1666         for (i = 0; trace_options[i]; i++) {
1667                 int len = strlen(trace_options[i]);
1668
1669                 if (strncmp(cmp, trace_options[i], len) == 0) {
1670                         if (neg)
1671                                 trace_flags &= ~(1 << i);
1672                         else
1673                                 trace_flags |= (1 << i);
1674                         break;
1675                 }
1676         }
1677
1678         filp->f_pos += cnt;
1679
1680         return cnt;
1681 }
1682
1683 static struct file_operations tracing_iter_fops = {
1684         .open = tracing_open_generic,
1685         .read = tracing_iter_ctrl_read,
1686         .write = tracing_iter_ctrl_write,
1687 };
1688
1689 static const char readme_msg[] =
1690         "tracing mini-HOWTO:\n\n"
1691         "# mkdir /debug\n"
1692         "# mount -t debugfs nodev /debug\n\n"
1693         "# cat /debug/tracing/available_tracers\n"
1694         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1695         "# cat /debug/tracing/current_tracer\n"
1696         "none\n"
1697         "# echo sched_switch > /debug/tracing/current_tracer\n"
1698         "# cat /debug/tracing/current_tracer\n"
1699         "sched_switch\n"
1700         "# cat /debug/tracing/iter_ctrl\n"
1701         "noprint-parent nosym-offset nosym-addr noverbose\n"
1702         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1703         "# echo 1 > /debug/tracing/tracing_enabled\n"
1704         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1705         "echo 0 > /debug/tracing/tracing_enabled\n"
1706 ;
1707
1708 static ssize_t
1709 tracing_readme_read(struct file *filp, char __user *ubuf,
1710                        size_t cnt, loff_t *ppos)
1711 {
1712         return simple_read_from_buffer(ubuf, cnt, ppos,
1713                                         readme_msg, strlen(readme_msg));
1714 }
1715
1716 static struct file_operations tracing_readme_fops = {
1717         .open = tracing_open_generic,
1718         .read = tracing_readme_read,
1719 };
1720
1721
1722 static ssize_t
1723 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1724                   size_t cnt, loff_t *ppos)
1725 {
1726         struct trace_array *tr = filp->private_data;
1727         char buf[64];
1728         int r;
1729
1730         r = sprintf(buf, "%ld\n", tr->ctrl);
1731         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1732 }
1733
1734 static ssize_t
1735 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1736                    size_t cnt, loff_t *ppos)
1737 {
1738         struct trace_array *tr = filp->private_data;
1739         long val;
1740         char buf[64];
1741
1742         if (cnt > 63)
1743                 cnt = 63;
1744
1745         if (copy_from_user(&buf, ubuf, cnt))
1746                 return -EFAULT;
1747
1748         buf[cnt] = 0;
1749
1750         val = simple_strtoul(buf, NULL, 10);
1751
1752         val = !!val;
1753
1754         mutex_lock(&trace_types_lock);
1755         if (tr->ctrl ^ val) {
1756                 if (val)
1757                         tracer_enabled = 1;
1758                 else
1759                         tracer_enabled = 0;
1760
1761                 tr->ctrl = val;
1762
1763                 if (current_trace && current_trace->ctrl_update)
1764                         current_trace->ctrl_update(tr);
1765         }
1766         mutex_unlock(&trace_types_lock);
1767
1768         filp->f_pos += cnt;
1769
1770         return cnt;
1771 }
1772
1773 static ssize_t
1774 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1775                        size_t cnt, loff_t *ppos)
1776 {
1777         char buf[max_tracer_type_len+2];
1778         int r;
1779
1780         mutex_lock(&trace_types_lock);
1781         if (current_trace)
1782                 r = sprintf(buf, "%s\n", current_trace->name);
1783         else
1784                 r = sprintf(buf, "\n");
1785         mutex_unlock(&trace_types_lock);
1786
1787         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1788 }
1789
1790 static ssize_t
1791 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1792                         size_t cnt, loff_t *ppos)
1793 {
1794         struct trace_array *tr = &global_trace;
1795         struct tracer *t;
1796         char buf[max_tracer_type_len+1];
1797         int i;
1798
1799         if (cnt > max_tracer_type_len)
1800                 cnt = max_tracer_type_len;
1801
1802         if (copy_from_user(&buf, ubuf, cnt))
1803                 return -EFAULT;
1804
1805         buf[cnt] = 0;
1806
1807         /* strip ending whitespace. */
1808         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1809                 buf[i] = 0;
1810
1811         mutex_lock(&trace_types_lock);
1812         for (t = trace_types; t; t = t->next) {
1813                 if (strcmp(t->name, buf) == 0)
1814                         break;
1815         }
1816         if (!t || t == current_trace)
1817                 goto out;
1818
1819         if (current_trace && current_trace->reset)
1820                 current_trace->reset(tr);
1821
1822         current_trace = t;
1823         if (t->init)
1824                 t->init(tr);
1825
1826  out:
1827         mutex_unlock(&trace_types_lock);
1828
1829         filp->f_pos += cnt;
1830
1831         return cnt;
1832 }
1833
1834 static ssize_t
1835 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1836                      size_t cnt, loff_t *ppos)
1837 {
1838         unsigned long *ptr = filp->private_data;
1839         char buf[64];
1840         int r;
1841
1842         r = snprintf(buf, 64, "%ld\n",
1843                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1844         if (r > 64)
1845                 r = 64;
1846         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1847 }
1848
1849 static ssize_t
1850 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1851                       size_t cnt, loff_t *ppos)
1852 {
1853         long *ptr = filp->private_data;
1854         long val;
1855         char buf[64];
1856
1857         if (cnt > 63)
1858                 cnt = 63;
1859
1860         if (copy_from_user(&buf, ubuf, cnt))
1861                 return -EFAULT;
1862
1863         buf[cnt] = 0;
1864
1865         val = simple_strtoul(buf, NULL, 10);
1866
1867         *ptr = val * 1000;
1868
1869         return cnt;
1870 }
1871
1872 static atomic_t tracing_reader;
1873
1874 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1875 {
1876         struct trace_iterator *iter;
1877
1878         if (tracing_disabled)
1879                 return -ENODEV;
1880
1881         /* We only allow for reader of the pipe */
1882         if (atomic_inc_return(&tracing_reader) != 1) {
1883                 atomic_dec(&tracing_reader);
1884                 return -EBUSY;
1885         }
1886
1887         /* create a buffer to store the information to pass to userspace */
1888         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1889         if (!iter)
1890                 return -ENOMEM;
1891
1892         iter->tr = &global_trace;
1893
1894         filp->private_data = iter;
1895
1896         return 0;
1897 }
1898
1899 static int tracing_release_pipe(struct inode *inode, struct file *file)
1900 {
1901         struct trace_iterator *iter = file->private_data;
1902
1903         kfree(iter);
1904         atomic_dec(&tracing_reader);
1905
1906         return 0;
1907 }
1908
1909 /*
1910  * Consumer reader.
1911  */
1912 static ssize_t
1913 tracing_read_pipe(struct file *filp, char __user *ubuf,
1914                   size_t cnt, loff_t *ppos)
1915 {
1916         struct trace_iterator *iter = filp->private_data;
1917         struct trace_array_cpu *data;
1918         static cpumask_t mask;
1919         static int start;
1920         unsigned long flags;
1921         int read = 0;
1922         int cpu;
1923         int len;
1924         int ret;
1925
1926         /* return any leftover data */
1927         if (iter->seq.len > start) {
1928                 len = iter->seq.len - start;
1929                 if (cnt > len)
1930                         cnt = len;
1931                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1932                 if (ret)
1933                         cnt = -EFAULT;
1934
1935                 start += len;
1936
1937                 return cnt;
1938         }
1939
1940         trace_seq_reset(&iter->seq);
1941         start = 0;
1942
1943         while (trace_empty(iter)) {
1944                 /*
1945                  * This is a make-shift waitqueue. The reason we don't use
1946                  * an actual wait queue is because:
1947                  *  1) we only ever have one waiter
1948                  *  2) the tracing, traces all functions, we don't want
1949                  *     the overhead of calling wake_up and friends
1950                  *     (and tracing them too)
1951                  *     Anyway, this is really very primitive wakeup.
1952                  */
1953                 set_current_state(TASK_INTERRUPTIBLE);
1954                 iter->tr->waiter = current;
1955
1956                 /* sleep for one second, and try again. */
1957                 schedule_timeout(HZ);
1958
1959                 iter->tr->waiter = NULL;
1960
1961                 if (signal_pending(current))
1962                         return -EINTR;
1963
1964                 /*
1965                  * We block until we read something and tracing is disabled.
1966                  * We still block if tracing is disabled, but we have never
1967                  * read anything. This allows a user to cat this file, and
1968                  * then enable tracing. But after we have read something,
1969                  * we give an EOF when tracing is again disabled.
1970                  *
1971                  * iter->pos will be 0 if we haven't read anything.
1972                  */
1973                 if (!tracer_enabled && iter->pos)
1974                         break;
1975
1976                 continue;
1977         }
1978
1979         /* stop when tracing is finished */
1980         if (trace_empty(iter))
1981                 return 0;
1982
1983         if (cnt >= PAGE_SIZE)
1984                 cnt = PAGE_SIZE - 1;
1985
1986         memset(iter, 0, sizeof(*iter));
1987         iter->tr = &global_trace;
1988         iter->pos = -1;
1989
1990         /*
1991          * We need to stop all tracing on all CPUS to read the
1992          * the next buffer. This is a bit expensive, but is
1993          * not done often. We fill all what we can read,
1994          * and then release the locks again.
1995          */
1996
1997         cpus_clear(mask);
1998         local_irq_save(flags);
1999         for_each_possible_cpu(cpu) {
2000                 data = iter->tr->data[cpu];
2001
2002                 if (!head_page(data) || !data->trace_idx)
2003                         continue;
2004
2005                 atomic_inc(&data->disabled);
2006                 spin_lock(&data->lock);
2007                 cpu_set(cpu, mask);
2008         }
2009
2010         while (find_next_entry_inc(iter) != NULL) {
2011                 int len = iter->seq.len;
2012
2013                 ret = print_trace_line(iter);
2014                 if (!ret) {
2015                         /* don't print partial lines */
2016                         iter->seq.len = len;
2017                         break;
2018                 }
2019
2020                 trace_consume(iter);
2021
2022                 if (iter->seq.len >= cnt)
2023                         break;
2024         }
2025
2026         for_each_cpu_mask(cpu, mask) {
2027                 data = iter->tr->data[cpu];
2028                 spin_unlock(&data->lock);
2029                 atomic_dec(&data->disabled);
2030         }
2031         local_irq_restore(flags);
2032
2033         /* Now copy what we have to the user */
2034         read = iter->seq.len;
2035         if (read > cnt)
2036                 read = cnt;
2037
2038         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2039
2040         if (read < iter->seq.len)
2041                 start = read;
2042         else
2043                 trace_seq_reset(&iter->seq);
2044
2045         if (ret)
2046                 read = -EFAULT;
2047
2048         return read;
2049 }
2050
2051 static struct file_operations tracing_max_lat_fops = {
2052         .open           = tracing_open_generic,
2053         .read           = tracing_max_lat_read,
2054         .write          = tracing_max_lat_write,
2055 };
2056
2057 static struct file_operations tracing_ctrl_fops = {
2058         .open           = tracing_open_generic,
2059         .read           = tracing_ctrl_read,
2060         .write          = tracing_ctrl_write,
2061 };
2062
2063 static struct file_operations set_tracer_fops = {
2064         .open           = tracing_open_generic,
2065         .read           = tracing_set_trace_read,
2066         .write          = tracing_set_trace_write,
2067 };
2068
2069 static struct file_operations tracing_pipe_fops = {
2070         .open           = tracing_open_pipe,
2071         .read           = tracing_read_pipe,
2072         .release        = tracing_release_pipe,
2073 };
2074
2075 #ifdef CONFIG_DYNAMIC_FTRACE
2076
2077 static ssize_t
2078 tracing_read_long(struct file *filp, char __user *ubuf,
2079                   size_t cnt, loff_t *ppos)
2080 {
2081         unsigned long *p = filp->private_data;
2082         char buf[64];
2083         int r;
2084
2085         r = sprintf(buf, "%ld\n", *p);
2086
2087         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2088 }
2089
2090 static struct file_operations tracing_read_long_fops = {
2091         .open           = tracing_open_generic,
2092         .read           = tracing_read_long,
2093 };
2094 #endif
2095
2096 static struct dentry *d_tracer;
2097
2098 struct dentry *tracing_init_dentry(void)
2099 {
2100         static int once;
2101
2102         if (d_tracer)
2103                 return d_tracer;
2104
2105         d_tracer = debugfs_create_dir("tracing", NULL);
2106
2107         if (!d_tracer && !once) {
2108                 once = 1;
2109                 pr_warning("Could not create debugfs directory 'tracing'\n");
2110                 return NULL;
2111         }
2112
2113         return d_tracer;
2114 }
2115
2116 #ifdef CONFIG_FTRACE_SELFTEST
2117 /* Let selftest have access to static functions in this file */
2118 #include "trace_selftest.c"
2119 #endif
2120
2121 static __init void tracer_init_debugfs(void)
2122 {
2123         struct dentry *d_tracer;
2124         struct dentry *entry;
2125
2126         d_tracer = tracing_init_dentry();
2127
2128         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2129                                     &global_trace, &tracing_ctrl_fops);
2130         if (!entry)
2131                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2132
2133         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2134                                     NULL, &tracing_iter_fops);
2135         if (!entry)
2136                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2137
2138         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2139                                     &global_trace, &tracing_lt_fops);
2140         if (!entry)
2141                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2142
2143         entry = debugfs_create_file("trace", 0444, d_tracer,
2144                                     &global_trace, &tracing_fops);
2145         if (!entry)
2146                 pr_warning("Could not create debugfs 'trace' entry\n");
2147
2148         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2149                                     &global_trace, &show_traces_fops);
2150         if (!entry)
2151                 pr_warning("Could not create debugfs 'trace' entry\n");
2152
2153         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2154                                     &global_trace, &set_tracer_fops);
2155         if (!entry)
2156                 pr_warning("Could not create debugfs 'trace' entry\n");
2157
2158         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2159                                     &tracing_max_latency,
2160                                     &tracing_max_lat_fops);
2161         if (!entry)
2162                 pr_warning("Could not create debugfs "
2163                            "'tracing_max_latency' entry\n");
2164
2165         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2166                                     &tracing_thresh, &tracing_max_lat_fops);
2167         if (!entry)
2168                 pr_warning("Could not create debugfs "
2169                            "'tracing_threash' entry\n");
2170         entry = debugfs_create_file("README", 0644, d_tracer,
2171                                     NULL, &tracing_readme_fops);
2172         if (!entry)
2173                 pr_warning("Could not create debugfs 'README' entry\n");
2174
2175         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2176                                     NULL, &tracing_pipe_fops);
2177         if (!entry)
2178                 pr_warning("Could not create debugfs "
2179                            "'tracing_threash' entry\n");
2180
2181 #ifdef CONFIG_DYNAMIC_FTRACE
2182         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2183                                     &ftrace_update_tot_cnt,
2184                                     &tracing_read_long_fops);
2185         if (!entry)
2186                 pr_warning("Could not create debugfs "
2187                            "'dyn_ftrace_total_info' entry\n");
2188 #endif
2189 }
2190
2191 /* dummy trace to disable tracing */
2192 static struct tracer no_tracer __read_mostly =
2193 {
2194         .name           = "none",
2195 };
2196
2197 static int trace_alloc_page(void)
2198 {
2199         struct trace_array_cpu *data;
2200         struct page *page, *tmp;
2201         LIST_HEAD(pages);
2202         void *array;
2203         int i;
2204
2205         /* first allocate a page for each CPU */
2206         for_each_possible_cpu(i) {
2207                 array = (void *)__get_free_page(GFP_KERNEL);
2208                 if (array == NULL) {
2209                         printk(KERN_ERR "tracer: failed to allocate page"
2210                                "for trace buffer!\n");
2211                         goto free_pages;
2212                 }
2213
2214                 page = virt_to_page(array);
2215                 list_add(&page->lru, &pages);
2216
2217 /* Only allocate if we are actually using the max trace */
2218 #ifdef CONFIG_TRACER_MAX_TRACE
2219                 array = (void *)__get_free_page(GFP_KERNEL);
2220                 if (array == NULL) {
2221                         printk(KERN_ERR "tracer: failed to allocate page"
2222                                "for trace buffer!\n");
2223                         goto free_pages;
2224                 }
2225                 page = virt_to_page(array);
2226                 list_add(&page->lru, &pages);
2227 #endif
2228         }
2229
2230         /* Now that we successfully allocate a page per CPU, add them */
2231         for_each_possible_cpu(i) {
2232                 data = global_trace.data[i];
2233                 spin_lock_init(&data->lock);
2234                 lockdep_set_class(&data->lock, &data->lock_key);
2235                 page = list_entry(pages.next, struct page, lru);
2236                 list_del_init(&page->lru);
2237                 list_add_tail(&page->lru, &data->trace_pages);
2238                 ClearPageLRU(page);
2239
2240 #ifdef CONFIG_TRACER_MAX_TRACE
2241                 data = max_tr.data[i];
2242                 spin_lock_init(&data->lock);
2243                 lockdep_set_class(&data->lock, &data->lock_key);
2244                 page = list_entry(pages.next, struct page, lru);
2245                 list_del_init(&page->lru);
2246                 list_add_tail(&page->lru, &data->trace_pages);
2247                 SetPageLRU(page);
2248 #endif
2249         }
2250         global_trace.entries += ENTRIES_PER_PAGE;
2251
2252         return 0;
2253
2254  free_pages:
2255         list_for_each_entry_safe(page, tmp, &pages, lru) {
2256                 list_del_init(&page->lru);
2257                 __free_page(page);
2258         }
2259         return -ENOMEM;
2260 }
2261
2262 __init static int tracer_alloc_buffers(void)
2263 {
2264         struct trace_array_cpu *data;
2265         void *array;
2266         struct page *page;
2267         int pages = 0;
2268         int ret = -ENOMEM;
2269         int i;
2270
2271         global_trace.ctrl = tracer_enabled;
2272
2273         /* Allocate the first page for all buffers */
2274         for_each_possible_cpu(i) {
2275                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2276                 max_tr.data[i] = &per_cpu(max_data, i);
2277
2278                 array = (void *)__get_free_page(GFP_KERNEL);
2279                 if (array == NULL) {
2280                         printk(KERN_ERR "tracer: failed to allocate page"
2281                                "for trace buffer!\n");
2282                         goto free_buffers;
2283                 }
2284
2285                 /* set the array to the list */
2286                 INIT_LIST_HEAD(&data->trace_pages);
2287                 page = virt_to_page(array);
2288                 list_add(&page->lru, &data->trace_pages);
2289                 /* use the LRU flag to differentiate the two buffers */
2290                 ClearPageLRU(page);
2291
2292 /* Only allocate if we are actually using the max trace */
2293 #ifdef CONFIG_TRACER_MAX_TRACE
2294                 array = (void *)__get_free_page(GFP_KERNEL);
2295                 if (array == NULL) {
2296                         printk(KERN_ERR "tracer: failed to allocate page"
2297                                "for trace buffer!\n");
2298                         goto free_buffers;
2299                 }
2300
2301                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2302                 page = virt_to_page(array);
2303                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2304                 SetPageLRU(page);
2305 #endif
2306         }
2307
2308         /*
2309          * Since we allocate by orders of pages, we may be able to
2310          * round up a bit.
2311          */
2312         global_trace.entries = ENTRIES_PER_PAGE;
2313         pages++;
2314
2315         while (global_trace.entries < trace_nr_entries) {
2316                 if (trace_alloc_page())
2317                         break;
2318                 pages++;
2319         }
2320         max_tr.entries = global_trace.entries;
2321
2322         pr_info("tracer: %d pages allocated for %ld",
2323                 pages, trace_nr_entries);
2324         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2325         pr_info("   actual entries %ld\n", global_trace.entries);
2326
2327         tracer_init_debugfs();
2328
2329         trace_init_cmdlines();
2330
2331         register_tracer(&no_tracer);
2332         current_trace = &no_tracer;
2333
2334         /* All seems OK, enable tracing */
2335         tracing_disabled = 0;
2336
2337         return 0;
2338
2339  free_buffers:
2340         for (i-- ; i >= 0; i--) {
2341                 struct page *page, *tmp;
2342                 struct trace_array_cpu *data = global_trace.data[i];
2343
2344                 if (data) {
2345                         list_for_each_entry_safe(page, tmp,
2346                                                  &data->trace_pages, lru) {
2347                                 list_del_init(&page->lru);
2348                                 __free_page(page);
2349                         }
2350                 }
2351
2352 #ifdef CONFIG_TRACER_MAX_TRACE
2353                 data = max_tr.data[i];
2354                 if (data) {
2355                         list_for_each_entry_safe(page, tmp,
2356                                                  &data->trace_pages, lru) {
2357                                 list_del_init(&page->lru);
2358                                 __free_page(page);
2359                         }
2360                 }
2361 #endif
2362         }
2363         return ret;
2364 }
2365 fs_initcall(tracer_alloc_buffers);