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