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/poll.h>
  28 #include <linux/gfp.h>
  29 #include <linux/fs.h>
  30
  31 #include <linux/stacktrace.h>
  32
  33 #include "trace.h"
  34
  35 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  36 unsigned long __read_mostly     tracing_thresh;
  37
  38 static int tracing_disabled = 1;
  39
  40 static long
  41 ns2usecs(cycle_t nsec)
  42 {
  43         nsec += 500;
  44         do_div(nsec, 1000);
  45         return nsec;
  46 }
  47
  48 cycle_t ftrace_now(int cpu)
  49 {
  50         return cpu_clock(cpu);
  51 }
  52
  53 static struct trace_array       global_trace;
  54
  55 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  56
  57 static struct trace_array       max_tr;
  58
  59 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  60
  61 static int                      tracer_enabled = 1;
  62 static unsigned long            trace_nr_entries = 65536UL;
  63
  64 static struct tracer            *trace_types __read_mostly;
  65 static struct tracer            *current_trace __read_mostly;
  66 static int                      max_tracer_type_len;
  67
  68 static DEFINE_MUTEX(trace_types_lock);
  69 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
  70
  71 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
  72
  73 void trace_wake_up(void)
  74 {
  75         /*
  76          * The runqueue_is_locked() can fail, but this is the best we
  77          * have for now:
  78          */
  79         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
  80                 wake_up(&trace_wait);
  81 }
  82
  83 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  84
  85 static int __init set_nr_entries(char *str)
  86 {
  87         if (!str)
  88                 return 0;
  89         trace_nr_entries = simple_strtoul(str, &str, 0);
  90         return 1;
  91 }
  92 __setup("trace_entries=", set_nr_entries);
  93
  94 unsigned long nsecs_to_usecs(unsigned long nsecs)
  95 {
  96         return nsecs / 1000;
  97 }
  98
  99 enum trace_type {
 100         __TRACE_FIRST_TYPE = 0,
 101
 102         TRACE_FN,
 103         TRACE_CTX,
 104         TRACE_WAKE,
 105         TRACE_STACK,
 106         TRACE_SPECIAL,
 107
 108         __TRACE_LAST_TYPE
 109 };
 110
 111 enum trace_flag_type {
 112         TRACE_FLAG_IRQS_OFF             = 0x01,
 113         TRACE_FLAG_NEED_RESCHED         = 0x02,
 114         TRACE_FLAG_HARDIRQ              = 0x04,
 115         TRACE_FLAG_SOFTIRQ              = 0x08,
 116 };
 117
 118 #define TRACE_ITER_SYM_MASK \
 119         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 120
 121 /* These must match the bit postions above */
 122 static const char *trace_options[] = {
 123         "print-parent",
 124         "sym-offset",
 125         "sym-addr",
 126         "verbose",
 127         "raw",
 128         "hex",
 129         "bin",
 130         "block",
 131         "stacktrace",
 132         NULL
 133 };
 134
 135 static DEFINE_SPINLOCK(ftrace_max_lock);
 136
 137 /*
 138  * Copy the new maximum trace into the separate maximum-trace
 139  * structure. (this way the maximum trace is permanently saved,
 140  * for later retrieval via /debugfs/tracing/latency_trace)
 141  */
 142 static void
 143 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 144 {
 145         struct trace_array_cpu *data = tr->data[cpu];
 146
 147         max_tr.cpu = cpu;
 148         max_tr.time_start = data->preempt_timestamp;
 149
 150         data = max_tr.data[cpu];
 151         data->saved_latency = tracing_max_latency;
 152
 153         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 154         data->pid = tsk->pid;
 155         data->uid = tsk->uid;
 156         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 157         data->policy = tsk->policy;
 158         data->rt_priority = tsk->rt_priority;
 159
 160         /* record this tasks comm */
 161         tracing_record_cmdline(current);
 162 }
 163
 164 void check_pages(struct trace_array_cpu *data)
 165 {
 166         struct page *page, *tmp;
 167
 168         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 169         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 170
 171         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 172                 BUG_ON(page->lru.next->prev != &page->lru);
 173                 BUG_ON(page->lru.prev->next != &page->lru);
 174         }
 175 }
 176
 177 void *head_page(struct trace_array_cpu *data)
 178 {
 179         struct page *page;
 180
 181         check_pages(data);
 182         if (list_empty(&data->trace_pages))
 183                 return NULL;
 184
 185         page = list_entry(data->trace_pages.next, struct page, lru);
 186         BUG_ON(&page->lru == &data->trace_pages);
 187
 188         return page_address(page);
 189 }
 190
 191 static int
 192 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 193 {
 194         int len = (PAGE_SIZE - 1) - s->len;
 195         va_list ap;
 196         int ret;
 197
 198         if (!len)
 199                 return 0;
 200
 201         va_start(ap, fmt);
 202         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 203         va_end(ap);
 204
 205         /* If we can't write it all, don't bother writing anything */
 206         if (ret > len)
 207                 return 0;
 208
 209         s->len += ret;
 210
 211         return len;
 212 }
 213
 214 static int
 215 trace_seq_puts(struct trace_seq *s, const char *str)
 216 {
 217         int len = strlen(str);
 218
 219         if (len > ((PAGE_SIZE - 1) - s->len))
 220                 return 0;
 221
 222         memcpy(s->buffer + s->len, str, len);
 223         s->len += len;
 224
 225         return len;
 226 }
 227
 228 static int
 229 trace_seq_putc(struct trace_seq *s, unsigned char c)
 230 {
 231         if (s->len >= (PAGE_SIZE - 1))
 232                 return 0;
 233
 234         s->buffer[s->len++] = c;
 235
 236         return 1;
 237 }
 238
 239 static int
 240 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 241 {
 242         if (len > ((PAGE_SIZE - 1) - s->len))
 243                 return 0;
 244
 245         memcpy(s->buffer + s->len, mem, len);
 246         s->len += len;
 247
 248         return len;
 249 }
 250
 251 #define HEX_CHARS 17
 252
 253 static int
 254 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 255 {
 256         unsigned char hex[HEX_CHARS];
 257         unsigned char *data;
 258         unsigned char byte;
 259         int i, j;
 260
 261         BUG_ON(len >= HEX_CHARS);
 262
 263         data = mem;
 264
 265 #ifdef __BIG_ENDIAN
 266         for (i = 0, j = 0; i < len; i++) {
 267 #else
 268         for (i = len-1, j = 0; i >= 0; i--) {
 269 #endif
 270                 byte = data[i];
 271
 272                 hex[j]   = byte & 0x0f;
 273                 if (hex[j] >= 10)
 274                         hex[j] += 'a' - 10;
 275                 else
 276                         hex[j] += '0';
 277                 j++;
 278
 279                 hex[j] = byte >> 4;
 280                 if (hex[j] >= 10)
 281                         hex[j] += 'a' - 10;
 282                 else
 283                         hex[j] += '0';
 284                 j++;
 285         }
 286         hex[j] = ' ';
 287         j++;
 288
 289         return trace_seq_putmem(s, hex, j);
 290 }
 291
 292 static void
 293 trace_seq_reset(struct trace_seq *s)
 294 {
 295         s->len = 0;
 296 }
 297
 298 static void
 299 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 300 {
 301         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 302
 303         s->buffer[len] = 0;
 304         seq_puts(m, s->buffer);
 305
 306         trace_seq_reset(s);
 307 }
 308
 309 static void
 310 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 311 {
 312         struct list_head flip_pages;
 313
 314         INIT_LIST_HEAD(&flip_pages);
 315
 316         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 317                 sizeof(struct trace_array_cpu) -
 318                 offsetof(struct trace_array_cpu, trace_head_idx));
 319
 320         check_pages(tr1);
 321         check_pages(tr2);
 322         list_splice_init(&tr1->trace_pages, &flip_pages);
 323         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 324         list_splice_init(&flip_pages, &tr2->trace_pages);
 325         BUG_ON(!list_empty(&flip_pages));
 326         check_pages(tr1);
 327         check_pages(tr2);
 328 }
 329
 330 void
 331 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 332 {
 333         struct trace_array_cpu *data;
 334         int i;
 335
 336         WARN_ON_ONCE(!irqs_disabled());
 337         spin_lock(&ftrace_max_lock);
 338         /* clear out all the previous traces */
 339         for_each_possible_cpu(i) {
 340                 data = tr->data[i];
 341                 flip_trace(max_tr.data[i], data);
 342                 tracing_reset(data);
 343         }
 344
 345         __update_max_tr(tr, tsk, cpu);
 346         spin_unlock(&ftrace_max_lock);
 347 }
 348
 349 /**
 350  * update_max_tr_single - only copy one trace over, and reset the rest
 351  * @tr - tracer
 352  * @tsk - task with the latency
 353  * @cpu - the cpu of the buffer to copy.
 354  */
 355 void
 356 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 357 {
 358         struct trace_array_cpu *data = tr->data[cpu];
 359         int i;
 360
 361         WARN_ON_ONCE(!irqs_disabled());
 362         spin_lock(&ftrace_max_lock);
 363         for_each_possible_cpu(i)
 364                 tracing_reset(max_tr.data[i]);
 365
 366         flip_trace(max_tr.data[cpu], data);
 367         tracing_reset(data);
 368
 369         __update_max_tr(tr, tsk, cpu);
 370         spin_unlock(&ftrace_max_lock);
 371 }
 372
 373 int register_tracer(struct tracer *type)
 374 {
 375         struct tracer *t;
 376         int len;
 377         int ret = 0;
 378
 379         if (!type->name) {
 380                 pr_info("Tracer must have a name\n");
 381                 return -1;
 382         }
 383
 384         mutex_lock(&trace_types_lock);
 385         for (t = trace_types; t; t = t->next) {
 386                 if (strcmp(type->name, t->name) == 0) {
 387                         /* already found */
 388                         pr_info("Trace %s already registered\n",
 389                                 type->name);
 390                         ret = -1;
 391                         goto out;
 392                 }
 393         }
 394
 395 #ifdef CONFIG_FTRACE_STARTUP_TEST
 396         if (type->selftest) {
 397                 struct tracer *saved_tracer = current_trace;
 398                 struct trace_array_cpu *data;
 399                 struct trace_array *tr = &global_trace;
 400                 int saved_ctrl = tr->ctrl;
 401                 int i;
 402                 /*
 403                  * Run a selftest on this tracer.
 404                  * Here we reset the trace buffer, and set the current
 405                  * tracer to be this tracer. The tracer can then run some
 406                  * internal tracing to verify that everything is in order.
 407                  * If we fail, we do not register this tracer.
 408                  */
 409                 for_each_possible_cpu(i) {
 410                         data = tr->data[i];
 411                         if (!head_page(data))
 412                                 continue;
 413                         tracing_reset(data);
 414                 }
 415                 current_trace = type;
 416                 tr->ctrl = 0;
 417                 /* the test is responsible for initializing and enabling */
 418                 pr_info("Testing tracer %s: ", type->name);
 419                 ret = type->selftest(type, tr);
 420                 /* the test is responsible for resetting too */
 421                 current_trace = saved_tracer;
 422                 tr->ctrl = saved_ctrl;
 423                 if (ret) {
 424                         printk(KERN_CONT "FAILED!\n");
 425                         goto out;
 426                 }
 427                 /* Only reset on passing, to avoid touching corrupted buffers */
 428                 for_each_possible_cpu(i) {
 429                         data = tr->data[i];
 430                         if (!head_page(data))
 431                                 continue;
 432                         tracing_reset(data);
 433                 }
 434                 printk(KERN_CONT "PASSED\n");
 435         }
 436 #endif
 437
 438         type->next = trace_types;
 439         trace_types = type;
 440         len = strlen(type->name);
 441         if (len > max_tracer_type_len)
 442                 max_tracer_type_len = len;
 443
 444  out:
 445         mutex_unlock(&trace_types_lock);
 446
 447         return ret;
 448 }
 449
 450 void unregister_tracer(struct tracer *type)
 451 {
 452         struct tracer **t;
 453         int len;
 454
 455         mutex_lock(&trace_types_lock);
 456         for (t = &trace_types; *t; t = &(*t)->next) {
 457                 if (*t == type)
 458                         goto found;
 459         }
 460         pr_info("Trace %s not registered\n", type->name);
 461         goto out;
 462
 463  found:
 464         *t = (*t)->next;
 465         if (strlen(type->name) != max_tracer_type_len)
 466                 goto out;
 467
 468         max_tracer_type_len = 0;
 469         for (t = &trace_types; *t; t = &(*t)->next) {
 470                 len = strlen((*t)->name);
 471                 if (len > max_tracer_type_len)
 472                         max_tracer_type_len = len;
 473         }
 474  out:
 475         mutex_unlock(&trace_types_lock);
 476 }
 477
 478 void tracing_reset(struct trace_array_cpu *data)
 479 {
 480         data->trace_idx = 0;
 481         data->trace_head = data->trace_tail = head_page(data);
 482         data->trace_head_idx = 0;
 483         data->trace_tail_idx = 0;
 484 }
 485
 486 #define SAVED_CMDLINES 128
 487 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 488 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 489 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 490 static int cmdline_idx;
 491 static DEFINE_SPINLOCK(trace_cmdline_lock);
 492 atomic_t trace_record_cmdline_disabled;
 493
 494 static void trace_init_cmdlines(void)
 495 {
 496         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 497         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 498         cmdline_idx = 0;
 499 }
 500
 501 void trace_stop_cmdline_recording(void);
 502
 503 static void trace_save_cmdline(struct task_struct *tsk)
 504 {
 505         unsigned map;
 506         unsigned idx;
 507
 508         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 509                 return;
 510
 511         /*
 512          * It's not the end of the world if we don't get
 513          * the lock, but we also don't want to spin
 514          * nor do we want to disable interrupts,
 515          * so if we miss here, then better luck next time.
 516          */
 517         if (!spin_trylock(&trace_cmdline_lock))
 518                 return;
 519
 520         idx = map_pid_to_cmdline[tsk->pid];
 521         if (idx >= SAVED_CMDLINES) {
 522                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 523
 524                 map = map_cmdline_to_pid[idx];
 525                 if (map <= PID_MAX_DEFAULT)
 526                         map_pid_to_cmdline[map] = (unsigned)-1;
 527
 528                 map_pid_to_cmdline[tsk->pid] = idx;
 529
 530                 cmdline_idx = idx;
 531         }
 532
 533         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 534
 535         spin_unlock(&trace_cmdline_lock);
 536 }
 537
 538 static char *trace_find_cmdline(int pid)
 539 {
 540         char *cmdline = "<...>";
 541         unsigned map;
 542
 543         if (!pid)
 544                 return "<idle>";
 545
 546         if (pid > PID_MAX_DEFAULT)
 547                 goto out;
 548
 549         map = map_pid_to_cmdline[pid];
 550         if (map >= SAVED_CMDLINES)
 551                 goto out;
 552
 553         cmdline = saved_cmdlines[map];
 554
 555  out:
 556         return cmdline;
 557 }
 558
 559 void tracing_record_cmdline(struct task_struct *tsk)
 560 {
 561         if (atomic_read(&trace_record_cmdline_disabled))
 562                 return;
 563
 564         trace_save_cmdline(tsk);
 565 }
 566
 567 static inline struct list_head *
 568 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 569 {
 570         /*
 571          * Roundrobin - but skip the head (which is not a real page):
 572          */
 573         next = next->next;
 574         if (unlikely(next == &data->trace_pages))
 575                 next = next->next;
 576         BUG_ON(next == &data->trace_pages);
 577
 578         return next;
 579 }
 580
 581 static inline void *
 582 trace_next_page(struct trace_array_cpu *data, void *addr)
 583 {
 584         struct list_head *next;
 585         struct page *page;
 586
 587         page = virt_to_page(addr);
 588
 589         next = trace_next_list(data, &page->lru);
 590         page = list_entry(next, struct page, lru);
 591
 592         return page_address(page);
 593 }
 594
 595 static inline struct trace_entry *
 596 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 597 {
 598         unsigned long idx, idx_next;
 599         struct trace_entry *entry;
 600
 601         data->trace_idx++;
 602         idx = data->trace_head_idx;
 603         idx_next = idx + 1;
 604
 605         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 606
 607         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 608
 609         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 610                 data->trace_head = trace_next_page(data, data->trace_head);
 611                 idx_next = 0;
 612         }
 613
 614         if (data->trace_head == data->trace_tail &&
 615             idx_next == data->trace_tail_idx) {
 616                 /* overrun */
 617                 data->trace_tail_idx++;
 618                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 619                         data->trace_tail =
 620                                 trace_next_page(data, data->trace_tail);
 621                         data->trace_tail_idx = 0;
 622                 }
 623         }
 624
 625         data->trace_head_idx = idx_next;
 626
 627         return entry;
 628 }
 629
 630 static inline void
 631 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 632 {
 633         struct task_struct *tsk = current;
 634         unsigned long pc;
 635
 636         pc = preempt_count();
 637
 638         entry->preempt_count    = pc & 0xff;
 639         entry->pid              = tsk->pid;
 640         entry->t                = ftrace_now(raw_smp_processor_id());
 641         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 642                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 643                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 644                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 645 }
 646
 647 void
 648 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 649                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 650 {
 651         struct trace_entry *entry;
 652         unsigned long irq_flags;
 653
 654         spin_lock_irqsave(&data->lock, irq_flags);
 655         entry                   = tracing_get_trace_entry(tr, data);
 656         tracing_generic_entry_update(entry, flags);
 657         entry->type             = TRACE_FN;
 658         entry->fn.ip            = ip;
 659         entry->fn.parent_ip     = parent_ip;
 660         spin_unlock_irqrestore(&data->lock, irq_flags);
 661
 662         trace_wake_up();
 663 }
 664
 665 void
 666 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 667        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 668 {
 669         if (likely(!atomic_read(&data->disabled)))
 670                 trace_function(tr, data, ip, parent_ip, flags);
 671 }
 672
 673 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
 674
 675 void
 676 __trace_special(void *__tr, void *__data,
 677                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 678 {
 679         struct trace_array_cpu *data = __data;
 680         struct trace_array *tr = __tr;
 681         struct trace_entry *entry;
 682         unsigned long irq_flags;
 683
 684         spin_lock_irqsave(&data->lock, irq_flags);
 685         entry                   = tracing_get_trace_entry(tr, data);
 686         tracing_generic_entry_update(entry, 0);
 687         entry->type             = TRACE_SPECIAL;
 688         entry->special.arg1     = arg1;
 689         entry->special.arg2     = arg2;
 690         entry->special.arg3     = arg3;
 691         spin_unlock_irqrestore(&data->lock, irq_flags);
 692
 693         trace_wake_up();
 694 }
 695
 696 #endif
 697
 698 void __trace_stack(struct trace_array *tr,
 699                    struct trace_array_cpu *data,
 700                    unsigned long flags,
 701                    int skip)
 702 {
 703         struct trace_entry *entry;
 704         struct stack_trace trace;
 705
 706         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 707                 return;
 708
 709         entry                   = tracing_get_trace_entry(tr, data);
 710         tracing_generic_entry_update(entry, flags);
 711         entry->type             = TRACE_STACK;
 712
 713         memset(&entry->stack, 0, sizeof(entry->stack));
 714
 715         trace.nr_entries        = 0;
 716         trace.max_entries       = FTRACE_STACK_ENTRIES;
 717         trace.skip              = skip;
 718         trace.entries           = entry->stack.caller;
 719
 720         save_stack_trace(&trace);
 721 }
 722
 723 void
 724 tracing_sched_switch_trace(struct trace_array *tr,
 725                            struct trace_array_cpu *data,
 726                            struct task_struct *prev,
 727                            struct task_struct *next,
 728                            unsigned long flags)
 729 {
 730         struct trace_entry *entry;
 731         unsigned long irq_flags;
 732
 733         spin_lock_irqsave(&data->lock, irq_flags);
 734         entry                   = tracing_get_trace_entry(tr, data);
 735         tracing_generic_entry_update(entry, flags);
 736         entry->type             = TRACE_CTX;
 737         entry->ctx.prev_pid     = prev->pid;
 738         entry->ctx.prev_prio    = prev->prio;
 739         entry->ctx.prev_state   = prev->state;
 740         entry->ctx.next_pid     = next->pid;
 741         entry->ctx.next_prio    = next->prio;
 742         __trace_stack(tr, data, flags, 4);
 743         spin_unlock_irqrestore(&data->lock, irq_flags);
 744 }
 745
 746 void
 747 tracing_sched_wakeup_trace(struct trace_array *tr,
 748                            struct trace_array_cpu *data,
 749                            struct task_struct *wakee,
 750                            struct task_struct *curr,
 751                            unsigned long flags)
 752 {
 753         struct trace_entry *entry;
 754         unsigned long irq_flags;
 755
 756         spin_lock_irqsave(&data->lock, irq_flags);
 757         entry                   = tracing_get_trace_entry(tr, data);
 758         tracing_generic_entry_update(entry, flags);
 759         entry->type             = TRACE_WAKE;
 760         entry->ctx.prev_pid     = curr->pid;
 761         entry->ctx.prev_prio    = curr->prio;
 762         entry->ctx.prev_state   = curr->state;
 763         entry->ctx.next_pid     = wakee->pid;
 764         entry->ctx.next_prio    = wakee->prio;
 765         __trace_stack(tr, data, flags, 5);
 766         spin_unlock_irqrestore(&data->lock, irq_flags);
 767
 768         trace_wake_up();
 769 }
 770
 771 #ifdef CONFIG_FTRACE
 772 static void
 773 function_trace_call(unsigned long ip, unsigned long parent_ip)
 774 {
 775         struct trace_array *tr = &global_trace;
 776         struct trace_array_cpu *data;
 777         unsigned long flags;
 778         long disabled;
 779         int cpu;
 780
 781         if (unlikely(!tracer_enabled))
 782                 return;
 783
 784         local_irq_save(flags);
 785         cpu = raw_smp_processor_id();
 786         data = tr->data[cpu];
 787         disabled = atomic_inc_return(&data->disabled);
 788
 789         if (likely(disabled == 1))
 790                 trace_function(tr, data, ip, parent_ip, flags);
 791
 792         atomic_dec(&data->disabled);
 793         local_irq_restore(flags);
 794 }
 795
 796 static struct ftrace_ops trace_ops __read_mostly =
 797 {
 798         .func = function_trace_call,
 799 };
 800
 801 void tracing_start_function_trace(void)
 802 {
 803         register_ftrace_function(&trace_ops);
 804 }
 805
 806 void tracing_stop_function_trace(void)
 807 {
 808         unregister_ftrace_function(&trace_ops);
 809 }
 810 #endif
 811
 812 enum trace_file_type {
 813         TRACE_FILE_LAT_FMT      = 1,
 814 };
 815
 816 static struct trace_entry *
 817 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 818                 struct trace_iterator *iter, int cpu)
 819 {
 820         struct page *page;
 821         struct trace_entry *array;
 822
 823         if (iter->next_idx[cpu] >= tr->entries ||
 824             iter->next_idx[cpu] >= data->trace_idx ||
 825             (data->trace_head == data->trace_tail &&
 826              data->trace_head_idx == data->trace_tail_idx))
 827                 return NULL;
 828
 829         if (!iter->next_page[cpu]) {
 830                 /* Initialize the iterator for this cpu trace buffer */
 831                 WARN_ON(!data->trace_tail);
 832                 page = virt_to_page(data->trace_tail);
 833                 iter->next_page[cpu] = &page->lru;
 834                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 835         }
 836
 837         page = list_entry(iter->next_page[cpu], struct page, lru);
 838         BUG_ON(&data->trace_pages == &page->lru);
 839
 840         array = page_address(page);
 841
 842         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 843         return &array[iter->next_page_idx[cpu]];
 844 }
 845
 846 static struct trace_entry *
 847 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 848 {
 849         struct trace_array *tr = iter->tr;
 850         struct trace_entry *ent, *next = NULL;
 851         int next_cpu = -1;
 852         int cpu;
 853
 854         for_each_possible_cpu(cpu) {
 855                 if (!head_page(tr->data[cpu]))
 856                         continue;
 857                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 858                 /*
 859                  * Pick the entry with the smallest timestamp:
 860                  */
 861                 if (ent && (!next || ent->t < next->t)) {
 862                         next = ent;
 863                         next_cpu = cpu;
 864                 }
 865         }
 866
 867         if (ent_cpu)
 868                 *ent_cpu = next_cpu;
 869
 870         return next;
 871 }
 872
 873 static void trace_iterator_increment(struct trace_iterator *iter)
 874 {
 875         iter->idx++;
 876         iter->next_idx[iter->cpu]++;
 877         iter->next_page_idx[iter->cpu]++;
 878
 879         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 880                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 881
 882                 iter->next_page_idx[iter->cpu] = 0;
 883                 iter->next_page[iter->cpu] =
 884                         trace_next_list(data, iter->next_page[iter->cpu]);
 885         }
 886 }
 887
 888 static void trace_consume(struct trace_iterator *iter)
 889 {
 890         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 891
 892         data->trace_tail_idx++;
 893         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 894                 data->trace_tail = trace_next_page(data, data->trace_tail);
 895                 data->trace_tail_idx = 0;
 896         }
 897
 898         /* Check if we empty it, then reset the index */
 899         if (data->trace_head == data->trace_tail &&
 900             data->trace_head_idx == data->trace_tail_idx)
 901                 data->trace_idx = 0;
 902 }
 903
 904 static void *find_next_entry_inc(struct trace_iterator *iter)
 905 {
 906         struct trace_entry *next;
 907         int next_cpu = -1;
 908
 909         next = find_next_entry(iter, &next_cpu);
 910
 911         iter->prev_ent = iter->ent;
 912         iter->prev_cpu = iter->cpu;
 913
 914         iter->ent = next;
 915         iter->cpu = next_cpu;
 916
 917         if (next)
 918                 trace_iterator_increment(iter);
 919
 920         return next ? iter : NULL;
 921 }
 922
 923 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 924 {
 925         struct trace_iterator *iter = m->private;
 926         void *last_ent = iter->ent;
 927         int i = (int)*pos;
 928         void *ent;
 929
 930         (*pos)++;
 931
 932         /* can't go backwards */
 933         if (iter->idx > i)
 934                 return NULL;
 935
 936         if (iter->idx < 0)
 937                 ent = find_next_entry_inc(iter);
 938         else
 939                 ent = iter;
 940
 941         while (ent && iter->idx < i)
 942                 ent = find_next_entry_inc(iter);
 943
 944         iter->pos = *pos;
 945
 946         if (last_ent && !ent)
 947                 seq_puts(m, "\n\nvim:ft=help\n");
 948
 949         return ent;
 950 }
 951
 952 static void *s_start(struct seq_file *m, loff_t *pos)
 953 {
 954         struct trace_iterator *iter = m->private;
 955         void *p = NULL;
 956         loff_t l = 0;
 957         int i;
 958
 959         mutex_lock(&trace_types_lock);
 960
 961         if (!current_trace || current_trace != iter->trace)
 962                 return NULL;
 963
 964         atomic_inc(&trace_record_cmdline_disabled);
 965
 966         /* let the tracer grab locks here if needed */
 967         if (current_trace->start)
 968                 current_trace->start(iter);
 969
 970         if (*pos != iter->pos) {
 971                 iter->ent = NULL;
 972                 iter->cpu = 0;
 973                 iter->idx = -1;
 974                 iter->prev_ent = NULL;
 975                 iter->prev_cpu = -1;
 976
 977                 for_each_possible_cpu(i) {
 978                         iter->next_idx[i] = 0;
 979                         iter->next_page[i] = NULL;
 980                 }
 981
 982                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 983                         ;
 984
 985         } else {
 986                 l = *pos - 1;
 987                 p = s_next(m, p, &l);
 988         }
 989
 990         return p;
 991 }
 992
 993 static void s_stop(struct seq_file *m, void *p)
 994 {
 995         struct trace_iterator *iter = m->private;
 996
 997         atomic_dec(&trace_record_cmdline_disabled);
 998
 999         /* let the tracer release locks here if needed */
1000         if (current_trace && current_trace == iter->trace && iter->trace->stop)
1001                 iter->trace->stop(iter);
1002
1003         mutex_unlock(&trace_types_lock);
1004 }
1005
1006 static int
1007 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1008 {
1009 #ifdef CONFIG_KALLSYMS
1010         char str[KSYM_SYMBOL_LEN];
1011
1012         kallsyms_lookup(address, NULL, NULL, NULL, str);
1013
1014         return trace_seq_printf(s, fmt, str);
1015 #endif
1016         return 1;
1017 }
1018
1019 static int
1020 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1021                      unsigned long address)
1022 {
1023 #ifdef CONFIG_KALLSYMS
1024         char str[KSYM_SYMBOL_LEN];
1025
1026         sprint_symbol(str, address);
1027         return trace_seq_printf(s, fmt, str);
1028 #endif
1029         return 1;
1030 }
1031
1032 #ifndef CONFIG_64BIT
1033 # define IP_FMT "%08lx"
1034 #else
1035 # define IP_FMT "%016lx"
1036 #endif
1037
1038 static int
1039 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1040 {
1041         int ret;
1042
1043         if (!ip)
1044                 return trace_seq_printf(s, "0");
1045
1046         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1047                 ret = seq_print_sym_offset(s, "%s", ip);
1048         else
1049                 ret = seq_print_sym_short(s, "%s", ip);
1050
1051         if (!ret)
1052                 return 0;
1053
1054         if (sym_flags & TRACE_ITER_SYM_ADDR)
1055                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1056         return ret;
1057 }
1058
1059 static void print_lat_help_header(struct seq_file *m)
1060 {
1061         seq_puts(m, "#                _------=> CPU#            \n");
1062         seq_puts(m, "#               / _-----=> irqs-off        \n");
1063         seq_puts(m, "#              | / _----=> need-resched    \n");
1064         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1065         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1066         seq_puts(m, "#              |||| /                      \n");
1067         seq_puts(m, "#              |||||     delay             \n");
1068         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1069         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1070 }
1071
1072 static void print_func_help_header(struct seq_file *m)
1073 {
1074         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1075         seq_puts(m, "#              | |      |          |         |\n");
1076 }
1077
1078
1079 static void
1080 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1081 {
1082         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1083         struct trace_array *tr = iter->tr;
1084         struct trace_array_cpu *data = tr->data[tr->cpu];
1085         struct tracer *type = current_trace;
1086         unsigned long total   = 0;
1087         unsigned long entries = 0;
1088         int cpu;
1089         const char *name = "preemption";
1090
1091         if (type)
1092                 name = type->name;
1093
1094         for_each_possible_cpu(cpu) {
1095                 if (head_page(tr->data[cpu])) {
1096                         total += tr->data[cpu]->trace_idx;
1097                         if (tr->data[cpu]->trace_idx > tr->entries)
1098                                 entries += tr->entries;
1099                         else
1100                                 entries += tr->data[cpu]->trace_idx;
1101                 }
1102         }
1103
1104         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1105                    name, UTS_RELEASE);
1106         seq_puts(m, "-----------------------------------"
1107                  "---------------------------------\n");
1108         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1109                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1110                    nsecs_to_usecs(data->saved_latency),
1111                    entries,
1112                    total,
1113                    tr->cpu,
1114 #if defined(CONFIG_PREEMPT_NONE)
1115                    "server",
1116 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1117                    "desktop",
1118 #elif defined(CONFIG_PREEMPT_DESKTOP)
1119                    "preempt",
1120 #else
1121                    "unknown",
1122 #endif
1123                    /* These are reserved for later use */
1124                    0, 0, 0, 0);
1125 #ifdef CONFIG_SMP
1126         seq_printf(m, " #P:%d)\n", num_online_cpus());
1127 #else
1128         seq_puts(m, ")\n");
1129 #endif
1130         seq_puts(m, "    -----------------\n");
1131         seq_printf(m, "    | task: %.16s-%d "
1132                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1133                    data->comm, data->pid, data->uid, data->nice,
1134                    data->policy, data->rt_priority);
1135         seq_puts(m, "    -----------------\n");
1136
1137         if (data->critical_start) {
1138                 seq_puts(m, " => started at: ");
1139                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1140                 trace_print_seq(m, &iter->seq);
1141                 seq_puts(m, "\n => ended at:   ");
1142                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1143                 trace_print_seq(m, &iter->seq);
1144                 seq_puts(m, "\n");
1145         }
1146
1147         seq_puts(m, "\n");
1148 }
1149
1150 static void
1151 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1152 {
1153         int hardirq, softirq;
1154         char *comm;
1155
1156         comm = trace_find_cmdline(entry->pid);
1157
1158         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1159         trace_seq_printf(s, "%d", cpu);
1160         trace_seq_printf(s, "%c%c",
1161                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1162                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1163
1164         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1165         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1166         if (hardirq && softirq)
1167                 trace_seq_putc(s, 'H');
1168         else {
1169                 if (hardirq)
1170                         trace_seq_putc(s, 'h');
1171                 else {
1172                         if (softirq)
1173                                 trace_seq_putc(s, 's');
1174                         else
1175                                 trace_seq_putc(s, '.');
1176                 }
1177         }
1178
1179         if (entry->preempt_count)
1180                 trace_seq_printf(s, "%x", entry->preempt_count);
1181         else
1182                 trace_seq_puts(s, ".");
1183 }
1184
1185 unsigned long preempt_mark_thresh = 100;
1186
1187 static void
1188 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1189                     unsigned long rel_usecs)
1190 {
1191         trace_seq_printf(s, " %4lldus", abs_usecs);
1192         if (rel_usecs > preempt_mark_thresh)
1193                 trace_seq_puts(s, "!: ");
1194         else if (rel_usecs > 1)
1195                 trace_seq_puts(s, "+: ");
1196         else
1197                 trace_seq_puts(s, " : ");
1198 }
1199
1200 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1201
1202 static int
1203 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1204 {
1205         struct trace_seq *s = &iter->seq;
1206         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1207         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1208         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1209         struct trace_entry *entry = iter->ent;
1210         unsigned long abs_usecs;
1211         unsigned long rel_usecs;
1212         char *comm;
1213         int S;
1214         int i;
1215
1216         if (!next_entry)
1217                 next_entry = entry;
1218         rel_usecs = ns2usecs(next_entry->t - entry->t);
1219         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1220
1221         if (verbose) {
1222                 comm = trace_find_cmdline(entry->pid);
1223                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1224                                  " %ld.%03ldms (+%ld.%03ldms): ",
1225                                  comm,
1226                                  entry->pid, cpu, entry->flags,
1227                                  entry->preempt_count, trace_idx,
1228                                  ns2usecs(entry->t),
1229                                  abs_usecs/1000,
1230                                  abs_usecs % 1000, rel_usecs/1000,
1231                                  rel_usecs % 1000);
1232         } else {
1233                 if (entry->type != TRACE_STACK) {
1234                         lat_print_generic(s, entry, cpu);
1235                         lat_print_timestamp(s, abs_usecs, rel_usecs);
1236                 }
1237         }
1238         switch (entry->type) {
1239         case TRACE_FN:
1240                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1241                 trace_seq_puts(s, " (");
1242                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1243                 trace_seq_puts(s, ")\n");
1244                 break;
1245         case TRACE_CTX:
1246         case TRACE_WAKE:
1247                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1248                         state_to_char[entry->ctx.prev_state] : 'X';
1249                 comm = trace_find_cmdline(entry->ctx.next_pid);
1250                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d %s\n",
1251                                  entry->ctx.prev_pid,
1252                                  entry->ctx.prev_prio,
1253                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1254                                  entry->ctx.next_pid,
1255                                  entry->ctx.next_prio,
1256                                  comm);
1257                 break;
1258         case TRACE_SPECIAL:
1259                 trace_seq_printf(s, " %ld %ld %ld\n",
1260                                  entry->special.arg1,
1261                                  entry->special.arg2,
1262                                  entry->special.arg3);
1263                 break;
1264         case TRACE_STACK:
1265                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1266                         if (i)
1267                                 trace_seq_puts(s, " <= ");
1268                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1269                 }
1270                 trace_seq_puts(s, "\n");
1271                 break;
1272         default:
1273                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1274         }
1275         return 1;
1276 }
1277
1278 static int print_trace_fmt(struct trace_iterator *iter)
1279 {
1280         struct trace_seq *s = &iter->seq;
1281         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1282         struct trace_entry *entry;
1283         unsigned long usec_rem;
1284         unsigned long long t;
1285         unsigned long secs;
1286         char *comm;
1287         int ret;
1288         int S;
1289         int i;
1290
1291         entry = iter->ent;
1292
1293         comm = trace_find_cmdline(iter->ent->pid);
1294
1295         t = ns2usecs(entry->t);
1296         usec_rem = do_div(t, 1000000ULL);
1297         secs = (unsigned long)t;
1298
1299         if (entry->type != TRACE_STACK) {
1300                 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1301                 if (!ret)
1302                         return 0;
1303                 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1304                 if (!ret)
1305                         return 0;
1306                 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1307                 if (!ret)
1308                         return 0;
1309         }
1310
1311         switch (entry->type) {
1312         case TRACE_FN:
1313                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1314                 if (!ret)
1315                         return 0;
1316                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1317                                                 entry->fn.parent_ip) {
1318                         ret = trace_seq_printf(s, " <-");
1319                         if (!ret)
1320                                 return 0;
1321                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1322                                                sym_flags);
1323                         if (!ret)
1324                                 return 0;
1325                 }
1326                 ret = trace_seq_printf(s, "\n");
1327                 if (!ret)
1328                         return 0;
1329                 break;
1330         case TRACE_CTX:
1331         case TRACE_WAKE:
1332                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1333                         state_to_char[entry->ctx.prev_state] : 'X';
1334                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d\n",
1335                                        entry->ctx.prev_pid,
1336                                        entry->ctx.prev_prio,
1337                                        S,
1338                                        entry->type == TRACE_CTX ? "==>" : "  +",
1339                                        entry->ctx.next_pid,
1340                                        entry->ctx.next_prio);
1341                 if (!ret)
1342                         return 0;
1343                 break;
1344         case TRACE_SPECIAL:
1345                 ret = trace_seq_printf(s, " %ld %ld %ld\n",
1346                                  entry->special.arg1,
1347                                  entry->special.arg2,
1348                                  entry->special.arg3);
1349                 if (!ret)
1350                         return 0;
1351                 break;
1352         case TRACE_STACK:
1353                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1354                         if (i) {
1355                                 ret = trace_seq_puts(s, " <= ");
1356                                 if (!ret)
1357                                         return 0;
1358                         }
1359                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1360                                                sym_flags);
1361                         if (!ret)
1362                                 return 0;
1363                 }
1364                 ret = trace_seq_puts(s, "\n");
1365                 if (!ret)
1366                         return 0;
1367                 break;
1368         }
1369         return 1;
1370 }
1371
1372 static int print_raw_fmt(struct trace_iterator *iter)
1373 {
1374         struct trace_seq *s = &iter->seq;
1375         struct trace_entry *entry;
1376         int ret;
1377         int S;
1378
1379         entry = iter->ent;
1380
1381         ret = trace_seq_printf(s, "%d %d %llu ",
1382                 entry->pid, iter->cpu, entry->t);
1383         if (!ret)
1384                 return 0;
1385
1386         switch (entry->type) {
1387         case TRACE_FN:
1388                 ret = trace_seq_printf(s, "%x %x\n",
1389                                         entry->fn.ip, entry->fn.parent_ip);
1390                 if (!ret)
1391                         return 0;
1392                 break;
1393         case TRACE_CTX:
1394         case TRACE_WAKE:
1395                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1396                         state_to_char[entry->ctx.prev_state] : 'X';
1397                 if (entry->type == TRACE_WAKE)
1398                         S = '+';
1399                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1400                                        entry->ctx.prev_pid,
1401                                        entry->ctx.prev_prio,
1402                                        S,
1403                                        entry->ctx.next_pid,
1404                                        entry->ctx.next_prio);
1405                 if (!ret)
1406                         return 0;
1407                 break;
1408         case TRACE_SPECIAL:
1409         case TRACE_STACK:
1410                 ret = trace_seq_printf(s, " %ld %ld %ld\n",
1411                                  entry->special.arg1,
1412                                  entry->special.arg2,
1413                                  entry->special.arg3);
1414                 if (!ret)
1415                         return 0;
1416                 break;
1417         }
1418         return 1;
1419 }
1420
1421 #define SEQ_PUT_FIELD_RET(s, x)                         \
1422 do {                                                    \
1423         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1424                 return 0;                               \
1425 } while (0)
1426
1427 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1428 do {                                                    \
1429         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1430                 return 0;                               \
1431 } while (0)
1432
1433 static int print_hex_fmt(struct trace_iterator *iter)
1434 {
1435         struct trace_seq *s = &iter->seq;
1436         unsigned char newline = '\n';
1437         struct trace_entry *entry;
1438         int S;
1439
1440         entry = iter->ent;
1441
1442         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1443         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1444         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1445
1446         switch (entry->type) {
1447         case TRACE_FN:
1448                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1449                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1450                 break;
1451         case TRACE_CTX:
1452         case TRACE_WAKE:
1453                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1454                         state_to_char[entry->ctx.prev_state] : 'X';
1455                 if (entry->type == TRACE_WAKE)
1456                         S = '+';
1457                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1458                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1459                 SEQ_PUT_HEX_FIELD_RET(s, S);
1460                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1461                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1462                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1463                 break;
1464         case TRACE_SPECIAL:
1465         case TRACE_STACK:
1466                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1467                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1468                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1469                 break;
1470         }
1471         SEQ_PUT_FIELD_RET(s, newline);
1472
1473         return 1;
1474 }
1475
1476 static int print_bin_fmt(struct trace_iterator *iter)
1477 {
1478         struct trace_seq *s = &iter->seq;
1479         struct trace_entry *entry;
1480
1481         entry = iter->ent;
1482
1483         SEQ_PUT_FIELD_RET(s, entry->pid);
1484         SEQ_PUT_FIELD_RET(s, entry->cpu);
1485         SEQ_PUT_FIELD_RET(s, entry->t);
1486
1487         switch (entry->type) {
1488         case TRACE_FN:
1489                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1490                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1491                 break;
1492         case TRACE_CTX:
1493                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1494                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1495                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1496                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1497                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1498                 break;
1499         case TRACE_SPECIAL:
1500         case TRACE_STACK:
1501                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1502                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1503                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1504                 break;
1505         }
1506         return 1;
1507 }
1508
1509 static int trace_empty(struct trace_iterator *iter)
1510 {
1511         struct trace_array_cpu *data;
1512         int cpu;
1513
1514         for_each_possible_cpu(cpu) {
1515                 data = iter->tr->data[cpu];
1516
1517                 if (head_page(data) && data->trace_idx &&
1518                     (data->trace_tail != data->trace_head ||
1519                      data->trace_tail_idx != data->trace_head_idx))
1520                         return 0;
1521         }
1522         return 1;
1523 }
1524
1525 static int print_trace_line(struct trace_iterator *iter)
1526 {
1527         if (trace_flags & TRACE_ITER_BIN)
1528                 return print_bin_fmt(iter);
1529
1530         if (trace_flags & TRACE_ITER_HEX)
1531                 return print_hex_fmt(iter);
1532
1533         if (trace_flags & TRACE_ITER_RAW)
1534                 return print_raw_fmt(iter);
1535
1536         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1537                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1538
1539         return print_trace_fmt(iter);
1540 }
1541
1542 static int s_show(struct seq_file *m, void *v)
1543 {
1544         struct trace_iterator *iter = v;
1545
1546         if (iter->ent == NULL) {
1547                 if (iter->tr) {
1548                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1549                         seq_puts(m, "#\n");
1550                 }
1551                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1552                         /* print nothing if the buffers are empty */
1553                         if (trace_empty(iter))
1554                                 return 0;
1555                         print_trace_header(m, iter);
1556                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1557                                 print_lat_help_header(m);
1558                 } else {
1559                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1560                                 print_func_help_header(m);
1561                 }
1562         } else {
1563                 print_trace_line(iter);
1564                 trace_print_seq(m, &iter->seq);
1565         }
1566
1567         return 0;
1568 }
1569
1570 static struct seq_operations tracer_seq_ops = {
1571         .start          = s_start,
1572         .next           = s_next,
1573         .stop           = s_stop,
1574         .show           = s_show,
1575 };
1576
1577 static struct trace_iterator *
1578 __tracing_open(struct inode *inode, struct file *file, int *ret)
1579 {
1580         struct trace_iterator *iter;
1581
1582         if (tracing_disabled) {
1583                 *ret = -ENODEV;
1584                 return NULL;
1585         }
1586
1587         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1588         if (!iter) {
1589                 *ret = -ENOMEM;
1590                 goto out;
1591         }
1592
1593         mutex_lock(&trace_types_lock);
1594         if (current_trace && current_trace->print_max)
1595                 iter->tr = &max_tr;
1596         else
1597                 iter->tr = inode->i_private;
1598         iter->trace = current_trace;
1599         iter->pos = -1;
1600
1601         /* TODO stop tracer */
1602         *ret = seq_open(file, &tracer_seq_ops);
1603         if (!*ret) {
1604                 struct seq_file *m = file->private_data;
1605                 m->private = iter;
1606
1607                 /* stop the trace while dumping */
1608                 if (iter->tr->ctrl)
1609                         tracer_enabled = 0;
1610
1611                 if (iter->trace && iter->trace->open)
1612                         iter->trace->open(iter);
1613         } else {
1614                 kfree(iter);
1615                 iter = NULL;
1616         }
1617         mutex_unlock(&trace_types_lock);
1618
1619  out:
1620         return iter;
1621 }
1622
1623 int tracing_open_generic(struct inode *inode, struct file *filp)
1624 {
1625         if (tracing_disabled)
1626                 return -ENODEV;
1627
1628         filp->private_data = inode->i_private;
1629         return 0;
1630 }
1631
1632 int tracing_release(struct inode *inode, struct file *file)
1633 {
1634         struct seq_file *m = (struct seq_file *)file->private_data;
1635         struct trace_iterator *iter = m->private;
1636
1637         mutex_lock(&trace_types_lock);
1638         if (iter->trace && iter->trace->close)
1639                 iter->trace->close(iter);
1640
1641         /* reenable tracing if it was previously enabled */
1642         if (iter->tr->ctrl)
1643                 tracer_enabled = 1;
1644         mutex_unlock(&trace_types_lock);
1645
1646         seq_release(inode, file);
1647         kfree(iter);
1648         return 0;
1649 }
1650
1651 static int tracing_open(struct inode *inode, struct file *file)
1652 {
1653         int ret;
1654
1655         __tracing_open(inode, file, &ret);
1656
1657         return ret;
1658 }
1659
1660 static int tracing_lt_open(struct inode *inode, struct file *file)
1661 {
1662         struct trace_iterator *iter;
1663         int ret;
1664
1665         iter = __tracing_open(inode, file, &ret);
1666
1667         if (!ret)
1668                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1669
1670         return ret;
1671 }
1672
1673
1674 static void *
1675 t_next(struct seq_file *m, void *v, loff_t *pos)
1676 {
1677         struct tracer *t = m->private;
1678
1679         (*pos)++;
1680
1681         if (t)
1682                 t = t->next;
1683
1684         m->private = t;
1685
1686         return t;
1687 }
1688
1689 static void *t_start(struct seq_file *m, loff_t *pos)
1690 {
1691         struct tracer *t = m->private;
1692         loff_t l = 0;
1693
1694         mutex_lock(&trace_types_lock);
1695         for (; t && l < *pos; t = t_next(m, t, &l))
1696                 ;
1697
1698         return t;
1699 }
1700
1701 static void t_stop(struct seq_file *m, void *p)
1702 {
1703         mutex_unlock(&trace_types_lock);
1704 }
1705
1706 static int t_show(struct seq_file *m, void *v)
1707 {
1708         struct tracer *t = v;
1709
1710         if (!t)
1711                 return 0;
1712
1713         seq_printf(m, "%s", t->name);
1714         if (t->next)
1715                 seq_putc(m, ' ');
1716         else
1717                 seq_putc(m, '\n');
1718
1719         return 0;
1720 }
1721
1722 static struct seq_operations show_traces_seq_ops = {
1723         .start          = t_start,
1724         .next           = t_next,
1725         .stop           = t_stop,
1726         .show           = t_show,
1727 };
1728
1729 static int show_traces_open(struct inode *inode, struct file *file)
1730 {
1731         int ret;
1732
1733         if (tracing_disabled)
1734                 return -ENODEV;
1735
1736         ret = seq_open(file, &show_traces_seq_ops);
1737         if (!ret) {
1738                 struct seq_file *m = file->private_data;
1739                 m->private = trace_types;
1740         }
1741
1742         return ret;
1743 }
1744
1745 static struct file_operations tracing_fops = {
1746         .open           = tracing_open,
1747         .read           = seq_read,
1748         .llseek         = seq_lseek,
1749         .release        = tracing_release,
1750 };
1751
1752 static struct file_operations tracing_lt_fops = {
1753         .open           = tracing_lt_open,
1754         .read           = seq_read,
1755         .llseek         = seq_lseek,
1756         .release        = tracing_release,
1757 };
1758
1759 static struct file_operations show_traces_fops = {
1760         .open = show_traces_open,
1761         .read = seq_read,
1762         .release = seq_release,
1763 };
1764
1765 static ssize_t
1766 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1767                        size_t cnt, loff_t *ppos)
1768 {
1769         char *buf;
1770         int r = 0;
1771         int len = 0;
1772         int i;
1773
1774         /* calulate max size */
1775         for (i = 0; trace_options[i]; i++) {
1776                 len += strlen(trace_options[i]);
1777                 len += 3; /* "no" and space */
1778         }
1779
1780         /* +2 for \n and \0 */
1781         buf = kmalloc(len + 2, GFP_KERNEL);
1782         if (!buf)
1783                 return -ENOMEM;
1784
1785         for (i = 0; trace_options[i]; i++) {
1786                 if (trace_flags & (1 << i))
1787                         r += sprintf(buf + r, "%s ", trace_options[i]);
1788                 else
1789                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1790         }
1791
1792         r += sprintf(buf + r, "\n");
1793         WARN_ON(r >= len + 2);
1794
1795         r = simple_read_from_buffer(ubuf, cnt, ppos,
1796                                     buf, r);
1797
1798         kfree(buf);
1799
1800         return r;
1801 }
1802
1803 static ssize_t
1804 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1805                         size_t cnt, loff_t *ppos)
1806 {
1807         char buf[64];
1808         char *cmp = buf;
1809         int neg = 0;
1810         int i;
1811
1812         if (cnt > 63)
1813                 cnt = 63;
1814
1815         if (copy_from_user(&buf, ubuf, cnt))
1816                 return -EFAULT;
1817
1818         buf[cnt] = 0;
1819
1820         if (strncmp(buf, "no", 2) == 0) {
1821                 neg = 1;
1822                 cmp += 2;
1823         }
1824
1825         for (i = 0; trace_options[i]; i++) {
1826                 int len = strlen(trace_options[i]);
1827
1828                 if (strncmp(cmp, trace_options[i], len) == 0) {
1829                         if (neg)
1830                                 trace_flags &= ~(1 << i);
1831                         else
1832                                 trace_flags |= (1 << i);
1833                         break;
1834                 }
1835         }
1836
1837         filp->f_pos += cnt;
1838
1839         return cnt;
1840 }
1841
1842 static struct file_operations tracing_iter_fops = {
1843         .open = tracing_open_generic,
1844         .read = tracing_iter_ctrl_read,
1845         .write = tracing_iter_ctrl_write,
1846 };
1847
1848 static const char readme_msg[] =
1849         "tracing mini-HOWTO:\n\n"
1850         "# mkdir /debug\n"
1851         "# mount -t debugfs nodev /debug\n\n"
1852         "# cat /debug/tracing/available_tracers\n"
1853         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1854         "# cat /debug/tracing/current_tracer\n"
1855         "none\n"
1856         "# echo sched_switch > /debug/tracing/current_tracer\n"
1857         "# cat /debug/tracing/current_tracer\n"
1858         "sched_switch\n"
1859         "# cat /debug/tracing/iter_ctrl\n"
1860         "noprint-parent nosym-offset nosym-addr noverbose\n"
1861         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1862         "# echo 1 > /debug/tracing/tracing_enabled\n"
1863         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1864         "echo 0 > /debug/tracing/tracing_enabled\n"
1865 ;
1866
1867 static ssize_t
1868 tracing_readme_read(struct file *filp, char __user *ubuf,
1869                        size_t cnt, loff_t *ppos)
1870 {
1871         return simple_read_from_buffer(ubuf, cnt, ppos,
1872                                         readme_msg, strlen(readme_msg));
1873 }
1874
1875 static struct file_operations tracing_readme_fops = {
1876         .open = tracing_open_generic,
1877         .read = tracing_readme_read,
1878 };
1879
1880 static ssize_t
1881 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1882                   size_t cnt, loff_t *ppos)
1883 {
1884         struct trace_array *tr = filp->private_data;
1885         char buf[64];
1886         int r;
1887
1888         r = sprintf(buf, "%ld\n", tr->ctrl);
1889         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1890 }
1891
1892 static ssize_t
1893 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1894                    size_t cnt, loff_t *ppos)
1895 {
1896         struct trace_array *tr = filp->private_data;
1897         long val;
1898         char buf[64];
1899
1900         if (cnt > 63)
1901                 cnt = 63;
1902
1903         if (copy_from_user(&buf, ubuf, cnt))
1904                 return -EFAULT;
1905
1906         buf[cnt] = 0;
1907
1908         val = simple_strtoul(buf, NULL, 10);
1909
1910         val = !!val;
1911
1912         mutex_lock(&trace_types_lock);
1913         if (tr->ctrl ^ val) {
1914                 if (val)
1915                         tracer_enabled = 1;
1916                 else
1917                         tracer_enabled = 0;
1918
1919                 tr->ctrl = val;
1920
1921                 if (current_trace && current_trace->ctrl_update)
1922                         current_trace->ctrl_update(tr);
1923         }
1924         mutex_unlock(&trace_types_lock);
1925
1926         filp->f_pos += cnt;
1927
1928         return cnt;
1929 }
1930
1931 static ssize_t
1932 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1933                        size_t cnt, loff_t *ppos)
1934 {
1935         char buf[max_tracer_type_len+2];
1936         int r;
1937
1938         mutex_lock(&trace_types_lock);
1939         if (current_trace)
1940                 r = sprintf(buf, "%s\n", current_trace->name);
1941         else
1942                 r = sprintf(buf, "\n");
1943         mutex_unlock(&trace_types_lock);
1944
1945         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1946 }
1947
1948 static ssize_t
1949 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1950                         size_t cnt, loff_t *ppos)
1951 {
1952         struct trace_array *tr = &global_trace;
1953         struct tracer *t;
1954         char buf[max_tracer_type_len+1];
1955         int i;
1956
1957         if (cnt > max_tracer_type_len)
1958                 cnt = max_tracer_type_len;
1959
1960         if (copy_from_user(&buf, ubuf, cnt))
1961                 return -EFAULT;
1962
1963         buf[cnt] = 0;
1964
1965         /* strip ending whitespace. */
1966         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1967                 buf[i] = 0;
1968
1969         mutex_lock(&trace_types_lock);
1970         for (t = trace_types; t; t = t->next) {
1971                 if (strcmp(t->name, buf) == 0)
1972                         break;
1973         }
1974         if (!t || t == current_trace)
1975                 goto out;
1976
1977         if (current_trace && current_trace->reset)
1978                 current_trace->reset(tr);
1979
1980         current_trace = t;
1981         if (t->init)
1982                 t->init(tr);
1983
1984  out:
1985         mutex_unlock(&trace_types_lock);
1986
1987         filp->f_pos += cnt;
1988
1989         return cnt;
1990 }
1991
1992 static ssize_t
1993 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1994                      size_t cnt, loff_t *ppos)
1995 {
1996         unsigned long *ptr = filp->private_data;
1997         char buf[64];
1998         int r;
1999
2000         r = snprintf(buf, 64, "%ld\n",
2001                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2002         if (r > 64)
2003                 r = 64;
2004         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2005 }
2006
2007 static ssize_t
2008 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2009                       size_t cnt, loff_t *ppos)
2010 {
2011         long *ptr = filp->private_data;
2012         long val;
2013         char buf[64];
2014
2015         if (cnt > 63)
2016                 cnt = 63;
2017
2018         if (copy_from_user(&buf, ubuf, cnt))
2019                 return -EFAULT;
2020
2021         buf[cnt] = 0;
2022
2023         val = simple_strtoul(buf, NULL, 10);
2024
2025         *ptr = val * 1000;
2026
2027         return cnt;
2028 }
2029
2030 static atomic_t tracing_reader;
2031
2032 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2033 {
2034         struct trace_iterator *iter;
2035
2036         if (tracing_disabled)
2037                 return -ENODEV;
2038
2039         /* We only allow for reader of the pipe */
2040         if (atomic_inc_return(&tracing_reader) != 1) {
2041                 atomic_dec(&tracing_reader);
2042                 return -EBUSY;
2043         }
2044
2045         /* create a buffer to store the information to pass to userspace */
2046         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2047         if (!iter)
2048                 return -ENOMEM;
2049
2050         iter->tr = &global_trace;
2051
2052         filp->private_data = iter;
2053
2054         return 0;
2055 }
2056
2057 static int tracing_release_pipe(struct inode *inode, struct file *file)
2058 {
2059         struct trace_iterator *iter = file->private_data;
2060
2061         kfree(iter);
2062         atomic_dec(&tracing_reader);
2063
2064         return 0;
2065 }
2066
2067 static unsigned int
2068 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2069 {
2070         struct trace_iterator *iter = filp->private_data;
2071
2072         if (trace_flags & TRACE_ITER_BLOCK) {
2073                 /*
2074                  * Always select as readable when in blocking mode
2075                  */
2076                 return POLLIN | POLLRDNORM;
2077         }
2078         else {
2079                 if (!trace_empty(iter))
2080                         return POLLIN | POLLRDNORM;
2081                 poll_wait(filp, &trace_wait, poll_table);
2082                 if (!trace_empty(iter))
2083                         return POLLIN | POLLRDNORM;
2084
2085                 return 0;
2086         }
2087 }
2088
2089 /*
2090  * Consumer reader.
2091  */
2092 static ssize_t
2093 tracing_read_pipe(struct file *filp, char __user *ubuf,
2094                   size_t cnt, loff_t *ppos)
2095 {
2096         struct trace_iterator *iter = filp->private_data;
2097         struct trace_array_cpu *data;
2098         static cpumask_t mask;
2099         static int start;
2100         unsigned long flags;
2101 #ifdef CONFIG_FTRACE
2102         int ftrace_save;
2103 #endif
2104         int read = 0;
2105         int cpu;
2106         int len;
2107         int ret;
2108
2109         /* return any leftover data */
2110         if (iter->seq.len > start) {
2111                 len = iter->seq.len - start;
2112                 if (cnt > len)
2113                         cnt = len;
2114                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2115                 if (ret)
2116                         cnt = -EFAULT;
2117
2118                 start += len;
2119
2120                 return cnt;
2121         }
2122
2123         trace_seq_reset(&iter->seq);
2124         start = 0;
2125
2126         while (trace_empty(iter)) {
2127                 if (!(trace_flags & TRACE_ITER_BLOCK))
2128                         return -EWOULDBLOCK;
2129                 /*
2130                  * This is a make-shift waitqueue. The reason we don't use
2131                  * an actual wait queue is because:
2132                  *  1) we only ever have one waiter
2133                  *  2) the tracing, traces all functions, we don't want
2134                  *     the overhead of calling wake_up and friends
2135                  *     (and tracing them too)
2136                  *     Anyway, this is really very primitive wakeup.
2137                  */
2138                 set_current_state(TASK_INTERRUPTIBLE);
2139                 iter->tr->waiter = current;
2140
2141                 /* sleep for one second, and try again. */
2142                 schedule_timeout(HZ);
2143
2144                 iter->tr->waiter = NULL;
2145
2146                 if (signal_pending(current))
2147                         return -EINTR;
2148
2149                 /*
2150                  * We block until we read something and tracing is disabled.
2151                  * We still block if tracing is disabled, but we have never
2152                  * read anything. This allows a user to cat this file, and
2153                  * then enable tracing. But after we have read something,
2154                  * we give an EOF when tracing is again disabled.
2155                  *
2156                  * iter->pos will be 0 if we haven't read anything.
2157                  */
2158                 if (!tracer_enabled && iter->pos)
2159                         break;
2160
2161                 continue;
2162         }
2163
2164         /* stop when tracing is finished */
2165         if (trace_empty(iter))
2166                 return 0;
2167
2168         if (cnt >= PAGE_SIZE)
2169                 cnt = PAGE_SIZE - 1;
2170
2171         memset(iter, 0, sizeof(*iter));
2172         iter->tr = &global_trace;
2173         iter->pos = -1;
2174
2175         /*
2176          * We need to stop all tracing on all CPUS to read the
2177          * the next buffer. This is a bit expensive, but is
2178          * not done often. We fill all what we can read,
2179          * and then release the locks again.
2180          */
2181
2182         cpus_clear(mask);
2183         local_irq_save(flags);
2184 #ifdef CONFIG_FTRACE
2185         ftrace_save = ftrace_enabled;
2186         ftrace_enabled = 0;
2187 #endif
2188         smp_wmb();
2189         for_each_possible_cpu(cpu) {
2190                 data = iter->tr->data[cpu];
2191
2192                 if (!head_page(data) || !data->trace_idx)
2193                         continue;
2194
2195                 atomic_inc(&data->disabled);
2196                 cpu_set(cpu, mask);
2197         }
2198
2199         for_each_cpu_mask(cpu, mask) {
2200                 data = iter->tr->data[cpu];
2201                 spin_lock(&data->lock);
2202         }
2203
2204         while (find_next_entry_inc(iter) != NULL) {
2205                 int len = iter->seq.len;
2206
2207                 ret = print_trace_line(iter);
2208                 if (!ret) {
2209                         /* don't print partial lines */
2210                         iter->seq.len = len;
2211                         break;
2212                 }
2213
2214                 trace_consume(iter);
2215
2216                 if (iter->seq.len >= cnt)
2217                         break;
2218         }
2219
2220         for_each_cpu_mask(cpu, mask) {
2221                 data = iter->tr->data[cpu];
2222                 spin_unlock(&data->lock);
2223         }
2224
2225         for_each_cpu_mask(cpu, mask) {
2226                 data = iter->tr->data[cpu];
2227                 atomic_dec(&data->disabled);
2228         }
2229 #ifdef CONFIG_FTRACE
2230         ftrace_enabled = ftrace_save;
2231 #endif
2232         local_irq_restore(flags);
2233
2234         /* Now copy what we have to the user */
2235         read = iter->seq.len;
2236         if (read > cnt)
2237                 read = cnt;
2238
2239         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2240
2241         if (read < iter->seq.len)
2242                 start = read;
2243         else
2244                 trace_seq_reset(&iter->seq);
2245
2246         if (ret)
2247                 read = -EFAULT;
2248
2249         return read;
2250 }
2251
2252 static struct file_operations tracing_max_lat_fops = {
2253         .open           = tracing_open_generic,
2254         .read           = tracing_max_lat_read,
2255         .write          = tracing_max_lat_write,
2256 };
2257
2258 static struct file_operations tracing_ctrl_fops = {
2259         .open           = tracing_open_generic,
2260         .read           = tracing_ctrl_read,
2261         .write          = tracing_ctrl_write,
2262 };
2263
2264 static struct file_operations set_tracer_fops = {
2265         .open           = tracing_open_generic,
2266         .read           = tracing_set_trace_read,
2267         .write          = tracing_set_trace_write,
2268 };
2269
2270 static struct file_operations tracing_pipe_fops = {
2271         .open           = tracing_open_pipe,
2272         .poll           = tracing_poll_pipe,
2273         .read           = tracing_read_pipe,
2274         .release        = tracing_release_pipe,
2275 };
2276
2277 #ifdef CONFIG_DYNAMIC_FTRACE
2278
2279 static ssize_t
2280 tracing_read_long(struct file *filp, char __user *ubuf,
2281                   size_t cnt, loff_t *ppos)
2282 {
2283         unsigned long *p = filp->private_data;
2284         char buf[64];
2285         int r;
2286
2287         r = sprintf(buf, "%ld\n", *p);
2288
2289         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2290 }
2291
2292 static struct file_operations tracing_read_long_fops = {
2293         .open           = tracing_open_generic,
2294         .read           = tracing_read_long,
2295 };
2296 #endif
2297
2298 static struct dentry *d_tracer;
2299
2300 struct dentry *tracing_init_dentry(void)
2301 {
2302         static int once;
2303
2304         if (d_tracer)
2305                 return d_tracer;
2306
2307         d_tracer = debugfs_create_dir("tracing", NULL);
2308
2309         if (!d_tracer && !once) {
2310                 once = 1;
2311                 pr_warning("Could not create debugfs directory 'tracing'\n");
2312                 return NULL;
2313         }
2314
2315         return d_tracer;
2316 }
2317
2318 #ifdef CONFIG_FTRACE_SELFTEST
2319 /* Let selftest have access to static functions in this file */
2320 #include "trace_selftest.c"
2321 #endif
2322
2323 static __init void tracer_init_debugfs(void)
2324 {
2325         struct dentry *d_tracer;
2326         struct dentry *entry;
2327
2328         d_tracer = tracing_init_dentry();
2329
2330         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2331                                     &global_trace, &tracing_ctrl_fops);
2332         if (!entry)
2333                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2334
2335         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2336                                     NULL, &tracing_iter_fops);
2337         if (!entry)
2338                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2339
2340         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2341                                     &global_trace, &tracing_lt_fops);
2342         if (!entry)
2343                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2344
2345         entry = debugfs_create_file("trace", 0444, d_tracer,
2346                                     &global_trace, &tracing_fops);
2347         if (!entry)
2348                 pr_warning("Could not create debugfs 'trace' entry\n");
2349
2350         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2351                                     &global_trace, &show_traces_fops);
2352         if (!entry)
2353                 pr_warning("Could not create debugfs 'trace' entry\n");
2354
2355         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2356                                     &global_trace, &set_tracer_fops);
2357         if (!entry)
2358                 pr_warning("Could not create debugfs 'trace' entry\n");
2359
2360         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2361                                     &tracing_max_latency,
2362                                     &tracing_max_lat_fops);
2363         if (!entry)
2364                 pr_warning("Could not create debugfs "
2365                            "'tracing_max_latency' entry\n");
2366
2367         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2368                                     &tracing_thresh, &tracing_max_lat_fops);
2369         if (!entry)
2370                 pr_warning("Could not create debugfs "
2371                            "'tracing_threash' entry\n");
2372         entry = debugfs_create_file("README", 0644, d_tracer,
2373                                     NULL, &tracing_readme_fops);
2374         if (!entry)
2375                 pr_warning("Could not create debugfs 'README' entry\n");
2376
2377         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2378                                     NULL, &tracing_pipe_fops);
2379         if (!entry)
2380                 pr_warning("Could not create debugfs "
2381                            "'tracing_threash' entry\n");
2382
2383 #ifdef CONFIG_DYNAMIC_FTRACE
2384         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2385                                     &ftrace_update_tot_cnt,
2386                                     &tracing_read_long_fops);
2387         if (!entry)
2388                 pr_warning("Could not create debugfs "
2389                            "'dyn_ftrace_total_info' entry\n");
2390 #endif
2391 }
2392
2393 /* dummy trace to disable tracing */
2394 static struct tracer no_tracer __read_mostly =
2395 {
2396         .name           = "none",
2397 };
2398
2399 static int trace_alloc_page(void)
2400 {
2401         struct trace_array_cpu *data;
2402         struct page *page, *tmp;
2403         LIST_HEAD(pages);
2404         void *array;
2405         int i;
2406
2407         /* first allocate a page for each CPU */
2408         for_each_possible_cpu(i) {
2409                 array = (void *)__get_free_page(GFP_KERNEL);
2410                 if (array == NULL) {
2411                         printk(KERN_ERR "tracer: failed to allocate page"
2412                                "for trace buffer!\n");
2413                         goto free_pages;
2414                 }
2415
2416                 page = virt_to_page(array);
2417                 list_add(&page->lru, &pages);
2418
2419 /* Only allocate if we are actually using the max trace */
2420 #ifdef CONFIG_TRACER_MAX_TRACE
2421                 array = (void *)__get_free_page(GFP_KERNEL);
2422                 if (array == NULL) {
2423                         printk(KERN_ERR "tracer: failed to allocate page"
2424                                "for trace buffer!\n");
2425                         goto free_pages;
2426                 }
2427                 page = virt_to_page(array);
2428                 list_add(&page->lru, &pages);
2429 #endif
2430         }
2431
2432         /* Now that we successfully allocate a page per CPU, add them */
2433         for_each_possible_cpu(i) {
2434                 data = global_trace.data[i];
2435                 spin_lock_init(&data->lock);
2436                 lockdep_set_class(&data->lock, &data->lock_key);
2437                 page = list_entry(pages.next, struct page, lru);
2438                 list_del_init(&page->lru);
2439                 list_add_tail(&page->lru, &data->trace_pages);
2440                 ClearPageLRU(page);
2441
2442 #ifdef CONFIG_TRACER_MAX_TRACE
2443                 data = max_tr.data[i];
2444                 spin_lock_init(&data->lock);
2445                 lockdep_set_class(&data->lock, &data->lock_key);
2446                 page = list_entry(pages.next, struct page, lru);
2447                 list_del_init(&page->lru);
2448                 list_add_tail(&page->lru, &data->trace_pages);
2449                 SetPageLRU(page);
2450 #endif
2451         }
2452         global_trace.entries += ENTRIES_PER_PAGE;
2453
2454         return 0;
2455
2456  free_pages:
2457         list_for_each_entry_safe(page, tmp, &pages, lru) {
2458                 list_del_init(&page->lru);
2459                 __free_page(page);
2460         }
2461         return -ENOMEM;
2462 }
2463
2464 __init static int tracer_alloc_buffers(void)
2465 {
2466         struct trace_array_cpu *data;
2467         void *array;
2468         struct page *page;
2469         int pages = 0;
2470         int ret = -ENOMEM;
2471         int i;
2472
2473         global_trace.ctrl = tracer_enabled;
2474
2475         /* Allocate the first page for all buffers */
2476         for_each_possible_cpu(i) {
2477                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2478                 max_tr.data[i] = &per_cpu(max_data, i);
2479
2480                 array = (void *)__get_free_page(GFP_KERNEL);
2481                 if (array == NULL) {
2482                         printk(KERN_ERR "tracer: failed to allocate page"
2483                                "for trace buffer!\n");
2484                         goto free_buffers;
2485                 }
2486
2487                 /* set the array to the list */
2488                 INIT_LIST_HEAD(&data->trace_pages);
2489                 page = virt_to_page(array);
2490                 list_add(&page->lru, &data->trace_pages);
2491                 /* use the LRU flag to differentiate the two buffers */
2492                 ClearPageLRU(page);
2493
2494 /* Only allocate if we are actually using the max trace */
2495 #ifdef CONFIG_TRACER_MAX_TRACE
2496                 array = (void *)__get_free_page(GFP_KERNEL);
2497                 if (array == NULL) {
2498                         printk(KERN_ERR "tracer: failed to allocate page"
2499                                "for trace buffer!\n");
2500                         goto free_buffers;
2501                 }
2502
2503                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2504                 page = virt_to_page(array);
2505                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2506                 SetPageLRU(page);
2507 #endif
2508         }
2509
2510         /*
2511          * Since we allocate by orders of pages, we may be able to
2512          * round up a bit.
2513          */
2514         global_trace.entries = ENTRIES_PER_PAGE;
2515         pages++;
2516
2517         while (global_trace.entries < trace_nr_entries) {
2518                 if (trace_alloc_page())
2519                         break;
2520                 pages++;
2521         }
2522         max_tr.entries = global_trace.entries;
2523
2524         pr_info("tracer: %d pages allocated for %ld",
2525                 pages, trace_nr_entries);
2526         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2527         pr_info("   actual entries %ld\n", global_trace.entries);
2528
2529         tracer_init_debugfs();
2530
2531         trace_init_cmdlines();
2532
2533         register_tracer(&no_tracer);
2534         current_trace = &no_tracer;
2535
2536         /* All seems OK, enable tracing */
2537         tracing_disabled = 0;
2538
2539         return 0;
2540
2541  free_buffers:
2542         for (i-- ; i >= 0; i--) {
2543                 struct page *page, *tmp;
2544                 struct trace_array_cpu *data = global_trace.data[i];
2545
2546                 if (data) {
2547                         list_for_each_entry_safe(page, tmp,
2548                                                  &data->trace_pages, lru) {
2549                                 list_del_init(&page->lru);
2550                                 __free_page(page);
2551                         }
2552                 }
2553
2554 #ifdef CONFIG_TRACER_MAX_TRACE
2555                 data = max_tr.data[i];
2556                 if (data) {
2557                         list_for_each_entry_safe(page, tmp,
2558                                                  &data->trace_pages, lru) {
2559                                 list_del_init(&page->lru);
2560                                 __free_page(page);
2561                         }
2562                 }
2563 #endif
2564         }
2565         return ret;
2566 }
2567 fs_initcall(tracer_alloc_buffers);