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 /* dummy trace to disable tracing */
  39 static struct tracer no_tracer __read_mostly =
  40 {
  41         .name           = "none",
  42 };
  43
  44 static int trace_alloc_page(void);
  45 static int trace_free_page(void);
  46
  47 static int tracing_disabled = 1;
  48
  49 long
  50 ns2usecs(cycle_t nsec)
  51 {
  52         nsec += 500;
  53         do_div(nsec, 1000);
  54         return nsec;
  55 }
  56
  57 cycle_t ftrace_now(int cpu)
  58 {
  59         return cpu_clock(cpu);
  60 }
  61
  62 static struct trace_array       global_trace;
  63
  64 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  65
  66 static struct trace_array       max_tr;
  67
  68 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  69
  70 static int                      tracer_enabled = 1;
  71 static unsigned long            trace_nr_entries = 65536UL;
  72
  73 static struct tracer            *trace_types __read_mostly;
  74 static struct tracer            *current_trace __read_mostly;
  75 static int                      max_tracer_type_len;
  76
  77 static DEFINE_MUTEX(trace_types_lock);
  78 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
  79
  80 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
  81
  82 void trace_wake_up(void)
  83 {
  84         /*
  85          * The runqueue_is_locked() can fail, but this is the best we
  86          * have for now:
  87          */
  88         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
  89                 wake_up(&trace_wait);
  90 }
  91
  92 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  93
  94 static int __init set_nr_entries(char *str)
  95 {
  96         if (!str)
  97                 return 0;
  98         trace_nr_entries = simple_strtoul(str, &str, 0);
  99         return 1;
 100 }
 101 __setup("trace_entries=", set_nr_entries);
 102
 103 unsigned long nsecs_to_usecs(unsigned long nsecs)
 104 {
 105         return nsecs / 1000;
 106 }
 107
 108 enum trace_flag_type {
 109         TRACE_FLAG_IRQS_OFF             = 0x01,
 110         TRACE_FLAG_NEED_RESCHED         = 0x02,
 111         TRACE_FLAG_HARDIRQ              = 0x04,
 112         TRACE_FLAG_SOFTIRQ              = 0x08,
 113 };
 114
 115 #define TRACE_ITER_SYM_MASK \
 116         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 117
 118 /* These must match the bit postions above */
 119 static const char *trace_options[] = {
 120         "print-parent",
 121         "sym-offset",
 122         "sym-addr",
 123         "verbose",
 124         "raw",
 125         "hex",
 126         "bin",
 127         "block",
 128         "stacktrace",
 129         "sched-tree",
 130         NULL
 131 };
 132
 133 static raw_spinlock_t ftrace_max_lock =
 134         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 135
 136 /*
 137  * Copy the new maximum trace into the separate maximum-trace
 138  * structure. (this way the maximum trace is permanently saved,
 139  * for later retrieval via /debugfs/tracing/latency_trace)
 140  */
 141 static void
 142 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 143 {
 144         struct trace_array_cpu *data = tr->data[cpu];
 145
 146         max_tr.cpu = cpu;
 147         max_tr.time_start = data->preempt_timestamp;
 148
 149         data = max_tr.data[cpu];
 150         data->saved_latency = tracing_max_latency;
 151
 152         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 153         data->pid = tsk->pid;
 154         data->uid = tsk->uid;
 155         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 156         data->policy = tsk->policy;
 157         data->rt_priority = tsk->rt_priority;
 158
 159         /* record this tasks comm */
 160         tracing_record_cmdline(current);
 161 }
 162
 163 void check_pages(struct trace_array_cpu *data)
 164 {
 165         struct page *page, *tmp;
 166
 167         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 168         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 169
 170         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 171                 BUG_ON(page->lru.next->prev != &page->lru);
 172                 BUG_ON(page->lru.prev->next != &page->lru);
 173         }
 174 }
 175
 176 void *head_page(struct trace_array_cpu *data)
 177 {
 178         struct page *page;
 179
 180         check_pages(data);
 181         if (list_empty(&data->trace_pages))
 182                 return NULL;
 183
 184         page = list_entry(data->trace_pages.next, struct page, lru);
 185         BUG_ON(&page->lru == &data->trace_pages);
 186
 187         return page_address(page);
 188 }
 189
 190 int
 191 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 192 {
 193         int len = (PAGE_SIZE - 1) - s->len;
 194         va_list ap;
 195         int ret;
 196
 197         if (!len)
 198                 return 0;
 199
 200         va_start(ap, fmt);
 201         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 202         va_end(ap);
 203
 204         /* If we can't write it all, don't bother writing anything */
 205         if (ret >= len)
 206                 return 0;
 207
 208         s->len += ret;
 209
 210         return len;
 211 }
 212
 213 static int
 214 trace_seq_puts(struct trace_seq *s, const char *str)
 215 {
 216         int len = strlen(str);
 217
 218         if (len > ((PAGE_SIZE - 1) - s->len))
 219                 return 0;
 220
 221         memcpy(s->buffer + s->len, str, len);
 222         s->len += len;
 223
 224         return len;
 225 }
 226
 227 static int
 228 trace_seq_putc(struct trace_seq *s, unsigned char c)
 229 {
 230         if (s->len >= (PAGE_SIZE - 1))
 231                 return 0;
 232
 233         s->buffer[s->len++] = c;
 234
 235         return 1;
 236 }
 237
 238 static int
 239 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 240 {
 241         if (len > ((PAGE_SIZE - 1) - s->len))
 242                 return 0;
 243
 244         memcpy(s->buffer + s->len, mem, len);
 245         s->len += len;
 246
 247         return len;
 248 }
 249
 250 #define HEX_CHARS 17
 251
 252 static int
 253 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 254 {
 255         unsigned char hex[HEX_CHARS];
 256         unsigned char *data;
 257         unsigned char byte;
 258         int i, j;
 259
 260         BUG_ON(len >= HEX_CHARS);
 261
 262         data = mem;
 263
 264 #ifdef __BIG_ENDIAN
 265         for (i = 0, j = 0; i < len; i++) {
 266 #else
 267         for (i = len-1, j = 0; i >= 0; i--) {
 268 #endif
 269                 byte = data[i];
 270
 271                 hex[j]   = byte & 0x0f;
 272                 if (hex[j] >= 10)
 273                         hex[j] += 'a' - 10;
 274                 else
 275                         hex[j] += '0';
 276                 j++;
 277
 278                 hex[j] = byte >> 4;
 279                 if (hex[j] >= 10)
 280                         hex[j] += 'a' - 10;
 281                 else
 282                         hex[j] += '0';
 283                 j++;
 284         }
 285         hex[j] = ' ';
 286         j++;
 287
 288         return trace_seq_putmem(s, hex, j);
 289 }
 290
 291 static void
 292 trace_seq_reset(struct trace_seq *s)
 293 {
 294         s->len = 0;
 295 }
 296
 297 static void
 298 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 299 {
 300         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 301
 302         s->buffer[len] = 0;
 303         seq_puts(m, s->buffer);
 304
 305         trace_seq_reset(s);
 306 }
 307
 308 static void
 309 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 310 {
 311         struct list_head flip_pages;
 312
 313         INIT_LIST_HEAD(&flip_pages);
 314
 315         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 316                 sizeof(struct trace_array_cpu) -
 317                 offsetof(struct trace_array_cpu, trace_head_idx));
 318
 319         check_pages(tr1);
 320         check_pages(tr2);
 321         list_splice_init(&tr1->trace_pages, &flip_pages);
 322         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 323         list_splice_init(&flip_pages, &tr2->trace_pages);
 324         BUG_ON(!list_empty(&flip_pages));
 325         check_pages(tr1);
 326         check_pages(tr2);
 327 }
 328
 329 void
 330 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 331 {
 332         struct trace_array_cpu *data;
 333         int i;
 334
 335         WARN_ON_ONCE(!irqs_disabled());
 336         __raw_spin_lock(&ftrace_max_lock);
 337         /* clear out all the previous traces */
 338         for_each_possible_cpu(i) {
 339                 data = tr->data[i];
 340                 flip_trace(max_tr.data[i], data);
 341                 tracing_reset(data);
 342         }
 343
 344         __update_max_tr(tr, tsk, cpu);
 345         __raw_spin_unlock(&ftrace_max_lock);
 346 }
 347
 348 /**
 349  * update_max_tr_single - only copy one trace over, and reset the rest
 350  * @tr - tracer
 351  * @tsk - task with the latency
 352  * @cpu - the cpu of the buffer to copy.
 353  */
 354 void
 355 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 356 {
 357         struct trace_array_cpu *data = tr->data[cpu];
 358         int i;
 359
 360         WARN_ON_ONCE(!irqs_disabled());
 361         __raw_spin_lock(&ftrace_max_lock);
 362         for_each_possible_cpu(i)
 363                 tracing_reset(max_tr.data[i]);
 364
 365         flip_trace(max_tr.data[cpu], data);
 366         tracing_reset(data);
 367
 368         __update_max_tr(tr, tsk, cpu);
 369         __raw_spin_unlock(&ftrace_max_lock);
 370 }
 371
 372 int register_tracer(struct tracer *type)
 373 {
 374         struct tracer *t;
 375         int len;
 376         int ret = 0;
 377
 378         if (!type->name) {
 379                 pr_info("Tracer must have a name\n");
 380                 return -1;
 381         }
 382
 383         mutex_lock(&trace_types_lock);
 384         for (t = trace_types; t; t = t->next) {
 385                 if (strcmp(type->name, t->name) == 0) {
 386                         /* already found */
 387                         pr_info("Trace %s already registered\n",
 388                                 type->name);
 389                         ret = -1;
 390                         goto out;
 391                 }
 392         }
 393
 394 #ifdef CONFIG_FTRACE_STARTUP_TEST
 395         if (type->selftest) {
 396                 struct tracer *saved_tracer = current_trace;
 397                 struct trace_array_cpu *data;
 398                 struct trace_array *tr = &global_trace;
 399                 int saved_ctrl = tr->ctrl;
 400                 int i;
 401                 /*
 402                  * Run a selftest on this tracer.
 403                  * Here we reset the trace buffer, and set the current
 404                  * tracer to be this tracer. The tracer can then run some
 405                  * internal tracing to verify that everything is in order.
 406                  * If we fail, we do not register this tracer.
 407                  */
 408                 for_each_possible_cpu(i) {
 409                         data = tr->data[i];
 410                         if (!head_page(data))
 411                                 continue;
 412                         tracing_reset(data);
 413                 }
 414                 current_trace = type;
 415                 tr->ctrl = 0;
 416                 /* the test is responsible for initializing and enabling */
 417                 pr_info("Testing tracer %s: ", type->name);
 418                 ret = type->selftest(type, tr);
 419                 /* the test is responsible for resetting too */
 420                 current_trace = saved_tracer;
 421                 tr->ctrl = saved_ctrl;
 422                 if (ret) {
 423                         printk(KERN_CONT "FAILED!\n");
 424                         goto out;
 425                 }
 426                 /* Only reset on passing, to avoid touching corrupted buffers */
 427                 for_each_possible_cpu(i) {
 428                         data = tr->data[i];
 429                         if (!head_page(data))
 430                                 continue;
 431                         tracing_reset(data);
 432                 }
 433                 printk(KERN_CONT "PASSED\n");
 434         }
 435 #endif
 436
 437         type->next = trace_types;
 438         trace_types = type;
 439         len = strlen(type->name);
 440         if (len > max_tracer_type_len)
 441                 max_tracer_type_len = len;
 442
 443  out:
 444         mutex_unlock(&trace_types_lock);
 445
 446         return ret;
 447 }
 448
 449 void unregister_tracer(struct tracer *type)
 450 {
 451         struct tracer **t;
 452         int len;
 453
 454         mutex_lock(&trace_types_lock);
 455         for (t = &trace_types; *t; t = &(*t)->next) {
 456                 if (*t == type)
 457                         goto found;
 458         }
 459         pr_info("Trace %s not registered\n", type->name);
 460         goto out;
 461
 462  found:
 463         *t = (*t)->next;
 464         if (strlen(type->name) != max_tracer_type_len)
 465                 goto out;
 466
 467         max_tracer_type_len = 0;
 468         for (t = &trace_types; *t; t = &(*t)->next) {
 469                 len = strlen((*t)->name);
 470                 if (len > max_tracer_type_len)
 471                         max_tracer_type_len = len;
 472         }
 473  out:
 474         mutex_unlock(&trace_types_lock);
 475 }
 476
 477 void tracing_reset(struct trace_array_cpu *data)
 478 {
 479         data->trace_idx = 0;
 480         data->trace_head = data->trace_tail = head_page(data);
 481         data->trace_head_idx = 0;
 482         data->trace_tail_idx = 0;
 483 }
 484
 485 #define SAVED_CMDLINES 128
 486 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 487 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 488 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 489 static int cmdline_idx;
 490 static DEFINE_SPINLOCK(trace_cmdline_lock);
 491 atomic_t trace_record_cmdline_disabled;
 492
 493 static void trace_init_cmdlines(void)
 494 {
 495         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 496         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 497         cmdline_idx = 0;
 498 }
 499
 500 void trace_stop_cmdline_recording(void);
 501
 502 static void trace_save_cmdline(struct task_struct *tsk)
 503 {
 504         unsigned map;
 505         unsigned idx;
 506
 507         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 508                 return;
 509
 510         /*
 511          * It's not the end of the world if we don't get
 512          * the lock, but we also don't want to spin
 513          * nor do we want to disable interrupts,
 514          * so if we miss here, then better luck next time.
 515          */
 516         if (!spin_trylock(&trace_cmdline_lock))
 517                 return;
 518
 519         idx = map_pid_to_cmdline[tsk->pid];
 520         if (idx >= SAVED_CMDLINES) {
 521                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 522
 523                 map = map_cmdline_to_pid[idx];
 524                 if (map <= PID_MAX_DEFAULT)
 525                         map_pid_to_cmdline[map] = (unsigned)-1;
 526
 527                 map_pid_to_cmdline[tsk->pid] = idx;
 528
 529                 cmdline_idx = idx;
 530         }
 531
 532         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 533
 534         spin_unlock(&trace_cmdline_lock);
 535 }
 536
 537 static char *trace_find_cmdline(int pid)
 538 {
 539         char *cmdline = "<...>";
 540         unsigned map;
 541
 542         if (!pid)
 543                 return "<idle>";
 544
 545         if (pid > PID_MAX_DEFAULT)
 546                 goto out;
 547
 548         map = map_pid_to_cmdline[pid];
 549         if (map >= SAVED_CMDLINES)
 550                 goto out;
 551
 552         cmdline = saved_cmdlines[map];
 553
 554  out:
 555         return cmdline;
 556 }
 557
 558 void tracing_record_cmdline(struct task_struct *tsk)
 559 {
 560         if (atomic_read(&trace_record_cmdline_disabled))
 561                 return;
 562
 563         trace_save_cmdline(tsk);
 564 }
 565
 566 static inline struct list_head *
 567 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 568 {
 569         /*
 570          * Roundrobin - but skip the head (which is not a real page):
 571          */
 572         next = next->next;
 573         if (unlikely(next == &data->trace_pages))
 574                 next = next->next;
 575         BUG_ON(next == &data->trace_pages);
 576
 577         return next;
 578 }
 579
 580 static inline void *
 581 trace_next_page(struct trace_array_cpu *data, void *addr)
 582 {
 583         struct list_head *next;
 584         struct page *page;
 585
 586         page = virt_to_page(addr);
 587
 588         next = trace_next_list(data, &page->lru);
 589         page = list_entry(next, struct page, lru);
 590
 591         return page_address(page);
 592 }
 593
 594 static inline struct trace_entry *
 595 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 596 {
 597         unsigned long idx, idx_next;
 598         struct trace_entry *entry;
 599
 600         data->trace_idx++;
 601         idx = data->trace_head_idx;
 602         idx_next = idx + 1;
 603
 604         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 605
 606         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 607
 608         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 609                 data->trace_head = trace_next_page(data, data->trace_head);
 610                 idx_next = 0;
 611         }
 612
 613         if (data->trace_head == data->trace_tail &&
 614             idx_next == data->trace_tail_idx) {
 615                 /* overrun */
 616                 data->trace_tail_idx++;
 617                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 618                         data->trace_tail =
 619                                 trace_next_page(data, data->trace_tail);
 620                         data->trace_tail_idx = 0;
 621                 }
 622         }
 623
 624         data->trace_head_idx = idx_next;
 625
 626         return entry;
 627 }
 628
 629 static inline void
 630 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 631 {
 632         struct task_struct *tsk = current;
 633         unsigned long pc;
 634
 635         pc = preempt_count();
 636
 637         entry->preempt_count    = pc & 0xff;
 638         entry->pid              = (tsk) ? tsk->pid : 0;
 639         entry->t                = ftrace_now(raw_smp_processor_id());
 640         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 641                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 642                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 643                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 644 }
 645
 646 void
 647 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 648                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 649 {
 650         struct trace_entry *entry;
 651         unsigned long irq_flags;
 652
 653         raw_local_irq_save(irq_flags);
 654         __raw_spin_lock(&data->lock);
 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         __raw_spin_unlock(&data->lock);
 661         raw_local_irq_restore(irq_flags);
 662 }
 663
 664 void
 665 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 666        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 667 {
 668         if (likely(!atomic_read(&data->disabled)))
 669                 trace_function(tr, data, ip, parent_ip, flags);
 670 }
 671
 672 void
 673 __trace_special(void *__tr, void *__data,
 674                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 675 {
 676         struct trace_array_cpu *data = __data;
 677         struct trace_array *tr = __tr;
 678         struct trace_entry *entry;
 679         unsigned long irq_flags;
 680
 681         raw_local_irq_save(irq_flags);
 682         __raw_spin_lock(&data->lock);
 683         entry                   = tracing_get_trace_entry(tr, data);
 684         tracing_generic_entry_update(entry, 0);
 685         entry->type             = TRACE_SPECIAL;
 686         entry->special.arg1     = arg1;
 687         entry->special.arg2     = arg2;
 688         entry->special.arg3     = arg3;
 689         __raw_spin_unlock(&data->lock);
 690         raw_local_irq_restore(irq_flags);
 691
 692         trace_wake_up();
 693 }
 694
 695 void __trace_stack(struct trace_array *tr,
 696                    struct trace_array_cpu *data,
 697                    unsigned long flags,
 698                    int skip)
 699 {
 700         struct trace_entry *entry;
 701         struct stack_trace trace;
 702
 703         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 704                 return;
 705
 706         entry                   = tracing_get_trace_entry(tr, data);
 707         tracing_generic_entry_update(entry, flags);
 708         entry->type             = TRACE_STACK;
 709
 710         memset(&entry->stack, 0, sizeof(entry->stack));
 711
 712         trace.nr_entries        = 0;
 713         trace.max_entries       = FTRACE_STACK_ENTRIES;
 714         trace.skip              = skip;
 715         trace.entries           = entry->stack.caller;
 716
 717         save_stack_trace(&trace);
 718 }
 719
 720 void
 721 tracing_sched_switch_trace(struct trace_array *tr,
 722                            struct trace_array_cpu *data,
 723                            struct task_struct *prev,
 724                            struct task_struct *next,
 725                            unsigned long flags)
 726 {
 727         struct trace_entry *entry;
 728         unsigned long irq_flags;
 729
 730         raw_local_irq_save(irq_flags);
 731         __raw_spin_lock(&data->lock);
 732         entry                   = tracing_get_trace_entry(tr, data);
 733         tracing_generic_entry_update(entry, flags);
 734         entry->type             = TRACE_CTX;
 735         entry->ctx.prev_pid     = prev->pid;
 736         entry->ctx.prev_prio    = prev->prio;
 737         entry->ctx.prev_state   = prev->state;
 738         entry->ctx.next_pid     = next->pid;
 739         entry->ctx.next_prio    = next->prio;
 740         entry->ctx.next_state   = next->state;
 741         __trace_stack(tr, data, flags, 4);
 742         __raw_spin_unlock(&data->lock);
 743         raw_local_irq_restore(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         raw_local_irq_save(irq_flags);
 757         __raw_spin_lock(&data->lock);
 758         entry                   = tracing_get_trace_entry(tr, data);
 759         tracing_generic_entry_update(entry, flags);
 760         entry->type             = TRACE_WAKE;
 761         entry->ctx.prev_pid     = curr->pid;
 762         entry->ctx.prev_prio    = curr->prio;
 763         entry->ctx.prev_state   = curr->state;
 764         entry->ctx.next_pid     = wakee->pid;
 765         entry->ctx.next_prio    = wakee->prio;
 766         entry->ctx.next_state   = wakee->state;
 767         __trace_stack(tr, data, flags, 5);
 768         __raw_spin_unlock(&data->lock);
 769         raw_local_irq_restore(irq_flags);
 770
 771         trace_wake_up();
 772 }
 773
 774 #ifdef CONFIG_FTRACE
 775 static void
 776 function_trace_call(unsigned long ip, unsigned long parent_ip)
 777 {
 778         struct trace_array *tr = &global_trace;
 779         struct trace_array_cpu *data;
 780         unsigned long flags;
 781         long disabled;
 782         int cpu;
 783
 784         if (unlikely(!tracer_enabled))
 785                 return;
 786
 787         local_irq_save(flags);
 788         cpu = raw_smp_processor_id();
 789         data = tr->data[cpu];
 790         disabled = atomic_inc_return(&data->disabled);
 791
 792         if (likely(disabled == 1))
 793                 trace_function(tr, data, ip, parent_ip, flags);
 794
 795         atomic_dec(&data->disabled);
 796         local_irq_restore(flags);
 797 }
 798
 799 static struct ftrace_ops trace_ops __read_mostly =
 800 {
 801         .func = function_trace_call,
 802 };
 803
 804 void tracing_start_function_trace(void)
 805 {
 806         register_ftrace_function(&trace_ops);
 807 }
 808
 809 void tracing_stop_function_trace(void)
 810 {
 811         unregister_ftrace_function(&trace_ops);
 812 }
 813 #endif
 814
 815 enum trace_file_type {
 816         TRACE_FILE_LAT_FMT      = 1,
 817 };
 818
 819 static struct trace_entry *
 820 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 821                 struct trace_iterator *iter, int cpu)
 822 {
 823         struct page *page;
 824         struct trace_entry *array;
 825
 826         if (iter->next_idx[cpu] >= tr->entries ||
 827             iter->next_idx[cpu] >= data->trace_idx ||
 828             (data->trace_head == data->trace_tail &&
 829              data->trace_head_idx == data->trace_tail_idx))
 830                 return NULL;
 831
 832         if (!iter->next_page[cpu]) {
 833                 /* Initialize the iterator for this cpu trace buffer */
 834                 WARN_ON(!data->trace_tail);
 835                 page = virt_to_page(data->trace_tail);
 836                 iter->next_page[cpu] = &page->lru;
 837                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 838         }
 839
 840         page = list_entry(iter->next_page[cpu], struct page, lru);
 841         BUG_ON(&data->trace_pages == &page->lru);
 842
 843         array = page_address(page);
 844
 845         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 846         return &array[iter->next_page_idx[cpu]];
 847 }
 848
 849 static struct trace_entry *
 850 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 851 {
 852         struct trace_array *tr = iter->tr;
 853         struct trace_entry *ent, *next = NULL;
 854         int next_cpu = -1;
 855         int cpu;
 856
 857         for_each_possible_cpu(cpu) {
 858                 if (!head_page(tr->data[cpu]))
 859                         continue;
 860                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 861                 /*
 862                  * Pick the entry with the smallest timestamp:
 863                  */
 864                 if (ent && (!next || ent->t < next->t)) {
 865                         next = ent;
 866                         next_cpu = cpu;
 867                 }
 868         }
 869
 870         if (ent_cpu)
 871                 *ent_cpu = next_cpu;
 872
 873         return next;
 874 }
 875
 876 static void trace_iterator_increment(struct trace_iterator *iter)
 877 {
 878         iter->idx++;
 879         iter->next_idx[iter->cpu]++;
 880         iter->next_page_idx[iter->cpu]++;
 881
 882         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 883                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 884
 885                 iter->next_page_idx[iter->cpu] = 0;
 886                 iter->next_page[iter->cpu] =
 887                         trace_next_list(data, iter->next_page[iter->cpu]);
 888         }
 889 }
 890
 891 static void trace_consume(struct trace_iterator *iter)
 892 {
 893         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 894
 895         data->trace_tail_idx++;
 896         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 897                 data->trace_tail = trace_next_page(data, data->trace_tail);
 898                 data->trace_tail_idx = 0;
 899         }
 900
 901         /* Check if we empty it, then reset the index */
 902         if (data->trace_head == data->trace_tail &&
 903             data->trace_head_idx == data->trace_tail_idx)
 904                 data->trace_idx = 0;
 905 }
 906
 907 static void *find_next_entry_inc(struct trace_iterator *iter)
 908 {
 909         struct trace_entry *next;
 910         int next_cpu = -1;
 911
 912         next = find_next_entry(iter, &next_cpu);
 913
 914         iter->prev_ent = iter->ent;
 915         iter->prev_cpu = iter->cpu;
 916
 917         iter->ent = next;
 918         iter->cpu = next_cpu;
 919
 920         if (next)
 921                 trace_iterator_increment(iter);
 922
 923         return next ? iter : NULL;
 924 }
 925
 926 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 927 {
 928         struct trace_iterator *iter = m->private;
 929         void *last_ent = iter->ent;
 930         int i = (int)*pos;
 931         void *ent;
 932
 933         (*pos)++;
 934
 935         /* can't go backwards */
 936         if (iter->idx > i)
 937                 return NULL;
 938
 939         if (iter->idx < 0)
 940                 ent = find_next_entry_inc(iter);
 941         else
 942                 ent = iter;
 943
 944         while (ent && iter->idx < i)
 945                 ent = find_next_entry_inc(iter);
 946
 947         iter->pos = *pos;
 948
 949         if (last_ent && !ent)
 950                 seq_puts(m, "\n\nvim:ft=help\n");
 951
 952         return ent;
 953 }
 954
 955 static void *s_start(struct seq_file *m, loff_t *pos)
 956 {
 957         struct trace_iterator *iter = m->private;
 958         void *p = NULL;
 959         loff_t l = 0;
 960         int i;
 961
 962         mutex_lock(&trace_types_lock);
 963
 964         if (!current_trace || current_trace != iter->trace) {
 965                 mutex_unlock(&trace_types_lock);
 966                 return NULL;
 967         }
 968
 969         atomic_inc(&trace_record_cmdline_disabled);
 970
 971         /* let the tracer grab locks here if needed */
 972         if (current_trace->start)
 973                 current_trace->start(iter);
 974
 975         if (*pos != iter->pos) {
 976                 iter->ent = NULL;
 977                 iter->cpu = 0;
 978                 iter->idx = -1;
 979                 iter->prev_ent = NULL;
 980                 iter->prev_cpu = -1;
 981
 982                 for_each_possible_cpu(i) {
 983                         iter->next_idx[i] = 0;
 984                         iter->next_page[i] = NULL;
 985                 }
 986
 987                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 988                         ;
 989
 990         } else {
 991                 l = *pos - 1;
 992                 p = s_next(m, p, &l);
 993         }
 994
 995         return p;
 996 }
 997
 998 static void s_stop(struct seq_file *m, void *p)
 999 {
1000         struct trace_iterator *iter = m->private;
1001
1002         atomic_dec(&trace_record_cmdline_disabled);
1003
1004         /* let the tracer release locks here if needed */
1005         if (current_trace && current_trace == iter->trace && iter->trace->stop)
1006                 iter->trace->stop(iter);
1007
1008         mutex_unlock(&trace_types_lock);
1009 }
1010
1011 static int
1012 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1013 {
1014 #ifdef CONFIG_KALLSYMS
1015         char str[KSYM_SYMBOL_LEN];
1016
1017         kallsyms_lookup(address, NULL, NULL, NULL, str);
1018
1019         return trace_seq_printf(s, fmt, str);
1020 #endif
1021         return 1;
1022 }
1023
1024 static int
1025 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1026                      unsigned long address)
1027 {
1028 #ifdef CONFIG_KALLSYMS
1029         char str[KSYM_SYMBOL_LEN];
1030
1031         sprint_symbol(str, address);
1032         return trace_seq_printf(s, fmt, str);
1033 #endif
1034         return 1;
1035 }
1036
1037 #ifndef CONFIG_64BIT
1038 # define IP_FMT "%08lx"
1039 #else
1040 # define IP_FMT "%016lx"
1041 #endif
1042
1043 static int
1044 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1045 {
1046         int ret;
1047
1048         if (!ip)
1049                 return trace_seq_printf(s, "0");
1050
1051         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1052                 ret = seq_print_sym_offset(s, "%s", ip);
1053         else
1054                 ret = seq_print_sym_short(s, "%s", ip);
1055
1056         if (!ret)
1057                 return 0;
1058
1059         if (sym_flags & TRACE_ITER_SYM_ADDR)
1060                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1061         return ret;
1062 }
1063
1064 static void print_lat_help_header(struct seq_file *m)
1065 {
1066         seq_puts(m, "#                _------=> CPU#            \n");
1067         seq_puts(m, "#               / _-----=> irqs-off        \n");
1068         seq_puts(m, "#              | / _----=> need-resched    \n");
1069         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1070         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1071         seq_puts(m, "#              |||| /                      \n");
1072         seq_puts(m, "#              |||||     delay             \n");
1073         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1074         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1075 }
1076
1077 static void print_func_help_header(struct seq_file *m)
1078 {
1079         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1080         seq_puts(m, "#              | |      |          |         |\n");
1081 }
1082
1083
1084 static void
1085 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1086 {
1087         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1088         struct trace_array *tr = iter->tr;
1089         struct trace_array_cpu *data = tr->data[tr->cpu];
1090         struct tracer *type = current_trace;
1091         unsigned long total   = 0;
1092         unsigned long entries = 0;
1093         int cpu;
1094         const char *name = "preemption";
1095
1096         if (type)
1097                 name = type->name;
1098
1099         for_each_possible_cpu(cpu) {
1100                 if (head_page(tr->data[cpu])) {
1101                         total += tr->data[cpu]->trace_idx;
1102                         if (tr->data[cpu]->trace_idx > tr->entries)
1103                                 entries += tr->entries;
1104                         else
1105                                 entries += tr->data[cpu]->trace_idx;
1106                 }
1107         }
1108
1109         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1110                    name, UTS_RELEASE);
1111         seq_puts(m, "-----------------------------------"
1112                  "---------------------------------\n");
1113         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1114                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1115                    nsecs_to_usecs(data->saved_latency),
1116                    entries,
1117                    total,
1118                    tr->cpu,
1119 #if defined(CONFIG_PREEMPT_NONE)
1120                    "server",
1121 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1122                    "desktop",
1123 #elif defined(CONFIG_PREEMPT_DESKTOP)
1124                    "preempt",
1125 #else
1126                    "unknown",
1127 #endif
1128                    /* These are reserved for later use */
1129                    0, 0, 0, 0);
1130 #ifdef CONFIG_SMP
1131         seq_printf(m, " #P:%d)\n", num_online_cpus());
1132 #else
1133         seq_puts(m, ")\n");
1134 #endif
1135         seq_puts(m, "    -----------------\n");
1136         seq_printf(m, "    | task: %.16s-%d "
1137                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1138                    data->comm, data->pid, data->uid, data->nice,
1139                    data->policy, data->rt_priority);
1140         seq_puts(m, "    -----------------\n");
1141
1142         if (data->critical_start) {
1143                 seq_puts(m, " => started at: ");
1144                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1145                 trace_print_seq(m, &iter->seq);
1146                 seq_puts(m, "\n => ended at:   ");
1147                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1148                 trace_print_seq(m, &iter->seq);
1149                 seq_puts(m, "\n");
1150         }
1151
1152         seq_puts(m, "\n");
1153 }
1154
1155 static void
1156 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1157 {
1158         int hardirq, softirq;
1159         char *comm;
1160
1161         comm = trace_find_cmdline(entry->pid);
1162
1163         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1164         trace_seq_printf(s, "%d", cpu);
1165         trace_seq_printf(s, "%c%c",
1166                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1167                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1168
1169         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1170         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1171         if (hardirq && softirq)
1172                 trace_seq_putc(s, 'H');
1173         else {
1174                 if (hardirq)
1175                         trace_seq_putc(s, 'h');
1176                 else {
1177                         if (softirq)
1178                                 trace_seq_putc(s, 's');
1179                         else
1180                                 trace_seq_putc(s, '.');
1181                 }
1182         }
1183
1184         if (entry->preempt_count)
1185                 trace_seq_printf(s, "%x", entry->preempt_count);
1186         else
1187                 trace_seq_puts(s, ".");
1188 }
1189
1190 unsigned long preempt_mark_thresh = 100;
1191
1192 static void
1193 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1194                     unsigned long rel_usecs)
1195 {
1196         trace_seq_printf(s, " %4lldus", abs_usecs);
1197         if (rel_usecs > preempt_mark_thresh)
1198                 trace_seq_puts(s, "!: ");
1199         else if (rel_usecs > 1)
1200                 trace_seq_puts(s, "+: ");
1201         else
1202                 trace_seq_puts(s, " : ");
1203 }
1204
1205 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1206
1207 static int
1208 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1209 {
1210         struct trace_seq *s = &iter->seq;
1211         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1212         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1213         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1214         struct trace_entry *entry = iter->ent;
1215         unsigned long abs_usecs;
1216         unsigned long rel_usecs;
1217         char *comm;
1218         int S, T;
1219         int i;
1220         unsigned state;
1221
1222         if (!next_entry)
1223                 next_entry = entry;
1224         rel_usecs = ns2usecs(next_entry->t - entry->t);
1225         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1226
1227         if (verbose) {
1228                 comm = trace_find_cmdline(entry->pid);
1229                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1230                                  " %ld.%03ldms (+%ld.%03ldms): ",
1231                                  comm,
1232                                  entry->pid, cpu, entry->flags,
1233                                  entry->preempt_count, trace_idx,
1234                                  ns2usecs(entry->t),
1235                                  abs_usecs/1000,
1236                                  abs_usecs % 1000, rel_usecs/1000,
1237                                  rel_usecs % 1000);
1238         } else {
1239                 lat_print_generic(s, entry, cpu);
1240                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1241         }
1242         switch (entry->type) {
1243         case TRACE_FN:
1244                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1245                 trace_seq_puts(s, " (");
1246                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1247                 trace_seq_puts(s, ")\n");
1248                 break;
1249         case TRACE_CTX:
1250         case TRACE_WAKE:
1251                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1252                         state_to_char[entry->ctx.next_state] : 'X';
1253
1254                 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1255                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1256                 comm = trace_find_cmdline(entry->ctx.next_pid);
1257                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1258                                  entry->ctx.prev_pid,
1259                                  entry->ctx.prev_prio,
1260                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1261                                  entry->ctx.next_pid,
1262                                  entry->ctx.next_prio,
1263                                  T, comm);
1264                 break;
1265         case TRACE_SPECIAL:
1266                 trace_seq_printf(s, "# %ld %ld %ld\n",
1267                                  entry->special.arg1,
1268                                  entry->special.arg2,
1269                                  entry->special.arg3);
1270                 break;
1271         case TRACE_STACK:
1272                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1273                         if (i)
1274                                 trace_seq_puts(s, " <= ");
1275                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1276                 }
1277                 trace_seq_puts(s, "\n");
1278                 break;
1279         default:
1280                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1281         }
1282         return 1;
1283 }
1284
1285 static int print_trace_fmt(struct trace_iterator *iter)
1286 {
1287         struct trace_seq *s = &iter->seq;
1288         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1289         struct trace_entry *entry;
1290         unsigned long usec_rem;
1291         unsigned long long t;
1292         unsigned long secs;
1293         char *comm;
1294         int ret;
1295         int S, T;
1296         int i;
1297
1298         entry = iter->ent;
1299
1300         comm = trace_find_cmdline(iter->ent->pid);
1301
1302         t = ns2usecs(entry->t);
1303         usec_rem = do_div(t, 1000000ULL);
1304         secs = (unsigned long)t;
1305
1306         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1307         if (!ret)
1308                 return 0;
1309         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1310         if (!ret)
1311                 return 0;
1312         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1313         if (!ret)
1314                 return 0;
1315
1316         switch (entry->type) {
1317         case TRACE_FN:
1318                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1319                 if (!ret)
1320                         return 0;
1321                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1322                                                 entry->fn.parent_ip) {
1323                         ret = trace_seq_printf(s, " <-");
1324                         if (!ret)
1325                                 return 0;
1326                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1327                                                sym_flags);
1328                         if (!ret)
1329                                 return 0;
1330                 }
1331                 ret = trace_seq_printf(s, "\n");
1332                 if (!ret)
1333                         return 0;
1334                 break;
1335         case TRACE_CTX:
1336         case TRACE_WAKE:
1337                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1338                         state_to_char[entry->ctx.prev_state] : 'X';
1339                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1340                         state_to_char[entry->ctx.next_state] : 'X';
1341                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1342                                        entry->ctx.prev_pid,
1343                                        entry->ctx.prev_prio,
1344                                        S,
1345                                        entry->type == TRACE_CTX ? "==>" : "  +",
1346                                        entry->ctx.next_pid,
1347                                        entry->ctx.next_prio,
1348                                        T);
1349                 if (!ret)
1350                         return 0;
1351                 break;
1352         case TRACE_SPECIAL:
1353                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1354                                  entry->special.arg1,
1355                                  entry->special.arg2,
1356                                  entry->special.arg3);
1357                 if (!ret)
1358                         return 0;
1359                 break;
1360         case TRACE_STACK:
1361                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1362                         if (i) {
1363                                 ret = trace_seq_puts(s, " <= ");
1364                                 if (!ret)
1365                                         return 0;
1366                         }
1367                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1368                                                sym_flags);
1369                         if (!ret)
1370                                 return 0;
1371                 }
1372                 ret = trace_seq_puts(s, "\n");
1373                 if (!ret)
1374                         return 0;
1375                 break;
1376         }
1377         return 1;
1378 }
1379
1380 static int print_raw_fmt(struct trace_iterator *iter)
1381 {
1382         struct trace_seq *s = &iter->seq;
1383         struct trace_entry *entry;
1384         int ret;
1385         int S, T;
1386
1387         entry = iter->ent;
1388
1389         ret = trace_seq_printf(s, "%d %d %llu ",
1390                 entry->pid, iter->cpu, entry->t);
1391         if (!ret)
1392                 return 0;
1393
1394         switch (entry->type) {
1395         case TRACE_FN:
1396                 ret = trace_seq_printf(s, "%x %x\n",
1397                                         entry->fn.ip, entry->fn.parent_ip);
1398                 if (!ret)
1399                         return 0;
1400                 break;
1401         case TRACE_CTX:
1402         case TRACE_WAKE:
1403                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1404                         state_to_char[entry->ctx.prev_state] : 'X';
1405                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1406                         state_to_char[entry->ctx.next_state] : 'X';
1407                 if (entry->type == TRACE_WAKE)
1408                         S = '+';
1409                 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1410                                        entry->ctx.prev_pid,
1411                                        entry->ctx.prev_prio,
1412                                        S,
1413                                        entry->ctx.next_pid,
1414                                        entry->ctx.next_prio,
1415                                        T);
1416                 if (!ret)
1417                         return 0;
1418                 break;
1419         case TRACE_SPECIAL:
1420         case TRACE_STACK:
1421                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1422                                  entry->special.arg1,
1423                                  entry->special.arg2,
1424                                  entry->special.arg3);
1425                 if (!ret)
1426                         return 0;
1427                 break;
1428         }
1429         return 1;
1430 }
1431
1432 #define SEQ_PUT_FIELD_RET(s, x)                         \
1433 do {                                                    \
1434         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1435                 return 0;                               \
1436 } while (0)
1437
1438 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1439 do {                                                    \
1440         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1441                 return 0;                               \
1442 } while (0)
1443
1444 static int print_hex_fmt(struct trace_iterator *iter)
1445 {
1446         struct trace_seq *s = &iter->seq;
1447         unsigned char newline = '\n';
1448         struct trace_entry *entry;
1449         int S, T;
1450
1451         entry = iter->ent;
1452
1453         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1454         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1455         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1456
1457         switch (entry->type) {
1458         case TRACE_FN:
1459                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1460                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1461                 break;
1462         case TRACE_CTX:
1463         case TRACE_WAKE:
1464                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1465                         state_to_char[entry->ctx.prev_state] : 'X';
1466                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1467                         state_to_char[entry->ctx.next_state] : 'X';
1468                 if (entry->type == TRACE_WAKE)
1469                         S = '+';
1470                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1471                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1472                 SEQ_PUT_HEX_FIELD_RET(s, S);
1473                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1474                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1475                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1476                 SEQ_PUT_HEX_FIELD_RET(s, T);
1477                 break;
1478         case TRACE_SPECIAL:
1479         case TRACE_STACK:
1480                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1481                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1482                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1483                 break;
1484         }
1485         SEQ_PUT_FIELD_RET(s, newline);
1486
1487         return 1;
1488 }
1489
1490 static int print_bin_fmt(struct trace_iterator *iter)
1491 {
1492         struct trace_seq *s = &iter->seq;
1493         struct trace_entry *entry;
1494
1495         entry = iter->ent;
1496
1497         SEQ_PUT_FIELD_RET(s, entry->pid);
1498         SEQ_PUT_FIELD_RET(s, entry->cpu);
1499         SEQ_PUT_FIELD_RET(s, entry->t);
1500
1501         switch (entry->type) {
1502         case TRACE_FN:
1503                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1504                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1505                 break;
1506         case TRACE_CTX:
1507                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1508                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1509                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1510                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1511                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1512                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1513                 break;
1514         case TRACE_SPECIAL:
1515         case TRACE_STACK:
1516                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1517                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1518                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1519                 break;
1520         }
1521         return 1;
1522 }
1523
1524 static int trace_empty(struct trace_iterator *iter)
1525 {
1526         struct trace_array_cpu *data;
1527         int cpu;
1528
1529         for_each_possible_cpu(cpu) {
1530                 data = iter->tr->data[cpu];
1531
1532                 if (head_page(data) && data->trace_idx &&
1533                     (data->trace_tail != data->trace_head ||
1534                      data->trace_tail_idx != data->trace_head_idx))
1535                         return 0;
1536         }
1537         return 1;
1538 }
1539
1540 static int print_trace_line(struct trace_iterator *iter)
1541 {
1542         if (iter->trace && iter->trace->print_line)
1543                 return iter->trace->print_line(iter);
1544
1545         if (trace_flags & TRACE_ITER_BIN)
1546                 return print_bin_fmt(iter);
1547
1548         if (trace_flags & TRACE_ITER_HEX)
1549                 return print_hex_fmt(iter);
1550
1551         if (trace_flags & TRACE_ITER_RAW)
1552                 return print_raw_fmt(iter);
1553
1554         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1555                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1556
1557         return print_trace_fmt(iter);
1558 }
1559
1560 static int s_show(struct seq_file *m, void *v)
1561 {
1562         struct trace_iterator *iter = v;
1563
1564         if (iter->ent == NULL) {
1565                 if (iter->tr) {
1566                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1567                         seq_puts(m, "#\n");
1568                 }
1569                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1570                         /* print nothing if the buffers are empty */
1571                         if (trace_empty(iter))
1572                                 return 0;
1573                         print_trace_header(m, iter);
1574                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1575                                 print_lat_help_header(m);
1576                 } else {
1577                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1578                                 print_func_help_header(m);
1579                 }
1580         } else {
1581                 print_trace_line(iter);
1582                 trace_print_seq(m, &iter->seq);
1583         }
1584
1585         return 0;
1586 }
1587
1588 static struct seq_operations tracer_seq_ops = {
1589         .start          = s_start,
1590         .next           = s_next,
1591         .stop           = s_stop,
1592         .show           = s_show,
1593 };
1594
1595 static struct trace_iterator *
1596 __tracing_open(struct inode *inode, struct file *file, int *ret)
1597 {
1598         struct trace_iterator *iter;
1599
1600         if (tracing_disabled) {
1601                 *ret = -ENODEV;
1602                 return NULL;
1603         }
1604
1605         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1606         if (!iter) {
1607                 *ret = -ENOMEM;
1608                 goto out;
1609         }
1610
1611         mutex_lock(&trace_types_lock);
1612         if (current_trace && current_trace->print_max)
1613                 iter->tr = &max_tr;
1614         else
1615                 iter->tr = inode->i_private;
1616         iter->trace = current_trace;
1617         iter->pos = -1;
1618
1619         /* TODO stop tracer */
1620         *ret = seq_open(file, &tracer_seq_ops);
1621         if (!*ret) {
1622                 struct seq_file *m = file->private_data;
1623                 m->private = iter;
1624
1625                 /* stop the trace while dumping */
1626                 if (iter->tr->ctrl)
1627                         tracer_enabled = 0;
1628
1629                 if (iter->trace && iter->trace->open)
1630                         iter->trace->open(iter);
1631         } else {
1632                 kfree(iter);
1633                 iter = NULL;
1634         }
1635         mutex_unlock(&trace_types_lock);
1636
1637  out:
1638         return iter;
1639 }
1640
1641 int tracing_open_generic(struct inode *inode, struct file *filp)
1642 {
1643         if (tracing_disabled)
1644                 return -ENODEV;
1645
1646         filp->private_data = inode->i_private;
1647         return 0;
1648 }
1649
1650 int tracing_release(struct inode *inode, struct file *file)
1651 {
1652         struct seq_file *m = (struct seq_file *)file->private_data;
1653         struct trace_iterator *iter = m->private;
1654
1655         mutex_lock(&trace_types_lock);
1656         if (iter->trace && iter->trace->close)
1657                 iter->trace->close(iter);
1658
1659         /* reenable tracing if it was previously enabled */
1660         if (iter->tr->ctrl)
1661                 tracer_enabled = 1;
1662         mutex_unlock(&trace_types_lock);
1663
1664         seq_release(inode, file);
1665         kfree(iter);
1666         return 0;
1667 }
1668
1669 static int tracing_open(struct inode *inode, struct file *file)
1670 {
1671         int ret;
1672
1673         __tracing_open(inode, file, &ret);
1674
1675         return ret;
1676 }
1677
1678 static int tracing_lt_open(struct inode *inode, struct file *file)
1679 {
1680         struct trace_iterator *iter;
1681         int ret;
1682
1683         iter = __tracing_open(inode, file, &ret);
1684
1685         if (!ret)
1686                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1687
1688         return ret;
1689 }
1690
1691
1692 static void *
1693 t_next(struct seq_file *m, void *v, loff_t *pos)
1694 {
1695         struct tracer *t = m->private;
1696
1697         (*pos)++;
1698
1699         if (t)
1700                 t = t->next;
1701
1702         m->private = t;
1703
1704         return t;
1705 }
1706
1707 static void *t_start(struct seq_file *m, loff_t *pos)
1708 {
1709         struct tracer *t = m->private;
1710         loff_t l = 0;
1711
1712         mutex_lock(&trace_types_lock);
1713         for (; t && l < *pos; t = t_next(m, t, &l))
1714                 ;
1715
1716         return t;
1717 }
1718
1719 static void t_stop(struct seq_file *m, void *p)
1720 {
1721         mutex_unlock(&trace_types_lock);
1722 }
1723
1724 static int t_show(struct seq_file *m, void *v)
1725 {
1726         struct tracer *t = v;
1727
1728         if (!t)
1729                 return 0;
1730
1731         seq_printf(m, "%s", t->name);
1732         if (t->next)
1733                 seq_putc(m, ' ');
1734         else
1735                 seq_putc(m, '\n');
1736
1737         return 0;
1738 }
1739
1740 static struct seq_operations show_traces_seq_ops = {
1741         .start          = t_start,
1742         .next           = t_next,
1743         .stop           = t_stop,
1744         .show           = t_show,
1745 };
1746
1747 static int show_traces_open(struct inode *inode, struct file *file)
1748 {
1749         int ret;
1750
1751         if (tracing_disabled)
1752                 return -ENODEV;
1753
1754         ret = seq_open(file, &show_traces_seq_ops);
1755         if (!ret) {
1756                 struct seq_file *m = file->private_data;
1757                 m->private = trace_types;
1758         }
1759
1760         return ret;
1761 }
1762
1763 static struct file_operations tracing_fops = {
1764         .open           = tracing_open,
1765         .read           = seq_read,
1766         .llseek         = seq_lseek,
1767         .release        = tracing_release,
1768 };
1769
1770 static struct file_operations tracing_lt_fops = {
1771         .open           = tracing_lt_open,
1772         .read           = seq_read,
1773         .llseek         = seq_lseek,
1774         .release        = tracing_release,
1775 };
1776
1777 static struct file_operations show_traces_fops = {
1778         .open           = show_traces_open,
1779         .read           = seq_read,
1780         .release        = seq_release,
1781 };
1782
1783 /*
1784  * Only trace on a CPU if the bitmask is set:
1785  */
1786 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1787
1788 /*
1789  * When tracing/tracing_cpu_mask is modified then this holds
1790  * the new bitmask we are about to install:
1791  */
1792 static cpumask_t tracing_cpumask_new;
1793
1794 /*
1795  * The tracer itself will not take this lock, but still we want
1796  * to provide a consistent cpumask to user-space:
1797  */
1798 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1799
1800 /*
1801  * Temporary storage for the character representation of the
1802  * CPU bitmask (and one more byte for the newline):
1803  */
1804 static char mask_str[NR_CPUS + 1];
1805
1806 static ssize_t
1807 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1808                      size_t count, loff_t *ppos)
1809 {
1810         int len;
1811
1812         mutex_lock(&tracing_cpumask_update_lock);
1813
1814         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1815         if (count - len < 2) {
1816                 count = -EINVAL;
1817                 goto out_err;
1818         }
1819         len += sprintf(mask_str + len, "\n");
1820         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1821
1822 out_err:
1823         mutex_unlock(&tracing_cpumask_update_lock);
1824
1825         return count;
1826 }
1827
1828 static ssize_t
1829 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1830                       size_t count, loff_t *ppos)
1831 {
1832         int err, cpu;
1833
1834         mutex_lock(&tracing_cpumask_update_lock);
1835         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1836         if (err)
1837                 goto err_unlock;
1838
1839         raw_local_irq_disable();
1840         __raw_spin_lock(&ftrace_max_lock);
1841         for_each_possible_cpu(cpu) {
1842                 /*
1843                  * Increase/decrease the disabled counter if we are
1844                  * about to flip a bit in the cpumask:
1845                  */
1846                 if (cpu_isset(cpu, tracing_cpumask) &&
1847                                 !cpu_isset(cpu, tracing_cpumask_new)) {
1848                         atomic_inc(&global_trace.data[cpu]->disabled);
1849                 }
1850                 if (!cpu_isset(cpu, tracing_cpumask) &&
1851                                 cpu_isset(cpu, tracing_cpumask_new)) {
1852                         atomic_dec(&global_trace.data[cpu]->disabled);
1853                 }
1854         }
1855         __raw_spin_unlock(&ftrace_max_lock);
1856         raw_local_irq_enable();
1857
1858         tracing_cpumask = tracing_cpumask_new;
1859
1860         mutex_unlock(&tracing_cpumask_update_lock);
1861
1862         return count;
1863
1864 err_unlock:
1865         mutex_unlock(&tracing_cpumask_update_lock);
1866
1867         return err;
1868 }
1869
1870 static struct file_operations tracing_cpumask_fops = {
1871         .open           = tracing_open_generic,
1872         .read           = tracing_cpumask_read,
1873         .write          = tracing_cpumask_write,
1874 };
1875
1876 static ssize_t
1877 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1878                        size_t cnt, loff_t *ppos)
1879 {
1880         char *buf;
1881         int r = 0;
1882         int len = 0;
1883         int i;
1884
1885         /* calulate max size */
1886         for (i = 0; trace_options[i]; i++) {
1887                 len += strlen(trace_options[i]);
1888                 len += 3; /* "no" and space */
1889         }
1890
1891         /* +2 for \n and \0 */
1892         buf = kmalloc(len + 2, GFP_KERNEL);
1893         if (!buf)
1894                 return -ENOMEM;
1895
1896         for (i = 0; trace_options[i]; i++) {
1897                 if (trace_flags & (1 << i))
1898                         r += sprintf(buf + r, "%s ", trace_options[i]);
1899                 else
1900                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1901         }
1902
1903         r += sprintf(buf + r, "\n");
1904         WARN_ON(r >= len + 2);
1905
1906         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1907
1908         kfree(buf);
1909
1910         return r;
1911 }
1912
1913 static ssize_t
1914 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1915                         size_t cnt, loff_t *ppos)
1916 {
1917         char buf[64];
1918         char *cmp = buf;
1919         int neg = 0;
1920         int i;
1921
1922         if (cnt > 63)
1923                 cnt = 63;
1924
1925         if (copy_from_user(&buf, ubuf, cnt))
1926                 return -EFAULT;
1927
1928         buf[cnt] = 0;
1929
1930         if (strncmp(buf, "no", 2) == 0) {
1931                 neg = 1;
1932                 cmp += 2;
1933         }
1934
1935         for (i = 0; trace_options[i]; i++) {
1936                 int len = strlen(trace_options[i]);
1937
1938                 if (strncmp(cmp, trace_options[i], len) == 0) {
1939                         if (neg)
1940                                 trace_flags &= ~(1 << i);
1941                         else
1942                                 trace_flags |= (1 << i);
1943                         break;
1944                 }
1945         }
1946         /*
1947          * If no option could be set, return an error:
1948          */
1949         if (!trace_options[i])
1950                 return -EINVAL;
1951
1952         filp->f_pos += cnt;
1953
1954         return cnt;
1955 }
1956
1957 static struct file_operations tracing_iter_fops = {
1958         .open           = tracing_open_generic,
1959         .read           = tracing_iter_ctrl_read,
1960         .write          = tracing_iter_ctrl_write,
1961 };
1962
1963 static const char readme_msg[] =
1964         "tracing mini-HOWTO:\n\n"
1965         "# mkdir /debug\n"
1966         "# mount -t debugfs nodev /debug\n\n"
1967         "# cat /debug/tracing/available_tracers\n"
1968         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1969         "# cat /debug/tracing/current_tracer\n"
1970         "none\n"
1971         "# echo sched_switch > /debug/tracing/current_tracer\n"
1972         "# cat /debug/tracing/current_tracer\n"
1973         "sched_switch\n"
1974         "# cat /debug/tracing/iter_ctrl\n"
1975         "noprint-parent nosym-offset nosym-addr noverbose\n"
1976         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1977         "# echo 1 > /debug/tracing/tracing_enabled\n"
1978         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1979         "echo 0 > /debug/tracing/tracing_enabled\n"
1980 ;
1981
1982 static ssize_t
1983 tracing_readme_read(struct file *filp, char __user *ubuf,
1984                        size_t cnt, loff_t *ppos)
1985 {
1986         return simple_read_from_buffer(ubuf, cnt, ppos,
1987                                         readme_msg, strlen(readme_msg));
1988 }
1989
1990 static struct file_operations tracing_readme_fops = {
1991         .open           = tracing_open_generic,
1992         .read           = tracing_readme_read,
1993 };
1994
1995 static ssize_t
1996 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1997                   size_t cnt, loff_t *ppos)
1998 {
1999         struct trace_array *tr = filp->private_data;
2000         char buf[64];
2001         int r;
2002
2003         r = sprintf(buf, "%ld\n", tr->ctrl);
2004         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2005 }
2006
2007 static ssize_t
2008 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2009                    size_t cnt, loff_t *ppos)
2010 {
2011         struct trace_array *tr = 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         val = !!val;
2026
2027         mutex_lock(&trace_types_lock);
2028         if (tr->ctrl ^ val) {
2029                 if (val)
2030                         tracer_enabled = 1;
2031                 else
2032                         tracer_enabled = 0;
2033
2034                 tr->ctrl = val;
2035
2036                 if (current_trace && current_trace->ctrl_update)
2037                         current_trace->ctrl_update(tr);
2038         }
2039         mutex_unlock(&trace_types_lock);
2040
2041         filp->f_pos += cnt;
2042
2043         return cnt;
2044 }
2045
2046 static ssize_t
2047 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2048                        size_t cnt, loff_t *ppos)
2049 {
2050         char buf[max_tracer_type_len+2];
2051         int r;
2052
2053         mutex_lock(&trace_types_lock);
2054         if (current_trace)
2055                 r = sprintf(buf, "%s\n", current_trace->name);
2056         else
2057                 r = sprintf(buf, "\n");
2058         mutex_unlock(&trace_types_lock);
2059
2060         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2061 }
2062
2063 static ssize_t
2064 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2065                         size_t cnt, loff_t *ppos)
2066 {
2067         struct trace_array *tr = &global_trace;
2068         struct tracer *t;
2069         char buf[max_tracer_type_len+1];
2070         int i;
2071
2072         if (cnt > max_tracer_type_len)
2073                 cnt = max_tracer_type_len;
2074
2075         if (copy_from_user(&buf, ubuf, cnt))
2076                 return -EFAULT;
2077
2078         buf[cnt] = 0;
2079
2080         /* strip ending whitespace. */
2081         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2082                 buf[i] = 0;
2083
2084         mutex_lock(&trace_types_lock);
2085         for (t = trace_types; t; t = t->next) {
2086                 if (strcmp(t->name, buf) == 0)
2087                         break;
2088         }
2089         if (!t || t == current_trace)
2090                 goto out;
2091
2092         if (current_trace && current_trace->reset)
2093                 current_trace->reset(tr);
2094
2095         current_trace = t;
2096         if (t->init)
2097                 t->init(tr);
2098
2099  out:
2100         mutex_unlock(&trace_types_lock);
2101
2102         filp->f_pos += cnt;
2103
2104         return cnt;
2105 }
2106
2107 static ssize_t
2108 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2109                      size_t cnt, loff_t *ppos)
2110 {
2111         unsigned long *ptr = filp->private_data;
2112         char buf[64];
2113         int r;
2114
2115         r = snprintf(buf, 64, "%ld\n",
2116                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2117         if (r > 64)
2118                 r = 64;
2119         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2120 }
2121
2122 static ssize_t
2123 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2124                       size_t cnt, loff_t *ppos)
2125 {
2126         long *ptr = filp->private_data;
2127         long val;
2128         char buf[64];
2129
2130         if (cnt > 63)
2131                 cnt = 63;
2132
2133         if (copy_from_user(&buf, ubuf, cnt))
2134                 return -EFAULT;
2135
2136         buf[cnt] = 0;
2137
2138         val = simple_strtoul(buf, NULL, 10);
2139
2140         *ptr = val * 1000;
2141
2142         return cnt;
2143 }
2144
2145 static atomic_t tracing_reader;
2146
2147 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2148 {
2149         struct trace_iterator *iter;
2150
2151         if (tracing_disabled)
2152                 return -ENODEV;
2153
2154         /* We only allow for reader of the pipe */
2155         if (atomic_inc_return(&tracing_reader) != 1) {
2156                 atomic_dec(&tracing_reader);
2157                 return -EBUSY;
2158         }
2159
2160         /* create a buffer to store the information to pass to userspace */
2161         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2162         if (!iter)
2163                 return -ENOMEM;
2164
2165         iter->tr = &global_trace;
2166         iter->trace = current_trace;
2167
2168         filp->private_data = iter;
2169
2170         return 0;
2171 }
2172
2173 static int tracing_release_pipe(struct inode *inode, struct file *file)
2174 {
2175         struct trace_iterator *iter = file->private_data;
2176
2177         kfree(iter);
2178         atomic_dec(&tracing_reader);
2179
2180         return 0;
2181 }
2182
2183 static unsigned int
2184 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2185 {
2186         struct trace_iterator *iter = filp->private_data;
2187
2188         if (trace_flags & TRACE_ITER_BLOCK) {
2189                 /*
2190                  * Always select as readable when in blocking mode
2191                  */
2192                 return POLLIN | POLLRDNORM;
2193         }
2194         else {
2195                 if (!trace_empty(iter))
2196                         return POLLIN | POLLRDNORM;
2197                 poll_wait(filp, &trace_wait, poll_table);
2198                 if (!trace_empty(iter))
2199                         return POLLIN | POLLRDNORM;
2200
2201                 return 0;
2202         }
2203 }
2204
2205 /*
2206  * Consumer reader.
2207  */
2208 static ssize_t
2209 tracing_read_pipe(struct file *filp, char __user *ubuf,
2210                   size_t cnt, loff_t *ppos)
2211 {
2212         struct trace_iterator *iter = filp->private_data;
2213         struct trace_array_cpu *data;
2214         struct trace_array *tr = iter->tr;
2215         struct tracer *tracer = iter->trace;
2216         static cpumask_t mask;
2217         static int start;
2218         unsigned long flags;
2219 #ifdef CONFIG_FTRACE
2220         int ftrace_save;
2221 #endif
2222         int read = 0;
2223         int cpu;
2224         int len;
2225         int ret;
2226
2227         /* return any leftover data */
2228         if (iter->seq.len > start) {
2229                 len = iter->seq.len - start;
2230                 if (cnt > len)
2231                         cnt = len;
2232                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2233                 if (ret)
2234                         cnt = -EFAULT;
2235
2236                 start += len;
2237
2238                 return cnt;
2239         }
2240
2241         trace_seq_reset(&iter->seq);
2242         start = 0;
2243
2244         while (trace_empty(iter)) {
2245
2246                 if ((filp->f_flags & O_NONBLOCK))
2247                         return -EAGAIN;
2248
2249                 /*
2250                  * This is a make-shift waitqueue. The reason we don't use
2251                  * an actual wait queue is because:
2252                  *  1) we only ever have one waiter
2253                  *  2) the tracing, traces all functions, we don't want
2254                  *     the overhead of calling wake_up and friends
2255                  *     (and tracing them too)
2256                  *     Anyway, this is really very primitive wakeup.
2257                  */
2258                 set_current_state(TASK_INTERRUPTIBLE);
2259                 iter->tr->waiter = current;
2260
2261                 /* sleep for one second, and try again. */
2262                 schedule_timeout(HZ);
2263
2264                 iter->tr->waiter = NULL;
2265
2266                 if (signal_pending(current))
2267                         return -EINTR;
2268
2269                 if (iter->trace != current_trace)
2270                         return 0;
2271
2272                 /*
2273                  * We block until we read something and tracing is disabled.
2274                  * We still block if tracing is disabled, but we have never
2275                  * read anything. This allows a user to cat this file, and
2276                  * then enable tracing. But after we have read something,
2277                  * we give an EOF when tracing is again disabled.
2278                  *
2279                  * iter->pos will be 0 if we haven't read anything.
2280                  */
2281                 if (!tracer_enabled && iter->pos)
2282                         break;
2283
2284                 continue;
2285         }
2286
2287         /* stop when tracing is finished */
2288         if (trace_empty(iter))
2289                 return 0;
2290
2291         if (cnt >= PAGE_SIZE)
2292                 cnt = PAGE_SIZE - 1;
2293
2294         memset(iter, 0, sizeof(*iter));
2295         iter->tr = tr;
2296         iter->trace = tracer;
2297         iter->pos = -1;
2298
2299         /*
2300          * We need to stop all tracing on all CPUS to read the
2301          * the next buffer. This is a bit expensive, but is
2302          * not done often. We fill all what we can read,
2303          * and then release the locks again.
2304          */
2305
2306         cpus_clear(mask);
2307         local_irq_save(flags);
2308 #ifdef CONFIG_FTRACE
2309         ftrace_save = ftrace_enabled;
2310         ftrace_enabled = 0;
2311 #endif
2312         smp_wmb();
2313         for_each_possible_cpu(cpu) {
2314                 data = iter->tr->data[cpu];
2315
2316                 if (!head_page(data) || !data->trace_idx)
2317                         continue;
2318
2319                 atomic_inc(&data->disabled);
2320                 cpu_set(cpu, mask);
2321         }
2322
2323         for_each_cpu_mask(cpu, mask) {
2324                 data = iter->tr->data[cpu];
2325                 __raw_spin_lock(&data->lock);
2326         }
2327
2328         while (find_next_entry_inc(iter) != NULL) {
2329                 int len = iter->seq.len;
2330
2331                 ret = print_trace_line(iter);
2332                 if (!ret) {
2333                         /* don't print partial lines */
2334                         iter->seq.len = len;
2335                         break;
2336                 }
2337
2338                 trace_consume(iter);
2339
2340                 if (iter->seq.len >= cnt)
2341                         break;
2342         }
2343
2344         for_each_cpu_mask(cpu, mask) {
2345                 data = iter->tr->data[cpu];
2346                 __raw_spin_unlock(&data->lock);
2347         }
2348
2349         for_each_cpu_mask(cpu, mask) {
2350                 data = iter->tr->data[cpu];
2351                 atomic_dec(&data->disabled);
2352         }
2353 #ifdef CONFIG_FTRACE
2354         ftrace_enabled = ftrace_save;
2355 #endif
2356         local_irq_restore(flags);
2357
2358         /* Now copy what we have to the user */
2359         read = iter->seq.len;
2360         if (read > cnt)
2361                 read = cnt;
2362
2363         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2364
2365         if (read < iter->seq.len)
2366                 start = read;
2367         else
2368                 trace_seq_reset(&iter->seq);
2369
2370         if (ret)
2371                 read = -EFAULT;
2372
2373         return read;
2374 }
2375
2376 static ssize_t
2377 tracing_entries_read(struct file *filp, char __user *ubuf,
2378                      size_t cnt, loff_t *ppos)
2379 {
2380         struct trace_array *tr = filp->private_data;
2381         char buf[64];
2382         int r;
2383
2384         r = sprintf(buf, "%lu\n", tr->entries);
2385         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2386 }
2387
2388 static ssize_t
2389 tracing_entries_write(struct file *filp, const char __user *ubuf,
2390                       size_t cnt, loff_t *ppos)
2391 {
2392         unsigned long val;
2393         char buf[64];
2394
2395         if (cnt > 63)
2396                 cnt = 63;
2397
2398         if (copy_from_user(&buf, ubuf, cnt))
2399                 return -EFAULT;
2400
2401         buf[cnt] = 0;
2402
2403         val = simple_strtoul(buf, NULL, 10);
2404
2405         /* must have at least 1 entry */
2406         if (!val)
2407                 return -EINVAL;
2408
2409         mutex_lock(&trace_types_lock);
2410
2411         if (current_trace != &no_tracer) {
2412                 cnt = -EBUSY;
2413                 pr_info("ftrace: set current_tracer to none"
2414                         " before modifying buffer size\n");
2415                 goto out;
2416         }
2417
2418         if (val > global_trace.entries) {
2419                 while (global_trace.entries < val) {
2420                         if (trace_alloc_page()) {
2421                                 cnt = -ENOMEM;
2422                                 goto out;
2423                         }
2424                 }
2425         } else {
2426                 /* include the number of entries in val (inc of page entries) */
2427                 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2428                         trace_free_page();
2429         }
2430
2431         filp->f_pos += cnt;
2432
2433  out:
2434         max_tr.entries = global_trace.entries;
2435         mutex_unlock(&trace_types_lock);
2436
2437         return cnt;
2438 }
2439
2440 static struct file_operations tracing_max_lat_fops = {
2441         .open           = tracing_open_generic,
2442         .read           = tracing_max_lat_read,
2443         .write          = tracing_max_lat_write,
2444 };
2445
2446 static struct file_operations tracing_ctrl_fops = {
2447         .open           = tracing_open_generic,
2448         .read           = tracing_ctrl_read,
2449         .write          = tracing_ctrl_write,
2450 };
2451
2452 static struct file_operations set_tracer_fops = {
2453         .open           = tracing_open_generic,
2454         .read           = tracing_set_trace_read,
2455         .write          = tracing_set_trace_write,
2456 };
2457
2458 static struct file_operations tracing_pipe_fops = {
2459         .open           = tracing_open_pipe,
2460         .poll           = tracing_poll_pipe,
2461         .read           = tracing_read_pipe,
2462         .release        = tracing_release_pipe,
2463 };
2464
2465 static struct file_operations tracing_entries_fops = {
2466         .open           = tracing_open_generic,
2467         .read           = tracing_entries_read,
2468         .write          = tracing_entries_write,
2469 };
2470
2471 #ifdef CONFIG_DYNAMIC_FTRACE
2472
2473 static ssize_t
2474 tracing_read_long(struct file *filp, char __user *ubuf,
2475                   size_t cnt, loff_t *ppos)
2476 {
2477         unsigned long *p = filp->private_data;
2478         char buf[64];
2479         int r;
2480
2481         r = sprintf(buf, "%ld\n", *p);
2482
2483         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2484 }
2485
2486 static struct file_operations tracing_read_long_fops = {
2487         .open           = tracing_open_generic,
2488         .read           = tracing_read_long,
2489 };
2490 #endif
2491
2492 static struct dentry *d_tracer;
2493
2494 struct dentry *tracing_init_dentry(void)
2495 {
2496         static int once;
2497
2498         if (d_tracer)
2499                 return d_tracer;
2500
2501         d_tracer = debugfs_create_dir("tracing", NULL);
2502
2503         if (!d_tracer && !once) {
2504                 once = 1;
2505                 pr_warning("Could not create debugfs directory 'tracing'\n");
2506                 return NULL;
2507         }
2508
2509         return d_tracer;
2510 }
2511
2512 #ifdef CONFIG_FTRACE_SELFTEST
2513 /* Let selftest have access to static functions in this file */
2514 #include "trace_selftest.c"
2515 #endif
2516
2517 static __init void tracer_init_debugfs(void)
2518 {
2519         struct dentry *d_tracer;
2520         struct dentry *entry;
2521
2522         d_tracer = tracing_init_dentry();
2523
2524         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2525                                     &global_trace, &tracing_ctrl_fops);
2526         if (!entry)
2527                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2528
2529         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2530                                     NULL, &tracing_iter_fops);
2531         if (!entry)
2532                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2533
2534         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2535                                     NULL, &tracing_cpumask_fops);
2536         if (!entry)
2537                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2538
2539         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2540                                     &global_trace, &tracing_lt_fops);
2541         if (!entry)
2542                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2543
2544         entry = debugfs_create_file("trace", 0444, d_tracer,
2545                                     &global_trace, &tracing_fops);
2546         if (!entry)
2547                 pr_warning("Could not create debugfs 'trace' entry\n");
2548
2549         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2550                                     &global_trace, &show_traces_fops);
2551         if (!entry)
2552                 pr_warning("Could not create debugfs 'trace' entry\n");
2553
2554         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2555                                     &global_trace, &set_tracer_fops);
2556         if (!entry)
2557                 pr_warning("Could not create debugfs 'trace' entry\n");
2558
2559         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2560                                     &tracing_max_latency,
2561                                     &tracing_max_lat_fops);
2562         if (!entry)
2563                 pr_warning("Could not create debugfs "
2564                            "'tracing_max_latency' entry\n");
2565
2566         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2567                                     &tracing_thresh, &tracing_max_lat_fops);
2568         if (!entry)
2569                 pr_warning("Could not create debugfs "
2570                            "'tracing_threash' entry\n");
2571         entry = debugfs_create_file("README", 0644, d_tracer,
2572                                     NULL, &tracing_readme_fops);
2573         if (!entry)
2574                 pr_warning("Could not create debugfs 'README' entry\n");
2575
2576         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2577                                     NULL, &tracing_pipe_fops);
2578         if (!entry)
2579                 pr_warning("Could not create debugfs "
2580                            "'tracing_threash' entry\n");
2581
2582         entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2583                                     &global_trace, &tracing_entries_fops);
2584         if (!entry)
2585                 pr_warning("Could not create debugfs "
2586                            "'tracing_threash' entry\n");
2587
2588 #ifdef CONFIG_DYNAMIC_FTRACE
2589         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2590                                     &ftrace_update_tot_cnt,
2591                                     &tracing_read_long_fops);
2592         if (!entry)
2593                 pr_warning("Could not create debugfs "
2594                            "'dyn_ftrace_total_info' entry\n");
2595 #endif
2596 }
2597
2598 static int trace_alloc_page(void)
2599 {
2600         struct trace_array_cpu *data;
2601         struct page *page, *tmp;
2602         LIST_HEAD(pages);
2603         void *array;
2604         int i;
2605
2606         /* first allocate a page for each CPU */
2607         for_each_possible_cpu(i) {
2608                 array = (void *)__get_free_page(GFP_KERNEL);
2609                 if (array == NULL) {
2610                         printk(KERN_ERR "tracer: failed to allocate page"
2611                                "for trace buffer!\n");
2612                         goto free_pages;
2613                 }
2614
2615                 page = virt_to_page(array);
2616                 list_add(&page->lru, &pages);
2617
2618 /* Only allocate if we are actually using the max trace */
2619 #ifdef CONFIG_TRACER_MAX_TRACE
2620                 array = (void *)__get_free_page(GFP_KERNEL);
2621                 if (array == NULL) {
2622                         printk(KERN_ERR "tracer: failed to allocate page"
2623                                "for trace buffer!\n");
2624                         goto free_pages;
2625                 }
2626                 page = virt_to_page(array);
2627                 list_add(&page->lru, &pages);
2628 #endif
2629         }
2630
2631         /* Now that we successfully allocate a page per CPU, add them */
2632         for_each_possible_cpu(i) {
2633                 data = global_trace.data[i];
2634                 page = list_entry(pages.next, struct page, lru);
2635                 list_del_init(&page->lru);
2636                 list_add_tail(&page->lru, &data->trace_pages);
2637                 ClearPageLRU(page);
2638
2639 #ifdef CONFIG_TRACER_MAX_TRACE
2640                 data = max_tr.data[i];
2641                 page = list_entry(pages.next, struct page, lru);
2642                 list_del_init(&page->lru);
2643                 list_add_tail(&page->lru, &data->trace_pages);
2644                 SetPageLRU(page);
2645 #endif
2646         }
2647         global_trace.entries += ENTRIES_PER_PAGE;
2648
2649         return 0;
2650
2651  free_pages:
2652         list_for_each_entry_safe(page, tmp, &pages, lru) {
2653                 list_del_init(&page->lru);
2654                 __free_page(page);
2655         }
2656         return -ENOMEM;
2657 }
2658
2659 static int trace_free_page(void)
2660 {
2661         struct trace_array_cpu *data;
2662         struct page *page;
2663         struct list_head *p;
2664         int i;
2665         int ret = 0;
2666
2667         /* free one page from each buffer */
2668         for_each_possible_cpu(i) {
2669                 data = global_trace.data[i];
2670                 p = data->trace_pages.next;
2671                 if (p == &data->trace_pages) {
2672                         /* should never happen */
2673                         WARN_ON(1);
2674                         tracing_disabled = 1;
2675                         ret = -1;
2676                         break;
2677                 }
2678                 page = list_entry(p, struct page, lru);
2679                 ClearPageLRU(page);
2680                 list_del(&page->lru);
2681                 __free_page(page);
2682
2683                 tracing_reset(data);
2684
2685 #ifdef CONFIG_TRACER_MAX_TRACE
2686                 data = max_tr.data[i];
2687                 p = data->trace_pages.next;
2688                 if (p == &data->trace_pages) {
2689                         /* should never happen */
2690                         WARN_ON(1);
2691                         tracing_disabled = 1;
2692                         ret = -1;
2693                         break;
2694                 }
2695                 page = list_entry(p, struct page, lru);
2696                 ClearPageLRU(page);
2697                 list_del(&page->lru);
2698                 __free_page(page);
2699
2700                 tracing_reset(data);
2701 #endif
2702         }
2703         global_trace.entries -= ENTRIES_PER_PAGE;
2704
2705         return ret;
2706 }
2707
2708 __init static int tracer_alloc_buffers(void)
2709 {
2710         struct trace_array_cpu *data;
2711         void *array;
2712         struct page *page;
2713         int pages = 0;
2714         int ret = -ENOMEM;
2715         int i;
2716
2717         global_trace.ctrl = tracer_enabled;
2718
2719         /* Allocate the first page for all buffers */
2720         for_each_possible_cpu(i) {
2721                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2722                 max_tr.data[i] = &per_cpu(max_data, i);
2723
2724                 array = (void *)__get_free_page(GFP_KERNEL);
2725                 if (array == NULL) {
2726                         printk(KERN_ERR "tracer: failed to allocate page"
2727                                "for trace buffer!\n");
2728                         goto free_buffers;
2729                 }
2730
2731                 /* set the array to the list */
2732                 INIT_LIST_HEAD(&data->trace_pages);
2733                 page = virt_to_page(array);
2734                 list_add(&page->lru, &data->trace_pages);
2735                 /* use the LRU flag to differentiate the two buffers */
2736                 ClearPageLRU(page);
2737
2738                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2739                 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2740
2741 /* Only allocate if we are actually using the max trace */
2742 #ifdef CONFIG_TRACER_MAX_TRACE
2743                 array = (void *)__get_free_page(GFP_KERNEL);
2744                 if (array == NULL) {
2745                         printk(KERN_ERR "tracer: failed to allocate page"
2746                                "for trace buffer!\n");
2747                         goto free_buffers;
2748                 }
2749
2750                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2751                 page = virt_to_page(array);
2752                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2753                 SetPageLRU(page);
2754 #endif
2755         }
2756
2757         /*
2758          * Since we allocate by orders of pages, we may be able to
2759          * round up a bit.
2760          */
2761         global_trace.entries = ENTRIES_PER_PAGE;
2762         pages++;
2763
2764         while (global_trace.entries < trace_nr_entries) {
2765                 if (trace_alloc_page())
2766                         break;
2767                 pages++;
2768         }
2769         max_tr.entries = global_trace.entries;
2770
2771         pr_info("tracer: %d pages allocated for %ld",
2772                 pages, trace_nr_entries);
2773         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2774         pr_info("   actual entries %ld\n", global_trace.entries);
2775
2776         tracer_init_debugfs();
2777
2778         trace_init_cmdlines();
2779
2780         register_tracer(&no_tracer);
2781         current_trace = &no_tracer;
2782
2783         /* All seems OK, enable tracing */
2784         tracing_disabled = 0;
2785
2786         return 0;
2787
2788  free_buffers:
2789         for (i-- ; i >= 0; i--) {
2790                 struct page *page, *tmp;
2791                 struct trace_array_cpu *data = global_trace.data[i];
2792
2793                 if (data) {
2794                         list_for_each_entry_safe(page, tmp,
2795                                                  &data->trace_pages, lru) {
2796                                 list_del_init(&page->lru);
2797                                 __free_page(page);
2798                         }
2799                 }
2800
2801 #ifdef CONFIG_TRACER_MAX_TRACE
2802                 data = max_tr.data[i];
2803                 if (data) {
2804                         list_for_each_entry_safe(page, tmp,
2805                                                  &data->trace_pages, lru) {
2806                                 list_del_init(&page->lru);
2807                                 __free_page(page);
2808                         }
2809                 }
2810 #endif
2811         }
2812         return ret;
2813 }
2814 fs_initcall(tracer_alloc_buffers);