git.oblomov.eu Git - linux-2.6/blob - kernel/trace/trace.c

   1 /*
   2  * ring buffer based function tracer
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally taken from the RT patch by:
   8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code from the latency_tracer, that is:
  11  *  Copyright (C) 2004-2006 Ingo Molnar
  12  *  Copyright (C) 2004 William Lee Irwin III
  13  */
  14 #include <linux/utsrelease.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/debugfs.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/linkage.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/ftrace.h>
  23 #include <linux/module.h>
  24 #include <linux/percpu.h>
  25 #include <linux/ctype.h>
  26 #include <linux/init.h>
  27 #include <linux/poll.h>
  28 #include <linux/gfp.h>
  29 #include <linux/fs.h>
  30
  31 #include <linux/stacktrace.h>
  32
  33 #include "trace.h"
  34
  35 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  36 unsigned long __read_mostly     tracing_thresh;
  37
  38 static int tracing_disabled = 1;
  39
  40 static long
  41 ns2usecs(cycle_t nsec)
  42 {
  43         nsec += 500;
  44         do_div(nsec, 1000);
  45         return nsec;
  46 }
  47
  48 cycle_t ftrace_now(int cpu)
  49 {
  50         return cpu_clock(cpu);
  51 }
  52
  53 static struct trace_array       global_trace;
  54
  55 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  56
  57 static struct trace_array       max_tr;
  58
  59 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  60
  61 static int                      tracer_enabled = 1;
  62 static unsigned long            trace_nr_entries = 65536UL;
  63
  64 static struct tracer            *trace_types __read_mostly;
  65 static struct tracer            *current_trace __read_mostly;
  66 static int                      max_tracer_type_len;
  67
  68 static DEFINE_MUTEX(trace_types_lock);
  69 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
  70
  71 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
  72
  73 void trace_wake_up(void)
  74 {
  75         /*
  76          * The runqueue_is_locked() can fail, but this is the best we
  77          * have for now:
  78          */
  79         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
  80                 wake_up(&trace_wait);
  81 }
  82
  83 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  84
  85 static int __init set_nr_entries(char *str)
  86 {
  87         if (!str)
  88                 return 0;
  89         trace_nr_entries = simple_strtoul(str, &str, 0);
  90         return 1;
  91 }
  92 __setup("trace_entries=", set_nr_entries);
  93
  94 unsigned long nsecs_to_usecs(unsigned long nsecs)
  95 {
  96         return nsecs / 1000;
  97 }
  98
  99 enum trace_type {
 100         __TRACE_FIRST_TYPE = 0,
 101
 102         TRACE_FN,
 103         TRACE_CTX,
 104         TRACE_WAKE,
 105         TRACE_STACK,
 106         TRACE_SPECIAL,
 107
 108         __TRACE_LAST_TYPE
 109 };
 110
 111 enum trace_flag_type {
 112         TRACE_FLAG_IRQS_OFF             = 0x01,
 113         TRACE_FLAG_NEED_RESCHED         = 0x02,
 114         TRACE_FLAG_HARDIRQ              = 0x04,
 115         TRACE_FLAG_SOFTIRQ              = 0x08,
 116 };
 117
 118 #define TRACE_ITER_SYM_MASK \
 119         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 120
 121 /* These must match the bit postions above */
 122 static const char *trace_options[] = {
 123         "print-parent",
 124         "sym-offset",
 125         "sym-addr",
 126         "verbose",
 127         "raw",
 128         "hex",
 129         "bin",
 130         "block",
 131         "stacktrace",
 132         "sched-tree",
 133         NULL
 134 };
 135
 136 static raw_spinlock_t ftrace_max_lock =
 137         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 138
 139 /*
 140  * Copy the new maximum trace into the separate maximum-trace
 141  * structure. (this way the maximum trace is permanently saved,
 142  * for later retrieval via /debugfs/tracing/latency_trace)
 143  */
 144 static void
 145 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 146 {
 147         struct trace_array_cpu *data = tr->data[cpu];
 148
 149         max_tr.cpu = cpu;
 150         max_tr.time_start = data->preempt_timestamp;
 151
 152         data = max_tr.data[cpu];
 153         data->saved_latency = tracing_max_latency;
 154
 155         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 156         data->pid = tsk->pid;
 157         data->uid = tsk->uid;
 158         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 159         data->policy = tsk->policy;
 160         data->rt_priority = tsk->rt_priority;
 161
 162         /* record this tasks comm */
 163         tracing_record_cmdline(current);
 164 }
 165
 166 void check_pages(struct trace_array_cpu *data)
 167 {
 168         struct page *page, *tmp;
 169
 170         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 171         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 172
 173         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 174                 BUG_ON(page->lru.next->prev != &page->lru);
 175                 BUG_ON(page->lru.prev->next != &page->lru);
 176         }
 177 }
 178
 179 void *head_page(struct trace_array_cpu *data)
 180 {
 181         struct page *page;
 182
 183         check_pages(data);
 184         if (list_empty(&data->trace_pages))
 185                 return NULL;
 186
 187         page = list_entry(data->trace_pages.next, struct page, lru);
 188         BUG_ON(&page->lru == &data->trace_pages);
 189
 190         return page_address(page);
 191 }
 192
 193 static int
 194 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 195 {
 196         int len = (PAGE_SIZE - 1) - s->len;
 197         va_list ap;
 198         int ret;
 199
 200         if (!len)
 201                 return 0;
 202
 203         va_start(ap, fmt);
 204         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 205         va_end(ap);
 206
 207         /* If we can't write it all, don't bother writing anything */
 208         if (ret > len)
 209                 return 0;
 210
 211         s->len += ret;
 212
 213         return len;
 214 }
 215
 216 static int
 217 trace_seq_puts(struct trace_seq *s, const char *str)
 218 {
 219         int len = strlen(str);
 220
 221         if (len > ((PAGE_SIZE - 1) - s->len))
 222                 return 0;
 223
 224         memcpy(s->buffer + s->len, str, len);
 225         s->len += len;
 226
 227         return len;
 228 }
 229
 230 static int
 231 trace_seq_putc(struct trace_seq *s, unsigned char c)
 232 {
 233         if (s->len >= (PAGE_SIZE - 1))
 234                 return 0;
 235
 236         s->buffer[s->len++] = c;
 237
 238         return 1;
 239 }
 240
 241 static int
 242 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 243 {
 244         if (len > ((PAGE_SIZE - 1) - s->len))
 245                 return 0;
 246
 247         memcpy(s->buffer + s->len, mem, len);
 248         s->len += len;
 249
 250         return len;
 251 }
 252
 253 #define HEX_CHARS 17
 254
 255 static int
 256 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 257 {
 258         unsigned char hex[HEX_CHARS];
 259         unsigned char *data;
 260         unsigned char byte;
 261         int i, j;
 262
 263         BUG_ON(len >= HEX_CHARS);
 264
 265         data = mem;
 266
 267 #ifdef __BIG_ENDIAN
 268         for (i = 0, j = 0; i < len; i++) {
 269 #else
 270         for (i = len-1, j = 0; i >= 0; i--) {
 271 #endif
 272                 byte = data[i];
 273
 274                 hex[j]   = byte & 0x0f;
 275                 if (hex[j] >= 10)
 276                         hex[j] += 'a' - 10;
 277                 else
 278                         hex[j] += '0';
 279                 j++;
 280
 281                 hex[j] = byte >> 4;
 282                 if (hex[j] >= 10)
 283                         hex[j] += 'a' - 10;
 284                 else
 285                         hex[j] += '0';
 286                 j++;
 287         }
 288         hex[j] = ' ';
 289         j++;
 290
 291         return trace_seq_putmem(s, hex, j);
 292 }
 293
 294 static void
 295 trace_seq_reset(struct trace_seq *s)
 296 {
 297         s->len = 0;
 298 }
 299
 300 static void
 301 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 302 {
 303         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 304
 305         s->buffer[len] = 0;
 306         seq_puts(m, s->buffer);
 307
 308         trace_seq_reset(s);
 309 }
 310
 311 static void
 312 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 313 {
 314         struct list_head flip_pages;
 315
 316         INIT_LIST_HEAD(&flip_pages);
 317
 318         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 319                 sizeof(struct trace_array_cpu) -
 320                 offsetof(struct trace_array_cpu, trace_head_idx));
 321
 322         check_pages(tr1);
 323         check_pages(tr2);
 324         list_splice_init(&tr1->trace_pages, &flip_pages);
 325         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 326         list_splice_init(&flip_pages, &tr2->trace_pages);
 327         BUG_ON(!list_empty(&flip_pages));
 328         check_pages(tr1);
 329         check_pages(tr2);
 330 }
 331
 332 void
 333 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 334 {
 335         struct trace_array_cpu *data;
 336         int i;
 337
 338         WARN_ON_ONCE(!irqs_disabled());
 339         __raw_spin_lock(&ftrace_max_lock);
 340         /* clear out all the previous traces */
 341         for_each_possible_cpu(i) {
 342                 data = tr->data[i];
 343                 flip_trace(max_tr.data[i], data);
 344                 tracing_reset(data);
 345         }
 346
 347         __update_max_tr(tr, tsk, cpu);
 348         __raw_spin_unlock(&ftrace_max_lock);
 349 }
 350
 351 /**
 352  * update_max_tr_single - only copy one trace over, and reset the rest
 353  * @tr - tracer
 354  * @tsk - task with the latency
 355  * @cpu - the cpu of the buffer to copy.
 356  */
 357 void
 358 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 359 {
 360         struct trace_array_cpu *data = tr->data[cpu];
 361         int i;
 362
 363         WARN_ON_ONCE(!irqs_disabled());
 364         __raw_spin_lock(&ftrace_max_lock);
 365         for_each_possible_cpu(i)
 366                 tracing_reset(max_tr.data[i]);
 367
 368         flip_trace(max_tr.data[cpu], data);
 369         tracing_reset(data);
 370
 371         __update_max_tr(tr, tsk, cpu);
 372         __raw_spin_unlock(&ftrace_max_lock);
 373 }
 374
 375 int register_tracer(struct tracer *type)
 376 {
 377         struct tracer *t;
 378         int len;
 379         int ret = 0;
 380
 381         if (!type->name) {
 382                 pr_info("Tracer must have a name\n");
 383                 return -1;
 384         }
 385
 386         mutex_lock(&trace_types_lock);
 387         for (t = trace_types; t; t = t->next) {
 388                 if (strcmp(type->name, t->name) == 0) {
 389                         /* already found */
 390                         pr_info("Trace %s already registered\n",
 391                                 type->name);
 392                         ret = -1;
 393                         goto out;
 394                 }
 395         }
 396
 397 #ifdef CONFIG_FTRACE_STARTUP_TEST
 398         if (type->selftest) {
 399                 struct tracer *saved_tracer = current_trace;
 400                 struct trace_array_cpu *data;
 401                 struct trace_array *tr = &global_trace;
 402                 int saved_ctrl = tr->ctrl;
 403                 int i;
 404                 /*
 405                  * Run a selftest on this tracer.
 406                  * Here we reset the trace buffer, and set the current
 407                  * tracer to be this tracer. The tracer can then run some
 408                  * internal tracing to verify that everything is in order.
 409                  * If we fail, we do not register this tracer.
 410                  */
 411                 for_each_possible_cpu(i) {
 412                         data = tr->data[i];
 413                         if (!head_page(data))
 414                                 continue;
 415                         tracing_reset(data);
 416                 }
 417                 current_trace = type;
 418                 tr->ctrl = 0;
 419                 /* the test is responsible for initializing and enabling */
 420                 pr_info("Testing tracer %s: ", type->name);
 421                 ret = type->selftest(type, tr);
 422                 /* the test is responsible for resetting too */
 423                 current_trace = saved_tracer;
 424                 tr->ctrl = saved_ctrl;
 425                 if (ret) {
 426                         printk(KERN_CONT "FAILED!\n");
 427                         goto out;
 428                 }
 429                 /* Only reset on passing, to avoid touching corrupted buffers */
 430                 for_each_possible_cpu(i) {
 431                         data = tr->data[i];
 432                         if (!head_page(data))
 433                                 continue;
 434                         tracing_reset(data);
 435                 }
 436                 printk(KERN_CONT "PASSED\n");
 437         }
 438 #endif
 439
 440         type->next = trace_types;
 441         trace_types = type;
 442         len = strlen(type->name);
 443         if (len > max_tracer_type_len)
 444                 max_tracer_type_len = len;
 445
 446  out:
 447         mutex_unlock(&trace_types_lock);
 448
 449         return ret;
 450 }
 451
 452 void unregister_tracer(struct tracer *type)
 453 {
 454         struct tracer **t;
 455         int len;
 456
 457         mutex_lock(&trace_types_lock);
 458         for (t = &trace_types; *t; t = &(*t)->next) {
 459                 if (*t == type)
 460                         goto found;
 461         }
 462         pr_info("Trace %s not registered\n", type->name);
 463         goto out;
 464
 465  found:
 466         *t = (*t)->next;
 467         if (strlen(type->name) != max_tracer_type_len)
 468                 goto out;
 469
 470         max_tracer_type_len = 0;
 471         for (t = &trace_types; *t; t = &(*t)->next) {
 472                 len = strlen((*t)->name);
 473                 if (len > max_tracer_type_len)
 474                         max_tracer_type_len = len;
 475         }
 476  out:
 477         mutex_unlock(&trace_types_lock);
 478 }
 479
 480 void tracing_reset(struct trace_array_cpu *data)
 481 {
 482         data->trace_idx = 0;
 483         data->trace_head = data->trace_tail = head_page(data);
 484         data->trace_head_idx = 0;
 485         data->trace_tail_idx = 0;
 486 }
 487
 488 #define SAVED_CMDLINES 128
 489 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 490 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 491 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 492 static int cmdline_idx;
 493 static DEFINE_SPINLOCK(trace_cmdline_lock);
 494 atomic_t trace_record_cmdline_disabled;
 495
 496 static void trace_init_cmdlines(void)
 497 {
 498         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 499         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 500         cmdline_idx = 0;
 501 }
 502
 503 void trace_stop_cmdline_recording(void);
 504
 505 static void trace_save_cmdline(struct task_struct *tsk)
 506 {
 507         unsigned map;
 508         unsigned idx;
 509
 510         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 511                 return;
 512
 513         /*
 514          * It's not the end of the world if we don't get
 515          * the lock, but we also don't want to spin
 516          * nor do we want to disable interrupts,
 517          * so if we miss here, then better luck next time.
 518          */
 519         if (!spin_trylock(&trace_cmdline_lock))
 520                 return;
 521
 522         idx = map_pid_to_cmdline[tsk->pid];
 523         if (idx >= SAVED_CMDLINES) {
 524                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 525
 526                 map = map_cmdline_to_pid[idx];
 527                 if (map <= PID_MAX_DEFAULT)
 528                         map_pid_to_cmdline[map] = (unsigned)-1;
 529
 530                 map_pid_to_cmdline[tsk->pid] = idx;
 531
 532                 cmdline_idx = idx;
 533         }
 534
 535         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 536
 537         spin_unlock(&trace_cmdline_lock);
 538 }
 539
 540 static char *trace_find_cmdline(int pid)
 541 {
 542         char *cmdline = "<...>";
 543         unsigned map;
 544
 545         if (!pid)
 546                 return "<idle>";
 547
 548         if (pid > PID_MAX_DEFAULT)
 549                 goto out;
 550
 551         map = map_pid_to_cmdline[pid];
 552         if (map >= SAVED_CMDLINES)
 553                 goto out;
 554
 555         cmdline = saved_cmdlines[map];
 556
 557  out:
 558         return cmdline;
 559 }
 560
 561 void tracing_record_cmdline(struct task_struct *tsk)
 562 {
 563         if (atomic_read(&trace_record_cmdline_disabled))
 564                 return;
 565
 566         trace_save_cmdline(tsk);
 567 }
 568
 569 static inline struct list_head *
 570 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 571 {
 572         /*
 573          * Roundrobin - but skip the head (which is not a real page):
 574          */
 575         next = next->next;
 576         if (unlikely(next == &data->trace_pages))
 577                 next = next->next;
 578         BUG_ON(next == &data->trace_pages);
 579
 580         return next;
 581 }
 582
 583 static inline void *
 584 trace_next_page(struct trace_array_cpu *data, void *addr)
 585 {
 586         struct list_head *next;
 587         struct page *page;
 588
 589         page = virt_to_page(addr);
 590
 591         next = trace_next_list(data, &page->lru);
 592         page = list_entry(next, struct page, lru);
 593
 594         return page_address(page);
 595 }
 596
 597 static inline struct trace_entry *
 598 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 599 {
 600         unsigned long idx, idx_next;
 601         struct trace_entry *entry;
 602
 603         data->trace_idx++;
 604         idx = data->trace_head_idx;
 605         idx_next = idx + 1;
 606
 607         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 608
 609         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 610
 611         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 612                 data->trace_head = trace_next_page(data, data->trace_head);
 613                 idx_next = 0;
 614         }
 615
 616         if (data->trace_head == data->trace_tail &&
 617             idx_next == data->trace_tail_idx) {
 618                 /* overrun */
 619                 data->trace_tail_idx++;
 620                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 621                         data->trace_tail =
 622                                 trace_next_page(data, data->trace_tail);
 623                         data->trace_tail_idx = 0;
 624                 }
 625         }
 626
 627         data->trace_head_idx = idx_next;
 628
 629         return entry;
 630 }
 631
 632 static inline void
 633 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 634 {
 635         struct task_struct *tsk = current;
 636         unsigned long pc;
 637
 638         pc = preempt_count();
 639
 640         entry->preempt_count    = pc & 0xff;
 641         entry->pid              = tsk->pid;
 642         entry->t                = ftrace_now(raw_smp_processor_id());
 643         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 644                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 645                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 646                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 647 }
 648
 649 void
 650 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 651                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 652 {
 653         struct trace_entry *entry;
 654         unsigned long irq_flags;
 655
 656         raw_local_irq_save(irq_flags);
 657         __raw_spin_lock(&data->lock);
 658         entry                   = tracing_get_trace_entry(tr, data);
 659         tracing_generic_entry_update(entry, flags);
 660         entry->type             = TRACE_FN;
 661         entry->fn.ip            = ip;
 662         entry->fn.parent_ip     = parent_ip;
 663         __raw_spin_unlock(&data->lock);
 664         raw_local_irq_restore(irq_flags);
 665 }
 666
 667 void
 668 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 669        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 670 {
 671         if (likely(!atomic_read(&data->disabled)))
 672                 trace_function(tr, data, ip, parent_ip, flags);
 673 }
 674
 675 void
 676 __trace_special(void *__tr, void *__data,
 677                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 678 {
 679         struct trace_array_cpu *data = __data;
 680         struct trace_array *tr = __tr;
 681         struct trace_entry *entry;
 682         unsigned long irq_flags;
 683
 684         raw_local_irq_save(irq_flags);
 685         __raw_spin_lock(&data->lock);
 686         entry                   = tracing_get_trace_entry(tr, data);
 687         tracing_generic_entry_update(entry, 0);
 688         entry->type             = TRACE_SPECIAL;
 689         entry->special.arg1     = arg1;
 690         entry->special.arg2     = arg2;
 691         entry->special.arg3     = arg3;
 692         __raw_spin_unlock(&data->lock);
 693         raw_local_irq_restore(irq_flags);
 694
 695         trace_wake_up();
 696 }
 697
 698 void __trace_stack(struct trace_array *tr,
 699                    struct trace_array_cpu *data,
 700                    unsigned long flags,
 701                    int skip)
 702 {
 703         struct trace_entry *entry;
 704         struct stack_trace trace;
 705
 706         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 707                 return;
 708
 709         entry                   = tracing_get_trace_entry(tr, data);
 710         tracing_generic_entry_update(entry, flags);
 711         entry->type             = TRACE_STACK;
 712
 713         memset(&entry->stack, 0, sizeof(entry->stack));
 714
 715         trace.nr_entries        = 0;
 716         trace.max_entries       = FTRACE_STACK_ENTRIES;
 717         trace.skip              = skip;
 718         trace.entries           = entry->stack.caller;
 719
 720         save_stack_trace(&trace);
 721 }
 722
 723 void
 724 tracing_sched_switch_trace(struct trace_array *tr,
 725                            struct trace_array_cpu *data,
 726                            struct task_struct *prev,
 727                            struct task_struct *next,
 728                            unsigned long flags)
 729 {
 730         struct trace_entry *entry;
 731         unsigned long irq_flags;
 732
 733         raw_local_irq_save(irq_flags);
 734         __raw_spin_lock(&data->lock);
 735         entry                   = tracing_get_trace_entry(tr, data);
 736         tracing_generic_entry_update(entry, flags);
 737         entry->type             = TRACE_CTX;
 738         entry->ctx.prev_pid     = prev->pid;
 739         entry->ctx.prev_prio    = prev->prio;
 740         entry->ctx.prev_state   = prev->state;
 741         entry->ctx.next_pid     = next->pid;
 742         entry->ctx.next_prio    = next->prio;
 743         entry->ctx.next_state   = next->state;
 744         __trace_stack(tr, data, flags, 4);
 745         __raw_spin_unlock(&data->lock);
 746         raw_local_irq_restore(irq_flags);
 747 }
 748
 749 void
 750 tracing_sched_wakeup_trace(struct trace_array *tr,
 751                            struct trace_array_cpu *data,
 752                            struct task_struct *wakee,
 753                            struct task_struct *curr,
 754                            unsigned long flags)
 755 {
 756         struct trace_entry *entry;
 757         unsigned long irq_flags;
 758
 759         raw_local_irq_save(irq_flags);
 760         __raw_spin_lock(&data->lock);
 761         entry                   = tracing_get_trace_entry(tr, data);
 762         tracing_generic_entry_update(entry, flags);
 763         entry->type             = TRACE_WAKE;
 764         entry->ctx.prev_pid     = curr->pid;
 765         entry->ctx.prev_prio    = curr->prio;
 766         entry->ctx.prev_state   = curr->state;
 767         entry->ctx.next_pid     = wakee->pid;
 768         entry->ctx.next_prio    = wakee->prio;
 769         entry->ctx.next_state   = wakee->state;
 770         __trace_stack(tr, data, flags, 5);
 771         __raw_spin_unlock(&data->lock);
 772         raw_local_irq_restore(irq_flags);
 773
 774         trace_wake_up();
 775 }
 776
 777 #ifdef CONFIG_FTRACE
 778 static void
 779 function_trace_call(unsigned long ip, unsigned long parent_ip)
 780 {
 781         struct trace_array *tr = &global_trace;
 782         struct trace_array_cpu *data;
 783         unsigned long flags;
 784         long disabled;
 785         int cpu;
 786
 787         if (unlikely(!tracer_enabled))
 788                 return;
 789
 790         local_irq_save(flags);
 791         cpu = raw_smp_processor_id();
 792         data = tr->data[cpu];
 793         disabled = atomic_inc_return(&data->disabled);
 794
 795         if (likely(disabled == 1))
 796                 trace_function(tr, data, ip, parent_ip, flags);
 797
 798         atomic_dec(&data->disabled);
 799         local_irq_restore(flags);
 800 }
 801
 802 static struct ftrace_ops trace_ops __read_mostly =
 803 {
 804         .func = function_trace_call,
 805 };
 806
 807 void tracing_start_function_trace(void)
 808 {
 809         register_ftrace_function(&trace_ops);
 810 }
 811
 812 void tracing_stop_function_trace(void)
 813 {
 814         unregister_ftrace_function(&trace_ops);
 815 }
 816 #endif
 817
 818 enum trace_file_type {
 819         TRACE_FILE_LAT_FMT      = 1,
 820 };
 821
 822 static struct trace_entry *
 823 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 824                 struct trace_iterator *iter, int cpu)
 825 {
 826         struct page *page;
 827         struct trace_entry *array;
 828
 829         if (iter->next_idx[cpu] >= tr->entries ||
 830             iter->next_idx[cpu] >= data->trace_idx ||
 831             (data->trace_head == data->trace_tail &&
 832              data->trace_head_idx == data->trace_tail_idx))
 833                 return NULL;
 834
 835         if (!iter->next_page[cpu]) {
 836                 /* Initialize the iterator for this cpu trace buffer */
 837                 WARN_ON(!data->trace_tail);
 838                 page = virt_to_page(data->trace_tail);
 839                 iter->next_page[cpu] = &page->lru;
 840                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 841         }
 842
 843         page = list_entry(iter->next_page[cpu], struct page, lru);
 844         BUG_ON(&data->trace_pages == &page->lru);
 845
 846         array = page_address(page);
 847
 848         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 849         return &array[iter->next_page_idx[cpu]];
 850 }
 851
 852 static struct trace_entry *
 853 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 854 {
 855         struct trace_array *tr = iter->tr;
 856         struct trace_entry *ent, *next = NULL;
 857         int next_cpu = -1;
 858         int cpu;
 859
 860         for_each_possible_cpu(cpu) {
 861                 if (!head_page(tr->data[cpu]))
 862                         continue;
 863                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 864                 /*
 865                  * Pick the entry with the smallest timestamp:
 866                  */
 867                 if (ent && (!next || ent->t < next->t)) {
 868                         next = ent;
 869                         next_cpu = cpu;
 870                 }
 871         }
 872
 873         if (ent_cpu)
 874                 *ent_cpu = next_cpu;
 875
 876         return next;
 877 }
 878
 879 static void trace_iterator_increment(struct trace_iterator *iter)
 880 {
 881         iter->idx++;
 882         iter->next_idx[iter->cpu]++;
 883         iter->next_page_idx[iter->cpu]++;
 884
 885         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 886                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 887
 888                 iter->next_page_idx[iter->cpu] = 0;
 889                 iter->next_page[iter->cpu] =
 890                         trace_next_list(data, iter->next_page[iter->cpu]);
 891         }
 892 }
 893
 894 static void trace_consume(struct trace_iterator *iter)
 895 {
 896         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 897
 898         data->trace_tail_idx++;
 899         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 900                 data->trace_tail = trace_next_page(data, data->trace_tail);
 901                 data->trace_tail_idx = 0;
 902         }
 903
 904         /* Check if we empty it, then reset the index */
 905         if (data->trace_head == data->trace_tail &&
 906             data->trace_head_idx == data->trace_tail_idx)
 907                 data->trace_idx = 0;
 908 }
 909
 910 static void *find_next_entry_inc(struct trace_iterator *iter)
 911 {
 912         struct trace_entry *next;
 913         int next_cpu = -1;
 914
 915         next = find_next_entry(iter, &next_cpu);
 916
 917         iter->prev_ent = iter->ent;
 918         iter->prev_cpu = iter->cpu;
 919
 920         iter->ent = next;
 921         iter->cpu = next_cpu;
 922
 923         if (next)
 924                 trace_iterator_increment(iter);
 925
 926         return next ? iter : NULL;
 927 }
 928
 929 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 930 {
 931         struct trace_iterator *iter = m->private;
 932         void *last_ent = iter->ent;
 933         int i = (int)*pos;
 934         void *ent;
 935
 936         (*pos)++;
 937
 938         /* can't go backwards */
 939         if (iter->idx > i)
 940                 return NULL;
 941
 942         if (iter->idx < 0)
 943                 ent = find_next_entry_inc(iter);
 944         else
 945                 ent = iter;
 946
 947         while (ent && iter->idx < i)
 948                 ent = find_next_entry_inc(iter);
 949
 950         iter->pos = *pos;
 951
 952         if (last_ent && !ent)
 953                 seq_puts(m, "\n\nvim:ft=help\n");
 954
 955         return ent;
 956 }
 957
 958 static void *s_start(struct seq_file *m, loff_t *pos)
 959 {
 960         struct trace_iterator *iter = m->private;
 961         void *p = NULL;
 962         loff_t l = 0;
 963         int i;
 964
 965         mutex_lock(&trace_types_lock);
 966
 967         if (!current_trace || current_trace != iter->trace)
 968                 return NULL;
 969
 970         atomic_inc(&trace_record_cmdline_disabled);
 971
 972         /* let the tracer grab locks here if needed */
 973         if (current_trace->start)
 974                 current_trace->start(iter);
 975
 976         if (*pos != iter->pos) {
 977                 iter->ent = NULL;
 978                 iter->cpu = 0;
 979                 iter->idx = -1;
 980                 iter->prev_ent = NULL;
 981                 iter->prev_cpu = -1;
 982
 983                 for_each_possible_cpu(i) {
 984                         iter->next_idx[i] = 0;
 985                         iter->next_page[i] = NULL;
 986                 }
 987
 988                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 989                         ;
 990
 991         } else {
 992                 l = *pos - 1;
 993                 p = s_next(m, p, &l);
 994         }
 995
 996         return p;
 997 }
 998
 999 static void s_stop(struct seq_file *m, void *p)
1000 {
1001         struct trace_iterator *iter = m->private;
1002
1003         atomic_dec(&trace_record_cmdline_disabled);
1004
1005         /* let the tracer release locks here if needed */
1006         if (current_trace && current_trace == iter->trace && iter->trace->stop)
1007                 iter->trace->stop(iter);
1008
1009         mutex_unlock(&trace_types_lock);
1010 }
1011
1012 static int
1013 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1014 {
1015 #ifdef CONFIG_KALLSYMS
1016         char str[KSYM_SYMBOL_LEN];
1017
1018         kallsyms_lookup(address, NULL, NULL, NULL, str);
1019
1020         return trace_seq_printf(s, fmt, str);
1021 #endif
1022         return 1;
1023 }
1024
1025 static int
1026 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1027                      unsigned long address)
1028 {
1029 #ifdef CONFIG_KALLSYMS
1030         char str[KSYM_SYMBOL_LEN];
1031
1032         sprint_symbol(str, address);
1033         return trace_seq_printf(s, fmt, str);
1034 #endif
1035         return 1;
1036 }
1037
1038 #ifndef CONFIG_64BIT
1039 # define IP_FMT "%08lx"
1040 #else
1041 # define IP_FMT "%016lx"
1042 #endif
1043
1044 static int
1045 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1046 {
1047         int ret;
1048
1049         if (!ip)
1050                 return trace_seq_printf(s, "0");
1051
1052         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1053                 ret = seq_print_sym_offset(s, "%s", ip);
1054         else
1055                 ret = seq_print_sym_short(s, "%s", ip);
1056
1057         if (!ret)
1058                 return 0;
1059
1060         if (sym_flags & TRACE_ITER_SYM_ADDR)
1061                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1062         return ret;
1063 }
1064
1065 static void print_lat_help_header(struct seq_file *m)
1066 {
1067         seq_puts(m, "#                _------=> CPU#            \n");
1068         seq_puts(m, "#               / _-----=> irqs-off        \n");
1069         seq_puts(m, "#              | / _----=> need-resched    \n");
1070         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1071         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1072         seq_puts(m, "#              |||| /                      \n");
1073         seq_puts(m, "#              |||||     delay             \n");
1074         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1075         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1076 }
1077
1078 static void print_func_help_header(struct seq_file *m)
1079 {
1080         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1081         seq_puts(m, "#              | |      |          |         |\n");
1082 }
1083
1084
1085 static void
1086 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1087 {
1088         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1089         struct trace_array *tr = iter->tr;
1090         struct trace_array_cpu *data = tr->data[tr->cpu];
1091         struct tracer *type = current_trace;
1092         unsigned long total   = 0;
1093         unsigned long entries = 0;
1094         int cpu;
1095         const char *name = "preemption";
1096
1097         if (type)
1098                 name = type->name;
1099
1100         for_each_possible_cpu(cpu) {
1101                 if (head_page(tr->data[cpu])) {
1102                         total += tr->data[cpu]->trace_idx;
1103                         if (tr->data[cpu]->trace_idx > tr->entries)
1104                                 entries += tr->entries;
1105                         else
1106                                 entries += tr->data[cpu]->trace_idx;
1107                 }
1108         }
1109
1110         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1111                    name, UTS_RELEASE);
1112         seq_puts(m, "-----------------------------------"
1113                  "---------------------------------\n");
1114         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1115                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1116                    nsecs_to_usecs(data->saved_latency),
1117                    entries,
1118                    total,
1119                    tr->cpu,
1120 #if defined(CONFIG_PREEMPT_NONE)
1121                    "server",
1122 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1123                    "desktop",
1124 #elif defined(CONFIG_PREEMPT_DESKTOP)
1125                    "preempt",
1126 #else
1127                    "unknown",
1128 #endif
1129                    /* These are reserved for later use */
1130                    0, 0, 0, 0);
1131 #ifdef CONFIG_SMP
1132         seq_printf(m, " #P:%d)\n", num_online_cpus());
1133 #else
1134         seq_puts(m, ")\n");
1135 #endif
1136         seq_puts(m, "    -----------------\n");
1137         seq_printf(m, "    | task: %.16s-%d "
1138                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1139                    data->comm, data->pid, data->uid, data->nice,
1140                    data->policy, data->rt_priority);
1141         seq_puts(m, "    -----------------\n");
1142
1143         if (data->critical_start) {
1144                 seq_puts(m, " => started at: ");
1145                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1146                 trace_print_seq(m, &iter->seq);
1147                 seq_puts(m, "\n => ended at:   ");
1148                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1149                 trace_print_seq(m, &iter->seq);
1150                 seq_puts(m, "\n");
1151         }
1152
1153         seq_puts(m, "\n");
1154 }
1155
1156 static void
1157 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1158 {
1159         int hardirq, softirq;
1160         char *comm;
1161
1162         comm = trace_find_cmdline(entry->pid);
1163
1164         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1165         trace_seq_printf(s, "%d", cpu);
1166         trace_seq_printf(s, "%c%c",
1167                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1168                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1169
1170         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1171         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1172         if (hardirq && softirq)
1173                 trace_seq_putc(s, 'H');
1174         else {
1175                 if (hardirq)
1176                         trace_seq_putc(s, 'h');
1177                 else {
1178                         if (softirq)
1179                                 trace_seq_putc(s, 's');
1180                         else
1181                                 trace_seq_putc(s, '.');
1182                 }
1183         }
1184
1185         if (entry->preempt_count)
1186                 trace_seq_printf(s, "%x", entry->preempt_count);
1187         else
1188                 trace_seq_puts(s, ".");
1189 }
1190
1191 unsigned long preempt_mark_thresh = 100;
1192
1193 static void
1194 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1195                     unsigned long rel_usecs)
1196 {
1197         trace_seq_printf(s, " %4lldus", abs_usecs);
1198         if (rel_usecs > preempt_mark_thresh)
1199                 trace_seq_puts(s, "!: ");
1200         else if (rel_usecs > 1)
1201                 trace_seq_puts(s, "+: ");
1202         else
1203                 trace_seq_puts(s, " : ");
1204 }
1205
1206 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1207
1208 static int
1209 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1210 {
1211         struct trace_seq *s = &iter->seq;
1212         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1213         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1214         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1215         struct trace_entry *entry = iter->ent;
1216         unsigned long abs_usecs;
1217         unsigned long rel_usecs;
1218         char *comm;
1219         int S, T;
1220         int i;
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                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1252                         state_to_char[entry->ctx.prev_state] : 'X';
1253                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1254                         state_to_char[entry->ctx.next_state] : 'X';
1255
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 (trace_flags & TRACE_ITER_BIN)
1543                 return print_bin_fmt(iter);
1544
1545         if (trace_flags & TRACE_ITER_HEX)
1546                 return print_hex_fmt(iter);
1547
1548         if (trace_flags & TRACE_ITER_RAW)
1549                 return print_raw_fmt(iter);
1550
1551         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1552                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1553
1554         return print_trace_fmt(iter);
1555 }
1556
1557 static int s_show(struct seq_file *m, void *v)
1558 {
1559         struct trace_iterator *iter = v;
1560
1561         if (iter->ent == NULL) {
1562                 if (iter->tr) {
1563                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1564                         seq_puts(m, "#\n");
1565                 }
1566                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1567                         /* print nothing if the buffers are empty */
1568                         if (trace_empty(iter))
1569                                 return 0;
1570                         print_trace_header(m, iter);
1571                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1572                                 print_lat_help_header(m);
1573                 } else {
1574                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1575                                 print_func_help_header(m);
1576                 }
1577         } else {
1578                 print_trace_line(iter);
1579                 trace_print_seq(m, &iter->seq);
1580         }
1581
1582         return 0;
1583 }
1584
1585 static struct seq_operations tracer_seq_ops = {
1586         .start          = s_start,
1587         .next           = s_next,
1588         .stop           = s_stop,
1589         .show           = s_show,
1590 };
1591
1592 static struct trace_iterator *
1593 __tracing_open(struct inode *inode, struct file *file, int *ret)
1594 {
1595         struct trace_iterator *iter;
1596
1597         if (tracing_disabled) {
1598                 *ret = -ENODEV;
1599                 return NULL;
1600         }
1601
1602         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1603         if (!iter) {
1604                 *ret = -ENOMEM;
1605                 goto out;
1606         }
1607
1608         mutex_lock(&trace_types_lock);
1609         if (current_trace && current_trace->print_max)
1610                 iter->tr = &max_tr;
1611         else
1612                 iter->tr = inode->i_private;
1613         iter->trace = current_trace;
1614         iter->pos = -1;
1615
1616         /* TODO stop tracer */
1617         *ret = seq_open(file, &tracer_seq_ops);
1618         if (!*ret) {
1619                 struct seq_file *m = file->private_data;
1620                 m->private = iter;
1621
1622                 /* stop the trace while dumping */
1623                 if (iter->tr->ctrl)
1624                         tracer_enabled = 0;
1625
1626                 if (iter->trace && iter->trace->open)
1627                         iter->trace->open(iter);
1628         } else {
1629                 kfree(iter);
1630                 iter = NULL;
1631         }
1632         mutex_unlock(&trace_types_lock);
1633
1634  out:
1635         return iter;
1636 }
1637
1638 int tracing_open_generic(struct inode *inode, struct file *filp)
1639 {
1640         if (tracing_disabled)
1641                 return -ENODEV;
1642
1643         filp->private_data = inode->i_private;
1644         return 0;
1645 }
1646
1647 int tracing_release(struct inode *inode, struct file *file)
1648 {
1649         struct seq_file *m = (struct seq_file *)file->private_data;
1650         struct trace_iterator *iter = m->private;
1651
1652         mutex_lock(&trace_types_lock);
1653         if (iter->trace && iter->trace->close)
1654                 iter->trace->close(iter);
1655
1656         /* reenable tracing if it was previously enabled */
1657         if (iter->tr->ctrl)
1658                 tracer_enabled = 1;
1659         mutex_unlock(&trace_types_lock);
1660
1661         seq_release(inode, file);
1662         kfree(iter);
1663         return 0;
1664 }
1665
1666 static int tracing_open(struct inode *inode, struct file *file)
1667 {
1668         int ret;
1669
1670         __tracing_open(inode, file, &ret);
1671
1672         return ret;
1673 }
1674
1675 static int tracing_lt_open(struct inode *inode, struct file *file)
1676 {
1677         struct trace_iterator *iter;
1678         int ret;
1679
1680         iter = __tracing_open(inode, file, &ret);
1681
1682         if (!ret)
1683                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1684
1685         return ret;
1686 }
1687
1688
1689 static void *
1690 t_next(struct seq_file *m, void *v, loff_t *pos)
1691 {
1692         struct tracer *t = m->private;
1693
1694         (*pos)++;
1695
1696         if (t)
1697                 t = t->next;
1698
1699         m->private = t;
1700
1701         return t;
1702 }
1703
1704 static void *t_start(struct seq_file *m, loff_t *pos)
1705 {
1706         struct tracer *t = m->private;
1707         loff_t l = 0;
1708
1709         mutex_lock(&trace_types_lock);
1710         for (; t && l < *pos; t = t_next(m, t, &l))
1711                 ;
1712
1713         return t;
1714 }
1715
1716 static void t_stop(struct seq_file *m, void *p)
1717 {
1718         mutex_unlock(&trace_types_lock);
1719 }
1720
1721 static int t_show(struct seq_file *m, void *v)
1722 {
1723         struct tracer *t = v;
1724
1725         if (!t)
1726                 return 0;
1727
1728         seq_printf(m, "%s", t->name);
1729         if (t->next)
1730                 seq_putc(m, ' ');
1731         else
1732                 seq_putc(m, '\n');
1733
1734         return 0;
1735 }
1736
1737 static struct seq_operations show_traces_seq_ops = {
1738         .start          = t_start,
1739         .next           = t_next,
1740         .stop           = t_stop,
1741         .show           = t_show,
1742 };
1743
1744 static int show_traces_open(struct inode *inode, struct file *file)
1745 {
1746         int ret;
1747
1748         if (tracing_disabled)
1749                 return -ENODEV;
1750
1751         ret = seq_open(file, &show_traces_seq_ops);
1752         if (!ret) {
1753                 struct seq_file *m = file->private_data;
1754                 m->private = trace_types;
1755         }
1756
1757         return ret;
1758 }
1759
1760 static struct file_operations tracing_fops = {
1761         .open           = tracing_open,
1762         .read           = seq_read,
1763         .llseek         = seq_lseek,
1764         .release        = tracing_release,
1765 };
1766
1767 static struct file_operations tracing_lt_fops = {
1768         .open           = tracing_lt_open,
1769         .read           = seq_read,
1770         .llseek         = seq_lseek,
1771         .release        = tracing_release,
1772 };
1773
1774 static struct file_operations show_traces_fops = {
1775         .open           = show_traces_open,
1776         .read           = seq_read,
1777         .release        = seq_release,
1778 };
1779
1780 /*
1781  * Only trace on a CPU if the bitmask is set:
1782  */
1783 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1784
1785 /*
1786  * When tracing/tracing_cpu_mask is modified then this holds
1787  * the new bitmask we are about to install:
1788  */
1789 static cpumask_t tracing_cpumask_new;
1790
1791 /*
1792  * The tracer itself will not take this lock, but still we want
1793  * to provide a consistent cpumask to user-space:
1794  */
1795 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1796
1797 /*
1798  * Temporary storage for the character representation of the
1799  * CPU bitmask (and one more byte for the newline):
1800  */
1801 static char mask_str[NR_CPUS + 1];
1802
1803 static ssize_t
1804 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1805                      size_t count, loff_t *ppos)
1806 {
1807         int len;
1808
1809         mutex_lock(&tracing_cpumask_update_lock);
1810
1811         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1812         if (count - len < 2) {
1813                 count = -EINVAL;
1814                 goto out_err;
1815         }
1816         len += sprintf(mask_str + len, "\n");
1817         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1818
1819 out_err:
1820         mutex_unlock(&tracing_cpumask_update_lock);
1821
1822         return count;
1823 }
1824
1825 static ssize_t
1826 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1827                       size_t count, loff_t *ppos)
1828 {
1829         int err, cpu;
1830
1831         mutex_lock(&tracing_cpumask_update_lock);
1832         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1833         if (err)
1834                 goto err_unlock;
1835
1836         raw_local_irq_disable();
1837         __raw_spin_lock(&ftrace_max_lock);
1838         for_each_possible_cpu(cpu) {
1839                 /*
1840                  * Increase/decrease the disabled counter if we are
1841                  * about to flip a bit in the cpumask:
1842                  */
1843                 if (cpu_isset(cpu, tracing_cpumask) &&
1844                                 !cpu_isset(cpu, tracing_cpumask_new)) {
1845                         atomic_inc(&global_trace.data[cpu]->disabled);
1846                 }
1847                 if (!cpu_isset(cpu, tracing_cpumask) &&
1848                                 cpu_isset(cpu, tracing_cpumask_new)) {
1849                         atomic_dec(&global_trace.data[cpu]->disabled);
1850                 }
1851         }
1852         __raw_spin_unlock(&ftrace_max_lock);
1853         raw_local_irq_enable();
1854
1855         tracing_cpumask = tracing_cpumask_new;
1856
1857         mutex_unlock(&tracing_cpumask_update_lock);
1858
1859         return count;
1860
1861 err_unlock:
1862         mutex_unlock(&tracing_cpumask_update_lock);
1863
1864         return err;
1865 }
1866
1867 static struct file_operations tracing_cpumask_fops = {
1868         .open           = tracing_open_generic,
1869         .read           = tracing_cpumask_read,
1870         .write          = tracing_cpumask_write,
1871 };
1872
1873 static ssize_t
1874 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1875                        size_t cnt, loff_t *ppos)
1876 {
1877         char *buf;
1878         int r = 0;
1879         int len = 0;
1880         int i;
1881
1882         /* calulate max size */
1883         for (i = 0; trace_options[i]; i++) {
1884                 len += strlen(trace_options[i]);
1885                 len += 3; /* "no" and space */
1886         }
1887
1888         /* +2 for \n and \0 */
1889         buf = kmalloc(len + 2, GFP_KERNEL);
1890         if (!buf)
1891                 return -ENOMEM;
1892
1893         for (i = 0; trace_options[i]; i++) {
1894                 if (trace_flags & (1 << i))
1895                         r += sprintf(buf + r, "%s ", trace_options[i]);
1896                 else
1897                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1898         }
1899
1900         r += sprintf(buf + r, "\n");
1901         WARN_ON(r >= len + 2);
1902
1903         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1904
1905         kfree(buf);
1906
1907         return r;
1908 }
1909
1910 static ssize_t
1911 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1912                         size_t cnt, loff_t *ppos)
1913 {
1914         char buf[64];
1915         char *cmp = buf;
1916         int neg = 0;
1917         int i;
1918
1919         if (cnt > 63)
1920                 cnt = 63;
1921
1922         if (copy_from_user(&buf, ubuf, cnt))
1923                 return -EFAULT;
1924
1925         buf[cnt] = 0;
1926
1927         if (strncmp(buf, "no", 2) == 0) {
1928                 neg = 1;
1929                 cmp += 2;
1930         }
1931
1932         for (i = 0; trace_options[i]; i++) {
1933                 int len = strlen(trace_options[i]);
1934
1935                 if (strncmp(cmp, trace_options[i], len) == 0) {
1936                         if (neg)
1937                                 trace_flags &= ~(1 << i);
1938                         else
1939                                 trace_flags |= (1 << i);
1940                         break;
1941                 }
1942         }
1943         /*
1944          * If no option could be set, return an error:
1945          */
1946         if (!trace_options[i])
1947                 return -EINVAL;
1948
1949         filp->f_pos += cnt;
1950
1951         return cnt;
1952 }
1953
1954 static struct file_operations tracing_iter_fops = {
1955         .open           = tracing_open_generic,
1956         .read           = tracing_iter_ctrl_read,
1957         .write          = tracing_iter_ctrl_write,
1958 };
1959
1960 static const char readme_msg[] =
1961         "tracing mini-HOWTO:\n\n"
1962         "# mkdir /debug\n"
1963         "# mount -t debugfs nodev /debug\n\n"
1964         "# cat /debug/tracing/available_tracers\n"
1965         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1966         "# cat /debug/tracing/current_tracer\n"
1967         "none\n"
1968         "# echo sched_switch > /debug/tracing/current_tracer\n"
1969         "# cat /debug/tracing/current_tracer\n"
1970         "sched_switch\n"
1971         "# cat /debug/tracing/iter_ctrl\n"
1972         "noprint-parent nosym-offset nosym-addr noverbose\n"
1973         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1974         "# echo 1 > /debug/tracing/tracing_enabled\n"
1975         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1976         "echo 0 > /debug/tracing/tracing_enabled\n"
1977 ;
1978
1979 static ssize_t
1980 tracing_readme_read(struct file *filp, char __user *ubuf,
1981                        size_t cnt, loff_t *ppos)
1982 {
1983         return simple_read_from_buffer(ubuf, cnt, ppos,
1984                                         readme_msg, strlen(readme_msg));
1985 }
1986
1987 static struct file_operations tracing_readme_fops = {
1988         .open           = tracing_open_generic,
1989         .read           = tracing_readme_read,
1990 };
1991
1992 static ssize_t
1993 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1994                   size_t cnt, loff_t *ppos)
1995 {
1996         struct trace_array *tr = filp->private_data;
1997         char buf[64];
1998         int r;
1999
2000         r = sprintf(buf, "%ld\n", tr->ctrl);
2001         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2002 }
2003
2004 static ssize_t
2005 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2006                    size_t cnt, loff_t *ppos)
2007 {
2008         struct trace_array *tr = filp->private_data;
2009         long val;
2010         char buf[64];
2011
2012         if (cnt > 63)
2013                 cnt = 63;
2014
2015         if (copy_from_user(&buf, ubuf, cnt))
2016                 return -EFAULT;
2017
2018         buf[cnt] = 0;
2019
2020         val = simple_strtoul(buf, NULL, 10);
2021
2022         val = !!val;
2023
2024         mutex_lock(&trace_types_lock);
2025         if (tr->ctrl ^ val) {
2026                 if (val)
2027                         tracer_enabled = 1;
2028                 else
2029                         tracer_enabled = 0;
2030
2031                 tr->ctrl = val;
2032
2033                 if (current_trace && current_trace->ctrl_update)
2034                         current_trace->ctrl_update(tr);
2035         }
2036         mutex_unlock(&trace_types_lock);
2037
2038         filp->f_pos += cnt;
2039
2040         return cnt;
2041 }
2042
2043 static ssize_t
2044 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2045                        size_t cnt, loff_t *ppos)
2046 {
2047         char buf[max_tracer_type_len+2];
2048         int r;
2049
2050         mutex_lock(&trace_types_lock);
2051         if (current_trace)
2052                 r = sprintf(buf, "%s\n", current_trace->name);
2053         else
2054                 r = sprintf(buf, "\n");
2055         mutex_unlock(&trace_types_lock);
2056
2057         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2058 }
2059
2060 static ssize_t
2061 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2062                         size_t cnt, loff_t *ppos)
2063 {
2064         struct trace_array *tr = &global_trace;
2065         struct tracer *t;
2066         char buf[max_tracer_type_len+1];
2067         int i;
2068
2069         if (cnt > max_tracer_type_len)
2070                 cnt = max_tracer_type_len;
2071
2072         if (copy_from_user(&buf, ubuf, cnt))
2073                 return -EFAULT;
2074
2075         buf[cnt] = 0;
2076
2077         /* strip ending whitespace. */
2078         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2079                 buf[i] = 0;
2080
2081         mutex_lock(&trace_types_lock);
2082         for (t = trace_types; t; t = t->next) {
2083                 if (strcmp(t->name, buf) == 0)
2084                         break;
2085         }
2086         if (!t || t == current_trace)
2087                 goto out;
2088
2089         if (current_trace && current_trace->reset)
2090                 current_trace->reset(tr);
2091
2092         current_trace = t;
2093         if (t->init)
2094                 t->init(tr);
2095
2096  out:
2097         mutex_unlock(&trace_types_lock);
2098
2099         filp->f_pos += cnt;
2100
2101         return cnt;
2102 }
2103
2104 static ssize_t
2105 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2106                      size_t cnt, loff_t *ppos)
2107 {
2108         unsigned long *ptr = filp->private_data;
2109         char buf[64];
2110         int r;
2111
2112         r = snprintf(buf, 64, "%ld\n",
2113                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2114         if (r > 64)
2115                 r = 64;
2116         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2117 }
2118
2119 static ssize_t
2120 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2121                       size_t cnt, loff_t *ppos)
2122 {
2123         long *ptr = filp->private_data;
2124         long val;
2125         char buf[64];
2126
2127         if (cnt > 63)
2128                 cnt = 63;
2129
2130         if (copy_from_user(&buf, ubuf, cnt))
2131                 return -EFAULT;
2132
2133         buf[cnt] = 0;
2134
2135         val = simple_strtoul(buf, NULL, 10);
2136
2137         *ptr = val * 1000;
2138
2139         return cnt;
2140 }
2141
2142 static atomic_t tracing_reader;
2143
2144 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2145 {
2146         struct trace_iterator *iter;
2147
2148         if (tracing_disabled)
2149                 return -ENODEV;
2150
2151         /* We only allow for reader of the pipe */
2152         if (atomic_inc_return(&tracing_reader) != 1) {
2153                 atomic_dec(&tracing_reader);
2154                 return -EBUSY;
2155         }
2156
2157         /* create a buffer to store the information to pass to userspace */
2158         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2159         if (!iter)
2160                 return -ENOMEM;
2161
2162         iter->tr = &global_trace;
2163
2164         filp->private_data = iter;
2165
2166         return 0;
2167 }
2168
2169 static int tracing_release_pipe(struct inode *inode, struct file *file)
2170 {
2171         struct trace_iterator *iter = file->private_data;
2172
2173         kfree(iter);
2174         atomic_dec(&tracing_reader);
2175
2176         return 0;
2177 }
2178
2179 static unsigned int
2180 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2181 {
2182         struct trace_iterator *iter = filp->private_data;
2183
2184         if (trace_flags & TRACE_ITER_BLOCK) {
2185                 /*
2186                  * Always select as readable when in blocking mode
2187                  */
2188                 return POLLIN | POLLRDNORM;
2189         }
2190         else {
2191                 if (!trace_empty(iter))
2192                         return POLLIN | POLLRDNORM;
2193                 poll_wait(filp, &trace_wait, poll_table);
2194                 if (!trace_empty(iter))
2195                         return POLLIN | POLLRDNORM;
2196
2197                 return 0;
2198         }
2199 }
2200
2201 /*
2202  * Consumer reader.
2203  */
2204 static ssize_t
2205 tracing_read_pipe(struct file *filp, char __user *ubuf,
2206                   size_t cnt, loff_t *ppos)
2207 {
2208         struct trace_iterator *iter = filp->private_data;
2209         struct trace_array_cpu *data;
2210         static cpumask_t mask;
2211         static int start;
2212         unsigned long flags;
2213 #ifdef CONFIG_FTRACE
2214         int ftrace_save;
2215 #endif
2216         int read = 0;
2217         int cpu;
2218         int len;
2219         int ret;
2220
2221         /* return any leftover data */
2222         if (iter->seq.len > start) {
2223                 len = iter->seq.len - start;
2224                 if (cnt > len)
2225                         cnt = len;
2226                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2227                 if (ret)
2228                         cnt = -EFAULT;
2229
2230                 start += len;
2231
2232                 return cnt;
2233         }
2234
2235         trace_seq_reset(&iter->seq);
2236         start = 0;
2237
2238         while (trace_empty(iter)) {
2239                 if (!(trace_flags & TRACE_ITER_BLOCK))
2240                         return -EWOULDBLOCK;
2241                 /*
2242                  * This is a make-shift waitqueue. The reason we don't use
2243                  * an actual wait queue is because:
2244                  *  1) we only ever have one waiter
2245                  *  2) the tracing, traces all functions, we don't want
2246                  *     the overhead of calling wake_up and friends
2247                  *     (and tracing them too)
2248                  *     Anyway, this is really very primitive wakeup.
2249                  */
2250                 set_current_state(TASK_INTERRUPTIBLE);
2251                 iter->tr->waiter = current;
2252
2253                 /* sleep for one second, and try again. */
2254                 schedule_timeout(HZ);
2255
2256                 iter->tr->waiter = NULL;
2257
2258                 if (signal_pending(current))
2259                         return -EINTR;
2260
2261                 /*
2262                  * We block until we read something and tracing is disabled.
2263                  * We still block if tracing is disabled, but we have never
2264                  * read anything. This allows a user to cat this file, and
2265                  * then enable tracing. But after we have read something,
2266                  * we give an EOF when tracing is again disabled.
2267                  *
2268                  * iter->pos will be 0 if we haven't read anything.
2269                  */
2270                 if (!tracer_enabled && iter->pos)
2271                         break;
2272
2273                 continue;
2274         }
2275
2276         /* stop when tracing is finished */
2277         if (trace_empty(iter))
2278                 return 0;
2279
2280         if (cnt >= PAGE_SIZE)
2281                 cnt = PAGE_SIZE - 1;
2282
2283         memset(iter, 0, sizeof(*iter));
2284         iter->tr = &global_trace;
2285         iter->pos = -1;
2286
2287         /*
2288          * We need to stop all tracing on all CPUS to read the
2289          * the next buffer. This is a bit expensive, but is
2290          * not done often. We fill all what we can read,
2291          * and then release the locks again.
2292          */
2293
2294         cpus_clear(mask);
2295         local_irq_save(flags);
2296 #ifdef CONFIG_FTRACE
2297         ftrace_save = ftrace_enabled;
2298         ftrace_enabled = 0;
2299 #endif
2300         smp_wmb();
2301         for_each_possible_cpu(cpu) {
2302                 data = iter->tr->data[cpu];
2303
2304                 if (!head_page(data) || !data->trace_idx)
2305                         continue;
2306
2307                 atomic_inc(&data->disabled);
2308                 cpu_set(cpu, mask);
2309         }
2310
2311         for_each_cpu_mask(cpu, mask) {
2312                 data = iter->tr->data[cpu];
2313                 __raw_spin_lock(&data->lock);
2314         }
2315
2316         while (find_next_entry_inc(iter) != NULL) {
2317                 int len = iter->seq.len;
2318
2319                 ret = print_trace_line(iter);
2320                 if (!ret) {
2321                         /* don't print partial lines */
2322                         iter->seq.len = len;
2323                         break;
2324                 }
2325
2326                 trace_consume(iter);
2327
2328                 if (iter->seq.len >= cnt)
2329                         break;
2330         }
2331
2332         for_each_cpu_mask(cpu, mask) {
2333                 data = iter->tr->data[cpu];
2334                 __raw_spin_unlock(&data->lock);
2335         }
2336
2337         for_each_cpu_mask(cpu, mask) {
2338                 data = iter->tr->data[cpu];
2339                 atomic_dec(&data->disabled);
2340         }
2341 #ifdef CONFIG_FTRACE
2342         ftrace_enabled = ftrace_save;
2343 #endif
2344         local_irq_restore(flags);
2345
2346         /* Now copy what we have to the user */
2347         read = iter->seq.len;
2348         if (read > cnt)
2349                 read = cnt;
2350
2351         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2352
2353         if (read < iter->seq.len)
2354                 start = read;
2355         else
2356                 trace_seq_reset(&iter->seq);
2357
2358         if (ret)
2359                 read = -EFAULT;
2360
2361         return read;
2362 }
2363
2364 static struct file_operations tracing_max_lat_fops = {
2365         .open           = tracing_open_generic,
2366         .read           = tracing_max_lat_read,
2367         .write          = tracing_max_lat_write,
2368 };
2369
2370 static struct file_operations tracing_ctrl_fops = {
2371         .open           = tracing_open_generic,
2372         .read           = tracing_ctrl_read,
2373         .write          = tracing_ctrl_write,
2374 };
2375
2376 static struct file_operations set_tracer_fops = {
2377         .open           = tracing_open_generic,
2378         .read           = tracing_set_trace_read,
2379         .write          = tracing_set_trace_write,
2380 };
2381
2382 static struct file_operations tracing_pipe_fops = {
2383         .open           = tracing_open_pipe,
2384         .poll           = tracing_poll_pipe,
2385         .read           = tracing_read_pipe,
2386         .release        = tracing_release_pipe,
2387 };
2388
2389 #ifdef CONFIG_DYNAMIC_FTRACE
2390
2391 static ssize_t
2392 tracing_read_long(struct file *filp, char __user *ubuf,
2393                   size_t cnt, loff_t *ppos)
2394 {
2395         unsigned long *p = filp->private_data;
2396         char buf[64];
2397         int r;
2398
2399         r = sprintf(buf, "%ld\n", *p);
2400
2401         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2402 }
2403
2404 static struct file_operations tracing_read_long_fops = {
2405         .open           = tracing_open_generic,
2406         .read           = tracing_read_long,
2407 };
2408 #endif
2409
2410 static struct dentry *d_tracer;
2411
2412 struct dentry *tracing_init_dentry(void)
2413 {
2414         static int once;
2415
2416         if (d_tracer)
2417                 return d_tracer;
2418
2419         d_tracer = debugfs_create_dir("tracing", NULL);
2420
2421         if (!d_tracer && !once) {
2422                 once = 1;
2423                 pr_warning("Could not create debugfs directory 'tracing'\n");
2424                 return NULL;
2425         }
2426
2427         return d_tracer;
2428 }
2429
2430 #ifdef CONFIG_FTRACE_SELFTEST
2431 /* Let selftest have access to static functions in this file */
2432 #include "trace_selftest.c"
2433 #endif
2434
2435 static __init void tracer_init_debugfs(void)
2436 {
2437         struct dentry *d_tracer;
2438         struct dentry *entry;
2439
2440         d_tracer = tracing_init_dentry();
2441
2442         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2443                                     &global_trace, &tracing_ctrl_fops);
2444         if (!entry)
2445                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2446
2447         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2448                                     NULL, &tracing_iter_fops);
2449         if (!entry)
2450                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2451
2452         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2453                                     NULL, &tracing_cpumask_fops);
2454         if (!entry)
2455                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2456
2457         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2458                                     &global_trace, &tracing_lt_fops);
2459         if (!entry)
2460                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2461
2462         entry = debugfs_create_file("trace", 0444, d_tracer,
2463                                     &global_trace, &tracing_fops);
2464         if (!entry)
2465                 pr_warning("Could not create debugfs 'trace' entry\n");
2466
2467         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2468                                     &global_trace, &show_traces_fops);
2469         if (!entry)
2470                 pr_warning("Could not create debugfs 'trace' entry\n");
2471
2472         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2473                                     &global_trace, &set_tracer_fops);
2474         if (!entry)
2475                 pr_warning("Could not create debugfs 'trace' entry\n");
2476
2477         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2478                                     &tracing_max_latency,
2479                                     &tracing_max_lat_fops);
2480         if (!entry)
2481                 pr_warning("Could not create debugfs "
2482                            "'tracing_max_latency' entry\n");
2483
2484         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2485                                     &tracing_thresh, &tracing_max_lat_fops);
2486         if (!entry)
2487                 pr_warning("Could not create debugfs "
2488                            "'tracing_threash' entry\n");
2489         entry = debugfs_create_file("README", 0644, d_tracer,
2490                                     NULL, &tracing_readme_fops);
2491         if (!entry)
2492                 pr_warning("Could not create debugfs 'README' entry\n");
2493
2494         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2495                                     NULL, &tracing_pipe_fops);
2496         if (!entry)
2497                 pr_warning("Could not create debugfs "
2498                            "'tracing_threash' entry\n");
2499
2500 #ifdef CONFIG_DYNAMIC_FTRACE
2501         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2502                                     &ftrace_update_tot_cnt,
2503                                     &tracing_read_long_fops);
2504         if (!entry)
2505                 pr_warning("Could not create debugfs "
2506                            "'dyn_ftrace_total_info' entry\n");
2507 #endif
2508 }
2509
2510 /* dummy trace to disable tracing */
2511 static struct tracer no_tracer __read_mostly =
2512 {
2513         .name           = "none",
2514 };
2515
2516 static int trace_alloc_page(void)
2517 {
2518         struct trace_array_cpu *data;
2519         struct page *page, *tmp;
2520         LIST_HEAD(pages);
2521         void *array;
2522         int i;
2523
2524         /* first allocate a page for each CPU */
2525         for_each_possible_cpu(i) {
2526                 array = (void *)__get_free_page(GFP_KERNEL);
2527                 if (array == NULL) {
2528                         printk(KERN_ERR "tracer: failed to allocate page"
2529                                "for trace buffer!\n");
2530                         goto free_pages;
2531                 }
2532
2533                 page = virt_to_page(array);
2534                 list_add(&page->lru, &pages);
2535
2536 /* Only allocate if we are actually using the max trace */
2537 #ifdef CONFIG_TRACER_MAX_TRACE
2538                 array = (void *)__get_free_page(GFP_KERNEL);
2539                 if (array == NULL) {
2540                         printk(KERN_ERR "tracer: failed to allocate page"
2541                                "for trace buffer!\n");
2542                         goto free_pages;
2543                 }
2544                 page = virt_to_page(array);
2545                 list_add(&page->lru, &pages);
2546 #endif
2547         }
2548
2549         /* Now that we successfully allocate a page per CPU, add them */
2550         for_each_possible_cpu(i) {
2551                 data = global_trace.data[i];
2552                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2553                 page = list_entry(pages.next, struct page, lru);
2554                 list_del_init(&page->lru);
2555                 list_add_tail(&page->lru, &data->trace_pages);
2556                 ClearPageLRU(page);
2557
2558 #ifdef CONFIG_TRACER_MAX_TRACE
2559                 data = max_tr.data[i];
2560                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2561                 page = list_entry(pages.next, struct page, lru);
2562                 list_del_init(&page->lru);
2563                 list_add_tail(&page->lru, &data->trace_pages);
2564                 SetPageLRU(page);
2565 #endif
2566         }
2567         global_trace.entries += ENTRIES_PER_PAGE;
2568
2569         return 0;
2570
2571  free_pages:
2572         list_for_each_entry_safe(page, tmp, &pages, lru) {
2573                 list_del_init(&page->lru);
2574                 __free_page(page);
2575         }
2576         return -ENOMEM;
2577 }
2578
2579 __init static int tracer_alloc_buffers(void)
2580 {
2581         struct trace_array_cpu *data;
2582         void *array;
2583         struct page *page;
2584         int pages = 0;
2585         int ret = -ENOMEM;
2586         int i;
2587
2588         global_trace.ctrl = tracer_enabled;
2589
2590         /* Allocate the first page for all buffers */
2591         for_each_possible_cpu(i) {
2592                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2593                 max_tr.data[i] = &per_cpu(max_data, i);
2594
2595                 array = (void *)__get_free_page(GFP_KERNEL);
2596                 if (array == NULL) {
2597                         printk(KERN_ERR "tracer: failed to allocate page"
2598                                "for trace buffer!\n");
2599                         goto free_buffers;
2600                 }
2601
2602                 /* set the array to the list */
2603                 INIT_LIST_HEAD(&data->trace_pages);
2604                 page = virt_to_page(array);
2605                 list_add(&page->lru, &data->trace_pages);
2606                 /* use the LRU flag to differentiate the two buffers */
2607                 ClearPageLRU(page);
2608
2609 /* Only allocate if we are actually using the max trace */
2610 #ifdef CONFIG_TRACER_MAX_TRACE
2611                 array = (void *)__get_free_page(GFP_KERNEL);
2612                 if (array == NULL) {
2613                         printk(KERN_ERR "tracer: failed to allocate page"
2614                                "for trace buffer!\n");
2615                         goto free_buffers;
2616                 }
2617
2618                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2619                 page = virt_to_page(array);
2620                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2621                 SetPageLRU(page);
2622 #endif
2623         }
2624
2625         /*
2626          * Since we allocate by orders of pages, we may be able to
2627          * round up a bit.
2628          */
2629         global_trace.entries = ENTRIES_PER_PAGE;
2630         pages++;
2631
2632         while (global_trace.entries < trace_nr_entries) {
2633                 if (trace_alloc_page())
2634                         break;
2635                 pages++;
2636         }
2637         max_tr.entries = global_trace.entries;
2638
2639         pr_info("tracer: %d pages allocated for %ld",
2640                 pages, trace_nr_entries);
2641         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2642         pr_info("   actual entries %ld\n", global_trace.entries);
2643
2644         tracer_init_debugfs();
2645
2646         trace_init_cmdlines();
2647
2648         register_tracer(&no_tracer);
2649         current_trace = &no_tracer;
2650
2651         /* All seems OK, enable tracing */
2652         tracing_disabled = 0;
2653
2654         return 0;
2655
2656  free_buffers:
2657         for (i-- ; i >= 0; i--) {
2658                 struct page *page, *tmp;
2659                 struct trace_array_cpu *data = global_trace.data[i];
2660
2661                 if (data) {
2662                         list_for_each_entry_safe(page, tmp,
2663                                                  &data->trace_pages, lru) {
2664                                 list_del_init(&page->lru);
2665                                 __free_page(page);
2666                         }
2667                 }
2668
2669 #ifdef CONFIG_TRACER_MAX_TRACE
2670                 data = max_tr.data[i];
2671                 if (data) {
2672                         list_for_each_entry_safe(page, tmp,
2673                                                  &data->trace_pages, lru) {
2674                                 list_del_init(&page->lru);
2675                                 __free_page(page);
2676                         }
2677                 }
2678 #endif
2679         }
2680         return ret;
2681 }
2682 fs_initcall(tracer_alloc_buffers);