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 "trace.h"
  32
  33 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  34 unsigned long __read_mostly     tracing_thresh;
  35
  36 static int tracing_disabled = 1;
  37
  38 static long
  39 ns2usecs(cycle_t nsec)
  40 {
  41         nsec += 500;
  42         do_div(nsec, 1000);
  43         return nsec;
  44 }
  45
  46 cycle_t ftrace_now(int cpu)
  47 {
  48         return cpu_clock(cpu);
  49 }
  50
  51 static struct trace_array       global_trace;
  52
  53 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  54
  55 static struct trace_array       max_tr;
  56
  57 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  58
  59 static int                      tracer_enabled = 1;
  60 static unsigned long            trace_nr_entries = 65536UL;
  61
  62 static struct tracer            *trace_types __read_mostly;
  63 static struct tracer            *current_trace __read_mostly;
  64 static int                      max_tracer_type_len;
  65
  66 static DEFINE_MUTEX(trace_types_lock);
  67 static DECLARE_WAIT_QUEUE_HEAD (trace_wait);
  68
  69 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  70
  71 static int __init set_nr_entries(char *str)
  72 {
  73         if (!str)
  74                 return 0;
  75         trace_nr_entries = simple_strtoul(str, &str, 0);
  76         return 1;
  77 }
  78 __setup("trace_entries=", set_nr_entries);
  79
  80 unsigned long nsecs_to_usecs(unsigned long nsecs)
  81 {
  82         return nsecs / 1000;
  83 }
  84
  85 enum trace_type {
  86         __TRACE_FIRST_TYPE = 0,
  87
  88         TRACE_FN,
  89         TRACE_CTX,
  90         TRACE_WAKE,
  91         TRACE_SPECIAL,
  92
  93         __TRACE_LAST_TYPE
  94 };
  95
  96 enum trace_flag_type {
  97         TRACE_FLAG_IRQS_OFF             = 0x01,
  98         TRACE_FLAG_NEED_RESCHED         = 0x02,
  99         TRACE_FLAG_HARDIRQ              = 0x04,
 100         TRACE_FLAG_SOFTIRQ              = 0x08,
 101 };
 102
 103 enum trace_iterator_flags {
 104         TRACE_ITER_PRINT_PARENT         = 0x01,
 105         TRACE_ITER_SYM_OFFSET           = 0x02,
 106         TRACE_ITER_SYM_ADDR             = 0x04,
 107         TRACE_ITER_VERBOSE              = 0x08,
 108         TRACE_ITER_RAW                  = 0x10,
 109         TRACE_ITER_HEX                  = 0x20,
 110         TRACE_ITER_BIN                  = 0x40,
 111         TRACE_ITER_BLOCK                = 0x80,
 112 };
 113
 114 #define TRACE_ITER_SYM_MASK \
 115         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 116
 117 /* These must match the bit postions above */
 118 static const char *trace_options[] = {
 119         "print-parent",
 120         "sym-offset",
 121         "sym-addr",
 122         "verbose",
 123         "raw",
 124         "hex",
 125         "bin",
 126         "block",
 127         NULL
 128 };
 129
 130 static unsigned trace_flags;
 131
 132 static DEFINE_SPINLOCK(ftrace_max_lock);
 133
 134 /*
 135  * Copy the new maximum trace into the separate maximum-trace
 136  * structure. (this way the maximum trace is permanently saved,
 137  * for later retrieval via /debugfs/tracing/latency_trace)
 138  */
 139 static void
 140 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 141 {
 142         struct trace_array_cpu *data = tr->data[cpu];
 143
 144         max_tr.cpu = cpu;
 145         max_tr.time_start = data->preempt_timestamp;
 146
 147         data = max_tr.data[cpu];
 148         data->saved_latency = tracing_max_latency;
 149
 150         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 151         data->pid = tsk->pid;
 152         data->uid = tsk->uid;
 153         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 154         data->policy = tsk->policy;
 155         data->rt_priority = tsk->rt_priority;
 156
 157         /* record this tasks comm */
 158         tracing_record_cmdline(current);
 159 }
 160
 161 void check_pages(struct trace_array_cpu *data)
 162 {
 163         struct page *page, *tmp;
 164
 165         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 166         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 167
 168         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 169                 BUG_ON(page->lru.next->prev != &page->lru);
 170                 BUG_ON(page->lru.prev->next != &page->lru);
 171         }
 172 }
 173
 174 void *head_page(struct trace_array_cpu *data)
 175 {
 176         struct page *page;
 177
 178         check_pages(data);
 179         if (list_empty(&data->trace_pages))
 180                 return NULL;
 181
 182         page = list_entry(data->trace_pages.next, struct page, lru);
 183         BUG_ON(&page->lru == &data->trace_pages);
 184
 185         return page_address(page);
 186 }
 187
 188 static int
 189 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 190 {
 191         int len = (PAGE_SIZE - 1) - s->len;
 192         va_list ap;
 193         int ret;
 194
 195         if (!len)
 196                 return 0;
 197
 198         va_start(ap, fmt);
 199         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 200         va_end(ap);
 201
 202         /* If we can't write it all, don't bother writing anything */
 203         if (ret > len)
 204                 return 0;
 205
 206         s->len += ret;
 207
 208         return len;
 209 }
 210
 211 static int
 212 trace_seq_puts(struct trace_seq *s, const char *str)
 213 {
 214         int len = strlen(str);
 215
 216         if (len > ((PAGE_SIZE - 1) - s->len))
 217                 return 0;
 218
 219         memcpy(s->buffer + s->len, str, len);
 220         s->len += len;
 221
 222         return len;
 223 }
 224
 225 static int
 226 trace_seq_putc(struct trace_seq *s, unsigned char c)
 227 {
 228         if (s->len >= (PAGE_SIZE - 1))
 229                 return 0;
 230
 231         s->buffer[s->len++] = c;
 232
 233         return 1;
 234 }
 235
 236 static int
 237 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 238 {
 239         if (len > ((PAGE_SIZE - 1) - s->len))
 240                 return 0;
 241
 242         memcpy(s->buffer + s->len, mem, len);
 243         s->len += len;
 244
 245         return len;
 246 }
 247
 248 #define HEX_CHARS 17
 249
 250 static int
 251 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 252 {
 253         unsigned char hex[HEX_CHARS];
 254         unsigned char *data;
 255         unsigned char byte;
 256         int i, j;
 257
 258         BUG_ON(len >= HEX_CHARS);
 259
 260         data = mem;
 261
 262 #ifdef __BIG_ENDIAN
 263         for (i = 0, j = 0; i < len; i++) {
 264 #else
 265         for (i = len-1, j = 0; i >= 0; i--) {
 266 #endif
 267                 byte = data[i];
 268
 269                 hex[j]   = byte & 0x0f;
 270                 if (hex[j] >= 10)
 271                         hex[j] += 'a' - 10;
 272                 else
 273                         hex[j] += '0';
 274                 j++;
 275
 276                 hex[j] = byte >> 4;
 277                 if (hex[j] >= 10)
 278                         hex[j] += 'a' - 10;
 279                 else
 280                         hex[j] += '0';
 281                 j++;
 282         }
 283         hex[j] = ' ';
 284         j++;
 285
 286         return trace_seq_putmem(s, hex, j);
 287 }
 288
 289 static void
 290 trace_seq_reset(struct trace_seq *s)
 291 {
 292         s->len = 0;
 293 }
 294
 295 static void
 296 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 297 {
 298         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 299
 300         s->buffer[len] = 0;
 301         seq_puts(m, s->buffer);
 302
 303         trace_seq_reset(s);
 304 }
 305
 306 static void
 307 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 308 {
 309         struct list_head flip_pages;
 310
 311         INIT_LIST_HEAD(&flip_pages);
 312
 313         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 314                 sizeof(struct trace_array_cpu) -
 315                 offsetof(struct trace_array_cpu, trace_head_idx));
 316
 317         check_pages(tr1);
 318         check_pages(tr2);
 319         list_splice_init(&tr1->trace_pages, &flip_pages);
 320         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 321         list_splice_init(&flip_pages, &tr2->trace_pages);
 322         BUG_ON(!list_empty(&flip_pages));
 323         check_pages(tr1);
 324         check_pages(tr2);
 325 }
 326
 327 void
 328 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 329 {
 330         struct trace_array_cpu *data;
 331         int i;
 332
 333         WARN_ON_ONCE(!irqs_disabled());
 334         spin_lock(&ftrace_max_lock);
 335         /* clear out all the previous traces */
 336         for_each_possible_cpu(i) {
 337                 data = tr->data[i];
 338                 flip_trace(max_tr.data[i], data);
 339                 tracing_reset(data);
 340         }
 341
 342         __update_max_tr(tr, tsk, cpu);
 343         spin_unlock(&ftrace_max_lock);
 344 }
 345
 346 /**
 347  * update_max_tr_single - only copy one trace over, and reset the rest
 348  * @tr - tracer
 349  * @tsk - task with the latency
 350  * @cpu - the cpu of the buffer to copy.
 351  */
 352 void
 353 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 354 {
 355         struct trace_array_cpu *data = tr->data[cpu];
 356         int i;
 357
 358         WARN_ON_ONCE(!irqs_disabled());
 359         spin_lock(&ftrace_max_lock);
 360         for_each_possible_cpu(i)
 361                 tracing_reset(max_tr.data[i]);
 362
 363         flip_trace(max_tr.data[cpu], data);
 364         tracing_reset(data);
 365
 366         __update_max_tr(tr, tsk, cpu);
 367         spin_unlock(&ftrace_max_lock);
 368 }
 369
 370 int register_tracer(struct tracer *type)
 371 {
 372         struct tracer *t;
 373         int len;
 374         int ret = 0;
 375
 376         if (!type->name) {
 377                 pr_info("Tracer must have a name\n");
 378                 return -1;
 379         }
 380
 381         mutex_lock(&trace_types_lock);
 382         for (t = trace_types; t; t = t->next) {
 383                 if (strcmp(type->name, t->name) == 0) {
 384                         /* already found */
 385                         pr_info("Trace %s already registered\n",
 386                                 type->name);
 387                         ret = -1;
 388                         goto out;
 389                 }
 390         }
 391
 392 #ifdef CONFIG_FTRACE_STARTUP_TEST
 393         if (type->selftest) {
 394                 struct tracer *saved_tracer = current_trace;
 395                 struct trace_array_cpu *data;
 396                 struct trace_array *tr = &global_trace;
 397                 int saved_ctrl = tr->ctrl;
 398                 int i;
 399                 /*
 400                  * Run a selftest on this tracer.
 401                  * Here we reset the trace buffer, and set the current
 402                  * tracer to be this tracer. The tracer can then run some
 403                  * internal tracing to verify that everything is in order.
 404                  * If we fail, we do not register this tracer.
 405                  */
 406                 for_each_possible_cpu(i) {
 407                         data = tr->data[i];
 408                         if (!head_page(data))
 409                                 continue;
 410                         tracing_reset(data);
 411                 }
 412                 current_trace = type;
 413                 tr->ctrl = 0;
 414                 /* the test is responsible for initializing and enabling */
 415                 pr_info("Testing tracer %s: ", type->name);
 416                 ret = type->selftest(type, tr);
 417                 /* the test is responsible for resetting too */
 418                 current_trace = saved_tracer;
 419                 tr->ctrl = saved_ctrl;
 420                 if (ret) {
 421                         printk(KERN_CONT "FAILED!\n");
 422                         goto out;
 423                 }
 424                 /* Only reset on passing, to avoid touching corrupted buffers */
 425                 for_each_possible_cpu(i) {
 426                         data = tr->data[i];
 427                         if (!head_page(data))
 428                                 continue;
 429                         tracing_reset(data);
 430                 }
 431                 printk(KERN_CONT "PASSED\n");
 432         }
 433 #endif
 434
 435         type->next = trace_types;
 436         trace_types = type;
 437         len = strlen(type->name);
 438         if (len > max_tracer_type_len)
 439                 max_tracer_type_len = len;
 440
 441  out:
 442         mutex_unlock(&trace_types_lock);
 443
 444         return ret;
 445 }
 446
 447 void unregister_tracer(struct tracer *type)
 448 {
 449         struct tracer **t;
 450         int len;
 451
 452         mutex_lock(&trace_types_lock);
 453         for (t = &trace_types; *t; t = &(*t)->next) {
 454                 if (*t == type)
 455                         goto found;
 456         }
 457         pr_info("Trace %s not registered\n", type->name);
 458         goto out;
 459
 460  found:
 461         *t = (*t)->next;
 462         if (strlen(type->name) != max_tracer_type_len)
 463                 goto out;
 464
 465         max_tracer_type_len = 0;
 466         for (t = &trace_types; *t; t = &(*t)->next) {
 467                 len = strlen((*t)->name);
 468                 if (len > max_tracer_type_len)
 469                         max_tracer_type_len = len;
 470         }
 471  out:
 472         mutex_unlock(&trace_types_lock);
 473 }
 474
 475 void tracing_reset(struct trace_array_cpu *data)
 476 {
 477         data->trace_idx = 0;
 478         data->trace_head = data->trace_tail = head_page(data);
 479         data->trace_head_idx = 0;
 480         data->trace_tail_idx = 0;
 481 }
 482
 483 #define SAVED_CMDLINES 128
 484 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 485 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 486 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 487 static int cmdline_idx;
 488 static DEFINE_SPINLOCK(trace_cmdline_lock);
 489 atomic_t trace_record_cmdline_disabled;
 490
 491 static void trace_init_cmdlines(void)
 492 {
 493         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 494         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 495         cmdline_idx = 0;
 496 }
 497
 498 void trace_stop_cmdline_recording(void);
 499
 500 static void trace_save_cmdline(struct task_struct *tsk)
 501 {
 502         unsigned map;
 503         unsigned idx;
 504
 505         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 506                 return;
 507
 508         /*
 509          * It's not the end of the world if we don't get
 510          * the lock, but we also don't want to spin
 511          * nor do we want to disable interrupts,
 512          * so if we miss here, then better luck next time.
 513          */
 514         if (!spin_trylock(&trace_cmdline_lock))
 515                 return;
 516
 517         idx = map_pid_to_cmdline[tsk->pid];
 518         if (idx >= SAVED_CMDLINES) {
 519                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 520
 521                 map = map_cmdline_to_pid[idx];
 522                 if (map <= PID_MAX_DEFAULT)
 523                         map_pid_to_cmdline[map] = (unsigned)-1;
 524
 525                 map_pid_to_cmdline[tsk->pid] = idx;
 526
 527                 cmdline_idx = idx;
 528         }
 529
 530         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 531
 532         spin_unlock(&trace_cmdline_lock);
 533 }
 534
 535 static char *trace_find_cmdline(int pid)
 536 {
 537         char *cmdline = "<...>";
 538         unsigned map;
 539
 540         if (!pid)
 541                 return "<idle>";
 542
 543         if (pid > PID_MAX_DEFAULT)
 544                 goto out;
 545
 546         map = map_pid_to_cmdline[pid];
 547         if (map >= SAVED_CMDLINES)
 548                 goto out;
 549
 550         cmdline = saved_cmdlines[map];
 551
 552  out:
 553         return cmdline;
 554 }
 555
 556 void tracing_record_cmdline(struct task_struct *tsk)
 557 {
 558         if (atomic_read(&trace_record_cmdline_disabled))
 559                 return;
 560
 561         trace_save_cmdline(tsk);
 562 }
 563
 564 static inline struct list_head *
 565 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 566 {
 567         /*
 568          * Roundrobin - but skip the head (which is not a real page):
 569          */
 570         next = next->next;
 571         if (unlikely(next == &data->trace_pages))
 572                 next = next->next;
 573         BUG_ON(next == &data->trace_pages);
 574
 575         return next;
 576 }
 577
 578 static inline void *
 579 trace_next_page(struct trace_array_cpu *data, void *addr)
 580 {
 581         struct list_head *next;
 582         struct page *page;
 583
 584         page = virt_to_page(addr);
 585
 586         next = trace_next_list(data, &page->lru);
 587         page = list_entry(next, struct page, lru);
 588
 589         return page_address(page);
 590 }
 591
 592 static inline struct trace_entry *
 593 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 594 {
 595         unsigned long idx, idx_next;
 596         struct trace_entry *entry;
 597
 598         data->trace_idx++;
 599         idx = data->trace_head_idx;
 600         idx_next = idx + 1;
 601
 602         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 603
 604         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 605
 606         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 607                 data->trace_head = trace_next_page(data, data->trace_head);
 608                 idx_next = 0;
 609         }
 610
 611         if (data->trace_head == data->trace_tail &&
 612             idx_next == data->trace_tail_idx) {
 613                 /* overrun */
 614                 data->trace_tail_idx++;
 615                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 616                         data->trace_tail =
 617                                 trace_next_page(data, data->trace_tail);
 618                         data->trace_tail_idx = 0;
 619                 }
 620         }
 621
 622         data->trace_head_idx = idx_next;
 623
 624         return entry;
 625 }
 626
 627 static inline void
 628 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 629 {
 630         struct task_struct *tsk = current;
 631         unsigned long pc;
 632
 633         pc = preempt_count();
 634
 635         entry->preempt_count    = pc & 0xff;
 636         entry->pid              = tsk->pid;
 637         entry->t                = ftrace_now(raw_smp_processor_id());
 638         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 639                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 640                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 641                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 642 }
 643
 644 void
 645 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 646                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 647 {
 648         struct trace_entry *entry;
 649         unsigned long irq_flags;
 650
 651         spin_lock_irqsave(&data->lock, irq_flags);
 652         entry                   = tracing_get_trace_entry(tr, data);
 653         tracing_generic_entry_update(entry, flags);
 654         entry->type             = TRACE_FN;
 655         entry->fn.ip            = ip;
 656         entry->fn.parent_ip     = parent_ip;
 657         spin_unlock_irqrestore(&data->lock, irq_flags);
 658
 659         if (!(trace_flags & TRACE_ITER_BLOCK))
 660                 wake_up (&trace_wait);
 661 }
 662
 663 void
 664 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 665        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 666 {
 667         if (likely(!atomic_read(&data->disabled)))
 668                 trace_function(tr, data, ip, parent_ip, flags);
 669 }
 670
 671 void
 672 trace_special(struct trace_array *tr, struct trace_array_cpu *data,
 673               unsigned long arg1, unsigned long arg2, unsigned long arg3)
 674 {
 675         struct trace_entry *entry;
 676         unsigned long irq_flags;
 677
 678         spin_lock_irqsave(&data->lock, irq_flags);
 679         entry                   = tracing_get_trace_entry(tr, data);
 680         tracing_generic_entry_update(entry, 0);
 681         entry->type             = TRACE_SPECIAL;
 682         entry->special.arg1     = arg1;
 683         entry->special.arg2     = arg2;
 684         entry->special.arg3     = arg3;
 685         spin_unlock_irqrestore(&data->lock, irq_flags);
 686
 687         if (!(trace_flags & TRACE_ITER_BLOCK))
 688                 wake_up (&trace_wait);
 689 }
 690
 691 void
 692 tracing_sched_switch_trace(struct trace_array *tr,
 693                            struct trace_array_cpu *data,
 694                            struct task_struct *prev, struct task_struct *next,
 695                            unsigned long flags)
 696 {
 697         struct trace_entry *entry;
 698         unsigned long irq_flags;
 699
 700         spin_lock_irqsave(&data->lock, irq_flags);
 701         entry                   = tracing_get_trace_entry(tr, data);
 702         tracing_generic_entry_update(entry, flags);
 703         entry->type             = TRACE_CTX;
 704         entry->ctx.prev_pid     = prev->pid;
 705         entry->ctx.prev_prio    = prev->prio;
 706         entry->ctx.prev_state   = prev->state;
 707         entry->ctx.next_pid     = next->pid;
 708         entry->ctx.next_prio    = next->prio;
 709         spin_unlock_irqrestore(&data->lock, irq_flags);
 710
 711         if (!(trace_flags & TRACE_ITER_BLOCK))
 712                 wake_up (&trace_wait);
 713 }
 714
 715 void
 716 tracing_sched_wakeup_trace(struct trace_array *tr,
 717                            struct trace_array_cpu *data,
 718                            struct task_struct *wakee, struct task_struct *curr,
 719                            unsigned long flags)
 720 {
 721         struct trace_entry *entry;
 722         unsigned long irq_flags;
 723
 724         spin_lock_irqsave(&data->lock, irq_flags);
 725         entry                   = tracing_get_trace_entry(tr, data);
 726         tracing_generic_entry_update(entry, flags);
 727         entry->type             = TRACE_WAKE;
 728         entry->ctx.prev_pid     = curr->pid;
 729         entry->ctx.prev_prio    = curr->prio;
 730         entry->ctx.prev_state   = curr->state;
 731         entry->ctx.next_pid     = wakee->pid;
 732         entry->ctx.next_prio    = wakee->prio;
 733         spin_unlock_irqrestore(&data->lock, irq_flags);
 734
 735         if (!(trace_flags & TRACE_ITER_BLOCK))
 736                 wake_up(&trace_wait);
 737 }
 738
 739 #ifdef CONFIG_FTRACE
 740 static void
 741 function_trace_call(unsigned long ip, unsigned long parent_ip)
 742 {
 743         struct trace_array *tr = &global_trace;
 744         struct trace_array_cpu *data;
 745         unsigned long flags;
 746         long disabled;
 747         int cpu;
 748
 749         if (unlikely(!tracer_enabled))
 750                 return;
 751
 752         local_irq_save(flags);
 753         cpu = raw_smp_processor_id();
 754         data = tr->data[cpu];
 755         disabled = atomic_inc_return(&data->disabled);
 756
 757         if (likely(disabled == 1))
 758                 trace_function(tr, data, ip, parent_ip, flags);
 759
 760         atomic_dec(&data->disabled);
 761         local_irq_restore(flags);
 762 }
 763
 764 static struct ftrace_ops trace_ops __read_mostly =
 765 {
 766         .func = function_trace_call,
 767 };
 768
 769 void tracing_start_function_trace(void)
 770 {
 771         register_ftrace_function(&trace_ops);
 772 }
 773
 774 void tracing_stop_function_trace(void)
 775 {
 776         unregister_ftrace_function(&trace_ops);
 777 }
 778 #endif
 779
 780 enum trace_file_type {
 781         TRACE_FILE_LAT_FMT      = 1,
 782 };
 783
 784 static struct trace_entry *
 785 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 786                 struct trace_iterator *iter, int cpu)
 787 {
 788         struct page *page;
 789         struct trace_entry *array;
 790
 791         if (iter->next_idx[cpu] >= tr->entries ||
 792             iter->next_idx[cpu] >= data->trace_idx ||
 793             (data->trace_head == data->trace_tail &&
 794              data->trace_head_idx == data->trace_tail_idx))
 795                 return NULL;
 796
 797         if (!iter->next_page[cpu]) {
 798                 /* Initialize the iterator for this cpu trace buffer */
 799                 WARN_ON(!data->trace_tail);
 800                 page = virt_to_page(data->trace_tail);
 801                 iter->next_page[cpu] = &page->lru;
 802                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 803         }
 804
 805         page = list_entry(iter->next_page[cpu], struct page, lru);
 806         BUG_ON(&data->trace_pages == &page->lru);
 807
 808         array = page_address(page);
 809
 810         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 811         return &array[iter->next_page_idx[cpu]];
 812 }
 813
 814 static struct trace_entry *
 815 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 816 {
 817         struct trace_array *tr = iter->tr;
 818         struct trace_entry *ent, *next = NULL;
 819         int next_cpu = -1;
 820         int cpu;
 821
 822         for_each_possible_cpu(cpu) {
 823                 if (!head_page(tr->data[cpu]))
 824                         continue;
 825                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 826                 /*
 827                  * Pick the entry with the smallest timestamp:
 828                  */
 829                 if (ent && (!next || ent->t < next->t)) {
 830                         next = ent;
 831                         next_cpu = cpu;
 832                 }
 833         }
 834
 835         if (ent_cpu)
 836                 *ent_cpu = next_cpu;
 837
 838         return next;
 839 }
 840
 841 static void trace_iterator_increment(struct trace_iterator *iter)
 842 {
 843         iter->idx++;
 844         iter->next_idx[iter->cpu]++;
 845         iter->next_page_idx[iter->cpu]++;
 846
 847         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 848                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 849
 850                 iter->next_page_idx[iter->cpu] = 0;
 851                 iter->next_page[iter->cpu] =
 852                         trace_next_list(data, iter->next_page[iter->cpu]);
 853         }
 854 }
 855
 856 static void trace_consume(struct trace_iterator *iter)
 857 {
 858         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 859
 860         data->trace_tail_idx++;
 861         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 862                 data->trace_tail = trace_next_page(data, data->trace_tail);
 863                 data->trace_tail_idx = 0;
 864         }
 865
 866         /* Check if we empty it, then reset the index */
 867         if (data->trace_head == data->trace_tail &&
 868             data->trace_head_idx == data->trace_tail_idx)
 869                 data->trace_idx = 0;
 870 }
 871
 872 static void *find_next_entry_inc(struct trace_iterator *iter)
 873 {
 874         struct trace_entry *next;
 875         int next_cpu = -1;
 876
 877         next = find_next_entry(iter, &next_cpu);
 878
 879         iter->prev_ent = iter->ent;
 880         iter->prev_cpu = iter->cpu;
 881
 882         iter->ent = next;
 883         iter->cpu = next_cpu;
 884
 885         if (next)
 886                 trace_iterator_increment(iter);
 887
 888         return next ? iter : NULL;
 889 }
 890
 891 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 892 {
 893         struct trace_iterator *iter = m->private;
 894         void *last_ent = iter->ent;
 895         int i = (int)*pos;
 896         void *ent;
 897
 898         (*pos)++;
 899
 900         /* can't go backwards */
 901         if (iter->idx > i)
 902                 return NULL;
 903
 904         if (iter->idx < 0)
 905                 ent = find_next_entry_inc(iter);
 906         else
 907                 ent = iter;
 908
 909         while (ent && iter->idx < i)
 910                 ent = find_next_entry_inc(iter);
 911
 912         iter->pos = *pos;
 913
 914         if (last_ent && !ent)
 915                 seq_puts(m, "\n\nvim:ft=help\n");
 916
 917         return ent;
 918 }
 919
 920 static void *s_start(struct seq_file *m, loff_t *pos)
 921 {
 922         struct trace_iterator *iter = m->private;
 923         void *p = NULL;
 924         loff_t l = 0;
 925         int i;
 926
 927         mutex_lock(&trace_types_lock);
 928
 929         if (!current_trace || current_trace != iter->trace)
 930                 return NULL;
 931
 932         atomic_inc(&trace_record_cmdline_disabled);
 933
 934         /* let the tracer grab locks here if needed */
 935         if (current_trace->start)
 936                 current_trace->start(iter);
 937
 938         if (*pos != iter->pos) {
 939                 iter->ent = NULL;
 940                 iter->cpu = 0;
 941                 iter->idx = -1;
 942                 iter->prev_ent = NULL;
 943                 iter->prev_cpu = -1;
 944
 945                 for_each_possible_cpu(i) {
 946                         iter->next_idx[i] = 0;
 947                         iter->next_page[i] = NULL;
 948                 }
 949
 950                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 951                         ;
 952
 953         } else {
 954                 l = *pos - 1;
 955                 p = s_next(m, p, &l);
 956         }
 957
 958         return p;
 959 }
 960
 961 static void s_stop(struct seq_file *m, void *p)
 962 {
 963         struct trace_iterator *iter = m->private;
 964
 965         atomic_dec(&trace_record_cmdline_disabled);
 966
 967         /* let the tracer release locks here if needed */
 968         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 969                 iter->trace->stop(iter);
 970
 971         mutex_unlock(&trace_types_lock);
 972 }
 973
 974 static int
 975 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 976 {
 977 #ifdef CONFIG_KALLSYMS
 978         char str[KSYM_SYMBOL_LEN];
 979
 980         kallsyms_lookup(address, NULL, NULL, NULL, str);
 981
 982         return trace_seq_printf(s, fmt, str);
 983 #endif
 984         return 1;
 985 }
 986
 987 static int
 988 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 989                      unsigned long address)
 990 {
 991 #ifdef CONFIG_KALLSYMS
 992         char str[KSYM_SYMBOL_LEN];
 993
 994         sprint_symbol(str, address);
 995         return trace_seq_printf(s, fmt, str);
 996 #endif
 997         return 1;
 998 }
 999
1000 #ifndef CONFIG_64BIT
1001 # define IP_FMT "%08lx"
1002 #else
1003 # define IP_FMT "%016lx"
1004 #endif
1005
1006 static int
1007 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1008 {
1009         int ret;
1010
1011         if (!ip)
1012                 return trace_seq_printf(s, "0");
1013
1014         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1015                 ret = seq_print_sym_offset(s, "%s", ip);
1016         else
1017                 ret = seq_print_sym_short(s, "%s", ip);
1018
1019         if (!ret)
1020                 return 0;
1021
1022         if (sym_flags & TRACE_ITER_SYM_ADDR)
1023                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1024         return ret;
1025 }
1026
1027 static void print_lat_help_header(struct seq_file *m)
1028 {
1029         seq_puts(m, "#                _------=> CPU#            \n");
1030         seq_puts(m, "#               / _-----=> irqs-off        \n");
1031         seq_puts(m, "#              | / _----=> need-resched    \n");
1032         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1033         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1034         seq_puts(m, "#              |||| /                      \n");
1035         seq_puts(m, "#              |||||     delay             \n");
1036         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1037         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1038 }
1039
1040 static void print_func_help_header(struct seq_file *m)
1041 {
1042         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1043         seq_puts(m, "#              | |      |          |         |\n");
1044 }
1045
1046
1047 static void
1048 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1049 {
1050         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1051         struct trace_array *tr = iter->tr;
1052         struct trace_array_cpu *data = tr->data[tr->cpu];
1053         struct tracer *type = current_trace;
1054         unsigned long total   = 0;
1055         unsigned long entries = 0;
1056         int cpu;
1057         const char *name = "preemption";
1058
1059         if (type)
1060                 name = type->name;
1061
1062         for_each_possible_cpu(cpu) {
1063                 if (head_page(tr->data[cpu])) {
1064                         total += tr->data[cpu]->trace_idx;
1065                         if (tr->data[cpu]->trace_idx > tr->entries)
1066                                 entries += tr->entries;
1067                         else
1068                                 entries += tr->data[cpu]->trace_idx;
1069                 }
1070         }
1071
1072         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1073                    name, UTS_RELEASE);
1074         seq_puts(m, "-----------------------------------"
1075                  "---------------------------------\n");
1076         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1077                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1078                    nsecs_to_usecs(data->saved_latency),
1079                    entries,
1080                    total,
1081                    tr->cpu,
1082 #if defined(CONFIG_PREEMPT_NONE)
1083                    "server",
1084 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1085                    "desktop",
1086 #elif defined(CONFIG_PREEMPT_DESKTOP)
1087                    "preempt",
1088 #else
1089                    "unknown",
1090 #endif
1091                    /* These are reserved for later use */
1092                    0, 0, 0, 0);
1093 #ifdef CONFIG_SMP
1094         seq_printf(m, " #P:%d)\n", num_online_cpus());
1095 #else
1096         seq_puts(m, ")\n");
1097 #endif
1098         seq_puts(m, "    -----------------\n");
1099         seq_printf(m, "    | task: %.16s-%d "
1100                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1101                    data->comm, data->pid, data->uid, data->nice,
1102                    data->policy, data->rt_priority);
1103         seq_puts(m, "    -----------------\n");
1104
1105         if (data->critical_start) {
1106                 seq_puts(m, " => started at: ");
1107                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1108                 trace_print_seq(m, &iter->seq);
1109                 seq_puts(m, "\n => ended at:   ");
1110                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1111                 trace_print_seq(m, &iter->seq);
1112                 seq_puts(m, "\n");
1113         }
1114
1115         seq_puts(m, "\n");
1116 }
1117
1118 static void
1119 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1120 {
1121         int hardirq, softirq;
1122         char *comm;
1123
1124         comm = trace_find_cmdline(entry->pid);
1125
1126         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1127         trace_seq_printf(s, "%d", cpu);
1128         trace_seq_printf(s, "%c%c",
1129                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1130                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1131
1132         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1133         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1134         if (hardirq && softirq)
1135                 trace_seq_putc(s, 'H');
1136         else {
1137                 if (hardirq)
1138                         trace_seq_putc(s, 'h');
1139                 else {
1140                         if (softirq)
1141                                 trace_seq_putc(s, 's');
1142                         else
1143                                 trace_seq_putc(s, '.');
1144                 }
1145         }
1146
1147         if (entry->preempt_count)
1148                 trace_seq_printf(s, "%x", entry->preempt_count);
1149         else
1150                 trace_seq_puts(s, ".");
1151 }
1152
1153 unsigned long preempt_mark_thresh = 100;
1154
1155 static void
1156 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1157                     unsigned long rel_usecs)
1158 {
1159         trace_seq_printf(s, " %4lldus", abs_usecs);
1160         if (rel_usecs > preempt_mark_thresh)
1161                 trace_seq_puts(s, "!: ");
1162         else if (rel_usecs > 1)
1163                 trace_seq_puts(s, "+: ");
1164         else
1165                 trace_seq_puts(s, " : ");
1166 }
1167
1168 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1169
1170 static int
1171 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1172 {
1173         struct trace_seq *s = &iter->seq;
1174         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1175         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1176         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1177         struct trace_entry *entry = iter->ent;
1178         unsigned long abs_usecs;
1179         unsigned long rel_usecs;
1180         char *comm;
1181         int S;
1182
1183         if (!next_entry)
1184                 next_entry = entry;
1185         rel_usecs = ns2usecs(next_entry->t - entry->t);
1186         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1187
1188         if (verbose) {
1189                 comm = trace_find_cmdline(entry->pid);
1190                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1191                                  " %ld.%03ldms (+%ld.%03ldms): ",
1192                                  comm,
1193                                  entry->pid, cpu, entry->flags,
1194                                  entry->preempt_count, trace_idx,
1195                                  ns2usecs(entry->t),
1196                                  abs_usecs/1000,
1197                                  abs_usecs % 1000, rel_usecs/1000,
1198                                  rel_usecs % 1000);
1199         } else {
1200                 lat_print_generic(s, entry, cpu);
1201                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1202         }
1203         switch (entry->type) {
1204         case TRACE_FN:
1205                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1206                 trace_seq_puts(s, " (");
1207                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1208                 trace_seq_puts(s, ")\n");
1209                 break;
1210         case TRACE_CTX:
1211         case TRACE_WAKE:
1212                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1213                         state_to_char[entry->ctx.prev_state] : 'X';
1214                 comm = trace_find_cmdline(entry->ctx.next_pid);
1215                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d %s\n",
1216                                  entry->ctx.prev_pid,
1217                                  entry->ctx.prev_prio,
1218                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1219                                  entry->ctx.next_pid,
1220                                  entry->ctx.next_prio,
1221                                  comm);
1222                 break;
1223         case TRACE_SPECIAL:
1224                 trace_seq_printf(s, " %lx %lx %lx\n",
1225                                  entry->special.arg1,
1226                                  entry->special.arg2,
1227                                  entry->special.arg3);
1228                 break;
1229         default:
1230                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1231         }
1232         return 1;
1233 }
1234
1235 static int print_trace_fmt(struct trace_iterator *iter)
1236 {
1237         struct trace_seq *s = &iter->seq;
1238         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1239         struct trace_entry *entry;
1240         unsigned long usec_rem;
1241         unsigned long long t;
1242         unsigned long secs;
1243         char *comm;
1244         int S;
1245         int ret;
1246
1247         entry = iter->ent;
1248
1249         comm = trace_find_cmdline(iter->ent->pid);
1250
1251         t = ns2usecs(entry->t);
1252         usec_rem = do_div(t, 1000000ULL);
1253         secs = (unsigned long)t;
1254
1255         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1256         if (!ret)
1257                 return 0;
1258         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1259         if (!ret)
1260                 return 0;
1261         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1262         if (!ret)
1263                 return 0;
1264
1265         switch (entry->type) {
1266         case TRACE_FN:
1267                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1268                 if (!ret)
1269                         return 0;
1270                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1271                                                 entry->fn.parent_ip) {
1272                         ret = trace_seq_printf(s, " <-");
1273                         if (!ret)
1274                                 return 0;
1275                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1276                                                sym_flags);
1277                         if (!ret)
1278                                 return 0;
1279                 }
1280                 ret = trace_seq_printf(s, "\n");
1281                 if (!ret)
1282                         return 0;
1283                 break;
1284         case TRACE_CTX:
1285         case TRACE_WAKE:
1286                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1287                         state_to_char[entry->ctx.prev_state] : 'X';
1288                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d\n",
1289                                        entry->ctx.prev_pid,
1290                                        entry->ctx.prev_prio,
1291                                        S,
1292                                        entry->type == TRACE_CTX ? "==>" : "  +",
1293                                        entry->ctx.next_pid,
1294                                        entry->ctx.next_prio);
1295                 if (!ret)
1296                         return 0;
1297                 break;
1298         case TRACE_SPECIAL:
1299                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1300                                  entry->special.arg1,
1301                                  entry->special.arg2,
1302                                  entry->special.arg3);
1303                 if (!ret)
1304                         return 0;
1305                 break;
1306         }
1307         return 1;
1308 }
1309
1310 static int print_raw_fmt(struct trace_iterator *iter)
1311 {
1312         struct trace_seq *s = &iter->seq;
1313         struct trace_entry *entry;
1314         int ret;
1315         int S;
1316
1317         entry = iter->ent;
1318
1319         ret = trace_seq_printf(s, "%d %d %llu ",
1320                 entry->pid, iter->cpu, entry->t);
1321         if (!ret)
1322                 return 0;
1323
1324         switch (entry->type) {
1325         case TRACE_FN:
1326                 ret = trace_seq_printf(s, "%x %x\n",
1327                                         entry->fn.ip, entry->fn.parent_ip);
1328                 if (!ret)
1329                         return 0;
1330                 break;
1331         case TRACE_CTX:
1332         case TRACE_WAKE:
1333                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1334                         state_to_char[entry->ctx.prev_state] : 'X';
1335                 if (entry->type == TRACE_WAKE)
1336                         S = '+';
1337                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1338                                        entry->ctx.prev_pid,
1339                                        entry->ctx.prev_prio,
1340                                        S,
1341                                        entry->ctx.next_pid,
1342                                        entry->ctx.next_prio);
1343                 if (!ret)
1344                         return 0;
1345                 break;
1346         case TRACE_SPECIAL:
1347                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1348                                  entry->special.arg1,
1349                                  entry->special.arg2,
1350                                  entry->special.arg3);
1351                 if (!ret)
1352                         return 0;
1353                 break;
1354         }
1355         return 1;
1356 }
1357
1358 #define SEQ_PUT_FIELD_RET(s, x)                         \
1359 do {                                                    \
1360         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1361                 return 0;                               \
1362 } while (0)
1363
1364 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1365 do {                                                    \
1366         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1367                 return 0;                               \
1368 } while (0)
1369
1370 static int print_hex_fmt(struct trace_iterator *iter)
1371 {
1372         struct trace_seq *s = &iter->seq;
1373         unsigned char newline = '\n';
1374         struct trace_entry *entry;
1375         int S;
1376
1377         entry = iter->ent;
1378
1379         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1380         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1381         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1382
1383         switch (entry->type) {
1384         case TRACE_FN:
1385                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1386                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1387                 break;
1388         case TRACE_CTX:
1389         case TRACE_WAKE:
1390                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1391                         state_to_char[entry->ctx.prev_state] : 'X';
1392                 if (entry->type == TRACE_WAKE)
1393                         S = '+';
1394                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1395                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1396                 SEQ_PUT_HEX_FIELD_RET(s, S);
1397                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1398                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1399                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1400                 break;
1401         case TRACE_SPECIAL:
1402                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1403                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1404                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1405                 break;
1406         }
1407         SEQ_PUT_FIELD_RET(s, newline);
1408
1409         return 1;
1410 }
1411
1412 static int print_bin_fmt(struct trace_iterator *iter)
1413 {
1414         struct trace_seq *s = &iter->seq;
1415         struct trace_entry *entry;
1416
1417         entry = iter->ent;
1418
1419         SEQ_PUT_FIELD_RET(s, entry->pid);
1420         SEQ_PUT_FIELD_RET(s, entry->cpu);
1421         SEQ_PUT_FIELD_RET(s, entry->t);
1422
1423         switch (entry->type) {
1424         case TRACE_FN:
1425                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1426                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1427                 break;
1428         case TRACE_CTX:
1429                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1430                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1431                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1432                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1433                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1434                 break;
1435         case TRACE_SPECIAL:
1436                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1437                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1438                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1439                 break;
1440         }
1441         return 1;
1442 }
1443
1444 static int trace_empty(struct trace_iterator *iter)
1445 {
1446         struct trace_array_cpu *data;
1447         int cpu;
1448
1449         for_each_possible_cpu(cpu) {
1450                 data = iter->tr->data[cpu];
1451
1452                 if (head_page(data) && data->trace_idx &&
1453                     (data->trace_tail != data->trace_head ||
1454                      data->trace_tail_idx != data->trace_head_idx))
1455                         return 0;
1456         }
1457         return 1;
1458 }
1459
1460 static int print_trace_line(struct trace_iterator *iter)
1461 {
1462         if (trace_flags & TRACE_ITER_BIN)
1463                 return print_bin_fmt(iter);
1464
1465         if (trace_flags & TRACE_ITER_HEX)
1466                 return print_hex_fmt(iter);
1467
1468         if (trace_flags & TRACE_ITER_RAW)
1469                 return print_raw_fmt(iter);
1470
1471         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1472                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1473
1474         return print_trace_fmt(iter);
1475 }
1476
1477 static int s_show(struct seq_file *m, void *v)
1478 {
1479         struct trace_iterator *iter = v;
1480
1481         if (iter->ent == NULL) {
1482                 if (iter->tr) {
1483                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1484                         seq_puts(m, "#\n");
1485                 }
1486                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1487                         /* print nothing if the buffers are empty */
1488                         if (trace_empty(iter))
1489                                 return 0;
1490                         print_trace_header(m, iter);
1491                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1492                                 print_lat_help_header(m);
1493                 } else {
1494                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1495                                 print_func_help_header(m);
1496                 }
1497         } else {
1498                 print_trace_line(iter);
1499                 trace_print_seq(m, &iter->seq);
1500         }
1501
1502         return 0;
1503 }
1504
1505 static struct seq_operations tracer_seq_ops = {
1506         .start          = s_start,
1507         .next           = s_next,
1508         .stop           = s_stop,
1509         .show           = s_show,
1510 };
1511
1512 static struct trace_iterator *
1513 __tracing_open(struct inode *inode, struct file *file, int *ret)
1514 {
1515         struct trace_iterator *iter;
1516
1517         if (tracing_disabled) {
1518                 *ret = -ENODEV;
1519                 return NULL;
1520         }
1521
1522         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1523         if (!iter) {
1524                 *ret = -ENOMEM;
1525                 goto out;
1526         }
1527
1528         mutex_lock(&trace_types_lock);
1529         if (current_trace && current_trace->print_max)
1530                 iter->tr = &max_tr;
1531         else
1532                 iter->tr = inode->i_private;
1533         iter->trace = current_trace;
1534         iter->pos = -1;
1535
1536         /* TODO stop tracer */
1537         *ret = seq_open(file, &tracer_seq_ops);
1538         if (!*ret) {
1539                 struct seq_file *m = file->private_data;
1540                 m->private = iter;
1541
1542                 /* stop the trace while dumping */
1543                 if (iter->tr->ctrl)
1544                         tracer_enabled = 0;
1545
1546                 if (iter->trace && iter->trace->open)
1547                         iter->trace->open(iter);
1548         } else {
1549                 kfree(iter);
1550                 iter = NULL;
1551         }
1552         mutex_unlock(&trace_types_lock);
1553
1554  out:
1555         return iter;
1556 }
1557
1558 int tracing_open_generic(struct inode *inode, struct file *filp)
1559 {
1560         if (tracing_disabled)
1561                 return -ENODEV;
1562
1563         filp->private_data = inode->i_private;
1564         return 0;
1565 }
1566
1567 int tracing_release(struct inode *inode, struct file *file)
1568 {
1569         struct seq_file *m = (struct seq_file *)file->private_data;
1570         struct trace_iterator *iter = m->private;
1571
1572         mutex_lock(&trace_types_lock);
1573         if (iter->trace && iter->trace->close)
1574                 iter->trace->close(iter);
1575
1576         /* reenable tracing if it was previously enabled */
1577         if (iter->tr->ctrl)
1578                 tracer_enabled = 1;
1579         mutex_unlock(&trace_types_lock);
1580
1581         seq_release(inode, file);
1582         kfree(iter);
1583         return 0;
1584 }
1585
1586 static int tracing_open(struct inode *inode, struct file *file)
1587 {
1588         int ret;
1589
1590         __tracing_open(inode, file, &ret);
1591
1592         return ret;
1593 }
1594
1595 static int tracing_lt_open(struct inode *inode, struct file *file)
1596 {
1597         struct trace_iterator *iter;
1598         int ret;
1599
1600         iter = __tracing_open(inode, file, &ret);
1601
1602         if (!ret)
1603                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1604
1605         return ret;
1606 }
1607
1608
1609 static void *
1610 t_next(struct seq_file *m, void *v, loff_t *pos)
1611 {
1612         struct tracer *t = m->private;
1613
1614         (*pos)++;
1615
1616         if (t)
1617                 t = t->next;
1618
1619         m->private = t;
1620
1621         return t;
1622 }
1623
1624 static void *t_start(struct seq_file *m, loff_t *pos)
1625 {
1626         struct tracer *t = m->private;
1627         loff_t l = 0;
1628
1629         mutex_lock(&trace_types_lock);
1630         for (; t && l < *pos; t = t_next(m, t, &l))
1631                 ;
1632
1633         return t;
1634 }
1635
1636 static void t_stop(struct seq_file *m, void *p)
1637 {
1638         mutex_unlock(&trace_types_lock);
1639 }
1640
1641 static int t_show(struct seq_file *m, void *v)
1642 {
1643         struct tracer *t = v;
1644
1645         if (!t)
1646                 return 0;
1647
1648         seq_printf(m, "%s", t->name);
1649         if (t->next)
1650                 seq_putc(m, ' ');
1651         else
1652                 seq_putc(m, '\n');
1653
1654         return 0;
1655 }
1656
1657 static struct seq_operations show_traces_seq_ops = {
1658         .start          = t_start,
1659         .next           = t_next,
1660         .stop           = t_stop,
1661         .show           = t_show,
1662 };
1663
1664 static int show_traces_open(struct inode *inode, struct file *file)
1665 {
1666         int ret;
1667
1668         if (tracing_disabled)
1669                 return -ENODEV;
1670
1671         ret = seq_open(file, &show_traces_seq_ops);
1672         if (!ret) {
1673                 struct seq_file *m = file->private_data;
1674                 m->private = trace_types;
1675         }
1676
1677         return ret;
1678 }
1679
1680 static struct file_operations tracing_fops = {
1681         .open           = tracing_open,
1682         .read           = seq_read,
1683         .llseek         = seq_lseek,
1684         .release        = tracing_release,
1685 };
1686
1687 static struct file_operations tracing_lt_fops = {
1688         .open           = tracing_lt_open,
1689         .read           = seq_read,
1690         .llseek         = seq_lseek,
1691         .release        = tracing_release,
1692 };
1693
1694 static struct file_operations show_traces_fops = {
1695         .open = show_traces_open,
1696         .read = seq_read,
1697         .release = seq_release,
1698 };
1699
1700 static ssize_t
1701 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1702                        size_t cnt, loff_t *ppos)
1703 {
1704         char *buf;
1705         int r = 0;
1706         int len = 0;
1707         int i;
1708
1709         /* calulate max size */
1710         for (i = 0; trace_options[i]; i++) {
1711                 len += strlen(trace_options[i]);
1712                 len += 3; /* "no" and space */
1713         }
1714
1715         /* +2 for \n and \0 */
1716         buf = kmalloc(len + 2, GFP_KERNEL);
1717         if (!buf)
1718                 return -ENOMEM;
1719
1720         for (i = 0; trace_options[i]; i++) {
1721                 if (trace_flags & (1 << i))
1722                         r += sprintf(buf + r, "%s ", trace_options[i]);
1723                 else
1724                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1725         }
1726
1727         r += sprintf(buf + r, "\n");
1728         WARN_ON(r >= len + 2);
1729
1730         r = simple_read_from_buffer(ubuf, cnt, ppos,
1731                                     buf, r);
1732
1733         kfree(buf);
1734
1735         return r;
1736 }
1737
1738 static ssize_t
1739 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1740                         size_t cnt, loff_t *ppos)
1741 {
1742         char buf[64];
1743         char *cmp = buf;
1744         int neg = 0;
1745         int i;
1746
1747         if (cnt > 63)
1748                 cnt = 63;
1749
1750         if (copy_from_user(&buf, ubuf, cnt))
1751                 return -EFAULT;
1752
1753         buf[cnt] = 0;
1754
1755         if (strncmp(buf, "no", 2) == 0) {
1756                 neg = 1;
1757                 cmp += 2;
1758         }
1759
1760         for (i = 0; trace_options[i]; i++) {
1761                 int len = strlen(trace_options[i]);
1762
1763                 if (strncmp(cmp, trace_options[i], len) == 0) {
1764                         if (neg)
1765                                 trace_flags &= ~(1 << i);
1766                         else
1767                                 trace_flags |= (1 << i);
1768                         break;
1769                 }
1770         }
1771
1772         filp->f_pos += cnt;
1773
1774         return cnt;
1775 }
1776
1777 static struct file_operations tracing_iter_fops = {
1778         .open = tracing_open_generic,
1779         .read = tracing_iter_ctrl_read,
1780         .write = tracing_iter_ctrl_write,
1781 };
1782
1783 static const char readme_msg[] =
1784         "tracing mini-HOWTO:\n\n"
1785         "# mkdir /debug\n"
1786         "# mount -t debugfs nodev /debug\n\n"
1787         "# cat /debug/tracing/available_tracers\n"
1788         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1789         "# cat /debug/tracing/current_tracer\n"
1790         "none\n"
1791         "# echo sched_switch > /debug/tracing/current_tracer\n"
1792         "# cat /debug/tracing/current_tracer\n"
1793         "sched_switch\n"
1794         "# cat /debug/tracing/iter_ctrl\n"
1795         "noprint-parent nosym-offset nosym-addr noverbose\n"
1796         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1797         "# echo 1 > /debug/tracing/tracing_enabled\n"
1798         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1799         "echo 0 > /debug/tracing/tracing_enabled\n"
1800 ;
1801
1802 static ssize_t
1803 tracing_readme_read(struct file *filp, char __user *ubuf,
1804                        size_t cnt, loff_t *ppos)
1805 {
1806         return simple_read_from_buffer(ubuf, cnt, ppos,
1807                                         readme_msg, strlen(readme_msg));
1808 }
1809
1810 static struct file_operations tracing_readme_fops = {
1811         .open = tracing_open_generic,
1812         .read = tracing_readme_read,
1813 };
1814
1815 static ssize_t
1816 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1817                   size_t cnt, loff_t *ppos)
1818 {
1819         struct trace_array *tr = filp->private_data;
1820         char buf[64];
1821         int r;
1822
1823         r = sprintf(buf, "%ld\n", tr->ctrl);
1824         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1825 }
1826
1827 static ssize_t
1828 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1829                    size_t cnt, loff_t *ppos)
1830 {
1831         struct trace_array *tr = filp->private_data;
1832         long val;
1833         char buf[64];
1834
1835         if (cnt > 63)
1836                 cnt = 63;
1837
1838         if (copy_from_user(&buf, ubuf, cnt))
1839                 return -EFAULT;
1840
1841         buf[cnt] = 0;
1842
1843         val = simple_strtoul(buf, NULL, 10);
1844
1845         val = !!val;
1846
1847         mutex_lock(&trace_types_lock);
1848         if (tr->ctrl ^ val) {
1849                 if (val)
1850                         tracer_enabled = 1;
1851                 else
1852                         tracer_enabled = 0;
1853
1854                 tr->ctrl = val;
1855
1856                 if (current_trace && current_trace->ctrl_update)
1857                         current_trace->ctrl_update(tr);
1858         }
1859         mutex_unlock(&trace_types_lock);
1860
1861         filp->f_pos += cnt;
1862
1863         return cnt;
1864 }
1865
1866 static ssize_t
1867 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1868                        size_t cnt, loff_t *ppos)
1869 {
1870         char buf[max_tracer_type_len+2];
1871         int r;
1872
1873         mutex_lock(&trace_types_lock);
1874         if (current_trace)
1875                 r = sprintf(buf, "%s\n", current_trace->name);
1876         else
1877                 r = sprintf(buf, "\n");
1878         mutex_unlock(&trace_types_lock);
1879
1880         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1881 }
1882
1883 static ssize_t
1884 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1885                         size_t cnt, loff_t *ppos)
1886 {
1887         struct trace_array *tr = &global_trace;
1888         struct tracer *t;
1889         char buf[max_tracer_type_len+1];
1890         int i;
1891
1892         if (cnt > max_tracer_type_len)
1893                 cnt = max_tracer_type_len;
1894
1895         if (copy_from_user(&buf, ubuf, cnt))
1896                 return -EFAULT;
1897
1898         buf[cnt] = 0;
1899
1900         /* strip ending whitespace. */
1901         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1902                 buf[i] = 0;
1903
1904         mutex_lock(&trace_types_lock);
1905         for (t = trace_types; t; t = t->next) {
1906                 if (strcmp(t->name, buf) == 0)
1907                         break;
1908         }
1909         if (!t || t == current_trace)
1910                 goto out;
1911
1912         if (current_trace && current_trace->reset)
1913                 current_trace->reset(tr);
1914
1915         current_trace = t;
1916         if (t->init)
1917                 t->init(tr);
1918
1919  out:
1920         mutex_unlock(&trace_types_lock);
1921
1922         filp->f_pos += cnt;
1923
1924         return cnt;
1925 }
1926
1927 static ssize_t
1928 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1929                      size_t cnt, loff_t *ppos)
1930 {
1931         unsigned long *ptr = filp->private_data;
1932         char buf[64];
1933         int r;
1934
1935         r = snprintf(buf, 64, "%ld\n",
1936                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1937         if (r > 64)
1938                 r = 64;
1939         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1940 }
1941
1942 static ssize_t
1943 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1944                       size_t cnt, loff_t *ppos)
1945 {
1946         long *ptr = filp->private_data;
1947         long val;
1948         char buf[64];
1949
1950         if (cnt > 63)
1951                 cnt = 63;
1952
1953         if (copy_from_user(&buf, ubuf, cnt))
1954                 return -EFAULT;
1955
1956         buf[cnt] = 0;
1957
1958         val = simple_strtoul(buf, NULL, 10);
1959
1960         *ptr = val * 1000;
1961
1962         return cnt;
1963 }
1964
1965 static atomic_t tracing_reader;
1966
1967 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1968 {
1969         struct trace_iterator *iter;
1970
1971         if (tracing_disabled)
1972                 return -ENODEV;
1973
1974         /* We only allow for reader of the pipe */
1975         if (atomic_inc_return(&tracing_reader) != 1) {
1976                 atomic_dec(&tracing_reader);
1977                 return -EBUSY;
1978         }
1979
1980         /* create a buffer to store the information to pass to userspace */
1981         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1982         if (!iter)
1983                 return -ENOMEM;
1984
1985         iter->tr = &global_trace;
1986
1987         filp->private_data = iter;
1988
1989         return 0;
1990 }
1991
1992 static int tracing_release_pipe(struct inode *inode, struct file *file)
1993 {
1994         struct trace_iterator *iter = file->private_data;
1995
1996         kfree(iter);
1997         atomic_dec(&tracing_reader);
1998
1999         return 0;
2000 }
2001
2002 static unsigned int
2003 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2004 {
2005         struct trace_iterator *iter = filp->private_data;
2006
2007         if (trace_flags & TRACE_ITER_BLOCK) {
2008                 /*
2009                  * Always select as readable when in blocking mode
2010                  */
2011                 return POLLIN | POLLRDNORM;
2012         }
2013         else {
2014                 if (!trace_empty(iter))
2015                         return POLLIN | POLLRDNORM;
2016                 poll_wait(filp, &trace_wait, poll_table);
2017                 if (!trace_empty(iter))
2018                         return POLLIN | POLLRDNORM;
2019
2020                 return 0;
2021         }
2022 }
2023
2024 /*
2025  * Consumer reader.
2026  */
2027 static ssize_t
2028 tracing_read_pipe(struct file *filp, char __user *ubuf,
2029                   size_t cnt, loff_t *ppos)
2030 {
2031         struct trace_iterator *iter = filp->private_data;
2032         struct trace_array_cpu *data;
2033         static cpumask_t mask;
2034         static int start;
2035         unsigned long flags;
2036 #ifdef CONFIG_FTRACE
2037         int ftrace_save;
2038 #endif
2039         int read = 0;
2040         int cpu;
2041         int len;
2042         int ret;
2043
2044         /* return any leftover data */
2045         if (iter->seq.len > start) {
2046                 len = iter->seq.len - start;
2047                 if (cnt > len)
2048                         cnt = len;
2049                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2050                 if (ret)
2051                         cnt = -EFAULT;
2052
2053                 start += len;
2054
2055                 return cnt;
2056         }
2057
2058         trace_seq_reset(&iter->seq);
2059         start = 0;
2060
2061         while (trace_empty(iter)) {
2062                 if (!(trace_flags & TRACE_ITER_BLOCK))
2063                         return -EWOULDBLOCK;
2064                 /*
2065                  * This is a make-shift waitqueue. The reason we don't use
2066                  * an actual wait queue is because:
2067                  *  1) we only ever have one waiter
2068                  *  2) the tracing, traces all functions, we don't want
2069                  *     the overhead of calling wake_up and friends
2070                  *     (and tracing them too)
2071                  *     Anyway, this is really very primitive wakeup.
2072                  */
2073                 set_current_state(TASK_INTERRUPTIBLE);
2074                 iter->tr->waiter = current;
2075
2076                 /* sleep for one second, and try again. */
2077                 schedule_timeout(HZ);
2078
2079                 iter->tr->waiter = NULL;
2080
2081                 if (signal_pending(current))
2082                         return -EINTR;
2083
2084                 /*
2085                  * We block until we read something and tracing is disabled.
2086                  * We still block if tracing is disabled, but we have never
2087                  * read anything. This allows a user to cat this file, and
2088                  * then enable tracing. But after we have read something,
2089                  * we give an EOF when tracing is again disabled.
2090                  *
2091                  * iter->pos will be 0 if we haven't read anything.
2092                  */
2093                 if (!tracer_enabled && iter->pos)
2094                         break;
2095
2096                 continue;
2097         }
2098
2099         /* stop when tracing is finished */
2100         if (trace_empty(iter))
2101                 return 0;
2102
2103         if (cnt >= PAGE_SIZE)
2104                 cnt = PAGE_SIZE - 1;
2105
2106         memset(iter, 0, sizeof(*iter));
2107         iter->tr = &global_trace;
2108         iter->pos = -1;
2109
2110         /*
2111          * We need to stop all tracing on all CPUS to read the
2112          * the next buffer. This is a bit expensive, but is
2113          * not done often. We fill all what we can read,
2114          * and then release the locks again.
2115          */
2116
2117         cpus_clear(mask);
2118         local_irq_save(flags);
2119 #ifdef CONFIG_FTRACE
2120         ftrace_save = ftrace_enabled;
2121         ftrace_enabled = 0;
2122 #endif
2123         smp_wmb();
2124         for_each_possible_cpu(cpu) {
2125                 data = iter->tr->data[cpu];
2126
2127                 if (!head_page(data) || !data->trace_idx)
2128                         continue;
2129
2130                 atomic_inc(&data->disabled);
2131                 cpu_set(cpu, mask);
2132         }
2133
2134         for_each_cpu_mask(cpu, mask) {
2135                 data = iter->tr->data[cpu];
2136                 spin_lock(&data->lock);
2137         }
2138
2139         while (find_next_entry_inc(iter) != NULL) {
2140                 int len = iter->seq.len;
2141
2142                 ret = print_trace_line(iter);
2143                 if (!ret) {
2144                         /* don't print partial lines */
2145                         iter->seq.len = len;
2146                         break;
2147                 }
2148
2149                 trace_consume(iter);
2150
2151                 if (iter->seq.len >= cnt)
2152                         break;
2153         }
2154
2155         for_each_cpu_mask(cpu, mask) {
2156                 data = iter->tr->data[cpu];
2157                 spin_unlock(&data->lock);
2158         }
2159
2160         for_each_cpu_mask(cpu, mask) {
2161                 data = iter->tr->data[cpu];
2162                 atomic_dec(&data->disabled);
2163         }
2164 #ifdef CONFIG_FTRACE
2165         ftrace_enabled = ftrace_save;
2166 #endif
2167         local_irq_restore(flags);
2168
2169         /* Now copy what we have to the user */
2170         read = iter->seq.len;
2171         if (read > cnt)
2172                 read = cnt;
2173
2174         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2175
2176         if (read < iter->seq.len)
2177                 start = read;
2178         else
2179                 trace_seq_reset(&iter->seq);
2180
2181         if (ret)
2182                 read = -EFAULT;
2183
2184         return read;
2185 }
2186
2187 static struct file_operations tracing_max_lat_fops = {
2188         .open           = tracing_open_generic,
2189         .read           = tracing_max_lat_read,
2190         .write          = tracing_max_lat_write,
2191 };
2192
2193 static struct file_operations tracing_ctrl_fops = {
2194         .open           = tracing_open_generic,
2195         .read           = tracing_ctrl_read,
2196         .write          = tracing_ctrl_write,
2197 };
2198
2199 static struct file_operations set_tracer_fops = {
2200         .open           = tracing_open_generic,
2201         .read           = tracing_set_trace_read,
2202         .write          = tracing_set_trace_write,
2203 };
2204
2205 static struct file_operations tracing_pipe_fops = {
2206         .open           = tracing_open_pipe,
2207         .poll           = tracing_poll_pipe,
2208         .read           = tracing_read_pipe,
2209         .release        = tracing_release_pipe,
2210 };
2211
2212 #ifdef CONFIG_DYNAMIC_FTRACE
2213
2214 static ssize_t
2215 tracing_read_long(struct file *filp, char __user *ubuf,
2216                   size_t cnt, loff_t *ppos)
2217 {
2218         unsigned long *p = filp->private_data;
2219         char buf[64];
2220         int r;
2221
2222         r = sprintf(buf, "%ld\n", *p);
2223
2224         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2225 }
2226
2227 static struct file_operations tracing_read_long_fops = {
2228         .open           = tracing_open_generic,
2229         .read           = tracing_read_long,
2230 };
2231 #endif
2232
2233 static struct dentry *d_tracer;
2234
2235 struct dentry *tracing_init_dentry(void)
2236 {
2237         static int once;
2238
2239         if (d_tracer)
2240                 return d_tracer;
2241
2242         d_tracer = debugfs_create_dir("tracing", NULL);
2243
2244         if (!d_tracer && !once) {
2245                 once = 1;
2246                 pr_warning("Could not create debugfs directory 'tracing'\n");
2247                 return NULL;
2248         }
2249
2250         return d_tracer;
2251 }
2252
2253 #ifdef CONFIG_FTRACE_SELFTEST
2254 /* Let selftest have access to static functions in this file */
2255 #include "trace_selftest.c"
2256 #endif
2257
2258 static __init void tracer_init_debugfs(void)
2259 {
2260         struct dentry *d_tracer;
2261         struct dentry *entry;
2262
2263         d_tracer = tracing_init_dentry();
2264
2265         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2266                                     &global_trace, &tracing_ctrl_fops);
2267         if (!entry)
2268                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2269
2270         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2271                                     NULL, &tracing_iter_fops);
2272         if (!entry)
2273                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2274
2275         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2276                                     &global_trace, &tracing_lt_fops);
2277         if (!entry)
2278                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2279
2280         entry = debugfs_create_file("trace", 0444, d_tracer,
2281                                     &global_trace, &tracing_fops);
2282         if (!entry)
2283                 pr_warning("Could not create debugfs 'trace' entry\n");
2284
2285         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2286                                     &global_trace, &show_traces_fops);
2287         if (!entry)
2288                 pr_warning("Could not create debugfs 'trace' entry\n");
2289
2290         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2291                                     &global_trace, &set_tracer_fops);
2292         if (!entry)
2293                 pr_warning("Could not create debugfs 'trace' entry\n");
2294
2295         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2296                                     &tracing_max_latency,
2297                                     &tracing_max_lat_fops);
2298         if (!entry)
2299                 pr_warning("Could not create debugfs "
2300                            "'tracing_max_latency' entry\n");
2301
2302         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2303                                     &tracing_thresh, &tracing_max_lat_fops);
2304         if (!entry)
2305                 pr_warning("Could not create debugfs "
2306                            "'tracing_threash' entry\n");
2307         entry = debugfs_create_file("README", 0644, d_tracer,
2308                                     NULL, &tracing_readme_fops);
2309         if (!entry)
2310                 pr_warning("Could not create debugfs 'README' entry\n");
2311
2312         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2313                                     NULL, &tracing_pipe_fops);
2314         if (!entry)
2315                 pr_warning("Could not create debugfs "
2316                            "'tracing_threash' entry\n");
2317
2318 #ifdef CONFIG_DYNAMIC_FTRACE
2319         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2320                                     &ftrace_update_tot_cnt,
2321                                     &tracing_read_long_fops);
2322         if (!entry)
2323                 pr_warning("Could not create debugfs "
2324                            "'dyn_ftrace_total_info' entry\n");
2325 #endif
2326 }
2327
2328 /* dummy trace to disable tracing */
2329 static struct tracer no_tracer __read_mostly =
2330 {
2331         .name           = "none",
2332 };
2333
2334 static int trace_alloc_page(void)
2335 {
2336         struct trace_array_cpu *data;
2337         struct page *page, *tmp;
2338         LIST_HEAD(pages);
2339         void *array;
2340         int i;
2341
2342         /* first allocate a page for each CPU */
2343         for_each_possible_cpu(i) {
2344                 array = (void *)__get_free_page(GFP_KERNEL);
2345                 if (array == NULL) {
2346                         printk(KERN_ERR "tracer: failed to allocate page"
2347                                "for trace buffer!\n");
2348                         goto free_pages;
2349                 }
2350
2351                 page = virt_to_page(array);
2352                 list_add(&page->lru, &pages);
2353
2354 /* Only allocate if we are actually using the max trace */
2355 #ifdef CONFIG_TRACER_MAX_TRACE
2356                 array = (void *)__get_free_page(GFP_KERNEL);
2357                 if (array == NULL) {
2358                         printk(KERN_ERR "tracer: failed to allocate page"
2359                                "for trace buffer!\n");
2360                         goto free_pages;
2361                 }
2362                 page = virt_to_page(array);
2363                 list_add(&page->lru, &pages);
2364 #endif
2365         }
2366
2367         /* Now that we successfully allocate a page per CPU, add them */
2368         for_each_possible_cpu(i) {
2369                 data = global_trace.data[i];
2370                 spin_lock_init(&data->lock);
2371                 lockdep_set_class(&data->lock, &data->lock_key);
2372                 page = list_entry(pages.next, struct page, lru);
2373                 list_del_init(&page->lru);
2374                 list_add_tail(&page->lru, &data->trace_pages);
2375                 ClearPageLRU(page);
2376
2377 #ifdef CONFIG_TRACER_MAX_TRACE
2378                 data = max_tr.data[i];
2379                 spin_lock_init(&data->lock);
2380                 lockdep_set_class(&data->lock, &data->lock_key);
2381                 page = list_entry(pages.next, struct page, lru);
2382                 list_del_init(&page->lru);
2383                 list_add_tail(&page->lru, &data->trace_pages);
2384                 SetPageLRU(page);
2385 #endif
2386         }
2387         global_trace.entries += ENTRIES_PER_PAGE;
2388
2389         return 0;
2390
2391  free_pages:
2392         list_for_each_entry_safe(page, tmp, &pages, lru) {
2393                 list_del_init(&page->lru);
2394                 __free_page(page);
2395         }
2396         return -ENOMEM;
2397 }
2398
2399 __init static int tracer_alloc_buffers(void)
2400 {
2401         struct trace_array_cpu *data;
2402         void *array;
2403         struct page *page;
2404         int pages = 0;
2405         int ret = -ENOMEM;
2406         int i;
2407
2408         global_trace.ctrl = tracer_enabled;
2409
2410         /* Allocate the first page for all buffers */
2411         for_each_possible_cpu(i) {
2412                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2413                 max_tr.data[i] = &per_cpu(max_data, i);
2414
2415                 array = (void *)__get_free_page(GFP_KERNEL);
2416                 if (array == NULL) {
2417                         printk(KERN_ERR "tracer: failed to allocate page"
2418                                "for trace buffer!\n");
2419                         goto free_buffers;
2420                 }
2421
2422                 /* set the array to the list */
2423                 INIT_LIST_HEAD(&data->trace_pages);
2424                 page = virt_to_page(array);
2425                 list_add(&page->lru, &data->trace_pages);
2426                 /* use the LRU flag to differentiate the two buffers */
2427                 ClearPageLRU(page);
2428
2429 /* Only allocate if we are actually using the max trace */
2430 #ifdef CONFIG_TRACER_MAX_TRACE
2431                 array = (void *)__get_free_page(GFP_KERNEL);
2432                 if (array == NULL) {
2433                         printk(KERN_ERR "tracer: failed to allocate page"
2434                                "for trace buffer!\n");
2435                         goto free_buffers;
2436                 }
2437
2438                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2439                 page = virt_to_page(array);
2440                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2441                 SetPageLRU(page);
2442 #endif
2443         }
2444
2445         /*
2446          * Since we allocate by orders of pages, we may be able to
2447          * round up a bit.
2448          */
2449         global_trace.entries = ENTRIES_PER_PAGE;
2450         pages++;
2451
2452         while (global_trace.entries < trace_nr_entries) {
2453                 if (trace_alloc_page())
2454                         break;
2455                 pages++;
2456         }
2457         max_tr.entries = global_trace.entries;
2458
2459         pr_info("tracer: %d pages allocated for %ld",
2460                 pages, trace_nr_entries);
2461         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2462         pr_info("   actual entries %ld\n", global_trace.entries);
2463
2464         tracer_init_debugfs();
2465
2466         trace_init_cmdlines();
2467
2468         register_tracer(&no_tracer);
2469         current_trace = &no_tracer;
2470
2471         /* All seems OK, enable tracing */
2472         tracing_disabled = 0;
2473
2474         return 0;
2475
2476  free_buffers:
2477         for (i-- ; i >= 0; i--) {
2478                 struct page *page, *tmp;
2479                 struct trace_array_cpu *data = global_trace.data[i];
2480
2481                 if (data) {
2482                         list_for_each_entry_safe(page, tmp,
2483                                                  &data->trace_pages, lru) {
2484                                 list_del_init(&page->lru);
2485                                 __free_page(page);
2486                         }
2487                 }
2488
2489 #ifdef CONFIG_TRACER_MAX_TRACE
2490                 data = max_tr.data[i];
2491                 if (data) {
2492                         list_for_each_entry_safe(page, tmp,
2493                                                  &data->trace_pages, lru) {
2494                                 list_del_init(&page->lru);
2495                                 __free_page(page);
2496                         }
2497                 }
2498 #endif
2499         }
2500         return ret;
2501 }
2502 fs_initcall(tracer_alloc_buffers);