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/gfp.h>
  28 #include <linux/fs.h>
  29
  30 #include "trace.h"
  31
  32 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  33 unsigned long __read_mostly     tracing_thresh;
  34
  35 static int tracing_disabled = 1;
  36
  37 static long notrace
  38 ns2usecs(cycle_t nsec)
  39 {
  40         nsec += 500;
  41         do_div(nsec, 1000);
  42         return nsec;
  43 }
  44
  45 notrace cycle_t ftrace_now(int cpu)
  46 {
  47         return cpu_clock(cpu);
  48 }
  49
  50 static struct trace_array       global_trace;
  51
  52 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  53
  54 static struct trace_array       max_tr;
  55
  56 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  57
  58 static int                      tracer_enabled = 1;
  59 static unsigned long            trace_nr_entries = 16384UL;
  60
  61 static struct tracer            *trace_types __read_mostly;
  62 static struct tracer            *current_trace __read_mostly;
  63 static int                      max_tracer_type_len;
  64
  65 static DEFINE_MUTEX(trace_types_lock);
  66
  67 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  68
  69 static int __init set_nr_entries(char *str)
  70 {
  71         if (!str)
  72                 return 0;
  73         trace_nr_entries = simple_strtoul(str, &str, 0);
  74         return 1;
  75 }
  76 __setup("trace_entries=", set_nr_entries);
  77
  78 unsigned long nsecs_to_usecs(unsigned long nsecs)
  79 {
  80         return nsecs / 1000;
  81 }
  82
  83 enum trace_type {
  84         __TRACE_FIRST_TYPE = 0,
  85
  86         TRACE_FN,
  87         TRACE_CTX,
  88         TRACE_SPECIAL,
  89
  90         __TRACE_LAST_TYPE
  91 };
  92
  93 enum trace_flag_type {
  94         TRACE_FLAG_IRQS_OFF             = 0x01,
  95         TRACE_FLAG_NEED_RESCHED         = 0x02,
  96         TRACE_FLAG_HARDIRQ              = 0x04,
  97         TRACE_FLAG_SOFTIRQ              = 0x08,
  98 };
  99
 100 enum trace_iterator_flags {
 101         TRACE_ITER_PRINT_PARENT         = 0x01,
 102         TRACE_ITER_SYM_OFFSET           = 0x02,
 103         TRACE_ITER_SYM_ADDR             = 0x04,
 104         TRACE_ITER_VERBOSE              = 0x08,
 105         TRACE_ITER_RAW                  = 0x10,
 106         TRACE_ITER_HEX                  = 0x20,
 107         TRACE_ITER_BIN                  = 0x40,
 108 };
 109
 110 #define TRACE_ITER_SYM_MASK \
 111         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 112
 113 /* These must match the bit postions above */
 114 static const char *trace_options[] = {
 115         "print-parent",
 116         "sym-offset",
 117         "sym-addr",
 118         "verbose",
 119         "raw",
 120         "hex",
 121         "bin",
 122         NULL
 123 };
 124
 125 static unsigned trace_flags;
 126
 127 static DEFINE_SPINLOCK(ftrace_max_lock);
 128
 129 /*
 130  * Copy the new maximum trace into the separate maximum-trace
 131  * structure. (this way the maximum trace is permanently saved,
 132  * for later retrieval via /debugfs/tracing/latency_trace)
 133  */
 134 static notrace void
 135 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 136 {
 137         struct trace_array_cpu *data = tr->data[cpu];
 138
 139         max_tr.cpu = cpu;
 140         max_tr.time_start = data->preempt_timestamp;
 141
 142         data = max_tr.data[cpu];
 143         data->saved_latency = tracing_max_latency;
 144
 145         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 146         data->pid = tsk->pid;
 147         data->uid = tsk->uid;
 148         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 149         data->policy = tsk->policy;
 150         data->rt_priority = tsk->rt_priority;
 151
 152         /* record this tasks comm */
 153         tracing_record_cmdline(current);
 154 }
 155
 156 void check_pages(struct trace_array_cpu *data)
 157 {
 158         struct page *page, *tmp;
 159
 160         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 161         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 162
 163         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 164                 BUG_ON(page->lru.next->prev != &page->lru);
 165                 BUG_ON(page->lru.prev->next != &page->lru);
 166         }
 167 }
 168
 169 void *head_page(struct trace_array_cpu *data)
 170 {
 171         struct page *page;
 172
 173         check_pages(data);
 174         if (list_empty(&data->trace_pages))
 175                 return NULL;
 176
 177         page = list_entry(data->trace_pages.next, struct page, lru);
 178         BUG_ON(&page->lru == &data->trace_pages);
 179
 180         return page_address(page);
 181 }
 182
 183 static notrace int
 184 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 185 {
 186         int len = (PAGE_SIZE - 1) - s->len;
 187         va_list ap;
 188         int ret;
 189
 190         if (!len)
 191                 return 0;
 192
 193         va_start(ap, fmt);
 194         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 195         va_end(ap);
 196
 197         /* If we can't write it all, don't bother writing anything */
 198         if (ret > len)
 199                 return 0;
 200
 201         s->len += ret;
 202
 203         return len;
 204 }
 205
 206 static notrace int
 207 trace_seq_puts(struct trace_seq *s, const char *str)
 208 {
 209         int len = strlen(str);
 210
 211         if (len > ((PAGE_SIZE - 1) - s->len))
 212                 return 0;
 213
 214         memcpy(s->buffer + s->len, str, len);
 215         s->len += len;
 216
 217         return len;
 218 }
 219
 220 static notrace int
 221 trace_seq_putc(struct trace_seq *s, unsigned char c)
 222 {
 223         if (s->len >= (PAGE_SIZE - 1))
 224                 return 0;
 225
 226         s->buffer[s->len++] = c;
 227
 228         return 1;
 229 }
 230
 231 static notrace int
 232 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 233 {
 234         if (len > ((PAGE_SIZE - 1) - s->len))
 235                 return 0;
 236
 237         memcpy(s->buffer + s->len, mem, len);
 238         s->len += len;
 239
 240         return len;
 241 }
 242
 243 #define HEX_CHARS 17
 244
 245 static notrace int
 246 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 247 {
 248         unsigned char hex[HEX_CHARS];
 249         unsigned char *data;
 250         unsigned char byte;
 251         int i, j;
 252
 253         BUG_ON(len >= HEX_CHARS);
 254
 255         data = mem;
 256
 257 #ifdef __BIG_ENDIAN
 258         for (i = 0, j = 0; i < len; i++) {
 259 #else
 260         for (i = len-1, j = 0; i >= 0; i--) {
 261 #endif
 262                 byte = data[i];
 263
 264                 hex[j]   = byte & 0x0f;
 265                 if (hex[j] >= 10)
 266                         hex[j] += 'a' - 10;
 267                 else
 268                         hex[j] += '0';
 269                 j++;
 270
 271                 hex[j] = byte >> 4;
 272                 if (hex[j] >= 10)
 273                         hex[j] += 'a' - 10;
 274                 else
 275                         hex[j] += '0';
 276                 j++;
 277         }
 278         hex[j] = ' ';
 279         j++;
 280
 281         return trace_seq_putmem(s, hex, j);
 282 }
 283
 284 static notrace void
 285 trace_seq_reset(struct trace_seq *s)
 286 {
 287         s->len = 0;
 288 }
 289
 290 static notrace void
 291 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 292 {
 293         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 294
 295         s->buffer[len] = 0;
 296         seq_puts(m, s->buffer);
 297
 298         trace_seq_reset(s);
 299 }
 300
 301 notrace static void
 302 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 303 {
 304         struct list_head flip_pages;
 305
 306         INIT_LIST_HEAD(&flip_pages);
 307
 308         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 309                 sizeof(struct trace_array_cpu) -
 310                 offsetof(struct trace_array_cpu, trace_head_idx));
 311
 312         check_pages(tr1);
 313         check_pages(tr2);
 314         list_splice_init(&tr1->trace_pages, &flip_pages);
 315         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 316         list_splice_init(&flip_pages, &tr2->trace_pages);
 317         BUG_ON(!list_empty(&flip_pages));
 318         check_pages(tr1);
 319         check_pages(tr2);
 320 }
 321
 322 notrace void
 323 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 324 {
 325         struct trace_array_cpu *data;
 326         int i;
 327
 328         WARN_ON_ONCE(!irqs_disabled());
 329         spin_lock(&ftrace_max_lock);
 330         /* clear out all the previous traces */
 331         for_each_possible_cpu(i) {
 332                 data = tr->data[i];
 333                 flip_trace(max_tr.data[i], data);
 334                 tracing_reset(data);
 335         }
 336
 337         __update_max_tr(tr, tsk, cpu);
 338         spin_unlock(&ftrace_max_lock);
 339 }
 340
 341 /**
 342  * update_max_tr_single - only copy one trace over, and reset the rest
 343  * @tr - tracer
 344  * @tsk - task with the latency
 345  * @cpu - the cpu of the buffer to copy.
 346  */
 347 notrace void
 348 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 349 {
 350         struct trace_array_cpu *data = tr->data[cpu];
 351         int i;
 352
 353         WARN_ON_ONCE(!irqs_disabled());
 354         spin_lock(&ftrace_max_lock);
 355         for_each_possible_cpu(i)
 356                 tracing_reset(max_tr.data[i]);
 357
 358         flip_trace(max_tr.data[cpu], data);
 359         tracing_reset(data);
 360
 361         __update_max_tr(tr, tsk, cpu);
 362         spin_unlock(&ftrace_max_lock);
 363 }
 364
 365 int register_tracer(struct tracer *type)
 366 {
 367         struct tracer *t;
 368         int len;
 369         int ret = 0;
 370
 371         if (!type->name) {
 372                 pr_info("Tracer must have a name\n");
 373                 return -1;
 374         }
 375
 376         mutex_lock(&trace_types_lock);
 377         for (t = trace_types; t; t = t->next) {
 378                 if (strcmp(type->name, t->name) == 0) {
 379                         /* already found */
 380                         pr_info("Trace %s already registered\n",
 381                                 type->name);
 382                         ret = -1;
 383                         goto out;
 384                 }
 385         }
 386
 387 #ifdef CONFIG_FTRACE_STARTUP_TEST
 388         if (type->selftest) {
 389                 struct tracer *saved_tracer = current_trace;
 390                 struct trace_array_cpu *data;
 391                 struct trace_array *tr = &global_trace;
 392                 int saved_ctrl = tr->ctrl;
 393                 int i;
 394                 /*
 395                  * Run a selftest on this tracer.
 396                  * Here we reset the trace buffer, and set the current
 397                  * tracer to be this tracer. The tracer can then run some
 398                  * internal tracing to verify that everything is in order.
 399                  * If we fail, we do not register this tracer.
 400                  */
 401                 for_each_possible_cpu(i) {
 402                         data = tr->data[i];
 403                         if (!head_page(data))
 404                                 continue;
 405                         tracing_reset(data);
 406                 }
 407                 current_trace = type;
 408                 tr->ctrl = 0;
 409                 /* the test is responsible for initializing and enabling */
 410                 pr_info("Testing tracer %s: ", type->name);
 411                 ret = type->selftest(type, tr);
 412                 /* the test is responsible for resetting too */
 413                 current_trace = saved_tracer;
 414                 tr->ctrl = saved_ctrl;
 415                 if (ret) {
 416                         printk(KERN_CONT "FAILED!\n");
 417                         goto out;
 418                 }
 419                 /* Only reset on passing, to avoid touching corrupted buffers */
 420                 for_each_possible_cpu(i) {
 421                         data = tr->data[i];
 422                         if (!head_page(data))
 423                                 continue;
 424                         tracing_reset(data);
 425                 }
 426                 printk(KERN_CONT "PASSED\n");
 427         }
 428 #endif
 429
 430         type->next = trace_types;
 431         trace_types = type;
 432         len = strlen(type->name);
 433         if (len > max_tracer_type_len)
 434                 max_tracer_type_len = len;
 435
 436  out:
 437         mutex_unlock(&trace_types_lock);
 438
 439         return ret;
 440 }
 441
 442 void unregister_tracer(struct tracer *type)
 443 {
 444         struct tracer **t;
 445         int len;
 446
 447         mutex_lock(&trace_types_lock);
 448         for (t = &trace_types; *t; t = &(*t)->next) {
 449                 if (*t == type)
 450                         goto found;
 451         }
 452         pr_info("Trace %s not registered\n", type->name);
 453         goto out;
 454
 455  found:
 456         *t = (*t)->next;
 457         if (strlen(type->name) != max_tracer_type_len)
 458                 goto out;
 459
 460         max_tracer_type_len = 0;
 461         for (t = &trace_types; *t; t = &(*t)->next) {
 462                 len = strlen((*t)->name);
 463                 if (len > max_tracer_type_len)
 464                         max_tracer_type_len = len;
 465         }
 466  out:
 467         mutex_unlock(&trace_types_lock);
 468 }
 469
 470 notrace void tracing_reset(struct trace_array_cpu *data)
 471 {
 472         data->trace_idx = 0;
 473         data->trace_head = data->trace_tail = head_page(data);
 474         data->trace_head_idx = 0;
 475         data->trace_tail_idx = 0;
 476 }
 477
 478 #define SAVED_CMDLINES 128
 479 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 480 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 481 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 482 static int cmdline_idx;
 483 static DEFINE_SPINLOCK(trace_cmdline_lock);
 484 atomic_t trace_record_cmdline_disabled;
 485
 486 static void trace_init_cmdlines(void)
 487 {
 488         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 489         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 490         cmdline_idx = 0;
 491 }
 492
 493 notrace void trace_stop_cmdline_recording(void);
 494
 495 static notrace void trace_save_cmdline(struct task_struct *tsk)
 496 {
 497         unsigned map;
 498         unsigned idx;
 499
 500         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 501                 return;
 502
 503         /*
 504          * It's not the end of the world if we don't get
 505          * the lock, but we also don't want to spin
 506          * nor do we want to disable interrupts,
 507          * so if we miss here, then better luck next time.
 508          */
 509         if (!spin_trylock(&trace_cmdline_lock))
 510                 return;
 511
 512         idx = map_pid_to_cmdline[tsk->pid];
 513         if (idx >= SAVED_CMDLINES) {
 514                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 515
 516                 map = map_cmdline_to_pid[idx];
 517                 if (map <= PID_MAX_DEFAULT)
 518                         map_pid_to_cmdline[map] = (unsigned)-1;
 519
 520                 map_pid_to_cmdline[tsk->pid] = idx;
 521
 522                 cmdline_idx = idx;
 523         }
 524
 525         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 526
 527         spin_unlock(&trace_cmdline_lock);
 528 }
 529
 530 static notrace char *trace_find_cmdline(int pid)
 531 {
 532         char *cmdline = "<...>";
 533         unsigned map;
 534
 535         if (!pid)
 536                 return "<idle>";
 537
 538         if (pid > PID_MAX_DEFAULT)
 539                 goto out;
 540
 541         map = map_pid_to_cmdline[pid];
 542         if (map >= SAVED_CMDLINES)
 543                 goto out;
 544
 545         cmdline = saved_cmdlines[map];
 546
 547  out:
 548         return cmdline;
 549 }
 550
 551 notrace void tracing_record_cmdline(struct task_struct *tsk)
 552 {
 553         if (atomic_read(&trace_record_cmdline_disabled))
 554                 return;
 555
 556         trace_save_cmdline(tsk);
 557 }
 558
 559 static inline notrace struct list_head *
 560 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 561 {
 562         /*
 563          * Roundrobin - but skip the head (which is not a real page):
 564          */
 565         next = next->next;
 566         if (unlikely(next == &data->trace_pages))
 567                 next = next->next;
 568         BUG_ON(next == &data->trace_pages);
 569
 570         return next;
 571 }
 572
 573 static inline notrace void *
 574 trace_next_page(struct trace_array_cpu *data, void *addr)
 575 {
 576         struct list_head *next;
 577         struct page *page;
 578
 579         page = virt_to_page(addr);
 580
 581         next = trace_next_list(data, &page->lru);
 582         page = list_entry(next, struct page, lru);
 583
 584         return page_address(page);
 585 }
 586
 587 static inline notrace struct trace_entry *
 588 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 589 {
 590         unsigned long idx, idx_next;
 591         struct trace_entry *entry;
 592
 593         data->trace_idx++;
 594         idx = data->trace_head_idx;
 595         idx_next = idx + 1;
 596
 597         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 598
 599         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 600
 601         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 602                 data->trace_head = trace_next_page(data, data->trace_head);
 603                 idx_next = 0;
 604         }
 605
 606         if (data->trace_head == data->trace_tail &&
 607             idx_next == data->trace_tail_idx) {
 608                 /* overrun */
 609                 data->trace_tail_idx++;
 610                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 611                         data->trace_tail =
 612                                 trace_next_page(data, data->trace_tail);
 613                         data->trace_tail_idx = 0;
 614                 }
 615         }
 616
 617         data->trace_head_idx = idx_next;
 618
 619         return entry;
 620 }
 621
 622 static inline notrace void
 623 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 624 {
 625         struct task_struct *tsk = current;
 626         unsigned long pc;
 627
 628         pc = preempt_count();
 629
 630         entry->preempt_count    = pc & 0xff;
 631         entry->pid              = tsk->pid;
 632         entry->t                = ftrace_now(raw_smp_processor_id());
 633         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 634                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 635                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 636                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 637 }
 638
 639 notrace void
 640 __ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 641          unsigned long ip, unsigned long parent_ip, unsigned long flags)
 642 {
 643         struct trace_entry *entry;
 644         unsigned long irq_flags;
 645
 646         spin_lock_irqsave(&data->lock, irq_flags);
 647         entry                   = tracing_get_trace_entry(tr, data);
 648         tracing_generic_entry_update(entry, flags);
 649         entry->type             = TRACE_FN;
 650         entry->fn.ip            = ip;
 651         entry->fn.parent_ip     = parent_ip;
 652         spin_unlock_irqrestore(&data->lock, irq_flags);
 653 }
 654
 655 notrace void
 656 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 657        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 658 {
 659         if (likely(!atomic_read(&data->disabled)))
 660                 __ftrace(tr, data, ip, parent_ip, flags);
 661 }
 662
 663 notrace void
 664 trace_special(struct trace_array *tr, struct trace_array_cpu *data,
 665               unsigned long arg1, unsigned long arg2, unsigned long arg3)
 666 {
 667         struct trace_entry *entry;
 668         unsigned long irq_flags;
 669
 670         spin_lock_irqsave(&data->lock, irq_flags);
 671         entry                   = tracing_get_trace_entry(tr, data);
 672         tracing_generic_entry_update(entry, 0);
 673         entry->type             = TRACE_SPECIAL;
 674         entry->special.arg1     = arg1;
 675         entry->special.arg2     = arg2;
 676         entry->special.arg3     = arg3;
 677         spin_unlock_irqrestore(&data->lock, irq_flags);
 678 }
 679
 680 notrace void
 681 tracing_sched_switch_trace(struct trace_array *tr,
 682                            struct trace_array_cpu *data,
 683                            struct task_struct *prev, struct task_struct *next,
 684                            unsigned long flags)
 685 {
 686         struct trace_entry *entry;
 687         unsigned long irq_flags;
 688
 689         spin_lock_irqsave(&data->lock, irq_flags);
 690         entry                   = tracing_get_trace_entry(tr, data);
 691         tracing_generic_entry_update(entry, flags);
 692         entry->type             = TRACE_CTX;
 693         entry->ctx.prev_pid     = prev->pid;
 694         entry->ctx.prev_prio    = prev->prio;
 695         entry->ctx.prev_state   = prev->state;
 696         entry->ctx.next_pid     = next->pid;
 697         entry->ctx.next_prio    = next->prio;
 698         spin_unlock_irqrestore(&data->lock, irq_flags);
 699 }
 700
 701 #ifdef CONFIG_FTRACE
 702 static notrace void
 703 function_trace_call(unsigned long ip, unsigned long parent_ip)
 704 {
 705         struct trace_array *tr = &global_trace;
 706         struct trace_array_cpu *data;
 707         unsigned long flags;
 708         long disabled;
 709         int cpu;
 710
 711         if (unlikely(!tracer_enabled))
 712                 return;
 713
 714         local_irq_save(flags);
 715         cpu = raw_smp_processor_id();
 716         data = tr->data[cpu];
 717         disabled = atomic_inc_return(&data->disabled);
 718
 719         if (likely(disabled == 1))
 720                 __ftrace(tr, data, ip, parent_ip, flags);
 721
 722         atomic_dec(&data->disabled);
 723         local_irq_restore(flags);
 724 }
 725
 726 static struct ftrace_ops trace_ops __read_mostly =
 727 {
 728         .func = function_trace_call,
 729 };
 730
 731 notrace void tracing_start_function_trace(void)
 732 {
 733         register_ftrace_function(&trace_ops);
 734 }
 735
 736 notrace void tracing_stop_function_trace(void)
 737 {
 738         unregister_ftrace_function(&trace_ops);
 739 }
 740 #endif
 741
 742 enum trace_file_type {
 743         TRACE_FILE_LAT_FMT      = 1,
 744 };
 745
 746 static struct trace_entry *
 747 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 748                 struct trace_iterator *iter, int cpu)
 749 {
 750         struct page *page;
 751         struct trace_entry *array;
 752
 753         if (iter->next_idx[cpu] >= tr->entries ||
 754             iter->next_idx[cpu] >= data->trace_idx ||
 755             (data->trace_head == data->trace_tail &&
 756              data->trace_head_idx == data->trace_tail_idx))
 757                 return NULL;
 758
 759         if (!iter->next_page[cpu]) {
 760                 /* Initialize the iterator for this cpu trace buffer */
 761                 WARN_ON(!data->trace_tail);
 762                 page = virt_to_page(data->trace_tail);
 763                 iter->next_page[cpu] = &page->lru;
 764                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 765         }
 766
 767         page = list_entry(iter->next_page[cpu], struct page, lru);
 768         BUG_ON(&data->trace_pages == &page->lru);
 769
 770         array = page_address(page);
 771
 772         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 773         return &array[iter->next_page_idx[cpu]];
 774 }
 775
 776 static struct trace_entry * notrace
 777 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 778 {
 779         struct trace_array *tr = iter->tr;
 780         struct trace_entry *ent, *next = NULL;
 781         int next_cpu = -1;
 782         int cpu;
 783
 784         for_each_possible_cpu(cpu) {
 785                 if (!head_page(tr->data[cpu]))
 786                         continue;
 787                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 788                 /*
 789                  * Pick the entry with the smallest timestamp:
 790                  */
 791                 if (ent && (!next || ent->t < next->t)) {
 792                         next = ent;
 793                         next_cpu = cpu;
 794                 }
 795         }
 796
 797         if (ent_cpu)
 798                 *ent_cpu = next_cpu;
 799
 800         return next;
 801 }
 802
 803 static notrace void trace_iterator_increment(struct trace_iterator *iter)
 804 {
 805         iter->idx++;
 806         iter->next_idx[iter->cpu]++;
 807         iter->next_page_idx[iter->cpu]++;
 808
 809         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 810                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 811
 812                 iter->next_page_idx[iter->cpu] = 0;
 813                 iter->next_page[iter->cpu] =
 814                         trace_next_list(data, iter->next_page[iter->cpu]);
 815         }
 816 }
 817
 818 static notrace void trace_consume(struct trace_iterator *iter)
 819 {
 820         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 821
 822         data->trace_tail_idx++;
 823         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 824                 data->trace_tail = trace_next_page(data, data->trace_tail);
 825                 data->trace_tail_idx = 0;
 826         }
 827
 828         /* Check if we empty it, then reset the index */
 829         if (data->trace_head == data->trace_tail &&
 830             data->trace_head_idx == data->trace_tail_idx)
 831                 data->trace_idx = 0;
 832 }
 833
 834 static notrace void *find_next_entry_inc(struct trace_iterator *iter)
 835 {
 836         struct trace_entry *next;
 837         int next_cpu = -1;
 838
 839         next = find_next_entry(iter, &next_cpu);
 840
 841         iter->prev_ent = iter->ent;
 842         iter->prev_cpu = iter->cpu;
 843
 844         iter->ent = next;
 845         iter->cpu = next_cpu;
 846
 847         if (next)
 848                 trace_iterator_increment(iter);
 849
 850         return next ? iter : NULL;
 851 }
 852
 853 static notrace void *s_next(struct seq_file *m, void *v, loff_t *pos)
 854 {
 855         struct trace_iterator *iter = m->private;
 856         void *last_ent = iter->ent;
 857         int i = (int)*pos;
 858         void *ent;
 859
 860         (*pos)++;
 861
 862         /* can't go backwards */
 863         if (iter->idx > i)
 864                 return NULL;
 865
 866         if (iter->idx < 0)
 867                 ent = find_next_entry_inc(iter);
 868         else
 869                 ent = iter;
 870
 871         while (ent && iter->idx < i)
 872                 ent = find_next_entry_inc(iter);
 873
 874         iter->pos = *pos;
 875
 876         if (last_ent && !ent)
 877                 seq_puts(m, "\n\nvim:ft=help\n");
 878
 879         return ent;
 880 }
 881
 882 static void *s_start(struct seq_file *m, loff_t *pos)
 883 {
 884         struct trace_iterator *iter = m->private;
 885         void *p = NULL;
 886         loff_t l = 0;
 887         int i;
 888
 889         mutex_lock(&trace_types_lock);
 890
 891         if (!current_trace || current_trace != iter->trace)
 892                 return NULL;
 893
 894         atomic_inc(&trace_record_cmdline_disabled);
 895
 896         /* let the tracer grab locks here if needed */
 897         if (current_trace->start)
 898                 current_trace->start(iter);
 899
 900         if (*pos != iter->pos) {
 901                 iter->ent = NULL;
 902                 iter->cpu = 0;
 903                 iter->idx = -1;
 904                 iter->prev_ent = NULL;
 905                 iter->prev_cpu = -1;
 906
 907                 for_each_possible_cpu(i) {
 908                         iter->next_idx[i] = 0;
 909                         iter->next_page[i] = NULL;
 910                 }
 911
 912                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 913                         ;
 914
 915         } else {
 916                 l = *pos - 1;
 917                 p = s_next(m, p, &l);
 918         }
 919
 920         return p;
 921 }
 922
 923 static void s_stop(struct seq_file *m, void *p)
 924 {
 925         struct trace_iterator *iter = m->private;
 926
 927         atomic_dec(&trace_record_cmdline_disabled);
 928
 929         /* let the tracer release locks here if needed */
 930         if (current_trace && current_trace == iter->trace && iter->trace->stop)
 931                 iter->trace->stop(iter);
 932
 933         mutex_unlock(&trace_types_lock);
 934 }
 935
 936 static int
 937 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
 938 {
 939 #ifdef CONFIG_KALLSYMS
 940         char str[KSYM_SYMBOL_LEN];
 941
 942         kallsyms_lookup(address, NULL, NULL, NULL, str);
 943
 944         return trace_seq_printf(s, fmt, str);
 945 #endif
 946         return 1;
 947 }
 948
 949 static int
 950 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
 951                      unsigned long address)
 952 {
 953 #ifdef CONFIG_KALLSYMS
 954         char str[KSYM_SYMBOL_LEN];
 955
 956         sprint_symbol(str, address);
 957         return trace_seq_printf(s, fmt, str);
 958 #endif
 959         return 1;
 960 }
 961
 962 #ifndef CONFIG_64BIT
 963 # define IP_FMT "%08lx"
 964 #else
 965 # define IP_FMT "%016lx"
 966 #endif
 967
 968 static notrace int
 969 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 970 {
 971         int ret;
 972
 973         if (!ip)
 974                 return trace_seq_printf(s, "0");
 975
 976         if (sym_flags & TRACE_ITER_SYM_OFFSET)
 977                 ret = seq_print_sym_offset(s, "%s", ip);
 978         else
 979                 ret = seq_print_sym_short(s, "%s", ip);
 980
 981         if (!ret)
 982                 return 0;
 983
 984         if (sym_flags & TRACE_ITER_SYM_ADDR)
 985                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
 986         return ret;
 987 }
 988
 989 static notrace void print_lat_help_header(struct seq_file *m)
 990 {
 991         seq_puts(m, "#                _------=> CPU#            \n");
 992         seq_puts(m, "#               / _-----=> irqs-off        \n");
 993         seq_puts(m, "#              | / _----=> need-resched    \n");
 994         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
 995         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
 996         seq_puts(m, "#              |||| /                      \n");
 997         seq_puts(m, "#              |||||     delay             \n");
 998         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
 999         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1000 }
1001
1002 static notrace void print_func_help_header(struct seq_file *m)
1003 {
1004         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1005         seq_puts(m, "#              | |      |          |         |\n");
1006 }
1007
1008
1009 static notrace void
1010 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1011 {
1012         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1013         struct trace_array *tr = iter->tr;
1014         struct trace_array_cpu *data = tr->data[tr->cpu];
1015         struct tracer *type = current_trace;
1016         unsigned long total   = 0;
1017         unsigned long entries = 0;
1018         int cpu;
1019         const char *name = "preemption";
1020
1021         if (type)
1022                 name = type->name;
1023
1024         for_each_possible_cpu(cpu) {
1025                 if (head_page(tr->data[cpu])) {
1026                         total += tr->data[cpu]->trace_idx;
1027                         if (tr->data[cpu]->trace_idx > tr->entries)
1028                                 entries += tr->entries;
1029                         else
1030                                 entries += tr->data[cpu]->trace_idx;
1031                 }
1032         }
1033
1034         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1035                    name, UTS_RELEASE);
1036         seq_puts(m, "-----------------------------------"
1037                  "---------------------------------\n");
1038         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1039                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1040                    nsecs_to_usecs(data->saved_latency),
1041                    entries,
1042                    total,
1043                    tr->cpu,
1044 #if defined(CONFIG_PREEMPT_NONE)
1045                    "server",
1046 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1047                    "desktop",
1048 #elif defined(CONFIG_PREEMPT_DESKTOP)
1049                    "preempt",
1050 #else
1051                    "unknown",
1052 #endif
1053                    /* These are reserved for later use */
1054                    0, 0, 0, 0);
1055 #ifdef CONFIG_SMP
1056         seq_printf(m, " #P:%d)\n", num_online_cpus());
1057 #else
1058         seq_puts(m, ")\n");
1059 #endif
1060         seq_puts(m, "    -----------------\n");
1061         seq_printf(m, "    | task: %.16s-%d "
1062                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1063                    data->comm, data->pid, data->uid, data->nice,
1064                    data->policy, data->rt_priority);
1065         seq_puts(m, "    -----------------\n");
1066
1067         if (data->critical_start) {
1068                 seq_puts(m, " => started at: ");
1069                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1070                 trace_print_seq(m, &iter->seq);
1071                 seq_puts(m, "\n => ended at:   ");
1072                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1073                 trace_print_seq(m, &iter->seq);
1074                 seq_puts(m, "\n");
1075         }
1076
1077         seq_puts(m, "\n");
1078 }
1079
1080 static notrace void
1081 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1082 {
1083         int hardirq, softirq;
1084         char *comm;
1085
1086         comm = trace_find_cmdline(entry->pid);
1087
1088         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1089         trace_seq_printf(s, "%d", cpu);
1090         trace_seq_printf(s, "%c%c",
1091                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1092                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1093
1094         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1095         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1096         if (hardirq && softirq)
1097                 trace_seq_putc(s, 'H');
1098         else {
1099                 if (hardirq)
1100                         trace_seq_putc(s, 'h');
1101                 else {
1102                         if (softirq)
1103                                 trace_seq_putc(s, 's');
1104                         else
1105                                 trace_seq_putc(s, '.');
1106                 }
1107         }
1108
1109         if (entry->preempt_count)
1110                 trace_seq_printf(s, "%x", entry->preempt_count);
1111         else
1112                 trace_seq_puts(s, ".");
1113 }
1114
1115 unsigned long preempt_mark_thresh = 100;
1116
1117 static notrace void
1118 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1119                     unsigned long rel_usecs)
1120 {
1121         trace_seq_printf(s, " %4lldus", abs_usecs);
1122         if (rel_usecs > preempt_mark_thresh)
1123                 trace_seq_puts(s, "!: ");
1124         else if (rel_usecs > 1)
1125                 trace_seq_puts(s, "+: ");
1126         else
1127                 trace_seq_puts(s, " : ");
1128 }
1129
1130 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1131
1132 static notrace int
1133 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1134 {
1135         struct trace_seq *s = &iter->seq;
1136         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1137         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1138         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1139         struct trace_entry *entry = iter->ent;
1140         unsigned long abs_usecs;
1141         unsigned long rel_usecs;
1142         char *comm;
1143         int S;
1144
1145         if (!next_entry)
1146                 next_entry = entry;
1147         rel_usecs = ns2usecs(next_entry->t - entry->t);
1148         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1149
1150         if (verbose) {
1151                 comm = trace_find_cmdline(entry->pid);
1152                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1153                                  " %ld.%03ldms (+%ld.%03ldms): ",
1154                                  comm,
1155                                  entry->pid, cpu, entry->flags,
1156                                  entry->preempt_count, trace_idx,
1157                                  ns2usecs(entry->t),
1158                                  abs_usecs/1000,
1159                                  abs_usecs % 1000, rel_usecs/1000,
1160                                  rel_usecs % 1000);
1161         } else {
1162                 lat_print_generic(s, entry, cpu);
1163                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1164         }
1165         switch (entry->type) {
1166         case TRACE_FN:
1167                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1168                 trace_seq_puts(s, " (");
1169                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1170                 trace_seq_puts(s, ")\n");
1171                 break;
1172         case TRACE_CTX:
1173                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1174                         state_to_char[entry->ctx.prev_state] : 'X';
1175                 comm = trace_find_cmdline(entry->ctx.next_pid);
1176                 trace_seq_printf(s, " %d:%d:%c --> %d:%d %s\n",
1177                                  entry->ctx.prev_pid,
1178                                  entry->ctx.prev_prio,
1179                                  S,
1180                                  entry->ctx.next_pid,
1181                                  entry->ctx.next_prio,
1182                                  comm);
1183                 break;
1184         case TRACE_SPECIAL:
1185                 trace_seq_printf(s, " %lx %lx %lx\n",
1186                                  entry->special.arg1,
1187                                  entry->special.arg2,
1188                                  entry->special.arg3);
1189                 break;
1190         default:
1191                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1192         }
1193         return 1;
1194 }
1195
1196 static notrace int print_trace_fmt(struct trace_iterator *iter)
1197 {
1198         struct trace_seq *s = &iter->seq;
1199         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1200         struct trace_entry *entry;
1201         unsigned long usec_rem;
1202         unsigned long long t;
1203         unsigned long secs;
1204         char *comm;
1205         int S;
1206         int ret;
1207
1208         entry = iter->ent;
1209
1210         comm = trace_find_cmdline(iter->ent->pid);
1211
1212         t = ns2usecs(entry->t);
1213         usec_rem = do_div(t, 1000000ULL);
1214         secs = (unsigned long)t;
1215
1216         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1217         if (!ret)
1218                 return 0;
1219         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1220         if (!ret)
1221                 return 0;
1222         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1223         if (!ret)
1224                 return 0;
1225
1226         switch (entry->type) {
1227         case TRACE_FN:
1228                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1229                 if (!ret)
1230                         return 0;
1231                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1232                                                 entry->fn.parent_ip) {
1233                         ret = trace_seq_printf(s, " <-");
1234                         if (!ret)
1235                                 return 0;
1236                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1237                                                sym_flags);
1238                         if (!ret)
1239                                 return 0;
1240                 }
1241                 ret = trace_seq_printf(s, "\n");
1242                 if (!ret)
1243                         return 0;
1244                 break;
1245         case TRACE_CTX:
1246                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1247                         state_to_char[entry->ctx.prev_state] : 'X';
1248                 ret = trace_seq_printf(s, " %d:%d:%c ==> %d:%d\n",
1249                                        entry->ctx.prev_pid,
1250                                        entry->ctx.prev_prio,
1251                                        S,
1252                                        entry->ctx.next_pid,
1253                                        entry->ctx.next_prio);
1254                 if (!ret)
1255                         return 0;
1256                 break;
1257         case TRACE_SPECIAL:
1258                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1259                                  entry->special.arg1,
1260                                  entry->special.arg2,
1261                                  entry->special.arg3);
1262                 if (!ret)
1263                         return 0;
1264                 break;
1265         }
1266         return 1;
1267 }
1268
1269 static notrace int print_raw_fmt(struct trace_iterator *iter)
1270 {
1271         struct trace_seq *s = &iter->seq;
1272         struct trace_entry *entry;
1273         int ret;
1274         int S;
1275
1276         entry = iter->ent;
1277
1278         ret = trace_seq_printf(s, "%d %d %llu ",
1279                 entry->pid, iter->cpu, entry->t);
1280         if (!ret)
1281                 return 0;
1282
1283         switch (entry->type) {
1284         case TRACE_FN:
1285                 ret = trace_seq_printf(s, "%x %x\n",
1286                                         entry->fn.ip, entry->fn.parent_ip);
1287                 if (!ret)
1288                         return 0;
1289                 break;
1290         case TRACE_CTX:
1291                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1292                         state_to_char[entry->ctx.prev_state] : 'X';
1293                 ret = trace_seq_printf(s, "%d %d %c %d %d\n",
1294                                        entry->ctx.prev_pid,
1295                                        entry->ctx.prev_prio,
1296                                        S,
1297                                        entry->ctx.next_pid,
1298                                        entry->ctx.next_prio);
1299                 if (!ret)
1300                         return 0;
1301                 break;
1302         case TRACE_SPECIAL:
1303                 ret = trace_seq_printf(s, " %lx %lx %lx\n",
1304                                  entry->special.arg1,
1305                                  entry->special.arg2,
1306                                  entry->special.arg3);
1307                 if (!ret)
1308                         return 0;
1309                 break;
1310         }
1311         return 1;
1312 }
1313
1314 #define SEQ_PUT_FIELD_RET(s, x)                         \
1315 do {                                                    \
1316         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1317                 return 0;                               \
1318 } while (0)
1319
1320 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1321 do {                                                    \
1322         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1323                 return 0;                               \
1324 } while (0)
1325
1326 static notrace int print_hex_fmt(struct trace_iterator *iter)
1327 {
1328         struct trace_seq *s = &iter->seq;
1329         unsigned char newline = '\n';
1330         struct trace_entry *entry;
1331         int S;
1332
1333         entry = iter->ent;
1334
1335         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1336         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1337         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1338
1339         switch (entry->type) {
1340         case TRACE_FN:
1341                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1342                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1343                 break;
1344         case TRACE_CTX:
1345                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1346                         state_to_char[entry->ctx.prev_state] : 'X';
1347                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1348                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1349                 SEQ_PUT_HEX_FIELD_RET(s, S);
1350                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1351                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1352                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1353                 break;
1354         case TRACE_SPECIAL:
1355                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1356                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1357                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1358                 break;
1359         }
1360         SEQ_PUT_FIELD_RET(s, newline);
1361
1362         return 1;
1363 }
1364
1365 static notrace int print_bin_fmt(struct trace_iterator *iter)
1366 {
1367         struct trace_seq *s = &iter->seq;
1368         struct trace_entry *entry;
1369
1370         entry = iter->ent;
1371
1372         SEQ_PUT_FIELD_RET(s, entry->pid);
1373         SEQ_PUT_FIELD_RET(s, entry->cpu);
1374         SEQ_PUT_FIELD_RET(s, entry->t);
1375
1376         switch (entry->type) {
1377         case TRACE_FN:
1378                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1379                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1380                 break;
1381         case TRACE_CTX:
1382                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1383                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1384                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1385                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1386                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1387                 break;
1388         case TRACE_SPECIAL:
1389                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1390                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1391                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1392                 break;
1393         }
1394         return 1;
1395 }
1396
1397 static int trace_empty(struct trace_iterator *iter)
1398 {
1399         struct trace_array_cpu *data;
1400         int cpu;
1401
1402         for_each_possible_cpu(cpu) {
1403                 data = iter->tr->data[cpu];
1404
1405                 if (head_page(data) && data->trace_idx &&
1406                     (data->trace_tail != data->trace_head ||
1407                      data->trace_tail_idx != data->trace_head_idx))
1408                         return 0;
1409         }
1410         return 1;
1411 }
1412
1413 static int print_trace_line(struct trace_iterator *iter)
1414 {
1415         if (trace_flags & TRACE_ITER_BIN)
1416                 return print_bin_fmt(iter);
1417
1418         if (trace_flags & TRACE_ITER_HEX)
1419                 return print_hex_fmt(iter);
1420
1421         if (trace_flags & TRACE_ITER_RAW)
1422                 return print_raw_fmt(iter);
1423
1424         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1425                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1426
1427         return print_trace_fmt(iter);
1428 }
1429
1430 static int s_show(struct seq_file *m, void *v)
1431 {
1432         struct trace_iterator *iter = v;
1433
1434         if (iter->ent == NULL) {
1435                 if (iter->tr) {
1436                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1437                         seq_puts(m, "#\n");
1438                 }
1439                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1440                         /* print nothing if the buffers are empty */
1441                         if (trace_empty(iter))
1442                                 return 0;
1443                         print_trace_header(m, iter);
1444                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1445                                 print_lat_help_header(m);
1446                 } else {
1447                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1448                                 print_func_help_header(m);
1449                 }
1450         } else {
1451                 print_trace_line(iter);
1452                 trace_print_seq(m, &iter->seq);
1453         }
1454
1455         return 0;
1456 }
1457
1458 static struct seq_operations tracer_seq_ops = {
1459         .start          = s_start,
1460         .next           = s_next,
1461         .stop           = s_stop,
1462         .show           = s_show,
1463 };
1464
1465 static struct trace_iterator notrace *
1466 __tracing_open(struct inode *inode, struct file *file, int *ret)
1467 {
1468         struct trace_iterator *iter;
1469
1470         if (tracing_disabled) {
1471                 *ret = -ENODEV;
1472                 return NULL;
1473         }
1474
1475         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1476         if (!iter) {
1477                 *ret = -ENOMEM;
1478                 goto out;
1479         }
1480
1481         mutex_lock(&trace_types_lock);
1482         if (current_trace && current_trace->print_max)
1483                 iter->tr = &max_tr;
1484         else
1485                 iter->tr = inode->i_private;
1486         iter->trace = current_trace;
1487         iter->pos = -1;
1488
1489         /* TODO stop tracer */
1490         *ret = seq_open(file, &tracer_seq_ops);
1491         if (!*ret) {
1492                 struct seq_file *m = file->private_data;
1493                 m->private = iter;
1494
1495                 /* stop the trace while dumping */
1496                 if (iter->tr->ctrl)
1497                         tracer_enabled = 0;
1498
1499                 if (iter->trace && iter->trace->open)
1500                         iter->trace->open(iter);
1501         } else {
1502                 kfree(iter);
1503                 iter = NULL;
1504         }
1505         mutex_unlock(&trace_types_lock);
1506
1507  out:
1508         return iter;
1509 }
1510
1511 int tracing_open_generic(struct inode *inode, struct file *filp)
1512 {
1513         if (tracing_disabled)
1514                 return -ENODEV;
1515
1516         filp->private_data = inode->i_private;
1517         return 0;
1518 }
1519
1520 int tracing_release(struct inode *inode, struct file *file)
1521 {
1522         struct seq_file *m = (struct seq_file *)file->private_data;
1523         struct trace_iterator *iter = m->private;
1524
1525         mutex_lock(&trace_types_lock);
1526         if (iter->trace && iter->trace->close)
1527                 iter->trace->close(iter);
1528
1529         /* reenable tracing if it was previously enabled */
1530         if (iter->tr->ctrl)
1531                 tracer_enabled = 1;
1532         mutex_unlock(&trace_types_lock);
1533
1534         seq_release(inode, file);
1535         kfree(iter);
1536         return 0;
1537 }
1538
1539 static int tracing_open(struct inode *inode, struct file *file)
1540 {
1541         int ret;
1542
1543         __tracing_open(inode, file, &ret);
1544
1545         return ret;
1546 }
1547
1548 static int tracing_lt_open(struct inode *inode, struct file *file)
1549 {
1550         struct trace_iterator *iter;
1551         int ret;
1552
1553         iter = __tracing_open(inode, file, &ret);
1554
1555         if (!ret)
1556                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1557
1558         return ret;
1559 }
1560
1561
1562 static notrace void *
1563 t_next(struct seq_file *m, void *v, loff_t *pos)
1564 {
1565         struct tracer *t = m->private;
1566
1567         (*pos)++;
1568
1569         if (t)
1570                 t = t->next;
1571
1572         m->private = t;
1573
1574         return t;
1575 }
1576
1577 static void *t_start(struct seq_file *m, loff_t *pos)
1578 {
1579         struct tracer *t = m->private;
1580         loff_t l = 0;
1581
1582         mutex_lock(&trace_types_lock);
1583         for (; t && l < *pos; t = t_next(m, t, &l))
1584                 ;
1585
1586         return t;
1587 }
1588
1589 static void t_stop(struct seq_file *m, void *p)
1590 {
1591         mutex_unlock(&trace_types_lock);
1592 }
1593
1594 static int t_show(struct seq_file *m, void *v)
1595 {
1596         struct tracer *t = v;
1597
1598         if (!t)
1599                 return 0;
1600
1601         seq_printf(m, "%s", t->name);
1602         if (t->next)
1603                 seq_putc(m, ' ');
1604         else
1605                 seq_putc(m, '\n');
1606
1607         return 0;
1608 }
1609
1610 static struct seq_operations show_traces_seq_ops = {
1611         .start          = t_start,
1612         .next           = t_next,
1613         .stop           = t_stop,
1614         .show           = t_show,
1615 };
1616
1617 static int show_traces_open(struct inode *inode, struct file *file)
1618 {
1619         int ret;
1620
1621         if (tracing_disabled)
1622                 return -ENODEV;
1623
1624         ret = seq_open(file, &show_traces_seq_ops);
1625         if (!ret) {
1626                 struct seq_file *m = file->private_data;
1627                 m->private = trace_types;
1628         }
1629
1630         return ret;
1631 }
1632
1633 static struct file_operations tracing_fops = {
1634         .open           = tracing_open,
1635         .read           = seq_read,
1636         .llseek         = seq_lseek,
1637         .release        = tracing_release,
1638 };
1639
1640 static struct file_operations tracing_lt_fops = {
1641         .open           = tracing_lt_open,
1642         .read           = seq_read,
1643         .llseek         = seq_lseek,
1644         .release        = tracing_release,
1645 };
1646
1647 static struct file_operations show_traces_fops = {
1648         .open = show_traces_open,
1649         .read = seq_read,
1650         .release = seq_release,
1651 };
1652
1653 static ssize_t
1654 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1655                        size_t cnt, loff_t *ppos)
1656 {
1657         char *buf;
1658         int r = 0;
1659         int len = 0;
1660         int i;
1661
1662         /* calulate max size */
1663         for (i = 0; trace_options[i]; i++) {
1664                 len += strlen(trace_options[i]);
1665                 len += 3; /* "no" and space */
1666         }
1667
1668         /* +2 for \n and \0 */
1669         buf = kmalloc(len + 2, GFP_KERNEL);
1670         if (!buf)
1671                 return -ENOMEM;
1672
1673         for (i = 0; trace_options[i]; i++) {
1674                 if (trace_flags & (1 << i))
1675                         r += sprintf(buf + r, "%s ", trace_options[i]);
1676                 else
1677                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1678         }
1679
1680         r += sprintf(buf + r, "\n");
1681         WARN_ON(r >= len + 2);
1682
1683         r = simple_read_from_buffer(ubuf, cnt, ppos,
1684                                     buf, r);
1685
1686         kfree(buf);
1687
1688         return r;
1689 }
1690
1691 static ssize_t
1692 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1693                         size_t cnt, loff_t *ppos)
1694 {
1695         char buf[64];
1696         char *cmp = buf;
1697         int neg = 0;
1698         int i;
1699
1700         if (cnt > 63)
1701                 cnt = 63;
1702
1703         if (copy_from_user(&buf, ubuf, cnt))
1704                 return -EFAULT;
1705
1706         buf[cnt] = 0;
1707
1708         if (strncmp(buf, "no", 2) == 0) {
1709                 neg = 1;
1710                 cmp += 2;
1711         }
1712
1713         for (i = 0; trace_options[i]; i++) {
1714                 int len = strlen(trace_options[i]);
1715
1716                 if (strncmp(cmp, trace_options[i], len) == 0) {
1717                         if (neg)
1718                                 trace_flags &= ~(1 << i);
1719                         else
1720                                 trace_flags |= (1 << i);
1721                         break;
1722                 }
1723         }
1724
1725         filp->f_pos += cnt;
1726
1727         return cnt;
1728 }
1729
1730 static struct file_operations tracing_iter_fops = {
1731         .open = tracing_open_generic,
1732         .read = tracing_iter_ctrl_read,
1733         .write = tracing_iter_ctrl_write,
1734 };
1735
1736 static const char readme_msg[] =
1737         "tracing mini-HOWTO:\n\n"
1738         "# mkdir /debug\n"
1739         "# mount -t debugfs nodev /debug\n\n"
1740         "# cat /debug/tracing/available_tracers\n"
1741         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1742         "# cat /debug/tracing/current_tracer\n"
1743         "none\n"
1744         "# echo sched_switch > /debug/tracing/current_tracer\n"
1745         "# cat /debug/tracing/current_tracer\n"
1746         "sched_switch\n"
1747         "# cat /debug/tracing/iter_ctrl\n"
1748         "noprint-parent nosym-offset nosym-addr noverbose\n"
1749         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1750         "# echo 1 > /debug/tracing/tracing_enabled\n"
1751         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1752         "echo 0 > /debug/tracing/tracing_enabled\n"
1753 ;
1754
1755 static ssize_t
1756 tracing_readme_read(struct file *filp, char __user *ubuf,
1757                        size_t cnt, loff_t *ppos)
1758 {
1759         return simple_read_from_buffer(ubuf, cnt, ppos,
1760                                         readme_msg, strlen(readme_msg));
1761 }
1762
1763 static struct file_operations tracing_readme_fops = {
1764         .open = tracing_open_generic,
1765         .read = tracing_readme_read,
1766 };
1767
1768
1769 static ssize_t
1770 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1771                   size_t cnt, loff_t *ppos)
1772 {
1773         struct trace_array *tr = filp->private_data;
1774         char buf[64];
1775         int r;
1776
1777         r = sprintf(buf, "%ld\n", tr->ctrl);
1778         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1779 }
1780
1781 static ssize_t
1782 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1783                    size_t cnt, loff_t *ppos)
1784 {
1785         struct trace_array *tr = filp->private_data;
1786         long val;
1787         char buf[64];
1788
1789         if (cnt > 63)
1790                 cnt = 63;
1791
1792         if (copy_from_user(&buf, ubuf, cnt))
1793                 return -EFAULT;
1794
1795         buf[cnt] = 0;
1796
1797         val = simple_strtoul(buf, NULL, 10);
1798
1799         val = !!val;
1800
1801         mutex_lock(&trace_types_lock);
1802         if (tr->ctrl ^ val) {
1803                 if (val)
1804                         tracer_enabled = 1;
1805                 else
1806                         tracer_enabled = 0;
1807
1808                 tr->ctrl = val;
1809
1810                 if (current_trace && current_trace->ctrl_update)
1811                         current_trace->ctrl_update(tr);
1812         }
1813         mutex_unlock(&trace_types_lock);
1814
1815         filp->f_pos += cnt;
1816
1817         return cnt;
1818 }
1819
1820 static ssize_t
1821 tracing_set_trace_read(struct file *filp, char __user *ubuf,
1822                        size_t cnt, loff_t *ppos)
1823 {
1824         char buf[max_tracer_type_len+2];
1825         int r;
1826
1827         mutex_lock(&trace_types_lock);
1828         if (current_trace)
1829                 r = sprintf(buf, "%s\n", current_trace->name);
1830         else
1831                 r = sprintf(buf, "\n");
1832         mutex_unlock(&trace_types_lock);
1833
1834         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1835 }
1836
1837 static ssize_t
1838 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
1839                         size_t cnt, loff_t *ppos)
1840 {
1841         struct trace_array *tr = &global_trace;
1842         struct tracer *t;
1843         char buf[max_tracer_type_len+1];
1844         int i;
1845
1846         if (cnt > max_tracer_type_len)
1847                 cnt = max_tracer_type_len;
1848
1849         if (copy_from_user(&buf, ubuf, cnt))
1850                 return -EFAULT;
1851
1852         buf[cnt] = 0;
1853
1854         /* strip ending whitespace. */
1855         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
1856                 buf[i] = 0;
1857
1858         mutex_lock(&trace_types_lock);
1859         for (t = trace_types; t; t = t->next) {
1860                 if (strcmp(t->name, buf) == 0)
1861                         break;
1862         }
1863         if (!t || t == current_trace)
1864                 goto out;
1865
1866         if (current_trace && current_trace->reset)
1867                 current_trace->reset(tr);
1868
1869         current_trace = t;
1870         if (t->init)
1871                 t->init(tr);
1872
1873  out:
1874         mutex_unlock(&trace_types_lock);
1875
1876         filp->f_pos += cnt;
1877
1878         return cnt;
1879 }
1880
1881 static ssize_t
1882 tracing_max_lat_read(struct file *filp, char __user *ubuf,
1883                      size_t cnt, loff_t *ppos)
1884 {
1885         unsigned long *ptr = filp->private_data;
1886         char buf[64];
1887         int r;
1888
1889         r = snprintf(buf, 64, "%ld\n",
1890                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
1891         if (r > 64)
1892                 r = 64;
1893         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1894 }
1895
1896 static ssize_t
1897 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
1898                       size_t cnt, loff_t *ppos)
1899 {
1900         long *ptr = filp->private_data;
1901         long val;
1902         char buf[64];
1903
1904         if (cnt > 63)
1905                 cnt = 63;
1906
1907         if (copy_from_user(&buf, ubuf, cnt))
1908                 return -EFAULT;
1909
1910         buf[cnt] = 0;
1911
1912         val = simple_strtoul(buf, NULL, 10);
1913
1914         *ptr = val * 1000;
1915
1916         return cnt;
1917 }
1918
1919 static atomic_t tracing_reader;
1920
1921 static int tracing_open_pipe(struct inode *inode, struct file *filp)
1922 {
1923         struct trace_iterator *iter;
1924
1925         if (tracing_disabled)
1926                 return -ENODEV;
1927
1928         /* We only allow for reader of the pipe */
1929         if (atomic_inc_return(&tracing_reader) != 1) {
1930                 atomic_dec(&tracing_reader);
1931                 return -EBUSY;
1932         }
1933
1934         /* create a buffer to store the information to pass to userspace */
1935         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1936         if (!iter)
1937                 return -ENOMEM;
1938
1939         iter->tr = &global_trace;
1940
1941         filp->private_data = iter;
1942
1943         return 0;
1944 }
1945
1946 static int tracing_release_pipe(struct inode *inode, struct file *file)
1947 {
1948         struct trace_iterator *iter = file->private_data;
1949
1950         kfree(iter);
1951         atomic_dec(&tracing_reader);
1952
1953         return 0;
1954 }
1955
1956 /*
1957  * Consumer reader.
1958  */
1959 static ssize_t
1960 tracing_read_pipe(struct file *filp, char __user *ubuf,
1961                   size_t cnt, loff_t *ppos)
1962 {
1963         struct trace_iterator *iter = filp->private_data;
1964         struct trace_array_cpu *data;
1965         static cpumask_t mask;
1966         static int start;
1967         unsigned long flags;
1968         int ftrace_save;
1969         int read = 0;
1970         int cpu;
1971         int len;
1972         int ret;
1973
1974         /* return any leftover data */
1975         if (iter->seq.len > start) {
1976                 len = iter->seq.len - start;
1977                 if (cnt > len)
1978                         cnt = len;
1979                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
1980                 if (ret)
1981                         cnt = -EFAULT;
1982
1983                 start += len;
1984
1985                 return cnt;
1986         }
1987
1988         trace_seq_reset(&iter->seq);
1989         start = 0;
1990
1991         while (trace_empty(iter)) {
1992                 /*
1993                  * This is a make-shift waitqueue. The reason we don't use
1994                  * an actual wait queue is because:
1995                  *  1) we only ever have one waiter
1996                  *  2) the tracing, traces all functions, we don't want
1997                  *     the overhead of calling wake_up and friends
1998                  *     (and tracing them too)
1999                  *     Anyway, this is really very primitive wakeup.
2000                  */
2001                 set_current_state(TASK_INTERRUPTIBLE);
2002                 iter->tr->waiter = current;
2003
2004                 /* sleep for one second, and try again. */
2005                 schedule_timeout(HZ);
2006
2007                 iter->tr->waiter = NULL;
2008
2009                 if (signal_pending(current))
2010                         return -EINTR;
2011
2012                 /*
2013                  * We block until we read something and tracing is disabled.
2014                  * We still block if tracing is disabled, but we have never
2015                  * read anything. This allows a user to cat this file, and
2016                  * then enable tracing. But after we have read something,
2017                  * we give an EOF when tracing is again disabled.
2018                  *
2019                  * iter->pos will be 0 if we haven't read anything.
2020                  */
2021                 if (!tracer_enabled && iter->pos)
2022                         break;
2023
2024                 continue;
2025         }
2026
2027         /* stop when tracing is finished */
2028         if (trace_empty(iter))
2029                 return 0;
2030
2031         if (cnt >= PAGE_SIZE)
2032                 cnt = PAGE_SIZE - 1;
2033
2034         memset(iter, 0, sizeof(*iter));
2035         iter->tr = &global_trace;
2036         iter->pos = -1;
2037
2038         /*
2039          * We need to stop all tracing on all CPUS to read the
2040          * the next buffer. This is a bit expensive, but is
2041          * not done often. We fill all what we can read,
2042          * and then release the locks again.
2043          */
2044
2045         cpus_clear(mask);
2046         local_irq_save(flags);
2047         ftrace_save = ftrace_enabled;
2048         ftrace_enabled = 0;
2049         smp_wmb();
2050         for_each_possible_cpu(cpu) {
2051                 data = iter->tr->data[cpu];
2052
2053                 if (!head_page(data) || !data->trace_idx)
2054                         continue;
2055
2056                 atomic_inc(&data->disabled);
2057                 cpu_set(cpu, mask);
2058         }
2059
2060         for_each_cpu_mask(cpu, mask) {
2061                 data = iter->tr->data[cpu];
2062                 spin_lock(&data->lock);
2063         }
2064
2065         while (find_next_entry_inc(iter) != NULL) {
2066                 int len = iter->seq.len;
2067
2068                 ret = print_trace_line(iter);
2069                 if (!ret) {
2070                         /* don't print partial lines */
2071                         iter->seq.len = len;
2072                         break;
2073                 }
2074
2075                 trace_consume(iter);
2076
2077                 if (iter->seq.len >= cnt)
2078                         break;
2079         }
2080
2081         for_each_cpu_mask(cpu, mask) {
2082                 data = iter->tr->data[cpu];
2083                 spin_unlock(&data->lock);
2084         }
2085
2086         for_each_cpu_mask(cpu, mask) {
2087                 data = iter->tr->data[cpu];
2088                 atomic_dec(&data->disabled);
2089         }
2090         ftrace_enabled = ftrace_save;
2091         local_irq_restore(flags);
2092
2093         /* Now copy what we have to the user */
2094         read = iter->seq.len;
2095         if (read > cnt)
2096                 read = cnt;
2097
2098         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2099
2100         if (read < iter->seq.len)
2101                 start = read;
2102         else
2103                 trace_seq_reset(&iter->seq);
2104
2105         if (ret)
2106                 read = -EFAULT;
2107
2108         return read;
2109 }
2110
2111 static struct file_operations tracing_max_lat_fops = {
2112         .open           = tracing_open_generic,
2113         .read           = tracing_max_lat_read,
2114         .write          = tracing_max_lat_write,
2115 };
2116
2117 static struct file_operations tracing_ctrl_fops = {
2118         .open           = tracing_open_generic,
2119         .read           = tracing_ctrl_read,
2120         .write          = tracing_ctrl_write,
2121 };
2122
2123 static struct file_operations set_tracer_fops = {
2124         .open           = tracing_open_generic,
2125         .read           = tracing_set_trace_read,
2126         .write          = tracing_set_trace_write,
2127 };
2128
2129 static struct file_operations tracing_pipe_fops = {
2130         .open           = tracing_open_pipe,
2131         .read           = tracing_read_pipe,
2132         .release        = tracing_release_pipe,
2133 };
2134
2135 #ifdef CONFIG_DYNAMIC_FTRACE
2136
2137 static ssize_t
2138 tracing_read_long(struct file *filp, char __user *ubuf,
2139                   size_t cnt, loff_t *ppos)
2140 {
2141         unsigned long *p = filp->private_data;
2142         char buf[64];
2143         int r;
2144
2145         r = sprintf(buf, "%ld\n", *p);
2146
2147         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2148 }
2149
2150 static struct file_operations tracing_read_long_fops = {
2151         .open           = tracing_open_generic,
2152         .read           = tracing_read_long,
2153 };
2154 #endif
2155
2156 static struct dentry *d_tracer;
2157
2158 struct dentry *tracing_init_dentry(void)
2159 {
2160         static int once;
2161
2162         if (d_tracer)
2163                 return d_tracer;
2164
2165         d_tracer = debugfs_create_dir("tracing", NULL);
2166
2167         if (!d_tracer && !once) {
2168                 once = 1;
2169                 pr_warning("Could not create debugfs directory 'tracing'\n");
2170                 return NULL;
2171         }
2172
2173         return d_tracer;
2174 }
2175
2176 #ifdef CONFIG_FTRACE_SELFTEST
2177 /* Let selftest have access to static functions in this file */
2178 #include "trace_selftest.c"
2179 #endif
2180
2181 static __init void tracer_init_debugfs(void)
2182 {
2183         struct dentry *d_tracer;
2184         struct dentry *entry;
2185
2186         d_tracer = tracing_init_dentry();
2187
2188         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2189                                     &global_trace, &tracing_ctrl_fops);
2190         if (!entry)
2191                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2192
2193         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2194                                     NULL, &tracing_iter_fops);
2195         if (!entry)
2196                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2197
2198         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2199                                     &global_trace, &tracing_lt_fops);
2200         if (!entry)
2201                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2202
2203         entry = debugfs_create_file("trace", 0444, d_tracer,
2204                                     &global_trace, &tracing_fops);
2205         if (!entry)
2206                 pr_warning("Could not create debugfs 'trace' entry\n");
2207
2208         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2209                                     &global_trace, &show_traces_fops);
2210         if (!entry)
2211                 pr_warning("Could not create debugfs 'trace' entry\n");
2212
2213         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2214                                     &global_trace, &set_tracer_fops);
2215         if (!entry)
2216                 pr_warning("Could not create debugfs 'trace' entry\n");
2217
2218         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2219                                     &tracing_max_latency,
2220                                     &tracing_max_lat_fops);
2221         if (!entry)
2222                 pr_warning("Could not create debugfs "
2223                            "'tracing_max_latency' entry\n");
2224
2225         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2226                                     &tracing_thresh, &tracing_max_lat_fops);
2227         if (!entry)
2228                 pr_warning("Could not create debugfs "
2229                            "'tracing_threash' entry\n");
2230         entry = debugfs_create_file("README", 0644, d_tracer,
2231                                     NULL, &tracing_readme_fops);
2232         if (!entry)
2233                 pr_warning("Could not create debugfs 'README' entry\n");
2234
2235         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2236                                     NULL, &tracing_pipe_fops);
2237         if (!entry)
2238                 pr_warning("Could not create debugfs "
2239                            "'tracing_threash' entry\n");
2240
2241 #ifdef CONFIG_DYNAMIC_FTRACE
2242         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2243                                     &ftrace_update_tot_cnt,
2244                                     &tracing_read_long_fops);
2245         if (!entry)
2246                 pr_warning("Could not create debugfs "
2247                            "'dyn_ftrace_total_info' entry\n");
2248 #endif
2249 }
2250
2251 /* dummy trace to disable tracing */
2252 static struct tracer no_tracer __read_mostly =
2253 {
2254         .name           = "none",
2255 };
2256
2257 static int trace_alloc_page(void)
2258 {
2259         struct trace_array_cpu *data;
2260         struct page *page, *tmp;
2261         LIST_HEAD(pages);
2262         void *array;
2263         int i;
2264
2265         /* first allocate a page for each CPU */
2266         for_each_possible_cpu(i) {
2267                 array = (void *)__get_free_page(GFP_KERNEL);
2268                 if (array == NULL) {
2269                         printk(KERN_ERR "tracer: failed to allocate page"
2270                                "for trace buffer!\n");
2271                         goto free_pages;
2272                 }
2273
2274                 page = virt_to_page(array);
2275                 list_add(&page->lru, &pages);
2276
2277 /* Only allocate if we are actually using the max trace */
2278 #ifdef CONFIG_TRACER_MAX_TRACE
2279                 array = (void *)__get_free_page(GFP_KERNEL);
2280                 if (array == NULL) {
2281                         printk(KERN_ERR "tracer: failed to allocate page"
2282                                "for trace buffer!\n");
2283                         goto free_pages;
2284                 }
2285                 page = virt_to_page(array);
2286                 list_add(&page->lru, &pages);
2287 #endif
2288         }
2289
2290         /* Now that we successfully allocate a page per CPU, add them */
2291         for_each_possible_cpu(i) {
2292                 data = global_trace.data[i];
2293                 spin_lock_init(&data->lock);
2294                 lockdep_set_class(&data->lock, &data->lock_key);
2295                 page = list_entry(pages.next, struct page, lru);
2296                 list_del_init(&page->lru);
2297                 list_add_tail(&page->lru, &data->trace_pages);
2298                 ClearPageLRU(page);
2299
2300 #ifdef CONFIG_TRACER_MAX_TRACE
2301                 data = max_tr.data[i];
2302                 spin_lock_init(&data->lock);
2303                 lockdep_set_class(&data->lock, &data->lock_key);
2304                 page = list_entry(pages.next, struct page, lru);
2305                 list_del_init(&page->lru);
2306                 list_add_tail(&page->lru, &data->trace_pages);
2307                 SetPageLRU(page);
2308 #endif
2309         }
2310         global_trace.entries += ENTRIES_PER_PAGE;
2311
2312         return 0;
2313
2314  free_pages:
2315         list_for_each_entry_safe(page, tmp, &pages, lru) {
2316                 list_del_init(&page->lru);
2317                 __free_page(page);
2318         }
2319         return -ENOMEM;
2320 }
2321
2322 __init static int tracer_alloc_buffers(void)
2323 {
2324         struct trace_array_cpu *data;
2325         void *array;
2326         struct page *page;
2327         int pages = 0;
2328         int ret = -ENOMEM;
2329         int i;
2330
2331         global_trace.ctrl = tracer_enabled;
2332
2333         /* Allocate the first page for all buffers */
2334         for_each_possible_cpu(i) {
2335                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2336                 max_tr.data[i] = &per_cpu(max_data, i);
2337
2338                 array = (void *)__get_free_page(GFP_KERNEL);
2339                 if (array == NULL) {
2340                         printk(KERN_ERR "tracer: failed to allocate page"
2341                                "for trace buffer!\n");
2342                         goto free_buffers;
2343                 }
2344
2345                 /* set the array to the list */
2346                 INIT_LIST_HEAD(&data->trace_pages);
2347                 page = virt_to_page(array);
2348                 list_add(&page->lru, &data->trace_pages);
2349                 /* use the LRU flag to differentiate the two buffers */
2350                 ClearPageLRU(page);
2351
2352 /* Only allocate if we are actually using the max trace */
2353 #ifdef CONFIG_TRACER_MAX_TRACE
2354                 array = (void *)__get_free_page(GFP_KERNEL);
2355                 if (array == NULL) {
2356                         printk(KERN_ERR "tracer: failed to allocate page"
2357                                "for trace buffer!\n");
2358                         goto free_buffers;
2359                 }
2360
2361                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2362                 page = virt_to_page(array);
2363                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2364                 SetPageLRU(page);
2365 #endif
2366         }
2367
2368         /*
2369          * Since we allocate by orders of pages, we may be able to
2370          * round up a bit.
2371          */
2372         global_trace.entries = ENTRIES_PER_PAGE;
2373         pages++;
2374
2375         while (global_trace.entries < trace_nr_entries) {
2376                 if (trace_alloc_page())
2377                         break;
2378                 pages++;
2379         }
2380         max_tr.entries = global_trace.entries;
2381
2382         pr_info("tracer: %d pages allocated for %ld",
2383                 pages, trace_nr_entries);
2384         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2385         pr_info("   actual entries %ld\n", global_trace.entries);
2386
2387         tracer_init_debugfs();
2388
2389         trace_init_cmdlines();
2390
2391         register_tracer(&no_tracer);
2392         current_trace = &no_tracer;
2393
2394         /* All seems OK, enable tracing */
2395         tracing_disabled = 0;
2396
2397         return 0;
2398
2399  free_buffers:
2400         for (i-- ; i >= 0; i--) {
2401                 struct page *page, *tmp;
2402                 struct trace_array_cpu *data = global_trace.data[i];
2403
2404                 if (data) {
2405                         list_for_each_entry_safe(page, tmp,
2406                                                  &data->trace_pages, lru) {
2407                                 list_del_init(&page->lru);
2408                                 __free_page(page);
2409                         }
2410                 }
2411
2412 #ifdef CONFIG_TRACER_MAX_TRACE
2413                 data = max_tr.data[i];
2414                 if (data) {
2415                         list_for_each_entry_safe(page, tmp,
2416                                                  &data->trace_pages, lru) {
2417                                 list_del_init(&page->lru);
2418                                 __free_page(page);
2419                         }
2420                 }
2421 #endif
2422         }
2423         return ret;
2424 }
2425 fs_initcall(tracer_alloc_buffers);