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

   1 /*
   2  * ring buffer based function tracer
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Originally taken from the RT patch by:
   8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
   9  *
  10  * Based on code from the latency_tracer, that is:
  11  *  Copyright (C) 2004-2006 Ingo Molnar
  12  *  Copyright (C) 2004 William Lee Irwin III
  13  */
  14 #include <linux/utsrelease.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/debugfs.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/linkage.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/ftrace.h>
  23 #include <linux/module.h>
  24 #include <linux/percpu.h>
  25 #include <linux/ctype.h>
  26 #include <linux/init.h>
  27 #include <linux/poll.h>
  28 #include <linux/gfp.h>
  29 #include <linux/fs.h>
  30
  31 #include <linux/stacktrace.h>
  32
  33 #include "trace.h"
  34
  35 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
  36 unsigned long __read_mostly     tracing_thresh;
  37
  38 static int tracing_disabled = 1;
  39
  40 static long
  41 ns2usecs(cycle_t nsec)
  42 {
  43         nsec += 500;
  44         do_div(nsec, 1000);
  45         return nsec;
  46 }
  47
  48 cycle_t ftrace_now(int cpu)
  49 {
  50         return cpu_clock(cpu);
  51 }
  52
  53 static struct trace_array       global_trace;
  54
  55 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
  56
  57 static struct trace_array       max_tr;
  58
  59 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
  60
  61 static int                      tracer_enabled = 1;
  62 static unsigned long            trace_nr_entries = 65536UL;
  63
  64 static struct tracer            *trace_types __read_mostly;
  65 static struct tracer            *current_trace __read_mostly;
  66 static int                      max_tracer_type_len;
  67
  68 static DEFINE_MUTEX(trace_types_lock);
  69 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
  70
  71 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
  72
  73 void trace_wake_up(void)
  74 {
  75         /*
  76          * The runqueue_is_locked() can fail, but this is the best we
  77          * have for now:
  78          */
  79         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
  80                 wake_up(&trace_wait);
  81 }
  82
  83 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
  84
  85 static int __init set_nr_entries(char *str)
  86 {
  87         if (!str)
  88                 return 0;
  89         trace_nr_entries = simple_strtoul(str, &str, 0);
  90         return 1;
  91 }
  92 __setup("trace_entries=", set_nr_entries);
  93
  94 unsigned long nsecs_to_usecs(unsigned long nsecs)
  95 {
  96         return nsecs / 1000;
  97 }
  98
  99 enum trace_type {
 100         __TRACE_FIRST_TYPE = 0,
 101
 102         TRACE_FN,
 103         TRACE_CTX,
 104         TRACE_WAKE,
 105         TRACE_STACK,
 106         TRACE_SPECIAL,
 107
 108         __TRACE_LAST_TYPE
 109 };
 110
 111 enum trace_flag_type {
 112         TRACE_FLAG_IRQS_OFF             = 0x01,
 113         TRACE_FLAG_NEED_RESCHED         = 0x02,
 114         TRACE_FLAG_HARDIRQ              = 0x04,
 115         TRACE_FLAG_SOFTIRQ              = 0x08,
 116 };
 117
 118 #define TRACE_ITER_SYM_MASK \
 119         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 120
 121 /* These must match the bit postions above */
 122 static const char *trace_options[] = {
 123         "print-parent",
 124         "sym-offset",
 125         "sym-addr",
 126         "verbose",
 127         "raw",
 128         "hex",
 129         "bin",
 130         "block",
 131         "stacktrace",
 132         "sched-tree",
 133         NULL
 134 };
 135
 136 static raw_spinlock_t ftrace_max_lock =
 137         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
 138
 139 /*
 140  * Copy the new maximum trace into the separate maximum-trace
 141  * structure. (this way the maximum trace is permanently saved,
 142  * for later retrieval via /debugfs/tracing/latency_trace)
 143  */
 144 static void
 145 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 146 {
 147         struct trace_array_cpu *data = tr->data[cpu];
 148
 149         max_tr.cpu = cpu;
 150         max_tr.time_start = data->preempt_timestamp;
 151
 152         data = max_tr.data[cpu];
 153         data->saved_latency = tracing_max_latency;
 154
 155         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
 156         data->pid = tsk->pid;
 157         data->uid = tsk->uid;
 158         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
 159         data->policy = tsk->policy;
 160         data->rt_priority = tsk->rt_priority;
 161
 162         /* record this tasks comm */
 163         tracing_record_cmdline(current);
 164 }
 165
 166 void check_pages(struct trace_array_cpu *data)
 167 {
 168         struct page *page, *tmp;
 169
 170         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
 171         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
 172
 173         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
 174                 BUG_ON(page->lru.next->prev != &page->lru);
 175                 BUG_ON(page->lru.prev->next != &page->lru);
 176         }
 177 }
 178
 179 void *head_page(struct trace_array_cpu *data)
 180 {
 181         struct page *page;
 182
 183         check_pages(data);
 184         if (list_empty(&data->trace_pages))
 185                 return NULL;
 186
 187         page = list_entry(data->trace_pages.next, struct page, lru);
 188         BUG_ON(&page->lru == &data->trace_pages);
 189
 190         return page_address(page);
 191 }
 192
 193 static int
 194 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
 195 {
 196         int len = (PAGE_SIZE - 1) - s->len;
 197         va_list ap;
 198         int ret;
 199
 200         if (!len)
 201                 return 0;
 202
 203         va_start(ap, fmt);
 204         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
 205         va_end(ap);
 206
 207         /* If we can't write it all, don't bother writing anything */
 208         if (ret > len)
 209                 return 0;
 210
 211         s->len += ret;
 212
 213         return len;
 214 }
 215
 216 static int
 217 trace_seq_puts(struct trace_seq *s, const char *str)
 218 {
 219         int len = strlen(str);
 220
 221         if (len > ((PAGE_SIZE - 1) - s->len))
 222                 return 0;
 223
 224         memcpy(s->buffer + s->len, str, len);
 225         s->len += len;
 226
 227         return len;
 228 }
 229
 230 static int
 231 trace_seq_putc(struct trace_seq *s, unsigned char c)
 232 {
 233         if (s->len >= (PAGE_SIZE - 1))
 234                 return 0;
 235
 236         s->buffer[s->len++] = c;
 237
 238         return 1;
 239 }
 240
 241 static int
 242 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
 243 {
 244         if (len > ((PAGE_SIZE - 1) - s->len))
 245                 return 0;
 246
 247         memcpy(s->buffer + s->len, mem, len);
 248         s->len += len;
 249
 250         return len;
 251 }
 252
 253 #define HEX_CHARS 17
 254
 255 static int
 256 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
 257 {
 258         unsigned char hex[HEX_CHARS];
 259         unsigned char *data;
 260         unsigned char byte;
 261         int i, j;
 262
 263         BUG_ON(len >= HEX_CHARS);
 264
 265         data = mem;
 266
 267 #ifdef __BIG_ENDIAN
 268         for (i = 0, j = 0; i < len; i++) {
 269 #else
 270         for (i = len-1, j = 0; i >= 0; i--) {
 271 #endif
 272                 byte = data[i];
 273
 274                 hex[j]   = byte & 0x0f;
 275                 if (hex[j] >= 10)
 276                         hex[j] += 'a' - 10;
 277                 else
 278                         hex[j] += '0';
 279                 j++;
 280
 281                 hex[j] = byte >> 4;
 282                 if (hex[j] >= 10)
 283                         hex[j] += 'a' - 10;
 284                 else
 285                         hex[j] += '0';
 286                 j++;
 287         }
 288         hex[j] = ' ';
 289         j++;
 290
 291         return trace_seq_putmem(s, hex, j);
 292 }
 293
 294 static void
 295 trace_seq_reset(struct trace_seq *s)
 296 {
 297         s->len = 0;
 298 }
 299
 300 static void
 301 trace_print_seq(struct seq_file *m, struct trace_seq *s)
 302 {
 303         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
 304
 305         s->buffer[len] = 0;
 306         seq_puts(m, s->buffer);
 307
 308         trace_seq_reset(s);
 309 }
 310
 311 static void
 312 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
 313 {
 314         struct list_head flip_pages;
 315
 316         INIT_LIST_HEAD(&flip_pages);
 317
 318         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
 319                 sizeof(struct trace_array_cpu) -
 320                 offsetof(struct trace_array_cpu, trace_head_idx));
 321
 322         check_pages(tr1);
 323         check_pages(tr2);
 324         list_splice_init(&tr1->trace_pages, &flip_pages);
 325         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
 326         list_splice_init(&flip_pages, &tr2->trace_pages);
 327         BUG_ON(!list_empty(&flip_pages));
 328         check_pages(tr1);
 329         check_pages(tr2);
 330 }
 331
 332 void
 333 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
 334 {
 335         struct trace_array_cpu *data;
 336         int i;
 337
 338         WARN_ON_ONCE(!irqs_disabled());
 339         __raw_spin_lock(&ftrace_max_lock);
 340         /* clear out all the previous traces */
 341         for_each_possible_cpu(i) {
 342                 data = tr->data[i];
 343                 flip_trace(max_tr.data[i], data);
 344                 tracing_reset(data);
 345         }
 346
 347         __update_max_tr(tr, tsk, cpu);
 348         __raw_spin_unlock(&ftrace_max_lock);
 349 }
 350
 351 /**
 352  * update_max_tr_single - only copy one trace over, and reset the rest
 353  * @tr - tracer
 354  * @tsk - task with the latency
 355  * @cpu - the cpu of the buffer to copy.
 356  */
 357 void
 358 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
 359 {
 360         struct trace_array_cpu *data = tr->data[cpu];
 361         int i;
 362
 363         WARN_ON_ONCE(!irqs_disabled());
 364         __raw_spin_lock(&ftrace_max_lock);
 365         for_each_possible_cpu(i)
 366                 tracing_reset(max_tr.data[i]);
 367
 368         flip_trace(max_tr.data[cpu], data);
 369         tracing_reset(data);
 370
 371         __update_max_tr(tr, tsk, cpu);
 372         __raw_spin_unlock(&ftrace_max_lock);
 373 }
 374
 375 int register_tracer(struct tracer *type)
 376 {
 377         struct tracer *t;
 378         int len;
 379         int ret = 0;
 380
 381         if (!type->name) {
 382                 pr_info("Tracer must have a name\n");
 383                 return -1;
 384         }
 385
 386         mutex_lock(&trace_types_lock);
 387         for (t = trace_types; t; t = t->next) {
 388                 if (strcmp(type->name, t->name) == 0) {
 389                         /* already found */
 390                         pr_info("Trace %s already registered\n",
 391                                 type->name);
 392                         ret = -1;
 393                         goto out;
 394                 }
 395         }
 396
 397 #ifdef CONFIG_FTRACE_STARTUP_TEST
 398         if (type->selftest) {
 399                 struct tracer *saved_tracer = current_trace;
 400                 struct trace_array_cpu *data;
 401                 struct trace_array *tr = &global_trace;
 402                 int saved_ctrl = tr->ctrl;
 403                 int i;
 404                 /*
 405                  * Run a selftest on this tracer.
 406                  * Here we reset the trace buffer, and set the current
 407                  * tracer to be this tracer. The tracer can then run some
 408                  * internal tracing to verify that everything is in order.
 409                  * If we fail, we do not register this tracer.
 410                  */
 411                 for_each_possible_cpu(i) {
 412                         data = tr->data[i];
 413                         if (!head_page(data))
 414                                 continue;
 415                         tracing_reset(data);
 416                 }
 417                 current_trace = type;
 418                 tr->ctrl = 0;
 419                 /* the test is responsible for initializing and enabling */
 420                 pr_info("Testing tracer %s: ", type->name);
 421                 ret = type->selftest(type, tr);
 422                 /* the test is responsible for resetting too */
 423                 current_trace = saved_tracer;
 424                 tr->ctrl = saved_ctrl;
 425                 if (ret) {
 426                         printk(KERN_CONT "FAILED!\n");
 427                         goto out;
 428                 }
 429                 /* Only reset on passing, to avoid touching corrupted buffers */
 430                 for_each_possible_cpu(i) {
 431                         data = tr->data[i];
 432                         if (!head_page(data))
 433                                 continue;
 434                         tracing_reset(data);
 435                 }
 436                 printk(KERN_CONT "PASSED\n");
 437         }
 438 #endif
 439
 440         type->next = trace_types;
 441         trace_types = type;
 442         len = strlen(type->name);
 443         if (len > max_tracer_type_len)
 444                 max_tracer_type_len = len;
 445
 446  out:
 447         mutex_unlock(&trace_types_lock);
 448
 449         return ret;
 450 }
 451
 452 void unregister_tracer(struct tracer *type)
 453 {
 454         struct tracer **t;
 455         int len;
 456
 457         mutex_lock(&trace_types_lock);
 458         for (t = &trace_types; *t; t = &(*t)->next) {
 459                 if (*t == type)
 460                         goto found;
 461         }
 462         pr_info("Trace %s not registered\n", type->name);
 463         goto out;
 464
 465  found:
 466         *t = (*t)->next;
 467         if (strlen(type->name) != max_tracer_type_len)
 468                 goto out;
 469
 470         max_tracer_type_len = 0;
 471         for (t = &trace_types; *t; t = &(*t)->next) {
 472                 len = strlen((*t)->name);
 473                 if (len > max_tracer_type_len)
 474                         max_tracer_type_len = len;
 475         }
 476  out:
 477         mutex_unlock(&trace_types_lock);
 478 }
 479
 480 void tracing_reset(struct trace_array_cpu *data)
 481 {
 482         data->trace_idx = 0;
 483         data->trace_head = data->trace_tail = head_page(data);
 484         data->trace_head_idx = 0;
 485         data->trace_tail_idx = 0;
 486 }
 487
 488 #define SAVED_CMDLINES 128
 489 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
 490 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
 491 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
 492 static int cmdline_idx;
 493 static DEFINE_SPINLOCK(trace_cmdline_lock);
 494 atomic_t trace_record_cmdline_disabled;
 495
 496 static void trace_init_cmdlines(void)
 497 {
 498         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
 499         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
 500         cmdline_idx = 0;
 501 }
 502
 503 void trace_stop_cmdline_recording(void);
 504
 505 static void trace_save_cmdline(struct task_struct *tsk)
 506 {
 507         unsigned map;
 508         unsigned idx;
 509
 510         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
 511                 return;
 512
 513         /*
 514          * It's not the end of the world if we don't get
 515          * the lock, but we also don't want to spin
 516          * nor do we want to disable interrupts,
 517          * so if we miss here, then better luck next time.
 518          */
 519         if (!spin_trylock(&trace_cmdline_lock))
 520                 return;
 521
 522         idx = map_pid_to_cmdline[tsk->pid];
 523         if (idx >= SAVED_CMDLINES) {
 524                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
 525
 526                 map = map_cmdline_to_pid[idx];
 527                 if (map <= PID_MAX_DEFAULT)
 528                         map_pid_to_cmdline[map] = (unsigned)-1;
 529
 530                 map_pid_to_cmdline[tsk->pid] = idx;
 531
 532                 cmdline_idx = idx;
 533         }
 534
 535         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
 536
 537         spin_unlock(&trace_cmdline_lock);
 538 }
 539
 540 static char *trace_find_cmdline(int pid)
 541 {
 542         char *cmdline = "<...>";
 543         unsigned map;
 544
 545         if (!pid)
 546                 return "<idle>";
 547
 548         if (pid > PID_MAX_DEFAULT)
 549                 goto out;
 550
 551         map = map_pid_to_cmdline[pid];
 552         if (map >= SAVED_CMDLINES)
 553                 goto out;
 554
 555         cmdline = saved_cmdlines[map];
 556
 557  out:
 558         return cmdline;
 559 }
 560
 561 void tracing_record_cmdline(struct task_struct *tsk)
 562 {
 563         if (atomic_read(&trace_record_cmdline_disabled))
 564                 return;
 565
 566         trace_save_cmdline(tsk);
 567 }
 568
 569 static inline struct list_head *
 570 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
 571 {
 572         /*
 573          * Roundrobin - but skip the head (which is not a real page):
 574          */
 575         next = next->next;
 576         if (unlikely(next == &data->trace_pages))
 577                 next = next->next;
 578         BUG_ON(next == &data->trace_pages);
 579
 580         return next;
 581 }
 582
 583 static inline void *
 584 trace_next_page(struct trace_array_cpu *data, void *addr)
 585 {
 586         struct list_head *next;
 587         struct page *page;
 588
 589         page = virt_to_page(addr);
 590
 591         next = trace_next_list(data, &page->lru);
 592         page = list_entry(next, struct page, lru);
 593
 594         return page_address(page);
 595 }
 596
 597 static inline struct trace_entry *
 598 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
 599 {
 600         unsigned long idx, idx_next;
 601         struct trace_entry *entry;
 602
 603         data->trace_idx++;
 604         idx = data->trace_head_idx;
 605         idx_next = idx + 1;
 606
 607         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
 608
 609         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
 610
 611         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
 612                 data->trace_head = trace_next_page(data, data->trace_head);
 613                 idx_next = 0;
 614         }
 615
 616         if (data->trace_head == data->trace_tail &&
 617             idx_next == data->trace_tail_idx) {
 618                 /* overrun */
 619                 data->trace_tail_idx++;
 620                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 621                         data->trace_tail =
 622                                 trace_next_page(data, data->trace_tail);
 623                         data->trace_tail_idx = 0;
 624                 }
 625         }
 626
 627         data->trace_head_idx = idx_next;
 628
 629         return entry;
 630 }
 631
 632 static inline void
 633 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
 634 {
 635         struct task_struct *tsk = current;
 636         unsigned long pc;
 637
 638         pc = preempt_count();
 639
 640         entry->preempt_count    = pc & 0xff;
 641         entry->pid              = tsk->pid;
 642         entry->t                = ftrace_now(raw_smp_processor_id());
 643         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
 644                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
 645                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 646                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 647 }
 648
 649 void
 650 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
 651                unsigned long ip, unsigned long parent_ip, unsigned long flags)
 652 {
 653         struct trace_entry *entry;
 654         unsigned long irq_flags;
 655
 656         raw_local_irq_save(irq_flags);
 657         __raw_spin_lock(&data->lock);
 658         entry                   = tracing_get_trace_entry(tr, data);
 659         tracing_generic_entry_update(entry, flags);
 660         entry->type             = TRACE_FN;
 661         entry->fn.ip            = ip;
 662         entry->fn.parent_ip     = parent_ip;
 663         __raw_spin_unlock(&data->lock);
 664         raw_local_irq_restore(irq_flags);
 665 }
 666
 667 void
 668 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
 669        unsigned long ip, unsigned long parent_ip, unsigned long flags)
 670 {
 671         if (likely(!atomic_read(&data->disabled)))
 672                 trace_function(tr, data, ip, parent_ip, flags);
 673 }
 674
 675 void
 676 __trace_special(void *__tr, void *__data,
 677                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
 678 {
 679         struct trace_array_cpu *data = __data;
 680         struct trace_array *tr = __tr;
 681         struct trace_entry *entry;
 682         unsigned long irq_flags;
 683
 684         raw_local_irq_save(irq_flags);
 685         __raw_spin_lock(&data->lock);
 686         entry                   = tracing_get_trace_entry(tr, data);
 687         tracing_generic_entry_update(entry, 0);
 688         entry->type             = TRACE_SPECIAL;
 689         entry->special.arg1     = arg1;
 690         entry->special.arg2     = arg2;
 691         entry->special.arg3     = arg3;
 692         __raw_spin_unlock(&data->lock);
 693         raw_local_irq_restore(irq_flags);
 694
 695         trace_wake_up();
 696 }
 697
 698 void __trace_stack(struct trace_array *tr,
 699                    struct trace_array_cpu *data,
 700                    unsigned long flags,
 701                    int skip)
 702 {
 703         struct trace_entry *entry;
 704         struct stack_trace trace;
 705
 706         if (!(trace_flags & TRACE_ITER_STACKTRACE))
 707                 return;
 708
 709         entry                   = tracing_get_trace_entry(tr, data);
 710         tracing_generic_entry_update(entry, flags);
 711         entry->type             = TRACE_STACK;
 712
 713         memset(&entry->stack, 0, sizeof(entry->stack));
 714
 715         trace.nr_entries        = 0;
 716         trace.max_entries       = FTRACE_STACK_ENTRIES;
 717         trace.skip              = skip;
 718         trace.entries           = entry->stack.caller;
 719
 720         save_stack_trace(&trace);
 721 }
 722
 723 void
 724 tracing_sched_switch_trace(struct trace_array *tr,
 725                            struct trace_array_cpu *data,
 726                            struct task_struct *prev,
 727                            struct task_struct *next,
 728                            unsigned long flags)
 729 {
 730         struct trace_entry *entry;
 731         unsigned long irq_flags;
 732
 733         raw_local_irq_save(irq_flags);
 734         __raw_spin_lock(&data->lock);
 735         entry                   = tracing_get_trace_entry(tr, data);
 736         tracing_generic_entry_update(entry, flags);
 737         entry->type             = TRACE_CTX;
 738         entry->ctx.prev_pid     = prev->pid;
 739         entry->ctx.prev_prio    = prev->prio;
 740         entry->ctx.prev_state   = prev->state;
 741         entry->ctx.next_pid     = next->pid;
 742         entry->ctx.next_prio    = next->prio;
 743         entry->ctx.next_state   = next->state;
 744         __trace_stack(tr, data, flags, 4);
 745         __raw_spin_unlock(&data->lock);
 746         raw_local_irq_restore(irq_flags);
 747 }
 748
 749 void
 750 tracing_sched_wakeup_trace(struct trace_array *tr,
 751                            struct trace_array_cpu *data,
 752                            struct task_struct *wakee,
 753                            struct task_struct *curr,
 754                            unsigned long flags)
 755 {
 756         struct trace_entry *entry;
 757         unsigned long irq_flags;
 758
 759         raw_local_irq_save(irq_flags);
 760         __raw_spin_lock(&data->lock);
 761         entry                   = tracing_get_trace_entry(tr, data);
 762         tracing_generic_entry_update(entry, flags);
 763         entry->type             = TRACE_WAKE;
 764         entry->ctx.prev_pid     = curr->pid;
 765         entry->ctx.prev_prio    = curr->prio;
 766         entry->ctx.prev_state   = curr->state;
 767         entry->ctx.next_pid     = wakee->pid;
 768         entry->ctx.next_prio    = wakee->prio;
 769         entry->ctx.next_state   = wakee->state;
 770         __trace_stack(tr, data, flags, 5);
 771         __raw_spin_unlock(&data->lock);
 772         raw_local_irq_restore(irq_flags);
 773
 774         trace_wake_up();
 775 }
 776
 777 #ifdef CONFIG_FTRACE
 778 static void
 779 function_trace_call(unsigned long ip, unsigned long parent_ip)
 780 {
 781         struct trace_array *tr = &global_trace;
 782         struct trace_array_cpu *data;
 783         unsigned long flags;
 784         long disabled;
 785         int cpu;
 786
 787         if (unlikely(!tracer_enabled))
 788                 return;
 789
 790         local_irq_save(flags);
 791         cpu = raw_smp_processor_id();
 792         data = tr->data[cpu];
 793         disabled = atomic_inc_return(&data->disabled);
 794
 795         if (likely(disabled == 1))
 796                 trace_function(tr, data, ip, parent_ip, flags);
 797
 798         atomic_dec(&data->disabled);
 799         local_irq_restore(flags);
 800 }
 801
 802 static struct ftrace_ops trace_ops __read_mostly =
 803 {
 804         .func = function_trace_call,
 805 };
 806
 807 void tracing_start_function_trace(void)
 808 {
 809         register_ftrace_function(&trace_ops);
 810 }
 811
 812 void tracing_stop_function_trace(void)
 813 {
 814         unregister_ftrace_function(&trace_ops);
 815 }
 816 #endif
 817
 818 enum trace_file_type {
 819         TRACE_FILE_LAT_FMT      = 1,
 820 };
 821
 822 static struct trace_entry *
 823 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
 824                 struct trace_iterator *iter, int cpu)
 825 {
 826         struct page *page;
 827         struct trace_entry *array;
 828
 829         if (iter->next_idx[cpu] >= tr->entries ||
 830             iter->next_idx[cpu] >= data->trace_idx ||
 831             (data->trace_head == data->trace_tail &&
 832              data->trace_head_idx == data->trace_tail_idx))
 833                 return NULL;
 834
 835         if (!iter->next_page[cpu]) {
 836                 /* Initialize the iterator for this cpu trace buffer */
 837                 WARN_ON(!data->trace_tail);
 838                 page = virt_to_page(data->trace_tail);
 839                 iter->next_page[cpu] = &page->lru;
 840                 iter->next_page_idx[cpu] = data->trace_tail_idx;
 841         }
 842
 843         page = list_entry(iter->next_page[cpu], struct page, lru);
 844         BUG_ON(&data->trace_pages == &page->lru);
 845
 846         array = page_address(page);
 847
 848         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
 849         return &array[iter->next_page_idx[cpu]];
 850 }
 851
 852 static struct trace_entry *
 853 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
 854 {
 855         struct trace_array *tr = iter->tr;
 856         struct trace_entry *ent, *next = NULL;
 857         int next_cpu = -1;
 858         int cpu;
 859
 860         for_each_possible_cpu(cpu) {
 861                 if (!head_page(tr->data[cpu]))
 862                         continue;
 863                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
 864                 /*
 865                  * Pick the entry with the smallest timestamp:
 866                  */
 867                 if (ent && (!next || ent->t < next->t)) {
 868                         next = ent;
 869                         next_cpu = cpu;
 870                 }
 871         }
 872
 873         if (ent_cpu)
 874                 *ent_cpu = next_cpu;
 875
 876         return next;
 877 }
 878
 879 static void trace_iterator_increment(struct trace_iterator *iter)
 880 {
 881         iter->idx++;
 882         iter->next_idx[iter->cpu]++;
 883         iter->next_page_idx[iter->cpu]++;
 884
 885         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
 886                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 887
 888                 iter->next_page_idx[iter->cpu] = 0;
 889                 iter->next_page[iter->cpu] =
 890                         trace_next_list(data, iter->next_page[iter->cpu]);
 891         }
 892 }
 893
 894 static void trace_consume(struct trace_iterator *iter)
 895 {
 896         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
 897
 898         data->trace_tail_idx++;
 899         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
 900                 data->trace_tail = trace_next_page(data, data->trace_tail);
 901                 data->trace_tail_idx = 0;
 902         }
 903
 904         /* Check if we empty it, then reset the index */
 905         if (data->trace_head == data->trace_tail &&
 906             data->trace_head_idx == data->trace_tail_idx)
 907                 data->trace_idx = 0;
 908 }
 909
 910 static void *find_next_entry_inc(struct trace_iterator *iter)
 911 {
 912         struct trace_entry *next;
 913         int next_cpu = -1;
 914
 915         next = find_next_entry(iter, &next_cpu);
 916
 917         iter->prev_ent = iter->ent;
 918         iter->prev_cpu = iter->cpu;
 919
 920         iter->ent = next;
 921         iter->cpu = next_cpu;
 922
 923         if (next)
 924                 trace_iterator_increment(iter);
 925
 926         return next ? iter : NULL;
 927 }
 928
 929 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
 930 {
 931         struct trace_iterator *iter = m->private;
 932         void *last_ent = iter->ent;
 933         int i = (int)*pos;
 934         void *ent;
 935
 936         (*pos)++;
 937
 938         /* can't go backwards */
 939         if (iter->idx > i)
 940                 return NULL;
 941
 942         if (iter->idx < 0)
 943                 ent = find_next_entry_inc(iter);
 944         else
 945                 ent = iter;
 946
 947         while (ent && iter->idx < i)
 948                 ent = find_next_entry_inc(iter);
 949
 950         iter->pos = *pos;
 951
 952         if (last_ent && !ent)
 953                 seq_puts(m, "\n\nvim:ft=help\n");
 954
 955         return ent;
 956 }
 957
 958 static void *s_start(struct seq_file *m, loff_t *pos)
 959 {
 960         struct trace_iterator *iter = m->private;
 961         void *p = NULL;
 962         loff_t l = 0;
 963         int i;
 964
 965         mutex_lock(&trace_types_lock);
 966
 967         if (!current_trace || current_trace != iter->trace) {
 968                 mutex_unlock(&trace_types_lock);
 969                 return NULL;
 970         }
 971
 972         atomic_inc(&trace_record_cmdline_disabled);
 973
 974         /* let the tracer grab locks here if needed */
 975         if (current_trace->start)
 976                 current_trace->start(iter);
 977
 978         if (*pos != iter->pos) {
 979                 iter->ent = NULL;
 980                 iter->cpu = 0;
 981                 iter->idx = -1;
 982                 iter->prev_ent = NULL;
 983                 iter->prev_cpu = -1;
 984
 985                 for_each_possible_cpu(i) {
 986                         iter->next_idx[i] = 0;
 987                         iter->next_page[i] = NULL;
 988                 }
 989
 990                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
 991                         ;
 992
 993         } else {
 994                 l = *pos - 1;
 995                 p = s_next(m, p, &l);
 996         }
 997
 998         return p;
 999 }
1000
1001 static void s_stop(struct seq_file *m, void *p)
1002 {
1003         struct trace_iterator *iter = m->private;
1004
1005         atomic_dec(&trace_record_cmdline_disabled);
1006
1007         /* let the tracer release locks here if needed */
1008         if (current_trace && current_trace == iter->trace && iter->trace->stop)
1009                 iter->trace->stop(iter);
1010
1011         mutex_unlock(&trace_types_lock);
1012 }
1013
1014 static int
1015 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1016 {
1017 #ifdef CONFIG_KALLSYMS
1018         char str[KSYM_SYMBOL_LEN];
1019
1020         kallsyms_lookup(address, NULL, NULL, NULL, str);
1021
1022         return trace_seq_printf(s, fmt, str);
1023 #endif
1024         return 1;
1025 }
1026
1027 static int
1028 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1029                      unsigned long address)
1030 {
1031 #ifdef CONFIG_KALLSYMS
1032         char str[KSYM_SYMBOL_LEN];
1033
1034         sprint_symbol(str, address);
1035         return trace_seq_printf(s, fmt, str);
1036 #endif
1037         return 1;
1038 }
1039
1040 #ifndef CONFIG_64BIT
1041 # define IP_FMT "%08lx"
1042 #else
1043 # define IP_FMT "%016lx"
1044 #endif
1045
1046 static int
1047 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1048 {
1049         int ret;
1050
1051         if (!ip)
1052                 return trace_seq_printf(s, "0");
1053
1054         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1055                 ret = seq_print_sym_offset(s, "%s", ip);
1056         else
1057                 ret = seq_print_sym_short(s, "%s", ip);
1058
1059         if (!ret)
1060                 return 0;
1061
1062         if (sym_flags & TRACE_ITER_SYM_ADDR)
1063                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1064         return ret;
1065 }
1066
1067 static void print_lat_help_header(struct seq_file *m)
1068 {
1069         seq_puts(m, "#                _------=> CPU#            \n");
1070         seq_puts(m, "#               / _-----=> irqs-off        \n");
1071         seq_puts(m, "#              | / _----=> need-resched    \n");
1072         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1073         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1074         seq_puts(m, "#              |||| /                      \n");
1075         seq_puts(m, "#              |||||     delay             \n");
1076         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1077         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1078 }
1079
1080 static void print_func_help_header(struct seq_file *m)
1081 {
1082         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1083         seq_puts(m, "#              | |      |          |         |\n");
1084 }
1085
1086
1087 static void
1088 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1089 {
1090         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1091         struct trace_array *tr = iter->tr;
1092         struct trace_array_cpu *data = tr->data[tr->cpu];
1093         struct tracer *type = current_trace;
1094         unsigned long total   = 0;
1095         unsigned long entries = 0;
1096         int cpu;
1097         const char *name = "preemption";
1098
1099         if (type)
1100                 name = type->name;
1101
1102         for_each_possible_cpu(cpu) {
1103                 if (head_page(tr->data[cpu])) {
1104                         total += tr->data[cpu]->trace_idx;
1105                         if (tr->data[cpu]->trace_idx > tr->entries)
1106                                 entries += tr->entries;
1107                         else
1108                                 entries += tr->data[cpu]->trace_idx;
1109                 }
1110         }
1111
1112         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1113                    name, UTS_RELEASE);
1114         seq_puts(m, "-----------------------------------"
1115                  "---------------------------------\n");
1116         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1117                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1118                    nsecs_to_usecs(data->saved_latency),
1119                    entries,
1120                    total,
1121                    tr->cpu,
1122 #if defined(CONFIG_PREEMPT_NONE)
1123                    "server",
1124 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1125                    "desktop",
1126 #elif defined(CONFIG_PREEMPT_DESKTOP)
1127                    "preempt",
1128 #else
1129                    "unknown",
1130 #endif
1131                    /* These are reserved for later use */
1132                    0, 0, 0, 0);
1133 #ifdef CONFIG_SMP
1134         seq_printf(m, " #P:%d)\n", num_online_cpus());
1135 #else
1136         seq_puts(m, ")\n");
1137 #endif
1138         seq_puts(m, "    -----------------\n");
1139         seq_printf(m, "    | task: %.16s-%d "
1140                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1141                    data->comm, data->pid, data->uid, data->nice,
1142                    data->policy, data->rt_priority);
1143         seq_puts(m, "    -----------------\n");
1144
1145         if (data->critical_start) {
1146                 seq_puts(m, " => started at: ");
1147                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1148                 trace_print_seq(m, &iter->seq);
1149                 seq_puts(m, "\n => ended at:   ");
1150                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1151                 trace_print_seq(m, &iter->seq);
1152                 seq_puts(m, "\n");
1153         }
1154
1155         seq_puts(m, "\n");
1156 }
1157
1158 static void
1159 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1160 {
1161         int hardirq, softirq;
1162         char *comm;
1163
1164         comm = trace_find_cmdline(entry->pid);
1165
1166         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1167         trace_seq_printf(s, "%d", cpu);
1168         trace_seq_printf(s, "%c%c",
1169                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1170                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1171
1172         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1173         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1174         if (hardirq && softirq)
1175                 trace_seq_putc(s, 'H');
1176         else {
1177                 if (hardirq)
1178                         trace_seq_putc(s, 'h');
1179                 else {
1180                         if (softirq)
1181                                 trace_seq_putc(s, 's');
1182                         else
1183                                 trace_seq_putc(s, '.');
1184                 }
1185         }
1186
1187         if (entry->preempt_count)
1188                 trace_seq_printf(s, "%x", entry->preempt_count);
1189         else
1190                 trace_seq_puts(s, ".");
1191 }
1192
1193 unsigned long preempt_mark_thresh = 100;
1194
1195 static void
1196 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1197                     unsigned long rel_usecs)
1198 {
1199         trace_seq_printf(s, " %4lldus", abs_usecs);
1200         if (rel_usecs > preempt_mark_thresh)
1201                 trace_seq_puts(s, "!: ");
1202         else if (rel_usecs > 1)
1203                 trace_seq_puts(s, "+: ");
1204         else
1205                 trace_seq_puts(s, " : ");
1206 }
1207
1208 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1209
1210 static int
1211 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1212 {
1213         struct trace_seq *s = &iter->seq;
1214         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1215         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1216         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1217         struct trace_entry *entry = iter->ent;
1218         unsigned long abs_usecs;
1219         unsigned long rel_usecs;
1220         char *comm;
1221         int S, T;
1222         int i;
1223
1224         if (!next_entry)
1225                 next_entry = entry;
1226         rel_usecs = ns2usecs(next_entry->t - entry->t);
1227         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1228
1229         if (verbose) {
1230                 comm = trace_find_cmdline(entry->pid);
1231                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1232                                  " %ld.%03ldms (+%ld.%03ldms): ",
1233                                  comm,
1234                                  entry->pid, cpu, entry->flags,
1235                                  entry->preempt_count, trace_idx,
1236                                  ns2usecs(entry->t),
1237                                  abs_usecs/1000,
1238                                  abs_usecs % 1000, rel_usecs/1000,
1239                                  rel_usecs % 1000);
1240         } else {
1241                 lat_print_generic(s, entry, cpu);
1242                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1243         }
1244         switch (entry->type) {
1245         case TRACE_FN:
1246                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1247                 trace_seq_puts(s, " (");
1248                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1249                 trace_seq_puts(s, ")\n");
1250                 break;
1251         case TRACE_CTX:
1252         case TRACE_WAKE:
1253                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1254                         state_to_char[entry->ctx.prev_state] : 'X';
1255                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1256                         state_to_char[entry->ctx.next_state] : 'X';
1257
1258                 comm = trace_find_cmdline(entry->ctx.next_pid);
1259                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1260                                  entry->ctx.prev_pid,
1261                                  entry->ctx.prev_prio,
1262                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1263                                  entry->ctx.next_pid,
1264                                  entry->ctx.next_prio,
1265                                  T, comm);
1266                 break;
1267         case TRACE_SPECIAL:
1268                 trace_seq_printf(s, "# %ld %ld %ld\n",
1269                                  entry->special.arg1,
1270                                  entry->special.arg2,
1271                                  entry->special.arg3);
1272                 break;
1273         case TRACE_STACK:
1274                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1275                         if (i)
1276                                 trace_seq_puts(s, " <= ");
1277                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1278                 }
1279                 trace_seq_puts(s, "\n");
1280                 break;
1281         default:
1282                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1283         }
1284         return 1;
1285 }
1286
1287 static int print_trace_fmt(struct trace_iterator *iter)
1288 {
1289         struct trace_seq *s = &iter->seq;
1290         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1291         struct trace_entry *entry;
1292         unsigned long usec_rem;
1293         unsigned long long t;
1294         unsigned long secs;
1295         char *comm;
1296         int ret;
1297         int S, T;
1298         int i;
1299
1300         entry = iter->ent;
1301
1302         comm = trace_find_cmdline(iter->ent->pid);
1303
1304         t = ns2usecs(entry->t);
1305         usec_rem = do_div(t, 1000000ULL);
1306         secs = (unsigned long)t;
1307
1308         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1309         if (!ret)
1310                 return 0;
1311         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1312         if (!ret)
1313                 return 0;
1314         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1315         if (!ret)
1316                 return 0;
1317
1318         switch (entry->type) {
1319         case TRACE_FN:
1320                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1321                 if (!ret)
1322                         return 0;
1323                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1324                                                 entry->fn.parent_ip) {
1325                         ret = trace_seq_printf(s, " <-");
1326                         if (!ret)
1327                                 return 0;
1328                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1329                                                sym_flags);
1330                         if (!ret)
1331                                 return 0;
1332                 }
1333                 ret = trace_seq_printf(s, "\n");
1334                 if (!ret)
1335                         return 0;
1336                 break;
1337         case TRACE_CTX:
1338         case TRACE_WAKE:
1339                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1340                         state_to_char[entry->ctx.prev_state] : 'X';
1341                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1342                         state_to_char[entry->ctx.next_state] : 'X';
1343                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1344                                        entry->ctx.prev_pid,
1345                                        entry->ctx.prev_prio,
1346                                        S,
1347                                        entry->type == TRACE_CTX ? "==>" : "  +",
1348                                        entry->ctx.next_pid,
1349                                        entry->ctx.next_prio,
1350                                        T);
1351                 if (!ret)
1352                         return 0;
1353                 break;
1354         case TRACE_SPECIAL:
1355                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1356                                  entry->special.arg1,
1357                                  entry->special.arg2,
1358                                  entry->special.arg3);
1359                 if (!ret)
1360                         return 0;
1361                 break;
1362         case TRACE_STACK:
1363                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1364                         if (i) {
1365                                 ret = trace_seq_puts(s, " <= ");
1366                                 if (!ret)
1367                                         return 0;
1368                         }
1369                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1370                                                sym_flags);
1371                         if (!ret)
1372                                 return 0;
1373                 }
1374                 ret = trace_seq_puts(s, "\n");
1375                 if (!ret)
1376                         return 0;
1377                 break;
1378         }
1379         return 1;
1380 }
1381
1382 static int print_raw_fmt(struct trace_iterator *iter)
1383 {
1384         struct trace_seq *s = &iter->seq;
1385         struct trace_entry *entry;
1386         int ret;
1387         int S, T;
1388
1389         entry = iter->ent;
1390
1391         ret = trace_seq_printf(s, "%d %d %llu ",
1392                 entry->pid, iter->cpu, entry->t);
1393         if (!ret)
1394                 return 0;
1395
1396         switch (entry->type) {
1397         case TRACE_FN:
1398                 ret = trace_seq_printf(s, "%x %x\n",
1399                                         entry->fn.ip, entry->fn.parent_ip);
1400                 if (!ret)
1401                         return 0;
1402                 break;
1403         case TRACE_CTX:
1404         case TRACE_WAKE:
1405                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1406                         state_to_char[entry->ctx.prev_state] : 'X';
1407                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1408                         state_to_char[entry->ctx.next_state] : 'X';
1409                 if (entry->type == TRACE_WAKE)
1410                         S = '+';
1411                 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1412                                        entry->ctx.prev_pid,
1413                                        entry->ctx.prev_prio,
1414                                        S,
1415                                        entry->ctx.next_pid,
1416                                        entry->ctx.next_prio,
1417                                        T);
1418                 if (!ret)
1419                         return 0;
1420                 break;
1421         case TRACE_SPECIAL:
1422         case TRACE_STACK:
1423                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1424                                  entry->special.arg1,
1425                                  entry->special.arg2,
1426                                  entry->special.arg3);
1427                 if (!ret)
1428                         return 0;
1429                 break;
1430         }
1431         return 1;
1432 }
1433
1434 #define SEQ_PUT_FIELD_RET(s, x)                         \
1435 do {                                                    \
1436         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1437                 return 0;                               \
1438 } while (0)
1439
1440 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1441 do {                                                    \
1442         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1443                 return 0;                               \
1444 } while (0)
1445
1446 static int print_hex_fmt(struct trace_iterator *iter)
1447 {
1448         struct trace_seq *s = &iter->seq;
1449         unsigned char newline = '\n';
1450         struct trace_entry *entry;
1451         int S, T;
1452
1453         entry = iter->ent;
1454
1455         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1456         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1457         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1458
1459         switch (entry->type) {
1460         case TRACE_FN:
1461                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1462                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1463                 break;
1464         case TRACE_CTX:
1465         case TRACE_WAKE:
1466                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1467                         state_to_char[entry->ctx.prev_state] : 'X';
1468                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1469                         state_to_char[entry->ctx.next_state] : 'X';
1470                 if (entry->type == TRACE_WAKE)
1471                         S = '+';
1472                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1473                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1474                 SEQ_PUT_HEX_FIELD_RET(s, S);
1475                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1476                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1477                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1478                 SEQ_PUT_HEX_FIELD_RET(s, T);
1479                 break;
1480         case TRACE_SPECIAL:
1481         case TRACE_STACK:
1482                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1483                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1484                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1485                 break;
1486         }
1487         SEQ_PUT_FIELD_RET(s, newline);
1488
1489         return 1;
1490 }
1491
1492 static int print_bin_fmt(struct trace_iterator *iter)
1493 {
1494         struct trace_seq *s = &iter->seq;
1495         struct trace_entry *entry;
1496
1497         entry = iter->ent;
1498
1499         SEQ_PUT_FIELD_RET(s, entry->pid);
1500         SEQ_PUT_FIELD_RET(s, entry->cpu);
1501         SEQ_PUT_FIELD_RET(s, entry->t);
1502
1503         switch (entry->type) {
1504         case TRACE_FN:
1505                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1506                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1507                 break;
1508         case TRACE_CTX:
1509                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1510                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1511                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1512                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1513                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1514                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1515                 break;
1516         case TRACE_SPECIAL:
1517         case TRACE_STACK:
1518                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1519                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1520                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1521                 break;
1522         }
1523         return 1;
1524 }
1525
1526 static int trace_empty(struct trace_iterator *iter)
1527 {
1528         struct trace_array_cpu *data;
1529         int cpu;
1530
1531         for_each_possible_cpu(cpu) {
1532                 data = iter->tr->data[cpu];
1533
1534                 if (head_page(data) && data->trace_idx &&
1535                     (data->trace_tail != data->trace_head ||
1536                      data->trace_tail_idx != data->trace_head_idx))
1537                         return 0;
1538         }
1539         return 1;
1540 }
1541
1542 static int print_trace_line(struct trace_iterator *iter)
1543 {
1544         if (trace_flags & TRACE_ITER_BIN)
1545                 return print_bin_fmt(iter);
1546
1547         if (trace_flags & TRACE_ITER_HEX)
1548                 return print_hex_fmt(iter);
1549
1550         if (trace_flags & TRACE_ITER_RAW)
1551                 return print_raw_fmt(iter);
1552
1553         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1554                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1555
1556         return print_trace_fmt(iter);
1557 }
1558
1559 static int s_show(struct seq_file *m, void *v)
1560 {
1561         struct trace_iterator *iter = v;
1562
1563         if (iter->ent == NULL) {
1564                 if (iter->tr) {
1565                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1566                         seq_puts(m, "#\n");
1567                 }
1568                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1569                         /* print nothing if the buffers are empty */
1570                         if (trace_empty(iter))
1571                                 return 0;
1572                         print_trace_header(m, iter);
1573                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1574                                 print_lat_help_header(m);
1575                 } else {
1576                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1577                                 print_func_help_header(m);
1578                 }
1579         } else {
1580                 print_trace_line(iter);
1581                 trace_print_seq(m, &iter->seq);
1582         }
1583
1584         return 0;
1585 }
1586
1587 static struct seq_operations tracer_seq_ops = {
1588         .start          = s_start,
1589         .next           = s_next,
1590         .stop           = s_stop,
1591         .show           = s_show,
1592 };
1593
1594 static struct trace_iterator *
1595 __tracing_open(struct inode *inode, struct file *file, int *ret)
1596 {
1597         struct trace_iterator *iter;
1598
1599         if (tracing_disabled) {
1600                 *ret = -ENODEV;
1601                 return NULL;
1602         }
1603
1604         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1605         if (!iter) {
1606                 *ret = -ENOMEM;
1607                 goto out;
1608         }
1609
1610         mutex_lock(&trace_types_lock);
1611         if (current_trace && current_trace->print_max)
1612                 iter->tr = &max_tr;
1613         else
1614                 iter->tr = inode->i_private;
1615         iter->trace = current_trace;
1616         iter->pos = -1;
1617
1618         /* TODO stop tracer */
1619         *ret = seq_open(file, &tracer_seq_ops);
1620         if (!*ret) {
1621                 struct seq_file *m = file->private_data;
1622                 m->private = iter;
1623
1624                 /* stop the trace while dumping */
1625                 if (iter->tr->ctrl)
1626                         tracer_enabled = 0;
1627
1628                 if (iter->trace && iter->trace->open)
1629                         iter->trace->open(iter);
1630         } else {
1631                 kfree(iter);
1632                 iter = NULL;
1633         }
1634         mutex_unlock(&trace_types_lock);
1635
1636  out:
1637         return iter;
1638 }
1639
1640 int tracing_open_generic(struct inode *inode, struct file *filp)
1641 {
1642         if (tracing_disabled)
1643                 return -ENODEV;
1644
1645         filp->private_data = inode->i_private;
1646         return 0;
1647 }
1648
1649 int tracing_release(struct inode *inode, struct file *file)
1650 {
1651         struct seq_file *m = (struct seq_file *)file->private_data;
1652         struct trace_iterator *iter = m->private;
1653
1654         mutex_lock(&trace_types_lock);
1655         if (iter->trace && iter->trace->close)
1656                 iter->trace->close(iter);
1657
1658         /* reenable tracing if it was previously enabled */
1659         if (iter->tr->ctrl)
1660                 tracer_enabled = 1;
1661         mutex_unlock(&trace_types_lock);
1662
1663         seq_release(inode, file);
1664         kfree(iter);
1665         return 0;
1666 }
1667
1668 static int tracing_open(struct inode *inode, struct file *file)
1669 {
1670         int ret;
1671
1672         __tracing_open(inode, file, &ret);
1673
1674         return ret;
1675 }
1676
1677 static int tracing_lt_open(struct inode *inode, struct file *file)
1678 {
1679         struct trace_iterator *iter;
1680         int ret;
1681
1682         iter = __tracing_open(inode, file, &ret);
1683
1684         if (!ret)
1685                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1686
1687         return ret;
1688 }
1689
1690
1691 static void *
1692 t_next(struct seq_file *m, void *v, loff_t *pos)
1693 {
1694         struct tracer *t = m->private;
1695
1696         (*pos)++;
1697
1698         if (t)
1699                 t = t->next;
1700
1701         m->private = t;
1702
1703         return t;
1704 }
1705
1706 static void *t_start(struct seq_file *m, loff_t *pos)
1707 {
1708         struct tracer *t = m->private;
1709         loff_t l = 0;
1710
1711         mutex_lock(&trace_types_lock);
1712         for (; t && l < *pos; t = t_next(m, t, &l))
1713                 ;
1714
1715         return t;
1716 }
1717
1718 static void t_stop(struct seq_file *m, void *p)
1719 {
1720         mutex_unlock(&trace_types_lock);
1721 }
1722
1723 static int t_show(struct seq_file *m, void *v)
1724 {
1725         struct tracer *t = v;
1726
1727         if (!t)
1728                 return 0;
1729
1730         seq_printf(m, "%s", t->name);
1731         if (t->next)
1732                 seq_putc(m, ' ');
1733         else
1734                 seq_putc(m, '\n');
1735
1736         return 0;
1737 }
1738
1739 static struct seq_operations show_traces_seq_ops = {
1740         .start          = t_start,
1741         .next           = t_next,
1742         .stop           = t_stop,
1743         .show           = t_show,
1744 };
1745
1746 static int show_traces_open(struct inode *inode, struct file *file)
1747 {
1748         int ret;
1749
1750         if (tracing_disabled)
1751                 return -ENODEV;
1752
1753         ret = seq_open(file, &show_traces_seq_ops);
1754         if (!ret) {
1755                 struct seq_file *m = file->private_data;
1756                 m->private = trace_types;
1757         }
1758
1759         return ret;
1760 }
1761
1762 static struct file_operations tracing_fops = {
1763         .open           = tracing_open,
1764         .read           = seq_read,
1765         .llseek         = seq_lseek,
1766         .release        = tracing_release,
1767 };
1768
1769 static struct file_operations tracing_lt_fops = {
1770         .open           = tracing_lt_open,
1771         .read           = seq_read,
1772         .llseek         = seq_lseek,
1773         .release        = tracing_release,
1774 };
1775
1776 static struct file_operations show_traces_fops = {
1777         .open           = show_traces_open,
1778         .read           = seq_read,
1779         .release        = seq_release,
1780 };
1781
1782 /*
1783  * Only trace on a CPU if the bitmask is set:
1784  */
1785 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1786
1787 /*
1788  * When tracing/tracing_cpu_mask is modified then this holds
1789  * the new bitmask we are about to install:
1790  */
1791 static cpumask_t tracing_cpumask_new;
1792
1793 /*
1794  * The tracer itself will not take this lock, but still we want
1795  * to provide a consistent cpumask to user-space:
1796  */
1797 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1798
1799 /*
1800  * Temporary storage for the character representation of the
1801  * CPU bitmask (and one more byte for the newline):
1802  */
1803 static char mask_str[NR_CPUS + 1];
1804
1805 static ssize_t
1806 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1807                      size_t count, loff_t *ppos)
1808 {
1809         int len;
1810
1811         mutex_lock(&tracing_cpumask_update_lock);
1812
1813         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1814         if (count - len < 2) {
1815                 count = -EINVAL;
1816                 goto out_err;
1817         }
1818         len += sprintf(mask_str + len, "\n");
1819         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1820
1821 out_err:
1822         mutex_unlock(&tracing_cpumask_update_lock);
1823
1824         return count;
1825 }
1826
1827 static ssize_t
1828 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1829                       size_t count, loff_t *ppos)
1830 {
1831         int err, cpu;
1832
1833         mutex_lock(&tracing_cpumask_update_lock);
1834         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1835         if (err)
1836                 goto err_unlock;
1837
1838         raw_local_irq_disable();
1839         __raw_spin_lock(&ftrace_max_lock);
1840         for_each_possible_cpu(cpu) {
1841                 /*
1842                  * Increase/decrease the disabled counter if we are
1843                  * about to flip a bit in the cpumask:
1844                  */
1845                 if (cpu_isset(cpu, tracing_cpumask) &&
1846                                 !cpu_isset(cpu, tracing_cpumask_new)) {
1847                         atomic_inc(&global_trace.data[cpu]->disabled);
1848                 }
1849                 if (!cpu_isset(cpu, tracing_cpumask) &&
1850                                 cpu_isset(cpu, tracing_cpumask_new)) {
1851                         atomic_dec(&global_trace.data[cpu]->disabled);
1852                 }
1853         }
1854         __raw_spin_unlock(&ftrace_max_lock);
1855         raw_local_irq_enable();
1856
1857         tracing_cpumask = tracing_cpumask_new;
1858
1859         mutex_unlock(&tracing_cpumask_update_lock);
1860
1861         return count;
1862
1863 err_unlock:
1864         mutex_unlock(&tracing_cpumask_update_lock);
1865
1866         return err;
1867 }
1868
1869 static struct file_operations tracing_cpumask_fops = {
1870         .open           = tracing_open_generic,
1871         .read           = tracing_cpumask_read,
1872         .write          = tracing_cpumask_write,
1873 };
1874
1875 static ssize_t
1876 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1877                        size_t cnt, loff_t *ppos)
1878 {
1879         char *buf;
1880         int r = 0;
1881         int len = 0;
1882         int i;
1883
1884         /* calulate max size */
1885         for (i = 0; trace_options[i]; i++) {
1886                 len += strlen(trace_options[i]);
1887                 len += 3; /* "no" and space */
1888         }
1889
1890         /* +2 for \n and \0 */
1891         buf = kmalloc(len + 2, GFP_KERNEL);
1892         if (!buf)
1893                 return -ENOMEM;
1894
1895         for (i = 0; trace_options[i]; i++) {
1896                 if (trace_flags & (1 << i))
1897                         r += sprintf(buf + r, "%s ", trace_options[i]);
1898                 else
1899                         r += sprintf(buf + r, "no%s ", trace_options[i]);
1900         }
1901
1902         r += sprintf(buf + r, "\n");
1903         WARN_ON(r >= len + 2);
1904
1905         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1906
1907         kfree(buf);
1908
1909         return r;
1910 }
1911
1912 static ssize_t
1913 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1914                         size_t cnt, loff_t *ppos)
1915 {
1916         char buf[64];
1917         char *cmp = buf;
1918         int neg = 0;
1919         int i;
1920
1921         if (cnt > 63)
1922                 cnt = 63;
1923
1924         if (copy_from_user(&buf, ubuf, cnt))
1925                 return -EFAULT;
1926
1927         buf[cnt] = 0;
1928
1929         if (strncmp(buf, "no", 2) == 0) {
1930                 neg = 1;
1931                 cmp += 2;
1932         }
1933
1934         for (i = 0; trace_options[i]; i++) {
1935                 int len = strlen(trace_options[i]);
1936
1937                 if (strncmp(cmp, trace_options[i], len) == 0) {
1938                         if (neg)
1939                                 trace_flags &= ~(1 << i);
1940                         else
1941                                 trace_flags |= (1 << i);
1942                         break;
1943                 }
1944         }
1945         /*
1946          * If no option could be set, return an error:
1947          */
1948         if (!trace_options[i])
1949                 return -EINVAL;
1950
1951         filp->f_pos += cnt;
1952
1953         return cnt;
1954 }
1955
1956 static struct file_operations tracing_iter_fops = {
1957         .open           = tracing_open_generic,
1958         .read           = tracing_iter_ctrl_read,
1959         .write          = tracing_iter_ctrl_write,
1960 };
1961
1962 static const char readme_msg[] =
1963         "tracing mini-HOWTO:\n\n"
1964         "# mkdir /debug\n"
1965         "# mount -t debugfs nodev /debug\n\n"
1966         "# cat /debug/tracing/available_tracers\n"
1967         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1968         "# cat /debug/tracing/current_tracer\n"
1969         "none\n"
1970         "# echo sched_switch > /debug/tracing/current_tracer\n"
1971         "# cat /debug/tracing/current_tracer\n"
1972         "sched_switch\n"
1973         "# cat /debug/tracing/iter_ctrl\n"
1974         "noprint-parent nosym-offset nosym-addr noverbose\n"
1975         "# echo print-parent > /debug/tracing/iter_ctrl\n"
1976         "# echo 1 > /debug/tracing/tracing_enabled\n"
1977         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1978         "echo 0 > /debug/tracing/tracing_enabled\n"
1979 ;
1980
1981 static ssize_t
1982 tracing_readme_read(struct file *filp, char __user *ubuf,
1983                        size_t cnt, loff_t *ppos)
1984 {
1985         return simple_read_from_buffer(ubuf, cnt, ppos,
1986                                         readme_msg, strlen(readme_msg));
1987 }
1988
1989 static struct file_operations tracing_readme_fops = {
1990         .open           = tracing_open_generic,
1991         .read           = tracing_readme_read,
1992 };
1993
1994 static ssize_t
1995 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1996                   size_t cnt, loff_t *ppos)
1997 {
1998         struct trace_array *tr = filp->private_data;
1999         char buf[64];
2000         int r;
2001
2002         r = sprintf(buf, "%ld\n", tr->ctrl);
2003         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2004 }
2005
2006 static ssize_t
2007 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2008                    size_t cnt, loff_t *ppos)
2009 {
2010         struct trace_array *tr = filp->private_data;
2011         long val;
2012         char buf[64];
2013
2014         if (cnt > 63)
2015                 cnt = 63;
2016
2017         if (copy_from_user(&buf, ubuf, cnt))
2018                 return -EFAULT;
2019
2020         buf[cnt] = 0;
2021
2022         val = simple_strtoul(buf, NULL, 10);
2023
2024         val = !!val;
2025
2026         mutex_lock(&trace_types_lock);
2027         if (tr->ctrl ^ val) {
2028                 if (val)
2029                         tracer_enabled = 1;
2030                 else
2031                         tracer_enabled = 0;
2032
2033                 tr->ctrl = val;
2034
2035                 if (current_trace && current_trace->ctrl_update)
2036                         current_trace->ctrl_update(tr);
2037         }
2038         mutex_unlock(&trace_types_lock);
2039
2040         filp->f_pos += cnt;
2041
2042         return cnt;
2043 }
2044
2045 static ssize_t
2046 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2047                        size_t cnt, loff_t *ppos)
2048 {
2049         char buf[max_tracer_type_len+2];
2050         int r;
2051
2052         mutex_lock(&trace_types_lock);
2053         if (current_trace)
2054                 r = sprintf(buf, "%s\n", current_trace->name);
2055         else
2056                 r = sprintf(buf, "\n");
2057         mutex_unlock(&trace_types_lock);
2058
2059         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2060 }
2061
2062 static ssize_t
2063 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2064                         size_t cnt, loff_t *ppos)
2065 {
2066         struct trace_array *tr = &global_trace;
2067         struct tracer *t;
2068         char buf[max_tracer_type_len+1];
2069         int i;
2070
2071         if (cnt > max_tracer_type_len)
2072                 cnt = max_tracer_type_len;
2073
2074         if (copy_from_user(&buf, ubuf, cnt))
2075                 return -EFAULT;
2076
2077         buf[cnt] = 0;
2078
2079         /* strip ending whitespace. */
2080         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2081                 buf[i] = 0;
2082
2083         mutex_lock(&trace_types_lock);
2084         for (t = trace_types; t; t = t->next) {
2085                 if (strcmp(t->name, buf) == 0)
2086                         break;
2087         }
2088         if (!t || t == current_trace)
2089                 goto out;
2090
2091         if (current_trace && current_trace->reset)
2092                 current_trace->reset(tr);
2093
2094         current_trace = t;
2095         if (t->init)
2096                 t->init(tr);
2097
2098  out:
2099         mutex_unlock(&trace_types_lock);
2100
2101         filp->f_pos += cnt;
2102
2103         return cnt;
2104 }
2105
2106 static ssize_t
2107 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2108                      size_t cnt, loff_t *ppos)
2109 {
2110         unsigned long *ptr = filp->private_data;
2111         char buf[64];
2112         int r;
2113
2114         r = snprintf(buf, 64, "%ld\n",
2115                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2116         if (r > 64)
2117                 r = 64;
2118         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2119 }
2120
2121 static ssize_t
2122 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2123                       size_t cnt, loff_t *ppos)
2124 {
2125         long *ptr = filp->private_data;
2126         long val;
2127         char buf[64];
2128
2129         if (cnt > 63)
2130                 cnt = 63;
2131
2132         if (copy_from_user(&buf, ubuf, cnt))
2133                 return -EFAULT;
2134
2135         buf[cnt] = 0;
2136
2137         val = simple_strtoul(buf, NULL, 10);
2138
2139         *ptr = val * 1000;
2140
2141         return cnt;
2142 }
2143
2144 static atomic_t tracing_reader;
2145
2146 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2147 {
2148         struct trace_iterator *iter;
2149
2150         if (tracing_disabled)
2151                 return -ENODEV;
2152
2153         /* We only allow for reader of the pipe */
2154         if (atomic_inc_return(&tracing_reader) != 1) {
2155                 atomic_dec(&tracing_reader);
2156                 return -EBUSY;
2157         }
2158
2159         /* create a buffer to store the information to pass to userspace */
2160         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2161         if (!iter)
2162                 return -ENOMEM;
2163
2164         iter->tr = &global_trace;
2165
2166         filp->private_data = iter;
2167
2168         return 0;
2169 }
2170
2171 static int tracing_release_pipe(struct inode *inode, struct file *file)
2172 {
2173         struct trace_iterator *iter = file->private_data;
2174
2175         kfree(iter);
2176         atomic_dec(&tracing_reader);
2177
2178         return 0;
2179 }
2180
2181 static unsigned int
2182 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2183 {
2184         struct trace_iterator *iter = filp->private_data;
2185
2186         if (trace_flags & TRACE_ITER_BLOCK) {
2187                 /*
2188                  * Always select as readable when in blocking mode
2189                  */
2190                 return POLLIN | POLLRDNORM;
2191         }
2192         else {
2193                 if (!trace_empty(iter))
2194                         return POLLIN | POLLRDNORM;
2195                 poll_wait(filp, &trace_wait, poll_table);
2196                 if (!trace_empty(iter))
2197                         return POLLIN | POLLRDNORM;
2198
2199                 return 0;
2200         }
2201 }
2202
2203 /*
2204  * Consumer reader.
2205  */
2206 static ssize_t
2207 tracing_read_pipe(struct file *filp, char __user *ubuf,
2208                   size_t cnt, loff_t *ppos)
2209 {
2210         struct trace_iterator *iter = filp->private_data;
2211         struct trace_array_cpu *data;
2212         static cpumask_t mask;
2213         static int start;
2214         unsigned long flags;
2215 #ifdef CONFIG_FTRACE
2216         int ftrace_save;
2217 #endif
2218         int read = 0;
2219         int cpu;
2220         int len;
2221         int ret;
2222
2223         /* return any leftover data */
2224         if (iter->seq.len > start) {
2225                 len = iter->seq.len - start;
2226                 if (cnt > len)
2227                         cnt = len;
2228                 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2229                 if (ret)
2230                         cnt = -EFAULT;
2231
2232                 start += len;
2233
2234                 return cnt;
2235         }
2236
2237         trace_seq_reset(&iter->seq);
2238         start = 0;
2239
2240         while (trace_empty(iter)) {
2241                 if (!(trace_flags & TRACE_ITER_BLOCK))
2242                         return -EWOULDBLOCK;
2243                 /*
2244                  * This is a make-shift waitqueue. The reason we don't use
2245                  * an actual wait queue is because:
2246                  *  1) we only ever have one waiter
2247                  *  2) the tracing, traces all functions, we don't want
2248                  *     the overhead of calling wake_up and friends
2249                  *     (and tracing them too)
2250                  *     Anyway, this is really very primitive wakeup.
2251                  */
2252                 set_current_state(TASK_INTERRUPTIBLE);
2253                 iter->tr->waiter = current;
2254
2255                 /* sleep for one second, and try again. */
2256                 schedule_timeout(HZ);
2257
2258                 iter->tr->waiter = NULL;
2259
2260                 if (signal_pending(current))
2261                         return -EINTR;
2262
2263                 /*
2264                  * We block until we read something and tracing is disabled.
2265                  * We still block if tracing is disabled, but we have never
2266                  * read anything. This allows a user to cat this file, and
2267                  * then enable tracing. But after we have read something,
2268                  * we give an EOF when tracing is again disabled.
2269                  *
2270                  * iter->pos will be 0 if we haven't read anything.
2271                  */
2272                 if (!tracer_enabled && iter->pos)
2273                         break;
2274
2275                 continue;
2276         }
2277
2278         /* stop when tracing is finished */
2279         if (trace_empty(iter))
2280                 return 0;
2281
2282         if (cnt >= PAGE_SIZE)
2283                 cnt = PAGE_SIZE - 1;
2284
2285         memset(iter, 0, sizeof(*iter));
2286         iter->tr = &global_trace;
2287         iter->pos = -1;
2288
2289         /*
2290          * We need to stop all tracing on all CPUS to read the
2291          * the next buffer. This is a bit expensive, but is
2292          * not done often. We fill all what we can read,
2293          * and then release the locks again.
2294          */
2295
2296         cpus_clear(mask);
2297         local_irq_save(flags);
2298 #ifdef CONFIG_FTRACE
2299         ftrace_save = ftrace_enabled;
2300         ftrace_enabled = 0;
2301 #endif
2302         smp_wmb();
2303         for_each_possible_cpu(cpu) {
2304                 data = iter->tr->data[cpu];
2305
2306                 if (!head_page(data) || !data->trace_idx)
2307                         continue;
2308
2309                 atomic_inc(&data->disabled);
2310                 cpu_set(cpu, mask);
2311         }
2312
2313         for_each_cpu_mask(cpu, mask) {
2314                 data = iter->tr->data[cpu];
2315                 __raw_spin_lock(&data->lock);
2316         }
2317
2318         while (find_next_entry_inc(iter) != NULL) {
2319                 int len = iter->seq.len;
2320
2321                 ret = print_trace_line(iter);
2322                 if (!ret) {
2323                         /* don't print partial lines */
2324                         iter->seq.len = len;
2325                         break;
2326                 }
2327
2328                 trace_consume(iter);
2329
2330                 if (iter->seq.len >= cnt)
2331                         break;
2332         }
2333
2334         for_each_cpu_mask(cpu, mask) {
2335                 data = iter->tr->data[cpu];
2336                 __raw_spin_unlock(&data->lock);
2337         }
2338
2339         for_each_cpu_mask(cpu, mask) {
2340                 data = iter->tr->data[cpu];
2341                 atomic_dec(&data->disabled);
2342         }
2343 #ifdef CONFIG_FTRACE
2344         ftrace_enabled = ftrace_save;
2345 #endif
2346         local_irq_restore(flags);
2347
2348         /* Now copy what we have to the user */
2349         read = iter->seq.len;
2350         if (read > cnt)
2351                 read = cnt;
2352
2353         ret = copy_to_user(ubuf, iter->seq.buffer, read);
2354
2355         if (read < iter->seq.len)
2356                 start = read;
2357         else
2358                 trace_seq_reset(&iter->seq);
2359
2360         if (ret)
2361                 read = -EFAULT;
2362
2363         return read;
2364 }
2365
2366 static struct file_operations tracing_max_lat_fops = {
2367         .open           = tracing_open_generic,
2368         .read           = tracing_max_lat_read,
2369         .write          = tracing_max_lat_write,
2370 };
2371
2372 static struct file_operations tracing_ctrl_fops = {
2373         .open           = tracing_open_generic,
2374         .read           = tracing_ctrl_read,
2375         .write          = tracing_ctrl_write,
2376 };
2377
2378 static struct file_operations set_tracer_fops = {
2379         .open           = tracing_open_generic,
2380         .read           = tracing_set_trace_read,
2381         .write          = tracing_set_trace_write,
2382 };
2383
2384 static struct file_operations tracing_pipe_fops = {
2385         .open           = tracing_open_pipe,
2386         .poll           = tracing_poll_pipe,
2387         .read           = tracing_read_pipe,
2388         .release        = tracing_release_pipe,
2389 };
2390
2391 #ifdef CONFIG_DYNAMIC_FTRACE
2392
2393 static ssize_t
2394 tracing_read_long(struct file *filp, char __user *ubuf,
2395                   size_t cnt, loff_t *ppos)
2396 {
2397         unsigned long *p = filp->private_data;
2398         char buf[64];
2399         int r;
2400
2401         r = sprintf(buf, "%ld\n", *p);
2402
2403         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2404 }
2405
2406 static struct file_operations tracing_read_long_fops = {
2407         .open           = tracing_open_generic,
2408         .read           = tracing_read_long,
2409 };
2410 #endif
2411
2412 static struct dentry *d_tracer;
2413
2414 struct dentry *tracing_init_dentry(void)
2415 {
2416         static int once;
2417
2418         if (d_tracer)
2419                 return d_tracer;
2420
2421         d_tracer = debugfs_create_dir("tracing", NULL);
2422
2423         if (!d_tracer && !once) {
2424                 once = 1;
2425                 pr_warning("Could not create debugfs directory 'tracing'\n");
2426                 return NULL;
2427         }
2428
2429         return d_tracer;
2430 }
2431
2432 #ifdef CONFIG_FTRACE_SELFTEST
2433 /* Let selftest have access to static functions in this file */
2434 #include "trace_selftest.c"
2435 #endif
2436
2437 static __init void tracer_init_debugfs(void)
2438 {
2439         struct dentry *d_tracer;
2440         struct dentry *entry;
2441
2442         d_tracer = tracing_init_dentry();
2443
2444         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2445                                     &global_trace, &tracing_ctrl_fops);
2446         if (!entry)
2447                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2448
2449         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2450                                     NULL, &tracing_iter_fops);
2451         if (!entry)
2452                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2453
2454         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2455                                     NULL, &tracing_cpumask_fops);
2456         if (!entry)
2457                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2458
2459         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2460                                     &global_trace, &tracing_lt_fops);
2461         if (!entry)
2462                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2463
2464         entry = debugfs_create_file("trace", 0444, d_tracer,
2465                                     &global_trace, &tracing_fops);
2466         if (!entry)
2467                 pr_warning("Could not create debugfs 'trace' entry\n");
2468
2469         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2470                                     &global_trace, &show_traces_fops);
2471         if (!entry)
2472                 pr_warning("Could not create debugfs 'trace' entry\n");
2473
2474         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2475                                     &global_trace, &set_tracer_fops);
2476         if (!entry)
2477                 pr_warning("Could not create debugfs 'trace' entry\n");
2478
2479         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2480                                     &tracing_max_latency,
2481                                     &tracing_max_lat_fops);
2482         if (!entry)
2483                 pr_warning("Could not create debugfs "
2484                            "'tracing_max_latency' entry\n");
2485
2486         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2487                                     &tracing_thresh, &tracing_max_lat_fops);
2488         if (!entry)
2489                 pr_warning("Could not create debugfs "
2490                            "'tracing_threash' entry\n");
2491         entry = debugfs_create_file("README", 0644, d_tracer,
2492                                     NULL, &tracing_readme_fops);
2493         if (!entry)
2494                 pr_warning("Could not create debugfs 'README' entry\n");
2495
2496         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2497                                     NULL, &tracing_pipe_fops);
2498         if (!entry)
2499                 pr_warning("Could not create debugfs "
2500                            "'tracing_threash' entry\n");
2501
2502 #ifdef CONFIG_DYNAMIC_FTRACE
2503         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2504                                     &ftrace_update_tot_cnt,
2505                                     &tracing_read_long_fops);
2506         if (!entry)
2507                 pr_warning("Could not create debugfs "
2508                            "'dyn_ftrace_total_info' entry\n");
2509 #endif
2510 }
2511
2512 /* dummy trace to disable tracing */
2513 static struct tracer no_tracer __read_mostly =
2514 {
2515         .name           = "none",
2516 };
2517
2518 static int trace_alloc_page(void)
2519 {
2520         struct trace_array_cpu *data;
2521         struct page *page, *tmp;
2522         LIST_HEAD(pages);
2523         void *array;
2524         int i;
2525
2526         /* first allocate a page for each CPU */
2527         for_each_possible_cpu(i) {
2528                 array = (void *)__get_free_page(GFP_KERNEL);
2529                 if (array == NULL) {
2530                         printk(KERN_ERR "tracer: failed to allocate page"
2531                                "for trace buffer!\n");
2532                         goto free_pages;
2533                 }
2534
2535                 page = virt_to_page(array);
2536                 list_add(&page->lru, &pages);
2537
2538 /* Only allocate if we are actually using the max trace */
2539 #ifdef CONFIG_TRACER_MAX_TRACE
2540                 array = (void *)__get_free_page(GFP_KERNEL);
2541                 if (array == NULL) {
2542                         printk(KERN_ERR "tracer: failed to allocate page"
2543                                "for trace buffer!\n");
2544                         goto free_pages;
2545                 }
2546                 page = virt_to_page(array);
2547                 list_add(&page->lru, &pages);
2548 #endif
2549         }
2550
2551         /* Now that we successfully allocate a page per CPU, add them */
2552         for_each_possible_cpu(i) {
2553                 data = global_trace.data[i];
2554                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2555                 page = list_entry(pages.next, struct page, lru);
2556                 list_del_init(&page->lru);
2557                 list_add_tail(&page->lru, &data->trace_pages);
2558                 ClearPageLRU(page);
2559
2560 #ifdef CONFIG_TRACER_MAX_TRACE
2561                 data = max_tr.data[i];
2562                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2563                 page = list_entry(pages.next, struct page, lru);
2564                 list_del_init(&page->lru);
2565                 list_add_tail(&page->lru, &data->trace_pages);
2566                 SetPageLRU(page);
2567 #endif
2568         }
2569         global_trace.entries += ENTRIES_PER_PAGE;
2570
2571         return 0;
2572
2573  free_pages:
2574         list_for_each_entry_safe(page, tmp, &pages, lru) {
2575                 list_del_init(&page->lru);
2576                 __free_page(page);
2577         }
2578         return -ENOMEM;
2579 }
2580
2581 __init static int tracer_alloc_buffers(void)
2582 {
2583         struct trace_array_cpu *data;
2584         void *array;
2585         struct page *page;
2586         int pages = 0;
2587         int ret = -ENOMEM;
2588         int i;
2589
2590         global_trace.ctrl = tracer_enabled;
2591
2592         /* Allocate the first page for all buffers */
2593         for_each_possible_cpu(i) {
2594                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2595                 max_tr.data[i] = &per_cpu(max_data, i);
2596
2597                 array = (void *)__get_free_page(GFP_KERNEL);
2598                 if (array == NULL) {
2599                         printk(KERN_ERR "tracer: failed to allocate page"
2600                                "for trace buffer!\n");
2601                         goto free_buffers;
2602                 }
2603
2604                 /* set the array to the list */
2605                 INIT_LIST_HEAD(&data->trace_pages);
2606                 page = virt_to_page(array);
2607                 list_add(&page->lru, &data->trace_pages);
2608                 /* use the LRU flag to differentiate the two buffers */
2609                 ClearPageLRU(page);
2610
2611 /* Only allocate if we are actually using the max trace */
2612 #ifdef CONFIG_TRACER_MAX_TRACE
2613                 array = (void *)__get_free_page(GFP_KERNEL);
2614                 if (array == NULL) {
2615                         printk(KERN_ERR "tracer: failed to allocate page"
2616                                "for trace buffer!\n");
2617                         goto free_buffers;
2618                 }
2619
2620                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2621                 page = virt_to_page(array);
2622                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2623                 SetPageLRU(page);
2624 #endif
2625         }
2626
2627         /*
2628          * Since we allocate by orders of pages, we may be able to
2629          * round up a bit.
2630          */
2631         global_trace.entries = ENTRIES_PER_PAGE;
2632         pages++;
2633
2634         while (global_trace.entries < trace_nr_entries) {
2635                 if (trace_alloc_page())
2636                         break;
2637                 pages++;
2638         }
2639         max_tr.entries = global_trace.entries;
2640
2641         pr_info("tracer: %d pages allocated for %ld",
2642                 pages, trace_nr_entries);
2643         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2644         pr_info("   actual entries %ld\n", global_trace.entries);
2645
2646         tracer_init_debugfs();
2647
2648         trace_init_cmdlines();
2649
2650         register_tracer(&no_tracer);
2651         current_trace = &no_tracer;
2652
2653         /* All seems OK, enable tracing */
2654         tracing_disabled = 0;
2655
2656         return 0;
2657
2658  free_buffers:
2659         for (i-- ; i >= 0; i--) {
2660                 struct page *page, *tmp;
2661                 struct trace_array_cpu *data = global_trace.data[i];
2662
2663                 if (data) {
2664                         list_for_each_entry_safe(page, tmp,
2665                                                  &data->trace_pages, lru) {
2666                                 list_del_init(&page->lru);
2667                                 __free_page(page);
2668                         }
2669                 }
2670
2671 #ifdef CONFIG_TRACER_MAX_TRACE
2672                 data = max_tr.data[i];
2673                 if (data) {
2674                         list_for_each_entry_safe(page, tmp,
2675                                                  &data->trace_pages, lru) {
2676                                 list_del_init(&page->lru);
2677                                 __free_page(page);
2678                         }
2679                 }
2680 #endif
2681         }
2682         return ret;
2683 }
2684 fs_initcall(tracer_alloc_buffers);