kernel/trace/trace_irqsoff.c

   1 /*
   2  * trace irqs off critical timings
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * From code in the latency_tracer, that is:
   8  *
   9  *  Copyright (C) 2004-2006 Ingo Molnar
  10  *  Copyright (C) 2004 Nadia Yvette Chambers
  11  */
  12 #include <linux/kallsyms.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/module.h>
  15 #include <linux/ftrace.h>
  16
  17 #include "trace.h"
  18
  19 static struct trace_array               *irqsoff_trace __read_mostly;
  20 static int                              tracer_enabled __read_mostly;
  21
  22 static DEFINE_PER_CPU(int, tracing_cpu);
  23
  24 static DEFINE_RAW_SPINLOCK(max_trace_lock);
  25
  26 enum {
  27         TRACER_IRQS_OFF         = (1 << 1),
  28         TRACER_PREEMPT_OFF      = (1 << 2),
  29 };
  30
  31 static int trace_type __read_mostly;
  32
  33 static int save_flags;
  34 static bool function_enabled;
  35
  36 static void stop_irqsoff_tracer(struct trace_array *tr, int graph);
  37 static int start_irqsoff_tracer(struct trace_array *tr, int graph);
  38
  39 #ifdef CONFIG_PREEMPT_TRACER
  40 static inline int
  41 preempt_trace(void)
  42 {
  43         return ((trace_type & TRACER_PREEMPT_OFF) && preempt_count());
  44 }
  45 #else
  46 # define preempt_trace() (0)
  47 #endif
  48
  49 #ifdef CONFIG_IRQSOFF_TRACER
  50 static inline int
  51 irq_trace(void)
  52 {
  53         return ((trace_type & TRACER_IRQS_OFF) &&
  54                 irqs_disabled());
  55 }
  56 #else
  57 # define irq_trace() (0)
  58 #endif
  59
  60 #define is_graph() (trace_flags & TRACE_ITER_DISPLAY_GRAPH)
  61
  62 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  63 static int irqsoff_display_graph(struct trace_array *tr, int set);
  64 #else
  65 static inline int irqsoff_display_graph(struct trace_array *tr, int set)
  66 {
  67         return -EINVAL;
  68 }
  69 #endif
  70
  71 /*
  72  * Sequence count - we record it when starting a measurement and
  73  * skip the latency if the sequence has changed - some other section
  74  * did a maximum and could disturb our measurement with serial console
  75  * printouts, etc. Truly coinciding maximum latencies should be rare
  76  * and what happens together happens separately as well, so this doesn't
  77  * decrease the validity of the maximum found:
  78  */
  79 static __cacheline_aligned_in_smp       unsigned long max_sequence;
  80
  81 #ifdef CONFIG_FUNCTION_TRACER
  82 /*
  83  * Prologue for the preempt and irqs off function tracers.
  84  *
  85  * Returns 1 if it is OK to continue, and data->disabled is
  86  *            incremented.
  87  *         0 if the trace is to be ignored, and data->disabled
  88  *            is kept the same.
  89  *
  90  * Note, this function is also used outside this ifdef but
  91  *  inside the #ifdef of the function graph tracer below.
  92  *  This is OK, since the function graph tracer is
  93  *  dependent on the function tracer.
  94  */
  95 static int func_prolog_dec(struct trace_array *tr,
  96                            struct trace_array_cpu **data,
  97                            unsigned long *flags)
  98 {
  99         long disabled;
 100         int cpu;
 101
 102         /*
 103          * Does not matter if we preempt. We test the flags
 104          * afterward, to see if irqs are disabled or not.
 105          * If we preempt and get a false positive, the flags
 106          * test will fail.
 107          */
 108         cpu = raw_smp_processor_id();
 109         if (likely(!per_cpu(tracing_cpu, cpu)))
 110                 return 0;
 111
 112         local_save_flags(*flags);
 113         /* slight chance to get a false positive on tracing_cpu */
 114         if (!irqs_disabled_flags(*flags))
 115                 return 0;
 116
 117         *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 118         disabled = atomic_inc_return(&(*data)->disabled);
 119
 120         if (likely(disabled == 1))
 121                 return 1;
 122
 123         atomic_dec(&(*data)->disabled);
 124
 125         return 0;
 126 }
 127
 128 /*
 129  * irqsoff uses its own tracer function to keep the overhead down:
 130  */
 131 static void
 132 irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip,
 133                     struct ftrace_ops *op, struct pt_regs *pt_regs)
 134 {
 135         struct trace_array *tr = irqsoff_trace;
 136         struct trace_array_cpu *data;
 137         unsigned long flags;
 138
 139         if (!func_prolog_dec(tr, &data, &flags))
 140                 return;
 141
 142         trace_function(tr, ip, parent_ip, flags, preempt_count());
 143
 144         atomic_dec(&data->disabled);
 145 }
 146 #endif /* CONFIG_FUNCTION_TRACER */
 147
 148 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 149 static int irqsoff_display_graph(struct trace_array *tr, int set)
 150 {
 151         int cpu;
 152
 153         if (!(is_graph() ^ set))
 154                 return 0;
 155
 156         stop_irqsoff_tracer(irqsoff_trace, !set);
 157
 158         for_each_possible_cpu(cpu)
 159                 per_cpu(tracing_cpu, cpu) = 0;
 160
 161         tr->max_latency = 0;
 162         tracing_reset_online_cpus(&irqsoff_trace->trace_buffer);
 163
 164         return start_irqsoff_tracer(irqsoff_trace, set);
 165 }
 166
 167 static int irqsoff_graph_entry(struct ftrace_graph_ent *trace)
 168 {
 169         struct trace_array *tr = irqsoff_trace;
 170         struct trace_array_cpu *data;
 171         unsigned long flags;
 172         int ret;
 173         int pc;
 174
 175         if (!func_prolog_dec(tr, &data, &flags))
 176                 return 0;
 177
 178         pc = preempt_count();
 179         ret = __trace_graph_entry(tr, trace, flags, pc);
 180         atomic_dec(&data->disabled);
 181
 182         return ret;
 183 }
 184
 185 static void irqsoff_graph_return(struct ftrace_graph_ret *trace)
 186 {
 187         struct trace_array *tr = irqsoff_trace;
 188         struct trace_array_cpu *data;
 189         unsigned long flags;
 190         int pc;
 191
 192         if (!func_prolog_dec(tr, &data, &flags))
 193                 return;
 194
 195         pc = preempt_count();
 196         __trace_graph_return(tr, trace, flags, pc);
 197         atomic_dec(&data->disabled);
 198 }
 199
 200 static void irqsoff_trace_open(struct trace_iterator *iter)
 201 {
 202         if (is_graph())
 203                 graph_trace_open(iter);
 204
 205 }
 206
 207 static void irqsoff_trace_close(struct trace_iterator *iter)
 208 {
 209         if (iter->private)
 210                 graph_trace_close(iter);
 211 }
 212
 213 #define GRAPH_TRACER_FLAGS (TRACE_GRAPH_PRINT_CPU | \
 214                             TRACE_GRAPH_PRINT_PROC | \
 215                             TRACE_GRAPH_PRINT_ABS_TIME | \
 216                             TRACE_GRAPH_PRINT_DURATION)
 217
 218 static enum print_line_t irqsoff_print_line(struct trace_iterator *iter)
 219 {
 220         /*
 221          * In graph mode call the graph tracer output function,
 222          * otherwise go with the TRACE_FN event handler
 223          */
 224         if (is_graph())
 225                 return print_graph_function_flags(iter, GRAPH_TRACER_FLAGS);
 226
 227         return TRACE_TYPE_UNHANDLED;
 228 }
 229
 230 static void irqsoff_print_header(struct seq_file *s)
 231 {
 232         if (is_graph())
 233                 print_graph_headers_flags(s, GRAPH_TRACER_FLAGS);
 234         else
 235                 trace_default_header(s);
 236 }
 237
 238 static void
 239 __trace_function(struct trace_array *tr,
 240                  unsigned long ip, unsigned long parent_ip,
 241                  unsigned long flags, int pc)
 242 {
 243         if (is_graph())
 244                 trace_graph_function(tr, ip, parent_ip, flags, pc);
 245         else
 246                 trace_function(tr, ip, parent_ip, flags, pc);
 247 }
 248
 249 #else
 250 #define __trace_function trace_function
 251
 252 static int irqsoff_graph_entry(struct ftrace_graph_ent *trace)
 253 {
 254         return -1;
 255 }
 256
 257 static enum print_line_t irqsoff_print_line(struct trace_iterator *iter)
 258 {
 259         return TRACE_TYPE_UNHANDLED;
 260 }
 261
 262 static void irqsoff_graph_return(struct ftrace_graph_ret *trace) { }
 263 static void irqsoff_trace_open(struct trace_iterator *iter) { }
 264 static void irqsoff_trace_close(struct trace_iterator *iter) { }
 265
 266 #ifdef CONFIG_FUNCTION_TRACER
 267 static void irqsoff_print_header(struct seq_file *s)
 268 {
 269         trace_default_header(s);
 270 }
 271 #else
 272 static void irqsoff_print_header(struct seq_file *s)
 273 {
 274         trace_latency_header(s);
 275 }
 276 #endif /* CONFIG_FUNCTION_TRACER */
 277 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 278
 279 /*
 280  * Should this new latency be reported/recorded?
 281  */
 282 static int report_latency(struct trace_array *tr, cycle_t delta)
 283 {
 284         if (tracing_thresh) {
 285                 if (delta < tracing_thresh)
 286                         return 0;
 287         } else {
 288                 if (delta <= tr->max_latency)
 289                         return 0;
 290         }
 291         return 1;
 292 }
 293
 294 static void
 295 check_critical_timing(struct trace_array *tr,
 296                       struct trace_array_cpu *data,
 297                       unsigned long parent_ip,
 298                       int cpu)
 299 {
 300         cycle_t T0, T1, delta;
 301         unsigned long flags;
 302         int pc;
 303
 304         T0 = data->preempt_timestamp;
 305         T1 = ftrace_now(cpu);
 306         delta = T1-T0;
 307
 308         local_save_flags(flags);
 309
 310         pc = preempt_count();
 311
 312         if (!report_latency(tr, delta))
 313                 goto out;
 314
 315         raw_spin_lock_irqsave(&max_trace_lock, flags);
 316
 317         /* check if we are still the max latency */
 318         if (!report_latency(tr, delta))
 319                 goto out_unlock;
 320
 321         __trace_function(tr, CALLER_ADDR0, parent_ip, flags, pc);
 322         /* Skip 5 functions to get to the irq/preempt enable function */
 323         __trace_stack(tr, flags, 5, pc);
 324
 325         if (data->critical_sequence != max_sequence)
 326                 goto out_unlock;
 327
 328         data->critical_end = parent_ip;
 329
 330         if (likely(!is_tracing_stopped())) {
 331                 tr->max_latency = delta;
 332                 update_max_tr_single(tr, current, cpu);
 333         }
 334
 335         max_sequence++;
 336
 337 out_unlock:
 338         raw_spin_unlock_irqrestore(&max_trace_lock, flags);
 339
 340 out:
 341         data->critical_sequence = max_sequence;
 342         data->preempt_timestamp = ftrace_now(cpu);
 343         __trace_function(tr, CALLER_ADDR0, parent_ip, flags, pc);
 344 }
 345
 346 static inline void
 347 start_critical_timing(unsigned long ip, unsigned long parent_ip)
 348 {
 349         int cpu;
 350         struct trace_array *tr = irqsoff_trace;
 351         struct trace_array_cpu *data;
 352         unsigned long flags;
 353
 354         if (!tracer_enabled || !tracing_is_enabled())
 355                 return;
 356
 357         cpu = raw_smp_processor_id();
 358
 359         if (per_cpu(tracing_cpu, cpu))
 360                 return;
 361
 362         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 363
 364         if (unlikely(!data) || atomic_read(&data->disabled))
 365                 return;
 366
 367         atomic_inc(&data->disabled);
 368
 369         data->critical_sequence = max_sequence;
 370         data->preempt_timestamp = ftrace_now(cpu);
 371         data->critical_start = parent_ip ? : ip;
 372
 373         local_save_flags(flags);
 374
 375         __trace_function(tr, ip, parent_ip, flags, preempt_count());
 376
 377         per_cpu(tracing_cpu, cpu) = 1;
 378
 379         atomic_dec(&data->disabled);
 380 }
 381
 382 static inline void
 383 stop_critical_timing(unsigned long ip, unsigned long parent_ip)
 384 {
 385         int cpu;
 386         struct trace_array *tr = irqsoff_trace;
 387         struct trace_array_cpu *data;
 388         unsigned long flags;
 389
 390         cpu = raw_smp_processor_id();
 391         /* Always clear the tracing cpu on stopping the trace */
 392         if (unlikely(per_cpu(tracing_cpu, cpu)))
 393                 per_cpu(tracing_cpu, cpu) = 0;
 394         else
 395                 return;
 396
 397         if (!tracer_enabled || !tracing_is_enabled())
 398                 return;
 399
 400         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 401
 402         if (unlikely(!data) ||
 403             !data->critical_start || atomic_read(&data->disabled))
 404                 return;
 405
 406         atomic_inc(&data->disabled);
 407
 408         local_save_flags(flags);
 409         __trace_function(tr, ip, parent_ip, flags, preempt_count());
 410         check_critical_timing(tr, data, parent_ip ? : ip, cpu);
 411         data->critical_start = 0;
 412         atomic_dec(&data->disabled);
 413 }
 414
 415 /* start and stop critical timings used to for stoppage (in idle) */
 416 void start_critical_timings(void)
 417 {
 418         if (preempt_trace() || irq_trace())
 419                 start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
 420 }
 421 EXPORT_SYMBOL_GPL(start_critical_timings);
 422
 423 void stop_critical_timings(void)
 424 {
 425         if (preempt_trace() || irq_trace())
 426                 stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
 427 }
 428 EXPORT_SYMBOL_GPL(stop_critical_timings);
 429
 430 #ifdef CONFIG_IRQSOFF_TRACER
 431 #ifdef CONFIG_PROVE_LOCKING
 432 void time_hardirqs_on(unsigned long a0, unsigned long a1)
 433 {
 434         if (!preempt_trace() && irq_trace())
 435                 stop_critical_timing(a0, a1);
 436 }
 437
 438 void time_hardirqs_off(unsigned long a0, unsigned long a1)
 439 {
 440         if (!preempt_trace() && irq_trace())
 441                 start_critical_timing(a0, a1);
 442 }
 443
 444 #else /* !CONFIG_PROVE_LOCKING */
 445
 446 /*
 447  * Stubs:
 448  */
 449
 450 void trace_softirqs_on(unsigned long ip)
 451 {
 452 }
 453
 454 void trace_softirqs_off(unsigned long ip)
 455 {
 456 }
 457
 458 inline void print_irqtrace_events(struct task_struct *curr)
 459 {
 460 }
 461
 462 /*
 463  * We are only interested in hardirq on/off events:
 464  */
 465 void trace_hardirqs_on(void)
 466 {
 467         if (!preempt_trace() && irq_trace())
 468                 stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
 469 }
 470 EXPORT_SYMBOL(trace_hardirqs_on);
 471
 472 void trace_hardirqs_off(void)
 473 {
 474         if (!preempt_trace() && irq_trace())
 475                 start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
 476 }
 477 EXPORT_SYMBOL(trace_hardirqs_off);
 478
 479 __visible void trace_hardirqs_on_caller(unsigned long caller_addr)
 480 {
 481         if (!preempt_trace() && irq_trace())
 482                 stop_critical_timing(CALLER_ADDR0, caller_addr);
 483 }
 484 EXPORT_SYMBOL(trace_hardirqs_on_caller);
 485
 486 __visible void trace_hardirqs_off_caller(unsigned long caller_addr)
 487 {
 488         if (!preempt_trace() && irq_trace())
 489                 start_critical_timing(CALLER_ADDR0, caller_addr);
 490 }
 491 EXPORT_SYMBOL(trace_hardirqs_off_caller);
 492
 493 #endif /* CONFIG_PROVE_LOCKING */
 494 #endif /*  CONFIG_IRQSOFF_TRACER */
 495
 496 #ifdef CONFIG_PREEMPT_TRACER
 497 void trace_preempt_on(unsigned long a0, unsigned long a1)
 498 {
 499         if (preempt_trace() && !irq_trace())
 500                 stop_critical_timing(a0, a1);
 501 }
 502
 503 void trace_preempt_off(unsigned long a0, unsigned long a1)
 504 {
 505         if (preempt_trace() && !irq_trace())
 506                 start_critical_timing(a0, a1);
 507 }
 508 #endif /* CONFIG_PREEMPT_TRACER */
 509
 510 static int register_irqsoff_function(struct trace_array *tr, int graph, int set)
 511 {
 512         int ret;
 513
 514         /* 'set' is set if TRACE_ITER_FUNCTION is about to be set */
 515         if (function_enabled || (!set && !(trace_flags & TRACE_ITER_FUNCTION)))
 516                 return 0;
 517
 518         if (graph)
 519                 ret = register_ftrace_graph(&irqsoff_graph_return,
 520                                             &irqsoff_graph_entry);
 521         else
 522                 ret = register_ftrace_function(tr->ops);
 523
 524         if (!ret)
 525                 function_enabled = true;
 526
 527         return ret;
 528 }
 529
 530 static void unregister_irqsoff_function(struct trace_array *tr, int graph)
 531 {
 532         if (!function_enabled)
 533                 return;
 534
 535         if (graph)
 536                 unregister_ftrace_graph();
 537         else
 538                 unregister_ftrace_function(tr->ops);
 539
 540         function_enabled = false;
 541 }
 542
 543 static int irqsoff_function_set(struct trace_array *tr, int set)
 544 {
 545         if (set)
 546                 register_irqsoff_function(tr, is_graph(), 1);
 547         else
 548                 unregister_irqsoff_function(tr, is_graph());
 549         return 0;
 550 }
 551
 552 static int irqsoff_flag_changed(struct trace_array *tr, u32 mask, int set)
 553 {
 554         struct tracer *tracer = tr->current_trace;
 555
 556         if (mask & TRACE_ITER_FUNCTION)
 557                 return irqsoff_function_set(tr, set);
 558
 559         if (mask & TRACE_ITER_DISPLAY_GRAPH)
 560                 return irqsoff_display_graph(tr, set);
 561
 562         return trace_keep_overwrite(tracer, mask, set);
 563 }
 564
 565 static int start_irqsoff_tracer(struct trace_array *tr, int graph)
 566 {
 567         int ret;
 568
 569         ret = register_irqsoff_function(tr, graph, 0);
 570
 571         if (!ret && tracing_is_enabled())
 572                 tracer_enabled = 1;
 573         else
 574                 tracer_enabled = 0;
 575
 576         return ret;
 577 }
 578
 579 static void stop_irqsoff_tracer(struct trace_array *tr, int graph)
 580 {
 581         tracer_enabled = 0;
 582
 583         unregister_irqsoff_function(tr, graph);
 584 }
 585
 586 static bool irqsoff_busy;
 587
 588 static int __irqsoff_tracer_init(struct trace_array *tr)
 589 {
 590         if (irqsoff_busy)
 591                 return -EBUSY;
 592
 593         save_flags = trace_flags;
 594
 595         /* non overwrite screws up the latency tracers */
 596         set_tracer_flag(tr, TRACE_ITER_OVERWRITE, 1);
 597         set_tracer_flag(tr, TRACE_ITER_LATENCY_FMT, 1);
 598
 599         tr->max_latency = 0;
 600         irqsoff_trace = tr;
 601         /* make sure that the tracer is visible */
 602         smp_wmb();
 603         tracing_reset_online_cpus(&tr->trace_buffer);
 604
 605         ftrace_init_array_ops(tr, irqsoff_tracer_call);
 606
 607         /* Only toplevel instance supports graph tracing */
 608         if (start_irqsoff_tracer(tr, (tr->flags & TRACE_ARRAY_FL_GLOBAL &&
 609                                       is_graph())))
 610                 printk(KERN_ERR "failed to start irqsoff tracer\n");
 611
 612         irqsoff_busy = true;
 613         return 0;
 614 }
 615
 616 static void irqsoff_tracer_reset(struct trace_array *tr)
 617 {
 618         int lat_flag = save_flags & TRACE_ITER_LATENCY_FMT;
 619         int overwrite_flag = save_flags & TRACE_ITER_OVERWRITE;
 620
 621         stop_irqsoff_tracer(tr, is_graph());
 622
 623         set_tracer_flag(tr, TRACE_ITER_LATENCY_FMT, lat_flag);
 624         set_tracer_flag(tr, TRACE_ITER_OVERWRITE, overwrite_flag);
 625         ftrace_reset_array_ops(tr);
 626
 627         irqsoff_busy = false;
 628 }
 629
 630 static void irqsoff_tracer_start(struct trace_array *tr)
 631 {
 632         tracer_enabled = 1;
 633 }
 634
 635 static void irqsoff_tracer_stop(struct trace_array *tr)
 636 {
 637         tracer_enabled = 0;
 638 }
 639
 640 #ifdef CONFIG_IRQSOFF_TRACER
 641 static int irqsoff_tracer_init(struct trace_array *tr)
 642 {
 643         trace_type = TRACER_IRQS_OFF;
 644
 645         return __irqsoff_tracer_init(tr);
 646 }
 647 static struct tracer irqsoff_tracer __read_mostly =
 648 {
 649         .name           = "irqsoff",
 650         .init           = irqsoff_tracer_init,
 651         .reset          = irqsoff_tracer_reset,
 652         .start          = irqsoff_tracer_start,
 653         .stop           = irqsoff_tracer_stop,
 654         .print_max      = true,
 655         .print_header   = irqsoff_print_header,
 656         .print_line     = irqsoff_print_line,
 657         .flag_changed   = irqsoff_flag_changed,
 658 #ifdef CONFIG_FTRACE_SELFTEST
 659         .selftest    = trace_selftest_startup_irqsoff,
 660 #endif
 661         .open           = irqsoff_trace_open,
 662         .close          = irqsoff_trace_close,
 663         .allow_instances = true,
 664         .use_max_tr     = true,
 665 };
 666 # define register_irqsoff(trace) register_tracer(&trace)
 667 #else
 668 # define register_irqsoff(trace) do { } while (0)
 669 #endif
 670
 671 #ifdef CONFIG_PREEMPT_TRACER
 672 static int preemptoff_tracer_init(struct trace_array *tr)
 673 {
 674         trace_type = TRACER_PREEMPT_OFF;
 675
 676         return __irqsoff_tracer_init(tr);
 677 }
 678
 679 static struct tracer preemptoff_tracer __read_mostly =
 680 {
 681         .name           = "preemptoff",
 682         .init           = preemptoff_tracer_init,
 683         .reset          = irqsoff_tracer_reset,
 684         .start          = irqsoff_tracer_start,
 685         .stop           = irqsoff_tracer_stop,
 686         .print_max      = true,
 687         .print_header   = irqsoff_print_header,
 688         .print_line     = irqsoff_print_line,
 689         .flag_changed   = irqsoff_flag_changed,
 690 #ifdef CONFIG_FTRACE_SELFTEST
 691         .selftest    = trace_selftest_startup_preemptoff,
 692 #endif
 693         .open           = irqsoff_trace_open,
 694         .close          = irqsoff_trace_close,
 695         .allow_instances = true,
 696         .use_max_tr     = true,
 697 };
 698 # define register_preemptoff(trace) register_tracer(&trace)
 699 #else
 700 # define register_preemptoff(trace) do { } while (0)
 701 #endif
 702
 703 #if defined(CONFIG_IRQSOFF_TRACER) && \
 704         defined(CONFIG_PREEMPT_TRACER)
 705
 706 static int preemptirqsoff_tracer_init(struct trace_array *tr)
 707 {
 708         trace_type = TRACER_IRQS_OFF | TRACER_PREEMPT_OFF;
 709
 710         return __irqsoff_tracer_init(tr);
 711 }
 712
 713 static struct tracer preemptirqsoff_tracer __read_mostly =
 714 {
 715         .name           = "preemptirqsoff",
 716         .init           = preemptirqsoff_tracer_init,
 717         .reset          = irqsoff_tracer_reset,
 718         .start          = irqsoff_tracer_start,
 719         .stop           = irqsoff_tracer_stop,
 720         .print_max      = true,
 721         .print_header   = irqsoff_print_header,
 722         .print_line     = irqsoff_print_line,
 723         .flag_changed   = irqsoff_flag_changed,
 724 #ifdef CONFIG_FTRACE_SELFTEST
 725         .selftest    = trace_selftest_startup_preemptirqsoff,
 726 #endif
 727         .open           = irqsoff_trace_open,
 728         .close          = irqsoff_trace_close,
 729         .allow_instances = true,
 730         .use_max_tr     = true,
 731 };
 732
 733 # define register_preemptirqsoff(trace) register_tracer(&trace)
 734 #else
 735 # define register_preemptirqsoff(trace) do { } while (0)
 736 #endif
 737
 738 __init static int init_irqsoff_tracer(void)
 739 {
 740         register_irqsoff(irqsoff_tracer);
 741         register_preemptoff(preemptoff_tracer);
 742         register_preemptirqsoff(preemptirqsoff_tracer);
 743
 744         return 0;
 745 }
 746 core_initcall(init_irqsoff_tracer);