kernel/trace/trace_sched_wakeup.c

   1 /*
   2  * trace task wakeup timings
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
   6  *
   7  * Based on code from the latency_tracer, that is:
   8  *
   9  *  Copyright (C) 2004-2006 Ingo Molnar
  10  *  Copyright (C) 2004 Nadia Yvette Chambers
  11  */
  12 #include <linux/module.h>
  13 #include <linux/fs.h>
  14 #include <linux/debugfs.h>
  15 #include <linux/kallsyms.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/ftrace.h>
  18 #include <linux/sched/rt.h>
  19 #include <trace/events/sched.h>
  20 #include "trace.h"
  21
  22 static struct trace_array       *wakeup_trace;
  23 static int __read_mostly        tracer_enabled;
  24
  25 static struct task_struct       *wakeup_task;
  26 static int                      wakeup_cpu;
  27 static int                      wakeup_current_cpu;
  28 static unsigned                 wakeup_prio = -1;
  29 static int                      wakeup_rt;
  30
  31 static arch_spinlock_t wakeup_lock =
  32         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
  33
  34 static void wakeup_reset(struct trace_array *tr);
  35 static void __wakeup_reset(struct trace_array *tr);
  36 static int wakeup_graph_entry(struct ftrace_graph_ent *trace);
  37 static void wakeup_graph_return(struct ftrace_graph_ret *trace);
  38
  39 static int save_flags;
  40
  41 #define TRACE_DISPLAY_GRAPH     1
  42
  43 static struct tracer_opt trace_opts[] = {
  44 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  45         /* display latency trace as call graph */
  46         { TRACER_OPT(display-graph, TRACE_DISPLAY_GRAPH) },
  47 #endif
  48         { } /* Empty entry */
  49 };
  50
  51 static struct tracer_flags tracer_flags = {
  52         .val  = 0,
  53         .opts = trace_opts,
  54 };
  55
  56 #define is_graph() (tracer_flags.val & TRACE_DISPLAY_GRAPH)
  57
  58 #ifdef CONFIG_FUNCTION_TRACER
  59
  60 /*
  61  * Prologue for the wakeup function tracers.
  62  *
  63  * Returns 1 if it is OK to continue, and preemption
  64  *            is disabled and data->disabled is incremented.
  65  *         0 if the trace is to be ignored, and preemption
  66  *            is not disabled and data->disabled is
  67  *            kept the same.
  68  *
  69  * Note, this function is also used outside this ifdef but
  70  *  inside the #ifdef of the function graph tracer below.
  71  *  This is OK, since the function graph tracer is
  72  *  dependent on the function tracer.
  73  */
  74 static int
  75 func_prolog_preempt_disable(struct trace_array *tr,
  76                             struct trace_array_cpu **data,
  77                             int *pc)
  78 {
  79         long disabled;
  80         int cpu;
  81
  82         if (likely(!wakeup_task))
  83                 return 0;
  84
  85         *pc = preempt_count();
  86         preempt_disable_notrace();
  87
  88         cpu = raw_smp_processor_id();
  89         if (cpu != wakeup_current_cpu)
  90                 goto out_enable;
  91
  92         *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
  93         disabled = atomic_inc_return(&(*data)->disabled);
  94         if (unlikely(disabled != 1))
  95                 goto out;
  96
  97         return 1;
  98
  99 out:
 100         atomic_dec(&(*data)->disabled);
 101
 102 out_enable:
 103         preempt_enable_notrace();
 104         return 0;
 105 }
 106
 107 /*
 108  * wakeup uses its own tracer function to keep the overhead down:
 109  */
 110 static void
 111 wakeup_tracer_call(unsigned long ip, unsigned long parent_ip,
 112                    struct ftrace_ops *op, struct pt_regs *pt_regs)
 113 {
 114         struct trace_array *tr = wakeup_trace;
 115         struct trace_array_cpu *data;
 116         unsigned long flags;
 117         int pc;
 118
 119         if (!func_prolog_preempt_disable(tr, &data, &pc))
 120                 return;
 121
 122         local_irq_save(flags);
 123         trace_function(tr, ip, parent_ip, flags, pc);
 124         local_irq_restore(flags);
 125
 126         atomic_dec(&data->disabled);
 127         preempt_enable_notrace();
 128 }
 129
 130 static struct ftrace_ops trace_ops __read_mostly =
 131 {
 132         .func = wakeup_tracer_call,
 133         .flags = FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_RECURSION_SAFE,
 134 };
 135 #endif /* CONFIG_FUNCTION_TRACER */
 136
 137 static int start_func_tracer(int graph)
 138 {
 139         int ret;
 140
 141         if (!graph)
 142                 ret = register_ftrace_function(&trace_ops);
 143         else
 144                 ret = register_ftrace_graph(&wakeup_graph_return,
 145                                             &wakeup_graph_entry);
 146
 147         if (!ret && tracing_is_enabled())
 148                 tracer_enabled = 1;
 149         else
 150                 tracer_enabled = 0;
 151
 152         return ret;
 153 }
 154
 155 static void stop_func_tracer(int graph)
 156 {
 157         tracer_enabled = 0;
 158
 159         if (!graph)
 160                 unregister_ftrace_function(&trace_ops);
 161         else
 162                 unregister_ftrace_graph();
 163 }
 164
 165 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 166 static int wakeup_set_flag(u32 old_flags, u32 bit, int set)
 167 {
 168
 169         if (!(bit & TRACE_DISPLAY_GRAPH))
 170                 return -EINVAL;
 171
 172         if (!(is_graph() ^ set))
 173                 return 0;
 174
 175         stop_func_tracer(!set);
 176
 177         wakeup_reset(wakeup_trace);
 178         tracing_max_latency = 0;
 179
 180         return start_func_tracer(set);
 181 }
 182
 183 static int wakeup_graph_entry(struct ftrace_graph_ent *trace)
 184 {
 185         struct trace_array *tr = wakeup_trace;
 186         struct trace_array_cpu *data;
 187         unsigned long flags;
 188         int pc, ret = 0;
 189
 190         if (!func_prolog_preempt_disable(tr, &data, &pc))
 191                 return 0;
 192
 193         local_save_flags(flags);
 194         ret = __trace_graph_entry(tr, trace, flags, pc);
 195         atomic_dec(&data->disabled);
 196         preempt_enable_notrace();
 197
 198         return ret;
 199 }
 200
 201 static void wakeup_graph_return(struct ftrace_graph_ret *trace)
 202 {
 203         struct trace_array *tr = wakeup_trace;
 204         struct trace_array_cpu *data;
 205         unsigned long flags;
 206         int pc;
 207
 208         if (!func_prolog_preempt_disable(tr, &data, &pc))
 209                 return;
 210
 211         local_save_flags(flags);
 212         __trace_graph_return(tr, trace, flags, pc);
 213         atomic_dec(&data->disabled);
 214
 215         preempt_enable_notrace();
 216         return;
 217 }
 218
 219 static void wakeup_trace_open(struct trace_iterator *iter)
 220 {
 221         if (is_graph())
 222                 graph_trace_open(iter);
 223 }
 224
 225 static void wakeup_trace_close(struct trace_iterator *iter)
 226 {
 227         if (iter->private)
 228                 graph_trace_close(iter);
 229 }
 230
 231 #define GRAPH_TRACER_FLAGS (TRACE_GRAPH_PRINT_PROC | \
 232                             TRACE_GRAPH_PRINT_ABS_TIME | \
 233                             TRACE_GRAPH_PRINT_DURATION)
 234
 235 static enum print_line_t wakeup_print_line(struct trace_iterator *iter)
 236 {
 237         /*
 238          * In graph mode call the graph tracer output function,
 239          * otherwise go with the TRACE_FN event handler
 240          */
 241         if (is_graph())
 242                 return print_graph_function_flags(iter, GRAPH_TRACER_FLAGS);
 243
 244         return TRACE_TYPE_UNHANDLED;
 245 }
 246
 247 static void wakeup_print_header(struct seq_file *s)
 248 {
 249         if (is_graph())
 250                 print_graph_headers_flags(s, GRAPH_TRACER_FLAGS);
 251         else
 252                 trace_default_header(s);
 253 }
 254
 255 static void
 256 __trace_function(struct trace_array *tr,
 257                  unsigned long ip, unsigned long parent_ip,
 258                  unsigned long flags, int pc)
 259 {
 260         if (is_graph())
 261                 trace_graph_function(tr, ip, parent_ip, flags, pc);
 262         else
 263                 trace_function(tr, ip, parent_ip, flags, pc);
 264 }
 265 #else
 266 #define __trace_function trace_function
 267
 268 static int wakeup_set_flag(u32 old_flags, u32 bit, int set)
 269 {
 270         return -EINVAL;
 271 }
 272
 273 static int wakeup_graph_entry(struct ftrace_graph_ent *trace)
 274 {
 275         return -1;
 276 }
 277
 278 static enum print_line_t wakeup_print_line(struct trace_iterator *iter)
 279 {
 280         return TRACE_TYPE_UNHANDLED;
 281 }
 282
 283 static void wakeup_graph_return(struct ftrace_graph_ret *trace) { }
 284 static void wakeup_trace_open(struct trace_iterator *iter) { }
 285 static void wakeup_trace_close(struct trace_iterator *iter) { }
 286
 287 #ifdef CONFIG_FUNCTION_TRACER
 288 static void wakeup_print_header(struct seq_file *s)
 289 {
 290         trace_default_header(s);
 291 }
 292 #else
 293 static void wakeup_print_header(struct seq_file *s)
 294 {
 295         trace_latency_header(s);
 296 }
 297 #endif /* CONFIG_FUNCTION_TRACER */
 298 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 299
 300 /*
 301  * Should this new latency be reported/recorded?
 302  */
 303 static int report_latency(cycle_t delta)
 304 {
 305         if (tracing_thresh) {
 306                 if (delta < tracing_thresh)
 307                         return 0;
 308         } else {
 309                 if (delta <= tracing_max_latency)
 310                         return 0;
 311         }
 312         return 1;
 313 }
 314
 315 static void
 316 probe_wakeup_migrate_task(void *ignore, struct task_struct *task, int cpu)
 317 {
 318         if (task != wakeup_task)
 319                 return;
 320
 321         wakeup_current_cpu = cpu;
 322 }
 323
 324 static void notrace
 325 probe_wakeup_sched_switch(void *ignore,
 326                           struct task_struct *prev, struct task_struct *next)
 327 {
 328         struct trace_array_cpu *data;
 329         cycle_t T0, T1, delta;
 330         unsigned long flags;
 331         long disabled;
 332         int cpu;
 333         int pc;
 334
 335         tracing_record_cmdline(prev);
 336
 337         if (unlikely(!tracer_enabled))
 338                 return;
 339
 340         /*
 341          * When we start a new trace, we set wakeup_task to NULL
 342          * and then set tracer_enabled = 1. We want to make sure
 343          * that another CPU does not see the tracer_enabled = 1
 344          * and the wakeup_task with an older task, that might
 345          * actually be the same as next.
 346          */
 347         smp_rmb();
 348
 349         if (next != wakeup_task)
 350                 return;
 351
 352         pc = preempt_count();
 353
 354         /* disable local data, not wakeup_cpu data */
 355         cpu = raw_smp_processor_id();
 356         disabled = atomic_inc_return(&per_cpu_ptr(wakeup_trace->trace_buffer.data, cpu)->disabled);
 357         if (likely(disabled != 1))
 358                 goto out;
 359
 360         local_irq_save(flags);
 361         arch_spin_lock(&wakeup_lock);
 362
 363         /* We could race with grabbing wakeup_lock */
 364         if (unlikely(!tracer_enabled || next != wakeup_task))
 365                 goto out_unlock;
 366
 367         /* The task we are waiting for is waking up */
 368         data = per_cpu_ptr(wakeup_trace->trace_buffer.data, wakeup_cpu);
 369
 370         __trace_function(wakeup_trace, CALLER_ADDR0, CALLER_ADDR1, flags, pc);
 371         tracing_sched_switch_trace(wakeup_trace, prev, next, flags, pc);
 372
 373         T0 = data->preempt_timestamp;
 374         T1 = ftrace_now(cpu);
 375         delta = T1-T0;
 376
 377         if (!report_latency(delta))
 378                 goto out_unlock;
 379
 380         if (likely(!is_tracing_stopped())) {
 381                 tracing_max_latency = delta;
 382                 update_max_tr(wakeup_trace, wakeup_task, wakeup_cpu);
 383         }
 384
 385 out_unlock:
 386         __wakeup_reset(wakeup_trace);
 387         arch_spin_unlock(&wakeup_lock);
 388         local_irq_restore(flags);
 389 out:
 390         atomic_dec(&per_cpu_ptr(wakeup_trace->trace_buffer.data, cpu)->disabled);
 391 }
 392
 393 static void __wakeup_reset(struct trace_array *tr)
 394 {
 395         wakeup_cpu = -1;
 396         wakeup_prio = -1;
 397
 398         if (wakeup_task)
 399                 put_task_struct(wakeup_task);
 400
 401         wakeup_task = NULL;
 402 }
 403
 404 static void wakeup_reset(struct trace_array *tr)
 405 {
 406         unsigned long flags;
 407
 408         tracing_reset_online_cpus(&tr->trace_buffer);
 409
 410         local_irq_save(flags);
 411         arch_spin_lock(&wakeup_lock);
 412         __wakeup_reset(tr);
 413         arch_spin_unlock(&wakeup_lock);
 414         local_irq_restore(flags);
 415 }
 416
 417 static void
 418 probe_wakeup(void *ignore, struct task_struct *p, int success)
 419 {
 420         struct trace_array_cpu *data;
 421         int cpu = smp_processor_id();
 422         unsigned long flags;
 423         long disabled;
 424         int pc;
 425
 426         if (likely(!tracer_enabled))
 427                 return;
 428
 429         tracing_record_cmdline(p);
 430         tracing_record_cmdline(current);
 431
 432         if ((wakeup_rt && !rt_task(p)) ||
 433                         p->prio >= wakeup_prio ||
 434                         p->prio >= current->prio)
 435                 return;
 436
 437         pc = preempt_count();
 438         disabled = atomic_inc_return(&per_cpu_ptr(wakeup_trace->trace_buffer.data, cpu)->disabled);
 439         if (unlikely(disabled != 1))
 440                 goto out;
 441
 442         /* interrupts should be off from try_to_wake_up */
 443         arch_spin_lock(&wakeup_lock);
 444
 445         /* check for races. */
 446         if (!tracer_enabled || p->prio >= wakeup_prio)
 447                 goto out_locked;
 448
 449         /* reset the trace */
 450         __wakeup_reset(wakeup_trace);
 451
 452         wakeup_cpu = task_cpu(p);
 453         wakeup_current_cpu = wakeup_cpu;
 454         wakeup_prio = p->prio;
 455
 456         wakeup_task = p;
 457         get_task_struct(wakeup_task);
 458
 459         local_save_flags(flags);
 460
 461         data = per_cpu_ptr(wakeup_trace->trace_buffer.data, wakeup_cpu);
 462         data->preempt_timestamp = ftrace_now(cpu);
 463         tracing_sched_wakeup_trace(wakeup_trace, p, current, flags, pc);
 464
 465         /*
 466          * We must be careful in using CALLER_ADDR2. But since wake_up
 467          * is not called by an assembly function  (where as schedule is)
 468          * it should be safe to use it here.
 469          */
 470         __trace_function(wakeup_trace, CALLER_ADDR1, CALLER_ADDR2, flags, pc);
 471
 472 out_locked:
 473         arch_spin_unlock(&wakeup_lock);
 474 out:
 475         atomic_dec(&per_cpu_ptr(wakeup_trace->trace_buffer.data, cpu)->disabled);
 476 }
 477
 478 static void start_wakeup_tracer(struct trace_array *tr)
 479 {
 480         int ret;
 481
 482         ret = register_trace_sched_wakeup(probe_wakeup, NULL);
 483         if (ret) {
 484                 pr_info("wakeup trace: Couldn't activate tracepoint"
 485                         " probe to kernel_sched_wakeup\n");
 486                 return;
 487         }
 488
 489         ret = register_trace_sched_wakeup_new(probe_wakeup, NULL);
 490         if (ret) {
 491                 pr_info("wakeup trace: Couldn't activate tracepoint"
 492                         " probe to kernel_sched_wakeup_new\n");
 493                 goto fail_deprobe;
 494         }
 495
 496         ret = register_trace_sched_switch(probe_wakeup_sched_switch, NULL);
 497         if (ret) {
 498                 pr_info("sched trace: Couldn't activate tracepoint"
 499                         " probe to kernel_sched_switch\n");
 500                 goto fail_deprobe_wake_new;
 501         }
 502
 503         ret = register_trace_sched_migrate_task(probe_wakeup_migrate_task, NULL);
 504         if (ret) {
 505                 pr_info("wakeup trace: Couldn't activate tracepoint"
 506                         " probe to kernel_sched_migrate_task\n");
 507                 return;
 508         }
 509
 510         wakeup_reset(tr);
 511
 512         /*
 513          * Don't let the tracer_enabled = 1 show up before
 514          * the wakeup_task is reset. This may be overkill since
 515          * wakeup_reset does a spin_unlock after setting the
 516          * wakeup_task to NULL, but I want to be safe.
 517          * This is a slow path anyway.
 518          */
 519         smp_wmb();
 520
 521         if (start_func_tracer(is_graph()))
 522                 printk(KERN_ERR "failed to start wakeup tracer\n");
 523
 524         return;
 525 fail_deprobe_wake_new:
 526         unregister_trace_sched_wakeup_new(probe_wakeup, NULL);
 527 fail_deprobe:
 528         unregister_trace_sched_wakeup(probe_wakeup, NULL);
 529 }
 530
 531 static void stop_wakeup_tracer(struct trace_array *tr)
 532 {
 533         tracer_enabled = 0;
 534         stop_func_tracer(is_graph());
 535         unregister_trace_sched_switch(probe_wakeup_sched_switch, NULL);
 536         unregister_trace_sched_wakeup_new(probe_wakeup, NULL);
 537         unregister_trace_sched_wakeup(probe_wakeup, NULL);
 538         unregister_trace_sched_migrate_task(probe_wakeup_migrate_task, NULL);
 539 }
 540
 541 static int __wakeup_tracer_init(struct trace_array *tr)
 542 {
 543         save_flags = trace_flags;
 544
 545         /* non overwrite screws up the latency tracers */
 546         set_tracer_flag(tr, TRACE_ITER_OVERWRITE, 1);
 547         set_tracer_flag(tr, TRACE_ITER_LATENCY_FMT, 1);
 548
 549         tracing_max_latency = 0;
 550         wakeup_trace = tr;
 551         start_wakeup_tracer(tr);
 552         return 0;
 553 }
 554
 555 static int wakeup_tracer_init(struct trace_array *tr)
 556 {
 557         wakeup_rt = 0;
 558         return __wakeup_tracer_init(tr);
 559 }
 560
 561 static int wakeup_rt_tracer_init(struct trace_array *tr)
 562 {
 563         wakeup_rt = 1;
 564         return __wakeup_tracer_init(tr);
 565 }
 566
 567 static void wakeup_tracer_reset(struct trace_array *tr)
 568 {
 569         int lat_flag = save_flags & TRACE_ITER_LATENCY_FMT;
 570         int overwrite_flag = save_flags & TRACE_ITER_OVERWRITE;
 571
 572         stop_wakeup_tracer(tr);
 573         /* make sure we put back any tasks we are tracing */
 574         wakeup_reset(tr);
 575
 576         set_tracer_flag(tr, TRACE_ITER_LATENCY_FMT, lat_flag);
 577         set_tracer_flag(tr, TRACE_ITER_OVERWRITE, overwrite_flag);
 578 }
 579
 580 static void wakeup_tracer_start(struct trace_array *tr)
 581 {
 582         wakeup_reset(tr);
 583         tracer_enabled = 1;
 584 }
 585
 586 static void wakeup_tracer_stop(struct trace_array *tr)
 587 {
 588         tracer_enabled = 0;
 589 }
 590
 591 static struct tracer wakeup_tracer __read_mostly =
 592 {
 593         .name           = "wakeup",
 594         .init           = wakeup_tracer_init,
 595         .reset          = wakeup_tracer_reset,
 596         .start          = wakeup_tracer_start,
 597         .stop           = wakeup_tracer_stop,
 598         .print_max      = true,
 599         .print_header   = wakeup_print_header,
 600         .print_line     = wakeup_print_line,
 601         .flags          = &tracer_flags,
 602         .set_flag       = wakeup_set_flag,
 603         .flag_changed   = trace_keep_overwrite,
 604 #ifdef CONFIG_FTRACE_SELFTEST
 605         .selftest    = trace_selftest_startup_wakeup,
 606 #endif
 607         .open           = wakeup_trace_open,
 608         .close          = wakeup_trace_close,
 609         .use_max_tr     = true,
 610 };
 611
 612 static struct tracer wakeup_rt_tracer __read_mostly =
 613 {
 614         .name           = "wakeup_rt",
 615         .init           = wakeup_rt_tracer_init,
 616         .reset          = wakeup_tracer_reset,
 617         .start          = wakeup_tracer_start,
 618         .stop           = wakeup_tracer_stop,
 619         .wait_pipe      = poll_wait_pipe,
 620         .print_max      = true,
 621         .print_header   = wakeup_print_header,
 622         .print_line     = wakeup_print_line,
 623         .flags          = &tracer_flags,
 624         .set_flag       = wakeup_set_flag,
 625         .flag_changed   = trace_keep_overwrite,
 626 #ifdef CONFIG_FTRACE_SELFTEST
 627         .selftest    = trace_selftest_startup_wakeup,
 628 #endif
 629         .open           = wakeup_trace_open,
 630         .close          = wakeup_trace_close,
 631         .use_max_tr     = true,
 632 };
 633
 634 __init static int init_wakeup_tracer(void)
 635 {
 636         int ret;
 637
 638         ret = register_tracer(&wakeup_tracer);
 639         if (ret)
 640                 return ret;
 641
 642         ret = register_tracer(&wakeup_rt_tracer);
 643         if (ret)
 644                 return ret;
 645
 646         return 0;
 647 }
 648 core_initcall(init_wakeup_tracer);