kernel/sched/idle.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Generic entry points for the idle threads and
   4  * implementation of the idle task scheduling class.
   5  *
   6  * (NOTE: these are not related to SCHED_IDLE batch scheduled
   7  *        tasks which are handled in sched/fair.c )
   8  */
   9
  10 /* Linker adds these: start and end of __cpuidle functions */
  11 extern char __cpuidle_text_start[], __cpuidle_text_end[];
  12
  13 /**
  14  * sched_idle_set_state - Record idle state for the current CPU.
  15  * @idle_state: State to record.
  16  */
  17 void sched_idle_set_state(struct cpuidle_state *idle_state)
  18 {
  19         idle_set_state(this_rq(), idle_state);
  20 }
  21
  22 static int __read_mostly cpu_idle_force_poll;
  23
  24 void cpu_idle_poll_ctrl(bool enable)
  25 {
  26         if (enable) {
  27                 cpu_idle_force_poll++;
  28         } else {
  29                 cpu_idle_force_poll--;
  30                 WARN_ON_ONCE(cpu_idle_force_poll < 0);
  31         }
  32 }
  33
  34 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
  35 static int __init cpu_idle_poll_setup(char *__unused)
  36 {
  37         cpu_idle_force_poll = 1;
  38
  39         return 1;
  40 }
  41 __setup("nohlt", cpu_idle_poll_setup);
  42
  43 static int __init cpu_idle_nopoll_setup(char *__unused)
  44 {
  45         cpu_idle_force_poll = 0;
  46
  47         return 1;
  48 }
  49 __setup("hlt", cpu_idle_nopoll_setup);
  50 #endif
  51
  52 static noinline int __cpuidle cpu_idle_poll(void)
  53 {
  54         instrumentation_begin();
  55         trace_cpu_idle(0, smp_processor_id());
  56         stop_critical_timings();
  57         ct_cpuidle_enter();
  58
  59         raw_local_irq_enable();
  60         while (!tif_need_resched() &&
  61                (cpu_idle_force_poll || tick_check_broadcast_expired()))
  62                 cpu_relax();
  63         raw_local_irq_disable();
  64
  65         ct_cpuidle_exit();
  66         start_critical_timings();
  67         trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
  68         local_irq_enable();
  69         instrumentation_end();
  70
  71         return 1;
  72 }
  73
  74 /* Weak implementations for optional arch specific functions */
  75 void __weak arch_cpu_idle_prepare(void) { }
  76 void __weak arch_cpu_idle_enter(void) { }
  77 void __weak arch_cpu_idle_exit(void) { }
  78 void __weak __noreturn arch_cpu_idle_dead(void) { while (1); }
  79 void __weak arch_cpu_idle(void)
  80 {
  81         cpu_idle_force_poll = 1;
  82 }
  83
  84 /**
  85  * default_idle_call - Default CPU idle routine.
  86  *
  87  * To use when the cpuidle framework cannot be used.
  88  */
  89 void __cpuidle default_idle_call(void)
  90 {
  91         instrumentation_begin();
  92         if (!current_clr_polling_and_test()) {
  93                 trace_cpu_idle(1, smp_processor_id());
  94                 stop_critical_timings();
  95
  96                 ct_cpuidle_enter();
  97                 arch_cpu_idle();
  98                 ct_cpuidle_exit();
  99
 100                 start_critical_timings();
 101                 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 102         }
 103         local_irq_enable();
 104         instrumentation_end();
 105 }
 106
 107 static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
 108                                struct cpuidle_device *dev)
 109 {
 110         if (current_clr_polling_and_test())
 111                 return -EBUSY;
 112
 113         return cpuidle_enter_s2idle(drv, dev);
 114 }
 115
 116 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 117                       int next_state)
 118 {
 119         /*
 120          * The idle task must be scheduled, it is pointless to go to idle, just
 121          * update no idle residency and return.
 122          */
 123         if (current_clr_polling_and_test()) {
 124                 dev->last_residency_ns = 0;
 125                 local_irq_enable();
 126                 return -EBUSY;
 127         }
 128
 129         /*
 130          * Enter the idle state previously returned by the governor decision.
 131          * This function will block until an interrupt occurs and will take
 132          * care of re-enabling the local interrupts
 133          */
 134         return cpuidle_enter(drv, dev, next_state);
 135 }
 136
 137 /**
 138  * cpuidle_idle_call - the main idle function
 139  *
 140  * NOTE: no locks or semaphores should be used here
 141  *
 142  * On architectures that support TIF_POLLING_NRFLAG, is called with polling
 143  * set, and it returns with polling set.  If it ever stops polling, it
 144  * must clear the polling bit.
 145  */
 146 static void cpuidle_idle_call(void)
 147 {
 148         struct cpuidle_device *dev = cpuidle_get_device();
 149         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 150         int next_state, entered_state;
 151
 152         /*
 153          * Check if the idle task must be rescheduled. If it is the
 154          * case, exit the function after re-enabling the local irq.
 155          */
 156         if (need_resched()) {
 157                 local_irq_enable();
 158                 return;
 159         }
 160
 161         /*
 162          * The RCU framework needs to be told that we are entering an idle
 163          * section, so no more rcu read side critical sections and one more
 164          * step to the grace period
 165          */
 166
 167         if (cpuidle_not_available(drv, dev)) {
 168                 tick_nohz_idle_stop_tick();
 169
 170                 default_idle_call();
 171                 goto exit_idle;
 172         }
 173
 174         /*
 175          * Suspend-to-idle ("s2idle") is a system state in which all user space
 176          * has been frozen, all I/O devices have been suspended and the only
 177          * activity happens here and in interrupts (if any). In that case bypass
 178          * the cpuidle governor and go straight for the deepest idle state
 179          * available.  Possibly also suspend the local tick and the entire
 180          * timekeeping to prevent timer interrupts from kicking us out of idle
 181          * until a proper wakeup interrupt happens.
 182          */
 183
 184         if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
 185                 u64 max_latency_ns;
 186
 187                 if (idle_should_enter_s2idle()) {
 188
 189                         entered_state = call_cpuidle_s2idle(drv, dev);
 190                         if (entered_state > 0)
 191                                 goto exit_idle;
 192
 193                         max_latency_ns = U64_MAX;
 194                 } else {
 195                         max_latency_ns = dev->forced_idle_latency_limit_ns;
 196                 }
 197
 198                 tick_nohz_idle_stop_tick();
 199
 200                 next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
 201                 call_cpuidle(drv, dev, next_state);
 202         } else {
 203                 bool stop_tick = true;
 204
 205                 /*
 206                  * Ask the cpuidle framework to choose a convenient idle state.
 207                  */
 208                 next_state = cpuidle_select(drv, dev, &stop_tick);
 209
 210                 if (stop_tick || tick_nohz_tick_stopped())
 211                         tick_nohz_idle_stop_tick();
 212                 else
 213                         tick_nohz_idle_retain_tick();
 214
 215                 entered_state = call_cpuidle(drv, dev, next_state);
 216                 /*
 217                  * Give the governor an opportunity to reflect on the outcome
 218                  */
 219                 cpuidle_reflect(dev, entered_state);
 220         }
 221
 222 exit_idle:
 223         __current_set_polling();
 224
 225         /*
 226          * It is up to the idle functions to reenable local interrupts
 227          */
 228         if (WARN_ON_ONCE(irqs_disabled()))
 229                 local_irq_enable();
 230 }
 231
 232 /*
 233  * Generic idle loop implementation
 234  *
 235  * Called with polling cleared.
 236  */
 237 static void do_idle(void)
 238 {
 239         int cpu = smp_processor_id();
 240
 241         /*
 242          * Check if we need to update blocked load
 243          */
 244         nohz_run_idle_balance(cpu);
 245
 246         /*
 247          * If the arch has a polling bit, we maintain an invariant:
 248          *
 249          * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
 250          * rq->idle). This means that, if rq->idle has the polling bit set,
 251          * then setting need_resched is guaranteed to cause the CPU to
 252          * reschedule.
 253          */
 254
 255         __current_set_polling();
 256         tick_nohz_idle_enter();
 257
 258         while (!need_resched()) {
 259                 rmb();
 260
 261                 local_irq_disable();
 262
 263                 if (cpu_is_offline(cpu)) {
 264                         tick_nohz_idle_stop_tick();
 265                         cpuhp_report_idle_dead();
 266                         arch_cpu_idle_dead();
 267                 }
 268
 269                 arch_cpu_idle_enter();
 270                 rcu_nocb_flush_deferred_wakeup();
 271
 272                 /*
 273                  * In poll mode we reenable interrupts and spin. Also if we
 274                  * detected in the wakeup from idle path that the tick
 275                  * broadcast device expired for us, we don't want to go deep
 276                  * idle as we know that the IPI is going to arrive right away.
 277                  */
 278                 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
 279                         tick_nohz_idle_restart_tick();
 280                         cpu_idle_poll();
 281                 } else {
 282                         cpuidle_idle_call();
 283                 }
 284                 arch_cpu_idle_exit();
 285         }
 286
 287         /*
 288          * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
 289          * be set, propagate it into PREEMPT_NEED_RESCHED.
 290          *
 291          * This is required because for polling idle loops we will not have had
 292          * an IPI to fold the state for us.
 293          */
 294         preempt_set_need_resched();
 295         tick_nohz_idle_exit();
 296         __current_clr_polling();
 297
 298         /*
 299          * We promise to call sched_ttwu_pending() and reschedule if
 300          * need_resched() is set while polling is set. That means that clearing
 301          * polling needs to be visible before doing these things.
 302          */
 303         smp_mb__after_atomic();
 304
 305         /*
 306          * RCU relies on this call to be done outside of an RCU read-side
 307          * critical section.
 308          */
 309         flush_smp_call_function_queue();
 310         schedule_idle();
 311
 312         if (unlikely(klp_patch_pending(current)))
 313                 klp_update_patch_state(current);
 314 }
 315
 316 bool cpu_in_idle(unsigned long pc)
 317 {
 318         return pc >= (unsigned long)__cpuidle_text_start &&
 319                 pc < (unsigned long)__cpuidle_text_end;
 320 }
 321
 322 struct idle_timer {
 323         struct hrtimer timer;
 324         int done;
 325 };
 326
 327 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
 328 {
 329         struct idle_timer *it = container_of(timer, struct idle_timer, timer);
 330
 331         WRITE_ONCE(it->done, 1);
 332         set_tsk_need_resched(current);
 333
 334         return HRTIMER_NORESTART;
 335 }
 336
 337 void play_idle_precise(u64 duration_ns, u64 latency_ns)
 338 {
 339         struct idle_timer it;
 340
 341         /*
 342          * Only FIFO tasks can disable the tick since they don't need the forced
 343          * preemption.
 344          */
 345         WARN_ON_ONCE(current->policy != SCHED_FIFO);
 346         WARN_ON_ONCE(current->nr_cpus_allowed != 1);
 347         WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
 348         WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
 349         WARN_ON_ONCE(!duration_ns);
 350         WARN_ON_ONCE(current->mm);
 351
 352         rcu_sleep_check();
 353         preempt_disable();
 354         current->flags |= PF_IDLE;
 355         cpuidle_use_deepest_state(latency_ns);
 356
 357         it.done = 0;
 358         hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
 359         it.timer.function = idle_inject_timer_fn;
 360         hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
 361                       HRTIMER_MODE_REL_PINNED_HARD);
 362
 363         while (!READ_ONCE(it.done))
 364                 do_idle();
 365
 366         cpuidle_use_deepest_state(0);
 367         current->flags &= ~PF_IDLE;
 368
 369         preempt_fold_need_resched();
 370         preempt_enable();
 371 }
 372 EXPORT_SYMBOL_GPL(play_idle_precise);
 373
 374 void cpu_startup_entry(enum cpuhp_state state)
 375 {
 376         arch_cpu_idle_prepare();
 377         cpuhp_online_idle(state);
 378         while (1)
 379                 do_idle();
 380 }
 381
 382 /*
 383  * idle-task scheduling class.
 384  */
 385
 386 #ifdef CONFIG_SMP
 387 static int
 388 select_task_rq_idle(struct task_struct *p, int cpu, int flags)
 389 {
 390         return task_cpu(p); /* IDLE tasks as never migrated */
 391 }
 392
 393 static int
 394 balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 395 {
 396         return WARN_ON_ONCE(1);
 397 }
 398 #endif
 399
 400 /*
 401  * Idle tasks are unconditionally rescheduled:
 402  */
 403 static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
 404 {
 405         resched_curr(rq);
 406 }
 407
 408 static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
 409 {
 410 }
 411
 412 static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
 413 {
 414         update_idle_core(rq);
 415         schedstat_inc(rq->sched_goidle);
 416 }
 417
 418 #ifdef CONFIG_SMP
 419 static struct task_struct *pick_task_idle(struct rq *rq)
 420 {
 421         return rq->idle;
 422 }
 423 #endif
 424
 425 struct task_struct *pick_next_task_idle(struct rq *rq)
 426 {
 427         struct task_struct *next = rq->idle;
 428
 429         set_next_task_idle(rq, next, true);
 430
 431         return next;
 432 }
 433
 434 /*
 435  * It is not legal to sleep in the idle task - print a warning
 436  * message if some code attempts to do it:
 437  */
 438 static void
 439 dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
 440 {
 441         raw_spin_rq_unlock_irq(rq);
 442         printk(KERN_ERR "bad: scheduling from the idle thread!\n");
 443         dump_stack();
 444         raw_spin_rq_lock_irq(rq);
 445 }
 446
 447 /*
 448  * scheduler tick hitting a task of our scheduling class.
 449  *
 450  * NOTE: This function can be called remotely by the tick offload that
 451  * goes along full dynticks. Therefore no local assumption can be made
 452  * and everything must be accessed through the @rq and @curr passed in
 453  * parameters.
 454  */
 455 static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
 456 {
 457 }
 458
 459 static void switched_to_idle(struct rq *rq, struct task_struct *p)
 460 {
 461         BUG();
 462 }
 463
 464 static void
 465 prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
 466 {
 467         BUG();
 468 }
 469
 470 static void update_curr_idle(struct rq *rq)
 471 {
 472 }
 473
 474 /*
 475  * Simple, special scheduling class for the per-CPU idle tasks:
 476  */
 477 DEFINE_SCHED_CLASS(idle) = {
 478
 479         /* no enqueue/yield_task for idle tasks */
 480
 481         /* dequeue is not valid, we print a debug message there: */
 482         .dequeue_task           = dequeue_task_idle,
 483
 484         .check_preempt_curr     = check_preempt_curr_idle,
 485
 486         .pick_next_task         = pick_next_task_idle,
 487         .put_prev_task          = put_prev_task_idle,
 488         .set_next_task          = set_next_task_idle,
 489
 490 #ifdef CONFIG_SMP
 491         .balance                = balance_idle,
 492         .pick_task              = pick_task_idle,
 493         .select_task_rq         = select_task_rq_idle,
 494         .set_cpus_allowed       = set_cpus_allowed_common,
 495 #endif
 496
 497         .task_tick              = task_tick_idle,
 498
 499         .prio_changed           = prio_changed_idle,
 500         .switched_to            = switched_to_idle,
 501         .update_curr            = update_curr_idle,
 502 };