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 #include "sched.h"
  10
  11 #include <trace/events/power.h>
  12
  13 /* Linker adds these: start and end of __cpuidle functions */
  14 extern char __cpuidle_text_start[], __cpuidle_text_end[];
  15
  16 /**
  17  * sched_idle_set_state - Record idle state for the current CPU.
  18  * @idle_state: State to record.
  19  */
  20 void sched_idle_set_state(struct cpuidle_state *idle_state)
  21 {
  22         idle_set_state(this_rq(), idle_state);
  23 }
  24
  25 static int __read_mostly cpu_idle_force_poll;
  26
  27 void cpu_idle_poll_ctrl(bool enable)
  28 {
  29         if (enable) {
  30                 cpu_idle_force_poll++;
  31         } else {
  32                 cpu_idle_force_poll--;
  33                 WARN_ON_ONCE(cpu_idle_force_poll < 0);
  34         }
  35 }
  36
  37 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
  38 static int __init cpu_idle_poll_setup(char *__unused)
  39 {
  40         cpu_idle_force_poll = 1;
  41
  42         return 1;
  43 }
  44 __setup("nohlt", cpu_idle_poll_setup);
  45
  46 static int __init cpu_idle_nopoll_setup(char *__unused)
  47 {
  48         cpu_idle_force_poll = 0;
  49
  50         return 1;
  51 }
  52 __setup("hlt", cpu_idle_nopoll_setup);
  53 #endif
  54
  55 static noinline int __cpuidle cpu_idle_poll(void)
  56 {
  57         trace_cpu_idle(0, smp_processor_id());
  58         stop_critical_timings();
  59         rcu_idle_enter();
  60         local_irq_enable();
  61
  62         while (!tif_need_resched() &&
  63                (cpu_idle_force_poll || tick_check_broadcast_expired()))
  64                 cpu_relax();
  65
  66         rcu_idle_exit();
  67         start_critical_timings();
  68         trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
  69
  70         return 1;
  71 }
  72
  73 /* Weak implementations for optional arch specific functions */
  74 void __weak arch_cpu_idle_prepare(void) { }
  75 void __weak arch_cpu_idle_enter(void) { }
  76 void __weak arch_cpu_idle_exit(void) { }
  77 void __weak arch_cpu_idle_dead(void) { }
  78 void __weak arch_cpu_idle(void)
  79 {
  80         cpu_idle_force_poll = 1;
  81         raw_local_irq_enable();
  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         if (current_clr_polling_and_test()) {
  92                 local_irq_enable();
  93         } else {
  94
  95                 trace_cpu_idle(1, smp_processor_id());
  96                 stop_critical_timings();
  97
  98                 /*
  99                  * arch_cpu_idle() is supposed to enable IRQs, however
 100                  * we can't do that because of RCU and tracing.
 101                  *
 102                  * Trace IRQs enable here, then switch off RCU, and have
 103                  * arch_cpu_idle() use raw_local_irq_enable(). Note that
 104                  * rcu_idle_enter() relies on lockdep IRQ state, so switch that
 105                  * last -- this is very similar to the entry code.
 106                  */
 107                 trace_hardirqs_on_prepare();
 108                 lockdep_hardirqs_on_prepare(_THIS_IP_);
 109                 rcu_idle_enter();
 110                 lockdep_hardirqs_on(_THIS_IP_);
 111
 112                 arch_cpu_idle();
 113
 114                 /*
 115                  * OK, so IRQs are enabled here, but RCU needs them disabled to
 116                  * turn itself back on.. funny thing is that disabling IRQs
 117                  * will cause tracing, which needs RCU. Jump through hoops to
 118                  * make it 'work'.
 119                  */
 120                 raw_local_irq_disable();
 121                 lockdep_hardirqs_off(_THIS_IP_);
 122                 rcu_idle_exit();
 123                 lockdep_hardirqs_on(_THIS_IP_);
 124                 raw_local_irq_enable();
 125
 126                 start_critical_timings();
 127                 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 128         }
 129 }
 130
 131 static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
 132                                struct cpuidle_device *dev)
 133 {
 134         if (current_clr_polling_and_test())
 135                 return -EBUSY;
 136
 137         return cpuidle_enter_s2idle(drv, dev);
 138 }
 139
 140 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 141                       int next_state)
 142 {
 143         /*
 144          * The idle task must be scheduled, it is pointless to go to idle, just
 145          * update no idle residency and return.
 146          */
 147         if (current_clr_polling_and_test()) {
 148                 dev->last_residency_ns = 0;
 149                 local_irq_enable();
 150                 return -EBUSY;
 151         }
 152
 153         /*
 154          * Enter the idle state previously returned by the governor decision.
 155          * This function will block until an interrupt occurs and will take
 156          * care of re-enabling the local interrupts
 157          */
 158         return cpuidle_enter(drv, dev, next_state);
 159 }
 160
 161 /**
 162  * cpuidle_idle_call - the main idle function
 163  *
 164  * NOTE: no locks or semaphores should be used here
 165  *
 166  * On archs that support TIF_POLLING_NRFLAG, is called with polling
 167  * set, and it returns with polling set.  If it ever stops polling, it
 168  * must clear the polling bit.
 169  */
 170 static void cpuidle_idle_call(void)
 171 {
 172         struct cpuidle_device *dev = cpuidle_get_device();
 173         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 174         int next_state, entered_state;
 175
 176         /*
 177          * Check if the idle task must be rescheduled. If it is the
 178          * case, exit the function after re-enabling the local irq.
 179          */
 180         if (need_resched()) {
 181                 local_irq_enable();
 182                 return;
 183         }
 184
 185         /*
 186          * The RCU framework needs to be told that we are entering an idle
 187          * section, so no more rcu read side critical sections and one more
 188          * step to the grace period
 189          */
 190
 191         if (cpuidle_not_available(drv, dev)) {
 192                 tick_nohz_idle_stop_tick();
 193
 194                 default_idle_call();
 195                 goto exit_idle;
 196         }
 197
 198         /*
 199          * Suspend-to-idle ("s2idle") is a system state in which all user space
 200          * has been frozen, all I/O devices have been suspended and the only
 201          * activity happens here and in interrupts (if any). In that case bypass
 202          * the cpuidle governor and go stratight for the deepest idle state
 203          * available.  Possibly also suspend the local tick and the entire
 204          * timekeeping to prevent timer interrupts from kicking us out of idle
 205          * until a proper wakeup interrupt happens.
 206          */
 207
 208         if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
 209                 u64 max_latency_ns;
 210
 211                 if (idle_should_enter_s2idle()) {
 212
 213                         entered_state = call_cpuidle_s2idle(drv, dev);
 214                         if (entered_state > 0)
 215                                 goto exit_idle;
 216
 217                         max_latency_ns = U64_MAX;
 218                 } else {
 219                         max_latency_ns = dev->forced_idle_latency_limit_ns;
 220                 }
 221
 222                 tick_nohz_idle_stop_tick();
 223
 224                 next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
 225                 call_cpuidle(drv, dev, next_state);
 226         } else {
 227                 bool stop_tick = true;
 228
 229                 /*
 230                  * Ask the cpuidle framework to choose a convenient idle state.
 231                  */
 232                 next_state = cpuidle_select(drv, dev, &stop_tick);
 233
 234                 if (stop_tick || tick_nohz_tick_stopped())
 235                         tick_nohz_idle_stop_tick();
 236                 else
 237                         tick_nohz_idle_retain_tick();
 238
 239                 entered_state = call_cpuidle(drv, dev, next_state);
 240                 /*
 241                  * Give the governor an opportunity to reflect on the outcome
 242                  */
 243                 cpuidle_reflect(dev, entered_state);
 244         }
 245
 246 exit_idle:
 247         __current_set_polling();
 248
 249         /*
 250          * It is up to the idle functions to reenable local interrupts
 251          */
 252         if (WARN_ON_ONCE(irqs_disabled()))
 253                 local_irq_enable();
 254 }
 255
 256 /*
 257  * Generic idle loop implementation
 258  *
 259  * Called with polling cleared.
 260  */
 261 static void do_idle(void)
 262 {
 263         int cpu = smp_processor_id();
 264         /*
 265          * If the arch has a polling bit, we maintain an invariant:
 266          *
 267          * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
 268          * rq->idle). This means that, if rq->idle has the polling bit set,
 269          * then setting need_resched is guaranteed to cause the CPU to
 270          * reschedule.
 271          */
 272
 273         __current_set_polling();
 274         tick_nohz_idle_enter();
 275
 276         while (!need_resched()) {
 277                 rmb();
 278
 279                 local_irq_disable();
 280
 281                 if (cpu_is_offline(cpu)) {
 282                         tick_nohz_idle_stop_tick();
 283                         cpuhp_report_idle_dead();
 284                         arch_cpu_idle_dead();
 285                 }
 286
 287                 arch_cpu_idle_enter();
 288                 rcu_nocb_flush_deferred_wakeup();
 289
 290                 /*
 291                  * In poll mode we reenable interrupts and spin. Also if we
 292                  * detected in the wakeup from idle path that the tick
 293                  * broadcast device expired for us, we don't want to go deep
 294                  * idle as we know that the IPI is going to arrive right away.
 295                  */
 296                 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
 297                         tick_nohz_idle_restart_tick();
 298                         cpu_idle_poll();
 299                 } else {
 300                         cpuidle_idle_call();
 301                 }
 302                 arch_cpu_idle_exit();
 303         }
 304
 305         /*
 306          * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
 307          * be set, propagate it into PREEMPT_NEED_RESCHED.
 308          *
 309          * This is required because for polling idle loops we will not have had
 310          * an IPI to fold the state for us.
 311          */
 312         preempt_set_need_resched();
 313         tick_nohz_idle_exit();
 314         __current_clr_polling();
 315
 316         /*
 317          * We promise to call sched_ttwu_pending() and reschedule if
 318          * need_resched() is set while polling is set. That means that clearing
 319          * polling needs to be visible before doing these things.
 320          */
 321         smp_mb__after_atomic();
 322
 323         /*
 324          * RCU relies on this call to be done outside of an RCU read-side
 325          * critical section.
 326          */
 327         flush_smp_call_function_from_idle();
 328         schedule_idle();
 329
 330         if (unlikely(klp_patch_pending(current)))
 331                 klp_update_patch_state(current);
 332 }
 333
 334 bool cpu_in_idle(unsigned long pc)
 335 {
 336         return pc >= (unsigned long)__cpuidle_text_start &&
 337                 pc < (unsigned long)__cpuidle_text_end;
 338 }
 339
 340 struct idle_timer {
 341         struct hrtimer timer;
 342         int done;
 343 };
 344
 345 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
 346 {
 347         struct idle_timer *it = container_of(timer, struct idle_timer, timer);
 348
 349         WRITE_ONCE(it->done, 1);
 350         set_tsk_need_resched(current);
 351
 352         return HRTIMER_NORESTART;
 353 }
 354
 355 void play_idle_precise(u64 duration_ns, u64 latency_ns)
 356 {
 357         struct idle_timer it;
 358
 359         /*
 360          * Only FIFO tasks can disable the tick since they don't need the forced
 361          * preemption.
 362          */
 363         WARN_ON_ONCE(current->policy != SCHED_FIFO);
 364         WARN_ON_ONCE(current->nr_cpus_allowed != 1);
 365         WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
 366         WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
 367         WARN_ON_ONCE(!duration_ns);
 368         WARN_ON_ONCE(current->mm);
 369
 370         rcu_sleep_check();
 371         preempt_disable();
 372         current->flags |= PF_IDLE;
 373         cpuidle_use_deepest_state(latency_ns);
 374
 375         it.done = 0;
 376         hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 377         it.timer.function = idle_inject_timer_fn;
 378         hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
 379                       HRTIMER_MODE_REL_PINNED);
 380
 381         while (!READ_ONCE(it.done))
 382                 do_idle();
 383
 384         cpuidle_use_deepest_state(0);
 385         current->flags &= ~PF_IDLE;
 386
 387         preempt_fold_need_resched();
 388         preempt_enable();
 389 }
 390 EXPORT_SYMBOL_GPL(play_idle_precise);
 391
 392 void cpu_startup_entry(enum cpuhp_state state)
 393 {
 394         arch_cpu_idle_prepare();
 395         cpuhp_online_idle(state);
 396         while (1)
 397                 do_idle();
 398 }
 399
 400 /*
 401  * idle-task scheduling class.
 402  */
 403
 404 #ifdef CONFIG_SMP
 405 static int
 406 select_task_rq_idle(struct task_struct *p, int cpu, int flags)
 407 {
 408         return task_cpu(p); /* IDLE tasks as never migrated */
 409 }
 410
 411 static int
 412 balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 413 {
 414         return WARN_ON_ONCE(1);
 415 }
 416 #endif
 417
 418 /*
 419  * Idle tasks are unconditionally rescheduled:
 420  */
 421 static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
 422 {
 423         resched_curr(rq);
 424 }
 425
 426 static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
 427 {
 428 }
 429
 430 static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
 431 {
 432         update_idle_core(rq);
 433         schedstat_inc(rq->sched_goidle);
 434 }
 435
 436 struct task_struct *pick_next_task_idle(struct rq *rq)
 437 {
 438         struct task_struct *next = rq->idle;
 439
 440         set_next_task_idle(rq, next, true);
 441
 442         return next;
 443 }
 444
 445 /*
 446  * It is not legal to sleep in the idle task - print a warning
 447  * message if some code attempts to do it:
 448  */
 449 static void
 450 dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
 451 {
 452         raw_spin_unlock_irq(&rq->lock);
 453         printk(KERN_ERR "bad: scheduling from the idle thread!\n");
 454         dump_stack();
 455         raw_spin_lock_irq(&rq->lock);
 456 }
 457
 458 /*
 459  * scheduler tick hitting a task of our scheduling class.
 460  *
 461  * NOTE: This function can be called remotely by the tick offload that
 462  * goes along full dynticks. Therefore no local assumption can be made
 463  * and everything must be accessed through the @rq and @curr passed in
 464  * parameters.
 465  */
 466 static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
 467 {
 468 }
 469
 470 static void switched_to_idle(struct rq *rq, struct task_struct *p)
 471 {
 472         BUG();
 473 }
 474
 475 static void
 476 prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
 477 {
 478         BUG();
 479 }
 480
 481 static void update_curr_idle(struct rq *rq)
 482 {
 483 }
 484
 485 /*
 486  * Simple, special scheduling class for the per-CPU idle tasks:
 487  */
 488 DEFINE_SCHED_CLASS(idle) = {
 489
 490         /* no enqueue/yield_task for idle tasks */
 491
 492         /* dequeue is not valid, we print a debug message there: */
 493         .dequeue_task           = dequeue_task_idle,
 494
 495         .check_preempt_curr     = check_preempt_curr_idle,
 496
 497         .pick_next_task         = pick_next_task_idle,
 498         .put_prev_task          = put_prev_task_idle,
 499         .set_next_task          = set_next_task_idle,
 500
 501 #ifdef CONFIG_SMP
 502         .balance                = balance_idle,
 503         .select_task_rq         = select_task_rq_idle,
 504         .set_cpus_allowed       = set_cpus_allowed_common,
 505 #endif
 506
 507         .task_tick              = task_tick_idle,
 508
 509         .prio_changed           = prio_changed_idle,
 510         .switched_to            = switched_to_idle,
 511         .update_curr            = update_curr_idle,
 512 };