kernel/sched/idle.c

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