1 // SPDX-License-Identifier: GPL-2.0-only
3 * Generic entry points for the idle threads and
4 * implementation of the idle task scheduling class.
6 * (NOTE: these are not related to SCHED_IDLE batch scheduled
7 * tasks which are handled in sched/fair.c )
11 #include <trace/events/power.h>
13 /* Linker adds these: start and end of __cpuidle functions */
14 extern char __cpuidle_text_start
[], __cpuidle_text_end
[];
17 * sched_idle_set_state - Record idle state for the current CPU.
18 * @idle_state: State to record.
20 void sched_idle_set_state(struct cpuidle_state
*idle_state
)
22 idle_set_state(this_rq(), idle_state
);
25 static int __read_mostly cpu_idle_force_poll
;
27 void cpu_idle_poll_ctrl(bool enable
)
30 cpu_idle_force_poll
++;
32 cpu_idle_force_poll
--;
33 WARN_ON_ONCE(cpu_idle_force_poll
< 0);
37 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
38 static int __init
cpu_idle_poll_setup(char *__unused
)
40 cpu_idle_force_poll
= 1;
44 __setup("nohlt", cpu_idle_poll_setup
);
46 static int __init
cpu_idle_nopoll_setup(char *__unused
)
48 cpu_idle_force_poll
= 0;
52 __setup("hlt", cpu_idle_nopoll_setup
);
55 static noinline
int __cpuidle
cpu_idle_poll(void)
57 trace_cpu_idle(0, smp_processor_id());
58 stop_critical_timings();
62 while (!tif_need_resched() &&
63 (cpu_idle_force_poll
|| tick_check_broadcast_expired()))
67 start_critical_timings();
68 trace_cpu_idle(PWR_EVENT_EXIT
, smp_processor_id());
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)
80 cpu_idle_force_poll
= 1;
85 * default_idle_call - Default CPU idle routine.
87 * To use when the cpuidle framework cannot be used.
89 void __cpuidle
default_idle_call(void)
91 if (current_clr_polling_and_test()) {
95 trace_cpu_idle(1, smp_processor_id());
96 stop_critical_timings();
100 start_critical_timings();
101 trace_cpu_idle(PWR_EVENT_EXIT
, smp_processor_id());
105 static int call_cpuidle_s2idle(struct cpuidle_driver
*drv
,
106 struct cpuidle_device
*dev
)
108 if (current_clr_polling_and_test())
111 return cpuidle_enter_s2idle(drv
, dev
);
114 static int call_cpuidle(struct cpuidle_driver
*drv
, struct cpuidle_device
*dev
,
118 * The idle task must be scheduled, it is pointless to go to idle, just
119 * update no idle residency and return.
121 if (current_clr_polling_and_test()) {
122 dev
->last_residency_ns
= 0;
128 * Enter the idle state previously returned by the governor decision.
129 * This function will block until an interrupt occurs and will take
130 * care of re-enabling the local interrupts
132 return cpuidle_enter(drv
, dev
, next_state
);
136 * cpuidle_idle_call - the main idle function
138 * NOTE: no locks or semaphores should be used here
140 * On archs that support TIF_POLLING_NRFLAG, is called with polling
141 * set, and it returns with polling set. If it ever stops polling, it
142 * must clear the polling bit.
144 static void cpuidle_idle_call(void)
146 struct cpuidle_device
*dev
= cpuidle_get_device();
147 struct cpuidle_driver
*drv
= cpuidle_get_cpu_driver(dev
);
148 int next_state
, entered_state
;
151 * Check if the idle task must be rescheduled. If it is the
152 * case, exit the function after re-enabling the local irq.
154 if (need_resched()) {
160 * The RCU framework needs to be told that we are entering an idle
161 * section, so no more rcu read side critical sections and one more
162 * step to the grace period
165 if (cpuidle_not_available(drv
, dev
)) {
166 tick_nohz_idle_stop_tick();
173 * Suspend-to-idle ("s2idle") is a system state in which all user space
174 * has been frozen, all I/O devices have been suspended and the only
175 * activity happens here and in interrupts (if any). In that case bypass
176 * the cpuidle governor and go stratight for the deepest idle state
177 * available. Possibly also suspend the local tick and the entire
178 * timekeeping to prevent timer interrupts from kicking us out of idle
179 * until a proper wakeup interrupt happens.
182 if (idle_should_enter_s2idle() || dev
->forced_idle_latency_limit_ns
) {
185 if (idle_should_enter_s2idle()) {
187 entered_state
= call_cpuidle_s2idle(drv
, dev
);
188 if (entered_state
> 0)
191 max_latency_ns
= U64_MAX
;
193 max_latency_ns
= dev
->forced_idle_latency_limit_ns
;
196 tick_nohz_idle_stop_tick();
198 next_state
= cpuidle_find_deepest_state(drv
, dev
, max_latency_ns
);
199 call_cpuidle(drv
, dev
, next_state
);
201 bool stop_tick
= true;
204 * Ask the cpuidle framework to choose a convenient idle state.
206 next_state
= cpuidle_select(drv
, dev
, &stop_tick
);
208 if (stop_tick
|| tick_nohz_tick_stopped())
209 tick_nohz_idle_stop_tick();
211 tick_nohz_idle_retain_tick();
213 entered_state
= call_cpuidle(drv
, dev
, next_state
);
215 * Give the governor an opportunity to reflect on the outcome
217 cpuidle_reflect(dev
, entered_state
);
221 __current_set_polling();
224 * It is up to the idle functions to reenable local interrupts
226 if (WARN_ON_ONCE(irqs_disabled()))
231 * Generic idle loop implementation
233 * Called with polling cleared.
235 static void do_idle(void)
237 int cpu
= smp_processor_id();
239 * If the arch has a polling bit, we maintain an invariant:
241 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
242 * rq->idle). This means that, if rq->idle has the polling bit set,
243 * then setting need_resched is guaranteed to cause the CPU to
247 __current_set_polling();
248 tick_nohz_idle_enter();
250 while (!need_resched()) {
255 if (cpu_is_offline(cpu
)) {
256 tick_nohz_idle_stop_tick();
257 cpuhp_report_idle_dead();
258 arch_cpu_idle_dead();
261 arch_cpu_idle_enter();
264 * In poll mode we reenable interrupts and spin. Also if we
265 * detected in the wakeup from idle path that the tick
266 * broadcast device expired for us, we don't want to go deep
267 * idle as we know that the IPI is going to arrive right away.
269 if (cpu_idle_force_poll
|| tick_check_broadcast_expired()) {
270 tick_nohz_idle_restart_tick();
275 arch_cpu_idle_exit();
279 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
280 * be set, propagate it into PREEMPT_NEED_RESCHED.
282 * This is required because for polling idle loops we will not have had
283 * an IPI to fold the state for us.
285 preempt_set_need_resched();
286 tick_nohz_idle_exit();
287 __current_clr_polling();
290 * We promise to call sched_ttwu_pending() and reschedule if
291 * need_resched() is set while polling is set. That means that clearing
292 * polling needs to be visible before doing these things.
294 smp_mb__after_atomic();
297 * RCU relies on this call to be done outside of an RCU read-side
300 flush_smp_call_function_from_idle();
303 if (unlikely(klp_patch_pending(current
)))
304 klp_update_patch_state(current
);
307 bool cpu_in_idle(unsigned long pc
)
309 return pc
>= (unsigned long)__cpuidle_text_start
&&
310 pc
< (unsigned long)__cpuidle_text_end
;
314 struct hrtimer timer
;
318 static enum hrtimer_restart
idle_inject_timer_fn(struct hrtimer
*timer
)
320 struct idle_timer
*it
= container_of(timer
, struct idle_timer
, timer
);
322 WRITE_ONCE(it
->done
, 1);
323 set_tsk_need_resched(current
);
325 return HRTIMER_NORESTART
;
328 void play_idle_precise(u64 duration_ns
, u64 latency_ns
)
330 struct idle_timer it
;
333 * Only FIFO tasks can disable the tick since they don't need the forced
336 WARN_ON_ONCE(current
->policy
!= SCHED_FIFO
);
337 WARN_ON_ONCE(current
->nr_cpus_allowed
!= 1);
338 WARN_ON_ONCE(!(current
->flags
& PF_KTHREAD
));
339 WARN_ON_ONCE(!(current
->flags
& PF_NO_SETAFFINITY
));
340 WARN_ON_ONCE(!duration_ns
);
344 current
->flags
|= PF_IDLE
;
345 cpuidle_use_deepest_state(latency_ns
);
348 hrtimer_init_on_stack(&it
.timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
349 it
.timer
.function
= idle_inject_timer_fn
;
350 hrtimer_start(&it
.timer
, ns_to_ktime(duration_ns
),
351 HRTIMER_MODE_REL_PINNED
);
353 while (!READ_ONCE(it
.done
))
356 cpuidle_use_deepest_state(0);
357 current
->flags
&= ~PF_IDLE
;
359 preempt_fold_need_resched();
362 EXPORT_SYMBOL_GPL(play_idle_precise
);
364 void cpu_startup_entry(enum cpuhp_state state
)
366 arch_cpu_idle_prepare();
367 cpuhp_online_idle(state
);
373 * idle-task scheduling class.
378 select_task_rq_idle(struct task_struct
*p
, int cpu
, int sd_flag
, int flags
)
380 return task_cpu(p
); /* IDLE tasks as never migrated */
384 balance_idle(struct rq
*rq
, struct task_struct
*prev
, struct rq_flags
*rf
)
386 return WARN_ON_ONCE(1);
391 * Idle tasks are unconditionally rescheduled:
393 static void check_preempt_curr_idle(struct rq
*rq
, struct task_struct
*p
, int flags
)
398 static void put_prev_task_idle(struct rq
*rq
, struct task_struct
*prev
)
402 static void set_next_task_idle(struct rq
*rq
, struct task_struct
*next
, bool first
)
404 update_idle_core(rq
);
405 schedstat_inc(rq
->sched_goidle
);
408 struct task_struct
*pick_next_task_idle(struct rq
*rq
)
410 struct task_struct
*next
= rq
->idle
;
412 set_next_task_idle(rq
, next
, true);
418 * It is not legal to sleep in the idle task - print a warning
419 * message if some code attempts to do it:
422 dequeue_task_idle(struct rq
*rq
, struct task_struct
*p
, int flags
)
424 raw_spin_unlock_irq(&rq
->lock
);
425 printk(KERN_ERR
"bad: scheduling from the idle thread!\n");
427 raw_spin_lock_irq(&rq
->lock
);
431 * scheduler tick hitting a task of our scheduling class.
433 * NOTE: This function can be called remotely by the tick offload that
434 * goes along full dynticks. Therefore no local assumption can be made
435 * and everything must be accessed through the @rq and @curr passed in
438 static void task_tick_idle(struct rq
*rq
, struct task_struct
*curr
, int queued
)
442 static void switched_to_idle(struct rq
*rq
, struct task_struct
*p
)
448 prio_changed_idle(struct rq
*rq
, struct task_struct
*p
, int oldprio
)
453 static void update_curr_idle(struct rq
*rq
)
458 * Simple, special scheduling class for the per-CPU idle tasks:
460 const struct sched_class idle_sched_class
461 __section("__idle_sched_class") = {
462 /* no enqueue/yield_task for idle tasks */
464 /* dequeue is not valid, we print a debug message there: */
465 .dequeue_task
= dequeue_task_idle
,
467 .check_preempt_curr
= check_preempt_curr_idle
,
469 .pick_next_task
= pick_next_task_idle
,
470 .put_prev_task
= put_prev_task_idle
,
471 .set_next_task
= set_next_task_idle
,
474 .balance
= balance_idle
,
475 .select_task_rq
= select_task_rq_idle
,
476 .set_cpus_allowed
= set_cpus_allowed_common
,
479 .task_tick
= task_tick_idle
,
481 .prio_changed
= prio_changed_idle
,
482 .switched_to
= switched_to_idle
,
483 .update_curr
= update_curr_idle
,