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 <linux/cpuidle.h>
10#include <linux/suspend.h>
11#include <linux/livepatch.h>
12#include "sched.h"
13#include "smp.h"
14
15/* Linker adds these: start and end of __cpuidle functions */
16extern char __cpuidle_text_start[], __cpuidle_text_end[];
17
18/**
19 * sched_idle_set_state - Record idle state for the current CPU.
20 * @idle_state: State to record.
21 */
22void sched_idle_set_state(struct cpuidle_state *idle_state)
23{
24 idle_set_state(this_rq(), idle_state);
25}
26
27static int __read_mostly cpu_idle_force_poll;
28
29void cpu_idle_poll_ctrl(bool enable)
30{
31 if (enable) {
32 cpu_idle_force_poll++;
33 } else {
34 cpu_idle_force_poll--;
35 WARN_ON_ONCE(cpu_idle_force_poll < 0);
36 }
37}
38
39#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
40static int __init cpu_idle_poll_setup(char *__unused)
41{
42 cpu_idle_force_poll = 1;
43
44 return 1;
45}
46__setup("nohlt", cpu_idle_poll_setup);
47
48static int __init cpu_idle_nopoll_setup(char *__unused)
49{
50 cpu_idle_force_poll = 0;
51
52 return 1;
53}
54__setup("hlt", cpu_idle_nopoll_setup);
55#endif /* CONFIG_GENERIC_IDLE_POLL_SETUP */
56
57static noinline int __cpuidle cpu_idle_poll(void)
58{
59 instrumentation_begin();
60 trace_cpu_idle(state: 0, smp_processor_id());
61 stop_critical_timings();
62 ct_cpuidle_enter();
63
64 raw_local_irq_enable();
65 while (!tif_need_resched() &&
66 (cpu_idle_force_poll || tick_check_broadcast_expired()))
67 cpu_relax();
68 raw_local_irq_disable();
69
70 ct_cpuidle_exit();
71 start_critical_timings();
72 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
73 local_irq_enable();
74 instrumentation_end();
75
76 return 1;
77}
78
79/* Weak implementations for optional arch specific functions */
80void __weak arch_cpu_idle_prepare(void) { }
81void __weak arch_cpu_idle_enter(void) { }
82void __weak arch_cpu_idle_exit(void) { }
83void __weak __noreturn arch_cpu_idle_dead(void) { while (1); }
84void __weak arch_cpu_idle(void)
85{
86 cpu_idle_force_poll = 1;
87}
88
89#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE
90DEFINE_STATIC_KEY_FALSE(arch_needs_tick_broadcast);
91
92static inline void cond_tick_broadcast_enter(void)
93{
94 if (static_branch_unlikely(&arch_needs_tick_broadcast))
95 tick_broadcast_enter();
96}
97
98static inline void cond_tick_broadcast_exit(void)
99{
100 if (static_branch_unlikely(&arch_needs_tick_broadcast))
101 tick_broadcast_exit();
102}
103#else /* !CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE: */
104static inline void cond_tick_broadcast_enter(void) { }
105static inline void cond_tick_broadcast_exit(void) { }
106#endif /* !CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE */
107
108/**
109 * default_idle_call - Default CPU idle routine.
110 *
111 * To use when the cpuidle framework cannot be used.
112 */
113void __cpuidle default_idle_call(void)
114{
115 instrumentation_begin();
116 if (!current_clr_polling_and_test()) {
117 cond_tick_broadcast_enter();
118 trace_cpu_idle(state: 1, smp_processor_id());
119 stop_critical_timings();
120
121 ct_cpuidle_enter();
122 arch_cpu_idle();
123 ct_cpuidle_exit();
124
125 start_critical_timings();
126 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
127 cond_tick_broadcast_exit();
128 }
129 local_irq_enable();
130 instrumentation_end();
131}
132
133static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
134 struct cpuidle_device *dev)
135{
136 if (current_clr_polling_and_test())
137 return -EBUSY;
138
139 return cpuidle_enter_s2idle(drv, dev);
140}
141
142static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
143 int next_state)
144{
145 /*
146 * The idle task must be scheduled, it is pointless to go to idle, just
147 * update no idle residency and return.
148 */
149 if (current_clr_polling_and_test()) {
150 dev->last_residency_ns = 0;
151 local_irq_enable();
152 return -EBUSY;
153 }
154
155 /*
156 * Enter the idle state previously returned by the governor decision.
157 * This function will block until an interrupt occurs and will take
158 * care of re-enabling the local interrupts
159 */
160 return cpuidle_enter(drv, dev, index: next_state);
161}
162
163/**
164 * cpuidle_idle_call - the main idle function
165 *
166 * NOTE: no locks or semaphores should be used here
167 *
168 * On architectures that support TIF_POLLING_NRFLAG, is called with polling
169 * set, and it returns with polling set. If it ever stops polling, it
170 * must clear the polling bit.
171 */
172static void cpuidle_idle_call(void)
173{
174 struct cpuidle_device *dev = cpuidle_get_device();
175 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
176 int next_state, entered_state;
177
178 /*
179 * Check if the idle task must be rescheduled. If it is the
180 * case, exit the function after re-enabling the local IRQ.
181 */
182 if (need_resched()) {
183 local_irq_enable();
184 return;
185 }
186
187 if (cpuidle_not_available(drv, dev)) {
188 tick_nohz_idle_stop_tick();
189
190 default_idle_call();
191 goto exit_idle;
192 }
193
194 /*
195 * Suspend-to-idle ("s2idle") is a system state in which all user space
196 * has been frozen, all I/O devices have been suspended and the only
197 * activity happens here and in interrupts (if any). In that case bypass
198 * the cpuidle governor and go straight for the deepest idle state
199 * available. Possibly also suspend the local tick and the entire
200 * timekeeping to prevent timer interrupts from kicking us out of idle
201 * until a proper wakeup interrupt happens.
202 */
203
204 if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
205 u64 max_latency_ns;
206
207 if (idle_should_enter_s2idle()) {
208
209 entered_state = call_cpuidle_s2idle(drv, dev);
210 if (entered_state > 0)
211 goto exit_idle;
212
213 max_latency_ns = U64_MAX;
214 } else {
215 max_latency_ns = dev->forced_idle_latency_limit_ns;
216 }
217
218 tick_nohz_idle_stop_tick();
219
220 next_state = cpuidle_find_deepest_state(drv, dev, latency_limit_ns: max_latency_ns);
221 call_cpuidle(drv, dev, next_state);
222 } else {
223 bool stop_tick = true;
224
225 /*
226 * Ask the cpuidle framework to choose a convenient idle state.
227 */
228 next_state = cpuidle_select(drv, dev, stop_tick: &stop_tick);
229
230 if (stop_tick || tick_nohz_tick_stopped())
231 tick_nohz_idle_stop_tick();
232 else
233 tick_nohz_idle_retain_tick();
234
235 entered_state = call_cpuidle(drv, dev, next_state);
236 /*
237 * Give the governor an opportunity to reflect on the outcome
238 */
239 cpuidle_reflect(dev, index: entered_state);
240 }
241
242exit_idle:
243 __current_set_polling();
244
245 /*
246 * It is up to the idle functions to re-enable local interrupts
247 */
248 if (WARN_ON_ONCE(irqs_disabled()))
249 local_irq_enable();
250}
251
252/*
253 * Generic idle loop implementation
254 *
255 * Called with polling cleared.
256 */
257static void do_idle(void)
258{
259 int cpu = smp_processor_id();
260
261 /*
262 * Check if we need to update blocked load
263 */
264 nohz_run_idle_balance(cpu);
265
266 /*
267 * If the arch has a polling bit, we maintain an invariant:
268 *
269 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
270 * rq->idle). This means that, if rq->idle has the polling bit set,
271 * then setting need_resched is guaranteed to cause the CPU to
272 * reschedule.
273 */
274
275 __current_set_polling();
276 tick_nohz_idle_enter();
277
278 while (!need_resched()) {
279
280 /*
281 * Interrupts shouldn't be re-enabled from that point on until
282 * the CPU sleeping instruction is reached. Otherwise an interrupt
283 * may fire and queue a timer that would be ignored until the CPU
284 * wakes from the sleeping instruction. And testing need_resched()
285 * doesn't tell about pending needed timer reprogram.
286 *
287 * Several cases to consider:
288 *
289 * - SLEEP-UNTIL-PENDING-INTERRUPT based instructions such as
290 * "wfi" or "mwait" are fine because they can be entered with
291 * interrupt disabled.
292 *
293 * - sti;mwait() couple is fine because the interrupts are
294 * re-enabled only upon the execution of mwait, leaving no gap
295 * in-between.
296 *
297 * - ROLLBACK based idle handlers with the sleeping instruction
298 * called with interrupts enabled are NOT fine. In this scheme
299 * when the interrupt detects it has interrupted an idle handler,
300 * it rolls back to its beginning which performs the
301 * need_resched() check before re-executing the sleeping
302 * instruction. This can leak a pending needed timer reprogram.
303 * If such a scheme is really mandatory due to the lack of an
304 * appropriate CPU sleeping instruction, then a FAST-FORWARD
305 * must instead be applied: when the interrupt detects it has
306 * interrupted an idle handler, it must resume to the end of
307 * this idle handler so that the generic idle loop is iterated
308 * again to reprogram the tick.
309 */
310 local_irq_disable();
311
312 if (cpu_is_offline(cpu)) {
313 cpuhp_report_idle_dead();
314 arch_cpu_idle_dead();
315 }
316
317 arch_cpu_idle_enter();
318 rcu_nocb_flush_deferred_wakeup();
319
320 /*
321 * In poll mode we re-enable interrupts and spin. Also if we
322 * detected in the wakeup from idle path that the tick
323 * broadcast device expired for us, we don't want to go deep
324 * idle as we know that the IPI is going to arrive right away.
325 */
326 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
327 tick_nohz_idle_restart_tick();
328 cpu_idle_poll();
329 } else {
330 cpuidle_idle_call();
331 }
332 arch_cpu_idle_exit();
333 }
334
335 /*
336 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
337 * be set, propagate it into PREEMPT_NEED_RESCHED.
338 *
339 * This is required because for polling idle loops we will not have had
340 * an IPI to fold the state for us.
341 */
342 preempt_set_need_resched();
343 tick_nohz_idle_exit();
344 __current_clr_polling();
345
346 /*
347 * We promise to call sched_ttwu_pending() and reschedule if
348 * need_resched() is set while polling is set. That means that clearing
349 * polling needs to be visible before doing these things.
350 */
351 smp_mb__after_atomic();
352
353 /*
354 * RCU relies on this call to be done outside of an RCU read-side
355 * critical section.
356 */
357 flush_smp_call_function_queue();
358 schedule_idle();
359
360 if (unlikely(klp_patch_pending(current)))
361 klp_update_patch_state(current);
362}
363
364bool cpu_in_idle(unsigned long pc)
365{
366 return pc >= (unsigned long)__cpuidle_text_start &&
367 pc < (unsigned long)__cpuidle_text_end;
368}
369
370struct idle_timer {
371 struct hrtimer timer;
372 int done;
373};
374
375static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
376{
377 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
378
379 WRITE_ONCE(it->done, 1);
380 set_tsk_need_resched(current);
381
382 return HRTIMER_NORESTART;
383}
384
385void play_idle_precise(u64 duration_ns, u64 latency_ns)
386{
387 struct idle_timer it;
388
389 /*
390 * Only FIFO tasks can disable the tick since they don't need the forced
391 * preemption.
392 */
393 WARN_ON_ONCE(current->policy != SCHED_FIFO);
394 WARN_ON_ONCE(current->nr_cpus_allowed != 1);
395 WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
396 WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
397 WARN_ON_ONCE(!duration_ns);
398 WARN_ON_ONCE(current->mm);
399
400 rcu_sleep_check();
401 preempt_disable();
402 current->flags |= PF_IDLE;
403 cpuidle_use_deepest_state(latency_limit_ns: latency_ns);
404
405 it.done = 0;
406 hrtimer_setup_on_stack(timer: &it.timer, function: idle_inject_timer_fn, CLOCK_MONOTONIC,
407 mode: HRTIMER_MODE_REL_HARD);
408 hrtimer_start(timer: &it.timer, tim: ns_to_ktime(ns: duration_ns),
409 mode: HRTIMER_MODE_REL_PINNED_HARD);
410
411 while (!READ_ONCE(it.done))
412 do_idle();
413
414 cpuidle_use_deepest_state(latency_limit_ns: 0);
415 current->flags &= ~PF_IDLE;
416
417 preempt_fold_need_resched();
418 preempt_enable();
419}
420EXPORT_SYMBOL_GPL(play_idle_precise);
421
422void cpu_startup_entry(enum cpuhp_state state)
423{
424 current->flags |= PF_IDLE;
425 arch_cpu_idle_prepare();
426 cpuhp_online_idle(state);
427 while (1)
428 do_idle();
429}
430
431/*
432 * idle-task scheduling class.
433 */
434
435static int
436select_task_rq_idle(struct task_struct *p, int cpu, int flags)
437{
438 return task_cpu(p); /* IDLE tasks as never migrated */
439}
440
441static int
442balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
443{
444 return WARN_ON_ONCE(1);
445}
446
447/*
448 * Idle tasks are unconditionally rescheduled:
449 */
450static void wakeup_preempt_idle(struct rq *rq, struct task_struct *p, int flags)
451{
452 resched_curr(rq);
453}
454
455static void put_prev_task_idle(struct rq *rq, struct task_struct *prev, struct task_struct *next)
456{
457 dl_server_update_idle_time(rq, p: prev);
458 scx_update_idle(rq, idle: false, do_notify: true);
459}
460
461static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
462{
463 update_idle_core(rq);
464 scx_update_idle(rq, idle: true, do_notify: true);
465 schedstat_inc(rq->sched_goidle);
466 next->se.exec_start = rq_clock_task(rq);
467}
468
469struct task_struct *pick_task_idle(struct rq *rq)
470{
471 scx_update_idle(rq, idle: true, do_notify: false);
472 return rq->idle;
473}
474
475/*
476 * It is not legal to sleep in the idle task - print a warning
477 * message if some code attempts to do it:
478 */
479static bool
480dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
481{
482 raw_spin_rq_unlock_irq(rq);
483 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
484 dump_stack();
485 raw_spin_rq_lock_irq(rq);
486 return true;
487}
488
489/*
490 * scheduler tick hitting a task of our scheduling class.
491 *
492 * NOTE: This function can be called remotely by the tick offload that
493 * goes along full dynticks. Therefore no local assumption can be made
494 * and everything must be accessed through the @rq and @curr passed in
495 * parameters.
496 */
497static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
498{
499}
500
501static void switched_to_idle(struct rq *rq, struct task_struct *p)
502{
503 BUG();
504}
505
506static void
507prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
508{
509 BUG();
510}
511
512static void update_curr_idle(struct rq *rq)
513{
514}
515
516/*
517 * Simple, special scheduling class for the per-CPU idle tasks:
518 */
519DEFINE_SCHED_CLASS(idle) = {
520
521 /* no enqueue/yield_task for idle tasks */
522
523 /* dequeue is not valid, we print a debug message there: */
524 .dequeue_task = dequeue_task_idle,
525
526 .wakeup_preempt = wakeup_preempt_idle,
527
528 .pick_task = pick_task_idle,
529 .put_prev_task = put_prev_task_idle,
530 .set_next_task = set_next_task_idle,
531
532 .balance = balance_idle,
533 .select_task_rq = select_task_rq_idle,
534 .set_cpus_allowed = set_cpus_allowed_common,
535
536 .task_tick = task_tick_idle,
537
538 .prio_changed = prio_changed_idle,
539 .switched_to = switched_to_idle,
540 .update_curr = update_curr_idle,
541};
542