Loading...
1/*
2 * Generic entry point for the idle threads
3 */
4#include <linux/sched.h>
5#include <linux/cpu.h>
6#include <linux/cpuidle.h>
7#include <linux/tick.h>
8#include <linux/mm.h>
9#include <linux/stackprotector.h>
10
11#include <asm/tlb.h>
12
13#include <trace/events/power.h>
14
15static int __read_mostly cpu_idle_force_poll;
16
17void cpu_idle_poll_ctrl(bool enable)
18{
19 if (enable) {
20 cpu_idle_force_poll++;
21 } else {
22 cpu_idle_force_poll--;
23 WARN_ON_ONCE(cpu_idle_force_poll < 0);
24 }
25}
26
27#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
28static int __init cpu_idle_poll_setup(char *__unused)
29{
30 cpu_idle_force_poll = 1;
31 return 1;
32}
33__setup("nohlt", cpu_idle_poll_setup);
34
35static int __init cpu_idle_nopoll_setup(char *__unused)
36{
37 cpu_idle_force_poll = 0;
38 return 1;
39}
40__setup("hlt", cpu_idle_nopoll_setup);
41#endif
42
43static inline int cpu_idle_poll(void)
44{
45 rcu_idle_enter();
46 trace_cpu_idle_rcuidle(0, smp_processor_id());
47 local_irq_enable();
48 while (!tif_need_resched())
49 cpu_relax();
50 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
51 rcu_idle_exit();
52 return 1;
53}
54
55/* Weak implementations for optional arch specific functions */
56void __weak arch_cpu_idle_prepare(void) { }
57void __weak arch_cpu_idle_enter(void) { }
58void __weak arch_cpu_idle_exit(void) { }
59void __weak arch_cpu_idle_dead(void) { }
60void __weak arch_cpu_idle(void)
61{
62 cpu_idle_force_poll = 1;
63 local_irq_enable();
64}
65
66/**
67 * cpuidle_idle_call - the main idle function
68 *
69 * NOTE: no locks or semaphores should be used here
70 * return non-zero on failure
71 */
72static int cpuidle_idle_call(void)
73{
74 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
75 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
76 int next_state, entered_state, ret;
77 bool broadcast;
78
79 /*
80 * Check if the idle task must be rescheduled. If it is the
81 * case, exit the function after re-enabling the local irq and
82 * set again the polling flag
83 */
84 if (current_clr_polling_and_test()) {
85 local_irq_enable();
86 __current_set_polling();
87 return 0;
88 }
89
90 /*
91 * During the idle period, stop measuring the disabled irqs
92 * critical sections latencies
93 */
94 stop_critical_timings();
95
96 /*
97 * Tell the RCU framework we are entering an idle section,
98 * so no more rcu read side critical sections and one more
99 * step to the grace period
100 */
101 rcu_idle_enter();
102
103 /*
104 * Check if the cpuidle framework is ready, otherwise fallback
105 * to the default arch specific idle method
106 */
107 ret = cpuidle_enabled(drv, dev);
108
109 if (!ret) {
110 /*
111 * Ask the governor to choose an idle state it thinks
112 * it is convenient to go to. There is *always* a
113 * convenient idle state
114 */
115 next_state = cpuidle_select(drv, dev);
116
117 /*
118 * The idle task must be scheduled, it is pointless to
119 * go to idle, just update no idle residency and get
120 * out of this function
121 */
122 if (current_clr_polling_and_test()) {
123 dev->last_residency = 0;
124 entered_state = next_state;
125 local_irq_enable();
126 } else {
127 broadcast = !!(drv->states[next_state].flags &
128 CPUIDLE_FLAG_TIMER_STOP);
129
130 if (broadcast)
131 /*
132 * Tell the time framework to switch
133 * to a broadcast timer because our
134 * local timer will be shutdown. If a
135 * local timer is used from another
136 * cpu as a broadcast timer, this call
137 * may fail if it is not available
138 */
139 ret = clockevents_notify(
140 CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
141 &dev->cpu);
142
143 if (!ret) {
144 trace_cpu_idle_rcuidle(next_state, dev->cpu);
145
146 /*
147 * Enter the idle state previously
148 * returned by the governor
149 * decision. This function will block
150 * until an interrupt occurs and will
151 * take care of re-enabling the local
152 * interrupts
153 */
154 entered_state = cpuidle_enter(drv, dev,
155 next_state);
156
157 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT,
158 dev->cpu);
159
160 if (broadcast)
161 clockevents_notify(
162 CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
163 &dev->cpu);
164
165 /*
166 * Give the governor an opportunity to reflect on the
167 * outcome
168 */
169 cpuidle_reflect(dev, entered_state);
170 }
171 }
172 }
173
174 /*
175 * We can't use the cpuidle framework, let's use the default
176 * idle routine
177 */
178 if (ret)
179 arch_cpu_idle();
180
181 __current_set_polling();
182
183 /*
184 * It is up to the idle functions to enable back the local
185 * interrupt
186 */
187 if (WARN_ON_ONCE(irqs_disabled()))
188 local_irq_enable();
189
190 rcu_idle_exit();
191 start_critical_timings();
192
193 return 0;
194}
195
196/*
197 * Generic idle loop implementation
198 */
199static void cpu_idle_loop(void)
200{
201 while (1) {
202 tick_nohz_idle_enter();
203
204 while (!need_resched()) {
205 check_pgt_cache();
206 rmb();
207
208 if (cpu_is_offline(smp_processor_id()))
209 arch_cpu_idle_dead();
210
211 local_irq_disable();
212 arch_cpu_idle_enter();
213
214 /*
215 * In poll mode we reenable interrupts and spin.
216 *
217 * Also if we detected in the wakeup from idle
218 * path that the tick broadcast device expired
219 * for us, we don't want to go deep idle as we
220 * know that the IPI is going to arrive right
221 * away
222 */
223 if (cpu_idle_force_poll || tick_check_broadcast_expired())
224 cpu_idle_poll();
225 else
226 cpuidle_idle_call();
227
228 arch_cpu_idle_exit();
229 }
230
231 /*
232 * Since we fell out of the loop above, we know
233 * TIF_NEED_RESCHED must be set, propagate it into
234 * PREEMPT_NEED_RESCHED.
235 *
236 * This is required because for polling idle loops we will
237 * not have had an IPI to fold the state for us.
238 */
239 preempt_set_need_resched();
240 tick_nohz_idle_exit();
241 schedule_preempt_disabled();
242 }
243}
244
245void cpu_startup_entry(enum cpuhp_state state)
246{
247 /*
248 * This #ifdef needs to die, but it's too late in the cycle to
249 * make this generic (arm and sh have never invoked the canary
250 * init for the non boot cpus!). Will be fixed in 3.11
251 */
252#ifdef CONFIG_X86
253 /*
254 * If we're the non-boot CPU, nothing set the stack canary up
255 * for us. The boot CPU already has it initialized but no harm
256 * in doing it again. This is a good place for updating it, as
257 * we wont ever return from this function (so the invalid
258 * canaries already on the stack wont ever trigger).
259 */
260 boot_init_stack_canary();
261#endif
262 __current_set_polling();
263 arch_cpu_idle_prepare();
264 cpu_idle_loop();
265}
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 */
11extern 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 */
17void sched_idle_set_state(struct cpuidle_state *idle_state)
18{
19 idle_set_state(this_rq(), idle_state);
20}
21
22static int __read_mostly cpu_idle_force_poll;
23
24void 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
35static 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
43static 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
52static 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 */
75void __weak arch_cpu_idle_prepare(void) { }
76void __weak arch_cpu_idle_enter(void) { }
77void __weak arch_cpu_idle_exit(void) { }
78void __weak __noreturn arch_cpu_idle_dead(void) { while (1); }
79void __weak arch_cpu_idle(void)
80{
81 cpu_idle_force_poll = 1;
82}
83
84#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE
85DEFINE_STATIC_KEY_FALSE(arch_needs_tick_broadcast);
86
87static inline void cond_tick_broadcast_enter(void)
88{
89 if (static_branch_unlikely(&arch_needs_tick_broadcast))
90 tick_broadcast_enter();
91}
92
93static inline void cond_tick_broadcast_exit(void)
94{
95 if (static_branch_unlikely(&arch_needs_tick_broadcast))
96 tick_broadcast_exit();
97}
98#else
99static inline void cond_tick_broadcast_enter(void) { }
100static inline void cond_tick_broadcast_exit(void) { }
101#endif
102
103/**
104 * default_idle_call - Default CPU idle routine.
105 *
106 * To use when the cpuidle framework cannot be used.
107 */
108void __cpuidle default_idle_call(void)
109{
110 instrumentation_begin();
111 if (!current_clr_polling_and_test()) {
112 cond_tick_broadcast_enter();
113 trace_cpu_idle(1, smp_processor_id());
114 stop_critical_timings();
115
116 ct_cpuidle_enter();
117 arch_cpu_idle();
118 ct_cpuidle_exit();
119
120 start_critical_timings();
121 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
122 cond_tick_broadcast_exit();
123 }
124 local_irq_enable();
125 instrumentation_end();
126}
127
128static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
129 struct cpuidle_device *dev)
130{
131 if (current_clr_polling_and_test())
132 return -EBUSY;
133
134 return cpuidle_enter_s2idle(drv, dev);
135}
136
137static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
138 int next_state)
139{
140 /*
141 * The idle task must be scheduled, it is pointless to go to idle, just
142 * update no idle residency and return.
143 */
144 if (current_clr_polling_and_test()) {
145 dev->last_residency_ns = 0;
146 local_irq_enable();
147 return -EBUSY;
148 }
149
150 /*
151 * Enter the idle state previously returned by the governor decision.
152 * This function will block until an interrupt occurs and will take
153 * care of re-enabling the local interrupts
154 */
155 return cpuidle_enter(drv, dev, next_state);
156}
157
158/**
159 * cpuidle_idle_call - the main idle function
160 *
161 * NOTE: no locks or semaphores should be used here
162 *
163 * On architectures that support TIF_POLLING_NRFLAG, is called with polling
164 * set, and it returns with polling set. If it ever stops polling, it
165 * must clear the polling bit.
166 */
167static void cpuidle_idle_call(void)
168{
169 struct cpuidle_device *dev = cpuidle_get_device();
170 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
171 int next_state, entered_state;
172
173 /*
174 * Check if the idle task must be rescheduled. If it is the
175 * case, exit the function after re-enabling the local IRQ.
176 */
177 if (need_resched()) {
178 local_irq_enable();
179 return;
180 }
181
182 if (cpuidle_not_available(drv, dev)) {
183 tick_nohz_idle_stop_tick();
184
185 default_idle_call();
186 goto exit_idle;
187 }
188
189 /*
190 * Suspend-to-idle ("s2idle") is a system state in which all user space
191 * has been frozen, all I/O devices have been suspended and the only
192 * activity happens here and in interrupts (if any). In that case bypass
193 * the cpuidle governor and go straight for the deepest idle state
194 * available. Possibly also suspend the local tick and the entire
195 * timekeeping to prevent timer interrupts from kicking us out of idle
196 * until a proper wakeup interrupt happens.
197 */
198
199 if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
200 u64 max_latency_ns;
201
202 if (idle_should_enter_s2idle()) {
203
204 entered_state = call_cpuidle_s2idle(drv, dev);
205 if (entered_state > 0)
206 goto exit_idle;
207
208 max_latency_ns = U64_MAX;
209 } else {
210 max_latency_ns = dev->forced_idle_latency_limit_ns;
211 }
212
213 tick_nohz_idle_stop_tick();
214
215 next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
216 call_cpuidle(drv, dev, next_state);
217 } else {
218 bool stop_tick = true;
219
220 /*
221 * Ask the cpuidle framework to choose a convenient idle state.
222 */
223 next_state = cpuidle_select(drv, dev, &stop_tick);
224
225 if (stop_tick || tick_nohz_tick_stopped())
226 tick_nohz_idle_stop_tick();
227 else
228 tick_nohz_idle_retain_tick();
229
230 entered_state = call_cpuidle(drv, dev, next_state);
231 /*
232 * Give the governor an opportunity to reflect on the outcome
233 */
234 cpuidle_reflect(dev, entered_state);
235 }
236
237exit_idle:
238 __current_set_polling();
239
240 /*
241 * It is up to the idle functions to re-enable local interrupts
242 */
243 if (WARN_ON_ONCE(irqs_disabled()))
244 local_irq_enable();
245}
246
247/*
248 * Generic idle loop implementation
249 *
250 * Called with polling cleared.
251 */
252static void do_idle(void)
253{
254 int cpu = smp_processor_id();
255
256 /*
257 * Check if we need to update blocked load
258 */
259 nohz_run_idle_balance(cpu);
260
261 /*
262 * If the arch has a polling bit, we maintain an invariant:
263 *
264 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
265 * rq->idle). This means that, if rq->idle has the polling bit set,
266 * then setting need_resched is guaranteed to cause the CPU to
267 * reschedule.
268 */
269
270 __current_set_polling();
271 tick_nohz_idle_enter();
272
273 while (!need_resched()) {
274
275 /*
276 * Interrupts shouldn't be re-enabled from that point on until
277 * the CPU sleeping instruction is reached. Otherwise an interrupt
278 * may fire and queue a timer that would be ignored until the CPU
279 * wakes from the sleeping instruction. And testing need_resched()
280 * doesn't tell about pending needed timer reprogram.
281 *
282 * Several cases to consider:
283 *
284 * - SLEEP-UNTIL-PENDING-INTERRUPT based instructions such as
285 * "wfi" or "mwait" are fine because they can be entered with
286 * interrupt disabled.
287 *
288 * - sti;mwait() couple is fine because the interrupts are
289 * re-enabled only upon the execution of mwait, leaving no gap
290 * in-between.
291 *
292 * - ROLLBACK based idle handlers with the sleeping instruction
293 * called with interrupts enabled are NOT fine. In this scheme
294 * when the interrupt detects it has interrupted an idle handler,
295 * it rolls back to its beginning which performs the
296 * need_resched() check before re-executing the sleeping
297 * instruction. This can leak a pending needed timer reprogram.
298 * If such a scheme is really mandatory due to the lack of an
299 * appropriate CPU sleeping instruction, then a FAST-FORWARD
300 * must instead be applied: when the interrupt detects it has
301 * interrupted an idle handler, it must resume to the end of
302 * this idle handler so that the generic idle loop is iterated
303 * again to reprogram the tick.
304 */
305 local_irq_disable();
306
307 if (cpu_is_offline(cpu)) {
308 cpuhp_report_idle_dead();
309 arch_cpu_idle_dead();
310 }
311
312 arch_cpu_idle_enter();
313 rcu_nocb_flush_deferred_wakeup();
314
315 /*
316 * In poll mode we re-enable interrupts and spin. Also if we
317 * detected in the wakeup from idle path that the tick
318 * broadcast device expired for us, we don't want to go deep
319 * idle as we know that the IPI is going to arrive right away.
320 */
321 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
322 tick_nohz_idle_restart_tick();
323 cpu_idle_poll();
324 } else {
325 cpuidle_idle_call();
326 }
327 arch_cpu_idle_exit();
328 }
329
330 /*
331 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
332 * be set, propagate it into PREEMPT_NEED_RESCHED.
333 *
334 * This is required because for polling idle loops we will not have had
335 * an IPI to fold the state for us.
336 */
337 preempt_set_need_resched();
338 tick_nohz_idle_exit();
339 __current_clr_polling();
340
341 /*
342 * We promise to call sched_ttwu_pending() and reschedule if
343 * need_resched() is set while polling is set. That means that clearing
344 * polling needs to be visible before doing these things.
345 */
346 smp_mb__after_atomic();
347
348 /*
349 * RCU relies on this call to be done outside of an RCU read-side
350 * critical section.
351 */
352 flush_smp_call_function_queue();
353 schedule_idle();
354
355 if (unlikely(klp_patch_pending(current)))
356 klp_update_patch_state(current);
357}
358
359bool cpu_in_idle(unsigned long pc)
360{
361 return pc >= (unsigned long)__cpuidle_text_start &&
362 pc < (unsigned long)__cpuidle_text_end;
363}
364
365struct idle_timer {
366 struct hrtimer timer;
367 int done;
368};
369
370static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
371{
372 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
373
374 WRITE_ONCE(it->done, 1);
375 set_tsk_need_resched(current);
376
377 return HRTIMER_NORESTART;
378}
379
380void play_idle_precise(u64 duration_ns, u64 latency_ns)
381{
382 struct idle_timer it;
383
384 /*
385 * Only FIFO tasks can disable the tick since they don't need the forced
386 * preemption.
387 */
388 WARN_ON_ONCE(current->policy != SCHED_FIFO);
389 WARN_ON_ONCE(current->nr_cpus_allowed != 1);
390 WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
391 WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
392 WARN_ON_ONCE(!duration_ns);
393 WARN_ON_ONCE(current->mm);
394
395 rcu_sleep_check();
396 preempt_disable();
397 current->flags |= PF_IDLE;
398 cpuidle_use_deepest_state(latency_ns);
399
400 it.done = 0;
401 hrtimer_setup_on_stack(&it.timer, idle_inject_timer_fn, CLOCK_MONOTONIC,
402 HRTIMER_MODE_REL_HARD);
403 hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
404 HRTIMER_MODE_REL_PINNED_HARD);
405
406 while (!READ_ONCE(it.done))
407 do_idle();
408
409 cpuidle_use_deepest_state(0);
410 current->flags &= ~PF_IDLE;
411
412 preempt_fold_need_resched();
413 preempt_enable();
414}
415EXPORT_SYMBOL_GPL(play_idle_precise);
416
417void cpu_startup_entry(enum cpuhp_state state)
418{
419 current->flags |= PF_IDLE;
420 arch_cpu_idle_prepare();
421 cpuhp_online_idle(state);
422 while (1)
423 do_idle();
424}
425
426/*
427 * idle-task scheduling class.
428 */
429
430#ifdef CONFIG_SMP
431static int
432select_task_rq_idle(struct task_struct *p, int cpu, int flags)
433{
434 return task_cpu(p); /* IDLE tasks as never migrated */
435}
436
437static int
438balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
439{
440 return WARN_ON_ONCE(1);
441}
442#endif
443
444/*
445 * Idle tasks are unconditionally rescheduled:
446 */
447static void wakeup_preempt_idle(struct rq *rq, struct task_struct *p, int flags)
448{
449 resched_curr(rq);
450}
451
452static void put_prev_task_idle(struct rq *rq, struct task_struct *prev, struct task_struct *next)
453{
454 dl_server_update_idle_time(rq, prev);
455 scx_update_idle(rq, false, true);
456}
457
458static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
459{
460 update_idle_core(rq);
461 scx_update_idle(rq, true, true);
462 schedstat_inc(rq->sched_goidle);
463 next->se.exec_start = rq_clock_task(rq);
464}
465
466struct task_struct *pick_task_idle(struct rq *rq)
467{
468 scx_update_idle(rq, true, false);
469 return rq->idle;
470}
471
472/*
473 * It is not legal to sleep in the idle task - print a warning
474 * message if some code attempts to do it:
475 */
476static bool
477dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
478{
479 raw_spin_rq_unlock_irq(rq);
480 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
481 dump_stack();
482 raw_spin_rq_lock_irq(rq);
483 return true;
484}
485
486/*
487 * scheduler tick hitting a task of our scheduling class.
488 *
489 * NOTE: This function can be called remotely by the tick offload that
490 * goes along full dynticks. Therefore no local assumption can be made
491 * and everything must be accessed through the @rq and @curr passed in
492 * parameters.
493 */
494static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
495{
496}
497
498static void switched_to_idle(struct rq *rq, struct task_struct *p)
499{
500 BUG();
501}
502
503static void
504prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
505{
506 BUG();
507}
508
509static void update_curr_idle(struct rq *rq)
510{
511}
512
513/*
514 * Simple, special scheduling class for the per-CPU idle tasks:
515 */
516DEFINE_SCHED_CLASS(idle) = {
517
518 /* no enqueue/yield_task for idle tasks */
519
520 /* dequeue is not valid, we print a debug message there: */
521 .dequeue_task = dequeue_task_idle,
522
523 .wakeup_preempt = wakeup_preempt_idle,
524
525 .pick_task = pick_task_idle,
526 .put_prev_task = put_prev_task_idle,
527 .set_next_task = set_next_task_idle,
528
529#ifdef CONFIG_SMP
530 .balance = balance_idle,
531 .select_task_rq = select_task_rq_idle,
532 .set_cpus_allowed = set_cpus_allowed_common,
533#endif
534
535 .task_tick = task_tick_idle,
536
537 .prio_changed = prio_changed_idle,
538 .switched_to = switched_to_idle,
539 .update_curr = update_curr_idle,
540};