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