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/cpuhotplug.h>
8#include <linux/tick.h>
9#include <linux/mm.h>
10#include <linux/stackprotector.h>
11#include <linux/suspend.h>
12
13#include <asm/tlb.h>
14
15#include <trace/events/power.h>
16
17#include "sched.h"
18
19/**
20 * sched_idle_set_state - Record idle state for the current CPU.
21 * @idle_state: State to record.
22 */
23void sched_idle_set_state(struct cpuidle_state *idle_state)
24{
25 idle_set_state(this_rq(), idle_state);
26}
27
28static int __read_mostly cpu_idle_force_poll;
29
30void cpu_idle_poll_ctrl(bool enable)
31{
32 if (enable) {
33 cpu_idle_force_poll++;
34 } else {
35 cpu_idle_force_poll--;
36 WARN_ON_ONCE(cpu_idle_force_poll < 0);
37 }
38}
39
40#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
41static int __init cpu_idle_poll_setup(char *__unused)
42{
43 cpu_idle_force_poll = 1;
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 return 1;
52}
53__setup("hlt", cpu_idle_nopoll_setup);
54#endif
55
56static inline int cpu_idle_poll(void)
57{
58 rcu_idle_enter();
59 trace_cpu_idle_rcuidle(0, smp_processor_id());
60 local_irq_enable();
61 stop_critical_timings();
62 while (!tif_need_resched() &&
63 (cpu_idle_force_poll || tick_check_broadcast_expired()))
64 cpu_relax();
65 start_critical_timings();
66 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
67 rcu_idle_exit();
68 return 1;
69}
70
71/* Weak implementations for optional arch specific functions */
72void __weak arch_cpu_idle_prepare(void) { }
73void __weak arch_cpu_idle_enter(void) { }
74void __weak arch_cpu_idle_exit(void) { }
75void __weak arch_cpu_idle_dead(void) { }
76void __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 */
87void 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
98static 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 */
128static void cpuidle_idle_call(void)
129{
130 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
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 * Tell the RCU framework we are entering an idle section,
145 * so no more rcu read side critical sections and one more
146 * step to the grace period
147 */
148 rcu_idle_enter();
149
150 if (cpuidle_not_available(drv, dev)) {
151 default_idle_call();
152 goto exit_idle;
153 }
154
155 /*
156 * Suspend-to-idle ("freeze") is a system state in which all user space
157 * has been frozen, all I/O devices have been suspended and the only
158 * activity happens here and in iterrupts (if any). In that case bypass
159 * the cpuidle governor and go stratight for the deepest idle state
160 * available. Possibly also suspend the local tick and the entire
161 * timekeeping to prevent timer interrupts from kicking us out of idle
162 * until a proper wakeup interrupt happens.
163 */
164 if (idle_should_freeze()) {
165 entered_state = cpuidle_enter_freeze(drv, dev);
166 if (entered_state > 0) {
167 local_irq_enable();
168 goto exit_idle;
169 }
170
171 next_state = cpuidle_find_deepest_state(drv, dev);
172 call_cpuidle(drv, dev, next_state);
173 } else {
174 /*
175 * Ask the cpuidle framework to choose a convenient idle state.
176 */
177 next_state = cpuidle_select(drv, dev);
178 entered_state = call_cpuidle(drv, dev, next_state);
179 /*
180 * Give the governor an opportunity to reflect on the outcome
181 */
182 cpuidle_reflect(dev, entered_state);
183 }
184
185exit_idle:
186 __current_set_polling();
187
188 /*
189 * It is up to the idle functions to reenable local interrupts
190 */
191 if (WARN_ON_ONCE(irqs_disabled()))
192 local_irq_enable();
193
194 rcu_idle_exit();
195}
196
197/*
198 * Generic idle loop implementation
199 *
200 * Called with polling cleared.
201 */
202static void cpu_idle_loop(void)
203{
204 while (1) {
205 /*
206 * If the arch has a polling bit, we maintain an invariant:
207 *
208 * Our polling bit is clear if we're not scheduled (i.e. if
209 * rq->curr != rq->idle). This means that, if rq->idle has
210 * the polling bit set, then setting need_resched is
211 * guaranteed to cause the cpu to reschedule.
212 */
213
214 __current_set_polling();
215 quiet_vmstat();
216 tick_nohz_idle_enter();
217
218 while (!need_resched()) {
219 check_pgt_cache();
220 rmb();
221
222 if (cpu_is_offline(smp_processor_id())) {
223 cpuhp_report_idle_dead();
224 arch_cpu_idle_dead();
225 }
226
227 local_irq_disable();
228 arch_cpu_idle_enter();
229
230 /*
231 * In poll mode we reenable interrupts and spin.
232 *
233 * Also if we detected in the wakeup from idle
234 * path that the tick broadcast device expired
235 * for us, we don't want to go deep idle as we
236 * know that the IPI is going to arrive right
237 * away
238 */
239 if (cpu_idle_force_poll || tick_check_broadcast_expired())
240 cpu_idle_poll();
241 else
242 cpuidle_idle_call();
243
244 arch_cpu_idle_exit();
245 }
246
247 /*
248 * Since we fell out of the loop above, we know
249 * TIF_NEED_RESCHED must be set, propagate it into
250 * PREEMPT_NEED_RESCHED.
251 *
252 * This is required because for polling idle loops we will
253 * not have had an IPI to fold the state for us.
254 */
255 preempt_set_need_resched();
256 tick_nohz_idle_exit();
257 __current_clr_polling();
258
259 /*
260 * We promise to call sched_ttwu_pending and reschedule
261 * if need_resched is set while polling is set. That
262 * means that clearing polling needs to be visible
263 * before doing these things.
264 */
265 smp_mb__after_atomic();
266
267 sched_ttwu_pending();
268 schedule_preempt_disabled();
269 }
270}
271
272void cpu_startup_entry(enum cpuhp_state state)
273{
274 /*
275 * This #ifdef needs to die, but it's too late in the cycle to
276 * make this generic (arm and sh have never invoked the canary
277 * init for the non boot cpus!). Will be fixed in 3.11
278 */
279#ifdef CONFIG_X86
280 /*
281 * If we're the non-boot CPU, nothing set the stack canary up
282 * for us. The boot CPU already has it initialized but no harm
283 * in doing it again. This is a good place for updating it, as
284 * we wont ever return from this function (so the invalid
285 * canaries already on the stack wont ever trigger).
286 */
287 boot_init_stack_canary();
288#endif
289 arch_cpu_idle_prepare();
290 cpuhp_online_idle(state);
291 cpu_idle_loop();
292}
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 "sched.h"
10
11#include <trace/events/power.h>
12
13/* Linker adds these: start and end of __cpuidle functions */
14extern char __cpuidle_text_start[], __cpuidle_text_end[];
15
16/**
17 * sched_idle_set_state - Record idle state for the current CPU.
18 * @idle_state: State to record.
19 */
20void sched_idle_set_state(struct cpuidle_state *idle_state)
21{
22 idle_set_state(this_rq(), idle_state);
23}
24
25static int __read_mostly cpu_idle_force_poll;
26
27void cpu_idle_poll_ctrl(bool enable)
28{
29 if (enable) {
30 cpu_idle_force_poll++;
31 } else {
32 cpu_idle_force_poll--;
33 WARN_ON_ONCE(cpu_idle_force_poll < 0);
34 }
35}
36
37#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
38static int __init cpu_idle_poll_setup(char *__unused)
39{
40 cpu_idle_force_poll = 1;
41
42 return 1;
43}
44__setup("nohlt", cpu_idle_poll_setup);
45
46static int __init cpu_idle_nopoll_setup(char *__unused)
47{
48 cpu_idle_force_poll = 0;
49
50 return 1;
51}
52__setup("hlt", cpu_idle_nopoll_setup);
53#endif
54
55static noinline int __cpuidle cpu_idle_poll(void)
56{
57 trace_cpu_idle(0, smp_processor_id());
58 stop_critical_timings();
59 rcu_idle_enter();
60 local_irq_enable();
61
62 while (!tif_need_resched() &&
63 (cpu_idle_force_poll || tick_check_broadcast_expired()))
64 cpu_relax();
65
66 rcu_idle_exit();
67 start_critical_timings();
68 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
69
70 return 1;
71}
72
73/* Weak implementations for optional arch specific functions */
74void __weak arch_cpu_idle_prepare(void) { }
75void __weak arch_cpu_idle_enter(void) { }
76void __weak arch_cpu_idle_exit(void) { }
77void __weak arch_cpu_idle_dead(void) { }
78void __weak arch_cpu_idle(void)
79{
80 cpu_idle_force_poll = 1;
81 local_irq_enable();
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 if (current_clr_polling_and_test()) {
92 local_irq_enable();
93 } else {
94
95 trace_cpu_idle(1, smp_processor_id());
96 stop_critical_timings();
97 rcu_idle_enter();
98 arch_cpu_idle();
99 rcu_idle_exit();
100 start_critical_timings();
101 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
102 }
103}
104
105static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
106 struct cpuidle_device *dev)
107{
108 if (current_clr_polling_and_test())
109 return -EBUSY;
110
111 return cpuidle_enter_s2idle(drv, dev);
112}
113
114static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
115 int next_state)
116{
117 /*
118 * The idle task must be scheduled, it is pointless to go to idle, just
119 * update no idle residency and return.
120 */
121 if (current_clr_polling_and_test()) {
122 dev->last_residency_ns = 0;
123 local_irq_enable();
124 return -EBUSY;
125 }
126
127 /*
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
131 */
132 return cpuidle_enter(drv, dev, next_state);
133}
134
135/**
136 * cpuidle_idle_call - the main idle function
137 *
138 * NOTE: no locks or semaphores should be used here
139 *
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.
143 */
144static void cpuidle_idle_call(void)
145{
146 struct cpuidle_device *dev = cpuidle_get_device();
147 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
148 int next_state, entered_state;
149
150 /*
151 * Check if the idle task must be rescheduled. If it is the
152 * case, exit the function after re-enabling the local irq.
153 */
154 if (need_resched()) {
155 local_irq_enable();
156 return;
157 }
158
159 /*
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
163 */
164
165 if (cpuidle_not_available(drv, dev)) {
166 tick_nohz_idle_stop_tick();
167
168 default_idle_call();
169 goto exit_idle;
170 }
171
172 /*
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.
180 */
181
182 if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
183 u64 max_latency_ns;
184
185 if (idle_should_enter_s2idle()) {
186
187 entered_state = call_cpuidle_s2idle(drv, dev);
188 if (entered_state > 0)
189 goto exit_idle;
190
191 max_latency_ns = U64_MAX;
192 } else {
193 max_latency_ns = dev->forced_idle_latency_limit_ns;
194 }
195
196 tick_nohz_idle_stop_tick();
197
198 next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
199 call_cpuidle(drv, dev, next_state);
200 } else {
201 bool stop_tick = true;
202
203 /*
204 * Ask the cpuidle framework to choose a convenient idle state.
205 */
206 next_state = cpuidle_select(drv, dev, &stop_tick);
207
208 if (stop_tick || tick_nohz_tick_stopped())
209 tick_nohz_idle_stop_tick();
210 else
211 tick_nohz_idle_retain_tick();
212
213 entered_state = call_cpuidle(drv, dev, next_state);
214 /*
215 * Give the governor an opportunity to reflect on the outcome
216 */
217 cpuidle_reflect(dev, entered_state);
218 }
219
220exit_idle:
221 __current_set_polling();
222
223 /*
224 * It is up to the idle functions to reenable local interrupts
225 */
226 if (WARN_ON_ONCE(irqs_disabled()))
227 local_irq_enable();
228}
229
230/*
231 * Generic idle loop implementation
232 *
233 * Called with polling cleared.
234 */
235static void do_idle(void)
236{
237 int cpu = smp_processor_id();
238 /*
239 * If the arch has a polling bit, we maintain an invariant:
240 *
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
244 * reschedule.
245 */
246
247 __current_set_polling();
248 tick_nohz_idle_enter();
249
250 while (!need_resched()) {
251 rmb();
252
253 local_irq_disable();
254
255 if (cpu_is_offline(cpu)) {
256 tick_nohz_idle_stop_tick();
257 cpuhp_report_idle_dead();
258 arch_cpu_idle_dead();
259 }
260
261 arch_cpu_idle_enter();
262
263 /*
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.
268 */
269 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
270 tick_nohz_idle_restart_tick();
271 cpu_idle_poll();
272 } else {
273 cpuidle_idle_call();
274 }
275 arch_cpu_idle_exit();
276 }
277
278 /*
279 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
280 * be set, propagate it into PREEMPT_NEED_RESCHED.
281 *
282 * This is required because for polling idle loops we will not have had
283 * an IPI to fold the state for us.
284 */
285 preempt_set_need_resched();
286 tick_nohz_idle_exit();
287 __current_clr_polling();
288
289 /*
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.
293 */
294 smp_mb__after_atomic();
295
296 /*
297 * RCU relies on this call to be done outside of an RCU read-side
298 * critical section.
299 */
300 flush_smp_call_function_from_idle();
301 schedule_idle();
302
303 if (unlikely(klp_patch_pending(current)))
304 klp_update_patch_state(current);
305}
306
307bool cpu_in_idle(unsigned long pc)
308{
309 return pc >= (unsigned long)__cpuidle_text_start &&
310 pc < (unsigned long)__cpuidle_text_end;
311}
312
313struct idle_timer {
314 struct hrtimer timer;
315 int done;
316};
317
318static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
319{
320 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
321
322 WRITE_ONCE(it->done, 1);
323 set_tsk_need_resched(current);
324
325 return HRTIMER_NORESTART;
326}
327
328void play_idle_precise(u64 duration_ns, u64 latency_ns)
329{
330 struct idle_timer it;
331
332 /*
333 * Only FIFO tasks can disable the tick since they don't need the forced
334 * preemption.
335 */
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);
341
342 rcu_sleep_check();
343 preempt_disable();
344 current->flags |= PF_IDLE;
345 cpuidle_use_deepest_state(latency_ns);
346
347 it.done = 0;
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);
352
353 while (!READ_ONCE(it.done))
354 do_idle();
355
356 cpuidle_use_deepest_state(0);
357 current->flags &= ~PF_IDLE;
358
359 preempt_fold_need_resched();
360 preempt_enable();
361}
362EXPORT_SYMBOL_GPL(play_idle_precise);
363
364void cpu_startup_entry(enum cpuhp_state state)
365{
366 arch_cpu_idle_prepare();
367 cpuhp_online_idle(state);
368 while (1)
369 do_idle();
370}
371
372/*
373 * idle-task scheduling class.
374 */
375
376#ifdef CONFIG_SMP
377static int
378select_task_rq_idle(struct task_struct *p, int cpu, int sd_flag, int flags)
379{
380 return task_cpu(p); /* IDLE tasks as never migrated */
381}
382
383static int
384balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
385{
386 return WARN_ON_ONCE(1);
387}
388#endif
389
390/*
391 * Idle tasks are unconditionally rescheduled:
392 */
393static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
394{
395 resched_curr(rq);
396}
397
398static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
399{
400}
401
402static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
403{
404 update_idle_core(rq);
405 schedstat_inc(rq->sched_goidle);
406}
407
408struct task_struct *pick_next_task_idle(struct rq *rq)
409{
410 struct task_struct *next = rq->idle;
411
412 set_next_task_idle(rq, next, true);
413
414 return next;
415}
416
417/*
418 * It is not legal to sleep in the idle task - print a warning
419 * message if some code attempts to do it:
420 */
421static void
422dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
423{
424 raw_spin_unlock_irq(&rq->lock);
425 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
426 dump_stack();
427 raw_spin_lock_irq(&rq->lock);
428}
429
430/*
431 * scheduler tick hitting a task of our scheduling class.
432 *
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
436 * parameters.
437 */
438static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
439{
440}
441
442static void switched_to_idle(struct rq *rq, struct task_struct *p)
443{
444 BUG();
445}
446
447static void
448prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
449{
450 BUG();
451}
452
453static void update_curr_idle(struct rq *rq)
454{
455}
456
457/*
458 * Simple, special scheduling class for the per-CPU idle tasks:
459 */
460const struct sched_class idle_sched_class
461 __attribute__((section("__idle_sched_class"))) = {
462 /* no enqueue/yield_task for idle tasks */
463
464 /* dequeue is not valid, we print a debug message there: */
465 .dequeue_task = dequeue_task_idle,
466
467 .check_preempt_curr = check_preempt_curr_idle,
468
469 .pick_next_task = pick_next_task_idle,
470 .put_prev_task = put_prev_task_idle,
471 .set_next_task = set_next_task_idle,
472
473#ifdef CONFIG_SMP
474 .balance = balance_idle,
475 .select_task_rq = select_task_rq_idle,
476 .set_cpus_allowed = set_cpus_allowed_common,
477#endif
478
479 .task_tick = task_tick_idle,
480
481 .prio_changed = prio_changed_idle,
482 .switched_to = switched_to_idle,
483 .update_curr = update_curr_idle,
484};