xref: /linux/kernel/sched/idle.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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 #include <linux/suspend.h>
11 
12 #include <asm/tlb.h>
13 
14 #include <trace/events/power.h>
15 
16 #include "sched.h"
17 
18 /**
19  * sched_idle_set_state - Record idle state for the current CPU.
20  * @idle_state: State to record.
21  */
22 void sched_idle_set_state(struct cpuidle_state *idle_state)
23 {
24 	idle_set_state(this_rq(), idle_state);
25 }
26 
27 static int __read_mostly cpu_idle_force_poll;
28 
29 void 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
40 static int __init cpu_idle_poll_setup(char *__unused)
41 {
42 	cpu_idle_force_poll = 1;
43 	return 1;
44 }
45 __setup("nohlt", cpu_idle_poll_setup);
46 
47 static int __init cpu_idle_nopoll_setup(char *__unused)
48 {
49 	cpu_idle_force_poll = 0;
50 	return 1;
51 }
52 __setup("hlt", cpu_idle_nopoll_setup);
53 #endif
54 
55 static inline int cpu_idle_poll(void)
56 {
57 	rcu_idle_enter();
58 	trace_cpu_idle_rcuidle(0, smp_processor_id());
59 	local_irq_enable();
60 	stop_critical_timings();
61 	while (!tif_need_resched() &&
62 		(cpu_idle_force_poll || tick_check_broadcast_expired()))
63 		cpu_relax();
64 	start_critical_timings();
65 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
66 	rcu_idle_exit();
67 	return 1;
68 }
69 
70 /* Weak implementations for optional arch specific functions */
71 void __weak arch_cpu_idle_prepare(void) { }
72 void __weak arch_cpu_idle_enter(void) { }
73 void __weak arch_cpu_idle_exit(void) { }
74 void __weak arch_cpu_idle_dead(void) { }
75 void __weak arch_cpu_idle(void)
76 {
77 	cpu_idle_force_poll = 1;
78 	local_irq_enable();
79 }
80 
81 /**
82  * default_idle_call - Default CPU idle routine.
83  *
84  * To use when the cpuidle framework cannot be used.
85  */
86 void default_idle_call(void)
87 {
88 	if (current_clr_polling_and_test()) {
89 		local_irq_enable();
90 	} else {
91 		stop_critical_timings();
92 		arch_cpu_idle();
93 		start_critical_timings();
94 	}
95 }
96 
97 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
98 		      int next_state)
99 {
100 	/*
101 	 * The idle task must be scheduled, it is pointless to go to idle, just
102 	 * update no idle residency and return.
103 	 */
104 	if (current_clr_polling_and_test()) {
105 		dev->last_residency = 0;
106 		local_irq_enable();
107 		return -EBUSY;
108 	}
109 
110 	/*
111 	 * Enter the idle state previously returned by the governor decision.
112 	 * This function will block until an interrupt occurs and will take
113 	 * care of re-enabling the local interrupts
114 	 */
115 	return cpuidle_enter(drv, dev, next_state);
116 }
117 
118 /**
119  * cpuidle_idle_call - the main idle function
120  *
121  * NOTE: no locks or semaphores should be used here
122  *
123  * On archs that support TIF_POLLING_NRFLAG, is called with polling
124  * set, and it returns with polling set.  If it ever stops polling, it
125  * must clear the polling bit.
126  */
127 static void cpuidle_idle_call(void)
128 {
129 	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
130 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
131 	int next_state, entered_state;
132 
133 	/*
134 	 * Check if the idle task must be rescheduled. If it is the
135 	 * case, exit the function after re-enabling the local irq.
136 	 */
137 	if (need_resched()) {
138 		local_irq_enable();
139 		return;
140 	}
141 
142 	/*
143 	 * Tell the RCU framework we are entering an idle section,
144 	 * so no more rcu read side critical sections and one more
145 	 * step to the grace period
146 	 */
147 	rcu_idle_enter();
148 
149 	if (cpuidle_not_available(drv, dev)) {
150 		default_idle_call();
151 		goto exit_idle;
152 	}
153 
154 	/*
155 	 * Suspend-to-idle ("freeze") is a system state in which all user space
156 	 * has been frozen, all I/O devices have been suspended and the only
157 	 * activity happens here and in iterrupts (if any).  In that case bypass
158 	 * the cpuidle governor and go stratight for the deepest idle state
159 	 * available.  Possibly also suspend the local tick and the entire
160 	 * timekeeping to prevent timer interrupts from kicking us out of idle
161 	 * until a proper wakeup interrupt happens.
162 	 */
163 	if (idle_should_freeze()) {
164 		entered_state = cpuidle_enter_freeze(drv, dev);
165 		if (entered_state > 0) {
166 			local_irq_enable();
167 			goto exit_idle;
168 		}
169 
170 		next_state = cpuidle_find_deepest_state(drv, dev);
171 		call_cpuidle(drv, dev, next_state);
172 	} else {
173 		/*
174 		 * Ask the cpuidle framework to choose a convenient idle state.
175 		 */
176 		next_state = cpuidle_select(drv, dev);
177 		entered_state = call_cpuidle(drv, dev, next_state);
178 		/*
179 		 * Give the governor an opportunity to reflect on the outcome
180 		 */
181 		cpuidle_reflect(dev, entered_state);
182 	}
183 
184 exit_idle:
185 	__current_set_polling();
186 
187 	/*
188 	 * It is up to the idle functions to reenable local interrupts
189 	 */
190 	if (WARN_ON_ONCE(irqs_disabled()))
191 		local_irq_enable();
192 
193 	rcu_idle_exit();
194 }
195 
196 DEFINE_PER_CPU(bool, cpu_dead_idle);
197 
198 /*
199  * Generic idle loop implementation
200  *
201  * Called with polling cleared.
202  */
203 static void cpu_idle_loop(void)
204 {
205 	while (1) {
206 		/*
207 		 * If the arch has a polling bit, we maintain an invariant:
208 		 *
209 		 * Our polling bit is clear if we're not scheduled (i.e. if
210 		 * rq->curr != rq->idle).  This means that, if rq->idle has
211 		 * the polling bit set, then setting need_resched is
212 		 * guaranteed to cause the cpu to reschedule.
213 		 */
214 
215 		__current_set_polling();
216 		quiet_vmstat();
217 		tick_nohz_idle_enter();
218 
219 		while (!need_resched()) {
220 			check_pgt_cache();
221 			rmb();
222 
223 			if (cpu_is_offline(smp_processor_id())) {
224 				rcu_cpu_notify(NULL, CPU_DYING_IDLE,
225 					       (void *)(long)smp_processor_id());
226 				smp_mb(); /* all activity before dead. */
227 				this_cpu_write(cpu_dead_idle, true);
228 				arch_cpu_idle_dead();
229 			}
230 
231 			local_irq_disable();
232 			arch_cpu_idle_enter();
233 
234 			/*
235 			 * In poll mode we reenable interrupts and spin.
236 			 *
237 			 * Also if we detected in the wakeup from idle
238 			 * path that the tick broadcast device expired
239 			 * for us, we don't want to go deep idle as we
240 			 * know that the IPI is going to arrive right
241 			 * away
242 			 */
243 			if (cpu_idle_force_poll || tick_check_broadcast_expired())
244 				cpu_idle_poll();
245 			else
246 				cpuidle_idle_call();
247 
248 			arch_cpu_idle_exit();
249 		}
250 
251 		/*
252 		 * Since we fell out of the loop above, we know
253 		 * TIF_NEED_RESCHED must be set, propagate it into
254 		 * PREEMPT_NEED_RESCHED.
255 		 *
256 		 * This is required because for polling idle loops we will
257 		 * not have had an IPI to fold the state for us.
258 		 */
259 		preempt_set_need_resched();
260 		tick_nohz_idle_exit();
261 		__current_clr_polling();
262 
263 		/*
264 		 * We promise to call sched_ttwu_pending and reschedule
265 		 * if need_resched is set while polling is set.  That
266 		 * means that clearing polling needs to be visible
267 		 * before doing these things.
268 		 */
269 		smp_mb__after_atomic();
270 
271 		sched_ttwu_pending();
272 		schedule_preempt_disabled();
273 	}
274 }
275 
276 void cpu_startup_entry(enum cpuhp_state state)
277 {
278 	/*
279 	 * This #ifdef needs to die, but it's too late in the cycle to
280 	 * make this generic (arm and sh have never invoked the canary
281 	 * init for the non boot cpus!). Will be fixed in 3.11
282 	 */
283 #ifdef CONFIG_X86
284 	/*
285 	 * If we're the non-boot CPU, nothing set the stack canary up
286 	 * for us. The boot CPU already has it initialized but no harm
287 	 * in doing it again. This is a good place for updating it, as
288 	 * we wont ever return from this function (so the invalid
289 	 * canaries already on the stack wont ever trigger).
290 	 */
291 	boot_init_stack_canary();
292 #endif
293 	arch_cpu_idle_prepare();
294 	cpu_idle_loop();
295 }
296