xref: /linux/kernel/sched/idle.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 	/* Fall back to the default arch idle method on errors. */
101 	if (next_state < 0) {
102 		default_idle_call();
103 		return next_state;
104 	}
105 
106 	/*
107 	 * The idle task must be scheduled, it is pointless to go to idle, just
108 	 * update no idle residency and return.
109 	 */
110 	if (current_clr_polling_and_test()) {
111 		dev->last_residency = 0;
112 		local_irq_enable();
113 		return -EBUSY;
114 	}
115 
116 	/*
117 	 * Enter the idle state previously returned by the governor decision.
118 	 * This function will block until an interrupt occurs and will take
119 	 * care of re-enabling the local interrupts
120 	 */
121 	return cpuidle_enter(drv, dev, next_state);
122 }
123 
124 /**
125  * cpuidle_idle_call - the main idle function
126  *
127  * NOTE: no locks or semaphores should be used here
128  *
129  * On archs that support TIF_POLLING_NRFLAG, is called with polling
130  * set, and it returns with polling set.  If it ever stops polling, it
131  * must clear the polling bit.
132  */
133 static void cpuidle_idle_call(void)
134 {
135 	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
136 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
137 	int next_state, entered_state;
138 
139 	/*
140 	 * Check if the idle task must be rescheduled. If it is the
141 	 * case, exit the function after re-enabling the local irq.
142 	 */
143 	if (need_resched()) {
144 		local_irq_enable();
145 		return;
146 	}
147 
148 	/*
149 	 * Tell the RCU framework we are entering an idle section,
150 	 * so no more rcu read side critical sections and one more
151 	 * step to the grace period
152 	 */
153 	rcu_idle_enter();
154 
155 	if (cpuidle_not_available(drv, dev)) {
156 		default_idle_call();
157 		goto exit_idle;
158 	}
159 
160 	/*
161 	 * Suspend-to-idle ("freeze") is a system state in which all user space
162 	 * has been frozen, all I/O devices have been suspended and the only
163 	 * activity happens here and in iterrupts (if any).  In that case bypass
164 	 * the cpuidle governor and go stratight for the deepest idle state
165 	 * available.  Possibly also suspend the local tick and the entire
166 	 * timekeeping to prevent timer interrupts from kicking us out of idle
167 	 * until a proper wakeup interrupt happens.
168 	 */
169 	if (idle_should_freeze()) {
170 		entered_state = cpuidle_enter_freeze(drv, dev);
171 		if (entered_state >= 0) {
172 			local_irq_enable();
173 			goto exit_idle;
174 		}
175 
176 		next_state = cpuidle_find_deepest_state(drv, dev);
177 		call_cpuidle(drv, dev, next_state);
178 	} else {
179 		/*
180 		 * Ask the cpuidle framework to choose a convenient idle state.
181 		 */
182 		next_state = cpuidle_select(drv, dev);
183 		entered_state = call_cpuidle(drv, dev, next_state);
184 		/*
185 		 * Give the governor an opportunity to reflect on the outcome
186 		 */
187 		cpuidle_reflect(dev, entered_state);
188 	}
189 
190 exit_idle:
191 	__current_set_polling();
192 
193 	/*
194 	 * It is up to the idle functions to reenable local interrupts
195 	 */
196 	if (WARN_ON_ONCE(irqs_disabled()))
197 		local_irq_enable();
198 
199 	rcu_idle_exit();
200 }
201 
202 DEFINE_PER_CPU(bool, cpu_dead_idle);
203 
204 /*
205  * Generic idle loop implementation
206  *
207  * Called with polling cleared.
208  */
209 static void cpu_idle_loop(void)
210 {
211 	while (1) {
212 		/*
213 		 * If the arch has a polling bit, we maintain an invariant:
214 		 *
215 		 * Our polling bit is clear if we're not scheduled (i.e. if
216 		 * rq->curr != rq->idle).  This means that, if rq->idle has
217 		 * the polling bit set, then setting need_resched is
218 		 * guaranteed to cause the cpu to reschedule.
219 		 */
220 
221 		__current_set_polling();
222 		tick_nohz_idle_enter();
223 
224 		while (!need_resched()) {
225 			check_pgt_cache();
226 			rmb();
227 
228 			if (cpu_is_offline(smp_processor_id())) {
229 				rcu_cpu_notify(NULL, CPU_DYING_IDLE,
230 					       (void *)(long)smp_processor_id());
231 				smp_mb(); /* all activity before dead. */
232 				this_cpu_write(cpu_dead_idle, true);
233 				arch_cpu_idle_dead();
234 			}
235 
236 			local_irq_disable();
237 			arch_cpu_idle_enter();
238 
239 			/*
240 			 * In poll mode we reenable interrupts and spin.
241 			 *
242 			 * Also if we detected in the wakeup from idle
243 			 * path that the tick broadcast device expired
244 			 * for us, we don't want to go deep idle as we
245 			 * know that the IPI is going to arrive right
246 			 * away
247 			 */
248 			if (cpu_idle_force_poll || tick_check_broadcast_expired())
249 				cpu_idle_poll();
250 			else
251 				cpuidle_idle_call();
252 
253 			arch_cpu_idle_exit();
254 		}
255 
256 		/*
257 		 * Since we fell out of the loop above, we know
258 		 * TIF_NEED_RESCHED must be set, propagate it into
259 		 * PREEMPT_NEED_RESCHED.
260 		 *
261 		 * This is required because for polling idle loops we will
262 		 * not have had an IPI to fold the state for us.
263 		 */
264 		preempt_set_need_resched();
265 		tick_nohz_idle_exit();
266 		__current_clr_polling();
267 
268 		/*
269 		 * We promise to call sched_ttwu_pending and reschedule
270 		 * if need_resched is set while polling is set.  That
271 		 * means that clearing polling needs to be visible
272 		 * before doing these things.
273 		 */
274 		smp_mb__after_atomic();
275 
276 		sched_ttwu_pending();
277 		schedule_preempt_disabled();
278 	}
279 }
280 
281 void cpu_startup_entry(enum cpuhp_state state)
282 {
283 	/*
284 	 * This #ifdef needs to die, but it's too late in the cycle to
285 	 * make this generic (arm and sh have never invoked the canary
286 	 * init for the non boot cpus!). Will be fixed in 3.11
287 	 */
288 #ifdef CONFIG_X86
289 	/*
290 	 * If we're the non-boot CPU, nothing set the stack canary up
291 	 * for us. The boot CPU already has it initialized but no harm
292 	 * in doing it again. This is a good place for updating it, as
293 	 * we wont ever return from this function (so the invalid
294 	 * canaries already on the stack wont ever trigger).
295 	 */
296 	boot_init_stack_canary();
297 #endif
298 	arch_cpu_idle_prepare();
299 	cpu_idle_loop();
300 }
301