xref: /linux/kernel/sched/idle.c (revision 04303f8ec14269b0ea2553863553bc7eaadca1f8)
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 
15 #include "sched.h"
16 
17 static int __read_mostly cpu_idle_force_poll;
18 
19 void cpu_idle_poll_ctrl(bool enable)
20 {
21 	if (enable) {
22 		cpu_idle_force_poll++;
23 	} else {
24 		cpu_idle_force_poll--;
25 		WARN_ON_ONCE(cpu_idle_force_poll < 0);
26 	}
27 }
28 
29 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
30 static int __init cpu_idle_poll_setup(char *__unused)
31 {
32 	cpu_idle_force_poll = 1;
33 	return 1;
34 }
35 __setup("nohlt", cpu_idle_poll_setup);
36 
37 static int __init cpu_idle_nopoll_setup(char *__unused)
38 {
39 	cpu_idle_force_poll = 0;
40 	return 1;
41 }
42 __setup("hlt", cpu_idle_nopoll_setup);
43 #endif
44 
45 static inline int cpu_idle_poll(void)
46 {
47 	rcu_idle_enter();
48 	trace_cpu_idle_rcuidle(0, smp_processor_id());
49 	local_irq_enable();
50 	while (!tif_need_resched() &&
51 		(cpu_idle_force_poll || tick_check_broadcast_expired()))
52 		cpu_relax();
53 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
54 	rcu_idle_exit();
55 	return 1;
56 }
57 
58 /* Weak implementations for optional arch specific functions */
59 void __weak arch_cpu_idle_prepare(void) { }
60 void __weak arch_cpu_idle_enter(void) { }
61 void __weak arch_cpu_idle_exit(void) { }
62 void __weak arch_cpu_idle_dead(void) { }
63 void __weak arch_cpu_idle(void)
64 {
65 	cpu_idle_force_poll = 1;
66 	local_irq_enable();
67 }
68 
69 /**
70  * cpuidle_idle_call - the main idle function
71  *
72  * NOTE: no locks or semaphores should be used here
73  *
74  * On archs that support TIF_POLLING_NRFLAG, is called with polling
75  * set, and it returns with polling set.  If it ever stops polling, it
76  * must clear the polling bit.
77  */
78 static void cpuidle_idle_call(void)
79 {
80 	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
81 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
82 	int next_state, entered_state;
83 	unsigned int broadcast;
84 
85 	/*
86 	 * Check if the idle task must be rescheduled. If it is the
87 	 * case, exit the function after re-enabling the local irq.
88 	 */
89 	if (need_resched()) {
90 		local_irq_enable();
91 		return;
92 	}
93 
94 	/*
95 	 * During the idle period, stop measuring the disabled irqs
96 	 * critical sections latencies
97 	 */
98 	stop_critical_timings();
99 
100 	/*
101 	 * Tell the RCU framework we are entering an idle section,
102 	 * so no more rcu read side critical sections and one more
103 	 * step to the grace period
104 	 */
105 	rcu_idle_enter();
106 
107 	/*
108 	 * Ask the cpuidle framework to choose a convenient idle state.
109 	 * Fall back to the default arch idle method on errors.
110 	 */
111 	next_state = cpuidle_select(drv, dev);
112 	if (next_state < 0) {
113 use_default:
114 		/*
115 		 * We can't use the cpuidle framework, let's use the default
116 		 * idle routine.
117 		 */
118 		if (current_clr_polling_and_test())
119 			local_irq_enable();
120 		else
121 			arch_cpu_idle();
122 
123 		goto exit_idle;
124 	}
125 
126 
127 	/*
128 	 * The idle task must be scheduled, it is pointless to
129 	 * go to idle, just update no idle residency and get
130 	 * out of this function
131 	 */
132 	if (current_clr_polling_and_test()) {
133 		dev->last_residency = 0;
134 		entered_state = next_state;
135 		local_irq_enable();
136 		goto exit_idle;
137 	}
138 
139 	broadcast = drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP;
140 
141 	/*
142 	 * Tell the time framework to switch to a broadcast timer
143 	 * because our local timer will be shutdown. If a local timer
144 	 * is used from another cpu as a broadcast timer, this call may
145 	 * fail if it is not available
146 	 */
147 	if (broadcast &&
148 	    clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu))
149 		goto use_default;
150 
151 	/* Take note of the planned idle state. */
152 	idle_set_state(this_rq(), &drv->states[next_state]);
153 
154 	/*
155 	 * Enter the idle state previously returned by the governor decision.
156 	 * This function will block until an interrupt occurs and will take
157 	 * care of re-enabling the local interrupts
158 	 */
159 	entered_state = cpuidle_enter(drv, dev, next_state);
160 
161 	/* The cpu is no longer idle or about to enter idle. */
162 	idle_set_state(this_rq(), NULL);
163 
164 	if (broadcast)
165 		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
166 
167 	/*
168 	 * Give the governor an opportunity to reflect on the outcome
169 	 */
170 	cpuidle_reflect(dev, entered_state);
171 
172 exit_idle:
173 	__current_set_polling();
174 
175 	/*
176 	 * It is up to the idle functions to reenable local interrupts
177 	 */
178 	if (WARN_ON_ONCE(irqs_disabled()))
179 		local_irq_enable();
180 
181 	rcu_idle_exit();
182 	start_critical_timings();
183 }
184 
185 /*
186  * Generic idle loop implementation
187  *
188  * Called with polling cleared.
189  */
190 static void cpu_idle_loop(void)
191 {
192 	while (1) {
193 		/*
194 		 * If the arch has a polling bit, we maintain an invariant:
195 		 *
196 		 * Our polling bit is clear if we're not scheduled (i.e. if
197 		 * rq->curr != rq->idle).  This means that, if rq->idle has
198 		 * the polling bit set, then setting need_resched is
199 		 * guaranteed to cause the cpu to reschedule.
200 		 */
201 
202 		__current_set_polling();
203 		tick_nohz_idle_enter();
204 
205 		while (!need_resched()) {
206 			check_pgt_cache();
207 			rmb();
208 
209 			if (cpu_is_offline(smp_processor_id()))
210 				arch_cpu_idle_dead();
211 
212 			local_irq_disable();
213 			arch_cpu_idle_enter();
214 
215 			/*
216 			 * In poll mode we reenable interrupts and spin.
217 			 *
218 			 * Also if we detected in the wakeup from idle
219 			 * path that the tick broadcast device expired
220 			 * for us, we don't want to go deep idle as we
221 			 * know that the IPI is going to arrive right
222 			 * away
223 			 */
224 			if (cpu_idle_force_poll || tick_check_broadcast_expired())
225 				cpu_idle_poll();
226 			else
227 				cpuidle_idle_call();
228 
229 			arch_cpu_idle_exit();
230 		}
231 
232 		/*
233 		 * Since we fell out of the loop above, we know
234 		 * TIF_NEED_RESCHED must be set, propagate it into
235 		 * PREEMPT_NEED_RESCHED.
236 		 *
237 		 * This is required because for polling idle loops we will
238 		 * not have had an IPI to fold the state for us.
239 		 */
240 		preempt_set_need_resched();
241 		tick_nohz_idle_exit();
242 		__current_clr_polling();
243 
244 		/*
245 		 * We promise to call sched_ttwu_pending and reschedule
246 		 * if need_resched is set while polling is set.  That
247 		 * means that clearing polling needs to be visible
248 		 * before doing these things.
249 		 */
250 		smp_mb__after_atomic();
251 
252 		sched_ttwu_pending();
253 		schedule_preempt_disabled();
254 	}
255 }
256 
257 void cpu_startup_entry(enum cpuhp_state state)
258 {
259 	/*
260 	 * This #ifdef needs to die, but it's too late in the cycle to
261 	 * make this generic (arm and sh have never invoked the canary
262 	 * init for the non boot cpus!). Will be fixed in 3.11
263 	 */
264 #ifdef CONFIG_X86
265 	/*
266 	 * If we're the non-boot CPU, nothing set the stack canary up
267 	 * for us. The boot CPU already has it initialized but no harm
268 	 * in doing it again. This is a good place for updating it, as
269 	 * we wont ever return from this function (so the invalid
270 	 * canaries already on the stack wont ever trigger).
271 	 */
272 	boot_init_stack_canary();
273 #endif
274 	arch_cpu_idle_prepare();
275 	cpu_idle_loop();
276 }
277