xref: /openbmc/linux/kernel/sched/idle.c (revision 62e7ca52)
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_relax();
52 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
53 	rcu_idle_exit();
54 	return 1;
55 }
56 
57 /* Weak implementations for optional arch specific functions */
58 void __weak arch_cpu_idle_prepare(void) { }
59 void __weak arch_cpu_idle_enter(void) { }
60 void __weak arch_cpu_idle_exit(void) { }
61 void __weak arch_cpu_idle_dead(void) { }
62 void __weak arch_cpu_idle(void)
63 {
64 	cpu_idle_force_poll = 1;
65 	local_irq_enable();
66 }
67 
68 /**
69  * cpuidle_idle_call - the main idle function
70  *
71  * NOTE: no locks or semaphores should be used here
72  *
73  * On archs that support TIF_POLLING_NRFLAG, is called with polling
74  * set, and it returns with polling set.  If it ever stops polling, it
75  * must clear the polling bit.
76  */
77 static void cpuidle_idle_call(void)
78 {
79 	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
80 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
81 	int next_state, entered_state;
82 	unsigned int broadcast;
83 
84 	/*
85 	 * Check if the idle task must be rescheduled. If it is the
86 	 * case, exit the function after re-enabling the local irq.
87 	 */
88 	if (need_resched()) {
89 		local_irq_enable();
90 		return;
91 	}
92 
93 	/*
94 	 * During the idle period, stop measuring the disabled irqs
95 	 * critical sections latencies
96 	 */
97 	stop_critical_timings();
98 
99 	/*
100 	 * Tell the RCU framework we are entering an idle section,
101 	 * so no more rcu read side critical sections and one more
102 	 * step to the grace period
103 	 */
104 	rcu_idle_enter();
105 
106 	/*
107 	 * Ask the cpuidle framework to choose a convenient idle state.
108 	 * Fall back to the default arch idle method on errors.
109 	 */
110 	next_state = cpuidle_select(drv, dev);
111 	if (next_state < 0) {
112 use_default:
113 		/*
114 		 * We can't use the cpuidle framework, let's use the default
115 		 * idle routine.
116 		 */
117 		if (current_clr_polling_and_test())
118 			local_irq_enable();
119 		else
120 			arch_cpu_idle();
121 
122 		goto exit_idle;
123 	}
124 
125 
126 	/*
127 	 * The idle task must be scheduled, it is pointless to
128 	 * go to idle, just update no idle residency and get
129 	 * out of this function
130 	 */
131 	if (current_clr_polling_and_test()) {
132 		dev->last_residency = 0;
133 		entered_state = next_state;
134 		local_irq_enable();
135 		goto exit_idle;
136 	}
137 
138 	broadcast = drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP;
139 
140 	/*
141 	 * Tell the time framework to switch to a broadcast timer
142 	 * because our local timer will be shutdown. If a local timer
143 	 * is used from another cpu as a broadcast timer, this call may
144 	 * fail if it is not available
145 	 */
146 	if (broadcast &&
147 	    clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu))
148 		goto use_default;
149 
150 	trace_cpu_idle_rcuidle(next_state, dev->cpu);
151 
152 	/*
153 	 * Enter the idle state previously returned by the governor decision.
154 	 * This function will block until an interrupt occurs and will take
155 	 * care of re-enabling the local interrupts
156 	 */
157 	entered_state = cpuidle_enter(drv, dev, next_state);
158 
159 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
160 
161 	if (broadcast)
162 		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
163 
164 	/*
165 	 * Give the governor an opportunity to reflect on the outcome
166 	 */
167 	cpuidle_reflect(dev, entered_state);
168 
169 exit_idle:
170 	__current_set_polling();
171 
172 	/*
173 	 * It is up to the idle functions to reenable local interrupts
174 	 */
175 	if (WARN_ON_ONCE(irqs_disabled()))
176 		local_irq_enable();
177 
178 	rcu_idle_exit();
179 	start_critical_timings();
180 }
181 
182 /*
183  * Generic idle loop implementation
184  *
185  * Called with polling cleared.
186  */
187 static void cpu_idle_loop(void)
188 {
189 	while (1) {
190 		/*
191 		 * If the arch has a polling bit, we maintain an invariant:
192 		 *
193 		 * Our polling bit is clear if we're not scheduled (i.e. if
194 		 * rq->curr != rq->idle).  This means that, if rq->idle has
195 		 * the polling bit set, then setting need_resched is
196 		 * guaranteed to cause the cpu to reschedule.
197 		 */
198 
199 		__current_set_polling();
200 		tick_nohz_idle_enter();
201 
202 		while (!need_resched()) {
203 			check_pgt_cache();
204 			rmb();
205 
206 			if (cpu_is_offline(smp_processor_id()))
207 				arch_cpu_idle_dead();
208 
209 			local_irq_disable();
210 			arch_cpu_idle_enter();
211 
212 			/*
213 			 * In poll mode we reenable interrupts and spin.
214 			 *
215 			 * Also if we detected in the wakeup from idle
216 			 * path that the tick broadcast device expired
217 			 * for us, we don't want to go deep idle as we
218 			 * know that the IPI is going to arrive right
219 			 * away
220 			 */
221 			if (cpu_idle_force_poll || tick_check_broadcast_expired())
222 				cpu_idle_poll();
223 			else
224 				cpuidle_idle_call();
225 
226 			arch_cpu_idle_exit();
227 		}
228 
229 		/*
230 		 * Since we fell out of the loop above, we know
231 		 * TIF_NEED_RESCHED must be set, propagate it into
232 		 * PREEMPT_NEED_RESCHED.
233 		 *
234 		 * This is required because for polling idle loops we will
235 		 * not have had an IPI to fold the state for us.
236 		 */
237 		preempt_set_need_resched();
238 		tick_nohz_idle_exit();
239 		__current_clr_polling();
240 
241 		/*
242 		 * We promise to call sched_ttwu_pending and reschedule
243 		 * if need_resched is set while polling is set.  That
244 		 * means that clearing polling needs to be visible
245 		 * before doing these things.
246 		 */
247 		smp_mb__after_atomic();
248 
249 		sched_ttwu_pending();
250 		schedule_preempt_disabled();
251 	}
252 }
253 
254 void cpu_startup_entry(enum cpuhp_state state)
255 {
256 	/*
257 	 * This #ifdef needs to die, but it's too late in the cycle to
258 	 * make this generic (arm and sh have never invoked the canary
259 	 * init for the non boot cpus!). Will be fixed in 3.11
260 	 */
261 #ifdef CONFIG_X86
262 	/*
263 	 * If we're the non-boot CPU, nothing set the stack canary up
264 	 * for us. The boot CPU already has it initialized but no harm
265 	 * in doing it again. This is a good place for updating it, as
266 	 * we wont ever return from this function (so the invalid
267 	 * canaries already on the stack wont ever trigger).
268 	 */
269 	boot_init_stack_canary();
270 #endif
271 	arch_cpu_idle_prepare();
272 	cpu_idle_loop();
273 }
274