xref: /openbmc/linux/kernel/sched/idle.c (revision d2c43ff1)
1 /*
2  * Generic entry point for the idle threads
3  */
4 #include <linux/sched.h>
5 #include <linux/sched/idle.h>
6 #include <linux/cpu.h>
7 #include <linux/cpuidle.h>
8 #include <linux/cpuhotplug.h>
9 #include <linux/tick.h>
10 #include <linux/mm.h>
11 #include <linux/stackprotector.h>
12 #include <linux/suspend.h>
13 #include <linux/livepatch.h>
14 
15 #include <asm/tlb.h>
16 
17 #include <trace/events/power.h>
18 
19 #include "sched.h"
20 
21 /* Linker adds these: start and end of __cpuidle functions */
22 extern char __cpuidle_text_start[], __cpuidle_text_end[];
23 
24 /**
25  * sched_idle_set_state - Record idle state for the current CPU.
26  * @idle_state: State to record.
27  */
28 void sched_idle_set_state(struct cpuidle_state *idle_state)
29 {
30 	idle_set_state(this_rq(), idle_state);
31 }
32 
33 static int __read_mostly cpu_idle_force_poll;
34 
35 void cpu_idle_poll_ctrl(bool enable)
36 {
37 	if (enable) {
38 		cpu_idle_force_poll++;
39 	} else {
40 		cpu_idle_force_poll--;
41 		WARN_ON_ONCE(cpu_idle_force_poll < 0);
42 	}
43 }
44 
45 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
46 static int __init cpu_idle_poll_setup(char *__unused)
47 {
48 	cpu_idle_force_poll = 1;
49 	return 1;
50 }
51 __setup("nohlt", cpu_idle_poll_setup);
52 
53 static int __init cpu_idle_nopoll_setup(char *__unused)
54 {
55 	cpu_idle_force_poll = 0;
56 	return 1;
57 }
58 __setup("hlt", cpu_idle_nopoll_setup);
59 #endif
60 
61 static noinline int __cpuidle cpu_idle_poll(void)
62 {
63 	rcu_idle_enter();
64 	trace_cpu_idle_rcuidle(0, smp_processor_id());
65 	local_irq_enable();
66 	stop_critical_timings();
67 	while (!tif_need_resched() &&
68 		(cpu_idle_force_poll || tick_check_broadcast_expired()))
69 		cpu_relax();
70 	start_critical_timings();
71 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
72 	rcu_idle_exit();
73 	return 1;
74 }
75 
76 /* Weak implementations for optional arch specific functions */
77 void __weak arch_cpu_idle_prepare(void) { }
78 void __weak arch_cpu_idle_enter(void) { }
79 void __weak arch_cpu_idle_exit(void) { }
80 void __weak arch_cpu_idle_dead(void) { }
81 void __weak arch_cpu_idle(void)
82 {
83 	cpu_idle_force_poll = 1;
84 	local_irq_enable();
85 }
86 
87 /**
88  * default_idle_call - Default CPU idle routine.
89  *
90  * To use when the cpuidle framework cannot be used.
91  */
92 void __cpuidle default_idle_call(void)
93 {
94 	if (current_clr_polling_and_test()) {
95 		local_irq_enable();
96 	} else {
97 		stop_critical_timings();
98 		arch_cpu_idle();
99 		start_critical_timings();
100 	}
101 }
102 
103 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
104 		      int next_state)
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 = cpuidle_get_device();
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 
170 	if (idle_should_freeze() || dev->use_deepest_state) {
171 		if (idle_should_freeze()) {
172 			entered_state = cpuidle_enter_freeze(drv, dev);
173 			if (entered_state > 0) {
174 				local_irq_enable();
175 				goto exit_idle;
176 			}
177 		}
178 
179 		next_state = cpuidle_find_deepest_state(drv, dev);
180 		call_cpuidle(drv, dev, next_state);
181 	} else {
182 		/*
183 		 * Ask the cpuidle framework to choose a convenient idle state.
184 		 */
185 		next_state = cpuidle_select(drv, dev);
186 		entered_state = call_cpuidle(drv, dev, next_state);
187 		/*
188 		 * Give the governor an opportunity to reflect on the outcome
189 		 */
190 		cpuidle_reflect(dev, entered_state);
191 	}
192 
193 exit_idle:
194 	__current_set_polling();
195 
196 	/*
197 	 * It is up to the idle functions to reenable local interrupts
198 	 */
199 	if (WARN_ON_ONCE(irqs_disabled()))
200 		local_irq_enable();
201 
202 	rcu_idle_exit();
203 }
204 
205 /*
206  * Generic idle loop implementation
207  *
208  * Called with polling cleared.
209  */
210 static void do_idle(void)
211 {
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 rq->curr !=
216 	 * rq->idle). This means that, if rq->idle has the polling bit set,
217 	 * then setting need_resched is guaranteed to cause the CPU to
218 	 * reschedule.
219 	 */
220 
221 	__current_set_polling();
222 	quiet_vmstat();
223 	tick_nohz_idle_enter();
224 
225 	while (!need_resched()) {
226 		check_pgt_cache();
227 		rmb();
228 
229 		if (cpu_is_offline(smp_processor_id())) {
230 			cpuhp_report_idle_dead();
231 			arch_cpu_idle_dead();
232 		}
233 
234 		local_irq_disable();
235 		arch_cpu_idle_enter();
236 
237 		/*
238 		 * In poll mode we reenable interrupts and spin. Also if we
239 		 * detected in the wakeup from idle path that the tick
240 		 * broadcast device expired for us, we don't want to go deep
241 		 * idle as we know that the IPI is going to arrive right away.
242 		 */
243 		if (cpu_idle_force_poll || tick_check_broadcast_expired())
244 			cpu_idle_poll();
245 		else
246 			cpuidle_idle_call();
247 		arch_cpu_idle_exit();
248 	}
249 
250 	/*
251 	 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
252 	 * be set, propagate it into PREEMPT_NEED_RESCHED.
253 	 *
254 	 * This is required because for polling idle loops we will not have had
255 	 * an IPI to fold the state for us.
256 	 */
257 	preempt_set_need_resched();
258 	tick_nohz_idle_exit();
259 	__current_clr_polling();
260 
261 	/*
262 	 * We promise to call sched_ttwu_pending() and reschedule if
263 	 * need_resched() is set while polling is set. That means that clearing
264 	 * polling needs to be visible before doing these things.
265 	 */
266 	smp_mb__after_atomic();
267 
268 	sched_ttwu_pending();
269 	schedule_idle();
270 
271 	if (unlikely(klp_patch_pending(current)))
272 		klp_update_patch_state(current);
273 }
274 
275 bool cpu_in_idle(unsigned long pc)
276 {
277 	return pc >= (unsigned long)__cpuidle_text_start &&
278 		pc < (unsigned long)__cpuidle_text_end;
279 }
280 
281 struct idle_timer {
282 	struct hrtimer timer;
283 	int done;
284 };
285 
286 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
287 {
288 	struct idle_timer *it = container_of(timer, struct idle_timer, timer);
289 
290 	WRITE_ONCE(it->done, 1);
291 	set_tsk_need_resched(current);
292 
293 	return HRTIMER_NORESTART;
294 }
295 
296 void play_idle(unsigned long duration_ms)
297 {
298 	struct idle_timer it;
299 
300 	/*
301 	 * Only FIFO tasks can disable the tick since they don't need the forced
302 	 * preemption.
303 	 */
304 	WARN_ON_ONCE(current->policy != SCHED_FIFO);
305 	WARN_ON_ONCE(current->nr_cpus_allowed != 1);
306 	WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
307 	WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
308 	WARN_ON_ONCE(!duration_ms);
309 
310 	rcu_sleep_check();
311 	preempt_disable();
312 	current->flags |= PF_IDLE;
313 	cpuidle_use_deepest_state(true);
314 
315 	it.done = 0;
316 	hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
317 	it.timer.function = idle_inject_timer_fn;
318 	hrtimer_start(&it.timer, ms_to_ktime(duration_ms), HRTIMER_MODE_REL_PINNED);
319 
320 	while (!READ_ONCE(it.done))
321 		do_idle();
322 
323 	cpuidle_use_deepest_state(false);
324 	current->flags &= ~PF_IDLE;
325 
326 	preempt_fold_need_resched();
327 	preempt_enable();
328 }
329 EXPORT_SYMBOL_GPL(play_idle);
330 
331 void cpu_startup_entry(enum cpuhp_state state)
332 {
333 	/*
334 	 * This #ifdef needs to die, but it's too late in the cycle to
335 	 * make this generic (arm and sh have never invoked the canary
336 	 * init for the non boot cpus!). Will be fixed in 3.11
337 	 */
338 #ifdef CONFIG_X86
339 	/*
340 	 * If we're the non-boot CPU, nothing set the stack canary up
341 	 * for us. The boot CPU already has it initialized but no harm
342 	 * in doing it again. This is a good place for updating it, as
343 	 * we wont ever return from this function (so the invalid
344 	 * canaries already on the stack wont ever trigger).
345 	 */
346 	boot_init_stack_canary();
347 #endif
348 	arch_cpu_idle_prepare();
349 	cpuhp_online_idle(state);
350 	while (1)
351 		do_idle();
352 }
353