Loading...
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
15static int __read_mostly cpu_idle_force_poll;
16
17void cpu_idle_poll_ctrl(bool enable)
18{
19 if (enable) {
20 cpu_idle_force_poll++;
21 } else {
22 cpu_idle_force_poll--;
23 WARN_ON_ONCE(cpu_idle_force_poll < 0);
24 }
25}
26
27#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
28static int __init cpu_idle_poll_setup(char *__unused)
29{
30 cpu_idle_force_poll = 1;
31 return 1;
32}
33__setup("nohlt", cpu_idle_poll_setup);
34
35static int __init cpu_idle_nopoll_setup(char *__unused)
36{
37 cpu_idle_force_poll = 0;
38 return 1;
39}
40__setup("hlt", cpu_idle_nopoll_setup);
41#endif
42
43static inline int cpu_idle_poll(void)
44{
45 rcu_idle_enter();
46 trace_cpu_idle_rcuidle(0, smp_processor_id());
47 local_irq_enable();
48 while (!tif_need_resched())
49 cpu_relax();
50 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
51 rcu_idle_exit();
52 return 1;
53}
54
55/* Weak implementations for optional arch specific functions */
56void __weak arch_cpu_idle_prepare(void) { }
57void __weak arch_cpu_idle_enter(void) { }
58void __weak arch_cpu_idle_exit(void) { }
59void __weak arch_cpu_idle_dead(void) { }
60void __weak arch_cpu_idle(void)
61{
62 cpu_idle_force_poll = 1;
63 local_irq_enable();
64}
65
66/**
67 * cpuidle_idle_call - the main idle function
68 *
69 * NOTE: no locks or semaphores should be used here
70 * return non-zero on failure
71 */
72static int cpuidle_idle_call(void)
73{
74 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
75 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
76 int next_state, entered_state, ret;
77 bool broadcast;
78
79 /*
80 * Check if the idle task must be rescheduled. If it is the
81 * case, exit the function after re-enabling the local irq and
82 * set again the polling flag
83 */
84 if (current_clr_polling_and_test()) {
85 local_irq_enable();
86 __current_set_polling();
87 return 0;
88 }
89
90 /*
91 * During the idle period, stop measuring the disabled irqs
92 * critical sections latencies
93 */
94 stop_critical_timings();
95
96 /*
97 * Tell the RCU framework we are entering an idle section,
98 * so no more rcu read side critical sections and one more
99 * step to the grace period
100 */
101 rcu_idle_enter();
102
103 /*
104 * Check if the cpuidle framework is ready, otherwise fallback
105 * to the default arch specific idle method
106 */
107 ret = cpuidle_enabled(drv, dev);
108
109 if (!ret) {
110 /*
111 * Ask the governor to choose an idle state it thinks
112 * it is convenient to go to. There is *always* a
113 * convenient idle state
114 */
115 next_state = cpuidle_select(drv, dev);
116
117 /*
118 * The idle task must be scheduled, it is pointless to
119 * go to idle, just update no idle residency and get
120 * out of this function
121 */
122 if (current_clr_polling_and_test()) {
123 dev->last_residency = 0;
124 entered_state = next_state;
125 local_irq_enable();
126 } else {
127 broadcast = !!(drv->states[next_state].flags &
128 CPUIDLE_FLAG_TIMER_STOP);
129
130 if (broadcast)
131 /*
132 * Tell the time framework to switch
133 * to a broadcast timer because our
134 * local timer will be shutdown. If a
135 * local timer is used from another
136 * cpu as a broadcast timer, this call
137 * may fail if it is not available
138 */
139 ret = clockevents_notify(
140 CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
141 &dev->cpu);
142
143 if (!ret) {
144 trace_cpu_idle_rcuidle(next_state, dev->cpu);
145
146 /*
147 * Enter the idle state previously
148 * returned by the governor
149 * decision. This function will block
150 * until an interrupt occurs and will
151 * take care of re-enabling the local
152 * interrupts
153 */
154 entered_state = cpuidle_enter(drv, dev,
155 next_state);
156
157 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT,
158 dev->cpu);
159
160 if (broadcast)
161 clockevents_notify(
162 CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
163 &dev->cpu);
164
165 /*
166 * Give the governor an opportunity to reflect on the
167 * outcome
168 */
169 cpuidle_reflect(dev, entered_state);
170 }
171 }
172 }
173
174 /*
175 * We can't use the cpuidle framework, let's use the default
176 * idle routine
177 */
178 if (ret)
179 arch_cpu_idle();
180
181 __current_set_polling();
182
183 /*
184 * It is up to the idle functions to enable back the local
185 * interrupt
186 */
187 if (WARN_ON_ONCE(irqs_disabled()))
188 local_irq_enable();
189
190 rcu_idle_exit();
191 start_critical_timings();
192
193 return 0;
194}
195
196/*
197 * Generic idle loop implementation
198 */
199static void cpu_idle_loop(void)
200{
201 while (1) {
202 tick_nohz_idle_enter();
203
204 while (!need_resched()) {
205 check_pgt_cache();
206 rmb();
207
208 if (cpu_is_offline(smp_processor_id()))
209 arch_cpu_idle_dead();
210
211 local_irq_disable();
212 arch_cpu_idle_enter();
213
214 /*
215 * In poll mode we reenable interrupts and spin.
216 *
217 * Also if we detected in the wakeup from idle
218 * path that the tick broadcast device expired
219 * for us, we don't want to go deep idle as we
220 * know that the IPI is going to arrive right
221 * away
222 */
223 if (cpu_idle_force_poll || tick_check_broadcast_expired())
224 cpu_idle_poll();
225 else
226 cpuidle_idle_call();
227
228 arch_cpu_idle_exit();
229 }
230
231 /*
232 * Since we fell out of the loop above, we know
233 * TIF_NEED_RESCHED must be set, propagate it into
234 * PREEMPT_NEED_RESCHED.
235 *
236 * This is required because for polling idle loops we will
237 * not have had an IPI to fold the state for us.
238 */
239 preempt_set_need_resched();
240 tick_nohz_idle_exit();
241 schedule_preempt_disabled();
242 }
243}
244
245void cpu_startup_entry(enum cpuhp_state state)
246{
247 /*
248 * This #ifdef needs to die, but it's too late in the cycle to
249 * make this generic (arm and sh have never invoked the canary
250 * init for the non boot cpus!). Will be fixed in 3.11
251 */
252#ifdef CONFIG_X86
253 /*
254 * If we're the non-boot CPU, nothing set the stack canary up
255 * for us. The boot CPU already has it initialized but no harm
256 * in doing it again. This is a good place for updating it, as
257 * we wont ever return from this function (so the invalid
258 * canaries already on the stack wont ever trigger).
259 */
260 boot_init_stack_canary();
261#endif
262 __current_set_polling();
263 arch_cpu_idle_prepare();
264 cpu_idle_loop();
265}
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/cpuhotplug.h>
8#include <linux/tick.h>
9#include <linux/mm.h>
10#include <linux/stackprotector.h>
11#include <linux/suspend.h>
12
13#include <asm/tlb.h>
14
15#include <trace/events/power.h>
16
17#include "sched.h"
18
19/**
20 * sched_idle_set_state - Record idle state for the current CPU.
21 * @idle_state: State to record.
22 */
23void sched_idle_set_state(struct cpuidle_state *idle_state)
24{
25 idle_set_state(this_rq(), idle_state);
26}
27
28static int __read_mostly cpu_idle_force_poll;
29
30void cpu_idle_poll_ctrl(bool enable)
31{
32 if (enable) {
33 cpu_idle_force_poll++;
34 } else {
35 cpu_idle_force_poll--;
36 WARN_ON_ONCE(cpu_idle_force_poll < 0);
37 }
38}
39
40#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
41static int __init cpu_idle_poll_setup(char *__unused)
42{
43 cpu_idle_force_poll = 1;
44 return 1;
45}
46__setup("nohlt", cpu_idle_poll_setup);
47
48static int __init cpu_idle_nopoll_setup(char *__unused)
49{
50 cpu_idle_force_poll = 0;
51 return 1;
52}
53__setup("hlt", cpu_idle_nopoll_setup);
54#endif
55
56static inline int cpu_idle_poll(void)
57{
58 rcu_idle_enter();
59 trace_cpu_idle_rcuidle(0, smp_processor_id());
60 local_irq_enable();
61 stop_critical_timings();
62 while (!tif_need_resched() &&
63 (cpu_idle_force_poll || tick_check_broadcast_expired()))
64 cpu_relax();
65 start_critical_timings();
66 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
67 rcu_idle_exit();
68 return 1;
69}
70
71/* Weak implementations for optional arch specific functions */
72void __weak arch_cpu_idle_prepare(void) { }
73void __weak arch_cpu_idle_enter(void) { }
74void __weak arch_cpu_idle_exit(void) { }
75void __weak arch_cpu_idle_dead(void) { }
76void __weak arch_cpu_idle(void)
77{
78 cpu_idle_force_poll = 1;
79 local_irq_enable();
80}
81
82/**
83 * default_idle_call - Default CPU idle routine.
84 *
85 * To use when the cpuidle framework cannot be used.
86 */
87void default_idle_call(void)
88{
89 if (current_clr_polling_and_test()) {
90 local_irq_enable();
91 } else {
92 stop_critical_timings();
93 arch_cpu_idle();
94 start_critical_timings();
95 }
96}
97
98static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
99 int next_state)
100{
101 /*
102 * The idle task must be scheduled, it is pointless to go to idle, just
103 * update no idle residency and return.
104 */
105 if (current_clr_polling_and_test()) {
106 dev->last_residency = 0;
107 local_irq_enable();
108 return -EBUSY;
109 }
110
111 /*
112 * Enter the idle state previously returned by the governor decision.
113 * This function will block until an interrupt occurs and will take
114 * care of re-enabling the local interrupts
115 */
116 return cpuidle_enter(drv, dev, next_state);
117}
118
119/**
120 * cpuidle_idle_call - the main idle function
121 *
122 * NOTE: no locks or semaphores should be used here
123 *
124 * On archs that support TIF_POLLING_NRFLAG, is called with polling
125 * set, and it returns with polling set. If it ever stops polling, it
126 * must clear the polling bit.
127 */
128static void cpuidle_idle_call(void)
129{
130 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
131 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
132 int next_state, entered_state;
133
134 /*
135 * Check if the idle task must be rescheduled. If it is the
136 * case, exit the function after re-enabling the local irq.
137 */
138 if (need_resched()) {
139 local_irq_enable();
140 return;
141 }
142
143 /*
144 * Tell the RCU framework we are entering an idle section,
145 * so no more rcu read side critical sections and one more
146 * step to the grace period
147 */
148 rcu_idle_enter();
149
150 if (cpuidle_not_available(drv, dev)) {
151 default_idle_call();
152 goto exit_idle;
153 }
154
155 /*
156 * Suspend-to-idle ("freeze") is a system state in which all user space
157 * has been frozen, all I/O devices have been suspended and the only
158 * activity happens here and in iterrupts (if any). In that case bypass
159 * the cpuidle governor and go stratight for the deepest idle state
160 * available. Possibly also suspend the local tick and the entire
161 * timekeeping to prevent timer interrupts from kicking us out of idle
162 * until a proper wakeup interrupt happens.
163 */
164 if (idle_should_freeze()) {
165 entered_state = cpuidle_enter_freeze(drv, dev);
166 if (entered_state > 0) {
167 local_irq_enable();
168 goto exit_idle;
169 }
170
171 next_state = cpuidle_find_deepest_state(drv, dev);
172 call_cpuidle(drv, dev, next_state);
173 } else {
174 /*
175 * Ask the cpuidle framework to choose a convenient idle state.
176 */
177 next_state = cpuidle_select(drv, dev);
178 entered_state = call_cpuidle(drv, dev, next_state);
179 /*
180 * Give the governor an opportunity to reflect on the outcome
181 */
182 cpuidle_reflect(dev, entered_state);
183 }
184
185exit_idle:
186 __current_set_polling();
187
188 /*
189 * It is up to the idle functions to reenable local interrupts
190 */
191 if (WARN_ON_ONCE(irqs_disabled()))
192 local_irq_enable();
193
194 rcu_idle_exit();
195}
196
197/*
198 * Generic idle loop implementation
199 *
200 * Called with polling cleared.
201 */
202static void cpu_idle_loop(void)
203{
204 while (1) {
205 /*
206 * If the arch has a polling bit, we maintain an invariant:
207 *
208 * Our polling bit is clear if we're not scheduled (i.e. if
209 * rq->curr != rq->idle). This means that, if rq->idle has
210 * the polling bit set, then setting need_resched is
211 * guaranteed to cause the cpu to reschedule.
212 */
213
214 __current_set_polling();
215 quiet_vmstat();
216 tick_nohz_idle_enter();
217
218 while (!need_resched()) {
219 check_pgt_cache();
220 rmb();
221
222 if (cpu_is_offline(smp_processor_id())) {
223 cpuhp_report_idle_dead();
224 arch_cpu_idle_dead();
225 }
226
227 local_irq_disable();
228 arch_cpu_idle_enter();
229
230 /*
231 * In poll mode we reenable interrupts and spin.
232 *
233 * Also if we detected in the wakeup from idle
234 * path that the tick broadcast device expired
235 * for us, we don't want to go deep idle as we
236 * know that the IPI is going to arrive right
237 * away
238 */
239 if (cpu_idle_force_poll || tick_check_broadcast_expired())
240 cpu_idle_poll();
241 else
242 cpuidle_idle_call();
243
244 arch_cpu_idle_exit();
245 }
246
247 /*
248 * Since we fell out of the loop above, we know
249 * TIF_NEED_RESCHED must be set, propagate it into
250 * PREEMPT_NEED_RESCHED.
251 *
252 * This is required because for polling idle loops we will
253 * not have had an IPI to fold the state for us.
254 */
255 preempt_set_need_resched();
256 tick_nohz_idle_exit();
257 __current_clr_polling();
258
259 /*
260 * We promise to call sched_ttwu_pending and reschedule
261 * if need_resched is set while polling is set. That
262 * means that clearing polling needs to be visible
263 * before doing these things.
264 */
265 smp_mb__after_atomic();
266
267 sched_ttwu_pending();
268 schedule_preempt_disabled();
269 }
270}
271
272void cpu_startup_entry(enum cpuhp_state state)
273{
274 /*
275 * This #ifdef needs to die, but it's too late in the cycle to
276 * make this generic (arm and sh have never invoked the canary
277 * init for the non boot cpus!). Will be fixed in 3.11
278 */
279#ifdef CONFIG_X86
280 /*
281 * If we're the non-boot CPU, nothing set the stack canary up
282 * for us. The boot CPU already has it initialized but no harm
283 * in doing it again. This is a good place for updating it, as
284 * we wont ever return from this function (so the invalid
285 * canaries already on the stack wont ever trigger).
286 */
287 boot_init_stack_canary();
288#endif
289 arch_cpu_idle_prepare();
290 cpuhp_online_idle(state);
291 cpu_idle_loop();
292}