Loading...
1/*
2 * cpuidle.c - core cpuidle infrastructure
3 *
4 * (C) 2006-2007 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5 * Shaohua Li <shaohua.li@intel.com>
6 * Adam Belay <abelay@novell.com>
7 *
8 * This code is licenced under the GPL.
9 */
10
11#include <linux/kernel.h>
12#include <linux/mutex.h>
13#include <linux/sched.h>
14#include <linux/notifier.h>
15#include <linux/pm_qos_params.h>
16#include <linux/cpu.h>
17#include <linux/cpuidle.h>
18#include <linux/ktime.h>
19#include <linux/hrtimer.h>
20#include <trace/events/power.h>
21
22#include "cpuidle.h"
23
24DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
25
26DEFINE_MUTEX(cpuidle_lock);
27LIST_HEAD(cpuidle_detected_devices);
28
29static int enabled_devices;
30static int off __read_mostly;
31static int initialized __read_mostly;
32
33int cpuidle_disabled(void)
34{
35 return off;
36}
37void disable_cpuidle(void)
38{
39 off = 1;
40}
41
42#if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
43static void cpuidle_kick_cpus(void)
44{
45 cpu_idle_wait();
46}
47#elif defined(CONFIG_SMP)
48# error "Arch needs cpu_idle_wait() equivalent here"
49#else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
50static void cpuidle_kick_cpus(void) {}
51#endif
52
53static int __cpuidle_register_device(struct cpuidle_device *dev);
54
55/**
56 * cpuidle_idle_call - the main idle loop
57 *
58 * NOTE: no locks or semaphores should be used here
59 * return non-zero on failure
60 */
61int cpuidle_idle_call(void)
62{
63 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
64 struct cpuidle_state *target_state;
65 int next_state;
66
67 if (off)
68 return -ENODEV;
69
70 if (!initialized)
71 return -ENODEV;
72
73 /* check if the device is ready */
74 if (!dev || !dev->enabled)
75 return -EBUSY;
76
77#if 0
78 /* shows regressions, re-enable for 2.6.29 */
79 /*
80 * run any timers that can be run now, at this point
81 * before calculating the idle duration etc.
82 */
83 hrtimer_peek_ahead_timers();
84#endif
85
86 /*
87 * Call the device's prepare function before calling the
88 * governor's select function. ->prepare gives the device's
89 * cpuidle driver a chance to update any dynamic information
90 * of its cpuidle states for the current idle period, e.g.
91 * state availability, latencies, residencies, etc.
92 */
93 if (dev->prepare)
94 dev->prepare(dev);
95
96 /* ask the governor for the next state */
97 next_state = cpuidle_curr_governor->select(dev);
98 if (need_resched()) {
99 local_irq_enable();
100 return 0;
101 }
102
103 target_state = &dev->states[next_state];
104
105 /* enter the state and update stats */
106 dev->last_state = target_state;
107
108 trace_power_start(POWER_CSTATE, next_state, dev->cpu);
109 trace_cpu_idle(next_state, dev->cpu);
110
111 dev->last_residency = target_state->enter(dev, target_state);
112
113 trace_power_end(dev->cpu);
114 trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
115
116 if (dev->last_state)
117 target_state = dev->last_state;
118
119 target_state->time += (unsigned long long)dev->last_residency;
120 target_state->usage++;
121
122 /* give the governor an opportunity to reflect on the outcome */
123 if (cpuidle_curr_governor->reflect)
124 cpuidle_curr_governor->reflect(dev);
125
126 return 0;
127}
128
129/**
130 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
131 */
132void cpuidle_install_idle_handler(void)
133{
134 if (enabled_devices) {
135 /* Make sure all changes finished before we switch to new idle */
136 smp_wmb();
137 initialized = 1;
138 }
139}
140
141/**
142 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
143 */
144void cpuidle_uninstall_idle_handler(void)
145{
146 if (enabled_devices) {
147 initialized = 0;
148 cpuidle_kick_cpus();
149 }
150}
151
152/**
153 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
154 */
155void cpuidle_pause_and_lock(void)
156{
157 mutex_lock(&cpuidle_lock);
158 cpuidle_uninstall_idle_handler();
159}
160
161EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
162
163/**
164 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
165 */
166void cpuidle_resume_and_unlock(void)
167{
168 cpuidle_install_idle_handler();
169 mutex_unlock(&cpuidle_lock);
170}
171
172EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
173
174#ifdef CONFIG_ARCH_HAS_CPU_RELAX
175static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
176{
177 ktime_t t1, t2;
178 s64 diff;
179 int ret;
180
181 t1 = ktime_get();
182 local_irq_enable();
183 while (!need_resched())
184 cpu_relax();
185
186 t2 = ktime_get();
187 diff = ktime_to_us(ktime_sub(t2, t1));
188 if (diff > INT_MAX)
189 diff = INT_MAX;
190
191 ret = (int) diff;
192 return ret;
193}
194
195static void poll_idle_init(struct cpuidle_device *dev)
196{
197 struct cpuidle_state *state = &dev->states[0];
198
199 cpuidle_set_statedata(state, NULL);
200
201 snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
202 snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
203 state->exit_latency = 0;
204 state->target_residency = 0;
205 state->power_usage = -1;
206 state->flags = 0;
207 state->enter = poll_idle;
208}
209#else
210static void poll_idle_init(struct cpuidle_device *dev) {}
211#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
212
213/**
214 * cpuidle_enable_device - enables idle PM for a CPU
215 * @dev: the CPU
216 *
217 * This function must be called between cpuidle_pause_and_lock and
218 * cpuidle_resume_and_unlock when used externally.
219 */
220int cpuidle_enable_device(struct cpuidle_device *dev)
221{
222 int ret, i;
223
224 if (dev->enabled)
225 return 0;
226 if (!cpuidle_get_driver() || !cpuidle_curr_governor)
227 return -EIO;
228 if (!dev->state_count)
229 return -EINVAL;
230
231 if (dev->registered == 0) {
232 ret = __cpuidle_register_device(dev);
233 if (ret)
234 return ret;
235 }
236
237 poll_idle_init(dev);
238
239 if ((ret = cpuidle_add_state_sysfs(dev)))
240 return ret;
241
242 if (cpuidle_curr_governor->enable &&
243 (ret = cpuidle_curr_governor->enable(dev)))
244 goto fail_sysfs;
245
246 for (i = 0; i < dev->state_count; i++) {
247 dev->states[i].usage = 0;
248 dev->states[i].time = 0;
249 }
250 dev->last_residency = 0;
251 dev->last_state = NULL;
252
253 smp_wmb();
254
255 dev->enabled = 1;
256
257 enabled_devices++;
258 return 0;
259
260fail_sysfs:
261 cpuidle_remove_state_sysfs(dev);
262
263 return ret;
264}
265
266EXPORT_SYMBOL_GPL(cpuidle_enable_device);
267
268/**
269 * cpuidle_disable_device - disables idle PM for a CPU
270 * @dev: the CPU
271 *
272 * This function must be called between cpuidle_pause_and_lock and
273 * cpuidle_resume_and_unlock when used externally.
274 */
275void cpuidle_disable_device(struct cpuidle_device *dev)
276{
277 if (!dev->enabled)
278 return;
279 if (!cpuidle_get_driver() || !cpuidle_curr_governor)
280 return;
281
282 dev->enabled = 0;
283
284 if (cpuidle_curr_governor->disable)
285 cpuidle_curr_governor->disable(dev);
286
287 cpuidle_remove_state_sysfs(dev);
288 enabled_devices--;
289}
290
291EXPORT_SYMBOL_GPL(cpuidle_disable_device);
292
293/**
294 * __cpuidle_register_device - internal register function called before register
295 * and enable routines
296 * @dev: the cpu
297 *
298 * cpuidle_lock mutex must be held before this is called
299 */
300static int __cpuidle_register_device(struct cpuidle_device *dev)
301{
302 int ret;
303 struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
304 struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
305
306 if (!sys_dev)
307 return -EINVAL;
308 if (!try_module_get(cpuidle_driver->owner))
309 return -EINVAL;
310
311 init_completion(&dev->kobj_unregister);
312
313 /*
314 * cpuidle driver should set the dev->power_specified bit
315 * before registering the device if the driver provides
316 * power_usage numbers.
317 *
318 * For those devices whose ->power_specified is not set,
319 * we fill in power_usage with decreasing values as the
320 * cpuidle code has an implicit assumption that state Cn
321 * uses less power than C(n-1).
322 *
323 * With CONFIG_ARCH_HAS_CPU_RELAX, C0 is already assigned
324 * an power value of -1. So we use -2, -3, etc, for other
325 * c-states.
326 */
327 if (!dev->power_specified) {
328 int i;
329 for (i = CPUIDLE_DRIVER_STATE_START; i < dev->state_count; i++)
330 dev->states[i].power_usage = -1 - i;
331 }
332
333 per_cpu(cpuidle_devices, dev->cpu) = dev;
334 list_add(&dev->device_list, &cpuidle_detected_devices);
335 if ((ret = cpuidle_add_sysfs(sys_dev))) {
336 module_put(cpuidle_driver->owner);
337 return ret;
338 }
339
340 dev->registered = 1;
341 return 0;
342}
343
344/**
345 * cpuidle_register_device - registers a CPU's idle PM feature
346 * @dev: the cpu
347 */
348int cpuidle_register_device(struct cpuidle_device *dev)
349{
350 int ret;
351
352 mutex_lock(&cpuidle_lock);
353
354 if ((ret = __cpuidle_register_device(dev))) {
355 mutex_unlock(&cpuidle_lock);
356 return ret;
357 }
358
359 cpuidle_enable_device(dev);
360 cpuidle_install_idle_handler();
361
362 mutex_unlock(&cpuidle_lock);
363
364 return 0;
365
366}
367
368EXPORT_SYMBOL_GPL(cpuidle_register_device);
369
370/**
371 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
372 * @dev: the cpu
373 */
374void cpuidle_unregister_device(struct cpuidle_device *dev)
375{
376 struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
377 struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
378
379 if (dev->registered == 0)
380 return;
381
382 cpuidle_pause_and_lock();
383
384 cpuidle_disable_device(dev);
385
386 cpuidle_remove_sysfs(sys_dev);
387 list_del(&dev->device_list);
388 wait_for_completion(&dev->kobj_unregister);
389 per_cpu(cpuidle_devices, dev->cpu) = NULL;
390
391 cpuidle_resume_and_unlock();
392
393 module_put(cpuidle_driver->owner);
394}
395
396EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
397
398#ifdef CONFIG_SMP
399
400static void smp_callback(void *v)
401{
402 /* we already woke the CPU up, nothing more to do */
403}
404
405/*
406 * This function gets called when a part of the kernel has a new latency
407 * requirement. This means we need to get all processors out of their C-state,
408 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
409 * wakes them all right up.
410 */
411static int cpuidle_latency_notify(struct notifier_block *b,
412 unsigned long l, void *v)
413{
414 smp_call_function(smp_callback, NULL, 1);
415 return NOTIFY_OK;
416}
417
418static struct notifier_block cpuidle_latency_notifier = {
419 .notifier_call = cpuidle_latency_notify,
420};
421
422static inline void latency_notifier_init(struct notifier_block *n)
423{
424 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
425}
426
427#else /* CONFIG_SMP */
428
429#define latency_notifier_init(x) do { } while (0)
430
431#endif /* CONFIG_SMP */
432
433/**
434 * cpuidle_init - core initializer
435 */
436static int __init cpuidle_init(void)
437{
438 int ret;
439
440 if (cpuidle_disabled())
441 return -ENODEV;
442
443 ret = cpuidle_add_class_sysfs(&cpu_sysdev_class);
444 if (ret)
445 return ret;
446
447 latency_notifier_init(&cpuidle_latency_notifier);
448
449 return 0;
450}
451
452module_param(off, int, 0444);
453core_initcall(cpuidle_init);
1/*
2 * cpuidle.c - core cpuidle infrastructure
3 *
4 * (C) 2006-2007 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5 * Shaohua Li <shaohua.li@intel.com>
6 * Adam Belay <abelay@novell.com>
7 *
8 * This code is licenced under the GPL.
9 */
10
11#include <linux/clockchips.h>
12#include <linux/kernel.h>
13#include <linux/mutex.h>
14#include <linux/sched.h>
15#include <linux/sched/clock.h>
16#include <linux/notifier.h>
17#include <linux/pm_qos.h>
18#include <linux/cpu.h>
19#include <linux/cpuidle.h>
20#include <linux/ktime.h>
21#include <linux/hrtimer.h>
22#include <linux/module.h>
23#include <linux/suspend.h>
24#include <linux/tick.h>
25#include <linux/mmu_context.h>
26#include <trace/events/power.h>
27
28#include "cpuidle.h"
29
30DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
31DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
32
33DEFINE_MUTEX(cpuidle_lock);
34LIST_HEAD(cpuidle_detected_devices);
35
36static int enabled_devices;
37static int off __read_mostly;
38static int initialized __read_mostly;
39
40int cpuidle_disabled(void)
41{
42 return off;
43}
44void disable_cpuidle(void)
45{
46 off = 1;
47}
48
49bool cpuidle_not_available(struct cpuidle_driver *drv,
50 struct cpuidle_device *dev)
51{
52 return off || !initialized || !drv || !dev || !dev->enabled;
53}
54
55/**
56 * cpuidle_play_dead - cpu off-lining
57 *
58 * Returns in case of an error or no driver
59 */
60int cpuidle_play_dead(void)
61{
62 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
63 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
64 int i;
65
66 if (!drv)
67 return -ENODEV;
68
69 /* Find lowest-power state that supports long-term idle */
70 for (i = drv->state_count - 1; i >= 0; i--)
71 if (drv->states[i].enter_dead)
72 return drv->states[i].enter_dead(dev, i);
73
74 return -ENODEV;
75}
76
77static int find_deepest_state(struct cpuidle_driver *drv,
78 struct cpuidle_device *dev,
79 u64 max_latency_ns,
80 unsigned int forbidden_flags,
81 bool s2idle)
82{
83 u64 latency_req = 0;
84 int i, ret = 0;
85
86 for (i = 1; i < drv->state_count; i++) {
87 struct cpuidle_state *s = &drv->states[i];
88
89 if (dev->states_usage[i].disable ||
90 s->exit_latency_ns <= latency_req ||
91 s->exit_latency_ns > max_latency_ns ||
92 (s->flags & forbidden_flags) ||
93 (s2idle && !s->enter_s2idle))
94 continue;
95
96 latency_req = s->exit_latency_ns;
97 ret = i;
98 }
99 return ret;
100}
101
102/**
103 * cpuidle_use_deepest_state - Set/unset governor override mode.
104 * @latency_limit_ns: Idle state exit latency limit (or no override if 0).
105 *
106 * If @latency_limit_ns is nonzero, set the current CPU to use the deepest idle
107 * state with exit latency within @latency_limit_ns (override governors going
108 * forward), or do not override governors if it is zero.
109 */
110void cpuidle_use_deepest_state(u64 latency_limit_ns)
111{
112 struct cpuidle_device *dev;
113
114 preempt_disable();
115 dev = cpuidle_get_device();
116 if (dev)
117 dev->forced_idle_latency_limit_ns = latency_limit_ns;
118 preempt_enable();
119}
120
121/**
122 * cpuidle_find_deepest_state - Find the deepest available idle state.
123 * @drv: cpuidle driver for the given CPU.
124 * @dev: cpuidle device for the given CPU.
125 * @latency_limit_ns: Idle state exit latency limit
126 *
127 * Return: the index of the deepest available idle state.
128 */
129int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
130 struct cpuidle_device *dev,
131 u64 latency_limit_ns)
132{
133 return find_deepest_state(drv, dev, latency_limit_ns, 0, false);
134}
135
136#ifdef CONFIG_SUSPEND
137static void enter_s2idle_proper(struct cpuidle_driver *drv,
138 struct cpuidle_device *dev, int index)
139{
140 ktime_t time_start, time_end;
141 struct cpuidle_state *target_state = &drv->states[index];
142
143 time_start = ns_to_ktime(local_clock());
144
145 tick_freeze();
146 /*
147 * The state used here cannot be a "coupled" one, because the "coupled"
148 * cpuidle mechanism enables interrupts and doing that with timekeeping
149 * suspended is generally unsafe.
150 */
151 stop_critical_timings();
152 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
153 rcu_idle_enter();
154 target_state->enter_s2idle(dev, drv, index);
155 if (WARN_ON_ONCE(!irqs_disabled()))
156 local_irq_disable();
157 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
158 rcu_idle_exit();
159 tick_unfreeze();
160 start_critical_timings();
161
162 time_end = ns_to_ktime(local_clock());
163
164 dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
165 dev->states_usage[index].s2idle_usage++;
166}
167
168/**
169 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
170 * @drv: cpuidle driver for the given CPU.
171 * @dev: cpuidle device for the given CPU.
172 *
173 * If there are states with the ->enter_s2idle callback, find the deepest of
174 * them and enter it with frozen tick.
175 */
176int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
177{
178 int index;
179
180 /*
181 * Find the deepest state with ->enter_s2idle present, which guarantees
182 * that interrupts won't be enabled when it exits and allows the tick to
183 * be frozen safely.
184 */
185 index = find_deepest_state(drv, dev, U64_MAX, 0, true);
186 if (index > 0) {
187 enter_s2idle_proper(drv, dev, index);
188 local_irq_enable();
189 }
190 return index;
191}
192#endif /* CONFIG_SUSPEND */
193
194/**
195 * cpuidle_enter_state - enter the state and update stats
196 * @dev: cpuidle device for this cpu
197 * @drv: cpuidle driver for this cpu
198 * @index: index into the states table in @drv of the state to enter
199 */
200int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
201 int index)
202{
203 int entered_state;
204
205 struct cpuidle_state *target_state = &drv->states[index];
206 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
207 ktime_t time_start, time_end;
208
209 /*
210 * Tell the time framework to switch to a broadcast timer because our
211 * local timer will be shut down. If a local timer is used from another
212 * CPU as a broadcast timer, this call may fail if it is not available.
213 */
214 if (broadcast && tick_broadcast_enter()) {
215 index = find_deepest_state(drv, dev, target_state->exit_latency_ns,
216 CPUIDLE_FLAG_TIMER_STOP, false);
217 if (index < 0) {
218 default_idle_call();
219 return -EBUSY;
220 }
221 target_state = &drv->states[index];
222 broadcast = false;
223 }
224
225 if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
226 leave_mm(dev->cpu);
227
228 /* Take note of the planned idle state. */
229 sched_idle_set_state(target_state);
230
231 trace_cpu_idle(index, dev->cpu);
232 time_start = ns_to_ktime(local_clock());
233
234 stop_critical_timings();
235 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
236 rcu_idle_enter();
237 entered_state = target_state->enter(dev, drv, index);
238 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
239 rcu_idle_exit();
240 start_critical_timings();
241
242 sched_clock_idle_wakeup_event();
243 time_end = ns_to_ktime(local_clock());
244 trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
245
246 /* The cpu is no longer idle or about to enter idle. */
247 sched_idle_set_state(NULL);
248
249 if (broadcast) {
250 if (WARN_ON_ONCE(!irqs_disabled()))
251 local_irq_disable();
252
253 tick_broadcast_exit();
254 }
255
256 if (!cpuidle_state_is_coupled(drv, index))
257 local_irq_enable();
258
259 if (entered_state >= 0) {
260 s64 diff, delay = drv->states[entered_state].exit_latency_ns;
261 int i;
262
263 /*
264 * Update cpuidle counters
265 * This can be moved to within driver enter routine,
266 * but that results in multiple copies of same code.
267 */
268 diff = ktime_sub(time_end, time_start);
269
270 dev->last_residency_ns = diff;
271 dev->states_usage[entered_state].time_ns += diff;
272 dev->states_usage[entered_state].usage++;
273
274 if (diff < drv->states[entered_state].target_residency_ns) {
275 for (i = entered_state - 1; i >= 0; i--) {
276 if (dev->states_usage[i].disable)
277 continue;
278
279 /* Shallower states are enabled, so update. */
280 dev->states_usage[entered_state].above++;
281 break;
282 }
283 } else if (diff > delay) {
284 for (i = entered_state + 1; i < drv->state_count; i++) {
285 if (dev->states_usage[i].disable)
286 continue;
287
288 /*
289 * Update if a deeper state would have been a
290 * better match for the observed idle duration.
291 */
292 if (diff - delay >= drv->states[i].target_residency_ns)
293 dev->states_usage[entered_state].below++;
294
295 break;
296 }
297 }
298 } else {
299 dev->last_residency_ns = 0;
300 }
301
302 return entered_state;
303}
304
305/**
306 * cpuidle_select - ask the cpuidle framework to choose an idle state
307 *
308 * @drv: the cpuidle driver
309 * @dev: the cpuidle device
310 * @stop_tick: indication on whether or not to stop the tick
311 *
312 * Returns the index of the idle state. The return value must not be negative.
313 *
314 * The memory location pointed to by @stop_tick is expected to be written the
315 * 'false' boolean value if the scheduler tick should not be stopped before
316 * entering the returned state.
317 */
318int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
319 bool *stop_tick)
320{
321 return cpuidle_curr_governor->select(drv, dev, stop_tick);
322}
323
324/**
325 * cpuidle_enter - enter into the specified idle state
326 *
327 * @drv: the cpuidle driver tied with the cpu
328 * @dev: the cpuidle device
329 * @index: the index in the idle state table
330 *
331 * Returns the index in the idle state, < 0 in case of error.
332 * The error code depends on the backend driver
333 */
334int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
335 int index)
336{
337 int ret = 0;
338
339 /*
340 * Store the next hrtimer, which becomes either next tick or the next
341 * timer event, whatever expires first. Additionally, to make this data
342 * useful for consumers outside cpuidle, we rely on that the governor's
343 * ->select() callback have decided, whether to stop the tick or not.
344 */
345 WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
346
347 if (cpuidle_state_is_coupled(drv, index))
348 ret = cpuidle_enter_state_coupled(dev, drv, index);
349 else
350 ret = cpuidle_enter_state(dev, drv, index);
351
352 WRITE_ONCE(dev->next_hrtimer, 0);
353 return ret;
354}
355
356/**
357 * cpuidle_reflect - tell the underlying governor what was the state
358 * we were in
359 *
360 * @dev : the cpuidle device
361 * @index: the index in the idle state table
362 *
363 */
364void cpuidle_reflect(struct cpuidle_device *dev, int index)
365{
366 if (cpuidle_curr_governor->reflect && index >= 0)
367 cpuidle_curr_governor->reflect(dev, index);
368}
369
370/**
371 * cpuidle_poll_time - return amount of time to poll for,
372 * governors can override dev->poll_limit_ns if necessary
373 *
374 * @drv: the cpuidle driver tied with the cpu
375 * @dev: the cpuidle device
376 *
377 */
378u64 cpuidle_poll_time(struct cpuidle_driver *drv,
379 struct cpuidle_device *dev)
380{
381 int i;
382 u64 limit_ns;
383
384 if (dev->poll_limit_ns)
385 return dev->poll_limit_ns;
386
387 limit_ns = TICK_NSEC;
388 for (i = 1; i < drv->state_count; i++) {
389 if (dev->states_usage[i].disable)
390 continue;
391
392 limit_ns = drv->states[i].target_residency_ns;
393 break;
394 }
395
396 dev->poll_limit_ns = limit_ns;
397
398 return dev->poll_limit_ns;
399}
400
401/**
402 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
403 */
404void cpuidle_install_idle_handler(void)
405{
406 if (enabled_devices) {
407 /* Make sure all changes finished before we switch to new idle */
408 smp_wmb();
409 initialized = 1;
410 }
411}
412
413/**
414 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
415 */
416void cpuidle_uninstall_idle_handler(void)
417{
418 if (enabled_devices) {
419 initialized = 0;
420 wake_up_all_idle_cpus();
421 }
422
423 /*
424 * Make sure external observers (such as the scheduler)
425 * are done looking at pointed idle states.
426 */
427 synchronize_rcu();
428}
429
430/**
431 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
432 */
433void cpuidle_pause_and_lock(void)
434{
435 mutex_lock(&cpuidle_lock);
436 cpuidle_uninstall_idle_handler();
437}
438
439EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
440
441/**
442 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
443 */
444void cpuidle_resume_and_unlock(void)
445{
446 cpuidle_install_idle_handler();
447 mutex_unlock(&cpuidle_lock);
448}
449
450EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
451
452/* Currently used in suspend/resume path to suspend cpuidle */
453void cpuidle_pause(void)
454{
455 mutex_lock(&cpuidle_lock);
456 cpuidle_uninstall_idle_handler();
457 mutex_unlock(&cpuidle_lock);
458}
459
460/* Currently used in suspend/resume path to resume cpuidle */
461void cpuidle_resume(void)
462{
463 mutex_lock(&cpuidle_lock);
464 cpuidle_install_idle_handler();
465 mutex_unlock(&cpuidle_lock);
466}
467
468/**
469 * cpuidle_enable_device - enables idle PM for a CPU
470 * @dev: the CPU
471 *
472 * This function must be called between cpuidle_pause_and_lock and
473 * cpuidle_resume_and_unlock when used externally.
474 */
475int cpuidle_enable_device(struct cpuidle_device *dev)
476{
477 int ret;
478 struct cpuidle_driver *drv;
479
480 if (!dev)
481 return -EINVAL;
482
483 if (dev->enabled)
484 return 0;
485
486 if (!cpuidle_curr_governor)
487 return -EIO;
488
489 drv = cpuidle_get_cpu_driver(dev);
490
491 if (!drv)
492 return -EIO;
493
494 if (!dev->registered)
495 return -EINVAL;
496
497 ret = cpuidle_add_device_sysfs(dev);
498 if (ret)
499 return ret;
500
501 if (cpuidle_curr_governor->enable) {
502 ret = cpuidle_curr_governor->enable(drv, dev);
503 if (ret)
504 goto fail_sysfs;
505 }
506
507 smp_wmb();
508
509 dev->enabled = 1;
510
511 enabled_devices++;
512 return 0;
513
514fail_sysfs:
515 cpuidle_remove_device_sysfs(dev);
516
517 return ret;
518}
519
520EXPORT_SYMBOL_GPL(cpuidle_enable_device);
521
522/**
523 * cpuidle_disable_device - disables idle PM for a CPU
524 * @dev: the CPU
525 *
526 * This function must be called between cpuidle_pause_and_lock and
527 * cpuidle_resume_and_unlock when used externally.
528 */
529void cpuidle_disable_device(struct cpuidle_device *dev)
530{
531 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
532
533 if (!dev || !dev->enabled)
534 return;
535
536 if (!drv || !cpuidle_curr_governor)
537 return;
538
539 dev->enabled = 0;
540
541 if (cpuidle_curr_governor->disable)
542 cpuidle_curr_governor->disable(drv, dev);
543
544 cpuidle_remove_device_sysfs(dev);
545 enabled_devices--;
546}
547
548EXPORT_SYMBOL_GPL(cpuidle_disable_device);
549
550static void __cpuidle_unregister_device(struct cpuidle_device *dev)
551{
552 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
553
554 list_del(&dev->device_list);
555 per_cpu(cpuidle_devices, dev->cpu) = NULL;
556 module_put(drv->owner);
557
558 dev->registered = 0;
559}
560
561static void __cpuidle_device_init(struct cpuidle_device *dev)
562{
563 memset(dev->states_usage, 0, sizeof(dev->states_usage));
564 dev->last_residency_ns = 0;
565 dev->next_hrtimer = 0;
566}
567
568/**
569 * __cpuidle_register_device - internal register function called before register
570 * and enable routines
571 * @dev: the cpu
572 *
573 * cpuidle_lock mutex must be held before this is called
574 */
575static int __cpuidle_register_device(struct cpuidle_device *dev)
576{
577 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
578 int i, ret;
579
580 if (!try_module_get(drv->owner))
581 return -EINVAL;
582
583 for (i = 0; i < drv->state_count; i++) {
584 if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
585 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
586
587 if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
588 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
589 }
590
591 per_cpu(cpuidle_devices, dev->cpu) = dev;
592 list_add(&dev->device_list, &cpuidle_detected_devices);
593
594 ret = cpuidle_coupled_register_device(dev);
595 if (ret)
596 __cpuidle_unregister_device(dev);
597 else
598 dev->registered = 1;
599
600 return ret;
601}
602
603/**
604 * cpuidle_register_device - registers a CPU's idle PM feature
605 * @dev: the cpu
606 */
607int cpuidle_register_device(struct cpuidle_device *dev)
608{
609 int ret = -EBUSY;
610
611 if (!dev)
612 return -EINVAL;
613
614 mutex_lock(&cpuidle_lock);
615
616 if (dev->registered)
617 goto out_unlock;
618
619 __cpuidle_device_init(dev);
620
621 ret = __cpuidle_register_device(dev);
622 if (ret)
623 goto out_unlock;
624
625 ret = cpuidle_add_sysfs(dev);
626 if (ret)
627 goto out_unregister;
628
629 ret = cpuidle_enable_device(dev);
630 if (ret)
631 goto out_sysfs;
632
633 cpuidle_install_idle_handler();
634
635out_unlock:
636 mutex_unlock(&cpuidle_lock);
637
638 return ret;
639
640out_sysfs:
641 cpuidle_remove_sysfs(dev);
642out_unregister:
643 __cpuidle_unregister_device(dev);
644 goto out_unlock;
645}
646
647EXPORT_SYMBOL_GPL(cpuidle_register_device);
648
649/**
650 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
651 * @dev: the cpu
652 */
653void cpuidle_unregister_device(struct cpuidle_device *dev)
654{
655 if (!dev || dev->registered == 0)
656 return;
657
658 cpuidle_pause_and_lock();
659
660 cpuidle_disable_device(dev);
661
662 cpuidle_remove_sysfs(dev);
663
664 __cpuidle_unregister_device(dev);
665
666 cpuidle_coupled_unregister_device(dev);
667
668 cpuidle_resume_and_unlock();
669}
670
671EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
672
673/**
674 * cpuidle_unregister: unregister a driver and the devices. This function
675 * can be used only if the driver has been previously registered through
676 * the cpuidle_register function.
677 *
678 * @drv: a valid pointer to a struct cpuidle_driver
679 */
680void cpuidle_unregister(struct cpuidle_driver *drv)
681{
682 int cpu;
683 struct cpuidle_device *device;
684
685 for_each_cpu(cpu, drv->cpumask) {
686 device = &per_cpu(cpuidle_dev, cpu);
687 cpuidle_unregister_device(device);
688 }
689
690 cpuidle_unregister_driver(drv);
691}
692EXPORT_SYMBOL_GPL(cpuidle_unregister);
693
694/**
695 * cpuidle_register: registers the driver and the cpu devices with the
696 * coupled_cpus passed as parameter. This function is used for all common
697 * initialization pattern there are in the arch specific drivers. The
698 * devices is globally defined in this file.
699 *
700 * @drv : a valid pointer to a struct cpuidle_driver
701 * @coupled_cpus: a cpumask for the coupled states
702 *
703 * Returns 0 on success, < 0 otherwise
704 */
705int cpuidle_register(struct cpuidle_driver *drv,
706 const struct cpumask *const coupled_cpus)
707{
708 int ret, cpu;
709 struct cpuidle_device *device;
710
711 ret = cpuidle_register_driver(drv);
712 if (ret) {
713 pr_err("failed to register cpuidle driver\n");
714 return ret;
715 }
716
717 for_each_cpu(cpu, drv->cpumask) {
718 device = &per_cpu(cpuidle_dev, cpu);
719 device->cpu = cpu;
720
721#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
722 /*
723 * On multiplatform for ARM, the coupled idle states could be
724 * enabled in the kernel even if the cpuidle driver does not
725 * use it. Note, coupled_cpus is a struct copy.
726 */
727 if (coupled_cpus)
728 device->coupled_cpus = *coupled_cpus;
729#endif
730 ret = cpuidle_register_device(device);
731 if (!ret)
732 continue;
733
734 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
735
736 cpuidle_unregister(drv);
737 break;
738 }
739
740 return ret;
741}
742EXPORT_SYMBOL_GPL(cpuidle_register);
743
744/**
745 * cpuidle_init - core initializer
746 */
747static int __init cpuidle_init(void)
748{
749 if (cpuidle_disabled())
750 return -ENODEV;
751
752 return cpuidle_add_interface(cpu_subsys.dev_root);
753}
754
755module_param(off, int, 0444);
756module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
757core_initcall(cpuidle_init);