2 * cpuidle.c - core cpuidle infrastructure
4 * (C) 2006-2007 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5 * Shaohua Li <shaohua.li@intel.com>
6 * Adam Belay <abelay@novell.com>
8 * This code is licenced under the GPL.
11 #include <linux/clockchips.h>
12 #include <linux/kernel.h>
13 #include <linux/mutex.h>
14 #include <linux/sched.h>
15 #include <linux/notifier.h>
16 #include <linux/pm_qos.h>
17 #include <linux/cpu.h>
18 #include <linux/cpuidle.h>
19 #include <linux/ktime.h>
20 #include <linux/hrtimer.h>
21 #include <linux/module.h>
22 #include <linux/suspend.h>
23 #include <linux/tick.h>
24 #include <trace/events/power.h>
28 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
29 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
31 DEFINE_MUTEX(cpuidle_lock);
32 LIST_HEAD(cpuidle_detected_devices);
34 static int enabled_devices;
35 static int off __read_mostly;
36 static int initialized __read_mostly;
38 int cpuidle_disabled(void)
42 void disable_cpuidle(void)
47 bool cpuidle_not_available(struct cpuidle_driver *drv,
48 struct cpuidle_device *dev)
50 return off || !initialized || !drv || !dev || !dev->enabled;
54 * cpuidle_play_dead - cpu off-lining
56 * Returns in case of an error or no driver
58 int cpuidle_play_dead(void)
60 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
61 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
67 /* Find lowest-power state that supports long-term idle */
68 for (i = drv->state_count - 1; i >= 0; i--)
69 if (drv->states[i].enter_dead)
70 return drv->states[i].enter_dead(dev, i);
75 static int find_deepest_state(struct cpuidle_driver *drv,
76 struct cpuidle_device *dev,
77 unsigned int max_latency,
78 unsigned int forbidden_flags,
81 unsigned int latency_req = 0;
84 for (i = 0; i < drv->state_count; i++) {
85 struct cpuidle_state *s = &drv->states[i];
86 struct cpuidle_state_usage *su = &dev->states_usage[i];
88 if (s->disabled || su->disable || s->exit_latency <= latency_req
89 || s->exit_latency > max_latency
90 || (s->flags & forbidden_flags)
91 || (freeze && !s->enter_freeze))
94 latency_req = s->exit_latency;
100 #ifdef CONFIG_SUSPEND
102 * cpuidle_find_deepest_state - Find the deepest available idle state.
103 * @drv: cpuidle driver for the given CPU.
104 * @dev: cpuidle device for the given CPU.
106 int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
107 struct cpuidle_device *dev)
109 return find_deepest_state(drv, dev, UINT_MAX, 0, false);
112 static void enter_freeze_proper(struct cpuidle_driver *drv,
113 struct cpuidle_device *dev, int index)
117 * The state used here cannot be a "coupled" one, because the "coupled"
118 * cpuidle mechanism enables interrupts and doing that with timekeeping
119 * suspended is generally unsafe.
121 stop_critical_timings();
122 drv->states[index].enter_freeze(dev, drv, index);
123 WARN_ON(!irqs_disabled());
125 * timekeeping_resume() that will be called by tick_unfreeze() for the
126 * last CPU executing it calls functions containing RCU read-side
127 * critical sections, so tell RCU about that.
129 RCU_NONIDLE(tick_unfreeze());
130 start_critical_timings();
134 * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle.
135 * @drv: cpuidle driver for the given CPU.
136 * @dev: cpuidle device for the given CPU.
138 * If there are states with the ->enter_freeze callback, find the deepest of
139 * them and enter it with frozen tick.
141 int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
146 * Find the deepest state with ->enter_freeze present, which guarantees
147 * that interrupts won't be enabled when it exits and allows the tick to
150 index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
152 enter_freeze_proper(drv, dev, index);
156 #endif /* CONFIG_SUSPEND */
159 * cpuidle_enter_state - enter the state and update stats
160 * @dev: cpuidle device for this cpu
161 * @drv: cpuidle driver for this cpu
162 * @index: index into the states table in @drv of the state to enter
164 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
169 struct cpuidle_state *target_state = &drv->states[index];
170 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
171 ktime_t time_start, time_end;
175 * Tell the time framework to switch to a broadcast timer because our
176 * local timer will be shut down. If a local timer is used from another
177 * CPU as a broadcast timer, this call may fail if it is not available.
179 if (broadcast && tick_broadcast_enter()) {
180 index = find_deepest_state(drv, dev, target_state->exit_latency,
181 CPUIDLE_FLAG_TIMER_STOP, false);
186 target_state = &drv->states[index];
189 /* Take note of the planned idle state. */
190 sched_idle_set_state(target_state);
192 trace_cpu_idle_rcuidle(index, dev->cpu);
193 time_start = ktime_get();
195 stop_critical_timings();
196 entered_state = target_state->enter(dev, drv, index);
197 start_critical_timings();
199 time_end = ktime_get();
200 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
202 /* The cpu is no longer idle or about to enter idle. */
203 sched_idle_set_state(NULL);
206 if (WARN_ON_ONCE(!irqs_disabled()))
209 tick_broadcast_exit();
212 if (!cpuidle_state_is_coupled(dev, drv, entered_state))
215 diff = ktime_to_us(ktime_sub(time_end, time_start));
219 dev->last_residency = (int) diff;
221 if (entered_state >= 0) {
222 /* Update cpuidle counters */
223 /* This can be moved to within driver enter routine
224 * but that results in multiple copies of same code.
226 dev->states_usage[entered_state].time += dev->last_residency;
227 dev->states_usage[entered_state].usage++;
229 dev->last_residency = 0;
232 return entered_state;
236 * cpuidle_select - ask the cpuidle framework to choose an idle state
238 * @drv: the cpuidle driver
239 * @dev: the cpuidle device
241 * Returns the index of the idle state.
243 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
245 return cpuidle_curr_governor->select(drv, dev);
249 * cpuidle_enter - enter into the specified idle state
251 * @drv: the cpuidle driver tied with the cpu
252 * @dev: the cpuidle device
253 * @index: the index in the idle state table
255 * Returns the index in the idle state, < 0 in case of error.
256 * The error code depends on the backend driver
258 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
261 if (cpuidle_state_is_coupled(dev, drv, index))
262 return cpuidle_enter_state_coupled(dev, drv, index);
263 return cpuidle_enter_state(dev, drv, index);
267 * cpuidle_reflect - tell the underlying governor what was the state
270 * @dev : the cpuidle device
271 * @index: the index in the idle state table
274 void cpuidle_reflect(struct cpuidle_device *dev, int index)
276 if (cpuidle_curr_governor->reflect && index >= 0)
277 cpuidle_curr_governor->reflect(dev, index);
281 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
283 void cpuidle_install_idle_handler(void)
285 if (enabled_devices) {
286 /* Make sure all changes finished before we switch to new idle */
293 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
295 void cpuidle_uninstall_idle_handler(void)
297 if (enabled_devices) {
299 wake_up_all_idle_cpus();
303 * Make sure external observers (such as the scheduler)
304 * are done looking at pointed idle states.
310 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
312 void cpuidle_pause_and_lock(void)
314 mutex_lock(&cpuidle_lock);
315 cpuidle_uninstall_idle_handler();
318 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
321 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
323 void cpuidle_resume_and_unlock(void)
325 cpuidle_install_idle_handler();
326 mutex_unlock(&cpuidle_lock);
329 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
331 /* Currently used in suspend/resume path to suspend cpuidle */
332 void cpuidle_pause(void)
334 mutex_lock(&cpuidle_lock);
335 cpuidle_uninstall_idle_handler();
336 mutex_unlock(&cpuidle_lock);
339 /* Currently used in suspend/resume path to resume cpuidle */
340 void cpuidle_resume(void)
342 mutex_lock(&cpuidle_lock);
343 cpuidle_install_idle_handler();
344 mutex_unlock(&cpuidle_lock);
348 * cpuidle_enable_device - enables idle PM for a CPU
351 * This function must be called between cpuidle_pause_and_lock and
352 * cpuidle_resume_and_unlock when used externally.
354 int cpuidle_enable_device(struct cpuidle_device *dev)
357 struct cpuidle_driver *drv;
365 drv = cpuidle_get_cpu_driver(dev);
367 if (!drv || !cpuidle_curr_governor)
370 if (!dev->registered)
373 ret = cpuidle_add_device_sysfs(dev);
377 if (cpuidle_curr_governor->enable &&
378 (ret = cpuidle_curr_governor->enable(drv, dev)))
389 cpuidle_remove_device_sysfs(dev);
394 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
397 * cpuidle_disable_device - disables idle PM for a CPU
400 * This function must be called between cpuidle_pause_and_lock and
401 * cpuidle_resume_and_unlock when used externally.
403 void cpuidle_disable_device(struct cpuidle_device *dev)
405 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
407 if (!dev || !dev->enabled)
410 if (!drv || !cpuidle_curr_governor)
415 if (cpuidle_curr_governor->disable)
416 cpuidle_curr_governor->disable(drv, dev);
418 cpuidle_remove_device_sysfs(dev);
422 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
424 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
426 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
428 list_del(&dev->device_list);
429 per_cpu(cpuidle_devices, dev->cpu) = NULL;
430 module_put(drv->owner);
433 static void __cpuidle_device_init(struct cpuidle_device *dev)
435 memset(dev->states_usage, 0, sizeof(dev->states_usage));
436 dev->last_residency = 0;
440 * __cpuidle_register_device - internal register function called before register
441 * and enable routines
444 * cpuidle_lock mutex must be held before this is called
446 static int __cpuidle_register_device(struct cpuidle_device *dev)
449 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
451 if (!try_module_get(drv->owner))
454 per_cpu(cpuidle_devices, dev->cpu) = dev;
455 list_add(&dev->device_list, &cpuidle_detected_devices);
457 ret = cpuidle_coupled_register_device(dev);
459 __cpuidle_unregister_device(dev);
467 * cpuidle_register_device - registers a CPU's idle PM feature
470 int cpuidle_register_device(struct cpuidle_device *dev)
477 mutex_lock(&cpuidle_lock);
482 __cpuidle_device_init(dev);
484 ret = __cpuidle_register_device(dev);
488 ret = cpuidle_add_sysfs(dev);
492 ret = cpuidle_enable_device(dev);
496 cpuidle_install_idle_handler();
499 mutex_unlock(&cpuidle_lock);
504 cpuidle_remove_sysfs(dev);
506 __cpuidle_unregister_device(dev);
510 EXPORT_SYMBOL_GPL(cpuidle_register_device);
513 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
516 void cpuidle_unregister_device(struct cpuidle_device *dev)
518 if (!dev || dev->registered == 0)
521 cpuidle_pause_and_lock();
523 cpuidle_disable_device(dev);
525 cpuidle_remove_sysfs(dev);
527 __cpuidle_unregister_device(dev);
529 cpuidle_coupled_unregister_device(dev);
531 cpuidle_resume_and_unlock();
534 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
537 * cpuidle_unregister: unregister a driver and the devices. This function
538 * can be used only if the driver has been previously registered through
539 * the cpuidle_register function.
541 * @drv: a valid pointer to a struct cpuidle_driver
543 void cpuidle_unregister(struct cpuidle_driver *drv)
546 struct cpuidle_device *device;
548 for_each_cpu(cpu, drv->cpumask) {
549 device = &per_cpu(cpuidle_dev, cpu);
550 cpuidle_unregister_device(device);
553 cpuidle_unregister_driver(drv);
555 EXPORT_SYMBOL_GPL(cpuidle_unregister);
558 * cpuidle_register: registers the driver and the cpu devices with the
559 * coupled_cpus passed as parameter. This function is used for all common
560 * initialization pattern there are in the arch specific drivers. The
561 * devices is globally defined in this file.
563 * @drv : a valid pointer to a struct cpuidle_driver
564 * @coupled_cpus: a cpumask for the coupled states
566 * Returns 0 on success, < 0 otherwise
568 int cpuidle_register(struct cpuidle_driver *drv,
569 const struct cpumask *const coupled_cpus)
572 struct cpuidle_device *device;
574 ret = cpuidle_register_driver(drv);
576 pr_err("failed to register cpuidle driver\n");
580 for_each_cpu(cpu, drv->cpumask) {
581 device = &per_cpu(cpuidle_dev, cpu);
584 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
586 * On multiplatform for ARM, the coupled idle states could be
587 * enabled in the kernel even if the cpuidle driver does not
588 * use it. Note, coupled_cpus is a struct copy.
591 device->coupled_cpus = *coupled_cpus;
593 ret = cpuidle_register_device(device);
597 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
599 cpuidle_unregister(drv);
605 EXPORT_SYMBOL_GPL(cpuidle_register);
610 * This function gets called when a part of the kernel has a new latency
611 * requirement. This means we need to get all processors out of their C-state,
612 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
613 * wakes them all right up.
615 static int cpuidle_latency_notify(struct notifier_block *b,
616 unsigned long l, void *v)
618 wake_up_all_idle_cpus();
622 static struct notifier_block cpuidle_latency_notifier = {
623 .notifier_call = cpuidle_latency_notify,
626 static inline void latency_notifier_init(struct notifier_block *n)
628 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
631 #else /* CONFIG_SMP */
633 #define latency_notifier_init(x) do { } while (0)
635 #endif /* CONFIG_SMP */
638 * cpuidle_init - core initializer
640 static int __init cpuidle_init(void)
644 if (cpuidle_disabled())
647 ret = cpuidle_add_interface(cpu_subsys.dev_root);
651 latency_notifier_init(&cpuidle_latency_notifier);
656 module_param(off, int, 0444);
657 core_initcall(cpuidle_init);