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 static 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 >= CPUIDLE_DRIVER_STATE_START; i--)
69 if (drv->states[i].enter_dead)
70 return drv->states[i].enter_dead(dev, i);
76 * cpuidle_find_deepest_state - Find deepest state meeting specific conditions.
77 * @drv: cpuidle driver for the given CPU.
78 * @dev: cpuidle device for the given CPU.
79 * @freeze: Whether or not the state should be suitable for suspend-to-idle.
81 static int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
82 struct cpuidle_device *dev, bool freeze)
84 unsigned int latency_req = 0;
85 int i, ret = freeze ? -1 : CPUIDLE_DRIVER_STATE_START - 1;
87 for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
88 struct cpuidle_state *s = &drv->states[i];
89 struct cpuidle_state_usage *su = &dev->states_usage[i];
91 if (s->disabled || su->disable || s->exit_latency <= latency_req
92 || (freeze && !s->enter_freeze))
95 latency_req = s->exit_latency;
101 static void enter_freeze_proper(struct cpuidle_driver *drv,
102 struct cpuidle_device *dev, int index)
106 * The state used here cannot be a "coupled" one, because the "coupled"
107 * cpuidle mechanism enables interrupts and doing that with timekeeping
108 * suspended is generally unsafe.
110 drv->states[index].enter_freeze(dev, drv, index);
111 WARN_ON(!irqs_disabled());
113 * timekeeping_resume() that will be called by tick_unfreeze() for the
114 * last CPU executing it calls functions containing RCU read-side
115 * critical sections, so tell RCU about that.
117 RCU_NONIDLE(tick_unfreeze());
121 * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle.
123 * If there are states with the ->enter_freeze callback, find the deepest of
124 * them and enter it with frozen tick. Otherwise, find the deepest state
125 * available and enter it normally.
127 * Returns with enabled interrupts.
129 void cpuidle_enter_freeze(void)
131 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
132 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
135 if (cpuidle_not_available(drv, dev))
139 * Find the deepest state with ->enter_freeze present, which guarantees
140 * that interrupts won't be enabled when it exits and allows the tick to
143 index = cpuidle_find_deepest_state(drv, dev, true);
145 enter_freeze_proper(drv, dev, index);
151 * It is not safe to freeze the tick, find the deepest state available
152 * at all and try to enter it normally.
154 index = cpuidle_find_deepest_state(drv, dev, false);
156 cpuidle_enter(drv, dev, index);
165 * cpuidle_enter_state - enter the state and update stats
166 * @dev: cpuidle device for this cpu
167 * @drv: cpuidle driver for this cpu
168 * @next_state: index into drv->states of the state to enter
170 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
175 struct cpuidle_state *target_state = &drv->states[index];
176 ktime_t time_start, time_end;
179 trace_cpu_idle_rcuidle(index, dev->cpu);
180 time_start = ktime_get();
182 entered_state = target_state->enter(dev, drv, index);
184 time_end = ktime_get();
185 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
187 if (!cpuidle_state_is_coupled(dev, drv, entered_state))
190 diff = ktime_to_us(ktime_sub(time_end, time_start));
194 dev->last_residency = (int) diff;
196 if (entered_state >= 0) {
197 /* Update cpuidle counters */
198 /* This can be moved to within driver enter routine
199 * but that results in multiple copies of same code.
201 dev->states_usage[entered_state].time += dev->last_residency;
202 dev->states_usage[entered_state].usage++;
204 dev->last_residency = 0;
207 return entered_state;
211 * cpuidle_select - ask the cpuidle framework to choose an idle state
213 * @drv: the cpuidle driver
214 * @dev: the cpuidle device
216 * Returns the index of the idle state.
218 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
220 if (cpuidle_not_available(drv, dev))
223 return cpuidle_curr_governor->select(drv, dev);
227 * cpuidle_enter - enter into the specified idle state
229 * @drv: the cpuidle driver tied with the cpu
230 * @dev: the cpuidle device
231 * @index: the index in the idle state table
233 * Returns the index in the idle state, < 0 in case of error.
234 * The error code depends on the backend driver
236 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
239 if (cpuidle_state_is_coupled(dev, drv, index))
240 return cpuidle_enter_state_coupled(dev, drv, index);
241 return cpuidle_enter_state(dev, drv, index);
245 * cpuidle_reflect - tell the underlying governor what was the state
248 * @dev : the cpuidle device
249 * @index: the index in the idle state table
252 void cpuidle_reflect(struct cpuidle_device *dev, int index)
254 if (cpuidle_curr_governor->reflect)
255 cpuidle_curr_governor->reflect(dev, index);
259 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
261 void cpuidle_install_idle_handler(void)
263 if (enabled_devices) {
264 /* Make sure all changes finished before we switch to new idle */
271 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
273 void cpuidle_uninstall_idle_handler(void)
275 if (enabled_devices) {
277 wake_up_all_idle_cpus();
281 * Make sure external observers (such as the scheduler)
282 * are done looking at pointed idle states.
288 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
290 void cpuidle_pause_and_lock(void)
292 mutex_lock(&cpuidle_lock);
293 cpuidle_uninstall_idle_handler();
296 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
299 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
301 void cpuidle_resume_and_unlock(void)
303 cpuidle_install_idle_handler();
304 mutex_unlock(&cpuidle_lock);
307 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
309 /* Currently used in suspend/resume path to suspend cpuidle */
310 void cpuidle_pause(void)
312 mutex_lock(&cpuidle_lock);
313 cpuidle_uninstall_idle_handler();
314 mutex_unlock(&cpuidle_lock);
317 /* Currently used in suspend/resume path to resume cpuidle */
318 void cpuidle_resume(void)
320 mutex_lock(&cpuidle_lock);
321 cpuidle_install_idle_handler();
322 mutex_unlock(&cpuidle_lock);
326 * cpuidle_enable_device - enables idle PM for a CPU
329 * This function must be called between cpuidle_pause_and_lock and
330 * cpuidle_resume_and_unlock when used externally.
332 int cpuidle_enable_device(struct cpuidle_device *dev)
335 struct cpuidle_driver *drv;
343 drv = cpuidle_get_cpu_driver(dev);
345 if (!drv || !cpuidle_curr_governor)
348 if (!dev->registered)
351 if (!dev->state_count)
352 dev->state_count = drv->state_count;
354 ret = cpuidle_add_device_sysfs(dev);
358 if (cpuidle_curr_governor->enable &&
359 (ret = cpuidle_curr_governor->enable(drv, dev)))
370 cpuidle_remove_device_sysfs(dev);
375 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
378 * cpuidle_disable_device - disables idle PM for a CPU
381 * This function must be called between cpuidle_pause_and_lock and
382 * cpuidle_resume_and_unlock when used externally.
384 void cpuidle_disable_device(struct cpuidle_device *dev)
386 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
388 if (!dev || !dev->enabled)
391 if (!drv || !cpuidle_curr_governor)
396 if (cpuidle_curr_governor->disable)
397 cpuidle_curr_governor->disable(drv, dev);
399 cpuidle_remove_device_sysfs(dev);
403 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
405 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
407 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
409 list_del(&dev->device_list);
410 per_cpu(cpuidle_devices, dev->cpu) = NULL;
411 module_put(drv->owner);
414 static void __cpuidle_device_init(struct cpuidle_device *dev)
416 memset(dev->states_usage, 0, sizeof(dev->states_usage));
417 dev->last_residency = 0;
421 * __cpuidle_register_device - internal register function called before register
422 * and enable routines
425 * cpuidle_lock mutex must be held before this is called
427 static int __cpuidle_register_device(struct cpuidle_device *dev)
430 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
432 if (!try_module_get(drv->owner))
435 per_cpu(cpuidle_devices, dev->cpu) = dev;
436 list_add(&dev->device_list, &cpuidle_detected_devices);
438 ret = cpuidle_coupled_register_device(dev);
440 __cpuidle_unregister_device(dev);
448 * cpuidle_register_device - registers a CPU's idle PM feature
451 int cpuidle_register_device(struct cpuidle_device *dev)
458 mutex_lock(&cpuidle_lock);
463 __cpuidle_device_init(dev);
465 ret = __cpuidle_register_device(dev);
469 ret = cpuidle_add_sysfs(dev);
473 ret = cpuidle_enable_device(dev);
477 cpuidle_install_idle_handler();
480 mutex_unlock(&cpuidle_lock);
485 cpuidle_remove_sysfs(dev);
487 __cpuidle_unregister_device(dev);
491 EXPORT_SYMBOL_GPL(cpuidle_register_device);
494 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
497 void cpuidle_unregister_device(struct cpuidle_device *dev)
499 if (!dev || dev->registered == 0)
502 cpuidle_pause_and_lock();
504 cpuidle_disable_device(dev);
506 cpuidle_remove_sysfs(dev);
508 __cpuidle_unregister_device(dev);
510 cpuidle_coupled_unregister_device(dev);
512 cpuidle_resume_and_unlock();
515 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
518 * cpuidle_unregister: unregister a driver and the devices. This function
519 * can be used only if the driver has been previously registered through
520 * the cpuidle_register function.
522 * @drv: a valid pointer to a struct cpuidle_driver
524 void cpuidle_unregister(struct cpuidle_driver *drv)
527 struct cpuidle_device *device;
529 for_each_cpu(cpu, drv->cpumask) {
530 device = &per_cpu(cpuidle_dev, cpu);
531 cpuidle_unregister_device(device);
534 cpuidle_unregister_driver(drv);
536 EXPORT_SYMBOL_GPL(cpuidle_unregister);
539 * cpuidle_register: registers the driver and the cpu devices with the
540 * coupled_cpus passed as parameter. This function is used for all common
541 * initialization pattern there are in the arch specific drivers. The
542 * devices is globally defined in this file.
544 * @drv : a valid pointer to a struct cpuidle_driver
545 * @coupled_cpus: a cpumask for the coupled states
547 * Returns 0 on success, < 0 otherwise
549 int cpuidle_register(struct cpuidle_driver *drv,
550 const struct cpumask *const coupled_cpus)
553 struct cpuidle_device *device;
555 ret = cpuidle_register_driver(drv);
557 pr_err("failed to register cpuidle driver\n");
561 for_each_cpu(cpu, drv->cpumask) {
562 device = &per_cpu(cpuidle_dev, cpu);
565 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
567 * On multiplatform for ARM, the coupled idle states could be
568 * enabled in the kernel even if the cpuidle driver does not
569 * use it. Note, coupled_cpus is a struct copy.
572 device->coupled_cpus = *coupled_cpus;
574 ret = cpuidle_register_device(device);
578 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
580 cpuidle_unregister(drv);
586 EXPORT_SYMBOL_GPL(cpuidle_register);
591 * This function gets called when a part of the kernel has a new latency
592 * requirement. This means we need to get all processors out of their C-state,
593 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
594 * wakes them all right up.
596 static int cpuidle_latency_notify(struct notifier_block *b,
597 unsigned long l, void *v)
599 wake_up_all_idle_cpus();
603 static struct notifier_block cpuidle_latency_notifier = {
604 .notifier_call = cpuidle_latency_notify,
607 static inline void latency_notifier_init(struct notifier_block *n)
609 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
612 #else /* CONFIG_SMP */
614 #define latency_notifier_init(x) do { } while (0)
616 #endif /* CONFIG_SMP */
619 * cpuidle_init - core initializer
621 static int __init cpuidle_init(void)
625 if (cpuidle_disabled())
628 ret = cpuidle_add_interface(cpu_subsys.dev_root);
632 latency_notifier_init(&cpuidle_latency_notifier);
637 module_param(off, int, 0444);
638 core_initcall(cpuidle_init);