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/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 <trace/events/power.h>
29 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
30 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
32 DEFINE_MUTEX(cpuidle_lock);
33 LIST_HEAD(cpuidle_detected_devices);
35 static int enabled_devices;
36 static int off __read_mostly;
37 static int initialized __read_mostly;
39 int cpuidle_disabled(void)
43 void disable_cpuidle(void)
48 bool cpuidle_not_available(struct cpuidle_driver *drv,
49 struct cpuidle_device *dev)
51 return off || !initialized || !drv || !dev || !dev->enabled;
55 * cpuidle_play_dead - cpu off-lining
57 * Returns in case of an error or no driver
59 int cpuidle_play_dead(void)
61 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
62 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
68 /* Find lowest-power state that supports long-term idle */
69 for (i = drv->state_count - 1; i >= 0; i--)
70 if (drv->states[i].enter_dead)
71 return drv->states[i].enter_dead(dev, i);
76 static int find_deepest_state(struct cpuidle_driver *drv,
77 struct cpuidle_device *dev,
78 unsigned int max_latency,
79 unsigned int forbidden_flags,
82 unsigned int latency_req = 0;
85 for (i = 1; i < drv->state_count; i++) {
86 struct cpuidle_state *s = &drv->states[i];
87 struct cpuidle_state_usage *su = &dev->states_usage[i];
89 if (s->disabled || su->disable || s->exit_latency <= latency_req
90 || s->exit_latency > max_latency
91 || (s->flags & forbidden_flags)
92 || (s2idle && !s->enter_s2idle))
95 latency_req = s->exit_latency;
102 * cpuidle_use_deepest_state - Set/clear governor override flag.
103 * @enable: New value of the flag.
105 * Set/unset the current CPU to use the deepest idle state (override governors
106 * going forward if set).
108 void cpuidle_use_deepest_state(bool enable)
110 struct cpuidle_device *dev;
113 dev = cpuidle_get_device();
115 dev->use_deepest_state = enable;
120 * cpuidle_find_deepest_state - Find the deepest available idle state.
121 * @drv: cpuidle driver for the given CPU.
122 * @dev: cpuidle device for the given CPU.
124 int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
125 struct cpuidle_device *dev)
127 return find_deepest_state(drv, dev, UINT_MAX, 0, false);
130 #ifdef CONFIG_SUSPEND
131 static void enter_s2idle_proper(struct cpuidle_driver *drv,
132 struct cpuidle_device *dev, int index)
134 ktime_t time_start, time_end;
136 time_start = ns_to_ktime(local_clock());
139 * trace_suspend_resume() called by tick_freeze() for the last CPU
140 * executing it contains RCU usage regarded as invalid in the idle
141 * context, so tell RCU about that.
143 RCU_NONIDLE(tick_freeze());
145 * The state used here cannot be a "coupled" one, because the "coupled"
146 * cpuidle mechanism enables interrupts and doing that with timekeeping
147 * suspended is generally unsafe.
149 stop_critical_timings();
150 drv->states[index].enter_s2idle(dev, drv, index);
151 WARN_ON(!irqs_disabled());
153 * timekeeping_resume() that will be called by tick_unfreeze() for the
154 * first CPU executing it calls functions containing RCU read-side
155 * critical sections, so tell RCU about that.
157 RCU_NONIDLE(tick_unfreeze());
158 start_critical_timings();
160 time_end = ns_to_ktime(local_clock());
162 dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
163 dev->states_usage[index].s2idle_usage++;
167 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
168 * @drv: cpuidle driver for the given CPU.
169 * @dev: cpuidle device for the given CPU.
171 * If there are states with the ->enter_s2idle callback, find the deepest of
172 * them and enter it with frozen tick.
174 int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
179 * Find the deepest state with ->enter_s2idle present, which guarantees
180 * that interrupts won't be enabled when it exits and allows the tick to
183 index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
185 enter_s2idle_proper(drv, dev, index);
189 #endif /* CONFIG_SUSPEND */
192 * cpuidle_enter_state - enter the state and update stats
193 * @dev: cpuidle device for this cpu
194 * @drv: cpuidle driver for this cpu
195 * @index: index into the states table in @drv of the state to enter
197 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
202 struct cpuidle_state *target_state = &drv->states[index];
203 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
204 ktime_t time_start, time_end;
207 * Tell the time framework to switch to a broadcast timer because our
208 * local timer will be shut down. If a local timer is used from another
209 * CPU as a broadcast timer, this call may fail if it is not available.
211 if (broadcast && tick_broadcast_enter()) {
212 index = find_deepest_state(drv, dev, target_state->exit_latency,
213 CPUIDLE_FLAG_TIMER_STOP, false);
218 target_state = &drv->states[index];
222 /* Take note of the planned idle state. */
223 sched_idle_set_state(target_state);
225 trace_cpu_idle_rcuidle(index, dev->cpu);
226 time_start = ns_to_ktime(local_clock());
228 stop_critical_timings();
229 entered_state = target_state->enter(dev, drv, index);
230 start_critical_timings();
232 sched_clock_idle_wakeup_event();
233 time_end = ns_to_ktime(local_clock());
234 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
236 /* The cpu is no longer idle or about to enter idle. */
237 sched_idle_set_state(NULL);
240 if (WARN_ON_ONCE(!irqs_disabled()))
243 tick_broadcast_exit();
246 if (!cpuidle_state_is_coupled(drv, index))
249 if (entered_state >= 0) {
250 s64 diff, delay = drv->states[entered_state].exit_latency;
254 * Update cpuidle counters
255 * This can be moved to within driver enter routine,
256 * but that results in multiple copies of same code.
258 diff = ktime_us_delta(time_end, time_start);
262 dev->last_residency = (int)diff;
263 dev->states_usage[entered_state].time += dev->last_residency;
264 dev->states_usage[entered_state].usage++;
266 if (diff < drv->states[entered_state].target_residency) {
267 for (i = entered_state - 1; i >= 0; i--) {
268 if (drv->states[i].disabled ||
269 dev->states_usage[i].disable)
272 /* Shallower states are enabled, so update. */
273 dev->states_usage[entered_state].above++;
276 } else if (diff > delay) {
277 for (i = entered_state + 1; i < drv->state_count; i++) {
278 if (drv->states[i].disabled ||
279 dev->states_usage[i].disable)
283 * Update if a deeper state would have been a
284 * better match for the observed idle duration.
286 if (diff - delay >= drv->states[i].target_residency)
287 dev->states_usage[entered_state].below++;
293 dev->last_residency = 0;
296 return entered_state;
300 * cpuidle_select - ask the cpuidle framework to choose an idle state
302 * @drv: the cpuidle driver
303 * @dev: the cpuidle device
304 * @stop_tick: indication on whether or not to stop the tick
306 * Returns the index of the idle state. The return value must not be negative.
308 * The memory location pointed to by @stop_tick is expected to be written the
309 * 'false' boolean value if the scheduler tick should not be stopped before
310 * entering the returned state.
312 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
315 return cpuidle_curr_governor->select(drv, dev, stop_tick);
319 * cpuidle_enter - enter into the specified idle state
321 * @drv: the cpuidle driver tied with the cpu
322 * @dev: the cpuidle device
323 * @index: the index in the idle state table
325 * Returns the index in the idle state, < 0 in case of error.
326 * The error code depends on the backend driver
328 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
334 * Store the next hrtimer, which becomes either next tick or the next
335 * timer event, whatever expires first. Additionally, to make this data
336 * useful for consumers outside cpuidle, we rely on that the governor's
337 * ->select() callback have decided, whether to stop the tick or not.
339 WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
341 if (cpuidle_state_is_coupled(drv, index))
342 ret = cpuidle_enter_state_coupled(dev, drv, index);
344 ret = cpuidle_enter_state(dev, drv, index);
346 WRITE_ONCE(dev->next_hrtimer, 0);
351 * cpuidle_reflect - tell the underlying governor what was the state
354 * @dev : the cpuidle device
355 * @index: the index in the idle state table
358 void cpuidle_reflect(struct cpuidle_device *dev, int index)
360 if (cpuidle_curr_governor->reflect && index >= 0)
361 cpuidle_curr_governor->reflect(dev, index);
365 * cpuidle_poll_time - return amount of time to poll for,
366 * governors can override dev->poll_limit_ns if necessary
368 * @drv: the cpuidle driver tied with the cpu
369 * @dev: the cpuidle device
372 u64 cpuidle_poll_time(struct cpuidle_driver *drv,
373 struct cpuidle_device *dev)
378 if (dev->poll_limit_ns)
379 return dev->poll_limit_ns;
381 limit_ns = TICK_NSEC;
382 for (i = 1; i < drv->state_count; i++) {
383 if (drv->states[i].disabled || dev->states_usage[i].disable)
386 limit_ns = (u64)drv->states[i].target_residency * NSEC_PER_USEC;
389 dev->poll_limit_ns = limit_ns;
391 return dev->poll_limit_ns;
395 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
397 void cpuidle_install_idle_handler(void)
399 if (enabled_devices) {
400 /* Make sure all changes finished before we switch to new idle */
407 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
409 void cpuidle_uninstall_idle_handler(void)
411 if (enabled_devices) {
413 wake_up_all_idle_cpus();
417 * Make sure external observers (such as the scheduler)
418 * are done looking at pointed idle states.
424 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
426 void cpuidle_pause_and_lock(void)
428 mutex_lock(&cpuidle_lock);
429 cpuidle_uninstall_idle_handler();
432 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
435 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
437 void cpuidle_resume_and_unlock(void)
439 cpuidle_install_idle_handler();
440 mutex_unlock(&cpuidle_lock);
443 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
445 /* Currently used in suspend/resume path to suspend cpuidle */
446 void cpuidle_pause(void)
448 mutex_lock(&cpuidle_lock);
449 cpuidle_uninstall_idle_handler();
450 mutex_unlock(&cpuidle_lock);
453 /* Currently used in suspend/resume path to resume cpuidle */
454 void cpuidle_resume(void)
456 mutex_lock(&cpuidle_lock);
457 cpuidle_install_idle_handler();
458 mutex_unlock(&cpuidle_lock);
462 * cpuidle_enable_device - enables idle PM for a CPU
465 * This function must be called between cpuidle_pause_and_lock and
466 * cpuidle_resume_and_unlock when used externally.
468 int cpuidle_enable_device(struct cpuidle_device *dev)
471 struct cpuidle_driver *drv;
479 if (!cpuidle_curr_governor)
482 drv = cpuidle_get_cpu_driver(dev);
487 if (!dev->registered)
490 ret = cpuidle_add_device_sysfs(dev);
494 if (cpuidle_curr_governor->enable) {
495 ret = cpuidle_curr_governor->enable(drv, dev);
508 cpuidle_remove_device_sysfs(dev);
513 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
516 * cpuidle_disable_device - disables idle PM for a CPU
519 * This function must be called between cpuidle_pause_and_lock and
520 * cpuidle_resume_and_unlock when used externally.
522 void cpuidle_disable_device(struct cpuidle_device *dev)
524 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
526 if (!dev || !dev->enabled)
529 if (!drv || !cpuidle_curr_governor)
534 if (cpuidle_curr_governor->disable)
535 cpuidle_curr_governor->disable(drv, dev);
537 cpuidle_remove_device_sysfs(dev);
541 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
543 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
545 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
547 list_del(&dev->device_list);
548 per_cpu(cpuidle_devices, dev->cpu) = NULL;
549 module_put(drv->owner);
554 static void __cpuidle_device_init(struct cpuidle_device *dev)
556 memset(dev->states_usage, 0, sizeof(dev->states_usage));
557 dev->last_residency = 0;
558 dev->next_hrtimer = 0;
562 * __cpuidle_register_device - internal register function called before register
563 * and enable routines
566 * cpuidle_lock mutex must be held before this is called
568 static int __cpuidle_register_device(struct cpuidle_device *dev)
571 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
573 if (!try_module_get(drv->owner))
576 per_cpu(cpuidle_devices, dev->cpu) = dev;
577 list_add(&dev->device_list, &cpuidle_detected_devices);
579 ret = cpuidle_coupled_register_device(dev);
581 __cpuidle_unregister_device(dev);
589 * cpuidle_register_device - registers a CPU's idle PM feature
592 int cpuidle_register_device(struct cpuidle_device *dev)
599 mutex_lock(&cpuidle_lock);
604 __cpuidle_device_init(dev);
606 ret = __cpuidle_register_device(dev);
610 ret = cpuidle_add_sysfs(dev);
614 ret = cpuidle_enable_device(dev);
618 cpuidle_install_idle_handler();
621 mutex_unlock(&cpuidle_lock);
626 cpuidle_remove_sysfs(dev);
628 __cpuidle_unregister_device(dev);
632 EXPORT_SYMBOL_GPL(cpuidle_register_device);
635 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
638 void cpuidle_unregister_device(struct cpuidle_device *dev)
640 if (!dev || dev->registered == 0)
643 cpuidle_pause_and_lock();
645 cpuidle_disable_device(dev);
647 cpuidle_remove_sysfs(dev);
649 __cpuidle_unregister_device(dev);
651 cpuidle_coupled_unregister_device(dev);
653 cpuidle_resume_and_unlock();
656 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
659 * cpuidle_unregister: unregister a driver and the devices. This function
660 * can be used only if the driver has been previously registered through
661 * the cpuidle_register function.
663 * @drv: a valid pointer to a struct cpuidle_driver
665 void cpuidle_unregister(struct cpuidle_driver *drv)
668 struct cpuidle_device *device;
670 for_each_cpu(cpu, drv->cpumask) {
671 device = &per_cpu(cpuidle_dev, cpu);
672 cpuidle_unregister_device(device);
675 cpuidle_unregister_driver(drv);
677 EXPORT_SYMBOL_GPL(cpuidle_unregister);
680 * cpuidle_register: registers the driver and the cpu devices with the
681 * coupled_cpus passed as parameter. This function is used for all common
682 * initialization pattern there are in the arch specific drivers. The
683 * devices is globally defined in this file.
685 * @drv : a valid pointer to a struct cpuidle_driver
686 * @coupled_cpus: a cpumask for the coupled states
688 * Returns 0 on success, < 0 otherwise
690 int cpuidle_register(struct cpuidle_driver *drv,
691 const struct cpumask *const coupled_cpus)
694 struct cpuidle_device *device;
696 ret = cpuidle_register_driver(drv);
698 pr_err("failed to register cpuidle driver\n");
702 for_each_cpu(cpu, drv->cpumask) {
703 device = &per_cpu(cpuidle_dev, cpu);
706 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
708 * On multiplatform for ARM, the coupled idle states could be
709 * enabled in the kernel even if the cpuidle driver does not
710 * use it. Note, coupled_cpus is a struct copy.
713 device->coupled_cpus = *coupled_cpus;
715 ret = cpuidle_register_device(device);
719 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
721 cpuidle_unregister(drv);
727 EXPORT_SYMBOL_GPL(cpuidle_register);
732 * This function gets called when a part of the kernel has a new latency
733 * requirement. This means we need to get all processors out of their C-state,
734 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
735 * wakes them all right up.
737 static int cpuidle_latency_notify(struct notifier_block *b,
738 unsigned long l, void *v)
740 wake_up_all_idle_cpus();
744 static struct notifier_block cpuidle_latency_notifier = {
745 .notifier_call = cpuidle_latency_notify,
748 static inline void latency_notifier_init(struct notifier_block *n)
750 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
753 #else /* CONFIG_SMP */
755 #define latency_notifier_init(x) do { } while (0)
757 #endif /* CONFIG_SMP */
760 * cpuidle_init - core initializer
762 static int __init cpuidle_init(void)
766 if (cpuidle_disabled())
769 ret = cpuidle_add_interface(cpu_subsys.dev_root);
773 latency_notifier_init(&cpuidle_latency_notifier);
778 module_param(off, int, 0444);
779 module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
780 core_initcall(cpuidle_init);