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 <linux/mmu_context.h>
26 #include <trace/events/power.h>
30 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
31 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
33 DEFINE_MUTEX(cpuidle_lock);
34 LIST_HEAD(cpuidle_detected_devices);
36 static int enabled_devices;
37 static int off __read_mostly;
38 static int initialized __read_mostly;
40 int cpuidle_disabled(void)
44 void disable_cpuidle(void)
49 bool cpuidle_not_available(struct cpuidle_driver *drv,
50 struct cpuidle_device *dev)
52 return off || !initialized || !drv || !dev || !dev->enabled;
56 * cpuidle_play_dead - cpu off-lining
58 * Returns in case of an error or no driver
60 int cpuidle_play_dead(void)
62 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
63 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
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);
77 static int find_deepest_state(struct cpuidle_driver *drv,
78 struct cpuidle_device *dev,
80 unsigned int forbidden_flags,
86 for (i = 1; i < drv->state_count; i++) {
87 struct cpuidle_state *s = &drv->states[i];
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))
96 latency_req = s->exit_latency_ns;
103 * cpuidle_use_deepest_state - Set/unset governor override mode.
104 * @latency_limit_ns: Idle state exit latency limit (or no override if 0).
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.
110 void cpuidle_use_deepest_state(u64 latency_limit_ns)
112 struct cpuidle_device *dev;
115 dev = cpuidle_get_device();
117 dev->forced_idle_latency_limit_ns = latency_limit_ns;
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
127 * Return: the index of the deepest available idle state.
129 int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
130 struct cpuidle_device *dev,
131 u64 latency_limit_ns)
133 return find_deepest_state(drv, dev, latency_limit_ns, 0, false);
136 #ifdef CONFIG_SUSPEND
137 static void enter_s2idle_proper(struct cpuidle_driver *drv,
138 struct cpuidle_device *dev, int index)
140 ktime_t time_start, time_end;
142 time_start = ns_to_ktime(local_clock());
145 * trace_suspend_resume() called by tick_freeze() for the last CPU
146 * executing it contains RCU usage regarded as invalid in the idle
147 * context, so tell RCU about that.
151 * The state used here cannot be a "coupled" one, because the "coupled"
152 * cpuidle mechanism enables interrupts and doing that with timekeeping
153 * suspended is generally unsafe.
155 stop_critical_timings();
157 drv->states[index].enter_s2idle(dev, drv, index);
158 if (WARN_ON_ONCE(!irqs_disabled()))
161 * timekeeping_resume() that will be called by tick_unfreeze() for the
162 * first CPU executing it calls functions containing RCU read-side
163 * critical sections, so tell RCU about that.
167 start_critical_timings();
169 time_end = ns_to_ktime(local_clock());
171 dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
172 dev->states_usage[index].s2idle_usage++;
176 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
177 * @drv: cpuidle driver for the given CPU.
178 * @dev: cpuidle device for the given CPU.
180 * If there are states with the ->enter_s2idle callback, find the deepest of
181 * them and enter it with frozen tick.
183 int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
188 * Find the deepest state with ->enter_s2idle present, which guarantees
189 * that interrupts won't be enabled when it exits and allows the tick to
192 index = find_deepest_state(drv, dev, U64_MAX, 0, true);
194 enter_s2idle_proper(drv, dev, index);
199 #endif /* CONFIG_SUSPEND */
202 * cpuidle_enter_state - enter the state and update stats
203 * @dev: cpuidle device for this cpu
204 * @drv: cpuidle driver for this cpu
205 * @index: index into the states table in @drv of the state to enter
207 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
212 struct cpuidle_state *target_state = &drv->states[index];
213 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
214 ktime_t time_start, time_end;
217 * Tell the time framework to switch to a broadcast timer because our
218 * local timer will be shut down. If a local timer is used from another
219 * CPU as a broadcast timer, this call may fail if it is not available.
221 if (broadcast && tick_broadcast_enter()) {
222 index = find_deepest_state(drv, dev, target_state->exit_latency_ns,
223 CPUIDLE_FLAG_TIMER_STOP, false);
228 target_state = &drv->states[index];
232 if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
235 /* Take note of the planned idle state. */
236 sched_idle_set_state(target_state);
238 trace_cpu_idle(index, dev->cpu);
239 time_start = ns_to_ktime(local_clock());
241 stop_critical_timings();
243 entered_state = target_state->enter(dev, drv, index);
245 start_critical_timings();
247 sched_clock_idle_wakeup_event();
248 time_end = ns_to_ktime(local_clock());
249 trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
251 /* The cpu is no longer idle or about to enter idle. */
252 sched_idle_set_state(NULL);
255 if (WARN_ON_ONCE(!irqs_disabled()))
258 tick_broadcast_exit();
261 if (!cpuidle_state_is_coupled(drv, index))
264 if (entered_state >= 0) {
265 s64 diff, delay = drv->states[entered_state].exit_latency_ns;
269 * Update cpuidle counters
270 * This can be moved to within driver enter routine,
271 * but that results in multiple copies of same code.
273 diff = ktime_sub(time_end, time_start);
275 dev->last_residency_ns = diff;
276 dev->states_usage[entered_state].time_ns += diff;
277 dev->states_usage[entered_state].usage++;
279 if (diff < drv->states[entered_state].target_residency_ns) {
280 for (i = entered_state - 1; i >= 0; i--) {
281 if (dev->states_usage[i].disable)
284 /* Shallower states are enabled, so update. */
285 dev->states_usage[entered_state].above++;
288 } else if (diff > delay) {
289 for (i = entered_state + 1; i < drv->state_count; i++) {
290 if (dev->states_usage[i].disable)
294 * Update if a deeper state would have been a
295 * better match for the observed idle duration.
297 if (diff - delay >= drv->states[i].target_residency_ns)
298 dev->states_usage[entered_state].below++;
304 dev->last_residency_ns = 0;
307 return entered_state;
311 * cpuidle_select - ask the cpuidle framework to choose an idle state
313 * @drv: the cpuidle driver
314 * @dev: the cpuidle device
315 * @stop_tick: indication on whether or not to stop the tick
317 * Returns the index of the idle state. The return value must not be negative.
319 * The memory location pointed to by @stop_tick is expected to be written the
320 * 'false' boolean value if the scheduler tick should not be stopped before
321 * entering the returned state.
323 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
326 return cpuidle_curr_governor->select(drv, dev, stop_tick);
330 * cpuidle_enter - enter into the specified idle state
332 * @drv: the cpuidle driver tied with the cpu
333 * @dev: the cpuidle device
334 * @index: the index in the idle state table
336 * Returns the index in the idle state, < 0 in case of error.
337 * The error code depends on the backend driver
339 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
345 * Store the next hrtimer, which becomes either next tick or the next
346 * timer event, whatever expires first. Additionally, to make this data
347 * useful for consumers outside cpuidle, we rely on that the governor's
348 * ->select() callback have decided, whether to stop the tick or not.
350 WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
352 if (cpuidle_state_is_coupled(drv, index))
353 ret = cpuidle_enter_state_coupled(dev, drv, index);
355 ret = cpuidle_enter_state(dev, drv, index);
357 WRITE_ONCE(dev->next_hrtimer, 0);
362 * cpuidle_reflect - tell the underlying governor what was the state
365 * @dev : the cpuidle device
366 * @index: the index in the idle state table
369 void cpuidle_reflect(struct cpuidle_device *dev, int index)
371 if (cpuidle_curr_governor->reflect && index >= 0)
372 cpuidle_curr_governor->reflect(dev, index);
376 * cpuidle_poll_time - return amount of time to poll for,
377 * governors can override dev->poll_limit_ns if necessary
379 * @drv: the cpuidle driver tied with the cpu
380 * @dev: the cpuidle device
383 u64 cpuidle_poll_time(struct cpuidle_driver *drv,
384 struct cpuidle_device *dev)
389 if (dev->poll_limit_ns)
390 return dev->poll_limit_ns;
392 limit_ns = TICK_NSEC;
393 for (i = 1; i < drv->state_count; i++) {
394 if (dev->states_usage[i].disable)
397 limit_ns = drv->states[i].target_residency_ns;
401 dev->poll_limit_ns = limit_ns;
403 return dev->poll_limit_ns;
407 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
409 void cpuidle_install_idle_handler(void)
411 if (enabled_devices) {
412 /* Make sure all changes finished before we switch to new idle */
419 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
421 void cpuidle_uninstall_idle_handler(void)
423 if (enabled_devices) {
425 wake_up_all_idle_cpus();
429 * Make sure external observers (such as the scheduler)
430 * are done looking at pointed idle states.
436 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
438 void cpuidle_pause_and_lock(void)
440 mutex_lock(&cpuidle_lock);
441 cpuidle_uninstall_idle_handler();
444 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
447 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
449 void cpuidle_resume_and_unlock(void)
451 cpuidle_install_idle_handler();
452 mutex_unlock(&cpuidle_lock);
455 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
457 /* Currently used in suspend/resume path to suspend cpuidle */
458 void cpuidle_pause(void)
460 mutex_lock(&cpuidle_lock);
461 cpuidle_uninstall_idle_handler();
462 mutex_unlock(&cpuidle_lock);
465 /* Currently used in suspend/resume path to resume cpuidle */
466 void cpuidle_resume(void)
468 mutex_lock(&cpuidle_lock);
469 cpuidle_install_idle_handler();
470 mutex_unlock(&cpuidle_lock);
474 * cpuidle_enable_device - enables idle PM for a CPU
477 * This function must be called between cpuidle_pause_and_lock and
478 * cpuidle_resume_and_unlock when used externally.
480 int cpuidle_enable_device(struct cpuidle_device *dev)
483 struct cpuidle_driver *drv;
491 if (!cpuidle_curr_governor)
494 drv = cpuidle_get_cpu_driver(dev);
499 if (!dev->registered)
502 ret = cpuidle_add_device_sysfs(dev);
506 if (cpuidle_curr_governor->enable) {
507 ret = cpuidle_curr_governor->enable(drv, dev);
520 cpuidle_remove_device_sysfs(dev);
525 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
528 * cpuidle_disable_device - disables idle PM for a CPU
531 * This function must be called between cpuidle_pause_and_lock and
532 * cpuidle_resume_and_unlock when used externally.
534 void cpuidle_disable_device(struct cpuidle_device *dev)
536 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
538 if (!dev || !dev->enabled)
541 if (!drv || !cpuidle_curr_governor)
546 if (cpuidle_curr_governor->disable)
547 cpuidle_curr_governor->disable(drv, dev);
549 cpuidle_remove_device_sysfs(dev);
553 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
555 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
557 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
559 list_del(&dev->device_list);
560 per_cpu(cpuidle_devices, dev->cpu) = NULL;
561 module_put(drv->owner);
566 static void __cpuidle_device_init(struct cpuidle_device *dev)
568 memset(dev->states_usage, 0, sizeof(dev->states_usage));
569 dev->last_residency_ns = 0;
570 dev->next_hrtimer = 0;
574 * __cpuidle_register_device - internal register function called before register
575 * and enable routines
578 * cpuidle_lock mutex must be held before this is called
580 static int __cpuidle_register_device(struct cpuidle_device *dev)
582 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
585 if (!try_module_get(drv->owner))
588 for (i = 0; i < drv->state_count; i++) {
589 if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
590 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
592 if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
593 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
596 per_cpu(cpuidle_devices, dev->cpu) = dev;
597 list_add(&dev->device_list, &cpuidle_detected_devices);
599 ret = cpuidle_coupled_register_device(dev);
601 __cpuidle_unregister_device(dev);
609 * cpuidle_register_device - registers a CPU's idle PM feature
612 int cpuidle_register_device(struct cpuidle_device *dev)
619 mutex_lock(&cpuidle_lock);
624 __cpuidle_device_init(dev);
626 ret = __cpuidle_register_device(dev);
630 ret = cpuidle_add_sysfs(dev);
634 ret = cpuidle_enable_device(dev);
638 cpuidle_install_idle_handler();
641 mutex_unlock(&cpuidle_lock);
646 cpuidle_remove_sysfs(dev);
648 __cpuidle_unregister_device(dev);
652 EXPORT_SYMBOL_GPL(cpuidle_register_device);
655 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
658 void cpuidle_unregister_device(struct cpuidle_device *dev)
660 if (!dev || dev->registered == 0)
663 cpuidle_pause_and_lock();
665 cpuidle_disable_device(dev);
667 cpuidle_remove_sysfs(dev);
669 __cpuidle_unregister_device(dev);
671 cpuidle_coupled_unregister_device(dev);
673 cpuidle_resume_and_unlock();
676 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
679 * cpuidle_unregister: unregister a driver and the devices. This function
680 * can be used only if the driver has been previously registered through
681 * the cpuidle_register function.
683 * @drv: a valid pointer to a struct cpuidle_driver
685 void cpuidle_unregister(struct cpuidle_driver *drv)
688 struct cpuidle_device *device;
690 for_each_cpu(cpu, drv->cpumask) {
691 device = &per_cpu(cpuidle_dev, cpu);
692 cpuidle_unregister_device(device);
695 cpuidle_unregister_driver(drv);
697 EXPORT_SYMBOL_GPL(cpuidle_unregister);
700 * cpuidle_register: registers the driver and the cpu devices with the
701 * coupled_cpus passed as parameter. This function is used for all common
702 * initialization pattern there are in the arch specific drivers. The
703 * devices is globally defined in this file.
705 * @drv : a valid pointer to a struct cpuidle_driver
706 * @coupled_cpus: a cpumask for the coupled states
708 * Returns 0 on success, < 0 otherwise
710 int cpuidle_register(struct cpuidle_driver *drv,
711 const struct cpumask *const coupled_cpus)
714 struct cpuidle_device *device;
716 ret = cpuidle_register_driver(drv);
718 pr_err("failed to register cpuidle driver\n");
722 for_each_cpu(cpu, drv->cpumask) {
723 device = &per_cpu(cpuidle_dev, cpu);
726 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
728 * On multiplatform for ARM, the coupled idle states could be
729 * enabled in the kernel even if the cpuidle driver does not
730 * use it. Note, coupled_cpus is a struct copy.
733 device->coupled_cpus = *coupled_cpus;
735 ret = cpuidle_register_device(device);
739 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
741 cpuidle_unregister(drv);
747 EXPORT_SYMBOL_GPL(cpuidle_register);
750 * cpuidle_init - core initializer
752 static int __init cpuidle_init(void)
754 if (cpuidle_disabled())
757 return cpuidle_add_interface(cpu_subsys.dev_root);
760 module_param(off, int, 0444);
761 module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
762 core_initcall(cpuidle_init);