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;
141 struct cpuidle_state *target_state = &drv->states[index];
143 time_start = ns_to_ktime(local_clock());
146 * trace_suspend_resume() called by tick_freeze() for the last CPU
147 * executing it contains RCU usage regarded as invalid in the idle
148 * context, so tell RCU about that.
152 * The state used here cannot be a "coupled" one, because the "coupled"
153 * cpuidle mechanism enables interrupts and doing that with timekeeping
154 * suspended is generally unsafe.
156 stop_critical_timings();
157 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
159 target_state->enter_s2idle(dev, drv, index);
160 if (WARN_ON_ONCE(!irqs_disabled()))
163 * timekeeping_resume() that will be called by tick_unfreeze() for the
164 * first CPU executing it calls functions containing RCU read-side
165 * critical sections, so tell RCU about that.
167 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
170 start_critical_timings();
172 time_end = ns_to_ktime(local_clock());
174 dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
175 dev->states_usage[index].s2idle_usage++;
179 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
180 * @drv: cpuidle driver for the given CPU.
181 * @dev: cpuidle device for the given CPU.
183 * If there are states with the ->enter_s2idle callback, find the deepest of
184 * them and enter it with frozen tick.
186 int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
191 * Find the deepest state with ->enter_s2idle present, which guarantees
192 * that interrupts won't be enabled when it exits and allows the tick to
195 index = find_deepest_state(drv, dev, U64_MAX, 0, true);
197 enter_s2idle_proper(drv, dev, index);
202 #endif /* CONFIG_SUSPEND */
205 * cpuidle_enter_state - enter the state and update stats
206 * @dev: cpuidle device for this cpu
207 * @drv: cpuidle driver for this cpu
208 * @index: index into the states table in @drv of the state to enter
210 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
215 struct cpuidle_state *target_state = &drv->states[index];
216 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
217 ktime_t time_start, time_end;
220 * Tell the time framework to switch to a broadcast timer because our
221 * local timer will be shut down. If a local timer is used from another
222 * CPU as a broadcast timer, this call may fail if it is not available.
224 if (broadcast && tick_broadcast_enter()) {
225 index = find_deepest_state(drv, dev, target_state->exit_latency_ns,
226 CPUIDLE_FLAG_TIMER_STOP, false);
231 target_state = &drv->states[index];
235 if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
238 /* Take note of the planned idle state. */
239 sched_idle_set_state(target_state);
241 trace_cpu_idle(index, dev->cpu);
242 time_start = ns_to_ktime(local_clock());
244 stop_critical_timings();
245 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
247 entered_state = target_state->enter(dev, drv, index);
248 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
250 start_critical_timings();
252 sched_clock_idle_wakeup_event();
253 time_end = ns_to_ktime(local_clock());
254 trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
256 /* The cpu is no longer idle or about to enter idle. */
257 sched_idle_set_state(NULL);
260 if (WARN_ON_ONCE(!irqs_disabled()))
263 tick_broadcast_exit();
266 if (!cpuidle_state_is_coupled(drv, index))
269 if (entered_state >= 0) {
270 s64 diff, delay = drv->states[entered_state].exit_latency_ns;
274 * Update cpuidle counters
275 * This can be moved to within driver enter routine,
276 * but that results in multiple copies of same code.
278 diff = ktime_sub(time_end, time_start);
280 dev->last_residency_ns = diff;
281 dev->states_usage[entered_state].time_ns += diff;
282 dev->states_usage[entered_state].usage++;
284 if (diff < drv->states[entered_state].target_residency_ns) {
285 for (i = entered_state - 1; i >= 0; i--) {
286 if (dev->states_usage[i].disable)
289 /* Shallower states are enabled, so update. */
290 dev->states_usage[entered_state].above++;
293 } else if (diff > delay) {
294 for (i = entered_state + 1; i < drv->state_count; i++) {
295 if (dev->states_usage[i].disable)
299 * Update if a deeper state would have been a
300 * better match for the observed idle duration.
302 if (diff - delay >= drv->states[i].target_residency_ns)
303 dev->states_usage[entered_state].below++;
309 dev->last_residency_ns = 0;
312 return entered_state;
316 * cpuidle_select - ask the cpuidle framework to choose an idle state
318 * @drv: the cpuidle driver
319 * @dev: the cpuidle device
320 * @stop_tick: indication on whether or not to stop the tick
322 * Returns the index of the idle state. The return value must not be negative.
324 * The memory location pointed to by @stop_tick is expected to be written the
325 * 'false' boolean value if the scheduler tick should not be stopped before
326 * entering the returned state.
328 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
331 return cpuidle_curr_governor->select(drv, dev, stop_tick);
335 * cpuidle_enter - enter into the specified idle state
337 * @drv: the cpuidle driver tied with the cpu
338 * @dev: the cpuidle device
339 * @index: the index in the idle state table
341 * Returns the index in the idle state, < 0 in case of error.
342 * The error code depends on the backend driver
344 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
350 * Store the next hrtimer, which becomes either next tick or the next
351 * timer event, whatever expires first. Additionally, to make this data
352 * useful for consumers outside cpuidle, we rely on that the governor's
353 * ->select() callback have decided, whether to stop the tick or not.
355 WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
357 if (cpuidle_state_is_coupled(drv, index))
358 ret = cpuidle_enter_state_coupled(dev, drv, index);
360 ret = cpuidle_enter_state(dev, drv, index);
362 WRITE_ONCE(dev->next_hrtimer, 0);
367 * cpuidle_reflect - tell the underlying governor what was the state
370 * @dev : the cpuidle device
371 * @index: the index in the idle state table
374 void cpuidle_reflect(struct cpuidle_device *dev, int index)
376 if (cpuidle_curr_governor->reflect && index >= 0)
377 cpuidle_curr_governor->reflect(dev, index);
381 * cpuidle_poll_time - return amount of time to poll for,
382 * governors can override dev->poll_limit_ns if necessary
384 * @drv: the cpuidle driver tied with the cpu
385 * @dev: the cpuidle device
388 u64 cpuidle_poll_time(struct cpuidle_driver *drv,
389 struct cpuidle_device *dev)
394 if (dev->poll_limit_ns)
395 return dev->poll_limit_ns;
397 limit_ns = TICK_NSEC;
398 for (i = 1; i < drv->state_count; i++) {
399 if (dev->states_usage[i].disable)
402 limit_ns = drv->states[i].target_residency_ns;
406 dev->poll_limit_ns = limit_ns;
408 return dev->poll_limit_ns;
412 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
414 void cpuidle_install_idle_handler(void)
416 if (enabled_devices) {
417 /* Make sure all changes finished before we switch to new idle */
424 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
426 void cpuidle_uninstall_idle_handler(void)
428 if (enabled_devices) {
430 wake_up_all_idle_cpus();
434 * Make sure external observers (such as the scheduler)
435 * are done looking at pointed idle states.
441 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
443 void cpuidle_pause_and_lock(void)
445 mutex_lock(&cpuidle_lock);
446 cpuidle_uninstall_idle_handler();
449 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
452 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
454 void cpuidle_resume_and_unlock(void)
456 cpuidle_install_idle_handler();
457 mutex_unlock(&cpuidle_lock);
460 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
462 /* Currently used in suspend/resume path to suspend cpuidle */
463 void cpuidle_pause(void)
465 mutex_lock(&cpuidle_lock);
466 cpuidle_uninstall_idle_handler();
467 mutex_unlock(&cpuidle_lock);
470 /* Currently used in suspend/resume path to resume cpuidle */
471 void cpuidle_resume(void)
473 mutex_lock(&cpuidle_lock);
474 cpuidle_install_idle_handler();
475 mutex_unlock(&cpuidle_lock);
479 * cpuidle_enable_device - enables idle PM for a CPU
482 * This function must be called between cpuidle_pause_and_lock and
483 * cpuidle_resume_and_unlock when used externally.
485 int cpuidle_enable_device(struct cpuidle_device *dev)
488 struct cpuidle_driver *drv;
496 if (!cpuidle_curr_governor)
499 drv = cpuidle_get_cpu_driver(dev);
504 if (!dev->registered)
507 ret = cpuidle_add_device_sysfs(dev);
511 if (cpuidle_curr_governor->enable) {
512 ret = cpuidle_curr_governor->enable(drv, dev);
525 cpuidle_remove_device_sysfs(dev);
530 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
533 * cpuidle_disable_device - disables idle PM for a CPU
536 * This function must be called between cpuidle_pause_and_lock and
537 * cpuidle_resume_and_unlock when used externally.
539 void cpuidle_disable_device(struct cpuidle_device *dev)
541 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
543 if (!dev || !dev->enabled)
546 if (!drv || !cpuidle_curr_governor)
551 if (cpuidle_curr_governor->disable)
552 cpuidle_curr_governor->disable(drv, dev);
554 cpuidle_remove_device_sysfs(dev);
558 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
560 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
562 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
564 list_del(&dev->device_list);
565 per_cpu(cpuidle_devices, dev->cpu) = NULL;
566 module_put(drv->owner);
571 static void __cpuidle_device_init(struct cpuidle_device *dev)
573 memset(dev->states_usage, 0, sizeof(dev->states_usage));
574 dev->last_residency_ns = 0;
575 dev->next_hrtimer = 0;
579 * __cpuidle_register_device - internal register function called before register
580 * and enable routines
583 * cpuidle_lock mutex must be held before this is called
585 static int __cpuidle_register_device(struct cpuidle_device *dev)
587 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
590 if (!try_module_get(drv->owner))
593 for (i = 0; i < drv->state_count; i++) {
594 if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
595 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
597 if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
598 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
601 per_cpu(cpuidle_devices, dev->cpu) = dev;
602 list_add(&dev->device_list, &cpuidle_detected_devices);
604 ret = cpuidle_coupled_register_device(dev);
606 __cpuidle_unregister_device(dev);
614 * cpuidle_register_device - registers a CPU's idle PM feature
617 int cpuidle_register_device(struct cpuidle_device *dev)
624 mutex_lock(&cpuidle_lock);
629 __cpuidle_device_init(dev);
631 ret = __cpuidle_register_device(dev);
635 ret = cpuidle_add_sysfs(dev);
639 ret = cpuidle_enable_device(dev);
643 cpuidle_install_idle_handler();
646 mutex_unlock(&cpuidle_lock);
651 cpuidle_remove_sysfs(dev);
653 __cpuidle_unregister_device(dev);
657 EXPORT_SYMBOL_GPL(cpuidle_register_device);
660 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
663 void cpuidle_unregister_device(struct cpuidle_device *dev)
665 if (!dev || dev->registered == 0)
668 cpuidle_pause_and_lock();
670 cpuidle_disable_device(dev);
672 cpuidle_remove_sysfs(dev);
674 __cpuidle_unregister_device(dev);
676 cpuidle_coupled_unregister_device(dev);
678 cpuidle_resume_and_unlock();
681 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
684 * cpuidle_unregister: unregister a driver and the devices. This function
685 * can be used only if the driver has been previously registered through
686 * the cpuidle_register function.
688 * @drv: a valid pointer to a struct cpuidle_driver
690 void cpuidle_unregister(struct cpuidle_driver *drv)
693 struct cpuidle_device *device;
695 for_each_cpu(cpu, drv->cpumask) {
696 device = &per_cpu(cpuidle_dev, cpu);
697 cpuidle_unregister_device(device);
700 cpuidle_unregister_driver(drv);
702 EXPORT_SYMBOL_GPL(cpuidle_unregister);
705 * cpuidle_register: registers the driver and the cpu devices with the
706 * coupled_cpus passed as parameter. This function is used for all common
707 * initialization pattern there are in the arch specific drivers. The
708 * devices is globally defined in this file.
710 * @drv : a valid pointer to a struct cpuidle_driver
711 * @coupled_cpus: a cpumask for the coupled states
713 * Returns 0 on success, < 0 otherwise
715 int cpuidle_register(struct cpuidle_driver *drv,
716 const struct cpumask *const coupled_cpus)
719 struct cpuidle_device *device;
721 ret = cpuidle_register_driver(drv);
723 pr_err("failed to register cpuidle driver\n");
727 for_each_cpu(cpu, drv->cpumask) {
728 device = &per_cpu(cpuidle_dev, cpu);
731 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
733 * On multiplatform for ARM, the coupled idle states could be
734 * enabled in the kernel even if the cpuidle driver does not
735 * use it. Note, coupled_cpus is a struct copy.
738 device->coupled_cpus = *coupled_cpus;
740 ret = cpuidle_register_device(device);
744 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
746 cpuidle_unregister(drv);
752 EXPORT_SYMBOL_GPL(cpuidle_register);
755 * cpuidle_init - core initializer
757 static int __init cpuidle_init(void)
759 if (cpuidle_disabled())
762 return cpuidle_add_interface(cpu_subsys.dev_root);
765 module_param(off, int, 0444);
766 module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
767 core_initcall(cpuidle_init);