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());
147 * The state used here cannot be a "coupled" one, because the "coupled"
148 * cpuidle mechanism enables interrupts and doing that with timekeeping
149 * suspended is generally unsafe.
151 stop_critical_timings();
152 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
154 target_state->enter_s2idle(dev, drv, index);
155 if (WARN_ON_ONCE(!irqs_disabled()))
157 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
160 start_critical_timings();
162 time_end = ns_to_ktime(local_clock());
164 dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
165 dev->states_usage[index].s2idle_usage++;
169 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
170 * @drv: cpuidle driver for the given CPU.
171 * @dev: cpuidle device for the given CPU.
173 * If there are states with the ->enter_s2idle callback, find the deepest of
174 * them and enter it with frozen tick.
176 int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
181 * Find the deepest state with ->enter_s2idle present, which guarantees
182 * that interrupts won't be enabled when it exits and allows the tick to
185 index = find_deepest_state(drv, dev, U64_MAX, 0, true);
187 enter_s2idle_proper(drv, dev, index);
192 #endif /* CONFIG_SUSPEND */
195 * cpuidle_enter_state - enter the state and update stats
196 * @dev: cpuidle device for this cpu
197 * @drv: cpuidle driver for this cpu
198 * @index: index into the states table in @drv of the state to enter
200 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
205 struct cpuidle_state *target_state = &drv->states[index];
206 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
207 ktime_t time_start, time_end;
210 * Tell the time framework to switch to a broadcast timer because our
211 * local timer will be shut down. If a local timer is used from another
212 * CPU as a broadcast timer, this call may fail if it is not available.
214 if (broadcast && tick_broadcast_enter()) {
215 index = find_deepest_state(drv, dev, target_state->exit_latency_ns,
216 CPUIDLE_FLAG_TIMER_STOP, false);
221 target_state = &drv->states[index];
225 if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
228 /* Take note of the planned idle state. */
229 sched_idle_set_state(target_state);
231 trace_cpu_idle(index, dev->cpu);
232 time_start = ns_to_ktime(local_clock());
234 stop_critical_timings();
235 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
237 entered_state = target_state->enter(dev, drv, index);
238 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
240 start_critical_timings();
242 sched_clock_idle_wakeup_event();
243 time_end = ns_to_ktime(local_clock());
244 trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
246 /* The cpu is no longer idle or about to enter idle. */
247 sched_idle_set_state(NULL);
250 if (WARN_ON_ONCE(!irqs_disabled()))
253 tick_broadcast_exit();
256 if (!cpuidle_state_is_coupled(drv, index))
259 if (entered_state >= 0) {
260 s64 diff, delay = drv->states[entered_state].exit_latency_ns;
264 * Update cpuidle counters
265 * This can be moved to within driver enter routine,
266 * but that results in multiple copies of same code.
268 diff = ktime_sub(time_end, time_start);
270 dev->last_residency_ns = diff;
271 dev->states_usage[entered_state].time_ns += diff;
272 dev->states_usage[entered_state].usage++;
274 if (diff < drv->states[entered_state].target_residency_ns) {
275 for (i = entered_state - 1; i >= 0; i--) {
276 if (dev->states_usage[i].disable)
279 /* Shallower states are enabled, so update. */
280 dev->states_usage[entered_state].above++;
283 } else if (diff > delay) {
284 for (i = entered_state + 1; i < drv->state_count; i++) {
285 if (dev->states_usage[i].disable)
289 * Update if a deeper state would have been a
290 * better match for the observed idle duration.
292 if (diff - delay >= drv->states[i].target_residency_ns)
293 dev->states_usage[entered_state].below++;
299 dev->last_residency_ns = 0;
300 dev->states_usage[index].rejected++;
303 return entered_state;
307 * cpuidle_select - ask the cpuidle framework to choose an idle state
309 * @drv: the cpuidle driver
310 * @dev: the cpuidle device
311 * @stop_tick: indication on whether or not to stop the tick
313 * Returns the index of the idle state. The return value must not be negative.
315 * The memory location pointed to by @stop_tick is expected to be written the
316 * 'false' boolean value if the scheduler tick should not be stopped before
317 * entering the returned state.
319 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
322 return cpuidle_curr_governor->select(drv, dev, stop_tick);
326 * cpuidle_enter - enter into the specified idle state
328 * @drv: the cpuidle driver tied with the cpu
329 * @dev: the cpuidle device
330 * @index: the index in the idle state table
332 * Returns the index in the idle state, < 0 in case of error.
333 * The error code depends on the backend driver
335 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
341 * Store the next hrtimer, which becomes either next tick or the next
342 * timer event, whatever expires first. Additionally, to make this data
343 * useful for consumers outside cpuidle, we rely on that the governor's
344 * ->select() callback have decided, whether to stop the tick or not.
346 WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
348 if (cpuidle_state_is_coupled(drv, index))
349 ret = cpuidle_enter_state_coupled(dev, drv, index);
351 ret = cpuidle_enter_state(dev, drv, index);
353 WRITE_ONCE(dev->next_hrtimer, 0);
358 * cpuidle_reflect - tell the underlying governor what was the state
361 * @dev : the cpuidle device
362 * @index: the index in the idle state table
365 void cpuidle_reflect(struct cpuidle_device *dev, int index)
367 if (cpuidle_curr_governor->reflect && index >= 0)
368 cpuidle_curr_governor->reflect(dev, index);
372 * Min polling interval of 10usec is a guess. It is assuming that
373 * for most users, the time for a single ping-pong workload like
374 * perf bench pipe would generally complete within 10usec but
375 * this is hardware dependant. Actual time can be estimated with
377 * perf bench sched pipe -l 10000
379 * Run multiple times to avoid cpufreq effects.
381 #define CPUIDLE_POLL_MIN 10000
382 #define CPUIDLE_POLL_MAX (TICK_NSEC / 16)
385 * cpuidle_poll_time - return amount of time to poll for,
386 * governors can override dev->poll_limit_ns if necessary
388 * @drv: the cpuidle driver tied with the cpu
389 * @dev: the cpuidle device
392 u64 cpuidle_poll_time(struct cpuidle_driver *drv,
393 struct cpuidle_device *dev)
398 BUILD_BUG_ON(CPUIDLE_POLL_MIN > CPUIDLE_POLL_MAX);
400 if (dev->poll_limit_ns)
401 return dev->poll_limit_ns;
403 limit_ns = CPUIDLE_POLL_MAX;
404 for (i = 1; i < drv->state_count; i++) {
407 if (dev->states_usage[i].disable)
410 state_limit = drv->states[i].target_residency_ns;
411 if (state_limit < CPUIDLE_POLL_MIN)
414 limit_ns = min_t(u64, state_limit, CPUIDLE_POLL_MAX);
418 dev->poll_limit_ns = limit_ns;
420 return dev->poll_limit_ns;
424 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
426 void cpuidle_install_idle_handler(void)
428 if (enabled_devices) {
429 /* Make sure all changes finished before we switch to new idle */
436 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
438 void cpuidle_uninstall_idle_handler(void)
440 if (enabled_devices) {
442 wake_up_all_idle_cpus();
446 * Make sure external observers (such as the scheduler)
447 * are done looking at pointed idle states.
453 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
455 void cpuidle_pause_and_lock(void)
457 mutex_lock(&cpuidle_lock);
458 cpuidle_uninstall_idle_handler();
461 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
464 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
466 void cpuidle_resume_and_unlock(void)
468 cpuidle_install_idle_handler();
469 mutex_unlock(&cpuidle_lock);
472 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
474 /* Currently used in suspend/resume path to suspend cpuidle */
475 void cpuidle_pause(void)
477 mutex_lock(&cpuidle_lock);
478 cpuidle_uninstall_idle_handler();
479 mutex_unlock(&cpuidle_lock);
482 /* Currently used in suspend/resume path to resume cpuidle */
483 void cpuidle_resume(void)
485 mutex_lock(&cpuidle_lock);
486 cpuidle_install_idle_handler();
487 mutex_unlock(&cpuidle_lock);
491 * cpuidle_enable_device - enables idle PM for a CPU
494 * This function must be called between cpuidle_pause_and_lock and
495 * cpuidle_resume_and_unlock when used externally.
497 int cpuidle_enable_device(struct cpuidle_device *dev)
500 struct cpuidle_driver *drv;
508 if (!cpuidle_curr_governor)
511 drv = cpuidle_get_cpu_driver(dev);
516 if (!dev->registered)
519 ret = cpuidle_add_device_sysfs(dev);
523 if (cpuidle_curr_governor->enable) {
524 ret = cpuidle_curr_governor->enable(drv, dev);
537 cpuidle_remove_device_sysfs(dev);
542 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
545 * cpuidle_disable_device - disables idle PM for a CPU
548 * This function must be called between cpuidle_pause_and_lock and
549 * cpuidle_resume_and_unlock when used externally.
551 void cpuidle_disable_device(struct cpuidle_device *dev)
553 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
555 if (!dev || !dev->enabled)
558 if (!drv || !cpuidle_curr_governor)
563 if (cpuidle_curr_governor->disable)
564 cpuidle_curr_governor->disable(drv, dev);
566 cpuidle_remove_device_sysfs(dev);
570 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
572 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
574 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
576 list_del(&dev->device_list);
577 per_cpu(cpuidle_devices, dev->cpu) = NULL;
578 module_put(drv->owner);
583 static void __cpuidle_device_init(struct cpuidle_device *dev)
585 memset(dev->states_usage, 0, sizeof(dev->states_usage));
586 dev->last_residency_ns = 0;
587 dev->next_hrtimer = 0;
591 * __cpuidle_register_device - internal register function called before register
592 * and enable routines
595 * cpuidle_lock mutex must be held before this is called
597 static int __cpuidle_register_device(struct cpuidle_device *dev)
599 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
602 if (!try_module_get(drv->owner))
605 for (i = 0; i < drv->state_count; i++) {
606 if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
607 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
609 if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
610 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
613 per_cpu(cpuidle_devices, dev->cpu) = dev;
614 list_add(&dev->device_list, &cpuidle_detected_devices);
616 ret = cpuidle_coupled_register_device(dev);
618 __cpuidle_unregister_device(dev);
626 * cpuidle_register_device - registers a CPU's idle PM feature
629 int cpuidle_register_device(struct cpuidle_device *dev)
636 mutex_lock(&cpuidle_lock);
641 __cpuidle_device_init(dev);
643 ret = __cpuidle_register_device(dev);
647 ret = cpuidle_add_sysfs(dev);
651 ret = cpuidle_enable_device(dev);
655 cpuidle_install_idle_handler();
658 mutex_unlock(&cpuidle_lock);
663 cpuidle_remove_sysfs(dev);
665 __cpuidle_unregister_device(dev);
669 EXPORT_SYMBOL_GPL(cpuidle_register_device);
672 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
675 void cpuidle_unregister_device(struct cpuidle_device *dev)
677 if (!dev || dev->registered == 0)
680 cpuidle_pause_and_lock();
682 cpuidle_disable_device(dev);
684 cpuidle_remove_sysfs(dev);
686 __cpuidle_unregister_device(dev);
688 cpuidle_coupled_unregister_device(dev);
690 cpuidle_resume_and_unlock();
693 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
696 * cpuidle_unregister: unregister a driver and the devices. This function
697 * can be used only if the driver has been previously registered through
698 * the cpuidle_register function.
700 * @drv: a valid pointer to a struct cpuidle_driver
702 void cpuidle_unregister(struct cpuidle_driver *drv)
705 struct cpuidle_device *device;
707 for_each_cpu(cpu, drv->cpumask) {
708 device = &per_cpu(cpuidle_dev, cpu);
709 cpuidle_unregister_device(device);
712 cpuidle_unregister_driver(drv);
714 EXPORT_SYMBOL_GPL(cpuidle_unregister);
717 * cpuidle_register: registers the driver and the cpu devices with the
718 * coupled_cpus passed as parameter. This function is used for all common
719 * initialization pattern there are in the arch specific drivers. The
720 * devices is globally defined in this file.
722 * @drv : a valid pointer to a struct cpuidle_driver
723 * @coupled_cpus: a cpumask for the coupled states
725 * Returns 0 on success, < 0 otherwise
727 int cpuidle_register(struct cpuidle_driver *drv,
728 const struct cpumask *const coupled_cpus)
731 struct cpuidle_device *device;
733 ret = cpuidle_register_driver(drv);
735 pr_err("failed to register cpuidle driver\n");
739 for_each_cpu(cpu, drv->cpumask) {
740 device = &per_cpu(cpuidle_dev, cpu);
743 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
745 * On multiplatform for ARM, the coupled idle states could be
746 * enabled in the kernel even if the cpuidle driver does not
747 * use it. Note, coupled_cpus is a struct copy.
750 device->coupled_cpus = *coupled_cpus;
752 ret = cpuidle_register_device(device);
756 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
758 cpuidle_unregister(drv);
764 EXPORT_SYMBOL_GPL(cpuidle_register);
767 * cpuidle_init - core initializer
769 static int __init cpuidle_init(void)
771 if (cpuidle_disabled())
774 return cpuidle_add_interface(cpu_subsys.dev_root);
777 module_param(off, int, 0444);
778 module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
779 core_initcall(cpuidle_init);