814 lines · c
1/*2 * cpuidle.c - core cpuidle infrastructure3 *4 * (C) 2006-2007 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>5 * Shaohua Li <shaohua.li@intel.com>6 * Adam Belay <abelay@novell.com>7 *8 * This code is licenced under the GPL.9 */10 11#include "linux/percpu-defs.h"12#include <linux/clockchips.h>13#include <linux/kernel.h>14#include <linux/mutex.h>15#include <linux/sched.h>16#include <linux/sched/clock.h>17#include <linux/sched/idle.h>18#include <linux/notifier.h>19#include <linux/pm_qos.h>20#include <linux/cpu.h>21#include <linux/cpuidle.h>22#include <linux/ktime.h>23#include <linux/hrtimer.h>24#include <linux/module.h>25#include <linux/suspend.h>26#include <linux/tick.h>27#include <linux/mmu_context.h>28#include <linux/context_tracking.h>29#include <trace/events/power.h>30 31#include "cpuidle.h"32 33DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);34DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);35 36DEFINE_MUTEX(cpuidle_lock);37LIST_HEAD(cpuidle_detected_devices);38 39static int enabled_devices;40static int off __read_mostly;41static int initialized __read_mostly;42 43int cpuidle_disabled(void)44{45 return off;46}47void disable_cpuidle(void)48{49 off = 1;50}51 52bool cpuidle_not_available(struct cpuidle_driver *drv,53 struct cpuidle_device *dev)54{55 return off || !initialized || !drv || !dev || !dev->enabled;56}57 58/**59 * cpuidle_play_dead - cpu off-lining60 *61 * Returns in case of an error or no driver62 */63int cpuidle_play_dead(void)64{65 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);66 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);67 int i;68 69 if (!drv)70 return -ENODEV;71 72 /* Find lowest-power state that supports long-term idle */73 for (i = drv->state_count - 1; i >= 0; i--)74 if (drv->states[i].enter_dead)75 return drv->states[i].enter_dead(dev, i);76 77 return -ENODEV;78}79 80static int find_deepest_state(struct cpuidle_driver *drv,81 struct cpuidle_device *dev,82 u64 max_latency_ns,83 unsigned int forbidden_flags,84 bool s2idle)85{86 u64 latency_req = 0;87 int i, ret = 0;88 89 for (i = 1; i < drv->state_count; i++) {90 struct cpuidle_state *s = &drv->states[i];91 92 if (dev->states_usage[i].disable ||93 s->exit_latency_ns <= latency_req ||94 s->exit_latency_ns > max_latency_ns ||95 (s->flags & forbidden_flags) ||96 (s2idle && !s->enter_s2idle))97 continue;98 99 latency_req = s->exit_latency_ns;100 ret = i;101 }102 return ret;103}104 105/**106 * cpuidle_use_deepest_state - Set/unset governor override mode.107 * @latency_limit_ns: Idle state exit latency limit (or no override if 0).108 *109 * If @latency_limit_ns is nonzero, set the current CPU to use the deepest idle110 * state with exit latency within @latency_limit_ns (override governors going111 * forward), or do not override governors if it is zero.112 */113void cpuidle_use_deepest_state(u64 latency_limit_ns)114{115 struct cpuidle_device *dev;116 117 preempt_disable();118 dev = cpuidle_get_device();119 if (dev)120 dev->forced_idle_latency_limit_ns = latency_limit_ns;121 preempt_enable();122}123 124/**125 * cpuidle_find_deepest_state - Find the deepest available idle state.126 * @drv: cpuidle driver for the given CPU.127 * @dev: cpuidle device for the given CPU.128 * @latency_limit_ns: Idle state exit latency limit129 *130 * Return: the index of the deepest available idle state.131 */132int cpuidle_find_deepest_state(struct cpuidle_driver *drv,133 struct cpuidle_device *dev,134 u64 latency_limit_ns)135{136 return find_deepest_state(drv, dev, latency_limit_ns, 0, false);137}138 139#ifdef CONFIG_SUSPEND140static noinstr void enter_s2idle_proper(struct cpuidle_driver *drv,141 struct cpuidle_device *dev, int index)142{143 struct cpuidle_state *target_state = &drv->states[index];144 ktime_t time_start, time_end;145 146 instrumentation_begin();147 148 time_start = ns_to_ktime(local_clock_noinstr());149 150 tick_freeze();151 /*152 * The state used here cannot be a "coupled" one, because the "coupled"153 * cpuidle mechanism enables interrupts and doing that with timekeeping154 * suspended is generally unsafe.155 */156 stop_critical_timings();157 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {158 ct_cpuidle_enter();159 /* Annotate away the indirect call */160 instrumentation_begin();161 }162 target_state->enter_s2idle(dev, drv, index);163 if (WARN_ON_ONCE(!irqs_disabled()))164 raw_local_irq_disable();165 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {166 instrumentation_end();167 ct_cpuidle_exit();168 }169 tick_unfreeze();170 start_critical_timings();171 172 time_end = ns_to_ktime(local_clock_noinstr());173 174 dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);175 dev->states_usage[index].s2idle_usage++;176 instrumentation_end();177}178 179/**180 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.181 * @drv: cpuidle driver for the given CPU.182 * @dev: cpuidle device for the given CPU.183 *184 * If there are states with the ->enter_s2idle callback, find the deepest of185 * them and enter it with frozen tick.186 */187int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)188{189 int index;190 191 /*192 * Find the deepest state with ->enter_s2idle present, which guarantees193 * that interrupts won't be enabled when it exits and allows the tick to194 * be frozen safely.195 */196 index = find_deepest_state(drv, dev, U64_MAX, 0, true);197 if (index > 0) {198 enter_s2idle_proper(drv, dev, index);199 local_irq_enable();200 }201 return index;202}203#endif /* CONFIG_SUSPEND */204 205/**206 * cpuidle_enter_state - enter the state and update stats207 * @dev: cpuidle device for this cpu208 * @drv: cpuidle driver for this cpu209 * @index: index into the states table in @drv of the state to enter210 */211noinstr int cpuidle_enter_state(struct cpuidle_device *dev,212 struct cpuidle_driver *drv,213 int index)214{215 int entered_state;216 217 struct cpuidle_state *target_state = &drv->states[index];218 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);219 ktime_t time_start, time_end;220 221 instrumentation_begin();222 223 /*224 * Tell the time framework to switch to a broadcast timer because our225 * local timer will be shut down. If a local timer is used from another226 * CPU as a broadcast timer, this call may fail if it is not available.227 */228 if (broadcast && tick_broadcast_enter()) {229 index = find_deepest_state(drv, dev, target_state->exit_latency_ns,230 CPUIDLE_FLAG_TIMER_STOP, false);231 232 target_state = &drv->states[index];233 broadcast = false;234 }235 236 if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)237 leave_mm();238 239 /* Take note of the planned idle state. */240 sched_idle_set_state(target_state);241 242 trace_cpu_idle(index, dev->cpu);243 time_start = ns_to_ktime(local_clock_noinstr());244 245 stop_critical_timings();246 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {247 ct_cpuidle_enter();248 /* Annotate away the indirect call */249 instrumentation_begin();250 }251 252 /*253 * NOTE!!254 *255 * For cpuidle_state::enter() methods that do *NOT* set256 * CPUIDLE_FLAG_RCU_IDLE RCU will be disabled here and these functions257 * must be marked either noinstr or __cpuidle.258 *259 * For cpuidle_state::enter() methods that *DO* set260 * CPUIDLE_FLAG_RCU_IDLE this isn't required, but they must mark the261 * function calling ct_cpuidle_enter() as noinstr/__cpuidle and all262 * functions called within the RCU-idle region.263 */264 entered_state = target_state->enter(dev, drv, index);265 266 if (WARN_ONCE(!irqs_disabled(), "%ps leaked IRQ state", target_state->enter))267 raw_local_irq_disable();268 269 if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {270 instrumentation_end();271 ct_cpuidle_exit();272 }273 start_critical_timings();274 275 sched_clock_idle_wakeup_event();276 time_end = ns_to_ktime(local_clock_noinstr());277 trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);278 279 /* The cpu is no longer idle or about to enter idle. */280 sched_idle_set_state(NULL);281 282 if (broadcast)283 tick_broadcast_exit();284 285 if (!cpuidle_state_is_coupled(drv, index))286 local_irq_enable();287 288 if (entered_state >= 0) {289 s64 diff, delay = drv->states[entered_state].exit_latency_ns;290 int i;291 292 /*293 * Update cpuidle counters294 * This can be moved to within driver enter routine,295 * but that results in multiple copies of same code.296 */297 diff = ktime_sub(time_end, time_start);298 299 dev->last_residency_ns = diff;300 dev->states_usage[entered_state].time_ns += diff;301 dev->states_usage[entered_state].usage++;302 303 if (diff < drv->states[entered_state].target_residency_ns) {304 for (i = entered_state - 1; i >= 0; i--) {305 if (dev->states_usage[i].disable)306 continue;307 308 /* Shallower states are enabled, so update. */309 dev->states_usage[entered_state].above++;310 trace_cpu_idle_miss(dev->cpu, entered_state, false);311 break;312 }313 } else if (diff > delay) {314 for (i = entered_state + 1; i < drv->state_count; i++) {315 if (dev->states_usage[i].disable)316 continue;317 318 /*319 * Update if a deeper state would have been a320 * better match for the observed idle duration.321 */322 if (diff - delay >= drv->states[i].target_residency_ns) {323 dev->states_usage[entered_state].below++;324 trace_cpu_idle_miss(dev->cpu, entered_state, true);325 }326 327 break;328 }329 }330 } else {331 dev->last_residency_ns = 0;332 dev->states_usage[index].rejected++;333 }334 335 instrumentation_end();336 337 return entered_state;338}339 340/**341 * cpuidle_select - ask the cpuidle framework to choose an idle state342 *343 * @drv: the cpuidle driver344 * @dev: the cpuidle device345 * @stop_tick: indication on whether or not to stop the tick346 *347 * Returns the index of the idle state. The return value must not be negative.348 *349 * The memory location pointed to by @stop_tick is expected to be written the350 * 'false' boolean value if the scheduler tick should not be stopped before351 * entering the returned state.352 */353int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,354 bool *stop_tick)355{356 return cpuidle_curr_governor->select(drv, dev, stop_tick);357}358 359/**360 * cpuidle_enter - enter into the specified idle state361 *362 * @drv: the cpuidle driver tied with the cpu363 * @dev: the cpuidle device364 * @index: the index in the idle state table365 *366 * Returns the index in the idle state, < 0 in case of error.367 * The error code depends on the backend driver368 */369int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,370 int index)371{372 int ret = 0;373 374 /*375 * Store the next hrtimer, which becomes either next tick or the next376 * timer event, whatever expires first. Additionally, to make this data377 * useful for consumers outside cpuidle, we rely on that the governor's378 * ->select() callback have decided, whether to stop the tick or not.379 */380 WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());381 382 if (cpuidle_state_is_coupled(drv, index))383 ret = cpuidle_enter_state_coupled(dev, drv, index);384 else385 ret = cpuidle_enter_state(dev, drv, index);386 387 WRITE_ONCE(dev->next_hrtimer, 0);388 return ret;389}390 391/**392 * cpuidle_reflect - tell the underlying governor what was the state393 * we were in394 *395 * @dev : the cpuidle device396 * @index: the index in the idle state table397 *398 */399void cpuidle_reflect(struct cpuidle_device *dev, int index)400{401 if (cpuidle_curr_governor->reflect && index >= 0)402 cpuidle_curr_governor->reflect(dev, index);403}404 405/*406 * Min polling interval of 10usec is a guess. It is assuming that407 * for most users, the time for a single ping-pong workload like408 * perf bench pipe would generally complete within 10usec but409 * this is hardware dependant. Actual time can be estimated with410 *411 * perf bench sched pipe -l 10000412 *413 * Run multiple times to avoid cpufreq effects.414 */415#define CPUIDLE_POLL_MIN 10000416#define CPUIDLE_POLL_MAX (TICK_NSEC / 16)417 418/**419 * cpuidle_poll_time - return amount of time to poll for,420 * governors can override dev->poll_limit_ns if necessary421 *422 * @drv: the cpuidle driver tied with the cpu423 * @dev: the cpuidle device424 *425 */426__cpuidle u64 cpuidle_poll_time(struct cpuidle_driver *drv,427 struct cpuidle_device *dev)428{429 int i;430 u64 limit_ns;431 432 BUILD_BUG_ON(CPUIDLE_POLL_MIN > CPUIDLE_POLL_MAX);433 434 if (dev->poll_limit_ns)435 return dev->poll_limit_ns;436 437 limit_ns = CPUIDLE_POLL_MAX;438 for (i = 1; i < drv->state_count; i++) {439 u64 state_limit;440 441 if (dev->states_usage[i].disable)442 continue;443 444 state_limit = drv->states[i].target_residency_ns;445 if (state_limit < CPUIDLE_POLL_MIN)446 continue;447 448 limit_ns = min_t(u64, state_limit, CPUIDLE_POLL_MAX);449 break;450 }451 452 dev->poll_limit_ns = limit_ns;453 454 return dev->poll_limit_ns;455}456 457/**458 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler459 */460void cpuidle_install_idle_handler(void)461{462 if (enabled_devices) {463 /* Make sure all changes finished before we switch to new idle */464 smp_wmb();465 initialized = 1;466 }467}468 469/**470 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler471 */472void cpuidle_uninstall_idle_handler(void)473{474 if (enabled_devices) {475 initialized = 0;476 wake_up_all_idle_cpus();477 }478 479 /*480 * Make sure external observers (such as the scheduler)481 * are done looking at pointed idle states.482 */483 synchronize_rcu();484}485 486/**487 * cpuidle_pause_and_lock - temporarily disables CPUIDLE488 */489void cpuidle_pause_and_lock(void)490{491 mutex_lock(&cpuidle_lock);492 cpuidle_uninstall_idle_handler();493}494 495EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);496 497/**498 * cpuidle_resume_and_unlock - resumes CPUIDLE operation499 */500void cpuidle_resume_and_unlock(void)501{502 cpuidle_install_idle_handler();503 mutex_unlock(&cpuidle_lock);504}505 506EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);507 508/* Currently used in suspend/resume path to suspend cpuidle */509void cpuidle_pause(void)510{511 mutex_lock(&cpuidle_lock);512 cpuidle_uninstall_idle_handler();513 mutex_unlock(&cpuidle_lock);514}515 516/* Currently used in suspend/resume path to resume cpuidle */517void cpuidle_resume(void)518{519 mutex_lock(&cpuidle_lock);520 cpuidle_install_idle_handler();521 mutex_unlock(&cpuidle_lock);522}523 524/**525 * cpuidle_enable_device - enables idle PM for a CPU526 * @dev: the CPU527 *528 * This function must be called between cpuidle_pause_and_lock and529 * cpuidle_resume_and_unlock when used externally.530 */531int cpuidle_enable_device(struct cpuidle_device *dev)532{533 int ret;534 struct cpuidle_driver *drv;535 536 if (!dev)537 return -EINVAL;538 539 if (dev->enabled)540 return 0;541 542 if (!cpuidle_curr_governor)543 return -EIO;544 545 drv = cpuidle_get_cpu_driver(dev);546 547 if (!drv)548 return -EIO;549 550 if (!dev->registered)551 return -EINVAL;552 553 ret = cpuidle_add_device_sysfs(dev);554 if (ret)555 return ret;556 557 if (cpuidle_curr_governor->enable) {558 ret = cpuidle_curr_governor->enable(drv, dev);559 if (ret)560 goto fail_sysfs;561 }562 563 smp_wmb();564 565 dev->enabled = 1;566 567 enabled_devices++;568 return 0;569 570fail_sysfs:571 cpuidle_remove_device_sysfs(dev);572 573 return ret;574}575 576EXPORT_SYMBOL_GPL(cpuidle_enable_device);577 578/**579 * cpuidle_disable_device - disables idle PM for a CPU580 * @dev: the CPU581 *582 * This function must be called between cpuidle_pause_and_lock and583 * cpuidle_resume_and_unlock when used externally.584 */585void cpuidle_disable_device(struct cpuidle_device *dev)586{587 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);588 589 if (!dev || !dev->enabled)590 return;591 592 if (!drv || !cpuidle_curr_governor)593 return;594 595 dev->enabled = 0;596 597 if (cpuidle_curr_governor->disable)598 cpuidle_curr_governor->disable(drv, dev);599 600 cpuidle_remove_device_sysfs(dev);601 enabled_devices--;602}603 604EXPORT_SYMBOL_GPL(cpuidle_disable_device);605 606static void __cpuidle_unregister_device(struct cpuidle_device *dev)607{608 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);609 610 list_del(&dev->device_list);611 per_cpu(cpuidle_devices, dev->cpu) = NULL;612 module_put(drv->owner);613 614 dev->registered = 0;615}616 617static void __cpuidle_device_init(struct cpuidle_device *dev)618{619 memset(dev->states_usage, 0, sizeof(dev->states_usage));620 dev->last_residency_ns = 0;621 dev->next_hrtimer = 0;622}623 624/**625 * __cpuidle_register_device - internal register function called before register626 * and enable routines627 * @dev: the cpu628 *629 * cpuidle_lock mutex must be held before this is called630 */631static int __cpuidle_register_device(struct cpuidle_device *dev)632{633 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);634 int i, ret;635 636 if (!try_module_get(drv->owner))637 return -EINVAL;638 639 for (i = 0; i < drv->state_count; i++) {640 if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)641 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;642 643 if (drv->states[i].flags & CPUIDLE_FLAG_OFF)644 dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;645 }646 647 per_cpu(cpuidle_devices, dev->cpu) = dev;648 list_add(&dev->device_list, &cpuidle_detected_devices);649 650 ret = cpuidle_coupled_register_device(dev);651 if (ret)652 __cpuidle_unregister_device(dev);653 else654 dev->registered = 1;655 656 return ret;657}658 659/**660 * cpuidle_register_device - registers a CPU's idle PM feature661 * @dev: the cpu662 */663int cpuidle_register_device(struct cpuidle_device *dev)664{665 int ret = -EBUSY;666 667 if (!dev)668 return -EINVAL;669 670 mutex_lock(&cpuidle_lock);671 672 if (dev->registered)673 goto out_unlock;674 675 __cpuidle_device_init(dev);676 677 ret = __cpuidle_register_device(dev);678 if (ret)679 goto out_unlock;680 681 ret = cpuidle_add_sysfs(dev);682 if (ret)683 goto out_unregister;684 685 ret = cpuidle_enable_device(dev);686 if (ret)687 goto out_sysfs;688 689 cpuidle_install_idle_handler();690 691out_unlock:692 mutex_unlock(&cpuidle_lock);693 694 return ret;695 696out_sysfs:697 cpuidle_remove_sysfs(dev);698out_unregister:699 __cpuidle_unregister_device(dev);700 goto out_unlock;701}702 703EXPORT_SYMBOL_GPL(cpuidle_register_device);704 705/**706 * cpuidle_unregister_device - unregisters a CPU's idle PM feature707 * @dev: the cpu708 */709void cpuidle_unregister_device(struct cpuidle_device *dev)710{711 if (!dev || dev->registered == 0)712 return;713 714 cpuidle_pause_and_lock();715 716 cpuidle_disable_device(dev);717 718 cpuidle_remove_sysfs(dev);719 720 __cpuidle_unregister_device(dev);721 722 cpuidle_coupled_unregister_device(dev);723 724 cpuidle_resume_and_unlock();725}726 727EXPORT_SYMBOL_GPL(cpuidle_unregister_device);728 729/**730 * cpuidle_unregister: unregister a driver and the devices. This function731 * can be used only if the driver has been previously registered through732 * the cpuidle_register function.733 *734 * @drv: a valid pointer to a struct cpuidle_driver735 */736void cpuidle_unregister(struct cpuidle_driver *drv)737{738 int cpu;739 struct cpuidle_device *device;740 741 for_each_cpu(cpu, drv->cpumask) {742 device = &per_cpu(cpuidle_dev, cpu);743 cpuidle_unregister_device(device);744 }745 746 cpuidle_unregister_driver(drv);747}748EXPORT_SYMBOL_GPL(cpuidle_unregister);749 750/**751 * cpuidle_register: registers the driver and the cpu devices with the752 * coupled_cpus passed as parameter. This function is used for all common753 * initialization pattern there are in the arch specific drivers. The754 * devices is globally defined in this file.755 *756 * @drv : a valid pointer to a struct cpuidle_driver757 * @coupled_cpus: a cpumask for the coupled states758 *759 * Returns 0 on success, < 0 otherwise760 */761int cpuidle_register(struct cpuidle_driver *drv,762 const struct cpumask *const coupled_cpus)763{764 int ret, cpu;765 struct cpuidle_device *device;766 767 ret = cpuidle_register_driver(drv);768 if (ret) {769 pr_err("failed to register cpuidle driver\n");770 return ret;771 }772 773 for_each_cpu(cpu, drv->cpumask) {774 device = &per_cpu(cpuidle_dev, cpu);775 device->cpu = cpu;776 777#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED778 /*779 * On multiplatform for ARM, the coupled idle states could be780 * enabled in the kernel even if the cpuidle driver does not781 * use it. Note, coupled_cpus is a struct copy.782 */783 if (coupled_cpus)784 device->coupled_cpus = *coupled_cpus;785#endif786 ret = cpuidle_register_device(device);787 if (!ret)788 continue;789 790 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);791 792 cpuidle_unregister(drv);793 break;794 }795 796 return ret;797}798EXPORT_SYMBOL_GPL(cpuidle_register);799 800/**801 * cpuidle_init - core initializer802 */803static int __init cpuidle_init(void)804{805 if (cpuidle_disabled())806 return -ENODEV;807 808 return cpuidle_add_interface();809}810 811module_param(off, int, 0444);812module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);813core_initcall(cpuidle_init);814