linux/drivers/cpuidle/cpuidle.c
<<
>>
Prefs
   1/*
   2 * cpuidle.c - core cpuidle infrastructure
   3 *
   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/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>
  26
  27#include "cpuidle.h"
  28
  29DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
  30DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
  31
  32DEFINE_MUTEX(cpuidle_lock);
  33LIST_HEAD(cpuidle_detected_devices);
  34
  35static int enabled_devices;
  36static int off __read_mostly;
  37static int initialized __read_mostly;
  38
  39int cpuidle_disabled(void)
  40{
  41        return off;
  42}
  43void disable_cpuidle(void)
  44{
  45        off = 1;
  46}
  47
  48bool cpuidle_not_available(struct cpuidle_driver *drv,
  49                           struct cpuidle_device *dev)
  50{
  51        return off || !initialized || !drv || !dev || !dev->enabled;
  52}
  53
  54/**
  55 * cpuidle_play_dead - cpu off-lining
  56 *
  57 * Returns in case of an error or no driver
  58 */
  59int cpuidle_play_dead(void)
  60{
  61        struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  62        struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  63        int i;
  64
  65        if (!drv)
  66                return -ENODEV;
  67
  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);
  72
  73        return -ENODEV;
  74}
  75
  76static int find_deepest_state(struct cpuidle_driver *drv,
  77                              struct cpuidle_device *dev,
  78                              unsigned int max_latency,
  79                              unsigned int forbidden_flags,
  80                              bool s2idle)
  81{
  82        unsigned int latency_req = 0;
  83        int i, ret = 0;
  84
  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];
  88
  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))
  93                        continue;
  94
  95                latency_req = s->exit_latency;
  96                ret = i;
  97        }
  98        return ret;
  99}
 100
 101/**
 102 * cpuidle_use_deepest_state - Set/clear governor override flag.
 103 * @enable: New value of the flag.
 104 *
 105 * Set/unset the current CPU to use the deepest idle state (override governors
 106 * going forward if set).
 107 */
 108void cpuidle_use_deepest_state(bool enable)
 109{
 110        struct cpuidle_device *dev;
 111
 112        preempt_disable();
 113        dev = cpuidle_get_device();
 114        if (dev)
 115                dev->use_deepest_state = enable;
 116        preempt_enable();
 117}
 118
 119/**
 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.
 123 */
 124int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
 125                               struct cpuidle_device *dev)
 126{
 127        return find_deepest_state(drv, dev, UINT_MAX, 0, false);
 128}
 129
 130#ifdef CONFIG_SUSPEND
 131static void enter_s2idle_proper(struct cpuidle_driver *drv,
 132                                struct cpuidle_device *dev, int index)
 133{
 134        /*
 135         * trace_suspend_resume() called by tick_freeze() for the last CPU
 136         * executing it contains RCU usage regarded as invalid in the idle
 137         * context, so tell RCU about that.
 138         */
 139        RCU_NONIDLE(tick_freeze());
 140        /*
 141         * The state used here cannot be a "coupled" one, because the "coupled"
 142         * cpuidle mechanism enables interrupts and doing that with timekeeping
 143         * suspended is generally unsafe.
 144         */
 145        stop_critical_timings();
 146        drv->states[index].enter_s2idle(dev, drv, index);
 147        WARN_ON(!irqs_disabled());
 148        /*
 149         * timekeeping_resume() that will be called by tick_unfreeze() for the
 150         * first CPU executing it calls functions containing RCU read-side
 151         * critical sections, so tell RCU about that.
 152         */
 153        RCU_NONIDLE(tick_unfreeze());
 154        start_critical_timings();
 155}
 156
 157/**
 158 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
 159 * @drv: cpuidle driver for the given CPU.
 160 * @dev: cpuidle device for the given CPU.
 161 *
 162 * If there are states with the ->enter_s2idle callback, find the deepest of
 163 * them and enter it with frozen tick.
 164 */
 165int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 166{
 167        int index;
 168
 169        /*
 170         * Find the deepest state with ->enter_s2idle present, which guarantees
 171         * that interrupts won't be enabled when it exits and allows the tick to
 172         * be frozen safely.
 173         */
 174        index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
 175        if (index > 0)
 176                enter_s2idle_proper(drv, dev, index);
 177
 178        return index;
 179}
 180#endif /* CONFIG_SUSPEND */
 181
 182/**
 183 * cpuidle_enter_state - enter the state and update stats
 184 * @dev: cpuidle device for this cpu
 185 * @drv: cpuidle driver for this cpu
 186 * @index: index into the states table in @drv of the state to enter
 187 */
 188int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
 189                        int index)
 190{
 191        int entered_state;
 192
 193        struct cpuidle_state *target_state = &drv->states[index];
 194        bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
 195        ktime_t time_start, time_end;
 196        s64 diff;
 197
 198        /*
 199         * Tell the time framework to switch to a broadcast timer because our
 200         * local timer will be shut down.  If a local timer is used from another
 201         * CPU as a broadcast timer, this call may fail if it is not available.
 202         */
 203        if (broadcast && tick_broadcast_enter()) {
 204                index = find_deepest_state(drv, dev, target_state->exit_latency,
 205                                           CPUIDLE_FLAG_TIMER_STOP, false);
 206                if (index < 0) {
 207                        default_idle_call();
 208                        return -EBUSY;
 209                }
 210                target_state = &drv->states[index];
 211                broadcast = false;
 212        }
 213
 214        /* Take note of the planned idle state. */
 215        sched_idle_set_state(target_state);
 216
 217        trace_cpu_idle_rcuidle(index, dev->cpu);
 218        time_start = ns_to_ktime(local_clock());
 219
 220        stop_critical_timings();
 221        entered_state = target_state->enter(dev, drv, index);
 222        start_critical_timings();
 223
 224        sched_clock_idle_wakeup_event();
 225        time_end = ns_to_ktime(local_clock());
 226        trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 227
 228        /* The cpu is no longer idle or about to enter idle. */
 229        sched_idle_set_state(NULL);
 230
 231        if (broadcast) {
 232                if (WARN_ON_ONCE(!irqs_disabled()))
 233                        local_irq_disable();
 234
 235                tick_broadcast_exit();
 236        }
 237
 238        if (!cpuidle_state_is_coupled(drv, index))
 239                local_irq_enable();
 240
 241        diff = ktime_us_delta(time_end, time_start);
 242        if (diff > INT_MAX)
 243                diff = INT_MAX;
 244
 245        dev->last_residency = (int) diff;
 246
 247        if (entered_state >= 0) {
 248                /* Update cpuidle counters */
 249                /* This can be moved to within driver enter routine
 250                 * but that results in multiple copies of same code.
 251                 */
 252                dev->states_usage[entered_state].time += dev->last_residency;
 253                dev->states_usage[entered_state].usage++;
 254        } else {
 255                dev->last_residency = 0;
 256        }
 257
 258        return entered_state;
 259}
 260
 261/**
 262 * cpuidle_select - ask the cpuidle framework to choose an idle state
 263 *
 264 * @drv: the cpuidle driver
 265 * @dev: the cpuidle device
 266 *
 267 * Returns the index of the idle state.  The return value must not be negative.
 268 */
 269int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 270{
 271        return cpuidle_curr_governor->select(drv, dev);
 272}
 273
 274/**
 275 * cpuidle_enter - enter into the specified idle state
 276 *
 277 * @drv:   the cpuidle driver tied with the cpu
 278 * @dev:   the cpuidle device
 279 * @index: the index in the idle state table
 280 *
 281 * Returns the index in the idle state, < 0 in case of error.
 282 * The error code depends on the backend driver
 283 */
 284int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 285                  int index)
 286{
 287        if (cpuidle_state_is_coupled(drv, index))
 288                return cpuidle_enter_state_coupled(dev, drv, index);
 289        return cpuidle_enter_state(dev, drv, index);
 290}
 291
 292/**
 293 * cpuidle_reflect - tell the underlying governor what was the state
 294 * we were in
 295 *
 296 * @dev  : the cpuidle device
 297 * @index: the index in the idle state table
 298 *
 299 */
 300void cpuidle_reflect(struct cpuidle_device *dev, int index)
 301{
 302        if (cpuidle_curr_governor->reflect && index >= 0)
 303                cpuidle_curr_governor->reflect(dev, index);
 304}
 305
 306/**
 307 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 308 */
 309void cpuidle_install_idle_handler(void)
 310{
 311        if (enabled_devices) {
 312                /* Make sure all changes finished before we switch to new idle */
 313                smp_wmb();
 314                initialized = 1;
 315        }
 316}
 317
 318/**
 319 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 320 */
 321void cpuidle_uninstall_idle_handler(void)
 322{
 323        if (enabled_devices) {
 324                initialized = 0;
 325                wake_up_all_idle_cpus();
 326        }
 327
 328        /*
 329         * Make sure external observers (such as the scheduler)
 330         * are done looking at pointed idle states.
 331         */
 332        synchronize_rcu();
 333}
 334
 335/**
 336 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 337 */
 338void cpuidle_pause_and_lock(void)
 339{
 340        mutex_lock(&cpuidle_lock);
 341        cpuidle_uninstall_idle_handler();
 342}
 343
 344EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 345
 346/**
 347 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 348 */
 349void cpuidle_resume_and_unlock(void)
 350{
 351        cpuidle_install_idle_handler();
 352        mutex_unlock(&cpuidle_lock);
 353}
 354
 355EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 356
 357/* Currently used in suspend/resume path to suspend cpuidle */
 358void cpuidle_pause(void)
 359{
 360        mutex_lock(&cpuidle_lock);
 361        cpuidle_uninstall_idle_handler();
 362        mutex_unlock(&cpuidle_lock);
 363}
 364
 365/* Currently used in suspend/resume path to resume cpuidle */
 366void cpuidle_resume(void)
 367{
 368        mutex_lock(&cpuidle_lock);
 369        cpuidle_install_idle_handler();
 370        mutex_unlock(&cpuidle_lock);
 371}
 372
 373/**
 374 * cpuidle_enable_device - enables idle PM for a CPU
 375 * @dev: the CPU
 376 *
 377 * This function must be called between cpuidle_pause_and_lock and
 378 * cpuidle_resume_and_unlock when used externally.
 379 */
 380int cpuidle_enable_device(struct cpuidle_device *dev)
 381{
 382        int ret;
 383        struct cpuidle_driver *drv;
 384
 385        if (!dev)
 386                return -EINVAL;
 387
 388        if (dev->enabled)
 389                return 0;
 390
 391        if (!cpuidle_curr_governor)
 392                return -EIO;
 393
 394        drv = cpuidle_get_cpu_driver(dev);
 395
 396        if (!drv)
 397                return -EIO;
 398
 399        if (!dev->registered)
 400                return -EINVAL;
 401
 402        ret = cpuidle_add_device_sysfs(dev);
 403        if (ret)
 404                return ret;
 405
 406        if (cpuidle_curr_governor->enable) {
 407                ret = cpuidle_curr_governor->enable(drv, dev);
 408                if (ret)
 409                        goto fail_sysfs;
 410        }
 411
 412        smp_wmb();
 413
 414        dev->enabled = 1;
 415
 416        enabled_devices++;
 417        return 0;
 418
 419fail_sysfs:
 420        cpuidle_remove_device_sysfs(dev);
 421
 422        return ret;
 423}
 424
 425EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 426
 427/**
 428 * cpuidle_disable_device - disables idle PM for a CPU
 429 * @dev: the CPU
 430 *
 431 * This function must be called between cpuidle_pause_and_lock and
 432 * cpuidle_resume_and_unlock when used externally.
 433 */
 434void cpuidle_disable_device(struct cpuidle_device *dev)
 435{
 436        struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 437
 438        if (!dev || !dev->enabled)
 439                return;
 440
 441        if (!drv || !cpuidle_curr_governor)
 442                return;
 443
 444        dev->enabled = 0;
 445
 446        if (cpuidle_curr_governor->disable)
 447                cpuidle_curr_governor->disable(drv, dev);
 448
 449        cpuidle_remove_device_sysfs(dev);
 450        enabled_devices--;
 451}
 452
 453EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 454
 455static void __cpuidle_unregister_device(struct cpuidle_device *dev)
 456{
 457        struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 458
 459        list_del(&dev->device_list);
 460        per_cpu(cpuidle_devices, dev->cpu) = NULL;
 461        module_put(drv->owner);
 462
 463        dev->registered = 0;
 464}
 465
 466static void __cpuidle_device_init(struct cpuidle_device *dev)
 467{
 468        memset(dev->states_usage, 0, sizeof(dev->states_usage));
 469        dev->last_residency = 0;
 470}
 471
 472/**
 473 * __cpuidle_register_device - internal register function called before register
 474 * and enable routines
 475 * @dev: the cpu
 476 *
 477 * cpuidle_lock mutex must be held before this is called
 478 */
 479static int __cpuidle_register_device(struct cpuidle_device *dev)
 480{
 481        int ret;
 482        struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 483
 484        if (!try_module_get(drv->owner))
 485                return -EINVAL;
 486
 487        per_cpu(cpuidle_devices, dev->cpu) = dev;
 488        list_add(&dev->device_list, &cpuidle_detected_devices);
 489
 490        ret = cpuidle_coupled_register_device(dev);
 491        if (ret)
 492                __cpuidle_unregister_device(dev);
 493        else
 494                dev->registered = 1;
 495
 496        return ret;
 497}
 498
 499/**
 500 * cpuidle_register_device - registers a CPU's idle PM feature
 501 * @dev: the cpu
 502 */
 503int cpuidle_register_device(struct cpuidle_device *dev)
 504{
 505        int ret = -EBUSY;
 506
 507        if (!dev)
 508                return -EINVAL;
 509
 510        mutex_lock(&cpuidle_lock);
 511
 512        if (dev->registered)
 513                goto out_unlock;
 514
 515        __cpuidle_device_init(dev);
 516
 517        ret = __cpuidle_register_device(dev);
 518        if (ret)
 519                goto out_unlock;
 520
 521        ret = cpuidle_add_sysfs(dev);
 522        if (ret)
 523                goto out_unregister;
 524
 525        ret = cpuidle_enable_device(dev);
 526        if (ret)
 527                goto out_sysfs;
 528
 529        cpuidle_install_idle_handler();
 530
 531out_unlock:
 532        mutex_unlock(&cpuidle_lock);
 533
 534        return ret;
 535
 536out_sysfs:
 537        cpuidle_remove_sysfs(dev);
 538out_unregister:
 539        __cpuidle_unregister_device(dev);
 540        goto out_unlock;
 541}
 542
 543EXPORT_SYMBOL_GPL(cpuidle_register_device);
 544
 545/**
 546 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 547 * @dev: the cpu
 548 */
 549void cpuidle_unregister_device(struct cpuidle_device *dev)
 550{
 551        if (!dev || dev->registered == 0)
 552                return;
 553
 554        cpuidle_pause_and_lock();
 555
 556        cpuidle_disable_device(dev);
 557
 558        cpuidle_remove_sysfs(dev);
 559
 560        __cpuidle_unregister_device(dev);
 561
 562        cpuidle_coupled_unregister_device(dev);
 563
 564        cpuidle_resume_and_unlock();
 565}
 566
 567EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 568
 569/**
 570 * cpuidle_unregister: unregister a driver and the devices. This function
 571 * can be used only if the driver has been previously registered through
 572 * the cpuidle_register function.
 573 *
 574 * @drv: a valid pointer to a struct cpuidle_driver
 575 */
 576void cpuidle_unregister(struct cpuidle_driver *drv)
 577{
 578        int cpu;
 579        struct cpuidle_device *device;
 580
 581        for_each_cpu(cpu, drv->cpumask) {
 582                device = &per_cpu(cpuidle_dev, cpu);
 583                cpuidle_unregister_device(device);
 584        }
 585
 586        cpuidle_unregister_driver(drv);
 587}
 588EXPORT_SYMBOL_GPL(cpuidle_unregister);
 589
 590/**
 591 * cpuidle_register: registers the driver and the cpu devices with the
 592 * coupled_cpus passed as parameter. This function is used for all common
 593 * initialization pattern there are in the arch specific drivers. The
 594 * devices is globally defined in this file.
 595 *
 596 * @drv         : a valid pointer to a struct cpuidle_driver
 597 * @coupled_cpus: a cpumask for the coupled states
 598 *
 599 * Returns 0 on success, < 0 otherwise
 600 */
 601int cpuidle_register(struct cpuidle_driver *drv,
 602                     const struct cpumask *const coupled_cpus)
 603{
 604        int ret, cpu;
 605        struct cpuidle_device *device;
 606
 607        ret = cpuidle_register_driver(drv);
 608        if (ret) {
 609                pr_err("failed to register cpuidle driver\n");
 610                return ret;
 611        }
 612
 613        for_each_cpu(cpu, drv->cpumask) {
 614                device = &per_cpu(cpuidle_dev, cpu);
 615                device->cpu = cpu;
 616
 617#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
 618                /*
 619                 * On multiplatform for ARM, the coupled idle states could be
 620                 * enabled in the kernel even if the cpuidle driver does not
 621                 * use it. Note, coupled_cpus is a struct copy.
 622                 */
 623                if (coupled_cpus)
 624                        device->coupled_cpus = *coupled_cpus;
 625#endif
 626                ret = cpuidle_register_device(device);
 627                if (!ret)
 628                        continue;
 629
 630                pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
 631
 632                cpuidle_unregister(drv);
 633                break;
 634        }
 635
 636        return ret;
 637}
 638EXPORT_SYMBOL_GPL(cpuidle_register);
 639
 640#ifdef CONFIG_SMP
 641
 642/*
 643 * This function gets called when a part of the kernel has a new latency
 644 * requirement.  This means we need to get all processors out of their C-state,
 645 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 646 * wakes them all right up.
 647 */
 648static int cpuidle_latency_notify(struct notifier_block *b,
 649                unsigned long l, void *v)
 650{
 651        wake_up_all_idle_cpus();
 652        return NOTIFY_OK;
 653}
 654
 655static struct notifier_block cpuidle_latency_notifier = {
 656        .notifier_call = cpuidle_latency_notify,
 657};
 658
 659static inline void latency_notifier_init(struct notifier_block *n)
 660{
 661        pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 662}
 663
 664#else /* CONFIG_SMP */
 665
 666#define latency_notifier_init(x) do { } while (0)
 667
 668#endif /* CONFIG_SMP */
 669
 670/**
 671 * cpuidle_init - core initializer
 672 */
 673static int __init cpuidle_init(void)
 674{
 675        int ret;
 676
 677        if (cpuidle_disabled())
 678                return -ENODEV;
 679
 680        ret = cpuidle_add_interface(cpu_subsys.dev_root);
 681        if (ret)
 682                return ret;
 683
 684        latency_notifier_init(&cpuidle_latency_notifier);
 685
 686        return 0;
 687}
 688
 689module_param(off, int, 0444);
 690core_initcall(cpuidle_init);
 691