drivers/cpuidle/cpuidle.c

   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
  29 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
  30 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
  31
  32 DEFINE_MUTEX(cpuidle_lock);
  33 LIST_HEAD(cpuidle_detected_devices);
  34
  35 static int enabled_devices;
  36 static int off __read_mostly;
  37 static int initialized __read_mostly;
  38
  39 int cpuidle_disabled(void)
  40 {
  41         return off;
  42 }
  43 void disable_cpuidle(void)
  44 {
  45         off = 1;
  46 }
  47
  48 bool 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  */
  59 int 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
  76 static 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 freeze)
  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                     || (freeze && !s->enter_freeze))
  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  */
 108 void 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  */
 124 int 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
 131 static void enter_freeze_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_freeze(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_freeze - 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_freeze callback, find the deepest of
 163  * them and enter it with frozen tick.
 164  */
 165 int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 166 {
 167         int index;
 168
 169         /*
 170          * Find the deepest state with ->enter_freeze 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_freeze_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  */
 188 int 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         }
 212
 213         /* Take note of the planned idle state. */
 214         sched_idle_set_state(target_state);
 215
 216         trace_cpu_idle_rcuidle(index, dev->cpu);
 217         time_start = ns_to_ktime(local_clock());
 218
 219         stop_critical_timings();
 220         entered_state = target_state->enter(dev, drv, index);
 221         start_critical_timings();
 222
 223         time_end = ns_to_ktime(local_clock());
 224         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 225
 226         /* The cpu is no longer idle or about to enter idle. */
 227         sched_idle_set_state(NULL);
 228
 229         if (broadcast) {
 230                 if (WARN_ON_ONCE(!irqs_disabled()))
 231                         local_irq_disable();
 232
 233                 tick_broadcast_exit();
 234         }
 235
 236         if (!cpuidle_state_is_coupled(drv, index))
 237                 local_irq_enable();
 238
 239         diff = ktime_us_delta(time_end, time_start);
 240         if (diff > INT_MAX)
 241                 diff = INT_MAX;
 242
 243         dev->last_residency = (int) diff;
 244
 245         if (entered_state >= 0) {
 246                 /* Update cpuidle counters */
 247                 /* This can be moved to within driver enter routine
 248                  * but that results in multiple copies of same code.
 249                  */
 250                 dev->states_usage[entered_state].time += dev->last_residency;
 251                 dev->states_usage[entered_state].usage++;
 252         } else {
 253                 dev->last_residency = 0;
 254         }
 255
 256         return entered_state;
 257 }
 258
 259 /**
 260  * cpuidle_select - ask the cpuidle framework to choose an idle state
 261  *
 262  * @drv: the cpuidle driver
 263  * @dev: the cpuidle device
 264  *
 265  * Returns the index of the idle state.  The return value must not be negative.
 266  */
 267 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 268 {
 269         return cpuidle_curr_governor->select(drv, dev);
 270 }
 271
 272 /**
 273  * cpuidle_enter - enter into the specified idle state
 274  *
 275  * @drv:   the cpuidle driver tied with the cpu
 276  * @dev:   the cpuidle device
 277  * @index: the index in the idle state table
 278  *
 279  * Returns the index in the idle state, < 0 in case of error.
 280  * The error code depends on the backend driver
 281  */
 282 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 283                   int index)
 284 {
 285         if (cpuidle_state_is_coupled(drv, index))
 286                 return cpuidle_enter_state_coupled(dev, drv, index);
 287         return cpuidle_enter_state(dev, drv, index);
 288 }
 289
 290 /**
 291  * cpuidle_reflect - tell the underlying governor what was the state
 292  * we were in
 293  *
 294  * @dev  : the cpuidle device
 295  * @index: the index in the idle state table
 296  *
 297  */
 298 void cpuidle_reflect(struct cpuidle_device *dev, int index)
 299 {
 300         if (cpuidle_curr_governor->reflect && index >= 0)
 301                 cpuidle_curr_governor->reflect(dev, index);
 302 }
 303
 304 /**
 305  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 306  */
 307 void cpuidle_install_idle_handler(void)
 308 {
 309         if (enabled_devices) {
 310                 /* Make sure all changes finished before we switch to new idle */
 311                 smp_wmb();
 312                 initialized = 1;
 313         }
 314 }
 315
 316 /**
 317  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 318  */
 319 void cpuidle_uninstall_idle_handler(void)
 320 {
 321         if (enabled_devices) {
 322                 initialized = 0;
 323                 wake_up_all_idle_cpus();
 324         }
 325
 326         /*
 327          * Make sure external observers (such as the scheduler)
 328          * are done looking at pointed idle states.
 329          */
 330         synchronize_rcu();
 331 }
 332
 333 /**
 334  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 335  */
 336 void cpuidle_pause_and_lock(void)
 337 {
 338         mutex_lock(&cpuidle_lock);
 339         cpuidle_uninstall_idle_handler();
 340 }
 341
 342 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 343
 344 /**
 345  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 346  */
 347 void cpuidle_resume_and_unlock(void)
 348 {
 349         cpuidle_install_idle_handler();
 350         mutex_unlock(&cpuidle_lock);
 351 }
 352
 353 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 354
 355 /* Currently used in suspend/resume path to suspend cpuidle */
 356 void cpuidle_pause(void)
 357 {
 358         mutex_lock(&cpuidle_lock);
 359         cpuidle_uninstall_idle_handler();
 360         mutex_unlock(&cpuidle_lock);
 361 }
 362
 363 /* Currently used in suspend/resume path to resume cpuidle */
 364 void cpuidle_resume(void)
 365 {
 366         mutex_lock(&cpuidle_lock);
 367         cpuidle_install_idle_handler();
 368         mutex_unlock(&cpuidle_lock);
 369 }
 370
 371 /**
 372  * cpuidle_enable_device - enables idle PM for a CPU
 373  * @dev: the CPU
 374  *
 375  * This function must be called between cpuidle_pause_and_lock and
 376  * cpuidle_resume_and_unlock when used externally.
 377  */
 378 int cpuidle_enable_device(struct cpuidle_device *dev)
 379 {
 380         int ret;
 381         struct cpuidle_driver *drv;
 382
 383         if (!dev)
 384                 return -EINVAL;
 385
 386         if (dev->enabled)
 387                 return 0;
 388
 389         drv = cpuidle_get_cpu_driver(dev);
 390
 391         if (!drv || !cpuidle_curr_governor)
 392                 return -EIO;
 393
 394         if (!dev->registered)
 395                 return -EINVAL;
 396
 397         ret = cpuidle_add_device_sysfs(dev);
 398         if (ret)
 399                 return ret;
 400
 401         if (cpuidle_curr_governor->enable &&
 402             (ret = cpuidle_curr_governor->enable(drv, dev)))
 403                 goto fail_sysfs;
 404
 405         smp_wmb();
 406
 407         dev->enabled = 1;
 408
 409         enabled_devices++;
 410         return 0;
 411
 412 fail_sysfs:
 413         cpuidle_remove_device_sysfs(dev);
 414
 415         return ret;
 416 }
 417
 418 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 419
 420 /**
 421  * cpuidle_disable_device - disables idle PM for a CPU
 422  * @dev: the CPU
 423  *
 424  * This function must be called between cpuidle_pause_and_lock and
 425  * cpuidle_resume_and_unlock when used externally.
 426  */
 427 void cpuidle_disable_device(struct cpuidle_device *dev)
 428 {
 429         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 430
 431         if (!dev || !dev->enabled)
 432                 return;
 433
 434         if (!drv || !cpuidle_curr_governor)
 435                 return;
 436
 437         dev->enabled = 0;
 438
 439         if (cpuidle_curr_governor->disable)
 440                 cpuidle_curr_governor->disable(drv, dev);
 441
 442         cpuidle_remove_device_sysfs(dev);
 443         enabled_devices--;
 444 }
 445
 446 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 447
 448 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
 449 {
 450         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 451
 452         list_del(&dev->device_list);
 453         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 454         module_put(drv->owner);
 455
 456         dev->registered = 0;
 457 }
 458
 459 static void __cpuidle_device_init(struct cpuidle_device *dev)
 460 {
 461         memset(dev->states_usage, 0, sizeof(dev->states_usage));
 462         dev->last_residency = 0;
 463 }
 464
 465 /**
 466  * __cpuidle_register_device - internal register function called before register
 467  * and enable routines
 468  * @dev: the cpu
 469  *
 470  * cpuidle_lock mutex must be held before this is called
 471  */
 472 static int __cpuidle_register_device(struct cpuidle_device *dev)
 473 {
 474         int ret;
 475         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 476
 477         if (!try_module_get(drv->owner))
 478                 return -EINVAL;
 479
 480         per_cpu(cpuidle_devices, dev->cpu) = dev;
 481         list_add(&dev->device_list, &cpuidle_detected_devices);
 482
 483         ret = cpuidle_coupled_register_device(dev);
 484         if (ret)
 485                 __cpuidle_unregister_device(dev);
 486         else
 487                 dev->registered = 1;
 488
 489         return ret;
 490 }
 491
 492 /**
 493  * cpuidle_register_device - registers a CPU's idle PM feature
 494  * @dev: the cpu
 495  */
 496 int cpuidle_register_device(struct cpuidle_device *dev)
 497 {
 498         int ret = -EBUSY;
 499
 500         if (!dev)
 501                 return -EINVAL;
 502
 503         mutex_lock(&cpuidle_lock);
 504
 505         if (dev->registered)
 506                 goto out_unlock;
 507
 508         __cpuidle_device_init(dev);
 509
 510         ret = __cpuidle_register_device(dev);
 511         if (ret)
 512                 goto out_unlock;
 513
 514         ret = cpuidle_add_sysfs(dev);
 515         if (ret)
 516                 goto out_unregister;
 517
 518         ret = cpuidle_enable_device(dev);
 519         if (ret)
 520                 goto out_sysfs;
 521
 522         cpuidle_install_idle_handler();
 523
 524 out_unlock:
 525         mutex_unlock(&cpuidle_lock);
 526
 527         return ret;
 528
 529 out_sysfs:
 530         cpuidle_remove_sysfs(dev);
 531 out_unregister:
 532         __cpuidle_unregister_device(dev);
 533         goto out_unlock;
 534 }
 535
 536 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 537
 538 /**
 539  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 540  * @dev: the cpu
 541  */
 542 void cpuidle_unregister_device(struct cpuidle_device *dev)
 543 {
 544         if (!dev || dev->registered == 0)
 545                 return;
 546
 547         cpuidle_pause_and_lock();
 548
 549         cpuidle_disable_device(dev);
 550
 551         cpuidle_remove_sysfs(dev);
 552
 553         __cpuidle_unregister_device(dev);
 554
 555         cpuidle_coupled_unregister_device(dev);
 556
 557         cpuidle_resume_and_unlock();
 558 }
 559
 560 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 561
 562 /**
 563  * cpuidle_unregister: unregister a driver and the devices. This function
 564  * can be used only if the driver has been previously registered through
 565  * the cpuidle_register function.
 566  *
 567  * @drv: a valid pointer to a struct cpuidle_driver
 568  */
 569 void cpuidle_unregister(struct cpuidle_driver *drv)
 570 {
 571         int cpu;
 572         struct cpuidle_device *device;
 573
 574         for_each_cpu(cpu, drv->cpumask) {
 575                 device = &per_cpu(cpuidle_dev, cpu);
 576                 cpuidle_unregister_device(device);
 577         }
 578
 579         cpuidle_unregister_driver(drv);
 580 }
 581 EXPORT_SYMBOL_GPL(cpuidle_unregister);
 582
 583 /**
 584  * cpuidle_register: registers the driver and the cpu devices with the
 585  * coupled_cpus passed as parameter. This function is used for all common
 586  * initialization pattern there are in the arch specific drivers. The
 587  * devices is globally defined in this file.
 588  *
 589  * @drv         : a valid pointer to a struct cpuidle_driver
 590  * @coupled_cpus: a cpumask for the coupled states
 591  *
 592  * Returns 0 on success, < 0 otherwise
 593  */
 594 int cpuidle_register(struct cpuidle_driver *drv,
 595                      const struct cpumask *const coupled_cpus)
 596 {
 597         int ret, cpu;
 598         struct cpuidle_device *device;
 599
 600         ret = cpuidle_register_driver(drv);
 601         if (ret) {
 602                 pr_err("failed to register cpuidle driver\n");
 603                 return ret;
 604         }
 605
 606         for_each_cpu(cpu, drv->cpumask) {
 607                 device = &per_cpu(cpuidle_dev, cpu);
 608                 device->cpu = cpu;
 609
 610 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
 611                 /*
 612                  * On multiplatform for ARM, the coupled idle states could be
 613                  * enabled in the kernel even if the cpuidle driver does not
 614                  * use it. Note, coupled_cpus is a struct copy.
 615                  */
 616                 if (coupled_cpus)
 617                         device->coupled_cpus = *coupled_cpus;
 618 #endif
 619                 ret = cpuidle_register_device(device);
 620                 if (!ret)
 621                         continue;
 622
 623                 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
 624
 625                 cpuidle_unregister(drv);
 626                 break;
 627         }
 628
 629         return ret;
 630 }
 631 EXPORT_SYMBOL_GPL(cpuidle_register);
 632
 633 #ifdef CONFIG_SMP
 634
 635 /*
 636  * This function gets called when a part of the kernel has a new latency
 637  * requirement.  This means we need to get all processors out of their C-state,
 638  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 639  * wakes them all right up.
 640  */
 641 static int cpuidle_latency_notify(struct notifier_block *b,
 642                 unsigned long l, void *v)
 643 {
 644         wake_up_all_idle_cpus();
 645         return NOTIFY_OK;
 646 }
 647
 648 static struct notifier_block cpuidle_latency_notifier = {
 649         .notifier_call = cpuidle_latency_notify,
 650 };
 651
 652 static inline void latency_notifier_init(struct notifier_block *n)
 653 {
 654         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 655 }
 656
 657 #else /* CONFIG_SMP */
 658
 659 #define latency_notifier_init(x) do { } while (0)
 660
 661 #endif /* CONFIG_SMP */
 662
 663 /**
 664  * cpuidle_init - core initializer
 665  */
 666 static int __init cpuidle_init(void)
 667 {
 668         int ret;
 669
 670         if (cpuidle_disabled())
 671                 return -ENODEV;
 672
 673         ret = cpuidle_add_interface(cpu_subsys.dev_root);
 674         if (ret)
 675                 return ret;
 676
 677         latency_notifier_init(&cpuidle_latency_notifier);
 678
 679         return 0;
 680 }
 681
 682 module_param(off, int, 0444);
 683 core_initcall(cpuidle_init);