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/notifier.h>
  16 #include <linux/pm_qos.h>
  17 #include <linux/cpu.h>
  18 #include <linux/cpuidle.h>
  19 #include <linux/ktime.h>
  20 #include <linux/hrtimer.h>
  21 #include <linux/module.h>
  22 #include <linux/suspend.h>
  23 #include <linux/tick.h>
  24 #include <trace/events/power.h>
  25
  26 #include "cpuidle.h"
  27
  28 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
  29 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
  30
  31 DEFINE_MUTEX(cpuidle_lock);
  32 LIST_HEAD(cpuidle_detected_devices);
  33
  34 static int enabled_devices;
  35 static int off __read_mostly;
  36 static int initialized __read_mostly;
  37
  38 int cpuidle_disabled(void)
  39 {
  40         return off;
  41 }
  42 void disable_cpuidle(void)
  43 {
  44         off = 1;
  45 }
  46
  47 bool cpuidle_not_available(struct cpuidle_driver *drv,
  48                            struct cpuidle_device *dev)
  49 {
  50         return off || !initialized || !drv || !dev || !dev->enabled;
  51 }
  52
  53 /**
  54  * cpuidle_play_dead - cpu off-lining
  55  *
  56  * Returns in case of an error or no driver
  57  */
  58 int cpuidle_play_dead(void)
  59 {
  60         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  61         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  62         int i;
  63
  64         if (!drv)
  65                 return -ENODEV;
  66
  67         /* Find lowest-power state that supports long-term idle */
  68         for (i = drv->state_count - 1; i >= 0; i--)
  69                 if (drv->states[i].enter_dead)
  70                         return drv->states[i].enter_dead(dev, i);
  71
  72         return -ENODEV;
  73 }
  74
  75 static int find_deepest_state(struct cpuidle_driver *drv,
  76                               struct cpuidle_device *dev,
  77                               unsigned int max_latency,
  78                               unsigned int forbidden_flags,
  79                               bool freeze)
  80 {
  81         unsigned int latency_req = 0;
  82         int i, ret = -ENXIO;
  83
  84         for (i = 0; i < drv->state_count; i++) {
  85                 struct cpuidle_state *s = &drv->states[i];
  86                 struct cpuidle_state_usage *su = &dev->states_usage[i];
  87
  88                 if (s->disabled || su->disable || s->exit_latency <= latency_req
  89                     || s->exit_latency > max_latency
  90                     || (s->flags & forbidden_flags)
  91                     || (freeze && !s->enter_freeze))
  92                         continue;
  93
  94                 latency_req = s->exit_latency;
  95                 ret = i;
  96         }
  97         return ret;
  98 }
  99
 100 #ifdef CONFIG_SUSPEND
 101 /**
 102  * cpuidle_find_deepest_state - Find the deepest available idle state.
 103  * @drv: cpuidle driver for the given CPU.
 104  * @dev: cpuidle device for the given CPU.
 105  */
 106 int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
 107                                struct cpuidle_device *dev)
 108 {
 109         return find_deepest_state(drv, dev, UINT_MAX, 0, false);
 110 }
 111
 112 static void enter_freeze_proper(struct cpuidle_driver *drv,
 113                                 struct cpuidle_device *dev, int index)
 114 {
 115         tick_freeze();
 116         /*
 117          * The state used here cannot be a "coupled" one, because the "coupled"
 118          * cpuidle mechanism enables interrupts and doing that with timekeeping
 119          * suspended is generally unsafe.
 120          */
 121         drv->states[index].enter_freeze(dev, drv, index);
 122         WARN_ON(!irqs_disabled());
 123         /*
 124          * timekeeping_resume() that will be called by tick_unfreeze() for the
 125          * last CPU executing it calls functions containing RCU read-side
 126          * critical sections, so tell RCU about that.
 127          */
 128         RCU_NONIDLE(tick_unfreeze());
 129 }
 130
 131 /**
 132  * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle.
 133  * @drv: cpuidle driver for the given CPU.
 134  * @dev: cpuidle device for the given CPU.
 135  *
 136  * If there are states with the ->enter_freeze callback, find the deepest of
 137  * them and enter it with frozen tick.
 138  */
 139 int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 140 {
 141         int index;
 142
 143         /*
 144          * Find the deepest state with ->enter_freeze present, which guarantees
 145          * that interrupts won't be enabled when it exits and allows the tick to
 146          * be frozen safely.
 147          */
 148         index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
 149         if (index >= 0)
 150                 enter_freeze_proper(drv, dev, index);
 151
 152         return index;
 153 }
 154 #endif /* CONFIG_SUSPEND */
 155
 156 /**
 157  * cpuidle_enter_state - enter the state and update stats
 158  * @dev: cpuidle device for this cpu
 159  * @drv: cpuidle driver for this cpu
 160  * @index: index into the states table in @drv of the state to enter
 161  */
 162 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
 163                         int index)
 164 {
 165         int entered_state;
 166
 167         struct cpuidle_state *target_state = &drv->states[index];
 168         bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
 169         ktime_t time_start, time_end;
 170         s64 diff;
 171
 172         /*
 173          * Tell the time framework to switch to a broadcast timer because our
 174          * local timer will be shut down.  If a local timer is used from another
 175          * CPU as a broadcast timer, this call may fail if it is not available.
 176          */
 177         if (broadcast && tick_broadcast_enter()) {
 178                 index = find_deepest_state(drv, dev, target_state->exit_latency,
 179                                            CPUIDLE_FLAG_TIMER_STOP, false);
 180                 if (index < 0) {
 181                         default_idle_call();
 182                         return -EBUSY;
 183                 }
 184                 target_state = &drv->states[index];
 185         }
 186
 187         /* Take note of the planned idle state. */
 188         sched_idle_set_state(target_state);
 189
 190         trace_cpu_idle_rcuidle(index, dev->cpu);
 191         time_start = ktime_get();
 192
 193         entered_state = target_state->enter(dev, drv, index);
 194
 195         time_end = ktime_get();
 196         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 197
 198         /* The cpu is no longer idle or about to enter idle. */
 199         sched_idle_set_state(NULL);
 200
 201         if (broadcast) {
 202                 if (WARN_ON_ONCE(!irqs_disabled()))
 203                         local_irq_disable();
 204
 205                 tick_broadcast_exit();
 206         }
 207
 208         if (!cpuidle_state_is_coupled(dev, drv, entered_state))
 209                 local_irq_enable();
 210
 211         diff = ktime_to_us(ktime_sub(time_end, time_start));
 212         if (diff > INT_MAX)
 213                 diff = INT_MAX;
 214
 215         dev->last_residency = (int) diff;
 216
 217         if (entered_state >= 0) {
 218                 /* Update cpuidle counters */
 219                 /* This can be moved to within driver enter routine
 220                  * but that results in multiple copies of same code.
 221                  */
 222                 dev->states_usage[entered_state].time += dev->last_residency;
 223                 dev->states_usage[entered_state].usage++;
 224         } else {
 225                 dev->last_residency = 0;
 226         }
 227
 228         return entered_state;
 229 }
 230
 231 /**
 232  * cpuidle_select - ask the cpuidle framework to choose an idle state
 233  *
 234  * @drv: the cpuidle driver
 235  * @dev: the cpuidle device
 236  *
 237  * Returns the index of the idle state.
 238  */
 239 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 240 {
 241         return cpuidle_curr_governor->select(drv, dev);
 242 }
 243
 244 /**
 245  * cpuidle_enter - enter into the specified idle state
 246  *
 247  * @drv:   the cpuidle driver tied with the cpu
 248  * @dev:   the cpuidle device
 249  * @index: the index in the idle state table
 250  *
 251  * Returns the index in the idle state, < 0 in case of error.
 252  * The error code depends on the backend driver
 253  */
 254 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 255                   int index)
 256 {
 257         if (cpuidle_state_is_coupled(dev, drv, index))
 258                 return cpuidle_enter_state_coupled(dev, drv, index);
 259         return cpuidle_enter_state(dev, drv, index);
 260 }
 261
 262 /**
 263  * cpuidle_reflect - tell the underlying governor what was the state
 264  * we were in
 265  *
 266  * @dev  : the cpuidle device
 267  * @index: the index in the idle state table
 268  *
 269  */
 270 void cpuidle_reflect(struct cpuidle_device *dev, int index)
 271 {
 272         if (cpuidle_curr_governor->reflect && index >= 0)
 273                 cpuidle_curr_governor->reflect(dev, index);
 274 }
 275
 276 /**
 277  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 278  */
 279 void cpuidle_install_idle_handler(void)
 280 {
 281         if (enabled_devices) {
 282                 /* Make sure all changes finished before we switch to new idle */
 283                 smp_wmb();
 284                 initialized = 1;
 285         }
 286 }
 287
 288 /**
 289  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 290  */
 291 void cpuidle_uninstall_idle_handler(void)
 292 {
 293         if (enabled_devices) {
 294                 initialized = 0;
 295                 wake_up_all_idle_cpus();
 296         }
 297
 298         /*
 299          * Make sure external observers (such as the scheduler)
 300          * are done looking at pointed idle states.
 301          */
 302         synchronize_rcu();
 303 }
 304
 305 /**
 306  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 307  */
 308 void cpuidle_pause_and_lock(void)
 309 {
 310         mutex_lock(&cpuidle_lock);
 311         cpuidle_uninstall_idle_handler();
 312 }
 313
 314 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 315
 316 /**
 317  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 318  */
 319 void cpuidle_resume_and_unlock(void)
 320 {
 321         cpuidle_install_idle_handler();
 322         mutex_unlock(&cpuidle_lock);
 323 }
 324
 325 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 326
 327 /* Currently used in suspend/resume path to suspend cpuidle */
 328 void cpuidle_pause(void)
 329 {
 330         mutex_lock(&cpuidle_lock);
 331         cpuidle_uninstall_idle_handler();
 332         mutex_unlock(&cpuidle_lock);
 333 }
 334
 335 /* Currently used in suspend/resume path to resume cpuidle */
 336 void cpuidle_resume(void)
 337 {
 338         mutex_lock(&cpuidle_lock);
 339         cpuidle_install_idle_handler();
 340         mutex_unlock(&cpuidle_lock);
 341 }
 342
 343 /**
 344  * cpuidle_enable_device - enables idle PM for a CPU
 345  * @dev: the CPU
 346  *
 347  * This function must be called between cpuidle_pause_and_lock and
 348  * cpuidle_resume_and_unlock when used externally.
 349  */
 350 int cpuidle_enable_device(struct cpuidle_device *dev)
 351 {
 352         int ret;
 353         struct cpuidle_driver *drv;
 354
 355         if (!dev)
 356                 return -EINVAL;
 357
 358         if (dev->enabled)
 359                 return 0;
 360
 361         drv = cpuidle_get_cpu_driver(dev);
 362
 363         if (!drv || !cpuidle_curr_governor)
 364                 return -EIO;
 365
 366         if (!dev->registered)
 367                 return -EINVAL;
 368
 369         ret = cpuidle_add_device_sysfs(dev);
 370         if (ret)
 371                 return ret;
 372
 373         if (cpuidle_curr_governor->enable &&
 374             (ret = cpuidle_curr_governor->enable(drv, dev)))
 375                 goto fail_sysfs;
 376
 377         smp_wmb();
 378
 379         dev->enabled = 1;
 380
 381         enabled_devices++;
 382         return 0;
 383
 384 fail_sysfs:
 385         cpuidle_remove_device_sysfs(dev);
 386
 387         return ret;
 388 }
 389
 390 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 391
 392 /**
 393  * cpuidle_disable_device - disables idle PM for a CPU
 394  * @dev: the CPU
 395  *
 396  * This function must be called between cpuidle_pause_and_lock and
 397  * cpuidle_resume_and_unlock when used externally.
 398  */
 399 void cpuidle_disable_device(struct cpuidle_device *dev)
 400 {
 401         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 402
 403         if (!dev || !dev->enabled)
 404                 return;
 405
 406         if (!drv || !cpuidle_curr_governor)
 407                 return;
 408
 409         dev->enabled = 0;
 410
 411         if (cpuidle_curr_governor->disable)
 412                 cpuidle_curr_governor->disable(drv, dev);
 413
 414         cpuidle_remove_device_sysfs(dev);
 415         enabled_devices--;
 416 }
 417
 418 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 419
 420 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
 421 {
 422         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 423
 424         list_del(&dev->device_list);
 425         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 426         module_put(drv->owner);
 427 }
 428
 429 static void __cpuidle_device_init(struct cpuidle_device *dev)
 430 {
 431         memset(dev->states_usage, 0, sizeof(dev->states_usage));
 432         dev->last_residency = 0;
 433 }
 434
 435 /**
 436  * __cpuidle_register_device - internal register function called before register
 437  * and enable routines
 438  * @dev: the cpu
 439  *
 440  * cpuidle_lock mutex must be held before this is called
 441  */
 442 static int __cpuidle_register_device(struct cpuidle_device *dev)
 443 {
 444         int ret;
 445         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 446
 447         if (!try_module_get(drv->owner))
 448                 return -EINVAL;
 449
 450         per_cpu(cpuidle_devices, dev->cpu) = dev;
 451         list_add(&dev->device_list, &cpuidle_detected_devices);
 452
 453         ret = cpuidle_coupled_register_device(dev);
 454         if (ret)
 455                 __cpuidle_unregister_device(dev);
 456         else
 457                 dev->registered = 1;
 458
 459         return ret;
 460 }
 461
 462 /**
 463  * cpuidle_register_device - registers a CPU's idle PM feature
 464  * @dev: the cpu
 465  */
 466 int cpuidle_register_device(struct cpuidle_device *dev)
 467 {
 468         int ret = -EBUSY;
 469
 470         if (!dev)
 471                 return -EINVAL;
 472
 473         mutex_lock(&cpuidle_lock);
 474
 475         if (dev->registered)
 476                 goto out_unlock;
 477
 478         __cpuidle_device_init(dev);
 479
 480         ret = __cpuidle_register_device(dev);
 481         if (ret)
 482                 goto out_unlock;
 483
 484         ret = cpuidle_add_sysfs(dev);
 485         if (ret)
 486                 goto out_unregister;
 487
 488         ret = cpuidle_enable_device(dev);
 489         if (ret)
 490                 goto out_sysfs;
 491
 492         cpuidle_install_idle_handler();
 493
 494 out_unlock:
 495         mutex_unlock(&cpuidle_lock);
 496
 497         return ret;
 498
 499 out_sysfs:
 500         cpuidle_remove_sysfs(dev);
 501 out_unregister:
 502         __cpuidle_unregister_device(dev);
 503         goto out_unlock;
 504 }
 505
 506 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 507
 508 /**
 509  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 510  * @dev: the cpu
 511  */
 512 void cpuidle_unregister_device(struct cpuidle_device *dev)
 513 {
 514         if (!dev || dev->registered == 0)
 515                 return;
 516
 517         cpuidle_pause_and_lock();
 518
 519         cpuidle_disable_device(dev);
 520
 521         cpuidle_remove_sysfs(dev);
 522
 523         __cpuidle_unregister_device(dev);
 524
 525         cpuidle_coupled_unregister_device(dev);
 526
 527         cpuidle_resume_and_unlock();
 528 }
 529
 530 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 531
 532 /**
 533  * cpuidle_unregister: unregister a driver and the devices. This function
 534  * can be used only if the driver has been previously registered through
 535  * the cpuidle_register function.
 536  *
 537  * @drv: a valid pointer to a struct cpuidle_driver
 538  */
 539 void cpuidle_unregister(struct cpuidle_driver *drv)
 540 {
 541         int cpu;
 542         struct cpuidle_device *device;
 543
 544         for_each_cpu(cpu, drv->cpumask) {
 545                 device = &per_cpu(cpuidle_dev, cpu);
 546                 cpuidle_unregister_device(device);
 547         }
 548
 549         cpuidle_unregister_driver(drv);
 550 }
 551 EXPORT_SYMBOL_GPL(cpuidle_unregister);
 552
 553 /**
 554  * cpuidle_register: registers the driver and the cpu devices with the
 555  * coupled_cpus passed as parameter. This function is used for all common
 556  * initialization pattern there are in the arch specific drivers. The
 557  * devices is globally defined in this file.
 558  *
 559  * @drv         : a valid pointer to a struct cpuidle_driver
 560  * @coupled_cpus: a cpumask for the coupled states
 561  *
 562  * Returns 0 on success, < 0 otherwise
 563  */
 564 int cpuidle_register(struct cpuidle_driver *drv,
 565                      const struct cpumask *const coupled_cpus)
 566 {
 567         int ret, cpu;
 568         struct cpuidle_device *device;
 569
 570         ret = cpuidle_register_driver(drv);
 571         if (ret) {
 572                 pr_err("failed to register cpuidle driver\n");
 573                 return ret;
 574         }
 575
 576         for_each_cpu(cpu, drv->cpumask) {
 577                 device = &per_cpu(cpuidle_dev, cpu);
 578                 device->cpu = cpu;
 579
 580 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
 581                 /*
 582                  * On multiplatform for ARM, the coupled idle states could be
 583                  * enabled in the kernel even if the cpuidle driver does not
 584                  * use it. Note, coupled_cpus is a struct copy.
 585                  */
 586                 if (coupled_cpus)
 587                         device->coupled_cpus = *coupled_cpus;
 588 #endif
 589                 ret = cpuidle_register_device(device);
 590                 if (!ret)
 591                         continue;
 592
 593                 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
 594
 595                 cpuidle_unregister(drv);
 596                 break;
 597         }
 598
 599         return ret;
 600 }
 601 EXPORT_SYMBOL_GPL(cpuidle_register);
 602
 603 #ifdef CONFIG_SMP
 604
 605 /*
 606  * This function gets called when a part of the kernel has a new latency
 607  * requirement.  This means we need to get all processors out of their C-state,
 608  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 609  * wakes them all right up.
 610  */
 611 static int cpuidle_latency_notify(struct notifier_block *b,
 612                 unsigned long l, void *v)
 613 {
 614         wake_up_all_idle_cpus();
 615         return NOTIFY_OK;
 616 }
 617
 618 static struct notifier_block cpuidle_latency_notifier = {
 619         .notifier_call = cpuidle_latency_notify,
 620 };
 621
 622 static inline void latency_notifier_init(struct notifier_block *n)
 623 {
 624         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 625 }
 626
 627 #else /* CONFIG_SMP */
 628
 629 #define latency_notifier_init(x) do { } while (0)
 630
 631 #endif /* CONFIG_SMP */
 632
 633 /**
 634  * cpuidle_init - core initializer
 635  */
 636 static int __init cpuidle_init(void)
 637 {
 638         int ret;
 639
 640         if (cpuidle_disabled())
 641                 return -ENODEV;
 642
 643         ret = cpuidle_add_interface(cpu_subsys.dev_root);
 644         if (ret)
 645                 return ret;
 646
 647         latency_notifier_init(&cpuidle_latency_notifier);
 648
 649         return 0;
 650 }
 651
 652 module_param(off, int, 0444);
 653 core_initcall(cpuidle_init);