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         /*
 116          * trace_suspend_resume() called by tick_freeze() for the last CPU
 117          * executing it contains RCU usage regarded as invalid in the idle
 118          * context, so tell RCU about that.
 119          */
 120         RCU_NONIDLE(tick_freeze());
 121         /*
 122          * The state used here cannot be a "coupled" one, because the "coupled"
 123          * cpuidle mechanism enables interrupts and doing that with timekeeping
 124          * suspended is generally unsafe.
 125          */
 126         drv->states[index].enter_freeze(dev, drv, index);
 127         WARN_ON(!irqs_disabled());
 128         /*
 129          * timekeeping_resume() that will be called by tick_unfreeze() for the
 130          * first CPU executing it calls functions containing RCU read-side
 131          * critical sections, so tell RCU about that.
 132          */
 133         RCU_NONIDLE(tick_unfreeze());
 134 }
 135
 136 /**
 137  * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle.
 138  * @drv: cpuidle driver for the given CPU.
 139  * @dev: cpuidle device for the given CPU.
 140  *
 141  * If there are states with the ->enter_freeze callback, find the deepest of
 142  * them and enter it with frozen tick.
 143  */
 144 int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 145 {
 146         int index;
 147
 148         /*
 149          * Find the deepest state with ->enter_freeze present, which guarantees
 150          * that interrupts won't be enabled when it exits and allows the tick to
 151          * be frozen safely.
 152          */
 153         index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
 154         if (index >= 0)
 155                 enter_freeze_proper(drv, dev, index);
 156
 157         return index;
 158 }
 159 #endif /* CONFIG_SUSPEND */
 160
 161 /**
 162  * cpuidle_enter_state - enter the state and update stats
 163  * @dev: cpuidle device for this cpu
 164  * @drv: cpuidle driver for this cpu
 165  * @index: index into the states table in @drv of the state to enter
 166  */
 167 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
 168                         int index)
 169 {
 170         int entered_state;
 171
 172         struct cpuidle_state *target_state = &drv->states[index];
 173         bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
 174         ktime_t time_start, time_end;
 175         s64 diff;
 176
 177         /*
 178          * Tell the time framework to switch to a broadcast timer because our
 179          * local timer will be shut down.  If a local timer is used from another
 180          * CPU as a broadcast timer, this call may fail if it is not available.
 181          */
 182         if (broadcast && tick_broadcast_enter()) {
 183                 index = find_deepest_state(drv, dev, target_state->exit_latency,
 184                                            CPUIDLE_FLAG_TIMER_STOP, false);
 185                 if (index < 0) {
 186                         default_idle_call();
 187                         return -EBUSY;
 188                 }
 189                 target_state = &drv->states[index];
 190         }
 191
 192         /* Take note of the planned idle state. */
 193         sched_idle_set_state(target_state);
 194
 195         trace_cpu_idle_rcuidle(index, dev->cpu);
 196         time_start = ktime_get();
 197
 198         entered_state = target_state->enter(dev, drv, index);
 199
 200         time_end = ktime_get();
 201         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 202
 203         /* The cpu is no longer idle or about to enter idle. */
 204         sched_idle_set_state(NULL);
 205
 206         if (broadcast) {
 207                 if (WARN_ON_ONCE(!irqs_disabled()))
 208                         local_irq_disable();
 209
 210                 tick_broadcast_exit();
 211         }
 212
 213         if (!cpuidle_state_is_coupled(drv, entered_state))
 214                 local_irq_enable();
 215
 216         diff = ktime_to_us(ktime_sub(time_end, time_start));
 217         if (diff > INT_MAX)
 218                 diff = INT_MAX;
 219
 220         dev->last_residency = (int) diff;
 221
 222         if (entered_state >= 0) {
 223                 /* Update cpuidle counters */
 224                 /* This can be moved to within driver enter routine
 225                  * but that results in multiple copies of same code.
 226                  */
 227                 dev->states_usage[entered_state].time += dev->last_residency;
 228                 dev->states_usage[entered_state].usage++;
 229         } else {
 230                 dev->last_residency = 0;
 231         }
 232
 233         return entered_state;
 234 }
 235
 236 /**
 237  * cpuidle_select - ask the cpuidle framework to choose an idle state
 238  *
 239  * @drv: the cpuidle driver
 240  * @dev: the cpuidle device
 241  *
 242  * Returns the index of the idle state.
 243  */
 244 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 245 {
 246         return cpuidle_curr_governor->select(drv, dev);
 247 }
 248
 249 /**
 250  * cpuidle_enter - enter into the specified idle state
 251  *
 252  * @drv:   the cpuidle driver tied with the cpu
 253  * @dev:   the cpuidle device
 254  * @index: the index in the idle state table
 255  *
 256  * Returns the index in the idle state, < 0 in case of error.
 257  * The error code depends on the backend driver
 258  */
 259 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 260                   int index)
 261 {
 262         if (cpuidle_state_is_coupled(drv, index))
 263                 return cpuidle_enter_state_coupled(dev, drv, index);
 264         return cpuidle_enter_state(dev, drv, index);
 265 }
 266
 267 /**
 268  * cpuidle_reflect - tell the underlying governor what was the state
 269  * we were in
 270  *
 271  * @dev  : the cpuidle device
 272  * @index: the index in the idle state table
 273  *
 274  */
 275 void cpuidle_reflect(struct cpuidle_device *dev, int index)
 276 {
 277         if (cpuidle_curr_governor->reflect && index >= 0)
 278                 cpuidle_curr_governor->reflect(dev, index);
 279 }
 280
 281 /**
 282  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 283  */
 284 void cpuidle_install_idle_handler(void)
 285 {
 286         if (enabled_devices) {
 287                 /* Make sure all changes finished before we switch to new idle */
 288                 smp_wmb();
 289                 initialized = 1;
 290         }
 291 }
 292
 293 /**
 294  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 295  */
 296 void cpuidle_uninstall_idle_handler(void)
 297 {
 298         if (enabled_devices) {
 299                 initialized = 0;
 300                 wake_up_all_idle_cpus();
 301         }
 302
 303         /*
 304          * Make sure external observers (such as the scheduler)
 305          * are done looking at pointed idle states.
 306          */
 307         synchronize_rcu();
 308 }
 309
 310 /**
 311  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 312  */
 313 void cpuidle_pause_and_lock(void)
 314 {
 315         mutex_lock(&cpuidle_lock);
 316         cpuidle_uninstall_idle_handler();
 317 }
 318
 319 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 320
 321 /**
 322  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 323  */
 324 void cpuidle_resume_and_unlock(void)
 325 {
 326         cpuidle_install_idle_handler();
 327         mutex_unlock(&cpuidle_lock);
 328 }
 329
 330 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 331
 332 /* Currently used in suspend/resume path to suspend cpuidle */
 333 void cpuidle_pause(void)
 334 {
 335         mutex_lock(&cpuidle_lock);
 336         cpuidle_uninstall_idle_handler();
 337         mutex_unlock(&cpuidle_lock);
 338 }
 339
 340 /* Currently used in suspend/resume path to resume cpuidle */
 341 void cpuidle_resume(void)
 342 {
 343         mutex_lock(&cpuidle_lock);
 344         cpuidle_install_idle_handler();
 345         mutex_unlock(&cpuidle_lock);
 346 }
 347
 348 /**
 349  * cpuidle_enable_device - enables idle PM for a CPU
 350  * @dev: the CPU
 351  *
 352  * This function must be called between cpuidle_pause_and_lock and
 353  * cpuidle_resume_and_unlock when used externally.
 354  */
 355 int cpuidle_enable_device(struct cpuidle_device *dev)
 356 {
 357         int ret;
 358         struct cpuidle_driver *drv;
 359
 360         if (!dev)
 361                 return -EINVAL;
 362
 363         if (dev->enabled)
 364                 return 0;
 365
 366         drv = cpuidle_get_cpu_driver(dev);
 367
 368         if (!drv || !cpuidle_curr_governor)
 369                 return -EIO;
 370
 371         if (!dev->registered)
 372                 return -EINVAL;
 373
 374         ret = cpuidle_add_device_sysfs(dev);
 375         if (ret)
 376                 return ret;
 377
 378         if (cpuidle_curr_governor->enable &&
 379             (ret = cpuidle_curr_governor->enable(drv, dev)))
 380                 goto fail_sysfs;
 381
 382         smp_wmb();
 383
 384         dev->enabled = 1;
 385
 386         enabled_devices++;
 387         return 0;
 388
 389 fail_sysfs:
 390         cpuidle_remove_device_sysfs(dev);
 391
 392         return ret;
 393 }
 394
 395 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 396
 397 /**
 398  * cpuidle_disable_device - disables idle PM for a CPU
 399  * @dev: the CPU
 400  *
 401  * This function must be called between cpuidle_pause_and_lock and
 402  * cpuidle_resume_and_unlock when used externally.
 403  */
 404 void cpuidle_disable_device(struct cpuidle_device *dev)
 405 {
 406         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 407
 408         if (!dev || !dev->enabled)
 409                 return;
 410
 411         if (!drv || !cpuidle_curr_governor)
 412                 return;
 413
 414         dev->enabled = 0;
 415
 416         if (cpuidle_curr_governor->disable)
 417                 cpuidle_curr_governor->disable(drv, dev);
 418
 419         cpuidle_remove_device_sysfs(dev);
 420         enabled_devices--;
 421 }
 422
 423 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 424
 425 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
 426 {
 427         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 428
 429         list_del(&dev->device_list);
 430         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 431         module_put(drv->owner);
 432 }
 433
 434 static void __cpuidle_device_init(struct cpuidle_device *dev)
 435 {
 436         memset(dev->states_usage, 0, sizeof(dev->states_usage));
 437         dev->last_residency = 0;
 438 }
 439
 440 /**
 441  * __cpuidle_register_device - internal register function called before register
 442  * and enable routines
 443  * @dev: the cpu
 444  *
 445  * cpuidle_lock mutex must be held before this is called
 446  */
 447 static int __cpuidle_register_device(struct cpuidle_device *dev)
 448 {
 449         int ret;
 450         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 451
 452         if (!try_module_get(drv->owner))
 453                 return -EINVAL;
 454
 455         per_cpu(cpuidle_devices, dev->cpu) = dev;
 456         list_add(&dev->device_list, &cpuidle_detected_devices);
 457
 458         ret = cpuidle_coupled_register_device(dev);
 459         if (ret)
 460                 __cpuidle_unregister_device(dev);
 461         else
 462                 dev->registered = 1;
 463
 464         return ret;
 465 }
 466
 467 /**
 468  * cpuidle_register_device - registers a CPU's idle PM feature
 469  * @dev: the cpu
 470  */
 471 int cpuidle_register_device(struct cpuidle_device *dev)
 472 {
 473         int ret = -EBUSY;
 474
 475         if (!dev)
 476                 return -EINVAL;
 477
 478         mutex_lock(&cpuidle_lock);
 479
 480         if (dev->registered)
 481                 goto out_unlock;
 482
 483         __cpuidle_device_init(dev);
 484
 485         ret = __cpuidle_register_device(dev);
 486         if (ret)
 487                 goto out_unlock;
 488
 489         ret = cpuidle_add_sysfs(dev);
 490         if (ret)
 491                 goto out_unregister;
 492
 493         ret = cpuidle_enable_device(dev);
 494         if (ret)
 495                 goto out_sysfs;
 496
 497         cpuidle_install_idle_handler();
 498
 499 out_unlock:
 500         mutex_unlock(&cpuidle_lock);
 501
 502         return ret;
 503
 504 out_sysfs:
 505         cpuidle_remove_sysfs(dev);
 506 out_unregister:
 507         __cpuidle_unregister_device(dev);
 508         goto out_unlock;
 509 }
 510
 511 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 512
 513 /**
 514  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 515  * @dev: the cpu
 516  */
 517 void cpuidle_unregister_device(struct cpuidle_device *dev)
 518 {
 519         if (!dev || dev->registered == 0)
 520                 return;
 521
 522         cpuidle_pause_and_lock();
 523
 524         cpuidle_disable_device(dev);
 525
 526         cpuidle_remove_sysfs(dev);
 527
 528         __cpuidle_unregister_device(dev);
 529
 530         cpuidle_coupled_unregister_device(dev);
 531
 532         cpuidle_resume_and_unlock();
 533 }
 534
 535 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 536
 537 /**
 538  * cpuidle_unregister: unregister a driver and the devices. This function
 539  * can be used only if the driver has been previously registered through
 540  * the cpuidle_register function.
 541  *
 542  * @drv: a valid pointer to a struct cpuidle_driver
 543  */
 544 void cpuidle_unregister(struct cpuidle_driver *drv)
 545 {
 546         int cpu;
 547         struct cpuidle_device *device;
 548
 549         for_each_cpu(cpu, drv->cpumask) {
 550                 device = &per_cpu(cpuidle_dev, cpu);
 551                 cpuidle_unregister_device(device);
 552         }
 553
 554         cpuidle_unregister_driver(drv);
 555 }
 556 EXPORT_SYMBOL_GPL(cpuidle_unregister);
 557
 558 /**
 559  * cpuidle_register: registers the driver and the cpu devices with the
 560  * coupled_cpus passed as parameter. This function is used for all common
 561  * initialization pattern there are in the arch specific drivers. The
 562  * devices is globally defined in this file.
 563  *
 564  * @drv         : a valid pointer to a struct cpuidle_driver
 565  * @coupled_cpus: a cpumask for the coupled states
 566  *
 567  * Returns 0 on success, < 0 otherwise
 568  */
 569 int cpuidle_register(struct cpuidle_driver *drv,
 570                      const struct cpumask *const coupled_cpus)
 571 {
 572         int ret, cpu;
 573         struct cpuidle_device *device;
 574
 575         ret = cpuidle_register_driver(drv);
 576         if (ret) {
 577                 pr_err("failed to register cpuidle driver\n");
 578                 return ret;
 579         }
 580
 581         for_each_cpu(cpu, drv->cpumask) {
 582                 device = &per_cpu(cpuidle_dev, cpu);
 583                 device->cpu = cpu;
 584
 585 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
 586                 /*
 587                  * On multiplatform for ARM, the coupled idle states could be
 588                  * enabled in the kernel even if the cpuidle driver does not
 589                  * use it. Note, coupled_cpus is a struct copy.
 590                  */
 591                 if (coupled_cpus)
 592                         device->coupled_cpus = *coupled_cpus;
 593 #endif
 594                 ret = cpuidle_register_device(device);
 595                 if (!ret)
 596                         continue;
 597
 598                 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
 599
 600                 cpuidle_unregister(drv);
 601                 break;
 602         }
 603
 604         return ret;
 605 }
 606 EXPORT_SYMBOL_GPL(cpuidle_register);
 607
 608 #ifdef CONFIG_SMP
 609
 610 /*
 611  * This function gets called when a part of the kernel has a new latency
 612  * requirement.  This means we need to get all processors out of their C-state,
 613  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 614  * wakes them all right up.
 615  */
 616 static int cpuidle_latency_notify(struct notifier_block *b,
 617                 unsigned long l, void *v)
 618 {
 619         wake_up_all_idle_cpus();
 620         return NOTIFY_OK;
 621 }
 622
 623 static struct notifier_block cpuidle_latency_notifier = {
 624         .notifier_call = cpuidle_latency_notify,
 625 };
 626
 627 static inline void latency_notifier_init(struct notifier_block *n)
 628 {
 629         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 630 }
 631
 632 #else /* CONFIG_SMP */
 633
 634 #define latency_notifier_init(x) do { } while (0)
 635
 636 #endif /* CONFIG_SMP */
 637
 638 /**
 639  * cpuidle_init - core initializer
 640  */
 641 static int __init cpuidle_init(void)
 642 {
 643         int ret;
 644
 645         if (cpuidle_disabled())
 646                 return -ENODEV;
 647
 648         ret = cpuidle_add_interface(cpu_subsys.dev_root);
 649         if (ret)
 650                 return ret;
 651
 652         latency_notifier_init(&cpuidle_latency_notifier);
 653
 654         return 0;
 655 }
 656
 657 module_param(off, int, 0444);
 658 core_initcall(cpuidle_init);