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