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