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 <trace/events/power.h>
  23
  24 #include "cpuidle.h"
  25
  26 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
  27 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
  28
  29 DEFINE_MUTEX(cpuidle_lock);
  30 LIST_HEAD(cpuidle_detected_devices);
  31
  32 static int enabled_devices;
  33 static int off __read_mostly;
  34 static int initialized __read_mostly;
  35
  36 int cpuidle_disabled(void)
  37 {
  38         return off;
  39 }
  40 void disable_cpuidle(void)
  41 {
  42         off = 1;
  43 }
  44
  45 /**
  46  * cpuidle_play_dead - cpu off-lining
  47  *
  48  * Returns in case of an error or no driver
  49  */
  50 int cpuidle_play_dead(void)
  51 {
  52         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  53         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  54         int i;
  55
  56         if (!drv)
  57                 return -ENODEV;
  58
  59         /* Find lowest-power state that supports long-term idle */
  60         for (i = drv->state_count - 1; i >= CPUIDLE_DRIVER_STATE_START; i--)
  61                 if (drv->states[i].enter_dead)
  62                         return drv->states[i].enter_dead(dev, i);
  63
  64         return -ENODEV;
  65 }
  66
  67 /**
  68  * cpuidle_enter_state - enter the state and update stats
  69  * @dev: cpuidle device for this cpu
  70  * @drv: cpuidle driver for this cpu
  71  * @next_state: index into drv->states of the state to enter
  72  */
  73 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
  74                         int index)
  75 {
  76         int entered_state;
  77
  78         struct cpuidle_state *target_state = &drv->states[index];
  79         ktime_t time_start, time_end;
  80         s64 diff;
  81
  82         time_start = ktime_get();
  83
  84         entered_state = target_state->enter(dev, drv, index);
  85
  86         time_end = ktime_get();
  87
  88         local_irq_enable();
  89
  90         diff = ktime_to_us(ktime_sub(time_end, time_start));
  91         if (diff > INT_MAX)
  92                 diff = INT_MAX;
  93
  94         dev->last_residency = (int) diff;
  95
  96         if (entered_state >= 0) {
  97                 /* Update cpuidle counters */
  98                 /* This can be moved to within driver enter routine
  99                  * but that results in multiple copies of same code.
 100                  */
 101                 dev->states_usage[entered_state].time += dev->last_residency;
 102                 dev->states_usage[entered_state].usage++;
 103         } else {
 104                 dev->last_residency = 0;
 105         }
 106
 107         return entered_state;
 108 }
 109
 110 /**
 111  * cpuidle_idle_call - the main idle loop
 112  *
 113  * NOTE: no locks or semaphores should be used here
 114  * return non-zero on failure
 115  */
 116 int cpuidle_idle_call(void)
 117 {
 118         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
 119         struct cpuidle_driver *drv;
 120         int next_state, entered_state;
 121         bool broadcast;
 122
 123         if (off || !initialized)
 124                 return -ENODEV;
 125
 126         /* check if the device is ready */
 127         if (!dev || !dev->enabled)
 128                 return -EBUSY;
 129
 130         drv = cpuidle_get_cpu_driver(dev);
 131
 132         /* ask the governor for the next state */
 133         next_state = cpuidle_curr_governor->select(drv, dev);
 134         if (need_resched()) {
 135                 dev->last_residency = 0;
 136                 /* give the governor an opportunity to reflect on the outcome */
 137                 if (cpuidle_curr_governor->reflect)
 138                         cpuidle_curr_governor->reflect(dev, next_state);
 139                 local_irq_enable();
 140                 return 0;
 141         }
 142
 143         trace_cpu_idle_rcuidle(next_state, dev->cpu);
 144
 145         broadcast = !!(drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP);
 146
 147         if (broadcast)
 148                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu);
 149
 150         if (cpuidle_state_is_coupled(dev, drv, next_state))
 151                 entered_state = cpuidle_enter_state_coupled(dev, drv,
 152                                                             next_state);
 153         else
 154                 entered_state = cpuidle_enter_state(dev, drv, next_state);
 155
 156         if (broadcast)
 157                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
 158
 159         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 160
 161         /* give the governor an opportunity to reflect on the outcome */
 162         if (cpuidle_curr_governor->reflect)
 163                 cpuidle_curr_governor->reflect(dev, entered_state);
 164
 165         return 0;
 166 }
 167
 168 /**
 169  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 170  */
 171 void cpuidle_install_idle_handler(void)
 172 {
 173         if (enabled_devices) {
 174                 /* Make sure all changes finished before we switch to new idle */
 175                 smp_wmb();
 176                 initialized = 1;
 177         }
 178 }
 179
 180 /**
 181  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 182  */
 183 void cpuidle_uninstall_idle_handler(void)
 184 {
 185         if (enabled_devices) {
 186                 initialized = 0;
 187                 kick_all_cpus_sync();
 188         }
 189 }
 190
 191 /**
 192  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 193  */
 194 void cpuidle_pause_and_lock(void)
 195 {
 196         mutex_lock(&cpuidle_lock);
 197         cpuidle_uninstall_idle_handler();
 198 }
 199
 200 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 201
 202 /**
 203  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 204  */
 205 void cpuidle_resume_and_unlock(void)
 206 {
 207         cpuidle_install_idle_handler();
 208         mutex_unlock(&cpuidle_lock);
 209 }
 210
 211 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 212
 213 /* Currently used in suspend/resume path to suspend cpuidle */
 214 void cpuidle_pause(void)
 215 {
 216         mutex_lock(&cpuidle_lock);
 217         cpuidle_uninstall_idle_handler();
 218         mutex_unlock(&cpuidle_lock);
 219 }
 220
 221 /* Currently used in suspend/resume path to resume cpuidle */
 222 void cpuidle_resume(void)
 223 {
 224         mutex_lock(&cpuidle_lock);
 225         cpuidle_install_idle_handler();
 226         mutex_unlock(&cpuidle_lock);
 227 }
 228
 229 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
 230 static int poll_idle(struct cpuidle_device *dev,
 231                 struct cpuidle_driver *drv, int index)
 232 {
 233         ktime_t t1, t2;
 234         s64 diff;
 235
 236         t1 = ktime_get();
 237         local_irq_enable();
 238         while (!need_resched())
 239                 cpu_relax();
 240
 241         t2 = ktime_get();
 242         diff = ktime_to_us(ktime_sub(t2, t1));
 243         if (diff > INT_MAX)
 244                 diff = INT_MAX;
 245
 246         dev->last_residency = (int) diff;
 247
 248         return index;
 249 }
 250
 251 static void poll_idle_init(struct cpuidle_driver *drv)
 252 {
 253         struct cpuidle_state *state = &drv->states[0];
 254
 255         snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
 256         snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
 257         state->exit_latency = 0;
 258         state->target_residency = 0;
 259         state->power_usage = -1;
 260         state->flags = 0;
 261         state->enter = poll_idle;
 262         state->disabled = false;
 263 }
 264 #else
 265 static void poll_idle_init(struct cpuidle_driver *drv) {}
 266 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
 267
 268 /**
 269  * cpuidle_enable_device - enables idle PM for a CPU
 270  * @dev: the CPU
 271  *
 272  * This function must be called between cpuidle_pause_and_lock and
 273  * cpuidle_resume_and_unlock when used externally.
 274  */
 275 int cpuidle_enable_device(struct cpuidle_device *dev)
 276 {
 277         int ret;
 278         struct cpuidle_driver *drv;
 279
 280         if (!dev)
 281                 return -EINVAL;
 282
 283         if (dev->enabled)
 284                 return 0;
 285
 286         drv = cpuidle_get_cpu_driver(dev);
 287
 288         if (!drv || !cpuidle_curr_governor)
 289                 return -EIO;
 290
 291         if (!dev->registered)
 292                 return -EINVAL;
 293
 294         if (!dev->state_count)
 295                 dev->state_count = drv->state_count;
 296
 297         poll_idle_init(drv);
 298
 299         ret = cpuidle_add_device_sysfs(dev);
 300         if (ret)
 301                 return ret;
 302
 303         if (cpuidle_curr_governor->enable &&
 304             (ret = cpuidle_curr_governor->enable(drv, dev)))
 305                 goto fail_sysfs;
 306
 307         smp_wmb();
 308
 309         dev->enabled = 1;
 310
 311         enabled_devices++;
 312         return 0;
 313
 314 fail_sysfs:
 315         cpuidle_remove_device_sysfs(dev);
 316
 317         return ret;
 318 }
 319
 320 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 321
 322 /**
 323  * cpuidle_disable_device - disables idle PM for a CPU
 324  * @dev: the CPU
 325  *
 326  * This function must be called between cpuidle_pause_and_lock and
 327  * cpuidle_resume_and_unlock when used externally.
 328  */
 329 void cpuidle_disable_device(struct cpuidle_device *dev)
 330 {
 331         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 332
 333         if (!dev || !dev->enabled)
 334                 return;
 335
 336         if (!drv || !cpuidle_curr_governor)
 337                 return;
 338
 339         dev->enabled = 0;
 340
 341         if (cpuidle_curr_governor->disable)
 342                 cpuidle_curr_governor->disable(drv, dev);
 343
 344         cpuidle_remove_device_sysfs(dev);
 345         enabled_devices--;
 346 }
 347
 348 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 349
 350 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
 351 {
 352         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 353
 354         list_del(&dev->device_list);
 355         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 356         module_put(drv->owner);
 357 }
 358
 359 static void __cpuidle_device_init(struct cpuidle_device *dev)
 360 {
 361         memset(dev->states_usage, 0, sizeof(dev->states_usage));
 362         dev->last_residency = 0;
 363 }
 364
 365 /**
 366  * __cpuidle_register_device - internal register function called before register
 367  * and enable routines
 368  * @dev: the cpu
 369  *
 370  * cpuidle_lock mutex must be held before this is called
 371  */
 372 static int __cpuidle_register_device(struct cpuidle_device *dev)
 373 {
 374         int ret;
 375         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 376
 377         if (!try_module_get(drv->owner))
 378                 return -EINVAL;
 379
 380         per_cpu(cpuidle_devices, dev->cpu) = dev;
 381         list_add(&dev->device_list, &cpuidle_detected_devices);
 382
 383         ret = cpuidle_coupled_register_device(dev);
 384         if (ret)
 385                 __cpuidle_unregister_device(dev);
 386         else
 387                 dev->registered = 1;
 388
 389         return ret;
 390 }
 391
 392 /**
 393  * cpuidle_register_device - registers a CPU's idle PM feature
 394  * @dev: the cpu
 395  */
 396 int cpuidle_register_device(struct cpuidle_device *dev)
 397 {
 398         int ret = -EBUSY;
 399
 400         if (!dev)
 401                 return -EINVAL;
 402
 403         mutex_lock(&cpuidle_lock);
 404
 405         if (dev->registered)
 406                 goto out_unlock;
 407
 408         __cpuidle_device_init(dev);
 409
 410         ret = __cpuidle_register_device(dev);
 411         if (ret)
 412                 goto out_unlock;
 413
 414         ret = cpuidle_add_sysfs(dev);
 415         if (ret)
 416                 goto out_unregister;
 417
 418         ret = cpuidle_enable_device(dev);
 419         if (ret)
 420                 goto out_sysfs;
 421
 422         cpuidle_install_idle_handler();
 423
 424 out_unlock:
 425         mutex_unlock(&cpuidle_lock);
 426
 427         return ret;
 428
 429 out_sysfs:
 430         cpuidle_remove_sysfs(dev);
 431 out_unregister:
 432         __cpuidle_unregister_device(dev);
 433         goto out_unlock;
 434 }
 435
 436 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 437
 438 /**
 439  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 440  * @dev: the cpu
 441  */
 442 void cpuidle_unregister_device(struct cpuidle_device *dev)
 443 {
 444         if (dev->registered == 0)
 445                 return;
 446
 447         cpuidle_pause_and_lock();
 448
 449         cpuidle_disable_device(dev);
 450
 451         cpuidle_remove_sysfs(dev);
 452
 453         __cpuidle_unregister_device(dev);
 454
 455         cpuidle_coupled_unregister_device(dev);
 456
 457         cpuidle_resume_and_unlock();
 458 }
 459
 460 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 461
 462 /**
 463  * cpuidle_unregister: unregister a driver and the devices. This function
 464  * can be used only if the driver has been previously registered through
 465  * the cpuidle_register function.
 466  *
 467  * @drv: a valid pointer to a struct cpuidle_driver
 468  */
 469 void cpuidle_unregister(struct cpuidle_driver *drv)
 470 {
 471         int cpu;
 472         struct cpuidle_device *device;
 473
 474         for_each_cpu(cpu, drv->cpumask) {
 475                 device = &per_cpu(cpuidle_dev, cpu);
 476                 cpuidle_unregister_device(device);
 477         }
 478
 479         cpuidle_unregister_driver(drv);
 480 }
 481 EXPORT_SYMBOL_GPL(cpuidle_unregister);
 482
 483 /**
 484  * cpuidle_register: registers the driver and the cpu devices with the
 485  * coupled_cpus passed as parameter. This function is used for all common
 486  * initialization pattern there are in the arch specific drivers. The
 487  * devices is globally defined in this file.
 488  *
 489  * @drv         : a valid pointer to a struct cpuidle_driver
 490  * @coupled_cpus: a cpumask for the coupled states
 491  *
 492  * Returns 0 on success, < 0 otherwise
 493  */
 494 int cpuidle_register(struct cpuidle_driver *drv,
 495                      const struct cpumask *const coupled_cpus)
 496 {
 497         int ret, cpu;
 498         struct cpuidle_device *device;
 499
 500         ret = cpuidle_register_driver(drv);
 501         if (ret) {
 502                 pr_err("failed to register cpuidle driver\n");
 503                 return ret;
 504         }
 505
 506         for_each_cpu(cpu, drv->cpumask) {
 507                 device = &per_cpu(cpuidle_dev, cpu);
 508                 device->cpu = cpu;
 509
 510 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
 511                 /*
 512                  * On multiplatform for ARM, the coupled idle states could be
 513                  * enabled in the kernel even if the cpuidle driver does not
 514                  * use it. Note, coupled_cpus is a struct copy.
 515                  */
 516                 if (coupled_cpus)
 517                         device->coupled_cpus = *coupled_cpus;
 518 #endif
 519                 ret = cpuidle_register_device(device);
 520                 if (!ret)
 521                         continue;
 522
 523                 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
 524
 525                 cpuidle_unregister(drv);
 526                 break;
 527         }
 528
 529         return ret;
 530 }
 531 EXPORT_SYMBOL_GPL(cpuidle_register);
 532
 533 #ifdef CONFIG_SMP
 534
 535 static void smp_callback(void *v)
 536 {
 537         /* we already woke the CPU up, nothing more to do */
 538 }
 539
 540 /*
 541  * This function gets called when a part of the kernel has a new latency
 542  * requirement.  This means we need to get all processors out of their C-state,
 543  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 544  * wakes them all right up.
 545  */
 546 static int cpuidle_latency_notify(struct notifier_block *b,
 547                 unsigned long l, void *v)
 548 {
 549         smp_call_function(smp_callback, NULL, 1);
 550         return NOTIFY_OK;
 551 }
 552
 553 static struct notifier_block cpuidle_latency_notifier = {
 554         .notifier_call = cpuidle_latency_notify,
 555 };
 556
 557 static inline void latency_notifier_init(struct notifier_block *n)
 558 {
 559         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 560 }
 561
 562 #else /* CONFIG_SMP */
 563
 564 #define latency_notifier_init(x) do { } while (0)
 565
 566 #endif /* CONFIG_SMP */
 567
 568 /**
 569  * cpuidle_init - core initializer
 570  */
 571 static int __init cpuidle_init(void)
 572 {
 573         int ret;
 574
 575         if (cpuidle_disabled())
 576                 return -ENODEV;
 577
 578         ret = cpuidle_add_interface(cpu_subsys.dev_root);
 579         if (ret)
 580                 return ret;
 581
 582         latency_notifier_init(&cpuidle_latency_notifier);
 583
 584         return 0;
 585 }
 586
 587 module_param(off, int, 0444);
 588 core_initcall(cpuidle_init);