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/kernel.h>
  12 #include <linux/mutex.h>
  13 #include <linux/sched.h>
  14 #include <linux/notifier.h>
  15 #include <linux/pm_qos.h>
  16 #include <linux/cpu.h>
  17 #include <linux/cpuidle.h>
  18 #include <linux/ktime.h>
  19 #include <linux/hrtimer.h>
  20 #include <linux/module.h>
  21 #include <trace/events/power.h>
  22
  23 #include "cpuidle.h"
  24
  25 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
  26
  27 DEFINE_MUTEX(cpuidle_lock);
  28 LIST_HEAD(cpuidle_detected_devices);
  29
  30 static int enabled_devices;
  31 static int off __read_mostly;
  32 static int initialized __read_mostly;
  33
  34 int cpuidle_disabled(void)
  35 {
  36         return off;
  37 }
  38 void disable_cpuidle(void)
  39 {
  40         off = 1;
  41 }
  42
  43 static int __cpuidle_register_device(struct cpuidle_device *dev);
  44
  45 static inline int cpuidle_enter(struct cpuidle_device *dev,
  46                                 struct cpuidle_driver *drv, int index)
  47 {
  48         struct cpuidle_state *target_state = &drv->states[index];
  49         return target_state->enter(dev, drv, index);
  50 }
  51
  52 static inline int cpuidle_enter_tk(struct cpuidle_device *dev,
  53                                struct cpuidle_driver *drv, int index)
  54 {
  55         return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter);
  56 }
  57
  58 typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev,
  59                                struct cpuidle_driver *drv, int index);
  60
  61 static cpuidle_enter_t cpuidle_enter_ops;
  62
  63 /**
  64  * cpuidle_play_dead - cpu off-lining
  65  *
  66  * Returns in case of an error or no driver
  67  */
  68 int cpuidle_play_dead(void)
  69 {
  70         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  71         struct cpuidle_driver *drv = cpuidle_get_driver();
  72         int i, dead_state = -1;
  73         int power_usage = -1;
  74
  75         if (!drv)
  76                 return -ENODEV;
  77
  78         /* Find lowest-power state that supports long-term idle */
  79         for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
  80                 struct cpuidle_state *s = &drv->states[i];
  81
  82                 if (s->power_usage < power_usage && s->enter_dead) {
  83                         power_usage = s->power_usage;
  84                         dead_state = i;
  85                 }
  86         }
  87
  88         if (dead_state != -1)
  89                 return drv->states[dead_state].enter_dead(dev, dead_state);
  90
  91         return -ENODEV;
  92 }
  93
  94 /**
  95  * cpuidle_idle_call - the main idle loop
  96  *
  97  * NOTE: no locks or semaphores should be used here
  98  * return non-zero on failure
  99  */
 100 int cpuidle_idle_call(void)
 101 {
 102         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
 103         struct cpuidle_driver *drv = cpuidle_get_driver();
 104         int next_state, entered_state;
 105
 106         if (off)
 107                 return -ENODEV;
 108
 109         if (!initialized)
 110                 return -ENODEV;
 111
 112         /* check if the device is ready */
 113         if (!dev || !dev->enabled)
 114                 return -EBUSY;
 115
 116 #if 0
 117         /* shows regressions, re-enable for 2.6.29 */
 118         /*
 119          * run any timers that can be run now, at this point
 120          * before calculating the idle duration etc.
 121          */
 122         hrtimer_peek_ahead_timers();
 123 #endif
 124
 125         /* ask the governor for the next state */
 126         next_state = cpuidle_curr_governor->select(drv, dev);
 127         if (need_resched()) {
 128                 local_irq_enable();
 129                 return 0;
 130         }
 131
 132         trace_power_start_rcuidle(POWER_CSTATE, next_state, dev->cpu);
 133         trace_cpu_idle_rcuidle(next_state, dev->cpu);
 134
 135         entered_state = cpuidle_enter_ops(dev, drv, next_state);
 136
 137         trace_power_end_rcuidle(dev->cpu);
 138         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 139
 140         if (entered_state >= 0) {
 141                 /* Update cpuidle counters */
 142                 /* This can be moved to within driver enter routine
 143                  * but that results in multiple copies of same code.
 144                  */
 145                 dev->states_usage[entered_state].time +=
 146                                 (unsigned long long)dev->last_residency;
 147                 dev->states_usage[entered_state].usage++;
 148         } else {
 149                 dev->last_residency = 0;
 150         }
 151
 152         /* give the governor an opportunity to reflect on the outcome */
 153         if (cpuidle_curr_governor->reflect)
 154                 cpuidle_curr_governor->reflect(dev, entered_state);
 155
 156         return 0;
 157 }
 158
 159 /**
 160  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 161  */
 162 void cpuidle_install_idle_handler(void)
 163 {
 164         if (enabled_devices) {
 165                 /* Make sure all changes finished before we switch to new idle */
 166                 smp_wmb();
 167                 initialized = 1;
 168         }
 169 }
 170
 171 /**
 172  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 173  */
 174 void cpuidle_uninstall_idle_handler(void)
 175 {
 176         if (enabled_devices) {
 177                 initialized = 0;
 178                 kick_all_cpus_sync();
 179         }
 180 }
 181
 182 /**
 183  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 184  */
 185 void cpuidle_pause_and_lock(void)
 186 {
 187         mutex_lock(&cpuidle_lock);
 188         cpuidle_uninstall_idle_handler();
 189 }
 190
 191 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 192
 193 /**
 194  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 195  */
 196 void cpuidle_resume_and_unlock(void)
 197 {
 198         cpuidle_install_idle_handler();
 199         mutex_unlock(&cpuidle_lock);
 200 }
 201
 202 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 203
 204 /* Currently used in suspend/resume path to suspend cpuidle */
 205 void cpuidle_pause(void)
 206 {
 207         mutex_lock(&cpuidle_lock);
 208         cpuidle_uninstall_idle_handler();
 209         mutex_unlock(&cpuidle_lock);
 210 }
 211
 212 /* Currently used in suspend/resume path to resume cpuidle */
 213 void cpuidle_resume(void)
 214 {
 215         mutex_lock(&cpuidle_lock);
 216         cpuidle_install_idle_handler();
 217         mutex_unlock(&cpuidle_lock);
 218 }
 219
 220 /**
 221  * cpuidle_wrap_enter - performs timekeeping and irqen around enter function
 222  * @dev: pointer to a valid cpuidle_device object
 223  * @drv: pointer to a valid cpuidle_driver object
 224  * @index: index of the target cpuidle state.
 225  */
 226 int cpuidle_wrap_enter(struct cpuidle_device *dev,
 227                                 struct cpuidle_driver *drv, int index,
 228                                 int (*enter)(struct cpuidle_device *dev,
 229                                         struct cpuidle_driver *drv, int index))
 230 {
 231         ktime_t time_start, time_end;
 232         s64 diff;
 233
 234         time_start = ktime_get();
 235
 236         index = enter(dev, drv, index);
 237
 238         time_end = ktime_get();
 239
 240         local_irq_enable();
 241
 242         diff = ktime_to_us(ktime_sub(time_end, time_start));
 243         if (diff > INT_MAX)
 244                 diff = INT_MAX;
 245
 246         dev->last_residency = (int) diff;
 247
 248         return index;
 249 }
 250
 251 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
 252 static int poll_idle(struct cpuidle_device *dev,
 253                 struct cpuidle_driver *drv, int index)
 254 {
 255         ktime_t t1, t2;
 256         s64 diff;
 257
 258         t1 = ktime_get();
 259         local_irq_enable();
 260         while (!need_resched())
 261                 cpu_relax();
 262
 263         t2 = ktime_get();
 264         diff = ktime_to_us(ktime_sub(t2, t1));
 265         if (diff > INT_MAX)
 266                 diff = INT_MAX;
 267
 268         dev->last_residency = (int) diff;
 269
 270         return index;
 271 }
 272
 273 static void poll_idle_init(struct cpuidle_driver *drv)
 274 {
 275         struct cpuidle_state *state = &drv->states[0];
 276
 277         snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
 278         snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
 279         state->exit_latency = 0;
 280         state->target_residency = 0;
 281         state->power_usage = -1;
 282         state->flags = 0;
 283         state->enter = poll_idle;
 284         state->disabled = false;
 285 }
 286 #else
 287 static void poll_idle_init(struct cpuidle_driver *drv) {}
 288 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
 289
 290 /**
 291  * cpuidle_enable_device - enables idle PM for a CPU
 292  * @dev: the CPU
 293  *
 294  * This function must be called between cpuidle_pause_and_lock and
 295  * cpuidle_resume_and_unlock when used externally.
 296  */
 297 int cpuidle_enable_device(struct cpuidle_device *dev)
 298 {
 299         int ret, i;
 300         struct cpuidle_driver *drv = cpuidle_get_driver();
 301
 302         if (dev->enabled)
 303                 return 0;
 304         if (!drv || !cpuidle_curr_governor)
 305                 return -EIO;
 306         if (!dev->state_count)
 307                 dev->state_count = drv->state_count;
 308
 309         if (dev->registered == 0) {
 310                 ret = __cpuidle_register_device(dev);
 311                 if (ret)
 312                         return ret;
 313         }
 314
 315         cpuidle_enter_ops = drv->en_core_tk_irqen ?
 316                 cpuidle_enter_tk : cpuidle_enter;
 317
 318         poll_idle_init(drv);
 319
 320         if ((ret = cpuidle_add_state_sysfs(dev)))
 321                 return ret;
 322
 323         if (cpuidle_curr_governor->enable &&
 324             (ret = cpuidle_curr_governor->enable(drv, dev)))
 325                 goto fail_sysfs;
 326
 327         for (i = 0; i < dev->state_count; i++) {
 328                 dev->states_usage[i].usage = 0;
 329                 dev->states_usage[i].time = 0;
 330         }
 331         dev->last_residency = 0;
 332
 333         smp_wmb();
 334
 335         dev->enabled = 1;
 336
 337         enabled_devices++;
 338         return 0;
 339
 340 fail_sysfs:
 341         cpuidle_remove_state_sysfs(dev);
 342
 343         return ret;
 344 }
 345
 346 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 347
 348 /**
 349  * cpuidle_disable_device - disables 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 void cpuidle_disable_device(struct cpuidle_device *dev)
 356 {
 357         if (!dev->enabled)
 358                 return;
 359         if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 360                 return;
 361
 362         dev->enabled = 0;
 363
 364         if (cpuidle_curr_governor->disable)
 365                 cpuidle_curr_governor->disable(cpuidle_get_driver(), dev);
 366
 367         cpuidle_remove_state_sysfs(dev);
 368         enabled_devices--;
 369 }
 370
 371 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 372
 373 /**
 374  * __cpuidle_register_device - internal register function called before register
 375  * and enable routines
 376  * @dev: the cpu
 377  *
 378  * cpuidle_lock mutex must be held before this is called
 379  */
 380 static int __cpuidle_register_device(struct cpuidle_device *dev)
 381 {
 382         int ret;
 383         struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
 384         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 385
 386         if (!dev)
 387                 return -EINVAL;
 388         if (!try_module_get(cpuidle_driver->owner))
 389                 return -EINVAL;
 390
 391         init_completion(&dev->kobj_unregister);
 392
 393         per_cpu(cpuidle_devices, dev->cpu) = dev;
 394         list_add(&dev->device_list, &cpuidle_detected_devices);
 395         if ((ret = cpuidle_add_sysfs(cpu_dev))) {
 396                 module_put(cpuidle_driver->owner);
 397                 return ret;
 398         }
 399
 400         dev->registered = 1;
 401         return 0;
 402 }
 403
 404 /**
 405  * cpuidle_register_device - registers a CPU's idle PM feature
 406  * @dev: the cpu
 407  */
 408 int cpuidle_register_device(struct cpuidle_device *dev)
 409 {
 410         int ret;
 411
 412         mutex_lock(&cpuidle_lock);
 413
 414         if ((ret = __cpuidle_register_device(dev))) {
 415                 mutex_unlock(&cpuidle_lock);
 416                 return ret;
 417         }
 418
 419         cpuidle_enable_device(dev);
 420         cpuidle_install_idle_handler();
 421
 422         mutex_unlock(&cpuidle_lock);
 423
 424         return 0;
 425
 426 }
 427
 428 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 429
 430 /**
 431  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 432  * @dev: the cpu
 433  */
 434 void cpuidle_unregister_device(struct cpuidle_device *dev)
 435 {
 436         struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
 437         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 438
 439         if (dev->registered == 0)
 440                 return;
 441
 442         cpuidle_pause_and_lock();
 443
 444         cpuidle_disable_device(dev);
 445
 446         cpuidle_remove_sysfs(cpu_dev);
 447         list_del(&dev->device_list);
 448         wait_for_completion(&dev->kobj_unregister);
 449         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 450
 451         cpuidle_resume_and_unlock();
 452
 453         module_put(cpuidle_driver->owner);
 454 }
 455
 456 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 457
 458 #ifdef CONFIG_SMP
 459
 460 static void smp_callback(void *v)
 461 {
 462         /* we already woke the CPU up, nothing more to do */
 463 }
 464
 465 /*
 466  * This function gets called when a part of the kernel has a new latency
 467  * requirement.  This means we need to get all processors out of their C-state,
 468  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 469  * wakes them all right up.
 470  */
 471 static int cpuidle_latency_notify(struct notifier_block *b,
 472                 unsigned long l, void *v)
 473 {
 474         smp_call_function(smp_callback, NULL, 1);
 475         return NOTIFY_OK;
 476 }
 477
 478 static struct notifier_block cpuidle_latency_notifier = {
 479         .notifier_call = cpuidle_latency_notify,
 480 };
 481
 482 static inline void latency_notifier_init(struct notifier_block *n)
 483 {
 484         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 485 }
 486
 487 #else /* CONFIG_SMP */
 488
 489 #define latency_notifier_init(x) do { } while (0)
 490
 491 #endif /* CONFIG_SMP */
 492
 493 /**
 494  * cpuidle_init - core initializer
 495  */
 496 static int __init cpuidle_init(void)
 497 {
 498         int ret;
 499
 500         if (cpuidle_disabled())
 501                 return -ENODEV;
 502
 503         ret = cpuidle_add_interface(cpu_subsys.dev_root);
 504         if (ret)
 505                 return ret;
 506
 507         latency_notifier_init(&cpuidle_latency_notifier);
 508
 509         return 0;
 510 }
 511
 512 module_param(off, int, 0444);
 513 core_initcall(cpuidle_init);