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 #if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
  44 static void cpuidle_kick_cpus(void)
  45 {
  46         cpu_idle_wait();
  47 }
  48 #elif defined(CONFIG_SMP)
  49 # error "Arch needs cpu_idle_wait() equivalent here"
  50 #else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
  51 static void cpuidle_kick_cpus(void) {}
  52 #endif
  53
  54 static int __cpuidle_register_device(struct cpuidle_device *dev);
  55
  56 static inline int cpuidle_enter(struct cpuidle_device *dev,
  57                                 struct cpuidle_driver *drv, int index)
  58 {
  59         struct cpuidle_state *target_state = &drv->states[index];
  60         return target_state->enter(dev, drv, index);
  61 }
  62
  63 static inline int cpuidle_enter_tk(struct cpuidle_device *dev,
  64                                struct cpuidle_driver *drv, int index)
  65 {
  66         return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter);
  67 }
  68
  69 typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev,
  70                                struct cpuidle_driver *drv, int index);
  71
  72 static cpuidle_enter_t cpuidle_enter_ops;
  73
  74 /**
  75  * cpuidle_play_dead - cpu off-lining
  76  *
  77  * Returns in case of an error or no driver
  78  */
  79 int cpuidle_play_dead(void)
  80 {
  81         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  82         struct cpuidle_driver *drv = cpuidle_get_driver();
  83         int i, dead_state = -1;
  84         int power_usage = -1;
  85
  86         if (!drv)
  87                 return -ENODEV;
  88
  89         /* Find lowest-power state that supports long-term idle */
  90         for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
  91                 struct cpuidle_state *s = &drv->states[i];
  92
  93                 if (s->power_usage < power_usage && s->enter_dead) {
  94                         power_usage = s->power_usage;
  95                         dead_state = i;
  96                 }
  97         }
  98
  99         if (dead_state != -1)
 100                 return drv->states[dead_state].enter_dead(dev, dead_state);
 101
 102         return -ENODEV;
 103 }
 104
 105 /**
 106  * cpuidle_idle_call - the main idle loop
 107  *
 108  * NOTE: no locks or semaphores should be used here
 109  * return non-zero on failure
 110  */
 111 int cpuidle_idle_call(void)
 112 {
 113         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
 114         struct cpuidle_driver *drv = cpuidle_get_driver();
 115         int next_state, entered_state;
 116
 117         if (off)
 118                 return -ENODEV;
 119
 120         if (!initialized)
 121                 return -ENODEV;
 122
 123         /* check if the device is ready */
 124         if (!dev || !dev->enabled)
 125                 return -EBUSY;
 126
 127 #if 0
 128         /* shows regressions, re-enable for 2.6.29 */
 129         /*
 130          * run any timers that can be run now, at this point
 131          * before calculating the idle duration etc.
 132          */
 133         hrtimer_peek_ahead_timers();
 134 #endif
 135
 136         /* ask the governor for the next state */
 137         next_state = cpuidle_curr_governor->select(drv, dev);
 138         if (need_resched()) {
 139                 local_irq_enable();
 140                 return 0;
 141         }
 142
 143         trace_power_start_rcuidle(POWER_CSTATE, next_state, dev->cpu);
 144         trace_cpu_idle_rcuidle(next_state, dev->cpu);
 145
 146         entered_state = cpuidle_enter_ops(dev, drv, next_state);
 147
 148         trace_power_end_rcuidle(dev->cpu);
 149         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 150
 151         if (entered_state >= 0) {
 152                 /* Update cpuidle counters */
 153                 /* This can be moved to within driver enter routine
 154                  * but that results in multiple copies of same code.
 155                  */
 156                 dev->states_usage[entered_state].time +=
 157                                 (unsigned long long)dev->last_residency;
 158                 dev->states_usage[entered_state].usage++;
 159         } else {
 160                 dev->last_residency = 0;
 161         }
 162
 163         /* give the governor an opportunity to reflect on the outcome */
 164         if (cpuidle_curr_governor->reflect)
 165                 cpuidle_curr_governor->reflect(dev, entered_state);
 166
 167         return 0;
 168 }
 169
 170 /**
 171  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 172  */
 173 void cpuidle_install_idle_handler(void)
 174 {
 175         if (enabled_devices) {
 176                 /* Make sure all changes finished before we switch to new idle */
 177                 smp_wmb();
 178                 initialized = 1;
 179         }
 180 }
 181
 182 /**
 183  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 184  */
 185 void cpuidle_uninstall_idle_handler(void)
 186 {
 187         if (enabled_devices) {
 188                 initialized = 0;
 189                 cpuidle_kick_cpus();
 190         }
 191 }
 192
 193 /**
 194  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 195  */
 196 void cpuidle_pause_and_lock(void)
 197 {
 198         mutex_lock(&cpuidle_lock);
 199         cpuidle_uninstall_idle_handler();
 200 }
 201
 202 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 203
 204 /**
 205  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 206  */
 207 void cpuidle_resume_and_unlock(void)
 208 {
 209         cpuidle_install_idle_handler();
 210         mutex_unlock(&cpuidle_lock);
 211 }
 212
 213 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 214
 215 /**
 216  * cpuidle_wrap_enter - performs timekeeping and irqen around enter function
 217  * @dev: pointer to a valid cpuidle_device object
 218  * @drv: pointer to a valid cpuidle_driver object
 219  * @index: index of the target cpuidle state.
 220  */
 221 int cpuidle_wrap_enter(struct cpuidle_device *dev,
 222                                 struct cpuidle_driver *drv, int index,
 223                                 int (*enter)(struct cpuidle_device *dev,
 224                                         struct cpuidle_driver *drv, int index))
 225 {
 226         ktime_t time_start, time_end;
 227         s64 diff;
 228
 229         time_start = ktime_get();
 230
 231         index = enter(dev, drv, index);
 232
 233         time_end = ktime_get();
 234
 235         local_irq_enable();
 236
 237         diff = ktime_to_us(ktime_sub(time_end, time_start));
 238         if (diff > INT_MAX)
 239                 diff = INT_MAX;
 240
 241         dev->last_residency = (int) diff;
 242
 243         return index;
 244 }
 245
 246 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
 247 static int poll_idle(struct cpuidle_device *dev,
 248                 struct cpuidle_driver *drv, int index)
 249 {
 250         ktime_t t1, t2;
 251         s64 diff;
 252
 253         t1 = ktime_get();
 254         local_irq_enable();
 255         while (!need_resched())
 256                 cpu_relax();
 257
 258         t2 = ktime_get();
 259         diff = ktime_to_us(ktime_sub(t2, t1));
 260         if (diff > INT_MAX)
 261                 diff = INT_MAX;
 262
 263         dev->last_residency = (int) diff;
 264
 265         return index;
 266 }
 267
 268 static void poll_idle_init(struct cpuidle_driver *drv)
 269 {
 270         struct cpuidle_state *state = &drv->states[0];
 271
 272         snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
 273         snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
 274         state->exit_latency = 0;
 275         state->target_residency = 0;
 276         state->power_usage = -1;
 277         state->flags = 0;
 278         state->enter = poll_idle;
 279         state->disable = 0;
 280 }
 281 #else
 282 static void poll_idle_init(struct cpuidle_driver *drv) {}
 283 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
 284
 285 /**
 286  * cpuidle_enable_device - enables idle PM for a CPU
 287  * @dev: the CPU
 288  *
 289  * This function must be called between cpuidle_pause_and_lock and
 290  * cpuidle_resume_and_unlock when used externally.
 291  */
 292 int cpuidle_enable_device(struct cpuidle_device *dev)
 293 {
 294         int ret, i;
 295         struct cpuidle_driver *drv = cpuidle_get_driver();
 296
 297         if (dev->enabled)
 298                 return 0;
 299         if (!drv || !cpuidle_curr_governor)
 300                 return -EIO;
 301         if (!dev->state_count)
 302                 dev->state_count = drv->state_count;
 303
 304         if (dev->registered == 0) {
 305                 ret = __cpuidle_register_device(dev);
 306                 if (ret)
 307                         return ret;
 308         }
 309
 310         cpuidle_enter_ops = drv->en_core_tk_irqen ?
 311                 cpuidle_enter_tk : cpuidle_enter;
 312
 313         poll_idle_init(drv);
 314
 315         if ((ret = cpuidle_add_state_sysfs(dev)))
 316                 return ret;
 317
 318         if (cpuidle_curr_governor->enable &&
 319             (ret = cpuidle_curr_governor->enable(drv, dev)))
 320                 goto fail_sysfs;
 321
 322         for (i = 0; i < dev->state_count; i++) {
 323                 dev->states_usage[i].usage = 0;
 324                 dev->states_usage[i].time = 0;
 325         }
 326         dev->last_residency = 0;
 327
 328         smp_wmb();
 329
 330         dev->enabled = 1;
 331
 332         enabled_devices++;
 333         return 0;
 334
 335 fail_sysfs:
 336         cpuidle_remove_state_sysfs(dev);
 337
 338         return ret;
 339 }
 340
 341 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 342
 343 /**
 344  * cpuidle_disable_device - disables idle PM for a CPU
 345  * @dev: the CPU
 346  *
 347  * This function must be called between cpuidle_pause_and_lock and
 348  * cpuidle_resume_and_unlock when used externally.
 349  */
 350 void cpuidle_disable_device(struct cpuidle_device *dev)
 351 {
 352         if (!dev->enabled)
 353                 return;
 354         if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 355                 return;
 356
 357         dev->enabled = 0;
 358
 359         if (cpuidle_curr_governor->disable)
 360                 cpuidle_curr_governor->disable(cpuidle_get_driver(), dev);
 361
 362         cpuidle_remove_state_sysfs(dev);
 363         enabled_devices--;
 364 }
 365
 366 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 367
 368 /**
 369  * __cpuidle_register_device - internal register function called before register
 370  * and enable routines
 371  * @dev: the cpu
 372  *
 373  * cpuidle_lock mutex must be held before this is called
 374  */
 375 static int __cpuidle_register_device(struct cpuidle_device *dev)
 376 {
 377         int ret;
 378         struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
 379         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 380
 381         if (!dev)
 382                 return -EINVAL;
 383         if (!try_module_get(cpuidle_driver->owner))
 384                 return -EINVAL;
 385
 386         init_completion(&dev->kobj_unregister);
 387
 388         per_cpu(cpuidle_devices, dev->cpu) = dev;
 389         list_add(&dev->device_list, &cpuidle_detected_devices);
 390         if ((ret = cpuidle_add_sysfs(cpu_dev))) {
 391                 module_put(cpuidle_driver->owner);
 392                 return ret;
 393         }
 394
 395         dev->registered = 1;
 396         return 0;
 397 }
 398
 399 /**
 400  * cpuidle_register_device - registers a CPU's idle PM feature
 401  * @dev: the cpu
 402  */
 403 int cpuidle_register_device(struct cpuidle_device *dev)
 404 {
 405         int ret;
 406
 407         mutex_lock(&cpuidle_lock);
 408
 409         if ((ret = __cpuidle_register_device(dev))) {
 410                 mutex_unlock(&cpuidle_lock);
 411                 return ret;
 412         }
 413
 414         cpuidle_enable_device(dev);
 415         cpuidle_install_idle_handler();
 416
 417         mutex_unlock(&cpuidle_lock);
 418
 419         return 0;
 420
 421 }
 422
 423 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 424
 425 /**
 426  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 427  * @dev: the cpu
 428  */
 429 void cpuidle_unregister_device(struct cpuidle_device *dev)
 430 {
 431         struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
 432         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 433
 434         if (dev->registered == 0)
 435                 return;
 436
 437         cpuidle_pause_and_lock();
 438
 439         cpuidle_disable_device(dev);
 440
 441         cpuidle_remove_sysfs(cpu_dev);
 442         list_del(&dev->device_list);
 443         wait_for_completion(&dev->kobj_unregister);
 444         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 445
 446         cpuidle_resume_and_unlock();
 447
 448         module_put(cpuidle_driver->owner);
 449 }
 450
 451 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 452
 453 #ifdef CONFIG_SMP
 454
 455 static void smp_callback(void *v)
 456 {
 457         /* we already woke the CPU up, nothing more to do */
 458 }
 459
 460 /*
 461  * This function gets called when a part of the kernel has a new latency
 462  * requirement.  This means we need to get all processors out of their C-state,
 463  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 464  * wakes them all right up.
 465  */
 466 static int cpuidle_latency_notify(struct notifier_block *b,
 467                 unsigned long l, void *v)
 468 {
 469         smp_call_function(smp_callback, NULL, 1);
 470         return NOTIFY_OK;
 471 }
 472
 473 static struct notifier_block cpuidle_latency_notifier = {
 474         .notifier_call = cpuidle_latency_notify,
 475 };
 476
 477 static inline void latency_notifier_init(struct notifier_block *n)
 478 {
 479         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 480 }
 481
 482 #else /* CONFIG_SMP */
 483
 484 #define latency_notifier_init(x) do { } while (0)
 485
 486 #endif /* CONFIG_SMP */
 487
 488 /**
 489  * cpuidle_init - core initializer
 490  */
 491 static int __init cpuidle_init(void)
 492 {
 493         int ret;
 494
 495         if (cpuidle_disabled())
 496                 return -ENODEV;
 497
 498         ret = cpuidle_add_interface(cpu_subsys.dev_root);
 499         if (ret)
 500                 return ret;
 501
 502         latency_notifier_init(&cpuidle_latency_notifier);
 503
 504         return 0;
 505 }
 506
 507 module_param(off, int, 0444);
 508 core_initcall(cpuidle_init);