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