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_params.h>
  16 #include <linux/cpu.h>
  17 #include <linux/cpuidle.h>
  18 #include <linux/ktime.h>
  19 #include <linux/hrtimer.h>
  20 #include <trace/events/power.h>
  21
  22 #include "cpuidle.h"
  23
  24 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
  25
  26 DEFINE_MUTEX(cpuidle_lock);
  27 LIST_HEAD(cpuidle_detected_devices);
  28 static void (*pm_idle_old)(void);
  29
  30 static int enabled_devices;
  31
  32 #if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
  33 static void cpuidle_kick_cpus(void)
  34 {
  35         cpu_idle_wait();
  36 }
  37 #elif defined(CONFIG_SMP)
  38 # error "Arch needs cpu_idle_wait() equivalent here"
  39 #else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
  40 static void cpuidle_kick_cpus(void) {}
  41 #endif
  42
  43 static int __cpuidle_register_device(struct cpuidle_device *dev);
  44
  45 /**
  46  * cpuidle_idle_call - the main idle loop
  47  *
  48  * NOTE: no locks or semaphores should be used here
  49  */
  50 static void cpuidle_idle_call(void)
  51 {
  52         struct cpuidle_device *dev = __get_cpu_var(cpuidle_devices);
  53         struct cpuidle_state *target_state;
  54         int next_state;
  55
  56         /* check if the device is ready */
  57         if (!dev || !dev->enabled) {
  58                 if (pm_idle_old)
  59                         pm_idle_old();
  60                 else
  61 #if defined(CONFIG_ARCH_HAS_DEFAULT_IDLE)
  62                         default_idle();
  63 #else
  64                         local_irq_enable();
  65 #endif
  66                 return;
  67         }
  68
  69 #if 0
  70         /* shows regressions, re-enable for 2.6.29 */
  71         /*
  72          * run any timers that can be run now, at this point
  73          * before calculating the idle duration etc.
  74          */
  75         hrtimer_peek_ahead_timers();
  76 #endif
  77
  78         /*
  79          * Call the device's prepare function before calling the
  80          * governor's select function.  ->prepare gives the device's
  81          * cpuidle driver a chance to update any dynamic information
  82          * of its cpuidle states for the current idle period, e.g.
  83          * state availability, latencies, residencies, etc.
  84          */
  85         if (dev->prepare)
  86                 dev->prepare(dev);
  87
  88         /* ask the governor for the next state */
  89         next_state = cpuidle_curr_governor->select(dev);
  90         if (need_resched()) {
  91                 local_irq_enable();
  92                 return;
  93         }
  94
  95         target_state = &dev->states[next_state];
  96
  97         /* enter the state and update stats */
  98         dev->last_state = target_state;
  99         dev->last_residency = target_state->enter(dev, target_state);
 100         if (dev->last_state)
 101                 target_state = dev->last_state;
 102
 103         target_state->time += (unsigned long long)dev->last_residency;
 104         target_state->usage++;
 105
 106         /* give the governor an opportunity to reflect on the outcome */
 107         if (cpuidle_curr_governor->reflect)
 108                 cpuidle_curr_governor->reflect(dev);
 109         trace_power_end(smp_processor_id());
 110 }
 111
 112 /**
 113  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 114  */
 115 void cpuidle_install_idle_handler(void)
 116 {
 117         if (enabled_devices && (pm_idle != cpuidle_idle_call)) {
 118                 /* Make sure all changes finished before we switch to new idle */
 119                 smp_wmb();
 120                 pm_idle = cpuidle_idle_call;
 121         }
 122 }
 123
 124 /**
 125  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 126  */
 127 void cpuidle_uninstall_idle_handler(void)
 128 {
 129         if (enabled_devices && pm_idle_old && (pm_idle != pm_idle_old)) {
 130                 pm_idle = pm_idle_old;
 131                 cpuidle_kick_cpus();
 132         }
 133 }
 134
 135 /**
 136  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 137  */
 138 void cpuidle_pause_and_lock(void)
 139 {
 140         mutex_lock(&cpuidle_lock);
 141         cpuidle_uninstall_idle_handler();
 142 }
 143
 144 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 145
 146 /**
 147  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 148  */
 149 void cpuidle_resume_and_unlock(void)
 150 {
 151         cpuidle_install_idle_handler();
 152         mutex_unlock(&cpuidle_lock);
 153 }
 154
 155 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 156
 157 /**
 158  * cpuidle_enable_device - enables idle PM for a CPU
 159  * @dev: the CPU
 160  *
 161  * This function must be called between cpuidle_pause_and_lock and
 162  * cpuidle_resume_and_unlock when used externally.
 163  */
 164 int cpuidle_enable_device(struct cpuidle_device *dev)
 165 {
 166         int ret, i;
 167
 168         if (dev->enabled)
 169                 return 0;
 170         if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 171                 return -EIO;
 172         if (!dev->state_count)
 173                 return -EINVAL;
 174
 175         if (dev->registered == 0) {
 176                 ret = __cpuidle_register_device(dev);
 177                 if (ret)
 178                         return ret;
 179         }
 180
 181         if ((ret = cpuidle_add_state_sysfs(dev)))
 182                 return ret;
 183
 184         if (cpuidle_curr_governor->enable &&
 185             (ret = cpuidle_curr_governor->enable(dev)))
 186                 goto fail_sysfs;
 187
 188         for (i = 0; i < dev->state_count; i++) {
 189                 dev->states[i].usage = 0;
 190                 dev->states[i].time = 0;
 191         }
 192         dev->last_residency = 0;
 193         dev->last_state = NULL;
 194
 195         smp_wmb();
 196
 197         dev->enabled = 1;
 198
 199         enabled_devices++;
 200         return 0;
 201
 202 fail_sysfs:
 203         cpuidle_remove_state_sysfs(dev);
 204
 205         return ret;
 206 }
 207
 208 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 209
 210 /**
 211  * cpuidle_disable_device - disables idle PM for a CPU
 212  * @dev: the CPU
 213  *
 214  * This function must be called between cpuidle_pause_and_lock and
 215  * cpuidle_resume_and_unlock when used externally.
 216  */
 217 void cpuidle_disable_device(struct cpuidle_device *dev)
 218 {
 219         if (!dev->enabled)
 220                 return;
 221         if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 222                 return;
 223
 224         dev->enabled = 0;
 225
 226         if (cpuidle_curr_governor->disable)
 227                 cpuidle_curr_governor->disable(dev);
 228
 229         cpuidle_remove_state_sysfs(dev);
 230         enabled_devices--;
 231 }
 232
 233 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 234
 235 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
 236 static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
 237 {
 238         ktime_t t1, t2;
 239         s64 diff;
 240         int ret;
 241
 242         t1 = ktime_get();
 243         local_irq_enable();
 244         while (!need_resched())
 245                 cpu_relax();
 246
 247         t2 = ktime_get();
 248         diff = ktime_to_us(ktime_sub(t2, t1));
 249         if (diff > INT_MAX)
 250                 diff = INT_MAX;
 251
 252         ret = (int) diff;
 253         return ret;
 254 }
 255
 256 static void poll_idle_init(struct cpuidle_device *dev)
 257 {
 258         struct cpuidle_state *state = &dev->states[0];
 259
 260         cpuidle_set_statedata(state, NULL);
 261
 262         snprintf(state->name, CPUIDLE_NAME_LEN, "C0");
 263         snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
 264         state->exit_latency = 0;
 265         state->target_residency = 0;
 266         state->power_usage = -1;
 267         state->flags = CPUIDLE_FLAG_POLL;
 268         state->enter = poll_idle;
 269 }
 270 #else
 271 static void poll_idle_init(struct cpuidle_device *dev) {}
 272 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
 273
 274 /**
 275  * __cpuidle_register_device - internal register function called before register
 276  * and enable routines
 277  * @dev: the cpu
 278  *
 279  * cpuidle_lock mutex must be held before this is called
 280  */
 281 static int __cpuidle_register_device(struct cpuidle_device *dev)
 282 {
 283         int ret;
 284         struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
 285         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 286
 287         if (!sys_dev)
 288                 return -EINVAL;
 289         if (!try_module_get(cpuidle_driver->owner))
 290                 return -EINVAL;
 291
 292         init_completion(&dev->kobj_unregister);
 293
 294         poll_idle_init(dev);
 295
 296         /*
 297          * cpuidle driver should set the dev->power_specified bit
 298          * before registering the device if the driver provides
 299          * power_usage numbers.
 300          *
 301          * For those devices whose ->power_specified is not set,
 302          * we fill in power_usage with decreasing values as the
 303          * cpuidle code has an implicit assumption that state Cn
 304          * uses less power than C(n-1).
 305          *
 306          * With CONFIG_ARCH_HAS_CPU_RELAX, C0 is already assigned
 307          * an power value of -1.  So we use -2, -3, etc, for other
 308          * c-states.
 309          */
 310         if (!dev->power_specified) {
 311                 int i;
 312                 for (i = CPUIDLE_DRIVER_STATE_START; i < dev->state_count; i++)
 313                         dev->states[i].power_usage = -1 - i;
 314         }
 315
 316         per_cpu(cpuidle_devices, dev->cpu) = dev;
 317         list_add(&dev->device_list, &cpuidle_detected_devices);
 318         if ((ret = cpuidle_add_sysfs(sys_dev))) {
 319                 module_put(cpuidle_driver->owner);
 320                 return ret;
 321         }
 322
 323         dev->registered = 1;
 324         return 0;
 325 }
 326
 327 /**
 328  * cpuidle_register_device - registers a CPU's idle PM feature
 329  * @dev: the cpu
 330  */
 331 int cpuidle_register_device(struct cpuidle_device *dev)
 332 {
 333         int ret;
 334
 335         mutex_lock(&cpuidle_lock);
 336
 337         if ((ret = __cpuidle_register_device(dev))) {
 338                 mutex_unlock(&cpuidle_lock);
 339                 return ret;
 340         }
 341
 342         cpuidle_enable_device(dev);
 343         cpuidle_install_idle_handler();
 344
 345         mutex_unlock(&cpuidle_lock);
 346
 347         return 0;
 348
 349 }
 350
 351 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 352
 353 /**
 354  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 355  * @dev: the cpu
 356  */
 357 void cpuidle_unregister_device(struct cpuidle_device *dev)
 358 {
 359         struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
 360         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 361
 362         if (dev->registered == 0)
 363                 return;
 364
 365         cpuidle_pause_and_lock();
 366
 367         cpuidle_disable_device(dev);
 368
 369         cpuidle_remove_sysfs(sys_dev);
 370         list_del(&dev->device_list);
 371         wait_for_completion(&dev->kobj_unregister);
 372         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 373
 374         cpuidle_resume_and_unlock();
 375
 376         module_put(cpuidle_driver->owner);
 377 }
 378
 379 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 380
 381 #ifdef CONFIG_SMP
 382
 383 static void smp_callback(void *v)
 384 {
 385         /* we already woke the CPU up, nothing more to do */
 386 }
 387
 388 /*
 389  * This function gets called when a part of the kernel has a new latency
 390  * requirement.  This means we need to get all processors out of their C-state,
 391  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 392  * wakes them all right up.
 393  */
 394 static int cpuidle_latency_notify(struct notifier_block *b,
 395                 unsigned long l, void *v)
 396 {
 397         smp_call_function(smp_callback, NULL, 1);
 398         return NOTIFY_OK;
 399 }
 400
 401 static struct notifier_block cpuidle_latency_notifier = {
 402         .notifier_call = cpuidle_latency_notify,
 403 };
 404
 405 static inline void latency_notifier_init(struct notifier_block *n)
 406 {
 407         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 408 }
 409
 410 #else /* CONFIG_SMP */
 411
 412 #define latency_notifier_init(x) do { } while (0)
 413
 414 #endif /* CONFIG_SMP */
 415
 416 /**
 417  * cpuidle_init - core initializer
 418  */
 419 static int __init cpuidle_init(void)
 420 {
 421         int ret;
 422
 423         pm_idle_old = pm_idle;
 424
 425         ret = cpuidle_add_class_sysfs(&cpu_sysdev_class);
 426         if (ret)
 427                 return ret;
 428
 429         latency_notifier_init(&cpuidle_latency_notifier);
 430
 431         return 0;
 432 }
 433
 434 core_initcall(cpuidle_init);