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 = __this_cpu_read(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         trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 111 }
 112
 113 /**
 114  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
 115  */
 116 void cpuidle_install_idle_handler(void)
 117 {
 118         if (enabled_devices && (pm_idle != cpuidle_idle_call)) {
 119                 /* Make sure all changes finished before we switch to new idle */
 120                 smp_wmb();
 121                 pm_idle = cpuidle_idle_call;
 122         }
 123 }
 124
 125 /**
 126  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
 127  */
 128 void cpuidle_uninstall_idle_handler(void)
 129 {
 130         if (enabled_devices && pm_idle_old && (pm_idle != pm_idle_old)) {
 131                 pm_idle = pm_idle_old;
 132                 cpuidle_kick_cpus();
 133         }
 134 }
 135
 136 /**
 137  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
 138  */
 139 void cpuidle_pause_and_lock(void)
 140 {
 141         mutex_lock(&cpuidle_lock);
 142         cpuidle_uninstall_idle_handler();
 143 }
 144
 145 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
 146
 147 /**
 148  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
 149  */
 150 void cpuidle_resume_and_unlock(void)
 151 {
 152         cpuidle_install_idle_handler();
 153         mutex_unlock(&cpuidle_lock);
 154 }
 155
 156 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
 157
 158 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
 159 static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
 160 {
 161         ktime_t t1, t2;
 162         s64 diff;
 163         int ret;
 164
 165         t1 = ktime_get();
 166         local_irq_enable();
 167         while (!need_resched())
 168                 cpu_relax();
 169
 170         t2 = ktime_get();
 171         diff = ktime_to_us(ktime_sub(t2, t1));
 172         if (diff > INT_MAX)
 173                 diff = INT_MAX;
 174
 175         ret = (int) diff;
 176         return ret;
 177 }
 178
 179 static void poll_idle_init(struct cpuidle_device *dev)
 180 {
 181         struct cpuidle_state *state = &dev->states[0];
 182
 183         cpuidle_set_statedata(state, NULL);
 184
 185         snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
 186         snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
 187         state->exit_latency = 0;
 188         state->target_residency = 0;
 189         state->power_usage = -1;
 190         state->flags = 0;
 191         state->enter = poll_idle;
 192 }
 193 #else
 194 static void poll_idle_init(struct cpuidle_device *dev) {}
 195 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
 196
 197 /**
 198  * cpuidle_enable_device - enables idle PM for a CPU
 199  * @dev: the CPU
 200  *
 201  * This function must be called between cpuidle_pause_and_lock and
 202  * cpuidle_resume_and_unlock when used externally.
 203  */
 204 int cpuidle_enable_device(struct cpuidle_device *dev)
 205 {
 206         int ret, i;
 207
 208         if (dev->enabled)
 209                 return 0;
 210         if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 211                 return -EIO;
 212         if (!dev->state_count)
 213                 return -EINVAL;
 214
 215         if (dev->registered == 0) {
 216                 ret = __cpuidle_register_device(dev);
 217                 if (ret)
 218                         return ret;
 219         }
 220
 221         poll_idle_init(dev);
 222
 223         if ((ret = cpuidle_add_state_sysfs(dev)))
 224                 return ret;
 225
 226         if (cpuidle_curr_governor->enable &&
 227             (ret = cpuidle_curr_governor->enable(dev)))
 228                 goto fail_sysfs;
 229
 230         for (i = 0; i < dev->state_count; i++) {
 231                 dev->states[i].usage = 0;
 232                 dev->states[i].time = 0;
 233         }
 234         dev->last_residency = 0;
 235         dev->last_state = NULL;
 236
 237         smp_wmb();
 238
 239         dev->enabled = 1;
 240
 241         enabled_devices++;
 242         return 0;
 243
 244 fail_sysfs:
 245         cpuidle_remove_state_sysfs(dev);
 246
 247         return ret;
 248 }
 249
 250 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
 251
 252 /**
 253  * cpuidle_disable_device - disables idle PM for a CPU
 254  * @dev: the CPU
 255  *
 256  * This function must be called between cpuidle_pause_and_lock and
 257  * cpuidle_resume_and_unlock when used externally.
 258  */
 259 void cpuidle_disable_device(struct cpuidle_device *dev)
 260 {
 261         if (!dev->enabled)
 262                 return;
 263         if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 264                 return;
 265
 266         dev->enabled = 0;
 267
 268         if (cpuidle_curr_governor->disable)
 269                 cpuidle_curr_governor->disable(dev);
 270
 271         cpuidle_remove_state_sysfs(dev);
 272         enabled_devices--;
 273 }
 274
 275 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
 276
 277 /**
 278  * __cpuidle_register_device - internal register function called before register
 279  * and enable routines
 280  * @dev: the cpu
 281  *
 282  * cpuidle_lock mutex must be held before this is called
 283  */
 284 static int __cpuidle_register_device(struct cpuidle_device *dev)
 285 {
 286         int ret;
 287         struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
 288         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 289
 290         if (!sys_dev)
 291                 return -EINVAL;
 292         if (!try_module_get(cpuidle_driver->owner))
 293                 return -EINVAL;
 294
 295         init_completion(&dev->kobj_unregister);
 296
 297         /*
 298          * cpuidle driver should set the dev->power_specified bit
 299          * before registering the device if the driver provides
 300          * power_usage numbers.
 301          *
 302          * For those devices whose ->power_specified is not set,
 303          * we fill in power_usage with decreasing values as the
 304          * cpuidle code has an implicit assumption that state Cn
 305          * uses less power than C(n-1).
 306          *
 307          * With CONFIG_ARCH_HAS_CPU_RELAX, C0 is already assigned
 308          * an power value of -1.  So we use -2, -3, etc, for other
 309          * c-states.
 310          */
 311         if (!dev->power_specified) {
 312                 int i;
 313                 for (i = CPUIDLE_DRIVER_STATE_START; i < dev->state_count; i++)
 314                         dev->states[i].power_usage = -1 - i;
 315         }
 316
 317         per_cpu(cpuidle_devices, dev->cpu) = dev;
 318         list_add(&dev->device_list, &cpuidle_detected_devices);
 319         if ((ret = cpuidle_add_sysfs(sys_dev))) {
 320                 module_put(cpuidle_driver->owner);
 321                 return ret;
 322         }
 323
 324         dev->registered = 1;
 325         return 0;
 326 }
 327
 328 /**
 329  * cpuidle_register_device - registers a CPU's idle PM feature
 330  * @dev: the cpu
 331  */
 332 int cpuidle_register_device(struct cpuidle_device *dev)
 333 {
 334         int ret;
 335
 336         mutex_lock(&cpuidle_lock);
 337
 338         if ((ret = __cpuidle_register_device(dev))) {
 339                 mutex_unlock(&cpuidle_lock);
 340                 return ret;
 341         }
 342
 343         cpuidle_enable_device(dev);
 344         cpuidle_install_idle_handler();
 345
 346         mutex_unlock(&cpuidle_lock);
 347
 348         return 0;
 349
 350 }
 351
 352 EXPORT_SYMBOL_GPL(cpuidle_register_device);
 353
 354 /**
 355  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
 356  * @dev: the cpu
 357  */
 358 void cpuidle_unregister_device(struct cpuidle_device *dev)
 359 {
 360         struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
 361         struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
 362
 363         if (dev->registered == 0)
 364                 return;
 365
 366         cpuidle_pause_and_lock();
 367
 368         cpuidle_disable_device(dev);
 369
 370         cpuidle_remove_sysfs(sys_dev);
 371         list_del(&dev->device_list);
 372         wait_for_completion(&dev->kobj_unregister);
 373         per_cpu(cpuidle_devices, dev->cpu) = NULL;
 374
 375         cpuidle_resume_and_unlock();
 376
 377         module_put(cpuidle_driver->owner);
 378 }
 379
 380 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
 381
 382 #ifdef CONFIG_SMP
 383
 384 static void smp_callback(void *v)
 385 {
 386         /* we already woke the CPU up, nothing more to do */
 387 }
 388
 389 /*
 390  * This function gets called when a part of the kernel has a new latency
 391  * requirement.  This means we need to get all processors out of their C-state,
 392  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
 393  * wakes them all right up.
 394  */
 395 static int cpuidle_latency_notify(struct notifier_block *b,
 396                 unsigned long l, void *v)
 397 {
 398         smp_call_function(smp_callback, NULL, 1);
 399         return NOTIFY_OK;
 400 }
 401
 402 static struct notifier_block cpuidle_latency_notifier = {
 403         .notifier_call = cpuidle_latency_notify,
 404 };
 405
 406 static inline void latency_notifier_init(struct notifier_block *n)
 407 {
 408         pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
 409 }
 410
 411 #else /* CONFIG_SMP */
 412
 413 #define latency_notifier_init(x) do { } while (0)
 414
 415 #endif /* CONFIG_SMP */
 416
 417 /**
 418  * cpuidle_init - core initializer
 419  */
 420 static int __init cpuidle_init(void)
 421 {
 422         int ret;
 423
 424         pm_idle_old = pm_idle;
 425
 426         ret = cpuidle_add_class_sysfs(&cpu_sysdev_class);
 427         if (ret)
 428                 return ret;
 429
 430         latency_notifier_init(&cpuidle_latency_notifier);
 431
 432         return 0;
 433 }
 434
 435 core_initcall(cpuidle_init);