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