Merge branch 'v28-range-hrtimers-for-linus-v2' of git://git.kernel.org/pub/scm/linux...
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / cpuidle / cpuidle.c
blob5bed73329ef825e2f423f04f4540d57330d30b73
1 /*
2 * cpuidle.c - core cpuidle infrastructure
4 * (C) 2006-2007 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5 * Shaohua Li <shaohua.li@intel.com>
6 * Adam Belay <abelay@novell.com>
8 * This code is licenced under the GPL.
9 */
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>
21 #include "cpuidle.h"
23 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
25 DEFINE_MUTEX(cpuidle_lock);
26 LIST_HEAD(cpuidle_detected_devices);
27 static void (*pm_idle_old)(void);
29 static int enabled_devices;
31 #if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
32 static void cpuidle_kick_cpus(void)
34 cpu_idle_wait();
36 #elif defined(CONFIG_SMP)
37 # error "Arch needs cpu_idle_wait() equivalent here"
38 #else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
39 static void cpuidle_kick_cpus(void) {}
40 #endif
42 static int __cpuidle_register_device(struct cpuidle_device *dev);
44 /**
45 * cpuidle_idle_call - the main idle loop
47 * NOTE: no locks or semaphores should be used here
49 static void cpuidle_idle_call(void)
51 struct cpuidle_device *dev = __get_cpu_var(cpuidle_devices);
52 struct cpuidle_state *target_state;
53 int next_state;
55 /* check if the device is ready */
56 if (!dev || !dev->enabled) {
57 if (pm_idle_old)
58 pm_idle_old();
59 else
60 #if defined(CONFIG_ARCH_HAS_DEFAULT_IDLE)
61 default_idle();
62 #else
63 local_irq_enable();
64 #endif
65 return;
69 * run any timers that can be run now, at this point
70 * before calculating the idle duration etc.
72 hrtimer_peek_ahead_timers();
74 /* ask the governor for the next state */
75 next_state = cpuidle_curr_governor->select(dev);
76 if (need_resched())
77 return;
78 target_state = &dev->states[next_state];
80 /* enter the state and update stats */
81 dev->last_state = target_state;
82 dev->last_residency = target_state->enter(dev, target_state);
83 if (dev->last_state)
84 target_state = dev->last_state;
86 target_state->time += (unsigned long long)dev->last_residency;
87 target_state->usage++;
89 /* give the governor an opportunity to reflect on the outcome */
90 if (cpuidle_curr_governor->reflect)
91 cpuidle_curr_governor->reflect(dev);
94 /**
95 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
97 void cpuidle_install_idle_handler(void)
99 if (enabled_devices && (pm_idle != cpuidle_idle_call)) {
100 /* Make sure all changes finished before we switch to new idle */
101 smp_wmb();
102 pm_idle = cpuidle_idle_call;
107 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
109 void cpuidle_uninstall_idle_handler(void)
111 if (enabled_devices && pm_idle_old && (pm_idle != pm_idle_old)) {
112 pm_idle = pm_idle_old;
113 cpuidle_kick_cpus();
118 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
120 void cpuidle_pause_and_lock(void)
122 mutex_lock(&cpuidle_lock);
123 cpuidle_uninstall_idle_handler();
126 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
129 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
131 void cpuidle_resume_and_unlock(void)
133 cpuidle_install_idle_handler();
134 mutex_unlock(&cpuidle_lock);
137 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
140 * cpuidle_enable_device - enables idle PM for a CPU
141 * @dev: the CPU
143 * This function must be called between cpuidle_pause_and_lock and
144 * cpuidle_resume_and_unlock when used externally.
146 int cpuidle_enable_device(struct cpuidle_device *dev)
148 int ret, i;
150 if (dev->enabled)
151 return 0;
152 if (!cpuidle_curr_driver || !cpuidle_curr_governor)
153 return -EIO;
154 if (!dev->state_count)
155 return -EINVAL;
157 if (dev->registered == 0) {
158 ret = __cpuidle_register_device(dev);
159 if (ret)
160 return ret;
163 if ((ret = cpuidle_add_state_sysfs(dev)))
164 return ret;
166 if (cpuidle_curr_governor->enable &&
167 (ret = cpuidle_curr_governor->enable(dev)))
168 goto fail_sysfs;
170 for (i = 0; i < dev->state_count; i++) {
171 dev->states[i].usage = 0;
172 dev->states[i].time = 0;
174 dev->last_residency = 0;
175 dev->last_state = NULL;
177 smp_wmb();
179 dev->enabled = 1;
181 enabled_devices++;
182 return 0;
184 fail_sysfs:
185 cpuidle_remove_state_sysfs(dev);
187 return ret;
190 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
193 * cpuidle_disable_device - disables idle PM for a CPU
194 * @dev: the CPU
196 * This function must be called between cpuidle_pause_and_lock and
197 * cpuidle_resume_and_unlock when used externally.
199 void cpuidle_disable_device(struct cpuidle_device *dev)
201 if (!dev->enabled)
202 return;
203 if (!cpuidle_curr_driver || !cpuidle_curr_governor)
204 return;
206 dev->enabled = 0;
208 if (cpuidle_curr_governor->disable)
209 cpuidle_curr_governor->disable(dev);
211 cpuidle_remove_state_sysfs(dev);
212 enabled_devices--;
215 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
217 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
218 static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
220 ktime_t t1, t2;
221 s64 diff;
222 int ret;
224 t1 = ktime_get();
225 local_irq_enable();
226 while (!need_resched())
227 cpu_relax();
229 t2 = ktime_get();
230 diff = ktime_to_us(ktime_sub(t2, t1));
231 if (diff > INT_MAX)
232 diff = INT_MAX;
234 ret = (int) diff;
235 return ret;
238 static void poll_idle_init(struct cpuidle_device *dev)
240 struct cpuidle_state *state = &dev->states[0];
242 cpuidle_set_statedata(state, NULL);
244 snprintf(state->name, CPUIDLE_NAME_LEN, "C0");
245 snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
246 state->exit_latency = 0;
247 state->target_residency = 0;
248 state->power_usage = -1;
249 state->flags = CPUIDLE_FLAG_POLL;
250 state->enter = poll_idle;
252 #else
253 static void poll_idle_init(struct cpuidle_device *dev) {}
254 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
257 * __cpuidle_register_device - internal register function called before register
258 * and enable routines
259 * @dev: the cpu
261 * cpuidle_lock mutex must be held before this is called
263 static int __cpuidle_register_device(struct cpuidle_device *dev)
265 int ret;
266 struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
268 if (!sys_dev)
269 return -EINVAL;
270 if (!try_module_get(cpuidle_curr_driver->owner))
271 return -EINVAL;
273 init_completion(&dev->kobj_unregister);
275 poll_idle_init(dev);
277 per_cpu(cpuidle_devices, dev->cpu) = dev;
278 list_add(&dev->device_list, &cpuidle_detected_devices);
279 if ((ret = cpuidle_add_sysfs(sys_dev))) {
280 module_put(cpuidle_curr_driver->owner);
281 return ret;
284 dev->registered = 1;
285 return 0;
289 * cpuidle_register_device - registers a CPU's idle PM feature
290 * @dev: the cpu
292 int cpuidle_register_device(struct cpuidle_device *dev)
294 int ret;
296 mutex_lock(&cpuidle_lock);
298 if ((ret = __cpuidle_register_device(dev))) {
299 mutex_unlock(&cpuidle_lock);
300 return ret;
303 cpuidle_enable_device(dev);
304 cpuidle_install_idle_handler();
306 mutex_unlock(&cpuidle_lock);
308 return 0;
312 EXPORT_SYMBOL_GPL(cpuidle_register_device);
315 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
316 * @dev: the cpu
318 void cpuidle_unregister_device(struct cpuidle_device *dev)
320 struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
322 if (dev->registered == 0)
323 return;
325 cpuidle_pause_and_lock();
327 cpuidle_disable_device(dev);
329 cpuidle_remove_sysfs(sys_dev);
330 list_del(&dev->device_list);
331 wait_for_completion(&dev->kobj_unregister);
332 per_cpu(cpuidle_devices, dev->cpu) = NULL;
334 cpuidle_resume_and_unlock();
336 module_put(cpuidle_curr_driver->owner);
339 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
341 #ifdef CONFIG_SMP
343 static void smp_callback(void *v)
345 /* we already woke the CPU up, nothing more to do */
349 * This function gets called when a part of the kernel has a new latency
350 * requirement. This means we need to get all processors out of their C-state,
351 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
352 * wakes them all right up.
354 static int cpuidle_latency_notify(struct notifier_block *b,
355 unsigned long l, void *v)
357 smp_call_function(smp_callback, NULL, 1);
358 return NOTIFY_OK;
361 static struct notifier_block cpuidle_latency_notifier = {
362 .notifier_call = cpuidle_latency_notify,
365 static inline void latency_notifier_init(struct notifier_block *n)
367 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
370 #else /* CONFIG_SMP */
372 #define latency_notifier_init(x) do { } while (0)
374 #endif /* CONFIG_SMP */
377 * cpuidle_init - core initializer
379 static int __init cpuidle_init(void)
381 int ret;
383 pm_idle_old = pm_idle;
385 ret = cpuidle_add_class_sysfs(&cpu_sysdev_class);
386 if (ret)
387 return ret;
389 latency_notifier_init(&cpuidle_latency_notifier);
391 return 0;
394 core_initcall(cpuidle_init);