x86, 64-bit: Move K8 B step iret fixup to fault entry asm
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / cpuidle / cpuidle.c
blobad41f19b8e3fa22d008a4c1e3c1ce9ec79fa025e
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>
20 #include <trace/events/power.h>
22 #include "cpuidle.h"
24 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
26 DEFINE_MUTEX(cpuidle_lock);
27 LIST_HEAD(cpuidle_detected_devices);
28 static void (*pm_idle_old)(void);
30 static int enabled_devices;
32 #if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
33 static void cpuidle_kick_cpus(void)
35 cpu_idle_wait();
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
43 static int __cpuidle_register_device(struct cpuidle_device *dev);
45 /**
46 * cpuidle_idle_call - the main idle loop
48 * NOTE: no locks or semaphores should be used here
50 static void cpuidle_idle_call(void)
52 struct cpuidle_device *dev = __get_cpu_var(cpuidle_devices);
53 struct cpuidle_state *target_state;
54 int next_state;
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;
69 #if 0
70 /* shows regressions, re-enable for 2.6.29 */
72 * run any timers that can be run now, at this point
73 * before calculating the idle duration etc.
75 hrtimer_peek_ahead_timers();
76 #endif
77 /* ask the governor for the next state */
78 next_state = cpuidle_curr_governor->select(dev);
79 if (need_resched())
80 return;
81 target_state = &dev->states[next_state];
83 /* enter the state and update stats */
84 dev->last_state = target_state;
85 dev->last_residency = target_state->enter(dev, target_state);
86 if (dev->last_state)
87 target_state = dev->last_state;
89 target_state->time += (unsigned long long)dev->last_residency;
90 target_state->usage++;
92 /* give the governor an opportunity to reflect on the outcome */
93 if (cpuidle_curr_governor->reflect)
94 cpuidle_curr_governor->reflect(dev);
95 trace_power_end(0);
98 /**
99 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
101 void cpuidle_install_idle_handler(void)
103 if (enabled_devices && (pm_idle != cpuidle_idle_call)) {
104 /* Make sure all changes finished before we switch to new idle */
105 smp_wmb();
106 pm_idle = cpuidle_idle_call;
111 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
113 void cpuidle_uninstall_idle_handler(void)
115 if (enabled_devices && pm_idle_old && (pm_idle != pm_idle_old)) {
116 pm_idle = pm_idle_old;
117 cpuidle_kick_cpus();
122 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
124 void cpuidle_pause_and_lock(void)
126 mutex_lock(&cpuidle_lock);
127 cpuidle_uninstall_idle_handler();
130 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
133 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
135 void cpuidle_resume_and_unlock(void)
137 cpuidle_install_idle_handler();
138 mutex_unlock(&cpuidle_lock);
141 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
144 * cpuidle_enable_device - enables idle PM for a CPU
145 * @dev: the CPU
147 * This function must be called between cpuidle_pause_and_lock and
148 * cpuidle_resume_and_unlock when used externally.
150 int cpuidle_enable_device(struct cpuidle_device *dev)
152 int ret, i;
154 if (dev->enabled)
155 return 0;
156 if (!cpuidle_curr_driver || !cpuidle_curr_governor)
157 return -EIO;
158 if (!dev->state_count)
159 return -EINVAL;
161 if (dev->registered == 0) {
162 ret = __cpuidle_register_device(dev);
163 if (ret)
164 return ret;
167 if ((ret = cpuidle_add_state_sysfs(dev)))
168 return ret;
170 if (cpuidle_curr_governor->enable &&
171 (ret = cpuidle_curr_governor->enable(dev)))
172 goto fail_sysfs;
174 for (i = 0; i < dev->state_count; i++) {
175 dev->states[i].usage = 0;
176 dev->states[i].time = 0;
178 dev->last_residency = 0;
179 dev->last_state = NULL;
181 smp_wmb();
183 dev->enabled = 1;
185 enabled_devices++;
186 return 0;
188 fail_sysfs:
189 cpuidle_remove_state_sysfs(dev);
191 return ret;
194 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
197 * cpuidle_disable_device - disables idle PM for a CPU
198 * @dev: the CPU
200 * This function must be called between cpuidle_pause_and_lock and
201 * cpuidle_resume_and_unlock when used externally.
203 void cpuidle_disable_device(struct cpuidle_device *dev)
205 if (!dev->enabled)
206 return;
207 if (!cpuidle_curr_driver || !cpuidle_curr_governor)
208 return;
210 dev->enabled = 0;
212 if (cpuidle_curr_governor->disable)
213 cpuidle_curr_governor->disable(dev);
215 cpuidle_remove_state_sysfs(dev);
216 enabled_devices--;
219 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
221 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
222 static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
224 ktime_t t1, t2;
225 s64 diff;
226 int ret;
228 t1 = ktime_get();
229 local_irq_enable();
230 while (!need_resched())
231 cpu_relax();
233 t2 = ktime_get();
234 diff = ktime_to_us(ktime_sub(t2, t1));
235 if (diff > INT_MAX)
236 diff = INT_MAX;
238 ret = (int) diff;
239 return ret;
242 static void poll_idle_init(struct cpuidle_device *dev)
244 struct cpuidle_state *state = &dev->states[0];
246 cpuidle_set_statedata(state, NULL);
248 snprintf(state->name, CPUIDLE_NAME_LEN, "C0");
249 snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
250 state->exit_latency = 0;
251 state->target_residency = 0;
252 state->power_usage = -1;
253 state->flags = CPUIDLE_FLAG_POLL;
254 state->enter = poll_idle;
256 #else
257 static void poll_idle_init(struct cpuidle_device *dev) {}
258 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
261 * __cpuidle_register_device - internal register function called before register
262 * and enable routines
263 * @dev: the cpu
265 * cpuidle_lock mutex must be held before this is called
267 static int __cpuidle_register_device(struct cpuidle_device *dev)
269 int ret;
270 struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
272 if (!sys_dev)
273 return -EINVAL;
274 if (!try_module_get(cpuidle_curr_driver->owner))
275 return -EINVAL;
277 init_completion(&dev->kobj_unregister);
279 poll_idle_init(dev);
281 per_cpu(cpuidle_devices, dev->cpu) = dev;
282 list_add(&dev->device_list, &cpuidle_detected_devices);
283 if ((ret = cpuidle_add_sysfs(sys_dev))) {
284 module_put(cpuidle_curr_driver->owner);
285 return ret;
288 dev->registered = 1;
289 return 0;
293 * cpuidle_register_device - registers a CPU's idle PM feature
294 * @dev: the cpu
296 int cpuidle_register_device(struct cpuidle_device *dev)
298 int ret;
300 mutex_lock(&cpuidle_lock);
302 if ((ret = __cpuidle_register_device(dev))) {
303 mutex_unlock(&cpuidle_lock);
304 return ret;
307 cpuidle_enable_device(dev);
308 cpuidle_install_idle_handler();
310 mutex_unlock(&cpuidle_lock);
312 return 0;
316 EXPORT_SYMBOL_GPL(cpuidle_register_device);
319 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
320 * @dev: the cpu
322 void cpuidle_unregister_device(struct cpuidle_device *dev)
324 struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
326 if (dev->registered == 0)
327 return;
329 cpuidle_pause_and_lock();
331 cpuidle_disable_device(dev);
333 cpuidle_remove_sysfs(sys_dev);
334 list_del(&dev->device_list);
335 wait_for_completion(&dev->kobj_unregister);
336 per_cpu(cpuidle_devices, dev->cpu) = NULL;
338 cpuidle_resume_and_unlock();
340 module_put(cpuidle_curr_driver->owner);
343 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
345 #ifdef CONFIG_SMP
347 static void smp_callback(void *v)
349 /* we already woke the CPU up, nothing more to do */
353 * This function gets called when a part of the kernel has a new latency
354 * requirement. This means we need to get all processors out of their C-state,
355 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
356 * wakes them all right up.
358 static int cpuidle_latency_notify(struct notifier_block *b,
359 unsigned long l, void *v)
361 smp_call_function(smp_callback, NULL, 1);
362 return NOTIFY_OK;
365 static struct notifier_block cpuidle_latency_notifier = {
366 .notifier_call = cpuidle_latency_notify,
369 static inline void latency_notifier_init(struct notifier_block *n)
371 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
374 #else /* CONFIG_SMP */
376 #define latency_notifier_init(x) do { } while (0)
378 #endif /* CONFIG_SMP */
381 * cpuidle_init - core initializer
383 static int __init cpuidle_init(void)
385 int ret;
387 pm_idle_old = pm_idle;
389 ret = cpuidle_add_class_sysfs(&cpu_sysdev_class);
390 if (ret)
391 return ret;
393 latency_notifier_init(&cpuidle_latency_notifier);
395 return 0;
398 core_initcall(cpuidle_init);