cpuidle.c revision d8c216cfa57e8a579f41729cbb88c97835d9ac8d 
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
24DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
25
26DEFINE_MUTEX(cpuidle_lock);
27LIST_HEAD(cpuidle_detected_devices);
28static void (*pm_idle_old)(void);
29
30static int enabled_devices;
31
32#if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
33static 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 */
40static void cpuidle_kick_cpus(void) {}
41#endif
42
43static 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 */
50static 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 */
115void 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 */
127void 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 */
138void cpuidle_pause_and_lock(void)
139{
140	mutex_lock(&cpuidle_lock);
141	cpuidle_uninstall_idle_handler();
142}
143
144EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
145
146/**
147 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
148 */
149void cpuidle_resume_and_unlock(void)
150{
151	cpuidle_install_idle_handler();
152	mutex_unlock(&cpuidle_lock);
153}
154
155EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
156
157#ifdef CONFIG_ARCH_HAS_CPU_RELAX
158static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
159{
160	ktime_t	t1, t2;
161	s64 diff;
162	int ret;
163
164	t1 = ktime_get();
165	local_irq_enable();
166	while (!need_resched())
167		cpu_relax();
168
169	t2 = ktime_get();
170	diff = ktime_to_us(ktime_sub(t2, t1));
171	if (diff > INT_MAX)
172		diff = INT_MAX;
173
174	ret = (int) diff;
175	return ret;
176}
177
178static void poll_idle_init(struct cpuidle_device *dev)
179{
180	struct cpuidle_state *state = &dev->states[0];
181
182	cpuidle_set_statedata(state, NULL);
183
184	snprintf(state->name, CPUIDLE_NAME_LEN, "C0");
185	snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
186	state->exit_latency = 0;
187	state->target_residency = 0;
188	state->power_usage = -1;
189	state->flags = CPUIDLE_FLAG_POLL;
190	state->enter = poll_idle;
191}
192#else
193static void poll_idle_init(struct cpuidle_device *dev) {}
194#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
195
196/**
197 * cpuidle_enable_device - enables idle PM for a CPU
198 * @dev: the CPU
199 *
200 * This function must be called between cpuidle_pause_and_lock and
201 * cpuidle_resume_and_unlock when used externally.
202 */
203int cpuidle_enable_device(struct cpuidle_device *dev)
204{
205	int ret, i;
206
207	if (dev->enabled)
208		return 0;
209	if (!cpuidle_get_driver() || !cpuidle_curr_governor)
210		return -EIO;
211	if (!dev->state_count)
212		return -EINVAL;
213
214	if (dev->registered == 0) {
215		ret = __cpuidle_register_device(dev);
216		if (ret)
217			return ret;
218	}
219
220	poll_idle_init(dev);
221
222	if ((ret = cpuidle_add_state_sysfs(dev)))
223		return ret;
224
225	if (cpuidle_curr_governor->enable &&
226	    (ret = cpuidle_curr_governor->enable(dev)))
227		goto fail_sysfs;
228
229	for (i = 0; i < dev->state_count; i++) {
230		dev->states[i].usage = 0;
231		dev->states[i].time = 0;
232	}
233	dev->last_residency = 0;
234	dev->last_state = NULL;
235
236	smp_wmb();
237
238	dev->enabled = 1;
239
240	enabled_devices++;
241	return 0;
242
243fail_sysfs:
244	cpuidle_remove_state_sysfs(dev);
245
246	return ret;
247}
248
249EXPORT_SYMBOL_GPL(cpuidle_enable_device);
250
251/**
252 * cpuidle_disable_device - disables idle PM for a CPU
253 * @dev: the CPU
254 *
255 * This function must be called between cpuidle_pause_and_lock and
256 * cpuidle_resume_and_unlock when used externally.
257 */
258void cpuidle_disable_device(struct cpuidle_device *dev)
259{
260	if (!dev->enabled)
261		return;
262	if (!cpuidle_get_driver() || !cpuidle_curr_governor)
263		return;
264
265	dev->enabled = 0;
266
267	if (cpuidle_curr_governor->disable)
268		cpuidle_curr_governor->disable(dev);
269
270	cpuidle_remove_state_sysfs(dev);
271	enabled_devices--;
272}
273
274EXPORT_SYMBOL_GPL(cpuidle_disable_device);
275
276/**
277 * __cpuidle_register_device - internal register function called before register
278 * and enable routines
279 * @dev: the cpu
280 *
281 * cpuidle_lock mutex must be held before this is called
282 */
283static int __cpuidle_register_device(struct cpuidle_device *dev)
284{
285	int ret;
286	struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
287	struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
288
289	if (!sys_dev)
290		return -EINVAL;
291	if (!try_module_get(cpuidle_driver->owner))
292		return -EINVAL;
293
294	init_completion(&dev->kobj_unregister);
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 */
331int 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
351EXPORT_SYMBOL_GPL(cpuidle_register_device);
352
353/**
354 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
355 * @dev: the cpu
356 */
357void 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
379EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
380
381#ifdef CONFIG_SMP
382
383static 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 */
394static 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
401static struct notifier_block cpuidle_latency_notifier = {
402	.notifier_call = cpuidle_latency_notify,
403};
404
405static 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 */
419static 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
434core_initcall(cpuidle_init);
435