cpuidle.c revision 8a25a2fd126c621f44f3aeaef80d51f00fc11639 
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.h>
16#include <linux/cpu.h>
17#include <linux/cpuidle.h>
18#include <linux/ktime.h>
19#include <linux/hrtimer.h>
20#include <linux/module.h>
21#include <trace/events/power.h>
22
23#include "cpuidle.h"
24
25DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
26
27DEFINE_MUTEX(cpuidle_lock);
28LIST_HEAD(cpuidle_detected_devices);
29
30static int enabled_devices;
31static int off __read_mostly;
32static int initialized __read_mostly;
33
34int cpuidle_disabled(void)
35{
36	return off;
37}
38void disable_cpuidle(void)
39{
40	off = 1;
41}
42
43#if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
44static void cpuidle_kick_cpus(void)
45{
46	cpu_idle_wait();
47}
48#elif defined(CONFIG_SMP)
49# error "Arch needs cpu_idle_wait() equivalent here"
50#else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
51static void cpuidle_kick_cpus(void) {}
52#endif
53
54static int __cpuidle_register_device(struct cpuidle_device *dev);
55
56/**
57 * cpuidle_idle_call - the main idle loop
58 *
59 * NOTE: no locks or semaphores should be used here
60 * return non-zero on failure
61 */
62int cpuidle_idle_call(void)
63{
64	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
65	struct cpuidle_driver *drv = cpuidle_get_driver();
66	struct cpuidle_state *target_state;
67	int next_state, entered_state;
68
69	if (off)
70		return -ENODEV;
71
72	if (!initialized)
73		return -ENODEV;
74
75	/* check if the device is ready */
76	if (!dev || !dev->enabled)
77		return -EBUSY;
78
79#if 0
80	/* shows regressions, re-enable for 2.6.29 */
81	/*
82	 * run any timers that can be run now, at this point
83	 * before calculating the idle duration etc.
84	 */
85	hrtimer_peek_ahead_timers();
86#endif
87
88	/* ask the governor for the next state */
89	next_state = cpuidle_curr_governor->select(drv, dev);
90	if (need_resched()) {
91		local_irq_enable();
92		return 0;
93	}
94
95	target_state = &drv->states[next_state];
96
97	trace_power_start(POWER_CSTATE, next_state, dev->cpu);
98	trace_cpu_idle(next_state, dev->cpu);
99
100	entered_state = target_state->enter(dev, drv, next_state);
101
102	trace_power_end(dev->cpu);
103	trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
104
105	if (entered_state >= 0) {
106		/* Update cpuidle counters */
107		/* This can be moved to within driver enter routine
108		 * but that results in multiple copies of same code.
109		 */
110		dev->states_usage[entered_state].time +=
111				(unsigned long long)dev->last_residency;
112		dev->states_usage[entered_state].usage++;
113	}
114
115	/* give the governor an opportunity to reflect on the outcome */
116	if (cpuidle_curr_governor->reflect)
117		cpuidle_curr_governor->reflect(dev, entered_state);
118
119	return 0;
120}
121
122/**
123 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
124 */
125void cpuidle_install_idle_handler(void)
126{
127	if (enabled_devices) {
128		/* Make sure all changes finished before we switch to new idle */
129		smp_wmb();
130		initialized = 1;
131	}
132}
133
134/**
135 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
136 */
137void cpuidle_uninstall_idle_handler(void)
138{
139	if (enabled_devices) {
140		initialized = 0;
141		cpuidle_kick_cpus();
142	}
143}
144
145/**
146 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
147 */
148void cpuidle_pause_and_lock(void)
149{
150	mutex_lock(&cpuidle_lock);
151	cpuidle_uninstall_idle_handler();
152}
153
154EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
155
156/**
157 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
158 */
159void cpuidle_resume_and_unlock(void)
160{
161	cpuidle_install_idle_handler();
162	mutex_unlock(&cpuidle_lock);
163}
164
165EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
166
167#ifdef CONFIG_ARCH_HAS_CPU_RELAX
168static int poll_idle(struct cpuidle_device *dev,
169		struct cpuidle_driver *drv, int index)
170{
171	ktime_t	t1, t2;
172	s64 diff;
173
174	t1 = ktime_get();
175	local_irq_enable();
176	while (!need_resched())
177		cpu_relax();
178
179	t2 = ktime_get();
180	diff = ktime_to_us(ktime_sub(t2, t1));
181	if (diff > INT_MAX)
182		diff = INT_MAX;
183
184	dev->last_residency = (int) diff;
185
186	return index;
187}
188
189static void poll_idle_init(struct cpuidle_driver *drv)
190{
191	struct cpuidle_state *state = &drv->states[0];
192
193	snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
194	snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
195	state->exit_latency = 0;
196	state->target_residency = 0;
197	state->power_usage = -1;
198	state->flags = 0;
199	state->enter = poll_idle;
200}
201#else
202static void poll_idle_init(struct cpuidle_driver *drv) {}
203#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
204
205/**
206 * cpuidle_enable_device - enables idle PM for a CPU
207 * @dev: the CPU
208 *
209 * This function must be called between cpuidle_pause_and_lock and
210 * cpuidle_resume_and_unlock when used externally.
211 */
212int cpuidle_enable_device(struct cpuidle_device *dev)
213{
214	int ret, i;
215
216	if (dev->enabled)
217		return 0;
218	if (!cpuidle_get_driver() || !cpuidle_curr_governor)
219		return -EIO;
220	if (!dev->state_count)
221		return -EINVAL;
222
223	if (dev->registered == 0) {
224		ret = __cpuidle_register_device(dev);
225		if (ret)
226			return ret;
227	}
228
229	poll_idle_init(cpuidle_get_driver());
230
231	if ((ret = cpuidle_add_state_sysfs(dev)))
232		return ret;
233
234	if (cpuidle_curr_governor->enable &&
235	    (ret = cpuidle_curr_governor->enable(cpuidle_get_driver(), dev)))
236		goto fail_sysfs;
237
238	for (i = 0; i < dev->state_count; i++) {
239		dev->states_usage[i].usage = 0;
240		dev->states_usage[i].time = 0;
241	}
242	dev->last_residency = 0;
243
244	smp_wmb();
245
246	dev->enabled = 1;
247
248	enabled_devices++;
249	return 0;
250
251fail_sysfs:
252	cpuidle_remove_state_sysfs(dev);
253
254	return ret;
255}
256
257EXPORT_SYMBOL_GPL(cpuidle_enable_device);
258
259/**
260 * cpuidle_disable_device - disables idle PM for a CPU
261 * @dev: the CPU
262 *
263 * This function must be called between cpuidle_pause_and_lock and
264 * cpuidle_resume_and_unlock when used externally.
265 */
266void cpuidle_disable_device(struct cpuidle_device *dev)
267{
268	if (!dev->enabled)
269		return;
270	if (!cpuidle_get_driver() || !cpuidle_curr_governor)
271		return;
272
273	dev->enabled = 0;
274
275	if (cpuidle_curr_governor->disable)
276		cpuidle_curr_governor->disable(cpuidle_get_driver(), dev);
277
278	cpuidle_remove_state_sysfs(dev);
279	enabled_devices--;
280}
281
282EXPORT_SYMBOL_GPL(cpuidle_disable_device);
283
284/**
285 * __cpuidle_register_device - internal register function called before register
286 * and enable routines
287 * @dev: the cpu
288 *
289 * cpuidle_lock mutex must be held before this is called
290 */
291static int __cpuidle_register_device(struct cpuidle_device *dev)
292{
293	int ret;
294	struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
295	struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
296
297	if (!dev)
298		return -EINVAL;
299	if (!try_module_get(cpuidle_driver->owner))
300		return -EINVAL;
301
302	init_completion(&dev->kobj_unregister);
303
304	per_cpu(cpuidle_devices, dev->cpu) = dev;
305	list_add(&dev->device_list, &cpuidle_detected_devices);
306	if ((ret = cpuidle_add_sysfs(cpu_dev))) {
307		module_put(cpuidle_driver->owner);
308		return ret;
309	}
310
311	dev->registered = 1;
312	return 0;
313}
314
315/**
316 * cpuidle_register_device - registers a CPU's idle PM feature
317 * @dev: the cpu
318 */
319int cpuidle_register_device(struct cpuidle_device *dev)
320{
321	int ret;
322
323	mutex_lock(&cpuidle_lock);
324
325	if ((ret = __cpuidle_register_device(dev))) {
326		mutex_unlock(&cpuidle_lock);
327		return ret;
328	}
329
330	cpuidle_enable_device(dev);
331	cpuidle_install_idle_handler();
332
333	mutex_unlock(&cpuidle_lock);
334
335	return 0;
336
337}
338
339EXPORT_SYMBOL_GPL(cpuidle_register_device);
340
341/**
342 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
343 * @dev: the cpu
344 */
345void cpuidle_unregister_device(struct cpuidle_device *dev)
346{
347	struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
348	struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
349
350	if (dev->registered == 0)
351		return;
352
353	cpuidle_pause_and_lock();
354
355	cpuidle_disable_device(dev);
356
357	cpuidle_remove_sysfs(cpu_dev);
358	list_del(&dev->device_list);
359	wait_for_completion(&dev->kobj_unregister);
360	per_cpu(cpuidle_devices, dev->cpu) = NULL;
361
362	cpuidle_resume_and_unlock();
363
364	module_put(cpuidle_driver->owner);
365}
366
367EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
368
369#ifdef CONFIG_SMP
370
371static void smp_callback(void *v)
372{
373	/* we already woke the CPU up, nothing more to do */
374}
375
376/*
377 * This function gets called when a part of the kernel has a new latency
378 * requirement.  This means we need to get all processors out of their C-state,
379 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
380 * wakes them all right up.
381 */
382static int cpuidle_latency_notify(struct notifier_block *b,
383		unsigned long l, void *v)
384{
385	smp_call_function(smp_callback, NULL, 1);
386	return NOTIFY_OK;
387}
388
389static struct notifier_block cpuidle_latency_notifier = {
390	.notifier_call = cpuidle_latency_notify,
391};
392
393static inline void latency_notifier_init(struct notifier_block *n)
394{
395	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
396}
397
398#else /* CONFIG_SMP */
399
400#define latency_notifier_init(x) do { } while (0)
401
402#endif /* CONFIG_SMP */
403
404/**
405 * cpuidle_init - core initializer
406 */
407static int __init cpuidle_init(void)
408{
409	int ret;
410
411	if (cpuidle_disabled())
412		return -ENODEV;
413
414	ret = cpuidle_add_interface(cpu_subsys.dev_root);
415	if (ret)
416		return ret;
417
418	latency_notifier_init(&cpuidle_latency_notifier);
419
420	return 0;
421}
422
423module_param(off, int, 0444);
424core_initcall(cpuidle_init);
425