cpuidle.c revision b60e6a0eb0273132cbb60a9806abf5f47a4aee1c 
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/clockchips.h>
12#include <linux/kernel.h>
13#include <linux/mutex.h>
14#include <linux/sched.h>
15#include <linux/notifier.h>
16#include <linux/pm_qos.h>
17#include <linux/cpu.h>
18#include <linux/cpuidle.h>
19#include <linux/ktime.h>
20#include <linux/hrtimer.h>
21#include <linux/module.h>
22#include <trace/events/power.h>
23
24#include "cpuidle.h"
25
26DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
27
28DEFINE_MUTEX(cpuidle_lock);
29LIST_HEAD(cpuidle_detected_devices);
30
31static int enabled_devices;
32static int off __read_mostly;
33static int initialized __read_mostly;
34
35int cpuidle_disabled(void)
36{
37	return off;
38}
39void disable_cpuidle(void)
40{
41	off = 1;
42}
43
44static int __cpuidle_register_device(struct cpuidle_device *dev);
45
46static inline int cpuidle_enter(struct cpuidle_device *dev,
47				struct cpuidle_driver *drv, int index)
48{
49	struct cpuidle_state *target_state = &drv->states[index];
50	return target_state->enter(dev, drv, index);
51}
52
53static inline int cpuidle_enter_tk(struct cpuidle_device *dev,
54			       struct cpuidle_driver *drv, int index)
55{
56	return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter);
57}
58
59typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev,
60			       struct cpuidle_driver *drv, int index);
61
62static cpuidle_enter_t cpuidle_enter_ops;
63
64/**
65 * cpuidle_play_dead - cpu off-lining
66 *
67 * Returns in case of an error or no driver
68 */
69int cpuidle_play_dead(void)
70{
71	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
72	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
73	int i;
74
75	if (!drv)
76		return -ENODEV;
77
78	/* Find lowest-power state that supports long-term idle */
79	for (i = drv->state_count - 1; i >= CPUIDLE_DRIVER_STATE_START; i--)
80		if (drv->states[i].enter_dead)
81			return drv->states[i].enter_dead(dev, i);
82
83	return -ENODEV;
84}
85
86/**
87 * cpuidle_enter_state - enter the state and update stats
88 * @dev: cpuidle device for this cpu
89 * @drv: cpuidle driver for this cpu
90 * @next_state: index into drv->states of the state to enter
91 */
92int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
93		int next_state)
94{
95	int entered_state;
96
97	entered_state = cpuidle_enter_ops(dev, drv, next_state);
98
99	if (entered_state >= 0) {
100		/* Update cpuidle counters */
101		/* This can be moved to within driver enter routine
102		 * but that results in multiple copies of same code.
103		 */
104		dev->states_usage[entered_state].time += dev->last_residency;
105		dev->states_usage[entered_state].usage++;
106	} else {
107		dev->last_residency = 0;
108	}
109
110	return entered_state;
111}
112
113/**
114 * cpuidle_idle_call - the main idle loop
115 *
116 * NOTE: no locks or semaphores should be used here
117 * return non-zero on failure
118 */
119int cpuidle_idle_call(void)
120{
121	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
122	struct cpuidle_driver *drv;
123	int next_state, entered_state;
124
125	if (off)
126		return -ENODEV;
127
128	if (!initialized)
129		return -ENODEV;
130
131	/* check if the device is ready */
132	if (!dev || !dev->enabled)
133		return -EBUSY;
134
135	drv = cpuidle_get_cpu_driver(dev);
136
137	/* ask the governor for the next state */
138	next_state = cpuidle_curr_governor->select(drv, dev);
139	if (need_resched()) {
140		dev->last_residency = 0;
141		/* give the governor an opportunity to reflect on the outcome */
142		if (cpuidle_curr_governor->reflect)
143			cpuidle_curr_governor->reflect(dev, next_state);
144		local_irq_enable();
145		return 0;
146	}
147
148	trace_cpu_idle_rcuidle(next_state, dev->cpu);
149
150	if (drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP)
151		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
152				   &dev->cpu);
153
154	if (cpuidle_state_is_coupled(dev, drv, next_state))
155		entered_state = cpuidle_enter_state_coupled(dev, drv,
156							    next_state);
157	else
158		entered_state = cpuidle_enter_state(dev, drv, next_state);
159
160	if (drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP)
161		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
162				   &dev->cpu);
163
164	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
165
166	/* give the governor an opportunity to reflect on the outcome */
167	if (cpuidle_curr_governor->reflect)
168		cpuidle_curr_governor->reflect(dev, entered_state);
169
170	return 0;
171}
172
173/**
174 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
175 */
176void cpuidle_install_idle_handler(void)
177{
178	if (enabled_devices) {
179		/* Make sure all changes finished before we switch to new idle */
180		smp_wmb();
181		initialized = 1;
182	}
183}
184
185/**
186 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
187 */
188void cpuidle_uninstall_idle_handler(void)
189{
190	if (enabled_devices) {
191		initialized = 0;
192		kick_all_cpus_sync();
193	}
194}
195
196/**
197 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
198 */
199void cpuidle_pause_and_lock(void)
200{
201	mutex_lock(&cpuidle_lock);
202	cpuidle_uninstall_idle_handler();
203}
204
205EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
206
207/**
208 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
209 */
210void cpuidle_resume_and_unlock(void)
211{
212	cpuidle_install_idle_handler();
213	mutex_unlock(&cpuidle_lock);
214}
215
216EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
217
218/* Currently used in suspend/resume path to suspend cpuidle */
219void cpuidle_pause(void)
220{
221	mutex_lock(&cpuidle_lock);
222	cpuidle_uninstall_idle_handler();
223	mutex_unlock(&cpuidle_lock);
224}
225
226/* Currently used in suspend/resume path to resume cpuidle */
227void cpuidle_resume(void)
228{
229	mutex_lock(&cpuidle_lock);
230	cpuidle_install_idle_handler();
231	mutex_unlock(&cpuidle_lock);
232}
233
234/**
235 * cpuidle_wrap_enter - performs timekeeping and irqen around enter function
236 * @dev: pointer to a valid cpuidle_device object
237 * @drv: pointer to a valid cpuidle_driver object
238 * @index: index of the target cpuidle state.
239 */
240int cpuidle_wrap_enter(struct cpuidle_device *dev,
241				struct cpuidle_driver *drv, int index,
242				int (*enter)(struct cpuidle_device *dev,
243					struct cpuidle_driver *drv, int index))
244{
245	ktime_t time_start, time_end;
246	s64 diff;
247
248	time_start = ktime_get();
249
250	index = enter(dev, drv, index);
251
252	time_end = ktime_get();
253
254	local_irq_enable();
255
256	diff = ktime_to_us(ktime_sub(time_end, time_start));
257	if (diff > INT_MAX)
258		diff = INT_MAX;
259
260	dev->last_residency = (int) diff;
261
262	return index;
263}
264
265#ifdef CONFIG_ARCH_HAS_CPU_RELAX
266static int poll_idle(struct cpuidle_device *dev,
267		struct cpuidle_driver *drv, int index)
268{
269	ktime_t	t1, t2;
270	s64 diff;
271
272	t1 = ktime_get();
273	local_irq_enable();
274	while (!need_resched())
275		cpu_relax();
276
277	t2 = ktime_get();
278	diff = ktime_to_us(ktime_sub(t2, t1));
279	if (diff > INT_MAX)
280		diff = INT_MAX;
281
282	dev->last_residency = (int) diff;
283
284	return index;
285}
286
287static void poll_idle_init(struct cpuidle_driver *drv)
288{
289	struct cpuidle_state *state = &drv->states[0];
290
291	snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
292	snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
293	state->exit_latency = 0;
294	state->target_residency = 0;
295	state->power_usage = -1;
296	state->flags = 0;
297	state->enter = poll_idle;
298	state->disabled = false;
299}
300#else
301static void poll_idle_init(struct cpuidle_driver *drv) {}
302#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
303
304/**
305 * cpuidle_enable_device - enables idle PM for a CPU
306 * @dev: the CPU
307 *
308 * This function must be called between cpuidle_pause_and_lock and
309 * cpuidle_resume_and_unlock when used externally.
310 */
311int cpuidle_enable_device(struct cpuidle_device *dev)
312{
313	int ret, i;
314	struct cpuidle_driver *drv;
315
316	if (!dev)
317		return -EINVAL;
318
319	if (dev->enabled)
320		return 0;
321
322	drv = cpuidle_get_cpu_driver(dev);
323
324	if (!drv || !cpuidle_curr_governor)
325		return -EIO;
326
327	if (!dev->state_count)
328		dev->state_count = drv->state_count;
329
330	if (dev->registered == 0) {
331		ret = __cpuidle_register_device(dev);
332		if (ret)
333			return ret;
334	}
335
336	cpuidle_enter_ops = drv->en_core_tk_irqen ?
337		cpuidle_enter_tk : cpuidle_enter;
338
339	poll_idle_init(drv);
340
341	ret = cpuidle_add_device_sysfs(dev);
342	if (ret)
343		return ret;
344
345	if (cpuidle_curr_governor->enable &&
346	    (ret = cpuidle_curr_governor->enable(drv, dev)))
347		goto fail_sysfs;
348
349	for (i = 0; i < dev->state_count; i++) {
350		dev->states_usage[i].usage = 0;
351		dev->states_usage[i].time = 0;
352	}
353	dev->last_residency = 0;
354
355	smp_wmb();
356
357	dev->enabled = 1;
358
359	enabled_devices++;
360	return 0;
361
362fail_sysfs:
363	cpuidle_remove_device_sysfs(dev);
364
365	return ret;
366}
367
368EXPORT_SYMBOL_GPL(cpuidle_enable_device);
369
370/**
371 * cpuidle_disable_device - disables idle PM for a CPU
372 * @dev: the CPU
373 *
374 * This function must be called between cpuidle_pause_and_lock and
375 * cpuidle_resume_and_unlock when used externally.
376 */
377void cpuidle_disable_device(struct cpuidle_device *dev)
378{
379	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
380
381	if (!dev || !dev->enabled)
382		return;
383
384	if (!drv || !cpuidle_curr_governor)
385		return;
386
387	dev->enabled = 0;
388
389	if (cpuidle_curr_governor->disable)
390		cpuidle_curr_governor->disable(drv, dev);
391
392	cpuidle_remove_device_sysfs(dev);
393	enabled_devices--;
394}
395
396EXPORT_SYMBOL_GPL(cpuidle_disable_device);
397
398/**
399 * __cpuidle_register_device - internal register function called before register
400 * and enable routines
401 * @dev: the cpu
402 *
403 * cpuidle_lock mutex must be held before this is called
404 */
405static int __cpuidle_register_device(struct cpuidle_device *dev)
406{
407	int ret;
408	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
409
410	if (!try_module_get(drv->owner))
411		return -EINVAL;
412
413	per_cpu(cpuidle_devices, dev->cpu) = dev;
414	list_add(&dev->device_list, &cpuidle_detected_devices);
415	ret = cpuidle_add_sysfs(dev);
416	if (ret)
417		goto err_sysfs;
418
419	ret = cpuidle_coupled_register_device(dev);
420	if (ret)
421		goto err_coupled;
422
423	dev->registered = 1;
424	return 0;
425
426err_coupled:
427	cpuidle_remove_sysfs(dev);
428err_sysfs:
429	list_del(&dev->device_list);
430	per_cpu(cpuidle_devices, dev->cpu) = NULL;
431	module_put(drv->owner);
432	return ret;
433}
434
435/**
436 * cpuidle_register_device - registers a CPU's idle PM feature
437 * @dev: the cpu
438 */
439int cpuidle_register_device(struct cpuidle_device *dev)
440{
441	int ret;
442
443	if (!dev)
444		return -EINVAL;
445
446	mutex_lock(&cpuidle_lock);
447
448	if ((ret = __cpuidle_register_device(dev))) {
449		mutex_unlock(&cpuidle_lock);
450		return ret;
451	}
452
453	cpuidle_enable_device(dev);
454	cpuidle_install_idle_handler();
455
456	mutex_unlock(&cpuidle_lock);
457
458	return 0;
459
460}
461
462EXPORT_SYMBOL_GPL(cpuidle_register_device);
463
464/**
465 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
466 * @dev: the cpu
467 */
468void cpuidle_unregister_device(struct cpuidle_device *dev)
469{
470	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
471
472	if (dev->registered == 0)
473		return;
474
475	cpuidle_pause_and_lock();
476
477	cpuidle_disable_device(dev);
478
479	cpuidle_remove_sysfs(dev);
480	list_del(&dev->device_list);
481	per_cpu(cpuidle_devices, dev->cpu) = NULL;
482
483	cpuidle_coupled_unregister_device(dev);
484
485	cpuidle_resume_and_unlock();
486
487	module_put(drv->owner);
488}
489
490EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
491
492#ifdef CONFIG_SMP
493
494static void smp_callback(void *v)
495{
496	/* we already woke the CPU up, nothing more to do */
497}
498
499/*
500 * This function gets called when a part of the kernel has a new latency
501 * requirement.  This means we need to get all processors out of their C-state,
502 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
503 * wakes them all right up.
504 */
505static int cpuidle_latency_notify(struct notifier_block *b,
506		unsigned long l, void *v)
507{
508	smp_call_function(smp_callback, NULL, 1);
509	return NOTIFY_OK;
510}
511
512static struct notifier_block cpuidle_latency_notifier = {
513	.notifier_call = cpuidle_latency_notify,
514};
515
516static inline void latency_notifier_init(struct notifier_block *n)
517{
518	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
519}
520
521#else /* CONFIG_SMP */
522
523#define latency_notifier_init(x) do { } while (0)
524
525#endif /* CONFIG_SMP */
526
527/**
528 * cpuidle_init - core initializer
529 */
530static int __init cpuidle_init(void)
531{
532	int ret;
533
534	if (cpuidle_disabled())
535		return -ENODEV;
536
537	ret = cpuidle_add_interface(cpu_subsys.dev_root);
538	if (ret)
539		return ret;
540
541	latency_notifier_init(&cpuidle_latency_notifier);
542
543	return 0;
544}
545
546module_param(off, int, 0444);
547core_initcall(cpuidle_init);
548