cpuidle.c revision 8aef33a7cf40ca9da188e8578b2abe7267a38c52 
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
43static int __cpuidle_register_device(struct cpuidle_device *dev);
44
45static inline int cpuidle_enter(struct cpuidle_device *dev,
46				struct cpuidle_driver *drv, int index)
47{
48	struct cpuidle_state *target_state = &drv->states[index];
49	return target_state->enter(dev, drv, index);
50}
51
52static inline int cpuidle_enter_tk(struct cpuidle_device *dev,
53			       struct cpuidle_driver *drv, int index)
54{
55	return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter);
56}
57
58typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev,
59			       struct cpuidle_driver *drv, int index);
60
61static cpuidle_enter_t cpuidle_enter_ops;
62
63/**
64 * cpuidle_play_dead - cpu off-lining
65 *
66 * Returns in case of an error or no driver
67 */
68int cpuidle_play_dead(void)
69{
70	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
71	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
72	int i;
73
74	if (!drv)
75		return -ENODEV;
76
77	/* Find lowest-power state that supports long-term idle */
78	for (i = drv->state_count - 1; i >= CPUIDLE_DRIVER_STATE_START; i--)
79		if (drv->states[i].enter_dead)
80			return drv->states[i].enter_dead(dev, i);
81
82	return -ENODEV;
83}
84
85/**
86 * cpuidle_enter_state - enter the state and update stats
87 * @dev: cpuidle device for this cpu
88 * @drv: cpuidle driver for this cpu
89 * @next_state: index into drv->states of the state to enter
90 */
91int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
92		int next_state)
93{
94	int entered_state;
95
96	entered_state = cpuidle_enter_ops(dev, drv, next_state);
97
98	if (entered_state >= 0) {
99		/* Update cpuidle counters */
100		/* This can be moved to within driver enter routine
101		 * but that results in multiple copies of same code.
102		 */
103		dev->states_usage[entered_state].time += dev->last_residency;
104		dev->states_usage[entered_state].usage++;
105	} else {
106		dev->last_residency = 0;
107	}
108
109	return entered_state;
110}
111
112/**
113 * cpuidle_idle_call - the main idle loop
114 *
115 * NOTE: no locks or semaphores should be used here
116 * return non-zero on failure
117 */
118int cpuidle_idle_call(void)
119{
120	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
121	struct cpuidle_driver *drv;
122	int next_state, entered_state;
123
124	if (off)
125		return -ENODEV;
126
127	if (!initialized)
128		return -ENODEV;
129
130	/* check if the device is ready */
131	if (!dev || !dev->enabled)
132		return -EBUSY;
133
134	drv = cpuidle_get_cpu_driver(dev);
135
136	/* ask the governor for the next state */
137	next_state = cpuidle_curr_governor->select(drv, dev);
138	if (need_resched()) {
139		dev->last_residency = 0;
140		/* give the governor an opportunity to reflect on the outcome */
141		if (cpuidle_curr_governor->reflect)
142			cpuidle_curr_governor->reflect(dev, next_state);
143		local_irq_enable();
144		return 0;
145	}
146
147	trace_power_start_rcuidle(POWER_CSTATE, next_state, dev->cpu);
148	trace_cpu_idle_rcuidle(next_state, dev->cpu);
149
150	if (cpuidle_state_is_coupled(dev, drv, next_state))
151		entered_state = cpuidle_enter_state_coupled(dev, drv,
152							    next_state);
153	else
154		entered_state = cpuidle_enter_state(dev, drv, next_state);
155
156	trace_power_end_rcuidle(dev->cpu);
157	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
158
159	/* give the governor an opportunity to reflect on the outcome */
160	if (cpuidle_curr_governor->reflect)
161		cpuidle_curr_governor->reflect(dev, entered_state);
162
163	return 0;
164}
165
166/**
167 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
168 */
169void cpuidle_install_idle_handler(void)
170{
171	if (enabled_devices) {
172		/* Make sure all changes finished before we switch to new idle */
173		smp_wmb();
174		initialized = 1;
175	}
176}
177
178/**
179 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
180 */
181void cpuidle_uninstall_idle_handler(void)
182{
183	if (enabled_devices) {
184		initialized = 0;
185		kick_all_cpus_sync();
186	}
187}
188
189/**
190 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
191 */
192void cpuidle_pause_and_lock(void)
193{
194	mutex_lock(&cpuidle_lock);
195	cpuidle_uninstall_idle_handler();
196}
197
198EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
199
200/**
201 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
202 */
203void cpuidle_resume_and_unlock(void)
204{
205	cpuidle_install_idle_handler();
206	mutex_unlock(&cpuidle_lock);
207}
208
209EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
210
211/* Currently used in suspend/resume path to suspend cpuidle */
212void cpuidle_pause(void)
213{
214	mutex_lock(&cpuidle_lock);
215	cpuidle_uninstall_idle_handler();
216	mutex_unlock(&cpuidle_lock);
217}
218
219/* Currently used in suspend/resume path to resume cpuidle */
220void cpuidle_resume(void)
221{
222	mutex_lock(&cpuidle_lock);
223	cpuidle_install_idle_handler();
224	mutex_unlock(&cpuidle_lock);
225}
226
227/**
228 * cpuidle_wrap_enter - performs timekeeping and irqen around enter function
229 * @dev: pointer to a valid cpuidle_device object
230 * @drv: pointer to a valid cpuidle_driver object
231 * @index: index of the target cpuidle state.
232 */
233int cpuidle_wrap_enter(struct cpuidle_device *dev,
234				struct cpuidle_driver *drv, int index,
235				int (*enter)(struct cpuidle_device *dev,
236					struct cpuidle_driver *drv, int index))
237{
238	ktime_t time_start, time_end;
239	s64 diff;
240
241	time_start = ktime_get();
242
243	index = enter(dev, drv, index);
244
245	time_end = ktime_get();
246
247	local_irq_enable();
248
249	diff = ktime_to_us(ktime_sub(time_end, time_start));
250	if (diff > INT_MAX)
251		diff = INT_MAX;
252
253	dev->last_residency = (int) diff;
254
255	return index;
256}
257
258#ifdef CONFIG_ARCH_HAS_CPU_RELAX
259static int poll_idle(struct cpuidle_device *dev,
260		struct cpuidle_driver *drv, int index)
261{
262	ktime_t	t1, t2;
263	s64 diff;
264
265	t1 = ktime_get();
266	local_irq_enable();
267	while (!need_resched())
268		cpu_relax();
269
270	t2 = ktime_get();
271	diff = ktime_to_us(ktime_sub(t2, t1));
272	if (diff > INT_MAX)
273		diff = INT_MAX;
274
275	dev->last_residency = (int) diff;
276
277	return index;
278}
279
280static void poll_idle_init(struct cpuidle_driver *drv)
281{
282	struct cpuidle_state *state = &drv->states[0];
283
284	snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
285	snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
286	state->exit_latency = 0;
287	state->target_residency = 0;
288	state->power_usage = -1;
289	state->flags = 0;
290	state->enter = poll_idle;
291	state->disabled = false;
292}
293#else
294static void poll_idle_init(struct cpuidle_driver *drv) {}
295#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
296
297/**
298 * cpuidle_enable_device - enables idle PM for a CPU
299 * @dev: the CPU
300 *
301 * This function must be called between cpuidle_pause_and_lock and
302 * cpuidle_resume_and_unlock when used externally.
303 */
304int cpuidle_enable_device(struct cpuidle_device *dev)
305{
306	int ret, i;
307	struct cpuidle_driver *drv;
308
309	if (!dev)
310		return -EINVAL;
311
312	if (dev->enabled)
313		return 0;
314
315	drv = cpuidle_get_cpu_driver(dev);
316
317	if (!drv || !cpuidle_curr_governor)
318		return -EIO;
319
320	if (!dev->state_count)
321		dev->state_count = drv->state_count;
322
323	if (dev->registered == 0) {
324		ret = __cpuidle_register_device(dev);
325		if (ret)
326			return ret;
327	}
328
329	cpuidle_enter_ops = drv->en_core_tk_irqen ?
330		cpuidle_enter_tk : cpuidle_enter;
331
332	poll_idle_init(drv);
333
334	ret = cpuidle_add_device_sysfs(dev);
335	if (ret)
336		return ret;
337
338	if (cpuidle_curr_governor->enable &&
339	    (ret = cpuidle_curr_governor->enable(drv, dev)))
340		goto fail_sysfs;
341
342	for (i = 0; i < dev->state_count; i++) {
343		dev->states_usage[i].usage = 0;
344		dev->states_usage[i].time = 0;
345	}
346	dev->last_residency = 0;
347
348	smp_wmb();
349
350	dev->enabled = 1;
351
352	enabled_devices++;
353	return 0;
354
355fail_sysfs:
356	cpuidle_remove_device_sysfs(dev);
357
358	return ret;
359}
360
361EXPORT_SYMBOL_GPL(cpuidle_enable_device);
362
363/**
364 * cpuidle_disable_device - disables idle PM for a CPU
365 * @dev: the CPU
366 *
367 * This function must be called between cpuidle_pause_and_lock and
368 * cpuidle_resume_and_unlock when used externally.
369 */
370void cpuidle_disable_device(struct cpuidle_device *dev)
371{
372	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
373
374	if (!dev || !dev->enabled)
375		return;
376
377	if (!drv || !cpuidle_curr_governor)
378		return;
379
380	dev->enabled = 0;
381
382	if (cpuidle_curr_governor->disable)
383		cpuidle_curr_governor->disable(drv, dev);
384
385	cpuidle_remove_device_sysfs(dev);
386	enabled_devices--;
387}
388
389EXPORT_SYMBOL_GPL(cpuidle_disable_device);
390
391/**
392 * __cpuidle_register_device - internal register function called before register
393 * and enable routines
394 * @dev: the cpu
395 *
396 * cpuidle_lock mutex must be held before this is called
397 */
398static int __cpuidle_register_device(struct cpuidle_device *dev)
399{
400	int ret;
401	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
402
403	if (!try_module_get(drv->owner))
404		return -EINVAL;
405
406	per_cpu(cpuidle_devices, dev->cpu) = dev;
407	list_add(&dev->device_list, &cpuidle_detected_devices);
408	ret = cpuidle_add_sysfs(dev);
409	if (ret)
410		goto err_sysfs;
411
412	ret = cpuidle_coupled_register_device(dev);
413	if (ret)
414		goto err_coupled;
415
416	dev->registered = 1;
417	return 0;
418
419err_coupled:
420	cpuidle_remove_sysfs(dev);
421err_sysfs:
422	list_del(&dev->device_list);
423	per_cpu(cpuidle_devices, dev->cpu) = NULL;
424	module_put(drv->owner);
425	return ret;
426}
427
428/**
429 * cpuidle_register_device - registers a CPU's idle PM feature
430 * @dev: the cpu
431 */
432int cpuidle_register_device(struct cpuidle_device *dev)
433{
434	int ret;
435
436	if (!dev)
437		return -EINVAL;
438
439	mutex_lock(&cpuidle_lock);
440
441	if ((ret = __cpuidle_register_device(dev))) {
442		mutex_unlock(&cpuidle_lock);
443		return ret;
444	}
445
446	cpuidle_enable_device(dev);
447	cpuidle_install_idle_handler();
448
449	mutex_unlock(&cpuidle_lock);
450
451	return 0;
452
453}
454
455EXPORT_SYMBOL_GPL(cpuidle_register_device);
456
457/**
458 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
459 * @dev: the cpu
460 */
461void cpuidle_unregister_device(struct cpuidle_device *dev)
462{
463	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
464
465	if (dev->registered == 0)
466		return;
467
468	cpuidle_pause_and_lock();
469
470	cpuidle_disable_device(dev);
471
472	cpuidle_remove_sysfs(dev);
473	list_del(&dev->device_list);
474	per_cpu(cpuidle_devices, dev->cpu) = NULL;
475
476	cpuidle_coupled_unregister_device(dev);
477
478	cpuidle_resume_and_unlock();
479
480	module_put(drv->owner);
481}
482
483EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
484
485#ifdef CONFIG_SMP
486
487static void smp_callback(void *v)
488{
489	/* we already woke the CPU up, nothing more to do */
490}
491
492/*
493 * This function gets called when a part of the kernel has a new latency
494 * requirement.  This means we need to get all processors out of their C-state,
495 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
496 * wakes them all right up.
497 */
498static int cpuidle_latency_notify(struct notifier_block *b,
499		unsigned long l, void *v)
500{
501	smp_call_function(smp_callback, NULL, 1);
502	return NOTIFY_OK;
503}
504
505static struct notifier_block cpuidle_latency_notifier = {
506	.notifier_call = cpuidle_latency_notify,
507};
508
509static inline void latency_notifier_init(struct notifier_block *n)
510{
511	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
512}
513
514#else /* CONFIG_SMP */
515
516#define latency_notifier_init(x) do { } while (0)
517
518#endif /* CONFIG_SMP */
519
520/**
521 * cpuidle_init - core initializer
522 */
523static int __init cpuidle_init(void)
524{
525	int ret;
526
527	if (cpuidle_disabled())
528		return -ENODEV;
529
530	ret = cpuidle_add_interface(cpu_subsys.dev_root);
531	if (ret)
532		return ret;
533
534	latency_notifier_init(&cpuidle_latency_notifier);
535
536	return 0;
537}
538
539module_param(off, int, 0444);
540core_initcall(cpuidle_init);
541