cpuidle.c revision 4126c0197bc8c58a0bb7fcda07b01b596b6fb4c5 
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
56static inline int cpuidle_enter(struct cpuidle_device *dev,
57				struct cpuidle_driver *drv, int index)
58{
59	struct cpuidle_state *target_state = &drv->states[index];
60	return target_state->enter(dev, drv, index);
61}
62
63static inline int cpuidle_enter_tk(struct cpuidle_device *dev,
64			       struct cpuidle_driver *drv, int index)
65{
66	return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter);
67}
68
69typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev,
70			       struct cpuidle_driver *drv, int index);
71
72static cpuidle_enter_t cpuidle_enter_ops;
73
74/**
75 * cpuidle_play_dead - cpu off-lining
76 *
77 * Returns in case of an error or no driver
78 */
79int cpuidle_play_dead(void)
80{
81	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
82	struct cpuidle_driver *drv = cpuidle_get_driver();
83	int i, dead_state = -1;
84	int power_usage = -1;
85
86	if (!drv)
87		return -ENODEV;
88
89	/* Find lowest-power state that supports long-term idle */
90	for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
91		struct cpuidle_state *s = &drv->states[i];
92
93		if (s->power_usage < power_usage && s->enter_dead) {
94			power_usage = s->power_usage;
95			dead_state = i;
96		}
97	}
98
99	if (dead_state != -1)
100		return drv->states[dead_state].enter_dead(dev, dead_state);
101
102	return -ENODEV;
103}
104
105/**
106 * cpuidle_enter_state - enter the state and update stats
107 * @dev: cpuidle device for this cpu
108 * @drv: cpuidle driver for this cpu
109 * @next_state: index into drv->states of the state to enter
110 */
111int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
112		int next_state)
113{
114	int entered_state;
115
116	entered_state = cpuidle_enter_ops(dev, drv, next_state);
117
118	if (entered_state >= 0) {
119		/* Update cpuidle counters */
120		/* This can be moved to within driver enter routine
121		 * but that results in multiple copies of same code.
122		 */
123		dev->states_usage[entered_state].time +=
124				(unsigned long long)dev->last_residency;
125		dev->states_usage[entered_state].usage++;
126	} else {
127		dev->last_residency = 0;
128	}
129
130	return entered_state;
131}
132
133/**
134 * cpuidle_idle_call - the main idle loop
135 *
136 * NOTE: no locks or semaphores should be used here
137 * return non-zero on failure
138 */
139int cpuidle_idle_call(void)
140{
141	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
142	struct cpuidle_driver *drv = cpuidle_get_driver();
143	int next_state, entered_state;
144
145	if (off)
146		return -ENODEV;
147
148	if (!initialized)
149		return -ENODEV;
150
151	/* check if the device is ready */
152	if (!dev || !dev->enabled)
153		return -EBUSY;
154
155#if 0
156	/* shows regressions, re-enable for 2.6.29 */
157	/*
158	 * run any timers that can be run now, at this point
159	 * before calculating the idle duration etc.
160	 */
161	hrtimer_peek_ahead_timers();
162#endif
163
164	/* ask the governor for the next state */
165	next_state = cpuidle_curr_governor->select(drv, dev);
166	if (need_resched()) {
167		local_irq_enable();
168		return 0;
169	}
170
171	trace_power_start_rcuidle(POWER_CSTATE, next_state, dev->cpu);
172	trace_cpu_idle_rcuidle(next_state, dev->cpu);
173
174	if (cpuidle_state_is_coupled(dev, drv, next_state))
175		entered_state = cpuidle_enter_state_coupled(dev, drv,
176							    next_state);
177	else
178		entered_state = cpuidle_enter_state(dev, drv, next_state);
179
180	trace_power_end_rcuidle(dev->cpu);
181	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
182
183	/* give the governor an opportunity to reflect on the outcome */
184	if (cpuidle_curr_governor->reflect)
185		cpuidle_curr_governor->reflect(dev, entered_state);
186
187	return 0;
188}
189
190/**
191 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
192 */
193void cpuidle_install_idle_handler(void)
194{
195	if (enabled_devices) {
196		/* Make sure all changes finished before we switch to new idle */
197		smp_wmb();
198		initialized = 1;
199	}
200}
201
202/**
203 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
204 */
205void cpuidle_uninstall_idle_handler(void)
206{
207	if (enabled_devices) {
208		initialized = 0;
209		cpuidle_kick_cpus();
210	}
211}
212
213/**
214 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
215 */
216void cpuidle_pause_and_lock(void)
217{
218	mutex_lock(&cpuidle_lock);
219	cpuidle_uninstall_idle_handler();
220}
221
222EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
223
224/**
225 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
226 */
227void cpuidle_resume_and_unlock(void)
228{
229	cpuidle_install_idle_handler();
230	mutex_unlock(&cpuidle_lock);
231}
232
233EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
234
235/**
236 * cpuidle_wrap_enter - performs timekeeping and irqen around enter function
237 * @dev: pointer to a valid cpuidle_device object
238 * @drv: pointer to a valid cpuidle_driver object
239 * @index: index of the target cpuidle state.
240 */
241int cpuidle_wrap_enter(struct cpuidle_device *dev,
242				struct cpuidle_driver *drv, int index,
243				int (*enter)(struct cpuidle_device *dev,
244					struct cpuidle_driver *drv, int index))
245{
246	ktime_t time_start, time_end;
247	s64 diff;
248
249	time_start = ktime_get();
250
251	index = enter(dev, drv, index);
252
253	time_end = ktime_get();
254
255	local_irq_enable();
256
257	diff = ktime_to_us(ktime_sub(time_end, time_start));
258	if (diff > INT_MAX)
259		diff = INT_MAX;
260
261	dev->last_residency = (int) diff;
262
263	return index;
264}
265
266#ifdef CONFIG_ARCH_HAS_CPU_RELAX
267static int poll_idle(struct cpuidle_device *dev,
268		struct cpuidle_driver *drv, int index)
269{
270	ktime_t	t1, t2;
271	s64 diff;
272
273	t1 = ktime_get();
274	local_irq_enable();
275	while (!need_resched())
276		cpu_relax();
277
278	t2 = ktime_get();
279	diff = ktime_to_us(ktime_sub(t2, t1));
280	if (diff > INT_MAX)
281		diff = INT_MAX;
282
283	dev->last_residency = (int) diff;
284
285	return index;
286}
287
288static void poll_idle_init(struct cpuidle_driver *drv)
289{
290	struct cpuidle_state *state = &drv->states[0];
291
292	snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
293	snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
294	state->exit_latency = 0;
295	state->target_residency = 0;
296	state->power_usage = -1;
297	state->flags = 0;
298	state->enter = poll_idle;
299	state->disable = 0;
300}
301#else
302static void poll_idle_init(struct cpuidle_driver *drv) {}
303#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
304
305/**
306 * cpuidle_enable_device - enables idle PM for a CPU
307 * @dev: the CPU
308 *
309 * This function must be called between cpuidle_pause_and_lock and
310 * cpuidle_resume_and_unlock when used externally.
311 */
312int cpuidle_enable_device(struct cpuidle_device *dev)
313{
314	int ret, i;
315	struct cpuidle_driver *drv = cpuidle_get_driver();
316
317	if (dev->enabled)
318		return 0;
319	if (!drv || !cpuidle_curr_governor)
320		return -EIO;
321	if (!dev->state_count)
322		dev->state_count = drv->state_count;
323
324	if (dev->registered == 0) {
325		ret = __cpuidle_register_device(dev);
326		if (ret)
327			return ret;
328	}
329
330	cpuidle_enter_ops = drv->en_core_tk_irqen ?
331		cpuidle_enter_tk : cpuidle_enter;
332
333	poll_idle_init(drv);
334
335	if ((ret = cpuidle_add_state_sysfs(dev)))
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_state_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	if (!dev->enabled)
373		return;
374	if (!cpuidle_get_driver() || !cpuidle_curr_governor)
375		return;
376
377	dev->enabled = 0;
378
379	if (cpuidle_curr_governor->disable)
380		cpuidle_curr_governor->disable(cpuidle_get_driver(), dev);
381
382	cpuidle_remove_state_sysfs(dev);
383	enabled_devices--;
384}
385
386EXPORT_SYMBOL_GPL(cpuidle_disable_device);
387
388/**
389 * __cpuidle_register_device - internal register function called before register
390 * and enable routines
391 * @dev: the cpu
392 *
393 * cpuidle_lock mutex must be held before this is called
394 */
395static int __cpuidle_register_device(struct cpuidle_device *dev)
396{
397	int ret;
398	struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
399	struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
400
401	if (!dev)
402		return -EINVAL;
403	if (!try_module_get(cpuidle_driver->owner))
404		return -EINVAL;
405
406	init_completion(&dev->kobj_unregister);
407
408	per_cpu(cpuidle_devices, dev->cpu) = dev;
409	list_add(&dev->device_list, &cpuidle_detected_devices);
410	ret = cpuidle_add_sysfs(cpu_dev);
411	if (ret)
412		goto err_sysfs;
413
414	ret = cpuidle_coupled_register_device(dev);
415	if (ret)
416		goto err_coupled;
417
418	dev->registered = 1;
419	return 0;
420
421err_coupled:
422	cpuidle_remove_sysfs(cpu_dev);
423	wait_for_completion(&dev->kobj_unregister);
424err_sysfs:
425	list_del(&dev->device_list);
426	per_cpu(cpuidle_devices, dev->cpu) = NULL;
427	module_put(cpuidle_driver->owner);
428	return ret;
429}
430
431/**
432 * cpuidle_register_device - registers a CPU's idle PM feature
433 * @dev: the cpu
434 */
435int cpuidle_register_device(struct cpuidle_device *dev)
436{
437	int ret;
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 device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
464	struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
465
466	if (dev->registered == 0)
467		return;
468
469	cpuidle_pause_and_lock();
470
471	cpuidle_disable_device(dev);
472
473	cpuidle_remove_sysfs(cpu_dev);
474	list_del(&dev->device_list);
475	wait_for_completion(&dev->kobj_unregister);
476	per_cpu(cpuidle_devices, dev->cpu) = NULL;
477
478	cpuidle_coupled_unregister_device(dev);
479
480	cpuidle_resume_and_unlock();
481
482	module_put(cpuidle_driver->owner);
483}
484
485EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
486
487#ifdef CONFIG_SMP
488
489static void smp_callback(void *v)
490{
491	/* we already woke the CPU up, nothing more to do */
492}
493
494/*
495 * This function gets called when a part of the kernel has a new latency
496 * requirement.  This means we need to get all processors out of their C-state,
497 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
498 * wakes them all right up.
499 */
500static int cpuidle_latency_notify(struct notifier_block *b,
501		unsigned long l, void *v)
502{
503	smp_call_function(smp_callback, NULL, 1);
504	return NOTIFY_OK;
505}
506
507static struct notifier_block cpuidle_latency_notifier = {
508	.notifier_call = cpuidle_latency_notify,
509};
510
511static inline void latency_notifier_init(struct notifier_block *n)
512{
513	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
514}
515
516#else /* CONFIG_SMP */
517
518#define latency_notifier_init(x) do { } while (0)
519
520#endif /* CONFIG_SMP */
521
522/**
523 * cpuidle_init - core initializer
524 */
525static int __init cpuidle_init(void)
526{
527	int ret;
528
529	if (cpuidle_disabled())
530		return -ENODEV;
531
532	ret = cpuidle_add_interface(cpu_subsys.dev_root);
533	if (ret)
534		return ret;
535
536	latency_notifier_init(&cpuidle_latency_notifier);
537
538	return 0;
539}
540
541module_param(off, int, 0444);
542core_initcall(cpuidle_init);
543