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);
27DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
28
29DEFINE_MUTEX(cpuidle_lock);
30LIST_HEAD(cpuidle_detected_devices);
31
32static int enabled_devices;
33static int off __read_mostly;
34static int initialized __read_mostly;
35static bool use_deepest_state __read_mostly;
36
37int cpuidle_disabled(void)
38{
39	return off;
40}
41void disable_cpuidle(void)
42{
43	off = 1;
44}
45
46/**
47 * cpuidle_play_dead - cpu off-lining
48 *
49 * Returns in case of an error or no driver
50 */
51int cpuidle_play_dead(void)
52{
53	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
54	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
55	int i;
56
57	if (!drv)
58		return -ENODEV;
59
60	/* Find lowest-power state that supports long-term idle */
61	for (i = drv->state_count - 1; i >= CPUIDLE_DRIVER_STATE_START; i--)
62		if (drv->states[i].enter_dead)
63			return drv->states[i].enter_dead(dev, i);
64
65	return -ENODEV;
66}
67
68/**
69 * cpuidle_use_deepest_state - Enable/disable the "deepest idle" mode.
70 * @enable: Whether enable or disable the feature.
71 *
72 * If the "deepest idle" mode is enabled, cpuidle will ignore the governor and
73 * always use the state with the greatest exit latency (out of the states that
74 * are not disabled).
75 *
76 * This function can only be called after cpuidle_pause() to avoid races.
77 */
78void cpuidle_use_deepest_state(bool enable)
79{
80	use_deepest_state = enable;
81}
82
83/**
84 * cpuidle_find_deepest_state - Find the state of the greatest exit latency.
85 * @drv: cpuidle driver for a given CPU.
86 * @dev: cpuidle device for a given CPU.
87 */
88static int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
89				      struct cpuidle_device *dev)
90{
91	unsigned int latency_req = 0;
92	int i, ret = CPUIDLE_DRIVER_STATE_START - 1;
93
94	for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
95		struct cpuidle_state *s = &drv->states[i];
96		struct cpuidle_state_usage *su = &dev->states_usage[i];
97
98		if (s->disabled || su->disable || s->exit_latency <= latency_req)
99			continue;
100
101		latency_req = s->exit_latency;
102		ret = i;
103	}
104	return ret;
105}
106
107/**
108 * cpuidle_enter_state - enter the state and update stats
109 * @dev: cpuidle device for this cpu
110 * @drv: cpuidle driver for this cpu
111 * @next_state: index into drv->states of the state to enter
112 */
113int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
114			int index)
115{
116	int entered_state;
117
118	struct cpuidle_state *target_state = &drv->states[index];
119	ktime_t time_start, time_end;
120	s64 diff;
121
122	trace_cpu_idle_rcuidle(index, dev->cpu);
123	time_start = ktime_get();
124
125	entered_state = target_state->enter(dev, drv, index);
126
127	time_end = ktime_get();
128	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
129
130	if (!cpuidle_state_is_coupled(dev, drv, entered_state))
131		local_irq_enable();
132
133	diff = ktime_to_us(ktime_sub(time_end, time_start));
134	if (diff > INT_MAX)
135		diff = INT_MAX;
136
137	dev->last_residency = (int) diff;
138
139	if (entered_state >= 0) {
140		/* Update cpuidle counters */
141		/* This can be moved to within driver enter routine
142		 * but that results in multiple copies of same code.
143		 */
144		dev->states_usage[entered_state].time += dev->last_residency;
145		dev->states_usage[entered_state].usage++;
146	} else {
147		dev->last_residency = 0;
148	}
149
150	return entered_state;
151}
152
153/**
154 * cpuidle_select - ask the cpuidle framework to choose an idle state
155 *
156 * @drv: the cpuidle driver
157 * @dev: the cpuidle device
158 *
159 * Returns the index of the idle state.
160 */
161int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
162{
163	if (off || !initialized)
164		return -ENODEV;
165
166	if (!drv || !dev || !dev->enabled)
167		return -EBUSY;
168
169	if (unlikely(use_deepest_state))
170		return cpuidle_find_deepest_state(drv, dev);
171
172	return cpuidle_curr_governor->select(drv, dev);
173}
174
175/**
176 * cpuidle_enter - enter into the specified idle state
177 *
178 * @drv:   the cpuidle driver tied with the cpu
179 * @dev:   the cpuidle device
180 * @index: the index in the idle state table
181 *
182 * Returns the index in the idle state, < 0 in case of error.
183 * The error code depends on the backend driver
184 */
185int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
186		  int index)
187{
188	if (cpuidle_state_is_coupled(dev, drv, index))
189		return cpuidle_enter_state_coupled(dev, drv, index);
190	return cpuidle_enter_state(dev, drv, index);
191}
192
193/**
194 * cpuidle_reflect - tell the underlying governor what was the state
195 * we were in
196 *
197 * @dev  : the cpuidle device
198 * @index: the index in the idle state table
199 *
200 */
201void cpuidle_reflect(struct cpuidle_device *dev, int index)
202{
203	if (cpuidle_curr_governor->reflect && !unlikely(use_deepest_state))
204		cpuidle_curr_governor->reflect(dev, index);
205}
206
207/**
208 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
209 */
210void cpuidle_install_idle_handler(void)
211{
212	if (enabled_devices) {
213		/* Make sure all changes finished before we switch to new idle */
214		smp_wmb();
215		initialized = 1;
216	}
217}
218
219/**
220 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
221 */
222void cpuidle_uninstall_idle_handler(void)
223{
224	if (enabled_devices) {
225		initialized = 0;
226		wake_up_all_idle_cpus();
227	}
228
229	/*
230	 * Make sure external observers (such as the scheduler)
231	 * are done looking at pointed idle states.
232	 */
233	synchronize_rcu();
234}
235
236/**
237 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
238 */
239void cpuidle_pause_and_lock(void)
240{
241	mutex_lock(&cpuidle_lock);
242	cpuidle_uninstall_idle_handler();
243}
244
245EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
246
247/**
248 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
249 */
250void cpuidle_resume_and_unlock(void)
251{
252	cpuidle_install_idle_handler();
253	mutex_unlock(&cpuidle_lock);
254}
255
256EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
257
258/* Currently used in suspend/resume path to suspend cpuidle */
259void cpuidle_pause(void)
260{
261	mutex_lock(&cpuidle_lock);
262	cpuidle_uninstall_idle_handler();
263	mutex_unlock(&cpuidle_lock);
264}
265
266/* Currently used in suspend/resume path to resume cpuidle */
267void cpuidle_resume(void)
268{
269	mutex_lock(&cpuidle_lock);
270	cpuidle_install_idle_handler();
271	mutex_unlock(&cpuidle_lock);
272}
273
274/**
275 * cpuidle_enable_device - enables idle PM for a CPU
276 * @dev: the CPU
277 *
278 * This function must be called between cpuidle_pause_and_lock and
279 * cpuidle_resume_and_unlock when used externally.
280 */
281int cpuidle_enable_device(struct cpuidle_device *dev)
282{
283	int ret;
284	struct cpuidle_driver *drv;
285
286	if (!dev)
287		return -EINVAL;
288
289	if (dev->enabled)
290		return 0;
291
292	drv = cpuidle_get_cpu_driver(dev);
293
294	if (!drv || !cpuidle_curr_governor)
295		return -EIO;
296
297	if (!dev->registered)
298		return -EINVAL;
299
300	if (!dev->state_count)
301		dev->state_count = drv->state_count;
302
303	ret = cpuidle_add_device_sysfs(dev);
304	if (ret)
305		return ret;
306
307	if (cpuidle_curr_governor->enable &&
308	    (ret = cpuidle_curr_governor->enable(drv, dev)))
309		goto fail_sysfs;
310
311	smp_wmb();
312
313	dev->enabled = 1;
314
315	enabled_devices++;
316	return 0;
317
318fail_sysfs:
319	cpuidle_remove_device_sysfs(dev);
320
321	return ret;
322}
323
324EXPORT_SYMBOL_GPL(cpuidle_enable_device);
325
326/**
327 * cpuidle_disable_device - disables idle PM for a CPU
328 * @dev: the CPU
329 *
330 * This function must be called between cpuidle_pause_and_lock and
331 * cpuidle_resume_and_unlock when used externally.
332 */
333void cpuidle_disable_device(struct cpuidle_device *dev)
334{
335	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
336
337	if (!dev || !dev->enabled)
338		return;
339
340	if (!drv || !cpuidle_curr_governor)
341		return;
342
343	dev->enabled = 0;
344
345	if (cpuidle_curr_governor->disable)
346		cpuidle_curr_governor->disable(drv, dev);
347
348	cpuidle_remove_device_sysfs(dev);
349	enabled_devices--;
350}
351
352EXPORT_SYMBOL_GPL(cpuidle_disable_device);
353
354static void __cpuidle_unregister_device(struct cpuidle_device *dev)
355{
356	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
357
358	list_del(&dev->device_list);
359	per_cpu(cpuidle_devices, dev->cpu) = NULL;
360	module_put(drv->owner);
361}
362
363static void __cpuidle_device_init(struct cpuidle_device *dev)
364{
365	memset(dev->states_usage, 0, sizeof(dev->states_usage));
366	dev->last_residency = 0;
367}
368
369/**
370 * __cpuidle_register_device - internal register function called before register
371 * and enable routines
372 * @dev: the cpu
373 *
374 * cpuidle_lock mutex must be held before this is called
375 */
376static int __cpuidle_register_device(struct cpuidle_device *dev)
377{
378	int ret;
379	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
380
381	if (!try_module_get(drv->owner))
382		return -EINVAL;
383
384	per_cpu(cpuidle_devices, dev->cpu) = dev;
385	list_add(&dev->device_list, &cpuidle_detected_devices);
386
387	ret = cpuidle_coupled_register_device(dev);
388	if (ret)
389		__cpuidle_unregister_device(dev);
390	else
391		dev->registered = 1;
392
393	return ret;
394}
395
396/**
397 * cpuidle_register_device - registers a CPU's idle PM feature
398 * @dev: the cpu
399 */
400int cpuidle_register_device(struct cpuidle_device *dev)
401{
402	int ret = -EBUSY;
403
404	if (!dev)
405		return -EINVAL;
406
407	mutex_lock(&cpuidle_lock);
408
409	if (dev->registered)
410		goto out_unlock;
411
412	__cpuidle_device_init(dev);
413
414	ret = __cpuidle_register_device(dev);
415	if (ret)
416		goto out_unlock;
417
418	ret = cpuidle_add_sysfs(dev);
419	if (ret)
420		goto out_unregister;
421
422	ret = cpuidle_enable_device(dev);
423	if (ret)
424		goto out_sysfs;
425
426	cpuidle_install_idle_handler();
427
428out_unlock:
429	mutex_unlock(&cpuidle_lock);
430
431	return ret;
432
433out_sysfs:
434	cpuidle_remove_sysfs(dev);
435out_unregister:
436	__cpuidle_unregister_device(dev);
437	goto out_unlock;
438}
439
440EXPORT_SYMBOL_GPL(cpuidle_register_device);
441
442/**
443 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
444 * @dev: the cpu
445 */
446void cpuidle_unregister_device(struct cpuidle_device *dev)
447{
448	if (!dev || dev->registered == 0)
449		return;
450
451	cpuidle_pause_and_lock();
452
453	cpuidle_disable_device(dev);
454
455	cpuidle_remove_sysfs(dev);
456
457	__cpuidle_unregister_device(dev);
458
459	cpuidle_coupled_unregister_device(dev);
460
461	cpuidle_resume_and_unlock();
462}
463
464EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
465
466/**
467 * cpuidle_unregister: unregister a driver and the devices. This function
468 * can be used only if the driver has been previously registered through
469 * the cpuidle_register function.
470 *
471 * @drv: a valid pointer to a struct cpuidle_driver
472 */
473void cpuidle_unregister(struct cpuidle_driver *drv)
474{
475	int cpu;
476	struct cpuidle_device *device;
477
478	for_each_cpu(cpu, drv->cpumask) {
479		device = &per_cpu(cpuidle_dev, cpu);
480		cpuidle_unregister_device(device);
481	}
482
483	cpuidle_unregister_driver(drv);
484}
485EXPORT_SYMBOL_GPL(cpuidle_unregister);
486
487/**
488 * cpuidle_register: registers the driver and the cpu devices with the
489 * coupled_cpus passed as parameter. This function is used for all common
490 * initialization pattern there are in the arch specific drivers. The
491 * devices is globally defined in this file.
492 *
493 * @drv         : a valid pointer to a struct cpuidle_driver
494 * @coupled_cpus: a cpumask for the coupled states
495 *
496 * Returns 0 on success, < 0 otherwise
497 */
498int cpuidle_register(struct cpuidle_driver *drv,
499		     const struct cpumask *const coupled_cpus)
500{
501	int ret, cpu;
502	struct cpuidle_device *device;
503
504	ret = cpuidle_register_driver(drv);
505	if (ret) {
506		pr_err("failed to register cpuidle driver\n");
507		return ret;
508	}
509
510	for_each_cpu(cpu, drv->cpumask) {
511		device = &per_cpu(cpuidle_dev, cpu);
512		device->cpu = cpu;
513
514#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
515		/*
516		 * On multiplatform for ARM, the coupled idle states could be
517		 * enabled in the kernel even if the cpuidle driver does not
518		 * use it. Note, coupled_cpus is a struct copy.
519		 */
520		if (coupled_cpus)
521			device->coupled_cpus = *coupled_cpus;
522#endif
523		ret = cpuidle_register_device(device);
524		if (!ret)
525			continue;
526
527		pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
528
529		cpuidle_unregister(drv);
530		break;
531	}
532
533	return ret;
534}
535EXPORT_SYMBOL_GPL(cpuidle_register);
536
537#ifdef CONFIG_SMP
538
539/*
540 * This function gets called when a part of the kernel has a new latency
541 * requirement.  This means we need to get all processors out of their C-state,
542 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
543 * wakes them all right up.
544 */
545static int cpuidle_latency_notify(struct notifier_block *b,
546		unsigned long l, void *v)
547{
548	wake_up_all_idle_cpus();
549	return NOTIFY_OK;
550}
551
552static struct notifier_block cpuidle_latency_notifier = {
553	.notifier_call = cpuidle_latency_notify,
554};
555
556static inline void latency_notifier_init(struct notifier_block *n)
557{
558	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
559}
560
561#else /* CONFIG_SMP */
562
563#define latency_notifier_init(x) do { } while (0)
564
565#endif /* CONFIG_SMP */
566
567/**
568 * cpuidle_init - core initializer
569 */
570static int __init cpuidle_init(void)
571{
572	int ret;
573
574	if (cpuidle_disabled())
575		return -ENODEV;
576
577	ret = cpuidle_add_interface(cpu_subsys.dev_root);
578	if (ret)
579		return ret;
580
581	latency_notifier_init(&cpuidle_latency_notifier);
582
583	return 0;
584}
585
586module_param(off, int, 0444);
587core_initcall(cpuidle_init);
588