Merge branch 'timers-nohz-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[deliverable/linux.git] / drivers / cpuidle / cpuidle.c
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 <linux/suspend.h>
23 #include <linux/tick.h>
24 #include <trace/events/power.h>
25
26 #include "cpuidle.h"
27
28 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
29 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
30
31 DEFINE_MUTEX(cpuidle_lock);
32 LIST_HEAD(cpuidle_detected_devices);
33
34 static int enabled_devices;
35 static int off __read_mostly;
36 static int initialized __read_mostly;
37
38 int cpuidle_disabled(void)
39 {
40 return off;
41 }
42 void disable_cpuidle(void)
43 {
44 off = 1;
45 }
46
47 bool cpuidle_not_available(struct cpuidle_driver *drv,
48 struct cpuidle_device *dev)
49 {
50 return off || !initialized || !drv || !dev || !dev->enabled;
51 }
52
53 /**
54 * cpuidle_play_dead - cpu off-lining
55 *
56 * Returns in case of an error or no driver
57 */
58 int cpuidle_play_dead(void)
59 {
60 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
61 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
62 int i;
63
64 if (!drv)
65 return -ENODEV;
66
67 /* Find lowest-power state that supports long-term idle */
68 for (i = drv->state_count - 1; i >= 0; i--)
69 if (drv->states[i].enter_dead)
70 return drv->states[i].enter_dead(dev, i);
71
72 return -ENODEV;
73 }
74
75 static int find_deepest_state(struct cpuidle_driver *drv,
76 struct cpuidle_device *dev,
77 unsigned int max_latency,
78 unsigned int forbidden_flags,
79 bool freeze)
80 {
81 unsigned int latency_req = 0;
82 int i, ret = -ENXIO;
83
84 for (i = 0; i < drv->state_count; i++) {
85 struct cpuidle_state *s = &drv->states[i];
86 struct cpuidle_state_usage *su = &dev->states_usage[i];
87
88 if (s->disabled || su->disable || s->exit_latency <= latency_req
89 || s->exit_latency > max_latency
90 || (s->flags & forbidden_flags)
91 || (freeze && !s->enter_freeze))
92 continue;
93
94 latency_req = s->exit_latency;
95 ret = i;
96 }
97 return ret;
98 }
99
100 #ifdef CONFIG_SUSPEND
101 /**
102 * cpuidle_find_deepest_state - Find the deepest available idle state.
103 * @drv: cpuidle driver for the given CPU.
104 * @dev: cpuidle device for the given CPU.
105 */
106 int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
107 struct cpuidle_device *dev)
108 {
109 return find_deepest_state(drv, dev, UINT_MAX, 0, false);
110 }
111
112 static void enter_freeze_proper(struct cpuidle_driver *drv,
113 struct cpuidle_device *dev, int index)
114 {
115 /*
116 * trace_suspend_resume() called by tick_freeze() for the last CPU
117 * executing it contains RCU usage regarded as invalid in the idle
118 * context, so tell RCU about that.
119 */
120 RCU_NONIDLE(tick_freeze());
121 /*
122 * The state used here cannot be a "coupled" one, because the "coupled"
123 * cpuidle mechanism enables interrupts and doing that with timekeeping
124 * suspended is generally unsafe.
125 */
126 stop_critical_timings();
127 drv->states[index].enter_freeze(dev, drv, index);
128 WARN_ON(!irqs_disabled());
129 /*
130 * timekeeping_resume() that will be called by tick_unfreeze() for the
131 * first CPU executing it calls functions containing RCU read-side
132 * critical sections, so tell RCU about that.
133 */
134 RCU_NONIDLE(tick_unfreeze());
135 start_critical_timings();
136 }
137
138 /**
139 * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle.
140 * @drv: cpuidle driver for the given CPU.
141 * @dev: cpuidle device for the given CPU.
142 *
143 * If there are states with the ->enter_freeze callback, find the deepest of
144 * them and enter it with frozen tick.
145 */
146 int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
147 {
148 int index;
149
150 /*
151 * Find the deepest state with ->enter_freeze present, which guarantees
152 * that interrupts won't be enabled when it exits and allows the tick to
153 * be frozen safely.
154 */
155 index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
156 if (index >= 0)
157 enter_freeze_proper(drv, dev, index);
158
159 return index;
160 }
161 #endif /* CONFIG_SUSPEND */
162
163 /**
164 * cpuidle_enter_state - enter the state and update stats
165 * @dev: cpuidle device for this cpu
166 * @drv: cpuidle driver for this cpu
167 * @index: index into the states table in @drv of the state to enter
168 */
169 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
170 int index)
171 {
172 int entered_state;
173
174 struct cpuidle_state *target_state = &drv->states[index];
175 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
176 ktime_t time_start, time_end;
177 s64 diff;
178
179 /*
180 * Tell the time framework to switch to a broadcast timer because our
181 * local timer will be shut down. If a local timer is used from another
182 * CPU as a broadcast timer, this call may fail if it is not available.
183 */
184 if (broadcast && tick_broadcast_enter()) {
185 index = find_deepest_state(drv, dev, target_state->exit_latency,
186 CPUIDLE_FLAG_TIMER_STOP, false);
187 if (index < 0) {
188 default_idle_call();
189 return -EBUSY;
190 }
191 target_state = &drv->states[index];
192 }
193
194 /* Take note of the planned idle state. */
195 sched_idle_set_state(target_state);
196
197 trace_cpu_idle_rcuidle(index, dev->cpu);
198 time_start = ktime_get();
199
200 stop_critical_timings();
201 entered_state = target_state->enter(dev, drv, index);
202 start_critical_timings();
203
204 time_end = ktime_get();
205 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
206
207 /* The cpu is no longer idle or about to enter idle. */
208 sched_idle_set_state(NULL);
209
210 if (broadcast) {
211 if (WARN_ON_ONCE(!irqs_disabled()))
212 local_irq_disable();
213
214 tick_broadcast_exit();
215 }
216
217 if (!cpuidle_state_is_coupled(dev, drv, entered_state))
218 local_irq_enable();
219
220 diff = ktime_to_us(ktime_sub(time_end, time_start));
221 if (diff > INT_MAX)
222 diff = INT_MAX;
223
224 dev->last_residency = (int) diff;
225
226 if (entered_state >= 0) {
227 /* Update cpuidle counters */
228 /* This can be moved to within driver enter routine
229 * but that results in multiple copies of same code.
230 */
231 dev->states_usage[entered_state].time += dev->last_residency;
232 dev->states_usage[entered_state].usage++;
233 } else {
234 dev->last_residency = 0;
235 }
236
237 return entered_state;
238 }
239
240 /**
241 * cpuidle_select - ask the cpuidle framework to choose an idle state
242 *
243 * @drv: the cpuidle driver
244 * @dev: the cpuidle device
245 *
246 * Returns the index of the idle state.
247 */
248 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
249 {
250 return cpuidle_curr_governor->select(drv, dev);
251 }
252
253 /**
254 * cpuidle_enter - enter into the specified idle state
255 *
256 * @drv: the cpuidle driver tied with the cpu
257 * @dev: the cpuidle device
258 * @index: the index in the idle state table
259 *
260 * Returns the index in the idle state, < 0 in case of error.
261 * The error code depends on the backend driver
262 */
263 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
264 int index)
265 {
266 if (cpuidle_state_is_coupled(dev, drv, index))
267 return cpuidle_enter_state_coupled(dev, drv, index);
268 return cpuidle_enter_state(dev, drv, index);
269 }
270
271 /**
272 * cpuidle_reflect - tell the underlying governor what was the state
273 * we were in
274 *
275 * @dev : the cpuidle device
276 * @index: the index in the idle state table
277 *
278 */
279 void cpuidle_reflect(struct cpuidle_device *dev, int index)
280 {
281 if (cpuidle_curr_governor->reflect && index >= 0)
282 cpuidle_curr_governor->reflect(dev, index);
283 }
284
285 /**
286 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
287 */
288 void cpuidle_install_idle_handler(void)
289 {
290 if (enabled_devices) {
291 /* Make sure all changes finished before we switch to new idle */
292 smp_wmb();
293 initialized = 1;
294 }
295 }
296
297 /**
298 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
299 */
300 void cpuidle_uninstall_idle_handler(void)
301 {
302 if (enabled_devices) {
303 initialized = 0;
304 wake_up_all_idle_cpus();
305 }
306
307 /*
308 * Make sure external observers (such as the scheduler)
309 * are done looking at pointed idle states.
310 */
311 synchronize_rcu();
312 }
313
314 /**
315 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
316 */
317 void cpuidle_pause_and_lock(void)
318 {
319 mutex_lock(&cpuidle_lock);
320 cpuidle_uninstall_idle_handler();
321 }
322
323 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
324
325 /**
326 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
327 */
328 void cpuidle_resume_and_unlock(void)
329 {
330 cpuidle_install_idle_handler();
331 mutex_unlock(&cpuidle_lock);
332 }
333
334 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
335
336 /* Currently used in suspend/resume path to suspend cpuidle */
337 void cpuidle_pause(void)
338 {
339 mutex_lock(&cpuidle_lock);
340 cpuidle_uninstall_idle_handler();
341 mutex_unlock(&cpuidle_lock);
342 }
343
344 /* Currently used in suspend/resume path to resume cpuidle */
345 void cpuidle_resume(void)
346 {
347 mutex_lock(&cpuidle_lock);
348 cpuidle_install_idle_handler();
349 mutex_unlock(&cpuidle_lock);
350 }
351
352 /**
353 * cpuidle_enable_device - enables idle PM for a CPU
354 * @dev: the CPU
355 *
356 * This function must be called between cpuidle_pause_and_lock and
357 * cpuidle_resume_and_unlock when used externally.
358 */
359 int cpuidle_enable_device(struct cpuidle_device *dev)
360 {
361 int ret;
362 struct cpuidle_driver *drv;
363
364 if (!dev)
365 return -EINVAL;
366
367 if (dev->enabled)
368 return 0;
369
370 drv = cpuidle_get_cpu_driver(dev);
371
372 if (!drv || !cpuidle_curr_governor)
373 return -EIO;
374
375 if (!dev->registered)
376 return -EINVAL;
377
378 ret = cpuidle_add_device_sysfs(dev);
379 if (ret)
380 return ret;
381
382 if (cpuidle_curr_governor->enable &&
383 (ret = cpuidle_curr_governor->enable(drv, dev)))
384 goto fail_sysfs;
385
386 smp_wmb();
387
388 dev->enabled = 1;
389
390 enabled_devices++;
391 return 0;
392
393 fail_sysfs:
394 cpuidle_remove_device_sysfs(dev);
395
396 return ret;
397 }
398
399 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
400
401 /**
402 * cpuidle_disable_device - disables idle PM for a CPU
403 * @dev: the CPU
404 *
405 * This function must be called between cpuidle_pause_and_lock and
406 * cpuidle_resume_and_unlock when used externally.
407 */
408 void cpuidle_disable_device(struct cpuidle_device *dev)
409 {
410 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
411
412 if (!dev || !dev->enabled)
413 return;
414
415 if (!drv || !cpuidle_curr_governor)
416 return;
417
418 dev->enabled = 0;
419
420 if (cpuidle_curr_governor->disable)
421 cpuidle_curr_governor->disable(drv, dev);
422
423 cpuidle_remove_device_sysfs(dev);
424 enabled_devices--;
425 }
426
427 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
428
429 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
430 {
431 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
432
433 list_del(&dev->device_list);
434 per_cpu(cpuidle_devices, dev->cpu) = NULL;
435 module_put(drv->owner);
436 }
437
438 static void __cpuidle_device_init(struct cpuidle_device *dev)
439 {
440 memset(dev->states_usage, 0, sizeof(dev->states_usage));
441 dev->last_residency = 0;
442 }
443
444 /**
445 * __cpuidle_register_device - internal register function called before register
446 * and enable routines
447 * @dev: the cpu
448 *
449 * cpuidle_lock mutex must be held before this is called
450 */
451 static int __cpuidle_register_device(struct cpuidle_device *dev)
452 {
453 int ret;
454 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
455
456 if (!try_module_get(drv->owner))
457 return -EINVAL;
458
459 per_cpu(cpuidle_devices, dev->cpu) = dev;
460 list_add(&dev->device_list, &cpuidle_detected_devices);
461
462 ret = cpuidle_coupled_register_device(dev);
463 if (ret)
464 __cpuidle_unregister_device(dev);
465 else
466 dev->registered = 1;
467
468 return ret;
469 }
470
471 /**
472 * cpuidle_register_device - registers a CPU's idle PM feature
473 * @dev: the cpu
474 */
475 int cpuidle_register_device(struct cpuidle_device *dev)
476 {
477 int ret = -EBUSY;
478
479 if (!dev)
480 return -EINVAL;
481
482 mutex_lock(&cpuidle_lock);
483
484 if (dev->registered)
485 goto out_unlock;
486
487 __cpuidle_device_init(dev);
488
489 ret = __cpuidle_register_device(dev);
490 if (ret)
491 goto out_unlock;
492
493 ret = cpuidle_add_sysfs(dev);
494 if (ret)
495 goto out_unregister;
496
497 ret = cpuidle_enable_device(dev);
498 if (ret)
499 goto out_sysfs;
500
501 cpuidle_install_idle_handler();
502
503 out_unlock:
504 mutex_unlock(&cpuidle_lock);
505
506 return ret;
507
508 out_sysfs:
509 cpuidle_remove_sysfs(dev);
510 out_unregister:
511 __cpuidle_unregister_device(dev);
512 goto out_unlock;
513 }
514
515 EXPORT_SYMBOL_GPL(cpuidle_register_device);
516
517 /**
518 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
519 * @dev: the cpu
520 */
521 void cpuidle_unregister_device(struct cpuidle_device *dev)
522 {
523 if (!dev || dev->registered == 0)
524 return;
525
526 cpuidle_pause_and_lock();
527
528 cpuidle_disable_device(dev);
529
530 cpuidle_remove_sysfs(dev);
531
532 __cpuidle_unregister_device(dev);
533
534 cpuidle_coupled_unregister_device(dev);
535
536 cpuidle_resume_and_unlock();
537 }
538
539 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
540
541 /**
542 * cpuidle_unregister: unregister a driver and the devices. This function
543 * can be used only if the driver has been previously registered through
544 * the cpuidle_register function.
545 *
546 * @drv: a valid pointer to a struct cpuidle_driver
547 */
548 void cpuidle_unregister(struct cpuidle_driver *drv)
549 {
550 int cpu;
551 struct cpuidle_device *device;
552
553 for_each_cpu(cpu, drv->cpumask) {
554 device = &per_cpu(cpuidle_dev, cpu);
555 cpuidle_unregister_device(device);
556 }
557
558 cpuidle_unregister_driver(drv);
559 }
560 EXPORT_SYMBOL_GPL(cpuidle_unregister);
561
562 /**
563 * cpuidle_register: registers the driver and the cpu devices with the
564 * coupled_cpus passed as parameter. This function is used for all common
565 * initialization pattern there are in the arch specific drivers. The
566 * devices is globally defined in this file.
567 *
568 * @drv : a valid pointer to a struct cpuidle_driver
569 * @coupled_cpus: a cpumask for the coupled states
570 *
571 * Returns 0 on success, < 0 otherwise
572 */
573 int cpuidle_register(struct cpuidle_driver *drv,
574 const struct cpumask *const coupled_cpus)
575 {
576 int ret, cpu;
577 struct cpuidle_device *device;
578
579 ret = cpuidle_register_driver(drv);
580 if (ret) {
581 pr_err("failed to register cpuidle driver\n");
582 return ret;
583 }
584
585 for_each_cpu(cpu, drv->cpumask) {
586 device = &per_cpu(cpuidle_dev, cpu);
587 device->cpu = cpu;
588
589 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
590 /*
591 * On multiplatform for ARM, the coupled idle states could be
592 * enabled in the kernel even if the cpuidle driver does not
593 * use it. Note, coupled_cpus is a struct copy.
594 */
595 if (coupled_cpus)
596 device->coupled_cpus = *coupled_cpus;
597 #endif
598 ret = cpuidle_register_device(device);
599 if (!ret)
600 continue;
601
602 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
603
604 cpuidle_unregister(drv);
605 break;
606 }
607
608 return ret;
609 }
610 EXPORT_SYMBOL_GPL(cpuidle_register);
611
612 #ifdef CONFIG_SMP
613
614 /*
615 * This function gets called when a part of the kernel has a new latency
616 * requirement. This means we need to get all processors out of their C-state,
617 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
618 * wakes them all right up.
619 */
620 static int cpuidle_latency_notify(struct notifier_block *b,
621 unsigned long l, void *v)
622 {
623 wake_up_all_idle_cpus();
624 return NOTIFY_OK;
625 }
626
627 static struct notifier_block cpuidle_latency_notifier = {
628 .notifier_call = cpuidle_latency_notify,
629 };
630
631 static inline void latency_notifier_init(struct notifier_block *n)
632 {
633 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
634 }
635
636 #else /* CONFIG_SMP */
637
638 #define latency_notifier_init(x) do { } while (0)
639
640 #endif /* CONFIG_SMP */
641
642 /**
643 * cpuidle_init - core initializer
644 */
645 static int __init cpuidle_init(void)
646 {
647 int ret;
648
649 if (cpuidle_disabled())
650 return -ENODEV;
651
652 ret = cpuidle_add_interface(cpu_subsys.dev_root);
653 if (ret)
654 return ret;
655
656 latency_notifier_init(&cpuidle_latency_notifier);
657
658 return 0;
659 }
660
661 module_param(off, int, 0444);
662 core_initcall(cpuidle_init);
This page took 0.046428 seconds and 5 git commands to generate.