]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/cpuidle/cpuidle.c
Merge branch 'drm-next-4.7' of git://people.freedesktop.org/~agd5f/linux into drm...
[mirror_ubuntu-bionic-kernel.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 = 0;
83
84 for (i = 1; 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 u64 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 = local_clock();
199
200 stop_critical_timings();
201 entered_state = target_state->enter(dev, drv, index);
202 start_critical_timings();
203
204 time_end = local_clock();
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(drv, entered_state))
218 local_irq_enable();
219
220 /*
221 * local_clock() returns the time in nanosecond, let's shift
222 * by 10 (divide by 1024) to have microsecond based time.
223 */
224 diff = (time_end - time_start) >> 10;
225 if (diff > INT_MAX)
226 diff = INT_MAX;
227
228 dev->last_residency = (int) diff;
229
230 if (entered_state >= 0) {
231 /* Update cpuidle counters */
232 /* This can be moved to within driver enter routine
233 * but that results in multiple copies of same code.
234 */
235 dev->states_usage[entered_state].time += dev->last_residency;
236 dev->states_usage[entered_state].usage++;
237 } else {
238 dev->last_residency = 0;
239 }
240
241 return entered_state;
242 }
243
244 /**
245 * cpuidle_select - ask the cpuidle framework to choose an idle state
246 *
247 * @drv: the cpuidle driver
248 * @dev: the cpuidle device
249 *
250 * Returns the index of the idle state. The return value must not be negative.
251 */
252 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
253 {
254 return cpuidle_curr_governor->select(drv, dev);
255 }
256
257 /**
258 * cpuidle_enter - enter into the specified idle state
259 *
260 * @drv: the cpuidle driver tied with the cpu
261 * @dev: the cpuidle device
262 * @index: the index in the idle state table
263 *
264 * Returns the index in the idle state, < 0 in case of error.
265 * The error code depends on the backend driver
266 */
267 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
268 int index)
269 {
270 if (cpuidle_state_is_coupled(drv, index))
271 return cpuidle_enter_state_coupled(dev, drv, index);
272 return cpuidle_enter_state(dev, drv, index);
273 }
274
275 /**
276 * cpuidle_reflect - tell the underlying governor what was the state
277 * we were in
278 *
279 * @dev : the cpuidle device
280 * @index: the index in the idle state table
281 *
282 */
283 void cpuidle_reflect(struct cpuidle_device *dev, int index)
284 {
285 if (cpuidle_curr_governor->reflect && index >= 0)
286 cpuidle_curr_governor->reflect(dev, index);
287 }
288
289 /**
290 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
291 */
292 void cpuidle_install_idle_handler(void)
293 {
294 if (enabled_devices) {
295 /* Make sure all changes finished before we switch to new idle */
296 smp_wmb();
297 initialized = 1;
298 }
299 }
300
301 /**
302 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
303 */
304 void cpuidle_uninstall_idle_handler(void)
305 {
306 if (enabled_devices) {
307 initialized = 0;
308 wake_up_all_idle_cpus();
309 }
310
311 /*
312 * Make sure external observers (such as the scheduler)
313 * are done looking at pointed idle states.
314 */
315 synchronize_rcu();
316 }
317
318 /**
319 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
320 */
321 void cpuidle_pause_and_lock(void)
322 {
323 mutex_lock(&cpuidle_lock);
324 cpuidle_uninstall_idle_handler();
325 }
326
327 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
328
329 /**
330 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
331 */
332 void cpuidle_resume_and_unlock(void)
333 {
334 cpuidle_install_idle_handler();
335 mutex_unlock(&cpuidle_lock);
336 }
337
338 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
339
340 /* Currently used in suspend/resume path to suspend cpuidle */
341 void cpuidle_pause(void)
342 {
343 mutex_lock(&cpuidle_lock);
344 cpuidle_uninstall_idle_handler();
345 mutex_unlock(&cpuidle_lock);
346 }
347
348 /* Currently used in suspend/resume path to resume cpuidle */
349 void cpuidle_resume(void)
350 {
351 mutex_lock(&cpuidle_lock);
352 cpuidle_install_idle_handler();
353 mutex_unlock(&cpuidle_lock);
354 }
355
356 /**
357 * cpuidle_enable_device - enables idle PM for a CPU
358 * @dev: the CPU
359 *
360 * This function must be called between cpuidle_pause_and_lock and
361 * cpuidle_resume_and_unlock when used externally.
362 */
363 int cpuidle_enable_device(struct cpuidle_device *dev)
364 {
365 int ret;
366 struct cpuidle_driver *drv;
367
368 if (!dev)
369 return -EINVAL;
370
371 if (dev->enabled)
372 return 0;
373
374 drv = cpuidle_get_cpu_driver(dev);
375
376 if (!drv || !cpuidle_curr_governor)
377 return -EIO;
378
379 if (!dev->registered)
380 return -EINVAL;
381
382 ret = cpuidle_add_device_sysfs(dev);
383 if (ret)
384 return ret;
385
386 if (cpuidle_curr_governor->enable &&
387 (ret = cpuidle_curr_governor->enable(drv, dev)))
388 goto fail_sysfs;
389
390 smp_wmb();
391
392 dev->enabled = 1;
393
394 enabled_devices++;
395 return 0;
396
397 fail_sysfs:
398 cpuidle_remove_device_sysfs(dev);
399
400 return ret;
401 }
402
403 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
404
405 /**
406 * cpuidle_disable_device - disables idle PM for a CPU
407 * @dev: the CPU
408 *
409 * This function must be called between cpuidle_pause_and_lock and
410 * cpuidle_resume_and_unlock when used externally.
411 */
412 void cpuidle_disable_device(struct cpuidle_device *dev)
413 {
414 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
415
416 if (!dev || !dev->enabled)
417 return;
418
419 if (!drv || !cpuidle_curr_governor)
420 return;
421
422 dev->enabled = 0;
423
424 if (cpuidle_curr_governor->disable)
425 cpuidle_curr_governor->disable(drv, dev);
426
427 cpuidle_remove_device_sysfs(dev);
428 enabled_devices--;
429 }
430
431 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
432
433 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
434 {
435 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
436
437 list_del(&dev->device_list);
438 per_cpu(cpuidle_devices, dev->cpu) = NULL;
439 module_put(drv->owner);
440
441 dev->registered = 0;
442 }
443
444 static void __cpuidle_device_init(struct cpuidle_device *dev)
445 {
446 memset(dev->states_usage, 0, sizeof(dev->states_usage));
447 dev->last_residency = 0;
448 }
449
450 /**
451 * __cpuidle_register_device - internal register function called before register
452 * and enable routines
453 * @dev: the cpu
454 *
455 * cpuidle_lock mutex must be held before this is called
456 */
457 static int __cpuidle_register_device(struct cpuidle_device *dev)
458 {
459 int ret;
460 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
461
462 if (!try_module_get(drv->owner))
463 return -EINVAL;
464
465 per_cpu(cpuidle_devices, dev->cpu) = dev;
466 list_add(&dev->device_list, &cpuidle_detected_devices);
467
468 ret = cpuidle_coupled_register_device(dev);
469 if (ret)
470 __cpuidle_unregister_device(dev);
471 else
472 dev->registered = 1;
473
474 return ret;
475 }
476
477 /**
478 * cpuidle_register_device - registers a CPU's idle PM feature
479 * @dev: the cpu
480 */
481 int cpuidle_register_device(struct cpuidle_device *dev)
482 {
483 int ret = -EBUSY;
484
485 if (!dev)
486 return -EINVAL;
487
488 mutex_lock(&cpuidle_lock);
489
490 if (dev->registered)
491 goto out_unlock;
492
493 __cpuidle_device_init(dev);
494
495 ret = __cpuidle_register_device(dev);
496 if (ret)
497 goto out_unlock;
498
499 ret = cpuidle_add_sysfs(dev);
500 if (ret)
501 goto out_unregister;
502
503 ret = cpuidle_enable_device(dev);
504 if (ret)
505 goto out_sysfs;
506
507 cpuidle_install_idle_handler();
508
509 out_unlock:
510 mutex_unlock(&cpuidle_lock);
511
512 return ret;
513
514 out_sysfs:
515 cpuidle_remove_sysfs(dev);
516 out_unregister:
517 __cpuidle_unregister_device(dev);
518 goto out_unlock;
519 }
520
521 EXPORT_SYMBOL_GPL(cpuidle_register_device);
522
523 /**
524 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
525 * @dev: the cpu
526 */
527 void cpuidle_unregister_device(struct cpuidle_device *dev)
528 {
529 if (!dev || dev->registered == 0)
530 return;
531
532 cpuidle_pause_and_lock();
533
534 cpuidle_disable_device(dev);
535
536 cpuidle_remove_sysfs(dev);
537
538 __cpuidle_unregister_device(dev);
539
540 cpuidle_coupled_unregister_device(dev);
541
542 cpuidle_resume_and_unlock();
543 }
544
545 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
546
547 /**
548 * cpuidle_unregister: unregister a driver and the devices. This function
549 * can be used only if the driver has been previously registered through
550 * the cpuidle_register function.
551 *
552 * @drv: a valid pointer to a struct cpuidle_driver
553 */
554 void cpuidle_unregister(struct cpuidle_driver *drv)
555 {
556 int cpu;
557 struct cpuidle_device *device;
558
559 for_each_cpu(cpu, drv->cpumask) {
560 device = &per_cpu(cpuidle_dev, cpu);
561 cpuidle_unregister_device(device);
562 }
563
564 cpuidle_unregister_driver(drv);
565 }
566 EXPORT_SYMBOL_GPL(cpuidle_unregister);
567
568 /**
569 * cpuidle_register: registers the driver and the cpu devices with the
570 * coupled_cpus passed as parameter. This function is used for all common
571 * initialization pattern there are in the arch specific drivers. The
572 * devices is globally defined in this file.
573 *
574 * @drv : a valid pointer to a struct cpuidle_driver
575 * @coupled_cpus: a cpumask for the coupled states
576 *
577 * Returns 0 on success, < 0 otherwise
578 */
579 int cpuidle_register(struct cpuidle_driver *drv,
580 const struct cpumask *const coupled_cpus)
581 {
582 int ret, cpu;
583 struct cpuidle_device *device;
584
585 ret = cpuidle_register_driver(drv);
586 if (ret) {
587 pr_err("failed to register cpuidle driver\n");
588 return ret;
589 }
590
591 for_each_cpu(cpu, drv->cpumask) {
592 device = &per_cpu(cpuidle_dev, cpu);
593 device->cpu = cpu;
594
595 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
596 /*
597 * On multiplatform for ARM, the coupled idle states could be
598 * enabled in the kernel even if the cpuidle driver does not
599 * use it. Note, coupled_cpus is a struct copy.
600 */
601 if (coupled_cpus)
602 device->coupled_cpus = *coupled_cpus;
603 #endif
604 ret = cpuidle_register_device(device);
605 if (!ret)
606 continue;
607
608 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
609
610 cpuidle_unregister(drv);
611 break;
612 }
613
614 return ret;
615 }
616 EXPORT_SYMBOL_GPL(cpuidle_register);
617
618 #ifdef CONFIG_SMP
619
620 /*
621 * This function gets called when a part of the kernel has a new latency
622 * requirement. This means we need to get all processors out of their C-state,
623 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
624 * wakes them all right up.
625 */
626 static int cpuidle_latency_notify(struct notifier_block *b,
627 unsigned long l, void *v)
628 {
629 wake_up_all_idle_cpus();
630 return NOTIFY_OK;
631 }
632
633 static struct notifier_block cpuidle_latency_notifier = {
634 .notifier_call = cpuidle_latency_notify,
635 };
636
637 static inline void latency_notifier_init(struct notifier_block *n)
638 {
639 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
640 }
641
642 #else /* CONFIG_SMP */
643
644 #define latency_notifier_init(x) do { } while (0)
645
646 #endif /* CONFIG_SMP */
647
648 /**
649 * cpuidle_init - core initializer
650 */
651 static int __init cpuidle_init(void)
652 {
653 int ret;
654
655 if (cpuidle_disabled())
656 return -ENODEV;
657
658 ret = cpuidle_add_interface(cpu_subsys.dev_root);
659 if (ret)
660 return ret;
661
662 latency_notifier_init(&cpuidle_latency_notifier);
663
664 return 0;
665 }
666
667 module_param(off, int, 0444);
668 core_initcall(cpuidle_init);