blob: ecf18feff16b35fed28ec1e3b0dc4b2768583b05 [file] [log] [blame]
rjw1f884582022-01-06 17:20:42 +08001/*
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/sched/clock.h>
16#include <linux/notifier.h>
17#include <linux/pm_qos.h>
18#include <linux/cpu.h>
19#include <linux/cpuidle.h>
20#include <linux/ktime.h>
21#include <linux/hrtimer.h>
22#include <linux/module.h>
23#include <linux/suspend.h>
24#include <linux/tick.h>
25#include <trace/events/power.h>
26
27#include "cpuidle.h"
28
29DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
30DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
31
32DEFINE_MUTEX(cpuidle_lock);
33LIST_HEAD(cpuidle_detected_devices);
34
35static int enabled_devices;
36static int off __read_mostly;
37static int initialized __read_mostly;
38
39int cpuidle_disabled(void)
40{
41 return off;
42}
43void disable_cpuidle(void)
44{
45 off = 1;
46}
47
48bool cpuidle_not_available(struct cpuidle_driver *drv,
49 struct cpuidle_device *dev)
50{
51 return off || !initialized || !drv || !dev || !dev->enabled;
52}
53
54/**
55 * cpuidle_play_dead - cpu off-lining
56 *
57 * Returns in case of an error or no driver
58 */
59int cpuidle_play_dead(void)
60{
61 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
62 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
63 int i;
64
65 if (!drv)
66 return -ENODEV;
67
68 /* Find lowest-power state that supports long-term idle */
69 for (i = drv->state_count - 1; i >= 0; i--)
70 if (drv->states[i].enter_dead)
71 return drv->states[i].enter_dead(dev, i);
72
73 return -ENODEV;
74}
75
76static int find_deepest_state(struct cpuidle_driver *drv,
77 struct cpuidle_device *dev,
78 unsigned int max_latency,
79 unsigned int forbidden_flags,
80 bool s2idle)
81{
82 unsigned int latency_req = 0;
83 int i, ret = 0;
84
85 for (i = 1; i < drv->state_count; i++) {
86 struct cpuidle_state *s = &drv->states[i];
87 struct cpuidle_state_usage *su = &dev->states_usage[i];
88
89 if (s->disabled || su->disable || s->exit_latency <= latency_req
90 || s->exit_latency > max_latency
91 || (s->flags & forbidden_flags)
92 || (s2idle && !s->enter_s2idle))
93 continue;
94
95 latency_req = s->exit_latency;
96 ret = i;
97 }
98 return ret;
99}
100
101/**
102 * cpuidle_use_deepest_state - Set/clear governor override flag.
103 * @enable: New value of the flag.
104 *
105 * Set/unset the current CPU to use the deepest idle state (override governors
106 * going forward if set).
107 */
108void cpuidle_use_deepest_state(bool enable)
109{
110 struct cpuidle_device *dev;
111
112 preempt_disable();
113 dev = cpuidle_get_device();
114 if (dev)
115 dev->use_deepest_state = enable;
116 preempt_enable();
117}
118
119/**
120 * cpuidle_find_deepest_state - Find the deepest available idle state.
121 * @drv: cpuidle driver for the given CPU.
122 * @dev: cpuidle device for the given CPU.
123 */
124int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
125 struct cpuidle_device *dev)
126{
127 return find_deepest_state(drv, dev, UINT_MAX, 0, false);
128}
129
130#ifdef CONFIG_SUSPEND
131static void enter_s2idle_proper(struct cpuidle_driver *drv,
132 struct cpuidle_device *dev, int index)
133{
134 /*
135 * trace_suspend_resume() called by tick_freeze() for the last CPU
136 * executing it contains RCU usage regarded as invalid in the idle
137 * context, so tell RCU about that.
138 */
139 RCU_NONIDLE(tick_freeze());
140 /*
141 * The state used here cannot be a "coupled" one, because the "coupled"
142 * cpuidle mechanism enables interrupts and doing that with timekeeping
143 * suspended is generally unsafe.
144 */
145 stop_critical_timings();
146 drv->states[index].enter_s2idle(dev, drv, index);
147 if (WARN_ON_ONCE(!irqs_disabled()))
148 local_irq_disable();
149 /*
150 * timekeeping_resume() that will be called by tick_unfreeze() for the
151 * first CPU executing it calls functions containing RCU read-side
152 * critical sections, so tell RCU about that.
153 */
154 RCU_NONIDLE(tick_unfreeze());
155 start_critical_timings();
156}
157
158/**
159 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
160 * @drv: cpuidle driver for the given CPU.
161 * @dev: cpuidle device for the given CPU.
162 *
163 * If there are states with the ->enter_s2idle callback, find the deepest of
164 * them and enter it with frozen tick.
165 */
166int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
167{
168 int index;
169
170 /*
171 * Find the deepest state with ->enter_s2idle present, which guarantees
172 * that interrupts won't be enabled when it exits and allows the tick to
173 * be frozen safely.
174 */
175 index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
176 if (index > 0)
177 enter_s2idle_proper(drv, dev, index);
178
179 return index;
180}
181#endif /* CONFIG_SUSPEND */
182
183/**
184 * cpuidle_enter_state - enter the state and update stats
185 * @dev: cpuidle device for this cpu
186 * @drv: cpuidle driver for this cpu
187 * @index: index into the states table in @drv of the state to enter
188 */
189int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
190 int index)
191{
192 int entered_state;
193
194 struct cpuidle_state *target_state = &drv->states[index];
195 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
196 ktime_t time_start, time_end;
197 s64 diff;
198
199 /*
200 * Tell the time framework to switch to a broadcast timer because our
201 * local timer will be shut down. If a local timer is used from another
202 * CPU as a broadcast timer, this call may fail if it is not available.
203 */
204 if (broadcast && tick_broadcast_enter()) {
205 index = find_deepest_state(drv, dev, target_state->exit_latency,
206 CPUIDLE_FLAG_TIMER_STOP, false);
207 if (index < 0) {
208 default_idle_call();
209 return -EBUSY;
210 }
211 target_state = &drv->states[index];
212 broadcast = false;
213 }
214
215 /* Take note of the planned idle state. */
216 sched_idle_set_state(target_state, index);
217
218 trace_cpu_idle_rcuidle(index, dev->cpu);
219 time_start = ns_to_ktime(local_clock());
220
221 stop_critical_timings();
222 entered_state = target_state->enter(dev, drv, index);
223 start_critical_timings();
224
225 sched_clock_idle_wakeup_event();
226 time_end = ns_to_ktime(local_clock());
227 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
228
229 /* The cpu is no longer idle or about to enter idle. */
230 sched_idle_set_state(NULL, -1);
231
232 if (broadcast) {
233 if (WARN_ON_ONCE(!irqs_disabled()))
234 local_irq_disable();
235
236 tick_broadcast_exit();
237 }
238
239 if (!cpuidle_state_is_coupled(drv, index))
240 local_irq_enable();
241
242 diff = ktime_us_delta(time_end, time_start);
243 if (diff > INT_MAX)
244 diff = INT_MAX;
245
246 dev->last_residency = (int) diff;
247
248 if (entered_state >= 0) {
249 /* Update cpuidle counters */
250 /* This can be moved to within driver enter routine
251 * but that results in multiple copies of same code.
252 */
253 dev->states_usage[entered_state].time += dev->last_residency;
254 dev->states_usage[entered_state].usage++;
255 } else {
256 dev->last_residency = 0;
257 }
258
259 return entered_state;
260}
261
262/**
263 * cpuidle_select - ask the cpuidle framework to choose an idle state
264 *
265 * @drv: the cpuidle driver
266 * @dev: the cpuidle device
267 * @stop_tick: indication on whether or not to stop the tick
268 *
269 * Returns the index of the idle state. The return value must not be negative.
270 *
271 * The memory location pointed to by @stop_tick is expected to be written the
272 * 'false' boolean value if the scheduler tick should not be stopped before
273 * entering the returned state.
274 */
275int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
276 bool *stop_tick)
277{
278 return cpuidle_curr_governor->select(drv, dev, stop_tick);
279}
280
281/**
282 * cpuidle_enter - enter into the specified idle state
283 *
284 * @drv: the cpuidle driver tied with the cpu
285 * @dev: the cpuidle device
286 * @index: the index in the idle state table
287 *
288 * Returns the index in the idle state, < 0 in case of error.
289 * The error code depends on the backend driver
290 */
291int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
292 int index)
293{
294 if (cpuidle_state_is_coupled(drv, index))
295 return cpuidle_enter_state_coupled(dev, drv, index);
296 return cpuidle_enter_state(dev, drv, index);
297}
298
299/**
300 * cpuidle_reflect - tell the underlying governor what was the state
301 * we were in
302 *
303 * @dev : the cpuidle device
304 * @index: the index in the idle state table
305 *
306 */
307void cpuidle_reflect(struct cpuidle_device *dev, int index)
308{
309 if (cpuidle_curr_governor->reflect && index >= 0)
310 cpuidle_curr_governor->reflect(dev, index);
311}
312
313/**
314 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
315 */
316void cpuidle_install_idle_handler(void)
317{
318 if (enabled_devices) {
319 /* Make sure all changes finished before we switch to new idle */
320 smp_wmb();
321 initialized = 1;
322 }
323}
324
325/**
326 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
327 */
328void cpuidle_uninstall_idle_handler(void)
329{
330 if (enabled_devices) {
331 initialized = 0;
332 wake_up_all_idle_cpus();
333 }
334
335 /*
336 * Make sure external observers (such as the scheduler)
337 * are done looking at pointed idle states.
338 */
339 synchronize_rcu();
340}
341
342/**
343 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
344 */
345void cpuidle_pause_and_lock(void)
346{
347 mutex_lock(&cpuidle_lock);
348 cpuidle_uninstall_idle_handler();
349}
350
351EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
352
353/**
354 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
355 */
356void cpuidle_resume_and_unlock(void)
357{
358 cpuidle_install_idle_handler();
359 mutex_unlock(&cpuidle_lock);
360}
361
362EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
363
364/* Currently used in suspend/resume path to suspend cpuidle */
365void cpuidle_pause(void)
366{
367 mutex_lock(&cpuidle_lock);
368 cpuidle_uninstall_idle_handler();
369 mutex_unlock(&cpuidle_lock);
370}
371
372/* Currently used in suspend/resume path to resume cpuidle */
373void cpuidle_resume(void)
374{
375 mutex_lock(&cpuidle_lock);
376 cpuidle_install_idle_handler();
377 mutex_unlock(&cpuidle_lock);
378}
379
380/**
381 * cpuidle_enable_device - enables idle PM for a CPU
382 * @dev: the CPU
383 *
384 * This function must be called between cpuidle_pause_and_lock and
385 * cpuidle_resume_and_unlock when used externally.
386 */
387int cpuidle_enable_device(struct cpuidle_device *dev)
388{
389 int ret;
390 struct cpuidle_driver *drv;
391
392 if (!dev)
393 return -EINVAL;
394
395 if (dev->enabled)
396 return 0;
397
398 drv = cpuidle_get_cpu_driver(dev);
399
400 if (!drv || !cpuidle_curr_governor)
401 return -EIO;
402
403 if (!dev->registered)
404 return -EINVAL;
405
406 ret = cpuidle_add_device_sysfs(dev);
407 if (ret)
408 return ret;
409
410 if (cpuidle_curr_governor->enable &&
411 (ret = cpuidle_curr_governor->enable(drv, dev)))
412 goto fail_sysfs;
413
414 smp_wmb();
415
416 dev->enabled = 1;
417
418 enabled_devices++;
419 return 0;
420
421fail_sysfs:
422 cpuidle_remove_device_sysfs(dev);
423
424 return ret;
425}
426
427EXPORT_SYMBOL_GPL(cpuidle_enable_device);
428
429/**
430 * cpuidle_disable_device - disables idle PM for a CPU
431 * @dev: the CPU
432 *
433 * This function must be called between cpuidle_pause_and_lock and
434 * cpuidle_resume_and_unlock when used externally.
435 */
436void cpuidle_disable_device(struct cpuidle_device *dev)
437{
438 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
439
440 if (!dev || !dev->enabled)
441 return;
442
443 if (!drv || !cpuidle_curr_governor)
444 return;
445
446 dev->enabled = 0;
447
448 if (cpuidle_curr_governor->disable)
449 cpuidle_curr_governor->disable(drv, dev);
450
451 cpuidle_remove_device_sysfs(dev);
452 enabled_devices--;
453}
454
455EXPORT_SYMBOL_GPL(cpuidle_disable_device);
456
457static void __cpuidle_unregister_device(struct cpuidle_device *dev)
458{
459 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
460
461 list_del(&dev->device_list);
462 per_cpu(cpuidle_devices, dev->cpu) = NULL;
463 module_put(drv->owner);
464
465 dev->registered = 0;
466}
467
468static void __cpuidle_device_init(struct cpuidle_device *dev)
469{
470 memset(dev->states_usage, 0, sizeof(dev->states_usage));
471 dev->last_residency = 0;
472}
473
474/**
475 * __cpuidle_register_device - internal register function called before register
476 * and enable routines
477 * @dev: the cpu
478 *
479 * cpuidle_lock mutex must be held before this is called
480 */
481static int __cpuidle_register_device(struct cpuidle_device *dev)
482{
483 int ret;
484 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
485
486 if (!try_module_get(drv->owner))
487 return -EINVAL;
488
489 per_cpu(cpuidle_devices, dev->cpu) = dev;
490 list_add(&dev->device_list, &cpuidle_detected_devices);
491
492 ret = cpuidle_coupled_register_device(dev);
493 if (ret)
494 __cpuidle_unregister_device(dev);
495 else
496 dev->registered = 1;
497
498 return ret;
499}
500
501/**
502 * cpuidle_register_device - registers a CPU's idle PM feature
503 * @dev: the cpu
504 */
505int cpuidle_register_device(struct cpuidle_device *dev)
506{
507 int ret = -EBUSY;
508
509 if (!dev)
510 return -EINVAL;
511
512 mutex_lock(&cpuidle_lock);
513
514 if (dev->registered)
515 goto out_unlock;
516
517 __cpuidle_device_init(dev);
518
519 ret = __cpuidle_register_device(dev);
520 if (ret)
521 goto out_unlock;
522
523 ret = cpuidle_add_sysfs(dev);
524 if (ret)
525 goto out_unregister;
526
527 ret = cpuidle_enable_device(dev);
528 if (ret)
529 goto out_sysfs;
530
531 cpuidle_install_idle_handler();
532
533out_unlock:
534 mutex_unlock(&cpuidle_lock);
535
536 return ret;
537
538out_sysfs:
539 cpuidle_remove_sysfs(dev);
540out_unregister:
541 __cpuidle_unregister_device(dev);
542 goto out_unlock;
543}
544
545EXPORT_SYMBOL_GPL(cpuidle_register_device);
546
547/**
548 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
549 * @dev: the cpu
550 */
551void cpuidle_unregister_device(struct cpuidle_device *dev)
552{
553 if (!dev || dev->registered == 0)
554 return;
555
556 cpuidle_pause_and_lock();
557
558 cpuidle_disable_device(dev);
559
560 cpuidle_remove_sysfs(dev);
561
562 __cpuidle_unregister_device(dev);
563
564 cpuidle_coupled_unregister_device(dev);
565
566 cpuidle_resume_and_unlock();
567}
568
569EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
570
571/**
572 * cpuidle_unregister: unregister a driver and the devices. This function
573 * can be used only if the driver has been previously registered through
574 * the cpuidle_register function.
575 *
576 * @drv: a valid pointer to a struct cpuidle_driver
577 */
578void cpuidle_unregister(struct cpuidle_driver *drv)
579{
580 int cpu;
581 struct cpuidle_device *device;
582
583 for_each_cpu(cpu, drv->cpumask) {
584 device = &per_cpu(cpuidle_dev, cpu);
585 cpuidle_unregister_device(device);
586 }
587
588 cpuidle_unregister_driver(drv);
589}
590EXPORT_SYMBOL_GPL(cpuidle_unregister);
591
592/**
593 * cpuidle_register: registers the driver and the cpu devices with the
594 * coupled_cpus passed as parameter. This function is used for all common
595 * initialization pattern there are in the arch specific drivers. The
596 * devices is globally defined in this file.
597 *
598 * @drv : a valid pointer to a struct cpuidle_driver
599 * @coupled_cpus: a cpumask for the coupled states
600 *
601 * Returns 0 on success, < 0 otherwise
602 */
603int cpuidle_register(struct cpuidle_driver *drv,
604 const struct cpumask *const coupled_cpus)
605{
606 int ret, cpu;
607 struct cpuidle_device *device;
608
609 ret = cpuidle_register_driver(drv);
610 if (ret) {
611 pr_err("failed to register cpuidle driver\n");
612 return ret;
613 }
614
615 for_each_cpu(cpu, drv->cpumask) {
616 device = &per_cpu(cpuidle_dev, cpu);
617 device->cpu = cpu;
618
619#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
620 /*
621 * On multiplatform for ARM, the coupled idle states could be
622 * enabled in the kernel even if the cpuidle driver does not
623 * use it. Note, coupled_cpus is a struct copy.
624 */
625 if (coupled_cpus)
626 device->coupled_cpus = *coupled_cpus;
627#endif
628 ret = cpuidle_register_device(device);
629 if (!ret)
630 continue;
631
632 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
633
634 cpuidle_unregister(drv);
635 break;
636 }
637
638 return ret;
639}
640EXPORT_SYMBOL_GPL(cpuidle_register);
641
642#ifdef CONFIG_SMP
643
644/*
645 * This function gets called when a part of the kernel has a new latency
646 * requirement. This means we need to get all processors out of their C-state,
647 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
648 * wakes them all right up.
649 */
650static int cpuidle_latency_notify(struct notifier_block *b,
651 unsigned long l, void *v)
652{
653 wake_up_all_idle_cpus();
654 return NOTIFY_OK;
655}
656
657static struct notifier_block cpuidle_latency_notifier = {
658 .notifier_call = cpuidle_latency_notify,
659};
660
661static inline void latency_notifier_init(struct notifier_block *n)
662{
663 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
664}
665
666#else /* CONFIG_SMP */
667
668#define latency_notifier_init(x) do { } while (0)
669
670#endif /* CONFIG_SMP */
671
672/**
673 * cpuidle_init - core initializer
674 */
675static int __init cpuidle_init(void)
676{
677 int ret;
678
679 if (cpuidle_disabled())
680 return -ENODEV;
681
682 ret = cpuidle_add_interface(cpu_subsys.dev_root);
683 if (ret)
684 return ret;
685
686 latency_notifier_init(&cpuidle_latency_notifier);
687
688 return 0;
689}
690
691module_param(off, int, 0444);
692core_initcall(cpuidle_init);