xref: /openbmc/linux/drivers/cpuidle/cpuidle.c (revision 7d51d979)
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 /**
101  * cpuidle_find_deepest_state - Find the deepest available idle state.
102  * @drv: cpuidle driver for the given CPU.
103  * @dev: cpuidle device for the given CPU.
104  */
105 int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
106 			       struct cpuidle_device *dev)
107 {
108 	return find_deepest_state(drv, dev, UINT_MAX, 0, false);
109 }
110 
111 static void enter_freeze_proper(struct cpuidle_driver *drv,
112 				struct cpuidle_device *dev, int index)
113 {
114 	tick_freeze();
115 	/*
116 	 * The state used here cannot be a "coupled" one, because the "coupled"
117 	 * cpuidle mechanism enables interrupts and doing that with timekeeping
118 	 * suspended is generally unsafe.
119 	 */
120 	drv->states[index].enter_freeze(dev, drv, index);
121 	WARN_ON(!irqs_disabled());
122 	/*
123 	 * timekeeping_resume() that will be called by tick_unfreeze() for the
124 	 * last CPU executing it calls functions containing RCU read-side
125 	 * critical sections, so tell RCU about that.
126 	 */
127 	RCU_NONIDLE(tick_unfreeze());
128 }
129 
130 /**
131  * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle.
132  * @drv: cpuidle driver for the given CPU.
133  * @dev: cpuidle device for the given CPU.
134  *
135  * If there are states with the ->enter_freeze callback, find the deepest of
136  * them and enter it with frozen tick.
137  */
138 int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
139 {
140 	int index;
141 
142 	/*
143 	 * Find the deepest state with ->enter_freeze present, which guarantees
144 	 * that interrupts won't be enabled when it exits and allows the tick to
145 	 * be frozen safely.
146 	 */
147 	index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
148 	if (index >= 0)
149 		enter_freeze_proper(drv, dev, index);
150 
151 	return index;
152 }
153 
154 /**
155  * cpuidle_enter_state - enter the state and update stats
156  * @dev: cpuidle device for this cpu
157  * @drv: cpuidle driver for this cpu
158  * @index: index into the states table in @drv of the state to enter
159  */
160 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
161 			int index)
162 {
163 	int entered_state;
164 
165 	struct cpuidle_state *target_state = &drv->states[index];
166 	bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
167 	ktime_t time_start, time_end;
168 	s64 diff;
169 
170 	/*
171 	 * Tell the time framework to switch to a broadcast timer because our
172 	 * local timer will be shut down.  If a local timer is used from another
173 	 * CPU as a broadcast timer, this call may fail if it is not available.
174 	 */
175 	if (broadcast && tick_broadcast_enter()) {
176 		index = find_deepest_state(drv, dev, target_state->exit_latency,
177 					   CPUIDLE_FLAG_TIMER_STOP, false);
178 		if (index < 0) {
179 			default_idle_call();
180 			return -EBUSY;
181 		}
182 		target_state = &drv->states[index];
183 	}
184 
185 	/* Take note of the planned idle state. */
186 	sched_idle_set_state(target_state);
187 
188 	trace_cpu_idle_rcuidle(index, dev->cpu);
189 	time_start = ktime_get();
190 
191 	entered_state = target_state->enter(dev, drv, index);
192 
193 	time_end = ktime_get();
194 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
195 
196 	/* The cpu is no longer idle or about to enter idle. */
197 	sched_idle_set_state(NULL);
198 
199 	if (broadcast) {
200 		if (WARN_ON_ONCE(!irqs_disabled()))
201 			local_irq_disable();
202 
203 		tick_broadcast_exit();
204 	}
205 
206 	if (!cpuidle_state_is_coupled(dev, drv, entered_state))
207 		local_irq_enable();
208 
209 	diff = ktime_to_us(ktime_sub(time_end, time_start));
210 	if (diff > INT_MAX)
211 		diff = INT_MAX;
212 
213 	dev->last_residency = (int) diff;
214 
215 	if (entered_state >= 0) {
216 		/* Update cpuidle counters */
217 		/* This can be moved to within driver enter routine
218 		 * but that results in multiple copies of same code.
219 		 */
220 		dev->states_usage[entered_state].time += dev->last_residency;
221 		dev->states_usage[entered_state].usage++;
222 	} else {
223 		dev->last_residency = 0;
224 	}
225 
226 	return entered_state;
227 }
228 
229 /**
230  * cpuidle_select - ask the cpuidle framework to choose an idle state
231  *
232  * @drv: the cpuidle driver
233  * @dev: the cpuidle device
234  *
235  * Returns the index of the idle state.
236  */
237 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
238 {
239 	return cpuidle_curr_governor->select(drv, dev);
240 }
241 
242 /**
243  * cpuidle_enter - enter into the specified idle state
244  *
245  * @drv:   the cpuidle driver tied with the cpu
246  * @dev:   the cpuidle device
247  * @index: the index in the idle state table
248  *
249  * Returns the index in the idle state, < 0 in case of error.
250  * The error code depends on the backend driver
251  */
252 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
253 		  int index)
254 {
255 	if (cpuidle_state_is_coupled(dev, drv, index))
256 		return cpuidle_enter_state_coupled(dev, drv, index);
257 	return cpuidle_enter_state(dev, drv, index);
258 }
259 
260 /**
261  * cpuidle_reflect - tell the underlying governor what was the state
262  * we were in
263  *
264  * @dev  : the cpuidle device
265  * @index: the index in the idle state table
266  *
267  */
268 void cpuidle_reflect(struct cpuidle_device *dev, int index)
269 {
270 	if (cpuidle_curr_governor->reflect && index >= 0)
271 		cpuidle_curr_governor->reflect(dev, index);
272 }
273 
274 /**
275  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
276  */
277 void cpuidle_install_idle_handler(void)
278 {
279 	if (enabled_devices) {
280 		/* Make sure all changes finished before we switch to new idle */
281 		smp_wmb();
282 		initialized = 1;
283 	}
284 }
285 
286 /**
287  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
288  */
289 void cpuidle_uninstall_idle_handler(void)
290 {
291 	if (enabled_devices) {
292 		initialized = 0;
293 		wake_up_all_idle_cpus();
294 	}
295 
296 	/*
297 	 * Make sure external observers (such as the scheduler)
298 	 * are done looking at pointed idle states.
299 	 */
300 	synchronize_rcu();
301 }
302 
303 /**
304  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
305  */
306 void cpuidle_pause_and_lock(void)
307 {
308 	mutex_lock(&cpuidle_lock);
309 	cpuidle_uninstall_idle_handler();
310 }
311 
312 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
313 
314 /**
315  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
316  */
317 void cpuidle_resume_and_unlock(void)
318 {
319 	cpuidle_install_idle_handler();
320 	mutex_unlock(&cpuidle_lock);
321 }
322 
323 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
324 
325 /* Currently used in suspend/resume path to suspend cpuidle */
326 void cpuidle_pause(void)
327 {
328 	mutex_lock(&cpuidle_lock);
329 	cpuidle_uninstall_idle_handler();
330 	mutex_unlock(&cpuidle_lock);
331 }
332 
333 /* Currently used in suspend/resume path to resume cpuidle */
334 void cpuidle_resume(void)
335 {
336 	mutex_lock(&cpuidle_lock);
337 	cpuidle_install_idle_handler();
338 	mutex_unlock(&cpuidle_lock);
339 }
340 
341 /**
342  * cpuidle_enable_device - enables idle PM for a CPU
343  * @dev: the CPU
344  *
345  * This function must be called between cpuidle_pause_and_lock and
346  * cpuidle_resume_and_unlock when used externally.
347  */
348 int cpuidle_enable_device(struct cpuidle_device *dev)
349 {
350 	int ret;
351 	struct cpuidle_driver *drv;
352 
353 	if (!dev)
354 		return -EINVAL;
355 
356 	if (dev->enabled)
357 		return 0;
358 
359 	drv = cpuidle_get_cpu_driver(dev);
360 
361 	if (!drv || !cpuidle_curr_governor)
362 		return -EIO;
363 
364 	if (!dev->registered)
365 		return -EINVAL;
366 
367 	ret = cpuidle_add_device_sysfs(dev);
368 	if (ret)
369 		return ret;
370 
371 	if (cpuidle_curr_governor->enable &&
372 	    (ret = cpuidle_curr_governor->enable(drv, dev)))
373 		goto fail_sysfs;
374 
375 	smp_wmb();
376 
377 	dev->enabled = 1;
378 
379 	enabled_devices++;
380 	return 0;
381 
382 fail_sysfs:
383 	cpuidle_remove_device_sysfs(dev);
384 
385 	return ret;
386 }
387 
388 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
389 
390 /**
391  * cpuidle_disable_device - disables idle PM for a CPU
392  * @dev: the CPU
393  *
394  * This function must be called between cpuidle_pause_and_lock and
395  * cpuidle_resume_and_unlock when used externally.
396  */
397 void cpuidle_disable_device(struct cpuidle_device *dev)
398 {
399 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
400 
401 	if (!dev || !dev->enabled)
402 		return;
403 
404 	if (!drv || !cpuidle_curr_governor)
405 		return;
406 
407 	dev->enabled = 0;
408 
409 	if (cpuidle_curr_governor->disable)
410 		cpuidle_curr_governor->disable(drv, dev);
411 
412 	cpuidle_remove_device_sysfs(dev);
413 	enabled_devices--;
414 }
415 
416 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
417 
418 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
419 {
420 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
421 
422 	list_del(&dev->device_list);
423 	per_cpu(cpuidle_devices, dev->cpu) = NULL;
424 	module_put(drv->owner);
425 }
426 
427 static void __cpuidle_device_init(struct cpuidle_device *dev)
428 {
429 	memset(dev->states_usage, 0, sizeof(dev->states_usage));
430 	dev->last_residency = 0;
431 }
432 
433 /**
434  * __cpuidle_register_device - internal register function called before register
435  * and enable routines
436  * @dev: the cpu
437  *
438  * cpuidle_lock mutex must be held before this is called
439  */
440 static int __cpuidle_register_device(struct cpuidle_device *dev)
441 {
442 	int ret;
443 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
444 
445 	if (!try_module_get(drv->owner))
446 		return -EINVAL;
447 
448 	per_cpu(cpuidle_devices, dev->cpu) = dev;
449 	list_add(&dev->device_list, &cpuidle_detected_devices);
450 
451 	ret = cpuidle_coupled_register_device(dev);
452 	if (ret)
453 		__cpuidle_unregister_device(dev);
454 	else
455 		dev->registered = 1;
456 
457 	return ret;
458 }
459 
460 /**
461  * cpuidle_register_device - registers a CPU's idle PM feature
462  * @dev: the cpu
463  */
464 int cpuidle_register_device(struct cpuidle_device *dev)
465 {
466 	int ret = -EBUSY;
467 
468 	if (!dev)
469 		return -EINVAL;
470 
471 	mutex_lock(&cpuidle_lock);
472 
473 	if (dev->registered)
474 		goto out_unlock;
475 
476 	__cpuidle_device_init(dev);
477 
478 	ret = __cpuidle_register_device(dev);
479 	if (ret)
480 		goto out_unlock;
481 
482 	ret = cpuidle_add_sysfs(dev);
483 	if (ret)
484 		goto out_unregister;
485 
486 	ret = cpuidle_enable_device(dev);
487 	if (ret)
488 		goto out_sysfs;
489 
490 	cpuidle_install_idle_handler();
491 
492 out_unlock:
493 	mutex_unlock(&cpuidle_lock);
494 
495 	return ret;
496 
497 out_sysfs:
498 	cpuidle_remove_sysfs(dev);
499 out_unregister:
500 	__cpuidle_unregister_device(dev);
501 	goto out_unlock;
502 }
503 
504 EXPORT_SYMBOL_GPL(cpuidle_register_device);
505 
506 /**
507  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
508  * @dev: the cpu
509  */
510 void cpuidle_unregister_device(struct cpuidle_device *dev)
511 {
512 	if (!dev || dev->registered == 0)
513 		return;
514 
515 	cpuidle_pause_and_lock();
516 
517 	cpuidle_disable_device(dev);
518 
519 	cpuidle_remove_sysfs(dev);
520 
521 	__cpuidle_unregister_device(dev);
522 
523 	cpuidle_coupled_unregister_device(dev);
524 
525 	cpuidle_resume_and_unlock();
526 }
527 
528 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
529 
530 /**
531  * cpuidle_unregister: unregister a driver and the devices. This function
532  * can be used only if the driver has been previously registered through
533  * the cpuidle_register function.
534  *
535  * @drv: a valid pointer to a struct cpuidle_driver
536  */
537 void cpuidle_unregister(struct cpuidle_driver *drv)
538 {
539 	int cpu;
540 	struct cpuidle_device *device;
541 
542 	for_each_cpu(cpu, drv->cpumask) {
543 		device = &per_cpu(cpuidle_dev, cpu);
544 		cpuidle_unregister_device(device);
545 	}
546 
547 	cpuidle_unregister_driver(drv);
548 }
549 EXPORT_SYMBOL_GPL(cpuidle_unregister);
550 
551 /**
552  * cpuidle_register: registers the driver and the cpu devices with the
553  * coupled_cpus passed as parameter. This function is used for all common
554  * initialization pattern there are in the arch specific drivers. The
555  * devices is globally defined in this file.
556  *
557  * @drv         : a valid pointer to a struct cpuidle_driver
558  * @coupled_cpus: a cpumask for the coupled states
559  *
560  * Returns 0 on success, < 0 otherwise
561  */
562 int cpuidle_register(struct cpuidle_driver *drv,
563 		     const struct cpumask *const coupled_cpus)
564 {
565 	int ret, cpu;
566 	struct cpuidle_device *device;
567 
568 	ret = cpuidle_register_driver(drv);
569 	if (ret) {
570 		pr_err("failed to register cpuidle driver\n");
571 		return ret;
572 	}
573 
574 	for_each_cpu(cpu, drv->cpumask) {
575 		device = &per_cpu(cpuidle_dev, cpu);
576 		device->cpu = cpu;
577 
578 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
579 		/*
580 		 * On multiplatform for ARM, the coupled idle states could be
581 		 * enabled in the kernel even if the cpuidle driver does not
582 		 * use it. Note, coupled_cpus is a struct copy.
583 		 */
584 		if (coupled_cpus)
585 			device->coupled_cpus = *coupled_cpus;
586 #endif
587 		ret = cpuidle_register_device(device);
588 		if (!ret)
589 			continue;
590 
591 		pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
592 
593 		cpuidle_unregister(drv);
594 		break;
595 	}
596 
597 	return ret;
598 }
599 EXPORT_SYMBOL_GPL(cpuidle_register);
600 
601 #ifdef CONFIG_SMP
602 
603 /*
604  * This function gets called when a part of the kernel has a new latency
605  * requirement.  This means we need to get all processors out of their C-state,
606  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
607  * wakes them all right up.
608  */
609 static int cpuidle_latency_notify(struct notifier_block *b,
610 		unsigned long l, void *v)
611 {
612 	wake_up_all_idle_cpus();
613 	return NOTIFY_OK;
614 }
615 
616 static struct notifier_block cpuidle_latency_notifier = {
617 	.notifier_call = cpuidle_latency_notify,
618 };
619 
620 static inline void latency_notifier_init(struct notifier_block *n)
621 {
622 	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
623 }
624 
625 #else /* CONFIG_SMP */
626 
627 #define latency_notifier_init(x) do { } while (0)
628 
629 #endif /* CONFIG_SMP */
630 
631 /**
632  * cpuidle_init - core initializer
633  */
634 static int __init cpuidle_init(void)
635 {
636 	int ret;
637 
638 	if (cpuidle_disabled())
639 		return -ENODEV;
640 
641 	ret = cpuidle_add_interface(cpu_subsys.dev_root);
642 	if (ret)
643 		return ret;
644 
645 	latency_notifier_init(&cpuidle_latency_notifier);
646 
647 	return 0;
648 }
649 
650 module_param(off, int, 0444);
651 core_initcall(cpuidle_init);
652