]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - drivers/cpuidle/cpuidle.c
sched/deadline: Fix inter- exclusive cpusets migrations
[mirror_ubuntu-jammy-kernel.git] / drivers / cpuidle / cpuidle.c
CommitLineData
4f86d3a8
LB
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
b60e6a0e 11#include <linux/clockchips.h>
4f86d3a8
LB
12#include <linux/kernel.h>
13#include <linux/mutex.h>
14#include <linux/sched.h>
15#include <linux/notifier.h>
e8db0be1 16#include <linux/pm_qos.h>
4f86d3a8
LB
17#include <linux/cpu.h>
18#include <linux/cpuidle.h>
9a0b8415 19#include <linux/ktime.h>
2e94d1f7 20#include <linux/hrtimer.h>
884b17e1 21#include <linux/module.h>
288f023e 22#include <trace/events/power.h>
4f86d3a8
LB
23
24#include "cpuidle.h"
25
26DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
4c637b21 27DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
4f86d3a8
LB
28
29DEFINE_MUTEX(cpuidle_lock);
30LIST_HEAD(cpuidle_detected_devices);
4f86d3a8
LB
31
32static int enabled_devices;
62027aea 33static int off __read_mostly;
a0bfa137 34static int initialized __read_mostly;
a6220fc1 35static bool use_deepest_state __read_mostly;
62027aea
LB
36
37int cpuidle_disabled(void)
38{
39 return off;
40}
d91ee586
LB
41void disable_cpuidle(void)
42{
43 off = 1;
44}
4f86d3a8 45
1a022e3f
BO
46/**
47 * cpuidle_play_dead - cpu off-lining
48 *
ee01e663 49 * Returns in case of an error or no driver
1a022e3f
BO
50 */
51int cpuidle_play_dead(void)
52{
53 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
bf4d1b5d 54 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
8aef33a7 55 int i;
1a022e3f 56
ee01e663
TK
57 if (!drv)
58 return -ENODEV;
59
1a022e3f 60 /* Find lowest-power state that supports long-term idle */
8aef33a7
DL
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);
1a022e3f
BO
64
65 return -ENODEV;
66}
67
a6220fc1
RW
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
56cfbf74
CC
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,
554c06ba 114 int index)
56cfbf74
CC
115{
116 int entered_state;
117
554c06ba
DL
118 struct cpuidle_state *target_state = &drv->states[index];
119 ktime_t time_start, time_end;
120 s64 diff;
121
30fe6884 122 trace_cpu_idle_rcuidle(index, dev->cpu);
554c06ba
DL
123 time_start = ktime_get();
124
125 entered_state = target_state->enter(dev, drv, index);
126
127 time_end = ktime_get();
30fe6884 128 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
554c06ba 129
0b89e9aa
PB
130 if (!cpuidle_state_is_coupled(dev, drv, entered_state))
131 local_irq_enable();
554c06ba
DL
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;
56cfbf74
CC
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 */
a474a515 144 dev->states_usage[entered_state].time += dev->last_residency;
56cfbf74
CC
145 dev->states_usage[entered_state].usage++;
146 } else {
147 dev->last_residency = 0;
148 }
149
150 return entered_state;
151}
152
4f86d3a8 153/**
907e30f1
DL
154 * cpuidle_select - ask the cpuidle framework to choose an idle state
155 *
156 * @drv: the cpuidle driver
157 * @dev: the cpuidle device
4f86d3a8 158 *
907e30f1 159 * Returns the index of the idle state.
4f86d3a8 160 */
907e30f1 161int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
4f86d3a8 162{
52c324f8
RW
163 if (off || !initialized)
164 return -ENODEV;
165
166 if (!drv || !dev || !dev->enabled)
167 return -EBUSY;
168
a6220fc1
RW
169 if (unlikely(use_deepest_state))
170 return cpuidle_find_deepest_state(drv, dev);
171
907e30f1
DL
172 return cpuidle_curr_governor->select(drv, dev);
173}
ba8f20c2 174
907e30f1
DL
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}
b60e6a0e 192
907e30f1
DL
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{
a6220fc1 203 if (cpuidle_curr_governor->reflect && !unlikely(use_deepest_state))
907e30f1 204 cpuidle_curr_governor->reflect(dev, index);
4f86d3a8
LB
205}
206
207/**
208 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
209 */
210void cpuidle_install_idle_handler(void)
211{
a0bfa137 212 if (enabled_devices) {
4f86d3a8
LB
213 /* Make sure all changes finished before we switch to new idle */
214 smp_wmb();
a0bfa137 215 initialized = 1;
4f86d3a8
LB
216 }
217}
218
219/**
220 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
221 */
222void cpuidle_uninstall_idle_handler(void)
223{
a0bfa137
LB
224 if (enabled_devices) {
225 initialized = 0;
2ed903c5 226 wake_up_all_idle_cpus();
4f86d3a8
LB
227 }
228}
229
230/**
231 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
232 */
233void cpuidle_pause_and_lock(void)
234{
235 mutex_lock(&cpuidle_lock);
236 cpuidle_uninstall_idle_handler();
237}
238
239EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
240
241/**
242 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
243 */
244void cpuidle_resume_and_unlock(void)
245{
246 cpuidle_install_idle_handler();
247 mutex_unlock(&cpuidle_lock);
248}
249
250EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
251
8651f97b
PM
252/* Currently used in suspend/resume path to suspend cpuidle */
253void cpuidle_pause(void)
254{
255 mutex_lock(&cpuidle_lock);
256 cpuidle_uninstall_idle_handler();
257 mutex_unlock(&cpuidle_lock);
258}
259
260/* Currently used in suspend/resume path to resume cpuidle */
261void cpuidle_resume(void)
262{
263 mutex_lock(&cpuidle_lock);
264 cpuidle_install_idle_handler();
265 mutex_unlock(&cpuidle_lock);
266}
267
4f86d3a8
LB
268/**
269 * cpuidle_enable_device - enables idle PM for a CPU
270 * @dev: the CPU
271 *
272 * This function must be called between cpuidle_pause_and_lock and
273 * cpuidle_resume_and_unlock when used externally.
274 */
275int cpuidle_enable_device(struct cpuidle_device *dev)
276{
5df0aa73 277 int ret;
bf4d1b5d 278 struct cpuidle_driver *drv;
4f86d3a8 279
1b0a0e9a
SB
280 if (!dev)
281 return -EINVAL;
282
4f86d3a8
LB
283 if (dev->enabled)
284 return 0;
bf4d1b5d
DL
285
286 drv = cpuidle_get_cpu_driver(dev);
287
e1689795 288 if (!drv || !cpuidle_curr_governor)
4f86d3a8 289 return -EIO;
bf4d1b5d 290
10b9d3f8
DL
291 if (!dev->registered)
292 return -EINVAL;
293
4f86d3a8 294 if (!dev->state_count)
fc850f39 295 dev->state_count = drv->state_count;
4f86d3a8 296
bf4d1b5d
DL
297 ret = cpuidle_add_device_sysfs(dev);
298 if (ret)
4f86d3a8
LB
299 return ret;
300
301 if (cpuidle_curr_governor->enable &&
e1689795 302 (ret = cpuidle_curr_governor->enable(drv, dev)))
4f86d3a8
LB
303 goto fail_sysfs;
304
4f86d3a8
LB
305 smp_wmb();
306
307 dev->enabled = 1;
308
309 enabled_devices++;
310 return 0;
311
312fail_sysfs:
bf4d1b5d 313 cpuidle_remove_device_sysfs(dev);
4f86d3a8
LB
314
315 return ret;
316}
317
318EXPORT_SYMBOL_GPL(cpuidle_enable_device);
319
320/**
321 * cpuidle_disable_device - disables idle PM for a CPU
322 * @dev: the CPU
323 *
324 * This function must be called between cpuidle_pause_and_lock and
325 * cpuidle_resume_and_unlock when used externally.
326 */
327void cpuidle_disable_device(struct cpuidle_device *dev)
328{
bf4d1b5d
DL
329 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
330
cf31cd1a 331 if (!dev || !dev->enabled)
4f86d3a8 332 return;
bf4d1b5d
DL
333
334 if (!drv || !cpuidle_curr_governor)
4f86d3a8
LB
335 return;
336
337 dev->enabled = 0;
338
339 if (cpuidle_curr_governor->disable)
bf4d1b5d 340 cpuidle_curr_governor->disable(drv, dev);
4f86d3a8 341
bf4d1b5d 342 cpuidle_remove_device_sysfs(dev);
4f86d3a8
LB
343 enabled_devices--;
344}
345
346EXPORT_SYMBOL_GPL(cpuidle_disable_device);
347
f6bb51a5
DL
348static void __cpuidle_unregister_device(struct cpuidle_device *dev)
349{
350 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
351
352 list_del(&dev->device_list);
353 per_cpu(cpuidle_devices, dev->cpu) = NULL;
354 module_put(drv->owner);
355}
356
267d4bf8 357static void __cpuidle_device_init(struct cpuidle_device *dev)
5df0aa73
DL
358{
359 memset(dev->states_usage, 0, sizeof(dev->states_usage));
360 dev->last_residency = 0;
5df0aa73
DL
361}
362
4f86d3a8 363/**
dcb84f33
VP
364 * __cpuidle_register_device - internal register function called before register
365 * and enable routines
4f86d3a8 366 * @dev: the cpu
dcb84f33
VP
367 *
368 * cpuidle_lock mutex must be held before this is called
4f86d3a8 369 */
dcb84f33 370static int __cpuidle_register_device(struct cpuidle_device *dev)
4f86d3a8
LB
371{
372 int ret;
bf4d1b5d 373 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
4f86d3a8 374
bf4d1b5d 375 if (!try_module_get(drv->owner))
4f86d3a8
LB
376 return -EINVAL;
377
4f86d3a8
LB
378 per_cpu(cpuidle_devices, dev->cpu) = dev;
379 list_add(&dev->device_list, &cpuidle_detected_devices);
4f86d3a8 380
4126c019 381 ret = cpuidle_coupled_register_device(dev);
47182668 382 if (ret)
f6bb51a5 383 __cpuidle_unregister_device(dev);
47182668
VK
384 else
385 dev->registered = 1;
4f86d3a8 386
47182668 387 return ret;
dcb84f33
VP
388}
389
390/**
391 * cpuidle_register_device - registers a CPU's idle PM feature
392 * @dev: the cpu
393 */
394int cpuidle_register_device(struct cpuidle_device *dev)
395{
c878a52d 396 int ret = -EBUSY;
dcb84f33 397
1b0a0e9a
SB
398 if (!dev)
399 return -EINVAL;
400
dcb84f33
VP
401 mutex_lock(&cpuidle_lock);
402
c878a52d
DL
403 if (dev->registered)
404 goto out_unlock;
405
267d4bf8 406 __cpuidle_device_init(dev);
5df0aa73 407
f6bb51a5
DL
408 ret = __cpuidle_register_device(dev);
409 if (ret)
410 goto out_unlock;
411
412 ret = cpuidle_add_sysfs(dev);
413 if (ret)
414 goto out_unregister;
dcb84f33 415
10b9d3f8 416 ret = cpuidle_enable_device(dev);
f6bb51a5
DL
417 if (ret)
418 goto out_sysfs;
10b9d3f8 419
4f86d3a8
LB
420 cpuidle_install_idle_handler();
421
f6bb51a5 422out_unlock:
4f86d3a8
LB
423 mutex_unlock(&cpuidle_lock);
424
f6bb51a5
DL
425 return ret;
426
427out_sysfs:
428 cpuidle_remove_sysfs(dev);
429out_unregister:
430 __cpuidle_unregister_device(dev);
431 goto out_unlock;
4f86d3a8
LB
432}
433
434EXPORT_SYMBOL_GPL(cpuidle_register_device);
435
436/**
437 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
438 * @dev: the cpu
439 */
440void cpuidle_unregister_device(struct cpuidle_device *dev)
441{
813e8e3d 442 if (!dev || dev->registered == 0)
dcb84f33
VP
443 return;
444
4f86d3a8
LB
445 cpuidle_pause_and_lock();
446
447 cpuidle_disable_device(dev);
448
1aef40e2 449 cpuidle_remove_sysfs(dev);
f6bb51a5
DL
450
451 __cpuidle_unregister_device(dev);
4f86d3a8 452
4126c019
CC
453 cpuidle_coupled_unregister_device(dev);
454
4f86d3a8 455 cpuidle_resume_and_unlock();
4f86d3a8
LB
456}
457
458EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
459
1c192d04 460/**
4c637b21
DL
461 * cpuidle_unregister: unregister a driver and the devices. This function
462 * can be used only if the driver has been previously registered through
463 * the cpuidle_register function.
464 *
465 * @drv: a valid pointer to a struct cpuidle_driver
466 */
467void cpuidle_unregister(struct cpuidle_driver *drv)
468{
469 int cpu;
470 struct cpuidle_device *device;
471
82467a5a 472 for_each_cpu(cpu, drv->cpumask) {
4c637b21
DL
473 device = &per_cpu(cpuidle_dev, cpu);
474 cpuidle_unregister_device(device);
475 }
476
477 cpuidle_unregister_driver(drv);
478}
479EXPORT_SYMBOL_GPL(cpuidle_unregister);
480
481/**
482 * cpuidle_register: registers the driver and the cpu devices with the
483 * coupled_cpus passed as parameter. This function is used for all common
484 * initialization pattern there are in the arch specific drivers. The
485 * devices is globally defined in this file.
486 *
487 * @drv : a valid pointer to a struct cpuidle_driver
488 * @coupled_cpus: a cpumask for the coupled states
489 *
490 * Returns 0 on success, < 0 otherwise
491 */
492int cpuidle_register(struct cpuidle_driver *drv,
493 const struct cpumask *const coupled_cpus)
494{
495 int ret, cpu;
496 struct cpuidle_device *device;
497
498 ret = cpuidle_register_driver(drv);
499 if (ret) {
500 pr_err("failed to register cpuidle driver\n");
501 return ret;
502 }
503
82467a5a 504 for_each_cpu(cpu, drv->cpumask) {
4c637b21
DL
505 device = &per_cpu(cpuidle_dev, cpu);
506 device->cpu = cpu;
507
508#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
509 /*
caf4a36e 510 * On multiplatform for ARM, the coupled idle states could be
4c637b21
DL
511 * enabled in the kernel even if the cpuidle driver does not
512 * use it. Note, coupled_cpus is a struct copy.
513 */
514 if (coupled_cpus)
515 device->coupled_cpus = *coupled_cpus;
516#endif
517 ret = cpuidle_register_device(device);
518 if (!ret)
519 continue;
520
521 pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
522
523 cpuidle_unregister(drv);
524 break;
525 }
526
527 return ret;
528}
529EXPORT_SYMBOL_GPL(cpuidle_register);
530
4f86d3a8
LB
531#ifdef CONFIG_SMP
532
4f86d3a8
LB
533/*
534 * This function gets called when a part of the kernel has a new latency
535 * requirement. This means we need to get all processors out of their C-state,
536 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
537 * wakes them all right up.
538 */
539static int cpuidle_latency_notify(struct notifier_block *b,
540 unsigned long l, void *v)
541{
2ed903c5 542 wake_up_all_idle_cpus();
4f86d3a8
LB
543 return NOTIFY_OK;
544}
545
546static struct notifier_block cpuidle_latency_notifier = {
547 .notifier_call = cpuidle_latency_notify,
548};
549
d82b3518
MG
550static inline void latency_notifier_init(struct notifier_block *n)
551{
552 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
553}
4f86d3a8
LB
554
555#else /* CONFIG_SMP */
556
557#define latency_notifier_init(x) do { } while (0)
558
559#endif /* CONFIG_SMP */
560
561/**
562 * cpuidle_init - core initializer
563 */
564static int __init cpuidle_init(void)
565{
566 int ret;
567
62027aea
LB
568 if (cpuidle_disabled())
569 return -ENODEV;
570
8a25a2fd 571 ret = cpuidle_add_interface(cpu_subsys.dev_root);
4f86d3a8
LB
572 if (ret)
573 return ret;
574
575 latency_notifier_init(&cpuidle_latency_notifier);
576
577 return 0;
578}
579
62027aea 580module_param(off, int, 0444);
4f86d3a8 581core_initcall(cpuidle_init);