Merge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
[linux-2.6-block.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
11#include <linux/kernel.h>
12#include <linux/mutex.h>
13#include <linux/sched.h>
14#include <linux/notifier.h>
e8db0be1 15#include <linux/pm_qos.h>
4f86d3a8
LB
16#include <linux/cpu.h>
17#include <linux/cpuidle.h>
9a0b8415 18#include <linux/ktime.h>
2e94d1f7 19#include <linux/hrtimer.h>
884b17e1 20#include <linux/module.h>
288f023e 21#include <trace/events/power.h>
4f86d3a8
LB
22
23#include "cpuidle.h"
24
25DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
4f86d3a8
LB
26
27DEFINE_MUTEX(cpuidle_lock);
28LIST_HEAD(cpuidle_detected_devices);
4f86d3a8
LB
29
30static int enabled_devices;
62027aea 31static int off __read_mostly;
a0bfa137 32static int initialized __read_mostly;
62027aea
LB
33
34int cpuidle_disabled(void)
35{
36 return off;
37}
d91ee586
LB
38void disable_cpuidle(void)
39{
40 off = 1;
41}
4f86d3a8 42
a6869cc4
VP
43#if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
44static void cpuidle_kick_cpus(void)
45{
46 cpu_idle_wait();
47}
48#elif defined(CONFIG_SMP)
49# error "Arch needs cpu_idle_wait() equivalent here"
50#else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
51static void cpuidle_kick_cpus(void) {}
52#endif
53
dcb84f33
VP
54static int __cpuidle_register_device(struct cpuidle_device *dev);
55
e1689795
RL
56static inline int cpuidle_enter(struct cpuidle_device *dev,
57 struct cpuidle_driver *drv, int index)
58{
59 struct cpuidle_state *target_state = &drv->states[index];
60 return target_state->enter(dev, drv, index);
61}
62
63static inline int cpuidle_enter_tk(struct cpuidle_device *dev,
64 struct cpuidle_driver *drv, int index)
65{
66 return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter);
67}
68
69typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev,
70 struct cpuidle_driver *drv, int index);
71
72static cpuidle_enter_t cpuidle_enter_ops;
73
1a022e3f
BO
74/**
75 * cpuidle_play_dead - cpu off-lining
76 *
77 * Only returns in case of an error
78 */
79int cpuidle_play_dead(void)
80{
81 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
82 struct cpuidle_driver *drv = cpuidle_get_driver();
83 int i, dead_state = -1;
84 int power_usage = -1;
85
86 /* Find lowest-power state that supports long-term idle */
87 for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
88 struct cpuidle_state *s = &drv->states[i];
89
90 if (s->power_usage < power_usage && s->enter_dead) {
91 power_usage = s->power_usage;
92 dead_state = i;
93 }
94 }
95
96 if (dead_state != -1)
97 return drv->states[dead_state].enter_dead(dev, dead_state);
98
99 return -ENODEV;
100}
101
4f86d3a8
LB
102/**
103 * cpuidle_idle_call - the main idle loop
104 *
105 * NOTE: no locks or semaphores should be used here
a0bfa137 106 * return non-zero on failure
4f86d3a8 107 */
a0bfa137 108int cpuidle_idle_call(void)
4f86d3a8 109{
4a6f4fe8 110 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
46bcfad7 111 struct cpuidle_driver *drv = cpuidle_get_driver();
e978aa7d 112 int next_state, entered_state;
4f86d3a8 113
a0bfa137
LB
114 if (off)
115 return -ENODEV;
116
117 if (!initialized)
118 return -ENODEV;
119
4f86d3a8 120 /* check if the device is ready */
a0bfa137
LB
121 if (!dev || !dev->enabled)
122 return -EBUSY;
4f86d3a8 123
9a655837
AV
124#if 0
125 /* shows regressions, re-enable for 2.6.29 */
2e94d1f7
AV
126 /*
127 * run any timers that can be run now, at this point
128 * before calculating the idle duration etc.
129 */
130 hrtimer_peek_ahead_timers();
9a655837 131#endif
71abbbf8 132
4f86d3a8 133 /* ask the governor for the next state */
46bcfad7 134 next_state = cpuidle_curr_governor->select(drv, dev);
246eb7f0
KH
135 if (need_resched()) {
136 local_irq_enable();
a0bfa137 137 return 0;
246eb7f0
KH
138 }
139
76027ea8
SR
140 trace_power_start_rcuidle(POWER_CSTATE, next_state, dev->cpu);
141 trace_cpu_idle_rcuidle(next_state, dev->cpu);
f77cfe4e 142
e1689795 143 entered_state = cpuidle_enter_ops(dev, drv, next_state);
f77cfe4e 144
76027ea8
SR
145 trace_power_end_rcuidle(dev->cpu);
146 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
f77cfe4e 147
e978aa7d
DD
148 if (entered_state >= 0) {
149 /* Update cpuidle counters */
150 /* This can be moved to within driver enter routine
151 * but that results in multiple copies of same code.
152 */
4202735e 153 dev->states_usage[entered_state].time +=
e978aa7d 154 (unsigned long long)dev->last_residency;
4202735e 155 dev->states_usage[entered_state].usage++;
e1689795
RL
156 } else {
157 dev->last_residency = 0;
e978aa7d 158 }
4f86d3a8
LB
159
160 /* give the governor an opportunity to reflect on the outcome */
161 if (cpuidle_curr_governor->reflect)
e978aa7d 162 cpuidle_curr_governor->reflect(dev, entered_state);
a0bfa137
LB
163
164 return 0;
4f86d3a8
LB
165}
166
167/**
168 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
169 */
170void cpuidle_install_idle_handler(void)
171{
a0bfa137 172 if (enabled_devices) {
4f86d3a8
LB
173 /* Make sure all changes finished before we switch to new idle */
174 smp_wmb();
a0bfa137 175 initialized = 1;
4f86d3a8
LB
176 }
177}
178
179/**
180 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
181 */
182void cpuidle_uninstall_idle_handler(void)
183{
a0bfa137
LB
184 if (enabled_devices) {
185 initialized = 0;
a6869cc4 186 cpuidle_kick_cpus();
4f86d3a8
LB
187 }
188}
189
190/**
191 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
192 */
193void cpuidle_pause_and_lock(void)
194{
195 mutex_lock(&cpuidle_lock);
196 cpuidle_uninstall_idle_handler();
197}
198
199EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
200
201/**
202 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
203 */
204void cpuidle_resume_and_unlock(void)
205{
206 cpuidle_install_idle_handler();
207 mutex_unlock(&cpuidle_lock);
208}
209
210EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
211
e1689795
RL
212/**
213 * cpuidle_wrap_enter - performs timekeeping and irqen around enter function
214 * @dev: pointer to a valid cpuidle_device object
215 * @drv: pointer to a valid cpuidle_driver object
216 * @index: index of the target cpuidle state.
217 */
218int cpuidle_wrap_enter(struct cpuidle_device *dev,
219 struct cpuidle_driver *drv, int index,
220 int (*enter)(struct cpuidle_device *dev,
221 struct cpuidle_driver *drv, int index))
222{
223 ktime_t time_start, time_end;
224 s64 diff;
225
226 time_start = ktime_get();
227
228 index = enter(dev, drv, index);
229
230 time_end = ktime_get();
231
232 local_irq_enable();
233
234 diff = ktime_to_us(ktime_sub(time_end, time_start));
235 if (diff > INT_MAX)
236 diff = INT_MAX;
237
238 dev->last_residency = (int) diff;
239
240 return index;
241}
242
d8c216cf 243#ifdef CONFIG_ARCH_HAS_CPU_RELAX
46bcfad7
DD
244static int poll_idle(struct cpuidle_device *dev,
245 struct cpuidle_driver *drv, int index)
d8c216cf
RW
246{
247 ktime_t t1, t2;
248 s64 diff;
d8c216cf
RW
249
250 t1 = ktime_get();
251 local_irq_enable();
252 while (!need_resched())
253 cpu_relax();
254
255 t2 = ktime_get();
256 diff = ktime_to_us(ktime_sub(t2, t1));
257 if (diff > INT_MAX)
258 diff = INT_MAX;
259
e978aa7d
DD
260 dev->last_residency = (int) diff;
261
262 return index;
d8c216cf
RW
263}
264
46bcfad7 265static void poll_idle_init(struct cpuidle_driver *drv)
d8c216cf 266{
46bcfad7 267 struct cpuidle_state *state = &drv->states[0];
d8c216cf 268
720f1c30 269 snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
d8c216cf
RW
270 snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
271 state->exit_latency = 0;
272 state->target_residency = 0;
273 state->power_usage = -1;
d247632c 274 state->flags = 0;
d8c216cf 275 state->enter = poll_idle;
3a53396b 276 state->disable = 0;
d8c216cf
RW
277}
278#else
46bcfad7 279static void poll_idle_init(struct cpuidle_driver *drv) {}
d8c216cf
RW
280#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
281
4f86d3a8
LB
282/**
283 * cpuidle_enable_device - enables idle PM for a CPU
284 * @dev: the CPU
285 *
286 * This function must be called between cpuidle_pause_and_lock and
287 * cpuidle_resume_and_unlock when used externally.
288 */
289int cpuidle_enable_device(struct cpuidle_device *dev)
290{
291 int ret, i;
e1689795 292 struct cpuidle_driver *drv = cpuidle_get_driver();
4f86d3a8
LB
293
294 if (dev->enabled)
295 return 0;
e1689795 296 if (!drv || !cpuidle_curr_governor)
4f86d3a8
LB
297 return -EIO;
298 if (!dev->state_count)
fc850f39 299 dev->state_count = drv->state_count;
4f86d3a8 300
dcb84f33
VP
301 if (dev->registered == 0) {
302 ret = __cpuidle_register_device(dev);
303 if (ret)
304 return ret;
305 }
306
e1689795
RL
307 cpuidle_enter_ops = drv->en_core_tk_irqen ?
308 cpuidle_enter_tk : cpuidle_enter;
309
310 poll_idle_init(drv);
d8c216cf 311
4f86d3a8
LB
312 if ((ret = cpuidle_add_state_sysfs(dev)))
313 return ret;
314
315 if (cpuidle_curr_governor->enable &&
e1689795 316 (ret = cpuidle_curr_governor->enable(drv, dev)))
4f86d3a8
LB
317 goto fail_sysfs;
318
319 for (i = 0; i < dev->state_count; i++) {
4202735e
DD
320 dev->states_usage[i].usage = 0;
321 dev->states_usage[i].time = 0;
4f86d3a8
LB
322 }
323 dev->last_residency = 0;
4f86d3a8
LB
324
325 smp_wmb();
326
327 dev->enabled = 1;
328
329 enabled_devices++;
330 return 0;
331
332fail_sysfs:
333 cpuidle_remove_state_sysfs(dev);
334
335 return ret;
336}
337
338EXPORT_SYMBOL_GPL(cpuidle_enable_device);
339
340/**
341 * cpuidle_disable_device - disables idle PM for a CPU
342 * @dev: the CPU
343 *
344 * This function must be called between cpuidle_pause_and_lock and
345 * cpuidle_resume_and_unlock when used externally.
346 */
347void cpuidle_disable_device(struct cpuidle_device *dev)
348{
349 if (!dev->enabled)
350 return;
752138df 351 if (!cpuidle_get_driver() || !cpuidle_curr_governor)
4f86d3a8
LB
352 return;
353
354 dev->enabled = 0;
355
356 if (cpuidle_curr_governor->disable)
46bcfad7 357 cpuidle_curr_governor->disable(cpuidle_get_driver(), dev);
4f86d3a8
LB
358
359 cpuidle_remove_state_sysfs(dev);
360 enabled_devices--;
361}
362
363EXPORT_SYMBOL_GPL(cpuidle_disable_device);
364
365/**
dcb84f33
VP
366 * __cpuidle_register_device - internal register function called before register
367 * and enable routines
4f86d3a8 368 * @dev: the cpu
dcb84f33
VP
369 *
370 * cpuidle_lock mutex must be held before this is called
4f86d3a8 371 */
dcb84f33 372static int __cpuidle_register_device(struct cpuidle_device *dev)
4f86d3a8
LB
373{
374 int ret;
8a25a2fd 375 struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
752138df 376 struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
4f86d3a8 377
8a25a2fd 378 if (!dev)
4f86d3a8 379 return -EINVAL;
752138df 380 if (!try_module_get(cpuidle_driver->owner))
4f86d3a8
LB
381 return -EINVAL;
382
383 init_completion(&dev->kobj_unregister);
384
4f86d3a8
LB
385 per_cpu(cpuidle_devices, dev->cpu) = dev;
386 list_add(&dev->device_list, &cpuidle_detected_devices);
8a25a2fd 387 if ((ret = cpuidle_add_sysfs(cpu_dev))) {
752138df 388 module_put(cpuidle_driver->owner);
4f86d3a8
LB
389 return ret;
390 }
391
dcb84f33
VP
392 dev->registered = 1;
393 return 0;
394}
395
396/**
397 * cpuidle_register_device - registers a CPU's idle PM feature
398 * @dev: the cpu
399 */
400int cpuidle_register_device(struct cpuidle_device *dev)
401{
402 int ret;
403
404 mutex_lock(&cpuidle_lock);
405
406 if ((ret = __cpuidle_register_device(dev))) {
407 mutex_unlock(&cpuidle_lock);
408 return ret;
409 }
410
4f86d3a8
LB
411 cpuidle_enable_device(dev);
412 cpuidle_install_idle_handler();
413
414 mutex_unlock(&cpuidle_lock);
415
416 return 0;
417
418}
419
420EXPORT_SYMBOL_GPL(cpuidle_register_device);
421
422/**
423 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
424 * @dev: the cpu
425 */
426void cpuidle_unregister_device(struct cpuidle_device *dev)
427{
8a25a2fd 428 struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu);
752138df 429 struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
4f86d3a8 430
dcb84f33
VP
431 if (dev->registered == 0)
432 return;
433
4f86d3a8
LB
434 cpuidle_pause_and_lock();
435
436 cpuidle_disable_device(dev);
437
8a25a2fd 438 cpuidle_remove_sysfs(cpu_dev);
4f86d3a8
LB
439 list_del(&dev->device_list);
440 wait_for_completion(&dev->kobj_unregister);
441 per_cpu(cpuidle_devices, dev->cpu) = NULL;
442
443 cpuidle_resume_and_unlock();
444
752138df 445 module_put(cpuidle_driver->owner);
4f86d3a8
LB
446}
447
448EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
449
450#ifdef CONFIG_SMP
451
452static void smp_callback(void *v)
453{
454 /* we already woke the CPU up, nothing more to do */
455}
456
457/*
458 * This function gets called when a part of the kernel has a new latency
459 * requirement. This means we need to get all processors out of their C-state,
460 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
461 * wakes them all right up.
462 */
463static int cpuidle_latency_notify(struct notifier_block *b,
464 unsigned long l, void *v)
465{
8691e5a8 466 smp_call_function(smp_callback, NULL, 1);
4f86d3a8
LB
467 return NOTIFY_OK;
468}
469
470static struct notifier_block cpuidle_latency_notifier = {
471 .notifier_call = cpuidle_latency_notify,
472};
473
d82b3518
MG
474static inline void latency_notifier_init(struct notifier_block *n)
475{
476 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
477}
4f86d3a8
LB
478
479#else /* CONFIG_SMP */
480
481#define latency_notifier_init(x) do { } while (0)
482
483#endif /* CONFIG_SMP */
484
485/**
486 * cpuidle_init - core initializer
487 */
488static int __init cpuidle_init(void)
489{
490 int ret;
491
62027aea
LB
492 if (cpuidle_disabled())
493 return -ENODEV;
494
8a25a2fd 495 ret = cpuidle_add_interface(cpu_subsys.dev_root);
4f86d3a8
LB
496 if (ret)
497 return ret;
498
499 latency_notifier_init(&cpuidle_latency_notifier);
500
501 return 0;
502}
503
62027aea 504module_param(off, int, 0444);
4f86d3a8 505core_initcall(cpuidle_init);