xref: /openbmc/linux/drivers/cpuidle/cpuidle.c (revision b6dcefde) 
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>
15 #include <linux/pm_qos_params.h>
16 #include <linux/cpu.h>
17 #include <linux/cpuidle.h>
18 #include <linux/ktime.h>
19 #include <linux/hrtimer.h>
20 #include <trace/events/power.h>
21 
22 #include "cpuidle.h"
23 
24 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
25 
26 DEFINE_MUTEX(cpuidle_lock);
27 LIST_HEAD(cpuidle_detected_devices);
28 static void (*pm_idle_old)(void);
29 
30 static int enabled_devices;
31 
32 #if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
33 static void cpuidle_kick_cpus(void)
34 {
35 	cpu_idle_wait();
36 }
37 #elif defined(CONFIG_SMP)
38 # error "Arch needs cpu_idle_wait() equivalent here"
39 #else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
40 static void cpuidle_kick_cpus(void) {}
41 #endif
42 
43 static int __cpuidle_register_device(struct cpuidle_device *dev);
44 
45 /**
46  * cpuidle_idle_call - the main idle loop
47  *
48  * NOTE: no locks or semaphores should be used here
49  */
50 static void cpuidle_idle_call(void)
51 {
52 	struct cpuidle_device *dev = __get_cpu_var(cpuidle_devices);
53 	struct cpuidle_state *target_state;
54 	int next_state;
55 
56 	/* check if the device is ready */
57 	if (!dev || !dev->enabled) {
58 		if (pm_idle_old)
59 			pm_idle_old();
60 		else
61 #if defined(CONFIG_ARCH_HAS_DEFAULT_IDLE)
62 			default_idle();
63 #else
64 			local_irq_enable();
65 #endif
66 		return;
67 	}
68 
69 #if 0
70 	/* shows regressions, re-enable for 2.6.29 */
71 	/*
72 	 * run any timers that can be run now, at this point
73 	 * before calculating the idle duration etc.
74 	 */
75 	hrtimer_peek_ahead_timers();
76 #endif
77 	/* ask the governor for the next state */
78 	next_state = cpuidle_curr_governor->select(dev);
79 	if (need_resched()) {
80 		local_irq_enable();
81 		return;
82 	}
83 
84 	target_state = &dev->states[next_state];
85 
86 	/* enter the state and update stats */
87 	dev->last_state = target_state;
88 	dev->last_residency = target_state->enter(dev, target_state);
89 	if (dev->last_state)
90 		target_state = dev->last_state;
91 
92 	target_state->time += (unsigned long long)dev->last_residency;
93 	target_state->usage++;
94 
95 	/* give the governor an opportunity to reflect on the outcome */
96 	if (cpuidle_curr_governor->reflect)
97 		cpuidle_curr_governor->reflect(dev);
98 	trace_power_end(0);
99 }
100 
101 /**
102  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
103  */
104 void cpuidle_install_idle_handler(void)
105 {
106 	if (enabled_devices && (pm_idle != cpuidle_idle_call)) {
107 		/* Make sure all changes finished before we switch to new idle */
108 		smp_wmb();
109 		pm_idle = cpuidle_idle_call;
110 	}
111 }
112 
113 /**
114  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
115  */
116 void cpuidle_uninstall_idle_handler(void)
117 {
118 	if (enabled_devices && pm_idle_old && (pm_idle != pm_idle_old)) {
119 		pm_idle = pm_idle_old;
120 		cpuidle_kick_cpus();
121 	}
122 }
123 
124 /**
125  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
126  */
127 void cpuidle_pause_and_lock(void)
128 {
129 	mutex_lock(&cpuidle_lock);
130 	cpuidle_uninstall_idle_handler();
131 }
132 
133 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
134 
135 /**
136  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
137  */
138 void cpuidle_resume_and_unlock(void)
139 {
140 	cpuidle_install_idle_handler();
141 	mutex_unlock(&cpuidle_lock);
142 }
143 
144 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
145 
146 /**
147  * cpuidle_enable_device - enables idle PM for a CPU
148  * @dev: the CPU
149  *
150  * This function must be called between cpuidle_pause_and_lock and
151  * cpuidle_resume_and_unlock when used externally.
152  */
153 int cpuidle_enable_device(struct cpuidle_device *dev)
154 {
155 	int ret, i;
156 
157 	if (dev->enabled)
158 		return 0;
159 	if (!cpuidle_curr_driver || !cpuidle_curr_governor)
160 		return -EIO;
161 	if (!dev->state_count)
162 		return -EINVAL;
163 
164 	if (dev->registered == 0) {
165 		ret = __cpuidle_register_device(dev);
166 		if (ret)
167 			return ret;
168 	}
169 
170 	if ((ret = cpuidle_add_state_sysfs(dev)))
171 		return ret;
172 
173 	if (cpuidle_curr_governor->enable &&
174 	    (ret = cpuidle_curr_governor->enable(dev)))
175 		goto fail_sysfs;
176 
177 	for (i = 0; i < dev->state_count; i++) {
178 		dev->states[i].usage = 0;
179 		dev->states[i].time = 0;
180 	}
181 	dev->last_residency = 0;
182 	dev->last_state = NULL;
183 
184 	smp_wmb();
185 
186 	dev->enabled = 1;
187 
188 	enabled_devices++;
189 	return 0;
190 
191 fail_sysfs:
192 	cpuidle_remove_state_sysfs(dev);
193 
194 	return ret;
195 }
196 
197 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
198 
199 /**
200  * cpuidle_disable_device - disables idle PM for a CPU
201  * @dev: the CPU
202  *
203  * This function must be called between cpuidle_pause_and_lock and
204  * cpuidle_resume_and_unlock when used externally.
205  */
206 void cpuidle_disable_device(struct cpuidle_device *dev)
207 {
208 	if (!dev->enabled)
209 		return;
210 	if (!cpuidle_curr_driver || !cpuidle_curr_governor)
211 		return;
212 
213 	dev->enabled = 0;
214 
215 	if (cpuidle_curr_governor->disable)
216 		cpuidle_curr_governor->disable(dev);
217 
218 	cpuidle_remove_state_sysfs(dev);
219 	enabled_devices--;
220 }
221 
222 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
223 
224 #ifdef CONFIG_ARCH_HAS_CPU_RELAX
225 static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
226 {
227 	ktime_t	t1, t2;
228 	s64 diff;
229 	int ret;
230 
231 	t1 = ktime_get();
232 	local_irq_enable();
233 	while (!need_resched())
234 		cpu_relax();
235 
236 	t2 = ktime_get();
237 	diff = ktime_to_us(ktime_sub(t2, t1));
238 	if (diff > INT_MAX)
239 		diff = INT_MAX;
240 
241 	ret = (int) diff;
242 	return ret;
243 }
244 
245 static void poll_idle_init(struct cpuidle_device *dev)
246 {
247 	struct cpuidle_state *state = &dev->states[0];
248 
249 	cpuidle_set_statedata(state, NULL);
250 
251 	snprintf(state->name, CPUIDLE_NAME_LEN, "C0");
252 	snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
253 	state->exit_latency = 0;
254 	state->target_residency = 0;
255 	state->power_usage = -1;
256 	state->flags = CPUIDLE_FLAG_POLL;
257 	state->enter = poll_idle;
258 }
259 #else
260 static void poll_idle_init(struct cpuidle_device *dev) {}
261 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */
262 
263 /**
264  * __cpuidle_register_device - internal register function called before register
265  * and enable routines
266  * @dev: the cpu
267  *
268  * cpuidle_lock mutex must be held before this is called
269  */
270 static int __cpuidle_register_device(struct cpuidle_device *dev)
271 {
272 	int ret;
273 	struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
274 
275 	if (!sys_dev)
276 		return -EINVAL;
277 	if (!try_module_get(cpuidle_curr_driver->owner))
278 		return -EINVAL;
279 
280 	init_completion(&dev->kobj_unregister);
281 
282 	poll_idle_init(dev);
283 
284 	per_cpu(cpuidle_devices, dev->cpu) = dev;
285 	list_add(&dev->device_list, &cpuidle_detected_devices);
286 	if ((ret = cpuidle_add_sysfs(sys_dev))) {
287 		module_put(cpuidle_curr_driver->owner);
288 		return ret;
289 	}
290 
291 	dev->registered = 1;
292 	return 0;
293 }
294 
295 /**
296  * cpuidle_register_device - registers a CPU's idle PM feature
297  * @dev: the cpu
298  */
299 int cpuidle_register_device(struct cpuidle_device *dev)
300 {
301 	int ret;
302 
303 	mutex_lock(&cpuidle_lock);
304 
305 	if ((ret = __cpuidle_register_device(dev))) {
306 		mutex_unlock(&cpuidle_lock);
307 		return ret;
308 	}
309 
310 	cpuidle_enable_device(dev);
311 	cpuidle_install_idle_handler();
312 
313 	mutex_unlock(&cpuidle_lock);
314 
315 	return 0;
316 
317 }
318 
319 EXPORT_SYMBOL_GPL(cpuidle_register_device);
320 
321 /**
322  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
323  * @dev: the cpu
324  */
325 void cpuidle_unregister_device(struct cpuidle_device *dev)
326 {
327 	struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
328 
329 	if (dev->registered == 0)
330 		return;
331 
332 	cpuidle_pause_and_lock();
333 
334 	cpuidle_disable_device(dev);
335 
336 	cpuidle_remove_sysfs(sys_dev);
337 	list_del(&dev->device_list);
338 	wait_for_completion(&dev->kobj_unregister);
339 	per_cpu(cpuidle_devices, dev->cpu) = NULL;
340 
341 	cpuidle_resume_and_unlock();
342 
343 	module_put(cpuidle_curr_driver->owner);
344 }
345 
346 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
347 
348 #ifdef CONFIG_SMP
349 
350 static void smp_callback(void *v)
351 {
352 	/* we already woke the CPU up, nothing more to do */
353 }
354 
355 /*
356  * This function gets called when a part of the kernel has a new latency
357  * requirement.  This means we need to get all processors out of their C-state,
358  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
359  * wakes them all right up.
360  */
361 static int cpuidle_latency_notify(struct notifier_block *b,
362 		unsigned long l, void *v)
363 {
364 	smp_call_function(smp_callback, NULL, 1);
365 	return NOTIFY_OK;
366 }
367 
368 static struct notifier_block cpuidle_latency_notifier = {
369 	.notifier_call = cpuidle_latency_notify,
370 };
371 
372 static inline void latency_notifier_init(struct notifier_block *n)
373 {
374 	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
375 }
376 
377 #else /* CONFIG_SMP */
378 
379 #define latency_notifier_init(x) do { } while (0)
380 
381 #endif /* CONFIG_SMP */
382 
383 /**
384  * cpuidle_init - core initializer
385  */
386 static int __init cpuidle_init(void)
387 {
388 	int ret;
389 
390 	pm_idle_old = pm_idle;
391 
392 	ret = cpuidle_add_class_sysfs(&cpu_sysdev_class);
393 	if (ret)
394 		return ret;
395 
396 	latency_notifier_init(&cpuidle_latency_notifier);
397 
398 	return 0;
399 }
400 
401 core_initcall(cpuidle_init);
402