xref: /linux/drivers/cpufreq/cpufreq.c (revision ed3174d93c342b8b2eeba6bbd124707d55304a7b)
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *
7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
8  *	Added handling for CPU hotplug
9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
10  *	Fix handling for CPU hotplug -- affected CPUs
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License version 2 as
14  * published by the Free Software Foundation.
15  *
16  */
17 
18 #include <linux/kernel.h>
19 #include <linux/module.h>
20 #include <linux/init.h>
21 #include <linux/notifier.h>
22 #include <linux/cpufreq.h>
23 #include <linux/delay.h>
24 #include <linux/interrupt.h>
25 #include <linux/spinlock.h>
26 #include <linux/device.h>
27 #include <linux/slab.h>
28 #include <linux/cpu.h>
29 #include <linux/completion.h>
30 #include <linux/mutex.h>
31 
32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
33 						"cpufreq-core", msg)
34 
35 /**
36  * The "cpufreq driver" - the arch- or hardware-dependent low
37  * level driver of CPUFreq support, and its spinlock. This lock
38  * also protects the cpufreq_cpu_data array.
39  */
40 static struct cpufreq_driver *cpufreq_driver;
41 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
42 #ifdef CONFIG_HOTPLUG_CPU
43 /* This one keeps track of the previously set governor of a removed CPU */
44 static struct cpufreq_governor *cpufreq_cpu_governor[NR_CPUS];
45 #endif
46 static DEFINE_SPINLOCK(cpufreq_driver_lock);
47 
48 /*
49  * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
50  * all cpufreq/hotplug/workqueue/etc related lock issues.
51  *
52  * The rules for this semaphore:
53  * - Any routine that wants to read from the policy structure will
54  *   do a down_read on this semaphore.
55  * - Any routine that will write to the policy structure and/or may take away
56  *   the policy altogether (eg. CPU hotplug), will hold this lock in write
57  *   mode before doing so.
58  *
59  * Additional rules:
60  * - All holders of the lock should check to make sure that the CPU they
61  *   are concerned with are online after they get the lock.
62  * - Governor routines that can be called in cpufreq hotplug path should not
63  *   take this sem as top level hotplug notifier handler takes this.
64  */
65 static DEFINE_PER_CPU(int, policy_cpu);
66 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
67 
68 #define lock_policy_rwsem(mode, cpu)					\
69 int lock_policy_rwsem_##mode						\
70 (int cpu)								\
71 {									\
72 	int policy_cpu = per_cpu(policy_cpu, cpu);			\
73 	BUG_ON(policy_cpu == -1);					\
74 	down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));		\
75 	if (unlikely(!cpu_online(cpu))) {				\
76 		up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));	\
77 		return -1;						\
78 	}								\
79 									\
80 	return 0;							\
81 }
82 
83 lock_policy_rwsem(read, cpu);
84 EXPORT_SYMBOL_GPL(lock_policy_rwsem_read);
85 
86 lock_policy_rwsem(write, cpu);
87 EXPORT_SYMBOL_GPL(lock_policy_rwsem_write);
88 
89 void unlock_policy_rwsem_read(int cpu)
90 {
91 	int policy_cpu = per_cpu(policy_cpu, cpu);
92 	BUG_ON(policy_cpu == -1);
93 	up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
94 }
95 EXPORT_SYMBOL_GPL(unlock_policy_rwsem_read);
96 
97 void unlock_policy_rwsem_write(int cpu)
98 {
99 	int policy_cpu = per_cpu(policy_cpu, cpu);
100 	BUG_ON(policy_cpu == -1);
101 	up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
102 }
103 EXPORT_SYMBOL_GPL(unlock_policy_rwsem_write);
104 
105 
106 /* internal prototypes */
107 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
108 static unsigned int __cpufreq_get(unsigned int cpu);
109 static void handle_update(struct work_struct *work);
110 
111 /**
112  * Two notifier lists: the "policy" list is involved in the
113  * validation process for a new CPU frequency policy; the
114  * "transition" list for kernel code that needs to handle
115  * changes to devices when the CPU clock speed changes.
116  * The mutex locks both lists.
117  */
118 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
119 static struct srcu_notifier_head cpufreq_transition_notifier_list;
120 
121 static int __init init_cpufreq_transition_notifier_list(void)
122 {
123 	srcu_init_notifier_head(&cpufreq_transition_notifier_list);
124 	return 0;
125 }
126 pure_initcall(init_cpufreq_transition_notifier_list);
127 
128 static LIST_HEAD(cpufreq_governor_list);
129 static DEFINE_MUTEX (cpufreq_governor_mutex);
130 
131 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
132 {
133 	struct cpufreq_policy *data;
134 	unsigned long flags;
135 
136 	if (cpu >= NR_CPUS)
137 		goto err_out;
138 
139 	/* get the cpufreq driver */
140 	spin_lock_irqsave(&cpufreq_driver_lock, flags);
141 
142 	if (!cpufreq_driver)
143 		goto err_out_unlock;
144 
145 	if (!try_module_get(cpufreq_driver->owner))
146 		goto err_out_unlock;
147 
148 
149 	/* get the CPU */
150 	data = cpufreq_cpu_data[cpu];
151 
152 	if (!data)
153 		goto err_out_put_module;
154 
155 	if (!kobject_get(&data->kobj))
156 		goto err_out_put_module;
157 
158 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
159 	return data;
160 
161 err_out_put_module:
162 	module_put(cpufreq_driver->owner);
163 err_out_unlock:
164 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
165 err_out:
166 	return NULL;
167 }
168 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
169 
170 
171 void cpufreq_cpu_put(struct cpufreq_policy *data)
172 {
173 	kobject_put(&data->kobj);
174 	module_put(cpufreq_driver->owner);
175 }
176 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
177 
178 
179 /*********************************************************************
180  *                     UNIFIED DEBUG HELPERS                         *
181  *********************************************************************/
182 #ifdef CONFIG_CPU_FREQ_DEBUG
183 
184 /* what part(s) of the CPUfreq subsystem are debugged? */
185 static unsigned int debug;
186 
187 /* is the debug output ratelimit'ed using printk_ratelimit? User can
188  * set or modify this value.
189  */
190 static unsigned int debug_ratelimit = 1;
191 
192 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
193  * loading of a cpufreq driver, temporarily disabled when a new policy
194  * is set, and disabled upon cpufreq driver removal
195  */
196 static unsigned int disable_ratelimit = 1;
197 static DEFINE_SPINLOCK(disable_ratelimit_lock);
198 
199 static void cpufreq_debug_enable_ratelimit(void)
200 {
201 	unsigned long flags;
202 
203 	spin_lock_irqsave(&disable_ratelimit_lock, flags);
204 	if (disable_ratelimit)
205 		disable_ratelimit--;
206 	spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
207 }
208 
209 static void cpufreq_debug_disable_ratelimit(void)
210 {
211 	unsigned long flags;
212 
213 	spin_lock_irqsave(&disable_ratelimit_lock, flags);
214 	disable_ratelimit++;
215 	spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
216 }
217 
218 void cpufreq_debug_printk(unsigned int type, const char *prefix,
219 							const char *fmt, ...)
220 {
221 	char s[256];
222 	va_list args;
223 	unsigned int len;
224 	unsigned long flags;
225 
226 	WARN_ON(!prefix);
227 	if (type & debug) {
228 		spin_lock_irqsave(&disable_ratelimit_lock, flags);
229 		if (!disable_ratelimit && debug_ratelimit
230 					&& !printk_ratelimit()) {
231 			spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
232 			return;
233 		}
234 		spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
235 
236 		len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
237 
238 		va_start(args, fmt);
239 		len += vsnprintf(&s[len], (256 - len), fmt, args);
240 		va_end(args);
241 
242 		printk(s);
243 
244 		WARN_ON(len < 5);
245 	}
246 }
247 EXPORT_SYMBOL(cpufreq_debug_printk);
248 
249 
250 module_param(debug, uint, 0644);
251 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core,"
252 			" 2 to debug drivers, and 4 to debug governors.");
253 
254 module_param(debug_ratelimit, uint, 0644);
255 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging:"
256 					" set to 0 to disable ratelimiting.");
257 
258 #else /* !CONFIG_CPU_FREQ_DEBUG */
259 
260 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
261 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
262 
263 #endif /* CONFIG_CPU_FREQ_DEBUG */
264 
265 
266 /*********************************************************************
267  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
268  *********************************************************************/
269 
270 /**
271  * adjust_jiffies - adjust the system "loops_per_jiffy"
272  *
273  * This function alters the system "loops_per_jiffy" for the clock
274  * speed change. Note that loops_per_jiffy cannot be updated on SMP
275  * systems as each CPU might be scaled differently. So, use the arch
276  * per-CPU loops_per_jiffy value wherever possible.
277  */
278 #ifndef CONFIG_SMP
279 static unsigned long l_p_j_ref;
280 static unsigned int  l_p_j_ref_freq;
281 
282 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
283 {
284 	if (ci->flags & CPUFREQ_CONST_LOOPS)
285 		return;
286 
287 	if (!l_p_j_ref_freq) {
288 		l_p_j_ref = loops_per_jiffy;
289 		l_p_j_ref_freq = ci->old;
290 		dprintk("saving %lu as reference value for loops_per_jiffy; "
291 			"freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
292 	}
293 	if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
294 	    (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
295 	    (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
296 		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297 								ci->new);
298 		dprintk("scaling loops_per_jiffy to %lu "
299 			"for frequency %u kHz\n", loops_per_jiffy, ci->new);
300 	}
301 }
302 #else
303 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
304 {
305 	return;
306 }
307 #endif
308 
309 
310 /**
311  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
312  * on frequency transition.
313  *
314  * This function calls the transition notifiers and the "adjust_jiffies"
315  * function. It is called twice on all CPU frequency changes that have
316  * external effects.
317  */
318 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
319 {
320 	struct cpufreq_policy *policy;
321 
322 	BUG_ON(irqs_disabled());
323 
324 	freqs->flags = cpufreq_driver->flags;
325 	dprintk("notification %u of frequency transition to %u kHz\n",
326 		state, freqs->new);
327 
328 	policy = cpufreq_cpu_data[freqs->cpu];
329 	switch (state) {
330 
331 	case CPUFREQ_PRECHANGE:
332 		/* detect if the driver reported a value as "old frequency"
333 		 * which is not equal to what the cpufreq core thinks is
334 		 * "old frequency".
335 		 */
336 		if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
337 			if ((policy) && (policy->cpu == freqs->cpu) &&
338 			    (policy->cur) && (policy->cur != freqs->old)) {
339 				dprintk("Warning: CPU frequency is"
340 					" %u, cpufreq assumed %u kHz.\n",
341 					freqs->old, policy->cur);
342 				freqs->old = policy->cur;
343 			}
344 		}
345 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
346 				CPUFREQ_PRECHANGE, freqs);
347 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
348 		break;
349 
350 	case CPUFREQ_POSTCHANGE:
351 		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
352 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
353 				CPUFREQ_POSTCHANGE, freqs);
354 		if (likely(policy) && likely(policy->cpu == freqs->cpu))
355 			policy->cur = freqs->new;
356 		break;
357 	}
358 }
359 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
360 
361 
362 
363 /*********************************************************************
364  *                          SYSFS INTERFACE                          *
365  *********************************************************************/
366 
367 static struct cpufreq_governor *__find_governor(const char *str_governor)
368 {
369 	struct cpufreq_governor *t;
370 
371 	list_for_each_entry(t, &cpufreq_governor_list, governor_list)
372 		if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
373 			return t;
374 
375 	return NULL;
376 }
377 
378 /**
379  * cpufreq_parse_governor - parse a governor string
380  */
381 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
382 				struct cpufreq_governor **governor)
383 {
384 	int err = -EINVAL;
385 
386 	if (!cpufreq_driver)
387 		goto out;
388 
389 	if (cpufreq_driver->setpolicy) {
390 		if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
391 			*policy = CPUFREQ_POLICY_PERFORMANCE;
392 			err = 0;
393 		} else if (!strnicmp(str_governor, "powersave",
394 						CPUFREQ_NAME_LEN)) {
395 			*policy = CPUFREQ_POLICY_POWERSAVE;
396 			err = 0;
397 		}
398 	} else if (cpufreq_driver->target) {
399 		struct cpufreq_governor *t;
400 
401 		mutex_lock(&cpufreq_governor_mutex);
402 
403 		t = __find_governor(str_governor);
404 
405 		if (t == NULL) {
406 			char *name = kasprintf(GFP_KERNEL, "cpufreq_%s",
407 								str_governor);
408 
409 			if (name) {
410 				int ret;
411 
412 				mutex_unlock(&cpufreq_governor_mutex);
413 				ret = request_module(name);
414 				mutex_lock(&cpufreq_governor_mutex);
415 
416 				if (ret == 0)
417 					t = __find_governor(str_governor);
418 			}
419 
420 			kfree(name);
421 		}
422 
423 		if (t != NULL) {
424 			*governor = t;
425 			err = 0;
426 		}
427 
428 		mutex_unlock(&cpufreq_governor_mutex);
429 	}
430   out:
431 	return err;
432 }
433 
434 
435 /* drivers/base/cpu.c */
436 extern struct sysdev_class cpu_sysdev_class;
437 
438 
439 /**
440  * cpufreq_per_cpu_attr_read() / show_##file_name() -
441  * print out cpufreq information
442  *
443  * Write out information from cpufreq_driver->policy[cpu]; object must be
444  * "unsigned int".
445  */
446 
447 #define show_one(file_name, object)			\
448 static ssize_t show_##file_name				\
449 (struct cpufreq_policy * policy, char *buf)		\
450 {							\
451 	return sprintf (buf, "%u\n", policy->object);	\
452 }
453 
454 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
455 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
456 show_one(scaling_min_freq, min);
457 show_one(scaling_max_freq, max);
458 show_one(scaling_cur_freq, cur);
459 
460 static int __cpufreq_set_policy(struct cpufreq_policy *data,
461 				struct cpufreq_policy *policy);
462 
463 /**
464  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
465  */
466 #define store_one(file_name, object)			\
467 static ssize_t store_##file_name					\
468 (struct cpufreq_policy * policy, const char *buf, size_t count)		\
469 {									\
470 	unsigned int ret = -EINVAL;					\
471 	struct cpufreq_policy new_policy;				\
472 									\
473 	ret = cpufreq_get_policy(&new_policy, policy->cpu);		\
474 	if (ret)							\
475 		return -EINVAL;						\
476 									\
477 	ret = sscanf (buf, "%u", &new_policy.object);			\
478 	if (ret != 1)							\
479 		return -EINVAL;						\
480 									\
481 	ret = __cpufreq_set_policy(policy, &new_policy);		\
482 	policy->user_policy.object = policy->object;			\
483 									\
484 	return ret ? ret : count;					\
485 }
486 
487 store_one(scaling_min_freq,min);
488 store_one(scaling_max_freq,max);
489 
490 /**
491  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
492  */
493 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy,
494 							char *buf)
495 {
496 	unsigned int cur_freq = __cpufreq_get(policy->cpu);
497 	if (!cur_freq)
498 		return sprintf(buf, "<unknown>");
499 	return sprintf(buf, "%u\n", cur_freq);
500 }
501 
502 
503 /**
504  * show_scaling_governor - show the current policy for the specified CPU
505  */
506 static ssize_t show_scaling_governor (struct cpufreq_policy * policy,
507 							char *buf)
508 {
509 	if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
510 		return sprintf(buf, "powersave\n");
511 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
512 		return sprintf(buf, "performance\n");
513 	else if (policy->governor)
514 		return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
515 	return -EINVAL;
516 }
517 
518 
519 /**
520  * store_scaling_governor - store policy for the specified CPU
521  */
522 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
523 				       const char *buf, size_t count)
524 {
525 	unsigned int ret = -EINVAL;
526 	char	str_governor[16];
527 	struct cpufreq_policy new_policy;
528 
529 	ret = cpufreq_get_policy(&new_policy, policy->cpu);
530 	if (ret)
531 		return ret;
532 
533 	ret = sscanf (buf, "%15s", str_governor);
534 	if (ret != 1)
535 		return -EINVAL;
536 
537 	if (cpufreq_parse_governor(str_governor, &new_policy.policy,
538 						&new_policy.governor))
539 		return -EINVAL;
540 
541 	/* Do not use cpufreq_set_policy here or the user_policy.max
542 	   will be wrongly overridden */
543 	ret = __cpufreq_set_policy(policy, &new_policy);
544 
545 	policy->user_policy.policy = policy->policy;
546 	policy->user_policy.governor = policy->governor;
547 
548 	if (ret)
549 		return ret;
550 	else
551 		return count;
552 }
553 
554 /**
555  * show_scaling_driver - show the cpufreq driver currently loaded
556  */
557 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
558 {
559 	return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
560 }
561 
562 /**
563  * show_scaling_available_governors - show the available CPUfreq governors
564  */
565 static ssize_t show_scaling_available_governors (struct cpufreq_policy *policy,
566 				char *buf)
567 {
568 	ssize_t i = 0;
569 	struct cpufreq_governor *t;
570 
571 	if (!cpufreq_driver->target) {
572 		i += sprintf(buf, "performance powersave");
573 		goto out;
574 	}
575 
576 	list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
577 		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
578 			goto out;
579 		i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
580 	}
581 out:
582 	i += sprintf(&buf[i], "\n");
583 	return i;
584 }
585 /**
586  * show_affected_cpus - show the CPUs affected by each transition
587  */
588 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
589 {
590 	ssize_t i = 0;
591 	unsigned int cpu;
592 
593 	for_each_cpu_mask(cpu, policy->cpus) {
594 		if (i)
595 			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
596 		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
597 		if (i >= (PAGE_SIZE - 5))
598 		    break;
599 	}
600 	i += sprintf(&buf[i], "\n");
601 	return i;
602 }
603 
604 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
605 		const char *buf, size_t count)
606 {
607 	unsigned int freq = 0;
608 	unsigned int ret;
609 
610 	if (!policy->governor->store_setspeed)
611 		return -EINVAL;
612 
613 	ret = sscanf(buf, "%u", &freq);
614 	if (ret != 1)
615 		return -EINVAL;
616 
617 	policy->governor->store_setspeed(policy, freq);
618 
619 	return count;
620 }
621 
622 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
623 {
624 	if (!policy->governor->show_setspeed)
625 		return sprintf(buf, "<unsupported>\n");
626 
627 	return policy->governor->show_setspeed(policy, buf);
628 }
629 
630 #define define_one_ro(_name) \
631 static struct freq_attr _name = \
632 __ATTR(_name, 0444, show_##_name, NULL)
633 
634 #define define_one_ro0400(_name) \
635 static struct freq_attr _name = \
636 __ATTR(_name, 0400, show_##_name, NULL)
637 
638 #define define_one_rw(_name) \
639 static struct freq_attr _name = \
640 __ATTR(_name, 0644, show_##_name, store_##_name)
641 
642 define_one_ro0400(cpuinfo_cur_freq);
643 define_one_ro(cpuinfo_min_freq);
644 define_one_ro(cpuinfo_max_freq);
645 define_one_ro(scaling_available_governors);
646 define_one_ro(scaling_driver);
647 define_one_ro(scaling_cur_freq);
648 define_one_ro(affected_cpus);
649 define_one_rw(scaling_min_freq);
650 define_one_rw(scaling_max_freq);
651 define_one_rw(scaling_governor);
652 define_one_rw(scaling_setspeed);
653 
654 static struct attribute * default_attrs[] = {
655 	&cpuinfo_min_freq.attr,
656 	&cpuinfo_max_freq.attr,
657 	&scaling_min_freq.attr,
658 	&scaling_max_freq.attr,
659 	&affected_cpus.attr,
660 	&scaling_governor.attr,
661 	&scaling_driver.attr,
662 	&scaling_available_governors.attr,
663 	&scaling_setspeed.attr,
664 	NULL
665 };
666 
667 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
668 #define to_attr(a) container_of(a,struct freq_attr,attr)
669 
670 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
671 {
672 	struct cpufreq_policy * policy = to_policy(kobj);
673 	struct freq_attr * fattr = to_attr(attr);
674 	ssize_t ret;
675 	policy = cpufreq_cpu_get(policy->cpu);
676 	if (!policy)
677 		return -EINVAL;
678 
679 	if (lock_policy_rwsem_read(policy->cpu) < 0)
680 		return -EINVAL;
681 
682 	if (fattr->show)
683 		ret = fattr->show(policy, buf);
684 	else
685 		ret = -EIO;
686 
687 	unlock_policy_rwsem_read(policy->cpu);
688 
689 	cpufreq_cpu_put(policy);
690 	return ret;
691 }
692 
693 static ssize_t store(struct kobject * kobj, struct attribute * attr,
694 		     const char * buf, size_t count)
695 {
696 	struct cpufreq_policy * policy = to_policy(kobj);
697 	struct freq_attr * fattr = to_attr(attr);
698 	ssize_t ret;
699 	policy = cpufreq_cpu_get(policy->cpu);
700 	if (!policy)
701 		return -EINVAL;
702 
703 	if (lock_policy_rwsem_write(policy->cpu) < 0)
704 		return -EINVAL;
705 
706 	if (fattr->store)
707 		ret = fattr->store(policy, buf, count);
708 	else
709 		ret = -EIO;
710 
711 	unlock_policy_rwsem_write(policy->cpu);
712 
713 	cpufreq_cpu_put(policy);
714 	return ret;
715 }
716 
717 static void cpufreq_sysfs_release(struct kobject * kobj)
718 {
719 	struct cpufreq_policy * policy = to_policy(kobj);
720 	dprintk("last reference is dropped\n");
721 	complete(&policy->kobj_unregister);
722 }
723 
724 static struct sysfs_ops sysfs_ops = {
725 	.show	= show,
726 	.store	= store,
727 };
728 
729 static struct kobj_type ktype_cpufreq = {
730 	.sysfs_ops	= &sysfs_ops,
731 	.default_attrs	= default_attrs,
732 	.release	= cpufreq_sysfs_release,
733 };
734 
735 
736 /**
737  * cpufreq_add_dev - add a CPU device
738  *
739  * Adds the cpufreq interface for a CPU device.
740  */
741 static int cpufreq_add_dev (struct sys_device * sys_dev)
742 {
743 	unsigned int cpu = sys_dev->id;
744 	int ret = 0;
745 	struct cpufreq_policy new_policy;
746 	struct cpufreq_policy *policy;
747 	struct freq_attr **drv_attr;
748 	struct sys_device *cpu_sys_dev;
749 	unsigned long flags;
750 	unsigned int j;
751 #ifdef CONFIG_SMP
752 	struct cpufreq_policy *managed_policy;
753 #endif
754 
755 	if (cpu_is_offline(cpu))
756 		return 0;
757 
758 	cpufreq_debug_disable_ratelimit();
759 	dprintk("adding CPU %u\n", cpu);
760 
761 #ifdef CONFIG_SMP
762 	/* check whether a different CPU already registered this
763 	 * CPU because it is in the same boat. */
764 	policy = cpufreq_cpu_get(cpu);
765 	if (unlikely(policy)) {
766 		cpufreq_cpu_put(policy);
767 		cpufreq_debug_enable_ratelimit();
768 		return 0;
769 	}
770 #endif
771 
772 	if (!try_module_get(cpufreq_driver->owner)) {
773 		ret = -EINVAL;
774 		goto module_out;
775 	}
776 
777 	policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
778 	if (!policy) {
779 		ret = -ENOMEM;
780 		goto nomem_out;
781 	}
782 
783 	policy->cpu = cpu;
784 	policy->cpus = cpumask_of_cpu(cpu);
785 
786 	/* Initially set CPU itself as the policy_cpu */
787 	per_cpu(policy_cpu, cpu) = cpu;
788 	lock_policy_rwsem_write(cpu);
789 
790 	init_completion(&policy->kobj_unregister);
791 	INIT_WORK(&policy->update, handle_update);
792 
793 	/* Set governor before ->init, so that driver could check it */
794 	policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
795 	/* call driver. From then on the cpufreq must be able
796 	 * to accept all calls to ->verify and ->setpolicy for this CPU
797 	 */
798 	ret = cpufreq_driver->init(policy);
799 	if (ret) {
800 		dprintk("initialization failed\n");
801 		unlock_policy_rwsem_write(cpu);
802 		goto err_out;
803 	}
804 	policy->user_policy.min = policy->cpuinfo.min_freq;
805 	policy->user_policy.max = policy->cpuinfo.max_freq;
806 
807 #ifdef CONFIG_SMP
808 
809 #ifdef CONFIG_HOTPLUG_CPU
810 	if (cpufreq_cpu_governor[cpu]){
811 		policy->governor = cpufreq_cpu_governor[cpu];
812 		dprintk("Restoring governor %s for cpu %d\n",
813 		       policy->governor->name, cpu);
814 	}
815 #endif
816 
817 	for_each_cpu_mask(j, policy->cpus) {
818 		if (cpu == j)
819 			continue;
820 
821 		/* check for existing affected CPUs.  They may not be aware
822 		 * of it due to CPU Hotplug.
823 		 */
824 		managed_policy = cpufreq_cpu_get(j);
825 		if (unlikely(managed_policy)) {
826 
827 			/* Set proper policy_cpu */
828 			unlock_policy_rwsem_write(cpu);
829 			per_cpu(policy_cpu, cpu) = managed_policy->cpu;
830 
831 			if (lock_policy_rwsem_write(cpu) < 0)
832 				goto err_out_driver_exit;
833 
834 			spin_lock_irqsave(&cpufreq_driver_lock, flags);
835 			managed_policy->cpus = policy->cpus;
836 			cpufreq_cpu_data[cpu] = managed_policy;
837 			spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
838 
839 			dprintk("CPU already managed, adding link\n");
840 			ret = sysfs_create_link(&sys_dev->kobj,
841 						&managed_policy->kobj,
842 						"cpufreq");
843 			if (ret) {
844 				unlock_policy_rwsem_write(cpu);
845 				goto err_out_driver_exit;
846 			}
847 
848 			cpufreq_debug_enable_ratelimit();
849 			ret = 0;
850 			unlock_policy_rwsem_write(cpu);
851 			goto err_out_driver_exit; /* call driver->exit() */
852 		}
853 	}
854 #endif
855 	memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
856 
857 	/* prepare interface data */
858 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &sys_dev->kobj,
859 				   "cpufreq");
860 	if (ret) {
861 		unlock_policy_rwsem_write(cpu);
862 		goto err_out_driver_exit;
863 	}
864 	/* set up files for this cpu device */
865 	drv_attr = cpufreq_driver->attr;
866 	while ((drv_attr) && (*drv_attr)) {
867 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
868 		if (ret) {
869 			unlock_policy_rwsem_write(cpu);
870 			goto err_out_driver_exit;
871 		}
872 		drv_attr++;
873 	}
874 	if (cpufreq_driver->get){
875 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
876 		if (ret) {
877 			unlock_policy_rwsem_write(cpu);
878 			goto err_out_driver_exit;
879 		}
880 	}
881 	if (cpufreq_driver->target){
882 		ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
883 		if (ret) {
884 			unlock_policy_rwsem_write(cpu);
885 			goto err_out_driver_exit;
886 		}
887 	}
888 
889 	spin_lock_irqsave(&cpufreq_driver_lock, flags);
890 	for_each_cpu_mask(j, policy->cpus) {
891 		cpufreq_cpu_data[j] = policy;
892 		per_cpu(policy_cpu, j) = policy->cpu;
893 	}
894 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
895 
896 	/* symlink affected CPUs */
897 	for_each_cpu_mask(j, policy->cpus) {
898 		if (j == cpu)
899 			continue;
900 		if (!cpu_online(j))
901 			continue;
902 
903 		dprintk("CPU %u already managed, adding link\n", j);
904 		cpufreq_cpu_get(cpu);
905 		cpu_sys_dev = get_cpu_sysdev(j);
906 		ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
907 					"cpufreq");
908 		if (ret) {
909 			unlock_policy_rwsem_write(cpu);
910 			goto err_out_unregister;
911 		}
912 	}
913 
914 	policy->governor = NULL; /* to assure that the starting sequence is
915 				  * run in cpufreq_set_policy */
916 
917 	/* set default policy */
918 	ret = __cpufreq_set_policy(policy, &new_policy);
919 	policy->user_policy.policy = policy->policy;
920 	policy->user_policy.governor = policy->governor;
921 
922 	unlock_policy_rwsem_write(cpu);
923 
924 	if (ret) {
925 		dprintk("setting policy failed\n");
926 		goto err_out_unregister;
927 	}
928 
929 	kobject_uevent(&policy->kobj, KOBJ_ADD);
930 	module_put(cpufreq_driver->owner);
931 	dprintk("initialization complete\n");
932 	cpufreq_debug_enable_ratelimit();
933 
934 	return 0;
935 
936 
937 err_out_unregister:
938 	spin_lock_irqsave(&cpufreq_driver_lock, flags);
939 	for_each_cpu_mask(j, policy->cpus)
940 		cpufreq_cpu_data[j] = NULL;
941 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
942 
943 	kobject_put(&policy->kobj);
944 	wait_for_completion(&policy->kobj_unregister);
945 
946 err_out_driver_exit:
947 	if (cpufreq_driver->exit)
948 		cpufreq_driver->exit(policy);
949 
950 err_out:
951 	kfree(policy);
952 
953 nomem_out:
954 	module_put(cpufreq_driver->owner);
955 module_out:
956 	cpufreq_debug_enable_ratelimit();
957 	return ret;
958 }
959 
960 
961 /**
962  * __cpufreq_remove_dev - remove a CPU device
963  *
964  * Removes the cpufreq interface for a CPU device.
965  * Caller should already have policy_rwsem in write mode for this CPU.
966  * This routine frees the rwsem before returning.
967  */
968 static int __cpufreq_remove_dev (struct sys_device * sys_dev)
969 {
970 	unsigned int cpu = sys_dev->id;
971 	unsigned long flags;
972 	struct cpufreq_policy *data;
973 #ifdef CONFIG_SMP
974 	struct sys_device *cpu_sys_dev;
975 	unsigned int j;
976 #endif
977 
978 	cpufreq_debug_disable_ratelimit();
979 	dprintk("unregistering CPU %u\n", cpu);
980 
981 	spin_lock_irqsave(&cpufreq_driver_lock, flags);
982 	data = cpufreq_cpu_data[cpu];
983 
984 	if (!data) {
985 		spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
986 		cpufreq_debug_enable_ratelimit();
987 		unlock_policy_rwsem_write(cpu);
988 		return -EINVAL;
989 	}
990 	cpufreq_cpu_data[cpu] = NULL;
991 
992 
993 #ifdef CONFIG_SMP
994 	/* if this isn't the CPU which is the parent of the kobj, we
995 	 * only need to unlink, put and exit
996 	 */
997 	if (unlikely(cpu != data->cpu)) {
998 		dprintk("removing link\n");
999 		cpu_clear(cpu, data->cpus);
1000 		spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1001 		sysfs_remove_link(&sys_dev->kobj, "cpufreq");
1002 		cpufreq_cpu_put(data);
1003 		cpufreq_debug_enable_ratelimit();
1004 		unlock_policy_rwsem_write(cpu);
1005 		return 0;
1006 	}
1007 #endif
1008 
1009 
1010 	if (!kobject_get(&data->kobj)) {
1011 		spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1012 		cpufreq_debug_enable_ratelimit();
1013 		unlock_policy_rwsem_write(cpu);
1014 		return -EFAULT;
1015 	}
1016 
1017 #ifdef CONFIG_SMP
1018 
1019 #ifdef CONFIG_HOTPLUG_CPU
1020 	cpufreq_cpu_governor[cpu] = data->governor;
1021 #endif
1022 
1023 	/* if we have other CPUs still registered, we need to unlink them,
1024 	 * or else wait_for_completion below will lock up. Clean the
1025 	 * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
1026 	 * links afterwards.
1027 	 */
1028 	if (unlikely(cpus_weight(data->cpus) > 1)) {
1029 		for_each_cpu_mask(j, data->cpus) {
1030 			if (j == cpu)
1031 				continue;
1032 			cpufreq_cpu_data[j] = NULL;
1033 		}
1034 	}
1035 
1036 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1037 
1038 	if (unlikely(cpus_weight(data->cpus) > 1)) {
1039 		for_each_cpu_mask(j, data->cpus) {
1040 			if (j == cpu)
1041 				continue;
1042 			dprintk("removing link for cpu %u\n", j);
1043 #ifdef CONFIG_HOTPLUG_CPU
1044 			cpufreq_cpu_governor[j] = data->governor;
1045 #endif
1046 			cpu_sys_dev = get_cpu_sysdev(j);
1047 			sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
1048 			cpufreq_cpu_put(data);
1049 		}
1050 	}
1051 #else
1052 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1053 #endif
1054 
1055 	if (cpufreq_driver->target)
1056 		__cpufreq_governor(data, CPUFREQ_GOV_STOP);
1057 
1058 	unlock_policy_rwsem_write(cpu);
1059 
1060 	kobject_put(&data->kobj);
1061 
1062 	/* we need to make sure that the underlying kobj is actually
1063 	 * not referenced anymore by anybody before we proceed with
1064 	 * unloading.
1065 	 */
1066 	dprintk("waiting for dropping of refcount\n");
1067 	wait_for_completion(&data->kobj_unregister);
1068 	dprintk("wait complete\n");
1069 
1070 	if (cpufreq_driver->exit)
1071 		cpufreq_driver->exit(data);
1072 
1073 	kfree(data);
1074 
1075 	cpufreq_debug_enable_ratelimit();
1076 	return 0;
1077 }
1078 
1079 
1080 static int cpufreq_remove_dev (struct sys_device * sys_dev)
1081 {
1082 	unsigned int cpu = sys_dev->id;
1083 	int retval;
1084 
1085 	if (cpu_is_offline(cpu))
1086 		return 0;
1087 
1088 	if (unlikely(lock_policy_rwsem_write(cpu)))
1089 		BUG();
1090 
1091 	retval = __cpufreq_remove_dev(sys_dev);
1092 	return retval;
1093 }
1094 
1095 
1096 static void handle_update(struct work_struct *work)
1097 {
1098 	struct cpufreq_policy *policy =
1099 		container_of(work, struct cpufreq_policy, update);
1100 	unsigned int cpu = policy->cpu;
1101 	dprintk("handle_update for cpu %u called\n", cpu);
1102 	cpufreq_update_policy(cpu);
1103 }
1104 
1105 /**
1106  *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1107  *	@cpu: cpu number
1108  *	@old_freq: CPU frequency the kernel thinks the CPU runs at
1109  *	@new_freq: CPU frequency the CPU actually runs at
1110  *
1111  *	We adjust to current frequency first, and need to clean up later. So either call
1112  *	to cpufreq_update_policy() or schedule handle_update()).
1113  */
1114 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1115 				unsigned int new_freq)
1116 {
1117 	struct cpufreq_freqs freqs;
1118 
1119 	dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
1120 	       "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1121 
1122 	freqs.cpu = cpu;
1123 	freqs.old = old_freq;
1124 	freqs.new = new_freq;
1125 	cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1126 	cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1127 }
1128 
1129 
1130 /**
1131  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1132  * @cpu: CPU number
1133  *
1134  * This is the last known freq, without actually getting it from the driver.
1135  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1136  */
1137 unsigned int cpufreq_quick_get(unsigned int cpu)
1138 {
1139 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1140 	unsigned int ret_freq = 0;
1141 
1142 	if (policy) {
1143 		ret_freq = policy->cur;
1144 		cpufreq_cpu_put(policy);
1145 	}
1146 
1147 	return (ret_freq);
1148 }
1149 EXPORT_SYMBOL(cpufreq_quick_get);
1150 
1151 
1152 static unsigned int __cpufreq_get(unsigned int cpu)
1153 {
1154 	struct cpufreq_policy *policy = cpufreq_cpu_data[cpu];
1155 	unsigned int ret_freq = 0;
1156 
1157 	if (!cpufreq_driver->get)
1158 		return (ret_freq);
1159 
1160 	ret_freq = cpufreq_driver->get(cpu);
1161 
1162 	if (ret_freq && policy->cur &&
1163 		!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1164 		/* verify no discrepancy between actual and
1165 					saved value exists */
1166 		if (unlikely(ret_freq != policy->cur)) {
1167 			cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1168 			schedule_work(&policy->update);
1169 		}
1170 	}
1171 
1172 	return (ret_freq);
1173 }
1174 
1175 /**
1176  * cpufreq_get - get the current CPU frequency (in kHz)
1177  * @cpu: CPU number
1178  *
1179  * Get the CPU current (static) CPU frequency
1180  */
1181 unsigned int cpufreq_get(unsigned int cpu)
1182 {
1183 	unsigned int ret_freq = 0;
1184 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1185 
1186 	if (!policy)
1187 		goto out;
1188 
1189 	if (unlikely(lock_policy_rwsem_read(cpu)))
1190 		goto out_policy;
1191 
1192 	ret_freq = __cpufreq_get(cpu);
1193 
1194 	unlock_policy_rwsem_read(cpu);
1195 
1196 out_policy:
1197 	cpufreq_cpu_put(policy);
1198 out:
1199 	return (ret_freq);
1200 }
1201 EXPORT_SYMBOL(cpufreq_get);
1202 
1203 
1204 /**
1205  *	cpufreq_suspend - let the low level driver prepare for suspend
1206  */
1207 
1208 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
1209 {
1210 	int cpu = sysdev->id;
1211 	int ret = 0;
1212 	unsigned int cur_freq = 0;
1213 	struct cpufreq_policy *cpu_policy;
1214 
1215 	dprintk("suspending cpu %u\n", cpu);
1216 
1217 	if (!cpu_online(cpu))
1218 		return 0;
1219 
1220 	/* we may be lax here as interrupts are off. Nonetheless
1221 	 * we need to grab the correct cpu policy, as to check
1222 	 * whether we really run on this CPU.
1223 	 */
1224 
1225 	cpu_policy = cpufreq_cpu_get(cpu);
1226 	if (!cpu_policy)
1227 		return -EINVAL;
1228 
1229 	/* only handle each CPU group once */
1230 	if (unlikely(cpu_policy->cpu != cpu)) {
1231 		cpufreq_cpu_put(cpu_policy);
1232 		return 0;
1233 	}
1234 
1235 	if (cpufreq_driver->suspend) {
1236 		ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1237 		if (ret) {
1238 			printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1239 					"step on CPU %u\n", cpu_policy->cpu);
1240 			cpufreq_cpu_put(cpu_policy);
1241 			return ret;
1242 		}
1243 	}
1244 
1245 
1246 	if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1247 		goto out;
1248 
1249 	if (cpufreq_driver->get)
1250 		cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1251 
1252 	if (!cur_freq || !cpu_policy->cur) {
1253 		printk(KERN_ERR "cpufreq: suspend failed to assert current "
1254 		       "frequency is what timing core thinks it is.\n");
1255 		goto out;
1256 	}
1257 
1258 	if (unlikely(cur_freq != cpu_policy->cur)) {
1259 		struct cpufreq_freqs freqs;
1260 
1261 		if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1262 			dprintk("Warning: CPU frequency is %u, "
1263 			       "cpufreq assumed %u kHz.\n",
1264 			       cur_freq, cpu_policy->cur);
1265 
1266 		freqs.cpu = cpu;
1267 		freqs.old = cpu_policy->cur;
1268 		freqs.new = cur_freq;
1269 
1270 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1271 				    CPUFREQ_SUSPENDCHANGE, &freqs);
1272 		adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1273 
1274 		cpu_policy->cur = cur_freq;
1275 	}
1276 
1277 out:
1278 	cpufreq_cpu_put(cpu_policy);
1279 	return 0;
1280 }
1281 
1282 /**
1283  *	cpufreq_resume -  restore proper CPU frequency handling after resume
1284  *
1285  *	1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1286  *	2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1287  *	3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1288  *	    restored.
1289  */
1290 static int cpufreq_resume(struct sys_device * sysdev)
1291 {
1292 	int cpu = sysdev->id;
1293 	int ret = 0;
1294 	struct cpufreq_policy *cpu_policy;
1295 
1296 	dprintk("resuming cpu %u\n", cpu);
1297 
1298 	if (!cpu_online(cpu))
1299 		return 0;
1300 
1301 	/* we may be lax here as interrupts are off. Nonetheless
1302 	 * we need to grab the correct cpu policy, as to check
1303 	 * whether we really run on this CPU.
1304 	 */
1305 
1306 	cpu_policy = cpufreq_cpu_get(cpu);
1307 	if (!cpu_policy)
1308 		return -EINVAL;
1309 
1310 	/* only handle each CPU group once */
1311 	if (unlikely(cpu_policy->cpu != cpu)) {
1312 		cpufreq_cpu_put(cpu_policy);
1313 		return 0;
1314 	}
1315 
1316 	if (cpufreq_driver->resume) {
1317 		ret = cpufreq_driver->resume(cpu_policy);
1318 		if (ret) {
1319 			printk(KERN_ERR "cpufreq: resume failed in ->resume "
1320 					"step on CPU %u\n", cpu_policy->cpu);
1321 			cpufreq_cpu_put(cpu_policy);
1322 			return ret;
1323 		}
1324 	}
1325 
1326 	if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1327 		unsigned int cur_freq = 0;
1328 
1329 		if (cpufreq_driver->get)
1330 			cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1331 
1332 		if (!cur_freq || !cpu_policy->cur) {
1333 			printk(KERN_ERR "cpufreq: resume failed to assert "
1334 					"current frequency is what timing core "
1335 					"thinks it is.\n");
1336 			goto out;
1337 		}
1338 
1339 		if (unlikely(cur_freq != cpu_policy->cur)) {
1340 			struct cpufreq_freqs freqs;
1341 
1342 			if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1343 				dprintk("Warning: CPU frequency "
1344 				       "is %u, cpufreq assumed %u kHz.\n",
1345 				       cur_freq, cpu_policy->cur);
1346 
1347 			freqs.cpu = cpu;
1348 			freqs.old = cpu_policy->cur;
1349 			freqs.new = cur_freq;
1350 
1351 			srcu_notifier_call_chain(
1352 					&cpufreq_transition_notifier_list,
1353 					CPUFREQ_RESUMECHANGE, &freqs);
1354 			adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1355 
1356 			cpu_policy->cur = cur_freq;
1357 		}
1358 	}
1359 
1360 out:
1361 	schedule_work(&cpu_policy->update);
1362 	cpufreq_cpu_put(cpu_policy);
1363 	return ret;
1364 }
1365 
1366 static struct sysdev_driver cpufreq_sysdev_driver = {
1367 	.add		= cpufreq_add_dev,
1368 	.remove		= cpufreq_remove_dev,
1369 	.suspend	= cpufreq_suspend,
1370 	.resume		= cpufreq_resume,
1371 };
1372 
1373 
1374 /*********************************************************************
1375  *                     NOTIFIER LISTS INTERFACE                      *
1376  *********************************************************************/
1377 
1378 /**
1379  *	cpufreq_register_notifier - register a driver with cpufreq
1380  *	@nb: notifier function to register
1381  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1382  *
1383  *	Add a driver to one of two lists: either a list of drivers that
1384  *      are notified about clock rate changes (once before and once after
1385  *      the transition), or a list of drivers that are notified about
1386  *      changes in cpufreq policy.
1387  *
1388  *	This function may sleep, and has the same return conditions as
1389  *	blocking_notifier_chain_register.
1390  */
1391 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1392 {
1393 	int ret;
1394 
1395 	switch (list) {
1396 	case CPUFREQ_TRANSITION_NOTIFIER:
1397 		ret = srcu_notifier_chain_register(
1398 				&cpufreq_transition_notifier_list, nb);
1399 		break;
1400 	case CPUFREQ_POLICY_NOTIFIER:
1401 		ret = blocking_notifier_chain_register(
1402 				&cpufreq_policy_notifier_list, nb);
1403 		break;
1404 	default:
1405 		ret = -EINVAL;
1406 	}
1407 
1408 	return ret;
1409 }
1410 EXPORT_SYMBOL(cpufreq_register_notifier);
1411 
1412 
1413 /**
1414  *	cpufreq_unregister_notifier - unregister a driver with cpufreq
1415  *	@nb: notifier block to be unregistered
1416  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1417  *
1418  *	Remove a driver from the CPU frequency notifier list.
1419  *
1420  *	This function may sleep, and has the same return conditions as
1421  *	blocking_notifier_chain_unregister.
1422  */
1423 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1424 {
1425 	int ret;
1426 
1427 	switch (list) {
1428 	case CPUFREQ_TRANSITION_NOTIFIER:
1429 		ret = srcu_notifier_chain_unregister(
1430 				&cpufreq_transition_notifier_list, nb);
1431 		break;
1432 	case CPUFREQ_POLICY_NOTIFIER:
1433 		ret = blocking_notifier_chain_unregister(
1434 				&cpufreq_policy_notifier_list, nb);
1435 		break;
1436 	default:
1437 		ret = -EINVAL;
1438 	}
1439 
1440 	return ret;
1441 }
1442 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1443 
1444 
1445 /*********************************************************************
1446  *                              GOVERNORS                            *
1447  *********************************************************************/
1448 
1449 
1450 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1451 			    unsigned int target_freq,
1452 			    unsigned int relation)
1453 {
1454 	int retval = -EINVAL;
1455 
1456 	dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1457 		target_freq, relation);
1458 	if (cpu_online(policy->cpu) && cpufreq_driver->target)
1459 		retval = cpufreq_driver->target(policy, target_freq, relation);
1460 
1461 	return retval;
1462 }
1463 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1464 
1465 int cpufreq_driver_target(struct cpufreq_policy *policy,
1466 			  unsigned int target_freq,
1467 			  unsigned int relation)
1468 {
1469 	int ret;
1470 
1471 	policy = cpufreq_cpu_get(policy->cpu);
1472 	if (!policy)
1473 		return -EINVAL;
1474 
1475 	if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1476 		return -EINVAL;
1477 
1478 	ret = __cpufreq_driver_target(policy, target_freq, relation);
1479 
1480 	unlock_policy_rwsem_write(policy->cpu);
1481 
1482 	cpufreq_cpu_put(policy);
1483 	return ret;
1484 }
1485 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1486 
1487 int __cpufreq_driver_getavg(struct cpufreq_policy *policy)
1488 {
1489 	int ret = 0;
1490 
1491 	policy = cpufreq_cpu_get(policy->cpu);
1492 	if (!policy)
1493 		return -EINVAL;
1494 
1495 	if (cpu_online(policy->cpu) && cpufreq_driver->getavg)
1496 		ret = cpufreq_driver->getavg(policy->cpu);
1497 
1498 	cpufreq_cpu_put(policy);
1499 	return ret;
1500 }
1501 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1502 
1503 /*
1504  * when "event" is CPUFREQ_GOV_LIMITS
1505  */
1506 
1507 static int __cpufreq_governor(struct cpufreq_policy *policy,
1508 					unsigned int event)
1509 {
1510 	int ret;
1511 
1512 	/* Only must be defined when default governor is known to have latency
1513 	   restrictions, like e.g. conservative or ondemand.
1514 	   That this is the case is already ensured in Kconfig
1515 	*/
1516 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1517 	struct cpufreq_governor *gov = &cpufreq_gov_performance;
1518 #else
1519 	struct cpufreq_governor *gov = NULL;
1520 #endif
1521 
1522 	if (policy->governor->max_transition_latency &&
1523 	    policy->cpuinfo.transition_latency >
1524 	    policy->governor->max_transition_latency) {
1525 		if (!gov)
1526 			return -EINVAL;
1527 		else {
1528 			printk(KERN_WARNING "%s governor failed, too long"
1529 			       " transition latency of HW, fallback"
1530 			       " to %s governor\n",
1531 			       policy->governor->name,
1532 			       gov->name);
1533 			policy->governor = gov;
1534 		}
1535 	}
1536 
1537 	if (!try_module_get(policy->governor->owner))
1538 		return -EINVAL;
1539 
1540 	dprintk("__cpufreq_governor for CPU %u, event %u\n",
1541 						policy->cpu, event);
1542 	ret = policy->governor->governor(policy, event);
1543 
1544 	/* we keep one module reference alive for
1545 			each CPU governed by this CPU */
1546 	if ((event != CPUFREQ_GOV_START) || ret)
1547 		module_put(policy->governor->owner);
1548 	if ((event == CPUFREQ_GOV_STOP) && !ret)
1549 		module_put(policy->governor->owner);
1550 
1551 	return ret;
1552 }
1553 
1554 
1555 int cpufreq_register_governor(struct cpufreq_governor *governor)
1556 {
1557 	int err;
1558 
1559 	if (!governor)
1560 		return -EINVAL;
1561 
1562 	mutex_lock(&cpufreq_governor_mutex);
1563 
1564 	err = -EBUSY;
1565 	if (__find_governor(governor->name) == NULL) {
1566 		err = 0;
1567 		list_add(&governor->governor_list, &cpufreq_governor_list);
1568 	}
1569 
1570 	mutex_unlock(&cpufreq_governor_mutex);
1571 	return err;
1572 }
1573 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1574 
1575 
1576 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1577 {
1578 	if (!governor)
1579 		return;
1580 
1581 	mutex_lock(&cpufreq_governor_mutex);
1582 	list_del(&governor->governor_list);
1583 	mutex_unlock(&cpufreq_governor_mutex);
1584 	return;
1585 }
1586 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1587 
1588 
1589 
1590 /*********************************************************************
1591  *                          POLICY INTERFACE                         *
1592  *********************************************************************/
1593 
1594 /**
1595  * cpufreq_get_policy - get the current cpufreq_policy
1596  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1597  *
1598  * Reads the current cpufreq policy.
1599  */
1600 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1601 {
1602 	struct cpufreq_policy *cpu_policy;
1603 	if (!policy)
1604 		return -EINVAL;
1605 
1606 	cpu_policy = cpufreq_cpu_get(cpu);
1607 	if (!cpu_policy)
1608 		return -EINVAL;
1609 
1610 	memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1611 
1612 	cpufreq_cpu_put(cpu_policy);
1613 	return 0;
1614 }
1615 EXPORT_SYMBOL(cpufreq_get_policy);
1616 
1617 
1618 /*
1619  * data   : current policy.
1620  * policy : policy to be set.
1621  */
1622 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1623 				struct cpufreq_policy *policy)
1624 {
1625 	int ret = 0;
1626 
1627 	cpufreq_debug_disable_ratelimit();
1628 	dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1629 		policy->min, policy->max);
1630 
1631 	memcpy(&policy->cpuinfo, &data->cpuinfo,
1632 				sizeof(struct cpufreq_cpuinfo));
1633 
1634 	if (policy->min > data->max || policy->max < data->min) {
1635 		ret = -EINVAL;
1636 		goto error_out;
1637 	}
1638 
1639 	/* verify the cpu speed can be set within this limit */
1640 	ret = cpufreq_driver->verify(policy);
1641 	if (ret)
1642 		goto error_out;
1643 
1644 	/* adjust if necessary - all reasons */
1645 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1646 			CPUFREQ_ADJUST, policy);
1647 
1648 	/* adjust if necessary - hardware incompatibility*/
1649 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1650 			CPUFREQ_INCOMPATIBLE, policy);
1651 
1652 	/* verify the cpu speed can be set within this limit,
1653 	   which might be different to the first one */
1654 	ret = cpufreq_driver->verify(policy);
1655 	if (ret)
1656 		goto error_out;
1657 
1658 	/* notification of the new policy */
1659 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1660 			CPUFREQ_NOTIFY, policy);
1661 
1662 	data->min = policy->min;
1663 	data->max = policy->max;
1664 
1665 	dprintk("new min and max freqs are %u - %u kHz\n",
1666 					data->min, data->max);
1667 
1668 	if (cpufreq_driver->setpolicy) {
1669 		data->policy = policy->policy;
1670 		dprintk("setting range\n");
1671 		ret = cpufreq_driver->setpolicy(policy);
1672 	} else {
1673 		if (policy->governor != data->governor) {
1674 			/* save old, working values */
1675 			struct cpufreq_governor *old_gov = data->governor;
1676 
1677 			dprintk("governor switch\n");
1678 
1679 			/* end old governor */
1680 			if (data->governor)
1681 				__cpufreq_governor(data, CPUFREQ_GOV_STOP);
1682 
1683 			/* start new governor */
1684 			data->governor = policy->governor;
1685 			if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1686 				/* new governor failed, so re-start old one */
1687 				dprintk("starting governor %s failed\n",
1688 							data->governor->name);
1689 				if (old_gov) {
1690 					data->governor = old_gov;
1691 					__cpufreq_governor(data,
1692 							   CPUFREQ_GOV_START);
1693 				}
1694 				ret = -EINVAL;
1695 				goto error_out;
1696 			}
1697 			/* might be a policy change, too, so fall through */
1698 		}
1699 		dprintk("governor: change or update limits\n");
1700 		__cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1701 	}
1702 
1703 error_out:
1704 	cpufreq_debug_enable_ratelimit();
1705 	return ret;
1706 }
1707 
1708 /**
1709  *	cpufreq_update_policy - re-evaluate an existing cpufreq policy
1710  *	@cpu: CPU which shall be re-evaluated
1711  *
1712  *	Usefull for policy notifiers which have different necessities
1713  *	at different times.
1714  */
1715 int cpufreq_update_policy(unsigned int cpu)
1716 {
1717 	struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1718 	struct cpufreq_policy policy;
1719 	int ret = 0;
1720 
1721 	if (!data)
1722 		return -ENODEV;
1723 
1724 	if (unlikely(lock_policy_rwsem_write(cpu)))
1725 		return -EINVAL;
1726 
1727 	dprintk("updating policy for CPU %u\n", cpu);
1728 	memcpy(&policy, data, sizeof(struct cpufreq_policy));
1729 	policy.min = data->user_policy.min;
1730 	policy.max = data->user_policy.max;
1731 	policy.policy = data->user_policy.policy;
1732 	policy.governor = data->user_policy.governor;
1733 
1734 	/* BIOS might change freq behind our back
1735 	  -> ask driver for current freq and notify governors about a change */
1736 	if (cpufreq_driver->get) {
1737 		policy.cur = cpufreq_driver->get(cpu);
1738 		if (!data->cur) {
1739 			dprintk("Driver did not initialize current freq");
1740 			data->cur = policy.cur;
1741 		} else {
1742 			if (data->cur != policy.cur)
1743 				cpufreq_out_of_sync(cpu, data->cur,
1744 								policy.cur);
1745 		}
1746 	}
1747 
1748 	ret = __cpufreq_set_policy(data, &policy);
1749 
1750 	unlock_policy_rwsem_write(cpu);
1751 
1752 	cpufreq_cpu_put(data);
1753 	return ret;
1754 }
1755 EXPORT_SYMBOL(cpufreq_update_policy);
1756 
1757 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1758 					unsigned long action, void *hcpu)
1759 {
1760 	unsigned int cpu = (unsigned long)hcpu;
1761 	struct sys_device *sys_dev;
1762 
1763 	sys_dev = get_cpu_sysdev(cpu);
1764 	if (sys_dev) {
1765 		switch (action) {
1766 		case CPU_ONLINE:
1767 		case CPU_ONLINE_FROZEN:
1768 			cpufreq_add_dev(sys_dev);
1769 			break;
1770 		case CPU_DOWN_PREPARE:
1771 		case CPU_DOWN_PREPARE_FROZEN:
1772 			if (unlikely(lock_policy_rwsem_write(cpu)))
1773 				BUG();
1774 
1775 			__cpufreq_remove_dev(sys_dev);
1776 			break;
1777 		case CPU_DOWN_FAILED:
1778 		case CPU_DOWN_FAILED_FROZEN:
1779 			cpufreq_add_dev(sys_dev);
1780 			break;
1781 		}
1782 	}
1783 	return NOTIFY_OK;
1784 }
1785 
1786 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1787 {
1788     .notifier_call = cpufreq_cpu_callback,
1789 };
1790 
1791 /*********************************************************************
1792  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1793  *********************************************************************/
1794 
1795 /**
1796  * cpufreq_register_driver - register a CPU Frequency driver
1797  * @driver_data: A struct cpufreq_driver containing the values#
1798  * submitted by the CPU Frequency driver.
1799  *
1800  *   Registers a CPU Frequency driver to this core code. This code
1801  * returns zero on success, -EBUSY when another driver got here first
1802  * (and isn't unregistered in the meantime).
1803  *
1804  */
1805 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1806 {
1807 	unsigned long flags;
1808 	int ret;
1809 
1810 	if (!driver_data || !driver_data->verify || !driver_data->init ||
1811 	    ((!driver_data->setpolicy) && (!driver_data->target)))
1812 		return -EINVAL;
1813 
1814 	dprintk("trying to register driver %s\n", driver_data->name);
1815 
1816 	if (driver_data->setpolicy)
1817 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
1818 
1819 	spin_lock_irqsave(&cpufreq_driver_lock, flags);
1820 	if (cpufreq_driver) {
1821 		spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1822 		return -EBUSY;
1823 	}
1824 	cpufreq_driver = driver_data;
1825 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1826 
1827 	ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1828 
1829 	if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1830 		int i;
1831 		ret = -ENODEV;
1832 
1833 		/* check for at least one working CPU */
1834 		for (i=0; i<NR_CPUS; i++)
1835 			if (cpufreq_cpu_data[i])
1836 				ret = 0;
1837 
1838 		/* if all ->init() calls failed, unregister */
1839 		if (ret) {
1840 			dprintk("no CPU initialized for driver %s\n",
1841 							driver_data->name);
1842 			sysdev_driver_unregister(&cpu_sysdev_class,
1843 						&cpufreq_sysdev_driver);
1844 
1845 			spin_lock_irqsave(&cpufreq_driver_lock, flags);
1846 			cpufreq_driver = NULL;
1847 			spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1848 		}
1849 	}
1850 
1851 	if (!ret) {
1852 		register_hotcpu_notifier(&cpufreq_cpu_notifier);
1853 		dprintk("driver %s up and running\n", driver_data->name);
1854 		cpufreq_debug_enable_ratelimit();
1855 	}
1856 
1857 	return (ret);
1858 }
1859 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1860 
1861 
1862 /**
1863  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1864  *
1865  *    Unregister the current CPUFreq driver. Only call this if you have
1866  * the right to do so, i.e. if you have succeeded in initialising before!
1867  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1868  * currently not initialised.
1869  */
1870 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1871 {
1872 	unsigned long flags;
1873 
1874 	cpufreq_debug_disable_ratelimit();
1875 
1876 	if (!cpufreq_driver || (driver != cpufreq_driver)) {
1877 		cpufreq_debug_enable_ratelimit();
1878 		return -EINVAL;
1879 	}
1880 
1881 	dprintk("unregistering driver %s\n", driver->name);
1882 
1883 	sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1884 	unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1885 
1886 	spin_lock_irqsave(&cpufreq_driver_lock, flags);
1887 	cpufreq_driver = NULL;
1888 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1889 
1890 	return 0;
1891 }
1892 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1893 
1894 static int __init cpufreq_core_init(void)
1895 {
1896 	int cpu;
1897 
1898 	for_each_possible_cpu(cpu) {
1899 		per_cpu(policy_cpu, cpu) = -1;
1900 		init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1901 	}
1902 	return 0;
1903 }
1904 
1905 core_initcall(cpufreq_core_init);
1906