xref: /openbmc/linux/drivers/cpufreq/cpufreq.c (revision eb3fcf00)
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
7  *
8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
9  *	Added handling for CPU hotplug
10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
11  *	Fix handling for CPU hotplug -- affected CPUs
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License version 2 as
15  * published by the Free Software Foundation.
16  */
17 
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 
20 #include <linux/cpu.h>
21 #include <linux/cpufreq.h>
22 #include <linux/delay.h>
23 #include <linux/device.h>
24 #include <linux/init.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/module.h>
27 #include <linux/mutex.h>
28 #include <linux/slab.h>
29 #include <linux/suspend.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/tick.h>
32 #include <trace/events/power.h>
33 
34 static LIST_HEAD(cpufreq_policy_list);
35 
36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
37 {
38 	return cpumask_empty(policy->cpus);
39 }
40 
41 static bool suitable_policy(struct cpufreq_policy *policy, bool active)
42 {
43 	return active == !policy_is_inactive(policy);
44 }
45 
46 /* Finds Next Acive/Inactive policy */
47 static struct cpufreq_policy *next_policy(struct cpufreq_policy *policy,
48 					  bool active)
49 {
50 	do {
51 		policy = list_next_entry(policy, policy_list);
52 
53 		/* No more policies in the list */
54 		if (&policy->policy_list == &cpufreq_policy_list)
55 			return NULL;
56 	} while (!suitable_policy(policy, active));
57 
58 	return policy;
59 }
60 
61 static struct cpufreq_policy *first_policy(bool active)
62 {
63 	struct cpufreq_policy *policy;
64 
65 	/* No policies in the list */
66 	if (list_empty(&cpufreq_policy_list))
67 		return NULL;
68 
69 	policy = list_first_entry(&cpufreq_policy_list, typeof(*policy),
70 				  policy_list);
71 
72 	if (!suitable_policy(policy, active))
73 		policy = next_policy(policy, active);
74 
75 	return policy;
76 }
77 
78 /* Macros to iterate over CPU policies */
79 #define for_each_suitable_policy(__policy, __active)	\
80 	for (__policy = first_policy(__active);		\
81 	     __policy;					\
82 	     __policy = next_policy(__policy, __active))
83 
84 #define for_each_active_policy(__policy)		\
85 	for_each_suitable_policy(__policy, true)
86 #define for_each_inactive_policy(__policy)		\
87 	for_each_suitable_policy(__policy, false)
88 
89 #define for_each_policy(__policy)			\
90 	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
91 
92 /* Iterate over governors */
93 static LIST_HEAD(cpufreq_governor_list);
94 #define for_each_governor(__governor)				\
95 	list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
96 
97 /**
98  * The "cpufreq driver" - the arch- or hardware-dependent low
99  * level driver of CPUFreq support, and its spinlock. This lock
100  * also protects the cpufreq_cpu_data array.
101  */
102 static struct cpufreq_driver *cpufreq_driver;
103 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
104 static DEFINE_RWLOCK(cpufreq_driver_lock);
105 DEFINE_MUTEX(cpufreq_governor_lock);
106 
107 /* Flag to suspend/resume CPUFreq governors */
108 static bool cpufreq_suspended;
109 
110 static inline bool has_target(void)
111 {
112 	return cpufreq_driver->target_index || cpufreq_driver->target;
113 }
114 
115 /* internal prototypes */
116 static int __cpufreq_governor(struct cpufreq_policy *policy,
117 		unsigned int event);
118 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
119 static void handle_update(struct work_struct *work);
120 
121 /**
122  * Two notifier lists: the "policy" list is involved in the
123  * validation process for a new CPU frequency policy; the
124  * "transition" list for kernel code that needs to handle
125  * changes to devices when the CPU clock speed changes.
126  * The mutex locks both lists.
127  */
128 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
129 static struct srcu_notifier_head cpufreq_transition_notifier_list;
130 
131 static bool init_cpufreq_transition_notifier_list_called;
132 static int __init init_cpufreq_transition_notifier_list(void)
133 {
134 	srcu_init_notifier_head(&cpufreq_transition_notifier_list);
135 	init_cpufreq_transition_notifier_list_called = true;
136 	return 0;
137 }
138 pure_initcall(init_cpufreq_transition_notifier_list);
139 
140 static int off __read_mostly;
141 static int cpufreq_disabled(void)
142 {
143 	return off;
144 }
145 void disable_cpufreq(void)
146 {
147 	off = 1;
148 }
149 static DEFINE_MUTEX(cpufreq_governor_mutex);
150 
151 bool have_governor_per_policy(void)
152 {
153 	return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
154 }
155 EXPORT_SYMBOL_GPL(have_governor_per_policy);
156 
157 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
158 {
159 	if (have_governor_per_policy())
160 		return &policy->kobj;
161 	else
162 		return cpufreq_global_kobject;
163 }
164 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
165 
166 struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu)
167 {
168 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
169 
170 	return policy && !policy_is_inactive(policy) ?
171 		policy->freq_table : NULL;
172 }
173 EXPORT_SYMBOL_GPL(cpufreq_frequency_get_table);
174 
175 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
176 {
177 	u64 idle_time;
178 	u64 cur_wall_time;
179 	u64 busy_time;
180 
181 	cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
182 
183 	busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
184 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
185 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
186 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
187 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
188 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
189 
190 	idle_time = cur_wall_time - busy_time;
191 	if (wall)
192 		*wall = cputime_to_usecs(cur_wall_time);
193 
194 	return cputime_to_usecs(idle_time);
195 }
196 
197 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
198 {
199 	u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
200 
201 	if (idle_time == -1ULL)
202 		return get_cpu_idle_time_jiffy(cpu, wall);
203 	else if (!io_busy)
204 		idle_time += get_cpu_iowait_time_us(cpu, wall);
205 
206 	return idle_time;
207 }
208 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
209 
210 /*
211  * This is a generic cpufreq init() routine which can be used by cpufreq
212  * drivers of SMP systems. It will do following:
213  * - validate & show freq table passed
214  * - set policies transition latency
215  * - policy->cpus with all possible CPUs
216  */
217 int cpufreq_generic_init(struct cpufreq_policy *policy,
218 		struct cpufreq_frequency_table *table,
219 		unsigned int transition_latency)
220 {
221 	int ret;
222 
223 	ret = cpufreq_table_validate_and_show(policy, table);
224 	if (ret) {
225 		pr_err("%s: invalid frequency table: %d\n", __func__, ret);
226 		return ret;
227 	}
228 
229 	policy->cpuinfo.transition_latency = transition_latency;
230 
231 	/*
232 	 * The driver only supports the SMP configuration where all processors
233 	 * share the clock and voltage and clock.
234 	 */
235 	cpumask_setall(policy->cpus);
236 
237 	return 0;
238 }
239 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
240 
241 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
242 {
243 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
244 
245 	return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
246 }
247 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
248 
249 unsigned int cpufreq_generic_get(unsigned int cpu)
250 {
251 	struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
252 
253 	if (!policy || IS_ERR(policy->clk)) {
254 		pr_err("%s: No %s associated to cpu: %d\n",
255 		       __func__, policy ? "clk" : "policy", cpu);
256 		return 0;
257 	}
258 
259 	return clk_get_rate(policy->clk) / 1000;
260 }
261 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
262 
263 /**
264  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
265  *
266  * @cpu: cpu to find policy for.
267  *
268  * This returns policy for 'cpu', returns NULL if it doesn't exist.
269  * It also increments the kobject reference count to mark it busy and so would
270  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
271  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
272  * freed as that depends on the kobj count.
273  *
274  * Return: A valid policy on success, otherwise NULL on failure.
275  */
276 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
277 {
278 	struct cpufreq_policy *policy = NULL;
279 	unsigned long flags;
280 
281 	if (WARN_ON(cpu >= nr_cpu_ids))
282 		return NULL;
283 
284 	/* get the cpufreq driver */
285 	read_lock_irqsave(&cpufreq_driver_lock, flags);
286 
287 	if (cpufreq_driver) {
288 		/* get the CPU */
289 		policy = cpufreq_cpu_get_raw(cpu);
290 		if (policy)
291 			kobject_get(&policy->kobj);
292 	}
293 
294 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
295 
296 	return policy;
297 }
298 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
299 
300 /**
301  * cpufreq_cpu_put: Decrements the usage count of a policy
302  *
303  * @policy: policy earlier returned by cpufreq_cpu_get().
304  *
305  * This decrements the kobject reference count incremented earlier by calling
306  * cpufreq_cpu_get().
307  */
308 void cpufreq_cpu_put(struct cpufreq_policy *policy)
309 {
310 	kobject_put(&policy->kobj);
311 }
312 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
313 
314 /*********************************************************************
315  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
316  *********************************************************************/
317 
318 /**
319  * adjust_jiffies - adjust the system "loops_per_jiffy"
320  *
321  * This function alters the system "loops_per_jiffy" for the clock
322  * speed change. Note that loops_per_jiffy cannot be updated on SMP
323  * systems as each CPU might be scaled differently. So, use the arch
324  * per-CPU loops_per_jiffy value wherever possible.
325  */
326 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
327 {
328 #ifndef CONFIG_SMP
329 	static unsigned long l_p_j_ref;
330 	static unsigned int l_p_j_ref_freq;
331 
332 	if (ci->flags & CPUFREQ_CONST_LOOPS)
333 		return;
334 
335 	if (!l_p_j_ref_freq) {
336 		l_p_j_ref = loops_per_jiffy;
337 		l_p_j_ref_freq = ci->old;
338 		pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
339 			 l_p_j_ref, l_p_j_ref_freq);
340 	}
341 	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
342 		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
343 								ci->new);
344 		pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
345 			 loops_per_jiffy, ci->new);
346 	}
347 #endif
348 }
349 
350 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
351 		struct cpufreq_freqs *freqs, unsigned int state)
352 {
353 	BUG_ON(irqs_disabled());
354 
355 	if (cpufreq_disabled())
356 		return;
357 
358 	freqs->flags = cpufreq_driver->flags;
359 	pr_debug("notification %u of frequency transition to %u kHz\n",
360 		 state, freqs->new);
361 
362 	switch (state) {
363 
364 	case CPUFREQ_PRECHANGE:
365 		/* detect if the driver reported a value as "old frequency"
366 		 * which is not equal to what the cpufreq core thinks is
367 		 * "old frequency".
368 		 */
369 		if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
370 			if ((policy) && (policy->cpu == freqs->cpu) &&
371 			    (policy->cur) && (policy->cur != freqs->old)) {
372 				pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
373 					 freqs->old, policy->cur);
374 				freqs->old = policy->cur;
375 			}
376 		}
377 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
378 				CPUFREQ_PRECHANGE, freqs);
379 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
380 		break;
381 
382 	case CPUFREQ_POSTCHANGE:
383 		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
384 		pr_debug("FREQ: %lu - CPU: %lu\n",
385 			 (unsigned long)freqs->new, (unsigned long)freqs->cpu);
386 		trace_cpu_frequency(freqs->new, freqs->cpu);
387 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
388 				CPUFREQ_POSTCHANGE, freqs);
389 		if (likely(policy) && likely(policy->cpu == freqs->cpu))
390 			policy->cur = freqs->new;
391 		break;
392 	}
393 }
394 
395 /**
396  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
397  * on frequency transition.
398  *
399  * This function calls the transition notifiers and the "adjust_jiffies"
400  * function. It is called twice on all CPU frequency changes that have
401  * external effects.
402  */
403 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
404 		struct cpufreq_freqs *freqs, unsigned int state)
405 {
406 	for_each_cpu(freqs->cpu, policy->cpus)
407 		__cpufreq_notify_transition(policy, freqs, state);
408 }
409 
410 /* Do post notifications when there are chances that transition has failed */
411 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
412 		struct cpufreq_freqs *freqs, int transition_failed)
413 {
414 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
415 	if (!transition_failed)
416 		return;
417 
418 	swap(freqs->old, freqs->new);
419 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
420 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
421 }
422 
423 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
424 		struct cpufreq_freqs *freqs)
425 {
426 
427 	/*
428 	 * Catch double invocations of _begin() which lead to self-deadlock.
429 	 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
430 	 * doesn't invoke _begin() on their behalf, and hence the chances of
431 	 * double invocations are very low. Moreover, there are scenarios
432 	 * where these checks can emit false-positive warnings in these
433 	 * drivers; so we avoid that by skipping them altogether.
434 	 */
435 	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
436 				&& current == policy->transition_task);
437 
438 wait:
439 	wait_event(policy->transition_wait, !policy->transition_ongoing);
440 
441 	spin_lock(&policy->transition_lock);
442 
443 	if (unlikely(policy->transition_ongoing)) {
444 		spin_unlock(&policy->transition_lock);
445 		goto wait;
446 	}
447 
448 	policy->transition_ongoing = true;
449 	policy->transition_task = current;
450 
451 	spin_unlock(&policy->transition_lock);
452 
453 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
454 }
455 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
456 
457 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
458 		struct cpufreq_freqs *freqs, int transition_failed)
459 {
460 	if (unlikely(WARN_ON(!policy->transition_ongoing)))
461 		return;
462 
463 	cpufreq_notify_post_transition(policy, freqs, transition_failed);
464 
465 	policy->transition_ongoing = false;
466 	policy->transition_task = NULL;
467 
468 	wake_up(&policy->transition_wait);
469 }
470 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
471 
472 
473 /*********************************************************************
474  *                          SYSFS INTERFACE                          *
475  *********************************************************************/
476 static ssize_t show_boost(struct kobject *kobj,
477 				 struct attribute *attr, char *buf)
478 {
479 	return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
480 }
481 
482 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
483 				  const char *buf, size_t count)
484 {
485 	int ret, enable;
486 
487 	ret = sscanf(buf, "%d", &enable);
488 	if (ret != 1 || enable < 0 || enable > 1)
489 		return -EINVAL;
490 
491 	if (cpufreq_boost_trigger_state(enable)) {
492 		pr_err("%s: Cannot %s BOOST!\n",
493 		       __func__, enable ? "enable" : "disable");
494 		return -EINVAL;
495 	}
496 
497 	pr_debug("%s: cpufreq BOOST %s\n",
498 		 __func__, enable ? "enabled" : "disabled");
499 
500 	return count;
501 }
502 define_one_global_rw(boost);
503 
504 static struct cpufreq_governor *find_governor(const char *str_governor)
505 {
506 	struct cpufreq_governor *t;
507 
508 	for_each_governor(t)
509 		if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
510 			return t;
511 
512 	return NULL;
513 }
514 
515 /**
516  * cpufreq_parse_governor - parse a governor string
517  */
518 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
519 				struct cpufreq_governor **governor)
520 {
521 	int err = -EINVAL;
522 
523 	if (cpufreq_driver->setpolicy) {
524 		if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
525 			*policy = CPUFREQ_POLICY_PERFORMANCE;
526 			err = 0;
527 		} else if (!strncasecmp(str_governor, "powersave",
528 						CPUFREQ_NAME_LEN)) {
529 			*policy = CPUFREQ_POLICY_POWERSAVE;
530 			err = 0;
531 		}
532 	} else {
533 		struct cpufreq_governor *t;
534 
535 		mutex_lock(&cpufreq_governor_mutex);
536 
537 		t = find_governor(str_governor);
538 
539 		if (t == NULL) {
540 			int ret;
541 
542 			mutex_unlock(&cpufreq_governor_mutex);
543 			ret = request_module("cpufreq_%s", str_governor);
544 			mutex_lock(&cpufreq_governor_mutex);
545 
546 			if (ret == 0)
547 				t = find_governor(str_governor);
548 		}
549 
550 		if (t != NULL) {
551 			*governor = t;
552 			err = 0;
553 		}
554 
555 		mutex_unlock(&cpufreq_governor_mutex);
556 	}
557 	return err;
558 }
559 
560 /**
561  * cpufreq_per_cpu_attr_read() / show_##file_name() -
562  * print out cpufreq information
563  *
564  * Write out information from cpufreq_driver->policy[cpu]; object must be
565  * "unsigned int".
566  */
567 
568 #define show_one(file_name, object)			\
569 static ssize_t show_##file_name				\
570 (struct cpufreq_policy *policy, char *buf)		\
571 {							\
572 	return sprintf(buf, "%u\n", policy->object);	\
573 }
574 
575 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
576 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
577 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
578 show_one(scaling_min_freq, min);
579 show_one(scaling_max_freq, max);
580 
581 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
582 {
583 	ssize_t ret;
584 
585 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
586 		ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
587 	else
588 		ret = sprintf(buf, "%u\n", policy->cur);
589 	return ret;
590 }
591 
592 static int cpufreq_set_policy(struct cpufreq_policy *policy,
593 				struct cpufreq_policy *new_policy);
594 
595 /**
596  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
597  */
598 #define store_one(file_name, object)			\
599 static ssize_t store_##file_name					\
600 (struct cpufreq_policy *policy, const char *buf, size_t count)		\
601 {									\
602 	int ret, temp;							\
603 	struct cpufreq_policy new_policy;				\
604 									\
605 	memcpy(&new_policy, policy, sizeof(*policy));			\
606 									\
607 	ret = sscanf(buf, "%u", &new_policy.object);			\
608 	if (ret != 1)							\
609 		return -EINVAL;						\
610 									\
611 	temp = new_policy.object;					\
612 	ret = cpufreq_set_policy(policy, &new_policy);		\
613 	if (!ret)							\
614 		policy->user_policy.object = temp;			\
615 									\
616 	return ret ? ret : count;					\
617 }
618 
619 store_one(scaling_min_freq, min);
620 store_one(scaling_max_freq, max);
621 
622 /**
623  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
624  */
625 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
626 					char *buf)
627 {
628 	unsigned int cur_freq = __cpufreq_get(policy);
629 	if (!cur_freq)
630 		return sprintf(buf, "<unknown>");
631 	return sprintf(buf, "%u\n", cur_freq);
632 }
633 
634 /**
635  * show_scaling_governor - show the current policy for the specified CPU
636  */
637 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
638 {
639 	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
640 		return sprintf(buf, "powersave\n");
641 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
642 		return sprintf(buf, "performance\n");
643 	else if (policy->governor)
644 		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
645 				policy->governor->name);
646 	return -EINVAL;
647 }
648 
649 /**
650  * store_scaling_governor - store policy for the specified CPU
651  */
652 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
653 					const char *buf, size_t count)
654 {
655 	int ret;
656 	char	str_governor[16];
657 	struct cpufreq_policy new_policy;
658 
659 	memcpy(&new_policy, policy, sizeof(*policy));
660 
661 	ret = sscanf(buf, "%15s", str_governor);
662 	if (ret != 1)
663 		return -EINVAL;
664 
665 	if (cpufreq_parse_governor(str_governor, &new_policy.policy,
666 						&new_policy.governor))
667 		return -EINVAL;
668 
669 	ret = cpufreq_set_policy(policy, &new_policy);
670 	return ret ? ret : count;
671 }
672 
673 /**
674  * show_scaling_driver - show the cpufreq driver currently loaded
675  */
676 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
677 {
678 	return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
679 }
680 
681 /**
682  * show_scaling_available_governors - show the available CPUfreq governors
683  */
684 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
685 						char *buf)
686 {
687 	ssize_t i = 0;
688 	struct cpufreq_governor *t;
689 
690 	if (!has_target()) {
691 		i += sprintf(buf, "performance powersave");
692 		goto out;
693 	}
694 
695 	for_each_governor(t) {
696 		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
697 		    - (CPUFREQ_NAME_LEN + 2)))
698 			goto out;
699 		i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
700 	}
701 out:
702 	i += sprintf(&buf[i], "\n");
703 	return i;
704 }
705 
706 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
707 {
708 	ssize_t i = 0;
709 	unsigned int cpu;
710 
711 	for_each_cpu(cpu, mask) {
712 		if (i)
713 			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
714 		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
715 		if (i >= (PAGE_SIZE - 5))
716 			break;
717 	}
718 	i += sprintf(&buf[i], "\n");
719 	return i;
720 }
721 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
722 
723 /**
724  * show_related_cpus - show the CPUs affected by each transition even if
725  * hw coordination is in use
726  */
727 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
728 {
729 	return cpufreq_show_cpus(policy->related_cpus, buf);
730 }
731 
732 /**
733  * show_affected_cpus - show the CPUs affected by each transition
734  */
735 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
736 {
737 	return cpufreq_show_cpus(policy->cpus, buf);
738 }
739 
740 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
741 					const char *buf, size_t count)
742 {
743 	unsigned int freq = 0;
744 	unsigned int ret;
745 
746 	if (!policy->governor || !policy->governor->store_setspeed)
747 		return -EINVAL;
748 
749 	ret = sscanf(buf, "%u", &freq);
750 	if (ret != 1)
751 		return -EINVAL;
752 
753 	policy->governor->store_setspeed(policy, freq);
754 
755 	return count;
756 }
757 
758 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
759 {
760 	if (!policy->governor || !policy->governor->show_setspeed)
761 		return sprintf(buf, "<unsupported>\n");
762 
763 	return policy->governor->show_setspeed(policy, buf);
764 }
765 
766 /**
767  * show_bios_limit - show the current cpufreq HW/BIOS limitation
768  */
769 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
770 {
771 	unsigned int limit;
772 	int ret;
773 	if (cpufreq_driver->bios_limit) {
774 		ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
775 		if (!ret)
776 			return sprintf(buf, "%u\n", limit);
777 	}
778 	return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
779 }
780 
781 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
782 cpufreq_freq_attr_ro(cpuinfo_min_freq);
783 cpufreq_freq_attr_ro(cpuinfo_max_freq);
784 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
785 cpufreq_freq_attr_ro(scaling_available_governors);
786 cpufreq_freq_attr_ro(scaling_driver);
787 cpufreq_freq_attr_ro(scaling_cur_freq);
788 cpufreq_freq_attr_ro(bios_limit);
789 cpufreq_freq_attr_ro(related_cpus);
790 cpufreq_freq_attr_ro(affected_cpus);
791 cpufreq_freq_attr_rw(scaling_min_freq);
792 cpufreq_freq_attr_rw(scaling_max_freq);
793 cpufreq_freq_attr_rw(scaling_governor);
794 cpufreq_freq_attr_rw(scaling_setspeed);
795 
796 static struct attribute *default_attrs[] = {
797 	&cpuinfo_min_freq.attr,
798 	&cpuinfo_max_freq.attr,
799 	&cpuinfo_transition_latency.attr,
800 	&scaling_min_freq.attr,
801 	&scaling_max_freq.attr,
802 	&affected_cpus.attr,
803 	&related_cpus.attr,
804 	&scaling_governor.attr,
805 	&scaling_driver.attr,
806 	&scaling_available_governors.attr,
807 	&scaling_setspeed.attr,
808 	NULL
809 };
810 
811 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
812 #define to_attr(a) container_of(a, struct freq_attr, attr)
813 
814 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
815 {
816 	struct cpufreq_policy *policy = to_policy(kobj);
817 	struct freq_attr *fattr = to_attr(attr);
818 	ssize_t ret;
819 
820 	down_read(&policy->rwsem);
821 
822 	if (fattr->show)
823 		ret = fattr->show(policy, buf);
824 	else
825 		ret = -EIO;
826 
827 	up_read(&policy->rwsem);
828 
829 	return ret;
830 }
831 
832 static ssize_t store(struct kobject *kobj, struct attribute *attr,
833 		     const char *buf, size_t count)
834 {
835 	struct cpufreq_policy *policy = to_policy(kobj);
836 	struct freq_attr *fattr = to_attr(attr);
837 	ssize_t ret = -EINVAL;
838 
839 	get_online_cpus();
840 
841 	if (!cpu_online(policy->cpu))
842 		goto unlock;
843 
844 	down_write(&policy->rwsem);
845 
846 	/* Updating inactive policies is invalid, so avoid doing that. */
847 	if (unlikely(policy_is_inactive(policy))) {
848 		ret = -EBUSY;
849 		goto unlock_policy_rwsem;
850 	}
851 
852 	if (fattr->store)
853 		ret = fattr->store(policy, buf, count);
854 	else
855 		ret = -EIO;
856 
857 unlock_policy_rwsem:
858 	up_write(&policy->rwsem);
859 unlock:
860 	put_online_cpus();
861 
862 	return ret;
863 }
864 
865 static void cpufreq_sysfs_release(struct kobject *kobj)
866 {
867 	struct cpufreq_policy *policy = to_policy(kobj);
868 	pr_debug("last reference is dropped\n");
869 	complete(&policy->kobj_unregister);
870 }
871 
872 static const struct sysfs_ops sysfs_ops = {
873 	.show	= show,
874 	.store	= store,
875 };
876 
877 static struct kobj_type ktype_cpufreq = {
878 	.sysfs_ops	= &sysfs_ops,
879 	.default_attrs	= default_attrs,
880 	.release	= cpufreq_sysfs_release,
881 };
882 
883 struct kobject *cpufreq_global_kobject;
884 EXPORT_SYMBOL(cpufreq_global_kobject);
885 
886 static int cpufreq_global_kobject_usage;
887 
888 int cpufreq_get_global_kobject(void)
889 {
890 	if (!cpufreq_global_kobject_usage++)
891 		return kobject_add(cpufreq_global_kobject,
892 				&cpu_subsys.dev_root->kobj, "%s", "cpufreq");
893 
894 	return 0;
895 }
896 EXPORT_SYMBOL(cpufreq_get_global_kobject);
897 
898 void cpufreq_put_global_kobject(void)
899 {
900 	if (!--cpufreq_global_kobject_usage)
901 		kobject_del(cpufreq_global_kobject);
902 }
903 EXPORT_SYMBOL(cpufreq_put_global_kobject);
904 
905 int cpufreq_sysfs_create_file(const struct attribute *attr)
906 {
907 	int ret = cpufreq_get_global_kobject();
908 
909 	if (!ret) {
910 		ret = sysfs_create_file(cpufreq_global_kobject, attr);
911 		if (ret)
912 			cpufreq_put_global_kobject();
913 	}
914 
915 	return ret;
916 }
917 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
918 
919 void cpufreq_sysfs_remove_file(const struct attribute *attr)
920 {
921 	sysfs_remove_file(cpufreq_global_kobject, attr);
922 	cpufreq_put_global_kobject();
923 }
924 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
925 
926 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
927 {
928 	struct device *cpu_dev;
929 
930 	pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
931 
932 	if (!policy)
933 		return 0;
934 
935 	cpu_dev = get_cpu_device(cpu);
936 	if (WARN_ON(!cpu_dev))
937 		return 0;
938 
939 	return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
940 }
941 
942 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
943 {
944 	struct device *cpu_dev;
945 
946 	pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
947 
948 	cpu_dev = get_cpu_device(cpu);
949 	if (WARN_ON(!cpu_dev))
950 		return;
951 
952 	sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
953 }
954 
955 /* Add/remove symlinks for all related CPUs */
956 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
957 {
958 	unsigned int j;
959 	int ret = 0;
960 
961 	/* Some related CPUs might not be present (physically hotplugged) */
962 	for_each_cpu(j, policy->real_cpus) {
963 		if (j == policy->kobj_cpu)
964 			continue;
965 
966 		ret = add_cpu_dev_symlink(policy, j);
967 		if (ret)
968 			break;
969 	}
970 
971 	return ret;
972 }
973 
974 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
975 {
976 	unsigned int j;
977 
978 	/* Some related CPUs might not be present (physically hotplugged) */
979 	for_each_cpu(j, policy->real_cpus) {
980 		if (j == policy->kobj_cpu)
981 			continue;
982 
983 		remove_cpu_dev_symlink(policy, j);
984 	}
985 }
986 
987 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
988 {
989 	struct freq_attr **drv_attr;
990 	int ret = 0;
991 
992 	/* set up files for this cpu device */
993 	drv_attr = cpufreq_driver->attr;
994 	while (drv_attr && *drv_attr) {
995 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
996 		if (ret)
997 			return ret;
998 		drv_attr++;
999 	}
1000 	if (cpufreq_driver->get) {
1001 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1002 		if (ret)
1003 			return ret;
1004 	}
1005 
1006 	ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1007 	if (ret)
1008 		return ret;
1009 
1010 	if (cpufreq_driver->bios_limit) {
1011 		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1012 		if (ret)
1013 			return ret;
1014 	}
1015 
1016 	return cpufreq_add_dev_symlink(policy);
1017 }
1018 
1019 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1020 {
1021 	struct cpufreq_governor *gov = NULL;
1022 	struct cpufreq_policy new_policy;
1023 
1024 	memcpy(&new_policy, policy, sizeof(*policy));
1025 
1026 	/* Update governor of new_policy to the governor used before hotplug */
1027 	gov = find_governor(policy->last_governor);
1028 	if (gov)
1029 		pr_debug("Restoring governor %s for cpu %d\n",
1030 				policy->governor->name, policy->cpu);
1031 	else
1032 		gov = CPUFREQ_DEFAULT_GOVERNOR;
1033 
1034 	new_policy.governor = gov;
1035 
1036 	/* Use the default policy if its valid. */
1037 	if (cpufreq_driver->setpolicy)
1038 		cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
1039 
1040 	/* set default policy */
1041 	return cpufreq_set_policy(policy, &new_policy);
1042 }
1043 
1044 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1045 {
1046 	int ret = 0;
1047 
1048 	/* Has this CPU been taken care of already? */
1049 	if (cpumask_test_cpu(cpu, policy->cpus))
1050 		return 0;
1051 
1052 	if (has_target()) {
1053 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1054 		if (ret) {
1055 			pr_err("%s: Failed to stop governor\n", __func__);
1056 			return ret;
1057 		}
1058 	}
1059 
1060 	down_write(&policy->rwsem);
1061 	cpumask_set_cpu(cpu, policy->cpus);
1062 	up_write(&policy->rwsem);
1063 
1064 	if (has_target()) {
1065 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1066 		if (!ret)
1067 			ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1068 
1069 		if (ret) {
1070 			pr_err("%s: Failed to start governor\n", __func__);
1071 			return ret;
1072 		}
1073 	}
1074 
1075 	return 0;
1076 }
1077 
1078 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1079 {
1080 	struct device *dev = get_cpu_device(cpu);
1081 	struct cpufreq_policy *policy;
1082 	int ret;
1083 
1084 	if (WARN_ON(!dev))
1085 		return NULL;
1086 
1087 	policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1088 	if (!policy)
1089 		return NULL;
1090 
1091 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1092 		goto err_free_policy;
1093 
1094 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1095 		goto err_free_cpumask;
1096 
1097 	if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1098 		goto err_free_rcpumask;
1099 
1100 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &dev->kobj,
1101 				   "cpufreq");
1102 	if (ret) {
1103 		pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1104 		goto err_free_real_cpus;
1105 	}
1106 
1107 	INIT_LIST_HEAD(&policy->policy_list);
1108 	init_rwsem(&policy->rwsem);
1109 	spin_lock_init(&policy->transition_lock);
1110 	init_waitqueue_head(&policy->transition_wait);
1111 	init_completion(&policy->kobj_unregister);
1112 	INIT_WORK(&policy->update, handle_update);
1113 
1114 	policy->cpu = cpu;
1115 
1116 	/* Set this once on allocation */
1117 	policy->kobj_cpu = cpu;
1118 
1119 	return policy;
1120 
1121 err_free_real_cpus:
1122 	free_cpumask_var(policy->real_cpus);
1123 err_free_rcpumask:
1124 	free_cpumask_var(policy->related_cpus);
1125 err_free_cpumask:
1126 	free_cpumask_var(policy->cpus);
1127 err_free_policy:
1128 	kfree(policy);
1129 
1130 	return NULL;
1131 }
1132 
1133 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1134 {
1135 	struct kobject *kobj;
1136 	struct completion *cmp;
1137 
1138 	if (notify)
1139 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1140 					     CPUFREQ_REMOVE_POLICY, policy);
1141 
1142 	down_write(&policy->rwsem);
1143 	cpufreq_remove_dev_symlink(policy);
1144 	kobj = &policy->kobj;
1145 	cmp = &policy->kobj_unregister;
1146 	up_write(&policy->rwsem);
1147 	kobject_put(kobj);
1148 
1149 	/*
1150 	 * We need to make sure that the underlying kobj is
1151 	 * actually not referenced anymore by anybody before we
1152 	 * proceed with unloading.
1153 	 */
1154 	pr_debug("waiting for dropping of refcount\n");
1155 	wait_for_completion(cmp);
1156 	pr_debug("wait complete\n");
1157 }
1158 
1159 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1160 {
1161 	unsigned long flags;
1162 	int cpu;
1163 
1164 	/* Remove policy from list */
1165 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1166 	list_del(&policy->policy_list);
1167 
1168 	for_each_cpu(cpu, policy->related_cpus)
1169 		per_cpu(cpufreq_cpu_data, cpu) = NULL;
1170 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1171 
1172 	cpufreq_policy_put_kobj(policy, notify);
1173 	free_cpumask_var(policy->real_cpus);
1174 	free_cpumask_var(policy->related_cpus);
1175 	free_cpumask_var(policy->cpus);
1176 	kfree(policy);
1177 }
1178 
1179 static int cpufreq_online(unsigned int cpu)
1180 {
1181 	struct cpufreq_policy *policy;
1182 	bool new_policy;
1183 	unsigned long flags;
1184 	unsigned int j;
1185 	int ret;
1186 
1187 	pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1188 
1189 	/* Check if this CPU already has a policy to manage it */
1190 	policy = per_cpu(cpufreq_cpu_data, cpu);
1191 	if (policy) {
1192 		WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1193 		if (!policy_is_inactive(policy))
1194 			return cpufreq_add_policy_cpu(policy, cpu);
1195 
1196 		/* This is the only online CPU for the policy.  Start over. */
1197 		new_policy = false;
1198 		down_write(&policy->rwsem);
1199 		policy->cpu = cpu;
1200 		policy->governor = NULL;
1201 		up_write(&policy->rwsem);
1202 	} else {
1203 		new_policy = true;
1204 		policy = cpufreq_policy_alloc(cpu);
1205 		if (!policy)
1206 			return -ENOMEM;
1207 	}
1208 
1209 	cpumask_copy(policy->cpus, cpumask_of(cpu));
1210 
1211 	/* call driver. From then on the cpufreq must be able
1212 	 * to accept all calls to ->verify and ->setpolicy for this CPU
1213 	 */
1214 	ret = cpufreq_driver->init(policy);
1215 	if (ret) {
1216 		pr_debug("initialization failed\n");
1217 		goto out_free_policy;
1218 	}
1219 
1220 	down_write(&policy->rwsem);
1221 
1222 	if (new_policy) {
1223 		/* related_cpus should at least include policy->cpus. */
1224 		cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1225 		/* Remember CPUs present at the policy creation time. */
1226 		cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1227 	}
1228 
1229 	/*
1230 	 * affected cpus must always be the one, which are online. We aren't
1231 	 * managing offline cpus here.
1232 	 */
1233 	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1234 
1235 	if (new_policy) {
1236 		policy->user_policy.min = policy->min;
1237 		policy->user_policy.max = policy->max;
1238 
1239 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1240 		for_each_cpu(j, policy->related_cpus)
1241 			per_cpu(cpufreq_cpu_data, j) = policy;
1242 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1243 	}
1244 
1245 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1246 		policy->cur = cpufreq_driver->get(policy->cpu);
1247 		if (!policy->cur) {
1248 			pr_err("%s: ->get() failed\n", __func__);
1249 			goto out_exit_policy;
1250 		}
1251 	}
1252 
1253 	/*
1254 	 * Sometimes boot loaders set CPU frequency to a value outside of
1255 	 * frequency table present with cpufreq core. In such cases CPU might be
1256 	 * unstable if it has to run on that frequency for long duration of time
1257 	 * and so its better to set it to a frequency which is specified in
1258 	 * freq-table. This also makes cpufreq stats inconsistent as
1259 	 * cpufreq-stats would fail to register because current frequency of CPU
1260 	 * isn't found in freq-table.
1261 	 *
1262 	 * Because we don't want this change to effect boot process badly, we go
1263 	 * for the next freq which is >= policy->cur ('cur' must be set by now,
1264 	 * otherwise we will end up setting freq to lowest of the table as 'cur'
1265 	 * is initialized to zero).
1266 	 *
1267 	 * We are passing target-freq as "policy->cur - 1" otherwise
1268 	 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1269 	 * equal to target-freq.
1270 	 */
1271 	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1272 	    && has_target()) {
1273 		/* Are we running at unknown frequency ? */
1274 		ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1275 		if (ret == -EINVAL) {
1276 			/* Warn user and fix it */
1277 			pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1278 				__func__, policy->cpu, policy->cur);
1279 			ret = __cpufreq_driver_target(policy, policy->cur - 1,
1280 				CPUFREQ_RELATION_L);
1281 
1282 			/*
1283 			 * Reaching here after boot in a few seconds may not
1284 			 * mean that system will remain stable at "unknown"
1285 			 * frequency for longer duration. Hence, a BUG_ON().
1286 			 */
1287 			BUG_ON(ret);
1288 			pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1289 				__func__, policy->cpu, policy->cur);
1290 		}
1291 	}
1292 
1293 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1294 				     CPUFREQ_START, policy);
1295 
1296 	if (new_policy) {
1297 		ret = cpufreq_add_dev_interface(policy);
1298 		if (ret)
1299 			goto out_exit_policy;
1300 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1301 				CPUFREQ_CREATE_POLICY, policy);
1302 
1303 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1304 		list_add(&policy->policy_list, &cpufreq_policy_list);
1305 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1306 	}
1307 
1308 	ret = cpufreq_init_policy(policy);
1309 	if (ret) {
1310 		pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1311 		       __func__, cpu, ret);
1312 		/* cpufreq_policy_free() will notify based on this */
1313 		new_policy = false;
1314 		goto out_exit_policy;
1315 	}
1316 
1317 	up_write(&policy->rwsem);
1318 
1319 	kobject_uevent(&policy->kobj, KOBJ_ADD);
1320 
1321 	/* Callback for handling stuff after policy is ready */
1322 	if (cpufreq_driver->ready)
1323 		cpufreq_driver->ready(policy);
1324 
1325 	pr_debug("initialization complete\n");
1326 
1327 	return 0;
1328 
1329 out_exit_policy:
1330 	up_write(&policy->rwsem);
1331 
1332 	if (cpufreq_driver->exit)
1333 		cpufreq_driver->exit(policy);
1334 out_free_policy:
1335 	cpufreq_policy_free(policy, !new_policy);
1336 	return ret;
1337 }
1338 
1339 /**
1340  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1341  * @dev: CPU device.
1342  * @sif: Subsystem interface structure pointer (not used)
1343  */
1344 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1345 {
1346 	unsigned cpu = dev->id;
1347 	int ret;
1348 
1349 	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1350 
1351 	if (cpu_online(cpu)) {
1352 		ret = cpufreq_online(cpu);
1353 	} else {
1354 		/*
1355 		 * A hotplug notifier will follow and we will handle it as CPU
1356 		 * online then.  For now, just create the sysfs link, unless
1357 		 * there is no policy or the link is already present.
1358 		 */
1359 		struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1360 
1361 		ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1362 			? add_cpu_dev_symlink(policy, cpu) : 0;
1363 	}
1364 
1365 	return ret;
1366 }
1367 
1368 static void cpufreq_offline_prepare(unsigned int cpu)
1369 {
1370 	struct cpufreq_policy *policy;
1371 
1372 	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1373 
1374 	policy = cpufreq_cpu_get_raw(cpu);
1375 	if (!policy) {
1376 		pr_debug("%s: No cpu_data found\n", __func__);
1377 		return;
1378 	}
1379 
1380 	if (has_target()) {
1381 		int ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1382 		if (ret)
1383 			pr_err("%s: Failed to stop governor\n", __func__);
1384 	}
1385 
1386 	down_write(&policy->rwsem);
1387 	cpumask_clear_cpu(cpu, policy->cpus);
1388 
1389 	if (policy_is_inactive(policy)) {
1390 		if (has_target())
1391 			strncpy(policy->last_governor, policy->governor->name,
1392 				CPUFREQ_NAME_LEN);
1393 	} else if (cpu == policy->cpu) {
1394 		/* Nominate new CPU */
1395 		policy->cpu = cpumask_any(policy->cpus);
1396 	}
1397 	up_write(&policy->rwsem);
1398 
1399 	/* Start governor again for active policy */
1400 	if (!policy_is_inactive(policy)) {
1401 		if (has_target()) {
1402 			int ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1403 			if (!ret)
1404 				ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1405 
1406 			if (ret)
1407 				pr_err("%s: Failed to start governor\n", __func__);
1408 		}
1409 	} else if (cpufreq_driver->stop_cpu) {
1410 		cpufreq_driver->stop_cpu(policy);
1411 	}
1412 }
1413 
1414 static void cpufreq_offline_finish(unsigned int cpu)
1415 {
1416 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1417 
1418 	if (!policy) {
1419 		pr_debug("%s: No cpu_data found\n", __func__);
1420 		return;
1421 	}
1422 
1423 	/* Only proceed for inactive policies */
1424 	if (!policy_is_inactive(policy))
1425 		return;
1426 
1427 	/* If cpu is last user of policy, free policy */
1428 	if (has_target()) {
1429 		int ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1430 		if (ret)
1431 			pr_err("%s: Failed to exit governor\n", __func__);
1432 	}
1433 
1434 	/*
1435 	 * Perform the ->exit() even during light-weight tear-down,
1436 	 * since this is a core component, and is essential for the
1437 	 * subsequent light-weight ->init() to succeed.
1438 	 */
1439 	if (cpufreq_driver->exit)
1440 		cpufreq_driver->exit(policy);
1441 }
1442 
1443 /**
1444  * cpufreq_remove_dev - remove a CPU device
1445  *
1446  * Removes the cpufreq interface for a CPU device.
1447  */
1448 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1449 {
1450 	unsigned int cpu = dev->id;
1451 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1452 
1453 	if (!policy)
1454 		return;
1455 
1456 	if (cpu_online(cpu)) {
1457 		cpufreq_offline_prepare(cpu);
1458 		cpufreq_offline_finish(cpu);
1459 	}
1460 
1461 	cpumask_clear_cpu(cpu, policy->real_cpus);
1462 
1463 	if (cpumask_empty(policy->real_cpus)) {
1464 		cpufreq_policy_free(policy, true);
1465 		return;
1466 	}
1467 
1468 	if (cpu != policy->kobj_cpu) {
1469 		remove_cpu_dev_symlink(policy, cpu);
1470 	} else {
1471 		/*
1472 		 * The CPU owning the policy object is going away.  Move it to
1473 		 * another suitable CPU.
1474 		 */
1475 		unsigned int new_cpu = cpumask_first(policy->real_cpus);
1476 		struct device *new_dev = get_cpu_device(new_cpu);
1477 
1478 		dev_dbg(dev, "%s: Moving policy object to CPU%u\n", __func__, new_cpu);
1479 
1480 		sysfs_remove_link(&new_dev->kobj, "cpufreq");
1481 		policy->kobj_cpu = new_cpu;
1482 		WARN_ON(kobject_move(&policy->kobj, &new_dev->kobj));
1483 	}
1484 }
1485 
1486 static void handle_update(struct work_struct *work)
1487 {
1488 	struct cpufreq_policy *policy =
1489 		container_of(work, struct cpufreq_policy, update);
1490 	unsigned int cpu = policy->cpu;
1491 	pr_debug("handle_update for cpu %u called\n", cpu);
1492 	cpufreq_update_policy(cpu);
1493 }
1494 
1495 /**
1496  *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1497  *	in deep trouble.
1498  *	@policy: policy managing CPUs
1499  *	@new_freq: CPU frequency the CPU actually runs at
1500  *
1501  *	We adjust to current frequency first, and need to clean up later.
1502  *	So either call to cpufreq_update_policy() or schedule handle_update()).
1503  */
1504 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1505 				unsigned int new_freq)
1506 {
1507 	struct cpufreq_freqs freqs;
1508 
1509 	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1510 		 policy->cur, new_freq);
1511 
1512 	freqs.old = policy->cur;
1513 	freqs.new = new_freq;
1514 
1515 	cpufreq_freq_transition_begin(policy, &freqs);
1516 	cpufreq_freq_transition_end(policy, &freqs, 0);
1517 }
1518 
1519 /**
1520  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1521  * @cpu: CPU number
1522  *
1523  * This is the last known freq, without actually getting it from the driver.
1524  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1525  */
1526 unsigned int cpufreq_quick_get(unsigned int cpu)
1527 {
1528 	struct cpufreq_policy *policy;
1529 	unsigned int ret_freq = 0;
1530 
1531 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1532 		return cpufreq_driver->get(cpu);
1533 
1534 	policy = cpufreq_cpu_get(cpu);
1535 	if (policy) {
1536 		ret_freq = policy->cur;
1537 		cpufreq_cpu_put(policy);
1538 	}
1539 
1540 	return ret_freq;
1541 }
1542 EXPORT_SYMBOL(cpufreq_quick_get);
1543 
1544 /**
1545  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1546  * @cpu: CPU number
1547  *
1548  * Just return the max possible frequency for a given CPU.
1549  */
1550 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1551 {
1552 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1553 	unsigned int ret_freq = 0;
1554 
1555 	if (policy) {
1556 		ret_freq = policy->max;
1557 		cpufreq_cpu_put(policy);
1558 	}
1559 
1560 	return ret_freq;
1561 }
1562 EXPORT_SYMBOL(cpufreq_quick_get_max);
1563 
1564 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1565 {
1566 	unsigned int ret_freq = 0;
1567 
1568 	if (!cpufreq_driver->get)
1569 		return ret_freq;
1570 
1571 	ret_freq = cpufreq_driver->get(policy->cpu);
1572 
1573 	/* Updating inactive policies is invalid, so avoid doing that. */
1574 	if (unlikely(policy_is_inactive(policy)))
1575 		return ret_freq;
1576 
1577 	if (ret_freq && policy->cur &&
1578 		!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1579 		/* verify no discrepancy between actual and
1580 					saved value exists */
1581 		if (unlikely(ret_freq != policy->cur)) {
1582 			cpufreq_out_of_sync(policy, ret_freq);
1583 			schedule_work(&policy->update);
1584 		}
1585 	}
1586 
1587 	return ret_freq;
1588 }
1589 
1590 /**
1591  * cpufreq_get - get the current CPU frequency (in kHz)
1592  * @cpu: CPU number
1593  *
1594  * Get the CPU current (static) CPU frequency
1595  */
1596 unsigned int cpufreq_get(unsigned int cpu)
1597 {
1598 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1599 	unsigned int ret_freq = 0;
1600 
1601 	if (policy) {
1602 		down_read(&policy->rwsem);
1603 		ret_freq = __cpufreq_get(policy);
1604 		up_read(&policy->rwsem);
1605 
1606 		cpufreq_cpu_put(policy);
1607 	}
1608 
1609 	return ret_freq;
1610 }
1611 EXPORT_SYMBOL(cpufreq_get);
1612 
1613 static struct subsys_interface cpufreq_interface = {
1614 	.name		= "cpufreq",
1615 	.subsys		= &cpu_subsys,
1616 	.add_dev	= cpufreq_add_dev,
1617 	.remove_dev	= cpufreq_remove_dev,
1618 };
1619 
1620 /*
1621  * In case platform wants some specific frequency to be configured
1622  * during suspend..
1623  */
1624 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1625 {
1626 	int ret;
1627 
1628 	if (!policy->suspend_freq) {
1629 		pr_debug("%s: suspend_freq not defined\n", __func__);
1630 		return 0;
1631 	}
1632 
1633 	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1634 			policy->suspend_freq);
1635 
1636 	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1637 			CPUFREQ_RELATION_H);
1638 	if (ret)
1639 		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1640 				__func__, policy->suspend_freq, ret);
1641 
1642 	return ret;
1643 }
1644 EXPORT_SYMBOL(cpufreq_generic_suspend);
1645 
1646 /**
1647  * cpufreq_suspend() - Suspend CPUFreq governors
1648  *
1649  * Called during system wide Suspend/Hibernate cycles for suspending governors
1650  * as some platforms can't change frequency after this point in suspend cycle.
1651  * Because some of the devices (like: i2c, regulators, etc) they use for
1652  * changing frequency are suspended quickly after this point.
1653  */
1654 void cpufreq_suspend(void)
1655 {
1656 	struct cpufreq_policy *policy;
1657 
1658 	if (!cpufreq_driver)
1659 		return;
1660 
1661 	if (!has_target())
1662 		goto suspend;
1663 
1664 	pr_debug("%s: Suspending Governors\n", __func__);
1665 
1666 	for_each_active_policy(policy) {
1667 		if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1668 			pr_err("%s: Failed to stop governor for policy: %p\n",
1669 				__func__, policy);
1670 		else if (cpufreq_driver->suspend
1671 		    && cpufreq_driver->suspend(policy))
1672 			pr_err("%s: Failed to suspend driver: %p\n", __func__,
1673 				policy);
1674 	}
1675 
1676 suspend:
1677 	cpufreq_suspended = true;
1678 }
1679 
1680 /**
1681  * cpufreq_resume() - Resume CPUFreq governors
1682  *
1683  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1684  * are suspended with cpufreq_suspend().
1685  */
1686 void cpufreq_resume(void)
1687 {
1688 	struct cpufreq_policy *policy;
1689 
1690 	if (!cpufreq_driver)
1691 		return;
1692 
1693 	cpufreq_suspended = false;
1694 
1695 	if (!has_target())
1696 		return;
1697 
1698 	pr_debug("%s: Resuming Governors\n", __func__);
1699 
1700 	for_each_active_policy(policy) {
1701 		if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1702 			pr_err("%s: Failed to resume driver: %p\n", __func__,
1703 				policy);
1704 		else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1705 		    || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1706 			pr_err("%s: Failed to start governor for policy: %p\n",
1707 				__func__, policy);
1708 	}
1709 
1710 	/*
1711 	 * schedule call cpufreq_update_policy() for first-online CPU, as that
1712 	 * wouldn't be hotplugged-out on suspend. It will verify that the
1713 	 * current freq is in sync with what we believe it to be.
1714 	 */
1715 	policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1716 	if (WARN_ON(!policy))
1717 		return;
1718 
1719 	schedule_work(&policy->update);
1720 }
1721 
1722 /**
1723  *	cpufreq_get_current_driver - return current driver's name
1724  *
1725  *	Return the name string of the currently loaded cpufreq driver
1726  *	or NULL, if none.
1727  */
1728 const char *cpufreq_get_current_driver(void)
1729 {
1730 	if (cpufreq_driver)
1731 		return cpufreq_driver->name;
1732 
1733 	return NULL;
1734 }
1735 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1736 
1737 /**
1738  *	cpufreq_get_driver_data - return current driver data
1739  *
1740  *	Return the private data of the currently loaded cpufreq
1741  *	driver, or NULL if no cpufreq driver is loaded.
1742  */
1743 void *cpufreq_get_driver_data(void)
1744 {
1745 	if (cpufreq_driver)
1746 		return cpufreq_driver->driver_data;
1747 
1748 	return NULL;
1749 }
1750 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1751 
1752 /*********************************************************************
1753  *                     NOTIFIER LISTS INTERFACE                      *
1754  *********************************************************************/
1755 
1756 /**
1757  *	cpufreq_register_notifier - register a driver with cpufreq
1758  *	@nb: notifier function to register
1759  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1760  *
1761  *	Add a driver to one of two lists: either a list of drivers that
1762  *      are notified about clock rate changes (once before and once after
1763  *      the transition), or a list of drivers that are notified about
1764  *      changes in cpufreq policy.
1765  *
1766  *	This function may sleep, and has the same return conditions as
1767  *	blocking_notifier_chain_register.
1768  */
1769 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1770 {
1771 	int ret;
1772 
1773 	if (cpufreq_disabled())
1774 		return -EINVAL;
1775 
1776 	WARN_ON(!init_cpufreq_transition_notifier_list_called);
1777 
1778 	switch (list) {
1779 	case CPUFREQ_TRANSITION_NOTIFIER:
1780 		ret = srcu_notifier_chain_register(
1781 				&cpufreq_transition_notifier_list, nb);
1782 		break;
1783 	case CPUFREQ_POLICY_NOTIFIER:
1784 		ret = blocking_notifier_chain_register(
1785 				&cpufreq_policy_notifier_list, nb);
1786 		break;
1787 	default:
1788 		ret = -EINVAL;
1789 	}
1790 
1791 	return ret;
1792 }
1793 EXPORT_SYMBOL(cpufreq_register_notifier);
1794 
1795 /**
1796  *	cpufreq_unregister_notifier - unregister a driver with cpufreq
1797  *	@nb: notifier block to be unregistered
1798  *	@list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1799  *
1800  *	Remove a driver from the CPU frequency notifier list.
1801  *
1802  *	This function may sleep, and has the same return conditions as
1803  *	blocking_notifier_chain_unregister.
1804  */
1805 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1806 {
1807 	int ret;
1808 
1809 	if (cpufreq_disabled())
1810 		return -EINVAL;
1811 
1812 	switch (list) {
1813 	case CPUFREQ_TRANSITION_NOTIFIER:
1814 		ret = srcu_notifier_chain_unregister(
1815 				&cpufreq_transition_notifier_list, nb);
1816 		break;
1817 	case CPUFREQ_POLICY_NOTIFIER:
1818 		ret = blocking_notifier_chain_unregister(
1819 				&cpufreq_policy_notifier_list, nb);
1820 		break;
1821 	default:
1822 		ret = -EINVAL;
1823 	}
1824 
1825 	return ret;
1826 }
1827 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1828 
1829 
1830 /*********************************************************************
1831  *                              GOVERNORS                            *
1832  *********************************************************************/
1833 
1834 /* Must set freqs->new to intermediate frequency */
1835 static int __target_intermediate(struct cpufreq_policy *policy,
1836 				 struct cpufreq_freqs *freqs, int index)
1837 {
1838 	int ret;
1839 
1840 	freqs->new = cpufreq_driver->get_intermediate(policy, index);
1841 
1842 	/* We don't need to switch to intermediate freq */
1843 	if (!freqs->new)
1844 		return 0;
1845 
1846 	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1847 		 __func__, policy->cpu, freqs->old, freqs->new);
1848 
1849 	cpufreq_freq_transition_begin(policy, freqs);
1850 	ret = cpufreq_driver->target_intermediate(policy, index);
1851 	cpufreq_freq_transition_end(policy, freqs, ret);
1852 
1853 	if (ret)
1854 		pr_err("%s: Failed to change to intermediate frequency: %d\n",
1855 		       __func__, ret);
1856 
1857 	return ret;
1858 }
1859 
1860 static int __target_index(struct cpufreq_policy *policy,
1861 			  struct cpufreq_frequency_table *freq_table, int index)
1862 {
1863 	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1864 	unsigned int intermediate_freq = 0;
1865 	int retval = -EINVAL;
1866 	bool notify;
1867 
1868 	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1869 	if (notify) {
1870 		/* Handle switching to intermediate frequency */
1871 		if (cpufreq_driver->get_intermediate) {
1872 			retval = __target_intermediate(policy, &freqs, index);
1873 			if (retval)
1874 				return retval;
1875 
1876 			intermediate_freq = freqs.new;
1877 			/* Set old freq to intermediate */
1878 			if (intermediate_freq)
1879 				freqs.old = freqs.new;
1880 		}
1881 
1882 		freqs.new = freq_table[index].frequency;
1883 		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1884 			 __func__, policy->cpu, freqs.old, freqs.new);
1885 
1886 		cpufreq_freq_transition_begin(policy, &freqs);
1887 	}
1888 
1889 	retval = cpufreq_driver->target_index(policy, index);
1890 	if (retval)
1891 		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1892 		       retval);
1893 
1894 	if (notify) {
1895 		cpufreq_freq_transition_end(policy, &freqs, retval);
1896 
1897 		/*
1898 		 * Failed after setting to intermediate freq? Driver should have
1899 		 * reverted back to initial frequency and so should we. Check
1900 		 * here for intermediate_freq instead of get_intermediate, in
1901 		 * case we haven't switched to intermediate freq at all.
1902 		 */
1903 		if (unlikely(retval && intermediate_freq)) {
1904 			freqs.old = intermediate_freq;
1905 			freqs.new = policy->restore_freq;
1906 			cpufreq_freq_transition_begin(policy, &freqs);
1907 			cpufreq_freq_transition_end(policy, &freqs, 0);
1908 		}
1909 	}
1910 
1911 	return retval;
1912 }
1913 
1914 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1915 			    unsigned int target_freq,
1916 			    unsigned int relation)
1917 {
1918 	unsigned int old_target_freq = target_freq;
1919 	int retval = -EINVAL;
1920 
1921 	if (cpufreq_disabled())
1922 		return -ENODEV;
1923 
1924 	/* Make sure that target_freq is within supported range */
1925 	if (target_freq > policy->max)
1926 		target_freq = policy->max;
1927 	if (target_freq < policy->min)
1928 		target_freq = policy->min;
1929 
1930 	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1931 		 policy->cpu, target_freq, relation, old_target_freq);
1932 
1933 	/*
1934 	 * This might look like a redundant call as we are checking it again
1935 	 * after finding index. But it is left intentionally for cases where
1936 	 * exactly same freq is called again and so we can save on few function
1937 	 * calls.
1938 	 */
1939 	if (target_freq == policy->cur)
1940 		return 0;
1941 
1942 	/* Save last value to restore later on errors */
1943 	policy->restore_freq = policy->cur;
1944 
1945 	if (cpufreq_driver->target)
1946 		retval = cpufreq_driver->target(policy, target_freq, relation);
1947 	else if (cpufreq_driver->target_index) {
1948 		struct cpufreq_frequency_table *freq_table;
1949 		int index;
1950 
1951 		freq_table = cpufreq_frequency_get_table(policy->cpu);
1952 		if (unlikely(!freq_table)) {
1953 			pr_err("%s: Unable to find freq_table\n", __func__);
1954 			goto out;
1955 		}
1956 
1957 		retval = cpufreq_frequency_table_target(policy, freq_table,
1958 				target_freq, relation, &index);
1959 		if (unlikely(retval)) {
1960 			pr_err("%s: Unable to find matching freq\n", __func__);
1961 			goto out;
1962 		}
1963 
1964 		if (freq_table[index].frequency == policy->cur) {
1965 			retval = 0;
1966 			goto out;
1967 		}
1968 
1969 		retval = __target_index(policy, freq_table, index);
1970 	}
1971 
1972 out:
1973 	return retval;
1974 }
1975 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1976 
1977 int cpufreq_driver_target(struct cpufreq_policy *policy,
1978 			  unsigned int target_freq,
1979 			  unsigned int relation)
1980 {
1981 	int ret = -EINVAL;
1982 
1983 	down_write(&policy->rwsem);
1984 
1985 	ret = __cpufreq_driver_target(policy, target_freq, relation);
1986 
1987 	up_write(&policy->rwsem);
1988 
1989 	return ret;
1990 }
1991 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1992 
1993 static int __cpufreq_governor(struct cpufreq_policy *policy,
1994 					unsigned int event)
1995 {
1996 	int ret;
1997 
1998 	/* Only must be defined when default governor is known to have latency
1999 	   restrictions, like e.g. conservative or ondemand.
2000 	   That this is the case is already ensured in Kconfig
2001 	*/
2002 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
2003 	struct cpufreq_governor *gov = &cpufreq_gov_performance;
2004 #else
2005 	struct cpufreq_governor *gov = NULL;
2006 #endif
2007 
2008 	/* Don't start any governor operations if we are entering suspend */
2009 	if (cpufreq_suspended)
2010 		return 0;
2011 	/*
2012 	 * Governor might not be initiated here if ACPI _PPC changed
2013 	 * notification happened, so check it.
2014 	 */
2015 	if (!policy->governor)
2016 		return -EINVAL;
2017 
2018 	if (policy->governor->max_transition_latency &&
2019 	    policy->cpuinfo.transition_latency >
2020 	    policy->governor->max_transition_latency) {
2021 		if (!gov)
2022 			return -EINVAL;
2023 		else {
2024 			pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2025 				policy->governor->name, gov->name);
2026 			policy->governor = gov;
2027 		}
2028 	}
2029 
2030 	if (event == CPUFREQ_GOV_POLICY_INIT)
2031 		if (!try_module_get(policy->governor->owner))
2032 			return -EINVAL;
2033 
2034 	pr_debug("%s: for CPU %u, event %u\n", __func__, policy->cpu, event);
2035 
2036 	mutex_lock(&cpufreq_governor_lock);
2037 	if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2038 	    || (!policy->governor_enabled
2039 	    && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2040 		mutex_unlock(&cpufreq_governor_lock);
2041 		return -EBUSY;
2042 	}
2043 
2044 	if (event == CPUFREQ_GOV_STOP)
2045 		policy->governor_enabled = false;
2046 	else if (event == CPUFREQ_GOV_START)
2047 		policy->governor_enabled = true;
2048 
2049 	mutex_unlock(&cpufreq_governor_lock);
2050 
2051 	ret = policy->governor->governor(policy, event);
2052 
2053 	if (!ret) {
2054 		if (event == CPUFREQ_GOV_POLICY_INIT)
2055 			policy->governor->initialized++;
2056 		else if (event == CPUFREQ_GOV_POLICY_EXIT)
2057 			policy->governor->initialized--;
2058 	} else {
2059 		/* Restore original values */
2060 		mutex_lock(&cpufreq_governor_lock);
2061 		if (event == CPUFREQ_GOV_STOP)
2062 			policy->governor_enabled = true;
2063 		else if (event == CPUFREQ_GOV_START)
2064 			policy->governor_enabled = false;
2065 		mutex_unlock(&cpufreq_governor_lock);
2066 	}
2067 
2068 	if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2069 			((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2070 		module_put(policy->governor->owner);
2071 
2072 	return ret;
2073 }
2074 
2075 int cpufreq_register_governor(struct cpufreq_governor *governor)
2076 {
2077 	int err;
2078 
2079 	if (!governor)
2080 		return -EINVAL;
2081 
2082 	if (cpufreq_disabled())
2083 		return -ENODEV;
2084 
2085 	mutex_lock(&cpufreq_governor_mutex);
2086 
2087 	governor->initialized = 0;
2088 	err = -EBUSY;
2089 	if (!find_governor(governor->name)) {
2090 		err = 0;
2091 		list_add(&governor->governor_list, &cpufreq_governor_list);
2092 	}
2093 
2094 	mutex_unlock(&cpufreq_governor_mutex);
2095 	return err;
2096 }
2097 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2098 
2099 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2100 {
2101 	struct cpufreq_policy *policy;
2102 	unsigned long flags;
2103 
2104 	if (!governor)
2105 		return;
2106 
2107 	if (cpufreq_disabled())
2108 		return;
2109 
2110 	/* clear last_governor for all inactive policies */
2111 	read_lock_irqsave(&cpufreq_driver_lock, flags);
2112 	for_each_inactive_policy(policy) {
2113 		if (!strcmp(policy->last_governor, governor->name)) {
2114 			policy->governor = NULL;
2115 			strcpy(policy->last_governor, "\0");
2116 		}
2117 	}
2118 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2119 
2120 	mutex_lock(&cpufreq_governor_mutex);
2121 	list_del(&governor->governor_list);
2122 	mutex_unlock(&cpufreq_governor_mutex);
2123 	return;
2124 }
2125 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2126 
2127 
2128 /*********************************************************************
2129  *                          POLICY INTERFACE                         *
2130  *********************************************************************/
2131 
2132 /**
2133  * cpufreq_get_policy - get the current cpufreq_policy
2134  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2135  *	is written
2136  *
2137  * Reads the current cpufreq policy.
2138  */
2139 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2140 {
2141 	struct cpufreq_policy *cpu_policy;
2142 	if (!policy)
2143 		return -EINVAL;
2144 
2145 	cpu_policy = cpufreq_cpu_get(cpu);
2146 	if (!cpu_policy)
2147 		return -EINVAL;
2148 
2149 	memcpy(policy, cpu_policy, sizeof(*policy));
2150 
2151 	cpufreq_cpu_put(cpu_policy);
2152 	return 0;
2153 }
2154 EXPORT_SYMBOL(cpufreq_get_policy);
2155 
2156 /*
2157  * policy : current policy.
2158  * new_policy: policy to be set.
2159  */
2160 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2161 				struct cpufreq_policy *new_policy)
2162 {
2163 	struct cpufreq_governor *old_gov;
2164 	int ret;
2165 
2166 	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2167 		 new_policy->cpu, new_policy->min, new_policy->max);
2168 
2169 	memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2170 
2171 	/*
2172 	* This check works well when we store new min/max freq attributes,
2173 	* because new_policy is a copy of policy with one field updated.
2174 	*/
2175 	if (new_policy->min > new_policy->max)
2176 		return -EINVAL;
2177 
2178 	/* verify the cpu speed can be set within this limit */
2179 	ret = cpufreq_driver->verify(new_policy);
2180 	if (ret)
2181 		return ret;
2182 
2183 	/* adjust if necessary - all reasons */
2184 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2185 			CPUFREQ_ADJUST, new_policy);
2186 
2187 	/*
2188 	 * verify the cpu speed can be set within this limit, which might be
2189 	 * different to the first one
2190 	 */
2191 	ret = cpufreq_driver->verify(new_policy);
2192 	if (ret)
2193 		return ret;
2194 
2195 	/* notification of the new policy */
2196 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2197 			CPUFREQ_NOTIFY, new_policy);
2198 
2199 	policy->min = new_policy->min;
2200 	policy->max = new_policy->max;
2201 
2202 	pr_debug("new min and max freqs are %u - %u kHz\n",
2203 		 policy->min, policy->max);
2204 
2205 	if (cpufreq_driver->setpolicy) {
2206 		policy->policy = new_policy->policy;
2207 		pr_debug("setting range\n");
2208 		return cpufreq_driver->setpolicy(new_policy);
2209 	}
2210 
2211 	if (new_policy->governor == policy->governor)
2212 		goto out;
2213 
2214 	pr_debug("governor switch\n");
2215 
2216 	/* save old, working values */
2217 	old_gov = policy->governor;
2218 	/* end old governor */
2219 	if (old_gov) {
2220 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2221 		if (ret) {
2222 			/* This can happen due to race with other operations */
2223 			pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2224 				 __func__, old_gov->name, ret);
2225 			return ret;
2226 		}
2227 
2228 		up_write(&policy->rwsem);
2229 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2230 		down_write(&policy->rwsem);
2231 
2232 		if (ret) {
2233 			pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2234 			       __func__, old_gov->name, ret);
2235 			return ret;
2236 		}
2237 	}
2238 
2239 	/* start new governor */
2240 	policy->governor = new_policy->governor;
2241 	ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2242 	if (!ret) {
2243 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
2244 		if (!ret)
2245 			goto out;
2246 
2247 		up_write(&policy->rwsem);
2248 		__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2249 		down_write(&policy->rwsem);
2250 	}
2251 
2252 	/* new governor failed, so re-start old one */
2253 	pr_debug("starting governor %s failed\n", policy->governor->name);
2254 	if (old_gov) {
2255 		policy->governor = old_gov;
2256 		if (__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2257 			policy->governor = NULL;
2258 		else
2259 			__cpufreq_governor(policy, CPUFREQ_GOV_START);
2260 	}
2261 
2262 	return ret;
2263 
2264  out:
2265 	pr_debug("governor: change or update limits\n");
2266 	return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2267 }
2268 
2269 /**
2270  *	cpufreq_update_policy - re-evaluate an existing cpufreq policy
2271  *	@cpu: CPU which shall be re-evaluated
2272  *
2273  *	Useful for policy notifiers which have different necessities
2274  *	at different times.
2275  */
2276 int cpufreq_update_policy(unsigned int cpu)
2277 {
2278 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2279 	struct cpufreq_policy new_policy;
2280 	int ret;
2281 
2282 	if (!policy)
2283 		return -ENODEV;
2284 
2285 	down_write(&policy->rwsem);
2286 
2287 	pr_debug("updating policy for CPU %u\n", cpu);
2288 	memcpy(&new_policy, policy, sizeof(*policy));
2289 	new_policy.min = policy->user_policy.min;
2290 	new_policy.max = policy->user_policy.max;
2291 
2292 	/*
2293 	 * BIOS might change freq behind our back
2294 	 * -> ask driver for current freq and notify governors about a change
2295 	 */
2296 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2297 		new_policy.cur = cpufreq_driver->get(cpu);
2298 		if (WARN_ON(!new_policy.cur)) {
2299 			ret = -EIO;
2300 			goto unlock;
2301 		}
2302 
2303 		if (!policy->cur) {
2304 			pr_debug("Driver did not initialize current freq\n");
2305 			policy->cur = new_policy.cur;
2306 		} else {
2307 			if (policy->cur != new_policy.cur && has_target())
2308 				cpufreq_out_of_sync(policy, new_policy.cur);
2309 		}
2310 	}
2311 
2312 	ret = cpufreq_set_policy(policy, &new_policy);
2313 
2314 unlock:
2315 	up_write(&policy->rwsem);
2316 
2317 	cpufreq_cpu_put(policy);
2318 	return ret;
2319 }
2320 EXPORT_SYMBOL(cpufreq_update_policy);
2321 
2322 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2323 					unsigned long action, void *hcpu)
2324 {
2325 	unsigned int cpu = (unsigned long)hcpu;
2326 
2327 	switch (action & ~CPU_TASKS_FROZEN) {
2328 	case CPU_ONLINE:
2329 		cpufreq_online(cpu);
2330 		break;
2331 
2332 	case CPU_DOWN_PREPARE:
2333 		cpufreq_offline_prepare(cpu);
2334 		break;
2335 
2336 	case CPU_POST_DEAD:
2337 		cpufreq_offline_finish(cpu);
2338 		break;
2339 
2340 	case CPU_DOWN_FAILED:
2341 		cpufreq_online(cpu);
2342 		break;
2343 	}
2344 	return NOTIFY_OK;
2345 }
2346 
2347 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2348 	.notifier_call = cpufreq_cpu_callback,
2349 };
2350 
2351 /*********************************************************************
2352  *               BOOST						     *
2353  *********************************************************************/
2354 static int cpufreq_boost_set_sw(int state)
2355 {
2356 	struct cpufreq_frequency_table *freq_table;
2357 	struct cpufreq_policy *policy;
2358 	int ret = -EINVAL;
2359 
2360 	for_each_active_policy(policy) {
2361 		freq_table = cpufreq_frequency_get_table(policy->cpu);
2362 		if (freq_table) {
2363 			ret = cpufreq_frequency_table_cpuinfo(policy,
2364 							freq_table);
2365 			if (ret) {
2366 				pr_err("%s: Policy frequency update failed\n",
2367 				       __func__);
2368 				break;
2369 			}
2370 			policy->user_policy.max = policy->max;
2371 			__cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2372 		}
2373 	}
2374 
2375 	return ret;
2376 }
2377 
2378 int cpufreq_boost_trigger_state(int state)
2379 {
2380 	unsigned long flags;
2381 	int ret = 0;
2382 
2383 	if (cpufreq_driver->boost_enabled == state)
2384 		return 0;
2385 
2386 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2387 	cpufreq_driver->boost_enabled = state;
2388 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2389 
2390 	ret = cpufreq_driver->set_boost(state);
2391 	if (ret) {
2392 		write_lock_irqsave(&cpufreq_driver_lock, flags);
2393 		cpufreq_driver->boost_enabled = !state;
2394 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2395 
2396 		pr_err("%s: Cannot %s BOOST\n",
2397 		       __func__, state ? "enable" : "disable");
2398 	}
2399 
2400 	return ret;
2401 }
2402 
2403 int cpufreq_boost_supported(void)
2404 {
2405 	if (likely(cpufreq_driver))
2406 		return cpufreq_driver->boost_supported;
2407 
2408 	return 0;
2409 }
2410 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2411 
2412 static int create_boost_sysfs_file(void)
2413 {
2414 	int ret;
2415 
2416 	if (!cpufreq_boost_supported())
2417 		return 0;
2418 
2419 	/*
2420 	 * Check if driver provides function to enable boost -
2421 	 * if not, use cpufreq_boost_set_sw as default
2422 	 */
2423 	if (!cpufreq_driver->set_boost)
2424 		cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2425 
2426 	ret = cpufreq_sysfs_create_file(&boost.attr);
2427 	if (ret)
2428 		pr_err("%s: cannot register global BOOST sysfs file\n",
2429 		       __func__);
2430 
2431 	return ret;
2432 }
2433 
2434 static void remove_boost_sysfs_file(void)
2435 {
2436 	if (cpufreq_boost_supported())
2437 		cpufreq_sysfs_remove_file(&boost.attr);
2438 }
2439 
2440 int cpufreq_enable_boost_support(void)
2441 {
2442 	if (!cpufreq_driver)
2443 		return -EINVAL;
2444 
2445 	if (cpufreq_boost_supported())
2446 		return 0;
2447 
2448 	cpufreq_driver->boost_supported = true;
2449 
2450 	/* This will get removed on driver unregister */
2451 	return create_boost_sysfs_file();
2452 }
2453 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2454 
2455 int cpufreq_boost_enabled(void)
2456 {
2457 	return cpufreq_driver->boost_enabled;
2458 }
2459 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2460 
2461 /*********************************************************************
2462  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2463  *********************************************************************/
2464 
2465 /**
2466  * cpufreq_register_driver - register a CPU Frequency driver
2467  * @driver_data: A struct cpufreq_driver containing the values#
2468  * submitted by the CPU Frequency driver.
2469  *
2470  * Registers a CPU Frequency driver to this core code. This code
2471  * returns zero on success, -EBUSY when another driver got here first
2472  * (and isn't unregistered in the meantime).
2473  *
2474  */
2475 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2476 {
2477 	unsigned long flags;
2478 	int ret;
2479 
2480 	if (cpufreq_disabled())
2481 		return -ENODEV;
2482 
2483 	if (!driver_data || !driver_data->verify || !driver_data->init ||
2484 	    !(driver_data->setpolicy || driver_data->target_index ||
2485 		    driver_data->target) ||
2486 	     (driver_data->setpolicy && (driver_data->target_index ||
2487 		    driver_data->target)) ||
2488 	     (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2489 		return -EINVAL;
2490 
2491 	pr_debug("trying to register driver %s\n", driver_data->name);
2492 
2493 	/* Protect against concurrent CPU online/offline. */
2494 	get_online_cpus();
2495 
2496 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2497 	if (cpufreq_driver) {
2498 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2499 		ret = -EEXIST;
2500 		goto out;
2501 	}
2502 	cpufreq_driver = driver_data;
2503 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2504 
2505 	if (driver_data->setpolicy)
2506 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2507 
2508 	ret = create_boost_sysfs_file();
2509 	if (ret)
2510 		goto err_null_driver;
2511 
2512 	ret = subsys_interface_register(&cpufreq_interface);
2513 	if (ret)
2514 		goto err_boost_unreg;
2515 
2516 	if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2517 	    list_empty(&cpufreq_policy_list)) {
2518 		/* if all ->init() calls failed, unregister */
2519 		pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2520 			 driver_data->name);
2521 		goto err_if_unreg;
2522 	}
2523 
2524 	register_hotcpu_notifier(&cpufreq_cpu_notifier);
2525 	pr_debug("driver %s up and running\n", driver_data->name);
2526 
2527 out:
2528 	put_online_cpus();
2529 	return ret;
2530 
2531 err_if_unreg:
2532 	subsys_interface_unregister(&cpufreq_interface);
2533 err_boost_unreg:
2534 	remove_boost_sysfs_file();
2535 err_null_driver:
2536 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2537 	cpufreq_driver = NULL;
2538 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2539 	goto out;
2540 }
2541 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2542 
2543 /**
2544  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2545  *
2546  * Unregister the current CPUFreq driver. Only call this if you have
2547  * the right to do so, i.e. if you have succeeded in initialising before!
2548  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2549  * currently not initialised.
2550  */
2551 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2552 {
2553 	unsigned long flags;
2554 
2555 	if (!cpufreq_driver || (driver != cpufreq_driver))
2556 		return -EINVAL;
2557 
2558 	pr_debug("unregistering driver %s\n", driver->name);
2559 
2560 	/* Protect against concurrent cpu hotplug */
2561 	get_online_cpus();
2562 	subsys_interface_unregister(&cpufreq_interface);
2563 	remove_boost_sysfs_file();
2564 	unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2565 
2566 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2567 
2568 	cpufreq_driver = NULL;
2569 
2570 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2571 	put_online_cpus();
2572 
2573 	return 0;
2574 }
2575 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2576 
2577 /*
2578  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2579  * or mutexes when secondary CPUs are halted.
2580  */
2581 static struct syscore_ops cpufreq_syscore_ops = {
2582 	.shutdown = cpufreq_suspend,
2583 };
2584 
2585 static int __init cpufreq_core_init(void)
2586 {
2587 	if (cpufreq_disabled())
2588 		return -ENODEV;
2589 
2590 	cpufreq_global_kobject = kobject_create();
2591 	BUG_ON(!cpufreq_global_kobject);
2592 
2593 	register_syscore_ops(&cpufreq_syscore_ops);
2594 
2595 	return 0;
2596 }
2597 core_initcall(cpufreq_core_init);
2598