xref: /openbmc/linux/drivers/cpufreq/cpufreq.c (revision 0984d159)
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 /* Macros to iterate over CPU policies */
42 #define for_each_suitable_policy(__policy, __active)			 \
43 	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
44 		if ((__active) == !policy_is_inactive(__policy))
45 
46 #define for_each_active_policy(__policy)		\
47 	for_each_suitable_policy(__policy, true)
48 #define for_each_inactive_policy(__policy)		\
49 	for_each_suitable_policy(__policy, false)
50 
51 #define for_each_policy(__policy)			\
52 	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
53 
54 /* Iterate over governors */
55 static LIST_HEAD(cpufreq_governor_list);
56 #define for_each_governor(__governor)				\
57 	list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
58 
59 /**
60  * The "cpufreq driver" - the arch- or hardware-dependent low
61  * level driver of CPUFreq support, and its spinlock. This lock
62  * also protects the cpufreq_cpu_data array.
63  */
64 static struct cpufreq_driver *cpufreq_driver;
65 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
66 static DEFINE_RWLOCK(cpufreq_driver_lock);
67 
68 /* Flag to suspend/resume CPUFreq governors */
69 static bool cpufreq_suspended;
70 
71 static inline bool has_target(void)
72 {
73 	return cpufreq_driver->target_index || cpufreq_driver->target;
74 }
75 
76 /* internal prototypes */
77 static int cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
78 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
79 static int cpufreq_start_governor(struct cpufreq_policy *policy);
80 
81 static inline void cpufreq_exit_governor(struct cpufreq_policy *policy)
82 {
83 	(void)cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
84 }
85 
86 static inline void cpufreq_stop_governor(struct cpufreq_policy *policy)
87 {
88 	(void)cpufreq_governor(policy, CPUFREQ_GOV_STOP);
89 }
90 
91 /**
92  * Two notifier lists: the "policy" list is involved in the
93  * validation process for a new CPU frequency policy; the
94  * "transition" list for kernel code that needs to handle
95  * changes to devices when the CPU clock speed changes.
96  * The mutex locks both lists.
97  */
98 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
99 static struct srcu_notifier_head cpufreq_transition_notifier_list;
100 
101 static bool init_cpufreq_transition_notifier_list_called;
102 static int __init init_cpufreq_transition_notifier_list(void)
103 {
104 	srcu_init_notifier_head(&cpufreq_transition_notifier_list);
105 	init_cpufreq_transition_notifier_list_called = true;
106 	return 0;
107 }
108 pure_initcall(init_cpufreq_transition_notifier_list);
109 
110 static int off __read_mostly;
111 static int cpufreq_disabled(void)
112 {
113 	return off;
114 }
115 void disable_cpufreq(void)
116 {
117 	off = 1;
118 }
119 static DEFINE_MUTEX(cpufreq_governor_mutex);
120 
121 bool have_governor_per_policy(void)
122 {
123 	return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
124 }
125 EXPORT_SYMBOL_GPL(have_governor_per_policy);
126 
127 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
128 {
129 	if (have_governor_per_policy())
130 		return &policy->kobj;
131 	else
132 		return cpufreq_global_kobject;
133 }
134 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
135 
136 struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu)
137 {
138 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
139 
140 	return policy && !policy_is_inactive(policy) ?
141 		policy->freq_table : NULL;
142 }
143 EXPORT_SYMBOL_GPL(cpufreq_frequency_get_table);
144 
145 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
146 {
147 	u64 idle_time;
148 	u64 cur_wall_time;
149 	u64 busy_time;
150 
151 	cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
152 
153 	busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
154 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
155 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
156 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
157 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
158 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
159 
160 	idle_time = cur_wall_time - busy_time;
161 	if (wall)
162 		*wall = cputime_to_usecs(cur_wall_time);
163 
164 	return cputime_to_usecs(idle_time);
165 }
166 
167 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
168 {
169 	u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
170 
171 	if (idle_time == -1ULL)
172 		return get_cpu_idle_time_jiffy(cpu, wall);
173 	else if (!io_busy)
174 		idle_time += get_cpu_iowait_time_us(cpu, wall);
175 
176 	return idle_time;
177 }
178 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
179 
180 /*
181  * This is a generic cpufreq init() routine which can be used by cpufreq
182  * drivers of SMP systems. It will do following:
183  * - validate & show freq table passed
184  * - set policies transition latency
185  * - policy->cpus with all possible CPUs
186  */
187 int cpufreq_generic_init(struct cpufreq_policy *policy,
188 		struct cpufreq_frequency_table *table,
189 		unsigned int transition_latency)
190 {
191 	int ret;
192 
193 	ret = cpufreq_table_validate_and_show(policy, table);
194 	if (ret) {
195 		pr_err("%s: invalid frequency table: %d\n", __func__, ret);
196 		return ret;
197 	}
198 
199 	policy->cpuinfo.transition_latency = transition_latency;
200 
201 	/*
202 	 * The driver only supports the SMP configuration where all processors
203 	 * share the clock and voltage and clock.
204 	 */
205 	cpumask_setall(policy->cpus);
206 
207 	return 0;
208 }
209 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
210 
211 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
212 {
213 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
214 
215 	return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
216 }
217 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
218 
219 unsigned int cpufreq_generic_get(unsigned int cpu)
220 {
221 	struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
222 
223 	if (!policy || IS_ERR(policy->clk)) {
224 		pr_err("%s: No %s associated to cpu: %d\n",
225 		       __func__, policy ? "clk" : "policy", cpu);
226 		return 0;
227 	}
228 
229 	return clk_get_rate(policy->clk) / 1000;
230 }
231 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
232 
233 /**
234  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
235  *
236  * @cpu: cpu to find policy for.
237  *
238  * This returns policy for 'cpu', returns NULL if it doesn't exist.
239  * It also increments the kobject reference count to mark it busy and so would
240  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
241  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
242  * freed as that depends on the kobj count.
243  *
244  * Return: A valid policy on success, otherwise NULL on failure.
245  */
246 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
247 {
248 	struct cpufreq_policy *policy = NULL;
249 	unsigned long flags;
250 
251 	if (WARN_ON(cpu >= nr_cpu_ids))
252 		return NULL;
253 
254 	/* get the cpufreq driver */
255 	read_lock_irqsave(&cpufreq_driver_lock, flags);
256 
257 	if (cpufreq_driver) {
258 		/* get the CPU */
259 		policy = cpufreq_cpu_get_raw(cpu);
260 		if (policy)
261 			kobject_get(&policy->kobj);
262 	}
263 
264 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
265 
266 	return policy;
267 }
268 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
269 
270 /**
271  * cpufreq_cpu_put: Decrements the usage count of a policy
272  *
273  * @policy: policy earlier returned by cpufreq_cpu_get().
274  *
275  * This decrements the kobject reference count incremented earlier by calling
276  * cpufreq_cpu_get().
277  */
278 void cpufreq_cpu_put(struct cpufreq_policy *policy)
279 {
280 	kobject_put(&policy->kobj);
281 }
282 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
283 
284 /*********************************************************************
285  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
286  *********************************************************************/
287 
288 /**
289  * adjust_jiffies - adjust the system "loops_per_jiffy"
290  *
291  * This function alters the system "loops_per_jiffy" for the clock
292  * speed change. Note that loops_per_jiffy cannot be updated on SMP
293  * systems as each CPU might be scaled differently. So, use the arch
294  * per-CPU loops_per_jiffy value wherever possible.
295  */
296 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
297 {
298 #ifndef CONFIG_SMP
299 	static unsigned long l_p_j_ref;
300 	static unsigned int l_p_j_ref_freq;
301 
302 	if (ci->flags & CPUFREQ_CONST_LOOPS)
303 		return;
304 
305 	if (!l_p_j_ref_freq) {
306 		l_p_j_ref = loops_per_jiffy;
307 		l_p_j_ref_freq = ci->old;
308 		pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
309 			 l_p_j_ref, l_p_j_ref_freq);
310 	}
311 	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
312 		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
313 								ci->new);
314 		pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
315 			 loops_per_jiffy, ci->new);
316 	}
317 #endif
318 }
319 
320 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
321 		struct cpufreq_freqs *freqs, unsigned int state)
322 {
323 	BUG_ON(irqs_disabled());
324 
325 	if (cpufreq_disabled())
326 		return;
327 
328 	freqs->flags = cpufreq_driver->flags;
329 	pr_debug("notification %u of frequency transition to %u kHz\n",
330 		 state, freqs->new);
331 
332 	switch (state) {
333 
334 	case CPUFREQ_PRECHANGE:
335 		/* detect if the driver reported a value as "old frequency"
336 		 * which is not equal to what the cpufreq core thinks is
337 		 * "old frequency".
338 		 */
339 		if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
340 			if ((policy) && (policy->cpu == freqs->cpu) &&
341 			    (policy->cur) && (policy->cur != freqs->old)) {
342 				pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
343 					 freqs->old, policy->cur);
344 				freqs->old = policy->cur;
345 			}
346 		}
347 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
348 				CPUFREQ_PRECHANGE, freqs);
349 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
350 		break;
351 
352 	case CPUFREQ_POSTCHANGE:
353 		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
354 		pr_debug("FREQ: %lu - CPU: %lu\n",
355 			 (unsigned long)freqs->new, (unsigned long)freqs->cpu);
356 		trace_cpu_frequency(freqs->new, freqs->cpu);
357 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
358 				CPUFREQ_POSTCHANGE, freqs);
359 		if (likely(policy) && likely(policy->cpu == freqs->cpu))
360 			policy->cur = freqs->new;
361 		break;
362 	}
363 }
364 
365 /**
366  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
367  * on frequency transition.
368  *
369  * This function calls the transition notifiers and the "adjust_jiffies"
370  * function. It is called twice on all CPU frequency changes that have
371  * external effects.
372  */
373 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
374 		struct cpufreq_freqs *freqs, unsigned int state)
375 {
376 	for_each_cpu(freqs->cpu, policy->cpus)
377 		__cpufreq_notify_transition(policy, freqs, state);
378 }
379 
380 /* Do post notifications when there are chances that transition has failed */
381 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
382 		struct cpufreq_freqs *freqs, int transition_failed)
383 {
384 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
385 	if (!transition_failed)
386 		return;
387 
388 	swap(freqs->old, freqs->new);
389 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
390 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
391 }
392 
393 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
394 		struct cpufreq_freqs *freqs)
395 {
396 
397 	/*
398 	 * Catch double invocations of _begin() which lead to self-deadlock.
399 	 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
400 	 * doesn't invoke _begin() on their behalf, and hence the chances of
401 	 * double invocations are very low. Moreover, there are scenarios
402 	 * where these checks can emit false-positive warnings in these
403 	 * drivers; so we avoid that by skipping them altogether.
404 	 */
405 	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
406 				&& current == policy->transition_task);
407 
408 wait:
409 	wait_event(policy->transition_wait, !policy->transition_ongoing);
410 
411 	spin_lock(&policy->transition_lock);
412 
413 	if (unlikely(policy->transition_ongoing)) {
414 		spin_unlock(&policy->transition_lock);
415 		goto wait;
416 	}
417 
418 	policy->transition_ongoing = true;
419 	policy->transition_task = current;
420 
421 	spin_unlock(&policy->transition_lock);
422 
423 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
424 }
425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
426 
427 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
428 		struct cpufreq_freqs *freqs, int transition_failed)
429 {
430 	if (unlikely(WARN_ON(!policy->transition_ongoing)))
431 		return;
432 
433 	cpufreq_notify_post_transition(policy, freqs, transition_failed);
434 
435 	policy->transition_ongoing = false;
436 	policy->transition_task = NULL;
437 
438 	wake_up(&policy->transition_wait);
439 }
440 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
441 
442 /*
443  * Fast frequency switching status count.  Positive means "enabled", negative
444  * means "disabled" and 0 means "not decided yet".
445  */
446 static int cpufreq_fast_switch_count;
447 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
448 
449 static void cpufreq_list_transition_notifiers(void)
450 {
451 	struct notifier_block *nb;
452 
453 	pr_info("Registered transition notifiers:\n");
454 
455 	mutex_lock(&cpufreq_transition_notifier_list.mutex);
456 
457 	for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
458 		pr_info("%pF\n", nb->notifier_call);
459 
460 	mutex_unlock(&cpufreq_transition_notifier_list.mutex);
461 }
462 
463 /**
464  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
465  * @policy: cpufreq policy to enable fast frequency switching for.
466  *
467  * Try to enable fast frequency switching for @policy.
468  *
469  * The attempt will fail if there is at least one transition notifier registered
470  * at this point, as fast frequency switching is quite fundamentally at odds
471  * with transition notifiers.  Thus if successful, it will make registration of
472  * transition notifiers fail going forward.
473  */
474 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
475 {
476 	lockdep_assert_held(&policy->rwsem);
477 
478 	if (!policy->fast_switch_possible)
479 		return;
480 
481 	mutex_lock(&cpufreq_fast_switch_lock);
482 	if (cpufreq_fast_switch_count >= 0) {
483 		cpufreq_fast_switch_count++;
484 		policy->fast_switch_enabled = true;
485 	} else {
486 		pr_warn("CPU%u: Fast frequency switching not enabled\n",
487 			policy->cpu);
488 		cpufreq_list_transition_notifiers();
489 	}
490 	mutex_unlock(&cpufreq_fast_switch_lock);
491 }
492 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
493 
494 /**
495  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
496  * @policy: cpufreq policy to disable fast frequency switching for.
497  */
498 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
499 {
500 	mutex_lock(&cpufreq_fast_switch_lock);
501 	if (policy->fast_switch_enabled) {
502 		policy->fast_switch_enabled = false;
503 		if (!WARN_ON(cpufreq_fast_switch_count <= 0))
504 			cpufreq_fast_switch_count--;
505 	}
506 	mutex_unlock(&cpufreq_fast_switch_lock);
507 }
508 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
509 
510 /*********************************************************************
511  *                          SYSFS INTERFACE                          *
512  *********************************************************************/
513 static ssize_t show_boost(struct kobject *kobj,
514 				 struct attribute *attr, char *buf)
515 {
516 	return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
517 }
518 
519 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
520 				  const char *buf, size_t count)
521 {
522 	int ret, enable;
523 
524 	ret = sscanf(buf, "%d", &enable);
525 	if (ret != 1 || enable < 0 || enable > 1)
526 		return -EINVAL;
527 
528 	if (cpufreq_boost_trigger_state(enable)) {
529 		pr_err("%s: Cannot %s BOOST!\n",
530 		       __func__, enable ? "enable" : "disable");
531 		return -EINVAL;
532 	}
533 
534 	pr_debug("%s: cpufreq BOOST %s\n",
535 		 __func__, enable ? "enabled" : "disabled");
536 
537 	return count;
538 }
539 define_one_global_rw(boost);
540 
541 static struct cpufreq_governor *find_governor(const char *str_governor)
542 {
543 	struct cpufreq_governor *t;
544 
545 	for_each_governor(t)
546 		if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
547 			return t;
548 
549 	return NULL;
550 }
551 
552 /**
553  * cpufreq_parse_governor - parse a governor string
554  */
555 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
556 				struct cpufreq_governor **governor)
557 {
558 	int err = -EINVAL;
559 
560 	if (cpufreq_driver->setpolicy) {
561 		if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
562 			*policy = CPUFREQ_POLICY_PERFORMANCE;
563 			err = 0;
564 		} else if (!strncasecmp(str_governor, "powersave",
565 						CPUFREQ_NAME_LEN)) {
566 			*policy = CPUFREQ_POLICY_POWERSAVE;
567 			err = 0;
568 		}
569 	} else {
570 		struct cpufreq_governor *t;
571 
572 		mutex_lock(&cpufreq_governor_mutex);
573 
574 		t = find_governor(str_governor);
575 
576 		if (t == NULL) {
577 			int ret;
578 
579 			mutex_unlock(&cpufreq_governor_mutex);
580 			ret = request_module("cpufreq_%s", str_governor);
581 			mutex_lock(&cpufreq_governor_mutex);
582 
583 			if (ret == 0)
584 				t = find_governor(str_governor);
585 		}
586 
587 		if (t != NULL) {
588 			*governor = t;
589 			err = 0;
590 		}
591 
592 		mutex_unlock(&cpufreq_governor_mutex);
593 	}
594 	return err;
595 }
596 
597 /**
598  * cpufreq_per_cpu_attr_read() / show_##file_name() -
599  * print out cpufreq information
600  *
601  * Write out information from cpufreq_driver->policy[cpu]; object must be
602  * "unsigned int".
603  */
604 
605 #define show_one(file_name, object)			\
606 static ssize_t show_##file_name				\
607 (struct cpufreq_policy *policy, char *buf)		\
608 {							\
609 	return sprintf(buf, "%u\n", policy->object);	\
610 }
611 
612 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
613 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
614 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
615 show_one(scaling_min_freq, min);
616 show_one(scaling_max_freq, max);
617 
618 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
619 {
620 	ssize_t ret;
621 
622 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
623 		ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
624 	else
625 		ret = sprintf(buf, "%u\n", policy->cur);
626 	return ret;
627 }
628 
629 static int cpufreq_set_policy(struct cpufreq_policy *policy,
630 				struct cpufreq_policy *new_policy);
631 
632 /**
633  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
634  */
635 #define store_one(file_name, object)			\
636 static ssize_t store_##file_name					\
637 (struct cpufreq_policy *policy, const char *buf, size_t count)		\
638 {									\
639 	int ret, temp;							\
640 	struct cpufreq_policy new_policy;				\
641 									\
642 	memcpy(&new_policy, policy, sizeof(*policy));			\
643 									\
644 	ret = sscanf(buf, "%u", &new_policy.object);			\
645 	if (ret != 1)							\
646 		return -EINVAL;						\
647 									\
648 	temp = new_policy.object;					\
649 	ret = cpufreq_set_policy(policy, &new_policy);		\
650 	if (!ret)							\
651 		policy->user_policy.object = temp;			\
652 									\
653 	return ret ? ret : count;					\
654 }
655 
656 store_one(scaling_min_freq, min);
657 store_one(scaling_max_freq, max);
658 
659 /**
660  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
661  */
662 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
663 					char *buf)
664 {
665 	unsigned int cur_freq = __cpufreq_get(policy);
666 	if (!cur_freq)
667 		return sprintf(buf, "<unknown>");
668 	return sprintf(buf, "%u\n", cur_freq);
669 }
670 
671 /**
672  * show_scaling_governor - show the current policy for the specified CPU
673  */
674 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
675 {
676 	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
677 		return sprintf(buf, "powersave\n");
678 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
679 		return sprintf(buf, "performance\n");
680 	else if (policy->governor)
681 		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
682 				policy->governor->name);
683 	return -EINVAL;
684 }
685 
686 /**
687  * store_scaling_governor - store policy for the specified CPU
688  */
689 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
690 					const char *buf, size_t count)
691 {
692 	int ret;
693 	char	str_governor[16];
694 	struct cpufreq_policy new_policy;
695 
696 	memcpy(&new_policy, policy, sizeof(*policy));
697 
698 	ret = sscanf(buf, "%15s", str_governor);
699 	if (ret != 1)
700 		return -EINVAL;
701 
702 	if (cpufreq_parse_governor(str_governor, &new_policy.policy,
703 						&new_policy.governor))
704 		return -EINVAL;
705 
706 	ret = cpufreq_set_policy(policy, &new_policy);
707 	return ret ? ret : count;
708 }
709 
710 /**
711  * show_scaling_driver - show the cpufreq driver currently loaded
712  */
713 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
714 {
715 	return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
716 }
717 
718 /**
719  * show_scaling_available_governors - show the available CPUfreq governors
720  */
721 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
722 						char *buf)
723 {
724 	ssize_t i = 0;
725 	struct cpufreq_governor *t;
726 
727 	if (!has_target()) {
728 		i += sprintf(buf, "performance powersave");
729 		goto out;
730 	}
731 
732 	for_each_governor(t) {
733 		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
734 		    - (CPUFREQ_NAME_LEN + 2)))
735 			goto out;
736 		i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
737 	}
738 out:
739 	i += sprintf(&buf[i], "\n");
740 	return i;
741 }
742 
743 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
744 {
745 	ssize_t i = 0;
746 	unsigned int cpu;
747 
748 	for_each_cpu(cpu, mask) {
749 		if (i)
750 			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
751 		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
752 		if (i >= (PAGE_SIZE - 5))
753 			break;
754 	}
755 	i += sprintf(&buf[i], "\n");
756 	return i;
757 }
758 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
759 
760 /**
761  * show_related_cpus - show the CPUs affected by each transition even if
762  * hw coordination is in use
763  */
764 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
765 {
766 	return cpufreq_show_cpus(policy->related_cpus, buf);
767 }
768 
769 /**
770  * show_affected_cpus - show the CPUs affected by each transition
771  */
772 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
773 {
774 	return cpufreq_show_cpus(policy->cpus, buf);
775 }
776 
777 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
778 					const char *buf, size_t count)
779 {
780 	unsigned int freq = 0;
781 	unsigned int ret;
782 
783 	if (!policy->governor || !policy->governor->store_setspeed)
784 		return -EINVAL;
785 
786 	ret = sscanf(buf, "%u", &freq);
787 	if (ret != 1)
788 		return -EINVAL;
789 
790 	policy->governor->store_setspeed(policy, freq);
791 
792 	return count;
793 }
794 
795 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
796 {
797 	if (!policy->governor || !policy->governor->show_setspeed)
798 		return sprintf(buf, "<unsupported>\n");
799 
800 	return policy->governor->show_setspeed(policy, buf);
801 }
802 
803 /**
804  * show_bios_limit - show the current cpufreq HW/BIOS limitation
805  */
806 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
807 {
808 	unsigned int limit;
809 	int ret;
810 	if (cpufreq_driver->bios_limit) {
811 		ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
812 		if (!ret)
813 			return sprintf(buf, "%u\n", limit);
814 	}
815 	return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
816 }
817 
818 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
819 cpufreq_freq_attr_ro(cpuinfo_min_freq);
820 cpufreq_freq_attr_ro(cpuinfo_max_freq);
821 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
822 cpufreq_freq_attr_ro(scaling_available_governors);
823 cpufreq_freq_attr_ro(scaling_driver);
824 cpufreq_freq_attr_ro(scaling_cur_freq);
825 cpufreq_freq_attr_ro(bios_limit);
826 cpufreq_freq_attr_ro(related_cpus);
827 cpufreq_freq_attr_ro(affected_cpus);
828 cpufreq_freq_attr_rw(scaling_min_freq);
829 cpufreq_freq_attr_rw(scaling_max_freq);
830 cpufreq_freq_attr_rw(scaling_governor);
831 cpufreq_freq_attr_rw(scaling_setspeed);
832 
833 static struct attribute *default_attrs[] = {
834 	&cpuinfo_min_freq.attr,
835 	&cpuinfo_max_freq.attr,
836 	&cpuinfo_transition_latency.attr,
837 	&scaling_min_freq.attr,
838 	&scaling_max_freq.attr,
839 	&affected_cpus.attr,
840 	&related_cpus.attr,
841 	&scaling_governor.attr,
842 	&scaling_driver.attr,
843 	&scaling_available_governors.attr,
844 	&scaling_setspeed.attr,
845 	NULL
846 };
847 
848 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
849 #define to_attr(a) container_of(a, struct freq_attr, attr)
850 
851 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
852 {
853 	struct cpufreq_policy *policy = to_policy(kobj);
854 	struct freq_attr *fattr = to_attr(attr);
855 	ssize_t ret;
856 
857 	down_read(&policy->rwsem);
858 	ret = fattr->show(policy, buf);
859 	up_read(&policy->rwsem);
860 
861 	return ret;
862 }
863 
864 static ssize_t store(struct kobject *kobj, struct attribute *attr,
865 		     const char *buf, size_t count)
866 {
867 	struct cpufreq_policy *policy = to_policy(kobj);
868 	struct freq_attr *fattr = to_attr(attr);
869 	ssize_t ret = -EINVAL;
870 
871 	get_online_cpus();
872 
873 	if (cpu_online(policy->cpu)) {
874 		down_write(&policy->rwsem);
875 		ret = fattr->store(policy, buf, count);
876 		up_write(&policy->rwsem);
877 	}
878 
879 	put_online_cpus();
880 
881 	return ret;
882 }
883 
884 static void cpufreq_sysfs_release(struct kobject *kobj)
885 {
886 	struct cpufreq_policy *policy = to_policy(kobj);
887 	pr_debug("last reference is dropped\n");
888 	complete(&policy->kobj_unregister);
889 }
890 
891 static const struct sysfs_ops sysfs_ops = {
892 	.show	= show,
893 	.store	= store,
894 };
895 
896 static struct kobj_type ktype_cpufreq = {
897 	.sysfs_ops	= &sysfs_ops,
898 	.default_attrs	= default_attrs,
899 	.release	= cpufreq_sysfs_release,
900 };
901 
902 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
903 {
904 	struct device *cpu_dev;
905 
906 	pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
907 
908 	if (!policy)
909 		return 0;
910 
911 	cpu_dev = get_cpu_device(cpu);
912 	if (WARN_ON(!cpu_dev))
913 		return 0;
914 
915 	return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
916 }
917 
918 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
919 {
920 	struct device *cpu_dev;
921 
922 	pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
923 
924 	cpu_dev = get_cpu_device(cpu);
925 	if (WARN_ON(!cpu_dev))
926 		return;
927 
928 	sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
929 }
930 
931 /* Add/remove symlinks for all related CPUs */
932 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
933 {
934 	unsigned int j;
935 	int ret = 0;
936 
937 	/* Some related CPUs might not be present (physically hotplugged) */
938 	for_each_cpu(j, policy->real_cpus) {
939 		ret = add_cpu_dev_symlink(policy, j);
940 		if (ret)
941 			break;
942 	}
943 
944 	return ret;
945 }
946 
947 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
948 {
949 	unsigned int j;
950 
951 	/* Some related CPUs might not be present (physically hotplugged) */
952 	for_each_cpu(j, policy->real_cpus)
953 		remove_cpu_dev_symlink(policy, j);
954 }
955 
956 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
957 {
958 	struct freq_attr **drv_attr;
959 	int ret = 0;
960 
961 	/* set up files for this cpu device */
962 	drv_attr = cpufreq_driver->attr;
963 	while (drv_attr && *drv_attr) {
964 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
965 		if (ret)
966 			return ret;
967 		drv_attr++;
968 	}
969 	if (cpufreq_driver->get) {
970 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
971 		if (ret)
972 			return ret;
973 	}
974 
975 	ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
976 	if (ret)
977 		return ret;
978 
979 	if (cpufreq_driver->bios_limit) {
980 		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
981 		if (ret)
982 			return ret;
983 	}
984 
985 	return cpufreq_add_dev_symlink(policy);
986 }
987 
988 __weak struct cpufreq_governor *cpufreq_default_governor(void)
989 {
990 	return NULL;
991 }
992 
993 static int cpufreq_init_policy(struct cpufreq_policy *policy)
994 {
995 	struct cpufreq_governor *gov = NULL;
996 	struct cpufreq_policy new_policy;
997 
998 	memcpy(&new_policy, policy, sizeof(*policy));
999 
1000 	/* Update governor of new_policy to the governor used before hotplug */
1001 	gov = find_governor(policy->last_governor);
1002 	if (gov) {
1003 		pr_debug("Restoring governor %s for cpu %d\n",
1004 				policy->governor->name, policy->cpu);
1005 	} else {
1006 		gov = cpufreq_default_governor();
1007 		if (!gov)
1008 			return -ENODATA;
1009 	}
1010 
1011 	new_policy.governor = gov;
1012 
1013 	/* Use the default policy if there is no last_policy. */
1014 	if (cpufreq_driver->setpolicy) {
1015 		if (policy->last_policy)
1016 			new_policy.policy = policy->last_policy;
1017 		else
1018 			cpufreq_parse_governor(gov->name, &new_policy.policy,
1019 					       NULL);
1020 	}
1021 	/* set default policy */
1022 	return cpufreq_set_policy(policy, &new_policy);
1023 }
1024 
1025 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1026 {
1027 	int ret = 0;
1028 
1029 	/* Has this CPU been taken care of already? */
1030 	if (cpumask_test_cpu(cpu, policy->cpus))
1031 		return 0;
1032 
1033 	down_write(&policy->rwsem);
1034 	if (has_target())
1035 		cpufreq_stop_governor(policy);
1036 
1037 	cpumask_set_cpu(cpu, policy->cpus);
1038 
1039 	if (has_target()) {
1040 		ret = cpufreq_start_governor(policy);
1041 		if (ret)
1042 			pr_err("%s: Failed to start governor\n", __func__);
1043 	}
1044 	up_write(&policy->rwsem);
1045 	return ret;
1046 }
1047 
1048 static void handle_update(struct work_struct *work)
1049 {
1050 	struct cpufreq_policy *policy =
1051 		container_of(work, struct cpufreq_policy, update);
1052 	unsigned int cpu = policy->cpu;
1053 	pr_debug("handle_update for cpu %u called\n", cpu);
1054 	cpufreq_update_policy(cpu);
1055 }
1056 
1057 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1058 {
1059 	struct device *dev = get_cpu_device(cpu);
1060 	struct cpufreq_policy *policy;
1061 	int ret;
1062 
1063 	if (WARN_ON(!dev))
1064 		return NULL;
1065 
1066 	policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1067 	if (!policy)
1068 		return NULL;
1069 
1070 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1071 		goto err_free_policy;
1072 
1073 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1074 		goto err_free_cpumask;
1075 
1076 	if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1077 		goto err_free_rcpumask;
1078 
1079 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1080 				   cpufreq_global_kobject, "policy%u", cpu);
1081 	if (ret) {
1082 		pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1083 		goto err_free_real_cpus;
1084 	}
1085 
1086 	INIT_LIST_HEAD(&policy->policy_list);
1087 	init_rwsem(&policy->rwsem);
1088 	spin_lock_init(&policy->transition_lock);
1089 	init_waitqueue_head(&policy->transition_wait);
1090 	init_completion(&policy->kobj_unregister);
1091 	INIT_WORK(&policy->update, handle_update);
1092 
1093 	policy->cpu = cpu;
1094 	return policy;
1095 
1096 err_free_real_cpus:
1097 	free_cpumask_var(policy->real_cpus);
1098 err_free_rcpumask:
1099 	free_cpumask_var(policy->related_cpus);
1100 err_free_cpumask:
1101 	free_cpumask_var(policy->cpus);
1102 err_free_policy:
1103 	kfree(policy);
1104 
1105 	return NULL;
1106 }
1107 
1108 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1109 {
1110 	struct kobject *kobj;
1111 	struct completion *cmp;
1112 
1113 	if (notify)
1114 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1115 					     CPUFREQ_REMOVE_POLICY, policy);
1116 
1117 	down_write(&policy->rwsem);
1118 	cpufreq_remove_dev_symlink(policy);
1119 	kobj = &policy->kobj;
1120 	cmp = &policy->kobj_unregister;
1121 	up_write(&policy->rwsem);
1122 	kobject_put(kobj);
1123 
1124 	/*
1125 	 * We need to make sure that the underlying kobj is
1126 	 * actually not referenced anymore by anybody before we
1127 	 * proceed with unloading.
1128 	 */
1129 	pr_debug("waiting for dropping of refcount\n");
1130 	wait_for_completion(cmp);
1131 	pr_debug("wait complete\n");
1132 }
1133 
1134 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1135 {
1136 	unsigned long flags;
1137 	int cpu;
1138 
1139 	/* Remove policy from list */
1140 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1141 	list_del(&policy->policy_list);
1142 
1143 	for_each_cpu(cpu, policy->related_cpus)
1144 		per_cpu(cpufreq_cpu_data, cpu) = NULL;
1145 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1146 
1147 	cpufreq_policy_put_kobj(policy, notify);
1148 	free_cpumask_var(policy->real_cpus);
1149 	free_cpumask_var(policy->related_cpus);
1150 	free_cpumask_var(policy->cpus);
1151 	kfree(policy);
1152 }
1153 
1154 static int cpufreq_online(unsigned int cpu)
1155 {
1156 	struct cpufreq_policy *policy;
1157 	bool new_policy;
1158 	unsigned long flags;
1159 	unsigned int j;
1160 	int ret;
1161 
1162 	pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1163 
1164 	/* Check if this CPU already has a policy to manage it */
1165 	policy = per_cpu(cpufreq_cpu_data, cpu);
1166 	if (policy) {
1167 		WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1168 		if (!policy_is_inactive(policy))
1169 			return cpufreq_add_policy_cpu(policy, cpu);
1170 
1171 		/* This is the only online CPU for the policy.  Start over. */
1172 		new_policy = false;
1173 		down_write(&policy->rwsem);
1174 		policy->cpu = cpu;
1175 		policy->governor = NULL;
1176 		up_write(&policy->rwsem);
1177 	} else {
1178 		new_policy = true;
1179 		policy = cpufreq_policy_alloc(cpu);
1180 		if (!policy)
1181 			return -ENOMEM;
1182 	}
1183 
1184 	cpumask_copy(policy->cpus, cpumask_of(cpu));
1185 
1186 	/* call driver. From then on the cpufreq must be able
1187 	 * to accept all calls to ->verify and ->setpolicy for this CPU
1188 	 */
1189 	ret = cpufreq_driver->init(policy);
1190 	if (ret) {
1191 		pr_debug("initialization failed\n");
1192 		goto out_free_policy;
1193 	}
1194 
1195 	down_write(&policy->rwsem);
1196 
1197 	if (new_policy) {
1198 		/* related_cpus should at least include policy->cpus. */
1199 		cpumask_copy(policy->related_cpus, policy->cpus);
1200 		/* Remember CPUs present at the policy creation time. */
1201 		cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1202 	}
1203 
1204 	/*
1205 	 * affected cpus must always be the one, which are online. We aren't
1206 	 * managing offline cpus here.
1207 	 */
1208 	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1209 
1210 	if (new_policy) {
1211 		policy->user_policy.min = policy->min;
1212 		policy->user_policy.max = policy->max;
1213 
1214 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1215 		for_each_cpu(j, policy->related_cpus)
1216 			per_cpu(cpufreq_cpu_data, j) = policy;
1217 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1218 	}
1219 
1220 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1221 		policy->cur = cpufreq_driver->get(policy->cpu);
1222 		if (!policy->cur) {
1223 			pr_err("%s: ->get() failed\n", __func__);
1224 			goto out_exit_policy;
1225 		}
1226 	}
1227 
1228 	/*
1229 	 * Sometimes boot loaders set CPU frequency to a value outside of
1230 	 * frequency table present with cpufreq core. In such cases CPU might be
1231 	 * unstable if it has to run on that frequency for long duration of time
1232 	 * and so its better to set it to a frequency which is specified in
1233 	 * freq-table. This also makes cpufreq stats inconsistent as
1234 	 * cpufreq-stats would fail to register because current frequency of CPU
1235 	 * isn't found in freq-table.
1236 	 *
1237 	 * Because we don't want this change to effect boot process badly, we go
1238 	 * for the next freq which is >= policy->cur ('cur' must be set by now,
1239 	 * otherwise we will end up setting freq to lowest of the table as 'cur'
1240 	 * is initialized to zero).
1241 	 *
1242 	 * We are passing target-freq as "policy->cur - 1" otherwise
1243 	 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1244 	 * equal to target-freq.
1245 	 */
1246 	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1247 	    && has_target()) {
1248 		/* Are we running at unknown frequency ? */
1249 		ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1250 		if (ret == -EINVAL) {
1251 			/* Warn user and fix it */
1252 			pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1253 				__func__, policy->cpu, policy->cur);
1254 			ret = __cpufreq_driver_target(policy, policy->cur - 1,
1255 				CPUFREQ_RELATION_L);
1256 
1257 			/*
1258 			 * Reaching here after boot in a few seconds may not
1259 			 * mean that system will remain stable at "unknown"
1260 			 * frequency for longer duration. Hence, a BUG_ON().
1261 			 */
1262 			BUG_ON(ret);
1263 			pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1264 				__func__, policy->cpu, policy->cur);
1265 		}
1266 	}
1267 
1268 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1269 				     CPUFREQ_START, policy);
1270 
1271 	if (new_policy) {
1272 		ret = cpufreq_add_dev_interface(policy);
1273 		if (ret)
1274 			goto out_exit_policy;
1275 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1276 				CPUFREQ_CREATE_POLICY, policy);
1277 
1278 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1279 		list_add(&policy->policy_list, &cpufreq_policy_list);
1280 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1281 	}
1282 
1283 	ret = cpufreq_init_policy(policy);
1284 	if (ret) {
1285 		pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1286 		       __func__, cpu, ret);
1287 		/* cpufreq_policy_free() will notify based on this */
1288 		new_policy = false;
1289 		goto out_exit_policy;
1290 	}
1291 
1292 	up_write(&policy->rwsem);
1293 
1294 	kobject_uevent(&policy->kobj, KOBJ_ADD);
1295 
1296 	/* Callback for handling stuff after policy is ready */
1297 	if (cpufreq_driver->ready)
1298 		cpufreq_driver->ready(policy);
1299 
1300 	pr_debug("initialization complete\n");
1301 
1302 	return 0;
1303 
1304 out_exit_policy:
1305 	up_write(&policy->rwsem);
1306 
1307 	if (cpufreq_driver->exit)
1308 		cpufreq_driver->exit(policy);
1309 out_free_policy:
1310 	cpufreq_policy_free(policy, !new_policy);
1311 	return ret;
1312 }
1313 
1314 /**
1315  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1316  * @dev: CPU device.
1317  * @sif: Subsystem interface structure pointer (not used)
1318  */
1319 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1320 {
1321 	struct cpufreq_policy *policy;
1322 	unsigned cpu = dev->id;
1323 
1324 	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1325 
1326 	if (cpu_online(cpu))
1327 		return cpufreq_online(cpu);
1328 
1329 	/*
1330 	 * A hotplug notifier will follow and we will handle it as CPU online
1331 	 * then.  For now, just create the sysfs link, unless there is no policy
1332 	 * or the link is already present.
1333 	 */
1334 	policy = per_cpu(cpufreq_cpu_data, cpu);
1335 	if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1336 		return 0;
1337 
1338 	return add_cpu_dev_symlink(policy, cpu);
1339 }
1340 
1341 static void cpufreq_offline(unsigned int cpu)
1342 {
1343 	struct cpufreq_policy *policy;
1344 	int ret;
1345 
1346 	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1347 
1348 	policy = cpufreq_cpu_get_raw(cpu);
1349 	if (!policy) {
1350 		pr_debug("%s: No cpu_data found\n", __func__);
1351 		return;
1352 	}
1353 
1354 	down_write(&policy->rwsem);
1355 	if (has_target())
1356 		cpufreq_stop_governor(policy);
1357 
1358 	cpumask_clear_cpu(cpu, policy->cpus);
1359 
1360 	if (policy_is_inactive(policy)) {
1361 		if (has_target())
1362 			strncpy(policy->last_governor, policy->governor->name,
1363 				CPUFREQ_NAME_LEN);
1364 		else
1365 			policy->last_policy = policy->policy;
1366 	} else if (cpu == policy->cpu) {
1367 		/* Nominate new CPU */
1368 		policy->cpu = cpumask_any(policy->cpus);
1369 	}
1370 
1371 	/* Start governor again for active policy */
1372 	if (!policy_is_inactive(policy)) {
1373 		if (has_target()) {
1374 			ret = cpufreq_start_governor(policy);
1375 			if (ret)
1376 				pr_err("%s: Failed to start governor\n", __func__);
1377 		}
1378 
1379 		goto unlock;
1380 	}
1381 
1382 	if (cpufreq_driver->stop_cpu)
1383 		cpufreq_driver->stop_cpu(policy);
1384 
1385 	if (has_target())
1386 		cpufreq_exit_governor(policy);
1387 
1388 	/*
1389 	 * Perform the ->exit() even during light-weight tear-down,
1390 	 * since this is a core component, and is essential for the
1391 	 * subsequent light-weight ->init() to succeed.
1392 	 */
1393 	if (cpufreq_driver->exit) {
1394 		cpufreq_driver->exit(policy);
1395 		policy->freq_table = NULL;
1396 	}
1397 
1398 unlock:
1399 	up_write(&policy->rwsem);
1400 }
1401 
1402 /**
1403  * cpufreq_remove_dev - remove a CPU device
1404  *
1405  * Removes the cpufreq interface for a CPU device.
1406  */
1407 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1408 {
1409 	unsigned int cpu = dev->id;
1410 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1411 
1412 	if (!policy)
1413 		return;
1414 
1415 	if (cpu_online(cpu))
1416 		cpufreq_offline(cpu);
1417 
1418 	cpumask_clear_cpu(cpu, policy->real_cpus);
1419 	remove_cpu_dev_symlink(policy, cpu);
1420 
1421 	if (cpumask_empty(policy->real_cpus))
1422 		cpufreq_policy_free(policy, true);
1423 }
1424 
1425 /**
1426  *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1427  *	in deep trouble.
1428  *	@policy: policy managing CPUs
1429  *	@new_freq: CPU frequency the CPU actually runs at
1430  *
1431  *	We adjust to current frequency first, and need to clean up later.
1432  *	So either call to cpufreq_update_policy() or schedule handle_update()).
1433  */
1434 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1435 				unsigned int new_freq)
1436 {
1437 	struct cpufreq_freqs freqs;
1438 
1439 	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1440 		 policy->cur, new_freq);
1441 
1442 	freqs.old = policy->cur;
1443 	freqs.new = new_freq;
1444 
1445 	cpufreq_freq_transition_begin(policy, &freqs);
1446 	cpufreq_freq_transition_end(policy, &freqs, 0);
1447 }
1448 
1449 /**
1450  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1451  * @cpu: CPU number
1452  *
1453  * This is the last known freq, without actually getting it from the driver.
1454  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1455  */
1456 unsigned int cpufreq_quick_get(unsigned int cpu)
1457 {
1458 	struct cpufreq_policy *policy;
1459 	unsigned int ret_freq = 0;
1460 	unsigned long flags;
1461 
1462 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1463 
1464 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1465 		ret_freq = cpufreq_driver->get(cpu);
1466 		read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1467 		return ret_freq;
1468 	}
1469 
1470 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1471 
1472 	policy = cpufreq_cpu_get(cpu);
1473 	if (policy) {
1474 		ret_freq = policy->cur;
1475 		cpufreq_cpu_put(policy);
1476 	}
1477 
1478 	return ret_freq;
1479 }
1480 EXPORT_SYMBOL(cpufreq_quick_get);
1481 
1482 /**
1483  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1484  * @cpu: CPU number
1485  *
1486  * Just return the max possible frequency for a given CPU.
1487  */
1488 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1489 {
1490 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1491 	unsigned int ret_freq = 0;
1492 
1493 	if (policy) {
1494 		ret_freq = policy->max;
1495 		cpufreq_cpu_put(policy);
1496 	}
1497 
1498 	return ret_freq;
1499 }
1500 EXPORT_SYMBOL(cpufreq_quick_get_max);
1501 
1502 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1503 {
1504 	unsigned int ret_freq = 0;
1505 
1506 	if (!cpufreq_driver->get)
1507 		return ret_freq;
1508 
1509 	ret_freq = cpufreq_driver->get(policy->cpu);
1510 
1511 	/*
1512 	 * Updating inactive policies is invalid, so avoid doing that.  Also
1513 	 * if fast frequency switching is used with the given policy, the check
1514 	 * against policy->cur is pointless, so skip it in that case too.
1515 	 */
1516 	if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1517 		return ret_freq;
1518 
1519 	if (ret_freq && policy->cur &&
1520 		!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1521 		/* verify no discrepancy between actual and
1522 					saved value exists */
1523 		if (unlikely(ret_freq != policy->cur)) {
1524 			cpufreq_out_of_sync(policy, ret_freq);
1525 			schedule_work(&policy->update);
1526 		}
1527 	}
1528 
1529 	return ret_freq;
1530 }
1531 
1532 /**
1533  * cpufreq_get - get the current CPU frequency (in kHz)
1534  * @cpu: CPU number
1535  *
1536  * Get the CPU current (static) CPU frequency
1537  */
1538 unsigned int cpufreq_get(unsigned int cpu)
1539 {
1540 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1541 	unsigned int ret_freq = 0;
1542 
1543 	if (policy) {
1544 		down_read(&policy->rwsem);
1545 		ret_freq = __cpufreq_get(policy);
1546 		up_read(&policy->rwsem);
1547 
1548 		cpufreq_cpu_put(policy);
1549 	}
1550 
1551 	return ret_freq;
1552 }
1553 EXPORT_SYMBOL(cpufreq_get);
1554 
1555 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1556 {
1557 	unsigned int new_freq;
1558 
1559 	if (cpufreq_suspended)
1560 		return 0;
1561 
1562 	new_freq = cpufreq_driver->get(policy->cpu);
1563 	if (!new_freq)
1564 		return 0;
1565 
1566 	if (!policy->cur) {
1567 		pr_debug("cpufreq: Driver did not initialize current freq\n");
1568 		policy->cur = new_freq;
1569 	} else if (policy->cur != new_freq && has_target()) {
1570 		cpufreq_out_of_sync(policy, new_freq);
1571 	}
1572 
1573 	return new_freq;
1574 }
1575 
1576 static struct subsys_interface cpufreq_interface = {
1577 	.name		= "cpufreq",
1578 	.subsys		= &cpu_subsys,
1579 	.add_dev	= cpufreq_add_dev,
1580 	.remove_dev	= cpufreq_remove_dev,
1581 };
1582 
1583 /*
1584  * In case platform wants some specific frequency to be configured
1585  * during suspend..
1586  */
1587 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1588 {
1589 	int ret;
1590 
1591 	if (!policy->suspend_freq) {
1592 		pr_debug("%s: suspend_freq not defined\n", __func__);
1593 		return 0;
1594 	}
1595 
1596 	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1597 			policy->suspend_freq);
1598 
1599 	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1600 			CPUFREQ_RELATION_H);
1601 	if (ret)
1602 		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1603 				__func__, policy->suspend_freq, ret);
1604 
1605 	return ret;
1606 }
1607 EXPORT_SYMBOL(cpufreq_generic_suspend);
1608 
1609 /**
1610  * cpufreq_suspend() - Suspend CPUFreq governors
1611  *
1612  * Called during system wide Suspend/Hibernate cycles for suspending governors
1613  * as some platforms can't change frequency after this point in suspend cycle.
1614  * Because some of the devices (like: i2c, regulators, etc) they use for
1615  * changing frequency are suspended quickly after this point.
1616  */
1617 void cpufreq_suspend(void)
1618 {
1619 	struct cpufreq_policy *policy;
1620 
1621 	if (!cpufreq_driver)
1622 		return;
1623 
1624 	if (!has_target() && !cpufreq_driver->suspend)
1625 		goto suspend;
1626 
1627 	pr_debug("%s: Suspending Governors\n", __func__);
1628 
1629 	for_each_active_policy(policy) {
1630 		if (has_target()) {
1631 			down_write(&policy->rwsem);
1632 			cpufreq_stop_governor(policy);
1633 			up_write(&policy->rwsem);
1634 		}
1635 
1636 		if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1637 			pr_err("%s: Failed to suspend driver: %p\n", __func__,
1638 				policy);
1639 	}
1640 
1641 suspend:
1642 	cpufreq_suspended = true;
1643 }
1644 
1645 /**
1646  * cpufreq_resume() - Resume CPUFreq governors
1647  *
1648  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1649  * are suspended with cpufreq_suspend().
1650  */
1651 void cpufreq_resume(void)
1652 {
1653 	struct cpufreq_policy *policy;
1654 	int ret;
1655 
1656 	if (!cpufreq_driver)
1657 		return;
1658 
1659 	cpufreq_suspended = false;
1660 
1661 	if (!has_target() && !cpufreq_driver->resume)
1662 		return;
1663 
1664 	pr_debug("%s: Resuming Governors\n", __func__);
1665 
1666 	for_each_active_policy(policy) {
1667 		if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1668 			pr_err("%s: Failed to resume driver: %p\n", __func__,
1669 				policy);
1670 		} else if (has_target()) {
1671 			down_write(&policy->rwsem);
1672 			ret = cpufreq_start_governor(policy);
1673 			up_write(&policy->rwsem);
1674 
1675 			if (ret)
1676 				pr_err("%s: Failed to start governor for policy: %p\n",
1677 				       __func__, policy);
1678 		}
1679 	}
1680 }
1681 
1682 /**
1683  *	cpufreq_get_current_driver - return current driver's name
1684  *
1685  *	Return the name string of the currently loaded cpufreq driver
1686  *	or NULL, if none.
1687  */
1688 const char *cpufreq_get_current_driver(void)
1689 {
1690 	if (cpufreq_driver)
1691 		return cpufreq_driver->name;
1692 
1693 	return NULL;
1694 }
1695 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1696 
1697 /**
1698  *	cpufreq_get_driver_data - return current driver data
1699  *
1700  *	Return the private data of the currently loaded cpufreq
1701  *	driver, or NULL if no cpufreq driver is loaded.
1702  */
1703 void *cpufreq_get_driver_data(void)
1704 {
1705 	if (cpufreq_driver)
1706 		return cpufreq_driver->driver_data;
1707 
1708 	return NULL;
1709 }
1710 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1711 
1712 /*********************************************************************
1713  *                     NOTIFIER LISTS INTERFACE                      *
1714  *********************************************************************/
1715 
1716 /**
1717  *	cpufreq_register_notifier - register a driver with cpufreq
1718  *	@nb: notifier function to register
1719  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1720  *
1721  *	Add a driver to one of two lists: either a list of drivers that
1722  *      are notified about clock rate changes (once before and once after
1723  *      the transition), or a list of drivers that are notified about
1724  *      changes in cpufreq policy.
1725  *
1726  *	This function may sleep, and has the same return conditions as
1727  *	blocking_notifier_chain_register.
1728  */
1729 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1730 {
1731 	int ret;
1732 
1733 	if (cpufreq_disabled())
1734 		return -EINVAL;
1735 
1736 	WARN_ON(!init_cpufreq_transition_notifier_list_called);
1737 
1738 	switch (list) {
1739 	case CPUFREQ_TRANSITION_NOTIFIER:
1740 		mutex_lock(&cpufreq_fast_switch_lock);
1741 
1742 		if (cpufreq_fast_switch_count > 0) {
1743 			mutex_unlock(&cpufreq_fast_switch_lock);
1744 			return -EBUSY;
1745 		}
1746 		ret = srcu_notifier_chain_register(
1747 				&cpufreq_transition_notifier_list, nb);
1748 		if (!ret)
1749 			cpufreq_fast_switch_count--;
1750 
1751 		mutex_unlock(&cpufreq_fast_switch_lock);
1752 		break;
1753 	case CPUFREQ_POLICY_NOTIFIER:
1754 		ret = blocking_notifier_chain_register(
1755 				&cpufreq_policy_notifier_list, nb);
1756 		break;
1757 	default:
1758 		ret = -EINVAL;
1759 	}
1760 
1761 	return ret;
1762 }
1763 EXPORT_SYMBOL(cpufreq_register_notifier);
1764 
1765 /**
1766  *	cpufreq_unregister_notifier - unregister a driver with cpufreq
1767  *	@nb: notifier block to be unregistered
1768  *	@list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1769  *
1770  *	Remove a driver from the CPU frequency notifier list.
1771  *
1772  *	This function may sleep, and has the same return conditions as
1773  *	blocking_notifier_chain_unregister.
1774  */
1775 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1776 {
1777 	int ret;
1778 
1779 	if (cpufreq_disabled())
1780 		return -EINVAL;
1781 
1782 	switch (list) {
1783 	case CPUFREQ_TRANSITION_NOTIFIER:
1784 		mutex_lock(&cpufreq_fast_switch_lock);
1785 
1786 		ret = srcu_notifier_chain_unregister(
1787 				&cpufreq_transition_notifier_list, nb);
1788 		if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1789 			cpufreq_fast_switch_count++;
1790 
1791 		mutex_unlock(&cpufreq_fast_switch_lock);
1792 		break;
1793 	case CPUFREQ_POLICY_NOTIFIER:
1794 		ret = blocking_notifier_chain_unregister(
1795 				&cpufreq_policy_notifier_list, nb);
1796 		break;
1797 	default:
1798 		ret = -EINVAL;
1799 	}
1800 
1801 	return ret;
1802 }
1803 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1804 
1805 
1806 /*********************************************************************
1807  *                              GOVERNORS                            *
1808  *********************************************************************/
1809 
1810 /**
1811  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1812  * @policy: cpufreq policy to switch the frequency for.
1813  * @target_freq: New frequency to set (may be approximate).
1814  *
1815  * Carry out a fast frequency switch without sleeping.
1816  *
1817  * The driver's ->fast_switch() callback invoked by this function must be
1818  * suitable for being called from within RCU-sched read-side critical sections
1819  * and it is expected to select the minimum available frequency greater than or
1820  * equal to @target_freq (CPUFREQ_RELATION_L).
1821  *
1822  * This function must not be called if policy->fast_switch_enabled is unset.
1823  *
1824  * Governors calling this function must guarantee that it will never be invoked
1825  * twice in parallel for the same policy and that it will never be called in
1826  * parallel with either ->target() or ->target_index() for the same policy.
1827  *
1828  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1829  * callback to indicate an error condition, the hardware configuration must be
1830  * preserved.
1831  */
1832 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1833 					unsigned int target_freq)
1834 {
1835 	target_freq = clamp_val(target_freq, policy->min, policy->max);
1836 
1837 	return cpufreq_driver->fast_switch(policy, target_freq);
1838 }
1839 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1840 
1841 /* Must set freqs->new to intermediate frequency */
1842 static int __target_intermediate(struct cpufreq_policy *policy,
1843 				 struct cpufreq_freqs *freqs, int index)
1844 {
1845 	int ret;
1846 
1847 	freqs->new = cpufreq_driver->get_intermediate(policy, index);
1848 
1849 	/* We don't need to switch to intermediate freq */
1850 	if (!freqs->new)
1851 		return 0;
1852 
1853 	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1854 		 __func__, policy->cpu, freqs->old, freqs->new);
1855 
1856 	cpufreq_freq_transition_begin(policy, freqs);
1857 	ret = cpufreq_driver->target_intermediate(policy, index);
1858 	cpufreq_freq_transition_end(policy, freqs, ret);
1859 
1860 	if (ret)
1861 		pr_err("%s: Failed to change to intermediate frequency: %d\n",
1862 		       __func__, ret);
1863 
1864 	return ret;
1865 }
1866 
1867 static int __target_index(struct cpufreq_policy *policy,
1868 			  struct cpufreq_frequency_table *freq_table, int index)
1869 {
1870 	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1871 	unsigned int intermediate_freq = 0;
1872 	int retval = -EINVAL;
1873 	bool notify;
1874 
1875 	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1876 	if (notify) {
1877 		/* Handle switching to intermediate frequency */
1878 		if (cpufreq_driver->get_intermediate) {
1879 			retval = __target_intermediate(policy, &freqs, index);
1880 			if (retval)
1881 				return retval;
1882 
1883 			intermediate_freq = freqs.new;
1884 			/* Set old freq to intermediate */
1885 			if (intermediate_freq)
1886 				freqs.old = freqs.new;
1887 		}
1888 
1889 		freqs.new = freq_table[index].frequency;
1890 		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1891 			 __func__, policy->cpu, freqs.old, freqs.new);
1892 
1893 		cpufreq_freq_transition_begin(policy, &freqs);
1894 	}
1895 
1896 	retval = cpufreq_driver->target_index(policy, index);
1897 	if (retval)
1898 		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1899 		       retval);
1900 
1901 	if (notify) {
1902 		cpufreq_freq_transition_end(policy, &freqs, retval);
1903 
1904 		/*
1905 		 * Failed after setting to intermediate freq? Driver should have
1906 		 * reverted back to initial frequency and so should we. Check
1907 		 * here for intermediate_freq instead of get_intermediate, in
1908 		 * case we haven't switched to intermediate freq at all.
1909 		 */
1910 		if (unlikely(retval && intermediate_freq)) {
1911 			freqs.old = intermediate_freq;
1912 			freqs.new = policy->restore_freq;
1913 			cpufreq_freq_transition_begin(policy, &freqs);
1914 			cpufreq_freq_transition_end(policy, &freqs, 0);
1915 		}
1916 	}
1917 
1918 	return retval;
1919 }
1920 
1921 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1922 			    unsigned int target_freq,
1923 			    unsigned int relation)
1924 {
1925 	unsigned int old_target_freq = target_freq;
1926 	struct cpufreq_frequency_table *freq_table;
1927 	int index, retval;
1928 
1929 	if (cpufreq_disabled())
1930 		return -ENODEV;
1931 
1932 	/* Make sure that target_freq is within supported range */
1933 	if (target_freq > policy->max)
1934 		target_freq = policy->max;
1935 	if (target_freq < policy->min)
1936 		target_freq = policy->min;
1937 
1938 	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1939 		 policy->cpu, target_freq, relation, old_target_freq);
1940 
1941 	/*
1942 	 * This might look like a redundant call as we are checking it again
1943 	 * after finding index. But it is left intentionally for cases where
1944 	 * exactly same freq is called again and so we can save on few function
1945 	 * calls.
1946 	 */
1947 	if (target_freq == policy->cur)
1948 		return 0;
1949 
1950 	/* Save last value to restore later on errors */
1951 	policy->restore_freq = policy->cur;
1952 
1953 	if (cpufreq_driver->target)
1954 		return cpufreq_driver->target(policy, target_freq, relation);
1955 
1956 	if (!cpufreq_driver->target_index)
1957 		return -EINVAL;
1958 
1959 	freq_table = cpufreq_frequency_get_table(policy->cpu);
1960 	if (unlikely(!freq_table)) {
1961 		pr_err("%s: Unable to find freq_table\n", __func__);
1962 		return -EINVAL;
1963 	}
1964 
1965 	retval = cpufreq_frequency_table_target(policy, freq_table, target_freq,
1966 						relation, &index);
1967 	if (unlikely(retval)) {
1968 		pr_err("%s: Unable to find matching freq\n", __func__);
1969 		return retval;
1970 	}
1971 
1972 	if (freq_table[index].frequency == policy->cur)
1973 		return 0;
1974 
1975 	return __target_index(policy, freq_table, index);
1976 }
1977 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1978 
1979 int cpufreq_driver_target(struct cpufreq_policy *policy,
1980 			  unsigned int target_freq,
1981 			  unsigned int relation)
1982 {
1983 	int ret = -EINVAL;
1984 
1985 	down_write(&policy->rwsem);
1986 
1987 	ret = __cpufreq_driver_target(policy, target_freq, relation);
1988 
1989 	up_write(&policy->rwsem);
1990 
1991 	return ret;
1992 }
1993 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1994 
1995 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1996 {
1997 	return NULL;
1998 }
1999 
2000 static int cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
2001 {
2002 	int ret;
2003 
2004 	/* Don't start any governor operations if we are entering suspend */
2005 	if (cpufreq_suspended)
2006 		return 0;
2007 	/*
2008 	 * Governor might not be initiated here if ACPI _PPC changed
2009 	 * notification happened, so check it.
2010 	 */
2011 	if (!policy->governor)
2012 		return -EINVAL;
2013 
2014 	if (policy->governor->max_transition_latency &&
2015 	    policy->cpuinfo.transition_latency >
2016 	    policy->governor->max_transition_latency) {
2017 		struct cpufreq_governor *gov = cpufreq_fallback_governor();
2018 
2019 		if (gov) {
2020 			pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2021 				policy->governor->name, gov->name);
2022 			policy->governor = gov;
2023 		} else {
2024 			return -EINVAL;
2025 		}
2026 	}
2027 
2028 	if (event == CPUFREQ_GOV_POLICY_INIT)
2029 		if (!try_module_get(policy->governor->owner))
2030 			return -EINVAL;
2031 
2032 	pr_debug("%s: for CPU %u, event %u\n", __func__, policy->cpu, event);
2033 
2034 	ret = policy->governor->governor(policy, event);
2035 
2036 	if (event == CPUFREQ_GOV_POLICY_INIT) {
2037 		if (ret)
2038 			module_put(policy->governor->owner);
2039 		else
2040 			policy->governor->initialized++;
2041 	} else if (event == CPUFREQ_GOV_POLICY_EXIT) {
2042 		policy->governor->initialized--;
2043 		module_put(policy->governor->owner);
2044 	}
2045 
2046 	return ret;
2047 }
2048 
2049 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2050 {
2051 	int ret;
2052 
2053 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2054 		cpufreq_update_current_freq(policy);
2055 
2056 	ret = cpufreq_governor(policy, CPUFREQ_GOV_START);
2057 	return ret ? ret : cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2058 }
2059 
2060 int cpufreq_register_governor(struct cpufreq_governor *governor)
2061 {
2062 	int err;
2063 
2064 	if (!governor)
2065 		return -EINVAL;
2066 
2067 	if (cpufreq_disabled())
2068 		return -ENODEV;
2069 
2070 	mutex_lock(&cpufreq_governor_mutex);
2071 
2072 	governor->initialized = 0;
2073 	err = -EBUSY;
2074 	if (!find_governor(governor->name)) {
2075 		err = 0;
2076 		list_add(&governor->governor_list, &cpufreq_governor_list);
2077 	}
2078 
2079 	mutex_unlock(&cpufreq_governor_mutex);
2080 	return err;
2081 }
2082 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2083 
2084 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2085 {
2086 	struct cpufreq_policy *policy;
2087 	unsigned long flags;
2088 
2089 	if (!governor)
2090 		return;
2091 
2092 	if (cpufreq_disabled())
2093 		return;
2094 
2095 	/* clear last_governor for all inactive policies */
2096 	read_lock_irqsave(&cpufreq_driver_lock, flags);
2097 	for_each_inactive_policy(policy) {
2098 		if (!strcmp(policy->last_governor, governor->name)) {
2099 			policy->governor = NULL;
2100 			strcpy(policy->last_governor, "\0");
2101 		}
2102 	}
2103 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2104 
2105 	mutex_lock(&cpufreq_governor_mutex);
2106 	list_del(&governor->governor_list);
2107 	mutex_unlock(&cpufreq_governor_mutex);
2108 	return;
2109 }
2110 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2111 
2112 
2113 /*********************************************************************
2114  *                          POLICY INTERFACE                         *
2115  *********************************************************************/
2116 
2117 /**
2118  * cpufreq_get_policy - get the current cpufreq_policy
2119  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2120  *	is written
2121  *
2122  * Reads the current cpufreq policy.
2123  */
2124 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2125 {
2126 	struct cpufreq_policy *cpu_policy;
2127 	if (!policy)
2128 		return -EINVAL;
2129 
2130 	cpu_policy = cpufreq_cpu_get(cpu);
2131 	if (!cpu_policy)
2132 		return -EINVAL;
2133 
2134 	memcpy(policy, cpu_policy, sizeof(*policy));
2135 
2136 	cpufreq_cpu_put(cpu_policy);
2137 	return 0;
2138 }
2139 EXPORT_SYMBOL(cpufreq_get_policy);
2140 
2141 /*
2142  * policy : current policy.
2143  * new_policy: policy to be set.
2144  */
2145 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2146 				struct cpufreq_policy *new_policy)
2147 {
2148 	struct cpufreq_governor *old_gov;
2149 	int ret;
2150 
2151 	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2152 		 new_policy->cpu, new_policy->min, new_policy->max);
2153 
2154 	memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2155 
2156 	/*
2157 	* This check works well when we store new min/max freq attributes,
2158 	* because new_policy is a copy of policy with one field updated.
2159 	*/
2160 	if (new_policy->min > new_policy->max)
2161 		return -EINVAL;
2162 
2163 	/* verify the cpu speed can be set within this limit */
2164 	ret = cpufreq_driver->verify(new_policy);
2165 	if (ret)
2166 		return ret;
2167 
2168 	/* adjust if necessary - all reasons */
2169 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2170 			CPUFREQ_ADJUST, new_policy);
2171 
2172 	/*
2173 	 * verify the cpu speed can be set within this limit, which might be
2174 	 * different to the first one
2175 	 */
2176 	ret = cpufreq_driver->verify(new_policy);
2177 	if (ret)
2178 		return ret;
2179 
2180 	/* notification of the new policy */
2181 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2182 			CPUFREQ_NOTIFY, new_policy);
2183 
2184 	policy->min = new_policy->min;
2185 	policy->max = new_policy->max;
2186 
2187 	pr_debug("new min and max freqs are %u - %u kHz\n",
2188 		 policy->min, policy->max);
2189 
2190 	if (cpufreq_driver->setpolicy) {
2191 		policy->policy = new_policy->policy;
2192 		pr_debug("setting range\n");
2193 		return cpufreq_driver->setpolicy(new_policy);
2194 	}
2195 
2196 	if (new_policy->governor == policy->governor) {
2197 		pr_debug("cpufreq: governor limits update\n");
2198 		return cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2199 	}
2200 
2201 	pr_debug("governor switch\n");
2202 
2203 	/* save old, working values */
2204 	old_gov = policy->governor;
2205 	/* end old governor */
2206 	if (old_gov) {
2207 		cpufreq_stop_governor(policy);
2208 		cpufreq_exit_governor(policy);
2209 	}
2210 
2211 	/* start new governor */
2212 	policy->governor = new_policy->governor;
2213 	ret = cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2214 	if (!ret) {
2215 		ret = cpufreq_start_governor(policy);
2216 		if (!ret) {
2217 			pr_debug("cpufreq: governor change\n");
2218 			return 0;
2219 		}
2220 		cpufreq_exit_governor(policy);
2221 	}
2222 
2223 	/* new governor failed, so re-start old one */
2224 	pr_debug("starting governor %s failed\n", policy->governor->name);
2225 	if (old_gov) {
2226 		policy->governor = old_gov;
2227 		if (cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2228 			policy->governor = NULL;
2229 		else
2230 			cpufreq_start_governor(policy);
2231 	}
2232 
2233 	return ret;
2234 }
2235 
2236 /**
2237  *	cpufreq_update_policy - re-evaluate an existing cpufreq policy
2238  *	@cpu: CPU which shall be re-evaluated
2239  *
2240  *	Useful for policy notifiers which have different necessities
2241  *	at different times.
2242  */
2243 int cpufreq_update_policy(unsigned int cpu)
2244 {
2245 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2246 	struct cpufreq_policy new_policy;
2247 	int ret;
2248 
2249 	if (!policy)
2250 		return -ENODEV;
2251 
2252 	down_write(&policy->rwsem);
2253 
2254 	pr_debug("updating policy for CPU %u\n", cpu);
2255 	memcpy(&new_policy, policy, sizeof(*policy));
2256 	new_policy.min = policy->user_policy.min;
2257 	new_policy.max = policy->user_policy.max;
2258 
2259 	/*
2260 	 * BIOS might change freq behind our back
2261 	 * -> ask driver for current freq and notify governors about a change
2262 	 */
2263 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2264 		if (cpufreq_suspended) {
2265 			ret = -EAGAIN;
2266 			goto unlock;
2267 		}
2268 		new_policy.cur = cpufreq_update_current_freq(policy);
2269 		if (WARN_ON(!new_policy.cur)) {
2270 			ret = -EIO;
2271 			goto unlock;
2272 		}
2273 	}
2274 
2275 	ret = cpufreq_set_policy(policy, &new_policy);
2276 
2277 unlock:
2278 	up_write(&policy->rwsem);
2279 
2280 	cpufreq_cpu_put(policy);
2281 	return ret;
2282 }
2283 EXPORT_SYMBOL(cpufreq_update_policy);
2284 
2285 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2286 					unsigned long action, void *hcpu)
2287 {
2288 	unsigned int cpu = (unsigned long)hcpu;
2289 
2290 	switch (action & ~CPU_TASKS_FROZEN) {
2291 	case CPU_ONLINE:
2292 	case CPU_DOWN_FAILED:
2293 		cpufreq_online(cpu);
2294 		break;
2295 
2296 	case CPU_DOWN_PREPARE:
2297 		cpufreq_offline(cpu);
2298 		break;
2299 	}
2300 	return NOTIFY_OK;
2301 }
2302 
2303 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2304 	.notifier_call = cpufreq_cpu_callback,
2305 };
2306 
2307 /*********************************************************************
2308  *               BOOST						     *
2309  *********************************************************************/
2310 static int cpufreq_boost_set_sw(int state)
2311 {
2312 	struct cpufreq_frequency_table *freq_table;
2313 	struct cpufreq_policy *policy;
2314 	int ret = -EINVAL;
2315 
2316 	for_each_active_policy(policy) {
2317 		freq_table = cpufreq_frequency_get_table(policy->cpu);
2318 		if (freq_table) {
2319 			ret = cpufreq_frequency_table_cpuinfo(policy,
2320 							freq_table);
2321 			if (ret) {
2322 				pr_err("%s: Policy frequency update failed\n",
2323 				       __func__);
2324 				break;
2325 			}
2326 
2327 			down_write(&policy->rwsem);
2328 			policy->user_policy.max = policy->max;
2329 			cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2330 			up_write(&policy->rwsem);
2331 		}
2332 	}
2333 
2334 	return ret;
2335 }
2336 
2337 int cpufreq_boost_trigger_state(int state)
2338 {
2339 	unsigned long flags;
2340 	int ret = 0;
2341 
2342 	if (cpufreq_driver->boost_enabled == state)
2343 		return 0;
2344 
2345 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2346 	cpufreq_driver->boost_enabled = state;
2347 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2348 
2349 	ret = cpufreq_driver->set_boost(state);
2350 	if (ret) {
2351 		write_lock_irqsave(&cpufreq_driver_lock, flags);
2352 		cpufreq_driver->boost_enabled = !state;
2353 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2354 
2355 		pr_err("%s: Cannot %s BOOST\n",
2356 		       __func__, state ? "enable" : "disable");
2357 	}
2358 
2359 	return ret;
2360 }
2361 
2362 static bool cpufreq_boost_supported(void)
2363 {
2364 	return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2365 }
2366 
2367 static int create_boost_sysfs_file(void)
2368 {
2369 	int ret;
2370 
2371 	ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2372 	if (ret)
2373 		pr_err("%s: cannot register global BOOST sysfs file\n",
2374 		       __func__);
2375 
2376 	return ret;
2377 }
2378 
2379 static void remove_boost_sysfs_file(void)
2380 {
2381 	if (cpufreq_boost_supported())
2382 		sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2383 }
2384 
2385 int cpufreq_enable_boost_support(void)
2386 {
2387 	if (!cpufreq_driver)
2388 		return -EINVAL;
2389 
2390 	if (cpufreq_boost_supported())
2391 		return 0;
2392 
2393 	cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2394 
2395 	/* This will get removed on driver unregister */
2396 	return create_boost_sysfs_file();
2397 }
2398 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2399 
2400 int cpufreq_boost_enabled(void)
2401 {
2402 	return cpufreq_driver->boost_enabled;
2403 }
2404 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2405 
2406 /*********************************************************************
2407  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2408  *********************************************************************/
2409 
2410 /**
2411  * cpufreq_register_driver - register a CPU Frequency driver
2412  * @driver_data: A struct cpufreq_driver containing the values#
2413  * submitted by the CPU Frequency driver.
2414  *
2415  * Registers a CPU Frequency driver to this core code. This code
2416  * returns zero on success, -EEXIST when another driver got here first
2417  * (and isn't unregistered in the meantime).
2418  *
2419  */
2420 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2421 {
2422 	unsigned long flags;
2423 	int ret;
2424 
2425 	if (cpufreq_disabled())
2426 		return -ENODEV;
2427 
2428 	if (!driver_data || !driver_data->verify || !driver_data->init ||
2429 	    !(driver_data->setpolicy || driver_data->target_index ||
2430 		    driver_data->target) ||
2431 	     (driver_data->setpolicy && (driver_data->target_index ||
2432 		    driver_data->target)) ||
2433 	     (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2434 		return -EINVAL;
2435 
2436 	pr_debug("trying to register driver %s\n", driver_data->name);
2437 
2438 	/* Protect against concurrent CPU online/offline. */
2439 	get_online_cpus();
2440 
2441 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2442 	if (cpufreq_driver) {
2443 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2444 		ret = -EEXIST;
2445 		goto out;
2446 	}
2447 	cpufreq_driver = driver_data;
2448 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2449 
2450 	if (driver_data->setpolicy)
2451 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2452 
2453 	if (cpufreq_boost_supported()) {
2454 		ret = create_boost_sysfs_file();
2455 		if (ret)
2456 			goto err_null_driver;
2457 	}
2458 
2459 	ret = subsys_interface_register(&cpufreq_interface);
2460 	if (ret)
2461 		goto err_boost_unreg;
2462 
2463 	if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2464 	    list_empty(&cpufreq_policy_list)) {
2465 		/* if all ->init() calls failed, unregister */
2466 		pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2467 			 driver_data->name);
2468 		goto err_if_unreg;
2469 	}
2470 
2471 	register_hotcpu_notifier(&cpufreq_cpu_notifier);
2472 	pr_debug("driver %s up and running\n", driver_data->name);
2473 	goto out;
2474 
2475 err_if_unreg:
2476 	subsys_interface_unregister(&cpufreq_interface);
2477 err_boost_unreg:
2478 	remove_boost_sysfs_file();
2479 err_null_driver:
2480 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2481 	cpufreq_driver = NULL;
2482 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2483 out:
2484 	put_online_cpus();
2485 	return ret;
2486 }
2487 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2488 
2489 /**
2490  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2491  *
2492  * Unregister the current CPUFreq driver. Only call this if you have
2493  * the right to do so, i.e. if you have succeeded in initialising before!
2494  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2495  * currently not initialised.
2496  */
2497 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2498 {
2499 	unsigned long flags;
2500 
2501 	if (!cpufreq_driver || (driver != cpufreq_driver))
2502 		return -EINVAL;
2503 
2504 	pr_debug("unregistering driver %s\n", driver->name);
2505 
2506 	/* Protect against concurrent cpu hotplug */
2507 	get_online_cpus();
2508 	subsys_interface_unregister(&cpufreq_interface);
2509 	remove_boost_sysfs_file();
2510 	unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2511 
2512 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2513 
2514 	cpufreq_driver = NULL;
2515 
2516 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2517 	put_online_cpus();
2518 
2519 	return 0;
2520 }
2521 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2522 
2523 /*
2524  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2525  * or mutexes when secondary CPUs are halted.
2526  */
2527 static struct syscore_ops cpufreq_syscore_ops = {
2528 	.shutdown = cpufreq_suspend,
2529 };
2530 
2531 struct kobject *cpufreq_global_kobject;
2532 EXPORT_SYMBOL(cpufreq_global_kobject);
2533 
2534 static int __init cpufreq_core_init(void)
2535 {
2536 	if (cpufreq_disabled())
2537 		return -ENODEV;
2538 
2539 	cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2540 	BUG_ON(!cpufreq_global_kobject);
2541 
2542 	register_syscore_ops(&cpufreq_syscore_ops);
2543 
2544 	return 0;
2545 }
2546 core_initcall(cpufreq_core_init);
2547