xref: /openbmc/linux/drivers/cpufreq/cpufreq.c (revision 4cff79e9)
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 unsigned int __cpufreq_get(struct cpufreq_policy *policy);
78 static int cpufreq_init_governor(struct cpufreq_policy *policy);
79 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
80 static int cpufreq_start_governor(struct cpufreq_policy *policy);
81 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
82 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
83 
84 /**
85  * Two notifier lists: the "policy" list is involved in the
86  * validation process for a new CPU frequency policy; the
87  * "transition" list for kernel code that needs to handle
88  * changes to devices when the CPU clock speed changes.
89  * The mutex locks both lists.
90  */
91 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
92 static struct srcu_notifier_head cpufreq_transition_notifier_list;
93 
94 static bool init_cpufreq_transition_notifier_list_called;
95 static int __init init_cpufreq_transition_notifier_list(void)
96 {
97 	srcu_init_notifier_head(&cpufreq_transition_notifier_list);
98 	init_cpufreq_transition_notifier_list_called = true;
99 	return 0;
100 }
101 pure_initcall(init_cpufreq_transition_notifier_list);
102 
103 static int off __read_mostly;
104 static int cpufreq_disabled(void)
105 {
106 	return off;
107 }
108 void disable_cpufreq(void)
109 {
110 	off = 1;
111 }
112 static DEFINE_MUTEX(cpufreq_governor_mutex);
113 
114 bool have_governor_per_policy(void)
115 {
116 	return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
117 }
118 EXPORT_SYMBOL_GPL(have_governor_per_policy);
119 
120 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
121 {
122 	if (have_governor_per_policy())
123 		return &policy->kobj;
124 	else
125 		return cpufreq_global_kobject;
126 }
127 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
128 
129 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
130 {
131 	u64 idle_time;
132 	u64 cur_wall_time;
133 	u64 busy_time;
134 
135 	cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
136 
137 	busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
138 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
139 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
140 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
141 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
142 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
143 
144 	idle_time = cur_wall_time - busy_time;
145 	if (wall)
146 		*wall = div_u64(cur_wall_time, NSEC_PER_USEC);
147 
148 	return div_u64(idle_time, NSEC_PER_USEC);
149 }
150 
151 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
152 {
153 	u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
154 
155 	if (idle_time == -1ULL)
156 		return get_cpu_idle_time_jiffy(cpu, wall);
157 	else if (!io_busy)
158 		idle_time += get_cpu_iowait_time_us(cpu, wall);
159 
160 	return idle_time;
161 }
162 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
163 
164 __weak void arch_set_freq_scale(struct cpumask *cpus, unsigned long cur_freq,
165 		unsigned long max_freq)
166 {
167 }
168 EXPORT_SYMBOL_GPL(arch_set_freq_scale);
169 
170 /*
171  * This is a generic cpufreq init() routine which can be used by cpufreq
172  * drivers of SMP systems. It will do following:
173  * - validate & show freq table passed
174  * - set policies transition latency
175  * - policy->cpus with all possible CPUs
176  */
177 int cpufreq_generic_init(struct cpufreq_policy *policy,
178 		struct cpufreq_frequency_table *table,
179 		unsigned int transition_latency)
180 {
181 	policy->freq_table = table;
182 	policy->cpuinfo.transition_latency = transition_latency;
183 
184 	/*
185 	 * The driver only supports the SMP configuration where all processors
186 	 * share the clock and voltage and clock.
187 	 */
188 	cpumask_setall(policy->cpus);
189 
190 	return 0;
191 }
192 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
193 
194 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
195 {
196 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
197 
198 	return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
199 }
200 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
201 
202 unsigned int cpufreq_generic_get(unsigned int cpu)
203 {
204 	struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
205 
206 	if (!policy || IS_ERR(policy->clk)) {
207 		pr_err("%s: No %s associated to cpu: %d\n",
208 		       __func__, policy ? "clk" : "policy", cpu);
209 		return 0;
210 	}
211 
212 	return clk_get_rate(policy->clk) / 1000;
213 }
214 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
215 
216 /**
217  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
218  *
219  * @cpu: cpu to find policy for.
220  *
221  * This returns policy for 'cpu', returns NULL if it doesn't exist.
222  * It also increments the kobject reference count to mark it busy and so would
223  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
224  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
225  * freed as that depends on the kobj count.
226  *
227  * Return: A valid policy on success, otherwise NULL on failure.
228  */
229 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
230 {
231 	struct cpufreq_policy *policy = NULL;
232 	unsigned long flags;
233 
234 	if (WARN_ON(cpu >= nr_cpu_ids))
235 		return NULL;
236 
237 	/* get the cpufreq driver */
238 	read_lock_irqsave(&cpufreq_driver_lock, flags);
239 
240 	if (cpufreq_driver) {
241 		/* get the CPU */
242 		policy = cpufreq_cpu_get_raw(cpu);
243 		if (policy)
244 			kobject_get(&policy->kobj);
245 	}
246 
247 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
248 
249 	return policy;
250 }
251 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
252 
253 /**
254  * cpufreq_cpu_put: Decrements the usage count of a policy
255  *
256  * @policy: policy earlier returned by cpufreq_cpu_get().
257  *
258  * This decrements the kobject reference count incremented earlier by calling
259  * cpufreq_cpu_get().
260  */
261 void cpufreq_cpu_put(struct cpufreq_policy *policy)
262 {
263 	kobject_put(&policy->kobj);
264 }
265 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
266 
267 /*********************************************************************
268  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
269  *********************************************************************/
270 
271 /**
272  * adjust_jiffies - adjust the system "loops_per_jiffy"
273  *
274  * This function alters the system "loops_per_jiffy" for the clock
275  * speed change. Note that loops_per_jiffy cannot be updated on SMP
276  * systems as each CPU might be scaled differently. So, use the arch
277  * per-CPU loops_per_jiffy value wherever possible.
278  */
279 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
280 {
281 #ifndef CONFIG_SMP
282 	static unsigned long l_p_j_ref;
283 	static unsigned int l_p_j_ref_freq;
284 
285 	if (ci->flags & CPUFREQ_CONST_LOOPS)
286 		return;
287 
288 	if (!l_p_j_ref_freq) {
289 		l_p_j_ref = loops_per_jiffy;
290 		l_p_j_ref_freq = ci->old;
291 		pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
292 			 l_p_j_ref, l_p_j_ref_freq);
293 	}
294 	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
295 		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
296 								ci->new);
297 		pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
298 			 loops_per_jiffy, ci->new);
299 	}
300 #endif
301 }
302 
303 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
304 		struct cpufreq_freqs *freqs, unsigned int state)
305 {
306 	BUG_ON(irqs_disabled());
307 
308 	if (cpufreq_disabled())
309 		return;
310 
311 	freqs->flags = cpufreq_driver->flags;
312 	pr_debug("notification %u of frequency transition to %u kHz\n",
313 		 state, freqs->new);
314 
315 	switch (state) {
316 
317 	case CPUFREQ_PRECHANGE:
318 		/* detect if the driver reported a value as "old frequency"
319 		 * which is not equal to what the cpufreq core thinks is
320 		 * "old frequency".
321 		 */
322 		if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
323 			if ((policy) && (policy->cpu == freqs->cpu) &&
324 			    (policy->cur) && (policy->cur != freqs->old)) {
325 				pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
326 					 freqs->old, policy->cur);
327 				freqs->old = policy->cur;
328 			}
329 		}
330 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
331 				CPUFREQ_PRECHANGE, freqs);
332 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
333 		break;
334 
335 	case CPUFREQ_POSTCHANGE:
336 		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
337 		pr_debug("FREQ: %lu - CPU: %lu\n",
338 			 (unsigned long)freqs->new, (unsigned long)freqs->cpu);
339 		trace_cpu_frequency(freqs->new, freqs->cpu);
340 		cpufreq_stats_record_transition(policy, freqs->new);
341 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
342 				CPUFREQ_POSTCHANGE, freqs);
343 		if (likely(policy) && likely(policy->cpu == freqs->cpu))
344 			policy->cur = freqs->new;
345 		break;
346 	}
347 }
348 
349 /**
350  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
351  * on frequency transition.
352  *
353  * This function calls the transition notifiers and the "adjust_jiffies"
354  * function. It is called twice on all CPU frequency changes that have
355  * external effects.
356  */
357 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
358 		struct cpufreq_freqs *freqs, unsigned int state)
359 {
360 	for_each_cpu(freqs->cpu, policy->cpus)
361 		__cpufreq_notify_transition(policy, freqs, state);
362 }
363 
364 /* Do post notifications when there are chances that transition has failed */
365 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
366 		struct cpufreq_freqs *freqs, int transition_failed)
367 {
368 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
369 	if (!transition_failed)
370 		return;
371 
372 	swap(freqs->old, freqs->new);
373 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
374 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
375 }
376 
377 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
378 		struct cpufreq_freqs *freqs)
379 {
380 
381 	/*
382 	 * Catch double invocations of _begin() which lead to self-deadlock.
383 	 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
384 	 * doesn't invoke _begin() on their behalf, and hence the chances of
385 	 * double invocations are very low. Moreover, there are scenarios
386 	 * where these checks can emit false-positive warnings in these
387 	 * drivers; so we avoid that by skipping them altogether.
388 	 */
389 	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
390 				&& current == policy->transition_task);
391 
392 wait:
393 	wait_event(policy->transition_wait, !policy->transition_ongoing);
394 
395 	spin_lock(&policy->transition_lock);
396 
397 	if (unlikely(policy->transition_ongoing)) {
398 		spin_unlock(&policy->transition_lock);
399 		goto wait;
400 	}
401 
402 	policy->transition_ongoing = true;
403 	policy->transition_task = current;
404 
405 	spin_unlock(&policy->transition_lock);
406 
407 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
408 }
409 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
410 
411 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
412 		struct cpufreq_freqs *freqs, int transition_failed)
413 {
414 	if (unlikely(WARN_ON(!policy->transition_ongoing)))
415 		return;
416 
417 	cpufreq_notify_post_transition(policy, freqs, transition_failed);
418 
419 	policy->transition_ongoing = false;
420 	policy->transition_task = NULL;
421 
422 	wake_up(&policy->transition_wait);
423 }
424 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
425 
426 /*
427  * Fast frequency switching status count.  Positive means "enabled", negative
428  * means "disabled" and 0 means "not decided yet".
429  */
430 static int cpufreq_fast_switch_count;
431 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
432 
433 static void cpufreq_list_transition_notifiers(void)
434 {
435 	struct notifier_block *nb;
436 
437 	pr_info("Registered transition notifiers:\n");
438 
439 	mutex_lock(&cpufreq_transition_notifier_list.mutex);
440 
441 	for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
442 		pr_info("%pF\n", nb->notifier_call);
443 
444 	mutex_unlock(&cpufreq_transition_notifier_list.mutex);
445 }
446 
447 /**
448  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
449  * @policy: cpufreq policy to enable fast frequency switching for.
450  *
451  * Try to enable fast frequency switching for @policy.
452  *
453  * The attempt will fail if there is at least one transition notifier registered
454  * at this point, as fast frequency switching is quite fundamentally at odds
455  * with transition notifiers.  Thus if successful, it will make registration of
456  * transition notifiers fail going forward.
457  */
458 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
459 {
460 	lockdep_assert_held(&policy->rwsem);
461 
462 	if (!policy->fast_switch_possible)
463 		return;
464 
465 	mutex_lock(&cpufreq_fast_switch_lock);
466 	if (cpufreq_fast_switch_count >= 0) {
467 		cpufreq_fast_switch_count++;
468 		policy->fast_switch_enabled = true;
469 	} else {
470 		pr_warn("CPU%u: Fast frequency switching not enabled\n",
471 			policy->cpu);
472 		cpufreq_list_transition_notifiers();
473 	}
474 	mutex_unlock(&cpufreq_fast_switch_lock);
475 }
476 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
477 
478 /**
479  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
480  * @policy: cpufreq policy to disable fast frequency switching for.
481  */
482 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
483 {
484 	mutex_lock(&cpufreq_fast_switch_lock);
485 	if (policy->fast_switch_enabled) {
486 		policy->fast_switch_enabled = false;
487 		if (!WARN_ON(cpufreq_fast_switch_count <= 0))
488 			cpufreq_fast_switch_count--;
489 	}
490 	mutex_unlock(&cpufreq_fast_switch_lock);
491 }
492 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
493 
494 /**
495  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
496  * one.
497  * @target_freq: target frequency to resolve.
498  *
499  * The target to driver frequency mapping is cached in the policy.
500  *
501  * Return: Lowest driver-supported frequency greater than or equal to the
502  * given target_freq, subject to policy (min/max) and driver limitations.
503  */
504 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
505 					 unsigned int target_freq)
506 {
507 	target_freq = clamp_val(target_freq, policy->min, policy->max);
508 	policy->cached_target_freq = target_freq;
509 
510 	if (cpufreq_driver->target_index) {
511 		int idx;
512 
513 		idx = cpufreq_frequency_table_target(policy, target_freq,
514 						     CPUFREQ_RELATION_L);
515 		policy->cached_resolved_idx = idx;
516 		return policy->freq_table[idx].frequency;
517 	}
518 
519 	if (cpufreq_driver->resolve_freq)
520 		return cpufreq_driver->resolve_freq(policy, target_freq);
521 
522 	return target_freq;
523 }
524 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
525 
526 unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
527 {
528 	unsigned int latency;
529 
530 	if (policy->transition_delay_us)
531 		return policy->transition_delay_us;
532 
533 	latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
534 	if (latency) {
535 		/*
536 		 * For platforms that can change the frequency very fast (< 10
537 		 * us), the above formula gives a decent transition delay. But
538 		 * for platforms where transition_latency is in milliseconds, it
539 		 * ends up giving unrealistic values.
540 		 *
541 		 * Cap the default transition delay to 10 ms, which seems to be
542 		 * a reasonable amount of time after which we should reevaluate
543 		 * the frequency.
544 		 */
545 		return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
546 	}
547 
548 	return LATENCY_MULTIPLIER;
549 }
550 EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
551 
552 /*********************************************************************
553  *                          SYSFS INTERFACE                          *
554  *********************************************************************/
555 static ssize_t show_boost(struct kobject *kobj,
556 				 struct attribute *attr, char *buf)
557 {
558 	return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
559 }
560 
561 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
562 				  const char *buf, size_t count)
563 {
564 	int ret, enable;
565 
566 	ret = sscanf(buf, "%d", &enable);
567 	if (ret != 1 || enable < 0 || enable > 1)
568 		return -EINVAL;
569 
570 	if (cpufreq_boost_trigger_state(enable)) {
571 		pr_err("%s: Cannot %s BOOST!\n",
572 		       __func__, enable ? "enable" : "disable");
573 		return -EINVAL;
574 	}
575 
576 	pr_debug("%s: cpufreq BOOST %s\n",
577 		 __func__, enable ? "enabled" : "disabled");
578 
579 	return count;
580 }
581 define_one_global_rw(boost);
582 
583 static struct cpufreq_governor *find_governor(const char *str_governor)
584 {
585 	struct cpufreq_governor *t;
586 
587 	for_each_governor(t)
588 		if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
589 			return t;
590 
591 	return NULL;
592 }
593 
594 /**
595  * cpufreq_parse_governor - parse a governor string
596  */
597 static int cpufreq_parse_governor(char *str_governor,
598 				  struct cpufreq_policy *policy)
599 {
600 	if (cpufreq_driver->setpolicy) {
601 		if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
602 			policy->policy = CPUFREQ_POLICY_PERFORMANCE;
603 			return 0;
604 		}
605 
606 		if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
607 			policy->policy = CPUFREQ_POLICY_POWERSAVE;
608 			return 0;
609 		}
610 	} else {
611 		struct cpufreq_governor *t;
612 
613 		mutex_lock(&cpufreq_governor_mutex);
614 
615 		t = find_governor(str_governor);
616 		if (!t) {
617 			int ret;
618 
619 			mutex_unlock(&cpufreq_governor_mutex);
620 
621 			ret = request_module("cpufreq_%s", str_governor);
622 			if (ret)
623 				return -EINVAL;
624 
625 			mutex_lock(&cpufreq_governor_mutex);
626 
627 			t = find_governor(str_governor);
628 		}
629 		if (t && !try_module_get(t->owner))
630 			t = NULL;
631 
632 		mutex_unlock(&cpufreq_governor_mutex);
633 
634 		if (t) {
635 			policy->governor = t;
636 			return 0;
637 		}
638 	}
639 
640 	return -EINVAL;
641 }
642 
643 /**
644  * cpufreq_per_cpu_attr_read() / show_##file_name() -
645  * print out cpufreq information
646  *
647  * Write out information from cpufreq_driver->policy[cpu]; object must be
648  * "unsigned int".
649  */
650 
651 #define show_one(file_name, object)			\
652 static ssize_t show_##file_name				\
653 (struct cpufreq_policy *policy, char *buf)		\
654 {							\
655 	return sprintf(buf, "%u\n", policy->object);	\
656 }
657 
658 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
659 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
660 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
661 show_one(scaling_min_freq, min);
662 show_one(scaling_max_freq, max);
663 
664 __weak unsigned int arch_freq_get_on_cpu(int cpu)
665 {
666 	return 0;
667 }
668 
669 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
670 {
671 	ssize_t ret;
672 	unsigned int freq;
673 
674 	freq = arch_freq_get_on_cpu(policy->cpu);
675 	if (freq)
676 		ret = sprintf(buf, "%u\n", freq);
677 	else if (cpufreq_driver && cpufreq_driver->setpolicy &&
678 			cpufreq_driver->get)
679 		ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
680 	else
681 		ret = sprintf(buf, "%u\n", policy->cur);
682 	return ret;
683 }
684 
685 static int cpufreq_set_policy(struct cpufreq_policy *policy,
686 				struct cpufreq_policy *new_policy);
687 
688 /**
689  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
690  */
691 #define store_one(file_name, object)			\
692 static ssize_t store_##file_name					\
693 (struct cpufreq_policy *policy, const char *buf, size_t count)		\
694 {									\
695 	int ret, temp;							\
696 	struct cpufreq_policy new_policy;				\
697 									\
698 	memcpy(&new_policy, policy, sizeof(*policy));			\
699 									\
700 	ret = sscanf(buf, "%u", &new_policy.object);			\
701 	if (ret != 1)							\
702 		return -EINVAL;						\
703 									\
704 	temp = new_policy.object;					\
705 	ret = cpufreq_set_policy(policy, &new_policy);		\
706 	if (!ret)							\
707 		policy->user_policy.object = temp;			\
708 									\
709 	return ret ? ret : count;					\
710 }
711 
712 store_one(scaling_min_freq, min);
713 store_one(scaling_max_freq, max);
714 
715 /**
716  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
717  */
718 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
719 					char *buf)
720 {
721 	unsigned int cur_freq = __cpufreq_get(policy);
722 
723 	if (cur_freq)
724 		return sprintf(buf, "%u\n", cur_freq);
725 
726 	return sprintf(buf, "<unknown>\n");
727 }
728 
729 /**
730  * show_scaling_governor - show the current policy for the specified CPU
731  */
732 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
733 {
734 	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
735 		return sprintf(buf, "powersave\n");
736 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
737 		return sprintf(buf, "performance\n");
738 	else if (policy->governor)
739 		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
740 				policy->governor->name);
741 	return -EINVAL;
742 }
743 
744 /**
745  * store_scaling_governor - store policy for the specified CPU
746  */
747 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
748 					const char *buf, size_t count)
749 {
750 	int ret;
751 	char	str_governor[16];
752 	struct cpufreq_policy new_policy;
753 
754 	memcpy(&new_policy, policy, sizeof(*policy));
755 
756 	ret = sscanf(buf, "%15s", str_governor);
757 	if (ret != 1)
758 		return -EINVAL;
759 
760 	if (cpufreq_parse_governor(str_governor, &new_policy))
761 		return -EINVAL;
762 
763 	ret = cpufreq_set_policy(policy, &new_policy);
764 
765 	if (new_policy.governor)
766 		module_put(new_policy.governor->owner);
767 
768 	return ret ? ret : count;
769 }
770 
771 /**
772  * show_scaling_driver - show the cpufreq driver currently loaded
773  */
774 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
775 {
776 	return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
777 }
778 
779 /**
780  * show_scaling_available_governors - show the available CPUfreq governors
781  */
782 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
783 						char *buf)
784 {
785 	ssize_t i = 0;
786 	struct cpufreq_governor *t;
787 
788 	if (!has_target()) {
789 		i += sprintf(buf, "performance powersave");
790 		goto out;
791 	}
792 
793 	for_each_governor(t) {
794 		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
795 		    - (CPUFREQ_NAME_LEN + 2)))
796 			goto out;
797 		i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
798 	}
799 out:
800 	i += sprintf(&buf[i], "\n");
801 	return i;
802 }
803 
804 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
805 {
806 	ssize_t i = 0;
807 	unsigned int cpu;
808 
809 	for_each_cpu(cpu, mask) {
810 		if (i)
811 			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
812 		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
813 		if (i >= (PAGE_SIZE - 5))
814 			break;
815 	}
816 	i += sprintf(&buf[i], "\n");
817 	return i;
818 }
819 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
820 
821 /**
822  * show_related_cpus - show the CPUs affected by each transition even if
823  * hw coordination is in use
824  */
825 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
826 {
827 	return cpufreq_show_cpus(policy->related_cpus, buf);
828 }
829 
830 /**
831  * show_affected_cpus - show the CPUs affected by each transition
832  */
833 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
834 {
835 	return cpufreq_show_cpus(policy->cpus, buf);
836 }
837 
838 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
839 					const char *buf, size_t count)
840 {
841 	unsigned int freq = 0;
842 	unsigned int ret;
843 
844 	if (!policy->governor || !policy->governor->store_setspeed)
845 		return -EINVAL;
846 
847 	ret = sscanf(buf, "%u", &freq);
848 	if (ret != 1)
849 		return -EINVAL;
850 
851 	policy->governor->store_setspeed(policy, freq);
852 
853 	return count;
854 }
855 
856 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
857 {
858 	if (!policy->governor || !policy->governor->show_setspeed)
859 		return sprintf(buf, "<unsupported>\n");
860 
861 	return policy->governor->show_setspeed(policy, buf);
862 }
863 
864 /**
865  * show_bios_limit - show the current cpufreq HW/BIOS limitation
866  */
867 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
868 {
869 	unsigned int limit;
870 	int ret;
871 	if (cpufreq_driver->bios_limit) {
872 		ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
873 		if (!ret)
874 			return sprintf(buf, "%u\n", limit);
875 	}
876 	return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
877 }
878 
879 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
880 cpufreq_freq_attr_ro(cpuinfo_min_freq);
881 cpufreq_freq_attr_ro(cpuinfo_max_freq);
882 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
883 cpufreq_freq_attr_ro(scaling_available_governors);
884 cpufreq_freq_attr_ro(scaling_driver);
885 cpufreq_freq_attr_ro(scaling_cur_freq);
886 cpufreq_freq_attr_ro(bios_limit);
887 cpufreq_freq_attr_ro(related_cpus);
888 cpufreq_freq_attr_ro(affected_cpus);
889 cpufreq_freq_attr_rw(scaling_min_freq);
890 cpufreq_freq_attr_rw(scaling_max_freq);
891 cpufreq_freq_attr_rw(scaling_governor);
892 cpufreq_freq_attr_rw(scaling_setspeed);
893 
894 static struct attribute *default_attrs[] = {
895 	&cpuinfo_min_freq.attr,
896 	&cpuinfo_max_freq.attr,
897 	&cpuinfo_transition_latency.attr,
898 	&scaling_min_freq.attr,
899 	&scaling_max_freq.attr,
900 	&affected_cpus.attr,
901 	&related_cpus.attr,
902 	&scaling_governor.attr,
903 	&scaling_driver.attr,
904 	&scaling_available_governors.attr,
905 	&scaling_setspeed.attr,
906 	NULL
907 };
908 
909 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
910 #define to_attr(a) container_of(a, struct freq_attr, attr)
911 
912 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
913 {
914 	struct cpufreq_policy *policy = to_policy(kobj);
915 	struct freq_attr *fattr = to_attr(attr);
916 	ssize_t ret;
917 
918 	down_read(&policy->rwsem);
919 	ret = fattr->show(policy, buf);
920 	up_read(&policy->rwsem);
921 
922 	return ret;
923 }
924 
925 static ssize_t store(struct kobject *kobj, struct attribute *attr,
926 		     const char *buf, size_t count)
927 {
928 	struct cpufreq_policy *policy = to_policy(kobj);
929 	struct freq_attr *fattr = to_attr(attr);
930 	ssize_t ret = -EINVAL;
931 
932 	cpus_read_lock();
933 
934 	if (cpu_online(policy->cpu)) {
935 		down_write(&policy->rwsem);
936 		ret = fattr->store(policy, buf, count);
937 		up_write(&policy->rwsem);
938 	}
939 
940 	cpus_read_unlock();
941 
942 	return ret;
943 }
944 
945 static void cpufreq_sysfs_release(struct kobject *kobj)
946 {
947 	struct cpufreq_policy *policy = to_policy(kobj);
948 	pr_debug("last reference is dropped\n");
949 	complete(&policy->kobj_unregister);
950 }
951 
952 static const struct sysfs_ops sysfs_ops = {
953 	.show	= show,
954 	.store	= store,
955 };
956 
957 static struct kobj_type ktype_cpufreq = {
958 	.sysfs_ops	= &sysfs_ops,
959 	.default_attrs	= default_attrs,
960 	.release	= cpufreq_sysfs_release,
961 };
962 
963 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
964 {
965 	struct device *dev = get_cpu_device(cpu);
966 
967 	if (!dev)
968 		return;
969 
970 	if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
971 		return;
972 
973 	dev_dbg(dev, "%s: Adding symlink\n", __func__);
974 	if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
975 		dev_err(dev, "cpufreq symlink creation failed\n");
976 }
977 
978 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
979 				   struct device *dev)
980 {
981 	dev_dbg(dev, "%s: Removing symlink\n", __func__);
982 	sysfs_remove_link(&dev->kobj, "cpufreq");
983 }
984 
985 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
986 {
987 	struct freq_attr **drv_attr;
988 	int ret = 0;
989 
990 	/* set up files for this cpu device */
991 	drv_attr = cpufreq_driver->attr;
992 	while (drv_attr && *drv_attr) {
993 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
994 		if (ret)
995 			return ret;
996 		drv_attr++;
997 	}
998 	if (cpufreq_driver->get) {
999 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1000 		if (ret)
1001 			return ret;
1002 	}
1003 
1004 	ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1005 	if (ret)
1006 		return ret;
1007 
1008 	if (cpufreq_driver->bios_limit) {
1009 		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1010 		if (ret)
1011 			return ret;
1012 	}
1013 
1014 	return 0;
1015 }
1016 
1017 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1018 {
1019 	return NULL;
1020 }
1021 
1022 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1023 {
1024 	struct cpufreq_governor *gov = NULL;
1025 	struct cpufreq_policy new_policy;
1026 
1027 	memcpy(&new_policy, policy, sizeof(*policy));
1028 
1029 	/* Update governor of new_policy to the governor used before hotplug */
1030 	gov = find_governor(policy->last_governor);
1031 	if (gov) {
1032 		pr_debug("Restoring governor %s for cpu %d\n",
1033 				policy->governor->name, policy->cpu);
1034 	} else {
1035 		gov = cpufreq_default_governor();
1036 		if (!gov)
1037 			return -ENODATA;
1038 	}
1039 
1040 	new_policy.governor = gov;
1041 
1042 	/* Use the default policy if there is no last_policy. */
1043 	if (cpufreq_driver->setpolicy) {
1044 		if (policy->last_policy)
1045 			new_policy.policy = policy->last_policy;
1046 		else
1047 			cpufreq_parse_governor(gov->name, &new_policy);
1048 	}
1049 	/* set default policy */
1050 	return cpufreq_set_policy(policy, &new_policy);
1051 }
1052 
1053 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1054 {
1055 	int ret = 0;
1056 
1057 	/* Has this CPU been taken care of already? */
1058 	if (cpumask_test_cpu(cpu, policy->cpus))
1059 		return 0;
1060 
1061 	down_write(&policy->rwsem);
1062 	if (has_target())
1063 		cpufreq_stop_governor(policy);
1064 
1065 	cpumask_set_cpu(cpu, policy->cpus);
1066 
1067 	if (has_target()) {
1068 		ret = cpufreq_start_governor(policy);
1069 		if (ret)
1070 			pr_err("%s: Failed to start governor\n", __func__);
1071 	}
1072 	up_write(&policy->rwsem);
1073 	return ret;
1074 }
1075 
1076 static void handle_update(struct work_struct *work)
1077 {
1078 	struct cpufreq_policy *policy =
1079 		container_of(work, struct cpufreq_policy, update);
1080 	unsigned int cpu = policy->cpu;
1081 	pr_debug("handle_update for cpu %u called\n", cpu);
1082 	cpufreq_update_policy(cpu);
1083 }
1084 
1085 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1086 {
1087 	struct cpufreq_policy *policy;
1088 	int ret;
1089 
1090 	policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1091 	if (!policy)
1092 		return NULL;
1093 
1094 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1095 		goto err_free_policy;
1096 
1097 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1098 		goto err_free_cpumask;
1099 
1100 	if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1101 		goto err_free_rcpumask;
1102 
1103 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1104 				   cpufreq_global_kobject, "policy%u", cpu);
1105 	if (ret) {
1106 		pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1107 		goto err_free_real_cpus;
1108 	}
1109 
1110 	INIT_LIST_HEAD(&policy->policy_list);
1111 	init_rwsem(&policy->rwsem);
1112 	spin_lock_init(&policy->transition_lock);
1113 	init_waitqueue_head(&policy->transition_wait);
1114 	init_completion(&policy->kobj_unregister);
1115 	INIT_WORK(&policy->update, handle_update);
1116 
1117 	policy->cpu = cpu;
1118 	return policy;
1119 
1120 err_free_real_cpus:
1121 	free_cpumask_var(policy->real_cpus);
1122 err_free_rcpumask:
1123 	free_cpumask_var(policy->related_cpus);
1124 err_free_cpumask:
1125 	free_cpumask_var(policy->cpus);
1126 err_free_policy:
1127 	kfree(policy);
1128 
1129 	return NULL;
1130 }
1131 
1132 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1133 {
1134 	struct kobject *kobj;
1135 	struct completion *cmp;
1136 
1137 	down_write(&policy->rwsem);
1138 	cpufreq_stats_free_table(policy);
1139 	kobj = &policy->kobj;
1140 	cmp = &policy->kobj_unregister;
1141 	up_write(&policy->rwsem);
1142 	kobject_put(kobj);
1143 
1144 	/*
1145 	 * We need to make sure that the underlying kobj is
1146 	 * actually not referenced anymore by anybody before we
1147 	 * proceed with unloading.
1148 	 */
1149 	pr_debug("waiting for dropping of refcount\n");
1150 	wait_for_completion(cmp);
1151 	pr_debug("wait complete\n");
1152 }
1153 
1154 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1155 {
1156 	unsigned long flags;
1157 	int cpu;
1158 
1159 	/* Remove policy from list */
1160 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1161 	list_del(&policy->policy_list);
1162 
1163 	for_each_cpu(cpu, policy->related_cpus)
1164 		per_cpu(cpufreq_cpu_data, cpu) = NULL;
1165 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1166 
1167 	cpufreq_policy_put_kobj(policy);
1168 	free_cpumask_var(policy->real_cpus);
1169 	free_cpumask_var(policy->related_cpus);
1170 	free_cpumask_var(policy->cpus);
1171 	kfree(policy);
1172 }
1173 
1174 static int cpufreq_online(unsigned int cpu)
1175 {
1176 	struct cpufreq_policy *policy;
1177 	bool new_policy;
1178 	unsigned long flags;
1179 	unsigned int j;
1180 	int ret;
1181 
1182 	pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1183 
1184 	/* Check if this CPU already has a policy to manage it */
1185 	policy = per_cpu(cpufreq_cpu_data, cpu);
1186 	if (policy) {
1187 		WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1188 		if (!policy_is_inactive(policy))
1189 			return cpufreq_add_policy_cpu(policy, cpu);
1190 
1191 		/* This is the only online CPU for the policy.  Start over. */
1192 		new_policy = false;
1193 		down_write(&policy->rwsem);
1194 		policy->cpu = cpu;
1195 		policy->governor = NULL;
1196 		up_write(&policy->rwsem);
1197 	} else {
1198 		new_policy = true;
1199 		policy = cpufreq_policy_alloc(cpu);
1200 		if (!policy)
1201 			return -ENOMEM;
1202 	}
1203 
1204 	cpumask_copy(policy->cpus, cpumask_of(cpu));
1205 
1206 	/* call driver. From then on the cpufreq must be able
1207 	 * to accept all calls to ->verify and ->setpolicy for this CPU
1208 	 */
1209 	ret = cpufreq_driver->init(policy);
1210 	if (ret) {
1211 		pr_debug("initialization failed\n");
1212 		goto out_free_policy;
1213 	}
1214 
1215 	ret = cpufreq_table_validate_and_sort(policy);
1216 	if (ret)
1217 		goto out_exit_policy;
1218 
1219 	down_write(&policy->rwsem);
1220 
1221 	if (new_policy) {
1222 		/* related_cpus should at least include policy->cpus. */
1223 		cpumask_copy(policy->related_cpus, policy->cpus);
1224 	}
1225 
1226 	/*
1227 	 * affected cpus must always be the one, which are online. We aren't
1228 	 * managing offline cpus here.
1229 	 */
1230 	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1231 
1232 	if (new_policy) {
1233 		policy->user_policy.min = policy->min;
1234 		policy->user_policy.max = policy->max;
1235 
1236 		for_each_cpu(j, policy->related_cpus) {
1237 			per_cpu(cpufreq_cpu_data, j) = policy;
1238 			add_cpu_dev_symlink(policy, j);
1239 		}
1240 	} else {
1241 		policy->min = policy->user_policy.min;
1242 		policy->max = policy->user_policy.max;
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_destroy_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 	if (new_policy) {
1294 		ret = cpufreq_add_dev_interface(policy);
1295 		if (ret)
1296 			goto out_destroy_policy;
1297 
1298 		cpufreq_stats_create_table(policy);
1299 
1300 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1301 		list_add(&policy->policy_list, &cpufreq_policy_list);
1302 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1303 	}
1304 
1305 	ret = cpufreq_init_policy(policy);
1306 	if (ret) {
1307 		pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1308 		       __func__, cpu, ret);
1309 		/* cpufreq_policy_free() will notify based on this */
1310 		new_policy = false;
1311 		goto out_destroy_policy;
1312 	}
1313 
1314 	up_write(&policy->rwsem);
1315 
1316 	kobject_uevent(&policy->kobj, KOBJ_ADD);
1317 
1318 	/* Callback for handling stuff after policy is ready */
1319 	if (cpufreq_driver->ready)
1320 		cpufreq_driver->ready(policy);
1321 
1322 	pr_debug("initialization complete\n");
1323 
1324 	return 0;
1325 
1326 out_destroy_policy:
1327 	for_each_cpu(j, policy->real_cpus)
1328 		remove_cpu_dev_symlink(policy, get_cpu_device(j));
1329 
1330 	up_write(&policy->rwsem);
1331 
1332 out_exit_policy:
1333 	if (cpufreq_driver->exit)
1334 		cpufreq_driver->exit(policy);
1335 
1336 out_free_policy:
1337 	cpufreq_policy_free(policy);
1338 	return ret;
1339 }
1340 
1341 /**
1342  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1343  * @dev: CPU device.
1344  * @sif: Subsystem interface structure pointer (not used)
1345  */
1346 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1347 {
1348 	struct cpufreq_policy *policy;
1349 	unsigned cpu = dev->id;
1350 	int ret;
1351 
1352 	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1353 
1354 	if (cpu_online(cpu)) {
1355 		ret = cpufreq_online(cpu);
1356 		if (ret)
1357 			return ret;
1358 	}
1359 
1360 	/* Create sysfs link on CPU registration */
1361 	policy = per_cpu(cpufreq_cpu_data, cpu);
1362 	if (policy)
1363 		add_cpu_dev_symlink(policy, cpu);
1364 
1365 	return 0;
1366 }
1367 
1368 static int cpufreq_offline(unsigned int cpu)
1369 {
1370 	struct cpufreq_policy *policy;
1371 	int ret;
1372 
1373 	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1374 
1375 	policy = cpufreq_cpu_get_raw(cpu);
1376 	if (!policy) {
1377 		pr_debug("%s: No cpu_data found\n", __func__);
1378 		return 0;
1379 	}
1380 
1381 	down_write(&policy->rwsem);
1382 	if (has_target())
1383 		cpufreq_stop_governor(policy);
1384 
1385 	cpumask_clear_cpu(cpu, policy->cpus);
1386 
1387 	if (policy_is_inactive(policy)) {
1388 		if (has_target())
1389 			strncpy(policy->last_governor, policy->governor->name,
1390 				CPUFREQ_NAME_LEN);
1391 		else
1392 			policy->last_policy = policy->policy;
1393 	} else if (cpu == policy->cpu) {
1394 		/* Nominate new CPU */
1395 		policy->cpu = cpumask_any(policy->cpus);
1396 	}
1397 
1398 	/* Start governor again for active policy */
1399 	if (!policy_is_inactive(policy)) {
1400 		if (has_target()) {
1401 			ret = cpufreq_start_governor(policy);
1402 			if (ret)
1403 				pr_err("%s: Failed to start governor\n", __func__);
1404 		}
1405 
1406 		goto unlock;
1407 	}
1408 
1409 	if (cpufreq_driver->stop_cpu)
1410 		cpufreq_driver->stop_cpu(policy);
1411 
1412 	if (has_target())
1413 		cpufreq_exit_governor(policy);
1414 
1415 	/*
1416 	 * Perform the ->exit() even during light-weight tear-down,
1417 	 * since this is a core component, and is essential for the
1418 	 * subsequent light-weight ->init() to succeed.
1419 	 */
1420 	if (cpufreq_driver->exit) {
1421 		cpufreq_driver->exit(policy);
1422 		policy->freq_table = NULL;
1423 	}
1424 
1425 unlock:
1426 	up_write(&policy->rwsem);
1427 	return 0;
1428 }
1429 
1430 /**
1431  * cpufreq_remove_dev - remove a CPU device
1432  *
1433  * Removes the cpufreq interface for a CPU device.
1434  */
1435 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1436 {
1437 	unsigned int cpu = dev->id;
1438 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1439 
1440 	if (!policy)
1441 		return;
1442 
1443 	if (cpu_online(cpu))
1444 		cpufreq_offline(cpu);
1445 
1446 	cpumask_clear_cpu(cpu, policy->real_cpus);
1447 	remove_cpu_dev_symlink(policy, dev);
1448 
1449 	if (cpumask_empty(policy->real_cpus))
1450 		cpufreq_policy_free(policy);
1451 }
1452 
1453 /**
1454  *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1455  *	in deep trouble.
1456  *	@policy: policy managing CPUs
1457  *	@new_freq: CPU frequency the CPU actually runs at
1458  *
1459  *	We adjust to current frequency first, and need to clean up later.
1460  *	So either call to cpufreq_update_policy() or schedule handle_update()).
1461  */
1462 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1463 				unsigned int new_freq)
1464 {
1465 	struct cpufreq_freqs freqs;
1466 
1467 	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1468 		 policy->cur, new_freq);
1469 
1470 	freqs.old = policy->cur;
1471 	freqs.new = new_freq;
1472 
1473 	cpufreq_freq_transition_begin(policy, &freqs);
1474 	cpufreq_freq_transition_end(policy, &freqs, 0);
1475 }
1476 
1477 /**
1478  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1479  * @cpu: CPU number
1480  *
1481  * This is the last known freq, without actually getting it from the driver.
1482  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1483  */
1484 unsigned int cpufreq_quick_get(unsigned int cpu)
1485 {
1486 	struct cpufreq_policy *policy;
1487 	unsigned int ret_freq = 0;
1488 	unsigned long flags;
1489 
1490 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1491 
1492 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1493 		ret_freq = cpufreq_driver->get(cpu);
1494 		read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1495 		return ret_freq;
1496 	}
1497 
1498 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1499 
1500 	policy = cpufreq_cpu_get(cpu);
1501 	if (policy) {
1502 		ret_freq = policy->cur;
1503 		cpufreq_cpu_put(policy);
1504 	}
1505 
1506 	return ret_freq;
1507 }
1508 EXPORT_SYMBOL(cpufreq_quick_get);
1509 
1510 /**
1511  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1512  * @cpu: CPU number
1513  *
1514  * Just return the max possible frequency for a given CPU.
1515  */
1516 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1517 {
1518 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1519 	unsigned int ret_freq = 0;
1520 
1521 	if (policy) {
1522 		ret_freq = policy->max;
1523 		cpufreq_cpu_put(policy);
1524 	}
1525 
1526 	return ret_freq;
1527 }
1528 EXPORT_SYMBOL(cpufreq_quick_get_max);
1529 
1530 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1531 {
1532 	unsigned int ret_freq = 0;
1533 
1534 	if (!cpufreq_driver->get)
1535 		return ret_freq;
1536 
1537 	ret_freq = cpufreq_driver->get(policy->cpu);
1538 
1539 	/*
1540 	 * Updating inactive policies is invalid, so avoid doing that.  Also
1541 	 * if fast frequency switching is used with the given policy, the check
1542 	 * against policy->cur is pointless, so skip it in that case too.
1543 	 */
1544 	if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1545 		return ret_freq;
1546 
1547 	if (ret_freq && policy->cur &&
1548 		!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1549 		/* verify no discrepancy between actual and
1550 					saved value exists */
1551 		if (unlikely(ret_freq != policy->cur)) {
1552 			cpufreq_out_of_sync(policy, ret_freq);
1553 			schedule_work(&policy->update);
1554 		}
1555 	}
1556 
1557 	return ret_freq;
1558 }
1559 
1560 /**
1561  * cpufreq_get - get the current CPU frequency (in kHz)
1562  * @cpu: CPU number
1563  *
1564  * Get the CPU current (static) CPU frequency
1565  */
1566 unsigned int cpufreq_get(unsigned int cpu)
1567 {
1568 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1569 	unsigned int ret_freq = 0;
1570 
1571 	if (policy) {
1572 		down_read(&policy->rwsem);
1573 
1574 		if (!policy_is_inactive(policy))
1575 			ret_freq = __cpufreq_get(policy);
1576 
1577 		up_read(&policy->rwsem);
1578 
1579 		cpufreq_cpu_put(policy);
1580 	}
1581 
1582 	return ret_freq;
1583 }
1584 EXPORT_SYMBOL(cpufreq_get);
1585 
1586 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1587 {
1588 	unsigned int new_freq;
1589 
1590 	new_freq = cpufreq_driver->get(policy->cpu);
1591 	if (!new_freq)
1592 		return 0;
1593 
1594 	if (!policy->cur) {
1595 		pr_debug("cpufreq: Driver did not initialize current freq\n");
1596 		policy->cur = new_freq;
1597 	} else if (policy->cur != new_freq && has_target()) {
1598 		cpufreq_out_of_sync(policy, new_freq);
1599 	}
1600 
1601 	return new_freq;
1602 }
1603 
1604 static struct subsys_interface cpufreq_interface = {
1605 	.name		= "cpufreq",
1606 	.subsys		= &cpu_subsys,
1607 	.add_dev	= cpufreq_add_dev,
1608 	.remove_dev	= cpufreq_remove_dev,
1609 };
1610 
1611 /*
1612  * In case platform wants some specific frequency to be configured
1613  * during suspend..
1614  */
1615 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1616 {
1617 	int ret;
1618 
1619 	if (!policy->suspend_freq) {
1620 		pr_debug("%s: suspend_freq not defined\n", __func__);
1621 		return 0;
1622 	}
1623 
1624 	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1625 			policy->suspend_freq);
1626 
1627 	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1628 			CPUFREQ_RELATION_H);
1629 	if (ret)
1630 		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1631 				__func__, policy->suspend_freq, ret);
1632 
1633 	return ret;
1634 }
1635 EXPORT_SYMBOL(cpufreq_generic_suspend);
1636 
1637 /**
1638  * cpufreq_suspend() - Suspend CPUFreq governors
1639  *
1640  * Called during system wide Suspend/Hibernate cycles for suspending governors
1641  * as some platforms can't change frequency after this point in suspend cycle.
1642  * Because some of the devices (like: i2c, regulators, etc) they use for
1643  * changing frequency are suspended quickly after this point.
1644  */
1645 void cpufreq_suspend(void)
1646 {
1647 	struct cpufreq_policy *policy;
1648 
1649 	if (!cpufreq_driver)
1650 		return;
1651 
1652 	if (!has_target() && !cpufreq_driver->suspend)
1653 		goto suspend;
1654 
1655 	pr_debug("%s: Suspending Governors\n", __func__);
1656 
1657 	for_each_active_policy(policy) {
1658 		if (has_target()) {
1659 			down_write(&policy->rwsem);
1660 			cpufreq_stop_governor(policy);
1661 			up_write(&policy->rwsem);
1662 		}
1663 
1664 		if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1665 			pr_err("%s: Failed to suspend driver: %p\n", __func__,
1666 				policy);
1667 	}
1668 
1669 suspend:
1670 	cpufreq_suspended = true;
1671 }
1672 
1673 /**
1674  * cpufreq_resume() - Resume CPUFreq governors
1675  *
1676  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1677  * are suspended with cpufreq_suspend().
1678  */
1679 void cpufreq_resume(void)
1680 {
1681 	struct cpufreq_policy *policy;
1682 	int ret;
1683 
1684 	if (!cpufreq_driver)
1685 		return;
1686 
1687 	if (unlikely(!cpufreq_suspended))
1688 		return;
1689 
1690 	cpufreq_suspended = false;
1691 
1692 	if (!has_target() && !cpufreq_driver->resume)
1693 		return;
1694 
1695 	pr_debug("%s: Resuming Governors\n", __func__);
1696 
1697 	for_each_active_policy(policy) {
1698 		if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1699 			pr_err("%s: Failed to resume driver: %p\n", __func__,
1700 				policy);
1701 		} else if (has_target()) {
1702 			down_write(&policy->rwsem);
1703 			ret = cpufreq_start_governor(policy);
1704 			up_write(&policy->rwsem);
1705 
1706 			if (ret)
1707 				pr_err("%s: Failed to start governor for policy: %p\n",
1708 				       __func__, policy);
1709 		}
1710 	}
1711 }
1712 
1713 /**
1714  *	cpufreq_get_current_driver - return current driver's name
1715  *
1716  *	Return the name string of the currently loaded cpufreq driver
1717  *	or NULL, if none.
1718  */
1719 const char *cpufreq_get_current_driver(void)
1720 {
1721 	if (cpufreq_driver)
1722 		return cpufreq_driver->name;
1723 
1724 	return NULL;
1725 }
1726 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1727 
1728 /**
1729  *	cpufreq_get_driver_data - return current driver data
1730  *
1731  *	Return the private data of the currently loaded cpufreq
1732  *	driver, or NULL if no cpufreq driver is loaded.
1733  */
1734 void *cpufreq_get_driver_data(void)
1735 {
1736 	if (cpufreq_driver)
1737 		return cpufreq_driver->driver_data;
1738 
1739 	return NULL;
1740 }
1741 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1742 
1743 /*********************************************************************
1744  *                     NOTIFIER LISTS INTERFACE                      *
1745  *********************************************************************/
1746 
1747 /**
1748  *	cpufreq_register_notifier - register a driver with cpufreq
1749  *	@nb: notifier function to register
1750  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1751  *
1752  *	Add a driver to one of two lists: either a list of drivers that
1753  *      are notified about clock rate changes (once before and once after
1754  *      the transition), or a list of drivers that are notified about
1755  *      changes in cpufreq policy.
1756  *
1757  *	This function may sleep, and has the same return conditions as
1758  *	blocking_notifier_chain_register.
1759  */
1760 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1761 {
1762 	int ret;
1763 
1764 	if (cpufreq_disabled())
1765 		return -EINVAL;
1766 
1767 	WARN_ON(!init_cpufreq_transition_notifier_list_called);
1768 
1769 	switch (list) {
1770 	case CPUFREQ_TRANSITION_NOTIFIER:
1771 		mutex_lock(&cpufreq_fast_switch_lock);
1772 
1773 		if (cpufreq_fast_switch_count > 0) {
1774 			mutex_unlock(&cpufreq_fast_switch_lock);
1775 			return -EBUSY;
1776 		}
1777 		ret = srcu_notifier_chain_register(
1778 				&cpufreq_transition_notifier_list, nb);
1779 		if (!ret)
1780 			cpufreq_fast_switch_count--;
1781 
1782 		mutex_unlock(&cpufreq_fast_switch_lock);
1783 		break;
1784 	case CPUFREQ_POLICY_NOTIFIER:
1785 		ret = blocking_notifier_chain_register(
1786 				&cpufreq_policy_notifier_list, nb);
1787 		break;
1788 	default:
1789 		ret = -EINVAL;
1790 	}
1791 
1792 	return ret;
1793 }
1794 EXPORT_SYMBOL(cpufreq_register_notifier);
1795 
1796 /**
1797  *	cpufreq_unregister_notifier - unregister a driver with cpufreq
1798  *	@nb: notifier block to be unregistered
1799  *	@list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1800  *
1801  *	Remove a driver from the CPU frequency notifier list.
1802  *
1803  *	This function may sleep, and has the same return conditions as
1804  *	blocking_notifier_chain_unregister.
1805  */
1806 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1807 {
1808 	int ret;
1809 
1810 	if (cpufreq_disabled())
1811 		return -EINVAL;
1812 
1813 	switch (list) {
1814 	case CPUFREQ_TRANSITION_NOTIFIER:
1815 		mutex_lock(&cpufreq_fast_switch_lock);
1816 
1817 		ret = srcu_notifier_chain_unregister(
1818 				&cpufreq_transition_notifier_list, nb);
1819 		if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1820 			cpufreq_fast_switch_count++;
1821 
1822 		mutex_unlock(&cpufreq_fast_switch_lock);
1823 		break;
1824 	case CPUFREQ_POLICY_NOTIFIER:
1825 		ret = blocking_notifier_chain_unregister(
1826 				&cpufreq_policy_notifier_list, nb);
1827 		break;
1828 	default:
1829 		ret = -EINVAL;
1830 	}
1831 
1832 	return ret;
1833 }
1834 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1835 
1836 
1837 /*********************************************************************
1838  *                              GOVERNORS                            *
1839  *********************************************************************/
1840 
1841 /**
1842  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1843  * @policy: cpufreq policy to switch the frequency for.
1844  * @target_freq: New frequency to set (may be approximate).
1845  *
1846  * Carry out a fast frequency switch without sleeping.
1847  *
1848  * The driver's ->fast_switch() callback invoked by this function must be
1849  * suitable for being called from within RCU-sched read-side critical sections
1850  * and it is expected to select the minimum available frequency greater than or
1851  * equal to @target_freq (CPUFREQ_RELATION_L).
1852  *
1853  * This function must not be called if policy->fast_switch_enabled is unset.
1854  *
1855  * Governors calling this function must guarantee that it will never be invoked
1856  * twice in parallel for the same policy and that it will never be called in
1857  * parallel with either ->target() or ->target_index() for the same policy.
1858  *
1859  * Returns the actual frequency set for the CPU.
1860  *
1861  * If 0 is returned by the driver's ->fast_switch() callback to indicate an
1862  * error condition, the hardware configuration must be preserved.
1863  */
1864 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1865 					unsigned int target_freq)
1866 {
1867 	target_freq = clamp_val(target_freq, policy->min, policy->max);
1868 
1869 	return cpufreq_driver->fast_switch(policy, target_freq);
1870 }
1871 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1872 
1873 /* Must set freqs->new to intermediate frequency */
1874 static int __target_intermediate(struct cpufreq_policy *policy,
1875 				 struct cpufreq_freqs *freqs, int index)
1876 {
1877 	int ret;
1878 
1879 	freqs->new = cpufreq_driver->get_intermediate(policy, index);
1880 
1881 	/* We don't need to switch to intermediate freq */
1882 	if (!freqs->new)
1883 		return 0;
1884 
1885 	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1886 		 __func__, policy->cpu, freqs->old, freqs->new);
1887 
1888 	cpufreq_freq_transition_begin(policy, freqs);
1889 	ret = cpufreq_driver->target_intermediate(policy, index);
1890 	cpufreq_freq_transition_end(policy, freqs, ret);
1891 
1892 	if (ret)
1893 		pr_err("%s: Failed to change to intermediate frequency: %d\n",
1894 		       __func__, ret);
1895 
1896 	return ret;
1897 }
1898 
1899 static int __target_index(struct cpufreq_policy *policy, int index)
1900 {
1901 	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1902 	unsigned int intermediate_freq = 0;
1903 	unsigned int newfreq = policy->freq_table[index].frequency;
1904 	int retval = -EINVAL;
1905 	bool notify;
1906 
1907 	if (newfreq == policy->cur)
1908 		return 0;
1909 
1910 	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1911 	if (notify) {
1912 		/* Handle switching to intermediate frequency */
1913 		if (cpufreq_driver->get_intermediate) {
1914 			retval = __target_intermediate(policy, &freqs, index);
1915 			if (retval)
1916 				return retval;
1917 
1918 			intermediate_freq = freqs.new;
1919 			/* Set old freq to intermediate */
1920 			if (intermediate_freq)
1921 				freqs.old = freqs.new;
1922 		}
1923 
1924 		freqs.new = newfreq;
1925 		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1926 			 __func__, policy->cpu, freqs.old, freqs.new);
1927 
1928 		cpufreq_freq_transition_begin(policy, &freqs);
1929 	}
1930 
1931 	retval = cpufreq_driver->target_index(policy, index);
1932 	if (retval)
1933 		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1934 		       retval);
1935 
1936 	if (notify) {
1937 		cpufreq_freq_transition_end(policy, &freqs, retval);
1938 
1939 		/*
1940 		 * Failed after setting to intermediate freq? Driver should have
1941 		 * reverted back to initial frequency and so should we. Check
1942 		 * here for intermediate_freq instead of get_intermediate, in
1943 		 * case we haven't switched to intermediate freq at all.
1944 		 */
1945 		if (unlikely(retval && intermediate_freq)) {
1946 			freqs.old = intermediate_freq;
1947 			freqs.new = policy->restore_freq;
1948 			cpufreq_freq_transition_begin(policy, &freqs);
1949 			cpufreq_freq_transition_end(policy, &freqs, 0);
1950 		}
1951 	}
1952 
1953 	return retval;
1954 }
1955 
1956 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1957 			    unsigned int target_freq,
1958 			    unsigned int relation)
1959 {
1960 	unsigned int old_target_freq = target_freq;
1961 	int index;
1962 
1963 	if (cpufreq_disabled())
1964 		return -ENODEV;
1965 
1966 	/* Make sure that target_freq is within supported range */
1967 	target_freq = clamp_val(target_freq, policy->min, policy->max);
1968 
1969 	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1970 		 policy->cpu, target_freq, relation, old_target_freq);
1971 
1972 	/*
1973 	 * This might look like a redundant call as we are checking it again
1974 	 * after finding index. But it is left intentionally for cases where
1975 	 * exactly same freq is called again and so we can save on few function
1976 	 * calls.
1977 	 */
1978 	if (target_freq == policy->cur)
1979 		return 0;
1980 
1981 	/* Save last value to restore later on errors */
1982 	policy->restore_freq = policy->cur;
1983 
1984 	if (cpufreq_driver->target)
1985 		return cpufreq_driver->target(policy, target_freq, relation);
1986 
1987 	if (!cpufreq_driver->target_index)
1988 		return -EINVAL;
1989 
1990 	index = cpufreq_frequency_table_target(policy, target_freq, relation);
1991 
1992 	return __target_index(policy, index);
1993 }
1994 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1995 
1996 int cpufreq_driver_target(struct cpufreq_policy *policy,
1997 			  unsigned int target_freq,
1998 			  unsigned int relation)
1999 {
2000 	int ret = -EINVAL;
2001 
2002 	down_write(&policy->rwsem);
2003 
2004 	ret = __cpufreq_driver_target(policy, target_freq, relation);
2005 
2006 	up_write(&policy->rwsem);
2007 
2008 	return ret;
2009 }
2010 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2011 
2012 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2013 {
2014 	return NULL;
2015 }
2016 
2017 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2018 {
2019 	int ret;
2020 
2021 	/* Don't start any governor operations if we are entering suspend */
2022 	if (cpufreq_suspended)
2023 		return 0;
2024 	/*
2025 	 * Governor might not be initiated here if ACPI _PPC changed
2026 	 * notification happened, so check it.
2027 	 */
2028 	if (!policy->governor)
2029 		return -EINVAL;
2030 
2031 	/* Platform doesn't want dynamic frequency switching ? */
2032 	if (policy->governor->dynamic_switching &&
2033 	    cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2034 		struct cpufreq_governor *gov = cpufreq_fallback_governor();
2035 
2036 		if (gov) {
2037 			pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2038 				policy->governor->name, gov->name);
2039 			policy->governor = gov;
2040 		} else {
2041 			return -EINVAL;
2042 		}
2043 	}
2044 
2045 	if (!try_module_get(policy->governor->owner))
2046 		return -EINVAL;
2047 
2048 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2049 
2050 	if (policy->governor->init) {
2051 		ret = policy->governor->init(policy);
2052 		if (ret) {
2053 			module_put(policy->governor->owner);
2054 			return ret;
2055 		}
2056 	}
2057 
2058 	return 0;
2059 }
2060 
2061 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2062 {
2063 	if (cpufreq_suspended || !policy->governor)
2064 		return;
2065 
2066 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2067 
2068 	if (policy->governor->exit)
2069 		policy->governor->exit(policy);
2070 
2071 	module_put(policy->governor->owner);
2072 }
2073 
2074 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2075 {
2076 	int ret;
2077 
2078 	if (cpufreq_suspended)
2079 		return 0;
2080 
2081 	if (!policy->governor)
2082 		return -EINVAL;
2083 
2084 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2085 
2086 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2087 		cpufreq_update_current_freq(policy);
2088 
2089 	if (policy->governor->start) {
2090 		ret = policy->governor->start(policy);
2091 		if (ret)
2092 			return ret;
2093 	}
2094 
2095 	if (policy->governor->limits)
2096 		policy->governor->limits(policy);
2097 
2098 	return 0;
2099 }
2100 
2101 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2102 {
2103 	if (cpufreq_suspended || !policy->governor)
2104 		return;
2105 
2106 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2107 
2108 	if (policy->governor->stop)
2109 		policy->governor->stop(policy);
2110 }
2111 
2112 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2113 {
2114 	if (cpufreq_suspended || !policy->governor)
2115 		return;
2116 
2117 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2118 
2119 	if (policy->governor->limits)
2120 		policy->governor->limits(policy);
2121 }
2122 
2123 int cpufreq_register_governor(struct cpufreq_governor *governor)
2124 {
2125 	int err;
2126 
2127 	if (!governor)
2128 		return -EINVAL;
2129 
2130 	if (cpufreq_disabled())
2131 		return -ENODEV;
2132 
2133 	mutex_lock(&cpufreq_governor_mutex);
2134 
2135 	err = -EBUSY;
2136 	if (!find_governor(governor->name)) {
2137 		err = 0;
2138 		list_add(&governor->governor_list, &cpufreq_governor_list);
2139 	}
2140 
2141 	mutex_unlock(&cpufreq_governor_mutex);
2142 	return err;
2143 }
2144 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2145 
2146 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2147 {
2148 	struct cpufreq_policy *policy;
2149 	unsigned long flags;
2150 
2151 	if (!governor)
2152 		return;
2153 
2154 	if (cpufreq_disabled())
2155 		return;
2156 
2157 	/* clear last_governor for all inactive policies */
2158 	read_lock_irqsave(&cpufreq_driver_lock, flags);
2159 	for_each_inactive_policy(policy) {
2160 		if (!strcmp(policy->last_governor, governor->name)) {
2161 			policy->governor = NULL;
2162 			strcpy(policy->last_governor, "\0");
2163 		}
2164 	}
2165 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2166 
2167 	mutex_lock(&cpufreq_governor_mutex);
2168 	list_del(&governor->governor_list);
2169 	mutex_unlock(&cpufreq_governor_mutex);
2170 }
2171 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2172 
2173 
2174 /*********************************************************************
2175  *                          POLICY INTERFACE                         *
2176  *********************************************************************/
2177 
2178 /**
2179  * cpufreq_get_policy - get the current cpufreq_policy
2180  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2181  *	is written
2182  *
2183  * Reads the current cpufreq policy.
2184  */
2185 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2186 {
2187 	struct cpufreq_policy *cpu_policy;
2188 	if (!policy)
2189 		return -EINVAL;
2190 
2191 	cpu_policy = cpufreq_cpu_get(cpu);
2192 	if (!cpu_policy)
2193 		return -EINVAL;
2194 
2195 	memcpy(policy, cpu_policy, sizeof(*policy));
2196 
2197 	cpufreq_cpu_put(cpu_policy);
2198 	return 0;
2199 }
2200 EXPORT_SYMBOL(cpufreq_get_policy);
2201 
2202 /*
2203  * policy : current policy.
2204  * new_policy: policy to be set.
2205  */
2206 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2207 				struct cpufreq_policy *new_policy)
2208 {
2209 	struct cpufreq_governor *old_gov;
2210 	int ret;
2211 
2212 	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2213 		 new_policy->cpu, new_policy->min, new_policy->max);
2214 
2215 	memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2216 
2217 	/*
2218 	* This check works well when we store new min/max freq attributes,
2219 	* because new_policy is a copy of policy with one field updated.
2220 	*/
2221 	if (new_policy->min > new_policy->max)
2222 		return -EINVAL;
2223 
2224 	/* verify the cpu speed can be set within this limit */
2225 	ret = cpufreq_driver->verify(new_policy);
2226 	if (ret)
2227 		return ret;
2228 
2229 	/* adjust if necessary - all reasons */
2230 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2231 			CPUFREQ_ADJUST, new_policy);
2232 
2233 	/*
2234 	 * verify the cpu speed can be set within this limit, which might be
2235 	 * different to the first one
2236 	 */
2237 	ret = cpufreq_driver->verify(new_policy);
2238 	if (ret)
2239 		return ret;
2240 
2241 	/* notification of the new policy */
2242 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2243 			CPUFREQ_NOTIFY, new_policy);
2244 
2245 	policy->min = new_policy->min;
2246 	policy->max = new_policy->max;
2247 
2248 	policy->cached_target_freq = UINT_MAX;
2249 
2250 	pr_debug("new min and max freqs are %u - %u kHz\n",
2251 		 policy->min, policy->max);
2252 
2253 	if (cpufreq_driver->setpolicy) {
2254 		policy->policy = new_policy->policy;
2255 		pr_debug("setting range\n");
2256 		return cpufreq_driver->setpolicy(new_policy);
2257 	}
2258 
2259 	if (new_policy->governor == policy->governor) {
2260 		pr_debug("cpufreq: governor limits update\n");
2261 		cpufreq_governor_limits(policy);
2262 		return 0;
2263 	}
2264 
2265 	pr_debug("governor switch\n");
2266 
2267 	/* save old, working values */
2268 	old_gov = policy->governor;
2269 	/* end old governor */
2270 	if (old_gov) {
2271 		cpufreq_stop_governor(policy);
2272 		cpufreq_exit_governor(policy);
2273 	}
2274 
2275 	/* start new governor */
2276 	policy->governor = new_policy->governor;
2277 	ret = cpufreq_init_governor(policy);
2278 	if (!ret) {
2279 		ret = cpufreq_start_governor(policy);
2280 		if (!ret) {
2281 			pr_debug("cpufreq: governor change\n");
2282 			return 0;
2283 		}
2284 		cpufreq_exit_governor(policy);
2285 	}
2286 
2287 	/* new governor failed, so re-start old one */
2288 	pr_debug("starting governor %s failed\n", policy->governor->name);
2289 	if (old_gov) {
2290 		policy->governor = old_gov;
2291 		if (cpufreq_init_governor(policy))
2292 			policy->governor = NULL;
2293 		else
2294 			cpufreq_start_governor(policy);
2295 	}
2296 
2297 	return ret;
2298 }
2299 
2300 /**
2301  *	cpufreq_update_policy - re-evaluate an existing cpufreq policy
2302  *	@cpu: CPU which shall be re-evaluated
2303  *
2304  *	Useful for policy notifiers which have different necessities
2305  *	at different times.
2306  */
2307 void cpufreq_update_policy(unsigned int cpu)
2308 {
2309 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2310 	struct cpufreq_policy new_policy;
2311 
2312 	if (!policy)
2313 		return;
2314 
2315 	down_write(&policy->rwsem);
2316 
2317 	if (policy_is_inactive(policy))
2318 		goto unlock;
2319 
2320 	pr_debug("updating policy for CPU %u\n", cpu);
2321 	memcpy(&new_policy, policy, sizeof(*policy));
2322 	new_policy.min = policy->user_policy.min;
2323 	new_policy.max = policy->user_policy.max;
2324 
2325 	/*
2326 	 * BIOS might change freq behind our back
2327 	 * -> ask driver for current freq and notify governors about a change
2328 	 */
2329 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2330 		if (cpufreq_suspended)
2331 			goto unlock;
2332 
2333 		new_policy.cur = cpufreq_update_current_freq(policy);
2334 		if (WARN_ON(!new_policy.cur))
2335 			goto unlock;
2336 	}
2337 
2338 	cpufreq_set_policy(policy, &new_policy);
2339 
2340 unlock:
2341 	up_write(&policy->rwsem);
2342 
2343 	cpufreq_cpu_put(policy);
2344 }
2345 EXPORT_SYMBOL(cpufreq_update_policy);
2346 
2347 /*********************************************************************
2348  *               BOOST						     *
2349  *********************************************************************/
2350 static int cpufreq_boost_set_sw(int state)
2351 {
2352 	struct cpufreq_policy *policy;
2353 	int ret = -EINVAL;
2354 
2355 	for_each_active_policy(policy) {
2356 		if (!policy->freq_table)
2357 			continue;
2358 
2359 		ret = cpufreq_frequency_table_cpuinfo(policy,
2360 						      policy->freq_table);
2361 		if (ret) {
2362 			pr_err("%s: Policy frequency update failed\n",
2363 			       __func__);
2364 			break;
2365 		}
2366 
2367 		down_write(&policy->rwsem);
2368 		policy->user_policy.max = policy->max;
2369 		cpufreq_governor_limits(policy);
2370 		up_write(&policy->rwsem);
2371 	}
2372 
2373 	return ret;
2374 }
2375 
2376 int cpufreq_boost_trigger_state(int state)
2377 {
2378 	unsigned long flags;
2379 	int ret = 0;
2380 
2381 	if (cpufreq_driver->boost_enabled == state)
2382 		return 0;
2383 
2384 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2385 	cpufreq_driver->boost_enabled = state;
2386 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2387 
2388 	ret = cpufreq_driver->set_boost(state);
2389 	if (ret) {
2390 		write_lock_irqsave(&cpufreq_driver_lock, flags);
2391 		cpufreq_driver->boost_enabled = !state;
2392 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2393 
2394 		pr_err("%s: Cannot %s BOOST\n",
2395 		       __func__, state ? "enable" : "disable");
2396 	}
2397 
2398 	return ret;
2399 }
2400 
2401 static bool cpufreq_boost_supported(void)
2402 {
2403 	return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2404 }
2405 
2406 static int create_boost_sysfs_file(void)
2407 {
2408 	int ret;
2409 
2410 	ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2411 	if (ret)
2412 		pr_err("%s: cannot register global BOOST sysfs file\n",
2413 		       __func__);
2414 
2415 	return ret;
2416 }
2417 
2418 static void remove_boost_sysfs_file(void)
2419 {
2420 	if (cpufreq_boost_supported())
2421 		sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2422 }
2423 
2424 int cpufreq_enable_boost_support(void)
2425 {
2426 	if (!cpufreq_driver)
2427 		return -EINVAL;
2428 
2429 	if (cpufreq_boost_supported())
2430 		return 0;
2431 
2432 	cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2433 
2434 	/* This will get removed on driver unregister */
2435 	return create_boost_sysfs_file();
2436 }
2437 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2438 
2439 int cpufreq_boost_enabled(void)
2440 {
2441 	return cpufreq_driver->boost_enabled;
2442 }
2443 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2444 
2445 /*********************************************************************
2446  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2447  *********************************************************************/
2448 static enum cpuhp_state hp_online;
2449 
2450 static int cpuhp_cpufreq_online(unsigned int cpu)
2451 {
2452 	cpufreq_online(cpu);
2453 
2454 	return 0;
2455 }
2456 
2457 static int cpuhp_cpufreq_offline(unsigned int cpu)
2458 {
2459 	cpufreq_offline(cpu);
2460 
2461 	return 0;
2462 }
2463 
2464 /**
2465  * cpufreq_register_driver - register a CPU Frequency driver
2466  * @driver_data: A struct cpufreq_driver containing the values#
2467  * submitted by the CPU Frequency driver.
2468  *
2469  * Registers a CPU Frequency driver to this core code. This code
2470  * returns zero on success, -EEXIST when another driver got here first
2471  * (and isn't unregistered in the meantime).
2472  *
2473  */
2474 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2475 {
2476 	unsigned long flags;
2477 	int ret;
2478 
2479 	if (cpufreq_disabled())
2480 		return -ENODEV;
2481 
2482 	if (!driver_data || !driver_data->verify || !driver_data->init ||
2483 	    !(driver_data->setpolicy || driver_data->target_index ||
2484 		    driver_data->target) ||
2485 	     (driver_data->setpolicy && (driver_data->target_index ||
2486 		    driver_data->target)) ||
2487 	     (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2488 		return -EINVAL;
2489 
2490 	pr_debug("trying to register driver %s\n", driver_data->name);
2491 
2492 	/* Protect against concurrent CPU online/offline. */
2493 	cpus_read_lock();
2494 
2495 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2496 	if (cpufreq_driver) {
2497 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2498 		ret = -EEXIST;
2499 		goto out;
2500 	}
2501 	cpufreq_driver = driver_data;
2502 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2503 
2504 	if (driver_data->setpolicy)
2505 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2506 
2507 	if (cpufreq_boost_supported()) {
2508 		ret = create_boost_sysfs_file();
2509 		if (ret)
2510 			goto err_null_driver;
2511 	}
2512 
2513 	ret = subsys_interface_register(&cpufreq_interface);
2514 	if (ret)
2515 		goto err_boost_unreg;
2516 
2517 	if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2518 	    list_empty(&cpufreq_policy_list)) {
2519 		/* if all ->init() calls failed, unregister */
2520 		ret = -ENODEV;
2521 		pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2522 			 driver_data->name);
2523 		goto err_if_unreg;
2524 	}
2525 
2526 	ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2527 						   "cpufreq:online",
2528 						   cpuhp_cpufreq_online,
2529 						   cpuhp_cpufreq_offline);
2530 	if (ret < 0)
2531 		goto err_if_unreg;
2532 	hp_online = ret;
2533 	ret = 0;
2534 
2535 	pr_debug("driver %s up and running\n", driver_data->name);
2536 	goto out;
2537 
2538 err_if_unreg:
2539 	subsys_interface_unregister(&cpufreq_interface);
2540 err_boost_unreg:
2541 	remove_boost_sysfs_file();
2542 err_null_driver:
2543 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2544 	cpufreq_driver = NULL;
2545 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2546 out:
2547 	cpus_read_unlock();
2548 	return ret;
2549 }
2550 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2551 
2552 /**
2553  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2554  *
2555  * Unregister the current CPUFreq driver. Only call this if you have
2556  * the right to do so, i.e. if you have succeeded in initialising before!
2557  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2558  * currently not initialised.
2559  */
2560 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2561 {
2562 	unsigned long flags;
2563 
2564 	if (!cpufreq_driver || (driver != cpufreq_driver))
2565 		return -EINVAL;
2566 
2567 	pr_debug("unregistering driver %s\n", driver->name);
2568 
2569 	/* Protect against concurrent cpu hotplug */
2570 	cpus_read_lock();
2571 	subsys_interface_unregister(&cpufreq_interface);
2572 	remove_boost_sysfs_file();
2573 	cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2574 
2575 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2576 
2577 	cpufreq_driver = NULL;
2578 
2579 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2580 	cpus_read_unlock();
2581 
2582 	return 0;
2583 }
2584 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2585 
2586 /*
2587  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2588  * or mutexes when secondary CPUs are halted.
2589  */
2590 static struct syscore_ops cpufreq_syscore_ops = {
2591 	.shutdown = cpufreq_suspend,
2592 };
2593 
2594 struct kobject *cpufreq_global_kobject;
2595 EXPORT_SYMBOL(cpufreq_global_kobject);
2596 
2597 static int __init cpufreq_core_init(void)
2598 {
2599 	if (cpufreq_disabled())
2600 		return -ENODEV;
2601 
2602 	cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2603 	BUG_ON(!cpufreq_global_kobject);
2604 
2605 	register_syscore_ops(&cpufreq_syscore_ops);
2606 
2607 	return 0;
2608 }
2609 module_param(off, int, 0444);
2610 core_initcall(cpufreq_core_init);
2611