xref: /openbmc/linux/drivers/cpufreq/cpufreq.c (revision 8cb5d748)
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 /*
165  * This is a generic cpufreq init() routine which can be used by cpufreq
166  * drivers of SMP systems. It will do following:
167  * - validate & show freq table passed
168  * - set policies transition latency
169  * - policy->cpus with all possible CPUs
170  */
171 int cpufreq_generic_init(struct cpufreq_policy *policy,
172 		struct cpufreq_frequency_table *table,
173 		unsigned int transition_latency)
174 {
175 	int ret;
176 
177 	ret = cpufreq_table_validate_and_show(policy, table);
178 	if (ret) {
179 		pr_err("%s: invalid frequency table: %d\n", __func__, ret);
180 		return ret;
181 	}
182 
183 	policy->cpuinfo.transition_latency = transition_latency;
184 
185 	/*
186 	 * The driver only supports the SMP configuration where all processors
187 	 * share the clock and voltage and clock.
188 	 */
189 	cpumask_setall(policy->cpus);
190 
191 	return 0;
192 }
193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
194 
195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
196 {
197 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
198 
199 	return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
200 }
201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
202 
203 unsigned int cpufreq_generic_get(unsigned int cpu)
204 {
205 	struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
206 
207 	if (!policy || IS_ERR(policy->clk)) {
208 		pr_err("%s: No %s associated to cpu: %d\n",
209 		       __func__, policy ? "clk" : "policy", cpu);
210 		return 0;
211 	}
212 
213 	return clk_get_rate(policy->clk) / 1000;
214 }
215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
216 
217 /**
218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
219  *
220  * @cpu: cpu to find policy for.
221  *
222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
223  * It also increments the kobject reference count to mark it busy and so would
224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
226  * freed as that depends on the kobj count.
227  *
228  * Return: A valid policy on success, otherwise NULL on failure.
229  */
230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
231 {
232 	struct cpufreq_policy *policy = NULL;
233 	unsigned long flags;
234 
235 	if (WARN_ON(cpu >= nr_cpu_ids))
236 		return NULL;
237 
238 	/* get the cpufreq driver */
239 	read_lock_irqsave(&cpufreq_driver_lock, flags);
240 
241 	if (cpufreq_driver) {
242 		/* get the CPU */
243 		policy = cpufreq_cpu_get_raw(cpu);
244 		if (policy)
245 			kobject_get(&policy->kobj);
246 	}
247 
248 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
249 
250 	return policy;
251 }
252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
253 
254 /**
255  * cpufreq_cpu_put: Decrements the usage count of a policy
256  *
257  * @policy: policy earlier returned by cpufreq_cpu_get().
258  *
259  * This decrements the kobject reference count incremented earlier by calling
260  * cpufreq_cpu_get().
261  */
262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
263 {
264 	kobject_put(&policy->kobj);
265 }
266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
267 
268 /*********************************************************************
269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
270  *********************************************************************/
271 
272 /**
273  * adjust_jiffies - adjust the system "loops_per_jiffy"
274  *
275  * This function alters the system "loops_per_jiffy" for the clock
276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
277  * systems as each CPU might be scaled differently. So, use the arch
278  * per-CPU loops_per_jiffy value wherever possible.
279  */
280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
281 {
282 #ifndef CONFIG_SMP
283 	static unsigned long l_p_j_ref;
284 	static unsigned int l_p_j_ref_freq;
285 
286 	if (ci->flags & CPUFREQ_CONST_LOOPS)
287 		return;
288 
289 	if (!l_p_j_ref_freq) {
290 		l_p_j_ref = loops_per_jiffy;
291 		l_p_j_ref_freq = ci->old;
292 		pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
293 			 l_p_j_ref, l_p_j_ref_freq);
294 	}
295 	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
296 		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297 								ci->new);
298 		pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
299 			 loops_per_jiffy, ci->new);
300 	}
301 #endif
302 }
303 
304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
305 		struct cpufreq_freqs *freqs, unsigned int state)
306 {
307 	BUG_ON(irqs_disabled());
308 
309 	if (cpufreq_disabled())
310 		return;
311 
312 	freqs->flags = cpufreq_driver->flags;
313 	pr_debug("notification %u of frequency transition to %u kHz\n",
314 		 state, freqs->new);
315 
316 	switch (state) {
317 
318 	case CPUFREQ_PRECHANGE:
319 		/* detect if the driver reported a value as "old frequency"
320 		 * which is not equal to what the cpufreq core thinks is
321 		 * "old frequency".
322 		 */
323 		if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
324 			if ((policy) && (policy->cpu == freqs->cpu) &&
325 			    (policy->cur) && (policy->cur != freqs->old)) {
326 				pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
327 					 freqs->old, policy->cur);
328 				freqs->old = policy->cur;
329 			}
330 		}
331 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
332 				CPUFREQ_PRECHANGE, freqs);
333 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
334 		break;
335 
336 	case CPUFREQ_POSTCHANGE:
337 		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
338 		pr_debug("FREQ: %lu - CPU: %lu\n",
339 			 (unsigned long)freqs->new, (unsigned long)freqs->cpu);
340 		trace_cpu_frequency(freqs->new, freqs->cpu);
341 		cpufreq_stats_record_transition(policy, freqs->new);
342 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
343 				CPUFREQ_POSTCHANGE, freqs);
344 		if (likely(policy) && likely(policy->cpu == freqs->cpu))
345 			policy->cur = freqs->new;
346 		break;
347 	}
348 }
349 
350 /**
351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
352  * on frequency transition.
353  *
354  * This function calls the transition notifiers and the "adjust_jiffies"
355  * function. It is called twice on all CPU frequency changes that have
356  * external effects.
357  */
358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
359 		struct cpufreq_freqs *freqs, unsigned int state)
360 {
361 	for_each_cpu(freqs->cpu, policy->cpus)
362 		__cpufreq_notify_transition(policy, freqs, state);
363 }
364 
365 /* Do post notifications when there are chances that transition has failed */
366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
367 		struct cpufreq_freqs *freqs, int transition_failed)
368 {
369 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
370 	if (!transition_failed)
371 		return;
372 
373 	swap(freqs->old, freqs->new);
374 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
375 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
376 }
377 
378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
379 		struct cpufreq_freqs *freqs)
380 {
381 
382 	/*
383 	 * Catch double invocations of _begin() which lead to self-deadlock.
384 	 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
385 	 * doesn't invoke _begin() on their behalf, and hence the chances of
386 	 * double invocations are very low. Moreover, there are scenarios
387 	 * where these checks can emit false-positive warnings in these
388 	 * drivers; so we avoid that by skipping them altogether.
389 	 */
390 	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
391 				&& current == policy->transition_task);
392 
393 wait:
394 	wait_event(policy->transition_wait, !policy->transition_ongoing);
395 
396 	spin_lock(&policy->transition_lock);
397 
398 	if (unlikely(policy->transition_ongoing)) {
399 		spin_unlock(&policy->transition_lock);
400 		goto wait;
401 	}
402 
403 	policy->transition_ongoing = true;
404 	policy->transition_task = current;
405 
406 	spin_unlock(&policy->transition_lock);
407 
408 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
409 }
410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
411 
412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
413 		struct cpufreq_freqs *freqs, int transition_failed)
414 {
415 	if (unlikely(WARN_ON(!policy->transition_ongoing)))
416 		return;
417 
418 	cpufreq_notify_post_transition(policy, freqs, transition_failed);
419 
420 	policy->transition_ongoing = false;
421 	policy->transition_task = NULL;
422 
423 	wake_up(&policy->transition_wait);
424 }
425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
426 
427 /*
428  * Fast frequency switching status count.  Positive means "enabled", negative
429  * means "disabled" and 0 means "not decided yet".
430  */
431 static int cpufreq_fast_switch_count;
432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
433 
434 static void cpufreq_list_transition_notifiers(void)
435 {
436 	struct notifier_block *nb;
437 
438 	pr_info("Registered transition notifiers:\n");
439 
440 	mutex_lock(&cpufreq_transition_notifier_list.mutex);
441 
442 	for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
443 		pr_info("%pF\n", nb->notifier_call);
444 
445 	mutex_unlock(&cpufreq_transition_notifier_list.mutex);
446 }
447 
448 /**
449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
450  * @policy: cpufreq policy to enable fast frequency switching for.
451  *
452  * Try to enable fast frequency switching for @policy.
453  *
454  * The attempt will fail if there is at least one transition notifier registered
455  * at this point, as fast frequency switching is quite fundamentally at odds
456  * with transition notifiers.  Thus if successful, it will make registration of
457  * transition notifiers fail going forward.
458  */
459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
460 {
461 	lockdep_assert_held(&policy->rwsem);
462 
463 	if (!policy->fast_switch_possible)
464 		return;
465 
466 	mutex_lock(&cpufreq_fast_switch_lock);
467 	if (cpufreq_fast_switch_count >= 0) {
468 		cpufreq_fast_switch_count++;
469 		policy->fast_switch_enabled = true;
470 	} else {
471 		pr_warn("CPU%u: Fast frequency switching not enabled\n",
472 			policy->cpu);
473 		cpufreq_list_transition_notifiers();
474 	}
475 	mutex_unlock(&cpufreq_fast_switch_lock);
476 }
477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
478 
479 /**
480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
481  * @policy: cpufreq policy to disable fast frequency switching for.
482  */
483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
484 {
485 	mutex_lock(&cpufreq_fast_switch_lock);
486 	if (policy->fast_switch_enabled) {
487 		policy->fast_switch_enabled = false;
488 		if (!WARN_ON(cpufreq_fast_switch_count <= 0))
489 			cpufreq_fast_switch_count--;
490 	}
491 	mutex_unlock(&cpufreq_fast_switch_lock);
492 }
493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
494 
495 /**
496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
497  * one.
498  * @target_freq: target frequency to resolve.
499  *
500  * The target to driver frequency mapping is cached in the policy.
501  *
502  * Return: Lowest driver-supported frequency greater than or equal to the
503  * given target_freq, subject to policy (min/max) and driver limitations.
504  */
505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
506 					 unsigned int target_freq)
507 {
508 	target_freq = clamp_val(target_freq, policy->min, policy->max);
509 	policy->cached_target_freq = target_freq;
510 
511 	if (cpufreq_driver->target_index) {
512 		int idx;
513 
514 		idx = cpufreq_frequency_table_target(policy, target_freq,
515 						     CPUFREQ_RELATION_L);
516 		policy->cached_resolved_idx = idx;
517 		return policy->freq_table[idx].frequency;
518 	}
519 
520 	if (cpufreq_driver->resolve_freq)
521 		return cpufreq_driver->resolve_freq(policy, target_freq);
522 
523 	return target_freq;
524 }
525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
526 
527 unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
528 {
529 	unsigned int latency;
530 
531 	if (policy->transition_delay_us)
532 		return policy->transition_delay_us;
533 
534 	latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
535 	if (latency) {
536 		/*
537 		 * For platforms that can change the frequency very fast (< 10
538 		 * us), the above formula gives a decent transition delay. But
539 		 * for platforms where transition_latency is in milliseconds, it
540 		 * ends up giving unrealistic values.
541 		 *
542 		 * Cap the default transition delay to 10 ms, which seems to be
543 		 * a reasonable amount of time after which we should reevaluate
544 		 * the frequency.
545 		 */
546 		return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
547 	}
548 
549 	return LATENCY_MULTIPLIER;
550 }
551 EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
552 
553 /*********************************************************************
554  *                          SYSFS INTERFACE                          *
555  *********************************************************************/
556 static ssize_t show_boost(struct kobject *kobj,
557 				 struct attribute *attr, char *buf)
558 {
559 	return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
560 }
561 
562 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
563 				  const char *buf, size_t count)
564 {
565 	int ret, enable;
566 
567 	ret = sscanf(buf, "%d", &enable);
568 	if (ret != 1 || enable < 0 || enable > 1)
569 		return -EINVAL;
570 
571 	if (cpufreq_boost_trigger_state(enable)) {
572 		pr_err("%s: Cannot %s BOOST!\n",
573 		       __func__, enable ? "enable" : "disable");
574 		return -EINVAL;
575 	}
576 
577 	pr_debug("%s: cpufreq BOOST %s\n",
578 		 __func__, enable ? "enabled" : "disabled");
579 
580 	return count;
581 }
582 define_one_global_rw(boost);
583 
584 static struct cpufreq_governor *find_governor(const char *str_governor)
585 {
586 	struct cpufreq_governor *t;
587 
588 	for_each_governor(t)
589 		if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
590 			return t;
591 
592 	return NULL;
593 }
594 
595 /**
596  * cpufreq_parse_governor - parse a governor string
597  */
598 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
599 				struct cpufreq_governor **governor)
600 {
601 	int err = -EINVAL;
602 
603 	if (cpufreq_driver->setpolicy) {
604 		if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
605 			*policy = CPUFREQ_POLICY_PERFORMANCE;
606 			err = 0;
607 		} else if (!strncasecmp(str_governor, "powersave",
608 						CPUFREQ_NAME_LEN)) {
609 			*policy = CPUFREQ_POLICY_POWERSAVE;
610 			err = 0;
611 		}
612 	} else {
613 		struct cpufreq_governor *t;
614 
615 		mutex_lock(&cpufreq_governor_mutex);
616 
617 		t = find_governor(str_governor);
618 
619 		if (t == NULL) {
620 			int ret;
621 
622 			mutex_unlock(&cpufreq_governor_mutex);
623 			ret = request_module("cpufreq_%s", str_governor);
624 			mutex_lock(&cpufreq_governor_mutex);
625 
626 			if (ret == 0)
627 				t = find_governor(str_governor);
628 		}
629 
630 		if (t != NULL) {
631 			*governor = t;
632 			err = 0;
633 		}
634 
635 		mutex_unlock(&cpufreq_governor_mutex);
636 	}
637 	return err;
638 }
639 
640 /**
641  * cpufreq_per_cpu_attr_read() / show_##file_name() -
642  * print out cpufreq information
643  *
644  * Write out information from cpufreq_driver->policy[cpu]; object must be
645  * "unsigned int".
646  */
647 
648 #define show_one(file_name, object)			\
649 static ssize_t show_##file_name				\
650 (struct cpufreq_policy *policy, char *buf)		\
651 {							\
652 	return sprintf(buf, "%u\n", policy->object);	\
653 }
654 
655 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
656 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
657 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
658 show_one(scaling_min_freq, min);
659 show_one(scaling_max_freq, max);
660 
661 __weak unsigned int arch_freq_get_on_cpu(int cpu)
662 {
663 	return 0;
664 }
665 
666 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
667 {
668 	ssize_t ret;
669 	unsigned int freq;
670 
671 	freq = arch_freq_get_on_cpu(policy->cpu);
672 	if (freq)
673 		ret = sprintf(buf, "%u\n", freq);
674 	else if (cpufreq_driver && cpufreq_driver->setpolicy &&
675 			cpufreq_driver->get)
676 		ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
677 	else
678 		ret = sprintf(buf, "%u\n", policy->cur);
679 	return ret;
680 }
681 
682 static int cpufreq_set_policy(struct cpufreq_policy *policy,
683 				struct cpufreq_policy *new_policy);
684 
685 /**
686  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
687  */
688 #define store_one(file_name, object)			\
689 static ssize_t store_##file_name					\
690 (struct cpufreq_policy *policy, const char *buf, size_t count)		\
691 {									\
692 	int ret, temp;							\
693 	struct cpufreq_policy new_policy;				\
694 									\
695 	memcpy(&new_policy, policy, sizeof(*policy));			\
696 									\
697 	ret = sscanf(buf, "%u", &new_policy.object);			\
698 	if (ret != 1)							\
699 		return -EINVAL;						\
700 									\
701 	temp = new_policy.object;					\
702 	ret = cpufreq_set_policy(policy, &new_policy);		\
703 	if (!ret)							\
704 		policy->user_policy.object = temp;			\
705 									\
706 	return ret ? ret : count;					\
707 }
708 
709 store_one(scaling_min_freq, min);
710 store_one(scaling_max_freq, max);
711 
712 /**
713  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
714  */
715 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
716 					char *buf)
717 {
718 	unsigned int cur_freq = __cpufreq_get(policy);
719 
720 	if (cur_freq)
721 		return sprintf(buf, "%u\n", cur_freq);
722 
723 	return sprintf(buf, "<unknown>\n");
724 }
725 
726 /**
727  * show_scaling_governor - show the current policy for the specified CPU
728  */
729 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
730 {
731 	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
732 		return sprintf(buf, "powersave\n");
733 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
734 		return sprintf(buf, "performance\n");
735 	else if (policy->governor)
736 		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
737 				policy->governor->name);
738 	return -EINVAL;
739 }
740 
741 /**
742  * store_scaling_governor - store policy for the specified CPU
743  */
744 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
745 					const char *buf, size_t count)
746 {
747 	int ret;
748 	char	str_governor[16];
749 	struct cpufreq_policy new_policy;
750 
751 	memcpy(&new_policy, policy, sizeof(*policy));
752 
753 	ret = sscanf(buf, "%15s", str_governor);
754 	if (ret != 1)
755 		return -EINVAL;
756 
757 	if (cpufreq_parse_governor(str_governor, &new_policy.policy,
758 						&new_policy.governor))
759 		return -EINVAL;
760 
761 	ret = cpufreq_set_policy(policy, &new_policy);
762 	return ret ? ret : count;
763 }
764 
765 /**
766  * show_scaling_driver - show the cpufreq driver currently loaded
767  */
768 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
769 {
770 	return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
771 }
772 
773 /**
774  * show_scaling_available_governors - show the available CPUfreq governors
775  */
776 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
777 						char *buf)
778 {
779 	ssize_t i = 0;
780 	struct cpufreq_governor *t;
781 
782 	if (!has_target()) {
783 		i += sprintf(buf, "performance powersave");
784 		goto out;
785 	}
786 
787 	for_each_governor(t) {
788 		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
789 		    - (CPUFREQ_NAME_LEN + 2)))
790 			goto out;
791 		i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
792 	}
793 out:
794 	i += sprintf(&buf[i], "\n");
795 	return i;
796 }
797 
798 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
799 {
800 	ssize_t i = 0;
801 	unsigned int cpu;
802 
803 	for_each_cpu(cpu, mask) {
804 		if (i)
805 			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
806 		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
807 		if (i >= (PAGE_SIZE - 5))
808 			break;
809 	}
810 	i += sprintf(&buf[i], "\n");
811 	return i;
812 }
813 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
814 
815 /**
816  * show_related_cpus - show the CPUs affected by each transition even if
817  * hw coordination is in use
818  */
819 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
820 {
821 	return cpufreq_show_cpus(policy->related_cpus, buf);
822 }
823 
824 /**
825  * show_affected_cpus - show the CPUs affected by each transition
826  */
827 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
828 {
829 	return cpufreq_show_cpus(policy->cpus, buf);
830 }
831 
832 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
833 					const char *buf, size_t count)
834 {
835 	unsigned int freq = 0;
836 	unsigned int ret;
837 
838 	if (!policy->governor || !policy->governor->store_setspeed)
839 		return -EINVAL;
840 
841 	ret = sscanf(buf, "%u", &freq);
842 	if (ret != 1)
843 		return -EINVAL;
844 
845 	policy->governor->store_setspeed(policy, freq);
846 
847 	return count;
848 }
849 
850 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
851 {
852 	if (!policy->governor || !policy->governor->show_setspeed)
853 		return sprintf(buf, "<unsupported>\n");
854 
855 	return policy->governor->show_setspeed(policy, buf);
856 }
857 
858 /**
859  * show_bios_limit - show the current cpufreq HW/BIOS limitation
860  */
861 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
862 {
863 	unsigned int limit;
864 	int ret;
865 	if (cpufreq_driver->bios_limit) {
866 		ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
867 		if (!ret)
868 			return sprintf(buf, "%u\n", limit);
869 	}
870 	return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
871 }
872 
873 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
874 cpufreq_freq_attr_ro(cpuinfo_min_freq);
875 cpufreq_freq_attr_ro(cpuinfo_max_freq);
876 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
877 cpufreq_freq_attr_ro(scaling_available_governors);
878 cpufreq_freq_attr_ro(scaling_driver);
879 cpufreq_freq_attr_ro(scaling_cur_freq);
880 cpufreq_freq_attr_ro(bios_limit);
881 cpufreq_freq_attr_ro(related_cpus);
882 cpufreq_freq_attr_ro(affected_cpus);
883 cpufreq_freq_attr_rw(scaling_min_freq);
884 cpufreq_freq_attr_rw(scaling_max_freq);
885 cpufreq_freq_attr_rw(scaling_governor);
886 cpufreq_freq_attr_rw(scaling_setspeed);
887 
888 static struct attribute *default_attrs[] = {
889 	&cpuinfo_min_freq.attr,
890 	&cpuinfo_max_freq.attr,
891 	&cpuinfo_transition_latency.attr,
892 	&scaling_min_freq.attr,
893 	&scaling_max_freq.attr,
894 	&affected_cpus.attr,
895 	&related_cpus.attr,
896 	&scaling_governor.attr,
897 	&scaling_driver.attr,
898 	&scaling_available_governors.attr,
899 	&scaling_setspeed.attr,
900 	NULL
901 };
902 
903 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
904 #define to_attr(a) container_of(a, struct freq_attr, attr)
905 
906 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
907 {
908 	struct cpufreq_policy *policy = to_policy(kobj);
909 	struct freq_attr *fattr = to_attr(attr);
910 	ssize_t ret;
911 
912 	down_read(&policy->rwsem);
913 	ret = fattr->show(policy, buf);
914 	up_read(&policy->rwsem);
915 
916 	return ret;
917 }
918 
919 static ssize_t store(struct kobject *kobj, struct attribute *attr,
920 		     const char *buf, size_t count)
921 {
922 	struct cpufreq_policy *policy = to_policy(kobj);
923 	struct freq_attr *fattr = to_attr(attr);
924 	ssize_t ret = -EINVAL;
925 
926 	cpus_read_lock();
927 
928 	if (cpu_online(policy->cpu)) {
929 		down_write(&policy->rwsem);
930 		ret = fattr->store(policy, buf, count);
931 		up_write(&policy->rwsem);
932 	}
933 
934 	cpus_read_unlock();
935 
936 	return ret;
937 }
938 
939 static void cpufreq_sysfs_release(struct kobject *kobj)
940 {
941 	struct cpufreq_policy *policy = to_policy(kobj);
942 	pr_debug("last reference is dropped\n");
943 	complete(&policy->kobj_unregister);
944 }
945 
946 static const struct sysfs_ops sysfs_ops = {
947 	.show	= show,
948 	.store	= store,
949 };
950 
951 static struct kobj_type ktype_cpufreq = {
952 	.sysfs_ops	= &sysfs_ops,
953 	.default_attrs	= default_attrs,
954 	.release	= cpufreq_sysfs_release,
955 };
956 
957 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
958 {
959 	struct device *dev = get_cpu_device(cpu);
960 
961 	if (!dev)
962 		return;
963 
964 	if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
965 		return;
966 
967 	dev_dbg(dev, "%s: Adding symlink\n", __func__);
968 	if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
969 		dev_err(dev, "cpufreq symlink creation failed\n");
970 }
971 
972 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
973 				   struct device *dev)
974 {
975 	dev_dbg(dev, "%s: Removing symlink\n", __func__);
976 	sysfs_remove_link(&dev->kobj, "cpufreq");
977 }
978 
979 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
980 {
981 	struct freq_attr **drv_attr;
982 	int ret = 0;
983 
984 	/* set up files for this cpu device */
985 	drv_attr = cpufreq_driver->attr;
986 	while (drv_attr && *drv_attr) {
987 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
988 		if (ret)
989 			return ret;
990 		drv_attr++;
991 	}
992 	if (cpufreq_driver->get) {
993 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
994 		if (ret)
995 			return ret;
996 	}
997 
998 	ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
999 	if (ret)
1000 		return ret;
1001 
1002 	if (cpufreq_driver->bios_limit) {
1003 		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1004 		if (ret)
1005 			return ret;
1006 	}
1007 
1008 	return 0;
1009 }
1010 
1011 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1012 {
1013 	return NULL;
1014 }
1015 
1016 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1017 {
1018 	struct cpufreq_governor *gov = NULL;
1019 	struct cpufreq_policy new_policy;
1020 
1021 	memcpy(&new_policy, policy, sizeof(*policy));
1022 
1023 	/* Update governor of new_policy to the governor used before hotplug */
1024 	gov = find_governor(policy->last_governor);
1025 	if (gov) {
1026 		pr_debug("Restoring governor %s for cpu %d\n",
1027 				policy->governor->name, policy->cpu);
1028 	} else {
1029 		gov = cpufreq_default_governor();
1030 		if (!gov)
1031 			return -ENODATA;
1032 	}
1033 
1034 	new_policy.governor = gov;
1035 
1036 	/* Use the default policy if there is no last_policy. */
1037 	if (cpufreq_driver->setpolicy) {
1038 		if (policy->last_policy)
1039 			new_policy.policy = policy->last_policy;
1040 		else
1041 			cpufreq_parse_governor(gov->name, &new_policy.policy,
1042 					       NULL);
1043 	}
1044 	/* set default policy */
1045 	return cpufreq_set_policy(policy, &new_policy);
1046 }
1047 
1048 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1049 {
1050 	int ret = 0;
1051 
1052 	/* Has this CPU been taken care of already? */
1053 	if (cpumask_test_cpu(cpu, policy->cpus))
1054 		return 0;
1055 
1056 	down_write(&policy->rwsem);
1057 	if (has_target())
1058 		cpufreq_stop_governor(policy);
1059 
1060 	cpumask_set_cpu(cpu, policy->cpus);
1061 
1062 	if (has_target()) {
1063 		ret = cpufreq_start_governor(policy);
1064 		if (ret)
1065 			pr_err("%s: Failed to start governor\n", __func__);
1066 	}
1067 	up_write(&policy->rwsem);
1068 	return ret;
1069 }
1070 
1071 static void handle_update(struct work_struct *work)
1072 {
1073 	struct cpufreq_policy *policy =
1074 		container_of(work, struct cpufreq_policy, update);
1075 	unsigned int cpu = policy->cpu;
1076 	pr_debug("handle_update for cpu %u called\n", cpu);
1077 	cpufreq_update_policy(cpu);
1078 }
1079 
1080 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1081 {
1082 	struct cpufreq_policy *policy;
1083 	int ret;
1084 
1085 	policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1086 	if (!policy)
1087 		return NULL;
1088 
1089 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1090 		goto err_free_policy;
1091 
1092 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1093 		goto err_free_cpumask;
1094 
1095 	if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1096 		goto err_free_rcpumask;
1097 
1098 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1099 				   cpufreq_global_kobject, "policy%u", cpu);
1100 	if (ret) {
1101 		pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1102 		goto err_free_real_cpus;
1103 	}
1104 
1105 	INIT_LIST_HEAD(&policy->policy_list);
1106 	init_rwsem(&policy->rwsem);
1107 	spin_lock_init(&policy->transition_lock);
1108 	init_waitqueue_head(&policy->transition_wait);
1109 	init_completion(&policy->kobj_unregister);
1110 	INIT_WORK(&policy->update, handle_update);
1111 
1112 	policy->cpu = cpu;
1113 	return policy;
1114 
1115 err_free_real_cpus:
1116 	free_cpumask_var(policy->real_cpus);
1117 err_free_rcpumask:
1118 	free_cpumask_var(policy->related_cpus);
1119 err_free_cpumask:
1120 	free_cpumask_var(policy->cpus);
1121 err_free_policy:
1122 	kfree(policy);
1123 
1124 	return NULL;
1125 }
1126 
1127 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1128 {
1129 	struct kobject *kobj;
1130 	struct completion *cmp;
1131 
1132 	down_write(&policy->rwsem);
1133 	cpufreq_stats_free_table(policy);
1134 	kobj = &policy->kobj;
1135 	cmp = &policy->kobj_unregister;
1136 	up_write(&policy->rwsem);
1137 	kobject_put(kobj);
1138 
1139 	/*
1140 	 * We need to make sure that the underlying kobj is
1141 	 * actually not referenced anymore by anybody before we
1142 	 * proceed with unloading.
1143 	 */
1144 	pr_debug("waiting for dropping of refcount\n");
1145 	wait_for_completion(cmp);
1146 	pr_debug("wait complete\n");
1147 }
1148 
1149 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1150 {
1151 	unsigned long flags;
1152 	int cpu;
1153 
1154 	/* Remove policy from list */
1155 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1156 	list_del(&policy->policy_list);
1157 
1158 	for_each_cpu(cpu, policy->related_cpus)
1159 		per_cpu(cpufreq_cpu_data, cpu) = NULL;
1160 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1161 
1162 	cpufreq_policy_put_kobj(policy);
1163 	free_cpumask_var(policy->real_cpus);
1164 	free_cpumask_var(policy->related_cpus);
1165 	free_cpumask_var(policy->cpus);
1166 	kfree(policy);
1167 }
1168 
1169 static int cpufreq_online(unsigned int cpu)
1170 {
1171 	struct cpufreq_policy *policy;
1172 	bool new_policy;
1173 	unsigned long flags;
1174 	unsigned int j;
1175 	int ret;
1176 
1177 	pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1178 
1179 	/* Check if this CPU already has a policy to manage it */
1180 	policy = per_cpu(cpufreq_cpu_data, cpu);
1181 	if (policy) {
1182 		WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1183 		if (!policy_is_inactive(policy))
1184 			return cpufreq_add_policy_cpu(policy, cpu);
1185 
1186 		/* This is the only online CPU for the policy.  Start over. */
1187 		new_policy = false;
1188 		down_write(&policy->rwsem);
1189 		policy->cpu = cpu;
1190 		policy->governor = NULL;
1191 		up_write(&policy->rwsem);
1192 	} else {
1193 		new_policy = true;
1194 		policy = cpufreq_policy_alloc(cpu);
1195 		if (!policy)
1196 			return -ENOMEM;
1197 	}
1198 
1199 	cpumask_copy(policy->cpus, cpumask_of(cpu));
1200 
1201 	/* call driver. From then on the cpufreq must be able
1202 	 * to accept all calls to ->verify and ->setpolicy for this CPU
1203 	 */
1204 	ret = cpufreq_driver->init(policy);
1205 	if (ret) {
1206 		pr_debug("initialization failed\n");
1207 		goto out_free_policy;
1208 	}
1209 
1210 	down_write(&policy->rwsem);
1211 
1212 	if (new_policy) {
1213 		/* related_cpus should at least include policy->cpus. */
1214 		cpumask_copy(policy->related_cpus, policy->cpus);
1215 	}
1216 
1217 	/*
1218 	 * affected cpus must always be the one, which are online. We aren't
1219 	 * managing offline cpus here.
1220 	 */
1221 	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1222 
1223 	if (new_policy) {
1224 		policy->user_policy.min = policy->min;
1225 		policy->user_policy.max = policy->max;
1226 
1227 		for_each_cpu(j, policy->related_cpus) {
1228 			per_cpu(cpufreq_cpu_data, j) = policy;
1229 			add_cpu_dev_symlink(policy, j);
1230 		}
1231 	} else {
1232 		policy->min = policy->user_policy.min;
1233 		policy->max = policy->user_policy.max;
1234 	}
1235 
1236 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1237 		policy->cur = cpufreq_driver->get(policy->cpu);
1238 		if (!policy->cur) {
1239 			pr_err("%s: ->get() failed\n", __func__);
1240 			goto out_exit_policy;
1241 		}
1242 	}
1243 
1244 	/*
1245 	 * Sometimes boot loaders set CPU frequency to a value outside of
1246 	 * frequency table present with cpufreq core. In such cases CPU might be
1247 	 * unstable if it has to run on that frequency for long duration of time
1248 	 * and so its better to set it to a frequency which is specified in
1249 	 * freq-table. This also makes cpufreq stats inconsistent as
1250 	 * cpufreq-stats would fail to register because current frequency of CPU
1251 	 * isn't found in freq-table.
1252 	 *
1253 	 * Because we don't want this change to effect boot process badly, we go
1254 	 * for the next freq which is >= policy->cur ('cur' must be set by now,
1255 	 * otherwise we will end up setting freq to lowest of the table as 'cur'
1256 	 * is initialized to zero).
1257 	 *
1258 	 * We are passing target-freq as "policy->cur - 1" otherwise
1259 	 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1260 	 * equal to target-freq.
1261 	 */
1262 	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1263 	    && has_target()) {
1264 		/* Are we running at unknown frequency ? */
1265 		ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1266 		if (ret == -EINVAL) {
1267 			/* Warn user and fix it */
1268 			pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1269 				__func__, policy->cpu, policy->cur);
1270 			ret = __cpufreq_driver_target(policy, policy->cur - 1,
1271 				CPUFREQ_RELATION_L);
1272 
1273 			/*
1274 			 * Reaching here after boot in a few seconds may not
1275 			 * mean that system will remain stable at "unknown"
1276 			 * frequency for longer duration. Hence, a BUG_ON().
1277 			 */
1278 			BUG_ON(ret);
1279 			pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1280 				__func__, policy->cpu, policy->cur);
1281 		}
1282 	}
1283 
1284 	if (new_policy) {
1285 		ret = cpufreq_add_dev_interface(policy);
1286 		if (ret)
1287 			goto out_exit_policy;
1288 
1289 		cpufreq_stats_create_table(policy);
1290 
1291 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1292 		list_add(&policy->policy_list, &cpufreq_policy_list);
1293 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1294 	}
1295 
1296 	ret = cpufreq_init_policy(policy);
1297 	if (ret) {
1298 		pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1299 		       __func__, cpu, ret);
1300 		/* cpufreq_policy_free() will notify based on this */
1301 		new_policy = false;
1302 		goto out_exit_policy;
1303 	}
1304 
1305 	up_write(&policy->rwsem);
1306 
1307 	kobject_uevent(&policy->kobj, KOBJ_ADD);
1308 
1309 	/* Callback for handling stuff after policy is ready */
1310 	if (cpufreq_driver->ready)
1311 		cpufreq_driver->ready(policy);
1312 
1313 	pr_debug("initialization complete\n");
1314 
1315 	return 0;
1316 
1317 out_exit_policy:
1318 	up_write(&policy->rwsem);
1319 
1320 	if (cpufreq_driver->exit)
1321 		cpufreq_driver->exit(policy);
1322 
1323 	for_each_cpu(j, policy->real_cpus)
1324 		remove_cpu_dev_symlink(policy, get_cpu_device(j));
1325 
1326 out_free_policy:
1327 	cpufreq_policy_free(policy);
1328 	return ret;
1329 }
1330 
1331 /**
1332  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1333  * @dev: CPU device.
1334  * @sif: Subsystem interface structure pointer (not used)
1335  */
1336 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1337 {
1338 	struct cpufreq_policy *policy;
1339 	unsigned cpu = dev->id;
1340 	int ret;
1341 
1342 	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1343 
1344 	if (cpu_online(cpu)) {
1345 		ret = cpufreq_online(cpu);
1346 		if (ret)
1347 			return ret;
1348 	}
1349 
1350 	/* Create sysfs link on CPU registration */
1351 	policy = per_cpu(cpufreq_cpu_data, cpu);
1352 	if (policy)
1353 		add_cpu_dev_symlink(policy, cpu);
1354 
1355 	return 0;
1356 }
1357 
1358 static int cpufreq_offline(unsigned int cpu)
1359 {
1360 	struct cpufreq_policy *policy;
1361 	int ret;
1362 
1363 	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1364 
1365 	policy = cpufreq_cpu_get_raw(cpu);
1366 	if (!policy) {
1367 		pr_debug("%s: No cpu_data found\n", __func__);
1368 		return 0;
1369 	}
1370 
1371 	down_write(&policy->rwsem);
1372 	if (has_target())
1373 		cpufreq_stop_governor(policy);
1374 
1375 	cpumask_clear_cpu(cpu, policy->cpus);
1376 
1377 	if (policy_is_inactive(policy)) {
1378 		if (has_target())
1379 			strncpy(policy->last_governor, policy->governor->name,
1380 				CPUFREQ_NAME_LEN);
1381 		else
1382 			policy->last_policy = policy->policy;
1383 	} else if (cpu == policy->cpu) {
1384 		/* Nominate new CPU */
1385 		policy->cpu = cpumask_any(policy->cpus);
1386 	}
1387 
1388 	/* Start governor again for active policy */
1389 	if (!policy_is_inactive(policy)) {
1390 		if (has_target()) {
1391 			ret = cpufreq_start_governor(policy);
1392 			if (ret)
1393 				pr_err("%s: Failed to start governor\n", __func__);
1394 		}
1395 
1396 		goto unlock;
1397 	}
1398 
1399 	if (cpufreq_driver->stop_cpu)
1400 		cpufreq_driver->stop_cpu(policy);
1401 
1402 	if (has_target())
1403 		cpufreq_exit_governor(policy);
1404 
1405 	/*
1406 	 * Perform the ->exit() even during light-weight tear-down,
1407 	 * since this is a core component, and is essential for the
1408 	 * subsequent light-weight ->init() to succeed.
1409 	 */
1410 	if (cpufreq_driver->exit) {
1411 		cpufreq_driver->exit(policy);
1412 		policy->freq_table = NULL;
1413 	}
1414 
1415 unlock:
1416 	up_write(&policy->rwsem);
1417 	return 0;
1418 }
1419 
1420 /**
1421  * cpufreq_remove_dev - remove a CPU device
1422  *
1423  * Removes the cpufreq interface for a CPU device.
1424  */
1425 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1426 {
1427 	unsigned int cpu = dev->id;
1428 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1429 
1430 	if (!policy)
1431 		return;
1432 
1433 	if (cpu_online(cpu))
1434 		cpufreq_offline(cpu);
1435 
1436 	cpumask_clear_cpu(cpu, policy->real_cpus);
1437 	remove_cpu_dev_symlink(policy, dev);
1438 
1439 	if (cpumask_empty(policy->real_cpus))
1440 		cpufreq_policy_free(policy);
1441 }
1442 
1443 /**
1444  *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1445  *	in deep trouble.
1446  *	@policy: policy managing CPUs
1447  *	@new_freq: CPU frequency the CPU actually runs at
1448  *
1449  *	We adjust to current frequency first, and need to clean up later.
1450  *	So either call to cpufreq_update_policy() or schedule handle_update()).
1451  */
1452 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1453 				unsigned int new_freq)
1454 {
1455 	struct cpufreq_freqs freqs;
1456 
1457 	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1458 		 policy->cur, new_freq);
1459 
1460 	freqs.old = policy->cur;
1461 	freqs.new = new_freq;
1462 
1463 	cpufreq_freq_transition_begin(policy, &freqs);
1464 	cpufreq_freq_transition_end(policy, &freqs, 0);
1465 }
1466 
1467 /**
1468  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1469  * @cpu: CPU number
1470  *
1471  * This is the last known freq, without actually getting it from the driver.
1472  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1473  */
1474 unsigned int cpufreq_quick_get(unsigned int cpu)
1475 {
1476 	struct cpufreq_policy *policy;
1477 	unsigned int ret_freq = 0;
1478 	unsigned long flags;
1479 
1480 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1481 
1482 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1483 		ret_freq = cpufreq_driver->get(cpu);
1484 		read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1485 		return ret_freq;
1486 	}
1487 
1488 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1489 
1490 	policy = cpufreq_cpu_get(cpu);
1491 	if (policy) {
1492 		ret_freq = policy->cur;
1493 		cpufreq_cpu_put(policy);
1494 	}
1495 
1496 	return ret_freq;
1497 }
1498 EXPORT_SYMBOL(cpufreq_quick_get);
1499 
1500 /**
1501  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1502  * @cpu: CPU number
1503  *
1504  * Just return the max possible frequency for a given CPU.
1505  */
1506 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1507 {
1508 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1509 	unsigned int ret_freq = 0;
1510 
1511 	if (policy) {
1512 		ret_freq = policy->max;
1513 		cpufreq_cpu_put(policy);
1514 	}
1515 
1516 	return ret_freq;
1517 }
1518 EXPORT_SYMBOL(cpufreq_quick_get_max);
1519 
1520 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1521 {
1522 	unsigned int ret_freq = 0;
1523 
1524 	if (!cpufreq_driver->get)
1525 		return ret_freq;
1526 
1527 	ret_freq = cpufreq_driver->get(policy->cpu);
1528 
1529 	/*
1530 	 * Updating inactive policies is invalid, so avoid doing that.  Also
1531 	 * if fast frequency switching is used with the given policy, the check
1532 	 * against policy->cur is pointless, so skip it in that case too.
1533 	 */
1534 	if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1535 		return ret_freq;
1536 
1537 	if (ret_freq && policy->cur &&
1538 		!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1539 		/* verify no discrepancy between actual and
1540 					saved value exists */
1541 		if (unlikely(ret_freq != policy->cur)) {
1542 			cpufreq_out_of_sync(policy, ret_freq);
1543 			schedule_work(&policy->update);
1544 		}
1545 	}
1546 
1547 	return ret_freq;
1548 }
1549 
1550 /**
1551  * cpufreq_get - get the current CPU frequency (in kHz)
1552  * @cpu: CPU number
1553  *
1554  * Get the CPU current (static) CPU frequency
1555  */
1556 unsigned int cpufreq_get(unsigned int cpu)
1557 {
1558 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1559 	unsigned int ret_freq = 0;
1560 
1561 	if (policy) {
1562 		down_read(&policy->rwsem);
1563 
1564 		if (!policy_is_inactive(policy))
1565 			ret_freq = __cpufreq_get(policy);
1566 
1567 		up_read(&policy->rwsem);
1568 
1569 		cpufreq_cpu_put(policy);
1570 	}
1571 
1572 	return ret_freq;
1573 }
1574 EXPORT_SYMBOL(cpufreq_get);
1575 
1576 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1577 {
1578 	unsigned int new_freq;
1579 
1580 	new_freq = cpufreq_driver->get(policy->cpu);
1581 	if (!new_freq)
1582 		return 0;
1583 
1584 	if (!policy->cur) {
1585 		pr_debug("cpufreq: Driver did not initialize current freq\n");
1586 		policy->cur = new_freq;
1587 	} else if (policy->cur != new_freq && has_target()) {
1588 		cpufreq_out_of_sync(policy, new_freq);
1589 	}
1590 
1591 	return new_freq;
1592 }
1593 
1594 static struct subsys_interface cpufreq_interface = {
1595 	.name		= "cpufreq",
1596 	.subsys		= &cpu_subsys,
1597 	.add_dev	= cpufreq_add_dev,
1598 	.remove_dev	= cpufreq_remove_dev,
1599 };
1600 
1601 /*
1602  * In case platform wants some specific frequency to be configured
1603  * during suspend..
1604  */
1605 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1606 {
1607 	int ret;
1608 
1609 	if (!policy->suspend_freq) {
1610 		pr_debug("%s: suspend_freq not defined\n", __func__);
1611 		return 0;
1612 	}
1613 
1614 	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1615 			policy->suspend_freq);
1616 
1617 	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1618 			CPUFREQ_RELATION_H);
1619 	if (ret)
1620 		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1621 				__func__, policy->suspend_freq, ret);
1622 
1623 	return ret;
1624 }
1625 EXPORT_SYMBOL(cpufreq_generic_suspend);
1626 
1627 /**
1628  * cpufreq_suspend() - Suspend CPUFreq governors
1629  *
1630  * Called during system wide Suspend/Hibernate cycles for suspending governors
1631  * as some platforms can't change frequency after this point in suspend cycle.
1632  * Because some of the devices (like: i2c, regulators, etc) they use for
1633  * changing frequency are suspended quickly after this point.
1634  */
1635 void cpufreq_suspend(void)
1636 {
1637 	struct cpufreq_policy *policy;
1638 
1639 	if (!cpufreq_driver)
1640 		return;
1641 
1642 	if (!has_target() && !cpufreq_driver->suspend)
1643 		goto suspend;
1644 
1645 	pr_debug("%s: Suspending Governors\n", __func__);
1646 
1647 	for_each_active_policy(policy) {
1648 		if (has_target()) {
1649 			down_write(&policy->rwsem);
1650 			cpufreq_stop_governor(policy);
1651 			up_write(&policy->rwsem);
1652 		}
1653 
1654 		if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1655 			pr_err("%s: Failed to suspend driver: %p\n", __func__,
1656 				policy);
1657 	}
1658 
1659 suspend:
1660 	cpufreq_suspended = true;
1661 }
1662 
1663 /**
1664  * cpufreq_resume() - Resume CPUFreq governors
1665  *
1666  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1667  * are suspended with cpufreq_suspend().
1668  */
1669 void cpufreq_resume(void)
1670 {
1671 	struct cpufreq_policy *policy;
1672 	int ret;
1673 
1674 	if (!cpufreq_driver)
1675 		return;
1676 
1677 	cpufreq_suspended = false;
1678 
1679 	if (!has_target() && !cpufreq_driver->resume)
1680 		return;
1681 
1682 	pr_debug("%s: Resuming Governors\n", __func__);
1683 
1684 	for_each_active_policy(policy) {
1685 		if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1686 			pr_err("%s: Failed to resume driver: %p\n", __func__,
1687 				policy);
1688 		} else if (has_target()) {
1689 			down_write(&policy->rwsem);
1690 			ret = cpufreq_start_governor(policy);
1691 			up_write(&policy->rwsem);
1692 
1693 			if (ret)
1694 				pr_err("%s: Failed to start governor for policy: %p\n",
1695 				       __func__, policy);
1696 		}
1697 	}
1698 }
1699 
1700 /**
1701  *	cpufreq_get_current_driver - return current driver's name
1702  *
1703  *	Return the name string of the currently loaded cpufreq driver
1704  *	or NULL, if none.
1705  */
1706 const char *cpufreq_get_current_driver(void)
1707 {
1708 	if (cpufreq_driver)
1709 		return cpufreq_driver->name;
1710 
1711 	return NULL;
1712 }
1713 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1714 
1715 /**
1716  *	cpufreq_get_driver_data - return current driver data
1717  *
1718  *	Return the private data of the currently loaded cpufreq
1719  *	driver, or NULL if no cpufreq driver is loaded.
1720  */
1721 void *cpufreq_get_driver_data(void)
1722 {
1723 	if (cpufreq_driver)
1724 		return cpufreq_driver->driver_data;
1725 
1726 	return NULL;
1727 }
1728 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1729 
1730 /*********************************************************************
1731  *                     NOTIFIER LISTS INTERFACE                      *
1732  *********************************************************************/
1733 
1734 /**
1735  *	cpufreq_register_notifier - register a driver with cpufreq
1736  *	@nb: notifier function to register
1737  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1738  *
1739  *	Add a driver to one of two lists: either a list of drivers that
1740  *      are notified about clock rate changes (once before and once after
1741  *      the transition), or a list of drivers that are notified about
1742  *      changes in cpufreq policy.
1743  *
1744  *	This function may sleep, and has the same return conditions as
1745  *	blocking_notifier_chain_register.
1746  */
1747 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1748 {
1749 	int ret;
1750 
1751 	if (cpufreq_disabled())
1752 		return -EINVAL;
1753 
1754 	WARN_ON(!init_cpufreq_transition_notifier_list_called);
1755 
1756 	switch (list) {
1757 	case CPUFREQ_TRANSITION_NOTIFIER:
1758 		mutex_lock(&cpufreq_fast_switch_lock);
1759 
1760 		if (cpufreq_fast_switch_count > 0) {
1761 			mutex_unlock(&cpufreq_fast_switch_lock);
1762 			return -EBUSY;
1763 		}
1764 		ret = srcu_notifier_chain_register(
1765 				&cpufreq_transition_notifier_list, nb);
1766 		if (!ret)
1767 			cpufreq_fast_switch_count--;
1768 
1769 		mutex_unlock(&cpufreq_fast_switch_lock);
1770 		break;
1771 	case CPUFREQ_POLICY_NOTIFIER:
1772 		ret = blocking_notifier_chain_register(
1773 				&cpufreq_policy_notifier_list, nb);
1774 		break;
1775 	default:
1776 		ret = -EINVAL;
1777 	}
1778 
1779 	return ret;
1780 }
1781 EXPORT_SYMBOL(cpufreq_register_notifier);
1782 
1783 /**
1784  *	cpufreq_unregister_notifier - unregister a driver with cpufreq
1785  *	@nb: notifier block to be unregistered
1786  *	@list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1787  *
1788  *	Remove a driver from the CPU frequency notifier list.
1789  *
1790  *	This function may sleep, and has the same return conditions as
1791  *	blocking_notifier_chain_unregister.
1792  */
1793 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1794 {
1795 	int ret;
1796 
1797 	if (cpufreq_disabled())
1798 		return -EINVAL;
1799 
1800 	switch (list) {
1801 	case CPUFREQ_TRANSITION_NOTIFIER:
1802 		mutex_lock(&cpufreq_fast_switch_lock);
1803 
1804 		ret = srcu_notifier_chain_unregister(
1805 				&cpufreq_transition_notifier_list, nb);
1806 		if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1807 			cpufreq_fast_switch_count++;
1808 
1809 		mutex_unlock(&cpufreq_fast_switch_lock);
1810 		break;
1811 	case CPUFREQ_POLICY_NOTIFIER:
1812 		ret = blocking_notifier_chain_unregister(
1813 				&cpufreq_policy_notifier_list, nb);
1814 		break;
1815 	default:
1816 		ret = -EINVAL;
1817 	}
1818 
1819 	return ret;
1820 }
1821 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1822 
1823 
1824 /*********************************************************************
1825  *                              GOVERNORS                            *
1826  *********************************************************************/
1827 
1828 /**
1829  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1830  * @policy: cpufreq policy to switch the frequency for.
1831  * @target_freq: New frequency to set (may be approximate).
1832  *
1833  * Carry out a fast frequency switch without sleeping.
1834  *
1835  * The driver's ->fast_switch() callback invoked by this function must be
1836  * suitable for being called from within RCU-sched read-side critical sections
1837  * and it is expected to select the minimum available frequency greater than or
1838  * equal to @target_freq (CPUFREQ_RELATION_L).
1839  *
1840  * This function must not be called if policy->fast_switch_enabled is unset.
1841  *
1842  * Governors calling this function must guarantee that it will never be invoked
1843  * twice in parallel for the same policy and that it will never be called in
1844  * parallel with either ->target() or ->target_index() for the same policy.
1845  *
1846  * Returns the actual frequency set for the CPU.
1847  *
1848  * If 0 is returned by the driver's ->fast_switch() callback to indicate an
1849  * error condition, the hardware configuration must be preserved.
1850  */
1851 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1852 					unsigned int target_freq)
1853 {
1854 	target_freq = clamp_val(target_freq, policy->min, policy->max);
1855 
1856 	return cpufreq_driver->fast_switch(policy, target_freq);
1857 }
1858 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1859 
1860 /* Must set freqs->new to intermediate frequency */
1861 static int __target_intermediate(struct cpufreq_policy *policy,
1862 				 struct cpufreq_freqs *freqs, int index)
1863 {
1864 	int ret;
1865 
1866 	freqs->new = cpufreq_driver->get_intermediate(policy, index);
1867 
1868 	/* We don't need to switch to intermediate freq */
1869 	if (!freqs->new)
1870 		return 0;
1871 
1872 	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1873 		 __func__, policy->cpu, freqs->old, freqs->new);
1874 
1875 	cpufreq_freq_transition_begin(policy, freqs);
1876 	ret = cpufreq_driver->target_intermediate(policy, index);
1877 	cpufreq_freq_transition_end(policy, freqs, ret);
1878 
1879 	if (ret)
1880 		pr_err("%s: Failed to change to intermediate frequency: %d\n",
1881 		       __func__, ret);
1882 
1883 	return ret;
1884 }
1885 
1886 static int __target_index(struct cpufreq_policy *policy, int index)
1887 {
1888 	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1889 	unsigned int intermediate_freq = 0;
1890 	unsigned int newfreq = policy->freq_table[index].frequency;
1891 	int retval = -EINVAL;
1892 	bool notify;
1893 
1894 	if (newfreq == policy->cur)
1895 		return 0;
1896 
1897 	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1898 	if (notify) {
1899 		/* Handle switching to intermediate frequency */
1900 		if (cpufreq_driver->get_intermediate) {
1901 			retval = __target_intermediate(policy, &freqs, index);
1902 			if (retval)
1903 				return retval;
1904 
1905 			intermediate_freq = freqs.new;
1906 			/* Set old freq to intermediate */
1907 			if (intermediate_freq)
1908 				freqs.old = freqs.new;
1909 		}
1910 
1911 		freqs.new = newfreq;
1912 		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1913 			 __func__, policy->cpu, freqs.old, freqs.new);
1914 
1915 		cpufreq_freq_transition_begin(policy, &freqs);
1916 	}
1917 
1918 	retval = cpufreq_driver->target_index(policy, index);
1919 	if (retval)
1920 		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1921 		       retval);
1922 
1923 	if (notify) {
1924 		cpufreq_freq_transition_end(policy, &freqs, retval);
1925 
1926 		/*
1927 		 * Failed after setting to intermediate freq? Driver should have
1928 		 * reverted back to initial frequency and so should we. Check
1929 		 * here for intermediate_freq instead of get_intermediate, in
1930 		 * case we haven't switched to intermediate freq at all.
1931 		 */
1932 		if (unlikely(retval && intermediate_freq)) {
1933 			freqs.old = intermediate_freq;
1934 			freqs.new = policy->restore_freq;
1935 			cpufreq_freq_transition_begin(policy, &freqs);
1936 			cpufreq_freq_transition_end(policy, &freqs, 0);
1937 		}
1938 	}
1939 
1940 	return retval;
1941 }
1942 
1943 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1944 			    unsigned int target_freq,
1945 			    unsigned int relation)
1946 {
1947 	unsigned int old_target_freq = target_freq;
1948 	int index;
1949 
1950 	if (cpufreq_disabled())
1951 		return -ENODEV;
1952 
1953 	/* Make sure that target_freq is within supported range */
1954 	target_freq = clamp_val(target_freq, policy->min, policy->max);
1955 
1956 	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1957 		 policy->cpu, target_freq, relation, old_target_freq);
1958 
1959 	/*
1960 	 * This might look like a redundant call as we are checking it again
1961 	 * after finding index. But it is left intentionally for cases where
1962 	 * exactly same freq is called again and so we can save on few function
1963 	 * calls.
1964 	 */
1965 	if (target_freq == policy->cur)
1966 		return 0;
1967 
1968 	/* Save last value to restore later on errors */
1969 	policy->restore_freq = policy->cur;
1970 
1971 	if (cpufreq_driver->target)
1972 		return cpufreq_driver->target(policy, target_freq, relation);
1973 
1974 	if (!cpufreq_driver->target_index)
1975 		return -EINVAL;
1976 
1977 	index = cpufreq_frequency_table_target(policy, target_freq, relation);
1978 
1979 	return __target_index(policy, index);
1980 }
1981 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1982 
1983 int cpufreq_driver_target(struct cpufreq_policy *policy,
1984 			  unsigned int target_freq,
1985 			  unsigned int relation)
1986 {
1987 	int ret = -EINVAL;
1988 
1989 	down_write(&policy->rwsem);
1990 
1991 	ret = __cpufreq_driver_target(policy, target_freq, relation);
1992 
1993 	up_write(&policy->rwsem);
1994 
1995 	return ret;
1996 }
1997 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1998 
1999 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2000 {
2001 	return NULL;
2002 }
2003 
2004 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2005 {
2006 	int ret;
2007 
2008 	/* Don't start any governor operations if we are entering suspend */
2009 	if (cpufreq_suspended)
2010 		return 0;
2011 	/*
2012 	 * Governor might not be initiated here if ACPI _PPC changed
2013 	 * notification happened, so check it.
2014 	 */
2015 	if (!policy->governor)
2016 		return -EINVAL;
2017 
2018 	/* Platform doesn't want dynamic frequency switching ? */
2019 	if (policy->governor->dynamic_switching &&
2020 	    cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2021 		struct cpufreq_governor *gov = cpufreq_fallback_governor();
2022 
2023 		if (gov) {
2024 			pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2025 				policy->governor->name, gov->name);
2026 			policy->governor = gov;
2027 		} else {
2028 			return -EINVAL;
2029 		}
2030 	}
2031 
2032 	if (!try_module_get(policy->governor->owner))
2033 		return -EINVAL;
2034 
2035 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2036 
2037 	if (policy->governor->init) {
2038 		ret = policy->governor->init(policy);
2039 		if (ret) {
2040 			module_put(policy->governor->owner);
2041 			return ret;
2042 		}
2043 	}
2044 
2045 	return 0;
2046 }
2047 
2048 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2049 {
2050 	if (cpufreq_suspended || !policy->governor)
2051 		return;
2052 
2053 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2054 
2055 	if (policy->governor->exit)
2056 		policy->governor->exit(policy);
2057 
2058 	module_put(policy->governor->owner);
2059 }
2060 
2061 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2062 {
2063 	int ret;
2064 
2065 	if (cpufreq_suspended)
2066 		return 0;
2067 
2068 	if (!policy->governor)
2069 		return -EINVAL;
2070 
2071 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2072 
2073 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2074 		cpufreq_update_current_freq(policy);
2075 
2076 	if (policy->governor->start) {
2077 		ret = policy->governor->start(policy);
2078 		if (ret)
2079 			return ret;
2080 	}
2081 
2082 	if (policy->governor->limits)
2083 		policy->governor->limits(policy);
2084 
2085 	return 0;
2086 }
2087 
2088 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2089 {
2090 	if (cpufreq_suspended || !policy->governor)
2091 		return;
2092 
2093 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2094 
2095 	if (policy->governor->stop)
2096 		policy->governor->stop(policy);
2097 }
2098 
2099 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2100 {
2101 	if (cpufreq_suspended || !policy->governor)
2102 		return;
2103 
2104 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2105 
2106 	if (policy->governor->limits)
2107 		policy->governor->limits(policy);
2108 }
2109 
2110 int cpufreq_register_governor(struct cpufreq_governor *governor)
2111 {
2112 	int err;
2113 
2114 	if (!governor)
2115 		return -EINVAL;
2116 
2117 	if (cpufreq_disabled())
2118 		return -ENODEV;
2119 
2120 	mutex_lock(&cpufreq_governor_mutex);
2121 
2122 	err = -EBUSY;
2123 	if (!find_governor(governor->name)) {
2124 		err = 0;
2125 		list_add(&governor->governor_list, &cpufreq_governor_list);
2126 	}
2127 
2128 	mutex_unlock(&cpufreq_governor_mutex);
2129 	return err;
2130 }
2131 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2132 
2133 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2134 {
2135 	struct cpufreq_policy *policy;
2136 	unsigned long flags;
2137 
2138 	if (!governor)
2139 		return;
2140 
2141 	if (cpufreq_disabled())
2142 		return;
2143 
2144 	/* clear last_governor for all inactive policies */
2145 	read_lock_irqsave(&cpufreq_driver_lock, flags);
2146 	for_each_inactive_policy(policy) {
2147 		if (!strcmp(policy->last_governor, governor->name)) {
2148 			policy->governor = NULL;
2149 			strcpy(policy->last_governor, "\0");
2150 		}
2151 	}
2152 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2153 
2154 	mutex_lock(&cpufreq_governor_mutex);
2155 	list_del(&governor->governor_list);
2156 	mutex_unlock(&cpufreq_governor_mutex);
2157 	return;
2158 }
2159 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2160 
2161 
2162 /*********************************************************************
2163  *                          POLICY INTERFACE                         *
2164  *********************************************************************/
2165 
2166 /**
2167  * cpufreq_get_policy - get the current cpufreq_policy
2168  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2169  *	is written
2170  *
2171  * Reads the current cpufreq policy.
2172  */
2173 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2174 {
2175 	struct cpufreq_policy *cpu_policy;
2176 	if (!policy)
2177 		return -EINVAL;
2178 
2179 	cpu_policy = cpufreq_cpu_get(cpu);
2180 	if (!cpu_policy)
2181 		return -EINVAL;
2182 
2183 	memcpy(policy, cpu_policy, sizeof(*policy));
2184 
2185 	cpufreq_cpu_put(cpu_policy);
2186 	return 0;
2187 }
2188 EXPORT_SYMBOL(cpufreq_get_policy);
2189 
2190 /*
2191  * policy : current policy.
2192  * new_policy: policy to be set.
2193  */
2194 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2195 				struct cpufreq_policy *new_policy)
2196 {
2197 	struct cpufreq_governor *old_gov;
2198 	int ret;
2199 
2200 	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2201 		 new_policy->cpu, new_policy->min, new_policy->max);
2202 
2203 	memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2204 
2205 	/*
2206 	* This check works well when we store new min/max freq attributes,
2207 	* because new_policy is a copy of policy with one field updated.
2208 	*/
2209 	if (new_policy->min > new_policy->max)
2210 		return -EINVAL;
2211 
2212 	/* verify the cpu speed can be set within this limit */
2213 	ret = cpufreq_driver->verify(new_policy);
2214 	if (ret)
2215 		return ret;
2216 
2217 	/* adjust if necessary - all reasons */
2218 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2219 			CPUFREQ_ADJUST, new_policy);
2220 
2221 	/*
2222 	 * verify the cpu speed can be set within this limit, which might be
2223 	 * different to the first one
2224 	 */
2225 	ret = cpufreq_driver->verify(new_policy);
2226 	if (ret)
2227 		return ret;
2228 
2229 	/* notification of the new policy */
2230 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2231 			CPUFREQ_NOTIFY, new_policy);
2232 
2233 	policy->min = new_policy->min;
2234 	policy->max = new_policy->max;
2235 
2236 	policy->cached_target_freq = UINT_MAX;
2237 
2238 	pr_debug("new min and max freqs are %u - %u kHz\n",
2239 		 policy->min, policy->max);
2240 
2241 	if (cpufreq_driver->setpolicy) {
2242 		policy->policy = new_policy->policy;
2243 		pr_debug("setting range\n");
2244 		return cpufreq_driver->setpolicy(new_policy);
2245 	}
2246 
2247 	if (new_policy->governor == policy->governor) {
2248 		pr_debug("cpufreq: governor limits update\n");
2249 		cpufreq_governor_limits(policy);
2250 		return 0;
2251 	}
2252 
2253 	pr_debug("governor switch\n");
2254 
2255 	/* save old, working values */
2256 	old_gov = policy->governor;
2257 	/* end old governor */
2258 	if (old_gov) {
2259 		cpufreq_stop_governor(policy);
2260 		cpufreq_exit_governor(policy);
2261 	}
2262 
2263 	/* start new governor */
2264 	policy->governor = new_policy->governor;
2265 	ret = cpufreq_init_governor(policy);
2266 	if (!ret) {
2267 		ret = cpufreq_start_governor(policy);
2268 		if (!ret) {
2269 			pr_debug("cpufreq: governor change\n");
2270 			return 0;
2271 		}
2272 		cpufreq_exit_governor(policy);
2273 	}
2274 
2275 	/* new governor failed, so re-start old one */
2276 	pr_debug("starting governor %s failed\n", policy->governor->name);
2277 	if (old_gov) {
2278 		policy->governor = old_gov;
2279 		if (cpufreq_init_governor(policy))
2280 			policy->governor = NULL;
2281 		else
2282 			cpufreq_start_governor(policy);
2283 	}
2284 
2285 	return ret;
2286 }
2287 
2288 /**
2289  *	cpufreq_update_policy - re-evaluate an existing cpufreq policy
2290  *	@cpu: CPU which shall be re-evaluated
2291  *
2292  *	Useful for policy notifiers which have different necessities
2293  *	at different times.
2294  */
2295 void cpufreq_update_policy(unsigned int cpu)
2296 {
2297 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2298 	struct cpufreq_policy new_policy;
2299 
2300 	if (!policy)
2301 		return;
2302 
2303 	down_write(&policy->rwsem);
2304 
2305 	if (policy_is_inactive(policy))
2306 		goto unlock;
2307 
2308 	pr_debug("updating policy for CPU %u\n", cpu);
2309 	memcpy(&new_policy, policy, sizeof(*policy));
2310 	new_policy.min = policy->user_policy.min;
2311 	new_policy.max = policy->user_policy.max;
2312 
2313 	/*
2314 	 * BIOS might change freq behind our back
2315 	 * -> ask driver for current freq and notify governors about a change
2316 	 */
2317 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2318 		if (cpufreq_suspended)
2319 			goto unlock;
2320 
2321 		new_policy.cur = cpufreq_update_current_freq(policy);
2322 		if (WARN_ON(!new_policy.cur))
2323 			goto unlock;
2324 	}
2325 
2326 	cpufreq_set_policy(policy, &new_policy);
2327 
2328 unlock:
2329 	up_write(&policy->rwsem);
2330 
2331 	cpufreq_cpu_put(policy);
2332 }
2333 EXPORT_SYMBOL(cpufreq_update_policy);
2334 
2335 /*********************************************************************
2336  *               BOOST						     *
2337  *********************************************************************/
2338 static int cpufreq_boost_set_sw(int state)
2339 {
2340 	struct cpufreq_policy *policy;
2341 	int ret = -EINVAL;
2342 
2343 	for_each_active_policy(policy) {
2344 		if (!policy->freq_table)
2345 			continue;
2346 
2347 		ret = cpufreq_frequency_table_cpuinfo(policy,
2348 						      policy->freq_table);
2349 		if (ret) {
2350 			pr_err("%s: Policy frequency update failed\n",
2351 			       __func__);
2352 			break;
2353 		}
2354 
2355 		down_write(&policy->rwsem);
2356 		policy->user_policy.max = policy->max;
2357 		cpufreq_governor_limits(policy);
2358 		up_write(&policy->rwsem);
2359 	}
2360 
2361 	return ret;
2362 }
2363 
2364 int cpufreq_boost_trigger_state(int state)
2365 {
2366 	unsigned long flags;
2367 	int ret = 0;
2368 
2369 	if (cpufreq_driver->boost_enabled == state)
2370 		return 0;
2371 
2372 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2373 	cpufreq_driver->boost_enabled = state;
2374 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2375 
2376 	ret = cpufreq_driver->set_boost(state);
2377 	if (ret) {
2378 		write_lock_irqsave(&cpufreq_driver_lock, flags);
2379 		cpufreq_driver->boost_enabled = !state;
2380 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2381 
2382 		pr_err("%s: Cannot %s BOOST\n",
2383 		       __func__, state ? "enable" : "disable");
2384 	}
2385 
2386 	return ret;
2387 }
2388 
2389 static bool cpufreq_boost_supported(void)
2390 {
2391 	return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2392 }
2393 
2394 static int create_boost_sysfs_file(void)
2395 {
2396 	int ret;
2397 
2398 	ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2399 	if (ret)
2400 		pr_err("%s: cannot register global BOOST sysfs file\n",
2401 		       __func__);
2402 
2403 	return ret;
2404 }
2405 
2406 static void remove_boost_sysfs_file(void)
2407 {
2408 	if (cpufreq_boost_supported())
2409 		sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2410 }
2411 
2412 int cpufreq_enable_boost_support(void)
2413 {
2414 	if (!cpufreq_driver)
2415 		return -EINVAL;
2416 
2417 	if (cpufreq_boost_supported())
2418 		return 0;
2419 
2420 	cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2421 
2422 	/* This will get removed on driver unregister */
2423 	return create_boost_sysfs_file();
2424 }
2425 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2426 
2427 int cpufreq_boost_enabled(void)
2428 {
2429 	return cpufreq_driver->boost_enabled;
2430 }
2431 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2432 
2433 /*********************************************************************
2434  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2435  *********************************************************************/
2436 static enum cpuhp_state hp_online;
2437 
2438 static int cpuhp_cpufreq_online(unsigned int cpu)
2439 {
2440 	cpufreq_online(cpu);
2441 
2442 	return 0;
2443 }
2444 
2445 static int cpuhp_cpufreq_offline(unsigned int cpu)
2446 {
2447 	cpufreq_offline(cpu);
2448 
2449 	return 0;
2450 }
2451 
2452 /**
2453  * cpufreq_register_driver - register a CPU Frequency driver
2454  * @driver_data: A struct cpufreq_driver containing the values#
2455  * submitted by the CPU Frequency driver.
2456  *
2457  * Registers a CPU Frequency driver to this core code. This code
2458  * returns zero on success, -EEXIST when another driver got here first
2459  * (and isn't unregistered in the meantime).
2460  *
2461  */
2462 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2463 {
2464 	unsigned long flags;
2465 	int ret;
2466 
2467 	if (cpufreq_disabled())
2468 		return -ENODEV;
2469 
2470 	if (!driver_data || !driver_data->verify || !driver_data->init ||
2471 	    !(driver_data->setpolicy || driver_data->target_index ||
2472 		    driver_data->target) ||
2473 	     (driver_data->setpolicy && (driver_data->target_index ||
2474 		    driver_data->target)) ||
2475 	     (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2476 		return -EINVAL;
2477 
2478 	pr_debug("trying to register driver %s\n", driver_data->name);
2479 
2480 	/* Protect against concurrent CPU online/offline. */
2481 	cpus_read_lock();
2482 
2483 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2484 	if (cpufreq_driver) {
2485 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2486 		ret = -EEXIST;
2487 		goto out;
2488 	}
2489 	cpufreq_driver = driver_data;
2490 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2491 
2492 	if (driver_data->setpolicy)
2493 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2494 
2495 	if (cpufreq_boost_supported()) {
2496 		ret = create_boost_sysfs_file();
2497 		if (ret)
2498 			goto err_null_driver;
2499 	}
2500 
2501 	ret = subsys_interface_register(&cpufreq_interface);
2502 	if (ret)
2503 		goto err_boost_unreg;
2504 
2505 	if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2506 	    list_empty(&cpufreq_policy_list)) {
2507 		/* if all ->init() calls failed, unregister */
2508 		ret = -ENODEV;
2509 		pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2510 			 driver_data->name);
2511 		goto err_if_unreg;
2512 	}
2513 
2514 	ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2515 						   "cpufreq:online",
2516 						   cpuhp_cpufreq_online,
2517 						   cpuhp_cpufreq_offline);
2518 	if (ret < 0)
2519 		goto err_if_unreg;
2520 	hp_online = ret;
2521 	ret = 0;
2522 
2523 	pr_debug("driver %s up and running\n", driver_data->name);
2524 	goto out;
2525 
2526 err_if_unreg:
2527 	subsys_interface_unregister(&cpufreq_interface);
2528 err_boost_unreg:
2529 	remove_boost_sysfs_file();
2530 err_null_driver:
2531 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2532 	cpufreq_driver = NULL;
2533 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2534 out:
2535 	cpus_read_unlock();
2536 	return ret;
2537 }
2538 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2539 
2540 /**
2541  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2542  *
2543  * Unregister the current CPUFreq driver. Only call this if you have
2544  * the right to do so, i.e. if you have succeeded in initialising before!
2545  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2546  * currently not initialised.
2547  */
2548 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2549 {
2550 	unsigned long flags;
2551 
2552 	if (!cpufreq_driver || (driver != cpufreq_driver))
2553 		return -EINVAL;
2554 
2555 	pr_debug("unregistering driver %s\n", driver->name);
2556 
2557 	/* Protect against concurrent cpu hotplug */
2558 	cpus_read_lock();
2559 	subsys_interface_unregister(&cpufreq_interface);
2560 	remove_boost_sysfs_file();
2561 	cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2562 
2563 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2564 
2565 	cpufreq_driver = NULL;
2566 
2567 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2568 	cpus_read_unlock();
2569 
2570 	return 0;
2571 }
2572 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2573 
2574 /*
2575  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2576  * or mutexes when secondary CPUs are halted.
2577  */
2578 static struct syscore_ops cpufreq_syscore_ops = {
2579 	.shutdown = cpufreq_suspend,
2580 };
2581 
2582 struct kobject *cpufreq_global_kobject;
2583 EXPORT_SYMBOL(cpufreq_global_kobject);
2584 
2585 static int __init cpufreq_core_init(void)
2586 {
2587 	if (cpufreq_disabled())
2588 		return -ENODEV;
2589 
2590 	cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2591 	BUG_ON(!cpufreq_global_kobject);
2592 
2593 	register_syscore_ops(&cpufreq_syscore_ops);
2594 
2595 	return 0;
2596 }
2597 module_param(off, int, 0444);
2598 core_initcall(cpufreq_core_init);
2599