1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 *  linux/drivers/cpufreq/cpufreq.c
4 *
5 *  Copyright (C) 2001 Russell King
6 *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
7 *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
8 *
9 *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
10 *	Added handling for CPU hotplug
11 *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
12 *	Fix handling for CPU hotplug -- affected CPUs
13 */
14
15#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16
17#include <linux/cpu.h>
18#include <linux/cpufreq.h>
19#include <linux/cpu_cooling.h>
20#include <linux/delay.h>
21#include <linux/device.h>
22#include <linux/init.h>
23#include <linux/kernel_stat.h>
24#include <linux/module.h>
25#include <linux/mutex.h>
26#include <linux/pm_qos.h>
27#include <linux/slab.h>
28#include <linux/suspend.h>
29#include <linux/syscore_ops.h>
30#include <linux/tick.h>
31#include <trace/events/power.h>
32
33static LIST_HEAD(cpufreq_policy_list);
34
35/* Macros to iterate over CPU policies */
36#define for_each_suitable_policy(__policy, __active)			 \
37	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
38		if ((__active) == !policy_is_inactive(__policy))
39
40#define for_each_active_policy(__policy)		\
41	for_each_suitable_policy(__policy, true)
42#define for_each_inactive_policy(__policy)		\
43	for_each_suitable_policy(__policy, false)
44
45#define for_each_policy(__policy)			\
46	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
47
48/* Iterate over governors */
49static LIST_HEAD(cpufreq_governor_list);
50#define for_each_governor(__governor)				\
51	list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
52
53static char default_governor[CPUFREQ_NAME_LEN];
54
55/*
56 * The "cpufreq driver" - the arch- or hardware-dependent low
57 * level driver of CPUFreq support, and its spinlock. This lock
58 * also protects the cpufreq_cpu_data array.
59 */
60static struct cpufreq_driver *cpufreq_driver;
61static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
62static DEFINE_RWLOCK(cpufreq_driver_lock);
63
64static DEFINE_STATIC_KEY_FALSE(cpufreq_freq_invariance);
65bool cpufreq_supports_freq_invariance(void)
66{
67	return static_branch_likely(&cpufreq_freq_invariance);
68}
69
70/* Flag to suspend/resume CPUFreq governors */
71static bool cpufreq_suspended;
72
73static inline bool has_target(void)
74{
75	return cpufreq_driver->target_index || cpufreq_driver->target;
76}
77
78/* internal prototypes */
79static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
80static int cpufreq_init_governor(struct cpufreq_policy *policy);
81static void cpufreq_exit_governor(struct cpufreq_policy *policy);
82static void cpufreq_governor_limits(struct cpufreq_policy *policy);
83static int cpufreq_set_policy(struct cpufreq_policy *policy,
84			      struct cpufreq_governor *new_gov,
85			      unsigned int new_pol);
86
87/*
88 * Two notifier lists: the "policy" list is involved in the
89 * validation process for a new CPU frequency policy; the
90 * "transition" list for kernel code that needs to handle
91 * changes to devices when the CPU clock speed changes.
92 * The mutex locks both lists.
93 */
94static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
95SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
96
97static int off __read_mostly;
98static int cpufreq_disabled(void)
99{
100	return off;
101}
102void disable_cpufreq(void)
103{
104	off = 1;
105}
106static DEFINE_MUTEX(cpufreq_governor_mutex);
107
108bool have_governor_per_policy(void)
109{
110	return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
111}
112EXPORT_SYMBOL_GPL(have_governor_per_policy);
113
114static struct kobject *cpufreq_global_kobject;
115
116struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
117{
118	if (have_governor_per_policy())
119		return &policy->kobj;
120	else
121		return cpufreq_global_kobject;
122}
123EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
124
125static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
126{
127	struct kernel_cpustat kcpustat;
128	u64 cur_wall_time;
129	u64 idle_time;
130	u64 busy_time;
131
132	cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
133
134	kcpustat_cpu_fetch(&kcpustat, cpu);
135
136	busy_time = kcpustat.cpustat[CPUTIME_USER];
137	busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
138	busy_time += kcpustat.cpustat[CPUTIME_IRQ];
139	busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
140	busy_time += kcpustat.cpustat[CPUTIME_STEAL];
141	busy_time += kcpustat.cpustat[CPUTIME_NICE];
142
143	idle_time = cur_wall_time - busy_time;
144	if (wall)
145		*wall = div_u64(cur_wall_time, NSEC_PER_USEC);
146
147	return div_u64(idle_time, NSEC_PER_USEC);
148}
149
150u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
151{
152	u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
153
154	if (idle_time == -1ULL)
155		return get_cpu_idle_time_jiffy(cpu, wall);
156	else if (!io_busy)
157		idle_time += get_cpu_iowait_time_us(cpu, wall);
158
159	return idle_time;
160}
161EXPORT_SYMBOL_GPL(get_cpu_idle_time);
162
163/*
164 * This is a generic cpufreq init() routine which can be used by cpufreq
165 * drivers of SMP systems. It will do following:
166 * - validate & show freq table passed
167 * - set policies transition latency
168 * - policy->cpus with all possible CPUs
169 */
170void cpufreq_generic_init(struct cpufreq_policy *policy,
171		struct cpufreq_frequency_table *table,
172		unsigned int transition_latency)
173{
174	policy->freq_table = table;
175	policy->cpuinfo.transition_latency = transition_latency;
176
177	/*
178	 * The driver only supports the SMP configuration where all processors
179	 * share the clock and voltage and clock.
180	 */
181	cpumask_setall(policy->cpus);
182}
183EXPORT_SYMBOL_GPL(cpufreq_generic_init);
184
185struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
186{
187	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
188
189	return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
190}
191EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
192
193unsigned int cpufreq_generic_get(unsigned int cpu)
194{
195	struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
196
197	if (!policy || IS_ERR(policy->clk)) {
198		pr_err("%s: No %s associated to cpu: %d\n",
199		       __func__, policy ? "clk" : "policy", cpu);
200		return 0;
201	}
202
203	return clk_get_rate(policy->clk) / 1000;
204}
205EXPORT_SYMBOL_GPL(cpufreq_generic_get);
206
207/**
208 * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
209 * @cpu: CPU to find the policy for.
210 *
211 * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
212 * the kobject reference counter of that policy.  Return a valid policy on
213 * success or NULL on failure.
214 *
215 * The policy returned by this function has to be released with the help of
216 * cpufreq_cpu_put() to balance its kobject reference counter properly.
217 */
218struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
219{
220	struct cpufreq_policy *policy = NULL;
221	unsigned long flags;
222
223	if (WARN_ON(cpu >= nr_cpu_ids))
224		return NULL;
225
226	/* get the cpufreq driver */
227	read_lock_irqsave(&cpufreq_driver_lock, flags);
228
229	if (cpufreq_driver) {
230		/* get the CPU */
231		policy = cpufreq_cpu_get_raw(cpu);
232		if (policy)
233			kobject_get(&policy->kobj);
234	}
235
236	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
237
238	return policy;
239}
240EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
241
242/**
243 * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
244 * @policy: cpufreq policy returned by cpufreq_cpu_get().
245 */
246void cpufreq_cpu_put(struct cpufreq_policy *policy)
247{
248	kobject_put(&policy->kobj);
249}
250EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
251
252/**
253 * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
254 * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
255 */
256void cpufreq_cpu_release(struct cpufreq_policy *policy)
257{
258	if (WARN_ON(!policy))
259		return;
260
261	lockdep_assert_held(&policy->rwsem);
262
263	up_write(&policy->rwsem);
264
265	cpufreq_cpu_put(policy);
266}
267
268/**
269 * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
270 * @cpu: CPU to find the policy for.
271 *
272 * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
273 * if the policy returned by it is not NULL, acquire its rwsem for writing.
274 * Return the policy if it is active or release it and return NULL otherwise.
275 *
276 * The policy returned by this function has to be released with the help of
277 * cpufreq_cpu_release() in order to release its rwsem and balance its usage
278 * counter properly.
279 */
280struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
281{
282	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
283
284	if (!policy)
285		return NULL;
286
287	down_write(&policy->rwsem);
288
289	if (policy_is_inactive(policy)) {
290		cpufreq_cpu_release(policy);
291		return NULL;
292	}
293
294	return policy;
295}
296
297/*********************************************************************
298 *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
299 *********************************************************************/
300
301/*
302 * adjust_jiffies - adjust the system "loops_per_jiffy"
303 *
304 * This function alters the system "loops_per_jiffy" for the clock
305 * speed change. Note that loops_per_jiffy cannot be updated on SMP
306 * systems as each CPU might be scaled differently. So, use the arch
307 * per-CPU loops_per_jiffy value wherever possible.
308 */
309static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
310{
311#ifndef CONFIG_SMP
312	static unsigned long l_p_j_ref;
313	static unsigned int l_p_j_ref_freq;
314
315	if (ci->flags & CPUFREQ_CONST_LOOPS)
316		return;
317
318	if (!l_p_j_ref_freq) {
319		l_p_j_ref = loops_per_jiffy;
320		l_p_j_ref_freq = ci->old;
321		pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
322			 l_p_j_ref, l_p_j_ref_freq);
323	}
324	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
325		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
326								ci->new);
327		pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
328			 loops_per_jiffy, ci->new);
329	}
330#endif
331}
332
333/**
334 * cpufreq_notify_transition - Notify frequency transition and adjust_jiffies.
335 * @policy: cpufreq policy to enable fast frequency switching for.
336 * @freqs: contain details of the frequency update.
337 * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
338 *
339 * This function calls the transition notifiers and the "adjust_jiffies"
340 * function. It is called twice on all CPU frequency changes that have
341 * external effects.
342 */
343static void cpufreq_notify_transition(struct cpufreq_policy *policy,
344				      struct cpufreq_freqs *freqs,
345				      unsigned int state)
346{
347	int cpu;
348
349	BUG_ON(irqs_disabled());
350
351	if (cpufreq_disabled())
352		return;
353
354	freqs->policy = policy;
355	freqs->flags = cpufreq_driver->flags;
356	pr_debug("notification %u of frequency transition to %u kHz\n",
357		 state, freqs->new);
358
359	switch (state) {
360	case CPUFREQ_PRECHANGE:
361		/*
362		 * Detect if the driver reported a value as "old frequency"
363		 * which is not equal to what the cpufreq core thinks is
364		 * "old frequency".
365		 */
366		if (policy->cur && policy->cur != freqs->old) {
367			pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
368				 freqs->old, policy->cur);
369			freqs->old = policy->cur;
370		}
371
372		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
373					 CPUFREQ_PRECHANGE, freqs);
374
375		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
376		break;
377
378	case CPUFREQ_POSTCHANGE:
379		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
380		pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
381			 cpumask_pr_args(policy->cpus));
382
383		for_each_cpu(cpu, policy->cpus)
384			trace_cpu_frequency(freqs->new, cpu);
385
386		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
387					 CPUFREQ_POSTCHANGE, freqs);
388
389		cpufreq_stats_record_transition(policy, freqs->new);
390		policy->cur = freqs->new;
391	}
392}
393
394/* Do post notifications when there are chances that transition has failed */
395static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
396		struct cpufreq_freqs *freqs, int transition_failed)
397{
398	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
399	if (!transition_failed)
400		return;
401
402	swap(freqs->old, freqs->new);
403	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
404	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
405}
406
407void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
408		struct cpufreq_freqs *freqs)
409{
410
411	/*
412	 * Catch double invocations of _begin() which lead to self-deadlock.
413	 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
414	 * doesn't invoke _begin() on their behalf, and hence the chances of
415	 * double invocations are very low. Moreover, there are scenarios
416	 * where these checks can emit false-positive warnings in these
417	 * drivers; so we avoid that by skipping them altogether.
418	 */
419	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
420				&& current == policy->transition_task);
421
422wait:
423	wait_event(policy->transition_wait, !policy->transition_ongoing);
424
425	spin_lock(&policy->transition_lock);
426
427	if (unlikely(policy->transition_ongoing)) {
428		spin_unlock(&policy->transition_lock);
429		goto wait;
430	}
431
432	policy->transition_ongoing = true;
433	policy->transition_task = current;
434
435	spin_unlock(&policy->transition_lock);
436
437	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
438}
439EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
440
441void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
442		struct cpufreq_freqs *freqs, int transition_failed)
443{
444	if (WARN_ON(!policy->transition_ongoing))
445		return;
446
447	cpufreq_notify_post_transition(policy, freqs, transition_failed);
448
449	arch_set_freq_scale(policy->related_cpus,
450			    policy->cur,
451			    policy->cpuinfo.max_freq);
452
453	spin_lock(&policy->transition_lock);
454	policy->transition_ongoing = false;
455	policy->transition_task = NULL;
456	spin_unlock(&policy->transition_lock);
457
458	wake_up(&policy->transition_wait);
459}
460EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
461
462/*
463 * Fast frequency switching status count.  Positive means "enabled", negative
464 * means "disabled" and 0 means "not decided yet".
465 */
466static int cpufreq_fast_switch_count;
467static DEFINE_MUTEX(cpufreq_fast_switch_lock);
468
469static void cpufreq_list_transition_notifiers(void)
470{
471	struct notifier_block *nb;
472
473	pr_info("Registered transition notifiers:\n");
474
475	mutex_lock(&cpufreq_transition_notifier_list.mutex);
476
477	for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
478		pr_info("%pS\n", nb->notifier_call);
479
480	mutex_unlock(&cpufreq_transition_notifier_list.mutex);
481}
482
483/**
484 * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
485 * @policy: cpufreq policy to enable fast frequency switching for.
486 *
487 * Try to enable fast frequency switching for @policy.
488 *
489 * The attempt will fail if there is at least one transition notifier registered
490 * at this point, as fast frequency switching is quite fundamentally at odds
491 * with transition notifiers.  Thus if successful, it will make registration of
492 * transition notifiers fail going forward.
493 */
494void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
495{
496	lockdep_assert_held(&policy->rwsem);
497
498	if (!policy->fast_switch_possible)
499		return;
500
501	mutex_lock(&cpufreq_fast_switch_lock);
502	if (cpufreq_fast_switch_count >= 0) {
503		cpufreq_fast_switch_count++;
504		policy->fast_switch_enabled = true;
505	} else {
506		pr_warn("CPU%u: Fast frequency switching not enabled\n",
507			policy->cpu);
508		cpufreq_list_transition_notifiers();
509	}
510	mutex_unlock(&cpufreq_fast_switch_lock);
511}
512EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
513
514/**
515 * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
516 * @policy: cpufreq policy to disable fast frequency switching for.
517 */
518void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
519{
520	mutex_lock(&cpufreq_fast_switch_lock);
521	if (policy->fast_switch_enabled) {
522		policy->fast_switch_enabled = false;
523		if (!WARN_ON(cpufreq_fast_switch_count <= 0))
524			cpufreq_fast_switch_count--;
525	}
526	mutex_unlock(&cpufreq_fast_switch_lock);
527}
528EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
529
530/**
531 * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
532 * one.
533 * @policy: associated policy to interrogate
534 * @target_freq: target frequency to resolve.
535 *
536 * The target to driver frequency mapping is cached in the policy.
537 *
538 * Return: Lowest driver-supported frequency greater than or equal to the
539 * given target_freq, subject to policy (min/max) and driver limitations.
540 */
541unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
542					 unsigned int target_freq)
543{
544	target_freq = clamp_val(target_freq, policy->min, policy->max);
545	policy->cached_target_freq = target_freq;
546
547	if (cpufreq_driver->target_index) {
548		unsigned int idx;
549
550		idx = cpufreq_frequency_table_target(policy, target_freq,
551						     CPUFREQ_RELATION_L);
552		policy->cached_resolved_idx = idx;
553		return policy->freq_table[idx].frequency;
554	}
555
556	if (cpufreq_driver->resolve_freq)
557		return cpufreq_driver->resolve_freq(policy, target_freq);
558
559	return target_freq;
560}
561EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
562
563unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
564{
565	unsigned int latency;
566
567	if (policy->transition_delay_us)
568		return policy->transition_delay_us;
569
570	latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
571	if (latency) {
572		/*
573		 * For platforms that can change the frequency very fast (< 10
574		 * us), the above formula gives a decent transition delay. But
575		 * for platforms where transition_latency is in milliseconds, it
576		 * ends up giving unrealistic values.
577		 *
578		 * Cap the default transition delay to 10 ms, which seems to be
579		 * a reasonable amount of time after which we should reevaluate
580		 * the frequency.
581		 */
582		return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
583	}
584
585	return LATENCY_MULTIPLIER;
586}
587EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
588
589/*********************************************************************
590 *                          SYSFS INTERFACE                          *
591 *********************************************************************/
592static ssize_t show_boost(struct kobject *kobj,
593			  struct kobj_attribute *attr, char *buf)
594{
595	return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
596}
597
598static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
599			   const char *buf, size_t count)
600{
601	int ret, enable;
602
603	ret = sscanf(buf, "%d", &enable);
604	if (ret != 1 || enable < 0 || enable > 1)
605		return -EINVAL;
606
607	if (cpufreq_boost_trigger_state(enable)) {
608		pr_err("%s: Cannot %s BOOST!\n",
609		       __func__, enable ? "enable" : "disable");
610		return -EINVAL;
611	}
612
613	pr_debug("%s: cpufreq BOOST %s\n",
614		 __func__, enable ? "enabled" : "disabled");
615
616	return count;
617}
618define_one_global_rw(boost);
619
620static struct cpufreq_governor *find_governor(const char *str_governor)
621{
622	struct cpufreq_governor *t;
623
624	for_each_governor(t)
625		if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
626			return t;
627
628	return NULL;
629}
630
631static struct cpufreq_governor *get_governor(const char *str_governor)
632{
633	struct cpufreq_governor *t;
634
635	mutex_lock(&cpufreq_governor_mutex);
636	t = find_governor(str_governor);
637	if (!t)
638		goto unlock;
639
640	if (!try_module_get(t->owner))
641		t = NULL;
642
643unlock:
644	mutex_unlock(&cpufreq_governor_mutex);
645
646	return t;
647}
648
649static unsigned int cpufreq_parse_policy(char *str_governor)
650{
651	if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
652		return CPUFREQ_POLICY_PERFORMANCE;
653
654	if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
655		return CPUFREQ_POLICY_POWERSAVE;
656
657	return CPUFREQ_POLICY_UNKNOWN;
658}
659
660/**
661 * cpufreq_parse_governor - parse a governor string only for has_target()
662 * @str_governor: Governor name.
663 */
664static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
665{
666	struct cpufreq_governor *t;
667
668	t = get_governor(str_governor);
669	if (t)
670		return t;
671
672	if (request_module("cpufreq_%s", str_governor))
673		return NULL;
674
675	return get_governor(str_governor);
676}
677
678/*
679 * cpufreq_per_cpu_attr_read() / show_##file_name() -
680 * print out cpufreq information
681 *
682 * Write out information from cpufreq_driver->policy[cpu]; object must be
683 * "unsigned int".
684 */
685
686#define show_one(file_name, object)			\
687static ssize_t show_##file_name				\
688(struct cpufreq_policy *policy, char *buf)		\
689{							\
690	return sprintf(buf, "%u\n", policy->object);	\
691}
692
693show_one(cpuinfo_min_freq, cpuinfo.min_freq);
694show_one(cpuinfo_max_freq, cpuinfo.max_freq);
695show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
696show_one(scaling_min_freq, min);
697show_one(scaling_max_freq, max);
698
699__weak unsigned int arch_freq_get_on_cpu(int cpu)
700{
701	return 0;
702}
703
704static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
705{
706	ssize_t ret;
707	unsigned int freq;
708
709	freq = arch_freq_get_on_cpu(policy->cpu);
710	if (freq)
711		ret = sprintf(buf, "%u\n", freq);
712	else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
713		ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
714	else
715		ret = sprintf(buf, "%u\n", policy->cur);
716	return ret;
717}
718
719/*
720 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
721 */
722#define store_one(file_name, object)			\
723static ssize_t store_##file_name					\
724(struct cpufreq_policy *policy, const char *buf, size_t count)		\
725{									\
726	unsigned long val;						\
727	int ret;							\
728									\
729	ret = sscanf(buf, "%lu", &val);					\
730	if (ret != 1)							\
731		return -EINVAL;						\
732									\
733	ret = freq_qos_update_request(policy->object##_freq_req, val);\
734	return ret >= 0 ? count : ret;					\
735}
736
737store_one(scaling_min_freq, min);
738store_one(scaling_max_freq, max);
739
740/*
741 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
742 */
743static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
744					char *buf)
745{
746	unsigned int cur_freq = __cpufreq_get(policy);
747
748	if (cur_freq)
749		return sprintf(buf, "%u\n", cur_freq);
750
751	return sprintf(buf, "<unknown>\n");
752}
753
754/*
755 * show_scaling_governor - show the current policy for the specified CPU
756 */
757static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
758{
759	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
760		return sprintf(buf, "powersave\n");
761	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
762		return sprintf(buf, "performance\n");
763	else if (policy->governor)
764		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
765				policy->governor->name);
766	return -EINVAL;
767}
768
769/*
770 * store_scaling_governor - store policy for the specified CPU
771 */
772static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
773					const char *buf, size_t count)
774{
775	char str_governor[16];
776	int ret;
777
778	ret = sscanf(buf, "%15s", str_governor);
779	if (ret != 1)
780		return -EINVAL;
781
782	if (cpufreq_driver->setpolicy) {
783		unsigned int new_pol;
784
785		new_pol = cpufreq_parse_policy(str_governor);
786		if (!new_pol)
787			return -EINVAL;
788
789		ret = cpufreq_set_policy(policy, NULL, new_pol);
790	} else {
791		struct cpufreq_governor *new_gov;
792
793		new_gov = cpufreq_parse_governor(str_governor);
794		if (!new_gov)
795			return -EINVAL;
796
797		ret = cpufreq_set_policy(policy, new_gov,
798					 CPUFREQ_POLICY_UNKNOWN);
799
800		module_put(new_gov->owner);
801	}
802
803	return ret ? ret : count;
804}
805
806/*
807 * show_scaling_driver - show the cpufreq driver currently loaded
808 */
809static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
810{
811	return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
812}
813
814/*
815 * show_scaling_available_governors - show the available CPUfreq governors
816 */
817static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
818						char *buf)
819{
820	ssize_t i = 0;
821	struct cpufreq_governor *t;
822
823	if (!has_target()) {
824		i += sprintf(buf, "performance powersave");
825		goto out;
826	}
827
828	mutex_lock(&cpufreq_governor_mutex);
829	for_each_governor(t) {
830		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
831		    - (CPUFREQ_NAME_LEN + 2)))
832			break;
833		i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
834	}
835	mutex_unlock(&cpufreq_governor_mutex);
836out:
837	i += sprintf(&buf[i], "\n");
838	return i;
839}
840
841ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
842{
843	ssize_t i = 0;
844	unsigned int cpu;
845
846	for_each_cpu(cpu, mask) {
847		if (i)
848			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
849		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
850		if (i >= (PAGE_SIZE - 5))
851			break;
852	}
853	i += sprintf(&buf[i], "\n");
854	return i;
855}
856EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
857
858/*
859 * show_related_cpus - show the CPUs affected by each transition even if
860 * hw coordination is in use
861 */
862static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
863{
864	return cpufreq_show_cpus(policy->related_cpus, buf);
865}
866
867/*
868 * show_affected_cpus - show the CPUs affected by each transition
869 */
870static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
871{
872	return cpufreq_show_cpus(policy->cpus, buf);
873}
874
875static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
876					const char *buf, size_t count)
877{
878	unsigned int freq = 0;
879	unsigned int ret;
880
881	if (!policy->governor || !policy->governor->store_setspeed)
882		return -EINVAL;
883
884	ret = sscanf(buf, "%u", &freq);
885	if (ret != 1)
886		return -EINVAL;
887
888	policy->governor->store_setspeed(policy, freq);
889
890	return count;
891}
892
893static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
894{
895	if (!policy->governor || !policy->governor->show_setspeed)
896		return sprintf(buf, "<unsupported>\n");
897
898	return policy->governor->show_setspeed(policy, buf);
899}
900
901/*
902 * show_bios_limit - show the current cpufreq HW/BIOS limitation
903 */
904static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
905{
906	unsigned int limit;
907	int ret;
908	ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
909	if (!ret)
910		return sprintf(buf, "%u\n", limit);
911	return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
912}
913
914cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
915cpufreq_freq_attr_ro(cpuinfo_min_freq);
916cpufreq_freq_attr_ro(cpuinfo_max_freq);
917cpufreq_freq_attr_ro(cpuinfo_transition_latency);
918cpufreq_freq_attr_ro(scaling_available_governors);
919cpufreq_freq_attr_ro(scaling_driver);
920cpufreq_freq_attr_ro(scaling_cur_freq);
921cpufreq_freq_attr_ro(bios_limit);
922cpufreq_freq_attr_ro(related_cpus);
923cpufreq_freq_attr_ro(affected_cpus);
924cpufreq_freq_attr_rw(scaling_min_freq);
925cpufreq_freq_attr_rw(scaling_max_freq);
926cpufreq_freq_attr_rw(scaling_governor);
927cpufreq_freq_attr_rw(scaling_setspeed);
928
929static struct attribute *default_attrs[] = {
930	&cpuinfo_min_freq.attr,
931	&cpuinfo_max_freq.attr,
932	&cpuinfo_transition_latency.attr,
933	&scaling_min_freq.attr,
934	&scaling_max_freq.attr,
935	&affected_cpus.attr,
936	&related_cpus.attr,
937	&scaling_governor.attr,
938	&scaling_driver.attr,
939	&scaling_available_governors.attr,
940	&scaling_setspeed.attr,
941	NULL
942};
943
944#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
945#define to_attr(a) container_of(a, struct freq_attr, attr)
946
947static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
948{
949	struct cpufreq_policy *policy = to_policy(kobj);
950	struct freq_attr *fattr = to_attr(attr);
951	ssize_t ret;
952
953	if (!fattr->show)
954		return -EIO;
955
956	down_read(&policy->rwsem);
957	ret = fattr->show(policy, buf);
958	up_read(&policy->rwsem);
959
960	return ret;
961}
962
963static ssize_t store(struct kobject *kobj, struct attribute *attr,
964		     const char *buf, size_t count)
965{
966	struct cpufreq_policy *policy = to_policy(kobj);
967	struct freq_attr *fattr = to_attr(attr);
968	ssize_t ret = -EINVAL;
969
970	if (!fattr->store)
971		return -EIO;
972
973	/*
974	 * cpus_read_trylock() is used here to work around a circular lock
975	 * dependency problem with respect to the cpufreq_register_driver().
976	 */
977	if (!cpus_read_trylock())
978		return -EBUSY;
979
980	if (cpu_online(policy->cpu)) {
981		down_write(&policy->rwsem);
982		ret = fattr->store(policy, buf, count);
983		up_write(&policy->rwsem);
984	}
985
986	cpus_read_unlock();
987
988	return ret;
989}
990
991static void cpufreq_sysfs_release(struct kobject *kobj)
992{
993	struct cpufreq_policy *policy = to_policy(kobj);
994	pr_debug("last reference is dropped\n");
995	complete(&policy->kobj_unregister);
996}
997
998static const struct sysfs_ops sysfs_ops = {
999	.show	= show,
1000	.store	= store,
1001};
1002
1003static struct kobj_type ktype_cpufreq = {
1004	.sysfs_ops	= &sysfs_ops,
1005	.default_attrs	= default_attrs,
1006	.release	= cpufreq_sysfs_release,
1007};
1008
1009static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
1010				struct device *dev)
1011{
1012	if (unlikely(!dev))
1013		return;
1014
1015	if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1016		return;
1017
1018	dev_dbg(dev, "%s: Adding symlink\n", __func__);
1019	if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1020		dev_err(dev, "cpufreq symlink creation failed\n");
1021}
1022
1023static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
1024				   struct device *dev)
1025{
1026	dev_dbg(dev, "%s: Removing symlink\n", __func__);
1027	sysfs_remove_link(&dev->kobj, "cpufreq");
1028}
1029
1030static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1031{
1032	struct freq_attr **drv_attr;
1033	int ret = 0;
1034
1035	/* set up files for this cpu device */
1036	drv_attr = cpufreq_driver->attr;
1037	while (drv_attr && *drv_attr) {
1038		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1039		if (ret)
1040			return ret;
1041		drv_attr++;
1042	}
1043	if (cpufreq_driver->get) {
1044		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1045		if (ret)
1046			return ret;
1047	}
1048
1049	ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1050	if (ret)
1051		return ret;
1052
1053	if (cpufreq_driver->bios_limit) {
1054		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1055		if (ret)
1056			return ret;
1057	}
1058
1059	return 0;
1060}
1061
1062static int cpufreq_init_policy(struct cpufreq_policy *policy)
1063{
1064	struct cpufreq_governor *gov = NULL;
1065	unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1066	int ret;
1067
1068	if (has_target()) {
1069		/* Update policy governor to the one used before hotplug. */
1070		gov = get_governor(policy->last_governor);
1071		if (gov) {
1072			pr_debug("Restoring governor %s for cpu %d\n",
1073				 gov->name, policy->cpu);
1074		} else {
1075			gov = get_governor(default_governor);
1076		}
1077
1078		if (!gov) {
1079			gov = cpufreq_default_governor();
1080			__module_get(gov->owner);
1081		}
1082
1083	} else {
1084
1085		/* Use the default policy if there is no last_policy. */
1086		if (policy->last_policy) {
1087			pol = policy->last_policy;
1088		} else {
1089			pol = cpufreq_parse_policy(default_governor);
1090			/*
1091			 * In case the default governor is neither "performance"
1092			 * nor "powersave", fall back to the initial policy
1093			 * value set by the driver.
1094			 */
1095			if (pol == CPUFREQ_POLICY_UNKNOWN)
1096				pol = policy->policy;
1097		}
1098		if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1099		    pol != CPUFREQ_POLICY_POWERSAVE)
1100			return -ENODATA;
1101	}
1102
1103	ret = cpufreq_set_policy(policy, gov, pol);
1104	if (gov)
1105		module_put(gov->owner);
1106
1107	return ret;
1108}
1109
1110static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1111{
1112	int ret = 0;
1113
1114	/* Has this CPU been taken care of already? */
1115	if (cpumask_test_cpu(cpu, policy->cpus))
1116		return 0;
1117
1118	down_write(&policy->rwsem);
1119	if (has_target())
1120		cpufreq_stop_governor(policy);
1121
1122	cpumask_set_cpu(cpu, policy->cpus);
1123
1124	if (has_target()) {
1125		ret = cpufreq_start_governor(policy);
1126		if (ret)
1127			pr_err("%s: Failed to start governor\n", __func__);
1128	}
1129	up_write(&policy->rwsem);
1130	return ret;
1131}
1132
1133void refresh_frequency_limits(struct cpufreq_policy *policy)
1134{
1135	if (!policy_is_inactive(policy)) {
1136		pr_debug("updating policy for CPU %u\n", policy->cpu);
1137
1138		cpufreq_set_policy(policy, policy->governor, policy->policy);
1139	}
1140}
1141EXPORT_SYMBOL(refresh_frequency_limits);
1142
1143static void handle_update(struct work_struct *work)
1144{
1145	struct cpufreq_policy *policy =
1146		container_of(work, struct cpufreq_policy, update);
1147
1148	pr_debug("handle_update for cpu %u called\n", policy->cpu);
1149	down_write(&policy->rwsem);
1150	refresh_frequency_limits(policy);
1151	up_write(&policy->rwsem);
1152}
1153
1154static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1155				void *data)
1156{
1157	struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1158
1159	schedule_work(&policy->update);
1160	return 0;
1161}
1162
1163static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1164				void *data)
1165{
1166	struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1167
1168	schedule_work(&policy->update);
1169	return 0;
1170}
1171
1172static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1173{
1174	struct kobject *kobj;
1175	struct completion *cmp;
1176
1177	down_write(&policy->rwsem);
1178	cpufreq_stats_free_table(policy);
1179	kobj = &policy->kobj;
1180	cmp = &policy->kobj_unregister;
1181	up_write(&policy->rwsem);
1182	kobject_put(kobj);
1183
1184	/*
1185	 * We need to make sure that the underlying kobj is
1186	 * actually not referenced anymore by anybody before we
1187	 * proceed with unloading.
1188	 */
1189	pr_debug("waiting for dropping of refcount\n");
1190	wait_for_completion(cmp);
1191	pr_debug("wait complete\n");
1192}
1193
1194static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1195{
1196	struct cpufreq_policy *policy;
1197	struct device *dev = get_cpu_device(cpu);
1198	int ret;
1199
1200	if (!dev)
1201		return NULL;
1202
1203	policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1204	if (!policy)
1205		return NULL;
1206
1207	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1208		goto err_free_policy;
1209
1210	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1211		goto err_free_cpumask;
1212
1213	if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1214		goto err_free_rcpumask;
1215
1216	init_completion(&policy->kobj_unregister);
1217	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1218				   cpufreq_global_kobject, "policy%u", cpu);
1219	if (ret) {
1220		dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1221		/*
1222		 * The entire policy object will be freed below, but the extra
1223		 * memory allocated for the kobject name needs to be freed by
1224		 * releasing the kobject.
1225		 */
1226		kobject_put(&policy->kobj);
1227		goto err_free_real_cpus;
1228	}
1229
1230	freq_constraints_init(&policy->constraints);
1231
1232	policy->nb_min.notifier_call = cpufreq_notifier_min;
1233	policy->nb_max.notifier_call = cpufreq_notifier_max;
1234
1235	ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1236				    &policy->nb_min);
1237	if (ret) {
1238		dev_err(dev, "Failed to register MIN QoS notifier: %d (%*pbl)\n",
1239			ret, cpumask_pr_args(policy->cpus));
1240		goto err_kobj_remove;
1241	}
1242
1243	ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1244				    &policy->nb_max);
1245	if (ret) {
1246		dev_err(dev, "Failed to register MAX QoS notifier: %d (%*pbl)\n",
1247			ret, cpumask_pr_args(policy->cpus));
1248		goto err_min_qos_notifier;
1249	}
1250
1251	INIT_LIST_HEAD(&policy->policy_list);
1252	init_rwsem(&policy->rwsem);
1253	spin_lock_init(&policy->transition_lock);
1254	init_waitqueue_head(&policy->transition_wait);
1255	INIT_WORK(&policy->update, handle_update);
1256
1257	policy->cpu = cpu;
1258	return policy;
1259
1260err_min_qos_notifier:
1261	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1262				 &policy->nb_min);
1263err_kobj_remove:
1264	cpufreq_policy_put_kobj(policy);
1265err_free_real_cpus:
1266	free_cpumask_var(policy->real_cpus);
1267err_free_rcpumask:
1268	free_cpumask_var(policy->related_cpus);
1269err_free_cpumask:
1270	free_cpumask_var(policy->cpus);
1271err_free_policy:
1272	kfree(policy);
1273
1274	return NULL;
1275}
1276
1277static void cpufreq_policy_free(struct cpufreq_policy *policy)
1278{
1279	unsigned long flags;
1280	int cpu;
1281
1282	/* Remove policy from list */
1283	write_lock_irqsave(&cpufreq_driver_lock, flags);
1284	list_del(&policy->policy_list);
1285
1286	for_each_cpu(cpu, policy->related_cpus)
1287		per_cpu(cpufreq_cpu_data, cpu) = NULL;
1288	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1289
1290	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1291				 &policy->nb_max);
1292	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1293				 &policy->nb_min);
1294
1295	/* Cancel any pending policy->update work before freeing the policy. */
1296	cancel_work_sync(&policy->update);
1297
1298	if (policy->max_freq_req) {
1299		/*
1300		 * CPUFREQ_CREATE_POLICY notification is sent only after
1301		 * successfully adding max_freq_req request.
1302		 */
1303		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1304					     CPUFREQ_REMOVE_POLICY, policy);
1305		freq_qos_remove_request(policy->max_freq_req);
1306	}
1307
1308	freq_qos_remove_request(policy->min_freq_req);
1309	kfree(policy->min_freq_req);
1310
1311	cpufreq_policy_put_kobj(policy);
1312	free_cpumask_var(policy->real_cpus);
1313	free_cpumask_var(policy->related_cpus);
1314	free_cpumask_var(policy->cpus);
1315	kfree(policy);
1316}
1317
1318static int cpufreq_online(unsigned int cpu)
1319{
1320	struct cpufreq_policy *policy;
1321	bool new_policy;
1322	unsigned long flags;
1323	unsigned int j;
1324	int ret;
1325
1326	pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1327
1328	/* Check if this CPU already has a policy to manage it */
1329	policy = per_cpu(cpufreq_cpu_data, cpu);
1330	if (policy) {
1331		WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1332		if (!policy_is_inactive(policy))
1333			return cpufreq_add_policy_cpu(policy, cpu);
1334
1335		/* This is the only online CPU for the policy.  Start over. */
1336		new_policy = false;
1337		down_write(&policy->rwsem);
1338		policy->cpu = cpu;
1339		policy->governor = NULL;
1340		up_write(&policy->rwsem);
1341	} else {
1342		new_policy = true;
1343		policy = cpufreq_policy_alloc(cpu);
1344		if (!policy)
1345			return -ENOMEM;
1346	}
1347
1348	if (!new_policy && cpufreq_driver->online) {
1349		ret = cpufreq_driver->online(policy);
1350		if (ret) {
1351			pr_debug("%s: %d: initialization failed\n", __func__,
1352				 __LINE__);
1353			goto out_exit_policy;
1354		}
1355
1356		/* Recover policy->cpus using related_cpus */
1357		cpumask_copy(policy->cpus, policy->related_cpus);
1358	} else {
1359		cpumask_copy(policy->cpus, cpumask_of(cpu));
1360
1361		/*
1362		 * Call driver. From then on the cpufreq must be able
1363		 * to accept all calls to ->verify and ->setpolicy for this CPU.
1364		 */
1365		ret = cpufreq_driver->init(policy);
1366		if (ret) {
1367			pr_debug("%s: %d: initialization failed\n", __func__,
1368				 __LINE__);
1369			goto out_free_policy;
1370		}
1371
1372		/*
1373		 * The initialization has succeeded and the policy is online.
1374		 * If there is a problem with its frequency table, take it
1375		 * offline and drop it.
1376		 */
1377		ret = cpufreq_table_validate_and_sort(policy);
1378		if (ret)
1379			goto out_offline_policy;
1380
1381		/* related_cpus should at least include policy->cpus. */
1382		cpumask_copy(policy->related_cpus, policy->cpus);
1383	}
1384
1385	down_write(&policy->rwsem);
1386	/*
1387	 * affected cpus must always be the one, which are online. We aren't
1388	 * managing offline cpus here.
1389	 */
1390	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1391
1392	if (new_policy) {
1393		for_each_cpu(j, policy->related_cpus) {
1394			per_cpu(cpufreq_cpu_data, j) = policy;
1395			add_cpu_dev_symlink(policy, j, get_cpu_device(j));
1396		}
1397
1398		policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1399					       GFP_KERNEL);
1400		if (!policy->min_freq_req)
1401			goto out_destroy_policy;
1402
1403		ret = freq_qos_add_request(&policy->constraints,
1404					   policy->min_freq_req, FREQ_QOS_MIN,
1405					   FREQ_QOS_MIN_DEFAULT_VALUE);
1406		if (ret < 0) {
1407			/*
1408			 * So we don't call freq_qos_remove_request() for an
1409			 * uninitialized request.
1410			 */
1411			kfree(policy->min_freq_req);
1412			policy->min_freq_req = NULL;
1413			goto out_destroy_policy;
1414		}
1415
1416		/*
1417		 * This must be initialized right here to avoid calling
1418		 * freq_qos_remove_request() on uninitialized request in case
1419		 * of errors.
1420		 */
1421		policy->max_freq_req = policy->min_freq_req + 1;
1422
1423		ret = freq_qos_add_request(&policy->constraints,
1424					   policy->max_freq_req, FREQ_QOS_MAX,
1425					   FREQ_QOS_MAX_DEFAULT_VALUE);
1426		if (ret < 0) {
1427			policy->max_freq_req = NULL;
1428			goto out_destroy_policy;
1429		}
1430
1431		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1432				CPUFREQ_CREATE_POLICY, policy);
1433	}
1434
1435	if (cpufreq_driver->get && has_target()) {
1436		policy->cur = cpufreq_driver->get(policy->cpu);
1437		if (!policy->cur) {
1438			pr_err("%s: ->get() failed\n", __func__);
1439			goto out_destroy_policy;
1440		}
1441	}
1442
1443	/*
1444	 * Sometimes boot loaders set CPU frequency to a value outside of
1445	 * frequency table present with cpufreq core. In such cases CPU might be
1446	 * unstable if it has to run on that frequency for long duration of time
1447	 * and so its better to set it to a frequency which is specified in
1448	 * freq-table. This also makes cpufreq stats inconsistent as
1449	 * cpufreq-stats would fail to register because current frequency of CPU
1450	 * isn't found in freq-table.
1451	 *
1452	 * Because we don't want this change to effect boot process badly, we go
1453	 * for the next freq which is >= policy->cur ('cur' must be set by now,
1454	 * otherwise we will end up setting freq to lowest of the table as 'cur'
1455	 * is initialized to zero).
1456	 *
1457	 * We are passing target-freq as "policy->cur - 1" otherwise
1458	 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1459	 * equal to target-freq.
1460	 */
1461	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1462	    && has_target()) {
1463		unsigned int old_freq = policy->cur;
1464
1465		/* Are we running at unknown frequency ? */
1466		ret = cpufreq_frequency_table_get_index(policy, old_freq);
1467		if (ret == -EINVAL) {
1468			ret = __cpufreq_driver_target(policy, old_freq - 1,
1469						      CPUFREQ_RELATION_L);
1470
1471			/*
1472			 * Reaching here after boot in a few seconds may not
1473			 * mean that system will remain stable at "unknown"
1474			 * frequency for longer duration. Hence, a BUG_ON().
1475			 */
1476			BUG_ON(ret);
1477			pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
1478				__func__, policy->cpu, old_freq, policy->cur);
1479		}
1480	}
1481
1482	if (new_policy) {
1483		ret = cpufreq_add_dev_interface(policy);
1484		if (ret)
1485			goto out_destroy_policy;
1486
1487		cpufreq_stats_create_table(policy);
1488
1489		write_lock_irqsave(&cpufreq_driver_lock, flags);
1490		list_add(&policy->policy_list, &cpufreq_policy_list);
1491		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1492	}
1493
1494	ret = cpufreq_init_policy(policy);
1495	if (ret) {
1496		pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1497		       __func__, cpu, ret);
1498		goto out_destroy_policy;
1499	}
1500
1501	up_write(&policy->rwsem);
1502
1503	kobject_uevent(&policy->kobj, KOBJ_ADD);
1504
1505	/* Callback for handling stuff after policy is ready */
1506	if (cpufreq_driver->ready)
1507		cpufreq_driver->ready(policy);
1508
1509	if (cpufreq_thermal_control_enabled(cpufreq_driver))
1510		policy->cdev = of_cpufreq_cooling_register(policy);
1511
1512	pr_debug("initialization complete\n");
1513
1514	return 0;
1515
1516out_destroy_policy:
1517	for_each_cpu(j, policy->real_cpus)
1518		remove_cpu_dev_symlink(policy, get_cpu_device(j));
1519
1520	up_write(&policy->rwsem);
1521
1522out_offline_policy:
1523	if (cpufreq_driver->offline)
1524		cpufreq_driver->offline(policy);
1525
1526out_exit_policy:
1527	if (cpufreq_driver->exit)
1528		cpufreq_driver->exit(policy);
1529
1530out_free_policy:
1531	cpufreq_policy_free(policy);
1532	return ret;
1533}
1534
1535/**
1536 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1537 * @dev: CPU device.
1538 * @sif: Subsystem interface structure pointer (not used)
1539 */
1540static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1541{
1542	struct cpufreq_policy *policy;
1543	unsigned cpu = dev->id;
1544	int ret;
1545
1546	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1547
1548	if (cpu_online(cpu)) {
1549		ret = cpufreq_online(cpu);
1550		if (ret)
1551			return ret;
1552	}
1553
1554	/* Create sysfs link on CPU registration */
1555	policy = per_cpu(cpufreq_cpu_data, cpu);
1556	if (policy)
1557		add_cpu_dev_symlink(policy, cpu, dev);
1558
1559	return 0;
1560}
1561
1562static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1563{
1564	int ret;
1565
1566	if (has_target())
1567		cpufreq_stop_governor(policy);
1568
1569	cpumask_clear_cpu(cpu, policy->cpus);
1570
1571	if (!policy_is_inactive(policy)) {
1572		/* Nominate a new CPU if necessary. */
1573		if (cpu == policy->cpu)
1574			policy->cpu = cpumask_any(policy->cpus);
1575
1576		/* Start the governor again for the active policy. */
1577		if (has_target()) {
1578			ret = cpufreq_start_governor(policy);
1579			if (ret)
1580				pr_err("%s: Failed to start governor\n", __func__);
1581		}
1582
1583		return;
1584	}
1585
1586	if (has_target())
1587		strncpy(policy->last_governor, policy->governor->name,
1588			CPUFREQ_NAME_LEN);
1589	else
1590		policy->last_policy = policy->policy;
1591
1592	if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1593		cpufreq_cooling_unregister(policy->cdev);
1594		policy->cdev = NULL;
1595	}
1596
1597	if (cpufreq_driver->stop_cpu)
1598		cpufreq_driver->stop_cpu(policy);
1599
1600	if (has_target())
1601		cpufreq_exit_governor(policy);
1602
1603	/*
1604	 * Perform the ->offline() during light-weight tear-down, as
1605	 * that allows fast recovery when the CPU comes back.
1606	 */
1607	if (cpufreq_driver->offline) {
1608		cpufreq_driver->offline(policy);
1609		return;
1610	}
1611
1612	if (cpufreq_driver->exit)
1613		cpufreq_driver->exit(policy);
1614
1615	policy->freq_table = NULL;
1616}
1617
1618static int cpufreq_offline(unsigned int cpu)
1619{
1620	struct cpufreq_policy *policy;
1621
1622	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1623
1624	policy = cpufreq_cpu_get_raw(cpu);
1625	if (!policy) {
1626		pr_debug("%s: No cpu_data found\n", __func__);
1627		return 0;
1628	}
1629
1630	down_write(&policy->rwsem);
1631
1632	__cpufreq_offline(cpu, policy);
1633
1634	up_write(&policy->rwsem);
1635	return 0;
1636}
1637
1638/*
1639 * cpufreq_remove_dev - remove a CPU device
1640 *
1641 * Removes the cpufreq interface for a CPU device.
1642 */
1643static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1644{
1645	unsigned int cpu = dev->id;
1646	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1647
1648	if (!policy)
1649		return;
1650
1651	down_write(&policy->rwsem);
1652
1653	if (cpu_online(cpu))
1654		__cpufreq_offline(cpu, policy);
1655
1656	cpumask_clear_cpu(cpu, policy->real_cpus);
1657	remove_cpu_dev_symlink(policy, dev);
1658
1659	if (!cpumask_empty(policy->real_cpus)) {
1660		up_write(&policy->rwsem);
1661		return;
1662	}
1663
1664	/* We did light-weight exit earlier, do full tear down now */
1665	if (cpufreq_driver->offline && cpufreq_driver->exit)
1666		cpufreq_driver->exit(policy);
1667
1668	up_write(&policy->rwsem);
1669
1670	cpufreq_policy_free(policy);
1671}
1672
1673/**
1674 *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1675 *	in deep trouble.
1676 *	@policy: policy managing CPUs
1677 *	@new_freq: CPU frequency the CPU actually runs at
1678 *
1679 *	We adjust to current frequency first, and need to clean up later.
1680 *	So either call to cpufreq_update_policy() or schedule handle_update()).
1681 */
1682static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1683				unsigned int new_freq)
1684{
1685	struct cpufreq_freqs freqs;
1686
1687	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1688		 policy->cur, new_freq);
1689
1690	freqs.old = policy->cur;
1691	freqs.new = new_freq;
1692
1693	cpufreq_freq_transition_begin(policy, &freqs);
1694	cpufreq_freq_transition_end(policy, &freqs, 0);
1695}
1696
1697static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1698{
1699	unsigned int new_freq;
1700
1701	new_freq = cpufreq_driver->get(policy->cpu);
1702	if (!new_freq)
1703		return 0;
1704
1705	/*
1706	 * If fast frequency switching is used with the given policy, the check
1707	 * against policy->cur is pointless, so skip it in that case.
1708	 */
1709	if (policy->fast_switch_enabled || !has_target())
1710		return new_freq;
1711
1712	if (policy->cur != new_freq) {
1713		cpufreq_out_of_sync(policy, new_freq);
1714		if (update)
1715			schedule_work(&policy->update);
1716	}
1717
1718	return new_freq;
1719}
1720
1721/**
1722 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1723 * @cpu: CPU number
1724 *
1725 * This is the last known freq, without actually getting it from the driver.
1726 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1727 */
1728unsigned int cpufreq_quick_get(unsigned int cpu)
1729{
1730	struct cpufreq_policy *policy;
1731	unsigned int ret_freq = 0;
1732	unsigned long flags;
1733
1734	read_lock_irqsave(&cpufreq_driver_lock, flags);
1735
1736	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1737		ret_freq = cpufreq_driver->get(cpu);
1738		read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1739		return ret_freq;
1740	}
1741
1742	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1743
1744	policy = cpufreq_cpu_get(cpu);
1745	if (policy) {
1746		ret_freq = policy->cur;
1747		cpufreq_cpu_put(policy);
1748	}
1749
1750	return ret_freq;
1751}
1752EXPORT_SYMBOL(cpufreq_quick_get);
1753
1754/**
1755 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1756 * @cpu: CPU number
1757 *
1758 * Just return the max possible frequency for a given CPU.
1759 */
1760unsigned int cpufreq_quick_get_max(unsigned int cpu)
1761{
1762	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1763	unsigned int ret_freq = 0;
1764
1765	if (policy) {
1766		ret_freq = policy->max;
1767		cpufreq_cpu_put(policy);
1768	}
1769
1770	return ret_freq;
1771}
1772EXPORT_SYMBOL(cpufreq_quick_get_max);
1773
1774/**
1775 * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1776 * @cpu: CPU number
1777 *
1778 * The default return value is the max_freq field of cpuinfo.
1779 */
1780__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1781{
1782	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1783	unsigned int ret_freq = 0;
1784
1785	if (policy) {
1786		ret_freq = policy->cpuinfo.max_freq;
1787		cpufreq_cpu_put(policy);
1788	}
1789
1790	return ret_freq;
1791}
1792EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1793
1794static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1795{
1796	if (unlikely(policy_is_inactive(policy)))
1797		return 0;
1798
1799	return cpufreq_verify_current_freq(policy, true);
1800}
1801
1802/**
1803 * cpufreq_get - get the current CPU frequency (in kHz)
1804 * @cpu: CPU number
1805 *
1806 * Get the CPU current (static) CPU frequency
1807 */
1808unsigned int cpufreq_get(unsigned int cpu)
1809{
1810	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1811	unsigned int ret_freq = 0;
1812
1813	if (policy) {
1814		down_read(&policy->rwsem);
1815		if (cpufreq_driver->get)
1816			ret_freq = __cpufreq_get(policy);
1817		up_read(&policy->rwsem);
1818
1819		cpufreq_cpu_put(policy);
1820	}
1821
1822	return ret_freq;
1823}
1824EXPORT_SYMBOL(cpufreq_get);
1825
1826static struct subsys_interface cpufreq_interface = {
1827	.name		= "cpufreq",
1828	.subsys		= &cpu_subsys,
1829	.add_dev	= cpufreq_add_dev,
1830	.remove_dev	= cpufreq_remove_dev,
1831};
1832
1833/*
1834 * In case platform wants some specific frequency to be configured
1835 * during suspend..
1836 */
1837int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1838{
1839	int ret;
1840
1841	if (!policy->suspend_freq) {
1842		pr_debug("%s: suspend_freq not defined\n", __func__);
1843		return 0;
1844	}
1845
1846	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1847			policy->suspend_freq);
1848
1849	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1850			CPUFREQ_RELATION_H);
1851	if (ret)
1852		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1853				__func__, policy->suspend_freq, ret);
1854
1855	return ret;
1856}
1857EXPORT_SYMBOL(cpufreq_generic_suspend);
1858
1859/**
1860 * cpufreq_suspend() - Suspend CPUFreq governors
1861 *
1862 * Called during system wide Suspend/Hibernate cycles for suspending governors
1863 * as some platforms can't change frequency after this point in suspend cycle.
1864 * Because some of the devices (like: i2c, regulators, etc) they use for
1865 * changing frequency are suspended quickly after this point.
1866 */
1867void cpufreq_suspend(void)
1868{
1869	struct cpufreq_policy *policy;
1870
1871	if (!cpufreq_driver)
1872		return;
1873
1874	if (!has_target() && !cpufreq_driver->suspend)
1875		goto suspend;
1876
1877	pr_debug("%s: Suspending Governors\n", __func__);
1878
1879	for_each_active_policy(policy) {
1880		if (has_target()) {
1881			down_write(&policy->rwsem);
1882			cpufreq_stop_governor(policy);
1883			up_write(&policy->rwsem);
1884		}
1885
1886		if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1887			pr_err("%s: Failed to suspend driver: %s\n", __func__,
1888				cpufreq_driver->name);
1889	}
1890
1891suspend:
1892	cpufreq_suspended = true;
1893}
1894
1895/**
1896 * cpufreq_resume() - Resume CPUFreq governors
1897 *
1898 * Called during system wide Suspend/Hibernate cycle for resuming governors that
1899 * are suspended with cpufreq_suspend().
1900 */
1901void cpufreq_resume(void)
1902{
1903	struct cpufreq_policy *policy;
1904	int ret;
1905
1906	if (!cpufreq_driver)
1907		return;
1908
1909	if (unlikely(!cpufreq_suspended))
1910		return;
1911
1912	cpufreq_suspended = false;
1913
1914	if (!has_target() && !cpufreq_driver->resume)
1915		return;
1916
1917	pr_debug("%s: Resuming Governors\n", __func__);
1918
1919	for_each_active_policy(policy) {
1920		if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1921			pr_err("%s: Failed to resume driver: %p\n", __func__,
1922				policy);
1923		} else if (has_target()) {
1924			down_write(&policy->rwsem);
1925			ret = cpufreq_start_governor(policy);
1926			up_write(&policy->rwsem);
1927
1928			if (ret)
1929				pr_err("%s: Failed to start governor for policy: %p\n",
1930				       __func__, policy);
1931		}
1932	}
1933}
1934
1935/**
1936 * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
1937 * @flags: Flags to test against the current cpufreq driver's flags.
1938 *
1939 * Assumes that the driver is there, so callers must ensure that this is the
1940 * case.
1941 */
1942bool cpufreq_driver_test_flags(u16 flags)
1943{
1944	return !!(cpufreq_driver->flags & flags);
1945}
1946
1947/**
1948 *	cpufreq_get_current_driver - return current driver's name
1949 *
1950 *	Return the name string of the currently loaded cpufreq driver
1951 *	or NULL, if none.
1952 */
1953const char *cpufreq_get_current_driver(void)
1954{
1955	if (cpufreq_driver)
1956		return cpufreq_driver->name;
1957
1958	return NULL;
1959}
1960EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1961
1962/**
1963 *	cpufreq_get_driver_data - return current driver data
1964 *
1965 *	Return the private data of the currently loaded cpufreq
1966 *	driver, or NULL if no cpufreq driver is loaded.
1967 */
1968void *cpufreq_get_driver_data(void)
1969{
1970	if (cpufreq_driver)
1971		return cpufreq_driver->driver_data;
1972
1973	return NULL;
1974}
1975EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1976
1977/*********************************************************************
1978 *                     NOTIFIER LISTS INTERFACE                      *
1979 *********************************************************************/
1980
1981/**
1982 *	cpufreq_register_notifier - register a driver with cpufreq
1983 *	@nb: notifier function to register
1984 *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1985 *
1986 *	Add a driver to one of two lists: either a list of drivers that
1987 *      are notified about clock rate changes (once before and once after
1988 *      the transition), or a list of drivers that are notified about
1989 *      changes in cpufreq policy.
1990 *
1991 *	This function may sleep, and has the same return conditions as
1992 *	blocking_notifier_chain_register.
1993 */
1994int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1995{
1996	int ret;
1997
1998	if (cpufreq_disabled())
1999		return -EINVAL;
2000
2001	switch (list) {
2002	case CPUFREQ_TRANSITION_NOTIFIER:
2003		mutex_lock(&cpufreq_fast_switch_lock);
2004
2005		if (cpufreq_fast_switch_count > 0) {
2006			mutex_unlock(&cpufreq_fast_switch_lock);
2007			return -EBUSY;
2008		}
2009		ret = srcu_notifier_chain_register(
2010				&cpufreq_transition_notifier_list, nb);
2011		if (!ret)
2012			cpufreq_fast_switch_count--;
2013
2014		mutex_unlock(&cpufreq_fast_switch_lock);
2015		break;
2016	case CPUFREQ_POLICY_NOTIFIER:
2017		ret = blocking_notifier_chain_register(
2018				&cpufreq_policy_notifier_list, nb);
2019		break;
2020	default:
2021		ret = -EINVAL;
2022	}
2023
2024	return ret;
2025}
2026EXPORT_SYMBOL(cpufreq_register_notifier);
2027
2028/**
2029 *	cpufreq_unregister_notifier - unregister a driver with cpufreq
2030 *	@nb: notifier block to be unregistered
2031 *	@list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
2032 *
2033 *	Remove a driver from the CPU frequency notifier list.
2034 *
2035 *	This function may sleep, and has the same return conditions as
2036 *	blocking_notifier_chain_unregister.
2037 */
2038int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2039{
2040	int ret;
2041
2042	if (cpufreq_disabled())
2043		return -EINVAL;
2044
2045	switch (list) {
2046	case CPUFREQ_TRANSITION_NOTIFIER:
2047		mutex_lock(&cpufreq_fast_switch_lock);
2048
2049		ret = srcu_notifier_chain_unregister(
2050				&cpufreq_transition_notifier_list, nb);
2051		if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2052			cpufreq_fast_switch_count++;
2053
2054		mutex_unlock(&cpufreq_fast_switch_lock);
2055		break;
2056	case CPUFREQ_POLICY_NOTIFIER:
2057		ret = blocking_notifier_chain_unregister(
2058				&cpufreq_policy_notifier_list, nb);
2059		break;
2060	default:
2061		ret = -EINVAL;
2062	}
2063
2064	return ret;
2065}
2066EXPORT_SYMBOL(cpufreq_unregister_notifier);
2067
2068
2069/*********************************************************************
2070 *                              GOVERNORS                            *
2071 *********************************************************************/
2072
2073/**
2074 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2075 * @policy: cpufreq policy to switch the frequency for.
2076 * @target_freq: New frequency to set (may be approximate).
2077 *
2078 * Carry out a fast frequency switch without sleeping.
2079 *
2080 * The driver's ->fast_switch() callback invoked by this function must be
2081 * suitable for being called from within RCU-sched read-side critical sections
2082 * and it is expected to select the minimum available frequency greater than or
2083 * equal to @target_freq (CPUFREQ_RELATION_L).
2084 *
2085 * This function must not be called if policy->fast_switch_enabled is unset.
2086 *
2087 * Governors calling this function must guarantee that it will never be invoked
2088 * twice in parallel for the same policy and that it will never be called in
2089 * parallel with either ->target() or ->target_index() for the same policy.
2090 *
2091 * Returns the actual frequency set for the CPU.
2092 *
2093 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2094 * error condition, the hardware configuration must be preserved.
2095 */
2096unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2097					unsigned int target_freq)
2098{
2099	unsigned int freq;
2100	int cpu;
2101
2102	target_freq = clamp_val(target_freq, policy->min, policy->max);
2103	freq = cpufreq_driver->fast_switch(policy, target_freq);
2104
2105	if (!freq)
2106		return 0;
2107
2108	policy->cur = freq;
2109	arch_set_freq_scale(policy->related_cpus, freq,
2110			    policy->cpuinfo.max_freq);
2111	cpufreq_stats_record_transition(policy, freq);
2112
2113	if (trace_cpu_frequency_enabled()) {
2114		for_each_cpu(cpu, policy->cpus)
2115			trace_cpu_frequency(freq, cpu);
2116	}
2117
2118	return freq;
2119}
2120EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2121
2122/* Must set freqs->new to intermediate frequency */
2123static int __target_intermediate(struct cpufreq_policy *policy,
2124				 struct cpufreq_freqs *freqs, int index)
2125{
2126	int ret;
2127
2128	freqs->new = cpufreq_driver->get_intermediate(policy, index);
2129
2130	/* We don't need to switch to intermediate freq */
2131	if (!freqs->new)
2132		return 0;
2133
2134	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2135		 __func__, policy->cpu, freqs->old, freqs->new);
2136
2137	cpufreq_freq_transition_begin(policy, freqs);
2138	ret = cpufreq_driver->target_intermediate(policy, index);
2139	cpufreq_freq_transition_end(policy, freqs, ret);
2140
2141	if (ret)
2142		pr_err("%s: Failed to change to intermediate frequency: %d\n",
2143		       __func__, ret);
2144
2145	return ret;
2146}
2147
2148static int __target_index(struct cpufreq_policy *policy, int index)
2149{
2150	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2151	unsigned int intermediate_freq = 0;
2152	unsigned int newfreq = policy->freq_table[index].frequency;
2153	int retval = -EINVAL;
2154	bool notify;
2155
2156	if (newfreq == policy->cur)
2157		return 0;
2158
2159	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2160	if (notify) {
2161		/* Handle switching to intermediate frequency */
2162		if (cpufreq_driver->get_intermediate) {
2163			retval = __target_intermediate(policy, &freqs, index);
2164			if (retval)
2165				return retval;
2166
2167			intermediate_freq = freqs.new;
2168			/* Set old freq to intermediate */
2169			if (intermediate_freq)
2170				freqs.old = freqs.new;
2171		}
2172
2173		freqs.new = newfreq;
2174		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2175			 __func__, policy->cpu, freqs.old, freqs.new);
2176
2177		cpufreq_freq_transition_begin(policy, &freqs);
2178	}
2179
2180	retval = cpufreq_driver->target_index(policy, index);
2181	if (retval)
2182		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2183		       retval);
2184
2185	if (notify) {
2186		cpufreq_freq_transition_end(policy, &freqs, retval);
2187
2188		/*
2189		 * Failed after setting to intermediate freq? Driver should have
2190		 * reverted back to initial frequency and so should we. Check
2191		 * here for intermediate_freq instead of get_intermediate, in
2192		 * case we haven't switched to intermediate freq at all.
2193		 */
2194		if (unlikely(retval && intermediate_freq)) {
2195			freqs.old = intermediate_freq;
2196			freqs.new = policy->restore_freq;
2197			cpufreq_freq_transition_begin(policy, &freqs);
2198			cpufreq_freq_transition_end(policy, &freqs, 0);
2199		}
2200	}
2201
2202	return retval;
2203}
2204
2205int __cpufreq_driver_target(struct cpufreq_policy *policy,
2206			    unsigned int target_freq,
2207			    unsigned int relation)
2208{
2209	unsigned int old_target_freq = target_freq;
2210	int index;
2211
2212	if (cpufreq_disabled())
2213		return -ENODEV;
2214
2215	/* Make sure that target_freq is within supported range */
2216	target_freq = clamp_val(target_freq, policy->min, policy->max);
2217
2218	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2219		 policy->cpu, target_freq, relation, old_target_freq);
2220
2221	/*
2222	 * This might look like a redundant call as we are checking it again
2223	 * after finding index. But it is left intentionally for cases where
2224	 * exactly same freq is called again and so we can save on few function
2225	 * calls.
2226	 */
2227	if (target_freq == policy->cur &&
2228	    !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2229		return 0;
2230
2231	/* Save last value to restore later on errors */
2232	policy->restore_freq = policy->cur;
2233
2234	if (cpufreq_driver->target)
2235		return cpufreq_driver->target(policy, target_freq, relation);
2236
2237	if (!cpufreq_driver->target_index)
2238		return -EINVAL;
2239
2240	index = cpufreq_frequency_table_target(policy, target_freq, relation);
2241
2242	return __target_index(policy, index);
2243}
2244EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2245
2246int cpufreq_driver_target(struct cpufreq_policy *policy,
2247			  unsigned int target_freq,
2248			  unsigned int relation)
2249{
2250	int ret;
2251
2252	down_write(&policy->rwsem);
2253
2254	ret = __cpufreq_driver_target(policy, target_freq, relation);
2255
2256	up_write(&policy->rwsem);
2257
2258	return ret;
2259}
2260EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2261
2262__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2263{
2264	return NULL;
2265}
2266
2267static int cpufreq_init_governor(struct cpufreq_policy *policy)
2268{
2269	int ret;
2270
2271	/* Don't start any governor operations if we are entering suspend */
2272	if (cpufreq_suspended)
2273		return 0;
2274	/*
2275	 * Governor might not be initiated here if ACPI _PPC changed
2276	 * notification happened, so check it.
2277	 */
2278	if (!policy->governor)
2279		return -EINVAL;
2280
2281	/* Platform doesn't want dynamic frequency switching ? */
2282	if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2283	    cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2284		struct cpufreq_governor *gov = cpufreq_fallback_governor();
2285
2286		if (gov) {
2287			pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2288				policy->governor->name, gov->name);
2289			policy->governor = gov;
2290		} else {
2291			return -EINVAL;
2292		}
2293	}
2294
2295	if (!try_module_get(policy->governor->owner))
2296		return -EINVAL;
2297
2298	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2299
2300	if (policy->governor->init) {
2301		ret = policy->governor->init(policy);
2302		if (ret) {
2303			module_put(policy->governor->owner);
2304			return ret;
2305		}
2306	}
2307
2308	policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2309
2310	return 0;
2311}
2312
2313static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2314{
2315	if (cpufreq_suspended || !policy->governor)
2316		return;
2317
2318	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2319
2320	if (policy->governor->exit)
2321		policy->governor->exit(policy);
2322
2323	module_put(policy->governor->owner);
2324}
2325
2326int cpufreq_start_governor(struct cpufreq_policy *policy)
2327{
2328	int ret;
2329
2330	if (cpufreq_suspended)
2331		return 0;
2332
2333	if (!policy->governor)
2334		return -EINVAL;
2335
2336	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2337
2338	if (cpufreq_driver->get)
2339		cpufreq_verify_current_freq(policy, false);
2340
2341	if (policy->governor->start) {
2342		ret = policy->governor->start(policy);
2343		if (ret)
2344			return ret;
2345	}
2346
2347	if (policy->governor->limits)
2348		policy->governor->limits(policy);
2349
2350	return 0;
2351}
2352
2353void cpufreq_stop_governor(struct cpufreq_policy *policy)
2354{
2355	if (cpufreq_suspended || !policy->governor)
2356		return;
2357
2358	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2359
2360	if (policy->governor->stop)
2361		policy->governor->stop(policy);
2362}
2363
2364static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2365{
2366	if (cpufreq_suspended || !policy->governor)
2367		return;
2368
2369	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2370
2371	if (policy->governor->limits)
2372		policy->governor->limits(policy);
2373}
2374
2375int cpufreq_register_governor(struct cpufreq_governor *governor)
2376{
2377	int err;
2378
2379	if (!governor)
2380		return -EINVAL;
2381
2382	if (cpufreq_disabled())
2383		return -ENODEV;
2384
2385	mutex_lock(&cpufreq_governor_mutex);
2386
2387	err = -EBUSY;
2388	if (!find_governor(governor->name)) {
2389		err = 0;
2390		list_add(&governor->governor_list, &cpufreq_governor_list);
2391	}
2392
2393	mutex_unlock(&cpufreq_governor_mutex);
2394	return err;
2395}
2396EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2397
2398void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2399{
2400	struct cpufreq_policy *policy;
2401	unsigned long flags;
2402
2403	if (!governor)
2404		return;
2405
2406	if (cpufreq_disabled())
2407		return;
2408
2409	/* clear last_governor for all inactive policies */
2410	read_lock_irqsave(&cpufreq_driver_lock, flags);
2411	for_each_inactive_policy(policy) {
2412		if (!strcmp(policy->last_governor, governor->name)) {
2413			policy->governor = NULL;
2414			strcpy(policy->last_governor, "\0");
2415		}
2416	}
2417	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2418
2419	mutex_lock(&cpufreq_governor_mutex);
2420	list_del(&governor->governor_list);
2421	mutex_unlock(&cpufreq_governor_mutex);
2422}
2423EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2424
2425
2426/*********************************************************************
2427 *                          POLICY INTERFACE                         *
2428 *********************************************************************/
2429
2430/**
2431 * cpufreq_get_policy - get the current cpufreq_policy
2432 * @policy: struct cpufreq_policy into which the current cpufreq_policy
2433 *	is written
2434 * @cpu: CPU to find the policy for
2435 *
2436 * Reads the current cpufreq policy.
2437 */
2438int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2439{
2440	struct cpufreq_policy *cpu_policy;
2441	if (!policy)
2442		return -EINVAL;
2443
2444	cpu_policy = cpufreq_cpu_get(cpu);
2445	if (!cpu_policy)
2446		return -EINVAL;
2447
2448	memcpy(policy, cpu_policy, sizeof(*policy));
2449
2450	cpufreq_cpu_put(cpu_policy);
2451	return 0;
2452}
2453EXPORT_SYMBOL(cpufreq_get_policy);
2454
2455/**
2456 * cpufreq_set_policy - Modify cpufreq policy parameters.
2457 * @policy: Policy object to modify.
2458 * @new_gov: Policy governor pointer.
2459 * @new_pol: Policy value (for drivers with built-in governors).
2460 *
2461 * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2462 * limits to be set for the policy, update @policy with the verified limits
2463 * values and either invoke the driver's ->setpolicy() callback (if present) or
2464 * carry out a governor update for @policy.  That is, run the current governor's
2465 * ->limits() callback (if @new_gov points to the same object as the one in
2466 * @policy) or replace the governor for @policy with @new_gov.
2467 *
2468 * The cpuinfo part of @policy is not updated by this function.
2469 */
2470static int cpufreq_set_policy(struct cpufreq_policy *policy,
2471			      struct cpufreq_governor *new_gov,
2472			      unsigned int new_pol)
2473{
2474	struct cpufreq_policy_data new_data;
2475	struct cpufreq_governor *old_gov;
2476	int ret;
2477
2478	memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2479	new_data.freq_table = policy->freq_table;
2480	new_data.cpu = policy->cpu;
2481	/*
2482	 * PM QoS framework collects all the requests from users and provide us
2483	 * the final aggregated value here.
2484	 */
2485	new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2486	new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2487
2488	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2489		 new_data.cpu, new_data.min, new_data.max);
2490
2491	/*
2492	 * Verify that the CPU speed can be set within these limits and make sure
2493	 * that min <= max.
2494	 */
2495	ret = cpufreq_driver->verify(&new_data);
2496	if (ret)
2497		return ret;
2498
2499	policy->min = new_data.min;
2500	policy->max = new_data.max;
2501	trace_cpu_frequency_limits(policy);
2502
2503	policy->cached_target_freq = UINT_MAX;
2504
2505	pr_debug("new min and max freqs are %u - %u kHz\n",
2506		 policy->min, policy->max);
2507
2508	if (cpufreq_driver->setpolicy) {
2509		policy->policy = new_pol;
2510		pr_debug("setting range\n");
2511		return cpufreq_driver->setpolicy(policy);
2512	}
2513
2514	if (new_gov == policy->governor) {
2515		pr_debug("governor limits update\n");
2516		cpufreq_governor_limits(policy);
2517		return 0;
2518	}
2519
2520	pr_debug("governor switch\n");
2521
2522	/* save old, working values */
2523	old_gov = policy->governor;
2524	/* end old governor */
2525	if (old_gov) {
2526		cpufreq_stop_governor(policy);
2527		cpufreq_exit_governor(policy);
2528	}
2529
2530	/* start new governor */
2531	policy->governor = new_gov;
2532	ret = cpufreq_init_governor(policy);
2533	if (!ret) {
2534		ret = cpufreq_start_governor(policy);
2535		if (!ret) {
2536			pr_debug("governor change\n");
2537			sched_cpufreq_governor_change(policy, old_gov);
2538			return 0;
2539		}
2540		cpufreq_exit_governor(policy);
2541	}
2542
2543	/* new governor failed, so re-start old one */
2544	pr_debug("starting governor %s failed\n", policy->governor->name);
2545	if (old_gov) {
2546		policy->governor = old_gov;
2547		if (cpufreq_init_governor(policy))
2548			policy->governor = NULL;
2549		else
2550			cpufreq_start_governor(policy);
2551	}
2552
2553	return ret;
2554}
2555
2556/**
2557 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2558 * @cpu: CPU to re-evaluate the policy for.
2559 *
2560 * Update the current frequency for the cpufreq policy of @cpu and use
2561 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2562 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2563 * for the policy in question, among other things.
2564 */
2565void cpufreq_update_policy(unsigned int cpu)
2566{
2567	struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2568
2569	if (!policy)
2570		return;
2571
2572	/*
2573	 * BIOS might change freq behind our back
2574	 * -> ask driver for current freq and notify governors about a change
2575	 */
2576	if (cpufreq_driver->get && has_target() &&
2577	    (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2578		goto unlock;
2579
2580	refresh_frequency_limits(policy);
2581
2582unlock:
2583	cpufreq_cpu_release(policy);
2584}
2585EXPORT_SYMBOL(cpufreq_update_policy);
2586
2587/**
2588 * cpufreq_update_limits - Update policy limits for a given CPU.
2589 * @cpu: CPU to update the policy limits for.
2590 *
2591 * Invoke the driver's ->update_limits callback if present or call
2592 * cpufreq_update_policy() for @cpu.
2593 */
2594void cpufreq_update_limits(unsigned int cpu)
2595{
2596	if (cpufreq_driver->update_limits)
2597		cpufreq_driver->update_limits(cpu);
2598	else
2599		cpufreq_update_policy(cpu);
2600}
2601EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2602
2603/*********************************************************************
2604 *               BOOST						     *
2605 *********************************************************************/
2606static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2607{
2608	int ret;
2609
2610	if (!policy->freq_table)
2611		return -ENXIO;
2612
2613	ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2614	if (ret) {
2615		pr_err("%s: Policy frequency update failed\n", __func__);
2616		return ret;
2617	}
2618
2619	ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2620	if (ret < 0)
2621		return ret;
2622
2623	return 0;
2624}
2625
2626int cpufreq_boost_trigger_state(int state)
2627{
2628	struct cpufreq_policy *policy;
2629	unsigned long flags;
2630	int ret = 0;
2631
2632	if (cpufreq_driver->boost_enabled == state)
2633		return 0;
2634
2635	write_lock_irqsave(&cpufreq_driver_lock, flags);
2636	cpufreq_driver->boost_enabled = state;
2637	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2638
2639	get_online_cpus();
2640	for_each_active_policy(policy) {
2641		ret = cpufreq_driver->set_boost(policy, state);
2642		if (ret)
2643			goto err_reset_state;
2644	}
2645	put_online_cpus();
2646
2647	return 0;
2648
2649err_reset_state:
2650	put_online_cpus();
2651
2652	write_lock_irqsave(&cpufreq_driver_lock, flags);
2653	cpufreq_driver->boost_enabled = !state;
2654	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2655
2656	pr_err("%s: Cannot %s BOOST\n",
2657	       __func__, state ? "enable" : "disable");
2658
2659	return ret;
2660}
2661
2662static bool cpufreq_boost_supported(void)
2663{
2664	return cpufreq_driver->set_boost;
2665}
2666
2667static int create_boost_sysfs_file(void)
2668{
2669	int ret;
2670
2671	ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2672	if (ret)
2673		pr_err("%s: cannot register global BOOST sysfs file\n",
2674		       __func__);
2675
2676	return ret;
2677}
2678
2679static void remove_boost_sysfs_file(void)
2680{
2681	if (cpufreq_boost_supported())
2682		sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2683}
2684
2685int cpufreq_enable_boost_support(void)
2686{
2687	if (!cpufreq_driver)
2688		return -EINVAL;
2689
2690	if (cpufreq_boost_supported())
2691		return 0;
2692
2693	cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2694
2695	/* This will get removed on driver unregister */
2696	return create_boost_sysfs_file();
2697}
2698EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2699
2700int cpufreq_boost_enabled(void)
2701{
2702	return cpufreq_driver->boost_enabled;
2703}
2704EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2705
2706/*********************************************************************
2707 *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2708 *********************************************************************/
2709static enum cpuhp_state hp_online;
2710
2711static int cpuhp_cpufreq_online(unsigned int cpu)
2712{
2713	cpufreq_online(cpu);
2714
2715	return 0;
2716}
2717
2718static int cpuhp_cpufreq_offline(unsigned int cpu)
2719{
2720	cpufreq_offline(cpu);
2721
2722	return 0;
2723}
2724
2725/**
2726 * cpufreq_register_driver - register a CPU Frequency driver
2727 * @driver_data: A struct cpufreq_driver containing the values#
2728 * submitted by the CPU Frequency driver.
2729 *
2730 * Registers a CPU Frequency driver to this core code. This code
2731 * returns zero on success, -EEXIST when another driver got here first
2732 * (and isn't unregistered in the meantime).
2733 *
2734 */
2735int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2736{
2737	unsigned long flags;
2738	int ret;
2739
2740	if (cpufreq_disabled())
2741		return -ENODEV;
2742
2743	/*
2744	 * The cpufreq core depends heavily on the availability of device
2745	 * structure, make sure they are available before proceeding further.
2746	 */
2747	if (!get_cpu_device(0))
2748		return -EPROBE_DEFER;
2749
2750	if (!driver_data || !driver_data->verify || !driver_data->init ||
2751	    !(driver_data->setpolicy || driver_data->target_index ||
2752		    driver_data->target) ||
2753	     (driver_data->setpolicy && (driver_data->target_index ||
2754		    driver_data->target)) ||
2755	     (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2756	     (!driver_data->online != !driver_data->offline))
2757		return -EINVAL;
2758
2759	pr_debug("trying to register driver %s\n", driver_data->name);
2760
2761	/* Protect against concurrent CPU online/offline. */
2762	cpus_read_lock();
2763
2764	write_lock_irqsave(&cpufreq_driver_lock, flags);
2765	if (cpufreq_driver) {
2766		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2767		ret = -EEXIST;
2768		goto out;
2769	}
2770	cpufreq_driver = driver_data;
2771	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2772
2773	/*
2774	 * Mark support for the scheduler's frequency invariance engine for
2775	 * drivers that implement target(), target_index() or fast_switch().
2776	 */
2777	if (!cpufreq_driver->setpolicy) {
2778		static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2779		pr_debug("supports frequency invariance");
2780	}
2781
2782	if (driver_data->setpolicy)
2783		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2784
2785	if (cpufreq_boost_supported()) {
2786		ret = create_boost_sysfs_file();
2787		if (ret)
2788			goto err_null_driver;
2789	}
2790
2791	ret = subsys_interface_register(&cpufreq_interface);
2792	if (ret)
2793		goto err_boost_unreg;
2794
2795	if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2796	    list_empty(&cpufreq_policy_list)) {
2797		/* if all ->init() calls failed, unregister */
2798		ret = -ENODEV;
2799		pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2800			 driver_data->name);
2801		goto err_if_unreg;
2802	}
2803
2804	ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2805						   "cpufreq:online",
2806						   cpuhp_cpufreq_online,
2807						   cpuhp_cpufreq_offline);
2808	if (ret < 0)
2809		goto err_if_unreg;
2810	hp_online = ret;
2811	ret = 0;
2812
2813	pr_debug("driver %s up and running\n", driver_data->name);
2814	goto out;
2815
2816err_if_unreg:
2817	subsys_interface_unregister(&cpufreq_interface);
2818err_boost_unreg:
2819	remove_boost_sysfs_file();
2820err_null_driver:
2821	write_lock_irqsave(&cpufreq_driver_lock, flags);
2822	cpufreq_driver = NULL;
2823	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2824out:
2825	cpus_read_unlock();
2826	return ret;
2827}
2828EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2829
2830/*
2831 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2832 *
2833 * Unregister the current CPUFreq driver. Only call this if you have
2834 * the right to do so, i.e. if you have succeeded in initialising before!
2835 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2836 * currently not initialised.
2837 */
2838int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2839{
2840	unsigned long flags;
2841
2842	if (!cpufreq_driver || (driver != cpufreq_driver))
2843		return -EINVAL;
2844
2845	pr_debug("unregistering driver %s\n", driver->name);
2846
2847	/* Protect against concurrent cpu hotplug */
2848	cpus_read_lock();
2849	subsys_interface_unregister(&cpufreq_interface);
2850	remove_boost_sysfs_file();
2851	static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
2852	cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2853
2854	write_lock_irqsave(&cpufreq_driver_lock, flags);
2855
2856	cpufreq_driver = NULL;
2857
2858	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2859	cpus_read_unlock();
2860
2861	return 0;
2862}
2863EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2864
2865static int __init cpufreq_core_init(void)
2866{
2867	struct cpufreq_governor *gov = cpufreq_default_governor();
2868
2869	if (cpufreq_disabled())
2870		return -ENODEV;
2871
2872	cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2873	BUG_ON(!cpufreq_global_kobject);
2874
2875	if (!strlen(default_governor))
2876		strncpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
2877
2878	return 0;
2879}
2880module_param(off, int, 0444);
2881module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
2882core_initcall(cpufreq_core_init);
2883