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 <linux/units.h>
32	#include <trace/events/power.h>
33
34	static LIST_HEAD(cpufreq_policy_list);
35
36	/* Macros to iterate over CPU policies */
37	#define for_each_suitable_policy(__policy, __active) \
38	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
39	if ((__active) == !policy_is_inactive(__policy))
40
41	#define for_each_active_policy(__policy) \
42	for_each_suitable_policy(__policy, true)
43	#define for_each_inactive_policy(__policy) \
44	for_each_suitable_policy(__policy, false)
45
46	/* Iterate over governors */
47	static LIST_HEAD(cpufreq_governor_list);
48	#define for_each_governor(__governor) \
49	list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
50
51	static char default_governor[CPUFREQ_NAME_LEN];
52
53	/*
54	* The "cpufreq driver" - the arch- or hardware-dependent low
55	* level driver of CPUFreq support, and its spinlock. This lock
56	* also protects the cpufreq_cpu_data array.
57	*/
58	static struct cpufreq_driver *cpufreq_driver;
59	static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
60	static DEFINE_RWLOCK(cpufreq_driver_lock);
61
62	static DEFINE_STATIC_KEY_FALSE(cpufreq_freq_invariance);
63	bool cpufreq_supports_freq_invariance(void)
64	{
65	return static_branch_likely(&cpufreq_freq_invariance);
66	}
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	bool has_target_index(void)
77	{
78	return !!cpufreq_driver->target_index;
79	}
80
81	/* internal prototypes */
82	static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
83	static int cpufreq_init_governor(struct cpufreq_policy *policy);
84	static void cpufreq_exit_governor(struct cpufreq_policy *policy);
85	static void cpufreq_governor_limits(struct cpufreq_policy *policy);
86	static int cpufreq_set_policy(struct cpufreq_policy *policy,
87	struct cpufreq_governor *new_gov,
88	unsigned int new_pol);
89	static bool cpufreq_boost_supported(void);
90
91	/*
92	* Two notifier lists: the "policy" list is involved in the
93	* validation process for a new CPU frequency policy; the
94	* "transition" list for kernel code that needs to handle
95	* changes to devices when the CPU clock speed changes.
96	* The mutex locks both lists.
97	*/
98	static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
99	SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
100
101	static int off __read_mostly;
102	static int cpufreq_disabled(void)
103	{
104	return off;
105	}
106	void disable_cpufreq(void)
107	{
108	off = 1;
109	}
110	static DEFINE_MUTEX(cpufreq_governor_mutex);
111
112	bool have_governor_per_policy(void)
113	{
114	return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
115	}
116	EXPORT_SYMBOL_GPL(have_governor_per_policy);
117
118	static struct kobject *cpufreq_global_kobject;
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	struct kernel_cpustat kcpustat;
132	u64 cur_wall_time;
133	u64 idle_time;
134	u64 busy_time;
135
136	cur_wall_time = jiffies64_to_nsecs(j: get_jiffies_64());
137
138	kcpustat_cpu_fetch(dst: &kcpustat, cpu);
139
140	busy_time = kcpustat.cpustat[CPUTIME_USER];
141	busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
142	busy_time += kcpustat.cpustat[CPUTIME_IRQ];
143	busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
144	busy_time += kcpustat.cpustat[CPUTIME_STEAL];
145	busy_time += kcpustat.cpustat[CPUTIME_NICE];
146
147	idle_time = cur_wall_time - busy_time;
148	if (wall)
149	*wall = div_u64(dividend: cur_wall_time, NSEC_PER_USEC);
150
151	return div_u64(dividend: idle_time, NSEC_PER_USEC);
152	}
153
154	u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
155	{
156	u64 idle_time = get_cpu_idle_time_us(cpu, last_update_time: io_busy ? wall : NULL);
157
158	if (idle_time == -1ULL)
159	return get_cpu_idle_time_jiffy(cpu, wall);
160	else if (!io_busy)
161	idle_time += get_cpu_iowait_time_us(cpu, last_update_time: wall);
162
163	return idle_time;
164	}
165	EXPORT_SYMBOL_GPL(get_cpu_idle_time);
166
167	/*
168	* This is a generic cpufreq init() routine which can be used by cpufreq
169	* drivers of SMP systems. It will do following:
170	* - validate & show freq table passed
171	* - set policies transition latency
172	* - policy->cpus with all possible CPUs
173	*/
174	void cpufreq_generic_init(struct cpufreq_policy *policy,
175	struct cpufreq_frequency_table *table,
176	unsigned int transition_latency)
177	{
178	policy->freq_table = table;
179	policy->cpuinfo.transition_latency = transition_latency;
180
181	/*
182	* The driver only supports the SMP configuration where all processors
183	* share the clock and voltage and clock.
184	*/
185	cpumask_setall(dstp: policy->cpus);
186	}
187	EXPORT_SYMBOL_GPL(cpufreq_generic_init);
188
189	struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
190	{
191	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
192
193	return policy && cpumask_test_cpu(cpu, cpumask: policy->cpus) ? policy : NULL;
194	}
195	EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
196
197	unsigned int cpufreq_generic_get(unsigned int cpu)
198	{
199	struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
200
201	if (!policy || IS_ERR(ptr: policy->clk)) {
202	pr_err("%s: No %s associated to cpu: %d\n",
203	__func__, policy ? "clk" : "policy", cpu);
204	return 0;
205	}
206
207	return clk_get_rate(clk: policy->clk) / 1000;
208	}
209	EXPORT_SYMBOL_GPL(cpufreq_generic_get);
210
211	/**
212	* cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
213	* @cpu: CPU to find the policy for.
214	*
215	* Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
216	* the kobject reference counter of that policy. Return a valid policy on
217	* success or NULL on failure.
218	*
219	* The policy returned by this function has to be released with the help of
220	* cpufreq_cpu_put() to balance its kobject reference counter properly.
221	*/
222	struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
223	{
224	struct cpufreq_policy *policy = NULL;
225	unsigned long flags;
226
227	if (WARN_ON(cpu >= nr_cpu_ids))
228	return NULL;
229
230	/* get the cpufreq driver */
231	read_lock_irqsave(&cpufreq_driver_lock, flags);
232
233	if (cpufreq_driver) {
234	/* get the CPU */
235	policy = cpufreq_cpu_get_raw(cpu);
236	if (policy)
237	kobject_get(kobj: &policy->kobj);
238	}
239
240	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
241
242	return policy;
243	}
244	EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
245
246	/**
247	* cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
248	* @policy: cpufreq policy returned by cpufreq_cpu_get().
249	*/
250	void cpufreq_cpu_put(struct cpufreq_policy *policy)
251	{
252	kobject_put(kobj: &policy->kobj);
253	}
254	EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
255
256	/**
257	* cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
258	* @policy: cpufreq policy returned by cpufreq_cpu_acquire().
259	*/
260	void cpufreq_cpu_release(struct cpufreq_policy *policy)
261	{
262	if (WARN_ON(!policy))
263	return;
264
265	lockdep_assert_held(&policy->rwsem);
266
267	up_write(sem: &policy->rwsem);
268
269	cpufreq_cpu_put(policy);
270	}
271
272	/**
273	* cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
274	* @cpu: CPU to find the policy for.
275	*
276	* Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
277	* if the policy returned by it is not NULL, acquire its rwsem for writing.
278	* Return the policy if it is active or release it and return NULL otherwise.
279	*
280	* The policy returned by this function has to be released with the help of
281	* cpufreq_cpu_release() in order to release its rwsem and balance its usage
282	* counter properly.
283	*/
284	struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
285	{
286	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
287
288	if (!policy)
289	return NULL;
290
291	down_write(sem: &policy->rwsem);
292
293	if (policy_is_inactive(policy)) {
294	cpufreq_cpu_release(policy);
295	return NULL;
296	}
297
298	return policy;
299	}
300
301	/*********************************************************************
302	* EXTERNALLY AFFECTING FREQUENCY CHANGES *
303	*********************************************************************/
304
305	/**
306	* adjust_jiffies - Adjust the system "loops_per_jiffy".
307	* @val: CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
308	* @ci: Frequency change information.
309	*
310	* This function alters the system "loops_per_jiffy" for the clock
311	* speed change. Note that loops_per_jiffy cannot be updated on SMP
312	* systems as each CPU might be scaled differently. So, use the arch
313	* per-CPU loops_per_jiffy value wherever possible.
314	*/
315	static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
316	{
317	#ifndef CONFIG_SMP
318	static unsigned long l_p_j_ref;
319	static unsigned int l_p_j_ref_freq;
320
321	if (ci->flags & CPUFREQ_CONST_LOOPS)
322	return;
323
324	if (!l_p_j_ref_freq) {
325	l_p_j_ref = loops_per_jiffy;
326	l_p_j_ref_freq = ci->old;
327	pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
328	l_p_j_ref, l_p_j_ref_freq);
329	}
330	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
331	loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
332	ci->new);
333	pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
334	loops_per_jiffy, ci->new);
335	}
336	#endif
337	}
338
339	/**
340	* cpufreq_notify_transition - Notify frequency transition and adjust jiffies.
341	* @policy: cpufreq policy to enable fast frequency switching for.
342	* @freqs: contain details of the frequency update.
343	* @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
344	*
345	* This function calls the transition notifiers and adjust_jiffies().
346	*
347	* It is called twice on all CPU frequency changes that have external effects.
348	*/
349	static void cpufreq_notify_transition(struct cpufreq_policy *policy,
350	struct cpufreq_freqs *freqs,
351	unsigned int state)
352	{
353	int cpu;
354
355	BUG_ON(irqs_disabled());
356
357	if (cpufreq_disabled())
358	return;
359
360	freqs->policy = policy;
361	freqs->flags = cpufreq_driver->flags;
362	pr_debug("notification %u of frequency transition to %u kHz\n",
363	state, freqs->new);
364
365	switch (state) {
366	case CPUFREQ_PRECHANGE:
367	/*
368	* Detect if the driver reported a value as "old frequency"
369	* which is not equal to what the cpufreq core thinks is
370	* "old frequency".
371	*/
372	if (policy->cur && policy->cur != freqs->old) {
373	pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
374	freqs->old, policy->cur);
375	freqs->old = policy->cur;
376	}
377
378	srcu_notifier_call_chain(nh: &cpufreq_transition_notifier_list,
379	CPUFREQ_PRECHANGE, v: freqs);
380
381	adjust_jiffies(CPUFREQ_PRECHANGE, ci: freqs);
382	break;
383
384	case CPUFREQ_POSTCHANGE:
385	adjust_jiffies(CPUFREQ_POSTCHANGE, ci: freqs);
386	pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
387	cpumask_pr_args(policy->cpus));
388
389	for_each_cpu(cpu, policy->cpus)
390	trace_cpu_frequency(frequency: freqs->new, cpu_id: cpu);
391
392	srcu_notifier_call_chain(nh: &cpufreq_transition_notifier_list,
393	CPUFREQ_POSTCHANGE, v: freqs);
394
395	cpufreq_stats_record_transition(policy, new_freq: freqs->new);
396	policy->cur = freqs->new;
397	}
398	}
399
400	/* Do post notifications when there are chances that transition has failed */
401	static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
402	struct cpufreq_freqs *freqs, int transition_failed)
403	{
404	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
405	if (!transition_failed)
406	return;
407
408	swap(freqs->old, freqs->new);
409	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
410	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
411	}
412
413	void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
414	struct cpufreq_freqs *freqs)
415	{
416
417	/*
418	* Catch double invocations of _begin() which lead to self-deadlock.
419	* ASYNC_NOTIFICATION drivers are left out because the cpufreq core
420	* doesn't invoke _begin() on their behalf, and hence the chances of
421	* double invocations are very low. Moreover, there are scenarios
422	* where these checks can emit false-positive warnings in these
423	* drivers; so we avoid that by skipping them altogether.
424	*/
425	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
426	&& current == policy->transition_task);
427
428	wait:
429	wait_event(policy->transition_wait, !policy->transition_ongoing);
430
431	spin_lock(lock: &policy->transition_lock);
432
433	if (unlikely(policy->transition_ongoing)) {
434	spin_unlock(lock: &policy->transition_lock);
435	goto wait;
436	}
437
438	policy->transition_ongoing = true;
439	policy->transition_task = current;
440
441	spin_unlock(lock: &policy->transition_lock);
442
443	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
444	}
445	EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
446
447	void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
448	struct cpufreq_freqs *freqs, int transition_failed)
449	{
450	if (WARN_ON(!policy->transition_ongoing))
451	return;
452
453	cpufreq_notify_post_transition(policy, freqs, transition_failed);
454
455	arch_set_freq_scale(cpus: policy->related_cpus,
456	cur_freq: policy->cur,
457	max_freq: arch_scale_freq_ref(cpu: policy->cpu));
458
459	spin_lock(lock: &policy->transition_lock);
460	policy->transition_ongoing = false;
461	policy->transition_task = NULL;
462	spin_unlock(lock: &policy->transition_lock);
463
464	wake_up(&policy->transition_wait);
465	}
466	EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
467
468	/*
469	* Fast frequency switching status count. Positive means "enabled", negative
470	* means "disabled" and 0 means "not decided yet".
471	*/
472	static int cpufreq_fast_switch_count;
473	static DEFINE_MUTEX(cpufreq_fast_switch_lock);
474
475	static void cpufreq_list_transition_notifiers(void)
476	{
477	struct notifier_block *nb;
478
479	pr_info("Registered transition notifiers:\n");
480
481	mutex_lock(&cpufreq_transition_notifier_list.mutex);
482
483	for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
484	pr_info("%pS\n", nb->notifier_call);
485
486	mutex_unlock(lock: &cpufreq_transition_notifier_list.mutex);
487	}
488
489	/**
490	* cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
491	* @policy: cpufreq policy to enable fast frequency switching for.
492	*
493	* Try to enable fast frequency switching for @policy.
494	*
495	* The attempt will fail if there is at least one transition notifier registered
496	* at this point, as fast frequency switching is quite fundamentally at odds
497	* with transition notifiers. Thus if successful, it will make registration of
498	* transition notifiers fail going forward.
499	*/
500	void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
501	{
502	lockdep_assert_held(&policy->rwsem);
503
504	if (!policy->fast_switch_possible)
505	return;
506
507	mutex_lock(&cpufreq_fast_switch_lock);
508	if (cpufreq_fast_switch_count >= 0) {
509	cpufreq_fast_switch_count++;
510	policy->fast_switch_enabled = true;
511	} else {
512	pr_warn("CPU%u: Fast frequency switching not enabled\n",
513	policy->cpu);
514	cpufreq_list_transition_notifiers();
515	}
516	mutex_unlock(lock: &cpufreq_fast_switch_lock);
517	}
518	EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
519
520	/**
521	* cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
522	* @policy: cpufreq policy to disable fast frequency switching for.
523	*/
524	void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
525	{
526	mutex_lock(&cpufreq_fast_switch_lock);
527	if (policy->fast_switch_enabled) {
528	policy->fast_switch_enabled = false;
529	if (!WARN_ON(cpufreq_fast_switch_count <= 0))
530	cpufreq_fast_switch_count--;
531	}
532	mutex_unlock(lock: &cpufreq_fast_switch_lock);
533	}
534	EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
535
536	static unsigned int __resolve_freq(struct cpufreq_policy *policy,
537	unsigned int target_freq, unsigned int relation)
538	{
539	unsigned int idx;
540
541	target_freq = clamp_val(target_freq, policy->min, policy->max);
542
543	if (!policy->freq_table)
544	return target_freq;
545
546	idx = cpufreq_frequency_table_target(policy, target_freq, relation);
547	policy->cached_resolved_idx = idx;
548	policy->cached_target_freq = target_freq;
549	return policy->freq_table[idx].frequency;
550	}
551
552	/**
553	* cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
554	* one.
555	* @policy: associated policy to interrogate
556	* @target_freq: target frequency to resolve.
557	*
558	* The target to driver frequency mapping is cached in the policy.
559	*
560	* Return: Lowest driver-supported frequency greater than or equal to the
561	* given target_freq, subject to policy (min/max) and driver limitations.
562	*/
563	unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
564	unsigned int target_freq)
565	{
566	return __resolve_freq(policy, target_freq, CPUFREQ_RELATION_LE);
567	}
568	EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
569
570	unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
571	{
572	unsigned int latency;
573
574	if (policy->transition_delay_us)
575	return policy->transition_delay_us;
576
577	latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
578	if (latency) {
579	unsigned int max_delay_us = 2 * MSEC_PER_SEC;
580
581	/*
582	* If the platform already has high transition_latency, use it
583	* as-is.
584	*/
585	if (latency > max_delay_us)
586	return latency;
587
588	/*
589	* For platforms that can change the frequency very fast (< 2
590	* us), the above formula gives a decent transition delay. But
591	* for platforms where transition_latency is in milliseconds, it
592	* ends up giving unrealistic values.
593	*
594	* Cap the default transition delay to 2 ms, which seems to be
595	* a reasonable amount of time after which we should reevaluate
596	* the frequency.
597	*/
598	return min(latency * LATENCY_MULTIPLIER, max_delay_us);
599	}
600
601	return LATENCY_MULTIPLIER;
602	}
603	EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
604
605	/*********************************************************************
606	* SYSFS INTERFACE *
607	*********************************************************************/
608	static ssize_t show_boost(struct kobject *kobj,
609	struct kobj_attribute *attr, char *buf)
610	{
611	return sprintf(buf, fmt: "%d\n", cpufreq_driver->boost_enabled);
612	}
613
614	static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
615	const char *buf, size_t count)
616	{
617	int ret, enable;
618
619	ret = sscanf(buf, "%d", &enable);
620	if (ret != 1 || enable < 0 || enable > 1)
621	return -EINVAL;
622
623	if (cpufreq_boost_trigger_state(state: enable)) {
624	pr_err("%s: Cannot %s BOOST!\n",
625	__func__, enable ? "enable" : "disable");
626	return -EINVAL;
627	}
628
629	pr_debug("%s: cpufreq BOOST %s\n",
630	__func__, enable ? "enabled" : "disabled");
631
632	return count;
633	}
634	define_one_global_rw(boost);
635
636	static ssize_t show_local_boost(struct cpufreq_policy *policy, char *buf)
637	{
638	return sysfs_emit(buf, fmt: "%d\n", policy->boost_enabled);
639	}
640
641	static ssize_t store_local_boost(struct cpufreq_policy *policy,
642	const char *buf, size_t count)
643	{
644	int ret, enable;
645
646	ret = kstrtoint(s: buf, base: 10, res: &enable);
647	if (ret || enable < 0 || enable > 1)
648	return -EINVAL;
649
650	if (!cpufreq_driver->boost_enabled)
651	return -EINVAL;
652
653	if (policy->boost_enabled == enable)
654	return count;
655
656	policy->boost_enabled = enable;
657
658	cpus_read_lock();
659	ret = cpufreq_driver->set_boost(policy, enable);
660	cpus_read_unlock();
661
662	if (ret) {
663	policy->boost_enabled = !policy->boost_enabled;
664	return ret;
665	}
666
667	return count;
668	}
669
670	static struct freq_attr local_boost = __ATTR(boost, 0644, show_local_boost, store_local_boost);
671
672	static struct cpufreq_governor *find_governor(const char *str_governor)
673	{
674	struct cpufreq_governor *t;
675
676	for_each_governor(t)
677	if (!strncasecmp(s1: str_governor, s2: t->name, CPUFREQ_NAME_LEN))
678	return t;
679
680	return NULL;
681	}
682
683	static struct cpufreq_governor *get_governor(const char *str_governor)
684	{
685	struct cpufreq_governor *t;
686
687	mutex_lock(&cpufreq_governor_mutex);
688	t = find_governor(str_governor);
689	if (!t)
690	goto unlock;
691
692	if (!try_module_get(module: t->owner))
693	t = NULL;
694
695	unlock:
696	mutex_unlock(lock: &cpufreq_governor_mutex);
697
698	return t;
699	}
700
701	static unsigned int cpufreq_parse_policy(char *str_governor)
702	{
703	if (!strncasecmp(s1: str_governor, s2: "performance", CPUFREQ_NAME_LEN))
704	return CPUFREQ_POLICY_PERFORMANCE;
705
706	if (!strncasecmp(s1: str_governor, s2: "powersave", CPUFREQ_NAME_LEN))
707	return CPUFREQ_POLICY_POWERSAVE;
708
709	return CPUFREQ_POLICY_UNKNOWN;
710	}
711
712	/**
713	* cpufreq_parse_governor - parse a governor string only for has_target()
714	* @str_governor: Governor name.
715	*/
716	static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
717	{
718	struct cpufreq_governor *t;
719
720	t = get_governor(str_governor);
721	if (t)
722	return t;
723
724	if (request_module("cpufreq_%s", str_governor))
725	return NULL;
726
727	return get_governor(str_governor);
728	}
729
730	/*
731	* cpufreq_per_cpu_attr_read() / show_##file_name() -
732	* print out cpufreq information
733	*
734	* Write out information from cpufreq_driver->policy[cpu]; object must be
735	* "unsigned int".
736	*/
737
738	#define show_one(file_name, object) \
739	static ssize_t show_##file_name \
740	(struct cpufreq_policy *policy, char *buf) \
741	{ \
742	return sprintf(buf, "%u\n", policy->object); \
743	}
744
745	show_one(cpuinfo_min_freq, cpuinfo.min_freq);
746	show_one(cpuinfo_max_freq, cpuinfo.max_freq);
747	show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
748	show_one(scaling_min_freq, min);
749	show_one(scaling_max_freq, max);
750
751	__weak unsigned int arch_freq_get_on_cpu(int cpu)
752	{
753	return 0;
754	}
755
756	static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
757	{
758	ssize_t ret;
759	unsigned int freq;
760
761	freq = arch_freq_get_on_cpu(cpu: policy->cpu);
762	if (freq)
763	ret = sprintf(buf, fmt: "%u\n", freq);
764	else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
765	ret = sprintf(buf, fmt: "%u\n", cpufreq_driver->get(policy->cpu));
766	else
767	ret = sprintf(buf, fmt: "%u\n", policy->cur);
768	return ret;
769	}
770
771	/*
772	* cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
773	*/
774	#define store_one(file_name, object) \
775	static ssize_t store_##file_name \
776	(struct cpufreq_policy *policy, const char *buf, size_t count) \
777	{ \
778	unsigned long val; \
779	int ret; \
780	\
781	ret = kstrtoul(buf, 0, &val); \
782	if (ret) \
783	return ret; \
784	\
785	ret = freq_qos_update_request(policy->object##_freq_req, val);\
786	return ret >= 0 ? count : ret; \
787	}
788
789	store_one(scaling_min_freq, min);
790	store_one(scaling_max_freq, max);
791
792	/*
793	* show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
794	*/
795	static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
796	char *buf)
797	{
798	unsigned int cur_freq = __cpufreq_get(policy);
799
800	if (cur_freq)
801	return sprintf(buf, fmt: "%u\n", cur_freq);
802
803	return sprintf(buf, fmt: "<unknown>\n");
804	}
805
806	/*
807	* show_scaling_governor - show the current policy for the specified CPU
808	*/
809	static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
810	{
811	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
812	return sprintf(buf, fmt: "powersave\n");
813	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
814	return sprintf(buf, fmt: "performance\n");
815	else if (policy->governor)
816	return scnprintf(buf, CPUFREQ_NAME_PLEN, fmt: "%s\n",
817	policy->governor->name);
818	return -EINVAL;
819	}
820
821	/*
822	* store_scaling_governor - store policy for the specified CPU
823	*/
824	static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
825	const char *buf, size_t count)
826	{
827	char str_governor[16];
828	int ret;
829
830	ret = sscanf(buf, "%15s", str_governor);
831	if (ret != 1)
832	return -EINVAL;
833
834	if (cpufreq_driver->setpolicy) {
835	unsigned int new_pol;
836
837	new_pol = cpufreq_parse_policy(str_governor);
838	if (!new_pol)
839	return -EINVAL;
840
841	ret = cpufreq_set_policy(policy, NULL, new_pol);
842	} else {
843	struct cpufreq_governor *new_gov;
844
845	new_gov = cpufreq_parse_governor(str_governor);
846	if (!new_gov)
847	return -EINVAL;
848
849	ret = cpufreq_set_policy(policy, new_gov,
850	CPUFREQ_POLICY_UNKNOWN);
851
852	module_put(module: new_gov->owner);
853	}
854
855	return ret ? ret : count;
856	}
857
858	/*
859	* show_scaling_driver - show the cpufreq driver currently loaded
860	*/
861	static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
862	{
863	return scnprintf(buf, CPUFREQ_NAME_PLEN, fmt: "%s\n", cpufreq_driver->name);
864	}
865
866	/*
867	* show_scaling_available_governors - show the available CPUfreq governors
868	*/
869	static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
870	char *buf)
871	{
872	ssize_t i = 0;
873	struct cpufreq_governor *t;
874
875	if (!has_target()) {
876	i += sprintf(buf, fmt: "performance powersave");
877	goto out;
878	}
879
880	mutex_lock(&cpufreq_governor_mutex);
881	for_each_governor(t) {
882	if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
883	- (CPUFREQ_NAME_LEN + 2)))
884	break;
885	i += scnprintf(buf: &buf[i], CPUFREQ_NAME_PLEN, fmt: "%s ", t->name);
886	}
887	mutex_unlock(lock: &cpufreq_governor_mutex);
888	out:
889	i += sprintf(buf: &buf[i], fmt: "\n");
890	return i;
891	}
892
893	ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
894	{
895	ssize_t i = 0;
896	unsigned int cpu;
897
898	for_each_cpu(cpu, mask) {
899	i += scnprintf(buf: &buf[i], size: (PAGE_SIZE - i - 2), fmt: "%u ", cpu);
900	if (i >= (PAGE_SIZE - 5))
901	break;
902	}
903
904	/* Remove the extra space at the end */
905	i--;
906
907	i += sprintf(buf: &buf[i], fmt: "\n");
908	return i;
909	}
910	EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
911
912	/*
913	* show_related_cpus - show the CPUs affected by each transition even if
914	* hw coordination is in use
915	*/
916	static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
917	{
918	return cpufreq_show_cpus(policy->related_cpus, buf);
919	}
920
921	/*
922	* show_affected_cpus - show the CPUs affected by each transition
923	*/
924	static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
925	{
926	return cpufreq_show_cpus(policy->cpus, buf);
927	}
928
929	static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
930	const char *buf, size_t count)
931	{
932	unsigned int freq = 0;
933	unsigned int ret;
934
935	if (!policy->governor || !policy->governor->store_setspeed)
936	return -EINVAL;
937
938	ret = sscanf(buf, "%u", &freq);
939	if (ret != 1)
940	return -EINVAL;
941
942	policy->governor->store_setspeed(policy, freq);
943
944	return count;
945	}
946
947	static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
948	{
949	if (!policy->governor || !policy->governor->show_setspeed)
950	return sprintf(buf, fmt: "<unsupported>\n");
951
952	return policy->governor->show_setspeed(policy, buf);
953	}
954
955	/*
956	* show_bios_limit - show the current cpufreq HW/BIOS limitation
957	*/
958	static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
959	{
960	unsigned int limit;
961	int ret;
962	ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
963	if (!ret)
964	return sprintf(buf, fmt: "%u\n", limit);
965	return sprintf(buf, fmt: "%u\n", policy->cpuinfo.max_freq);
966	}
967
968	cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
969	cpufreq_freq_attr_ro(cpuinfo_min_freq);
970	cpufreq_freq_attr_ro(cpuinfo_max_freq);
971	cpufreq_freq_attr_ro(cpuinfo_transition_latency);
972	cpufreq_freq_attr_ro(scaling_available_governors);
973	cpufreq_freq_attr_ro(scaling_driver);
974	cpufreq_freq_attr_ro(scaling_cur_freq);
975	cpufreq_freq_attr_ro(bios_limit);
976	cpufreq_freq_attr_ro(related_cpus);
977	cpufreq_freq_attr_ro(affected_cpus);
978	cpufreq_freq_attr_rw(scaling_min_freq);
979	cpufreq_freq_attr_rw(scaling_max_freq);
980	cpufreq_freq_attr_rw(scaling_governor);
981	cpufreq_freq_attr_rw(scaling_setspeed);
982
983	static struct attribute *cpufreq_attrs[] = {
984	&cpuinfo_min_freq.attr,
985	&cpuinfo_max_freq.attr,
986	&cpuinfo_transition_latency.attr,
987	&scaling_min_freq.attr,
988	&scaling_max_freq.attr,
989	&affected_cpus.attr,
990	&related_cpus.attr,
991	&scaling_governor.attr,
992	&scaling_driver.attr,
993	&scaling_available_governors.attr,
994	&scaling_setspeed.attr,
995	NULL
996	};
997	ATTRIBUTE_GROUPS(cpufreq);
998
999	#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
1000	#define to_attr(a) container_of(a, struct freq_attr, attr)
1001
1002	static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1003	{
1004	struct cpufreq_policy *policy = to_policy(kobj);
1005	struct freq_attr *fattr = to_attr(attr);
1006	ssize_t ret = -EBUSY;
1007
1008	if (!fattr->show)
1009	return -EIO;
1010
1011	down_read(sem: &policy->rwsem);
1012	if (likely(!policy_is_inactive(policy)))
1013	ret = fattr->show(policy, buf);
1014	up_read(sem: &policy->rwsem);
1015
1016	return ret;
1017	}
1018
1019	static ssize_t store(struct kobject *kobj, struct attribute *attr,
1020	const char *buf, size_t count)
1021	{
1022	struct cpufreq_policy *policy = to_policy(kobj);
1023	struct freq_attr *fattr = to_attr(attr);
1024	ssize_t ret = -EBUSY;
1025
1026	if (!fattr->store)
1027	return -EIO;
1028
1029	down_write(sem: &policy->rwsem);
1030	if (likely(!policy_is_inactive(policy)))
1031	ret = fattr->store(policy, buf, count);
1032	up_write(sem: &policy->rwsem);
1033
1034	return ret;
1035	}
1036
1037	static void cpufreq_sysfs_release(struct kobject *kobj)
1038	{
1039	struct cpufreq_policy *policy = to_policy(kobj);
1040	pr_debug("last reference is dropped\n");
1041	complete(&policy->kobj_unregister);
1042	}
1043
1044	static const struct sysfs_ops sysfs_ops = {
1045	.show = show,
1046	.store = store,
1047	};
1048
1049	static const struct kobj_type ktype_cpufreq = {
1050	.sysfs_ops = &sysfs_ops,
1051	.default_groups = cpufreq_groups,
1052	.release = cpufreq_sysfs_release,
1053	};
1054
1055	static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
1056	struct device *dev)
1057	{
1058	if (unlikely(!dev))
1059	return;
1060
1061	if (cpumask_test_and_set_cpu(cpu, cpumask: policy->real_cpus))
1062	return;
1063
1064	dev_dbg(dev, "%s: Adding symlink\n", __func__);
1065	if (sysfs_create_link(kobj: &dev->kobj, target: &policy->kobj, name: "cpufreq"))
1066	dev_err(dev, "cpufreq symlink creation failed\n");
1067	}
1068
1069	static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu,
1070	struct device *dev)
1071	{
1072	dev_dbg(dev, "%s: Removing symlink\n", __func__);
1073	sysfs_remove_link(kobj: &dev->kobj, name: "cpufreq");
1074	cpumask_clear_cpu(cpu, dstp: policy->real_cpus);
1075	}
1076
1077	static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1078	{
1079	struct freq_attr **drv_attr;
1080	int ret = 0;
1081
1082	/* set up files for this cpu device */
1083	drv_attr = cpufreq_driver->attr;
1084	while (drv_attr && *drv_attr) {
1085	ret = sysfs_create_file(kobj: &policy->kobj, attr: &((*drv_attr)->attr));
1086	if (ret)
1087	return ret;
1088	drv_attr++;
1089	}
1090	if (cpufreq_driver->get) {
1091	ret = sysfs_create_file(kobj: &policy->kobj, attr: &cpuinfo_cur_freq.attr);
1092	if (ret)
1093	return ret;
1094	}
1095
1096	ret = sysfs_create_file(kobj: &policy->kobj, attr: &scaling_cur_freq.attr);
1097	if (ret)
1098	return ret;
1099
1100	if (cpufreq_driver->bios_limit) {
1101	ret = sysfs_create_file(kobj: &policy->kobj, attr: &bios_limit.attr);
1102	if (ret)
1103	return ret;
1104	}
1105
1106	if (cpufreq_boost_supported()) {
1107	ret = sysfs_create_file(kobj: &policy->kobj, attr: &local_boost.attr);
1108	if (ret)
1109	return ret;
1110	}
1111
1112	return 0;
1113	}
1114
1115	static int cpufreq_init_policy(struct cpufreq_policy *policy)
1116	{
1117	struct cpufreq_governor *gov = NULL;
1118	unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1119	int ret;
1120
1121	if (has_target()) {
1122	/* Update policy governor to the one used before hotplug. */
1123	gov = get_governor(str_governor: policy->last_governor);
1124	if (gov) {
1125	pr_debug("Restoring governor %s for cpu %d\n",
1126	gov->name, policy->cpu);
1127	} else {
1128	gov = get_governor(str_governor: default_governor);
1129	}
1130
1131	if (!gov) {
1132	gov = cpufreq_default_governor();
1133	__module_get(module: gov->owner);
1134	}
1135
1136	} else {
1137
1138	/* Use the default policy if there is no last_policy. */
1139	if (policy->last_policy) {
1140	pol = policy->last_policy;
1141	} else {
1142	pol = cpufreq_parse_policy(str_governor: default_governor);
1143	/*
1144	* In case the default governor is neither "performance"
1145	* nor "powersave", fall back to the initial policy
1146	* value set by the driver.
1147	*/
1148	if (pol == CPUFREQ_POLICY_UNKNOWN)
1149	pol = policy->policy;
1150	}
1151	if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1152	pol != CPUFREQ_POLICY_POWERSAVE)
1153	return -ENODATA;
1154	}
1155
1156	ret = cpufreq_set_policy(policy, new_gov: gov, new_pol: pol);
1157	if (gov)
1158	module_put(module: gov->owner);
1159
1160	return ret;
1161	}
1162
1163	static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1164	{
1165	int ret = 0;
1166
1167	/* Has this CPU been taken care of already? */
1168	if (cpumask_test_cpu(cpu, cpumask: policy->cpus))
1169	return 0;
1170
1171	down_write(sem: &policy->rwsem);
1172	if (has_target())
1173	cpufreq_stop_governor(policy);
1174
1175	cpumask_set_cpu(cpu, dstp: policy->cpus);
1176
1177	if (has_target()) {
1178	ret = cpufreq_start_governor(policy);
1179	if (ret)
1180	pr_err("%s: Failed to start governor\n", __func__);
1181	}
1182	up_write(sem: &policy->rwsem);
1183	return ret;
1184	}
1185
1186	void refresh_frequency_limits(struct cpufreq_policy *policy)
1187	{
1188	if (!policy_is_inactive(policy)) {
1189	pr_debug("updating policy for CPU %u\n", policy->cpu);
1190
1191	cpufreq_set_policy(policy, new_gov: policy->governor, new_pol: policy->policy);
1192	}
1193	}
1194	EXPORT_SYMBOL(refresh_frequency_limits);
1195
1196	static void handle_update(struct work_struct *work)
1197	{
1198	struct cpufreq_policy *policy =
1199	container_of(work, struct cpufreq_policy, update);
1200
1201	pr_debug("handle_update for cpu %u called\n", policy->cpu);
1202	down_write(sem: &policy->rwsem);
1203	refresh_frequency_limits(policy);
1204	up_write(sem: &policy->rwsem);
1205	}
1206
1207	static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1208	void *data)
1209	{
1210	struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1211
1212	schedule_work(work: &policy->update);
1213	return 0;
1214	}
1215
1216	static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1217	void *data)
1218	{
1219	struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1220
1221	schedule_work(work: &policy->update);
1222	return 0;
1223	}
1224
1225	static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1226	{
1227	struct kobject *kobj;
1228	struct completion *cmp;
1229
1230	down_write(sem: &policy->rwsem);
1231	cpufreq_stats_free_table(policy);
1232	kobj = &policy->kobj;
1233	cmp = &policy->kobj_unregister;
1234	up_write(sem: &policy->rwsem);
1235	kobject_put(kobj);
1236
1237	/*
1238	* We need to make sure that the underlying kobj is
1239	* actually not referenced anymore by anybody before we
1240	* proceed with unloading.
1241	*/
1242	pr_debug("waiting for dropping of refcount\n");
1243	wait_for_completion(cmp);
1244	pr_debug("wait complete\n");
1245	}
1246
1247	static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1248	{
1249	struct cpufreq_policy *policy;
1250	struct device *dev = get_cpu_device(cpu);
1251	int ret;
1252
1253	if (!dev)
1254	return NULL;
1255
1256	policy = kzalloc(size: sizeof(*policy), GFP_KERNEL);
1257	if (!policy)
1258	return NULL;
1259
1260	if (!alloc_cpumask_var(mask: &policy->cpus, GFP_KERNEL))
1261	goto err_free_policy;
1262
1263	if (!zalloc_cpumask_var(mask: &policy->related_cpus, GFP_KERNEL))
1264	goto err_free_cpumask;
1265
1266	if (!zalloc_cpumask_var(mask: &policy->real_cpus, GFP_KERNEL))
1267	goto err_free_rcpumask;
1268
1269	init_completion(x: &policy->kobj_unregister);
1270	ret = kobject_init_and_add(kobj: &policy->kobj, ktype: &ktype_cpufreq,
1271	parent: cpufreq_global_kobject, fmt: "policy%u", cpu);
1272	if (ret) {
1273	dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1274	/*
1275	* The entire policy object will be freed below, but the extra
1276	* memory allocated for the kobject name needs to be freed by
1277	* releasing the kobject.
1278	*/
1279	kobject_put(kobj: &policy->kobj);
1280	goto err_free_real_cpus;
1281	}
1282
1283	freq_constraints_init(qos: &policy->constraints);
1284
1285	policy->nb_min.notifier_call = cpufreq_notifier_min;
1286	policy->nb_max.notifier_call = cpufreq_notifier_max;
1287
1288	ret = freq_qos_add_notifier(qos: &policy->constraints, type: FREQ_QOS_MIN,
1289	notifier: &policy->nb_min);
1290	if (ret) {
1291	dev_err(dev, "Failed to register MIN QoS notifier: %d (CPU%u)\n",
1292	ret, cpu);
1293	goto err_kobj_remove;
1294	}
1295
1296	ret = freq_qos_add_notifier(qos: &policy->constraints, type: FREQ_QOS_MAX,
1297	notifier: &policy->nb_max);
1298	if (ret) {
1299	dev_err(dev, "Failed to register MAX QoS notifier: %d (CPU%u)\n",
1300	ret, cpu);
1301	goto err_min_qos_notifier;
1302	}
1303
1304	INIT_LIST_HEAD(list: &policy->policy_list);
1305	init_rwsem(&policy->rwsem);
1306	spin_lock_init(&policy->transition_lock);
1307	init_waitqueue_head(&policy->transition_wait);
1308	INIT_WORK(&policy->update, handle_update);
1309
1310	policy->cpu = cpu;
1311	return policy;
1312
1313	err_min_qos_notifier:
1314	freq_qos_remove_notifier(qos: &policy->constraints, type: FREQ_QOS_MIN,
1315	notifier: &policy->nb_min);
1316	err_kobj_remove:
1317	cpufreq_policy_put_kobj(policy);
1318	err_free_real_cpus:
1319	free_cpumask_var(mask: policy->real_cpus);
1320	err_free_rcpumask:
1321	free_cpumask_var(mask: policy->related_cpus);
1322	err_free_cpumask:
1323	free_cpumask_var(mask: policy->cpus);
1324	err_free_policy:
1325	kfree(objp: policy);
1326
1327	return NULL;
1328	}
1329
1330	static void cpufreq_policy_free(struct cpufreq_policy *policy)
1331	{
1332	unsigned long flags;
1333	int cpu;
1334
1335	/*
1336	* The callers must ensure the policy is inactive by now, to avoid any
1337	* races with show()/store() callbacks.
1338	*/
1339	if (unlikely(!policy_is_inactive(policy)))
1340	pr_warn("%s: Freeing active policy\n", __func__);
1341
1342	/* Remove policy from list */
1343	write_lock_irqsave(&cpufreq_driver_lock, flags);
1344	list_del(entry: &policy->policy_list);
1345
1346	for_each_cpu(cpu, policy->related_cpus)
1347	per_cpu(cpufreq_cpu_data, cpu) = NULL;
1348	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1349
1350	freq_qos_remove_notifier(qos: &policy->constraints, type: FREQ_QOS_MAX,
1351	notifier: &policy->nb_max);
1352	freq_qos_remove_notifier(qos: &policy->constraints, type: FREQ_QOS_MIN,
1353	notifier: &policy->nb_min);
1354
1355	/* Cancel any pending policy->update work before freeing the policy. */
1356	cancel_work_sync(work: &policy->update);
1357
1358	if (policy->max_freq_req) {
1359	/*
1360	* Remove max_freq_req after sending CPUFREQ_REMOVE_POLICY
1361	* notification, since CPUFREQ_CREATE_POLICY notification was
1362	* sent after adding max_freq_req earlier.
1363	*/
1364	blocking_notifier_call_chain(nh: &cpufreq_policy_notifier_list,
1365	CPUFREQ_REMOVE_POLICY, v: policy);
1366	freq_qos_remove_request(req: policy->max_freq_req);
1367	}
1368
1369	freq_qos_remove_request(req: policy->min_freq_req);
1370	kfree(objp: policy->min_freq_req);
1371
1372	cpufreq_policy_put_kobj(policy);
1373	free_cpumask_var(mask: policy->real_cpus);
1374	free_cpumask_var(mask: policy->related_cpus);
1375	free_cpumask_var(mask: policy->cpus);
1376	kfree(objp: policy);
1377	}
1378
1379	static int cpufreq_online(unsigned int cpu)
1380	{
1381	struct cpufreq_policy *policy;
1382	bool new_policy;
1383	unsigned long flags;
1384	unsigned int j;
1385	int ret;
1386
1387	pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1388
1389	/* Check if this CPU already has a policy to manage it */
1390	policy = per_cpu(cpufreq_cpu_data, cpu);
1391	if (policy) {
1392	WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1393	if (!policy_is_inactive(policy))
1394	return cpufreq_add_policy_cpu(policy, cpu);
1395
1396	/* This is the only online CPU for the policy. Start over. */
1397	new_policy = false;
1398	down_write(sem: &policy->rwsem);
1399	policy->cpu = cpu;
1400	policy->governor = NULL;
1401	} else {
1402	new_policy = true;
1403	policy = cpufreq_policy_alloc(cpu);
1404	if (!policy)
1405	return -ENOMEM;
1406	down_write(sem: &policy->rwsem);
1407	}
1408
1409	if (!new_policy && cpufreq_driver->online) {
1410	/* Recover policy->cpus using related_cpus */
1411	cpumask_copy(dstp: policy->cpus, srcp: policy->related_cpus);
1412
1413	ret = cpufreq_driver->online(policy);
1414	if (ret) {
1415	pr_debug("%s: %d: initialization failed\n", __func__,
1416	__LINE__);
1417	goto out_exit_policy;
1418	}
1419	} else {
1420	cpumask_copy(dstp: policy->cpus, cpumask_of(cpu));
1421
1422	/*
1423	* Call driver. From then on the cpufreq must be able
1424	* to accept all calls to ->verify and ->setpolicy for this CPU.
1425	*/
1426	ret = cpufreq_driver->init(policy);
1427	if (ret) {
1428	pr_debug("%s: %d: initialization failed\n", __func__,
1429	__LINE__);
1430	goto out_free_policy;
1431	}
1432
1433	/* Let the per-policy boost flag mirror the cpufreq_driver boost during init */
1434	policy->boost_enabled = cpufreq_boost_enabled() && policy_has_boost_freq(policy);
1435
1436	/*
1437	* The initialization has succeeded and the policy is online.
1438	* If there is a problem with its frequency table, take it
1439	* offline and drop it.
1440	*/
1441	ret = cpufreq_table_validate_and_sort(policy);
1442	if (ret)
1443	goto out_offline_policy;
1444
1445	/* related_cpus should at least include policy->cpus. */
1446	cpumask_copy(dstp: policy->related_cpus, srcp: policy->cpus);
1447	}
1448
1449	/*
1450	* affected cpus must always be the one, which are online. We aren't
1451	* managing offline cpus here.
1452	*/
1453	cpumask_and(dstp: policy->cpus, src1p: policy->cpus, cpu_online_mask);
1454
1455	if (new_policy) {
1456	for_each_cpu(j, policy->related_cpus) {
1457	per_cpu(cpufreq_cpu_data, j) = policy;
1458	add_cpu_dev_symlink(policy, cpu: j, dev: get_cpu_device(cpu: j));
1459	}
1460
1461	policy->min_freq_req = kzalloc(size: 2 * sizeof(*policy->min_freq_req),
1462	GFP_KERNEL);
1463	if (!policy->min_freq_req) {
1464	ret = -ENOMEM;
1465	goto out_destroy_policy;
1466	}
1467
1468	ret = freq_qos_add_request(qos: &policy->constraints,
1469	req: policy->min_freq_req, type: FREQ_QOS_MIN,
1470	FREQ_QOS_MIN_DEFAULT_VALUE);
1471	if (ret < 0) {
1472	/*
1473	* So we don't call freq_qos_remove_request() for an
1474	* uninitialized request.
1475	*/
1476	kfree(objp: policy->min_freq_req);
1477	policy->min_freq_req = NULL;
1478	goto out_destroy_policy;
1479	}
1480
1481	/*
1482	* This must be initialized right here to avoid calling
1483	* freq_qos_remove_request() on uninitialized request in case
1484	* of errors.
1485	*/
1486	policy->max_freq_req = policy->min_freq_req + 1;
1487
1488	ret = freq_qos_add_request(qos: &policy->constraints,
1489	req: policy->max_freq_req, type: FREQ_QOS_MAX,
1490	FREQ_QOS_MAX_DEFAULT_VALUE);
1491	if (ret < 0) {
1492	policy->max_freq_req = NULL;
1493	goto out_destroy_policy;
1494	}
1495
1496	blocking_notifier_call_chain(nh: &cpufreq_policy_notifier_list,
1497	CPUFREQ_CREATE_POLICY, v: policy);
1498	}
1499
1500	if (cpufreq_driver->get && has_target()) {
1501	policy->cur = cpufreq_driver->get(policy->cpu);
1502	if (!policy->cur) {
1503	ret = -EIO;
1504	pr_err("%s: ->get() failed\n", __func__);
1505	goto out_destroy_policy;
1506	}
1507	}
1508
1509	/*
1510	* Sometimes boot loaders set CPU frequency to a value outside of
1511	* frequency table present with cpufreq core. In such cases CPU might be
1512	* unstable if it has to run on that frequency for long duration of time
1513	* and so its better to set it to a frequency which is specified in
1514	* freq-table. This also makes cpufreq stats inconsistent as
1515	* cpufreq-stats would fail to register because current frequency of CPU
1516	* isn't found in freq-table.
1517	*
1518	* Because we don't want this change to effect boot process badly, we go
1519	* for the next freq which is >= policy->cur ('cur' must be set by now,
1520	* otherwise we will end up setting freq to lowest of the table as 'cur'
1521	* is initialized to zero).
1522	*
1523	* We are passing target-freq as "policy->cur - 1" otherwise
1524	* __cpufreq_driver_target() would simply fail, as policy->cur will be
1525	* equal to target-freq.
1526	*/
1527	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1528	&& has_target()) {
1529	unsigned int old_freq = policy->cur;
1530
1531	/* Are we running at unknown frequency ? */
1532	ret = cpufreq_frequency_table_get_index(policy, freq: old_freq);
1533	if (ret == -EINVAL) {
1534	ret = __cpufreq_driver_target(policy, target_freq: old_freq - 1,
1535	CPUFREQ_RELATION_L);
1536
1537	/*
1538	* Reaching here after boot in a few seconds may not
1539	* mean that system will remain stable at "unknown"
1540	* frequency for longer duration. Hence, a BUG_ON().
1541	*/
1542	BUG_ON(ret);
1543	pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
1544	__func__, policy->cpu, old_freq, policy->cur);
1545	}
1546	}
1547
1548	if (new_policy) {
1549	ret = cpufreq_add_dev_interface(policy);
1550	if (ret)
1551	goto out_destroy_policy;
1552
1553	cpufreq_stats_create_table(policy);
1554
1555	write_lock_irqsave(&cpufreq_driver_lock, flags);
1556	list_add(new: &policy->policy_list, head: &cpufreq_policy_list);
1557	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1558
1559	/*
1560	* Register with the energy model before
1561	* sugov_eas_rebuild_sd() is called, which will result
1562	* in rebuilding of the sched domains, which should only be done
1563	* once the energy model is properly initialized for the policy
1564	* first.
1565	*
1566	* Also, this should be called before the policy is registered
1567	* with cooling framework.
1568	*/
1569	if (cpufreq_driver->register_em)
1570	cpufreq_driver->register_em(policy);
1571	}
1572
1573	ret = cpufreq_init_policy(policy);
1574	if (ret) {
1575	pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1576	__func__, cpu, ret);
1577	goto out_destroy_policy;
1578	}
1579
1580	up_write(sem: &policy->rwsem);
1581
1582	kobject_uevent(kobj: &policy->kobj, action: KOBJ_ADD);
1583
1584	/* Callback for handling stuff after policy is ready */
1585	if (cpufreq_driver->ready)
1586	cpufreq_driver->ready(policy);
1587
1588	/* Register cpufreq cooling only for a new policy */
1589	if (new_policy && cpufreq_thermal_control_enabled(drv: cpufreq_driver))
1590	policy->cdev = of_cpufreq_cooling_register(policy);
1591
1592	pr_debug("initialization complete\n");
1593
1594	return 0;
1595
1596	out_destroy_policy:
1597	for_each_cpu(j, policy->real_cpus)
1598	remove_cpu_dev_symlink(policy, cpu: j, dev: get_cpu_device(cpu: j));
1599
1600	out_offline_policy:
1601	if (cpufreq_driver->offline)
1602	cpufreq_driver->offline(policy);
1603
1604	out_exit_policy:
1605	if (cpufreq_driver->exit)
1606	cpufreq_driver->exit(policy);
1607
1608	out_free_policy:
1609	cpumask_clear(dstp: policy->cpus);
1610	up_write(sem: &policy->rwsem);
1611
1612	cpufreq_policy_free(policy);
1613	return ret;
1614	}
1615
1616	/**
1617	* cpufreq_add_dev - the cpufreq interface for a CPU device.
1618	* @dev: CPU device.
1619	* @sif: Subsystem interface structure pointer (not used)
1620	*/
1621	static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1622	{
1623	struct cpufreq_policy *policy;
1624	unsigned cpu = dev->id;
1625	int ret;
1626
1627	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1628
1629	if (cpu_online(cpu)) {
1630	ret = cpufreq_online(cpu);
1631	if (ret)
1632	return ret;
1633	}
1634
1635	/* Create sysfs link on CPU registration */
1636	policy = per_cpu(cpufreq_cpu_data, cpu);
1637	if (policy)
1638	add_cpu_dev_symlink(policy, cpu, dev);
1639
1640	return 0;
1641	}
1642
1643	static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1644	{
1645	int ret;
1646
1647	if (has_target())
1648	cpufreq_stop_governor(policy);
1649
1650	cpumask_clear_cpu(cpu, dstp: policy->cpus);
1651
1652	if (!policy_is_inactive(policy)) {
1653	/* Nominate a new CPU if necessary. */
1654	if (cpu == policy->cpu)
1655	policy->cpu = cpumask_any(policy->cpus);
1656
1657	/* Start the governor again for the active policy. */
1658	if (has_target()) {
1659	ret = cpufreq_start_governor(policy);
1660	if (ret)
1661	pr_err("%s: Failed to start governor\n", __func__);
1662	}
1663
1664	return;
1665	}
1666
1667	if (has_target())
1668	strscpy(policy->last_governor, policy->governor->name,
1669	CPUFREQ_NAME_LEN);
1670	else
1671	policy->last_policy = policy->policy;
1672
1673	if (has_target())
1674	cpufreq_exit_governor(policy);
1675
1676	/*
1677	* Perform the ->offline() during light-weight tear-down, as
1678	* that allows fast recovery when the CPU comes back.
1679	*/
1680	if (cpufreq_driver->offline) {
1681	cpufreq_driver->offline(policy);
1682	} else if (cpufreq_driver->exit) {
1683	cpufreq_driver->exit(policy);
1684	policy->freq_table = NULL;
1685	}
1686	}
1687
1688	static int cpufreq_offline(unsigned int cpu)
1689	{
1690	struct cpufreq_policy *policy;
1691
1692	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1693
1694	policy = cpufreq_cpu_get_raw(cpu);
1695	if (!policy) {
1696	pr_debug("%s: No cpu_data found\n", __func__);
1697	return 0;
1698	}
1699
1700	down_write(sem: &policy->rwsem);
1701
1702	__cpufreq_offline(cpu, policy);
1703
1704	up_write(sem: &policy->rwsem);
1705	return 0;
1706	}
1707
1708	/*
1709	* cpufreq_remove_dev - remove a CPU device
1710	*
1711	* Removes the cpufreq interface for a CPU device.
1712	*/
1713	static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1714	{
1715	unsigned int cpu = dev->id;
1716	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1717
1718	if (!policy)
1719	return;
1720
1721	down_write(sem: &policy->rwsem);
1722
1723	if (cpu_online(cpu))
1724	__cpufreq_offline(cpu, policy);
1725
1726	remove_cpu_dev_symlink(policy, cpu, dev);
1727
1728	if (!cpumask_empty(srcp: policy->real_cpus)) {
1729	up_write(sem: &policy->rwsem);
1730	return;
1731	}
1732
1733	/*
1734	* Unregister cpufreq cooling once all the CPUs of the policy are
1735	* removed.
1736	*/
1737	if (cpufreq_thermal_control_enabled(drv: cpufreq_driver)) {
1738	cpufreq_cooling_unregister(cdev: policy->cdev);
1739	policy->cdev = NULL;
1740	}
1741
1742	/* We did light-weight exit earlier, do full tear down now */
1743	if (cpufreq_driver->offline)
1744	cpufreq_driver->exit(policy);
1745
1746	up_write(sem: &policy->rwsem);
1747
1748	cpufreq_policy_free(policy);
1749	}
1750
1751	/**
1752	* cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1753	* @policy: Policy managing CPUs.
1754	* @new_freq: New CPU frequency.
1755	*
1756	* Adjust to the current frequency first and clean up later by either calling
1757	* cpufreq_update_policy(), or scheduling handle_update().
1758	*/
1759	static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1760	unsigned int new_freq)
1761	{
1762	struct cpufreq_freqs freqs;
1763
1764	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1765	policy->cur, new_freq);
1766
1767	freqs.old = policy->cur;
1768	freqs.new = new_freq;
1769
1770	cpufreq_freq_transition_begin(policy, &freqs);
1771	cpufreq_freq_transition_end(policy, &freqs, 0);
1772	}
1773
1774	static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1775	{
1776	unsigned int new_freq;
1777
1778	new_freq = cpufreq_driver->get(policy->cpu);
1779	if (!new_freq)
1780	return 0;
1781
1782	/*
1783	* If fast frequency switching is used with the given policy, the check
1784	* against policy->cur is pointless, so skip it in that case.
1785	*/
1786	if (policy->fast_switch_enabled || !has_target())
1787	return new_freq;
1788
1789	if (policy->cur != new_freq) {
1790	/*
1791	* For some platforms, the frequency returned by hardware may be
1792	* slightly different from what is provided in the frequency
1793	* table, for example hardware may return 499 MHz instead of 500
1794	* MHz. In such cases it is better to avoid getting into
1795	* unnecessary frequency updates.
1796	*/
1797	if (abs(policy->cur - new_freq) < KHZ_PER_MHZ)
1798	return policy->cur;
1799
1800	cpufreq_out_of_sync(policy, new_freq);
1801	if (update)
1802	schedule_work(work: &policy->update);
1803	}
1804
1805	return new_freq;
1806	}
1807
1808	/**
1809	* cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1810	* @cpu: CPU number
1811	*
1812	* This is the last known freq, without actually getting it from the driver.
1813	* Return value will be same as what is shown in scaling_cur_freq in sysfs.
1814	*/
1815	unsigned int cpufreq_quick_get(unsigned int cpu)
1816	{
1817	struct cpufreq_policy *policy;
1818	unsigned int ret_freq = 0;
1819	unsigned long flags;
1820
1821	read_lock_irqsave(&cpufreq_driver_lock, flags);
1822
1823	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1824	ret_freq = cpufreq_driver->get(cpu);
1825	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1826	return ret_freq;
1827	}
1828
1829	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1830
1831	policy = cpufreq_cpu_get(cpu);
1832	if (policy) {
1833	ret_freq = policy->cur;
1834	cpufreq_cpu_put(policy);
1835	}
1836
1837	return ret_freq;
1838	}
1839	EXPORT_SYMBOL(cpufreq_quick_get);
1840
1841	/**
1842	* cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1843	* @cpu: CPU number
1844	*
1845	* Just return the max possible frequency for a given CPU.
1846	*/
1847	unsigned int cpufreq_quick_get_max(unsigned int cpu)
1848	{
1849	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1850	unsigned int ret_freq = 0;
1851
1852	if (policy) {
1853	ret_freq = policy->max;
1854	cpufreq_cpu_put(policy);
1855	}
1856
1857	return ret_freq;
1858	}
1859	EXPORT_SYMBOL(cpufreq_quick_get_max);
1860
1861	/**
1862	* cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1863	* @cpu: CPU number
1864	*
1865	* The default return value is the max_freq field of cpuinfo.
1866	*/
1867	__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1868	{
1869	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1870	unsigned int ret_freq = 0;
1871
1872	if (policy) {
1873	ret_freq = policy->cpuinfo.max_freq;
1874	cpufreq_cpu_put(policy);
1875	}
1876
1877	return ret_freq;
1878	}
1879	EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1880
1881	static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1882	{
1883	if (unlikely(policy_is_inactive(policy)))
1884	return 0;
1885
1886	return cpufreq_verify_current_freq(policy, update: true);
1887	}
1888
1889	/**
1890	* cpufreq_get - get the current CPU frequency (in kHz)
1891	* @cpu: CPU number
1892	*
1893	* Get the CPU current (static) CPU frequency
1894	*/
1895	unsigned int cpufreq_get(unsigned int cpu)
1896	{
1897	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1898	unsigned int ret_freq = 0;
1899
1900	if (policy) {
1901	down_read(sem: &policy->rwsem);
1902	if (cpufreq_driver->get)
1903	ret_freq = __cpufreq_get(policy);
1904	up_read(sem: &policy->rwsem);
1905
1906	cpufreq_cpu_put(policy);
1907	}
1908
1909	return ret_freq;
1910	}
1911	EXPORT_SYMBOL(cpufreq_get);
1912
1913	static struct subsys_interface cpufreq_interface = {
1914	.name = "cpufreq",
1915	.subsys = &cpu_subsys,
1916	.add_dev = cpufreq_add_dev,
1917	.remove_dev = cpufreq_remove_dev,
1918	};
1919
1920	/*
1921	* In case platform wants some specific frequency to be configured
1922	* during suspend..
1923	*/
1924	int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1925	{
1926	int ret;
1927
1928	if (!policy->suspend_freq) {
1929	pr_debug("%s: suspend_freq not defined\n", __func__);
1930	return 0;
1931	}
1932
1933	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1934	policy->suspend_freq);
1935
1936	ret = __cpufreq_driver_target(policy, target_freq: policy->suspend_freq,
1937	CPUFREQ_RELATION_H);
1938	if (ret)
1939	pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1940	__func__, policy->suspend_freq, ret);
1941
1942	return ret;
1943	}
1944	EXPORT_SYMBOL(cpufreq_generic_suspend);
1945
1946	/**
1947	* cpufreq_suspend() - Suspend CPUFreq governors.
1948	*
1949	* Called during system wide Suspend/Hibernate cycles for suspending governors
1950	* as some platforms can't change frequency after this point in suspend cycle.
1951	* Because some of the devices (like: i2c, regulators, etc) they use for
1952	* changing frequency are suspended quickly after this point.
1953	*/
1954	void cpufreq_suspend(void)
1955	{
1956	struct cpufreq_policy *policy;
1957
1958	if (!cpufreq_driver)
1959	return;
1960
1961	if (!has_target() && !cpufreq_driver->suspend)
1962	goto suspend;
1963
1964	pr_debug("%s: Suspending Governors\n", __func__);
1965
1966	for_each_active_policy(policy) {
1967	if (has_target()) {
1968	down_write(sem: &policy->rwsem);
1969	cpufreq_stop_governor(policy);
1970	up_write(sem: &policy->rwsem);
1971	}
1972
1973	if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1974	pr_err("%s: Failed to suspend driver: %s\n", __func__,
1975	cpufreq_driver->name);
1976	}
1977
1978	suspend:
1979	cpufreq_suspended = true;
1980	}
1981
1982	/**
1983	* cpufreq_resume() - Resume CPUFreq governors.
1984	*
1985	* Called during system wide Suspend/Hibernate cycle for resuming governors that
1986	* are suspended with cpufreq_suspend().
1987	*/
1988	void cpufreq_resume(void)
1989	{
1990	struct cpufreq_policy *policy;
1991	int ret;
1992
1993	if (!cpufreq_driver)
1994	return;
1995
1996	if (unlikely(!cpufreq_suspended))
1997	return;
1998
1999	cpufreq_suspended = false;
2000
2001	if (!has_target() && !cpufreq_driver->resume)
2002	return;
2003
2004	pr_debug("%s: Resuming Governors\n", __func__);
2005
2006	for_each_active_policy(policy) {
2007	if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
2008	pr_err("%s: Failed to resume driver: %s\n", __func__,
2009	cpufreq_driver->name);
2010	} else if (has_target()) {
2011	down_write(sem: &policy->rwsem);
2012	ret = cpufreq_start_governor(policy);
2013	up_write(sem: &policy->rwsem);
2014
2015	if (ret)
2016	pr_err("%s: Failed to start governor for CPU%u's policy\n",
2017	__func__, policy->cpu);
2018	}
2019	}
2020	}
2021
2022	/**
2023	* cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
2024	* @flags: Flags to test against the current cpufreq driver's flags.
2025	*
2026	* Assumes that the driver is there, so callers must ensure that this is the
2027	* case.
2028	*/
2029	bool cpufreq_driver_test_flags(u16 flags)
2030	{
2031	return !!(cpufreq_driver->flags & flags);
2032	}
2033
2034	/**
2035	* cpufreq_get_current_driver - Return the current driver's name.
2036	*
2037	* Return the name string of the currently registered cpufreq driver or NULL if
2038	* none.
2039	*/
2040	const char *cpufreq_get_current_driver(void)
2041	{
2042	if (cpufreq_driver)
2043	return cpufreq_driver->name;
2044
2045	return NULL;
2046	}
2047	EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
2048
2049	/**
2050	* cpufreq_get_driver_data - Return current driver data.
2051	*
2052	* Return the private data of the currently registered cpufreq driver, or NULL
2053	* if no cpufreq driver has been registered.
2054	*/
2055	void *cpufreq_get_driver_data(void)
2056	{
2057	if (cpufreq_driver)
2058	return cpufreq_driver->driver_data;
2059
2060	return NULL;
2061	}
2062	EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
2063
2064	/*********************************************************************
2065	* NOTIFIER LISTS INTERFACE *
2066	*********************************************************************/
2067
2068	/**
2069	* cpufreq_register_notifier - Register a notifier with cpufreq.
2070	* @nb: notifier function to register.
2071	* @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2072	*
2073	* Add a notifier to one of two lists: either a list of notifiers that run on
2074	* clock rate changes (once before and once after every transition), or a list
2075	* of notifiers that ron on cpufreq policy changes.
2076	*
2077	* This function may sleep and it has the same return values as
2078	* blocking_notifier_chain_register().
2079	*/
2080	int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
2081	{
2082	int ret;
2083
2084	if (cpufreq_disabled())
2085	return -EINVAL;
2086
2087	switch (list) {
2088	case CPUFREQ_TRANSITION_NOTIFIER:
2089	mutex_lock(&cpufreq_fast_switch_lock);
2090
2091	if (cpufreq_fast_switch_count > 0) {
2092	mutex_unlock(lock: &cpufreq_fast_switch_lock);
2093	return -EBUSY;
2094	}
2095	ret = srcu_notifier_chain_register(
2096	nh: &cpufreq_transition_notifier_list, nb);
2097	if (!ret)
2098	cpufreq_fast_switch_count--;
2099
2100	mutex_unlock(lock: &cpufreq_fast_switch_lock);
2101	break;
2102	case CPUFREQ_POLICY_NOTIFIER:
2103	ret = blocking_notifier_chain_register(
2104	nh: &cpufreq_policy_notifier_list, nb);
2105	break;
2106	default:
2107	ret = -EINVAL;
2108	}
2109
2110	return ret;
2111	}
2112	EXPORT_SYMBOL(cpufreq_register_notifier);
2113
2114	/**
2115	* cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2116	* @nb: notifier block to be unregistered.
2117	* @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2118	*
2119	* Remove a notifier from one of the cpufreq notifier lists.
2120	*
2121	* This function may sleep and it has the same return values as
2122	* blocking_notifier_chain_unregister().
2123	*/
2124	int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2125	{
2126	int ret;
2127
2128	if (cpufreq_disabled())
2129	return -EINVAL;
2130
2131	switch (list) {
2132	case CPUFREQ_TRANSITION_NOTIFIER:
2133	mutex_lock(&cpufreq_fast_switch_lock);
2134
2135	ret = srcu_notifier_chain_unregister(
2136	nh: &cpufreq_transition_notifier_list, nb);
2137	if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2138	cpufreq_fast_switch_count++;
2139
2140	mutex_unlock(lock: &cpufreq_fast_switch_lock);
2141	break;
2142	case CPUFREQ_POLICY_NOTIFIER:
2143	ret = blocking_notifier_chain_unregister(
2144	nh: &cpufreq_policy_notifier_list, nb);
2145	break;
2146	default:
2147	ret = -EINVAL;
2148	}
2149
2150	return ret;
2151	}
2152	EXPORT_SYMBOL(cpufreq_unregister_notifier);
2153
2154
2155	/*********************************************************************
2156	* GOVERNORS *
2157	*********************************************************************/
2158
2159	/**
2160	* cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2161	* @policy: cpufreq policy to switch the frequency for.
2162	* @target_freq: New frequency to set (may be approximate).
2163	*
2164	* Carry out a fast frequency switch without sleeping.
2165	*
2166	* The driver's ->fast_switch() callback invoked by this function must be
2167	* suitable for being called from within RCU-sched read-side critical sections
2168	* and it is expected to select the minimum available frequency greater than or
2169	* equal to @target_freq (CPUFREQ_RELATION_L).
2170	*
2171	* This function must not be called if policy->fast_switch_enabled is unset.
2172	*
2173	* Governors calling this function must guarantee that it will never be invoked
2174	* twice in parallel for the same policy and that it will never be called in
2175	* parallel with either ->target() or ->target_index() for the same policy.
2176	*
2177	* Returns the actual frequency set for the CPU.
2178	*
2179	* If 0 is returned by the driver's ->fast_switch() callback to indicate an
2180	* error condition, the hardware configuration must be preserved.
2181	*/
2182	unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2183	unsigned int target_freq)
2184	{
2185	unsigned int freq;
2186	int cpu;
2187
2188	target_freq = clamp_val(target_freq, policy->min, policy->max);
2189	freq = cpufreq_driver->fast_switch(policy, target_freq);
2190
2191	if (!freq)
2192	return 0;
2193
2194	policy->cur = freq;
2195	arch_set_freq_scale(cpus: policy->related_cpus, cur_freq: freq,
2196	max_freq: arch_scale_freq_ref(cpu: policy->cpu));
2197	cpufreq_stats_record_transition(policy, new_freq: freq);
2198
2199	if (trace_cpu_frequency_enabled()) {
2200	for_each_cpu(cpu, policy->cpus)
2201	trace_cpu_frequency(frequency: freq, cpu_id: cpu);
2202	}
2203
2204	return freq;
2205	}
2206	EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2207
2208	/**
2209	* cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2210	* @cpu: Target CPU.
2211	* @min_perf: Minimum (required) performance level (units of @capacity).
2212	* @target_perf: Target (desired) performance level (units of @capacity).
2213	* @capacity: Capacity of the target CPU.
2214	*
2215	* Carry out a fast performance level switch of @cpu without sleeping.
2216	*
2217	* The driver's ->adjust_perf() callback invoked by this function must be
2218	* suitable for being called from within RCU-sched read-side critical sections
2219	* and it is expected to select a suitable performance level equal to or above
2220	* @min_perf and preferably equal to or below @target_perf.
2221	*
2222	* This function must not be called if policy->fast_switch_enabled is unset.
2223	*
2224	* Governors calling this function must guarantee that it will never be invoked
2225	* twice in parallel for the same CPU and that it will never be called in
2226	* parallel with either ->target() or ->target_index() or ->fast_switch() for
2227	* the same CPU.
2228	*/
2229	void cpufreq_driver_adjust_perf(unsigned int cpu,
2230	unsigned long min_perf,
2231	unsigned long target_perf,
2232	unsigned long capacity)
2233	{
2234	cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2235	}
2236
2237	/**
2238	* cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2239	*
2240	* Return 'true' if the ->adjust_perf callback is present for the
2241	* current driver or 'false' otherwise.
2242	*/
2243	bool cpufreq_driver_has_adjust_perf(void)
2244	{
2245	return !!cpufreq_driver->adjust_perf;
2246	}
2247
2248	/* Must set freqs->new to intermediate frequency */
2249	static int __target_intermediate(struct cpufreq_policy *policy,
2250	struct cpufreq_freqs *freqs, int index)
2251	{
2252	int ret;
2253
2254	freqs->new = cpufreq_driver->get_intermediate(policy, index);
2255
2256	/* We don't need to switch to intermediate freq */
2257	if (!freqs->new)
2258	return 0;
2259
2260	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2261	__func__, policy->cpu, freqs->old, freqs->new);
2262
2263	cpufreq_freq_transition_begin(policy, freqs);
2264	ret = cpufreq_driver->target_intermediate(policy, index);
2265	cpufreq_freq_transition_end(policy, freqs, ret);
2266
2267	if (ret)
2268	pr_err("%s: Failed to change to intermediate frequency: %d\n",
2269	__func__, ret);
2270
2271	return ret;
2272	}
2273
2274	static int __target_index(struct cpufreq_policy *policy, int index)
2275	{
2276	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2277	unsigned int restore_freq, intermediate_freq = 0;
2278	unsigned int newfreq = policy->freq_table[index].frequency;
2279	int retval = -EINVAL;
2280	bool notify;
2281
2282	if (newfreq == policy->cur)
2283	return 0;
2284
2285	/* Save last value to restore later on errors */
2286	restore_freq = policy->cur;
2287
2288	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2289	if (notify) {
2290	/* Handle switching to intermediate frequency */
2291	if (cpufreq_driver->get_intermediate) {
2292	retval = __target_intermediate(policy, freqs: &freqs, index);
2293	if (retval)
2294	return retval;
2295
2296	intermediate_freq = freqs.new;
2297	/* Set old freq to intermediate */
2298	if (intermediate_freq)
2299	freqs.old = freqs.new;
2300	}
2301
2302	freqs.new = newfreq;
2303	pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2304	__func__, policy->cpu, freqs.old, freqs.new);
2305
2306	cpufreq_freq_transition_begin(policy, &freqs);
2307	}
2308
2309	retval = cpufreq_driver->target_index(policy, index);
2310	if (retval)
2311	pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2312	retval);
2313
2314	if (notify) {
2315	cpufreq_freq_transition_end(policy, &freqs, retval);
2316
2317	/*
2318	* Failed after setting to intermediate freq? Driver should have
2319	* reverted back to initial frequency and so should we. Check
2320	* here for intermediate_freq instead of get_intermediate, in
2321	* case we haven't switched to intermediate freq at all.
2322	*/
2323	if (unlikely(retval && intermediate_freq)) {
2324	freqs.old = intermediate_freq;
2325	freqs.new = restore_freq;
2326	cpufreq_freq_transition_begin(policy, &freqs);
2327	cpufreq_freq_transition_end(policy, &freqs, 0);
2328	}
2329	}
2330
2331	return retval;
2332	}
2333
2334	int __cpufreq_driver_target(struct cpufreq_policy *policy,
2335	unsigned int target_freq,
2336	unsigned int relation)
2337	{
2338	unsigned int old_target_freq = target_freq;
2339
2340	if (cpufreq_disabled())
2341	return -ENODEV;
2342
2343	target_freq = __resolve_freq(policy, target_freq, relation);
2344
2345	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2346	policy->cpu, target_freq, relation, old_target_freq);
2347
2348	/*
2349	* This might look like a redundant call as we are checking it again
2350	* after finding index. But it is left intentionally for cases where
2351	* exactly same freq is called again and so we can save on few function
2352	* calls.
2353	*/
2354	if (target_freq == policy->cur &&
2355	!(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2356	return 0;
2357
2358	if (cpufreq_driver->target) {
2359	/*
2360	* If the driver hasn't setup a single inefficient frequency,
2361	* it's unlikely it knows how to decode CPUFREQ_RELATION_E.
2362	*/
2363	if (!policy->efficiencies_available)
2364	relation &= ~CPUFREQ_RELATION_E;
2365
2366	return cpufreq_driver->target(policy, target_freq, relation);
2367	}
2368
2369	if (!cpufreq_driver->target_index)
2370	return -EINVAL;
2371
2372	return __target_index(policy, index: policy->cached_resolved_idx);
2373	}
2374	EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2375
2376	int cpufreq_driver_target(struct cpufreq_policy *policy,
2377	unsigned int target_freq,
2378	unsigned int relation)
2379	{
2380	int ret;
2381
2382	down_write(sem: &policy->rwsem);
2383
2384	ret = __cpufreq_driver_target(policy, target_freq, relation);
2385
2386	up_write(sem: &policy->rwsem);
2387
2388	return ret;
2389	}
2390	EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2391
2392	__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2393	{
2394	return NULL;
2395	}
2396
2397	static int cpufreq_init_governor(struct cpufreq_policy *policy)
2398	{
2399	int ret;
2400
2401	/* Don't start any governor operations if we are entering suspend */
2402	if (cpufreq_suspended)
2403	return 0;
2404	/*
2405	* Governor might not be initiated here if ACPI _PPC changed
2406	* notification happened, so check it.
2407	*/
2408	if (!policy->governor)
2409	return -EINVAL;
2410
2411	/* Platform doesn't want dynamic frequency switching ? */
2412	if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2413	cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2414	struct cpufreq_governor *gov = cpufreq_fallback_governor();
2415
2416	if (gov) {
2417	pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2418	policy->governor->name, gov->name);
2419	policy->governor = gov;
2420	} else {
2421	return -EINVAL;
2422	}
2423	}
2424
2425	if (!try_module_get(module: policy->governor->owner))
2426	return -EINVAL;
2427
2428	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2429
2430	if (policy->governor->init) {
2431	ret = policy->governor->init(policy);
2432	if (ret) {
2433	module_put(module: policy->governor->owner);
2434	return ret;
2435	}
2436	}
2437
2438	policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2439
2440	return 0;
2441	}
2442
2443	static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2444	{
2445	if (cpufreq_suspended || !policy->governor)
2446	return;
2447
2448	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2449
2450	if (policy->governor->exit)
2451	policy->governor->exit(policy);
2452
2453	module_put(module: policy->governor->owner);
2454	}
2455
2456	int cpufreq_start_governor(struct cpufreq_policy *policy)
2457	{
2458	int ret;
2459
2460	if (cpufreq_suspended)
2461	return 0;
2462
2463	if (!policy->governor)
2464	return -EINVAL;
2465
2466	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2467
2468	if (cpufreq_driver->get)
2469	cpufreq_verify_current_freq(policy, update: false);
2470
2471	if (policy->governor->start) {
2472	ret = policy->governor->start(policy);
2473	if (ret)
2474	return ret;
2475	}
2476
2477	if (policy->governor->limits)
2478	policy->governor->limits(policy);
2479
2480	return 0;
2481	}
2482
2483	void cpufreq_stop_governor(struct cpufreq_policy *policy)
2484	{
2485	if (cpufreq_suspended || !policy->governor)
2486	return;
2487
2488	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2489
2490	if (policy->governor->stop)
2491	policy->governor->stop(policy);
2492	}
2493
2494	static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2495	{
2496	if (cpufreq_suspended || !policy->governor)
2497	return;
2498
2499	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2500
2501	if (policy->governor->limits)
2502	policy->governor->limits(policy);
2503	}
2504
2505	int cpufreq_register_governor(struct cpufreq_governor *governor)
2506	{
2507	int err;
2508
2509	if (!governor)
2510	return -EINVAL;
2511
2512	if (cpufreq_disabled())
2513	return -ENODEV;
2514
2515	mutex_lock(&cpufreq_governor_mutex);
2516
2517	err = -EBUSY;
2518	if (!find_governor(str_governor: governor->name)) {
2519	err = 0;
2520	list_add(new: &governor->governor_list, head: &cpufreq_governor_list);
2521	}
2522
2523	mutex_unlock(lock: &cpufreq_governor_mutex);
2524	return err;
2525	}
2526	EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2527
2528	void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2529	{
2530	struct cpufreq_policy *policy;
2531	unsigned long flags;
2532
2533	if (!governor)
2534	return;
2535
2536	if (cpufreq_disabled())
2537	return;
2538
2539	/* clear last_governor for all inactive policies */
2540	read_lock_irqsave(&cpufreq_driver_lock, flags);
2541	for_each_inactive_policy(policy) {
2542	if (!strcmp(policy->last_governor, governor->name)) {
2543	policy->governor = NULL;
2544	strcpy(p: policy->last_governor, q: "\0");
2545	}
2546	}
2547	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2548
2549	mutex_lock(&cpufreq_governor_mutex);
2550	list_del(entry: &governor->governor_list);
2551	mutex_unlock(lock: &cpufreq_governor_mutex);
2552	}
2553	EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2554
2555
2556	/*********************************************************************
2557	* POLICY INTERFACE *
2558	*********************************************************************/
2559
2560	/**
2561	* cpufreq_get_policy - get the current cpufreq_policy
2562	* @policy: struct cpufreq_policy into which the current cpufreq_policy
2563	* is written
2564	* @cpu: CPU to find the policy for
2565	*
2566	* Reads the current cpufreq policy.
2567	*/
2568	int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2569	{
2570	struct cpufreq_policy *cpu_policy;
2571	if (!policy)
2572	return -EINVAL;
2573
2574	cpu_policy = cpufreq_cpu_get(cpu);
2575	if (!cpu_policy)
2576	return -EINVAL;
2577
2578	memcpy(policy, cpu_policy, sizeof(*policy));
2579
2580	cpufreq_cpu_put(cpu_policy);
2581	return 0;
2582	}
2583	EXPORT_SYMBOL(cpufreq_get_policy);
2584
2585	/**
2586	* cpufreq_set_policy - Modify cpufreq policy parameters.
2587	* @policy: Policy object to modify.
2588	* @new_gov: Policy governor pointer.
2589	* @new_pol: Policy value (for drivers with built-in governors).
2590	*
2591	* Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2592	* limits to be set for the policy, update @policy with the verified limits
2593	* values and either invoke the driver's ->setpolicy() callback (if present) or
2594	* carry out a governor update for @policy. That is, run the current governor's
2595	* ->limits() callback (if @new_gov points to the same object as the one in
2596	* @policy) or replace the governor for @policy with @new_gov.
2597	*
2598	* The cpuinfo part of @policy is not updated by this function.
2599	*/
2600	static int cpufreq_set_policy(struct cpufreq_policy *policy,
2601	struct cpufreq_governor *new_gov,
2602	unsigned int new_pol)
2603	{
2604	struct cpufreq_policy_data new_data;
2605	struct cpufreq_governor *old_gov;
2606	int ret;
2607
2608	memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2609	new_data.freq_table = policy->freq_table;
2610	new_data.cpu = policy->cpu;
2611	/*
2612	* PM QoS framework collects all the requests from users and provide us
2613	* the final aggregated value here.
2614	*/
2615	new_data.min = freq_qos_read_value(qos: &policy->constraints, type: FREQ_QOS_MIN);
2616	new_data.max = freq_qos_read_value(qos: &policy->constraints, type: FREQ_QOS_MAX);
2617
2618	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2619	new_data.cpu, new_data.min, new_data.max);
2620
2621	/*
2622	* Verify that the CPU speed can be set within these limits and make sure
2623	* that min <= max.
2624	*/
2625	ret = cpufreq_driver->verify(&new_data);
2626	if (ret)
2627	return ret;
2628
2629	/*
2630	* Resolve policy min/max to available frequencies. It ensures
2631	* no frequency resolution will neither overshoot the requested maximum
2632	* nor undershoot the requested minimum.
2633	*/
2634	policy->min = new_data.min;
2635	policy->max = new_data.max;
2636	policy->min = __resolve_freq(policy, target_freq: policy->min, CPUFREQ_RELATION_L);
2637	policy->max = __resolve_freq(policy, target_freq: policy->max, CPUFREQ_RELATION_H);
2638	trace_cpu_frequency_limits(policy);
2639
2640	policy->cached_target_freq = UINT_MAX;
2641
2642	pr_debug("new min and max freqs are %u - %u kHz\n",
2643	policy->min, policy->max);
2644
2645	if (cpufreq_driver->setpolicy) {
2646	policy->policy = new_pol;
2647	pr_debug("setting range\n");
2648	return cpufreq_driver->setpolicy(policy);
2649	}
2650
2651	if (new_gov == policy->governor) {
2652	pr_debug("governor limits update\n");
2653	cpufreq_governor_limits(policy);
2654	return 0;
2655	}
2656
2657	pr_debug("governor switch\n");
2658
2659	/* save old, working values */
2660	old_gov = policy->governor;
2661	/* end old governor */
2662	if (old_gov) {
2663	cpufreq_stop_governor(policy);
2664	cpufreq_exit_governor(policy);
2665	}
2666
2667	/* start new governor */
2668	policy->governor = new_gov;
2669	ret = cpufreq_init_governor(policy);
2670	if (!ret) {
2671	ret = cpufreq_start_governor(policy);
2672	if (!ret) {
2673	pr_debug("governor change\n");
2674	return 0;
2675	}
2676	cpufreq_exit_governor(policy);
2677	}
2678
2679	/* new governor failed, so re-start old one */
2680	pr_debug("starting governor %s failed\n", policy->governor->name);
2681	if (old_gov) {
2682	policy->governor = old_gov;
2683	if (cpufreq_init_governor(policy))
2684	policy->governor = NULL;
2685	else
2686	cpufreq_start_governor(policy);
2687	}
2688
2689	return ret;
2690	}
2691
2692	/**
2693	* cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2694	* @cpu: CPU to re-evaluate the policy for.
2695	*
2696	* Update the current frequency for the cpufreq policy of @cpu and use
2697	* cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2698	* evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2699	* for the policy in question, among other things.
2700	*/
2701	void cpufreq_update_policy(unsigned int cpu)
2702	{
2703	struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2704
2705	if (!policy)
2706	return;
2707
2708	/*
2709	* BIOS might change freq behind our back
2710	* -> ask driver for current freq and notify governors about a change
2711	*/
2712	if (cpufreq_driver->get && has_target() &&
2713	(cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2714	goto unlock;
2715
2716	refresh_frequency_limits(policy);
2717
2718	unlock:
2719	cpufreq_cpu_release(policy);
2720	}
2721	EXPORT_SYMBOL(cpufreq_update_policy);
2722
2723	/**
2724	* cpufreq_update_limits - Update policy limits for a given CPU.
2725	* @cpu: CPU to update the policy limits for.
2726	*
2727	* Invoke the driver's ->update_limits callback if present or call
2728	* cpufreq_update_policy() for @cpu.
2729	*/
2730	void cpufreq_update_limits(unsigned int cpu)
2731	{
2732	if (cpufreq_driver->update_limits)
2733	cpufreq_driver->update_limits(cpu);
2734	else
2735	cpufreq_update_policy(cpu);
2736	}
2737	EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2738
2739	/*********************************************************************
2740	* BOOST *
2741	*********************************************************************/
2742	static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2743	{
2744	int ret;
2745
2746	if (!policy->freq_table)
2747	return -ENXIO;
2748
2749	ret = cpufreq_frequency_table_cpuinfo(policy, table: policy->freq_table);
2750	if (ret) {
2751	pr_err("%s: Policy frequency update failed\n", __func__);
2752	return ret;
2753	}
2754
2755	ret = freq_qos_update_request(req: policy->max_freq_req, new_value: policy->max);
2756	if (ret < 0)
2757	return ret;
2758
2759	return 0;
2760	}
2761
2762	int cpufreq_boost_trigger_state(int state)
2763	{
2764	struct cpufreq_policy *policy;
2765	unsigned long flags;
2766	int ret = 0;
2767
2768	if (cpufreq_driver->boost_enabled == state)
2769	return 0;
2770
2771	write_lock_irqsave(&cpufreq_driver_lock, flags);
2772	cpufreq_driver->boost_enabled = state;
2773	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2774
2775	cpus_read_lock();
2776	for_each_active_policy(policy) {
2777	policy->boost_enabled = state;
2778	ret = cpufreq_driver->set_boost(policy, state);
2779	if (ret) {
2780	policy->boost_enabled = !policy->boost_enabled;
2781	goto err_reset_state;
2782	}
2783	}
2784	cpus_read_unlock();
2785
2786	return 0;
2787
2788	err_reset_state:
2789	cpus_read_unlock();
2790
2791	write_lock_irqsave(&cpufreq_driver_lock, flags);
2792	cpufreq_driver->boost_enabled = !state;
2793	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2794
2795	pr_err("%s: Cannot %s BOOST\n",
2796	__func__, state ? "enable" : "disable");
2797
2798	return ret;
2799	}
2800
2801	static bool cpufreq_boost_supported(void)
2802	{
2803	return cpufreq_driver->set_boost;
2804	}
2805
2806	static int create_boost_sysfs_file(void)
2807	{
2808	int ret;
2809
2810	ret = sysfs_create_file(kobj: cpufreq_global_kobject, attr: &boost.attr);
2811	if (ret)
2812	pr_err("%s: cannot register global BOOST sysfs file\n",
2813	__func__);
2814
2815	return ret;
2816	}
2817
2818	static void remove_boost_sysfs_file(void)
2819	{
2820	if (cpufreq_boost_supported())
2821	sysfs_remove_file(kobj: cpufreq_global_kobject, attr: &boost.attr);
2822	}
2823
2824	int cpufreq_enable_boost_support(void)
2825	{
2826	if (!cpufreq_driver)
2827	return -EINVAL;
2828
2829	if (cpufreq_boost_supported())
2830	return 0;
2831
2832	cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2833
2834	/* This will get removed on driver unregister */
2835	return create_boost_sysfs_file();
2836	}
2837	EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2838
2839	int cpufreq_boost_enabled(void)
2840	{
2841	return cpufreq_driver->boost_enabled;
2842	}
2843	EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2844
2845	/*********************************************************************
2846	* REGISTER / UNREGISTER CPUFREQ DRIVER *
2847	*********************************************************************/
2848	static enum cpuhp_state hp_online;
2849
2850	static int cpuhp_cpufreq_online(unsigned int cpu)
2851	{
2852	cpufreq_online(cpu);
2853
2854	return 0;
2855	}
2856
2857	static int cpuhp_cpufreq_offline(unsigned int cpu)
2858	{
2859	cpufreq_offline(cpu);
2860
2861	return 0;
2862	}
2863
2864	/**
2865	* cpufreq_register_driver - register a CPU Frequency driver
2866	* @driver_data: A struct cpufreq_driver containing the values#
2867	* submitted by the CPU Frequency driver.
2868	*
2869	* Registers a CPU Frequency driver to this core code. This code
2870	* returns zero on success, -EEXIST when another driver got here first
2871	* (and isn't unregistered in the meantime).
2872	*
2873	*/
2874	int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2875	{
2876	unsigned long flags;
2877	int ret;
2878
2879	if (cpufreq_disabled())
2880	return -ENODEV;
2881
2882	/*
2883	* The cpufreq core depends heavily on the availability of device
2884	* structure, make sure they are available before proceeding further.
2885	*/
2886	if (!get_cpu_device(cpu: 0))
2887	return -EPROBE_DEFER;
2888
2889	if (!driver_data || !driver_data->verify || !driver_data->init ||
2890	!(driver_data->setpolicy || driver_data->target_index ||
2891	driver_data->target) ||
2892	(driver_data->setpolicy && (driver_data->target_index ||
2893	driver_data->target)) ||
2894	(!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2895	(!driver_data->online != !driver_data->offline) ||
2896	(driver_data->adjust_perf && !driver_data->fast_switch))
2897	return -EINVAL;
2898
2899	pr_debug("trying to register driver %s\n", driver_data->name);
2900
2901	/* Protect against concurrent CPU online/offline. */
2902	cpus_read_lock();
2903
2904	write_lock_irqsave(&cpufreq_driver_lock, flags);
2905	if (cpufreq_driver) {
2906	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2907	ret = -EEXIST;
2908	goto out;
2909	}
2910	cpufreq_driver = driver_data;
2911	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2912
2913	/*
2914	* Mark support for the scheduler's frequency invariance engine for
2915	* drivers that implement target(), target_index() or fast_switch().
2916	*/
2917	if (!cpufreq_driver->setpolicy) {
2918	static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2919	pr_debug("supports frequency invariance");
2920	}
2921
2922	if (driver_data->setpolicy)
2923	driver_data->flags |= CPUFREQ_CONST_LOOPS;
2924
2925	if (cpufreq_boost_supported()) {
2926	ret = create_boost_sysfs_file();
2927	if (ret)
2928	goto err_null_driver;
2929	}
2930
2931	ret = subsys_interface_register(sif: &cpufreq_interface);
2932	if (ret)
2933	goto err_boost_unreg;
2934
2935	if (unlikely(list_empty(&cpufreq_policy_list))) {
2936	/* if all ->init() calls failed, unregister */
2937	ret = -ENODEV;
2938	pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2939	driver_data->name);
2940	goto err_if_unreg;
2941	}
2942
2943	ret = cpuhp_setup_state_nocalls_cpuslocked(state: CPUHP_AP_ONLINE_DYN,
2944	name: "cpufreq:online",
2945	startup: cpuhp_cpufreq_online,
2946	teardown: cpuhp_cpufreq_offline);
2947	if (ret < 0)
2948	goto err_if_unreg;
2949	hp_online = ret;
2950	ret = 0;
2951
2952	pr_debug("driver %s up and running\n", driver_data->name);
2953	goto out;
2954
2955	err_if_unreg:
2956	subsys_interface_unregister(sif: &cpufreq_interface);
2957	err_boost_unreg:
2958	remove_boost_sysfs_file();
2959	err_null_driver:
2960	write_lock_irqsave(&cpufreq_driver_lock, flags);
2961	cpufreq_driver = NULL;
2962	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2963	out:
2964	cpus_read_unlock();
2965	return ret;
2966	}
2967	EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2968
2969	/*
2970	* cpufreq_unregister_driver - unregister the current CPUFreq driver
2971	*
2972	* Unregister the current CPUFreq driver. Only call this if you have
2973	* the right to do so, i.e. if you have succeeded in initialising before!
2974	* Returns zero if successful, and -EINVAL if the cpufreq_driver is
2975	* currently not initialised.
2976	*/
2977	void cpufreq_unregister_driver(struct cpufreq_driver *driver)
2978	{
2979	unsigned long flags;
2980
2981	if (WARN_ON(!cpufreq_driver || (driver != cpufreq_driver)))
2982	return;
2983
2984	pr_debug("unregistering driver %s\n", driver->name);
2985
2986	/* Protect against concurrent cpu hotplug */
2987	cpus_read_lock();
2988	subsys_interface_unregister(sif: &cpufreq_interface);
2989	remove_boost_sysfs_file();
2990	static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
2991	cpuhp_remove_state_nocalls_cpuslocked(state: hp_online);
2992
2993	write_lock_irqsave(&cpufreq_driver_lock, flags);
2994
2995	cpufreq_driver = NULL;
2996
2997	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2998	cpus_read_unlock();
2999	}
3000	EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
3001
3002	static int __init cpufreq_core_init(void)
3003	{
3004	struct cpufreq_governor *gov = cpufreq_default_governor();
3005	struct device *dev_root;
3006
3007	if (cpufreq_disabled())
3008	return -ENODEV;
3009
3010	dev_root = bus_get_dev_root(bus: &cpu_subsys);
3011	if (dev_root) {
3012	cpufreq_global_kobject = kobject_create_and_add(name: "cpufreq", parent: &dev_root->kobj);
3013	put_device(dev: dev_root);
3014	}
3015	BUG_ON(!cpufreq_global_kobject);
3016
3017	if (!strlen(default_governor))
3018	strscpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
3019
3020	return 0;
3021	}
3022	module_param(off, int, 0444);
3023	module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
3024	core_initcall(cpufreq_core_init);
3025