xref: /linux/drivers/cpufreq/cpufreq.c (revision 3e7819886281e077e82006fe4804b0d6b0f5643b)
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(get_jiffies_64());
137 
138 	kcpustat_cpu_fetch(&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(cur_wall_time, NSEC_PER_USEC);
150 
151 	return div_u64(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, 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, 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(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, 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(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(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(&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(&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(&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(&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(&cpufreq_transition_notifier_list,
379 					 CPUFREQ_PRECHANGE, freqs);
380 
381 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
382 		break;
383 
384 	case CPUFREQ_POSTCHANGE:
385 		adjust_jiffies(CPUFREQ_POSTCHANGE, 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(freqs->new, cpu);
391 
392 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
393 					 CPUFREQ_POSTCHANGE, freqs);
394 
395 		cpufreq_stats_record_transition(policy, 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(&policy->transition_lock);
432 
433 	if (unlikely(policy->transition_ongoing)) {
434 		spin_unlock(&policy->transition_lock);
435 		goto wait;
436 	}
437 
438 	policy->transition_ongoing = true;
439 	policy->transition_task = current;
440 
441 	spin_unlock(&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(policy->related_cpus,
456 			    policy->cur,
457 			    arch_scale_freq_ref(policy->cpu));
458 
459 	spin_lock(&policy->transition_lock);
460 	policy->transition_ongoing = false;
461 	policy->transition_task = NULL;
462 	spin_unlock(&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(&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(&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(&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, "%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(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, "%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(buf, 10, &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(str_governor, 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(t->owner))
693 		t = NULL;
694 
695 unlock:
696 	mutex_unlock(&cpufreq_governor_mutex);
697 
698 	return t;
699 }
700 
701 static unsigned int cpufreq_parse_policy(char *str_governor)
702 {
703 	if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
704 		return CPUFREQ_POLICY_PERFORMANCE;
705 
706 	if (!strncasecmp(str_governor, "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(policy->cpu);
762 	if (freq)
763 		ret = sprintf(buf, "%u\n", freq);
764 	else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
765 		ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
766 	else
767 		ret = sprintf(buf, "%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, "%u\n", cur_freq);
802 
803 	return sprintf(buf, "<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, "powersave\n");
813 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
814 		return sprintf(buf, "performance\n");
815 	else if (policy->governor)
816 		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%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(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, "%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, "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[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
886 	}
887 	mutex_unlock(&cpufreq_governor_mutex);
888 out:
889 	i += sprintf(&buf[i], "\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[i], (PAGE_SIZE - i - 2), "%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[i], "\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, "<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, "%u\n", limit);
965 	return sprintf(buf, "%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(&policy->rwsem);
1012 	if (likely(!policy_is_inactive(policy)))
1013 		ret = fattr->show(policy, buf);
1014 	up_read(&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(&policy->rwsem);
1030 	if (likely(!policy_is_inactive(policy)))
1031 		ret = fattr->store(policy, buf, count);
1032 	up_write(&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, policy->real_cpus))
1062 		return;
1063 
1064 	dev_dbg(dev, "%s: Adding symlink\n", __func__);
1065 	if (sysfs_create_link(&dev->kobj, &policy->kobj, "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(&dev->kobj, "cpufreq");
1074 	cpumask_clear_cpu(cpu, 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(&policy->kobj, &((*drv_attr)->attr));
1086 		if (ret)
1087 			return ret;
1088 		drv_attr++;
1089 	}
1090 	if (cpufreq_driver->get) {
1091 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1092 		if (ret)
1093 			return ret;
1094 	}
1095 
1096 	ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1097 	if (ret)
1098 		return ret;
1099 
1100 	if (cpufreq_driver->bios_limit) {
1101 		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1102 		if (ret)
1103 			return ret;
1104 	}
1105 
1106 	if (cpufreq_boost_supported()) {
1107 		ret = sysfs_create_file(&policy->kobj, &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(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(default_governor);
1129 		}
1130 
1131 		if (!gov) {
1132 			gov = cpufreq_default_governor();
1133 			__module_get(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(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, gov, pol);
1157 	if (gov)
1158 		module_put(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, policy->cpus))
1169 		return 0;
1170 
1171 	down_write(&policy->rwsem);
1172 	if (has_target())
1173 		cpufreq_stop_governor(policy);
1174 
1175 	cpumask_set_cpu(cpu, 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(&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, policy->governor, 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(&policy->rwsem);
1203 	refresh_frequency_limits(policy);
1204 	up_write(&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(&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(&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(&policy->rwsem);
1231 	cpufreq_stats_free_table(policy);
1232 	kobj = &policy->kobj;
1233 	cmp = &policy->kobj_unregister;
1234 	up_write(&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(sizeof(*policy), GFP_KERNEL);
1257 	if (!policy)
1258 		return NULL;
1259 
1260 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1261 		goto err_free_policy;
1262 
1263 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1264 		goto err_free_cpumask;
1265 
1266 	if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1267 		goto err_free_rcpumask;
1268 
1269 	init_completion(&policy->kobj_unregister);
1270 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1271 				   cpufreq_global_kobject, "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(&policy->kobj);
1280 		goto err_free_real_cpus;
1281 	}
1282 
1283 	freq_constraints_init(&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(&policy->constraints, FREQ_QOS_MIN,
1289 				    &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(&policy->constraints, FREQ_QOS_MAX,
1297 				    &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(&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(&policy->constraints, FREQ_QOS_MIN,
1315 				 &policy->nb_min);
1316 err_kobj_remove:
1317 	cpufreq_policy_put_kobj(policy);
1318 err_free_real_cpus:
1319 	free_cpumask_var(policy->real_cpus);
1320 err_free_rcpumask:
1321 	free_cpumask_var(policy->related_cpus);
1322 err_free_cpumask:
1323 	free_cpumask_var(policy->cpus);
1324 err_free_policy:
1325 	kfree(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(&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(&policy->constraints, FREQ_QOS_MAX,
1351 				 &policy->nb_max);
1352 	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1353 				 &policy->nb_min);
1354 
1355 	/* Cancel any pending policy->update work before freeing the policy. */
1356 	cancel_work_sync(&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(&cpufreq_policy_notifier_list,
1365 					     CPUFREQ_REMOVE_POLICY, policy);
1366 		freq_qos_remove_request(policy->max_freq_req);
1367 	}
1368 
1369 	freq_qos_remove_request(policy->min_freq_req);
1370 	kfree(policy->min_freq_req);
1371 
1372 	cpufreq_policy_put_kobj(policy);
1373 	free_cpumask_var(policy->real_cpus);
1374 	free_cpumask_var(policy->related_cpus);
1375 	free_cpumask_var(policy->cpus);
1376 	kfree(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(&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(&policy->rwsem);
1407 	}
1408 
1409 	if (!new_policy && cpufreq_driver->online) {
1410 		/* Recover policy->cpus using related_cpus */
1411 		cpumask_copy(policy->cpus, 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(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 		if (cpufreq_boost_enabled() && policy_has_boost_freq(policy))
1435 			policy->boost_enabled = true;
1436 
1437 		/*
1438 		 * The initialization has succeeded and the policy is online.
1439 		 * If there is a problem with its frequency table, take it
1440 		 * offline and drop it.
1441 		 */
1442 		ret = cpufreq_table_validate_and_sort(policy);
1443 		if (ret)
1444 			goto out_offline_policy;
1445 
1446 		/* related_cpus should at least include policy->cpus. */
1447 		cpumask_copy(policy->related_cpus, policy->cpus);
1448 	}
1449 
1450 	/*
1451 	 * affected cpus must always be the one, which are online. We aren't
1452 	 * managing offline cpus here.
1453 	 */
1454 	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1455 
1456 	if (new_policy) {
1457 		for_each_cpu(j, policy->related_cpus) {
1458 			per_cpu(cpufreq_cpu_data, j) = policy;
1459 			add_cpu_dev_symlink(policy, j, get_cpu_device(j));
1460 		}
1461 
1462 		policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1463 					       GFP_KERNEL);
1464 		if (!policy->min_freq_req) {
1465 			ret = -ENOMEM;
1466 			goto out_destroy_policy;
1467 		}
1468 
1469 		ret = freq_qos_add_request(&policy->constraints,
1470 					   policy->min_freq_req, FREQ_QOS_MIN,
1471 					   FREQ_QOS_MIN_DEFAULT_VALUE);
1472 		if (ret < 0) {
1473 			/*
1474 			 * So we don't call freq_qos_remove_request() for an
1475 			 * uninitialized request.
1476 			 */
1477 			kfree(policy->min_freq_req);
1478 			policy->min_freq_req = NULL;
1479 			goto out_destroy_policy;
1480 		}
1481 
1482 		/*
1483 		 * This must be initialized right here to avoid calling
1484 		 * freq_qos_remove_request() on uninitialized request in case
1485 		 * of errors.
1486 		 */
1487 		policy->max_freq_req = policy->min_freq_req + 1;
1488 
1489 		ret = freq_qos_add_request(&policy->constraints,
1490 					   policy->max_freq_req, FREQ_QOS_MAX,
1491 					   FREQ_QOS_MAX_DEFAULT_VALUE);
1492 		if (ret < 0) {
1493 			policy->max_freq_req = NULL;
1494 			goto out_destroy_policy;
1495 		}
1496 
1497 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1498 				CPUFREQ_CREATE_POLICY, policy);
1499 	}
1500 
1501 	if (cpufreq_driver->get && has_target()) {
1502 		policy->cur = cpufreq_driver->get(policy->cpu);
1503 		if (!policy->cur) {
1504 			ret = -EIO;
1505 			pr_err("%s: ->get() failed\n", __func__);
1506 			goto out_destroy_policy;
1507 		}
1508 	}
1509 
1510 	/*
1511 	 * Sometimes boot loaders set CPU frequency to a value outside of
1512 	 * frequency table present with cpufreq core. In such cases CPU might be
1513 	 * unstable if it has to run on that frequency for long duration of time
1514 	 * and so its better to set it to a frequency which is specified in
1515 	 * freq-table. This also makes cpufreq stats inconsistent as
1516 	 * cpufreq-stats would fail to register because current frequency of CPU
1517 	 * isn't found in freq-table.
1518 	 *
1519 	 * Because we don't want this change to effect boot process badly, we go
1520 	 * for the next freq which is >= policy->cur ('cur' must be set by now,
1521 	 * otherwise we will end up setting freq to lowest of the table as 'cur'
1522 	 * is initialized to zero).
1523 	 *
1524 	 * We are passing target-freq as "policy->cur - 1" otherwise
1525 	 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1526 	 * equal to target-freq.
1527 	 */
1528 	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1529 	    && has_target()) {
1530 		unsigned int old_freq = policy->cur;
1531 
1532 		/* Are we running at unknown frequency ? */
1533 		ret = cpufreq_frequency_table_get_index(policy, old_freq);
1534 		if (ret == -EINVAL) {
1535 			ret = __cpufreq_driver_target(policy, old_freq - 1,
1536 						      CPUFREQ_RELATION_L);
1537 
1538 			/*
1539 			 * Reaching here after boot in a few seconds may not
1540 			 * mean that system will remain stable at "unknown"
1541 			 * frequency for longer duration. Hence, a BUG_ON().
1542 			 */
1543 			BUG_ON(ret);
1544 			pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
1545 				__func__, policy->cpu, old_freq, policy->cur);
1546 		}
1547 	}
1548 
1549 	if (new_policy) {
1550 		ret = cpufreq_add_dev_interface(policy);
1551 		if (ret)
1552 			goto out_destroy_policy;
1553 
1554 		cpufreq_stats_create_table(policy);
1555 
1556 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1557 		list_add(&policy->policy_list, &cpufreq_policy_list);
1558 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1559 
1560 		/*
1561 		 * Register with the energy model before
1562 		 * sugov_eas_rebuild_sd() is called, which will result
1563 		 * in rebuilding of the sched domains, which should only be done
1564 		 * once the energy model is properly initialized for the policy
1565 		 * first.
1566 		 *
1567 		 * Also, this should be called before the policy is registered
1568 		 * with cooling framework.
1569 		 */
1570 		if (cpufreq_driver->register_em)
1571 			cpufreq_driver->register_em(policy);
1572 	}
1573 
1574 	ret = cpufreq_init_policy(policy);
1575 	if (ret) {
1576 		pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1577 		       __func__, cpu, ret);
1578 		goto out_destroy_policy;
1579 	}
1580 
1581 	up_write(&policy->rwsem);
1582 
1583 	kobject_uevent(&policy->kobj, KOBJ_ADD);
1584 
1585 	/* Callback for handling stuff after policy is ready */
1586 	if (cpufreq_driver->ready)
1587 		cpufreq_driver->ready(policy);
1588 
1589 	/* Register cpufreq cooling only for a new policy */
1590 	if (new_policy && cpufreq_thermal_control_enabled(cpufreq_driver))
1591 		policy->cdev = of_cpufreq_cooling_register(policy);
1592 
1593 	pr_debug("initialization complete\n");
1594 
1595 	return 0;
1596 
1597 out_destroy_policy:
1598 	for_each_cpu(j, policy->real_cpus)
1599 		remove_cpu_dev_symlink(policy, j, get_cpu_device(j));
1600 
1601 out_offline_policy:
1602 	if (cpufreq_driver->offline)
1603 		cpufreq_driver->offline(policy);
1604 
1605 out_exit_policy:
1606 	if (cpufreq_driver->exit)
1607 		cpufreq_driver->exit(policy);
1608 
1609 out_free_policy:
1610 	cpumask_clear(policy->cpus);
1611 	up_write(&policy->rwsem);
1612 
1613 	cpufreq_policy_free(policy);
1614 	return ret;
1615 }
1616 
1617 /**
1618  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1619  * @dev: CPU device.
1620  * @sif: Subsystem interface structure pointer (not used)
1621  */
1622 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1623 {
1624 	struct cpufreq_policy *policy;
1625 	unsigned cpu = dev->id;
1626 	int ret;
1627 
1628 	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1629 
1630 	if (cpu_online(cpu)) {
1631 		ret = cpufreq_online(cpu);
1632 		if (ret)
1633 			return ret;
1634 	}
1635 
1636 	/* Create sysfs link on CPU registration */
1637 	policy = per_cpu(cpufreq_cpu_data, cpu);
1638 	if (policy)
1639 		add_cpu_dev_symlink(policy, cpu, dev);
1640 
1641 	return 0;
1642 }
1643 
1644 static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1645 {
1646 	int ret;
1647 
1648 	if (has_target())
1649 		cpufreq_stop_governor(policy);
1650 
1651 	cpumask_clear_cpu(cpu, policy->cpus);
1652 
1653 	if (!policy_is_inactive(policy)) {
1654 		/* Nominate a new CPU if necessary. */
1655 		if (cpu == policy->cpu)
1656 			policy->cpu = cpumask_any(policy->cpus);
1657 
1658 		/* Start the governor again for the active policy. */
1659 		if (has_target()) {
1660 			ret = cpufreq_start_governor(policy);
1661 			if (ret)
1662 				pr_err("%s: Failed to start governor\n", __func__);
1663 		}
1664 
1665 		return;
1666 	}
1667 
1668 	if (has_target())
1669 		strscpy(policy->last_governor, policy->governor->name,
1670 			CPUFREQ_NAME_LEN);
1671 	else
1672 		policy->last_policy = policy->policy;
1673 
1674 	if (has_target())
1675 		cpufreq_exit_governor(policy);
1676 
1677 	/*
1678 	 * Perform the ->offline() during light-weight tear-down, as
1679 	 * that allows fast recovery when the CPU comes back.
1680 	 */
1681 	if (cpufreq_driver->offline) {
1682 		cpufreq_driver->offline(policy);
1683 		return;
1684 	}
1685 
1686 	if (cpufreq_driver->exit)
1687 		cpufreq_driver->exit(policy);
1688 
1689 	policy->freq_table = NULL;
1690 }
1691 
1692 static int cpufreq_offline(unsigned int cpu)
1693 {
1694 	struct cpufreq_policy *policy;
1695 
1696 	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1697 
1698 	policy = cpufreq_cpu_get_raw(cpu);
1699 	if (!policy) {
1700 		pr_debug("%s: No cpu_data found\n", __func__);
1701 		return 0;
1702 	}
1703 
1704 	down_write(&policy->rwsem);
1705 
1706 	__cpufreq_offline(cpu, policy);
1707 
1708 	up_write(&policy->rwsem);
1709 	return 0;
1710 }
1711 
1712 /*
1713  * cpufreq_remove_dev - remove a CPU device
1714  *
1715  * Removes the cpufreq interface for a CPU device.
1716  */
1717 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1718 {
1719 	unsigned int cpu = dev->id;
1720 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1721 
1722 	if (!policy)
1723 		return;
1724 
1725 	down_write(&policy->rwsem);
1726 
1727 	if (cpu_online(cpu))
1728 		__cpufreq_offline(cpu, policy);
1729 
1730 	remove_cpu_dev_symlink(policy, cpu, dev);
1731 
1732 	if (!cpumask_empty(policy->real_cpus)) {
1733 		up_write(&policy->rwsem);
1734 		return;
1735 	}
1736 
1737 	/*
1738 	 * Unregister cpufreq cooling once all the CPUs of the policy are
1739 	 * removed.
1740 	 */
1741 	if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1742 		cpufreq_cooling_unregister(policy->cdev);
1743 		policy->cdev = NULL;
1744 	}
1745 
1746 	/* We did light-weight exit earlier, do full tear down now */
1747 	if (cpufreq_driver->offline && cpufreq_driver->exit)
1748 		cpufreq_driver->exit(policy);
1749 
1750 	up_write(&policy->rwsem);
1751 
1752 	cpufreq_policy_free(policy);
1753 }
1754 
1755 /**
1756  * cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1757  * @policy: Policy managing CPUs.
1758  * @new_freq: New CPU frequency.
1759  *
1760  * Adjust to the current frequency first and clean up later by either calling
1761  * cpufreq_update_policy(), or scheduling handle_update().
1762  */
1763 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1764 				unsigned int new_freq)
1765 {
1766 	struct cpufreq_freqs freqs;
1767 
1768 	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1769 		 policy->cur, new_freq);
1770 
1771 	freqs.old = policy->cur;
1772 	freqs.new = new_freq;
1773 
1774 	cpufreq_freq_transition_begin(policy, &freqs);
1775 	cpufreq_freq_transition_end(policy, &freqs, 0);
1776 }
1777 
1778 static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1779 {
1780 	unsigned int new_freq;
1781 
1782 	new_freq = cpufreq_driver->get(policy->cpu);
1783 	if (!new_freq)
1784 		return 0;
1785 
1786 	/*
1787 	 * If fast frequency switching is used with the given policy, the check
1788 	 * against policy->cur is pointless, so skip it in that case.
1789 	 */
1790 	if (policy->fast_switch_enabled || !has_target())
1791 		return new_freq;
1792 
1793 	if (policy->cur != new_freq) {
1794 		/*
1795 		 * For some platforms, the frequency returned by hardware may be
1796 		 * slightly different from what is provided in the frequency
1797 		 * table, for example hardware may return 499 MHz instead of 500
1798 		 * MHz. In such cases it is better to avoid getting into
1799 		 * unnecessary frequency updates.
1800 		 */
1801 		if (abs(policy->cur - new_freq) < KHZ_PER_MHZ)
1802 			return policy->cur;
1803 
1804 		cpufreq_out_of_sync(policy, new_freq);
1805 		if (update)
1806 			schedule_work(&policy->update);
1807 	}
1808 
1809 	return new_freq;
1810 }
1811 
1812 /**
1813  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1814  * @cpu: CPU number
1815  *
1816  * This is the last known freq, without actually getting it from the driver.
1817  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1818  */
1819 unsigned int cpufreq_quick_get(unsigned int cpu)
1820 {
1821 	struct cpufreq_policy *policy;
1822 	unsigned int ret_freq = 0;
1823 	unsigned long flags;
1824 
1825 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1826 
1827 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1828 		ret_freq = cpufreq_driver->get(cpu);
1829 		read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1830 		return ret_freq;
1831 	}
1832 
1833 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1834 
1835 	policy = cpufreq_cpu_get(cpu);
1836 	if (policy) {
1837 		ret_freq = policy->cur;
1838 		cpufreq_cpu_put(policy);
1839 	}
1840 
1841 	return ret_freq;
1842 }
1843 EXPORT_SYMBOL(cpufreq_quick_get);
1844 
1845 /**
1846  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1847  * @cpu: CPU number
1848  *
1849  * Just return the max possible frequency for a given CPU.
1850  */
1851 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1852 {
1853 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1854 	unsigned int ret_freq = 0;
1855 
1856 	if (policy) {
1857 		ret_freq = policy->max;
1858 		cpufreq_cpu_put(policy);
1859 	}
1860 
1861 	return ret_freq;
1862 }
1863 EXPORT_SYMBOL(cpufreq_quick_get_max);
1864 
1865 /**
1866  * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1867  * @cpu: CPU number
1868  *
1869  * The default return value is the max_freq field of cpuinfo.
1870  */
1871 __weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1872 {
1873 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1874 	unsigned int ret_freq = 0;
1875 
1876 	if (policy) {
1877 		ret_freq = policy->cpuinfo.max_freq;
1878 		cpufreq_cpu_put(policy);
1879 	}
1880 
1881 	return ret_freq;
1882 }
1883 EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1884 
1885 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1886 {
1887 	if (unlikely(policy_is_inactive(policy)))
1888 		return 0;
1889 
1890 	return cpufreq_verify_current_freq(policy, true);
1891 }
1892 
1893 /**
1894  * cpufreq_get - get the current CPU frequency (in kHz)
1895  * @cpu: CPU number
1896  *
1897  * Get the CPU current (static) CPU frequency
1898  */
1899 unsigned int cpufreq_get(unsigned int cpu)
1900 {
1901 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1902 	unsigned int ret_freq = 0;
1903 
1904 	if (policy) {
1905 		down_read(&policy->rwsem);
1906 		if (cpufreq_driver->get)
1907 			ret_freq = __cpufreq_get(policy);
1908 		up_read(&policy->rwsem);
1909 
1910 		cpufreq_cpu_put(policy);
1911 	}
1912 
1913 	return ret_freq;
1914 }
1915 EXPORT_SYMBOL(cpufreq_get);
1916 
1917 static struct subsys_interface cpufreq_interface = {
1918 	.name		= "cpufreq",
1919 	.subsys		= &cpu_subsys,
1920 	.add_dev	= cpufreq_add_dev,
1921 	.remove_dev	= cpufreq_remove_dev,
1922 };
1923 
1924 /*
1925  * In case platform wants some specific frequency to be configured
1926  * during suspend..
1927  */
1928 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1929 {
1930 	int ret;
1931 
1932 	if (!policy->suspend_freq) {
1933 		pr_debug("%s: suspend_freq not defined\n", __func__);
1934 		return 0;
1935 	}
1936 
1937 	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1938 			policy->suspend_freq);
1939 
1940 	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1941 			CPUFREQ_RELATION_H);
1942 	if (ret)
1943 		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1944 				__func__, policy->suspend_freq, ret);
1945 
1946 	return ret;
1947 }
1948 EXPORT_SYMBOL(cpufreq_generic_suspend);
1949 
1950 /**
1951  * cpufreq_suspend() - Suspend CPUFreq governors.
1952  *
1953  * Called during system wide Suspend/Hibernate cycles for suspending governors
1954  * as some platforms can't change frequency after this point in suspend cycle.
1955  * Because some of the devices (like: i2c, regulators, etc) they use for
1956  * changing frequency are suspended quickly after this point.
1957  */
1958 void cpufreq_suspend(void)
1959 {
1960 	struct cpufreq_policy *policy;
1961 
1962 	if (!cpufreq_driver)
1963 		return;
1964 
1965 	if (!has_target() && !cpufreq_driver->suspend)
1966 		goto suspend;
1967 
1968 	pr_debug("%s: Suspending Governors\n", __func__);
1969 
1970 	for_each_active_policy(policy) {
1971 		if (has_target()) {
1972 			down_write(&policy->rwsem);
1973 			cpufreq_stop_governor(policy);
1974 			up_write(&policy->rwsem);
1975 		}
1976 
1977 		if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1978 			pr_err("%s: Failed to suspend driver: %s\n", __func__,
1979 				cpufreq_driver->name);
1980 	}
1981 
1982 suspend:
1983 	cpufreq_suspended = true;
1984 }
1985 
1986 /**
1987  * cpufreq_resume() - Resume CPUFreq governors.
1988  *
1989  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1990  * are suspended with cpufreq_suspend().
1991  */
1992 void cpufreq_resume(void)
1993 {
1994 	struct cpufreq_policy *policy;
1995 	int ret;
1996 
1997 	if (!cpufreq_driver)
1998 		return;
1999 
2000 	if (unlikely(!cpufreq_suspended))
2001 		return;
2002 
2003 	cpufreq_suspended = false;
2004 
2005 	if (!has_target() && !cpufreq_driver->resume)
2006 		return;
2007 
2008 	pr_debug("%s: Resuming Governors\n", __func__);
2009 
2010 	for_each_active_policy(policy) {
2011 		if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
2012 			pr_err("%s: Failed to resume driver: %s\n", __func__,
2013 				cpufreq_driver->name);
2014 		} else if (has_target()) {
2015 			down_write(&policy->rwsem);
2016 			ret = cpufreq_start_governor(policy);
2017 			up_write(&policy->rwsem);
2018 
2019 			if (ret)
2020 				pr_err("%s: Failed to start governor for CPU%u's policy\n",
2021 				       __func__, policy->cpu);
2022 		}
2023 	}
2024 }
2025 
2026 /**
2027  * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
2028  * @flags: Flags to test against the current cpufreq driver's flags.
2029  *
2030  * Assumes that the driver is there, so callers must ensure that this is the
2031  * case.
2032  */
2033 bool cpufreq_driver_test_flags(u16 flags)
2034 {
2035 	return !!(cpufreq_driver->flags & flags);
2036 }
2037 
2038 /**
2039  * cpufreq_get_current_driver - Return the current driver's name.
2040  *
2041  * Return the name string of the currently registered cpufreq driver or NULL if
2042  * none.
2043  */
2044 const char *cpufreq_get_current_driver(void)
2045 {
2046 	if (cpufreq_driver)
2047 		return cpufreq_driver->name;
2048 
2049 	return NULL;
2050 }
2051 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
2052 
2053 /**
2054  * cpufreq_get_driver_data - Return current driver data.
2055  *
2056  * Return the private data of the currently registered cpufreq driver, or NULL
2057  * if no cpufreq driver has been registered.
2058  */
2059 void *cpufreq_get_driver_data(void)
2060 {
2061 	if (cpufreq_driver)
2062 		return cpufreq_driver->driver_data;
2063 
2064 	return NULL;
2065 }
2066 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
2067 
2068 /*********************************************************************
2069  *                     NOTIFIER LISTS INTERFACE                      *
2070  *********************************************************************/
2071 
2072 /**
2073  * cpufreq_register_notifier - Register a notifier with cpufreq.
2074  * @nb: notifier function to register.
2075  * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2076  *
2077  * Add a notifier to one of two lists: either a list of notifiers that run on
2078  * clock rate changes (once before and once after every transition), or a list
2079  * of notifiers that ron on cpufreq policy changes.
2080  *
2081  * This function may sleep and it has the same return values as
2082  * blocking_notifier_chain_register().
2083  */
2084 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
2085 {
2086 	int ret;
2087 
2088 	if (cpufreq_disabled())
2089 		return -EINVAL;
2090 
2091 	switch (list) {
2092 	case CPUFREQ_TRANSITION_NOTIFIER:
2093 		mutex_lock(&cpufreq_fast_switch_lock);
2094 
2095 		if (cpufreq_fast_switch_count > 0) {
2096 			mutex_unlock(&cpufreq_fast_switch_lock);
2097 			return -EBUSY;
2098 		}
2099 		ret = srcu_notifier_chain_register(
2100 				&cpufreq_transition_notifier_list, nb);
2101 		if (!ret)
2102 			cpufreq_fast_switch_count--;
2103 
2104 		mutex_unlock(&cpufreq_fast_switch_lock);
2105 		break;
2106 	case CPUFREQ_POLICY_NOTIFIER:
2107 		ret = blocking_notifier_chain_register(
2108 				&cpufreq_policy_notifier_list, nb);
2109 		break;
2110 	default:
2111 		ret = -EINVAL;
2112 	}
2113 
2114 	return ret;
2115 }
2116 EXPORT_SYMBOL(cpufreq_register_notifier);
2117 
2118 /**
2119  * cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2120  * @nb: notifier block to be unregistered.
2121  * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2122  *
2123  * Remove a notifier from one of the cpufreq notifier lists.
2124  *
2125  * This function may sleep and it has the same return values as
2126  * blocking_notifier_chain_unregister().
2127  */
2128 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2129 {
2130 	int ret;
2131 
2132 	if (cpufreq_disabled())
2133 		return -EINVAL;
2134 
2135 	switch (list) {
2136 	case CPUFREQ_TRANSITION_NOTIFIER:
2137 		mutex_lock(&cpufreq_fast_switch_lock);
2138 
2139 		ret = srcu_notifier_chain_unregister(
2140 				&cpufreq_transition_notifier_list, nb);
2141 		if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2142 			cpufreq_fast_switch_count++;
2143 
2144 		mutex_unlock(&cpufreq_fast_switch_lock);
2145 		break;
2146 	case CPUFREQ_POLICY_NOTIFIER:
2147 		ret = blocking_notifier_chain_unregister(
2148 				&cpufreq_policy_notifier_list, nb);
2149 		break;
2150 	default:
2151 		ret = -EINVAL;
2152 	}
2153 
2154 	return ret;
2155 }
2156 EXPORT_SYMBOL(cpufreq_unregister_notifier);
2157 
2158 
2159 /*********************************************************************
2160  *                              GOVERNORS                            *
2161  *********************************************************************/
2162 
2163 /**
2164  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2165  * @policy: cpufreq policy to switch the frequency for.
2166  * @target_freq: New frequency to set (may be approximate).
2167  *
2168  * Carry out a fast frequency switch without sleeping.
2169  *
2170  * The driver's ->fast_switch() callback invoked by this function must be
2171  * suitable for being called from within RCU-sched read-side critical sections
2172  * and it is expected to select the minimum available frequency greater than or
2173  * equal to @target_freq (CPUFREQ_RELATION_L).
2174  *
2175  * This function must not be called if policy->fast_switch_enabled is unset.
2176  *
2177  * Governors calling this function must guarantee that it will never be invoked
2178  * twice in parallel for the same policy and that it will never be called in
2179  * parallel with either ->target() or ->target_index() for the same policy.
2180  *
2181  * Returns the actual frequency set for the CPU.
2182  *
2183  * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2184  * error condition, the hardware configuration must be preserved.
2185  */
2186 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2187 					unsigned int target_freq)
2188 {
2189 	unsigned int freq;
2190 	int cpu;
2191 
2192 	target_freq = clamp_val(target_freq, policy->min, policy->max);
2193 	freq = cpufreq_driver->fast_switch(policy, target_freq);
2194 
2195 	if (!freq)
2196 		return 0;
2197 
2198 	policy->cur = freq;
2199 	arch_set_freq_scale(policy->related_cpus, freq,
2200 			    arch_scale_freq_ref(policy->cpu));
2201 	cpufreq_stats_record_transition(policy, freq);
2202 
2203 	if (trace_cpu_frequency_enabled()) {
2204 		for_each_cpu(cpu, policy->cpus)
2205 			trace_cpu_frequency(freq, cpu);
2206 	}
2207 
2208 	return freq;
2209 }
2210 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2211 
2212 /**
2213  * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2214  * @cpu: Target CPU.
2215  * @min_perf: Minimum (required) performance level (units of @capacity).
2216  * @target_perf: Target (desired) performance level (units of @capacity).
2217  * @capacity: Capacity of the target CPU.
2218  *
2219  * Carry out a fast performance level switch of @cpu without sleeping.
2220  *
2221  * The driver's ->adjust_perf() callback invoked by this function must be
2222  * suitable for being called from within RCU-sched read-side critical sections
2223  * and it is expected to select a suitable performance level equal to or above
2224  * @min_perf and preferably equal to or below @target_perf.
2225  *
2226  * This function must not be called if policy->fast_switch_enabled is unset.
2227  *
2228  * Governors calling this function must guarantee that it will never be invoked
2229  * twice in parallel for the same CPU and that it will never be called in
2230  * parallel with either ->target() or ->target_index() or ->fast_switch() for
2231  * the same CPU.
2232  */
2233 void cpufreq_driver_adjust_perf(unsigned int cpu,
2234 				 unsigned long min_perf,
2235 				 unsigned long target_perf,
2236 				 unsigned long capacity)
2237 {
2238 	cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2239 }
2240 
2241 /**
2242  * cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2243  *
2244  * Return 'true' if the ->adjust_perf callback is present for the
2245  * current driver or 'false' otherwise.
2246  */
2247 bool cpufreq_driver_has_adjust_perf(void)
2248 {
2249 	return !!cpufreq_driver->adjust_perf;
2250 }
2251 
2252 /* Must set freqs->new to intermediate frequency */
2253 static int __target_intermediate(struct cpufreq_policy *policy,
2254 				 struct cpufreq_freqs *freqs, int index)
2255 {
2256 	int ret;
2257 
2258 	freqs->new = cpufreq_driver->get_intermediate(policy, index);
2259 
2260 	/* We don't need to switch to intermediate freq */
2261 	if (!freqs->new)
2262 		return 0;
2263 
2264 	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2265 		 __func__, policy->cpu, freqs->old, freqs->new);
2266 
2267 	cpufreq_freq_transition_begin(policy, freqs);
2268 	ret = cpufreq_driver->target_intermediate(policy, index);
2269 	cpufreq_freq_transition_end(policy, freqs, ret);
2270 
2271 	if (ret)
2272 		pr_err("%s: Failed to change to intermediate frequency: %d\n",
2273 		       __func__, ret);
2274 
2275 	return ret;
2276 }
2277 
2278 static int __target_index(struct cpufreq_policy *policy, int index)
2279 {
2280 	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2281 	unsigned int restore_freq, intermediate_freq = 0;
2282 	unsigned int newfreq = policy->freq_table[index].frequency;
2283 	int retval = -EINVAL;
2284 	bool notify;
2285 
2286 	if (newfreq == policy->cur)
2287 		return 0;
2288 
2289 	/* Save last value to restore later on errors */
2290 	restore_freq = policy->cur;
2291 
2292 	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2293 	if (notify) {
2294 		/* Handle switching to intermediate frequency */
2295 		if (cpufreq_driver->get_intermediate) {
2296 			retval = __target_intermediate(policy, &freqs, index);
2297 			if (retval)
2298 				return retval;
2299 
2300 			intermediate_freq = freqs.new;
2301 			/* Set old freq to intermediate */
2302 			if (intermediate_freq)
2303 				freqs.old = freqs.new;
2304 		}
2305 
2306 		freqs.new = newfreq;
2307 		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2308 			 __func__, policy->cpu, freqs.old, freqs.new);
2309 
2310 		cpufreq_freq_transition_begin(policy, &freqs);
2311 	}
2312 
2313 	retval = cpufreq_driver->target_index(policy, index);
2314 	if (retval)
2315 		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2316 		       retval);
2317 
2318 	if (notify) {
2319 		cpufreq_freq_transition_end(policy, &freqs, retval);
2320 
2321 		/*
2322 		 * Failed after setting to intermediate freq? Driver should have
2323 		 * reverted back to initial frequency and so should we. Check
2324 		 * here for intermediate_freq instead of get_intermediate, in
2325 		 * case we haven't switched to intermediate freq at all.
2326 		 */
2327 		if (unlikely(retval && intermediate_freq)) {
2328 			freqs.old = intermediate_freq;
2329 			freqs.new = restore_freq;
2330 			cpufreq_freq_transition_begin(policy, &freqs);
2331 			cpufreq_freq_transition_end(policy, &freqs, 0);
2332 		}
2333 	}
2334 
2335 	return retval;
2336 }
2337 
2338 int __cpufreq_driver_target(struct cpufreq_policy *policy,
2339 			    unsigned int target_freq,
2340 			    unsigned int relation)
2341 {
2342 	unsigned int old_target_freq = target_freq;
2343 
2344 	if (cpufreq_disabled())
2345 		return -ENODEV;
2346 
2347 	target_freq = __resolve_freq(policy, target_freq, relation);
2348 
2349 	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2350 		 policy->cpu, target_freq, relation, old_target_freq);
2351 
2352 	/*
2353 	 * This might look like a redundant call as we are checking it again
2354 	 * after finding index. But it is left intentionally for cases where
2355 	 * exactly same freq is called again and so we can save on few function
2356 	 * calls.
2357 	 */
2358 	if (target_freq == policy->cur &&
2359 	    !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2360 		return 0;
2361 
2362 	if (cpufreq_driver->target) {
2363 		/*
2364 		 * If the driver hasn't setup a single inefficient frequency,
2365 		 * it's unlikely it knows how to decode CPUFREQ_RELATION_E.
2366 		 */
2367 		if (!policy->efficiencies_available)
2368 			relation &= ~CPUFREQ_RELATION_E;
2369 
2370 		return cpufreq_driver->target(policy, target_freq, relation);
2371 	}
2372 
2373 	if (!cpufreq_driver->target_index)
2374 		return -EINVAL;
2375 
2376 	return __target_index(policy, policy->cached_resolved_idx);
2377 }
2378 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2379 
2380 int cpufreq_driver_target(struct cpufreq_policy *policy,
2381 			  unsigned int target_freq,
2382 			  unsigned int relation)
2383 {
2384 	int ret;
2385 
2386 	down_write(&policy->rwsem);
2387 
2388 	ret = __cpufreq_driver_target(policy, target_freq, relation);
2389 
2390 	up_write(&policy->rwsem);
2391 
2392 	return ret;
2393 }
2394 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2395 
2396 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2397 {
2398 	return NULL;
2399 }
2400 
2401 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2402 {
2403 	int ret;
2404 
2405 	/* Don't start any governor operations if we are entering suspend */
2406 	if (cpufreq_suspended)
2407 		return 0;
2408 	/*
2409 	 * Governor might not be initiated here if ACPI _PPC changed
2410 	 * notification happened, so check it.
2411 	 */
2412 	if (!policy->governor)
2413 		return -EINVAL;
2414 
2415 	/* Platform doesn't want dynamic frequency switching ? */
2416 	if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2417 	    cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2418 		struct cpufreq_governor *gov = cpufreq_fallback_governor();
2419 
2420 		if (gov) {
2421 			pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2422 				policy->governor->name, gov->name);
2423 			policy->governor = gov;
2424 		} else {
2425 			return -EINVAL;
2426 		}
2427 	}
2428 
2429 	if (!try_module_get(policy->governor->owner))
2430 		return -EINVAL;
2431 
2432 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2433 
2434 	if (policy->governor->init) {
2435 		ret = policy->governor->init(policy);
2436 		if (ret) {
2437 			module_put(policy->governor->owner);
2438 			return ret;
2439 		}
2440 	}
2441 
2442 	policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2443 
2444 	return 0;
2445 }
2446 
2447 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2448 {
2449 	if (cpufreq_suspended || !policy->governor)
2450 		return;
2451 
2452 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2453 
2454 	if (policy->governor->exit)
2455 		policy->governor->exit(policy);
2456 
2457 	module_put(policy->governor->owner);
2458 }
2459 
2460 int cpufreq_start_governor(struct cpufreq_policy *policy)
2461 {
2462 	int ret;
2463 
2464 	if (cpufreq_suspended)
2465 		return 0;
2466 
2467 	if (!policy->governor)
2468 		return -EINVAL;
2469 
2470 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2471 
2472 	if (cpufreq_driver->get)
2473 		cpufreq_verify_current_freq(policy, false);
2474 
2475 	if (policy->governor->start) {
2476 		ret = policy->governor->start(policy);
2477 		if (ret)
2478 			return ret;
2479 	}
2480 
2481 	if (policy->governor->limits)
2482 		policy->governor->limits(policy);
2483 
2484 	return 0;
2485 }
2486 
2487 void cpufreq_stop_governor(struct cpufreq_policy *policy)
2488 {
2489 	if (cpufreq_suspended || !policy->governor)
2490 		return;
2491 
2492 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2493 
2494 	if (policy->governor->stop)
2495 		policy->governor->stop(policy);
2496 }
2497 
2498 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2499 {
2500 	if (cpufreq_suspended || !policy->governor)
2501 		return;
2502 
2503 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2504 
2505 	if (policy->governor->limits)
2506 		policy->governor->limits(policy);
2507 }
2508 
2509 int cpufreq_register_governor(struct cpufreq_governor *governor)
2510 {
2511 	int err;
2512 
2513 	if (!governor)
2514 		return -EINVAL;
2515 
2516 	if (cpufreq_disabled())
2517 		return -ENODEV;
2518 
2519 	mutex_lock(&cpufreq_governor_mutex);
2520 
2521 	err = -EBUSY;
2522 	if (!find_governor(governor->name)) {
2523 		err = 0;
2524 		list_add(&governor->governor_list, &cpufreq_governor_list);
2525 	}
2526 
2527 	mutex_unlock(&cpufreq_governor_mutex);
2528 	return err;
2529 }
2530 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2531 
2532 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2533 {
2534 	struct cpufreq_policy *policy;
2535 	unsigned long flags;
2536 
2537 	if (!governor)
2538 		return;
2539 
2540 	if (cpufreq_disabled())
2541 		return;
2542 
2543 	/* clear last_governor for all inactive policies */
2544 	read_lock_irqsave(&cpufreq_driver_lock, flags);
2545 	for_each_inactive_policy(policy) {
2546 		if (!strcmp(policy->last_governor, governor->name)) {
2547 			policy->governor = NULL;
2548 			strcpy(policy->last_governor, "\0");
2549 		}
2550 	}
2551 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2552 
2553 	mutex_lock(&cpufreq_governor_mutex);
2554 	list_del(&governor->governor_list);
2555 	mutex_unlock(&cpufreq_governor_mutex);
2556 }
2557 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2558 
2559 
2560 /*********************************************************************
2561  *                          POLICY INTERFACE                         *
2562  *********************************************************************/
2563 
2564 /**
2565  * cpufreq_get_policy - get the current cpufreq_policy
2566  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2567  *	is written
2568  * @cpu: CPU to find the policy for
2569  *
2570  * Reads the current cpufreq policy.
2571  */
2572 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2573 {
2574 	struct cpufreq_policy *cpu_policy;
2575 	if (!policy)
2576 		return -EINVAL;
2577 
2578 	cpu_policy = cpufreq_cpu_get(cpu);
2579 	if (!cpu_policy)
2580 		return -EINVAL;
2581 
2582 	memcpy(policy, cpu_policy, sizeof(*policy));
2583 
2584 	cpufreq_cpu_put(cpu_policy);
2585 	return 0;
2586 }
2587 EXPORT_SYMBOL(cpufreq_get_policy);
2588 
2589 DEFINE_PER_CPU(unsigned long, cpufreq_pressure);
2590 
2591 /**
2592  * cpufreq_update_pressure() - Update cpufreq pressure for CPUs
2593  * @policy: cpufreq policy of the CPUs.
2594  *
2595  * Update the value of cpufreq pressure for all @cpus in the policy.
2596  */
2597 static void cpufreq_update_pressure(struct cpufreq_policy *policy)
2598 {
2599 	unsigned long max_capacity, capped_freq, pressure;
2600 	u32 max_freq;
2601 	int cpu;
2602 
2603 	cpu = cpumask_first(policy->related_cpus);
2604 	max_freq = arch_scale_freq_ref(cpu);
2605 	capped_freq = policy->max;
2606 
2607 	/*
2608 	 * Handle properly the boost frequencies, which should simply clean
2609 	 * the cpufreq pressure value.
2610 	 */
2611 	if (max_freq <= capped_freq) {
2612 		pressure = 0;
2613 	} else {
2614 		max_capacity = arch_scale_cpu_capacity(cpu);
2615 		pressure = max_capacity -
2616 			   mult_frac(max_capacity, capped_freq, max_freq);
2617 	}
2618 
2619 	for_each_cpu(cpu, policy->related_cpus)
2620 		WRITE_ONCE(per_cpu(cpufreq_pressure, cpu), pressure);
2621 }
2622 
2623 /**
2624  * cpufreq_set_policy - Modify cpufreq policy parameters.
2625  * @policy: Policy object to modify.
2626  * @new_gov: Policy governor pointer.
2627  * @new_pol: Policy value (for drivers with built-in governors).
2628  *
2629  * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2630  * limits to be set for the policy, update @policy with the verified limits
2631  * values and either invoke the driver's ->setpolicy() callback (if present) or
2632  * carry out a governor update for @policy.  That is, run the current governor's
2633  * ->limits() callback (if @new_gov points to the same object as the one in
2634  * @policy) or replace the governor for @policy with @new_gov.
2635  *
2636  * The cpuinfo part of @policy is not updated by this function.
2637  */
2638 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2639 			      struct cpufreq_governor *new_gov,
2640 			      unsigned int new_pol)
2641 {
2642 	struct cpufreq_policy_data new_data;
2643 	struct cpufreq_governor *old_gov;
2644 	int ret;
2645 
2646 	memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2647 	new_data.freq_table = policy->freq_table;
2648 	new_data.cpu = policy->cpu;
2649 	/*
2650 	 * PM QoS framework collects all the requests from users and provide us
2651 	 * the final aggregated value here.
2652 	 */
2653 	new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2654 	new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2655 
2656 	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2657 		 new_data.cpu, new_data.min, new_data.max);
2658 
2659 	/*
2660 	 * Verify that the CPU speed can be set within these limits and make sure
2661 	 * that min <= max.
2662 	 */
2663 	ret = cpufreq_driver->verify(&new_data);
2664 	if (ret)
2665 		return ret;
2666 
2667 	/*
2668 	 * Resolve policy min/max to available frequencies. It ensures
2669 	 * no frequency resolution will neither overshoot the requested maximum
2670 	 * nor undershoot the requested minimum.
2671 	 */
2672 	policy->min = new_data.min;
2673 	policy->max = new_data.max;
2674 	policy->min = __resolve_freq(policy, policy->min, CPUFREQ_RELATION_L);
2675 	policy->max = __resolve_freq(policy, policy->max, CPUFREQ_RELATION_H);
2676 	trace_cpu_frequency_limits(policy);
2677 
2678 	cpufreq_update_pressure(policy);
2679 
2680 	policy->cached_target_freq = UINT_MAX;
2681 
2682 	pr_debug("new min and max freqs are %u - %u kHz\n",
2683 		 policy->min, policy->max);
2684 
2685 	if (cpufreq_driver->setpolicy) {
2686 		policy->policy = new_pol;
2687 		pr_debug("setting range\n");
2688 		return cpufreq_driver->setpolicy(policy);
2689 	}
2690 
2691 	if (new_gov == policy->governor) {
2692 		pr_debug("governor limits update\n");
2693 		cpufreq_governor_limits(policy);
2694 		return 0;
2695 	}
2696 
2697 	pr_debug("governor switch\n");
2698 
2699 	/* save old, working values */
2700 	old_gov = policy->governor;
2701 	/* end old governor */
2702 	if (old_gov) {
2703 		cpufreq_stop_governor(policy);
2704 		cpufreq_exit_governor(policy);
2705 	}
2706 
2707 	/* start new governor */
2708 	policy->governor = new_gov;
2709 	ret = cpufreq_init_governor(policy);
2710 	if (!ret) {
2711 		ret = cpufreq_start_governor(policy);
2712 		if (!ret) {
2713 			pr_debug("governor change\n");
2714 			return 0;
2715 		}
2716 		cpufreq_exit_governor(policy);
2717 	}
2718 
2719 	/* new governor failed, so re-start old one */
2720 	pr_debug("starting governor %s failed\n", policy->governor->name);
2721 	if (old_gov) {
2722 		policy->governor = old_gov;
2723 		if (cpufreq_init_governor(policy))
2724 			policy->governor = NULL;
2725 		else
2726 			cpufreq_start_governor(policy);
2727 	}
2728 
2729 	return ret;
2730 }
2731 
2732 /**
2733  * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2734  * @cpu: CPU to re-evaluate the policy for.
2735  *
2736  * Update the current frequency for the cpufreq policy of @cpu and use
2737  * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2738  * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2739  * for the policy in question, among other things.
2740  */
2741 void cpufreq_update_policy(unsigned int cpu)
2742 {
2743 	struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2744 
2745 	if (!policy)
2746 		return;
2747 
2748 	/*
2749 	 * BIOS might change freq behind our back
2750 	 * -> ask driver for current freq and notify governors about a change
2751 	 */
2752 	if (cpufreq_driver->get && has_target() &&
2753 	    (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2754 		goto unlock;
2755 
2756 	refresh_frequency_limits(policy);
2757 
2758 unlock:
2759 	cpufreq_cpu_release(policy);
2760 }
2761 EXPORT_SYMBOL(cpufreq_update_policy);
2762 
2763 /**
2764  * cpufreq_update_limits - Update policy limits for a given CPU.
2765  * @cpu: CPU to update the policy limits for.
2766  *
2767  * Invoke the driver's ->update_limits callback if present or call
2768  * cpufreq_update_policy() for @cpu.
2769  */
2770 void cpufreq_update_limits(unsigned int cpu)
2771 {
2772 	if (cpufreq_driver->update_limits)
2773 		cpufreq_driver->update_limits(cpu);
2774 	else
2775 		cpufreq_update_policy(cpu);
2776 }
2777 EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2778 
2779 /*********************************************************************
2780  *               BOOST						     *
2781  *********************************************************************/
2782 static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2783 {
2784 	int ret;
2785 
2786 	if (!policy->freq_table)
2787 		return -ENXIO;
2788 
2789 	ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2790 	if (ret) {
2791 		pr_err("%s: Policy frequency update failed\n", __func__);
2792 		return ret;
2793 	}
2794 
2795 	ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2796 	if (ret < 0)
2797 		return ret;
2798 
2799 	return 0;
2800 }
2801 
2802 int cpufreq_boost_trigger_state(int state)
2803 {
2804 	struct cpufreq_policy *policy;
2805 	unsigned long flags;
2806 	int ret = 0;
2807 
2808 	if (cpufreq_driver->boost_enabled == state)
2809 		return 0;
2810 
2811 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2812 	cpufreq_driver->boost_enabled = state;
2813 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2814 
2815 	cpus_read_lock();
2816 	for_each_active_policy(policy) {
2817 		policy->boost_enabled = state;
2818 		ret = cpufreq_driver->set_boost(policy, state);
2819 		if (ret) {
2820 			policy->boost_enabled = !policy->boost_enabled;
2821 			goto err_reset_state;
2822 		}
2823 	}
2824 	cpus_read_unlock();
2825 
2826 	return 0;
2827 
2828 err_reset_state:
2829 	cpus_read_unlock();
2830 
2831 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2832 	cpufreq_driver->boost_enabled = !state;
2833 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2834 
2835 	pr_err("%s: Cannot %s BOOST\n",
2836 	       __func__, state ? "enable" : "disable");
2837 
2838 	return ret;
2839 }
2840 
2841 static bool cpufreq_boost_supported(void)
2842 {
2843 	return cpufreq_driver->set_boost;
2844 }
2845 
2846 static int create_boost_sysfs_file(void)
2847 {
2848 	int ret;
2849 
2850 	ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2851 	if (ret)
2852 		pr_err("%s: cannot register global BOOST sysfs file\n",
2853 		       __func__);
2854 
2855 	return ret;
2856 }
2857 
2858 static void remove_boost_sysfs_file(void)
2859 {
2860 	if (cpufreq_boost_supported())
2861 		sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2862 }
2863 
2864 int cpufreq_enable_boost_support(void)
2865 {
2866 	if (!cpufreq_driver)
2867 		return -EINVAL;
2868 
2869 	if (cpufreq_boost_supported())
2870 		return 0;
2871 
2872 	cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2873 
2874 	/* This will get removed on driver unregister */
2875 	return create_boost_sysfs_file();
2876 }
2877 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2878 
2879 int cpufreq_boost_enabled(void)
2880 {
2881 	return cpufreq_driver->boost_enabled;
2882 }
2883 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2884 
2885 /*********************************************************************
2886  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2887  *********************************************************************/
2888 static enum cpuhp_state hp_online;
2889 
2890 static int cpuhp_cpufreq_online(unsigned int cpu)
2891 {
2892 	cpufreq_online(cpu);
2893 
2894 	return 0;
2895 }
2896 
2897 static int cpuhp_cpufreq_offline(unsigned int cpu)
2898 {
2899 	cpufreq_offline(cpu);
2900 
2901 	return 0;
2902 }
2903 
2904 /**
2905  * cpufreq_register_driver - register a CPU Frequency driver
2906  * @driver_data: A struct cpufreq_driver containing the values#
2907  * submitted by the CPU Frequency driver.
2908  *
2909  * Registers a CPU Frequency driver to this core code. This code
2910  * returns zero on success, -EEXIST when another driver got here first
2911  * (and isn't unregistered in the meantime).
2912  *
2913  */
2914 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2915 {
2916 	unsigned long flags;
2917 	int ret;
2918 
2919 	if (cpufreq_disabled())
2920 		return -ENODEV;
2921 
2922 	/*
2923 	 * The cpufreq core depends heavily on the availability of device
2924 	 * structure, make sure they are available before proceeding further.
2925 	 */
2926 	if (!get_cpu_device(0))
2927 		return -EPROBE_DEFER;
2928 
2929 	if (!driver_data || !driver_data->verify || !driver_data->init ||
2930 	    !(driver_data->setpolicy || driver_data->target_index ||
2931 		    driver_data->target) ||
2932 	     (driver_data->setpolicy && (driver_data->target_index ||
2933 		    driver_data->target)) ||
2934 	     (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2935 	     (!driver_data->online != !driver_data->offline) ||
2936 		 (driver_data->adjust_perf && !driver_data->fast_switch))
2937 		return -EINVAL;
2938 
2939 	pr_debug("trying to register driver %s\n", driver_data->name);
2940 
2941 	/* Protect against concurrent CPU online/offline. */
2942 	cpus_read_lock();
2943 
2944 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2945 	if (cpufreq_driver) {
2946 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2947 		ret = -EEXIST;
2948 		goto out;
2949 	}
2950 	cpufreq_driver = driver_data;
2951 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2952 
2953 	/*
2954 	 * Mark support for the scheduler's frequency invariance engine for
2955 	 * drivers that implement target(), target_index() or fast_switch().
2956 	 */
2957 	if (!cpufreq_driver->setpolicy) {
2958 		static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2959 		pr_debug("supports frequency invariance");
2960 	}
2961 
2962 	if (driver_data->setpolicy)
2963 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2964 
2965 	if (cpufreq_boost_supported()) {
2966 		ret = create_boost_sysfs_file();
2967 		if (ret)
2968 			goto err_null_driver;
2969 	}
2970 
2971 	ret = subsys_interface_register(&cpufreq_interface);
2972 	if (ret)
2973 		goto err_boost_unreg;
2974 
2975 	if (unlikely(list_empty(&cpufreq_policy_list))) {
2976 		/* if all ->init() calls failed, unregister */
2977 		ret = -ENODEV;
2978 		pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2979 			 driver_data->name);
2980 		goto err_if_unreg;
2981 	}
2982 
2983 	ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2984 						   "cpufreq:online",
2985 						   cpuhp_cpufreq_online,
2986 						   cpuhp_cpufreq_offline);
2987 	if (ret < 0)
2988 		goto err_if_unreg;
2989 	hp_online = ret;
2990 	ret = 0;
2991 
2992 	pr_debug("driver %s up and running\n", driver_data->name);
2993 	goto out;
2994 
2995 err_if_unreg:
2996 	subsys_interface_unregister(&cpufreq_interface);
2997 err_boost_unreg:
2998 	remove_boost_sysfs_file();
2999 err_null_driver:
3000 	write_lock_irqsave(&cpufreq_driver_lock, flags);
3001 	cpufreq_driver = NULL;
3002 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
3003 out:
3004 	cpus_read_unlock();
3005 	return ret;
3006 }
3007 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
3008 
3009 /*
3010  * cpufreq_unregister_driver - unregister the current CPUFreq driver
3011  *
3012  * Unregister the current CPUFreq driver. Only call this if you have
3013  * the right to do so, i.e. if you have succeeded in initialising before!
3014  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
3015  * currently not initialised.
3016  */
3017 void cpufreq_unregister_driver(struct cpufreq_driver *driver)
3018 {
3019 	unsigned long flags;
3020 
3021 	if (WARN_ON(!cpufreq_driver || (driver != cpufreq_driver)))
3022 		return;
3023 
3024 	pr_debug("unregistering driver %s\n", driver->name);
3025 
3026 	/* Protect against concurrent cpu hotplug */
3027 	cpus_read_lock();
3028 	subsys_interface_unregister(&cpufreq_interface);
3029 	remove_boost_sysfs_file();
3030 	static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
3031 	cpuhp_remove_state_nocalls_cpuslocked(hp_online);
3032 
3033 	write_lock_irqsave(&cpufreq_driver_lock, flags);
3034 
3035 	cpufreq_driver = NULL;
3036 
3037 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
3038 	cpus_read_unlock();
3039 }
3040 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
3041 
3042 static int __init cpufreq_core_init(void)
3043 {
3044 	struct cpufreq_governor *gov = cpufreq_default_governor();
3045 	struct device *dev_root;
3046 
3047 	if (cpufreq_disabled())
3048 		return -ENODEV;
3049 
3050 	dev_root = bus_get_dev_root(&cpu_subsys);
3051 	if (dev_root) {
3052 		cpufreq_global_kobject = kobject_create_and_add("cpufreq", &dev_root->kobj);
3053 		put_device(dev_root);
3054 	}
3055 	BUG_ON(!cpufreq_global_kobject);
3056 
3057 	if (!strlen(default_governor))
3058 		strscpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
3059 
3060 	return 0;
3061 }
3062 module_param(off, int, 0444);
3063 module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
3064 core_initcall(cpufreq_core_init);
3065