xref: /openbmc/linux/drivers/cpufreq/cpufreq.c (revision bb6d3fb3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/drivers/cpufreq/cpufreq.c
4  *
5  *  Copyright (C) 2001 Russell King
6  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
7  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
8  *
9  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
10  *	Added handling for CPU hotplug
11  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
12  *	Fix handling for CPU hotplug -- affected CPUs
13  */
14 
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 
17 #include <linux/cpu.h>
18 #include <linux/cpufreq.h>
19 #include <linux/cpu_cooling.h>
20 #include <linux/delay.h>
21 #include <linux/device.h>
22 #include <linux/init.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/module.h>
25 #include <linux/mutex.h>
26 #include <linux/pm_qos.h>
27 #include <linux/slab.h>
28 #include <linux/suspend.h>
29 #include <linux/syscore_ops.h>
30 #include <linux/tick.h>
31 #include <trace/events/power.h>
32 
33 static LIST_HEAD(cpufreq_policy_list);
34 
35 /* Macros to iterate over CPU policies */
36 #define for_each_suitable_policy(__policy, __active)			 \
37 	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
38 		if ((__active) == !policy_is_inactive(__policy))
39 
40 #define for_each_active_policy(__policy)		\
41 	for_each_suitable_policy(__policy, true)
42 #define for_each_inactive_policy(__policy)		\
43 	for_each_suitable_policy(__policy, false)
44 
45 #define for_each_policy(__policy)			\
46 	list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
47 
48 /* Iterate over governors */
49 static LIST_HEAD(cpufreq_governor_list);
50 #define for_each_governor(__governor)				\
51 	list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
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 /* Flag to suspend/resume CPUFreq governors */
63 static bool cpufreq_suspended;
64 
65 static inline bool has_target(void)
66 {
67 	return cpufreq_driver->target_index || cpufreq_driver->target;
68 }
69 
70 /* internal prototypes */
71 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
72 static int cpufreq_init_governor(struct cpufreq_policy *policy);
73 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
74 static int cpufreq_start_governor(struct cpufreq_policy *policy);
75 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
76 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
77 static int cpufreq_set_policy(struct cpufreq_policy *policy,
78 			      struct cpufreq_governor *new_gov,
79 			      unsigned int new_pol);
80 
81 /**
82  * Two notifier lists: the "policy" list is involved in the
83  * validation process for a new CPU frequency policy; the
84  * "transition" list for kernel code that needs to handle
85  * changes to devices when the CPU clock speed changes.
86  * The mutex locks both lists.
87  */
88 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
89 SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
90 
91 static int off __read_mostly;
92 static int cpufreq_disabled(void)
93 {
94 	return off;
95 }
96 void disable_cpufreq(void)
97 {
98 	off = 1;
99 }
100 static DEFINE_MUTEX(cpufreq_governor_mutex);
101 
102 bool have_governor_per_policy(void)
103 {
104 	return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
105 }
106 EXPORT_SYMBOL_GPL(have_governor_per_policy);
107 
108 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
109 {
110 	if (have_governor_per_policy())
111 		return &policy->kobj;
112 	else
113 		return cpufreq_global_kobject;
114 }
115 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
116 
117 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
118 {
119 	struct kernel_cpustat kcpustat;
120 	u64 cur_wall_time;
121 	u64 idle_time;
122 	u64 busy_time;
123 
124 	cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
125 
126 	kcpustat_cpu_fetch(&kcpustat, cpu);
127 
128 	busy_time = kcpustat.cpustat[CPUTIME_USER];
129 	busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
130 	busy_time += kcpustat.cpustat[CPUTIME_IRQ];
131 	busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
132 	busy_time += kcpustat.cpustat[CPUTIME_STEAL];
133 	busy_time += kcpustat.cpustat[CPUTIME_NICE];
134 
135 	idle_time = cur_wall_time - busy_time;
136 	if (wall)
137 		*wall = div_u64(cur_wall_time, NSEC_PER_USEC);
138 
139 	return div_u64(idle_time, NSEC_PER_USEC);
140 }
141 
142 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
143 {
144 	u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
145 
146 	if (idle_time == -1ULL)
147 		return get_cpu_idle_time_jiffy(cpu, wall);
148 	else if (!io_busy)
149 		idle_time += get_cpu_iowait_time_us(cpu, wall);
150 
151 	return idle_time;
152 }
153 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
154 
155 __weak void arch_set_freq_scale(struct cpumask *cpus, unsigned long cur_freq,
156 		unsigned long max_freq)
157 {
158 }
159 EXPORT_SYMBOL_GPL(arch_set_freq_scale);
160 
161 /*
162  * This is a generic cpufreq init() routine which can be used by cpufreq
163  * drivers of SMP systems. It will do following:
164  * - validate & show freq table passed
165  * - set policies transition latency
166  * - policy->cpus with all possible CPUs
167  */
168 void cpufreq_generic_init(struct cpufreq_policy *policy,
169 		struct cpufreq_frequency_table *table,
170 		unsigned int transition_latency)
171 {
172 	policy->freq_table = table;
173 	policy->cpuinfo.transition_latency = transition_latency;
174 
175 	/*
176 	 * The driver only supports the SMP configuration where all processors
177 	 * share the clock and voltage and clock.
178 	 */
179 	cpumask_setall(policy->cpus);
180 }
181 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
182 
183 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
184 {
185 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
186 
187 	return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
188 }
189 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
190 
191 unsigned int cpufreq_generic_get(unsigned int cpu)
192 {
193 	struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
194 
195 	if (!policy || IS_ERR(policy->clk)) {
196 		pr_err("%s: No %s associated to cpu: %d\n",
197 		       __func__, policy ? "clk" : "policy", cpu);
198 		return 0;
199 	}
200 
201 	return clk_get_rate(policy->clk) / 1000;
202 }
203 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
204 
205 /**
206  * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
207  * @cpu: CPU to find the policy for.
208  *
209  * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
210  * the kobject reference counter of that policy.  Return a valid policy on
211  * success or NULL on failure.
212  *
213  * The policy returned by this function has to be released with the help of
214  * cpufreq_cpu_put() to balance its kobject reference counter properly.
215  */
216 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
217 {
218 	struct cpufreq_policy *policy = NULL;
219 	unsigned long flags;
220 
221 	if (WARN_ON(cpu >= nr_cpu_ids))
222 		return NULL;
223 
224 	/* get the cpufreq driver */
225 	read_lock_irqsave(&cpufreq_driver_lock, flags);
226 
227 	if (cpufreq_driver) {
228 		/* get the CPU */
229 		policy = cpufreq_cpu_get_raw(cpu);
230 		if (policy)
231 			kobject_get(&policy->kobj);
232 	}
233 
234 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
235 
236 	return policy;
237 }
238 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
239 
240 /**
241  * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
242  * @policy: cpufreq policy returned by cpufreq_cpu_get().
243  */
244 void cpufreq_cpu_put(struct cpufreq_policy *policy)
245 {
246 	kobject_put(&policy->kobj);
247 }
248 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
249 
250 /**
251  * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
252  * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
253  */
254 void cpufreq_cpu_release(struct cpufreq_policy *policy)
255 {
256 	if (WARN_ON(!policy))
257 		return;
258 
259 	lockdep_assert_held(&policy->rwsem);
260 
261 	up_write(&policy->rwsem);
262 
263 	cpufreq_cpu_put(policy);
264 }
265 
266 /**
267  * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
268  * @cpu: CPU to find the policy for.
269  *
270  * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
271  * if the policy returned by it is not NULL, acquire its rwsem for writing.
272  * Return the policy if it is active or release it and return NULL otherwise.
273  *
274  * The policy returned by this function has to be released with the help of
275  * cpufreq_cpu_release() in order to release its rwsem and balance its usage
276  * counter properly.
277  */
278 struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
279 {
280 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
281 
282 	if (!policy)
283 		return NULL;
284 
285 	down_write(&policy->rwsem);
286 
287 	if (policy_is_inactive(policy)) {
288 		cpufreq_cpu_release(policy);
289 		return NULL;
290 	}
291 
292 	return policy;
293 }
294 
295 /*********************************************************************
296  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
297  *********************************************************************/
298 
299 /**
300  * adjust_jiffies - adjust the system "loops_per_jiffy"
301  *
302  * This function alters the system "loops_per_jiffy" for the clock
303  * speed change. Note that loops_per_jiffy cannot be updated on SMP
304  * systems as each CPU might be scaled differently. So, use the arch
305  * per-CPU loops_per_jiffy value wherever possible.
306  */
307 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
308 {
309 #ifndef CONFIG_SMP
310 	static unsigned long l_p_j_ref;
311 	static unsigned int l_p_j_ref_freq;
312 
313 	if (ci->flags & CPUFREQ_CONST_LOOPS)
314 		return;
315 
316 	if (!l_p_j_ref_freq) {
317 		l_p_j_ref = loops_per_jiffy;
318 		l_p_j_ref_freq = ci->old;
319 		pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
320 			 l_p_j_ref, l_p_j_ref_freq);
321 	}
322 	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
323 		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
324 								ci->new);
325 		pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
326 			 loops_per_jiffy, ci->new);
327 	}
328 #endif
329 }
330 
331 /**
332  * cpufreq_notify_transition - Notify frequency transition and adjust_jiffies.
333  * @policy: cpufreq policy to enable fast frequency switching for.
334  * @freqs: contain details of the frequency update.
335  * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
336  *
337  * This function calls the transition notifiers and the "adjust_jiffies"
338  * function. It is called twice on all CPU frequency changes that have
339  * external effects.
340  */
341 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
342 				      struct cpufreq_freqs *freqs,
343 				      unsigned int state)
344 {
345 	int cpu;
346 
347 	BUG_ON(irqs_disabled());
348 
349 	if (cpufreq_disabled())
350 		return;
351 
352 	freqs->policy = policy;
353 	freqs->flags = cpufreq_driver->flags;
354 	pr_debug("notification %u of frequency transition to %u kHz\n",
355 		 state, freqs->new);
356 
357 	switch (state) {
358 	case CPUFREQ_PRECHANGE:
359 		/*
360 		 * Detect if the driver reported a value as "old frequency"
361 		 * which is not equal to what the cpufreq core thinks is
362 		 * "old frequency".
363 		 */
364 		if (policy->cur && policy->cur != freqs->old) {
365 			pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
366 				 freqs->old, policy->cur);
367 			freqs->old = policy->cur;
368 		}
369 
370 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
371 					 CPUFREQ_PRECHANGE, freqs);
372 
373 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
374 		break;
375 
376 	case CPUFREQ_POSTCHANGE:
377 		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
378 		pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
379 			 cpumask_pr_args(policy->cpus));
380 
381 		for_each_cpu(cpu, policy->cpus)
382 			trace_cpu_frequency(freqs->new, cpu);
383 
384 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
385 					 CPUFREQ_POSTCHANGE, freqs);
386 
387 		cpufreq_stats_record_transition(policy, freqs->new);
388 		policy->cur = freqs->new;
389 	}
390 }
391 
392 /* Do post notifications when there are chances that transition has failed */
393 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
394 		struct cpufreq_freqs *freqs, int transition_failed)
395 {
396 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
397 	if (!transition_failed)
398 		return;
399 
400 	swap(freqs->old, freqs->new);
401 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
402 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
403 }
404 
405 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
406 		struct cpufreq_freqs *freqs)
407 {
408 
409 	/*
410 	 * Catch double invocations of _begin() which lead to self-deadlock.
411 	 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
412 	 * doesn't invoke _begin() on their behalf, and hence the chances of
413 	 * double invocations are very low. Moreover, there are scenarios
414 	 * where these checks can emit false-positive warnings in these
415 	 * drivers; so we avoid that by skipping them altogether.
416 	 */
417 	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
418 				&& current == policy->transition_task);
419 
420 wait:
421 	wait_event(policy->transition_wait, !policy->transition_ongoing);
422 
423 	spin_lock(&policy->transition_lock);
424 
425 	if (unlikely(policy->transition_ongoing)) {
426 		spin_unlock(&policy->transition_lock);
427 		goto wait;
428 	}
429 
430 	policy->transition_ongoing = true;
431 	policy->transition_task = current;
432 
433 	spin_unlock(&policy->transition_lock);
434 
435 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
436 }
437 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
438 
439 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
440 		struct cpufreq_freqs *freqs, int transition_failed)
441 {
442 	if (WARN_ON(!policy->transition_ongoing))
443 		return;
444 
445 	cpufreq_notify_post_transition(policy, freqs, transition_failed);
446 
447 	policy->transition_ongoing = false;
448 	policy->transition_task = NULL;
449 
450 	wake_up(&policy->transition_wait);
451 }
452 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
453 
454 /*
455  * Fast frequency switching status count.  Positive means "enabled", negative
456  * means "disabled" and 0 means "not decided yet".
457  */
458 static int cpufreq_fast_switch_count;
459 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
460 
461 static void cpufreq_list_transition_notifiers(void)
462 {
463 	struct notifier_block *nb;
464 
465 	pr_info("Registered transition notifiers:\n");
466 
467 	mutex_lock(&cpufreq_transition_notifier_list.mutex);
468 
469 	for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
470 		pr_info("%pS\n", nb->notifier_call);
471 
472 	mutex_unlock(&cpufreq_transition_notifier_list.mutex);
473 }
474 
475 /**
476  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
477  * @policy: cpufreq policy to enable fast frequency switching for.
478  *
479  * Try to enable fast frequency switching for @policy.
480  *
481  * The attempt will fail if there is at least one transition notifier registered
482  * at this point, as fast frequency switching is quite fundamentally at odds
483  * with transition notifiers.  Thus if successful, it will make registration of
484  * transition notifiers fail going forward.
485  */
486 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
487 {
488 	lockdep_assert_held(&policy->rwsem);
489 
490 	if (!policy->fast_switch_possible)
491 		return;
492 
493 	mutex_lock(&cpufreq_fast_switch_lock);
494 	if (cpufreq_fast_switch_count >= 0) {
495 		cpufreq_fast_switch_count++;
496 		policy->fast_switch_enabled = true;
497 	} else {
498 		pr_warn("CPU%u: Fast frequency switching not enabled\n",
499 			policy->cpu);
500 		cpufreq_list_transition_notifiers();
501 	}
502 	mutex_unlock(&cpufreq_fast_switch_lock);
503 }
504 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
505 
506 /**
507  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
508  * @policy: cpufreq policy to disable fast frequency switching for.
509  */
510 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
511 {
512 	mutex_lock(&cpufreq_fast_switch_lock);
513 	if (policy->fast_switch_enabled) {
514 		policy->fast_switch_enabled = false;
515 		if (!WARN_ON(cpufreq_fast_switch_count <= 0))
516 			cpufreq_fast_switch_count--;
517 	}
518 	mutex_unlock(&cpufreq_fast_switch_lock);
519 }
520 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
521 
522 /**
523  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
524  * one.
525  * @target_freq: target frequency to resolve.
526  *
527  * The target to driver frequency mapping is cached in the policy.
528  *
529  * Return: Lowest driver-supported frequency greater than or equal to the
530  * given target_freq, subject to policy (min/max) and driver limitations.
531  */
532 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
533 					 unsigned int target_freq)
534 {
535 	target_freq = clamp_val(target_freq, policy->min, policy->max);
536 	policy->cached_target_freq = target_freq;
537 
538 	if (cpufreq_driver->target_index) {
539 		int idx;
540 
541 		idx = cpufreq_frequency_table_target(policy, target_freq,
542 						     CPUFREQ_RELATION_L);
543 		policy->cached_resolved_idx = idx;
544 		return policy->freq_table[idx].frequency;
545 	}
546 
547 	if (cpufreq_driver->resolve_freq)
548 		return cpufreq_driver->resolve_freq(policy, target_freq);
549 
550 	return target_freq;
551 }
552 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
553 
554 unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
555 {
556 	unsigned int latency;
557 
558 	if (policy->transition_delay_us)
559 		return policy->transition_delay_us;
560 
561 	latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
562 	if (latency) {
563 		/*
564 		 * For platforms that can change the frequency very fast (< 10
565 		 * us), the above formula gives a decent transition delay. But
566 		 * for platforms where transition_latency is in milliseconds, it
567 		 * ends up giving unrealistic values.
568 		 *
569 		 * Cap the default transition delay to 10 ms, which seems to be
570 		 * a reasonable amount of time after which we should reevaluate
571 		 * the frequency.
572 		 */
573 		return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
574 	}
575 
576 	return LATENCY_MULTIPLIER;
577 }
578 EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
579 
580 /*********************************************************************
581  *                          SYSFS INTERFACE                          *
582  *********************************************************************/
583 static ssize_t show_boost(struct kobject *kobj,
584 			  struct kobj_attribute *attr, char *buf)
585 {
586 	return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
587 }
588 
589 static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
590 			   const char *buf, size_t count)
591 {
592 	int ret, enable;
593 
594 	ret = sscanf(buf, "%d", &enable);
595 	if (ret != 1 || enable < 0 || enable > 1)
596 		return -EINVAL;
597 
598 	if (cpufreq_boost_trigger_state(enable)) {
599 		pr_err("%s: Cannot %s BOOST!\n",
600 		       __func__, enable ? "enable" : "disable");
601 		return -EINVAL;
602 	}
603 
604 	pr_debug("%s: cpufreq BOOST %s\n",
605 		 __func__, enable ? "enabled" : "disabled");
606 
607 	return count;
608 }
609 define_one_global_rw(boost);
610 
611 static struct cpufreq_governor *find_governor(const char *str_governor)
612 {
613 	struct cpufreq_governor *t;
614 
615 	for_each_governor(t)
616 		if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
617 			return t;
618 
619 	return NULL;
620 }
621 
622 static unsigned int cpufreq_parse_policy(char *str_governor)
623 {
624 	if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
625 		return CPUFREQ_POLICY_PERFORMANCE;
626 
627 	if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
628 		return CPUFREQ_POLICY_POWERSAVE;
629 
630 	return CPUFREQ_POLICY_UNKNOWN;
631 }
632 
633 /**
634  * cpufreq_parse_governor - parse a governor string only for has_target()
635  * @str_governor: Governor name.
636  */
637 static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
638 {
639 	struct cpufreq_governor *t;
640 
641 	mutex_lock(&cpufreq_governor_mutex);
642 
643 	t = find_governor(str_governor);
644 	if (!t) {
645 		int ret;
646 
647 		mutex_unlock(&cpufreq_governor_mutex);
648 
649 		ret = request_module("cpufreq_%s", str_governor);
650 		if (ret)
651 			return NULL;
652 
653 		mutex_lock(&cpufreq_governor_mutex);
654 
655 		t = find_governor(str_governor);
656 	}
657 	if (t && !try_module_get(t->owner))
658 		t = NULL;
659 
660 	mutex_unlock(&cpufreq_governor_mutex);
661 
662 	return t;
663 }
664 
665 /**
666  * cpufreq_per_cpu_attr_read() / show_##file_name() -
667  * print out cpufreq information
668  *
669  * Write out information from cpufreq_driver->policy[cpu]; object must be
670  * "unsigned int".
671  */
672 
673 #define show_one(file_name, object)			\
674 static ssize_t show_##file_name				\
675 (struct cpufreq_policy *policy, char *buf)		\
676 {							\
677 	return sprintf(buf, "%u\n", policy->object);	\
678 }
679 
680 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
681 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
682 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
683 show_one(scaling_min_freq, min);
684 show_one(scaling_max_freq, max);
685 
686 __weak unsigned int arch_freq_get_on_cpu(int cpu)
687 {
688 	return 0;
689 }
690 
691 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
692 {
693 	ssize_t ret;
694 	unsigned int freq;
695 
696 	freq = arch_freq_get_on_cpu(policy->cpu);
697 	if (freq)
698 		ret = sprintf(buf, "%u\n", freq);
699 	else if (cpufreq_driver && cpufreq_driver->setpolicy &&
700 			cpufreq_driver->get)
701 		ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
702 	else
703 		ret = sprintf(buf, "%u\n", policy->cur);
704 	return ret;
705 }
706 
707 /**
708  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
709  */
710 #define store_one(file_name, object)			\
711 static ssize_t store_##file_name					\
712 (struct cpufreq_policy *policy, const char *buf, size_t count)		\
713 {									\
714 	unsigned long val;						\
715 	int ret;							\
716 									\
717 	ret = sscanf(buf, "%lu", &val);					\
718 	if (ret != 1)							\
719 		return -EINVAL;						\
720 									\
721 	ret = freq_qos_update_request(policy->object##_freq_req, val);\
722 	return ret >= 0 ? count : ret;					\
723 }
724 
725 store_one(scaling_min_freq, min);
726 store_one(scaling_max_freq, max);
727 
728 /**
729  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
730  */
731 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
732 					char *buf)
733 {
734 	unsigned int cur_freq = __cpufreq_get(policy);
735 
736 	if (cur_freq)
737 		return sprintf(buf, "%u\n", cur_freq);
738 
739 	return sprintf(buf, "<unknown>\n");
740 }
741 
742 /**
743  * show_scaling_governor - show the current policy for the specified CPU
744  */
745 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
746 {
747 	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
748 		return sprintf(buf, "powersave\n");
749 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
750 		return sprintf(buf, "performance\n");
751 	else if (policy->governor)
752 		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
753 				policy->governor->name);
754 	return -EINVAL;
755 }
756 
757 /**
758  * store_scaling_governor - store policy for the specified CPU
759  */
760 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
761 					const char *buf, size_t count)
762 {
763 	char str_governor[16];
764 	int ret;
765 
766 	ret = sscanf(buf, "%15s", str_governor);
767 	if (ret != 1)
768 		return -EINVAL;
769 
770 	if (cpufreq_driver->setpolicy) {
771 		unsigned int new_pol;
772 
773 		new_pol = cpufreq_parse_policy(str_governor);
774 		if (!new_pol)
775 			return -EINVAL;
776 
777 		ret = cpufreq_set_policy(policy, NULL, new_pol);
778 	} else {
779 		struct cpufreq_governor *new_gov;
780 
781 		new_gov = cpufreq_parse_governor(str_governor);
782 		if (!new_gov)
783 			return -EINVAL;
784 
785 		ret = cpufreq_set_policy(policy, new_gov,
786 					 CPUFREQ_POLICY_UNKNOWN);
787 
788 		module_put(new_gov->owner);
789 	}
790 
791 	return ret ? ret : count;
792 }
793 
794 /**
795  * show_scaling_driver - show the cpufreq driver currently loaded
796  */
797 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
798 {
799 	return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
800 }
801 
802 /**
803  * show_scaling_available_governors - show the available CPUfreq governors
804  */
805 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
806 						char *buf)
807 {
808 	ssize_t i = 0;
809 	struct cpufreq_governor *t;
810 
811 	if (!has_target()) {
812 		i += sprintf(buf, "performance powersave");
813 		goto out;
814 	}
815 
816 	for_each_governor(t) {
817 		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
818 		    - (CPUFREQ_NAME_LEN + 2)))
819 			goto out;
820 		i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
821 	}
822 out:
823 	i += sprintf(&buf[i], "\n");
824 	return i;
825 }
826 
827 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
828 {
829 	ssize_t i = 0;
830 	unsigned int cpu;
831 
832 	for_each_cpu(cpu, mask) {
833 		if (i)
834 			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
835 		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
836 		if (i >= (PAGE_SIZE - 5))
837 			break;
838 	}
839 	i += sprintf(&buf[i], "\n");
840 	return i;
841 }
842 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
843 
844 /**
845  * show_related_cpus - show the CPUs affected by each transition even if
846  * hw coordination is in use
847  */
848 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
849 {
850 	return cpufreq_show_cpus(policy->related_cpus, buf);
851 }
852 
853 /**
854  * show_affected_cpus - show the CPUs affected by each transition
855  */
856 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
857 {
858 	return cpufreq_show_cpus(policy->cpus, buf);
859 }
860 
861 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
862 					const char *buf, size_t count)
863 {
864 	unsigned int freq = 0;
865 	unsigned int ret;
866 
867 	if (!policy->governor || !policy->governor->store_setspeed)
868 		return -EINVAL;
869 
870 	ret = sscanf(buf, "%u", &freq);
871 	if (ret != 1)
872 		return -EINVAL;
873 
874 	policy->governor->store_setspeed(policy, freq);
875 
876 	return count;
877 }
878 
879 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
880 {
881 	if (!policy->governor || !policy->governor->show_setspeed)
882 		return sprintf(buf, "<unsupported>\n");
883 
884 	return policy->governor->show_setspeed(policy, buf);
885 }
886 
887 /**
888  * show_bios_limit - show the current cpufreq HW/BIOS limitation
889  */
890 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
891 {
892 	unsigned int limit;
893 	int ret;
894 	ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
895 	if (!ret)
896 		return sprintf(buf, "%u\n", limit);
897 	return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
898 }
899 
900 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
901 cpufreq_freq_attr_ro(cpuinfo_min_freq);
902 cpufreq_freq_attr_ro(cpuinfo_max_freq);
903 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
904 cpufreq_freq_attr_ro(scaling_available_governors);
905 cpufreq_freq_attr_ro(scaling_driver);
906 cpufreq_freq_attr_ro(scaling_cur_freq);
907 cpufreq_freq_attr_ro(bios_limit);
908 cpufreq_freq_attr_ro(related_cpus);
909 cpufreq_freq_attr_ro(affected_cpus);
910 cpufreq_freq_attr_rw(scaling_min_freq);
911 cpufreq_freq_attr_rw(scaling_max_freq);
912 cpufreq_freq_attr_rw(scaling_governor);
913 cpufreq_freq_attr_rw(scaling_setspeed);
914 
915 static struct attribute *default_attrs[] = {
916 	&cpuinfo_min_freq.attr,
917 	&cpuinfo_max_freq.attr,
918 	&cpuinfo_transition_latency.attr,
919 	&scaling_min_freq.attr,
920 	&scaling_max_freq.attr,
921 	&affected_cpus.attr,
922 	&related_cpus.attr,
923 	&scaling_governor.attr,
924 	&scaling_driver.attr,
925 	&scaling_available_governors.attr,
926 	&scaling_setspeed.attr,
927 	NULL
928 };
929 
930 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
931 #define to_attr(a) container_of(a, struct freq_attr, attr)
932 
933 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
934 {
935 	struct cpufreq_policy *policy = to_policy(kobj);
936 	struct freq_attr *fattr = to_attr(attr);
937 	ssize_t ret;
938 
939 	if (!fattr->show)
940 		return -EIO;
941 
942 	down_read(&policy->rwsem);
943 	ret = fattr->show(policy, buf);
944 	up_read(&policy->rwsem);
945 
946 	return ret;
947 }
948 
949 static ssize_t store(struct kobject *kobj, struct attribute *attr,
950 		     const char *buf, size_t count)
951 {
952 	struct cpufreq_policy *policy = to_policy(kobj);
953 	struct freq_attr *fattr = to_attr(attr);
954 	ssize_t ret = -EINVAL;
955 
956 	if (!fattr->store)
957 		return -EIO;
958 
959 	/*
960 	 * cpus_read_trylock() is used here to work around a circular lock
961 	 * dependency problem with respect to the cpufreq_register_driver().
962 	 */
963 	if (!cpus_read_trylock())
964 		return -EBUSY;
965 
966 	if (cpu_online(policy->cpu)) {
967 		down_write(&policy->rwsem);
968 		ret = fattr->store(policy, buf, count);
969 		up_write(&policy->rwsem);
970 	}
971 
972 	cpus_read_unlock();
973 
974 	return ret;
975 }
976 
977 static void cpufreq_sysfs_release(struct kobject *kobj)
978 {
979 	struct cpufreq_policy *policy = to_policy(kobj);
980 	pr_debug("last reference is dropped\n");
981 	complete(&policy->kobj_unregister);
982 }
983 
984 static const struct sysfs_ops sysfs_ops = {
985 	.show	= show,
986 	.store	= store,
987 };
988 
989 static struct kobj_type ktype_cpufreq = {
990 	.sysfs_ops	= &sysfs_ops,
991 	.default_attrs	= default_attrs,
992 	.release	= cpufreq_sysfs_release,
993 };
994 
995 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
996 {
997 	struct device *dev = get_cpu_device(cpu);
998 
999 	if (unlikely(!dev))
1000 		return;
1001 
1002 	if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1003 		return;
1004 
1005 	dev_dbg(dev, "%s: Adding symlink\n", __func__);
1006 	if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1007 		dev_err(dev, "cpufreq symlink creation failed\n");
1008 }
1009 
1010 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
1011 				   struct device *dev)
1012 {
1013 	dev_dbg(dev, "%s: Removing symlink\n", __func__);
1014 	sysfs_remove_link(&dev->kobj, "cpufreq");
1015 }
1016 
1017 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1018 {
1019 	struct freq_attr **drv_attr;
1020 	int ret = 0;
1021 
1022 	/* set up files for this cpu device */
1023 	drv_attr = cpufreq_driver->attr;
1024 	while (drv_attr && *drv_attr) {
1025 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1026 		if (ret)
1027 			return ret;
1028 		drv_attr++;
1029 	}
1030 	if (cpufreq_driver->get) {
1031 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1032 		if (ret)
1033 			return ret;
1034 	}
1035 
1036 	ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1037 	if (ret)
1038 		return ret;
1039 
1040 	if (cpufreq_driver->bios_limit) {
1041 		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1042 		if (ret)
1043 			return ret;
1044 	}
1045 
1046 	return 0;
1047 }
1048 
1049 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1050 {
1051 	return NULL;
1052 }
1053 
1054 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1055 {
1056 	struct cpufreq_governor *def_gov = cpufreq_default_governor();
1057 	struct cpufreq_governor *gov = NULL;
1058 	unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1059 
1060 	if (has_target()) {
1061 		/* Update policy governor to the one used before hotplug. */
1062 		gov = find_governor(policy->last_governor);
1063 		if (gov) {
1064 			pr_debug("Restoring governor %s for cpu %d\n",
1065 				 policy->governor->name, policy->cpu);
1066 		} else if (def_gov) {
1067 			gov = def_gov;
1068 		} else {
1069 			return -ENODATA;
1070 		}
1071 	} else {
1072 		/* Use the default policy if there is no last_policy. */
1073 		if (policy->last_policy) {
1074 			pol = policy->last_policy;
1075 		} else if (def_gov) {
1076 			pol = cpufreq_parse_policy(def_gov->name);
1077 		} else {
1078 			return -ENODATA;
1079 		}
1080 	}
1081 
1082 	return cpufreq_set_policy(policy, gov, pol);
1083 }
1084 
1085 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1086 {
1087 	int ret = 0;
1088 
1089 	/* Has this CPU been taken care of already? */
1090 	if (cpumask_test_cpu(cpu, policy->cpus))
1091 		return 0;
1092 
1093 	down_write(&policy->rwsem);
1094 	if (has_target())
1095 		cpufreq_stop_governor(policy);
1096 
1097 	cpumask_set_cpu(cpu, policy->cpus);
1098 
1099 	if (has_target()) {
1100 		ret = cpufreq_start_governor(policy);
1101 		if (ret)
1102 			pr_err("%s: Failed to start governor\n", __func__);
1103 	}
1104 	up_write(&policy->rwsem);
1105 	return ret;
1106 }
1107 
1108 void refresh_frequency_limits(struct cpufreq_policy *policy)
1109 {
1110 	if (!policy_is_inactive(policy)) {
1111 		pr_debug("updating policy for CPU %u\n", policy->cpu);
1112 
1113 		cpufreq_set_policy(policy, policy->governor, policy->policy);
1114 	}
1115 }
1116 EXPORT_SYMBOL(refresh_frequency_limits);
1117 
1118 static void handle_update(struct work_struct *work)
1119 {
1120 	struct cpufreq_policy *policy =
1121 		container_of(work, struct cpufreq_policy, update);
1122 
1123 	pr_debug("handle_update for cpu %u called\n", policy->cpu);
1124 	down_write(&policy->rwsem);
1125 	refresh_frequency_limits(policy);
1126 	up_write(&policy->rwsem);
1127 }
1128 
1129 static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1130 				void *data)
1131 {
1132 	struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1133 
1134 	schedule_work(&policy->update);
1135 	return 0;
1136 }
1137 
1138 static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1139 				void *data)
1140 {
1141 	struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1142 
1143 	schedule_work(&policy->update);
1144 	return 0;
1145 }
1146 
1147 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1148 {
1149 	struct kobject *kobj;
1150 	struct completion *cmp;
1151 
1152 	down_write(&policy->rwsem);
1153 	cpufreq_stats_free_table(policy);
1154 	kobj = &policy->kobj;
1155 	cmp = &policy->kobj_unregister;
1156 	up_write(&policy->rwsem);
1157 	kobject_put(kobj);
1158 
1159 	/*
1160 	 * We need to make sure that the underlying kobj is
1161 	 * actually not referenced anymore by anybody before we
1162 	 * proceed with unloading.
1163 	 */
1164 	pr_debug("waiting for dropping of refcount\n");
1165 	wait_for_completion(cmp);
1166 	pr_debug("wait complete\n");
1167 }
1168 
1169 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1170 {
1171 	struct cpufreq_policy *policy;
1172 	struct device *dev = get_cpu_device(cpu);
1173 	int ret;
1174 
1175 	if (!dev)
1176 		return NULL;
1177 
1178 	policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1179 	if (!policy)
1180 		return NULL;
1181 
1182 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1183 		goto err_free_policy;
1184 
1185 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1186 		goto err_free_cpumask;
1187 
1188 	if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1189 		goto err_free_rcpumask;
1190 
1191 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1192 				   cpufreq_global_kobject, "policy%u", cpu);
1193 	if (ret) {
1194 		dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1195 		/*
1196 		 * The entire policy object will be freed below, but the extra
1197 		 * memory allocated for the kobject name needs to be freed by
1198 		 * releasing the kobject.
1199 		 */
1200 		kobject_put(&policy->kobj);
1201 		goto err_free_real_cpus;
1202 	}
1203 
1204 	freq_constraints_init(&policy->constraints);
1205 
1206 	policy->nb_min.notifier_call = cpufreq_notifier_min;
1207 	policy->nb_max.notifier_call = cpufreq_notifier_max;
1208 
1209 	ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1210 				    &policy->nb_min);
1211 	if (ret) {
1212 		dev_err(dev, "Failed to register MIN QoS notifier: %d (%*pbl)\n",
1213 			ret, cpumask_pr_args(policy->cpus));
1214 		goto err_kobj_remove;
1215 	}
1216 
1217 	ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1218 				    &policy->nb_max);
1219 	if (ret) {
1220 		dev_err(dev, "Failed to register MAX QoS notifier: %d (%*pbl)\n",
1221 			ret, cpumask_pr_args(policy->cpus));
1222 		goto err_min_qos_notifier;
1223 	}
1224 
1225 	INIT_LIST_HEAD(&policy->policy_list);
1226 	init_rwsem(&policy->rwsem);
1227 	spin_lock_init(&policy->transition_lock);
1228 	init_waitqueue_head(&policy->transition_wait);
1229 	init_completion(&policy->kobj_unregister);
1230 	INIT_WORK(&policy->update, handle_update);
1231 
1232 	policy->cpu = cpu;
1233 	return policy;
1234 
1235 err_min_qos_notifier:
1236 	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1237 				 &policy->nb_min);
1238 err_kobj_remove:
1239 	cpufreq_policy_put_kobj(policy);
1240 err_free_real_cpus:
1241 	free_cpumask_var(policy->real_cpus);
1242 err_free_rcpumask:
1243 	free_cpumask_var(policy->related_cpus);
1244 err_free_cpumask:
1245 	free_cpumask_var(policy->cpus);
1246 err_free_policy:
1247 	kfree(policy);
1248 
1249 	return NULL;
1250 }
1251 
1252 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1253 {
1254 	unsigned long flags;
1255 	int cpu;
1256 
1257 	/* Remove policy from list */
1258 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1259 	list_del(&policy->policy_list);
1260 
1261 	for_each_cpu(cpu, policy->related_cpus)
1262 		per_cpu(cpufreq_cpu_data, cpu) = NULL;
1263 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1264 
1265 	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1266 				 &policy->nb_max);
1267 	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1268 				 &policy->nb_min);
1269 
1270 	/* Cancel any pending policy->update work before freeing the policy. */
1271 	cancel_work_sync(&policy->update);
1272 
1273 	if (policy->max_freq_req) {
1274 		/*
1275 		 * CPUFREQ_CREATE_POLICY notification is sent only after
1276 		 * successfully adding max_freq_req request.
1277 		 */
1278 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1279 					     CPUFREQ_REMOVE_POLICY, policy);
1280 		freq_qos_remove_request(policy->max_freq_req);
1281 	}
1282 
1283 	freq_qos_remove_request(policy->min_freq_req);
1284 	kfree(policy->min_freq_req);
1285 
1286 	cpufreq_policy_put_kobj(policy);
1287 	free_cpumask_var(policy->real_cpus);
1288 	free_cpumask_var(policy->related_cpus);
1289 	free_cpumask_var(policy->cpus);
1290 	kfree(policy);
1291 }
1292 
1293 static int cpufreq_online(unsigned int cpu)
1294 {
1295 	struct cpufreq_policy *policy;
1296 	bool new_policy;
1297 	unsigned long flags;
1298 	unsigned int j;
1299 	int ret;
1300 
1301 	pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1302 
1303 	/* Check if this CPU already has a policy to manage it */
1304 	policy = per_cpu(cpufreq_cpu_data, cpu);
1305 	if (policy) {
1306 		WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1307 		if (!policy_is_inactive(policy))
1308 			return cpufreq_add_policy_cpu(policy, cpu);
1309 
1310 		/* This is the only online CPU for the policy.  Start over. */
1311 		new_policy = false;
1312 		down_write(&policy->rwsem);
1313 		policy->cpu = cpu;
1314 		policy->governor = NULL;
1315 		up_write(&policy->rwsem);
1316 	} else {
1317 		new_policy = true;
1318 		policy = cpufreq_policy_alloc(cpu);
1319 		if (!policy)
1320 			return -ENOMEM;
1321 	}
1322 
1323 	if (!new_policy && cpufreq_driver->online) {
1324 		ret = cpufreq_driver->online(policy);
1325 		if (ret) {
1326 			pr_debug("%s: %d: initialization failed\n", __func__,
1327 				 __LINE__);
1328 			goto out_exit_policy;
1329 		}
1330 
1331 		/* Recover policy->cpus using related_cpus */
1332 		cpumask_copy(policy->cpus, policy->related_cpus);
1333 	} else {
1334 		cpumask_copy(policy->cpus, cpumask_of(cpu));
1335 
1336 		/*
1337 		 * Call driver. From then on the cpufreq must be able
1338 		 * to accept all calls to ->verify and ->setpolicy for this CPU.
1339 		 */
1340 		ret = cpufreq_driver->init(policy);
1341 		if (ret) {
1342 			pr_debug("%s: %d: initialization failed\n", __func__,
1343 				 __LINE__);
1344 			goto out_free_policy;
1345 		}
1346 
1347 		ret = cpufreq_table_validate_and_sort(policy);
1348 		if (ret)
1349 			goto out_exit_policy;
1350 
1351 		/* related_cpus should at least include policy->cpus. */
1352 		cpumask_copy(policy->related_cpus, policy->cpus);
1353 	}
1354 
1355 	down_write(&policy->rwsem);
1356 	/*
1357 	 * affected cpus must always be the one, which are online. We aren't
1358 	 * managing offline cpus here.
1359 	 */
1360 	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1361 
1362 	if (new_policy) {
1363 		for_each_cpu(j, policy->related_cpus) {
1364 			per_cpu(cpufreq_cpu_data, j) = policy;
1365 			add_cpu_dev_symlink(policy, j);
1366 		}
1367 
1368 		policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1369 					       GFP_KERNEL);
1370 		if (!policy->min_freq_req)
1371 			goto out_destroy_policy;
1372 
1373 		ret = freq_qos_add_request(&policy->constraints,
1374 					   policy->min_freq_req, FREQ_QOS_MIN,
1375 					   policy->min);
1376 		if (ret < 0) {
1377 			/*
1378 			 * So we don't call freq_qos_remove_request() for an
1379 			 * uninitialized request.
1380 			 */
1381 			kfree(policy->min_freq_req);
1382 			policy->min_freq_req = NULL;
1383 			goto out_destroy_policy;
1384 		}
1385 
1386 		/*
1387 		 * This must be initialized right here to avoid calling
1388 		 * freq_qos_remove_request() on uninitialized request in case
1389 		 * of errors.
1390 		 */
1391 		policy->max_freq_req = policy->min_freq_req + 1;
1392 
1393 		ret = freq_qos_add_request(&policy->constraints,
1394 					   policy->max_freq_req, FREQ_QOS_MAX,
1395 					   policy->max);
1396 		if (ret < 0) {
1397 			policy->max_freq_req = NULL;
1398 			goto out_destroy_policy;
1399 		}
1400 
1401 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1402 				CPUFREQ_CREATE_POLICY, policy);
1403 	}
1404 
1405 	if (cpufreq_driver->get && has_target()) {
1406 		policy->cur = cpufreq_driver->get(policy->cpu);
1407 		if (!policy->cur) {
1408 			pr_err("%s: ->get() failed\n", __func__);
1409 			goto out_destroy_policy;
1410 		}
1411 	}
1412 
1413 	/*
1414 	 * Sometimes boot loaders set CPU frequency to a value outside of
1415 	 * frequency table present with cpufreq core. In such cases CPU might be
1416 	 * unstable if it has to run on that frequency for long duration of time
1417 	 * and so its better to set it to a frequency which is specified in
1418 	 * freq-table. This also makes cpufreq stats inconsistent as
1419 	 * cpufreq-stats would fail to register because current frequency of CPU
1420 	 * isn't found in freq-table.
1421 	 *
1422 	 * Because we don't want this change to effect boot process badly, we go
1423 	 * for the next freq which is >= policy->cur ('cur' must be set by now,
1424 	 * otherwise we will end up setting freq to lowest of the table as 'cur'
1425 	 * is initialized to zero).
1426 	 *
1427 	 * We are passing target-freq as "policy->cur - 1" otherwise
1428 	 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1429 	 * equal to target-freq.
1430 	 */
1431 	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1432 	    && has_target()) {
1433 		/* Are we running at unknown frequency ? */
1434 		ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1435 		if (ret == -EINVAL) {
1436 			/* Warn user and fix it */
1437 			pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1438 				__func__, policy->cpu, policy->cur);
1439 			ret = __cpufreq_driver_target(policy, policy->cur - 1,
1440 				CPUFREQ_RELATION_L);
1441 
1442 			/*
1443 			 * Reaching here after boot in a few seconds may not
1444 			 * mean that system will remain stable at "unknown"
1445 			 * frequency for longer duration. Hence, a BUG_ON().
1446 			 */
1447 			BUG_ON(ret);
1448 			pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1449 				__func__, policy->cpu, policy->cur);
1450 		}
1451 	}
1452 
1453 	if (new_policy) {
1454 		ret = cpufreq_add_dev_interface(policy);
1455 		if (ret)
1456 			goto out_destroy_policy;
1457 
1458 		cpufreq_stats_create_table(policy);
1459 
1460 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1461 		list_add(&policy->policy_list, &cpufreq_policy_list);
1462 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1463 	}
1464 
1465 	ret = cpufreq_init_policy(policy);
1466 	if (ret) {
1467 		pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1468 		       __func__, cpu, ret);
1469 		goto out_destroy_policy;
1470 	}
1471 
1472 	up_write(&policy->rwsem);
1473 
1474 	kobject_uevent(&policy->kobj, KOBJ_ADD);
1475 
1476 	/* Callback for handling stuff after policy is ready */
1477 	if (cpufreq_driver->ready)
1478 		cpufreq_driver->ready(policy);
1479 
1480 	if (cpufreq_thermal_control_enabled(cpufreq_driver))
1481 		policy->cdev = of_cpufreq_cooling_register(policy);
1482 
1483 	pr_debug("initialization complete\n");
1484 
1485 	return 0;
1486 
1487 out_destroy_policy:
1488 	for_each_cpu(j, policy->real_cpus)
1489 		remove_cpu_dev_symlink(policy, get_cpu_device(j));
1490 
1491 	up_write(&policy->rwsem);
1492 
1493 out_exit_policy:
1494 	if (cpufreq_driver->exit)
1495 		cpufreq_driver->exit(policy);
1496 
1497 out_free_policy:
1498 	cpufreq_policy_free(policy);
1499 	return ret;
1500 }
1501 
1502 /**
1503  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1504  * @dev: CPU device.
1505  * @sif: Subsystem interface structure pointer (not used)
1506  */
1507 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1508 {
1509 	struct cpufreq_policy *policy;
1510 	unsigned cpu = dev->id;
1511 	int ret;
1512 
1513 	dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1514 
1515 	if (cpu_online(cpu)) {
1516 		ret = cpufreq_online(cpu);
1517 		if (ret)
1518 			return ret;
1519 	}
1520 
1521 	/* Create sysfs link on CPU registration */
1522 	policy = per_cpu(cpufreq_cpu_data, cpu);
1523 	if (policy)
1524 		add_cpu_dev_symlink(policy, cpu);
1525 
1526 	return 0;
1527 }
1528 
1529 static int cpufreq_offline(unsigned int cpu)
1530 {
1531 	struct cpufreq_policy *policy;
1532 	int ret;
1533 
1534 	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1535 
1536 	policy = cpufreq_cpu_get_raw(cpu);
1537 	if (!policy) {
1538 		pr_debug("%s: No cpu_data found\n", __func__);
1539 		return 0;
1540 	}
1541 
1542 	down_write(&policy->rwsem);
1543 	if (has_target())
1544 		cpufreq_stop_governor(policy);
1545 
1546 	cpumask_clear_cpu(cpu, policy->cpus);
1547 
1548 	if (policy_is_inactive(policy)) {
1549 		if (has_target())
1550 			strncpy(policy->last_governor, policy->governor->name,
1551 				CPUFREQ_NAME_LEN);
1552 		else
1553 			policy->last_policy = policy->policy;
1554 	} else if (cpu == policy->cpu) {
1555 		/* Nominate new CPU */
1556 		policy->cpu = cpumask_any(policy->cpus);
1557 	}
1558 
1559 	/* Start governor again for active policy */
1560 	if (!policy_is_inactive(policy)) {
1561 		if (has_target()) {
1562 			ret = cpufreq_start_governor(policy);
1563 			if (ret)
1564 				pr_err("%s: Failed to start governor\n", __func__);
1565 		}
1566 
1567 		goto unlock;
1568 	}
1569 
1570 	if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1571 		cpufreq_cooling_unregister(policy->cdev);
1572 		policy->cdev = NULL;
1573 	}
1574 
1575 	if (cpufreq_driver->stop_cpu)
1576 		cpufreq_driver->stop_cpu(policy);
1577 
1578 	if (has_target())
1579 		cpufreq_exit_governor(policy);
1580 
1581 	/*
1582 	 * Perform the ->offline() during light-weight tear-down, as
1583 	 * that allows fast recovery when the CPU comes back.
1584 	 */
1585 	if (cpufreq_driver->offline) {
1586 		cpufreq_driver->offline(policy);
1587 	} else if (cpufreq_driver->exit) {
1588 		cpufreq_driver->exit(policy);
1589 		policy->freq_table = NULL;
1590 	}
1591 
1592 unlock:
1593 	up_write(&policy->rwsem);
1594 	return 0;
1595 }
1596 
1597 /**
1598  * cpufreq_remove_dev - remove a CPU device
1599  *
1600  * Removes the cpufreq interface for a CPU device.
1601  */
1602 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1603 {
1604 	unsigned int cpu = dev->id;
1605 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1606 
1607 	if (!policy)
1608 		return;
1609 
1610 	if (cpu_online(cpu))
1611 		cpufreq_offline(cpu);
1612 
1613 	cpumask_clear_cpu(cpu, policy->real_cpus);
1614 	remove_cpu_dev_symlink(policy, dev);
1615 
1616 	if (cpumask_empty(policy->real_cpus)) {
1617 		/* We did light-weight exit earlier, do full tear down now */
1618 		if (cpufreq_driver->offline)
1619 			cpufreq_driver->exit(policy);
1620 
1621 		cpufreq_policy_free(policy);
1622 	}
1623 }
1624 
1625 /**
1626  *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1627  *	in deep trouble.
1628  *	@policy: policy managing CPUs
1629  *	@new_freq: CPU frequency the CPU actually runs at
1630  *
1631  *	We adjust to current frequency first, and need to clean up later.
1632  *	So either call to cpufreq_update_policy() or schedule handle_update()).
1633  */
1634 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1635 				unsigned int new_freq)
1636 {
1637 	struct cpufreq_freqs freqs;
1638 
1639 	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1640 		 policy->cur, new_freq);
1641 
1642 	freqs.old = policy->cur;
1643 	freqs.new = new_freq;
1644 
1645 	cpufreq_freq_transition_begin(policy, &freqs);
1646 	cpufreq_freq_transition_end(policy, &freqs, 0);
1647 }
1648 
1649 static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1650 {
1651 	unsigned int new_freq;
1652 
1653 	new_freq = cpufreq_driver->get(policy->cpu);
1654 	if (!new_freq)
1655 		return 0;
1656 
1657 	/*
1658 	 * If fast frequency switching is used with the given policy, the check
1659 	 * against policy->cur is pointless, so skip it in that case.
1660 	 */
1661 	if (policy->fast_switch_enabled || !has_target())
1662 		return new_freq;
1663 
1664 	if (policy->cur != new_freq) {
1665 		cpufreq_out_of_sync(policy, new_freq);
1666 		if (update)
1667 			schedule_work(&policy->update);
1668 	}
1669 
1670 	return new_freq;
1671 }
1672 
1673 /**
1674  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1675  * @cpu: CPU number
1676  *
1677  * This is the last known freq, without actually getting it from the driver.
1678  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1679  */
1680 unsigned int cpufreq_quick_get(unsigned int cpu)
1681 {
1682 	struct cpufreq_policy *policy;
1683 	unsigned int ret_freq = 0;
1684 	unsigned long flags;
1685 
1686 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1687 
1688 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1689 		ret_freq = cpufreq_driver->get(cpu);
1690 		read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1691 		return ret_freq;
1692 	}
1693 
1694 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1695 
1696 	policy = cpufreq_cpu_get(cpu);
1697 	if (policy) {
1698 		ret_freq = policy->cur;
1699 		cpufreq_cpu_put(policy);
1700 	}
1701 
1702 	return ret_freq;
1703 }
1704 EXPORT_SYMBOL(cpufreq_quick_get);
1705 
1706 /**
1707  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1708  * @cpu: CPU number
1709  *
1710  * Just return the max possible frequency for a given CPU.
1711  */
1712 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1713 {
1714 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1715 	unsigned int ret_freq = 0;
1716 
1717 	if (policy) {
1718 		ret_freq = policy->max;
1719 		cpufreq_cpu_put(policy);
1720 	}
1721 
1722 	return ret_freq;
1723 }
1724 EXPORT_SYMBOL(cpufreq_quick_get_max);
1725 
1726 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1727 {
1728 	if (unlikely(policy_is_inactive(policy)))
1729 		return 0;
1730 
1731 	return cpufreq_verify_current_freq(policy, true);
1732 }
1733 
1734 /**
1735  * cpufreq_get - get the current CPU frequency (in kHz)
1736  * @cpu: CPU number
1737  *
1738  * Get the CPU current (static) CPU frequency
1739  */
1740 unsigned int cpufreq_get(unsigned int cpu)
1741 {
1742 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1743 	unsigned int ret_freq = 0;
1744 
1745 	if (policy) {
1746 		down_read(&policy->rwsem);
1747 		if (cpufreq_driver->get)
1748 			ret_freq = __cpufreq_get(policy);
1749 		up_read(&policy->rwsem);
1750 
1751 		cpufreq_cpu_put(policy);
1752 	}
1753 
1754 	return ret_freq;
1755 }
1756 EXPORT_SYMBOL(cpufreq_get);
1757 
1758 static struct subsys_interface cpufreq_interface = {
1759 	.name		= "cpufreq",
1760 	.subsys		= &cpu_subsys,
1761 	.add_dev	= cpufreq_add_dev,
1762 	.remove_dev	= cpufreq_remove_dev,
1763 };
1764 
1765 /*
1766  * In case platform wants some specific frequency to be configured
1767  * during suspend..
1768  */
1769 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1770 {
1771 	int ret;
1772 
1773 	if (!policy->suspend_freq) {
1774 		pr_debug("%s: suspend_freq not defined\n", __func__);
1775 		return 0;
1776 	}
1777 
1778 	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1779 			policy->suspend_freq);
1780 
1781 	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1782 			CPUFREQ_RELATION_H);
1783 	if (ret)
1784 		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1785 				__func__, policy->suspend_freq, ret);
1786 
1787 	return ret;
1788 }
1789 EXPORT_SYMBOL(cpufreq_generic_suspend);
1790 
1791 /**
1792  * cpufreq_suspend() - Suspend CPUFreq governors
1793  *
1794  * Called during system wide Suspend/Hibernate cycles for suspending governors
1795  * as some platforms can't change frequency after this point in suspend cycle.
1796  * Because some of the devices (like: i2c, regulators, etc) they use for
1797  * changing frequency are suspended quickly after this point.
1798  */
1799 void cpufreq_suspend(void)
1800 {
1801 	struct cpufreq_policy *policy;
1802 
1803 	if (!cpufreq_driver)
1804 		return;
1805 
1806 	if (!has_target() && !cpufreq_driver->suspend)
1807 		goto suspend;
1808 
1809 	pr_debug("%s: Suspending Governors\n", __func__);
1810 
1811 	for_each_active_policy(policy) {
1812 		if (has_target()) {
1813 			down_write(&policy->rwsem);
1814 			cpufreq_stop_governor(policy);
1815 			up_write(&policy->rwsem);
1816 		}
1817 
1818 		if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1819 			pr_err("%s: Failed to suspend driver: %s\n", __func__,
1820 				cpufreq_driver->name);
1821 	}
1822 
1823 suspend:
1824 	cpufreq_suspended = true;
1825 }
1826 
1827 /**
1828  * cpufreq_resume() - Resume CPUFreq governors
1829  *
1830  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1831  * are suspended with cpufreq_suspend().
1832  */
1833 void cpufreq_resume(void)
1834 {
1835 	struct cpufreq_policy *policy;
1836 	int ret;
1837 
1838 	if (!cpufreq_driver)
1839 		return;
1840 
1841 	if (unlikely(!cpufreq_suspended))
1842 		return;
1843 
1844 	cpufreq_suspended = false;
1845 
1846 	if (!has_target() && !cpufreq_driver->resume)
1847 		return;
1848 
1849 	pr_debug("%s: Resuming Governors\n", __func__);
1850 
1851 	for_each_active_policy(policy) {
1852 		if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1853 			pr_err("%s: Failed to resume driver: %p\n", __func__,
1854 				policy);
1855 		} else if (has_target()) {
1856 			down_write(&policy->rwsem);
1857 			ret = cpufreq_start_governor(policy);
1858 			up_write(&policy->rwsem);
1859 
1860 			if (ret)
1861 				pr_err("%s: Failed to start governor for policy: %p\n",
1862 				       __func__, policy);
1863 		}
1864 	}
1865 }
1866 
1867 /**
1868  *	cpufreq_get_current_driver - return current driver's name
1869  *
1870  *	Return the name string of the currently loaded cpufreq driver
1871  *	or NULL, if none.
1872  */
1873 const char *cpufreq_get_current_driver(void)
1874 {
1875 	if (cpufreq_driver)
1876 		return cpufreq_driver->name;
1877 
1878 	return NULL;
1879 }
1880 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1881 
1882 /**
1883  *	cpufreq_get_driver_data - return current driver data
1884  *
1885  *	Return the private data of the currently loaded cpufreq
1886  *	driver, or NULL if no cpufreq driver is loaded.
1887  */
1888 void *cpufreq_get_driver_data(void)
1889 {
1890 	if (cpufreq_driver)
1891 		return cpufreq_driver->driver_data;
1892 
1893 	return NULL;
1894 }
1895 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1896 
1897 /*********************************************************************
1898  *                     NOTIFIER LISTS INTERFACE                      *
1899  *********************************************************************/
1900 
1901 /**
1902  *	cpufreq_register_notifier - register a driver with cpufreq
1903  *	@nb: notifier function to register
1904  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1905  *
1906  *	Add a driver to one of two lists: either a list of drivers that
1907  *      are notified about clock rate changes (once before and once after
1908  *      the transition), or a list of drivers that are notified about
1909  *      changes in cpufreq policy.
1910  *
1911  *	This function may sleep, and has the same return conditions as
1912  *	blocking_notifier_chain_register.
1913  */
1914 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1915 {
1916 	int ret;
1917 
1918 	if (cpufreq_disabled())
1919 		return -EINVAL;
1920 
1921 	switch (list) {
1922 	case CPUFREQ_TRANSITION_NOTIFIER:
1923 		mutex_lock(&cpufreq_fast_switch_lock);
1924 
1925 		if (cpufreq_fast_switch_count > 0) {
1926 			mutex_unlock(&cpufreq_fast_switch_lock);
1927 			return -EBUSY;
1928 		}
1929 		ret = srcu_notifier_chain_register(
1930 				&cpufreq_transition_notifier_list, nb);
1931 		if (!ret)
1932 			cpufreq_fast_switch_count--;
1933 
1934 		mutex_unlock(&cpufreq_fast_switch_lock);
1935 		break;
1936 	case CPUFREQ_POLICY_NOTIFIER:
1937 		ret = blocking_notifier_chain_register(
1938 				&cpufreq_policy_notifier_list, nb);
1939 		break;
1940 	default:
1941 		ret = -EINVAL;
1942 	}
1943 
1944 	return ret;
1945 }
1946 EXPORT_SYMBOL(cpufreq_register_notifier);
1947 
1948 /**
1949  *	cpufreq_unregister_notifier - unregister a driver with cpufreq
1950  *	@nb: notifier block to be unregistered
1951  *	@list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1952  *
1953  *	Remove a driver from the CPU frequency notifier list.
1954  *
1955  *	This function may sleep, and has the same return conditions as
1956  *	blocking_notifier_chain_unregister.
1957  */
1958 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1959 {
1960 	int ret;
1961 
1962 	if (cpufreq_disabled())
1963 		return -EINVAL;
1964 
1965 	switch (list) {
1966 	case CPUFREQ_TRANSITION_NOTIFIER:
1967 		mutex_lock(&cpufreq_fast_switch_lock);
1968 
1969 		ret = srcu_notifier_chain_unregister(
1970 				&cpufreq_transition_notifier_list, nb);
1971 		if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1972 			cpufreq_fast_switch_count++;
1973 
1974 		mutex_unlock(&cpufreq_fast_switch_lock);
1975 		break;
1976 	case CPUFREQ_POLICY_NOTIFIER:
1977 		ret = blocking_notifier_chain_unregister(
1978 				&cpufreq_policy_notifier_list, nb);
1979 		break;
1980 	default:
1981 		ret = -EINVAL;
1982 	}
1983 
1984 	return ret;
1985 }
1986 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1987 
1988 
1989 /*********************************************************************
1990  *                              GOVERNORS                            *
1991  *********************************************************************/
1992 
1993 /**
1994  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1995  * @policy: cpufreq policy to switch the frequency for.
1996  * @target_freq: New frequency to set (may be approximate).
1997  *
1998  * Carry out a fast frequency switch without sleeping.
1999  *
2000  * The driver's ->fast_switch() callback invoked by this function must be
2001  * suitable for being called from within RCU-sched read-side critical sections
2002  * and it is expected to select the minimum available frequency greater than or
2003  * equal to @target_freq (CPUFREQ_RELATION_L).
2004  *
2005  * This function must not be called if policy->fast_switch_enabled is unset.
2006  *
2007  * Governors calling this function must guarantee that it will never be invoked
2008  * twice in parallel for the same policy and that it will never be called in
2009  * parallel with either ->target() or ->target_index() for the same policy.
2010  *
2011  * Returns the actual frequency set for the CPU.
2012  *
2013  * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2014  * error condition, the hardware configuration must be preserved.
2015  */
2016 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2017 					unsigned int target_freq)
2018 {
2019 	target_freq = clamp_val(target_freq, policy->min, policy->max);
2020 
2021 	return cpufreq_driver->fast_switch(policy, target_freq);
2022 }
2023 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2024 
2025 /* Must set freqs->new to intermediate frequency */
2026 static int __target_intermediate(struct cpufreq_policy *policy,
2027 				 struct cpufreq_freqs *freqs, int index)
2028 {
2029 	int ret;
2030 
2031 	freqs->new = cpufreq_driver->get_intermediate(policy, index);
2032 
2033 	/* We don't need to switch to intermediate freq */
2034 	if (!freqs->new)
2035 		return 0;
2036 
2037 	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2038 		 __func__, policy->cpu, freqs->old, freqs->new);
2039 
2040 	cpufreq_freq_transition_begin(policy, freqs);
2041 	ret = cpufreq_driver->target_intermediate(policy, index);
2042 	cpufreq_freq_transition_end(policy, freqs, ret);
2043 
2044 	if (ret)
2045 		pr_err("%s: Failed to change to intermediate frequency: %d\n",
2046 		       __func__, ret);
2047 
2048 	return ret;
2049 }
2050 
2051 static int __target_index(struct cpufreq_policy *policy, int index)
2052 {
2053 	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2054 	unsigned int intermediate_freq = 0;
2055 	unsigned int newfreq = policy->freq_table[index].frequency;
2056 	int retval = -EINVAL;
2057 	bool notify;
2058 
2059 	if (newfreq == policy->cur)
2060 		return 0;
2061 
2062 	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2063 	if (notify) {
2064 		/* Handle switching to intermediate frequency */
2065 		if (cpufreq_driver->get_intermediate) {
2066 			retval = __target_intermediate(policy, &freqs, index);
2067 			if (retval)
2068 				return retval;
2069 
2070 			intermediate_freq = freqs.new;
2071 			/* Set old freq to intermediate */
2072 			if (intermediate_freq)
2073 				freqs.old = freqs.new;
2074 		}
2075 
2076 		freqs.new = newfreq;
2077 		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2078 			 __func__, policy->cpu, freqs.old, freqs.new);
2079 
2080 		cpufreq_freq_transition_begin(policy, &freqs);
2081 	}
2082 
2083 	retval = cpufreq_driver->target_index(policy, index);
2084 	if (retval)
2085 		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2086 		       retval);
2087 
2088 	if (notify) {
2089 		cpufreq_freq_transition_end(policy, &freqs, retval);
2090 
2091 		/*
2092 		 * Failed after setting to intermediate freq? Driver should have
2093 		 * reverted back to initial frequency and so should we. Check
2094 		 * here for intermediate_freq instead of get_intermediate, in
2095 		 * case we haven't switched to intermediate freq at all.
2096 		 */
2097 		if (unlikely(retval && intermediate_freq)) {
2098 			freqs.old = intermediate_freq;
2099 			freqs.new = policy->restore_freq;
2100 			cpufreq_freq_transition_begin(policy, &freqs);
2101 			cpufreq_freq_transition_end(policy, &freqs, 0);
2102 		}
2103 	}
2104 
2105 	return retval;
2106 }
2107 
2108 int __cpufreq_driver_target(struct cpufreq_policy *policy,
2109 			    unsigned int target_freq,
2110 			    unsigned int relation)
2111 {
2112 	unsigned int old_target_freq = target_freq;
2113 	int index;
2114 
2115 	if (cpufreq_disabled())
2116 		return -ENODEV;
2117 
2118 	/* Make sure that target_freq is within supported range */
2119 	target_freq = clamp_val(target_freq, policy->min, policy->max);
2120 
2121 	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2122 		 policy->cpu, target_freq, relation, old_target_freq);
2123 
2124 	/*
2125 	 * This might look like a redundant call as we are checking it again
2126 	 * after finding index. But it is left intentionally for cases where
2127 	 * exactly same freq is called again and so we can save on few function
2128 	 * calls.
2129 	 */
2130 	if (target_freq == policy->cur)
2131 		return 0;
2132 
2133 	/* Save last value to restore later on errors */
2134 	policy->restore_freq = policy->cur;
2135 
2136 	if (cpufreq_driver->target)
2137 		return cpufreq_driver->target(policy, target_freq, relation);
2138 
2139 	if (!cpufreq_driver->target_index)
2140 		return -EINVAL;
2141 
2142 	index = cpufreq_frequency_table_target(policy, target_freq, relation);
2143 
2144 	return __target_index(policy, index);
2145 }
2146 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2147 
2148 int cpufreq_driver_target(struct cpufreq_policy *policy,
2149 			  unsigned int target_freq,
2150 			  unsigned int relation)
2151 {
2152 	int ret;
2153 
2154 	down_write(&policy->rwsem);
2155 
2156 	ret = __cpufreq_driver_target(policy, target_freq, relation);
2157 
2158 	up_write(&policy->rwsem);
2159 
2160 	return ret;
2161 }
2162 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2163 
2164 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2165 {
2166 	return NULL;
2167 }
2168 
2169 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2170 {
2171 	int ret;
2172 
2173 	/* Don't start any governor operations if we are entering suspend */
2174 	if (cpufreq_suspended)
2175 		return 0;
2176 	/*
2177 	 * Governor might not be initiated here if ACPI _PPC changed
2178 	 * notification happened, so check it.
2179 	 */
2180 	if (!policy->governor)
2181 		return -EINVAL;
2182 
2183 	/* Platform doesn't want dynamic frequency switching ? */
2184 	if (policy->governor->dynamic_switching &&
2185 	    cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2186 		struct cpufreq_governor *gov = cpufreq_fallback_governor();
2187 
2188 		if (gov) {
2189 			pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2190 				policy->governor->name, gov->name);
2191 			policy->governor = gov;
2192 		} else {
2193 			return -EINVAL;
2194 		}
2195 	}
2196 
2197 	if (!try_module_get(policy->governor->owner))
2198 		return -EINVAL;
2199 
2200 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2201 
2202 	if (policy->governor->init) {
2203 		ret = policy->governor->init(policy);
2204 		if (ret) {
2205 			module_put(policy->governor->owner);
2206 			return ret;
2207 		}
2208 	}
2209 
2210 	return 0;
2211 }
2212 
2213 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2214 {
2215 	if (cpufreq_suspended || !policy->governor)
2216 		return;
2217 
2218 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2219 
2220 	if (policy->governor->exit)
2221 		policy->governor->exit(policy);
2222 
2223 	module_put(policy->governor->owner);
2224 }
2225 
2226 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2227 {
2228 	int ret;
2229 
2230 	if (cpufreq_suspended)
2231 		return 0;
2232 
2233 	if (!policy->governor)
2234 		return -EINVAL;
2235 
2236 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2237 
2238 	if (cpufreq_driver->get)
2239 		cpufreq_verify_current_freq(policy, false);
2240 
2241 	if (policy->governor->start) {
2242 		ret = policy->governor->start(policy);
2243 		if (ret)
2244 			return ret;
2245 	}
2246 
2247 	if (policy->governor->limits)
2248 		policy->governor->limits(policy);
2249 
2250 	return 0;
2251 }
2252 
2253 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2254 {
2255 	if (cpufreq_suspended || !policy->governor)
2256 		return;
2257 
2258 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2259 
2260 	if (policy->governor->stop)
2261 		policy->governor->stop(policy);
2262 }
2263 
2264 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2265 {
2266 	if (cpufreq_suspended || !policy->governor)
2267 		return;
2268 
2269 	pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2270 
2271 	if (policy->governor->limits)
2272 		policy->governor->limits(policy);
2273 }
2274 
2275 int cpufreq_register_governor(struct cpufreq_governor *governor)
2276 {
2277 	int err;
2278 
2279 	if (!governor)
2280 		return -EINVAL;
2281 
2282 	if (cpufreq_disabled())
2283 		return -ENODEV;
2284 
2285 	mutex_lock(&cpufreq_governor_mutex);
2286 
2287 	err = -EBUSY;
2288 	if (!find_governor(governor->name)) {
2289 		err = 0;
2290 		list_add(&governor->governor_list, &cpufreq_governor_list);
2291 	}
2292 
2293 	mutex_unlock(&cpufreq_governor_mutex);
2294 	return err;
2295 }
2296 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2297 
2298 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2299 {
2300 	struct cpufreq_policy *policy;
2301 	unsigned long flags;
2302 
2303 	if (!governor)
2304 		return;
2305 
2306 	if (cpufreq_disabled())
2307 		return;
2308 
2309 	/* clear last_governor for all inactive policies */
2310 	read_lock_irqsave(&cpufreq_driver_lock, flags);
2311 	for_each_inactive_policy(policy) {
2312 		if (!strcmp(policy->last_governor, governor->name)) {
2313 			policy->governor = NULL;
2314 			strcpy(policy->last_governor, "\0");
2315 		}
2316 	}
2317 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2318 
2319 	mutex_lock(&cpufreq_governor_mutex);
2320 	list_del(&governor->governor_list);
2321 	mutex_unlock(&cpufreq_governor_mutex);
2322 }
2323 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2324 
2325 
2326 /*********************************************************************
2327  *                          POLICY INTERFACE                         *
2328  *********************************************************************/
2329 
2330 /**
2331  * cpufreq_get_policy - get the current cpufreq_policy
2332  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2333  *	is written
2334  *
2335  * Reads the current cpufreq policy.
2336  */
2337 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2338 {
2339 	struct cpufreq_policy *cpu_policy;
2340 	if (!policy)
2341 		return -EINVAL;
2342 
2343 	cpu_policy = cpufreq_cpu_get(cpu);
2344 	if (!cpu_policy)
2345 		return -EINVAL;
2346 
2347 	memcpy(policy, cpu_policy, sizeof(*policy));
2348 
2349 	cpufreq_cpu_put(cpu_policy);
2350 	return 0;
2351 }
2352 EXPORT_SYMBOL(cpufreq_get_policy);
2353 
2354 /**
2355  * cpufreq_set_policy - Modify cpufreq policy parameters.
2356  * @policy: Policy object to modify.
2357  * @new_gov: Policy governor pointer.
2358  * @new_pol: Policy value (for drivers with built-in governors).
2359  *
2360  * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2361  * limits to be set for the policy, update @policy with the verified limits
2362  * values and either invoke the driver's ->setpolicy() callback (if present) or
2363  * carry out a governor update for @policy.  That is, run the current governor's
2364  * ->limits() callback (if @new_gov points to the same object as the one in
2365  * @policy) or replace the governor for @policy with @new_gov.
2366  *
2367  * The cpuinfo part of @policy is not updated by this function.
2368  */
2369 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2370 			      struct cpufreq_governor *new_gov,
2371 			      unsigned int new_pol)
2372 {
2373 	struct cpufreq_policy_data new_data;
2374 	struct cpufreq_governor *old_gov;
2375 	int ret;
2376 
2377 	memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2378 	new_data.freq_table = policy->freq_table;
2379 	new_data.cpu = policy->cpu;
2380 	/*
2381 	 * PM QoS framework collects all the requests from users and provide us
2382 	 * the final aggregated value here.
2383 	 */
2384 	new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2385 	new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2386 
2387 	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2388 		 new_data.cpu, new_data.min, new_data.max);
2389 
2390 	/*
2391 	 * Verify that the CPU speed can be set within these limits and make sure
2392 	 * that min <= max.
2393 	 */
2394 	ret = cpufreq_driver->verify(&new_data);
2395 	if (ret)
2396 		return ret;
2397 
2398 	policy->min = new_data.min;
2399 	policy->max = new_data.max;
2400 	trace_cpu_frequency_limits(policy);
2401 
2402 	policy->cached_target_freq = UINT_MAX;
2403 
2404 	pr_debug("new min and max freqs are %u - %u kHz\n",
2405 		 policy->min, policy->max);
2406 
2407 	if (cpufreq_driver->setpolicy) {
2408 		policy->policy = new_pol;
2409 		pr_debug("setting range\n");
2410 		return cpufreq_driver->setpolicy(policy);
2411 	}
2412 
2413 	if (new_gov == policy->governor) {
2414 		pr_debug("governor limits update\n");
2415 		cpufreq_governor_limits(policy);
2416 		return 0;
2417 	}
2418 
2419 	pr_debug("governor switch\n");
2420 
2421 	/* save old, working values */
2422 	old_gov = policy->governor;
2423 	/* end old governor */
2424 	if (old_gov) {
2425 		cpufreq_stop_governor(policy);
2426 		cpufreq_exit_governor(policy);
2427 	}
2428 
2429 	/* start new governor */
2430 	policy->governor = new_gov;
2431 	ret = cpufreq_init_governor(policy);
2432 	if (!ret) {
2433 		ret = cpufreq_start_governor(policy);
2434 		if (!ret) {
2435 			pr_debug("governor change\n");
2436 			sched_cpufreq_governor_change(policy, old_gov);
2437 			return 0;
2438 		}
2439 		cpufreq_exit_governor(policy);
2440 	}
2441 
2442 	/* new governor failed, so re-start old one */
2443 	pr_debug("starting governor %s failed\n", policy->governor->name);
2444 	if (old_gov) {
2445 		policy->governor = old_gov;
2446 		if (cpufreq_init_governor(policy))
2447 			policy->governor = NULL;
2448 		else
2449 			cpufreq_start_governor(policy);
2450 	}
2451 
2452 	return ret;
2453 }
2454 
2455 /**
2456  * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2457  * @cpu: CPU to re-evaluate the policy for.
2458  *
2459  * Update the current frequency for the cpufreq policy of @cpu and use
2460  * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2461  * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2462  * for the policy in question, among other things.
2463  */
2464 void cpufreq_update_policy(unsigned int cpu)
2465 {
2466 	struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2467 
2468 	if (!policy)
2469 		return;
2470 
2471 	/*
2472 	 * BIOS might change freq behind our back
2473 	 * -> ask driver for current freq and notify governors about a change
2474 	 */
2475 	if (cpufreq_driver->get && has_target() &&
2476 	    (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2477 		goto unlock;
2478 
2479 	refresh_frequency_limits(policy);
2480 
2481 unlock:
2482 	cpufreq_cpu_release(policy);
2483 }
2484 EXPORT_SYMBOL(cpufreq_update_policy);
2485 
2486 /**
2487  * cpufreq_update_limits - Update policy limits for a given CPU.
2488  * @cpu: CPU to update the policy limits for.
2489  *
2490  * Invoke the driver's ->update_limits callback if present or call
2491  * cpufreq_update_policy() for @cpu.
2492  */
2493 void cpufreq_update_limits(unsigned int cpu)
2494 {
2495 	if (cpufreq_driver->update_limits)
2496 		cpufreq_driver->update_limits(cpu);
2497 	else
2498 		cpufreq_update_policy(cpu);
2499 }
2500 EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2501 
2502 /*********************************************************************
2503  *               BOOST						     *
2504  *********************************************************************/
2505 static int cpufreq_boost_set_sw(int state)
2506 {
2507 	struct cpufreq_policy *policy;
2508 	int ret = -EINVAL;
2509 
2510 	for_each_active_policy(policy) {
2511 		if (!policy->freq_table)
2512 			continue;
2513 
2514 		ret = cpufreq_frequency_table_cpuinfo(policy,
2515 						      policy->freq_table);
2516 		if (ret) {
2517 			pr_err("%s: Policy frequency update failed\n",
2518 			       __func__);
2519 			break;
2520 		}
2521 
2522 		ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2523 		if (ret < 0)
2524 			break;
2525 	}
2526 
2527 	return ret;
2528 }
2529 
2530 int cpufreq_boost_trigger_state(int state)
2531 {
2532 	unsigned long flags;
2533 	int ret = 0;
2534 
2535 	if (cpufreq_driver->boost_enabled == state)
2536 		return 0;
2537 
2538 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2539 	cpufreq_driver->boost_enabled = state;
2540 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2541 
2542 	ret = cpufreq_driver->set_boost(state);
2543 	if (ret) {
2544 		write_lock_irqsave(&cpufreq_driver_lock, flags);
2545 		cpufreq_driver->boost_enabled = !state;
2546 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2547 
2548 		pr_err("%s: Cannot %s BOOST\n",
2549 		       __func__, state ? "enable" : "disable");
2550 	}
2551 
2552 	return ret;
2553 }
2554 
2555 static bool cpufreq_boost_supported(void)
2556 {
2557 	return cpufreq_driver->set_boost;
2558 }
2559 
2560 static int create_boost_sysfs_file(void)
2561 {
2562 	int ret;
2563 
2564 	ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2565 	if (ret)
2566 		pr_err("%s: cannot register global BOOST sysfs file\n",
2567 		       __func__);
2568 
2569 	return ret;
2570 }
2571 
2572 static void remove_boost_sysfs_file(void)
2573 {
2574 	if (cpufreq_boost_supported())
2575 		sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2576 }
2577 
2578 int cpufreq_enable_boost_support(void)
2579 {
2580 	if (!cpufreq_driver)
2581 		return -EINVAL;
2582 
2583 	if (cpufreq_boost_supported())
2584 		return 0;
2585 
2586 	cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2587 
2588 	/* This will get removed on driver unregister */
2589 	return create_boost_sysfs_file();
2590 }
2591 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2592 
2593 int cpufreq_boost_enabled(void)
2594 {
2595 	return cpufreq_driver->boost_enabled;
2596 }
2597 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2598 
2599 /*********************************************************************
2600  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2601  *********************************************************************/
2602 static enum cpuhp_state hp_online;
2603 
2604 static int cpuhp_cpufreq_online(unsigned int cpu)
2605 {
2606 	cpufreq_online(cpu);
2607 
2608 	return 0;
2609 }
2610 
2611 static int cpuhp_cpufreq_offline(unsigned int cpu)
2612 {
2613 	cpufreq_offline(cpu);
2614 
2615 	return 0;
2616 }
2617 
2618 /**
2619  * cpufreq_register_driver - register a CPU Frequency driver
2620  * @driver_data: A struct cpufreq_driver containing the values#
2621  * submitted by the CPU Frequency driver.
2622  *
2623  * Registers a CPU Frequency driver to this core code. This code
2624  * returns zero on success, -EEXIST when another driver got here first
2625  * (and isn't unregistered in the meantime).
2626  *
2627  */
2628 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2629 {
2630 	unsigned long flags;
2631 	int ret;
2632 
2633 	if (cpufreq_disabled())
2634 		return -ENODEV;
2635 
2636 	/*
2637 	 * The cpufreq core depends heavily on the availability of device
2638 	 * structure, make sure they are available before proceeding further.
2639 	 */
2640 	if (!get_cpu_device(0))
2641 		return -EPROBE_DEFER;
2642 
2643 	if (!driver_data || !driver_data->verify || !driver_data->init ||
2644 	    !(driver_data->setpolicy || driver_data->target_index ||
2645 		    driver_data->target) ||
2646 	     (driver_data->setpolicy && (driver_data->target_index ||
2647 		    driver_data->target)) ||
2648 	     (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2649 	     (!driver_data->online != !driver_data->offline))
2650 		return -EINVAL;
2651 
2652 	pr_debug("trying to register driver %s\n", driver_data->name);
2653 
2654 	/* Protect against concurrent CPU online/offline. */
2655 	cpus_read_lock();
2656 
2657 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2658 	if (cpufreq_driver) {
2659 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2660 		ret = -EEXIST;
2661 		goto out;
2662 	}
2663 	cpufreq_driver = driver_data;
2664 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2665 
2666 	if (driver_data->setpolicy)
2667 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2668 
2669 	if (cpufreq_boost_supported()) {
2670 		ret = create_boost_sysfs_file();
2671 		if (ret)
2672 			goto err_null_driver;
2673 	}
2674 
2675 	ret = subsys_interface_register(&cpufreq_interface);
2676 	if (ret)
2677 		goto err_boost_unreg;
2678 
2679 	if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2680 	    list_empty(&cpufreq_policy_list)) {
2681 		/* if all ->init() calls failed, unregister */
2682 		ret = -ENODEV;
2683 		pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2684 			 driver_data->name);
2685 		goto err_if_unreg;
2686 	}
2687 
2688 	ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2689 						   "cpufreq:online",
2690 						   cpuhp_cpufreq_online,
2691 						   cpuhp_cpufreq_offline);
2692 	if (ret < 0)
2693 		goto err_if_unreg;
2694 	hp_online = ret;
2695 	ret = 0;
2696 
2697 	pr_debug("driver %s up and running\n", driver_data->name);
2698 	goto out;
2699 
2700 err_if_unreg:
2701 	subsys_interface_unregister(&cpufreq_interface);
2702 err_boost_unreg:
2703 	remove_boost_sysfs_file();
2704 err_null_driver:
2705 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2706 	cpufreq_driver = NULL;
2707 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2708 out:
2709 	cpus_read_unlock();
2710 	return ret;
2711 }
2712 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2713 
2714 /**
2715  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2716  *
2717  * Unregister the current CPUFreq driver. Only call this if you have
2718  * the right to do so, i.e. if you have succeeded in initialising before!
2719  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2720  * currently not initialised.
2721  */
2722 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2723 {
2724 	unsigned long flags;
2725 
2726 	if (!cpufreq_driver || (driver != cpufreq_driver))
2727 		return -EINVAL;
2728 
2729 	pr_debug("unregistering driver %s\n", driver->name);
2730 
2731 	/* Protect against concurrent cpu hotplug */
2732 	cpus_read_lock();
2733 	subsys_interface_unregister(&cpufreq_interface);
2734 	remove_boost_sysfs_file();
2735 	cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2736 
2737 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2738 
2739 	cpufreq_driver = NULL;
2740 
2741 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2742 	cpus_read_unlock();
2743 
2744 	return 0;
2745 }
2746 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2747 
2748 struct kobject *cpufreq_global_kobject;
2749 EXPORT_SYMBOL(cpufreq_global_kobject);
2750 
2751 static int __init cpufreq_core_init(void)
2752 {
2753 	if (cpufreq_disabled())
2754 		return -ENODEV;
2755 
2756 	cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2757 	BUG_ON(!cpufreq_global_kobject);
2758 
2759 	return 0;
2760 }
2761 module_param(off, int, 0444);
2762 core_initcall(cpufreq_core_init);
2763