xref: /openbmc/linux/drivers/cpufreq/cpufreq.c (revision 275876e2)
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
7  *
8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
9  *	Added handling for CPU hotplug
10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
11  *	Fix handling for CPU hotplug -- affected CPUs
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License version 2 as
15  * published by the Free Software Foundation.
16  */
17 
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 
20 #include <linux/cpu.h>
21 #include <linux/cpufreq.h>
22 #include <linux/delay.h>
23 #include <linux/device.h>
24 #include <linux/init.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/module.h>
27 #include <linux/mutex.h>
28 #include <linux/slab.h>
29 #include <linux/suspend.h>
30 #include <linux/tick.h>
31 #include <trace/events/power.h>
32 
33 /**
34  * The "cpufreq driver" - the arch- or hardware-dependent low
35  * level driver of CPUFreq support, and its spinlock. This lock
36  * also protects the cpufreq_cpu_data array.
37  */
38 static struct cpufreq_driver *cpufreq_driver;
39 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
40 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data_fallback);
41 static DEFINE_RWLOCK(cpufreq_driver_lock);
42 DEFINE_MUTEX(cpufreq_governor_lock);
43 static LIST_HEAD(cpufreq_policy_list);
44 
45 /* This one keeps track of the previously set governor of a removed CPU */
46 static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
47 
48 /* Flag to suspend/resume CPUFreq governors */
49 static bool cpufreq_suspended;
50 
51 static inline bool has_target(void)
52 {
53 	return cpufreq_driver->target_index || cpufreq_driver->target;
54 }
55 
56 /*
57  * rwsem to guarantee that cpufreq driver module doesn't unload during critical
58  * sections
59  */
60 static DECLARE_RWSEM(cpufreq_rwsem);
61 
62 /* internal prototypes */
63 static int __cpufreq_governor(struct cpufreq_policy *policy,
64 		unsigned int event);
65 static unsigned int __cpufreq_get(unsigned int cpu);
66 static void handle_update(struct work_struct *work);
67 
68 /**
69  * Two notifier lists: the "policy" list is involved in the
70  * validation process for a new CPU frequency policy; the
71  * "transition" list for kernel code that needs to handle
72  * changes to devices when the CPU clock speed changes.
73  * The mutex locks both lists.
74  */
75 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
76 static struct srcu_notifier_head cpufreq_transition_notifier_list;
77 
78 static bool init_cpufreq_transition_notifier_list_called;
79 static int __init init_cpufreq_transition_notifier_list(void)
80 {
81 	srcu_init_notifier_head(&cpufreq_transition_notifier_list);
82 	init_cpufreq_transition_notifier_list_called = true;
83 	return 0;
84 }
85 pure_initcall(init_cpufreq_transition_notifier_list);
86 
87 static int off __read_mostly;
88 static int cpufreq_disabled(void)
89 {
90 	return off;
91 }
92 void disable_cpufreq(void)
93 {
94 	off = 1;
95 }
96 static LIST_HEAD(cpufreq_governor_list);
97 static DEFINE_MUTEX(cpufreq_governor_mutex);
98 
99 bool have_governor_per_policy(void)
100 {
101 	return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
102 }
103 EXPORT_SYMBOL_GPL(have_governor_per_policy);
104 
105 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
106 {
107 	if (have_governor_per_policy())
108 		return &policy->kobj;
109 	else
110 		return cpufreq_global_kobject;
111 }
112 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
113 
114 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
115 {
116 	u64 idle_time;
117 	u64 cur_wall_time;
118 	u64 busy_time;
119 
120 	cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
121 
122 	busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
123 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
124 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
125 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
126 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
127 	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
128 
129 	idle_time = cur_wall_time - busy_time;
130 	if (wall)
131 		*wall = cputime_to_usecs(cur_wall_time);
132 
133 	return cputime_to_usecs(idle_time);
134 }
135 
136 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
137 {
138 	u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
139 
140 	if (idle_time == -1ULL)
141 		return get_cpu_idle_time_jiffy(cpu, wall);
142 	else if (!io_busy)
143 		idle_time += get_cpu_iowait_time_us(cpu, wall);
144 
145 	return idle_time;
146 }
147 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
148 
149 /*
150  * This is a generic cpufreq init() routine which can be used by cpufreq
151  * drivers of SMP systems. It will do following:
152  * - validate & show freq table passed
153  * - set policies transition latency
154  * - policy->cpus with all possible CPUs
155  */
156 int cpufreq_generic_init(struct cpufreq_policy *policy,
157 		struct cpufreq_frequency_table *table,
158 		unsigned int transition_latency)
159 {
160 	int ret;
161 
162 	ret = cpufreq_table_validate_and_show(policy, table);
163 	if (ret) {
164 		pr_err("%s: invalid frequency table: %d\n", __func__, ret);
165 		return ret;
166 	}
167 
168 	policy->cpuinfo.transition_latency = transition_latency;
169 
170 	/*
171 	 * The driver only supports the SMP configuartion where all processors
172 	 * share the clock and voltage and clock.
173 	 */
174 	cpumask_setall(policy->cpus);
175 
176 	return 0;
177 }
178 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
179 
180 unsigned int cpufreq_generic_get(unsigned int cpu)
181 {
182 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
183 
184 	if (!policy || IS_ERR(policy->clk)) {
185 		pr_err("%s: No %s associated to cpu: %d\n",
186 		       __func__, policy ? "clk" : "policy", cpu);
187 		return 0;
188 	}
189 
190 	return clk_get_rate(policy->clk) / 1000;
191 }
192 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
193 
194 /* Only for cpufreq core internal use */
195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
196 {
197 	return per_cpu(cpufreq_cpu_data, cpu);
198 }
199 
200 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
201 {
202 	struct cpufreq_policy *policy = NULL;
203 	unsigned long flags;
204 
205 	if (cpufreq_disabled() || (cpu >= nr_cpu_ids))
206 		return NULL;
207 
208 	if (!down_read_trylock(&cpufreq_rwsem))
209 		return NULL;
210 
211 	/* get the cpufreq driver */
212 	read_lock_irqsave(&cpufreq_driver_lock, flags);
213 
214 	if (cpufreq_driver) {
215 		/* get the CPU */
216 		policy = per_cpu(cpufreq_cpu_data, cpu);
217 		if (policy)
218 			kobject_get(&policy->kobj);
219 	}
220 
221 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
222 
223 	if (!policy)
224 		up_read(&cpufreq_rwsem);
225 
226 	return policy;
227 }
228 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
229 
230 void cpufreq_cpu_put(struct cpufreq_policy *policy)
231 {
232 	if (cpufreq_disabled())
233 		return;
234 
235 	kobject_put(&policy->kobj);
236 	up_read(&cpufreq_rwsem);
237 }
238 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
239 
240 /*********************************************************************
241  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
242  *********************************************************************/
243 
244 /**
245  * adjust_jiffies - adjust the system "loops_per_jiffy"
246  *
247  * This function alters the system "loops_per_jiffy" for the clock
248  * speed change. Note that loops_per_jiffy cannot be updated on SMP
249  * systems as each CPU might be scaled differently. So, use the arch
250  * per-CPU loops_per_jiffy value wherever possible.
251  */
252 #ifndef CONFIG_SMP
253 static unsigned long l_p_j_ref;
254 static unsigned int l_p_j_ref_freq;
255 
256 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
257 {
258 	if (ci->flags & CPUFREQ_CONST_LOOPS)
259 		return;
260 
261 	if (!l_p_j_ref_freq) {
262 		l_p_j_ref = loops_per_jiffy;
263 		l_p_j_ref_freq = ci->old;
264 		pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
265 			 l_p_j_ref, l_p_j_ref_freq);
266 	}
267 	if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
268 		loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
269 								ci->new);
270 		pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
271 			 loops_per_jiffy, ci->new);
272 	}
273 }
274 #else
275 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
276 {
277 	return;
278 }
279 #endif
280 
281 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
282 		struct cpufreq_freqs *freqs, unsigned int state)
283 {
284 	BUG_ON(irqs_disabled());
285 
286 	if (cpufreq_disabled())
287 		return;
288 
289 	freqs->flags = cpufreq_driver->flags;
290 	pr_debug("notification %u of frequency transition to %u kHz\n",
291 		 state, freqs->new);
292 
293 	switch (state) {
294 
295 	case CPUFREQ_PRECHANGE:
296 		/* detect if the driver reported a value as "old frequency"
297 		 * which is not equal to what the cpufreq core thinks is
298 		 * "old frequency".
299 		 */
300 		if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
301 			if ((policy) && (policy->cpu == freqs->cpu) &&
302 			    (policy->cur) && (policy->cur != freqs->old)) {
303 				pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
304 					 freqs->old, policy->cur);
305 				freqs->old = policy->cur;
306 			}
307 		}
308 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
309 				CPUFREQ_PRECHANGE, freqs);
310 		adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
311 		break;
312 
313 	case CPUFREQ_POSTCHANGE:
314 		adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
315 		pr_debug("FREQ: %lu - CPU: %lu\n",
316 			 (unsigned long)freqs->new, (unsigned long)freqs->cpu);
317 		trace_cpu_frequency(freqs->new, freqs->cpu);
318 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
319 				CPUFREQ_POSTCHANGE, freqs);
320 		if (likely(policy) && likely(policy->cpu == freqs->cpu))
321 			policy->cur = freqs->new;
322 		break;
323 	}
324 }
325 
326 /**
327  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
328  * on frequency transition.
329  *
330  * This function calls the transition notifiers and the "adjust_jiffies"
331  * function. It is called twice on all CPU frequency changes that have
332  * external effects.
333  */
334 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
335 		struct cpufreq_freqs *freqs, unsigned int state)
336 {
337 	for_each_cpu(freqs->cpu, policy->cpus)
338 		__cpufreq_notify_transition(policy, freqs, state);
339 }
340 
341 /* Do post notifications when there are chances that transition has failed */
342 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
343 		struct cpufreq_freqs *freqs, int transition_failed)
344 {
345 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
346 	if (!transition_failed)
347 		return;
348 
349 	swap(freqs->old, freqs->new);
350 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
351 	cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
352 }
353 
354 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
355 		struct cpufreq_freqs *freqs)
356 {
357 
358 	/*
359 	 * Catch double invocations of _begin() which lead to self-deadlock.
360 	 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
361 	 * doesn't invoke _begin() on their behalf, and hence the chances of
362 	 * double invocations are very low. Moreover, there are scenarios
363 	 * where these checks can emit false-positive warnings in these
364 	 * drivers; so we avoid that by skipping them altogether.
365 	 */
366 	WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
367 				&& current == policy->transition_task);
368 
369 wait:
370 	wait_event(policy->transition_wait, !policy->transition_ongoing);
371 
372 	spin_lock(&policy->transition_lock);
373 
374 	if (unlikely(policy->transition_ongoing)) {
375 		spin_unlock(&policy->transition_lock);
376 		goto wait;
377 	}
378 
379 	policy->transition_ongoing = true;
380 	policy->transition_task = current;
381 
382 	spin_unlock(&policy->transition_lock);
383 
384 	cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
385 }
386 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
387 
388 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
389 		struct cpufreq_freqs *freqs, int transition_failed)
390 {
391 	if (unlikely(WARN_ON(!policy->transition_ongoing)))
392 		return;
393 
394 	cpufreq_notify_post_transition(policy, freqs, transition_failed);
395 
396 	policy->transition_ongoing = false;
397 	policy->transition_task = NULL;
398 
399 	wake_up(&policy->transition_wait);
400 }
401 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
402 
403 
404 /*********************************************************************
405  *                          SYSFS INTERFACE                          *
406  *********************************************************************/
407 static ssize_t show_boost(struct kobject *kobj,
408 				 struct attribute *attr, char *buf)
409 {
410 	return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
411 }
412 
413 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
414 				  const char *buf, size_t count)
415 {
416 	int ret, enable;
417 
418 	ret = sscanf(buf, "%d", &enable);
419 	if (ret != 1 || enable < 0 || enable > 1)
420 		return -EINVAL;
421 
422 	if (cpufreq_boost_trigger_state(enable)) {
423 		pr_err("%s: Cannot %s BOOST!\n",
424 		       __func__, enable ? "enable" : "disable");
425 		return -EINVAL;
426 	}
427 
428 	pr_debug("%s: cpufreq BOOST %s\n",
429 		 __func__, enable ? "enabled" : "disabled");
430 
431 	return count;
432 }
433 define_one_global_rw(boost);
434 
435 static struct cpufreq_governor *__find_governor(const char *str_governor)
436 {
437 	struct cpufreq_governor *t;
438 
439 	list_for_each_entry(t, &cpufreq_governor_list, governor_list)
440 		if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
441 			return t;
442 
443 	return NULL;
444 }
445 
446 /**
447  * cpufreq_parse_governor - parse a governor string
448  */
449 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
450 				struct cpufreq_governor **governor)
451 {
452 	int err = -EINVAL;
453 
454 	if (!cpufreq_driver)
455 		goto out;
456 
457 	if (cpufreq_driver->setpolicy) {
458 		if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
459 			*policy = CPUFREQ_POLICY_PERFORMANCE;
460 			err = 0;
461 		} else if (!strnicmp(str_governor, "powersave",
462 						CPUFREQ_NAME_LEN)) {
463 			*policy = CPUFREQ_POLICY_POWERSAVE;
464 			err = 0;
465 		}
466 	} else if (has_target()) {
467 		struct cpufreq_governor *t;
468 
469 		mutex_lock(&cpufreq_governor_mutex);
470 
471 		t = __find_governor(str_governor);
472 
473 		if (t == NULL) {
474 			int ret;
475 
476 			mutex_unlock(&cpufreq_governor_mutex);
477 			ret = request_module("cpufreq_%s", str_governor);
478 			mutex_lock(&cpufreq_governor_mutex);
479 
480 			if (ret == 0)
481 				t = __find_governor(str_governor);
482 		}
483 
484 		if (t != NULL) {
485 			*governor = t;
486 			err = 0;
487 		}
488 
489 		mutex_unlock(&cpufreq_governor_mutex);
490 	}
491 out:
492 	return err;
493 }
494 
495 /**
496  * cpufreq_per_cpu_attr_read() / show_##file_name() -
497  * print out cpufreq information
498  *
499  * Write out information from cpufreq_driver->policy[cpu]; object must be
500  * "unsigned int".
501  */
502 
503 #define show_one(file_name, object)			\
504 static ssize_t show_##file_name				\
505 (struct cpufreq_policy *policy, char *buf)		\
506 {							\
507 	return sprintf(buf, "%u\n", policy->object);	\
508 }
509 
510 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
511 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
512 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
513 show_one(scaling_min_freq, min);
514 show_one(scaling_max_freq, max);
515 show_one(scaling_cur_freq, cur);
516 
517 static int cpufreq_set_policy(struct cpufreq_policy *policy,
518 				struct cpufreq_policy *new_policy);
519 
520 /**
521  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
522  */
523 #define store_one(file_name, object)			\
524 static ssize_t store_##file_name					\
525 (struct cpufreq_policy *policy, const char *buf, size_t count)		\
526 {									\
527 	int ret;							\
528 	struct cpufreq_policy new_policy;				\
529 									\
530 	ret = cpufreq_get_policy(&new_policy, policy->cpu);		\
531 	if (ret)							\
532 		return -EINVAL;						\
533 									\
534 	ret = sscanf(buf, "%u", &new_policy.object);			\
535 	if (ret != 1)							\
536 		return -EINVAL;						\
537 									\
538 	ret = cpufreq_set_policy(policy, &new_policy);		\
539 	policy->user_policy.object = policy->object;			\
540 									\
541 	return ret ? ret : count;					\
542 }
543 
544 store_one(scaling_min_freq, min);
545 store_one(scaling_max_freq, max);
546 
547 /**
548  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
549  */
550 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
551 					char *buf)
552 {
553 	unsigned int cur_freq = __cpufreq_get(policy->cpu);
554 	if (!cur_freq)
555 		return sprintf(buf, "<unknown>");
556 	return sprintf(buf, "%u\n", cur_freq);
557 }
558 
559 /**
560  * show_scaling_governor - show the current policy for the specified CPU
561  */
562 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
563 {
564 	if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
565 		return sprintf(buf, "powersave\n");
566 	else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
567 		return sprintf(buf, "performance\n");
568 	else if (policy->governor)
569 		return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
570 				policy->governor->name);
571 	return -EINVAL;
572 }
573 
574 /**
575  * store_scaling_governor - store policy for the specified CPU
576  */
577 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
578 					const char *buf, size_t count)
579 {
580 	int ret;
581 	char	str_governor[16];
582 	struct cpufreq_policy new_policy;
583 
584 	ret = cpufreq_get_policy(&new_policy, policy->cpu);
585 	if (ret)
586 		return ret;
587 
588 	ret = sscanf(buf, "%15s", str_governor);
589 	if (ret != 1)
590 		return -EINVAL;
591 
592 	if (cpufreq_parse_governor(str_governor, &new_policy.policy,
593 						&new_policy.governor))
594 		return -EINVAL;
595 
596 	ret = cpufreq_set_policy(policy, &new_policy);
597 
598 	policy->user_policy.policy = policy->policy;
599 	policy->user_policy.governor = policy->governor;
600 
601 	if (ret)
602 		return ret;
603 	else
604 		return count;
605 }
606 
607 /**
608  * show_scaling_driver - show the cpufreq driver currently loaded
609  */
610 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
611 {
612 	return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
613 }
614 
615 /**
616  * show_scaling_available_governors - show the available CPUfreq governors
617  */
618 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
619 						char *buf)
620 {
621 	ssize_t i = 0;
622 	struct cpufreq_governor *t;
623 
624 	if (!has_target()) {
625 		i += sprintf(buf, "performance powersave");
626 		goto out;
627 	}
628 
629 	list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
630 		if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
631 		    - (CPUFREQ_NAME_LEN + 2)))
632 			goto out;
633 		i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
634 	}
635 out:
636 	i += sprintf(&buf[i], "\n");
637 	return i;
638 }
639 
640 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
641 {
642 	ssize_t i = 0;
643 	unsigned int cpu;
644 
645 	for_each_cpu(cpu, mask) {
646 		if (i)
647 			i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
648 		i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
649 		if (i >= (PAGE_SIZE - 5))
650 			break;
651 	}
652 	i += sprintf(&buf[i], "\n");
653 	return i;
654 }
655 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
656 
657 /**
658  * show_related_cpus - show the CPUs affected by each transition even if
659  * hw coordination is in use
660  */
661 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
662 {
663 	return cpufreq_show_cpus(policy->related_cpus, buf);
664 }
665 
666 /**
667  * show_affected_cpus - show the CPUs affected by each transition
668  */
669 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
670 {
671 	return cpufreq_show_cpus(policy->cpus, buf);
672 }
673 
674 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
675 					const char *buf, size_t count)
676 {
677 	unsigned int freq = 0;
678 	unsigned int ret;
679 
680 	if (!policy->governor || !policy->governor->store_setspeed)
681 		return -EINVAL;
682 
683 	ret = sscanf(buf, "%u", &freq);
684 	if (ret != 1)
685 		return -EINVAL;
686 
687 	policy->governor->store_setspeed(policy, freq);
688 
689 	return count;
690 }
691 
692 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
693 {
694 	if (!policy->governor || !policy->governor->show_setspeed)
695 		return sprintf(buf, "<unsupported>\n");
696 
697 	return policy->governor->show_setspeed(policy, buf);
698 }
699 
700 /**
701  * show_bios_limit - show the current cpufreq HW/BIOS limitation
702  */
703 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
704 {
705 	unsigned int limit;
706 	int ret;
707 	if (cpufreq_driver->bios_limit) {
708 		ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
709 		if (!ret)
710 			return sprintf(buf, "%u\n", limit);
711 	}
712 	return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
713 }
714 
715 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
716 cpufreq_freq_attr_ro(cpuinfo_min_freq);
717 cpufreq_freq_attr_ro(cpuinfo_max_freq);
718 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
719 cpufreq_freq_attr_ro(scaling_available_governors);
720 cpufreq_freq_attr_ro(scaling_driver);
721 cpufreq_freq_attr_ro(scaling_cur_freq);
722 cpufreq_freq_attr_ro(bios_limit);
723 cpufreq_freq_attr_ro(related_cpus);
724 cpufreq_freq_attr_ro(affected_cpus);
725 cpufreq_freq_attr_rw(scaling_min_freq);
726 cpufreq_freq_attr_rw(scaling_max_freq);
727 cpufreq_freq_attr_rw(scaling_governor);
728 cpufreq_freq_attr_rw(scaling_setspeed);
729 
730 static struct attribute *default_attrs[] = {
731 	&cpuinfo_min_freq.attr,
732 	&cpuinfo_max_freq.attr,
733 	&cpuinfo_transition_latency.attr,
734 	&scaling_min_freq.attr,
735 	&scaling_max_freq.attr,
736 	&affected_cpus.attr,
737 	&related_cpus.attr,
738 	&scaling_governor.attr,
739 	&scaling_driver.attr,
740 	&scaling_available_governors.attr,
741 	&scaling_setspeed.attr,
742 	NULL
743 };
744 
745 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
746 #define to_attr(a) container_of(a, struct freq_attr, attr)
747 
748 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
749 {
750 	struct cpufreq_policy *policy = to_policy(kobj);
751 	struct freq_attr *fattr = to_attr(attr);
752 	ssize_t ret;
753 
754 	if (!down_read_trylock(&cpufreq_rwsem))
755 		return -EINVAL;
756 
757 	down_read(&policy->rwsem);
758 
759 	if (fattr->show)
760 		ret = fattr->show(policy, buf);
761 	else
762 		ret = -EIO;
763 
764 	up_read(&policy->rwsem);
765 	up_read(&cpufreq_rwsem);
766 
767 	return ret;
768 }
769 
770 static ssize_t store(struct kobject *kobj, struct attribute *attr,
771 		     const char *buf, size_t count)
772 {
773 	struct cpufreq_policy *policy = to_policy(kobj);
774 	struct freq_attr *fattr = to_attr(attr);
775 	ssize_t ret = -EINVAL;
776 
777 	get_online_cpus();
778 
779 	if (!cpu_online(policy->cpu))
780 		goto unlock;
781 
782 	if (!down_read_trylock(&cpufreq_rwsem))
783 		goto unlock;
784 
785 	down_write(&policy->rwsem);
786 
787 	if (fattr->store)
788 		ret = fattr->store(policy, buf, count);
789 	else
790 		ret = -EIO;
791 
792 	up_write(&policy->rwsem);
793 
794 	up_read(&cpufreq_rwsem);
795 unlock:
796 	put_online_cpus();
797 
798 	return ret;
799 }
800 
801 static void cpufreq_sysfs_release(struct kobject *kobj)
802 {
803 	struct cpufreq_policy *policy = to_policy(kobj);
804 	pr_debug("last reference is dropped\n");
805 	complete(&policy->kobj_unregister);
806 }
807 
808 static const struct sysfs_ops sysfs_ops = {
809 	.show	= show,
810 	.store	= store,
811 };
812 
813 static struct kobj_type ktype_cpufreq = {
814 	.sysfs_ops	= &sysfs_ops,
815 	.default_attrs	= default_attrs,
816 	.release	= cpufreq_sysfs_release,
817 };
818 
819 struct kobject *cpufreq_global_kobject;
820 EXPORT_SYMBOL(cpufreq_global_kobject);
821 
822 static int cpufreq_global_kobject_usage;
823 
824 int cpufreq_get_global_kobject(void)
825 {
826 	if (!cpufreq_global_kobject_usage++)
827 		return kobject_add(cpufreq_global_kobject,
828 				&cpu_subsys.dev_root->kobj, "%s", "cpufreq");
829 
830 	return 0;
831 }
832 EXPORT_SYMBOL(cpufreq_get_global_kobject);
833 
834 void cpufreq_put_global_kobject(void)
835 {
836 	if (!--cpufreq_global_kobject_usage)
837 		kobject_del(cpufreq_global_kobject);
838 }
839 EXPORT_SYMBOL(cpufreq_put_global_kobject);
840 
841 int cpufreq_sysfs_create_file(const struct attribute *attr)
842 {
843 	int ret = cpufreq_get_global_kobject();
844 
845 	if (!ret) {
846 		ret = sysfs_create_file(cpufreq_global_kobject, attr);
847 		if (ret)
848 			cpufreq_put_global_kobject();
849 	}
850 
851 	return ret;
852 }
853 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
854 
855 void cpufreq_sysfs_remove_file(const struct attribute *attr)
856 {
857 	sysfs_remove_file(cpufreq_global_kobject, attr);
858 	cpufreq_put_global_kobject();
859 }
860 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
861 
862 /* symlink affected CPUs */
863 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
864 {
865 	unsigned int j;
866 	int ret = 0;
867 
868 	for_each_cpu(j, policy->cpus) {
869 		struct device *cpu_dev;
870 
871 		if (j == policy->cpu)
872 			continue;
873 
874 		pr_debug("Adding link for CPU: %u\n", j);
875 		cpu_dev = get_cpu_device(j);
876 		ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
877 					"cpufreq");
878 		if (ret)
879 			break;
880 	}
881 	return ret;
882 }
883 
884 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy,
885 				     struct device *dev)
886 {
887 	struct freq_attr **drv_attr;
888 	int ret = 0;
889 
890 	/* prepare interface data */
891 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
892 				   &dev->kobj, "cpufreq");
893 	if (ret)
894 		return ret;
895 
896 	/* set up files for this cpu device */
897 	drv_attr = cpufreq_driver->attr;
898 	while ((drv_attr) && (*drv_attr)) {
899 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
900 		if (ret)
901 			goto err_out_kobj_put;
902 		drv_attr++;
903 	}
904 	if (cpufreq_driver->get) {
905 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
906 		if (ret)
907 			goto err_out_kobj_put;
908 	}
909 	if (has_target()) {
910 		ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
911 		if (ret)
912 			goto err_out_kobj_put;
913 	}
914 	if (cpufreq_driver->bios_limit) {
915 		ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
916 		if (ret)
917 			goto err_out_kobj_put;
918 	}
919 
920 	ret = cpufreq_add_dev_symlink(policy);
921 	if (ret)
922 		goto err_out_kobj_put;
923 
924 	return ret;
925 
926 err_out_kobj_put:
927 	kobject_put(&policy->kobj);
928 	wait_for_completion(&policy->kobj_unregister);
929 	return ret;
930 }
931 
932 static void cpufreq_init_policy(struct cpufreq_policy *policy)
933 {
934 	struct cpufreq_governor *gov = NULL;
935 	struct cpufreq_policy new_policy;
936 	int ret = 0;
937 
938 	memcpy(&new_policy, policy, sizeof(*policy));
939 
940 	/* Update governor of new_policy to the governor used before hotplug */
941 	gov = __find_governor(per_cpu(cpufreq_cpu_governor, policy->cpu));
942 	if (gov)
943 		pr_debug("Restoring governor %s for cpu %d\n",
944 				policy->governor->name, policy->cpu);
945 	else
946 		gov = CPUFREQ_DEFAULT_GOVERNOR;
947 
948 	new_policy.governor = gov;
949 
950 	/* Use the default policy if its valid. */
951 	if (cpufreq_driver->setpolicy)
952 		cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
953 
954 	/* set default policy */
955 	ret = cpufreq_set_policy(policy, &new_policy);
956 	if (ret) {
957 		pr_debug("setting policy failed\n");
958 		if (cpufreq_driver->exit)
959 			cpufreq_driver->exit(policy);
960 	}
961 }
962 
963 #ifdef CONFIG_HOTPLUG_CPU
964 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy,
965 				  unsigned int cpu, struct device *dev)
966 {
967 	int ret = 0;
968 	unsigned long flags;
969 
970 	if (has_target()) {
971 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
972 		if (ret) {
973 			pr_err("%s: Failed to stop governor\n", __func__);
974 			return ret;
975 		}
976 	}
977 
978 	down_write(&policy->rwsem);
979 
980 	write_lock_irqsave(&cpufreq_driver_lock, flags);
981 
982 	cpumask_set_cpu(cpu, policy->cpus);
983 	per_cpu(cpufreq_cpu_data, cpu) = policy;
984 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
985 
986 	up_write(&policy->rwsem);
987 
988 	if (has_target()) {
989 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
990 		if (!ret)
991 			ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
992 
993 		if (ret) {
994 			pr_err("%s: Failed to start governor\n", __func__);
995 			return ret;
996 		}
997 	}
998 
999 	return sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
1000 }
1001 #endif
1002 
1003 static struct cpufreq_policy *cpufreq_policy_restore(unsigned int cpu)
1004 {
1005 	struct cpufreq_policy *policy;
1006 	unsigned long flags;
1007 
1008 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1009 
1010 	policy = per_cpu(cpufreq_cpu_data_fallback, cpu);
1011 
1012 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1013 
1014 	policy->governor = NULL;
1015 
1016 	return policy;
1017 }
1018 
1019 static struct cpufreq_policy *cpufreq_policy_alloc(void)
1020 {
1021 	struct cpufreq_policy *policy;
1022 
1023 	policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1024 	if (!policy)
1025 		return NULL;
1026 
1027 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1028 		goto err_free_policy;
1029 
1030 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1031 		goto err_free_cpumask;
1032 
1033 	INIT_LIST_HEAD(&policy->policy_list);
1034 	init_rwsem(&policy->rwsem);
1035 	spin_lock_init(&policy->transition_lock);
1036 	init_waitqueue_head(&policy->transition_wait);
1037 
1038 	return policy;
1039 
1040 err_free_cpumask:
1041 	free_cpumask_var(policy->cpus);
1042 err_free_policy:
1043 	kfree(policy);
1044 
1045 	return NULL;
1046 }
1047 
1048 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1049 {
1050 	struct kobject *kobj;
1051 	struct completion *cmp;
1052 
1053 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1054 			CPUFREQ_REMOVE_POLICY, policy);
1055 
1056 	down_read(&policy->rwsem);
1057 	kobj = &policy->kobj;
1058 	cmp = &policy->kobj_unregister;
1059 	up_read(&policy->rwsem);
1060 	kobject_put(kobj);
1061 
1062 	/*
1063 	 * We need to make sure that the underlying kobj is
1064 	 * actually not referenced anymore by anybody before we
1065 	 * proceed with unloading.
1066 	 */
1067 	pr_debug("waiting for dropping of refcount\n");
1068 	wait_for_completion(cmp);
1069 	pr_debug("wait complete\n");
1070 }
1071 
1072 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1073 {
1074 	free_cpumask_var(policy->related_cpus);
1075 	free_cpumask_var(policy->cpus);
1076 	kfree(policy);
1077 }
1078 
1079 static int update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu,
1080 			     struct device *cpu_dev)
1081 {
1082 	int ret;
1083 
1084 	if (WARN_ON(cpu == policy->cpu))
1085 		return 0;
1086 
1087 	/* Move kobject to the new policy->cpu */
1088 	ret = kobject_move(&policy->kobj, &cpu_dev->kobj);
1089 	if (ret) {
1090 		pr_err("%s: Failed to move kobj: %d\n", __func__, ret);
1091 		return ret;
1092 	}
1093 
1094 	down_write(&policy->rwsem);
1095 
1096 	policy->last_cpu = policy->cpu;
1097 	policy->cpu = cpu;
1098 
1099 	up_write(&policy->rwsem);
1100 
1101 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1102 			CPUFREQ_UPDATE_POLICY_CPU, policy);
1103 
1104 	return 0;
1105 }
1106 
1107 static int __cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1108 {
1109 	unsigned int j, cpu = dev->id;
1110 	int ret = -ENOMEM;
1111 	struct cpufreq_policy *policy;
1112 	unsigned long flags;
1113 	bool recover_policy = cpufreq_suspended;
1114 #ifdef CONFIG_HOTPLUG_CPU
1115 	struct cpufreq_policy *tpolicy;
1116 #endif
1117 
1118 	if (cpu_is_offline(cpu))
1119 		return 0;
1120 
1121 	pr_debug("adding CPU %u\n", cpu);
1122 
1123 #ifdef CONFIG_SMP
1124 	/* check whether a different CPU already registered this
1125 	 * CPU because it is in the same boat. */
1126 	policy = cpufreq_cpu_get(cpu);
1127 	if (unlikely(policy)) {
1128 		cpufreq_cpu_put(policy);
1129 		return 0;
1130 	}
1131 #endif
1132 
1133 	if (!down_read_trylock(&cpufreq_rwsem))
1134 		return 0;
1135 
1136 #ifdef CONFIG_HOTPLUG_CPU
1137 	/* Check if this cpu was hot-unplugged earlier and has siblings */
1138 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1139 	list_for_each_entry(tpolicy, &cpufreq_policy_list, policy_list) {
1140 		if (cpumask_test_cpu(cpu, tpolicy->related_cpus)) {
1141 			read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1142 			ret = cpufreq_add_policy_cpu(tpolicy, cpu, dev);
1143 			up_read(&cpufreq_rwsem);
1144 			return ret;
1145 		}
1146 	}
1147 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1148 #endif
1149 
1150 	/*
1151 	 * Restore the saved policy when doing light-weight init and fall back
1152 	 * to the full init if that fails.
1153 	 */
1154 	policy = recover_policy ? cpufreq_policy_restore(cpu) : NULL;
1155 	if (!policy) {
1156 		recover_policy = false;
1157 		policy = cpufreq_policy_alloc();
1158 		if (!policy)
1159 			goto nomem_out;
1160 	}
1161 
1162 	/*
1163 	 * In the resume path, since we restore a saved policy, the assignment
1164 	 * to policy->cpu is like an update of the existing policy, rather than
1165 	 * the creation of a brand new one. So we need to perform this update
1166 	 * by invoking update_policy_cpu().
1167 	 */
1168 	if (recover_policy && cpu != policy->cpu)
1169 		WARN_ON(update_policy_cpu(policy, cpu, dev));
1170 	else
1171 		policy->cpu = cpu;
1172 
1173 	cpumask_copy(policy->cpus, cpumask_of(cpu));
1174 
1175 	init_completion(&policy->kobj_unregister);
1176 	INIT_WORK(&policy->update, handle_update);
1177 
1178 	/* call driver. From then on the cpufreq must be able
1179 	 * to accept all calls to ->verify and ->setpolicy for this CPU
1180 	 */
1181 	ret = cpufreq_driver->init(policy);
1182 	if (ret) {
1183 		pr_debug("initialization failed\n");
1184 		goto err_set_policy_cpu;
1185 	}
1186 
1187 	/* related cpus should atleast have policy->cpus */
1188 	cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1189 
1190 	/*
1191 	 * affected cpus must always be the one, which are online. We aren't
1192 	 * managing offline cpus here.
1193 	 */
1194 	cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1195 
1196 	if (!recover_policy) {
1197 		policy->user_policy.min = policy->min;
1198 		policy->user_policy.max = policy->max;
1199 	}
1200 
1201 	down_write(&policy->rwsem);
1202 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1203 	for_each_cpu(j, policy->cpus)
1204 		per_cpu(cpufreq_cpu_data, j) = policy;
1205 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1206 
1207 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1208 		policy->cur = cpufreq_driver->get(policy->cpu);
1209 		if (!policy->cur) {
1210 			pr_err("%s: ->get() failed\n", __func__);
1211 			goto err_get_freq;
1212 		}
1213 	}
1214 
1215 	/*
1216 	 * Sometimes boot loaders set CPU frequency to a value outside of
1217 	 * frequency table present with cpufreq core. In such cases CPU might be
1218 	 * unstable if it has to run on that frequency for long duration of time
1219 	 * and so its better to set it to a frequency which is specified in
1220 	 * freq-table. This also makes cpufreq stats inconsistent as
1221 	 * cpufreq-stats would fail to register because current frequency of CPU
1222 	 * isn't found in freq-table.
1223 	 *
1224 	 * Because we don't want this change to effect boot process badly, we go
1225 	 * for the next freq which is >= policy->cur ('cur' must be set by now,
1226 	 * otherwise we will end up setting freq to lowest of the table as 'cur'
1227 	 * is initialized to zero).
1228 	 *
1229 	 * We are passing target-freq as "policy->cur - 1" otherwise
1230 	 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1231 	 * equal to target-freq.
1232 	 */
1233 	if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1234 	    && has_target()) {
1235 		/* Are we running at unknown frequency ? */
1236 		ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1237 		if (ret == -EINVAL) {
1238 			/* Warn user and fix it */
1239 			pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1240 				__func__, policy->cpu, policy->cur);
1241 			ret = __cpufreq_driver_target(policy, policy->cur - 1,
1242 				CPUFREQ_RELATION_L);
1243 
1244 			/*
1245 			 * Reaching here after boot in a few seconds may not
1246 			 * mean that system will remain stable at "unknown"
1247 			 * frequency for longer duration. Hence, a BUG_ON().
1248 			 */
1249 			BUG_ON(ret);
1250 			pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1251 				__func__, policy->cpu, policy->cur);
1252 		}
1253 	}
1254 
1255 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1256 				     CPUFREQ_START, policy);
1257 
1258 	if (!recover_policy) {
1259 		ret = cpufreq_add_dev_interface(policy, dev);
1260 		if (ret)
1261 			goto err_out_unregister;
1262 		blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1263 				CPUFREQ_CREATE_POLICY, policy);
1264 	}
1265 
1266 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1267 	list_add(&policy->policy_list, &cpufreq_policy_list);
1268 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1269 
1270 	cpufreq_init_policy(policy);
1271 
1272 	if (!recover_policy) {
1273 		policy->user_policy.policy = policy->policy;
1274 		policy->user_policy.governor = policy->governor;
1275 	}
1276 	up_write(&policy->rwsem);
1277 
1278 	kobject_uevent(&policy->kobj, KOBJ_ADD);
1279 	up_read(&cpufreq_rwsem);
1280 
1281 	pr_debug("initialization complete\n");
1282 
1283 	return 0;
1284 
1285 err_out_unregister:
1286 err_get_freq:
1287 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1288 	for_each_cpu(j, policy->cpus)
1289 		per_cpu(cpufreq_cpu_data, j) = NULL;
1290 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1291 
1292 	if (cpufreq_driver->exit)
1293 		cpufreq_driver->exit(policy);
1294 err_set_policy_cpu:
1295 	if (recover_policy) {
1296 		/* Do not leave stale fallback data behind. */
1297 		per_cpu(cpufreq_cpu_data_fallback, cpu) = NULL;
1298 		cpufreq_policy_put_kobj(policy);
1299 	}
1300 	cpufreq_policy_free(policy);
1301 
1302 nomem_out:
1303 	up_read(&cpufreq_rwsem);
1304 
1305 	return ret;
1306 }
1307 
1308 /**
1309  * cpufreq_add_dev - add a CPU device
1310  *
1311  * Adds the cpufreq interface for a CPU device.
1312  *
1313  * The Oracle says: try running cpufreq registration/unregistration concurrently
1314  * with with cpu hotplugging and all hell will break loose. Tried to clean this
1315  * mess up, but more thorough testing is needed. - Mathieu
1316  */
1317 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1318 {
1319 	return __cpufreq_add_dev(dev, sif);
1320 }
1321 
1322 static int __cpufreq_remove_dev_prepare(struct device *dev,
1323 					struct subsys_interface *sif)
1324 {
1325 	unsigned int cpu = dev->id, cpus;
1326 	int ret;
1327 	unsigned long flags;
1328 	struct cpufreq_policy *policy;
1329 
1330 	pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1331 
1332 	write_lock_irqsave(&cpufreq_driver_lock, flags);
1333 
1334 	policy = per_cpu(cpufreq_cpu_data, cpu);
1335 
1336 	/* Save the policy somewhere when doing a light-weight tear-down */
1337 	if (cpufreq_suspended)
1338 		per_cpu(cpufreq_cpu_data_fallback, cpu) = policy;
1339 
1340 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1341 
1342 	if (!policy) {
1343 		pr_debug("%s: No cpu_data found\n", __func__);
1344 		return -EINVAL;
1345 	}
1346 
1347 	if (has_target()) {
1348 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1349 		if (ret) {
1350 			pr_err("%s: Failed to stop governor\n", __func__);
1351 			return ret;
1352 		}
1353 	}
1354 
1355 	if (!cpufreq_driver->setpolicy)
1356 		strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1357 			policy->governor->name, CPUFREQ_NAME_LEN);
1358 
1359 	down_read(&policy->rwsem);
1360 	cpus = cpumask_weight(policy->cpus);
1361 	up_read(&policy->rwsem);
1362 
1363 	if (cpu != policy->cpu) {
1364 		sysfs_remove_link(&dev->kobj, "cpufreq");
1365 	} else if (cpus > 1) {
1366 		/* Nominate new CPU */
1367 		int new_cpu = cpumask_any_but(policy->cpus, cpu);
1368 		struct device *cpu_dev = get_cpu_device(new_cpu);
1369 
1370 		sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1371 		ret = update_policy_cpu(policy, new_cpu, cpu_dev);
1372 		if (ret) {
1373 			if (sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
1374 					      "cpufreq"))
1375 				pr_err("%s: Failed to restore kobj link to cpu:%d\n",
1376 				       __func__, cpu_dev->id);
1377 			return ret;
1378 		}
1379 
1380 		if (!cpufreq_suspended)
1381 			pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1382 				 __func__, new_cpu, cpu);
1383 	} else if (cpufreq_driver->stop_cpu && cpufreq_driver->setpolicy) {
1384 		cpufreq_driver->stop_cpu(policy);
1385 	}
1386 
1387 	return 0;
1388 }
1389 
1390 static int __cpufreq_remove_dev_finish(struct device *dev,
1391 				       struct subsys_interface *sif)
1392 {
1393 	unsigned int cpu = dev->id, cpus;
1394 	int ret;
1395 	unsigned long flags;
1396 	struct cpufreq_policy *policy;
1397 
1398 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1399 	policy = per_cpu(cpufreq_cpu_data, cpu);
1400 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1401 
1402 	if (!policy) {
1403 		pr_debug("%s: No cpu_data found\n", __func__);
1404 		return -EINVAL;
1405 	}
1406 
1407 	down_write(&policy->rwsem);
1408 	cpus = cpumask_weight(policy->cpus);
1409 
1410 	if (cpus > 1)
1411 		cpumask_clear_cpu(cpu, policy->cpus);
1412 	up_write(&policy->rwsem);
1413 
1414 	/* If cpu is last user of policy, free policy */
1415 	if (cpus == 1) {
1416 		if (has_target()) {
1417 			ret = __cpufreq_governor(policy,
1418 					CPUFREQ_GOV_POLICY_EXIT);
1419 			if (ret) {
1420 				pr_err("%s: Failed to exit governor\n",
1421 				       __func__);
1422 				return ret;
1423 			}
1424 		}
1425 
1426 		if (!cpufreq_suspended)
1427 			cpufreq_policy_put_kobj(policy);
1428 
1429 		/*
1430 		 * Perform the ->exit() even during light-weight tear-down,
1431 		 * since this is a core component, and is essential for the
1432 		 * subsequent light-weight ->init() to succeed.
1433 		 */
1434 		if (cpufreq_driver->exit)
1435 			cpufreq_driver->exit(policy);
1436 
1437 		/* Remove policy from list of active policies */
1438 		write_lock_irqsave(&cpufreq_driver_lock, flags);
1439 		list_del(&policy->policy_list);
1440 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1441 
1442 		if (!cpufreq_suspended)
1443 			cpufreq_policy_free(policy);
1444 	} else if (has_target()) {
1445 		ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1446 		if (!ret)
1447 			ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1448 
1449 		if (ret) {
1450 			pr_err("%s: Failed to start governor\n", __func__);
1451 			return ret;
1452 		}
1453 	}
1454 
1455 	per_cpu(cpufreq_cpu_data, cpu) = NULL;
1456 	return 0;
1457 }
1458 
1459 /**
1460  * cpufreq_remove_dev - remove a CPU device
1461  *
1462  * Removes the cpufreq interface for a CPU device.
1463  */
1464 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1465 {
1466 	unsigned int cpu = dev->id;
1467 	int ret;
1468 
1469 	if (cpu_is_offline(cpu))
1470 		return 0;
1471 
1472 	ret = __cpufreq_remove_dev_prepare(dev, sif);
1473 
1474 	if (!ret)
1475 		ret = __cpufreq_remove_dev_finish(dev, sif);
1476 
1477 	return ret;
1478 }
1479 
1480 static void handle_update(struct work_struct *work)
1481 {
1482 	struct cpufreq_policy *policy =
1483 		container_of(work, struct cpufreq_policy, update);
1484 	unsigned int cpu = policy->cpu;
1485 	pr_debug("handle_update for cpu %u called\n", cpu);
1486 	cpufreq_update_policy(cpu);
1487 }
1488 
1489 /**
1490  *	cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1491  *	in deep trouble.
1492  *	@cpu: cpu number
1493  *	@old_freq: CPU frequency the kernel thinks the CPU runs at
1494  *	@new_freq: CPU frequency the CPU actually runs at
1495  *
1496  *	We adjust to current frequency first, and need to clean up later.
1497  *	So either call to cpufreq_update_policy() or schedule handle_update()).
1498  */
1499 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1500 				unsigned int new_freq)
1501 {
1502 	struct cpufreq_policy *policy;
1503 	struct cpufreq_freqs freqs;
1504 	unsigned long flags;
1505 
1506 	pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1507 		 old_freq, new_freq);
1508 
1509 	freqs.old = old_freq;
1510 	freqs.new = new_freq;
1511 
1512 	read_lock_irqsave(&cpufreq_driver_lock, flags);
1513 	policy = per_cpu(cpufreq_cpu_data, cpu);
1514 	read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1515 
1516 	cpufreq_freq_transition_begin(policy, &freqs);
1517 	cpufreq_freq_transition_end(policy, &freqs, 0);
1518 }
1519 
1520 /**
1521  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1522  * @cpu: CPU number
1523  *
1524  * This is the last known freq, without actually getting it from the driver.
1525  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1526  */
1527 unsigned int cpufreq_quick_get(unsigned int cpu)
1528 {
1529 	struct cpufreq_policy *policy;
1530 	unsigned int ret_freq = 0;
1531 
1532 	if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1533 		return cpufreq_driver->get(cpu);
1534 
1535 	policy = cpufreq_cpu_get(cpu);
1536 	if (policy) {
1537 		ret_freq = policy->cur;
1538 		cpufreq_cpu_put(policy);
1539 	}
1540 
1541 	return ret_freq;
1542 }
1543 EXPORT_SYMBOL(cpufreq_quick_get);
1544 
1545 /**
1546  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1547  * @cpu: CPU number
1548  *
1549  * Just return the max possible frequency for a given CPU.
1550  */
1551 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1552 {
1553 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1554 	unsigned int ret_freq = 0;
1555 
1556 	if (policy) {
1557 		ret_freq = policy->max;
1558 		cpufreq_cpu_put(policy);
1559 	}
1560 
1561 	return ret_freq;
1562 }
1563 EXPORT_SYMBOL(cpufreq_quick_get_max);
1564 
1565 static unsigned int __cpufreq_get(unsigned int cpu)
1566 {
1567 	struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1568 	unsigned int ret_freq = 0;
1569 
1570 	if (!cpufreq_driver->get)
1571 		return ret_freq;
1572 
1573 	ret_freq = cpufreq_driver->get(cpu);
1574 
1575 	if (ret_freq && policy->cur &&
1576 		!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1577 		/* verify no discrepancy between actual and
1578 					saved value exists */
1579 		if (unlikely(ret_freq != policy->cur)) {
1580 			cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1581 			schedule_work(&policy->update);
1582 		}
1583 	}
1584 
1585 	return ret_freq;
1586 }
1587 
1588 /**
1589  * cpufreq_get - get the current CPU frequency (in kHz)
1590  * @cpu: CPU number
1591  *
1592  * Get the CPU current (static) CPU frequency
1593  */
1594 unsigned int cpufreq_get(unsigned int cpu)
1595 {
1596 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1597 	unsigned int ret_freq = 0;
1598 
1599 	if (policy) {
1600 		down_read(&policy->rwsem);
1601 		ret_freq = __cpufreq_get(cpu);
1602 		up_read(&policy->rwsem);
1603 
1604 		cpufreq_cpu_put(policy);
1605 	}
1606 
1607 	return ret_freq;
1608 }
1609 EXPORT_SYMBOL(cpufreq_get);
1610 
1611 static struct subsys_interface cpufreq_interface = {
1612 	.name		= "cpufreq",
1613 	.subsys		= &cpu_subsys,
1614 	.add_dev	= cpufreq_add_dev,
1615 	.remove_dev	= cpufreq_remove_dev,
1616 };
1617 
1618 /*
1619  * In case platform wants some specific frequency to be configured
1620  * during suspend..
1621  */
1622 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1623 {
1624 	int ret;
1625 
1626 	if (!policy->suspend_freq) {
1627 		pr_err("%s: suspend_freq can't be zero\n", __func__);
1628 		return -EINVAL;
1629 	}
1630 
1631 	pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1632 			policy->suspend_freq);
1633 
1634 	ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1635 			CPUFREQ_RELATION_H);
1636 	if (ret)
1637 		pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1638 				__func__, policy->suspend_freq, ret);
1639 
1640 	return ret;
1641 }
1642 EXPORT_SYMBOL(cpufreq_generic_suspend);
1643 
1644 /**
1645  * cpufreq_suspend() - Suspend CPUFreq governors
1646  *
1647  * Called during system wide Suspend/Hibernate cycles for suspending governors
1648  * as some platforms can't change frequency after this point in suspend cycle.
1649  * Because some of the devices (like: i2c, regulators, etc) they use for
1650  * changing frequency are suspended quickly after this point.
1651  */
1652 void cpufreq_suspend(void)
1653 {
1654 	struct cpufreq_policy *policy;
1655 
1656 	if (!cpufreq_driver)
1657 		return;
1658 
1659 	if (!has_target())
1660 		return;
1661 
1662 	pr_debug("%s: Suspending Governors\n", __func__);
1663 
1664 	list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
1665 		if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1666 			pr_err("%s: Failed to stop governor for policy: %p\n",
1667 				__func__, policy);
1668 		else if (cpufreq_driver->suspend
1669 		    && cpufreq_driver->suspend(policy))
1670 			pr_err("%s: Failed to suspend driver: %p\n", __func__,
1671 				policy);
1672 	}
1673 
1674 	cpufreq_suspended = true;
1675 }
1676 
1677 /**
1678  * cpufreq_resume() - Resume CPUFreq governors
1679  *
1680  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1681  * are suspended with cpufreq_suspend().
1682  */
1683 void cpufreq_resume(void)
1684 {
1685 	struct cpufreq_policy *policy;
1686 
1687 	if (!cpufreq_driver)
1688 		return;
1689 
1690 	if (!has_target())
1691 		return;
1692 
1693 	pr_debug("%s: Resuming Governors\n", __func__);
1694 
1695 	cpufreq_suspended = false;
1696 
1697 	list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
1698 		if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1699 			pr_err("%s: Failed to resume driver: %p\n", __func__,
1700 				policy);
1701 		else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1702 		    || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1703 			pr_err("%s: Failed to start governor for policy: %p\n",
1704 				__func__, policy);
1705 
1706 		/*
1707 		 * schedule call cpufreq_update_policy() for boot CPU, i.e. last
1708 		 * policy in list. It will verify that the current freq is in
1709 		 * sync with what we believe it to be.
1710 		 */
1711 		if (list_is_last(&policy->policy_list, &cpufreq_policy_list))
1712 			schedule_work(&policy->update);
1713 	}
1714 }
1715 
1716 /**
1717  *	cpufreq_get_current_driver - return current driver's name
1718  *
1719  *	Return the name string of the currently loaded cpufreq driver
1720  *	or NULL, if none.
1721  */
1722 const char *cpufreq_get_current_driver(void)
1723 {
1724 	if (cpufreq_driver)
1725 		return cpufreq_driver->name;
1726 
1727 	return NULL;
1728 }
1729 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1730 
1731 /*********************************************************************
1732  *                     NOTIFIER LISTS INTERFACE                      *
1733  *********************************************************************/
1734 
1735 /**
1736  *	cpufreq_register_notifier - register a driver with cpufreq
1737  *	@nb: notifier function to register
1738  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1739  *
1740  *	Add a driver to one of two lists: either a list of drivers that
1741  *      are notified about clock rate changes (once before and once after
1742  *      the transition), or a list of drivers that are notified about
1743  *      changes in cpufreq policy.
1744  *
1745  *	This function may sleep, and has the same return conditions as
1746  *	blocking_notifier_chain_register.
1747  */
1748 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1749 {
1750 	int ret;
1751 
1752 	if (cpufreq_disabled())
1753 		return -EINVAL;
1754 
1755 	WARN_ON(!init_cpufreq_transition_notifier_list_called);
1756 
1757 	switch (list) {
1758 	case CPUFREQ_TRANSITION_NOTIFIER:
1759 		ret = srcu_notifier_chain_register(
1760 				&cpufreq_transition_notifier_list, nb);
1761 		break;
1762 	case CPUFREQ_POLICY_NOTIFIER:
1763 		ret = blocking_notifier_chain_register(
1764 				&cpufreq_policy_notifier_list, nb);
1765 		break;
1766 	default:
1767 		ret = -EINVAL;
1768 	}
1769 
1770 	return ret;
1771 }
1772 EXPORT_SYMBOL(cpufreq_register_notifier);
1773 
1774 /**
1775  *	cpufreq_unregister_notifier - unregister a driver with cpufreq
1776  *	@nb: notifier block to be unregistered
1777  *	@list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1778  *
1779  *	Remove a driver from the CPU frequency notifier list.
1780  *
1781  *	This function may sleep, and has the same return conditions as
1782  *	blocking_notifier_chain_unregister.
1783  */
1784 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1785 {
1786 	int ret;
1787 
1788 	if (cpufreq_disabled())
1789 		return -EINVAL;
1790 
1791 	switch (list) {
1792 	case CPUFREQ_TRANSITION_NOTIFIER:
1793 		ret = srcu_notifier_chain_unregister(
1794 				&cpufreq_transition_notifier_list, nb);
1795 		break;
1796 	case CPUFREQ_POLICY_NOTIFIER:
1797 		ret = blocking_notifier_chain_unregister(
1798 				&cpufreq_policy_notifier_list, nb);
1799 		break;
1800 	default:
1801 		ret = -EINVAL;
1802 	}
1803 
1804 	return ret;
1805 }
1806 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1807 
1808 
1809 /*********************************************************************
1810  *                              GOVERNORS                            *
1811  *********************************************************************/
1812 
1813 /* Must set freqs->new to intermediate frequency */
1814 static int __target_intermediate(struct cpufreq_policy *policy,
1815 				 struct cpufreq_freqs *freqs, int index)
1816 {
1817 	int ret;
1818 
1819 	freqs->new = cpufreq_driver->get_intermediate(policy, index);
1820 
1821 	/* We don't need to switch to intermediate freq */
1822 	if (!freqs->new)
1823 		return 0;
1824 
1825 	pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1826 		 __func__, policy->cpu, freqs->old, freqs->new);
1827 
1828 	cpufreq_freq_transition_begin(policy, freqs);
1829 	ret = cpufreq_driver->target_intermediate(policy, index);
1830 	cpufreq_freq_transition_end(policy, freqs, ret);
1831 
1832 	if (ret)
1833 		pr_err("%s: Failed to change to intermediate frequency: %d\n",
1834 		       __func__, ret);
1835 
1836 	return ret;
1837 }
1838 
1839 static int __target_index(struct cpufreq_policy *policy,
1840 			  struct cpufreq_frequency_table *freq_table, int index)
1841 {
1842 	struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1843 	unsigned int intermediate_freq = 0;
1844 	int retval = -EINVAL;
1845 	bool notify;
1846 
1847 	notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1848 	if (notify) {
1849 		/* Handle switching to intermediate frequency */
1850 		if (cpufreq_driver->get_intermediate) {
1851 			retval = __target_intermediate(policy, &freqs, index);
1852 			if (retval)
1853 				return retval;
1854 
1855 			intermediate_freq = freqs.new;
1856 			/* Set old freq to intermediate */
1857 			if (intermediate_freq)
1858 				freqs.old = freqs.new;
1859 		}
1860 
1861 		freqs.new = freq_table[index].frequency;
1862 		pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1863 			 __func__, policy->cpu, freqs.old, freqs.new);
1864 
1865 		cpufreq_freq_transition_begin(policy, &freqs);
1866 	}
1867 
1868 	retval = cpufreq_driver->target_index(policy, index);
1869 	if (retval)
1870 		pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1871 		       retval);
1872 
1873 	if (notify) {
1874 		cpufreq_freq_transition_end(policy, &freqs, retval);
1875 
1876 		/*
1877 		 * Failed after setting to intermediate freq? Driver should have
1878 		 * reverted back to initial frequency and so should we. Check
1879 		 * here for intermediate_freq instead of get_intermediate, in
1880 		 * case we have't switched to intermediate freq at all.
1881 		 */
1882 		if (unlikely(retval && intermediate_freq)) {
1883 			freqs.old = intermediate_freq;
1884 			freqs.new = policy->restore_freq;
1885 			cpufreq_freq_transition_begin(policy, &freqs);
1886 			cpufreq_freq_transition_end(policy, &freqs, 0);
1887 		}
1888 	}
1889 
1890 	return retval;
1891 }
1892 
1893 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1894 			    unsigned int target_freq,
1895 			    unsigned int relation)
1896 {
1897 	unsigned int old_target_freq = target_freq;
1898 	int retval = -EINVAL;
1899 
1900 	if (cpufreq_disabled())
1901 		return -ENODEV;
1902 
1903 	/* Make sure that target_freq is within supported range */
1904 	if (target_freq > policy->max)
1905 		target_freq = policy->max;
1906 	if (target_freq < policy->min)
1907 		target_freq = policy->min;
1908 
1909 	pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1910 		 policy->cpu, target_freq, relation, old_target_freq);
1911 
1912 	/*
1913 	 * This might look like a redundant call as we are checking it again
1914 	 * after finding index. But it is left intentionally for cases where
1915 	 * exactly same freq is called again and so we can save on few function
1916 	 * calls.
1917 	 */
1918 	if (target_freq == policy->cur)
1919 		return 0;
1920 
1921 	/* Save last value to restore later on errors */
1922 	policy->restore_freq = policy->cur;
1923 
1924 	if (cpufreq_driver->target)
1925 		retval = cpufreq_driver->target(policy, target_freq, relation);
1926 	else if (cpufreq_driver->target_index) {
1927 		struct cpufreq_frequency_table *freq_table;
1928 		int index;
1929 
1930 		freq_table = cpufreq_frequency_get_table(policy->cpu);
1931 		if (unlikely(!freq_table)) {
1932 			pr_err("%s: Unable to find freq_table\n", __func__);
1933 			goto out;
1934 		}
1935 
1936 		retval = cpufreq_frequency_table_target(policy, freq_table,
1937 				target_freq, relation, &index);
1938 		if (unlikely(retval)) {
1939 			pr_err("%s: Unable to find matching freq\n", __func__);
1940 			goto out;
1941 		}
1942 
1943 		if (freq_table[index].frequency == policy->cur) {
1944 			retval = 0;
1945 			goto out;
1946 		}
1947 
1948 		retval = __target_index(policy, freq_table, index);
1949 	}
1950 
1951 out:
1952 	return retval;
1953 }
1954 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1955 
1956 int cpufreq_driver_target(struct cpufreq_policy *policy,
1957 			  unsigned int target_freq,
1958 			  unsigned int relation)
1959 {
1960 	int ret = -EINVAL;
1961 
1962 	down_write(&policy->rwsem);
1963 
1964 	ret = __cpufreq_driver_target(policy, target_freq, relation);
1965 
1966 	up_write(&policy->rwsem);
1967 
1968 	return ret;
1969 }
1970 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1971 
1972 /*
1973  * when "event" is CPUFREQ_GOV_LIMITS
1974  */
1975 
1976 static int __cpufreq_governor(struct cpufreq_policy *policy,
1977 					unsigned int event)
1978 {
1979 	int ret;
1980 
1981 	/* Only must be defined when default governor is known to have latency
1982 	   restrictions, like e.g. conservative or ondemand.
1983 	   That this is the case is already ensured in Kconfig
1984 	*/
1985 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1986 	struct cpufreq_governor *gov = &cpufreq_gov_performance;
1987 #else
1988 	struct cpufreq_governor *gov = NULL;
1989 #endif
1990 
1991 	/* Don't start any governor operations if we are entering suspend */
1992 	if (cpufreq_suspended)
1993 		return 0;
1994 
1995 	if (policy->governor->max_transition_latency &&
1996 	    policy->cpuinfo.transition_latency >
1997 	    policy->governor->max_transition_latency) {
1998 		if (!gov)
1999 			return -EINVAL;
2000 		else {
2001 			pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2002 				policy->governor->name, gov->name);
2003 			policy->governor = gov;
2004 		}
2005 	}
2006 
2007 	if (event == CPUFREQ_GOV_POLICY_INIT)
2008 		if (!try_module_get(policy->governor->owner))
2009 			return -EINVAL;
2010 
2011 	pr_debug("__cpufreq_governor for CPU %u, event %u\n",
2012 		 policy->cpu, event);
2013 
2014 	mutex_lock(&cpufreq_governor_lock);
2015 	if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2016 	    || (!policy->governor_enabled
2017 	    && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2018 		mutex_unlock(&cpufreq_governor_lock);
2019 		return -EBUSY;
2020 	}
2021 
2022 	if (event == CPUFREQ_GOV_STOP)
2023 		policy->governor_enabled = false;
2024 	else if (event == CPUFREQ_GOV_START)
2025 		policy->governor_enabled = true;
2026 
2027 	mutex_unlock(&cpufreq_governor_lock);
2028 
2029 	ret = policy->governor->governor(policy, event);
2030 
2031 	if (!ret) {
2032 		if (event == CPUFREQ_GOV_POLICY_INIT)
2033 			policy->governor->initialized++;
2034 		else if (event == CPUFREQ_GOV_POLICY_EXIT)
2035 			policy->governor->initialized--;
2036 	} else {
2037 		/* Restore original values */
2038 		mutex_lock(&cpufreq_governor_lock);
2039 		if (event == CPUFREQ_GOV_STOP)
2040 			policy->governor_enabled = true;
2041 		else if (event == CPUFREQ_GOV_START)
2042 			policy->governor_enabled = false;
2043 		mutex_unlock(&cpufreq_governor_lock);
2044 	}
2045 
2046 	if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2047 			((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2048 		module_put(policy->governor->owner);
2049 
2050 	return ret;
2051 }
2052 
2053 int cpufreq_register_governor(struct cpufreq_governor *governor)
2054 {
2055 	int err;
2056 
2057 	if (!governor)
2058 		return -EINVAL;
2059 
2060 	if (cpufreq_disabled())
2061 		return -ENODEV;
2062 
2063 	mutex_lock(&cpufreq_governor_mutex);
2064 
2065 	governor->initialized = 0;
2066 	err = -EBUSY;
2067 	if (__find_governor(governor->name) == NULL) {
2068 		err = 0;
2069 		list_add(&governor->governor_list, &cpufreq_governor_list);
2070 	}
2071 
2072 	mutex_unlock(&cpufreq_governor_mutex);
2073 	return err;
2074 }
2075 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2076 
2077 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2078 {
2079 	int cpu;
2080 
2081 	if (!governor)
2082 		return;
2083 
2084 	if (cpufreq_disabled())
2085 		return;
2086 
2087 	for_each_present_cpu(cpu) {
2088 		if (cpu_online(cpu))
2089 			continue;
2090 		if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
2091 			strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
2092 	}
2093 
2094 	mutex_lock(&cpufreq_governor_mutex);
2095 	list_del(&governor->governor_list);
2096 	mutex_unlock(&cpufreq_governor_mutex);
2097 	return;
2098 }
2099 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2100 
2101 
2102 /*********************************************************************
2103  *                          POLICY INTERFACE                         *
2104  *********************************************************************/
2105 
2106 /**
2107  * cpufreq_get_policy - get the current cpufreq_policy
2108  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2109  *	is written
2110  *
2111  * Reads the current cpufreq policy.
2112  */
2113 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2114 {
2115 	struct cpufreq_policy *cpu_policy;
2116 	if (!policy)
2117 		return -EINVAL;
2118 
2119 	cpu_policy = cpufreq_cpu_get(cpu);
2120 	if (!cpu_policy)
2121 		return -EINVAL;
2122 
2123 	memcpy(policy, cpu_policy, sizeof(*policy));
2124 
2125 	cpufreq_cpu_put(cpu_policy);
2126 	return 0;
2127 }
2128 EXPORT_SYMBOL(cpufreq_get_policy);
2129 
2130 /*
2131  * policy : current policy.
2132  * new_policy: policy to be set.
2133  */
2134 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2135 				struct cpufreq_policy *new_policy)
2136 {
2137 	struct cpufreq_governor *old_gov;
2138 	int ret;
2139 
2140 	pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2141 		 new_policy->cpu, new_policy->min, new_policy->max);
2142 
2143 	memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2144 
2145 	if (new_policy->min > policy->max || new_policy->max < policy->min)
2146 		return -EINVAL;
2147 
2148 	/* verify the cpu speed can be set within this limit */
2149 	ret = cpufreq_driver->verify(new_policy);
2150 	if (ret)
2151 		return ret;
2152 
2153 	/* adjust if necessary - all reasons */
2154 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2155 			CPUFREQ_ADJUST, new_policy);
2156 
2157 	/* adjust if necessary - hardware incompatibility*/
2158 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2159 			CPUFREQ_INCOMPATIBLE, new_policy);
2160 
2161 	/*
2162 	 * verify the cpu speed can be set within this limit, which might be
2163 	 * different to the first one
2164 	 */
2165 	ret = cpufreq_driver->verify(new_policy);
2166 	if (ret)
2167 		return ret;
2168 
2169 	/* notification of the new policy */
2170 	blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2171 			CPUFREQ_NOTIFY, new_policy);
2172 
2173 	policy->min = new_policy->min;
2174 	policy->max = new_policy->max;
2175 
2176 	pr_debug("new min and max freqs are %u - %u kHz\n",
2177 		 policy->min, policy->max);
2178 
2179 	if (cpufreq_driver->setpolicy) {
2180 		policy->policy = new_policy->policy;
2181 		pr_debug("setting range\n");
2182 		return cpufreq_driver->setpolicy(new_policy);
2183 	}
2184 
2185 	if (new_policy->governor == policy->governor)
2186 		goto out;
2187 
2188 	pr_debug("governor switch\n");
2189 
2190 	/* save old, working values */
2191 	old_gov = policy->governor;
2192 	/* end old governor */
2193 	if (old_gov) {
2194 		__cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2195 		up_write(&policy->rwsem);
2196 		__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2197 		down_write(&policy->rwsem);
2198 	}
2199 
2200 	/* start new governor */
2201 	policy->governor = new_policy->governor;
2202 	if (!__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT)) {
2203 		if (!__cpufreq_governor(policy, CPUFREQ_GOV_START))
2204 			goto out;
2205 
2206 		up_write(&policy->rwsem);
2207 		__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2208 		down_write(&policy->rwsem);
2209 	}
2210 
2211 	/* new governor failed, so re-start old one */
2212 	pr_debug("starting governor %s failed\n", policy->governor->name);
2213 	if (old_gov) {
2214 		policy->governor = old_gov;
2215 		__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2216 		__cpufreq_governor(policy, CPUFREQ_GOV_START);
2217 	}
2218 
2219 	return -EINVAL;
2220 
2221  out:
2222 	pr_debug("governor: change or update limits\n");
2223 	return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2224 }
2225 
2226 /**
2227  *	cpufreq_update_policy - re-evaluate an existing cpufreq policy
2228  *	@cpu: CPU which shall be re-evaluated
2229  *
2230  *	Useful for policy notifiers which have different necessities
2231  *	at different times.
2232  */
2233 int cpufreq_update_policy(unsigned int cpu)
2234 {
2235 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2236 	struct cpufreq_policy new_policy;
2237 	int ret;
2238 
2239 	if (!policy)
2240 		return -ENODEV;
2241 
2242 	down_write(&policy->rwsem);
2243 
2244 	pr_debug("updating policy for CPU %u\n", cpu);
2245 	memcpy(&new_policy, policy, sizeof(*policy));
2246 	new_policy.min = policy->user_policy.min;
2247 	new_policy.max = policy->user_policy.max;
2248 	new_policy.policy = policy->user_policy.policy;
2249 	new_policy.governor = policy->user_policy.governor;
2250 
2251 	/*
2252 	 * BIOS might change freq behind our back
2253 	 * -> ask driver for current freq and notify governors about a change
2254 	 */
2255 	if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2256 		new_policy.cur = cpufreq_driver->get(cpu);
2257 		if (WARN_ON(!new_policy.cur)) {
2258 			ret = -EIO;
2259 			goto unlock;
2260 		}
2261 
2262 		if (!policy->cur) {
2263 			pr_debug("Driver did not initialize current freq\n");
2264 			policy->cur = new_policy.cur;
2265 		} else {
2266 			if (policy->cur != new_policy.cur && has_target())
2267 				cpufreq_out_of_sync(cpu, policy->cur,
2268 								new_policy.cur);
2269 		}
2270 	}
2271 
2272 	ret = cpufreq_set_policy(policy, &new_policy);
2273 
2274 unlock:
2275 	up_write(&policy->rwsem);
2276 
2277 	cpufreq_cpu_put(policy);
2278 	return ret;
2279 }
2280 EXPORT_SYMBOL(cpufreq_update_policy);
2281 
2282 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2283 					unsigned long action, void *hcpu)
2284 {
2285 	unsigned int cpu = (unsigned long)hcpu;
2286 	struct device *dev;
2287 
2288 	dev = get_cpu_device(cpu);
2289 	if (dev) {
2290 		switch (action & ~CPU_TASKS_FROZEN) {
2291 		case CPU_ONLINE:
2292 			__cpufreq_add_dev(dev, NULL);
2293 			break;
2294 
2295 		case CPU_DOWN_PREPARE:
2296 			__cpufreq_remove_dev_prepare(dev, NULL);
2297 			break;
2298 
2299 		case CPU_POST_DEAD:
2300 			__cpufreq_remove_dev_finish(dev, NULL);
2301 			break;
2302 
2303 		case CPU_DOWN_FAILED:
2304 			__cpufreq_add_dev(dev, NULL);
2305 			break;
2306 		}
2307 	}
2308 	return NOTIFY_OK;
2309 }
2310 
2311 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2312 	.notifier_call = cpufreq_cpu_callback,
2313 };
2314 
2315 /*********************************************************************
2316  *               BOOST						     *
2317  *********************************************************************/
2318 static int cpufreq_boost_set_sw(int state)
2319 {
2320 	struct cpufreq_frequency_table *freq_table;
2321 	struct cpufreq_policy *policy;
2322 	int ret = -EINVAL;
2323 
2324 	list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
2325 		freq_table = cpufreq_frequency_get_table(policy->cpu);
2326 		if (freq_table) {
2327 			ret = cpufreq_frequency_table_cpuinfo(policy,
2328 							freq_table);
2329 			if (ret) {
2330 				pr_err("%s: Policy frequency update failed\n",
2331 				       __func__);
2332 				break;
2333 			}
2334 			policy->user_policy.max = policy->max;
2335 			__cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2336 		}
2337 	}
2338 
2339 	return ret;
2340 }
2341 
2342 int cpufreq_boost_trigger_state(int state)
2343 {
2344 	unsigned long flags;
2345 	int ret = 0;
2346 
2347 	if (cpufreq_driver->boost_enabled == state)
2348 		return 0;
2349 
2350 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2351 	cpufreq_driver->boost_enabled = state;
2352 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2353 
2354 	ret = cpufreq_driver->set_boost(state);
2355 	if (ret) {
2356 		write_lock_irqsave(&cpufreq_driver_lock, flags);
2357 		cpufreq_driver->boost_enabled = !state;
2358 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2359 
2360 		pr_err("%s: Cannot %s BOOST\n",
2361 		       __func__, state ? "enable" : "disable");
2362 	}
2363 
2364 	return ret;
2365 }
2366 
2367 int cpufreq_boost_supported(void)
2368 {
2369 	if (likely(cpufreq_driver))
2370 		return cpufreq_driver->boost_supported;
2371 
2372 	return 0;
2373 }
2374 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2375 
2376 int cpufreq_boost_enabled(void)
2377 {
2378 	return cpufreq_driver->boost_enabled;
2379 }
2380 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2381 
2382 /*********************************************************************
2383  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2384  *********************************************************************/
2385 
2386 /**
2387  * cpufreq_register_driver - register a CPU Frequency driver
2388  * @driver_data: A struct cpufreq_driver containing the values#
2389  * submitted by the CPU Frequency driver.
2390  *
2391  * Registers a CPU Frequency driver to this core code. This code
2392  * returns zero on success, -EBUSY when another driver got here first
2393  * (and isn't unregistered in the meantime).
2394  *
2395  */
2396 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2397 {
2398 	unsigned long flags;
2399 	int ret;
2400 
2401 	if (cpufreq_disabled())
2402 		return -ENODEV;
2403 
2404 	if (!driver_data || !driver_data->verify || !driver_data->init ||
2405 	    !(driver_data->setpolicy || driver_data->target_index ||
2406 		    driver_data->target) ||
2407 	     (driver_data->setpolicy && (driver_data->target_index ||
2408 		    driver_data->target)) ||
2409 	     (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2410 		return -EINVAL;
2411 
2412 	pr_debug("trying to register driver %s\n", driver_data->name);
2413 
2414 	if (driver_data->setpolicy)
2415 		driver_data->flags |= CPUFREQ_CONST_LOOPS;
2416 
2417 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2418 	if (cpufreq_driver) {
2419 		write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2420 		return -EEXIST;
2421 	}
2422 	cpufreq_driver = driver_data;
2423 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2424 
2425 	if (cpufreq_boost_supported()) {
2426 		/*
2427 		 * Check if driver provides function to enable boost -
2428 		 * if not, use cpufreq_boost_set_sw as default
2429 		 */
2430 		if (!cpufreq_driver->set_boost)
2431 			cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2432 
2433 		ret = cpufreq_sysfs_create_file(&boost.attr);
2434 		if (ret) {
2435 			pr_err("%s: cannot register global BOOST sysfs file\n",
2436 			       __func__);
2437 			goto err_null_driver;
2438 		}
2439 	}
2440 
2441 	ret = subsys_interface_register(&cpufreq_interface);
2442 	if (ret)
2443 		goto err_boost_unreg;
2444 
2445 	if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
2446 		int i;
2447 		ret = -ENODEV;
2448 
2449 		/* check for at least one working CPU */
2450 		for (i = 0; i < nr_cpu_ids; i++)
2451 			if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
2452 				ret = 0;
2453 				break;
2454 			}
2455 
2456 		/* if all ->init() calls failed, unregister */
2457 		if (ret) {
2458 			pr_debug("no CPU initialized for driver %s\n",
2459 				 driver_data->name);
2460 			goto err_if_unreg;
2461 		}
2462 	}
2463 
2464 	register_hotcpu_notifier(&cpufreq_cpu_notifier);
2465 	pr_debug("driver %s up and running\n", driver_data->name);
2466 
2467 	return 0;
2468 err_if_unreg:
2469 	subsys_interface_unregister(&cpufreq_interface);
2470 err_boost_unreg:
2471 	if (cpufreq_boost_supported())
2472 		cpufreq_sysfs_remove_file(&boost.attr);
2473 err_null_driver:
2474 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2475 	cpufreq_driver = NULL;
2476 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2477 	return ret;
2478 }
2479 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2480 
2481 /**
2482  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2483  *
2484  * Unregister the current CPUFreq driver. Only call this if you have
2485  * the right to do so, i.e. if you have succeeded in initialising before!
2486  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2487  * currently not initialised.
2488  */
2489 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2490 {
2491 	unsigned long flags;
2492 
2493 	if (!cpufreq_driver || (driver != cpufreq_driver))
2494 		return -EINVAL;
2495 
2496 	pr_debug("unregistering driver %s\n", driver->name);
2497 
2498 	subsys_interface_unregister(&cpufreq_interface);
2499 	if (cpufreq_boost_supported())
2500 		cpufreq_sysfs_remove_file(&boost.attr);
2501 
2502 	unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2503 
2504 	down_write(&cpufreq_rwsem);
2505 	write_lock_irqsave(&cpufreq_driver_lock, flags);
2506 
2507 	cpufreq_driver = NULL;
2508 
2509 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2510 	up_write(&cpufreq_rwsem);
2511 
2512 	return 0;
2513 }
2514 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2515 
2516 static int __init cpufreq_core_init(void)
2517 {
2518 	if (cpufreq_disabled())
2519 		return -ENODEV;
2520 
2521 	cpufreq_global_kobject = kobject_create();
2522 	BUG_ON(!cpufreq_global_kobject);
2523 
2524 	return 0;
2525 }
2526 core_initcall(cpufreq_core_init);
2527