Merge branch 'linux-linaro-lsk-v4.4' into linux-linaro-lsk-v4.4-android
[firefly-linux-kernel-4.4.55.git] / drivers / cpufreq / cpufreq.c
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/syscore_ops.h>
31 #include <linux/tick.h>
32 #ifdef CONFIG_SMP
33 #include <linux/sched.h>
34 #endif
35 #include <trace/events/power.h>
36
37 static LIST_HEAD(cpufreq_policy_list);
38
39 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
40 {
41         return cpumask_empty(policy->cpus);
42 }
43
44 static bool suitable_policy(struct cpufreq_policy *policy, bool active)
45 {
46         return active == !policy_is_inactive(policy);
47 }
48
49 /* Finds Next Acive/Inactive policy */
50 static struct cpufreq_policy *next_policy(struct cpufreq_policy *policy,
51                                           bool active)
52 {
53         do {
54                 policy = list_next_entry(policy, policy_list);
55
56                 /* No more policies in the list */
57                 if (&policy->policy_list == &cpufreq_policy_list)
58                         return NULL;
59         } while (!suitable_policy(policy, active));
60
61         return policy;
62 }
63
64 static struct cpufreq_policy *first_policy(bool active)
65 {
66         struct cpufreq_policy *policy;
67
68         /* No policies in the list */
69         if (list_empty(&cpufreq_policy_list))
70                 return NULL;
71
72         policy = list_first_entry(&cpufreq_policy_list, typeof(*policy),
73                                   policy_list);
74
75         if (!suitable_policy(policy, active))
76                 policy = next_policy(policy, active);
77
78         return policy;
79 }
80
81 /* Macros to iterate over CPU policies */
82 #define for_each_suitable_policy(__policy, __active)    \
83         for (__policy = first_policy(__active);         \
84              __policy;                                  \
85              __policy = next_policy(__policy, __active))
86
87 #define for_each_active_policy(__policy)                \
88         for_each_suitable_policy(__policy, true)
89 #define for_each_inactive_policy(__policy)              \
90         for_each_suitable_policy(__policy, false)
91
92 #define for_each_policy(__policy)                       \
93         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
94
95 /* Iterate over governors */
96 static LIST_HEAD(cpufreq_governor_list);
97 #define for_each_governor(__governor)                           \
98         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
99
100 /**
101  * The "cpufreq driver" - the arch- or hardware-dependent low
102  * level driver of CPUFreq support, and its spinlock. This lock
103  * also protects the cpufreq_cpu_data array.
104  */
105 static struct cpufreq_driver *cpufreq_driver;
106 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
107 static DEFINE_RWLOCK(cpufreq_driver_lock);
108 DEFINE_MUTEX(cpufreq_governor_lock);
109
110 /* Flag to suspend/resume CPUFreq governors */
111 static bool cpufreq_suspended;
112
113 static inline bool has_target(void)
114 {
115         return cpufreq_driver->target_index || cpufreq_driver->target;
116 }
117
118 /* internal prototypes */
119 static int __cpufreq_governor(struct cpufreq_policy *policy,
120                 unsigned int event);
121 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
122 static void handle_update(struct work_struct *work);
123
124 /**
125  * Two notifier lists: the "policy" list is involved in the
126  * validation process for a new CPU frequency policy; the
127  * "transition" list for kernel code that needs to handle
128  * changes to devices when the CPU clock speed changes.
129  * The mutex locks both lists.
130  */
131 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
132 static struct srcu_notifier_head cpufreq_transition_notifier_list;
133
134 static bool init_cpufreq_transition_notifier_list_called;
135 static int __init init_cpufreq_transition_notifier_list(void)
136 {
137         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
138         init_cpufreq_transition_notifier_list_called = true;
139         return 0;
140 }
141 pure_initcall(init_cpufreq_transition_notifier_list);
142
143 static int off __read_mostly;
144 static int cpufreq_disabled(void)
145 {
146         return off;
147 }
148 void disable_cpufreq(void)
149 {
150         off = 1;
151 }
152 static DEFINE_MUTEX(cpufreq_governor_mutex);
153
154 bool have_governor_per_policy(void)
155 {
156         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
157 }
158 EXPORT_SYMBOL_GPL(have_governor_per_policy);
159
160 bool cpufreq_driver_is_slow(void)
161 {
162         return !(cpufreq_driver->flags & CPUFREQ_DRIVER_FAST);
163 }
164 EXPORT_SYMBOL_GPL(cpufreq_driver_is_slow);
165
166 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
167 {
168         if (have_governor_per_policy())
169                 return &policy->kobj;
170         else
171                 return cpufreq_global_kobject;
172 }
173 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
174
175 struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu)
176 {
177         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
178
179         return policy && !policy_is_inactive(policy) ?
180                 policy->freq_table : NULL;
181 }
182 EXPORT_SYMBOL_GPL(cpufreq_frequency_get_table);
183
184 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
185 {
186         u64 idle_time;
187         u64 cur_wall_time;
188         u64 busy_time;
189
190         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
191
192         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
193         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
194         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
195         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
196         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
197         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
198
199         idle_time = cur_wall_time - busy_time;
200         if (wall)
201                 *wall = cputime_to_usecs(cur_wall_time);
202
203         return cputime_to_usecs(idle_time);
204 }
205
206 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
207 {
208         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
209
210         if (idle_time == -1ULL)
211                 return get_cpu_idle_time_jiffy(cpu, wall);
212         else if (!io_busy)
213                 idle_time += get_cpu_iowait_time_us(cpu, wall);
214
215         return idle_time;
216 }
217 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
218
219 /*
220  * This is a generic cpufreq init() routine which can be used by cpufreq
221  * drivers of SMP systems. It will do following:
222  * - validate & show freq table passed
223  * - set policies transition latency
224  * - policy->cpus with all possible CPUs
225  */
226 int cpufreq_generic_init(struct cpufreq_policy *policy,
227                 struct cpufreq_frequency_table *table,
228                 unsigned int transition_latency)
229 {
230         int ret;
231
232         ret = cpufreq_table_validate_and_show(policy, table);
233         if (ret) {
234                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
235                 return ret;
236         }
237
238         policy->cpuinfo.transition_latency = transition_latency;
239
240         /*
241          * The driver only supports the SMP configuration where all processors
242          * share the clock and voltage and clock.
243          */
244         cpumask_setall(policy->cpus);
245
246         return 0;
247 }
248 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
249
250 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
251 {
252         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
253
254         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
255 }
256 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
257
258 unsigned int cpufreq_generic_get(unsigned int cpu)
259 {
260         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
261
262         if (!policy || IS_ERR(policy->clk)) {
263                 pr_err("%s: No %s associated to cpu: %d\n",
264                        __func__, policy ? "clk" : "policy", cpu);
265                 return 0;
266         }
267
268         return clk_get_rate(policy->clk) / 1000;
269 }
270 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
271
272 /**
273  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
274  *
275  * @cpu: cpu to find policy for.
276  *
277  * This returns policy for 'cpu', returns NULL if it doesn't exist.
278  * It also increments the kobject reference count to mark it busy and so would
279  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
280  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
281  * freed as that depends on the kobj count.
282  *
283  * Return: A valid policy on success, otherwise NULL on failure.
284  */
285 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
286 {
287         struct cpufreq_policy *policy = NULL;
288         unsigned long flags;
289
290         if (WARN_ON(cpu >= nr_cpu_ids))
291                 return NULL;
292
293         /* get the cpufreq driver */
294         read_lock_irqsave(&cpufreq_driver_lock, flags);
295
296         if (cpufreq_driver) {
297                 /* get the CPU */
298                 policy = cpufreq_cpu_get_raw(cpu);
299                 if (policy)
300                         kobject_get(&policy->kobj);
301         }
302
303         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
304
305         return policy;
306 }
307 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
308
309 /**
310  * cpufreq_cpu_put: Decrements the usage count of a policy
311  *
312  * @policy: policy earlier returned by cpufreq_cpu_get().
313  *
314  * This decrements the kobject reference count incremented earlier by calling
315  * cpufreq_cpu_get().
316  */
317 void cpufreq_cpu_put(struct cpufreq_policy *policy)
318 {
319         kobject_put(&policy->kobj);
320 }
321 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
322
323 /*********************************************************************
324  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
325  *********************************************************************/
326
327 /**
328  * adjust_jiffies - adjust the system "loops_per_jiffy"
329  *
330  * This function alters the system "loops_per_jiffy" for the clock
331  * speed change. Note that loops_per_jiffy cannot be updated on SMP
332  * systems as each CPU might be scaled differently. So, use the arch
333  * per-CPU loops_per_jiffy value wherever possible.
334  */
335 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
336 {
337 #ifndef CONFIG_SMP
338         static unsigned long l_p_j_ref;
339         static unsigned int l_p_j_ref_freq;
340
341         if (ci->flags & CPUFREQ_CONST_LOOPS)
342                 return;
343
344         if (!l_p_j_ref_freq) {
345                 l_p_j_ref = loops_per_jiffy;
346                 l_p_j_ref_freq = ci->old;
347                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
348                          l_p_j_ref, l_p_j_ref_freq);
349         }
350         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
351                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
352                                                                 ci->new);
353                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
354                          loops_per_jiffy, ci->new);
355         }
356 #endif
357 }
358
359 /*********************************************************************
360  *               FREQUENCY INVARIANT CPU CAPACITY                    *
361  *********************************************************************/
362
363 static DEFINE_PER_CPU(unsigned long, freq_scale) = SCHED_CAPACITY_SCALE;
364 static DEFINE_PER_CPU(unsigned long, max_freq_scale) = SCHED_CAPACITY_SCALE;
365
366 static void
367 scale_freq_capacity(struct cpufreq_policy *policy, struct cpufreq_freqs *freqs)
368 {
369         unsigned long cur = freqs ? freqs->new : policy->cur;
370         unsigned long scale = (cur << SCHED_CAPACITY_SHIFT) / policy->max;
371         struct cpufreq_cpuinfo *cpuinfo = &policy->cpuinfo;
372         int cpu;
373
374         pr_debug("cpus %*pbl cur/cur max freq %lu/%u kHz freq scale %lu\n",
375                  cpumask_pr_args(policy->cpus), cur, policy->max, scale);
376
377         for_each_cpu(cpu, policy->cpus)
378                 per_cpu(freq_scale, cpu) = scale;
379
380         if (freqs)
381                 return;
382
383         scale = (policy->max << SCHED_CAPACITY_SHIFT) / cpuinfo->max_freq;
384
385         pr_debug("cpus %*pbl cur max/max freq %u/%u kHz max freq scale %lu\n",
386                  cpumask_pr_args(policy->cpus), policy->max, cpuinfo->max_freq,
387                  scale);
388
389         for_each_cpu(cpu, policy->cpus)
390                 per_cpu(max_freq_scale, cpu) = scale;
391 }
392
393 unsigned long cpufreq_scale_freq_capacity(struct sched_domain *sd, int cpu)
394 {
395         return per_cpu(freq_scale, cpu);
396 }
397
398 unsigned long cpufreq_scale_max_freq_capacity(int cpu)
399 {
400         return per_cpu(max_freq_scale, cpu);
401 }
402
403 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
404                 struct cpufreq_freqs *freqs, unsigned int state)
405 {
406         BUG_ON(irqs_disabled());
407
408         if (cpufreq_disabled())
409                 return;
410
411         freqs->flags = cpufreq_driver->flags;
412         pr_debug("notification %u of frequency transition to %u kHz\n",
413                  state, freqs->new);
414
415         switch (state) {
416
417         case CPUFREQ_PRECHANGE:
418                 /* detect if the driver reported a value as "old frequency"
419                  * which is not equal to what the cpufreq core thinks is
420                  * "old frequency".
421                  */
422                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
423                         if ((policy) && (policy->cpu == freqs->cpu) &&
424                             (policy->cur) && (policy->cur != freqs->old)) {
425                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
426                                          freqs->old, policy->cur);
427                                 freqs->old = policy->cur;
428                         }
429                 }
430                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
431                                 CPUFREQ_PRECHANGE, freqs);
432                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
433                 break;
434
435         case CPUFREQ_POSTCHANGE:
436                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
437                 pr_debug("FREQ: %lu - CPU: %lu\n",
438                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
439                 trace_cpu_frequency(freqs->new, freqs->cpu);
440                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
441                                 CPUFREQ_POSTCHANGE, freqs);
442                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
443                         policy->cur = freqs->new;
444                 break;
445         }
446 }
447
448 /**
449  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
450  * on frequency transition.
451  *
452  * This function calls the transition notifiers and the "adjust_jiffies"
453  * function. It is called twice on all CPU frequency changes that have
454  * external effects.
455  */
456 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
457                 struct cpufreq_freqs *freqs, unsigned int state)
458 {
459         for_each_cpu(freqs->cpu, policy->cpus)
460                 __cpufreq_notify_transition(policy, freqs, state);
461 }
462
463 /* Do post notifications when there are chances that transition has failed */
464 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
465                 struct cpufreq_freqs *freqs, int transition_failed)
466 {
467         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
468         if (!transition_failed)
469                 return;
470
471         swap(freqs->old, freqs->new);
472         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
473         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
474 }
475
476 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
477                 struct cpufreq_freqs *freqs)
478 {
479 #ifdef CONFIG_SMP
480         int cpu;
481 #endif
482
483         /*
484          * Catch double invocations of _begin() which lead to self-deadlock.
485          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
486          * doesn't invoke _begin() on their behalf, and hence the chances of
487          * double invocations are very low. Moreover, there are scenarios
488          * where these checks can emit false-positive warnings in these
489          * drivers; so we avoid that by skipping them altogether.
490          */
491         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
492                                 && current == policy->transition_task);
493
494 wait:
495         wait_event(policy->transition_wait, !policy->transition_ongoing);
496
497         spin_lock(&policy->transition_lock);
498
499         if (unlikely(policy->transition_ongoing)) {
500                 spin_unlock(&policy->transition_lock);
501                 goto wait;
502         }
503
504         policy->transition_ongoing = true;
505         policy->transition_task = current;
506
507         spin_unlock(&policy->transition_lock);
508
509         scale_freq_capacity(policy, freqs);
510 #ifdef CONFIG_SMP
511         for_each_cpu(cpu, policy->cpus)
512                 trace_cpu_capacity(capacity_curr_of(cpu), cpu);
513 #endif
514
515         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
516 }
517 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
518
519 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
520                 struct cpufreq_freqs *freqs, int transition_failed)
521 {
522         if (unlikely(WARN_ON(!policy->transition_ongoing)))
523                 return;
524
525         cpufreq_notify_post_transition(policy, freqs, transition_failed);
526
527         policy->transition_ongoing = false;
528         policy->transition_task = NULL;
529
530         wake_up(&policy->transition_wait);
531 }
532 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
533
534
535 /*********************************************************************
536  *                          SYSFS INTERFACE                          *
537  *********************************************************************/
538 static ssize_t show_boost(struct kobject *kobj,
539                                  struct attribute *attr, char *buf)
540 {
541         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
542 }
543
544 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
545                                   const char *buf, size_t count)
546 {
547         int ret, enable;
548
549         ret = sscanf(buf, "%d", &enable);
550         if (ret != 1 || enable < 0 || enable > 1)
551                 return -EINVAL;
552
553         if (cpufreq_boost_trigger_state(enable)) {
554                 pr_err("%s: Cannot %s BOOST!\n",
555                        __func__, enable ? "enable" : "disable");
556                 return -EINVAL;
557         }
558
559         pr_debug("%s: cpufreq BOOST %s\n",
560                  __func__, enable ? "enabled" : "disabled");
561
562         return count;
563 }
564 define_one_global_rw(boost);
565
566 static struct cpufreq_governor *find_governor(const char *str_governor)
567 {
568         struct cpufreq_governor *t;
569
570         for_each_governor(t)
571                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
572                         return t;
573
574         return NULL;
575 }
576
577 /**
578  * cpufreq_parse_governor - parse a governor string
579  */
580 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
581                                 struct cpufreq_governor **governor)
582 {
583         int err = -EINVAL;
584
585         if (cpufreq_driver->setpolicy) {
586                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
587                         *policy = CPUFREQ_POLICY_PERFORMANCE;
588                         err = 0;
589                 } else if (!strncasecmp(str_governor, "powersave",
590                                                 CPUFREQ_NAME_LEN)) {
591                         *policy = CPUFREQ_POLICY_POWERSAVE;
592                         err = 0;
593                 }
594         } else {
595                 struct cpufreq_governor *t;
596
597                 mutex_lock(&cpufreq_governor_mutex);
598
599                 t = find_governor(str_governor);
600
601                 if (t == NULL) {
602                         int ret;
603
604                         mutex_unlock(&cpufreq_governor_mutex);
605                         ret = request_module("cpufreq_%s", str_governor);
606                         mutex_lock(&cpufreq_governor_mutex);
607
608                         if (ret == 0)
609                                 t = find_governor(str_governor);
610                 }
611
612                 if (t != NULL) {
613                         *governor = t;
614                         err = 0;
615                 }
616
617                 mutex_unlock(&cpufreq_governor_mutex);
618         }
619         return err;
620 }
621
622 /**
623  * cpufreq_per_cpu_attr_read() / show_##file_name() -
624  * print out cpufreq information
625  *
626  * Write out information from cpufreq_driver->policy[cpu]; object must be
627  * "unsigned int".
628  */
629
630 #define show_one(file_name, object)                     \
631 static ssize_t show_##file_name                         \
632 (struct cpufreq_policy *policy, char *buf)              \
633 {                                                       \
634         return sprintf(buf, "%u\n", policy->object);    \
635 }
636
637 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
638 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
639 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
640 show_one(scaling_min_freq, min);
641 show_one(scaling_max_freq, max);
642
643 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
644 {
645         ssize_t ret;
646
647         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
648                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
649         else
650                 ret = sprintf(buf, "%u\n", policy->cur);
651         return ret;
652 }
653
654 static int cpufreq_set_policy(struct cpufreq_policy *policy,
655                                 struct cpufreq_policy *new_policy);
656
657 /**
658  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
659  */
660 #define store_one(file_name, object)                    \
661 static ssize_t store_##file_name                                        \
662 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
663 {                                                                       \
664         int ret, temp;                                                  \
665         struct cpufreq_policy new_policy;                               \
666                                                                         \
667         memcpy(&new_policy, policy, sizeof(*policy));                   \
668                                                                         \
669         ret = sscanf(buf, "%u", &new_policy.object);                    \
670         if (ret != 1)                                                   \
671                 return -EINVAL;                                         \
672                                                                         \
673         temp = new_policy.object;                                       \
674         ret = cpufreq_set_policy(policy, &new_policy);          \
675         if (!ret)                                                       \
676                 policy->user_policy.object = temp;                      \
677                                                                         \
678         return ret ? ret : count;                                       \
679 }
680
681 store_one(scaling_min_freq, min);
682 store_one(scaling_max_freq, max);
683
684 /**
685  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
686  */
687 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
688                                         char *buf)
689 {
690         unsigned int cur_freq = __cpufreq_get(policy);
691         if (!cur_freq)
692                 return sprintf(buf, "<unknown>");
693         return sprintf(buf, "%u\n", cur_freq);
694 }
695
696 /**
697  * show_scaling_governor - show the current policy for the specified CPU
698  */
699 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
700 {
701         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
702                 return sprintf(buf, "powersave\n");
703         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
704                 return sprintf(buf, "performance\n");
705         else if (policy->governor)
706                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
707                                 policy->governor->name);
708         return -EINVAL;
709 }
710
711 /**
712  * store_scaling_governor - store policy for the specified CPU
713  */
714 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
715                                         const char *buf, size_t count)
716 {
717         int ret;
718         char    str_governor[16];
719         struct cpufreq_policy new_policy;
720
721         memcpy(&new_policy, policy, sizeof(*policy));
722
723         ret = sscanf(buf, "%15s", str_governor);
724         if (ret != 1)
725                 return -EINVAL;
726
727         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
728                                                 &new_policy.governor))
729                 return -EINVAL;
730
731         ret = cpufreq_set_policy(policy, &new_policy);
732         return ret ? ret : count;
733 }
734
735 /**
736  * show_scaling_driver - show the cpufreq driver currently loaded
737  */
738 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
739 {
740         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
741 }
742
743 /**
744  * show_scaling_available_governors - show the available CPUfreq governors
745  */
746 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
747                                                 char *buf)
748 {
749         ssize_t i = 0;
750         struct cpufreq_governor *t;
751
752         if (!has_target()) {
753                 i += sprintf(buf, "performance powersave");
754                 goto out;
755         }
756
757         for_each_governor(t) {
758                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
759                     - (CPUFREQ_NAME_LEN + 2)))
760                         goto out;
761                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
762         }
763 out:
764         i += sprintf(&buf[i], "\n");
765         return i;
766 }
767
768 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
769 {
770         ssize_t i = 0;
771         unsigned int cpu;
772
773         for_each_cpu(cpu, mask) {
774                 if (i)
775                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
776                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
777                 if (i >= (PAGE_SIZE - 5))
778                         break;
779         }
780         i += sprintf(&buf[i], "\n");
781         return i;
782 }
783 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
784
785 /**
786  * show_related_cpus - show the CPUs affected by each transition even if
787  * hw coordination is in use
788  */
789 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
790 {
791         return cpufreq_show_cpus(policy->related_cpus, buf);
792 }
793
794 /**
795  * show_affected_cpus - show the CPUs affected by each transition
796  */
797 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
798 {
799         return cpufreq_show_cpus(policy->cpus, buf);
800 }
801
802 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
803                                         const char *buf, size_t count)
804 {
805         unsigned int freq = 0;
806         unsigned int ret;
807
808         if (!policy->governor || !policy->governor->store_setspeed)
809                 return -EINVAL;
810
811         ret = sscanf(buf, "%u", &freq);
812         if (ret != 1)
813                 return -EINVAL;
814
815         policy->governor->store_setspeed(policy, freq);
816
817         return count;
818 }
819
820 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
821 {
822         if (!policy->governor || !policy->governor->show_setspeed)
823                 return sprintf(buf, "<unsupported>\n");
824
825         return policy->governor->show_setspeed(policy, buf);
826 }
827
828 /**
829  * show_bios_limit - show the current cpufreq HW/BIOS limitation
830  */
831 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
832 {
833         unsigned int limit;
834         int ret;
835         if (cpufreq_driver->bios_limit) {
836                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
837                 if (!ret)
838                         return sprintf(buf, "%u\n", limit);
839         }
840         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
841 }
842
843 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
844 cpufreq_freq_attr_ro(cpuinfo_min_freq);
845 cpufreq_freq_attr_ro(cpuinfo_max_freq);
846 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
847 cpufreq_freq_attr_ro(scaling_available_governors);
848 cpufreq_freq_attr_ro(scaling_driver);
849 cpufreq_freq_attr_ro(scaling_cur_freq);
850 cpufreq_freq_attr_ro(bios_limit);
851 cpufreq_freq_attr_ro(related_cpus);
852 cpufreq_freq_attr_ro(affected_cpus);
853 cpufreq_freq_attr_rw(scaling_min_freq);
854 cpufreq_freq_attr_rw(scaling_max_freq);
855 cpufreq_freq_attr_rw(scaling_governor);
856 cpufreq_freq_attr_rw(scaling_setspeed);
857
858 static struct attribute *default_attrs[] = {
859         &cpuinfo_min_freq.attr,
860         &cpuinfo_max_freq.attr,
861         &cpuinfo_transition_latency.attr,
862         &scaling_min_freq.attr,
863         &scaling_max_freq.attr,
864         &affected_cpus.attr,
865         &related_cpus.attr,
866         &scaling_governor.attr,
867         &scaling_driver.attr,
868         &scaling_available_governors.attr,
869         &scaling_setspeed.attr,
870         NULL
871 };
872
873 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
874 #define to_attr(a) container_of(a, struct freq_attr, attr)
875
876 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
877 {
878         struct cpufreq_policy *policy = to_policy(kobj);
879         struct freq_attr *fattr = to_attr(attr);
880         ssize_t ret;
881
882         down_read(&policy->rwsem);
883
884         if (fattr->show)
885                 ret = fattr->show(policy, buf);
886         else
887                 ret = -EIO;
888
889         up_read(&policy->rwsem);
890
891         return ret;
892 }
893
894 static ssize_t store(struct kobject *kobj, struct attribute *attr,
895                      const char *buf, size_t count)
896 {
897         struct cpufreq_policy *policy = to_policy(kobj);
898         struct freq_attr *fattr = to_attr(attr);
899         ssize_t ret = -EINVAL;
900
901         get_online_cpus();
902
903         if (!cpu_online(policy->cpu))
904                 goto unlock;
905
906         down_write(&policy->rwsem);
907
908         if (fattr->store)
909                 ret = fattr->store(policy, buf, count);
910         else
911                 ret = -EIO;
912
913         up_write(&policy->rwsem);
914 unlock:
915         put_online_cpus();
916
917         return ret;
918 }
919
920 static void cpufreq_sysfs_release(struct kobject *kobj)
921 {
922         struct cpufreq_policy *policy = to_policy(kobj);
923         pr_debug("last reference is dropped\n");
924         complete(&policy->kobj_unregister);
925 }
926
927 static const struct sysfs_ops sysfs_ops = {
928         .show   = show,
929         .store  = store,
930 };
931
932 static struct kobj_type ktype_cpufreq = {
933         .sysfs_ops      = &sysfs_ops,
934         .default_attrs  = default_attrs,
935         .release        = cpufreq_sysfs_release,
936 };
937
938 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
939 {
940         struct device *cpu_dev;
941
942         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
943
944         if (!policy)
945                 return 0;
946
947         cpu_dev = get_cpu_device(cpu);
948         if (WARN_ON(!cpu_dev))
949                 return 0;
950
951         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
952 }
953
954 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
955 {
956         struct device *cpu_dev;
957
958         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
959
960         cpu_dev = get_cpu_device(cpu);
961         if (WARN_ON(!cpu_dev))
962                 return;
963
964         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
965 }
966
967 /* Add/remove symlinks for all related CPUs */
968 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
969 {
970         unsigned int j;
971         int ret = 0;
972
973         /* Some related CPUs might not be present (physically hotplugged) */
974         for_each_cpu(j, policy->real_cpus) {
975                 ret = add_cpu_dev_symlink(policy, j);
976                 if (ret)
977                         break;
978         }
979
980         return ret;
981 }
982
983 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
984 {
985         unsigned int j;
986
987         /* Some related CPUs might not be present (physically hotplugged) */
988         for_each_cpu(j, policy->real_cpus)
989                 remove_cpu_dev_symlink(policy, j);
990 }
991
992 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
993 {
994         struct freq_attr **drv_attr;
995         int ret = 0;
996
997         /* set up files for this cpu device */
998         drv_attr = cpufreq_driver->attr;
999         while (drv_attr && *drv_attr) {
1000                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1001                 if (ret)
1002                         return ret;
1003                 drv_attr++;
1004         }
1005         if (cpufreq_driver->get) {
1006                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1007                 if (ret)
1008                         return ret;
1009         }
1010
1011         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1012         if (ret)
1013                 return ret;
1014
1015         if (cpufreq_driver->bios_limit) {
1016                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1017                 if (ret)
1018                         return ret;
1019         }
1020
1021         return cpufreq_add_dev_symlink(policy);
1022 }
1023
1024 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1025 {
1026         struct cpufreq_governor *gov = NULL;
1027         struct cpufreq_policy new_policy;
1028
1029         memcpy(&new_policy, policy, sizeof(*policy));
1030
1031         /* Update governor of new_policy to the governor used before hotplug */
1032         gov = find_governor(policy->last_governor);
1033         if (gov)
1034                 pr_debug("Restoring governor %s for cpu %d\n",
1035                                 policy->governor->name, policy->cpu);
1036         else
1037                 gov = CPUFREQ_DEFAULT_GOVERNOR;
1038
1039         new_policy.governor = gov;
1040
1041         /* Use the default policy if there is no last_policy. */
1042         if (cpufreq_driver->setpolicy) {
1043                 if (policy->last_policy)
1044                         new_policy.policy = policy->last_policy;
1045                 else
1046                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1047                                                NULL);
1048         }
1049         /* set default policy */
1050         return cpufreq_set_policy(policy, &new_policy);
1051 }
1052
1053 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1054 {
1055         int ret = 0;
1056
1057         /* Has this CPU been taken care of already? */
1058         if (cpumask_test_cpu(cpu, policy->cpus))
1059                 return 0;
1060
1061         if (has_target()) {
1062                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1063                 if (ret) {
1064                         pr_err("%s: Failed to stop governor\n", __func__);
1065                         return ret;
1066                 }
1067         }
1068
1069         down_write(&policy->rwsem);
1070         cpumask_set_cpu(cpu, policy->cpus);
1071         up_write(&policy->rwsem);
1072
1073         if (has_target()) {
1074                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1075                 if (!ret)
1076                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1077
1078                 if (ret) {
1079                         pr_err("%s: Failed to start governor\n", __func__);
1080                         return ret;
1081                 }
1082         }
1083
1084         return 0;
1085 }
1086
1087 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1088 {
1089         struct device *dev = get_cpu_device(cpu);
1090         struct cpufreq_policy *policy;
1091
1092         if (WARN_ON(!dev))
1093                 return NULL;
1094
1095         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1096         if (!policy)
1097                 return NULL;
1098
1099         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1100                 goto err_free_policy;
1101
1102         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1103                 goto err_free_cpumask;
1104
1105         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1106                 goto err_free_rcpumask;
1107
1108         kobject_init(&policy->kobj, &ktype_cpufreq);
1109         INIT_LIST_HEAD(&policy->policy_list);
1110         init_rwsem(&policy->rwsem);
1111         spin_lock_init(&policy->transition_lock);
1112         init_waitqueue_head(&policy->transition_wait);
1113         init_completion(&policy->kobj_unregister);
1114         INIT_WORK(&policy->update, handle_update);
1115
1116         policy->cpu = cpu;
1117         return policy;
1118
1119 err_free_rcpumask:
1120         free_cpumask_var(policy->related_cpus);
1121 err_free_cpumask:
1122         free_cpumask_var(policy->cpus);
1123 err_free_policy:
1124         kfree(policy);
1125
1126         return NULL;
1127 }
1128
1129 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1130 {
1131         struct kobject *kobj;
1132         struct completion *cmp;
1133
1134         if (notify)
1135                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1136                                              CPUFREQ_REMOVE_POLICY, policy);
1137
1138         down_write(&policy->rwsem);
1139         cpufreq_remove_dev_symlink(policy);
1140         kobj = &policy->kobj;
1141         cmp = &policy->kobj_unregister;
1142         up_write(&policy->rwsem);
1143         kobject_put(kobj);
1144
1145         /*
1146          * We need to make sure that the underlying kobj is
1147          * actually not referenced anymore by anybody before we
1148          * proceed with unloading.
1149          */
1150         pr_debug("waiting for dropping of refcount\n");
1151         wait_for_completion(cmp);
1152         pr_debug("wait complete\n");
1153 }
1154
1155 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1156 {
1157         unsigned long flags;
1158         int cpu;
1159
1160         /* Remove policy from list */
1161         write_lock_irqsave(&cpufreq_driver_lock, flags);
1162         list_del(&policy->policy_list);
1163
1164         for_each_cpu(cpu, policy->related_cpus)
1165                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1166         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1167
1168         cpufreq_policy_put_kobj(policy, notify);
1169         free_cpumask_var(policy->real_cpus);
1170         free_cpumask_var(policy->related_cpus);
1171         free_cpumask_var(policy->cpus);
1172         kfree(policy);
1173 }
1174
1175 static int cpufreq_online(unsigned int cpu)
1176 {
1177         struct cpufreq_policy *policy;
1178         bool new_policy;
1179         unsigned long flags;
1180         unsigned int j;
1181         int ret;
1182
1183         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1184
1185         /* Check if this CPU already has a policy to manage it */
1186         policy = per_cpu(cpufreq_cpu_data, cpu);
1187         if (policy) {
1188                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1189                 if (!policy_is_inactive(policy))
1190                         return cpufreq_add_policy_cpu(policy, cpu);
1191
1192                 /* This is the only online CPU for the policy.  Start over. */
1193                 new_policy = false;
1194                 down_write(&policy->rwsem);
1195                 policy->cpu = cpu;
1196                 policy->governor = NULL;
1197                 up_write(&policy->rwsem);
1198         } else {
1199                 new_policy = true;
1200                 policy = cpufreq_policy_alloc(cpu);
1201                 if (!policy)
1202                         return -ENOMEM;
1203         }
1204
1205         cpumask_copy(policy->cpus, cpumask_of(cpu));
1206
1207         /* call driver. From then on the cpufreq must be able
1208          * to accept all calls to ->verify and ->setpolicy for this CPU
1209          */
1210         ret = cpufreq_driver->init(policy);
1211         if (ret) {
1212                 pr_debug("initialization failed\n");
1213                 goto out_free_policy;
1214         }
1215
1216         down_write(&policy->rwsem);
1217
1218         if (new_policy) {
1219                 /* related_cpus should at least include policy->cpus. */
1220                 cpumask_copy(policy->related_cpus, policy->cpus);
1221                 /* Remember CPUs present at the policy creation time. */
1222                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1223
1224                 /* Name and add the kobject */
1225                 ret = kobject_add(&policy->kobj, cpufreq_global_kobject,
1226                                   "policy%u",
1227                                   cpumask_first(policy->related_cpus));
1228                 if (ret) {
1229                         pr_err("%s: failed to add policy->kobj: %d\n", __func__,
1230                                ret);
1231                         goto out_exit_policy;
1232                 }
1233         }
1234
1235         /*
1236          * affected cpus must always be the one, which are online. We aren't
1237          * managing offline cpus here.
1238          */
1239         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1240
1241         if (new_policy) {
1242                 policy->user_policy.min = policy->min;
1243                 policy->user_policy.max = policy->max;
1244
1245                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1246                 for_each_cpu(j, policy->related_cpus)
1247                         per_cpu(cpufreq_cpu_data, j) = policy;
1248                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1249         }
1250
1251         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1252                 policy->cur = cpufreq_driver->get(policy->cpu);
1253                 if (!policy->cur) {
1254                         pr_err("%s: ->get() failed\n", __func__);
1255                         goto out_exit_policy;
1256                 }
1257         }
1258
1259         /*
1260          * Sometimes boot loaders set CPU frequency to a value outside of
1261          * frequency table present with cpufreq core. In such cases CPU might be
1262          * unstable if it has to run on that frequency for long duration of time
1263          * and so its better to set it to a frequency which is specified in
1264          * freq-table. This also makes cpufreq stats inconsistent as
1265          * cpufreq-stats would fail to register because current frequency of CPU
1266          * isn't found in freq-table.
1267          *
1268          * Because we don't want this change to effect boot process badly, we go
1269          * for the next freq which is >= policy->cur ('cur' must be set by now,
1270          * otherwise we will end up setting freq to lowest of the table as 'cur'
1271          * is initialized to zero).
1272          *
1273          * We are passing target-freq as "policy->cur - 1" otherwise
1274          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1275          * equal to target-freq.
1276          */
1277         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1278             && has_target()) {
1279                 /* Are we running at unknown frequency ? */
1280                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1281                 if (ret == -EINVAL) {
1282                         /* Warn user and fix it */
1283                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1284                                 __func__, policy->cpu, policy->cur);
1285                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1286                                 CPUFREQ_RELATION_L);
1287
1288                         /*
1289                          * Reaching here after boot in a few seconds may not
1290                          * mean that system will remain stable at "unknown"
1291                          * frequency for longer duration. Hence, a BUG_ON().
1292                          */
1293                         BUG_ON(ret);
1294                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1295                                 __func__, policy->cpu, policy->cur);
1296                 }
1297         }
1298
1299         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1300                                      CPUFREQ_START, policy);
1301
1302         if (new_policy) {
1303                 ret = cpufreq_add_dev_interface(policy);
1304                 if (ret)
1305                         goto out_exit_policy;
1306                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1307                                 CPUFREQ_CREATE_POLICY, policy);
1308
1309                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1310                 list_add(&policy->policy_list, &cpufreq_policy_list);
1311                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1312         }
1313
1314         ret = cpufreq_init_policy(policy);
1315         if (ret) {
1316                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1317                        __func__, cpu, ret);
1318                 /* cpufreq_policy_free() will notify based on this */
1319                 new_policy = false;
1320                 goto out_exit_policy;
1321         }
1322
1323         up_write(&policy->rwsem);
1324
1325         kobject_uevent(&policy->kobj, KOBJ_ADD);
1326
1327         /* Callback for handling stuff after policy is ready */
1328         if (cpufreq_driver->ready)
1329                 cpufreq_driver->ready(policy);
1330
1331         pr_debug("initialization complete\n");
1332
1333         return 0;
1334
1335 out_exit_policy:
1336         up_write(&policy->rwsem);
1337
1338         if (cpufreq_driver->exit)
1339                 cpufreq_driver->exit(policy);
1340 out_free_policy:
1341         cpufreq_policy_free(policy, !new_policy);
1342         return ret;
1343 }
1344
1345 /**
1346  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1347  * @dev: CPU device.
1348  * @sif: Subsystem interface structure pointer (not used)
1349  */
1350 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1351 {
1352         unsigned cpu = dev->id;
1353         int ret;
1354
1355         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1356
1357         if (cpu_online(cpu)) {
1358                 ret = cpufreq_online(cpu);
1359         } else {
1360                 /*
1361                  * A hotplug notifier will follow and we will handle it as CPU
1362                  * online then.  For now, just create the sysfs link, unless
1363                  * there is no policy or the link is already present.
1364                  */
1365                 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1366
1367                 ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1368                         ? add_cpu_dev_symlink(policy, cpu) : 0;
1369         }
1370
1371         return ret;
1372 }
1373
1374 static void cpufreq_offline_prepare(unsigned int cpu)
1375 {
1376         struct cpufreq_policy *policy;
1377
1378         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1379
1380         policy = cpufreq_cpu_get_raw(cpu);
1381         if (!policy) {
1382                 pr_debug("%s: No cpu_data found\n", __func__);
1383                 return;
1384         }
1385
1386         if (has_target()) {
1387                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1388                 if (ret)
1389                         pr_err("%s: Failed to stop governor\n", __func__);
1390         }
1391
1392         down_write(&policy->rwsem);
1393         cpumask_clear_cpu(cpu, policy->cpus);
1394
1395         if (policy_is_inactive(policy)) {
1396                 if (has_target())
1397                         strncpy(policy->last_governor, policy->governor->name,
1398                                 CPUFREQ_NAME_LEN);
1399                 else
1400                         policy->last_policy = policy->policy;
1401         } else if (cpu == policy->cpu) {
1402                 /* Nominate new CPU */
1403                 policy->cpu = cpumask_any(policy->cpus);
1404         }
1405         up_write(&policy->rwsem);
1406
1407         /* Start governor again for active policy */
1408         if (!policy_is_inactive(policy)) {
1409                 if (has_target()) {
1410                         int ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1411                         if (!ret)
1412                                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1413
1414                         if (ret)
1415                                 pr_err("%s: Failed to start governor\n", __func__);
1416                 }
1417         } else if (cpufreq_driver->stop_cpu) {
1418                 cpufreq_driver->stop_cpu(policy);
1419         }
1420 }
1421
1422 static void cpufreq_offline_finish(unsigned int cpu)
1423 {
1424         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1425
1426         if (!policy) {
1427                 pr_debug("%s: No cpu_data found\n", __func__);
1428                 return;
1429         }
1430
1431         /* Only proceed for inactive policies */
1432         if (!policy_is_inactive(policy))
1433                 return;
1434
1435         /* If cpu is last user of policy, free policy */
1436         if (has_target()) {
1437                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1438                 if (ret)
1439                         pr_err("%s: Failed to exit governor\n", __func__);
1440         }
1441
1442         /*
1443          * Perform the ->exit() even during light-weight tear-down,
1444          * since this is a core component, and is essential for the
1445          * subsequent light-weight ->init() to succeed.
1446          */
1447         if (cpufreq_driver->exit) {
1448                 cpufreq_driver->exit(policy);
1449                 policy->freq_table = NULL;
1450         }
1451 }
1452
1453 /**
1454  * cpufreq_remove_dev - remove a CPU device
1455  *
1456  * Removes the cpufreq interface for a CPU device.
1457  */
1458 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1459 {
1460         unsigned int cpu = dev->id;
1461         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1462
1463         if (!policy)
1464                 return;
1465
1466         if (cpu_online(cpu)) {
1467                 cpufreq_offline_prepare(cpu);
1468                 cpufreq_offline_finish(cpu);
1469         }
1470
1471         cpumask_clear_cpu(cpu, policy->real_cpus);
1472         remove_cpu_dev_symlink(policy, cpu);
1473
1474         if (cpumask_empty(policy->real_cpus))
1475                 cpufreq_policy_free(policy, true);
1476 }
1477
1478 static void handle_update(struct work_struct *work)
1479 {
1480         struct cpufreq_policy *policy =
1481                 container_of(work, struct cpufreq_policy, update);
1482         unsigned int cpu = policy->cpu;
1483         pr_debug("handle_update for cpu %u called\n", cpu);
1484         cpufreq_update_policy(cpu);
1485 }
1486
1487 /**
1488  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1489  *      in deep trouble.
1490  *      @policy: policy managing CPUs
1491  *      @new_freq: CPU frequency the CPU actually runs at
1492  *
1493  *      We adjust to current frequency first, and need to clean up later.
1494  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1495  */
1496 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1497                                 unsigned int new_freq)
1498 {
1499         struct cpufreq_freqs freqs;
1500
1501         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1502                  policy->cur, new_freq);
1503
1504         freqs.old = policy->cur;
1505         freqs.new = new_freq;
1506
1507         cpufreq_freq_transition_begin(policy, &freqs);
1508         cpufreq_freq_transition_end(policy, &freqs, 0);
1509 }
1510
1511 /**
1512  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1513  * @cpu: CPU number
1514  *
1515  * This is the last known freq, without actually getting it from the driver.
1516  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1517  */
1518 unsigned int cpufreq_quick_get(unsigned int cpu)
1519 {
1520         struct cpufreq_policy *policy;
1521         unsigned int ret_freq = 0;
1522
1523         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1524                 return cpufreq_driver->get(cpu);
1525
1526         policy = cpufreq_cpu_get(cpu);
1527         if (policy) {
1528                 ret_freq = policy->cur;
1529                 cpufreq_cpu_put(policy);
1530         }
1531
1532         return ret_freq;
1533 }
1534 EXPORT_SYMBOL(cpufreq_quick_get);
1535
1536 /**
1537  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1538  * @cpu: CPU number
1539  *
1540  * Just return the max possible frequency for a given CPU.
1541  */
1542 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1543 {
1544         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1545         unsigned int ret_freq = 0;
1546
1547         if (policy) {
1548                 ret_freq = policy->max;
1549                 cpufreq_cpu_put(policy);
1550         }
1551
1552         return ret_freq;
1553 }
1554 EXPORT_SYMBOL(cpufreq_quick_get_max);
1555
1556 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1557 {
1558         unsigned int ret_freq = 0;
1559
1560         if (!cpufreq_driver->get)
1561                 return ret_freq;
1562
1563         ret_freq = cpufreq_driver->get(policy->cpu);
1564
1565         /* Updating inactive policies is invalid, so avoid doing that. */
1566         if (unlikely(policy_is_inactive(policy)))
1567                 return ret_freq;
1568
1569         if (ret_freq && policy->cur &&
1570                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1571                 /* verify no discrepancy between actual and
1572                                         saved value exists */
1573                 if (unlikely(ret_freq != policy->cur)) {
1574                         cpufreq_out_of_sync(policy, ret_freq);
1575                         schedule_work(&policy->update);
1576                 }
1577         }
1578
1579         return ret_freq;
1580 }
1581
1582 /**
1583  * cpufreq_get - get the current CPU frequency (in kHz)
1584  * @cpu: CPU number
1585  *
1586  * Get the CPU current (static) CPU frequency
1587  */
1588 unsigned int cpufreq_get(unsigned int cpu)
1589 {
1590         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1591         unsigned int ret_freq = 0;
1592
1593         if (policy) {
1594                 down_read(&policy->rwsem);
1595                 ret_freq = __cpufreq_get(policy);
1596                 up_read(&policy->rwsem);
1597
1598                 cpufreq_cpu_put(policy);
1599         }
1600
1601         return ret_freq;
1602 }
1603 EXPORT_SYMBOL(cpufreq_get);
1604
1605 static struct subsys_interface cpufreq_interface = {
1606         .name           = "cpufreq",
1607         .subsys         = &cpu_subsys,
1608         .add_dev        = cpufreq_add_dev,
1609         .remove_dev     = cpufreq_remove_dev,
1610 };
1611
1612 /*
1613  * In case platform wants some specific frequency to be configured
1614  * during suspend..
1615  */
1616 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1617 {
1618         int ret;
1619
1620         if (!policy->suspend_freq) {
1621                 pr_debug("%s: suspend_freq not defined\n", __func__);
1622                 return 0;
1623         }
1624
1625         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1626                         policy->suspend_freq);
1627
1628         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1629                         CPUFREQ_RELATION_H);
1630         if (ret)
1631                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1632                                 __func__, policy->suspend_freq, ret);
1633
1634         return ret;
1635 }
1636 EXPORT_SYMBOL(cpufreq_generic_suspend);
1637
1638 /**
1639  * cpufreq_suspend() - Suspend CPUFreq governors
1640  *
1641  * Called during system wide Suspend/Hibernate cycles for suspending governors
1642  * as some platforms can't change frequency after this point in suspend cycle.
1643  * Because some of the devices (like: i2c, regulators, etc) they use for
1644  * changing frequency are suspended quickly after this point.
1645  */
1646 void cpufreq_suspend(void)
1647 {
1648         struct cpufreq_policy *policy;
1649
1650         if (!cpufreq_driver)
1651                 return;
1652
1653         if (!has_target())
1654                 goto suspend;
1655
1656         pr_debug("%s: Suspending Governors\n", __func__);
1657
1658         for_each_active_policy(policy) {
1659                 if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1660                         pr_err("%s: Failed to stop governor for policy: %p\n",
1661                                 __func__, policy);
1662                 else if (cpufreq_driver->suspend
1663                     && cpufreq_driver->suspend(policy))
1664                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1665                                 policy);
1666         }
1667
1668 suspend:
1669         cpufreq_suspended = true;
1670 }
1671
1672 /**
1673  * cpufreq_resume() - Resume CPUFreq governors
1674  *
1675  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1676  * are suspended with cpufreq_suspend().
1677  */
1678 void cpufreq_resume(void)
1679 {
1680         struct cpufreq_policy *policy;
1681
1682         if (!cpufreq_driver)
1683                 return;
1684
1685         cpufreq_suspended = false;
1686
1687         if (!has_target())
1688                 return;
1689
1690         pr_debug("%s: Resuming Governors\n", __func__);
1691
1692         for_each_active_policy(policy) {
1693                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1694                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1695                                 policy);
1696                 else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1697                     || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1698                         pr_err("%s: Failed to start governor for policy: %p\n",
1699                                 __func__, policy);
1700         }
1701
1702         /*
1703          * schedule call cpufreq_update_policy() for first-online CPU, as that
1704          * wouldn't be hotplugged-out on suspend. It will verify that the
1705          * current freq is in sync with what we believe it to be.
1706          */
1707         policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1708         if (WARN_ON(!policy))
1709                 return;
1710
1711         schedule_work(&policy->update);
1712 }
1713
1714 /**
1715  *      cpufreq_get_current_driver - return current driver's name
1716  *
1717  *      Return the name string of the currently loaded cpufreq driver
1718  *      or NULL, if none.
1719  */
1720 const char *cpufreq_get_current_driver(void)
1721 {
1722         if (cpufreq_driver)
1723                 return cpufreq_driver->name;
1724
1725         return NULL;
1726 }
1727 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1728
1729 /**
1730  *      cpufreq_get_driver_data - return current driver data
1731  *
1732  *      Return the private data of the currently loaded cpufreq
1733  *      driver, or NULL if no cpufreq driver is loaded.
1734  */
1735 void *cpufreq_get_driver_data(void)
1736 {
1737         if (cpufreq_driver)
1738                 return cpufreq_driver->driver_data;
1739
1740         return NULL;
1741 }
1742 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1743
1744 /*********************************************************************
1745  *                     NOTIFIER LISTS INTERFACE                      *
1746  *********************************************************************/
1747
1748 /**
1749  *      cpufreq_register_notifier - register a driver with cpufreq
1750  *      @nb: notifier function to register
1751  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1752  *
1753  *      Add a driver to one of two lists: either a list of drivers that
1754  *      are notified about clock rate changes (once before and once after
1755  *      the transition), or a list of drivers that are notified about
1756  *      changes in cpufreq policy.
1757  *
1758  *      This function may sleep, and has the same return conditions as
1759  *      blocking_notifier_chain_register.
1760  */
1761 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1762 {
1763         int ret;
1764
1765         if (cpufreq_disabled())
1766                 return -EINVAL;
1767
1768         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1769
1770         switch (list) {
1771         case CPUFREQ_TRANSITION_NOTIFIER:
1772                 ret = srcu_notifier_chain_register(
1773                                 &cpufreq_transition_notifier_list, nb);
1774                 break;
1775         case CPUFREQ_POLICY_NOTIFIER:
1776                 ret = blocking_notifier_chain_register(
1777                                 &cpufreq_policy_notifier_list, nb);
1778                 break;
1779         default:
1780                 ret = -EINVAL;
1781         }
1782
1783         return ret;
1784 }
1785 EXPORT_SYMBOL(cpufreq_register_notifier);
1786
1787 /**
1788  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1789  *      @nb: notifier block to be unregistered
1790  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1791  *
1792  *      Remove a driver from the CPU frequency notifier list.
1793  *
1794  *      This function may sleep, and has the same return conditions as
1795  *      blocking_notifier_chain_unregister.
1796  */
1797 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1798 {
1799         int ret;
1800
1801         if (cpufreq_disabled())
1802                 return -EINVAL;
1803
1804         switch (list) {
1805         case CPUFREQ_TRANSITION_NOTIFIER:
1806                 ret = srcu_notifier_chain_unregister(
1807                                 &cpufreq_transition_notifier_list, nb);
1808                 break;
1809         case CPUFREQ_POLICY_NOTIFIER:
1810                 ret = blocking_notifier_chain_unregister(
1811                                 &cpufreq_policy_notifier_list, nb);
1812                 break;
1813         default:
1814                 ret = -EINVAL;
1815         }
1816
1817         return ret;
1818 }
1819 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1820
1821
1822 /*********************************************************************
1823  *                              GOVERNORS                            *
1824  *********************************************************************/
1825
1826 /* Must set freqs->new to intermediate frequency */
1827 static int __target_intermediate(struct cpufreq_policy *policy,
1828                                  struct cpufreq_freqs *freqs, int index)
1829 {
1830         int ret;
1831
1832         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1833
1834         /* We don't need to switch to intermediate freq */
1835         if (!freqs->new)
1836                 return 0;
1837
1838         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1839                  __func__, policy->cpu, freqs->old, freqs->new);
1840
1841         cpufreq_freq_transition_begin(policy, freqs);
1842         ret = cpufreq_driver->target_intermediate(policy, index);
1843         cpufreq_freq_transition_end(policy, freqs, ret);
1844
1845         if (ret)
1846                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1847                        __func__, ret);
1848
1849         return ret;
1850 }
1851
1852 static int __target_index(struct cpufreq_policy *policy,
1853                           struct cpufreq_frequency_table *freq_table, int index)
1854 {
1855         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1856         unsigned int intermediate_freq = 0;
1857         int retval = -EINVAL;
1858         bool notify;
1859
1860         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1861         if (notify) {
1862                 /* Handle switching to intermediate frequency */
1863                 if (cpufreq_driver->get_intermediate) {
1864                         retval = __target_intermediate(policy, &freqs, index);
1865                         if (retval)
1866                                 return retval;
1867
1868                         intermediate_freq = freqs.new;
1869                         /* Set old freq to intermediate */
1870                         if (intermediate_freq)
1871                                 freqs.old = freqs.new;
1872                 }
1873
1874                 freqs.new = freq_table[index].frequency;
1875                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1876                          __func__, policy->cpu, freqs.old, freqs.new);
1877
1878                 cpufreq_freq_transition_begin(policy, &freqs);
1879         }
1880
1881         retval = cpufreq_driver->target_index(policy, index);
1882         if (retval)
1883                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1884                        retval);
1885
1886         if (notify) {
1887                 cpufreq_freq_transition_end(policy, &freqs, retval);
1888
1889                 /*
1890                  * Failed after setting to intermediate freq? Driver should have
1891                  * reverted back to initial frequency and so should we. Check
1892                  * here for intermediate_freq instead of get_intermediate, in
1893                  * case we haven't switched to intermediate freq at all.
1894                  */
1895                 if (unlikely(retval && intermediate_freq)) {
1896                         freqs.old = intermediate_freq;
1897                         freqs.new = policy->restore_freq;
1898                         cpufreq_freq_transition_begin(policy, &freqs);
1899                         cpufreq_freq_transition_end(policy, &freqs, 0);
1900                 }
1901         }
1902
1903         return retval;
1904 }
1905
1906 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1907                             unsigned int target_freq,
1908                             unsigned int relation)
1909 {
1910         unsigned int old_target_freq = target_freq;
1911         int retval = -EINVAL;
1912
1913         if (cpufreq_disabled())
1914                 return -ENODEV;
1915
1916         /* Make sure that target_freq is within supported range */
1917         if (target_freq > policy->max)
1918                 target_freq = policy->max;
1919         if (target_freq < policy->min)
1920                 target_freq = policy->min;
1921
1922         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1923                  policy->cpu, target_freq, relation, old_target_freq);
1924
1925         /*
1926          * This might look like a redundant call as we are checking it again
1927          * after finding index. But it is left intentionally for cases where
1928          * exactly same freq is called again and so we can save on few function
1929          * calls.
1930          */
1931         if (target_freq == policy->cur)
1932                 return 0;
1933
1934         /* Save last value to restore later on errors */
1935         policy->restore_freq = policy->cur;
1936
1937         if (cpufreq_driver->target)
1938                 retval = cpufreq_driver->target(policy, target_freq, relation);
1939         else if (cpufreq_driver->target_index) {
1940                 struct cpufreq_frequency_table *freq_table;
1941                 int index;
1942
1943                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1944                 if (unlikely(!freq_table)) {
1945                         pr_err("%s: Unable to find freq_table\n", __func__);
1946                         goto out;
1947                 }
1948
1949                 retval = cpufreq_frequency_table_target(policy, freq_table,
1950                                 target_freq, relation, &index);
1951                 if (unlikely(retval)) {
1952                         pr_err("%s: Unable to find matching freq\n", __func__);
1953                         goto out;
1954                 }
1955
1956                 if (freq_table[index].frequency == policy->cur) {
1957                         retval = 0;
1958                         goto out;
1959                 }
1960
1961                 retval = __target_index(policy, freq_table, index);
1962         }
1963
1964 out:
1965         return retval;
1966 }
1967 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1968
1969 int cpufreq_driver_target(struct cpufreq_policy *policy,
1970                           unsigned int target_freq,
1971                           unsigned int relation)
1972 {
1973         int ret = -EINVAL;
1974
1975         down_write(&policy->rwsem);
1976
1977         ret = __cpufreq_driver_target(policy, target_freq, relation);
1978
1979         up_write(&policy->rwsem);
1980
1981         return ret;
1982 }
1983 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1984
1985 static int __cpufreq_governor(struct cpufreq_policy *policy,
1986                                         unsigned int event)
1987 {
1988         int ret;
1989
1990         /* Only must be defined when default governor is known to have latency
1991            restrictions, like e.g. conservative or ondemand.
1992            That this is the case is already ensured in Kconfig
1993         */
1994 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1995         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1996 #else
1997         struct cpufreq_governor *gov = NULL;
1998 #endif
1999
2000         /* Don't start any governor operations if we are entering suspend */
2001         if (cpufreq_suspended)
2002                 return 0;
2003         /*
2004          * Governor might not be initiated here if ACPI _PPC changed
2005          * notification happened, so check it.
2006          */
2007         if (!policy->governor)
2008                 return -EINVAL;
2009
2010         if (policy->governor->max_transition_latency &&
2011             policy->cpuinfo.transition_latency >
2012             policy->governor->max_transition_latency) {
2013                 if (!gov)
2014                         return -EINVAL;
2015                 else {
2016                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2017                                 policy->governor->name, gov->name);
2018                         policy->governor = gov;
2019                 }
2020         }
2021
2022         if (event == CPUFREQ_GOV_POLICY_INIT)
2023                 if (!try_module_get(policy->governor->owner))
2024                         return -EINVAL;
2025
2026         pr_debug("%s: for CPU %u, event %u\n", __func__, policy->cpu, event);
2027
2028         mutex_lock(&cpufreq_governor_lock);
2029         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2030             || (!policy->governor_enabled
2031             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2032                 mutex_unlock(&cpufreq_governor_lock);
2033                 return -EBUSY;
2034         }
2035
2036         if (event == CPUFREQ_GOV_STOP)
2037                 policy->governor_enabled = false;
2038         else if (event == CPUFREQ_GOV_START)
2039                 policy->governor_enabled = true;
2040
2041         mutex_unlock(&cpufreq_governor_lock);
2042
2043         ret = policy->governor->governor(policy, event);
2044
2045         if (!ret) {
2046                 if (event == CPUFREQ_GOV_POLICY_INIT)
2047                         policy->governor->initialized++;
2048                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
2049                         policy->governor->initialized--;
2050         } else {
2051                 /* Restore original values */
2052                 mutex_lock(&cpufreq_governor_lock);
2053                 if (event == CPUFREQ_GOV_STOP)
2054                         policy->governor_enabled = true;
2055                 else if (event == CPUFREQ_GOV_START)
2056                         policy->governor_enabled = false;
2057                 mutex_unlock(&cpufreq_governor_lock);
2058         }
2059
2060         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2061                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2062                 module_put(policy->governor->owner);
2063
2064         return ret;
2065 }
2066
2067 int cpufreq_register_governor(struct cpufreq_governor *governor)
2068 {
2069         int err;
2070
2071         if (!governor)
2072                 return -EINVAL;
2073
2074         if (cpufreq_disabled())
2075                 return -ENODEV;
2076
2077         mutex_lock(&cpufreq_governor_mutex);
2078
2079         governor->initialized = 0;
2080         err = -EBUSY;
2081         if (!find_governor(governor->name)) {
2082                 err = 0;
2083                 list_add(&governor->governor_list, &cpufreq_governor_list);
2084         }
2085
2086         mutex_unlock(&cpufreq_governor_mutex);
2087         return err;
2088 }
2089 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2090
2091 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2092 {
2093         struct cpufreq_policy *policy;
2094         unsigned long flags;
2095
2096         if (!governor)
2097                 return;
2098
2099         if (cpufreq_disabled())
2100                 return;
2101
2102         /* clear last_governor for all inactive policies */
2103         read_lock_irqsave(&cpufreq_driver_lock, flags);
2104         for_each_inactive_policy(policy) {
2105                 if (!strcmp(policy->last_governor, governor->name)) {
2106                         policy->governor = NULL;
2107                         strcpy(policy->last_governor, "\0");
2108                 }
2109         }
2110         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2111
2112         mutex_lock(&cpufreq_governor_mutex);
2113         list_del(&governor->governor_list);
2114         mutex_unlock(&cpufreq_governor_mutex);
2115         return;
2116 }
2117 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2118
2119
2120 /*********************************************************************
2121  *                          POLICY INTERFACE                         *
2122  *********************************************************************/
2123
2124 /**
2125  * cpufreq_get_policy - get the current cpufreq_policy
2126  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2127  *      is written
2128  *
2129  * Reads the current cpufreq policy.
2130  */
2131 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2132 {
2133         struct cpufreq_policy *cpu_policy;
2134         if (!policy)
2135                 return -EINVAL;
2136
2137         cpu_policy = cpufreq_cpu_get(cpu);
2138         if (!cpu_policy)
2139                 return -EINVAL;
2140
2141         memcpy(policy, cpu_policy, sizeof(*policy));
2142
2143         cpufreq_cpu_put(cpu_policy);
2144         return 0;
2145 }
2146 EXPORT_SYMBOL(cpufreq_get_policy);
2147
2148 /*
2149  * policy : current policy.
2150  * new_policy: policy to be set.
2151  */
2152 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2153                                 struct cpufreq_policy *new_policy)
2154 {
2155         struct cpufreq_governor *old_gov;
2156         int ret;
2157
2158         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2159                  new_policy->cpu, new_policy->min, new_policy->max);
2160
2161         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2162
2163         /*
2164         * This check works well when we store new min/max freq attributes,
2165         * because new_policy is a copy of policy with one field updated.
2166         */
2167         if (new_policy->min > new_policy->max)
2168                 return -EINVAL;
2169
2170         /* verify the cpu speed can be set within this limit */
2171         ret = cpufreq_driver->verify(new_policy);
2172         if (ret)
2173                 return ret;
2174
2175         /* adjust if necessary - all reasons */
2176         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2177                         CPUFREQ_ADJUST, new_policy);
2178
2179         /*
2180          * verify the cpu speed can be set within this limit, which might be
2181          * different to the first one
2182          */
2183         ret = cpufreq_driver->verify(new_policy);
2184         if (ret)
2185                 return ret;
2186
2187         /* notification of the new policy */
2188         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2189                         CPUFREQ_NOTIFY, new_policy);
2190
2191         scale_freq_capacity(new_policy, NULL);
2192
2193         policy->min = new_policy->min;
2194         policy->max = new_policy->max;
2195         trace_cpu_frequency_limits(policy->max, policy->min, policy->cpu);
2196
2197         pr_debug("new min and max freqs are %u - %u kHz\n",
2198                  policy->min, policy->max);
2199
2200         if (cpufreq_driver->setpolicy) {
2201                 policy->policy = new_policy->policy;
2202                 pr_debug("setting range\n");
2203                 return cpufreq_driver->setpolicy(new_policy);
2204         }
2205
2206         if (new_policy->governor == policy->governor)
2207                 goto out;
2208
2209         pr_debug("governor switch\n");
2210
2211         /* save old, working values */
2212         old_gov = policy->governor;
2213         /* end old governor */
2214         if (old_gov) {
2215                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2216                 if (ret) {
2217                         /* This can happen due to race with other operations */
2218                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2219                                  __func__, old_gov->name, ret);
2220                         return ret;
2221                 }
2222
2223                 up_write(&policy->rwsem);
2224                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2225                 down_write(&policy->rwsem);
2226
2227                 if (ret) {
2228                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2229                                __func__, old_gov->name, ret);
2230                         return ret;
2231                 }
2232         }
2233
2234         /* start new governor */
2235         policy->governor = new_policy->governor;
2236         ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2237         if (!ret) {
2238                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
2239                 if (!ret)
2240                         goto out;
2241
2242                 up_write(&policy->rwsem);
2243                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2244                 down_write(&policy->rwsem);
2245         }
2246
2247         /* new governor failed, so re-start old one */
2248         pr_debug("starting governor %s failed\n", policy->governor->name);
2249         if (old_gov) {
2250                 policy->governor = old_gov;
2251                 if (__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2252                         policy->governor = NULL;
2253                 else
2254                         __cpufreq_governor(policy, CPUFREQ_GOV_START);
2255         }
2256
2257         return ret;
2258
2259  out:
2260         pr_debug("governor: change or update limits\n");
2261         return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2262 }
2263
2264 /**
2265  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2266  *      @cpu: CPU which shall be re-evaluated
2267  *
2268  *      Useful for policy notifiers which have different necessities
2269  *      at different times.
2270  */
2271 int cpufreq_update_policy(unsigned int cpu)
2272 {
2273         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2274         struct cpufreq_policy new_policy;
2275         int ret;
2276
2277         if (!policy)
2278                 return -ENODEV;
2279
2280         down_write(&policy->rwsem);
2281
2282         pr_debug("updating policy for CPU %u\n", cpu);
2283         memcpy(&new_policy, policy, sizeof(*policy));
2284         new_policy.min = policy->user_policy.min;
2285         new_policy.max = policy->user_policy.max;
2286
2287         /*
2288          * BIOS might change freq behind our back
2289          * -> ask driver for current freq and notify governors about a change
2290          */
2291         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2292                 new_policy.cur = cpufreq_driver->get(cpu);
2293                 if (WARN_ON(!new_policy.cur)) {
2294                         ret = -EIO;
2295                         goto unlock;
2296                 }
2297
2298                 if (!policy->cur) {
2299                         pr_debug("Driver did not initialize current freq\n");
2300                         policy->cur = new_policy.cur;
2301                 } else {
2302                         if (policy->cur != new_policy.cur && has_target())
2303                                 cpufreq_out_of_sync(policy, new_policy.cur);
2304                 }
2305         }
2306
2307         ret = cpufreq_set_policy(policy, &new_policy);
2308
2309 unlock:
2310         up_write(&policy->rwsem);
2311
2312         cpufreq_cpu_put(policy);
2313         return ret;
2314 }
2315 EXPORT_SYMBOL(cpufreq_update_policy);
2316
2317 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2318                                         unsigned long action, void *hcpu)
2319 {
2320         unsigned int cpu = (unsigned long)hcpu;
2321
2322         switch (action & ~CPU_TASKS_FROZEN) {
2323         case CPU_ONLINE:
2324                 cpufreq_online(cpu);
2325                 break;
2326
2327         case CPU_DOWN_PREPARE:
2328                 cpufreq_offline_prepare(cpu);
2329                 break;
2330
2331         case CPU_POST_DEAD:
2332                 cpufreq_offline_finish(cpu);
2333                 break;
2334
2335         case CPU_DOWN_FAILED:
2336                 cpufreq_online(cpu);
2337                 break;
2338         }
2339         return NOTIFY_OK;
2340 }
2341
2342 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2343         .notifier_call = cpufreq_cpu_callback,
2344 };
2345
2346 /*********************************************************************
2347  *               BOOST                                               *
2348  *********************************************************************/
2349 static int cpufreq_boost_set_sw(int state)
2350 {
2351         struct cpufreq_frequency_table *freq_table;
2352         struct cpufreq_policy *policy;
2353         int ret = -EINVAL;
2354
2355         for_each_active_policy(policy) {
2356                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2357                 if (freq_table) {
2358                         ret = cpufreq_frequency_table_cpuinfo(policy,
2359                                                         freq_table);
2360                         if (ret) {
2361                                 pr_err("%s: Policy frequency update failed\n",
2362                                        __func__);
2363                                 break;
2364                         }
2365                         policy->user_policy.max = policy->max;
2366                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2367                 }
2368         }
2369
2370         return ret;
2371 }
2372
2373 int cpufreq_boost_trigger_state(int state)
2374 {
2375         unsigned long flags;
2376         int ret = 0;
2377
2378         if (cpufreq_driver->boost_enabled == state)
2379                 return 0;
2380
2381         write_lock_irqsave(&cpufreq_driver_lock, flags);
2382         cpufreq_driver->boost_enabled = state;
2383         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2384
2385         ret = cpufreq_driver->set_boost(state);
2386         if (ret) {
2387                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2388                 cpufreq_driver->boost_enabled = !state;
2389                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2390
2391                 pr_err("%s: Cannot %s BOOST\n",
2392                        __func__, state ? "enable" : "disable");
2393         }
2394
2395         return ret;
2396 }
2397
2398 int cpufreq_boost_supported(void)
2399 {
2400         if (likely(cpufreq_driver))
2401                 return cpufreq_driver->boost_supported;
2402
2403         return 0;
2404 }
2405 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2406
2407 static int create_boost_sysfs_file(void)
2408 {
2409         int ret;
2410
2411         if (!cpufreq_boost_supported())
2412                 return 0;
2413
2414         /*
2415          * Check if driver provides function to enable boost -
2416          * if not, use cpufreq_boost_set_sw as default
2417          */
2418         if (!cpufreq_driver->set_boost)
2419                 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2420
2421         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2422         if (ret)
2423                 pr_err("%s: cannot register global BOOST sysfs file\n",
2424                        __func__);
2425
2426         return ret;
2427 }
2428
2429 static void remove_boost_sysfs_file(void)
2430 {
2431         if (cpufreq_boost_supported())
2432                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2433 }
2434
2435 int cpufreq_enable_boost_support(void)
2436 {
2437         if (!cpufreq_driver)
2438                 return -EINVAL;
2439
2440         if (cpufreq_boost_supported())
2441                 return 0;
2442
2443         cpufreq_driver->boost_supported = true;
2444
2445         /* This will get removed on driver unregister */
2446         return create_boost_sysfs_file();
2447 }
2448 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2449
2450 int cpufreq_boost_enabled(void)
2451 {
2452         return cpufreq_driver->boost_enabled;
2453 }
2454 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2455
2456 /*********************************************************************
2457  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2458  *********************************************************************/
2459
2460 /**
2461  * cpufreq_register_driver - register a CPU Frequency driver
2462  * @driver_data: A struct cpufreq_driver containing the values#
2463  * submitted by the CPU Frequency driver.
2464  *
2465  * Registers a CPU Frequency driver to this core code. This code
2466  * returns zero on success, -EBUSY when another driver got here first
2467  * (and isn't unregistered in the meantime).
2468  *
2469  */
2470 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2471 {
2472         unsigned long flags;
2473         int ret;
2474
2475         if (cpufreq_disabled())
2476                 return -ENODEV;
2477
2478         if (!driver_data || !driver_data->verify || !driver_data->init ||
2479             !(driver_data->setpolicy || driver_data->target_index ||
2480                     driver_data->target) ||
2481              (driver_data->setpolicy && (driver_data->target_index ||
2482                     driver_data->target)) ||
2483              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2484                 return -EINVAL;
2485
2486         pr_debug("trying to register driver %s\n", driver_data->name);
2487
2488         /* Protect against concurrent CPU online/offline. */
2489         get_online_cpus();
2490
2491         write_lock_irqsave(&cpufreq_driver_lock, flags);
2492         if (cpufreq_driver) {
2493                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2494                 ret = -EEXIST;
2495                 goto out;
2496         }
2497         cpufreq_driver = driver_data;
2498         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2499
2500         if (driver_data->setpolicy)
2501                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2502
2503         ret = create_boost_sysfs_file();
2504         if (ret)
2505                 goto err_null_driver;
2506
2507         ret = subsys_interface_register(&cpufreq_interface);
2508         if (ret)
2509                 goto err_boost_unreg;
2510
2511         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2512             list_empty(&cpufreq_policy_list)) {
2513                 /* if all ->init() calls failed, unregister */
2514                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2515                          driver_data->name);
2516                 goto err_if_unreg;
2517         }
2518
2519         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2520         pr_debug("driver %s up and running\n", driver_data->name);
2521
2522 out:
2523         put_online_cpus();
2524         return ret;
2525
2526 err_if_unreg:
2527         subsys_interface_unregister(&cpufreq_interface);
2528 err_boost_unreg:
2529         remove_boost_sysfs_file();
2530 err_null_driver:
2531         write_lock_irqsave(&cpufreq_driver_lock, flags);
2532         cpufreq_driver = NULL;
2533         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2534         goto out;
2535 }
2536 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2537
2538 /**
2539  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2540  *
2541  * Unregister the current CPUFreq driver. Only call this if you have
2542  * the right to do so, i.e. if you have succeeded in initialising before!
2543  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2544  * currently not initialised.
2545  */
2546 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2547 {
2548         unsigned long flags;
2549
2550         if (!cpufreq_driver || (driver != cpufreq_driver))
2551                 return -EINVAL;
2552
2553         pr_debug("unregistering driver %s\n", driver->name);
2554
2555         /* Protect against concurrent cpu hotplug */
2556         get_online_cpus();
2557         subsys_interface_unregister(&cpufreq_interface);
2558         remove_boost_sysfs_file();
2559         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2560
2561         write_lock_irqsave(&cpufreq_driver_lock, flags);
2562
2563         cpufreq_driver = NULL;
2564
2565         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2566         put_online_cpus();
2567
2568         return 0;
2569 }
2570 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2571
2572 /*
2573  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2574  * or mutexes when secondary CPUs are halted.
2575  */
2576 static struct syscore_ops cpufreq_syscore_ops = {
2577         .shutdown = cpufreq_suspend,
2578 };
2579
2580 struct kobject *cpufreq_global_kobject;
2581 EXPORT_SYMBOL(cpufreq_global_kobject);
2582
2583 static int __init cpufreq_core_init(void)
2584 {
2585         if (cpufreq_disabled())
2586                 return -ENODEV;
2587
2588         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2589         BUG_ON(!cpufreq_global_kobject);
2590
2591         register_syscore_ops(&cpufreq_syscore_ops);
2592
2593         return 0;
2594 }
2595 core_initcall(cpufreq_core_init);