Merge branch 'pm-cpufreq'
[linux-2.6-block.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 #include <trace/events/power.h>
33
34 static LIST_HEAD(cpufreq_policy_list);
35
36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
37 {
38         return cpumask_empty(policy->cpus);
39 }
40
41 /* Macros to iterate over CPU policies */
42 #define for_each_suitable_policy(__policy, __active)                     \
43         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
44                 if ((__active) == !policy_is_inactive(__policy))
45
46 #define for_each_active_policy(__policy)                \
47         for_each_suitable_policy(__policy, true)
48 #define for_each_inactive_policy(__policy)              \
49         for_each_suitable_policy(__policy, false)
50
51 #define for_each_policy(__policy)                       \
52         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
53
54 /* Iterate over governors */
55 static LIST_HEAD(cpufreq_governor_list);
56 #define for_each_governor(__governor)                           \
57         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
58
59 /**
60  * The "cpufreq driver" - the arch- or hardware-dependent low
61  * level driver of CPUFreq support, and its spinlock. This lock
62  * also protects the cpufreq_cpu_data array.
63  */
64 static struct cpufreq_driver *cpufreq_driver;
65 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
66 static DEFINE_RWLOCK(cpufreq_driver_lock);
67
68 /* Flag to suspend/resume CPUFreq governors */
69 static bool cpufreq_suspended;
70
71 static inline bool has_target(void)
72 {
73         return cpufreq_driver->target_index || cpufreq_driver->target;
74 }
75
76 /* internal prototypes */
77 static int cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
78 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
79 static int cpufreq_start_governor(struct cpufreq_policy *policy);
80
81 static inline int cpufreq_exit_governor(struct cpufreq_policy *policy)
82 {
83         return cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
84 }
85
86 /**
87  * Two notifier lists: the "policy" list is involved in the
88  * validation process for a new CPU frequency policy; the
89  * "transition" list for kernel code that needs to handle
90  * changes to devices when the CPU clock speed changes.
91  * The mutex locks both lists.
92  */
93 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
94 static struct srcu_notifier_head cpufreq_transition_notifier_list;
95
96 static bool init_cpufreq_transition_notifier_list_called;
97 static int __init init_cpufreq_transition_notifier_list(void)
98 {
99         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
100         init_cpufreq_transition_notifier_list_called = true;
101         return 0;
102 }
103 pure_initcall(init_cpufreq_transition_notifier_list);
104
105 static int off __read_mostly;
106 static int cpufreq_disabled(void)
107 {
108         return off;
109 }
110 void disable_cpufreq(void)
111 {
112         off = 1;
113 }
114 static DEFINE_MUTEX(cpufreq_governor_mutex);
115
116 bool have_governor_per_policy(void)
117 {
118         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
119 }
120 EXPORT_SYMBOL_GPL(have_governor_per_policy);
121
122 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
123 {
124         if (have_governor_per_policy())
125                 return &policy->kobj;
126         else
127                 return cpufreq_global_kobject;
128 }
129 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
130
131 struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu)
132 {
133         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
134
135         return policy && !policy_is_inactive(policy) ?
136                 policy->freq_table : NULL;
137 }
138 EXPORT_SYMBOL_GPL(cpufreq_frequency_get_table);
139
140 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
141 {
142         u64 idle_time;
143         u64 cur_wall_time;
144         u64 busy_time;
145
146         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
147
148         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
149         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
150         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
151         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
152         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
153         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
154
155         idle_time = cur_wall_time - busy_time;
156         if (wall)
157                 *wall = cputime_to_usecs(cur_wall_time);
158
159         return cputime_to_usecs(idle_time);
160 }
161
162 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
163 {
164         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
165
166         if (idle_time == -1ULL)
167                 return get_cpu_idle_time_jiffy(cpu, wall);
168         else if (!io_busy)
169                 idle_time += get_cpu_iowait_time_us(cpu, wall);
170
171         return idle_time;
172 }
173 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
174
175 /*
176  * This is a generic cpufreq init() routine which can be used by cpufreq
177  * drivers of SMP systems. It will do following:
178  * - validate & show freq table passed
179  * - set policies transition latency
180  * - policy->cpus with all possible CPUs
181  */
182 int cpufreq_generic_init(struct cpufreq_policy *policy,
183                 struct cpufreq_frequency_table *table,
184                 unsigned int transition_latency)
185 {
186         int ret;
187
188         ret = cpufreq_table_validate_and_show(policy, table);
189         if (ret) {
190                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
191                 return ret;
192         }
193
194         policy->cpuinfo.transition_latency = transition_latency;
195
196         /*
197          * The driver only supports the SMP configuration where all processors
198          * share the clock and voltage and clock.
199          */
200         cpumask_setall(policy->cpus);
201
202         return 0;
203 }
204 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
205
206 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
207 {
208         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
209
210         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
211 }
212 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
213
214 unsigned int cpufreq_generic_get(unsigned int cpu)
215 {
216         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
217
218         if (!policy || IS_ERR(policy->clk)) {
219                 pr_err("%s: No %s associated to cpu: %d\n",
220                        __func__, policy ? "clk" : "policy", cpu);
221                 return 0;
222         }
223
224         return clk_get_rate(policy->clk) / 1000;
225 }
226 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
227
228 /**
229  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
230  *
231  * @cpu: cpu to find policy for.
232  *
233  * This returns policy for 'cpu', returns NULL if it doesn't exist.
234  * It also increments the kobject reference count to mark it busy and so would
235  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
236  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
237  * freed as that depends on the kobj count.
238  *
239  * Return: A valid policy on success, otherwise NULL on failure.
240  */
241 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
242 {
243         struct cpufreq_policy *policy = NULL;
244         unsigned long flags;
245
246         if (WARN_ON(cpu >= nr_cpu_ids))
247                 return NULL;
248
249         /* get the cpufreq driver */
250         read_lock_irqsave(&cpufreq_driver_lock, flags);
251
252         if (cpufreq_driver) {
253                 /* get the CPU */
254                 policy = cpufreq_cpu_get_raw(cpu);
255                 if (policy)
256                         kobject_get(&policy->kobj);
257         }
258
259         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
260
261         return policy;
262 }
263 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
264
265 /**
266  * cpufreq_cpu_put: Decrements the usage count of a policy
267  *
268  * @policy: policy earlier returned by cpufreq_cpu_get().
269  *
270  * This decrements the kobject reference count incremented earlier by calling
271  * cpufreq_cpu_get().
272  */
273 void cpufreq_cpu_put(struct cpufreq_policy *policy)
274 {
275         kobject_put(&policy->kobj);
276 }
277 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
278
279 /*********************************************************************
280  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
281  *********************************************************************/
282
283 /**
284  * adjust_jiffies - adjust the system "loops_per_jiffy"
285  *
286  * This function alters the system "loops_per_jiffy" for the clock
287  * speed change. Note that loops_per_jiffy cannot be updated on SMP
288  * systems as each CPU might be scaled differently. So, use the arch
289  * per-CPU loops_per_jiffy value wherever possible.
290  */
291 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
292 {
293 #ifndef CONFIG_SMP
294         static unsigned long l_p_j_ref;
295         static unsigned int l_p_j_ref_freq;
296
297         if (ci->flags & CPUFREQ_CONST_LOOPS)
298                 return;
299
300         if (!l_p_j_ref_freq) {
301                 l_p_j_ref = loops_per_jiffy;
302                 l_p_j_ref_freq = ci->old;
303                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
304                          l_p_j_ref, l_p_j_ref_freq);
305         }
306         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
307                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
308                                                                 ci->new);
309                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
310                          loops_per_jiffy, ci->new);
311         }
312 #endif
313 }
314
315 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
316                 struct cpufreq_freqs *freqs, unsigned int state)
317 {
318         BUG_ON(irqs_disabled());
319
320         if (cpufreq_disabled())
321                 return;
322
323         freqs->flags = cpufreq_driver->flags;
324         pr_debug("notification %u of frequency transition to %u kHz\n",
325                  state, freqs->new);
326
327         switch (state) {
328
329         case CPUFREQ_PRECHANGE:
330                 /* detect if the driver reported a value as "old frequency"
331                  * which is not equal to what the cpufreq core thinks is
332                  * "old frequency".
333                  */
334                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
335                         if ((policy) && (policy->cpu == freqs->cpu) &&
336                             (policy->cur) && (policy->cur != freqs->old)) {
337                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
338                                          freqs->old, policy->cur);
339                                 freqs->old = policy->cur;
340                         }
341                 }
342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
343                                 CPUFREQ_PRECHANGE, freqs);
344                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
345                 break;
346
347         case CPUFREQ_POSTCHANGE:
348                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
349                 pr_debug("FREQ: %lu - CPU: %lu\n",
350                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
351                 trace_cpu_frequency(freqs->new, freqs->cpu);
352                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
353                                 CPUFREQ_POSTCHANGE, freqs);
354                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
355                         policy->cur = freqs->new;
356                 break;
357         }
358 }
359
360 /**
361  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
362  * on frequency transition.
363  *
364  * This function calls the transition notifiers and the "adjust_jiffies"
365  * function. It is called twice on all CPU frequency changes that have
366  * external effects.
367  */
368 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
369                 struct cpufreq_freqs *freqs, unsigned int state)
370 {
371         for_each_cpu(freqs->cpu, policy->cpus)
372                 __cpufreq_notify_transition(policy, freqs, state);
373 }
374
375 /* Do post notifications when there are chances that transition has failed */
376 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
377                 struct cpufreq_freqs *freqs, int transition_failed)
378 {
379         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
380         if (!transition_failed)
381                 return;
382
383         swap(freqs->old, freqs->new);
384         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
385         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
386 }
387
388 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
389                 struct cpufreq_freqs *freqs)
390 {
391
392         /*
393          * Catch double invocations of _begin() which lead to self-deadlock.
394          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
395          * doesn't invoke _begin() on their behalf, and hence the chances of
396          * double invocations are very low. Moreover, there are scenarios
397          * where these checks can emit false-positive warnings in these
398          * drivers; so we avoid that by skipping them altogether.
399          */
400         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
401                                 && current == policy->transition_task);
402
403 wait:
404         wait_event(policy->transition_wait, !policy->transition_ongoing);
405
406         spin_lock(&policy->transition_lock);
407
408         if (unlikely(policy->transition_ongoing)) {
409                 spin_unlock(&policy->transition_lock);
410                 goto wait;
411         }
412
413         policy->transition_ongoing = true;
414         policy->transition_task = current;
415
416         spin_unlock(&policy->transition_lock);
417
418         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
419 }
420 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
421
422 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
423                 struct cpufreq_freqs *freqs, int transition_failed)
424 {
425         if (unlikely(WARN_ON(!policy->transition_ongoing)))
426                 return;
427
428         cpufreq_notify_post_transition(policy, freqs, transition_failed);
429
430         policy->transition_ongoing = false;
431         policy->transition_task = NULL;
432
433         wake_up(&policy->transition_wait);
434 }
435 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
436
437 /*
438  * Fast frequency switching status count.  Positive means "enabled", negative
439  * means "disabled" and 0 means "not decided yet".
440  */
441 static int cpufreq_fast_switch_count;
442 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
443
444 static void cpufreq_list_transition_notifiers(void)
445 {
446         struct notifier_block *nb;
447
448         pr_info("Registered transition notifiers:\n");
449
450         mutex_lock(&cpufreq_transition_notifier_list.mutex);
451
452         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
453                 pr_info("%pF\n", nb->notifier_call);
454
455         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
456 }
457
458 /**
459  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
460  * @policy: cpufreq policy to enable fast frequency switching for.
461  *
462  * Try to enable fast frequency switching for @policy.
463  *
464  * The attempt will fail if there is at least one transition notifier registered
465  * at this point, as fast frequency switching is quite fundamentally at odds
466  * with transition notifiers.  Thus if successful, it will make registration of
467  * transition notifiers fail going forward.
468  */
469 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
470 {
471         lockdep_assert_held(&policy->rwsem);
472
473         if (!policy->fast_switch_possible)
474                 return;
475
476         mutex_lock(&cpufreq_fast_switch_lock);
477         if (cpufreq_fast_switch_count >= 0) {
478                 cpufreq_fast_switch_count++;
479                 policy->fast_switch_enabled = true;
480         } else {
481                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
482                         policy->cpu);
483                 cpufreq_list_transition_notifiers();
484         }
485         mutex_unlock(&cpufreq_fast_switch_lock);
486 }
487 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
488
489 /**
490  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
491  * @policy: cpufreq policy to disable fast frequency switching for.
492  */
493 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
494 {
495         mutex_lock(&cpufreq_fast_switch_lock);
496         if (policy->fast_switch_enabled) {
497                 policy->fast_switch_enabled = false;
498                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
499                         cpufreq_fast_switch_count--;
500         }
501         mutex_unlock(&cpufreq_fast_switch_lock);
502 }
503 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
504
505 /*********************************************************************
506  *                          SYSFS INTERFACE                          *
507  *********************************************************************/
508 static ssize_t show_boost(struct kobject *kobj,
509                                  struct attribute *attr, char *buf)
510 {
511         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
512 }
513
514 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
515                                   const char *buf, size_t count)
516 {
517         int ret, enable;
518
519         ret = sscanf(buf, "%d", &enable);
520         if (ret != 1 || enable < 0 || enable > 1)
521                 return -EINVAL;
522
523         if (cpufreq_boost_trigger_state(enable)) {
524                 pr_err("%s: Cannot %s BOOST!\n",
525                        __func__, enable ? "enable" : "disable");
526                 return -EINVAL;
527         }
528
529         pr_debug("%s: cpufreq BOOST %s\n",
530                  __func__, enable ? "enabled" : "disabled");
531
532         return count;
533 }
534 define_one_global_rw(boost);
535
536 static struct cpufreq_governor *find_governor(const char *str_governor)
537 {
538         struct cpufreq_governor *t;
539
540         for_each_governor(t)
541                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
542                         return t;
543
544         return NULL;
545 }
546
547 /**
548  * cpufreq_parse_governor - parse a governor string
549  */
550 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
551                                 struct cpufreq_governor **governor)
552 {
553         int err = -EINVAL;
554
555         if (cpufreq_driver->setpolicy) {
556                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
557                         *policy = CPUFREQ_POLICY_PERFORMANCE;
558                         err = 0;
559                 } else if (!strncasecmp(str_governor, "powersave",
560                                                 CPUFREQ_NAME_LEN)) {
561                         *policy = CPUFREQ_POLICY_POWERSAVE;
562                         err = 0;
563                 }
564         } else {
565                 struct cpufreq_governor *t;
566
567                 mutex_lock(&cpufreq_governor_mutex);
568
569                 t = find_governor(str_governor);
570
571                 if (t == NULL) {
572                         int ret;
573
574                         mutex_unlock(&cpufreq_governor_mutex);
575                         ret = request_module("cpufreq_%s", str_governor);
576                         mutex_lock(&cpufreq_governor_mutex);
577
578                         if (ret == 0)
579                                 t = find_governor(str_governor);
580                 }
581
582                 if (t != NULL) {
583                         *governor = t;
584                         err = 0;
585                 }
586
587                 mutex_unlock(&cpufreq_governor_mutex);
588         }
589         return err;
590 }
591
592 /**
593  * cpufreq_per_cpu_attr_read() / show_##file_name() -
594  * print out cpufreq information
595  *
596  * Write out information from cpufreq_driver->policy[cpu]; object must be
597  * "unsigned int".
598  */
599
600 #define show_one(file_name, object)                     \
601 static ssize_t show_##file_name                         \
602 (struct cpufreq_policy *policy, char *buf)              \
603 {                                                       \
604         return sprintf(buf, "%u\n", policy->object);    \
605 }
606
607 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
608 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
609 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
610 show_one(scaling_min_freq, min);
611 show_one(scaling_max_freq, max);
612
613 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
614 {
615         ssize_t ret;
616
617         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
618                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
619         else
620                 ret = sprintf(buf, "%u\n", policy->cur);
621         return ret;
622 }
623
624 static int cpufreq_set_policy(struct cpufreq_policy *policy,
625                                 struct cpufreq_policy *new_policy);
626
627 /**
628  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
629  */
630 #define store_one(file_name, object)                    \
631 static ssize_t store_##file_name                                        \
632 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
633 {                                                                       \
634         int ret, temp;                                                  \
635         struct cpufreq_policy new_policy;                               \
636                                                                         \
637         memcpy(&new_policy, policy, sizeof(*policy));                   \
638                                                                         \
639         ret = sscanf(buf, "%u", &new_policy.object);                    \
640         if (ret != 1)                                                   \
641                 return -EINVAL;                                         \
642                                                                         \
643         temp = new_policy.object;                                       \
644         ret = cpufreq_set_policy(policy, &new_policy);          \
645         if (!ret)                                                       \
646                 policy->user_policy.object = temp;                      \
647                                                                         \
648         return ret ? ret : count;                                       \
649 }
650
651 store_one(scaling_min_freq, min);
652 store_one(scaling_max_freq, max);
653
654 /**
655  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
656  */
657 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
658                                         char *buf)
659 {
660         unsigned int cur_freq = __cpufreq_get(policy);
661         if (!cur_freq)
662                 return sprintf(buf, "<unknown>");
663         return sprintf(buf, "%u\n", cur_freq);
664 }
665
666 /**
667  * show_scaling_governor - show the current policy for the specified CPU
668  */
669 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
670 {
671         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
672                 return sprintf(buf, "powersave\n");
673         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
674                 return sprintf(buf, "performance\n");
675         else if (policy->governor)
676                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
677                                 policy->governor->name);
678         return -EINVAL;
679 }
680
681 /**
682  * store_scaling_governor - store policy for the specified CPU
683  */
684 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
685                                         const char *buf, size_t count)
686 {
687         int ret;
688         char    str_governor[16];
689         struct cpufreq_policy new_policy;
690
691         memcpy(&new_policy, policy, sizeof(*policy));
692
693         ret = sscanf(buf, "%15s", str_governor);
694         if (ret != 1)
695                 return -EINVAL;
696
697         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
698                                                 &new_policy.governor))
699                 return -EINVAL;
700
701         ret = cpufreq_set_policy(policy, &new_policy);
702         return ret ? ret : count;
703 }
704
705 /**
706  * show_scaling_driver - show the cpufreq driver currently loaded
707  */
708 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
709 {
710         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
711 }
712
713 /**
714  * show_scaling_available_governors - show the available CPUfreq governors
715  */
716 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
717                                                 char *buf)
718 {
719         ssize_t i = 0;
720         struct cpufreq_governor *t;
721
722         if (!has_target()) {
723                 i += sprintf(buf, "performance powersave");
724                 goto out;
725         }
726
727         for_each_governor(t) {
728                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
729                     - (CPUFREQ_NAME_LEN + 2)))
730                         goto out;
731                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
732         }
733 out:
734         i += sprintf(&buf[i], "\n");
735         return i;
736 }
737
738 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
739 {
740         ssize_t i = 0;
741         unsigned int cpu;
742
743         for_each_cpu(cpu, mask) {
744                 if (i)
745                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
746                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
747                 if (i >= (PAGE_SIZE - 5))
748                         break;
749         }
750         i += sprintf(&buf[i], "\n");
751         return i;
752 }
753 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
754
755 /**
756  * show_related_cpus - show the CPUs affected by each transition even if
757  * hw coordination is in use
758  */
759 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
760 {
761         return cpufreq_show_cpus(policy->related_cpus, buf);
762 }
763
764 /**
765  * show_affected_cpus - show the CPUs affected by each transition
766  */
767 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
768 {
769         return cpufreq_show_cpus(policy->cpus, buf);
770 }
771
772 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
773                                         const char *buf, size_t count)
774 {
775         unsigned int freq = 0;
776         unsigned int ret;
777
778         if (!policy->governor || !policy->governor->store_setspeed)
779                 return -EINVAL;
780
781         ret = sscanf(buf, "%u", &freq);
782         if (ret != 1)
783                 return -EINVAL;
784
785         policy->governor->store_setspeed(policy, freq);
786
787         return count;
788 }
789
790 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
791 {
792         if (!policy->governor || !policy->governor->show_setspeed)
793                 return sprintf(buf, "<unsupported>\n");
794
795         return policy->governor->show_setspeed(policy, buf);
796 }
797
798 /**
799  * show_bios_limit - show the current cpufreq HW/BIOS limitation
800  */
801 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
802 {
803         unsigned int limit;
804         int ret;
805         if (cpufreq_driver->bios_limit) {
806                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
807                 if (!ret)
808                         return sprintf(buf, "%u\n", limit);
809         }
810         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
811 }
812
813 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
814 cpufreq_freq_attr_ro(cpuinfo_min_freq);
815 cpufreq_freq_attr_ro(cpuinfo_max_freq);
816 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
817 cpufreq_freq_attr_ro(scaling_available_governors);
818 cpufreq_freq_attr_ro(scaling_driver);
819 cpufreq_freq_attr_ro(scaling_cur_freq);
820 cpufreq_freq_attr_ro(bios_limit);
821 cpufreq_freq_attr_ro(related_cpus);
822 cpufreq_freq_attr_ro(affected_cpus);
823 cpufreq_freq_attr_rw(scaling_min_freq);
824 cpufreq_freq_attr_rw(scaling_max_freq);
825 cpufreq_freq_attr_rw(scaling_governor);
826 cpufreq_freq_attr_rw(scaling_setspeed);
827
828 static struct attribute *default_attrs[] = {
829         &cpuinfo_min_freq.attr,
830         &cpuinfo_max_freq.attr,
831         &cpuinfo_transition_latency.attr,
832         &scaling_min_freq.attr,
833         &scaling_max_freq.attr,
834         &affected_cpus.attr,
835         &related_cpus.attr,
836         &scaling_governor.attr,
837         &scaling_driver.attr,
838         &scaling_available_governors.attr,
839         &scaling_setspeed.attr,
840         NULL
841 };
842
843 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
844 #define to_attr(a) container_of(a, struct freq_attr, attr)
845
846 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
847 {
848         struct cpufreq_policy *policy = to_policy(kobj);
849         struct freq_attr *fattr = to_attr(attr);
850         ssize_t ret;
851
852         down_read(&policy->rwsem);
853         ret = fattr->show(policy, buf);
854         up_read(&policy->rwsem);
855
856         return ret;
857 }
858
859 static ssize_t store(struct kobject *kobj, struct attribute *attr,
860                      const char *buf, size_t count)
861 {
862         struct cpufreq_policy *policy = to_policy(kobj);
863         struct freq_attr *fattr = to_attr(attr);
864         ssize_t ret = -EINVAL;
865
866         get_online_cpus();
867
868         if (cpu_online(policy->cpu)) {
869                 down_write(&policy->rwsem);
870                 ret = fattr->store(policy, buf, count);
871                 up_write(&policy->rwsem);
872         }
873
874         put_online_cpus();
875
876         return ret;
877 }
878
879 static void cpufreq_sysfs_release(struct kobject *kobj)
880 {
881         struct cpufreq_policy *policy = to_policy(kobj);
882         pr_debug("last reference is dropped\n");
883         complete(&policy->kobj_unregister);
884 }
885
886 static const struct sysfs_ops sysfs_ops = {
887         .show   = show,
888         .store  = store,
889 };
890
891 static struct kobj_type ktype_cpufreq = {
892         .sysfs_ops      = &sysfs_ops,
893         .default_attrs  = default_attrs,
894         .release        = cpufreq_sysfs_release,
895 };
896
897 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
898 {
899         struct device *cpu_dev;
900
901         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
902
903         if (!policy)
904                 return 0;
905
906         cpu_dev = get_cpu_device(cpu);
907         if (WARN_ON(!cpu_dev))
908                 return 0;
909
910         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
911 }
912
913 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
914 {
915         struct device *cpu_dev;
916
917         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
918
919         cpu_dev = get_cpu_device(cpu);
920         if (WARN_ON(!cpu_dev))
921                 return;
922
923         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
924 }
925
926 /* Add/remove symlinks for all related CPUs */
927 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
928 {
929         unsigned int j;
930         int ret = 0;
931
932         /* Some related CPUs might not be present (physically hotplugged) */
933         for_each_cpu(j, policy->real_cpus) {
934                 ret = add_cpu_dev_symlink(policy, j);
935                 if (ret)
936                         break;
937         }
938
939         return ret;
940 }
941
942 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
943 {
944         unsigned int j;
945
946         /* Some related CPUs might not be present (physically hotplugged) */
947         for_each_cpu(j, policy->real_cpus)
948                 remove_cpu_dev_symlink(policy, j);
949 }
950
951 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
952 {
953         struct freq_attr **drv_attr;
954         int ret = 0;
955
956         /* set up files for this cpu device */
957         drv_attr = cpufreq_driver->attr;
958         while (drv_attr && *drv_attr) {
959                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
960                 if (ret)
961                         return ret;
962                 drv_attr++;
963         }
964         if (cpufreq_driver->get) {
965                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
966                 if (ret)
967                         return ret;
968         }
969
970         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
971         if (ret)
972                 return ret;
973
974         if (cpufreq_driver->bios_limit) {
975                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
976                 if (ret)
977                         return ret;
978         }
979
980         return cpufreq_add_dev_symlink(policy);
981 }
982
983 __weak struct cpufreq_governor *cpufreq_default_governor(void)
984 {
985         return NULL;
986 }
987
988 static int cpufreq_init_policy(struct cpufreq_policy *policy)
989 {
990         struct cpufreq_governor *gov = NULL;
991         struct cpufreq_policy new_policy;
992
993         memcpy(&new_policy, policy, sizeof(*policy));
994
995         /* Update governor of new_policy to the governor used before hotplug */
996         gov = find_governor(policy->last_governor);
997         if (gov) {
998                 pr_debug("Restoring governor %s for cpu %d\n",
999                                 policy->governor->name, policy->cpu);
1000         } else {
1001                 gov = cpufreq_default_governor();
1002                 if (!gov)
1003                         return -ENODATA;
1004         }
1005
1006         new_policy.governor = gov;
1007
1008         /* Use the default policy if there is no last_policy. */
1009         if (cpufreq_driver->setpolicy) {
1010                 if (policy->last_policy)
1011                         new_policy.policy = policy->last_policy;
1012                 else
1013                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1014                                                NULL);
1015         }
1016         /* set default policy */
1017         return cpufreq_set_policy(policy, &new_policy);
1018 }
1019
1020 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1021 {
1022         int ret = 0;
1023
1024         /* Has this CPU been taken care of already? */
1025         if (cpumask_test_cpu(cpu, policy->cpus))
1026                 return 0;
1027
1028         down_write(&policy->rwsem);
1029         if (has_target()) {
1030                 ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1031                 if (ret) {
1032                         pr_err("%s: Failed to stop governor\n", __func__);
1033                         goto unlock;
1034                 }
1035         }
1036
1037         cpumask_set_cpu(cpu, policy->cpus);
1038
1039         if (has_target()) {
1040                 ret = cpufreq_start_governor(policy);
1041                 if (ret)
1042                         pr_err("%s: Failed to start governor\n", __func__);
1043         }
1044
1045 unlock:
1046         up_write(&policy->rwsem);
1047         return ret;
1048 }
1049
1050 static void handle_update(struct work_struct *work)
1051 {
1052         struct cpufreq_policy *policy =
1053                 container_of(work, struct cpufreq_policy, update);
1054         unsigned int cpu = policy->cpu;
1055         pr_debug("handle_update for cpu %u called\n", cpu);
1056         cpufreq_update_policy(cpu);
1057 }
1058
1059 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1060 {
1061         struct device *dev = get_cpu_device(cpu);
1062         struct cpufreq_policy *policy;
1063         int ret;
1064
1065         if (WARN_ON(!dev))
1066                 return NULL;
1067
1068         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1069         if (!policy)
1070                 return NULL;
1071
1072         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1073                 goto err_free_policy;
1074
1075         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1076                 goto err_free_cpumask;
1077
1078         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1079                 goto err_free_rcpumask;
1080
1081         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1082                                    cpufreq_global_kobject, "policy%u", cpu);
1083         if (ret) {
1084                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1085                 goto err_free_real_cpus;
1086         }
1087
1088         INIT_LIST_HEAD(&policy->policy_list);
1089         init_rwsem(&policy->rwsem);
1090         spin_lock_init(&policy->transition_lock);
1091         init_waitqueue_head(&policy->transition_wait);
1092         init_completion(&policy->kobj_unregister);
1093         INIT_WORK(&policy->update, handle_update);
1094
1095         policy->cpu = cpu;
1096         return policy;
1097
1098 err_free_real_cpus:
1099         free_cpumask_var(policy->real_cpus);
1100 err_free_rcpumask:
1101         free_cpumask_var(policy->related_cpus);
1102 err_free_cpumask:
1103         free_cpumask_var(policy->cpus);
1104 err_free_policy:
1105         kfree(policy);
1106
1107         return NULL;
1108 }
1109
1110 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1111 {
1112         struct kobject *kobj;
1113         struct completion *cmp;
1114
1115         if (notify)
1116                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1117                                              CPUFREQ_REMOVE_POLICY, policy);
1118
1119         down_write(&policy->rwsem);
1120         cpufreq_remove_dev_symlink(policy);
1121         kobj = &policy->kobj;
1122         cmp = &policy->kobj_unregister;
1123         up_write(&policy->rwsem);
1124         kobject_put(kobj);
1125
1126         /*
1127          * We need to make sure that the underlying kobj is
1128          * actually not referenced anymore by anybody before we
1129          * proceed with unloading.
1130          */
1131         pr_debug("waiting for dropping of refcount\n");
1132         wait_for_completion(cmp);
1133         pr_debug("wait complete\n");
1134 }
1135
1136 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1137 {
1138         unsigned long flags;
1139         int cpu;
1140
1141         /* Remove policy from list */
1142         write_lock_irqsave(&cpufreq_driver_lock, flags);
1143         list_del(&policy->policy_list);
1144
1145         for_each_cpu(cpu, policy->related_cpus)
1146                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1147         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1148
1149         cpufreq_policy_put_kobj(policy, notify);
1150         free_cpumask_var(policy->real_cpus);
1151         free_cpumask_var(policy->related_cpus);
1152         free_cpumask_var(policy->cpus);
1153         kfree(policy);
1154 }
1155
1156 static int cpufreq_online(unsigned int cpu)
1157 {
1158         struct cpufreq_policy *policy;
1159         bool new_policy;
1160         unsigned long flags;
1161         unsigned int j;
1162         int ret;
1163
1164         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1165
1166         /* Check if this CPU already has a policy to manage it */
1167         policy = per_cpu(cpufreq_cpu_data, cpu);
1168         if (policy) {
1169                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1170                 if (!policy_is_inactive(policy))
1171                         return cpufreq_add_policy_cpu(policy, cpu);
1172
1173                 /* This is the only online CPU for the policy.  Start over. */
1174                 new_policy = false;
1175                 down_write(&policy->rwsem);
1176                 policy->cpu = cpu;
1177                 policy->governor = NULL;
1178                 up_write(&policy->rwsem);
1179         } else {
1180                 new_policy = true;
1181                 policy = cpufreq_policy_alloc(cpu);
1182                 if (!policy)
1183                         return -ENOMEM;
1184         }
1185
1186         cpumask_copy(policy->cpus, cpumask_of(cpu));
1187
1188         /* call driver. From then on the cpufreq must be able
1189          * to accept all calls to ->verify and ->setpolicy for this CPU
1190          */
1191         ret = cpufreq_driver->init(policy);
1192         if (ret) {
1193                 pr_debug("initialization failed\n");
1194                 goto out_free_policy;
1195         }
1196
1197         down_write(&policy->rwsem);
1198
1199         if (new_policy) {
1200                 /* related_cpus should at least include policy->cpus. */
1201                 cpumask_copy(policy->related_cpus, policy->cpus);
1202                 /* Remember CPUs present at the policy creation time. */
1203                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1204         }
1205
1206         /*
1207          * affected cpus must always be the one, which are online. We aren't
1208          * managing offline cpus here.
1209          */
1210         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1211
1212         if (new_policy) {
1213                 policy->user_policy.min = policy->min;
1214                 policy->user_policy.max = policy->max;
1215
1216                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1217                 for_each_cpu(j, policy->related_cpus)
1218                         per_cpu(cpufreq_cpu_data, j) = policy;
1219                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1220         }
1221
1222         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1223                 policy->cur = cpufreq_driver->get(policy->cpu);
1224                 if (!policy->cur) {
1225                         pr_err("%s: ->get() failed\n", __func__);
1226                         goto out_exit_policy;
1227                 }
1228         }
1229
1230         /*
1231          * Sometimes boot loaders set CPU frequency to a value outside of
1232          * frequency table present with cpufreq core. In such cases CPU might be
1233          * unstable if it has to run on that frequency for long duration of time
1234          * and so its better to set it to a frequency which is specified in
1235          * freq-table. This also makes cpufreq stats inconsistent as
1236          * cpufreq-stats would fail to register because current frequency of CPU
1237          * isn't found in freq-table.
1238          *
1239          * Because we don't want this change to effect boot process badly, we go
1240          * for the next freq which is >= policy->cur ('cur' must be set by now,
1241          * otherwise we will end up setting freq to lowest of the table as 'cur'
1242          * is initialized to zero).
1243          *
1244          * We are passing target-freq as "policy->cur - 1" otherwise
1245          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1246          * equal to target-freq.
1247          */
1248         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1249             && has_target()) {
1250                 /* Are we running at unknown frequency ? */
1251                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1252                 if (ret == -EINVAL) {
1253                         /* Warn user and fix it */
1254                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1255                                 __func__, policy->cpu, policy->cur);
1256                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1257                                 CPUFREQ_RELATION_L);
1258
1259                         /*
1260                          * Reaching here after boot in a few seconds may not
1261                          * mean that system will remain stable at "unknown"
1262                          * frequency for longer duration. Hence, a BUG_ON().
1263                          */
1264                         BUG_ON(ret);
1265                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1266                                 __func__, policy->cpu, policy->cur);
1267                 }
1268         }
1269
1270         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1271                                      CPUFREQ_START, policy);
1272
1273         if (new_policy) {
1274                 ret = cpufreq_add_dev_interface(policy);
1275                 if (ret)
1276                         goto out_exit_policy;
1277                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1278                                 CPUFREQ_CREATE_POLICY, policy);
1279
1280                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1281                 list_add(&policy->policy_list, &cpufreq_policy_list);
1282                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1283         }
1284
1285         ret = cpufreq_init_policy(policy);
1286         if (ret) {
1287                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1288                        __func__, cpu, ret);
1289                 /* cpufreq_policy_free() will notify based on this */
1290                 new_policy = false;
1291                 goto out_exit_policy;
1292         }
1293
1294         up_write(&policy->rwsem);
1295
1296         kobject_uevent(&policy->kobj, KOBJ_ADD);
1297
1298         /* Callback for handling stuff after policy is ready */
1299         if (cpufreq_driver->ready)
1300                 cpufreq_driver->ready(policy);
1301
1302         pr_debug("initialization complete\n");
1303
1304         return 0;
1305
1306 out_exit_policy:
1307         up_write(&policy->rwsem);
1308
1309         if (cpufreq_driver->exit)
1310                 cpufreq_driver->exit(policy);
1311 out_free_policy:
1312         cpufreq_policy_free(policy, !new_policy);
1313         return ret;
1314 }
1315
1316 /**
1317  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1318  * @dev: CPU device.
1319  * @sif: Subsystem interface structure pointer (not used)
1320  */
1321 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1322 {
1323         struct cpufreq_policy *policy;
1324         unsigned cpu = dev->id;
1325
1326         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1327
1328         if (cpu_online(cpu))
1329                 return cpufreq_online(cpu);
1330
1331         /*
1332          * A hotplug notifier will follow and we will handle it as CPU online
1333          * then.  For now, just create the sysfs link, unless there is no policy
1334          * or the link is already present.
1335          */
1336         policy = per_cpu(cpufreq_cpu_data, cpu);
1337         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1338                 return 0;
1339
1340         return add_cpu_dev_symlink(policy, cpu);
1341 }
1342
1343 static void cpufreq_offline(unsigned int cpu)
1344 {
1345         struct cpufreq_policy *policy;
1346         int ret;
1347
1348         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1349
1350         policy = cpufreq_cpu_get_raw(cpu);
1351         if (!policy) {
1352                 pr_debug("%s: No cpu_data found\n", __func__);
1353                 return;
1354         }
1355
1356         down_write(&policy->rwsem);
1357         if (has_target()) {
1358                 ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1359                 if (ret)
1360                         pr_err("%s: Failed to stop governor\n", __func__);
1361         }
1362
1363         cpumask_clear_cpu(cpu, policy->cpus);
1364
1365         if (policy_is_inactive(policy)) {
1366                 if (has_target())
1367                         strncpy(policy->last_governor, policy->governor->name,
1368                                 CPUFREQ_NAME_LEN);
1369                 else
1370                         policy->last_policy = policy->policy;
1371         } else if (cpu == policy->cpu) {
1372                 /* Nominate new CPU */
1373                 policy->cpu = cpumask_any(policy->cpus);
1374         }
1375
1376         /* Start governor again for active policy */
1377         if (!policy_is_inactive(policy)) {
1378                 if (has_target()) {
1379                         ret = cpufreq_start_governor(policy);
1380                         if (ret)
1381                                 pr_err("%s: Failed to start governor\n", __func__);
1382                 }
1383
1384                 goto unlock;
1385         }
1386
1387         if (cpufreq_driver->stop_cpu)
1388                 cpufreq_driver->stop_cpu(policy);
1389
1390         /* If cpu is last user of policy, free policy */
1391         if (has_target()) {
1392                 ret = cpufreq_exit_governor(policy);
1393                 if (ret)
1394                         pr_err("%s: Failed to exit governor\n", __func__);
1395         }
1396
1397         /*
1398          * Perform the ->exit() even during light-weight tear-down,
1399          * since this is a core component, and is essential for the
1400          * subsequent light-weight ->init() to succeed.
1401          */
1402         if (cpufreq_driver->exit) {
1403                 cpufreq_driver->exit(policy);
1404                 policy->freq_table = NULL;
1405         }
1406
1407 unlock:
1408         up_write(&policy->rwsem);
1409 }
1410
1411 /**
1412  * cpufreq_remove_dev - remove a CPU device
1413  *
1414  * Removes the cpufreq interface for a CPU device.
1415  */
1416 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1417 {
1418         unsigned int cpu = dev->id;
1419         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1420
1421         if (!policy)
1422                 return;
1423
1424         if (cpu_online(cpu))
1425                 cpufreq_offline(cpu);
1426
1427         cpumask_clear_cpu(cpu, policy->real_cpus);
1428         remove_cpu_dev_symlink(policy, cpu);
1429
1430         if (cpumask_empty(policy->real_cpus))
1431                 cpufreq_policy_free(policy, true);
1432 }
1433
1434 /**
1435  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1436  *      in deep trouble.
1437  *      @policy: policy managing CPUs
1438  *      @new_freq: CPU frequency the CPU actually runs at
1439  *
1440  *      We adjust to current frequency first, and need to clean up later.
1441  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1442  */
1443 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1444                                 unsigned int new_freq)
1445 {
1446         struct cpufreq_freqs freqs;
1447
1448         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1449                  policy->cur, new_freq);
1450
1451         freqs.old = policy->cur;
1452         freqs.new = new_freq;
1453
1454         cpufreq_freq_transition_begin(policy, &freqs);
1455         cpufreq_freq_transition_end(policy, &freqs, 0);
1456 }
1457
1458 /**
1459  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1460  * @cpu: CPU number
1461  *
1462  * This is the last known freq, without actually getting it from the driver.
1463  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1464  */
1465 unsigned int cpufreq_quick_get(unsigned int cpu)
1466 {
1467         struct cpufreq_policy *policy;
1468         unsigned int ret_freq = 0;
1469         unsigned long flags;
1470
1471         read_lock_irqsave(&cpufreq_driver_lock, flags);
1472
1473         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1474                 ret_freq = cpufreq_driver->get(cpu);
1475                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1476                 return ret_freq;
1477         }
1478
1479         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1480
1481         policy = cpufreq_cpu_get(cpu);
1482         if (policy) {
1483                 ret_freq = policy->cur;
1484                 cpufreq_cpu_put(policy);
1485         }
1486
1487         return ret_freq;
1488 }
1489 EXPORT_SYMBOL(cpufreq_quick_get);
1490
1491 /**
1492  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1493  * @cpu: CPU number
1494  *
1495  * Just return the max possible frequency for a given CPU.
1496  */
1497 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1498 {
1499         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1500         unsigned int ret_freq = 0;
1501
1502         if (policy) {
1503                 ret_freq = policy->max;
1504                 cpufreq_cpu_put(policy);
1505         }
1506
1507         return ret_freq;
1508 }
1509 EXPORT_SYMBOL(cpufreq_quick_get_max);
1510
1511 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1512 {
1513         unsigned int ret_freq = 0;
1514
1515         if (!cpufreq_driver->get)
1516                 return ret_freq;
1517
1518         ret_freq = cpufreq_driver->get(policy->cpu);
1519
1520         /*
1521          * Updating inactive policies is invalid, so avoid doing that.  Also
1522          * if fast frequency switching is used with the given policy, the check
1523          * against policy->cur is pointless, so skip it in that case too.
1524          */
1525         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1526                 return ret_freq;
1527
1528         if (ret_freq && policy->cur &&
1529                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1530                 /* verify no discrepancy between actual and
1531                                         saved value exists */
1532                 if (unlikely(ret_freq != policy->cur)) {
1533                         cpufreq_out_of_sync(policy, ret_freq);
1534                         schedule_work(&policy->update);
1535                 }
1536         }
1537
1538         return ret_freq;
1539 }
1540
1541 /**
1542  * cpufreq_get - get the current CPU frequency (in kHz)
1543  * @cpu: CPU number
1544  *
1545  * Get the CPU current (static) CPU frequency
1546  */
1547 unsigned int cpufreq_get(unsigned int cpu)
1548 {
1549         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1550         unsigned int ret_freq = 0;
1551
1552         if (policy) {
1553                 down_read(&policy->rwsem);
1554                 ret_freq = __cpufreq_get(policy);
1555                 up_read(&policy->rwsem);
1556
1557                 cpufreq_cpu_put(policy);
1558         }
1559
1560         return ret_freq;
1561 }
1562 EXPORT_SYMBOL(cpufreq_get);
1563
1564 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1565 {
1566         unsigned int new_freq;
1567
1568         if (cpufreq_suspended)
1569                 return 0;
1570
1571         new_freq = cpufreq_driver->get(policy->cpu);
1572         if (!new_freq)
1573                 return 0;
1574
1575         if (!policy->cur) {
1576                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1577                 policy->cur = new_freq;
1578         } else if (policy->cur != new_freq && has_target()) {
1579                 cpufreq_out_of_sync(policy, new_freq);
1580         }
1581
1582         return new_freq;
1583 }
1584
1585 static struct subsys_interface cpufreq_interface = {
1586         .name           = "cpufreq",
1587         .subsys         = &cpu_subsys,
1588         .add_dev        = cpufreq_add_dev,
1589         .remove_dev     = cpufreq_remove_dev,
1590 };
1591
1592 /*
1593  * In case platform wants some specific frequency to be configured
1594  * during suspend..
1595  */
1596 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1597 {
1598         int ret;
1599
1600         if (!policy->suspend_freq) {
1601                 pr_debug("%s: suspend_freq not defined\n", __func__);
1602                 return 0;
1603         }
1604
1605         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1606                         policy->suspend_freq);
1607
1608         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1609                         CPUFREQ_RELATION_H);
1610         if (ret)
1611                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1612                                 __func__, policy->suspend_freq, ret);
1613
1614         return ret;
1615 }
1616 EXPORT_SYMBOL(cpufreq_generic_suspend);
1617
1618 /**
1619  * cpufreq_suspend() - Suspend CPUFreq governors
1620  *
1621  * Called during system wide Suspend/Hibernate cycles for suspending governors
1622  * as some platforms can't change frequency after this point in suspend cycle.
1623  * Because some of the devices (like: i2c, regulators, etc) they use for
1624  * changing frequency are suspended quickly after this point.
1625  */
1626 void cpufreq_suspend(void)
1627 {
1628         struct cpufreq_policy *policy;
1629         int ret;
1630
1631         if (!cpufreq_driver)
1632                 return;
1633
1634         if (!has_target() && !cpufreq_driver->suspend)
1635                 goto suspend;
1636
1637         pr_debug("%s: Suspending Governors\n", __func__);
1638
1639         for_each_active_policy(policy) {
1640                 if (has_target()) {
1641                         down_write(&policy->rwsem);
1642                         ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1643                         up_write(&policy->rwsem);
1644
1645                         if (ret) {
1646                                 pr_err("%s: Failed to stop governor for policy: %p\n",
1647                                         __func__, policy);
1648                                 continue;
1649                         }
1650                 }
1651
1652                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1653                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1654                                 policy);
1655         }
1656
1657 suspend:
1658         cpufreq_suspended = true;
1659 }
1660
1661 /**
1662  * cpufreq_resume() - Resume CPUFreq governors
1663  *
1664  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1665  * are suspended with cpufreq_suspend().
1666  */
1667 void cpufreq_resume(void)
1668 {
1669         struct cpufreq_policy *policy;
1670         int ret;
1671
1672         if (!cpufreq_driver)
1673                 return;
1674
1675         cpufreq_suspended = false;
1676
1677         if (!has_target() && !cpufreq_driver->resume)
1678                 return;
1679
1680         pr_debug("%s: Resuming Governors\n", __func__);
1681
1682         for_each_active_policy(policy) {
1683                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1684                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1685                                 policy);
1686                 } else if (has_target()) {
1687                         down_write(&policy->rwsem);
1688                         ret = cpufreq_start_governor(policy);
1689                         up_write(&policy->rwsem);
1690
1691                         if (ret)
1692                                 pr_err("%s: Failed to start governor for policy: %p\n",
1693                                        __func__, policy);
1694                 }
1695         }
1696 }
1697
1698 /**
1699  *      cpufreq_get_current_driver - return current driver's name
1700  *
1701  *      Return the name string of the currently loaded cpufreq driver
1702  *      or NULL, if none.
1703  */
1704 const char *cpufreq_get_current_driver(void)
1705 {
1706         if (cpufreq_driver)
1707                 return cpufreq_driver->name;
1708
1709         return NULL;
1710 }
1711 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1712
1713 /**
1714  *      cpufreq_get_driver_data - return current driver data
1715  *
1716  *      Return the private data of the currently loaded cpufreq
1717  *      driver, or NULL if no cpufreq driver is loaded.
1718  */
1719 void *cpufreq_get_driver_data(void)
1720 {
1721         if (cpufreq_driver)
1722                 return cpufreq_driver->driver_data;
1723
1724         return NULL;
1725 }
1726 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1727
1728 /*********************************************************************
1729  *                     NOTIFIER LISTS INTERFACE                      *
1730  *********************************************************************/
1731
1732 /**
1733  *      cpufreq_register_notifier - register a driver with cpufreq
1734  *      @nb: notifier function to register
1735  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1736  *
1737  *      Add a driver to one of two lists: either a list of drivers that
1738  *      are notified about clock rate changes (once before and once after
1739  *      the transition), or a list of drivers that are notified about
1740  *      changes in cpufreq policy.
1741  *
1742  *      This function may sleep, and has the same return conditions as
1743  *      blocking_notifier_chain_register.
1744  */
1745 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1746 {
1747         int ret;
1748
1749         if (cpufreq_disabled())
1750                 return -EINVAL;
1751
1752         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1753
1754         switch (list) {
1755         case CPUFREQ_TRANSITION_NOTIFIER:
1756                 mutex_lock(&cpufreq_fast_switch_lock);
1757
1758                 if (cpufreq_fast_switch_count > 0) {
1759                         mutex_unlock(&cpufreq_fast_switch_lock);
1760                         return -EBUSY;
1761                 }
1762                 ret = srcu_notifier_chain_register(
1763                                 &cpufreq_transition_notifier_list, nb);
1764                 if (!ret)
1765                         cpufreq_fast_switch_count--;
1766
1767                 mutex_unlock(&cpufreq_fast_switch_lock);
1768                 break;
1769         case CPUFREQ_POLICY_NOTIFIER:
1770                 ret = blocking_notifier_chain_register(
1771                                 &cpufreq_policy_notifier_list, nb);
1772                 break;
1773         default:
1774                 ret = -EINVAL;
1775         }
1776
1777         return ret;
1778 }
1779 EXPORT_SYMBOL(cpufreq_register_notifier);
1780
1781 /**
1782  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1783  *      @nb: notifier block to be unregistered
1784  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1785  *
1786  *      Remove a driver from the CPU frequency notifier list.
1787  *
1788  *      This function may sleep, and has the same return conditions as
1789  *      blocking_notifier_chain_unregister.
1790  */
1791 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1792 {
1793         int ret;
1794
1795         if (cpufreq_disabled())
1796                 return -EINVAL;
1797
1798         switch (list) {
1799         case CPUFREQ_TRANSITION_NOTIFIER:
1800                 mutex_lock(&cpufreq_fast_switch_lock);
1801
1802                 ret = srcu_notifier_chain_unregister(
1803                                 &cpufreq_transition_notifier_list, nb);
1804                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1805                         cpufreq_fast_switch_count++;
1806
1807                 mutex_unlock(&cpufreq_fast_switch_lock);
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 /**
1827  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1828  * @policy: cpufreq policy to switch the frequency for.
1829  * @target_freq: New frequency to set (may be approximate).
1830  *
1831  * Carry out a fast frequency switch without sleeping.
1832  *
1833  * The driver's ->fast_switch() callback invoked by this function must be
1834  * suitable for being called from within RCU-sched read-side critical sections
1835  * and it is expected to select the minimum available frequency greater than or
1836  * equal to @target_freq (CPUFREQ_RELATION_L).
1837  *
1838  * This function must not be called if policy->fast_switch_enabled is unset.
1839  *
1840  * Governors calling this function must guarantee that it will never be invoked
1841  * twice in parallel for the same policy and that it will never be called in
1842  * parallel with either ->target() or ->target_index() for the same policy.
1843  *
1844  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1845  * callback to indicate an error condition, the hardware configuration must be
1846  * preserved.
1847  */
1848 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1849                                         unsigned int target_freq)
1850 {
1851         clamp_val(target_freq, policy->min, policy->max);
1852
1853         return cpufreq_driver->fast_switch(policy, target_freq);
1854 }
1855 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1856
1857 /* Must set freqs->new to intermediate frequency */
1858 static int __target_intermediate(struct cpufreq_policy *policy,
1859                                  struct cpufreq_freqs *freqs, int index)
1860 {
1861         int ret;
1862
1863         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1864
1865         /* We don't need to switch to intermediate freq */
1866         if (!freqs->new)
1867                 return 0;
1868
1869         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1870                  __func__, policy->cpu, freqs->old, freqs->new);
1871
1872         cpufreq_freq_transition_begin(policy, freqs);
1873         ret = cpufreq_driver->target_intermediate(policy, index);
1874         cpufreq_freq_transition_end(policy, freqs, ret);
1875
1876         if (ret)
1877                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1878                        __func__, ret);
1879
1880         return ret;
1881 }
1882
1883 static int __target_index(struct cpufreq_policy *policy,
1884                           struct cpufreq_frequency_table *freq_table, int index)
1885 {
1886         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1887         unsigned int intermediate_freq = 0;
1888         int retval = -EINVAL;
1889         bool notify;
1890
1891         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1892         if (notify) {
1893                 /* Handle switching to intermediate frequency */
1894                 if (cpufreq_driver->get_intermediate) {
1895                         retval = __target_intermediate(policy, &freqs, index);
1896                         if (retval)
1897                                 return retval;
1898
1899                         intermediate_freq = freqs.new;
1900                         /* Set old freq to intermediate */
1901                         if (intermediate_freq)
1902                                 freqs.old = freqs.new;
1903                 }
1904
1905                 freqs.new = freq_table[index].frequency;
1906                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1907                          __func__, policy->cpu, freqs.old, freqs.new);
1908
1909                 cpufreq_freq_transition_begin(policy, &freqs);
1910         }
1911
1912         retval = cpufreq_driver->target_index(policy, index);
1913         if (retval)
1914                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1915                        retval);
1916
1917         if (notify) {
1918                 cpufreq_freq_transition_end(policy, &freqs, retval);
1919
1920                 /*
1921                  * Failed after setting to intermediate freq? Driver should have
1922                  * reverted back to initial frequency and so should we. Check
1923                  * here for intermediate_freq instead of get_intermediate, in
1924                  * case we haven't switched to intermediate freq at all.
1925                  */
1926                 if (unlikely(retval && intermediate_freq)) {
1927                         freqs.old = intermediate_freq;
1928                         freqs.new = policy->restore_freq;
1929                         cpufreq_freq_transition_begin(policy, &freqs);
1930                         cpufreq_freq_transition_end(policy, &freqs, 0);
1931                 }
1932         }
1933
1934         return retval;
1935 }
1936
1937 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1938                             unsigned int target_freq,
1939                             unsigned int relation)
1940 {
1941         unsigned int old_target_freq = target_freq;
1942         struct cpufreq_frequency_table *freq_table;
1943         int index, retval;
1944
1945         if (cpufreq_disabled())
1946                 return -ENODEV;
1947
1948         /* Make sure that target_freq is within supported range */
1949         if (target_freq > policy->max)
1950                 target_freq = policy->max;
1951         if (target_freq < policy->min)
1952                 target_freq = policy->min;
1953
1954         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1955                  policy->cpu, target_freq, relation, old_target_freq);
1956
1957         /*
1958          * This might look like a redundant call as we are checking it again
1959          * after finding index. But it is left intentionally for cases where
1960          * exactly same freq is called again and so we can save on few function
1961          * calls.
1962          */
1963         if (target_freq == policy->cur)
1964                 return 0;
1965
1966         /* Save last value to restore later on errors */
1967         policy->restore_freq = policy->cur;
1968
1969         if (cpufreq_driver->target)
1970                 return cpufreq_driver->target(policy, target_freq, relation);
1971
1972         if (!cpufreq_driver->target_index)
1973                 return -EINVAL;
1974
1975         freq_table = cpufreq_frequency_get_table(policy->cpu);
1976         if (unlikely(!freq_table)) {
1977                 pr_err("%s: Unable to find freq_table\n", __func__);
1978                 return -EINVAL;
1979         }
1980
1981         retval = cpufreq_frequency_table_target(policy, freq_table, target_freq,
1982                                                 relation, &index);
1983         if (unlikely(retval)) {
1984                 pr_err("%s: Unable to find matching freq\n", __func__);
1985                 return retval;
1986         }
1987
1988         if (freq_table[index].frequency == policy->cur)
1989                 return 0;
1990
1991         return __target_index(policy, freq_table, index);
1992 }
1993 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1994
1995 int cpufreq_driver_target(struct cpufreq_policy *policy,
1996                           unsigned int target_freq,
1997                           unsigned int relation)
1998 {
1999         int ret = -EINVAL;
2000
2001         down_write(&policy->rwsem);
2002
2003         ret = __cpufreq_driver_target(policy, target_freq, relation);
2004
2005         up_write(&policy->rwsem);
2006
2007         return ret;
2008 }
2009 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2010
2011 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2012 {
2013         return NULL;
2014 }
2015
2016 static int cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
2017 {
2018         int ret;
2019
2020         /* Don't start any governor operations if we are entering suspend */
2021         if (cpufreq_suspended)
2022                 return 0;
2023         /*
2024          * Governor might not be initiated here if ACPI _PPC changed
2025          * notification happened, so check it.
2026          */
2027         if (!policy->governor)
2028                 return -EINVAL;
2029
2030         if (policy->governor->max_transition_latency &&
2031             policy->cpuinfo.transition_latency >
2032             policy->governor->max_transition_latency) {
2033                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2034
2035                 if (gov) {
2036                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2037                                 policy->governor->name, gov->name);
2038                         policy->governor = gov;
2039                 } else {
2040                         return -EINVAL;
2041                 }
2042         }
2043
2044         if (event == CPUFREQ_GOV_POLICY_INIT)
2045                 if (!try_module_get(policy->governor->owner))
2046                         return -EINVAL;
2047
2048         pr_debug("%s: for CPU %u, event %u\n", __func__, policy->cpu, event);
2049
2050         ret = policy->governor->governor(policy, event);
2051
2052         if (!ret) {
2053                 if (event == CPUFREQ_GOV_POLICY_INIT)
2054                         policy->governor->initialized++;
2055                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
2056                         policy->governor->initialized--;
2057         }
2058
2059         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2060                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2061                 module_put(policy->governor->owner);
2062
2063         return ret;
2064 }
2065
2066 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2067 {
2068         int ret;
2069
2070         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2071                 cpufreq_update_current_freq(policy);
2072
2073         ret = cpufreq_governor(policy, CPUFREQ_GOV_START);
2074         return ret ? ret : cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2075 }
2076
2077 int cpufreq_register_governor(struct cpufreq_governor *governor)
2078 {
2079         int err;
2080
2081         if (!governor)
2082                 return -EINVAL;
2083
2084         if (cpufreq_disabled())
2085                 return -ENODEV;
2086
2087         mutex_lock(&cpufreq_governor_mutex);
2088
2089         governor->initialized = 0;
2090         err = -EBUSY;
2091         if (!find_governor(governor->name)) {
2092                 err = 0;
2093                 list_add(&governor->governor_list, &cpufreq_governor_list);
2094         }
2095
2096         mutex_unlock(&cpufreq_governor_mutex);
2097         return err;
2098 }
2099 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2100
2101 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2102 {
2103         struct cpufreq_policy *policy;
2104         unsigned long flags;
2105
2106         if (!governor)
2107                 return;
2108
2109         if (cpufreq_disabled())
2110                 return;
2111
2112         /* clear last_governor for all inactive policies */
2113         read_lock_irqsave(&cpufreq_driver_lock, flags);
2114         for_each_inactive_policy(policy) {
2115                 if (!strcmp(policy->last_governor, governor->name)) {
2116                         policy->governor = NULL;
2117                         strcpy(policy->last_governor, "\0");
2118                 }
2119         }
2120         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2121
2122         mutex_lock(&cpufreq_governor_mutex);
2123         list_del(&governor->governor_list);
2124         mutex_unlock(&cpufreq_governor_mutex);
2125         return;
2126 }
2127 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2128
2129
2130 /*********************************************************************
2131  *                          POLICY INTERFACE                         *
2132  *********************************************************************/
2133
2134 /**
2135  * cpufreq_get_policy - get the current cpufreq_policy
2136  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2137  *      is written
2138  *
2139  * Reads the current cpufreq policy.
2140  */
2141 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2142 {
2143         struct cpufreq_policy *cpu_policy;
2144         if (!policy)
2145                 return -EINVAL;
2146
2147         cpu_policy = cpufreq_cpu_get(cpu);
2148         if (!cpu_policy)
2149                 return -EINVAL;
2150
2151         memcpy(policy, cpu_policy, sizeof(*policy));
2152
2153         cpufreq_cpu_put(cpu_policy);
2154         return 0;
2155 }
2156 EXPORT_SYMBOL(cpufreq_get_policy);
2157
2158 /*
2159  * policy : current policy.
2160  * new_policy: policy to be set.
2161  */
2162 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2163                                 struct cpufreq_policy *new_policy)
2164 {
2165         struct cpufreq_governor *old_gov;
2166         int ret;
2167
2168         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2169                  new_policy->cpu, new_policy->min, new_policy->max);
2170
2171         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2172
2173         /*
2174         * This check works well when we store new min/max freq attributes,
2175         * because new_policy is a copy of policy with one field updated.
2176         */
2177         if (new_policy->min > new_policy->max)
2178                 return -EINVAL;
2179
2180         /* verify the cpu speed can be set within this limit */
2181         ret = cpufreq_driver->verify(new_policy);
2182         if (ret)
2183                 return ret;
2184
2185         /* adjust if necessary - all reasons */
2186         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2187                         CPUFREQ_ADJUST, new_policy);
2188
2189         /*
2190          * verify the cpu speed can be set within this limit, which might be
2191          * different to the first one
2192          */
2193         ret = cpufreq_driver->verify(new_policy);
2194         if (ret)
2195                 return ret;
2196
2197         /* notification of the new policy */
2198         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2199                         CPUFREQ_NOTIFY, new_policy);
2200
2201         policy->min = new_policy->min;
2202         policy->max = new_policy->max;
2203
2204         pr_debug("new min and max freqs are %u - %u kHz\n",
2205                  policy->min, policy->max);
2206
2207         if (cpufreq_driver->setpolicy) {
2208                 policy->policy = new_policy->policy;
2209                 pr_debug("setting range\n");
2210                 return cpufreq_driver->setpolicy(new_policy);
2211         }
2212
2213         if (new_policy->governor == policy->governor) {
2214                 pr_debug("cpufreq: governor limits update\n");
2215                 return cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2216         }
2217
2218         pr_debug("governor switch\n");
2219
2220         /* save old, working values */
2221         old_gov = policy->governor;
2222         /* end old governor */
2223         if (old_gov) {
2224                 ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2225                 if (ret) {
2226                         /* This can happen due to race with other operations */
2227                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2228                                  __func__, old_gov->name, ret);
2229                         return ret;
2230                 }
2231
2232                 ret = cpufreq_exit_governor(policy);
2233                 if (ret) {
2234                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2235                                __func__, old_gov->name, ret);
2236                         return ret;
2237                 }
2238         }
2239
2240         /* start new governor */
2241         policy->governor = new_policy->governor;
2242         ret = cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2243         if (!ret) {
2244                 ret = cpufreq_start_governor(policy);
2245                 if (!ret) {
2246                         pr_debug("cpufreq: governor change\n");
2247                         return 0;
2248                 }
2249                 cpufreq_exit_governor(policy);
2250         }
2251
2252         /* new governor failed, so re-start old one */
2253         pr_debug("starting governor %s failed\n", policy->governor->name);
2254         if (old_gov) {
2255                 policy->governor = old_gov;
2256                 if (cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2257                         policy->governor = NULL;
2258                 else
2259                         cpufreq_start_governor(policy);
2260         }
2261
2262         return ret;
2263 }
2264
2265 /**
2266  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2267  *      @cpu: CPU which shall be re-evaluated
2268  *
2269  *      Useful for policy notifiers which have different necessities
2270  *      at different times.
2271  */
2272 int cpufreq_update_policy(unsigned int cpu)
2273 {
2274         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2275         struct cpufreq_policy new_policy;
2276         int ret;
2277
2278         if (!policy)
2279                 return -ENODEV;
2280
2281         down_write(&policy->rwsem);
2282
2283         pr_debug("updating policy for CPU %u\n", cpu);
2284         memcpy(&new_policy, policy, sizeof(*policy));
2285         new_policy.min = policy->user_policy.min;
2286         new_policy.max = policy->user_policy.max;
2287
2288         /*
2289          * BIOS might change freq behind our back
2290          * -> ask driver for current freq and notify governors about a change
2291          */
2292         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2293                 new_policy.cur = cpufreq_update_current_freq(policy);
2294                 if (WARN_ON(!new_policy.cur)) {
2295                         ret = -EIO;
2296                         goto unlock;
2297                 }
2298         }
2299
2300         ret = cpufreq_set_policy(policy, &new_policy);
2301
2302 unlock:
2303         up_write(&policy->rwsem);
2304
2305         cpufreq_cpu_put(policy);
2306         return ret;
2307 }
2308 EXPORT_SYMBOL(cpufreq_update_policy);
2309
2310 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2311                                         unsigned long action, void *hcpu)
2312 {
2313         unsigned int cpu = (unsigned long)hcpu;
2314
2315         switch (action & ~CPU_TASKS_FROZEN) {
2316         case CPU_ONLINE:
2317         case CPU_DOWN_FAILED:
2318                 cpufreq_online(cpu);
2319                 break;
2320
2321         case CPU_DOWN_PREPARE:
2322                 cpufreq_offline(cpu);
2323                 break;
2324         }
2325         return NOTIFY_OK;
2326 }
2327
2328 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2329         .notifier_call = cpufreq_cpu_callback,
2330 };
2331
2332 /*********************************************************************
2333  *               BOOST                                               *
2334  *********************************************************************/
2335 static int cpufreq_boost_set_sw(int state)
2336 {
2337         struct cpufreq_frequency_table *freq_table;
2338         struct cpufreq_policy *policy;
2339         int ret = -EINVAL;
2340
2341         for_each_active_policy(policy) {
2342                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2343                 if (freq_table) {
2344                         ret = cpufreq_frequency_table_cpuinfo(policy,
2345                                                         freq_table);
2346                         if (ret) {
2347                                 pr_err("%s: Policy frequency update failed\n",
2348                                        __func__);
2349                                 break;
2350                         }
2351
2352                         down_write(&policy->rwsem);
2353                         policy->user_policy.max = policy->max;
2354                         cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2355                         up_write(&policy->rwsem);
2356                 }
2357         }
2358
2359         return ret;
2360 }
2361
2362 int cpufreq_boost_trigger_state(int state)
2363 {
2364         unsigned long flags;
2365         int ret = 0;
2366
2367         if (cpufreq_driver->boost_enabled == state)
2368                 return 0;
2369
2370         write_lock_irqsave(&cpufreq_driver_lock, flags);
2371         cpufreq_driver->boost_enabled = state;
2372         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2373
2374         ret = cpufreq_driver->set_boost(state);
2375         if (ret) {
2376                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2377                 cpufreq_driver->boost_enabled = !state;
2378                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2379
2380                 pr_err("%s: Cannot %s BOOST\n",
2381                        __func__, state ? "enable" : "disable");
2382         }
2383
2384         return ret;
2385 }
2386
2387 static bool cpufreq_boost_supported(void)
2388 {
2389         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2390 }
2391
2392 static int create_boost_sysfs_file(void)
2393 {
2394         int ret;
2395
2396         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2397         if (ret)
2398                 pr_err("%s: cannot register global BOOST sysfs file\n",
2399                        __func__);
2400
2401         return ret;
2402 }
2403
2404 static void remove_boost_sysfs_file(void)
2405 {
2406         if (cpufreq_boost_supported())
2407                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2408 }
2409
2410 int cpufreq_enable_boost_support(void)
2411 {
2412         if (!cpufreq_driver)
2413                 return -EINVAL;
2414
2415         if (cpufreq_boost_supported())
2416                 return 0;
2417
2418         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2419
2420         /* This will get removed on driver unregister */
2421         return create_boost_sysfs_file();
2422 }
2423 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2424
2425 int cpufreq_boost_enabled(void)
2426 {
2427         return cpufreq_driver->boost_enabled;
2428 }
2429 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2430
2431 /*********************************************************************
2432  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2433  *********************************************************************/
2434
2435 /**
2436  * cpufreq_register_driver - register a CPU Frequency driver
2437  * @driver_data: A struct cpufreq_driver containing the values#
2438  * submitted by the CPU Frequency driver.
2439  *
2440  * Registers a CPU Frequency driver to this core code. This code
2441  * returns zero on success, -EEXIST when another driver got here first
2442  * (and isn't unregistered in the meantime).
2443  *
2444  */
2445 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2446 {
2447         unsigned long flags;
2448         int ret;
2449
2450         if (cpufreq_disabled())
2451                 return -ENODEV;
2452
2453         if (!driver_data || !driver_data->verify || !driver_data->init ||
2454             !(driver_data->setpolicy || driver_data->target_index ||
2455                     driver_data->target) ||
2456              (driver_data->setpolicy && (driver_data->target_index ||
2457                     driver_data->target)) ||
2458              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2459                 return -EINVAL;
2460
2461         pr_debug("trying to register driver %s\n", driver_data->name);
2462
2463         /* Protect against concurrent CPU online/offline. */
2464         get_online_cpus();
2465
2466         write_lock_irqsave(&cpufreq_driver_lock, flags);
2467         if (cpufreq_driver) {
2468                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2469                 ret = -EEXIST;
2470                 goto out;
2471         }
2472         cpufreq_driver = driver_data;
2473         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2474
2475         if (driver_data->setpolicy)
2476                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2477
2478         if (cpufreq_boost_supported()) {
2479                 ret = create_boost_sysfs_file();
2480                 if (ret)
2481                         goto err_null_driver;
2482         }
2483
2484         ret = subsys_interface_register(&cpufreq_interface);
2485         if (ret)
2486                 goto err_boost_unreg;
2487
2488         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2489             list_empty(&cpufreq_policy_list)) {
2490                 /* if all ->init() calls failed, unregister */
2491                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2492                          driver_data->name);
2493                 goto err_if_unreg;
2494         }
2495
2496         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2497         pr_debug("driver %s up and running\n", driver_data->name);
2498
2499 out:
2500         put_online_cpus();
2501         return ret;
2502
2503 err_if_unreg:
2504         subsys_interface_unregister(&cpufreq_interface);
2505 err_boost_unreg:
2506         remove_boost_sysfs_file();
2507 err_null_driver:
2508         write_lock_irqsave(&cpufreq_driver_lock, flags);
2509         cpufreq_driver = NULL;
2510         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2511         goto out;
2512 }
2513 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2514
2515 /**
2516  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2517  *
2518  * Unregister the current CPUFreq driver. Only call this if you have
2519  * the right to do so, i.e. if you have succeeded in initialising before!
2520  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2521  * currently not initialised.
2522  */
2523 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2524 {
2525         unsigned long flags;
2526
2527         if (!cpufreq_driver || (driver != cpufreq_driver))
2528                 return -EINVAL;
2529
2530         pr_debug("unregistering driver %s\n", driver->name);
2531
2532         /* Protect against concurrent cpu hotplug */
2533         get_online_cpus();
2534         subsys_interface_unregister(&cpufreq_interface);
2535         remove_boost_sysfs_file();
2536         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2537
2538         write_lock_irqsave(&cpufreq_driver_lock, flags);
2539
2540         cpufreq_driver = NULL;
2541
2542         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2543         put_online_cpus();
2544
2545         return 0;
2546 }
2547 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2548
2549 /*
2550  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2551  * or mutexes when secondary CPUs are halted.
2552  */
2553 static struct syscore_ops cpufreq_syscore_ops = {
2554         .shutdown = cpufreq_suspend,
2555 };
2556
2557 struct kobject *cpufreq_global_kobject;
2558 EXPORT_SYMBOL(cpufreq_global_kobject);
2559
2560 static int __init cpufreq_core_init(void)
2561 {
2562         if (cpufreq_disabled())
2563                 return -ENODEV;
2564
2565         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2566         BUG_ON(!cpufreq_global_kobject);
2567
2568         register_syscore_ops(&cpufreq_syscore_ops);
2569
2570         return 0;
2571 }
2572 core_initcall(cpufreq_core_init);