293f47b814bf0111661d2529689267511857d486
[linux-2.6-microblaze.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 static bool suitable_policy(struct cpufreq_policy *policy, bool active)
42 {
43         return active == !policy_is_inactive(policy);
44 }
45
46 /* Finds Next Acive/Inactive policy */
47 static struct cpufreq_policy *next_policy(struct cpufreq_policy *policy,
48                                           bool active)
49 {
50         do {
51                 policy = list_next_entry(policy, policy_list);
52
53                 /* No more policies in the list */
54                 if (&policy->policy_list == &cpufreq_policy_list)
55                         return NULL;
56         } while (!suitable_policy(policy, active));
57
58         return policy;
59 }
60
61 static struct cpufreq_policy *first_policy(bool active)
62 {
63         struct cpufreq_policy *policy;
64
65         /* No policies in the list */
66         if (list_empty(&cpufreq_policy_list))
67                 return NULL;
68
69         policy = list_first_entry(&cpufreq_policy_list, typeof(*policy),
70                                   policy_list);
71
72         if (!suitable_policy(policy, active))
73                 policy = next_policy(policy, active);
74
75         return policy;
76 }
77
78 /* Macros to iterate over CPU policies */
79 #define for_each_suitable_policy(__policy, __active)    \
80         for (__policy = first_policy(__active);         \
81              __policy;                                  \
82              __policy = next_policy(__policy, __active))
83
84 #define for_each_active_policy(__policy)                \
85         for_each_suitable_policy(__policy, true)
86 #define for_each_inactive_policy(__policy)              \
87         for_each_suitable_policy(__policy, false)
88
89 #define for_each_policy(__policy)                       \
90         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
91
92 /* Iterate over governors */
93 static LIST_HEAD(cpufreq_governor_list);
94 #define for_each_governor(__governor)                           \
95         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
96
97 /**
98  * The "cpufreq driver" - the arch- or hardware-dependent low
99  * level driver of CPUFreq support, and its spinlock. This lock
100  * also protects the cpufreq_cpu_data array.
101  */
102 static struct cpufreq_driver *cpufreq_driver;
103 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
104 static DEFINE_RWLOCK(cpufreq_driver_lock);
105 DEFINE_MUTEX(cpufreq_governor_lock);
106
107 /* Flag to suspend/resume CPUFreq governors */
108 static bool cpufreq_suspended;
109
110 static inline bool has_target(void)
111 {
112         return cpufreq_driver->target_index || cpufreq_driver->target;
113 }
114
115 /* internal prototypes */
116 static int __cpufreq_governor(struct cpufreq_policy *policy,
117                 unsigned int event);
118 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
119 static void handle_update(struct work_struct *work);
120
121 /**
122  * Two notifier lists: the "policy" list is involved in the
123  * validation process for a new CPU frequency policy; the
124  * "transition" list for kernel code that needs to handle
125  * changes to devices when the CPU clock speed changes.
126  * The mutex locks both lists.
127  */
128 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
129 static struct srcu_notifier_head cpufreq_transition_notifier_list;
130
131 static bool init_cpufreq_transition_notifier_list_called;
132 static int __init init_cpufreq_transition_notifier_list(void)
133 {
134         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
135         init_cpufreq_transition_notifier_list_called = true;
136         return 0;
137 }
138 pure_initcall(init_cpufreq_transition_notifier_list);
139
140 static int off __read_mostly;
141 static int cpufreq_disabled(void)
142 {
143         return off;
144 }
145 void disable_cpufreq(void)
146 {
147         off = 1;
148 }
149 static DEFINE_MUTEX(cpufreq_governor_mutex);
150
151 bool have_governor_per_policy(void)
152 {
153         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
154 }
155 EXPORT_SYMBOL_GPL(have_governor_per_policy);
156
157 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
158 {
159         if (have_governor_per_policy())
160                 return &policy->kobj;
161         else
162                 return cpufreq_global_kobject;
163 }
164 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
165
166 struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu)
167 {
168         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
169
170         return policy && !policy_is_inactive(policy) ?
171                 policy->freq_table : NULL;
172 }
173 EXPORT_SYMBOL_GPL(cpufreq_frequency_get_table);
174
175 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
176 {
177         u64 idle_time;
178         u64 cur_wall_time;
179         u64 busy_time;
180
181         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
182
183         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
184         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
185         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
186         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
187         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
188         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
189
190         idle_time = cur_wall_time - busy_time;
191         if (wall)
192                 *wall = cputime_to_usecs(cur_wall_time);
193
194         return cputime_to_usecs(idle_time);
195 }
196
197 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
198 {
199         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
200
201         if (idle_time == -1ULL)
202                 return get_cpu_idle_time_jiffy(cpu, wall);
203         else if (!io_busy)
204                 idle_time += get_cpu_iowait_time_us(cpu, wall);
205
206         return idle_time;
207 }
208 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
209
210 /*
211  * This is a generic cpufreq init() routine which can be used by cpufreq
212  * drivers of SMP systems. It will do following:
213  * - validate & show freq table passed
214  * - set policies transition latency
215  * - policy->cpus with all possible CPUs
216  */
217 int cpufreq_generic_init(struct cpufreq_policy *policy,
218                 struct cpufreq_frequency_table *table,
219                 unsigned int transition_latency)
220 {
221         int ret;
222
223         ret = cpufreq_table_validate_and_show(policy, table);
224         if (ret) {
225                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
226                 return ret;
227         }
228
229         policy->cpuinfo.transition_latency = transition_latency;
230
231         /*
232          * The driver only supports the SMP configuration where all processors
233          * share the clock and voltage and clock.
234          */
235         cpumask_setall(policy->cpus);
236
237         return 0;
238 }
239 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
240
241 /* Only for cpufreq core internal use */
242 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
243 {
244         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
245
246         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
247 }
248
249 unsigned int cpufreq_generic_get(unsigned int cpu)
250 {
251         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
252
253         if (!policy || IS_ERR(policy->clk)) {
254                 pr_err("%s: No %s associated to cpu: %d\n",
255                        __func__, policy ? "clk" : "policy", cpu);
256                 return 0;
257         }
258
259         return clk_get_rate(policy->clk) / 1000;
260 }
261 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
262
263 /**
264  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
265  *
266  * @cpu: cpu to find policy for.
267  *
268  * This returns policy for 'cpu', returns NULL if it doesn't exist.
269  * It also increments the kobject reference count to mark it busy and so would
270  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
271  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
272  * freed as that depends on the kobj count.
273  *
274  * Return: A valid policy on success, otherwise NULL on failure.
275  */
276 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
277 {
278         struct cpufreq_policy *policy = NULL;
279         unsigned long flags;
280
281         if (WARN_ON(cpu >= nr_cpu_ids))
282                 return NULL;
283
284         /* get the cpufreq driver */
285         read_lock_irqsave(&cpufreq_driver_lock, flags);
286
287         if (cpufreq_driver) {
288                 /* get the CPU */
289                 policy = cpufreq_cpu_get_raw(cpu);
290                 if (policy)
291                         kobject_get(&policy->kobj);
292         }
293
294         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
295
296         return policy;
297 }
298 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
299
300 /**
301  * cpufreq_cpu_put: Decrements the usage count of a policy
302  *
303  * @policy: policy earlier returned by cpufreq_cpu_get().
304  *
305  * This decrements the kobject reference count incremented earlier by calling
306  * cpufreq_cpu_get().
307  */
308 void cpufreq_cpu_put(struct cpufreq_policy *policy)
309 {
310         kobject_put(&policy->kobj);
311 }
312 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
313
314 /*********************************************************************
315  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
316  *********************************************************************/
317
318 /**
319  * adjust_jiffies - adjust the system "loops_per_jiffy"
320  *
321  * This function alters the system "loops_per_jiffy" for the clock
322  * speed change. Note that loops_per_jiffy cannot be updated on SMP
323  * systems as each CPU might be scaled differently. So, use the arch
324  * per-CPU loops_per_jiffy value wherever possible.
325  */
326 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
327 {
328 #ifndef CONFIG_SMP
329         static unsigned long l_p_j_ref;
330         static unsigned int l_p_j_ref_freq;
331
332         if (ci->flags & CPUFREQ_CONST_LOOPS)
333                 return;
334
335         if (!l_p_j_ref_freq) {
336                 l_p_j_ref = loops_per_jiffy;
337                 l_p_j_ref_freq = ci->old;
338                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
339                          l_p_j_ref, l_p_j_ref_freq);
340         }
341         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
342                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
343                                                                 ci->new);
344                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
345                          loops_per_jiffy, ci->new);
346         }
347 #endif
348 }
349
350 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
351                 struct cpufreq_freqs *freqs, unsigned int state)
352 {
353         BUG_ON(irqs_disabled());
354
355         if (cpufreq_disabled())
356                 return;
357
358         freqs->flags = cpufreq_driver->flags;
359         pr_debug("notification %u of frequency transition to %u kHz\n",
360                  state, freqs->new);
361
362         switch (state) {
363
364         case CPUFREQ_PRECHANGE:
365                 /* detect if the driver reported a value as "old frequency"
366                  * which is not equal to what the cpufreq core thinks is
367                  * "old frequency".
368                  */
369                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
370                         if ((policy) && (policy->cpu == freqs->cpu) &&
371                             (policy->cur) && (policy->cur != freqs->old)) {
372                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
373                                          freqs->old, policy->cur);
374                                 freqs->old = policy->cur;
375                         }
376                 }
377                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
378                                 CPUFREQ_PRECHANGE, freqs);
379                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
380                 break;
381
382         case CPUFREQ_POSTCHANGE:
383                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
384                 pr_debug("FREQ: %lu - CPU: %lu\n",
385                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
386                 trace_cpu_frequency(freqs->new, freqs->cpu);
387                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
388                                 CPUFREQ_POSTCHANGE, freqs);
389                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
390                         policy->cur = freqs->new;
391                 break;
392         }
393 }
394
395 /**
396  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
397  * on frequency transition.
398  *
399  * This function calls the transition notifiers and the "adjust_jiffies"
400  * function. It is called twice on all CPU frequency changes that have
401  * external effects.
402  */
403 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
404                 struct cpufreq_freqs *freqs, unsigned int state)
405 {
406         for_each_cpu(freqs->cpu, policy->cpus)
407                 __cpufreq_notify_transition(policy, freqs, state);
408 }
409
410 /* Do post notifications when there are chances that transition has failed */
411 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
412                 struct cpufreq_freqs *freqs, int transition_failed)
413 {
414         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
415         if (!transition_failed)
416                 return;
417
418         swap(freqs->old, freqs->new);
419         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
420         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
421 }
422
423 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
424                 struct cpufreq_freqs *freqs)
425 {
426
427         /*
428          * Catch double invocations of _begin() which lead to self-deadlock.
429          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
430          * doesn't invoke _begin() on their behalf, and hence the chances of
431          * double invocations are very low. Moreover, there are scenarios
432          * where these checks can emit false-positive warnings in these
433          * drivers; so we avoid that by skipping them altogether.
434          */
435         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
436                                 && current == policy->transition_task);
437
438 wait:
439         wait_event(policy->transition_wait, !policy->transition_ongoing);
440
441         spin_lock(&policy->transition_lock);
442
443         if (unlikely(policy->transition_ongoing)) {
444                 spin_unlock(&policy->transition_lock);
445                 goto wait;
446         }
447
448         policy->transition_ongoing = true;
449         policy->transition_task = current;
450
451         spin_unlock(&policy->transition_lock);
452
453         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
454 }
455 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
456
457 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
458                 struct cpufreq_freqs *freqs, int transition_failed)
459 {
460         if (unlikely(WARN_ON(!policy->transition_ongoing)))
461                 return;
462
463         cpufreq_notify_post_transition(policy, freqs, transition_failed);
464
465         policy->transition_ongoing = false;
466         policy->transition_task = NULL;
467
468         wake_up(&policy->transition_wait);
469 }
470 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
471
472
473 /*********************************************************************
474  *                          SYSFS INTERFACE                          *
475  *********************************************************************/
476 static ssize_t show_boost(struct kobject *kobj,
477                                  struct attribute *attr, char *buf)
478 {
479         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
480 }
481
482 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
483                                   const char *buf, size_t count)
484 {
485         int ret, enable;
486
487         ret = sscanf(buf, "%d", &enable);
488         if (ret != 1 || enable < 0 || enable > 1)
489                 return -EINVAL;
490
491         if (cpufreq_boost_trigger_state(enable)) {
492                 pr_err("%s: Cannot %s BOOST!\n",
493                        __func__, enable ? "enable" : "disable");
494                 return -EINVAL;
495         }
496
497         pr_debug("%s: cpufreq BOOST %s\n",
498                  __func__, enable ? "enabled" : "disabled");
499
500         return count;
501 }
502 define_one_global_rw(boost);
503
504 static struct cpufreq_governor *find_governor(const char *str_governor)
505 {
506         struct cpufreq_governor *t;
507
508         for_each_governor(t)
509                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
510                         return t;
511
512         return NULL;
513 }
514
515 /**
516  * cpufreq_parse_governor - parse a governor string
517  */
518 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
519                                 struct cpufreq_governor **governor)
520 {
521         int err = -EINVAL;
522
523         if (!cpufreq_driver)
524                 goto out;
525
526         if (cpufreq_driver->setpolicy) {
527                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
528                         *policy = CPUFREQ_POLICY_PERFORMANCE;
529                         err = 0;
530                 } else if (!strncasecmp(str_governor, "powersave",
531                                                 CPUFREQ_NAME_LEN)) {
532                         *policy = CPUFREQ_POLICY_POWERSAVE;
533                         err = 0;
534                 }
535         } else {
536                 struct cpufreq_governor *t;
537
538                 mutex_lock(&cpufreq_governor_mutex);
539
540                 t = find_governor(str_governor);
541
542                 if (t == NULL) {
543                         int ret;
544
545                         mutex_unlock(&cpufreq_governor_mutex);
546                         ret = request_module("cpufreq_%s", str_governor);
547                         mutex_lock(&cpufreq_governor_mutex);
548
549                         if (ret == 0)
550                                 t = find_governor(str_governor);
551                 }
552
553                 if (t != NULL) {
554                         *governor = t;
555                         err = 0;
556                 }
557
558                 mutex_unlock(&cpufreq_governor_mutex);
559         }
560 out:
561         return err;
562 }
563
564 /**
565  * cpufreq_per_cpu_attr_read() / show_##file_name() -
566  * print out cpufreq information
567  *
568  * Write out information from cpufreq_driver->policy[cpu]; object must be
569  * "unsigned int".
570  */
571
572 #define show_one(file_name, object)                     \
573 static ssize_t show_##file_name                         \
574 (struct cpufreq_policy *policy, char *buf)              \
575 {                                                       \
576         return sprintf(buf, "%u\n", policy->object);    \
577 }
578
579 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
580 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
581 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
582 show_one(scaling_min_freq, min);
583 show_one(scaling_max_freq, max);
584
585 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
586 {
587         ssize_t ret;
588
589         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
590                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
591         else
592                 ret = sprintf(buf, "%u\n", policy->cur);
593         return ret;
594 }
595
596 static int cpufreq_set_policy(struct cpufreq_policy *policy,
597                                 struct cpufreq_policy *new_policy);
598
599 /**
600  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
601  */
602 #define store_one(file_name, object)                    \
603 static ssize_t store_##file_name                                        \
604 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
605 {                                                                       \
606         int ret, temp;                                                  \
607         struct cpufreq_policy new_policy;                               \
608                                                                         \
609         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
610         if (ret)                                                        \
611                 return -EINVAL;                                         \
612                                                                         \
613         ret = sscanf(buf, "%u", &new_policy.object);                    \
614         if (ret != 1)                                                   \
615                 return -EINVAL;                                         \
616                                                                         \
617         temp = new_policy.object;                                       \
618         ret = cpufreq_set_policy(policy, &new_policy);          \
619         if (!ret)                                                       \
620                 policy->user_policy.object = temp;                      \
621                                                                         \
622         return ret ? ret : count;                                       \
623 }
624
625 store_one(scaling_min_freq, min);
626 store_one(scaling_max_freq, max);
627
628 /**
629  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
630  */
631 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
632                                         char *buf)
633 {
634         unsigned int cur_freq = __cpufreq_get(policy);
635         if (!cur_freq)
636                 return sprintf(buf, "<unknown>");
637         return sprintf(buf, "%u\n", cur_freq);
638 }
639
640 /**
641  * show_scaling_governor - show the current policy for the specified CPU
642  */
643 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
644 {
645         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
646                 return sprintf(buf, "powersave\n");
647         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
648                 return sprintf(buf, "performance\n");
649         else if (policy->governor)
650                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
651                                 policy->governor->name);
652         return -EINVAL;
653 }
654
655 /**
656  * store_scaling_governor - store policy for the specified CPU
657  */
658 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
659                                         const char *buf, size_t count)
660 {
661         int ret;
662         char    str_governor[16];
663         struct cpufreq_policy new_policy;
664
665         ret = cpufreq_get_policy(&new_policy, policy->cpu);
666         if (ret)
667                 return ret;
668
669         ret = sscanf(buf, "%15s", str_governor);
670         if (ret != 1)
671                 return -EINVAL;
672
673         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
674                                                 &new_policy.governor))
675                 return -EINVAL;
676
677         ret = cpufreq_set_policy(policy, &new_policy);
678
679         policy->user_policy.policy = policy->policy;
680         policy->user_policy.governor = policy->governor;
681
682         if (ret)
683                 return ret;
684         else
685                 return count;
686 }
687
688 /**
689  * show_scaling_driver - show the cpufreq driver currently loaded
690  */
691 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
692 {
693         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
694 }
695
696 /**
697  * show_scaling_available_governors - show the available CPUfreq governors
698  */
699 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
700                                                 char *buf)
701 {
702         ssize_t i = 0;
703         struct cpufreq_governor *t;
704
705         if (!has_target()) {
706                 i += sprintf(buf, "performance powersave");
707                 goto out;
708         }
709
710         for_each_governor(t) {
711                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
712                     - (CPUFREQ_NAME_LEN + 2)))
713                         goto out;
714                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
715         }
716 out:
717         i += sprintf(&buf[i], "\n");
718         return i;
719 }
720
721 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
722 {
723         ssize_t i = 0;
724         unsigned int cpu;
725
726         for_each_cpu(cpu, mask) {
727                 if (i)
728                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
729                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
730                 if (i >= (PAGE_SIZE - 5))
731                         break;
732         }
733         i += sprintf(&buf[i], "\n");
734         return i;
735 }
736 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
737
738 /**
739  * show_related_cpus - show the CPUs affected by each transition even if
740  * hw coordination is in use
741  */
742 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
743 {
744         return cpufreq_show_cpus(policy->related_cpus, buf);
745 }
746
747 /**
748  * show_affected_cpus - show the CPUs affected by each transition
749  */
750 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
751 {
752         return cpufreq_show_cpus(policy->cpus, buf);
753 }
754
755 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
756                                         const char *buf, size_t count)
757 {
758         unsigned int freq = 0;
759         unsigned int ret;
760
761         if (!policy->governor || !policy->governor->store_setspeed)
762                 return -EINVAL;
763
764         ret = sscanf(buf, "%u", &freq);
765         if (ret != 1)
766                 return -EINVAL;
767
768         policy->governor->store_setspeed(policy, freq);
769
770         return count;
771 }
772
773 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
774 {
775         if (!policy->governor || !policy->governor->show_setspeed)
776                 return sprintf(buf, "<unsupported>\n");
777
778         return policy->governor->show_setspeed(policy, buf);
779 }
780
781 /**
782  * show_bios_limit - show the current cpufreq HW/BIOS limitation
783  */
784 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
785 {
786         unsigned int limit;
787         int ret;
788         if (cpufreq_driver->bios_limit) {
789                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
790                 if (!ret)
791                         return sprintf(buf, "%u\n", limit);
792         }
793         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
794 }
795
796 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
797 cpufreq_freq_attr_ro(cpuinfo_min_freq);
798 cpufreq_freq_attr_ro(cpuinfo_max_freq);
799 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
800 cpufreq_freq_attr_ro(scaling_available_governors);
801 cpufreq_freq_attr_ro(scaling_driver);
802 cpufreq_freq_attr_ro(scaling_cur_freq);
803 cpufreq_freq_attr_ro(bios_limit);
804 cpufreq_freq_attr_ro(related_cpus);
805 cpufreq_freq_attr_ro(affected_cpus);
806 cpufreq_freq_attr_rw(scaling_min_freq);
807 cpufreq_freq_attr_rw(scaling_max_freq);
808 cpufreq_freq_attr_rw(scaling_governor);
809 cpufreq_freq_attr_rw(scaling_setspeed);
810
811 static struct attribute *default_attrs[] = {
812         &cpuinfo_min_freq.attr,
813         &cpuinfo_max_freq.attr,
814         &cpuinfo_transition_latency.attr,
815         &scaling_min_freq.attr,
816         &scaling_max_freq.attr,
817         &affected_cpus.attr,
818         &related_cpus.attr,
819         &scaling_governor.attr,
820         &scaling_driver.attr,
821         &scaling_available_governors.attr,
822         &scaling_setspeed.attr,
823         NULL
824 };
825
826 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
827 #define to_attr(a) container_of(a, struct freq_attr, attr)
828
829 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
830 {
831         struct cpufreq_policy *policy = to_policy(kobj);
832         struct freq_attr *fattr = to_attr(attr);
833         ssize_t ret;
834
835         down_read(&policy->rwsem);
836
837         if (fattr->show)
838                 ret = fattr->show(policy, buf);
839         else
840                 ret = -EIO;
841
842         up_read(&policy->rwsem);
843
844         return ret;
845 }
846
847 static ssize_t store(struct kobject *kobj, struct attribute *attr,
848                      const char *buf, size_t count)
849 {
850         struct cpufreq_policy *policy = to_policy(kobj);
851         struct freq_attr *fattr = to_attr(attr);
852         ssize_t ret = -EINVAL;
853
854         get_online_cpus();
855
856         if (!cpu_online(policy->cpu))
857                 goto unlock;
858
859         down_write(&policy->rwsem);
860
861         /* Updating inactive policies is invalid, so avoid doing that. */
862         if (unlikely(policy_is_inactive(policy))) {
863                 ret = -EBUSY;
864                 goto unlock_policy_rwsem;
865         }
866
867         if (fattr->store)
868                 ret = fattr->store(policy, buf, count);
869         else
870                 ret = -EIO;
871
872 unlock_policy_rwsem:
873         up_write(&policy->rwsem);
874 unlock:
875         put_online_cpus();
876
877         return ret;
878 }
879
880 static void cpufreq_sysfs_release(struct kobject *kobj)
881 {
882         struct cpufreq_policy *policy = to_policy(kobj);
883         pr_debug("last reference is dropped\n");
884         complete(&policy->kobj_unregister);
885 }
886
887 static const struct sysfs_ops sysfs_ops = {
888         .show   = show,
889         .store  = store,
890 };
891
892 static struct kobj_type ktype_cpufreq = {
893         .sysfs_ops      = &sysfs_ops,
894         .default_attrs  = default_attrs,
895         .release        = cpufreq_sysfs_release,
896 };
897
898 struct kobject *cpufreq_global_kobject;
899 EXPORT_SYMBOL(cpufreq_global_kobject);
900
901 static int cpufreq_global_kobject_usage;
902
903 int cpufreq_get_global_kobject(void)
904 {
905         if (!cpufreq_global_kobject_usage++)
906                 return kobject_add(cpufreq_global_kobject,
907                                 &cpu_subsys.dev_root->kobj, "%s", "cpufreq");
908
909         return 0;
910 }
911 EXPORT_SYMBOL(cpufreq_get_global_kobject);
912
913 void cpufreq_put_global_kobject(void)
914 {
915         if (!--cpufreq_global_kobject_usage)
916                 kobject_del(cpufreq_global_kobject);
917 }
918 EXPORT_SYMBOL(cpufreq_put_global_kobject);
919
920 int cpufreq_sysfs_create_file(const struct attribute *attr)
921 {
922         int ret = cpufreq_get_global_kobject();
923
924         if (!ret) {
925                 ret = sysfs_create_file(cpufreq_global_kobject, attr);
926                 if (ret)
927                         cpufreq_put_global_kobject();
928         }
929
930         return ret;
931 }
932 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
933
934 void cpufreq_sysfs_remove_file(const struct attribute *attr)
935 {
936         sysfs_remove_file(cpufreq_global_kobject, attr);
937         cpufreq_put_global_kobject();
938 }
939 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
940
941 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
942 {
943         struct device *cpu_dev;
944
945         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
946
947         if (!policy)
948                 return 0;
949
950         cpu_dev = get_cpu_device(cpu);
951         if (WARN_ON(!cpu_dev))
952                 return 0;
953
954         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
955 }
956
957 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
958 {
959         struct device *cpu_dev;
960
961         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
962
963         cpu_dev = get_cpu_device(cpu);
964         if (WARN_ON(!cpu_dev))
965                 return;
966
967         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
968 }
969
970 /* Add/remove symlinks for all related CPUs */
971 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
972 {
973         unsigned int j;
974         int ret = 0;
975
976         /* Some related CPUs might not be present (physically hotplugged) */
977         for_each_cpu(j, policy->real_cpus) {
978                 if (j == policy->kobj_cpu)
979                         continue;
980
981                 ret = add_cpu_dev_symlink(policy, j);
982                 if (ret)
983                         break;
984         }
985
986         return ret;
987 }
988
989 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
990 {
991         unsigned int j;
992
993         /* Some related CPUs might not be present (physically hotplugged) */
994         for_each_cpu(j, policy->real_cpus) {
995                 if (j == policy->kobj_cpu)
996                         continue;
997
998                 remove_cpu_dev_symlink(policy, j);
999         }
1000 }
1001
1002 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1003 {
1004         struct freq_attr **drv_attr;
1005         int ret = 0;
1006
1007         /* set up files for this cpu device */
1008         drv_attr = cpufreq_driver->attr;
1009         while (drv_attr && *drv_attr) {
1010                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1011                 if (ret)
1012                         return ret;
1013                 drv_attr++;
1014         }
1015         if (cpufreq_driver->get) {
1016                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1017                 if (ret)
1018                         return ret;
1019         }
1020
1021         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1022         if (ret)
1023                 return ret;
1024
1025         if (cpufreq_driver->bios_limit) {
1026                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1027                 if (ret)
1028                         return ret;
1029         }
1030
1031         return cpufreq_add_dev_symlink(policy);
1032 }
1033
1034 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1035 {
1036         struct cpufreq_governor *gov = NULL;
1037         struct cpufreq_policy new_policy;
1038
1039         memcpy(&new_policy, policy, sizeof(*policy));
1040
1041         /* Update governor of new_policy to the governor used before hotplug */
1042         gov = find_governor(policy->last_governor);
1043         if (gov)
1044                 pr_debug("Restoring governor %s for cpu %d\n",
1045                                 policy->governor->name, policy->cpu);
1046         else
1047                 gov = CPUFREQ_DEFAULT_GOVERNOR;
1048
1049         new_policy.governor = gov;
1050
1051         /* Use the default policy if its valid. */
1052         if (cpufreq_driver->setpolicy)
1053                 cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
1054
1055         /* set default policy */
1056         return cpufreq_set_policy(policy, &new_policy);
1057 }
1058
1059 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1060 {
1061         int ret = 0;
1062
1063         /* Has this CPU been taken care of already? */
1064         if (cpumask_test_cpu(cpu, policy->cpus))
1065                 return 0;
1066
1067         if (has_target()) {
1068                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1069                 if (ret) {
1070                         pr_err("%s: Failed to stop governor\n", __func__);
1071                         return ret;
1072                 }
1073         }
1074
1075         down_write(&policy->rwsem);
1076         cpumask_set_cpu(cpu, policy->cpus);
1077         up_write(&policy->rwsem);
1078
1079         if (has_target()) {
1080                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1081                 if (!ret)
1082                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1083
1084                 if (ret) {
1085                         pr_err("%s: Failed to start governor\n", __func__);
1086                         return ret;
1087                 }
1088         }
1089
1090         return 0;
1091 }
1092
1093 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1094 {
1095         struct device *dev = get_cpu_device(cpu);
1096         struct cpufreq_policy *policy;
1097         int ret;
1098
1099         if (WARN_ON(!dev))
1100                 return NULL;
1101
1102         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1103         if (!policy)
1104                 return NULL;
1105
1106         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1107                 goto err_free_policy;
1108
1109         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1110                 goto err_free_cpumask;
1111
1112         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1113                 goto err_free_rcpumask;
1114
1115         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &dev->kobj,
1116                                    "cpufreq");
1117         if (ret) {
1118                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1119                 goto err_free_real_cpus;
1120         }
1121
1122         INIT_LIST_HEAD(&policy->policy_list);
1123         init_rwsem(&policy->rwsem);
1124         spin_lock_init(&policy->transition_lock);
1125         init_waitqueue_head(&policy->transition_wait);
1126         init_completion(&policy->kobj_unregister);
1127         INIT_WORK(&policy->update, handle_update);
1128
1129         policy->cpu = cpu;
1130
1131         /* Set this once on allocation */
1132         policy->kobj_cpu = cpu;
1133
1134         return policy;
1135
1136 err_free_real_cpus:
1137         free_cpumask_var(policy->real_cpus);
1138 err_free_rcpumask:
1139         free_cpumask_var(policy->related_cpus);
1140 err_free_cpumask:
1141         free_cpumask_var(policy->cpus);
1142 err_free_policy:
1143         kfree(policy);
1144
1145         return NULL;
1146 }
1147
1148 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1149 {
1150         struct kobject *kobj;
1151         struct completion *cmp;
1152
1153         if (notify)
1154                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1155                                              CPUFREQ_REMOVE_POLICY, policy);
1156
1157         down_write(&policy->rwsem);
1158         cpufreq_remove_dev_symlink(policy);
1159         kobj = &policy->kobj;
1160         cmp = &policy->kobj_unregister;
1161         up_write(&policy->rwsem);
1162         kobject_put(kobj);
1163
1164         /*
1165          * We need to make sure that the underlying kobj is
1166          * actually not referenced anymore by anybody before we
1167          * proceed with unloading.
1168          */
1169         pr_debug("waiting for dropping of refcount\n");
1170         wait_for_completion(cmp);
1171         pr_debug("wait complete\n");
1172 }
1173
1174 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1175 {
1176         unsigned long flags;
1177         int cpu;
1178
1179         /* Remove policy from list */
1180         write_lock_irqsave(&cpufreq_driver_lock, flags);
1181         list_del(&policy->policy_list);
1182
1183         for_each_cpu(cpu, policy->related_cpus)
1184                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1185         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1186
1187         cpufreq_policy_put_kobj(policy, notify);
1188         free_cpumask_var(policy->real_cpus);
1189         free_cpumask_var(policy->related_cpus);
1190         free_cpumask_var(policy->cpus);
1191         kfree(policy);
1192 }
1193
1194 static int cpufreq_online(unsigned int cpu)
1195 {
1196         struct cpufreq_policy *policy;
1197         bool new_policy;
1198         unsigned long flags;
1199         unsigned int j;
1200         int ret;
1201
1202         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1203
1204         /* Check if this CPU already has a policy to manage it */
1205         policy = per_cpu(cpufreq_cpu_data, cpu);
1206         if (policy) {
1207                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1208                 if (!policy_is_inactive(policy))
1209                         return cpufreq_add_policy_cpu(policy, cpu);
1210
1211                 /* This is the only online CPU for the policy.  Start over. */
1212                 new_policy = false;
1213                 down_write(&policy->rwsem);
1214                 policy->cpu = cpu;
1215                 policy->governor = NULL;
1216                 up_write(&policy->rwsem);
1217         } else {
1218                 new_policy = true;
1219                 policy = cpufreq_policy_alloc(cpu);
1220                 if (!policy)
1221                         return -ENOMEM;
1222         }
1223
1224         cpumask_copy(policy->cpus, cpumask_of(cpu));
1225
1226         /* call driver. From then on the cpufreq must be able
1227          * to accept all calls to ->verify and ->setpolicy for this CPU
1228          */
1229         ret = cpufreq_driver->init(policy);
1230         if (ret) {
1231                 pr_debug("initialization failed\n");
1232                 goto out_free_policy;
1233         }
1234
1235         down_write(&policy->rwsem);
1236
1237         if (new_policy) {
1238                 /* related_cpus should at least include policy->cpus. */
1239                 cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1240                 /* Remember CPUs present at the policy creation time. */
1241                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1242         }
1243
1244         /*
1245          * affected cpus must always be the one, which are online. We aren't
1246          * managing offline cpus here.
1247          */
1248         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1249
1250         if (new_policy) {
1251                 policy->user_policy.min = policy->min;
1252                 policy->user_policy.max = policy->max;
1253
1254                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1255                 for_each_cpu(j, policy->related_cpus)
1256                         per_cpu(cpufreq_cpu_data, j) = policy;
1257                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1258         }
1259
1260         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1261                 policy->cur = cpufreq_driver->get(policy->cpu);
1262                 if (!policy->cur) {
1263                         pr_err("%s: ->get() failed\n", __func__);
1264                         goto out_exit_policy;
1265                 }
1266         }
1267
1268         /*
1269          * Sometimes boot loaders set CPU frequency to a value outside of
1270          * frequency table present with cpufreq core. In such cases CPU might be
1271          * unstable if it has to run on that frequency for long duration of time
1272          * and so its better to set it to a frequency which is specified in
1273          * freq-table. This also makes cpufreq stats inconsistent as
1274          * cpufreq-stats would fail to register because current frequency of CPU
1275          * isn't found in freq-table.
1276          *
1277          * Because we don't want this change to effect boot process badly, we go
1278          * for the next freq which is >= policy->cur ('cur' must be set by now,
1279          * otherwise we will end up setting freq to lowest of the table as 'cur'
1280          * is initialized to zero).
1281          *
1282          * We are passing target-freq as "policy->cur - 1" otherwise
1283          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1284          * equal to target-freq.
1285          */
1286         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1287             && has_target()) {
1288                 /* Are we running at unknown frequency ? */
1289                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1290                 if (ret == -EINVAL) {
1291                         /* Warn user and fix it */
1292                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1293                                 __func__, policy->cpu, policy->cur);
1294                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1295                                 CPUFREQ_RELATION_L);
1296
1297                         /*
1298                          * Reaching here after boot in a few seconds may not
1299                          * mean that system will remain stable at "unknown"
1300                          * frequency for longer duration. Hence, a BUG_ON().
1301                          */
1302                         BUG_ON(ret);
1303                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1304                                 __func__, policy->cpu, policy->cur);
1305                 }
1306         }
1307
1308         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1309                                      CPUFREQ_START, policy);
1310
1311         if (new_policy) {
1312                 ret = cpufreq_add_dev_interface(policy);
1313                 if (ret)
1314                         goto out_exit_policy;
1315                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1316                                 CPUFREQ_CREATE_POLICY, policy);
1317
1318                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1319                 list_add(&policy->policy_list, &cpufreq_policy_list);
1320                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1321         }
1322
1323         ret = cpufreq_init_policy(policy);
1324         if (ret) {
1325                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1326                        __func__, cpu, ret);
1327                 /* cpufreq_policy_free() will notify based on this */
1328                 new_policy = false;
1329                 goto out_exit_policy;
1330         }
1331
1332         if (new_policy) {
1333                 policy->user_policy.policy = policy->policy;
1334                 policy->user_policy.governor = policy->governor;
1335         }
1336         up_write(&policy->rwsem);
1337
1338         kobject_uevent(&policy->kobj, KOBJ_ADD);
1339
1340         /* Callback for handling stuff after policy is ready */
1341         if (cpufreq_driver->ready)
1342                 cpufreq_driver->ready(policy);
1343
1344         pr_debug("initialization complete\n");
1345
1346         return 0;
1347
1348 out_exit_policy:
1349         up_write(&policy->rwsem);
1350
1351         if (cpufreq_driver->exit)
1352                 cpufreq_driver->exit(policy);
1353 out_free_policy:
1354         cpufreq_policy_free(policy, !new_policy);
1355         return ret;
1356 }
1357
1358 /**
1359  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1360  * @dev: CPU device.
1361  * @sif: Subsystem interface structure pointer (not used)
1362  */
1363 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1364 {
1365         unsigned cpu = dev->id;
1366         int ret;
1367
1368         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1369
1370         if (cpu_online(cpu)) {
1371                 ret = cpufreq_online(cpu);
1372         } else {
1373                 /*
1374                  * A hotplug notifier will follow and we will handle it as CPU
1375                  * online then.  For now, just create the sysfs link, unless
1376                  * there is no policy or the link is already present.
1377                  */
1378                 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1379
1380                 ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1381                         ? add_cpu_dev_symlink(policy, cpu) : 0;
1382         }
1383
1384         return ret;
1385 }
1386
1387 static void cpufreq_offline_prepare(unsigned int cpu)
1388 {
1389         struct cpufreq_policy *policy;
1390
1391         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1392
1393         policy = cpufreq_cpu_get_raw(cpu);
1394         if (!policy) {
1395                 pr_debug("%s: No cpu_data found\n", __func__);
1396                 return;
1397         }
1398
1399         if (has_target()) {
1400                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1401                 if (ret)
1402                         pr_err("%s: Failed to stop governor\n", __func__);
1403         }
1404
1405         down_write(&policy->rwsem);
1406         cpumask_clear_cpu(cpu, policy->cpus);
1407
1408         if (policy_is_inactive(policy)) {
1409                 if (has_target())
1410                         strncpy(policy->last_governor, policy->governor->name,
1411                                 CPUFREQ_NAME_LEN);
1412         } else if (cpu == policy->cpu) {
1413                 /* Nominate new CPU */
1414                 policy->cpu = cpumask_any(policy->cpus);
1415         }
1416         up_write(&policy->rwsem);
1417
1418         /* Start governor again for active policy */
1419         if (!policy_is_inactive(policy)) {
1420                 if (has_target()) {
1421                         int ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1422                         if (!ret)
1423                                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1424
1425                         if (ret)
1426                                 pr_err("%s: Failed to start governor\n", __func__);
1427                 }
1428         } else if (cpufreq_driver->stop_cpu) {
1429                 cpufreq_driver->stop_cpu(policy);
1430         }
1431 }
1432
1433 static void cpufreq_offline_finish(unsigned int cpu)
1434 {
1435         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1436
1437         if (!policy) {
1438                 pr_debug("%s: No cpu_data found\n", __func__);
1439                 return;
1440         }
1441
1442         /* Only proceed for inactive policies */
1443         if (!policy_is_inactive(policy))
1444                 return;
1445
1446         /* If cpu is last user of policy, free policy */
1447         if (has_target()) {
1448                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1449                 if (ret)
1450                         pr_err("%s: Failed to exit governor\n", __func__);
1451         }
1452
1453         /*
1454          * Perform the ->exit() even during light-weight tear-down,
1455          * since this is a core component, and is essential for the
1456          * subsequent light-weight ->init() to succeed.
1457          */
1458         if (cpufreq_driver->exit)
1459                 cpufreq_driver->exit(policy);
1460 }
1461
1462 /**
1463  * cpufreq_remove_dev - remove a CPU device
1464  *
1465  * Removes the cpufreq interface for a CPU device.
1466  */
1467 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1468 {
1469         unsigned int cpu = dev->id;
1470         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1471
1472         if (!policy)
1473                 return 0;
1474
1475         if (cpu_online(cpu)) {
1476                 cpufreq_offline_prepare(cpu);
1477                 cpufreq_offline_finish(cpu);
1478         }
1479
1480         cpumask_clear_cpu(cpu, policy->real_cpus);
1481
1482         if (cpumask_empty(policy->real_cpus)) {
1483                 cpufreq_policy_free(policy, true);
1484                 return 0;
1485         }
1486
1487         if (cpu != policy->kobj_cpu) {
1488                 remove_cpu_dev_symlink(policy, cpu);
1489         } else {
1490                 /*
1491                  * The CPU owning the policy object is going away.  Move it to
1492                  * another suitable CPU.
1493                  */
1494                 unsigned int new_cpu = cpumask_first(policy->real_cpus);
1495                 struct device *new_dev = get_cpu_device(new_cpu);
1496
1497                 dev_dbg(dev, "%s: Moving policy object to CPU%u\n", __func__, new_cpu);
1498
1499                 sysfs_remove_link(&new_dev->kobj, "cpufreq");
1500                 policy->kobj_cpu = new_cpu;
1501                 WARN_ON(kobject_move(&policy->kobj, &new_dev->kobj));
1502         }
1503
1504         return 0;
1505 }
1506
1507 static void handle_update(struct work_struct *work)
1508 {
1509         struct cpufreq_policy *policy =
1510                 container_of(work, struct cpufreq_policy, update);
1511         unsigned int cpu = policy->cpu;
1512         pr_debug("handle_update for cpu %u called\n", cpu);
1513         cpufreq_update_policy(cpu);
1514 }
1515
1516 /**
1517  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1518  *      in deep trouble.
1519  *      @policy: policy managing CPUs
1520  *      @new_freq: CPU frequency the CPU actually runs at
1521  *
1522  *      We adjust to current frequency first, and need to clean up later.
1523  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1524  */
1525 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1526                                 unsigned int new_freq)
1527 {
1528         struct cpufreq_freqs freqs;
1529
1530         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1531                  policy->cur, new_freq);
1532
1533         freqs.old = policy->cur;
1534         freqs.new = new_freq;
1535
1536         cpufreq_freq_transition_begin(policy, &freqs);
1537         cpufreq_freq_transition_end(policy, &freqs, 0);
1538 }
1539
1540 /**
1541  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1542  * @cpu: CPU number
1543  *
1544  * This is the last known freq, without actually getting it from the driver.
1545  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1546  */
1547 unsigned int cpufreq_quick_get(unsigned int cpu)
1548 {
1549         struct cpufreq_policy *policy;
1550         unsigned int ret_freq = 0;
1551
1552         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1553                 return cpufreq_driver->get(cpu);
1554
1555         policy = cpufreq_cpu_get(cpu);
1556         if (policy) {
1557                 ret_freq = policy->cur;
1558                 cpufreq_cpu_put(policy);
1559         }
1560
1561         return ret_freq;
1562 }
1563 EXPORT_SYMBOL(cpufreq_quick_get);
1564
1565 /**
1566  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1567  * @cpu: CPU number
1568  *
1569  * Just return the max possible frequency for a given CPU.
1570  */
1571 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1572 {
1573         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1574         unsigned int ret_freq = 0;
1575
1576         if (policy) {
1577                 ret_freq = policy->max;
1578                 cpufreq_cpu_put(policy);
1579         }
1580
1581         return ret_freq;
1582 }
1583 EXPORT_SYMBOL(cpufreq_quick_get_max);
1584
1585 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1586 {
1587         unsigned int ret_freq = 0;
1588
1589         if (!cpufreq_driver->get)
1590                 return ret_freq;
1591
1592         ret_freq = cpufreq_driver->get(policy->cpu);
1593
1594         /* Updating inactive policies is invalid, so avoid doing that. */
1595         if (unlikely(policy_is_inactive(policy)))
1596                 return ret_freq;
1597
1598         if (ret_freq && policy->cur &&
1599                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1600                 /* verify no discrepancy between actual and
1601                                         saved value exists */
1602                 if (unlikely(ret_freq != policy->cur)) {
1603                         cpufreq_out_of_sync(policy, ret_freq);
1604                         schedule_work(&policy->update);
1605                 }
1606         }
1607
1608         return ret_freq;
1609 }
1610
1611 /**
1612  * cpufreq_get - get the current CPU frequency (in kHz)
1613  * @cpu: CPU number
1614  *
1615  * Get the CPU current (static) CPU frequency
1616  */
1617 unsigned int cpufreq_get(unsigned int cpu)
1618 {
1619         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1620         unsigned int ret_freq = 0;
1621
1622         if (policy) {
1623                 down_read(&policy->rwsem);
1624                 ret_freq = __cpufreq_get(policy);
1625                 up_read(&policy->rwsem);
1626
1627                 cpufreq_cpu_put(policy);
1628         }
1629
1630         return ret_freq;
1631 }
1632 EXPORT_SYMBOL(cpufreq_get);
1633
1634 static struct subsys_interface cpufreq_interface = {
1635         .name           = "cpufreq",
1636         .subsys         = &cpu_subsys,
1637         .add_dev        = cpufreq_add_dev,
1638         .remove_dev     = cpufreq_remove_dev,
1639 };
1640
1641 /*
1642  * In case platform wants some specific frequency to be configured
1643  * during suspend..
1644  */
1645 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1646 {
1647         int ret;
1648
1649         if (!policy->suspend_freq) {
1650                 pr_err("%s: suspend_freq can't be zero\n", __func__);
1651                 return -EINVAL;
1652         }
1653
1654         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1655                         policy->suspend_freq);
1656
1657         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1658                         CPUFREQ_RELATION_H);
1659         if (ret)
1660                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1661                                 __func__, policy->suspend_freq, ret);
1662
1663         return ret;
1664 }
1665 EXPORT_SYMBOL(cpufreq_generic_suspend);
1666
1667 /**
1668  * cpufreq_suspend() - Suspend CPUFreq governors
1669  *
1670  * Called during system wide Suspend/Hibernate cycles for suspending governors
1671  * as some platforms can't change frequency after this point in suspend cycle.
1672  * Because some of the devices (like: i2c, regulators, etc) they use for
1673  * changing frequency are suspended quickly after this point.
1674  */
1675 void cpufreq_suspend(void)
1676 {
1677         struct cpufreq_policy *policy;
1678
1679         if (!cpufreq_driver)
1680                 return;
1681
1682         if (!has_target())
1683                 goto suspend;
1684
1685         pr_debug("%s: Suspending Governors\n", __func__);
1686
1687         for_each_active_policy(policy) {
1688                 if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1689                         pr_err("%s: Failed to stop governor for policy: %p\n",
1690                                 __func__, policy);
1691                 else if (cpufreq_driver->suspend
1692                     && cpufreq_driver->suspend(policy))
1693                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1694                                 policy);
1695         }
1696
1697 suspend:
1698         cpufreq_suspended = true;
1699 }
1700
1701 /**
1702  * cpufreq_resume() - Resume CPUFreq governors
1703  *
1704  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1705  * are suspended with cpufreq_suspend().
1706  */
1707 void cpufreq_resume(void)
1708 {
1709         struct cpufreq_policy *policy;
1710
1711         if (!cpufreq_driver)
1712                 return;
1713
1714         cpufreq_suspended = false;
1715
1716         if (!has_target())
1717                 return;
1718
1719         pr_debug("%s: Resuming Governors\n", __func__);
1720
1721         for_each_active_policy(policy) {
1722                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1723                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1724                                 policy);
1725                 else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1726                     || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1727                         pr_err("%s: Failed to start governor for policy: %p\n",
1728                                 __func__, policy);
1729         }
1730
1731         /*
1732          * schedule call cpufreq_update_policy() for first-online CPU, as that
1733          * wouldn't be hotplugged-out on suspend. It will verify that the
1734          * current freq is in sync with what we believe it to be.
1735          */
1736         policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1737         if (WARN_ON(!policy))
1738                 return;
1739
1740         schedule_work(&policy->update);
1741 }
1742
1743 /**
1744  *      cpufreq_get_current_driver - return current driver's name
1745  *
1746  *      Return the name string of the currently loaded cpufreq driver
1747  *      or NULL, if none.
1748  */
1749 const char *cpufreq_get_current_driver(void)
1750 {
1751         if (cpufreq_driver)
1752                 return cpufreq_driver->name;
1753
1754         return NULL;
1755 }
1756 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1757
1758 /**
1759  *      cpufreq_get_driver_data - return current driver data
1760  *
1761  *      Return the private data of the currently loaded cpufreq
1762  *      driver, or NULL if no cpufreq driver is loaded.
1763  */
1764 void *cpufreq_get_driver_data(void)
1765 {
1766         if (cpufreq_driver)
1767                 return cpufreq_driver->driver_data;
1768
1769         return NULL;
1770 }
1771 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1772
1773 /*********************************************************************
1774  *                     NOTIFIER LISTS INTERFACE                      *
1775  *********************************************************************/
1776
1777 /**
1778  *      cpufreq_register_notifier - register a driver with cpufreq
1779  *      @nb: notifier function to register
1780  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1781  *
1782  *      Add a driver to one of two lists: either a list of drivers that
1783  *      are notified about clock rate changes (once before and once after
1784  *      the transition), or a list of drivers that are notified about
1785  *      changes in cpufreq policy.
1786  *
1787  *      This function may sleep, and has the same return conditions as
1788  *      blocking_notifier_chain_register.
1789  */
1790 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1791 {
1792         int ret;
1793
1794         if (cpufreq_disabled())
1795                 return -EINVAL;
1796
1797         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1798
1799         switch (list) {
1800         case CPUFREQ_TRANSITION_NOTIFIER:
1801                 ret = srcu_notifier_chain_register(
1802                                 &cpufreq_transition_notifier_list, nb);
1803                 break;
1804         case CPUFREQ_POLICY_NOTIFIER:
1805                 ret = blocking_notifier_chain_register(
1806                                 &cpufreq_policy_notifier_list, nb);
1807                 break;
1808         default:
1809                 ret = -EINVAL;
1810         }
1811
1812         return ret;
1813 }
1814 EXPORT_SYMBOL(cpufreq_register_notifier);
1815
1816 /**
1817  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1818  *      @nb: notifier block to be unregistered
1819  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1820  *
1821  *      Remove a driver from the CPU frequency notifier list.
1822  *
1823  *      This function may sleep, and has the same return conditions as
1824  *      blocking_notifier_chain_unregister.
1825  */
1826 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1827 {
1828         int ret;
1829
1830         if (cpufreq_disabled())
1831                 return -EINVAL;
1832
1833         switch (list) {
1834         case CPUFREQ_TRANSITION_NOTIFIER:
1835                 ret = srcu_notifier_chain_unregister(
1836                                 &cpufreq_transition_notifier_list, nb);
1837                 break;
1838         case CPUFREQ_POLICY_NOTIFIER:
1839                 ret = blocking_notifier_chain_unregister(
1840                                 &cpufreq_policy_notifier_list, nb);
1841                 break;
1842         default:
1843                 ret = -EINVAL;
1844         }
1845
1846         return ret;
1847 }
1848 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1849
1850
1851 /*********************************************************************
1852  *                              GOVERNORS                            *
1853  *********************************************************************/
1854
1855 /* Must set freqs->new to intermediate frequency */
1856 static int __target_intermediate(struct cpufreq_policy *policy,
1857                                  struct cpufreq_freqs *freqs, int index)
1858 {
1859         int ret;
1860
1861         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1862
1863         /* We don't need to switch to intermediate freq */
1864         if (!freqs->new)
1865                 return 0;
1866
1867         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1868                  __func__, policy->cpu, freqs->old, freqs->new);
1869
1870         cpufreq_freq_transition_begin(policy, freqs);
1871         ret = cpufreq_driver->target_intermediate(policy, index);
1872         cpufreq_freq_transition_end(policy, freqs, ret);
1873
1874         if (ret)
1875                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1876                        __func__, ret);
1877
1878         return ret;
1879 }
1880
1881 static int __target_index(struct cpufreq_policy *policy,
1882                           struct cpufreq_frequency_table *freq_table, int index)
1883 {
1884         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1885         unsigned int intermediate_freq = 0;
1886         int retval = -EINVAL;
1887         bool notify;
1888
1889         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1890         if (notify) {
1891                 /* Handle switching to intermediate frequency */
1892                 if (cpufreq_driver->get_intermediate) {
1893                         retval = __target_intermediate(policy, &freqs, index);
1894                         if (retval)
1895                                 return retval;
1896
1897                         intermediate_freq = freqs.new;
1898                         /* Set old freq to intermediate */
1899                         if (intermediate_freq)
1900                                 freqs.old = freqs.new;
1901                 }
1902
1903                 freqs.new = freq_table[index].frequency;
1904                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1905                          __func__, policy->cpu, freqs.old, freqs.new);
1906
1907                 cpufreq_freq_transition_begin(policy, &freqs);
1908         }
1909
1910         retval = cpufreq_driver->target_index(policy, index);
1911         if (retval)
1912                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1913                        retval);
1914
1915         if (notify) {
1916                 cpufreq_freq_transition_end(policy, &freqs, retval);
1917
1918                 /*
1919                  * Failed after setting to intermediate freq? Driver should have
1920                  * reverted back to initial frequency and so should we. Check
1921                  * here for intermediate_freq instead of get_intermediate, in
1922                  * case we haven't switched to intermediate freq at all.
1923                  */
1924                 if (unlikely(retval && intermediate_freq)) {
1925                         freqs.old = intermediate_freq;
1926                         freqs.new = policy->restore_freq;
1927                         cpufreq_freq_transition_begin(policy, &freqs);
1928                         cpufreq_freq_transition_end(policy, &freqs, 0);
1929                 }
1930         }
1931
1932         return retval;
1933 }
1934
1935 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1936                             unsigned int target_freq,
1937                             unsigned int relation)
1938 {
1939         unsigned int old_target_freq = target_freq;
1940         int retval = -EINVAL;
1941
1942         if (cpufreq_disabled())
1943                 return -ENODEV;
1944
1945         /* Make sure that target_freq is within supported range */
1946         if (target_freq > policy->max)
1947                 target_freq = policy->max;
1948         if (target_freq < policy->min)
1949                 target_freq = policy->min;
1950
1951         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1952                  policy->cpu, target_freq, relation, old_target_freq);
1953
1954         /*
1955          * This might look like a redundant call as we are checking it again
1956          * after finding index. But it is left intentionally for cases where
1957          * exactly same freq is called again and so we can save on few function
1958          * calls.
1959          */
1960         if (target_freq == policy->cur)
1961                 return 0;
1962
1963         /* Save last value to restore later on errors */
1964         policy->restore_freq = policy->cur;
1965
1966         if (cpufreq_driver->target)
1967                 retval = cpufreq_driver->target(policy, target_freq, relation);
1968         else if (cpufreq_driver->target_index) {
1969                 struct cpufreq_frequency_table *freq_table;
1970                 int index;
1971
1972                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1973                 if (unlikely(!freq_table)) {
1974                         pr_err("%s: Unable to find freq_table\n", __func__);
1975                         goto out;
1976                 }
1977
1978                 retval = cpufreq_frequency_table_target(policy, freq_table,
1979                                 target_freq, relation, &index);
1980                 if (unlikely(retval)) {
1981                         pr_err("%s: Unable to find matching freq\n", __func__);
1982                         goto out;
1983                 }
1984
1985                 if (freq_table[index].frequency == policy->cur) {
1986                         retval = 0;
1987                         goto out;
1988                 }
1989
1990                 retval = __target_index(policy, freq_table, index);
1991         }
1992
1993 out:
1994         return retval;
1995 }
1996 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1997
1998 int cpufreq_driver_target(struct cpufreq_policy *policy,
1999                           unsigned int target_freq,
2000                           unsigned int relation)
2001 {
2002         int ret = -EINVAL;
2003
2004         down_write(&policy->rwsem);
2005
2006         ret = __cpufreq_driver_target(policy, target_freq, relation);
2007
2008         up_write(&policy->rwsem);
2009
2010         return ret;
2011 }
2012 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2013
2014 static int __cpufreq_governor(struct cpufreq_policy *policy,
2015                                         unsigned int event)
2016 {
2017         int ret;
2018
2019         /* Only must be defined when default governor is known to have latency
2020            restrictions, like e.g. conservative or ondemand.
2021            That this is the case is already ensured in Kconfig
2022         */
2023 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
2024         struct cpufreq_governor *gov = &cpufreq_gov_performance;
2025 #else
2026         struct cpufreq_governor *gov = NULL;
2027 #endif
2028
2029         /* Don't start any governor operations if we are entering suspend */
2030         if (cpufreq_suspended)
2031                 return 0;
2032         /*
2033          * Governor might not be initiated here if ACPI _PPC changed
2034          * notification happened, so check it.
2035          */
2036         if (!policy->governor)
2037                 return -EINVAL;
2038
2039         if (policy->governor->max_transition_latency &&
2040             policy->cpuinfo.transition_latency >
2041             policy->governor->max_transition_latency) {
2042                 if (!gov)
2043                         return -EINVAL;
2044                 else {
2045                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2046                                 policy->governor->name, gov->name);
2047                         policy->governor = gov;
2048                 }
2049         }
2050
2051         if (event == CPUFREQ_GOV_POLICY_INIT)
2052                 if (!try_module_get(policy->governor->owner))
2053                         return -EINVAL;
2054
2055         pr_debug("__cpufreq_governor for CPU %u, event %u\n",
2056                  policy->cpu, event);
2057
2058         mutex_lock(&cpufreq_governor_lock);
2059         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2060             || (!policy->governor_enabled
2061             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2062                 mutex_unlock(&cpufreq_governor_lock);
2063                 return -EBUSY;
2064         }
2065
2066         if (event == CPUFREQ_GOV_STOP)
2067                 policy->governor_enabled = false;
2068         else if (event == CPUFREQ_GOV_START)
2069                 policy->governor_enabled = true;
2070
2071         mutex_unlock(&cpufreq_governor_lock);
2072
2073         ret = policy->governor->governor(policy, event);
2074
2075         if (!ret) {
2076                 if (event == CPUFREQ_GOV_POLICY_INIT)
2077                         policy->governor->initialized++;
2078                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
2079                         policy->governor->initialized--;
2080         } else {
2081                 /* Restore original values */
2082                 mutex_lock(&cpufreq_governor_lock);
2083                 if (event == CPUFREQ_GOV_STOP)
2084                         policy->governor_enabled = true;
2085                 else if (event == CPUFREQ_GOV_START)
2086                         policy->governor_enabled = false;
2087                 mutex_unlock(&cpufreq_governor_lock);
2088         }
2089
2090         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2091                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2092                 module_put(policy->governor->owner);
2093
2094         return ret;
2095 }
2096
2097 int cpufreq_register_governor(struct cpufreq_governor *governor)
2098 {
2099         int err;
2100
2101         if (!governor)
2102                 return -EINVAL;
2103
2104         if (cpufreq_disabled())
2105                 return -ENODEV;
2106
2107         mutex_lock(&cpufreq_governor_mutex);
2108
2109         governor->initialized = 0;
2110         err = -EBUSY;
2111         if (!find_governor(governor->name)) {
2112                 err = 0;
2113                 list_add(&governor->governor_list, &cpufreq_governor_list);
2114         }
2115
2116         mutex_unlock(&cpufreq_governor_mutex);
2117         return err;
2118 }
2119 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2120
2121 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2122 {
2123         struct cpufreq_policy *policy;
2124         unsigned long flags;
2125
2126         if (!governor)
2127                 return;
2128
2129         if (cpufreq_disabled())
2130                 return;
2131
2132         /* clear last_governor for all inactive policies */
2133         read_lock_irqsave(&cpufreq_driver_lock, flags);
2134         for_each_inactive_policy(policy) {
2135                 if (!strcmp(policy->last_governor, governor->name)) {
2136                         policy->governor = NULL;
2137                         strcpy(policy->last_governor, "\0");
2138                 }
2139         }
2140         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2141
2142         mutex_lock(&cpufreq_governor_mutex);
2143         list_del(&governor->governor_list);
2144         mutex_unlock(&cpufreq_governor_mutex);
2145         return;
2146 }
2147 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2148
2149
2150 /*********************************************************************
2151  *                          POLICY INTERFACE                         *
2152  *********************************************************************/
2153
2154 /**
2155  * cpufreq_get_policy - get the current cpufreq_policy
2156  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2157  *      is written
2158  *
2159  * Reads the current cpufreq policy.
2160  */
2161 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2162 {
2163         struct cpufreq_policy *cpu_policy;
2164         if (!policy)
2165                 return -EINVAL;
2166
2167         cpu_policy = cpufreq_cpu_get(cpu);
2168         if (!cpu_policy)
2169                 return -EINVAL;
2170
2171         memcpy(policy, cpu_policy, sizeof(*policy));
2172
2173         cpufreq_cpu_put(cpu_policy);
2174         return 0;
2175 }
2176 EXPORT_SYMBOL(cpufreq_get_policy);
2177
2178 /*
2179  * policy : current policy.
2180  * new_policy: policy to be set.
2181  */
2182 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2183                                 struct cpufreq_policy *new_policy)
2184 {
2185         struct cpufreq_governor *old_gov;
2186         int ret;
2187
2188         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2189                  new_policy->cpu, new_policy->min, new_policy->max);
2190
2191         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2192
2193         /*
2194         * This check works well when we store new min/max freq attributes,
2195         * because new_policy is a copy of policy with one field updated.
2196         */
2197         if (new_policy->min > new_policy->max)
2198                 return -EINVAL;
2199
2200         /* verify the cpu speed can be set within this limit */
2201         ret = cpufreq_driver->verify(new_policy);
2202         if (ret)
2203                 return ret;
2204
2205         /* adjust if necessary - all reasons */
2206         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2207                         CPUFREQ_ADJUST, new_policy);
2208
2209         /*
2210          * verify the cpu speed can be set within this limit, which might be
2211          * different to the first one
2212          */
2213         ret = cpufreq_driver->verify(new_policy);
2214         if (ret)
2215                 return ret;
2216
2217         /* notification of the new policy */
2218         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2219                         CPUFREQ_NOTIFY, new_policy);
2220
2221         policy->min = new_policy->min;
2222         policy->max = new_policy->max;
2223
2224         pr_debug("new min and max freqs are %u - %u kHz\n",
2225                  policy->min, policy->max);
2226
2227         if (cpufreq_driver->setpolicy) {
2228                 policy->policy = new_policy->policy;
2229                 pr_debug("setting range\n");
2230                 return cpufreq_driver->setpolicy(new_policy);
2231         }
2232
2233         if (new_policy->governor == policy->governor)
2234                 goto out;
2235
2236         pr_debug("governor switch\n");
2237
2238         /* save old, working values */
2239         old_gov = policy->governor;
2240         /* end old governor */
2241         if (old_gov) {
2242                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2243                 if (ret) {
2244                         /* This can happen due to race with other operations */
2245                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2246                                  __func__, old_gov->name, ret);
2247                         return ret;
2248                 }
2249
2250                 up_write(&policy->rwsem);
2251                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2252                 down_write(&policy->rwsem);
2253
2254                 if (ret) {
2255                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2256                                __func__, old_gov->name, ret);
2257                         return ret;
2258                 }
2259         }
2260
2261         /* start new governor */
2262         policy->governor = new_policy->governor;
2263         ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2264         if (!ret) {
2265                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
2266                 if (!ret)
2267                         goto out;
2268
2269                 up_write(&policy->rwsem);
2270                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2271                 down_write(&policy->rwsem);
2272         }
2273
2274         /* new governor failed, so re-start old one */
2275         pr_debug("starting governor %s failed\n", policy->governor->name);
2276         if (old_gov) {
2277                 policy->governor = old_gov;
2278                 if (__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2279                         policy->governor = NULL;
2280                 else
2281                         __cpufreq_governor(policy, CPUFREQ_GOV_START);
2282         }
2283
2284         return ret;
2285
2286  out:
2287         pr_debug("governor: change or update limits\n");
2288         return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2289 }
2290
2291 /**
2292  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2293  *      @cpu: CPU which shall be re-evaluated
2294  *
2295  *      Useful for policy notifiers which have different necessities
2296  *      at different times.
2297  */
2298 int cpufreq_update_policy(unsigned int cpu)
2299 {
2300         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2301         struct cpufreq_policy new_policy;
2302         int ret;
2303
2304         if (!policy)
2305                 return -ENODEV;
2306
2307         down_write(&policy->rwsem);
2308
2309         pr_debug("updating policy for CPU %u\n", cpu);
2310         memcpy(&new_policy, policy, sizeof(*policy));
2311         new_policy.min = policy->user_policy.min;
2312         new_policy.max = policy->user_policy.max;
2313         new_policy.policy = policy->user_policy.policy;
2314         new_policy.governor = policy->user_policy.governor;
2315
2316         /*
2317          * BIOS might change freq behind our back
2318          * -> ask driver for current freq and notify governors about a change
2319          */
2320         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2321                 new_policy.cur = cpufreq_driver->get(cpu);
2322                 if (WARN_ON(!new_policy.cur)) {
2323                         ret = -EIO;
2324                         goto unlock;
2325                 }
2326
2327                 if (!policy->cur) {
2328                         pr_debug("Driver did not initialize current freq\n");
2329                         policy->cur = new_policy.cur;
2330                 } else {
2331                         if (policy->cur != new_policy.cur && has_target())
2332                                 cpufreq_out_of_sync(policy, new_policy.cur);
2333                 }
2334         }
2335
2336         ret = cpufreq_set_policy(policy, &new_policy);
2337
2338 unlock:
2339         up_write(&policy->rwsem);
2340
2341         cpufreq_cpu_put(policy);
2342         return ret;
2343 }
2344 EXPORT_SYMBOL(cpufreq_update_policy);
2345
2346 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2347                                         unsigned long action, void *hcpu)
2348 {
2349         unsigned int cpu = (unsigned long)hcpu;
2350
2351         switch (action & ~CPU_TASKS_FROZEN) {
2352         case CPU_ONLINE:
2353                 cpufreq_online(cpu);
2354                 break;
2355
2356         case CPU_DOWN_PREPARE:
2357                 cpufreq_offline_prepare(cpu);
2358                 break;
2359
2360         case CPU_POST_DEAD:
2361                 cpufreq_offline_finish(cpu);
2362                 break;
2363
2364         case CPU_DOWN_FAILED:
2365                 cpufreq_online(cpu);
2366                 break;
2367         }
2368         return NOTIFY_OK;
2369 }
2370
2371 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2372         .notifier_call = cpufreq_cpu_callback,
2373 };
2374
2375 /*********************************************************************
2376  *               BOOST                                               *
2377  *********************************************************************/
2378 static int cpufreq_boost_set_sw(int state)
2379 {
2380         struct cpufreq_frequency_table *freq_table;
2381         struct cpufreq_policy *policy;
2382         int ret = -EINVAL;
2383
2384         for_each_active_policy(policy) {
2385                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2386                 if (freq_table) {
2387                         ret = cpufreq_frequency_table_cpuinfo(policy,
2388                                                         freq_table);
2389                         if (ret) {
2390                                 pr_err("%s: Policy frequency update failed\n",
2391                                        __func__);
2392                                 break;
2393                         }
2394                         policy->user_policy.max = policy->max;
2395                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2396                 }
2397         }
2398
2399         return ret;
2400 }
2401
2402 int cpufreq_boost_trigger_state(int state)
2403 {
2404         unsigned long flags;
2405         int ret = 0;
2406
2407         if (cpufreq_driver->boost_enabled == state)
2408                 return 0;
2409
2410         write_lock_irqsave(&cpufreq_driver_lock, flags);
2411         cpufreq_driver->boost_enabled = state;
2412         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2413
2414         ret = cpufreq_driver->set_boost(state);
2415         if (ret) {
2416                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2417                 cpufreq_driver->boost_enabled = !state;
2418                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2419
2420                 pr_err("%s: Cannot %s BOOST\n",
2421                        __func__, state ? "enable" : "disable");
2422         }
2423
2424         return ret;
2425 }
2426
2427 int cpufreq_boost_supported(void)
2428 {
2429         if (likely(cpufreq_driver))
2430                 return cpufreq_driver->boost_supported;
2431
2432         return 0;
2433 }
2434 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2435
2436 int cpufreq_boost_enabled(void)
2437 {
2438         return cpufreq_driver->boost_enabled;
2439 }
2440 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2441
2442 /*********************************************************************
2443  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2444  *********************************************************************/
2445
2446 /**
2447  * cpufreq_register_driver - register a CPU Frequency driver
2448  * @driver_data: A struct cpufreq_driver containing the values#
2449  * submitted by the CPU Frequency driver.
2450  *
2451  * Registers a CPU Frequency driver to this core code. This code
2452  * returns zero on success, -EBUSY when another driver got here first
2453  * (and isn't unregistered in the meantime).
2454  *
2455  */
2456 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2457 {
2458         unsigned long flags;
2459         int ret;
2460
2461         if (cpufreq_disabled())
2462                 return -ENODEV;
2463
2464         if (!driver_data || !driver_data->verify || !driver_data->init ||
2465             !(driver_data->setpolicy || driver_data->target_index ||
2466                     driver_data->target) ||
2467              (driver_data->setpolicy && (driver_data->target_index ||
2468                     driver_data->target)) ||
2469              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2470                 return -EINVAL;
2471
2472         pr_debug("trying to register driver %s\n", driver_data->name);
2473
2474         /* Protect against concurrent CPU online/offline. */
2475         get_online_cpus();
2476
2477         write_lock_irqsave(&cpufreq_driver_lock, flags);
2478         if (cpufreq_driver) {
2479                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2480                 ret = -EEXIST;
2481                 goto out;
2482         }
2483         cpufreq_driver = driver_data;
2484         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2485
2486         if (driver_data->setpolicy)
2487                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2488
2489         if (cpufreq_boost_supported()) {
2490                 /*
2491                  * Check if driver provides function to enable boost -
2492                  * if not, use cpufreq_boost_set_sw as default
2493                  */
2494                 if (!cpufreq_driver->set_boost)
2495                         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2496
2497                 ret = cpufreq_sysfs_create_file(&boost.attr);
2498                 if (ret) {
2499                         pr_err("%s: cannot register global BOOST sysfs file\n",
2500                                __func__);
2501                         goto err_null_driver;
2502                 }
2503         }
2504
2505         ret = subsys_interface_register(&cpufreq_interface);
2506         if (ret)
2507                 goto err_boost_unreg;
2508
2509         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2510             list_empty(&cpufreq_policy_list)) {
2511                 /* if all ->init() calls failed, unregister */
2512                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2513                          driver_data->name);
2514                 goto err_if_unreg;
2515         }
2516
2517         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2518         pr_debug("driver %s up and running\n", driver_data->name);
2519
2520 out:
2521         put_online_cpus();
2522         return ret;
2523
2524 err_if_unreg:
2525         subsys_interface_unregister(&cpufreq_interface);
2526 err_boost_unreg:
2527         if (cpufreq_boost_supported())
2528                 cpufreq_sysfs_remove_file(&boost.attr);
2529 err_null_driver:
2530         write_lock_irqsave(&cpufreq_driver_lock, flags);
2531         cpufreq_driver = NULL;
2532         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2533         goto out;
2534 }
2535 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2536
2537 /**
2538  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2539  *
2540  * Unregister the current CPUFreq driver. Only call this if you have
2541  * the right to do so, i.e. if you have succeeded in initialising before!
2542  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2543  * currently not initialised.
2544  */
2545 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2546 {
2547         unsigned long flags;
2548
2549         if (!cpufreq_driver || (driver != cpufreq_driver))
2550                 return -EINVAL;
2551
2552         pr_debug("unregistering driver %s\n", driver->name);
2553
2554         /* Protect against concurrent cpu hotplug */
2555         get_online_cpus();
2556         subsys_interface_unregister(&cpufreq_interface);
2557         if (cpufreq_boost_supported())
2558                 cpufreq_sysfs_remove_file(&boost.attr);
2559
2560         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2561
2562         write_lock_irqsave(&cpufreq_driver_lock, flags);
2563
2564         cpufreq_driver = NULL;
2565
2566         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2567         put_online_cpus();
2568
2569         return 0;
2570 }
2571 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2572
2573 /*
2574  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2575  * or mutexes when secondary CPUs are halted.
2576  */
2577 static struct syscore_ops cpufreq_syscore_ops = {
2578         .shutdown = cpufreq_suspend,
2579 };
2580
2581 static int __init cpufreq_core_init(void)
2582 {
2583         if (cpufreq_disabled())
2584                 return -ENODEV;
2585
2586         cpufreq_global_kobject = kobject_create();
2587         BUG_ON(!cpufreq_global_kobject);
2588
2589         register_syscore_ops(&cpufreq_syscore_ops);
2590
2591         return 0;
2592 }
2593 core_initcall(cpufreq_core_init);