OSDN Git Service

Linux 4.4.187
[sagit-ice-cold/kernel_xiaomi_msm8998.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 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
242 {
243         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
244
245         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
246 }
247 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
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 kobj_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 kobj_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->setpolicy) {
524                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
525                         *policy = CPUFREQ_POLICY_PERFORMANCE;
526                         err = 0;
527                 } else if (!strncasecmp(str_governor, "powersave",
528                                                 CPUFREQ_NAME_LEN)) {
529                         *policy = CPUFREQ_POLICY_POWERSAVE;
530                         err = 0;
531                 }
532         } else {
533                 struct cpufreq_governor *t;
534
535                 mutex_lock(&cpufreq_governor_mutex);
536
537                 t = find_governor(str_governor);
538
539                 if (t == NULL) {
540                         int ret;
541
542                         mutex_unlock(&cpufreq_governor_mutex);
543                         ret = request_module("cpufreq_%s", str_governor);
544                         mutex_lock(&cpufreq_governor_mutex);
545
546                         if (ret == 0)
547                                 t = find_governor(str_governor);
548                 }
549
550                 if (t != NULL) {
551                         *governor = t;
552                         err = 0;
553                 }
554
555                 mutex_unlock(&cpufreq_governor_mutex);
556         }
557         return err;
558 }
559
560 /**
561  * cpufreq_per_cpu_attr_read() / show_##file_name() -
562  * print out cpufreq information
563  *
564  * Write out information from cpufreq_driver->policy[cpu]; object must be
565  * "unsigned int".
566  */
567
568 #define show_one(file_name, object)                     \
569 static ssize_t show_##file_name                         \
570 (struct cpufreq_policy *policy, char *buf)              \
571 {                                                       \
572         return sprintf(buf, "%u\n", policy->object);    \
573 }
574
575 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
576 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
577 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
578 show_one(scaling_min_freq, min);
579 show_one(scaling_max_freq, max);
580
581 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
582 {
583         ssize_t ret;
584
585         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
586                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
587         else
588                 ret = sprintf(buf, "%u\n", policy->cur);
589         return ret;
590 }
591
592 static int cpufreq_set_policy(struct cpufreq_policy *policy,
593                                 struct cpufreq_policy *new_policy);
594
595 /**
596  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
597  */
598 #define store_one(file_name, object)                    \
599 static ssize_t store_##file_name                                        \
600 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
601 {                                                                       \
602         int ret, temp;                                                  \
603         struct cpufreq_policy new_policy;                               \
604                                                                         \
605         memcpy(&new_policy, policy, sizeof(*policy));                   \
606         new_policy.min = policy->user_policy.min;                       \
607         new_policy.max = policy->user_policy.max;                       \
608                                                                         \
609         ret = sscanf(buf, "%u", &new_policy.object);                    \
610         if (ret != 1)                                                   \
611                 return -EINVAL;                                         \
612                                                                         \
613         temp = new_policy.object;                                       \
614         ret = cpufreq_set_policy(policy, &new_policy);          \
615         if (!ret)                                                       \
616                 policy->user_policy.object = temp;                      \
617                                                                         \
618         return ret ? ret : count;                                       \
619 }
620
621 store_one(scaling_min_freq, min);
622 store_one(scaling_max_freq, max);
623
624 /**
625  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
626  */
627 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
628                                         char *buf)
629 {
630         unsigned int cur_freq = __cpufreq_get(policy);
631
632         if (cur_freq)
633                 return sprintf(buf, "%u\n", cur_freq);
634
635         return sprintf(buf, "<unknown>\n");
636 }
637
638 /**
639  * show_scaling_governor - show the current policy for the specified CPU
640  */
641 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
642 {
643         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
644                 return sprintf(buf, "powersave\n");
645         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
646                 return sprintf(buf, "performance\n");
647         else if (policy->governor)
648                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
649                                 policy->governor->name);
650         return -EINVAL;
651 }
652
653 /**
654  * store_scaling_governor - store policy for the specified CPU
655  */
656 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
657                                         const char *buf, size_t count)
658 {
659         int ret;
660         char    str_governor[16];
661         struct cpufreq_policy new_policy;
662
663         memcpy(&new_policy, policy, sizeof(*policy));
664
665         ret = sscanf(buf, "%15s", str_governor);
666         if (ret != 1)
667                 return -EINVAL;
668
669         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
670                                                 &new_policy.governor))
671                 return -EINVAL;
672
673         ret = cpufreq_set_policy(policy, &new_policy);
674         return ret ? ret : count;
675 }
676
677 /**
678  * show_scaling_driver - show the cpufreq driver currently loaded
679  */
680 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
681 {
682         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
683 }
684
685 /**
686  * show_scaling_available_governors - show the available CPUfreq governors
687  */
688 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
689                                                 char *buf)
690 {
691         ssize_t i = 0;
692         struct cpufreq_governor *t;
693
694         if (!has_target()) {
695                 i += sprintf(buf, "performance powersave");
696                 goto out;
697         }
698
699         for_each_governor(t) {
700                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
701                     - (CPUFREQ_NAME_LEN + 2)))
702                         goto out;
703                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
704         }
705 out:
706         i += sprintf(&buf[i], "\n");
707         return i;
708 }
709
710 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
711 {
712         ssize_t i = 0;
713         unsigned int cpu;
714
715         for_each_cpu(cpu, mask) {
716                 if (i)
717                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
718                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
719                 if (i >= (PAGE_SIZE - 5))
720                         break;
721         }
722         i += sprintf(&buf[i], "\n");
723         return i;
724 }
725 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
726
727 /**
728  * show_related_cpus - show the CPUs affected by each transition even if
729  * hw coordination is in use
730  */
731 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
732 {
733         return cpufreq_show_cpus(policy->related_cpus, buf);
734 }
735
736 /**
737  * show_affected_cpus - show the CPUs affected by each transition
738  */
739 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
740 {
741         return cpufreq_show_cpus(policy->cpus, buf);
742 }
743
744 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
745                                         const char *buf, size_t count)
746 {
747         unsigned int freq = 0;
748         unsigned int ret;
749
750         if (!policy->governor || !policy->governor->store_setspeed)
751                 return -EINVAL;
752
753         ret = sscanf(buf, "%u", &freq);
754         if (ret != 1)
755                 return -EINVAL;
756
757         policy->governor->store_setspeed(policy, freq);
758
759         return count;
760 }
761
762 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
763 {
764         if (!policy->governor || !policy->governor->show_setspeed)
765                 return sprintf(buf, "<unsupported>\n");
766
767         return policy->governor->show_setspeed(policy, buf);
768 }
769
770 /**
771  * show_bios_limit - show the current cpufreq HW/BIOS limitation
772  */
773 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
774 {
775         unsigned int limit;
776         int ret;
777         if (cpufreq_driver->bios_limit) {
778                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
779                 if (!ret)
780                         return sprintf(buf, "%u\n", limit);
781         }
782         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
783 }
784
785 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
786 cpufreq_freq_attr_ro(cpuinfo_min_freq);
787 cpufreq_freq_attr_ro(cpuinfo_max_freq);
788 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
789 cpufreq_freq_attr_ro(scaling_available_governors);
790 cpufreq_freq_attr_ro(scaling_driver);
791 cpufreq_freq_attr_ro(scaling_cur_freq);
792 cpufreq_freq_attr_ro(bios_limit);
793 cpufreq_freq_attr_ro(related_cpus);
794 cpufreq_freq_attr_ro(affected_cpus);
795 cpufreq_freq_attr_rw(scaling_min_freq);
796 cpufreq_freq_attr_rw(scaling_max_freq);
797 cpufreq_freq_attr_rw(scaling_governor);
798 cpufreq_freq_attr_rw(scaling_setspeed);
799
800 static struct attribute *default_attrs[] = {
801         &cpuinfo_min_freq.attr,
802         &cpuinfo_max_freq.attr,
803         &cpuinfo_transition_latency.attr,
804         &scaling_min_freq.attr,
805         &scaling_max_freq.attr,
806         &affected_cpus.attr,
807         &related_cpus.attr,
808         &scaling_governor.attr,
809         &scaling_driver.attr,
810         &scaling_available_governors.attr,
811         &scaling_setspeed.attr,
812         NULL
813 };
814
815 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
816 #define to_attr(a) container_of(a, struct freq_attr, attr)
817
818 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
819 {
820         struct cpufreq_policy *policy = to_policy(kobj);
821         struct freq_attr *fattr = to_attr(attr);
822         ssize_t ret;
823
824         down_read(&policy->rwsem);
825
826         if (fattr->show)
827                 ret = fattr->show(policy, buf);
828         else
829                 ret = -EIO;
830
831         up_read(&policy->rwsem);
832
833         return ret;
834 }
835
836 static ssize_t store(struct kobject *kobj, struct attribute *attr,
837                      const char *buf, size_t count)
838 {
839         struct cpufreq_policy *policy = to_policy(kobj);
840         struct freq_attr *fattr = to_attr(attr);
841         ssize_t ret = -EINVAL;
842
843         get_online_cpus();
844
845         if (!cpu_online(policy->cpu))
846                 goto unlock;
847
848         down_write(&policy->rwsem);
849
850         if (fattr->store)
851                 ret = fattr->store(policy, buf, count);
852         else
853                 ret = -EIO;
854
855         up_write(&policy->rwsem);
856 unlock:
857         put_online_cpus();
858
859         return ret;
860 }
861
862 static void cpufreq_sysfs_release(struct kobject *kobj)
863 {
864         struct cpufreq_policy *policy = to_policy(kobj);
865         pr_debug("last reference is dropped\n");
866         complete(&policy->kobj_unregister);
867 }
868
869 static const struct sysfs_ops sysfs_ops = {
870         .show   = show,
871         .store  = store,
872 };
873
874 static struct kobj_type ktype_cpufreq = {
875         .sysfs_ops      = &sysfs_ops,
876         .default_attrs  = default_attrs,
877         .release        = cpufreq_sysfs_release,
878 };
879
880 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
881 {
882         struct device *cpu_dev;
883
884         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
885
886         if (!policy)
887                 return 0;
888
889         cpu_dev = get_cpu_device(cpu);
890         if (WARN_ON(!cpu_dev))
891                 return 0;
892
893         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
894 }
895
896 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
897 {
898         struct device *cpu_dev;
899
900         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
901
902         cpu_dev = get_cpu_device(cpu);
903         if (WARN_ON(!cpu_dev))
904                 return;
905
906         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
907 }
908
909 /* Add/remove symlinks for all related CPUs */
910 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
911 {
912         unsigned int j;
913         int ret = 0;
914
915         /* Some related CPUs might not be present (physically hotplugged) */
916         for_each_cpu(j, policy->real_cpus) {
917                 ret = add_cpu_dev_symlink(policy, j);
918                 if (ret)
919                         break;
920         }
921
922         return ret;
923 }
924
925 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
926 {
927         unsigned int j;
928
929         /* Some related CPUs might not be present (physically hotplugged) */
930         for_each_cpu(j, policy->real_cpus)
931                 remove_cpu_dev_symlink(policy, j);
932 }
933
934 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
935 {
936         struct freq_attr **drv_attr;
937         int ret = 0;
938
939         /* set up files for this cpu device */
940         drv_attr = cpufreq_driver->attr;
941         while (drv_attr && *drv_attr) {
942                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
943                 if (ret)
944                         return ret;
945                 drv_attr++;
946         }
947         if (cpufreq_driver->get) {
948                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
949                 if (ret)
950                         return ret;
951         }
952
953         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
954         if (ret)
955                 return ret;
956
957         if (cpufreq_driver->bios_limit) {
958                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
959                 if (ret)
960                         return ret;
961         }
962
963         return cpufreq_add_dev_symlink(policy);
964 }
965
966 static int cpufreq_init_policy(struct cpufreq_policy *policy)
967 {
968         struct cpufreq_governor *gov = NULL;
969         struct cpufreq_policy new_policy;
970
971         memcpy(&new_policy, policy, sizeof(*policy));
972
973         /* Update governor of new_policy to the governor used before hotplug */
974         gov = find_governor(policy->last_governor);
975         if (gov)
976                 pr_debug("Restoring governor %s for cpu %d\n",
977                                 policy->governor->name, policy->cpu);
978         else
979                 gov = CPUFREQ_DEFAULT_GOVERNOR;
980
981         new_policy.governor = gov;
982
983         /* Use the default policy if there is no last_policy. */
984         if (cpufreq_driver->setpolicy) {
985                 if (policy->last_policy)
986                         new_policy.policy = policy->last_policy;
987                 else
988                         cpufreq_parse_governor(gov->name, &new_policy.policy,
989                                                NULL);
990         }
991         /* set default policy */
992         return cpufreq_set_policy(policy, &new_policy);
993 }
994
995 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
996 {
997         int ret = 0;
998
999         /* Has this CPU been taken care of already? */
1000         if (cpumask_test_cpu(cpu, policy->cpus))
1001                 return 0;
1002
1003         if (has_target()) {
1004                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1005                 if (ret) {
1006                         pr_err("%s: Failed to stop governor\n", __func__);
1007                         return ret;
1008                 }
1009         }
1010
1011         down_write(&policy->rwsem);
1012         cpumask_set_cpu(cpu, policy->cpus);
1013         up_write(&policy->rwsem);
1014
1015         if (has_target()) {
1016                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1017                 if (!ret)
1018                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1019
1020                 if (ret) {
1021                         pr_err("%s: Failed to start governor\n", __func__);
1022                         return ret;
1023                 }
1024         }
1025
1026         return 0;
1027 }
1028
1029 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1030 {
1031         struct device *dev = get_cpu_device(cpu);
1032         struct cpufreq_policy *policy;
1033
1034         if (WARN_ON(!dev))
1035                 return NULL;
1036
1037         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1038         if (!policy)
1039                 return NULL;
1040
1041         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1042                 goto err_free_policy;
1043
1044         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1045                 goto err_free_cpumask;
1046
1047         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1048                 goto err_free_rcpumask;
1049
1050         kobject_init(&policy->kobj, &ktype_cpufreq);
1051         INIT_LIST_HEAD(&policy->policy_list);
1052         init_rwsem(&policy->rwsem);
1053         spin_lock_init(&policy->transition_lock);
1054         init_waitqueue_head(&policy->transition_wait);
1055         init_completion(&policy->kobj_unregister);
1056         INIT_WORK(&policy->update, handle_update);
1057
1058         policy->cpu = cpu;
1059         return policy;
1060
1061 err_free_rcpumask:
1062         free_cpumask_var(policy->related_cpus);
1063 err_free_cpumask:
1064         free_cpumask_var(policy->cpus);
1065 err_free_policy:
1066         kfree(policy);
1067
1068         return NULL;
1069 }
1070
1071 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1072 {
1073         struct kobject *kobj;
1074         struct completion *cmp;
1075
1076         if (notify)
1077                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1078                                              CPUFREQ_REMOVE_POLICY, policy);
1079
1080         down_write(&policy->rwsem);
1081         cpufreq_remove_dev_symlink(policy);
1082         kobj = &policy->kobj;
1083         cmp = &policy->kobj_unregister;
1084         up_write(&policy->rwsem);
1085         kobject_put(kobj);
1086
1087         /*
1088          * We need to make sure that the underlying kobj is
1089          * actually not referenced anymore by anybody before we
1090          * proceed with unloading.
1091          */
1092         pr_debug("waiting for dropping of refcount\n");
1093         wait_for_completion(cmp);
1094         pr_debug("wait complete\n");
1095 }
1096
1097 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1098 {
1099         unsigned long flags;
1100         int cpu;
1101
1102         /* Remove policy from list */
1103         write_lock_irqsave(&cpufreq_driver_lock, flags);
1104         list_del(&policy->policy_list);
1105
1106         for_each_cpu(cpu, policy->related_cpus)
1107                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1108         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1109
1110         cpufreq_policy_put_kobj(policy, notify);
1111         free_cpumask_var(policy->real_cpus);
1112         free_cpumask_var(policy->related_cpus);
1113         free_cpumask_var(policy->cpus);
1114         kfree(policy);
1115 }
1116
1117 static int cpufreq_online(unsigned int cpu)
1118 {
1119         struct cpufreq_policy *policy;
1120         bool new_policy;
1121         unsigned long flags;
1122         unsigned int j;
1123         int ret;
1124
1125         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1126
1127         /* Check if this CPU already has a policy to manage it */
1128         policy = per_cpu(cpufreq_cpu_data, cpu);
1129         if (policy) {
1130                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1131                 if (!policy_is_inactive(policy))
1132                         return cpufreq_add_policy_cpu(policy, cpu);
1133
1134                 /* This is the only online CPU for the policy.  Start over. */
1135                 new_policy = false;
1136                 down_write(&policy->rwsem);
1137                 policy->cpu = cpu;
1138                 policy->governor = NULL;
1139                 up_write(&policy->rwsem);
1140         } else {
1141                 new_policy = true;
1142                 policy = cpufreq_policy_alloc(cpu);
1143                 if (!policy)
1144                         return -ENOMEM;
1145         }
1146
1147         cpumask_copy(policy->cpus, cpumask_of(cpu));
1148
1149         /* call driver. From then on the cpufreq must be able
1150          * to accept all calls to ->verify and ->setpolicy for this CPU
1151          */
1152         ret = cpufreq_driver->init(policy);
1153         if (ret) {
1154                 pr_debug("initialization failed\n");
1155                 goto out_free_policy;
1156         }
1157
1158         down_write(&policy->rwsem);
1159
1160         if (new_policy) {
1161                 /* related_cpus should at least include policy->cpus. */
1162                 cpumask_copy(policy->related_cpus, policy->cpus);
1163                 /* Remember CPUs present at the policy creation time. */
1164                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1165
1166                 /* Name and add the kobject */
1167                 ret = kobject_add(&policy->kobj, cpufreq_global_kobject,
1168                                   "policy%u",
1169                                   cpumask_first(policy->related_cpus));
1170                 if (ret) {
1171                         pr_err("%s: failed to add policy->kobj: %d\n", __func__,
1172                                ret);
1173                         goto out_exit_policy;
1174                 }
1175         }
1176
1177         /*
1178          * affected cpus must always be the one, which are online. We aren't
1179          * managing offline cpus here.
1180          */
1181         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1182
1183         if (new_policy) {
1184                 policy->user_policy.min = policy->min;
1185                 policy->user_policy.max = policy->max;
1186
1187                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1188                 for_each_cpu(j, policy->related_cpus)
1189                         per_cpu(cpufreq_cpu_data, j) = policy;
1190                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1191         } else {
1192                 policy->min = policy->user_policy.min;
1193                 policy->max = policy->user_policy.max;
1194         }
1195
1196         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1197                 policy->cur = cpufreq_driver->get(policy->cpu);
1198                 if (!policy->cur) {
1199                         pr_err("%s: ->get() failed\n", __func__);
1200                         goto out_exit_policy;
1201                 }
1202         }
1203
1204         /*
1205          * Sometimes boot loaders set CPU frequency to a value outside of
1206          * frequency table present with cpufreq core. In such cases CPU might be
1207          * unstable if it has to run on that frequency for long duration of time
1208          * and so its better to set it to a frequency which is specified in
1209          * freq-table. This also makes cpufreq stats inconsistent as
1210          * cpufreq-stats would fail to register because current frequency of CPU
1211          * isn't found in freq-table.
1212          *
1213          * Because we don't want this change to effect boot process badly, we go
1214          * for the next freq which is >= policy->cur ('cur' must be set by now,
1215          * otherwise we will end up setting freq to lowest of the table as 'cur'
1216          * is initialized to zero).
1217          *
1218          * We are passing target-freq as "policy->cur - 1" otherwise
1219          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1220          * equal to target-freq.
1221          */
1222         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1223             && has_target()) {
1224                 /* Are we running at unknown frequency ? */
1225                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1226                 if (ret == -EINVAL) {
1227                         /* Warn user and fix it */
1228                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1229                                 __func__, policy->cpu, policy->cur);
1230                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1231                                 CPUFREQ_RELATION_L);
1232
1233                         /*
1234                          * Reaching here after boot in a few seconds may not
1235                          * mean that system will remain stable at "unknown"
1236                          * frequency for longer duration. Hence, a BUG_ON().
1237                          */
1238                         BUG_ON(ret);
1239                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1240                                 __func__, policy->cpu, policy->cur);
1241                 }
1242         }
1243
1244         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1245                                      CPUFREQ_START, policy);
1246
1247         if (new_policy) {
1248                 ret = cpufreq_add_dev_interface(policy);
1249                 if (ret)
1250                         goto out_exit_policy;
1251                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1252                                 CPUFREQ_CREATE_POLICY, policy);
1253
1254                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1255                 list_add(&policy->policy_list, &cpufreq_policy_list);
1256                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1257         }
1258
1259         ret = cpufreq_init_policy(policy);
1260         if (ret) {
1261                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1262                        __func__, cpu, ret);
1263                 /* cpufreq_policy_free() will notify based on this */
1264                 new_policy = false;
1265                 goto out_exit_policy;
1266         }
1267
1268         up_write(&policy->rwsem);
1269
1270         kobject_uevent(&policy->kobj, KOBJ_ADD);
1271
1272         /* Callback for handling stuff after policy is ready */
1273         if (cpufreq_driver->ready)
1274                 cpufreq_driver->ready(policy);
1275
1276         pr_debug("initialization complete\n");
1277
1278         return 0;
1279
1280 out_exit_policy:
1281         up_write(&policy->rwsem);
1282
1283         if (cpufreq_driver->exit)
1284                 cpufreq_driver->exit(policy);
1285 out_free_policy:
1286         cpufreq_policy_free(policy, !new_policy);
1287         return ret;
1288 }
1289
1290 /**
1291  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1292  * @dev: CPU device.
1293  * @sif: Subsystem interface structure pointer (not used)
1294  */
1295 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1296 {
1297         unsigned cpu = dev->id;
1298         int ret;
1299
1300         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1301
1302         if (cpu_online(cpu)) {
1303                 ret = cpufreq_online(cpu);
1304         } else {
1305                 /*
1306                  * A hotplug notifier will follow and we will handle it as CPU
1307                  * online then.  For now, just create the sysfs link, unless
1308                  * there is no policy or the link is already present.
1309                  */
1310                 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1311
1312                 ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1313                         ? add_cpu_dev_symlink(policy, cpu) : 0;
1314         }
1315
1316         return ret;
1317 }
1318
1319 static void cpufreq_offline_prepare(unsigned int cpu)
1320 {
1321         struct cpufreq_policy *policy;
1322
1323         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1324
1325         policy = cpufreq_cpu_get_raw(cpu);
1326         if (!policy) {
1327                 pr_debug("%s: No cpu_data found\n", __func__);
1328                 return;
1329         }
1330
1331         if (has_target()) {
1332                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1333                 if (ret)
1334                         pr_err("%s: Failed to stop governor\n", __func__);
1335         }
1336
1337         down_write(&policy->rwsem);
1338         cpumask_clear_cpu(cpu, policy->cpus);
1339
1340         if (policy_is_inactive(policy)) {
1341                 if (has_target())
1342                         strncpy(policy->last_governor, policy->governor->name,
1343                                 CPUFREQ_NAME_LEN);
1344                 else
1345                         policy->last_policy = policy->policy;
1346         } else if (cpu == policy->cpu) {
1347                 /* Nominate new CPU */
1348                 policy->cpu = cpumask_any(policy->cpus);
1349         }
1350         up_write(&policy->rwsem);
1351
1352         /* Start governor again for active policy */
1353         if (!policy_is_inactive(policy)) {
1354                 if (has_target()) {
1355                         int ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1356                         if (!ret)
1357                                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1358
1359                         if (ret)
1360                                 pr_err("%s: Failed to start governor\n", __func__);
1361                 }
1362         } else if (cpufreq_driver->stop_cpu) {
1363                 cpufreq_driver->stop_cpu(policy);
1364         }
1365 }
1366
1367 static void cpufreq_offline_finish(unsigned int cpu)
1368 {
1369         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1370
1371         if (!policy) {
1372                 pr_debug("%s: No cpu_data found\n", __func__);
1373                 return;
1374         }
1375
1376         /* Only proceed for inactive policies */
1377         if (!policy_is_inactive(policy))
1378                 return;
1379
1380         /* If cpu is last user of policy, free policy */
1381         if (has_target()) {
1382                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1383                 if (ret)
1384                         pr_err("%s: Failed to exit governor\n", __func__);
1385         }
1386
1387         /*
1388          * Perform the ->exit() even during light-weight tear-down,
1389          * since this is a core component, and is essential for the
1390          * subsequent light-weight ->init() to succeed.
1391          */
1392         if (cpufreq_driver->exit) {
1393                 cpufreq_driver->exit(policy);
1394                 policy->freq_table = NULL;
1395         }
1396 }
1397
1398 /**
1399  * cpufreq_remove_dev - remove a CPU device
1400  *
1401  * Removes the cpufreq interface for a CPU device.
1402  */
1403 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1404 {
1405         unsigned int cpu = dev->id;
1406         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1407
1408         if (!policy)
1409                 return;
1410
1411         if (cpu_online(cpu)) {
1412                 cpufreq_offline_prepare(cpu);
1413                 cpufreq_offline_finish(cpu);
1414         }
1415
1416         cpumask_clear_cpu(cpu, policy->real_cpus);
1417         remove_cpu_dev_symlink(policy, cpu);
1418
1419         if (cpumask_empty(policy->real_cpus))
1420                 cpufreq_policy_free(policy, true);
1421 }
1422
1423 static void handle_update(struct work_struct *work)
1424 {
1425         struct cpufreq_policy *policy =
1426                 container_of(work, struct cpufreq_policy, update);
1427         unsigned int cpu = policy->cpu;
1428         pr_debug("handle_update for cpu %u called\n", cpu);
1429         cpufreq_update_policy(cpu);
1430 }
1431
1432 /**
1433  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1434  *      in deep trouble.
1435  *      @policy: policy managing CPUs
1436  *      @new_freq: CPU frequency the CPU actually runs at
1437  *
1438  *      We adjust to current frequency first, and need to clean up later.
1439  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1440  */
1441 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1442                                 unsigned int new_freq)
1443 {
1444         struct cpufreq_freqs freqs;
1445
1446         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1447                  policy->cur, new_freq);
1448
1449         freqs.old = policy->cur;
1450         freqs.new = new_freq;
1451
1452         cpufreq_freq_transition_begin(policy, &freqs);
1453         cpufreq_freq_transition_end(policy, &freqs, 0);
1454 }
1455
1456 /**
1457  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1458  * @cpu: CPU number
1459  *
1460  * This is the last known freq, without actually getting it from the driver.
1461  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1462  */
1463 unsigned int cpufreq_quick_get(unsigned int cpu)
1464 {
1465         struct cpufreq_policy *policy;
1466         unsigned int ret_freq = 0;
1467
1468         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1469                 return cpufreq_driver->get(cpu);
1470
1471         policy = cpufreq_cpu_get(cpu);
1472         if (policy) {
1473                 ret_freq = policy->cur;
1474                 cpufreq_cpu_put(policy);
1475         }
1476
1477         return ret_freq;
1478 }
1479 EXPORT_SYMBOL(cpufreq_quick_get);
1480
1481 /**
1482  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1483  * @cpu: CPU number
1484  *
1485  * Just return the max possible frequency for a given CPU.
1486  */
1487 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1488 {
1489         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1490         unsigned int ret_freq = 0;
1491
1492         if (policy) {
1493                 ret_freq = policy->max;
1494                 cpufreq_cpu_put(policy);
1495         }
1496
1497         return ret_freq;
1498 }
1499 EXPORT_SYMBOL(cpufreq_quick_get_max);
1500
1501 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1502 {
1503         unsigned int ret_freq = 0;
1504
1505         if (!cpufreq_driver->get)
1506                 return ret_freq;
1507
1508         ret_freq = cpufreq_driver->get(policy->cpu);
1509
1510         /* Updating inactive policies is invalid, so avoid doing that. */
1511         if (unlikely(policy_is_inactive(policy)))
1512                 return ret_freq;
1513
1514         if (ret_freq && policy->cur &&
1515                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1516                 /* verify no discrepancy between actual and
1517                                         saved value exists */
1518                 if (unlikely(ret_freq != policy->cur)) {
1519                         cpufreq_out_of_sync(policy, ret_freq);
1520                         schedule_work(&policy->update);
1521                 }
1522         }
1523
1524         return ret_freq;
1525 }
1526
1527 /**
1528  * cpufreq_get - get the current CPU frequency (in kHz)
1529  * @cpu: CPU number
1530  *
1531  * Get the CPU current (static) CPU frequency
1532  */
1533 unsigned int cpufreq_get(unsigned int cpu)
1534 {
1535         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1536         unsigned int ret_freq = 0;
1537
1538         if (policy) {
1539                 down_read(&policy->rwsem);
1540                 ret_freq = __cpufreq_get(policy);
1541                 up_read(&policy->rwsem);
1542
1543                 cpufreq_cpu_put(policy);
1544         }
1545
1546         return ret_freq;
1547 }
1548 EXPORT_SYMBOL(cpufreq_get);
1549
1550 static struct subsys_interface cpufreq_interface = {
1551         .name           = "cpufreq",
1552         .subsys         = &cpu_subsys,
1553         .add_dev        = cpufreq_add_dev,
1554         .remove_dev     = cpufreq_remove_dev,
1555 };
1556
1557 /*
1558  * In case platform wants some specific frequency to be configured
1559  * during suspend..
1560  */
1561 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1562 {
1563         int ret;
1564
1565         if (!policy->suspend_freq) {
1566                 pr_debug("%s: suspend_freq not defined\n", __func__);
1567                 return 0;
1568         }
1569
1570         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1571                         policy->suspend_freq);
1572
1573         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1574                         CPUFREQ_RELATION_H);
1575         if (ret)
1576                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1577                                 __func__, policy->suspend_freq, ret);
1578
1579         return ret;
1580 }
1581 EXPORT_SYMBOL(cpufreq_generic_suspend);
1582
1583 /**
1584  * cpufreq_suspend() - Suspend CPUFreq governors
1585  *
1586  * Called during system wide Suspend/Hibernate cycles for suspending governors
1587  * as some platforms can't change frequency after this point in suspend cycle.
1588  * Because some of the devices (like: i2c, regulators, etc) they use for
1589  * changing frequency are suspended quickly after this point.
1590  */
1591 void cpufreq_suspend(void)
1592 {
1593         struct cpufreq_policy *policy;
1594
1595         if (!cpufreq_driver)
1596                 return;
1597
1598         if (!has_target())
1599                 goto suspend;
1600
1601         pr_debug("%s: Suspending Governors\n", __func__);
1602
1603         for_each_active_policy(policy) {
1604                 if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1605                         pr_err("%s: Failed to stop governor for policy: %p\n",
1606                                 __func__, policy);
1607                 else if (cpufreq_driver->suspend
1608                     && cpufreq_driver->suspend(policy))
1609                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1610                                 policy);
1611         }
1612
1613 suspend:
1614         cpufreq_suspended = true;
1615 }
1616
1617 /**
1618  * cpufreq_resume() - Resume CPUFreq governors
1619  *
1620  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1621  * are suspended with cpufreq_suspend().
1622  */
1623 void cpufreq_resume(void)
1624 {
1625         struct cpufreq_policy *policy;
1626
1627         if (!cpufreq_driver)
1628                 return;
1629
1630         cpufreq_suspended = false;
1631
1632         if (!has_target())
1633                 return;
1634
1635         pr_debug("%s: Resuming Governors\n", __func__);
1636
1637         for_each_active_policy(policy) {
1638                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1639                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1640                                 policy);
1641                 else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1642                     || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1643                         pr_err("%s: Failed to start governor for policy: %p\n",
1644                                 __func__, policy);
1645         }
1646
1647         /*
1648          * schedule call cpufreq_update_policy() for first-online CPU, as that
1649          * wouldn't be hotplugged-out on suspend. It will verify that the
1650          * current freq is in sync with what we believe it to be.
1651          */
1652         policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1653         if (WARN_ON(!policy))
1654                 return;
1655
1656         schedule_work(&policy->update);
1657 }
1658
1659 /**
1660  *      cpufreq_get_current_driver - return current driver's name
1661  *
1662  *      Return the name string of the currently loaded cpufreq driver
1663  *      or NULL, if none.
1664  */
1665 const char *cpufreq_get_current_driver(void)
1666 {
1667         if (cpufreq_driver)
1668                 return cpufreq_driver->name;
1669
1670         return NULL;
1671 }
1672 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1673
1674 /**
1675  *      cpufreq_get_driver_data - return current driver data
1676  *
1677  *      Return the private data of the currently loaded cpufreq
1678  *      driver, or NULL if no cpufreq driver is loaded.
1679  */
1680 void *cpufreq_get_driver_data(void)
1681 {
1682         if (cpufreq_driver)
1683                 return cpufreq_driver->driver_data;
1684
1685         return NULL;
1686 }
1687 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1688
1689 /*********************************************************************
1690  *                     NOTIFIER LISTS INTERFACE                      *
1691  *********************************************************************/
1692
1693 /**
1694  *      cpufreq_register_notifier - register a driver with cpufreq
1695  *      @nb: notifier function to register
1696  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1697  *
1698  *      Add a driver to one of two lists: either a list of drivers that
1699  *      are notified about clock rate changes (once before and once after
1700  *      the transition), or a list of drivers that are notified about
1701  *      changes in cpufreq policy.
1702  *
1703  *      This function may sleep, and has the same return conditions as
1704  *      blocking_notifier_chain_register.
1705  */
1706 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1707 {
1708         int ret;
1709
1710         if (cpufreq_disabled())
1711                 return -EINVAL;
1712
1713         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1714
1715         switch (list) {
1716         case CPUFREQ_TRANSITION_NOTIFIER:
1717                 ret = srcu_notifier_chain_register(
1718                                 &cpufreq_transition_notifier_list, nb);
1719                 break;
1720         case CPUFREQ_POLICY_NOTIFIER:
1721                 ret = blocking_notifier_chain_register(
1722                                 &cpufreq_policy_notifier_list, nb);
1723                 break;
1724         default:
1725                 ret = -EINVAL;
1726         }
1727
1728         return ret;
1729 }
1730 EXPORT_SYMBOL(cpufreq_register_notifier);
1731
1732 /**
1733  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1734  *      @nb: notifier block to be unregistered
1735  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1736  *
1737  *      Remove a driver from the CPU frequency notifier list.
1738  *
1739  *      This function may sleep, and has the same return conditions as
1740  *      blocking_notifier_chain_unregister.
1741  */
1742 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1743 {
1744         int ret;
1745
1746         if (cpufreq_disabled())
1747                 return -EINVAL;
1748
1749         switch (list) {
1750         case CPUFREQ_TRANSITION_NOTIFIER:
1751                 ret = srcu_notifier_chain_unregister(
1752                                 &cpufreq_transition_notifier_list, nb);
1753                 break;
1754         case CPUFREQ_POLICY_NOTIFIER:
1755                 ret = blocking_notifier_chain_unregister(
1756                                 &cpufreq_policy_notifier_list, nb);
1757                 break;
1758         default:
1759                 ret = -EINVAL;
1760         }
1761
1762         return ret;
1763 }
1764 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1765
1766
1767 /*********************************************************************
1768  *                              GOVERNORS                            *
1769  *********************************************************************/
1770
1771 /* Must set freqs->new to intermediate frequency */
1772 static int __target_intermediate(struct cpufreq_policy *policy,
1773                                  struct cpufreq_freqs *freqs, int index)
1774 {
1775         int ret;
1776
1777         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1778
1779         /* We don't need to switch to intermediate freq */
1780         if (!freqs->new)
1781                 return 0;
1782
1783         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1784                  __func__, policy->cpu, freqs->old, freqs->new);
1785
1786         cpufreq_freq_transition_begin(policy, freqs);
1787         ret = cpufreq_driver->target_intermediate(policy, index);
1788         cpufreq_freq_transition_end(policy, freqs, ret);
1789
1790         if (ret)
1791                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1792                        __func__, ret);
1793
1794         return ret;
1795 }
1796
1797 static int __target_index(struct cpufreq_policy *policy,
1798                           struct cpufreq_frequency_table *freq_table, int index)
1799 {
1800         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1801         unsigned int intermediate_freq = 0;
1802         int retval = -EINVAL;
1803         bool notify;
1804
1805         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1806         if (notify) {
1807                 /* Handle switching to intermediate frequency */
1808                 if (cpufreq_driver->get_intermediate) {
1809                         retval = __target_intermediate(policy, &freqs, index);
1810                         if (retval)
1811                                 return retval;
1812
1813                         intermediate_freq = freqs.new;
1814                         /* Set old freq to intermediate */
1815                         if (intermediate_freq)
1816                                 freqs.old = freqs.new;
1817                 }
1818
1819                 freqs.new = freq_table[index].frequency;
1820                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1821                          __func__, policy->cpu, freqs.old, freqs.new);
1822
1823                 cpufreq_freq_transition_begin(policy, &freqs);
1824         }
1825
1826         retval = cpufreq_driver->target_index(policy, index);
1827         if (retval)
1828                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1829                        retval);
1830
1831         if (notify) {
1832                 cpufreq_freq_transition_end(policy, &freqs, retval);
1833
1834                 /*
1835                  * Failed after setting to intermediate freq? Driver should have
1836                  * reverted back to initial frequency and so should we. Check
1837                  * here for intermediate_freq instead of get_intermediate, in
1838                  * case we haven't switched to intermediate freq at all.
1839                  */
1840                 if (unlikely(retval && intermediate_freq)) {
1841                         freqs.old = intermediate_freq;
1842                         freqs.new = policy->restore_freq;
1843                         cpufreq_freq_transition_begin(policy, &freqs);
1844                         cpufreq_freq_transition_end(policy, &freqs, 0);
1845                 }
1846         }
1847
1848         return retval;
1849 }
1850
1851 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1852                             unsigned int target_freq,
1853                             unsigned int relation)
1854 {
1855         unsigned int old_target_freq = target_freq;
1856         int retval = -EINVAL;
1857
1858         if (cpufreq_disabled())
1859                 return -ENODEV;
1860
1861         /* Make sure that target_freq is within supported range */
1862         if (target_freq > policy->max)
1863                 target_freq = policy->max;
1864         if (target_freq < policy->min)
1865                 target_freq = policy->min;
1866
1867         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1868                  policy->cpu, target_freq, relation, old_target_freq);
1869
1870         /*
1871          * This might look like a redundant call as we are checking it again
1872          * after finding index. But it is left intentionally for cases where
1873          * exactly same freq is called again and so we can save on few function
1874          * calls.
1875          */
1876         if (target_freq == policy->cur)
1877                 return 0;
1878
1879         /* Save last value to restore later on errors */
1880         policy->restore_freq = policy->cur;
1881
1882         if (cpufreq_driver->target)
1883                 retval = cpufreq_driver->target(policy, target_freq, relation);
1884         else if (cpufreq_driver->target_index) {
1885                 struct cpufreq_frequency_table *freq_table;
1886                 int index;
1887
1888                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1889                 if (unlikely(!freq_table)) {
1890                         pr_err("%s: Unable to find freq_table\n", __func__);
1891                         goto out;
1892                 }
1893
1894                 retval = cpufreq_frequency_table_target(policy, freq_table,
1895                                 target_freq, relation, &index);
1896                 if (unlikely(retval)) {
1897                         pr_err("%s: Unable to find matching freq\n", __func__);
1898                         goto out;
1899                 }
1900
1901                 if (freq_table[index].frequency == policy->cur) {
1902                         retval = 0;
1903                         goto out;
1904                 }
1905
1906                 retval = __target_index(policy, freq_table, index);
1907         }
1908
1909 out:
1910         return retval;
1911 }
1912 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1913
1914 int cpufreq_driver_target(struct cpufreq_policy *policy,
1915                           unsigned int target_freq,
1916                           unsigned int relation)
1917 {
1918         int ret = -EINVAL;
1919
1920         down_write(&policy->rwsem);
1921
1922         ret = __cpufreq_driver_target(policy, target_freq, relation);
1923
1924         up_write(&policy->rwsem);
1925
1926         return ret;
1927 }
1928 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1929
1930 static int __cpufreq_governor(struct cpufreq_policy *policy,
1931                                         unsigned int event)
1932 {
1933         int ret;
1934
1935         /* Only must be defined when default governor is known to have latency
1936            restrictions, like e.g. conservative or ondemand.
1937            That this is the case is already ensured in Kconfig
1938         */
1939 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1940         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1941 #else
1942         struct cpufreq_governor *gov = NULL;
1943 #endif
1944
1945         /* Don't start any governor operations if we are entering suspend */
1946         if (cpufreq_suspended)
1947                 return 0;
1948         /*
1949          * Governor might not be initiated here if ACPI _PPC changed
1950          * notification happened, so check it.
1951          */
1952         if (!policy->governor)
1953                 return -EINVAL;
1954
1955         if (policy->governor->max_transition_latency &&
1956             policy->cpuinfo.transition_latency >
1957             policy->governor->max_transition_latency) {
1958                 if (!gov)
1959                         return -EINVAL;
1960                 else {
1961                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1962                                 policy->governor->name, gov->name);
1963                         policy->governor = gov;
1964                 }
1965         }
1966
1967         if (event == CPUFREQ_GOV_POLICY_INIT)
1968                 if (!try_module_get(policy->governor->owner))
1969                         return -EINVAL;
1970
1971         pr_debug("%s: for CPU %u, event %u\n", __func__, policy->cpu, event);
1972
1973         mutex_lock(&cpufreq_governor_lock);
1974         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
1975             || (!policy->governor_enabled
1976             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
1977                 mutex_unlock(&cpufreq_governor_lock);
1978                 return -EBUSY;
1979         }
1980
1981         if (event == CPUFREQ_GOV_STOP)
1982                 policy->governor_enabled = false;
1983         else if (event == CPUFREQ_GOV_START)
1984                 policy->governor_enabled = true;
1985
1986         mutex_unlock(&cpufreq_governor_lock);
1987
1988         ret = policy->governor->governor(policy, event);
1989
1990         if (!ret) {
1991                 if (event == CPUFREQ_GOV_POLICY_INIT)
1992                         policy->governor->initialized++;
1993                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1994                         policy->governor->initialized--;
1995         } else {
1996                 /* Restore original values */
1997                 mutex_lock(&cpufreq_governor_lock);
1998                 if (event == CPUFREQ_GOV_STOP)
1999                         policy->governor_enabled = true;
2000                 else if (event == CPUFREQ_GOV_START)
2001                         policy->governor_enabled = false;
2002                 mutex_unlock(&cpufreq_governor_lock);
2003         }
2004
2005         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2006                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2007                 module_put(policy->governor->owner);
2008
2009         return ret;
2010 }
2011
2012 int cpufreq_register_governor(struct cpufreq_governor *governor)
2013 {
2014         int err;
2015
2016         if (!governor)
2017                 return -EINVAL;
2018
2019         if (cpufreq_disabled())
2020                 return -ENODEV;
2021
2022         mutex_lock(&cpufreq_governor_mutex);
2023
2024         governor->initialized = 0;
2025         err = -EBUSY;
2026         if (!find_governor(governor->name)) {
2027                 err = 0;
2028                 list_add(&governor->governor_list, &cpufreq_governor_list);
2029         }
2030
2031         mutex_unlock(&cpufreq_governor_mutex);
2032         return err;
2033 }
2034 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2035
2036 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2037 {
2038         struct cpufreq_policy *policy;
2039         unsigned long flags;
2040
2041         if (!governor)
2042                 return;
2043
2044         if (cpufreq_disabled())
2045                 return;
2046
2047         /* clear last_governor for all inactive policies */
2048         read_lock_irqsave(&cpufreq_driver_lock, flags);
2049         for_each_inactive_policy(policy) {
2050                 if (!strcmp(policy->last_governor, governor->name)) {
2051                         policy->governor = NULL;
2052                         strcpy(policy->last_governor, "\0");
2053                 }
2054         }
2055         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2056
2057         mutex_lock(&cpufreq_governor_mutex);
2058         list_del(&governor->governor_list);
2059         mutex_unlock(&cpufreq_governor_mutex);
2060         return;
2061 }
2062 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2063
2064
2065 /*********************************************************************
2066  *                          POLICY INTERFACE                         *
2067  *********************************************************************/
2068
2069 /**
2070  * cpufreq_get_policy - get the current cpufreq_policy
2071  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2072  *      is written
2073  *
2074  * Reads the current cpufreq policy.
2075  */
2076 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2077 {
2078         struct cpufreq_policy *cpu_policy;
2079         if (!policy)
2080                 return -EINVAL;
2081
2082         cpu_policy = cpufreq_cpu_get(cpu);
2083         if (!cpu_policy)
2084                 return -EINVAL;
2085
2086         memcpy(policy, cpu_policy, sizeof(*policy));
2087
2088         cpufreq_cpu_put(cpu_policy);
2089         return 0;
2090 }
2091 EXPORT_SYMBOL(cpufreq_get_policy);
2092
2093 /*
2094  * policy : current policy.
2095  * new_policy: policy to be set.
2096  */
2097 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2098                                 struct cpufreq_policy *new_policy)
2099 {
2100         struct cpufreq_governor *old_gov;
2101         int ret;
2102
2103         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2104                  new_policy->cpu, new_policy->min, new_policy->max);
2105
2106         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2107
2108         /*
2109         * This check works well when we store new min/max freq attributes,
2110         * because new_policy is a copy of policy with one field updated.
2111         */
2112         if (new_policy->min > new_policy->max)
2113                 return -EINVAL;
2114
2115         /* verify the cpu speed can be set within this limit */
2116         ret = cpufreq_driver->verify(new_policy);
2117         if (ret)
2118                 return ret;
2119
2120         /* adjust if necessary - all reasons */
2121         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2122                         CPUFREQ_ADJUST, new_policy);
2123
2124         /*
2125          * verify the cpu speed can be set within this limit, which might be
2126          * different to the first one
2127          */
2128         ret = cpufreq_driver->verify(new_policy);
2129         if (ret)
2130                 return ret;
2131
2132         /* notification of the new policy */
2133         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2134                         CPUFREQ_NOTIFY, new_policy);
2135
2136         policy->min = new_policy->min;
2137         policy->max = new_policy->max;
2138
2139         pr_debug("new min and max freqs are %u - %u kHz\n",
2140                  policy->min, policy->max);
2141
2142         if (cpufreq_driver->setpolicy) {
2143                 policy->policy = new_policy->policy;
2144                 pr_debug("setting range\n");
2145                 return cpufreq_driver->setpolicy(new_policy);
2146         }
2147
2148         if (new_policy->governor == policy->governor)
2149                 goto out;
2150
2151         pr_debug("governor switch\n");
2152
2153         /* save old, working values */
2154         old_gov = policy->governor;
2155         /* end old governor */
2156         if (old_gov) {
2157                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2158                 if (ret) {
2159                         /* This can happen due to race with other operations */
2160                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2161                                  __func__, old_gov->name, ret);
2162                         return ret;
2163                 }
2164
2165                 up_write(&policy->rwsem);
2166                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2167                 down_write(&policy->rwsem);
2168
2169                 if (ret) {
2170                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2171                                __func__, old_gov->name, ret);
2172                         return ret;
2173                 }
2174         }
2175
2176         /* start new governor */
2177         policy->governor = new_policy->governor;
2178         ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2179         if (!ret) {
2180                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
2181                 if (!ret)
2182                         goto out;
2183
2184                 up_write(&policy->rwsem);
2185                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2186                 down_write(&policy->rwsem);
2187         }
2188
2189         /* new governor failed, so re-start old one */
2190         pr_debug("starting governor %s failed\n", policy->governor->name);
2191         if (old_gov) {
2192                 policy->governor = old_gov;
2193                 if (__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2194                         policy->governor = NULL;
2195                 else
2196                         __cpufreq_governor(policy, CPUFREQ_GOV_START);
2197         }
2198
2199         return ret;
2200
2201  out:
2202         pr_debug("governor: change or update limits\n");
2203         return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2204 }
2205
2206 /**
2207  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2208  *      @cpu: CPU which shall be re-evaluated
2209  *
2210  *      Useful for policy notifiers which have different necessities
2211  *      at different times.
2212  */
2213 int cpufreq_update_policy(unsigned int cpu)
2214 {
2215         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2216         struct cpufreq_policy new_policy;
2217         int ret;
2218
2219         if (!policy)
2220                 return -ENODEV;
2221
2222         down_write(&policy->rwsem);
2223
2224         pr_debug("updating policy for CPU %u\n", cpu);
2225         memcpy(&new_policy, policy, sizeof(*policy));
2226         new_policy.min = policy->user_policy.min;
2227         new_policy.max = policy->user_policy.max;
2228
2229         /*
2230          * BIOS might change freq behind our back
2231          * -> ask driver for current freq and notify governors about a change
2232          */
2233         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2234                 new_policy.cur = cpufreq_driver->get(cpu);
2235                 if (WARN_ON(!new_policy.cur)) {
2236                         ret = -EIO;
2237                         goto unlock;
2238                 }
2239
2240                 if (!policy->cur) {
2241                         pr_debug("Driver did not initialize current freq\n");
2242                         policy->cur = new_policy.cur;
2243                 } else {
2244                         if (policy->cur != new_policy.cur && has_target())
2245                                 cpufreq_out_of_sync(policy, new_policy.cur);
2246                 }
2247         }
2248
2249         ret = cpufreq_set_policy(policy, &new_policy);
2250
2251 unlock:
2252         up_write(&policy->rwsem);
2253
2254         cpufreq_cpu_put(policy);
2255         return ret;
2256 }
2257 EXPORT_SYMBOL(cpufreq_update_policy);
2258
2259 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2260                                         unsigned long action, void *hcpu)
2261 {
2262         unsigned int cpu = (unsigned long)hcpu;
2263
2264         switch (action & ~CPU_TASKS_FROZEN) {
2265         case CPU_ONLINE:
2266                 cpufreq_online(cpu);
2267                 break;
2268
2269         case CPU_DOWN_PREPARE:
2270                 cpufreq_offline_prepare(cpu);
2271                 break;
2272
2273         case CPU_POST_DEAD:
2274                 cpufreq_offline_finish(cpu);
2275                 break;
2276
2277         case CPU_DOWN_FAILED:
2278                 cpufreq_online(cpu);
2279                 break;
2280         }
2281         return NOTIFY_OK;
2282 }
2283
2284 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2285         .notifier_call = cpufreq_cpu_callback,
2286 };
2287
2288 /*********************************************************************
2289  *               BOOST                                               *
2290  *********************************************************************/
2291 static int cpufreq_boost_set_sw(int state)
2292 {
2293         struct cpufreq_frequency_table *freq_table;
2294         struct cpufreq_policy *policy;
2295         int ret = -EINVAL;
2296
2297         for_each_active_policy(policy) {
2298                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2299                 if (freq_table) {
2300                         ret = cpufreq_frequency_table_cpuinfo(policy,
2301                                                         freq_table);
2302                         if (ret) {
2303                                 pr_err("%s: Policy frequency update failed\n",
2304                                        __func__);
2305                                 break;
2306                         }
2307                         policy->user_policy.max = policy->max;
2308                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2309                 }
2310         }
2311
2312         return ret;
2313 }
2314
2315 int cpufreq_boost_trigger_state(int state)
2316 {
2317         unsigned long flags;
2318         int ret = 0;
2319
2320         if (cpufreq_driver->boost_enabled == state)
2321                 return 0;
2322
2323         write_lock_irqsave(&cpufreq_driver_lock, flags);
2324         cpufreq_driver->boost_enabled = state;
2325         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2326
2327         ret = cpufreq_driver->set_boost(state);
2328         if (ret) {
2329                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2330                 cpufreq_driver->boost_enabled = !state;
2331                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2332
2333                 pr_err("%s: Cannot %s BOOST\n",
2334                        __func__, state ? "enable" : "disable");
2335         }
2336
2337         return ret;
2338 }
2339
2340 int cpufreq_boost_supported(void)
2341 {
2342         if (likely(cpufreq_driver))
2343                 return cpufreq_driver->boost_supported;
2344
2345         return 0;
2346 }
2347 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2348
2349 static int create_boost_sysfs_file(void)
2350 {
2351         int ret;
2352
2353         if (!cpufreq_boost_supported())
2354                 return 0;
2355
2356         /*
2357          * Check if driver provides function to enable boost -
2358          * if not, use cpufreq_boost_set_sw as default
2359          */
2360         if (!cpufreq_driver->set_boost)
2361                 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2362
2363         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2364         if (ret)
2365                 pr_err("%s: cannot register global BOOST sysfs file\n",
2366                        __func__);
2367
2368         return ret;
2369 }
2370
2371 static void remove_boost_sysfs_file(void)
2372 {
2373         if (cpufreq_boost_supported())
2374                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2375 }
2376
2377 int cpufreq_enable_boost_support(void)
2378 {
2379         if (!cpufreq_driver)
2380                 return -EINVAL;
2381
2382         if (cpufreq_boost_supported())
2383                 return 0;
2384
2385         cpufreq_driver->boost_supported = true;
2386
2387         /* This will get removed on driver unregister */
2388         return create_boost_sysfs_file();
2389 }
2390 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2391
2392 int cpufreq_boost_enabled(void)
2393 {
2394         return cpufreq_driver->boost_enabled;
2395 }
2396 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2397
2398 /*********************************************************************
2399  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2400  *********************************************************************/
2401
2402 /**
2403  * cpufreq_register_driver - register a CPU Frequency driver
2404  * @driver_data: A struct cpufreq_driver containing the values#
2405  * submitted by the CPU Frequency driver.
2406  *
2407  * Registers a CPU Frequency driver to this core code. This code
2408  * returns zero on success, -EBUSY when another driver got here first
2409  * (and isn't unregistered in the meantime).
2410  *
2411  */
2412 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2413 {
2414         unsigned long flags;
2415         int ret;
2416
2417         if (cpufreq_disabled())
2418                 return -ENODEV;
2419
2420         if (!driver_data || !driver_data->verify || !driver_data->init ||
2421             !(driver_data->setpolicy || driver_data->target_index ||
2422                     driver_data->target) ||
2423              (driver_data->setpolicy && (driver_data->target_index ||
2424                     driver_data->target)) ||
2425              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2426                 return -EINVAL;
2427
2428         pr_debug("trying to register driver %s\n", driver_data->name);
2429
2430         /* Protect against concurrent CPU online/offline. */
2431         get_online_cpus();
2432
2433         write_lock_irqsave(&cpufreq_driver_lock, flags);
2434         if (cpufreq_driver) {
2435                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2436                 ret = -EEXIST;
2437                 goto out;
2438         }
2439         cpufreq_driver = driver_data;
2440         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2441
2442         if (driver_data->setpolicy)
2443                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2444
2445         ret = create_boost_sysfs_file();
2446         if (ret)
2447                 goto err_null_driver;
2448
2449         ret = subsys_interface_register(&cpufreq_interface);
2450         if (ret)
2451                 goto err_boost_unreg;
2452
2453         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2454             list_empty(&cpufreq_policy_list)) {
2455                 /* if all ->init() calls failed, unregister */
2456                 ret = -ENODEV;
2457                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2458                          driver_data->name);
2459                 goto err_if_unreg;
2460         }
2461
2462         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2463         pr_debug("driver %s up and running\n", driver_data->name);
2464
2465 out:
2466         put_online_cpus();
2467         return ret;
2468
2469 err_if_unreg:
2470         subsys_interface_unregister(&cpufreq_interface);
2471 err_boost_unreg:
2472         remove_boost_sysfs_file();
2473 err_null_driver:
2474         write_lock_irqsave(&cpufreq_driver_lock, flags);
2475         cpufreq_driver = NULL;
2476         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2477         goto out;
2478 }
2479 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2480
2481 /**
2482  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2483  *
2484  * Unregister the current CPUFreq driver. Only call this if you have
2485  * the right to do so, i.e. if you have succeeded in initialising before!
2486  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2487  * currently not initialised.
2488  */
2489 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2490 {
2491         unsigned long flags;
2492
2493         if (!cpufreq_driver || (driver != cpufreq_driver))
2494                 return -EINVAL;
2495
2496         pr_debug("unregistering driver %s\n", driver->name);
2497
2498         /* Protect against concurrent cpu hotplug */
2499         get_online_cpus();
2500         subsys_interface_unregister(&cpufreq_interface);
2501         remove_boost_sysfs_file();
2502         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2503
2504         write_lock_irqsave(&cpufreq_driver_lock, flags);
2505
2506         cpufreq_driver = NULL;
2507
2508         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2509         put_online_cpus();
2510
2511         return 0;
2512 }
2513 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2514
2515 /*
2516  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2517  * or mutexes when secondary CPUs are halted.
2518  */
2519 static struct syscore_ops cpufreq_syscore_ops = {
2520         .shutdown = cpufreq_suspend,
2521 };
2522
2523 struct kobject *cpufreq_global_kobject;
2524 EXPORT_SYMBOL(cpufreq_global_kobject);
2525
2526 static int __init cpufreq_core_init(void)
2527 {
2528         if (cpufreq_disabled())
2529                 return -ENODEV;
2530
2531         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2532         BUG_ON(!cpufreq_global_kobject);
2533
2534         register_syscore_ops(&cpufreq_syscore_ops);
2535
2536         return 0;
2537 }
2538 core_initcall(cpufreq_core_init);