]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - drivers/cpufreq/cpufreq.c
Merge tag 'mfd-fixes-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd
[karo-tx-linux.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 attribute *attr, char *buf)
478 {
479         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
480 }
481
482 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
483                                   const char *buf, size_t count)
484 {
485         int ret, enable;
486
487         ret = sscanf(buf, "%d", &enable);
488         if (ret != 1 || enable < 0 || enable > 1)
489                 return -EINVAL;
490
491         if (cpufreq_boost_trigger_state(enable)) {
492                 pr_err("%s: Cannot %s BOOST!\n",
493                        __func__, enable ? "enable" : "disable");
494                 return -EINVAL;
495         }
496
497         pr_debug("%s: cpufreq BOOST %s\n",
498                  __func__, enable ? "enabled" : "disabled");
499
500         return count;
501 }
502 define_one_global_rw(boost);
503
504 static struct cpufreq_governor *find_governor(const char *str_governor)
505 {
506         struct cpufreq_governor *t;
507
508         for_each_governor(t)
509                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
510                         return t;
511
512         return NULL;
513 }
514
515 /**
516  * cpufreq_parse_governor - parse a governor string
517  */
518 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
519                                 struct cpufreq_governor **governor)
520 {
521         int err = -EINVAL;
522
523         if (cpufreq_driver->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                                                                         \
607         ret = sscanf(buf, "%u", &new_policy.object);                    \
608         if (ret != 1)                                                   \
609                 return -EINVAL;                                         \
610                                                                         \
611         temp = new_policy.object;                                       \
612         ret = cpufreq_set_policy(policy, &new_policy);          \
613         if (!ret)                                                       \
614                 policy->user_policy.object = temp;                      \
615                                                                         \
616         return ret ? ret : count;                                       \
617 }
618
619 store_one(scaling_min_freq, min);
620 store_one(scaling_max_freq, max);
621
622 /**
623  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
624  */
625 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
626                                         char *buf)
627 {
628         unsigned int cur_freq = __cpufreq_get(policy);
629         if (!cur_freq)
630                 return sprintf(buf, "<unknown>");
631         return sprintf(buf, "%u\n", cur_freq);
632 }
633
634 /**
635  * show_scaling_governor - show the current policy for the specified CPU
636  */
637 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
638 {
639         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
640                 return sprintf(buf, "powersave\n");
641         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
642                 return sprintf(buf, "performance\n");
643         else if (policy->governor)
644                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
645                                 policy->governor->name);
646         return -EINVAL;
647 }
648
649 /**
650  * store_scaling_governor - store policy for the specified CPU
651  */
652 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
653                                         const char *buf, size_t count)
654 {
655         int ret;
656         char    str_governor[16];
657         struct cpufreq_policy new_policy;
658
659         memcpy(&new_policy, policy, sizeof(*policy));
660
661         ret = sscanf(buf, "%15s", str_governor);
662         if (ret != 1)
663                 return -EINVAL;
664
665         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
666                                                 &new_policy.governor))
667                 return -EINVAL;
668
669         ret = cpufreq_set_policy(policy, &new_policy);
670         return ret ? ret : count;
671 }
672
673 /**
674  * show_scaling_driver - show the cpufreq driver currently loaded
675  */
676 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
677 {
678         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
679 }
680
681 /**
682  * show_scaling_available_governors - show the available CPUfreq governors
683  */
684 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
685                                                 char *buf)
686 {
687         ssize_t i = 0;
688         struct cpufreq_governor *t;
689
690         if (!has_target()) {
691                 i += sprintf(buf, "performance powersave");
692                 goto out;
693         }
694
695         for_each_governor(t) {
696                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
697                     - (CPUFREQ_NAME_LEN + 2)))
698                         goto out;
699                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
700         }
701 out:
702         i += sprintf(&buf[i], "\n");
703         return i;
704 }
705
706 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
707 {
708         ssize_t i = 0;
709         unsigned int cpu;
710
711         for_each_cpu(cpu, mask) {
712                 if (i)
713                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
714                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
715                 if (i >= (PAGE_SIZE - 5))
716                         break;
717         }
718         i += sprintf(&buf[i], "\n");
719         return i;
720 }
721 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
722
723 /**
724  * show_related_cpus - show the CPUs affected by each transition even if
725  * hw coordination is in use
726  */
727 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
728 {
729         return cpufreq_show_cpus(policy->related_cpus, buf);
730 }
731
732 /**
733  * show_affected_cpus - show the CPUs affected by each transition
734  */
735 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
736 {
737         return cpufreq_show_cpus(policy->cpus, buf);
738 }
739
740 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
741                                         const char *buf, size_t count)
742 {
743         unsigned int freq = 0;
744         unsigned int ret;
745
746         if (!policy->governor || !policy->governor->store_setspeed)
747                 return -EINVAL;
748
749         ret = sscanf(buf, "%u", &freq);
750         if (ret != 1)
751                 return -EINVAL;
752
753         policy->governor->store_setspeed(policy, freq);
754
755         return count;
756 }
757
758 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
759 {
760         if (!policy->governor || !policy->governor->show_setspeed)
761                 return sprintf(buf, "<unsupported>\n");
762
763         return policy->governor->show_setspeed(policy, buf);
764 }
765
766 /**
767  * show_bios_limit - show the current cpufreq HW/BIOS limitation
768  */
769 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
770 {
771         unsigned int limit;
772         int ret;
773         if (cpufreq_driver->bios_limit) {
774                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
775                 if (!ret)
776                         return sprintf(buf, "%u\n", limit);
777         }
778         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
779 }
780
781 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
782 cpufreq_freq_attr_ro(cpuinfo_min_freq);
783 cpufreq_freq_attr_ro(cpuinfo_max_freq);
784 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
785 cpufreq_freq_attr_ro(scaling_available_governors);
786 cpufreq_freq_attr_ro(scaling_driver);
787 cpufreq_freq_attr_ro(scaling_cur_freq);
788 cpufreq_freq_attr_ro(bios_limit);
789 cpufreq_freq_attr_ro(related_cpus);
790 cpufreq_freq_attr_ro(affected_cpus);
791 cpufreq_freq_attr_rw(scaling_min_freq);
792 cpufreq_freq_attr_rw(scaling_max_freq);
793 cpufreq_freq_attr_rw(scaling_governor);
794 cpufreq_freq_attr_rw(scaling_setspeed);
795
796 static struct attribute *default_attrs[] = {
797         &cpuinfo_min_freq.attr,
798         &cpuinfo_max_freq.attr,
799         &cpuinfo_transition_latency.attr,
800         &scaling_min_freq.attr,
801         &scaling_max_freq.attr,
802         &affected_cpus.attr,
803         &related_cpus.attr,
804         &scaling_governor.attr,
805         &scaling_driver.attr,
806         &scaling_available_governors.attr,
807         &scaling_setspeed.attr,
808         NULL
809 };
810
811 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
812 #define to_attr(a) container_of(a, struct freq_attr, attr)
813
814 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
815 {
816         struct cpufreq_policy *policy = to_policy(kobj);
817         struct freq_attr *fattr = to_attr(attr);
818         ssize_t ret;
819
820         down_read(&policy->rwsem);
821
822         if (fattr->show)
823                 ret = fattr->show(policy, buf);
824         else
825                 ret = -EIO;
826
827         up_read(&policy->rwsem);
828
829         return ret;
830 }
831
832 static ssize_t store(struct kobject *kobj, struct attribute *attr,
833                      const char *buf, size_t count)
834 {
835         struct cpufreq_policy *policy = to_policy(kobj);
836         struct freq_attr *fattr = to_attr(attr);
837         ssize_t ret = -EINVAL;
838
839         get_online_cpus();
840
841         if (!cpu_online(policy->cpu))
842                 goto unlock;
843
844         down_write(&policy->rwsem);
845
846         /* Updating inactive policies is invalid, so avoid doing that. */
847         if (unlikely(policy_is_inactive(policy))) {
848                 ret = -EBUSY;
849                 goto unlock_policy_rwsem;
850         }
851
852         if (fattr->store)
853                 ret = fattr->store(policy, buf, count);
854         else
855                 ret = -EIO;
856
857 unlock_policy_rwsem:
858         up_write(&policy->rwsem);
859 unlock:
860         put_online_cpus();
861
862         return ret;
863 }
864
865 static void cpufreq_sysfs_release(struct kobject *kobj)
866 {
867         struct cpufreq_policy *policy = to_policy(kobj);
868         pr_debug("last reference is dropped\n");
869         complete(&policy->kobj_unregister);
870 }
871
872 static const struct sysfs_ops sysfs_ops = {
873         .show   = show,
874         .store  = store,
875 };
876
877 static struct kobj_type ktype_cpufreq = {
878         .sysfs_ops      = &sysfs_ops,
879         .default_attrs  = default_attrs,
880         .release        = cpufreq_sysfs_release,
881 };
882
883 struct kobject *cpufreq_global_kobject;
884 EXPORT_SYMBOL(cpufreq_global_kobject);
885
886 static int cpufreq_global_kobject_usage;
887
888 int cpufreq_get_global_kobject(void)
889 {
890         if (!cpufreq_global_kobject_usage++)
891                 return kobject_add(cpufreq_global_kobject,
892                                 &cpu_subsys.dev_root->kobj, "%s", "cpufreq");
893
894         return 0;
895 }
896 EXPORT_SYMBOL(cpufreq_get_global_kobject);
897
898 void cpufreq_put_global_kobject(void)
899 {
900         if (!--cpufreq_global_kobject_usage)
901                 kobject_del(cpufreq_global_kobject);
902 }
903 EXPORT_SYMBOL(cpufreq_put_global_kobject);
904
905 int cpufreq_sysfs_create_file(const struct attribute *attr)
906 {
907         int ret = cpufreq_get_global_kobject();
908
909         if (!ret) {
910                 ret = sysfs_create_file(cpufreq_global_kobject, attr);
911                 if (ret)
912                         cpufreq_put_global_kobject();
913         }
914
915         return ret;
916 }
917 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
918
919 void cpufreq_sysfs_remove_file(const struct attribute *attr)
920 {
921         sysfs_remove_file(cpufreq_global_kobject, attr);
922         cpufreq_put_global_kobject();
923 }
924 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
925
926 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
927 {
928         struct device *cpu_dev;
929
930         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
931
932         if (!policy)
933                 return 0;
934
935         cpu_dev = get_cpu_device(cpu);
936         if (WARN_ON(!cpu_dev))
937                 return 0;
938
939         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
940 }
941
942 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
943 {
944         struct device *cpu_dev;
945
946         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
947
948         cpu_dev = get_cpu_device(cpu);
949         if (WARN_ON(!cpu_dev))
950                 return;
951
952         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
953 }
954
955 /* Add/remove symlinks for all related CPUs */
956 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
957 {
958         unsigned int j;
959         int ret = 0;
960
961         /* Some related CPUs might not be present (physically hotplugged) */
962         for_each_cpu(j, policy->real_cpus) {
963                 if (j == policy->kobj_cpu)
964                         continue;
965
966                 ret = add_cpu_dev_symlink(policy, j);
967                 if (ret)
968                         break;
969         }
970
971         return ret;
972 }
973
974 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
975 {
976         unsigned int j;
977
978         /* Some related CPUs might not be present (physically hotplugged) */
979         for_each_cpu(j, policy->real_cpus) {
980                 if (j == policy->kobj_cpu)
981                         continue;
982
983                 remove_cpu_dev_symlink(policy, j);
984         }
985 }
986
987 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
988 {
989         struct freq_attr **drv_attr;
990         int ret = 0;
991
992         /* set up files for this cpu device */
993         drv_attr = cpufreq_driver->attr;
994         while (drv_attr && *drv_attr) {
995                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
996                 if (ret)
997                         return ret;
998                 drv_attr++;
999         }
1000         if (cpufreq_driver->get) {
1001                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1002                 if (ret)
1003                         return ret;
1004         }
1005
1006         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1007         if (ret)
1008                 return ret;
1009
1010         if (cpufreq_driver->bios_limit) {
1011                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1012                 if (ret)
1013                         return ret;
1014         }
1015
1016         return cpufreq_add_dev_symlink(policy);
1017 }
1018
1019 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1020 {
1021         struct cpufreq_governor *gov = NULL;
1022         struct cpufreq_policy new_policy;
1023
1024         memcpy(&new_policy, policy, sizeof(*policy));
1025
1026         /* Update governor of new_policy to the governor used before hotplug */
1027         gov = find_governor(policy->last_governor);
1028         if (gov)
1029                 pr_debug("Restoring governor %s for cpu %d\n",
1030                                 policy->governor->name, policy->cpu);
1031         else
1032                 gov = CPUFREQ_DEFAULT_GOVERNOR;
1033
1034         new_policy.governor = gov;
1035
1036         /* Use the default policy if its valid. */
1037         if (cpufreq_driver->setpolicy)
1038                 cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
1039
1040         /* set default policy */
1041         return cpufreq_set_policy(policy, &new_policy);
1042 }
1043
1044 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1045 {
1046         int ret = 0;
1047
1048         /* Has this CPU been taken care of already? */
1049         if (cpumask_test_cpu(cpu, policy->cpus))
1050                 return 0;
1051
1052         if (has_target()) {
1053                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1054                 if (ret) {
1055                         pr_err("%s: Failed to stop governor\n", __func__);
1056                         return ret;
1057                 }
1058         }
1059
1060         down_write(&policy->rwsem);
1061         cpumask_set_cpu(cpu, policy->cpus);
1062         up_write(&policy->rwsem);
1063
1064         if (has_target()) {
1065                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1066                 if (!ret)
1067                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1068
1069                 if (ret) {
1070                         pr_err("%s: Failed to start governor\n", __func__);
1071                         return ret;
1072                 }
1073         }
1074
1075         return 0;
1076 }
1077
1078 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1079 {
1080         struct device *dev = get_cpu_device(cpu);
1081         struct cpufreq_policy *policy;
1082         int ret;
1083
1084         if (WARN_ON(!dev))
1085                 return NULL;
1086
1087         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1088         if (!policy)
1089                 return NULL;
1090
1091         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1092                 goto err_free_policy;
1093
1094         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1095                 goto err_free_cpumask;
1096
1097         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1098                 goto err_free_rcpumask;
1099
1100         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &dev->kobj,
1101                                    "cpufreq");
1102         if (ret) {
1103                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1104                 goto err_free_real_cpus;
1105         }
1106
1107         INIT_LIST_HEAD(&policy->policy_list);
1108         init_rwsem(&policy->rwsem);
1109         spin_lock_init(&policy->transition_lock);
1110         init_waitqueue_head(&policy->transition_wait);
1111         init_completion(&policy->kobj_unregister);
1112         INIT_WORK(&policy->update, handle_update);
1113
1114         policy->cpu = cpu;
1115
1116         /* Set this once on allocation */
1117         policy->kobj_cpu = cpu;
1118
1119         return policy;
1120
1121 err_free_real_cpus:
1122         free_cpumask_var(policy->real_cpus);
1123 err_free_rcpumask:
1124         free_cpumask_var(policy->related_cpus);
1125 err_free_cpumask:
1126         free_cpumask_var(policy->cpus);
1127 err_free_policy:
1128         kfree(policy);
1129
1130         return NULL;
1131 }
1132
1133 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1134 {
1135         struct kobject *kobj;
1136         struct completion *cmp;
1137
1138         if (notify)
1139                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1140                                              CPUFREQ_REMOVE_POLICY, policy);
1141
1142         down_write(&policy->rwsem);
1143         cpufreq_remove_dev_symlink(policy);
1144         kobj = &policy->kobj;
1145         cmp = &policy->kobj_unregister;
1146         up_write(&policy->rwsem);
1147         kobject_put(kobj);
1148
1149         /*
1150          * We need to make sure that the underlying kobj is
1151          * actually not referenced anymore by anybody before we
1152          * proceed with unloading.
1153          */
1154         pr_debug("waiting for dropping of refcount\n");
1155         wait_for_completion(cmp);
1156         pr_debug("wait complete\n");
1157 }
1158
1159 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1160 {
1161         unsigned long flags;
1162         int cpu;
1163
1164         /* Remove policy from list */
1165         write_lock_irqsave(&cpufreq_driver_lock, flags);
1166         list_del(&policy->policy_list);
1167
1168         for_each_cpu(cpu, policy->related_cpus)
1169                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1170         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1171
1172         cpufreq_policy_put_kobj(policy, notify);
1173         free_cpumask_var(policy->real_cpus);
1174         free_cpumask_var(policy->related_cpus);
1175         free_cpumask_var(policy->cpus);
1176         kfree(policy);
1177 }
1178
1179 static int cpufreq_online(unsigned int cpu)
1180 {
1181         struct cpufreq_policy *policy;
1182         bool new_policy;
1183         unsigned long flags;
1184         unsigned int j;
1185         int ret;
1186
1187         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1188
1189         /* Check if this CPU already has a policy to manage it */
1190         policy = per_cpu(cpufreq_cpu_data, cpu);
1191         if (policy) {
1192                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1193                 if (!policy_is_inactive(policy))
1194                         return cpufreq_add_policy_cpu(policy, cpu);
1195
1196                 /* This is the only online CPU for the policy.  Start over. */
1197                 new_policy = false;
1198                 down_write(&policy->rwsem);
1199                 policy->cpu = cpu;
1200                 policy->governor = NULL;
1201                 up_write(&policy->rwsem);
1202         } else {
1203                 new_policy = true;
1204                 policy = cpufreq_policy_alloc(cpu);
1205                 if (!policy)
1206                         return -ENOMEM;
1207         }
1208
1209         cpumask_copy(policy->cpus, cpumask_of(cpu));
1210
1211         /* call driver. From then on the cpufreq must be able
1212          * to accept all calls to ->verify and ->setpolicy for this CPU
1213          */
1214         ret = cpufreq_driver->init(policy);
1215         if (ret) {
1216                 pr_debug("initialization failed\n");
1217                 goto out_free_policy;
1218         }
1219
1220         down_write(&policy->rwsem);
1221
1222         if (new_policy) {
1223                 /* related_cpus should at least include policy->cpus. */
1224                 cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1225                 /* Remember CPUs present at the policy creation time. */
1226                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1227         }
1228
1229         /*
1230          * affected cpus must always be the one, which are online. We aren't
1231          * managing offline cpus here.
1232          */
1233         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1234
1235         if (new_policy) {
1236                 policy->user_policy.min = policy->min;
1237                 policy->user_policy.max = policy->max;
1238
1239                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1240                 for_each_cpu(j, policy->related_cpus)
1241                         per_cpu(cpufreq_cpu_data, j) = policy;
1242                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1243         }
1244
1245         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1246                 policy->cur = cpufreq_driver->get(policy->cpu);
1247                 if (!policy->cur) {
1248                         pr_err("%s: ->get() failed\n", __func__);
1249                         goto out_exit_policy;
1250                 }
1251         }
1252
1253         /*
1254          * Sometimes boot loaders set CPU frequency to a value outside of
1255          * frequency table present with cpufreq core. In such cases CPU might be
1256          * unstable if it has to run on that frequency for long duration of time
1257          * and so its better to set it to a frequency which is specified in
1258          * freq-table. This also makes cpufreq stats inconsistent as
1259          * cpufreq-stats would fail to register because current frequency of CPU
1260          * isn't found in freq-table.
1261          *
1262          * Because we don't want this change to effect boot process badly, we go
1263          * for the next freq which is >= policy->cur ('cur' must be set by now,
1264          * otherwise we will end up setting freq to lowest of the table as 'cur'
1265          * is initialized to zero).
1266          *
1267          * We are passing target-freq as "policy->cur - 1" otherwise
1268          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1269          * equal to target-freq.
1270          */
1271         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1272             && has_target()) {
1273                 /* Are we running at unknown frequency ? */
1274                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1275                 if (ret == -EINVAL) {
1276                         /* Warn user and fix it */
1277                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1278                                 __func__, policy->cpu, policy->cur);
1279                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1280                                 CPUFREQ_RELATION_L);
1281
1282                         /*
1283                          * Reaching here after boot in a few seconds may not
1284                          * mean that system will remain stable at "unknown"
1285                          * frequency for longer duration. Hence, a BUG_ON().
1286                          */
1287                         BUG_ON(ret);
1288                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1289                                 __func__, policy->cpu, policy->cur);
1290                 }
1291         }
1292
1293         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1294                                      CPUFREQ_START, policy);
1295
1296         if (new_policy) {
1297                 ret = cpufreq_add_dev_interface(policy);
1298                 if (ret)
1299                         goto out_exit_policy;
1300                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1301                                 CPUFREQ_CREATE_POLICY, policy);
1302
1303                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1304                 list_add(&policy->policy_list, &cpufreq_policy_list);
1305                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1306         }
1307
1308         ret = cpufreq_init_policy(policy);
1309         if (ret) {
1310                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1311                        __func__, cpu, ret);
1312                 /* cpufreq_policy_free() will notify based on this */
1313                 new_policy = false;
1314                 goto out_exit_policy;
1315         }
1316
1317         up_write(&policy->rwsem);
1318
1319         kobject_uevent(&policy->kobj, KOBJ_ADD);
1320
1321         /* Callback for handling stuff after policy is ready */
1322         if (cpufreq_driver->ready)
1323                 cpufreq_driver->ready(policy);
1324
1325         pr_debug("initialization complete\n");
1326
1327         return 0;
1328
1329 out_exit_policy:
1330         up_write(&policy->rwsem);
1331
1332         if (cpufreq_driver->exit)
1333                 cpufreq_driver->exit(policy);
1334 out_free_policy:
1335         cpufreq_policy_free(policy, !new_policy);
1336         return ret;
1337 }
1338
1339 /**
1340  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1341  * @dev: CPU device.
1342  * @sif: Subsystem interface structure pointer (not used)
1343  */
1344 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1345 {
1346         unsigned cpu = dev->id;
1347         int ret;
1348
1349         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1350
1351         if (cpu_online(cpu)) {
1352                 ret = cpufreq_online(cpu);
1353         } else {
1354                 /*
1355                  * A hotplug notifier will follow and we will handle it as CPU
1356                  * online then.  For now, just create the sysfs link, unless
1357                  * there is no policy or the link is already present.
1358                  */
1359                 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1360
1361                 ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1362                         ? add_cpu_dev_symlink(policy, cpu) : 0;
1363         }
1364
1365         return ret;
1366 }
1367
1368 static void cpufreq_offline_prepare(unsigned int cpu)
1369 {
1370         struct cpufreq_policy *policy;
1371
1372         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1373
1374         policy = cpufreq_cpu_get_raw(cpu);
1375         if (!policy) {
1376                 pr_debug("%s: No cpu_data found\n", __func__);
1377                 return;
1378         }
1379
1380         if (has_target()) {
1381                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1382                 if (ret)
1383                         pr_err("%s: Failed to stop governor\n", __func__);
1384         }
1385
1386         down_write(&policy->rwsem);
1387         cpumask_clear_cpu(cpu, policy->cpus);
1388
1389         if (policy_is_inactive(policy)) {
1390                 if (has_target())
1391                         strncpy(policy->last_governor, policy->governor->name,
1392                                 CPUFREQ_NAME_LEN);
1393         } else if (cpu == policy->cpu) {
1394                 /* Nominate new CPU */
1395                 policy->cpu = cpumask_any(policy->cpus);
1396         }
1397         up_write(&policy->rwsem);
1398
1399         /* Start governor again for active policy */
1400         if (!policy_is_inactive(policy)) {
1401                 if (has_target()) {
1402                         int ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1403                         if (!ret)
1404                                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1405
1406                         if (ret)
1407                                 pr_err("%s: Failed to start governor\n", __func__);
1408                 }
1409         } else if (cpufreq_driver->stop_cpu) {
1410                 cpufreq_driver->stop_cpu(policy);
1411         }
1412 }
1413
1414 static void cpufreq_offline_finish(unsigned int cpu)
1415 {
1416         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1417
1418         if (!policy) {
1419                 pr_debug("%s: No cpu_data found\n", __func__);
1420                 return;
1421         }
1422
1423         /* Only proceed for inactive policies */
1424         if (!policy_is_inactive(policy))
1425                 return;
1426
1427         /* If cpu is last user of policy, free policy */
1428         if (has_target()) {
1429                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1430                 if (ret)
1431                         pr_err("%s: Failed to exit governor\n", __func__);
1432         }
1433
1434         /*
1435          * Perform the ->exit() even during light-weight tear-down,
1436          * since this is a core component, and is essential for the
1437          * subsequent light-weight ->init() to succeed.
1438          */
1439         if (cpufreq_driver->exit) {
1440                 cpufreq_driver->exit(policy);
1441                 policy->freq_table = NULL;
1442         }
1443 }
1444
1445 /**
1446  * cpufreq_remove_dev - remove a CPU device
1447  *
1448  * Removes the cpufreq interface for a CPU device.
1449  */
1450 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1451 {
1452         unsigned int cpu = dev->id;
1453         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1454
1455         if (!policy)
1456                 return;
1457
1458         if (cpu_online(cpu)) {
1459                 cpufreq_offline_prepare(cpu);
1460                 cpufreq_offline_finish(cpu);
1461         }
1462
1463         cpumask_clear_cpu(cpu, policy->real_cpus);
1464
1465         if (cpumask_empty(policy->real_cpus)) {
1466                 cpufreq_policy_free(policy, true);
1467                 return;
1468         }
1469
1470         if (cpu != policy->kobj_cpu) {
1471                 remove_cpu_dev_symlink(policy, cpu);
1472         } else {
1473                 /*
1474                  * The CPU owning the policy object is going away.  Move it to
1475                  * another suitable CPU.
1476                  */
1477                 unsigned int new_cpu = cpumask_first(policy->real_cpus);
1478                 struct device *new_dev = get_cpu_device(new_cpu);
1479
1480                 dev_dbg(dev, "%s: Moving policy object to CPU%u\n", __func__, new_cpu);
1481
1482                 sysfs_remove_link(&new_dev->kobj, "cpufreq");
1483                 policy->kobj_cpu = new_cpu;
1484                 WARN_ON(kobject_move(&policy->kobj, &new_dev->kobj));
1485         }
1486 }
1487
1488 static void handle_update(struct work_struct *work)
1489 {
1490         struct cpufreq_policy *policy =
1491                 container_of(work, struct cpufreq_policy, update);
1492         unsigned int cpu = policy->cpu;
1493         pr_debug("handle_update for cpu %u called\n", cpu);
1494         cpufreq_update_policy(cpu);
1495 }
1496
1497 /**
1498  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1499  *      in deep trouble.
1500  *      @policy: policy managing CPUs
1501  *      @new_freq: CPU frequency the CPU actually runs at
1502  *
1503  *      We adjust to current frequency first, and need to clean up later.
1504  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1505  */
1506 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1507                                 unsigned int new_freq)
1508 {
1509         struct cpufreq_freqs freqs;
1510
1511         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1512                  policy->cur, new_freq);
1513
1514         freqs.old = policy->cur;
1515         freqs.new = new_freq;
1516
1517         cpufreq_freq_transition_begin(policy, &freqs);
1518         cpufreq_freq_transition_end(policy, &freqs, 0);
1519 }
1520
1521 /**
1522  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1523  * @cpu: CPU number
1524  *
1525  * This is the last known freq, without actually getting it from the driver.
1526  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1527  */
1528 unsigned int cpufreq_quick_get(unsigned int cpu)
1529 {
1530         struct cpufreq_policy *policy;
1531         unsigned int ret_freq = 0;
1532
1533         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1534                 return cpufreq_driver->get(cpu);
1535
1536         policy = cpufreq_cpu_get(cpu);
1537         if (policy) {
1538                 ret_freq = policy->cur;
1539                 cpufreq_cpu_put(policy);
1540         }
1541
1542         return ret_freq;
1543 }
1544 EXPORT_SYMBOL(cpufreq_quick_get);
1545
1546 /**
1547  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1548  * @cpu: CPU number
1549  *
1550  * Just return the max possible frequency for a given CPU.
1551  */
1552 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1553 {
1554         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1555         unsigned int ret_freq = 0;
1556
1557         if (policy) {
1558                 ret_freq = policy->max;
1559                 cpufreq_cpu_put(policy);
1560         }
1561
1562         return ret_freq;
1563 }
1564 EXPORT_SYMBOL(cpufreq_quick_get_max);
1565
1566 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1567 {
1568         unsigned int ret_freq = 0;
1569
1570         if (!cpufreq_driver->get)
1571                 return ret_freq;
1572
1573         ret_freq = cpufreq_driver->get(policy->cpu);
1574
1575         /* Updating inactive policies is invalid, so avoid doing that. */
1576         if (unlikely(policy_is_inactive(policy)))
1577                 return ret_freq;
1578
1579         if (ret_freq && policy->cur &&
1580                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1581                 /* verify no discrepancy between actual and
1582                                         saved value exists */
1583                 if (unlikely(ret_freq != policy->cur)) {
1584                         cpufreq_out_of_sync(policy, ret_freq);
1585                         schedule_work(&policy->update);
1586                 }
1587         }
1588
1589         return ret_freq;
1590 }
1591
1592 /**
1593  * cpufreq_get - get the current CPU frequency (in kHz)
1594  * @cpu: CPU number
1595  *
1596  * Get the CPU current (static) CPU frequency
1597  */
1598 unsigned int cpufreq_get(unsigned int cpu)
1599 {
1600         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1601         unsigned int ret_freq = 0;
1602
1603         if (policy) {
1604                 down_read(&policy->rwsem);
1605                 ret_freq = __cpufreq_get(policy);
1606                 up_read(&policy->rwsem);
1607
1608                 cpufreq_cpu_put(policy);
1609         }
1610
1611         return ret_freq;
1612 }
1613 EXPORT_SYMBOL(cpufreq_get);
1614
1615 static struct subsys_interface cpufreq_interface = {
1616         .name           = "cpufreq",
1617         .subsys         = &cpu_subsys,
1618         .add_dev        = cpufreq_add_dev,
1619         .remove_dev     = cpufreq_remove_dev,
1620 };
1621
1622 /*
1623  * In case platform wants some specific frequency to be configured
1624  * during suspend..
1625  */
1626 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1627 {
1628         int ret;
1629
1630         if (!policy->suspend_freq) {
1631                 pr_debug("%s: suspend_freq not defined\n", __func__);
1632                 return 0;
1633         }
1634
1635         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1636                         policy->suspend_freq);
1637
1638         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1639                         CPUFREQ_RELATION_H);
1640         if (ret)
1641                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1642                                 __func__, policy->suspend_freq, ret);
1643
1644         return ret;
1645 }
1646 EXPORT_SYMBOL(cpufreq_generic_suspend);
1647
1648 /**
1649  * cpufreq_suspend() - Suspend CPUFreq governors
1650  *
1651  * Called during system wide Suspend/Hibernate cycles for suspending governors
1652  * as some platforms can't change frequency after this point in suspend cycle.
1653  * Because some of the devices (like: i2c, regulators, etc) they use for
1654  * changing frequency are suspended quickly after this point.
1655  */
1656 void cpufreq_suspend(void)
1657 {
1658         struct cpufreq_policy *policy;
1659
1660         if (!cpufreq_driver)
1661                 return;
1662
1663         if (!has_target())
1664                 goto suspend;
1665
1666         pr_debug("%s: Suspending Governors\n", __func__);
1667
1668         for_each_active_policy(policy) {
1669                 if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1670                         pr_err("%s: Failed to stop governor for policy: %p\n",
1671                                 __func__, policy);
1672                 else if (cpufreq_driver->suspend
1673                     && cpufreq_driver->suspend(policy))
1674                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1675                                 policy);
1676         }
1677
1678 suspend:
1679         cpufreq_suspended = true;
1680 }
1681
1682 /**
1683  * cpufreq_resume() - Resume CPUFreq governors
1684  *
1685  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1686  * are suspended with cpufreq_suspend().
1687  */
1688 void cpufreq_resume(void)
1689 {
1690         struct cpufreq_policy *policy;
1691
1692         if (!cpufreq_driver)
1693                 return;
1694
1695         cpufreq_suspended = false;
1696
1697         if (!has_target())
1698                 return;
1699
1700         pr_debug("%s: Resuming Governors\n", __func__);
1701
1702         for_each_active_policy(policy) {
1703                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1704                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1705                                 policy);
1706                 else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1707                     || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1708                         pr_err("%s: Failed to start governor for policy: %p\n",
1709                                 __func__, policy);
1710         }
1711
1712         /*
1713          * schedule call cpufreq_update_policy() for first-online CPU, as that
1714          * wouldn't be hotplugged-out on suspend. It will verify that the
1715          * current freq is in sync with what we believe it to be.
1716          */
1717         policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1718         if (WARN_ON(!policy))
1719                 return;
1720
1721         schedule_work(&policy->update);
1722 }
1723
1724 /**
1725  *      cpufreq_get_current_driver - return current driver's name
1726  *
1727  *      Return the name string of the currently loaded cpufreq driver
1728  *      or NULL, if none.
1729  */
1730 const char *cpufreq_get_current_driver(void)
1731 {
1732         if (cpufreq_driver)
1733                 return cpufreq_driver->name;
1734
1735         return NULL;
1736 }
1737 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1738
1739 /**
1740  *      cpufreq_get_driver_data - return current driver data
1741  *
1742  *      Return the private data of the currently loaded cpufreq
1743  *      driver, or NULL if no cpufreq driver is loaded.
1744  */
1745 void *cpufreq_get_driver_data(void)
1746 {
1747         if (cpufreq_driver)
1748                 return cpufreq_driver->driver_data;
1749
1750         return NULL;
1751 }
1752 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1753
1754 /*********************************************************************
1755  *                     NOTIFIER LISTS INTERFACE                      *
1756  *********************************************************************/
1757
1758 /**
1759  *      cpufreq_register_notifier - register a driver with cpufreq
1760  *      @nb: notifier function to register
1761  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1762  *
1763  *      Add a driver to one of two lists: either a list of drivers that
1764  *      are notified about clock rate changes (once before and once after
1765  *      the transition), or a list of drivers that are notified about
1766  *      changes in cpufreq policy.
1767  *
1768  *      This function may sleep, and has the same return conditions as
1769  *      blocking_notifier_chain_register.
1770  */
1771 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1772 {
1773         int ret;
1774
1775         if (cpufreq_disabled())
1776                 return -EINVAL;
1777
1778         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1779
1780         switch (list) {
1781         case CPUFREQ_TRANSITION_NOTIFIER:
1782                 ret = srcu_notifier_chain_register(
1783                                 &cpufreq_transition_notifier_list, nb);
1784                 break;
1785         case CPUFREQ_POLICY_NOTIFIER:
1786                 ret = blocking_notifier_chain_register(
1787                                 &cpufreq_policy_notifier_list, nb);
1788                 break;
1789         default:
1790                 ret = -EINVAL;
1791         }
1792
1793         return ret;
1794 }
1795 EXPORT_SYMBOL(cpufreq_register_notifier);
1796
1797 /**
1798  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1799  *      @nb: notifier block to be unregistered
1800  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1801  *
1802  *      Remove a driver from the CPU frequency notifier list.
1803  *
1804  *      This function may sleep, and has the same return conditions as
1805  *      blocking_notifier_chain_unregister.
1806  */
1807 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1808 {
1809         int ret;
1810
1811         if (cpufreq_disabled())
1812                 return -EINVAL;
1813
1814         switch (list) {
1815         case CPUFREQ_TRANSITION_NOTIFIER:
1816                 ret = srcu_notifier_chain_unregister(
1817                                 &cpufreq_transition_notifier_list, nb);
1818                 break;
1819         case CPUFREQ_POLICY_NOTIFIER:
1820                 ret = blocking_notifier_chain_unregister(
1821                                 &cpufreq_policy_notifier_list, nb);
1822                 break;
1823         default:
1824                 ret = -EINVAL;
1825         }
1826
1827         return ret;
1828 }
1829 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1830
1831
1832 /*********************************************************************
1833  *                              GOVERNORS                            *
1834  *********************************************************************/
1835
1836 /* Must set freqs->new to intermediate frequency */
1837 static int __target_intermediate(struct cpufreq_policy *policy,
1838                                  struct cpufreq_freqs *freqs, int index)
1839 {
1840         int ret;
1841
1842         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1843
1844         /* We don't need to switch to intermediate freq */
1845         if (!freqs->new)
1846                 return 0;
1847
1848         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1849                  __func__, policy->cpu, freqs->old, freqs->new);
1850
1851         cpufreq_freq_transition_begin(policy, freqs);
1852         ret = cpufreq_driver->target_intermediate(policy, index);
1853         cpufreq_freq_transition_end(policy, freqs, ret);
1854
1855         if (ret)
1856                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1857                        __func__, ret);
1858
1859         return ret;
1860 }
1861
1862 static int __target_index(struct cpufreq_policy *policy,
1863                           struct cpufreq_frequency_table *freq_table, int index)
1864 {
1865         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1866         unsigned int intermediate_freq = 0;
1867         int retval = -EINVAL;
1868         bool notify;
1869
1870         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1871         if (notify) {
1872                 /* Handle switching to intermediate frequency */
1873                 if (cpufreq_driver->get_intermediate) {
1874                         retval = __target_intermediate(policy, &freqs, index);
1875                         if (retval)
1876                                 return retval;
1877
1878                         intermediate_freq = freqs.new;
1879                         /* Set old freq to intermediate */
1880                         if (intermediate_freq)
1881                                 freqs.old = freqs.new;
1882                 }
1883
1884                 freqs.new = freq_table[index].frequency;
1885                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1886                          __func__, policy->cpu, freqs.old, freqs.new);
1887
1888                 cpufreq_freq_transition_begin(policy, &freqs);
1889         }
1890
1891         retval = cpufreq_driver->target_index(policy, index);
1892         if (retval)
1893                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1894                        retval);
1895
1896         if (notify) {
1897                 cpufreq_freq_transition_end(policy, &freqs, retval);
1898
1899                 /*
1900                  * Failed after setting to intermediate freq? Driver should have
1901                  * reverted back to initial frequency and so should we. Check
1902                  * here for intermediate_freq instead of get_intermediate, in
1903                  * case we haven't switched to intermediate freq at all.
1904                  */
1905                 if (unlikely(retval && intermediate_freq)) {
1906                         freqs.old = intermediate_freq;
1907                         freqs.new = policy->restore_freq;
1908                         cpufreq_freq_transition_begin(policy, &freqs);
1909                         cpufreq_freq_transition_end(policy, &freqs, 0);
1910                 }
1911         }
1912
1913         return retval;
1914 }
1915
1916 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1917                             unsigned int target_freq,
1918                             unsigned int relation)
1919 {
1920         unsigned int old_target_freq = target_freq;
1921         int retval = -EINVAL;
1922
1923         if (cpufreq_disabled())
1924                 return -ENODEV;
1925
1926         /* Make sure that target_freq is within supported range */
1927         if (target_freq > policy->max)
1928                 target_freq = policy->max;
1929         if (target_freq < policy->min)
1930                 target_freq = policy->min;
1931
1932         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1933                  policy->cpu, target_freq, relation, old_target_freq);
1934
1935         /*
1936          * This might look like a redundant call as we are checking it again
1937          * after finding index. But it is left intentionally for cases where
1938          * exactly same freq is called again and so we can save on few function
1939          * calls.
1940          */
1941         if (target_freq == policy->cur)
1942                 return 0;
1943
1944         /* Save last value to restore later on errors */
1945         policy->restore_freq = policy->cur;
1946
1947         if (cpufreq_driver->target)
1948                 retval = cpufreq_driver->target(policy, target_freq, relation);
1949         else if (cpufreq_driver->target_index) {
1950                 struct cpufreq_frequency_table *freq_table;
1951                 int index;
1952
1953                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1954                 if (unlikely(!freq_table)) {
1955                         pr_err("%s: Unable to find freq_table\n", __func__);
1956                         goto out;
1957                 }
1958
1959                 retval = cpufreq_frequency_table_target(policy, freq_table,
1960                                 target_freq, relation, &index);
1961                 if (unlikely(retval)) {
1962                         pr_err("%s: Unable to find matching freq\n", __func__);
1963                         goto out;
1964                 }
1965
1966                 if (freq_table[index].frequency == policy->cur) {
1967                         retval = 0;
1968                         goto out;
1969                 }
1970
1971                 retval = __target_index(policy, freq_table, index);
1972         }
1973
1974 out:
1975         return retval;
1976 }
1977 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1978
1979 int cpufreq_driver_target(struct cpufreq_policy *policy,
1980                           unsigned int target_freq,
1981                           unsigned int relation)
1982 {
1983         int ret = -EINVAL;
1984
1985         down_write(&policy->rwsem);
1986
1987         ret = __cpufreq_driver_target(policy, target_freq, relation);
1988
1989         up_write(&policy->rwsem);
1990
1991         return ret;
1992 }
1993 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1994
1995 static int __cpufreq_governor(struct cpufreq_policy *policy,
1996                                         unsigned int event)
1997 {
1998         int ret;
1999
2000         /* Only must be defined when default governor is known to have latency
2001            restrictions, like e.g. conservative or ondemand.
2002            That this is the case is already ensured in Kconfig
2003         */
2004 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
2005         struct cpufreq_governor *gov = &cpufreq_gov_performance;
2006 #else
2007         struct cpufreq_governor *gov = NULL;
2008 #endif
2009
2010         /* Don't start any governor operations if we are entering suspend */
2011         if (cpufreq_suspended)
2012                 return 0;
2013         /*
2014          * Governor might not be initiated here if ACPI _PPC changed
2015          * notification happened, so check it.
2016          */
2017         if (!policy->governor)
2018                 return -EINVAL;
2019
2020         if (policy->governor->max_transition_latency &&
2021             policy->cpuinfo.transition_latency >
2022             policy->governor->max_transition_latency) {
2023                 if (!gov)
2024                         return -EINVAL;
2025                 else {
2026                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2027                                 policy->governor->name, gov->name);
2028                         policy->governor = gov;
2029                 }
2030         }
2031
2032         if (event == CPUFREQ_GOV_POLICY_INIT)
2033                 if (!try_module_get(policy->governor->owner))
2034                         return -EINVAL;
2035
2036         pr_debug("%s: for CPU %u, event %u\n", __func__, policy->cpu, event);
2037
2038         mutex_lock(&cpufreq_governor_lock);
2039         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2040             || (!policy->governor_enabled
2041             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2042                 mutex_unlock(&cpufreq_governor_lock);
2043                 return -EBUSY;
2044         }
2045
2046         if (event == CPUFREQ_GOV_STOP)
2047                 policy->governor_enabled = false;
2048         else if (event == CPUFREQ_GOV_START)
2049                 policy->governor_enabled = true;
2050
2051         mutex_unlock(&cpufreq_governor_lock);
2052
2053         ret = policy->governor->governor(policy, event);
2054
2055         if (!ret) {
2056                 if (event == CPUFREQ_GOV_POLICY_INIT)
2057                         policy->governor->initialized++;
2058                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
2059                         policy->governor->initialized--;
2060         } else {
2061                 /* Restore original values */
2062                 mutex_lock(&cpufreq_governor_lock);
2063                 if (event == CPUFREQ_GOV_STOP)
2064                         policy->governor_enabled = true;
2065                 else if (event == CPUFREQ_GOV_START)
2066                         policy->governor_enabled = false;
2067                 mutex_unlock(&cpufreq_governor_lock);
2068         }
2069
2070         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2071                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2072                 module_put(policy->governor->owner);
2073
2074         return ret;
2075 }
2076
2077 int cpufreq_register_governor(struct cpufreq_governor *governor)
2078 {
2079         int err;
2080
2081         if (!governor)
2082                 return -EINVAL;
2083
2084         if (cpufreq_disabled())
2085                 return -ENODEV;
2086
2087         mutex_lock(&cpufreq_governor_mutex);
2088
2089         governor->initialized = 0;
2090         err = -EBUSY;
2091         if (!find_governor(governor->name)) {
2092                 err = 0;
2093                 list_add(&governor->governor_list, &cpufreq_governor_list);
2094         }
2095
2096         mutex_unlock(&cpufreq_governor_mutex);
2097         return err;
2098 }
2099 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2100
2101 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2102 {
2103         struct cpufreq_policy *policy;
2104         unsigned long flags;
2105
2106         if (!governor)
2107                 return;
2108
2109         if (cpufreq_disabled())
2110                 return;
2111
2112         /* clear last_governor for all inactive policies */
2113         read_lock_irqsave(&cpufreq_driver_lock, flags);
2114         for_each_inactive_policy(policy) {
2115                 if (!strcmp(policy->last_governor, governor->name)) {
2116                         policy->governor = NULL;
2117                         strcpy(policy->last_governor, "\0");
2118                 }
2119         }
2120         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2121
2122         mutex_lock(&cpufreq_governor_mutex);
2123         list_del(&governor->governor_list);
2124         mutex_unlock(&cpufreq_governor_mutex);
2125         return;
2126 }
2127 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2128
2129
2130 /*********************************************************************
2131  *                          POLICY INTERFACE                         *
2132  *********************************************************************/
2133
2134 /**
2135  * cpufreq_get_policy - get the current cpufreq_policy
2136  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2137  *      is written
2138  *
2139  * Reads the current cpufreq policy.
2140  */
2141 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2142 {
2143         struct cpufreq_policy *cpu_policy;
2144         if (!policy)
2145                 return -EINVAL;
2146
2147         cpu_policy = cpufreq_cpu_get(cpu);
2148         if (!cpu_policy)
2149                 return -EINVAL;
2150
2151         memcpy(policy, cpu_policy, sizeof(*policy));
2152
2153         cpufreq_cpu_put(cpu_policy);
2154         return 0;
2155 }
2156 EXPORT_SYMBOL(cpufreq_get_policy);
2157
2158 /*
2159  * policy : current policy.
2160  * new_policy: policy to be set.
2161  */
2162 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2163                                 struct cpufreq_policy *new_policy)
2164 {
2165         struct cpufreq_governor *old_gov;
2166         int ret;
2167
2168         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2169                  new_policy->cpu, new_policy->min, new_policy->max);
2170
2171         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2172
2173         /*
2174         * This check works well when we store new min/max freq attributes,
2175         * because new_policy is a copy of policy with one field updated.
2176         */
2177         if (new_policy->min > new_policy->max)
2178                 return -EINVAL;
2179
2180         /* verify the cpu speed can be set within this limit */
2181         ret = cpufreq_driver->verify(new_policy);
2182         if (ret)
2183                 return ret;
2184
2185         /* adjust if necessary - all reasons */
2186         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2187                         CPUFREQ_ADJUST, new_policy);
2188
2189         /*
2190          * verify the cpu speed can be set within this limit, which might be
2191          * different to the first one
2192          */
2193         ret = cpufreq_driver->verify(new_policy);
2194         if (ret)
2195                 return ret;
2196
2197         /* notification of the new policy */
2198         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2199                         CPUFREQ_NOTIFY, new_policy);
2200
2201         policy->min = new_policy->min;
2202         policy->max = new_policy->max;
2203
2204         pr_debug("new min and max freqs are %u - %u kHz\n",
2205                  policy->min, policy->max);
2206
2207         if (cpufreq_driver->setpolicy) {
2208                 policy->policy = new_policy->policy;
2209                 pr_debug("setting range\n");
2210                 return cpufreq_driver->setpolicy(new_policy);
2211         }
2212
2213         if (new_policy->governor == policy->governor)
2214                 goto out;
2215
2216         pr_debug("governor switch\n");
2217
2218         /* save old, working values */
2219         old_gov = policy->governor;
2220         /* end old governor */
2221         if (old_gov) {
2222                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2223                 if (ret) {
2224                         /* This can happen due to race with other operations */
2225                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2226                                  __func__, old_gov->name, ret);
2227                         return ret;
2228                 }
2229
2230                 up_write(&policy->rwsem);
2231                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2232                 down_write(&policy->rwsem);
2233
2234                 if (ret) {
2235                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2236                                __func__, old_gov->name, ret);
2237                         return ret;
2238                 }
2239         }
2240
2241         /* start new governor */
2242         policy->governor = new_policy->governor;
2243         ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2244         if (!ret) {
2245                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
2246                 if (!ret)
2247                         goto out;
2248
2249                 up_write(&policy->rwsem);
2250                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2251                 down_write(&policy->rwsem);
2252         }
2253
2254         /* new governor failed, so re-start old one */
2255         pr_debug("starting governor %s failed\n", policy->governor->name);
2256         if (old_gov) {
2257                 policy->governor = old_gov;
2258                 if (__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2259                         policy->governor = NULL;
2260                 else
2261                         __cpufreq_governor(policy, CPUFREQ_GOV_START);
2262         }
2263
2264         return ret;
2265
2266  out:
2267         pr_debug("governor: change or update limits\n");
2268         return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2269 }
2270
2271 /**
2272  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2273  *      @cpu: CPU which shall be re-evaluated
2274  *
2275  *      Useful for policy notifiers which have different necessities
2276  *      at different times.
2277  */
2278 int cpufreq_update_policy(unsigned int cpu)
2279 {
2280         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2281         struct cpufreq_policy new_policy;
2282         int ret;
2283
2284         if (!policy)
2285                 return -ENODEV;
2286
2287         down_write(&policy->rwsem);
2288
2289         pr_debug("updating policy for CPU %u\n", cpu);
2290         memcpy(&new_policy, policy, sizeof(*policy));
2291         new_policy.min = policy->user_policy.min;
2292         new_policy.max = policy->user_policy.max;
2293
2294         /*
2295          * BIOS might change freq behind our back
2296          * -> ask driver for current freq and notify governors about a change
2297          */
2298         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2299                 new_policy.cur = cpufreq_driver->get(cpu);
2300                 if (WARN_ON(!new_policy.cur)) {
2301                         ret = -EIO;
2302                         goto unlock;
2303                 }
2304
2305                 if (!policy->cur) {
2306                         pr_debug("Driver did not initialize current freq\n");
2307                         policy->cur = new_policy.cur;
2308                 } else {
2309                         if (policy->cur != new_policy.cur && has_target())
2310                                 cpufreq_out_of_sync(policy, new_policy.cur);
2311                 }
2312         }
2313
2314         ret = cpufreq_set_policy(policy, &new_policy);
2315
2316 unlock:
2317         up_write(&policy->rwsem);
2318
2319         cpufreq_cpu_put(policy);
2320         return ret;
2321 }
2322 EXPORT_SYMBOL(cpufreq_update_policy);
2323
2324 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2325                                         unsigned long action, void *hcpu)
2326 {
2327         unsigned int cpu = (unsigned long)hcpu;
2328
2329         switch (action & ~CPU_TASKS_FROZEN) {
2330         case CPU_ONLINE:
2331                 cpufreq_online(cpu);
2332                 break;
2333
2334         case CPU_DOWN_PREPARE:
2335                 cpufreq_offline_prepare(cpu);
2336                 break;
2337
2338         case CPU_POST_DEAD:
2339                 cpufreq_offline_finish(cpu);
2340                 break;
2341
2342         case CPU_DOWN_FAILED:
2343                 cpufreq_online(cpu);
2344                 break;
2345         }
2346         return NOTIFY_OK;
2347 }
2348
2349 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2350         .notifier_call = cpufreq_cpu_callback,
2351 };
2352
2353 /*********************************************************************
2354  *               BOOST                                               *
2355  *********************************************************************/
2356 static int cpufreq_boost_set_sw(int state)
2357 {
2358         struct cpufreq_frequency_table *freq_table;
2359         struct cpufreq_policy *policy;
2360         int ret = -EINVAL;
2361
2362         for_each_active_policy(policy) {
2363                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2364                 if (freq_table) {
2365                         ret = cpufreq_frequency_table_cpuinfo(policy,
2366                                                         freq_table);
2367                         if (ret) {
2368                                 pr_err("%s: Policy frequency update failed\n",
2369                                        __func__);
2370                                 break;
2371                         }
2372                         policy->user_policy.max = policy->max;
2373                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2374                 }
2375         }
2376
2377         return ret;
2378 }
2379
2380 int cpufreq_boost_trigger_state(int state)
2381 {
2382         unsigned long flags;
2383         int ret = 0;
2384
2385         if (cpufreq_driver->boost_enabled == state)
2386                 return 0;
2387
2388         write_lock_irqsave(&cpufreq_driver_lock, flags);
2389         cpufreq_driver->boost_enabled = state;
2390         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2391
2392         ret = cpufreq_driver->set_boost(state);
2393         if (ret) {
2394                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2395                 cpufreq_driver->boost_enabled = !state;
2396                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2397
2398                 pr_err("%s: Cannot %s BOOST\n",
2399                        __func__, state ? "enable" : "disable");
2400         }
2401
2402         return ret;
2403 }
2404
2405 int cpufreq_boost_supported(void)
2406 {
2407         if (likely(cpufreq_driver))
2408                 return cpufreq_driver->boost_supported;
2409
2410         return 0;
2411 }
2412 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2413
2414 static int create_boost_sysfs_file(void)
2415 {
2416         int ret;
2417
2418         if (!cpufreq_boost_supported())
2419                 return 0;
2420
2421         /*
2422          * Check if driver provides function to enable boost -
2423          * if not, use cpufreq_boost_set_sw as default
2424          */
2425         if (!cpufreq_driver->set_boost)
2426                 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2427
2428         ret = cpufreq_sysfs_create_file(&boost.attr);
2429         if (ret)
2430                 pr_err("%s: cannot register global BOOST sysfs file\n",
2431                        __func__);
2432
2433         return ret;
2434 }
2435
2436 static void remove_boost_sysfs_file(void)
2437 {
2438         if (cpufreq_boost_supported())
2439                 cpufreq_sysfs_remove_file(&boost.attr);
2440 }
2441
2442 int cpufreq_enable_boost_support(void)
2443 {
2444         if (!cpufreq_driver)
2445                 return -EINVAL;
2446
2447         if (cpufreq_boost_supported())
2448                 return 0;
2449
2450         cpufreq_driver->boost_supported = true;
2451
2452         /* This will get removed on driver unregister */
2453         return create_boost_sysfs_file();
2454 }
2455 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2456
2457 int cpufreq_boost_enabled(void)
2458 {
2459         return cpufreq_driver->boost_enabled;
2460 }
2461 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2462
2463 /*********************************************************************
2464  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2465  *********************************************************************/
2466
2467 /**
2468  * cpufreq_register_driver - register a CPU Frequency driver
2469  * @driver_data: A struct cpufreq_driver containing the values#
2470  * submitted by the CPU Frequency driver.
2471  *
2472  * Registers a CPU Frequency driver to this core code. This code
2473  * returns zero on success, -EBUSY when another driver got here first
2474  * (and isn't unregistered in the meantime).
2475  *
2476  */
2477 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2478 {
2479         unsigned long flags;
2480         int ret;
2481
2482         if (cpufreq_disabled())
2483                 return -ENODEV;
2484
2485         if (!driver_data || !driver_data->verify || !driver_data->init ||
2486             !(driver_data->setpolicy || driver_data->target_index ||
2487                     driver_data->target) ||
2488              (driver_data->setpolicy && (driver_data->target_index ||
2489                     driver_data->target)) ||
2490              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2491                 return -EINVAL;
2492
2493         pr_debug("trying to register driver %s\n", driver_data->name);
2494
2495         /* Protect against concurrent CPU online/offline. */
2496         get_online_cpus();
2497
2498         write_lock_irqsave(&cpufreq_driver_lock, flags);
2499         if (cpufreq_driver) {
2500                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2501                 ret = -EEXIST;
2502                 goto out;
2503         }
2504         cpufreq_driver = driver_data;
2505         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2506
2507         if (driver_data->setpolicy)
2508                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2509
2510         ret = create_boost_sysfs_file();
2511         if (ret)
2512                 goto err_null_driver;
2513
2514         ret = subsys_interface_register(&cpufreq_interface);
2515         if (ret)
2516                 goto err_boost_unreg;
2517
2518         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2519             list_empty(&cpufreq_policy_list)) {
2520                 /* if all ->init() calls failed, unregister */
2521                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2522                          driver_data->name);
2523                 goto err_if_unreg;
2524         }
2525
2526         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2527         pr_debug("driver %s up and running\n", driver_data->name);
2528
2529 out:
2530         put_online_cpus();
2531         return ret;
2532
2533 err_if_unreg:
2534         subsys_interface_unregister(&cpufreq_interface);
2535 err_boost_unreg:
2536         remove_boost_sysfs_file();
2537 err_null_driver:
2538         write_lock_irqsave(&cpufreq_driver_lock, flags);
2539         cpufreq_driver = NULL;
2540         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2541         goto out;
2542 }
2543 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2544
2545 /**
2546  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2547  *
2548  * Unregister the current CPUFreq driver. Only call this if you have
2549  * the right to do so, i.e. if you have succeeded in initialising before!
2550  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2551  * currently not initialised.
2552  */
2553 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2554 {
2555         unsigned long flags;
2556
2557         if (!cpufreq_driver || (driver != cpufreq_driver))
2558                 return -EINVAL;
2559
2560         pr_debug("unregistering driver %s\n", driver->name);
2561
2562         /* Protect against concurrent cpu hotplug */
2563         get_online_cpus();
2564         subsys_interface_unregister(&cpufreq_interface);
2565         remove_boost_sysfs_file();
2566         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2567
2568         write_lock_irqsave(&cpufreq_driver_lock, flags);
2569
2570         cpufreq_driver = NULL;
2571
2572         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2573         put_online_cpus();
2574
2575         return 0;
2576 }
2577 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2578
2579 /*
2580  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2581  * or mutexes when secondary CPUs are halted.
2582  */
2583 static struct syscore_ops cpufreq_syscore_ops = {
2584         .shutdown = cpufreq_suspend,
2585 };
2586
2587 static int __init cpufreq_core_init(void)
2588 {
2589         if (cpufreq_disabled())
2590                 return -ENODEV;
2591
2592         cpufreq_global_kobject = kobject_create();
2593         BUG_ON(!cpufreq_global_kobject);
2594
2595         register_syscore_ops(&cpufreq_syscore_ops);
2596
2597         return 0;
2598 }
2599 core_initcall(cpufreq_core_init);