3dd4884c6f9ef5e609959d26c026606ad85bf1b3
[cascardo/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 /* Macros to iterate over CPU policies */
42 #define for_each_suitable_policy(__policy, __active)                     \
43         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
44                 if ((__active) == !policy_is_inactive(__policy))
45
46 #define for_each_active_policy(__policy)                \
47         for_each_suitable_policy(__policy, true)
48 #define for_each_inactive_policy(__policy)              \
49         for_each_suitable_policy(__policy, false)
50
51 #define for_each_policy(__policy)                       \
52         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
53
54 /* Iterate over governors */
55 static LIST_HEAD(cpufreq_governor_list);
56 #define for_each_governor(__governor)                           \
57         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
58
59 /**
60  * The "cpufreq driver" - the arch- or hardware-dependent low
61  * level driver of CPUFreq support, and its spinlock. This lock
62  * also protects the cpufreq_cpu_data array.
63  */
64 static struct cpufreq_driver *cpufreq_driver;
65 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
66 static DEFINE_RWLOCK(cpufreq_driver_lock);
67
68 /* Flag to suspend/resume CPUFreq governors */
69 static bool cpufreq_suspended;
70
71 static inline bool has_target(void)
72 {
73         return cpufreq_driver->target_index || cpufreq_driver->target;
74 }
75
76 /* internal prototypes */
77 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
78 static int cpufreq_init_governor(struct cpufreq_policy *policy);
79 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
80 static int cpufreq_start_governor(struct cpufreq_policy *policy);
81 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
82 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
83
84 /**
85  * Two notifier lists: the "policy" list is involved in the
86  * validation process for a new CPU frequency policy; the
87  * "transition" list for kernel code that needs to handle
88  * changes to devices when the CPU clock speed changes.
89  * The mutex locks both lists.
90  */
91 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
92 static struct srcu_notifier_head cpufreq_transition_notifier_list;
93
94 static bool init_cpufreq_transition_notifier_list_called;
95 static int __init init_cpufreq_transition_notifier_list(void)
96 {
97         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
98         init_cpufreq_transition_notifier_list_called = true;
99         return 0;
100 }
101 pure_initcall(init_cpufreq_transition_notifier_list);
102
103 static int off __read_mostly;
104 static int cpufreq_disabled(void)
105 {
106         return off;
107 }
108 void disable_cpufreq(void)
109 {
110         off = 1;
111 }
112 static DEFINE_MUTEX(cpufreq_governor_mutex);
113
114 bool have_governor_per_policy(void)
115 {
116         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
117 }
118 EXPORT_SYMBOL_GPL(have_governor_per_policy);
119
120 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
121 {
122         if (have_governor_per_policy())
123                 return &policy->kobj;
124         else
125                 return cpufreq_global_kobject;
126 }
127 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
128
129 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
130 {
131         u64 idle_time;
132         u64 cur_wall_time;
133         u64 busy_time;
134
135         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
136
137         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
138         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
139         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
140         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
141         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
142         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
143
144         idle_time = cur_wall_time - busy_time;
145         if (wall)
146                 *wall = cputime_to_usecs(cur_wall_time);
147
148         return cputime_to_usecs(idle_time);
149 }
150
151 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
152 {
153         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
154
155         if (idle_time == -1ULL)
156                 return get_cpu_idle_time_jiffy(cpu, wall);
157         else if (!io_busy)
158                 idle_time += get_cpu_iowait_time_us(cpu, wall);
159
160         return idle_time;
161 }
162 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
163
164 /*
165  * This is a generic cpufreq init() routine which can be used by cpufreq
166  * drivers of SMP systems. It will do following:
167  * - validate & show freq table passed
168  * - set policies transition latency
169  * - policy->cpus with all possible CPUs
170  */
171 int cpufreq_generic_init(struct cpufreq_policy *policy,
172                 struct cpufreq_frequency_table *table,
173                 unsigned int transition_latency)
174 {
175         int ret;
176
177         ret = cpufreq_table_validate_and_show(policy, table);
178         if (ret) {
179                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
180                 return ret;
181         }
182
183         policy->cpuinfo.transition_latency = transition_latency;
184
185         /*
186          * The driver only supports the SMP configuration where all processors
187          * share the clock and voltage and clock.
188          */
189         cpumask_setall(policy->cpus);
190
191         return 0;
192 }
193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
194
195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
196 {
197         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
198
199         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
200 }
201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
202
203 unsigned int cpufreq_generic_get(unsigned int cpu)
204 {
205         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
206
207         if (!policy || IS_ERR(policy->clk)) {
208                 pr_err("%s: No %s associated to cpu: %d\n",
209                        __func__, policy ? "clk" : "policy", cpu);
210                 return 0;
211         }
212
213         return clk_get_rate(policy->clk) / 1000;
214 }
215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
216
217 /**
218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
219  *
220  * @cpu: cpu to find policy for.
221  *
222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
223  * It also increments the kobject reference count to mark it busy and so would
224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
226  * freed as that depends on the kobj count.
227  *
228  * Return: A valid policy on success, otherwise NULL on failure.
229  */
230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
231 {
232         struct cpufreq_policy *policy = NULL;
233         unsigned long flags;
234
235         if (WARN_ON(cpu >= nr_cpu_ids))
236                 return NULL;
237
238         /* get the cpufreq driver */
239         read_lock_irqsave(&cpufreq_driver_lock, flags);
240
241         if (cpufreq_driver) {
242                 /* get the CPU */
243                 policy = cpufreq_cpu_get_raw(cpu);
244                 if (policy)
245                         kobject_get(&policy->kobj);
246         }
247
248         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
249
250         return policy;
251 }
252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
253
254 /**
255  * cpufreq_cpu_put: Decrements the usage count of a policy
256  *
257  * @policy: policy earlier returned by cpufreq_cpu_get().
258  *
259  * This decrements the kobject reference count incremented earlier by calling
260  * cpufreq_cpu_get().
261  */
262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
263 {
264         kobject_put(&policy->kobj);
265 }
266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
267
268 /*********************************************************************
269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
270  *********************************************************************/
271
272 /**
273  * adjust_jiffies - adjust the system "loops_per_jiffy"
274  *
275  * This function alters the system "loops_per_jiffy" for the clock
276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
277  * systems as each CPU might be scaled differently. So, use the arch
278  * per-CPU loops_per_jiffy value wherever possible.
279  */
280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
281 {
282 #ifndef CONFIG_SMP
283         static unsigned long l_p_j_ref;
284         static unsigned int l_p_j_ref_freq;
285
286         if (ci->flags & CPUFREQ_CONST_LOOPS)
287                 return;
288
289         if (!l_p_j_ref_freq) {
290                 l_p_j_ref = loops_per_jiffy;
291                 l_p_j_ref_freq = ci->old;
292                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
293                          l_p_j_ref, l_p_j_ref_freq);
294         }
295         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297                                                                 ci->new);
298                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
299                          loops_per_jiffy, ci->new);
300         }
301 #endif
302 }
303
304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
305                 struct cpufreq_freqs *freqs, unsigned int state)
306 {
307         BUG_ON(irqs_disabled());
308
309         if (cpufreq_disabled())
310                 return;
311
312         freqs->flags = cpufreq_driver->flags;
313         pr_debug("notification %u of frequency transition to %u kHz\n",
314                  state, freqs->new);
315
316         switch (state) {
317
318         case CPUFREQ_PRECHANGE:
319                 /* detect if the driver reported a value as "old frequency"
320                  * which is not equal to what the cpufreq core thinks is
321                  * "old frequency".
322                  */
323                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
324                         if ((policy) && (policy->cpu == freqs->cpu) &&
325                             (policy->cur) && (policy->cur != freqs->old)) {
326                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
327                                          freqs->old, policy->cur);
328                                 freqs->old = policy->cur;
329                         }
330                 }
331                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
332                                 CPUFREQ_PRECHANGE, freqs);
333                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
334                 break;
335
336         case CPUFREQ_POSTCHANGE:
337                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
338                 pr_debug("FREQ: %lu - CPU: %lu\n",
339                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
340                 trace_cpu_frequency(freqs->new, freqs->cpu);
341                 cpufreq_stats_record_transition(policy, freqs->new);
342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
343                                 CPUFREQ_POSTCHANGE, freqs);
344                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
345                         policy->cur = freqs->new;
346                 break;
347         }
348 }
349
350 /**
351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
352  * on frequency transition.
353  *
354  * This function calls the transition notifiers and the "adjust_jiffies"
355  * function. It is called twice on all CPU frequency changes that have
356  * external effects.
357  */
358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
359                 struct cpufreq_freqs *freqs, unsigned int state)
360 {
361         for_each_cpu(freqs->cpu, policy->cpus)
362                 __cpufreq_notify_transition(policy, freqs, state);
363 }
364
365 /* Do post notifications when there are chances that transition has failed */
366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
367                 struct cpufreq_freqs *freqs, int transition_failed)
368 {
369         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
370         if (!transition_failed)
371                 return;
372
373         swap(freqs->old, freqs->new);
374         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
376 }
377
378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
379                 struct cpufreq_freqs *freqs)
380 {
381
382         /*
383          * Catch double invocations of _begin() which lead to self-deadlock.
384          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
385          * doesn't invoke _begin() on their behalf, and hence the chances of
386          * double invocations are very low. Moreover, there are scenarios
387          * where these checks can emit false-positive warnings in these
388          * drivers; so we avoid that by skipping them altogether.
389          */
390         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
391                                 && current == policy->transition_task);
392
393 wait:
394         wait_event(policy->transition_wait, !policy->transition_ongoing);
395
396         spin_lock(&policy->transition_lock);
397
398         if (unlikely(policy->transition_ongoing)) {
399                 spin_unlock(&policy->transition_lock);
400                 goto wait;
401         }
402
403         policy->transition_ongoing = true;
404         policy->transition_task = current;
405
406         spin_unlock(&policy->transition_lock);
407
408         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
409 }
410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
411
412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
413                 struct cpufreq_freqs *freqs, int transition_failed)
414 {
415         if (unlikely(WARN_ON(!policy->transition_ongoing)))
416                 return;
417
418         cpufreq_notify_post_transition(policy, freqs, transition_failed);
419
420         policy->transition_ongoing = false;
421         policy->transition_task = NULL;
422
423         wake_up(&policy->transition_wait);
424 }
425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
426
427 /*
428  * Fast frequency switching status count.  Positive means "enabled", negative
429  * means "disabled" and 0 means "not decided yet".
430  */
431 static int cpufreq_fast_switch_count;
432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
433
434 static void cpufreq_list_transition_notifiers(void)
435 {
436         struct notifier_block *nb;
437
438         pr_info("Registered transition notifiers:\n");
439
440         mutex_lock(&cpufreq_transition_notifier_list.mutex);
441
442         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
443                 pr_info("%pF\n", nb->notifier_call);
444
445         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
446 }
447
448 /**
449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
450  * @policy: cpufreq policy to enable fast frequency switching for.
451  *
452  * Try to enable fast frequency switching for @policy.
453  *
454  * The attempt will fail if there is at least one transition notifier registered
455  * at this point, as fast frequency switching is quite fundamentally at odds
456  * with transition notifiers.  Thus if successful, it will make registration of
457  * transition notifiers fail going forward.
458  */
459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
460 {
461         lockdep_assert_held(&policy->rwsem);
462
463         if (!policy->fast_switch_possible)
464                 return;
465
466         mutex_lock(&cpufreq_fast_switch_lock);
467         if (cpufreq_fast_switch_count >= 0) {
468                 cpufreq_fast_switch_count++;
469                 policy->fast_switch_enabled = true;
470         } else {
471                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
472                         policy->cpu);
473                 cpufreq_list_transition_notifiers();
474         }
475         mutex_unlock(&cpufreq_fast_switch_lock);
476 }
477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
478
479 /**
480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
481  * @policy: cpufreq policy to disable fast frequency switching for.
482  */
483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
484 {
485         mutex_lock(&cpufreq_fast_switch_lock);
486         if (policy->fast_switch_enabled) {
487                 policy->fast_switch_enabled = false;
488                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
489                         cpufreq_fast_switch_count--;
490         }
491         mutex_unlock(&cpufreq_fast_switch_lock);
492 }
493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
494
495 /**
496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
497  * one.
498  * @target_freq: target frequency to resolve.
499  *
500  * The target to driver frequency mapping is cached in the policy.
501  *
502  * Return: Lowest driver-supported frequency greater than or equal to the
503  * given target_freq, subject to policy (min/max) and driver limitations.
504  */
505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
506                                          unsigned int target_freq)
507 {
508         target_freq = clamp_val(target_freq, policy->min, policy->max);
509         policy->cached_target_freq = target_freq;
510
511         if (cpufreq_driver->target_index) {
512                 int idx;
513
514                 idx = cpufreq_frequency_table_target(policy, target_freq,
515                                                      CPUFREQ_RELATION_L);
516                 policy->cached_resolved_idx = idx;
517                 return policy->freq_table[idx].frequency;
518         }
519
520         if (cpufreq_driver->resolve_freq)
521                 return cpufreq_driver->resolve_freq(policy, target_freq);
522
523         return target_freq;
524 }
525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
526
527 /*********************************************************************
528  *                          SYSFS INTERFACE                          *
529  *********************************************************************/
530 static ssize_t show_boost(struct kobject *kobj,
531                                  struct attribute *attr, char *buf)
532 {
533         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
534 }
535
536 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
537                                   const char *buf, size_t count)
538 {
539         int ret, enable;
540
541         ret = sscanf(buf, "%d", &enable);
542         if (ret != 1 || enable < 0 || enable > 1)
543                 return -EINVAL;
544
545         if (cpufreq_boost_trigger_state(enable)) {
546                 pr_err("%s: Cannot %s BOOST!\n",
547                        __func__, enable ? "enable" : "disable");
548                 return -EINVAL;
549         }
550
551         pr_debug("%s: cpufreq BOOST %s\n",
552                  __func__, enable ? "enabled" : "disabled");
553
554         return count;
555 }
556 define_one_global_rw(boost);
557
558 static struct cpufreq_governor *find_governor(const char *str_governor)
559 {
560         struct cpufreq_governor *t;
561
562         for_each_governor(t)
563                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
564                         return t;
565
566         return NULL;
567 }
568
569 /**
570  * cpufreq_parse_governor - parse a governor string
571  */
572 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
573                                 struct cpufreq_governor **governor)
574 {
575         int err = -EINVAL;
576
577         if (cpufreq_driver->setpolicy) {
578                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
579                         *policy = CPUFREQ_POLICY_PERFORMANCE;
580                         err = 0;
581                 } else if (!strncasecmp(str_governor, "powersave",
582                                                 CPUFREQ_NAME_LEN)) {
583                         *policy = CPUFREQ_POLICY_POWERSAVE;
584                         err = 0;
585                 }
586         } else {
587                 struct cpufreq_governor *t;
588
589                 mutex_lock(&cpufreq_governor_mutex);
590
591                 t = find_governor(str_governor);
592
593                 if (t == NULL) {
594                         int ret;
595
596                         mutex_unlock(&cpufreq_governor_mutex);
597                         ret = request_module("cpufreq_%s", str_governor);
598                         mutex_lock(&cpufreq_governor_mutex);
599
600                         if (ret == 0)
601                                 t = find_governor(str_governor);
602                 }
603
604                 if (t != NULL) {
605                         *governor = t;
606                         err = 0;
607                 }
608
609                 mutex_unlock(&cpufreq_governor_mutex);
610         }
611         return err;
612 }
613
614 /**
615  * cpufreq_per_cpu_attr_read() / show_##file_name() -
616  * print out cpufreq information
617  *
618  * Write out information from cpufreq_driver->policy[cpu]; object must be
619  * "unsigned int".
620  */
621
622 #define show_one(file_name, object)                     \
623 static ssize_t show_##file_name                         \
624 (struct cpufreq_policy *policy, char *buf)              \
625 {                                                       \
626         return sprintf(buf, "%u\n", policy->object);    \
627 }
628
629 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
630 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
631 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
632 show_one(scaling_min_freq, min);
633 show_one(scaling_max_freq, max);
634
635 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
636 {
637         ssize_t ret;
638
639         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
640                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
641         else
642                 ret = sprintf(buf, "%u\n", policy->cur);
643         return ret;
644 }
645
646 static int cpufreq_set_policy(struct cpufreq_policy *policy,
647                                 struct cpufreq_policy *new_policy);
648
649 /**
650  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
651  */
652 #define store_one(file_name, object)                    \
653 static ssize_t store_##file_name                                        \
654 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
655 {                                                                       \
656         int ret, temp;                                                  \
657         struct cpufreq_policy new_policy;                               \
658                                                                         \
659         memcpy(&new_policy, policy, sizeof(*policy));                   \
660                                                                         \
661         ret = sscanf(buf, "%u", &new_policy.object);                    \
662         if (ret != 1)                                                   \
663                 return -EINVAL;                                         \
664                                                                         \
665         temp = new_policy.object;                                       \
666         ret = cpufreq_set_policy(policy, &new_policy);          \
667         if (!ret)                                                       \
668                 policy->user_policy.object = temp;                      \
669                                                                         \
670         return ret ? ret : count;                                       \
671 }
672
673 store_one(scaling_min_freq, min);
674 store_one(scaling_max_freq, max);
675
676 /**
677  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
678  */
679 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
680                                         char *buf)
681 {
682         unsigned int cur_freq = __cpufreq_get(policy);
683         if (!cur_freq)
684                 return sprintf(buf, "<unknown>");
685         return sprintf(buf, "%u\n", cur_freq);
686 }
687
688 /**
689  * show_scaling_governor - show the current policy for the specified CPU
690  */
691 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
692 {
693         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
694                 return sprintf(buf, "powersave\n");
695         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
696                 return sprintf(buf, "performance\n");
697         else if (policy->governor)
698                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
699                                 policy->governor->name);
700         return -EINVAL;
701 }
702
703 /**
704  * store_scaling_governor - store policy for the specified CPU
705  */
706 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
707                                         const char *buf, size_t count)
708 {
709         int ret;
710         char    str_governor[16];
711         struct cpufreq_policy new_policy;
712
713         memcpy(&new_policy, policy, sizeof(*policy));
714
715         ret = sscanf(buf, "%15s", str_governor);
716         if (ret != 1)
717                 return -EINVAL;
718
719         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
720                                                 &new_policy.governor))
721                 return -EINVAL;
722
723         ret = cpufreq_set_policy(policy, &new_policy);
724         return ret ? ret : count;
725 }
726
727 /**
728  * show_scaling_driver - show the cpufreq driver currently loaded
729  */
730 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
731 {
732         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
733 }
734
735 /**
736  * show_scaling_available_governors - show the available CPUfreq governors
737  */
738 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
739                                                 char *buf)
740 {
741         ssize_t i = 0;
742         struct cpufreq_governor *t;
743
744         if (!has_target()) {
745                 i += sprintf(buf, "performance powersave");
746                 goto out;
747         }
748
749         for_each_governor(t) {
750                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
751                     - (CPUFREQ_NAME_LEN + 2)))
752                         goto out;
753                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
754         }
755 out:
756         i += sprintf(&buf[i], "\n");
757         return i;
758 }
759
760 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
761 {
762         ssize_t i = 0;
763         unsigned int cpu;
764
765         for_each_cpu(cpu, mask) {
766                 if (i)
767                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
768                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
769                 if (i >= (PAGE_SIZE - 5))
770                         break;
771         }
772         i += sprintf(&buf[i], "\n");
773         return i;
774 }
775 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
776
777 /**
778  * show_related_cpus - show the CPUs affected by each transition even if
779  * hw coordination is in use
780  */
781 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
782 {
783         return cpufreq_show_cpus(policy->related_cpus, buf);
784 }
785
786 /**
787  * show_affected_cpus - show the CPUs affected by each transition
788  */
789 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
790 {
791         return cpufreq_show_cpus(policy->cpus, buf);
792 }
793
794 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
795                                         const char *buf, size_t count)
796 {
797         unsigned int freq = 0;
798         unsigned int ret;
799
800         if (!policy->governor || !policy->governor->store_setspeed)
801                 return -EINVAL;
802
803         ret = sscanf(buf, "%u", &freq);
804         if (ret != 1)
805                 return -EINVAL;
806
807         policy->governor->store_setspeed(policy, freq);
808
809         return count;
810 }
811
812 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
813 {
814         if (!policy->governor || !policy->governor->show_setspeed)
815                 return sprintf(buf, "<unsupported>\n");
816
817         return policy->governor->show_setspeed(policy, buf);
818 }
819
820 /**
821  * show_bios_limit - show the current cpufreq HW/BIOS limitation
822  */
823 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
824 {
825         unsigned int limit;
826         int ret;
827         if (cpufreq_driver->bios_limit) {
828                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
829                 if (!ret)
830                         return sprintf(buf, "%u\n", limit);
831         }
832         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
833 }
834
835 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
836 cpufreq_freq_attr_ro(cpuinfo_min_freq);
837 cpufreq_freq_attr_ro(cpuinfo_max_freq);
838 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
839 cpufreq_freq_attr_ro(scaling_available_governors);
840 cpufreq_freq_attr_ro(scaling_driver);
841 cpufreq_freq_attr_ro(scaling_cur_freq);
842 cpufreq_freq_attr_ro(bios_limit);
843 cpufreq_freq_attr_ro(related_cpus);
844 cpufreq_freq_attr_ro(affected_cpus);
845 cpufreq_freq_attr_rw(scaling_min_freq);
846 cpufreq_freq_attr_rw(scaling_max_freq);
847 cpufreq_freq_attr_rw(scaling_governor);
848 cpufreq_freq_attr_rw(scaling_setspeed);
849
850 static struct attribute *default_attrs[] = {
851         &cpuinfo_min_freq.attr,
852         &cpuinfo_max_freq.attr,
853         &cpuinfo_transition_latency.attr,
854         &scaling_min_freq.attr,
855         &scaling_max_freq.attr,
856         &affected_cpus.attr,
857         &related_cpus.attr,
858         &scaling_governor.attr,
859         &scaling_driver.attr,
860         &scaling_available_governors.attr,
861         &scaling_setspeed.attr,
862         NULL
863 };
864
865 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
866 #define to_attr(a) container_of(a, struct freq_attr, attr)
867
868 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
869 {
870         struct cpufreq_policy *policy = to_policy(kobj);
871         struct freq_attr *fattr = to_attr(attr);
872         ssize_t ret;
873
874         down_read(&policy->rwsem);
875         ret = fattr->show(policy, buf);
876         up_read(&policy->rwsem);
877
878         return ret;
879 }
880
881 static ssize_t store(struct kobject *kobj, struct attribute *attr,
882                      const char *buf, size_t count)
883 {
884         struct cpufreq_policy *policy = to_policy(kobj);
885         struct freq_attr *fattr = to_attr(attr);
886         ssize_t ret = -EINVAL;
887
888         get_online_cpus();
889
890         if (cpu_online(policy->cpu)) {
891                 down_write(&policy->rwsem);
892                 ret = fattr->store(policy, buf, count);
893                 up_write(&policy->rwsem);
894         }
895
896         put_online_cpus();
897
898         return ret;
899 }
900
901 static void cpufreq_sysfs_release(struct kobject *kobj)
902 {
903         struct cpufreq_policy *policy = to_policy(kobj);
904         pr_debug("last reference is dropped\n");
905         complete(&policy->kobj_unregister);
906 }
907
908 static const struct sysfs_ops sysfs_ops = {
909         .show   = show,
910         .store  = store,
911 };
912
913 static struct kobj_type ktype_cpufreq = {
914         .sysfs_ops      = &sysfs_ops,
915         .default_attrs  = default_attrs,
916         .release        = cpufreq_sysfs_release,
917 };
918
919 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
920 {
921         struct device *cpu_dev;
922
923         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
924
925         if (!policy)
926                 return 0;
927
928         cpu_dev = get_cpu_device(cpu);
929         if (WARN_ON(!cpu_dev))
930                 return 0;
931
932         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
933 }
934
935 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
936 {
937         struct device *cpu_dev;
938
939         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
940
941         cpu_dev = get_cpu_device(cpu);
942         if (WARN_ON(!cpu_dev))
943                 return;
944
945         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
946 }
947
948 /* Add/remove symlinks for all related CPUs */
949 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
950 {
951         unsigned int j;
952         int ret = 0;
953
954         /* Some related CPUs might not be present (physically hotplugged) */
955         for_each_cpu(j, policy->real_cpus) {
956                 ret = add_cpu_dev_symlink(policy, j);
957                 if (ret)
958                         break;
959         }
960
961         return ret;
962 }
963
964 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
965 {
966         unsigned int j;
967
968         /* Some related CPUs might not be present (physically hotplugged) */
969         for_each_cpu(j, policy->real_cpus)
970                 remove_cpu_dev_symlink(policy, j);
971 }
972
973 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
974 {
975         struct freq_attr **drv_attr;
976         int ret = 0;
977
978         /* set up files for this cpu device */
979         drv_attr = cpufreq_driver->attr;
980         while (drv_attr && *drv_attr) {
981                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
982                 if (ret)
983                         return ret;
984                 drv_attr++;
985         }
986         if (cpufreq_driver->get) {
987                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
988                 if (ret)
989                         return ret;
990         }
991
992         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
993         if (ret)
994                 return ret;
995
996         if (cpufreq_driver->bios_limit) {
997                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
998                 if (ret)
999                         return ret;
1000         }
1001
1002         return cpufreq_add_dev_symlink(policy);
1003 }
1004
1005 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1006 {
1007         return NULL;
1008 }
1009
1010 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1011 {
1012         struct cpufreq_governor *gov = NULL;
1013         struct cpufreq_policy new_policy;
1014
1015         memcpy(&new_policy, policy, sizeof(*policy));
1016
1017         /* Update governor of new_policy to the governor used before hotplug */
1018         gov = find_governor(policy->last_governor);
1019         if (gov) {
1020                 pr_debug("Restoring governor %s for cpu %d\n",
1021                                 policy->governor->name, policy->cpu);
1022         } else {
1023                 gov = cpufreq_default_governor();
1024                 if (!gov)
1025                         return -ENODATA;
1026         }
1027
1028         new_policy.governor = gov;
1029
1030         /* Use the default policy if there is no last_policy. */
1031         if (cpufreq_driver->setpolicy) {
1032                 if (policy->last_policy)
1033                         new_policy.policy = policy->last_policy;
1034                 else
1035                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1036                                                NULL);
1037         }
1038         /* set default policy */
1039         return cpufreq_set_policy(policy, &new_policy);
1040 }
1041
1042 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1043 {
1044         int ret = 0;
1045
1046         /* Has this CPU been taken care of already? */
1047         if (cpumask_test_cpu(cpu, policy->cpus))
1048                 return 0;
1049
1050         down_write(&policy->rwsem);
1051         if (has_target())
1052                 cpufreq_stop_governor(policy);
1053
1054         cpumask_set_cpu(cpu, policy->cpus);
1055
1056         if (has_target()) {
1057                 ret = cpufreq_start_governor(policy);
1058                 if (ret)
1059                         pr_err("%s: Failed to start governor\n", __func__);
1060         }
1061         up_write(&policy->rwsem);
1062         return ret;
1063 }
1064
1065 static void handle_update(struct work_struct *work)
1066 {
1067         struct cpufreq_policy *policy =
1068                 container_of(work, struct cpufreq_policy, update);
1069         unsigned int cpu = policy->cpu;
1070         pr_debug("handle_update for cpu %u called\n", cpu);
1071         cpufreq_update_policy(cpu);
1072 }
1073
1074 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1075 {
1076         struct device *dev = get_cpu_device(cpu);
1077         struct cpufreq_policy *policy;
1078         int ret;
1079
1080         if (WARN_ON(!dev))
1081                 return NULL;
1082
1083         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1084         if (!policy)
1085                 return NULL;
1086
1087         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1088                 goto err_free_policy;
1089
1090         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1091                 goto err_free_cpumask;
1092
1093         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1094                 goto err_free_rcpumask;
1095
1096         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1097                                    cpufreq_global_kobject, "policy%u", cpu);
1098         if (ret) {
1099                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1100                 goto err_free_real_cpus;
1101         }
1102
1103         INIT_LIST_HEAD(&policy->policy_list);
1104         init_rwsem(&policy->rwsem);
1105         spin_lock_init(&policy->transition_lock);
1106         init_waitqueue_head(&policy->transition_wait);
1107         init_completion(&policy->kobj_unregister);
1108         INIT_WORK(&policy->update, handle_update);
1109
1110         policy->cpu = cpu;
1111         return policy;
1112
1113 err_free_real_cpus:
1114         free_cpumask_var(policy->real_cpus);
1115 err_free_rcpumask:
1116         free_cpumask_var(policy->related_cpus);
1117 err_free_cpumask:
1118         free_cpumask_var(policy->cpus);
1119 err_free_policy:
1120         kfree(policy);
1121
1122         return NULL;
1123 }
1124
1125 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1126 {
1127         struct kobject *kobj;
1128         struct completion *cmp;
1129
1130         if (notify)
1131                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1132                                              CPUFREQ_REMOVE_POLICY, policy);
1133
1134         down_write(&policy->rwsem);
1135         cpufreq_stats_free_table(policy);
1136         cpufreq_remove_dev_symlink(policy);
1137         kobj = &policy->kobj;
1138         cmp = &policy->kobj_unregister;
1139         up_write(&policy->rwsem);
1140         kobject_put(kobj);
1141
1142         /*
1143          * We need to make sure that the underlying kobj is
1144          * actually not referenced anymore by anybody before we
1145          * proceed with unloading.
1146          */
1147         pr_debug("waiting for dropping of refcount\n");
1148         wait_for_completion(cmp);
1149         pr_debug("wait complete\n");
1150 }
1151
1152 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1153 {
1154         unsigned long flags;
1155         int cpu;
1156
1157         /* Remove policy from list */
1158         write_lock_irqsave(&cpufreq_driver_lock, flags);
1159         list_del(&policy->policy_list);
1160
1161         for_each_cpu(cpu, policy->related_cpus)
1162                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1163         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1164
1165         cpufreq_policy_put_kobj(policy, notify);
1166         free_cpumask_var(policy->real_cpus);
1167         free_cpumask_var(policy->related_cpus);
1168         free_cpumask_var(policy->cpus);
1169         kfree(policy);
1170 }
1171
1172 static int cpufreq_online(unsigned int cpu)
1173 {
1174         struct cpufreq_policy *policy;
1175         bool new_policy;
1176         unsigned long flags;
1177         unsigned int j;
1178         int ret;
1179
1180         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1181
1182         /* Check if this CPU already has a policy to manage it */
1183         policy = per_cpu(cpufreq_cpu_data, cpu);
1184         if (policy) {
1185                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1186                 if (!policy_is_inactive(policy))
1187                         return cpufreq_add_policy_cpu(policy, cpu);
1188
1189                 /* This is the only online CPU for the policy.  Start over. */
1190                 new_policy = false;
1191                 down_write(&policy->rwsem);
1192                 policy->cpu = cpu;
1193                 policy->governor = NULL;
1194                 up_write(&policy->rwsem);
1195         } else {
1196                 new_policy = true;
1197                 policy = cpufreq_policy_alloc(cpu);
1198                 if (!policy)
1199                         return -ENOMEM;
1200         }
1201
1202         cpumask_copy(policy->cpus, cpumask_of(cpu));
1203
1204         /* call driver. From then on the cpufreq must be able
1205          * to accept all calls to ->verify and ->setpolicy for this CPU
1206          */
1207         ret = cpufreq_driver->init(policy);
1208         if (ret) {
1209                 pr_debug("initialization failed\n");
1210                 goto out_free_policy;
1211         }
1212
1213         down_write(&policy->rwsem);
1214
1215         if (new_policy) {
1216                 /* related_cpus should at least include policy->cpus. */
1217                 cpumask_copy(policy->related_cpus, policy->cpus);
1218                 /* Remember CPUs present at the policy creation time. */
1219                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1220         }
1221
1222         /*
1223          * affected cpus must always be the one, which are online. We aren't
1224          * managing offline cpus here.
1225          */
1226         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1227
1228         if (new_policy) {
1229                 policy->user_policy.min = policy->min;
1230                 policy->user_policy.max = policy->max;
1231
1232                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1233                 for_each_cpu(j, policy->related_cpus)
1234                         per_cpu(cpufreq_cpu_data, j) = policy;
1235                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1236         }
1237
1238         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1239                 policy->cur = cpufreq_driver->get(policy->cpu);
1240                 if (!policy->cur) {
1241                         pr_err("%s: ->get() failed\n", __func__);
1242                         goto out_exit_policy;
1243                 }
1244         }
1245
1246         /*
1247          * Sometimes boot loaders set CPU frequency to a value outside of
1248          * frequency table present with cpufreq core. In such cases CPU might be
1249          * unstable if it has to run on that frequency for long duration of time
1250          * and so its better to set it to a frequency which is specified in
1251          * freq-table. This also makes cpufreq stats inconsistent as
1252          * cpufreq-stats would fail to register because current frequency of CPU
1253          * isn't found in freq-table.
1254          *
1255          * Because we don't want this change to effect boot process badly, we go
1256          * for the next freq which is >= policy->cur ('cur' must be set by now,
1257          * otherwise we will end up setting freq to lowest of the table as 'cur'
1258          * is initialized to zero).
1259          *
1260          * We are passing target-freq as "policy->cur - 1" otherwise
1261          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1262          * equal to target-freq.
1263          */
1264         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1265             && has_target()) {
1266                 /* Are we running at unknown frequency ? */
1267                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1268                 if (ret == -EINVAL) {
1269                         /* Warn user and fix it */
1270                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1271                                 __func__, policy->cpu, policy->cur);
1272                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1273                                 CPUFREQ_RELATION_L);
1274
1275                         /*
1276                          * Reaching here after boot in a few seconds may not
1277                          * mean that system will remain stable at "unknown"
1278                          * frequency for longer duration. Hence, a BUG_ON().
1279                          */
1280                         BUG_ON(ret);
1281                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1282                                 __func__, policy->cpu, policy->cur);
1283                 }
1284         }
1285
1286         if (new_policy) {
1287                 ret = cpufreq_add_dev_interface(policy);
1288                 if (ret)
1289                         goto out_exit_policy;
1290
1291                 cpufreq_stats_create_table(policy);
1292                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1293                                 CPUFREQ_CREATE_POLICY, policy);
1294
1295                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1296                 list_add(&policy->policy_list, &cpufreq_policy_list);
1297                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1298         }
1299
1300         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1301                                      CPUFREQ_START, policy);
1302
1303         ret = cpufreq_init_policy(policy);
1304         if (ret) {
1305                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1306                        __func__, cpu, ret);
1307                 /* cpufreq_policy_free() will notify based on this */
1308                 new_policy = false;
1309                 goto out_exit_policy;
1310         }
1311
1312         up_write(&policy->rwsem);
1313
1314         kobject_uevent(&policy->kobj, KOBJ_ADD);
1315
1316         /* Callback for handling stuff after policy is ready */
1317         if (cpufreq_driver->ready)
1318                 cpufreq_driver->ready(policy);
1319
1320         pr_debug("initialization complete\n");
1321
1322         return 0;
1323
1324 out_exit_policy:
1325         up_write(&policy->rwsem);
1326
1327         if (cpufreq_driver->exit)
1328                 cpufreq_driver->exit(policy);
1329 out_free_policy:
1330         cpufreq_policy_free(policy, !new_policy);
1331         return ret;
1332 }
1333
1334 /**
1335  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1336  * @dev: CPU device.
1337  * @sif: Subsystem interface structure pointer (not used)
1338  */
1339 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1340 {
1341         struct cpufreq_policy *policy;
1342         unsigned cpu = dev->id;
1343
1344         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1345
1346         if (cpu_online(cpu))
1347                 return cpufreq_online(cpu);
1348
1349         /*
1350          * A hotplug notifier will follow and we will handle it as CPU online
1351          * then.  For now, just create the sysfs link, unless there is no policy
1352          * or the link is already present.
1353          */
1354         policy = per_cpu(cpufreq_cpu_data, cpu);
1355         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1356                 return 0;
1357
1358         return add_cpu_dev_symlink(policy, cpu);
1359 }
1360
1361 static void cpufreq_offline(unsigned int cpu)
1362 {
1363         struct cpufreq_policy *policy;
1364         int ret;
1365
1366         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1367
1368         policy = cpufreq_cpu_get_raw(cpu);
1369         if (!policy) {
1370                 pr_debug("%s: No cpu_data found\n", __func__);
1371                 return;
1372         }
1373
1374         down_write(&policy->rwsem);
1375         if (has_target())
1376                 cpufreq_stop_governor(policy);
1377
1378         cpumask_clear_cpu(cpu, policy->cpus);
1379
1380         if (policy_is_inactive(policy)) {
1381                 if (has_target())
1382                         strncpy(policy->last_governor, policy->governor->name,
1383                                 CPUFREQ_NAME_LEN);
1384                 else
1385                         policy->last_policy = policy->policy;
1386         } else if (cpu == policy->cpu) {
1387                 /* Nominate new CPU */
1388                 policy->cpu = cpumask_any(policy->cpus);
1389         }
1390
1391         /* Start governor again for active policy */
1392         if (!policy_is_inactive(policy)) {
1393                 if (has_target()) {
1394                         ret = cpufreq_start_governor(policy);
1395                         if (ret)
1396                                 pr_err("%s: Failed to start governor\n", __func__);
1397                 }
1398
1399                 goto unlock;
1400         }
1401
1402         if (cpufreq_driver->stop_cpu)
1403                 cpufreq_driver->stop_cpu(policy);
1404
1405         if (has_target())
1406                 cpufreq_exit_governor(policy);
1407
1408         /*
1409          * Perform the ->exit() even during light-weight tear-down,
1410          * since this is a core component, and is essential for the
1411          * subsequent light-weight ->init() to succeed.
1412          */
1413         if (cpufreq_driver->exit) {
1414                 cpufreq_driver->exit(policy);
1415                 policy->freq_table = NULL;
1416         }
1417
1418 unlock:
1419         up_write(&policy->rwsem);
1420 }
1421
1422 /**
1423  * cpufreq_remove_dev - remove a CPU device
1424  *
1425  * Removes the cpufreq interface for a CPU device.
1426  */
1427 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1428 {
1429         unsigned int cpu = dev->id;
1430         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1431
1432         if (!policy)
1433                 return;
1434
1435         if (cpu_online(cpu))
1436                 cpufreq_offline(cpu);
1437
1438         cpumask_clear_cpu(cpu, policy->real_cpus);
1439         remove_cpu_dev_symlink(policy, cpu);
1440
1441         if (cpumask_empty(policy->real_cpus))
1442                 cpufreq_policy_free(policy, true);
1443 }
1444
1445 /**
1446  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1447  *      in deep trouble.
1448  *      @policy: policy managing CPUs
1449  *      @new_freq: CPU frequency the CPU actually runs at
1450  *
1451  *      We adjust to current frequency first, and need to clean up later.
1452  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1453  */
1454 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1455                                 unsigned int new_freq)
1456 {
1457         struct cpufreq_freqs freqs;
1458
1459         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1460                  policy->cur, new_freq);
1461
1462         freqs.old = policy->cur;
1463         freqs.new = new_freq;
1464
1465         cpufreq_freq_transition_begin(policy, &freqs);
1466         cpufreq_freq_transition_end(policy, &freqs, 0);
1467 }
1468
1469 /**
1470  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1471  * @cpu: CPU number
1472  *
1473  * This is the last known freq, without actually getting it from the driver.
1474  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1475  */
1476 unsigned int cpufreq_quick_get(unsigned int cpu)
1477 {
1478         struct cpufreq_policy *policy;
1479         unsigned int ret_freq = 0;
1480         unsigned long flags;
1481
1482         read_lock_irqsave(&cpufreq_driver_lock, flags);
1483
1484         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1485                 ret_freq = cpufreq_driver->get(cpu);
1486                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1487                 return ret_freq;
1488         }
1489
1490         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1491
1492         policy = cpufreq_cpu_get(cpu);
1493         if (policy) {
1494                 ret_freq = policy->cur;
1495                 cpufreq_cpu_put(policy);
1496         }
1497
1498         return ret_freq;
1499 }
1500 EXPORT_SYMBOL(cpufreq_quick_get);
1501
1502 /**
1503  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1504  * @cpu: CPU number
1505  *
1506  * Just return the max possible frequency for a given CPU.
1507  */
1508 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1509 {
1510         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1511         unsigned int ret_freq = 0;
1512
1513         if (policy) {
1514                 ret_freq = policy->max;
1515                 cpufreq_cpu_put(policy);
1516         }
1517
1518         return ret_freq;
1519 }
1520 EXPORT_SYMBOL(cpufreq_quick_get_max);
1521
1522 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1523 {
1524         unsigned int ret_freq = 0;
1525
1526         if (!cpufreq_driver->get)
1527                 return ret_freq;
1528
1529         ret_freq = cpufreq_driver->get(policy->cpu);
1530
1531         /*
1532          * Updating inactive policies is invalid, so avoid doing that.  Also
1533          * if fast frequency switching is used with the given policy, the check
1534          * against policy->cur is pointless, so skip it in that case too.
1535          */
1536         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1537                 return ret_freq;
1538
1539         if (ret_freq && policy->cur &&
1540                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1541                 /* verify no discrepancy between actual and
1542                                         saved value exists */
1543                 if (unlikely(ret_freq != policy->cur)) {
1544                         cpufreq_out_of_sync(policy, ret_freq);
1545                         schedule_work(&policy->update);
1546                 }
1547         }
1548
1549         return ret_freq;
1550 }
1551
1552 /**
1553  * cpufreq_get - get the current CPU frequency (in kHz)
1554  * @cpu: CPU number
1555  *
1556  * Get the CPU current (static) CPU frequency
1557  */
1558 unsigned int cpufreq_get(unsigned int cpu)
1559 {
1560         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1561         unsigned int ret_freq = 0;
1562
1563         if (policy) {
1564                 down_read(&policy->rwsem);
1565                 ret_freq = __cpufreq_get(policy);
1566                 up_read(&policy->rwsem);
1567
1568                 cpufreq_cpu_put(policy);
1569         }
1570
1571         return ret_freq;
1572 }
1573 EXPORT_SYMBOL(cpufreq_get);
1574
1575 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1576 {
1577         unsigned int new_freq;
1578
1579         new_freq = cpufreq_driver->get(policy->cpu);
1580         if (!new_freq)
1581                 return 0;
1582
1583         if (!policy->cur) {
1584                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1585                 policy->cur = new_freq;
1586         } else if (policy->cur != new_freq && has_target()) {
1587                 cpufreq_out_of_sync(policy, new_freq);
1588         }
1589
1590         return new_freq;
1591 }
1592
1593 static struct subsys_interface cpufreq_interface = {
1594         .name           = "cpufreq",
1595         .subsys         = &cpu_subsys,
1596         .add_dev        = cpufreq_add_dev,
1597         .remove_dev     = cpufreq_remove_dev,
1598 };
1599
1600 /*
1601  * In case platform wants some specific frequency to be configured
1602  * during suspend..
1603  */
1604 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1605 {
1606         int ret;
1607
1608         if (!policy->suspend_freq) {
1609                 pr_debug("%s: suspend_freq not defined\n", __func__);
1610                 return 0;
1611         }
1612
1613         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1614                         policy->suspend_freq);
1615
1616         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1617                         CPUFREQ_RELATION_H);
1618         if (ret)
1619                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1620                                 __func__, policy->suspend_freq, ret);
1621
1622         return ret;
1623 }
1624 EXPORT_SYMBOL(cpufreq_generic_suspend);
1625
1626 /**
1627  * cpufreq_suspend() - Suspend CPUFreq governors
1628  *
1629  * Called during system wide Suspend/Hibernate cycles for suspending governors
1630  * as some platforms can't change frequency after this point in suspend cycle.
1631  * Because some of the devices (like: i2c, regulators, etc) they use for
1632  * changing frequency are suspended quickly after this point.
1633  */
1634 void cpufreq_suspend(void)
1635 {
1636         struct cpufreq_policy *policy;
1637
1638         if (!cpufreq_driver)
1639                 return;
1640
1641         if (!has_target() && !cpufreq_driver->suspend)
1642                 goto suspend;
1643
1644         pr_debug("%s: Suspending Governors\n", __func__);
1645
1646         for_each_active_policy(policy) {
1647                 if (has_target()) {
1648                         down_write(&policy->rwsem);
1649                         cpufreq_stop_governor(policy);
1650                         up_write(&policy->rwsem);
1651                 }
1652
1653                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1654                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1655                                 policy);
1656         }
1657
1658 suspend:
1659         cpufreq_suspended = true;
1660 }
1661
1662 /**
1663  * cpufreq_resume() - Resume CPUFreq governors
1664  *
1665  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1666  * are suspended with cpufreq_suspend().
1667  */
1668 void cpufreq_resume(void)
1669 {
1670         struct cpufreq_policy *policy;
1671         int ret;
1672
1673         if (!cpufreq_driver)
1674                 return;
1675
1676         cpufreq_suspended = false;
1677
1678         if (!has_target() && !cpufreq_driver->resume)
1679                 return;
1680
1681         pr_debug("%s: Resuming Governors\n", __func__);
1682
1683         for_each_active_policy(policy) {
1684                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1685                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1686                                 policy);
1687                 } else if (has_target()) {
1688                         down_write(&policy->rwsem);
1689                         ret = cpufreq_start_governor(policy);
1690                         up_write(&policy->rwsem);
1691
1692                         if (ret)
1693                                 pr_err("%s: Failed to start governor for policy: %p\n",
1694                                        __func__, policy);
1695                 }
1696         }
1697 }
1698
1699 /**
1700  *      cpufreq_get_current_driver - return current driver's name
1701  *
1702  *      Return the name string of the currently loaded cpufreq driver
1703  *      or NULL, if none.
1704  */
1705 const char *cpufreq_get_current_driver(void)
1706 {
1707         if (cpufreq_driver)
1708                 return cpufreq_driver->name;
1709
1710         return NULL;
1711 }
1712 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1713
1714 /**
1715  *      cpufreq_get_driver_data - return current driver data
1716  *
1717  *      Return the private data of the currently loaded cpufreq
1718  *      driver, or NULL if no cpufreq driver is loaded.
1719  */
1720 void *cpufreq_get_driver_data(void)
1721 {
1722         if (cpufreq_driver)
1723                 return cpufreq_driver->driver_data;
1724
1725         return NULL;
1726 }
1727 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1728
1729 /*********************************************************************
1730  *                     NOTIFIER LISTS INTERFACE                      *
1731  *********************************************************************/
1732
1733 /**
1734  *      cpufreq_register_notifier - register a driver with cpufreq
1735  *      @nb: notifier function to register
1736  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1737  *
1738  *      Add a driver to one of two lists: either a list of drivers that
1739  *      are notified about clock rate changes (once before and once after
1740  *      the transition), or a list of drivers that are notified about
1741  *      changes in cpufreq policy.
1742  *
1743  *      This function may sleep, and has the same return conditions as
1744  *      blocking_notifier_chain_register.
1745  */
1746 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1747 {
1748         int ret;
1749
1750         if (cpufreq_disabled())
1751                 return -EINVAL;
1752
1753         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1754
1755         switch (list) {
1756         case CPUFREQ_TRANSITION_NOTIFIER:
1757                 mutex_lock(&cpufreq_fast_switch_lock);
1758
1759                 if (cpufreq_fast_switch_count > 0) {
1760                         mutex_unlock(&cpufreq_fast_switch_lock);
1761                         return -EBUSY;
1762                 }
1763                 ret = srcu_notifier_chain_register(
1764                                 &cpufreq_transition_notifier_list, nb);
1765                 if (!ret)
1766                         cpufreq_fast_switch_count--;
1767
1768                 mutex_unlock(&cpufreq_fast_switch_lock);
1769                 break;
1770         case CPUFREQ_POLICY_NOTIFIER:
1771                 ret = blocking_notifier_chain_register(
1772                                 &cpufreq_policy_notifier_list, nb);
1773                 break;
1774         default:
1775                 ret = -EINVAL;
1776         }
1777
1778         return ret;
1779 }
1780 EXPORT_SYMBOL(cpufreq_register_notifier);
1781
1782 /**
1783  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1784  *      @nb: notifier block to be unregistered
1785  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1786  *
1787  *      Remove a driver from the CPU frequency notifier list.
1788  *
1789  *      This function may sleep, and has the same return conditions as
1790  *      blocking_notifier_chain_unregister.
1791  */
1792 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1793 {
1794         int ret;
1795
1796         if (cpufreq_disabled())
1797                 return -EINVAL;
1798
1799         switch (list) {
1800         case CPUFREQ_TRANSITION_NOTIFIER:
1801                 mutex_lock(&cpufreq_fast_switch_lock);
1802
1803                 ret = srcu_notifier_chain_unregister(
1804                                 &cpufreq_transition_notifier_list, nb);
1805                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1806                         cpufreq_fast_switch_count++;
1807
1808                 mutex_unlock(&cpufreq_fast_switch_lock);
1809                 break;
1810         case CPUFREQ_POLICY_NOTIFIER:
1811                 ret = blocking_notifier_chain_unregister(
1812                                 &cpufreq_policy_notifier_list, nb);
1813                 break;
1814         default:
1815                 ret = -EINVAL;
1816         }
1817
1818         return ret;
1819 }
1820 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1821
1822
1823 /*********************************************************************
1824  *                              GOVERNORS                            *
1825  *********************************************************************/
1826
1827 /**
1828  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1829  * @policy: cpufreq policy to switch the frequency for.
1830  * @target_freq: New frequency to set (may be approximate).
1831  *
1832  * Carry out a fast frequency switch without sleeping.
1833  *
1834  * The driver's ->fast_switch() callback invoked by this function must be
1835  * suitable for being called from within RCU-sched read-side critical sections
1836  * and it is expected to select the minimum available frequency greater than or
1837  * equal to @target_freq (CPUFREQ_RELATION_L).
1838  *
1839  * This function must not be called if policy->fast_switch_enabled is unset.
1840  *
1841  * Governors calling this function must guarantee that it will never be invoked
1842  * twice in parallel for the same policy and that it will never be called in
1843  * parallel with either ->target() or ->target_index() for the same policy.
1844  *
1845  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1846  * callback to indicate an error condition, the hardware configuration must be
1847  * preserved.
1848  */
1849 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1850                                         unsigned int target_freq)
1851 {
1852         target_freq = clamp_val(target_freq, policy->min, policy->max);
1853
1854         return cpufreq_driver->fast_switch(policy, target_freq);
1855 }
1856 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1857
1858 /* Must set freqs->new to intermediate frequency */
1859 static int __target_intermediate(struct cpufreq_policy *policy,
1860                                  struct cpufreq_freqs *freqs, int index)
1861 {
1862         int ret;
1863
1864         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1865
1866         /* We don't need to switch to intermediate freq */
1867         if (!freqs->new)
1868                 return 0;
1869
1870         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1871                  __func__, policy->cpu, freqs->old, freqs->new);
1872
1873         cpufreq_freq_transition_begin(policy, freqs);
1874         ret = cpufreq_driver->target_intermediate(policy, index);
1875         cpufreq_freq_transition_end(policy, freqs, ret);
1876
1877         if (ret)
1878                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1879                        __func__, ret);
1880
1881         return ret;
1882 }
1883
1884 static int __target_index(struct cpufreq_policy *policy, int index)
1885 {
1886         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1887         unsigned int intermediate_freq = 0;
1888         unsigned int newfreq = policy->freq_table[index].frequency;
1889         int retval = -EINVAL;
1890         bool notify;
1891
1892         if (newfreq == policy->cur)
1893                 return 0;
1894
1895         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1896         if (notify) {
1897                 /* Handle switching to intermediate frequency */
1898                 if (cpufreq_driver->get_intermediate) {
1899                         retval = __target_intermediate(policy, &freqs, index);
1900                         if (retval)
1901                                 return retval;
1902
1903                         intermediate_freq = freqs.new;
1904                         /* Set old freq to intermediate */
1905                         if (intermediate_freq)
1906                                 freqs.old = freqs.new;
1907                 }
1908
1909                 freqs.new = newfreq;
1910                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1911                          __func__, policy->cpu, freqs.old, freqs.new);
1912
1913                 cpufreq_freq_transition_begin(policy, &freqs);
1914         }
1915
1916         retval = cpufreq_driver->target_index(policy, index);
1917         if (retval)
1918                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1919                        retval);
1920
1921         if (notify) {
1922                 cpufreq_freq_transition_end(policy, &freqs, retval);
1923
1924                 /*
1925                  * Failed after setting to intermediate freq? Driver should have
1926                  * reverted back to initial frequency and so should we. Check
1927                  * here for intermediate_freq instead of get_intermediate, in
1928                  * case we haven't switched to intermediate freq at all.
1929                  */
1930                 if (unlikely(retval && intermediate_freq)) {
1931                         freqs.old = intermediate_freq;
1932                         freqs.new = policy->restore_freq;
1933                         cpufreq_freq_transition_begin(policy, &freqs);
1934                         cpufreq_freq_transition_end(policy, &freqs, 0);
1935                 }
1936         }
1937
1938         return retval;
1939 }
1940
1941 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1942                             unsigned int target_freq,
1943                             unsigned int relation)
1944 {
1945         unsigned int old_target_freq = target_freq;
1946         int index;
1947
1948         if (cpufreq_disabled())
1949                 return -ENODEV;
1950
1951         /* Make sure that target_freq is within supported range */
1952         target_freq = clamp_val(target_freq, policy->min, policy->max);
1953
1954         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1955                  policy->cpu, target_freq, relation, old_target_freq);
1956
1957         /*
1958          * This might look like a redundant call as we are checking it again
1959          * after finding index. But it is left intentionally for cases where
1960          * exactly same freq is called again and so we can save on few function
1961          * calls.
1962          */
1963         if (target_freq == policy->cur)
1964                 return 0;
1965
1966         /* Save last value to restore later on errors */
1967         policy->restore_freq = policy->cur;
1968
1969         if (cpufreq_driver->target)
1970                 return cpufreq_driver->target(policy, target_freq, relation);
1971
1972         if (!cpufreq_driver->target_index)
1973                 return -EINVAL;
1974
1975         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1976
1977         return __target_index(policy, index);
1978 }
1979 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1980
1981 int cpufreq_driver_target(struct cpufreq_policy *policy,
1982                           unsigned int target_freq,
1983                           unsigned int relation)
1984 {
1985         int ret = -EINVAL;
1986
1987         down_write(&policy->rwsem);
1988
1989         ret = __cpufreq_driver_target(policy, target_freq, relation);
1990
1991         up_write(&policy->rwsem);
1992
1993         return ret;
1994 }
1995 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1996
1997 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1998 {
1999         return NULL;
2000 }
2001
2002 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2003 {
2004         int ret;
2005
2006         /* Don't start any governor operations if we are entering suspend */
2007         if (cpufreq_suspended)
2008                 return 0;
2009         /*
2010          * Governor might not be initiated here if ACPI _PPC changed
2011          * notification happened, so check it.
2012          */
2013         if (!policy->governor)
2014                 return -EINVAL;
2015
2016         if (policy->governor->max_transition_latency &&
2017             policy->cpuinfo.transition_latency >
2018             policy->governor->max_transition_latency) {
2019                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2020
2021                 if (gov) {
2022                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2023                                 policy->governor->name, gov->name);
2024                         policy->governor = gov;
2025                 } else {
2026                         return -EINVAL;
2027                 }
2028         }
2029
2030         if (!try_module_get(policy->governor->owner))
2031                 return -EINVAL;
2032
2033         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2034
2035         if (policy->governor->init) {
2036                 ret = policy->governor->init(policy);
2037                 if (ret) {
2038                         module_put(policy->governor->owner);
2039                         return ret;
2040                 }
2041         }
2042
2043         return 0;
2044 }
2045
2046 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2047 {
2048         if (cpufreq_suspended || !policy->governor)
2049                 return;
2050
2051         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2052
2053         if (policy->governor->exit)
2054                 policy->governor->exit(policy);
2055
2056         module_put(policy->governor->owner);
2057 }
2058
2059 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2060 {
2061         int ret;
2062
2063         if (cpufreq_suspended)
2064                 return 0;
2065
2066         if (!policy->governor)
2067                 return -EINVAL;
2068
2069         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2070
2071         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2072                 cpufreq_update_current_freq(policy);
2073
2074         if (policy->governor->start) {
2075                 ret = policy->governor->start(policy);
2076                 if (ret)
2077                         return ret;
2078         }
2079
2080         if (policy->governor->limits)
2081                 policy->governor->limits(policy);
2082
2083         return 0;
2084 }
2085
2086 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2087 {
2088         if (cpufreq_suspended || !policy->governor)
2089                 return;
2090
2091         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2092
2093         if (policy->governor->stop)
2094                 policy->governor->stop(policy);
2095 }
2096
2097 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2098 {
2099         if (cpufreq_suspended || !policy->governor)
2100                 return;
2101
2102         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2103
2104         if (policy->governor->limits)
2105                 policy->governor->limits(policy);
2106 }
2107
2108 int cpufreq_register_governor(struct cpufreq_governor *governor)
2109 {
2110         int err;
2111
2112         if (!governor)
2113                 return -EINVAL;
2114
2115         if (cpufreq_disabled())
2116                 return -ENODEV;
2117
2118         mutex_lock(&cpufreq_governor_mutex);
2119
2120         err = -EBUSY;
2121         if (!find_governor(governor->name)) {
2122                 err = 0;
2123                 list_add(&governor->governor_list, &cpufreq_governor_list);
2124         }
2125
2126         mutex_unlock(&cpufreq_governor_mutex);
2127         return err;
2128 }
2129 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2130
2131 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2132 {
2133         struct cpufreq_policy *policy;
2134         unsigned long flags;
2135
2136         if (!governor)
2137                 return;
2138
2139         if (cpufreq_disabled())
2140                 return;
2141
2142         /* clear last_governor for all inactive policies */
2143         read_lock_irqsave(&cpufreq_driver_lock, flags);
2144         for_each_inactive_policy(policy) {
2145                 if (!strcmp(policy->last_governor, governor->name)) {
2146                         policy->governor = NULL;
2147                         strcpy(policy->last_governor, "\0");
2148                 }
2149         }
2150         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2151
2152         mutex_lock(&cpufreq_governor_mutex);
2153         list_del(&governor->governor_list);
2154         mutex_unlock(&cpufreq_governor_mutex);
2155         return;
2156 }
2157 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2158
2159
2160 /*********************************************************************
2161  *                          POLICY INTERFACE                         *
2162  *********************************************************************/
2163
2164 /**
2165  * cpufreq_get_policy - get the current cpufreq_policy
2166  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2167  *      is written
2168  *
2169  * Reads the current cpufreq policy.
2170  */
2171 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2172 {
2173         struct cpufreq_policy *cpu_policy;
2174         if (!policy)
2175                 return -EINVAL;
2176
2177         cpu_policy = cpufreq_cpu_get(cpu);
2178         if (!cpu_policy)
2179                 return -EINVAL;
2180
2181         memcpy(policy, cpu_policy, sizeof(*policy));
2182
2183         cpufreq_cpu_put(cpu_policy);
2184         return 0;
2185 }
2186 EXPORT_SYMBOL(cpufreq_get_policy);
2187
2188 /*
2189  * policy : current policy.
2190  * new_policy: policy to be set.
2191  */
2192 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2193                                 struct cpufreq_policy *new_policy)
2194 {
2195         struct cpufreq_governor *old_gov;
2196         int ret;
2197
2198         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2199                  new_policy->cpu, new_policy->min, new_policy->max);
2200
2201         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2202
2203         /*
2204         * This check works well when we store new min/max freq attributes,
2205         * because new_policy is a copy of policy with one field updated.
2206         */
2207         if (new_policy->min > new_policy->max)
2208                 return -EINVAL;
2209
2210         /* verify the cpu speed can be set within this limit */
2211         ret = cpufreq_driver->verify(new_policy);
2212         if (ret)
2213                 return ret;
2214
2215         /* adjust if necessary - all reasons */
2216         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2217                         CPUFREQ_ADJUST, new_policy);
2218
2219         /*
2220          * verify the cpu speed can be set within this limit, which might be
2221          * different to the first one
2222          */
2223         ret = cpufreq_driver->verify(new_policy);
2224         if (ret)
2225                 return ret;
2226
2227         /* notification of the new policy */
2228         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2229                         CPUFREQ_NOTIFY, new_policy);
2230
2231         policy->min = new_policy->min;
2232         policy->max = new_policy->max;
2233
2234         policy->cached_target_freq = UINT_MAX;
2235
2236         pr_debug("new min and max freqs are %u - %u kHz\n",
2237                  policy->min, policy->max);
2238
2239         if (cpufreq_driver->setpolicy) {
2240                 policy->policy = new_policy->policy;
2241                 pr_debug("setting range\n");
2242                 return cpufreq_driver->setpolicy(new_policy);
2243         }
2244
2245         if (new_policy->governor == policy->governor) {
2246                 pr_debug("cpufreq: governor limits update\n");
2247                 cpufreq_governor_limits(policy);
2248                 return 0;
2249         }
2250
2251         pr_debug("governor switch\n");
2252
2253         /* save old, working values */
2254         old_gov = policy->governor;
2255         /* end old governor */
2256         if (old_gov) {
2257                 cpufreq_stop_governor(policy);
2258                 cpufreq_exit_governor(policy);
2259         }
2260
2261         /* start new governor */
2262         policy->governor = new_policy->governor;
2263         ret = cpufreq_init_governor(policy);
2264         if (!ret) {
2265                 ret = cpufreq_start_governor(policy);
2266                 if (!ret) {
2267                         pr_debug("cpufreq: governor change\n");
2268                         return 0;
2269                 }
2270                 cpufreq_exit_governor(policy);
2271         }
2272
2273         /* new governor failed, so re-start old one */
2274         pr_debug("starting governor %s failed\n", policy->governor->name);
2275         if (old_gov) {
2276                 policy->governor = old_gov;
2277                 if (cpufreq_init_governor(policy))
2278                         policy->governor = NULL;
2279                 else
2280                         cpufreq_start_governor(policy);
2281         }
2282
2283         return ret;
2284 }
2285
2286 /**
2287  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2288  *      @cpu: CPU which shall be re-evaluated
2289  *
2290  *      Useful for policy notifiers which have different necessities
2291  *      at different times.
2292  */
2293 int cpufreq_update_policy(unsigned int cpu)
2294 {
2295         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2296         struct cpufreq_policy new_policy;
2297         int ret;
2298
2299         if (!policy)
2300                 return -ENODEV;
2301
2302         down_write(&policy->rwsem);
2303
2304         pr_debug("updating policy for CPU %u\n", cpu);
2305         memcpy(&new_policy, policy, sizeof(*policy));
2306         new_policy.min = policy->user_policy.min;
2307         new_policy.max = policy->user_policy.max;
2308
2309         /*
2310          * BIOS might change freq behind our back
2311          * -> ask driver for current freq and notify governors about a change
2312          */
2313         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2314                 if (cpufreq_suspended) {
2315                         ret = -EAGAIN;
2316                         goto unlock;
2317                 }
2318                 new_policy.cur = cpufreq_update_current_freq(policy);
2319                 if (WARN_ON(!new_policy.cur)) {
2320                         ret = -EIO;
2321                         goto unlock;
2322                 }
2323         }
2324
2325         ret = cpufreq_set_policy(policy, &new_policy);
2326
2327 unlock:
2328         up_write(&policy->rwsem);
2329
2330         cpufreq_cpu_put(policy);
2331         return ret;
2332 }
2333 EXPORT_SYMBOL(cpufreq_update_policy);
2334
2335 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2336                                         unsigned long action, void *hcpu)
2337 {
2338         unsigned int cpu = (unsigned long)hcpu;
2339
2340         switch (action & ~CPU_TASKS_FROZEN) {
2341         case CPU_ONLINE:
2342         case CPU_DOWN_FAILED:
2343                 cpufreq_online(cpu);
2344                 break;
2345
2346         case CPU_DOWN_PREPARE:
2347                 cpufreq_offline(cpu);
2348                 break;
2349         }
2350         return NOTIFY_OK;
2351 }
2352
2353 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2354         .notifier_call = cpufreq_cpu_callback,
2355 };
2356
2357 /*********************************************************************
2358  *               BOOST                                               *
2359  *********************************************************************/
2360 static int cpufreq_boost_set_sw(int state)
2361 {
2362         struct cpufreq_policy *policy;
2363         int ret = -EINVAL;
2364
2365         for_each_active_policy(policy) {
2366                 if (!policy->freq_table)
2367                         continue;
2368
2369                 ret = cpufreq_frequency_table_cpuinfo(policy,
2370                                                       policy->freq_table);
2371                 if (ret) {
2372                         pr_err("%s: Policy frequency update failed\n",
2373                                __func__);
2374                         break;
2375                 }
2376
2377                 down_write(&policy->rwsem);
2378                 policy->user_policy.max = policy->max;
2379                 cpufreq_governor_limits(policy);
2380                 up_write(&policy->rwsem);
2381         }
2382
2383         return ret;
2384 }
2385
2386 int cpufreq_boost_trigger_state(int state)
2387 {
2388         unsigned long flags;
2389         int ret = 0;
2390
2391         if (cpufreq_driver->boost_enabled == state)
2392                 return 0;
2393
2394         write_lock_irqsave(&cpufreq_driver_lock, flags);
2395         cpufreq_driver->boost_enabled = state;
2396         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2397
2398         ret = cpufreq_driver->set_boost(state);
2399         if (ret) {
2400                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2401                 cpufreq_driver->boost_enabled = !state;
2402                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2403
2404                 pr_err("%s: Cannot %s BOOST\n",
2405                        __func__, state ? "enable" : "disable");
2406         }
2407
2408         return ret;
2409 }
2410
2411 static bool cpufreq_boost_supported(void)
2412 {
2413         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2414 }
2415
2416 static int create_boost_sysfs_file(void)
2417 {
2418         int ret;
2419
2420         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2421         if (ret)
2422                 pr_err("%s: cannot register global BOOST sysfs file\n",
2423                        __func__);
2424
2425         return ret;
2426 }
2427
2428 static void remove_boost_sysfs_file(void)
2429 {
2430         if (cpufreq_boost_supported())
2431                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2432 }
2433
2434 int cpufreq_enable_boost_support(void)
2435 {
2436         if (!cpufreq_driver)
2437                 return -EINVAL;
2438
2439         if (cpufreq_boost_supported())
2440                 return 0;
2441
2442         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2443
2444         /* This will get removed on driver unregister */
2445         return create_boost_sysfs_file();
2446 }
2447 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2448
2449 int cpufreq_boost_enabled(void)
2450 {
2451         return cpufreq_driver->boost_enabled;
2452 }
2453 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2454
2455 /*********************************************************************
2456  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2457  *********************************************************************/
2458
2459 /**
2460  * cpufreq_register_driver - register a CPU Frequency driver
2461  * @driver_data: A struct cpufreq_driver containing the values#
2462  * submitted by the CPU Frequency driver.
2463  *
2464  * Registers a CPU Frequency driver to this core code. This code
2465  * returns zero on success, -EEXIST when another driver got here first
2466  * (and isn't unregistered in the meantime).
2467  *
2468  */
2469 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2470 {
2471         unsigned long flags;
2472         int ret;
2473
2474         if (cpufreq_disabled())
2475                 return -ENODEV;
2476
2477         if (!driver_data || !driver_data->verify || !driver_data->init ||
2478             !(driver_data->setpolicy || driver_data->target_index ||
2479                     driver_data->target) ||
2480              (driver_data->setpolicy && (driver_data->target_index ||
2481                     driver_data->target)) ||
2482              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2483                 return -EINVAL;
2484
2485         pr_debug("trying to register driver %s\n", driver_data->name);
2486
2487         /* Protect against concurrent CPU online/offline. */
2488         get_online_cpus();
2489
2490         write_lock_irqsave(&cpufreq_driver_lock, flags);
2491         if (cpufreq_driver) {
2492                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2493                 ret = -EEXIST;
2494                 goto out;
2495         }
2496         cpufreq_driver = driver_data;
2497         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2498
2499         if (driver_data->setpolicy)
2500                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2501
2502         if (cpufreq_boost_supported()) {
2503                 ret = create_boost_sysfs_file();
2504                 if (ret)
2505                         goto err_null_driver;
2506         }
2507
2508         ret = subsys_interface_register(&cpufreq_interface);
2509         if (ret)
2510                 goto err_boost_unreg;
2511
2512         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2513             list_empty(&cpufreq_policy_list)) {
2514                 /* if all ->init() calls failed, unregister */
2515                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2516                          driver_data->name);
2517                 goto err_if_unreg;
2518         }
2519
2520         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2521         pr_debug("driver %s up and running\n", driver_data->name);
2522         goto out;
2523
2524 err_if_unreg:
2525         subsys_interface_unregister(&cpufreq_interface);
2526 err_boost_unreg:
2527         remove_boost_sysfs_file();
2528 err_null_driver:
2529         write_lock_irqsave(&cpufreq_driver_lock, flags);
2530         cpufreq_driver = NULL;
2531         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2532 out:
2533         put_online_cpus();
2534         return ret;
2535 }
2536 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2537
2538 /**
2539  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2540  *
2541  * Unregister the current CPUFreq driver. Only call this if you have
2542  * the right to do so, i.e. if you have succeeded in initialising before!
2543  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2544  * currently not initialised.
2545  */
2546 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2547 {
2548         unsigned long flags;
2549
2550         if (!cpufreq_driver || (driver != cpufreq_driver))
2551                 return -EINVAL;
2552
2553         pr_debug("unregistering driver %s\n", driver->name);
2554
2555         /* Protect against concurrent cpu hotplug */
2556         get_online_cpus();
2557         subsys_interface_unregister(&cpufreq_interface);
2558         remove_boost_sysfs_file();
2559         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2560
2561         write_lock_irqsave(&cpufreq_driver_lock, flags);
2562
2563         cpufreq_driver = NULL;
2564
2565         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2566         put_online_cpus();
2567
2568         return 0;
2569 }
2570 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2571
2572 /*
2573  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2574  * or mutexes when secondary CPUs are halted.
2575  */
2576 static struct syscore_ops cpufreq_syscore_ops = {
2577         .shutdown = cpufreq_suspend,
2578 };
2579
2580 struct kobject *cpufreq_global_kobject;
2581 EXPORT_SYMBOL(cpufreq_global_kobject);
2582
2583 static int __init cpufreq_core_init(void)
2584 {
2585         if (cpufreq_disabled())
2586                 return -ENODEV;
2587
2588         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2589         BUG_ON(!cpufreq_global_kobject);
2590
2591         register_syscore_ops(&cpufreq_syscore_ops);
2592
2593         return 0;
2594 }
2595 core_initcall(cpufreq_core_init);