Merge tag 'hsi-for-4.16' of git://git.kernel.org/pub/scm/linux/kernel/git/sre/linux-hsi
[linux-2.6-microblaze.git] / drivers / cpufreq / cpufreq.c
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
7  *
8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
9  *      Added handling for CPU hotplug
10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
11  *      Fix handling for CPU hotplug -- affected CPUs
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License version 2 as
15  * published by the Free Software Foundation.
16  */
17
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20 #include <linux/cpu.h>
21 #include <linux/cpufreq.h>
22 #include <linux/delay.h>
23 #include <linux/device.h>
24 #include <linux/init.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/module.h>
27 #include <linux/mutex.h>
28 #include <linux/slab.h>
29 #include <linux/suspend.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/tick.h>
32 #include <trace/events/power.h>
33
34 static LIST_HEAD(cpufreq_policy_list);
35
36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
37 {
38         return cpumask_empty(policy->cpus);
39 }
40
41 /* 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_nsecs(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 = div_u64(cur_wall_time, NSEC_PER_USEC);
147
148         return div_u64(idle_time, NSEC_PER_USEC);
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 __weak void arch_set_freq_scale(struct cpumask *cpus, unsigned long cur_freq,
165                 unsigned long max_freq)
166 {
167 }
168 EXPORT_SYMBOL_GPL(arch_set_freq_scale);
169
170 /*
171  * This is a generic cpufreq init() routine which can be used by cpufreq
172  * drivers of SMP systems. It will do following:
173  * - validate & show freq table passed
174  * - set policies transition latency
175  * - policy->cpus with all possible CPUs
176  */
177 int cpufreq_generic_init(struct cpufreq_policy *policy,
178                 struct cpufreq_frequency_table *table,
179                 unsigned int transition_latency)
180 {
181         int ret;
182
183         ret = cpufreq_table_validate_and_show(policy, table);
184         if (ret) {
185                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
186                 return ret;
187         }
188
189         policy->cpuinfo.transition_latency = transition_latency;
190
191         /*
192          * The driver only supports the SMP configuration where all processors
193          * share the clock and voltage and clock.
194          */
195         cpumask_setall(policy->cpus);
196
197         return 0;
198 }
199 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
200
201 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
202 {
203         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
204
205         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
206 }
207 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
208
209 unsigned int cpufreq_generic_get(unsigned int cpu)
210 {
211         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
212
213         if (!policy || IS_ERR(policy->clk)) {
214                 pr_err("%s: No %s associated to cpu: %d\n",
215                        __func__, policy ? "clk" : "policy", cpu);
216                 return 0;
217         }
218
219         return clk_get_rate(policy->clk) / 1000;
220 }
221 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
222
223 /**
224  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
225  *
226  * @cpu: cpu to find policy for.
227  *
228  * This returns policy for 'cpu', returns NULL if it doesn't exist.
229  * It also increments the kobject reference count to mark it busy and so would
230  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
231  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
232  * freed as that depends on the kobj count.
233  *
234  * Return: A valid policy on success, otherwise NULL on failure.
235  */
236 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
237 {
238         struct cpufreq_policy *policy = NULL;
239         unsigned long flags;
240
241         if (WARN_ON(cpu >= nr_cpu_ids))
242                 return NULL;
243
244         /* get the cpufreq driver */
245         read_lock_irqsave(&cpufreq_driver_lock, flags);
246
247         if (cpufreq_driver) {
248                 /* get the CPU */
249                 policy = cpufreq_cpu_get_raw(cpu);
250                 if (policy)
251                         kobject_get(&policy->kobj);
252         }
253
254         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
255
256         return policy;
257 }
258 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
259
260 /**
261  * cpufreq_cpu_put: Decrements the usage count of a policy
262  *
263  * @policy: policy earlier returned by cpufreq_cpu_get().
264  *
265  * This decrements the kobject reference count incremented earlier by calling
266  * cpufreq_cpu_get().
267  */
268 void cpufreq_cpu_put(struct cpufreq_policy *policy)
269 {
270         kobject_put(&policy->kobj);
271 }
272 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
273
274 /*********************************************************************
275  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
276  *********************************************************************/
277
278 /**
279  * adjust_jiffies - adjust the system "loops_per_jiffy"
280  *
281  * This function alters the system "loops_per_jiffy" for the clock
282  * speed change. Note that loops_per_jiffy cannot be updated on SMP
283  * systems as each CPU might be scaled differently. So, use the arch
284  * per-CPU loops_per_jiffy value wherever possible.
285  */
286 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
287 {
288 #ifndef CONFIG_SMP
289         static unsigned long l_p_j_ref;
290         static unsigned int l_p_j_ref_freq;
291
292         if (ci->flags & CPUFREQ_CONST_LOOPS)
293                 return;
294
295         if (!l_p_j_ref_freq) {
296                 l_p_j_ref = loops_per_jiffy;
297                 l_p_j_ref_freq = ci->old;
298                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
299                          l_p_j_ref, l_p_j_ref_freq);
300         }
301         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
302                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
303                                                                 ci->new);
304                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
305                          loops_per_jiffy, ci->new);
306         }
307 #endif
308 }
309
310 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
311                 struct cpufreq_freqs *freqs, unsigned int state)
312 {
313         BUG_ON(irqs_disabled());
314
315         if (cpufreq_disabled())
316                 return;
317
318         freqs->flags = cpufreq_driver->flags;
319         pr_debug("notification %u of frequency transition to %u kHz\n",
320                  state, freqs->new);
321
322         switch (state) {
323
324         case CPUFREQ_PRECHANGE:
325                 /* detect if the driver reported a value as "old frequency"
326                  * which is not equal to what the cpufreq core thinks is
327                  * "old frequency".
328                  */
329                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
330                         if ((policy) && (policy->cpu == freqs->cpu) &&
331                             (policy->cur) && (policy->cur != freqs->old)) {
332                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
333                                          freqs->old, policy->cur);
334                                 freqs->old = policy->cur;
335                         }
336                 }
337                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
338                                 CPUFREQ_PRECHANGE, freqs);
339                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
340                 break;
341
342         case CPUFREQ_POSTCHANGE:
343                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
344                 pr_debug("FREQ: %lu - CPU: %lu\n",
345                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
346                 trace_cpu_frequency(freqs->new, freqs->cpu);
347                 cpufreq_stats_record_transition(policy, freqs->new);
348                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
349                                 CPUFREQ_POSTCHANGE, freqs);
350                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
351                         policy->cur = freqs->new;
352                 break;
353         }
354 }
355
356 /**
357  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
358  * on frequency transition.
359  *
360  * This function calls the transition notifiers and the "adjust_jiffies"
361  * function. It is called twice on all CPU frequency changes that have
362  * external effects.
363  */
364 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
365                 struct cpufreq_freqs *freqs, unsigned int state)
366 {
367         for_each_cpu(freqs->cpu, policy->cpus)
368                 __cpufreq_notify_transition(policy, freqs, state);
369 }
370
371 /* Do post notifications when there are chances that transition has failed */
372 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
373                 struct cpufreq_freqs *freqs, int transition_failed)
374 {
375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
376         if (!transition_failed)
377                 return;
378
379         swap(freqs->old, freqs->new);
380         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
381         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
382 }
383
384 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
385                 struct cpufreq_freqs *freqs)
386 {
387
388         /*
389          * Catch double invocations of _begin() which lead to self-deadlock.
390          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
391          * doesn't invoke _begin() on their behalf, and hence the chances of
392          * double invocations are very low. Moreover, there are scenarios
393          * where these checks can emit false-positive warnings in these
394          * drivers; so we avoid that by skipping them altogether.
395          */
396         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
397                                 && current == policy->transition_task);
398
399 wait:
400         wait_event(policy->transition_wait, !policy->transition_ongoing);
401
402         spin_lock(&policy->transition_lock);
403
404         if (unlikely(policy->transition_ongoing)) {
405                 spin_unlock(&policy->transition_lock);
406                 goto wait;
407         }
408
409         policy->transition_ongoing = true;
410         policy->transition_task = current;
411
412         spin_unlock(&policy->transition_lock);
413
414         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
415 }
416 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
417
418 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
419                 struct cpufreq_freqs *freqs, int transition_failed)
420 {
421         if (unlikely(WARN_ON(!policy->transition_ongoing)))
422                 return;
423
424         cpufreq_notify_post_transition(policy, freqs, transition_failed);
425
426         policy->transition_ongoing = false;
427         policy->transition_task = NULL;
428
429         wake_up(&policy->transition_wait);
430 }
431 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
432
433 /*
434  * Fast frequency switching status count.  Positive means "enabled", negative
435  * means "disabled" and 0 means "not decided yet".
436  */
437 static int cpufreq_fast_switch_count;
438 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
439
440 static void cpufreq_list_transition_notifiers(void)
441 {
442         struct notifier_block *nb;
443
444         pr_info("Registered transition notifiers:\n");
445
446         mutex_lock(&cpufreq_transition_notifier_list.mutex);
447
448         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
449                 pr_info("%pF\n", nb->notifier_call);
450
451         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
452 }
453
454 /**
455  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
456  * @policy: cpufreq policy to enable fast frequency switching for.
457  *
458  * Try to enable fast frequency switching for @policy.
459  *
460  * The attempt will fail if there is at least one transition notifier registered
461  * at this point, as fast frequency switching is quite fundamentally at odds
462  * with transition notifiers.  Thus if successful, it will make registration of
463  * transition notifiers fail going forward.
464  */
465 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
466 {
467         lockdep_assert_held(&policy->rwsem);
468
469         if (!policy->fast_switch_possible)
470                 return;
471
472         mutex_lock(&cpufreq_fast_switch_lock);
473         if (cpufreq_fast_switch_count >= 0) {
474                 cpufreq_fast_switch_count++;
475                 policy->fast_switch_enabled = true;
476         } else {
477                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
478                         policy->cpu);
479                 cpufreq_list_transition_notifiers();
480         }
481         mutex_unlock(&cpufreq_fast_switch_lock);
482 }
483 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
484
485 /**
486  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
487  * @policy: cpufreq policy to disable fast frequency switching for.
488  */
489 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
490 {
491         mutex_lock(&cpufreq_fast_switch_lock);
492         if (policy->fast_switch_enabled) {
493                 policy->fast_switch_enabled = false;
494                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
495                         cpufreq_fast_switch_count--;
496         }
497         mutex_unlock(&cpufreq_fast_switch_lock);
498 }
499 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
500
501 /**
502  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
503  * one.
504  * @target_freq: target frequency to resolve.
505  *
506  * The target to driver frequency mapping is cached in the policy.
507  *
508  * Return: Lowest driver-supported frequency greater than or equal to the
509  * given target_freq, subject to policy (min/max) and driver limitations.
510  */
511 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
512                                          unsigned int target_freq)
513 {
514         target_freq = clamp_val(target_freq, policy->min, policy->max);
515         policy->cached_target_freq = target_freq;
516
517         if (cpufreq_driver->target_index) {
518                 int idx;
519
520                 idx = cpufreq_frequency_table_target(policy, target_freq,
521                                                      CPUFREQ_RELATION_L);
522                 policy->cached_resolved_idx = idx;
523                 return policy->freq_table[idx].frequency;
524         }
525
526         if (cpufreq_driver->resolve_freq)
527                 return cpufreq_driver->resolve_freq(policy, target_freq);
528
529         return target_freq;
530 }
531 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
532
533 unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
534 {
535         unsigned int latency;
536
537         if (policy->transition_delay_us)
538                 return policy->transition_delay_us;
539
540         latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
541         if (latency) {
542                 /*
543                  * For platforms that can change the frequency very fast (< 10
544                  * us), the above formula gives a decent transition delay. But
545                  * for platforms where transition_latency is in milliseconds, it
546                  * ends up giving unrealistic values.
547                  *
548                  * Cap the default transition delay to 10 ms, which seems to be
549                  * a reasonable amount of time after which we should reevaluate
550                  * the frequency.
551                  */
552                 return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
553         }
554
555         return LATENCY_MULTIPLIER;
556 }
557 EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
558
559 /*********************************************************************
560  *                          SYSFS INTERFACE                          *
561  *********************************************************************/
562 static ssize_t show_boost(struct kobject *kobj,
563                                  struct attribute *attr, char *buf)
564 {
565         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
566 }
567
568 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
569                                   const char *buf, size_t count)
570 {
571         int ret, enable;
572
573         ret = sscanf(buf, "%d", &enable);
574         if (ret != 1 || enable < 0 || enable > 1)
575                 return -EINVAL;
576
577         if (cpufreq_boost_trigger_state(enable)) {
578                 pr_err("%s: Cannot %s BOOST!\n",
579                        __func__, enable ? "enable" : "disable");
580                 return -EINVAL;
581         }
582
583         pr_debug("%s: cpufreq BOOST %s\n",
584                  __func__, enable ? "enabled" : "disabled");
585
586         return count;
587 }
588 define_one_global_rw(boost);
589
590 static struct cpufreq_governor *find_governor(const char *str_governor)
591 {
592         struct cpufreq_governor *t;
593
594         for_each_governor(t)
595                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
596                         return t;
597
598         return NULL;
599 }
600
601 /**
602  * cpufreq_parse_governor - parse a governor string
603  */
604 static int cpufreq_parse_governor(char *str_governor,
605                                   struct cpufreq_policy *policy)
606 {
607         if (cpufreq_driver->setpolicy) {
608                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
609                         policy->policy = CPUFREQ_POLICY_PERFORMANCE;
610                         return 0;
611                 }
612
613                 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
614                         policy->policy = CPUFREQ_POLICY_POWERSAVE;
615                         return 0;
616                 }
617         } else {
618                 struct cpufreq_governor *t;
619
620                 mutex_lock(&cpufreq_governor_mutex);
621
622                 t = find_governor(str_governor);
623                 if (!t) {
624                         int ret;
625
626                         mutex_unlock(&cpufreq_governor_mutex);
627
628                         ret = request_module("cpufreq_%s", str_governor);
629                         if (ret)
630                                 return -EINVAL;
631
632                         mutex_lock(&cpufreq_governor_mutex);
633
634                         t = find_governor(str_governor);
635                 }
636                 if (t && !try_module_get(t->owner))
637                         t = NULL;
638
639                 mutex_unlock(&cpufreq_governor_mutex);
640
641                 if (t) {
642                         policy->governor = t;
643                         return 0;
644                 }
645         }
646
647         return -EINVAL;
648 }
649
650 /**
651  * cpufreq_per_cpu_attr_read() / show_##file_name() -
652  * print out cpufreq information
653  *
654  * Write out information from cpufreq_driver->policy[cpu]; object must be
655  * "unsigned int".
656  */
657
658 #define show_one(file_name, object)                     \
659 static ssize_t show_##file_name                         \
660 (struct cpufreq_policy *policy, char *buf)              \
661 {                                                       \
662         return sprintf(buf, "%u\n", policy->object);    \
663 }
664
665 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
666 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
667 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
668 show_one(scaling_min_freq, min);
669 show_one(scaling_max_freq, max);
670
671 __weak unsigned int arch_freq_get_on_cpu(int cpu)
672 {
673         return 0;
674 }
675
676 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
677 {
678         ssize_t ret;
679         unsigned int freq;
680
681         freq = arch_freq_get_on_cpu(policy->cpu);
682         if (freq)
683                 ret = sprintf(buf, "%u\n", freq);
684         else if (cpufreq_driver && cpufreq_driver->setpolicy &&
685                         cpufreq_driver->get)
686                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
687         else
688                 ret = sprintf(buf, "%u\n", policy->cur);
689         return ret;
690 }
691
692 static int cpufreq_set_policy(struct cpufreq_policy *policy,
693                                 struct cpufreq_policy *new_policy);
694
695 /**
696  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
697  */
698 #define store_one(file_name, object)                    \
699 static ssize_t store_##file_name                                        \
700 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
701 {                                                                       \
702         int ret, temp;                                                  \
703         struct cpufreq_policy new_policy;                               \
704                                                                         \
705         memcpy(&new_policy, policy, sizeof(*policy));                   \
706                                                                         \
707         ret = sscanf(buf, "%u", &new_policy.object);                    \
708         if (ret != 1)                                                   \
709                 return -EINVAL;                                         \
710                                                                         \
711         temp = new_policy.object;                                       \
712         ret = cpufreq_set_policy(policy, &new_policy);          \
713         if (!ret)                                                       \
714                 policy->user_policy.object = temp;                      \
715                                                                         \
716         return ret ? ret : count;                                       \
717 }
718
719 store_one(scaling_min_freq, min);
720 store_one(scaling_max_freq, max);
721
722 /**
723  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
724  */
725 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
726                                         char *buf)
727 {
728         unsigned int cur_freq = __cpufreq_get(policy);
729
730         if (cur_freq)
731                 return sprintf(buf, "%u\n", cur_freq);
732
733         return sprintf(buf, "<unknown>\n");
734 }
735
736 /**
737  * show_scaling_governor - show the current policy for the specified CPU
738  */
739 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
740 {
741         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
742                 return sprintf(buf, "powersave\n");
743         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
744                 return sprintf(buf, "performance\n");
745         else if (policy->governor)
746                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
747                                 policy->governor->name);
748         return -EINVAL;
749 }
750
751 /**
752  * store_scaling_governor - store policy for the specified CPU
753  */
754 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
755                                         const char *buf, size_t count)
756 {
757         int ret;
758         char    str_governor[16];
759         struct cpufreq_policy new_policy;
760
761         memcpy(&new_policy, policy, sizeof(*policy));
762
763         ret = sscanf(buf, "%15s", str_governor);
764         if (ret != 1)
765                 return -EINVAL;
766
767         if (cpufreq_parse_governor(str_governor, &new_policy))
768                 return -EINVAL;
769
770         ret = cpufreq_set_policy(policy, &new_policy);
771
772         if (new_policy.governor)
773                 module_put(new_policy.governor->owner);
774
775         return ret ? ret : count;
776 }
777
778 /**
779  * show_scaling_driver - show the cpufreq driver currently loaded
780  */
781 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
782 {
783         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
784 }
785
786 /**
787  * show_scaling_available_governors - show the available CPUfreq governors
788  */
789 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
790                                                 char *buf)
791 {
792         ssize_t i = 0;
793         struct cpufreq_governor *t;
794
795         if (!has_target()) {
796                 i += sprintf(buf, "performance powersave");
797                 goto out;
798         }
799
800         for_each_governor(t) {
801                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
802                     - (CPUFREQ_NAME_LEN + 2)))
803                         goto out;
804                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
805         }
806 out:
807         i += sprintf(&buf[i], "\n");
808         return i;
809 }
810
811 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
812 {
813         ssize_t i = 0;
814         unsigned int cpu;
815
816         for_each_cpu(cpu, mask) {
817                 if (i)
818                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
819                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
820                 if (i >= (PAGE_SIZE - 5))
821                         break;
822         }
823         i += sprintf(&buf[i], "\n");
824         return i;
825 }
826 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
827
828 /**
829  * show_related_cpus - show the CPUs affected by each transition even if
830  * hw coordination is in use
831  */
832 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
833 {
834         return cpufreq_show_cpus(policy->related_cpus, buf);
835 }
836
837 /**
838  * show_affected_cpus - show the CPUs affected by each transition
839  */
840 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
841 {
842         return cpufreq_show_cpus(policy->cpus, buf);
843 }
844
845 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
846                                         const char *buf, size_t count)
847 {
848         unsigned int freq = 0;
849         unsigned int ret;
850
851         if (!policy->governor || !policy->governor->store_setspeed)
852                 return -EINVAL;
853
854         ret = sscanf(buf, "%u", &freq);
855         if (ret != 1)
856                 return -EINVAL;
857
858         policy->governor->store_setspeed(policy, freq);
859
860         return count;
861 }
862
863 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
864 {
865         if (!policy->governor || !policy->governor->show_setspeed)
866                 return sprintf(buf, "<unsupported>\n");
867
868         return policy->governor->show_setspeed(policy, buf);
869 }
870
871 /**
872  * show_bios_limit - show the current cpufreq HW/BIOS limitation
873  */
874 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
875 {
876         unsigned int limit;
877         int ret;
878         if (cpufreq_driver->bios_limit) {
879                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
880                 if (!ret)
881                         return sprintf(buf, "%u\n", limit);
882         }
883         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
884 }
885
886 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
887 cpufreq_freq_attr_ro(cpuinfo_min_freq);
888 cpufreq_freq_attr_ro(cpuinfo_max_freq);
889 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
890 cpufreq_freq_attr_ro(scaling_available_governors);
891 cpufreq_freq_attr_ro(scaling_driver);
892 cpufreq_freq_attr_ro(scaling_cur_freq);
893 cpufreq_freq_attr_ro(bios_limit);
894 cpufreq_freq_attr_ro(related_cpus);
895 cpufreq_freq_attr_ro(affected_cpus);
896 cpufreq_freq_attr_rw(scaling_min_freq);
897 cpufreq_freq_attr_rw(scaling_max_freq);
898 cpufreq_freq_attr_rw(scaling_governor);
899 cpufreq_freq_attr_rw(scaling_setspeed);
900
901 static struct attribute *default_attrs[] = {
902         &cpuinfo_min_freq.attr,
903         &cpuinfo_max_freq.attr,
904         &cpuinfo_transition_latency.attr,
905         &scaling_min_freq.attr,
906         &scaling_max_freq.attr,
907         &affected_cpus.attr,
908         &related_cpus.attr,
909         &scaling_governor.attr,
910         &scaling_driver.attr,
911         &scaling_available_governors.attr,
912         &scaling_setspeed.attr,
913         NULL
914 };
915
916 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
917 #define to_attr(a) container_of(a, struct freq_attr, attr)
918
919 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
920 {
921         struct cpufreq_policy *policy = to_policy(kobj);
922         struct freq_attr *fattr = to_attr(attr);
923         ssize_t ret;
924
925         down_read(&policy->rwsem);
926         ret = fattr->show(policy, buf);
927         up_read(&policy->rwsem);
928
929         return ret;
930 }
931
932 static ssize_t store(struct kobject *kobj, struct attribute *attr,
933                      const char *buf, size_t count)
934 {
935         struct cpufreq_policy *policy = to_policy(kobj);
936         struct freq_attr *fattr = to_attr(attr);
937         ssize_t ret = -EINVAL;
938
939         cpus_read_lock();
940
941         if (cpu_online(policy->cpu)) {
942                 down_write(&policy->rwsem);
943                 ret = fattr->store(policy, buf, count);
944                 up_write(&policy->rwsem);
945         }
946
947         cpus_read_unlock();
948
949         return ret;
950 }
951
952 static void cpufreq_sysfs_release(struct kobject *kobj)
953 {
954         struct cpufreq_policy *policy = to_policy(kobj);
955         pr_debug("last reference is dropped\n");
956         complete(&policy->kobj_unregister);
957 }
958
959 static const struct sysfs_ops sysfs_ops = {
960         .show   = show,
961         .store  = store,
962 };
963
964 static struct kobj_type ktype_cpufreq = {
965         .sysfs_ops      = &sysfs_ops,
966         .default_attrs  = default_attrs,
967         .release        = cpufreq_sysfs_release,
968 };
969
970 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
971 {
972         struct device *dev = get_cpu_device(cpu);
973
974         if (!dev)
975                 return;
976
977         if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
978                 return;
979
980         dev_dbg(dev, "%s: Adding symlink\n", __func__);
981         if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
982                 dev_err(dev, "cpufreq symlink creation failed\n");
983 }
984
985 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
986                                    struct device *dev)
987 {
988         dev_dbg(dev, "%s: Removing symlink\n", __func__);
989         sysfs_remove_link(&dev->kobj, "cpufreq");
990 }
991
992 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
993 {
994         struct freq_attr **drv_attr;
995         int ret = 0;
996
997         /* set up files for this cpu device */
998         drv_attr = cpufreq_driver->attr;
999         while (drv_attr && *drv_attr) {
1000                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1001                 if (ret)
1002                         return ret;
1003                 drv_attr++;
1004         }
1005         if (cpufreq_driver->get) {
1006                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1007                 if (ret)
1008                         return ret;
1009         }
1010
1011         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1012         if (ret)
1013                 return ret;
1014
1015         if (cpufreq_driver->bios_limit) {
1016                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1017                 if (ret)
1018                         return ret;
1019         }
1020
1021         return 0;
1022 }
1023
1024 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1025 {
1026         return NULL;
1027 }
1028
1029 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1030 {
1031         struct cpufreq_governor *gov = NULL;
1032         struct cpufreq_policy new_policy;
1033
1034         memcpy(&new_policy, policy, sizeof(*policy));
1035
1036         /* Update governor of new_policy to the governor used before hotplug */
1037         gov = find_governor(policy->last_governor);
1038         if (gov) {
1039                 pr_debug("Restoring governor %s for cpu %d\n",
1040                                 policy->governor->name, policy->cpu);
1041         } else {
1042                 gov = cpufreq_default_governor();
1043                 if (!gov)
1044                         return -ENODATA;
1045         }
1046
1047         new_policy.governor = gov;
1048
1049         /* Use the default policy if there is no last_policy. */
1050         if (cpufreq_driver->setpolicy) {
1051                 if (policy->last_policy)
1052                         new_policy.policy = policy->last_policy;
1053                 else
1054                         cpufreq_parse_governor(gov->name, &new_policy);
1055         }
1056         /* set default policy */
1057         return cpufreq_set_policy(policy, &new_policy);
1058 }
1059
1060 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1061 {
1062         int ret = 0;
1063
1064         /* Has this CPU been taken care of already? */
1065         if (cpumask_test_cpu(cpu, policy->cpus))
1066                 return 0;
1067
1068         down_write(&policy->rwsem);
1069         if (has_target())
1070                 cpufreq_stop_governor(policy);
1071
1072         cpumask_set_cpu(cpu, policy->cpus);
1073
1074         if (has_target()) {
1075                 ret = cpufreq_start_governor(policy);
1076                 if (ret)
1077                         pr_err("%s: Failed to start governor\n", __func__);
1078         }
1079         up_write(&policy->rwsem);
1080         return ret;
1081 }
1082
1083 static void handle_update(struct work_struct *work)
1084 {
1085         struct cpufreq_policy *policy =
1086                 container_of(work, struct cpufreq_policy, update);
1087         unsigned int cpu = policy->cpu;
1088         pr_debug("handle_update for cpu %u called\n", cpu);
1089         cpufreq_update_policy(cpu);
1090 }
1091
1092 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1093 {
1094         struct cpufreq_policy *policy;
1095         int ret;
1096
1097         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1098         if (!policy)
1099                 return NULL;
1100
1101         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1102                 goto err_free_policy;
1103
1104         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1105                 goto err_free_cpumask;
1106
1107         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1108                 goto err_free_rcpumask;
1109
1110         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1111                                    cpufreq_global_kobject, "policy%u", cpu);
1112         if (ret) {
1113                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1114                 goto err_free_real_cpus;
1115         }
1116
1117         INIT_LIST_HEAD(&policy->policy_list);
1118         init_rwsem(&policy->rwsem);
1119         spin_lock_init(&policy->transition_lock);
1120         init_waitqueue_head(&policy->transition_wait);
1121         init_completion(&policy->kobj_unregister);
1122         INIT_WORK(&policy->update, handle_update);
1123
1124         policy->cpu = cpu;
1125         return policy;
1126
1127 err_free_real_cpus:
1128         free_cpumask_var(policy->real_cpus);
1129 err_free_rcpumask:
1130         free_cpumask_var(policy->related_cpus);
1131 err_free_cpumask:
1132         free_cpumask_var(policy->cpus);
1133 err_free_policy:
1134         kfree(policy);
1135
1136         return NULL;
1137 }
1138
1139 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1140 {
1141         struct kobject *kobj;
1142         struct completion *cmp;
1143
1144         down_write(&policy->rwsem);
1145         cpufreq_stats_free_table(policy);
1146         kobj = &policy->kobj;
1147         cmp = &policy->kobj_unregister;
1148         up_write(&policy->rwsem);
1149         kobject_put(kobj);
1150
1151         /*
1152          * We need to make sure that the underlying kobj is
1153          * actually not referenced anymore by anybody before we
1154          * proceed with unloading.
1155          */
1156         pr_debug("waiting for dropping of refcount\n");
1157         wait_for_completion(cmp);
1158         pr_debug("wait complete\n");
1159 }
1160
1161 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1162 {
1163         unsigned long flags;
1164         int cpu;
1165
1166         /* Remove policy from list */
1167         write_lock_irqsave(&cpufreq_driver_lock, flags);
1168         list_del(&policy->policy_list);
1169
1170         for_each_cpu(cpu, policy->related_cpus)
1171                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1172         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1173
1174         cpufreq_policy_put_kobj(policy);
1175         free_cpumask_var(policy->real_cpus);
1176         free_cpumask_var(policy->related_cpus);
1177         free_cpumask_var(policy->cpus);
1178         kfree(policy);
1179 }
1180
1181 static int cpufreq_online(unsigned int cpu)
1182 {
1183         struct cpufreq_policy *policy;
1184         bool new_policy;
1185         unsigned long flags;
1186         unsigned int j;
1187         int ret;
1188
1189         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1190
1191         /* Check if this CPU already has a policy to manage it */
1192         policy = per_cpu(cpufreq_cpu_data, cpu);
1193         if (policy) {
1194                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1195                 if (!policy_is_inactive(policy))
1196                         return cpufreq_add_policy_cpu(policy, cpu);
1197
1198                 /* This is the only online CPU for the policy.  Start over. */
1199                 new_policy = false;
1200                 down_write(&policy->rwsem);
1201                 policy->cpu = cpu;
1202                 policy->governor = NULL;
1203                 up_write(&policy->rwsem);
1204         } else {
1205                 new_policy = true;
1206                 policy = cpufreq_policy_alloc(cpu);
1207                 if (!policy)
1208                         return -ENOMEM;
1209         }
1210
1211         cpumask_copy(policy->cpus, cpumask_of(cpu));
1212
1213         /* call driver. From then on the cpufreq must be able
1214          * to accept all calls to ->verify and ->setpolicy for this CPU
1215          */
1216         ret = cpufreq_driver->init(policy);
1217         if (ret) {
1218                 pr_debug("initialization failed\n");
1219                 goto out_free_policy;
1220         }
1221
1222         down_write(&policy->rwsem);
1223
1224         if (new_policy) {
1225                 /* related_cpus should at least include policy->cpus. */
1226                 cpumask_copy(policy->related_cpus, policy->cpus);
1227         }
1228
1229         /*
1230          * affected cpus must always be the one, which are online. We aren't
1231          * managing offline cpus here.
1232          */
1233         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1234
1235         if (new_policy) {
1236                 policy->user_policy.min = policy->min;
1237                 policy->user_policy.max = policy->max;
1238
1239                 for_each_cpu(j, policy->related_cpus) {
1240                         per_cpu(cpufreq_cpu_data, j) = policy;
1241                         add_cpu_dev_symlink(policy, j);
1242                 }
1243         } else {
1244                 policy->min = policy->user_policy.min;
1245                 policy->max = policy->user_policy.max;
1246         }
1247
1248         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1249                 policy->cur = cpufreq_driver->get(policy->cpu);
1250                 if (!policy->cur) {
1251                         pr_err("%s: ->get() failed\n", __func__);
1252                         goto out_exit_policy;
1253                 }
1254         }
1255
1256         /*
1257          * Sometimes boot loaders set CPU frequency to a value outside of
1258          * frequency table present with cpufreq core. In such cases CPU might be
1259          * unstable if it has to run on that frequency for long duration of time
1260          * and so its better to set it to a frequency which is specified in
1261          * freq-table. This also makes cpufreq stats inconsistent as
1262          * cpufreq-stats would fail to register because current frequency of CPU
1263          * isn't found in freq-table.
1264          *
1265          * Because we don't want this change to effect boot process badly, we go
1266          * for the next freq which is >= policy->cur ('cur' must be set by now,
1267          * otherwise we will end up setting freq to lowest of the table as 'cur'
1268          * is initialized to zero).
1269          *
1270          * We are passing target-freq as "policy->cur - 1" otherwise
1271          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1272          * equal to target-freq.
1273          */
1274         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1275             && has_target()) {
1276                 /* Are we running at unknown frequency ? */
1277                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1278                 if (ret == -EINVAL) {
1279                         /* Warn user and fix it */
1280                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1281                                 __func__, policy->cpu, policy->cur);
1282                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1283                                 CPUFREQ_RELATION_L);
1284
1285                         /*
1286                          * Reaching here after boot in a few seconds may not
1287                          * mean that system will remain stable at "unknown"
1288                          * frequency for longer duration. Hence, a BUG_ON().
1289                          */
1290                         BUG_ON(ret);
1291                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1292                                 __func__, policy->cpu, policy->cur);
1293                 }
1294         }
1295
1296         if (new_policy) {
1297                 ret = cpufreq_add_dev_interface(policy);
1298                 if (ret)
1299                         goto out_exit_policy;
1300
1301                 cpufreq_stats_create_table(policy);
1302
1303                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1304                 list_add(&policy->policy_list, &cpufreq_policy_list);
1305                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1306         }
1307
1308         ret = cpufreq_init_policy(policy);
1309         if (ret) {
1310                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1311                        __func__, cpu, ret);
1312                 /* cpufreq_policy_free() will notify based on this */
1313                 new_policy = false;
1314                 goto out_exit_policy;
1315         }
1316
1317         up_write(&policy->rwsem);
1318
1319         kobject_uevent(&policy->kobj, KOBJ_ADD);
1320
1321         /* Callback for handling stuff after policy is ready */
1322         if (cpufreq_driver->ready)
1323                 cpufreq_driver->ready(policy);
1324
1325         pr_debug("initialization complete\n");
1326
1327         return 0;
1328
1329 out_exit_policy:
1330         up_write(&policy->rwsem);
1331
1332         if (cpufreq_driver->exit)
1333                 cpufreq_driver->exit(policy);
1334
1335         for_each_cpu(j, policy->real_cpus)
1336                 remove_cpu_dev_symlink(policy, get_cpu_device(j));
1337
1338 out_free_policy:
1339         cpufreq_policy_free(policy);
1340         return ret;
1341 }
1342
1343 /**
1344  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1345  * @dev: CPU device.
1346  * @sif: Subsystem interface structure pointer (not used)
1347  */
1348 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1349 {
1350         struct cpufreq_policy *policy;
1351         unsigned cpu = dev->id;
1352         int ret;
1353
1354         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1355
1356         if (cpu_online(cpu)) {
1357                 ret = cpufreq_online(cpu);
1358                 if (ret)
1359                         return ret;
1360         }
1361
1362         /* Create sysfs link on CPU registration */
1363         policy = per_cpu(cpufreq_cpu_data, cpu);
1364         if (policy)
1365                 add_cpu_dev_symlink(policy, cpu);
1366
1367         return 0;
1368 }
1369
1370 static int cpufreq_offline(unsigned int cpu)
1371 {
1372         struct cpufreq_policy *policy;
1373         int ret;
1374
1375         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1376
1377         policy = cpufreq_cpu_get_raw(cpu);
1378         if (!policy) {
1379                 pr_debug("%s: No cpu_data found\n", __func__);
1380                 return 0;
1381         }
1382
1383         down_write(&policy->rwsem);
1384         if (has_target())
1385                 cpufreq_stop_governor(policy);
1386
1387         cpumask_clear_cpu(cpu, policy->cpus);
1388
1389         if (policy_is_inactive(policy)) {
1390                 if (has_target())
1391                         strncpy(policy->last_governor, policy->governor->name,
1392                                 CPUFREQ_NAME_LEN);
1393                 else
1394                         policy->last_policy = policy->policy;
1395         } else if (cpu == policy->cpu) {
1396                 /* Nominate new CPU */
1397                 policy->cpu = cpumask_any(policy->cpus);
1398         }
1399
1400         /* Start governor again for active policy */
1401         if (!policy_is_inactive(policy)) {
1402                 if (has_target()) {
1403                         ret = cpufreq_start_governor(policy);
1404                         if (ret)
1405                                 pr_err("%s: Failed to start governor\n", __func__);
1406                 }
1407
1408                 goto unlock;
1409         }
1410
1411         if (cpufreq_driver->stop_cpu)
1412                 cpufreq_driver->stop_cpu(policy);
1413
1414         if (has_target())
1415                 cpufreq_exit_governor(policy);
1416
1417         /*
1418          * Perform the ->exit() even during light-weight tear-down,
1419          * since this is a core component, and is essential for the
1420          * subsequent light-weight ->init() to succeed.
1421          */
1422         if (cpufreq_driver->exit) {
1423                 cpufreq_driver->exit(policy);
1424                 policy->freq_table = NULL;
1425         }
1426
1427 unlock:
1428         up_write(&policy->rwsem);
1429         return 0;
1430 }
1431
1432 /**
1433  * cpufreq_remove_dev - remove a CPU device
1434  *
1435  * Removes the cpufreq interface for a CPU device.
1436  */
1437 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1438 {
1439         unsigned int cpu = dev->id;
1440         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1441
1442         if (!policy)
1443                 return;
1444
1445         if (cpu_online(cpu))
1446                 cpufreq_offline(cpu);
1447
1448         cpumask_clear_cpu(cpu, policy->real_cpus);
1449         remove_cpu_dev_symlink(policy, dev);
1450
1451         if (cpumask_empty(policy->real_cpus))
1452                 cpufreq_policy_free(policy);
1453 }
1454
1455 /**
1456  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1457  *      in deep trouble.
1458  *      @policy: policy managing CPUs
1459  *      @new_freq: CPU frequency the CPU actually runs at
1460  *
1461  *      We adjust to current frequency first, and need to clean up later.
1462  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1463  */
1464 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1465                                 unsigned int new_freq)
1466 {
1467         struct cpufreq_freqs freqs;
1468
1469         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1470                  policy->cur, new_freq);
1471
1472         freqs.old = policy->cur;
1473         freqs.new = new_freq;
1474
1475         cpufreq_freq_transition_begin(policy, &freqs);
1476         cpufreq_freq_transition_end(policy, &freqs, 0);
1477 }
1478
1479 /**
1480  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1481  * @cpu: CPU number
1482  *
1483  * This is the last known freq, without actually getting it from the driver.
1484  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1485  */
1486 unsigned int cpufreq_quick_get(unsigned int cpu)
1487 {
1488         struct cpufreq_policy *policy;
1489         unsigned int ret_freq = 0;
1490         unsigned long flags;
1491
1492         read_lock_irqsave(&cpufreq_driver_lock, flags);
1493
1494         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1495                 ret_freq = cpufreq_driver->get(cpu);
1496                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1497                 return ret_freq;
1498         }
1499
1500         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1501
1502         policy = cpufreq_cpu_get(cpu);
1503         if (policy) {
1504                 ret_freq = policy->cur;
1505                 cpufreq_cpu_put(policy);
1506         }
1507
1508         return ret_freq;
1509 }
1510 EXPORT_SYMBOL(cpufreq_quick_get);
1511
1512 /**
1513  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1514  * @cpu: CPU number
1515  *
1516  * Just return the max possible frequency for a given CPU.
1517  */
1518 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1519 {
1520         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1521         unsigned int ret_freq = 0;
1522
1523         if (policy) {
1524                 ret_freq = policy->max;
1525                 cpufreq_cpu_put(policy);
1526         }
1527
1528         return ret_freq;
1529 }
1530 EXPORT_SYMBOL(cpufreq_quick_get_max);
1531
1532 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1533 {
1534         unsigned int ret_freq = 0;
1535
1536         if (!cpufreq_driver->get)
1537                 return ret_freq;
1538
1539         ret_freq = cpufreq_driver->get(policy->cpu);
1540
1541         /*
1542          * Updating inactive policies is invalid, so avoid doing that.  Also
1543          * if fast frequency switching is used with the given policy, the check
1544          * against policy->cur is pointless, so skip it in that case too.
1545          */
1546         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1547                 return ret_freq;
1548
1549         if (ret_freq && policy->cur &&
1550                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1551                 /* verify no discrepancy between actual and
1552                                         saved value exists */
1553                 if (unlikely(ret_freq != policy->cur)) {
1554                         cpufreq_out_of_sync(policy, ret_freq);
1555                         schedule_work(&policy->update);
1556                 }
1557         }
1558
1559         return ret_freq;
1560 }
1561
1562 /**
1563  * cpufreq_get - get the current CPU frequency (in kHz)
1564  * @cpu: CPU number
1565  *
1566  * Get the CPU current (static) CPU frequency
1567  */
1568 unsigned int cpufreq_get(unsigned int cpu)
1569 {
1570         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1571         unsigned int ret_freq = 0;
1572
1573         if (policy) {
1574                 down_read(&policy->rwsem);
1575
1576                 if (!policy_is_inactive(policy))
1577                         ret_freq = __cpufreq_get(policy);
1578
1579                 up_read(&policy->rwsem);
1580
1581                 cpufreq_cpu_put(policy);
1582         }
1583
1584         return ret_freq;
1585 }
1586 EXPORT_SYMBOL(cpufreq_get);
1587
1588 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1589 {
1590         unsigned int new_freq;
1591
1592         new_freq = cpufreq_driver->get(policy->cpu);
1593         if (!new_freq)
1594                 return 0;
1595
1596         if (!policy->cur) {
1597                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1598                 policy->cur = new_freq;
1599         } else if (policy->cur != new_freq && has_target()) {
1600                 cpufreq_out_of_sync(policy, new_freq);
1601         }
1602
1603         return new_freq;
1604 }
1605
1606 static struct subsys_interface cpufreq_interface = {
1607         .name           = "cpufreq",
1608         .subsys         = &cpu_subsys,
1609         .add_dev        = cpufreq_add_dev,
1610         .remove_dev     = cpufreq_remove_dev,
1611 };
1612
1613 /*
1614  * In case platform wants some specific frequency to be configured
1615  * during suspend..
1616  */
1617 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1618 {
1619         int ret;
1620
1621         if (!policy->suspend_freq) {
1622                 pr_debug("%s: suspend_freq not defined\n", __func__);
1623                 return 0;
1624         }
1625
1626         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1627                         policy->suspend_freq);
1628
1629         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1630                         CPUFREQ_RELATION_H);
1631         if (ret)
1632                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1633                                 __func__, policy->suspend_freq, ret);
1634
1635         return ret;
1636 }
1637 EXPORT_SYMBOL(cpufreq_generic_suspend);
1638
1639 /**
1640  * cpufreq_suspend() - Suspend CPUFreq governors
1641  *
1642  * Called during system wide Suspend/Hibernate cycles for suspending governors
1643  * as some platforms can't change frequency after this point in suspend cycle.
1644  * Because some of the devices (like: i2c, regulators, etc) they use for
1645  * changing frequency are suspended quickly after this point.
1646  */
1647 void cpufreq_suspend(void)
1648 {
1649         struct cpufreq_policy *policy;
1650
1651         if (!cpufreq_driver)
1652                 return;
1653
1654         if (!has_target() && !cpufreq_driver->suspend)
1655                 goto suspend;
1656
1657         pr_debug("%s: Suspending Governors\n", __func__);
1658
1659         for_each_active_policy(policy) {
1660                 if (has_target()) {
1661                         down_write(&policy->rwsem);
1662                         cpufreq_stop_governor(policy);
1663                         up_write(&policy->rwsem);
1664                 }
1665
1666                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1667                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1668                                 policy);
1669         }
1670
1671 suspend:
1672         cpufreq_suspended = true;
1673 }
1674
1675 /**
1676  * cpufreq_resume() - Resume CPUFreq governors
1677  *
1678  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1679  * are suspended with cpufreq_suspend().
1680  */
1681 void cpufreq_resume(void)
1682 {
1683         struct cpufreq_policy *policy;
1684         int ret;
1685
1686         if (!cpufreq_driver)
1687                 return;
1688
1689         cpufreq_suspended = false;
1690
1691         if (!has_target() && !cpufreq_driver->resume)
1692                 return;
1693
1694         pr_debug("%s: Resuming Governors\n", __func__);
1695
1696         for_each_active_policy(policy) {
1697                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1698                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1699                                 policy);
1700                 } else if (has_target()) {
1701                         down_write(&policy->rwsem);
1702                         ret = cpufreq_start_governor(policy);
1703                         up_write(&policy->rwsem);
1704
1705                         if (ret)
1706                                 pr_err("%s: Failed to start governor for policy: %p\n",
1707                                        __func__, policy);
1708                 }
1709         }
1710 }
1711
1712 /**
1713  *      cpufreq_get_current_driver - return current driver's name
1714  *
1715  *      Return the name string of the currently loaded cpufreq driver
1716  *      or NULL, if none.
1717  */
1718 const char *cpufreq_get_current_driver(void)
1719 {
1720         if (cpufreq_driver)
1721                 return cpufreq_driver->name;
1722
1723         return NULL;
1724 }
1725 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1726
1727 /**
1728  *      cpufreq_get_driver_data - return current driver data
1729  *
1730  *      Return the private data of the currently loaded cpufreq
1731  *      driver, or NULL if no cpufreq driver is loaded.
1732  */
1733 void *cpufreq_get_driver_data(void)
1734 {
1735         if (cpufreq_driver)
1736                 return cpufreq_driver->driver_data;
1737
1738         return NULL;
1739 }
1740 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1741
1742 /*********************************************************************
1743  *                     NOTIFIER LISTS INTERFACE                      *
1744  *********************************************************************/
1745
1746 /**
1747  *      cpufreq_register_notifier - register a driver with cpufreq
1748  *      @nb: notifier function to register
1749  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1750  *
1751  *      Add a driver to one of two lists: either a list of drivers that
1752  *      are notified about clock rate changes (once before and once after
1753  *      the transition), or a list of drivers that are notified about
1754  *      changes in cpufreq policy.
1755  *
1756  *      This function may sleep, and has the same return conditions as
1757  *      blocking_notifier_chain_register.
1758  */
1759 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1760 {
1761         int ret;
1762
1763         if (cpufreq_disabled())
1764                 return -EINVAL;
1765
1766         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1767
1768         switch (list) {
1769         case CPUFREQ_TRANSITION_NOTIFIER:
1770                 mutex_lock(&cpufreq_fast_switch_lock);
1771
1772                 if (cpufreq_fast_switch_count > 0) {
1773                         mutex_unlock(&cpufreq_fast_switch_lock);
1774                         return -EBUSY;
1775                 }
1776                 ret = srcu_notifier_chain_register(
1777                                 &cpufreq_transition_notifier_list, nb);
1778                 if (!ret)
1779                         cpufreq_fast_switch_count--;
1780
1781                 mutex_unlock(&cpufreq_fast_switch_lock);
1782                 break;
1783         case CPUFREQ_POLICY_NOTIFIER:
1784                 ret = blocking_notifier_chain_register(
1785                                 &cpufreq_policy_notifier_list, nb);
1786                 break;
1787         default:
1788                 ret = -EINVAL;
1789         }
1790
1791         return ret;
1792 }
1793 EXPORT_SYMBOL(cpufreq_register_notifier);
1794
1795 /**
1796  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1797  *      @nb: notifier block to be unregistered
1798  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1799  *
1800  *      Remove a driver from the CPU frequency notifier list.
1801  *
1802  *      This function may sleep, and has the same return conditions as
1803  *      blocking_notifier_chain_unregister.
1804  */
1805 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1806 {
1807         int ret;
1808
1809         if (cpufreq_disabled())
1810                 return -EINVAL;
1811
1812         switch (list) {
1813         case CPUFREQ_TRANSITION_NOTIFIER:
1814                 mutex_lock(&cpufreq_fast_switch_lock);
1815
1816                 ret = srcu_notifier_chain_unregister(
1817                                 &cpufreq_transition_notifier_list, nb);
1818                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1819                         cpufreq_fast_switch_count++;
1820
1821                 mutex_unlock(&cpufreq_fast_switch_lock);
1822                 break;
1823         case CPUFREQ_POLICY_NOTIFIER:
1824                 ret = blocking_notifier_chain_unregister(
1825                                 &cpufreq_policy_notifier_list, nb);
1826                 break;
1827         default:
1828                 ret = -EINVAL;
1829         }
1830
1831         return ret;
1832 }
1833 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1834
1835
1836 /*********************************************************************
1837  *                              GOVERNORS                            *
1838  *********************************************************************/
1839
1840 /**
1841  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1842  * @policy: cpufreq policy to switch the frequency for.
1843  * @target_freq: New frequency to set (may be approximate).
1844  *
1845  * Carry out a fast frequency switch without sleeping.
1846  *
1847  * The driver's ->fast_switch() callback invoked by this function must be
1848  * suitable for being called from within RCU-sched read-side critical sections
1849  * and it is expected to select the minimum available frequency greater than or
1850  * equal to @target_freq (CPUFREQ_RELATION_L).
1851  *
1852  * This function must not be called if policy->fast_switch_enabled is unset.
1853  *
1854  * Governors calling this function must guarantee that it will never be invoked
1855  * twice in parallel for the same policy and that it will never be called in
1856  * parallel with either ->target() or ->target_index() for the same policy.
1857  *
1858  * Returns the actual frequency set for the CPU.
1859  *
1860  * If 0 is returned by the driver's ->fast_switch() callback to indicate an
1861  * error condition, the hardware configuration must be preserved.
1862  */
1863 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1864                                         unsigned int target_freq)
1865 {
1866         target_freq = clamp_val(target_freq, policy->min, policy->max);
1867
1868         return cpufreq_driver->fast_switch(policy, target_freq);
1869 }
1870 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1871
1872 /* Must set freqs->new to intermediate frequency */
1873 static int __target_intermediate(struct cpufreq_policy *policy,
1874                                  struct cpufreq_freqs *freqs, int index)
1875 {
1876         int ret;
1877
1878         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1879
1880         /* We don't need to switch to intermediate freq */
1881         if (!freqs->new)
1882                 return 0;
1883
1884         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1885                  __func__, policy->cpu, freqs->old, freqs->new);
1886
1887         cpufreq_freq_transition_begin(policy, freqs);
1888         ret = cpufreq_driver->target_intermediate(policy, index);
1889         cpufreq_freq_transition_end(policy, freqs, ret);
1890
1891         if (ret)
1892                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1893                        __func__, ret);
1894
1895         return ret;
1896 }
1897
1898 static int __target_index(struct cpufreq_policy *policy, int index)
1899 {
1900         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1901         unsigned int intermediate_freq = 0;
1902         unsigned int newfreq = policy->freq_table[index].frequency;
1903         int retval = -EINVAL;
1904         bool notify;
1905
1906         if (newfreq == policy->cur)
1907                 return 0;
1908
1909         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1910         if (notify) {
1911                 /* Handle switching to intermediate frequency */
1912                 if (cpufreq_driver->get_intermediate) {
1913                         retval = __target_intermediate(policy, &freqs, index);
1914                         if (retval)
1915                                 return retval;
1916
1917                         intermediate_freq = freqs.new;
1918                         /* Set old freq to intermediate */
1919                         if (intermediate_freq)
1920                                 freqs.old = freqs.new;
1921                 }
1922
1923                 freqs.new = newfreq;
1924                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1925                          __func__, policy->cpu, freqs.old, freqs.new);
1926
1927                 cpufreq_freq_transition_begin(policy, &freqs);
1928         }
1929
1930         retval = cpufreq_driver->target_index(policy, index);
1931         if (retval)
1932                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1933                        retval);
1934
1935         if (notify) {
1936                 cpufreq_freq_transition_end(policy, &freqs, retval);
1937
1938                 /*
1939                  * Failed after setting to intermediate freq? Driver should have
1940                  * reverted back to initial frequency and so should we. Check
1941                  * here for intermediate_freq instead of get_intermediate, in
1942                  * case we haven't switched to intermediate freq at all.
1943                  */
1944                 if (unlikely(retval && intermediate_freq)) {
1945                         freqs.old = intermediate_freq;
1946                         freqs.new = policy->restore_freq;
1947                         cpufreq_freq_transition_begin(policy, &freqs);
1948                         cpufreq_freq_transition_end(policy, &freqs, 0);
1949                 }
1950         }
1951
1952         return retval;
1953 }
1954
1955 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1956                             unsigned int target_freq,
1957                             unsigned int relation)
1958 {
1959         unsigned int old_target_freq = target_freq;
1960         int index;
1961
1962         if (cpufreq_disabled())
1963                 return -ENODEV;
1964
1965         /* Make sure that target_freq is within supported range */
1966         target_freq = clamp_val(target_freq, policy->min, policy->max);
1967
1968         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1969                  policy->cpu, target_freq, relation, old_target_freq);
1970
1971         /*
1972          * This might look like a redundant call as we are checking it again
1973          * after finding index. But it is left intentionally for cases where
1974          * exactly same freq is called again and so we can save on few function
1975          * calls.
1976          */
1977         if (target_freq == policy->cur)
1978                 return 0;
1979
1980         /* Save last value to restore later on errors */
1981         policy->restore_freq = policy->cur;
1982
1983         if (cpufreq_driver->target)
1984                 return cpufreq_driver->target(policy, target_freq, relation);
1985
1986         if (!cpufreq_driver->target_index)
1987                 return -EINVAL;
1988
1989         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1990
1991         return __target_index(policy, index);
1992 }
1993 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1994
1995 int cpufreq_driver_target(struct cpufreq_policy *policy,
1996                           unsigned int target_freq,
1997                           unsigned int relation)
1998 {
1999         int ret = -EINVAL;
2000
2001         down_write(&policy->rwsem);
2002
2003         ret = __cpufreq_driver_target(policy, target_freq, relation);
2004
2005         up_write(&policy->rwsem);
2006
2007         return ret;
2008 }
2009 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2010
2011 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2012 {
2013         return NULL;
2014 }
2015
2016 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2017 {
2018         int ret;
2019
2020         /* Don't start any governor operations if we are entering suspend */
2021         if (cpufreq_suspended)
2022                 return 0;
2023         /*
2024          * Governor might not be initiated here if ACPI _PPC changed
2025          * notification happened, so check it.
2026          */
2027         if (!policy->governor)
2028                 return -EINVAL;
2029
2030         /* Platform doesn't want dynamic frequency switching ? */
2031         if (policy->governor->dynamic_switching &&
2032             cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2033                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2034
2035                 if (gov) {
2036                         pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2037                                 policy->governor->name, gov->name);
2038                         policy->governor = gov;
2039                 } else {
2040                         return -EINVAL;
2041                 }
2042         }
2043
2044         if (!try_module_get(policy->governor->owner))
2045                 return -EINVAL;
2046
2047         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2048
2049         if (policy->governor->init) {
2050                 ret = policy->governor->init(policy);
2051                 if (ret) {
2052                         module_put(policy->governor->owner);
2053                         return ret;
2054                 }
2055         }
2056
2057         return 0;
2058 }
2059
2060 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2061 {
2062         if (cpufreq_suspended || !policy->governor)
2063                 return;
2064
2065         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2066
2067         if (policy->governor->exit)
2068                 policy->governor->exit(policy);
2069
2070         module_put(policy->governor->owner);
2071 }
2072
2073 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2074 {
2075         int ret;
2076
2077         if (cpufreq_suspended)
2078                 return 0;
2079
2080         if (!policy->governor)
2081                 return -EINVAL;
2082
2083         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2084
2085         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2086                 cpufreq_update_current_freq(policy);
2087
2088         if (policy->governor->start) {
2089                 ret = policy->governor->start(policy);
2090                 if (ret)
2091                         return ret;
2092         }
2093
2094         if (policy->governor->limits)
2095                 policy->governor->limits(policy);
2096
2097         return 0;
2098 }
2099
2100 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2101 {
2102         if (cpufreq_suspended || !policy->governor)
2103                 return;
2104
2105         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2106
2107         if (policy->governor->stop)
2108                 policy->governor->stop(policy);
2109 }
2110
2111 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2112 {
2113         if (cpufreq_suspended || !policy->governor)
2114                 return;
2115
2116         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2117
2118         if (policy->governor->limits)
2119                 policy->governor->limits(policy);
2120 }
2121
2122 int cpufreq_register_governor(struct cpufreq_governor *governor)
2123 {
2124         int err;
2125
2126         if (!governor)
2127                 return -EINVAL;
2128
2129         if (cpufreq_disabled())
2130                 return -ENODEV;
2131
2132         mutex_lock(&cpufreq_governor_mutex);
2133
2134         err = -EBUSY;
2135         if (!find_governor(governor->name)) {
2136                 err = 0;
2137                 list_add(&governor->governor_list, &cpufreq_governor_list);
2138         }
2139
2140         mutex_unlock(&cpufreq_governor_mutex);
2141         return err;
2142 }
2143 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2144
2145 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2146 {
2147         struct cpufreq_policy *policy;
2148         unsigned long flags;
2149
2150         if (!governor)
2151                 return;
2152
2153         if (cpufreq_disabled())
2154                 return;
2155
2156         /* clear last_governor for all inactive policies */
2157         read_lock_irqsave(&cpufreq_driver_lock, flags);
2158         for_each_inactive_policy(policy) {
2159                 if (!strcmp(policy->last_governor, governor->name)) {
2160                         policy->governor = NULL;
2161                         strcpy(policy->last_governor, "\0");
2162                 }
2163         }
2164         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2165
2166         mutex_lock(&cpufreq_governor_mutex);
2167         list_del(&governor->governor_list);
2168         mutex_unlock(&cpufreq_governor_mutex);
2169 }
2170 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2171
2172
2173 /*********************************************************************
2174  *                          POLICY INTERFACE                         *
2175  *********************************************************************/
2176
2177 /**
2178  * cpufreq_get_policy - get the current cpufreq_policy
2179  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2180  *      is written
2181  *
2182  * Reads the current cpufreq policy.
2183  */
2184 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2185 {
2186         struct cpufreq_policy *cpu_policy;
2187         if (!policy)
2188                 return -EINVAL;
2189
2190         cpu_policy = cpufreq_cpu_get(cpu);
2191         if (!cpu_policy)
2192                 return -EINVAL;
2193
2194         memcpy(policy, cpu_policy, sizeof(*policy));
2195
2196         cpufreq_cpu_put(cpu_policy);
2197         return 0;
2198 }
2199 EXPORT_SYMBOL(cpufreq_get_policy);
2200
2201 /*
2202  * policy : current policy.
2203  * new_policy: policy to be set.
2204  */
2205 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2206                                 struct cpufreq_policy *new_policy)
2207 {
2208         struct cpufreq_governor *old_gov;
2209         int ret;
2210
2211         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2212                  new_policy->cpu, new_policy->min, new_policy->max);
2213
2214         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2215
2216         /*
2217         * This check works well when we store new min/max freq attributes,
2218         * because new_policy is a copy of policy with one field updated.
2219         */
2220         if (new_policy->min > new_policy->max)
2221                 return -EINVAL;
2222
2223         /* verify the cpu speed can be set within this limit */
2224         ret = cpufreq_driver->verify(new_policy);
2225         if (ret)
2226                 return ret;
2227
2228         /* adjust if necessary - all reasons */
2229         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2230                         CPUFREQ_ADJUST, new_policy);
2231
2232         /*
2233          * verify the cpu speed can be set within this limit, which might be
2234          * different to the first one
2235          */
2236         ret = cpufreq_driver->verify(new_policy);
2237         if (ret)
2238                 return ret;
2239
2240         /* notification of the new policy */
2241         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2242                         CPUFREQ_NOTIFY, new_policy);
2243
2244         policy->min = new_policy->min;
2245         policy->max = new_policy->max;
2246
2247         policy->cached_target_freq = UINT_MAX;
2248
2249         pr_debug("new min and max freqs are %u - %u kHz\n",
2250                  policy->min, policy->max);
2251
2252         if (cpufreq_driver->setpolicy) {
2253                 policy->policy = new_policy->policy;
2254                 pr_debug("setting range\n");
2255                 return cpufreq_driver->setpolicy(new_policy);
2256         }
2257
2258         if (new_policy->governor == policy->governor) {
2259                 pr_debug("cpufreq: governor limits update\n");
2260                 cpufreq_governor_limits(policy);
2261                 return 0;
2262         }
2263
2264         pr_debug("governor switch\n");
2265
2266         /* save old, working values */
2267         old_gov = policy->governor;
2268         /* end old governor */
2269         if (old_gov) {
2270                 cpufreq_stop_governor(policy);
2271                 cpufreq_exit_governor(policy);
2272         }
2273
2274         /* start new governor */
2275         policy->governor = new_policy->governor;
2276         ret = cpufreq_init_governor(policy);
2277         if (!ret) {
2278                 ret = cpufreq_start_governor(policy);
2279                 if (!ret) {
2280                         pr_debug("cpufreq: governor change\n");
2281                         return 0;
2282                 }
2283                 cpufreq_exit_governor(policy);
2284         }
2285
2286         /* new governor failed, so re-start old one */
2287         pr_debug("starting governor %s failed\n", policy->governor->name);
2288         if (old_gov) {
2289                 policy->governor = old_gov;
2290                 if (cpufreq_init_governor(policy))
2291                         policy->governor = NULL;
2292                 else
2293                         cpufreq_start_governor(policy);
2294         }
2295
2296         return ret;
2297 }
2298
2299 /**
2300  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2301  *      @cpu: CPU which shall be re-evaluated
2302  *
2303  *      Useful for policy notifiers which have different necessities
2304  *      at different times.
2305  */
2306 void cpufreq_update_policy(unsigned int cpu)
2307 {
2308         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2309         struct cpufreq_policy new_policy;
2310
2311         if (!policy)
2312                 return;
2313
2314         down_write(&policy->rwsem);
2315
2316         if (policy_is_inactive(policy))
2317                 goto unlock;
2318
2319         pr_debug("updating policy for CPU %u\n", cpu);
2320         memcpy(&new_policy, policy, sizeof(*policy));
2321         new_policy.min = policy->user_policy.min;
2322         new_policy.max = policy->user_policy.max;
2323
2324         /*
2325          * BIOS might change freq behind our back
2326          * -> ask driver for current freq and notify governors about a change
2327          */
2328         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2329                 if (cpufreq_suspended)
2330                         goto unlock;
2331
2332                 new_policy.cur = cpufreq_update_current_freq(policy);
2333                 if (WARN_ON(!new_policy.cur))
2334                         goto unlock;
2335         }
2336
2337         cpufreq_set_policy(policy, &new_policy);
2338
2339 unlock:
2340         up_write(&policy->rwsem);
2341
2342         cpufreq_cpu_put(policy);
2343 }
2344 EXPORT_SYMBOL(cpufreq_update_policy);
2345
2346 /*********************************************************************
2347  *               BOOST                                               *
2348  *********************************************************************/
2349 static int cpufreq_boost_set_sw(int state)
2350 {
2351         struct cpufreq_policy *policy;
2352         int ret = -EINVAL;
2353
2354         for_each_active_policy(policy) {
2355                 if (!policy->freq_table)
2356                         continue;
2357
2358                 ret = cpufreq_frequency_table_cpuinfo(policy,
2359                                                       policy->freq_table);
2360                 if (ret) {
2361                         pr_err("%s: Policy frequency update failed\n",
2362                                __func__);
2363                         break;
2364                 }
2365
2366                 down_write(&policy->rwsem);
2367                 policy->user_policy.max = policy->max;
2368                 cpufreq_governor_limits(policy);
2369                 up_write(&policy->rwsem);
2370         }
2371
2372         return ret;
2373 }
2374
2375 int cpufreq_boost_trigger_state(int state)
2376 {
2377         unsigned long flags;
2378         int ret = 0;
2379
2380         if (cpufreq_driver->boost_enabled == state)
2381                 return 0;
2382
2383         write_lock_irqsave(&cpufreq_driver_lock, flags);
2384         cpufreq_driver->boost_enabled = state;
2385         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2386
2387         ret = cpufreq_driver->set_boost(state);
2388         if (ret) {
2389                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2390                 cpufreq_driver->boost_enabled = !state;
2391                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2392
2393                 pr_err("%s: Cannot %s BOOST\n",
2394                        __func__, state ? "enable" : "disable");
2395         }
2396
2397         return ret;
2398 }
2399
2400 static bool cpufreq_boost_supported(void)
2401 {
2402         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2403 }
2404
2405 static int create_boost_sysfs_file(void)
2406 {
2407         int ret;
2408
2409         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2410         if (ret)
2411                 pr_err("%s: cannot register global BOOST sysfs file\n",
2412                        __func__);
2413
2414         return ret;
2415 }
2416
2417 static void remove_boost_sysfs_file(void)
2418 {
2419         if (cpufreq_boost_supported())
2420                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2421 }
2422
2423 int cpufreq_enable_boost_support(void)
2424 {
2425         if (!cpufreq_driver)
2426                 return -EINVAL;
2427
2428         if (cpufreq_boost_supported())
2429                 return 0;
2430
2431         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2432
2433         /* This will get removed on driver unregister */
2434         return create_boost_sysfs_file();
2435 }
2436 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2437
2438 int cpufreq_boost_enabled(void)
2439 {
2440         return cpufreq_driver->boost_enabled;
2441 }
2442 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2443
2444 /*********************************************************************
2445  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2446  *********************************************************************/
2447 static enum cpuhp_state hp_online;
2448
2449 static int cpuhp_cpufreq_online(unsigned int cpu)
2450 {
2451         cpufreq_online(cpu);
2452
2453         return 0;
2454 }
2455
2456 static int cpuhp_cpufreq_offline(unsigned int cpu)
2457 {
2458         cpufreq_offline(cpu);
2459
2460         return 0;
2461 }
2462
2463 /**
2464  * cpufreq_register_driver - register a CPU Frequency driver
2465  * @driver_data: A struct cpufreq_driver containing the values#
2466  * submitted by the CPU Frequency driver.
2467  *
2468  * Registers a CPU Frequency driver to this core code. This code
2469  * returns zero on success, -EEXIST when another driver got here first
2470  * (and isn't unregistered in the meantime).
2471  *
2472  */
2473 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2474 {
2475         unsigned long flags;
2476         int ret;
2477
2478         if (cpufreq_disabled())
2479                 return -ENODEV;
2480
2481         if (!driver_data || !driver_data->verify || !driver_data->init ||
2482             !(driver_data->setpolicy || driver_data->target_index ||
2483                     driver_data->target) ||
2484              (driver_data->setpolicy && (driver_data->target_index ||
2485                     driver_data->target)) ||
2486              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2487                 return -EINVAL;
2488
2489         pr_debug("trying to register driver %s\n", driver_data->name);
2490
2491         /* Protect against concurrent CPU online/offline. */
2492         cpus_read_lock();
2493
2494         write_lock_irqsave(&cpufreq_driver_lock, flags);
2495         if (cpufreq_driver) {
2496                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2497                 ret = -EEXIST;
2498                 goto out;
2499         }
2500         cpufreq_driver = driver_data;
2501         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2502
2503         if (driver_data->setpolicy)
2504                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2505
2506         if (cpufreq_boost_supported()) {
2507                 ret = create_boost_sysfs_file();
2508                 if (ret)
2509                         goto err_null_driver;
2510         }
2511
2512         ret = subsys_interface_register(&cpufreq_interface);
2513         if (ret)
2514                 goto err_boost_unreg;
2515
2516         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2517             list_empty(&cpufreq_policy_list)) {
2518                 /* if all ->init() calls failed, unregister */
2519                 ret = -ENODEV;
2520                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2521                          driver_data->name);
2522                 goto err_if_unreg;
2523         }
2524
2525         ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2526                                                    "cpufreq:online",
2527                                                    cpuhp_cpufreq_online,
2528                                                    cpuhp_cpufreq_offline);
2529         if (ret < 0)
2530                 goto err_if_unreg;
2531         hp_online = ret;
2532         ret = 0;
2533
2534         pr_debug("driver %s up and running\n", driver_data->name);
2535         goto out;
2536
2537 err_if_unreg:
2538         subsys_interface_unregister(&cpufreq_interface);
2539 err_boost_unreg:
2540         remove_boost_sysfs_file();
2541 err_null_driver:
2542         write_lock_irqsave(&cpufreq_driver_lock, flags);
2543         cpufreq_driver = NULL;
2544         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2545 out:
2546         cpus_read_unlock();
2547         return ret;
2548 }
2549 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2550
2551 /**
2552  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2553  *
2554  * Unregister the current CPUFreq driver. Only call this if you have
2555  * the right to do so, i.e. if you have succeeded in initialising before!
2556  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2557  * currently not initialised.
2558  */
2559 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2560 {
2561         unsigned long flags;
2562
2563         if (!cpufreq_driver || (driver != cpufreq_driver))
2564                 return -EINVAL;
2565
2566         pr_debug("unregistering driver %s\n", driver->name);
2567
2568         /* Protect against concurrent cpu hotplug */
2569         cpus_read_lock();
2570         subsys_interface_unregister(&cpufreq_interface);
2571         remove_boost_sysfs_file();
2572         cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2573
2574         write_lock_irqsave(&cpufreq_driver_lock, flags);
2575
2576         cpufreq_driver = NULL;
2577
2578         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2579         cpus_read_unlock();
2580
2581         return 0;
2582 }
2583 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2584
2585 /*
2586  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2587  * or mutexes when secondary CPUs are halted.
2588  */
2589 static struct syscore_ops cpufreq_syscore_ops = {
2590         .shutdown = cpufreq_suspend,
2591 };
2592
2593 struct kobject *cpufreq_global_kobject;
2594 EXPORT_SYMBOL(cpufreq_global_kobject);
2595
2596 static int __init cpufreq_core_init(void)
2597 {
2598         if (cpufreq_disabled())
2599                 return -ENODEV;
2600
2601         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2602         BUG_ON(!cpufreq_global_kobject);
2603
2604         register_syscore_ops(&cpufreq_syscore_ops);
2605
2606         return 0;
2607 }
2608 module_param(off, int, 0444);
2609 core_initcall(cpufreq_core_init);