GNU Linux-libre 4.9.314-gnu1
[releases.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 /*
165  * This is a generic cpufreq init() routine which can be used by cpufreq
166  * drivers of SMP systems. It will do following:
167  * - validate & show freq table passed
168  * - set policies transition latency
169  * - policy->cpus with all possible CPUs
170  */
171 int cpufreq_generic_init(struct cpufreq_policy *policy,
172                 struct cpufreq_frequency_table *table,
173                 unsigned int transition_latency)
174 {
175         int ret;
176
177         ret = cpufreq_table_validate_and_show(policy, table);
178         if (ret) {
179                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
180                 return ret;
181         }
182
183         policy->cpuinfo.transition_latency = transition_latency;
184
185         /*
186          * The driver only supports the SMP configuration where all processors
187          * share the clock and voltage and clock.
188          */
189         cpumask_setall(policy->cpus);
190
191         return 0;
192 }
193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
194
195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
196 {
197         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
198
199         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
200 }
201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
202
203 unsigned int cpufreq_generic_get(unsigned int cpu)
204 {
205         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
206
207         if (!policy || IS_ERR(policy->clk)) {
208                 pr_err("%s: No %s associated to cpu: %d\n",
209                        __func__, policy ? "clk" : "policy", cpu);
210                 return 0;
211         }
212
213         return clk_get_rate(policy->clk) / 1000;
214 }
215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
216
217 /**
218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
219  *
220  * @cpu: cpu to find policy for.
221  *
222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
223  * It also increments the kobject reference count to mark it busy and so would
224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
226  * freed as that depends on the kobj count.
227  *
228  * Return: A valid policy on success, otherwise NULL on failure.
229  */
230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
231 {
232         struct cpufreq_policy *policy = NULL;
233         unsigned long flags;
234
235         if (WARN_ON(cpu >= nr_cpu_ids))
236                 return NULL;
237
238         /* get the cpufreq driver */
239         read_lock_irqsave(&cpufreq_driver_lock, flags);
240
241         if (cpufreq_driver) {
242                 /* get the CPU */
243                 policy = cpufreq_cpu_get_raw(cpu);
244                 if (policy)
245                         kobject_get(&policy->kobj);
246         }
247
248         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
249
250         return policy;
251 }
252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
253
254 /**
255  * cpufreq_cpu_put: Decrements the usage count of a policy
256  *
257  * @policy: policy earlier returned by cpufreq_cpu_get().
258  *
259  * This decrements the kobject reference count incremented earlier by calling
260  * cpufreq_cpu_get().
261  */
262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
263 {
264         kobject_put(&policy->kobj);
265 }
266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
267
268 /*********************************************************************
269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
270  *********************************************************************/
271
272 /**
273  * adjust_jiffies - adjust the system "loops_per_jiffy"
274  *
275  * This function alters the system "loops_per_jiffy" for the clock
276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
277  * systems as each CPU might be scaled differently. So, use the arch
278  * per-CPU loops_per_jiffy value wherever possible.
279  */
280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
281 {
282 #ifndef CONFIG_SMP
283         static unsigned long l_p_j_ref;
284         static unsigned int l_p_j_ref_freq;
285
286         if (ci->flags & CPUFREQ_CONST_LOOPS)
287                 return;
288
289         if (!l_p_j_ref_freq) {
290                 l_p_j_ref = loops_per_jiffy;
291                 l_p_j_ref_freq = ci->old;
292                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
293                          l_p_j_ref, l_p_j_ref_freq);
294         }
295         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297                                                                 ci->new);
298                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
299                          loops_per_jiffy, ci->new);
300         }
301 #endif
302 }
303
304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
305                 struct cpufreq_freqs *freqs, unsigned int state)
306 {
307         BUG_ON(irqs_disabled());
308
309         if (cpufreq_disabled())
310                 return;
311
312         freqs->flags = cpufreq_driver->flags;
313         pr_debug("notification %u of frequency transition to %u kHz\n",
314                  state, freqs->new);
315
316         switch (state) {
317
318         case CPUFREQ_PRECHANGE:
319                 /* detect if the driver reported a value as "old frequency"
320                  * which is not equal to what the cpufreq core thinks is
321                  * "old frequency".
322                  */
323                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
324                         if ((policy) && (policy->cpu == freqs->cpu) &&
325                             (policy->cur) && (policy->cur != freqs->old)) {
326                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
327                                          freqs->old, policy->cur);
328                                 freqs->old = policy->cur;
329                         }
330                 }
331                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
332                                 CPUFREQ_PRECHANGE, freqs);
333                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
334                 break;
335
336         case CPUFREQ_POSTCHANGE:
337                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
338                 pr_debug("FREQ: %lu - CPU: %lu\n",
339                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
340                 trace_cpu_frequency(freqs->new, freqs->cpu);
341                 cpufreq_stats_record_transition(policy, freqs->new);
342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
343                                 CPUFREQ_POSTCHANGE, freqs);
344                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
345                         policy->cur = freqs->new;
346                 break;
347         }
348 }
349
350 /**
351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
352  * on frequency transition.
353  *
354  * This function calls the transition notifiers and the "adjust_jiffies"
355  * function. It is called twice on all CPU frequency changes that have
356  * external effects.
357  */
358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
359                 struct cpufreq_freqs *freqs, unsigned int state)
360 {
361         for_each_cpu(freqs->cpu, policy->cpus)
362                 __cpufreq_notify_transition(policy, freqs, state);
363 }
364
365 /* Do post notifications when there are chances that transition has failed */
366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
367                 struct cpufreq_freqs *freqs, int transition_failed)
368 {
369         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
370         if (!transition_failed)
371                 return;
372
373         swap(freqs->old, freqs->new);
374         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
376 }
377
378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
379                 struct cpufreq_freqs *freqs)
380 {
381
382         /*
383          * Catch double invocations of _begin() which lead to self-deadlock.
384          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
385          * doesn't invoke _begin() on their behalf, and hence the chances of
386          * double invocations are very low. Moreover, there are scenarios
387          * where these checks can emit false-positive warnings in these
388          * drivers; so we avoid that by skipping them altogether.
389          */
390         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
391                                 && current == policy->transition_task);
392
393 wait:
394         wait_event(policy->transition_wait, !policy->transition_ongoing);
395
396         spin_lock(&policy->transition_lock);
397
398         if (unlikely(policy->transition_ongoing)) {
399                 spin_unlock(&policy->transition_lock);
400                 goto wait;
401         }
402
403         policy->transition_ongoing = true;
404         policy->transition_task = current;
405
406         spin_unlock(&policy->transition_lock);
407
408         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
409 }
410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
411
412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
413                 struct cpufreq_freqs *freqs, int transition_failed)
414 {
415         if (unlikely(WARN_ON(!policy->transition_ongoing)))
416                 return;
417
418         cpufreq_notify_post_transition(policy, freqs, transition_failed);
419
420         policy->transition_ongoing = false;
421         policy->transition_task = NULL;
422
423         wake_up(&policy->transition_wait);
424 }
425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
426
427 /*
428  * Fast frequency switching status count.  Positive means "enabled", negative
429  * means "disabled" and 0 means "not decided yet".
430  */
431 static int cpufreq_fast_switch_count;
432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
433
434 static void cpufreq_list_transition_notifiers(void)
435 {
436         struct notifier_block *nb;
437
438         pr_info("Registered transition notifiers:\n");
439
440         mutex_lock(&cpufreq_transition_notifier_list.mutex);
441
442         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
443                 pr_info("%pF\n", nb->notifier_call);
444
445         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
446 }
447
448 /**
449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
450  * @policy: cpufreq policy to enable fast frequency switching for.
451  *
452  * Try to enable fast frequency switching for @policy.
453  *
454  * The attempt will fail if there is at least one transition notifier registered
455  * at this point, as fast frequency switching is quite fundamentally at odds
456  * with transition notifiers.  Thus if successful, it will make registration of
457  * transition notifiers fail going forward.
458  */
459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
460 {
461         lockdep_assert_held(&policy->rwsem);
462
463         if (!policy->fast_switch_possible)
464                 return;
465
466         mutex_lock(&cpufreq_fast_switch_lock);
467         if (cpufreq_fast_switch_count >= 0) {
468                 cpufreq_fast_switch_count++;
469                 policy->fast_switch_enabled = true;
470         } else {
471                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
472                         policy->cpu);
473                 cpufreq_list_transition_notifiers();
474         }
475         mutex_unlock(&cpufreq_fast_switch_lock);
476 }
477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
478
479 /**
480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
481  * @policy: cpufreq policy to disable fast frequency switching for.
482  */
483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
484 {
485         mutex_lock(&cpufreq_fast_switch_lock);
486         if (policy->fast_switch_enabled) {
487                 policy->fast_switch_enabled = false;
488                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
489                         cpufreq_fast_switch_count--;
490         }
491         mutex_unlock(&cpufreq_fast_switch_lock);
492 }
493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
494
495 /**
496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
497  * one.
498  * @target_freq: target frequency to resolve.
499  *
500  * The target to driver frequency mapping is cached in the policy.
501  *
502  * Return: Lowest driver-supported frequency greater than or equal to the
503  * given target_freq, subject to policy (min/max) and driver limitations.
504  */
505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
506                                          unsigned int target_freq)
507 {
508         target_freq = clamp_val(target_freq, policy->min, policy->max);
509         policy->cached_target_freq = target_freq;
510
511         if (cpufreq_driver->target_index) {
512                 int idx;
513
514                 idx = cpufreq_frequency_table_target(policy, target_freq,
515                                                      CPUFREQ_RELATION_L);
516                 policy->cached_resolved_idx = idx;
517                 return policy->freq_table[idx].frequency;
518         }
519
520         if (cpufreq_driver->resolve_freq)
521                 return cpufreq_driver->resolve_freq(policy, target_freq);
522
523         return target_freq;
524 }
525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
526
527 /*********************************************************************
528  *                          SYSFS INTERFACE                          *
529  *********************************************************************/
530 static ssize_t show_boost(struct kobject *kobj,
531                           struct kobj_attribute *attr, char *buf)
532 {
533         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
534 }
535
536 static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
537                            const char *buf, size_t count)
538 {
539         int ret, enable;
540
541         ret = sscanf(buf, "%d", &enable);
542         if (ret != 1 || enable < 0 || enable > 1)
543                 return -EINVAL;
544
545         if (cpufreq_boost_trigger_state(enable)) {
546                 pr_err("%s: Cannot %s BOOST!\n",
547                        __func__, enable ? "enable" : "disable");
548                 return -EINVAL;
549         }
550
551         pr_debug("%s: cpufreq BOOST %s\n",
552                  __func__, enable ? "enabled" : "disabled");
553
554         return count;
555 }
556 define_one_global_rw(boost);
557
558 static struct cpufreq_governor *find_governor(const char *str_governor)
559 {
560         struct cpufreq_governor *t;
561
562         for_each_governor(t)
563                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
564                         return t;
565
566         return NULL;
567 }
568
569 /**
570  * cpufreq_parse_governor - parse a governor string
571  */
572 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
573                                 struct cpufreq_governor **governor)
574 {
575         int err = -EINVAL;
576
577         if (cpufreq_driver->setpolicy) {
578                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
579                         *policy = CPUFREQ_POLICY_PERFORMANCE;
580                         err = 0;
581                 } else if (!strncasecmp(str_governor, "powersave",
582                                                 CPUFREQ_NAME_LEN)) {
583                         *policy = CPUFREQ_POLICY_POWERSAVE;
584                         err = 0;
585                 }
586         } else {
587                 struct cpufreq_governor *t;
588
589                 mutex_lock(&cpufreq_governor_mutex);
590
591                 t = find_governor(str_governor);
592
593                 if (t == NULL) {
594                         int ret;
595
596                         mutex_unlock(&cpufreq_governor_mutex);
597                         ret = request_module("cpufreq_%s", str_governor);
598                         mutex_lock(&cpufreq_governor_mutex);
599
600                         if (ret == 0)
601                                 t = find_governor(str_governor);
602                 }
603
604                 if (t != NULL) {
605                         *governor = t;
606                         err = 0;
607                 }
608
609                 mutex_unlock(&cpufreq_governor_mutex);
610         }
611         return err;
612 }
613
614 /**
615  * cpufreq_per_cpu_attr_read() / show_##file_name() -
616  * print out cpufreq information
617  *
618  * Write out information from cpufreq_driver->policy[cpu]; object must be
619  * "unsigned int".
620  */
621
622 #define show_one(file_name, object)                     \
623 static ssize_t show_##file_name                         \
624 (struct cpufreq_policy *policy, char *buf)              \
625 {                                                       \
626         return sprintf(buf, "%u\n", policy->object);    \
627 }
628
629 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
630 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
631 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
632 show_one(scaling_min_freq, min);
633 show_one(scaling_max_freq, max);
634
635 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
636 {
637         ssize_t ret;
638
639         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
640                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
641         else
642                 ret = sprintf(buf, "%u\n", policy->cur);
643         return ret;
644 }
645
646 static int cpufreq_set_policy(struct cpufreq_policy *policy,
647                                 struct cpufreq_policy *new_policy);
648
649 /**
650  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
651  */
652 #define store_one(file_name, object)                    \
653 static ssize_t store_##file_name                                        \
654 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
655 {                                                                       \
656         int ret, temp;                                                  \
657         struct cpufreq_policy new_policy;                               \
658                                                                         \
659         memcpy(&new_policy, policy, sizeof(*policy));                   \
660         new_policy.min = policy->user_policy.min;                       \
661         new_policy.max = policy->user_policy.max;                       \
662                                                                         \
663         ret = sscanf(buf, "%u", &new_policy.object);                    \
664         if (ret != 1)                                                   \
665                 return -EINVAL;                                         \
666                                                                         \
667         temp = new_policy.object;                                       \
668         ret = cpufreq_set_policy(policy, &new_policy);          \
669         if (!ret)                                                       \
670                 policy->user_policy.object = temp;                      \
671                                                                         \
672         return ret ? ret : count;                                       \
673 }
674
675 store_one(scaling_min_freq, min);
676 store_one(scaling_max_freq, max);
677
678 /**
679  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
680  */
681 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
682                                         char *buf)
683 {
684         unsigned int cur_freq = __cpufreq_get(policy);
685
686         if (cur_freq)
687                 return sprintf(buf, "%u\n", cur_freq);
688
689         return sprintf(buf, "<unknown>\n");
690 }
691
692 /**
693  * show_scaling_governor - show the current policy for the specified CPU
694  */
695 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
696 {
697         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
698                 return sprintf(buf, "powersave\n");
699         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
700                 return sprintf(buf, "performance\n");
701         else if (policy->governor)
702                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
703                                 policy->governor->name);
704         return -EINVAL;
705 }
706
707 /**
708  * store_scaling_governor - store policy for the specified CPU
709  */
710 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
711                                         const char *buf, size_t count)
712 {
713         int ret;
714         char    str_governor[16];
715         struct cpufreq_policy new_policy;
716
717         memcpy(&new_policy, policy, sizeof(*policy));
718
719         ret = sscanf(buf, "%15s", str_governor);
720         if (ret != 1)
721                 return -EINVAL;
722
723         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
724                                                 &new_policy.governor))
725                 return -EINVAL;
726
727         ret = cpufreq_set_policy(policy, &new_policy);
728         return ret ? ret : count;
729 }
730
731 /**
732  * show_scaling_driver - show the cpufreq driver currently loaded
733  */
734 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
735 {
736         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
737 }
738
739 /**
740  * show_scaling_available_governors - show the available CPUfreq governors
741  */
742 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
743                                                 char *buf)
744 {
745         ssize_t i = 0;
746         struct cpufreq_governor *t;
747
748         if (!has_target()) {
749                 i += sprintf(buf, "performance powersave");
750                 goto out;
751         }
752
753         for_each_governor(t) {
754                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
755                     - (CPUFREQ_NAME_LEN + 2)))
756                         goto out;
757                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
758         }
759 out:
760         i += sprintf(&buf[i], "\n");
761         return i;
762 }
763
764 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
765 {
766         ssize_t i = 0;
767         unsigned int cpu;
768
769         for_each_cpu(cpu, mask) {
770                 if (i)
771                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
772                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
773                 if (i >= (PAGE_SIZE - 5))
774                         break;
775         }
776         i += sprintf(&buf[i], "\n");
777         return i;
778 }
779 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
780
781 /**
782  * show_related_cpus - show the CPUs affected by each transition even if
783  * hw coordination is in use
784  */
785 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
786 {
787         return cpufreq_show_cpus(policy->related_cpus, buf);
788 }
789
790 /**
791  * show_affected_cpus - show the CPUs affected by each transition
792  */
793 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
794 {
795         return cpufreq_show_cpus(policy->cpus, buf);
796 }
797
798 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
799                                         const char *buf, size_t count)
800 {
801         unsigned int freq = 0;
802         unsigned int ret;
803
804         if (!policy->governor || !policy->governor->store_setspeed)
805                 return -EINVAL;
806
807         ret = sscanf(buf, "%u", &freq);
808         if (ret != 1)
809                 return -EINVAL;
810
811         policy->governor->store_setspeed(policy, freq);
812
813         return count;
814 }
815
816 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
817 {
818         if (!policy->governor || !policy->governor->show_setspeed)
819                 return sprintf(buf, "<unsupported>\n");
820
821         return policy->governor->show_setspeed(policy, buf);
822 }
823
824 /**
825  * show_bios_limit - show the current cpufreq HW/BIOS limitation
826  */
827 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
828 {
829         unsigned int limit;
830         int ret;
831         if (cpufreq_driver->bios_limit) {
832                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
833                 if (!ret)
834                         return sprintf(buf, "%u\n", limit);
835         }
836         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
837 }
838
839 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
840 cpufreq_freq_attr_ro(cpuinfo_min_freq);
841 cpufreq_freq_attr_ro(cpuinfo_max_freq);
842 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
843 cpufreq_freq_attr_ro(scaling_available_governors);
844 cpufreq_freq_attr_ro(scaling_driver);
845 cpufreq_freq_attr_ro(scaling_cur_freq);
846 cpufreq_freq_attr_ro(bios_limit);
847 cpufreq_freq_attr_ro(related_cpus);
848 cpufreq_freq_attr_ro(affected_cpus);
849 cpufreq_freq_attr_rw(scaling_min_freq);
850 cpufreq_freq_attr_rw(scaling_max_freq);
851 cpufreq_freq_attr_rw(scaling_governor);
852 cpufreq_freq_attr_rw(scaling_setspeed);
853
854 static struct attribute *default_attrs[] = {
855         &cpuinfo_min_freq.attr,
856         &cpuinfo_max_freq.attr,
857         &cpuinfo_transition_latency.attr,
858         &scaling_min_freq.attr,
859         &scaling_max_freq.attr,
860         &affected_cpus.attr,
861         &related_cpus.attr,
862         &scaling_governor.attr,
863         &scaling_driver.attr,
864         &scaling_available_governors.attr,
865         &scaling_setspeed.attr,
866         NULL
867 };
868
869 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
870 #define to_attr(a) container_of(a, struct freq_attr, attr)
871
872 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
873 {
874         struct cpufreq_policy *policy = to_policy(kobj);
875         struct freq_attr *fattr = to_attr(attr);
876         ssize_t ret;
877
878         if (!fattr->show)
879                 return -EIO;
880
881         down_read(&policy->rwsem);
882         ret = fattr->show(policy, buf);
883         up_read(&policy->rwsem);
884
885         return ret;
886 }
887
888 static ssize_t store(struct kobject *kobj, struct attribute *attr,
889                      const char *buf, size_t count)
890 {
891         struct cpufreq_policy *policy = to_policy(kobj);
892         struct freq_attr *fattr = to_attr(attr);
893         ssize_t ret = -EINVAL;
894
895         if (!fattr->store)
896                 return -EIO;
897
898         get_online_cpus();
899
900         if (cpu_online(policy->cpu)) {
901                 down_write(&policy->rwsem);
902                 ret = fattr->store(policy, buf, count);
903                 up_write(&policy->rwsem);
904         }
905
906         put_online_cpus();
907
908         return ret;
909 }
910
911 static void cpufreq_sysfs_release(struct kobject *kobj)
912 {
913         struct cpufreq_policy *policy = to_policy(kobj);
914         pr_debug("last reference is dropped\n");
915         complete(&policy->kobj_unregister);
916 }
917
918 static const struct sysfs_ops sysfs_ops = {
919         .show   = show,
920         .store  = store,
921 };
922
923 static struct kobj_type ktype_cpufreq = {
924         .sysfs_ops      = &sysfs_ops,
925         .default_attrs  = default_attrs,
926         .release        = cpufreq_sysfs_release,
927 };
928
929 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
930 {
931         struct device *dev = get_cpu_device(cpu);
932
933         if (!dev)
934                 return;
935
936         if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
937                 return;
938
939         dev_dbg(dev, "%s: Adding symlink\n", __func__);
940         if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
941                 dev_err(dev, "cpufreq symlink creation failed\n");
942 }
943
944 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
945                                    struct device *dev)
946 {
947         dev_dbg(dev, "%s: Removing symlink\n", __func__);
948         sysfs_remove_link(&dev->kobj, "cpufreq");
949 }
950
951 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
952 {
953         struct freq_attr **drv_attr;
954         int ret = 0;
955
956         /* set up files for this cpu device */
957         drv_attr = cpufreq_driver->attr;
958         while (drv_attr && *drv_attr) {
959                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
960                 if (ret)
961                         return ret;
962                 drv_attr++;
963         }
964         if (cpufreq_driver->get) {
965                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
966                 if (ret)
967                         return ret;
968         }
969
970         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
971         if (ret)
972                 return ret;
973
974         if (cpufreq_driver->bios_limit) {
975                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
976                 if (ret)
977                         return ret;
978         }
979
980         return 0;
981 }
982
983 __weak struct cpufreq_governor *cpufreq_default_governor(void)
984 {
985         return NULL;
986 }
987
988 static int cpufreq_init_policy(struct cpufreq_policy *policy)
989 {
990         struct cpufreq_governor *gov = NULL;
991         struct cpufreq_policy new_policy;
992
993         memcpy(&new_policy, policy, sizeof(*policy));
994
995         /* Update governor of new_policy to the governor used before hotplug */
996         gov = find_governor(policy->last_governor);
997         if (gov) {
998                 pr_debug("Restoring governor %s for cpu %d\n",
999                                 policy->governor->name, policy->cpu);
1000         } else {
1001                 gov = cpufreq_default_governor();
1002                 if (!gov)
1003                         return -ENODATA;
1004         }
1005
1006         new_policy.governor = gov;
1007
1008         /* Use the default policy if there is no last_policy. */
1009         if (cpufreq_driver->setpolicy) {
1010                 if (policy->last_policy)
1011                         new_policy.policy = policy->last_policy;
1012                 else
1013                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1014                                                NULL);
1015         }
1016         /* set default policy */
1017         return cpufreq_set_policy(policy, &new_policy);
1018 }
1019
1020 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1021 {
1022         int ret = 0;
1023
1024         /* Has this CPU been taken care of already? */
1025         if (cpumask_test_cpu(cpu, policy->cpus))
1026                 return 0;
1027
1028         down_write(&policy->rwsem);
1029         if (has_target())
1030                 cpufreq_stop_governor(policy);
1031
1032         cpumask_set_cpu(cpu, policy->cpus);
1033
1034         if (has_target()) {
1035                 ret = cpufreq_start_governor(policy);
1036                 if (ret)
1037                         pr_err("%s: Failed to start governor\n", __func__);
1038         }
1039         up_write(&policy->rwsem);
1040         return ret;
1041 }
1042
1043 static void handle_update(struct work_struct *work)
1044 {
1045         struct cpufreq_policy *policy =
1046                 container_of(work, struct cpufreq_policy, update);
1047         unsigned int cpu = policy->cpu;
1048         pr_debug("handle_update for cpu %u called\n", cpu);
1049         cpufreq_update_policy(cpu);
1050 }
1051
1052 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1053 {
1054         struct cpufreq_policy *policy;
1055         int ret;
1056
1057         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1058         if (!policy)
1059                 return NULL;
1060
1061         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1062                 goto err_free_policy;
1063
1064         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1065                 goto err_free_cpumask;
1066
1067         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1068                 goto err_free_rcpumask;
1069
1070         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1071                                    cpufreq_global_kobject, "policy%u", cpu);
1072         if (ret) {
1073                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1074                 kobject_put(&policy->kobj);
1075                 goto err_free_real_cpus;
1076         }
1077
1078         INIT_LIST_HEAD(&policy->policy_list);
1079         init_rwsem(&policy->rwsem);
1080         spin_lock_init(&policy->transition_lock);
1081         init_waitqueue_head(&policy->transition_wait);
1082         init_completion(&policy->kobj_unregister);
1083         INIT_WORK(&policy->update, handle_update);
1084
1085         policy->cpu = cpu;
1086         return policy;
1087
1088 err_free_real_cpus:
1089         free_cpumask_var(policy->real_cpus);
1090 err_free_rcpumask:
1091         free_cpumask_var(policy->related_cpus);
1092 err_free_cpumask:
1093         free_cpumask_var(policy->cpus);
1094 err_free_policy:
1095         kfree(policy);
1096
1097         return NULL;
1098 }
1099
1100 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1101 {
1102         struct kobject *kobj;
1103         struct completion *cmp;
1104
1105         if (notify)
1106                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1107                                              CPUFREQ_REMOVE_POLICY, policy);
1108
1109         down_write(&policy->rwsem);
1110         cpufreq_stats_free_table(policy);
1111         kobj = &policy->kobj;
1112         cmp = &policy->kobj_unregister;
1113         up_write(&policy->rwsem);
1114         kobject_put(kobj);
1115
1116         /*
1117          * We need to make sure that the underlying kobj is
1118          * actually not referenced anymore by anybody before we
1119          * proceed with unloading.
1120          */
1121         pr_debug("waiting for dropping of refcount\n");
1122         wait_for_completion(cmp);
1123         pr_debug("wait complete\n");
1124 }
1125
1126 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1127 {
1128         unsigned long flags;
1129         int cpu;
1130
1131         /* Remove policy from list */
1132         write_lock_irqsave(&cpufreq_driver_lock, flags);
1133         list_del(&policy->policy_list);
1134
1135         for_each_cpu(cpu, policy->related_cpus)
1136                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1137         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1138
1139         cpufreq_policy_put_kobj(policy, notify);
1140         free_cpumask_var(policy->real_cpus);
1141         free_cpumask_var(policy->related_cpus);
1142         free_cpumask_var(policy->cpus);
1143         kfree(policy);
1144 }
1145
1146 static int cpufreq_online(unsigned int cpu)
1147 {
1148         struct cpufreq_policy *policy;
1149         bool new_policy;
1150         unsigned long flags;
1151         unsigned int j;
1152         int ret;
1153
1154         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1155
1156         /* Check if this CPU already has a policy to manage it */
1157         policy = per_cpu(cpufreq_cpu_data, cpu);
1158         if (policy) {
1159                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1160                 if (!policy_is_inactive(policy))
1161                         return cpufreq_add_policy_cpu(policy, cpu);
1162
1163                 /* This is the only online CPU for the policy.  Start over. */
1164                 new_policy = false;
1165                 down_write(&policy->rwsem);
1166                 policy->cpu = cpu;
1167                 policy->governor = NULL;
1168                 up_write(&policy->rwsem);
1169         } else {
1170                 new_policy = true;
1171                 policy = cpufreq_policy_alloc(cpu);
1172                 if (!policy)
1173                         return -ENOMEM;
1174         }
1175
1176         cpumask_copy(policy->cpus, cpumask_of(cpu));
1177
1178         /* call driver. From then on the cpufreq must be able
1179          * to accept all calls to ->verify and ->setpolicy for this CPU
1180          */
1181         ret = cpufreq_driver->init(policy);
1182         if (ret) {
1183                 pr_debug("initialization failed\n");
1184                 goto out_free_policy;
1185         }
1186
1187         down_write(&policy->rwsem);
1188
1189         if (new_policy) {
1190                 /* related_cpus should at least include policy->cpus. */
1191                 cpumask_copy(policy->related_cpus, policy->cpus);
1192         }
1193
1194         /*
1195          * affected cpus must always be the one, which are online. We aren't
1196          * managing offline cpus here.
1197          */
1198         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1199
1200         if (new_policy) {
1201                 policy->user_policy.min = policy->min;
1202                 policy->user_policy.max = policy->max;
1203
1204                 for_each_cpu(j, policy->related_cpus) {
1205                         per_cpu(cpufreq_cpu_data, j) = policy;
1206                         add_cpu_dev_symlink(policy, j);
1207                 }
1208         } else {
1209                 policy->min = policy->user_policy.min;
1210                 policy->max = policy->user_policy.max;
1211         }
1212
1213         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1214                 policy->cur = cpufreq_driver->get(policy->cpu);
1215                 if (!policy->cur) {
1216                         pr_err("%s: ->get() failed\n", __func__);
1217                         goto out_exit_policy;
1218                 }
1219         }
1220
1221         /*
1222          * Sometimes boot loaders set CPU frequency to a value outside of
1223          * frequency table present with cpufreq core. In such cases CPU might be
1224          * unstable if it has to run on that frequency for long duration of time
1225          * and so its better to set it to a frequency which is specified in
1226          * freq-table. This also makes cpufreq stats inconsistent as
1227          * cpufreq-stats would fail to register because current frequency of CPU
1228          * isn't found in freq-table.
1229          *
1230          * Because we don't want this change to effect boot process badly, we go
1231          * for the next freq which is >= policy->cur ('cur' must be set by now,
1232          * otherwise we will end up setting freq to lowest of the table as 'cur'
1233          * is initialized to zero).
1234          *
1235          * We are passing target-freq as "policy->cur - 1" otherwise
1236          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1237          * equal to target-freq.
1238          */
1239         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1240             && has_target()) {
1241                 /* Are we running at unknown frequency ? */
1242                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1243                 if (ret == -EINVAL) {
1244                         /* Warn user and fix it */
1245                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1246                                 __func__, policy->cpu, policy->cur);
1247                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1248                                 CPUFREQ_RELATION_L);
1249
1250                         /*
1251                          * Reaching here after boot in a few seconds may not
1252                          * mean that system will remain stable at "unknown"
1253                          * frequency for longer duration. Hence, a BUG_ON().
1254                          */
1255                         BUG_ON(ret);
1256                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1257                                 __func__, policy->cpu, policy->cur);
1258                 }
1259         }
1260
1261         if (new_policy) {
1262                 ret = cpufreq_add_dev_interface(policy);
1263                 if (ret)
1264                         goto out_exit_policy;
1265
1266                 cpufreq_stats_create_table(policy);
1267                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1268                                 CPUFREQ_CREATE_POLICY, policy);
1269
1270                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1271                 list_add(&policy->policy_list, &cpufreq_policy_list);
1272                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1273         }
1274
1275         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1276                                      CPUFREQ_START, policy);
1277
1278         ret = cpufreq_init_policy(policy);
1279         if (ret) {
1280                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1281                        __func__, cpu, ret);
1282                 /* cpufreq_policy_free() will notify based on this */
1283                 new_policy = false;
1284                 goto out_exit_policy;
1285         }
1286
1287         up_write(&policy->rwsem);
1288
1289         kobject_uevent(&policy->kobj, KOBJ_ADD);
1290
1291         /* Callback for handling stuff after policy is ready */
1292         if (cpufreq_driver->ready)
1293                 cpufreq_driver->ready(policy);
1294
1295         pr_debug("initialization complete\n");
1296
1297         return 0;
1298
1299 out_exit_policy:
1300         for_each_cpu(j, policy->real_cpus)
1301                 remove_cpu_dev_symlink(policy, get_cpu_device(j));
1302
1303         up_write(&policy->rwsem);
1304
1305         if (cpufreq_driver->exit)
1306                 cpufreq_driver->exit(policy);
1307
1308 out_free_policy:
1309         cpufreq_policy_free(policy, !new_policy);
1310         return ret;
1311 }
1312
1313 /**
1314  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1315  * @dev: CPU device.
1316  * @sif: Subsystem interface structure pointer (not used)
1317  */
1318 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1319 {
1320         struct cpufreq_policy *policy;
1321         unsigned cpu = dev->id;
1322         int ret;
1323
1324         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1325
1326         if (cpu_online(cpu)) {
1327                 ret = cpufreq_online(cpu);
1328                 if (ret)
1329                         return ret;
1330         }
1331
1332         /* Create sysfs link on CPU registration */
1333         policy = per_cpu(cpufreq_cpu_data, cpu);
1334         if (policy)
1335                 add_cpu_dev_symlink(policy, cpu);
1336
1337         return 0;
1338 }
1339
1340 static int cpufreq_offline(unsigned int cpu)
1341 {
1342         struct cpufreq_policy *policy;
1343         int ret;
1344
1345         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1346
1347         policy = cpufreq_cpu_get_raw(cpu);
1348         if (!policy) {
1349                 pr_debug("%s: No cpu_data found\n", __func__);
1350                 return 0;
1351         }
1352
1353         down_write(&policy->rwsem);
1354         if (has_target())
1355                 cpufreq_stop_governor(policy);
1356
1357         cpumask_clear_cpu(cpu, policy->cpus);
1358
1359         if (policy_is_inactive(policy)) {
1360                 if (has_target())
1361                         strncpy(policy->last_governor, policy->governor->name,
1362                                 CPUFREQ_NAME_LEN);
1363                 else
1364                         policy->last_policy = policy->policy;
1365         } else if (cpu == policy->cpu) {
1366                 /* Nominate new CPU */
1367                 policy->cpu = cpumask_any(policy->cpus);
1368         }
1369
1370         /* Start governor again for active policy */
1371         if (!policy_is_inactive(policy)) {
1372                 if (has_target()) {
1373                         ret = cpufreq_start_governor(policy);
1374                         if (ret)
1375                                 pr_err("%s: Failed to start governor\n", __func__);
1376                 }
1377
1378                 goto unlock;
1379         }
1380
1381         if (cpufreq_driver->stop_cpu)
1382                 cpufreq_driver->stop_cpu(policy);
1383
1384         if (has_target())
1385                 cpufreq_exit_governor(policy);
1386
1387         /*
1388          * Perform the ->exit() even during light-weight tear-down,
1389          * since this is a core component, and is essential for the
1390          * subsequent light-weight ->init() to succeed.
1391          */
1392         if (cpufreq_driver->exit) {
1393                 cpufreq_driver->exit(policy);
1394                 policy->freq_table = NULL;
1395         }
1396
1397 unlock:
1398         up_write(&policy->rwsem);
1399         return 0;
1400 }
1401
1402 /**
1403  * cpufreq_remove_dev - remove a CPU device
1404  *
1405  * Removes the cpufreq interface for a CPU device.
1406  */
1407 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1408 {
1409         unsigned int cpu = dev->id;
1410         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1411
1412         if (!policy)
1413                 return;
1414
1415         if (cpu_online(cpu))
1416                 cpufreq_offline(cpu);
1417
1418         cpumask_clear_cpu(cpu, policy->real_cpus);
1419         remove_cpu_dev_symlink(policy, dev);
1420
1421         if (cpumask_empty(policy->real_cpus))
1422                 cpufreq_policy_free(policy, true);
1423 }
1424
1425 /**
1426  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1427  *      in deep trouble.
1428  *      @policy: policy managing CPUs
1429  *      @new_freq: CPU frequency the CPU actually runs at
1430  *
1431  *      We adjust to current frequency first, and need to clean up later.
1432  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1433  */
1434 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1435                                 unsigned int new_freq)
1436 {
1437         struct cpufreq_freqs freqs;
1438
1439         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1440                  policy->cur, new_freq);
1441
1442         freqs.old = policy->cur;
1443         freqs.new = new_freq;
1444
1445         cpufreq_freq_transition_begin(policy, &freqs);
1446         cpufreq_freq_transition_end(policy, &freqs, 0);
1447 }
1448
1449 /**
1450  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1451  * @cpu: CPU number
1452  *
1453  * This is the last known freq, without actually getting it from the driver.
1454  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1455  */
1456 unsigned int cpufreq_quick_get(unsigned int cpu)
1457 {
1458         struct cpufreq_policy *policy;
1459         unsigned int ret_freq = 0;
1460         unsigned long flags;
1461
1462         read_lock_irqsave(&cpufreq_driver_lock, flags);
1463
1464         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1465                 ret_freq = cpufreq_driver->get(cpu);
1466                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1467                 return ret_freq;
1468         }
1469
1470         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1471
1472         policy = cpufreq_cpu_get(cpu);
1473         if (policy) {
1474                 ret_freq = policy->cur;
1475                 cpufreq_cpu_put(policy);
1476         }
1477
1478         return ret_freq;
1479 }
1480 EXPORT_SYMBOL(cpufreq_quick_get);
1481
1482 /**
1483  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1484  * @cpu: CPU number
1485  *
1486  * Just return the max possible frequency for a given CPU.
1487  */
1488 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1489 {
1490         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1491         unsigned int ret_freq = 0;
1492
1493         if (policy) {
1494                 ret_freq = policy->max;
1495                 cpufreq_cpu_put(policy);
1496         }
1497
1498         return ret_freq;
1499 }
1500 EXPORT_SYMBOL(cpufreq_quick_get_max);
1501
1502 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1503 {
1504         unsigned int ret_freq = 0;
1505
1506         if (unlikely(policy_is_inactive(policy)) || !cpufreq_driver->get)
1507                 return ret_freq;
1508
1509         ret_freq = cpufreq_driver->get(policy->cpu);
1510
1511         /*
1512          * If fast frequency switching is used with the given policy, the check
1513          * against policy->cur is pointless, so skip it in that case too.
1514          */
1515         if (policy->fast_switch_enabled)
1516                 return ret_freq;
1517
1518         if (ret_freq && policy->cur &&
1519                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1520                 /* verify no discrepancy between actual and
1521                                         saved value exists */
1522                 if (unlikely(ret_freq != policy->cur)) {
1523                         cpufreq_out_of_sync(policy, ret_freq);
1524                         schedule_work(&policy->update);
1525                 }
1526         }
1527
1528         return ret_freq;
1529 }
1530
1531 /**
1532  * cpufreq_get - get the current CPU frequency (in kHz)
1533  * @cpu: CPU number
1534  *
1535  * Get the CPU current (static) CPU frequency
1536  */
1537 unsigned int cpufreq_get(unsigned int cpu)
1538 {
1539         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1540         unsigned int ret_freq = 0;
1541
1542         if (policy) {
1543                 down_read(&policy->rwsem);
1544                 ret_freq = __cpufreq_get(policy);
1545                 up_read(&policy->rwsem);
1546
1547                 cpufreq_cpu_put(policy);
1548         }
1549
1550         return ret_freq;
1551 }
1552 EXPORT_SYMBOL(cpufreq_get);
1553
1554 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1555 {
1556         unsigned int new_freq;
1557
1558         new_freq = cpufreq_driver->get(policy->cpu);
1559         if (!new_freq)
1560                 return 0;
1561
1562         if (!policy->cur) {
1563                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1564                 policy->cur = new_freq;
1565         } else if (policy->cur != new_freq && has_target()) {
1566                 cpufreq_out_of_sync(policy, new_freq);
1567         }
1568
1569         return new_freq;
1570 }
1571
1572 static struct subsys_interface cpufreq_interface = {
1573         .name           = "cpufreq",
1574         .subsys         = &cpu_subsys,
1575         .add_dev        = cpufreq_add_dev,
1576         .remove_dev     = cpufreq_remove_dev,
1577 };
1578
1579 /*
1580  * In case platform wants some specific frequency to be configured
1581  * during suspend..
1582  */
1583 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1584 {
1585         int ret;
1586
1587         if (!policy->suspend_freq) {
1588                 pr_debug("%s: suspend_freq not defined\n", __func__);
1589                 return 0;
1590         }
1591
1592         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1593                         policy->suspend_freq);
1594
1595         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1596                         CPUFREQ_RELATION_H);
1597         if (ret)
1598                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1599                                 __func__, policy->suspend_freq, ret);
1600
1601         return ret;
1602 }
1603 EXPORT_SYMBOL(cpufreq_generic_suspend);
1604
1605 /**
1606  * cpufreq_suspend() - Suspend CPUFreq governors
1607  *
1608  * Called during system wide Suspend/Hibernate cycles for suspending governors
1609  * as some platforms can't change frequency after this point in suspend cycle.
1610  * Because some of the devices (like: i2c, regulators, etc) they use for
1611  * changing frequency are suspended quickly after this point.
1612  */
1613 void cpufreq_suspend(void)
1614 {
1615         struct cpufreq_policy *policy;
1616
1617         if (!cpufreq_driver)
1618                 return;
1619
1620         if (!has_target() && !cpufreq_driver->suspend)
1621                 goto suspend;
1622
1623         pr_debug("%s: Suspending Governors\n", __func__);
1624
1625         for_each_active_policy(policy) {
1626                 if (has_target()) {
1627                         down_write(&policy->rwsem);
1628                         cpufreq_stop_governor(policy);
1629                         up_write(&policy->rwsem);
1630                 }
1631
1632                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1633                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1634                                 policy);
1635         }
1636
1637 suspend:
1638         cpufreq_suspended = true;
1639 }
1640
1641 /**
1642  * cpufreq_resume() - Resume CPUFreq governors
1643  *
1644  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1645  * are suspended with cpufreq_suspend().
1646  */
1647 void cpufreq_resume(void)
1648 {
1649         struct cpufreq_policy *policy;
1650         int ret;
1651
1652         if (!cpufreq_driver)
1653                 return;
1654
1655         if (unlikely(!cpufreq_suspended))
1656                 return;
1657
1658         cpufreq_suspended = false;
1659
1660         if (!has_target() && !cpufreq_driver->resume)
1661                 return;
1662
1663         pr_debug("%s: Resuming Governors\n", __func__);
1664
1665         for_each_active_policy(policy) {
1666                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1667                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1668                                 policy);
1669                 } else if (has_target()) {
1670                         down_write(&policy->rwsem);
1671                         ret = cpufreq_start_governor(policy);
1672                         up_write(&policy->rwsem);
1673
1674                         if (ret)
1675                                 pr_err("%s: Failed to start governor for policy: %p\n",
1676                                        __func__, policy);
1677                 }
1678         }
1679 }
1680
1681 /**
1682  *      cpufreq_get_current_driver - return current driver's name
1683  *
1684  *      Return the name string of the currently loaded cpufreq driver
1685  *      or NULL, if none.
1686  */
1687 const char *cpufreq_get_current_driver(void)
1688 {
1689         if (cpufreq_driver)
1690                 return cpufreq_driver->name;
1691
1692         return NULL;
1693 }
1694 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1695
1696 /**
1697  *      cpufreq_get_driver_data - return current driver data
1698  *
1699  *      Return the private data of the currently loaded cpufreq
1700  *      driver, or NULL if no cpufreq driver is loaded.
1701  */
1702 void *cpufreq_get_driver_data(void)
1703 {
1704         if (cpufreq_driver)
1705                 return cpufreq_driver->driver_data;
1706
1707         return NULL;
1708 }
1709 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1710
1711 /*********************************************************************
1712  *                     NOTIFIER LISTS INTERFACE                      *
1713  *********************************************************************/
1714
1715 /**
1716  *      cpufreq_register_notifier - register a driver with cpufreq
1717  *      @nb: notifier function to register
1718  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1719  *
1720  *      Add a driver to one of two lists: either a list of drivers that
1721  *      are notified about clock rate changes (once before and once after
1722  *      the transition), or a list of drivers that are notified about
1723  *      changes in cpufreq policy.
1724  *
1725  *      This function may sleep, and has the same return conditions as
1726  *      blocking_notifier_chain_register.
1727  */
1728 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1729 {
1730         int ret;
1731
1732         if (cpufreq_disabled())
1733                 return -EINVAL;
1734
1735         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1736
1737         switch (list) {
1738         case CPUFREQ_TRANSITION_NOTIFIER:
1739                 mutex_lock(&cpufreq_fast_switch_lock);
1740
1741                 if (cpufreq_fast_switch_count > 0) {
1742                         mutex_unlock(&cpufreq_fast_switch_lock);
1743                         return -EBUSY;
1744                 }
1745                 ret = srcu_notifier_chain_register(
1746                                 &cpufreq_transition_notifier_list, nb);
1747                 if (!ret)
1748                         cpufreq_fast_switch_count--;
1749
1750                 mutex_unlock(&cpufreq_fast_switch_lock);
1751                 break;
1752         case CPUFREQ_POLICY_NOTIFIER:
1753                 ret = blocking_notifier_chain_register(
1754                                 &cpufreq_policy_notifier_list, nb);
1755                 break;
1756         default:
1757                 ret = -EINVAL;
1758         }
1759
1760         return ret;
1761 }
1762 EXPORT_SYMBOL(cpufreq_register_notifier);
1763
1764 /**
1765  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1766  *      @nb: notifier block to be unregistered
1767  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1768  *
1769  *      Remove a driver from the CPU frequency notifier list.
1770  *
1771  *      This function may sleep, and has the same return conditions as
1772  *      blocking_notifier_chain_unregister.
1773  */
1774 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1775 {
1776         int ret;
1777
1778         if (cpufreq_disabled())
1779                 return -EINVAL;
1780
1781         switch (list) {
1782         case CPUFREQ_TRANSITION_NOTIFIER:
1783                 mutex_lock(&cpufreq_fast_switch_lock);
1784
1785                 ret = srcu_notifier_chain_unregister(
1786                                 &cpufreq_transition_notifier_list, nb);
1787                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1788                         cpufreq_fast_switch_count++;
1789
1790                 mutex_unlock(&cpufreq_fast_switch_lock);
1791                 break;
1792         case CPUFREQ_POLICY_NOTIFIER:
1793                 ret = blocking_notifier_chain_unregister(
1794                                 &cpufreq_policy_notifier_list, nb);
1795                 break;
1796         default:
1797                 ret = -EINVAL;
1798         }
1799
1800         return ret;
1801 }
1802 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1803
1804
1805 /*********************************************************************
1806  *                              GOVERNORS                            *
1807  *********************************************************************/
1808
1809 /**
1810  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1811  * @policy: cpufreq policy to switch the frequency for.
1812  * @target_freq: New frequency to set (may be approximate).
1813  *
1814  * Carry out a fast frequency switch without sleeping.
1815  *
1816  * The driver's ->fast_switch() callback invoked by this function must be
1817  * suitable for being called from within RCU-sched read-side critical sections
1818  * and it is expected to select the minimum available frequency greater than or
1819  * equal to @target_freq (CPUFREQ_RELATION_L).
1820  *
1821  * This function must not be called if policy->fast_switch_enabled is unset.
1822  *
1823  * Governors calling this function must guarantee that it will never be invoked
1824  * twice in parallel for the same policy and that it will never be called in
1825  * parallel with either ->target() or ->target_index() for the same policy.
1826  *
1827  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1828  * callback to indicate an error condition, the hardware configuration must be
1829  * preserved.
1830  */
1831 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1832                                         unsigned int target_freq)
1833 {
1834         target_freq = clamp_val(target_freq, policy->min, policy->max);
1835
1836         return cpufreq_driver->fast_switch(policy, target_freq);
1837 }
1838 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1839
1840 /* Must set freqs->new to intermediate frequency */
1841 static int __target_intermediate(struct cpufreq_policy *policy,
1842                                  struct cpufreq_freqs *freqs, int index)
1843 {
1844         int ret;
1845
1846         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1847
1848         /* We don't need to switch to intermediate freq */
1849         if (!freqs->new)
1850                 return 0;
1851
1852         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1853                  __func__, policy->cpu, freqs->old, freqs->new);
1854
1855         cpufreq_freq_transition_begin(policy, freqs);
1856         ret = cpufreq_driver->target_intermediate(policy, index);
1857         cpufreq_freq_transition_end(policy, freqs, ret);
1858
1859         if (ret)
1860                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1861                        __func__, ret);
1862
1863         return ret;
1864 }
1865
1866 static int __target_index(struct cpufreq_policy *policy, int index)
1867 {
1868         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1869         unsigned int intermediate_freq = 0;
1870         unsigned int newfreq = policy->freq_table[index].frequency;
1871         int retval = -EINVAL;
1872         bool notify;
1873
1874         if (newfreq == policy->cur)
1875                 return 0;
1876
1877         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1878         if (notify) {
1879                 /* Handle switching to intermediate frequency */
1880                 if (cpufreq_driver->get_intermediate) {
1881                         retval = __target_intermediate(policy, &freqs, index);
1882                         if (retval)
1883                                 return retval;
1884
1885                         intermediate_freq = freqs.new;
1886                         /* Set old freq to intermediate */
1887                         if (intermediate_freq)
1888                                 freqs.old = freqs.new;
1889                 }
1890
1891                 freqs.new = newfreq;
1892                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1893                          __func__, policy->cpu, freqs.old, freqs.new);
1894
1895                 cpufreq_freq_transition_begin(policy, &freqs);
1896         }
1897
1898         retval = cpufreq_driver->target_index(policy, index);
1899         if (retval)
1900                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1901                        retval);
1902
1903         if (notify) {
1904                 cpufreq_freq_transition_end(policy, &freqs, retval);
1905
1906                 /*
1907                  * Failed after setting to intermediate freq? Driver should have
1908                  * reverted back to initial frequency and so should we. Check
1909                  * here for intermediate_freq instead of get_intermediate, in
1910                  * case we haven't switched to intermediate freq at all.
1911                  */
1912                 if (unlikely(retval && intermediate_freq)) {
1913                         freqs.old = intermediate_freq;
1914                         freqs.new = policy->restore_freq;
1915                         cpufreq_freq_transition_begin(policy, &freqs);
1916                         cpufreq_freq_transition_end(policy, &freqs, 0);
1917                 }
1918         }
1919
1920         return retval;
1921 }
1922
1923 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1924                             unsigned int target_freq,
1925                             unsigned int relation)
1926 {
1927         unsigned int old_target_freq = target_freq;
1928         int index;
1929
1930         if (cpufreq_disabled())
1931                 return -ENODEV;
1932
1933         /* Make sure that target_freq is within supported range */
1934         target_freq = clamp_val(target_freq, policy->min, policy->max);
1935
1936         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1937                  policy->cpu, target_freq, relation, old_target_freq);
1938
1939         /*
1940          * This might look like a redundant call as we are checking it again
1941          * after finding index. But it is left intentionally for cases where
1942          * exactly same freq is called again and so we can save on few function
1943          * calls.
1944          */
1945         if (target_freq == policy->cur)
1946                 return 0;
1947
1948         /* Save last value to restore later on errors */
1949         policy->restore_freq = policy->cur;
1950
1951         if (cpufreq_driver->target)
1952                 return cpufreq_driver->target(policy, target_freq, relation);
1953
1954         if (!cpufreq_driver->target_index)
1955                 return -EINVAL;
1956
1957         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1958
1959         return __target_index(policy, index);
1960 }
1961 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1962
1963 int cpufreq_driver_target(struct cpufreq_policy *policy,
1964                           unsigned int target_freq,
1965                           unsigned int relation)
1966 {
1967         int ret = -EINVAL;
1968
1969         down_write(&policy->rwsem);
1970
1971         ret = __cpufreq_driver_target(policy, target_freq, relation);
1972
1973         up_write(&policy->rwsem);
1974
1975         return ret;
1976 }
1977 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1978
1979 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1980 {
1981         return NULL;
1982 }
1983
1984 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1985 {
1986         int ret;
1987
1988         /* Don't start any governor operations if we are entering suspend */
1989         if (cpufreq_suspended)
1990                 return 0;
1991         /*
1992          * Governor might not be initiated here if ACPI _PPC changed
1993          * notification happened, so check it.
1994          */
1995         if (!policy->governor)
1996                 return -EINVAL;
1997
1998         if (policy->governor->max_transition_latency &&
1999             policy->cpuinfo.transition_latency >
2000             policy->governor->max_transition_latency) {
2001                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2002
2003                 if (gov) {
2004                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2005                                 policy->governor->name, gov->name);
2006                         policy->governor = gov;
2007                 } else {
2008                         return -EINVAL;
2009                 }
2010         }
2011
2012         if (!try_module_get(policy->governor->owner))
2013                 return -EINVAL;
2014
2015         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2016
2017         if (policy->governor->init) {
2018                 ret = policy->governor->init(policy);
2019                 if (ret) {
2020                         module_put(policy->governor->owner);
2021                         return ret;
2022                 }
2023         }
2024
2025         return 0;
2026 }
2027
2028 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2029 {
2030         if (cpufreq_suspended || !policy->governor)
2031                 return;
2032
2033         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2034
2035         if (policy->governor->exit)
2036                 policy->governor->exit(policy);
2037
2038         module_put(policy->governor->owner);
2039 }
2040
2041 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2042 {
2043         int ret;
2044
2045         if (cpufreq_suspended)
2046                 return 0;
2047
2048         if (!policy->governor)
2049                 return -EINVAL;
2050
2051         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2052
2053         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2054                 cpufreq_update_current_freq(policy);
2055
2056         if (policy->governor->start) {
2057                 ret = policy->governor->start(policy);
2058                 if (ret)
2059                         return ret;
2060         }
2061
2062         if (policy->governor->limits)
2063                 policy->governor->limits(policy);
2064
2065         return 0;
2066 }
2067
2068 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2069 {
2070         if (cpufreq_suspended || !policy->governor)
2071                 return;
2072
2073         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2074
2075         if (policy->governor->stop)
2076                 policy->governor->stop(policy);
2077 }
2078
2079 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2080 {
2081         if (cpufreq_suspended || !policy->governor)
2082                 return;
2083
2084         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2085
2086         if (policy->governor->limits)
2087                 policy->governor->limits(policy);
2088 }
2089
2090 int cpufreq_register_governor(struct cpufreq_governor *governor)
2091 {
2092         int err;
2093
2094         if (!governor)
2095                 return -EINVAL;
2096
2097         if (cpufreq_disabled())
2098                 return -ENODEV;
2099
2100         mutex_lock(&cpufreq_governor_mutex);
2101
2102         err = -EBUSY;
2103         if (!find_governor(governor->name)) {
2104                 err = 0;
2105                 list_add(&governor->governor_list, &cpufreq_governor_list);
2106         }
2107
2108         mutex_unlock(&cpufreq_governor_mutex);
2109         return err;
2110 }
2111 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2112
2113 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2114 {
2115         struct cpufreq_policy *policy;
2116         unsigned long flags;
2117
2118         if (!governor)
2119                 return;
2120
2121         if (cpufreq_disabled())
2122                 return;
2123
2124         /* clear last_governor for all inactive policies */
2125         read_lock_irqsave(&cpufreq_driver_lock, flags);
2126         for_each_inactive_policy(policy) {
2127                 if (!strcmp(policy->last_governor, governor->name)) {
2128                         policy->governor = NULL;
2129                         strcpy(policy->last_governor, "\0");
2130                 }
2131         }
2132         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2133
2134         mutex_lock(&cpufreq_governor_mutex);
2135         list_del(&governor->governor_list);
2136         mutex_unlock(&cpufreq_governor_mutex);
2137         return;
2138 }
2139 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2140
2141
2142 /*********************************************************************
2143  *                          POLICY INTERFACE                         *
2144  *********************************************************************/
2145
2146 /**
2147  * cpufreq_get_policy - get the current cpufreq_policy
2148  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2149  *      is written
2150  *
2151  * Reads the current cpufreq policy.
2152  */
2153 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2154 {
2155         struct cpufreq_policy *cpu_policy;
2156         if (!policy)
2157                 return -EINVAL;
2158
2159         cpu_policy = cpufreq_cpu_get(cpu);
2160         if (!cpu_policy)
2161                 return -EINVAL;
2162
2163         memcpy(policy, cpu_policy, sizeof(*policy));
2164
2165         cpufreq_cpu_put(cpu_policy);
2166         return 0;
2167 }
2168 EXPORT_SYMBOL(cpufreq_get_policy);
2169
2170 /*
2171  * policy : current policy.
2172  * new_policy: policy to be set.
2173  */
2174 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2175                                 struct cpufreq_policy *new_policy)
2176 {
2177         struct cpufreq_governor *old_gov;
2178         int ret;
2179
2180         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2181                  new_policy->cpu, new_policy->min, new_policy->max);
2182
2183         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2184
2185         /*
2186         * This check works well when we store new min/max freq attributes,
2187         * because new_policy is a copy of policy with one field updated.
2188         */
2189         if (new_policy->min > new_policy->max)
2190                 return -EINVAL;
2191
2192         /* verify the cpu speed can be set within this limit */
2193         ret = cpufreq_driver->verify(new_policy);
2194         if (ret)
2195                 return ret;
2196
2197         /* adjust if necessary - all reasons */
2198         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2199                         CPUFREQ_ADJUST, new_policy);
2200
2201         /*
2202          * verify the cpu speed can be set within this limit, which might be
2203          * different to the first one
2204          */
2205         ret = cpufreq_driver->verify(new_policy);
2206         if (ret)
2207                 return ret;
2208
2209         /* notification of the new policy */
2210         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2211                         CPUFREQ_NOTIFY, new_policy);
2212
2213         policy->min = new_policy->min;
2214         policy->max = new_policy->max;
2215
2216         policy->cached_target_freq = UINT_MAX;
2217
2218         pr_debug("new min and max freqs are %u - %u kHz\n",
2219                  policy->min, policy->max);
2220
2221         if (cpufreq_driver->setpolicy) {
2222                 policy->policy = new_policy->policy;
2223                 pr_debug("setting range\n");
2224                 return cpufreq_driver->setpolicy(new_policy);
2225         }
2226
2227         if (new_policy->governor == policy->governor) {
2228                 pr_debug("cpufreq: governor limits update\n");
2229                 cpufreq_governor_limits(policy);
2230                 return 0;
2231         }
2232
2233         pr_debug("governor switch\n");
2234
2235         /* save old, working values */
2236         old_gov = policy->governor;
2237         /* end old governor */
2238         if (old_gov) {
2239                 cpufreq_stop_governor(policy);
2240                 cpufreq_exit_governor(policy);
2241         }
2242
2243         /* start new governor */
2244         policy->governor = new_policy->governor;
2245         ret = cpufreq_init_governor(policy);
2246         if (!ret) {
2247                 ret = cpufreq_start_governor(policy);
2248                 if (!ret) {
2249                         pr_debug("cpufreq: governor change\n");
2250                         return 0;
2251                 }
2252                 cpufreq_exit_governor(policy);
2253         }
2254
2255         /* new governor failed, so re-start old one */
2256         pr_debug("starting governor %s failed\n", policy->governor->name);
2257         if (old_gov) {
2258                 policy->governor = old_gov;
2259                 if (cpufreq_init_governor(policy))
2260                         policy->governor = NULL;
2261                 else
2262                         cpufreq_start_governor(policy);
2263         }
2264
2265         return ret;
2266 }
2267
2268 /**
2269  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2270  *      @cpu: CPU which shall be re-evaluated
2271  *
2272  *      Useful for policy notifiers which have different necessities
2273  *      at different times.
2274  */
2275 int cpufreq_update_policy(unsigned int cpu)
2276 {
2277         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2278         struct cpufreq_policy new_policy;
2279         int ret;
2280
2281         if (!policy)
2282                 return -ENODEV;
2283
2284         down_write(&policy->rwsem);
2285
2286         pr_debug("updating policy for CPU %u\n", cpu);
2287         memcpy(&new_policy, policy, sizeof(*policy));
2288         new_policy.min = policy->user_policy.min;
2289         new_policy.max = policy->user_policy.max;
2290
2291         /*
2292          * BIOS might change freq behind our back
2293          * -> ask driver for current freq and notify governors about a change
2294          */
2295         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2296                 if (cpufreq_suspended) {
2297                         ret = -EAGAIN;
2298                         goto unlock;
2299                 }
2300                 new_policy.cur = cpufreq_update_current_freq(policy);
2301                 if (WARN_ON(!new_policy.cur)) {
2302                         ret = -EIO;
2303                         goto unlock;
2304                 }
2305         }
2306
2307         ret = cpufreq_set_policy(policy, &new_policy);
2308
2309 unlock:
2310         up_write(&policy->rwsem);
2311
2312         cpufreq_cpu_put(policy);
2313         return ret;
2314 }
2315 EXPORT_SYMBOL(cpufreq_update_policy);
2316
2317 /*********************************************************************
2318  *               BOOST                                               *
2319  *********************************************************************/
2320 static int cpufreq_boost_set_sw(int state)
2321 {
2322         struct cpufreq_policy *policy;
2323         int ret = -EINVAL;
2324
2325         for_each_active_policy(policy) {
2326                 if (!policy->freq_table)
2327                         continue;
2328
2329                 ret = cpufreq_frequency_table_cpuinfo(policy,
2330                                                       policy->freq_table);
2331                 if (ret) {
2332                         pr_err("%s: Policy frequency update failed\n",
2333                                __func__);
2334                         break;
2335                 }
2336
2337                 down_write(&policy->rwsem);
2338                 policy->user_policy.max = policy->max;
2339                 cpufreq_governor_limits(policy);
2340                 up_write(&policy->rwsem);
2341         }
2342
2343         return ret;
2344 }
2345
2346 int cpufreq_boost_trigger_state(int state)
2347 {
2348         unsigned long flags;
2349         int ret = 0;
2350
2351         if (cpufreq_driver->boost_enabled == state)
2352                 return 0;
2353
2354         write_lock_irqsave(&cpufreq_driver_lock, flags);
2355         cpufreq_driver->boost_enabled = state;
2356         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2357
2358         ret = cpufreq_driver->set_boost(state);
2359         if (ret) {
2360                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2361                 cpufreq_driver->boost_enabled = !state;
2362                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2363
2364                 pr_err("%s: Cannot %s BOOST\n",
2365                        __func__, state ? "enable" : "disable");
2366         }
2367
2368         return ret;
2369 }
2370
2371 static bool cpufreq_boost_supported(void)
2372 {
2373         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2374 }
2375
2376 static int create_boost_sysfs_file(void)
2377 {
2378         int ret;
2379
2380         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2381         if (ret)
2382                 pr_err("%s: cannot register global BOOST sysfs file\n",
2383                        __func__);
2384
2385         return ret;
2386 }
2387
2388 static void remove_boost_sysfs_file(void)
2389 {
2390         if (cpufreq_boost_supported())
2391                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2392 }
2393
2394 int cpufreq_enable_boost_support(void)
2395 {
2396         if (!cpufreq_driver)
2397                 return -EINVAL;
2398
2399         if (cpufreq_boost_supported())
2400                 return 0;
2401
2402         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2403
2404         /* This will get removed on driver unregister */
2405         return create_boost_sysfs_file();
2406 }
2407 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2408
2409 int cpufreq_boost_enabled(void)
2410 {
2411         return cpufreq_driver->boost_enabled;
2412 }
2413 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2414
2415 /*********************************************************************
2416  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2417  *********************************************************************/
2418 static enum cpuhp_state hp_online;
2419
2420 static int cpuhp_cpufreq_online(unsigned int cpu)
2421 {
2422         cpufreq_online(cpu);
2423
2424         return 0;
2425 }
2426
2427 static int cpuhp_cpufreq_offline(unsigned int cpu)
2428 {
2429         cpufreq_offline(cpu);
2430
2431         return 0;
2432 }
2433
2434 /**
2435  * cpufreq_register_driver - register a CPU Frequency driver
2436  * @driver_data: A struct cpufreq_driver containing the values#
2437  * submitted by the CPU Frequency driver.
2438  *
2439  * Registers a CPU Frequency driver to this core code. This code
2440  * returns zero on success, -EEXIST when another driver got here first
2441  * (and isn't unregistered in the meantime).
2442  *
2443  */
2444 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2445 {
2446         unsigned long flags;
2447         int ret;
2448
2449         if (cpufreq_disabled())
2450                 return -ENODEV;
2451
2452         /*
2453          * The cpufreq core depends heavily on the availability of device
2454          * structure, make sure they are available before proceeding further.
2455          */
2456         if (!get_cpu_device(0))
2457                 return -EPROBE_DEFER;
2458
2459         if (!driver_data || !driver_data->verify || !driver_data->init ||
2460             !(driver_data->setpolicy || driver_data->target_index ||
2461                     driver_data->target) ||
2462              (driver_data->setpolicy && (driver_data->target_index ||
2463                     driver_data->target)) ||
2464              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2465                 return -EINVAL;
2466
2467         pr_debug("trying to register driver %s\n", driver_data->name);
2468
2469         /* Protect against concurrent CPU online/offline. */
2470         get_online_cpus();
2471
2472         write_lock_irqsave(&cpufreq_driver_lock, flags);
2473         if (cpufreq_driver) {
2474                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2475                 ret = -EEXIST;
2476                 goto out;
2477         }
2478         cpufreq_driver = driver_data;
2479         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2480
2481         if (driver_data->setpolicy)
2482                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2483
2484         if (cpufreq_boost_supported()) {
2485                 ret = create_boost_sysfs_file();
2486                 if (ret)
2487                         goto err_null_driver;
2488         }
2489
2490         ret = subsys_interface_register(&cpufreq_interface);
2491         if (ret)
2492                 goto err_boost_unreg;
2493
2494         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2495             list_empty(&cpufreq_policy_list)) {
2496                 /* if all ->init() calls failed, unregister */
2497                 ret = -ENODEV;
2498                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2499                          driver_data->name);
2500                 goto err_if_unreg;
2501         }
2502
2503         ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "cpufreq:online",
2504                                         cpuhp_cpufreq_online,
2505                                         cpuhp_cpufreq_offline);
2506         if (ret < 0)
2507                 goto err_if_unreg;
2508         hp_online = ret;
2509         ret = 0;
2510
2511         pr_debug("driver %s up and running\n", driver_data->name);
2512         goto out;
2513
2514 err_if_unreg:
2515         subsys_interface_unregister(&cpufreq_interface);
2516 err_boost_unreg:
2517         remove_boost_sysfs_file();
2518 err_null_driver:
2519         write_lock_irqsave(&cpufreq_driver_lock, flags);
2520         cpufreq_driver = NULL;
2521         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2522 out:
2523         put_online_cpus();
2524         return ret;
2525 }
2526 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2527
2528 /**
2529  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2530  *
2531  * Unregister the current CPUFreq driver. Only call this if you have
2532  * the right to do so, i.e. if you have succeeded in initialising before!
2533  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2534  * currently not initialised.
2535  */
2536 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2537 {
2538         unsigned long flags;
2539
2540         if (!cpufreq_driver || (driver != cpufreq_driver))
2541                 return -EINVAL;
2542
2543         pr_debug("unregistering driver %s\n", driver->name);
2544
2545         /* Protect against concurrent cpu hotplug */
2546         get_online_cpus();
2547         subsys_interface_unregister(&cpufreq_interface);
2548         remove_boost_sysfs_file();
2549         cpuhp_remove_state_nocalls(hp_online);
2550
2551         write_lock_irqsave(&cpufreq_driver_lock, flags);
2552
2553         cpufreq_driver = NULL;
2554
2555         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2556         put_online_cpus();
2557
2558         return 0;
2559 }
2560 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2561
2562 struct kobject *cpufreq_global_kobject;
2563 EXPORT_SYMBOL(cpufreq_global_kobject);
2564
2565 static int __init cpufreq_core_init(void)
2566 {
2567         if (cpufreq_disabled())
2568                 return -ENODEV;
2569
2570         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2571         BUG_ON(!cpufreq_global_kobject);
2572
2573         return 0;
2574 }
2575 core_initcall(cpufreq_core_init);