Merge tag 'imx-fixes-5.0-2' of git://git.kernel.org/pub/scm/linux/kernel/git/shawnguo...
[sfrench/cifs-2.6.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 SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
93
94 static int off __read_mostly;
95 static int cpufreq_disabled(void)
96 {
97         return off;
98 }
99 void disable_cpufreq(void)
100 {
101         off = 1;
102 }
103 static DEFINE_MUTEX(cpufreq_governor_mutex);
104
105 bool have_governor_per_policy(void)
106 {
107         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
108 }
109 EXPORT_SYMBOL_GPL(have_governor_per_policy);
110
111 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
112 {
113         if (have_governor_per_policy())
114                 return &policy->kobj;
115         else
116                 return cpufreq_global_kobject;
117 }
118 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
119
120 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
121 {
122         u64 idle_time;
123         u64 cur_wall_time;
124         u64 busy_time;
125
126         cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
127
128         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
129         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
130         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
131         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
132         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
133         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
134
135         idle_time = cur_wall_time - busy_time;
136         if (wall)
137                 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
138
139         return div_u64(idle_time, NSEC_PER_USEC);
140 }
141
142 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
143 {
144         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
145
146         if (idle_time == -1ULL)
147                 return get_cpu_idle_time_jiffy(cpu, wall);
148         else if (!io_busy)
149                 idle_time += get_cpu_iowait_time_us(cpu, wall);
150
151         return idle_time;
152 }
153 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
154
155 __weak void arch_set_freq_scale(struct cpumask *cpus, unsigned long cur_freq,
156                 unsigned long max_freq)
157 {
158 }
159 EXPORT_SYMBOL_GPL(arch_set_freq_scale);
160
161 /*
162  * This is a generic cpufreq init() routine which can be used by cpufreq
163  * drivers of SMP systems. It will do following:
164  * - validate & show freq table passed
165  * - set policies transition latency
166  * - policy->cpus with all possible CPUs
167  */
168 int cpufreq_generic_init(struct cpufreq_policy *policy,
169                 struct cpufreq_frequency_table *table,
170                 unsigned int transition_latency)
171 {
172         policy->freq_table = table;
173         policy->cpuinfo.transition_latency = transition_latency;
174
175         /*
176          * The driver only supports the SMP configuration where all processors
177          * share the clock and voltage and clock.
178          */
179         cpumask_setall(policy->cpus);
180
181         return 0;
182 }
183 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
184
185 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
186 {
187         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
188
189         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
190 }
191 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
192
193 unsigned int cpufreq_generic_get(unsigned int cpu)
194 {
195         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
196
197         if (!policy || IS_ERR(policy->clk)) {
198                 pr_err("%s: No %s associated to cpu: %d\n",
199                        __func__, policy ? "clk" : "policy", cpu);
200                 return 0;
201         }
202
203         return clk_get_rate(policy->clk) / 1000;
204 }
205 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
206
207 /**
208  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
209  *
210  * @cpu: cpu to find policy for.
211  *
212  * This returns policy for 'cpu', returns NULL if it doesn't exist.
213  * It also increments the kobject reference count to mark it busy and so would
214  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
215  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
216  * freed as that depends on the kobj count.
217  *
218  * Return: A valid policy on success, otherwise NULL on failure.
219  */
220 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
221 {
222         struct cpufreq_policy *policy = NULL;
223         unsigned long flags;
224
225         if (WARN_ON(cpu >= nr_cpu_ids))
226                 return NULL;
227
228         /* get the cpufreq driver */
229         read_lock_irqsave(&cpufreq_driver_lock, flags);
230
231         if (cpufreq_driver) {
232                 /* get the CPU */
233                 policy = cpufreq_cpu_get_raw(cpu);
234                 if (policy)
235                         kobject_get(&policy->kobj);
236         }
237
238         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
239
240         return policy;
241 }
242 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
243
244 /**
245  * cpufreq_cpu_put: Decrements the usage count of a policy
246  *
247  * @policy: policy earlier returned by cpufreq_cpu_get().
248  *
249  * This decrements the kobject reference count incremented earlier by calling
250  * cpufreq_cpu_get().
251  */
252 void cpufreq_cpu_put(struct cpufreq_policy *policy)
253 {
254         kobject_put(&policy->kobj);
255 }
256 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
257
258 /*********************************************************************
259  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
260  *********************************************************************/
261
262 /**
263  * adjust_jiffies - adjust the system "loops_per_jiffy"
264  *
265  * This function alters the system "loops_per_jiffy" for the clock
266  * speed change. Note that loops_per_jiffy cannot be updated on SMP
267  * systems as each CPU might be scaled differently. So, use the arch
268  * per-CPU loops_per_jiffy value wherever possible.
269  */
270 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
271 {
272 #ifndef CONFIG_SMP
273         static unsigned long l_p_j_ref;
274         static unsigned int l_p_j_ref_freq;
275
276         if (ci->flags & CPUFREQ_CONST_LOOPS)
277                 return;
278
279         if (!l_p_j_ref_freq) {
280                 l_p_j_ref = loops_per_jiffy;
281                 l_p_j_ref_freq = ci->old;
282                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
283                          l_p_j_ref, l_p_j_ref_freq);
284         }
285         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
286                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
287                                                                 ci->new);
288                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
289                          loops_per_jiffy, ci->new);
290         }
291 #endif
292 }
293
294 /**
295  * cpufreq_notify_transition - Notify frequency transition and adjust_jiffies.
296  * @policy: cpufreq policy to enable fast frequency switching for.
297  * @freqs: contain details of the frequency update.
298  * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
299  *
300  * This function calls the transition notifiers and the "adjust_jiffies"
301  * function. It is called twice on all CPU frequency changes that have
302  * external effects.
303  */
304 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
305                                       struct cpufreq_freqs *freqs,
306                                       unsigned int state)
307 {
308         BUG_ON(irqs_disabled());
309
310         if (cpufreq_disabled())
311                 return;
312
313         freqs->flags = cpufreq_driver->flags;
314         pr_debug("notification %u of frequency transition to %u kHz\n",
315                  state, freqs->new);
316
317         switch (state) {
318         case CPUFREQ_PRECHANGE:
319                 /*
320                  * Detect if the driver reported a value as "old frequency"
321                  * which is not equal to what the cpufreq core thinks is
322                  * "old frequency".
323                  */
324                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
325                         if (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
332                 for_each_cpu(freqs->cpu, policy->cpus) {
333                         srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
334                                                  CPUFREQ_PRECHANGE, freqs);
335                 }
336
337                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
338                 break;
339
340         case CPUFREQ_POSTCHANGE:
341                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
342                 pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
343                          cpumask_pr_args(policy->cpus));
344
345                 for_each_cpu(freqs->cpu, policy->cpus) {
346                         trace_cpu_frequency(freqs->new, freqs->cpu);
347                         srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
348                                                  CPUFREQ_POSTCHANGE, freqs);
349                 }
350
351                 cpufreq_stats_record_transition(policy, freqs->new);
352                 policy->cur = freqs->new;
353         }
354 }
355
356 /* Do post notifications when there are chances that transition has failed */
357 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
358                 struct cpufreq_freqs *freqs, int transition_failed)
359 {
360         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
361         if (!transition_failed)
362                 return;
363
364         swap(freqs->old, freqs->new);
365         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
366         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
367 }
368
369 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
370                 struct cpufreq_freqs *freqs)
371 {
372
373         /*
374          * Catch double invocations of _begin() which lead to self-deadlock.
375          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
376          * doesn't invoke _begin() on their behalf, and hence the chances of
377          * double invocations are very low. Moreover, there are scenarios
378          * where these checks can emit false-positive warnings in these
379          * drivers; so we avoid that by skipping them altogether.
380          */
381         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
382                                 && current == policy->transition_task);
383
384 wait:
385         wait_event(policy->transition_wait, !policy->transition_ongoing);
386
387         spin_lock(&policy->transition_lock);
388
389         if (unlikely(policy->transition_ongoing)) {
390                 spin_unlock(&policy->transition_lock);
391                 goto wait;
392         }
393
394         policy->transition_ongoing = true;
395         policy->transition_task = current;
396
397         spin_unlock(&policy->transition_lock);
398
399         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
400 }
401 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
402
403 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
404                 struct cpufreq_freqs *freqs, int transition_failed)
405 {
406         if (WARN_ON(!policy->transition_ongoing))
407                 return;
408
409         cpufreq_notify_post_transition(policy, freqs, transition_failed);
410
411         policy->transition_ongoing = false;
412         policy->transition_task = NULL;
413
414         wake_up(&policy->transition_wait);
415 }
416 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
417
418 /*
419  * Fast frequency switching status count.  Positive means "enabled", negative
420  * means "disabled" and 0 means "not decided yet".
421  */
422 static int cpufreq_fast_switch_count;
423 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
424
425 static void cpufreq_list_transition_notifiers(void)
426 {
427         struct notifier_block *nb;
428
429         pr_info("Registered transition notifiers:\n");
430
431         mutex_lock(&cpufreq_transition_notifier_list.mutex);
432
433         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
434                 pr_info("%pF\n", nb->notifier_call);
435
436         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
437 }
438
439 /**
440  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
441  * @policy: cpufreq policy to enable fast frequency switching for.
442  *
443  * Try to enable fast frequency switching for @policy.
444  *
445  * The attempt will fail if there is at least one transition notifier registered
446  * at this point, as fast frequency switching is quite fundamentally at odds
447  * with transition notifiers.  Thus if successful, it will make registration of
448  * transition notifiers fail going forward.
449  */
450 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
451 {
452         lockdep_assert_held(&policy->rwsem);
453
454         if (!policy->fast_switch_possible)
455                 return;
456
457         mutex_lock(&cpufreq_fast_switch_lock);
458         if (cpufreq_fast_switch_count >= 0) {
459                 cpufreq_fast_switch_count++;
460                 policy->fast_switch_enabled = true;
461         } else {
462                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
463                         policy->cpu);
464                 cpufreq_list_transition_notifiers();
465         }
466         mutex_unlock(&cpufreq_fast_switch_lock);
467 }
468 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
469
470 /**
471  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
472  * @policy: cpufreq policy to disable fast frequency switching for.
473  */
474 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
475 {
476         mutex_lock(&cpufreq_fast_switch_lock);
477         if (policy->fast_switch_enabled) {
478                 policy->fast_switch_enabled = false;
479                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
480                         cpufreq_fast_switch_count--;
481         }
482         mutex_unlock(&cpufreq_fast_switch_lock);
483 }
484 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
485
486 /**
487  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
488  * one.
489  * @target_freq: target frequency to resolve.
490  *
491  * The target to driver frequency mapping is cached in the policy.
492  *
493  * Return: Lowest driver-supported frequency greater than or equal to the
494  * given target_freq, subject to policy (min/max) and driver limitations.
495  */
496 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
497                                          unsigned int target_freq)
498 {
499         target_freq = clamp_val(target_freq, policy->min, policy->max);
500         policy->cached_target_freq = target_freq;
501
502         if (cpufreq_driver->target_index) {
503                 int idx;
504
505                 idx = cpufreq_frequency_table_target(policy, target_freq,
506                                                      CPUFREQ_RELATION_L);
507                 policy->cached_resolved_idx = idx;
508                 return policy->freq_table[idx].frequency;
509         }
510
511         if (cpufreq_driver->resolve_freq)
512                 return cpufreq_driver->resolve_freq(policy, target_freq);
513
514         return target_freq;
515 }
516 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
517
518 unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
519 {
520         unsigned int latency;
521
522         if (policy->transition_delay_us)
523                 return policy->transition_delay_us;
524
525         latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
526         if (latency) {
527                 /*
528                  * For platforms that can change the frequency very fast (< 10
529                  * us), the above formula gives a decent transition delay. But
530                  * for platforms where transition_latency is in milliseconds, it
531                  * ends up giving unrealistic values.
532                  *
533                  * Cap the default transition delay to 10 ms, which seems to be
534                  * a reasonable amount of time after which we should reevaluate
535                  * the frequency.
536                  */
537                 return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
538         }
539
540         return LATENCY_MULTIPLIER;
541 }
542 EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
543
544 /*********************************************************************
545  *                          SYSFS INTERFACE                          *
546  *********************************************************************/
547 static ssize_t show_boost(struct kobject *kobj,
548                                  struct attribute *attr, char *buf)
549 {
550         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
551 }
552
553 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
554                                   const char *buf, size_t count)
555 {
556         int ret, enable;
557
558         ret = sscanf(buf, "%d", &enable);
559         if (ret != 1 || enable < 0 || enable > 1)
560                 return -EINVAL;
561
562         if (cpufreq_boost_trigger_state(enable)) {
563                 pr_err("%s: Cannot %s BOOST!\n",
564                        __func__, enable ? "enable" : "disable");
565                 return -EINVAL;
566         }
567
568         pr_debug("%s: cpufreq BOOST %s\n",
569                  __func__, enable ? "enabled" : "disabled");
570
571         return count;
572 }
573 define_one_global_rw(boost);
574
575 static struct cpufreq_governor *find_governor(const char *str_governor)
576 {
577         struct cpufreq_governor *t;
578
579         for_each_governor(t)
580                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
581                         return t;
582
583         return NULL;
584 }
585
586 /**
587  * cpufreq_parse_governor - parse a governor string
588  */
589 static int cpufreq_parse_governor(char *str_governor,
590                                   struct cpufreq_policy *policy)
591 {
592         if (cpufreq_driver->setpolicy) {
593                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
594                         policy->policy = CPUFREQ_POLICY_PERFORMANCE;
595                         return 0;
596                 }
597
598                 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
599                         policy->policy = CPUFREQ_POLICY_POWERSAVE;
600                         return 0;
601                 }
602         } else {
603                 struct cpufreq_governor *t;
604
605                 mutex_lock(&cpufreq_governor_mutex);
606
607                 t = find_governor(str_governor);
608                 if (!t) {
609                         int ret;
610
611                         mutex_unlock(&cpufreq_governor_mutex);
612
613                         ret = request_module("cpufreq_%s", str_governor);
614                         if (ret)
615                                 return -EINVAL;
616
617                         mutex_lock(&cpufreq_governor_mutex);
618
619                         t = find_governor(str_governor);
620                 }
621                 if (t && !try_module_get(t->owner))
622                         t = NULL;
623
624                 mutex_unlock(&cpufreq_governor_mutex);
625
626                 if (t) {
627                         policy->governor = t;
628                         return 0;
629                 }
630         }
631
632         return -EINVAL;
633 }
634
635 /**
636  * cpufreq_per_cpu_attr_read() / show_##file_name() -
637  * print out cpufreq information
638  *
639  * Write out information from cpufreq_driver->policy[cpu]; object must be
640  * "unsigned int".
641  */
642
643 #define show_one(file_name, object)                     \
644 static ssize_t show_##file_name                         \
645 (struct cpufreq_policy *policy, char *buf)              \
646 {                                                       \
647         return sprintf(buf, "%u\n", policy->object);    \
648 }
649
650 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
651 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
652 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
653 show_one(scaling_min_freq, min);
654 show_one(scaling_max_freq, max);
655
656 __weak unsigned int arch_freq_get_on_cpu(int cpu)
657 {
658         return 0;
659 }
660
661 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
662 {
663         ssize_t ret;
664         unsigned int freq;
665
666         freq = arch_freq_get_on_cpu(policy->cpu);
667         if (freq)
668                 ret = sprintf(buf, "%u\n", freq);
669         else if (cpufreq_driver && cpufreq_driver->setpolicy &&
670                         cpufreq_driver->get)
671                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
672         else
673                 ret = sprintf(buf, "%u\n", policy->cur);
674         return ret;
675 }
676
677 static int cpufreq_set_policy(struct cpufreq_policy *policy,
678                                 struct cpufreq_policy *new_policy);
679
680 /**
681  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
682  */
683 #define store_one(file_name, object)                    \
684 static ssize_t store_##file_name                                        \
685 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
686 {                                                                       \
687         int ret, temp;                                                  \
688         struct cpufreq_policy new_policy;                               \
689                                                                         \
690         memcpy(&new_policy, policy, sizeof(*policy));                   \
691         new_policy.min = policy->user_policy.min;                       \
692         new_policy.max = policy->user_policy.max;                       \
693                                                                         \
694         ret = sscanf(buf, "%u", &new_policy.object);                    \
695         if (ret != 1)                                                   \
696                 return -EINVAL;                                         \
697                                                                         \
698         temp = new_policy.object;                                       \
699         ret = cpufreq_set_policy(policy, &new_policy);          \
700         if (!ret)                                                       \
701                 policy->user_policy.object = temp;                      \
702                                                                         \
703         return ret ? ret : count;                                       \
704 }
705
706 store_one(scaling_min_freq, min);
707 store_one(scaling_max_freq, max);
708
709 /**
710  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
711  */
712 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
713                                         char *buf)
714 {
715         unsigned int cur_freq = __cpufreq_get(policy);
716
717         if (cur_freq)
718                 return sprintf(buf, "%u\n", cur_freq);
719
720         return sprintf(buf, "<unknown>\n");
721 }
722
723 /**
724  * show_scaling_governor - show the current policy for the specified CPU
725  */
726 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
727 {
728         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
729                 return sprintf(buf, "powersave\n");
730         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
731                 return sprintf(buf, "performance\n");
732         else if (policy->governor)
733                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
734                                 policy->governor->name);
735         return -EINVAL;
736 }
737
738 /**
739  * store_scaling_governor - store policy for the specified CPU
740  */
741 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
742                                         const char *buf, size_t count)
743 {
744         int ret;
745         char    str_governor[16];
746         struct cpufreq_policy new_policy;
747
748         memcpy(&new_policy, policy, sizeof(*policy));
749
750         ret = sscanf(buf, "%15s", str_governor);
751         if (ret != 1)
752                 return -EINVAL;
753
754         if (cpufreq_parse_governor(str_governor, &new_policy))
755                 return -EINVAL;
756
757         ret = cpufreq_set_policy(policy, &new_policy);
758
759         if (new_policy.governor)
760                 module_put(new_policy.governor->owner);
761
762         return ret ? ret : count;
763 }
764
765 /**
766  * show_scaling_driver - show the cpufreq driver currently loaded
767  */
768 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
769 {
770         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
771 }
772
773 /**
774  * show_scaling_available_governors - show the available CPUfreq governors
775  */
776 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
777                                                 char *buf)
778 {
779         ssize_t i = 0;
780         struct cpufreq_governor *t;
781
782         if (!has_target()) {
783                 i += sprintf(buf, "performance powersave");
784                 goto out;
785         }
786
787         for_each_governor(t) {
788                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
789                     - (CPUFREQ_NAME_LEN + 2)))
790                         goto out;
791                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
792         }
793 out:
794         i += sprintf(&buf[i], "\n");
795         return i;
796 }
797
798 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
799 {
800         ssize_t i = 0;
801         unsigned int cpu;
802
803         for_each_cpu(cpu, mask) {
804                 if (i)
805                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
806                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
807                 if (i >= (PAGE_SIZE - 5))
808                         break;
809         }
810         i += sprintf(&buf[i], "\n");
811         return i;
812 }
813 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
814
815 /**
816  * show_related_cpus - show the CPUs affected by each transition even if
817  * hw coordination is in use
818  */
819 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
820 {
821         return cpufreq_show_cpus(policy->related_cpus, buf);
822 }
823
824 /**
825  * show_affected_cpus - show the CPUs affected by each transition
826  */
827 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
828 {
829         return cpufreq_show_cpus(policy->cpus, buf);
830 }
831
832 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
833                                         const char *buf, size_t count)
834 {
835         unsigned int freq = 0;
836         unsigned int ret;
837
838         if (!policy->governor || !policy->governor->store_setspeed)
839                 return -EINVAL;
840
841         ret = sscanf(buf, "%u", &freq);
842         if (ret != 1)
843                 return -EINVAL;
844
845         policy->governor->store_setspeed(policy, freq);
846
847         return count;
848 }
849
850 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
851 {
852         if (!policy->governor || !policy->governor->show_setspeed)
853                 return sprintf(buf, "<unsupported>\n");
854
855         return policy->governor->show_setspeed(policy, buf);
856 }
857
858 /**
859  * show_bios_limit - show the current cpufreq HW/BIOS limitation
860  */
861 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
862 {
863         unsigned int limit;
864         int ret;
865         if (cpufreq_driver->bios_limit) {
866                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
867                 if (!ret)
868                         return sprintf(buf, "%u\n", limit);
869         }
870         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
871 }
872
873 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
874 cpufreq_freq_attr_ro(cpuinfo_min_freq);
875 cpufreq_freq_attr_ro(cpuinfo_max_freq);
876 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
877 cpufreq_freq_attr_ro(scaling_available_governors);
878 cpufreq_freq_attr_ro(scaling_driver);
879 cpufreq_freq_attr_ro(scaling_cur_freq);
880 cpufreq_freq_attr_ro(bios_limit);
881 cpufreq_freq_attr_ro(related_cpus);
882 cpufreq_freq_attr_ro(affected_cpus);
883 cpufreq_freq_attr_rw(scaling_min_freq);
884 cpufreq_freq_attr_rw(scaling_max_freq);
885 cpufreq_freq_attr_rw(scaling_governor);
886 cpufreq_freq_attr_rw(scaling_setspeed);
887
888 static struct attribute *default_attrs[] = {
889         &cpuinfo_min_freq.attr,
890         &cpuinfo_max_freq.attr,
891         &cpuinfo_transition_latency.attr,
892         &scaling_min_freq.attr,
893         &scaling_max_freq.attr,
894         &affected_cpus.attr,
895         &related_cpus.attr,
896         &scaling_governor.attr,
897         &scaling_driver.attr,
898         &scaling_available_governors.attr,
899         &scaling_setspeed.attr,
900         NULL
901 };
902
903 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
904 #define to_attr(a) container_of(a, struct freq_attr, attr)
905
906 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
907 {
908         struct cpufreq_policy *policy = to_policy(kobj);
909         struct freq_attr *fattr = to_attr(attr);
910         ssize_t ret;
911
912         down_read(&policy->rwsem);
913         ret = fattr->show(policy, buf);
914         up_read(&policy->rwsem);
915
916         return ret;
917 }
918
919 static ssize_t store(struct kobject *kobj, struct attribute *attr,
920                      const char *buf, size_t count)
921 {
922         struct cpufreq_policy *policy = to_policy(kobj);
923         struct freq_attr *fattr = to_attr(attr);
924         ssize_t ret = -EINVAL;
925
926         /*
927          * cpus_read_trylock() is used here to work around a circular lock
928          * dependency problem with respect to the cpufreq_register_driver().
929          */
930         if (!cpus_read_trylock())
931                 return -EBUSY;
932
933         if (cpu_online(policy->cpu)) {
934                 down_write(&policy->rwsem);
935                 ret = fattr->store(policy, buf, count);
936                 up_write(&policy->rwsem);
937         }
938
939         cpus_read_unlock();
940
941         return ret;
942 }
943
944 static void cpufreq_sysfs_release(struct kobject *kobj)
945 {
946         struct cpufreq_policy *policy = to_policy(kobj);
947         pr_debug("last reference is dropped\n");
948         complete(&policy->kobj_unregister);
949 }
950
951 static const struct sysfs_ops sysfs_ops = {
952         .show   = show,
953         .store  = store,
954 };
955
956 static struct kobj_type ktype_cpufreq = {
957         .sysfs_ops      = &sysfs_ops,
958         .default_attrs  = default_attrs,
959         .release        = cpufreq_sysfs_release,
960 };
961
962 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
963 {
964         struct device *dev = get_cpu_device(cpu);
965
966         if (!dev)
967                 return;
968
969         if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
970                 return;
971
972         dev_dbg(dev, "%s: Adding symlink\n", __func__);
973         if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
974                 dev_err(dev, "cpufreq symlink creation failed\n");
975 }
976
977 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
978                                    struct device *dev)
979 {
980         dev_dbg(dev, "%s: Removing symlink\n", __func__);
981         sysfs_remove_link(&dev->kobj, "cpufreq");
982 }
983
984 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
985 {
986         struct freq_attr **drv_attr;
987         int ret = 0;
988
989         /* set up files for this cpu device */
990         drv_attr = cpufreq_driver->attr;
991         while (drv_attr && *drv_attr) {
992                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
993                 if (ret)
994                         return ret;
995                 drv_attr++;
996         }
997         if (cpufreq_driver->get) {
998                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
999                 if (ret)
1000                         return ret;
1001         }
1002
1003         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1004         if (ret)
1005                 return ret;
1006
1007         if (cpufreq_driver->bios_limit) {
1008                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1009                 if (ret)
1010                         return ret;
1011         }
1012
1013         return 0;
1014 }
1015
1016 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1017 {
1018         return NULL;
1019 }
1020
1021 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1022 {
1023         struct cpufreq_governor *gov = NULL;
1024         struct cpufreq_policy new_policy;
1025
1026         memcpy(&new_policy, policy, sizeof(*policy));
1027
1028         /* Update governor of new_policy to the governor used before hotplug */
1029         gov = find_governor(policy->last_governor);
1030         if (gov) {
1031                 pr_debug("Restoring governor %s for cpu %d\n",
1032                                 policy->governor->name, policy->cpu);
1033         } else {
1034                 gov = cpufreq_default_governor();
1035                 if (!gov)
1036                         return -ENODATA;
1037         }
1038
1039         new_policy.governor = gov;
1040
1041         /* Use the default policy if there is no last_policy. */
1042         if (cpufreq_driver->setpolicy) {
1043                 if (policy->last_policy)
1044                         new_policy.policy = policy->last_policy;
1045                 else
1046                         cpufreq_parse_governor(gov->name, &new_policy);
1047         }
1048         /* set default policy */
1049         return cpufreq_set_policy(policy, &new_policy);
1050 }
1051
1052 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1053 {
1054         int ret = 0;
1055
1056         /* Has this CPU been taken care of already? */
1057         if (cpumask_test_cpu(cpu, policy->cpus))
1058                 return 0;
1059
1060         down_write(&policy->rwsem);
1061         if (has_target())
1062                 cpufreq_stop_governor(policy);
1063
1064         cpumask_set_cpu(cpu, policy->cpus);
1065
1066         if (has_target()) {
1067                 ret = cpufreq_start_governor(policy);
1068                 if (ret)
1069                         pr_err("%s: Failed to start governor\n", __func__);
1070         }
1071         up_write(&policy->rwsem);
1072         return ret;
1073 }
1074
1075 static void handle_update(struct work_struct *work)
1076 {
1077         struct cpufreq_policy *policy =
1078                 container_of(work, struct cpufreq_policy, update);
1079         unsigned int cpu = policy->cpu;
1080         pr_debug("handle_update for cpu %u called\n", cpu);
1081         cpufreq_update_policy(cpu);
1082 }
1083
1084 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1085 {
1086         struct cpufreq_policy *policy;
1087         int ret;
1088
1089         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1090         if (!policy)
1091                 return NULL;
1092
1093         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1094                 goto err_free_policy;
1095
1096         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1097                 goto err_free_cpumask;
1098
1099         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1100                 goto err_free_rcpumask;
1101
1102         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1103                                    cpufreq_global_kobject, "policy%u", cpu);
1104         if (ret) {
1105                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1106                 goto err_free_real_cpus;
1107         }
1108
1109         INIT_LIST_HEAD(&policy->policy_list);
1110         init_rwsem(&policy->rwsem);
1111         spin_lock_init(&policy->transition_lock);
1112         init_waitqueue_head(&policy->transition_wait);
1113         init_completion(&policy->kobj_unregister);
1114         INIT_WORK(&policy->update, handle_update);
1115
1116         policy->cpu = cpu;
1117         return policy;
1118
1119 err_free_real_cpus:
1120         free_cpumask_var(policy->real_cpus);
1121 err_free_rcpumask:
1122         free_cpumask_var(policy->related_cpus);
1123 err_free_cpumask:
1124         free_cpumask_var(policy->cpus);
1125 err_free_policy:
1126         kfree(policy);
1127
1128         return NULL;
1129 }
1130
1131 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1132 {
1133         struct kobject *kobj;
1134         struct completion *cmp;
1135
1136         down_write(&policy->rwsem);
1137         cpufreq_stats_free_table(policy);
1138         kobj = &policy->kobj;
1139         cmp = &policy->kobj_unregister;
1140         up_write(&policy->rwsem);
1141         kobject_put(kobj);
1142
1143         /*
1144          * We need to make sure that the underlying kobj is
1145          * actually not referenced anymore by anybody before we
1146          * proceed with unloading.
1147          */
1148         pr_debug("waiting for dropping of refcount\n");
1149         wait_for_completion(cmp);
1150         pr_debug("wait complete\n");
1151 }
1152
1153 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1154 {
1155         unsigned long flags;
1156         int cpu;
1157
1158         /* Remove policy from list */
1159         write_lock_irqsave(&cpufreq_driver_lock, flags);
1160         list_del(&policy->policy_list);
1161
1162         for_each_cpu(cpu, policy->related_cpus)
1163                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1164         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1165
1166         cpufreq_policy_put_kobj(policy);
1167         free_cpumask_var(policy->real_cpus);
1168         free_cpumask_var(policy->related_cpus);
1169         free_cpumask_var(policy->cpus);
1170         kfree(policy);
1171 }
1172
1173 static int cpufreq_online(unsigned int cpu)
1174 {
1175         struct cpufreq_policy *policy;
1176         bool new_policy;
1177         unsigned long flags;
1178         unsigned int j;
1179         int ret;
1180
1181         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1182
1183         /* Check if this CPU already has a policy to manage it */
1184         policy = per_cpu(cpufreq_cpu_data, cpu);
1185         if (policy) {
1186                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1187                 if (!policy_is_inactive(policy))
1188                         return cpufreq_add_policy_cpu(policy, cpu);
1189
1190                 /* This is the only online CPU for the policy.  Start over. */
1191                 new_policy = false;
1192                 down_write(&policy->rwsem);
1193                 policy->cpu = cpu;
1194                 policy->governor = NULL;
1195                 up_write(&policy->rwsem);
1196         } else {
1197                 new_policy = true;
1198                 policy = cpufreq_policy_alloc(cpu);
1199                 if (!policy)
1200                         return -ENOMEM;
1201         }
1202
1203         cpumask_copy(policy->cpus, cpumask_of(cpu));
1204
1205         /* call driver. From then on the cpufreq must be able
1206          * to accept all calls to ->verify and ->setpolicy for this CPU
1207          */
1208         ret = cpufreq_driver->init(policy);
1209         if (ret) {
1210                 pr_debug("initialization failed\n");
1211                 goto out_free_policy;
1212         }
1213
1214         ret = cpufreq_table_validate_and_sort(policy);
1215         if (ret)
1216                 goto out_exit_policy;
1217
1218         down_write(&policy->rwsem);
1219
1220         if (new_policy) {
1221                 /* related_cpus should at least include policy->cpus. */
1222                 cpumask_copy(policy->related_cpus, policy->cpus);
1223         }
1224
1225         /*
1226          * affected cpus must always be the one, which are online. We aren't
1227          * managing offline cpus here.
1228          */
1229         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1230
1231         if (new_policy) {
1232                 policy->user_policy.min = policy->min;
1233                 policy->user_policy.max = policy->max;
1234
1235                 for_each_cpu(j, policy->related_cpus) {
1236                         per_cpu(cpufreq_cpu_data, j) = policy;
1237                         add_cpu_dev_symlink(policy, j);
1238                 }
1239         } else {
1240                 policy->min = policy->user_policy.min;
1241                 policy->max = policy->user_policy.max;
1242         }
1243
1244         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1245                 policy->cur = cpufreq_driver->get(policy->cpu);
1246                 if (!policy->cur) {
1247                         pr_err("%s: ->get() failed\n", __func__);
1248                         goto out_destroy_policy;
1249                 }
1250         }
1251
1252         /*
1253          * Sometimes boot loaders set CPU frequency to a value outside of
1254          * frequency table present with cpufreq core. In such cases CPU might be
1255          * unstable if it has to run on that frequency for long duration of time
1256          * and so its better to set it to a frequency which is specified in
1257          * freq-table. This also makes cpufreq stats inconsistent as
1258          * cpufreq-stats would fail to register because current frequency of CPU
1259          * isn't found in freq-table.
1260          *
1261          * Because we don't want this change to effect boot process badly, we go
1262          * for the next freq which is >= policy->cur ('cur' must be set by now,
1263          * otherwise we will end up setting freq to lowest of the table as 'cur'
1264          * is initialized to zero).
1265          *
1266          * We are passing target-freq as "policy->cur - 1" otherwise
1267          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1268          * equal to target-freq.
1269          */
1270         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1271             && has_target()) {
1272                 /* Are we running at unknown frequency ? */
1273                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1274                 if (ret == -EINVAL) {
1275                         /* Warn user and fix it */
1276                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1277                                 __func__, policy->cpu, policy->cur);
1278                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1279                                 CPUFREQ_RELATION_L);
1280
1281                         /*
1282                          * Reaching here after boot in a few seconds may not
1283                          * mean that system will remain stable at "unknown"
1284                          * frequency for longer duration. Hence, a BUG_ON().
1285                          */
1286                         BUG_ON(ret);
1287                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1288                                 __func__, policy->cpu, policy->cur);
1289                 }
1290         }
1291
1292         if (new_policy) {
1293                 ret = cpufreq_add_dev_interface(policy);
1294                 if (ret)
1295                         goto out_destroy_policy;
1296
1297                 cpufreq_stats_create_table(policy);
1298
1299                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1300                 list_add(&policy->policy_list, &cpufreq_policy_list);
1301                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1302         }
1303
1304         ret = cpufreq_init_policy(policy);
1305         if (ret) {
1306                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1307                        __func__, cpu, ret);
1308                 /* cpufreq_policy_free() will notify based on this */
1309                 new_policy = false;
1310                 goto out_destroy_policy;
1311         }
1312
1313         up_write(&policy->rwsem);
1314
1315         kobject_uevent(&policy->kobj, KOBJ_ADD);
1316
1317         /* Callback for handling stuff after policy is ready */
1318         if (cpufreq_driver->ready)
1319                 cpufreq_driver->ready(policy);
1320
1321         pr_debug("initialization complete\n");
1322
1323         return 0;
1324
1325 out_destroy_policy:
1326         for_each_cpu(j, policy->real_cpus)
1327                 remove_cpu_dev_symlink(policy, get_cpu_device(j));
1328
1329         up_write(&policy->rwsem);
1330
1331 out_exit_policy:
1332         if (cpufreq_driver->exit)
1333                 cpufreq_driver->exit(policy);
1334
1335 out_free_policy:
1336         cpufreq_policy_free(policy);
1337         return ret;
1338 }
1339
1340 /**
1341  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1342  * @dev: CPU device.
1343  * @sif: Subsystem interface structure pointer (not used)
1344  */
1345 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1346 {
1347         struct cpufreq_policy *policy;
1348         unsigned cpu = dev->id;
1349         int ret;
1350
1351         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1352
1353         if (cpu_online(cpu)) {
1354                 ret = cpufreq_online(cpu);
1355                 if (ret)
1356                         return ret;
1357         }
1358
1359         /* Create sysfs link on CPU registration */
1360         policy = per_cpu(cpufreq_cpu_data, cpu);
1361         if (policy)
1362                 add_cpu_dev_symlink(policy, cpu);
1363
1364         return 0;
1365 }
1366
1367 static int cpufreq_offline(unsigned int cpu)
1368 {
1369         struct cpufreq_policy *policy;
1370         int ret;
1371
1372         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1373
1374         policy = cpufreq_cpu_get_raw(cpu);
1375         if (!policy) {
1376                 pr_debug("%s: No cpu_data found\n", __func__);
1377                 return 0;
1378         }
1379
1380         down_write(&policy->rwsem);
1381         if (has_target())
1382                 cpufreq_stop_governor(policy);
1383
1384         cpumask_clear_cpu(cpu, policy->cpus);
1385
1386         if (policy_is_inactive(policy)) {
1387                 if (has_target())
1388                         strncpy(policy->last_governor, policy->governor->name,
1389                                 CPUFREQ_NAME_LEN);
1390                 else
1391                         policy->last_policy = policy->policy;
1392         } else if (cpu == policy->cpu) {
1393                 /* Nominate new CPU */
1394                 policy->cpu = cpumask_any(policy->cpus);
1395         }
1396
1397         /* Start governor again for active policy */
1398         if (!policy_is_inactive(policy)) {
1399                 if (has_target()) {
1400                         ret = cpufreq_start_governor(policy);
1401                         if (ret)
1402                                 pr_err("%s: Failed to start governor\n", __func__);
1403                 }
1404
1405                 goto unlock;
1406         }
1407
1408         if (cpufreq_driver->stop_cpu)
1409                 cpufreq_driver->stop_cpu(policy);
1410
1411         if (has_target())
1412                 cpufreq_exit_governor(policy);
1413
1414         /*
1415          * Perform the ->exit() even during light-weight tear-down,
1416          * since this is a core component, and is essential for the
1417          * subsequent light-weight ->init() to succeed.
1418          */
1419         if (cpufreq_driver->exit) {
1420                 cpufreq_driver->exit(policy);
1421                 policy->freq_table = NULL;
1422         }
1423
1424 unlock:
1425         up_write(&policy->rwsem);
1426         return 0;
1427 }
1428
1429 /**
1430  * cpufreq_remove_dev - remove a CPU device
1431  *
1432  * Removes the cpufreq interface for a CPU device.
1433  */
1434 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1435 {
1436         unsigned int cpu = dev->id;
1437         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1438
1439         if (!policy)
1440                 return;
1441
1442         if (cpu_online(cpu))
1443                 cpufreq_offline(cpu);
1444
1445         cpumask_clear_cpu(cpu, policy->real_cpus);
1446         remove_cpu_dev_symlink(policy, dev);
1447
1448         if (cpumask_empty(policy->real_cpus))
1449                 cpufreq_policy_free(policy);
1450 }
1451
1452 /**
1453  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1454  *      in deep trouble.
1455  *      @policy: policy managing CPUs
1456  *      @new_freq: CPU frequency the CPU actually runs at
1457  *
1458  *      We adjust to current frequency first, and need to clean up later.
1459  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1460  */
1461 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1462                                 unsigned int new_freq)
1463 {
1464         struct cpufreq_freqs freqs;
1465
1466         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1467                  policy->cur, new_freq);
1468
1469         freqs.old = policy->cur;
1470         freqs.new = new_freq;
1471
1472         cpufreq_freq_transition_begin(policy, &freqs);
1473         cpufreq_freq_transition_end(policy, &freqs, 0);
1474 }
1475
1476 /**
1477  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1478  * @cpu: CPU number
1479  *
1480  * This is the last known freq, without actually getting it from the driver.
1481  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1482  */
1483 unsigned int cpufreq_quick_get(unsigned int cpu)
1484 {
1485         struct cpufreq_policy *policy;
1486         unsigned int ret_freq = 0;
1487         unsigned long flags;
1488
1489         read_lock_irqsave(&cpufreq_driver_lock, flags);
1490
1491         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1492                 ret_freq = cpufreq_driver->get(cpu);
1493                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1494                 return ret_freq;
1495         }
1496
1497         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1498
1499         policy = cpufreq_cpu_get(cpu);
1500         if (policy) {
1501                 ret_freq = policy->cur;
1502                 cpufreq_cpu_put(policy);
1503         }
1504
1505         return ret_freq;
1506 }
1507 EXPORT_SYMBOL(cpufreq_quick_get);
1508
1509 /**
1510  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1511  * @cpu: CPU number
1512  *
1513  * Just return the max possible frequency for a given CPU.
1514  */
1515 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1516 {
1517         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1518         unsigned int ret_freq = 0;
1519
1520         if (policy) {
1521                 ret_freq = policy->max;
1522                 cpufreq_cpu_put(policy);
1523         }
1524
1525         return ret_freq;
1526 }
1527 EXPORT_SYMBOL(cpufreq_quick_get_max);
1528
1529 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1530 {
1531         unsigned int ret_freq = 0;
1532
1533         if (unlikely(policy_is_inactive(policy)) || !cpufreq_driver->get)
1534                 return ret_freq;
1535
1536         ret_freq = cpufreq_driver->get(policy->cpu);
1537
1538         /*
1539          * If fast frequency switching is used with the given policy, the check
1540          * against policy->cur is pointless, so skip it in that case too.
1541          */
1542         if (policy->fast_switch_enabled)
1543                 return ret_freq;
1544
1545         if (ret_freq && policy->cur &&
1546                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1547                 /* verify no discrepancy between actual and
1548                                         saved value exists */
1549                 if (unlikely(ret_freq != policy->cur)) {
1550                         cpufreq_out_of_sync(policy, ret_freq);
1551                         schedule_work(&policy->update);
1552                 }
1553         }
1554
1555         return ret_freq;
1556 }
1557
1558 /**
1559  * cpufreq_get - get the current CPU frequency (in kHz)
1560  * @cpu: CPU number
1561  *
1562  * Get the CPU current (static) CPU frequency
1563  */
1564 unsigned int cpufreq_get(unsigned int cpu)
1565 {
1566         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1567         unsigned int ret_freq = 0;
1568
1569         if (policy) {
1570                 down_read(&policy->rwsem);
1571                 ret_freq = __cpufreq_get(policy);
1572                 up_read(&policy->rwsem);
1573
1574                 cpufreq_cpu_put(policy);
1575         }
1576
1577         return ret_freq;
1578 }
1579 EXPORT_SYMBOL(cpufreq_get);
1580
1581 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1582 {
1583         unsigned int new_freq;
1584
1585         new_freq = cpufreq_driver->get(policy->cpu);
1586         if (!new_freq)
1587                 return 0;
1588
1589         if (!policy->cur) {
1590                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1591                 policy->cur = new_freq;
1592         } else if (policy->cur != new_freq && has_target()) {
1593                 cpufreq_out_of_sync(policy, new_freq);
1594         }
1595
1596         return new_freq;
1597 }
1598
1599 static struct subsys_interface cpufreq_interface = {
1600         .name           = "cpufreq",
1601         .subsys         = &cpu_subsys,
1602         .add_dev        = cpufreq_add_dev,
1603         .remove_dev     = cpufreq_remove_dev,
1604 };
1605
1606 /*
1607  * In case platform wants some specific frequency to be configured
1608  * during suspend..
1609  */
1610 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1611 {
1612         int ret;
1613
1614         if (!policy->suspend_freq) {
1615                 pr_debug("%s: suspend_freq not defined\n", __func__);
1616                 return 0;
1617         }
1618
1619         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1620                         policy->suspend_freq);
1621
1622         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1623                         CPUFREQ_RELATION_H);
1624         if (ret)
1625                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1626                                 __func__, policy->suspend_freq, ret);
1627
1628         return ret;
1629 }
1630 EXPORT_SYMBOL(cpufreq_generic_suspend);
1631
1632 /**
1633  * cpufreq_suspend() - Suspend CPUFreq governors
1634  *
1635  * Called during system wide Suspend/Hibernate cycles for suspending governors
1636  * as some platforms can't change frequency after this point in suspend cycle.
1637  * Because some of the devices (like: i2c, regulators, etc) they use for
1638  * changing frequency are suspended quickly after this point.
1639  */
1640 void cpufreq_suspend(void)
1641 {
1642         struct cpufreq_policy *policy;
1643
1644         if (!cpufreq_driver)
1645                 return;
1646
1647         if (!has_target() && !cpufreq_driver->suspend)
1648                 goto suspend;
1649
1650         pr_debug("%s: Suspending Governors\n", __func__);
1651
1652         for_each_active_policy(policy) {
1653                 if (has_target()) {
1654                         down_write(&policy->rwsem);
1655                         cpufreq_stop_governor(policy);
1656                         up_write(&policy->rwsem);
1657                 }
1658
1659                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1660                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1661                                 policy);
1662         }
1663
1664 suspend:
1665         cpufreq_suspended = true;
1666 }
1667
1668 /**
1669  * cpufreq_resume() - Resume CPUFreq governors
1670  *
1671  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1672  * are suspended with cpufreq_suspend().
1673  */
1674 void cpufreq_resume(void)
1675 {
1676         struct cpufreq_policy *policy;
1677         int ret;
1678
1679         if (!cpufreq_driver)
1680                 return;
1681
1682         if (unlikely(!cpufreq_suspended))
1683                 return;
1684
1685         cpufreq_suspended = false;
1686
1687         if (!has_target() && !cpufreq_driver->resume)
1688                 return;
1689
1690         pr_debug("%s: Resuming Governors\n", __func__);
1691
1692         for_each_active_policy(policy) {
1693                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1694                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1695                                 policy);
1696                 } else if (has_target()) {
1697                         down_write(&policy->rwsem);
1698                         ret = cpufreq_start_governor(policy);
1699                         up_write(&policy->rwsem);
1700
1701                         if (ret)
1702                                 pr_err("%s: Failed to start governor for policy: %p\n",
1703                                        __func__, policy);
1704                 }
1705         }
1706 }
1707
1708 /**
1709  *      cpufreq_get_current_driver - return current driver's name
1710  *
1711  *      Return the name string of the currently loaded cpufreq driver
1712  *      or NULL, if none.
1713  */
1714 const char *cpufreq_get_current_driver(void)
1715 {
1716         if (cpufreq_driver)
1717                 return cpufreq_driver->name;
1718
1719         return NULL;
1720 }
1721 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1722
1723 /**
1724  *      cpufreq_get_driver_data - return current driver data
1725  *
1726  *      Return the private data of the currently loaded cpufreq
1727  *      driver, or NULL if no cpufreq driver is loaded.
1728  */
1729 void *cpufreq_get_driver_data(void)
1730 {
1731         if (cpufreq_driver)
1732                 return cpufreq_driver->driver_data;
1733
1734         return NULL;
1735 }
1736 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1737
1738 /*********************************************************************
1739  *                     NOTIFIER LISTS INTERFACE                      *
1740  *********************************************************************/
1741
1742 /**
1743  *      cpufreq_register_notifier - register a driver with cpufreq
1744  *      @nb: notifier function to register
1745  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1746  *
1747  *      Add a driver to one of two lists: either a list of drivers that
1748  *      are notified about clock rate changes (once before and once after
1749  *      the transition), or a list of drivers that are notified about
1750  *      changes in cpufreq policy.
1751  *
1752  *      This function may sleep, and has the same return conditions as
1753  *      blocking_notifier_chain_register.
1754  */
1755 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1756 {
1757         int ret;
1758
1759         if (cpufreq_disabled())
1760                 return -EINVAL;
1761
1762         switch (list) {
1763         case CPUFREQ_TRANSITION_NOTIFIER:
1764                 mutex_lock(&cpufreq_fast_switch_lock);
1765
1766                 if (cpufreq_fast_switch_count > 0) {
1767                         mutex_unlock(&cpufreq_fast_switch_lock);
1768                         return -EBUSY;
1769                 }
1770                 ret = srcu_notifier_chain_register(
1771                                 &cpufreq_transition_notifier_list, nb);
1772                 if (!ret)
1773                         cpufreq_fast_switch_count--;
1774
1775                 mutex_unlock(&cpufreq_fast_switch_lock);
1776                 break;
1777         case CPUFREQ_POLICY_NOTIFIER:
1778                 ret = blocking_notifier_chain_register(
1779                                 &cpufreq_policy_notifier_list, nb);
1780                 break;
1781         default:
1782                 ret = -EINVAL;
1783         }
1784
1785         return ret;
1786 }
1787 EXPORT_SYMBOL(cpufreq_register_notifier);
1788
1789 /**
1790  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1791  *      @nb: notifier block to be unregistered
1792  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1793  *
1794  *      Remove a driver from the CPU frequency notifier list.
1795  *
1796  *      This function may sleep, and has the same return conditions as
1797  *      blocking_notifier_chain_unregister.
1798  */
1799 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1800 {
1801         int ret;
1802
1803         if (cpufreq_disabled())
1804                 return -EINVAL;
1805
1806         switch (list) {
1807         case CPUFREQ_TRANSITION_NOTIFIER:
1808                 mutex_lock(&cpufreq_fast_switch_lock);
1809
1810                 ret = srcu_notifier_chain_unregister(
1811                                 &cpufreq_transition_notifier_list, nb);
1812                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1813                         cpufreq_fast_switch_count++;
1814
1815                 mutex_unlock(&cpufreq_fast_switch_lock);
1816                 break;
1817         case CPUFREQ_POLICY_NOTIFIER:
1818                 ret = blocking_notifier_chain_unregister(
1819                                 &cpufreq_policy_notifier_list, nb);
1820                 break;
1821         default:
1822                 ret = -EINVAL;
1823         }
1824
1825         return ret;
1826 }
1827 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1828
1829
1830 /*********************************************************************
1831  *                              GOVERNORS                            *
1832  *********************************************************************/
1833
1834 /**
1835  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1836  * @policy: cpufreq policy to switch the frequency for.
1837  * @target_freq: New frequency to set (may be approximate).
1838  *
1839  * Carry out a fast frequency switch without sleeping.
1840  *
1841  * The driver's ->fast_switch() callback invoked by this function must be
1842  * suitable for being called from within RCU-sched read-side critical sections
1843  * and it is expected to select the minimum available frequency greater than or
1844  * equal to @target_freq (CPUFREQ_RELATION_L).
1845  *
1846  * This function must not be called if policy->fast_switch_enabled is unset.
1847  *
1848  * Governors calling this function must guarantee that it will never be invoked
1849  * twice in parallel for the same policy and that it will never be called in
1850  * parallel with either ->target() or ->target_index() for the same policy.
1851  *
1852  * Returns the actual frequency set for the CPU.
1853  *
1854  * If 0 is returned by the driver's ->fast_switch() callback to indicate an
1855  * error condition, the hardware configuration must be preserved.
1856  */
1857 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1858                                         unsigned int target_freq)
1859 {
1860         target_freq = clamp_val(target_freq, policy->min, policy->max);
1861
1862         return cpufreq_driver->fast_switch(policy, target_freq);
1863 }
1864 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1865
1866 /* Must set freqs->new to intermediate frequency */
1867 static int __target_intermediate(struct cpufreq_policy *policy,
1868                                  struct cpufreq_freqs *freqs, int index)
1869 {
1870         int ret;
1871
1872         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1873
1874         /* We don't need to switch to intermediate freq */
1875         if (!freqs->new)
1876                 return 0;
1877
1878         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1879                  __func__, policy->cpu, freqs->old, freqs->new);
1880
1881         cpufreq_freq_transition_begin(policy, freqs);
1882         ret = cpufreq_driver->target_intermediate(policy, index);
1883         cpufreq_freq_transition_end(policy, freqs, ret);
1884
1885         if (ret)
1886                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1887                        __func__, ret);
1888
1889         return ret;
1890 }
1891
1892 static int __target_index(struct cpufreq_policy *policy, int index)
1893 {
1894         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1895         unsigned int intermediate_freq = 0;
1896         unsigned int newfreq = policy->freq_table[index].frequency;
1897         int retval = -EINVAL;
1898         bool notify;
1899
1900         if (newfreq == policy->cur)
1901                 return 0;
1902
1903         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1904         if (notify) {
1905                 /* Handle switching to intermediate frequency */
1906                 if (cpufreq_driver->get_intermediate) {
1907                         retval = __target_intermediate(policy, &freqs, index);
1908                         if (retval)
1909                                 return retval;
1910
1911                         intermediate_freq = freqs.new;
1912                         /* Set old freq to intermediate */
1913                         if (intermediate_freq)
1914                                 freqs.old = freqs.new;
1915                 }
1916
1917                 freqs.new = newfreq;
1918                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1919                          __func__, policy->cpu, freqs.old, freqs.new);
1920
1921                 cpufreq_freq_transition_begin(policy, &freqs);
1922         }
1923
1924         retval = cpufreq_driver->target_index(policy, index);
1925         if (retval)
1926                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1927                        retval);
1928
1929         if (notify) {
1930                 cpufreq_freq_transition_end(policy, &freqs, retval);
1931
1932                 /*
1933                  * Failed after setting to intermediate freq? Driver should have
1934                  * reverted back to initial frequency and so should we. Check
1935                  * here for intermediate_freq instead of get_intermediate, in
1936                  * case we haven't switched to intermediate freq at all.
1937                  */
1938                 if (unlikely(retval && intermediate_freq)) {
1939                         freqs.old = intermediate_freq;
1940                         freqs.new = policy->restore_freq;
1941                         cpufreq_freq_transition_begin(policy, &freqs);
1942                         cpufreq_freq_transition_end(policy, &freqs, 0);
1943                 }
1944         }
1945
1946         return retval;
1947 }
1948
1949 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1950                             unsigned int target_freq,
1951                             unsigned int relation)
1952 {
1953         unsigned int old_target_freq = target_freq;
1954         int index;
1955
1956         if (cpufreq_disabled())
1957                 return -ENODEV;
1958
1959         /* Make sure that target_freq is within supported range */
1960         target_freq = clamp_val(target_freq, policy->min, policy->max);
1961
1962         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1963                  policy->cpu, target_freq, relation, old_target_freq);
1964
1965         /*
1966          * This might look like a redundant call as we are checking it again
1967          * after finding index. But it is left intentionally for cases where
1968          * exactly same freq is called again and so we can save on few function
1969          * calls.
1970          */
1971         if (target_freq == policy->cur)
1972                 return 0;
1973
1974         /* Save last value to restore later on errors */
1975         policy->restore_freq = policy->cur;
1976
1977         if (cpufreq_driver->target)
1978                 return cpufreq_driver->target(policy, target_freq, relation);
1979
1980         if (!cpufreq_driver->target_index)
1981                 return -EINVAL;
1982
1983         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1984
1985         return __target_index(policy, index);
1986 }
1987 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1988
1989 int cpufreq_driver_target(struct cpufreq_policy *policy,
1990                           unsigned int target_freq,
1991                           unsigned int relation)
1992 {
1993         int ret = -EINVAL;
1994
1995         down_write(&policy->rwsem);
1996
1997         ret = __cpufreq_driver_target(policy, target_freq, relation);
1998
1999         up_write(&policy->rwsem);
2000
2001         return ret;
2002 }
2003 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2004
2005 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2006 {
2007         return NULL;
2008 }
2009
2010 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2011 {
2012         int ret;
2013
2014         /* Don't start any governor operations if we are entering suspend */
2015         if (cpufreq_suspended)
2016                 return 0;
2017         /*
2018          * Governor might not be initiated here if ACPI _PPC changed
2019          * notification happened, so check it.
2020          */
2021         if (!policy->governor)
2022                 return -EINVAL;
2023
2024         /* Platform doesn't want dynamic frequency switching ? */
2025         if (policy->governor->dynamic_switching &&
2026             cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2027                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2028
2029                 if (gov) {
2030                         pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2031                                 policy->governor->name, gov->name);
2032                         policy->governor = gov;
2033                 } else {
2034                         return -EINVAL;
2035                 }
2036         }
2037
2038         if (!try_module_get(policy->governor->owner))
2039                 return -EINVAL;
2040
2041         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2042
2043         if (policy->governor->init) {
2044                 ret = policy->governor->init(policy);
2045                 if (ret) {
2046                         module_put(policy->governor->owner);
2047                         return ret;
2048                 }
2049         }
2050
2051         return 0;
2052 }
2053
2054 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2055 {
2056         if (cpufreq_suspended || !policy->governor)
2057                 return;
2058
2059         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2060
2061         if (policy->governor->exit)
2062                 policy->governor->exit(policy);
2063
2064         module_put(policy->governor->owner);
2065 }
2066
2067 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2068 {
2069         int ret;
2070
2071         if (cpufreq_suspended)
2072                 return 0;
2073
2074         if (!policy->governor)
2075                 return -EINVAL;
2076
2077         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2078
2079         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2080                 cpufreq_update_current_freq(policy);
2081
2082         if (policy->governor->start) {
2083                 ret = policy->governor->start(policy);
2084                 if (ret)
2085                         return ret;
2086         }
2087
2088         if (policy->governor->limits)
2089                 policy->governor->limits(policy);
2090
2091         return 0;
2092 }
2093
2094 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2095 {
2096         if (cpufreq_suspended || !policy->governor)
2097                 return;
2098
2099         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2100
2101         if (policy->governor->stop)
2102                 policy->governor->stop(policy);
2103 }
2104
2105 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2106 {
2107         if (cpufreq_suspended || !policy->governor)
2108                 return;
2109
2110         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2111
2112         if (policy->governor->limits)
2113                 policy->governor->limits(policy);
2114 }
2115
2116 int cpufreq_register_governor(struct cpufreq_governor *governor)
2117 {
2118         int err;
2119
2120         if (!governor)
2121                 return -EINVAL;
2122
2123         if (cpufreq_disabled())
2124                 return -ENODEV;
2125
2126         mutex_lock(&cpufreq_governor_mutex);
2127
2128         err = -EBUSY;
2129         if (!find_governor(governor->name)) {
2130                 err = 0;
2131                 list_add(&governor->governor_list, &cpufreq_governor_list);
2132         }
2133
2134         mutex_unlock(&cpufreq_governor_mutex);
2135         return err;
2136 }
2137 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2138
2139 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2140 {
2141         struct cpufreq_policy *policy;
2142         unsigned long flags;
2143
2144         if (!governor)
2145                 return;
2146
2147         if (cpufreq_disabled())
2148                 return;
2149
2150         /* clear last_governor for all inactive policies */
2151         read_lock_irqsave(&cpufreq_driver_lock, flags);
2152         for_each_inactive_policy(policy) {
2153                 if (!strcmp(policy->last_governor, governor->name)) {
2154                         policy->governor = NULL;
2155                         strcpy(policy->last_governor, "\0");
2156                 }
2157         }
2158         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2159
2160         mutex_lock(&cpufreq_governor_mutex);
2161         list_del(&governor->governor_list);
2162         mutex_unlock(&cpufreq_governor_mutex);
2163 }
2164 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2165
2166
2167 /*********************************************************************
2168  *                          POLICY INTERFACE                         *
2169  *********************************************************************/
2170
2171 /**
2172  * cpufreq_get_policy - get the current cpufreq_policy
2173  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2174  *      is written
2175  *
2176  * Reads the current cpufreq policy.
2177  */
2178 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2179 {
2180         struct cpufreq_policy *cpu_policy;
2181         if (!policy)
2182                 return -EINVAL;
2183
2184         cpu_policy = cpufreq_cpu_get(cpu);
2185         if (!cpu_policy)
2186                 return -EINVAL;
2187
2188         memcpy(policy, cpu_policy, sizeof(*policy));
2189
2190         cpufreq_cpu_put(cpu_policy);
2191         return 0;
2192 }
2193 EXPORT_SYMBOL(cpufreq_get_policy);
2194
2195 /*
2196  * policy : current policy.
2197  * new_policy: policy to be set.
2198  */
2199 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2200                                 struct cpufreq_policy *new_policy)
2201 {
2202         struct cpufreq_governor *old_gov;
2203         int ret;
2204
2205         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2206                  new_policy->cpu, new_policy->min, new_policy->max);
2207
2208         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2209
2210         /*
2211         * This check works well when we store new min/max freq attributes,
2212         * because new_policy is a copy of policy with one field updated.
2213         */
2214         if (new_policy->min > new_policy->max)
2215                 return -EINVAL;
2216
2217         /* verify the cpu speed can be set within this limit */
2218         ret = cpufreq_driver->verify(new_policy);
2219         if (ret)
2220                 return ret;
2221
2222         /* adjust if necessary - all reasons */
2223         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2224                         CPUFREQ_ADJUST, new_policy);
2225
2226         /*
2227          * verify the cpu speed can be set within this limit, which might be
2228          * different to the first one
2229          */
2230         ret = cpufreq_driver->verify(new_policy);
2231         if (ret)
2232                 return ret;
2233
2234         /* notification of the new policy */
2235         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2236                         CPUFREQ_NOTIFY, new_policy);
2237
2238         policy->min = new_policy->min;
2239         policy->max = new_policy->max;
2240         trace_cpu_frequency_limits(policy);
2241
2242         policy->cached_target_freq = UINT_MAX;
2243
2244         pr_debug("new min and max freqs are %u - %u kHz\n",
2245                  policy->min, policy->max);
2246
2247         if (cpufreq_driver->setpolicy) {
2248                 policy->policy = new_policy->policy;
2249                 pr_debug("setting range\n");
2250                 return cpufreq_driver->setpolicy(new_policy);
2251         }
2252
2253         if (new_policy->governor == policy->governor) {
2254                 pr_debug("cpufreq: governor limits update\n");
2255                 cpufreq_governor_limits(policy);
2256                 return 0;
2257         }
2258
2259         pr_debug("governor switch\n");
2260
2261         /* save old, working values */
2262         old_gov = policy->governor;
2263         /* end old governor */
2264         if (old_gov) {
2265                 cpufreq_stop_governor(policy);
2266                 cpufreq_exit_governor(policy);
2267         }
2268
2269         /* start new governor */
2270         policy->governor = new_policy->governor;
2271         ret = cpufreq_init_governor(policy);
2272         if (!ret) {
2273                 ret = cpufreq_start_governor(policy);
2274                 if (!ret) {
2275                         pr_debug("cpufreq: governor change\n");
2276                         sched_cpufreq_governor_change(policy, old_gov);
2277                         return 0;
2278                 }
2279                 cpufreq_exit_governor(policy);
2280         }
2281
2282         /* new governor failed, so re-start old one */
2283         pr_debug("starting governor %s failed\n", policy->governor->name);
2284         if (old_gov) {
2285                 policy->governor = old_gov;
2286                 if (cpufreq_init_governor(policy))
2287                         policy->governor = NULL;
2288                 else
2289                         cpufreq_start_governor(policy);
2290         }
2291
2292         return ret;
2293 }
2294
2295 /**
2296  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2297  *      @cpu: CPU which shall be re-evaluated
2298  *
2299  *      Useful for policy notifiers which have different necessities
2300  *      at different times.
2301  */
2302 void cpufreq_update_policy(unsigned int cpu)
2303 {
2304         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2305         struct cpufreq_policy new_policy;
2306
2307         if (!policy)
2308                 return;
2309
2310         down_write(&policy->rwsem);
2311
2312         if (policy_is_inactive(policy))
2313                 goto unlock;
2314
2315         pr_debug("updating policy for CPU %u\n", cpu);
2316         memcpy(&new_policy, policy, sizeof(*policy));
2317         new_policy.min = policy->user_policy.min;
2318         new_policy.max = policy->user_policy.max;
2319
2320         /*
2321          * BIOS might change freq behind our back
2322          * -> ask driver for current freq and notify governors about a change
2323          */
2324         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2325                 if (cpufreq_suspended)
2326                         goto unlock;
2327
2328                 new_policy.cur = cpufreq_update_current_freq(policy);
2329                 if (WARN_ON(!new_policy.cur))
2330                         goto unlock;
2331         }
2332
2333         cpufreq_set_policy(policy, &new_policy);
2334
2335 unlock:
2336         up_write(&policy->rwsem);
2337
2338         cpufreq_cpu_put(policy);
2339 }
2340 EXPORT_SYMBOL(cpufreq_update_policy);
2341
2342 /*********************************************************************
2343  *               BOOST                                               *
2344  *********************************************************************/
2345 static int cpufreq_boost_set_sw(int state)
2346 {
2347         struct cpufreq_policy *policy;
2348         int ret = -EINVAL;
2349
2350         for_each_active_policy(policy) {
2351                 if (!policy->freq_table)
2352                         continue;
2353
2354                 ret = cpufreq_frequency_table_cpuinfo(policy,
2355                                                       policy->freq_table);
2356                 if (ret) {
2357                         pr_err("%s: Policy frequency update failed\n",
2358                                __func__);
2359                         break;
2360                 }
2361
2362                 down_write(&policy->rwsem);
2363                 policy->user_policy.max = policy->max;
2364                 cpufreq_governor_limits(policy);
2365                 up_write(&policy->rwsem);
2366         }
2367
2368         return ret;
2369 }
2370
2371 int cpufreq_boost_trigger_state(int state)
2372 {
2373         unsigned long flags;
2374         int ret = 0;
2375
2376         if (cpufreq_driver->boost_enabled == state)
2377                 return 0;
2378
2379         write_lock_irqsave(&cpufreq_driver_lock, flags);
2380         cpufreq_driver->boost_enabled = state;
2381         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2382
2383         ret = cpufreq_driver->set_boost(state);
2384         if (ret) {
2385                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2386                 cpufreq_driver->boost_enabled = !state;
2387                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2388
2389                 pr_err("%s: Cannot %s BOOST\n",
2390                        __func__, state ? "enable" : "disable");
2391         }
2392
2393         return ret;
2394 }
2395
2396 static bool cpufreq_boost_supported(void)
2397 {
2398         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2399 }
2400
2401 static int create_boost_sysfs_file(void)
2402 {
2403         int ret;
2404
2405         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2406         if (ret)
2407                 pr_err("%s: cannot register global BOOST sysfs file\n",
2408                        __func__);
2409
2410         return ret;
2411 }
2412
2413 static void remove_boost_sysfs_file(void)
2414 {
2415         if (cpufreq_boost_supported())
2416                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2417 }
2418
2419 int cpufreq_enable_boost_support(void)
2420 {
2421         if (!cpufreq_driver)
2422                 return -EINVAL;
2423
2424         if (cpufreq_boost_supported())
2425                 return 0;
2426
2427         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2428
2429         /* This will get removed on driver unregister */
2430         return create_boost_sysfs_file();
2431 }
2432 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2433
2434 int cpufreq_boost_enabled(void)
2435 {
2436         return cpufreq_driver->boost_enabled;
2437 }
2438 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2439
2440 /*********************************************************************
2441  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2442  *********************************************************************/
2443 static enum cpuhp_state hp_online;
2444
2445 static int cpuhp_cpufreq_online(unsigned int cpu)
2446 {
2447         cpufreq_online(cpu);
2448
2449         return 0;
2450 }
2451
2452 static int cpuhp_cpufreq_offline(unsigned int cpu)
2453 {
2454         cpufreq_offline(cpu);
2455
2456         return 0;
2457 }
2458
2459 /**
2460  * cpufreq_register_driver - register a CPU Frequency driver
2461  * @driver_data: A struct cpufreq_driver containing the values#
2462  * submitted by the CPU Frequency driver.
2463  *
2464  * Registers a CPU Frequency driver to this core code. This code
2465  * returns zero on success, -EEXIST when another driver got here first
2466  * (and isn't unregistered in the meantime).
2467  *
2468  */
2469 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2470 {
2471         unsigned long flags;
2472         int ret;
2473
2474         if (cpufreq_disabled())
2475                 return -ENODEV;
2476
2477         if (!driver_data || !driver_data->verify || !driver_data->init ||
2478             !(driver_data->setpolicy || driver_data->target_index ||
2479                     driver_data->target) ||
2480              (driver_data->setpolicy && (driver_data->target_index ||
2481                     driver_data->target)) ||
2482              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2483                 return -EINVAL;
2484
2485         pr_debug("trying to register driver %s\n", driver_data->name);
2486
2487         /* Protect against concurrent CPU online/offline. */
2488         cpus_read_lock();
2489
2490         write_lock_irqsave(&cpufreq_driver_lock, flags);
2491         if (cpufreq_driver) {
2492                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2493                 ret = -EEXIST;
2494                 goto out;
2495         }
2496         cpufreq_driver = driver_data;
2497         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2498
2499         if (driver_data->setpolicy)
2500                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2501
2502         if (cpufreq_boost_supported()) {
2503                 ret = create_boost_sysfs_file();
2504                 if (ret)
2505                         goto err_null_driver;
2506         }
2507
2508         ret = subsys_interface_register(&cpufreq_interface);
2509         if (ret)
2510                 goto err_boost_unreg;
2511
2512         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2513             list_empty(&cpufreq_policy_list)) {
2514                 /* if all ->init() calls failed, unregister */
2515                 ret = -ENODEV;
2516                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2517                          driver_data->name);
2518                 goto err_if_unreg;
2519         }
2520
2521         ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2522                                                    "cpufreq:online",
2523                                                    cpuhp_cpufreq_online,
2524                                                    cpuhp_cpufreq_offline);
2525         if (ret < 0)
2526                 goto err_if_unreg;
2527         hp_online = ret;
2528         ret = 0;
2529
2530         pr_debug("driver %s up and running\n", driver_data->name);
2531         goto out;
2532
2533 err_if_unreg:
2534         subsys_interface_unregister(&cpufreq_interface);
2535 err_boost_unreg:
2536         remove_boost_sysfs_file();
2537 err_null_driver:
2538         write_lock_irqsave(&cpufreq_driver_lock, flags);
2539         cpufreq_driver = NULL;
2540         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2541 out:
2542         cpus_read_unlock();
2543         return ret;
2544 }
2545 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2546
2547 /**
2548  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2549  *
2550  * Unregister the current CPUFreq driver. Only call this if you have
2551  * the right to do so, i.e. if you have succeeded in initialising before!
2552  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2553  * currently not initialised.
2554  */
2555 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2556 {
2557         unsigned long flags;
2558
2559         if (!cpufreq_driver || (driver != cpufreq_driver))
2560                 return -EINVAL;
2561
2562         pr_debug("unregistering driver %s\n", driver->name);
2563
2564         /* Protect against concurrent cpu hotplug */
2565         cpus_read_lock();
2566         subsys_interface_unregister(&cpufreq_interface);
2567         remove_boost_sysfs_file();
2568         cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2569
2570         write_lock_irqsave(&cpufreq_driver_lock, flags);
2571
2572         cpufreq_driver = NULL;
2573
2574         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2575         cpus_read_unlock();
2576
2577         return 0;
2578 }
2579 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2580
2581 /*
2582  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2583  * or mutexes when secondary CPUs are halted.
2584  */
2585 static struct syscore_ops cpufreq_syscore_ops = {
2586         .shutdown = cpufreq_suspend,
2587 };
2588
2589 struct kobject *cpufreq_global_kobject;
2590 EXPORT_SYMBOL(cpufreq_global_kobject);
2591
2592 static int __init cpufreq_core_init(void)
2593 {
2594         if (cpufreq_disabled())
2595                 return -ENODEV;
2596
2597         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2598         BUG_ON(!cpufreq_global_kobject);
2599
2600         register_syscore_ops(&cpufreq_syscore_ops);
2601
2602         return 0;
2603 }
2604 module_param(off, int, 0444);
2605 core_initcall(cpufreq_core_init);