Merge tag 'modules-for-v4.13' of git://git.kernel.org/pub/scm/linux/kernel/git/jeyu...
[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 static struct srcu_notifier_head cpufreq_transition_notifier_list;
93
94 static bool init_cpufreq_transition_notifier_list_called;
95 static int __init init_cpufreq_transition_notifier_list(void)
96 {
97         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
98         init_cpufreq_transition_notifier_list_called = true;
99         return 0;
100 }
101 pure_initcall(init_cpufreq_transition_notifier_list);
102
103 static int off __read_mostly;
104 static int cpufreq_disabled(void)
105 {
106         return off;
107 }
108 void disable_cpufreq(void)
109 {
110         off = 1;
111 }
112 static DEFINE_MUTEX(cpufreq_governor_mutex);
113
114 bool have_governor_per_policy(void)
115 {
116         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
117 }
118 EXPORT_SYMBOL_GPL(have_governor_per_policy);
119
120 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
121 {
122         if (have_governor_per_policy())
123                 return &policy->kobj;
124         else
125                 return cpufreq_global_kobject;
126 }
127 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
128
129 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
130 {
131         u64 idle_time;
132         u64 cur_wall_time;
133         u64 busy_time;
134
135         cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
136
137         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
138         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
139         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
140         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
141         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
142         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
143
144         idle_time = cur_wall_time - busy_time;
145         if (wall)
146                 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
147
148         return div_u64(idle_time, NSEC_PER_USEC);
149 }
150
151 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
152 {
153         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
154
155         if (idle_time == -1ULL)
156                 return get_cpu_idle_time_jiffy(cpu, wall);
157         else if (!io_busy)
158                 idle_time += get_cpu_iowait_time_us(cpu, wall);
159
160         return idle_time;
161 }
162 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
163
164 /*
165  * This is a generic cpufreq init() routine which can be used by cpufreq
166  * drivers of SMP systems. It will do following:
167  * - validate & show freq table passed
168  * - set policies transition latency
169  * - policy->cpus with all possible CPUs
170  */
171 int cpufreq_generic_init(struct cpufreq_policy *policy,
172                 struct cpufreq_frequency_table *table,
173                 unsigned int transition_latency)
174 {
175         int ret;
176
177         ret = cpufreq_table_validate_and_show(policy, table);
178         if (ret) {
179                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
180                 return ret;
181         }
182
183         policy->cpuinfo.transition_latency = transition_latency;
184
185         /*
186          * The driver only supports the SMP configuration where all processors
187          * share the clock and voltage and clock.
188          */
189         cpumask_setall(policy->cpus);
190
191         return 0;
192 }
193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
194
195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
196 {
197         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
198
199         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
200 }
201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
202
203 unsigned int cpufreq_generic_get(unsigned int cpu)
204 {
205         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
206
207         if (!policy || IS_ERR(policy->clk)) {
208                 pr_err("%s: No %s associated to cpu: %d\n",
209                        __func__, policy ? "clk" : "policy", cpu);
210                 return 0;
211         }
212
213         return clk_get_rate(policy->clk) / 1000;
214 }
215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
216
217 /**
218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
219  *
220  * @cpu: cpu to find policy for.
221  *
222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
223  * It also increments the kobject reference count to mark it busy and so would
224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
226  * freed as that depends on the kobj count.
227  *
228  * Return: A valid policy on success, otherwise NULL on failure.
229  */
230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
231 {
232         struct cpufreq_policy *policy = NULL;
233         unsigned long flags;
234
235         if (WARN_ON(cpu >= nr_cpu_ids))
236                 return NULL;
237
238         /* get the cpufreq driver */
239         read_lock_irqsave(&cpufreq_driver_lock, flags);
240
241         if (cpufreq_driver) {
242                 /* get the CPU */
243                 policy = cpufreq_cpu_get_raw(cpu);
244                 if (policy)
245                         kobject_get(&policy->kobj);
246         }
247
248         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
249
250         return policy;
251 }
252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
253
254 /**
255  * cpufreq_cpu_put: Decrements the usage count of a policy
256  *
257  * @policy: policy earlier returned by cpufreq_cpu_get().
258  *
259  * This decrements the kobject reference count incremented earlier by calling
260  * cpufreq_cpu_get().
261  */
262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
263 {
264         kobject_put(&policy->kobj);
265 }
266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
267
268 /*********************************************************************
269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
270  *********************************************************************/
271
272 /**
273  * adjust_jiffies - adjust the system "loops_per_jiffy"
274  *
275  * This function alters the system "loops_per_jiffy" for the clock
276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
277  * systems as each CPU might be scaled differently. So, use the arch
278  * per-CPU loops_per_jiffy value wherever possible.
279  */
280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
281 {
282 #ifndef CONFIG_SMP
283         static unsigned long l_p_j_ref;
284         static unsigned int l_p_j_ref_freq;
285
286         if (ci->flags & CPUFREQ_CONST_LOOPS)
287                 return;
288
289         if (!l_p_j_ref_freq) {
290                 l_p_j_ref = loops_per_jiffy;
291                 l_p_j_ref_freq = ci->old;
292                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
293                          l_p_j_ref, l_p_j_ref_freq);
294         }
295         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297                                                                 ci->new);
298                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
299                          loops_per_jiffy, ci->new);
300         }
301 #endif
302 }
303
304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
305                 struct cpufreq_freqs *freqs, unsigned int state)
306 {
307         BUG_ON(irqs_disabled());
308
309         if (cpufreq_disabled())
310                 return;
311
312         freqs->flags = cpufreq_driver->flags;
313         pr_debug("notification %u of frequency transition to %u kHz\n",
314                  state, freqs->new);
315
316         switch (state) {
317
318         case CPUFREQ_PRECHANGE:
319                 /* detect if the driver reported a value as "old frequency"
320                  * which is not equal to what the cpufreq core thinks is
321                  * "old frequency".
322                  */
323                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
324                         if ((policy) && (policy->cpu == freqs->cpu) &&
325                             (policy->cur) && (policy->cur != freqs->old)) {
326                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
327                                          freqs->old, policy->cur);
328                                 freqs->old = policy->cur;
329                         }
330                 }
331                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
332                                 CPUFREQ_PRECHANGE, freqs);
333                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
334                 break;
335
336         case CPUFREQ_POSTCHANGE:
337                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
338                 pr_debug("FREQ: %lu - CPU: %lu\n",
339                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
340                 trace_cpu_frequency(freqs->new, freqs->cpu);
341                 cpufreq_stats_record_transition(policy, freqs->new);
342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
343                                 CPUFREQ_POSTCHANGE, freqs);
344                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
345                         policy->cur = freqs->new;
346                 break;
347         }
348 }
349
350 /**
351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
352  * on frequency transition.
353  *
354  * This function calls the transition notifiers and the "adjust_jiffies"
355  * function. It is called twice on all CPU frequency changes that have
356  * external effects.
357  */
358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
359                 struct cpufreq_freqs *freqs, unsigned int state)
360 {
361         for_each_cpu(freqs->cpu, policy->cpus)
362                 __cpufreq_notify_transition(policy, freqs, state);
363 }
364
365 /* Do post notifications when there are chances that transition has failed */
366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
367                 struct cpufreq_freqs *freqs, int transition_failed)
368 {
369         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
370         if (!transition_failed)
371                 return;
372
373         swap(freqs->old, freqs->new);
374         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
376 }
377
378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
379                 struct cpufreq_freqs *freqs)
380 {
381
382         /*
383          * Catch double invocations of _begin() which lead to self-deadlock.
384          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
385          * doesn't invoke _begin() on their behalf, and hence the chances of
386          * double invocations are very low. Moreover, there are scenarios
387          * where these checks can emit false-positive warnings in these
388          * drivers; so we avoid that by skipping them altogether.
389          */
390         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
391                                 && current == policy->transition_task);
392
393 wait:
394         wait_event(policy->transition_wait, !policy->transition_ongoing);
395
396         spin_lock(&policy->transition_lock);
397
398         if (unlikely(policy->transition_ongoing)) {
399                 spin_unlock(&policy->transition_lock);
400                 goto wait;
401         }
402
403         policy->transition_ongoing = true;
404         policy->transition_task = current;
405
406         spin_unlock(&policy->transition_lock);
407
408         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
409 }
410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
411
412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
413                 struct cpufreq_freqs *freqs, int transition_failed)
414 {
415         if (unlikely(WARN_ON(!policy->transition_ongoing)))
416                 return;
417
418         cpufreq_notify_post_transition(policy, freqs, transition_failed);
419
420         policy->transition_ongoing = false;
421         policy->transition_task = NULL;
422
423         wake_up(&policy->transition_wait);
424 }
425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
426
427 /*
428  * Fast frequency switching status count.  Positive means "enabled", negative
429  * means "disabled" and 0 means "not decided yet".
430  */
431 static int cpufreq_fast_switch_count;
432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
433
434 static void cpufreq_list_transition_notifiers(void)
435 {
436         struct notifier_block *nb;
437
438         pr_info("Registered transition notifiers:\n");
439
440         mutex_lock(&cpufreq_transition_notifier_list.mutex);
441
442         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
443                 pr_info("%pF\n", nb->notifier_call);
444
445         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
446 }
447
448 /**
449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
450  * @policy: cpufreq policy to enable fast frequency switching for.
451  *
452  * Try to enable fast frequency switching for @policy.
453  *
454  * The attempt will fail if there is at least one transition notifier registered
455  * at this point, as fast frequency switching is quite fundamentally at odds
456  * with transition notifiers.  Thus if successful, it will make registration of
457  * transition notifiers fail going forward.
458  */
459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
460 {
461         lockdep_assert_held(&policy->rwsem);
462
463         if (!policy->fast_switch_possible)
464                 return;
465
466         mutex_lock(&cpufreq_fast_switch_lock);
467         if (cpufreq_fast_switch_count >= 0) {
468                 cpufreq_fast_switch_count++;
469                 policy->fast_switch_enabled = true;
470         } else {
471                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
472                         policy->cpu);
473                 cpufreq_list_transition_notifiers();
474         }
475         mutex_unlock(&cpufreq_fast_switch_lock);
476 }
477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
478
479 /**
480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
481  * @policy: cpufreq policy to disable fast frequency switching for.
482  */
483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
484 {
485         mutex_lock(&cpufreq_fast_switch_lock);
486         if (policy->fast_switch_enabled) {
487                 policy->fast_switch_enabled = false;
488                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
489                         cpufreq_fast_switch_count--;
490         }
491         mutex_unlock(&cpufreq_fast_switch_lock);
492 }
493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
494
495 /**
496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
497  * one.
498  * @target_freq: target frequency to resolve.
499  *
500  * The target to driver frequency mapping is cached in the policy.
501  *
502  * Return: Lowest driver-supported frequency greater than or equal to the
503  * given target_freq, subject to policy (min/max) and driver limitations.
504  */
505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
506                                          unsigned int target_freq)
507 {
508         target_freq = clamp_val(target_freq, policy->min, policy->max);
509         policy->cached_target_freq = target_freq;
510
511         if (cpufreq_driver->target_index) {
512                 int idx;
513
514                 idx = cpufreq_frequency_table_target(policy, target_freq,
515                                                      CPUFREQ_RELATION_L);
516                 policy->cached_resolved_idx = idx;
517                 return policy->freq_table[idx].frequency;
518         }
519
520         if (cpufreq_driver->resolve_freq)
521                 return cpufreq_driver->resolve_freq(policy, target_freq);
522
523         return target_freq;
524 }
525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
526
527 /*********************************************************************
528  *                          SYSFS INTERFACE                          *
529  *********************************************************************/
530 static ssize_t show_boost(struct kobject *kobj,
531                                  struct attribute *attr, char *buf)
532 {
533         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
534 }
535
536 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
537                                   const char *buf, size_t count)
538 {
539         int ret, enable;
540
541         ret = sscanf(buf, "%d", &enable);
542         if (ret != 1 || enable < 0 || enable > 1)
543                 return -EINVAL;
544
545         if (cpufreq_boost_trigger_state(enable)) {
546                 pr_err("%s: Cannot %s BOOST!\n",
547                        __func__, enable ? "enable" : "disable");
548                 return -EINVAL;
549         }
550
551         pr_debug("%s: cpufreq BOOST %s\n",
552                  __func__, enable ? "enabled" : "disabled");
553
554         return count;
555 }
556 define_one_global_rw(boost);
557
558 static struct cpufreq_governor *find_governor(const char *str_governor)
559 {
560         struct cpufreq_governor *t;
561
562         for_each_governor(t)
563                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
564                         return t;
565
566         return NULL;
567 }
568
569 /**
570  * cpufreq_parse_governor - parse a governor string
571  */
572 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
573                                 struct cpufreq_governor **governor)
574 {
575         int err = -EINVAL;
576
577         if (cpufreq_driver->setpolicy) {
578                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
579                         *policy = CPUFREQ_POLICY_PERFORMANCE;
580                         err = 0;
581                 } else if (!strncasecmp(str_governor, "powersave",
582                                                 CPUFREQ_NAME_LEN)) {
583                         *policy = CPUFREQ_POLICY_POWERSAVE;
584                         err = 0;
585                 }
586         } else {
587                 struct cpufreq_governor *t;
588
589                 mutex_lock(&cpufreq_governor_mutex);
590
591                 t = find_governor(str_governor);
592
593                 if (t == NULL) {
594                         int ret;
595
596                         mutex_unlock(&cpufreq_governor_mutex);
597                         ret = request_module("cpufreq_%s", str_governor);
598                         mutex_lock(&cpufreq_governor_mutex);
599
600                         if (ret == 0)
601                                 t = find_governor(str_governor);
602                 }
603
604                 if (t != NULL) {
605                         *governor = t;
606                         err = 0;
607                 }
608
609                 mutex_unlock(&cpufreq_governor_mutex);
610         }
611         return err;
612 }
613
614 /**
615  * cpufreq_per_cpu_attr_read() / show_##file_name() -
616  * print out cpufreq information
617  *
618  * Write out information from cpufreq_driver->policy[cpu]; object must be
619  * "unsigned int".
620  */
621
622 #define show_one(file_name, object)                     \
623 static ssize_t show_##file_name                         \
624 (struct cpufreq_policy *policy, char *buf)              \
625 {                                                       \
626         return sprintf(buf, "%u\n", policy->object);    \
627 }
628
629 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
630 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
631 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
632 show_one(scaling_min_freq, min);
633 show_one(scaling_max_freq, max);
634
635 __weak unsigned int arch_freq_get_on_cpu(int cpu)
636 {
637         return 0;
638 }
639
640 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
641 {
642         ssize_t ret;
643         unsigned int freq;
644
645         freq = arch_freq_get_on_cpu(policy->cpu);
646         if (freq)
647                 ret = sprintf(buf, "%u\n", freq);
648         else if (cpufreq_driver && cpufreq_driver->setpolicy &&
649                         cpufreq_driver->get)
650                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
651         else
652                 ret = sprintf(buf, "%u\n", policy->cur);
653         return ret;
654 }
655
656 static int cpufreq_set_policy(struct cpufreq_policy *policy,
657                                 struct cpufreq_policy *new_policy);
658
659 /**
660  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
661  */
662 #define store_one(file_name, object)                    \
663 static ssize_t store_##file_name                                        \
664 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
665 {                                                                       \
666         int ret, temp;                                                  \
667         struct cpufreq_policy new_policy;                               \
668                                                                         \
669         memcpy(&new_policy, policy, sizeof(*policy));                   \
670                                                                         \
671         ret = sscanf(buf, "%u", &new_policy.object);                    \
672         if (ret != 1)                                                   \
673                 return -EINVAL;                                         \
674                                                                         \
675         temp = new_policy.object;                                       \
676         ret = cpufreq_set_policy(policy, &new_policy);          \
677         if (!ret)                                                       \
678                 policy->user_policy.object = temp;                      \
679                                                                         \
680         return ret ? ret : count;                                       \
681 }
682
683 store_one(scaling_min_freq, min);
684 store_one(scaling_max_freq, max);
685
686 /**
687  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
688  */
689 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
690                                         char *buf)
691 {
692         unsigned int cur_freq = __cpufreq_get(policy);
693
694         if (cur_freq)
695                 return sprintf(buf, "%u\n", cur_freq);
696
697         return sprintf(buf, "<unknown>\n");
698 }
699
700 /**
701  * show_scaling_governor - show the current policy for the specified CPU
702  */
703 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
704 {
705         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
706                 return sprintf(buf, "powersave\n");
707         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
708                 return sprintf(buf, "performance\n");
709         else if (policy->governor)
710                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
711                                 policy->governor->name);
712         return -EINVAL;
713 }
714
715 /**
716  * store_scaling_governor - store policy for the specified CPU
717  */
718 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
719                                         const char *buf, size_t count)
720 {
721         int ret;
722         char    str_governor[16];
723         struct cpufreq_policy new_policy;
724
725         memcpy(&new_policy, policy, sizeof(*policy));
726
727         ret = sscanf(buf, "%15s", str_governor);
728         if (ret != 1)
729                 return -EINVAL;
730
731         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
732                                                 &new_policy.governor))
733                 return -EINVAL;
734
735         ret = cpufreq_set_policy(policy, &new_policy);
736         return ret ? ret : count;
737 }
738
739 /**
740  * show_scaling_driver - show the cpufreq driver currently loaded
741  */
742 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
743 {
744         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
745 }
746
747 /**
748  * show_scaling_available_governors - show the available CPUfreq governors
749  */
750 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
751                                                 char *buf)
752 {
753         ssize_t i = 0;
754         struct cpufreq_governor *t;
755
756         if (!has_target()) {
757                 i += sprintf(buf, "performance powersave");
758                 goto out;
759         }
760
761         for_each_governor(t) {
762                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
763                     - (CPUFREQ_NAME_LEN + 2)))
764                         goto out;
765                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
766         }
767 out:
768         i += sprintf(&buf[i], "\n");
769         return i;
770 }
771
772 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
773 {
774         ssize_t i = 0;
775         unsigned int cpu;
776
777         for_each_cpu(cpu, mask) {
778                 if (i)
779                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
780                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
781                 if (i >= (PAGE_SIZE - 5))
782                         break;
783         }
784         i += sprintf(&buf[i], "\n");
785         return i;
786 }
787 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
788
789 /**
790  * show_related_cpus - show the CPUs affected by each transition even if
791  * hw coordination is in use
792  */
793 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
794 {
795         return cpufreq_show_cpus(policy->related_cpus, buf);
796 }
797
798 /**
799  * show_affected_cpus - show the CPUs affected by each transition
800  */
801 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
802 {
803         return cpufreq_show_cpus(policy->cpus, buf);
804 }
805
806 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
807                                         const char *buf, size_t count)
808 {
809         unsigned int freq = 0;
810         unsigned int ret;
811
812         if (!policy->governor || !policy->governor->store_setspeed)
813                 return -EINVAL;
814
815         ret = sscanf(buf, "%u", &freq);
816         if (ret != 1)
817                 return -EINVAL;
818
819         policy->governor->store_setspeed(policy, freq);
820
821         return count;
822 }
823
824 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
825 {
826         if (!policy->governor || !policy->governor->show_setspeed)
827                 return sprintf(buf, "<unsupported>\n");
828
829         return policy->governor->show_setspeed(policy, buf);
830 }
831
832 /**
833  * show_bios_limit - show the current cpufreq HW/BIOS limitation
834  */
835 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
836 {
837         unsigned int limit;
838         int ret;
839         if (cpufreq_driver->bios_limit) {
840                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
841                 if (!ret)
842                         return sprintf(buf, "%u\n", limit);
843         }
844         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
845 }
846
847 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
848 cpufreq_freq_attr_ro(cpuinfo_min_freq);
849 cpufreq_freq_attr_ro(cpuinfo_max_freq);
850 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
851 cpufreq_freq_attr_ro(scaling_available_governors);
852 cpufreq_freq_attr_ro(scaling_driver);
853 cpufreq_freq_attr_ro(scaling_cur_freq);
854 cpufreq_freq_attr_ro(bios_limit);
855 cpufreq_freq_attr_ro(related_cpus);
856 cpufreq_freq_attr_ro(affected_cpus);
857 cpufreq_freq_attr_rw(scaling_min_freq);
858 cpufreq_freq_attr_rw(scaling_max_freq);
859 cpufreq_freq_attr_rw(scaling_governor);
860 cpufreq_freq_attr_rw(scaling_setspeed);
861
862 static struct attribute *default_attrs[] = {
863         &cpuinfo_min_freq.attr,
864         &cpuinfo_max_freq.attr,
865         &cpuinfo_transition_latency.attr,
866         &scaling_min_freq.attr,
867         &scaling_max_freq.attr,
868         &affected_cpus.attr,
869         &related_cpus.attr,
870         &scaling_governor.attr,
871         &scaling_driver.attr,
872         &scaling_available_governors.attr,
873         &scaling_setspeed.attr,
874         NULL
875 };
876
877 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
878 #define to_attr(a) container_of(a, struct freq_attr, attr)
879
880 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
881 {
882         struct cpufreq_policy *policy = to_policy(kobj);
883         struct freq_attr *fattr = to_attr(attr);
884         ssize_t ret;
885
886         down_read(&policy->rwsem);
887         ret = fattr->show(policy, buf);
888         up_read(&policy->rwsem);
889
890         return ret;
891 }
892
893 static ssize_t store(struct kobject *kobj, struct attribute *attr,
894                      const char *buf, size_t count)
895 {
896         struct cpufreq_policy *policy = to_policy(kobj);
897         struct freq_attr *fattr = to_attr(attr);
898         ssize_t ret = -EINVAL;
899
900         cpus_read_lock();
901
902         if (cpu_online(policy->cpu)) {
903                 down_write(&policy->rwsem);
904                 ret = fattr->store(policy, buf, count);
905                 up_write(&policy->rwsem);
906         }
907
908         cpus_read_unlock();
909
910         return ret;
911 }
912
913 static void cpufreq_sysfs_release(struct kobject *kobj)
914 {
915         struct cpufreq_policy *policy = to_policy(kobj);
916         pr_debug("last reference is dropped\n");
917         complete(&policy->kobj_unregister);
918 }
919
920 static const struct sysfs_ops sysfs_ops = {
921         .show   = show,
922         .store  = store,
923 };
924
925 static struct kobj_type ktype_cpufreq = {
926         .sysfs_ops      = &sysfs_ops,
927         .default_attrs  = default_attrs,
928         .release        = cpufreq_sysfs_release,
929 };
930
931 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
932 {
933         struct device *dev = get_cpu_device(cpu);
934
935         if (!dev)
936                 return;
937
938         if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
939                 return;
940
941         dev_dbg(dev, "%s: Adding symlink\n", __func__);
942         if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
943                 dev_err(dev, "cpufreq symlink creation failed\n");
944 }
945
946 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
947                                    struct device *dev)
948 {
949         dev_dbg(dev, "%s: Removing symlink\n", __func__);
950         sysfs_remove_link(&dev->kobj, "cpufreq");
951 }
952
953 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
954 {
955         struct freq_attr **drv_attr;
956         int ret = 0;
957
958         /* set up files for this cpu device */
959         drv_attr = cpufreq_driver->attr;
960         while (drv_attr && *drv_attr) {
961                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
962                 if (ret)
963                         return ret;
964                 drv_attr++;
965         }
966         if (cpufreq_driver->get) {
967                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
968                 if (ret)
969                         return ret;
970         }
971
972         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
973         if (ret)
974                 return ret;
975
976         if (cpufreq_driver->bios_limit) {
977                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
978                 if (ret)
979                         return ret;
980         }
981
982         return 0;
983 }
984
985 __weak struct cpufreq_governor *cpufreq_default_governor(void)
986 {
987         return NULL;
988 }
989
990 static int cpufreq_init_policy(struct cpufreq_policy *policy)
991 {
992         struct cpufreq_governor *gov = NULL;
993         struct cpufreq_policy new_policy;
994
995         memcpy(&new_policy, policy, sizeof(*policy));
996
997         /* Update governor of new_policy to the governor used before hotplug */
998         gov = find_governor(policy->last_governor);
999         if (gov) {
1000                 pr_debug("Restoring governor %s for cpu %d\n",
1001                                 policy->governor->name, policy->cpu);
1002         } else {
1003                 gov = cpufreq_default_governor();
1004                 if (!gov)
1005                         return -ENODATA;
1006         }
1007
1008         new_policy.governor = gov;
1009
1010         /* Use the default policy if there is no last_policy. */
1011         if (cpufreq_driver->setpolicy) {
1012                 if (policy->last_policy)
1013                         new_policy.policy = policy->last_policy;
1014                 else
1015                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1016                                                NULL);
1017         }
1018         /* set default policy */
1019         return cpufreq_set_policy(policy, &new_policy);
1020 }
1021
1022 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1023 {
1024         int ret = 0;
1025
1026         /* Has this CPU been taken care of already? */
1027         if (cpumask_test_cpu(cpu, policy->cpus))
1028                 return 0;
1029
1030         down_write(&policy->rwsem);
1031         if (has_target())
1032                 cpufreq_stop_governor(policy);
1033
1034         cpumask_set_cpu(cpu, policy->cpus);
1035
1036         if (has_target()) {
1037                 ret = cpufreq_start_governor(policy);
1038                 if (ret)
1039                         pr_err("%s: Failed to start governor\n", __func__);
1040         }
1041         up_write(&policy->rwsem);
1042         return ret;
1043 }
1044
1045 static void handle_update(struct work_struct *work)
1046 {
1047         struct cpufreq_policy *policy =
1048                 container_of(work, struct cpufreq_policy, update);
1049         unsigned int cpu = policy->cpu;
1050         pr_debug("handle_update for cpu %u called\n", cpu);
1051         cpufreq_update_policy(cpu);
1052 }
1053
1054 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1055 {
1056         struct cpufreq_policy *policy;
1057         int ret;
1058
1059         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1060         if (!policy)
1061                 return NULL;
1062
1063         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1064                 goto err_free_policy;
1065
1066         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1067                 goto err_free_cpumask;
1068
1069         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1070                 goto err_free_rcpumask;
1071
1072         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1073                                    cpufreq_global_kobject, "policy%u", cpu);
1074         if (ret) {
1075                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1076                 goto err_free_real_cpus;
1077         }
1078
1079         INIT_LIST_HEAD(&policy->policy_list);
1080         init_rwsem(&policy->rwsem);
1081         spin_lock_init(&policy->transition_lock);
1082         init_waitqueue_head(&policy->transition_wait);
1083         init_completion(&policy->kobj_unregister);
1084         INIT_WORK(&policy->update, handle_update);
1085
1086         policy->cpu = cpu;
1087         return policy;
1088
1089 err_free_real_cpus:
1090         free_cpumask_var(policy->real_cpus);
1091 err_free_rcpumask:
1092         free_cpumask_var(policy->related_cpus);
1093 err_free_cpumask:
1094         free_cpumask_var(policy->cpus);
1095 err_free_policy:
1096         kfree(policy);
1097
1098         return NULL;
1099 }
1100
1101 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1102 {
1103         struct kobject *kobj;
1104         struct completion *cmp;
1105
1106         down_write(&policy->rwsem);
1107         cpufreq_stats_free_table(policy);
1108         kobj = &policy->kobj;
1109         cmp = &policy->kobj_unregister;
1110         up_write(&policy->rwsem);
1111         kobject_put(kobj);
1112
1113         /*
1114          * We need to make sure that the underlying kobj is
1115          * actually not referenced anymore by anybody before we
1116          * proceed with unloading.
1117          */
1118         pr_debug("waiting for dropping of refcount\n");
1119         wait_for_completion(cmp);
1120         pr_debug("wait complete\n");
1121 }
1122
1123 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1124 {
1125         unsigned long flags;
1126         int cpu;
1127
1128         /* Remove policy from list */
1129         write_lock_irqsave(&cpufreq_driver_lock, flags);
1130         list_del(&policy->policy_list);
1131
1132         for_each_cpu(cpu, policy->related_cpus)
1133                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1134         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1135
1136         cpufreq_policy_put_kobj(policy);
1137         free_cpumask_var(policy->real_cpus);
1138         free_cpumask_var(policy->related_cpus);
1139         free_cpumask_var(policy->cpus);
1140         kfree(policy);
1141 }
1142
1143 static int cpufreq_online(unsigned int cpu)
1144 {
1145         struct cpufreq_policy *policy;
1146         bool new_policy;
1147         unsigned long flags;
1148         unsigned int j;
1149         int ret;
1150
1151         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1152
1153         /* Check if this CPU already has a policy to manage it */
1154         policy = per_cpu(cpufreq_cpu_data, cpu);
1155         if (policy) {
1156                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1157                 if (!policy_is_inactive(policy))
1158                         return cpufreq_add_policy_cpu(policy, cpu);
1159
1160                 /* This is the only online CPU for the policy.  Start over. */
1161                 new_policy = false;
1162                 down_write(&policy->rwsem);
1163                 policy->cpu = cpu;
1164                 policy->governor = NULL;
1165                 up_write(&policy->rwsem);
1166         } else {
1167                 new_policy = true;
1168                 policy = cpufreq_policy_alloc(cpu);
1169                 if (!policy)
1170                         return -ENOMEM;
1171         }
1172
1173         cpumask_copy(policy->cpus, cpumask_of(cpu));
1174
1175         /* call driver. From then on the cpufreq must be able
1176          * to accept all calls to ->verify and ->setpolicy for this CPU
1177          */
1178         ret = cpufreq_driver->init(policy);
1179         if (ret) {
1180                 pr_debug("initialization failed\n");
1181                 goto out_free_policy;
1182         }
1183
1184         down_write(&policy->rwsem);
1185
1186         if (new_policy) {
1187                 /* related_cpus should at least include policy->cpus. */
1188                 cpumask_copy(policy->related_cpus, policy->cpus);
1189         }
1190
1191         /*
1192          * affected cpus must always be the one, which are online. We aren't
1193          * managing offline cpus here.
1194          */
1195         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1196
1197         if (new_policy) {
1198                 policy->user_policy.min = policy->min;
1199                 policy->user_policy.max = policy->max;
1200
1201                 for_each_cpu(j, policy->related_cpus) {
1202                         per_cpu(cpufreq_cpu_data, j) = policy;
1203                         add_cpu_dev_symlink(policy, j);
1204                 }
1205         } else {
1206                 policy->min = policy->user_policy.min;
1207                 policy->max = policy->user_policy.max;
1208         }
1209
1210         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1211                 policy->cur = cpufreq_driver->get(policy->cpu);
1212                 if (!policy->cur) {
1213                         pr_err("%s: ->get() failed\n", __func__);
1214                         goto out_exit_policy;
1215                 }
1216         }
1217
1218         /*
1219          * Sometimes boot loaders set CPU frequency to a value outside of
1220          * frequency table present with cpufreq core. In such cases CPU might be
1221          * unstable if it has to run on that frequency for long duration of time
1222          * and so its better to set it to a frequency which is specified in
1223          * freq-table. This also makes cpufreq stats inconsistent as
1224          * cpufreq-stats would fail to register because current frequency of CPU
1225          * isn't found in freq-table.
1226          *
1227          * Because we don't want this change to effect boot process badly, we go
1228          * for the next freq which is >= policy->cur ('cur' must be set by now,
1229          * otherwise we will end up setting freq to lowest of the table as 'cur'
1230          * is initialized to zero).
1231          *
1232          * We are passing target-freq as "policy->cur - 1" otherwise
1233          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1234          * equal to target-freq.
1235          */
1236         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1237             && has_target()) {
1238                 /* Are we running at unknown frequency ? */
1239                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1240                 if (ret == -EINVAL) {
1241                         /* Warn user and fix it */
1242                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1243                                 __func__, policy->cpu, policy->cur);
1244                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1245                                 CPUFREQ_RELATION_L);
1246
1247                         /*
1248                          * Reaching here after boot in a few seconds may not
1249                          * mean that system will remain stable at "unknown"
1250                          * frequency for longer duration. Hence, a BUG_ON().
1251                          */
1252                         BUG_ON(ret);
1253                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1254                                 __func__, policy->cpu, policy->cur);
1255                 }
1256         }
1257
1258         if (new_policy) {
1259                 ret = cpufreq_add_dev_interface(policy);
1260                 if (ret)
1261                         goto out_exit_policy;
1262
1263                 cpufreq_stats_create_table(policy);
1264
1265                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1266                 list_add(&policy->policy_list, &cpufreq_policy_list);
1267                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1268         }
1269
1270         ret = cpufreq_init_policy(policy);
1271         if (ret) {
1272                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1273                        __func__, cpu, ret);
1274                 /* cpufreq_policy_free() will notify based on this */
1275                 new_policy = false;
1276                 goto out_exit_policy;
1277         }
1278
1279         up_write(&policy->rwsem);
1280
1281         kobject_uevent(&policy->kobj, KOBJ_ADD);
1282
1283         /* Callback for handling stuff after policy is ready */
1284         if (cpufreq_driver->ready)
1285                 cpufreq_driver->ready(policy);
1286
1287         pr_debug("initialization complete\n");
1288
1289         return 0;
1290
1291 out_exit_policy:
1292         up_write(&policy->rwsem);
1293
1294         if (cpufreq_driver->exit)
1295                 cpufreq_driver->exit(policy);
1296
1297         for_each_cpu(j, policy->real_cpus)
1298                 remove_cpu_dev_symlink(policy, get_cpu_device(j));
1299
1300 out_free_policy:
1301         cpufreq_policy_free(policy);
1302         return ret;
1303 }
1304
1305 /**
1306  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1307  * @dev: CPU device.
1308  * @sif: Subsystem interface structure pointer (not used)
1309  */
1310 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1311 {
1312         struct cpufreq_policy *policy;
1313         unsigned cpu = dev->id;
1314         int ret;
1315
1316         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1317
1318         if (cpu_online(cpu)) {
1319                 ret = cpufreq_online(cpu);
1320                 if (ret)
1321                         return ret;
1322         }
1323
1324         /* Create sysfs link on CPU registration */
1325         policy = per_cpu(cpufreq_cpu_data, cpu);
1326         if (policy)
1327                 add_cpu_dev_symlink(policy, cpu);
1328
1329         return 0;
1330 }
1331
1332 static int cpufreq_offline(unsigned int cpu)
1333 {
1334         struct cpufreq_policy *policy;
1335         int ret;
1336
1337         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1338
1339         policy = cpufreq_cpu_get_raw(cpu);
1340         if (!policy) {
1341                 pr_debug("%s: No cpu_data found\n", __func__);
1342                 return 0;
1343         }
1344
1345         down_write(&policy->rwsem);
1346         if (has_target())
1347                 cpufreq_stop_governor(policy);
1348
1349         cpumask_clear_cpu(cpu, policy->cpus);
1350
1351         if (policy_is_inactive(policy)) {
1352                 if (has_target())
1353                         strncpy(policy->last_governor, policy->governor->name,
1354                                 CPUFREQ_NAME_LEN);
1355                 else
1356                         policy->last_policy = policy->policy;
1357         } else if (cpu == policy->cpu) {
1358                 /* Nominate new CPU */
1359                 policy->cpu = cpumask_any(policy->cpus);
1360         }
1361
1362         /* Start governor again for active policy */
1363         if (!policy_is_inactive(policy)) {
1364                 if (has_target()) {
1365                         ret = cpufreq_start_governor(policy);
1366                         if (ret)
1367                                 pr_err("%s: Failed to start governor\n", __func__);
1368                 }
1369
1370                 goto unlock;
1371         }
1372
1373         if (cpufreq_driver->stop_cpu)
1374                 cpufreq_driver->stop_cpu(policy);
1375
1376         if (has_target())
1377                 cpufreq_exit_governor(policy);
1378
1379         /*
1380          * Perform the ->exit() even during light-weight tear-down,
1381          * since this is a core component, and is essential for the
1382          * subsequent light-weight ->init() to succeed.
1383          */
1384         if (cpufreq_driver->exit) {
1385                 cpufreq_driver->exit(policy);
1386                 policy->freq_table = NULL;
1387         }
1388
1389 unlock:
1390         up_write(&policy->rwsem);
1391         return 0;
1392 }
1393
1394 /**
1395  * cpufreq_remove_dev - remove a CPU device
1396  *
1397  * Removes the cpufreq interface for a CPU device.
1398  */
1399 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1400 {
1401         unsigned int cpu = dev->id;
1402         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1403
1404         if (!policy)
1405                 return;
1406
1407         if (cpu_online(cpu))
1408                 cpufreq_offline(cpu);
1409
1410         cpumask_clear_cpu(cpu, policy->real_cpus);
1411         remove_cpu_dev_symlink(policy, dev);
1412
1413         if (cpumask_empty(policy->real_cpus))
1414                 cpufreq_policy_free(policy);
1415 }
1416
1417 /**
1418  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1419  *      in deep trouble.
1420  *      @policy: policy managing CPUs
1421  *      @new_freq: CPU frequency the CPU actually runs at
1422  *
1423  *      We adjust to current frequency first, and need to clean up later.
1424  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1425  */
1426 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1427                                 unsigned int new_freq)
1428 {
1429         struct cpufreq_freqs freqs;
1430
1431         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1432                  policy->cur, new_freq);
1433
1434         freqs.old = policy->cur;
1435         freqs.new = new_freq;
1436
1437         cpufreq_freq_transition_begin(policy, &freqs);
1438         cpufreq_freq_transition_end(policy, &freqs, 0);
1439 }
1440
1441 /**
1442  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1443  * @cpu: CPU number
1444  *
1445  * This is the last known freq, without actually getting it from the driver.
1446  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1447  */
1448 unsigned int cpufreq_quick_get(unsigned int cpu)
1449 {
1450         struct cpufreq_policy *policy;
1451         unsigned int ret_freq = 0;
1452         unsigned long flags;
1453
1454         read_lock_irqsave(&cpufreq_driver_lock, flags);
1455
1456         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1457                 ret_freq = cpufreq_driver->get(cpu);
1458                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1459                 return ret_freq;
1460         }
1461
1462         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1463
1464         policy = cpufreq_cpu_get(cpu);
1465         if (policy) {
1466                 ret_freq = policy->cur;
1467                 cpufreq_cpu_put(policy);
1468         }
1469
1470         return ret_freq;
1471 }
1472 EXPORT_SYMBOL(cpufreq_quick_get);
1473
1474 /**
1475  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1476  * @cpu: CPU number
1477  *
1478  * Just return the max possible frequency for a given CPU.
1479  */
1480 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1481 {
1482         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1483         unsigned int ret_freq = 0;
1484
1485         if (policy) {
1486                 ret_freq = policy->max;
1487                 cpufreq_cpu_put(policy);
1488         }
1489
1490         return ret_freq;
1491 }
1492 EXPORT_SYMBOL(cpufreq_quick_get_max);
1493
1494 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1495 {
1496         unsigned int ret_freq = 0;
1497
1498         if (!cpufreq_driver->get)
1499                 return ret_freq;
1500
1501         ret_freq = cpufreq_driver->get(policy->cpu);
1502
1503         /*
1504          * Updating inactive policies is invalid, so avoid doing that.  Also
1505          * if fast frequency switching is used with the given policy, the check
1506          * against policy->cur is pointless, so skip it in that case too.
1507          */
1508         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1509                 return ret_freq;
1510
1511         if (ret_freq && policy->cur &&
1512                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1513                 /* verify no discrepancy between actual and
1514                                         saved value exists */
1515                 if (unlikely(ret_freq != policy->cur)) {
1516                         cpufreq_out_of_sync(policy, ret_freq);
1517                         schedule_work(&policy->update);
1518                 }
1519         }
1520
1521         return ret_freq;
1522 }
1523
1524 /**
1525  * cpufreq_get - get the current CPU frequency (in kHz)
1526  * @cpu: CPU number
1527  *
1528  * Get the CPU current (static) CPU frequency
1529  */
1530 unsigned int cpufreq_get(unsigned int cpu)
1531 {
1532         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1533         unsigned int ret_freq = 0;
1534
1535         if (policy) {
1536                 down_read(&policy->rwsem);
1537
1538                 if (!policy_is_inactive(policy))
1539                         ret_freq = __cpufreq_get(policy);
1540
1541                 up_read(&policy->rwsem);
1542
1543                 cpufreq_cpu_put(policy);
1544         }
1545
1546         return ret_freq;
1547 }
1548 EXPORT_SYMBOL(cpufreq_get);
1549
1550 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1551 {
1552         unsigned int new_freq;
1553
1554         new_freq = cpufreq_driver->get(policy->cpu);
1555         if (!new_freq)
1556                 return 0;
1557
1558         if (!policy->cur) {
1559                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1560                 policy->cur = new_freq;
1561         } else if (policy->cur != new_freq && has_target()) {
1562                 cpufreq_out_of_sync(policy, new_freq);
1563         }
1564
1565         return new_freq;
1566 }
1567
1568 static struct subsys_interface cpufreq_interface = {
1569         .name           = "cpufreq",
1570         .subsys         = &cpu_subsys,
1571         .add_dev        = cpufreq_add_dev,
1572         .remove_dev     = cpufreq_remove_dev,
1573 };
1574
1575 /*
1576  * In case platform wants some specific frequency to be configured
1577  * during suspend..
1578  */
1579 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1580 {
1581         int ret;
1582
1583         if (!policy->suspend_freq) {
1584                 pr_debug("%s: suspend_freq not defined\n", __func__);
1585                 return 0;
1586         }
1587
1588         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1589                         policy->suspend_freq);
1590
1591         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1592                         CPUFREQ_RELATION_H);
1593         if (ret)
1594                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1595                                 __func__, policy->suspend_freq, ret);
1596
1597         return ret;
1598 }
1599 EXPORT_SYMBOL(cpufreq_generic_suspend);
1600
1601 /**
1602  * cpufreq_suspend() - Suspend CPUFreq governors
1603  *
1604  * Called during system wide Suspend/Hibernate cycles for suspending governors
1605  * as some platforms can't change frequency after this point in suspend cycle.
1606  * Because some of the devices (like: i2c, regulators, etc) they use for
1607  * changing frequency are suspended quickly after this point.
1608  */
1609 void cpufreq_suspend(void)
1610 {
1611         struct cpufreq_policy *policy;
1612
1613         if (!cpufreq_driver)
1614                 return;
1615
1616         if (!has_target() && !cpufreq_driver->suspend)
1617                 goto suspend;
1618
1619         pr_debug("%s: Suspending Governors\n", __func__);
1620
1621         for_each_active_policy(policy) {
1622                 if (has_target()) {
1623                         down_write(&policy->rwsem);
1624                         cpufreq_stop_governor(policy);
1625                         up_write(&policy->rwsem);
1626                 }
1627
1628                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1629                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1630                                 policy);
1631         }
1632
1633 suspend:
1634         cpufreq_suspended = true;
1635 }
1636
1637 /**
1638  * cpufreq_resume() - Resume CPUFreq governors
1639  *
1640  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1641  * are suspended with cpufreq_suspend().
1642  */
1643 void cpufreq_resume(void)
1644 {
1645         struct cpufreq_policy *policy;
1646         int ret;
1647
1648         if (!cpufreq_driver)
1649                 return;
1650
1651         cpufreq_suspended = false;
1652
1653         if (!has_target() && !cpufreq_driver->resume)
1654                 return;
1655
1656         pr_debug("%s: Resuming Governors\n", __func__);
1657
1658         for_each_active_policy(policy) {
1659                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1660                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1661                                 policy);
1662                 } else if (has_target()) {
1663                         down_write(&policy->rwsem);
1664                         ret = cpufreq_start_governor(policy);
1665                         up_write(&policy->rwsem);
1666
1667                         if (ret)
1668                                 pr_err("%s: Failed to start governor for policy: %p\n",
1669                                        __func__, policy);
1670                 }
1671         }
1672 }
1673
1674 /**
1675  *      cpufreq_get_current_driver - return current driver's name
1676  *
1677  *      Return the name string of the currently loaded cpufreq driver
1678  *      or NULL, if none.
1679  */
1680 const char *cpufreq_get_current_driver(void)
1681 {
1682         if (cpufreq_driver)
1683                 return cpufreq_driver->name;
1684
1685         return NULL;
1686 }
1687 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1688
1689 /**
1690  *      cpufreq_get_driver_data - return current driver data
1691  *
1692  *      Return the private data of the currently loaded cpufreq
1693  *      driver, or NULL if no cpufreq driver is loaded.
1694  */
1695 void *cpufreq_get_driver_data(void)
1696 {
1697         if (cpufreq_driver)
1698                 return cpufreq_driver->driver_data;
1699
1700         return NULL;
1701 }
1702 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1703
1704 /*********************************************************************
1705  *                     NOTIFIER LISTS INTERFACE                      *
1706  *********************************************************************/
1707
1708 /**
1709  *      cpufreq_register_notifier - register a driver with cpufreq
1710  *      @nb: notifier function to register
1711  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1712  *
1713  *      Add a driver to one of two lists: either a list of drivers that
1714  *      are notified about clock rate changes (once before and once after
1715  *      the transition), or a list of drivers that are notified about
1716  *      changes in cpufreq policy.
1717  *
1718  *      This function may sleep, and has the same return conditions as
1719  *      blocking_notifier_chain_register.
1720  */
1721 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1722 {
1723         int ret;
1724
1725         if (cpufreq_disabled())
1726                 return -EINVAL;
1727
1728         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1729
1730         switch (list) {
1731         case CPUFREQ_TRANSITION_NOTIFIER:
1732                 mutex_lock(&cpufreq_fast_switch_lock);
1733
1734                 if (cpufreq_fast_switch_count > 0) {
1735                         mutex_unlock(&cpufreq_fast_switch_lock);
1736                         return -EBUSY;
1737                 }
1738                 ret = srcu_notifier_chain_register(
1739                                 &cpufreq_transition_notifier_list, nb);
1740                 if (!ret)
1741                         cpufreq_fast_switch_count--;
1742
1743                 mutex_unlock(&cpufreq_fast_switch_lock);
1744                 break;
1745         case CPUFREQ_POLICY_NOTIFIER:
1746                 ret = blocking_notifier_chain_register(
1747                                 &cpufreq_policy_notifier_list, nb);
1748                 break;
1749         default:
1750                 ret = -EINVAL;
1751         }
1752
1753         return ret;
1754 }
1755 EXPORT_SYMBOL(cpufreq_register_notifier);
1756
1757 /**
1758  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1759  *      @nb: notifier block to be unregistered
1760  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1761  *
1762  *      Remove a driver from the CPU frequency notifier list.
1763  *
1764  *      This function may sleep, and has the same return conditions as
1765  *      blocking_notifier_chain_unregister.
1766  */
1767 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1768 {
1769         int ret;
1770
1771         if (cpufreq_disabled())
1772                 return -EINVAL;
1773
1774         switch (list) {
1775         case CPUFREQ_TRANSITION_NOTIFIER:
1776                 mutex_lock(&cpufreq_fast_switch_lock);
1777
1778                 ret = srcu_notifier_chain_unregister(
1779                                 &cpufreq_transition_notifier_list, nb);
1780                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1781                         cpufreq_fast_switch_count++;
1782
1783                 mutex_unlock(&cpufreq_fast_switch_lock);
1784                 break;
1785         case CPUFREQ_POLICY_NOTIFIER:
1786                 ret = blocking_notifier_chain_unregister(
1787                                 &cpufreq_policy_notifier_list, nb);
1788                 break;
1789         default:
1790                 ret = -EINVAL;
1791         }
1792
1793         return ret;
1794 }
1795 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1796
1797
1798 /*********************************************************************
1799  *                              GOVERNORS                            *
1800  *********************************************************************/
1801
1802 /**
1803  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1804  * @policy: cpufreq policy to switch the frequency for.
1805  * @target_freq: New frequency to set (may be approximate).
1806  *
1807  * Carry out a fast frequency switch without sleeping.
1808  *
1809  * The driver's ->fast_switch() callback invoked by this function must be
1810  * suitable for being called from within RCU-sched read-side critical sections
1811  * and it is expected to select the minimum available frequency greater than or
1812  * equal to @target_freq (CPUFREQ_RELATION_L).
1813  *
1814  * This function must not be called if policy->fast_switch_enabled is unset.
1815  *
1816  * Governors calling this function must guarantee that it will never be invoked
1817  * twice in parallel for the same policy and that it will never be called in
1818  * parallel with either ->target() or ->target_index() for the same policy.
1819  *
1820  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1821  * callback to indicate an error condition, the hardware configuration must be
1822  * preserved.
1823  */
1824 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1825                                         unsigned int target_freq)
1826 {
1827         target_freq = clamp_val(target_freq, policy->min, policy->max);
1828
1829         return cpufreq_driver->fast_switch(policy, target_freq);
1830 }
1831 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1832
1833 /* Must set freqs->new to intermediate frequency */
1834 static int __target_intermediate(struct cpufreq_policy *policy,
1835                                  struct cpufreq_freqs *freqs, int index)
1836 {
1837         int ret;
1838
1839         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1840
1841         /* We don't need to switch to intermediate freq */
1842         if (!freqs->new)
1843                 return 0;
1844
1845         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1846                  __func__, policy->cpu, freqs->old, freqs->new);
1847
1848         cpufreq_freq_transition_begin(policy, freqs);
1849         ret = cpufreq_driver->target_intermediate(policy, index);
1850         cpufreq_freq_transition_end(policy, freqs, ret);
1851
1852         if (ret)
1853                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1854                        __func__, ret);
1855
1856         return ret;
1857 }
1858
1859 static int __target_index(struct cpufreq_policy *policy, int index)
1860 {
1861         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1862         unsigned int intermediate_freq = 0;
1863         unsigned int newfreq = policy->freq_table[index].frequency;
1864         int retval = -EINVAL;
1865         bool notify;
1866
1867         if (newfreq == policy->cur)
1868                 return 0;
1869
1870         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1871         if (notify) {
1872                 /* Handle switching to intermediate frequency */
1873                 if (cpufreq_driver->get_intermediate) {
1874                         retval = __target_intermediate(policy, &freqs, index);
1875                         if (retval)
1876                                 return retval;
1877
1878                         intermediate_freq = freqs.new;
1879                         /* Set old freq to intermediate */
1880                         if (intermediate_freq)
1881                                 freqs.old = freqs.new;
1882                 }
1883
1884                 freqs.new = newfreq;
1885                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1886                          __func__, policy->cpu, freqs.old, freqs.new);
1887
1888                 cpufreq_freq_transition_begin(policy, &freqs);
1889         }
1890
1891         retval = cpufreq_driver->target_index(policy, index);
1892         if (retval)
1893                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1894                        retval);
1895
1896         if (notify) {
1897                 cpufreq_freq_transition_end(policy, &freqs, retval);
1898
1899                 /*
1900                  * Failed after setting to intermediate freq? Driver should have
1901                  * reverted back to initial frequency and so should we. Check
1902                  * here for intermediate_freq instead of get_intermediate, in
1903                  * case we haven't switched to intermediate freq at all.
1904                  */
1905                 if (unlikely(retval && intermediate_freq)) {
1906                         freqs.old = intermediate_freq;
1907                         freqs.new = policy->restore_freq;
1908                         cpufreq_freq_transition_begin(policy, &freqs);
1909                         cpufreq_freq_transition_end(policy, &freqs, 0);
1910                 }
1911         }
1912
1913         return retval;
1914 }
1915
1916 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1917                             unsigned int target_freq,
1918                             unsigned int relation)
1919 {
1920         unsigned int old_target_freq = target_freq;
1921         int index;
1922
1923         if (cpufreq_disabled())
1924                 return -ENODEV;
1925
1926         /* Make sure that target_freq is within supported range */
1927         target_freq = clamp_val(target_freq, policy->min, policy->max);
1928
1929         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1930                  policy->cpu, target_freq, relation, old_target_freq);
1931
1932         /*
1933          * This might look like a redundant call as we are checking it again
1934          * after finding index. But it is left intentionally for cases where
1935          * exactly same freq is called again and so we can save on few function
1936          * calls.
1937          */
1938         if (target_freq == policy->cur)
1939                 return 0;
1940
1941         /* Save last value to restore later on errors */
1942         policy->restore_freq = policy->cur;
1943
1944         if (cpufreq_driver->target)
1945                 return cpufreq_driver->target(policy, target_freq, relation);
1946
1947         if (!cpufreq_driver->target_index)
1948                 return -EINVAL;
1949
1950         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1951
1952         return __target_index(policy, index);
1953 }
1954 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1955
1956 int cpufreq_driver_target(struct cpufreq_policy *policy,
1957                           unsigned int target_freq,
1958                           unsigned int relation)
1959 {
1960         int ret = -EINVAL;
1961
1962         down_write(&policy->rwsem);
1963
1964         ret = __cpufreq_driver_target(policy, target_freq, relation);
1965
1966         up_write(&policy->rwsem);
1967
1968         return ret;
1969 }
1970 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1971
1972 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1973 {
1974         return NULL;
1975 }
1976
1977 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1978 {
1979         int ret;
1980
1981         /* Don't start any governor operations if we are entering suspend */
1982         if (cpufreq_suspended)
1983                 return 0;
1984         /*
1985          * Governor might not be initiated here if ACPI _PPC changed
1986          * notification happened, so check it.
1987          */
1988         if (!policy->governor)
1989                 return -EINVAL;
1990
1991         if (policy->governor->max_transition_latency &&
1992             policy->cpuinfo.transition_latency >
1993             policy->governor->max_transition_latency) {
1994                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
1995
1996                 if (gov) {
1997                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1998                                 policy->governor->name, gov->name);
1999                         policy->governor = gov;
2000                 } else {
2001                         return -EINVAL;
2002                 }
2003         }
2004
2005         if (!try_module_get(policy->governor->owner))
2006                 return -EINVAL;
2007
2008         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2009
2010         if (policy->governor->init) {
2011                 ret = policy->governor->init(policy);
2012                 if (ret) {
2013                         module_put(policy->governor->owner);
2014                         return ret;
2015                 }
2016         }
2017
2018         return 0;
2019 }
2020
2021 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2022 {
2023         if (cpufreq_suspended || !policy->governor)
2024                 return;
2025
2026         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2027
2028         if (policy->governor->exit)
2029                 policy->governor->exit(policy);
2030
2031         module_put(policy->governor->owner);
2032 }
2033
2034 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2035 {
2036         int ret;
2037
2038         if (cpufreq_suspended)
2039                 return 0;
2040
2041         if (!policy->governor)
2042                 return -EINVAL;
2043
2044         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2045
2046         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2047                 cpufreq_update_current_freq(policy);
2048
2049         if (policy->governor->start) {
2050                 ret = policy->governor->start(policy);
2051                 if (ret)
2052                         return ret;
2053         }
2054
2055         if (policy->governor->limits)
2056                 policy->governor->limits(policy);
2057
2058         return 0;
2059 }
2060
2061 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2062 {
2063         if (cpufreq_suspended || !policy->governor)
2064                 return;
2065
2066         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2067
2068         if (policy->governor->stop)
2069                 policy->governor->stop(policy);
2070 }
2071
2072 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2073 {
2074         if (cpufreq_suspended || !policy->governor)
2075                 return;
2076
2077         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2078
2079         if (policy->governor->limits)
2080                 policy->governor->limits(policy);
2081 }
2082
2083 int cpufreq_register_governor(struct cpufreq_governor *governor)
2084 {
2085         int err;
2086
2087         if (!governor)
2088                 return -EINVAL;
2089
2090         if (cpufreq_disabled())
2091                 return -ENODEV;
2092
2093         mutex_lock(&cpufreq_governor_mutex);
2094
2095         err = -EBUSY;
2096         if (!find_governor(governor->name)) {
2097                 err = 0;
2098                 list_add(&governor->governor_list, &cpufreq_governor_list);
2099         }
2100
2101         mutex_unlock(&cpufreq_governor_mutex);
2102         return err;
2103 }
2104 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2105
2106 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2107 {
2108         struct cpufreq_policy *policy;
2109         unsigned long flags;
2110
2111         if (!governor)
2112                 return;
2113
2114         if (cpufreq_disabled())
2115                 return;
2116
2117         /* clear last_governor for all inactive policies */
2118         read_lock_irqsave(&cpufreq_driver_lock, flags);
2119         for_each_inactive_policy(policy) {
2120                 if (!strcmp(policy->last_governor, governor->name)) {
2121                         policy->governor = NULL;
2122                         strcpy(policy->last_governor, "\0");
2123                 }
2124         }
2125         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2126
2127         mutex_lock(&cpufreq_governor_mutex);
2128         list_del(&governor->governor_list);
2129         mutex_unlock(&cpufreq_governor_mutex);
2130         return;
2131 }
2132 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2133
2134
2135 /*********************************************************************
2136  *                          POLICY INTERFACE                         *
2137  *********************************************************************/
2138
2139 /**
2140  * cpufreq_get_policy - get the current cpufreq_policy
2141  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2142  *      is written
2143  *
2144  * Reads the current cpufreq policy.
2145  */
2146 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2147 {
2148         struct cpufreq_policy *cpu_policy;
2149         if (!policy)
2150                 return -EINVAL;
2151
2152         cpu_policy = cpufreq_cpu_get(cpu);
2153         if (!cpu_policy)
2154                 return -EINVAL;
2155
2156         memcpy(policy, cpu_policy, sizeof(*policy));
2157
2158         cpufreq_cpu_put(cpu_policy);
2159         return 0;
2160 }
2161 EXPORT_SYMBOL(cpufreq_get_policy);
2162
2163 /*
2164  * policy : current policy.
2165  * new_policy: policy to be set.
2166  */
2167 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2168                                 struct cpufreq_policy *new_policy)
2169 {
2170         struct cpufreq_governor *old_gov;
2171         int ret;
2172
2173         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2174                  new_policy->cpu, new_policy->min, new_policy->max);
2175
2176         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2177
2178         /*
2179         * This check works well when we store new min/max freq attributes,
2180         * because new_policy is a copy of policy with one field updated.
2181         */
2182         if (new_policy->min > new_policy->max)
2183                 return -EINVAL;
2184
2185         /* verify the cpu speed can be set within this limit */
2186         ret = cpufreq_driver->verify(new_policy);
2187         if (ret)
2188                 return ret;
2189
2190         /* adjust if necessary - all reasons */
2191         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2192                         CPUFREQ_ADJUST, new_policy);
2193
2194         /*
2195          * verify the cpu speed can be set within this limit, which might be
2196          * different to the first one
2197          */
2198         ret = cpufreq_driver->verify(new_policy);
2199         if (ret)
2200                 return ret;
2201
2202         /* notification of the new policy */
2203         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2204                         CPUFREQ_NOTIFY, new_policy);
2205
2206         policy->min = new_policy->min;
2207         policy->max = new_policy->max;
2208
2209         policy->cached_target_freq = UINT_MAX;
2210
2211         pr_debug("new min and max freqs are %u - %u kHz\n",
2212                  policy->min, policy->max);
2213
2214         if (cpufreq_driver->setpolicy) {
2215                 policy->policy = new_policy->policy;
2216                 pr_debug("setting range\n");
2217                 return cpufreq_driver->setpolicy(new_policy);
2218         }
2219
2220         if (new_policy->governor == policy->governor) {
2221                 pr_debug("cpufreq: governor limits update\n");
2222                 cpufreq_governor_limits(policy);
2223                 return 0;
2224         }
2225
2226         pr_debug("governor switch\n");
2227
2228         /* save old, working values */
2229         old_gov = policy->governor;
2230         /* end old governor */
2231         if (old_gov) {
2232                 cpufreq_stop_governor(policy);
2233                 cpufreq_exit_governor(policy);
2234         }
2235
2236         /* start new governor */
2237         policy->governor = new_policy->governor;
2238         ret = cpufreq_init_governor(policy);
2239         if (!ret) {
2240                 ret = cpufreq_start_governor(policy);
2241                 if (!ret) {
2242                         pr_debug("cpufreq: governor change\n");
2243                         return 0;
2244                 }
2245                 cpufreq_exit_governor(policy);
2246         }
2247
2248         /* new governor failed, so re-start old one */
2249         pr_debug("starting governor %s failed\n", policy->governor->name);
2250         if (old_gov) {
2251                 policy->governor = old_gov;
2252                 if (cpufreq_init_governor(policy))
2253                         policy->governor = NULL;
2254                 else
2255                         cpufreq_start_governor(policy);
2256         }
2257
2258         return ret;
2259 }
2260
2261 /**
2262  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2263  *      @cpu: CPU which shall be re-evaluated
2264  *
2265  *      Useful for policy notifiers which have different necessities
2266  *      at different times.
2267  */
2268 void cpufreq_update_policy(unsigned int cpu)
2269 {
2270         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2271         struct cpufreq_policy new_policy;
2272
2273         if (!policy)
2274                 return;
2275
2276         down_write(&policy->rwsem);
2277
2278         if (policy_is_inactive(policy))
2279                 goto unlock;
2280
2281         pr_debug("updating policy for CPU %u\n", cpu);
2282         memcpy(&new_policy, policy, sizeof(*policy));
2283         new_policy.min = policy->user_policy.min;
2284         new_policy.max = policy->user_policy.max;
2285
2286         /*
2287          * BIOS might change freq behind our back
2288          * -> ask driver for current freq and notify governors about a change
2289          */
2290         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2291                 if (cpufreq_suspended)
2292                         goto unlock;
2293
2294                 new_policy.cur = cpufreq_update_current_freq(policy);
2295                 if (WARN_ON(!new_policy.cur))
2296                         goto unlock;
2297         }
2298
2299         cpufreq_set_policy(policy, &new_policy);
2300
2301 unlock:
2302         up_write(&policy->rwsem);
2303
2304         cpufreq_cpu_put(policy);
2305 }
2306 EXPORT_SYMBOL(cpufreq_update_policy);
2307
2308 /*********************************************************************
2309  *               BOOST                                               *
2310  *********************************************************************/
2311 static int cpufreq_boost_set_sw(int state)
2312 {
2313         struct cpufreq_policy *policy;
2314         int ret = -EINVAL;
2315
2316         for_each_active_policy(policy) {
2317                 if (!policy->freq_table)
2318                         continue;
2319
2320                 ret = cpufreq_frequency_table_cpuinfo(policy,
2321                                                       policy->freq_table);
2322                 if (ret) {
2323                         pr_err("%s: Policy frequency update failed\n",
2324                                __func__);
2325                         break;
2326                 }
2327
2328                 down_write(&policy->rwsem);
2329                 policy->user_policy.max = policy->max;
2330                 cpufreq_governor_limits(policy);
2331                 up_write(&policy->rwsem);
2332         }
2333
2334         return ret;
2335 }
2336
2337 int cpufreq_boost_trigger_state(int state)
2338 {
2339         unsigned long flags;
2340         int ret = 0;
2341
2342         if (cpufreq_driver->boost_enabled == state)
2343                 return 0;
2344
2345         write_lock_irqsave(&cpufreq_driver_lock, flags);
2346         cpufreq_driver->boost_enabled = state;
2347         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2348
2349         ret = cpufreq_driver->set_boost(state);
2350         if (ret) {
2351                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2352                 cpufreq_driver->boost_enabled = !state;
2353                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2354
2355                 pr_err("%s: Cannot %s BOOST\n",
2356                        __func__, state ? "enable" : "disable");
2357         }
2358
2359         return ret;
2360 }
2361
2362 static bool cpufreq_boost_supported(void)
2363 {
2364         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2365 }
2366
2367 static int create_boost_sysfs_file(void)
2368 {
2369         int ret;
2370
2371         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2372         if (ret)
2373                 pr_err("%s: cannot register global BOOST sysfs file\n",
2374                        __func__);
2375
2376         return ret;
2377 }
2378
2379 static void remove_boost_sysfs_file(void)
2380 {
2381         if (cpufreq_boost_supported())
2382                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2383 }
2384
2385 int cpufreq_enable_boost_support(void)
2386 {
2387         if (!cpufreq_driver)
2388                 return -EINVAL;
2389
2390         if (cpufreq_boost_supported())
2391                 return 0;
2392
2393         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2394
2395         /* This will get removed on driver unregister */
2396         return create_boost_sysfs_file();
2397 }
2398 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2399
2400 int cpufreq_boost_enabled(void)
2401 {
2402         return cpufreq_driver->boost_enabled;
2403 }
2404 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2405
2406 /*********************************************************************
2407  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2408  *********************************************************************/
2409 static enum cpuhp_state hp_online;
2410
2411 static int cpuhp_cpufreq_online(unsigned int cpu)
2412 {
2413         cpufreq_online(cpu);
2414
2415         return 0;
2416 }
2417
2418 static int cpuhp_cpufreq_offline(unsigned int cpu)
2419 {
2420         cpufreq_offline(cpu);
2421
2422         return 0;
2423 }
2424
2425 /**
2426  * cpufreq_register_driver - register a CPU Frequency driver
2427  * @driver_data: A struct cpufreq_driver containing the values#
2428  * submitted by the CPU Frequency driver.
2429  *
2430  * Registers a CPU Frequency driver to this core code. This code
2431  * returns zero on success, -EEXIST when another driver got here first
2432  * (and isn't unregistered in the meantime).
2433  *
2434  */
2435 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2436 {
2437         unsigned long flags;
2438         int ret;
2439
2440         if (cpufreq_disabled())
2441                 return -ENODEV;
2442
2443         if (!driver_data || !driver_data->verify || !driver_data->init ||
2444             !(driver_data->setpolicy || driver_data->target_index ||
2445                     driver_data->target) ||
2446              (driver_data->setpolicy && (driver_data->target_index ||
2447                     driver_data->target)) ||
2448              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2449                 return -EINVAL;
2450
2451         pr_debug("trying to register driver %s\n", driver_data->name);
2452
2453         /* Protect against concurrent CPU online/offline. */
2454         cpus_read_lock();
2455
2456         write_lock_irqsave(&cpufreq_driver_lock, flags);
2457         if (cpufreq_driver) {
2458                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2459                 ret = -EEXIST;
2460                 goto out;
2461         }
2462         cpufreq_driver = driver_data;
2463         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2464
2465         if (driver_data->setpolicy)
2466                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2467
2468         if (cpufreq_boost_supported()) {
2469                 ret = create_boost_sysfs_file();
2470                 if (ret)
2471                         goto err_null_driver;
2472         }
2473
2474         ret = subsys_interface_register(&cpufreq_interface);
2475         if (ret)
2476                 goto err_boost_unreg;
2477
2478         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2479             list_empty(&cpufreq_policy_list)) {
2480                 /* if all ->init() calls failed, unregister */
2481                 ret = -ENODEV;
2482                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2483                          driver_data->name);
2484                 goto err_if_unreg;
2485         }
2486
2487         ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2488                                                    "cpufreq:online",
2489                                                    cpuhp_cpufreq_online,
2490                                                    cpuhp_cpufreq_offline);
2491         if (ret < 0)
2492                 goto err_if_unreg;
2493         hp_online = ret;
2494         ret = 0;
2495
2496         pr_debug("driver %s up and running\n", driver_data->name);
2497         goto out;
2498
2499 err_if_unreg:
2500         subsys_interface_unregister(&cpufreq_interface);
2501 err_boost_unreg:
2502         remove_boost_sysfs_file();
2503 err_null_driver:
2504         write_lock_irqsave(&cpufreq_driver_lock, flags);
2505         cpufreq_driver = NULL;
2506         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2507 out:
2508         cpus_read_unlock();
2509         return ret;
2510 }
2511 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2512
2513 /**
2514  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2515  *
2516  * Unregister the current CPUFreq driver. Only call this if you have
2517  * the right to do so, i.e. if you have succeeded in initialising before!
2518  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2519  * currently not initialised.
2520  */
2521 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2522 {
2523         unsigned long flags;
2524
2525         if (!cpufreq_driver || (driver != cpufreq_driver))
2526                 return -EINVAL;
2527
2528         pr_debug("unregistering driver %s\n", driver->name);
2529
2530         /* Protect against concurrent cpu hotplug */
2531         cpus_read_lock();
2532         subsys_interface_unregister(&cpufreq_interface);
2533         remove_boost_sysfs_file();
2534         cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2535
2536         write_lock_irqsave(&cpufreq_driver_lock, flags);
2537
2538         cpufreq_driver = NULL;
2539
2540         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2541         cpus_read_unlock();
2542
2543         return 0;
2544 }
2545 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2546
2547 /*
2548  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2549  * or mutexes when secondary CPUs are halted.
2550  */
2551 static struct syscore_ops cpufreq_syscore_ops = {
2552         .shutdown = cpufreq_suspend,
2553 };
2554
2555 struct kobject *cpufreq_global_kobject;
2556 EXPORT_SYMBOL(cpufreq_global_kobject);
2557
2558 static int __init cpufreq_core_init(void)
2559 {
2560         if (cpufreq_disabled())
2561                 return -ENODEV;
2562
2563         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2564         BUG_ON(!cpufreq_global_kobject);
2565
2566         register_syscore_ops(&cpufreq_syscore_ops);
2567
2568         return 0;
2569 }
2570 module_param(off, int, 0444);
2571 core_initcall(cpufreq_core_init);