net: phy: Export mdiobus_register_board_info()
[linux-2.6-microblaze.git] / drivers / cpufreq / cpufreq.c
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
7  *
8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
9  *      Added handling for CPU hotplug
10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
11  *      Fix handling for CPU hotplug -- affected CPUs
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License version 2 as
15  * published by the Free Software Foundation.
16  */
17
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20 #include <linux/cpu.h>
21 #include <linux/cpufreq.h>
22 #include <linux/delay.h>
23 #include <linux/device.h>
24 #include <linux/init.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/module.h>
27 #include <linux/mutex.h>
28 #include <linux/slab.h>
29 #include <linux/suspend.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/tick.h>
32 #include <trace/events/power.h>
33
34 static LIST_HEAD(cpufreq_policy_list);
35
36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
37 {
38         return cpumask_empty(policy->cpus);
39 }
40
41 /* Macros to iterate over CPU policies */
42 #define for_each_suitable_policy(__policy, __active)                     \
43         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
44                 if ((__active) == !policy_is_inactive(__policy))
45
46 #define for_each_active_policy(__policy)                \
47         for_each_suitable_policy(__policy, true)
48 #define for_each_inactive_policy(__policy)              \
49         for_each_suitable_policy(__policy, false)
50
51 #define for_each_policy(__policy)                       \
52         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
53
54 /* Iterate over governors */
55 static LIST_HEAD(cpufreq_governor_list);
56 #define for_each_governor(__governor)                           \
57         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
58
59 /**
60  * The "cpufreq driver" - the arch- or hardware-dependent low
61  * level driver of CPUFreq support, and its spinlock. This lock
62  * also protects the cpufreq_cpu_data array.
63  */
64 static struct cpufreq_driver *cpufreq_driver;
65 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
66 static DEFINE_RWLOCK(cpufreq_driver_lock);
67
68 /* Flag to suspend/resume CPUFreq governors */
69 static bool cpufreq_suspended;
70
71 static inline bool has_target(void)
72 {
73         return cpufreq_driver->target_index || cpufreq_driver->target;
74 }
75
76 /* internal prototypes */
77 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
78 static int cpufreq_init_governor(struct cpufreq_policy *policy);
79 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
80 static int cpufreq_start_governor(struct cpufreq_policy *policy);
81 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
82 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
83
84 /**
85  * Two notifier lists: the "policy" list is involved in the
86  * validation process for a new CPU frequency policy; the
87  * "transition" list for kernel code that needs to handle
88  * changes to devices when the CPU clock speed changes.
89  * The mutex locks both lists.
90  */
91 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
92 static struct srcu_notifier_head cpufreq_transition_notifier_list;
93
94 static bool init_cpufreq_transition_notifier_list_called;
95 static int __init init_cpufreq_transition_notifier_list(void)
96 {
97         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
98         init_cpufreq_transition_notifier_list_called = true;
99         return 0;
100 }
101 pure_initcall(init_cpufreq_transition_notifier_list);
102
103 static int off __read_mostly;
104 static int cpufreq_disabled(void)
105 {
106         return off;
107 }
108 void disable_cpufreq(void)
109 {
110         off = 1;
111 }
112 static DEFINE_MUTEX(cpufreq_governor_mutex);
113
114 bool have_governor_per_policy(void)
115 {
116         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
117 }
118 EXPORT_SYMBOL_GPL(have_governor_per_policy);
119
120 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
121 {
122         if (have_governor_per_policy())
123                 return &policy->kobj;
124         else
125                 return cpufreq_global_kobject;
126 }
127 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
128
129 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
130 {
131         u64 idle_time;
132         u64 cur_wall_time;
133         u64 busy_time;
134
135         cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
136
137         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
138         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
139         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
140         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
141         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
142         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
143
144         idle_time = cur_wall_time - busy_time;
145         if (wall)
146                 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
147
148         return div_u64(idle_time, NSEC_PER_USEC);
149 }
150
151 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
152 {
153         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
154
155         if (idle_time == -1ULL)
156                 return get_cpu_idle_time_jiffy(cpu, wall);
157         else if (!io_busy)
158                 idle_time += get_cpu_iowait_time_us(cpu, wall);
159
160         return idle_time;
161 }
162 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
163
164 /*
165  * This is a generic cpufreq init() routine which can be used by cpufreq
166  * drivers of SMP systems. It will do following:
167  * - validate & show freq table passed
168  * - set policies transition latency
169  * - policy->cpus with all possible CPUs
170  */
171 int cpufreq_generic_init(struct cpufreq_policy *policy,
172                 struct cpufreq_frequency_table *table,
173                 unsigned int transition_latency)
174 {
175         int ret;
176
177         ret = cpufreq_table_validate_and_show(policy, table);
178         if (ret) {
179                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
180                 return ret;
181         }
182
183         policy->cpuinfo.transition_latency = transition_latency;
184
185         /*
186          * The driver only supports the SMP configuration where all processors
187          * share the clock and voltage and clock.
188          */
189         cpumask_setall(policy->cpus);
190
191         return 0;
192 }
193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
194
195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
196 {
197         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
198
199         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
200 }
201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
202
203 unsigned int cpufreq_generic_get(unsigned int cpu)
204 {
205         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
206
207         if (!policy || IS_ERR(policy->clk)) {
208                 pr_err("%s: No %s associated to cpu: %d\n",
209                        __func__, policy ? "clk" : "policy", cpu);
210                 return 0;
211         }
212
213         return clk_get_rate(policy->clk) / 1000;
214 }
215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
216
217 /**
218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
219  *
220  * @cpu: cpu to find policy for.
221  *
222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
223  * It also increments the kobject reference count to mark it busy and so would
224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
226  * freed as that depends on the kobj count.
227  *
228  * Return: A valid policy on success, otherwise NULL on failure.
229  */
230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
231 {
232         struct cpufreq_policy *policy = NULL;
233         unsigned long flags;
234
235         if (WARN_ON(cpu >= nr_cpu_ids))
236                 return NULL;
237
238         /* get the cpufreq driver */
239         read_lock_irqsave(&cpufreq_driver_lock, flags);
240
241         if (cpufreq_driver) {
242                 /* get the CPU */
243                 policy = cpufreq_cpu_get_raw(cpu);
244                 if (policy)
245                         kobject_get(&policy->kobj);
246         }
247
248         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
249
250         return policy;
251 }
252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
253
254 /**
255  * cpufreq_cpu_put: Decrements the usage count of a policy
256  *
257  * @policy: policy earlier returned by cpufreq_cpu_get().
258  *
259  * This decrements the kobject reference count incremented earlier by calling
260  * cpufreq_cpu_get().
261  */
262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
263 {
264         kobject_put(&policy->kobj);
265 }
266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
267
268 /*********************************************************************
269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
270  *********************************************************************/
271
272 /**
273  * adjust_jiffies - adjust the system "loops_per_jiffy"
274  *
275  * This function alters the system "loops_per_jiffy" for the clock
276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
277  * systems as each CPU might be scaled differently. So, use the arch
278  * per-CPU loops_per_jiffy value wherever possible.
279  */
280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
281 {
282 #ifndef CONFIG_SMP
283         static unsigned long l_p_j_ref;
284         static unsigned int l_p_j_ref_freq;
285
286         if (ci->flags & CPUFREQ_CONST_LOOPS)
287                 return;
288
289         if (!l_p_j_ref_freq) {
290                 l_p_j_ref = loops_per_jiffy;
291                 l_p_j_ref_freq = ci->old;
292                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
293                          l_p_j_ref, l_p_j_ref_freq);
294         }
295         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297                                                                 ci->new);
298                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
299                          loops_per_jiffy, ci->new);
300         }
301 #endif
302 }
303
304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
305                 struct cpufreq_freqs *freqs, unsigned int state)
306 {
307         BUG_ON(irqs_disabled());
308
309         if (cpufreq_disabled())
310                 return;
311
312         freqs->flags = cpufreq_driver->flags;
313         pr_debug("notification %u of frequency transition to %u kHz\n",
314                  state, freqs->new);
315
316         switch (state) {
317
318         case CPUFREQ_PRECHANGE:
319                 /* detect if the driver reported a value as "old frequency"
320                  * which is not equal to what the cpufreq core thinks is
321                  * "old frequency".
322                  */
323                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
324                         if ((policy) && (policy->cpu == freqs->cpu) &&
325                             (policy->cur) && (policy->cur != freqs->old)) {
326                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
327                                          freqs->old, policy->cur);
328                                 freqs->old = policy->cur;
329                         }
330                 }
331                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
332                                 CPUFREQ_PRECHANGE, freqs);
333                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
334                 break;
335
336         case CPUFREQ_POSTCHANGE:
337                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
338                 pr_debug("FREQ: %lu - CPU: %lu\n",
339                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
340                 trace_cpu_frequency(freqs->new, freqs->cpu);
341                 cpufreq_stats_record_transition(policy, freqs->new);
342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
343                                 CPUFREQ_POSTCHANGE, freqs);
344                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
345                         policy->cur = freqs->new;
346                 break;
347         }
348 }
349
350 /**
351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
352  * on frequency transition.
353  *
354  * This function calls the transition notifiers and the "adjust_jiffies"
355  * function. It is called twice on all CPU frequency changes that have
356  * external effects.
357  */
358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
359                 struct cpufreq_freqs *freqs, unsigned int state)
360 {
361         for_each_cpu(freqs->cpu, policy->cpus)
362                 __cpufreq_notify_transition(policy, freqs, state);
363 }
364
365 /* Do post notifications when there are chances that transition has failed */
366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
367                 struct cpufreq_freqs *freqs, int transition_failed)
368 {
369         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
370         if (!transition_failed)
371                 return;
372
373         swap(freqs->old, freqs->new);
374         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
376 }
377
378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
379                 struct cpufreq_freqs *freqs)
380 {
381
382         /*
383          * Catch double invocations of _begin() which lead to self-deadlock.
384          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
385          * doesn't invoke _begin() on their behalf, and hence the chances of
386          * double invocations are very low. Moreover, there are scenarios
387          * where these checks can emit false-positive warnings in these
388          * drivers; so we avoid that by skipping them altogether.
389          */
390         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
391                                 && current == policy->transition_task);
392
393 wait:
394         wait_event(policy->transition_wait, !policy->transition_ongoing);
395
396         spin_lock(&policy->transition_lock);
397
398         if (unlikely(policy->transition_ongoing)) {
399                 spin_unlock(&policy->transition_lock);
400                 goto wait;
401         }
402
403         policy->transition_ongoing = true;
404         policy->transition_task = current;
405
406         spin_unlock(&policy->transition_lock);
407
408         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
409 }
410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
411
412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
413                 struct cpufreq_freqs *freqs, int transition_failed)
414 {
415         if (unlikely(WARN_ON(!policy->transition_ongoing)))
416                 return;
417
418         cpufreq_notify_post_transition(policy, freqs, transition_failed);
419
420         policy->transition_ongoing = false;
421         policy->transition_task = NULL;
422
423         wake_up(&policy->transition_wait);
424 }
425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
426
427 /*
428  * Fast frequency switching status count.  Positive means "enabled", negative
429  * means "disabled" and 0 means "not decided yet".
430  */
431 static int cpufreq_fast_switch_count;
432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
433
434 static void cpufreq_list_transition_notifiers(void)
435 {
436         struct notifier_block *nb;
437
438         pr_info("Registered transition notifiers:\n");
439
440         mutex_lock(&cpufreq_transition_notifier_list.mutex);
441
442         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
443                 pr_info("%pF\n", nb->notifier_call);
444
445         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
446 }
447
448 /**
449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
450  * @policy: cpufreq policy to enable fast frequency switching for.
451  *
452  * Try to enable fast frequency switching for @policy.
453  *
454  * The attempt will fail if there is at least one transition notifier registered
455  * at this point, as fast frequency switching is quite fundamentally at odds
456  * with transition notifiers.  Thus if successful, it will make registration of
457  * transition notifiers fail going forward.
458  */
459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
460 {
461         lockdep_assert_held(&policy->rwsem);
462
463         if (!policy->fast_switch_possible)
464                 return;
465
466         mutex_lock(&cpufreq_fast_switch_lock);
467         if (cpufreq_fast_switch_count >= 0) {
468                 cpufreq_fast_switch_count++;
469                 policy->fast_switch_enabled = true;
470         } else {
471                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
472                         policy->cpu);
473                 cpufreq_list_transition_notifiers();
474         }
475         mutex_unlock(&cpufreq_fast_switch_lock);
476 }
477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
478
479 /**
480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
481  * @policy: cpufreq policy to disable fast frequency switching for.
482  */
483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
484 {
485         mutex_lock(&cpufreq_fast_switch_lock);
486         if (policy->fast_switch_enabled) {
487                 policy->fast_switch_enabled = false;
488                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
489                         cpufreq_fast_switch_count--;
490         }
491         mutex_unlock(&cpufreq_fast_switch_lock);
492 }
493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
494
495 /**
496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
497  * one.
498  * @target_freq: target frequency to resolve.
499  *
500  * The target to driver frequency mapping is cached in the policy.
501  *
502  * Return: Lowest driver-supported frequency greater than or equal to the
503  * given target_freq, subject to policy (min/max) and driver limitations.
504  */
505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
506                                          unsigned int target_freq)
507 {
508         target_freq = clamp_val(target_freq, policy->min, policy->max);
509         policy->cached_target_freq = target_freq;
510
511         if (cpufreq_driver->target_index) {
512                 int idx;
513
514                 idx = cpufreq_frequency_table_target(policy, target_freq,
515                                                      CPUFREQ_RELATION_L);
516                 policy->cached_resolved_idx = idx;
517                 return policy->freq_table[idx].frequency;
518         }
519
520         if (cpufreq_driver->resolve_freq)
521                 return cpufreq_driver->resolve_freq(policy, target_freq);
522
523         return target_freq;
524 }
525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
526
527 /*********************************************************************
528  *                          SYSFS INTERFACE                          *
529  *********************************************************************/
530 static ssize_t show_boost(struct kobject *kobj,
531                                  struct attribute *attr, char *buf)
532 {
533         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
534 }
535
536 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
537                                   const char *buf, size_t count)
538 {
539         int ret, enable;
540
541         ret = sscanf(buf, "%d", &enable);
542         if (ret != 1 || enable < 0 || enable > 1)
543                 return -EINVAL;
544
545         if (cpufreq_boost_trigger_state(enable)) {
546                 pr_err("%s: Cannot %s BOOST!\n",
547                        __func__, enable ? "enable" : "disable");
548                 return -EINVAL;
549         }
550
551         pr_debug("%s: cpufreq BOOST %s\n",
552                  __func__, enable ? "enabled" : "disabled");
553
554         return count;
555 }
556 define_one_global_rw(boost);
557
558 static struct cpufreq_governor *find_governor(const char *str_governor)
559 {
560         struct cpufreq_governor *t;
561
562         for_each_governor(t)
563                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
564                         return t;
565
566         return NULL;
567 }
568
569 /**
570  * cpufreq_parse_governor - parse a governor string
571  */
572 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
573                                 struct cpufreq_governor **governor)
574 {
575         int err = -EINVAL;
576
577         if (cpufreq_driver->setpolicy) {
578                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
579                         *policy = CPUFREQ_POLICY_PERFORMANCE;
580                         err = 0;
581                 } else if (!strncasecmp(str_governor, "powersave",
582                                                 CPUFREQ_NAME_LEN)) {
583                         *policy = CPUFREQ_POLICY_POWERSAVE;
584                         err = 0;
585                 }
586         } else {
587                 struct cpufreq_governor *t;
588
589                 mutex_lock(&cpufreq_governor_mutex);
590
591                 t = find_governor(str_governor);
592
593                 if (t == NULL) {
594                         int ret;
595
596                         mutex_unlock(&cpufreq_governor_mutex);
597                         ret = request_module("cpufreq_%s", str_governor);
598                         mutex_lock(&cpufreq_governor_mutex);
599
600                         if (ret == 0)
601                                 t = find_governor(str_governor);
602                 }
603
604                 if (t != NULL) {
605                         *governor = t;
606                         err = 0;
607                 }
608
609                 mutex_unlock(&cpufreq_governor_mutex);
610         }
611         return err;
612 }
613
614 /**
615  * cpufreq_per_cpu_attr_read() / show_##file_name() -
616  * print out cpufreq information
617  *
618  * Write out information from cpufreq_driver->policy[cpu]; object must be
619  * "unsigned int".
620  */
621
622 #define show_one(file_name, object)                     \
623 static ssize_t show_##file_name                         \
624 (struct cpufreq_policy *policy, char *buf)              \
625 {                                                       \
626         return sprintf(buf, "%u\n", policy->object);    \
627 }
628
629 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
630 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
631 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
632 show_one(scaling_min_freq, min);
633 show_one(scaling_max_freq, max);
634
635 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
636 {
637         ssize_t ret;
638
639         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
640                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
641         else
642                 ret = sprintf(buf, "%u\n", policy->cur);
643         return ret;
644 }
645
646 static int cpufreq_set_policy(struct cpufreq_policy *policy,
647                                 struct cpufreq_policy *new_policy);
648
649 /**
650  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
651  */
652 #define store_one(file_name, object)                    \
653 static ssize_t store_##file_name                                        \
654 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
655 {                                                                       \
656         int ret, temp;                                                  \
657         struct cpufreq_policy new_policy;                               \
658                                                                         \
659         memcpy(&new_policy, policy, sizeof(*policy));                   \
660                                                                         \
661         ret = sscanf(buf, "%u", &new_policy.object);                    \
662         if (ret != 1)                                                   \
663                 return -EINVAL;                                         \
664                                                                         \
665         temp = new_policy.object;                                       \
666         ret = cpufreq_set_policy(policy, &new_policy);          \
667         if (!ret)                                                       \
668                 policy->user_policy.object = temp;                      \
669                                                                         \
670         return ret ? ret : count;                                       \
671 }
672
673 store_one(scaling_min_freq, min);
674 store_one(scaling_max_freq, max);
675
676 /**
677  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
678  */
679 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
680                                         char *buf)
681 {
682         unsigned int cur_freq = __cpufreq_get(policy);
683
684         if (cur_freq)
685                 return sprintf(buf, "%u\n", cur_freq);
686
687         return sprintf(buf, "<unknown>\n");
688 }
689
690 /**
691  * show_scaling_governor - show the current policy for the specified CPU
692  */
693 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
694 {
695         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
696                 return sprintf(buf, "powersave\n");
697         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
698                 return sprintf(buf, "performance\n");
699         else if (policy->governor)
700                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
701                                 policy->governor->name);
702         return -EINVAL;
703 }
704
705 /**
706  * store_scaling_governor - store policy for the specified CPU
707  */
708 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
709                                         const char *buf, size_t count)
710 {
711         int ret;
712         char    str_governor[16];
713         struct cpufreq_policy new_policy;
714
715         memcpy(&new_policy, policy, sizeof(*policy));
716
717         ret = sscanf(buf, "%15s", str_governor);
718         if (ret != 1)
719                 return -EINVAL;
720
721         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
722                                                 &new_policy.governor))
723                 return -EINVAL;
724
725         ret = cpufreq_set_policy(policy, &new_policy);
726         return ret ? ret : count;
727 }
728
729 /**
730  * show_scaling_driver - show the cpufreq driver currently loaded
731  */
732 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
733 {
734         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
735 }
736
737 /**
738  * show_scaling_available_governors - show the available CPUfreq governors
739  */
740 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
741                                                 char *buf)
742 {
743         ssize_t i = 0;
744         struct cpufreq_governor *t;
745
746         if (!has_target()) {
747                 i += sprintf(buf, "performance powersave");
748                 goto out;
749         }
750
751         for_each_governor(t) {
752                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
753                     - (CPUFREQ_NAME_LEN + 2)))
754                         goto out;
755                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
756         }
757 out:
758         i += sprintf(&buf[i], "\n");
759         return i;
760 }
761
762 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
763 {
764         ssize_t i = 0;
765         unsigned int cpu;
766
767         for_each_cpu(cpu, mask) {
768                 if (i)
769                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
770                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
771                 if (i >= (PAGE_SIZE - 5))
772                         break;
773         }
774         i += sprintf(&buf[i], "\n");
775         return i;
776 }
777 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
778
779 /**
780  * show_related_cpus - show the CPUs affected by each transition even if
781  * hw coordination is in use
782  */
783 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
784 {
785         return cpufreq_show_cpus(policy->related_cpus, buf);
786 }
787
788 /**
789  * show_affected_cpus - show the CPUs affected by each transition
790  */
791 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
792 {
793         return cpufreq_show_cpus(policy->cpus, buf);
794 }
795
796 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
797                                         const char *buf, size_t count)
798 {
799         unsigned int freq = 0;
800         unsigned int ret;
801
802         if (!policy->governor || !policy->governor->store_setspeed)
803                 return -EINVAL;
804
805         ret = sscanf(buf, "%u", &freq);
806         if (ret != 1)
807                 return -EINVAL;
808
809         policy->governor->store_setspeed(policy, freq);
810
811         return count;
812 }
813
814 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
815 {
816         if (!policy->governor || !policy->governor->show_setspeed)
817                 return sprintf(buf, "<unsupported>\n");
818
819         return policy->governor->show_setspeed(policy, buf);
820 }
821
822 /**
823  * show_bios_limit - show the current cpufreq HW/BIOS limitation
824  */
825 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
826 {
827         unsigned int limit;
828         int ret;
829         if (cpufreq_driver->bios_limit) {
830                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
831                 if (!ret)
832                         return sprintf(buf, "%u\n", limit);
833         }
834         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
835 }
836
837 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
838 cpufreq_freq_attr_ro(cpuinfo_min_freq);
839 cpufreq_freq_attr_ro(cpuinfo_max_freq);
840 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
841 cpufreq_freq_attr_ro(scaling_available_governors);
842 cpufreq_freq_attr_ro(scaling_driver);
843 cpufreq_freq_attr_ro(scaling_cur_freq);
844 cpufreq_freq_attr_ro(bios_limit);
845 cpufreq_freq_attr_ro(related_cpus);
846 cpufreq_freq_attr_ro(affected_cpus);
847 cpufreq_freq_attr_rw(scaling_min_freq);
848 cpufreq_freq_attr_rw(scaling_max_freq);
849 cpufreq_freq_attr_rw(scaling_governor);
850 cpufreq_freq_attr_rw(scaling_setspeed);
851
852 static struct attribute *default_attrs[] = {
853         &cpuinfo_min_freq.attr,
854         &cpuinfo_max_freq.attr,
855         &cpuinfo_transition_latency.attr,
856         &scaling_min_freq.attr,
857         &scaling_max_freq.attr,
858         &affected_cpus.attr,
859         &related_cpus.attr,
860         &scaling_governor.attr,
861         &scaling_driver.attr,
862         &scaling_available_governors.attr,
863         &scaling_setspeed.attr,
864         NULL
865 };
866
867 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
868 #define to_attr(a) container_of(a, struct freq_attr, attr)
869
870 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
871 {
872         struct cpufreq_policy *policy = to_policy(kobj);
873         struct freq_attr *fattr = to_attr(attr);
874         ssize_t ret;
875
876         down_read(&policy->rwsem);
877         ret = fattr->show(policy, buf);
878         up_read(&policy->rwsem);
879
880         return ret;
881 }
882
883 static ssize_t store(struct kobject *kobj, struct attribute *attr,
884                      const char *buf, size_t count)
885 {
886         struct cpufreq_policy *policy = to_policy(kobj);
887         struct freq_attr *fattr = to_attr(attr);
888         ssize_t ret = -EINVAL;
889
890         get_online_cpus();
891
892         if (cpu_online(policy->cpu)) {
893                 down_write(&policy->rwsem);
894                 ret = fattr->store(policy, buf, count);
895                 up_write(&policy->rwsem);
896         }
897
898         put_online_cpus();
899
900         return ret;
901 }
902
903 static void cpufreq_sysfs_release(struct kobject *kobj)
904 {
905         struct cpufreq_policy *policy = to_policy(kobj);
906         pr_debug("last reference is dropped\n");
907         complete(&policy->kobj_unregister);
908 }
909
910 static const struct sysfs_ops sysfs_ops = {
911         .show   = show,
912         .store  = store,
913 };
914
915 static struct kobj_type ktype_cpufreq = {
916         .sysfs_ops      = &sysfs_ops,
917         .default_attrs  = default_attrs,
918         .release        = cpufreq_sysfs_release,
919 };
920
921 static int add_cpu_dev_symlink(struct cpufreq_policy *policy,
922                                struct device *dev)
923 {
924         dev_dbg(dev, "%s: Adding symlink\n", __func__);
925         return sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
926 }
927
928 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
929                                    struct device *dev)
930 {
931         dev_dbg(dev, "%s: Removing symlink\n", __func__);
932         sysfs_remove_link(&dev->kobj, "cpufreq");
933 }
934
935 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
936 {
937         struct freq_attr **drv_attr;
938         int ret = 0;
939
940         /* set up files for this cpu device */
941         drv_attr = cpufreq_driver->attr;
942         while (drv_attr && *drv_attr) {
943                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
944                 if (ret)
945                         return ret;
946                 drv_attr++;
947         }
948         if (cpufreq_driver->get) {
949                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
950                 if (ret)
951                         return ret;
952         }
953
954         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
955         if (ret)
956                 return ret;
957
958         if (cpufreq_driver->bios_limit) {
959                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
960                 if (ret)
961                         return ret;
962         }
963
964         return 0;
965 }
966
967 __weak struct cpufreq_governor *cpufreq_default_governor(void)
968 {
969         return NULL;
970 }
971
972 static int cpufreq_init_policy(struct cpufreq_policy *policy)
973 {
974         struct cpufreq_governor *gov = NULL;
975         struct cpufreq_policy new_policy;
976
977         memcpy(&new_policy, policy, sizeof(*policy));
978
979         /* Update governor of new_policy to the governor used before hotplug */
980         gov = find_governor(policy->last_governor);
981         if (gov) {
982                 pr_debug("Restoring governor %s for cpu %d\n",
983                                 policy->governor->name, policy->cpu);
984         } else {
985                 gov = cpufreq_default_governor();
986                 if (!gov)
987                         return -ENODATA;
988         }
989
990         new_policy.governor = gov;
991
992         /* Use the default policy if there is no last_policy. */
993         if (cpufreq_driver->setpolicy) {
994                 if (policy->last_policy)
995                         new_policy.policy = policy->last_policy;
996                 else
997                         cpufreq_parse_governor(gov->name, &new_policy.policy,
998                                                NULL);
999         }
1000         /* set default policy */
1001         return cpufreq_set_policy(policy, &new_policy);
1002 }
1003
1004 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1005 {
1006         int ret = 0;
1007
1008         /* Has this CPU been taken care of already? */
1009         if (cpumask_test_cpu(cpu, policy->cpus))
1010                 return 0;
1011
1012         down_write(&policy->rwsem);
1013         if (has_target())
1014                 cpufreq_stop_governor(policy);
1015
1016         cpumask_set_cpu(cpu, policy->cpus);
1017
1018         if (has_target()) {
1019                 ret = cpufreq_start_governor(policy);
1020                 if (ret)
1021                         pr_err("%s: Failed to start governor\n", __func__);
1022         }
1023         up_write(&policy->rwsem);
1024         return ret;
1025 }
1026
1027 static void handle_update(struct work_struct *work)
1028 {
1029         struct cpufreq_policy *policy =
1030                 container_of(work, struct cpufreq_policy, update);
1031         unsigned int cpu = policy->cpu;
1032         pr_debug("handle_update for cpu %u called\n", cpu);
1033         cpufreq_update_policy(cpu);
1034 }
1035
1036 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1037 {
1038         struct cpufreq_policy *policy;
1039         int ret;
1040
1041         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1042         if (!policy)
1043                 return NULL;
1044
1045         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1046                 goto err_free_policy;
1047
1048         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1049                 goto err_free_cpumask;
1050
1051         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1052                 goto err_free_rcpumask;
1053
1054         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1055                                    cpufreq_global_kobject, "policy%u", cpu);
1056         if (ret) {
1057                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1058                 goto err_free_real_cpus;
1059         }
1060
1061         INIT_LIST_HEAD(&policy->policy_list);
1062         init_rwsem(&policy->rwsem);
1063         spin_lock_init(&policy->transition_lock);
1064         init_waitqueue_head(&policy->transition_wait);
1065         init_completion(&policy->kobj_unregister);
1066         INIT_WORK(&policy->update, handle_update);
1067
1068         policy->cpu = cpu;
1069         return policy;
1070
1071 err_free_real_cpus:
1072         free_cpumask_var(policy->real_cpus);
1073 err_free_rcpumask:
1074         free_cpumask_var(policy->related_cpus);
1075 err_free_cpumask:
1076         free_cpumask_var(policy->cpus);
1077 err_free_policy:
1078         kfree(policy);
1079
1080         return NULL;
1081 }
1082
1083 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1084 {
1085         struct kobject *kobj;
1086         struct completion *cmp;
1087
1088         down_write(&policy->rwsem);
1089         cpufreq_stats_free_table(policy);
1090         kobj = &policy->kobj;
1091         cmp = &policy->kobj_unregister;
1092         up_write(&policy->rwsem);
1093         kobject_put(kobj);
1094
1095         /*
1096          * We need to make sure that the underlying kobj is
1097          * actually not referenced anymore by anybody before we
1098          * proceed with unloading.
1099          */
1100         pr_debug("waiting for dropping of refcount\n");
1101         wait_for_completion(cmp);
1102         pr_debug("wait complete\n");
1103 }
1104
1105 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1106 {
1107         unsigned long flags;
1108         int cpu;
1109
1110         /* Remove policy from list */
1111         write_lock_irqsave(&cpufreq_driver_lock, flags);
1112         list_del(&policy->policy_list);
1113
1114         for_each_cpu(cpu, policy->related_cpus)
1115                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1116         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1117
1118         cpufreq_policy_put_kobj(policy);
1119         free_cpumask_var(policy->real_cpus);
1120         free_cpumask_var(policy->related_cpus);
1121         free_cpumask_var(policy->cpus);
1122         kfree(policy);
1123 }
1124
1125 static int cpufreq_online(unsigned int cpu)
1126 {
1127         struct cpufreq_policy *policy;
1128         bool new_policy;
1129         unsigned long flags;
1130         unsigned int j;
1131         int ret;
1132
1133         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1134
1135         /* Check if this CPU already has a policy to manage it */
1136         policy = per_cpu(cpufreq_cpu_data, cpu);
1137         if (policy) {
1138                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1139                 if (!policy_is_inactive(policy))
1140                         return cpufreq_add_policy_cpu(policy, cpu);
1141
1142                 /* This is the only online CPU for the policy.  Start over. */
1143                 new_policy = false;
1144                 down_write(&policy->rwsem);
1145                 policy->cpu = cpu;
1146                 policy->governor = NULL;
1147                 up_write(&policy->rwsem);
1148         } else {
1149                 new_policy = true;
1150                 policy = cpufreq_policy_alloc(cpu);
1151                 if (!policy)
1152                         return -ENOMEM;
1153         }
1154
1155         cpumask_copy(policy->cpus, cpumask_of(cpu));
1156
1157         /* call driver. From then on the cpufreq must be able
1158          * to accept all calls to ->verify and ->setpolicy for this CPU
1159          */
1160         ret = cpufreq_driver->init(policy);
1161         if (ret) {
1162                 pr_debug("initialization failed\n");
1163                 goto out_free_policy;
1164         }
1165
1166         down_write(&policy->rwsem);
1167
1168         if (new_policy) {
1169                 /* related_cpus should at least include policy->cpus. */
1170                 cpumask_copy(policy->related_cpus, policy->cpus);
1171         }
1172
1173         /*
1174          * affected cpus must always be the one, which are online. We aren't
1175          * managing offline cpus here.
1176          */
1177         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1178
1179         if (new_policy) {
1180                 policy->user_policy.min = policy->min;
1181                 policy->user_policy.max = policy->max;
1182
1183                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1184                 for_each_cpu(j, policy->related_cpus)
1185                         per_cpu(cpufreq_cpu_data, j) = policy;
1186                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1187         }
1188
1189         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1190                 policy->cur = cpufreq_driver->get(policy->cpu);
1191                 if (!policy->cur) {
1192                         pr_err("%s: ->get() failed\n", __func__);
1193                         goto out_exit_policy;
1194                 }
1195         }
1196
1197         /*
1198          * Sometimes boot loaders set CPU frequency to a value outside of
1199          * frequency table present with cpufreq core. In such cases CPU might be
1200          * unstable if it has to run on that frequency for long duration of time
1201          * and so its better to set it to a frequency which is specified in
1202          * freq-table. This also makes cpufreq stats inconsistent as
1203          * cpufreq-stats would fail to register because current frequency of CPU
1204          * isn't found in freq-table.
1205          *
1206          * Because we don't want this change to effect boot process badly, we go
1207          * for the next freq which is >= policy->cur ('cur' must be set by now,
1208          * otherwise we will end up setting freq to lowest of the table as 'cur'
1209          * is initialized to zero).
1210          *
1211          * We are passing target-freq as "policy->cur - 1" otherwise
1212          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1213          * equal to target-freq.
1214          */
1215         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1216             && has_target()) {
1217                 /* Are we running at unknown frequency ? */
1218                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1219                 if (ret == -EINVAL) {
1220                         /* Warn user and fix it */
1221                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1222                                 __func__, policy->cpu, policy->cur);
1223                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1224                                 CPUFREQ_RELATION_L);
1225
1226                         /*
1227                          * Reaching here after boot in a few seconds may not
1228                          * mean that system will remain stable at "unknown"
1229                          * frequency for longer duration. Hence, a BUG_ON().
1230                          */
1231                         BUG_ON(ret);
1232                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1233                                 __func__, policy->cpu, policy->cur);
1234                 }
1235         }
1236
1237         if (new_policy) {
1238                 ret = cpufreq_add_dev_interface(policy);
1239                 if (ret)
1240                         goto out_exit_policy;
1241
1242                 cpufreq_stats_create_table(policy);
1243
1244                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1245                 list_add(&policy->policy_list, &cpufreq_policy_list);
1246                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1247         }
1248
1249         ret = cpufreq_init_policy(policy);
1250         if (ret) {
1251                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1252                        __func__, cpu, ret);
1253                 /* cpufreq_policy_free() will notify based on this */
1254                 new_policy = false;
1255                 goto out_exit_policy;
1256         }
1257
1258         up_write(&policy->rwsem);
1259
1260         kobject_uevent(&policy->kobj, KOBJ_ADD);
1261
1262         /* Callback for handling stuff after policy is ready */
1263         if (cpufreq_driver->ready)
1264                 cpufreq_driver->ready(policy);
1265
1266         pr_debug("initialization complete\n");
1267
1268         return 0;
1269
1270 out_exit_policy:
1271         up_write(&policy->rwsem);
1272
1273         if (cpufreq_driver->exit)
1274                 cpufreq_driver->exit(policy);
1275 out_free_policy:
1276         cpufreq_policy_free(policy);
1277         return ret;
1278 }
1279
1280 static int cpufreq_offline(unsigned int cpu);
1281
1282 /**
1283  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1284  * @dev: CPU device.
1285  * @sif: Subsystem interface structure pointer (not used)
1286  */
1287 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1288 {
1289         struct cpufreq_policy *policy;
1290         unsigned cpu = dev->id;
1291         int ret;
1292
1293         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1294
1295         if (cpu_online(cpu)) {
1296                 ret = cpufreq_online(cpu);
1297                 if (ret)
1298                         return ret;
1299         }
1300
1301         /* Create sysfs link on CPU registration */
1302         policy = per_cpu(cpufreq_cpu_data, cpu);
1303         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1304                 return 0;
1305
1306         ret = add_cpu_dev_symlink(policy, dev);
1307         if (ret) {
1308                 cpumask_clear_cpu(cpu, policy->real_cpus);
1309                 cpufreq_offline(cpu);
1310         }
1311
1312         return ret;
1313 }
1314
1315 static int cpufreq_offline(unsigned int cpu)
1316 {
1317         struct cpufreq_policy *policy;
1318         int ret;
1319
1320         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1321
1322         policy = cpufreq_cpu_get_raw(cpu);
1323         if (!policy) {
1324                 pr_debug("%s: No cpu_data found\n", __func__);
1325                 return 0;
1326         }
1327
1328         down_write(&policy->rwsem);
1329         if (has_target())
1330                 cpufreq_stop_governor(policy);
1331
1332         cpumask_clear_cpu(cpu, policy->cpus);
1333
1334         if (policy_is_inactive(policy)) {
1335                 if (has_target())
1336                         strncpy(policy->last_governor, policy->governor->name,
1337                                 CPUFREQ_NAME_LEN);
1338                 else
1339                         policy->last_policy = policy->policy;
1340         } else if (cpu == policy->cpu) {
1341                 /* Nominate new CPU */
1342                 policy->cpu = cpumask_any(policy->cpus);
1343         }
1344
1345         /* Start governor again for active policy */
1346         if (!policy_is_inactive(policy)) {
1347                 if (has_target()) {
1348                         ret = cpufreq_start_governor(policy);
1349                         if (ret)
1350                                 pr_err("%s: Failed to start governor\n", __func__);
1351                 }
1352
1353                 goto unlock;
1354         }
1355
1356         if (cpufreq_driver->stop_cpu)
1357                 cpufreq_driver->stop_cpu(policy);
1358
1359         if (has_target())
1360                 cpufreq_exit_governor(policy);
1361
1362         /*
1363          * Perform the ->exit() even during light-weight tear-down,
1364          * since this is a core component, and is essential for the
1365          * subsequent light-weight ->init() to succeed.
1366          */
1367         if (cpufreq_driver->exit) {
1368                 cpufreq_driver->exit(policy);
1369                 policy->freq_table = NULL;
1370         }
1371
1372 unlock:
1373         up_write(&policy->rwsem);
1374         return 0;
1375 }
1376
1377 /**
1378  * cpufreq_remove_dev - remove a CPU device
1379  *
1380  * Removes the cpufreq interface for a CPU device.
1381  */
1382 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1383 {
1384         unsigned int cpu = dev->id;
1385         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1386
1387         if (!policy)
1388                 return;
1389
1390         if (cpu_online(cpu))
1391                 cpufreq_offline(cpu);
1392
1393         cpumask_clear_cpu(cpu, policy->real_cpus);
1394         remove_cpu_dev_symlink(policy, dev);
1395
1396         if (cpumask_empty(policy->real_cpus))
1397                 cpufreq_policy_free(policy);
1398 }
1399
1400 /**
1401  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1402  *      in deep trouble.
1403  *      @policy: policy managing CPUs
1404  *      @new_freq: CPU frequency the CPU actually runs at
1405  *
1406  *      We adjust to current frequency first, and need to clean up later.
1407  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1408  */
1409 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1410                                 unsigned int new_freq)
1411 {
1412         struct cpufreq_freqs freqs;
1413
1414         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1415                  policy->cur, new_freq);
1416
1417         freqs.old = policy->cur;
1418         freqs.new = new_freq;
1419
1420         cpufreq_freq_transition_begin(policy, &freqs);
1421         cpufreq_freq_transition_end(policy, &freqs, 0);
1422 }
1423
1424 /**
1425  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1426  * @cpu: CPU number
1427  *
1428  * This is the last known freq, without actually getting it from the driver.
1429  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1430  */
1431 unsigned int cpufreq_quick_get(unsigned int cpu)
1432 {
1433         struct cpufreq_policy *policy;
1434         unsigned int ret_freq = 0;
1435         unsigned long flags;
1436
1437         read_lock_irqsave(&cpufreq_driver_lock, flags);
1438
1439         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1440                 ret_freq = cpufreq_driver->get(cpu);
1441                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1442                 return ret_freq;
1443         }
1444
1445         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1446
1447         policy = cpufreq_cpu_get(cpu);
1448         if (policy) {
1449                 ret_freq = policy->cur;
1450                 cpufreq_cpu_put(policy);
1451         }
1452
1453         return ret_freq;
1454 }
1455 EXPORT_SYMBOL(cpufreq_quick_get);
1456
1457 /**
1458  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1459  * @cpu: CPU number
1460  *
1461  * Just return the max possible frequency for a given CPU.
1462  */
1463 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1464 {
1465         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1466         unsigned int ret_freq = 0;
1467
1468         if (policy) {
1469                 ret_freq = policy->max;
1470                 cpufreq_cpu_put(policy);
1471         }
1472
1473         return ret_freq;
1474 }
1475 EXPORT_SYMBOL(cpufreq_quick_get_max);
1476
1477 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1478 {
1479         unsigned int ret_freq = 0;
1480
1481         if (!cpufreq_driver->get)
1482                 return ret_freq;
1483
1484         ret_freq = cpufreq_driver->get(policy->cpu);
1485
1486         /*
1487          * Updating inactive policies is invalid, so avoid doing that.  Also
1488          * if fast frequency switching is used with the given policy, the check
1489          * against policy->cur is pointless, so skip it in that case too.
1490          */
1491         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1492                 return ret_freq;
1493
1494         if (ret_freq && policy->cur &&
1495                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1496                 /* verify no discrepancy between actual and
1497                                         saved value exists */
1498                 if (unlikely(ret_freq != policy->cur)) {
1499                         cpufreq_out_of_sync(policy, ret_freq);
1500                         schedule_work(&policy->update);
1501                 }
1502         }
1503
1504         return ret_freq;
1505 }
1506
1507 /**
1508  * cpufreq_get - get the current CPU frequency (in kHz)
1509  * @cpu: CPU number
1510  *
1511  * Get the CPU current (static) CPU frequency
1512  */
1513 unsigned int cpufreq_get(unsigned int cpu)
1514 {
1515         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1516         unsigned int ret_freq = 0;
1517
1518         if (policy) {
1519                 down_read(&policy->rwsem);
1520
1521                 if (!policy_is_inactive(policy))
1522                         ret_freq = __cpufreq_get(policy);
1523
1524                 up_read(&policy->rwsem);
1525
1526                 cpufreq_cpu_put(policy);
1527         }
1528
1529         return ret_freq;
1530 }
1531 EXPORT_SYMBOL(cpufreq_get);
1532
1533 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1534 {
1535         unsigned int new_freq;
1536
1537         new_freq = cpufreq_driver->get(policy->cpu);
1538         if (!new_freq)
1539                 return 0;
1540
1541         if (!policy->cur) {
1542                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1543                 policy->cur = new_freq;
1544         } else if (policy->cur != new_freq && has_target()) {
1545                 cpufreq_out_of_sync(policy, new_freq);
1546         }
1547
1548         return new_freq;
1549 }
1550
1551 static struct subsys_interface cpufreq_interface = {
1552         .name           = "cpufreq",
1553         .subsys         = &cpu_subsys,
1554         .add_dev        = cpufreq_add_dev,
1555         .remove_dev     = cpufreq_remove_dev,
1556 };
1557
1558 /*
1559  * In case platform wants some specific frequency to be configured
1560  * during suspend..
1561  */
1562 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1563 {
1564         int ret;
1565
1566         if (!policy->suspend_freq) {
1567                 pr_debug("%s: suspend_freq not defined\n", __func__);
1568                 return 0;
1569         }
1570
1571         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1572                         policy->suspend_freq);
1573
1574         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1575                         CPUFREQ_RELATION_H);
1576         if (ret)
1577                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1578                                 __func__, policy->suspend_freq, ret);
1579
1580         return ret;
1581 }
1582 EXPORT_SYMBOL(cpufreq_generic_suspend);
1583
1584 /**
1585  * cpufreq_suspend() - Suspend CPUFreq governors
1586  *
1587  * Called during system wide Suspend/Hibernate cycles for suspending governors
1588  * as some platforms can't change frequency after this point in suspend cycle.
1589  * Because some of the devices (like: i2c, regulators, etc) they use for
1590  * changing frequency are suspended quickly after this point.
1591  */
1592 void cpufreq_suspend(void)
1593 {
1594         struct cpufreq_policy *policy;
1595
1596         if (!cpufreq_driver)
1597                 return;
1598
1599         if (!has_target() && !cpufreq_driver->suspend)
1600                 goto suspend;
1601
1602         pr_debug("%s: Suspending Governors\n", __func__);
1603
1604         for_each_active_policy(policy) {
1605                 if (has_target()) {
1606                         down_write(&policy->rwsem);
1607                         cpufreq_stop_governor(policy);
1608                         up_write(&policy->rwsem);
1609                 }
1610
1611                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1612                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1613                                 policy);
1614         }
1615
1616 suspend:
1617         cpufreq_suspended = true;
1618 }
1619
1620 /**
1621  * cpufreq_resume() - Resume CPUFreq governors
1622  *
1623  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1624  * are suspended with cpufreq_suspend().
1625  */
1626 void cpufreq_resume(void)
1627 {
1628         struct cpufreq_policy *policy;
1629         int ret;
1630
1631         if (!cpufreq_driver)
1632                 return;
1633
1634         cpufreq_suspended = false;
1635
1636         if (!has_target() && !cpufreq_driver->resume)
1637                 return;
1638
1639         pr_debug("%s: Resuming Governors\n", __func__);
1640
1641         for_each_active_policy(policy) {
1642                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1643                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1644                                 policy);
1645                 } else if (has_target()) {
1646                         down_write(&policy->rwsem);
1647                         ret = cpufreq_start_governor(policy);
1648                         up_write(&policy->rwsem);
1649
1650                         if (ret)
1651                                 pr_err("%s: Failed to start governor for policy: %p\n",
1652                                        __func__, policy);
1653                 }
1654         }
1655 }
1656
1657 /**
1658  *      cpufreq_get_current_driver - return current driver's name
1659  *
1660  *      Return the name string of the currently loaded cpufreq driver
1661  *      or NULL, if none.
1662  */
1663 const char *cpufreq_get_current_driver(void)
1664 {
1665         if (cpufreq_driver)
1666                 return cpufreq_driver->name;
1667
1668         return NULL;
1669 }
1670 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1671
1672 /**
1673  *      cpufreq_get_driver_data - return current driver data
1674  *
1675  *      Return the private data of the currently loaded cpufreq
1676  *      driver, or NULL if no cpufreq driver is loaded.
1677  */
1678 void *cpufreq_get_driver_data(void)
1679 {
1680         if (cpufreq_driver)
1681                 return cpufreq_driver->driver_data;
1682
1683         return NULL;
1684 }
1685 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1686
1687 /*********************************************************************
1688  *                     NOTIFIER LISTS INTERFACE                      *
1689  *********************************************************************/
1690
1691 /**
1692  *      cpufreq_register_notifier - register a driver with cpufreq
1693  *      @nb: notifier function to register
1694  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1695  *
1696  *      Add a driver to one of two lists: either a list of drivers that
1697  *      are notified about clock rate changes (once before and once after
1698  *      the transition), or a list of drivers that are notified about
1699  *      changes in cpufreq policy.
1700  *
1701  *      This function may sleep, and has the same return conditions as
1702  *      blocking_notifier_chain_register.
1703  */
1704 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1705 {
1706         int ret;
1707
1708         if (cpufreq_disabled())
1709                 return -EINVAL;
1710
1711         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1712
1713         switch (list) {
1714         case CPUFREQ_TRANSITION_NOTIFIER:
1715                 mutex_lock(&cpufreq_fast_switch_lock);
1716
1717                 if (cpufreq_fast_switch_count > 0) {
1718                         mutex_unlock(&cpufreq_fast_switch_lock);
1719                         return -EBUSY;
1720                 }
1721                 ret = srcu_notifier_chain_register(
1722                                 &cpufreq_transition_notifier_list, nb);
1723                 if (!ret)
1724                         cpufreq_fast_switch_count--;
1725
1726                 mutex_unlock(&cpufreq_fast_switch_lock);
1727                 break;
1728         case CPUFREQ_POLICY_NOTIFIER:
1729                 ret = blocking_notifier_chain_register(
1730                                 &cpufreq_policy_notifier_list, nb);
1731                 break;
1732         default:
1733                 ret = -EINVAL;
1734         }
1735
1736         return ret;
1737 }
1738 EXPORT_SYMBOL(cpufreq_register_notifier);
1739
1740 /**
1741  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1742  *      @nb: notifier block to be unregistered
1743  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1744  *
1745  *      Remove a driver from the CPU frequency notifier list.
1746  *
1747  *      This function may sleep, and has the same return conditions as
1748  *      blocking_notifier_chain_unregister.
1749  */
1750 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1751 {
1752         int ret;
1753
1754         if (cpufreq_disabled())
1755                 return -EINVAL;
1756
1757         switch (list) {
1758         case CPUFREQ_TRANSITION_NOTIFIER:
1759                 mutex_lock(&cpufreq_fast_switch_lock);
1760
1761                 ret = srcu_notifier_chain_unregister(
1762                                 &cpufreq_transition_notifier_list, nb);
1763                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1764                         cpufreq_fast_switch_count++;
1765
1766                 mutex_unlock(&cpufreq_fast_switch_lock);
1767                 break;
1768         case CPUFREQ_POLICY_NOTIFIER:
1769                 ret = blocking_notifier_chain_unregister(
1770                                 &cpufreq_policy_notifier_list, nb);
1771                 break;
1772         default:
1773                 ret = -EINVAL;
1774         }
1775
1776         return ret;
1777 }
1778 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1779
1780
1781 /*********************************************************************
1782  *                              GOVERNORS                            *
1783  *********************************************************************/
1784
1785 /**
1786  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1787  * @policy: cpufreq policy to switch the frequency for.
1788  * @target_freq: New frequency to set (may be approximate).
1789  *
1790  * Carry out a fast frequency switch without sleeping.
1791  *
1792  * The driver's ->fast_switch() callback invoked by this function must be
1793  * suitable for being called from within RCU-sched read-side critical sections
1794  * and it is expected to select the minimum available frequency greater than or
1795  * equal to @target_freq (CPUFREQ_RELATION_L).
1796  *
1797  * This function must not be called if policy->fast_switch_enabled is unset.
1798  *
1799  * Governors calling this function must guarantee that it will never be invoked
1800  * twice in parallel for the same policy and that it will never be called in
1801  * parallel with either ->target() or ->target_index() for the same policy.
1802  *
1803  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1804  * callback to indicate an error condition, the hardware configuration must be
1805  * preserved.
1806  */
1807 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1808                                         unsigned int target_freq)
1809 {
1810         target_freq = clamp_val(target_freq, policy->min, policy->max);
1811
1812         return cpufreq_driver->fast_switch(policy, target_freq);
1813 }
1814 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1815
1816 /* Must set freqs->new to intermediate frequency */
1817 static int __target_intermediate(struct cpufreq_policy *policy,
1818                                  struct cpufreq_freqs *freqs, int index)
1819 {
1820         int ret;
1821
1822         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1823
1824         /* We don't need to switch to intermediate freq */
1825         if (!freqs->new)
1826                 return 0;
1827
1828         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1829                  __func__, policy->cpu, freqs->old, freqs->new);
1830
1831         cpufreq_freq_transition_begin(policy, freqs);
1832         ret = cpufreq_driver->target_intermediate(policy, index);
1833         cpufreq_freq_transition_end(policy, freqs, ret);
1834
1835         if (ret)
1836                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1837                        __func__, ret);
1838
1839         return ret;
1840 }
1841
1842 static int __target_index(struct cpufreq_policy *policy, int index)
1843 {
1844         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1845         unsigned int intermediate_freq = 0;
1846         unsigned int newfreq = policy->freq_table[index].frequency;
1847         int retval = -EINVAL;
1848         bool notify;
1849
1850         if (newfreq == policy->cur)
1851                 return 0;
1852
1853         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1854         if (notify) {
1855                 /* Handle switching to intermediate frequency */
1856                 if (cpufreq_driver->get_intermediate) {
1857                         retval = __target_intermediate(policy, &freqs, index);
1858                         if (retval)
1859                                 return retval;
1860
1861                         intermediate_freq = freqs.new;
1862                         /* Set old freq to intermediate */
1863                         if (intermediate_freq)
1864                                 freqs.old = freqs.new;
1865                 }
1866
1867                 freqs.new = newfreq;
1868                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1869                          __func__, policy->cpu, freqs.old, freqs.new);
1870
1871                 cpufreq_freq_transition_begin(policy, &freqs);
1872         }
1873
1874         retval = cpufreq_driver->target_index(policy, index);
1875         if (retval)
1876                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1877                        retval);
1878
1879         if (notify) {
1880                 cpufreq_freq_transition_end(policy, &freqs, retval);
1881
1882                 /*
1883                  * Failed after setting to intermediate freq? Driver should have
1884                  * reverted back to initial frequency and so should we. Check
1885                  * here for intermediate_freq instead of get_intermediate, in
1886                  * case we haven't switched to intermediate freq at all.
1887                  */
1888                 if (unlikely(retval && intermediate_freq)) {
1889                         freqs.old = intermediate_freq;
1890                         freqs.new = policy->restore_freq;
1891                         cpufreq_freq_transition_begin(policy, &freqs);
1892                         cpufreq_freq_transition_end(policy, &freqs, 0);
1893                 }
1894         }
1895
1896         return retval;
1897 }
1898
1899 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1900                             unsigned int target_freq,
1901                             unsigned int relation)
1902 {
1903         unsigned int old_target_freq = target_freq;
1904         int index;
1905
1906         if (cpufreq_disabled())
1907                 return -ENODEV;
1908
1909         /* Make sure that target_freq is within supported range */
1910         target_freq = clamp_val(target_freq, policy->min, policy->max);
1911
1912         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1913                  policy->cpu, target_freq, relation, old_target_freq);
1914
1915         /*
1916          * This might look like a redundant call as we are checking it again
1917          * after finding index. But it is left intentionally for cases where
1918          * exactly same freq is called again and so we can save on few function
1919          * calls.
1920          */
1921         if (target_freq == policy->cur)
1922                 return 0;
1923
1924         /* Save last value to restore later on errors */
1925         policy->restore_freq = policy->cur;
1926
1927         if (cpufreq_driver->target)
1928                 return cpufreq_driver->target(policy, target_freq, relation);
1929
1930         if (!cpufreq_driver->target_index)
1931                 return -EINVAL;
1932
1933         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1934
1935         return __target_index(policy, index);
1936 }
1937 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1938
1939 int cpufreq_driver_target(struct cpufreq_policy *policy,
1940                           unsigned int target_freq,
1941                           unsigned int relation)
1942 {
1943         int ret = -EINVAL;
1944
1945         down_write(&policy->rwsem);
1946
1947         ret = __cpufreq_driver_target(policy, target_freq, relation);
1948
1949         up_write(&policy->rwsem);
1950
1951         return ret;
1952 }
1953 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1954
1955 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1956 {
1957         return NULL;
1958 }
1959
1960 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1961 {
1962         int ret;
1963
1964         /* Don't start any governor operations if we are entering suspend */
1965         if (cpufreq_suspended)
1966                 return 0;
1967         /*
1968          * Governor might not be initiated here if ACPI _PPC changed
1969          * notification happened, so check it.
1970          */
1971         if (!policy->governor)
1972                 return -EINVAL;
1973
1974         if (policy->governor->max_transition_latency &&
1975             policy->cpuinfo.transition_latency >
1976             policy->governor->max_transition_latency) {
1977                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
1978
1979                 if (gov) {
1980                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1981                                 policy->governor->name, gov->name);
1982                         policy->governor = gov;
1983                 } else {
1984                         return -EINVAL;
1985                 }
1986         }
1987
1988         if (!try_module_get(policy->governor->owner))
1989                 return -EINVAL;
1990
1991         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
1992
1993         if (policy->governor->init) {
1994                 ret = policy->governor->init(policy);
1995                 if (ret) {
1996                         module_put(policy->governor->owner);
1997                         return ret;
1998                 }
1999         }
2000
2001         return 0;
2002 }
2003
2004 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2005 {
2006         if (cpufreq_suspended || !policy->governor)
2007                 return;
2008
2009         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2010
2011         if (policy->governor->exit)
2012                 policy->governor->exit(policy);
2013
2014         module_put(policy->governor->owner);
2015 }
2016
2017 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2018 {
2019         int ret;
2020
2021         if (cpufreq_suspended)
2022                 return 0;
2023
2024         if (!policy->governor)
2025                 return -EINVAL;
2026
2027         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2028
2029         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2030                 cpufreq_update_current_freq(policy);
2031
2032         if (policy->governor->start) {
2033                 ret = policy->governor->start(policy);
2034                 if (ret)
2035                         return ret;
2036         }
2037
2038         if (policy->governor->limits)
2039                 policy->governor->limits(policy);
2040
2041         return 0;
2042 }
2043
2044 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2045 {
2046         if (cpufreq_suspended || !policy->governor)
2047                 return;
2048
2049         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2050
2051         if (policy->governor->stop)
2052                 policy->governor->stop(policy);
2053 }
2054
2055 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2056 {
2057         if (cpufreq_suspended || !policy->governor)
2058                 return;
2059
2060         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2061
2062         if (policy->governor->limits)
2063                 policy->governor->limits(policy);
2064 }
2065
2066 int cpufreq_register_governor(struct cpufreq_governor *governor)
2067 {
2068         int err;
2069
2070         if (!governor)
2071                 return -EINVAL;
2072
2073         if (cpufreq_disabled())
2074                 return -ENODEV;
2075
2076         mutex_lock(&cpufreq_governor_mutex);
2077
2078         err = -EBUSY;
2079         if (!find_governor(governor->name)) {
2080                 err = 0;
2081                 list_add(&governor->governor_list, &cpufreq_governor_list);
2082         }
2083
2084         mutex_unlock(&cpufreq_governor_mutex);
2085         return err;
2086 }
2087 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2088
2089 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2090 {
2091         struct cpufreq_policy *policy;
2092         unsigned long flags;
2093
2094         if (!governor)
2095                 return;
2096
2097         if (cpufreq_disabled())
2098                 return;
2099
2100         /* clear last_governor for all inactive policies */
2101         read_lock_irqsave(&cpufreq_driver_lock, flags);
2102         for_each_inactive_policy(policy) {
2103                 if (!strcmp(policy->last_governor, governor->name)) {
2104                         policy->governor = NULL;
2105                         strcpy(policy->last_governor, "\0");
2106                 }
2107         }
2108         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2109
2110         mutex_lock(&cpufreq_governor_mutex);
2111         list_del(&governor->governor_list);
2112         mutex_unlock(&cpufreq_governor_mutex);
2113         return;
2114 }
2115 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2116
2117
2118 /*********************************************************************
2119  *                          POLICY INTERFACE                         *
2120  *********************************************************************/
2121
2122 /**
2123  * cpufreq_get_policy - get the current cpufreq_policy
2124  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2125  *      is written
2126  *
2127  * Reads the current cpufreq policy.
2128  */
2129 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2130 {
2131         struct cpufreq_policy *cpu_policy;
2132         if (!policy)
2133                 return -EINVAL;
2134
2135         cpu_policy = cpufreq_cpu_get(cpu);
2136         if (!cpu_policy)
2137                 return -EINVAL;
2138
2139         memcpy(policy, cpu_policy, sizeof(*policy));
2140
2141         cpufreq_cpu_put(cpu_policy);
2142         return 0;
2143 }
2144 EXPORT_SYMBOL(cpufreq_get_policy);
2145
2146 /*
2147  * policy : current policy.
2148  * new_policy: policy to be set.
2149  */
2150 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2151                                 struct cpufreq_policy *new_policy)
2152 {
2153         struct cpufreq_governor *old_gov;
2154         int ret;
2155
2156         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2157                  new_policy->cpu, new_policy->min, new_policy->max);
2158
2159         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2160
2161         /*
2162         * This check works well when we store new min/max freq attributes,
2163         * because new_policy is a copy of policy with one field updated.
2164         */
2165         if (new_policy->min > new_policy->max)
2166                 return -EINVAL;
2167
2168         /* verify the cpu speed can be set within this limit */
2169         ret = cpufreq_driver->verify(new_policy);
2170         if (ret)
2171                 return ret;
2172
2173         /* adjust if necessary - all reasons */
2174         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2175                         CPUFREQ_ADJUST, new_policy);
2176
2177         /*
2178          * verify the cpu speed can be set within this limit, which might be
2179          * different to the first one
2180          */
2181         ret = cpufreq_driver->verify(new_policy);
2182         if (ret)
2183                 return ret;
2184
2185         /* notification of the new policy */
2186         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2187                         CPUFREQ_NOTIFY, new_policy);
2188
2189         policy->min = new_policy->min;
2190         policy->max = new_policy->max;
2191
2192         policy->cached_target_freq = UINT_MAX;
2193
2194         pr_debug("new min and max freqs are %u - %u kHz\n",
2195                  policy->min, policy->max);
2196
2197         if (cpufreq_driver->setpolicy) {
2198                 policy->policy = new_policy->policy;
2199                 pr_debug("setting range\n");
2200                 return cpufreq_driver->setpolicy(new_policy);
2201         }
2202
2203         if (new_policy->governor == policy->governor) {
2204                 pr_debug("cpufreq: governor limits update\n");
2205                 cpufreq_governor_limits(policy);
2206                 return 0;
2207         }
2208
2209         pr_debug("governor switch\n");
2210
2211         /* save old, working values */
2212         old_gov = policy->governor;
2213         /* end old governor */
2214         if (old_gov) {
2215                 cpufreq_stop_governor(policy);
2216                 cpufreq_exit_governor(policy);
2217         }
2218
2219         /* start new governor */
2220         policy->governor = new_policy->governor;
2221         ret = cpufreq_init_governor(policy);
2222         if (!ret) {
2223                 ret = cpufreq_start_governor(policy);
2224                 if (!ret) {
2225                         pr_debug("cpufreq: governor change\n");
2226                         return 0;
2227                 }
2228                 cpufreq_exit_governor(policy);
2229         }
2230
2231         /* new governor failed, so re-start old one */
2232         pr_debug("starting governor %s failed\n", policy->governor->name);
2233         if (old_gov) {
2234                 policy->governor = old_gov;
2235                 if (cpufreq_init_governor(policy))
2236                         policy->governor = NULL;
2237                 else
2238                         cpufreq_start_governor(policy);
2239         }
2240
2241         return ret;
2242 }
2243
2244 /**
2245  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2246  *      @cpu: CPU which shall be re-evaluated
2247  *
2248  *      Useful for policy notifiers which have different necessities
2249  *      at different times.
2250  */
2251 void cpufreq_update_policy(unsigned int cpu)
2252 {
2253         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2254         struct cpufreq_policy new_policy;
2255
2256         if (!policy)
2257                 return;
2258
2259         down_write(&policy->rwsem);
2260
2261         if (policy_is_inactive(policy))
2262                 goto unlock;
2263
2264         pr_debug("updating policy for CPU %u\n", cpu);
2265         memcpy(&new_policy, policy, sizeof(*policy));
2266         new_policy.min = policy->user_policy.min;
2267         new_policy.max = policy->user_policy.max;
2268
2269         /*
2270          * BIOS might change freq behind our back
2271          * -> ask driver for current freq and notify governors about a change
2272          */
2273         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2274                 if (cpufreq_suspended)
2275                         goto unlock;
2276
2277                 new_policy.cur = cpufreq_update_current_freq(policy);
2278                 if (WARN_ON(!new_policy.cur))
2279                         goto unlock;
2280         }
2281
2282         cpufreq_set_policy(policy, &new_policy);
2283
2284 unlock:
2285         up_write(&policy->rwsem);
2286
2287         cpufreq_cpu_put(policy);
2288 }
2289 EXPORT_SYMBOL(cpufreq_update_policy);
2290
2291 /*********************************************************************
2292  *               BOOST                                               *
2293  *********************************************************************/
2294 static int cpufreq_boost_set_sw(int state)
2295 {
2296         struct cpufreq_policy *policy;
2297         int ret = -EINVAL;
2298
2299         for_each_active_policy(policy) {
2300                 if (!policy->freq_table)
2301                         continue;
2302
2303                 ret = cpufreq_frequency_table_cpuinfo(policy,
2304                                                       policy->freq_table);
2305                 if (ret) {
2306                         pr_err("%s: Policy frequency update failed\n",
2307                                __func__);
2308                         break;
2309                 }
2310
2311                 down_write(&policy->rwsem);
2312                 policy->user_policy.max = policy->max;
2313                 cpufreq_governor_limits(policy);
2314                 up_write(&policy->rwsem);
2315         }
2316
2317         return ret;
2318 }
2319
2320 int cpufreq_boost_trigger_state(int state)
2321 {
2322         unsigned long flags;
2323         int ret = 0;
2324
2325         if (cpufreq_driver->boost_enabled == state)
2326                 return 0;
2327
2328         write_lock_irqsave(&cpufreq_driver_lock, flags);
2329         cpufreq_driver->boost_enabled = state;
2330         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2331
2332         ret = cpufreq_driver->set_boost(state);
2333         if (ret) {
2334                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2335                 cpufreq_driver->boost_enabled = !state;
2336                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2337
2338                 pr_err("%s: Cannot %s BOOST\n",
2339                        __func__, state ? "enable" : "disable");
2340         }
2341
2342         return ret;
2343 }
2344
2345 static bool cpufreq_boost_supported(void)
2346 {
2347         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2348 }
2349
2350 static int create_boost_sysfs_file(void)
2351 {
2352         int ret;
2353
2354         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2355         if (ret)
2356                 pr_err("%s: cannot register global BOOST sysfs file\n",
2357                        __func__);
2358
2359         return ret;
2360 }
2361
2362 static void remove_boost_sysfs_file(void)
2363 {
2364         if (cpufreq_boost_supported())
2365                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2366 }
2367
2368 int cpufreq_enable_boost_support(void)
2369 {
2370         if (!cpufreq_driver)
2371                 return -EINVAL;
2372
2373         if (cpufreq_boost_supported())
2374                 return 0;
2375
2376         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2377
2378         /* This will get removed on driver unregister */
2379         return create_boost_sysfs_file();
2380 }
2381 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2382
2383 int cpufreq_boost_enabled(void)
2384 {
2385         return cpufreq_driver->boost_enabled;
2386 }
2387 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2388
2389 /*********************************************************************
2390  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2391  *********************************************************************/
2392 static enum cpuhp_state hp_online;
2393
2394 /**
2395  * cpufreq_register_driver - register a CPU Frequency driver
2396  * @driver_data: A struct cpufreq_driver containing the values#
2397  * submitted by the CPU Frequency driver.
2398  *
2399  * Registers a CPU Frequency driver to this core code. This code
2400  * returns zero on success, -EEXIST when another driver got here first
2401  * (and isn't unregistered in the meantime).
2402  *
2403  */
2404 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2405 {
2406         unsigned long flags;
2407         int ret;
2408
2409         if (cpufreq_disabled())
2410                 return -ENODEV;
2411
2412         if (!driver_data || !driver_data->verify || !driver_data->init ||
2413             !(driver_data->setpolicy || driver_data->target_index ||
2414                     driver_data->target) ||
2415              (driver_data->setpolicy && (driver_data->target_index ||
2416                     driver_data->target)) ||
2417              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2418                 return -EINVAL;
2419
2420         pr_debug("trying to register driver %s\n", driver_data->name);
2421
2422         /* Protect against concurrent CPU online/offline. */
2423         get_online_cpus();
2424
2425         write_lock_irqsave(&cpufreq_driver_lock, flags);
2426         if (cpufreq_driver) {
2427                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2428                 ret = -EEXIST;
2429                 goto out;
2430         }
2431         cpufreq_driver = driver_data;
2432         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2433
2434         if (driver_data->setpolicy)
2435                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2436
2437         if (cpufreq_boost_supported()) {
2438                 ret = create_boost_sysfs_file();
2439                 if (ret)
2440                         goto err_null_driver;
2441         }
2442
2443         ret = subsys_interface_register(&cpufreq_interface);
2444         if (ret)
2445                 goto err_boost_unreg;
2446
2447         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2448             list_empty(&cpufreq_policy_list)) {
2449                 /* if all ->init() calls failed, unregister */
2450                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2451                          driver_data->name);
2452                 goto err_if_unreg;
2453         }
2454
2455         ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "cpufreq:online",
2456                                         cpufreq_online,
2457                                         cpufreq_offline);
2458         if (ret < 0)
2459                 goto err_if_unreg;
2460         hp_online = ret;
2461         ret = 0;
2462
2463         pr_debug("driver %s up and running\n", driver_data->name);
2464         goto out;
2465
2466 err_if_unreg:
2467         subsys_interface_unregister(&cpufreq_interface);
2468 err_boost_unreg:
2469         remove_boost_sysfs_file();
2470 err_null_driver:
2471         write_lock_irqsave(&cpufreq_driver_lock, flags);
2472         cpufreq_driver = NULL;
2473         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2474 out:
2475         put_online_cpus();
2476         return ret;
2477 }
2478 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2479
2480 /**
2481  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2482  *
2483  * Unregister the current CPUFreq driver. Only call this if you have
2484  * the right to do so, i.e. if you have succeeded in initialising before!
2485  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2486  * currently not initialised.
2487  */
2488 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2489 {
2490         unsigned long flags;
2491
2492         if (!cpufreq_driver || (driver != cpufreq_driver))
2493                 return -EINVAL;
2494
2495         pr_debug("unregistering driver %s\n", driver->name);
2496
2497         /* Protect against concurrent cpu hotplug */
2498         get_online_cpus();
2499         subsys_interface_unregister(&cpufreq_interface);
2500         remove_boost_sysfs_file();
2501         cpuhp_remove_state_nocalls(hp_online);
2502
2503         write_lock_irqsave(&cpufreq_driver_lock, flags);
2504
2505         cpufreq_driver = NULL;
2506
2507         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2508         put_online_cpus();
2509
2510         return 0;
2511 }
2512 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2513
2514 /*
2515  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2516  * or mutexes when secondary CPUs are halted.
2517  */
2518 static struct syscore_ops cpufreq_syscore_ops = {
2519         .shutdown = cpufreq_suspend,
2520 };
2521
2522 struct kobject *cpufreq_global_kobject;
2523 EXPORT_SYMBOL(cpufreq_global_kobject);
2524
2525 static int __init cpufreq_core_init(void)
2526 {
2527         if (cpufreq_disabled())
2528                 return -ENODEV;
2529
2530         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2531         BUG_ON(!cpufreq_global_kobject);
2532
2533         register_syscore_ops(&cpufreq_syscore_ops);
2534
2535         return 0;
2536 }
2537 module_param(off, int, 0444);
2538 core_initcall(cpufreq_core_init);