Merge back earlier cpufreq material for v4.8.
[cascardo/linux.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_cputime64(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 = cputime_to_usecs(cur_wall_time);
147
148         return cputime_to_usecs(idle_time);
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  *                          SYSFS INTERFACE                          *
497  *********************************************************************/
498 static ssize_t show_boost(struct kobject *kobj,
499                                  struct attribute *attr, char *buf)
500 {
501         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
502 }
503
504 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
505                                   const char *buf, size_t count)
506 {
507         int ret, enable;
508
509         ret = sscanf(buf, "%d", &enable);
510         if (ret != 1 || enable < 0 || enable > 1)
511                 return -EINVAL;
512
513         if (cpufreq_boost_trigger_state(enable)) {
514                 pr_err("%s: Cannot %s BOOST!\n",
515                        __func__, enable ? "enable" : "disable");
516                 return -EINVAL;
517         }
518
519         pr_debug("%s: cpufreq BOOST %s\n",
520                  __func__, enable ? "enabled" : "disabled");
521
522         return count;
523 }
524 define_one_global_rw(boost);
525
526 static struct cpufreq_governor *find_governor(const char *str_governor)
527 {
528         struct cpufreq_governor *t;
529
530         for_each_governor(t)
531                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
532                         return t;
533
534         return NULL;
535 }
536
537 /**
538  * cpufreq_parse_governor - parse a governor string
539  */
540 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
541                                 struct cpufreq_governor **governor)
542 {
543         int err = -EINVAL;
544
545         if (cpufreq_driver->setpolicy) {
546                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
547                         *policy = CPUFREQ_POLICY_PERFORMANCE;
548                         err = 0;
549                 } else if (!strncasecmp(str_governor, "powersave",
550                                                 CPUFREQ_NAME_LEN)) {
551                         *policy = CPUFREQ_POLICY_POWERSAVE;
552                         err = 0;
553                 }
554         } else {
555                 struct cpufreq_governor *t;
556
557                 mutex_lock(&cpufreq_governor_mutex);
558
559                 t = find_governor(str_governor);
560
561                 if (t == NULL) {
562                         int ret;
563
564                         mutex_unlock(&cpufreq_governor_mutex);
565                         ret = request_module("cpufreq_%s", str_governor);
566                         mutex_lock(&cpufreq_governor_mutex);
567
568                         if (ret == 0)
569                                 t = find_governor(str_governor);
570                 }
571
572                 if (t != NULL) {
573                         *governor = t;
574                         err = 0;
575                 }
576
577                 mutex_unlock(&cpufreq_governor_mutex);
578         }
579         return err;
580 }
581
582 /**
583  * cpufreq_per_cpu_attr_read() / show_##file_name() -
584  * print out cpufreq information
585  *
586  * Write out information from cpufreq_driver->policy[cpu]; object must be
587  * "unsigned int".
588  */
589
590 #define show_one(file_name, object)                     \
591 static ssize_t show_##file_name                         \
592 (struct cpufreq_policy *policy, char *buf)              \
593 {                                                       \
594         return sprintf(buf, "%u\n", policy->object);    \
595 }
596
597 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
598 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
599 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
600 show_one(scaling_min_freq, min);
601 show_one(scaling_max_freq, max);
602
603 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
604 {
605         ssize_t ret;
606
607         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
608                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
609         else
610                 ret = sprintf(buf, "%u\n", policy->cur);
611         return ret;
612 }
613
614 static int cpufreq_set_policy(struct cpufreq_policy *policy,
615                                 struct cpufreq_policy *new_policy);
616
617 /**
618  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
619  */
620 #define store_one(file_name, object)                    \
621 static ssize_t store_##file_name                                        \
622 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
623 {                                                                       \
624         int ret, temp;                                                  \
625         struct cpufreq_policy new_policy;                               \
626                                                                         \
627         memcpy(&new_policy, policy, sizeof(*policy));                   \
628                                                                         \
629         ret = sscanf(buf, "%u", &new_policy.object);                    \
630         if (ret != 1)                                                   \
631                 return -EINVAL;                                         \
632                                                                         \
633         temp = new_policy.object;                                       \
634         ret = cpufreq_set_policy(policy, &new_policy);          \
635         if (!ret)                                                       \
636                 policy->user_policy.object = temp;                      \
637                                                                         \
638         return ret ? ret : count;                                       \
639 }
640
641 store_one(scaling_min_freq, min);
642 store_one(scaling_max_freq, max);
643
644 /**
645  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
646  */
647 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
648                                         char *buf)
649 {
650         unsigned int cur_freq = __cpufreq_get(policy);
651         if (!cur_freq)
652                 return sprintf(buf, "<unknown>");
653         return sprintf(buf, "%u\n", cur_freq);
654 }
655
656 /**
657  * show_scaling_governor - show the current policy for the specified CPU
658  */
659 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
660 {
661         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
662                 return sprintf(buf, "powersave\n");
663         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
664                 return sprintf(buf, "performance\n");
665         else if (policy->governor)
666                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
667                                 policy->governor->name);
668         return -EINVAL;
669 }
670
671 /**
672  * store_scaling_governor - store policy for the specified CPU
673  */
674 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
675                                         const char *buf, size_t count)
676 {
677         int ret;
678         char    str_governor[16];
679         struct cpufreq_policy new_policy;
680
681         memcpy(&new_policy, policy, sizeof(*policy));
682
683         ret = sscanf(buf, "%15s", str_governor);
684         if (ret != 1)
685                 return -EINVAL;
686
687         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
688                                                 &new_policy.governor))
689                 return -EINVAL;
690
691         ret = cpufreq_set_policy(policy, &new_policy);
692         return ret ? ret : count;
693 }
694
695 /**
696  * show_scaling_driver - show the cpufreq driver currently loaded
697  */
698 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
699 {
700         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
701 }
702
703 /**
704  * show_scaling_available_governors - show the available CPUfreq governors
705  */
706 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
707                                                 char *buf)
708 {
709         ssize_t i = 0;
710         struct cpufreq_governor *t;
711
712         if (!has_target()) {
713                 i += sprintf(buf, "performance powersave");
714                 goto out;
715         }
716
717         for_each_governor(t) {
718                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
719                     - (CPUFREQ_NAME_LEN + 2)))
720                         goto out;
721                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
722         }
723 out:
724         i += sprintf(&buf[i], "\n");
725         return i;
726 }
727
728 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
729 {
730         ssize_t i = 0;
731         unsigned int cpu;
732
733         for_each_cpu(cpu, mask) {
734                 if (i)
735                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
736                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
737                 if (i >= (PAGE_SIZE - 5))
738                         break;
739         }
740         i += sprintf(&buf[i], "\n");
741         return i;
742 }
743 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
744
745 /**
746  * show_related_cpus - show the CPUs affected by each transition even if
747  * hw coordination is in use
748  */
749 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
750 {
751         return cpufreq_show_cpus(policy->related_cpus, buf);
752 }
753
754 /**
755  * show_affected_cpus - show the CPUs affected by each transition
756  */
757 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
758 {
759         return cpufreq_show_cpus(policy->cpus, buf);
760 }
761
762 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
763                                         const char *buf, size_t count)
764 {
765         unsigned int freq = 0;
766         unsigned int ret;
767
768         if (!policy->governor || !policy->governor->store_setspeed)
769                 return -EINVAL;
770
771         ret = sscanf(buf, "%u", &freq);
772         if (ret != 1)
773                 return -EINVAL;
774
775         policy->governor->store_setspeed(policy, freq);
776
777         return count;
778 }
779
780 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
781 {
782         if (!policy->governor || !policy->governor->show_setspeed)
783                 return sprintf(buf, "<unsupported>\n");
784
785         return policy->governor->show_setspeed(policy, buf);
786 }
787
788 /**
789  * show_bios_limit - show the current cpufreq HW/BIOS limitation
790  */
791 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
792 {
793         unsigned int limit;
794         int ret;
795         if (cpufreq_driver->bios_limit) {
796                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
797                 if (!ret)
798                         return sprintf(buf, "%u\n", limit);
799         }
800         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
801 }
802
803 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
804 cpufreq_freq_attr_ro(cpuinfo_min_freq);
805 cpufreq_freq_attr_ro(cpuinfo_max_freq);
806 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
807 cpufreq_freq_attr_ro(scaling_available_governors);
808 cpufreq_freq_attr_ro(scaling_driver);
809 cpufreq_freq_attr_ro(scaling_cur_freq);
810 cpufreq_freq_attr_ro(bios_limit);
811 cpufreq_freq_attr_ro(related_cpus);
812 cpufreq_freq_attr_ro(affected_cpus);
813 cpufreq_freq_attr_rw(scaling_min_freq);
814 cpufreq_freq_attr_rw(scaling_max_freq);
815 cpufreq_freq_attr_rw(scaling_governor);
816 cpufreq_freq_attr_rw(scaling_setspeed);
817
818 static struct attribute *default_attrs[] = {
819         &cpuinfo_min_freq.attr,
820         &cpuinfo_max_freq.attr,
821         &cpuinfo_transition_latency.attr,
822         &scaling_min_freq.attr,
823         &scaling_max_freq.attr,
824         &affected_cpus.attr,
825         &related_cpus.attr,
826         &scaling_governor.attr,
827         &scaling_driver.attr,
828         &scaling_available_governors.attr,
829         &scaling_setspeed.attr,
830         NULL
831 };
832
833 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
834 #define to_attr(a) container_of(a, struct freq_attr, attr)
835
836 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
837 {
838         struct cpufreq_policy *policy = to_policy(kobj);
839         struct freq_attr *fattr = to_attr(attr);
840         ssize_t ret;
841
842         down_read(&policy->rwsem);
843         ret = fattr->show(policy, buf);
844         up_read(&policy->rwsem);
845
846         return ret;
847 }
848
849 static ssize_t store(struct kobject *kobj, struct attribute *attr,
850                      const char *buf, size_t count)
851 {
852         struct cpufreq_policy *policy = to_policy(kobj);
853         struct freq_attr *fattr = to_attr(attr);
854         ssize_t ret = -EINVAL;
855
856         get_online_cpus();
857
858         if (cpu_online(policy->cpu)) {
859                 down_write(&policy->rwsem);
860                 ret = fattr->store(policy, buf, count);
861                 up_write(&policy->rwsem);
862         }
863
864         put_online_cpus();
865
866         return ret;
867 }
868
869 static void cpufreq_sysfs_release(struct kobject *kobj)
870 {
871         struct cpufreq_policy *policy = to_policy(kobj);
872         pr_debug("last reference is dropped\n");
873         complete(&policy->kobj_unregister);
874 }
875
876 static const struct sysfs_ops sysfs_ops = {
877         .show   = show,
878         .store  = store,
879 };
880
881 static struct kobj_type ktype_cpufreq = {
882         .sysfs_ops      = &sysfs_ops,
883         .default_attrs  = default_attrs,
884         .release        = cpufreq_sysfs_release,
885 };
886
887 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
888 {
889         struct device *cpu_dev;
890
891         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
892
893         if (!policy)
894                 return 0;
895
896         cpu_dev = get_cpu_device(cpu);
897         if (WARN_ON(!cpu_dev))
898                 return 0;
899
900         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
901 }
902
903 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
904 {
905         struct device *cpu_dev;
906
907         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
908
909         cpu_dev = get_cpu_device(cpu);
910         if (WARN_ON(!cpu_dev))
911                 return;
912
913         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
914 }
915
916 /* Add/remove symlinks for all related CPUs */
917 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
918 {
919         unsigned int j;
920         int ret = 0;
921
922         /* Some related CPUs might not be present (physically hotplugged) */
923         for_each_cpu(j, policy->real_cpus) {
924                 ret = add_cpu_dev_symlink(policy, j);
925                 if (ret)
926                         break;
927         }
928
929         return ret;
930 }
931
932 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
933 {
934         unsigned int j;
935
936         /* Some related CPUs might not be present (physically hotplugged) */
937         for_each_cpu(j, policy->real_cpus)
938                 remove_cpu_dev_symlink(policy, j);
939 }
940
941 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
942 {
943         struct freq_attr **drv_attr;
944         int ret = 0;
945
946         /* set up files for this cpu device */
947         drv_attr = cpufreq_driver->attr;
948         while (drv_attr && *drv_attr) {
949                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
950                 if (ret)
951                         return ret;
952                 drv_attr++;
953         }
954         if (cpufreq_driver->get) {
955                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
956                 if (ret)
957                         return ret;
958         }
959
960         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
961         if (ret)
962                 return ret;
963
964         if (cpufreq_driver->bios_limit) {
965                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
966                 if (ret)
967                         return ret;
968         }
969
970         return cpufreq_add_dev_symlink(policy);
971 }
972
973 __weak struct cpufreq_governor *cpufreq_default_governor(void)
974 {
975         return NULL;
976 }
977
978 static int cpufreq_init_policy(struct cpufreq_policy *policy)
979 {
980         struct cpufreq_governor *gov = NULL;
981         struct cpufreq_policy new_policy;
982
983         memcpy(&new_policy, policy, sizeof(*policy));
984
985         /* Update governor of new_policy to the governor used before hotplug */
986         gov = find_governor(policy->last_governor);
987         if (gov) {
988                 pr_debug("Restoring governor %s for cpu %d\n",
989                                 policy->governor->name, policy->cpu);
990         } else {
991                 gov = cpufreq_default_governor();
992                 if (!gov)
993                         return -ENODATA;
994         }
995
996         new_policy.governor = gov;
997
998         /* Use the default policy if there is no last_policy. */
999         if (cpufreq_driver->setpolicy) {
1000                 if (policy->last_policy)
1001                         new_policy.policy = policy->last_policy;
1002                 else
1003                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1004                                                NULL);
1005         }
1006         /* set default policy */
1007         return cpufreq_set_policy(policy, &new_policy);
1008 }
1009
1010 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1011 {
1012         int ret = 0;
1013
1014         /* Has this CPU been taken care of already? */
1015         if (cpumask_test_cpu(cpu, policy->cpus))
1016                 return 0;
1017
1018         down_write(&policy->rwsem);
1019         if (has_target())
1020                 cpufreq_stop_governor(policy);
1021
1022         cpumask_set_cpu(cpu, policy->cpus);
1023
1024         if (has_target()) {
1025                 ret = cpufreq_start_governor(policy);
1026                 if (ret)
1027                         pr_err("%s: Failed to start governor\n", __func__);
1028         }
1029         up_write(&policy->rwsem);
1030         return ret;
1031 }
1032
1033 static void handle_update(struct work_struct *work)
1034 {
1035         struct cpufreq_policy *policy =
1036                 container_of(work, struct cpufreq_policy, update);
1037         unsigned int cpu = policy->cpu;
1038         pr_debug("handle_update for cpu %u called\n", cpu);
1039         cpufreq_update_policy(cpu);
1040 }
1041
1042 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1043 {
1044         struct device *dev = get_cpu_device(cpu);
1045         struct cpufreq_policy *policy;
1046         int ret;
1047
1048         if (WARN_ON(!dev))
1049                 return NULL;
1050
1051         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1052         if (!policy)
1053                 return NULL;
1054
1055         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1056                 goto err_free_policy;
1057
1058         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1059                 goto err_free_cpumask;
1060
1061         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1062                 goto err_free_rcpumask;
1063
1064         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1065                                    cpufreq_global_kobject, "policy%u", cpu);
1066         if (ret) {
1067                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1068                 goto err_free_real_cpus;
1069         }
1070
1071         INIT_LIST_HEAD(&policy->policy_list);
1072         init_rwsem(&policy->rwsem);
1073         spin_lock_init(&policy->transition_lock);
1074         init_waitqueue_head(&policy->transition_wait);
1075         init_completion(&policy->kobj_unregister);
1076         INIT_WORK(&policy->update, handle_update);
1077
1078         policy->cpu = cpu;
1079         return policy;
1080
1081 err_free_real_cpus:
1082         free_cpumask_var(policy->real_cpus);
1083 err_free_rcpumask:
1084         free_cpumask_var(policy->related_cpus);
1085 err_free_cpumask:
1086         free_cpumask_var(policy->cpus);
1087 err_free_policy:
1088         kfree(policy);
1089
1090         return NULL;
1091 }
1092
1093 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1094 {
1095         struct kobject *kobj;
1096         struct completion *cmp;
1097
1098         if (notify)
1099                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1100                                              CPUFREQ_REMOVE_POLICY, policy);
1101
1102         down_write(&policy->rwsem);
1103         cpufreq_stats_free_table(policy);
1104         cpufreq_remove_dev_symlink(policy);
1105         kobj = &policy->kobj;
1106         cmp = &policy->kobj_unregister;
1107         up_write(&policy->rwsem);
1108         kobject_put(kobj);
1109
1110         /*
1111          * We need to make sure that the underlying kobj is
1112          * actually not referenced anymore by anybody before we
1113          * proceed with unloading.
1114          */
1115         pr_debug("waiting for dropping of refcount\n");
1116         wait_for_completion(cmp);
1117         pr_debug("wait complete\n");
1118 }
1119
1120 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1121 {
1122         unsigned long flags;
1123         int cpu;
1124
1125         /* Remove policy from list */
1126         write_lock_irqsave(&cpufreq_driver_lock, flags);
1127         list_del(&policy->policy_list);
1128
1129         for_each_cpu(cpu, policy->related_cpus)
1130                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1131         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1132
1133         cpufreq_policy_put_kobj(policy, notify);
1134         free_cpumask_var(policy->real_cpus);
1135         free_cpumask_var(policy->related_cpus);
1136         free_cpumask_var(policy->cpus);
1137         kfree(policy);
1138 }
1139
1140 static int cpufreq_online(unsigned int cpu)
1141 {
1142         struct cpufreq_policy *policy;
1143         bool new_policy;
1144         unsigned long flags;
1145         unsigned int j;
1146         int ret;
1147
1148         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1149
1150         /* Check if this CPU already has a policy to manage it */
1151         policy = per_cpu(cpufreq_cpu_data, cpu);
1152         if (policy) {
1153                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1154                 if (!policy_is_inactive(policy))
1155                         return cpufreq_add_policy_cpu(policy, cpu);
1156
1157                 /* This is the only online CPU for the policy.  Start over. */
1158                 new_policy = false;
1159                 down_write(&policy->rwsem);
1160                 policy->cpu = cpu;
1161                 policy->governor = NULL;
1162                 up_write(&policy->rwsem);
1163         } else {
1164                 new_policy = true;
1165                 policy = cpufreq_policy_alloc(cpu);
1166                 if (!policy)
1167                         return -ENOMEM;
1168         }
1169
1170         cpumask_copy(policy->cpus, cpumask_of(cpu));
1171
1172         /* call driver. From then on the cpufreq must be able
1173          * to accept all calls to ->verify and ->setpolicy for this CPU
1174          */
1175         ret = cpufreq_driver->init(policy);
1176         if (ret) {
1177                 pr_debug("initialization failed\n");
1178                 goto out_free_policy;
1179         }
1180
1181         down_write(&policy->rwsem);
1182
1183         if (new_policy) {
1184                 /* related_cpus should at least include policy->cpus. */
1185                 cpumask_copy(policy->related_cpus, policy->cpus);
1186                 /* Remember CPUs present at the policy creation time. */
1187                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1188         }
1189
1190         /*
1191          * affected cpus must always be the one, which are online. We aren't
1192          * managing offline cpus here.
1193          */
1194         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1195
1196         if (new_policy) {
1197                 policy->user_policy.min = policy->min;
1198                 policy->user_policy.max = policy->max;
1199
1200                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1201                 for_each_cpu(j, policy->related_cpus)
1202                         per_cpu(cpufreq_cpu_data, j) = policy;
1203                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1204         }
1205
1206         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1207                 policy->cur = cpufreq_driver->get(policy->cpu);
1208                 if (!policy->cur) {
1209                         pr_err("%s: ->get() failed\n", __func__);
1210                         goto out_exit_policy;
1211                 }
1212         }
1213
1214         /*
1215          * Sometimes boot loaders set CPU frequency to a value outside of
1216          * frequency table present with cpufreq core. In such cases CPU might be
1217          * unstable if it has to run on that frequency for long duration of time
1218          * and so its better to set it to a frequency which is specified in
1219          * freq-table. This also makes cpufreq stats inconsistent as
1220          * cpufreq-stats would fail to register because current frequency of CPU
1221          * isn't found in freq-table.
1222          *
1223          * Because we don't want this change to effect boot process badly, we go
1224          * for the next freq which is >= policy->cur ('cur' must be set by now,
1225          * otherwise we will end up setting freq to lowest of the table as 'cur'
1226          * is initialized to zero).
1227          *
1228          * We are passing target-freq as "policy->cur - 1" otherwise
1229          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1230          * equal to target-freq.
1231          */
1232         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1233             && has_target()) {
1234                 /* Are we running at unknown frequency ? */
1235                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1236                 if (ret == -EINVAL) {
1237                         /* Warn user and fix it */
1238                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1239                                 __func__, policy->cpu, policy->cur);
1240                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1241                                 CPUFREQ_RELATION_L);
1242
1243                         /*
1244                          * Reaching here after boot in a few seconds may not
1245                          * mean that system will remain stable at "unknown"
1246                          * frequency for longer duration. Hence, a BUG_ON().
1247                          */
1248                         BUG_ON(ret);
1249                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1250                                 __func__, policy->cpu, policy->cur);
1251                 }
1252         }
1253
1254         if (new_policy) {
1255                 ret = cpufreq_add_dev_interface(policy);
1256                 if (ret)
1257                         goto out_exit_policy;
1258
1259                 cpufreq_stats_create_table(policy);
1260                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1261                                 CPUFREQ_CREATE_POLICY, policy);
1262
1263                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1264                 list_add(&policy->policy_list, &cpufreq_policy_list);
1265                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1266         }
1267
1268         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1269                                      CPUFREQ_START, policy);
1270
1271         ret = cpufreq_init_policy(policy);
1272         if (ret) {
1273                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1274                        __func__, cpu, ret);
1275                 /* cpufreq_policy_free() will notify based on this */
1276                 new_policy = false;
1277                 goto out_exit_policy;
1278         }
1279
1280         up_write(&policy->rwsem);
1281
1282         kobject_uevent(&policy->kobj, KOBJ_ADD);
1283
1284         /* Callback for handling stuff after policy is ready */
1285         if (cpufreq_driver->ready)
1286                 cpufreq_driver->ready(policy);
1287
1288         pr_debug("initialization complete\n");
1289
1290         return 0;
1291
1292 out_exit_policy:
1293         up_write(&policy->rwsem);
1294
1295         if (cpufreq_driver->exit)
1296                 cpufreq_driver->exit(policy);
1297 out_free_policy:
1298         cpufreq_policy_free(policy, !new_policy);
1299         return ret;
1300 }
1301
1302 /**
1303  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1304  * @dev: CPU device.
1305  * @sif: Subsystem interface structure pointer (not used)
1306  */
1307 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1308 {
1309         struct cpufreq_policy *policy;
1310         unsigned cpu = dev->id;
1311
1312         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1313
1314         if (cpu_online(cpu))
1315                 return cpufreq_online(cpu);
1316
1317         /*
1318          * A hotplug notifier will follow and we will handle it as CPU online
1319          * then.  For now, just create the sysfs link, unless there is no policy
1320          * or the link is already present.
1321          */
1322         policy = per_cpu(cpufreq_cpu_data, cpu);
1323         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1324                 return 0;
1325
1326         return add_cpu_dev_symlink(policy, cpu);
1327 }
1328
1329 static void cpufreq_offline(unsigned int cpu)
1330 {
1331         struct cpufreq_policy *policy;
1332         int ret;
1333
1334         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1335
1336         policy = cpufreq_cpu_get_raw(cpu);
1337         if (!policy) {
1338                 pr_debug("%s: No cpu_data found\n", __func__);
1339                 return;
1340         }
1341
1342         down_write(&policy->rwsem);
1343         if (has_target())
1344                 cpufreq_stop_governor(policy);
1345
1346         cpumask_clear_cpu(cpu, policy->cpus);
1347
1348         if (policy_is_inactive(policy)) {
1349                 if (has_target())
1350                         strncpy(policy->last_governor, policy->governor->name,
1351                                 CPUFREQ_NAME_LEN);
1352                 else
1353                         policy->last_policy = policy->policy;
1354         } else if (cpu == policy->cpu) {
1355                 /* Nominate new CPU */
1356                 policy->cpu = cpumask_any(policy->cpus);
1357         }
1358
1359         /* Start governor again for active policy */
1360         if (!policy_is_inactive(policy)) {
1361                 if (has_target()) {
1362                         ret = cpufreq_start_governor(policy);
1363                         if (ret)
1364                                 pr_err("%s: Failed to start governor\n", __func__);
1365                 }
1366
1367                 goto unlock;
1368         }
1369
1370         if (cpufreq_driver->stop_cpu)
1371                 cpufreq_driver->stop_cpu(policy);
1372
1373         if (has_target())
1374                 cpufreq_exit_governor(policy);
1375
1376         /*
1377          * Perform the ->exit() even during light-weight tear-down,
1378          * since this is a core component, and is essential for the
1379          * subsequent light-weight ->init() to succeed.
1380          */
1381         if (cpufreq_driver->exit) {
1382                 cpufreq_driver->exit(policy);
1383                 policy->freq_table = NULL;
1384         }
1385
1386 unlock:
1387         up_write(&policy->rwsem);
1388 }
1389
1390 /**
1391  * cpufreq_remove_dev - remove a CPU device
1392  *
1393  * Removes the cpufreq interface for a CPU device.
1394  */
1395 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1396 {
1397         unsigned int cpu = dev->id;
1398         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1399
1400         if (!policy)
1401                 return;
1402
1403         if (cpu_online(cpu))
1404                 cpufreq_offline(cpu);
1405
1406         cpumask_clear_cpu(cpu, policy->real_cpus);
1407         remove_cpu_dev_symlink(policy, cpu);
1408
1409         if (cpumask_empty(policy->real_cpus))
1410                 cpufreq_policy_free(policy, true);
1411 }
1412
1413 /**
1414  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1415  *      in deep trouble.
1416  *      @policy: policy managing CPUs
1417  *      @new_freq: CPU frequency the CPU actually runs at
1418  *
1419  *      We adjust to current frequency first, and need to clean up later.
1420  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1421  */
1422 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1423                                 unsigned int new_freq)
1424 {
1425         struct cpufreq_freqs freqs;
1426
1427         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1428                  policy->cur, new_freq);
1429
1430         freqs.old = policy->cur;
1431         freqs.new = new_freq;
1432
1433         cpufreq_freq_transition_begin(policy, &freqs);
1434         cpufreq_freq_transition_end(policy, &freqs, 0);
1435 }
1436
1437 /**
1438  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1439  * @cpu: CPU number
1440  *
1441  * This is the last known freq, without actually getting it from the driver.
1442  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1443  */
1444 unsigned int cpufreq_quick_get(unsigned int cpu)
1445 {
1446         struct cpufreq_policy *policy;
1447         unsigned int ret_freq = 0;
1448         unsigned long flags;
1449
1450         read_lock_irqsave(&cpufreq_driver_lock, flags);
1451
1452         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1453                 ret_freq = cpufreq_driver->get(cpu);
1454                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1455                 return ret_freq;
1456         }
1457
1458         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1459
1460         policy = cpufreq_cpu_get(cpu);
1461         if (policy) {
1462                 ret_freq = policy->cur;
1463                 cpufreq_cpu_put(policy);
1464         }
1465
1466         return ret_freq;
1467 }
1468 EXPORT_SYMBOL(cpufreq_quick_get);
1469
1470 /**
1471  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1472  * @cpu: CPU number
1473  *
1474  * Just return the max possible frequency for a given CPU.
1475  */
1476 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1477 {
1478         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1479         unsigned int ret_freq = 0;
1480
1481         if (policy) {
1482                 ret_freq = policy->max;
1483                 cpufreq_cpu_put(policy);
1484         }
1485
1486         return ret_freq;
1487 }
1488 EXPORT_SYMBOL(cpufreq_quick_get_max);
1489
1490 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1491 {
1492         unsigned int ret_freq = 0;
1493
1494         if (!cpufreq_driver->get)
1495                 return ret_freq;
1496
1497         ret_freq = cpufreq_driver->get(policy->cpu);
1498
1499         /*
1500          * Updating inactive policies is invalid, so avoid doing that.  Also
1501          * if fast frequency switching is used with the given policy, the check
1502          * against policy->cur is pointless, so skip it in that case too.
1503          */
1504         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1505                 return ret_freq;
1506
1507         if (ret_freq && policy->cur &&
1508                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1509                 /* verify no discrepancy between actual and
1510                                         saved value exists */
1511                 if (unlikely(ret_freq != policy->cur)) {
1512                         cpufreq_out_of_sync(policy, ret_freq);
1513                         schedule_work(&policy->update);
1514                 }
1515         }
1516
1517         return ret_freq;
1518 }
1519
1520 /**
1521  * cpufreq_get - get the current CPU frequency (in kHz)
1522  * @cpu: CPU number
1523  *
1524  * Get the CPU current (static) CPU frequency
1525  */
1526 unsigned int cpufreq_get(unsigned int cpu)
1527 {
1528         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1529         unsigned int ret_freq = 0;
1530
1531         if (policy) {
1532                 down_read(&policy->rwsem);
1533                 ret_freq = __cpufreq_get(policy);
1534                 up_read(&policy->rwsem);
1535
1536                 cpufreq_cpu_put(policy);
1537         }
1538
1539         return ret_freq;
1540 }
1541 EXPORT_SYMBOL(cpufreq_get);
1542
1543 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1544 {
1545         unsigned int new_freq;
1546
1547         if (cpufreq_suspended)
1548                 return 0;
1549
1550         new_freq = cpufreq_driver->get(policy->cpu);
1551         if (!new_freq)
1552                 return 0;
1553
1554         if (!policy->cur) {
1555                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1556                 policy->cur = new_freq;
1557         } else if (policy->cur != new_freq && has_target()) {
1558                 cpufreq_out_of_sync(policy, new_freq);
1559         }
1560
1561         return new_freq;
1562 }
1563
1564 static struct subsys_interface cpufreq_interface = {
1565         .name           = "cpufreq",
1566         .subsys         = &cpu_subsys,
1567         .add_dev        = cpufreq_add_dev,
1568         .remove_dev     = cpufreq_remove_dev,
1569 };
1570
1571 /*
1572  * In case platform wants some specific frequency to be configured
1573  * during suspend..
1574  */
1575 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1576 {
1577         int ret;
1578
1579         if (!policy->suspend_freq) {
1580                 pr_debug("%s: suspend_freq not defined\n", __func__);
1581                 return 0;
1582         }
1583
1584         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1585                         policy->suspend_freq);
1586
1587         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1588                         CPUFREQ_RELATION_H);
1589         if (ret)
1590                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1591                                 __func__, policy->suspend_freq, ret);
1592
1593         return ret;
1594 }
1595 EXPORT_SYMBOL(cpufreq_generic_suspend);
1596
1597 /**
1598  * cpufreq_suspend() - Suspend CPUFreq governors
1599  *
1600  * Called during system wide Suspend/Hibernate cycles for suspending governors
1601  * as some platforms can't change frequency after this point in suspend cycle.
1602  * Because some of the devices (like: i2c, regulators, etc) they use for
1603  * changing frequency are suspended quickly after this point.
1604  */
1605 void cpufreq_suspend(void)
1606 {
1607         struct cpufreq_policy *policy;
1608
1609         if (!cpufreq_driver)
1610                 return;
1611
1612         if (!has_target() && !cpufreq_driver->suspend)
1613                 goto suspend;
1614
1615         pr_debug("%s: Suspending Governors\n", __func__);
1616
1617         for_each_active_policy(policy) {
1618                 if (has_target()) {
1619                         down_write(&policy->rwsem);
1620                         cpufreq_stop_governor(policy);
1621                         up_write(&policy->rwsem);
1622                 }
1623
1624                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1625                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1626                                 policy);
1627         }
1628
1629 suspend:
1630         cpufreq_suspended = true;
1631 }
1632
1633 /**
1634  * cpufreq_resume() - Resume CPUFreq governors
1635  *
1636  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1637  * are suspended with cpufreq_suspend().
1638  */
1639 void cpufreq_resume(void)
1640 {
1641         struct cpufreq_policy *policy;
1642         int ret;
1643
1644         if (!cpufreq_driver)
1645                 return;
1646
1647         cpufreq_suspended = false;
1648
1649         if (!has_target() && !cpufreq_driver->resume)
1650                 return;
1651
1652         pr_debug("%s: Resuming Governors\n", __func__);
1653
1654         for_each_active_policy(policy) {
1655                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1656                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1657                                 policy);
1658                 } else if (has_target()) {
1659                         down_write(&policy->rwsem);
1660                         ret = cpufreq_start_governor(policy);
1661                         up_write(&policy->rwsem);
1662
1663                         if (ret)
1664                                 pr_err("%s: Failed to start governor for policy: %p\n",
1665                                        __func__, policy);
1666                 }
1667         }
1668 }
1669
1670 /**
1671  *      cpufreq_get_current_driver - return current driver's name
1672  *
1673  *      Return the name string of the currently loaded cpufreq driver
1674  *      or NULL, if none.
1675  */
1676 const char *cpufreq_get_current_driver(void)
1677 {
1678         if (cpufreq_driver)
1679                 return cpufreq_driver->name;
1680
1681         return NULL;
1682 }
1683 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1684
1685 /**
1686  *      cpufreq_get_driver_data - return current driver data
1687  *
1688  *      Return the private data of the currently loaded cpufreq
1689  *      driver, or NULL if no cpufreq driver is loaded.
1690  */
1691 void *cpufreq_get_driver_data(void)
1692 {
1693         if (cpufreq_driver)
1694                 return cpufreq_driver->driver_data;
1695
1696         return NULL;
1697 }
1698 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1699
1700 /*********************************************************************
1701  *                     NOTIFIER LISTS INTERFACE                      *
1702  *********************************************************************/
1703
1704 /**
1705  *      cpufreq_register_notifier - register a driver with cpufreq
1706  *      @nb: notifier function to register
1707  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1708  *
1709  *      Add a driver to one of two lists: either a list of drivers that
1710  *      are notified about clock rate changes (once before and once after
1711  *      the transition), or a list of drivers that are notified about
1712  *      changes in cpufreq policy.
1713  *
1714  *      This function may sleep, and has the same return conditions as
1715  *      blocking_notifier_chain_register.
1716  */
1717 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1718 {
1719         int ret;
1720
1721         if (cpufreq_disabled())
1722                 return -EINVAL;
1723
1724         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1725
1726         switch (list) {
1727         case CPUFREQ_TRANSITION_NOTIFIER:
1728                 mutex_lock(&cpufreq_fast_switch_lock);
1729
1730                 if (cpufreq_fast_switch_count > 0) {
1731                         mutex_unlock(&cpufreq_fast_switch_lock);
1732                         return -EBUSY;
1733                 }
1734                 ret = srcu_notifier_chain_register(
1735                                 &cpufreq_transition_notifier_list, nb);
1736                 if (!ret)
1737                         cpufreq_fast_switch_count--;
1738
1739                 mutex_unlock(&cpufreq_fast_switch_lock);
1740                 break;
1741         case CPUFREQ_POLICY_NOTIFIER:
1742                 ret = blocking_notifier_chain_register(
1743                                 &cpufreq_policy_notifier_list, nb);
1744                 break;
1745         default:
1746                 ret = -EINVAL;
1747         }
1748
1749         return ret;
1750 }
1751 EXPORT_SYMBOL(cpufreq_register_notifier);
1752
1753 /**
1754  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1755  *      @nb: notifier block to be unregistered
1756  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1757  *
1758  *      Remove a driver from the CPU frequency notifier list.
1759  *
1760  *      This function may sleep, and has the same return conditions as
1761  *      blocking_notifier_chain_unregister.
1762  */
1763 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1764 {
1765         int ret;
1766
1767         if (cpufreq_disabled())
1768                 return -EINVAL;
1769
1770         switch (list) {
1771         case CPUFREQ_TRANSITION_NOTIFIER:
1772                 mutex_lock(&cpufreq_fast_switch_lock);
1773
1774                 ret = srcu_notifier_chain_unregister(
1775                                 &cpufreq_transition_notifier_list, nb);
1776                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1777                         cpufreq_fast_switch_count++;
1778
1779                 mutex_unlock(&cpufreq_fast_switch_lock);
1780                 break;
1781         case CPUFREQ_POLICY_NOTIFIER:
1782                 ret = blocking_notifier_chain_unregister(
1783                                 &cpufreq_policy_notifier_list, nb);
1784                 break;
1785         default:
1786                 ret = -EINVAL;
1787         }
1788
1789         return ret;
1790 }
1791 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1792
1793
1794 /*********************************************************************
1795  *                              GOVERNORS                            *
1796  *********************************************************************/
1797
1798 /**
1799  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1800  * @policy: cpufreq policy to switch the frequency for.
1801  * @target_freq: New frequency to set (may be approximate).
1802  *
1803  * Carry out a fast frequency switch without sleeping.
1804  *
1805  * The driver's ->fast_switch() callback invoked by this function must be
1806  * suitable for being called from within RCU-sched read-side critical sections
1807  * and it is expected to select the minimum available frequency greater than or
1808  * equal to @target_freq (CPUFREQ_RELATION_L).
1809  *
1810  * This function must not be called if policy->fast_switch_enabled is unset.
1811  *
1812  * Governors calling this function must guarantee that it will never be invoked
1813  * twice in parallel for the same policy and that it will never be called in
1814  * parallel with either ->target() or ->target_index() for the same policy.
1815  *
1816  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1817  * callback to indicate an error condition, the hardware configuration must be
1818  * preserved.
1819  */
1820 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1821                                         unsigned int target_freq)
1822 {
1823         target_freq = clamp_val(target_freq, policy->min, policy->max);
1824
1825         return cpufreq_driver->fast_switch(policy, target_freq);
1826 }
1827 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1828
1829 /* Must set freqs->new to intermediate frequency */
1830 static int __target_intermediate(struct cpufreq_policy *policy,
1831                                  struct cpufreq_freqs *freqs, int index)
1832 {
1833         int ret;
1834
1835         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1836
1837         /* We don't need to switch to intermediate freq */
1838         if (!freqs->new)
1839                 return 0;
1840
1841         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1842                  __func__, policy->cpu, freqs->old, freqs->new);
1843
1844         cpufreq_freq_transition_begin(policy, freqs);
1845         ret = cpufreq_driver->target_intermediate(policy, index);
1846         cpufreq_freq_transition_end(policy, freqs, ret);
1847
1848         if (ret)
1849                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1850                        __func__, ret);
1851
1852         return ret;
1853 }
1854
1855 static int __target_index(struct cpufreq_policy *policy, int index)
1856 {
1857         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1858         unsigned int intermediate_freq = 0;
1859         unsigned int newfreq = policy->freq_table[index].frequency;
1860         int retval = -EINVAL;
1861         bool notify;
1862
1863         if (newfreq == policy->cur)
1864                 return 0;
1865
1866         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1867         if (notify) {
1868                 /* Handle switching to intermediate frequency */
1869                 if (cpufreq_driver->get_intermediate) {
1870                         retval = __target_intermediate(policy, &freqs, index);
1871                         if (retval)
1872                                 return retval;
1873
1874                         intermediate_freq = freqs.new;
1875                         /* Set old freq to intermediate */
1876                         if (intermediate_freq)
1877                                 freqs.old = freqs.new;
1878                 }
1879
1880                 freqs.new = newfreq;
1881                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1882                          __func__, policy->cpu, freqs.old, freqs.new);
1883
1884                 cpufreq_freq_transition_begin(policy, &freqs);
1885         }
1886
1887         retval = cpufreq_driver->target_index(policy, index);
1888         if (retval)
1889                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1890                        retval);
1891
1892         if (notify) {
1893                 cpufreq_freq_transition_end(policy, &freqs, retval);
1894
1895                 /*
1896                  * Failed after setting to intermediate freq? Driver should have
1897                  * reverted back to initial frequency and so should we. Check
1898                  * here for intermediate_freq instead of get_intermediate, in
1899                  * case we haven't switched to intermediate freq at all.
1900                  */
1901                 if (unlikely(retval && intermediate_freq)) {
1902                         freqs.old = intermediate_freq;
1903                         freqs.new = policy->restore_freq;
1904                         cpufreq_freq_transition_begin(policy, &freqs);
1905                         cpufreq_freq_transition_end(policy, &freqs, 0);
1906                 }
1907         }
1908
1909         return retval;
1910 }
1911
1912 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1913                             unsigned int target_freq,
1914                             unsigned int relation)
1915 {
1916         unsigned int old_target_freq = target_freq;
1917         int index;
1918
1919         if (cpufreq_disabled())
1920                 return -ENODEV;
1921
1922         /* Make sure that target_freq is within supported range */
1923         target_freq = clamp_val(target_freq, policy->min, policy->max);
1924
1925         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1926                  policy->cpu, target_freq, relation, old_target_freq);
1927
1928         /*
1929          * This might look like a redundant call as we are checking it again
1930          * after finding index. But it is left intentionally for cases where
1931          * exactly same freq is called again and so we can save on few function
1932          * calls.
1933          */
1934         if (target_freq == policy->cur)
1935                 return 0;
1936
1937         /* Save last value to restore later on errors */
1938         policy->restore_freq = policy->cur;
1939
1940         if (cpufreq_driver->target)
1941                 return cpufreq_driver->target(policy, target_freq, relation);
1942
1943         if (!cpufreq_driver->target_index)
1944                 return -EINVAL;
1945
1946         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1947
1948         return __target_index(policy, index);
1949 }
1950 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1951
1952 int cpufreq_driver_target(struct cpufreq_policy *policy,
1953                           unsigned int target_freq,
1954                           unsigned int relation)
1955 {
1956         int ret = -EINVAL;
1957
1958         down_write(&policy->rwsem);
1959
1960         ret = __cpufreq_driver_target(policy, target_freq, relation);
1961
1962         up_write(&policy->rwsem);
1963
1964         return ret;
1965 }
1966 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1967
1968 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1969 {
1970         return NULL;
1971 }
1972
1973 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1974 {
1975         int ret;
1976
1977         /* Don't start any governor operations if we are entering suspend */
1978         if (cpufreq_suspended)
1979                 return 0;
1980         /*
1981          * Governor might not be initiated here if ACPI _PPC changed
1982          * notification happened, so check it.
1983          */
1984         if (!policy->governor)
1985                 return -EINVAL;
1986
1987         if (policy->governor->max_transition_latency &&
1988             policy->cpuinfo.transition_latency >
1989             policy->governor->max_transition_latency) {
1990                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
1991
1992                 if (gov) {
1993                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1994                                 policy->governor->name, gov->name);
1995                         policy->governor = gov;
1996                 } else {
1997                         return -EINVAL;
1998                 }
1999         }
2000
2001         if (!try_module_get(policy->governor->owner))
2002                 return -EINVAL;
2003
2004         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2005
2006         if (policy->governor->init) {
2007                 ret = policy->governor->init(policy);
2008                 if (ret) {
2009                         module_put(policy->governor->owner);
2010                         return ret;
2011                 }
2012         }
2013
2014         return 0;
2015 }
2016
2017 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2018 {
2019         if (cpufreq_suspended || !policy->governor)
2020                 return;
2021
2022         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2023
2024         if (policy->governor->exit)
2025                 policy->governor->exit(policy);
2026
2027         module_put(policy->governor->owner);
2028 }
2029
2030 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2031 {
2032         int ret;
2033
2034         if (cpufreq_suspended)
2035                 return 0;
2036
2037         if (!policy->governor)
2038                 return -EINVAL;
2039
2040         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2041
2042         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2043                 cpufreq_update_current_freq(policy);
2044
2045         if (policy->governor->start) {
2046                 ret = policy->governor->start(policy);
2047                 if (ret)
2048                         return ret;
2049         }
2050
2051         if (policy->governor->limits)
2052                 policy->governor->limits(policy);
2053
2054         return 0;
2055 }
2056
2057 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2058 {
2059         if (cpufreq_suspended || !policy->governor)
2060                 return;
2061
2062         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2063
2064         if (policy->governor->stop)
2065                 policy->governor->stop(policy);
2066 }
2067
2068 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2069 {
2070         if (cpufreq_suspended || !policy->governor)
2071                 return;
2072
2073         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2074
2075         if (policy->governor->limits)
2076                 policy->governor->limits(policy);
2077 }
2078
2079 int cpufreq_register_governor(struct cpufreq_governor *governor)
2080 {
2081         int err;
2082
2083         if (!governor)
2084                 return -EINVAL;
2085
2086         if (cpufreq_disabled())
2087                 return -ENODEV;
2088
2089         mutex_lock(&cpufreq_governor_mutex);
2090
2091         err = -EBUSY;
2092         if (!find_governor(governor->name)) {
2093                 err = 0;
2094                 list_add(&governor->governor_list, &cpufreq_governor_list);
2095         }
2096
2097         mutex_unlock(&cpufreq_governor_mutex);
2098         return err;
2099 }
2100 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2101
2102 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2103 {
2104         struct cpufreq_policy *policy;
2105         unsigned long flags;
2106
2107         if (!governor)
2108                 return;
2109
2110         if (cpufreq_disabled())
2111                 return;
2112
2113         /* clear last_governor for all inactive policies */
2114         read_lock_irqsave(&cpufreq_driver_lock, flags);
2115         for_each_inactive_policy(policy) {
2116                 if (!strcmp(policy->last_governor, governor->name)) {
2117                         policy->governor = NULL;
2118                         strcpy(policy->last_governor, "\0");
2119                 }
2120         }
2121         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2122
2123         mutex_lock(&cpufreq_governor_mutex);
2124         list_del(&governor->governor_list);
2125         mutex_unlock(&cpufreq_governor_mutex);
2126         return;
2127 }
2128 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2129
2130
2131 /*********************************************************************
2132  *                          POLICY INTERFACE                         *
2133  *********************************************************************/
2134
2135 /**
2136  * cpufreq_get_policy - get the current cpufreq_policy
2137  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2138  *      is written
2139  *
2140  * Reads the current cpufreq policy.
2141  */
2142 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2143 {
2144         struct cpufreq_policy *cpu_policy;
2145         if (!policy)
2146                 return -EINVAL;
2147
2148         cpu_policy = cpufreq_cpu_get(cpu);
2149         if (!cpu_policy)
2150                 return -EINVAL;
2151
2152         memcpy(policy, cpu_policy, sizeof(*policy));
2153
2154         cpufreq_cpu_put(cpu_policy);
2155         return 0;
2156 }
2157 EXPORT_SYMBOL(cpufreq_get_policy);
2158
2159 /*
2160  * policy : current policy.
2161  * new_policy: policy to be set.
2162  */
2163 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2164                                 struct cpufreq_policy *new_policy)
2165 {
2166         struct cpufreq_governor *old_gov;
2167         int ret;
2168
2169         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2170                  new_policy->cpu, new_policy->min, new_policy->max);
2171
2172         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2173
2174         /*
2175         * This check works well when we store new min/max freq attributes,
2176         * because new_policy is a copy of policy with one field updated.
2177         */
2178         if (new_policy->min > new_policy->max)
2179                 return -EINVAL;
2180
2181         /* verify the cpu speed can be set within this limit */
2182         ret = cpufreq_driver->verify(new_policy);
2183         if (ret)
2184                 return ret;
2185
2186         /* adjust if necessary - all reasons */
2187         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2188                         CPUFREQ_ADJUST, new_policy);
2189
2190         /*
2191          * verify the cpu speed can be set within this limit, which might be
2192          * different to the first one
2193          */
2194         ret = cpufreq_driver->verify(new_policy);
2195         if (ret)
2196                 return ret;
2197
2198         /* notification of the new policy */
2199         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2200                         CPUFREQ_NOTIFY, new_policy);
2201
2202         policy->min = new_policy->min;
2203         policy->max = new_policy->max;
2204
2205         pr_debug("new min and max freqs are %u - %u kHz\n",
2206                  policy->min, policy->max);
2207
2208         if (cpufreq_driver->setpolicy) {
2209                 policy->policy = new_policy->policy;
2210                 pr_debug("setting range\n");
2211                 return cpufreq_driver->setpolicy(new_policy);
2212         }
2213
2214         if (new_policy->governor == policy->governor) {
2215                 pr_debug("cpufreq: governor limits update\n");
2216                 cpufreq_governor_limits(policy);
2217                 return 0;
2218         }
2219
2220         pr_debug("governor switch\n");
2221
2222         /* save old, working values */
2223         old_gov = policy->governor;
2224         /* end old governor */
2225         if (old_gov) {
2226                 cpufreq_stop_governor(policy);
2227                 cpufreq_exit_governor(policy);
2228         }
2229
2230         /* start new governor */
2231         policy->governor = new_policy->governor;
2232         ret = cpufreq_init_governor(policy);
2233         if (!ret) {
2234                 ret = cpufreq_start_governor(policy);
2235                 if (!ret) {
2236                         pr_debug("cpufreq: governor change\n");
2237                         return 0;
2238                 }
2239                 cpufreq_exit_governor(policy);
2240         }
2241
2242         /* new governor failed, so re-start old one */
2243         pr_debug("starting governor %s failed\n", policy->governor->name);
2244         if (old_gov) {
2245                 policy->governor = old_gov;
2246                 if (cpufreq_init_governor(policy))
2247                         policy->governor = NULL;
2248                 else
2249                         cpufreq_start_governor(policy);
2250         }
2251
2252         return ret;
2253 }
2254
2255 /**
2256  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2257  *      @cpu: CPU which shall be re-evaluated
2258  *
2259  *      Useful for policy notifiers which have different necessities
2260  *      at different times.
2261  */
2262 int cpufreq_update_policy(unsigned int cpu)
2263 {
2264         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2265         struct cpufreq_policy new_policy;
2266         int ret;
2267
2268         if (!policy)
2269                 return -ENODEV;
2270
2271         down_write(&policy->rwsem);
2272
2273         pr_debug("updating policy for CPU %u\n", cpu);
2274         memcpy(&new_policy, policy, sizeof(*policy));
2275         new_policy.min = policy->user_policy.min;
2276         new_policy.max = policy->user_policy.max;
2277
2278         /*
2279          * BIOS might change freq behind our back
2280          * -> ask driver for current freq and notify governors about a change
2281          */
2282         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2283                 if (cpufreq_suspended) {
2284                         ret = -EAGAIN;
2285                         goto unlock;
2286                 }
2287                 new_policy.cur = cpufreq_update_current_freq(policy);
2288                 if (WARN_ON(!new_policy.cur)) {
2289                         ret = -EIO;
2290                         goto unlock;
2291                 }
2292         }
2293
2294         ret = cpufreq_set_policy(policy, &new_policy);
2295
2296 unlock:
2297         up_write(&policy->rwsem);
2298
2299         cpufreq_cpu_put(policy);
2300         return ret;
2301 }
2302 EXPORT_SYMBOL(cpufreq_update_policy);
2303
2304 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2305                                         unsigned long action, void *hcpu)
2306 {
2307         unsigned int cpu = (unsigned long)hcpu;
2308
2309         switch (action & ~CPU_TASKS_FROZEN) {
2310         case CPU_ONLINE:
2311         case CPU_DOWN_FAILED:
2312                 cpufreq_online(cpu);
2313                 break;
2314
2315         case CPU_DOWN_PREPARE:
2316                 cpufreq_offline(cpu);
2317                 break;
2318         }
2319         return NOTIFY_OK;
2320 }
2321
2322 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2323         .notifier_call = cpufreq_cpu_callback,
2324 };
2325
2326 /*********************************************************************
2327  *               BOOST                                               *
2328  *********************************************************************/
2329 static int cpufreq_boost_set_sw(int state)
2330 {
2331         struct cpufreq_policy *policy;
2332         int ret = -EINVAL;
2333
2334         for_each_active_policy(policy) {
2335                 if (!policy->freq_table)
2336                         continue;
2337
2338                 ret = cpufreq_frequency_table_cpuinfo(policy,
2339                                                       policy->freq_table);
2340                 if (ret) {
2341                         pr_err("%s: Policy frequency update failed\n",
2342                                __func__);
2343                         break;
2344                 }
2345
2346                 down_write(&policy->rwsem);
2347                 policy->user_policy.max = policy->max;
2348                 cpufreq_governor_limits(policy);
2349                 up_write(&policy->rwsem);
2350         }
2351
2352         return ret;
2353 }
2354
2355 int cpufreq_boost_trigger_state(int state)
2356 {
2357         unsigned long flags;
2358         int ret = 0;
2359
2360         if (cpufreq_driver->boost_enabled == state)
2361                 return 0;
2362
2363         write_lock_irqsave(&cpufreq_driver_lock, flags);
2364         cpufreq_driver->boost_enabled = state;
2365         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2366
2367         ret = cpufreq_driver->set_boost(state);
2368         if (ret) {
2369                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2370                 cpufreq_driver->boost_enabled = !state;
2371                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2372
2373                 pr_err("%s: Cannot %s BOOST\n",
2374                        __func__, state ? "enable" : "disable");
2375         }
2376
2377         return ret;
2378 }
2379
2380 static bool cpufreq_boost_supported(void)
2381 {
2382         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2383 }
2384
2385 static int create_boost_sysfs_file(void)
2386 {
2387         int ret;
2388
2389         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2390         if (ret)
2391                 pr_err("%s: cannot register global BOOST sysfs file\n",
2392                        __func__);
2393
2394         return ret;
2395 }
2396
2397 static void remove_boost_sysfs_file(void)
2398 {
2399         if (cpufreq_boost_supported())
2400                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2401 }
2402
2403 int cpufreq_enable_boost_support(void)
2404 {
2405         if (!cpufreq_driver)
2406                 return -EINVAL;
2407
2408         if (cpufreq_boost_supported())
2409                 return 0;
2410
2411         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2412
2413         /* This will get removed on driver unregister */
2414         return create_boost_sysfs_file();
2415 }
2416 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2417
2418 int cpufreq_boost_enabled(void)
2419 {
2420         return cpufreq_driver->boost_enabled;
2421 }
2422 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2423
2424 /*********************************************************************
2425  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2426  *********************************************************************/
2427
2428 /**
2429  * cpufreq_register_driver - register a CPU Frequency driver
2430  * @driver_data: A struct cpufreq_driver containing the values#
2431  * submitted by the CPU Frequency driver.
2432  *
2433  * Registers a CPU Frequency driver to this core code. This code
2434  * returns zero on success, -EEXIST when another driver got here first
2435  * (and isn't unregistered in the meantime).
2436  *
2437  */
2438 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2439 {
2440         unsigned long flags;
2441         int ret;
2442
2443         if (cpufreq_disabled())
2444                 return -ENODEV;
2445
2446         if (!driver_data || !driver_data->verify || !driver_data->init ||
2447             !(driver_data->setpolicy || driver_data->target_index ||
2448                     driver_data->target) ||
2449              (driver_data->setpolicy && (driver_data->target_index ||
2450                     driver_data->target)) ||
2451              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2452                 return -EINVAL;
2453
2454         pr_debug("trying to register driver %s\n", driver_data->name);
2455
2456         /* Protect against concurrent CPU online/offline. */
2457         get_online_cpus();
2458
2459         write_lock_irqsave(&cpufreq_driver_lock, flags);
2460         if (cpufreq_driver) {
2461                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2462                 ret = -EEXIST;
2463                 goto out;
2464         }
2465         cpufreq_driver = driver_data;
2466         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2467
2468         if (driver_data->setpolicy)
2469                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2470
2471         if (cpufreq_boost_supported()) {
2472                 ret = create_boost_sysfs_file();
2473                 if (ret)
2474                         goto err_null_driver;
2475         }
2476
2477         ret = subsys_interface_register(&cpufreq_interface);
2478         if (ret)
2479                 goto err_boost_unreg;
2480
2481         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2482             list_empty(&cpufreq_policy_list)) {
2483                 /* if all ->init() calls failed, unregister */
2484                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2485                          driver_data->name);
2486                 goto err_if_unreg;
2487         }
2488
2489         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2490         pr_debug("driver %s up and running\n", driver_data->name);
2491         goto out;
2492
2493 err_if_unreg:
2494         subsys_interface_unregister(&cpufreq_interface);
2495 err_boost_unreg:
2496         remove_boost_sysfs_file();
2497 err_null_driver:
2498         write_lock_irqsave(&cpufreq_driver_lock, flags);
2499         cpufreq_driver = NULL;
2500         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2501 out:
2502         put_online_cpus();
2503         return ret;
2504 }
2505 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2506
2507 /**
2508  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2509  *
2510  * Unregister the current CPUFreq driver. Only call this if you have
2511  * the right to do so, i.e. if you have succeeded in initialising before!
2512  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2513  * currently not initialised.
2514  */
2515 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2516 {
2517         unsigned long flags;
2518
2519         if (!cpufreq_driver || (driver != cpufreq_driver))
2520                 return -EINVAL;
2521
2522         pr_debug("unregistering driver %s\n", driver->name);
2523
2524         /* Protect against concurrent cpu hotplug */
2525         get_online_cpus();
2526         subsys_interface_unregister(&cpufreq_interface);
2527         remove_boost_sysfs_file();
2528         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2529
2530         write_lock_irqsave(&cpufreq_driver_lock, flags);
2531
2532         cpufreq_driver = NULL;
2533
2534         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2535         put_online_cpus();
2536
2537         return 0;
2538 }
2539 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2540
2541 /*
2542  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2543  * or mutexes when secondary CPUs are halted.
2544  */
2545 static struct syscore_ops cpufreq_syscore_ops = {
2546         .shutdown = cpufreq_suspend,
2547 };
2548
2549 struct kobject *cpufreq_global_kobject;
2550 EXPORT_SYMBOL(cpufreq_global_kobject);
2551
2552 static int __init cpufreq_core_init(void)
2553 {
2554         if (cpufreq_disabled())
2555                 return -ENODEV;
2556
2557         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2558         BUG_ON(!cpufreq_global_kobject);
2559
2560         register_syscore_ops(&cpufreq_syscore_ops);
2561
2562         return 0;
2563 }
2564 core_initcall(cpufreq_core_init);