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