Merge branch 'for-next-merge' of git://git.kernel.org/pub/scm/linux/kernel/git/nab...
[linux-2.6-block.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 int cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
78 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
79
80 /**
81  * Two notifier lists: the "policy" list is involved in the
82  * validation process for a new CPU frequency policy; the
83  * "transition" list for kernel code that needs to handle
84  * changes to devices when the CPU clock speed changes.
85  * The mutex locks both lists.
86  */
87 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
88 static struct srcu_notifier_head cpufreq_transition_notifier_list;
89
90 static bool init_cpufreq_transition_notifier_list_called;
91 static int __init init_cpufreq_transition_notifier_list(void)
92 {
93         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
94         init_cpufreq_transition_notifier_list_called = true;
95         return 0;
96 }
97 pure_initcall(init_cpufreq_transition_notifier_list);
98
99 static int off __read_mostly;
100 static int cpufreq_disabled(void)
101 {
102         return off;
103 }
104 void disable_cpufreq(void)
105 {
106         off = 1;
107 }
108 static DEFINE_MUTEX(cpufreq_governor_mutex);
109
110 bool have_governor_per_policy(void)
111 {
112         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
113 }
114 EXPORT_SYMBOL_GPL(have_governor_per_policy);
115
116 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
117 {
118         if (have_governor_per_policy())
119                 return &policy->kobj;
120         else
121                 return cpufreq_global_kobject;
122 }
123 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
124
125 struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu)
126 {
127         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
128
129         return policy && !policy_is_inactive(policy) ?
130                 policy->freq_table : NULL;
131 }
132 EXPORT_SYMBOL_GPL(cpufreq_frequency_get_table);
133
134 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
135 {
136         u64 idle_time;
137         u64 cur_wall_time;
138         u64 busy_time;
139
140         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
141
142         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
143         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
144         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
145         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
146         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
147         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
148
149         idle_time = cur_wall_time - busy_time;
150         if (wall)
151                 *wall = cputime_to_usecs(cur_wall_time);
152
153         return cputime_to_usecs(idle_time);
154 }
155
156 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
157 {
158         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
159
160         if (idle_time == -1ULL)
161                 return get_cpu_idle_time_jiffy(cpu, wall);
162         else if (!io_busy)
163                 idle_time += get_cpu_iowait_time_us(cpu, wall);
164
165         return idle_time;
166 }
167 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
168
169 /*
170  * This is a generic cpufreq init() routine which can be used by cpufreq
171  * drivers of SMP systems. It will do following:
172  * - validate & show freq table passed
173  * - set policies transition latency
174  * - policy->cpus with all possible CPUs
175  */
176 int cpufreq_generic_init(struct cpufreq_policy *policy,
177                 struct cpufreq_frequency_table *table,
178                 unsigned int transition_latency)
179 {
180         int ret;
181
182         ret = cpufreq_table_validate_and_show(policy, table);
183         if (ret) {
184                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
185                 return ret;
186         }
187
188         policy->cpuinfo.transition_latency = transition_latency;
189
190         /*
191          * The driver only supports the SMP configuration where all processors
192          * share the clock and voltage and clock.
193          */
194         cpumask_setall(policy->cpus);
195
196         return 0;
197 }
198 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
199
200 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
201 {
202         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
203
204         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
205 }
206 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
207
208 unsigned int cpufreq_generic_get(unsigned int cpu)
209 {
210         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
211
212         if (!policy || IS_ERR(policy->clk)) {
213                 pr_err("%s: No %s associated to cpu: %d\n",
214                        __func__, policy ? "clk" : "policy", cpu);
215                 return 0;
216         }
217
218         return clk_get_rate(policy->clk) / 1000;
219 }
220 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
221
222 /**
223  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
224  *
225  * @cpu: cpu to find policy for.
226  *
227  * This returns policy for 'cpu', returns NULL if it doesn't exist.
228  * It also increments the kobject reference count to mark it busy and so would
229  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
230  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
231  * freed as that depends on the kobj count.
232  *
233  * Return: A valid policy on success, otherwise NULL on failure.
234  */
235 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
236 {
237         struct cpufreq_policy *policy = NULL;
238         unsigned long flags;
239
240         if (WARN_ON(cpu >= nr_cpu_ids))
241                 return NULL;
242
243         /* get the cpufreq driver */
244         read_lock_irqsave(&cpufreq_driver_lock, flags);
245
246         if (cpufreq_driver) {
247                 /* get the CPU */
248                 policy = cpufreq_cpu_get_raw(cpu);
249                 if (policy)
250                         kobject_get(&policy->kobj);
251         }
252
253         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
254
255         return policy;
256 }
257 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
258
259 /**
260  * cpufreq_cpu_put: Decrements the usage count of a policy
261  *
262  * @policy: policy earlier returned by cpufreq_cpu_get().
263  *
264  * This decrements the kobject reference count incremented earlier by calling
265  * cpufreq_cpu_get().
266  */
267 void cpufreq_cpu_put(struct cpufreq_policy *policy)
268 {
269         kobject_put(&policy->kobj);
270 }
271 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
272
273 /*********************************************************************
274  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
275  *********************************************************************/
276
277 /**
278  * adjust_jiffies - adjust the system "loops_per_jiffy"
279  *
280  * This function alters the system "loops_per_jiffy" for the clock
281  * speed change. Note that loops_per_jiffy cannot be updated on SMP
282  * systems as each CPU might be scaled differently. So, use the arch
283  * per-CPU loops_per_jiffy value wherever possible.
284  */
285 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
286 {
287 #ifndef CONFIG_SMP
288         static unsigned long l_p_j_ref;
289         static unsigned int l_p_j_ref_freq;
290
291         if (ci->flags & CPUFREQ_CONST_LOOPS)
292                 return;
293
294         if (!l_p_j_ref_freq) {
295                 l_p_j_ref = loops_per_jiffy;
296                 l_p_j_ref_freq = ci->old;
297                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
298                          l_p_j_ref, l_p_j_ref_freq);
299         }
300         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
301                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
302                                                                 ci->new);
303                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
304                          loops_per_jiffy, ci->new);
305         }
306 #endif
307 }
308
309 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
310                 struct cpufreq_freqs *freqs, unsigned int state)
311 {
312         BUG_ON(irqs_disabled());
313
314         if (cpufreq_disabled())
315                 return;
316
317         freqs->flags = cpufreq_driver->flags;
318         pr_debug("notification %u of frequency transition to %u kHz\n",
319                  state, freqs->new);
320
321         switch (state) {
322
323         case CPUFREQ_PRECHANGE:
324                 /* detect if the driver reported a value as "old frequency"
325                  * which is not equal to what the cpufreq core thinks is
326                  * "old frequency".
327                  */
328                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
329                         if ((policy) && (policy->cpu == freqs->cpu) &&
330                             (policy->cur) && (policy->cur != freqs->old)) {
331                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
332                                          freqs->old, policy->cur);
333                                 freqs->old = policy->cur;
334                         }
335                 }
336                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
337                                 CPUFREQ_PRECHANGE, freqs);
338                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
339                 break;
340
341         case CPUFREQ_POSTCHANGE:
342                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
343                 pr_debug("FREQ: %lu - CPU: %lu\n",
344                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
345                 trace_cpu_frequency(freqs->new, freqs->cpu);
346                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
347                                 CPUFREQ_POSTCHANGE, freqs);
348                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
349                         policy->cur = freqs->new;
350                 break;
351         }
352 }
353
354 /**
355  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
356  * on frequency transition.
357  *
358  * This function calls the transition notifiers and the "adjust_jiffies"
359  * function. It is called twice on all CPU frequency changes that have
360  * external effects.
361  */
362 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
363                 struct cpufreq_freqs *freqs, unsigned int state)
364 {
365         for_each_cpu(freqs->cpu, policy->cpus)
366                 __cpufreq_notify_transition(policy, freqs, state);
367 }
368
369 /* Do post notifications when there are chances that transition has failed */
370 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
371                 struct cpufreq_freqs *freqs, int transition_failed)
372 {
373         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
374         if (!transition_failed)
375                 return;
376
377         swap(freqs->old, freqs->new);
378         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
379         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
380 }
381
382 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
383                 struct cpufreq_freqs *freqs)
384 {
385
386         /*
387          * Catch double invocations of _begin() which lead to self-deadlock.
388          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
389          * doesn't invoke _begin() on their behalf, and hence the chances of
390          * double invocations are very low. Moreover, there are scenarios
391          * where these checks can emit false-positive warnings in these
392          * drivers; so we avoid that by skipping them altogether.
393          */
394         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
395                                 && current == policy->transition_task);
396
397 wait:
398         wait_event(policy->transition_wait, !policy->transition_ongoing);
399
400         spin_lock(&policy->transition_lock);
401
402         if (unlikely(policy->transition_ongoing)) {
403                 spin_unlock(&policy->transition_lock);
404                 goto wait;
405         }
406
407         policy->transition_ongoing = true;
408         policy->transition_task = current;
409
410         spin_unlock(&policy->transition_lock);
411
412         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
413 }
414 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
415
416 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
417                 struct cpufreq_freqs *freqs, int transition_failed)
418 {
419         if (unlikely(WARN_ON(!policy->transition_ongoing)))
420                 return;
421
422         cpufreq_notify_post_transition(policy, freqs, transition_failed);
423
424         policy->transition_ongoing = false;
425         policy->transition_task = NULL;
426
427         wake_up(&policy->transition_wait);
428 }
429 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
430
431
432 /*********************************************************************
433  *                          SYSFS INTERFACE                          *
434  *********************************************************************/
435 static ssize_t show_boost(struct kobject *kobj,
436                                  struct attribute *attr, char *buf)
437 {
438         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
439 }
440
441 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
442                                   const char *buf, size_t count)
443 {
444         int ret, enable;
445
446         ret = sscanf(buf, "%d", &enable);
447         if (ret != 1 || enable < 0 || enable > 1)
448                 return -EINVAL;
449
450         if (cpufreq_boost_trigger_state(enable)) {
451                 pr_err("%s: Cannot %s BOOST!\n",
452                        __func__, enable ? "enable" : "disable");
453                 return -EINVAL;
454         }
455
456         pr_debug("%s: cpufreq BOOST %s\n",
457                  __func__, enable ? "enabled" : "disabled");
458
459         return count;
460 }
461 define_one_global_rw(boost);
462
463 static struct cpufreq_governor *find_governor(const char *str_governor)
464 {
465         struct cpufreq_governor *t;
466
467         for_each_governor(t)
468                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
469                         return t;
470
471         return NULL;
472 }
473
474 /**
475  * cpufreq_parse_governor - parse a governor string
476  */
477 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
478                                 struct cpufreq_governor **governor)
479 {
480         int err = -EINVAL;
481
482         if (cpufreq_driver->setpolicy) {
483                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
484                         *policy = CPUFREQ_POLICY_PERFORMANCE;
485                         err = 0;
486                 } else if (!strncasecmp(str_governor, "powersave",
487                                                 CPUFREQ_NAME_LEN)) {
488                         *policy = CPUFREQ_POLICY_POWERSAVE;
489                         err = 0;
490                 }
491         } else {
492                 struct cpufreq_governor *t;
493
494                 mutex_lock(&cpufreq_governor_mutex);
495
496                 t = find_governor(str_governor);
497
498                 if (t == NULL) {
499                         int ret;
500
501                         mutex_unlock(&cpufreq_governor_mutex);
502                         ret = request_module("cpufreq_%s", str_governor);
503                         mutex_lock(&cpufreq_governor_mutex);
504
505                         if (ret == 0)
506                                 t = find_governor(str_governor);
507                 }
508
509                 if (t != NULL) {
510                         *governor = t;
511                         err = 0;
512                 }
513
514                 mutex_unlock(&cpufreq_governor_mutex);
515         }
516         return err;
517 }
518
519 /**
520  * cpufreq_per_cpu_attr_read() / show_##file_name() -
521  * print out cpufreq information
522  *
523  * Write out information from cpufreq_driver->policy[cpu]; object must be
524  * "unsigned int".
525  */
526
527 #define show_one(file_name, object)                     \
528 static ssize_t show_##file_name                         \
529 (struct cpufreq_policy *policy, char *buf)              \
530 {                                                       \
531         return sprintf(buf, "%u\n", policy->object);    \
532 }
533
534 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
535 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
536 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
537 show_one(scaling_min_freq, min);
538 show_one(scaling_max_freq, max);
539
540 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
541 {
542         ssize_t ret;
543
544         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
545                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
546         else
547                 ret = sprintf(buf, "%u\n", policy->cur);
548         return ret;
549 }
550
551 static int cpufreq_set_policy(struct cpufreq_policy *policy,
552                                 struct cpufreq_policy *new_policy);
553
554 /**
555  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
556  */
557 #define store_one(file_name, object)                    \
558 static ssize_t store_##file_name                                        \
559 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
560 {                                                                       \
561         int ret, temp;                                                  \
562         struct cpufreq_policy new_policy;                               \
563                                                                         \
564         memcpy(&new_policy, policy, sizeof(*policy));                   \
565                                                                         \
566         ret = sscanf(buf, "%u", &new_policy.object);                    \
567         if (ret != 1)                                                   \
568                 return -EINVAL;                                         \
569                                                                         \
570         temp = new_policy.object;                                       \
571         ret = cpufreq_set_policy(policy, &new_policy);          \
572         if (!ret)                                                       \
573                 policy->user_policy.object = temp;                      \
574                                                                         \
575         return ret ? ret : count;                                       \
576 }
577
578 store_one(scaling_min_freq, min);
579 store_one(scaling_max_freq, max);
580
581 /**
582  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
583  */
584 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
585                                         char *buf)
586 {
587         unsigned int cur_freq = __cpufreq_get(policy);
588         if (!cur_freq)
589                 return sprintf(buf, "<unknown>");
590         return sprintf(buf, "%u\n", cur_freq);
591 }
592
593 /**
594  * show_scaling_governor - show the current policy for the specified CPU
595  */
596 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
597 {
598         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
599                 return sprintf(buf, "powersave\n");
600         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
601                 return sprintf(buf, "performance\n");
602         else if (policy->governor)
603                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
604                                 policy->governor->name);
605         return -EINVAL;
606 }
607
608 /**
609  * store_scaling_governor - store policy for the specified CPU
610  */
611 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
612                                         const char *buf, size_t count)
613 {
614         int ret;
615         char    str_governor[16];
616         struct cpufreq_policy new_policy;
617
618         memcpy(&new_policy, policy, sizeof(*policy));
619
620         ret = sscanf(buf, "%15s", str_governor);
621         if (ret != 1)
622                 return -EINVAL;
623
624         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
625                                                 &new_policy.governor))
626                 return -EINVAL;
627
628         ret = cpufreq_set_policy(policy, &new_policy);
629         return ret ? ret : count;
630 }
631
632 /**
633  * show_scaling_driver - show the cpufreq driver currently loaded
634  */
635 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
636 {
637         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
638 }
639
640 /**
641  * show_scaling_available_governors - show the available CPUfreq governors
642  */
643 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
644                                                 char *buf)
645 {
646         ssize_t i = 0;
647         struct cpufreq_governor *t;
648
649         if (!has_target()) {
650                 i += sprintf(buf, "performance powersave");
651                 goto out;
652         }
653
654         for_each_governor(t) {
655                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
656                     - (CPUFREQ_NAME_LEN + 2)))
657                         goto out;
658                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
659         }
660 out:
661         i += sprintf(&buf[i], "\n");
662         return i;
663 }
664
665 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
666 {
667         ssize_t i = 0;
668         unsigned int cpu;
669
670         for_each_cpu(cpu, mask) {
671                 if (i)
672                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
673                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
674                 if (i >= (PAGE_SIZE - 5))
675                         break;
676         }
677         i += sprintf(&buf[i], "\n");
678         return i;
679 }
680 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
681
682 /**
683  * show_related_cpus - show the CPUs affected by each transition even if
684  * hw coordination is in use
685  */
686 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
687 {
688         return cpufreq_show_cpus(policy->related_cpus, buf);
689 }
690
691 /**
692  * show_affected_cpus - show the CPUs affected by each transition
693  */
694 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
695 {
696         return cpufreq_show_cpus(policy->cpus, buf);
697 }
698
699 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
700                                         const char *buf, size_t count)
701 {
702         unsigned int freq = 0;
703         unsigned int ret;
704
705         if (!policy->governor || !policy->governor->store_setspeed)
706                 return -EINVAL;
707
708         ret = sscanf(buf, "%u", &freq);
709         if (ret != 1)
710                 return -EINVAL;
711
712         policy->governor->store_setspeed(policy, freq);
713
714         return count;
715 }
716
717 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
718 {
719         if (!policy->governor || !policy->governor->show_setspeed)
720                 return sprintf(buf, "<unsupported>\n");
721
722         return policy->governor->show_setspeed(policy, buf);
723 }
724
725 /**
726  * show_bios_limit - show the current cpufreq HW/BIOS limitation
727  */
728 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
729 {
730         unsigned int limit;
731         int ret;
732         if (cpufreq_driver->bios_limit) {
733                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
734                 if (!ret)
735                         return sprintf(buf, "%u\n", limit);
736         }
737         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
738 }
739
740 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
741 cpufreq_freq_attr_ro(cpuinfo_min_freq);
742 cpufreq_freq_attr_ro(cpuinfo_max_freq);
743 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
744 cpufreq_freq_attr_ro(scaling_available_governors);
745 cpufreq_freq_attr_ro(scaling_driver);
746 cpufreq_freq_attr_ro(scaling_cur_freq);
747 cpufreq_freq_attr_ro(bios_limit);
748 cpufreq_freq_attr_ro(related_cpus);
749 cpufreq_freq_attr_ro(affected_cpus);
750 cpufreq_freq_attr_rw(scaling_min_freq);
751 cpufreq_freq_attr_rw(scaling_max_freq);
752 cpufreq_freq_attr_rw(scaling_governor);
753 cpufreq_freq_attr_rw(scaling_setspeed);
754
755 static struct attribute *default_attrs[] = {
756         &cpuinfo_min_freq.attr,
757         &cpuinfo_max_freq.attr,
758         &cpuinfo_transition_latency.attr,
759         &scaling_min_freq.attr,
760         &scaling_max_freq.attr,
761         &affected_cpus.attr,
762         &related_cpus.attr,
763         &scaling_governor.attr,
764         &scaling_driver.attr,
765         &scaling_available_governors.attr,
766         &scaling_setspeed.attr,
767         NULL
768 };
769
770 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
771 #define to_attr(a) container_of(a, struct freq_attr, attr)
772
773 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
774 {
775         struct cpufreq_policy *policy = to_policy(kobj);
776         struct freq_attr *fattr = to_attr(attr);
777         ssize_t ret;
778
779         down_read(&policy->rwsem);
780         ret = fattr->show(policy, buf);
781         up_read(&policy->rwsem);
782
783         return ret;
784 }
785
786 static ssize_t store(struct kobject *kobj, struct attribute *attr,
787                      const char *buf, size_t count)
788 {
789         struct cpufreq_policy *policy = to_policy(kobj);
790         struct freq_attr *fattr = to_attr(attr);
791         ssize_t ret = -EINVAL;
792
793         get_online_cpus();
794
795         if (cpu_online(policy->cpu)) {
796                 down_write(&policy->rwsem);
797                 ret = fattr->store(policy, buf, count);
798                 up_write(&policy->rwsem);
799         }
800
801         put_online_cpus();
802
803         return ret;
804 }
805
806 static void cpufreq_sysfs_release(struct kobject *kobj)
807 {
808         struct cpufreq_policy *policy = to_policy(kobj);
809         pr_debug("last reference is dropped\n");
810         complete(&policy->kobj_unregister);
811 }
812
813 static const struct sysfs_ops sysfs_ops = {
814         .show   = show,
815         .store  = store,
816 };
817
818 static struct kobj_type ktype_cpufreq = {
819         .sysfs_ops      = &sysfs_ops,
820         .default_attrs  = default_attrs,
821         .release        = cpufreq_sysfs_release,
822 };
823
824 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
825 {
826         struct device *cpu_dev;
827
828         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
829
830         if (!policy)
831                 return 0;
832
833         cpu_dev = get_cpu_device(cpu);
834         if (WARN_ON(!cpu_dev))
835                 return 0;
836
837         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
838 }
839
840 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
841 {
842         struct device *cpu_dev;
843
844         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
845
846         cpu_dev = get_cpu_device(cpu);
847         if (WARN_ON(!cpu_dev))
848                 return;
849
850         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
851 }
852
853 /* Add/remove symlinks for all related CPUs */
854 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
855 {
856         unsigned int j;
857         int ret = 0;
858
859         /* Some related CPUs might not be present (physically hotplugged) */
860         for_each_cpu(j, policy->real_cpus) {
861                 ret = add_cpu_dev_symlink(policy, j);
862                 if (ret)
863                         break;
864         }
865
866         return ret;
867 }
868
869 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
870 {
871         unsigned int j;
872
873         /* Some related CPUs might not be present (physically hotplugged) */
874         for_each_cpu(j, policy->real_cpus)
875                 remove_cpu_dev_symlink(policy, j);
876 }
877
878 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
879 {
880         struct freq_attr **drv_attr;
881         int ret = 0;
882
883         /* set up files for this cpu device */
884         drv_attr = cpufreq_driver->attr;
885         while (drv_attr && *drv_attr) {
886                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
887                 if (ret)
888                         return ret;
889                 drv_attr++;
890         }
891         if (cpufreq_driver->get) {
892                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
893                 if (ret)
894                         return ret;
895         }
896
897         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
898         if (ret)
899                 return ret;
900
901         if (cpufreq_driver->bios_limit) {
902                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
903                 if (ret)
904                         return ret;
905         }
906
907         return cpufreq_add_dev_symlink(policy);
908 }
909
910 __weak struct cpufreq_governor *cpufreq_default_governor(void)
911 {
912         return NULL;
913 }
914
915 static int cpufreq_init_policy(struct cpufreq_policy *policy)
916 {
917         struct cpufreq_governor *gov = NULL;
918         struct cpufreq_policy new_policy;
919
920         memcpy(&new_policy, policy, sizeof(*policy));
921
922         /* Update governor of new_policy to the governor used before hotplug */
923         gov = find_governor(policy->last_governor);
924         if (gov) {
925                 pr_debug("Restoring governor %s for cpu %d\n",
926                                 policy->governor->name, policy->cpu);
927         } else {
928                 gov = cpufreq_default_governor();
929                 if (!gov)
930                         return -ENODATA;
931         }
932
933         new_policy.governor = gov;
934
935         /* Use the default policy if there is no last_policy. */
936         if (cpufreq_driver->setpolicy) {
937                 if (policy->last_policy)
938                         new_policy.policy = policy->last_policy;
939                 else
940                         cpufreq_parse_governor(gov->name, &new_policy.policy,
941                                                NULL);
942         }
943         /* set default policy */
944         return cpufreq_set_policy(policy, &new_policy);
945 }
946
947 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
948 {
949         int ret = 0;
950
951         /* Has this CPU been taken care of already? */
952         if (cpumask_test_cpu(cpu, policy->cpus))
953                 return 0;
954
955         down_write(&policy->rwsem);
956         if (has_target()) {
957                 ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
958                 if (ret) {
959                         pr_err("%s: Failed to stop governor\n", __func__);
960                         goto unlock;
961                 }
962         }
963
964         cpumask_set_cpu(cpu, policy->cpus);
965
966         if (has_target()) {
967                 ret = cpufreq_governor(policy, CPUFREQ_GOV_START);
968                 if (!ret)
969                         ret = cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
970
971                 if (ret)
972                         pr_err("%s: Failed to start governor\n", __func__);
973         }
974
975 unlock:
976         up_write(&policy->rwsem);
977         return ret;
978 }
979
980 static void handle_update(struct work_struct *work)
981 {
982         struct cpufreq_policy *policy =
983                 container_of(work, struct cpufreq_policy, update);
984         unsigned int cpu = policy->cpu;
985         pr_debug("handle_update for cpu %u called\n", cpu);
986         cpufreq_update_policy(cpu);
987 }
988
989 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
990 {
991         struct device *dev = get_cpu_device(cpu);
992         struct cpufreq_policy *policy;
993         int ret;
994
995         if (WARN_ON(!dev))
996                 return NULL;
997
998         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
999         if (!policy)
1000                 return NULL;
1001
1002         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1003                 goto err_free_policy;
1004
1005         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1006                 goto err_free_cpumask;
1007
1008         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1009                 goto err_free_rcpumask;
1010
1011         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1012                                    cpufreq_global_kobject, "policy%u", cpu);
1013         if (ret) {
1014                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1015                 goto err_free_real_cpus;
1016         }
1017
1018         INIT_LIST_HEAD(&policy->policy_list);
1019         init_rwsem(&policy->rwsem);
1020         spin_lock_init(&policy->transition_lock);
1021         init_waitqueue_head(&policy->transition_wait);
1022         init_completion(&policy->kobj_unregister);
1023         INIT_WORK(&policy->update, handle_update);
1024
1025         policy->cpu = cpu;
1026         return policy;
1027
1028 err_free_real_cpus:
1029         free_cpumask_var(policy->real_cpus);
1030 err_free_rcpumask:
1031         free_cpumask_var(policy->related_cpus);
1032 err_free_cpumask:
1033         free_cpumask_var(policy->cpus);
1034 err_free_policy:
1035         kfree(policy);
1036
1037         return NULL;
1038 }
1039
1040 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1041 {
1042         struct kobject *kobj;
1043         struct completion *cmp;
1044
1045         if (notify)
1046                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1047                                              CPUFREQ_REMOVE_POLICY, policy);
1048
1049         down_write(&policy->rwsem);
1050         cpufreq_remove_dev_symlink(policy);
1051         kobj = &policy->kobj;
1052         cmp = &policy->kobj_unregister;
1053         up_write(&policy->rwsem);
1054         kobject_put(kobj);
1055
1056         /*
1057          * We need to make sure that the underlying kobj is
1058          * actually not referenced anymore by anybody before we
1059          * proceed with unloading.
1060          */
1061         pr_debug("waiting for dropping of refcount\n");
1062         wait_for_completion(cmp);
1063         pr_debug("wait complete\n");
1064 }
1065
1066 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1067 {
1068         unsigned long flags;
1069         int cpu;
1070
1071         /* Remove policy from list */
1072         write_lock_irqsave(&cpufreq_driver_lock, flags);
1073         list_del(&policy->policy_list);
1074
1075         for_each_cpu(cpu, policy->related_cpus)
1076                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1077         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1078
1079         cpufreq_policy_put_kobj(policy, notify);
1080         free_cpumask_var(policy->real_cpus);
1081         free_cpumask_var(policy->related_cpus);
1082         free_cpumask_var(policy->cpus);
1083         kfree(policy);
1084 }
1085
1086 static int cpufreq_online(unsigned int cpu)
1087 {
1088         struct cpufreq_policy *policy;
1089         bool new_policy;
1090         unsigned long flags;
1091         unsigned int j;
1092         int ret;
1093
1094         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1095
1096         /* Check if this CPU already has a policy to manage it */
1097         policy = per_cpu(cpufreq_cpu_data, cpu);
1098         if (policy) {
1099                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1100                 if (!policy_is_inactive(policy))
1101                         return cpufreq_add_policy_cpu(policy, cpu);
1102
1103                 /* This is the only online CPU for the policy.  Start over. */
1104                 new_policy = false;
1105                 down_write(&policy->rwsem);
1106                 policy->cpu = cpu;
1107                 policy->governor = NULL;
1108                 up_write(&policy->rwsem);
1109         } else {
1110                 new_policy = true;
1111                 policy = cpufreq_policy_alloc(cpu);
1112                 if (!policy)
1113                         return -ENOMEM;
1114         }
1115
1116         cpumask_copy(policy->cpus, cpumask_of(cpu));
1117
1118         /* call driver. From then on the cpufreq must be able
1119          * to accept all calls to ->verify and ->setpolicy for this CPU
1120          */
1121         ret = cpufreq_driver->init(policy);
1122         if (ret) {
1123                 pr_debug("initialization failed\n");
1124                 goto out_free_policy;
1125         }
1126
1127         down_write(&policy->rwsem);
1128
1129         if (new_policy) {
1130                 /* related_cpus should at least include policy->cpus. */
1131                 cpumask_copy(policy->related_cpus, policy->cpus);
1132                 /* Remember CPUs present at the policy creation time. */
1133                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1134         }
1135
1136         /*
1137          * affected cpus must always be the one, which are online. We aren't
1138          * managing offline cpus here.
1139          */
1140         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1141
1142         if (new_policy) {
1143                 policy->user_policy.min = policy->min;
1144                 policy->user_policy.max = policy->max;
1145
1146                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1147                 for_each_cpu(j, policy->related_cpus)
1148                         per_cpu(cpufreq_cpu_data, j) = policy;
1149                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1150         }
1151
1152         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1153                 policy->cur = cpufreq_driver->get(policy->cpu);
1154                 if (!policy->cur) {
1155                         pr_err("%s: ->get() failed\n", __func__);
1156                         goto out_exit_policy;
1157                 }
1158         }
1159
1160         /*
1161          * Sometimes boot loaders set CPU frequency to a value outside of
1162          * frequency table present with cpufreq core. In such cases CPU might be
1163          * unstable if it has to run on that frequency for long duration of time
1164          * and so its better to set it to a frequency which is specified in
1165          * freq-table. This also makes cpufreq stats inconsistent as
1166          * cpufreq-stats would fail to register because current frequency of CPU
1167          * isn't found in freq-table.
1168          *
1169          * Because we don't want this change to effect boot process badly, we go
1170          * for the next freq which is >= policy->cur ('cur' must be set by now,
1171          * otherwise we will end up setting freq to lowest of the table as 'cur'
1172          * is initialized to zero).
1173          *
1174          * We are passing target-freq as "policy->cur - 1" otherwise
1175          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1176          * equal to target-freq.
1177          */
1178         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1179             && has_target()) {
1180                 /* Are we running at unknown frequency ? */
1181                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1182                 if (ret == -EINVAL) {
1183                         /* Warn user and fix it */
1184                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1185                                 __func__, policy->cpu, policy->cur);
1186                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1187                                 CPUFREQ_RELATION_L);
1188
1189                         /*
1190                          * Reaching here after boot in a few seconds may not
1191                          * mean that system will remain stable at "unknown"
1192                          * frequency for longer duration. Hence, a BUG_ON().
1193                          */
1194                         BUG_ON(ret);
1195                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1196                                 __func__, policy->cpu, policy->cur);
1197                 }
1198         }
1199
1200         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1201                                      CPUFREQ_START, policy);
1202
1203         if (new_policy) {
1204                 ret = cpufreq_add_dev_interface(policy);
1205                 if (ret)
1206                         goto out_exit_policy;
1207                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1208                                 CPUFREQ_CREATE_POLICY, policy);
1209
1210                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1211                 list_add(&policy->policy_list, &cpufreq_policy_list);
1212                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1213         }
1214
1215         ret = cpufreq_init_policy(policy);
1216         if (ret) {
1217                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1218                        __func__, cpu, ret);
1219                 /* cpufreq_policy_free() will notify based on this */
1220                 new_policy = false;
1221                 goto out_exit_policy;
1222         }
1223
1224         up_write(&policy->rwsem);
1225
1226         kobject_uevent(&policy->kobj, KOBJ_ADD);
1227
1228         /* Callback for handling stuff after policy is ready */
1229         if (cpufreq_driver->ready)
1230                 cpufreq_driver->ready(policy);
1231
1232         pr_debug("initialization complete\n");
1233
1234         return 0;
1235
1236 out_exit_policy:
1237         up_write(&policy->rwsem);
1238
1239         if (cpufreq_driver->exit)
1240                 cpufreq_driver->exit(policy);
1241 out_free_policy:
1242         cpufreq_policy_free(policy, !new_policy);
1243         return ret;
1244 }
1245
1246 /**
1247  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1248  * @dev: CPU device.
1249  * @sif: Subsystem interface structure pointer (not used)
1250  */
1251 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1252 {
1253         unsigned cpu = dev->id;
1254         int ret;
1255
1256         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1257
1258         if (cpu_online(cpu)) {
1259                 ret = cpufreq_online(cpu);
1260         } else {
1261                 /*
1262                  * A hotplug notifier will follow and we will handle it as CPU
1263                  * online then.  For now, just create the sysfs link, unless
1264                  * there is no policy or the link is already present.
1265                  */
1266                 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1267
1268                 ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1269                         ? add_cpu_dev_symlink(policy, cpu) : 0;
1270         }
1271
1272         return ret;
1273 }
1274
1275 static void cpufreq_offline(unsigned int cpu)
1276 {
1277         struct cpufreq_policy *policy;
1278         int ret;
1279
1280         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1281
1282         policy = cpufreq_cpu_get_raw(cpu);
1283         if (!policy) {
1284                 pr_debug("%s: No cpu_data found\n", __func__);
1285                 return;
1286         }
1287
1288         down_write(&policy->rwsem);
1289         if (has_target()) {
1290                 ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1291                 if (ret)
1292                         pr_err("%s: Failed to stop governor\n", __func__);
1293         }
1294
1295         cpumask_clear_cpu(cpu, policy->cpus);
1296
1297         if (policy_is_inactive(policy)) {
1298                 if (has_target())
1299                         strncpy(policy->last_governor, policy->governor->name,
1300                                 CPUFREQ_NAME_LEN);
1301                 else
1302                         policy->last_policy = policy->policy;
1303         } else if (cpu == policy->cpu) {
1304                 /* Nominate new CPU */
1305                 policy->cpu = cpumask_any(policy->cpus);
1306         }
1307
1308         /* Start governor again for active policy */
1309         if (!policy_is_inactive(policy)) {
1310                 if (has_target()) {
1311                         ret = cpufreq_governor(policy, CPUFREQ_GOV_START);
1312                         if (!ret)
1313                                 ret = cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1314
1315                         if (ret)
1316                                 pr_err("%s: Failed to start governor\n", __func__);
1317                 }
1318
1319                 goto unlock;
1320         }
1321
1322         if (cpufreq_driver->stop_cpu)
1323                 cpufreq_driver->stop_cpu(policy);
1324
1325         /* If cpu is last user of policy, free policy */
1326         if (has_target()) {
1327                 ret = cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1328                 if (ret)
1329                         pr_err("%s: Failed to exit governor\n", __func__);
1330         }
1331
1332         /*
1333          * Perform the ->exit() even during light-weight tear-down,
1334          * since this is a core component, and is essential for the
1335          * subsequent light-weight ->init() to succeed.
1336          */
1337         if (cpufreq_driver->exit) {
1338                 cpufreq_driver->exit(policy);
1339                 policy->freq_table = NULL;
1340         }
1341
1342 unlock:
1343         up_write(&policy->rwsem);
1344 }
1345
1346 /**
1347  * cpufreq_remove_dev - remove a CPU device
1348  *
1349  * Removes the cpufreq interface for a CPU device.
1350  */
1351 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1352 {
1353         unsigned int cpu = dev->id;
1354         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1355
1356         if (!policy)
1357                 return;
1358
1359         if (cpu_online(cpu))
1360                 cpufreq_offline(cpu);
1361
1362         cpumask_clear_cpu(cpu, policy->real_cpus);
1363         remove_cpu_dev_symlink(policy, cpu);
1364
1365         if (cpumask_empty(policy->real_cpus))
1366                 cpufreq_policy_free(policy, true);
1367 }
1368
1369 /**
1370  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1371  *      in deep trouble.
1372  *      @policy: policy managing CPUs
1373  *      @new_freq: CPU frequency the CPU actually runs at
1374  *
1375  *      We adjust to current frequency first, and need to clean up later.
1376  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1377  */
1378 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1379                                 unsigned int new_freq)
1380 {
1381         struct cpufreq_freqs freqs;
1382
1383         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1384                  policy->cur, new_freq);
1385
1386         freqs.old = policy->cur;
1387         freqs.new = new_freq;
1388
1389         cpufreq_freq_transition_begin(policy, &freqs);
1390         cpufreq_freq_transition_end(policy, &freqs, 0);
1391 }
1392
1393 /**
1394  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1395  * @cpu: CPU number
1396  *
1397  * This is the last known freq, without actually getting it from the driver.
1398  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1399  */
1400 unsigned int cpufreq_quick_get(unsigned int cpu)
1401 {
1402         struct cpufreq_policy *policy;
1403         unsigned int ret_freq = 0;
1404
1405         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1406                 return cpufreq_driver->get(cpu);
1407
1408         policy = cpufreq_cpu_get(cpu);
1409         if (policy) {
1410                 ret_freq = policy->cur;
1411                 cpufreq_cpu_put(policy);
1412         }
1413
1414         return ret_freq;
1415 }
1416 EXPORT_SYMBOL(cpufreq_quick_get);
1417
1418 /**
1419  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1420  * @cpu: CPU number
1421  *
1422  * Just return the max possible frequency for a given CPU.
1423  */
1424 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1425 {
1426         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1427         unsigned int ret_freq = 0;
1428
1429         if (policy) {
1430                 ret_freq = policy->max;
1431                 cpufreq_cpu_put(policy);
1432         }
1433
1434         return ret_freq;
1435 }
1436 EXPORT_SYMBOL(cpufreq_quick_get_max);
1437
1438 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1439 {
1440         unsigned int ret_freq = 0;
1441
1442         if (!cpufreq_driver->get)
1443                 return ret_freq;
1444
1445         ret_freq = cpufreq_driver->get(policy->cpu);
1446
1447         /* Updating inactive policies is invalid, so avoid doing that. */
1448         if (unlikely(policy_is_inactive(policy)))
1449                 return ret_freq;
1450
1451         if (ret_freq && policy->cur &&
1452                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1453                 /* verify no discrepancy between actual and
1454                                         saved value exists */
1455                 if (unlikely(ret_freq != policy->cur)) {
1456                         cpufreq_out_of_sync(policy, ret_freq);
1457                         schedule_work(&policy->update);
1458                 }
1459         }
1460
1461         return ret_freq;
1462 }
1463
1464 /**
1465  * cpufreq_get - get the current CPU frequency (in kHz)
1466  * @cpu: CPU number
1467  *
1468  * Get the CPU current (static) CPU frequency
1469  */
1470 unsigned int cpufreq_get(unsigned int cpu)
1471 {
1472         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1473         unsigned int ret_freq = 0;
1474
1475         if (policy) {
1476                 down_read(&policy->rwsem);
1477                 ret_freq = __cpufreq_get(policy);
1478                 up_read(&policy->rwsem);
1479
1480                 cpufreq_cpu_put(policy);
1481         }
1482
1483         return ret_freq;
1484 }
1485 EXPORT_SYMBOL(cpufreq_get);
1486
1487 static struct subsys_interface cpufreq_interface = {
1488         .name           = "cpufreq",
1489         .subsys         = &cpu_subsys,
1490         .add_dev        = cpufreq_add_dev,
1491         .remove_dev     = cpufreq_remove_dev,
1492 };
1493
1494 /*
1495  * In case platform wants some specific frequency to be configured
1496  * during suspend..
1497  */
1498 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1499 {
1500         int ret;
1501
1502         if (!policy->suspend_freq) {
1503                 pr_debug("%s: suspend_freq not defined\n", __func__);
1504                 return 0;
1505         }
1506
1507         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1508                         policy->suspend_freq);
1509
1510         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1511                         CPUFREQ_RELATION_H);
1512         if (ret)
1513                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1514                                 __func__, policy->suspend_freq, ret);
1515
1516         return ret;
1517 }
1518 EXPORT_SYMBOL(cpufreq_generic_suspend);
1519
1520 /**
1521  * cpufreq_suspend() - Suspend CPUFreq governors
1522  *
1523  * Called during system wide Suspend/Hibernate cycles for suspending governors
1524  * as some platforms can't change frequency after this point in suspend cycle.
1525  * Because some of the devices (like: i2c, regulators, etc) they use for
1526  * changing frequency are suspended quickly after this point.
1527  */
1528 void cpufreq_suspend(void)
1529 {
1530         struct cpufreq_policy *policy;
1531         int ret;
1532
1533         if (!cpufreq_driver)
1534                 return;
1535
1536         if (!has_target())
1537                 goto suspend;
1538
1539         pr_debug("%s: Suspending Governors\n", __func__);
1540
1541         for_each_active_policy(policy) {
1542                 down_write(&policy->rwsem);
1543                 ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1544                 up_write(&policy->rwsem);
1545
1546                 if (ret)
1547                         pr_err("%s: Failed to stop governor for policy: %p\n",
1548                                 __func__, policy);
1549                 else if (cpufreq_driver->suspend
1550                     && cpufreq_driver->suspend(policy))
1551                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1552                                 policy);
1553         }
1554
1555 suspend:
1556         cpufreq_suspended = true;
1557 }
1558
1559 /**
1560  * cpufreq_resume() - Resume CPUFreq governors
1561  *
1562  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1563  * are suspended with cpufreq_suspend().
1564  */
1565 void cpufreq_resume(void)
1566 {
1567         struct cpufreq_policy *policy;
1568         int ret;
1569
1570         if (!cpufreq_driver)
1571                 return;
1572
1573         cpufreq_suspended = false;
1574
1575         if (!has_target())
1576                 return;
1577
1578         pr_debug("%s: Resuming Governors\n", __func__);
1579
1580         for_each_active_policy(policy) {
1581                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1582                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1583                                 policy);
1584                 } else {
1585                         down_write(&policy->rwsem);
1586                         ret = cpufreq_governor(policy, CPUFREQ_GOV_START);
1587                         if (!ret)
1588                                 cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1589                         up_write(&policy->rwsem);
1590
1591                         if (ret)
1592                                 pr_err("%s: Failed to start governor for policy: %p\n",
1593                                        __func__, policy);
1594                 }
1595         }
1596
1597         /*
1598          * schedule call cpufreq_update_policy() for first-online CPU, as that
1599          * wouldn't be hotplugged-out on suspend. It will verify that the
1600          * current freq is in sync with what we believe it to be.
1601          */
1602         policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1603         if (WARN_ON(!policy))
1604                 return;
1605
1606         schedule_work(&policy->update);
1607 }
1608
1609 /**
1610  *      cpufreq_get_current_driver - return current driver's name
1611  *
1612  *      Return the name string of the currently loaded cpufreq driver
1613  *      or NULL, if none.
1614  */
1615 const char *cpufreq_get_current_driver(void)
1616 {
1617         if (cpufreq_driver)
1618                 return cpufreq_driver->name;
1619
1620         return NULL;
1621 }
1622 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1623
1624 /**
1625  *      cpufreq_get_driver_data - return current driver data
1626  *
1627  *      Return the private data of the currently loaded cpufreq
1628  *      driver, or NULL if no cpufreq driver is loaded.
1629  */
1630 void *cpufreq_get_driver_data(void)
1631 {
1632         if (cpufreq_driver)
1633                 return cpufreq_driver->driver_data;
1634
1635         return NULL;
1636 }
1637 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1638
1639 /*********************************************************************
1640  *                     NOTIFIER LISTS INTERFACE                      *
1641  *********************************************************************/
1642
1643 /**
1644  *      cpufreq_register_notifier - register a driver with cpufreq
1645  *      @nb: notifier function to register
1646  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1647  *
1648  *      Add a driver to one of two lists: either a list of drivers that
1649  *      are notified about clock rate changes (once before and once after
1650  *      the transition), or a list of drivers that are notified about
1651  *      changes in cpufreq policy.
1652  *
1653  *      This function may sleep, and has the same return conditions as
1654  *      blocking_notifier_chain_register.
1655  */
1656 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1657 {
1658         int ret;
1659
1660         if (cpufreq_disabled())
1661                 return -EINVAL;
1662
1663         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1664
1665         switch (list) {
1666         case CPUFREQ_TRANSITION_NOTIFIER:
1667                 ret = srcu_notifier_chain_register(
1668                                 &cpufreq_transition_notifier_list, nb);
1669                 break;
1670         case CPUFREQ_POLICY_NOTIFIER:
1671                 ret = blocking_notifier_chain_register(
1672                                 &cpufreq_policy_notifier_list, nb);
1673                 break;
1674         default:
1675                 ret = -EINVAL;
1676         }
1677
1678         return ret;
1679 }
1680 EXPORT_SYMBOL(cpufreq_register_notifier);
1681
1682 /**
1683  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1684  *      @nb: notifier block to be unregistered
1685  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1686  *
1687  *      Remove a driver from the CPU frequency notifier list.
1688  *
1689  *      This function may sleep, and has the same return conditions as
1690  *      blocking_notifier_chain_unregister.
1691  */
1692 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1693 {
1694         int ret;
1695
1696         if (cpufreq_disabled())
1697                 return -EINVAL;
1698
1699         switch (list) {
1700         case CPUFREQ_TRANSITION_NOTIFIER:
1701                 ret = srcu_notifier_chain_unregister(
1702                                 &cpufreq_transition_notifier_list, nb);
1703                 break;
1704         case CPUFREQ_POLICY_NOTIFIER:
1705                 ret = blocking_notifier_chain_unregister(
1706                                 &cpufreq_policy_notifier_list, nb);
1707                 break;
1708         default:
1709                 ret = -EINVAL;
1710         }
1711
1712         return ret;
1713 }
1714 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1715
1716
1717 /*********************************************************************
1718  *                              GOVERNORS                            *
1719  *********************************************************************/
1720
1721 /* Must set freqs->new to intermediate frequency */
1722 static int __target_intermediate(struct cpufreq_policy *policy,
1723                                  struct cpufreq_freqs *freqs, int index)
1724 {
1725         int ret;
1726
1727         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1728
1729         /* We don't need to switch to intermediate freq */
1730         if (!freqs->new)
1731                 return 0;
1732
1733         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1734                  __func__, policy->cpu, freqs->old, freqs->new);
1735
1736         cpufreq_freq_transition_begin(policy, freqs);
1737         ret = cpufreq_driver->target_intermediate(policy, index);
1738         cpufreq_freq_transition_end(policy, freqs, ret);
1739
1740         if (ret)
1741                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1742                        __func__, ret);
1743
1744         return ret;
1745 }
1746
1747 static int __target_index(struct cpufreq_policy *policy,
1748                           struct cpufreq_frequency_table *freq_table, int index)
1749 {
1750         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1751         unsigned int intermediate_freq = 0;
1752         int retval = -EINVAL;
1753         bool notify;
1754
1755         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1756         if (notify) {
1757                 /* Handle switching to intermediate frequency */
1758                 if (cpufreq_driver->get_intermediate) {
1759                         retval = __target_intermediate(policy, &freqs, index);
1760                         if (retval)
1761                                 return retval;
1762
1763                         intermediate_freq = freqs.new;
1764                         /* Set old freq to intermediate */
1765                         if (intermediate_freq)
1766                                 freqs.old = freqs.new;
1767                 }
1768
1769                 freqs.new = freq_table[index].frequency;
1770                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1771                          __func__, policy->cpu, freqs.old, freqs.new);
1772
1773                 cpufreq_freq_transition_begin(policy, &freqs);
1774         }
1775
1776         retval = cpufreq_driver->target_index(policy, index);
1777         if (retval)
1778                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1779                        retval);
1780
1781         if (notify) {
1782                 cpufreq_freq_transition_end(policy, &freqs, retval);
1783
1784                 /*
1785                  * Failed after setting to intermediate freq? Driver should have
1786                  * reverted back to initial frequency and so should we. Check
1787                  * here for intermediate_freq instead of get_intermediate, in
1788                  * case we haven't switched to intermediate freq at all.
1789                  */
1790                 if (unlikely(retval && intermediate_freq)) {
1791                         freqs.old = intermediate_freq;
1792                         freqs.new = policy->restore_freq;
1793                         cpufreq_freq_transition_begin(policy, &freqs);
1794                         cpufreq_freq_transition_end(policy, &freqs, 0);
1795                 }
1796         }
1797
1798         return retval;
1799 }
1800
1801 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1802                             unsigned int target_freq,
1803                             unsigned int relation)
1804 {
1805         unsigned int old_target_freq = target_freq;
1806         struct cpufreq_frequency_table *freq_table;
1807         int index, retval;
1808
1809         if (cpufreq_disabled())
1810                 return -ENODEV;
1811
1812         /* Make sure that target_freq is within supported range */
1813         if (target_freq > policy->max)
1814                 target_freq = policy->max;
1815         if (target_freq < policy->min)
1816                 target_freq = policy->min;
1817
1818         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1819                  policy->cpu, target_freq, relation, old_target_freq);
1820
1821         /*
1822          * This might look like a redundant call as we are checking it again
1823          * after finding index. But it is left intentionally for cases where
1824          * exactly same freq is called again and so we can save on few function
1825          * calls.
1826          */
1827         if (target_freq == policy->cur)
1828                 return 0;
1829
1830         /* Save last value to restore later on errors */
1831         policy->restore_freq = policy->cur;
1832
1833         if (cpufreq_driver->target)
1834                 return cpufreq_driver->target(policy, target_freq, relation);
1835
1836         if (!cpufreq_driver->target_index)
1837                 return -EINVAL;
1838
1839         freq_table = cpufreq_frequency_get_table(policy->cpu);
1840         if (unlikely(!freq_table)) {
1841                 pr_err("%s: Unable to find freq_table\n", __func__);
1842                 return -EINVAL;
1843         }
1844
1845         retval = cpufreq_frequency_table_target(policy, freq_table, target_freq,
1846                                                 relation, &index);
1847         if (unlikely(retval)) {
1848                 pr_err("%s: Unable to find matching freq\n", __func__);
1849                 return retval;
1850         }
1851
1852         if (freq_table[index].frequency == policy->cur)
1853                 return 0;
1854
1855         return __target_index(policy, freq_table, index);
1856 }
1857 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1858
1859 int cpufreq_driver_target(struct cpufreq_policy *policy,
1860                           unsigned int target_freq,
1861                           unsigned int relation)
1862 {
1863         int ret = -EINVAL;
1864
1865         down_write(&policy->rwsem);
1866
1867         ret = __cpufreq_driver_target(policy, target_freq, relation);
1868
1869         up_write(&policy->rwsem);
1870
1871         return ret;
1872 }
1873 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1874
1875 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1876 {
1877         return NULL;
1878 }
1879
1880 static int cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1881 {
1882         int ret;
1883
1884         /* Don't start any governor operations if we are entering suspend */
1885         if (cpufreq_suspended)
1886                 return 0;
1887         /*
1888          * Governor might not be initiated here if ACPI _PPC changed
1889          * notification happened, so check it.
1890          */
1891         if (!policy->governor)
1892                 return -EINVAL;
1893
1894         if (policy->governor->max_transition_latency &&
1895             policy->cpuinfo.transition_latency >
1896             policy->governor->max_transition_latency) {
1897                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
1898
1899                 if (gov) {
1900                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1901                                 policy->governor->name, gov->name);
1902                         policy->governor = gov;
1903                 } else {
1904                         return -EINVAL;
1905                 }
1906         }
1907
1908         if (event == CPUFREQ_GOV_POLICY_INIT)
1909                 if (!try_module_get(policy->governor->owner))
1910                         return -EINVAL;
1911
1912         pr_debug("%s: for CPU %u, event %u\n", __func__, policy->cpu, event);
1913
1914         ret = policy->governor->governor(policy, event);
1915
1916         if (!ret) {
1917                 if (event == CPUFREQ_GOV_POLICY_INIT)
1918                         policy->governor->initialized++;
1919                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1920                         policy->governor->initialized--;
1921         }
1922
1923         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
1924                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
1925                 module_put(policy->governor->owner);
1926
1927         return ret;
1928 }
1929
1930 int cpufreq_register_governor(struct cpufreq_governor *governor)
1931 {
1932         int err;
1933
1934         if (!governor)
1935                 return -EINVAL;
1936
1937         if (cpufreq_disabled())
1938                 return -ENODEV;
1939
1940         mutex_lock(&cpufreq_governor_mutex);
1941
1942         governor->initialized = 0;
1943         err = -EBUSY;
1944         if (!find_governor(governor->name)) {
1945                 err = 0;
1946                 list_add(&governor->governor_list, &cpufreq_governor_list);
1947         }
1948
1949         mutex_unlock(&cpufreq_governor_mutex);
1950         return err;
1951 }
1952 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1953
1954 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1955 {
1956         struct cpufreq_policy *policy;
1957         unsigned long flags;
1958
1959         if (!governor)
1960                 return;
1961
1962         if (cpufreq_disabled())
1963                 return;
1964
1965         /* clear last_governor for all inactive policies */
1966         read_lock_irqsave(&cpufreq_driver_lock, flags);
1967         for_each_inactive_policy(policy) {
1968                 if (!strcmp(policy->last_governor, governor->name)) {
1969                         policy->governor = NULL;
1970                         strcpy(policy->last_governor, "\0");
1971                 }
1972         }
1973         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1974
1975         mutex_lock(&cpufreq_governor_mutex);
1976         list_del(&governor->governor_list);
1977         mutex_unlock(&cpufreq_governor_mutex);
1978         return;
1979 }
1980 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1981
1982
1983 /*********************************************************************
1984  *                          POLICY INTERFACE                         *
1985  *********************************************************************/
1986
1987 /**
1988  * cpufreq_get_policy - get the current cpufreq_policy
1989  * @policy: struct cpufreq_policy into which the current cpufreq_policy
1990  *      is written
1991  *
1992  * Reads the current cpufreq policy.
1993  */
1994 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1995 {
1996         struct cpufreq_policy *cpu_policy;
1997         if (!policy)
1998                 return -EINVAL;
1999
2000         cpu_policy = cpufreq_cpu_get(cpu);
2001         if (!cpu_policy)
2002                 return -EINVAL;
2003
2004         memcpy(policy, cpu_policy, sizeof(*policy));
2005
2006         cpufreq_cpu_put(cpu_policy);
2007         return 0;
2008 }
2009 EXPORT_SYMBOL(cpufreq_get_policy);
2010
2011 /*
2012  * policy : current policy.
2013  * new_policy: policy to be set.
2014  */
2015 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2016                                 struct cpufreq_policy *new_policy)
2017 {
2018         struct cpufreq_governor *old_gov;
2019         int ret;
2020
2021         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2022                  new_policy->cpu, new_policy->min, new_policy->max);
2023
2024         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2025
2026         /*
2027         * This check works well when we store new min/max freq attributes,
2028         * because new_policy is a copy of policy with one field updated.
2029         */
2030         if (new_policy->min > new_policy->max)
2031                 return -EINVAL;
2032
2033         /* verify the cpu speed can be set within this limit */
2034         ret = cpufreq_driver->verify(new_policy);
2035         if (ret)
2036                 return ret;
2037
2038         /* adjust if necessary - all reasons */
2039         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2040                         CPUFREQ_ADJUST, new_policy);
2041
2042         /*
2043          * verify the cpu speed can be set within this limit, which might be
2044          * different to the first one
2045          */
2046         ret = cpufreq_driver->verify(new_policy);
2047         if (ret)
2048                 return ret;
2049
2050         /* notification of the new policy */
2051         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2052                         CPUFREQ_NOTIFY, new_policy);
2053
2054         policy->min = new_policy->min;
2055         policy->max = new_policy->max;
2056
2057         pr_debug("new min and max freqs are %u - %u kHz\n",
2058                  policy->min, policy->max);
2059
2060         if (cpufreq_driver->setpolicy) {
2061                 policy->policy = new_policy->policy;
2062                 pr_debug("setting range\n");
2063                 return cpufreq_driver->setpolicy(new_policy);
2064         }
2065
2066         if (new_policy->governor == policy->governor)
2067                 goto out;
2068
2069         pr_debug("governor switch\n");
2070
2071         /* save old, working values */
2072         old_gov = policy->governor;
2073         /* end old governor */
2074         if (old_gov) {
2075                 ret = cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2076                 if (ret) {
2077                         /* This can happen due to race with other operations */
2078                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2079                                  __func__, old_gov->name, ret);
2080                         return ret;
2081                 }
2082
2083                 ret = cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2084                 if (ret) {
2085                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2086                                __func__, old_gov->name, ret);
2087                         return ret;
2088                 }
2089         }
2090
2091         /* start new governor */
2092         policy->governor = new_policy->governor;
2093         ret = cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2094         if (!ret) {
2095                 ret = cpufreq_governor(policy, CPUFREQ_GOV_START);
2096                 if (!ret)
2097                         goto out;
2098
2099                 cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2100         }
2101
2102         /* new governor failed, so re-start old one */
2103         pr_debug("starting governor %s failed\n", policy->governor->name);
2104         if (old_gov) {
2105                 policy->governor = old_gov;
2106                 if (cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2107                         policy->governor = NULL;
2108                 else
2109                         cpufreq_governor(policy, CPUFREQ_GOV_START);
2110         }
2111
2112         return ret;
2113
2114  out:
2115         pr_debug("governor: change or update limits\n");
2116         return cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2117 }
2118
2119 /**
2120  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2121  *      @cpu: CPU which shall be re-evaluated
2122  *
2123  *      Useful for policy notifiers which have different necessities
2124  *      at different times.
2125  */
2126 int cpufreq_update_policy(unsigned int cpu)
2127 {
2128         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2129         struct cpufreq_policy new_policy;
2130         int ret;
2131
2132         if (!policy)
2133                 return -ENODEV;
2134
2135         down_write(&policy->rwsem);
2136
2137         pr_debug("updating policy for CPU %u\n", cpu);
2138         memcpy(&new_policy, policy, sizeof(*policy));
2139         new_policy.min = policy->user_policy.min;
2140         new_policy.max = policy->user_policy.max;
2141
2142         /*
2143          * BIOS might change freq behind our back
2144          * -> ask driver for current freq and notify governors about a change
2145          */
2146         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2147                 new_policy.cur = cpufreq_driver->get(cpu);
2148                 if (WARN_ON(!new_policy.cur)) {
2149                         ret = -EIO;
2150                         goto unlock;
2151                 }
2152
2153                 if (!policy->cur) {
2154                         pr_debug("Driver did not initialize current freq\n");
2155                         policy->cur = new_policy.cur;
2156                 } else {
2157                         if (policy->cur != new_policy.cur && has_target())
2158                                 cpufreq_out_of_sync(policy, new_policy.cur);
2159                 }
2160         }
2161
2162         ret = cpufreq_set_policy(policy, &new_policy);
2163
2164 unlock:
2165         up_write(&policy->rwsem);
2166
2167         cpufreq_cpu_put(policy);
2168         return ret;
2169 }
2170 EXPORT_SYMBOL(cpufreq_update_policy);
2171
2172 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2173                                         unsigned long action, void *hcpu)
2174 {
2175         unsigned int cpu = (unsigned long)hcpu;
2176
2177         switch (action & ~CPU_TASKS_FROZEN) {
2178         case CPU_ONLINE:
2179                 cpufreq_online(cpu);
2180                 break;
2181
2182         case CPU_DOWN_PREPARE:
2183                 cpufreq_offline(cpu);
2184                 break;
2185
2186         case CPU_DOWN_FAILED:
2187                 cpufreq_online(cpu);
2188                 break;
2189         }
2190         return NOTIFY_OK;
2191 }
2192
2193 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2194         .notifier_call = cpufreq_cpu_callback,
2195 };
2196
2197 /*********************************************************************
2198  *               BOOST                                               *
2199  *********************************************************************/
2200 static int cpufreq_boost_set_sw(int state)
2201 {
2202         struct cpufreq_frequency_table *freq_table;
2203         struct cpufreq_policy *policy;
2204         int ret = -EINVAL;
2205
2206         for_each_active_policy(policy) {
2207                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2208                 if (freq_table) {
2209                         ret = cpufreq_frequency_table_cpuinfo(policy,
2210                                                         freq_table);
2211                         if (ret) {
2212                                 pr_err("%s: Policy frequency update failed\n",
2213                                        __func__);
2214                                 break;
2215                         }
2216
2217                         down_write(&policy->rwsem);
2218                         policy->user_policy.max = policy->max;
2219                         cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2220                         up_write(&policy->rwsem);
2221                 }
2222         }
2223
2224         return ret;
2225 }
2226
2227 int cpufreq_boost_trigger_state(int state)
2228 {
2229         unsigned long flags;
2230         int ret = 0;
2231
2232         if (cpufreq_driver->boost_enabled == state)
2233                 return 0;
2234
2235         write_lock_irqsave(&cpufreq_driver_lock, flags);
2236         cpufreq_driver->boost_enabled = state;
2237         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2238
2239         ret = cpufreq_driver->set_boost(state);
2240         if (ret) {
2241                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2242                 cpufreq_driver->boost_enabled = !state;
2243                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2244
2245                 pr_err("%s: Cannot %s BOOST\n",
2246                        __func__, state ? "enable" : "disable");
2247         }
2248
2249         return ret;
2250 }
2251
2252 static bool cpufreq_boost_supported(void)
2253 {
2254         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2255 }
2256
2257 static int create_boost_sysfs_file(void)
2258 {
2259         int ret;
2260
2261         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2262         if (ret)
2263                 pr_err("%s: cannot register global BOOST sysfs file\n",
2264                        __func__);
2265
2266         return ret;
2267 }
2268
2269 static void remove_boost_sysfs_file(void)
2270 {
2271         if (cpufreq_boost_supported())
2272                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2273 }
2274
2275 int cpufreq_enable_boost_support(void)
2276 {
2277         if (!cpufreq_driver)
2278                 return -EINVAL;
2279
2280         if (cpufreq_boost_supported())
2281                 return 0;
2282
2283         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2284
2285         /* This will get removed on driver unregister */
2286         return create_boost_sysfs_file();
2287 }
2288 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2289
2290 int cpufreq_boost_enabled(void)
2291 {
2292         return cpufreq_driver->boost_enabled;
2293 }
2294 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2295
2296 /*********************************************************************
2297  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2298  *********************************************************************/
2299
2300 /**
2301  * cpufreq_register_driver - register a CPU Frequency driver
2302  * @driver_data: A struct cpufreq_driver containing the values#
2303  * submitted by the CPU Frequency driver.
2304  *
2305  * Registers a CPU Frequency driver to this core code. This code
2306  * returns zero on success, -EEXIST when another driver got here first
2307  * (and isn't unregistered in the meantime).
2308  *
2309  */
2310 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2311 {
2312         unsigned long flags;
2313         int ret;
2314
2315         if (cpufreq_disabled())
2316                 return -ENODEV;
2317
2318         if (!driver_data || !driver_data->verify || !driver_data->init ||
2319             !(driver_data->setpolicy || driver_data->target_index ||
2320                     driver_data->target) ||
2321              (driver_data->setpolicy && (driver_data->target_index ||
2322                     driver_data->target)) ||
2323              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2324                 return -EINVAL;
2325
2326         pr_debug("trying to register driver %s\n", driver_data->name);
2327
2328         /* Protect against concurrent CPU online/offline. */
2329         get_online_cpus();
2330
2331         write_lock_irqsave(&cpufreq_driver_lock, flags);
2332         if (cpufreq_driver) {
2333                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2334                 ret = -EEXIST;
2335                 goto out;
2336         }
2337         cpufreq_driver = driver_data;
2338         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2339
2340         if (driver_data->setpolicy)
2341                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2342
2343         if (cpufreq_boost_supported()) {
2344                 ret = create_boost_sysfs_file();
2345                 if (ret)
2346                         goto err_null_driver;
2347         }
2348
2349         ret = subsys_interface_register(&cpufreq_interface);
2350         if (ret)
2351                 goto err_boost_unreg;
2352
2353         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2354             list_empty(&cpufreq_policy_list)) {
2355                 /* if all ->init() calls failed, unregister */
2356                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2357                          driver_data->name);
2358                 goto err_if_unreg;
2359         }
2360
2361         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2362         pr_debug("driver %s up and running\n", driver_data->name);
2363
2364 out:
2365         put_online_cpus();
2366         return ret;
2367
2368 err_if_unreg:
2369         subsys_interface_unregister(&cpufreq_interface);
2370 err_boost_unreg:
2371         remove_boost_sysfs_file();
2372 err_null_driver:
2373         write_lock_irqsave(&cpufreq_driver_lock, flags);
2374         cpufreq_driver = NULL;
2375         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2376         goto out;
2377 }
2378 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2379
2380 /**
2381  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2382  *
2383  * Unregister the current CPUFreq driver. Only call this if you have
2384  * the right to do so, i.e. if you have succeeded in initialising before!
2385  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2386  * currently not initialised.
2387  */
2388 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2389 {
2390         unsigned long flags;
2391
2392         if (!cpufreq_driver || (driver != cpufreq_driver))
2393                 return -EINVAL;
2394
2395         pr_debug("unregistering driver %s\n", driver->name);
2396
2397         /* Protect against concurrent cpu hotplug */
2398         get_online_cpus();
2399         subsys_interface_unregister(&cpufreq_interface);
2400         remove_boost_sysfs_file();
2401         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2402
2403         write_lock_irqsave(&cpufreq_driver_lock, flags);
2404
2405         cpufreq_driver = NULL;
2406
2407         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2408         put_online_cpus();
2409
2410         return 0;
2411 }
2412 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2413
2414 /*
2415  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2416  * or mutexes when secondary CPUs are halted.
2417  */
2418 static struct syscore_ops cpufreq_syscore_ops = {
2419         .shutdown = cpufreq_suspend,
2420 };
2421
2422 struct kobject *cpufreq_global_kobject;
2423 EXPORT_SYMBOL(cpufreq_global_kobject);
2424
2425 static int __init cpufreq_core_init(void)
2426 {
2427         if (cpufreq_disabled())
2428                 return -ENODEV;
2429
2430         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2431         BUG_ON(!cpufreq_global_kobject);
2432
2433         register_syscore_ops(&cpufreq_syscore_ops);
2434
2435         return 0;
2436 }
2437 core_initcall(cpufreq_core_init);