Merge branches 'pm-cpufreq', 'pm-cpuidle', 'pm-devfreq', 'pm-opp' and 'pm-tools'
[linux-drm-fsl-dcu.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 /* Macros to iterate over lists */
35 /* Iterate over online CPUs policies */
36 static LIST_HEAD(cpufreq_policy_list);
37 #define for_each_policy(__policy)                               \
38         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
39
40 /* Iterate over governors */
41 static LIST_HEAD(cpufreq_governor_list);
42 #define for_each_governor(__governor)                           \
43         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
44
45 /**
46  * The "cpufreq driver" - the arch- or hardware-dependent low
47  * level driver of CPUFreq support, and its spinlock. This lock
48  * also protects the cpufreq_cpu_data array.
49  */
50 static struct cpufreq_driver *cpufreq_driver;
51 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
52 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data_fallback);
53 static DEFINE_RWLOCK(cpufreq_driver_lock);
54 DEFINE_MUTEX(cpufreq_governor_lock);
55
56 /* This one keeps track of the previously set governor of a removed CPU */
57 static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
58
59 /* Flag to suspend/resume CPUFreq governors */
60 static bool cpufreq_suspended;
61
62 static inline bool has_target(void)
63 {
64         return cpufreq_driver->target_index || cpufreq_driver->target;
65 }
66
67 /*
68  * rwsem to guarantee that cpufreq driver module doesn't unload during critical
69  * sections
70  */
71 static DECLARE_RWSEM(cpufreq_rwsem);
72
73 /* internal prototypes */
74 static int __cpufreq_governor(struct cpufreq_policy *policy,
75                 unsigned int event);
76 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
77 static void handle_update(struct work_struct *work);
78
79 /**
80  * Two notifier lists: the "policy" list is involved in the
81  * validation process for a new CPU frequency policy; the
82  * "transition" list for kernel code that needs to handle
83  * changes to devices when the CPU clock speed changes.
84  * The mutex locks both lists.
85  */
86 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
87 static struct srcu_notifier_head cpufreq_transition_notifier_list;
88
89 static bool init_cpufreq_transition_notifier_list_called;
90 static int __init init_cpufreq_transition_notifier_list(void)
91 {
92         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
93         init_cpufreq_transition_notifier_list_called = true;
94         return 0;
95 }
96 pure_initcall(init_cpufreq_transition_notifier_list);
97
98 static int off __read_mostly;
99 static int cpufreq_disabled(void)
100 {
101         return off;
102 }
103 void disable_cpufreq(void)
104 {
105         off = 1;
106 }
107 static DEFINE_MUTEX(cpufreq_governor_mutex);
108
109 bool have_governor_per_policy(void)
110 {
111         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
112 }
113 EXPORT_SYMBOL_GPL(have_governor_per_policy);
114
115 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
116 {
117         if (have_governor_per_policy())
118                 return &policy->kobj;
119         else
120                 return cpufreq_global_kobject;
121 }
122 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
123
124 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
125 {
126         u64 idle_time;
127         u64 cur_wall_time;
128         u64 busy_time;
129
130         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
131
132         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
133         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
134         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
135         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
136         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
137         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
138
139         idle_time = cur_wall_time - busy_time;
140         if (wall)
141                 *wall = cputime_to_usecs(cur_wall_time);
142
143         return cputime_to_usecs(idle_time);
144 }
145
146 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
147 {
148         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
149
150         if (idle_time == -1ULL)
151                 return get_cpu_idle_time_jiffy(cpu, wall);
152         else if (!io_busy)
153                 idle_time += get_cpu_iowait_time_us(cpu, wall);
154
155         return idle_time;
156 }
157 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
158
159 /*
160  * This is a generic cpufreq init() routine which can be used by cpufreq
161  * drivers of SMP systems. It will do following:
162  * - validate & show freq table passed
163  * - set policies transition latency
164  * - policy->cpus with all possible CPUs
165  */
166 int cpufreq_generic_init(struct cpufreq_policy *policy,
167                 struct cpufreq_frequency_table *table,
168                 unsigned int transition_latency)
169 {
170         int ret;
171
172         ret = cpufreq_table_validate_and_show(policy, table);
173         if (ret) {
174                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
175                 return ret;
176         }
177
178         policy->cpuinfo.transition_latency = transition_latency;
179
180         /*
181          * The driver only supports the SMP configuartion where all processors
182          * share the clock and voltage and clock.
183          */
184         cpumask_setall(policy->cpus);
185
186         return 0;
187 }
188 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
189
190 unsigned int cpufreq_generic_get(unsigned int cpu)
191 {
192         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
193
194         if (!policy || IS_ERR(policy->clk)) {
195                 pr_err("%s: No %s associated to cpu: %d\n",
196                        __func__, policy ? "clk" : "policy", cpu);
197                 return 0;
198         }
199
200         return clk_get_rate(policy->clk) / 1000;
201 }
202 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
203
204 /* Only for cpufreq core internal use */
205 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
206 {
207         return per_cpu(cpufreq_cpu_data, cpu);
208 }
209
210 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
211 {
212         struct cpufreq_policy *policy = NULL;
213         unsigned long flags;
214
215         if (cpu >= nr_cpu_ids)
216                 return NULL;
217
218         if (!down_read_trylock(&cpufreq_rwsem))
219                 return NULL;
220
221         /* get the cpufreq driver */
222         read_lock_irqsave(&cpufreq_driver_lock, flags);
223
224         if (cpufreq_driver) {
225                 /* get the CPU */
226                 policy = per_cpu(cpufreq_cpu_data, cpu);
227                 if (policy)
228                         kobject_get(&policy->kobj);
229         }
230
231         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
232
233         if (!policy)
234                 up_read(&cpufreq_rwsem);
235
236         return policy;
237 }
238 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
239
240 void cpufreq_cpu_put(struct cpufreq_policy *policy)
241 {
242         kobject_put(&policy->kobj);
243         up_read(&cpufreq_rwsem);
244 }
245 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
246
247 /*********************************************************************
248  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
249  *********************************************************************/
250
251 /**
252  * adjust_jiffies - adjust the system "loops_per_jiffy"
253  *
254  * This function alters the system "loops_per_jiffy" for the clock
255  * speed change. Note that loops_per_jiffy cannot be updated on SMP
256  * systems as each CPU might be scaled differently. So, use the arch
257  * per-CPU loops_per_jiffy value wherever possible.
258  */
259 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
260 {
261 #ifndef CONFIG_SMP
262         static unsigned long l_p_j_ref;
263         static unsigned int l_p_j_ref_freq;
264
265         if (ci->flags & CPUFREQ_CONST_LOOPS)
266                 return;
267
268         if (!l_p_j_ref_freq) {
269                 l_p_j_ref = loops_per_jiffy;
270                 l_p_j_ref_freq = ci->old;
271                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
272                          l_p_j_ref, l_p_j_ref_freq);
273         }
274         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
275                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
276                                                                 ci->new);
277                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
278                          loops_per_jiffy, ci->new);
279         }
280 #endif
281 }
282
283 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
284                 struct cpufreq_freqs *freqs, unsigned int state)
285 {
286         BUG_ON(irqs_disabled());
287
288         if (cpufreq_disabled())
289                 return;
290
291         freqs->flags = cpufreq_driver->flags;
292         pr_debug("notification %u of frequency transition to %u kHz\n",
293                  state, freqs->new);
294
295         switch (state) {
296
297         case CPUFREQ_PRECHANGE:
298                 /* detect if the driver reported a value as "old frequency"
299                  * which is not equal to what the cpufreq core thinks is
300                  * "old frequency".
301                  */
302                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
303                         if ((policy) && (policy->cpu == freqs->cpu) &&
304                             (policy->cur) && (policy->cur != freqs->old)) {
305                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
306                                          freqs->old, policy->cur);
307                                 freqs->old = policy->cur;
308                         }
309                 }
310                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
311                                 CPUFREQ_PRECHANGE, freqs);
312                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
313                 break;
314
315         case CPUFREQ_POSTCHANGE:
316                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
317                 pr_debug("FREQ: %lu - CPU: %lu\n",
318                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
319                 trace_cpu_frequency(freqs->new, freqs->cpu);
320                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
321                                 CPUFREQ_POSTCHANGE, freqs);
322                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
323                         policy->cur = freqs->new;
324                 break;
325         }
326 }
327
328 /**
329  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
330  * on frequency transition.
331  *
332  * This function calls the transition notifiers and the "adjust_jiffies"
333  * function. It is called twice on all CPU frequency changes that have
334  * external effects.
335  */
336 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
337                 struct cpufreq_freqs *freqs, unsigned int state)
338 {
339         for_each_cpu(freqs->cpu, policy->cpus)
340                 __cpufreq_notify_transition(policy, freqs, state);
341 }
342
343 /* Do post notifications when there are chances that transition has failed */
344 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
345                 struct cpufreq_freqs *freqs, int transition_failed)
346 {
347         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
348         if (!transition_failed)
349                 return;
350
351         swap(freqs->old, freqs->new);
352         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
353         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
354 }
355
356 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
357                 struct cpufreq_freqs *freqs)
358 {
359
360         /*
361          * Catch double invocations of _begin() which lead to self-deadlock.
362          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
363          * doesn't invoke _begin() on their behalf, and hence the chances of
364          * double invocations are very low. Moreover, there are scenarios
365          * where these checks can emit false-positive warnings in these
366          * drivers; so we avoid that by skipping them altogether.
367          */
368         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
369                                 && current == policy->transition_task);
370
371 wait:
372         wait_event(policy->transition_wait, !policy->transition_ongoing);
373
374         spin_lock(&policy->transition_lock);
375
376         if (unlikely(policy->transition_ongoing)) {
377                 spin_unlock(&policy->transition_lock);
378                 goto wait;
379         }
380
381         policy->transition_ongoing = true;
382         policy->transition_task = current;
383
384         spin_unlock(&policy->transition_lock);
385
386         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
387 }
388 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
389
390 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
391                 struct cpufreq_freqs *freqs, int transition_failed)
392 {
393         if (unlikely(WARN_ON(!policy->transition_ongoing)))
394                 return;
395
396         cpufreq_notify_post_transition(policy, freqs, transition_failed);
397
398         policy->transition_ongoing = false;
399         policy->transition_task = NULL;
400
401         wake_up(&policy->transition_wait);
402 }
403 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
404
405
406 /*********************************************************************
407  *                          SYSFS INTERFACE                          *
408  *********************************************************************/
409 static ssize_t show_boost(struct kobject *kobj,
410                                  struct attribute *attr, char *buf)
411 {
412         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
413 }
414
415 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
416                                   const char *buf, size_t count)
417 {
418         int ret, enable;
419
420         ret = sscanf(buf, "%d", &enable);
421         if (ret != 1 || enable < 0 || enable > 1)
422                 return -EINVAL;
423
424         if (cpufreq_boost_trigger_state(enable)) {
425                 pr_err("%s: Cannot %s BOOST!\n",
426                        __func__, enable ? "enable" : "disable");
427                 return -EINVAL;
428         }
429
430         pr_debug("%s: cpufreq BOOST %s\n",
431                  __func__, enable ? "enabled" : "disabled");
432
433         return count;
434 }
435 define_one_global_rw(boost);
436
437 static struct cpufreq_governor *find_governor(const char *str_governor)
438 {
439         struct cpufreq_governor *t;
440
441         for_each_governor(t)
442                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
443                         return t;
444
445         return NULL;
446 }
447
448 /**
449  * cpufreq_parse_governor - parse a governor string
450  */
451 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
452                                 struct cpufreq_governor **governor)
453 {
454         int err = -EINVAL;
455
456         if (!cpufreq_driver)
457                 goto out;
458
459         if (cpufreq_driver->setpolicy) {
460                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
461                         *policy = CPUFREQ_POLICY_PERFORMANCE;
462                         err = 0;
463                 } else if (!strncasecmp(str_governor, "powersave",
464                                                 CPUFREQ_NAME_LEN)) {
465                         *policy = CPUFREQ_POLICY_POWERSAVE;
466                         err = 0;
467                 }
468         } else {
469                 struct cpufreq_governor *t;
470
471                 mutex_lock(&cpufreq_governor_mutex);
472
473                 t = find_governor(str_governor);
474
475                 if (t == NULL) {
476                         int ret;
477
478                         mutex_unlock(&cpufreq_governor_mutex);
479                         ret = request_module("cpufreq_%s", str_governor);
480                         mutex_lock(&cpufreq_governor_mutex);
481
482                         if (ret == 0)
483                                 t = find_governor(str_governor);
484                 }
485
486                 if (t != NULL) {
487                         *governor = t;
488                         err = 0;
489                 }
490
491                 mutex_unlock(&cpufreq_governor_mutex);
492         }
493 out:
494         return err;
495 }
496
497 /**
498  * cpufreq_per_cpu_attr_read() / show_##file_name() -
499  * print out cpufreq information
500  *
501  * Write out information from cpufreq_driver->policy[cpu]; object must be
502  * "unsigned int".
503  */
504
505 #define show_one(file_name, object)                     \
506 static ssize_t show_##file_name                         \
507 (struct cpufreq_policy *policy, char *buf)              \
508 {                                                       \
509         return sprintf(buf, "%u\n", policy->object);    \
510 }
511
512 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
513 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
514 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
515 show_one(scaling_min_freq, min);
516 show_one(scaling_max_freq, max);
517
518 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
519 {
520         ssize_t ret;
521
522         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
523                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
524         else
525                 ret = sprintf(buf, "%u\n", policy->cur);
526         return ret;
527 }
528
529 static int cpufreq_set_policy(struct cpufreq_policy *policy,
530                                 struct cpufreq_policy *new_policy);
531
532 /**
533  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
534  */
535 #define store_one(file_name, object)                    \
536 static ssize_t store_##file_name                                        \
537 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
538 {                                                                       \
539         int ret, temp;                                                  \
540         struct cpufreq_policy new_policy;                               \
541                                                                         \
542         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
543         if (ret)                                                        \
544                 return -EINVAL;                                         \
545                                                                         \
546         ret = sscanf(buf, "%u", &new_policy.object);                    \
547         if (ret != 1)                                                   \
548                 return -EINVAL;                                         \
549                                                                         \
550         temp = new_policy.object;                                       \
551         ret = cpufreq_set_policy(policy, &new_policy);          \
552         if (!ret)                                                       \
553                 policy->user_policy.object = temp;                      \
554                                                                         \
555         return ret ? ret : count;                                       \
556 }
557
558 store_one(scaling_min_freq, min);
559 store_one(scaling_max_freq, max);
560
561 /**
562  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
563  */
564 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
565                                         char *buf)
566 {
567         unsigned int cur_freq = __cpufreq_get(policy);
568         if (!cur_freq)
569                 return sprintf(buf, "<unknown>");
570         return sprintf(buf, "%u\n", cur_freq);
571 }
572
573 /**
574  * show_scaling_governor - show the current policy for the specified CPU
575  */
576 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
577 {
578         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
579                 return sprintf(buf, "powersave\n");
580         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
581                 return sprintf(buf, "performance\n");
582         else if (policy->governor)
583                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
584                                 policy->governor->name);
585         return -EINVAL;
586 }
587
588 /**
589  * store_scaling_governor - store policy for the specified CPU
590  */
591 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
592                                         const char *buf, size_t count)
593 {
594         int ret;
595         char    str_governor[16];
596         struct cpufreq_policy new_policy;
597
598         ret = cpufreq_get_policy(&new_policy, policy->cpu);
599         if (ret)
600                 return ret;
601
602         ret = sscanf(buf, "%15s", str_governor);
603         if (ret != 1)
604                 return -EINVAL;
605
606         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
607                                                 &new_policy.governor))
608                 return -EINVAL;
609
610         ret = cpufreq_set_policy(policy, &new_policy);
611
612         policy->user_policy.policy = policy->policy;
613         policy->user_policy.governor = policy->governor;
614
615         if (ret)
616                 return ret;
617         else
618                 return count;
619 }
620
621 /**
622  * show_scaling_driver - show the cpufreq driver currently loaded
623  */
624 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
625 {
626         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
627 }
628
629 /**
630  * show_scaling_available_governors - show the available CPUfreq governors
631  */
632 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
633                                                 char *buf)
634 {
635         ssize_t i = 0;
636         struct cpufreq_governor *t;
637
638         if (!has_target()) {
639                 i += sprintf(buf, "performance powersave");
640                 goto out;
641         }
642
643         for_each_governor(t) {
644                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
645                     - (CPUFREQ_NAME_LEN + 2)))
646                         goto out;
647                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
648         }
649 out:
650         i += sprintf(&buf[i], "\n");
651         return i;
652 }
653
654 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
655 {
656         ssize_t i = 0;
657         unsigned int cpu;
658
659         for_each_cpu(cpu, mask) {
660                 if (i)
661                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
662                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
663                 if (i >= (PAGE_SIZE - 5))
664                         break;
665         }
666         i += sprintf(&buf[i], "\n");
667         return i;
668 }
669 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
670
671 /**
672  * show_related_cpus - show the CPUs affected by each transition even if
673  * hw coordination is in use
674  */
675 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
676 {
677         return cpufreq_show_cpus(policy->related_cpus, buf);
678 }
679
680 /**
681  * show_affected_cpus - show the CPUs affected by each transition
682  */
683 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
684 {
685         return cpufreq_show_cpus(policy->cpus, buf);
686 }
687
688 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
689                                         const char *buf, size_t count)
690 {
691         unsigned int freq = 0;
692         unsigned int ret;
693
694         if (!policy->governor || !policy->governor->store_setspeed)
695                 return -EINVAL;
696
697         ret = sscanf(buf, "%u", &freq);
698         if (ret != 1)
699                 return -EINVAL;
700
701         policy->governor->store_setspeed(policy, freq);
702
703         return count;
704 }
705
706 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
707 {
708         if (!policy->governor || !policy->governor->show_setspeed)
709                 return sprintf(buf, "<unsupported>\n");
710
711         return policy->governor->show_setspeed(policy, buf);
712 }
713
714 /**
715  * show_bios_limit - show the current cpufreq HW/BIOS limitation
716  */
717 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
718 {
719         unsigned int limit;
720         int ret;
721         if (cpufreq_driver->bios_limit) {
722                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
723                 if (!ret)
724                         return sprintf(buf, "%u\n", limit);
725         }
726         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
727 }
728
729 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
730 cpufreq_freq_attr_ro(cpuinfo_min_freq);
731 cpufreq_freq_attr_ro(cpuinfo_max_freq);
732 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
733 cpufreq_freq_attr_ro(scaling_available_governors);
734 cpufreq_freq_attr_ro(scaling_driver);
735 cpufreq_freq_attr_ro(scaling_cur_freq);
736 cpufreq_freq_attr_ro(bios_limit);
737 cpufreq_freq_attr_ro(related_cpus);
738 cpufreq_freq_attr_ro(affected_cpus);
739 cpufreq_freq_attr_rw(scaling_min_freq);
740 cpufreq_freq_attr_rw(scaling_max_freq);
741 cpufreq_freq_attr_rw(scaling_governor);
742 cpufreq_freq_attr_rw(scaling_setspeed);
743
744 static struct attribute *default_attrs[] = {
745         &cpuinfo_min_freq.attr,
746         &cpuinfo_max_freq.attr,
747         &cpuinfo_transition_latency.attr,
748         &scaling_min_freq.attr,
749         &scaling_max_freq.attr,
750         &affected_cpus.attr,
751         &related_cpus.attr,
752         &scaling_governor.attr,
753         &scaling_driver.attr,
754         &scaling_available_governors.attr,
755         &scaling_setspeed.attr,
756         NULL
757 };
758
759 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
760 #define to_attr(a) container_of(a, struct freq_attr, attr)
761
762 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
763 {
764         struct cpufreq_policy *policy = to_policy(kobj);
765         struct freq_attr *fattr = to_attr(attr);
766         ssize_t ret;
767
768         if (!down_read_trylock(&cpufreq_rwsem))
769                 return -EINVAL;
770
771         down_read(&policy->rwsem);
772
773         if (fattr->show)
774                 ret = fattr->show(policy, buf);
775         else
776                 ret = -EIO;
777
778         up_read(&policy->rwsem);
779         up_read(&cpufreq_rwsem);
780
781         return ret;
782 }
783
784 static ssize_t store(struct kobject *kobj, struct attribute *attr,
785                      const char *buf, size_t count)
786 {
787         struct cpufreq_policy *policy = to_policy(kobj);
788         struct freq_attr *fattr = to_attr(attr);
789         ssize_t ret = -EINVAL;
790
791         get_online_cpus();
792
793         if (!cpu_online(policy->cpu))
794                 goto unlock;
795
796         if (!down_read_trylock(&cpufreq_rwsem))
797                 goto unlock;
798
799         down_write(&policy->rwsem);
800
801         if (fattr->store)
802                 ret = fattr->store(policy, buf, count);
803         else
804                 ret = -EIO;
805
806         up_write(&policy->rwsem);
807
808         up_read(&cpufreq_rwsem);
809 unlock:
810         put_online_cpus();
811
812         return ret;
813 }
814
815 static void cpufreq_sysfs_release(struct kobject *kobj)
816 {
817         struct cpufreq_policy *policy = to_policy(kobj);
818         pr_debug("last reference is dropped\n");
819         complete(&policy->kobj_unregister);
820 }
821
822 static const struct sysfs_ops sysfs_ops = {
823         .show   = show,
824         .store  = store,
825 };
826
827 static struct kobj_type ktype_cpufreq = {
828         .sysfs_ops      = &sysfs_ops,
829         .default_attrs  = default_attrs,
830         .release        = cpufreq_sysfs_release,
831 };
832
833 struct kobject *cpufreq_global_kobject;
834 EXPORT_SYMBOL(cpufreq_global_kobject);
835
836 static int cpufreq_global_kobject_usage;
837
838 int cpufreq_get_global_kobject(void)
839 {
840         if (!cpufreq_global_kobject_usage++)
841                 return kobject_add(cpufreq_global_kobject,
842                                 &cpu_subsys.dev_root->kobj, "%s", "cpufreq");
843
844         return 0;
845 }
846 EXPORT_SYMBOL(cpufreq_get_global_kobject);
847
848 void cpufreq_put_global_kobject(void)
849 {
850         if (!--cpufreq_global_kobject_usage)
851                 kobject_del(cpufreq_global_kobject);
852 }
853 EXPORT_SYMBOL(cpufreq_put_global_kobject);
854
855 int cpufreq_sysfs_create_file(const struct attribute *attr)
856 {
857         int ret = cpufreq_get_global_kobject();
858
859         if (!ret) {
860                 ret = sysfs_create_file(cpufreq_global_kobject, attr);
861                 if (ret)
862                         cpufreq_put_global_kobject();
863         }
864
865         return ret;
866 }
867 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
868
869 void cpufreq_sysfs_remove_file(const struct attribute *attr)
870 {
871         sysfs_remove_file(cpufreq_global_kobject, attr);
872         cpufreq_put_global_kobject();
873 }
874 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
875
876 /* symlink affected CPUs */
877 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
878 {
879         unsigned int j;
880         int ret = 0;
881
882         for_each_cpu(j, policy->cpus) {
883                 struct device *cpu_dev;
884
885                 if (j == policy->cpu)
886                         continue;
887
888                 pr_debug("Adding link for CPU: %u\n", j);
889                 cpu_dev = get_cpu_device(j);
890                 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
891                                         "cpufreq");
892                 if (ret)
893                         break;
894         }
895         return ret;
896 }
897
898 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy,
899                                      struct device *dev)
900 {
901         struct freq_attr **drv_attr;
902         int ret = 0;
903
904         /* set up files for this cpu device */
905         drv_attr = cpufreq_driver->attr;
906         while (drv_attr && *drv_attr) {
907                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
908                 if (ret)
909                         return ret;
910                 drv_attr++;
911         }
912         if (cpufreq_driver->get) {
913                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
914                 if (ret)
915                         return ret;
916         }
917
918         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
919         if (ret)
920                 return ret;
921
922         if (cpufreq_driver->bios_limit) {
923                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
924                 if (ret)
925                         return ret;
926         }
927
928         return cpufreq_add_dev_symlink(policy);
929 }
930
931 static void cpufreq_init_policy(struct cpufreq_policy *policy)
932 {
933         struct cpufreq_governor *gov = NULL;
934         struct cpufreq_policy new_policy;
935         int ret = 0;
936
937         memcpy(&new_policy, policy, sizeof(*policy));
938
939         /* Update governor of new_policy to the governor used before hotplug */
940         gov = find_governor(per_cpu(cpufreq_cpu_governor, policy->cpu));
941         if (gov)
942                 pr_debug("Restoring governor %s for cpu %d\n",
943                                 policy->governor->name, policy->cpu);
944         else
945                 gov = CPUFREQ_DEFAULT_GOVERNOR;
946
947         new_policy.governor = gov;
948
949         /* Use the default policy if its valid. */
950         if (cpufreq_driver->setpolicy)
951                 cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
952
953         /* set default policy */
954         ret = cpufreq_set_policy(policy, &new_policy);
955         if (ret) {
956                 pr_debug("setting policy failed\n");
957                 if (cpufreq_driver->exit)
958                         cpufreq_driver->exit(policy);
959         }
960 }
961
962 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy,
963                                   unsigned int cpu, struct device *dev)
964 {
965         int ret = 0;
966         unsigned long flags;
967
968         if (has_target()) {
969                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
970                 if (ret) {
971                         pr_err("%s: Failed to stop governor\n", __func__);
972                         return ret;
973                 }
974         }
975
976         down_write(&policy->rwsem);
977
978         write_lock_irqsave(&cpufreq_driver_lock, flags);
979
980         cpumask_set_cpu(cpu, policy->cpus);
981         per_cpu(cpufreq_cpu_data, cpu) = policy;
982         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
983
984         up_write(&policy->rwsem);
985
986         if (has_target()) {
987                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
988                 if (!ret)
989                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
990
991                 if (ret) {
992                         pr_err("%s: Failed to start governor\n", __func__);
993                         return ret;
994                 }
995         }
996
997         return sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
998 }
999
1000 static struct cpufreq_policy *cpufreq_policy_restore(unsigned int cpu)
1001 {
1002         struct cpufreq_policy *policy;
1003         unsigned long flags;
1004
1005         read_lock_irqsave(&cpufreq_driver_lock, flags);
1006
1007         policy = per_cpu(cpufreq_cpu_data_fallback, cpu);
1008
1009         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1010
1011         if (policy)
1012                 policy->governor = NULL;
1013
1014         return policy;
1015 }
1016
1017 static struct cpufreq_policy *cpufreq_policy_alloc(void)
1018 {
1019         struct cpufreq_policy *policy;
1020
1021         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1022         if (!policy)
1023                 return NULL;
1024
1025         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1026                 goto err_free_policy;
1027
1028         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1029                 goto err_free_cpumask;
1030
1031         INIT_LIST_HEAD(&policy->policy_list);
1032         init_rwsem(&policy->rwsem);
1033         spin_lock_init(&policy->transition_lock);
1034         init_waitqueue_head(&policy->transition_wait);
1035         init_completion(&policy->kobj_unregister);
1036         INIT_WORK(&policy->update, handle_update);
1037
1038         return policy;
1039
1040 err_free_cpumask:
1041         free_cpumask_var(policy->cpus);
1042 err_free_policy:
1043         kfree(policy);
1044
1045         return NULL;
1046 }
1047
1048 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1049 {
1050         struct kobject *kobj;
1051         struct completion *cmp;
1052
1053         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1054                         CPUFREQ_REMOVE_POLICY, policy);
1055
1056         down_read(&policy->rwsem);
1057         kobj = &policy->kobj;
1058         cmp = &policy->kobj_unregister;
1059         up_read(&policy->rwsem);
1060         kobject_put(kobj);
1061
1062         /*
1063          * We need to make sure that the underlying kobj is
1064          * actually not referenced anymore by anybody before we
1065          * proceed with unloading.
1066          */
1067         pr_debug("waiting for dropping of refcount\n");
1068         wait_for_completion(cmp);
1069         pr_debug("wait complete\n");
1070 }
1071
1072 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1073 {
1074         free_cpumask_var(policy->related_cpus);
1075         free_cpumask_var(policy->cpus);
1076         kfree(policy);
1077 }
1078
1079 static int update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu,
1080                              struct device *cpu_dev)
1081 {
1082         int ret;
1083
1084         if (WARN_ON(cpu == policy->cpu))
1085                 return 0;
1086
1087         /* Move kobject to the new policy->cpu */
1088         ret = kobject_move(&policy->kobj, &cpu_dev->kobj);
1089         if (ret) {
1090                 pr_err("%s: Failed to move kobj: %d\n", __func__, ret);
1091                 return ret;
1092         }
1093
1094         down_write(&policy->rwsem);
1095         policy->cpu = cpu;
1096         up_write(&policy->rwsem);
1097
1098         return 0;
1099 }
1100
1101 static int __cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1102 {
1103         unsigned int j, cpu = dev->id;
1104         int ret = -ENOMEM;
1105         struct cpufreq_policy *policy;
1106         unsigned long flags;
1107         bool recover_policy = cpufreq_suspended;
1108
1109         if (cpu_is_offline(cpu))
1110                 return 0;
1111
1112         pr_debug("adding CPU %u\n", cpu);
1113
1114         /* check whether a different CPU already registered this
1115          * CPU because it is in the same boat. */
1116         policy = cpufreq_cpu_get_raw(cpu);
1117         if (unlikely(policy))
1118                 return 0;
1119
1120         if (!down_read_trylock(&cpufreq_rwsem))
1121                 return 0;
1122
1123         /* Check if this cpu was hot-unplugged earlier and has siblings */
1124         read_lock_irqsave(&cpufreq_driver_lock, flags);
1125         for_each_policy(policy) {
1126                 if (cpumask_test_cpu(cpu, policy->related_cpus)) {
1127                         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1128                         ret = cpufreq_add_policy_cpu(policy, cpu, dev);
1129                         up_read(&cpufreq_rwsem);
1130                         return ret;
1131                 }
1132         }
1133         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1134
1135         /*
1136          * Restore the saved policy when doing light-weight init and fall back
1137          * to the full init if that fails.
1138          */
1139         policy = recover_policy ? cpufreq_policy_restore(cpu) : NULL;
1140         if (!policy) {
1141                 recover_policy = false;
1142                 policy = cpufreq_policy_alloc();
1143                 if (!policy)
1144                         goto nomem_out;
1145         }
1146
1147         /*
1148          * In the resume path, since we restore a saved policy, the assignment
1149          * to policy->cpu is like an update of the existing policy, rather than
1150          * the creation of a brand new one. So we need to perform this update
1151          * by invoking update_policy_cpu().
1152          */
1153         if (recover_policy && cpu != policy->cpu)
1154                 WARN_ON(update_policy_cpu(policy, cpu, dev));
1155         else
1156                 policy->cpu = cpu;
1157
1158         cpumask_copy(policy->cpus, cpumask_of(cpu));
1159
1160         /* call driver. From then on the cpufreq must be able
1161          * to accept all calls to ->verify and ->setpolicy for this CPU
1162          */
1163         ret = cpufreq_driver->init(policy);
1164         if (ret) {
1165                 pr_debug("initialization failed\n");
1166                 goto err_set_policy_cpu;
1167         }
1168
1169         down_write(&policy->rwsem);
1170
1171         /* related cpus should atleast have policy->cpus */
1172         cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1173
1174         /*
1175          * affected cpus must always be the one, which are online. We aren't
1176          * managing offline cpus here.
1177          */
1178         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1179
1180         if (!recover_policy) {
1181                 policy->user_policy.min = policy->min;
1182                 policy->user_policy.max = policy->max;
1183
1184                 /* prepare interface data */
1185                 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1186                                            &dev->kobj, "cpufreq");
1187                 if (ret) {
1188                         pr_err("%s: failed to init policy->kobj: %d\n",
1189                                __func__, ret);
1190                         goto err_init_policy_kobj;
1191                 }
1192         }
1193
1194         write_lock_irqsave(&cpufreq_driver_lock, flags);
1195         for_each_cpu(j, policy->cpus)
1196                 per_cpu(cpufreq_cpu_data, j) = policy;
1197         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1198
1199         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1200                 policy->cur = cpufreq_driver->get(policy->cpu);
1201                 if (!policy->cur) {
1202                         pr_err("%s: ->get() failed\n", __func__);
1203                         goto err_get_freq;
1204                 }
1205         }
1206
1207         /*
1208          * Sometimes boot loaders set CPU frequency to a value outside of
1209          * frequency table present with cpufreq core. In such cases CPU might be
1210          * unstable if it has to run on that frequency for long duration of time
1211          * and so its better to set it to a frequency which is specified in
1212          * freq-table. This also makes cpufreq stats inconsistent as
1213          * cpufreq-stats would fail to register because current frequency of CPU
1214          * isn't found in freq-table.
1215          *
1216          * Because we don't want this change to effect boot process badly, we go
1217          * for the next freq which is >= policy->cur ('cur' must be set by now,
1218          * otherwise we will end up setting freq to lowest of the table as 'cur'
1219          * is initialized to zero).
1220          *
1221          * We are passing target-freq as "policy->cur - 1" otherwise
1222          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1223          * equal to target-freq.
1224          */
1225         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1226             && has_target()) {
1227                 /* Are we running at unknown frequency ? */
1228                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1229                 if (ret == -EINVAL) {
1230                         /* Warn user and fix it */
1231                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1232                                 __func__, policy->cpu, policy->cur);
1233                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1234                                 CPUFREQ_RELATION_L);
1235
1236                         /*
1237                          * Reaching here after boot in a few seconds may not
1238                          * mean that system will remain stable at "unknown"
1239                          * frequency for longer duration. Hence, a BUG_ON().
1240                          */
1241                         BUG_ON(ret);
1242                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1243                                 __func__, policy->cpu, policy->cur);
1244                 }
1245         }
1246
1247         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1248                                      CPUFREQ_START, policy);
1249
1250         if (!recover_policy) {
1251                 ret = cpufreq_add_dev_interface(policy, dev);
1252                 if (ret)
1253                         goto err_out_unregister;
1254                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1255                                 CPUFREQ_CREATE_POLICY, policy);
1256         }
1257
1258         write_lock_irqsave(&cpufreq_driver_lock, flags);
1259         list_add(&policy->policy_list, &cpufreq_policy_list);
1260         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1261
1262         cpufreq_init_policy(policy);
1263
1264         if (!recover_policy) {
1265                 policy->user_policy.policy = policy->policy;
1266                 policy->user_policy.governor = policy->governor;
1267         }
1268         up_write(&policy->rwsem);
1269
1270         kobject_uevent(&policy->kobj, KOBJ_ADD);
1271
1272         up_read(&cpufreq_rwsem);
1273
1274         /* Callback for handling stuff after policy is ready */
1275         if (cpufreq_driver->ready)
1276                 cpufreq_driver->ready(policy);
1277
1278         pr_debug("initialization complete\n");
1279
1280         return 0;
1281
1282 err_out_unregister:
1283 err_get_freq:
1284         write_lock_irqsave(&cpufreq_driver_lock, flags);
1285         for_each_cpu(j, policy->cpus)
1286                 per_cpu(cpufreq_cpu_data, j) = NULL;
1287         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1288
1289         if (!recover_policy) {
1290                 kobject_put(&policy->kobj);
1291                 wait_for_completion(&policy->kobj_unregister);
1292         }
1293 err_init_policy_kobj:
1294         up_write(&policy->rwsem);
1295
1296         if (cpufreq_driver->exit)
1297                 cpufreq_driver->exit(policy);
1298 err_set_policy_cpu:
1299         if (recover_policy) {
1300                 /* Do not leave stale fallback data behind. */
1301                 per_cpu(cpufreq_cpu_data_fallback, cpu) = NULL;
1302                 cpufreq_policy_put_kobj(policy);
1303         }
1304         cpufreq_policy_free(policy);
1305
1306 nomem_out:
1307         up_read(&cpufreq_rwsem);
1308
1309         return ret;
1310 }
1311
1312 /**
1313  * cpufreq_add_dev - add a CPU device
1314  *
1315  * Adds the cpufreq interface for a CPU device.
1316  *
1317  * The Oracle says: try running cpufreq registration/unregistration concurrently
1318  * with with cpu hotplugging and all hell will break loose. Tried to clean this
1319  * mess up, but more thorough testing is needed. - Mathieu
1320  */
1321 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1322 {
1323         return __cpufreq_add_dev(dev, sif);
1324 }
1325
1326 static int __cpufreq_remove_dev_prepare(struct device *dev,
1327                                         struct subsys_interface *sif)
1328 {
1329         unsigned int cpu = dev->id, cpus;
1330         int ret;
1331         unsigned long flags;
1332         struct cpufreq_policy *policy;
1333
1334         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1335
1336         write_lock_irqsave(&cpufreq_driver_lock, flags);
1337
1338         policy = per_cpu(cpufreq_cpu_data, cpu);
1339
1340         /* Save the policy somewhere when doing a light-weight tear-down */
1341         if (cpufreq_suspended)
1342                 per_cpu(cpufreq_cpu_data_fallback, cpu) = policy;
1343
1344         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1345
1346         if (!policy) {
1347                 pr_debug("%s: No cpu_data found\n", __func__);
1348                 return -EINVAL;
1349         }
1350
1351         if (has_target()) {
1352                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1353                 if (ret) {
1354                         pr_err("%s: Failed to stop governor\n", __func__);
1355                         return ret;
1356                 }
1357
1358                 strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1359                         policy->governor->name, CPUFREQ_NAME_LEN);
1360         }
1361
1362         down_read(&policy->rwsem);
1363         cpus = cpumask_weight(policy->cpus);
1364         up_read(&policy->rwsem);
1365
1366         if (cpu != policy->cpu) {
1367                 sysfs_remove_link(&dev->kobj, "cpufreq");
1368         } else if (cpus > 1) {
1369                 /* Nominate new CPU */
1370                 int new_cpu = cpumask_any_but(policy->cpus, cpu);
1371                 struct device *cpu_dev = get_cpu_device(new_cpu);
1372
1373                 sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1374                 ret = update_policy_cpu(policy, new_cpu, cpu_dev);
1375                 if (ret) {
1376                         if (sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
1377                                               "cpufreq"))
1378                                 pr_err("%s: Failed to restore kobj link to cpu:%d\n",
1379                                        __func__, cpu_dev->id);
1380                         return ret;
1381                 }
1382
1383                 if (!cpufreq_suspended)
1384                         pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1385                                  __func__, new_cpu, cpu);
1386         } else if (cpufreq_driver->stop_cpu) {
1387                 cpufreq_driver->stop_cpu(policy);
1388         }
1389
1390         return 0;
1391 }
1392
1393 static int __cpufreq_remove_dev_finish(struct device *dev,
1394                                        struct subsys_interface *sif)
1395 {
1396         unsigned int cpu = dev->id, cpus;
1397         int ret;
1398         unsigned long flags;
1399         struct cpufreq_policy *policy;
1400
1401         write_lock_irqsave(&cpufreq_driver_lock, flags);
1402         policy = per_cpu(cpufreq_cpu_data, cpu);
1403         per_cpu(cpufreq_cpu_data, cpu) = NULL;
1404         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1405
1406         if (!policy) {
1407                 pr_debug("%s: No cpu_data found\n", __func__);
1408                 return -EINVAL;
1409         }
1410
1411         down_write(&policy->rwsem);
1412         cpus = cpumask_weight(policy->cpus);
1413
1414         if (cpus > 1)
1415                 cpumask_clear_cpu(cpu, policy->cpus);
1416         up_write(&policy->rwsem);
1417
1418         /* If cpu is last user of policy, free policy */
1419         if (cpus == 1) {
1420                 if (has_target()) {
1421                         ret = __cpufreq_governor(policy,
1422                                         CPUFREQ_GOV_POLICY_EXIT);
1423                         if (ret) {
1424                                 pr_err("%s: Failed to exit governor\n",
1425                                        __func__);
1426                                 return ret;
1427                         }
1428                 }
1429
1430                 if (!cpufreq_suspended)
1431                         cpufreq_policy_put_kobj(policy);
1432
1433                 /*
1434                  * Perform the ->exit() even during light-weight tear-down,
1435                  * since this is a core component, and is essential for the
1436                  * subsequent light-weight ->init() to succeed.
1437                  */
1438                 if (cpufreq_driver->exit)
1439                         cpufreq_driver->exit(policy);
1440
1441                 /* Remove policy from list of active policies */
1442                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1443                 list_del(&policy->policy_list);
1444                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1445
1446                 if (!cpufreq_suspended)
1447                         cpufreq_policy_free(policy);
1448         } else if (has_target()) {
1449                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1450                 if (!ret)
1451                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1452
1453                 if (ret) {
1454                         pr_err("%s: Failed to start governor\n", __func__);
1455                         return ret;
1456                 }
1457         }
1458
1459         return 0;
1460 }
1461
1462 /**
1463  * cpufreq_remove_dev - remove a CPU device
1464  *
1465  * Removes the cpufreq interface for a CPU device.
1466  */
1467 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1468 {
1469         unsigned int cpu = dev->id;
1470         int ret;
1471
1472         if (cpu_is_offline(cpu))
1473                 return 0;
1474
1475         ret = __cpufreq_remove_dev_prepare(dev, sif);
1476
1477         if (!ret)
1478                 ret = __cpufreq_remove_dev_finish(dev, sif);
1479
1480         return ret;
1481 }
1482
1483 static void handle_update(struct work_struct *work)
1484 {
1485         struct cpufreq_policy *policy =
1486                 container_of(work, struct cpufreq_policy, update);
1487         unsigned int cpu = policy->cpu;
1488         pr_debug("handle_update for cpu %u called\n", cpu);
1489         cpufreq_update_policy(cpu);
1490 }
1491
1492 /**
1493  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1494  *      in deep trouble.
1495  *      @policy: policy managing CPUs
1496  *      @new_freq: CPU frequency the CPU actually runs at
1497  *
1498  *      We adjust to current frequency first, and need to clean up later.
1499  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1500  */
1501 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1502                                 unsigned int new_freq)
1503 {
1504         struct cpufreq_freqs freqs;
1505
1506         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1507                  policy->cur, new_freq);
1508
1509         freqs.old = policy->cur;
1510         freqs.new = new_freq;
1511
1512         cpufreq_freq_transition_begin(policy, &freqs);
1513         cpufreq_freq_transition_end(policy, &freqs, 0);
1514 }
1515
1516 /**
1517  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1518  * @cpu: CPU number
1519  *
1520  * This is the last known freq, without actually getting it from the driver.
1521  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1522  */
1523 unsigned int cpufreq_quick_get(unsigned int cpu)
1524 {
1525         struct cpufreq_policy *policy;
1526         unsigned int ret_freq = 0;
1527
1528         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1529                 return cpufreq_driver->get(cpu);
1530
1531         policy = cpufreq_cpu_get(cpu);
1532         if (policy) {
1533                 ret_freq = policy->cur;
1534                 cpufreq_cpu_put(policy);
1535         }
1536
1537         return ret_freq;
1538 }
1539 EXPORT_SYMBOL(cpufreq_quick_get);
1540
1541 /**
1542  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1543  * @cpu: CPU number
1544  *
1545  * Just return the max possible frequency for a given CPU.
1546  */
1547 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1548 {
1549         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1550         unsigned int ret_freq = 0;
1551
1552         if (policy) {
1553                 ret_freq = policy->max;
1554                 cpufreq_cpu_put(policy);
1555         }
1556
1557         return ret_freq;
1558 }
1559 EXPORT_SYMBOL(cpufreq_quick_get_max);
1560
1561 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1562 {
1563         unsigned int ret_freq = 0;
1564
1565         if (!cpufreq_driver->get)
1566                 return ret_freq;
1567
1568         ret_freq = cpufreq_driver->get(policy->cpu);
1569
1570         if (ret_freq && policy->cur &&
1571                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1572                 /* verify no discrepancy between actual and
1573                                         saved value exists */
1574                 if (unlikely(ret_freq != policy->cur)) {
1575                         cpufreq_out_of_sync(policy, ret_freq);
1576                         schedule_work(&policy->update);
1577                 }
1578         }
1579
1580         return ret_freq;
1581 }
1582
1583 /**
1584  * cpufreq_get - get the current CPU frequency (in kHz)
1585  * @cpu: CPU number
1586  *
1587  * Get the CPU current (static) CPU frequency
1588  */
1589 unsigned int cpufreq_get(unsigned int cpu)
1590 {
1591         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1592         unsigned int ret_freq = 0;
1593
1594         if (policy) {
1595                 down_read(&policy->rwsem);
1596                 ret_freq = __cpufreq_get(policy);
1597                 up_read(&policy->rwsem);
1598
1599                 cpufreq_cpu_put(policy);
1600         }
1601
1602         return ret_freq;
1603 }
1604 EXPORT_SYMBOL(cpufreq_get);
1605
1606 static struct subsys_interface cpufreq_interface = {
1607         .name           = "cpufreq",
1608         .subsys         = &cpu_subsys,
1609         .add_dev        = cpufreq_add_dev,
1610         .remove_dev     = cpufreq_remove_dev,
1611 };
1612
1613 /*
1614  * In case platform wants some specific frequency to be configured
1615  * during suspend..
1616  */
1617 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1618 {
1619         int ret;
1620
1621         if (!policy->suspend_freq) {
1622                 pr_err("%s: suspend_freq can't be zero\n", __func__);
1623                 return -EINVAL;
1624         }
1625
1626         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1627                         policy->suspend_freq);
1628
1629         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1630                         CPUFREQ_RELATION_H);
1631         if (ret)
1632                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1633                                 __func__, policy->suspend_freq, ret);
1634
1635         return ret;
1636 }
1637 EXPORT_SYMBOL(cpufreq_generic_suspend);
1638
1639 /**
1640  * cpufreq_suspend() - Suspend CPUFreq governors
1641  *
1642  * Called during system wide Suspend/Hibernate cycles for suspending governors
1643  * as some platforms can't change frequency after this point in suspend cycle.
1644  * Because some of the devices (like: i2c, regulators, etc) they use for
1645  * changing frequency are suspended quickly after this point.
1646  */
1647 void cpufreq_suspend(void)
1648 {
1649         struct cpufreq_policy *policy;
1650
1651         if (!cpufreq_driver)
1652                 return;
1653
1654         if (!has_target())
1655                 goto suspend;
1656
1657         pr_debug("%s: Suspending Governors\n", __func__);
1658
1659         for_each_policy(policy) {
1660                 if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1661                         pr_err("%s: Failed to stop governor for policy: %p\n",
1662                                 __func__, policy);
1663                 else if (cpufreq_driver->suspend
1664                     && cpufreq_driver->suspend(policy))
1665                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1666                                 policy);
1667         }
1668
1669 suspend:
1670         cpufreq_suspended = true;
1671 }
1672
1673 /**
1674  * cpufreq_resume() - Resume CPUFreq governors
1675  *
1676  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1677  * are suspended with cpufreq_suspend().
1678  */
1679 void cpufreq_resume(void)
1680 {
1681         struct cpufreq_policy *policy;
1682
1683         if (!cpufreq_driver)
1684                 return;
1685
1686         cpufreq_suspended = false;
1687
1688         if (!has_target())
1689                 return;
1690
1691         pr_debug("%s: Resuming Governors\n", __func__);
1692
1693         for_each_policy(policy) {
1694                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1695                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1696                                 policy);
1697                 else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1698                     || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1699                         pr_err("%s: Failed to start governor for policy: %p\n",
1700                                 __func__, policy);
1701
1702                 /*
1703                  * schedule call cpufreq_update_policy() for boot CPU, i.e. last
1704                  * policy in list. It will verify that the current freq is in
1705                  * sync with what we believe it to be.
1706                  */
1707                 if (list_is_last(&policy->policy_list, &cpufreq_policy_list))
1708                         schedule_work(&policy->update);
1709         }
1710 }
1711
1712 /**
1713  *      cpufreq_get_current_driver - return current driver's name
1714  *
1715  *      Return the name string of the currently loaded cpufreq driver
1716  *      or NULL, if none.
1717  */
1718 const char *cpufreq_get_current_driver(void)
1719 {
1720         if (cpufreq_driver)
1721                 return cpufreq_driver->name;
1722
1723         return NULL;
1724 }
1725 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1726
1727 /**
1728  *      cpufreq_get_driver_data - return current driver data
1729  *
1730  *      Return the private data of the currently loaded cpufreq
1731  *      driver, or NULL if no cpufreq driver is loaded.
1732  */
1733 void *cpufreq_get_driver_data(void)
1734 {
1735         if (cpufreq_driver)
1736                 return cpufreq_driver->driver_data;
1737
1738         return NULL;
1739 }
1740 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1741
1742 /*********************************************************************
1743  *                     NOTIFIER LISTS INTERFACE                      *
1744  *********************************************************************/
1745
1746 /**
1747  *      cpufreq_register_notifier - register a driver with cpufreq
1748  *      @nb: notifier function to register
1749  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1750  *
1751  *      Add a driver to one of two lists: either a list of drivers that
1752  *      are notified about clock rate changes (once before and once after
1753  *      the transition), or a list of drivers that are notified about
1754  *      changes in cpufreq policy.
1755  *
1756  *      This function may sleep, and has the same return conditions as
1757  *      blocking_notifier_chain_register.
1758  */
1759 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1760 {
1761         int ret;
1762
1763         if (cpufreq_disabled())
1764                 return -EINVAL;
1765
1766         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1767
1768         switch (list) {
1769         case CPUFREQ_TRANSITION_NOTIFIER:
1770                 ret = srcu_notifier_chain_register(
1771                                 &cpufreq_transition_notifier_list, nb);
1772                 break;
1773         case CPUFREQ_POLICY_NOTIFIER:
1774                 ret = blocking_notifier_chain_register(
1775                                 &cpufreq_policy_notifier_list, nb);
1776                 break;
1777         default:
1778                 ret = -EINVAL;
1779         }
1780
1781         return ret;
1782 }
1783 EXPORT_SYMBOL(cpufreq_register_notifier);
1784
1785 /**
1786  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1787  *      @nb: notifier block to be unregistered
1788  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1789  *
1790  *      Remove a driver from the CPU frequency notifier list.
1791  *
1792  *      This function may sleep, and has the same return conditions as
1793  *      blocking_notifier_chain_unregister.
1794  */
1795 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1796 {
1797         int ret;
1798
1799         if (cpufreq_disabled())
1800                 return -EINVAL;
1801
1802         switch (list) {
1803         case CPUFREQ_TRANSITION_NOTIFIER:
1804                 ret = srcu_notifier_chain_unregister(
1805                                 &cpufreq_transition_notifier_list, nb);
1806                 break;
1807         case CPUFREQ_POLICY_NOTIFIER:
1808                 ret = blocking_notifier_chain_unregister(
1809                                 &cpufreq_policy_notifier_list, nb);
1810                 break;
1811         default:
1812                 ret = -EINVAL;
1813         }
1814
1815         return ret;
1816 }
1817 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1818
1819
1820 /*********************************************************************
1821  *                              GOVERNORS                            *
1822  *********************************************************************/
1823
1824 /* Must set freqs->new to intermediate frequency */
1825 static int __target_intermediate(struct cpufreq_policy *policy,
1826                                  struct cpufreq_freqs *freqs, int index)
1827 {
1828         int ret;
1829
1830         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1831
1832         /* We don't need to switch to intermediate freq */
1833         if (!freqs->new)
1834                 return 0;
1835
1836         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1837                  __func__, policy->cpu, freqs->old, freqs->new);
1838
1839         cpufreq_freq_transition_begin(policy, freqs);
1840         ret = cpufreq_driver->target_intermediate(policy, index);
1841         cpufreq_freq_transition_end(policy, freqs, ret);
1842
1843         if (ret)
1844                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1845                        __func__, ret);
1846
1847         return ret;
1848 }
1849
1850 static int __target_index(struct cpufreq_policy *policy,
1851                           struct cpufreq_frequency_table *freq_table, int index)
1852 {
1853         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1854         unsigned int intermediate_freq = 0;
1855         int retval = -EINVAL;
1856         bool notify;
1857
1858         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1859         if (notify) {
1860                 /* Handle switching to intermediate frequency */
1861                 if (cpufreq_driver->get_intermediate) {
1862                         retval = __target_intermediate(policy, &freqs, index);
1863                         if (retval)
1864                                 return retval;
1865
1866                         intermediate_freq = freqs.new;
1867                         /* Set old freq to intermediate */
1868                         if (intermediate_freq)
1869                                 freqs.old = freqs.new;
1870                 }
1871
1872                 freqs.new = freq_table[index].frequency;
1873                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1874                          __func__, policy->cpu, freqs.old, freqs.new);
1875
1876                 cpufreq_freq_transition_begin(policy, &freqs);
1877         }
1878
1879         retval = cpufreq_driver->target_index(policy, index);
1880         if (retval)
1881                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1882                        retval);
1883
1884         if (notify) {
1885                 cpufreq_freq_transition_end(policy, &freqs, retval);
1886
1887                 /*
1888                  * Failed after setting to intermediate freq? Driver should have
1889                  * reverted back to initial frequency and so should we. Check
1890                  * here for intermediate_freq instead of get_intermediate, in
1891                  * case we have't switched to intermediate freq at all.
1892                  */
1893                 if (unlikely(retval && intermediate_freq)) {
1894                         freqs.old = intermediate_freq;
1895                         freqs.new = policy->restore_freq;
1896                         cpufreq_freq_transition_begin(policy, &freqs);
1897                         cpufreq_freq_transition_end(policy, &freqs, 0);
1898                 }
1899         }
1900
1901         return retval;
1902 }
1903
1904 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1905                             unsigned int target_freq,
1906                             unsigned int relation)
1907 {
1908         unsigned int old_target_freq = target_freq;
1909         int retval = -EINVAL;
1910
1911         if (cpufreq_disabled())
1912                 return -ENODEV;
1913
1914         /* Make sure that target_freq is within supported range */
1915         if (target_freq > policy->max)
1916                 target_freq = policy->max;
1917         if (target_freq < policy->min)
1918                 target_freq = policy->min;
1919
1920         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1921                  policy->cpu, target_freq, relation, old_target_freq);
1922
1923         /*
1924          * This might look like a redundant call as we are checking it again
1925          * after finding index. But it is left intentionally for cases where
1926          * exactly same freq is called again and so we can save on few function
1927          * calls.
1928          */
1929         if (target_freq == policy->cur)
1930                 return 0;
1931
1932         /* Save last value to restore later on errors */
1933         policy->restore_freq = policy->cur;
1934
1935         if (cpufreq_driver->target)
1936                 retval = cpufreq_driver->target(policy, target_freq, relation);
1937         else if (cpufreq_driver->target_index) {
1938                 struct cpufreq_frequency_table *freq_table;
1939                 int index;
1940
1941                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1942                 if (unlikely(!freq_table)) {
1943                         pr_err("%s: Unable to find freq_table\n", __func__);
1944                         goto out;
1945                 }
1946
1947                 retval = cpufreq_frequency_table_target(policy, freq_table,
1948                                 target_freq, relation, &index);
1949                 if (unlikely(retval)) {
1950                         pr_err("%s: Unable to find matching freq\n", __func__);
1951                         goto out;
1952                 }
1953
1954                 if (freq_table[index].frequency == policy->cur) {
1955                         retval = 0;
1956                         goto out;
1957                 }
1958
1959                 retval = __target_index(policy, freq_table, index);
1960         }
1961
1962 out:
1963         return retval;
1964 }
1965 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1966
1967 int cpufreq_driver_target(struct cpufreq_policy *policy,
1968                           unsigned int target_freq,
1969                           unsigned int relation)
1970 {
1971         int ret = -EINVAL;
1972
1973         down_write(&policy->rwsem);
1974
1975         ret = __cpufreq_driver_target(policy, target_freq, relation);
1976
1977         up_write(&policy->rwsem);
1978
1979         return ret;
1980 }
1981 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1982
1983 static int __cpufreq_governor(struct cpufreq_policy *policy,
1984                                         unsigned int event)
1985 {
1986         int ret;
1987
1988         /* Only must be defined when default governor is known to have latency
1989            restrictions, like e.g. conservative or ondemand.
1990            That this is the case is already ensured in Kconfig
1991         */
1992 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1993         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1994 #else
1995         struct cpufreq_governor *gov = NULL;
1996 #endif
1997
1998         /* Don't start any governor operations if we are entering suspend */
1999         if (cpufreq_suspended)
2000                 return 0;
2001         /*
2002          * Governor might not be initiated here if ACPI _PPC changed
2003          * notification happened, so check it.
2004          */
2005         if (!policy->governor)
2006                 return -EINVAL;
2007
2008         if (policy->governor->max_transition_latency &&
2009             policy->cpuinfo.transition_latency >
2010             policy->governor->max_transition_latency) {
2011                 if (!gov)
2012                         return -EINVAL;
2013                 else {
2014                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2015                                 policy->governor->name, gov->name);
2016                         policy->governor = gov;
2017                 }
2018         }
2019
2020         if (event == CPUFREQ_GOV_POLICY_INIT)
2021                 if (!try_module_get(policy->governor->owner))
2022                         return -EINVAL;
2023
2024         pr_debug("__cpufreq_governor for CPU %u, event %u\n",
2025                  policy->cpu, event);
2026
2027         mutex_lock(&cpufreq_governor_lock);
2028         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2029             || (!policy->governor_enabled
2030             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2031                 mutex_unlock(&cpufreq_governor_lock);
2032                 return -EBUSY;
2033         }
2034
2035         if (event == CPUFREQ_GOV_STOP)
2036                 policy->governor_enabled = false;
2037         else if (event == CPUFREQ_GOV_START)
2038                 policy->governor_enabled = true;
2039
2040         mutex_unlock(&cpufreq_governor_lock);
2041
2042         ret = policy->governor->governor(policy, event);
2043
2044         if (!ret) {
2045                 if (event == CPUFREQ_GOV_POLICY_INIT)
2046                         policy->governor->initialized++;
2047                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
2048                         policy->governor->initialized--;
2049         } else {
2050                 /* Restore original values */
2051                 mutex_lock(&cpufreq_governor_lock);
2052                 if (event == CPUFREQ_GOV_STOP)
2053                         policy->governor_enabled = true;
2054                 else if (event == CPUFREQ_GOV_START)
2055                         policy->governor_enabled = false;
2056                 mutex_unlock(&cpufreq_governor_lock);
2057         }
2058
2059         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2060                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2061                 module_put(policy->governor->owner);
2062
2063         return ret;
2064 }
2065
2066 int cpufreq_register_governor(struct cpufreq_governor *governor)
2067 {
2068         int err;
2069
2070         if (!governor)
2071                 return -EINVAL;
2072
2073         if (cpufreq_disabled())
2074                 return -ENODEV;
2075
2076         mutex_lock(&cpufreq_governor_mutex);
2077
2078         governor->initialized = 0;
2079         err = -EBUSY;
2080         if (!find_governor(governor->name)) {
2081                 err = 0;
2082                 list_add(&governor->governor_list, &cpufreq_governor_list);
2083         }
2084
2085         mutex_unlock(&cpufreq_governor_mutex);
2086         return err;
2087 }
2088 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2089
2090 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2091 {
2092         int cpu;
2093
2094         if (!governor)
2095                 return;
2096
2097         if (cpufreq_disabled())
2098                 return;
2099
2100         for_each_present_cpu(cpu) {
2101                 if (cpu_online(cpu))
2102                         continue;
2103                 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
2104                         strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
2105         }
2106
2107         mutex_lock(&cpufreq_governor_mutex);
2108         list_del(&governor->governor_list);
2109         mutex_unlock(&cpufreq_governor_mutex);
2110         return;
2111 }
2112 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2113
2114
2115 /*********************************************************************
2116  *                          POLICY INTERFACE                         *
2117  *********************************************************************/
2118
2119 /**
2120  * cpufreq_get_policy - get the current cpufreq_policy
2121  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2122  *      is written
2123  *
2124  * Reads the current cpufreq policy.
2125  */
2126 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2127 {
2128         struct cpufreq_policy *cpu_policy;
2129         if (!policy)
2130                 return -EINVAL;
2131
2132         cpu_policy = cpufreq_cpu_get(cpu);
2133         if (!cpu_policy)
2134                 return -EINVAL;
2135
2136         memcpy(policy, cpu_policy, sizeof(*policy));
2137
2138         cpufreq_cpu_put(cpu_policy);
2139         return 0;
2140 }
2141 EXPORT_SYMBOL(cpufreq_get_policy);
2142
2143 /*
2144  * policy : current policy.
2145  * new_policy: policy to be set.
2146  */
2147 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2148                                 struct cpufreq_policy *new_policy)
2149 {
2150         struct cpufreq_governor *old_gov;
2151         int ret;
2152
2153         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2154                  new_policy->cpu, new_policy->min, new_policy->max);
2155
2156         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2157
2158         if (new_policy->min > policy->max || new_policy->max < policy->min)
2159                 return -EINVAL;
2160
2161         /* verify the cpu speed can be set within this limit */
2162         ret = cpufreq_driver->verify(new_policy);
2163         if (ret)
2164                 return ret;
2165
2166         /* adjust if necessary - all reasons */
2167         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2168                         CPUFREQ_ADJUST, new_policy);
2169
2170         /* adjust if necessary - hardware incompatibility*/
2171         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2172                         CPUFREQ_INCOMPATIBLE, new_policy);
2173
2174         /*
2175          * verify the cpu speed can be set within this limit, which might be
2176          * different to the first one
2177          */
2178         ret = cpufreq_driver->verify(new_policy);
2179         if (ret)
2180                 return ret;
2181
2182         /* notification of the new policy */
2183         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2184                         CPUFREQ_NOTIFY, new_policy);
2185
2186         policy->min = new_policy->min;
2187         policy->max = new_policy->max;
2188
2189         pr_debug("new min and max freqs are %u - %u kHz\n",
2190                  policy->min, policy->max);
2191
2192         if (cpufreq_driver->setpolicy) {
2193                 policy->policy = new_policy->policy;
2194                 pr_debug("setting range\n");
2195                 return cpufreq_driver->setpolicy(new_policy);
2196         }
2197
2198         if (new_policy->governor == policy->governor)
2199                 goto out;
2200
2201         pr_debug("governor switch\n");
2202
2203         /* save old, working values */
2204         old_gov = policy->governor;
2205         /* end old governor */
2206         if (old_gov) {
2207                 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2208                 up_write(&policy->rwsem);
2209                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2210                 down_write(&policy->rwsem);
2211         }
2212
2213         /* start new governor */
2214         policy->governor = new_policy->governor;
2215         if (!__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT)) {
2216                 if (!__cpufreq_governor(policy, CPUFREQ_GOV_START))
2217                         goto out;
2218
2219                 up_write(&policy->rwsem);
2220                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2221                 down_write(&policy->rwsem);
2222         }
2223
2224         /* new governor failed, so re-start old one */
2225         pr_debug("starting governor %s failed\n", policy->governor->name);
2226         if (old_gov) {
2227                 policy->governor = old_gov;
2228                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2229                 __cpufreq_governor(policy, CPUFREQ_GOV_START);
2230         }
2231
2232         return -EINVAL;
2233
2234  out:
2235         pr_debug("governor: change or update limits\n");
2236         return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2237 }
2238
2239 /**
2240  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2241  *      @cpu: CPU which shall be re-evaluated
2242  *
2243  *      Useful for policy notifiers which have different necessities
2244  *      at different times.
2245  */
2246 int cpufreq_update_policy(unsigned int cpu)
2247 {
2248         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2249         struct cpufreq_policy new_policy;
2250         int ret;
2251
2252         if (!policy)
2253                 return -ENODEV;
2254
2255         down_write(&policy->rwsem);
2256
2257         pr_debug("updating policy for CPU %u\n", cpu);
2258         memcpy(&new_policy, policy, sizeof(*policy));
2259         new_policy.min = policy->user_policy.min;
2260         new_policy.max = policy->user_policy.max;
2261         new_policy.policy = policy->user_policy.policy;
2262         new_policy.governor = policy->user_policy.governor;
2263
2264         /*
2265          * BIOS might change freq behind our back
2266          * -> ask driver for current freq and notify governors about a change
2267          */
2268         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2269                 new_policy.cur = cpufreq_driver->get(cpu);
2270                 if (WARN_ON(!new_policy.cur)) {
2271                         ret = -EIO;
2272                         goto unlock;
2273                 }
2274
2275                 if (!policy->cur) {
2276                         pr_debug("Driver did not initialize current freq\n");
2277                         policy->cur = new_policy.cur;
2278                 } else {
2279                         if (policy->cur != new_policy.cur && has_target())
2280                                 cpufreq_out_of_sync(policy, new_policy.cur);
2281                 }
2282         }
2283
2284         ret = cpufreq_set_policy(policy, &new_policy);
2285
2286 unlock:
2287         up_write(&policy->rwsem);
2288
2289         cpufreq_cpu_put(policy);
2290         return ret;
2291 }
2292 EXPORT_SYMBOL(cpufreq_update_policy);
2293
2294 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2295                                         unsigned long action, void *hcpu)
2296 {
2297         unsigned int cpu = (unsigned long)hcpu;
2298         struct device *dev;
2299
2300         dev = get_cpu_device(cpu);
2301         if (dev) {
2302                 switch (action & ~CPU_TASKS_FROZEN) {
2303                 case CPU_ONLINE:
2304                         __cpufreq_add_dev(dev, NULL);
2305                         break;
2306
2307                 case CPU_DOWN_PREPARE:
2308                         __cpufreq_remove_dev_prepare(dev, NULL);
2309                         break;
2310
2311                 case CPU_POST_DEAD:
2312                         __cpufreq_remove_dev_finish(dev, NULL);
2313                         break;
2314
2315                 case CPU_DOWN_FAILED:
2316                         __cpufreq_add_dev(dev, NULL);
2317                         break;
2318                 }
2319         }
2320         return NOTIFY_OK;
2321 }
2322
2323 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2324         .notifier_call = cpufreq_cpu_callback,
2325 };
2326
2327 /*********************************************************************
2328  *               BOOST                                               *
2329  *********************************************************************/
2330 static int cpufreq_boost_set_sw(int state)
2331 {
2332         struct cpufreq_frequency_table *freq_table;
2333         struct cpufreq_policy *policy;
2334         int ret = -EINVAL;
2335
2336         for_each_policy(policy) {
2337                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2338                 if (freq_table) {
2339                         ret = cpufreq_frequency_table_cpuinfo(policy,
2340                                                         freq_table);
2341                         if (ret) {
2342                                 pr_err("%s: Policy frequency update failed\n",
2343                                        __func__);
2344                                 break;
2345                         }
2346                         policy->user_policy.max = policy->max;
2347                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2348                 }
2349         }
2350
2351         return ret;
2352 }
2353
2354 int cpufreq_boost_trigger_state(int state)
2355 {
2356         unsigned long flags;
2357         int ret = 0;
2358
2359         if (cpufreq_driver->boost_enabled == state)
2360                 return 0;
2361
2362         write_lock_irqsave(&cpufreq_driver_lock, flags);
2363         cpufreq_driver->boost_enabled = state;
2364         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2365
2366         ret = cpufreq_driver->set_boost(state);
2367         if (ret) {
2368                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2369                 cpufreq_driver->boost_enabled = !state;
2370                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2371
2372                 pr_err("%s: Cannot %s BOOST\n",
2373                        __func__, state ? "enable" : "disable");
2374         }
2375
2376         return ret;
2377 }
2378
2379 int cpufreq_boost_supported(void)
2380 {
2381         if (likely(cpufreq_driver))
2382                 return cpufreq_driver->boost_supported;
2383
2384         return 0;
2385 }
2386 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2387
2388 int cpufreq_boost_enabled(void)
2389 {
2390         return cpufreq_driver->boost_enabled;
2391 }
2392 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2393
2394 /*********************************************************************
2395  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2396  *********************************************************************/
2397
2398 /**
2399  * cpufreq_register_driver - register a CPU Frequency driver
2400  * @driver_data: A struct cpufreq_driver containing the values#
2401  * submitted by the CPU Frequency driver.
2402  *
2403  * Registers a CPU Frequency driver to this core code. This code
2404  * returns zero on success, -EBUSY when another driver got here first
2405  * (and isn't unregistered in the meantime).
2406  *
2407  */
2408 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2409 {
2410         unsigned long flags;
2411         int ret;
2412
2413         if (cpufreq_disabled())
2414                 return -ENODEV;
2415
2416         if (!driver_data || !driver_data->verify || !driver_data->init ||
2417             !(driver_data->setpolicy || driver_data->target_index ||
2418                     driver_data->target) ||
2419              (driver_data->setpolicy && (driver_data->target_index ||
2420                     driver_data->target)) ||
2421              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2422                 return -EINVAL;
2423
2424         pr_debug("trying to register driver %s\n", driver_data->name);
2425
2426         write_lock_irqsave(&cpufreq_driver_lock, flags);
2427         if (cpufreq_driver) {
2428                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2429                 return -EEXIST;
2430         }
2431         cpufreq_driver = driver_data;
2432         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2433
2434         if (driver_data->setpolicy)
2435                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2436
2437         if (cpufreq_boost_supported()) {
2438                 /*
2439                  * Check if driver provides function to enable boost -
2440                  * if not, use cpufreq_boost_set_sw as default
2441                  */
2442                 if (!cpufreq_driver->set_boost)
2443                         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2444
2445                 ret = cpufreq_sysfs_create_file(&boost.attr);
2446                 if (ret) {
2447                         pr_err("%s: cannot register global BOOST sysfs file\n",
2448                                __func__);
2449                         goto err_null_driver;
2450                 }
2451         }
2452
2453         ret = subsys_interface_register(&cpufreq_interface);
2454         if (ret)
2455                 goto err_boost_unreg;
2456
2457         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2458             list_empty(&cpufreq_policy_list)) {
2459                 /* if all ->init() calls failed, unregister */
2460                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2461                          driver_data->name);
2462                 goto err_if_unreg;
2463         }
2464
2465         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2466         pr_debug("driver %s up and running\n", driver_data->name);
2467
2468         return 0;
2469 err_if_unreg:
2470         subsys_interface_unregister(&cpufreq_interface);
2471 err_boost_unreg:
2472         if (cpufreq_boost_supported())
2473                 cpufreq_sysfs_remove_file(&boost.attr);
2474 err_null_driver:
2475         write_lock_irqsave(&cpufreq_driver_lock, flags);
2476         cpufreq_driver = NULL;
2477         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2478         return ret;
2479 }
2480 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2481
2482 /**
2483  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2484  *
2485  * Unregister the current CPUFreq driver. Only call this if you have
2486  * the right to do so, i.e. if you have succeeded in initialising before!
2487  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2488  * currently not initialised.
2489  */
2490 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2491 {
2492         unsigned long flags;
2493
2494         if (!cpufreq_driver || (driver != cpufreq_driver))
2495                 return -EINVAL;
2496
2497         pr_debug("unregistering driver %s\n", driver->name);
2498
2499         subsys_interface_unregister(&cpufreq_interface);
2500         if (cpufreq_boost_supported())
2501                 cpufreq_sysfs_remove_file(&boost.attr);
2502
2503         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2504
2505         down_write(&cpufreq_rwsem);
2506         write_lock_irqsave(&cpufreq_driver_lock, flags);
2507
2508         cpufreq_driver = NULL;
2509
2510         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2511         up_write(&cpufreq_rwsem);
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 static int __init cpufreq_core_init(void)
2526 {
2527         if (cpufreq_disabled())
2528                 return -ENODEV;
2529
2530         cpufreq_global_kobject = kobject_create();
2531         BUG_ON(!cpufreq_global_kobject);
2532
2533         register_syscore_ops(&cpufreq_syscore_ops);
2534
2535         return 0;
2536 }
2537 core_initcall(cpufreq_core_init);