cpufreq: update user_policy.* on success
[firefly-linux-kernel-4.4.55.git] / drivers / cpufreq / cpufreq.c
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
7  *
8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
9  *      Added handling for CPU hotplug
10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
11  *      Fix handling for CPU hotplug -- affected CPUs
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License version 2 as
15  * published by the Free Software Foundation.
16  */
17
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20 #include <linux/cpu.h>
21 #include <linux/cpufreq.h>
22 #include <linux/delay.h>
23 #include <linux/device.h>
24 #include <linux/init.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/module.h>
27 #include <linux/mutex.h>
28 #include <linux/slab.h>
29 #include <linux/suspend.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/tick.h>
32 #include <trace/events/power.h>
33
34 static LIST_HEAD(cpufreq_policy_list);
35
36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
37 {
38         return cpumask_empty(policy->cpus);
39 }
40
41 static bool suitable_policy(struct cpufreq_policy *policy, bool active)
42 {
43         return active == !policy_is_inactive(policy);
44 }
45
46 /* Finds Next Acive/Inactive policy */
47 static struct cpufreq_policy *next_policy(struct cpufreq_policy *policy,
48                                           bool active)
49 {
50         do {
51                 policy = list_next_entry(policy, policy_list);
52
53                 /* No more policies in the list */
54                 if (&policy->policy_list == &cpufreq_policy_list)
55                         return NULL;
56         } while (!suitable_policy(policy, active));
57
58         return policy;
59 }
60
61 static struct cpufreq_policy *first_policy(bool active)
62 {
63         struct cpufreq_policy *policy;
64
65         /* No policies in the list */
66         if (list_empty(&cpufreq_policy_list))
67                 return NULL;
68
69         policy = list_first_entry(&cpufreq_policy_list, typeof(*policy),
70                                   policy_list);
71
72         if (!suitable_policy(policy, active))
73                 policy = next_policy(policy, active);
74
75         return policy;
76 }
77
78 /* Macros to iterate over CPU policies */
79 #define for_each_suitable_policy(__policy, __active)    \
80         for (__policy = first_policy(__active);         \
81              __policy;                                  \
82              __policy = next_policy(__policy, __active))
83
84 #define for_each_active_policy(__policy)                \
85         for_each_suitable_policy(__policy, true)
86 #define for_each_inactive_policy(__policy)              \
87         for_each_suitable_policy(__policy, false)
88
89 #define for_each_policy(__policy)                       \
90         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
91
92 /* Iterate over governors */
93 static LIST_HEAD(cpufreq_governor_list);
94 #define for_each_governor(__governor)                           \
95         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
96
97 /**
98  * The "cpufreq driver" - the arch- or hardware-dependent low
99  * level driver of CPUFreq support, and its spinlock. This lock
100  * also protects the cpufreq_cpu_data array.
101  */
102 static struct cpufreq_driver *cpufreq_driver;
103 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
104 static DEFINE_RWLOCK(cpufreq_driver_lock);
105 DEFINE_MUTEX(cpufreq_governor_lock);
106
107 /* Flag to suspend/resume CPUFreq governors */
108 static bool cpufreq_suspended;
109
110 static inline bool has_target(void)
111 {
112         return cpufreq_driver->target_index || cpufreq_driver->target;
113 }
114
115 /* internal prototypes */
116 static int __cpufreq_governor(struct cpufreq_policy *policy,
117                 unsigned int event);
118 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
119 static void handle_update(struct work_struct *work);
120
121 /**
122  * Two notifier lists: the "policy" list is involved in the
123  * validation process for a new CPU frequency policy; the
124  * "transition" list for kernel code that needs to handle
125  * changes to devices when the CPU clock speed changes.
126  * The mutex locks both lists.
127  */
128 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
129 static struct srcu_notifier_head cpufreq_transition_notifier_list;
130
131 static bool init_cpufreq_transition_notifier_list_called;
132 static int __init init_cpufreq_transition_notifier_list(void)
133 {
134         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
135         init_cpufreq_transition_notifier_list_called = true;
136         return 0;
137 }
138 pure_initcall(init_cpufreq_transition_notifier_list);
139
140 static int off __read_mostly;
141 static int cpufreq_disabled(void)
142 {
143         return off;
144 }
145 void disable_cpufreq(void)
146 {
147         off = 1;
148 }
149 static DEFINE_MUTEX(cpufreq_governor_mutex);
150
151 bool have_governor_per_policy(void)
152 {
153         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
154 }
155 EXPORT_SYMBOL_GPL(have_governor_per_policy);
156
157 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
158 {
159         if (have_governor_per_policy())
160                 return &policy->kobj;
161         else
162                 return cpufreq_global_kobject;
163 }
164 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
165
166 struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu)
167 {
168         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
169
170         return policy && !policy_is_inactive(policy) ?
171                 policy->freq_table : NULL;
172 }
173 EXPORT_SYMBOL_GPL(cpufreq_frequency_get_table);
174
175 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
176 {
177         u64 idle_time;
178         u64 cur_wall_time;
179         u64 busy_time;
180
181         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
182
183         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
184         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
185         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
186         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
187         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
188         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
189
190         idle_time = cur_wall_time - busy_time;
191         if (wall)
192                 *wall = cputime_to_usecs(cur_wall_time);
193
194         return cputime_to_usecs(idle_time);
195 }
196
197 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
198 {
199         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
200
201         if (idle_time == -1ULL)
202                 return get_cpu_idle_time_jiffy(cpu, wall);
203         else if (!io_busy)
204                 idle_time += get_cpu_iowait_time_us(cpu, wall);
205
206         return idle_time;
207 }
208 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
209
210 /*
211  * This is a generic cpufreq init() routine which can be used by cpufreq
212  * drivers of SMP systems. It will do following:
213  * - validate & show freq table passed
214  * - set policies transition latency
215  * - policy->cpus with all possible CPUs
216  */
217 int cpufreq_generic_init(struct cpufreq_policy *policy,
218                 struct cpufreq_frequency_table *table,
219                 unsigned int transition_latency)
220 {
221         int ret;
222
223         ret = cpufreq_table_validate_and_show(policy, table);
224         if (ret) {
225                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
226                 return ret;
227         }
228
229         policy->cpuinfo.transition_latency = transition_latency;
230
231         /*
232          * The driver only supports the SMP configuration where all processors
233          * share the clock and voltage and clock.
234          */
235         cpumask_setall(policy->cpus);
236
237         return 0;
238 }
239 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
240
241 /* Only for cpufreq core internal use */
242 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
243 {
244         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
245
246         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
247 }
248
249 unsigned int cpufreq_generic_get(unsigned int cpu)
250 {
251         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
252
253         if (!policy || IS_ERR(policy->clk)) {
254                 pr_err("%s: No %s associated to cpu: %d\n",
255                        __func__, policy ? "clk" : "policy", cpu);
256                 return 0;
257         }
258
259         return clk_get_rate(policy->clk) / 1000;
260 }
261 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
262
263 /**
264  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
265  *
266  * @cpu: cpu to find policy for.
267  *
268  * This returns policy for 'cpu', returns NULL if it doesn't exist.
269  * It also increments the kobject reference count to mark it busy and so would
270  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
271  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
272  * freed as that depends on the kobj count.
273  *
274  * Return: A valid policy on success, otherwise NULL on failure.
275  */
276 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
277 {
278         struct cpufreq_policy *policy = NULL;
279         unsigned long flags;
280
281         if (WARN_ON(cpu >= nr_cpu_ids))
282                 return NULL;
283
284         /* get the cpufreq driver */
285         read_lock_irqsave(&cpufreq_driver_lock, flags);
286
287         if (cpufreq_driver) {
288                 /* get the CPU */
289                 policy = cpufreq_cpu_get_raw(cpu);
290                 if (policy)
291                         kobject_get(&policy->kobj);
292         }
293
294         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
295
296         return policy;
297 }
298 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
299
300 /**
301  * cpufreq_cpu_put: Decrements the usage count of a policy
302  *
303  * @policy: policy earlier returned by cpufreq_cpu_get().
304  *
305  * This decrements the kobject reference count incremented earlier by calling
306  * cpufreq_cpu_get().
307  */
308 void cpufreq_cpu_put(struct cpufreq_policy *policy)
309 {
310         kobject_put(&policy->kobj);
311 }
312 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
313
314 /*********************************************************************
315  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
316  *********************************************************************/
317
318 /**
319  * adjust_jiffies - adjust the system "loops_per_jiffy"
320  *
321  * This function alters the system "loops_per_jiffy" for the clock
322  * speed change. Note that loops_per_jiffy cannot be updated on SMP
323  * systems as each CPU might be scaled differently. So, use the arch
324  * per-CPU loops_per_jiffy value wherever possible.
325  */
326 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
327 {
328 #ifndef CONFIG_SMP
329         static unsigned long l_p_j_ref;
330         static unsigned int l_p_j_ref_freq;
331
332         if (ci->flags & CPUFREQ_CONST_LOOPS)
333                 return;
334
335         if (!l_p_j_ref_freq) {
336                 l_p_j_ref = loops_per_jiffy;
337                 l_p_j_ref_freq = ci->old;
338                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
339                          l_p_j_ref, l_p_j_ref_freq);
340         }
341         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
342                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
343                                                                 ci->new);
344                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
345                          loops_per_jiffy, ci->new);
346         }
347 #endif
348 }
349
350 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
351                 struct cpufreq_freqs *freqs, unsigned int state)
352 {
353         BUG_ON(irqs_disabled());
354
355         if (cpufreq_disabled())
356                 return;
357
358         freqs->flags = cpufreq_driver->flags;
359         pr_debug("notification %u of frequency transition to %u kHz\n",
360                  state, freqs->new);
361
362         switch (state) {
363
364         case CPUFREQ_PRECHANGE:
365                 /* detect if the driver reported a value as "old frequency"
366                  * which is not equal to what the cpufreq core thinks is
367                  * "old frequency".
368                  */
369                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
370                         if ((policy) && (policy->cpu == freqs->cpu) &&
371                             (policy->cur) && (policy->cur != freqs->old)) {
372                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
373                                          freqs->old, policy->cur);
374                                 freqs->old = policy->cur;
375                         }
376                 }
377                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
378                                 CPUFREQ_PRECHANGE, freqs);
379                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
380                 break;
381
382         case CPUFREQ_POSTCHANGE:
383                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
384                 pr_debug("FREQ: %lu - CPU: %lu\n",
385                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
386                 trace_cpu_frequency(freqs->new, freqs->cpu);
387                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
388                                 CPUFREQ_POSTCHANGE, freqs);
389                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
390                         policy->cur = freqs->new;
391                 break;
392         }
393 }
394
395 /**
396  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
397  * on frequency transition.
398  *
399  * This function calls the transition notifiers and the "adjust_jiffies"
400  * function. It is called twice on all CPU frequency changes that have
401  * external effects.
402  */
403 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
404                 struct cpufreq_freqs *freqs, unsigned int state)
405 {
406         for_each_cpu(freqs->cpu, policy->cpus)
407                 __cpufreq_notify_transition(policy, freqs, state);
408 }
409
410 /* Do post notifications when there are chances that transition has failed */
411 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
412                 struct cpufreq_freqs *freqs, int transition_failed)
413 {
414         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
415         if (!transition_failed)
416                 return;
417
418         swap(freqs->old, freqs->new);
419         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
420         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
421 }
422
423 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
424                 struct cpufreq_freqs *freqs)
425 {
426
427         /*
428          * Catch double invocations of _begin() which lead to self-deadlock.
429          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
430          * doesn't invoke _begin() on their behalf, and hence the chances of
431          * double invocations are very low. Moreover, there are scenarios
432          * where these checks can emit false-positive warnings in these
433          * drivers; so we avoid that by skipping them altogether.
434          */
435         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
436                                 && current == policy->transition_task);
437
438 wait:
439         wait_event(policy->transition_wait, !policy->transition_ongoing);
440
441         spin_lock(&policy->transition_lock);
442
443         if (unlikely(policy->transition_ongoing)) {
444                 spin_unlock(&policy->transition_lock);
445                 goto wait;
446         }
447
448         policy->transition_ongoing = true;
449         policy->transition_task = current;
450
451         spin_unlock(&policy->transition_lock);
452
453         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
454 }
455 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
456
457 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
458                 struct cpufreq_freqs *freqs, int transition_failed)
459 {
460         if (unlikely(WARN_ON(!policy->transition_ongoing)))
461                 return;
462
463         cpufreq_notify_post_transition(policy, freqs, transition_failed);
464
465         policy->transition_ongoing = false;
466         policy->transition_task = NULL;
467
468         wake_up(&policy->transition_wait);
469 }
470 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
471
472
473 /*********************************************************************
474  *                          SYSFS INTERFACE                          *
475  *********************************************************************/
476 static ssize_t show_boost(struct kobject *kobj,
477                                  struct attribute *attr, char *buf)
478 {
479         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
480 }
481
482 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
483                                   const char *buf, size_t count)
484 {
485         int ret, enable;
486
487         ret = sscanf(buf, "%d", &enable);
488         if (ret != 1 || enable < 0 || enable > 1)
489                 return -EINVAL;
490
491         if (cpufreq_boost_trigger_state(enable)) {
492                 pr_err("%s: Cannot %s BOOST!\n",
493                        __func__, enable ? "enable" : "disable");
494                 return -EINVAL;
495         }
496
497         pr_debug("%s: cpufreq BOOST %s\n",
498                  __func__, enable ? "enabled" : "disabled");
499
500         return count;
501 }
502 define_one_global_rw(boost);
503
504 static struct cpufreq_governor *find_governor(const char *str_governor)
505 {
506         struct cpufreq_governor *t;
507
508         for_each_governor(t)
509                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
510                         return t;
511
512         return NULL;
513 }
514
515 /**
516  * cpufreq_parse_governor - parse a governor string
517  */
518 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
519                                 struct cpufreq_governor **governor)
520 {
521         int err = -EINVAL;
522
523         if (!cpufreq_driver)
524                 goto out;
525
526         if (cpufreq_driver->setpolicy) {
527                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
528                         *policy = CPUFREQ_POLICY_PERFORMANCE;
529                         err = 0;
530                 } else if (!strncasecmp(str_governor, "powersave",
531                                                 CPUFREQ_NAME_LEN)) {
532                         *policy = CPUFREQ_POLICY_POWERSAVE;
533                         err = 0;
534                 }
535         } else {
536                 struct cpufreq_governor *t;
537
538                 mutex_lock(&cpufreq_governor_mutex);
539
540                 t = find_governor(str_governor);
541
542                 if (t == NULL) {
543                         int ret;
544
545                         mutex_unlock(&cpufreq_governor_mutex);
546                         ret = request_module("cpufreq_%s", str_governor);
547                         mutex_lock(&cpufreq_governor_mutex);
548
549                         if (ret == 0)
550                                 t = find_governor(str_governor);
551                 }
552
553                 if (t != NULL) {
554                         *governor = t;
555                         err = 0;
556                 }
557
558                 mutex_unlock(&cpufreq_governor_mutex);
559         }
560 out:
561         return err;
562 }
563
564 /**
565  * cpufreq_per_cpu_attr_read() / show_##file_name() -
566  * print out cpufreq information
567  *
568  * Write out information from cpufreq_driver->policy[cpu]; object must be
569  * "unsigned int".
570  */
571
572 #define show_one(file_name, object)                     \
573 static ssize_t show_##file_name                         \
574 (struct cpufreq_policy *policy, char *buf)              \
575 {                                                       \
576         return sprintf(buf, "%u\n", policy->object);    \
577 }
578
579 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
580 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
581 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
582 show_one(scaling_min_freq, min);
583 show_one(scaling_max_freq, max);
584
585 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
586 {
587         ssize_t ret;
588
589         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
590                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
591         else
592                 ret = sprintf(buf, "%u\n", policy->cur);
593         return ret;
594 }
595
596 static int cpufreq_set_policy(struct cpufreq_policy *policy,
597                                 struct cpufreq_policy *new_policy);
598
599 /**
600  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
601  */
602 #define store_one(file_name, object)                    \
603 static ssize_t store_##file_name                                        \
604 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
605 {                                                                       \
606         int ret, temp;                                                  \
607         struct cpufreq_policy new_policy;                               \
608                                                                         \
609         memcpy(&new_policy, policy, sizeof(*policy));                   \
610                                                                         \
611         ret = sscanf(buf, "%u", &new_policy.object);                    \
612         if (ret != 1)                                                   \
613                 return -EINVAL;                                         \
614                                                                         \
615         temp = new_policy.object;                                       \
616         ret = cpufreq_set_policy(policy, &new_policy);          \
617         if (!ret)                                                       \
618                 policy->user_policy.object = temp;                      \
619                                                                         \
620         return ret ? ret : count;                                       \
621 }
622
623 store_one(scaling_min_freq, min);
624 store_one(scaling_max_freq, max);
625
626 /**
627  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
628  */
629 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
630                                         char *buf)
631 {
632         unsigned int cur_freq = __cpufreq_get(policy);
633         if (!cur_freq)
634                 return sprintf(buf, "<unknown>");
635         return sprintf(buf, "%u\n", cur_freq);
636 }
637
638 /**
639  * show_scaling_governor - show the current policy for the specified CPU
640  */
641 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
642 {
643         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
644                 return sprintf(buf, "powersave\n");
645         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
646                 return sprintf(buf, "performance\n");
647         else if (policy->governor)
648                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
649                                 policy->governor->name);
650         return -EINVAL;
651 }
652
653 /**
654  * store_scaling_governor - store policy for the specified CPU
655  */
656 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
657                                         const char *buf, size_t count)
658 {
659         int ret;
660         char    str_governor[16];
661         struct cpufreq_policy new_policy;
662
663         memcpy(&new_policy, policy, sizeof(*policy));
664
665         ret = sscanf(buf, "%15s", str_governor);
666         if (ret != 1)
667                 return -EINVAL;
668
669         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
670                                                 &new_policy.governor))
671                 return -EINVAL;
672
673         ret = cpufreq_set_policy(policy, &new_policy);
674         if (ret)
675                 return ret;
676
677         policy->user_policy.policy = policy->policy;
678         policy->user_policy.governor = policy->governor;
679         return count;
680 }
681
682 /**
683  * show_scaling_driver - show the cpufreq driver currently loaded
684  */
685 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
686 {
687         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
688 }
689
690 /**
691  * show_scaling_available_governors - show the available CPUfreq governors
692  */
693 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
694                                                 char *buf)
695 {
696         ssize_t i = 0;
697         struct cpufreq_governor *t;
698
699         if (!has_target()) {
700                 i += sprintf(buf, "performance powersave");
701                 goto out;
702         }
703
704         for_each_governor(t) {
705                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
706                     - (CPUFREQ_NAME_LEN + 2)))
707                         goto out;
708                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
709         }
710 out:
711         i += sprintf(&buf[i], "\n");
712         return i;
713 }
714
715 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
716 {
717         ssize_t i = 0;
718         unsigned int cpu;
719
720         for_each_cpu(cpu, mask) {
721                 if (i)
722                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
723                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
724                 if (i >= (PAGE_SIZE - 5))
725                         break;
726         }
727         i += sprintf(&buf[i], "\n");
728         return i;
729 }
730 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
731
732 /**
733  * show_related_cpus - show the CPUs affected by each transition even if
734  * hw coordination is in use
735  */
736 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
737 {
738         return cpufreq_show_cpus(policy->related_cpus, buf);
739 }
740
741 /**
742  * show_affected_cpus - show the CPUs affected by each transition
743  */
744 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
745 {
746         return cpufreq_show_cpus(policy->cpus, buf);
747 }
748
749 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
750                                         const char *buf, size_t count)
751 {
752         unsigned int freq = 0;
753         unsigned int ret;
754
755         if (!policy->governor || !policy->governor->store_setspeed)
756                 return -EINVAL;
757
758         ret = sscanf(buf, "%u", &freq);
759         if (ret != 1)
760                 return -EINVAL;
761
762         policy->governor->store_setspeed(policy, freq);
763
764         return count;
765 }
766
767 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
768 {
769         if (!policy->governor || !policy->governor->show_setspeed)
770                 return sprintf(buf, "<unsupported>\n");
771
772         return policy->governor->show_setspeed(policy, buf);
773 }
774
775 /**
776  * show_bios_limit - show the current cpufreq HW/BIOS limitation
777  */
778 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
779 {
780         unsigned int limit;
781         int ret;
782         if (cpufreq_driver->bios_limit) {
783                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
784                 if (!ret)
785                         return sprintf(buf, "%u\n", limit);
786         }
787         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
788 }
789
790 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
791 cpufreq_freq_attr_ro(cpuinfo_min_freq);
792 cpufreq_freq_attr_ro(cpuinfo_max_freq);
793 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
794 cpufreq_freq_attr_ro(scaling_available_governors);
795 cpufreq_freq_attr_ro(scaling_driver);
796 cpufreq_freq_attr_ro(scaling_cur_freq);
797 cpufreq_freq_attr_ro(bios_limit);
798 cpufreq_freq_attr_ro(related_cpus);
799 cpufreq_freq_attr_ro(affected_cpus);
800 cpufreq_freq_attr_rw(scaling_min_freq);
801 cpufreq_freq_attr_rw(scaling_max_freq);
802 cpufreq_freq_attr_rw(scaling_governor);
803 cpufreq_freq_attr_rw(scaling_setspeed);
804
805 static struct attribute *default_attrs[] = {
806         &cpuinfo_min_freq.attr,
807         &cpuinfo_max_freq.attr,
808         &cpuinfo_transition_latency.attr,
809         &scaling_min_freq.attr,
810         &scaling_max_freq.attr,
811         &affected_cpus.attr,
812         &related_cpus.attr,
813         &scaling_governor.attr,
814         &scaling_driver.attr,
815         &scaling_available_governors.attr,
816         &scaling_setspeed.attr,
817         NULL
818 };
819
820 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
821 #define to_attr(a) container_of(a, struct freq_attr, attr)
822
823 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
824 {
825         struct cpufreq_policy *policy = to_policy(kobj);
826         struct freq_attr *fattr = to_attr(attr);
827         ssize_t ret;
828
829         down_read(&policy->rwsem);
830
831         if (fattr->show)
832                 ret = fattr->show(policy, buf);
833         else
834                 ret = -EIO;
835
836         up_read(&policy->rwsem);
837
838         return ret;
839 }
840
841 static ssize_t store(struct kobject *kobj, struct attribute *attr,
842                      const char *buf, size_t count)
843 {
844         struct cpufreq_policy *policy = to_policy(kobj);
845         struct freq_attr *fattr = to_attr(attr);
846         ssize_t ret = -EINVAL;
847
848         get_online_cpus();
849
850         if (!cpu_online(policy->cpu))
851                 goto unlock;
852
853         down_write(&policy->rwsem);
854
855         /* Updating inactive policies is invalid, so avoid doing that. */
856         if (unlikely(policy_is_inactive(policy))) {
857                 ret = -EBUSY;
858                 goto unlock_policy_rwsem;
859         }
860
861         if (fattr->store)
862                 ret = fattr->store(policy, buf, count);
863         else
864                 ret = -EIO;
865
866 unlock_policy_rwsem:
867         up_write(&policy->rwsem);
868 unlock:
869         put_online_cpus();
870
871         return ret;
872 }
873
874 static void cpufreq_sysfs_release(struct kobject *kobj)
875 {
876         struct cpufreq_policy *policy = to_policy(kobj);
877         pr_debug("last reference is dropped\n");
878         complete(&policy->kobj_unregister);
879 }
880
881 static const struct sysfs_ops sysfs_ops = {
882         .show   = show,
883         .store  = store,
884 };
885
886 static struct kobj_type ktype_cpufreq = {
887         .sysfs_ops      = &sysfs_ops,
888         .default_attrs  = default_attrs,
889         .release        = cpufreq_sysfs_release,
890 };
891
892 struct kobject *cpufreq_global_kobject;
893 EXPORT_SYMBOL(cpufreq_global_kobject);
894
895 static int cpufreq_global_kobject_usage;
896
897 int cpufreq_get_global_kobject(void)
898 {
899         if (!cpufreq_global_kobject_usage++)
900                 return kobject_add(cpufreq_global_kobject,
901                                 &cpu_subsys.dev_root->kobj, "%s", "cpufreq");
902
903         return 0;
904 }
905 EXPORT_SYMBOL(cpufreq_get_global_kobject);
906
907 void cpufreq_put_global_kobject(void)
908 {
909         if (!--cpufreq_global_kobject_usage)
910                 kobject_del(cpufreq_global_kobject);
911 }
912 EXPORT_SYMBOL(cpufreq_put_global_kobject);
913
914 int cpufreq_sysfs_create_file(const struct attribute *attr)
915 {
916         int ret = cpufreq_get_global_kobject();
917
918         if (!ret) {
919                 ret = sysfs_create_file(cpufreq_global_kobject, attr);
920                 if (ret)
921                         cpufreq_put_global_kobject();
922         }
923
924         return ret;
925 }
926 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
927
928 void cpufreq_sysfs_remove_file(const struct attribute *attr)
929 {
930         sysfs_remove_file(cpufreq_global_kobject, attr);
931         cpufreq_put_global_kobject();
932 }
933 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
934
935 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
936 {
937         struct device *cpu_dev;
938
939         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
940
941         if (!policy)
942                 return 0;
943
944         cpu_dev = get_cpu_device(cpu);
945         if (WARN_ON(!cpu_dev))
946                 return 0;
947
948         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
949 }
950
951 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
952 {
953         struct device *cpu_dev;
954
955         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
956
957         cpu_dev = get_cpu_device(cpu);
958         if (WARN_ON(!cpu_dev))
959                 return;
960
961         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
962 }
963
964 /* Add/remove symlinks for all related CPUs */
965 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
966 {
967         unsigned int j;
968         int ret = 0;
969
970         /* Some related CPUs might not be present (physically hotplugged) */
971         for_each_cpu(j, policy->real_cpus) {
972                 if (j == policy->kobj_cpu)
973                         continue;
974
975                 ret = add_cpu_dev_symlink(policy, j);
976                 if (ret)
977                         break;
978         }
979
980         return ret;
981 }
982
983 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
984 {
985         unsigned int j;
986
987         /* Some related CPUs might not be present (physically hotplugged) */
988         for_each_cpu(j, policy->real_cpus) {
989                 if (j == policy->kobj_cpu)
990                         continue;
991
992                 remove_cpu_dev_symlink(policy, j);
993         }
994 }
995
996 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
997 {
998         struct freq_attr **drv_attr;
999         int ret = 0;
1000
1001         /* set up files for this cpu device */
1002         drv_attr = cpufreq_driver->attr;
1003         while (drv_attr && *drv_attr) {
1004                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1005                 if (ret)
1006                         return ret;
1007                 drv_attr++;
1008         }
1009         if (cpufreq_driver->get) {
1010                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1011                 if (ret)
1012                         return ret;
1013         }
1014
1015         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1016         if (ret)
1017                 return ret;
1018
1019         if (cpufreq_driver->bios_limit) {
1020                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1021                 if (ret)
1022                         return ret;
1023         }
1024
1025         return cpufreq_add_dev_symlink(policy);
1026 }
1027
1028 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1029 {
1030         struct cpufreq_governor *gov = NULL;
1031         struct cpufreq_policy new_policy;
1032
1033         memcpy(&new_policy, policy, sizeof(*policy));
1034
1035         /* Update governor of new_policy to the governor used before hotplug */
1036         gov = find_governor(policy->last_governor);
1037         if (gov)
1038                 pr_debug("Restoring governor %s for cpu %d\n",
1039                                 policy->governor->name, policy->cpu);
1040         else
1041                 gov = CPUFREQ_DEFAULT_GOVERNOR;
1042
1043         new_policy.governor = gov;
1044
1045         /* Use the default policy if its valid. */
1046         if (cpufreq_driver->setpolicy)
1047                 cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
1048
1049         /* set default policy */
1050         return cpufreq_set_policy(policy, &new_policy);
1051 }
1052
1053 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1054 {
1055         int ret = 0;
1056
1057         /* Has this CPU been taken care of already? */
1058         if (cpumask_test_cpu(cpu, policy->cpus))
1059                 return 0;
1060
1061         if (has_target()) {
1062                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1063                 if (ret) {
1064                         pr_err("%s: Failed to stop governor\n", __func__);
1065                         return ret;
1066                 }
1067         }
1068
1069         down_write(&policy->rwsem);
1070         cpumask_set_cpu(cpu, policy->cpus);
1071         up_write(&policy->rwsem);
1072
1073         if (has_target()) {
1074                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1075                 if (!ret)
1076                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1077
1078                 if (ret) {
1079                         pr_err("%s: Failed to start governor\n", __func__);
1080                         return ret;
1081                 }
1082         }
1083
1084         return 0;
1085 }
1086
1087 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1088 {
1089         struct device *dev = get_cpu_device(cpu);
1090         struct cpufreq_policy *policy;
1091         int ret;
1092
1093         if (WARN_ON(!dev))
1094                 return NULL;
1095
1096         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1097         if (!policy)
1098                 return NULL;
1099
1100         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1101                 goto err_free_policy;
1102
1103         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1104                 goto err_free_cpumask;
1105
1106         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1107                 goto err_free_rcpumask;
1108
1109         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &dev->kobj,
1110                                    "cpufreq");
1111         if (ret) {
1112                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1113                 goto err_free_real_cpus;
1114         }
1115
1116         INIT_LIST_HEAD(&policy->policy_list);
1117         init_rwsem(&policy->rwsem);
1118         spin_lock_init(&policy->transition_lock);
1119         init_waitqueue_head(&policy->transition_wait);
1120         init_completion(&policy->kobj_unregister);
1121         INIT_WORK(&policy->update, handle_update);
1122
1123         policy->cpu = cpu;
1124
1125         /* Set this once on allocation */
1126         policy->kobj_cpu = cpu;
1127
1128         return policy;
1129
1130 err_free_real_cpus:
1131         free_cpumask_var(policy->real_cpus);
1132 err_free_rcpumask:
1133         free_cpumask_var(policy->related_cpus);
1134 err_free_cpumask:
1135         free_cpumask_var(policy->cpus);
1136 err_free_policy:
1137         kfree(policy);
1138
1139         return NULL;
1140 }
1141
1142 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1143 {
1144         struct kobject *kobj;
1145         struct completion *cmp;
1146
1147         if (notify)
1148                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1149                                              CPUFREQ_REMOVE_POLICY, policy);
1150
1151         down_write(&policy->rwsem);
1152         cpufreq_remove_dev_symlink(policy);
1153         kobj = &policy->kobj;
1154         cmp = &policy->kobj_unregister;
1155         up_write(&policy->rwsem);
1156         kobject_put(kobj);
1157
1158         /*
1159          * We need to make sure that the underlying kobj is
1160          * actually not referenced anymore by anybody before we
1161          * proceed with unloading.
1162          */
1163         pr_debug("waiting for dropping of refcount\n");
1164         wait_for_completion(cmp);
1165         pr_debug("wait complete\n");
1166 }
1167
1168 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1169 {
1170         unsigned long flags;
1171         int cpu;
1172
1173         /* Remove policy from list */
1174         write_lock_irqsave(&cpufreq_driver_lock, flags);
1175         list_del(&policy->policy_list);
1176
1177         for_each_cpu(cpu, policy->related_cpus)
1178                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1179         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1180
1181         cpufreq_policy_put_kobj(policy, notify);
1182         free_cpumask_var(policy->real_cpus);
1183         free_cpumask_var(policy->related_cpus);
1184         free_cpumask_var(policy->cpus);
1185         kfree(policy);
1186 }
1187
1188 static int cpufreq_online(unsigned int cpu)
1189 {
1190         struct cpufreq_policy *policy;
1191         bool new_policy;
1192         unsigned long flags;
1193         unsigned int j;
1194         int ret;
1195
1196         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1197
1198         /* Check if this CPU already has a policy to manage it */
1199         policy = per_cpu(cpufreq_cpu_data, cpu);
1200         if (policy) {
1201                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1202                 if (!policy_is_inactive(policy))
1203                         return cpufreq_add_policy_cpu(policy, cpu);
1204
1205                 /* This is the only online CPU for the policy.  Start over. */
1206                 new_policy = false;
1207                 down_write(&policy->rwsem);
1208                 policy->cpu = cpu;
1209                 policy->governor = NULL;
1210                 up_write(&policy->rwsem);
1211         } else {
1212                 new_policy = true;
1213                 policy = cpufreq_policy_alloc(cpu);
1214                 if (!policy)
1215                         return -ENOMEM;
1216         }
1217
1218         cpumask_copy(policy->cpus, cpumask_of(cpu));
1219
1220         /* call driver. From then on the cpufreq must be able
1221          * to accept all calls to ->verify and ->setpolicy for this CPU
1222          */
1223         ret = cpufreq_driver->init(policy);
1224         if (ret) {
1225                 pr_debug("initialization failed\n");
1226                 goto out_free_policy;
1227         }
1228
1229         down_write(&policy->rwsem);
1230
1231         if (new_policy) {
1232                 /* related_cpus should at least include policy->cpus. */
1233                 cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1234                 /* Remember CPUs present at the policy creation time. */
1235                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1236         }
1237
1238         /*
1239          * affected cpus must always be the one, which are online. We aren't
1240          * managing offline cpus here.
1241          */
1242         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1243
1244         if (new_policy) {
1245                 policy->user_policy.min = policy->min;
1246                 policy->user_policy.max = policy->max;
1247
1248                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1249                 for_each_cpu(j, policy->related_cpus)
1250                         per_cpu(cpufreq_cpu_data, j) = policy;
1251                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1252         }
1253
1254         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1255                 policy->cur = cpufreq_driver->get(policy->cpu);
1256                 if (!policy->cur) {
1257                         pr_err("%s: ->get() failed\n", __func__);
1258                         goto out_exit_policy;
1259                 }
1260         }
1261
1262         /*
1263          * Sometimes boot loaders set CPU frequency to a value outside of
1264          * frequency table present with cpufreq core. In such cases CPU might be
1265          * unstable if it has to run on that frequency for long duration of time
1266          * and so its better to set it to a frequency which is specified in
1267          * freq-table. This also makes cpufreq stats inconsistent as
1268          * cpufreq-stats would fail to register because current frequency of CPU
1269          * isn't found in freq-table.
1270          *
1271          * Because we don't want this change to effect boot process badly, we go
1272          * for the next freq which is >= policy->cur ('cur' must be set by now,
1273          * otherwise we will end up setting freq to lowest of the table as 'cur'
1274          * is initialized to zero).
1275          *
1276          * We are passing target-freq as "policy->cur - 1" otherwise
1277          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1278          * equal to target-freq.
1279          */
1280         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1281             && has_target()) {
1282                 /* Are we running at unknown frequency ? */
1283                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1284                 if (ret == -EINVAL) {
1285                         /* Warn user and fix it */
1286                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1287                                 __func__, policy->cpu, policy->cur);
1288                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1289                                 CPUFREQ_RELATION_L);
1290
1291                         /*
1292                          * Reaching here after boot in a few seconds may not
1293                          * mean that system will remain stable at "unknown"
1294                          * frequency for longer duration. Hence, a BUG_ON().
1295                          */
1296                         BUG_ON(ret);
1297                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1298                                 __func__, policy->cpu, policy->cur);
1299                 }
1300         }
1301
1302         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1303                                      CPUFREQ_START, policy);
1304
1305         if (new_policy) {
1306                 ret = cpufreq_add_dev_interface(policy);
1307                 if (ret)
1308                         goto out_exit_policy;
1309                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1310                                 CPUFREQ_CREATE_POLICY, policy);
1311
1312                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1313                 list_add(&policy->policy_list, &cpufreq_policy_list);
1314                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1315         }
1316
1317         ret = cpufreq_init_policy(policy);
1318         if (ret) {
1319                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1320                        __func__, cpu, ret);
1321                 /* cpufreq_policy_free() will notify based on this */
1322                 new_policy = false;
1323                 goto out_exit_policy;
1324         }
1325
1326         if (new_policy) {
1327                 policy->user_policy.policy = policy->policy;
1328                 policy->user_policy.governor = policy->governor;
1329         }
1330         up_write(&policy->rwsem);
1331
1332         kobject_uevent(&policy->kobj, KOBJ_ADD);
1333
1334         /* Callback for handling stuff after policy is ready */
1335         if (cpufreq_driver->ready)
1336                 cpufreq_driver->ready(policy);
1337
1338         pr_debug("initialization complete\n");
1339
1340         return 0;
1341
1342 out_exit_policy:
1343         up_write(&policy->rwsem);
1344
1345         if (cpufreq_driver->exit)
1346                 cpufreq_driver->exit(policy);
1347 out_free_policy:
1348         cpufreq_policy_free(policy, !new_policy);
1349         return ret;
1350 }
1351
1352 /**
1353  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1354  * @dev: CPU device.
1355  * @sif: Subsystem interface structure pointer (not used)
1356  */
1357 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1358 {
1359         unsigned cpu = dev->id;
1360         int ret;
1361
1362         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1363
1364         if (cpu_online(cpu)) {
1365                 ret = cpufreq_online(cpu);
1366         } else {
1367                 /*
1368                  * A hotplug notifier will follow and we will handle it as CPU
1369                  * online then.  For now, just create the sysfs link, unless
1370                  * there is no policy or the link is already present.
1371                  */
1372                 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1373
1374                 ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1375                         ? add_cpu_dev_symlink(policy, cpu) : 0;
1376         }
1377
1378         return ret;
1379 }
1380
1381 static void cpufreq_offline_prepare(unsigned int cpu)
1382 {
1383         struct cpufreq_policy *policy;
1384
1385         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1386
1387         policy = cpufreq_cpu_get_raw(cpu);
1388         if (!policy) {
1389                 pr_debug("%s: No cpu_data found\n", __func__);
1390                 return;
1391         }
1392
1393         if (has_target()) {
1394                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1395                 if (ret)
1396                         pr_err("%s: Failed to stop governor\n", __func__);
1397         }
1398
1399         down_write(&policy->rwsem);
1400         cpumask_clear_cpu(cpu, policy->cpus);
1401
1402         if (policy_is_inactive(policy)) {
1403                 if (has_target())
1404                         strncpy(policy->last_governor, policy->governor->name,
1405                                 CPUFREQ_NAME_LEN);
1406         } else if (cpu == policy->cpu) {
1407                 /* Nominate new CPU */
1408                 policy->cpu = cpumask_any(policy->cpus);
1409         }
1410         up_write(&policy->rwsem);
1411
1412         /* Start governor again for active policy */
1413         if (!policy_is_inactive(policy)) {
1414                 if (has_target()) {
1415                         int ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1416                         if (!ret)
1417                                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1418
1419                         if (ret)
1420                                 pr_err("%s: Failed to start governor\n", __func__);
1421                 }
1422         } else if (cpufreq_driver->stop_cpu) {
1423                 cpufreq_driver->stop_cpu(policy);
1424         }
1425 }
1426
1427 static void cpufreq_offline_finish(unsigned int cpu)
1428 {
1429         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1430
1431         if (!policy) {
1432                 pr_debug("%s: No cpu_data found\n", __func__);
1433                 return;
1434         }
1435
1436         /* Only proceed for inactive policies */
1437         if (!policy_is_inactive(policy))
1438                 return;
1439
1440         /* If cpu is last user of policy, free policy */
1441         if (has_target()) {
1442                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1443                 if (ret)
1444                         pr_err("%s: Failed to exit governor\n", __func__);
1445         }
1446
1447         /*
1448          * Perform the ->exit() even during light-weight tear-down,
1449          * since this is a core component, and is essential for the
1450          * subsequent light-weight ->init() to succeed.
1451          */
1452         if (cpufreq_driver->exit)
1453                 cpufreq_driver->exit(policy);
1454 }
1455
1456 /**
1457  * cpufreq_remove_dev - remove a CPU device
1458  *
1459  * Removes the cpufreq interface for a CPU device.
1460  */
1461 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1462 {
1463         unsigned int cpu = dev->id;
1464         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1465
1466         if (!policy)
1467                 return 0;
1468
1469         if (cpu_online(cpu)) {
1470                 cpufreq_offline_prepare(cpu);
1471                 cpufreq_offline_finish(cpu);
1472         }
1473
1474         cpumask_clear_cpu(cpu, policy->real_cpus);
1475
1476         if (cpumask_empty(policy->real_cpus)) {
1477                 cpufreq_policy_free(policy, true);
1478                 return 0;
1479         }
1480
1481         if (cpu != policy->kobj_cpu) {
1482                 remove_cpu_dev_symlink(policy, cpu);
1483         } else {
1484                 /*
1485                  * The CPU owning the policy object is going away.  Move it to
1486                  * another suitable CPU.
1487                  */
1488                 unsigned int new_cpu = cpumask_first(policy->real_cpus);
1489                 struct device *new_dev = get_cpu_device(new_cpu);
1490
1491                 dev_dbg(dev, "%s: Moving policy object to CPU%u\n", __func__, new_cpu);
1492
1493                 sysfs_remove_link(&new_dev->kobj, "cpufreq");
1494                 policy->kobj_cpu = new_cpu;
1495                 WARN_ON(kobject_move(&policy->kobj, &new_dev->kobj));
1496         }
1497
1498         return 0;
1499 }
1500
1501 static void handle_update(struct work_struct *work)
1502 {
1503         struct cpufreq_policy *policy =
1504                 container_of(work, struct cpufreq_policy, update);
1505         unsigned int cpu = policy->cpu;
1506         pr_debug("handle_update for cpu %u called\n", cpu);
1507         cpufreq_update_policy(cpu);
1508 }
1509
1510 /**
1511  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1512  *      in deep trouble.
1513  *      @policy: policy managing CPUs
1514  *      @new_freq: CPU frequency the CPU actually runs at
1515  *
1516  *      We adjust to current frequency first, and need to clean up later.
1517  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1518  */
1519 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1520                                 unsigned int new_freq)
1521 {
1522         struct cpufreq_freqs freqs;
1523
1524         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1525                  policy->cur, new_freq);
1526
1527         freqs.old = policy->cur;
1528         freqs.new = new_freq;
1529
1530         cpufreq_freq_transition_begin(policy, &freqs);
1531         cpufreq_freq_transition_end(policy, &freqs, 0);
1532 }
1533
1534 /**
1535  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1536  * @cpu: CPU number
1537  *
1538  * This is the last known freq, without actually getting it from the driver.
1539  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1540  */
1541 unsigned int cpufreq_quick_get(unsigned int cpu)
1542 {
1543         struct cpufreq_policy *policy;
1544         unsigned int ret_freq = 0;
1545
1546         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1547                 return cpufreq_driver->get(cpu);
1548
1549         policy = cpufreq_cpu_get(cpu);
1550         if (policy) {
1551                 ret_freq = policy->cur;
1552                 cpufreq_cpu_put(policy);
1553         }
1554
1555         return ret_freq;
1556 }
1557 EXPORT_SYMBOL(cpufreq_quick_get);
1558
1559 /**
1560  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1561  * @cpu: CPU number
1562  *
1563  * Just return the max possible frequency for a given CPU.
1564  */
1565 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1566 {
1567         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1568         unsigned int ret_freq = 0;
1569
1570         if (policy) {
1571                 ret_freq = policy->max;
1572                 cpufreq_cpu_put(policy);
1573         }
1574
1575         return ret_freq;
1576 }
1577 EXPORT_SYMBOL(cpufreq_quick_get_max);
1578
1579 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1580 {
1581         unsigned int ret_freq = 0;
1582
1583         if (!cpufreq_driver->get)
1584                 return ret_freq;
1585
1586         ret_freq = cpufreq_driver->get(policy->cpu);
1587
1588         /* Updating inactive policies is invalid, so avoid doing that. */
1589         if (unlikely(policy_is_inactive(policy)))
1590                 return ret_freq;
1591
1592         if (ret_freq && policy->cur &&
1593                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1594                 /* verify no discrepancy between actual and
1595                                         saved value exists */
1596                 if (unlikely(ret_freq != policy->cur)) {
1597                         cpufreq_out_of_sync(policy, ret_freq);
1598                         schedule_work(&policy->update);
1599                 }
1600         }
1601
1602         return ret_freq;
1603 }
1604
1605 /**
1606  * cpufreq_get - get the current CPU frequency (in kHz)
1607  * @cpu: CPU number
1608  *
1609  * Get the CPU current (static) CPU frequency
1610  */
1611 unsigned int cpufreq_get(unsigned int cpu)
1612 {
1613         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1614         unsigned int ret_freq = 0;
1615
1616         if (policy) {
1617                 down_read(&policy->rwsem);
1618                 ret_freq = __cpufreq_get(policy);
1619                 up_read(&policy->rwsem);
1620
1621                 cpufreq_cpu_put(policy);
1622         }
1623
1624         return ret_freq;
1625 }
1626 EXPORT_SYMBOL(cpufreq_get);
1627
1628 static struct subsys_interface cpufreq_interface = {
1629         .name           = "cpufreq",
1630         .subsys         = &cpu_subsys,
1631         .add_dev        = cpufreq_add_dev,
1632         .remove_dev     = cpufreq_remove_dev,
1633 };
1634
1635 /*
1636  * In case platform wants some specific frequency to be configured
1637  * during suspend..
1638  */
1639 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1640 {
1641         int ret;
1642
1643         if (!policy->suspend_freq) {
1644                 pr_err("%s: suspend_freq can't be zero\n", __func__);
1645                 return -EINVAL;
1646         }
1647
1648         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1649                         policy->suspend_freq);
1650
1651         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1652                         CPUFREQ_RELATION_H);
1653         if (ret)
1654                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1655                                 __func__, policy->suspend_freq, ret);
1656
1657         return ret;
1658 }
1659 EXPORT_SYMBOL(cpufreq_generic_suspend);
1660
1661 /**
1662  * cpufreq_suspend() - Suspend CPUFreq governors
1663  *
1664  * Called during system wide Suspend/Hibernate cycles for suspending governors
1665  * as some platforms can't change frequency after this point in suspend cycle.
1666  * Because some of the devices (like: i2c, regulators, etc) they use for
1667  * changing frequency are suspended quickly after this point.
1668  */
1669 void cpufreq_suspend(void)
1670 {
1671         struct cpufreq_policy *policy;
1672
1673         if (!cpufreq_driver)
1674                 return;
1675
1676         if (!has_target())
1677                 goto suspend;
1678
1679         pr_debug("%s: Suspending Governors\n", __func__);
1680
1681         for_each_active_policy(policy) {
1682                 if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1683                         pr_err("%s: Failed to stop governor for policy: %p\n",
1684                                 __func__, policy);
1685                 else if (cpufreq_driver->suspend
1686                     && cpufreq_driver->suspend(policy))
1687                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1688                                 policy);
1689         }
1690
1691 suspend:
1692         cpufreq_suspended = true;
1693 }
1694
1695 /**
1696  * cpufreq_resume() - Resume CPUFreq governors
1697  *
1698  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1699  * are suspended with cpufreq_suspend().
1700  */
1701 void cpufreq_resume(void)
1702 {
1703         struct cpufreq_policy *policy;
1704
1705         if (!cpufreq_driver)
1706                 return;
1707
1708         cpufreq_suspended = false;
1709
1710         if (!has_target())
1711                 return;
1712
1713         pr_debug("%s: Resuming Governors\n", __func__);
1714
1715         for_each_active_policy(policy) {
1716                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1717                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1718                                 policy);
1719                 else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1720                     || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1721                         pr_err("%s: Failed to start governor for policy: %p\n",
1722                                 __func__, policy);
1723         }
1724
1725         /*
1726          * schedule call cpufreq_update_policy() for first-online CPU, as that
1727          * wouldn't be hotplugged-out on suspend. It will verify that the
1728          * current freq is in sync with what we believe it to be.
1729          */
1730         policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1731         if (WARN_ON(!policy))
1732                 return;
1733
1734         schedule_work(&policy->update);
1735 }
1736
1737 /**
1738  *      cpufreq_get_current_driver - return current driver's name
1739  *
1740  *      Return the name string of the currently loaded cpufreq driver
1741  *      or NULL, if none.
1742  */
1743 const char *cpufreq_get_current_driver(void)
1744 {
1745         if (cpufreq_driver)
1746                 return cpufreq_driver->name;
1747
1748         return NULL;
1749 }
1750 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1751
1752 /**
1753  *      cpufreq_get_driver_data - return current driver data
1754  *
1755  *      Return the private data of the currently loaded cpufreq
1756  *      driver, or NULL if no cpufreq driver is loaded.
1757  */
1758 void *cpufreq_get_driver_data(void)
1759 {
1760         if (cpufreq_driver)
1761                 return cpufreq_driver->driver_data;
1762
1763         return NULL;
1764 }
1765 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1766
1767 /*********************************************************************
1768  *                     NOTIFIER LISTS INTERFACE                      *
1769  *********************************************************************/
1770
1771 /**
1772  *      cpufreq_register_notifier - register a driver with cpufreq
1773  *      @nb: notifier function to register
1774  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1775  *
1776  *      Add a driver to one of two lists: either a list of drivers that
1777  *      are notified about clock rate changes (once before and once after
1778  *      the transition), or a list of drivers that are notified about
1779  *      changes in cpufreq policy.
1780  *
1781  *      This function may sleep, and has the same return conditions as
1782  *      blocking_notifier_chain_register.
1783  */
1784 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1785 {
1786         int ret;
1787
1788         if (cpufreq_disabled())
1789                 return -EINVAL;
1790
1791         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1792
1793         switch (list) {
1794         case CPUFREQ_TRANSITION_NOTIFIER:
1795                 ret = srcu_notifier_chain_register(
1796                                 &cpufreq_transition_notifier_list, nb);
1797                 break;
1798         case CPUFREQ_POLICY_NOTIFIER:
1799                 ret = blocking_notifier_chain_register(
1800                                 &cpufreq_policy_notifier_list, nb);
1801                 break;
1802         default:
1803                 ret = -EINVAL;
1804         }
1805
1806         return ret;
1807 }
1808 EXPORT_SYMBOL(cpufreq_register_notifier);
1809
1810 /**
1811  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1812  *      @nb: notifier block to be unregistered
1813  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1814  *
1815  *      Remove a driver from the CPU frequency notifier list.
1816  *
1817  *      This function may sleep, and has the same return conditions as
1818  *      blocking_notifier_chain_unregister.
1819  */
1820 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1821 {
1822         int ret;
1823
1824         if (cpufreq_disabled())
1825                 return -EINVAL;
1826
1827         switch (list) {
1828         case CPUFREQ_TRANSITION_NOTIFIER:
1829                 ret = srcu_notifier_chain_unregister(
1830                                 &cpufreq_transition_notifier_list, nb);
1831                 break;
1832         case CPUFREQ_POLICY_NOTIFIER:
1833                 ret = blocking_notifier_chain_unregister(
1834                                 &cpufreq_policy_notifier_list, nb);
1835                 break;
1836         default:
1837                 ret = -EINVAL;
1838         }
1839
1840         return ret;
1841 }
1842 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1843
1844
1845 /*********************************************************************
1846  *                              GOVERNORS                            *
1847  *********************************************************************/
1848
1849 /* Must set freqs->new to intermediate frequency */
1850 static int __target_intermediate(struct cpufreq_policy *policy,
1851                                  struct cpufreq_freqs *freqs, int index)
1852 {
1853         int ret;
1854
1855         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1856
1857         /* We don't need to switch to intermediate freq */
1858         if (!freqs->new)
1859                 return 0;
1860
1861         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1862                  __func__, policy->cpu, freqs->old, freqs->new);
1863
1864         cpufreq_freq_transition_begin(policy, freqs);
1865         ret = cpufreq_driver->target_intermediate(policy, index);
1866         cpufreq_freq_transition_end(policy, freqs, ret);
1867
1868         if (ret)
1869                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1870                        __func__, ret);
1871
1872         return ret;
1873 }
1874
1875 static int __target_index(struct cpufreq_policy *policy,
1876                           struct cpufreq_frequency_table *freq_table, int index)
1877 {
1878         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1879         unsigned int intermediate_freq = 0;
1880         int retval = -EINVAL;
1881         bool notify;
1882
1883         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1884         if (notify) {
1885                 /* Handle switching to intermediate frequency */
1886                 if (cpufreq_driver->get_intermediate) {
1887                         retval = __target_intermediate(policy, &freqs, index);
1888                         if (retval)
1889                                 return retval;
1890
1891                         intermediate_freq = freqs.new;
1892                         /* Set old freq to intermediate */
1893                         if (intermediate_freq)
1894                                 freqs.old = freqs.new;
1895                 }
1896
1897                 freqs.new = freq_table[index].frequency;
1898                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1899                          __func__, policy->cpu, freqs.old, freqs.new);
1900
1901                 cpufreq_freq_transition_begin(policy, &freqs);
1902         }
1903
1904         retval = cpufreq_driver->target_index(policy, index);
1905         if (retval)
1906                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1907                        retval);
1908
1909         if (notify) {
1910                 cpufreq_freq_transition_end(policy, &freqs, retval);
1911
1912                 /*
1913                  * Failed after setting to intermediate freq? Driver should have
1914                  * reverted back to initial frequency and so should we. Check
1915                  * here for intermediate_freq instead of get_intermediate, in
1916                  * case we haven't switched to intermediate freq at all.
1917                  */
1918                 if (unlikely(retval && intermediate_freq)) {
1919                         freqs.old = intermediate_freq;
1920                         freqs.new = policy->restore_freq;
1921                         cpufreq_freq_transition_begin(policy, &freqs);
1922                         cpufreq_freq_transition_end(policy, &freqs, 0);
1923                 }
1924         }
1925
1926         return retval;
1927 }
1928
1929 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1930                             unsigned int target_freq,
1931                             unsigned int relation)
1932 {
1933         unsigned int old_target_freq = target_freq;
1934         int retval = -EINVAL;
1935
1936         if (cpufreq_disabled())
1937                 return -ENODEV;
1938
1939         /* Make sure that target_freq is within supported range */
1940         if (target_freq > policy->max)
1941                 target_freq = policy->max;
1942         if (target_freq < policy->min)
1943                 target_freq = policy->min;
1944
1945         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1946                  policy->cpu, target_freq, relation, old_target_freq);
1947
1948         /*
1949          * This might look like a redundant call as we are checking it again
1950          * after finding index. But it is left intentionally for cases where
1951          * exactly same freq is called again and so we can save on few function
1952          * calls.
1953          */
1954         if (target_freq == policy->cur)
1955                 return 0;
1956
1957         /* Save last value to restore later on errors */
1958         policy->restore_freq = policy->cur;
1959
1960         if (cpufreq_driver->target)
1961                 retval = cpufreq_driver->target(policy, target_freq, relation);
1962         else if (cpufreq_driver->target_index) {
1963                 struct cpufreq_frequency_table *freq_table;
1964                 int index;
1965
1966                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1967                 if (unlikely(!freq_table)) {
1968                         pr_err("%s: Unable to find freq_table\n", __func__);
1969                         goto out;
1970                 }
1971
1972                 retval = cpufreq_frequency_table_target(policy, freq_table,
1973                                 target_freq, relation, &index);
1974                 if (unlikely(retval)) {
1975                         pr_err("%s: Unable to find matching freq\n", __func__);
1976                         goto out;
1977                 }
1978
1979                 if (freq_table[index].frequency == policy->cur) {
1980                         retval = 0;
1981                         goto out;
1982                 }
1983
1984                 retval = __target_index(policy, freq_table, index);
1985         }
1986
1987 out:
1988         return retval;
1989 }
1990 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1991
1992 int cpufreq_driver_target(struct cpufreq_policy *policy,
1993                           unsigned int target_freq,
1994                           unsigned int relation)
1995 {
1996         int ret = -EINVAL;
1997
1998         down_write(&policy->rwsem);
1999
2000         ret = __cpufreq_driver_target(policy, target_freq, relation);
2001
2002         up_write(&policy->rwsem);
2003
2004         return ret;
2005 }
2006 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2007
2008 static int __cpufreq_governor(struct cpufreq_policy *policy,
2009                                         unsigned int event)
2010 {
2011         int ret;
2012
2013         /* Only must be defined when default governor is known to have latency
2014            restrictions, like e.g. conservative or ondemand.
2015            That this is the case is already ensured in Kconfig
2016         */
2017 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
2018         struct cpufreq_governor *gov = &cpufreq_gov_performance;
2019 #else
2020         struct cpufreq_governor *gov = NULL;
2021 #endif
2022
2023         /* Don't start any governor operations if we are entering suspend */
2024         if (cpufreq_suspended)
2025                 return 0;
2026         /*
2027          * Governor might not be initiated here if ACPI _PPC changed
2028          * notification happened, so check it.
2029          */
2030         if (!policy->governor)
2031                 return -EINVAL;
2032
2033         if (policy->governor->max_transition_latency &&
2034             policy->cpuinfo.transition_latency >
2035             policy->governor->max_transition_latency) {
2036                 if (!gov)
2037                         return -EINVAL;
2038                 else {
2039                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2040                                 policy->governor->name, gov->name);
2041                         policy->governor = gov;
2042                 }
2043         }
2044
2045         if (event == CPUFREQ_GOV_POLICY_INIT)
2046                 if (!try_module_get(policy->governor->owner))
2047                         return -EINVAL;
2048
2049         pr_debug("__cpufreq_governor for CPU %u, event %u\n",
2050                  policy->cpu, event);
2051
2052         mutex_lock(&cpufreq_governor_lock);
2053         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2054             || (!policy->governor_enabled
2055             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2056                 mutex_unlock(&cpufreq_governor_lock);
2057                 return -EBUSY;
2058         }
2059
2060         if (event == CPUFREQ_GOV_STOP)
2061                 policy->governor_enabled = false;
2062         else if (event == CPUFREQ_GOV_START)
2063                 policy->governor_enabled = true;
2064
2065         mutex_unlock(&cpufreq_governor_lock);
2066
2067         ret = policy->governor->governor(policy, event);
2068
2069         if (!ret) {
2070                 if (event == CPUFREQ_GOV_POLICY_INIT)
2071                         policy->governor->initialized++;
2072                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
2073                         policy->governor->initialized--;
2074         } else {
2075                 /* Restore original values */
2076                 mutex_lock(&cpufreq_governor_lock);
2077                 if (event == CPUFREQ_GOV_STOP)
2078                         policy->governor_enabled = true;
2079                 else if (event == CPUFREQ_GOV_START)
2080                         policy->governor_enabled = false;
2081                 mutex_unlock(&cpufreq_governor_lock);
2082         }
2083
2084         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2085                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2086                 module_put(policy->governor->owner);
2087
2088         return ret;
2089 }
2090
2091 int cpufreq_register_governor(struct cpufreq_governor *governor)
2092 {
2093         int err;
2094
2095         if (!governor)
2096                 return -EINVAL;
2097
2098         if (cpufreq_disabled())
2099                 return -ENODEV;
2100
2101         mutex_lock(&cpufreq_governor_mutex);
2102
2103         governor->initialized = 0;
2104         err = -EBUSY;
2105         if (!find_governor(governor->name)) {
2106                 err = 0;
2107                 list_add(&governor->governor_list, &cpufreq_governor_list);
2108         }
2109
2110         mutex_unlock(&cpufreq_governor_mutex);
2111         return err;
2112 }
2113 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2114
2115 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2116 {
2117         struct cpufreq_policy *policy;
2118         unsigned long flags;
2119
2120         if (!governor)
2121                 return;
2122
2123         if (cpufreq_disabled())
2124                 return;
2125
2126         /* clear last_governor for all inactive policies */
2127         read_lock_irqsave(&cpufreq_driver_lock, flags);
2128         for_each_inactive_policy(policy) {
2129                 if (!strcmp(policy->last_governor, governor->name)) {
2130                         policy->governor = NULL;
2131                         strcpy(policy->last_governor, "\0");
2132                 }
2133         }
2134         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2135
2136         mutex_lock(&cpufreq_governor_mutex);
2137         list_del(&governor->governor_list);
2138         mutex_unlock(&cpufreq_governor_mutex);
2139         return;
2140 }
2141 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2142
2143
2144 /*********************************************************************
2145  *                          POLICY INTERFACE                         *
2146  *********************************************************************/
2147
2148 /**
2149  * cpufreq_get_policy - get the current cpufreq_policy
2150  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2151  *      is written
2152  *
2153  * Reads the current cpufreq policy.
2154  */
2155 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2156 {
2157         struct cpufreq_policy *cpu_policy;
2158         if (!policy)
2159                 return -EINVAL;
2160
2161         cpu_policy = cpufreq_cpu_get(cpu);
2162         if (!cpu_policy)
2163                 return -EINVAL;
2164
2165         memcpy(policy, cpu_policy, sizeof(*policy));
2166
2167         cpufreq_cpu_put(cpu_policy);
2168         return 0;
2169 }
2170 EXPORT_SYMBOL(cpufreq_get_policy);
2171
2172 /*
2173  * policy : current policy.
2174  * new_policy: policy to be set.
2175  */
2176 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2177                                 struct cpufreq_policy *new_policy)
2178 {
2179         struct cpufreq_governor *old_gov;
2180         int ret;
2181
2182         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2183                  new_policy->cpu, new_policy->min, new_policy->max);
2184
2185         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2186
2187         /*
2188         * This check works well when we store new min/max freq attributes,
2189         * because new_policy is a copy of policy with one field updated.
2190         */
2191         if (new_policy->min > new_policy->max)
2192                 return -EINVAL;
2193
2194         /* verify the cpu speed can be set within this limit */
2195         ret = cpufreq_driver->verify(new_policy);
2196         if (ret)
2197                 return ret;
2198
2199         /* adjust if necessary - all reasons */
2200         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2201                         CPUFREQ_ADJUST, new_policy);
2202
2203         /*
2204          * verify the cpu speed can be set within this limit, which might be
2205          * different to the first one
2206          */
2207         ret = cpufreq_driver->verify(new_policy);
2208         if (ret)
2209                 return ret;
2210
2211         /* notification of the new policy */
2212         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2213                         CPUFREQ_NOTIFY, new_policy);
2214
2215         policy->min = new_policy->min;
2216         policy->max = new_policy->max;
2217
2218         pr_debug("new min and max freqs are %u - %u kHz\n",
2219                  policy->min, policy->max);
2220
2221         if (cpufreq_driver->setpolicy) {
2222                 policy->policy = new_policy->policy;
2223                 pr_debug("setting range\n");
2224                 return cpufreq_driver->setpolicy(new_policy);
2225         }
2226
2227         if (new_policy->governor == policy->governor)
2228                 goto out;
2229
2230         pr_debug("governor switch\n");
2231
2232         /* save old, working values */
2233         old_gov = policy->governor;
2234         /* end old governor */
2235         if (old_gov) {
2236                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2237                 if (ret) {
2238                         /* This can happen due to race with other operations */
2239                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2240                                  __func__, old_gov->name, ret);
2241                         return ret;
2242                 }
2243
2244                 up_write(&policy->rwsem);
2245                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2246                 down_write(&policy->rwsem);
2247
2248                 if (ret) {
2249                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2250                                __func__, old_gov->name, ret);
2251                         return ret;
2252                 }
2253         }
2254
2255         /* start new governor */
2256         policy->governor = new_policy->governor;
2257         ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2258         if (!ret) {
2259                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
2260                 if (!ret)
2261                         goto out;
2262
2263                 up_write(&policy->rwsem);
2264                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2265                 down_write(&policy->rwsem);
2266         }
2267
2268         /* new governor failed, so re-start old one */
2269         pr_debug("starting governor %s failed\n", policy->governor->name);
2270         if (old_gov) {
2271                 policy->governor = old_gov;
2272                 if (__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2273                         policy->governor = NULL;
2274                 else
2275                         __cpufreq_governor(policy, CPUFREQ_GOV_START);
2276         }
2277
2278         return ret;
2279
2280  out:
2281         pr_debug("governor: change or update limits\n");
2282         return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2283 }
2284
2285 /**
2286  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2287  *      @cpu: CPU which shall be re-evaluated
2288  *
2289  *      Useful for policy notifiers which have different necessities
2290  *      at different times.
2291  */
2292 int cpufreq_update_policy(unsigned int cpu)
2293 {
2294         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2295         struct cpufreq_policy new_policy;
2296         int ret;
2297
2298         if (!policy)
2299                 return -ENODEV;
2300
2301         down_write(&policy->rwsem);
2302
2303         pr_debug("updating policy for CPU %u\n", cpu);
2304         memcpy(&new_policy, policy, sizeof(*policy));
2305         new_policy.min = policy->user_policy.min;
2306         new_policy.max = policy->user_policy.max;
2307         new_policy.policy = policy->user_policy.policy;
2308         new_policy.governor = policy->user_policy.governor;
2309
2310         /*
2311          * BIOS might change freq behind our back
2312          * -> ask driver for current freq and notify governors about a change
2313          */
2314         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2315                 new_policy.cur = cpufreq_driver->get(cpu);
2316                 if (WARN_ON(!new_policy.cur)) {
2317                         ret = -EIO;
2318                         goto unlock;
2319                 }
2320
2321                 if (!policy->cur) {
2322                         pr_debug("Driver did not initialize current freq\n");
2323                         policy->cur = new_policy.cur;
2324                 } else {
2325                         if (policy->cur != new_policy.cur && has_target())
2326                                 cpufreq_out_of_sync(policy, new_policy.cur);
2327                 }
2328         }
2329
2330         ret = cpufreq_set_policy(policy, &new_policy);
2331
2332 unlock:
2333         up_write(&policy->rwsem);
2334
2335         cpufreq_cpu_put(policy);
2336         return ret;
2337 }
2338 EXPORT_SYMBOL(cpufreq_update_policy);
2339
2340 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2341                                         unsigned long action, void *hcpu)
2342 {
2343         unsigned int cpu = (unsigned long)hcpu;
2344
2345         switch (action & ~CPU_TASKS_FROZEN) {
2346         case CPU_ONLINE:
2347                 cpufreq_online(cpu);
2348                 break;
2349
2350         case CPU_DOWN_PREPARE:
2351                 cpufreq_offline_prepare(cpu);
2352                 break;
2353
2354         case CPU_POST_DEAD:
2355                 cpufreq_offline_finish(cpu);
2356                 break;
2357
2358         case CPU_DOWN_FAILED:
2359                 cpufreq_online(cpu);
2360                 break;
2361         }
2362         return NOTIFY_OK;
2363 }
2364
2365 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2366         .notifier_call = cpufreq_cpu_callback,
2367 };
2368
2369 /*********************************************************************
2370  *               BOOST                                               *
2371  *********************************************************************/
2372 static int cpufreq_boost_set_sw(int state)
2373 {
2374         struct cpufreq_frequency_table *freq_table;
2375         struct cpufreq_policy *policy;
2376         int ret = -EINVAL;
2377
2378         for_each_active_policy(policy) {
2379                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2380                 if (freq_table) {
2381                         ret = cpufreq_frequency_table_cpuinfo(policy,
2382                                                         freq_table);
2383                         if (ret) {
2384                                 pr_err("%s: Policy frequency update failed\n",
2385                                        __func__);
2386                                 break;
2387                         }
2388                         policy->user_policy.max = policy->max;
2389                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2390                 }
2391         }
2392
2393         return ret;
2394 }
2395
2396 int cpufreq_boost_trigger_state(int state)
2397 {
2398         unsigned long flags;
2399         int ret = 0;
2400
2401         if (cpufreq_driver->boost_enabled == state)
2402                 return 0;
2403
2404         write_lock_irqsave(&cpufreq_driver_lock, flags);
2405         cpufreq_driver->boost_enabled = state;
2406         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2407
2408         ret = cpufreq_driver->set_boost(state);
2409         if (ret) {
2410                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2411                 cpufreq_driver->boost_enabled = !state;
2412                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2413
2414                 pr_err("%s: Cannot %s BOOST\n",
2415                        __func__, state ? "enable" : "disable");
2416         }
2417
2418         return ret;
2419 }
2420
2421 int cpufreq_boost_supported(void)
2422 {
2423         if (likely(cpufreq_driver))
2424                 return cpufreq_driver->boost_supported;
2425
2426         return 0;
2427 }
2428 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2429
2430 int cpufreq_boost_enabled(void)
2431 {
2432         return cpufreq_driver->boost_enabled;
2433 }
2434 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2435
2436 /*********************************************************************
2437  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2438  *********************************************************************/
2439
2440 /**
2441  * cpufreq_register_driver - register a CPU Frequency driver
2442  * @driver_data: A struct cpufreq_driver containing the values#
2443  * submitted by the CPU Frequency driver.
2444  *
2445  * Registers a CPU Frequency driver to this core code. This code
2446  * returns zero on success, -EBUSY when another driver got here first
2447  * (and isn't unregistered in the meantime).
2448  *
2449  */
2450 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2451 {
2452         unsigned long flags;
2453         int ret;
2454
2455         if (cpufreq_disabled())
2456                 return -ENODEV;
2457
2458         if (!driver_data || !driver_data->verify || !driver_data->init ||
2459             !(driver_data->setpolicy || driver_data->target_index ||
2460                     driver_data->target) ||
2461              (driver_data->setpolicy && (driver_data->target_index ||
2462                     driver_data->target)) ||
2463              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2464                 return -EINVAL;
2465
2466         pr_debug("trying to register driver %s\n", driver_data->name);
2467
2468         /* Protect against concurrent CPU online/offline. */
2469         get_online_cpus();
2470
2471         write_lock_irqsave(&cpufreq_driver_lock, flags);
2472         if (cpufreq_driver) {
2473                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2474                 ret = -EEXIST;
2475                 goto out;
2476         }
2477         cpufreq_driver = driver_data;
2478         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2479
2480         if (driver_data->setpolicy)
2481                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2482
2483         if (cpufreq_boost_supported()) {
2484                 /*
2485                  * Check if driver provides function to enable boost -
2486                  * if not, use cpufreq_boost_set_sw as default
2487                  */
2488                 if (!cpufreq_driver->set_boost)
2489                         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2490
2491                 ret = cpufreq_sysfs_create_file(&boost.attr);
2492                 if (ret) {
2493                         pr_err("%s: cannot register global BOOST sysfs file\n",
2494                                __func__);
2495                         goto err_null_driver;
2496                 }
2497         }
2498
2499         ret = subsys_interface_register(&cpufreq_interface);
2500         if (ret)
2501                 goto err_boost_unreg;
2502
2503         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2504             list_empty(&cpufreq_policy_list)) {
2505                 /* if all ->init() calls failed, unregister */
2506                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2507                          driver_data->name);
2508                 goto err_if_unreg;
2509         }
2510
2511         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2512         pr_debug("driver %s up and running\n", driver_data->name);
2513
2514 out:
2515         put_online_cpus();
2516         return ret;
2517
2518 err_if_unreg:
2519         subsys_interface_unregister(&cpufreq_interface);
2520 err_boost_unreg:
2521         if (cpufreq_boost_supported())
2522                 cpufreq_sysfs_remove_file(&boost.attr);
2523 err_null_driver:
2524         write_lock_irqsave(&cpufreq_driver_lock, flags);
2525         cpufreq_driver = NULL;
2526         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2527         goto out;
2528 }
2529 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2530
2531 /**
2532  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2533  *
2534  * Unregister the current CPUFreq driver. Only call this if you have
2535  * the right to do so, i.e. if you have succeeded in initialising before!
2536  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2537  * currently not initialised.
2538  */
2539 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2540 {
2541         unsigned long flags;
2542
2543         if (!cpufreq_driver || (driver != cpufreq_driver))
2544                 return -EINVAL;
2545
2546         pr_debug("unregistering driver %s\n", driver->name);
2547
2548         /* Protect against concurrent cpu hotplug */
2549         get_online_cpus();
2550         subsys_interface_unregister(&cpufreq_interface);
2551         if (cpufreq_boost_supported())
2552                 cpufreq_sysfs_remove_file(&boost.attr);
2553
2554         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2555
2556         write_lock_irqsave(&cpufreq_driver_lock, flags);
2557
2558         cpufreq_driver = NULL;
2559
2560         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2561         put_online_cpus();
2562
2563         return 0;
2564 }
2565 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2566
2567 /*
2568  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2569  * or mutexes when secondary CPUs are halted.
2570  */
2571 static struct syscore_ops cpufreq_syscore_ops = {
2572         .shutdown = cpufreq_suspend,
2573 };
2574
2575 static int __init cpufreq_core_init(void)
2576 {
2577         if (cpufreq_disabled())
2578                 return -ENODEV;
2579
2580         cpufreq_global_kobject = kobject_create();
2581         BUG_ON(!cpufreq_global_kobject);
2582
2583         register_syscore_ops(&cpufreq_syscore_ops);
2584
2585         return 0;
2586 }
2587 core_initcall(cpufreq_core_init);