a80bd68bbd747be4e2216ef5d3f9587edb497901
[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         return ret ? ret : count;
675 }
676
677 /**
678  * show_scaling_driver - show the cpufreq driver currently loaded
679  */
680 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
681 {
682         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
683 }
684
685 /**
686  * show_scaling_available_governors - show the available CPUfreq governors
687  */
688 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
689                                                 char *buf)
690 {
691         ssize_t i = 0;
692         struct cpufreq_governor *t;
693
694         if (!has_target()) {
695                 i += sprintf(buf, "performance powersave");
696                 goto out;
697         }
698
699         for_each_governor(t) {
700                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
701                     - (CPUFREQ_NAME_LEN + 2)))
702                         goto out;
703                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
704         }
705 out:
706         i += sprintf(&buf[i], "\n");
707         return i;
708 }
709
710 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
711 {
712         ssize_t i = 0;
713         unsigned int cpu;
714
715         for_each_cpu(cpu, mask) {
716                 if (i)
717                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
718                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
719                 if (i >= (PAGE_SIZE - 5))
720                         break;
721         }
722         i += sprintf(&buf[i], "\n");
723         return i;
724 }
725 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
726
727 /**
728  * show_related_cpus - show the CPUs affected by each transition even if
729  * hw coordination is in use
730  */
731 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
732 {
733         return cpufreq_show_cpus(policy->related_cpus, buf);
734 }
735
736 /**
737  * show_affected_cpus - show the CPUs affected by each transition
738  */
739 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
740 {
741         return cpufreq_show_cpus(policy->cpus, buf);
742 }
743
744 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
745                                         const char *buf, size_t count)
746 {
747         unsigned int freq = 0;
748         unsigned int ret;
749
750         if (!policy->governor || !policy->governor->store_setspeed)
751                 return -EINVAL;
752
753         ret = sscanf(buf, "%u", &freq);
754         if (ret != 1)
755                 return -EINVAL;
756
757         policy->governor->store_setspeed(policy, freq);
758
759         return count;
760 }
761
762 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
763 {
764         if (!policy->governor || !policy->governor->show_setspeed)
765                 return sprintf(buf, "<unsupported>\n");
766
767         return policy->governor->show_setspeed(policy, buf);
768 }
769
770 /**
771  * show_bios_limit - show the current cpufreq HW/BIOS limitation
772  */
773 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
774 {
775         unsigned int limit;
776         int ret;
777         if (cpufreq_driver->bios_limit) {
778                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
779                 if (!ret)
780                         return sprintf(buf, "%u\n", limit);
781         }
782         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
783 }
784
785 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
786 cpufreq_freq_attr_ro(cpuinfo_min_freq);
787 cpufreq_freq_attr_ro(cpuinfo_max_freq);
788 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
789 cpufreq_freq_attr_ro(scaling_available_governors);
790 cpufreq_freq_attr_ro(scaling_driver);
791 cpufreq_freq_attr_ro(scaling_cur_freq);
792 cpufreq_freq_attr_ro(bios_limit);
793 cpufreq_freq_attr_ro(related_cpus);
794 cpufreq_freq_attr_ro(affected_cpus);
795 cpufreq_freq_attr_rw(scaling_min_freq);
796 cpufreq_freq_attr_rw(scaling_max_freq);
797 cpufreq_freq_attr_rw(scaling_governor);
798 cpufreq_freq_attr_rw(scaling_setspeed);
799
800 static struct attribute *default_attrs[] = {
801         &cpuinfo_min_freq.attr,
802         &cpuinfo_max_freq.attr,
803         &cpuinfo_transition_latency.attr,
804         &scaling_min_freq.attr,
805         &scaling_max_freq.attr,
806         &affected_cpus.attr,
807         &related_cpus.attr,
808         &scaling_governor.attr,
809         &scaling_driver.attr,
810         &scaling_available_governors.attr,
811         &scaling_setspeed.attr,
812         NULL
813 };
814
815 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
816 #define to_attr(a) container_of(a, struct freq_attr, attr)
817
818 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
819 {
820         struct cpufreq_policy *policy = to_policy(kobj);
821         struct freq_attr *fattr = to_attr(attr);
822         ssize_t ret;
823
824         down_read(&policy->rwsem);
825
826         if (fattr->show)
827                 ret = fattr->show(policy, buf);
828         else
829                 ret = -EIO;
830
831         up_read(&policy->rwsem);
832
833         return ret;
834 }
835
836 static ssize_t store(struct kobject *kobj, struct attribute *attr,
837                      const char *buf, size_t count)
838 {
839         struct cpufreq_policy *policy = to_policy(kobj);
840         struct freq_attr *fattr = to_attr(attr);
841         ssize_t ret = -EINVAL;
842
843         get_online_cpus();
844
845         if (!cpu_online(policy->cpu))
846                 goto unlock;
847
848         down_write(&policy->rwsem);
849
850         /* Updating inactive policies is invalid, so avoid doing that. */
851         if (unlikely(policy_is_inactive(policy))) {
852                 ret = -EBUSY;
853                 goto unlock_policy_rwsem;
854         }
855
856         if (fattr->store)
857                 ret = fattr->store(policy, buf, count);
858         else
859                 ret = -EIO;
860
861 unlock_policy_rwsem:
862         up_write(&policy->rwsem);
863 unlock:
864         put_online_cpus();
865
866         return ret;
867 }
868
869 static void cpufreq_sysfs_release(struct kobject *kobj)
870 {
871         struct cpufreq_policy *policy = to_policy(kobj);
872         pr_debug("last reference is dropped\n");
873         complete(&policy->kobj_unregister);
874 }
875
876 static const struct sysfs_ops sysfs_ops = {
877         .show   = show,
878         .store  = store,
879 };
880
881 static struct kobj_type ktype_cpufreq = {
882         .sysfs_ops      = &sysfs_ops,
883         .default_attrs  = default_attrs,
884         .release        = cpufreq_sysfs_release,
885 };
886
887 struct kobject *cpufreq_global_kobject;
888 EXPORT_SYMBOL(cpufreq_global_kobject);
889
890 static int cpufreq_global_kobject_usage;
891
892 int cpufreq_get_global_kobject(void)
893 {
894         if (!cpufreq_global_kobject_usage++)
895                 return kobject_add(cpufreq_global_kobject,
896                                 &cpu_subsys.dev_root->kobj, "%s", "cpufreq");
897
898         return 0;
899 }
900 EXPORT_SYMBOL(cpufreq_get_global_kobject);
901
902 void cpufreq_put_global_kobject(void)
903 {
904         if (!--cpufreq_global_kobject_usage)
905                 kobject_del(cpufreq_global_kobject);
906 }
907 EXPORT_SYMBOL(cpufreq_put_global_kobject);
908
909 int cpufreq_sysfs_create_file(const struct attribute *attr)
910 {
911         int ret = cpufreq_get_global_kobject();
912
913         if (!ret) {
914                 ret = sysfs_create_file(cpufreq_global_kobject, attr);
915                 if (ret)
916                         cpufreq_put_global_kobject();
917         }
918
919         return ret;
920 }
921 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
922
923 void cpufreq_sysfs_remove_file(const struct attribute *attr)
924 {
925         sysfs_remove_file(cpufreq_global_kobject, attr);
926         cpufreq_put_global_kobject();
927 }
928 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
929
930 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
931 {
932         struct device *cpu_dev;
933
934         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
935
936         if (!policy)
937                 return 0;
938
939         cpu_dev = get_cpu_device(cpu);
940         if (WARN_ON(!cpu_dev))
941                 return 0;
942
943         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
944 }
945
946 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
947 {
948         struct device *cpu_dev;
949
950         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
951
952         cpu_dev = get_cpu_device(cpu);
953         if (WARN_ON(!cpu_dev))
954                 return;
955
956         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
957 }
958
959 /* Add/remove symlinks for all related CPUs */
960 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
961 {
962         unsigned int j;
963         int ret = 0;
964
965         /* Some related CPUs might not be present (physically hotplugged) */
966         for_each_cpu(j, policy->real_cpus) {
967                 if (j == policy->kobj_cpu)
968                         continue;
969
970                 ret = add_cpu_dev_symlink(policy, j);
971                 if (ret)
972                         break;
973         }
974
975         return ret;
976 }
977
978 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
979 {
980         unsigned int j;
981
982         /* Some related CPUs might not be present (physically hotplugged) */
983         for_each_cpu(j, policy->real_cpus) {
984                 if (j == policy->kobj_cpu)
985                         continue;
986
987                 remove_cpu_dev_symlink(policy, j);
988         }
989 }
990
991 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
992 {
993         struct freq_attr **drv_attr;
994         int ret = 0;
995
996         /* set up files for this cpu device */
997         drv_attr = cpufreq_driver->attr;
998         while (drv_attr && *drv_attr) {
999                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1000                 if (ret)
1001                         return ret;
1002                 drv_attr++;
1003         }
1004         if (cpufreq_driver->get) {
1005                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1006                 if (ret)
1007                         return ret;
1008         }
1009
1010         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1011         if (ret)
1012                 return ret;
1013
1014         if (cpufreq_driver->bios_limit) {
1015                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1016                 if (ret)
1017                         return ret;
1018         }
1019
1020         return cpufreq_add_dev_symlink(policy);
1021 }
1022
1023 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1024 {
1025         struct cpufreq_governor *gov = NULL;
1026         struct cpufreq_policy new_policy;
1027
1028         memcpy(&new_policy, policy, sizeof(*policy));
1029
1030         /* Update governor of new_policy to the governor used before hotplug */
1031         gov = find_governor(policy->last_governor);
1032         if (gov)
1033                 pr_debug("Restoring governor %s for cpu %d\n",
1034                                 policy->governor->name, policy->cpu);
1035         else
1036                 gov = CPUFREQ_DEFAULT_GOVERNOR;
1037
1038         new_policy.governor = gov;
1039
1040         /* Use the default policy if its valid. */
1041         if (cpufreq_driver->setpolicy)
1042                 cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
1043
1044         /* set default policy */
1045         return cpufreq_set_policy(policy, &new_policy);
1046 }
1047
1048 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1049 {
1050         int ret = 0;
1051
1052         /* Has this CPU been taken care of already? */
1053         if (cpumask_test_cpu(cpu, policy->cpus))
1054                 return 0;
1055
1056         if (has_target()) {
1057                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1058                 if (ret) {
1059                         pr_err("%s: Failed to stop governor\n", __func__);
1060                         return ret;
1061                 }
1062         }
1063
1064         down_write(&policy->rwsem);
1065         cpumask_set_cpu(cpu, policy->cpus);
1066         up_write(&policy->rwsem);
1067
1068         if (has_target()) {
1069                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1070                 if (!ret)
1071                         ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1072
1073                 if (ret) {
1074                         pr_err("%s: Failed to start governor\n", __func__);
1075                         return ret;
1076                 }
1077         }
1078
1079         return 0;
1080 }
1081
1082 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1083 {
1084         struct device *dev = get_cpu_device(cpu);
1085         struct cpufreq_policy *policy;
1086         int ret;
1087
1088         if (WARN_ON(!dev))
1089                 return NULL;
1090
1091         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1092         if (!policy)
1093                 return NULL;
1094
1095         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1096                 goto err_free_policy;
1097
1098         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1099                 goto err_free_cpumask;
1100
1101         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1102                 goto err_free_rcpumask;
1103
1104         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &dev->kobj,
1105                                    "cpufreq");
1106         if (ret) {
1107                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1108                 goto err_free_real_cpus;
1109         }
1110
1111         INIT_LIST_HEAD(&policy->policy_list);
1112         init_rwsem(&policy->rwsem);
1113         spin_lock_init(&policy->transition_lock);
1114         init_waitqueue_head(&policy->transition_wait);
1115         init_completion(&policy->kobj_unregister);
1116         INIT_WORK(&policy->update, handle_update);
1117
1118         policy->cpu = cpu;
1119
1120         /* Set this once on allocation */
1121         policy->kobj_cpu = cpu;
1122
1123         return policy;
1124
1125 err_free_real_cpus:
1126         free_cpumask_var(policy->real_cpus);
1127 err_free_rcpumask:
1128         free_cpumask_var(policy->related_cpus);
1129 err_free_cpumask:
1130         free_cpumask_var(policy->cpus);
1131 err_free_policy:
1132         kfree(policy);
1133
1134         return NULL;
1135 }
1136
1137 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1138 {
1139         struct kobject *kobj;
1140         struct completion *cmp;
1141
1142         if (notify)
1143                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1144                                              CPUFREQ_REMOVE_POLICY, policy);
1145
1146         down_write(&policy->rwsem);
1147         cpufreq_remove_dev_symlink(policy);
1148         kobj = &policy->kobj;
1149         cmp = &policy->kobj_unregister;
1150         up_write(&policy->rwsem);
1151         kobject_put(kobj);
1152
1153         /*
1154          * We need to make sure that the underlying kobj is
1155          * actually not referenced anymore by anybody before we
1156          * proceed with unloading.
1157          */
1158         pr_debug("waiting for dropping of refcount\n");
1159         wait_for_completion(cmp);
1160         pr_debug("wait complete\n");
1161 }
1162
1163 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1164 {
1165         unsigned long flags;
1166         int cpu;
1167
1168         /* Remove policy from list */
1169         write_lock_irqsave(&cpufreq_driver_lock, flags);
1170         list_del(&policy->policy_list);
1171
1172         for_each_cpu(cpu, policy->related_cpus)
1173                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1174         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1175
1176         cpufreq_policy_put_kobj(policy, notify);
1177         free_cpumask_var(policy->real_cpus);
1178         free_cpumask_var(policy->related_cpus);
1179         free_cpumask_var(policy->cpus);
1180         kfree(policy);
1181 }
1182
1183 static int cpufreq_online(unsigned int cpu)
1184 {
1185         struct cpufreq_policy *policy;
1186         bool new_policy;
1187         unsigned long flags;
1188         unsigned int j;
1189         int ret;
1190
1191         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1192
1193         /* Check if this CPU already has a policy to manage it */
1194         policy = per_cpu(cpufreq_cpu_data, cpu);
1195         if (policy) {
1196                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1197                 if (!policy_is_inactive(policy))
1198                         return cpufreq_add_policy_cpu(policy, cpu);
1199
1200                 /* This is the only online CPU for the policy.  Start over. */
1201                 new_policy = false;
1202                 down_write(&policy->rwsem);
1203                 policy->cpu = cpu;
1204                 policy->governor = NULL;
1205                 up_write(&policy->rwsem);
1206         } else {
1207                 new_policy = true;
1208                 policy = cpufreq_policy_alloc(cpu);
1209                 if (!policy)
1210                         return -ENOMEM;
1211         }
1212
1213         cpumask_copy(policy->cpus, cpumask_of(cpu));
1214
1215         /* call driver. From then on the cpufreq must be able
1216          * to accept all calls to ->verify and ->setpolicy for this CPU
1217          */
1218         ret = cpufreq_driver->init(policy);
1219         if (ret) {
1220                 pr_debug("initialization failed\n");
1221                 goto out_free_policy;
1222         }
1223
1224         down_write(&policy->rwsem);
1225
1226         if (new_policy) {
1227                 /* related_cpus should at least include policy->cpus. */
1228                 cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1229                 /* Remember CPUs present at the policy creation time. */
1230                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1231         }
1232
1233         /*
1234          * affected cpus must always be the one, which are online. We aren't
1235          * managing offline cpus here.
1236          */
1237         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1238
1239         if (new_policy) {
1240                 policy->user_policy.min = policy->min;
1241                 policy->user_policy.max = policy->max;
1242
1243                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1244                 for_each_cpu(j, policy->related_cpus)
1245                         per_cpu(cpufreq_cpu_data, j) = policy;
1246                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1247         }
1248
1249         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1250                 policy->cur = cpufreq_driver->get(policy->cpu);
1251                 if (!policy->cur) {
1252                         pr_err("%s: ->get() failed\n", __func__);
1253                         goto out_exit_policy;
1254                 }
1255         }
1256
1257         /*
1258          * Sometimes boot loaders set CPU frequency to a value outside of
1259          * frequency table present with cpufreq core. In such cases CPU might be
1260          * unstable if it has to run on that frequency for long duration of time
1261          * and so its better to set it to a frequency which is specified in
1262          * freq-table. This also makes cpufreq stats inconsistent as
1263          * cpufreq-stats would fail to register because current frequency of CPU
1264          * isn't found in freq-table.
1265          *
1266          * Because we don't want this change to effect boot process badly, we go
1267          * for the next freq which is >= policy->cur ('cur' must be set by now,
1268          * otherwise we will end up setting freq to lowest of the table as 'cur'
1269          * is initialized to zero).
1270          *
1271          * We are passing target-freq as "policy->cur - 1" otherwise
1272          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1273          * equal to target-freq.
1274          */
1275         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1276             && has_target()) {
1277                 /* Are we running at unknown frequency ? */
1278                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1279                 if (ret == -EINVAL) {
1280                         /* Warn user and fix it */
1281                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1282                                 __func__, policy->cpu, policy->cur);
1283                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1284                                 CPUFREQ_RELATION_L);
1285
1286                         /*
1287                          * Reaching here after boot in a few seconds may not
1288                          * mean that system will remain stable at "unknown"
1289                          * frequency for longer duration. Hence, a BUG_ON().
1290                          */
1291                         BUG_ON(ret);
1292                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1293                                 __func__, policy->cpu, policy->cur);
1294                 }
1295         }
1296
1297         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1298                                      CPUFREQ_START, policy);
1299
1300         if (new_policy) {
1301                 ret = cpufreq_add_dev_interface(policy);
1302                 if (ret)
1303                         goto out_exit_policy;
1304                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1305                                 CPUFREQ_CREATE_POLICY, policy);
1306
1307                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1308                 list_add(&policy->policy_list, &cpufreq_policy_list);
1309                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1310         }
1311
1312         ret = cpufreq_init_policy(policy);
1313         if (ret) {
1314                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1315                        __func__, cpu, ret);
1316                 /* cpufreq_policy_free() will notify based on this */
1317                 new_policy = false;
1318                 goto out_exit_policy;
1319         }
1320
1321         up_write(&policy->rwsem);
1322
1323         kobject_uevent(&policy->kobj, KOBJ_ADD);
1324
1325         /* Callback for handling stuff after policy is ready */
1326         if (cpufreq_driver->ready)
1327                 cpufreq_driver->ready(policy);
1328
1329         pr_debug("initialization complete\n");
1330
1331         return 0;
1332
1333 out_exit_policy:
1334         up_write(&policy->rwsem);
1335
1336         if (cpufreq_driver->exit)
1337                 cpufreq_driver->exit(policy);
1338 out_free_policy:
1339         cpufreq_policy_free(policy, !new_policy);
1340         return ret;
1341 }
1342
1343 /**
1344  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1345  * @dev: CPU device.
1346  * @sif: Subsystem interface structure pointer (not used)
1347  */
1348 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1349 {
1350         unsigned cpu = dev->id;
1351         int ret;
1352
1353         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1354
1355         if (cpu_online(cpu)) {
1356                 ret = cpufreq_online(cpu);
1357         } else {
1358                 /*
1359                  * A hotplug notifier will follow and we will handle it as CPU
1360                  * online then.  For now, just create the sysfs link, unless
1361                  * there is no policy or the link is already present.
1362                  */
1363                 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1364
1365                 ret = policy && !cpumask_test_and_set_cpu(cpu, policy->real_cpus)
1366                         ? add_cpu_dev_symlink(policy, cpu) : 0;
1367         }
1368
1369         return ret;
1370 }
1371
1372 static void cpufreq_offline_prepare(unsigned int cpu)
1373 {
1374         struct cpufreq_policy *policy;
1375
1376         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1377
1378         policy = cpufreq_cpu_get_raw(cpu);
1379         if (!policy) {
1380                 pr_debug("%s: No cpu_data found\n", __func__);
1381                 return;
1382         }
1383
1384         if (has_target()) {
1385                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1386                 if (ret)
1387                         pr_err("%s: Failed to stop governor\n", __func__);
1388         }
1389
1390         down_write(&policy->rwsem);
1391         cpumask_clear_cpu(cpu, policy->cpus);
1392
1393         if (policy_is_inactive(policy)) {
1394                 if (has_target())
1395                         strncpy(policy->last_governor, policy->governor->name,
1396                                 CPUFREQ_NAME_LEN);
1397         } else if (cpu == policy->cpu) {
1398                 /* Nominate new CPU */
1399                 policy->cpu = cpumask_any(policy->cpus);
1400         }
1401         up_write(&policy->rwsem);
1402
1403         /* Start governor again for active policy */
1404         if (!policy_is_inactive(policy)) {
1405                 if (has_target()) {
1406                         int ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
1407                         if (!ret)
1408                                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
1409
1410                         if (ret)
1411                                 pr_err("%s: Failed to start governor\n", __func__);
1412                 }
1413         } else if (cpufreq_driver->stop_cpu) {
1414                 cpufreq_driver->stop_cpu(policy);
1415         }
1416 }
1417
1418 static void cpufreq_offline_finish(unsigned int cpu)
1419 {
1420         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1421
1422         if (!policy) {
1423                 pr_debug("%s: No cpu_data found\n", __func__);
1424                 return;
1425         }
1426
1427         /* Only proceed for inactive policies */
1428         if (!policy_is_inactive(policy))
1429                 return;
1430
1431         /* If cpu is last user of policy, free policy */
1432         if (has_target()) {
1433                 int ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
1434                 if (ret)
1435                         pr_err("%s: Failed to exit governor\n", __func__);
1436         }
1437
1438         /*
1439          * Perform the ->exit() even during light-weight tear-down,
1440          * since this is a core component, and is essential for the
1441          * subsequent light-weight ->init() to succeed.
1442          */
1443         if (cpufreq_driver->exit)
1444                 cpufreq_driver->exit(policy);
1445 }
1446
1447 /**
1448  * cpufreq_remove_dev - remove a CPU device
1449  *
1450  * Removes the cpufreq interface for a CPU device.
1451  */
1452 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1453 {
1454         unsigned int cpu = dev->id;
1455         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1456
1457         if (!policy)
1458                 return 0;
1459
1460         if (cpu_online(cpu)) {
1461                 cpufreq_offline_prepare(cpu);
1462                 cpufreq_offline_finish(cpu);
1463         }
1464
1465         cpumask_clear_cpu(cpu, policy->real_cpus);
1466
1467         if (cpumask_empty(policy->real_cpus)) {
1468                 cpufreq_policy_free(policy, true);
1469                 return 0;
1470         }
1471
1472         if (cpu != policy->kobj_cpu) {
1473                 remove_cpu_dev_symlink(policy, cpu);
1474         } else {
1475                 /*
1476                  * The CPU owning the policy object is going away.  Move it to
1477                  * another suitable CPU.
1478                  */
1479                 unsigned int new_cpu = cpumask_first(policy->real_cpus);
1480                 struct device *new_dev = get_cpu_device(new_cpu);
1481
1482                 dev_dbg(dev, "%s: Moving policy object to CPU%u\n", __func__, new_cpu);
1483
1484                 sysfs_remove_link(&new_dev->kobj, "cpufreq");
1485                 policy->kobj_cpu = new_cpu;
1486                 WARN_ON(kobject_move(&policy->kobj, &new_dev->kobj));
1487         }
1488
1489         return 0;
1490 }
1491
1492 static void handle_update(struct work_struct *work)
1493 {
1494         struct cpufreq_policy *policy =
1495                 container_of(work, struct cpufreq_policy, update);
1496         unsigned int cpu = policy->cpu;
1497         pr_debug("handle_update for cpu %u called\n", cpu);
1498         cpufreq_update_policy(cpu);
1499 }
1500
1501 /**
1502  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1503  *      in deep trouble.
1504  *      @policy: policy managing CPUs
1505  *      @new_freq: CPU frequency the CPU actually runs at
1506  *
1507  *      We adjust to current frequency first, and need to clean up later.
1508  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1509  */
1510 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1511                                 unsigned int new_freq)
1512 {
1513         struct cpufreq_freqs freqs;
1514
1515         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1516                  policy->cur, new_freq);
1517
1518         freqs.old = policy->cur;
1519         freqs.new = new_freq;
1520
1521         cpufreq_freq_transition_begin(policy, &freqs);
1522         cpufreq_freq_transition_end(policy, &freqs, 0);
1523 }
1524
1525 /**
1526  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1527  * @cpu: CPU number
1528  *
1529  * This is the last known freq, without actually getting it from the driver.
1530  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1531  */
1532 unsigned int cpufreq_quick_get(unsigned int cpu)
1533 {
1534         struct cpufreq_policy *policy;
1535         unsigned int ret_freq = 0;
1536
1537         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1538                 return cpufreq_driver->get(cpu);
1539
1540         policy = cpufreq_cpu_get(cpu);
1541         if (policy) {
1542                 ret_freq = policy->cur;
1543                 cpufreq_cpu_put(policy);
1544         }
1545
1546         return ret_freq;
1547 }
1548 EXPORT_SYMBOL(cpufreq_quick_get);
1549
1550 /**
1551  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1552  * @cpu: CPU number
1553  *
1554  * Just return the max possible frequency for a given CPU.
1555  */
1556 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1557 {
1558         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1559         unsigned int ret_freq = 0;
1560
1561         if (policy) {
1562                 ret_freq = policy->max;
1563                 cpufreq_cpu_put(policy);
1564         }
1565
1566         return ret_freq;
1567 }
1568 EXPORT_SYMBOL(cpufreq_quick_get_max);
1569
1570 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1571 {
1572         unsigned int ret_freq = 0;
1573
1574         if (!cpufreq_driver->get)
1575                 return ret_freq;
1576
1577         ret_freq = cpufreq_driver->get(policy->cpu);
1578
1579         /* Updating inactive policies is invalid, so avoid doing that. */
1580         if (unlikely(policy_is_inactive(policy)))
1581                 return ret_freq;
1582
1583         if (ret_freq && policy->cur &&
1584                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1585                 /* verify no discrepancy between actual and
1586                                         saved value exists */
1587                 if (unlikely(ret_freq != policy->cur)) {
1588                         cpufreq_out_of_sync(policy, ret_freq);
1589                         schedule_work(&policy->update);
1590                 }
1591         }
1592
1593         return ret_freq;
1594 }
1595
1596 /**
1597  * cpufreq_get - get the current CPU frequency (in kHz)
1598  * @cpu: CPU number
1599  *
1600  * Get the CPU current (static) CPU frequency
1601  */
1602 unsigned int cpufreq_get(unsigned int cpu)
1603 {
1604         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1605         unsigned int ret_freq = 0;
1606
1607         if (policy) {
1608                 down_read(&policy->rwsem);
1609                 ret_freq = __cpufreq_get(policy);
1610                 up_read(&policy->rwsem);
1611
1612                 cpufreq_cpu_put(policy);
1613         }
1614
1615         return ret_freq;
1616 }
1617 EXPORT_SYMBOL(cpufreq_get);
1618
1619 static struct subsys_interface cpufreq_interface = {
1620         .name           = "cpufreq",
1621         .subsys         = &cpu_subsys,
1622         .add_dev        = cpufreq_add_dev,
1623         .remove_dev     = cpufreq_remove_dev,
1624 };
1625
1626 /*
1627  * In case platform wants some specific frequency to be configured
1628  * during suspend..
1629  */
1630 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1631 {
1632         int ret;
1633
1634         if (!policy->suspend_freq) {
1635                 pr_err("%s: suspend_freq can't be zero\n", __func__);
1636                 return -EINVAL;
1637         }
1638
1639         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1640                         policy->suspend_freq);
1641
1642         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1643                         CPUFREQ_RELATION_H);
1644         if (ret)
1645                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1646                                 __func__, policy->suspend_freq, ret);
1647
1648         return ret;
1649 }
1650 EXPORT_SYMBOL(cpufreq_generic_suspend);
1651
1652 /**
1653  * cpufreq_suspend() - Suspend CPUFreq governors
1654  *
1655  * Called during system wide Suspend/Hibernate cycles for suspending governors
1656  * as some platforms can't change frequency after this point in suspend cycle.
1657  * Because some of the devices (like: i2c, regulators, etc) they use for
1658  * changing frequency are suspended quickly after this point.
1659  */
1660 void cpufreq_suspend(void)
1661 {
1662         struct cpufreq_policy *policy;
1663
1664         if (!cpufreq_driver)
1665                 return;
1666
1667         if (!has_target())
1668                 goto suspend;
1669
1670         pr_debug("%s: Suspending Governors\n", __func__);
1671
1672         for_each_active_policy(policy) {
1673                 if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
1674                         pr_err("%s: Failed to stop governor for policy: %p\n",
1675                                 __func__, policy);
1676                 else if (cpufreq_driver->suspend
1677                     && cpufreq_driver->suspend(policy))
1678                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1679                                 policy);
1680         }
1681
1682 suspend:
1683         cpufreq_suspended = true;
1684 }
1685
1686 /**
1687  * cpufreq_resume() - Resume CPUFreq governors
1688  *
1689  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1690  * are suspended with cpufreq_suspend().
1691  */
1692 void cpufreq_resume(void)
1693 {
1694         struct cpufreq_policy *policy;
1695
1696         if (!cpufreq_driver)
1697                 return;
1698
1699         cpufreq_suspended = false;
1700
1701         if (!has_target())
1702                 return;
1703
1704         pr_debug("%s: Resuming Governors\n", __func__);
1705
1706         for_each_active_policy(policy) {
1707                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
1708                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1709                                 policy);
1710                 else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
1711                     || __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
1712                         pr_err("%s: Failed to start governor for policy: %p\n",
1713                                 __func__, policy);
1714         }
1715
1716         /*
1717          * schedule call cpufreq_update_policy() for first-online CPU, as that
1718          * wouldn't be hotplugged-out on suspend. It will verify that the
1719          * current freq is in sync with what we believe it to be.
1720          */
1721         policy = cpufreq_cpu_get_raw(cpumask_first(cpu_online_mask));
1722         if (WARN_ON(!policy))
1723                 return;
1724
1725         schedule_work(&policy->update);
1726 }
1727
1728 /**
1729  *      cpufreq_get_current_driver - return current driver's name
1730  *
1731  *      Return the name string of the currently loaded cpufreq driver
1732  *      or NULL, if none.
1733  */
1734 const char *cpufreq_get_current_driver(void)
1735 {
1736         if (cpufreq_driver)
1737                 return cpufreq_driver->name;
1738
1739         return NULL;
1740 }
1741 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1742
1743 /**
1744  *      cpufreq_get_driver_data - return current driver data
1745  *
1746  *      Return the private data of the currently loaded cpufreq
1747  *      driver, or NULL if no cpufreq driver is loaded.
1748  */
1749 void *cpufreq_get_driver_data(void)
1750 {
1751         if (cpufreq_driver)
1752                 return cpufreq_driver->driver_data;
1753
1754         return NULL;
1755 }
1756 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1757
1758 /*********************************************************************
1759  *                     NOTIFIER LISTS INTERFACE                      *
1760  *********************************************************************/
1761
1762 /**
1763  *      cpufreq_register_notifier - register a driver with cpufreq
1764  *      @nb: notifier function to register
1765  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1766  *
1767  *      Add a driver to one of two lists: either a list of drivers that
1768  *      are notified about clock rate changes (once before and once after
1769  *      the transition), or a list of drivers that are notified about
1770  *      changes in cpufreq policy.
1771  *
1772  *      This function may sleep, and has the same return conditions as
1773  *      blocking_notifier_chain_register.
1774  */
1775 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1776 {
1777         int ret;
1778
1779         if (cpufreq_disabled())
1780                 return -EINVAL;
1781
1782         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1783
1784         switch (list) {
1785         case CPUFREQ_TRANSITION_NOTIFIER:
1786                 ret = srcu_notifier_chain_register(
1787                                 &cpufreq_transition_notifier_list, nb);
1788                 break;
1789         case CPUFREQ_POLICY_NOTIFIER:
1790                 ret = blocking_notifier_chain_register(
1791                                 &cpufreq_policy_notifier_list, nb);
1792                 break;
1793         default:
1794                 ret = -EINVAL;
1795         }
1796
1797         return ret;
1798 }
1799 EXPORT_SYMBOL(cpufreq_register_notifier);
1800
1801 /**
1802  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1803  *      @nb: notifier block to be unregistered
1804  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1805  *
1806  *      Remove a driver from the CPU frequency notifier list.
1807  *
1808  *      This function may sleep, and has the same return conditions as
1809  *      blocking_notifier_chain_unregister.
1810  */
1811 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1812 {
1813         int ret;
1814
1815         if (cpufreq_disabled())
1816                 return -EINVAL;
1817
1818         switch (list) {
1819         case CPUFREQ_TRANSITION_NOTIFIER:
1820                 ret = srcu_notifier_chain_unregister(
1821                                 &cpufreq_transition_notifier_list, nb);
1822                 break;
1823         case CPUFREQ_POLICY_NOTIFIER:
1824                 ret = blocking_notifier_chain_unregister(
1825                                 &cpufreq_policy_notifier_list, nb);
1826                 break;
1827         default:
1828                 ret = -EINVAL;
1829         }
1830
1831         return ret;
1832 }
1833 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1834
1835
1836 /*********************************************************************
1837  *                              GOVERNORS                            *
1838  *********************************************************************/
1839
1840 /* Must set freqs->new to intermediate frequency */
1841 static int __target_intermediate(struct cpufreq_policy *policy,
1842                                  struct cpufreq_freqs *freqs, int index)
1843 {
1844         int ret;
1845
1846         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1847
1848         /* We don't need to switch to intermediate freq */
1849         if (!freqs->new)
1850                 return 0;
1851
1852         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1853                  __func__, policy->cpu, freqs->old, freqs->new);
1854
1855         cpufreq_freq_transition_begin(policy, freqs);
1856         ret = cpufreq_driver->target_intermediate(policy, index);
1857         cpufreq_freq_transition_end(policy, freqs, ret);
1858
1859         if (ret)
1860                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1861                        __func__, ret);
1862
1863         return ret;
1864 }
1865
1866 static int __target_index(struct cpufreq_policy *policy,
1867                           struct cpufreq_frequency_table *freq_table, int index)
1868 {
1869         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1870         unsigned int intermediate_freq = 0;
1871         int retval = -EINVAL;
1872         bool notify;
1873
1874         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1875         if (notify) {
1876                 /* Handle switching to intermediate frequency */
1877                 if (cpufreq_driver->get_intermediate) {
1878                         retval = __target_intermediate(policy, &freqs, index);
1879                         if (retval)
1880                                 return retval;
1881
1882                         intermediate_freq = freqs.new;
1883                         /* Set old freq to intermediate */
1884                         if (intermediate_freq)
1885                                 freqs.old = freqs.new;
1886                 }
1887
1888                 freqs.new = freq_table[index].frequency;
1889                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1890                          __func__, policy->cpu, freqs.old, freqs.new);
1891
1892                 cpufreq_freq_transition_begin(policy, &freqs);
1893         }
1894
1895         retval = cpufreq_driver->target_index(policy, index);
1896         if (retval)
1897                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1898                        retval);
1899
1900         if (notify) {
1901                 cpufreq_freq_transition_end(policy, &freqs, retval);
1902
1903                 /*
1904                  * Failed after setting to intermediate freq? Driver should have
1905                  * reverted back to initial frequency and so should we. Check
1906                  * here for intermediate_freq instead of get_intermediate, in
1907                  * case we haven't switched to intermediate freq at all.
1908                  */
1909                 if (unlikely(retval && intermediate_freq)) {
1910                         freqs.old = intermediate_freq;
1911                         freqs.new = policy->restore_freq;
1912                         cpufreq_freq_transition_begin(policy, &freqs);
1913                         cpufreq_freq_transition_end(policy, &freqs, 0);
1914                 }
1915         }
1916
1917         return retval;
1918 }
1919
1920 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1921                             unsigned int target_freq,
1922                             unsigned int relation)
1923 {
1924         unsigned int old_target_freq = target_freq;
1925         int retval = -EINVAL;
1926
1927         if (cpufreq_disabled())
1928                 return -ENODEV;
1929
1930         /* Make sure that target_freq is within supported range */
1931         if (target_freq > policy->max)
1932                 target_freq = policy->max;
1933         if (target_freq < policy->min)
1934                 target_freq = policy->min;
1935
1936         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1937                  policy->cpu, target_freq, relation, old_target_freq);
1938
1939         /*
1940          * This might look like a redundant call as we are checking it again
1941          * after finding index. But it is left intentionally for cases where
1942          * exactly same freq is called again and so we can save on few function
1943          * calls.
1944          */
1945         if (target_freq == policy->cur)
1946                 return 0;
1947
1948         /* Save last value to restore later on errors */
1949         policy->restore_freq = policy->cur;
1950
1951         if (cpufreq_driver->target)
1952                 retval = cpufreq_driver->target(policy, target_freq, relation);
1953         else if (cpufreq_driver->target_index) {
1954                 struct cpufreq_frequency_table *freq_table;
1955                 int index;
1956
1957                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1958                 if (unlikely(!freq_table)) {
1959                         pr_err("%s: Unable to find freq_table\n", __func__);
1960                         goto out;
1961                 }
1962
1963                 retval = cpufreq_frequency_table_target(policy, freq_table,
1964                                 target_freq, relation, &index);
1965                 if (unlikely(retval)) {
1966                         pr_err("%s: Unable to find matching freq\n", __func__);
1967                         goto out;
1968                 }
1969
1970                 if (freq_table[index].frequency == policy->cur) {
1971                         retval = 0;
1972                         goto out;
1973                 }
1974
1975                 retval = __target_index(policy, freq_table, index);
1976         }
1977
1978 out:
1979         return retval;
1980 }
1981 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1982
1983 int cpufreq_driver_target(struct cpufreq_policy *policy,
1984                           unsigned int target_freq,
1985                           unsigned int relation)
1986 {
1987         int ret = -EINVAL;
1988
1989         down_write(&policy->rwsem);
1990
1991         ret = __cpufreq_driver_target(policy, target_freq, relation);
1992
1993         up_write(&policy->rwsem);
1994
1995         return ret;
1996 }
1997 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1998
1999 static int __cpufreq_governor(struct cpufreq_policy *policy,
2000                                         unsigned int event)
2001 {
2002         int ret;
2003
2004         /* Only must be defined when default governor is known to have latency
2005            restrictions, like e.g. conservative or ondemand.
2006            That this is the case is already ensured in Kconfig
2007         */
2008 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
2009         struct cpufreq_governor *gov = &cpufreq_gov_performance;
2010 #else
2011         struct cpufreq_governor *gov = NULL;
2012 #endif
2013
2014         /* Don't start any governor operations if we are entering suspend */
2015         if (cpufreq_suspended)
2016                 return 0;
2017         /*
2018          * Governor might not be initiated here if ACPI _PPC changed
2019          * notification happened, so check it.
2020          */
2021         if (!policy->governor)
2022                 return -EINVAL;
2023
2024         if (policy->governor->max_transition_latency &&
2025             policy->cpuinfo.transition_latency >
2026             policy->governor->max_transition_latency) {
2027                 if (!gov)
2028                         return -EINVAL;
2029                 else {
2030                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2031                                 policy->governor->name, gov->name);
2032                         policy->governor = gov;
2033                 }
2034         }
2035
2036         if (event == CPUFREQ_GOV_POLICY_INIT)
2037                 if (!try_module_get(policy->governor->owner))
2038                         return -EINVAL;
2039
2040         pr_debug("__cpufreq_governor for CPU %u, event %u\n",
2041                  policy->cpu, event);
2042
2043         mutex_lock(&cpufreq_governor_lock);
2044         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
2045             || (!policy->governor_enabled
2046             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
2047                 mutex_unlock(&cpufreq_governor_lock);
2048                 return -EBUSY;
2049         }
2050
2051         if (event == CPUFREQ_GOV_STOP)
2052                 policy->governor_enabled = false;
2053         else if (event == CPUFREQ_GOV_START)
2054                 policy->governor_enabled = true;
2055
2056         mutex_unlock(&cpufreq_governor_lock);
2057
2058         ret = policy->governor->governor(policy, event);
2059
2060         if (!ret) {
2061                 if (event == CPUFREQ_GOV_POLICY_INIT)
2062                         policy->governor->initialized++;
2063                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
2064                         policy->governor->initialized--;
2065         } else {
2066                 /* Restore original values */
2067                 mutex_lock(&cpufreq_governor_lock);
2068                 if (event == CPUFREQ_GOV_STOP)
2069                         policy->governor_enabled = true;
2070                 else if (event == CPUFREQ_GOV_START)
2071                         policy->governor_enabled = false;
2072                 mutex_unlock(&cpufreq_governor_lock);
2073         }
2074
2075         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
2076                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
2077                 module_put(policy->governor->owner);
2078
2079         return ret;
2080 }
2081
2082 int cpufreq_register_governor(struct cpufreq_governor *governor)
2083 {
2084         int err;
2085
2086         if (!governor)
2087                 return -EINVAL;
2088
2089         if (cpufreq_disabled())
2090                 return -ENODEV;
2091
2092         mutex_lock(&cpufreq_governor_mutex);
2093
2094         governor->initialized = 0;
2095         err = -EBUSY;
2096         if (!find_governor(governor->name)) {
2097                 err = 0;
2098                 list_add(&governor->governor_list, &cpufreq_governor_list);
2099         }
2100
2101         mutex_unlock(&cpufreq_governor_mutex);
2102         return err;
2103 }
2104 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2105
2106 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2107 {
2108         struct cpufreq_policy *policy;
2109         unsigned long flags;
2110
2111         if (!governor)
2112                 return;
2113
2114         if (cpufreq_disabled())
2115                 return;
2116
2117         /* clear last_governor for all inactive policies */
2118         read_lock_irqsave(&cpufreq_driver_lock, flags);
2119         for_each_inactive_policy(policy) {
2120                 if (!strcmp(policy->last_governor, governor->name)) {
2121                         policy->governor = NULL;
2122                         strcpy(policy->last_governor, "\0");
2123                 }
2124         }
2125         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2126
2127         mutex_lock(&cpufreq_governor_mutex);
2128         list_del(&governor->governor_list);
2129         mutex_unlock(&cpufreq_governor_mutex);
2130         return;
2131 }
2132 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2133
2134
2135 /*********************************************************************
2136  *                          POLICY INTERFACE                         *
2137  *********************************************************************/
2138
2139 /**
2140  * cpufreq_get_policy - get the current cpufreq_policy
2141  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2142  *      is written
2143  *
2144  * Reads the current cpufreq policy.
2145  */
2146 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2147 {
2148         struct cpufreq_policy *cpu_policy;
2149         if (!policy)
2150                 return -EINVAL;
2151
2152         cpu_policy = cpufreq_cpu_get(cpu);
2153         if (!cpu_policy)
2154                 return -EINVAL;
2155
2156         memcpy(policy, cpu_policy, sizeof(*policy));
2157
2158         cpufreq_cpu_put(cpu_policy);
2159         return 0;
2160 }
2161 EXPORT_SYMBOL(cpufreq_get_policy);
2162
2163 /*
2164  * policy : current policy.
2165  * new_policy: policy to be set.
2166  */
2167 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2168                                 struct cpufreq_policy *new_policy)
2169 {
2170         struct cpufreq_governor *old_gov;
2171         int ret;
2172
2173         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2174                  new_policy->cpu, new_policy->min, new_policy->max);
2175
2176         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2177
2178         /*
2179         * This check works well when we store new min/max freq attributes,
2180         * because new_policy is a copy of policy with one field updated.
2181         */
2182         if (new_policy->min > new_policy->max)
2183                 return -EINVAL;
2184
2185         /* verify the cpu speed can be set within this limit */
2186         ret = cpufreq_driver->verify(new_policy);
2187         if (ret)
2188                 return ret;
2189
2190         /* adjust if necessary - all reasons */
2191         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2192                         CPUFREQ_ADJUST, new_policy);
2193
2194         /*
2195          * verify the cpu speed can be set within this limit, which might be
2196          * different to the first one
2197          */
2198         ret = cpufreq_driver->verify(new_policy);
2199         if (ret)
2200                 return ret;
2201
2202         /* notification of the new policy */
2203         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2204                         CPUFREQ_NOTIFY, new_policy);
2205
2206         policy->min = new_policy->min;
2207         policy->max = new_policy->max;
2208
2209         pr_debug("new min and max freqs are %u - %u kHz\n",
2210                  policy->min, policy->max);
2211
2212         if (cpufreq_driver->setpolicy) {
2213                 policy->policy = new_policy->policy;
2214                 pr_debug("setting range\n");
2215                 return cpufreq_driver->setpolicy(new_policy);
2216         }
2217
2218         if (new_policy->governor == policy->governor)
2219                 goto out;
2220
2221         pr_debug("governor switch\n");
2222
2223         /* save old, working values */
2224         old_gov = policy->governor;
2225         /* end old governor */
2226         if (old_gov) {
2227                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2228                 if (ret) {
2229                         /* This can happen due to race with other operations */
2230                         pr_debug("%s: Failed to Stop Governor: %s (%d)\n",
2231                                  __func__, old_gov->name, ret);
2232                         return ret;
2233                 }
2234
2235                 up_write(&policy->rwsem);
2236                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2237                 down_write(&policy->rwsem);
2238
2239                 if (ret) {
2240                         pr_err("%s: Failed to Exit Governor: %s (%d)\n",
2241                                __func__, old_gov->name, ret);
2242                         return ret;
2243                 }
2244         }
2245
2246         /* start new governor */
2247         policy->governor = new_policy->governor;
2248         ret = __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
2249         if (!ret) {
2250                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
2251                 if (!ret)
2252                         goto out;
2253
2254                 up_write(&policy->rwsem);
2255                 __cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
2256                 down_write(&policy->rwsem);
2257         }
2258
2259         /* new governor failed, so re-start old one */
2260         pr_debug("starting governor %s failed\n", policy->governor->name);
2261         if (old_gov) {
2262                 policy->governor = old_gov;
2263                 if (__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT))
2264                         policy->governor = NULL;
2265                 else
2266                         __cpufreq_governor(policy, CPUFREQ_GOV_START);
2267         }
2268
2269         return ret;
2270
2271  out:
2272         pr_debug("governor: change or update limits\n");
2273         return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2274 }
2275
2276 /**
2277  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2278  *      @cpu: CPU which shall be re-evaluated
2279  *
2280  *      Useful for policy notifiers which have different necessities
2281  *      at different times.
2282  */
2283 int cpufreq_update_policy(unsigned int cpu)
2284 {
2285         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2286         struct cpufreq_policy new_policy;
2287         int ret;
2288
2289         if (!policy)
2290                 return -ENODEV;
2291
2292         down_write(&policy->rwsem);
2293
2294         pr_debug("updating policy for CPU %u\n", cpu);
2295         memcpy(&new_policy, policy, sizeof(*policy));
2296         new_policy.min = policy->user_policy.min;
2297         new_policy.max = policy->user_policy.max;
2298
2299         /*
2300          * BIOS might change freq behind our back
2301          * -> ask driver for current freq and notify governors about a change
2302          */
2303         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2304                 new_policy.cur = cpufreq_driver->get(cpu);
2305                 if (WARN_ON(!new_policy.cur)) {
2306                         ret = -EIO;
2307                         goto unlock;
2308                 }
2309
2310                 if (!policy->cur) {
2311                         pr_debug("Driver did not initialize current freq\n");
2312                         policy->cur = new_policy.cur;
2313                 } else {
2314                         if (policy->cur != new_policy.cur && has_target())
2315                                 cpufreq_out_of_sync(policy, new_policy.cur);
2316                 }
2317         }
2318
2319         ret = cpufreq_set_policy(policy, &new_policy);
2320
2321 unlock:
2322         up_write(&policy->rwsem);
2323
2324         cpufreq_cpu_put(policy);
2325         return ret;
2326 }
2327 EXPORT_SYMBOL(cpufreq_update_policy);
2328
2329 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2330                                         unsigned long action, void *hcpu)
2331 {
2332         unsigned int cpu = (unsigned long)hcpu;
2333
2334         switch (action & ~CPU_TASKS_FROZEN) {
2335         case CPU_ONLINE:
2336                 cpufreq_online(cpu);
2337                 break;
2338
2339         case CPU_DOWN_PREPARE:
2340                 cpufreq_offline_prepare(cpu);
2341                 break;
2342
2343         case CPU_POST_DEAD:
2344                 cpufreq_offline_finish(cpu);
2345                 break;
2346
2347         case CPU_DOWN_FAILED:
2348                 cpufreq_online(cpu);
2349                 break;
2350         }
2351         return NOTIFY_OK;
2352 }
2353
2354 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2355         .notifier_call = cpufreq_cpu_callback,
2356 };
2357
2358 /*********************************************************************
2359  *               BOOST                                               *
2360  *********************************************************************/
2361 static int cpufreq_boost_set_sw(int state)
2362 {
2363         struct cpufreq_frequency_table *freq_table;
2364         struct cpufreq_policy *policy;
2365         int ret = -EINVAL;
2366
2367         for_each_active_policy(policy) {
2368                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2369                 if (freq_table) {
2370                         ret = cpufreq_frequency_table_cpuinfo(policy,
2371                                                         freq_table);
2372                         if (ret) {
2373                                 pr_err("%s: Policy frequency update failed\n",
2374                                        __func__);
2375                                 break;
2376                         }
2377                         policy->user_policy.max = policy->max;
2378                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2379                 }
2380         }
2381
2382         return ret;
2383 }
2384
2385 int cpufreq_boost_trigger_state(int state)
2386 {
2387         unsigned long flags;
2388         int ret = 0;
2389
2390         if (cpufreq_driver->boost_enabled == state)
2391                 return 0;
2392
2393         write_lock_irqsave(&cpufreq_driver_lock, flags);
2394         cpufreq_driver->boost_enabled = state;
2395         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2396
2397         ret = cpufreq_driver->set_boost(state);
2398         if (ret) {
2399                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2400                 cpufreq_driver->boost_enabled = !state;
2401                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2402
2403                 pr_err("%s: Cannot %s BOOST\n",
2404                        __func__, state ? "enable" : "disable");
2405         }
2406
2407         return ret;
2408 }
2409
2410 int cpufreq_boost_supported(void)
2411 {
2412         if (likely(cpufreq_driver))
2413                 return cpufreq_driver->boost_supported;
2414
2415         return 0;
2416 }
2417 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2418
2419 int cpufreq_boost_enabled(void)
2420 {
2421         return cpufreq_driver->boost_enabled;
2422 }
2423 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2424
2425 /*********************************************************************
2426  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2427  *********************************************************************/
2428
2429 /**
2430  * cpufreq_register_driver - register a CPU Frequency driver
2431  * @driver_data: A struct cpufreq_driver containing the values#
2432  * submitted by the CPU Frequency driver.
2433  *
2434  * Registers a CPU Frequency driver to this core code. This code
2435  * returns zero on success, -EBUSY when another driver got here first
2436  * (and isn't unregistered in the meantime).
2437  *
2438  */
2439 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2440 {
2441         unsigned long flags;
2442         int ret;
2443
2444         if (cpufreq_disabled())
2445                 return -ENODEV;
2446
2447         if (!driver_data || !driver_data->verify || !driver_data->init ||
2448             !(driver_data->setpolicy || driver_data->target_index ||
2449                     driver_data->target) ||
2450              (driver_data->setpolicy && (driver_data->target_index ||
2451                     driver_data->target)) ||
2452              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2453                 return -EINVAL;
2454
2455         pr_debug("trying to register driver %s\n", driver_data->name);
2456
2457         /* Protect against concurrent CPU online/offline. */
2458         get_online_cpus();
2459
2460         write_lock_irqsave(&cpufreq_driver_lock, flags);
2461         if (cpufreq_driver) {
2462                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2463                 ret = -EEXIST;
2464                 goto out;
2465         }
2466         cpufreq_driver = driver_data;
2467         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2468
2469         if (driver_data->setpolicy)
2470                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2471
2472         if (cpufreq_boost_supported()) {
2473                 /*
2474                  * Check if driver provides function to enable boost -
2475                  * if not, use cpufreq_boost_set_sw as default
2476                  */
2477                 if (!cpufreq_driver->set_boost)
2478                         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2479
2480                 ret = cpufreq_sysfs_create_file(&boost.attr);
2481                 if (ret) {
2482                         pr_err("%s: cannot register global BOOST sysfs file\n",
2483                                __func__);
2484                         goto err_null_driver;
2485                 }
2486         }
2487
2488         ret = subsys_interface_register(&cpufreq_interface);
2489         if (ret)
2490                 goto err_boost_unreg;
2491
2492         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2493             list_empty(&cpufreq_policy_list)) {
2494                 /* if all ->init() calls failed, unregister */
2495                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2496                          driver_data->name);
2497                 goto err_if_unreg;
2498         }
2499
2500         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2501         pr_debug("driver %s up and running\n", driver_data->name);
2502
2503 out:
2504         put_online_cpus();
2505         return ret;
2506
2507 err_if_unreg:
2508         subsys_interface_unregister(&cpufreq_interface);
2509 err_boost_unreg:
2510         if (cpufreq_boost_supported())
2511                 cpufreq_sysfs_remove_file(&boost.attr);
2512 err_null_driver:
2513         write_lock_irqsave(&cpufreq_driver_lock, flags);
2514         cpufreq_driver = NULL;
2515         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2516         goto out;
2517 }
2518 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2519
2520 /**
2521  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2522  *
2523  * Unregister the current CPUFreq driver. Only call this if you have
2524  * the right to do so, i.e. if you have succeeded in initialising before!
2525  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2526  * currently not initialised.
2527  */
2528 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2529 {
2530         unsigned long flags;
2531
2532         if (!cpufreq_driver || (driver != cpufreq_driver))
2533                 return -EINVAL;
2534
2535         pr_debug("unregistering driver %s\n", driver->name);
2536
2537         /* Protect against concurrent cpu hotplug */
2538         get_online_cpus();
2539         subsys_interface_unregister(&cpufreq_interface);
2540         if (cpufreq_boost_supported())
2541                 cpufreq_sysfs_remove_file(&boost.attr);
2542
2543         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2544
2545         write_lock_irqsave(&cpufreq_driver_lock, flags);
2546
2547         cpufreq_driver = NULL;
2548
2549         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2550         put_online_cpus();
2551
2552         return 0;
2553 }
2554 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2555
2556 /*
2557  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2558  * or mutexes when secondary CPUs are halted.
2559  */
2560 static struct syscore_ops cpufreq_syscore_ops = {
2561         .shutdown = cpufreq_suspend,
2562 };
2563
2564 static int __init cpufreq_core_init(void)
2565 {
2566         if (cpufreq_disabled())
2567                 return -ENODEV;
2568
2569         cpufreq_global_kobject = kobject_create();
2570         BUG_ON(!cpufreq_global_kobject);
2571
2572         register_syscore_ops(&cpufreq_syscore_ops);
2573
2574         return 0;
2575 }
2576 core_initcall(cpufreq_core_init);