usb: host: xhci-plat: Fix timeout on removal of hot pluggable xhci controllers
[sfrench/cifs-2.6.git] / drivers / cpufreq / cpufreq.c
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
7  *
8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
9  *      Added handling for CPU hotplug
10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
11  *      Fix handling for CPU hotplug -- affected CPUs
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License version 2 as
15  * published by the Free Software Foundation.
16  */
17
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20 #include <linux/cpu.h>
21 #include <linux/cpufreq.h>
22 #include <linux/delay.h>
23 #include <linux/device.h>
24 #include <linux/init.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/module.h>
27 #include <linux/mutex.h>
28 #include <linux/slab.h>
29 #include <linux/suspend.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/tick.h>
32 #include <trace/events/power.h>
33
34 static LIST_HEAD(cpufreq_policy_list);
35
36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
37 {
38         return cpumask_empty(policy->cpus);
39 }
40
41 /* Macros to iterate over CPU policies */
42 #define for_each_suitable_policy(__policy, __active)                     \
43         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
44                 if ((__active) == !policy_is_inactive(__policy))
45
46 #define for_each_active_policy(__policy)                \
47         for_each_suitable_policy(__policy, true)
48 #define for_each_inactive_policy(__policy)              \
49         for_each_suitable_policy(__policy, false)
50
51 #define for_each_policy(__policy)                       \
52         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
53
54 /* Iterate over governors */
55 static LIST_HEAD(cpufreq_governor_list);
56 #define for_each_governor(__governor)                           \
57         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
58
59 /**
60  * The "cpufreq driver" - the arch- or hardware-dependent low
61  * level driver of CPUFreq support, and its spinlock. This lock
62  * also protects the cpufreq_cpu_data array.
63  */
64 static struct cpufreq_driver *cpufreq_driver;
65 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
66 static DEFINE_RWLOCK(cpufreq_driver_lock);
67
68 /* Flag to suspend/resume CPUFreq governors */
69 static bool cpufreq_suspended;
70
71 static inline bool has_target(void)
72 {
73         return cpufreq_driver->target_index || cpufreq_driver->target;
74 }
75
76 /* internal prototypes */
77 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
78 static int cpufreq_init_governor(struct cpufreq_policy *policy);
79 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
80 static int cpufreq_start_governor(struct cpufreq_policy *policy);
81 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
82 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
83
84 /**
85  * Two notifier lists: the "policy" list is involved in the
86  * validation process for a new CPU frequency policy; the
87  * "transition" list for kernel code that needs to handle
88  * changes to devices when the CPU clock speed changes.
89  * The mutex locks both lists.
90  */
91 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
92 static struct srcu_notifier_head cpufreq_transition_notifier_list;
93
94 static bool init_cpufreq_transition_notifier_list_called;
95 static int __init init_cpufreq_transition_notifier_list(void)
96 {
97         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
98         init_cpufreq_transition_notifier_list_called = true;
99         return 0;
100 }
101 pure_initcall(init_cpufreq_transition_notifier_list);
102
103 static int off __read_mostly;
104 static int cpufreq_disabled(void)
105 {
106         return off;
107 }
108 void disable_cpufreq(void)
109 {
110         off = 1;
111 }
112 static DEFINE_MUTEX(cpufreq_governor_mutex);
113
114 bool have_governor_per_policy(void)
115 {
116         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
117 }
118 EXPORT_SYMBOL_GPL(have_governor_per_policy);
119
120 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
121 {
122         if (have_governor_per_policy())
123                 return &policy->kobj;
124         else
125                 return cpufreq_global_kobject;
126 }
127 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
128
129 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
130 {
131         u64 idle_time;
132         u64 cur_wall_time;
133         u64 busy_time;
134
135         cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
136
137         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
138         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
139         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
140         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
141         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
142         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
143
144         idle_time = cur_wall_time - busy_time;
145         if (wall)
146                 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
147
148         return div_u64(idle_time, NSEC_PER_USEC);
149 }
150
151 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
152 {
153         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
154
155         if (idle_time == -1ULL)
156                 return get_cpu_idle_time_jiffy(cpu, wall);
157         else if (!io_busy)
158                 idle_time += get_cpu_iowait_time_us(cpu, wall);
159
160         return idle_time;
161 }
162 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
163
164 /*
165  * This is a generic cpufreq init() routine which can be used by cpufreq
166  * drivers of SMP systems. It will do following:
167  * - validate & show freq table passed
168  * - set policies transition latency
169  * - policy->cpus with all possible CPUs
170  */
171 int cpufreq_generic_init(struct cpufreq_policy *policy,
172                 struct cpufreq_frequency_table *table,
173                 unsigned int transition_latency)
174 {
175         int ret;
176
177         ret = cpufreq_table_validate_and_show(policy, table);
178         if (ret) {
179                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
180                 return ret;
181         }
182
183         policy->cpuinfo.transition_latency = transition_latency;
184
185         /*
186          * The driver only supports the SMP configuration where all processors
187          * share the clock and voltage and clock.
188          */
189         cpumask_setall(policy->cpus);
190
191         return 0;
192 }
193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
194
195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
196 {
197         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
198
199         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
200 }
201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
202
203 unsigned int cpufreq_generic_get(unsigned int cpu)
204 {
205         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
206
207         if (!policy || IS_ERR(policy->clk)) {
208                 pr_err("%s: No %s associated to cpu: %d\n",
209                        __func__, policy ? "clk" : "policy", cpu);
210                 return 0;
211         }
212
213         return clk_get_rate(policy->clk) / 1000;
214 }
215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
216
217 /**
218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
219  *
220  * @cpu: cpu to find policy for.
221  *
222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
223  * It also increments the kobject reference count to mark it busy and so would
224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
226  * freed as that depends on the kobj count.
227  *
228  * Return: A valid policy on success, otherwise NULL on failure.
229  */
230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
231 {
232         struct cpufreq_policy *policy = NULL;
233         unsigned long flags;
234
235         if (WARN_ON(cpu >= nr_cpu_ids))
236                 return NULL;
237
238         /* get the cpufreq driver */
239         read_lock_irqsave(&cpufreq_driver_lock, flags);
240
241         if (cpufreq_driver) {
242                 /* get the CPU */
243                 policy = cpufreq_cpu_get_raw(cpu);
244                 if (policy)
245                         kobject_get(&policy->kobj);
246         }
247
248         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
249
250         return policy;
251 }
252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
253
254 /**
255  * cpufreq_cpu_put: Decrements the usage count of a policy
256  *
257  * @policy: policy earlier returned by cpufreq_cpu_get().
258  *
259  * This decrements the kobject reference count incremented earlier by calling
260  * cpufreq_cpu_get().
261  */
262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
263 {
264         kobject_put(&policy->kobj);
265 }
266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
267
268 /*********************************************************************
269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
270  *********************************************************************/
271
272 /**
273  * adjust_jiffies - adjust the system "loops_per_jiffy"
274  *
275  * This function alters the system "loops_per_jiffy" for the clock
276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
277  * systems as each CPU might be scaled differently. So, use the arch
278  * per-CPU loops_per_jiffy value wherever possible.
279  */
280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
281 {
282 #ifndef CONFIG_SMP
283         static unsigned long l_p_j_ref;
284         static unsigned int l_p_j_ref_freq;
285
286         if (ci->flags & CPUFREQ_CONST_LOOPS)
287                 return;
288
289         if (!l_p_j_ref_freq) {
290                 l_p_j_ref = loops_per_jiffy;
291                 l_p_j_ref_freq = ci->old;
292                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
293                          l_p_j_ref, l_p_j_ref_freq);
294         }
295         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297                                                                 ci->new);
298                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
299                          loops_per_jiffy, ci->new);
300         }
301 #endif
302 }
303
304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
305                 struct cpufreq_freqs *freqs, unsigned int state)
306 {
307         BUG_ON(irqs_disabled());
308
309         if (cpufreq_disabled())
310                 return;
311
312         freqs->flags = cpufreq_driver->flags;
313         pr_debug("notification %u of frequency transition to %u kHz\n",
314                  state, freqs->new);
315
316         switch (state) {
317
318         case CPUFREQ_PRECHANGE:
319                 /* detect if the driver reported a value as "old frequency"
320                  * which is not equal to what the cpufreq core thinks is
321                  * "old frequency".
322                  */
323                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
324                         if ((policy) && (policy->cpu == freqs->cpu) &&
325                             (policy->cur) && (policy->cur != freqs->old)) {
326                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
327                                          freqs->old, policy->cur);
328                                 freqs->old = policy->cur;
329                         }
330                 }
331                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
332                                 CPUFREQ_PRECHANGE, freqs);
333                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
334                 break;
335
336         case CPUFREQ_POSTCHANGE:
337                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
338                 pr_debug("FREQ: %lu - CPU: %lu\n",
339                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
340                 trace_cpu_frequency(freqs->new, freqs->cpu);
341                 cpufreq_stats_record_transition(policy, freqs->new);
342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
343                                 CPUFREQ_POSTCHANGE, freqs);
344                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
345                         policy->cur = freqs->new;
346                 break;
347         }
348 }
349
350 /**
351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
352  * on frequency transition.
353  *
354  * This function calls the transition notifiers and the "adjust_jiffies"
355  * function. It is called twice on all CPU frequency changes that have
356  * external effects.
357  */
358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
359                 struct cpufreq_freqs *freqs, unsigned int state)
360 {
361         for_each_cpu(freqs->cpu, policy->cpus)
362                 __cpufreq_notify_transition(policy, freqs, state);
363 }
364
365 /* Do post notifications when there are chances that transition has failed */
366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
367                 struct cpufreq_freqs *freqs, int transition_failed)
368 {
369         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
370         if (!transition_failed)
371                 return;
372
373         swap(freqs->old, freqs->new);
374         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
376 }
377
378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
379                 struct cpufreq_freqs *freqs)
380 {
381
382         /*
383          * Catch double invocations of _begin() which lead to self-deadlock.
384          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
385          * doesn't invoke _begin() on their behalf, and hence the chances of
386          * double invocations are very low. Moreover, there are scenarios
387          * where these checks can emit false-positive warnings in these
388          * drivers; so we avoid that by skipping them altogether.
389          */
390         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
391                                 && current == policy->transition_task);
392
393 wait:
394         wait_event(policy->transition_wait, !policy->transition_ongoing);
395
396         spin_lock(&policy->transition_lock);
397
398         if (unlikely(policy->transition_ongoing)) {
399                 spin_unlock(&policy->transition_lock);
400                 goto wait;
401         }
402
403         policy->transition_ongoing = true;
404         policy->transition_task = current;
405
406         spin_unlock(&policy->transition_lock);
407
408         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
409 }
410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
411
412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
413                 struct cpufreq_freqs *freqs, int transition_failed)
414 {
415         if (unlikely(WARN_ON(!policy->transition_ongoing)))
416                 return;
417
418         cpufreq_notify_post_transition(policy, freqs, transition_failed);
419
420         policy->transition_ongoing = false;
421         policy->transition_task = NULL;
422
423         wake_up(&policy->transition_wait);
424 }
425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
426
427 /*
428  * Fast frequency switching status count.  Positive means "enabled", negative
429  * means "disabled" and 0 means "not decided yet".
430  */
431 static int cpufreq_fast_switch_count;
432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
433
434 static void cpufreq_list_transition_notifiers(void)
435 {
436         struct notifier_block *nb;
437
438         pr_info("Registered transition notifiers:\n");
439
440         mutex_lock(&cpufreq_transition_notifier_list.mutex);
441
442         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
443                 pr_info("%pF\n", nb->notifier_call);
444
445         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
446 }
447
448 /**
449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
450  * @policy: cpufreq policy to enable fast frequency switching for.
451  *
452  * Try to enable fast frequency switching for @policy.
453  *
454  * The attempt will fail if there is at least one transition notifier registered
455  * at this point, as fast frequency switching is quite fundamentally at odds
456  * with transition notifiers.  Thus if successful, it will make registration of
457  * transition notifiers fail going forward.
458  */
459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
460 {
461         lockdep_assert_held(&policy->rwsem);
462
463         if (!policy->fast_switch_possible)
464                 return;
465
466         mutex_lock(&cpufreq_fast_switch_lock);
467         if (cpufreq_fast_switch_count >= 0) {
468                 cpufreq_fast_switch_count++;
469                 policy->fast_switch_enabled = true;
470         } else {
471                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
472                         policy->cpu);
473                 cpufreq_list_transition_notifiers();
474         }
475         mutex_unlock(&cpufreq_fast_switch_lock);
476 }
477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
478
479 /**
480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
481  * @policy: cpufreq policy to disable fast frequency switching for.
482  */
483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
484 {
485         mutex_lock(&cpufreq_fast_switch_lock);
486         if (policy->fast_switch_enabled) {
487                 policy->fast_switch_enabled = false;
488                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
489                         cpufreq_fast_switch_count--;
490         }
491         mutex_unlock(&cpufreq_fast_switch_lock);
492 }
493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
494
495 /**
496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
497  * one.
498  * @target_freq: target frequency to resolve.
499  *
500  * The target to driver frequency mapping is cached in the policy.
501  *
502  * Return: Lowest driver-supported frequency greater than or equal to the
503  * given target_freq, subject to policy (min/max) and driver limitations.
504  */
505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
506                                          unsigned int target_freq)
507 {
508         target_freq = clamp_val(target_freq, policy->min, policy->max);
509         policy->cached_target_freq = target_freq;
510
511         if (cpufreq_driver->target_index) {
512                 int idx;
513
514                 idx = cpufreq_frequency_table_target(policy, target_freq,
515                                                      CPUFREQ_RELATION_L);
516                 policy->cached_resolved_idx = idx;
517                 return policy->freq_table[idx].frequency;
518         }
519
520         if (cpufreq_driver->resolve_freq)
521                 return cpufreq_driver->resolve_freq(policy, target_freq);
522
523         return target_freq;
524 }
525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
526
527 /*********************************************************************
528  *                          SYSFS INTERFACE                          *
529  *********************************************************************/
530 static ssize_t show_boost(struct kobject *kobj,
531                                  struct attribute *attr, char *buf)
532 {
533         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
534 }
535
536 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
537                                   const char *buf, size_t count)
538 {
539         int ret, enable;
540
541         ret = sscanf(buf, "%d", &enable);
542         if (ret != 1 || enable < 0 || enable > 1)
543                 return -EINVAL;
544
545         if (cpufreq_boost_trigger_state(enable)) {
546                 pr_err("%s: Cannot %s BOOST!\n",
547                        __func__, enable ? "enable" : "disable");
548                 return -EINVAL;
549         }
550
551         pr_debug("%s: cpufreq BOOST %s\n",
552                  __func__, enable ? "enabled" : "disabled");
553
554         return count;
555 }
556 define_one_global_rw(boost);
557
558 static struct cpufreq_governor *find_governor(const char *str_governor)
559 {
560         struct cpufreq_governor *t;
561
562         for_each_governor(t)
563                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
564                         return t;
565
566         return NULL;
567 }
568
569 /**
570  * cpufreq_parse_governor - parse a governor string
571  */
572 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
573                                 struct cpufreq_governor **governor)
574 {
575         int err = -EINVAL;
576
577         if (cpufreq_driver->setpolicy) {
578                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
579                         *policy = CPUFREQ_POLICY_PERFORMANCE;
580                         err = 0;
581                 } else if (!strncasecmp(str_governor, "powersave",
582                                                 CPUFREQ_NAME_LEN)) {
583                         *policy = CPUFREQ_POLICY_POWERSAVE;
584                         err = 0;
585                 }
586         } else {
587                 struct cpufreq_governor *t;
588
589                 mutex_lock(&cpufreq_governor_mutex);
590
591                 t = find_governor(str_governor);
592
593                 if (t == NULL) {
594                         int ret;
595
596                         mutex_unlock(&cpufreq_governor_mutex);
597                         ret = request_module("cpufreq_%s", str_governor);
598                         mutex_lock(&cpufreq_governor_mutex);
599
600                         if (ret == 0)
601                                 t = find_governor(str_governor);
602                 }
603
604                 if (t != NULL) {
605                         *governor = t;
606                         err = 0;
607                 }
608
609                 mutex_unlock(&cpufreq_governor_mutex);
610         }
611         return err;
612 }
613
614 /**
615  * cpufreq_per_cpu_attr_read() / show_##file_name() -
616  * print out cpufreq information
617  *
618  * Write out information from cpufreq_driver->policy[cpu]; object must be
619  * "unsigned int".
620  */
621
622 #define show_one(file_name, object)                     \
623 static ssize_t show_##file_name                         \
624 (struct cpufreq_policy *policy, char *buf)              \
625 {                                                       \
626         return sprintf(buf, "%u\n", policy->object);    \
627 }
628
629 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
630 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
631 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
632 show_one(scaling_min_freq, min);
633 show_one(scaling_max_freq, max);
634
635 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
636 {
637         ssize_t ret;
638
639         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
640                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
641         else
642                 ret = sprintf(buf, "%u\n", policy->cur);
643         return ret;
644 }
645
646 static int cpufreq_set_policy(struct cpufreq_policy *policy,
647                                 struct cpufreq_policy *new_policy);
648
649 /**
650  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
651  */
652 #define store_one(file_name, object)                    \
653 static ssize_t store_##file_name                                        \
654 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
655 {                                                                       \
656         int ret, temp;                                                  \
657         struct cpufreq_policy new_policy;                               \
658                                                                         \
659         memcpy(&new_policy, policy, sizeof(*policy));                   \
660                                                                         \
661         ret = sscanf(buf, "%u", &new_policy.object);                    \
662         if (ret != 1)                                                   \
663                 return -EINVAL;                                         \
664                                                                         \
665         temp = new_policy.object;                                       \
666         ret = cpufreq_set_policy(policy, &new_policy);          \
667         if (!ret)                                                       \
668                 policy->user_policy.object = temp;                      \
669                                                                         \
670         return ret ? ret : count;                                       \
671 }
672
673 store_one(scaling_min_freq, min);
674 store_one(scaling_max_freq, max);
675
676 /**
677  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
678  */
679 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
680                                         char *buf)
681 {
682         unsigned int cur_freq = __cpufreq_get(policy);
683         if (!cur_freq)
684                 return sprintf(buf, "<unknown>");
685         return sprintf(buf, "%u\n", cur_freq);
686 }
687
688 /**
689  * show_scaling_governor - show the current policy for the specified CPU
690  */
691 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
692 {
693         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
694                 return sprintf(buf, "powersave\n");
695         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
696                 return sprintf(buf, "performance\n");
697         else if (policy->governor)
698                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
699                                 policy->governor->name);
700         return -EINVAL;
701 }
702
703 /**
704  * store_scaling_governor - store policy for the specified CPU
705  */
706 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
707                                         const char *buf, size_t count)
708 {
709         int ret;
710         char    str_governor[16];
711         struct cpufreq_policy new_policy;
712
713         memcpy(&new_policy, policy, sizeof(*policy));
714
715         ret = sscanf(buf, "%15s", str_governor);
716         if (ret != 1)
717                 return -EINVAL;
718
719         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
720                                                 &new_policy.governor))
721                 return -EINVAL;
722
723         ret = cpufreq_set_policy(policy, &new_policy);
724         return ret ? ret : count;
725 }
726
727 /**
728  * show_scaling_driver - show the cpufreq driver currently loaded
729  */
730 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
731 {
732         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
733 }
734
735 /**
736  * show_scaling_available_governors - show the available CPUfreq governors
737  */
738 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
739                                                 char *buf)
740 {
741         ssize_t i = 0;
742         struct cpufreq_governor *t;
743
744         if (!has_target()) {
745                 i += sprintf(buf, "performance powersave");
746                 goto out;
747         }
748
749         for_each_governor(t) {
750                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
751                     - (CPUFREQ_NAME_LEN + 2)))
752                         goto out;
753                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
754         }
755 out:
756         i += sprintf(&buf[i], "\n");
757         return i;
758 }
759
760 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
761 {
762         ssize_t i = 0;
763         unsigned int cpu;
764
765         for_each_cpu(cpu, mask) {
766                 if (i)
767                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
768                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
769                 if (i >= (PAGE_SIZE - 5))
770                         break;
771         }
772         i += sprintf(&buf[i], "\n");
773         return i;
774 }
775 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
776
777 /**
778  * show_related_cpus - show the CPUs affected by each transition even if
779  * hw coordination is in use
780  */
781 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
782 {
783         return cpufreq_show_cpus(policy->related_cpus, buf);
784 }
785
786 /**
787  * show_affected_cpus - show the CPUs affected by each transition
788  */
789 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
790 {
791         return cpufreq_show_cpus(policy->cpus, buf);
792 }
793
794 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
795                                         const char *buf, size_t count)
796 {
797         unsigned int freq = 0;
798         unsigned int ret;
799
800         if (!policy->governor || !policy->governor->store_setspeed)
801                 return -EINVAL;
802
803         ret = sscanf(buf, "%u", &freq);
804         if (ret != 1)
805                 return -EINVAL;
806
807         policy->governor->store_setspeed(policy, freq);
808
809         return count;
810 }
811
812 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
813 {
814         if (!policy->governor || !policy->governor->show_setspeed)
815                 return sprintf(buf, "<unsupported>\n");
816
817         return policy->governor->show_setspeed(policy, buf);
818 }
819
820 /**
821  * show_bios_limit - show the current cpufreq HW/BIOS limitation
822  */
823 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
824 {
825         unsigned int limit;
826         int ret;
827         if (cpufreq_driver->bios_limit) {
828                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
829                 if (!ret)
830                         return sprintf(buf, "%u\n", limit);
831         }
832         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
833 }
834
835 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
836 cpufreq_freq_attr_ro(cpuinfo_min_freq);
837 cpufreq_freq_attr_ro(cpuinfo_max_freq);
838 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
839 cpufreq_freq_attr_ro(scaling_available_governors);
840 cpufreq_freq_attr_ro(scaling_driver);
841 cpufreq_freq_attr_ro(scaling_cur_freq);
842 cpufreq_freq_attr_ro(bios_limit);
843 cpufreq_freq_attr_ro(related_cpus);
844 cpufreq_freq_attr_ro(affected_cpus);
845 cpufreq_freq_attr_rw(scaling_min_freq);
846 cpufreq_freq_attr_rw(scaling_max_freq);
847 cpufreq_freq_attr_rw(scaling_governor);
848 cpufreq_freq_attr_rw(scaling_setspeed);
849
850 static struct attribute *default_attrs[] = {
851         &cpuinfo_min_freq.attr,
852         &cpuinfo_max_freq.attr,
853         &cpuinfo_transition_latency.attr,
854         &scaling_min_freq.attr,
855         &scaling_max_freq.attr,
856         &affected_cpus.attr,
857         &related_cpus.attr,
858         &scaling_governor.attr,
859         &scaling_driver.attr,
860         &scaling_available_governors.attr,
861         &scaling_setspeed.attr,
862         NULL
863 };
864
865 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
866 #define to_attr(a) container_of(a, struct freq_attr, attr)
867
868 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
869 {
870         struct cpufreq_policy *policy = to_policy(kobj);
871         struct freq_attr *fattr = to_attr(attr);
872         ssize_t ret;
873
874         down_read(&policy->rwsem);
875         ret = fattr->show(policy, buf);
876         up_read(&policy->rwsem);
877
878         return ret;
879 }
880
881 static ssize_t store(struct kobject *kobj, struct attribute *attr,
882                      const char *buf, size_t count)
883 {
884         struct cpufreq_policy *policy = to_policy(kobj);
885         struct freq_attr *fattr = to_attr(attr);
886         ssize_t ret = -EINVAL;
887
888         get_online_cpus();
889
890         if (cpu_online(policy->cpu)) {
891                 down_write(&policy->rwsem);
892                 ret = fattr->store(policy, buf, count);
893                 up_write(&policy->rwsem);
894         }
895
896         put_online_cpus();
897
898         return ret;
899 }
900
901 static void cpufreq_sysfs_release(struct kobject *kobj)
902 {
903         struct cpufreq_policy *policy = to_policy(kobj);
904         pr_debug("last reference is dropped\n");
905         complete(&policy->kobj_unregister);
906 }
907
908 static const struct sysfs_ops sysfs_ops = {
909         .show   = show,
910         .store  = store,
911 };
912
913 static struct kobj_type ktype_cpufreq = {
914         .sysfs_ops      = &sysfs_ops,
915         .default_attrs  = default_attrs,
916         .release        = cpufreq_sysfs_release,
917 };
918
919 static int add_cpu_dev_symlink(struct cpufreq_policy *policy,
920                                struct device *dev)
921 {
922         dev_dbg(dev, "%s: Adding symlink\n", __func__);
923         return sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
924 }
925
926 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
927                                    struct device *dev)
928 {
929         dev_dbg(dev, "%s: Removing symlink\n", __func__);
930         sysfs_remove_link(&dev->kobj, "cpufreq");
931 }
932
933 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
934 {
935         struct freq_attr **drv_attr;
936         int ret = 0;
937
938         /* set up files for this cpu device */
939         drv_attr = cpufreq_driver->attr;
940         while (drv_attr && *drv_attr) {
941                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
942                 if (ret)
943                         return ret;
944                 drv_attr++;
945         }
946         if (cpufreq_driver->get) {
947                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
948                 if (ret)
949                         return ret;
950         }
951
952         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
953         if (ret)
954                 return ret;
955
956         if (cpufreq_driver->bios_limit) {
957                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
958                 if (ret)
959                         return ret;
960         }
961
962         return 0;
963 }
964
965 __weak struct cpufreq_governor *cpufreq_default_governor(void)
966 {
967         return NULL;
968 }
969
970 static int cpufreq_init_policy(struct cpufreq_policy *policy)
971 {
972         struct cpufreq_governor *gov = NULL;
973         struct cpufreq_policy new_policy;
974
975         memcpy(&new_policy, policy, sizeof(*policy));
976
977         /* Update governor of new_policy to the governor used before hotplug */
978         gov = find_governor(policy->last_governor);
979         if (gov) {
980                 pr_debug("Restoring governor %s for cpu %d\n",
981                                 policy->governor->name, policy->cpu);
982         } else {
983                 gov = cpufreq_default_governor();
984                 if (!gov)
985                         return -ENODATA;
986         }
987
988         new_policy.governor = gov;
989
990         /* Use the default policy if there is no last_policy. */
991         if (cpufreq_driver->setpolicy) {
992                 if (policy->last_policy)
993                         new_policy.policy = policy->last_policy;
994                 else
995                         cpufreq_parse_governor(gov->name, &new_policy.policy,
996                                                NULL);
997         }
998         /* set default policy */
999         return cpufreq_set_policy(policy, &new_policy);
1000 }
1001
1002 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1003 {
1004         int ret = 0;
1005
1006         /* Has this CPU been taken care of already? */
1007         if (cpumask_test_cpu(cpu, policy->cpus))
1008                 return 0;
1009
1010         down_write(&policy->rwsem);
1011         if (has_target())
1012                 cpufreq_stop_governor(policy);
1013
1014         cpumask_set_cpu(cpu, policy->cpus);
1015
1016         if (has_target()) {
1017                 ret = cpufreq_start_governor(policy);
1018                 if (ret)
1019                         pr_err("%s: Failed to start governor\n", __func__);
1020         }
1021         up_write(&policy->rwsem);
1022         return ret;
1023 }
1024
1025 static void handle_update(struct work_struct *work)
1026 {
1027         struct cpufreq_policy *policy =
1028                 container_of(work, struct cpufreq_policy, update);
1029         unsigned int cpu = policy->cpu;
1030         pr_debug("handle_update for cpu %u called\n", cpu);
1031         cpufreq_update_policy(cpu);
1032 }
1033
1034 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1035 {
1036         struct cpufreq_policy *policy;
1037         int ret;
1038
1039         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1040         if (!policy)
1041                 return NULL;
1042
1043         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1044                 goto err_free_policy;
1045
1046         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1047                 goto err_free_cpumask;
1048
1049         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1050                 goto err_free_rcpumask;
1051
1052         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1053                                    cpufreq_global_kobject, "policy%u", cpu);
1054         if (ret) {
1055                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1056                 goto err_free_real_cpus;
1057         }
1058
1059         INIT_LIST_HEAD(&policy->policy_list);
1060         init_rwsem(&policy->rwsem);
1061         spin_lock_init(&policy->transition_lock);
1062         init_waitqueue_head(&policy->transition_wait);
1063         init_completion(&policy->kobj_unregister);
1064         INIT_WORK(&policy->update, handle_update);
1065
1066         policy->cpu = cpu;
1067         return policy;
1068
1069 err_free_real_cpus:
1070         free_cpumask_var(policy->real_cpus);
1071 err_free_rcpumask:
1072         free_cpumask_var(policy->related_cpus);
1073 err_free_cpumask:
1074         free_cpumask_var(policy->cpus);
1075 err_free_policy:
1076         kfree(policy);
1077
1078         return NULL;
1079 }
1080
1081 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1082 {
1083         struct kobject *kobj;
1084         struct completion *cmp;
1085
1086         down_write(&policy->rwsem);
1087         cpufreq_stats_free_table(policy);
1088         kobj = &policy->kobj;
1089         cmp = &policy->kobj_unregister;
1090         up_write(&policy->rwsem);
1091         kobject_put(kobj);
1092
1093         /*
1094          * We need to make sure that the underlying kobj is
1095          * actually not referenced anymore by anybody before we
1096          * proceed with unloading.
1097          */
1098         pr_debug("waiting for dropping of refcount\n");
1099         wait_for_completion(cmp);
1100         pr_debug("wait complete\n");
1101 }
1102
1103 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1104 {
1105         unsigned long flags;
1106         int cpu;
1107
1108         /* Remove policy from list */
1109         write_lock_irqsave(&cpufreq_driver_lock, flags);
1110         list_del(&policy->policy_list);
1111
1112         for_each_cpu(cpu, policy->related_cpus)
1113                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1114         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1115
1116         cpufreq_policy_put_kobj(policy);
1117         free_cpumask_var(policy->real_cpus);
1118         free_cpumask_var(policy->related_cpus);
1119         free_cpumask_var(policy->cpus);
1120         kfree(policy);
1121 }
1122
1123 static int cpufreq_online(unsigned int cpu)
1124 {
1125         struct cpufreq_policy *policy;
1126         bool new_policy;
1127         unsigned long flags;
1128         unsigned int j;
1129         int ret;
1130
1131         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1132
1133         /* Check if this CPU already has a policy to manage it */
1134         policy = per_cpu(cpufreq_cpu_data, cpu);
1135         if (policy) {
1136                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1137                 if (!policy_is_inactive(policy))
1138                         return cpufreq_add_policy_cpu(policy, cpu);
1139
1140                 /* This is the only online CPU for the policy.  Start over. */
1141                 new_policy = false;
1142                 down_write(&policy->rwsem);
1143                 policy->cpu = cpu;
1144                 policy->governor = NULL;
1145                 up_write(&policy->rwsem);
1146         } else {
1147                 new_policy = true;
1148                 policy = cpufreq_policy_alloc(cpu);
1149                 if (!policy)
1150                         return -ENOMEM;
1151         }
1152
1153         cpumask_copy(policy->cpus, cpumask_of(cpu));
1154
1155         /* call driver. From then on the cpufreq must be able
1156          * to accept all calls to ->verify and ->setpolicy for this CPU
1157          */
1158         ret = cpufreq_driver->init(policy);
1159         if (ret) {
1160                 pr_debug("initialization failed\n");
1161                 goto out_free_policy;
1162         }
1163
1164         down_write(&policy->rwsem);
1165
1166         if (new_policy) {
1167                 /* related_cpus should at least include policy->cpus. */
1168                 cpumask_copy(policy->related_cpus, policy->cpus);
1169         }
1170
1171         /*
1172          * affected cpus must always be the one, which are online. We aren't
1173          * managing offline cpus here.
1174          */
1175         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1176
1177         if (new_policy) {
1178                 policy->user_policy.min = policy->min;
1179                 policy->user_policy.max = policy->max;
1180
1181                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1182                 for_each_cpu(j, policy->related_cpus)
1183                         per_cpu(cpufreq_cpu_data, j) = policy;
1184                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1185         }
1186
1187         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1188                 policy->cur = cpufreq_driver->get(policy->cpu);
1189                 if (!policy->cur) {
1190                         pr_err("%s: ->get() failed\n", __func__);
1191                         goto out_exit_policy;
1192                 }
1193         }
1194
1195         /*
1196          * Sometimes boot loaders set CPU frequency to a value outside of
1197          * frequency table present with cpufreq core. In such cases CPU might be
1198          * unstable if it has to run on that frequency for long duration of time
1199          * and so its better to set it to a frequency which is specified in
1200          * freq-table. This also makes cpufreq stats inconsistent as
1201          * cpufreq-stats would fail to register because current frequency of CPU
1202          * isn't found in freq-table.
1203          *
1204          * Because we don't want this change to effect boot process badly, we go
1205          * for the next freq which is >= policy->cur ('cur' must be set by now,
1206          * otherwise we will end up setting freq to lowest of the table as 'cur'
1207          * is initialized to zero).
1208          *
1209          * We are passing target-freq as "policy->cur - 1" otherwise
1210          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1211          * equal to target-freq.
1212          */
1213         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1214             && has_target()) {
1215                 /* Are we running at unknown frequency ? */
1216                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1217                 if (ret == -EINVAL) {
1218                         /* Warn user and fix it */
1219                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1220                                 __func__, policy->cpu, policy->cur);
1221                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1222                                 CPUFREQ_RELATION_L);
1223
1224                         /*
1225                          * Reaching here after boot in a few seconds may not
1226                          * mean that system will remain stable at "unknown"
1227                          * frequency for longer duration. Hence, a BUG_ON().
1228                          */
1229                         BUG_ON(ret);
1230                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1231                                 __func__, policy->cpu, policy->cur);
1232                 }
1233         }
1234
1235         if (new_policy) {
1236                 ret = cpufreq_add_dev_interface(policy);
1237                 if (ret)
1238                         goto out_exit_policy;
1239
1240                 cpufreq_stats_create_table(policy);
1241
1242                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1243                 list_add(&policy->policy_list, &cpufreq_policy_list);
1244                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1245         }
1246
1247         ret = cpufreq_init_policy(policy);
1248         if (ret) {
1249                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1250                        __func__, cpu, ret);
1251                 /* cpufreq_policy_free() will notify based on this */
1252                 new_policy = false;
1253                 goto out_exit_policy;
1254         }
1255
1256         up_write(&policy->rwsem);
1257
1258         kobject_uevent(&policy->kobj, KOBJ_ADD);
1259
1260         /* Callback for handling stuff after policy is ready */
1261         if (cpufreq_driver->ready)
1262                 cpufreq_driver->ready(policy);
1263
1264         pr_debug("initialization complete\n");
1265
1266         return 0;
1267
1268 out_exit_policy:
1269         up_write(&policy->rwsem);
1270
1271         if (cpufreq_driver->exit)
1272                 cpufreq_driver->exit(policy);
1273 out_free_policy:
1274         cpufreq_policy_free(policy);
1275         return ret;
1276 }
1277
1278 static int cpufreq_offline(unsigned int cpu);
1279
1280 /**
1281  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1282  * @dev: CPU device.
1283  * @sif: Subsystem interface structure pointer (not used)
1284  */
1285 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1286 {
1287         struct cpufreq_policy *policy;
1288         unsigned cpu = dev->id;
1289         int ret;
1290
1291         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1292
1293         if (cpu_online(cpu)) {
1294                 ret = cpufreq_online(cpu);
1295                 if (ret)
1296                         return ret;
1297         }
1298
1299         /* Create sysfs link on CPU registration */
1300         policy = per_cpu(cpufreq_cpu_data, cpu);
1301         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1302                 return 0;
1303
1304         ret = add_cpu_dev_symlink(policy, dev);
1305         if (ret) {
1306                 cpumask_clear_cpu(cpu, policy->real_cpus);
1307                 cpufreq_offline(cpu);
1308         }
1309
1310         return ret;
1311 }
1312
1313 static int cpufreq_offline(unsigned int cpu)
1314 {
1315         struct cpufreq_policy *policy;
1316         int ret;
1317
1318         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1319
1320         policy = cpufreq_cpu_get_raw(cpu);
1321         if (!policy) {
1322                 pr_debug("%s: No cpu_data found\n", __func__);
1323                 return 0;
1324         }
1325
1326         down_write(&policy->rwsem);
1327         if (has_target())
1328                 cpufreq_stop_governor(policy);
1329
1330         cpumask_clear_cpu(cpu, policy->cpus);
1331
1332         if (policy_is_inactive(policy)) {
1333                 if (has_target())
1334                         strncpy(policy->last_governor, policy->governor->name,
1335                                 CPUFREQ_NAME_LEN);
1336                 else
1337                         policy->last_policy = policy->policy;
1338         } else if (cpu == policy->cpu) {
1339                 /* Nominate new CPU */
1340                 policy->cpu = cpumask_any(policy->cpus);
1341         }
1342
1343         /* Start governor again for active policy */
1344         if (!policy_is_inactive(policy)) {
1345                 if (has_target()) {
1346                         ret = cpufreq_start_governor(policy);
1347                         if (ret)
1348                                 pr_err("%s: Failed to start governor\n", __func__);
1349                 }
1350
1351                 goto unlock;
1352         }
1353
1354         if (cpufreq_driver->stop_cpu)
1355                 cpufreq_driver->stop_cpu(policy);
1356
1357         if (has_target())
1358                 cpufreq_exit_governor(policy);
1359
1360         /*
1361          * Perform the ->exit() even during light-weight tear-down,
1362          * since this is a core component, and is essential for the
1363          * subsequent light-weight ->init() to succeed.
1364          */
1365         if (cpufreq_driver->exit) {
1366                 cpufreq_driver->exit(policy);
1367                 policy->freq_table = NULL;
1368         }
1369
1370 unlock:
1371         up_write(&policy->rwsem);
1372         return 0;
1373 }
1374
1375 /**
1376  * cpufreq_remove_dev - remove a CPU device
1377  *
1378  * Removes the cpufreq interface for a CPU device.
1379  */
1380 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1381 {
1382         unsigned int cpu = dev->id;
1383         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1384
1385         if (!policy)
1386                 return;
1387
1388         if (cpu_online(cpu))
1389                 cpufreq_offline(cpu);
1390
1391         cpumask_clear_cpu(cpu, policy->real_cpus);
1392         remove_cpu_dev_symlink(policy, dev);
1393
1394         if (cpumask_empty(policy->real_cpus))
1395                 cpufreq_policy_free(policy);
1396 }
1397
1398 /**
1399  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1400  *      in deep trouble.
1401  *      @policy: policy managing CPUs
1402  *      @new_freq: CPU frequency the CPU actually runs at
1403  *
1404  *      We adjust to current frequency first, and need to clean up later.
1405  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1406  */
1407 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1408                                 unsigned int new_freq)
1409 {
1410         struct cpufreq_freqs freqs;
1411
1412         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1413                  policy->cur, new_freq);
1414
1415         freqs.old = policy->cur;
1416         freqs.new = new_freq;
1417
1418         cpufreq_freq_transition_begin(policy, &freqs);
1419         cpufreq_freq_transition_end(policy, &freqs, 0);
1420 }
1421
1422 /**
1423  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1424  * @cpu: CPU number
1425  *
1426  * This is the last known freq, without actually getting it from the driver.
1427  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1428  */
1429 unsigned int cpufreq_quick_get(unsigned int cpu)
1430 {
1431         struct cpufreq_policy *policy;
1432         unsigned int ret_freq = 0;
1433         unsigned long flags;
1434
1435         read_lock_irqsave(&cpufreq_driver_lock, flags);
1436
1437         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1438                 ret_freq = cpufreq_driver->get(cpu);
1439                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1440                 return ret_freq;
1441         }
1442
1443         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1444
1445         policy = cpufreq_cpu_get(cpu);
1446         if (policy) {
1447                 ret_freq = policy->cur;
1448                 cpufreq_cpu_put(policy);
1449         }
1450
1451         return ret_freq;
1452 }
1453 EXPORT_SYMBOL(cpufreq_quick_get);
1454
1455 /**
1456  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1457  * @cpu: CPU number
1458  *
1459  * Just return the max possible frequency for a given CPU.
1460  */
1461 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1462 {
1463         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1464         unsigned int ret_freq = 0;
1465
1466         if (policy) {
1467                 ret_freq = policy->max;
1468                 cpufreq_cpu_put(policy);
1469         }
1470
1471         return ret_freq;
1472 }
1473 EXPORT_SYMBOL(cpufreq_quick_get_max);
1474
1475 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1476 {
1477         unsigned int ret_freq = 0;
1478
1479         if (!cpufreq_driver->get)
1480                 return ret_freq;
1481
1482         ret_freq = cpufreq_driver->get(policy->cpu);
1483
1484         /*
1485          * Updating inactive policies is invalid, so avoid doing that.  Also
1486          * if fast frequency switching is used with the given policy, the check
1487          * against policy->cur is pointless, so skip it in that case too.
1488          */
1489         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1490                 return ret_freq;
1491
1492         if (ret_freq && policy->cur &&
1493                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1494                 /* verify no discrepancy between actual and
1495                                         saved value exists */
1496                 if (unlikely(ret_freq != policy->cur)) {
1497                         cpufreq_out_of_sync(policy, ret_freq);
1498                         schedule_work(&policy->update);
1499                 }
1500         }
1501
1502         return ret_freq;
1503 }
1504
1505 /**
1506  * cpufreq_get - get the current CPU frequency (in kHz)
1507  * @cpu: CPU number
1508  *
1509  * Get the CPU current (static) CPU frequency
1510  */
1511 unsigned int cpufreq_get(unsigned int cpu)
1512 {
1513         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1514         unsigned int ret_freq = 0;
1515
1516         if (policy) {
1517                 down_read(&policy->rwsem);
1518
1519                 if (!policy_is_inactive(policy))
1520                         ret_freq = __cpufreq_get(policy);
1521
1522                 up_read(&policy->rwsem);
1523
1524                 cpufreq_cpu_put(policy);
1525         }
1526
1527         return ret_freq;
1528 }
1529 EXPORT_SYMBOL(cpufreq_get);
1530
1531 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1532 {
1533         unsigned int new_freq;
1534
1535         new_freq = cpufreq_driver->get(policy->cpu);
1536         if (!new_freq)
1537                 return 0;
1538
1539         if (!policy->cur) {
1540                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1541                 policy->cur = new_freq;
1542         } else if (policy->cur != new_freq && has_target()) {
1543                 cpufreq_out_of_sync(policy, new_freq);
1544         }
1545
1546         return new_freq;
1547 }
1548
1549 static struct subsys_interface cpufreq_interface = {
1550         .name           = "cpufreq",
1551         .subsys         = &cpu_subsys,
1552         .add_dev        = cpufreq_add_dev,
1553         .remove_dev     = cpufreq_remove_dev,
1554 };
1555
1556 /*
1557  * In case platform wants some specific frequency to be configured
1558  * during suspend..
1559  */
1560 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1561 {
1562         int ret;
1563
1564         if (!policy->suspend_freq) {
1565                 pr_debug("%s: suspend_freq not defined\n", __func__);
1566                 return 0;
1567         }
1568
1569         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1570                         policy->suspend_freq);
1571
1572         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1573                         CPUFREQ_RELATION_H);
1574         if (ret)
1575                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1576                                 __func__, policy->suspend_freq, ret);
1577
1578         return ret;
1579 }
1580 EXPORT_SYMBOL(cpufreq_generic_suspend);
1581
1582 /**
1583  * cpufreq_suspend() - Suspend CPUFreq governors
1584  *
1585  * Called during system wide Suspend/Hibernate cycles for suspending governors
1586  * as some platforms can't change frequency after this point in suspend cycle.
1587  * Because some of the devices (like: i2c, regulators, etc) they use for
1588  * changing frequency are suspended quickly after this point.
1589  */
1590 void cpufreq_suspend(void)
1591 {
1592         struct cpufreq_policy *policy;
1593
1594         if (!cpufreq_driver)
1595                 return;
1596
1597         if (!has_target() && !cpufreq_driver->suspend)
1598                 goto suspend;
1599
1600         pr_debug("%s: Suspending Governors\n", __func__);
1601
1602         for_each_active_policy(policy) {
1603                 if (has_target()) {
1604                         down_write(&policy->rwsem);
1605                         cpufreq_stop_governor(policy);
1606                         up_write(&policy->rwsem);
1607                 }
1608
1609                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1610                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1611                                 policy);
1612         }
1613
1614 suspend:
1615         cpufreq_suspended = true;
1616 }
1617
1618 /**
1619  * cpufreq_resume() - Resume CPUFreq governors
1620  *
1621  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1622  * are suspended with cpufreq_suspend().
1623  */
1624 void cpufreq_resume(void)
1625 {
1626         struct cpufreq_policy *policy;
1627         int ret;
1628
1629         if (!cpufreq_driver)
1630                 return;
1631
1632         cpufreq_suspended = false;
1633
1634         if (!has_target() && !cpufreq_driver->resume)
1635                 return;
1636
1637         pr_debug("%s: Resuming Governors\n", __func__);
1638
1639         for_each_active_policy(policy) {
1640                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1641                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1642                                 policy);
1643                 } else if (has_target()) {
1644                         down_write(&policy->rwsem);
1645                         ret = cpufreq_start_governor(policy);
1646                         up_write(&policy->rwsem);
1647
1648                         if (ret)
1649                                 pr_err("%s: Failed to start governor for policy: %p\n",
1650                                        __func__, policy);
1651                 }
1652         }
1653 }
1654
1655 /**
1656  *      cpufreq_get_current_driver - return current driver's name
1657  *
1658  *      Return the name string of the currently loaded cpufreq driver
1659  *      or NULL, if none.
1660  */
1661 const char *cpufreq_get_current_driver(void)
1662 {
1663         if (cpufreq_driver)
1664                 return cpufreq_driver->name;
1665
1666         return NULL;
1667 }
1668 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1669
1670 /**
1671  *      cpufreq_get_driver_data - return current driver data
1672  *
1673  *      Return the private data of the currently loaded cpufreq
1674  *      driver, or NULL if no cpufreq driver is loaded.
1675  */
1676 void *cpufreq_get_driver_data(void)
1677 {
1678         if (cpufreq_driver)
1679                 return cpufreq_driver->driver_data;
1680
1681         return NULL;
1682 }
1683 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1684
1685 /*********************************************************************
1686  *                     NOTIFIER LISTS INTERFACE                      *
1687  *********************************************************************/
1688
1689 /**
1690  *      cpufreq_register_notifier - register a driver with cpufreq
1691  *      @nb: notifier function to register
1692  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1693  *
1694  *      Add a driver to one of two lists: either a list of drivers that
1695  *      are notified about clock rate changes (once before and once after
1696  *      the transition), or a list of drivers that are notified about
1697  *      changes in cpufreq policy.
1698  *
1699  *      This function may sleep, and has the same return conditions as
1700  *      blocking_notifier_chain_register.
1701  */
1702 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1703 {
1704         int ret;
1705
1706         if (cpufreq_disabled())
1707                 return -EINVAL;
1708
1709         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1710
1711         switch (list) {
1712         case CPUFREQ_TRANSITION_NOTIFIER:
1713                 mutex_lock(&cpufreq_fast_switch_lock);
1714
1715                 if (cpufreq_fast_switch_count > 0) {
1716                         mutex_unlock(&cpufreq_fast_switch_lock);
1717                         return -EBUSY;
1718                 }
1719                 ret = srcu_notifier_chain_register(
1720                                 &cpufreq_transition_notifier_list, nb);
1721                 if (!ret)
1722                         cpufreq_fast_switch_count--;
1723
1724                 mutex_unlock(&cpufreq_fast_switch_lock);
1725                 break;
1726         case CPUFREQ_POLICY_NOTIFIER:
1727                 ret = blocking_notifier_chain_register(
1728                                 &cpufreq_policy_notifier_list, nb);
1729                 break;
1730         default:
1731                 ret = -EINVAL;
1732         }
1733
1734         return ret;
1735 }
1736 EXPORT_SYMBOL(cpufreq_register_notifier);
1737
1738 /**
1739  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1740  *      @nb: notifier block to be unregistered
1741  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1742  *
1743  *      Remove a driver from the CPU frequency notifier list.
1744  *
1745  *      This function may sleep, and has the same return conditions as
1746  *      blocking_notifier_chain_unregister.
1747  */
1748 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1749 {
1750         int ret;
1751
1752         if (cpufreq_disabled())
1753                 return -EINVAL;
1754
1755         switch (list) {
1756         case CPUFREQ_TRANSITION_NOTIFIER:
1757                 mutex_lock(&cpufreq_fast_switch_lock);
1758
1759                 ret = srcu_notifier_chain_unregister(
1760                                 &cpufreq_transition_notifier_list, nb);
1761                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1762                         cpufreq_fast_switch_count++;
1763
1764                 mutex_unlock(&cpufreq_fast_switch_lock);
1765                 break;
1766         case CPUFREQ_POLICY_NOTIFIER:
1767                 ret = blocking_notifier_chain_unregister(
1768                                 &cpufreq_policy_notifier_list, nb);
1769                 break;
1770         default:
1771                 ret = -EINVAL;
1772         }
1773
1774         return ret;
1775 }
1776 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1777
1778
1779 /*********************************************************************
1780  *                              GOVERNORS                            *
1781  *********************************************************************/
1782
1783 /**
1784  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1785  * @policy: cpufreq policy to switch the frequency for.
1786  * @target_freq: New frequency to set (may be approximate).
1787  *
1788  * Carry out a fast frequency switch without sleeping.
1789  *
1790  * The driver's ->fast_switch() callback invoked by this function must be
1791  * suitable for being called from within RCU-sched read-side critical sections
1792  * and it is expected to select the minimum available frequency greater than or
1793  * equal to @target_freq (CPUFREQ_RELATION_L).
1794  *
1795  * This function must not be called if policy->fast_switch_enabled is unset.
1796  *
1797  * Governors calling this function must guarantee that it will never be invoked
1798  * twice in parallel for the same policy and that it will never be called in
1799  * parallel with either ->target() or ->target_index() for the same policy.
1800  *
1801  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1802  * callback to indicate an error condition, the hardware configuration must be
1803  * preserved.
1804  */
1805 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1806                                         unsigned int target_freq)
1807 {
1808         target_freq = clamp_val(target_freq, policy->min, policy->max);
1809
1810         return cpufreq_driver->fast_switch(policy, target_freq);
1811 }
1812 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1813
1814 /* Must set freqs->new to intermediate frequency */
1815 static int __target_intermediate(struct cpufreq_policy *policy,
1816                                  struct cpufreq_freqs *freqs, int index)
1817 {
1818         int ret;
1819
1820         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1821
1822         /* We don't need to switch to intermediate freq */
1823         if (!freqs->new)
1824                 return 0;
1825
1826         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1827                  __func__, policy->cpu, freqs->old, freqs->new);
1828
1829         cpufreq_freq_transition_begin(policy, freqs);
1830         ret = cpufreq_driver->target_intermediate(policy, index);
1831         cpufreq_freq_transition_end(policy, freqs, ret);
1832
1833         if (ret)
1834                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1835                        __func__, ret);
1836
1837         return ret;
1838 }
1839
1840 static int __target_index(struct cpufreq_policy *policy, int index)
1841 {
1842         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1843         unsigned int intermediate_freq = 0;
1844         unsigned int newfreq = policy->freq_table[index].frequency;
1845         int retval = -EINVAL;
1846         bool notify;
1847
1848         if (newfreq == policy->cur)
1849                 return 0;
1850
1851         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1852         if (notify) {
1853                 /* Handle switching to intermediate frequency */
1854                 if (cpufreq_driver->get_intermediate) {
1855                         retval = __target_intermediate(policy, &freqs, index);
1856                         if (retval)
1857                                 return retval;
1858
1859                         intermediate_freq = freqs.new;
1860                         /* Set old freq to intermediate */
1861                         if (intermediate_freq)
1862                                 freqs.old = freqs.new;
1863                 }
1864
1865                 freqs.new = newfreq;
1866                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1867                          __func__, policy->cpu, freqs.old, freqs.new);
1868
1869                 cpufreq_freq_transition_begin(policy, &freqs);
1870         }
1871
1872         retval = cpufreq_driver->target_index(policy, index);
1873         if (retval)
1874                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1875                        retval);
1876
1877         if (notify) {
1878                 cpufreq_freq_transition_end(policy, &freqs, retval);
1879
1880                 /*
1881                  * Failed after setting to intermediate freq? Driver should have
1882                  * reverted back to initial frequency and so should we. Check
1883                  * here for intermediate_freq instead of get_intermediate, in
1884                  * case we haven't switched to intermediate freq at all.
1885                  */
1886                 if (unlikely(retval && intermediate_freq)) {
1887                         freqs.old = intermediate_freq;
1888                         freqs.new = policy->restore_freq;
1889                         cpufreq_freq_transition_begin(policy, &freqs);
1890                         cpufreq_freq_transition_end(policy, &freqs, 0);
1891                 }
1892         }
1893
1894         return retval;
1895 }
1896
1897 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1898                             unsigned int target_freq,
1899                             unsigned int relation)
1900 {
1901         unsigned int old_target_freq = target_freq;
1902         int index;
1903
1904         if (cpufreq_disabled())
1905                 return -ENODEV;
1906
1907         /* Make sure that target_freq is within supported range */
1908         target_freq = clamp_val(target_freq, policy->min, policy->max);
1909
1910         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1911                  policy->cpu, target_freq, relation, old_target_freq);
1912
1913         /*
1914          * This might look like a redundant call as we are checking it again
1915          * after finding index. But it is left intentionally for cases where
1916          * exactly same freq is called again and so we can save on few function
1917          * calls.
1918          */
1919         if (target_freq == policy->cur)
1920                 return 0;
1921
1922         /* Save last value to restore later on errors */
1923         policy->restore_freq = policy->cur;
1924
1925         if (cpufreq_driver->target)
1926                 return cpufreq_driver->target(policy, target_freq, relation);
1927
1928         if (!cpufreq_driver->target_index)
1929                 return -EINVAL;
1930
1931         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1932
1933         return __target_index(policy, index);
1934 }
1935 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1936
1937 int cpufreq_driver_target(struct cpufreq_policy *policy,
1938                           unsigned int target_freq,
1939                           unsigned int relation)
1940 {
1941         int ret = -EINVAL;
1942
1943         down_write(&policy->rwsem);
1944
1945         ret = __cpufreq_driver_target(policy, target_freq, relation);
1946
1947         up_write(&policy->rwsem);
1948
1949         return ret;
1950 }
1951 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1952
1953 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1954 {
1955         return NULL;
1956 }
1957
1958 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1959 {
1960         int ret;
1961
1962         /* Don't start any governor operations if we are entering suspend */
1963         if (cpufreq_suspended)
1964                 return 0;
1965         /*
1966          * Governor might not be initiated here if ACPI _PPC changed
1967          * notification happened, so check it.
1968          */
1969         if (!policy->governor)
1970                 return -EINVAL;
1971
1972         if (policy->governor->max_transition_latency &&
1973             policy->cpuinfo.transition_latency >
1974             policy->governor->max_transition_latency) {
1975                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
1976
1977                 if (gov) {
1978                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1979                                 policy->governor->name, gov->name);
1980                         policy->governor = gov;
1981                 } else {
1982                         return -EINVAL;
1983                 }
1984         }
1985
1986         if (!try_module_get(policy->governor->owner))
1987                 return -EINVAL;
1988
1989         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
1990
1991         if (policy->governor->init) {
1992                 ret = policy->governor->init(policy);
1993                 if (ret) {
1994                         module_put(policy->governor->owner);
1995                         return ret;
1996                 }
1997         }
1998
1999         return 0;
2000 }
2001
2002 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2003 {
2004         if (cpufreq_suspended || !policy->governor)
2005                 return;
2006
2007         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2008
2009         if (policy->governor->exit)
2010                 policy->governor->exit(policy);
2011
2012         module_put(policy->governor->owner);
2013 }
2014
2015 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2016 {
2017         int ret;
2018
2019         if (cpufreq_suspended)
2020                 return 0;
2021
2022         if (!policy->governor)
2023                 return -EINVAL;
2024
2025         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2026
2027         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2028                 cpufreq_update_current_freq(policy);
2029
2030         if (policy->governor->start) {
2031                 ret = policy->governor->start(policy);
2032                 if (ret)
2033                         return ret;
2034         }
2035
2036         if (policy->governor->limits)
2037                 policy->governor->limits(policy);
2038
2039         return 0;
2040 }
2041
2042 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2043 {
2044         if (cpufreq_suspended || !policy->governor)
2045                 return;
2046
2047         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2048
2049         if (policy->governor->stop)
2050                 policy->governor->stop(policy);
2051 }
2052
2053 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2054 {
2055         if (cpufreq_suspended || !policy->governor)
2056                 return;
2057
2058         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2059
2060         if (policy->governor->limits)
2061                 policy->governor->limits(policy);
2062 }
2063
2064 int cpufreq_register_governor(struct cpufreq_governor *governor)
2065 {
2066         int err;
2067
2068         if (!governor)
2069                 return -EINVAL;
2070
2071         if (cpufreq_disabled())
2072                 return -ENODEV;
2073
2074         mutex_lock(&cpufreq_governor_mutex);
2075
2076         err = -EBUSY;
2077         if (!find_governor(governor->name)) {
2078                 err = 0;
2079                 list_add(&governor->governor_list, &cpufreq_governor_list);
2080         }
2081
2082         mutex_unlock(&cpufreq_governor_mutex);
2083         return err;
2084 }
2085 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2086
2087 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2088 {
2089         struct cpufreq_policy *policy;
2090         unsigned long flags;
2091
2092         if (!governor)
2093                 return;
2094
2095         if (cpufreq_disabled())
2096                 return;
2097
2098         /* clear last_governor for all inactive policies */
2099         read_lock_irqsave(&cpufreq_driver_lock, flags);
2100         for_each_inactive_policy(policy) {
2101                 if (!strcmp(policy->last_governor, governor->name)) {
2102                         policy->governor = NULL;
2103                         strcpy(policy->last_governor, "\0");
2104                 }
2105         }
2106         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2107
2108         mutex_lock(&cpufreq_governor_mutex);
2109         list_del(&governor->governor_list);
2110         mutex_unlock(&cpufreq_governor_mutex);
2111         return;
2112 }
2113 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2114
2115
2116 /*********************************************************************
2117  *                          POLICY INTERFACE                         *
2118  *********************************************************************/
2119
2120 /**
2121  * cpufreq_get_policy - get the current cpufreq_policy
2122  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2123  *      is written
2124  *
2125  * Reads the current cpufreq policy.
2126  */
2127 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2128 {
2129         struct cpufreq_policy *cpu_policy;
2130         if (!policy)
2131                 return -EINVAL;
2132
2133         cpu_policy = cpufreq_cpu_get(cpu);
2134         if (!cpu_policy)
2135                 return -EINVAL;
2136
2137         memcpy(policy, cpu_policy, sizeof(*policy));
2138
2139         cpufreq_cpu_put(cpu_policy);
2140         return 0;
2141 }
2142 EXPORT_SYMBOL(cpufreq_get_policy);
2143
2144 /*
2145  * policy : current policy.
2146  * new_policy: policy to be set.
2147  */
2148 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2149                                 struct cpufreq_policy *new_policy)
2150 {
2151         struct cpufreq_governor *old_gov;
2152         int ret;
2153
2154         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2155                  new_policy->cpu, new_policy->min, new_policy->max);
2156
2157         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2158
2159         /*
2160         * This check works well when we store new min/max freq attributes,
2161         * because new_policy is a copy of policy with one field updated.
2162         */
2163         if (new_policy->min > new_policy->max)
2164                 return -EINVAL;
2165
2166         /* verify the cpu speed can be set within this limit */
2167         ret = cpufreq_driver->verify(new_policy);
2168         if (ret)
2169                 return ret;
2170
2171         /* adjust if necessary - all reasons */
2172         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2173                         CPUFREQ_ADJUST, new_policy);
2174
2175         /*
2176          * verify the cpu speed can be set within this limit, which might be
2177          * different to the first one
2178          */
2179         ret = cpufreq_driver->verify(new_policy);
2180         if (ret)
2181                 return ret;
2182
2183         /* notification of the new policy */
2184         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2185                         CPUFREQ_NOTIFY, new_policy);
2186
2187         policy->min = new_policy->min;
2188         policy->max = new_policy->max;
2189
2190         policy->cached_target_freq = UINT_MAX;
2191
2192         pr_debug("new min and max freqs are %u - %u kHz\n",
2193                  policy->min, policy->max);
2194
2195         if (cpufreq_driver->setpolicy) {
2196                 policy->policy = new_policy->policy;
2197                 pr_debug("setting range\n");
2198                 return cpufreq_driver->setpolicy(new_policy);
2199         }
2200
2201         if (new_policy->governor == policy->governor) {
2202                 pr_debug("cpufreq: governor limits update\n");
2203                 cpufreq_governor_limits(policy);
2204                 return 0;
2205         }
2206
2207         pr_debug("governor switch\n");
2208
2209         /* save old, working values */
2210         old_gov = policy->governor;
2211         /* end old governor */
2212         if (old_gov) {
2213                 cpufreq_stop_governor(policy);
2214                 cpufreq_exit_governor(policy);
2215         }
2216
2217         /* start new governor */
2218         policy->governor = new_policy->governor;
2219         ret = cpufreq_init_governor(policy);
2220         if (!ret) {
2221                 ret = cpufreq_start_governor(policy);
2222                 if (!ret) {
2223                         pr_debug("cpufreq: governor change\n");
2224                         return 0;
2225                 }
2226                 cpufreq_exit_governor(policy);
2227         }
2228
2229         /* new governor failed, so re-start old one */
2230         pr_debug("starting governor %s failed\n", policy->governor->name);
2231         if (old_gov) {
2232                 policy->governor = old_gov;
2233                 if (cpufreq_init_governor(policy))
2234                         policy->governor = NULL;
2235                 else
2236                         cpufreq_start_governor(policy);
2237         }
2238
2239         return ret;
2240 }
2241
2242 /**
2243  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2244  *      @cpu: CPU which shall be re-evaluated
2245  *
2246  *      Useful for policy notifiers which have different necessities
2247  *      at different times.
2248  */
2249 void cpufreq_update_policy(unsigned int cpu)
2250 {
2251         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2252         struct cpufreq_policy new_policy;
2253
2254         if (!policy)
2255                 return;
2256
2257         down_write(&policy->rwsem);
2258
2259         if (policy_is_inactive(policy))
2260                 goto unlock;
2261
2262         pr_debug("updating policy for CPU %u\n", cpu);
2263         memcpy(&new_policy, policy, sizeof(*policy));
2264         new_policy.min = policy->user_policy.min;
2265         new_policy.max = policy->user_policy.max;
2266
2267         /*
2268          * BIOS might change freq behind our back
2269          * -> ask driver for current freq and notify governors about a change
2270          */
2271         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2272                 if (cpufreq_suspended)
2273                         goto unlock;
2274
2275                 new_policy.cur = cpufreq_update_current_freq(policy);
2276                 if (WARN_ON(!new_policy.cur))
2277                         goto unlock;
2278         }
2279
2280         cpufreq_set_policy(policy, &new_policy);
2281
2282 unlock:
2283         up_write(&policy->rwsem);
2284
2285         cpufreq_cpu_put(policy);
2286 }
2287 EXPORT_SYMBOL(cpufreq_update_policy);
2288
2289 /*********************************************************************
2290  *               BOOST                                               *
2291  *********************************************************************/
2292 static int cpufreq_boost_set_sw(int state)
2293 {
2294         struct cpufreq_policy *policy;
2295         int ret = -EINVAL;
2296
2297         for_each_active_policy(policy) {
2298                 if (!policy->freq_table)
2299                         continue;
2300
2301                 ret = cpufreq_frequency_table_cpuinfo(policy,
2302                                                       policy->freq_table);
2303                 if (ret) {
2304                         pr_err("%s: Policy frequency update failed\n",
2305                                __func__);
2306                         break;
2307                 }
2308
2309                 down_write(&policy->rwsem);
2310                 policy->user_policy.max = policy->max;
2311                 cpufreq_governor_limits(policy);
2312                 up_write(&policy->rwsem);
2313         }
2314
2315         return ret;
2316 }
2317
2318 int cpufreq_boost_trigger_state(int state)
2319 {
2320         unsigned long flags;
2321         int ret = 0;
2322
2323         if (cpufreq_driver->boost_enabled == state)
2324                 return 0;
2325
2326         write_lock_irqsave(&cpufreq_driver_lock, flags);
2327         cpufreq_driver->boost_enabled = state;
2328         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2329
2330         ret = cpufreq_driver->set_boost(state);
2331         if (ret) {
2332                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2333                 cpufreq_driver->boost_enabled = !state;
2334                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2335
2336                 pr_err("%s: Cannot %s BOOST\n",
2337                        __func__, state ? "enable" : "disable");
2338         }
2339
2340         return ret;
2341 }
2342
2343 static bool cpufreq_boost_supported(void)
2344 {
2345         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2346 }
2347
2348 static int create_boost_sysfs_file(void)
2349 {
2350         int ret;
2351
2352         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2353         if (ret)
2354                 pr_err("%s: cannot register global BOOST sysfs file\n",
2355                        __func__);
2356
2357         return ret;
2358 }
2359
2360 static void remove_boost_sysfs_file(void)
2361 {
2362         if (cpufreq_boost_supported())
2363                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2364 }
2365
2366 int cpufreq_enable_boost_support(void)
2367 {
2368         if (!cpufreq_driver)
2369                 return -EINVAL;
2370
2371         if (cpufreq_boost_supported())
2372                 return 0;
2373
2374         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2375
2376         /* This will get removed on driver unregister */
2377         return create_boost_sysfs_file();
2378 }
2379 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2380
2381 int cpufreq_boost_enabled(void)
2382 {
2383         return cpufreq_driver->boost_enabled;
2384 }
2385 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2386
2387 /*********************************************************************
2388  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2389  *********************************************************************/
2390 static enum cpuhp_state hp_online;
2391
2392 /**
2393  * cpufreq_register_driver - register a CPU Frequency driver
2394  * @driver_data: A struct cpufreq_driver containing the values#
2395  * submitted by the CPU Frequency driver.
2396  *
2397  * Registers a CPU Frequency driver to this core code. This code
2398  * returns zero on success, -EEXIST when another driver got here first
2399  * (and isn't unregistered in the meantime).
2400  *
2401  */
2402 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2403 {
2404         unsigned long flags;
2405         int ret;
2406
2407         if (cpufreq_disabled())
2408                 return -ENODEV;
2409
2410         if (!driver_data || !driver_data->verify || !driver_data->init ||
2411             !(driver_data->setpolicy || driver_data->target_index ||
2412                     driver_data->target) ||
2413              (driver_data->setpolicy && (driver_data->target_index ||
2414                     driver_data->target)) ||
2415              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2416                 return -EINVAL;
2417
2418         pr_debug("trying to register driver %s\n", driver_data->name);
2419
2420         /* Protect against concurrent CPU online/offline. */
2421         get_online_cpus();
2422
2423         write_lock_irqsave(&cpufreq_driver_lock, flags);
2424         if (cpufreq_driver) {
2425                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2426                 ret = -EEXIST;
2427                 goto out;
2428         }
2429         cpufreq_driver = driver_data;
2430         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2431
2432         if (driver_data->setpolicy)
2433                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2434
2435         if (cpufreq_boost_supported()) {
2436                 ret = create_boost_sysfs_file();
2437                 if (ret)
2438                         goto err_null_driver;
2439         }
2440
2441         ret = subsys_interface_register(&cpufreq_interface);
2442         if (ret)
2443                 goto err_boost_unreg;
2444
2445         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2446             list_empty(&cpufreq_policy_list)) {
2447                 /* if all ->init() calls failed, unregister */
2448                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2449                          driver_data->name);
2450                 goto err_if_unreg;
2451         }
2452
2453         ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "cpufreq:online",
2454                                         cpufreq_online,
2455                                         cpufreq_offline);
2456         if (ret < 0)
2457                 goto err_if_unreg;
2458         hp_online = ret;
2459         ret = 0;
2460
2461         pr_debug("driver %s up and running\n", driver_data->name);
2462         goto out;
2463
2464 err_if_unreg:
2465         subsys_interface_unregister(&cpufreq_interface);
2466 err_boost_unreg:
2467         remove_boost_sysfs_file();
2468 err_null_driver:
2469         write_lock_irqsave(&cpufreq_driver_lock, flags);
2470         cpufreq_driver = NULL;
2471         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2472 out:
2473         put_online_cpus();
2474         return ret;
2475 }
2476 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2477
2478 /**
2479  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2480  *
2481  * Unregister the current CPUFreq driver. Only call this if you have
2482  * the right to do so, i.e. if you have succeeded in initialising before!
2483  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2484  * currently not initialised.
2485  */
2486 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2487 {
2488         unsigned long flags;
2489
2490         if (!cpufreq_driver || (driver != cpufreq_driver))
2491                 return -EINVAL;
2492
2493         pr_debug("unregistering driver %s\n", driver->name);
2494
2495         /* Protect against concurrent cpu hotplug */
2496         get_online_cpus();
2497         subsys_interface_unregister(&cpufreq_interface);
2498         remove_boost_sysfs_file();
2499         cpuhp_remove_state_nocalls(hp_online);
2500
2501         write_lock_irqsave(&cpufreq_driver_lock, flags);
2502
2503         cpufreq_driver = NULL;
2504
2505         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2506         put_online_cpus();
2507
2508         return 0;
2509 }
2510 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2511
2512 /*
2513  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2514  * or mutexes when secondary CPUs are halted.
2515  */
2516 static struct syscore_ops cpufreq_syscore_ops = {
2517         .shutdown = cpufreq_suspend,
2518 };
2519
2520 struct kobject *cpufreq_global_kobject;
2521 EXPORT_SYMBOL(cpufreq_global_kobject);
2522
2523 static int __init cpufreq_core_init(void)
2524 {
2525         if (cpufreq_disabled())
2526                 return -ENODEV;
2527
2528         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2529         BUG_ON(!cpufreq_global_kobject);
2530
2531         register_syscore_ops(&cpufreq_syscore_ops);
2532
2533         return 0;
2534 }
2535 core_initcall(cpufreq_core_init);