Merge tag 'hwlock-v5.3' of git://github.com/andersson/remoteproc
[linux-2.6-block.git] / drivers / cpufreq / cpufreq.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/drivers/cpufreq/cpufreq.c
4  *
5  *  Copyright (C) 2001 Russell King
6  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
7  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
8  *
9  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
10  *      Added handling for CPU hotplug
11  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
12  *      Fix handling for CPU hotplug -- affected CPUs
13  */
14
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16
17 #include <linux/cpu.h>
18 #include <linux/cpufreq.h>
19 #include <linux/cpu_cooling.h>
20 #include <linux/delay.h>
21 #include <linux/device.h>
22 #include <linux/init.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/module.h>
25 #include <linux/mutex.h>
26 #include <linux/slab.h>
27 #include <linux/suspend.h>
28 #include <linux/syscore_ops.h>
29 #include <linux/tick.h>
30 #include <trace/events/power.h>
31
32 static LIST_HEAD(cpufreq_policy_list);
33
34 /* Macros to iterate over CPU policies */
35 #define for_each_suitable_policy(__policy, __active)                     \
36         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
37                 if ((__active) == !policy_is_inactive(__policy))
38
39 #define for_each_active_policy(__policy)                \
40         for_each_suitable_policy(__policy, true)
41 #define for_each_inactive_policy(__policy)              \
42         for_each_suitable_policy(__policy, false)
43
44 #define for_each_policy(__policy)                       \
45         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
46
47 /* Iterate over governors */
48 static LIST_HEAD(cpufreq_governor_list);
49 #define for_each_governor(__governor)                           \
50         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
51
52 /**
53  * The "cpufreq driver" - the arch- or hardware-dependent low
54  * level driver of CPUFreq support, and its spinlock. This lock
55  * also protects the cpufreq_cpu_data array.
56  */
57 static struct cpufreq_driver *cpufreq_driver;
58 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
59 static DEFINE_RWLOCK(cpufreq_driver_lock);
60
61 /* Flag to suspend/resume CPUFreq governors */
62 static bool cpufreq_suspended;
63
64 static inline bool has_target(void)
65 {
66         return cpufreq_driver->target_index || cpufreq_driver->target;
67 }
68
69 /* internal prototypes */
70 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
71 static int cpufreq_init_governor(struct cpufreq_policy *policy);
72 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
73 static int cpufreq_start_governor(struct cpufreq_policy *policy);
74 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
75 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
76
77 /**
78  * Two notifier lists: the "policy" list is involved in the
79  * validation process for a new CPU frequency policy; the
80  * "transition" list for kernel code that needs to handle
81  * changes to devices when the CPU clock speed changes.
82  * The mutex locks both lists.
83  */
84 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
85 SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
86
87 static int off __read_mostly;
88 static int cpufreq_disabled(void)
89 {
90         return off;
91 }
92 void disable_cpufreq(void)
93 {
94         off = 1;
95 }
96 static DEFINE_MUTEX(cpufreq_governor_mutex);
97
98 bool have_governor_per_policy(void)
99 {
100         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
101 }
102 EXPORT_SYMBOL_GPL(have_governor_per_policy);
103
104 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
105 {
106         if (have_governor_per_policy())
107                 return &policy->kobj;
108         else
109                 return cpufreq_global_kobject;
110 }
111 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
112
113 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
114 {
115         u64 idle_time;
116         u64 cur_wall_time;
117         u64 busy_time;
118
119         cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
120
121         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
122         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
123         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
124         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
125         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
126         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
127
128         idle_time = cur_wall_time - busy_time;
129         if (wall)
130                 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
131
132         return div_u64(idle_time, NSEC_PER_USEC);
133 }
134
135 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
136 {
137         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
138
139         if (idle_time == -1ULL)
140                 return get_cpu_idle_time_jiffy(cpu, wall);
141         else if (!io_busy)
142                 idle_time += get_cpu_iowait_time_us(cpu, wall);
143
144         return idle_time;
145 }
146 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
147
148 __weak void arch_set_freq_scale(struct cpumask *cpus, unsigned long cur_freq,
149                 unsigned long max_freq)
150 {
151 }
152 EXPORT_SYMBOL_GPL(arch_set_freq_scale);
153
154 /*
155  * This is a generic cpufreq init() routine which can be used by cpufreq
156  * drivers of SMP systems. It will do following:
157  * - validate & show freq table passed
158  * - set policies transition latency
159  * - policy->cpus with all possible CPUs
160  */
161 int cpufreq_generic_init(struct cpufreq_policy *policy,
162                 struct cpufreq_frequency_table *table,
163                 unsigned int transition_latency)
164 {
165         policy->freq_table = table;
166         policy->cpuinfo.transition_latency = transition_latency;
167
168         /*
169          * The driver only supports the SMP configuration where all processors
170          * share the clock and voltage and clock.
171          */
172         cpumask_setall(policy->cpus);
173
174         return 0;
175 }
176 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
177
178 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
179 {
180         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
181
182         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
183 }
184 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
185
186 unsigned int cpufreq_generic_get(unsigned int cpu)
187 {
188         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
189
190         if (!policy || IS_ERR(policy->clk)) {
191                 pr_err("%s: No %s associated to cpu: %d\n",
192                        __func__, policy ? "clk" : "policy", cpu);
193                 return 0;
194         }
195
196         return clk_get_rate(policy->clk) / 1000;
197 }
198 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
199
200 /**
201  * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
202  * @cpu: CPU to find the policy for.
203  *
204  * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
205  * the kobject reference counter of that policy.  Return a valid policy on
206  * success or NULL on failure.
207  *
208  * The policy returned by this function has to be released with the help of
209  * cpufreq_cpu_put() to balance its kobject reference counter properly.
210  */
211 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
212 {
213         struct cpufreq_policy *policy = NULL;
214         unsigned long flags;
215
216         if (WARN_ON(cpu >= nr_cpu_ids))
217                 return NULL;
218
219         /* get the cpufreq driver */
220         read_lock_irqsave(&cpufreq_driver_lock, flags);
221
222         if (cpufreq_driver) {
223                 /* get the CPU */
224                 policy = cpufreq_cpu_get_raw(cpu);
225                 if (policy)
226                         kobject_get(&policy->kobj);
227         }
228
229         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
230
231         return policy;
232 }
233 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
234
235 /**
236  * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
237  * @policy: cpufreq policy returned by cpufreq_cpu_get().
238  */
239 void cpufreq_cpu_put(struct cpufreq_policy *policy)
240 {
241         kobject_put(&policy->kobj);
242 }
243 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
244
245 /**
246  * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
247  * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
248  */
249 void cpufreq_cpu_release(struct cpufreq_policy *policy)
250 {
251         if (WARN_ON(!policy))
252                 return;
253
254         lockdep_assert_held(&policy->rwsem);
255
256         up_write(&policy->rwsem);
257
258         cpufreq_cpu_put(policy);
259 }
260
261 /**
262  * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
263  * @cpu: CPU to find the policy for.
264  *
265  * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
266  * if the policy returned by it is not NULL, acquire its rwsem for writing.
267  * Return the policy if it is active or release it and return NULL otherwise.
268  *
269  * The policy returned by this function has to be released with the help of
270  * cpufreq_cpu_release() in order to release its rwsem and balance its usage
271  * counter properly.
272  */
273 struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
274 {
275         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
276
277         if (!policy)
278                 return NULL;
279
280         down_write(&policy->rwsem);
281
282         if (policy_is_inactive(policy)) {
283                 cpufreq_cpu_release(policy);
284                 return NULL;
285         }
286
287         return policy;
288 }
289
290 /*********************************************************************
291  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
292  *********************************************************************/
293
294 /**
295  * adjust_jiffies - adjust the system "loops_per_jiffy"
296  *
297  * This function alters the system "loops_per_jiffy" for the clock
298  * speed change. Note that loops_per_jiffy cannot be updated on SMP
299  * systems as each CPU might be scaled differently. So, use the arch
300  * per-CPU loops_per_jiffy value wherever possible.
301  */
302 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
303 {
304 #ifndef CONFIG_SMP
305         static unsigned long l_p_j_ref;
306         static unsigned int l_p_j_ref_freq;
307
308         if (ci->flags & CPUFREQ_CONST_LOOPS)
309                 return;
310
311         if (!l_p_j_ref_freq) {
312                 l_p_j_ref = loops_per_jiffy;
313                 l_p_j_ref_freq = ci->old;
314                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
315                          l_p_j_ref, l_p_j_ref_freq);
316         }
317         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
318                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
319                                                                 ci->new);
320                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
321                          loops_per_jiffy, ci->new);
322         }
323 #endif
324 }
325
326 /**
327  * cpufreq_notify_transition - Notify frequency transition and adjust_jiffies.
328  * @policy: cpufreq policy to enable fast frequency switching for.
329  * @freqs: contain details of the frequency update.
330  * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
331  *
332  * This function calls the transition notifiers and the "adjust_jiffies"
333  * function. It is called twice on all CPU frequency changes that have
334  * external effects.
335  */
336 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
337                                       struct cpufreq_freqs *freqs,
338                                       unsigned int state)
339 {
340         int cpu;
341
342         BUG_ON(irqs_disabled());
343
344         if (cpufreq_disabled())
345                 return;
346
347         freqs->policy = policy;
348         freqs->flags = cpufreq_driver->flags;
349         pr_debug("notification %u of frequency transition to %u kHz\n",
350                  state, freqs->new);
351
352         switch (state) {
353         case CPUFREQ_PRECHANGE:
354                 /*
355                  * Detect if the driver reported a value as "old frequency"
356                  * which is not equal to what the cpufreq core thinks is
357                  * "old frequency".
358                  */
359                 if (policy->cur && policy->cur != freqs->old) {
360                         pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
361                                  freqs->old, policy->cur);
362                         freqs->old = policy->cur;
363                 }
364
365                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
366                                          CPUFREQ_PRECHANGE, freqs);
367
368                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
369                 break;
370
371         case CPUFREQ_POSTCHANGE:
372                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
373                 pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
374                          cpumask_pr_args(policy->cpus));
375
376                 for_each_cpu(cpu, policy->cpus)
377                         trace_cpu_frequency(freqs->new, cpu);
378
379                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
380                                          CPUFREQ_POSTCHANGE, freqs);
381
382                 cpufreq_stats_record_transition(policy, freqs->new);
383                 policy->cur = freqs->new;
384         }
385 }
386
387 /* Do post notifications when there are chances that transition has failed */
388 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
389                 struct cpufreq_freqs *freqs, int transition_failed)
390 {
391         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
392         if (!transition_failed)
393                 return;
394
395         swap(freqs->old, freqs->new);
396         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
397         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
398 }
399
400 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
401                 struct cpufreq_freqs *freqs)
402 {
403
404         /*
405          * Catch double invocations of _begin() which lead to self-deadlock.
406          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
407          * doesn't invoke _begin() on their behalf, and hence the chances of
408          * double invocations are very low. Moreover, there are scenarios
409          * where these checks can emit false-positive warnings in these
410          * drivers; so we avoid that by skipping them altogether.
411          */
412         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
413                                 && current == policy->transition_task);
414
415 wait:
416         wait_event(policy->transition_wait, !policy->transition_ongoing);
417
418         spin_lock(&policy->transition_lock);
419
420         if (unlikely(policy->transition_ongoing)) {
421                 spin_unlock(&policy->transition_lock);
422                 goto wait;
423         }
424
425         policy->transition_ongoing = true;
426         policy->transition_task = current;
427
428         spin_unlock(&policy->transition_lock);
429
430         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
431 }
432 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
433
434 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
435                 struct cpufreq_freqs *freqs, int transition_failed)
436 {
437         if (WARN_ON(!policy->transition_ongoing))
438                 return;
439
440         cpufreq_notify_post_transition(policy, freqs, transition_failed);
441
442         policy->transition_ongoing = false;
443         policy->transition_task = NULL;
444
445         wake_up(&policy->transition_wait);
446 }
447 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
448
449 /*
450  * Fast frequency switching status count.  Positive means "enabled", negative
451  * means "disabled" and 0 means "not decided yet".
452  */
453 static int cpufreq_fast_switch_count;
454 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
455
456 static void cpufreq_list_transition_notifiers(void)
457 {
458         struct notifier_block *nb;
459
460         pr_info("Registered transition notifiers:\n");
461
462         mutex_lock(&cpufreq_transition_notifier_list.mutex);
463
464         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
465                 pr_info("%pS\n", nb->notifier_call);
466
467         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
468 }
469
470 /**
471  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
472  * @policy: cpufreq policy to enable fast frequency switching for.
473  *
474  * Try to enable fast frequency switching for @policy.
475  *
476  * The attempt will fail if there is at least one transition notifier registered
477  * at this point, as fast frequency switching is quite fundamentally at odds
478  * with transition notifiers.  Thus if successful, it will make registration of
479  * transition notifiers fail going forward.
480  */
481 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
482 {
483         lockdep_assert_held(&policy->rwsem);
484
485         if (!policy->fast_switch_possible)
486                 return;
487
488         mutex_lock(&cpufreq_fast_switch_lock);
489         if (cpufreq_fast_switch_count >= 0) {
490                 cpufreq_fast_switch_count++;
491                 policy->fast_switch_enabled = true;
492         } else {
493                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
494                         policy->cpu);
495                 cpufreq_list_transition_notifiers();
496         }
497         mutex_unlock(&cpufreq_fast_switch_lock);
498 }
499 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
500
501 /**
502  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
503  * @policy: cpufreq policy to disable fast frequency switching for.
504  */
505 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
506 {
507         mutex_lock(&cpufreq_fast_switch_lock);
508         if (policy->fast_switch_enabled) {
509                 policy->fast_switch_enabled = false;
510                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
511                         cpufreq_fast_switch_count--;
512         }
513         mutex_unlock(&cpufreq_fast_switch_lock);
514 }
515 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
516
517 /**
518  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
519  * one.
520  * @target_freq: target frequency to resolve.
521  *
522  * The target to driver frequency mapping is cached in the policy.
523  *
524  * Return: Lowest driver-supported frequency greater than or equal to the
525  * given target_freq, subject to policy (min/max) and driver limitations.
526  */
527 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
528                                          unsigned int target_freq)
529 {
530         target_freq = clamp_val(target_freq, policy->min, policy->max);
531         policy->cached_target_freq = target_freq;
532
533         if (cpufreq_driver->target_index) {
534                 int idx;
535
536                 idx = cpufreq_frequency_table_target(policy, target_freq,
537                                                      CPUFREQ_RELATION_L);
538                 policy->cached_resolved_idx = idx;
539                 return policy->freq_table[idx].frequency;
540         }
541
542         if (cpufreq_driver->resolve_freq)
543                 return cpufreq_driver->resolve_freq(policy, target_freq);
544
545         return target_freq;
546 }
547 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
548
549 unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
550 {
551         unsigned int latency;
552
553         if (policy->transition_delay_us)
554                 return policy->transition_delay_us;
555
556         latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
557         if (latency) {
558                 /*
559                  * For platforms that can change the frequency very fast (< 10
560                  * us), the above formula gives a decent transition delay. But
561                  * for platforms where transition_latency is in milliseconds, it
562                  * ends up giving unrealistic values.
563                  *
564                  * Cap the default transition delay to 10 ms, which seems to be
565                  * a reasonable amount of time after which we should reevaluate
566                  * the frequency.
567                  */
568                 return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
569         }
570
571         return LATENCY_MULTIPLIER;
572 }
573 EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
574
575 /*********************************************************************
576  *                          SYSFS INTERFACE                          *
577  *********************************************************************/
578 static ssize_t show_boost(struct kobject *kobj,
579                           struct kobj_attribute *attr, char *buf)
580 {
581         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
582 }
583
584 static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
585                            const char *buf, size_t count)
586 {
587         int ret, enable;
588
589         ret = sscanf(buf, "%d", &enable);
590         if (ret != 1 || enable < 0 || enable > 1)
591                 return -EINVAL;
592
593         if (cpufreq_boost_trigger_state(enable)) {
594                 pr_err("%s: Cannot %s BOOST!\n",
595                        __func__, enable ? "enable" : "disable");
596                 return -EINVAL;
597         }
598
599         pr_debug("%s: cpufreq BOOST %s\n",
600                  __func__, enable ? "enabled" : "disabled");
601
602         return count;
603 }
604 define_one_global_rw(boost);
605
606 static struct cpufreq_governor *find_governor(const char *str_governor)
607 {
608         struct cpufreq_governor *t;
609
610         for_each_governor(t)
611                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
612                         return t;
613
614         return NULL;
615 }
616
617 static int cpufreq_parse_policy(char *str_governor,
618                                 struct cpufreq_policy *policy)
619 {
620         if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
621                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
622                 return 0;
623         }
624         if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
625                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
626                 return 0;
627         }
628         return -EINVAL;
629 }
630
631 /**
632  * cpufreq_parse_governor - parse a governor string only for has_target()
633  */
634 static int cpufreq_parse_governor(char *str_governor,
635                                   struct cpufreq_policy *policy)
636 {
637         struct cpufreq_governor *t;
638
639         mutex_lock(&cpufreq_governor_mutex);
640
641         t = find_governor(str_governor);
642         if (!t) {
643                 int ret;
644
645                 mutex_unlock(&cpufreq_governor_mutex);
646
647                 ret = request_module("cpufreq_%s", str_governor);
648                 if (ret)
649                         return -EINVAL;
650
651                 mutex_lock(&cpufreq_governor_mutex);
652
653                 t = find_governor(str_governor);
654         }
655         if (t && !try_module_get(t->owner))
656                 t = NULL;
657
658         mutex_unlock(&cpufreq_governor_mutex);
659
660         if (t) {
661                 policy->governor = t;
662                 return 0;
663         }
664
665         return -EINVAL;
666 }
667
668 /**
669  * cpufreq_per_cpu_attr_read() / show_##file_name() -
670  * print out cpufreq information
671  *
672  * Write out information from cpufreq_driver->policy[cpu]; object must be
673  * "unsigned int".
674  */
675
676 #define show_one(file_name, object)                     \
677 static ssize_t show_##file_name                         \
678 (struct cpufreq_policy *policy, char *buf)              \
679 {                                                       \
680         return sprintf(buf, "%u\n", policy->object);    \
681 }
682
683 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
684 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
685 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
686 show_one(scaling_min_freq, min);
687 show_one(scaling_max_freq, max);
688
689 __weak unsigned int arch_freq_get_on_cpu(int cpu)
690 {
691         return 0;
692 }
693
694 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
695 {
696         ssize_t ret;
697         unsigned int freq;
698
699         freq = arch_freq_get_on_cpu(policy->cpu);
700         if (freq)
701                 ret = sprintf(buf, "%u\n", freq);
702         else if (cpufreq_driver && cpufreq_driver->setpolicy &&
703                         cpufreq_driver->get)
704                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
705         else
706                 ret = sprintf(buf, "%u\n", policy->cur);
707         return ret;
708 }
709
710 /**
711  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
712  */
713 #define store_one(file_name, object)                    \
714 static ssize_t store_##file_name                                        \
715 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
716 {                                                                       \
717         int ret, temp;                                                  \
718         struct cpufreq_policy new_policy;                               \
719                                                                         \
720         memcpy(&new_policy, policy, sizeof(*policy));                   \
721         new_policy.min = policy->user_policy.min;                       \
722         new_policy.max = policy->user_policy.max;                       \
723                                                                         \
724         ret = sscanf(buf, "%u", &new_policy.object);                    \
725         if (ret != 1)                                                   \
726                 return -EINVAL;                                         \
727                                                                         \
728         temp = new_policy.object;                                       \
729         ret = cpufreq_set_policy(policy, &new_policy);          \
730         if (!ret)                                                       \
731                 policy->user_policy.object = temp;                      \
732                                                                         \
733         return ret ? ret : count;                                       \
734 }
735
736 store_one(scaling_min_freq, min);
737 store_one(scaling_max_freq, max);
738
739 /**
740  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
741  */
742 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
743                                         char *buf)
744 {
745         unsigned int cur_freq = __cpufreq_get(policy);
746
747         if (cur_freq)
748                 return sprintf(buf, "%u\n", cur_freq);
749
750         return sprintf(buf, "<unknown>\n");
751 }
752
753 /**
754  * show_scaling_governor - show the current policy for the specified CPU
755  */
756 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
757 {
758         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
759                 return sprintf(buf, "powersave\n");
760         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
761                 return sprintf(buf, "performance\n");
762         else if (policy->governor)
763                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
764                                 policy->governor->name);
765         return -EINVAL;
766 }
767
768 /**
769  * store_scaling_governor - store policy for the specified CPU
770  */
771 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
772                                         const char *buf, size_t count)
773 {
774         int ret;
775         char    str_governor[16];
776         struct cpufreq_policy new_policy;
777
778         memcpy(&new_policy, policy, sizeof(*policy));
779
780         ret = sscanf(buf, "%15s", str_governor);
781         if (ret != 1)
782                 return -EINVAL;
783
784         if (cpufreq_driver->setpolicy) {
785                 if (cpufreq_parse_policy(str_governor, &new_policy))
786                         return -EINVAL;
787         } else {
788                 if (cpufreq_parse_governor(str_governor, &new_policy))
789                         return -EINVAL;
790         }
791
792         ret = cpufreq_set_policy(policy, &new_policy);
793
794         if (new_policy.governor)
795                 module_put(new_policy.governor->owner);
796
797         return ret ? ret : count;
798 }
799
800 /**
801  * show_scaling_driver - show the cpufreq driver currently loaded
802  */
803 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
804 {
805         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
806 }
807
808 /**
809  * show_scaling_available_governors - show the available CPUfreq governors
810  */
811 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
812                                                 char *buf)
813 {
814         ssize_t i = 0;
815         struct cpufreq_governor *t;
816
817         if (!has_target()) {
818                 i += sprintf(buf, "performance powersave");
819                 goto out;
820         }
821
822         for_each_governor(t) {
823                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
824                     - (CPUFREQ_NAME_LEN + 2)))
825                         goto out;
826                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
827         }
828 out:
829         i += sprintf(&buf[i], "\n");
830         return i;
831 }
832
833 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
834 {
835         ssize_t i = 0;
836         unsigned int cpu;
837
838         for_each_cpu(cpu, mask) {
839                 if (i)
840                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
841                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
842                 if (i >= (PAGE_SIZE - 5))
843                         break;
844         }
845         i += sprintf(&buf[i], "\n");
846         return i;
847 }
848 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
849
850 /**
851  * show_related_cpus - show the CPUs affected by each transition even if
852  * hw coordination is in use
853  */
854 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
855 {
856         return cpufreq_show_cpus(policy->related_cpus, buf);
857 }
858
859 /**
860  * show_affected_cpus - show the CPUs affected by each transition
861  */
862 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
863 {
864         return cpufreq_show_cpus(policy->cpus, buf);
865 }
866
867 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
868                                         const char *buf, size_t count)
869 {
870         unsigned int freq = 0;
871         unsigned int ret;
872
873         if (!policy->governor || !policy->governor->store_setspeed)
874                 return -EINVAL;
875
876         ret = sscanf(buf, "%u", &freq);
877         if (ret != 1)
878                 return -EINVAL;
879
880         policy->governor->store_setspeed(policy, freq);
881
882         return count;
883 }
884
885 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
886 {
887         if (!policy->governor || !policy->governor->show_setspeed)
888                 return sprintf(buf, "<unsupported>\n");
889
890         return policy->governor->show_setspeed(policy, buf);
891 }
892
893 /**
894  * show_bios_limit - show the current cpufreq HW/BIOS limitation
895  */
896 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
897 {
898         unsigned int limit;
899         int ret;
900         ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
901         if (!ret)
902                 return sprintf(buf, "%u\n", limit);
903         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
904 }
905
906 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
907 cpufreq_freq_attr_ro(cpuinfo_min_freq);
908 cpufreq_freq_attr_ro(cpuinfo_max_freq);
909 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
910 cpufreq_freq_attr_ro(scaling_available_governors);
911 cpufreq_freq_attr_ro(scaling_driver);
912 cpufreq_freq_attr_ro(scaling_cur_freq);
913 cpufreq_freq_attr_ro(bios_limit);
914 cpufreq_freq_attr_ro(related_cpus);
915 cpufreq_freq_attr_ro(affected_cpus);
916 cpufreq_freq_attr_rw(scaling_min_freq);
917 cpufreq_freq_attr_rw(scaling_max_freq);
918 cpufreq_freq_attr_rw(scaling_governor);
919 cpufreq_freq_attr_rw(scaling_setspeed);
920
921 static struct attribute *default_attrs[] = {
922         &cpuinfo_min_freq.attr,
923         &cpuinfo_max_freq.attr,
924         &cpuinfo_transition_latency.attr,
925         &scaling_min_freq.attr,
926         &scaling_max_freq.attr,
927         &affected_cpus.attr,
928         &related_cpus.attr,
929         &scaling_governor.attr,
930         &scaling_driver.attr,
931         &scaling_available_governors.attr,
932         &scaling_setspeed.attr,
933         NULL
934 };
935
936 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
937 #define to_attr(a) container_of(a, struct freq_attr, attr)
938
939 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
940 {
941         struct cpufreq_policy *policy = to_policy(kobj);
942         struct freq_attr *fattr = to_attr(attr);
943         ssize_t ret;
944
945         down_read(&policy->rwsem);
946         ret = fattr->show(policy, buf);
947         up_read(&policy->rwsem);
948
949         return ret;
950 }
951
952 static ssize_t store(struct kobject *kobj, struct attribute *attr,
953                      const char *buf, size_t count)
954 {
955         struct cpufreq_policy *policy = to_policy(kobj);
956         struct freq_attr *fattr = to_attr(attr);
957         ssize_t ret = -EINVAL;
958
959         /*
960          * cpus_read_trylock() is used here to work around a circular lock
961          * dependency problem with respect to the cpufreq_register_driver().
962          */
963         if (!cpus_read_trylock())
964                 return -EBUSY;
965
966         if (cpu_online(policy->cpu)) {
967                 down_write(&policy->rwsem);
968                 ret = fattr->store(policy, buf, count);
969                 up_write(&policy->rwsem);
970         }
971
972         cpus_read_unlock();
973
974         return ret;
975 }
976
977 static void cpufreq_sysfs_release(struct kobject *kobj)
978 {
979         struct cpufreq_policy *policy = to_policy(kobj);
980         pr_debug("last reference is dropped\n");
981         complete(&policy->kobj_unregister);
982 }
983
984 static const struct sysfs_ops sysfs_ops = {
985         .show   = show,
986         .store  = store,
987 };
988
989 static struct kobj_type ktype_cpufreq = {
990         .sysfs_ops      = &sysfs_ops,
991         .default_attrs  = default_attrs,
992         .release        = cpufreq_sysfs_release,
993 };
994
995 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
996 {
997         struct device *dev = get_cpu_device(cpu);
998
999         if (!dev)
1000                 return;
1001
1002         if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1003                 return;
1004
1005         dev_dbg(dev, "%s: Adding symlink\n", __func__);
1006         if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1007                 dev_err(dev, "cpufreq symlink creation failed\n");
1008 }
1009
1010 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
1011                                    struct device *dev)
1012 {
1013         dev_dbg(dev, "%s: Removing symlink\n", __func__);
1014         sysfs_remove_link(&dev->kobj, "cpufreq");
1015 }
1016
1017 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1018 {
1019         struct freq_attr **drv_attr;
1020         int ret = 0;
1021
1022         /* set up files for this cpu device */
1023         drv_attr = cpufreq_driver->attr;
1024         while (drv_attr && *drv_attr) {
1025                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1026                 if (ret)
1027                         return ret;
1028                 drv_attr++;
1029         }
1030         if (cpufreq_driver->get) {
1031                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1032                 if (ret)
1033                         return ret;
1034         }
1035
1036         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1037         if (ret)
1038                 return ret;
1039
1040         if (cpufreq_driver->bios_limit) {
1041                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1042                 if (ret)
1043                         return ret;
1044         }
1045
1046         return 0;
1047 }
1048
1049 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1050 {
1051         return NULL;
1052 }
1053
1054 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1055 {
1056         struct cpufreq_governor *gov = NULL, *def_gov = NULL;
1057         struct cpufreq_policy new_policy;
1058
1059         memcpy(&new_policy, policy, sizeof(*policy));
1060
1061         def_gov = cpufreq_default_governor();
1062
1063         if (has_target()) {
1064                 /*
1065                  * Update governor of new_policy to the governor used before
1066                  * hotplug
1067                  */
1068                 gov = find_governor(policy->last_governor);
1069                 if (gov) {
1070                         pr_debug("Restoring governor %s for cpu %d\n",
1071                                 policy->governor->name, policy->cpu);
1072                 } else {
1073                         if (!def_gov)
1074                                 return -ENODATA;
1075                         gov = def_gov;
1076                 }
1077                 new_policy.governor = gov;
1078         } else {
1079                 /* Use the default policy if there is no last_policy. */
1080                 if (policy->last_policy) {
1081                         new_policy.policy = policy->last_policy;
1082                 } else {
1083                         if (!def_gov)
1084                                 return -ENODATA;
1085                         cpufreq_parse_policy(def_gov->name, &new_policy);
1086                 }
1087         }
1088
1089         return cpufreq_set_policy(policy, &new_policy);
1090 }
1091
1092 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1093 {
1094         int ret = 0;
1095
1096         /* Has this CPU been taken care of already? */
1097         if (cpumask_test_cpu(cpu, policy->cpus))
1098                 return 0;
1099
1100         down_write(&policy->rwsem);
1101         if (has_target())
1102                 cpufreq_stop_governor(policy);
1103
1104         cpumask_set_cpu(cpu, policy->cpus);
1105
1106         if (has_target()) {
1107                 ret = cpufreq_start_governor(policy);
1108                 if (ret)
1109                         pr_err("%s: Failed to start governor\n", __func__);
1110         }
1111         up_write(&policy->rwsem);
1112         return ret;
1113 }
1114
1115 static void refresh_frequency_limits(struct cpufreq_policy *policy)
1116 {
1117         struct cpufreq_policy new_policy = *policy;
1118
1119         pr_debug("updating policy for CPU %u\n", policy->cpu);
1120
1121         new_policy.min = policy->user_policy.min;
1122         new_policy.max = policy->user_policy.max;
1123
1124         cpufreq_set_policy(policy, &new_policy);
1125 }
1126
1127 static void handle_update(struct work_struct *work)
1128 {
1129         struct cpufreq_policy *policy =
1130                 container_of(work, struct cpufreq_policy, update);
1131
1132         pr_debug("handle_update for cpu %u called\n", policy->cpu);
1133         refresh_frequency_limits(policy);
1134 }
1135
1136 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1137 {
1138         struct cpufreq_policy *policy;
1139         int ret;
1140
1141         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1142         if (!policy)
1143                 return NULL;
1144
1145         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1146                 goto err_free_policy;
1147
1148         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1149                 goto err_free_cpumask;
1150
1151         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1152                 goto err_free_rcpumask;
1153
1154         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1155                                    cpufreq_global_kobject, "policy%u", cpu);
1156         if (ret) {
1157                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1158                 /*
1159                  * The entire policy object will be freed below, but the extra
1160                  * memory allocated for the kobject name needs to be freed by
1161                  * releasing the kobject.
1162                  */
1163                 kobject_put(&policy->kobj);
1164                 goto err_free_real_cpus;
1165         }
1166
1167         INIT_LIST_HEAD(&policy->policy_list);
1168         init_rwsem(&policy->rwsem);
1169         spin_lock_init(&policy->transition_lock);
1170         init_waitqueue_head(&policy->transition_wait);
1171         init_completion(&policy->kobj_unregister);
1172         INIT_WORK(&policy->update, handle_update);
1173
1174         policy->cpu = cpu;
1175         return policy;
1176
1177 err_free_real_cpus:
1178         free_cpumask_var(policy->real_cpus);
1179 err_free_rcpumask:
1180         free_cpumask_var(policy->related_cpus);
1181 err_free_cpumask:
1182         free_cpumask_var(policy->cpus);
1183 err_free_policy:
1184         kfree(policy);
1185
1186         return NULL;
1187 }
1188
1189 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1190 {
1191         struct kobject *kobj;
1192         struct completion *cmp;
1193
1194         down_write(&policy->rwsem);
1195         cpufreq_stats_free_table(policy);
1196         kobj = &policy->kobj;
1197         cmp = &policy->kobj_unregister;
1198         up_write(&policy->rwsem);
1199         kobject_put(kobj);
1200
1201         /*
1202          * We need to make sure that the underlying kobj is
1203          * actually not referenced anymore by anybody before we
1204          * proceed with unloading.
1205          */
1206         pr_debug("waiting for dropping of refcount\n");
1207         wait_for_completion(cmp);
1208         pr_debug("wait complete\n");
1209 }
1210
1211 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1212 {
1213         unsigned long flags;
1214         int cpu;
1215
1216         /* Remove policy from list */
1217         write_lock_irqsave(&cpufreq_driver_lock, flags);
1218         list_del(&policy->policy_list);
1219
1220         for_each_cpu(cpu, policy->related_cpus)
1221                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1222         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1223
1224         cpufreq_policy_put_kobj(policy);
1225         free_cpumask_var(policy->real_cpus);
1226         free_cpumask_var(policy->related_cpus);
1227         free_cpumask_var(policy->cpus);
1228         kfree(policy);
1229 }
1230
1231 static int cpufreq_online(unsigned int cpu)
1232 {
1233         struct cpufreq_policy *policy;
1234         bool new_policy;
1235         unsigned long flags;
1236         unsigned int j;
1237         int ret;
1238
1239         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1240
1241         /* Check if this CPU already has a policy to manage it */
1242         policy = per_cpu(cpufreq_cpu_data, cpu);
1243         if (policy) {
1244                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1245                 if (!policy_is_inactive(policy))
1246                         return cpufreq_add_policy_cpu(policy, cpu);
1247
1248                 /* This is the only online CPU for the policy.  Start over. */
1249                 new_policy = false;
1250                 down_write(&policy->rwsem);
1251                 policy->cpu = cpu;
1252                 policy->governor = NULL;
1253                 up_write(&policy->rwsem);
1254         } else {
1255                 new_policy = true;
1256                 policy = cpufreq_policy_alloc(cpu);
1257                 if (!policy)
1258                         return -ENOMEM;
1259         }
1260
1261         if (!new_policy && cpufreq_driver->online) {
1262                 ret = cpufreq_driver->online(policy);
1263                 if (ret) {
1264                         pr_debug("%s: %d: initialization failed\n", __func__,
1265                                  __LINE__);
1266                         goto out_exit_policy;
1267                 }
1268
1269                 /* Recover policy->cpus using related_cpus */
1270                 cpumask_copy(policy->cpus, policy->related_cpus);
1271         } else {
1272                 cpumask_copy(policy->cpus, cpumask_of(cpu));
1273
1274                 /*
1275                  * Call driver. From then on the cpufreq must be able
1276                  * to accept all calls to ->verify and ->setpolicy for this CPU.
1277                  */
1278                 ret = cpufreq_driver->init(policy);
1279                 if (ret) {
1280                         pr_debug("%s: %d: initialization failed\n", __func__,
1281                                  __LINE__);
1282                         goto out_free_policy;
1283                 }
1284
1285                 ret = cpufreq_table_validate_and_sort(policy);
1286                 if (ret)
1287                         goto out_exit_policy;
1288
1289                 /* related_cpus should at least include policy->cpus. */
1290                 cpumask_copy(policy->related_cpus, policy->cpus);
1291         }
1292
1293         down_write(&policy->rwsem);
1294         /*
1295          * affected cpus must always be the one, which are online. We aren't
1296          * managing offline cpus here.
1297          */
1298         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1299
1300         if (new_policy) {
1301                 policy->user_policy.min = policy->min;
1302                 policy->user_policy.max = policy->max;
1303
1304                 for_each_cpu(j, policy->related_cpus) {
1305                         per_cpu(cpufreq_cpu_data, j) = policy;
1306                         add_cpu_dev_symlink(policy, j);
1307                 }
1308         } else {
1309                 policy->min = policy->user_policy.min;
1310                 policy->max = policy->user_policy.max;
1311         }
1312
1313         if (cpufreq_driver->get && has_target()) {
1314                 policy->cur = cpufreq_driver->get(policy->cpu);
1315                 if (!policy->cur) {
1316                         pr_err("%s: ->get() failed\n", __func__);
1317                         goto out_destroy_policy;
1318                 }
1319         }
1320
1321         /*
1322          * Sometimes boot loaders set CPU frequency to a value outside of
1323          * frequency table present with cpufreq core. In such cases CPU might be
1324          * unstable if it has to run on that frequency for long duration of time
1325          * and so its better to set it to a frequency which is specified in
1326          * freq-table. This also makes cpufreq stats inconsistent as
1327          * cpufreq-stats would fail to register because current frequency of CPU
1328          * isn't found in freq-table.
1329          *
1330          * Because we don't want this change to effect boot process badly, we go
1331          * for the next freq which is >= policy->cur ('cur' must be set by now,
1332          * otherwise we will end up setting freq to lowest of the table as 'cur'
1333          * is initialized to zero).
1334          *
1335          * We are passing target-freq as "policy->cur - 1" otherwise
1336          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1337          * equal to target-freq.
1338          */
1339         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1340             && has_target()) {
1341                 /* Are we running at unknown frequency ? */
1342                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1343                 if (ret == -EINVAL) {
1344                         /* Warn user and fix it */
1345                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1346                                 __func__, policy->cpu, policy->cur);
1347                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1348                                 CPUFREQ_RELATION_L);
1349
1350                         /*
1351                          * Reaching here after boot in a few seconds may not
1352                          * mean that system will remain stable at "unknown"
1353                          * frequency for longer duration. Hence, a BUG_ON().
1354                          */
1355                         BUG_ON(ret);
1356                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1357                                 __func__, policy->cpu, policy->cur);
1358                 }
1359         }
1360
1361         if (new_policy) {
1362                 ret = cpufreq_add_dev_interface(policy);
1363                 if (ret)
1364                         goto out_destroy_policy;
1365
1366                 cpufreq_stats_create_table(policy);
1367
1368                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1369                 list_add(&policy->policy_list, &cpufreq_policy_list);
1370                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1371         }
1372
1373         ret = cpufreq_init_policy(policy);
1374         if (ret) {
1375                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1376                        __func__, cpu, ret);
1377                 goto out_destroy_policy;
1378         }
1379
1380         up_write(&policy->rwsem);
1381
1382         kobject_uevent(&policy->kobj, KOBJ_ADD);
1383
1384         /* Callback for handling stuff after policy is ready */
1385         if (cpufreq_driver->ready)
1386                 cpufreq_driver->ready(policy);
1387
1388         if (cpufreq_thermal_control_enabled(cpufreq_driver))
1389                 policy->cdev = of_cpufreq_cooling_register(policy);
1390
1391         pr_debug("initialization complete\n");
1392
1393         return 0;
1394
1395 out_destroy_policy:
1396         for_each_cpu(j, policy->real_cpus)
1397                 remove_cpu_dev_symlink(policy, get_cpu_device(j));
1398
1399         up_write(&policy->rwsem);
1400
1401 out_exit_policy:
1402         if (cpufreq_driver->exit)
1403                 cpufreq_driver->exit(policy);
1404
1405 out_free_policy:
1406         cpufreq_policy_free(policy);
1407         return ret;
1408 }
1409
1410 /**
1411  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1412  * @dev: CPU device.
1413  * @sif: Subsystem interface structure pointer (not used)
1414  */
1415 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1416 {
1417         struct cpufreq_policy *policy;
1418         unsigned cpu = dev->id;
1419         int ret;
1420
1421         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1422
1423         if (cpu_online(cpu)) {
1424                 ret = cpufreq_online(cpu);
1425                 if (ret)
1426                         return ret;
1427         }
1428
1429         /* Create sysfs link on CPU registration */
1430         policy = per_cpu(cpufreq_cpu_data, cpu);
1431         if (policy)
1432                 add_cpu_dev_symlink(policy, cpu);
1433
1434         return 0;
1435 }
1436
1437 static int cpufreq_offline(unsigned int cpu)
1438 {
1439         struct cpufreq_policy *policy;
1440         int ret;
1441
1442         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1443
1444         policy = cpufreq_cpu_get_raw(cpu);
1445         if (!policy) {
1446                 pr_debug("%s: No cpu_data found\n", __func__);
1447                 return 0;
1448         }
1449
1450         down_write(&policy->rwsem);
1451         if (has_target())
1452                 cpufreq_stop_governor(policy);
1453
1454         cpumask_clear_cpu(cpu, policy->cpus);
1455
1456         if (policy_is_inactive(policy)) {
1457                 if (has_target())
1458                         strncpy(policy->last_governor, policy->governor->name,
1459                                 CPUFREQ_NAME_LEN);
1460                 else
1461                         policy->last_policy = policy->policy;
1462         } else if (cpu == policy->cpu) {
1463                 /* Nominate new CPU */
1464                 policy->cpu = cpumask_any(policy->cpus);
1465         }
1466
1467         /* Start governor again for active policy */
1468         if (!policy_is_inactive(policy)) {
1469                 if (has_target()) {
1470                         ret = cpufreq_start_governor(policy);
1471                         if (ret)
1472                                 pr_err("%s: Failed to start governor\n", __func__);
1473                 }
1474
1475                 goto unlock;
1476         }
1477
1478         if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1479                 cpufreq_cooling_unregister(policy->cdev);
1480                 policy->cdev = NULL;
1481         }
1482
1483         if (cpufreq_driver->stop_cpu)
1484                 cpufreq_driver->stop_cpu(policy);
1485
1486         if (has_target())
1487                 cpufreq_exit_governor(policy);
1488
1489         /*
1490          * Perform the ->offline() during light-weight tear-down, as
1491          * that allows fast recovery when the CPU comes back.
1492          */
1493         if (cpufreq_driver->offline) {
1494                 cpufreq_driver->offline(policy);
1495         } else if (cpufreq_driver->exit) {
1496                 cpufreq_driver->exit(policy);
1497                 policy->freq_table = NULL;
1498         }
1499
1500 unlock:
1501         up_write(&policy->rwsem);
1502         return 0;
1503 }
1504
1505 /**
1506  * cpufreq_remove_dev - remove a CPU device
1507  *
1508  * Removes the cpufreq interface for a CPU device.
1509  */
1510 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1511 {
1512         unsigned int cpu = dev->id;
1513         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1514
1515         if (!policy)
1516                 return;
1517
1518         if (cpu_online(cpu))
1519                 cpufreq_offline(cpu);
1520
1521         cpumask_clear_cpu(cpu, policy->real_cpus);
1522         remove_cpu_dev_symlink(policy, dev);
1523
1524         if (cpumask_empty(policy->real_cpus)) {
1525                 /* We did light-weight exit earlier, do full tear down now */
1526                 if (cpufreq_driver->offline)
1527                         cpufreq_driver->exit(policy);
1528
1529                 cpufreq_policy_free(policy);
1530         }
1531 }
1532
1533 /**
1534  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1535  *      in deep trouble.
1536  *      @policy: policy managing CPUs
1537  *      @new_freq: CPU frequency the CPU actually runs at
1538  *
1539  *      We adjust to current frequency first, and need to clean up later.
1540  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1541  */
1542 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1543                                 unsigned int new_freq)
1544 {
1545         struct cpufreq_freqs freqs;
1546
1547         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1548                  policy->cur, new_freq);
1549
1550         freqs.old = policy->cur;
1551         freqs.new = new_freq;
1552
1553         cpufreq_freq_transition_begin(policy, &freqs);
1554         cpufreq_freq_transition_end(policy, &freqs, 0);
1555 }
1556
1557 static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1558 {
1559         unsigned int new_freq;
1560
1561         new_freq = cpufreq_driver->get(policy->cpu);
1562         if (!new_freq)
1563                 return 0;
1564
1565         /*
1566          * If fast frequency switching is used with the given policy, the check
1567          * against policy->cur is pointless, so skip it in that case.
1568          */
1569         if (policy->fast_switch_enabled || !has_target())
1570                 return new_freq;
1571
1572         if (policy->cur != new_freq) {
1573                 cpufreq_out_of_sync(policy, new_freq);
1574                 if (update)
1575                         schedule_work(&policy->update);
1576         }
1577
1578         return new_freq;
1579 }
1580
1581 /**
1582  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1583  * @cpu: CPU number
1584  *
1585  * This is the last known freq, without actually getting it from the driver.
1586  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1587  */
1588 unsigned int cpufreq_quick_get(unsigned int cpu)
1589 {
1590         struct cpufreq_policy *policy;
1591         unsigned int ret_freq = 0;
1592         unsigned long flags;
1593
1594         read_lock_irqsave(&cpufreq_driver_lock, flags);
1595
1596         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1597                 ret_freq = cpufreq_driver->get(cpu);
1598                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1599                 return ret_freq;
1600         }
1601
1602         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1603
1604         policy = cpufreq_cpu_get(cpu);
1605         if (policy) {
1606                 ret_freq = policy->cur;
1607                 cpufreq_cpu_put(policy);
1608         }
1609
1610         return ret_freq;
1611 }
1612 EXPORT_SYMBOL(cpufreq_quick_get);
1613
1614 /**
1615  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1616  * @cpu: CPU number
1617  *
1618  * Just return the max possible frequency for a given CPU.
1619  */
1620 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1621 {
1622         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1623         unsigned int ret_freq = 0;
1624
1625         if (policy) {
1626                 ret_freq = policy->max;
1627                 cpufreq_cpu_put(policy);
1628         }
1629
1630         return ret_freq;
1631 }
1632 EXPORT_SYMBOL(cpufreq_quick_get_max);
1633
1634 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1635 {
1636         if (unlikely(policy_is_inactive(policy)))
1637                 return 0;
1638
1639         return cpufreq_verify_current_freq(policy, true);
1640 }
1641
1642 /**
1643  * cpufreq_get - get the current CPU frequency (in kHz)
1644  * @cpu: CPU number
1645  *
1646  * Get the CPU current (static) CPU frequency
1647  */
1648 unsigned int cpufreq_get(unsigned int cpu)
1649 {
1650         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1651         unsigned int ret_freq = 0;
1652
1653         if (policy) {
1654                 down_read(&policy->rwsem);
1655                 if (cpufreq_driver->get)
1656                         ret_freq = __cpufreq_get(policy);
1657                 up_read(&policy->rwsem);
1658
1659                 cpufreq_cpu_put(policy);
1660         }
1661
1662         return ret_freq;
1663 }
1664 EXPORT_SYMBOL(cpufreq_get);
1665
1666 static struct subsys_interface cpufreq_interface = {
1667         .name           = "cpufreq",
1668         .subsys         = &cpu_subsys,
1669         .add_dev        = cpufreq_add_dev,
1670         .remove_dev     = cpufreq_remove_dev,
1671 };
1672
1673 /*
1674  * In case platform wants some specific frequency to be configured
1675  * during suspend..
1676  */
1677 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1678 {
1679         int ret;
1680
1681         if (!policy->suspend_freq) {
1682                 pr_debug("%s: suspend_freq not defined\n", __func__);
1683                 return 0;
1684         }
1685
1686         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1687                         policy->suspend_freq);
1688
1689         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1690                         CPUFREQ_RELATION_H);
1691         if (ret)
1692                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1693                                 __func__, policy->suspend_freq, ret);
1694
1695         return ret;
1696 }
1697 EXPORT_SYMBOL(cpufreq_generic_suspend);
1698
1699 /**
1700  * cpufreq_suspend() - Suspend CPUFreq governors
1701  *
1702  * Called during system wide Suspend/Hibernate cycles for suspending governors
1703  * as some platforms can't change frequency after this point in suspend cycle.
1704  * Because some of the devices (like: i2c, regulators, etc) they use for
1705  * changing frequency are suspended quickly after this point.
1706  */
1707 void cpufreq_suspend(void)
1708 {
1709         struct cpufreq_policy *policy;
1710
1711         if (!cpufreq_driver)
1712                 return;
1713
1714         if (!has_target() && !cpufreq_driver->suspend)
1715                 goto suspend;
1716
1717         pr_debug("%s: Suspending Governors\n", __func__);
1718
1719         for_each_active_policy(policy) {
1720                 if (has_target()) {
1721                         down_write(&policy->rwsem);
1722                         cpufreq_stop_governor(policy);
1723                         up_write(&policy->rwsem);
1724                 }
1725
1726                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1727                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1728                                 policy);
1729         }
1730
1731 suspend:
1732         cpufreq_suspended = true;
1733 }
1734
1735 /**
1736  * cpufreq_resume() - Resume CPUFreq governors
1737  *
1738  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1739  * are suspended with cpufreq_suspend().
1740  */
1741 void cpufreq_resume(void)
1742 {
1743         struct cpufreq_policy *policy;
1744         int ret;
1745
1746         if (!cpufreq_driver)
1747                 return;
1748
1749         if (unlikely(!cpufreq_suspended))
1750                 return;
1751
1752         cpufreq_suspended = false;
1753
1754         if (!has_target() && !cpufreq_driver->resume)
1755                 return;
1756
1757         pr_debug("%s: Resuming Governors\n", __func__);
1758
1759         for_each_active_policy(policy) {
1760                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1761                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1762                                 policy);
1763                 } else if (has_target()) {
1764                         down_write(&policy->rwsem);
1765                         ret = cpufreq_start_governor(policy);
1766                         up_write(&policy->rwsem);
1767
1768                         if (ret)
1769                                 pr_err("%s: Failed to start governor for policy: %p\n",
1770                                        __func__, policy);
1771                 }
1772         }
1773 }
1774
1775 /**
1776  *      cpufreq_get_current_driver - return current driver's name
1777  *
1778  *      Return the name string of the currently loaded cpufreq driver
1779  *      or NULL, if none.
1780  */
1781 const char *cpufreq_get_current_driver(void)
1782 {
1783         if (cpufreq_driver)
1784                 return cpufreq_driver->name;
1785
1786         return NULL;
1787 }
1788 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1789
1790 /**
1791  *      cpufreq_get_driver_data - return current driver data
1792  *
1793  *      Return the private data of the currently loaded cpufreq
1794  *      driver, or NULL if no cpufreq driver is loaded.
1795  */
1796 void *cpufreq_get_driver_data(void)
1797 {
1798         if (cpufreq_driver)
1799                 return cpufreq_driver->driver_data;
1800
1801         return NULL;
1802 }
1803 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1804
1805 /*********************************************************************
1806  *                     NOTIFIER LISTS INTERFACE                      *
1807  *********************************************************************/
1808
1809 /**
1810  *      cpufreq_register_notifier - register a driver with cpufreq
1811  *      @nb: notifier function to register
1812  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1813  *
1814  *      Add a driver to one of two lists: either a list of drivers that
1815  *      are notified about clock rate changes (once before and once after
1816  *      the transition), or a list of drivers that are notified about
1817  *      changes in cpufreq policy.
1818  *
1819  *      This function may sleep, and has the same return conditions as
1820  *      blocking_notifier_chain_register.
1821  */
1822 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1823 {
1824         int ret;
1825
1826         if (cpufreq_disabled())
1827                 return -EINVAL;
1828
1829         switch (list) {
1830         case CPUFREQ_TRANSITION_NOTIFIER:
1831                 mutex_lock(&cpufreq_fast_switch_lock);
1832
1833                 if (cpufreq_fast_switch_count > 0) {
1834                         mutex_unlock(&cpufreq_fast_switch_lock);
1835                         return -EBUSY;
1836                 }
1837                 ret = srcu_notifier_chain_register(
1838                                 &cpufreq_transition_notifier_list, nb);
1839                 if (!ret)
1840                         cpufreq_fast_switch_count--;
1841
1842                 mutex_unlock(&cpufreq_fast_switch_lock);
1843                 break;
1844         case CPUFREQ_POLICY_NOTIFIER:
1845                 ret = blocking_notifier_chain_register(
1846                                 &cpufreq_policy_notifier_list, nb);
1847                 break;
1848         default:
1849                 ret = -EINVAL;
1850         }
1851
1852         return ret;
1853 }
1854 EXPORT_SYMBOL(cpufreq_register_notifier);
1855
1856 /**
1857  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1858  *      @nb: notifier block to be unregistered
1859  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1860  *
1861  *      Remove a driver from the CPU frequency notifier list.
1862  *
1863  *      This function may sleep, and has the same return conditions as
1864  *      blocking_notifier_chain_unregister.
1865  */
1866 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1867 {
1868         int ret;
1869
1870         if (cpufreq_disabled())
1871                 return -EINVAL;
1872
1873         switch (list) {
1874         case CPUFREQ_TRANSITION_NOTIFIER:
1875                 mutex_lock(&cpufreq_fast_switch_lock);
1876
1877                 ret = srcu_notifier_chain_unregister(
1878                                 &cpufreq_transition_notifier_list, nb);
1879                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1880                         cpufreq_fast_switch_count++;
1881
1882                 mutex_unlock(&cpufreq_fast_switch_lock);
1883                 break;
1884         case CPUFREQ_POLICY_NOTIFIER:
1885                 ret = blocking_notifier_chain_unregister(
1886                                 &cpufreq_policy_notifier_list, nb);
1887                 break;
1888         default:
1889                 ret = -EINVAL;
1890         }
1891
1892         return ret;
1893 }
1894 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1895
1896
1897 /*********************************************************************
1898  *                              GOVERNORS                            *
1899  *********************************************************************/
1900
1901 /**
1902  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1903  * @policy: cpufreq policy to switch the frequency for.
1904  * @target_freq: New frequency to set (may be approximate).
1905  *
1906  * Carry out a fast frequency switch without sleeping.
1907  *
1908  * The driver's ->fast_switch() callback invoked by this function must be
1909  * suitable for being called from within RCU-sched read-side critical sections
1910  * and it is expected to select the minimum available frequency greater than or
1911  * equal to @target_freq (CPUFREQ_RELATION_L).
1912  *
1913  * This function must not be called if policy->fast_switch_enabled is unset.
1914  *
1915  * Governors calling this function must guarantee that it will never be invoked
1916  * twice in parallel for the same policy and that it will never be called in
1917  * parallel with either ->target() or ->target_index() for the same policy.
1918  *
1919  * Returns the actual frequency set for the CPU.
1920  *
1921  * If 0 is returned by the driver's ->fast_switch() callback to indicate an
1922  * error condition, the hardware configuration must be preserved.
1923  */
1924 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1925                                         unsigned int target_freq)
1926 {
1927         target_freq = clamp_val(target_freq, policy->min, policy->max);
1928
1929         return cpufreq_driver->fast_switch(policy, target_freq);
1930 }
1931 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1932
1933 /* Must set freqs->new to intermediate frequency */
1934 static int __target_intermediate(struct cpufreq_policy *policy,
1935                                  struct cpufreq_freqs *freqs, int index)
1936 {
1937         int ret;
1938
1939         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1940
1941         /* We don't need to switch to intermediate freq */
1942         if (!freqs->new)
1943                 return 0;
1944
1945         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1946                  __func__, policy->cpu, freqs->old, freqs->new);
1947
1948         cpufreq_freq_transition_begin(policy, freqs);
1949         ret = cpufreq_driver->target_intermediate(policy, index);
1950         cpufreq_freq_transition_end(policy, freqs, ret);
1951
1952         if (ret)
1953                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1954                        __func__, ret);
1955
1956         return ret;
1957 }
1958
1959 static int __target_index(struct cpufreq_policy *policy, int index)
1960 {
1961         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1962         unsigned int intermediate_freq = 0;
1963         unsigned int newfreq = policy->freq_table[index].frequency;
1964         int retval = -EINVAL;
1965         bool notify;
1966
1967         if (newfreq == policy->cur)
1968                 return 0;
1969
1970         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1971         if (notify) {
1972                 /* Handle switching to intermediate frequency */
1973                 if (cpufreq_driver->get_intermediate) {
1974                         retval = __target_intermediate(policy, &freqs, index);
1975                         if (retval)
1976                                 return retval;
1977
1978                         intermediate_freq = freqs.new;
1979                         /* Set old freq to intermediate */
1980                         if (intermediate_freq)
1981                                 freqs.old = freqs.new;
1982                 }
1983
1984                 freqs.new = newfreq;
1985                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1986                          __func__, policy->cpu, freqs.old, freqs.new);
1987
1988                 cpufreq_freq_transition_begin(policy, &freqs);
1989         }
1990
1991         retval = cpufreq_driver->target_index(policy, index);
1992         if (retval)
1993                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1994                        retval);
1995
1996         if (notify) {
1997                 cpufreq_freq_transition_end(policy, &freqs, retval);
1998
1999                 /*
2000                  * Failed after setting to intermediate freq? Driver should have
2001                  * reverted back to initial frequency and so should we. Check
2002                  * here for intermediate_freq instead of get_intermediate, in
2003                  * case we haven't switched to intermediate freq at all.
2004                  */
2005                 if (unlikely(retval && intermediate_freq)) {
2006                         freqs.old = intermediate_freq;
2007                         freqs.new = policy->restore_freq;
2008                         cpufreq_freq_transition_begin(policy, &freqs);
2009                         cpufreq_freq_transition_end(policy, &freqs, 0);
2010                 }
2011         }
2012
2013         return retval;
2014 }
2015
2016 int __cpufreq_driver_target(struct cpufreq_policy *policy,
2017                             unsigned int target_freq,
2018                             unsigned int relation)
2019 {
2020         unsigned int old_target_freq = target_freq;
2021         int index;
2022
2023         if (cpufreq_disabled())
2024                 return -ENODEV;
2025
2026         /* Make sure that target_freq is within supported range */
2027         target_freq = clamp_val(target_freq, policy->min, policy->max);
2028
2029         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2030                  policy->cpu, target_freq, relation, old_target_freq);
2031
2032         /*
2033          * This might look like a redundant call as we are checking it again
2034          * after finding index. But it is left intentionally for cases where
2035          * exactly same freq is called again and so we can save on few function
2036          * calls.
2037          */
2038         if (target_freq == policy->cur)
2039                 return 0;
2040
2041         /* Save last value to restore later on errors */
2042         policy->restore_freq = policy->cur;
2043
2044         if (cpufreq_driver->target)
2045                 return cpufreq_driver->target(policy, target_freq, relation);
2046
2047         if (!cpufreq_driver->target_index)
2048                 return -EINVAL;
2049
2050         index = cpufreq_frequency_table_target(policy, target_freq, relation);
2051
2052         return __target_index(policy, index);
2053 }
2054 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2055
2056 int cpufreq_driver_target(struct cpufreq_policy *policy,
2057                           unsigned int target_freq,
2058                           unsigned int relation)
2059 {
2060         int ret = -EINVAL;
2061
2062         down_write(&policy->rwsem);
2063
2064         ret = __cpufreq_driver_target(policy, target_freq, relation);
2065
2066         up_write(&policy->rwsem);
2067
2068         return ret;
2069 }
2070 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2071
2072 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2073 {
2074         return NULL;
2075 }
2076
2077 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2078 {
2079         int ret;
2080
2081         /* Don't start any governor operations if we are entering suspend */
2082         if (cpufreq_suspended)
2083                 return 0;
2084         /*
2085          * Governor might not be initiated here if ACPI _PPC changed
2086          * notification happened, so check it.
2087          */
2088         if (!policy->governor)
2089                 return -EINVAL;
2090
2091         /* Platform doesn't want dynamic frequency switching ? */
2092         if (policy->governor->dynamic_switching &&
2093             cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2094                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2095
2096                 if (gov) {
2097                         pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2098                                 policy->governor->name, gov->name);
2099                         policy->governor = gov;
2100                 } else {
2101                         return -EINVAL;
2102                 }
2103         }
2104
2105         if (!try_module_get(policy->governor->owner))
2106                 return -EINVAL;
2107
2108         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2109
2110         if (policy->governor->init) {
2111                 ret = policy->governor->init(policy);
2112                 if (ret) {
2113                         module_put(policy->governor->owner);
2114                         return ret;
2115                 }
2116         }
2117
2118         return 0;
2119 }
2120
2121 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2122 {
2123         if (cpufreq_suspended || !policy->governor)
2124                 return;
2125
2126         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2127
2128         if (policy->governor->exit)
2129                 policy->governor->exit(policy);
2130
2131         module_put(policy->governor->owner);
2132 }
2133
2134 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2135 {
2136         int ret;
2137
2138         if (cpufreq_suspended)
2139                 return 0;
2140
2141         if (!policy->governor)
2142                 return -EINVAL;
2143
2144         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2145
2146         if (cpufreq_driver->get)
2147                 cpufreq_verify_current_freq(policy, false);
2148
2149         if (policy->governor->start) {
2150                 ret = policy->governor->start(policy);
2151                 if (ret)
2152                         return ret;
2153         }
2154
2155         if (policy->governor->limits)
2156                 policy->governor->limits(policy);
2157
2158         return 0;
2159 }
2160
2161 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2162 {
2163         if (cpufreq_suspended || !policy->governor)
2164                 return;
2165
2166         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2167
2168         if (policy->governor->stop)
2169                 policy->governor->stop(policy);
2170 }
2171
2172 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2173 {
2174         if (cpufreq_suspended || !policy->governor)
2175                 return;
2176
2177         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2178
2179         if (policy->governor->limits)
2180                 policy->governor->limits(policy);
2181 }
2182
2183 int cpufreq_register_governor(struct cpufreq_governor *governor)
2184 {
2185         int err;
2186
2187         if (!governor)
2188                 return -EINVAL;
2189
2190         if (cpufreq_disabled())
2191                 return -ENODEV;
2192
2193         mutex_lock(&cpufreq_governor_mutex);
2194
2195         err = -EBUSY;
2196         if (!find_governor(governor->name)) {
2197                 err = 0;
2198                 list_add(&governor->governor_list, &cpufreq_governor_list);
2199         }
2200
2201         mutex_unlock(&cpufreq_governor_mutex);
2202         return err;
2203 }
2204 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2205
2206 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2207 {
2208         struct cpufreq_policy *policy;
2209         unsigned long flags;
2210
2211         if (!governor)
2212                 return;
2213
2214         if (cpufreq_disabled())
2215                 return;
2216
2217         /* clear last_governor for all inactive policies */
2218         read_lock_irqsave(&cpufreq_driver_lock, flags);
2219         for_each_inactive_policy(policy) {
2220                 if (!strcmp(policy->last_governor, governor->name)) {
2221                         policy->governor = NULL;
2222                         strcpy(policy->last_governor, "\0");
2223                 }
2224         }
2225         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2226
2227         mutex_lock(&cpufreq_governor_mutex);
2228         list_del(&governor->governor_list);
2229         mutex_unlock(&cpufreq_governor_mutex);
2230 }
2231 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2232
2233
2234 /*********************************************************************
2235  *                          POLICY INTERFACE                         *
2236  *********************************************************************/
2237
2238 /**
2239  * cpufreq_get_policy - get the current cpufreq_policy
2240  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2241  *      is written
2242  *
2243  * Reads the current cpufreq policy.
2244  */
2245 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2246 {
2247         struct cpufreq_policy *cpu_policy;
2248         if (!policy)
2249                 return -EINVAL;
2250
2251         cpu_policy = cpufreq_cpu_get(cpu);
2252         if (!cpu_policy)
2253                 return -EINVAL;
2254
2255         memcpy(policy, cpu_policy, sizeof(*policy));
2256
2257         cpufreq_cpu_put(cpu_policy);
2258         return 0;
2259 }
2260 EXPORT_SYMBOL(cpufreq_get_policy);
2261
2262 /**
2263  * cpufreq_set_policy - Modify cpufreq policy parameters.
2264  * @policy: Policy object to modify.
2265  * @new_policy: New policy data.
2266  *
2267  * Pass @new_policy to the cpufreq driver's ->verify() callback, run the
2268  * installed policy notifiers for it with the CPUFREQ_ADJUST value, pass it to
2269  * the driver's ->verify() callback again and run the notifiers for it again
2270  * with the CPUFREQ_NOTIFY value.  Next, copy the min and max parameters
2271  * of @new_policy to @policy and either invoke the driver's ->setpolicy()
2272  * callback (if present) or carry out a governor update for @policy.  That is,
2273  * run the current governor's ->limits() callback (if the governor field in
2274  * @new_policy points to the same object as the one in @policy) or replace the
2275  * governor for @policy with the new one stored in @new_policy.
2276  *
2277  * The cpuinfo part of @policy is not updated by this function.
2278  */
2279 int cpufreq_set_policy(struct cpufreq_policy *policy,
2280                        struct cpufreq_policy *new_policy)
2281 {
2282         struct cpufreq_governor *old_gov;
2283         int ret;
2284
2285         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2286                  new_policy->cpu, new_policy->min, new_policy->max);
2287
2288         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2289
2290         /*
2291         * This check works well when we store new min/max freq attributes,
2292         * because new_policy is a copy of policy with one field updated.
2293         */
2294         if (new_policy->min > new_policy->max)
2295                 return -EINVAL;
2296
2297         /* verify the cpu speed can be set within this limit */
2298         ret = cpufreq_driver->verify(new_policy);
2299         if (ret)
2300                 return ret;
2301
2302         /* adjust if necessary - all reasons */
2303         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2304                         CPUFREQ_ADJUST, new_policy);
2305
2306         /*
2307          * verify the cpu speed can be set within this limit, which might be
2308          * different to the first one
2309          */
2310         ret = cpufreq_driver->verify(new_policy);
2311         if (ret)
2312                 return ret;
2313
2314         /* notification of the new policy */
2315         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2316                         CPUFREQ_NOTIFY, new_policy);
2317
2318         policy->min = new_policy->min;
2319         policy->max = new_policy->max;
2320         trace_cpu_frequency_limits(policy);
2321
2322         policy->cached_target_freq = UINT_MAX;
2323
2324         pr_debug("new min and max freqs are %u - %u kHz\n",
2325                  policy->min, policy->max);
2326
2327         if (cpufreq_driver->setpolicy) {
2328                 policy->policy = new_policy->policy;
2329                 pr_debug("setting range\n");
2330                 return cpufreq_driver->setpolicy(policy);
2331         }
2332
2333         if (new_policy->governor == policy->governor) {
2334                 pr_debug("governor limits update\n");
2335                 cpufreq_governor_limits(policy);
2336                 return 0;
2337         }
2338
2339         pr_debug("governor switch\n");
2340
2341         /* save old, working values */
2342         old_gov = policy->governor;
2343         /* end old governor */
2344         if (old_gov) {
2345                 cpufreq_stop_governor(policy);
2346                 cpufreq_exit_governor(policy);
2347         }
2348
2349         /* start new governor */
2350         policy->governor = new_policy->governor;
2351         ret = cpufreq_init_governor(policy);
2352         if (!ret) {
2353                 ret = cpufreq_start_governor(policy);
2354                 if (!ret) {
2355                         pr_debug("governor change\n");
2356                         sched_cpufreq_governor_change(policy, old_gov);
2357                         return 0;
2358                 }
2359                 cpufreq_exit_governor(policy);
2360         }
2361
2362         /* new governor failed, so re-start old one */
2363         pr_debug("starting governor %s failed\n", policy->governor->name);
2364         if (old_gov) {
2365                 policy->governor = old_gov;
2366                 if (cpufreq_init_governor(policy))
2367                         policy->governor = NULL;
2368                 else
2369                         cpufreq_start_governor(policy);
2370         }
2371
2372         return ret;
2373 }
2374
2375 /**
2376  * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2377  * @cpu: CPU to re-evaluate the policy for.
2378  *
2379  * Update the current frequency for the cpufreq policy of @cpu and use
2380  * cpufreq_set_policy() to re-apply the min and max limits saved in the
2381  * user_policy sub-structure of that policy, which triggers the evaluation
2382  * of policy notifiers and the cpufreq driver's ->verify() callback for the
2383  * policy in question, among other things.
2384  */
2385 void cpufreq_update_policy(unsigned int cpu)
2386 {
2387         struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2388
2389         if (!policy)
2390                 return;
2391
2392         /*
2393          * BIOS might change freq behind our back
2394          * -> ask driver for current freq and notify governors about a change
2395          */
2396         if (cpufreq_driver->get && has_target() &&
2397             (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2398                 goto unlock;
2399
2400         refresh_frequency_limits(policy);
2401
2402 unlock:
2403         cpufreq_cpu_release(policy);
2404 }
2405 EXPORT_SYMBOL(cpufreq_update_policy);
2406
2407 /**
2408  * cpufreq_update_limits - Update policy limits for a given CPU.
2409  * @cpu: CPU to update the policy limits for.
2410  *
2411  * Invoke the driver's ->update_limits callback if present or call
2412  * cpufreq_update_policy() for @cpu.
2413  */
2414 void cpufreq_update_limits(unsigned int cpu)
2415 {
2416         if (cpufreq_driver->update_limits)
2417                 cpufreq_driver->update_limits(cpu);
2418         else
2419                 cpufreq_update_policy(cpu);
2420 }
2421 EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2422
2423 /*********************************************************************
2424  *               BOOST                                               *
2425  *********************************************************************/
2426 static int cpufreq_boost_set_sw(int state)
2427 {
2428         struct cpufreq_policy *policy;
2429         int ret = -EINVAL;
2430
2431         for_each_active_policy(policy) {
2432                 if (!policy->freq_table)
2433                         continue;
2434
2435                 ret = cpufreq_frequency_table_cpuinfo(policy,
2436                                                       policy->freq_table);
2437                 if (ret) {
2438                         pr_err("%s: Policy frequency update failed\n",
2439                                __func__);
2440                         break;
2441                 }
2442
2443                 down_write(&policy->rwsem);
2444                 policy->user_policy.max = policy->max;
2445                 cpufreq_governor_limits(policy);
2446                 up_write(&policy->rwsem);
2447         }
2448
2449         return ret;
2450 }
2451
2452 int cpufreq_boost_trigger_state(int state)
2453 {
2454         unsigned long flags;
2455         int ret = 0;
2456
2457         if (cpufreq_driver->boost_enabled == state)
2458                 return 0;
2459
2460         write_lock_irqsave(&cpufreq_driver_lock, flags);
2461         cpufreq_driver->boost_enabled = state;
2462         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2463
2464         ret = cpufreq_driver->set_boost(state);
2465         if (ret) {
2466                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2467                 cpufreq_driver->boost_enabled = !state;
2468                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2469
2470                 pr_err("%s: Cannot %s BOOST\n",
2471                        __func__, state ? "enable" : "disable");
2472         }
2473
2474         return ret;
2475 }
2476
2477 static bool cpufreq_boost_supported(void)
2478 {
2479         return cpufreq_driver->set_boost;
2480 }
2481
2482 static int create_boost_sysfs_file(void)
2483 {
2484         int ret;
2485
2486         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2487         if (ret)
2488                 pr_err("%s: cannot register global BOOST sysfs file\n",
2489                        __func__);
2490
2491         return ret;
2492 }
2493
2494 static void remove_boost_sysfs_file(void)
2495 {
2496         if (cpufreq_boost_supported())
2497                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2498 }
2499
2500 int cpufreq_enable_boost_support(void)
2501 {
2502         if (!cpufreq_driver)
2503                 return -EINVAL;
2504
2505         if (cpufreq_boost_supported())
2506                 return 0;
2507
2508         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2509
2510         /* This will get removed on driver unregister */
2511         return create_boost_sysfs_file();
2512 }
2513 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2514
2515 int cpufreq_boost_enabled(void)
2516 {
2517         return cpufreq_driver->boost_enabled;
2518 }
2519 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2520
2521 /*********************************************************************
2522  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2523  *********************************************************************/
2524 static enum cpuhp_state hp_online;
2525
2526 static int cpuhp_cpufreq_online(unsigned int cpu)
2527 {
2528         cpufreq_online(cpu);
2529
2530         return 0;
2531 }
2532
2533 static int cpuhp_cpufreq_offline(unsigned int cpu)
2534 {
2535         cpufreq_offline(cpu);
2536
2537         return 0;
2538 }
2539
2540 /**
2541  * cpufreq_register_driver - register a CPU Frequency driver
2542  * @driver_data: A struct cpufreq_driver containing the values#
2543  * submitted by the CPU Frequency driver.
2544  *
2545  * Registers a CPU Frequency driver to this core code. This code
2546  * returns zero on success, -EEXIST when another driver got here first
2547  * (and isn't unregistered in the meantime).
2548  *
2549  */
2550 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2551 {
2552         unsigned long flags;
2553         int ret;
2554
2555         if (cpufreq_disabled())
2556                 return -ENODEV;
2557
2558         if (!driver_data || !driver_data->verify || !driver_data->init ||
2559             !(driver_data->setpolicy || driver_data->target_index ||
2560                     driver_data->target) ||
2561              (driver_data->setpolicy && (driver_data->target_index ||
2562                     driver_data->target)) ||
2563              (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2564              (!driver_data->online != !driver_data->offline))
2565                 return -EINVAL;
2566
2567         pr_debug("trying to register driver %s\n", driver_data->name);
2568
2569         /* Protect against concurrent CPU online/offline. */
2570         cpus_read_lock();
2571
2572         write_lock_irqsave(&cpufreq_driver_lock, flags);
2573         if (cpufreq_driver) {
2574                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2575                 ret = -EEXIST;
2576                 goto out;
2577         }
2578         cpufreq_driver = driver_data;
2579         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2580
2581         if (driver_data->setpolicy)
2582                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2583
2584         if (cpufreq_boost_supported()) {
2585                 ret = create_boost_sysfs_file();
2586                 if (ret)
2587                         goto err_null_driver;
2588         }
2589
2590         ret = subsys_interface_register(&cpufreq_interface);
2591         if (ret)
2592                 goto err_boost_unreg;
2593
2594         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2595             list_empty(&cpufreq_policy_list)) {
2596                 /* if all ->init() calls failed, unregister */
2597                 ret = -ENODEV;
2598                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2599                          driver_data->name);
2600                 goto err_if_unreg;
2601         }
2602
2603         ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2604                                                    "cpufreq:online",
2605                                                    cpuhp_cpufreq_online,
2606                                                    cpuhp_cpufreq_offline);
2607         if (ret < 0)
2608                 goto err_if_unreg;
2609         hp_online = ret;
2610         ret = 0;
2611
2612         pr_debug("driver %s up and running\n", driver_data->name);
2613         goto out;
2614
2615 err_if_unreg:
2616         subsys_interface_unregister(&cpufreq_interface);
2617 err_boost_unreg:
2618         remove_boost_sysfs_file();
2619 err_null_driver:
2620         write_lock_irqsave(&cpufreq_driver_lock, flags);
2621         cpufreq_driver = NULL;
2622         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2623 out:
2624         cpus_read_unlock();
2625         return ret;
2626 }
2627 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2628
2629 /**
2630  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2631  *
2632  * Unregister the current CPUFreq driver. Only call this if you have
2633  * the right to do so, i.e. if you have succeeded in initialising before!
2634  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2635  * currently not initialised.
2636  */
2637 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2638 {
2639         unsigned long flags;
2640
2641         if (!cpufreq_driver || (driver != cpufreq_driver))
2642                 return -EINVAL;
2643
2644         pr_debug("unregistering driver %s\n", driver->name);
2645
2646         /* Protect against concurrent cpu hotplug */
2647         cpus_read_lock();
2648         subsys_interface_unregister(&cpufreq_interface);
2649         remove_boost_sysfs_file();
2650         cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2651
2652         write_lock_irqsave(&cpufreq_driver_lock, flags);
2653
2654         cpufreq_driver = NULL;
2655
2656         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2657         cpus_read_unlock();
2658
2659         return 0;
2660 }
2661 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2662
2663 /*
2664  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2665  * or mutexes when secondary CPUs are halted.
2666  */
2667 static struct syscore_ops cpufreq_syscore_ops = {
2668         .shutdown = cpufreq_suspend,
2669 };
2670
2671 struct kobject *cpufreq_global_kobject;
2672 EXPORT_SYMBOL(cpufreq_global_kobject);
2673
2674 static int __init cpufreq_core_init(void)
2675 {
2676         if (cpufreq_disabled())
2677                 return -ENODEV;
2678
2679         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2680         BUG_ON(!cpufreq_global_kobject);
2681
2682         register_syscore_ops(&cpufreq_syscore_ops);
2683
2684         return 0;
2685 }
2686 module_param(off, int, 0444);
2687 core_initcall(cpufreq_core_init);