Loading...
1/*
2 * linux/drivers/cpufreq/cpufreq.c
3 *
4 * Copyright (C) 2001 Russell King
5 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6 *
7 * Oct 2005 - Ashok Raj <ashok.raj@intel.com>
8 * Added handling for CPU hotplug
9 * Feb 2006 - Jacob Shin <jacob.shin@amd.com>
10 * Fix handling for CPU hotplug -- affected CPUs
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 *
16 */
17
18#include <linux/kernel.h>
19#include <linux/module.h>
20#include <linux/init.h>
21#include <linux/notifier.h>
22#include <linux/cpufreq.h>
23#include <linux/delay.h>
24#include <linux/interrupt.h>
25#include <linux/spinlock.h>
26#include <linux/device.h>
27#include <linux/slab.h>
28#include <linux/cpu.h>
29#include <linux/completion.h>
30#include <linux/mutex.h>
31#include <linux/syscore_ops.h>
32
33#include <trace/events/power.h>
34
35/**
36 * The "cpufreq driver" - the arch- or hardware-dependent low
37 * level driver of CPUFreq support, and its spinlock. This lock
38 * also protects the cpufreq_cpu_data array.
39 */
40static struct cpufreq_driver *cpufreq_driver;
41static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
42#ifdef CONFIG_HOTPLUG_CPU
43/* This one keeps track of the previously set governor of a removed CPU */
44static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
45#endif
46static DEFINE_SPINLOCK(cpufreq_driver_lock);
47
48/*
49 * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
50 * all cpufreq/hotplug/workqueue/etc related lock issues.
51 *
52 * The rules for this semaphore:
53 * - Any routine that wants to read from the policy structure will
54 * do a down_read on this semaphore.
55 * - Any routine that will write to the policy structure and/or may take away
56 * the policy altogether (eg. CPU hotplug), will hold this lock in write
57 * mode before doing so.
58 *
59 * Additional rules:
60 * - All holders of the lock should check to make sure that the CPU they
61 * are concerned with are online after they get the lock.
62 * - Governor routines that can be called in cpufreq hotplug path should not
63 * take this sem as top level hotplug notifier handler takes this.
64 * - Lock should not be held across
65 * __cpufreq_governor(data, CPUFREQ_GOV_STOP);
66 */
67static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
68static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
69
70#define lock_policy_rwsem(mode, cpu) \
71static int lock_policy_rwsem_##mode \
72(int cpu) \
73{ \
74 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
75 BUG_ON(policy_cpu == -1); \
76 down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
77 if (unlikely(!cpu_online(cpu))) { \
78 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
79 return -1; \
80 } \
81 \
82 return 0; \
83}
84
85lock_policy_rwsem(read, cpu);
86
87lock_policy_rwsem(write, cpu);
88
89static void unlock_policy_rwsem_read(int cpu)
90{
91 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
92 BUG_ON(policy_cpu == -1);
93 up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
94}
95
96static void unlock_policy_rwsem_write(int cpu)
97{
98 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
99 BUG_ON(policy_cpu == -1);
100 up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
101}
102
103
104/* internal prototypes */
105static int __cpufreq_governor(struct cpufreq_policy *policy,
106 unsigned int event);
107static unsigned int __cpufreq_get(unsigned int cpu);
108static void handle_update(struct work_struct *work);
109
110/**
111 * Two notifier lists: the "policy" list is involved in the
112 * validation process for a new CPU frequency policy; the
113 * "transition" list for kernel code that needs to handle
114 * changes to devices when the CPU clock speed changes.
115 * The mutex locks both lists.
116 */
117static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
118static struct srcu_notifier_head cpufreq_transition_notifier_list;
119
120static bool init_cpufreq_transition_notifier_list_called;
121static int __init init_cpufreq_transition_notifier_list(void)
122{
123 srcu_init_notifier_head(&cpufreq_transition_notifier_list);
124 init_cpufreq_transition_notifier_list_called = true;
125 return 0;
126}
127pure_initcall(init_cpufreq_transition_notifier_list);
128
129static LIST_HEAD(cpufreq_governor_list);
130static DEFINE_MUTEX(cpufreq_governor_mutex);
131
132struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
133{
134 struct cpufreq_policy *data;
135 unsigned long flags;
136
137 if (cpu >= nr_cpu_ids)
138 goto err_out;
139
140 /* get the cpufreq driver */
141 spin_lock_irqsave(&cpufreq_driver_lock, flags);
142
143 if (!cpufreq_driver)
144 goto err_out_unlock;
145
146 if (!try_module_get(cpufreq_driver->owner))
147 goto err_out_unlock;
148
149
150 /* get the CPU */
151 data = per_cpu(cpufreq_cpu_data, cpu);
152
153 if (!data)
154 goto err_out_put_module;
155
156 if (!kobject_get(&data->kobj))
157 goto err_out_put_module;
158
159 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
160 return data;
161
162err_out_put_module:
163 module_put(cpufreq_driver->owner);
164err_out_unlock:
165 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
166err_out:
167 return NULL;
168}
169EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
170
171
172void cpufreq_cpu_put(struct cpufreq_policy *data)
173{
174 kobject_put(&data->kobj);
175 module_put(cpufreq_driver->owner);
176}
177EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
178
179
180/*********************************************************************
181 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
182 *********************************************************************/
183
184/**
185 * adjust_jiffies - adjust the system "loops_per_jiffy"
186 *
187 * This function alters the system "loops_per_jiffy" for the clock
188 * speed change. Note that loops_per_jiffy cannot be updated on SMP
189 * systems as each CPU might be scaled differently. So, use the arch
190 * per-CPU loops_per_jiffy value wherever possible.
191 */
192#ifndef CONFIG_SMP
193static unsigned long l_p_j_ref;
194static unsigned int l_p_j_ref_freq;
195
196static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
197{
198 if (ci->flags & CPUFREQ_CONST_LOOPS)
199 return;
200
201 if (!l_p_j_ref_freq) {
202 l_p_j_ref = loops_per_jiffy;
203 l_p_j_ref_freq = ci->old;
204 pr_debug("saving %lu as reference value for loops_per_jiffy; "
205 "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
206 }
207 if ((val == CPUFREQ_PRECHANGE && ci->old < ci->new) ||
208 (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
209 (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
210 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
211 ci->new);
212 pr_debug("scaling loops_per_jiffy to %lu "
213 "for frequency %u kHz\n", loops_per_jiffy, ci->new);
214 }
215}
216#else
217static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
218{
219 return;
220}
221#endif
222
223
224/**
225 * cpufreq_notify_transition - call notifier chain and adjust_jiffies
226 * on frequency transition.
227 *
228 * This function calls the transition notifiers and the "adjust_jiffies"
229 * function. It is called twice on all CPU frequency changes that have
230 * external effects.
231 */
232void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
233{
234 struct cpufreq_policy *policy;
235
236 BUG_ON(irqs_disabled());
237
238 freqs->flags = cpufreq_driver->flags;
239 pr_debug("notification %u of frequency transition to %u kHz\n",
240 state, freqs->new);
241
242 policy = per_cpu(cpufreq_cpu_data, freqs->cpu);
243 switch (state) {
244
245 case CPUFREQ_PRECHANGE:
246 /* detect if the driver reported a value as "old frequency"
247 * which is not equal to what the cpufreq core thinks is
248 * "old frequency".
249 */
250 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
251 if ((policy) && (policy->cpu == freqs->cpu) &&
252 (policy->cur) && (policy->cur != freqs->old)) {
253 pr_debug("Warning: CPU frequency is"
254 " %u, cpufreq assumed %u kHz.\n",
255 freqs->old, policy->cur);
256 freqs->old = policy->cur;
257 }
258 }
259 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
260 CPUFREQ_PRECHANGE, freqs);
261 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
262 break;
263
264 case CPUFREQ_POSTCHANGE:
265 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
266 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
267 (unsigned long)freqs->cpu);
268 trace_power_frequency(POWER_PSTATE, freqs->new, freqs->cpu);
269 trace_cpu_frequency(freqs->new, freqs->cpu);
270 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
271 CPUFREQ_POSTCHANGE, freqs);
272 if (likely(policy) && likely(policy->cpu == freqs->cpu))
273 policy->cur = freqs->new;
274 break;
275 }
276}
277EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
278
279
280
281/*********************************************************************
282 * SYSFS INTERFACE *
283 *********************************************************************/
284
285static struct cpufreq_governor *__find_governor(const char *str_governor)
286{
287 struct cpufreq_governor *t;
288
289 list_for_each_entry(t, &cpufreq_governor_list, governor_list)
290 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
291 return t;
292
293 return NULL;
294}
295
296/**
297 * cpufreq_parse_governor - parse a governor string
298 */
299static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
300 struct cpufreq_governor **governor)
301{
302 int err = -EINVAL;
303
304 if (!cpufreq_driver)
305 goto out;
306
307 if (cpufreq_driver->setpolicy) {
308 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
309 *policy = CPUFREQ_POLICY_PERFORMANCE;
310 err = 0;
311 } else if (!strnicmp(str_governor, "powersave",
312 CPUFREQ_NAME_LEN)) {
313 *policy = CPUFREQ_POLICY_POWERSAVE;
314 err = 0;
315 }
316 } else if (cpufreq_driver->target) {
317 struct cpufreq_governor *t;
318
319 mutex_lock(&cpufreq_governor_mutex);
320
321 t = __find_governor(str_governor);
322
323 if (t == NULL) {
324 int ret;
325
326 mutex_unlock(&cpufreq_governor_mutex);
327 ret = request_module("cpufreq_%s", str_governor);
328 mutex_lock(&cpufreq_governor_mutex);
329
330 if (ret == 0)
331 t = __find_governor(str_governor);
332 }
333
334 if (t != NULL) {
335 *governor = t;
336 err = 0;
337 }
338
339 mutex_unlock(&cpufreq_governor_mutex);
340 }
341out:
342 return err;
343}
344
345
346/**
347 * cpufreq_per_cpu_attr_read() / show_##file_name() -
348 * print out cpufreq information
349 *
350 * Write out information from cpufreq_driver->policy[cpu]; object must be
351 * "unsigned int".
352 */
353
354#define show_one(file_name, object) \
355static ssize_t show_##file_name \
356(struct cpufreq_policy *policy, char *buf) \
357{ \
358 return sprintf(buf, "%u\n", policy->object); \
359}
360
361show_one(cpuinfo_min_freq, cpuinfo.min_freq);
362show_one(cpuinfo_max_freq, cpuinfo.max_freq);
363show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
364show_one(scaling_min_freq, min);
365show_one(scaling_max_freq, max);
366show_one(scaling_cur_freq, cur);
367
368static int __cpufreq_set_policy(struct cpufreq_policy *data,
369 struct cpufreq_policy *policy);
370
371/**
372 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
373 */
374#define store_one(file_name, object) \
375static ssize_t store_##file_name \
376(struct cpufreq_policy *policy, const char *buf, size_t count) \
377{ \
378 unsigned int ret = -EINVAL; \
379 struct cpufreq_policy new_policy; \
380 \
381 ret = cpufreq_get_policy(&new_policy, policy->cpu); \
382 if (ret) \
383 return -EINVAL; \
384 \
385 ret = sscanf(buf, "%u", &new_policy.object); \
386 if (ret != 1) \
387 return -EINVAL; \
388 \
389 ret = __cpufreq_set_policy(policy, &new_policy); \
390 policy->user_policy.object = policy->object; \
391 \
392 return ret ? ret : count; \
393}
394
395store_one(scaling_min_freq, min);
396store_one(scaling_max_freq, max);
397
398/**
399 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
400 */
401static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
402 char *buf)
403{
404 unsigned int cur_freq = __cpufreq_get(policy->cpu);
405 if (!cur_freq)
406 return sprintf(buf, "<unknown>");
407 return sprintf(buf, "%u\n", cur_freq);
408}
409
410
411/**
412 * show_scaling_governor - show the current policy for the specified CPU
413 */
414static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
415{
416 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
417 return sprintf(buf, "powersave\n");
418 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
419 return sprintf(buf, "performance\n");
420 else if (policy->governor)
421 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n",
422 policy->governor->name);
423 return -EINVAL;
424}
425
426
427/**
428 * store_scaling_governor - store policy for the specified CPU
429 */
430static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
431 const char *buf, size_t count)
432{
433 unsigned int ret = -EINVAL;
434 char str_governor[16];
435 struct cpufreq_policy new_policy;
436
437 ret = cpufreq_get_policy(&new_policy, policy->cpu);
438 if (ret)
439 return ret;
440
441 ret = sscanf(buf, "%15s", str_governor);
442 if (ret != 1)
443 return -EINVAL;
444
445 if (cpufreq_parse_governor(str_governor, &new_policy.policy,
446 &new_policy.governor))
447 return -EINVAL;
448
449 /* Do not use cpufreq_set_policy here or the user_policy.max
450 will be wrongly overridden */
451 ret = __cpufreq_set_policy(policy, &new_policy);
452
453 policy->user_policy.policy = policy->policy;
454 policy->user_policy.governor = policy->governor;
455
456 if (ret)
457 return ret;
458 else
459 return count;
460}
461
462/**
463 * show_scaling_driver - show the cpufreq driver currently loaded
464 */
465static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
466{
467 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
468}
469
470/**
471 * show_scaling_available_governors - show the available CPUfreq governors
472 */
473static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
474 char *buf)
475{
476 ssize_t i = 0;
477 struct cpufreq_governor *t;
478
479 if (!cpufreq_driver->target) {
480 i += sprintf(buf, "performance powersave");
481 goto out;
482 }
483
484 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
485 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
486 - (CPUFREQ_NAME_LEN + 2)))
487 goto out;
488 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
489 }
490out:
491 i += sprintf(&buf[i], "\n");
492 return i;
493}
494
495static ssize_t show_cpus(const struct cpumask *mask, char *buf)
496{
497 ssize_t i = 0;
498 unsigned int cpu;
499
500 for_each_cpu(cpu, mask) {
501 if (i)
502 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
503 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
504 if (i >= (PAGE_SIZE - 5))
505 break;
506 }
507 i += sprintf(&buf[i], "\n");
508 return i;
509}
510
511/**
512 * show_related_cpus - show the CPUs affected by each transition even if
513 * hw coordination is in use
514 */
515static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
516{
517 if (cpumask_empty(policy->related_cpus))
518 return show_cpus(policy->cpus, buf);
519 return show_cpus(policy->related_cpus, buf);
520}
521
522/**
523 * show_affected_cpus - show the CPUs affected by each transition
524 */
525static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
526{
527 return show_cpus(policy->cpus, buf);
528}
529
530static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
531 const char *buf, size_t count)
532{
533 unsigned int freq = 0;
534 unsigned int ret;
535
536 if (!policy->governor || !policy->governor->store_setspeed)
537 return -EINVAL;
538
539 ret = sscanf(buf, "%u", &freq);
540 if (ret != 1)
541 return -EINVAL;
542
543 policy->governor->store_setspeed(policy, freq);
544
545 return count;
546}
547
548static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
549{
550 if (!policy->governor || !policy->governor->show_setspeed)
551 return sprintf(buf, "<unsupported>\n");
552
553 return policy->governor->show_setspeed(policy, buf);
554}
555
556/**
557 * show_scaling_driver - show the current cpufreq HW/BIOS limitation
558 */
559static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
560{
561 unsigned int limit;
562 int ret;
563 if (cpufreq_driver->bios_limit) {
564 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
565 if (!ret)
566 return sprintf(buf, "%u\n", limit);
567 }
568 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
569}
570
571cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
572cpufreq_freq_attr_ro(cpuinfo_min_freq);
573cpufreq_freq_attr_ro(cpuinfo_max_freq);
574cpufreq_freq_attr_ro(cpuinfo_transition_latency);
575cpufreq_freq_attr_ro(scaling_available_governors);
576cpufreq_freq_attr_ro(scaling_driver);
577cpufreq_freq_attr_ro(scaling_cur_freq);
578cpufreq_freq_attr_ro(bios_limit);
579cpufreq_freq_attr_ro(related_cpus);
580cpufreq_freq_attr_ro(affected_cpus);
581cpufreq_freq_attr_rw(scaling_min_freq);
582cpufreq_freq_attr_rw(scaling_max_freq);
583cpufreq_freq_attr_rw(scaling_governor);
584cpufreq_freq_attr_rw(scaling_setspeed);
585
586static struct attribute *default_attrs[] = {
587 &cpuinfo_min_freq.attr,
588 &cpuinfo_max_freq.attr,
589 &cpuinfo_transition_latency.attr,
590 &scaling_min_freq.attr,
591 &scaling_max_freq.attr,
592 &affected_cpus.attr,
593 &related_cpus.attr,
594 &scaling_governor.attr,
595 &scaling_driver.attr,
596 &scaling_available_governors.attr,
597 &scaling_setspeed.attr,
598 NULL
599};
600
601struct kobject *cpufreq_global_kobject;
602EXPORT_SYMBOL(cpufreq_global_kobject);
603
604#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
605#define to_attr(a) container_of(a, struct freq_attr, attr)
606
607static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
608{
609 struct cpufreq_policy *policy = to_policy(kobj);
610 struct freq_attr *fattr = to_attr(attr);
611 ssize_t ret = -EINVAL;
612 policy = cpufreq_cpu_get(policy->cpu);
613 if (!policy)
614 goto no_policy;
615
616 if (lock_policy_rwsem_read(policy->cpu) < 0)
617 goto fail;
618
619 if (fattr->show)
620 ret = fattr->show(policy, buf);
621 else
622 ret = -EIO;
623
624 unlock_policy_rwsem_read(policy->cpu);
625fail:
626 cpufreq_cpu_put(policy);
627no_policy:
628 return ret;
629}
630
631static ssize_t store(struct kobject *kobj, struct attribute *attr,
632 const char *buf, size_t count)
633{
634 struct cpufreq_policy *policy = to_policy(kobj);
635 struct freq_attr *fattr = to_attr(attr);
636 ssize_t ret = -EINVAL;
637 policy = cpufreq_cpu_get(policy->cpu);
638 if (!policy)
639 goto no_policy;
640
641 if (lock_policy_rwsem_write(policy->cpu) < 0)
642 goto fail;
643
644 if (fattr->store)
645 ret = fattr->store(policy, buf, count);
646 else
647 ret = -EIO;
648
649 unlock_policy_rwsem_write(policy->cpu);
650fail:
651 cpufreq_cpu_put(policy);
652no_policy:
653 return ret;
654}
655
656static void cpufreq_sysfs_release(struct kobject *kobj)
657{
658 struct cpufreq_policy *policy = to_policy(kobj);
659 pr_debug("last reference is dropped\n");
660 complete(&policy->kobj_unregister);
661}
662
663static const struct sysfs_ops sysfs_ops = {
664 .show = show,
665 .store = store,
666};
667
668static struct kobj_type ktype_cpufreq = {
669 .sysfs_ops = &sysfs_ops,
670 .default_attrs = default_attrs,
671 .release = cpufreq_sysfs_release,
672};
673
674/*
675 * Returns:
676 * Negative: Failure
677 * 0: Success
678 * Positive: When we have a managed CPU and the sysfs got symlinked
679 */
680static int cpufreq_add_dev_policy(unsigned int cpu,
681 struct cpufreq_policy *policy,
682 struct sys_device *sys_dev)
683{
684 int ret = 0;
685#ifdef CONFIG_SMP
686 unsigned long flags;
687 unsigned int j;
688#ifdef CONFIG_HOTPLUG_CPU
689 struct cpufreq_governor *gov;
690
691 gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
692 if (gov) {
693 policy->governor = gov;
694 pr_debug("Restoring governor %s for cpu %d\n",
695 policy->governor->name, cpu);
696 }
697#endif
698
699 for_each_cpu(j, policy->cpus) {
700 struct cpufreq_policy *managed_policy;
701
702 if (cpu == j)
703 continue;
704
705 /* Check for existing affected CPUs.
706 * They may not be aware of it due to CPU Hotplug.
707 * cpufreq_cpu_put is called when the device is removed
708 * in __cpufreq_remove_dev()
709 */
710 managed_policy = cpufreq_cpu_get(j);
711 if (unlikely(managed_policy)) {
712
713 /* Set proper policy_cpu */
714 unlock_policy_rwsem_write(cpu);
715 per_cpu(cpufreq_policy_cpu, cpu) = managed_policy->cpu;
716
717 if (lock_policy_rwsem_write(cpu) < 0) {
718 /* Should not go through policy unlock path */
719 if (cpufreq_driver->exit)
720 cpufreq_driver->exit(policy);
721 cpufreq_cpu_put(managed_policy);
722 return -EBUSY;
723 }
724
725 spin_lock_irqsave(&cpufreq_driver_lock, flags);
726 cpumask_copy(managed_policy->cpus, policy->cpus);
727 per_cpu(cpufreq_cpu_data, cpu) = managed_policy;
728 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
729
730 pr_debug("CPU already managed, adding link\n");
731 ret = sysfs_create_link(&sys_dev->kobj,
732 &managed_policy->kobj,
733 "cpufreq");
734 if (ret)
735 cpufreq_cpu_put(managed_policy);
736 /*
737 * Success. We only needed to be added to the mask.
738 * Call driver->exit() because only the cpu parent of
739 * the kobj needed to call init().
740 */
741 if (cpufreq_driver->exit)
742 cpufreq_driver->exit(policy);
743
744 if (!ret)
745 return 1;
746 else
747 return ret;
748 }
749 }
750#endif
751 return ret;
752}
753
754
755/* symlink affected CPUs */
756static int cpufreq_add_dev_symlink(unsigned int cpu,
757 struct cpufreq_policy *policy)
758{
759 unsigned int j;
760 int ret = 0;
761
762 for_each_cpu(j, policy->cpus) {
763 struct cpufreq_policy *managed_policy;
764 struct sys_device *cpu_sys_dev;
765
766 if (j == cpu)
767 continue;
768 if (!cpu_online(j))
769 continue;
770
771 pr_debug("CPU %u already managed, adding link\n", j);
772 managed_policy = cpufreq_cpu_get(cpu);
773 cpu_sys_dev = get_cpu_sysdev(j);
774 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
775 "cpufreq");
776 if (ret) {
777 cpufreq_cpu_put(managed_policy);
778 return ret;
779 }
780 }
781 return ret;
782}
783
784static int cpufreq_add_dev_interface(unsigned int cpu,
785 struct cpufreq_policy *policy,
786 struct sys_device *sys_dev)
787{
788 struct cpufreq_policy new_policy;
789 struct freq_attr **drv_attr;
790 unsigned long flags;
791 int ret = 0;
792 unsigned int j;
793
794 /* prepare interface data */
795 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
796 &sys_dev->kobj, "cpufreq");
797 if (ret)
798 return ret;
799
800 /* set up files for this cpu device */
801 drv_attr = cpufreq_driver->attr;
802 while ((drv_attr) && (*drv_attr)) {
803 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
804 if (ret)
805 goto err_out_kobj_put;
806 drv_attr++;
807 }
808 if (cpufreq_driver->get) {
809 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
810 if (ret)
811 goto err_out_kobj_put;
812 }
813 if (cpufreq_driver->target) {
814 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
815 if (ret)
816 goto err_out_kobj_put;
817 }
818 if (cpufreq_driver->bios_limit) {
819 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
820 if (ret)
821 goto err_out_kobj_put;
822 }
823
824 spin_lock_irqsave(&cpufreq_driver_lock, flags);
825 for_each_cpu(j, policy->cpus) {
826 if (!cpu_online(j))
827 continue;
828 per_cpu(cpufreq_cpu_data, j) = policy;
829 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
830 }
831 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
832
833 ret = cpufreq_add_dev_symlink(cpu, policy);
834 if (ret)
835 goto err_out_kobj_put;
836
837 memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
838 /* assure that the starting sequence is run in __cpufreq_set_policy */
839 policy->governor = NULL;
840
841 /* set default policy */
842 ret = __cpufreq_set_policy(policy, &new_policy);
843 policy->user_policy.policy = policy->policy;
844 policy->user_policy.governor = policy->governor;
845
846 if (ret) {
847 pr_debug("setting policy failed\n");
848 if (cpufreq_driver->exit)
849 cpufreq_driver->exit(policy);
850 }
851 return ret;
852
853err_out_kobj_put:
854 kobject_put(&policy->kobj);
855 wait_for_completion(&policy->kobj_unregister);
856 return ret;
857}
858
859
860/**
861 * cpufreq_add_dev - add a CPU device
862 *
863 * Adds the cpufreq interface for a CPU device.
864 *
865 * The Oracle says: try running cpufreq registration/unregistration concurrently
866 * with with cpu hotplugging and all hell will break loose. Tried to clean this
867 * mess up, but more thorough testing is needed. - Mathieu
868 */
869static int cpufreq_add_dev(struct sys_device *sys_dev)
870{
871 unsigned int cpu = sys_dev->id;
872 int ret = 0, found = 0;
873 struct cpufreq_policy *policy;
874 unsigned long flags;
875 unsigned int j;
876#ifdef CONFIG_HOTPLUG_CPU
877 int sibling;
878#endif
879
880 if (cpu_is_offline(cpu))
881 return 0;
882
883 pr_debug("adding CPU %u\n", cpu);
884
885#ifdef CONFIG_SMP
886 /* check whether a different CPU already registered this
887 * CPU because it is in the same boat. */
888 policy = cpufreq_cpu_get(cpu);
889 if (unlikely(policy)) {
890 cpufreq_cpu_put(policy);
891 return 0;
892 }
893#endif
894
895 if (!try_module_get(cpufreq_driver->owner)) {
896 ret = -EINVAL;
897 goto module_out;
898 }
899
900 ret = -ENOMEM;
901 policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
902 if (!policy)
903 goto nomem_out;
904
905 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
906 goto err_free_policy;
907
908 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
909 goto err_free_cpumask;
910
911 policy->cpu = cpu;
912 cpumask_copy(policy->cpus, cpumask_of(cpu));
913
914 /* Initially set CPU itself as the policy_cpu */
915 per_cpu(cpufreq_policy_cpu, cpu) = cpu;
916 ret = (lock_policy_rwsem_write(cpu) < 0);
917 WARN_ON(ret);
918
919 init_completion(&policy->kobj_unregister);
920 INIT_WORK(&policy->update, handle_update);
921
922 /* Set governor before ->init, so that driver could check it */
923#ifdef CONFIG_HOTPLUG_CPU
924 for_each_online_cpu(sibling) {
925 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
926 if (cp && cp->governor &&
927 (cpumask_test_cpu(cpu, cp->related_cpus))) {
928 policy->governor = cp->governor;
929 found = 1;
930 break;
931 }
932 }
933#endif
934 if (!found)
935 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
936 /* call driver. From then on the cpufreq must be able
937 * to accept all calls to ->verify and ->setpolicy for this CPU
938 */
939 ret = cpufreq_driver->init(policy);
940 if (ret) {
941 pr_debug("initialization failed\n");
942 goto err_unlock_policy;
943 }
944 policy->user_policy.min = policy->min;
945 policy->user_policy.max = policy->max;
946
947 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
948 CPUFREQ_START, policy);
949
950 ret = cpufreq_add_dev_policy(cpu, policy, sys_dev);
951 if (ret) {
952 if (ret > 0)
953 /* This is a managed cpu, symlink created,
954 exit with 0 */
955 ret = 0;
956 goto err_unlock_policy;
957 }
958
959 ret = cpufreq_add_dev_interface(cpu, policy, sys_dev);
960 if (ret)
961 goto err_out_unregister;
962
963 unlock_policy_rwsem_write(cpu);
964
965 kobject_uevent(&policy->kobj, KOBJ_ADD);
966 module_put(cpufreq_driver->owner);
967 pr_debug("initialization complete\n");
968
969 return 0;
970
971
972err_out_unregister:
973 spin_lock_irqsave(&cpufreq_driver_lock, flags);
974 for_each_cpu(j, policy->cpus)
975 per_cpu(cpufreq_cpu_data, j) = NULL;
976 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
977
978 kobject_put(&policy->kobj);
979 wait_for_completion(&policy->kobj_unregister);
980
981err_unlock_policy:
982 unlock_policy_rwsem_write(cpu);
983 free_cpumask_var(policy->related_cpus);
984err_free_cpumask:
985 free_cpumask_var(policy->cpus);
986err_free_policy:
987 kfree(policy);
988nomem_out:
989 module_put(cpufreq_driver->owner);
990module_out:
991 return ret;
992}
993
994
995/**
996 * __cpufreq_remove_dev - remove a CPU device
997 *
998 * Removes the cpufreq interface for a CPU device.
999 * Caller should already have policy_rwsem in write mode for this CPU.
1000 * This routine frees the rwsem before returning.
1001 */
1002static int __cpufreq_remove_dev(struct sys_device *sys_dev)
1003{
1004 unsigned int cpu = sys_dev->id;
1005 unsigned long flags;
1006 struct cpufreq_policy *data;
1007 struct kobject *kobj;
1008 struct completion *cmp;
1009#ifdef CONFIG_SMP
1010 struct sys_device *cpu_sys_dev;
1011 unsigned int j;
1012#endif
1013
1014 pr_debug("unregistering CPU %u\n", cpu);
1015
1016 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1017 data = per_cpu(cpufreq_cpu_data, cpu);
1018
1019 if (!data) {
1020 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1021 unlock_policy_rwsem_write(cpu);
1022 return -EINVAL;
1023 }
1024 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1025
1026
1027#ifdef CONFIG_SMP
1028 /* if this isn't the CPU which is the parent of the kobj, we
1029 * only need to unlink, put and exit
1030 */
1031 if (unlikely(cpu != data->cpu)) {
1032 pr_debug("removing link\n");
1033 cpumask_clear_cpu(cpu, data->cpus);
1034 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1035 kobj = &sys_dev->kobj;
1036 cpufreq_cpu_put(data);
1037 unlock_policy_rwsem_write(cpu);
1038 sysfs_remove_link(kobj, "cpufreq");
1039 return 0;
1040 }
1041#endif
1042
1043#ifdef CONFIG_SMP
1044
1045#ifdef CONFIG_HOTPLUG_CPU
1046 strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name,
1047 CPUFREQ_NAME_LEN);
1048#endif
1049
1050 /* if we have other CPUs still registered, we need to unlink them,
1051 * or else wait_for_completion below will lock up. Clean the
1052 * per_cpu(cpufreq_cpu_data) while holding the lock, and remove
1053 * the sysfs links afterwards.
1054 */
1055 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1056 for_each_cpu(j, data->cpus) {
1057 if (j == cpu)
1058 continue;
1059 per_cpu(cpufreq_cpu_data, j) = NULL;
1060 }
1061 }
1062
1063 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1064
1065 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1066 for_each_cpu(j, data->cpus) {
1067 if (j == cpu)
1068 continue;
1069 pr_debug("removing link for cpu %u\n", j);
1070#ifdef CONFIG_HOTPLUG_CPU
1071 strncpy(per_cpu(cpufreq_cpu_governor, j),
1072 data->governor->name, CPUFREQ_NAME_LEN);
1073#endif
1074 cpu_sys_dev = get_cpu_sysdev(j);
1075 kobj = &cpu_sys_dev->kobj;
1076 unlock_policy_rwsem_write(cpu);
1077 sysfs_remove_link(kobj, "cpufreq");
1078 lock_policy_rwsem_write(cpu);
1079 cpufreq_cpu_put(data);
1080 }
1081 }
1082#else
1083 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1084#endif
1085
1086 if (cpufreq_driver->target)
1087 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1088
1089 kobj = &data->kobj;
1090 cmp = &data->kobj_unregister;
1091 unlock_policy_rwsem_write(cpu);
1092 kobject_put(kobj);
1093
1094 /* we need to make sure that the underlying kobj is actually
1095 * not referenced anymore by anybody before we proceed with
1096 * unloading.
1097 */
1098 pr_debug("waiting for dropping of refcount\n");
1099 wait_for_completion(cmp);
1100 pr_debug("wait complete\n");
1101
1102 lock_policy_rwsem_write(cpu);
1103 if (cpufreq_driver->exit)
1104 cpufreq_driver->exit(data);
1105 unlock_policy_rwsem_write(cpu);
1106
1107#ifdef CONFIG_HOTPLUG_CPU
1108 /* when the CPU which is the parent of the kobj is hotplugged
1109 * offline, check for siblings, and create cpufreq sysfs interface
1110 * and symlinks
1111 */
1112 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1113 /* first sibling now owns the new sysfs dir */
1114 cpumask_clear_cpu(cpu, data->cpus);
1115 cpufreq_add_dev(get_cpu_sysdev(cpumask_first(data->cpus)));
1116
1117 /* finally remove our own symlink */
1118 lock_policy_rwsem_write(cpu);
1119 __cpufreq_remove_dev(sys_dev);
1120 }
1121#endif
1122
1123 free_cpumask_var(data->related_cpus);
1124 free_cpumask_var(data->cpus);
1125 kfree(data);
1126
1127 return 0;
1128}
1129
1130
1131static int cpufreq_remove_dev(struct sys_device *sys_dev)
1132{
1133 unsigned int cpu = sys_dev->id;
1134 int retval;
1135
1136 if (cpu_is_offline(cpu))
1137 return 0;
1138
1139 if (unlikely(lock_policy_rwsem_write(cpu)))
1140 BUG();
1141
1142 retval = __cpufreq_remove_dev(sys_dev);
1143 return retval;
1144}
1145
1146
1147static void handle_update(struct work_struct *work)
1148{
1149 struct cpufreq_policy *policy =
1150 container_of(work, struct cpufreq_policy, update);
1151 unsigned int cpu = policy->cpu;
1152 pr_debug("handle_update for cpu %u called\n", cpu);
1153 cpufreq_update_policy(cpu);
1154}
1155
1156/**
1157 * cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1158 * @cpu: cpu number
1159 * @old_freq: CPU frequency the kernel thinks the CPU runs at
1160 * @new_freq: CPU frequency the CPU actually runs at
1161 *
1162 * We adjust to current frequency first, and need to clean up later.
1163 * So either call to cpufreq_update_policy() or schedule handle_update()).
1164 */
1165static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1166 unsigned int new_freq)
1167{
1168 struct cpufreq_freqs freqs;
1169
1170 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1171 "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1172
1173 freqs.cpu = cpu;
1174 freqs.old = old_freq;
1175 freqs.new = new_freq;
1176 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1177 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1178}
1179
1180
1181/**
1182 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1183 * @cpu: CPU number
1184 *
1185 * This is the last known freq, without actually getting it from the driver.
1186 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1187 */
1188unsigned int cpufreq_quick_get(unsigned int cpu)
1189{
1190 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1191 unsigned int ret_freq = 0;
1192
1193 if (policy) {
1194 ret_freq = policy->cur;
1195 cpufreq_cpu_put(policy);
1196 }
1197
1198 return ret_freq;
1199}
1200EXPORT_SYMBOL(cpufreq_quick_get);
1201
1202/**
1203 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1204 * @cpu: CPU number
1205 *
1206 * Just return the max possible frequency for a given CPU.
1207 */
1208unsigned int cpufreq_quick_get_max(unsigned int cpu)
1209{
1210 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1211 unsigned int ret_freq = 0;
1212
1213 if (policy) {
1214 ret_freq = policy->max;
1215 cpufreq_cpu_put(policy);
1216 }
1217
1218 return ret_freq;
1219}
1220EXPORT_SYMBOL(cpufreq_quick_get_max);
1221
1222
1223static unsigned int __cpufreq_get(unsigned int cpu)
1224{
1225 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1226 unsigned int ret_freq = 0;
1227
1228 if (!cpufreq_driver->get)
1229 return ret_freq;
1230
1231 ret_freq = cpufreq_driver->get(cpu);
1232
1233 if (ret_freq && policy->cur &&
1234 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1235 /* verify no discrepancy between actual and
1236 saved value exists */
1237 if (unlikely(ret_freq != policy->cur)) {
1238 cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1239 schedule_work(&policy->update);
1240 }
1241 }
1242
1243 return ret_freq;
1244}
1245
1246/**
1247 * cpufreq_get - get the current CPU frequency (in kHz)
1248 * @cpu: CPU number
1249 *
1250 * Get the CPU current (static) CPU frequency
1251 */
1252unsigned int cpufreq_get(unsigned int cpu)
1253{
1254 unsigned int ret_freq = 0;
1255 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1256
1257 if (!policy)
1258 goto out;
1259
1260 if (unlikely(lock_policy_rwsem_read(cpu)))
1261 goto out_policy;
1262
1263 ret_freq = __cpufreq_get(cpu);
1264
1265 unlock_policy_rwsem_read(cpu);
1266
1267out_policy:
1268 cpufreq_cpu_put(policy);
1269out:
1270 return ret_freq;
1271}
1272EXPORT_SYMBOL(cpufreq_get);
1273
1274static struct sysdev_driver cpufreq_sysdev_driver = {
1275 .add = cpufreq_add_dev,
1276 .remove = cpufreq_remove_dev,
1277};
1278
1279
1280/**
1281 * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1282 *
1283 * This function is only executed for the boot processor. The other CPUs
1284 * have been put offline by means of CPU hotplug.
1285 */
1286static int cpufreq_bp_suspend(void)
1287{
1288 int ret = 0;
1289
1290 int cpu = smp_processor_id();
1291 struct cpufreq_policy *cpu_policy;
1292
1293 pr_debug("suspending cpu %u\n", cpu);
1294
1295 /* If there's no policy for the boot CPU, we have nothing to do. */
1296 cpu_policy = cpufreq_cpu_get(cpu);
1297 if (!cpu_policy)
1298 return 0;
1299
1300 if (cpufreq_driver->suspend) {
1301 ret = cpufreq_driver->suspend(cpu_policy);
1302 if (ret)
1303 printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1304 "step on CPU %u\n", cpu_policy->cpu);
1305 }
1306
1307 cpufreq_cpu_put(cpu_policy);
1308 return ret;
1309}
1310
1311/**
1312 * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1313 *
1314 * 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1315 * 2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1316 * restored. It will verify that the current freq is in sync with
1317 * what we believe it to be. This is a bit later than when it
1318 * should be, but nonethteless it's better than calling
1319 * cpufreq_driver->get() here which might re-enable interrupts...
1320 *
1321 * This function is only executed for the boot CPU. The other CPUs have not
1322 * been turned on yet.
1323 */
1324static void cpufreq_bp_resume(void)
1325{
1326 int ret = 0;
1327
1328 int cpu = smp_processor_id();
1329 struct cpufreq_policy *cpu_policy;
1330
1331 pr_debug("resuming cpu %u\n", cpu);
1332
1333 /* If there's no policy for the boot CPU, we have nothing to do. */
1334 cpu_policy = cpufreq_cpu_get(cpu);
1335 if (!cpu_policy)
1336 return;
1337
1338 if (cpufreq_driver->resume) {
1339 ret = cpufreq_driver->resume(cpu_policy);
1340 if (ret) {
1341 printk(KERN_ERR "cpufreq: resume failed in ->resume "
1342 "step on CPU %u\n", cpu_policy->cpu);
1343 goto fail;
1344 }
1345 }
1346
1347 schedule_work(&cpu_policy->update);
1348
1349fail:
1350 cpufreq_cpu_put(cpu_policy);
1351}
1352
1353static struct syscore_ops cpufreq_syscore_ops = {
1354 .suspend = cpufreq_bp_suspend,
1355 .resume = cpufreq_bp_resume,
1356};
1357
1358
1359/*********************************************************************
1360 * NOTIFIER LISTS INTERFACE *
1361 *********************************************************************/
1362
1363/**
1364 * cpufreq_register_notifier - register a driver with cpufreq
1365 * @nb: notifier function to register
1366 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1367 *
1368 * Add a driver to one of two lists: either a list of drivers that
1369 * are notified about clock rate changes (once before and once after
1370 * the transition), or a list of drivers that are notified about
1371 * changes in cpufreq policy.
1372 *
1373 * This function may sleep, and has the same return conditions as
1374 * blocking_notifier_chain_register.
1375 */
1376int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1377{
1378 int ret;
1379
1380 WARN_ON(!init_cpufreq_transition_notifier_list_called);
1381
1382 switch (list) {
1383 case CPUFREQ_TRANSITION_NOTIFIER:
1384 ret = srcu_notifier_chain_register(
1385 &cpufreq_transition_notifier_list, nb);
1386 break;
1387 case CPUFREQ_POLICY_NOTIFIER:
1388 ret = blocking_notifier_chain_register(
1389 &cpufreq_policy_notifier_list, nb);
1390 break;
1391 default:
1392 ret = -EINVAL;
1393 }
1394
1395 return ret;
1396}
1397EXPORT_SYMBOL(cpufreq_register_notifier);
1398
1399
1400/**
1401 * cpufreq_unregister_notifier - unregister a driver with cpufreq
1402 * @nb: notifier block to be unregistered
1403 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1404 *
1405 * Remove a driver from the CPU frequency notifier list.
1406 *
1407 * This function may sleep, and has the same return conditions as
1408 * blocking_notifier_chain_unregister.
1409 */
1410int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1411{
1412 int ret;
1413
1414 switch (list) {
1415 case CPUFREQ_TRANSITION_NOTIFIER:
1416 ret = srcu_notifier_chain_unregister(
1417 &cpufreq_transition_notifier_list, nb);
1418 break;
1419 case CPUFREQ_POLICY_NOTIFIER:
1420 ret = blocking_notifier_chain_unregister(
1421 &cpufreq_policy_notifier_list, nb);
1422 break;
1423 default:
1424 ret = -EINVAL;
1425 }
1426
1427 return ret;
1428}
1429EXPORT_SYMBOL(cpufreq_unregister_notifier);
1430
1431
1432/*********************************************************************
1433 * GOVERNORS *
1434 *********************************************************************/
1435
1436
1437int __cpufreq_driver_target(struct cpufreq_policy *policy,
1438 unsigned int target_freq,
1439 unsigned int relation)
1440{
1441 int retval = -EINVAL;
1442
1443 pr_debug("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1444 target_freq, relation);
1445 if (cpu_online(policy->cpu) && cpufreq_driver->target)
1446 retval = cpufreq_driver->target(policy, target_freq, relation);
1447
1448 return retval;
1449}
1450EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1451
1452int cpufreq_driver_target(struct cpufreq_policy *policy,
1453 unsigned int target_freq,
1454 unsigned int relation)
1455{
1456 int ret = -EINVAL;
1457
1458 policy = cpufreq_cpu_get(policy->cpu);
1459 if (!policy)
1460 goto no_policy;
1461
1462 if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1463 goto fail;
1464
1465 ret = __cpufreq_driver_target(policy, target_freq, relation);
1466
1467 unlock_policy_rwsem_write(policy->cpu);
1468
1469fail:
1470 cpufreq_cpu_put(policy);
1471no_policy:
1472 return ret;
1473}
1474EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1475
1476int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
1477{
1478 int ret = 0;
1479
1480 policy = cpufreq_cpu_get(policy->cpu);
1481 if (!policy)
1482 return -EINVAL;
1483
1484 if (cpu_online(cpu) && cpufreq_driver->getavg)
1485 ret = cpufreq_driver->getavg(policy, cpu);
1486
1487 cpufreq_cpu_put(policy);
1488 return ret;
1489}
1490EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1491
1492/*
1493 * when "event" is CPUFREQ_GOV_LIMITS
1494 */
1495
1496static int __cpufreq_governor(struct cpufreq_policy *policy,
1497 unsigned int event)
1498{
1499 int ret;
1500
1501 /* Only must be defined when default governor is known to have latency
1502 restrictions, like e.g. conservative or ondemand.
1503 That this is the case is already ensured in Kconfig
1504 */
1505#ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1506 struct cpufreq_governor *gov = &cpufreq_gov_performance;
1507#else
1508 struct cpufreq_governor *gov = NULL;
1509#endif
1510
1511 if (policy->governor->max_transition_latency &&
1512 policy->cpuinfo.transition_latency >
1513 policy->governor->max_transition_latency) {
1514 if (!gov)
1515 return -EINVAL;
1516 else {
1517 printk(KERN_WARNING "%s governor failed, too long"
1518 " transition latency of HW, fallback"
1519 " to %s governor\n",
1520 policy->governor->name,
1521 gov->name);
1522 policy->governor = gov;
1523 }
1524 }
1525
1526 if (!try_module_get(policy->governor->owner))
1527 return -EINVAL;
1528
1529 pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1530 policy->cpu, event);
1531 ret = policy->governor->governor(policy, event);
1532
1533 /* we keep one module reference alive for
1534 each CPU governed by this CPU */
1535 if ((event != CPUFREQ_GOV_START) || ret)
1536 module_put(policy->governor->owner);
1537 if ((event == CPUFREQ_GOV_STOP) && !ret)
1538 module_put(policy->governor->owner);
1539
1540 return ret;
1541}
1542
1543
1544int cpufreq_register_governor(struct cpufreq_governor *governor)
1545{
1546 int err;
1547
1548 if (!governor)
1549 return -EINVAL;
1550
1551 mutex_lock(&cpufreq_governor_mutex);
1552
1553 err = -EBUSY;
1554 if (__find_governor(governor->name) == NULL) {
1555 err = 0;
1556 list_add(&governor->governor_list, &cpufreq_governor_list);
1557 }
1558
1559 mutex_unlock(&cpufreq_governor_mutex);
1560 return err;
1561}
1562EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1563
1564
1565void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1566{
1567#ifdef CONFIG_HOTPLUG_CPU
1568 int cpu;
1569#endif
1570
1571 if (!governor)
1572 return;
1573
1574#ifdef CONFIG_HOTPLUG_CPU
1575 for_each_present_cpu(cpu) {
1576 if (cpu_online(cpu))
1577 continue;
1578 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1579 strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1580 }
1581#endif
1582
1583 mutex_lock(&cpufreq_governor_mutex);
1584 list_del(&governor->governor_list);
1585 mutex_unlock(&cpufreq_governor_mutex);
1586 return;
1587}
1588EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1589
1590
1591
1592/*********************************************************************
1593 * POLICY INTERFACE *
1594 *********************************************************************/
1595
1596/**
1597 * cpufreq_get_policy - get the current cpufreq_policy
1598 * @policy: struct cpufreq_policy into which the current cpufreq_policy
1599 * is written
1600 *
1601 * Reads the current cpufreq policy.
1602 */
1603int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1604{
1605 struct cpufreq_policy *cpu_policy;
1606 if (!policy)
1607 return -EINVAL;
1608
1609 cpu_policy = cpufreq_cpu_get(cpu);
1610 if (!cpu_policy)
1611 return -EINVAL;
1612
1613 memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1614
1615 cpufreq_cpu_put(cpu_policy);
1616 return 0;
1617}
1618EXPORT_SYMBOL(cpufreq_get_policy);
1619
1620
1621/*
1622 * data : current policy.
1623 * policy : policy to be set.
1624 */
1625static int __cpufreq_set_policy(struct cpufreq_policy *data,
1626 struct cpufreq_policy *policy)
1627{
1628 int ret = 0;
1629
1630 pr_debug("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1631 policy->min, policy->max);
1632
1633 memcpy(&policy->cpuinfo, &data->cpuinfo,
1634 sizeof(struct cpufreq_cpuinfo));
1635
1636 if (policy->min > data->max || policy->max < data->min) {
1637 ret = -EINVAL;
1638 goto error_out;
1639 }
1640
1641 /* verify the cpu speed can be set within this limit */
1642 ret = cpufreq_driver->verify(policy);
1643 if (ret)
1644 goto error_out;
1645
1646 /* adjust if necessary - all reasons */
1647 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1648 CPUFREQ_ADJUST, policy);
1649
1650 /* adjust if necessary - hardware incompatibility*/
1651 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1652 CPUFREQ_INCOMPATIBLE, policy);
1653
1654 /* verify the cpu speed can be set within this limit,
1655 which might be different to the first one */
1656 ret = cpufreq_driver->verify(policy);
1657 if (ret)
1658 goto error_out;
1659
1660 /* notification of the new policy */
1661 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1662 CPUFREQ_NOTIFY, policy);
1663
1664 data->min = policy->min;
1665 data->max = policy->max;
1666
1667 pr_debug("new min and max freqs are %u - %u kHz\n",
1668 data->min, data->max);
1669
1670 if (cpufreq_driver->setpolicy) {
1671 data->policy = policy->policy;
1672 pr_debug("setting range\n");
1673 ret = cpufreq_driver->setpolicy(policy);
1674 } else {
1675 if (policy->governor != data->governor) {
1676 /* save old, working values */
1677 struct cpufreq_governor *old_gov = data->governor;
1678
1679 pr_debug("governor switch\n");
1680
1681 /* end old governor */
1682 if (data->governor)
1683 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1684
1685 /* start new governor */
1686 data->governor = policy->governor;
1687 if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1688 /* new governor failed, so re-start old one */
1689 pr_debug("starting governor %s failed\n",
1690 data->governor->name);
1691 if (old_gov) {
1692 data->governor = old_gov;
1693 __cpufreq_governor(data,
1694 CPUFREQ_GOV_START);
1695 }
1696 ret = -EINVAL;
1697 goto error_out;
1698 }
1699 /* might be a policy change, too, so fall through */
1700 }
1701 pr_debug("governor: change or update limits\n");
1702 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1703 }
1704
1705error_out:
1706 return ret;
1707}
1708
1709/**
1710 * cpufreq_update_policy - re-evaluate an existing cpufreq policy
1711 * @cpu: CPU which shall be re-evaluated
1712 *
1713 * Useful for policy notifiers which have different necessities
1714 * at different times.
1715 */
1716int cpufreq_update_policy(unsigned int cpu)
1717{
1718 struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1719 struct cpufreq_policy policy;
1720 int ret;
1721
1722 if (!data) {
1723 ret = -ENODEV;
1724 goto no_policy;
1725 }
1726
1727 if (unlikely(lock_policy_rwsem_write(cpu))) {
1728 ret = -EINVAL;
1729 goto fail;
1730 }
1731
1732 pr_debug("updating policy for CPU %u\n", cpu);
1733 memcpy(&policy, data, sizeof(struct cpufreq_policy));
1734 policy.min = data->user_policy.min;
1735 policy.max = data->user_policy.max;
1736 policy.policy = data->user_policy.policy;
1737 policy.governor = data->user_policy.governor;
1738
1739 /* BIOS might change freq behind our back
1740 -> ask driver for current freq and notify governors about a change */
1741 if (cpufreq_driver->get) {
1742 policy.cur = cpufreq_driver->get(cpu);
1743 if (!data->cur) {
1744 pr_debug("Driver did not initialize current freq");
1745 data->cur = policy.cur;
1746 } else {
1747 if (data->cur != policy.cur)
1748 cpufreq_out_of_sync(cpu, data->cur,
1749 policy.cur);
1750 }
1751 }
1752
1753 ret = __cpufreq_set_policy(data, &policy);
1754
1755 unlock_policy_rwsem_write(cpu);
1756
1757fail:
1758 cpufreq_cpu_put(data);
1759no_policy:
1760 return ret;
1761}
1762EXPORT_SYMBOL(cpufreq_update_policy);
1763
1764static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1765 unsigned long action, void *hcpu)
1766{
1767 unsigned int cpu = (unsigned long)hcpu;
1768 struct sys_device *sys_dev;
1769
1770 sys_dev = get_cpu_sysdev(cpu);
1771 if (sys_dev) {
1772 switch (action) {
1773 case CPU_ONLINE:
1774 case CPU_ONLINE_FROZEN:
1775 cpufreq_add_dev(sys_dev);
1776 break;
1777 case CPU_DOWN_PREPARE:
1778 case CPU_DOWN_PREPARE_FROZEN:
1779 if (unlikely(lock_policy_rwsem_write(cpu)))
1780 BUG();
1781
1782 __cpufreq_remove_dev(sys_dev);
1783 break;
1784 case CPU_DOWN_FAILED:
1785 case CPU_DOWN_FAILED_FROZEN:
1786 cpufreq_add_dev(sys_dev);
1787 break;
1788 }
1789 }
1790 return NOTIFY_OK;
1791}
1792
1793static struct notifier_block __refdata cpufreq_cpu_notifier = {
1794 .notifier_call = cpufreq_cpu_callback,
1795};
1796
1797/*********************************************************************
1798 * REGISTER / UNREGISTER CPUFREQ DRIVER *
1799 *********************************************************************/
1800
1801/**
1802 * cpufreq_register_driver - register a CPU Frequency driver
1803 * @driver_data: A struct cpufreq_driver containing the values#
1804 * submitted by the CPU Frequency driver.
1805 *
1806 * Registers a CPU Frequency driver to this core code. This code
1807 * returns zero on success, -EBUSY when another driver got here first
1808 * (and isn't unregistered in the meantime).
1809 *
1810 */
1811int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1812{
1813 unsigned long flags;
1814 int ret;
1815
1816 if (!driver_data || !driver_data->verify || !driver_data->init ||
1817 ((!driver_data->setpolicy) && (!driver_data->target)))
1818 return -EINVAL;
1819
1820 pr_debug("trying to register driver %s\n", driver_data->name);
1821
1822 if (driver_data->setpolicy)
1823 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1824
1825 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1826 if (cpufreq_driver) {
1827 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1828 return -EBUSY;
1829 }
1830 cpufreq_driver = driver_data;
1831 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1832
1833 ret = sysdev_driver_register(&cpu_sysdev_class,
1834 &cpufreq_sysdev_driver);
1835 if (ret)
1836 goto err_null_driver;
1837
1838 if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1839 int i;
1840 ret = -ENODEV;
1841
1842 /* check for at least one working CPU */
1843 for (i = 0; i < nr_cpu_ids; i++)
1844 if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
1845 ret = 0;
1846 break;
1847 }
1848
1849 /* if all ->init() calls failed, unregister */
1850 if (ret) {
1851 pr_debug("no CPU initialized for driver %s\n",
1852 driver_data->name);
1853 goto err_sysdev_unreg;
1854 }
1855 }
1856
1857 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1858 pr_debug("driver %s up and running\n", driver_data->name);
1859
1860 return 0;
1861err_sysdev_unreg:
1862 sysdev_driver_unregister(&cpu_sysdev_class,
1863 &cpufreq_sysdev_driver);
1864err_null_driver:
1865 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1866 cpufreq_driver = NULL;
1867 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1868 return ret;
1869}
1870EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1871
1872
1873/**
1874 * cpufreq_unregister_driver - unregister the current CPUFreq driver
1875 *
1876 * Unregister the current CPUFreq driver. Only call this if you have
1877 * the right to do so, i.e. if you have succeeded in initialising before!
1878 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1879 * currently not initialised.
1880 */
1881int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1882{
1883 unsigned long flags;
1884
1885 if (!cpufreq_driver || (driver != cpufreq_driver))
1886 return -EINVAL;
1887
1888 pr_debug("unregistering driver %s\n", driver->name);
1889
1890 sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1891 unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1892
1893 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1894 cpufreq_driver = NULL;
1895 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1896
1897 return 0;
1898}
1899EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1900
1901static int __init cpufreq_core_init(void)
1902{
1903 int cpu;
1904
1905 for_each_possible_cpu(cpu) {
1906 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1907 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1908 }
1909
1910 cpufreq_global_kobject = kobject_create_and_add("cpufreq",
1911 &cpu_sysdev_class.kset.kobj);
1912 BUG_ON(!cpufreq_global_kobject);
1913 register_syscore_ops(&cpufreq_syscore_ops);
1914
1915 return 0;
1916}
1917core_initcall(cpufreq_core_init);
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/pm_qos.h>
27#include <linux/slab.h>
28#include <linux/suspend.h>
29#include <linux/syscore_ops.h>
30#include <linux/tick.h>
31#include <linux/units.h>
32#include <trace/events/power.h>
33
34static LIST_HEAD(cpufreq_policy_list);
35
36/* Macros to iterate over CPU policies */
37#define for_each_suitable_policy(__policy, __active) \
38 list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
39 if ((__active) == !policy_is_inactive(__policy))
40
41#define for_each_active_policy(__policy) \
42 for_each_suitable_policy(__policy, true)
43#define for_each_inactive_policy(__policy) \
44 for_each_suitable_policy(__policy, false)
45
46/* Iterate over governors */
47static LIST_HEAD(cpufreq_governor_list);
48#define for_each_governor(__governor) \
49 list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
50
51static char default_governor[CPUFREQ_NAME_LEN];
52
53/*
54 * The "cpufreq driver" - the arch- or hardware-dependent low
55 * level driver of CPUFreq support, and its spinlock. This lock
56 * also protects the cpufreq_cpu_data array.
57 */
58static struct cpufreq_driver *cpufreq_driver;
59static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
60static DEFINE_RWLOCK(cpufreq_driver_lock);
61
62static DEFINE_STATIC_KEY_FALSE(cpufreq_freq_invariance);
63bool cpufreq_supports_freq_invariance(void)
64{
65 return static_branch_likely(&cpufreq_freq_invariance);
66}
67
68/* Flag to suspend/resume CPUFreq governors */
69static bool cpufreq_suspended;
70
71static inline bool has_target(void)
72{
73 return cpufreq_driver->target_index || cpufreq_driver->target;
74}
75
76bool has_target_index(void)
77{
78 return !!cpufreq_driver->target_index;
79}
80
81/* internal prototypes */
82static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
83static int cpufreq_init_governor(struct cpufreq_policy *policy);
84static void cpufreq_exit_governor(struct cpufreq_policy *policy);
85static void cpufreq_governor_limits(struct cpufreq_policy *policy);
86static int cpufreq_set_policy(struct cpufreq_policy *policy,
87 struct cpufreq_governor *new_gov,
88 unsigned int new_pol);
89static bool cpufreq_boost_supported(void);
90
91/*
92 * Two notifier lists: the "policy" list is involved in the
93 * validation process for a new CPU frequency policy; the
94 * "transition" list for kernel code that needs to handle
95 * changes to devices when the CPU clock speed changes.
96 * The mutex locks both lists.
97 */
98static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
99SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
100
101static int off __read_mostly;
102static int cpufreq_disabled(void)
103{
104 return off;
105}
106void disable_cpufreq(void)
107{
108 off = 1;
109}
110static DEFINE_MUTEX(cpufreq_governor_mutex);
111
112bool have_governor_per_policy(void)
113{
114 return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
115}
116EXPORT_SYMBOL_GPL(have_governor_per_policy);
117
118static struct kobject *cpufreq_global_kobject;
119
120struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
121{
122 if (have_governor_per_policy())
123 return &policy->kobj;
124 else
125 return cpufreq_global_kobject;
126}
127EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
128
129static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
130{
131 struct kernel_cpustat kcpustat;
132 u64 cur_wall_time;
133 u64 idle_time;
134 u64 busy_time;
135
136 cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
137
138 kcpustat_cpu_fetch(&kcpustat, cpu);
139
140 busy_time = kcpustat.cpustat[CPUTIME_USER];
141 busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
142 busy_time += kcpustat.cpustat[CPUTIME_IRQ];
143 busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
144 busy_time += kcpustat.cpustat[CPUTIME_STEAL];
145 busy_time += kcpustat.cpustat[CPUTIME_NICE];
146
147 idle_time = cur_wall_time - busy_time;
148 if (wall)
149 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
150
151 return div_u64(idle_time, NSEC_PER_USEC);
152}
153
154u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
155{
156 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
157
158 if (idle_time == -1ULL)
159 return get_cpu_idle_time_jiffy(cpu, wall);
160 else if (!io_busy)
161 idle_time += get_cpu_iowait_time_us(cpu, wall);
162
163 return idle_time;
164}
165EXPORT_SYMBOL_GPL(get_cpu_idle_time);
166
167/*
168 * This is a generic cpufreq init() routine which can be used by cpufreq
169 * drivers of SMP systems. It will do following:
170 * - validate & show freq table passed
171 * - set policies transition latency
172 * - policy->cpus with all possible CPUs
173 */
174void cpufreq_generic_init(struct cpufreq_policy *policy,
175 struct cpufreq_frequency_table *table,
176 unsigned int transition_latency)
177{
178 policy->freq_table = table;
179 policy->cpuinfo.transition_latency = transition_latency;
180
181 /*
182 * The driver only supports the SMP configuration where all processors
183 * share the clock and voltage and clock.
184 */
185 cpumask_setall(policy->cpus);
186}
187EXPORT_SYMBOL_GPL(cpufreq_generic_init);
188
189struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
190{
191 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
192
193 return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
194}
195EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
196
197unsigned int cpufreq_generic_get(unsigned int cpu)
198{
199 struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
200
201 if (!policy || IS_ERR(policy->clk)) {
202 pr_err("%s: No %s associated to cpu: %d\n",
203 __func__, policy ? "clk" : "policy", cpu);
204 return 0;
205 }
206
207 return clk_get_rate(policy->clk) / 1000;
208}
209EXPORT_SYMBOL_GPL(cpufreq_generic_get);
210
211/**
212 * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
213 * @cpu: CPU to find the policy for.
214 *
215 * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
216 * the kobject reference counter of that policy. Return a valid policy on
217 * success or NULL on failure.
218 *
219 * The policy returned by this function has to be released with the help of
220 * cpufreq_cpu_put() to balance its kobject reference counter properly.
221 */
222struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
223{
224 struct cpufreq_policy *policy = NULL;
225 unsigned long flags;
226
227 if (WARN_ON(cpu >= nr_cpu_ids))
228 return NULL;
229
230 /* get the cpufreq driver */
231 read_lock_irqsave(&cpufreq_driver_lock, flags);
232
233 if (cpufreq_driver) {
234 /* get the CPU */
235 policy = cpufreq_cpu_get_raw(cpu);
236 if (policy)
237 kobject_get(&policy->kobj);
238 }
239
240 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
241
242 return policy;
243}
244EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
245
246/**
247 * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
248 * @policy: cpufreq policy returned by cpufreq_cpu_get().
249 */
250void cpufreq_cpu_put(struct cpufreq_policy *policy)
251{
252 kobject_put(&policy->kobj);
253}
254EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
255
256/**
257 * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
258 * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
259 */
260void cpufreq_cpu_release(struct cpufreq_policy *policy)
261{
262 if (WARN_ON(!policy))
263 return;
264
265 lockdep_assert_held(&policy->rwsem);
266
267 up_write(&policy->rwsem);
268
269 cpufreq_cpu_put(policy);
270}
271
272/**
273 * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
274 * @cpu: CPU to find the policy for.
275 *
276 * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
277 * if the policy returned by it is not NULL, acquire its rwsem for writing.
278 * Return the policy if it is active or release it and return NULL otherwise.
279 *
280 * The policy returned by this function has to be released with the help of
281 * cpufreq_cpu_release() in order to release its rwsem and balance its usage
282 * counter properly.
283 */
284struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
285{
286 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
287
288 if (!policy)
289 return NULL;
290
291 down_write(&policy->rwsem);
292
293 if (policy_is_inactive(policy)) {
294 cpufreq_cpu_release(policy);
295 return NULL;
296 }
297
298 return policy;
299}
300
301/*********************************************************************
302 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
303 *********************************************************************/
304
305/**
306 * adjust_jiffies - Adjust the system "loops_per_jiffy".
307 * @val: CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
308 * @ci: Frequency change information.
309 *
310 * This function alters the system "loops_per_jiffy" for the clock
311 * speed change. Note that loops_per_jiffy cannot be updated on SMP
312 * systems as each CPU might be scaled differently. So, use the arch
313 * per-CPU loops_per_jiffy value wherever possible.
314 */
315static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
316{
317#ifndef CONFIG_SMP
318 static unsigned long l_p_j_ref;
319 static unsigned int l_p_j_ref_freq;
320
321 if (ci->flags & CPUFREQ_CONST_LOOPS)
322 return;
323
324 if (!l_p_j_ref_freq) {
325 l_p_j_ref = loops_per_jiffy;
326 l_p_j_ref_freq = ci->old;
327 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
328 l_p_j_ref, l_p_j_ref_freq);
329 }
330 if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
331 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
332 ci->new);
333 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
334 loops_per_jiffy, ci->new);
335 }
336#endif
337}
338
339/**
340 * cpufreq_notify_transition - Notify frequency transition and adjust jiffies.
341 * @policy: cpufreq policy to enable fast frequency switching for.
342 * @freqs: contain details of the frequency update.
343 * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
344 *
345 * This function calls the transition notifiers and adjust_jiffies().
346 *
347 * It is called twice on all CPU frequency changes that have external effects.
348 */
349static void cpufreq_notify_transition(struct cpufreq_policy *policy,
350 struct cpufreq_freqs *freqs,
351 unsigned int state)
352{
353 int cpu;
354
355 BUG_ON(irqs_disabled());
356
357 if (cpufreq_disabled())
358 return;
359
360 freqs->policy = policy;
361 freqs->flags = cpufreq_driver->flags;
362 pr_debug("notification %u of frequency transition to %u kHz\n",
363 state, freqs->new);
364
365 switch (state) {
366 case CPUFREQ_PRECHANGE:
367 /*
368 * Detect if the driver reported a value as "old frequency"
369 * which is not equal to what the cpufreq core thinks is
370 * "old frequency".
371 */
372 if (policy->cur && policy->cur != freqs->old) {
373 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
374 freqs->old, policy->cur);
375 freqs->old = policy->cur;
376 }
377
378 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
379 CPUFREQ_PRECHANGE, freqs);
380
381 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
382 break;
383
384 case CPUFREQ_POSTCHANGE:
385 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
386 pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
387 cpumask_pr_args(policy->cpus));
388
389 for_each_cpu(cpu, policy->cpus)
390 trace_cpu_frequency(freqs->new, cpu);
391
392 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
393 CPUFREQ_POSTCHANGE, freqs);
394
395 cpufreq_stats_record_transition(policy, freqs->new);
396 policy->cur = freqs->new;
397 }
398}
399
400/* Do post notifications when there are chances that transition has failed */
401static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
402 struct cpufreq_freqs *freqs, int transition_failed)
403{
404 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
405 if (!transition_failed)
406 return;
407
408 swap(freqs->old, freqs->new);
409 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
410 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
411}
412
413void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
414 struct cpufreq_freqs *freqs)
415{
416
417 /*
418 * Catch double invocations of _begin() which lead to self-deadlock.
419 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
420 * doesn't invoke _begin() on their behalf, and hence the chances of
421 * double invocations are very low. Moreover, there are scenarios
422 * where these checks can emit false-positive warnings in these
423 * drivers; so we avoid that by skipping them altogether.
424 */
425 WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
426 && current == policy->transition_task);
427
428wait:
429 wait_event(policy->transition_wait, !policy->transition_ongoing);
430
431 spin_lock(&policy->transition_lock);
432
433 if (unlikely(policy->transition_ongoing)) {
434 spin_unlock(&policy->transition_lock);
435 goto wait;
436 }
437
438 policy->transition_ongoing = true;
439 policy->transition_task = current;
440
441 spin_unlock(&policy->transition_lock);
442
443 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
444}
445EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
446
447void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
448 struct cpufreq_freqs *freqs, int transition_failed)
449{
450 if (WARN_ON(!policy->transition_ongoing))
451 return;
452
453 cpufreq_notify_post_transition(policy, freqs, transition_failed);
454
455 arch_set_freq_scale(policy->related_cpus,
456 policy->cur,
457 arch_scale_freq_ref(policy->cpu));
458
459 spin_lock(&policy->transition_lock);
460 policy->transition_ongoing = false;
461 policy->transition_task = NULL;
462 spin_unlock(&policy->transition_lock);
463
464 wake_up(&policy->transition_wait);
465}
466EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
467
468/*
469 * Fast frequency switching status count. Positive means "enabled", negative
470 * means "disabled" and 0 means "not decided yet".
471 */
472static int cpufreq_fast_switch_count;
473static DEFINE_MUTEX(cpufreq_fast_switch_lock);
474
475static void cpufreq_list_transition_notifiers(void)
476{
477 struct notifier_block *nb;
478
479 pr_info("Registered transition notifiers:\n");
480
481 mutex_lock(&cpufreq_transition_notifier_list.mutex);
482
483 for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
484 pr_info("%pS\n", nb->notifier_call);
485
486 mutex_unlock(&cpufreq_transition_notifier_list.mutex);
487}
488
489/**
490 * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
491 * @policy: cpufreq policy to enable fast frequency switching for.
492 *
493 * Try to enable fast frequency switching for @policy.
494 *
495 * The attempt will fail if there is at least one transition notifier registered
496 * at this point, as fast frequency switching is quite fundamentally at odds
497 * with transition notifiers. Thus if successful, it will make registration of
498 * transition notifiers fail going forward.
499 */
500void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
501{
502 lockdep_assert_held(&policy->rwsem);
503
504 if (!policy->fast_switch_possible)
505 return;
506
507 mutex_lock(&cpufreq_fast_switch_lock);
508 if (cpufreq_fast_switch_count >= 0) {
509 cpufreq_fast_switch_count++;
510 policy->fast_switch_enabled = true;
511 } else {
512 pr_warn("CPU%u: Fast frequency switching not enabled\n",
513 policy->cpu);
514 cpufreq_list_transition_notifiers();
515 }
516 mutex_unlock(&cpufreq_fast_switch_lock);
517}
518EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
519
520/**
521 * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
522 * @policy: cpufreq policy to disable fast frequency switching for.
523 */
524void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
525{
526 mutex_lock(&cpufreq_fast_switch_lock);
527 if (policy->fast_switch_enabled) {
528 policy->fast_switch_enabled = false;
529 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
530 cpufreq_fast_switch_count--;
531 }
532 mutex_unlock(&cpufreq_fast_switch_lock);
533}
534EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
535
536static unsigned int __resolve_freq(struct cpufreq_policy *policy,
537 unsigned int target_freq, unsigned int relation)
538{
539 unsigned int idx;
540
541 target_freq = clamp_val(target_freq, policy->min, policy->max);
542
543 if (!policy->freq_table)
544 return target_freq;
545
546 idx = cpufreq_frequency_table_target(policy, target_freq, relation);
547 policy->cached_resolved_idx = idx;
548 policy->cached_target_freq = target_freq;
549 return policy->freq_table[idx].frequency;
550}
551
552/**
553 * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
554 * one.
555 * @policy: associated policy to interrogate
556 * @target_freq: target frequency to resolve.
557 *
558 * The target to driver frequency mapping is cached in the policy.
559 *
560 * Return: Lowest driver-supported frequency greater than or equal to the
561 * given target_freq, subject to policy (min/max) and driver limitations.
562 */
563unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
564 unsigned int target_freq)
565{
566 return __resolve_freq(policy, target_freq, CPUFREQ_RELATION_LE);
567}
568EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
569
570unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
571{
572 unsigned int latency;
573
574 if (policy->transition_delay_us)
575 return policy->transition_delay_us;
576
577 latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
578 if (latency) {
579 unsigned int max_delay_us = 2 * MSEC_PER_SEC;
580
581 /*
582 * If the platform already has high transition_latency, use it
583 * as-is.
584 */
585 if (latency > max_delay_us)
586 return latency;
587
588 /*
589 * For platforms that can change the frequency very fast (< 2
590 * us), the above formula gives a decent transition delay. But
591 * for platforms where transition_latency is in milliseconds, it
592 * ends up giving unrealistic values.
593 *
594 * Cap the default transition delay to 2 ms, which seems to be
595 * a reasonable amount of time after which we should reevaluate
596 * the frequency.
597 */
598 return min(latency * LATENCY_MULTIPLIER, max_delay_us);
599 }
600
601 return LATENCY_MULTIPLIER;
602}
603EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
604
605/*********************************************************************
606 * SYSFS INTERFACE *
607 *********************************************************************/
608static ssize_t show_boost(struct kobject *kobj,
609 struct kobj_attribute *attr, char *buf)
610{
611 return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
612}
613
614static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
615 const char *buf, size_t count)
616{
617 int ret, enable;
618
619 ret = sscanf(buf, "%d", &enable);
620 if (ret != 1 || enable < 0 || enable > 1)
621 return -EINVAL;
622
623 if (cpufreq_boost_trigger_state(enable)) {
624 pr_err("%s: Cannot %s BOOST!\n",
625 __func__, enable ? "enable" : "disable");
626 return -EINVAL;
627 }
628
629 pr_debug("%s: cpufreq BOOST %s\n",
630 __func__, enable ? "enabled" : "disabled");
631
632 return count;
633}
634define_one_global_rw(boost);
635
636static ssize_t show_local_boost(struct cpufreq_policy *policy, char *buf)
637{
638 return sysfs_emit(buf, "%d\n", policy->boost_enabled);
639}
640
641static ssize_t store_local_boost(struct cpufreq_policy *policy,
642 const char *buf, size_t count)
643{
644 int ret, enable;
645
646 ret = kstrtoint(buf, 10, &enable);
647 if (ret || enable < 0 || enable > 1)
648 return -EINVAL;
649
650 if (!cpufreq_driver->boost_enabled)
651 return -EINVAL;
652
653 if (policy->boost_enabled == enable)
654 return count;
655
656 policy->boost_enabled = enable;
657
658 cpus_read_lock();
659 ret = cpufreq_driver->set_boost(policy, enable);
660 cpus_read_unlock();
661
662 if (ret) {
663 policy->boost_enabled = !policy->boost_enabled;
664 return ret;
665 }
666
667 return count;
668}
669
670static struct freq_attr local_boost = __ATTR(boost, 0644, show_local_boost, store_local_boost);
671
672static struct cpufreq_governor *find_governor(const char *str_governor)
673{
674 struct cpufreq_governor *t;
675
676 for_each_governor(t)
677 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
678 return t;
679
680 return NULL;
681}
682
683static struct cpufreq_governor *get_governor(const char *str_governor)
684{
685 struct cpufreq_governor *t;
686
687 mutex_lock(&cpufreq_governor_mutex);
688 t = find_governor(str_governor);
689 if (!t)
690 goto unlock;
691
692 if (!try_module_get(t->owner))
693 t = NULL;
694
695unlock:
696 mutex_unlock(&cpufreq_governor_mutex);
697
698 return t;
699}
700
701static unsigned int cpufreq_parse_policy(char *str_governor)
702{
703 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
704 return CPUFREQ_POLICY_PERFORMANCE;
705
706 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
707 return CPUFREQ_POLICY_POWERSAVE;
708
709 return CPUFREQ_POLICY_UNKNOWN;
710}
711
712/**
713 * cpufreq_parse_governor - parse a governor string only for has_target()
714 * @str_governor: Governor name.
715 */
716static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
717{
718 struct cpufreq_governor *t;
719
720 t = get_governor(str_governor);
721 if (t)
722 return t;
723
724 if (request_module("cpufreq_%s", str_governor))
725 return NULL;
726
727 return get_governor(str_governor);
728}
729
730/*
731 * cpufreq_per_cpu_attr_read() / show_##file_name() -
732 * print out cpufreq information
733 *
734 * Write out information from cpufreq_driver->policy[cpu]; object must be
735 * "unsigned int".
736 */
737
738#define show_one(file_name, object) \
739static ssize_t show_##file_name \
740(struct cpufreq_policy *policy, char *buf) \
741{ \
742 return sprintf(buf, "%u\n", policy->object); \
743}
744
745show_one(cpuinfo_min_freq, cpuinfo.min_freq);
746show_one(cpuinfo_max_freq, cpuinfo.max_freq);
747show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
748show_one(scaling_min_freq, min);
749show_one(scaling_max_freq, max);
750
751__weak unsigned int arch_freq_get_on_cpu(int cpu)
752{
753 return 0;
754}
755
756static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
757{
758 ssize_t ret;
759 unsigned int freq;
760
761 freq = arch_freq_get_on_cpu(policy->cpu);
762 if (freq)
763 ret = sprintf(buf, "%u\n", freq);
764 else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
765 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
766 else
767 ret = sprintf(buf, "%u\n", policy->cur);
768 return ret;
769}
770
771/*
772 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
773 */
774#define store_one(file_name, object) \
775static ssize_t store_##file_name \
776(struct cpufreq_policy *policy, const char *buf, size_t count) \
777{ \
778 unsigned long val; \
779 int ret; \
780 \
781 ret = kstrtoul(buf, 0, &val); \
782 if (ret) \
783 return ret; \
784 \
785 ret = freq_qos_update_request(policy->object##_freq_req, val);\
786 return ret >= 0 ? count : ret; \
787}
788
789store_one(scaling_min_freq, min);
790store_one(scaling_max_freq, max);
791
792/*
793 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
794 */
795static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
796 char *buf)
797{
798 unsigned int cur_freq = __cpufreq_get(policy);
799
800 if (cur_freq)
801 return sprintf(buf, "%u\n", cur_freq);
802
803 return sprintf(buf, "<unknown>\n");
804}
805
806/*
807 * show_scaling_governor - show the current policy for the specified CPU
808 */
809static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
810{
811 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
812 return sprintf(buf, "powersave\n");
813 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
814 return sprintf(buf, "performance\n");
815 else if (policy->governor)
816 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
817 policy->governor->name);
818 return -EINVAL;
819}
820
821/*
822 * store_scaling_governor - store policy for the specified CPU
823 */
824static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
825 const char *buf, size_t count)
826{
827 char str_governor[16];
828 int ret;
829
830 ret = sscanf(buf, "%15s", str_governor);
831 if (ret != 1)
832 return -EINVAL;
833
834 if (cpufreq_driver->setpolicy) {
835 unsigned int new_pol;
836
837 new_pol = cpufreq_parse_policy(str_governor);
838 if (!new_pol)
839 return -EINVAL;
840
841 ret = cpufreq_set_policy(policy, NULL, new_pol);
842 } else {
843 struct cpufreq_governor *new_gov;
844
845 new_gov = cpufreq_parse_governor(str_governor);
846 if (!new_gov)
847 return -EINVAL;
848
849 ret = cpufreq_set_policy(policy, new_gov,
850 CPUFREQ_POLICY_UNKNOWN);
851
852 module_put(new_gov->owner);
853 }
854
855 return ret ? ret : count;
856}
857
858/*
859 * show_scaling_driver - show the cpufreq driver currently loaded
860 */
861static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
862{
863 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
864}
865
866/*
867 * show_scaling_available_governors - show the available CPUfreq governors
868 */
869static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
870 char *buf)
871{
872 ssize_t i = 0;
873 struct cpufreq_governor *t;
874
875 if (!has_target()) {
876 i += sprintf(buf, "performance powersave");
877 goto out;
878 }
879
880 mutex_lock(&cpufreq_governor_mutex);
881 for_each_governor(t) {
882 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
883 - (CPUFREQ_NAME_LEN + 2)))
884 break;
885 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
886 }
887 mutex_unlock(&cpufreq_governor_mutex);
888out:
889 i += sprintf(&buf[i], "\n");
890 return i;
891}
892
893ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
894{
895 ssize_t i = 0;
896 unsigned int cpu;
897
898 for_each_cpu(cpu, mask) {
899 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u ", cpu);
900 if (i >= (PAGE_SIZE - 5))
901 break;
902 }
903
904 /* Remove the extra space at the end */
905 i--;
906
907 i += sprintf(&buf[i], "\n");
908 return i;
909}
910EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
911
912/*
913 * show_related_cpus - show the CPUs affected by each transition even if
914 * hw coordination is in use
915 */
916static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
917{
918 return cpufreq_show_cpus(policy->related_cpus, buf);
919}
920
921/*
922 * show_affected_cpus - show the CPUs affected by each transition
923 */
924static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
925{
926 return cpufreq_show_cpus(policy->cpus, buf);
927}
928
929static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
930 const char *buf, size_t count)
931{
932 unsigned int freq = 0;
933 unsigned int ret;
934
935 if (!policy->governor || !policy->governor->store_setspeed)
936 return -EINVAL;
937
938 ret = sscanf(buf, "%u", &freq);
939 if (ret != 1)
940 return -EINVAL;
941
942 policy->governor->store_setspeed(policy, freq);
943
944 return count;
945}
946
947static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
948{
949 if (!policy->governor || !policy->governor->show_setspeed)
950 return sprintf(buf, "<unsupported>\n");
951
952 return policy->governor->show_setspeed(policy, buf);
953}
954
955/*
956 * show_bios_limit - show the current cpufreq HW/BIOS limitation
957 */
958static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
959{
960 unsigned int limit;
961 int ret;
962 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
963 if (!ret)
964 return sprintf(buf, "%u\n", limit);
965 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
966}
967
968cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
969cpufreq_freq_attr_ro(cpuinfo_min_freq);
970cpufreq_freq_attr_ro(cpuinfo_max_freq);
971cpufreq_freq_attr_ro(cpuinfo_transition_latency);
972cpufreq_freq_attr_ro(scaling_available_governors);
973cpufreq_freq_attr_ro(scaling_driver);
974cpufreq_freq_attr_ro(scaling_cur_freq);
975cpufreq_freq_attr_ro(bios_limit);
976cpufreq_freq_attr_ro(related_cpus);
977cpufreq_freq_attr_ro(affected_cpus);
978cpufreq_freq_attr_rw(scaling_min_freq);
979cpufreq_freq_attr_rw(scaling_max_freq);
980cpufreq_freq_attr_rw(scaling_governor);
981cpufreq_freq_attr_rw(scaling_setspeed);
982
983static struct attribute *cpufreq_attrs[] = {
984 &cpuinfo_min_freq.attr,
985 &cpuinfo_max_freq.attr,
986 &cpuinfo_transition_latency.attr,
987 &scaling_min_freq.attr,
988 &scaling_max_freq.attr,
989 &affected_cpus.attr,
990 &related_cpus.attr,
991 &scaling_governor.attr,
992 &scaling_driver.attr,
993 &scaling_available_governors.attr,
994 &scaling_setspeed.attr,
995 NULL
996};
997ATTRIBUTE_GROUPS(cpufreq);
998
999#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
1000#define to_attr(a) container_of(a, struct freq_attr, attr)
1001
1002static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1003{
1004 struct cpufreq_policy *policy = to_policy(kobj);
1005 struct freq_attr *fattr = to_attr(attr);
1006 ssize_t ret = -EBUSY;
1007
1008 if (!fattr->show)
1009 return -EIO;
1010
1011 down_read(&policy->rwsem);
1012 if (likely(!policy_is_inactive(policy)))
1013 ret = fattr->show(policy, buf);
1014 up_read(&policy->rwsem);
1015
1016 return ret;
1017}
1018
1019static ssize_t store(struct kobject *kobj, struct attribute *attr,
1020 const char *buf, size_t count)
1021{
1022 struct cpufreq_policy *policy = to_policy(kobj);
1023 struct freq_attr *fattr = to_attr(attr);
1024 ssize_t ret = -EBUSY;
1025
1026 if (!fattr->store)
1027 return -EIO;
1028
1029 down_write(&policy->rwsem);
1030 if (likely(!policy_is_inactive(policy)))
1031 ret = fattr->store(policy, buf, count);
1032 up_write(&policy->rwsem);
1033
1034 return ret;
1035}
1036
1037static void cpufreq_sysfs_release(struct kobject *kobj)
1038{
1039 struct cpufreq_policy *policy = to_policy(kobj);
1040 pr_debug("last reference is dropped\n");
1041 complete(&policy->kobj_unregister);
1042}
1043
1044static const struct sysfs_ops sysfs_ops = {
1045 .show = show,
1046 .store = store,
1047};
1048
1049static const struct kobj_type ktype_cpufreq = {
1050 .sysfs_ops = &sysfs_ops,
1051 .default_groups = cpufreq_groups,
1052 .release = cpufreq_sysfs_release,
1053};
1054
1055static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
1056 struct device *dev)
1057{
1058 if (unlikely(!dev))
1059 return;
1060
1061 if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1062 return;
1063
1064 dev_dbg(dev, "%s: Adding symlink\n", __func__);
1065 if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1066 dev_err(dev, "cpufreq symlink creation failed\n");
1067}
1068
1069static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu,
1070 struct device *dev)
1071{
1072 dev_dbg(dev, "%s: Removing symlink\n", __func__);
1073 sysfs_remove_link(&dev->kobj, "cpufreq");
1074 cpumask_clear_cpu(cpu, policy->real_cpus);
1075}
1076
1077static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1078{
1079 struct freq_attr **drv_attr;
1080 int ret = 0;
1081
1082 /* set up files for this cpu device */
1083 drv_attr = cpufreq_driver->attr;
1084 while (drv_attr && *drv_attr) {
1085 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1086 if (ret)
1087 return ret;
1088 drv_attr++;
1089 }
1090 if (cpufreq_driver->get) {
1091 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1092 if (ret)
1093 return ret;
1094 }
1095
1096 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1097 if (ret)
1098 return ret;
1099
1100 if (cpufreq_driver->bios_limit) {
1101 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1102 if (ret)
1103 return ret;
1104 }
1105
1106 if (cpufreq_boost_supported()) {
1107 ret = sysfs_create_file(&policy->kobj, &local_boost.attr);
1108 if (ret)
1109 return ret;
1110 }
1111
1112 return 0;
1113}
1114
1115static int cpufreq_init_policy(struct cpufreq_policy *policy)
1116{
1117 struct cpufreq_governor *gov = NULL;
1118 unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1119 int ret;
1120
1121 if (has_target()) {
1122 /* Update policy governor to the one used before hotplug. */
1123 gov = get_governor(policy->last_governor);
1124 if (gov) {
1125 pr_debug("Restoring governor %s for cpu %d\n",
1126 gov->name, policy->cpu);
1127 } else {
1128 gov = get_governor(default_governor);
1129 }
1130
1131 if (!gov) {
1132 gov = cpufreq_default_governor();
1133 __module_get(gov->owner);
1134 }
1135
1136 } else {
1137
1138 /* Use the default policy if there is no last_policy. */
1139 if (policy->last_policy) {
1140 pol = policy->last_policy;
1141 } else {
1142 pol = cpufreq_parse_policy(default_governor);
1143 /*
1144 * In case the default governor is neither "performance"
1145 * nor "powersave", fall back to the initial policy
1146 * value set by the driver.
1147 */
1148 if (pol == CPUFREQ_POLICY_UNKNOWN)
1149 pol = policy->policy;
1150 }
1151 if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1152 pol != CPUFREQ_POLICY_POWERSAVE)
1153 return -ENODATA;
1154 }
1155
1156 ret = cpufreq_set_policy(policy, gov, pol);
1157 if (gov)
1158 module_put(gov->owner);
1159
1160 return ret;
1161}
1162
1163static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1164{
1165 int ret = 0;
1166
1167 /* Has this CPU been taken care of already? */
1168 if (cpumask_test_cpu(cpu, policy->cpus))
1169 return 0;
1170
1171 down_write(&policy->rwsem);
1172 if (has_target())
1173 cpufreq_stop_governor(policy);
1174
1175 cpumask_set_cpu(cpu, policy->cpus);
1176
1177 if (has_target()) {
1178 ret = cpufreq_start_governor(policy);
1179 if (ret)
1180 pr_err("%s: Failed to start governor\n", __func__);
1181 }
1182 up_write(&policy->rwsem);
1183 return ret;
1184}
1185
1186void refresh_frequency_limits(struct cpufreq_policy *policy)
1187{
1188 if (!policy_is_inactive(policy)) {
1189 pr_debug("updating policy for CPU %u\n", policy->cpu);
1190
1191 cpufreq_set_policy(policy, policy->governor, policy->policy);
1192 }
1193}
1194EXPORT_SYMBOL(refresh_frequency_limits);
1195
1196static void handle_update(struct work_struct *work)
1197{
1198 struct cpufreq_policy *policy =
1199 container_of(work, struct cpufreq_policy, update);
1200
1201 pr_debug("handle_update for cpu %u called\n", policy->cpu);
1202 down_write(&policy->rwsem);
1203 refresh_frequency_limits(policy);
1204 up_write(&policy->rwsem);
1205}
1206
1207static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1208 void *data)
1209{
1210 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1211
1212 schedule_work(&policy->update);
1213 return 0;
1214}
1215
1216static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1217 void *data)
1218{
1219 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1220
1221 schedule_work(&policy->update);
1222 return 0;
1223}
1224
1225static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1226{
1227 struct kobject *kobj;
1228 struct completion *cmp;
1229
1230 down_write(&policy->rwsem);
1231 cpufreq_stats_free_table(policy);
1232 kobj = &policy->kobj;
1233 cmp = &policy->kobj_unregister;
1234 up_write(&policy->rwsem);
1235 kobject_put(kobj);
1236
1237 /*
1238 * We need to make sure that the underlying kobj is
1239 * actually not referenced anymore by anybody before we
1240 * proceed with unloading.
1241 */
1242 pr_debug("waiting for dropping of refcount\n");
1243 wait_for_completion(cmp);
1244 pr_debug("wait complete\n");
1245}
1246
1247static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1248{
1249 struct cpufreq_policy *policy;
1250 struct device *dev = get_cpu_device(cpu);
1251 int ret;
1252
1253 if (!dev)
1254 return NULL;
1255
1256 policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1257 if (!policy)
1258 return NULL;
1259
1260 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1261 goto err_free_policy;
1262
1263 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1264 goto err_free_cpumask;
1265
1266 if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1267 goto err_free_rcpumask;
1268
1269 init_completion(&policy->kobj_unregister);
1270 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1271 cpufreq_global_kobject, "policy%u", cpu);
1272 if (ret) {
1273 dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1274 /*
1275 * The entire policy object will be freed below, but the extra
1276 * memory allocated for the kobject name needs to be freed by
1277 * releasing the kobject.
1278 */
1279 kobject_put(&policy->kobj);
1280 goto err_free_real_cpus;
1281 }
1282
1283 freq_constraints_init(&policy->constraints);
1284
1285 policy->nb_min.notifier_call = cpufreq_notifier_min;
1286 policy->nb_max.notifier_call = cpufreq_notifier_max;
1287
1288 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1289 &policy->nb_min);
1290 if (ret) {
1291 dev_err(dev, "Failed to register MIN QoS notifier: %d (CPU%u)\n",
1292 ret, cpu);
1293 goto err_kobj_remove;
1294 }
1295
1296 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1297 &policy->nb_max);
1298 if (ret) {
1299 dev_err(dev, "Failed to register MAX QoS notifier: %d (CPU%u)\n",
1300 ret, cpu);
1301 goto err_min_qos_notifier;
1302 }
1303
1304 INIT_LIST_HEAD(&policy->policy_list);
1305 init_rwsem(&policy->rwsem);
1306 spin_lock_init(&policy->transition_lock);
1307 init_waitqueue_head(&policy->transition_wait);
1308 INIT_WORK(&policy->update, handle_update);
1309
1310 policy->cpu = cpu;
1311 return policy;
1312
1313err_min_qos_notifier:
1314 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1315 &policy->nb_min);
1316err_kobj_remove:
1317 cpufreq_policy_put_kobj(policy);
1318err_free_real_cpus:
1319 free_cpumask_var(policy->real_cpus);
1320err_free_rcpumask:
1321 free_cpumask_var(policy->related_cpus);
1322err_free_cpumask:
1323 free_cpumask_var(policy->cpus);
1324err_free_policy:
1325 kfree(policy);
1326
1327 return NULL;
1328}
1329
1330static void cpufreq_policy_free(struct cpufreq_policy *policy)
1331{
1332 unsigned long flags;
1333 int cpu;
1334
1335 /*
1336 * The callers must ensure the policy is inactive by now, to avoid any
1337 * races with show()/store() callbacks.
1338 */
1339 if (unlikely(!policy_is_inactive(policy)))
1340 pr_warn("%s: Freeing active policy\n", __func__);
1341
1342 /* Remove policy from list */
1343 write_lock_irqsave(&cpufreq_driver_lock, flags);
1344 list_del(&policy->policy_list);
1345
1346 for_each_cpu(cpu, policy->related_cpus)
1347 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1348 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1349
1350 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1351 &policy->nb_max);
1352 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1353 &policy->nb_min);
1354
1355 /* Cancel any pending policy->update work before freeing the policy. */
1356 cancel_work_sync(&policy->update);
1357
1358 if (policy->max_freq_req) {
1359 /*
1360 * Remove max_freq_req after sending CPUFREQ_REMOVE_POLICY
1361 * notification, since CPUFREQ_CREATE_POLICY notification was
1362 * sent after adding max_freq_req earlier.
1363 */
1364 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1365 CPUFREQ_REMOVE_POLICY, policy);
1366 freq_qos_remove_request(policy->max_freq_req);
1367 }
1368
1369 freq_qos_remove_request(policy->min_freq_req);
1370 kfree(policy->min_freq_req);
1371
1372 cpufreq_policy_put_kobj(policy);
1373 free_cpumask_var(policy->real_cpus);
1374 free_cpumask_var(policy->related_cpus);
1375 free_cpumask_var(policy->cpus);
1376 kfree(policy);
1377}
1378
1379static int cpufreq_online(unsigned int cpu)
1380{
1381 struct cpufreq_policy *policy;
1382 bool new_policy;
1383 unsigned long flags;
1384 unsigned int j;
1385 int ret;
1386
1387 pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1388
1389 /* Check if this CPU already has a policy to manage it */
1390 policy = per_cpu(cpufreq_cpu_data, cpu);
1391 if (policy) {
1392 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1393 if (!policy_is_inactive(policy))
1394 return cpufreq_add_policy_cpu(policy, cpu);
1395
1396 /* This is the only online CPU for the policy. Start over. */
1397 new_policy = false;
1398 down_write(&policy->rwsem);
1399 policy->cpu = cpu;
1400 policy->governor = NULL;
1401 } else {
1402 new_policy = true;
1403 policy = cpufreq_policy_alloc(cpu);
1404 if (!policy)
1405 return -ENOMEM;
1406 down_write(&policy->rwsem);
1407 }
1408
1409 if (!new_policy && cpufreq_driver->online) {
1410 /* Recover policy->cpus using related_cpus */
1411 cpumask_copy(policy->cpus, policy->related_cpus);
1412
1413 ret = cpufreq_driver->online(policy);
1414 if (ret) {
1415 pr_debug("%s: %d: initialization failed\n", __func__,
1416 __LINE__);
1417 goto out_exit_policy;
1418 }
1419 } else {
1420 cpumask_copy(policy->cpus, cpumask_of(cpu));
1421
1422 /*
1423 * Call driver. From then on the cpufreq must be able
1424 * to accept all calls to ->verify and ->setpolicy for this CPU.
1425 */
1426 ret = cpufreq_driver->init(policy);
1427 if (ret) {
1428 pr_debug("%s: %d: initialization failed\n", __func__,
1429 __LINE__);
1430 goto out_free_policy;
1431 }
1432
1433 /* Let the per-policy boost flag mirror the cpufreq_driver boost during init */
1434 policy->boost_enabled = cpufreq_boost_enabled() && policy_has_boost_freq(policy);
1435
1436 /*
1437 * The initialization has succeeded and the policy is online.
1438 * If there is a problem with its frequency table, take it
1439 * offline and drop it.
1440 */
1441 ret = cpufreq_table_validate_and_sort(policy);
1442 if (ret)
1443 goto out_offline_policy;
1444
1445 /* related_cpus should at least include policy->cpus. */
1446 cpumask_copy(policy->related_cpus, policy->cpus);
1447 }
1448
1449 /*
1450 * affected cpus must always be the one, which are online. We aren't
1451 * managing offline cpus here.
1452 */
1453 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1454
1455 if (new_policy) {
1456 for_each_cpu(j, policy->related_cpus) {
1457 per_cpu(cpufreq_cpu_data, j) = policy;
1458 add_cpu_dev_symlink(policy, j, get_cpu_device(j));
1459 }
1460
1461 policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1462 GFP_KERNEL);
1463 if (!policy->min_freq_req) {
1464 ret = -ENOMEM;
1465 goto out_destroy_policy;
1466 }
1467
1468 ret = freq_qos_add_request(&policy->constraints,
1469 policy->min_freq_req, FREQ_QOS_MIN,
1470 FREQ_QOS_MIN_DEFAULT_VALUE);
1471 if (ret < 0) {
1472 /*
1473 * So we don't call freq_qos_remove_request() for an
1474 * uninitialized request.
1475 */
1476 kfree(policy->min_freq_req);
1477 policy->min_freq_req = NULL;
1478 goto out_destroy_policy;
1479 }
1480
1481 /*
1482 * This must be initialized right here to avoid calling
1483 * freq_qos_remove_request() on uninitialized request in case
1484 * of errors.
1485 */
1486 policy->max_freq_req = policy->min_freq_req + 1;
1487
1488 ret = freq_qos_add_request(&policy->constraints,
1489 policy->max_freq_req, FREQ_QOS_MAX,
1490 FREQ_QOS_MAX_DEFAULT_VALUE);
1491 if (ret < 0) {
1492 policy->max_freq_req = NULL;
1493 goto out_destroy_policy;
1494 }
1495
1496 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1497 CPUFREQ_CREATE_POLICY, policy);
1498 }
1499
1500 if (cpufreq_driver->get && has_target()) {
1501 policy->cur = cpufreq_driver->get(policy->cpu);
1502 if (!policy->cur) {
1503 ret = -EIO;
1504 pr_err("%s: ->get() failed\n", __func__);
1505 goto out_destroy_policy;
1506 }
1507 }
1508
1509 /*
1510 * Sometimes boot loaders set CPU frequency to a value outside of
1511 * frequency table present with cpufreq core. In such cases CPU might be
1512 * unstable if it has to run on that frequency for long duration of time
1513 * and so its better to set it to a frequency which is specified in
1514 * freq-table. This also makes cpufreq stats inconsistent as
1515 * cpufreq-stats would fail to register because current frequency of CPU
1516 * isn't found in freq-table.
1517 *
1518 * Because we don't want this change to effect boot process badly, we go
1519 * for the next freq which is >= policy->cur ('cur' must be set by now,
1520 * otherwise we will end up setting freq to lowest of the table as 'cur'
1521 * is initialized to zero).
1522 *
1523 * We are passing target-freq as "policy->cur - 1" otherwise
1524 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1525 * equal to target-freq.
1526 */
1527 if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1528 && has_target()) {
1529 unsigned int old_freq = policy->cur;
1530
1531 /* Are we running at unknown frequency ? */
1532 ret = cpufreq_frequency_table_get_index(policy, old_freq);
1533 if (ret == -EINVAL) {
1534 ret = __cpufreq_driver_target(policy, old_freq - 1,
1535 CPUFREQ_RELATION_L);
1536
1537 /*
1538 * Reaching here after boot in a few seconds may not
1539 * mean that system will remain stable at "unknown"
1540 * frequency for longer duration. Hence, a BUG_ON().
1541 */
1542 BUG_ON(ret);
1543 pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
1544 __func__, policy->cpu, old_freq, policy->cur);
1545 }
1546 }
1547
1548 if (new_policy) {
1549 ret = cpufreq_add_dev_interface(policy);
1550 if (ret)
1551 goto out_destroy_policy;
1552
1553 cpufreq_stats_create_table(policy);
1554
1555 write_lock_irqsave(&cpufreq_driver_lock, flags);
1556 list_add(&policy->policy_list, &cpufreq_policy_list);
1557 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1558
1559 /*
1560 * Register with the energy model before
1561 * sugov_eas_rebuild_sd() is called, which will result
1562 * in rebuilding of the sched domains, which should only be done
1563 * once the energy model is properly initialized for the policy
1564 * first.
1565 *
1566 * Also, this should be called before the policy is registered
1567 * with cooling framework.
1568 */
1569 if (cpufreq_driver->register_em)
1570 cpufreq_driver->register_em(policy);
1571 }
1572
1573 ret = cpufreq_init_policy(policy);
1574 if (ret) {
1575 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1576 __func__, cpu, ret);
1577 goto out_destroy_policy;
1578 }
1579
1580 up_write(&policy->rwsem);
1581
1582 kobject_uevent(&policy->kobj, KOBJ_ADD);
1583
1584 /* Callback for handling stuff after policy is ready */
1585 if (cpufreq_driver->ready)
1586 cpufreq_driver->ready(policy);
1587
1588 /* Register cpufreq cooling only for a new policy */
1589 if (new_policy && cpufreq_thermal_control_enabled(cpufreq_driver))
1590 policy->cdev = of_cpufreq_cooling_register(policy);
1591
1592 pr_debug("initialization complete\n");
1593
1594 return 0;
1595
1596out_destroy_policy:
1597 for_each_cpu(j, policy->real_cpus)
1598 remove_cpu_dev_symlink(policy, j, get_cpu_device(j));
1599
1600out_offline_policy:
1601 if (cpufreq_driver->offline)
1602 cpufreq_driver->offline(policy);
1603
1604out_exit_policy:
1605 if (cpufreq_driver->exit)
1606 cpufreq_driver->exit(policy);
1607
1608out_free_policy:
1609 cpumask_clear(policy->cpus);
1610 up_write(&policy->rwsem);
1611
1612 cpufreq_policy_free(policy);
1613 return ret;
1614}
1615
1616/**
1617 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1618 * @dev: CPU device.
1619 * @sif: Subsystem interface structure pointer (not used)
1620 */
1621static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1622{
1623 struct cpufreq_policy *policy;
1624 unsigned cpu = dev->id;
1625 int ret;
1626
1627 dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1628
1629 if (cpu_online(cpu)) {
1630 ret = cpufreq_online(cpu);
1631 if (ret)
1632 return ret;
1633 }
1634
1635 /* Create sysfs link on CPU registration */
1636 policy = per_cpu(cpufreq_cpu_data, cpu);
1637 if (policy)
1638 add_cpu_dev_symlink(policy, cpu, dev);
1639
1640 return 0;
1641}
1642
1643static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1644{
1645 int ret;
1646
1647 if (has_target())
1648 cpufreq_stop_governor(policy);
1649
1650 cpumask_clear_cpu(cpu, policy->cpus);
1651
1652 if (!policy_is_inactive(policy)) {
1653 /* Nominate a new CPU if necessary. */
1654 if (cpu == policy->cpu)
1655 policy->cpu = cpumask_any(policy->cpus);
1656
1657 /* Start the governor again for the active policy. */
1658 if (has_target()) {
1659 ret = cpufreq_start_governor(policy);
1660 if (ret)
1661 pr_err("%s: Failed to start governor\n", __func__);
1662 }
1663
1664 return;
1665 }
1666
1667 if (has_target())
1668 strscpy(policy->last_governor, policy->governor->name,
1669 CPUFREQ_NAME_LEN);
1670 else
1671 policy->last_policy = policy->policy;
1672
1673 if (has_target())
1674 cpufreq_exit_governor(policy);
1675
1676 /*
1677 * Perform the ->offline() during light-weight tear-down, as
1678 * that allows fast recovery when the CPU comes back.
1679 */
1680 if (cpufreq_driver->offline) {
1681 cpufreq_driver->offline(policy);
1682 return;
1683 }
1684
1685 if (cpufreq_driver->exit)
1686 cpufreq_driver->exit(policy);
1687
1688 policy->freq_table = NULL;
1689}
1690
1691static int cpufreq_offline(unsigned int cpu)
1692{
1693 struct cpufreq_policy *policy;
1694
1695 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1696
1697 policy = cpufreq_cpu_get_raw(cpu);
1698 if (!policy) {
1699 pr_debug("%s: No cpu_data found\n", __func__);
1700 return 0;
1701 }
1702
1703 down_write(&policy->rwsem);
1704
1705 __cpufreq_offline(cpu, policy);
1706
1707 up_write(&policy->rwsem);
1708 return 0;
1709}
1710
1711/*
1712 * cpufreq_remove_dev - remove a CPU device
1713 *
1714 * Removes the cpufreq interface for a CPU device.
1715 */
1716static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1717{
1718 unsigned int cpu = dev->id;
1719 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1720
1721 if (!policy)
1722 return;
1723
1724 down_write(&policy->rwsem);
1725
1726 if (cpu_online(cpu))
1727 __cpufreq_offline(cpu, policy);
1728
1729 remove_cpu_dev_symlink(policy, cpu, dev);
1730
1731 if (!cpumask_empty(policy->real_cpus)) {
1732 up_write(&policy->rwsem);
1733 return;
1734 }
1735
1736 /*
1737 * Unregister cpufreq cooling once all the CPUs of the policy are
1738 * removed.
1739 */
1740 if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1741 cpufreq_cooling_unregister(policy->cdev);
1742 policy->cdev = NULL;
1743 }
1744
1745 /* We did light-weight exit earlier, do full tear down now */
1746 if (cpufreq_driver->offline && cpufreq_driver->exit)
1747 cpufreq_driver->exit(policy);
1748
1749 up_write(&policy->rwsem);
1750
1751 cpufreq_policy_free(policy);
1752}
1753
1754/**
1755 * cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1756 * @policy: Policy managing CPUs.
1757 * @new_freq: New CPU frequency.
1758 *
1759 * Adjust to the current frequency first and clean up later by either calling
1760 * cpufreq_update_policy(), or scheduling handle_update().
1761 */
1762static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1763 unsigned int new_freq)
1764{
1765 struct cpufreq_freqs freqs;
1766
1767 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1768 policy->cur, new_freq);
1769
1770 freqs.old = policy->cur;
1771 freqs.new = new_freq;
1772
1773 cpufreq_freq_transition_begin(policy, &freqs);
1774 cpufreq_freq_transition_end(policy, &freqs, 0);
1775}
1776
1777static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1778{
1779 unsigned int new_freq;
1780
1781 new_freq = cpufreq_driver->get(policy->cpu);
1782 if (!new_freq)
1783 return 0;
1784
1785 /*
1786 * If fast frequency switching is used with the given policy, the check
1787 * against policy->cur is pointless, so skip it in that case.
1788 */
1789 if (policy->fast_switch_enabled || !has_target())
1790 return new_freq;
1791
1792 if (policy->cur != new_freq) {
1793 /*
1794 * For some platforms, the frequency returned by hardware may be
1795 * slightly different from what is provided in the frequency
1796 * table, for example hardware may return 499 MHz instead of 500
1797 * MHz. In such cases it is better to avoid getting into
1798 * unnecessary frequency updates.
1799 */
1800 if (abs(policy->cur - new_freq) < KHZ_PER_MHZ)
1801 return policy->cur;
1802
1803 cpufreq_out_of_sync(policy, new_freq);
1804 if (update)
1805 schedule_work(&policy->update);
1806 }
1807
1808 return new_freq;
1809}
1810
1811/**
1812 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1813 * @cpu: CPU number
1814 *
1815 * This is the last known freq, without actually getting it from the driver.
1816 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1817 */
1818unsigned int cpufreq_quick_get(unsigned int cpu)
1819{
1820 struct cpufreq_policy *policy;
1821 unsigned int ret_freq = 0;
1822 unsigned long flags;
1823
1824 read_lock_irqsave(&cpufreq_driver_lock, flags);
1825
1826 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1827 ret_freq = cpufreq_driver->get(cpu);
1828 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1829 return ret_freq;
1830 }
1831
1832 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1833
1834 policy = cpufreq_cpu_get(cpu);
1835 if (policy) {
1836 ret_freq = policy->cur;
1837 cpufreq_cpu_put(policy);
1838 }
1839
1840 return ret_freq;
1841}
1842EXPORT_SYMBOL(cpufreq_quick_get);
1843
1844/**
1845 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1846 * @cpu: CPU number
1847 *
1848 * Just return the max possible frequency for a given CPU.
1849 */
1850unsigned int cpufreq_quick_get_max(unsigned int cpu)
1851{
1852 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1853 unsigned int ret_freq = 0;
1854
1855 if (policy) {
1856 ret_freq = policy->max;
1857 cpufreq_cpu_put(policy);
1858 }
1859
1860 return ret_freq;
1861}
1862EXPORT_SYMBOL(cpufreq_quick_get_max);
1863
1864/**
1865 * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1866 * @cpu: CPU number
1867 *
1868 * The default return value is the max_freq field of cpuinfo.
1869 */
1870__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1871{
1872 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1873 unsigned int ret_freq = 0;
1874
1875 if (policy) {
1876 ret_freq = policy->cpuinfo.max_freq;
1877 cpufreq_cpu_put(policy);
1878 }
1879
1880 return ret_freq;
1881}
1882EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1883
1884static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1885{
1886 if (unlikely(policy_is_inactive(policy)))
1887 return 0;
1888
1889 return cpufreq_verify_current_freq(policy, true);
1890}
1891
1892/**
1893 * cpufreq_get - get the current CPU frequency (in kHz)
1894 * @cpu: CPU number
1895 *
1896 * Get the CPU current (static) CPU frequency
1897 */
1898unsigned int cpufreq_get(unsigned int cpu)
1899{
1900 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1901 unsigned int ret_freq = 0;
1902
1903 if (policy) {
1904 down_read(&policy->rwsem);
1905 if (cpufreq_driver->get)
1906 ret_freq = __cpufreq_get(policy);
1907 up_read(&policy->rwsem);
1908
1909 cpufreq_cpu_put(policy);
1910 }
1911
1912 return ret_freq;
1913}
1914EXPORT_SYMBOL(cpufreq_get);
1915
1916static struct subsys_interface cpufreq_interface = {
1917 .name = "cpufreq",
1918 .subsys = &cpu_subsys,
1919 .add_dev = cpufreq_add_dev,
1920 .remove_dev = cpufreq_remove_dev,
1921};
1922
1923/*
1924 * In case platform wants some specific frequency to be configured
1925 * during suspend..
1926 */
1927int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1928{
1929 int ret;
1930
1931 if (!policy->suspend_freq) {
1932 pr_debug("%s: suspend_freq not defined\n", __func__);
1933 return 0;
1934 }
1935
1936 pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1937 policy->suspend_freq);
1938
1939 ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1940 CPUFREQ_RELATION_H);
1941 if (ret)
1942 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1943 __func__, policy->suspend_freq, ret);
1944
1945 return ret;
1946}
1947EXPORT_SYMBOL(cpufreq_generic_suspend);
1948
1949/**
1950 * cpufreq_suspend() - Suspend CPUFreq governors.
1951 *
1952 * Called during system wide Suspend/Hibernate cycles for suspending governors
1953 * as some platforms can't change frequency after this point in suspend cycle.
1954 * Because some of the devices (like: i2c, regulators, etc) they use for
1955 * changing frequency are suspended quickly after this point.
1956 */
1957void cpufreq_suspend(void)
1958{
1959 struct cpufreq_policy *policy;
1960
1961 if (!cpufreq_driver)
1962 return;
1963
1964 if (!has_target() && !cpufreq_driver->suspend)
1965 goto suspend;
1966
1967 pr_debug("%s: Suspending Governors\n", __func__);
1968
1969 for_each_active_policy(policy) {
1970 if (has_target()) {
1971 down_write(&policy->rwsem);
1972 cpufreq_stop_governor(policy);
1973 up_write(&policy->rwsem);
1974 }
1975
1976 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1977 pr_err("%s: Failed to suspend driver: %s\n", __func__,
1978 cpufreq_driver->name);
1979 }
1980
1981suspend:
1982 cpufreq_suspended = true;
1983}
1984
1985/**
1986 * cpufreq_resume() - Resume CPUFreq governors.
1987 *
1988 * Called during system wide Suspend/Hibernate cycle for resuming governors that
1989 * are suspended with cpufreq_suspend().
1990 */
1991void cpufreq_resume(void)
1992{
1993 struct cpufreq_policy *policy;
1994 int ret;
1995
1996 if (!cpufreq_driver)
1997 return;
1998
1999 if (unlikely(!cpufreq_suspended))
2000 return;
2001
2002 cpufreq_suspended = false;
2003
2004 if (!has_target() && !cpufreq_driver->resume)
2005 return;
2006
2007 pr_debug("%s: Resuming Governors\n", __func__);
2008
2009 for_each_active_policy(policy) {
2010 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
2011 pr_err("%s: Failed to resume driver: %s\n", __func__,
2012 cpufreq_driver->name);
2013 } else if (has_target()) {
2014 down_write(&policy->rwsem);
2015 ret = cpufreq_start_governor(policy);
2016 up_write(&policy->rwsem);
2017
2018 if (ret)
2019 pr_err("%s: Failed to start governor for CPU%u's policy\n",
2020 __func__, policy->cpu);
2021 }
2022 }
2023}
2024
2025/**
2026 * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
2027 * @flags: Flags to test against the current cpufreq driver's flags.
2028 *
2029 * Assumes that the driver is there, so callers must ensure that this is the
2030 * case.
2031 */
2032bool cpufreq_driver_test_flags(u16 flags)
2033{
2034 return !!(cpufreq_driver->flags & flags);
2035}
2036
2037/**
2038 * cpufreq_get_current_driver - Return the current driver's name.
2039 *
2040 * Return the name string of the currently registered cpufreq driver or NULL if
2041 * none.
2042 */
2043const char *cpufreq_get_current_driver(void)
2044{
2045 if (cpufreq_driver)
2046 return cpufreq_driver->name;
2047
2048 return NULL;
2049}
2050EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
2051
2052/**
2053 * cpufreq_get_driver_data - Return current driver data.
2054 *
2055 * Return the private data of the currently registered cpufreq driver, or NULL
2056 * if no cpufreq driver has been registered.
2057 */
2058void *cpufreq_get_driver_data(void)
2059{
2060 if (cpufreq_driver)
2061 return cpufreq_driver->driver_data;
2062
2063 return NULL;
2064}
2065EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
2066
2067/*********************************************************************
2068 * NOTIFIER LISTS INTERFACE *
2069 *********************************************************************/
2070
2071/**
2072 * cpufreq_register_notifier - Register a notifier with cpufreq.
2073 * @nb: notifier function to register.
2074 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2075 *
2076 * Add a notifier to one of two lists: either a list of notifiers that run on
2077 * clock rate changes (once before and once after every transition), or a list
2078 * of notifiers that ron on cpufreq policy changes.
2079 *
2080 * This function may sleep and it has the same return values as
2081 * blocking_notifier_chain_register().
2082 */
2083int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
2084{
2085 int ret;
2086
2087 if (cpufreq_disabled())
2088 return -EINVAL;
2089
2090 switch (list) {
2091 case CPUFREQ_TRANSITION_NOTIFIER:
2092 mutex_lock(&cpufreq_fast_switch_lock);
2093
2094 if (cpufreq_fast_switch_count > 0) {
2095 mutex_unlock(&cpufreq_fast_switch_lock);
2096 return -EBUSY;
2097 }
2098 ret = srcu_notifier_chain_register(
2099 &cpufreq_transition_notifier_list, nb);
2100 if (!ret)
2101 cpufreq_fast_switch_count--;
2102
2103 mutex_unlock(&cpufreq_fast_switch_lock);
2104 break;
2105 case CPUFREQ_POLICY_NOTIFIER:
2106 ret = blocking_notifier_chain_register(
2107 &cpufreq_policy_notifier_list, nb);
2108 break;
2109 default:
2110 ret = -EINVAL;
2111 }
2112
2113 return ret;
2114}
2115EXPORT_SYMBOL(cpufreq_register_notifier);
2116
2117/**
2118 * cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2119 * @nb: notifier block to be unregistered.
2120 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2121 *
2122 * Remove a notifier from one of the cpufreq notifier lists.
2123 *
2124 * This function may sleep and it has the same return values as
2125 * blocking_notifier_chain_unregister().
2126 */
2127int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2128{
2129 int ret;
2130
2131 if (cpufreq_disabled())
2132 return -EINVAL;
2133
2134 switch (list) {
2135 case CPUFREQ_TRANSITION_NOTIFIER:
2136 mutex_lock(&cpufreq_fast_switch_lock);
2137
2138 ret = srcu_notifier_chain_unregister(
2139 &cpufreq_transition_notifier_list, nb);
2140 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2141 cpufreq_fast_switch_count++;
2142
2143 mutex_unlock(&cpufreq_fast_switch_lock);
2144 break;
2145 case CPUFREQ_POLICY_NOTIFIER:
2146 ret = blocking_notifier_chain_unregister(
2147 &cpufreq_policy_notifier_list, nb);
2148 break;
2149 default:
2150 ret = -EINVAL;
2151 }
2152
2153 return ret;
2154}
2155EXPORT_SYMBOL(cpufreq_unregister_notifier);
2156
2157
2158/*********************************************************************
2159 * GOVERNORS *
2160 *********************************************************************/
2161
2162/**
2163 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2164 * @policy: cpufreq policy to switch the frequency for.
2165 * @target_freq: New frequency to set (may be approximate).
2166 *
2167 * Carry out a fast frequency switch without sleeping.
2168 *
2169 * The driver's ->fast_switch() callback invoked by this function must be
2170 * suitable for being called from within RCU-sched read-side critical sections
2171 * and it is expected to select the minimum available frequency greater than or
2172 * equal to @target_freq (CPUFREQ_RELATION_L).
2173 *
2174 * This function must not be called if policy->fast_switch_enabled is unset.
2175 *
2176 * Governors calling this function must guarantee that it will never be invoked
2177 * twice in parallel for the same policy and that it will never be called in
2178 * parallel with either ->target() or ->target_index() for the same policy.
2179 *
2180 * Returns the actual frequency set for the CPU.
2181 *
2182 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2183 * error condition, the hardware configuration must be preserved.
2184 */
2185unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2186 unsigned int target_freq)
2187{
2188 unsigned int freq;
2189 int cpu;
2190
2191 target_freq = clamp_val(target_freq, policy->min, policy->max);
2192 freq = cpufreq_driver->fast_switch(policy, target_freq);
2193
2194 if (!freq)
2195 return 0;
2196
2197 policy->cur = freq;
2198 arch_set_freq_scale(policy->related_cpus, freq,
2199 arch_scale_freq_ref(policy->cpu));
2200 cpufreq_stats_record_transition(policy, freq);
2201
2202 if (trace_cpu_frequency_enabled()) {
2203 for_each_cpu(cpu, policy->cpus)
2204 trace_cpu_frequency(freq, cpu);
2205 }
2206
2207 return freq;
2208}
2209EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2210
2211/**
2212 * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2213 * @cpu: Target CPU.
2214 * @min_perf: Minimum (required) performance level (units of @capacity).
2215 * @target_perf: Target (desired) performance level (units of @capacity).
2216 * @capacity: Capacity of the target CPU.
2217 *
2218 * Carry out a fast performance level switch of @cpu without sleeping.
2219 *
2220 * The driver's ->adjust_perf() callback invoked by this function must be
2221 * suitable for being called from within RCU-sched read-side critical sections
2222 * and it is expected to select a suitable performance level equal to or above
2223 * @min_perf and preferably equal to or below @target_perf.
2224 *
2225 * This function must not be called if policy->fast_switch_enabled is unset.
2226 *
2227 * Governors calling this function must guarantee that it will never be invoked
2228 * twice in parallel for the same CPU and that it will never be called in
2229 * parallel with either ->target() or ->target_index() or ->fast_switch() for
2230 * the same CPU.
2231 */
2232void cpufreq_driver_adjust_perf(unsigned int cpu,
2233 unsigned long min_perf,
2234 unsigned long target_perf,
2235 unsigned long capacity)
2236{
2237 cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2238}
2239
2240/**
2241 * cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2242 *
2243 * Return 'true' if the ->adjust_perf callback is present for the
2244 * current driver or 'false' otherwise.
2245 */
2246bool cpufreq_driver_has_adjust_perf(void)
2247{
2248 return !!cpufreq_driver->adjust_perf;
2249}
2250
2251/* Must set freqs->new to intermediate frequency */
2252static int __target_intermediate(struct cpufreq_policy *policy,
2253 struct cpufreq_freqs *freqs, int index)
2254{
2255 int ret;
2256
2257 freqs->new = cpufreq_driver->get_intermediate(policy, index);
2258
2259 /* We don't need to switch to intermediate freq */
2260 if (!freqs->new)
2261 return 0;
2262
2263 pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2264 __func__, policy->cpu, freqs->old, freqs->new);
2265
2266 cpufreq_freq_transition_begin(policy, freqs);
2267 ret = cpufreq_driver->target_intermediate(policy, index);
2268 cpufreq_freq_transition_end(policy, freqs, ret);
2269
2270 if (ret)
2271 pr_err("%s: Failed to change to intermediate frequency: %d\n",
2272 __func__, ret);
2273
2274 return ret;
2275}
2276
2277static int __target_index(struct cpufreq_policy *policy, int index)
2278{
2279 struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2280 unsigned int restore_freq, intermediate_freq = 0;
2281 unsigned int newfreq = policy->freq_table[index].frequency;
2282 int retval = -EINVAL;
2283 bool notify;
2284
2285 if (newfreq == policy->cur)
2286 return 0;
2287
2288 /* Save last value to restore later on errors */
2289 restore_freq = policy->cur;
2290
2291 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2292 if (notify) {
2293 /* Handle switching to intermediate frequency */
2294 if (cpufreq_driver->get_intermediate) {
2295 retval = __target_intermediate(policy, &freqs, index);
2296 if (retval)
2297 return retval;
2298
2299 intermediate_freq = freqs.new;
2300 /* Set old freq to intermediate */
2301 if (intermediate_freq)
2302 freqs.old = freqs.new;
2303 }
2304
2305 freqs.new = newfreq;
2306 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2307 __func__, policy->cpu, freqs.old, freqs.new);
2308
2309 cpufreq_freq_transition_begin(policy, &freqs);
2310 }
2311
2312 retval = cpufreq_driver->target_index(policy, index);
2313 if (retval)
2314 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2315 retval);
2316
2317 if (notify) {
2318 cpufreq_freq_transition_end(policy, &freqs, retval);
2319
2320 /*
2321 * Failed after setting to intermediate freq? Driver should have
2322 * reverted back to initial frequency and so should we. Check
2323 * here for intermediate_freq instead of get_intermediate, in
2324 * case we haven't switched to intermediate freq at all.
2325 */
2326 if (unlikely(retval && intermediate_freq)) {
2327 freqs.old = intermediate_freq;
2328 freqs.new = restore_freq;
2329 cpufreq_freq_transition_begin(policy, &freqs);
2330 cpufreq_freq_transition_end(policy, &freqs, 0);
2331 }
2332 }
2333
2334 return retval;
2335}
2336
2337int __cpufreq_driver_target(struct cpufreq_policy *policy,
2338 unsigned int target_freq,
2339 unsigned int relation)
2340{
2341 unsigned int old_target_freq = target_freq;
2342
2343 if (cpufreq_disabled())
2344 return -ENODEV;
2345
2346 target_freq = __resolve_freq(policy, target_freq, relation);
2347
2348 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2349 policy->cpu, target_freq, relation, old_target_freq);
2350
2351 /*
2352 * This might look like a redundant call as we are checking it again
2353 * after finding index. But it is left intentionally for cases where
2354 * exactly same freq is called again and so we can save on few function
2355 * calls.
2356 */
2357 if (target_freq == policy->cur &&
2358 !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2359 return 0;
2360
2361 if (cpufreq_driver->target) {
2362 /*
2363 * If the driver hasn't setup a single inefficient frequency,
2364 * it's unlikely it knows how to decode CPUFREQ_RELATION_E.
2365 */
2366 if (!policy->efficiencies_available)
2367 relation &= ~CPUFREQ_RELATION_E;
2368
2369 return cpufreq_driver->target(policy, target_freq, relation);
2370 }
2371
2372 if (!cpufreq_driver->target_index)
2373 return -EINVAL;
2374
2375 return __target_index(policy, policy->cached_resolved_idx);
2376}
2377EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2378
2379int cpufreq_driver_target(struct cpufreq_policy *policy,
2380 unsigned int target_freq,
2381 unsigned int relation)
2382{
2383 int ret;
2384
2385 down_write(&policy->rwsem);
2386
2387 ret = __cpufreq_driver_target(policy, target_freq, relation);
2388
2389 up_write(&policy->rwsem);
2390
2391 return ret;
2392}
2393EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2394
2395__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2396{
2397 return NULL;
2398}
2399
2400static int cpufreq_init_governor(struct cpufreq_policy *policy)
2401{
2402 int ret;
2403
2404 /* Don't start any governor operations if we are entering suspend */
2405 if (cpufreq_suspended)
2406 return 0;
2407 /*
2408 * Governor might not be initiated here if ACPI _PPC changed
2409 * notification happened, so check it.
2410 */
2411 if (!policy->governor)
2412 return -EINVAL;
2413
2414 /* Platform doesn't want dynamic frequency switching ? */
2415 if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2416 cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2417 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2418
2419 if (gov) {
2420 pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2421 policy->governor->name, gov->name);
2422 policy->governor = gov;
2423 } else {
2424 return -EINVAL;
2425 }
2426 }
2427
2428 if (!try_module_get(policy->governor->owner))
2429 return -EINVAL;
2430
2431 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2432
2433 if (policy->governor->init) {
2434 ret = policy->governor->init(policy);
2435 if (ret) {
2436 module_put(policy->governor->owner);
2437 return ret;
2438 }
2439 }
2440
2441 policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2442
2443 return 0;
2444}
2445
2446static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2447{
2448 if (cpufreq_suspended || !policy->governor)
2449 return;
2450
2451 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2452
2453 if (policy->governor->exit)
2454 policy->governor->exit(policy);
2455
2456 module_put(policy->governor->owner);
2457}
2458
2459int cpufreq_start_governor(struct cpufreq_policy *policy)
2460{
2461 int ret;
2462
2463 if (cpufreq_suspended)
2464 return 0;
2465
2466 if (!policy->governor)
2467 return -EINVAL;
2468
2469 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2470
2471 if (cpufreq_driver->get)
2472 cpufreq_verify_current_freq(policy, false);
2473
2474 if (policy->governor->start) {
2475 ret = policy->governor->start(policy);
2476 if (ret)
2477 return ret;
2478 }
2479
2480 if (policy->governor->limits)
2481 policy->governor->limits(policy);
2482
2483 return 0;
2484}
2485
2486void cpufreq_stop_governor(struct cpufreq_policy *policy)
2487{
2488 if (cpufreq_suspended || !policy->governor)
2489 return;
2490
2491 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2492
2493 if (policy->governor->stop)
2494 policy->governor->stop(policy);
2495}
2496
2497static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2498{
2499 if (cpufreq_suspended || !policy->governor)
2500 return;
2501
2502 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2503
2504 if (policy->governor->limits)
2505 policy->governor->limits(policy);
2506}
2507
2508int cpufreq_register_governor(struct cpufreq_governor *governor)
2509{
2510 int err;
2511
2512 if (!governor)
2513 return -EINVAL;
2514
2515 if (cpufreq_disabled())
2516 return -ENODEV;
2517
2518 mutex_lock(&cpufreq_governor_mutex);
2519
2520 err = -EBUSY;
2521 if (!find_governor(governor->name)) {
2522 err = 0;
2523 list_add(&governor->governor_list, &cpufreq_governor_list);
2524 }
2525
2526 mutex_unlock(&cpufreq_governor_mutex);
2527 return err;
2528}
2529EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2530
2531void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2532{
2533 struct cpufreq_policy *policy;
2534 unsigned long flags;
2535
2536 if (!governor)
2537 return;
2538
2539 if (cpufreq_disabled())
2540 return;
2541
2542 /* clear last_governor for all inactive policies */
2543 read_lock_irqsave(&cpufreq_driver_lock, flags);
2544 for_each_inactive_policy(policy) {
2545 if (!strcmp(policy->last_governor, governor->name)) {
2546 policy->governor = NULL;
2547 strcpy(policy->last_governor, "\0");
2548 }
2549 }
2550 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2551
2552 mutex_lock(&cpufreq_governor_mutex);
2553 list_del(&governor->governor_list);
2554 mutex_unlock(&cpufreq_governor_mutex);
2555}
2556EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2557
2558
2559/*********************************************************************
2560 * POLICY INTERFACE *
2561 *********************************************************************/
2562
2563/**
2564 * cpufreq_get_policy - get the current cpufreq_policy
2565 * @policy: struct cpufreq_policy into which the current cpufreq_policy
2566 * is written
2567 * @cpu: CPU to find the policy for
2568 *
2569 * Reads the current cpufreq policy.
2570 */
2571int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2572{
2573 struct cpufreq_policy *cpu_policy;
2574 if (!policy)
2575 return -EINVAL;
2576
2577 cpu_policy = cpufreq_cpu_get(cpu);
2578 if (!cpu_policy)
2579 return -EINVAL;
2580
2581 memcpy(policy, cpu_policy, sizeof(*policy));
2582
2583 cpufreq_cpu_put(cpu_policy);
2584 return 0;
2585}
2586EXPORT_SYMBOL(cpufreq_get_policy);
2587
2588/**
2589 * cpufreq_set_policy - Modify cpufreq policy parameters.
2590 * @policy: Policy object to modify.
2591 * @new_gov: Policy governor pointer.
2592 * @new_pol: Policy value (for drivers with built-in governors).
2593 *
2594 * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2595 * limits to be set for the policy, update @policy with the verified limits
2596 * values and either invoke the driver's ->setpolicy() callback (if present) or
2597 * carry out a governor update for @policy. That is, run the current governor's
2598 * ->limits() callback (if @new_gov points to the same object as the one in
2599 * @policy) or replace the governor for @policy with @new_gov.
2600 *
2601 * The cpuinfo part of @policy is not updated by this function.
2602 */
2603static int cpufreq_set_policy(struct cpufreq_policy *policy,
2604 struct cpufreq_governor *new_gov,
2605 unsigned int new_pol)
2606{
2607 struct cpufreq_policy_data new_data;
2608 struct cpufreq_governor *old_gov;
2609 int ret;
2610
2611 memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2612 new_data.freq_table = policy->freq_table;
2613 new_data.cpu = policy->cpu;
2614 /*
2615 * PM QoS framework collects all the requests from users and provide us
2616 * the final aggregated value here.
2617 */
2618 new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2619 new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2620
2621 pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2622 new_data.cpu, new_data.min, new_data.max);
2623
2624 /*
2625 * Verify that the CPU speed can be set within these limits and make sure
2626 * that min <= max.
2627 */
2628 ret = cpufreq_driver->verify(&new_data);
2629 if (ret)
2630 return ret;
2631
2632 /*
2633 * Resolve policy min/max to available frequencies. It ensures
2634 * no frequency resolution will neither overshoot the requested maximum
2635 * nor undershoot the requested minimum.
2636 */
2637 policy->min = new_data.min;
2638 policy->max = new_data.max;
2639 policy->min = __resolve_freq(policy, policy->min, CPUFREQ_RELATION_L);
2640 policy->max = __resolve_freq(policy, policy->max, CPUFREQ_RELATION_H);
2641 trace_cpu_frequency_limits(policy);
2642
2643 policy->cached_target_freq = UINT_MAX;
2644
2645 pr_debug("new min and max freqs are %u - %u kHz\n",
2646 policy->min, policy->max);
2647
2648 if (cpufreq_driver->setpolicy) {
2649 policy->policy = new_pol;
2650 pr_debug("setting range\n");
2651 return cpufreq_driver->setpolicy(policy);
2652 }
2653
2654 if (new_gov == policy->governor) {
2655 pr_debug("governor limits update\n");
2656 cpufreq_governor_limits(policy);
2657 return 0;
2658 }
2659
2660 pr_debug("governor switch\n");
2661
2662 /* save old, working values */
2663 old_gov = policy->governor;
2664 /* end old governor */
2665 if (old_gov) {
2666 cpufreq_stop_governor(policy);
2667 cpufreq_exit_governor(policy);
2668 }
2669
2670 /* start new governor */
2671 policy->governor = new_gov;
2672 ret = cpufreq_init_governor(policy);
2673 if (!ret) {
2674 ret = cpufreq_start_governor(policy);
2675 if (!ret) {
2676 pr_debug("governor change\n");
2677 return 0;
2678 }
2679 cpufreq_exit_governor(policy);
2680 }
2681
2682 /* new governor failed, so re-start old one */
2683 pr_debug("starting governor %s failed\n", policy->governor->name);
2684 if (old_gov) {
2685 policy->governor = old_gov;
2686 if (cpufreq_init_governor(policy))
2687 policy->governor = NULL;
2688 else
2689 cpufreq_start_governor(policy);
2690 }
2691
2692 return ret;
2693}
2694
2695/**
2696 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2697 * @cpu: CPU to re-evaluate the policy for.
2698 *
2699 * Update the current frequency for the cpufreq policy of @cpu and use
2700 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2701 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2702 * for the policy in question, among other things.
2703 */
2704void cpufreq_update_policy(unsigned int cpu)
2705{
2706 struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2707
2708 if (!policy)
2709 return;
2710
2711 /*
2712 * BIOS might change freq behind our back
2713 * -> ask driver for current freq and notify governors about a change
2714 */
2715 if (cpufreq_driver->get && has_target() &&
2716 (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2717 goto unlock;
2718
2719 refresh_frequency_limits(policy);
2720
2721unlock:
2722 cpufreq_cpu_release(policy);
2723}
2724EXPORT_SYMBOL(cpufreq_update_policy);
2725
2726/**
2727 * cpufreq_update_limits - Update policy limits for a given CPU.
2728 * @cpu: CPU to update the policy limits for.
2729 *
2730 * Invoke the driver's ->update_limits callback if present or call
2731 * cpufreq_update_policy() for @cpu.
2732 */
2733void cpufreq_update_limits(unsigned int cpu)
2734{
2735 if (cpufreq_driver->update_limits)
2736 cpufreq_driver->update_limits(cpu);
2737 else
2738 cpufreq_update_policy(cpu);
2739}
2740EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2741
2742/*********************************************************************
2743 * BOOST *
2744 *********************************************************************/
2745static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2746{
2747 int ret;
2748
2749 if (!policy->freq_table)
2750 return -ENXIO;
2751
2752 ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2753 if (ret) {
2754 pr_err("%s: Policy frequency update failed\n", __func__);
2755 return ret;
2756 }
2757
2758 ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2759 if (ret < 0)
2760 return ret;
2761
2762 return 0;
2763}
2764
2765int cpufreq_boost_trigger_state(int state)
2766{
2767 struct cpufreq_policy *policy;
2768 unsigned long flags;
2769 int ret = 0;
2770
2771 if (cpufreq_driver->boost_enabled == state)
2772 return 0;
2773
2774 write_lock_irqsave(&cpufreq_driver_lock, flags);
2775 cpufreq_driver->boost_enabled = state;
2776 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2777
2778 cpus_read_lock();
2779 for_each_active_policy(policy) {
2780 policy->boost_enabled = state;
2781 ret = cpufreq_driver->set_boost(policy, state);
2782 if (ret) {
2783 policy->boost_enabled = !policy->boost_enabled;
2784 goto err_reset_state;
2785 }
2786 }
2787 cpus_read_unlock();
2788
2789 return 0;
2790
2791err_reset_state:
2792 cpus_read_unlock();
2793
2794 write_lock_irqsave(&cpufreq_driver_lock, flags);
2795 cpufreq_driver->boost_enabled = !state;
2796 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2797
2798 pr_err("%s: Cannot %s BOOST\n",
2799 __func__, state ? "enable" : "disable");
2800
2801 return ret;
2802}
2803
2804static bool cpufreq_boost_supported(void)
2805{
2806 return cpufreq_driver->set_boost;
2807}
2808
2809static int create_boost_sysfs_file(void)
2810{
2811 int ret;
2812
2813 ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2814 if (ret)
2815 pr_err("%s: cannot register global BOOST sysfs file\n",
2816 __func__);
2817
2818 return ret;
2819}
2820
2821static void remove_boost_sysfs_file(void)
2822{
2823 if (cpufreq_boost_supported())
2824 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2825}
2826
2827int cpufreq_enable_boost_support(void)
2828{
2829 if (!cpufreq_driver)
2830 return -EINVAL;
2831
2832 if (cpufreq_boost_supported())
2833 return 0;
2834
2835 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2836
2837 /* This will get removed on driver unregister */
2838 return create_boost_sysfs_file();
2839}
2840EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2841
2842int cpufreq_boost_enabled(void)
2843{
2844 return cpufreq_driver->boost_enabled;
2845}
2846EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2847
2848/*********************************************************************
2849 * REGISTER / UNREGISTER CPUFREQ DRIVER *
2850 *********************************************************************/
2851static enum cpuhp_state hp_online;
2852
2853static int cpuhp_cpufreq_online(unsigned int cpu)
2854{
2855 cpufreq_online(cpu);
2856
2857 return 0;
2858}
2859
2860static int cpuhp_cpufreq_offline(unsigned int cpu)
2861{
2862 cpufreq_offline(cpu);
2863
2864 return 0;
2865}
2866
2867/**
2868 * cpufreq_register_driver - register a CPU Frequency driver
2869 * @driver_data: A struct cpufreq_driver containing the values#
2870 * submitted by the CPU Frequency driver.
2871 *
2872 * Registers a CPU Frequency driver to this core code. This code
2873 * returns zero on success, -EEXIST when another driver got here first
2874 * (and isn't unregistered in the meantime).
2875 *
2876 */
2877int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2878{
2879 unsigned long flags;
2880 int ret;
2881
2882 if (cpufreq_disabled())
2883 return -ENODEV;
2884
2885 /*
2886 * The cpufreq core depends heavily on the availability of device
2887 * structure, make sure they are available before proceeding further.
2888 */
2889 if (!get_cpu_device(0))
2890 return -EPROBE_DEFER;
2891
2892 if (!driver_data || !driver_data->verify || !driver_data->init ||
2893 !(driver_data->setpolicy || driver_data->target_index ||
2894 driver_data->target) ||
2895 (driver_data->setpolicy && (driver_data->target_index ||
2896 driver_data->target)) ||
2897 (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2898 (!driver_data->online != !driver_data->offline) ||
2899 (driver_data->adjust_perf && !driver_data->fast_switch))
2900 return -EINVAL;
2901
2902 pr_debug("trying to register driver %s\n", driver_data->name);
2903
2904 /* Protect against concurrent CPU online/offline. */
2905 cpus_read_lock();
2906
2907 write_lock_irqsave(&cpufreq_driver_lock, flags);
2908 if (cpufreq_driver) {
2909 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2910 ret = -EEXIST;
2911 goto out;
2912 }
2913 cpufreq_driver = driver_data;
2914 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2915
2916 /*
2917 * Mark support for the scheduler's frequency invariance engine for
2918 * drivers that implement target(), target_index() or fast_switch().
2919 */
2920 if (!cpufreq_driver->setpolicy) {
2921 static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2922 pr_debug("supports frequency invariance");
2923 }
2924
2925 if (driver_data->setpolicy)
2926 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2927
2928 if (cpufreq_boost_supported()) {
2929 ret = create_boost_sysfs_file();
2930 if (ret)
2931 goto err_null_driver;
2932 }
2933
2934 ret = subsys_interface_register(&cpufreq_interface);
2935 if (ret)
2936 goto err_boost_unreg;
2937
2938 if (unlikely(list_empty(&cpufreq_policy_list))) {
2939 /* if all ->init() calls failed, unregister */
2940 ret = -ENODEV;
2941 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2942 driver_data->name);
2943 goto err_if_unreg;
2944 }
2945
2946 ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2947 "cpufreq:online",
2948 cpuhp_cpufreq_online,
2949 cpuhp_cpufreq_offline);
2950 if (ret < 0)
2951 goto err_if_unreg;
2952 hp_online = ret;
2953 ret = 0;
2954
2955 pr_debug("driver %s up and running\n", driver_data->name);
2956 goto out;
2957
2958err_if_unreg:
2959 subsys_interface_unregister(&cpufreq_interface);
2960err_boost_unreg:
2961 remove_boost_sysfs_file();
2962err_null_driver:
2963 write_lock_irqsave(&cpufreq_driver_lock, flags);
2964 cpufreq_driver = NULL;
2965 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2966out:
2967 cpus_read_unlock();
2968 return ret;
2969}
2970EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2971
2972/*
2973 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2974 *
2975 * Unregister the current CPUFreq driver. Only call this if you have
2976 * the right to do so, i.e. if you have succeeded in initialising before!
2977 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2978 * currently not initialised.
2979 */
2980void cpufreq_unregister_driver(struct cpufreq_driver *driver)
2981{
2982 unsigned long flags;
2983
2984 if (WARN_ON(!cpufreq_driver || (driver != cpufreq_driver)))
2985 return;
2986
2987 pr_debug("unregistering driver %s\n", driver->name);
2988
2989 /* Protect against concurrent cpu hotplug */
2990 cpus_read_lock();
2991 subsys_interface_unregister(&cpufreq_interface);
2992 remove_boost_sysfs_file();
2993 static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
2994 cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2995
2996 write_lock_irqsave(&cpufreq_driver_lock, flags);
2997
2998 cpufreq_driver = NULL;
2999
3000 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
3001 cpus_read_unlock();
3002}
3003EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
3004
3005static int __init cpufreq_core_init(void)
3006{
3007 struct cpufreq_governor *gov = cpufreq_default_governor();
3008 struct device *dev_root;
3009
3010 if (cpufreq_disabled())
3011 return -ENODEV;
3012
3013 dev_root = bus_get_dev_root(&cpu_subsys);
3014 if (dev_root) {
3015 cpufreq_global_kobject = kobject_create_and_add("cpufreq", &dev_root->kobj);
3016 put_device(dev_root);
3017 }
3018 BUG_ON(!cpufreq_global_kobject);
3019
3020 if (!strlen(default_governor))
3021 strscpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
3022
3023 return 0;
3024}
3025module_param(off, int, 0444);
3026module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
3027core_initcall(cpufreq_core_init);