2 * linux/drivers/cpufreq/cpufreq.c
4 * Copyright (C) 2001 Russell King
5 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6 * (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
8 * Oct 2005 - Ashok Raj <ashok.raj@intel.com>
9 * Added handling for CPU hotplug
10 * Feb 2006 - Jacob Shin <jacob.shin@amd.com>
11 * Fix handling for CPU hotplug -- affected CPUs
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License version 2 as
15 * published by the Free Software Foundation.
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <asm/cputime.h>
21 #include <linux/kernel.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/notifier.h>
26 #include <linux/cpufreq.h>
27 #include <linux/delay.h>
28 #include <linux/interrupt.h>
29 #include <linux/spinlock.h>
30 #include <linux/tick.h>
31 #include <linux/device.h>
32 #include <linux/slab.h>
33 #include <linux/cpu.h>
34 #include <linux/completion.h>
35 #include <linux/mutex.h>
36 #include <linux/syscore_ops.h>
38 #include <trace/events/power.h>
41 * The "cpufreq driver" - the arch- or hardware-dependent low
42 * level driver of CPUFreq support, and its spinlock. This lock
43 * also protects the cpufreq_cpu_data array.
45 static struct cpufreq_driver *cpufreq_driver;
46 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
47 static DEFINE_RWLOCK(cpufreq_driver_lock);
48 static DEFINE_MUTEX(cpufreq_governor_lock);
50 #ifdef CONFIG_HOTPLUG_CPU
51 /* This one keeps track of the previously set governor of a removed CPU */
52 static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
56 * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
57 * all cpufreq/hotplug/workqueue/etc related lock issues.
59 * The rules for this semaphore:
60 * - Any routine that wants to read from the policy structure will
61 * do a down_read on this semaphore.
62 * - Any routine that will write to the policy structure and/or may take away
63 * the policy altogether (eg. CPU hotplug), will hold this lock in write
64 * mode before doing so.
67 * - Governor routines that can be called in cpufreq hotplug path should not
68 * take this sem as top level hotplug notifier handler takes this.
69 * - Lock should not be held across
70 * __cpufreq_governor(data, CPUFREQ_GOV_STOP);
72 static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
73 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
75 #define lock_policy_rwsem(mode, cpu) \
76 static int lock_policy_rwsem_##mode(int cpu) \
78 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
79 BUG_ON(policy_cpu == -1); \
80 down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
85 lock_policy_rwsem(read, cpu);
86 lock_policy_rwsem(write, cpu);
88 #define unlock_policy_rwsem(mode, cpu) \
89 static void unlock_policy_rwsem_##mode(int cpu) \
91 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
92 BUG_ON(policy_cpu == -1); \
93 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
96 unlock_policy_rwsem(read, cpu);
97 unlock_policy_rwsem(write, cpu);
99 /* internal prototypes */
100 static int __cpufreq_governor(struct cpufreq_policy *policy,
102 static unsigned int __cpufreq_get(unsigned int cpu);
103 static void handle_update(struct work_struct *work);
106 * Two notifier lists: the "policy" list is involved in the
107 * validation process for a new CPU frequency policy; the
108 * "transition" list for kernel code that needs to handle
109 * changes to devices when the CPU clock speed changes.
110 * The mutex locks both lists.
112 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
113 static struct srcu_notifier_head cpufreq_transition_notifier_list;
115 static bool init_cpufreq_transition_notifier_list_called;
116 static int __init init_cpufreq_transition_notifier_list(void)
118 srcu_init_notifier_head(&cpufreq_transition_notifier_list);
119 init_cpufreq_transition_notifier_list_called = true;
122 pure_initcall(init_cpufreq_transition_notifier_list);
124 static int off __read_mostly;
125 static int cpufreq_disabled(void)
129 void disable_cpufreq(void)
133 static LIST_HEAD(cpufreq_governor_list);
134 static DEFINE_MUTEX(cpufreq_governor_mutex);
136 bool have_governor_per_policy(void)
138 return cpufreq_driver->have_governor_per_policy;
140 EXPORT_SYMBOL_GPL(have_governor_per_policy);
142 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
144 if (have_governor_per_policy())
145 return &policy->kobj;
147 return cpufreq_global_kobject;
149 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
151 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
157 cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
159 busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
160 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
161 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
162 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
163 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
164 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
166 idle_time = cur_wall_time - busy_time;
168 *wall = cputime_to_usecs(cur_wall_time);
170 return cputime_to_usecs(idle_time);
173 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
175 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
177 if (idle_time == -1ULL)
178 return get_cpu_idle_time_jiffy(cpu, wall);
180 idle_time += get_cpu_iowait_time_us(cpu, wall);
184 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
186 static struct cpufreq_policy *__cpufreq_cpu_get(unsigned int cpu, bool sysfs)
188 struct cpufreq_policy *data;
191 if (cpu >= nr_cpu_ids)
194 /* get the cpufreq driver */
195 read_lock_irqsave(&cpufreq_driver_lock, flags);
200 if (!try_module_get(cpufreq_driver->owner))
204 data = per_cpu(cpufreq_cpu_data, cpu);
207 goto err_out_put_module;
209 if (!sysfs && !kobject_get(&data->kobj))
210 goto err_out_put_module;
212 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
216 module_put(cpufreq_driver->owner);
218 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
223 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
225 if (cpufreq_disabled())
228 return __cpufreq_cpu_get(cpu, false);
230 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
232 static struct cpufreq_policy *cpufreq_cpu_get_sysfs(unsigned int cpu)
234 return __cpufreq_cpu_get(cpu, true);
237 static void __cpufreq_cpu_put(struct cpufreq_policy *data, bool sysfs)
240 kobject_put(&data->kobj);
241 module_put(cpufreq_driver->owner);
244 void cpufreq_cpu_put(struct cpufreq_policy *data)
246 if (cpufreq_disabled())
249 __cpufreq_cpu_put(data, false);
251 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
253 static void cpufreq_cpu_put_sysfs(struct cpufreq_policy *data)
255 __cpufreq_cpu_put(data, true);
258 /*********************************************************************
259 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
260 *********************************************************************/
263 * adjust_jiffies - adjust the system "loops_per_jiffy"
265 * This function alters the system "loops_per_jiffy" for the clock
266 * speed change. Note that loops_per_jiffy cannot be updated on SMP
267 * systems as each CPU might be scaled differently. So, use the arch
268 * per-CPU loops_per_jiffy value wherever possible.
271 static unsigned long l_p_j_ref;
272 static unsigned int l_p_j_ref_freq;
274 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
276 if (ci->flags & CPUFREQ_CONST_LOOPS)
279 if (!l_p_j_ref_freq) {
280 l_p_j_ref = loops_per_jiffy;
281 l_p_j_ref_freq = ci->old;
282 pr_debug("saving %lu as reference value for loops_per_jiffy; "
283 "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
285 if ((val == CPUFREQ_POSTCHANGE && ci->old != ci->new) ||
286 (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
287 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
289 pr_debug("scaling loops_per_jiffy to %lu "
290 "for frequency %u kHz\n", loops_per_jiffy, ci->new);
294 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
300 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
301 struct cpufreq_freqs *freqs, unsigned int state)
303 BUG_ON(irqs_disabled());
305 if (cpufreq_disabled())
308 freqs->flags = cpufreq_driver->flags;
309 pr_debug("notification %u of frequency transition to %u kHz\n",
314 case CPUFREQ_PRECHANGE:
315 if (WARN(policy->transition_ongoing ==
316 cpumask_weight(policy->cpus),
317 "In middle of another frequency transition\n"))
320 policy->transition_ongoing++;
322 /* detect if the driver reported a value as "old frequency"
323 * which is not equal to what the cpufreq core thinks is
326 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
327 if ((policy) && (policy->cpu == freqs->cpu) &&
328 (policy->cur) && (policy->cur != freqs->old)) {
329 pr_debug("Warning: CPU frequency is"
330 " %u, cpufreq assumed %u kHz.\n",
331 freqs->old, policy->cur);
332 freqs->old = policy->cur;
335 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
336 CPUFREQ_PRECHANGE, freqs);
337 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
340 case CPUFREQ_POSTCHANGE:
341 if (WARN(!policy->transition_ongoing,
342 "No frequency transition in progress\n"))
345 policy->transition_ongoing--;
347 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
348 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
349 (unsigned long)freqs->cpu);
350 trace_cpu_frequency(freqs->new, freqs->cpu);
351 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
352 CPUFREQ_POSTCHANGE, freqs);
353 if (likely(policy) && likely(policy->cpu == freqs->cpu))
354 policy->cur = freqs->new;
360 * cpufreq_notify_transition - call notifier chain and adjust_jiffies
361 * on frequency transition.
363 * This function calls the transition notifiers and the "adjust_jiffies"
364 * function. It is called twice on all CPU frequency changes that have
367 void cpufreq_notify_transition(struct cpufreq_policy *policy,
368 struct cpufreq_freqs *freqs, unsigned int state)
370 for_each_cpu(freqs->cpu, policy->cpus)
371 __cpufreq_notify_transition(policy, freqs, state);
373 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
376 /*********************************************************************
378 *********************************************************************/
380 static struct cpufreq_governor *__find_governor(const char *str_governor)
382 struct cpufreq_governor *t;
384 list_for_each_entry(t, &cpufreq_governor_list, governor_list)
385 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
392 * cpufreq_parse_governor - parse a governor string
394 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
395 struct cpufreq_governor **governor)
402 if (cpufreq_driver->setpolicy) {
403 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
404 *policy = CPUFREQ_POLICY_PERFORMANCE;
406 } else if (!strnicmp(str_governor, "powersave",
408 *policy = CPUFREQ_POLICY_POWERSAVE;
411 } else if (cpufreq_driver->target) {
412 struct cpufreq_governor *t;
414 mutex_lock(&cpufreq_governor_mutex);
416 t = __find_governor(str_governor);
421 mutex_unlock(&cpufreq_governor_mutex);
422 ret = request_module("cpufreq_%s", str_governor);
423 mutex_lock(&cpufreq_governor_mutex);
426 t = __find_governor(str_governor);
434 mutex_unlock(&cpufreq_governor_mutex);
441 * cpufreq_per_cpu_attr_read() / show_##file_name() -
442 * print out cpufreq information
444 * Write out information from cpufreq_driver->policy[cpu]; object must be
448 #define show_one(file_name, object) \
449 static ssize_t show_##file_name \
450 (struct cpufreq_policy *policy, char *buf) \
452 return sprintf(buf, "%u\n", policy->object); \
455 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
456 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
457 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
458 show_one(scaling_min_freq, min);
459 show_one(scaling_max_freq, max);
460 show_one(scaling_cur_freq, cur);
462 static int __cpufreq_set_policy(struct cpufreq_policy *data,
463 struct cpufreq_policy *policy);
466 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
468 #define store_one(file_name, object) \
469 static ssize_t store_##file_name \
470 (struct cpufreq_policy *policy, const char *buf, size_t count) \
473 struct cpufreq_policy new_policy; \
475 ret = cpufreq_get_policy(&new_policy, policy->cpu); \
479 ret = sscanf(buf, "%u", &new_policy.object); \
483 ret = __cpufreq_set_policy(policy, &new_policy); \
484 policy->user_policy.object = policy->object; \
486 return ret ? ret : count; \
489 store_one(scaling_min_freq, min);
490 store_one(scaling_max_freq, max);
493 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
495 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
498 unsigned int cur_freq = __cpufreq_get(policy->cpu);
500 return sprintf(buf, "<unknown>");
501 return sprintf(buf, "%u\n", cur_freq);
505 * show_scaling_governor - show the current policy for the specified CPU
507 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
509 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
510 return sprintf(buf, "powersave\n");
511 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
512 return sprintf(buf, "performance\n");
513 else if (policy->governor)
514 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
515 policy->governor->name);
520 * store_scaling_governor - store policy for the specified CPU
522 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
523 const char *buf, size_t count)
526 char str_governor[16];
527 struct cpufreq_policy new_policy;
529 ret = cpufreq_get_policy(&new_policy, policy->cpu);
533 ret = sscanf(buf, "%15s", str_governor);
537 if (cpufreq_parse_governor(str_governor, &new_policy.policy,
538 &new_policy.governor))
542 * Do not use cpufreq_set_policy here or the user_policy.max
543 * will be wrongly overridden
545 ret = __cpufreq_set_policy(policy, &new_policy);
547 policy->user_policy.policy = policy->policy;
548 policy->user_policy.governor = policy->governor;
557 * show_scaling_driver - show the cpufreq driver currently loaded
559 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
561 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
565 * show_scaling_available_governors - show the available CPUfreq governors
567 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
571 struct cpufreq_governor *t;
573 if (!cpufreq_driver->target) {
574 i += sprintf(buf, "performance powersave");
578 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
579 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
580 - (CPUFREQ_NAME_LEN + 2)))
582 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
585 i += sprintf(&buf[i], "\n");
589 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
594 for_each_cpu(cpu, mask) {
596 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
597 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
598 if (i >= (PAGE_SIZE - 5))
601 i += sprintf(&buf[i], "\n");
604 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
607 * show_related_cpus - show the CPUs affected by each transition even if
608 * hw coordination is in use
610 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
612 return cpufreq_show_cpus(policy->related_cpus, buf);
616 * show_affected_cpus - show the CPUs affected by each transition
618 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
620 return cpufreq_show_cpus(policy->cpus, buf);
623 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
624 const char *buf, size_t count)
626 unsigned int freq = 0;
629 if (!policy->governor || !policy->governor->store_setspeed)
632 ret = sscanf(buf, "%u", &freq);
636 policy->governor->store_setspeed(policy, freq);
641 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
643 if (!policy->governor || !policy->governor->show_setspeed)
644 return sprintf(buf, "<unsupported>\n");
646 return policy->governor->show_setspeed(policy, buf);
650 * show_bios_limit - show the current cpufreq HW/BIOS limitation
652 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
656 if (cpufreq_driver->bios_limit) {
657 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
659 return sprintf(buf, "%u\n", limit);
661 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
664 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
665 cpufreq_freq_attr_ro(cpuinfo_min_freq);
666 cpufreq_freq_attr_ro(cpuinfo_max_freq);
667 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
668 cpufreq_freq_attr_ro(scaling_available_governors);
669 cpufreq_freq_attr_ro(scaling_driver);
670 cpufreq_freq_attr_ro(scaling_cur_freq);
671 cpufreq_freq_attr_ro(bios_limit);
672 cpufreq_freq_attr_ro(related_cpus);
673 cpufreq_freq_attr_ro(affected_cpus);
674 cpufreq_freq_attr_rw(scaling_min_freq);
675 cpufreq_freq_attr_rw(scaling_max_freq);
676 cpufreq_freq_attr_rw(scaling_governor);
677 cpufreq_freq_attr_rw(scaling_setspeed);
679 static struct attribute *default_attrs[] = {
680 &cpuinfo_min_freq.attr,
681 &cpuinfo_max_freq.attr,
682 &cpuinfo_transition_latency.attr,
683 &scaling_min_freq.attr,
684 &scaling_max_freq.attr,
687 &scaling_governor.attr,
688 &scaling_driver.attr,
689 &scaling_available_governors.attr,
690 &scaling_setspeed.attr,
694 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
695 #define to_attr(a) container_of(a, struct freq_attr, attr)
697 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
699 struct cpufreq_policy *policy = to_policy(kobj);
700 struct freq_attr *fattr = to_attr(attr);
701 ssize_t ret = -EINVAL;
702 policy = cpufreq_cpu_get_sysfs(policy->cpu);
706 if (lock_policy_rwsem_read(policy->cpu) < 0)
710 ret = fattr->show(policy, buf);
714 unlock_policy_rwsem_read(policy->cpu);
716 cpufreq_cpu_put_sysfs(policy);
721 static ssize_t store(struct kobject *kobj, struct attribute *attr,
722 const char *buf, size_t count)
724 struct cpufreq_policy *policy = to_policy(kobj);
725 struct freq_attr *fattr = to_attr(attr);
726 ssize_t ret = -EINVAL;
727 policy = cpufreq_cpu_get_sysfs(policy->cpu);
731 if (lock_policy_rwsem_write(policy->cpu) < 0)
735 ret = fattr->store(policy, buf, count);
739 unlock_policy_rwsem_write(policy->cpu);
741 cpufreq_cpu_put_sysfs(policy);
746 static void cpufreq_sysfs_release(struct kobject *kobj)
748 struct cpufreq_policy *policy = to_policy(kobj);
749 pr_debug("last reference is dropped\n");
750 complete(&policy->kobj_unregister);
753 static const struct sysfs_ops sysfs_ops = {
758 static struct kobj_type ktype_cpufreq = {
759 .sysfs_ops = &sysfs_ops,
760 .default_attrs = default_attrs,
761 .release = cpufreq_sysfs_release,
764 struct kobject *cpufreq_global_kobject;
765 EXPORT_SYMBOL(cpufreq_global_kobject);
767 static int cpufreq_global_kobject_usage;
769 int cpufreq_get_global_kobject(void)
771 if (!cpufreq_global_kobject_usage++)
772 return kobject_add(cpufreq_global_kobject,
773 &cpu_subsys.dev_root->kobj, "%s", "cpufreq");
777 EXPORT_SYMBOL(cpufreq_get_global_kobject);
779 void cpufreq_put_global_kobject(void)
781 if (!--cpufreq_global_kobject_usage)
782 kobject_del(cpufreq_global_kobject);
784 EXPORT_SYMBOL(cpufreq_put_global_kobject);
786 int cpufreq_sysfs_create_file(const struct attribute *attr)
788 int ret = cpufreq_get_global_kobject();
791 ret = sysfs_create_file(cpufreq_global_kobject, attr);
793 cpufreq_put_global_kobject();
798 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
800 void cpufreq_sysfs_remove_file(const struct attribute *attr)
802 sysfs_remove_file(cpufreq_global_kobject, attr);
803 cpufreq_put_global_kobject();
805 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
807 /* symlink affected CPUs */
808 static int cpufreq_add_dev_symlink(unsigned int cpu,
809 struct cpufreq_policy *policy)
814 for_each_cpu(j, policy->cpus) {
815 struct cpufreq_policy *managed_policy;
816 struct device *cpu_dev;
821 pr_debug("CPU %u already managed, adding link\n", j);
822 managed_policy = cpufreq_cpu_get(cpu);
823 cpu_dev = get_cpu_device(j);
824 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
827 cpufreq_cpu_put(managed_policy);
834 static int cpufreq_add_dev_interface(unsigned int cpu,
835 struct cpufreq_policy *policy,
838 struct cpufreq_policy new_policy;
839 struct freq_attr **drv_attr;
844 /* prepare interface data */
845 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
846 &dev->kobj, "cpufreq");
850 /* set up files for this cpu device */
851 drv_attr = cpufreq_driver->attr;
852 while ((drv_attr) && (*drv_attr)) {
853 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
855 goto err_out_kobj_put;
858 if (cpufreq_driver->get) {
859 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
861 goto err_out_kobj_put;
863 if (cpufreq_driver->target) {
864 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
866 goto err_out_kobj_put;
868 if (cpufreq_driver->bios_limit) {
869 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
871 goto err_out_kobj_put;
874 write_lock_irqsave(&cpufreq_driver_lock, flags);
875 for_each_cpu(j, policy->cpus) {
876 per_cpu(cpufreq_cpu_data, j) = policy;
877 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
879 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
881 ret = cpufreq_add_dev_symlink(cpu, policy);
883 goto err_out_kobj_put;
885 memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
886 /* assure that the starting sequence is run in __cpufreq_set_policy */
887 policy->governor = NULL;
889 /* set default policy */
890 ret = __cpufreq_set_policy(policy, &new_policy);
891 policy->user_policy.policy = policy->policy;
892 policy->user_policy.governor = policy->governor;
895 pr_debug("setting policy failed\n");
896 if (cpufreq_driver->exit)
897 cpufreq_driver->exit(policy);
902 kobject_put(&policy->kobj);
903 wait_for_completion(&policy->kobj_unregister);
907 #ifdef CONFIG_HOTPLUG_CPU
908 static int cpufreq_add_policy_cpu(unsigned int cpu, unsigned int sibling,
911 struct cpufreq_policy *policy;
912 int ret = 0, has_target = !!cpufreq_driver->target;
915 policy = cpufreq_cpu_get(sibling);
919 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
921 lock_policy_rwsem_write(sibling);
923 write_lock_irqsave(&cpufreq_driver_lock, flags);
925 cpumask_set_cpu(cpu, policy->cpus);
926 per_cpu(cpufreq_policy_cpu, cpu) = policy->cpu;
927 per_cpu(cpufreq_cpu_data, cpu) = policy;
928 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
930 unlock_policy_rwsem_write(sibling);
933 __cpufreq_governor(policy, CPUFREQ_GOV_START);
934 __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
937 ret = sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
939 cpufreq_cpu_put(policy);
948 * cpufreq_add_dev - add a CPU device
950 * Adds the cpufreq interface for a CPU device.
952 * The Oracle says: try running cpufreq registration/unregistration concurrently
953 * with with cpu hotplugging and all hell will break loose. Tried to clean this
954 * mess up, but more thorough testing is needed. - Mathieu
956 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
958 unsigned int j, cpu = dev->id;
960 struct cpufreq_policy *policy;
962 #ifdef CONFIG_HOTPLUG_CPU
963 struct cpufreq_governor *gov;
967 if (cpu_is_offline(cpu))
970 pr_debug("adding CPU %u\n", cpu);
973 /* check whether a different CPU already registered this
974 * CPU because it is in the same boat. */
975 policy = cpufreq_cpu_get(cpu);
976 if (unlikely(policy)) {
977 cpufreq_cpu_put(policy);
981 #ifdef CONFIG_HOTPLUG_CPU
982 /* Check if this cpu was hot-unplugged earlier and has siblings */
983 read_lock_irqsave(&cpufreq_driver_lock, flags);
984 for_each_online_cpu(sibling) {
985 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
986 if (cp && cpumask_test_cpu(cpu, cp->related_cpus)) {
987 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
988 return cpufreq_add_policy_cpu(cpu, sibling, dev);
991 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
995 if (!try_module_get(cpufreq_driver->owner)) {
1000 policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
1004 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1005 goto err_free_policy;
1007 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1008 goto err_free_cpumask;
1011 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
1012 cpumask_copy(policy->cpus, cpumask_of(cpu));
1014 /* Initially set CPU itself as the policy_cpu */
1015 per_cpu(cpufreq_policy_cpu, cpu) = cpu;
1017 init_completion(&policy->kobj_unregister);
1018 INIT_WORK(&policy->update, handle_update);
1020 /* call driver. From then on the cpufreq must be able
1021 * to accept all calls to ->verify and ->setpolicy for this CPU
1023 ret = cpufreq_driver->init(policy);
1025 pr_debug("initialization failed\n");
1026 goto err_set_policy_cpu;
1029 /* related cpus should atleast have policy->cpus */
1030 cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1033 * affected cpus must always be the one, which are online. We aren't
1034 * managing offline cpus here.
1036 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1038 policy->user_policy.min = policy->min;
1039 policy->user_policy.max = policy->max;
1041 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1042 CPUFREQ_START, policy);
1044 #ifdef CONFIG_HOTPLUG_CPU
1045 gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
1047 policy->governor = gov;
1048 pr_debug("Restoring governor %s for cpu %d\n",
1049 policy->governor->name, cpu);
1053 ret = cpufreq_add_dev_interface(cpu, policy, dev);
1055 goto err_out_unregister;
1057 kobject_uevent(&policy->kobj, KOBJ_ADD);
1058 module_put(cpufreq_driver->owner);
1059 pr_debug("initialization complete\n");
1064 write_lock_irqsave(&cpufreq_driver_lock, flags);
1065 for_each_cpu(j, policy->cpus)
1066 per_cpu(cpufreq_cpu_data, j) = NULL;
1067 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1069 kobject_put(&policy->kobj);
1070 wait_for_completion(&policy->kobj_unregister);
1073 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1074 free_cpumask_var(policy->related_cpus);
1076 free_cpumask_var(policy->cpus);
1080 module_put(cpufreq_driver->owner);
1085 static void update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1089 policy->last_cpu = policy->cpu;
1092 for_each_cpu(j, policy->cpus)
1093 per_cpu(cpufreq_policy_cpu, j) = cpu;
1095 #ifdef CONFIG_CPU_FREQ_TABLE
1096 cpufreq_frequency_table_update_policy_cpu(policy);
1098 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1099 CPUFREQ_UPDATE_POLICY_CPU, policy);
1103 * __cpufreq_remove_dev - remove a CPU device
1105 * Removes the cpufreq interface for a CPU device.
1106 * Caller should already have policy_rwsem in write mode for this CPU.
1107 * This routine frees the rwsem before returning.
1109 static int __cpufreq_remove_dev(struct device *dev,
1110 struct subsys_interface *sif)
1112 unsigned int cpu = dev->id, ret, cpus;
1113 unsigned long flags;
1114 struct cpufreq_policy *data;
1115 struct kobject *kobj;
1116 struct completion *cmp;
1117 struct device *cpu_dev;
1119 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1121 write_lock_irqsave(&cpufreq_driver_lock, flags);
1123 data = per_cpu(cpufreq_cpu_data, cpu);
1124 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1126 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1129 pr_debug("%s: No cpu_data found\n", __func__);
1133 if (cpufreq_driver->target)
1134 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1136 #ifdef CONFIG_HOTPLUG_CPU
1137 if (!cpufreq_driver->setpolicy)
1138 strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1139 data->governor->name, CPUFREQ_NAME_LEN);
1142 WARN_ON(lock_policy_rwsem_write(cpu));
1143 cpus = cpumask_weight(data->cpus);
1146 cpumask_clear_cpu(cpu, data->cpus);
1147 unlock_policy_rwsem_write(cpu);
1149 if (cpu != data->cpu) {
1150 sysfs_remove_link(&dev->kobj, "cpufreq");
1151 } else if (cpus > 1) {
1152 /* first sibling now owns the new sysfs dir */
1153 cpu_dev = get_cpu_device(cpumask_first(data->cpus));
1154 sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1155 ret = kobject_move(&data->kobj, &cpu_dev->kobj);
1157 pr_err("%s: Failed to move kobj: %d", __func__, ret);
1159 WARN_ON(lock_policy_rwsem_write(cpu));
1160 cpumask_set_cpu(cpu, data->cpus);
1162 write_lock_irqsave(&cpufreq_driver_lock, flags);
1163 per_cpu(cpufreq_cpu_data, cpu) = data;
1164 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1166 unlock_policy_rwsem_write(cpu);
1168 ret = sysfs_create_link(&cpu_dev->kobj, &data->kobj,
1173 WARN_ON(lock_policy_rwsem_write(cpu));
1174 update_policy_cpu(data, cpu_dev->id);
1175 unlock_policy_rwsem_write(cpu);
1176 pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1177 __func__, cpu_dev->id, cpu);
1180 if ((cpus == 1) && (cpufreq_driver->target))
1181 __cpufreq_governor(data, CPUFREQ_GOV_POLICY_EXIT);
1183 pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
1184 cpufreq_cpu_put(data);
1186 /* If cpu is last user of policy, free policy */
1188 lock_policy_rwsem_read(cpu);
1190 cmp = &data->kobj_unregister;
1191 unlock_policy_rwsem_read(cpu);
1194 /* we need to make sure that the underlying kobj is actually
1195 * not referenced anymore by anybody before we proceed with
1198 pr_debug("waiting for dropping of refcount\n");
1199 wait_for_completion(cmp);
1200 pr_debug("wait complete\n");
1202 if (cpufreq_driver->exit)
1203 cpufreq_driver->exit(data);
1205 free_cpumask_var(data->related_cpus);
1206 free_cpumask_var(data->cpus);
1208 } else if (cpufreq_driver->target) {
1209 __cpufreq_governor(data, CPUFREQ_GOV_START);
1210 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1213 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1217 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1219 unsigned int cpu = dev->id;
1222 if (cpu_is_offline(cpu))
1225 retval = __cpufreq_remove_dev(dev, sif);
1229 static void handle_update(struct work_struct *work)
1231 struct cpufreq_policy *policy =
1232 container_of(work, struct cpufreq_policy, update);
1233 unsigned int cpu = policy->cpu;
1234 pr_debug("handle_update for cpu %u called\n", cpu);
1235 cpufreq_update_policy(cpu);
1239 * cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1242 * @old_freq: CPU frequency the kernel thinks the CPU runs at
1243 * @new_freq: CPU frequency the CPU actually runs at
1245 * We adjust to current frequency first, and need to clean up later.
1246 * So either call to cpufreq_update_policy() or schedule handle_update()).
1248 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1249 unsigned int new_freq)
1251 struct cpufreq_policy *policy;
1252 struct cpufreq_freqs freqs;
1253 unsigned long flags;
1255 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1256 "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1258 freqs.old = old_freq;
1259 freqs.new = new_freq;
1261 read_lock_irqsave(&cpufreq_driver_lock, flags);
1262 policy = per_cpu(cpufreq_cpu_data, cpu);
1263 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1265 cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
1266 cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
1270 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1273 * This is the last known freq, without actually getting it from the driver.
1274 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1276 unsigned int cpufreq_quick_get(unsigned int cpu)
1278 struct cpufreq_policy *policy;
1279 unsigned int ret_freq = 0;
1281 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1282 return cpufreq_driver->get(cpu);
1284 policy = cpufreq_cpu_get(cpu);
1286 ret_freq = policy->cur;
1287 cpufreq_cpu_put(policy);
1292 EXPORT_SYMBOL(cpufreq_quick_get);
1295 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1298 * Just return the max possible frequency for a given CPU.
1300 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1302 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1303 unsigned int ret_freq = 0;
1306 ret_freq = policy->max;
1307 cpufreq_cpu_put(policy);
1312 EXPORT_SYMBOL(cpufreq_quick_get_max);
1314 static unsigned int __cpufreq_get(unsigned int cpu)
1316 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1317 unsigned int ret_freq = 0;
1319 if (!cpufreq_driver->get)
1322 ret_freq = cpufreq_driver->get(cpu);
1324 if (ret_freq && policy->cur &&
1325 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1326 /* verify no discrepancy between actual and
1327 saved value exists */
1328 if (unlikely(ret_freq != policy->cur)) {
1329 cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1330 schedule_work(&policy->update);
1338 * cpufreq_get - get the current CPU frequency (in kHz)
1341 * Get the CPU current (static) CPU frequency
1343 unsigned int cpufreq_get(unsigned int cpu)
1345 unsigned int ret_freq = 0;
1346 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1351 if (unlikely(lock_policy_rwsem_read(cpu)))
1354 ret_freq = __cpufreq_get(cpu);
1356 unlock_policy_rwsem_read(cpu);
1359 cpufreq_cpu_put(policy);
1363 EXPORT_SYMBOL(cpufreq_get);
1365 static struct subsys_interface cpufreq_interface = {
1367 .subsys = &cpu_subsys,
1368 .add_dev = cpufreq_add_dev,
1369 .remove_dev = cpufreq_remove_dev,
1373 * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1375 * This function is only executed for the boot processor. The other CPUs
1376 * have been put offline by means of CPU hotplug.
1378 static int cpufreq_bp_suspend(void)
1382 int cpu = smp_processor_id();
1383 struct cpufreq_policy *cpu_policy;
1385 pr_debug("suspending cpu %u\n", cpu);
1387 /* If there's no policy for the boot CPU, we have nothing to do. */
1388 cpu_policy = cpufreq_cpu_get(cpu);
1392 if (cpufreq_driver->suspend) {
1393 ret = cpufreq_driver->suspend(cpu_policy);
1395 printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1396 "step on CPU %u\n", cpu_policy->cpu);
1399 cpufreq_cpu_put(cpu_policy);
1404 * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1406 * 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1407 * 2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1408 * restored. It will verify that the current freq is in sync with
1409 * what we believe it to be. This is a bit later than when it
1410 * should be, but nonethteless it's better than calling
1411 * cpufreq_driver->get() here which might re-enable interrupts...
1413 * This function is only executed for the boot CPU. The other CPUs have not
1414 * been turned on yet.
1416 static void cpufreq_bp_resume(void)
1420 int cpu = smp_processor_id();
1421 struct cpufreq_policy *cpu_policy;
1423 pr_debug("resuming cpu %u\n", cpu);
1425 /* If there's no policy for the boot CPU, we have nothing to do. */
1426 cpu_policy = cpufreq_cpu_get(cpu);
1430 if (cpufreq_driver->resume) {
1431 ret = cpufreq_driver->resume(cpu_policy);
1433 printk(KERN_ERR "cpufreq: resume failed in ->resume "
1434 "step on CPU %u\n", cpu_policy->cpu);
1439 schedule_work(&cpu_policy->update);
1442 cpufreq_cpu_put(cpu_policy);
1445 static struct syscore_ops cpufreq_syscore_ops = {
1446 .suspend = cpufreq_bp_suspend,
1447 .resume = cpufreq_bp_resume,
1451 * cpufreq_get_current_driver - return current driver's name
1453 * Return the name string of the currently loaded cpufreq driver
1456 const char *cpufreq_get_current_driver(void)
1459 return cpufreq_driver->name;
1463 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1465 /*********************************************************************
1466 * NOTIFIER LISTS INTERFACE *
1467 *********************************************************************/
1470 * cpufreq_register_notifier - register a driver with cpufreq
1471 * @nb: notifier function to register
1472 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1474 * Add a driver to one of two lists: either a list of drivers that
1475 * are notified about clock rate changes (once before and once after
1476 * the transition), or a list of drivers that are notified about
1477 * changes in cpufreq policy.
1479 * This function may sleep, and has the same return conditions as
1480 * blocking_notifier_chain_register.
1482 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1486 if (cpufreq_disabled())
1489 WARN_ON(!init_cpufreq_transition_notifier_list_called);
1492 case CPUFREQ_TRANSITION_NOTIFIER:
1493 ret = srcu_notifier_chain_register(
1494 &cpufreq_transition_notifier_list, nb);
1496 case CPUFREQ_POLICY_NOTIFIER:
1497 ret = blocking_notifier_chain_register(
1498 &cpufreq_policy_notifier_list, nb);
1506 EXPORT_SYMBOL(cpufreq_register_notifier);
1509 * cpufreq_unregister_notifier - unregister a driver with cpufreq
1510 * @nb: notifier block to be unregistered
1511 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1513 * Remove a driver from the CPU frequency notifier list.
1515 * This function may sleep, and has the same return conditions as
1516 * blocking_notifier_chain_unregister.
1518 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1522 if (cpufreq_disabled())
1526 case CPUFREQ_TRANSITION_NOTIFIER:
1527 ret = srcu_notifier_chain_unregister(
1528 &cpufreq_transition_notifier_list, nb);
1530 case CPUFREQ_POLICY_NOTIFIER:
1531 ret = blocking_notifier_chain_unregister(
1532 &cpufreq_policy_notifier_list, nb);
1540 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1543 /*********************************************************************
1545 *********************************************************************/
1547 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1548 unsigned int target_freq,
1549 unsigned int relation)
1551 int retval = -EINVAL;
1552 unsigned int old_target_freq = target_freq;
1554 if (cpufreq_disabled())
1556 if (policy->transition_ongoing)
1559 /* Make sure that target_freq is within supported range */
1560 if (target_freq > policy->max)
1561 target_freq = policy->max;
1562 if (target_freq < policy->min)
1563 target_freq = policy->min;
1565 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1566 policy->cpu, target_freq, relation, old_target_freq);
1568 if (target_freq == policy->cur)
1571 if (cpufreq_driver->target)
1572 retval = cpufreq_driver->target(policy, target_freq, relation);
1576 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1578 int cpufreq_driver_target(struct cpufreq_policy *policy,
1579 unsigned int target_freq,
1580 unsigned int relation)
1584 if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1587 ret = __cpufreq_driver_target(policy, target_freq, relation);
1589 unlock_policy_rwsem_write(policy->cpu);
1594 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1597 * when "event" is CPUFREQ_GOV_LIMITS
1600 static int __cpufreq_governor(struct cpufreq_policy *policy,
1605 /* Only must be defined when default governor is known to have latency
1606 restrictions, like e.g. conservative or ondemand.
1607 That this is the case is already ensured in Kconfig
1609 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1610 struct cpufreq_governor *gov = &cpufreq_gov_performance;
1612 struct cpufreq_governor *gov = NULL;
1615 if (policy->governor->max_transition_latency &&
1616 policy->cpuinfo.transition_latency >
1617 policy->governor->max_transition_latency) {
1621 printk(KERN_WARNING "%s governor failed, too long"
1622 " transition latency of HW, fallback"
1623 " to %s governor\n",
1624 policy->governor->name,
1626 policy->governor = gov;
1630 if (!try_module_get(policy->governor->owner))
1633 pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1634 policy->cpu, event);
1636 mutex_lock(&cpufreq_governor_lock);
1637 if ((!policy->governor_enabled && (event == CPUFREQ_GOV_STOP)) ||
1638 (policy->governor_enabled && (event == CPUFREQ_GOV_START))) {
1639 mutex_unlock(&cpufreq_governor_lock);
1643 if (event == CPUFREQ_GOV_STOP)
1644 policy->governor_enabled = false;
1645 else if (event == CPUFREQ_GOV_START)
1646 policy->governor_enabled = true;
1648 mutex_unlock(&cpufreq_governor_lock);
1650 ret = policy->governor->governor(policy, event);
1653 if (event == CPUFREQ_GOV_POLICY_INIT)
1654 policy->governor->initialized++;
1655 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1656 policy->governor->initialized--;
1658 /* Restore original values */
1659 mutex_lock(&cpufreq_governor_lock);
1660 if (event == CPUFREQ_GOV_STOP)
1661 policy->governor_enabled = true;
1662 else if (event == CPUFREQ_GOV_START)
1663 policy->governor_enabled = false;
1664 mutex_unlock(&cpufreq_governor_lock);
1667 /* we keep one module reference alive for
1668 each CPU governed by this CPU */
1669 if ((event != CPUFREQ_GOV_START) || ret)
1670 module_put(policy->governor->owner);
1671 if ((event == CPUFREQ_GOV_STOP) && !ret)
1672 module_put(policy->governor->owner);
1677 int cpufreq_register_governor(struct cpufreq_governor *governor)
1684 if (cpufreq_disabled())
1687 mutex_lock(&cpufreq_governor_mutex);
1689 governor->initialized = 0;
1691 if (__find_governor(governor->name) == NULL) {
1693 list_add(&governor->governor_list, &cpufreq_governor_list);
1696 mutex_unlock(&cpufreq_governor_mutex);
1699 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1701 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1703 #ifdef CONFIG_HOTPLUG_CPU
1710 if (cpufreq_disabled())
1713 #ifdef CONFIG_HOTPLUG_CPU
1714 for_each_present_cpu(cpu) {
1715 if (cpu_online(cpu))
1717 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1718 strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1722 mutex_lock(&cpufreq_governor_mutex);
1723 list_del(&governor->governor_list);
1724 mutex_unlock(&cpufreq_governor_mutex);
1727 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1730 /*********************************************************************
1731 * POLICY INTERFACE *
1732 *********************************************************************/
1735 * cpufreq_get_policy - get the current cpufreq_policy
1736 * @policy: struct cpufreq_policy into which the current cpufreq_policy
1739 * Reads the current cpufreq policy.
1741 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1743 struct cpufreq_policy *cpu_policy;
1747 cpu_policy = cpufreq_cpu_get(cpu);
1751 memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1753 cpufreq_cpu_put(cpu_policy);
1756 EXPORT_SYMBOL(cpufreq_get_policy);
1759 * data : current policy.
1760 * policy : policy to be set.
1762 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1763 struct cpufreq_policy *policy)
1765 int ret = 0, failed = 1;
1767 pr_debug("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1768 policy->min, policy->max);
1770 memcpy(&policy->cpuinfo, &data->cpuinfo,
1771 sizeof(struct cpufreq_cpuinfo));
1773 if (policy->min > data->max || policy->max < data->min) {
1778 /* verify the cpu speed can be set within this limit */
1779 ret = cpufreq_driver->verify(policy);
1783 /* adjust if necessary - all reasons */
1784 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1785 CPUFREQ_ADJUST, policy);
1787 /* adjust if necessary - hardware incompatibility*/
1788 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1789 CPUFREQ_INCOMPATIBLE, policy);
1792 * verify the cpu speed can be set within this limit, which might be
1793 * different to the first one
1795 ret = cpufreq_driver->verify(policy);
1799 /* notification of the new policy */
1800 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1801 CPUFREQ_NOTIFY, policy);
1803 data->min = policy->min;
1804 data->max = policy->max;
1806 pr_debug("new min and max freqs are %u - %u kHz\n",
1807 data->min, data->max);
1809 if (cpufreq_driver->setpolicy) {
1810 data->policy = policy->policy;
1811 pr_debug("setting range\n");
1812 ret = cpufreq_driver->setpolicy(policy);
1814 if (policy->governor != data->governor) {
1815 /* save old, working values */
1816 struct cpufreq_governor *old_gov = data->governor;
1818 pr_debug("governor switch\n");
1820 /* end old governor */
1821 if (data->governor) {
1822 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1823 unlock_policy_rwsem_write(policy->cpu);
1824 __cpufreq_governor(data,
1825 CPUFREQ_GOV_POLICY_EXIT);
1826 lock_policy_rwsem_write(policy->cpu);
1829 /* start new governor */
1830 data->governor = policy->governor;
1831 if (!__cpufreq_governor(data, CPUFREQ_GOV_POLICY_INIT)) {
1832 if (!__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1835 unlock_policy_rwsem_write(policy->cpu);
1836 __cpufreq_governor(data,
1837 CPUFREQ_GOV_POLICY_EXIT);
1838 lock_policy_rwsem_write(policy->cpu);
1843 /* new governor failed, so re-start old one */
1844 pr_debug("starting governor %s failed\n",
1845 data->governor->name);
1847 data->governor = old_gov;
1848 __cpufreq_governor(data,
1849 CPUFREQ_GOV_POLICY_INIT);
1850 __cpufreq_governor(data,
1856 /* might be a policy change, too, so fall through */
1858 pr_debug("governor: change or update limits\n");
1859 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1867 * cpufreq_update_policy - re-evaluate an existing cpufreq policy
1868 * @cpu: CPU which shall be re-evaluated
1870 * Useful for policy notifiers which have different necessities
1871 * at different times.
1873 int cpufreq_update_policy(unsigned int cpu)
1875 struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1876 struct cpufreq_policy policy;
1884 if (unlikely(lock_policy_rwsem_write(cpu))) {
1889 pr_debug("updating policy for CPU %u\n", cpu);
1890 memcpy(&policy, data, sizeof(struct cpufreq_policy));
1891 policy.min = data->user_policy.min;
1892 policy.max = data->user_policy.max;
1893 policy.policy = data->user_policy.policy;
1894 policy.governor = data->user_policy.governor;
1897 * BIOS might change freq behind our back
1898 * -> ask driver for current freq and notify governors about a change
1900 if (cpufreq_driver->get) {
1901 policy.cur = cpufreq_driver->get(cpu);
1903 pr_debug("Driver did not initialize current freq");
1904 data->cur = policy.cur;
1906 if (data->cur != policy.cur && cpufreq_driver->target)
1907 cpufreq_out_of_sync(cpu, data->cur,
1912 ret = __cpufreq_set_policy(data, &policy);
1914 unlock_policy_rwsem_write(cpu);
1917 cpufreq_cpu_put(data);
1921 EXPORT_SYMBOL(cpufreq_update_policy);
1923 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1924 unsigned long action, void *hcpu)
1926 unsigned int cpu = (unsigned long)hcpu;
1929 dev = get_cpu_device(cpu);
1933 case CPU_ONLINE_FROZEN:
1934 cpufreq_add_dev(dev, NULL);
1936 case CPU_DOWN_PREPARE:
1937 case CPU_DOWN_PREPARE_FROZEN:
1938 __cpufreq_remove_dev(dev, NULL);
1940 case CPU_DOWN_FAILED:
1941 case CPU_DOWN_FAILED_FROZEN:
1942 cpufreq_add_dev(dev, NULL);
1949 static struct notifier_block __refdata cpufreq_cpu_notifier = {
1950 .notifier_call = cpufreq_cpu_callback,
1953 /*********************************************************************
1954 * REGISTER / UNREGISTER CPUFREQ DRIVER *
1955 *********************************************************************/
1958 * cpufreq_register_driver - register a CPU Frequency driver
1959 * @driver_data: A struct cpufreq_driver containing the values#
1960 * submitted by the CPU Frequency driver.
1962 * Registers a CPU Frequency driver to this core code. This code
1963 * returns zero on success, -EBUSY when another driver got here first
1964 * (and isn't unregistered in the meantime).
1967 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1969 unsigned long flags;
1972 if (cpufreq_disabled())
1975 if (!driver_data || !driver_data->verify || !driver_data->init ||
1976 ((!driver_data->setpolicy) && (!driver_data->target)))
1979 pr_debug("trying to register driver %s\n", driver_data->name);
1981 if (driver_data->setpolicy)
1982 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1984 write_lock_irqsave(&cpufreq_driver_lock, flags);
1985 if (cpufreq_driver) {
1986 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1989 cpufreq_driver = driver_data;
1990 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1992 ret = subsys_interface_register(&cpufreq_interface);
1994 goto err_null_driver;
1996 if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
2000 /* check for at least one working CPU */
2001 for (i = 0; i < nr_cpu_ids; i++)
2002 if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
2007 /* if all ->init() calls failed, unregister */
2009 pr_debug("no CPU initialized for driver %s\n",
2015 register_hotcpu_notifier(&cpufreq_cpu_notifier);
2016 pr_debug("driver %s up and running\n", driver_data->name);
2020 subsys_interface_unregister(&cpufreq_interface);
2022 write_lock_irqsave(&cpufreq_driver_lock, flags);
2023 cpufreq_driver = NULL;
2024 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2027 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2030 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2032 * Unregister the current CPUFreq driver. Only call this if you have
2033 * the right to do so, i.e. if you have succeeded in initialising before!
2034 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2035 * currently not initialised.
2037 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2039 unsigned long flags;
2041 if (!cpufreq_driver || (driver != cpufreq_driver))
2044 pr_debug("unregistering driver %s\n", driver->name);
2046 subsys_interface_unregister(&cpufreq_interface);
2047 unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2049 write_lock_irqsave(&cpufreq_driver_lock, flags);
2050 cpufreq_driver = NULL;
2051 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2055 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2057 static int __init cpufreq_core_init(void)
2061 if (cpufreq_disabled())
2064 for_each_possible_cpu(cpu) {
2065 per_cpu(cpufreq_policy_cpu, cpu) = -1;
2066 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
2069 cpufreq_global_kobject = kobject_create();
2070 BUG_ON(!cpufreq_global_kobject);
2071 register_syscore_ops(&cpufreq_syscore_ops);
2075 core_initcall(cpufreq_core_init);