2 * linux/drivers/cpufreq/cpufreq.c
4 * Copyright (C) 2001 Russell King
5 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
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
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.
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 #ifdef CONFIG_HOTPLUG_CPU
48 /* This one keeps track of the previously set governor of a removed CPU */
49 static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
51 static DEFINE_RWLOCK(cpufreq_driver_lock);
52 static DEFINE_MUTEX(cpufreq_governor_lock);
55 * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
56 * all cpufreq/hotplug/workqueue/etc related lock issues.
58 * The rules for this semaphore:
59 * - Any routine that wants to read from the policy structure will
60 * do a down_read on this semaphore.
61 * - Any routine that will write to the policy structure and/or may take away
62 * the policy altogether (eg. CPU hotplug), will hold this lock in write
63 * mode before doing so.
66 * - Governor routines that can be called in cpufreq hotplug path should not
67 * take this sem as top level hotplug notifier handler takes this.
68 * - Lock should not be held across
69 * __cpufreq_governor(data, CPUFREQ_GOV_STOP);
71 static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
72 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
74 #define lock_policy_rwsem(mode, cpu) \
75 static int lock_policy_rwsem_##mode(int cpu) \
77 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
78 BUG_ON(policy_cpu == -1); \
79 down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
84 lock_policy_rwsem(read, cpu);
85 lock_policy_rwsem(write, cpu);
87 #define unlock_policy_rwsem(mode, cpu) \
88 static void unlock_policy_rwsem_##mode(int cpu) \
90 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
91 BUG_ON(policy_cpu == -1); \
92 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
95 unlock_policy_rwsem(read, cpu);
96 unlock_policy_rwsem(write, cpu);
98 /* internal prototypes */
99 static int __cpufreq_governor(struct cpufreq_policy *policy,
101 static unsigned int __cpufreq_get(unsigned int cpu);
102 static void handle_update(struct work_struct *work);
105 * Two notifier lists: the "policy" list is involved in the
106 * validation process for a new CPU frequency policy; the
107 * "transition" list for kernel code that needs to handle
108 * changes to devices when the CPU clock speed changes.
109 * The mutex locks both lists.
111 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
112 static struct srcu_notifier_head cpufreq_transition_notifier_list;
114 static bool init_cpufreq_transition_notifier_list_called;
115 static int __init init_cpufreq_transition_notifier_list(void)
117 srcu_init_notifier_head(&cpufreq_transition_notifier_list);
118 init_cpufreq_transition_notifier_list_called = true;
121 pure_initcall(init_cpufreq_transition_notifier_list);
123 static int off __read_mostly;
124 static int cpufreq_disabled(void)
128 void disable_cpufreq(void)
132 static LIST_HEAD(cpufreq_governor_list);
133 static DEFINE_MUTEX(cpufreq_governor_mutex);
135 bool have_governor_per_policy(void)
137 return cpufreq_driver->have_governor_per_policy;
139 EXPORT_SYMBOL_GPL(have_governor_per_policy);
141 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
143 if (have_governor_per_policy())
144 return &policy->kobj;
146 return cpufreq_global_kobject;
148 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
150 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
156 cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
158 busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
159 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
160 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
161 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
162 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
163 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
165 idle_time = cur_wall_time - busy_time;
167 *wall = cputime_to_usecs(cur_wall_time);
169 return cputime_to_usecs(idle_time);
172 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
174 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
176 if (idle_time == -1ULL)
177 return get_cpu_idle_time_jiffy(cpu, wall);
179 idle_time += get_cpu_iowait_time_us(cpu, wall);
183 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
185 static struct cpufreq_policy *__cpufreq_cpu_get(unsigned int cpu, bool sysfs)
187 struct cpufreq_policy *data;
190 if (cpu >= nr_cpu_ids)
193 /* get the cpufreq driver */
194 read_lock_irqsave(&cpufreq_driver_lock, flags);
199 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)
301 void __cpufreq_notify_transition(struct cpufreq_policy *policy,
302 struct cpufreq_freqs *freqs, unsigned int state)
304 BUG_ON(irqs_disabled());
306 if (cpufreq_disabled())
309 freqs->flags = cpufreq_driver->flags;
310 pr_debug("notification %u of frequency transition to %u kHz\n",
315 case CPUFREQ_PRECHANGE:
316 /* detect if the driver reported a value as "old frequency"
317 * which is not equal to what the cpufreq core thinks is
320 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
321 if ((policy) && (policy->cpu == freqs->cpu) &&
322 (policy->cur) && (policy->cur != freqs->old)) {
323 pr_debug("Warning: CPU frequency is"
324 " %u, cpufreq assumed %u kHz.\n",
325 freqs->old, policy->cur);
326 freqs->old = policy->cur;
329 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
330 CPUFREQ_PRECHANGE, freqs);
331 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
334 case CPUFREQ_POSTCHANGE:
335 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
336 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
337 (unsigned long)freqs->cpu);
338 trace_cpu_frequency(freqs->new, freqs->cpu);
339 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
340 CPUFREQ_POSTCHANGE, freqs);
341 if (likely(policy) && likely(policy->cpu == freqs->cpu))
342 policy->cur = freqs->new;
347 * cpufreq_notify_transition - call notifier chain and adjust_jiffies
348 * on frequency transition.
350 * This function calls the transition notifiers and the "adjust_jiffies"
351 * function. It is called twice on all CPU frequency changes that have
354 void cpufreq_notify_transition(struct cpufreq_policy *policy,
355 struct cpufreq_freqs *freqs, unsigned int state)
357 for_each_cpu(freqs->cpu, policy->cpus)
358 __cpufreq_notify_transition(policy, freqs, state);
360 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
364 /*********************************************************************
366 *********************************************************************/
368 static struct cpufreq_governor *__find_governor(const char *str_governor)
370 struct cpufreq_governor *t;
372 list_for_each_entry(t, &cpufreq_governor_list, governor_list)
373 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
380 * cpufreq_parse_governor - parse a governor string
382 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
383 struct cpufreq_governor **governor)
390 if (cpufreq_driver->setpolicy) {
391 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
392 *policy = CPUFREQ_POLICY_PERFORMANCE;
394 } else if (!strnicmp(str_governor, "powersave",
396 *policy = CPUFREQ_POLICY_POWERSAVE;
399 } else if (cpufreq_driver->target) {
400 struct cpufreq_governor *t;
402 mutex_lock(&cpufreq_governor_mutex);
404 t = __find_governor(str_governor);
409 mutex_unlock(&cpufreq_governor_mutex);
410 ret = request_module("cpufreq_%s", str_governor);
411 mutex_lock(&cpufreq_governor_mutex);
414 t = __find_governor(str_governor);
422 mutex_unlock(&cpufreq_governor_mutex);
430 * cpufreq_per_cpu_attr_read() / show_##file_name() -
431 * print out cpufreq information
433 * Write out information from cpufreq_driver->policy[cpu]; object must be
437 #define show_one(file_name, object) \
438 static ssize_t show_##file_name \
439 (struct cpufreq_policy *policy, char *buf) \
441 return sprintf(buf, "%u\n", policy->object); \
444 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
445 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
446 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
447 show_one(scaling_min_freq, min);
448 show_one(scaling_max_freq, max);
449 show_one(scaling_cur_freq, cur);
451 static int __cpufreq_set_policy(struct cpufreq_policy *data,
452 struct cpufreq_policy *policy);
455 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
457 #define store_one(file_name, object) \
458 static ssize_t store_##file_name \
459 (struct cpufreq_policy *policy, const char *buf, size_t count) \
462 struct cpufreq_policy new_policy; \
464 ret = cpufreq_get_policy(&new_policy, policy->cpu); \
468 ret = sscanf(buf, "%u", &new_policy.object); \
472 ret = __cpufreq_set_policy(policy, &new_policy); \
473 policy->user_policy.object = policy->object; \
475 return ret ? ret : count; \
478 store_one(scaling_min_freq, min);
479 store_one(scaling_max_freq, max);
482 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
484 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
487 unsigned int cur_freq = __cpufreq_get(policy->cpu);
489 return sprintf(buf, "<unknown>");
490 return sprintf(buf, "%u\n", cur_freq);
495 * show_scaling_governor - show the current policy for the specified CPU
497 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
499 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
500 return sprintf(buf, "powersave\n");
501 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
502 return sprintf(buf, "performance\n");
503 else if (policy->governor)
504 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
505 policy->governor->name);
511 * store_scaling_governor - store policy for the specified CPU
513 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
514 const char *buf, size_t count)
517 char str_governor[16];
518 struct cpufreq_policy new_policy;
520 ret = cpufreq_get_policy(&new_policy, policy->cpu);
524 ret = sscanf(buf, "%15s", str_governor);
528 if (cpufreq_parse_governor(str_governor, &new_policy.policy,
529 &new_policy.governor))
532 /* Do not use cpufreq_set_policy here or the user_policy.max
533 will be wrongly overridden */
534 ret = __cpufreq_set_policy(policy, &new_policy);
536 policy->user_policy.policy = policy->policy;
537 policy->user_policy.governor = policy->governor;
546 * show_scaling_driver - show the cpufreq driver currently loaded
548 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
550 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
554 * show_scaling_available_governors - show the available CPUfreq governors
556 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
560 struct cpufreq_governor *t;
562 if (!cpufreq_driver->target) {
563 i += sprintf(buf, "performance powersave");
567 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
568 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
569 - (CPUFREQ_NAME_LEN + 2)))
571 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
574 i += sprintf(&buf[i], "\n");
578 static ssize_t show_cpus(const struct cpumask *mask, char *buf)
583 for_each_cpu(cpu, mask) {
585 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
586 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
587 if (i >= (PAGE_SIZE - 5))
590 i += sprintf(&buf[i], "\n");
595 * show_related_cpus - show the CPUs affected by each transition even if
596 * hw coordination is in use
598 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
600 return show_cpus(policy->related_cpus, buf);
604 * show_affected_cpus - show the CPUs affected by each transition
606 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
608 return show_cpus(policy->cpus, buf);
611 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
612 const char *buf, size_t count)
614 unsigned int freq = 0;
617 if (!policy->governor || !policy->governor->store_setspeed)
620 ret = sscanf(buf, "%u", &freq);
624 policy->governor->store_setspeed(policy, freq);
629 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
631 if (!policy->governor || !policy->governor->show_setspeed)
632 return sprintf(buf, "<unsupported>\n");
634 return policy->governor->show_setspeed(policy, buf);
638 * show_bios_limit - show the current cpufreq HW/BIOS limitation
640 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
644 if (cpufreq_driver->bios_limit) {
645 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
647 return sprintf(buf, "%u\n", limit);
649 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
652 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
653 cpufreq_freq_attr_ro(cpuinfo_min_freq);
654 cpufreq_freq_attr_ro(cpuinfo_max_freq);
655 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
656 cpufreq_freq_attr_ro(scaling_available_governors);
657 cpufreq_freq_attr_ro(scaling_driver);
658 cpufreq_freq_attr_ro(scaling_cur_freq);
659 cpufreq_freq_attr_ro(bios_limit);
660 cpufreq_freq_attr_ro(related_cpus);
661 cpufreq_freq_attr_ro(affected_cpus);
662 cpufreq_freq_attr_rw(scaling_min_freq);
663 cpufreq_freq_attr_rw(scaling_max_freq);
664 cpufreq_freq_attr_rw(scaling_governor);
665 cpufreq_freq_attr_rw(scaling_setspeed);
667 static struct attribute *default_attrs[] = {
668 &cpuinfo_min_freq.attr,
669 &cpuinfo_max_freq.attr,
670 &cpuinfo_transition_latency.attr,
671 &scaling_min_freq.attr,
672 &scaling_max_freq.attr,
675 &scaling_governor.attr,
676 &scaling_driver.attr,
677 &scaling_available_governors.attr,
678 &scaling_setspeed.attr,
682 struct kobject *cpufreq_global_kobject;
683 EXPORT_SYMBOL(cpufreq_global_kobject);
685 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
686 #define to_attr(a) container_of(a, struct freq_attr, attr)
688 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
690 struct cpufreq_policy *policy = to_policy(kobj);
691 struct freq_attr *fattr = to_attr(attr);
692 ssize_t ret = -EINVAL;
693 policy = cpufreq_cpu_get_sysfs(policy->cpu);
697 if (lock_policy_rwsem_read(policy->cpu) < 0)
701 ret = fattr->show(policy, buf);
705 unlock_policy_rwsem_read(policy->cpu);
707 cpufreq_cpu_put_sysfs(policy);
712 static ssize_t store(struct kobject *kobj, struct attribute *attr,
713 const char *buf, size_t count)
715 struct cpufreq_policy *policy = to_policy(kobj);
716 struct freq_attr *fattr = to_attr(attr);
717 ssize_t ret = -EINVAL;
718 policy = cpufreq_cpu_get_sysfs(policy->cpu);
722 if (lock_policy_rwsem_write(policy->cpu) < 0)
726 ret = fattr->store(policy, buf, count);
730 unlock_policy_rwsem_write(policy->cpu);
732 cpufreq_cpu_put_sysfs(policy);
737 static void cpufreq_sysfs_release(struct kobject *kobj)
739 struct cpufreq_policy *policy = to_policy(kobj);
740 pr_debug("last reference is dropped\n");
741 complete(&policy->kobj_unregister);
744 static const struct sysfs_ops sysfs_ops = {
749 static struct kobj_type ktype_cpufreq = {
750 .sysfs_ops = &sysfs_ops,
751 .default_attrs = default_attrs,
752 .release = cpufreq_sysfs_release,
755 /* symlink affected CPUs */
756 static int cpufreq_add_dev_symlink(unsigned int cpu,
757 struct cpufreq_policy *policy)
762 for_each_cpu(j, policy->cpus) {
763 struct cpufreq_policy *managed_policy;
764 struct device *cpu_dev;
769 pr_debug("CPU %u already managed, adding link\n", j);
770 managed_policy = cpufreq_cpu_get(cpu);
771 cpu_dev = get_cpu_device(j);
772 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
775 cpufreq_cpu_put(managed_policy);
782 static int cpufreq_add_dev_interface(unsigned int cpu,
783 struct cpufreq_policy *policy,
786 struct cpufreq_policy new_policy;
787 struct freq_attr **drv_attr;
792 /* prepare interface data */
793 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
794 &dev->kobj, "cpufreq");
798 /* set up files for this cpu device */
799 drv_attr = cpufreq_driver->attr;
800 while ((drv_attr) && (*drv_attr)) {
801 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
803 goto err_out_kobj_put;
806 if (cpufreq_driver->get) {
807 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
809 goto err_out_kobj_put;
811 if (cpufreq_driver->target) {
812 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
814 goto err_out_kobj_put;
816 if (cpufreq_driver->bios_limit) {
817 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
819 goto err_out_kobj_put;
822 write_lock_irqsave(&cpufreq_driver_lock, flags);
823 for_each_cpu(j, policy->cpus) {
824 per_cpu(cpufreq_cpu_data, j) = policy;
825 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
827 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
829 ret = cpufreq_add_dev_symlink(cpu, policy);
831 goto err_out_kobj_put;
833 memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
834 /* assure that the starting sequence is run in __cpufreq_set_policy */
835 policy->governor = NULL;
837 /* set default policy */
838 ret = __cpufreq_set_policy(policy, &new_policy);
839 policy->user_policy.policy = policy->policy;
840 policy->user_policy.governor = policy->governor;
843 pr_debug("setting policy failed\n");
844 if (cpufreq_driver->exit)
845 cpufreq_driver->exit(policy);
850 kobject_put(&policy->kobj);
851 wait_for_completion(&policy->kobj_unregister);
855 #ifdef CONFIG_HOTPLUG_CPU
856 static int cpufreq_add_policy_cpu(unsigned int cpu, unsigned int sibling,
859 struct cpufreq_policy *policy;
860 int ret = 0, has_target = !!cpufreq_driver->target;
863 policy = cpufreq_cpu_get(sibling);
867 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
869 lock_policy_rwsem_write(sibling);
871 write_lock_irqsave(&cpufreq_driver_lock, flags);
873 cpumask_set_cpu(cpu, policy->cpus);
874 per_cpu(cpufreq_policy_cpu, cpu) = policy->cpu;
875 per_cpu(cpufreq_cpu_data, cpu) = policy;
876 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
878 unlock_policy_rwsem_write(sibling);
881 __cpufreq_governor(policy, CPUFREQ_GOV_START);
882 __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
885 ret = sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
887 cpufreq_cpu_put(policy);
896 * cpufreq_add_dev - add a CPU device
898 * Adds the cpufreq interface for a CPU device.
900 * The Oracle says: try running cpufreq registration/unregistration concurrently
901 * with with cpu hotplugging and all hell will break loose. Tried to clean this
902 * mess up, but more thorough testing is needed. - Mathieu
904 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
906 unsigned int j, cpu = dev->id;
908 struct cpufreq_policy *policy;
910 #ifdef CONFIG_HOTPLUG_CPU
911 struct cpufreq_governor *gov;
915 if (cpu_is_offline(cpu))
918 pr_debug("adding CPU %u\n", cpu);
921 /* check whether a different CPU already registered this
922 * CPU because it is in the same boat. */
923 policy = cpufreq_cpu_get(cpu);
924 if (unlikely(policy)) {
925 cpufreq_cpu_put(policy);
929 #ifdef CONFIG_HOTPLUG_CPU
930 /* Check if this cpu was hot-unplugged earlier and has siblings */
931 read_lock_irqsave(&cpufreq_driver_lock, flags);
932 for_each_online_cpu(sibling) {
933 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
934 if (cp && cpumask_test_cpu(cpu, cp->related_cpus)) {
935 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
936 return cpufreq_add_policy_cpu(cpu, sibling, dev);
939 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
943 if (!try_module_get(cpufreq_driver->owner)) {
948 policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
952 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
953 goto err_free_policy;
955 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
956 goto err_free_cpumask;
959 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
960 cpumask_copy(policy->cpus, cpumask_of(cpu));
962 /* Initially set CPU itself as the policy_cpu */
963 per_cpu(cpufreq_policy_cpu, cpu) = cpu;
965 init_completion(&policy->kobj_unregister);
966 INIT_WORK(&policy->update, handle_update);
968 /* call driver. From then on the cpufreq must be able
969 * to accept all calls to ->verify and ->setpolicy for this CPU
971 ret = cpufreq_driver->init(policy);
973 pr_debug("initialization failed\n");
974 goto err_set_policy_cpu;
977 /* related cpus should atleast have policy->cpus */
978 cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
981 * affected cpus must always be the one, which are online. We aren't
982 * managing offline cpus here.
984 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
986 policy->user_policy.min = policy->min;
987 policy->user_policy.max = policy->max;
989 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
990 CPUFREQ_START, policy);
992 #ifdef CONFIG_HOTPLUG_CPU
993 gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
995 policy->governor = gov;
996 pr_debug("Restoring governor %s for cpu %d\n",
997 policy->governor->name, cpu);
1001 ret = cpufreq_add_dev_interface(cpu, policy, dev);
1003 goto err_out_unregister;
1005 kobject_uevent(&policy->kobj, KOBJ_ADD);
1006 module_put(cpufreq_driver->owner);
1007 pr_debug("initialization complete\n");
1012 write_lock_irqsave(&cpufreq_driver_lock, flags);
1013 for_each_cpu(j, policy->cpus)
1014 per_cpu(cpufreq_cpu_data, j) = NULL;
1015 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1017 kobject_put(&policy->kobj);
1018 wait_for_completion(&policy->kobj_unregister);
1021 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1022 free_cpumask_var(policy->related_cpus);
1024 free_cpumask_var(policy->cpus);
1028 module_put(cpufreq_driver->owner);
1033 static void update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1037 policy->last_cpu = policy->cpu;
1040 for_each_cpu(j, policy->cpus)
1041 per_cpu(cpufreq_policy_cpu, j) = cpu;
1043 #ifdef CONFIG_CPU_FREQ_TABLE
1044 cpufreq_frequency_table_update_policy_cpu(policy);
1046 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1047 CPUFREQ_UPDATE_POLICY_CPU, policy);
1051 * __cpufreq_remove_dev - remove a CPU device
1053 * Removes the cpufreq interface for a CPU device.
1054 * Caller should already have policy_rwsem in write mode for this CPU.
1055 * This routine frees the rwsem before returning.
1057 static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1059 unsigned int cpu = dev->id, ret, cpus;
1060 unsigned long flags;
1061 struct cpufreq_policy *data;
1062 struct kobject *kobj;
1063 struct completion *cmp;
1064 struct device *cpu_dev;
1066 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1068 write_lock_irqsave(&cpufreq_driver_lock, flags);
1070 data = per_cpu(cpufreq_cpu_data, cpu);
1071 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1073 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1076 pr_debug("%s: No cpu_data found\n", __func__);
1080 if (cpufreq_driver->target)
1081 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1083 #ifdef CONFIG_HOTPLUG_CPU
1084 if (!cpufreq_driver->setpolicy)
1085 strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1086 data->governor->name, CPUFREQ_NAME_LEN);
1089 WARN_ON(lock_policy_rwsem_write(cpu));
1090 cpus = cpumask_weight(data->cpus);
1093 cpumask_clear_cpu(cpu, data->cpus);
1094 unlock_policy_rwsem_write(cpu);
1096 if (cpu != data->cpu) {
1097 sysfs_remove_link(&dev->kobj, "cpufreq");
1098 } else if (cpus > 1) {
1099 /* first sibling now owns the new sysfs dir */
1100 cpu_dev = get_cpu_device(cpumask_first(data->cpus));
1101 sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1102 ret = kobject_move(&data->kobj, &cpu_dev->kobj);
1104 pr_err("%s: Failed to move kobj: %d", __func__, ret);
1106 WARN_ON(lock_policy_rwsem_write(cpu));
1107 cpumask_set_cpu(cpu, data->cpus);
1109 write_lock_irqsave(&cpufreq_driver_lock, flags);
1110 per_cpu(cpufreq_cpu_data, cpu) = data;
1111 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1113 unlock_policy_rwsem_write(cpu);
1115 ret = sysfs_create_link(&cpu_dev->kobj, &data->kobj,
1120 WARN_ON(lock_policy_rwsem_write(cpu));
1121 update_policy_cpu(data, cpu_dev->id);
1122 unlock_policy_rwsem_write(cpu);
1123 pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1124 __func__, cpu_dev->id, cpu);
1127 /* If cpu is last user of policy, free policy */
1129 if (cpufreq_driver->target)
1130 __cpufreq_governor(data, CPUFREQ_GOV_POLICY_EXIT);
1132 lock_policy_rwsem_read(cpu);
1134 cmp = &data->kobj_unregister;
1135 unlock_policy_rwsem_read(cpu);
1138 /* we need to make sure that the underlying kobj is actually
1139 * not referenced anymore by anybody before we proceed with
1142 pr_debug("waiting for dropping of refcount\n");
1143 wait_for_completion(cmp);
1144 pr_debug("wait complete\n");
1146 if (cpufreq_driver->exit)
1147 cpufreq_driver->exit(data);
1149 free_cpumask_var(data->related_cpus);
1150 free_cpumask_var(data->cpus);
1153 pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
1154 cpufreq_cpu_put(data);
1155 if (cpufreq_driver->target) {
1156 __cpufreq_governor(data, CPUFREQ_GOV_START);
1157 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1161 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1166 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1168 unsigned int cpu = dev->id;
1171 if (cpu_is_offline(cpu))
1174 retval = __cpufreq_remove_dev(dev, sif);
1179 static void handle_update(struct work_struct *work)
1181 struct cpufreq_policy *policy =
1182 container_of(work, struct cpufreq_policy, update);
1183 unsigned int cpu = policy->cpu;
1184 pr_debug("handle_update for cpu %u called\n", cpu);
1185 cpufreq_update_policy(cpu);
1189 * cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1191 * @old_freq: CPU frequency the kernel thinks the CPU runs at
1192 * @new_freq: CPU frequency the CPU actually runs at
1194 * We adjust to current frequency first, and need to clean up later.
1195 * So either call to cpufreq_update_policy() or schedule handle_update()).
1197 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1198 unsigned int new_freq)
1200 struct cpufreq_policy *policy;
1201 struct cpufreq_freqs freqs;
1202 unsigned long flags;
1205 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1206 "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1208 freqs.old = old_freq;
1209 freqs.new = new_freq;
1211 read_lock_irqsave(&cpufreq_driver_lock, flags);
1212 policy = per_cpu(cpufreq_cpu_data, cpu);
1213 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1215 cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
1216 cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
1221 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1224 * This is the last known freq, without actually getting it from the driver.
1225 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1227 unsigned int cpufreq_quick_get(unsigned int cpu)
1229 struct cpufreq_policy *policy;
1230 unsigned int ret_freq = 0;
1232 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1233 return cpufreq_driver->get(cpu);
1235 policy = cpufreq_cpu_get(cpu);
1237 ret_freq = policy->cur;
1238 cpufreq_cpu_put(policy);
1243 EXPORT_SYMBOL(cpufreq_quick_get);
1246 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1249 * Just return the max possible frequency for a given CPU.
1251 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1253 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1254 unsigned int ret_freq = 0;
1257 ret_freq = policy->max;
1258 cpufreq_cpu_put(policy);
1263 EXPORT_SYMBOL(cpufreq_quick_get_max);
1266 static unsigned int __cpufreq_get(unsigned int cpu)
1268 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1269 unsigned int ret_freq = 0;
1271 if (!cpufreq_driver->get)
1274 ret_freq = cpufreq_driver->get(cpu);
1276 if (ret_freq && policy->cur &&
1277 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1278 /* verify no discrepancy between actual and
1279 saved value exists */
1280 if (unlikely(ret_freq != policy->cur)) {
1281 cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1282 schedule_work(&policy->update);
1290 * cpufreq_get - get the current CPU frequency (in kHz)
1293 * Get the CPU current (static) CPU frequency
1295 unsigned int cpufreq_get(unsigned int cpu)
1297 unsigned int ret_freq = 0;
1298 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1303 if (unlikely(lock_policy_rwsem_read(cpu)))
1306 ret_freq = __cpufreq_get(cpu);
1308 unlock_policy_rwsem_read(cpu);
1311 cpufreq_cpu_put(policy);
1315 EXPORT_SYMBOL(cpufreq_get);
1317 static struct subsys_interface cpufreq_interface = {
1319 .subsys = &cpu_subsys,
1320 .add_dev = cpufreq_add_dev,
1321 .remove_dev = cpufreq_remove_dev,
1326 * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1328 * This function is only executed for the boot processor. The other CPUs
1329 * have been put offline by means of CPU hotplug.
1331 static int cpufreq_bp_suspend(void)
1335 int cpu = smp_processor_id();
1336 struct cpufreq_policy *cpu_policy;
1338 pr_debug("suspending cpu %u\n", cpu);
1340 /* If there's no policy for the boot CPU, we have nothing to do. */
1341 cpu_policy = cpufreq_cpu_get(cpu);
1345 if (cpufreq_driver->suspend) {
1346 ret = cpufreq_driver->suspend(cpu_policy);
1348 printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1349 "step on CPU %u\n", cpu_policy->cpu);
1352 cpufreq_cpu_put(cpu_policy);
1357 * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1359 * 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1360 * 2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1361 * restored. It will verify that the current freq is in sync with
1362 * what we believe it to be. This is a bit later than when it
1363 * should be, but nonethteless it's better than calling
1364 * cpufreq_driver->get() here which might re-enable interrupts...
1366 * This function is only executed for the boot CPU. The other CPUs have not
1367 * been turned on yet.
1369 static void cpufreq_bp_resume(void)
1373 int cpu = smp_processor_id();
1374 struct cpufreq_policy *cpu_policy;
1376 pr_debug("resuming cpu %u\n", cpu);
1378 /* If there's no policy for the boot CPU, we have nothing to do. */
1379 cpu_policy = cpufreq_cpu_get(cpu);
1383 if (cpufreq_driver->resume) {
1384 ret = cpufreq_driver->resume(cpu_policy);
1386 printk(KERN_ERR "cpufreq: resume failed in ->resume "
1387 "step on CPU %u\n", cpu_policy->cpu);
1392 schedule_work(&cpu_policy->update);
1395 cpufreq_cpu_put(cpu_policy);
1398 static struct syscore_ops cpufreq_syscore_ops = {
1399 .suspend = cpufreq_bp_suspend,
1400 .resume = cpufreq_bp_resume,
1404 * cpufreq_get_current_driver - return current driver's name
1406 * Return the name string of the currently loaded cpufreq driver
1409 const char *cpufreq_get_current_driver(void)
1412 return cpufreq_driver->name;
1416 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1418 /*********************************************************************
1419 * NOTIFIER LISTS INTERFACE *
1420 *********************************************************************/
1423 * cpufreq_register_notifier - register a driver with cpufreq
1424 * @nb: notifier function to register
1425 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1427 * Add a driver to one of two lists: either a list of drivers that
1428 * are notified about clock rate changes (once before and once after
1429 * the transition), or a list of drivers that are notified about
1430 * changes in cpufreq policy.
1432 * This function may sleep, and has the same return conditions as
1433 * blocking_notifier_chain_register.
1435 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1439 if (cpufreq_disabled())
1442 WARN_ON(!init_cpufreq_transition_notifier_list_called);
1445 case CPUFREQ_TRANSITION_NOTIFIER:
1446 ret = srcu_notifier_chain_register(
1447 &cpufreq_transition_notifier_list, nb);
1449 case CPUFREQ_POLICY_NOTIFIER:
1450 ret = blocking_notifier_chain_register(
1451 &cpufreq_policy_notifier_list, nb);
1459 EXPORT_SYMBOL(cpufreq_register_notifier);
1463 * cpufreq_unregister_notifier - unregister a driver with cpufreq
1464 * @nb: notifier block to be unregistered
1465 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1467 * Remove a driver from the CPU frequency notifier list.
1469 * This function may sleep, and has the same return conditions as
1470 * blocking_notifier_chain_unregister.
1472 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1476 if (cpufreq_disabled())
1480 case CPUFREQ_TRANSITION_NOTIFIER:
1481 ret = srcu_notifier_chain_unregister(
1482 &cpufreq_transition_notifier_list, nb);
1484 case CPUFREQ_POLICY_NOTIFIER:
1485 ret = blocking_notifier_chain_unregister(
1486 &cpufreq_policy_notifier_list, nb);
1494 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1497 /*********************************************************************
1499 *********************************************************************/
1502 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1503 unsigned int target_freq,
1504 unsigned int relation)
1506 int retval = -EINVAL;
1507 unsigned int old_target_freq = target_freq;
1509 if (cpufreq_disabled())
1512 /* Make sure that target_freq is within supported range */
1513 if (target_freq > policy->max)
1514 target_freq = policy->max;
1515 if (target_freq < policy->min)
1516 target_freq = policy->min;
1518 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1519 policy->cpu, target_freq, relation, old_target_freq);
1521 if (target_freq == policy->cur)
1524 if (cpufreq_driver->target)
1525 retval = cpufreq_driver->target(policy, target_freq, relation);
1529 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1531 int cpufreq_driver_target(struct cpufreq_policy *policy,
1532 unsigned int target_freq,
1533 unsigned int relation)
1537 policy = cpufreq_cpu_get(policy->cpu);
1541 if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1544 ret = __cpufreq_driver_target(policy, target_freq, relation);
1546 unlock_policy_rwsem_write(policy->cpu);
1549 cpufreq_cpu_put(policy);
1553 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1555 int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
1559 if (cpufreq_disabled())
1562 if (!cpufreq_driver->getavg)
1565 policy = cpufreq_cpu_get(policy->cpu);
1569 ret = cpufreq_driver->getavg(policy, cpu);
1571 cpufreq_cpu_put(policy);
1574 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1577 * when "event" is CPUFREQ_GOV_LIMITS
1580 static int __cpufreq_governor(struct cpufreq_policy *policy,
1585 /* Only must be defined when default governor is known to have latency
1586 restrictions, like e.g. conservative or ondemand.
1587 That this is the case is already ensured in Kconfig
1589 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1590 struct cpufreq_governor *gov = &cpufreq_gov_performance;
1592 struct cpufreq_governor *gov = NULL;
1595 if (policy->governor->max_transition_latency &&
1596 policy->cpuinfo.transition_latency >
1597 policy->governor->max_transition_latency) {
1601 printk(KERN_WARNING "%s governor failed, too long"
1602 " transition latency of HW, fallback"
1603 " to %s governor\n",
1604 policy->governor->name,
1606 policy->governor = gov;
1610 if (!try_module_get(policy->governor->owner))
1613 pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1614 policy->cpu, event);
1616 mutex_lock(&cpufreq_governor_lock);
1617 if ((!policy->governor_enabled && (event == CPUFREQ_GOV_STOP)) ||
1618 (policy->governor_enabled && (event == CPUFREQ_GOV_START))) {
1619 mutex_unlock(&cpufreq_governor_lock);
1623 if (event == CPUFREQ_GOV_STOP)
1624 policy->governor_enabled = false;
1625 else if (event == CPUFREQ_GOV_START)
1626 policy->governor_enabled = true;
1628 mutex_unlock(&cpufreq_governor_lock);
1630 ret = policy->governor->governor(policy, event);
1633 if (event == CPUFREQ_GOV_POLICY_INIT)
1634 policy->governor->initialized++;
1635 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1636 policy->governor->initialized--;
1638 /* Restore original values */
1639 mutex_lock(&cpufreq_governor_lock);
1640 if (event == CPUFREQ_GOV_STOP)
1641 policy->governor_enabled = true;
1642 else if (event == CPUFREQ_GOV_START)
1643 policy->governor_enabled = false;
1644 mutex_unlock(&cpufreq_governor_lock);
1647 /* we keep one module reference alive for
1648 each CPU governed by this CPU */
1649 if ((event != CPUFREQ_GOV_START) || ret)
1650 module_put(policy->governor->owner);
1651 if ((event == CPUFREQ_GOV_STOP) && !ret)
1652 module_put(policy->governor->owner);
1658 int cpufreq_register_governor(struct cpufreq_governor *governor)
1665 if (cpufreq_disabled())
1668 mutex_lock(&cpufreq_governor_mutex);
1670 governor->initialized = 0;
1672 if (__find_governor(governor->name) == NULL) {
1674 list_add(&governor->governor_list, &cpufreq_governor_list);
1677 mutex_unlock(&cpufreq_governor_mutex);
1680 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1683 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1685 #ifdef CONFIG_HOTPLUG_CPU
1692 if (cpufreq_disabled())
1695 #ifdef CONFIG_HOTPLUG_CPU
1696 for_each_present_cpu(cpu) {
1697 if (cpu_online(cpu))
1699 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1700 strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1704 mutex_lock(&cpufreq_governor_mutex);
1705 list_del(&governor->governor_list);
1706 mutex_unlock(&cpufreq_governor_mutex);
1709 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1713 /*********************************************************************
1714 * POLICY INTERFACE *
1715 *********************************************************************/
1718 * cpufreq_get_policy - get the current cpufreq_policy
1719 * @policy: struct cpufreq_policy into which the current cpufreq_policy
1722 * Reads the current cpufreq policy.
1724 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1726 struct cpufreq_policy *cpu_policy;
1730 cpu_policy = cpufreq_cpu_get(cpu);
1734 memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1736 cpufreq_cpu_put(cpu_policy);
1739 EXPORT_SYMBOL(cpufreq_get_policy);
1743 * data : current policy.
1744 * policy : policy to be set.
1746 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1747 struct cpufreq_policy *policy)
1749 int ret = 0, failed = 1;
1751 pr_debug("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1752 policy->min, policy->max);
1754 memcpy(&policy->cpuinfo, &data->cpuinfo,
1755 sizeof(struct cpufreq_cpuinfo));
1757 if (policy->min > data->max || policy->max < data->min) {
1762 /* verify the cpu speed can be set within this limit */
1763 ret = cpufreq_driver->verify(policy);
1767 /* adjust if necessary - all reasons */
1768 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1769 CPUFREQ_ADJUST, policy);
1771 /* adjust if necessary - hardware incompatibility*/
1772 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1773 CPUFREQ_INCOMPATIBLE, policy);
1775 /* verify the cpu speed can be set within this limit,
1776 which might be different to the first one */
1777 ret = cpufreq_driver->verify(policy);
1781 /* notification of the new policy */
1782 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1783 CPUFREQ_NOTIFY, policy);
1785 data->min = policy->min;
1786 data->max = policy->max;
1788 pr_debug("new min and max freqs are %u - %u kHz\n",
1789 data->min, data->max);
1791 if (cpufreq_driver->setpolicy) {
1792 data->policy = policy->policy;
1793 pr_debug("setting range\n");
1794 ret = cpufreq_driver->setpolicy(policy);
1796 if (policy->governor != data->governor) {
1797 /* save old, working values */
1798 struct cpufreq_governor *old_gov = data->governor;
1800 pr_debug("governor switch\n");
1802 /* end old governor */
1803 if (data->governor) {
1804 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1805 unlock_policy_rwsem_write(policy->cpu);
1806 __cpufreq_governor(data,
1807 CPUFREQ_GOV_POLICY_EXIT);
1808 lock_policy_rwsem_write(policy->cpu);
1811 /* start new governor */
1812 data->governor = policy->governor;
1813 if (!__cpufreq_governor(data, CPUFREQ_GOV_POLICY_INIT)) {
1814 if (!__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1817 unlock_policy_rwsem_write(policy->cpu);
1818 __cpufreq_governor(data,
1819 CPUFREQ_GOV_POLICY_EXIT);
1820 lock_policy_rwsem_write(policy->cpu);
1825 /* new governor failed, so re-start old one */
1826 pr_debug("starting governor %s failed\n",
1827 data->governor->name);
1829 data->governor = old_gov;
1830 __cpufreq_governor(data,
1831 CPUFREQ_GOV_POLICY_INIT);
1832 __cpufreq_governor(data,
1838 /* might be a policy change, too, so fall through */
1840 pr_debug("governor: change or update limits\n");
1841 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1849 * cpufreq_update_policy - re-evaluate an existing cpufreq policy
1850 * @cpu: CPU which shall be re-evaluated
1852 * Useful for policy notifiers which have different necessities
1853 * at different times.
1855 int cpufreq_update_policy(unsigned int cpu)
1857 struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1858 struct cpufreq_policy policy;
1866 if (unlikely(lock_policy_rwsem_write(cpu))) {
1871 pr_debug("updating policy for CPU %u\n", cpu);
1872 memcpy(&policy, data, sizeof(struct cpufreq_policy));
1873 policy.min = data->user_policy.min;
1874 policy.max = data->user_policy.max;
1875 policy.policy = data->user_policy.policy;
1876 policy.governor = data->user_policy.governor;
1878 /* BIOS might change freq behind our back
1879 -> ask driver for current freq and notify governors about a change */
1880 if (cpufreq_driver->get) {
1881 policy.cur = cpufreq_driver->get(cpu);
1883 pr_debug("Driver did not initialize current freq");
1884 data->cur = policy.cur;
1886 if (data->cur != policy.cur && cpufreq_driver->target)
1887 cpufreq_out_of_sync(cpu, data->cur,
1892 ret = __cpufreq_set_policy(data, &policy);
1894 unlock_policy_rwsem_write(cpu);
1897 cpufreq_cpu_put(data);
1901 EXPORT_SYMBOL(cpufreq_update_policy);
1903 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1904 unsigned long action, void *hcpu)
1906 unsigned int cpu = (unsigned long)hcpu;
1909 dev = get_cpu_device(cpu);
1913 case CPU_ONLINE_FROZEN:
1914 cpufreq_add_dev(dev, NULL);
1916 case CPU_DOWN_PREPARE:
1917 case CPU_DOWN_PREPARE_FROZEN:
1918 __cpufreq_remove_dev(dev, NULL);
1920 case CPU_DOWN_FAILED:
1921 case CPU_DOWN_FAILED_FROZEN:
1922 cpufreq_add_dev(dev, NULL);
1929 static struct notifier_block __refdata cpufreq_cpu_notifier = {
1930 .notifier_call = cpufreq_cpu_callback,
1933 /*********************************************************************
1934 * REGISTER / UNREGISTER CPUFREQ DRIVER *
1935 *********************************************************************/
1938 * cpufreq_register_driver - register a CPU Frequency driver
1939 * @driver_data: A struct cpufreq_driver containing the values#
1940 * submitted by the CPU Frequency driver.
1942 * Registers a CPU Frequency driver to this core code. This code
1943 * returns zero on success, -EBUSY when another driver got here first
1944 * (and isn't unregistered in the meantime).
1947 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1949 unsigned long flags;
1952 if (cpufreq_disabled())
1955 if (!driver_data || !driver_data->verify || !driver_data->init ||
1956 ((!driver_data->setpolicy) && (!driver_data->target)))
1959 pr_debug("trying to register driver %s\n", driver_data->name);
1961 if (driver_data->setpolicy)
1962 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1964 write_lock_irqsave(&cpufreq_driver_lock, flags);
1965 if (cpufreq_driver) {
1966 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1969 cpufreq_driver = driver_data;
1970 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1972 ret = subsys_interface_register(&cpufreq_interface);
1974 goto err_null_driver;
1976 if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1980 /* check for at least one working CPU */
1981 for (i = 0; i < nr_cpu_ids; i++)
1982 if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
1987 /* if all ->init() calls failed, unregister */
1989 pr_debug("no CPU initialized for driver %s\n",
1995 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1996 pr_debug("driver %s up and running\n", driver_data->name);
2000 subsys_interface_unregister(&cpufreq_interface);
2002 write_lock_irqsave(&cpufreq_driver_lock, flags);
2003 cpufreq_driver = NULL;
2004 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2007 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2011 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2013 * Unregister the current CPUFreq driver. Only call this if you have
2014 * the right to do so, i.e. if you have succeeded in initialising before!
2015 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2016 * currently not initialised.
2018 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2020 unsigned long flags;
2022 if (!cpufreq_driver || (driver != cpufreq_driver))
2025 pr_debug("unregistering driver %s\n", driver->name);
2027 subsys_interface_unregister(&cpufreq_interface);
2028 unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2030 write_lock_irqsave(&cpufreq_driver_lock, flags);
2031 cpufreq_driver = NULL;
2032 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2036 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2038 static int __init cpufreq_core_init(void)
2042 if (cpufreq_disabled())
2045 for_each_possible_cpu(cpu) {
2046 per_cpu(cpufreq_policy_cpu, cpu) = -1;
2047 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
2050 cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2051 BUG_ON(!cpufreq_global_kobject);
2052 register_syscore_ops(&cpufreq_syscore_ops);
2056 core_initcall(cpufreq_core_init);