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 <linux/cpu.h>
21 #include <linux/cpufreq.h>
22 #include <linux/delay.h>
23 #include <linux/device.h>
24 #include <linux/init.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/module.h>
27 #include <linux/mutex.h>
28 #include <linux/slab.h>
29 #include <linux/syscore_ops.h>
30 #include <linux/tick.h>
31 #include <trace/events/power.h>
34 * The "cpufreq driver" - the arch- or hardware-dependent low
35 * level driver of CPUFreq support, and its spinlock. This lock
36 * also protects the cpufreq_cpu_data array.
38 static struct cpufreq_driver *cpufreq_driver;
39 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
40 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data_fallback);
41 static DEFINE_RWLOCK(cpufreq_driver_lock);
42 static DEFINE_MUTEX(cpufreq_governor_lock);
43 static LIST_HEAD(cpufreq_policy_list);
45 #ifdef CONFIG_HOTPLUG_CPU
46 /* This one keeps track of the previously set governor of a removed CPU */
47 static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
50 static inline bool has_target(void)
52 return cpufreq_driver->target_index || cpufreq_driver->target;
56 * rwsem to guarantee that cpufreq driver module doesn't unload during critical
59 static DECLARE_RWSEM(cpufreq_rwsem);
61 /* internal prototypes */
62 static int __cpufreq_governor(struct cpufreq_policy *policy,
64 static unsigned int __cpufreq_get(unsigned int cpu);
65 static void handle_update(struct work_struct *work);
68 * Two notifier lists: the "policy" list is involved in the
69 * validation process for a new CPU frequency policy; the
70 * "transition" list for kernel code that needs to handle
71 * changes to devices when the CPU clock speed changes.
72 * The mutex locks both lists.
74 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
75 static struct srcu_notifier_head cpufreq_transition_notifier_list;
77 static bool init_cpufreq_transition_notifier_list_called;
78 static int __init init_cpufreq_transition_notifier_list(void)
80 srcu_init_notifier_head(&cpufreq_transition_notifier_list);
81 init_cpufreq_transition_notifier_list_called = true;
84 pure_initcall(init_cpufreq_transition_notifier_list);
86 static int off __read_mostly;
87 static int cpufreq_disabled(void)
91 void disable_cpufreq(void)
95 static LIST_HEAD(cpufreq_governor_list);
96 static DEFINE_MUTEX(cpufreq_governor_mutex);
98 bool have_governor_per_policy(void)
100 return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
102 EXPORT_SYMBOL_GPL(have_governor_per_policy);
104 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
106 if (have_governor_per_policy())
107 return &policy->kobj;
109 return cpufreq_global_kobject;
111 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
113 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
119 cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
121 busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
122 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
123 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
124 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
125 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
126 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
128 idle_time = cur_wall_time - busy_time;
130 *wall = cputime_to_usecs(cur_wall_time);
132 return cputime_to_usecs(idle_time);
135 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
137 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
139 if (idle_time == -1ULL)
140 return get_cpu_idle_time_jiffy(cpu, wall);
142 idle_time += get_cpu_iowait_time_us(cpu, wall);
146 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
149 * This is a generic cpufreq init() routine which can be used by cpufreq
150 * drivers of SMP systems. It will do following:
151 * - validate & show freq table passed
152 * - set policies transition latency
153 * - policy->cpus with all possible CPUs
155 int cpufreq_generic_init(struct cpufreq_policy *policy,
156 struct cpufreq_frequency_table *table,
157 unsigned int transition_latency)
161 ret = cpufreq_table_validate_and_show(policy, table);
163 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
167 policy->cpuinfo.transition_latency = transition_latency;
170 * The driver only supports the SMP configuartion where all processors
171 * share the clock and voltage and clock.
173 cpumask_setall(policy->cpus);
177 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
179 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
181 struct cpufreq_policy *policy = NULL;
184 if (cpufreq_disabled() || (cpu >= nr_cpu_ids))
187 if (!down_read_trylock(&cpufreq_rwsem))
190 /* get the cpufreq driver */
191 read_lock_irqsave(&cpufreq_driver_lock, flags);
193 if (cpufreq_driver) {
195 policy = per_cpu(cpufreq_cpu_data, cpu);
197 kobject_get(&policy->kobj);
200 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
203 up_read(&cpufreq_rwsem);
207 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
209 void cpufreq_cpu_put(struct cpufreq_policy *policy)
211 if (cpufreq_disabled())
214 kobject_put(&policy->kobj);
215 up_read(&cpufreq_rwsem);
217 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
219 /*********************************************************************
220 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
221 *********************************************************************/
224 * adjust_jiffies - adjust the system "loops_per_jiffy"
226 * This function alters the system "loops_per_jiffy" for the clock
227 * speed change. Note that loops_per_jiffy cannot be updated on SMP
228 * systems as each CPU might be scaled differently. So, use the arch
229 * per-CPU loops_per_jiffy value wherever possible.
232 static unsigned long l_p_j_ref;
233 static unsigned int l_p_j_ref_freq;
235 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
237 if (ci->flags & CPUFREQ_CONST_LOOPS)
240 if (!l_p_j_ref_freq) {
241 l_p_j_ref = loops_per_jiffy;
242 l_p_j_ref_freq = ci->old;
243 pr_debug("saving %lu as reference value for loops_per_jiffy; "
244 "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
246 if ((val == CPUFREQ_POSTCHANGE && ci->old != ci->new) ||
247 (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
248 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
250 pr_debug("scaling loops_per_jiffy to %lu "
251 "for frequency %u kHz\n", loops_per_jiffy, ci->new);
255 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
261 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
262 struct cpufreq_freqs *freqs, unsigned int state)
264 BUG_ON(irqs_disabled());
266 if (cpufreq_disabled())
269 freqs->flags = cpufreq_driver->flags;
270 pr_debug("notification %u of frequency transition to %u kHz\n",
275 case CPUFREQ_PRECHANGE:
276 /* detect if the driver reported a value as "old frequency"
277 * which is not equal to what the cpufreq core thinks is
280 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
281 if ((policy) && (policy->cpu == freqs->cpu) &&
282 (policy->cur) && (policy->cur != freqs->old)) {
283 pr_debug("Warning: CPU frequency is"
284 " %u, cpufreq assumed %u kHz.\n",
285 freqs->old, policy->cur);
286 freqs->old = policy->cur;
289 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
290 CPUFREQ_PRECHANGE, freqs);
291 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
294 case CPUFREQ_POSTCHANGE:
295 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
296 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
297 (unsigned long)freqs->cpu);
298 trace_cpu_frequency(freqs->new, freqs->cpu);
299 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
300 CPUFREQ_POSTCHANGE, freqs);
301 if (likely(policy) && likely(policy->cpu == freqs->cpu))
302 policy->cur = freqs->new;
308 * cpufreq_notify_transition - call notifier chain and adjust_jiffies
309 * on frequency transition.
311 * This function calls the transition notifiers and the "adjust_jiffies"
312 * function. It is called twice on all CPU frequency changes that have
315 void cpufreq_notify_transition(struct cpufreq_policy *policy,
316 struct cpufreq_freqs *freqs, unsigned int state)
318 for_each_cpu(freqs->cpu, policy->cpus)
319 __cpufreq_notify_transition(policy, freqs, state);
321 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
324 /*********************************************************************
326 *********************************************************************/
328 static struct cpufreq_governor *__find_governor(const char *str_governor)
330 struct cpufreq_governor *t;
332 list_for_each_entry(t, &cpufreq_governor_list, governor_list)
333 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
340 * cpufreq_parse_governor - parse a governor string
342 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
343 struct cpufreq_governor **governor)
350 if (cpufreq_driver->setpolicy) {
351 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
352 *policy = CPUFREQ_POLICY_PERFORMANCE;
354 } else if (!strnicmp(str_governor, "powersave",
356 *policy = CPUFREQ_POLICY_POWERSAVE;
359 } else if (has_target()) {
360 struct cpufreq_governor *t;
362 mutex_lock(&cpufreq_governor_mutex);
364 t = __find_governor(str_governor);
369 mutex_unlock(&cpufreq_governor_mutex);
370 ret = request_module("cpufreq_%s", str_governor);
371 mutex_lock(&cpufreq_governor_mutex);
374 t = __find_governor(str_governor);
382 mutex_unlock(&cpufreq_governor_mutex);
389 * cpufreq_per_cpu_attr_read() / show_##file_name() -
390 * print out cpufreq information
392 * Write out information from cpufreq_driver->policy[cpu]; object must be
396 #define show_one(file_name, object) \
397 static ssize_t show_##file_name \
398 (struct cpufreq_policy *policy, char *buf) \
400 return sprintf(buf, "%u\n", policy->object); \
403 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
404 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
405 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
406 show_one(scaling_min_freq, min);
407 show_one(scaling_max_freq, max);
408 show_one(scaling_cur_freq, cur);
410 static int cpufreq_set_policy(struct cpufreq_policy *policy,
411 struct cpufreq_policy *new_policy);
414 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
416 #define store_one(file_name, object) \
417 static ssize_t store_##file_name \
418 (struct cpufreq_policy *policy, const char *buf, size_t count) \
421 struct cpufreq_policy new_policy; \
423 ret = cpufreq_get_policy(&new_policy, policy->cpu); \
427 ret = sscanf(buf, "%u", &new_policy.object); \
431 ret = cpufreq_set_policy(policy, &new_policy); \
432 policy->user_policy.object = policy->object; \
434 return ret ? ret : count; \
437 store_one(scaling_min_freq, min);
438 store_one(scaling_max_freq, max);
441 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
443 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
446 unsigned int cur_freq = __cpufreq_get(policy->cpu);
448 return sprintf(buf, "<unknown>");
449 return sprintf(buf, "%u\n", cur_freq);
453 * show_scaling_governor - show the current policy for the specified CPU
455 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
457 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
458 return sprintf(buf, "powersave\n");
459 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
460 return sprintf(buf, "performance\n");
461 else if (policy->governor)
462 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
463 policy->governor->name);
468 * store_scaling_governor - store policy for the specified CPU
470 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
471 const char *buf, size_t count)
474 char str_governor[16];
475 struct cpufreq_policy new_policy;
477 ret = cpufreq_get_policy(&new_policy, policy->cpu);
481 ret = sscanf(buf, "%15s", str_governor);
485 if (cpufreq_parse_governor(str_governor, &new_policy.policy,
486 &new_policy.governor))
489 ret = cpufreq_set_policy(policy, &new_policy);
491 policy->user_policy.policy = policy->policy;
492 policy->user_policy.governor = policy->governor;
501 * show_scaling_driver - show the cpufreq driver currently loaded
503 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
505 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
509 * show_scaling_available_governors - show the available CPUfreq governors
511 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
515 struct cpufreq_governor *t;
518 i += sprintf(buf, "performance powersave");
522 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
523 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
524 - (CPUFREQ_NAME_LEN + 2)))
526 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
529 i += sprintf(&buf[i], "\n");
533 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
538 for_each_cpu(cpu, mask) {
540 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
541 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
542 if (i >= (PAGE_SIZE - 5))
545 i += sprintf(&buf[i], "\n");
548 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
551 * show_related_cpus - show the CPUs affected by each transition even if
552 * hw coordination is in use
554 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
556 return cpufreq_show_cpus(policy->related_cpus, buf);
560 * show_affected_cpus - show the CPUs affected by each transition
562 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
564 return cpufreq_show_cpus(policy->cpus, buf);
567 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
568 const char *buf, size_t count)
570 unsigned int freq = 0;
573 if (!policy->governor || !policy->governor->store_setspeed)
576 ret = sscanf(buf, "%u", &freq);
580 policy->governor->store_setspeed(policy, freq);
585 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
587 if (!policy->governor || !policy->governor->show_setspeed)
588 return sprintf(buf, "<unsupported>\n");
590 return policy->governor->show_setspeed(policy, buf);
594 * show_bios_limit - show the current cpufreq HW/BIOS limitation
596 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
600 if (cpufreq_driver->bios_limit) {
601 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
603 return sprintf(buf, "%u\n", limit);
605 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
608 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
609 cpufreq_freq_attr_ro(cpuinfo_min_freq);
610 cpufreq_freq_attr_ro(cpuinfo_max_freq);
611 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
612 cpufreq_freq_attr_ro(scaling_available_governors);
613 cpufreq_freq_attr_ro(scaling_driver);
614 cpufreq_freq_attr_ro(scaling_cur_freq);
615 cpufreq_freq_attr_ro(bios_limit);
616 cpufreq_freq_attr_ro(related_cpus);
617 cpufreq_freq_attr_ro(affected_cpus);
618 cpufreq_freq_attr_rw(scaling_min_freq);
619 cpufreq_freq_attr_rw(scaling_max_freq);
620 cpufreq_freq_attr_rw(scaling_governor);
621 cpufreq_freq_attr_rw(scaling_setspeed);
623 static struct attribute *default_attrs[] = {
624 &cpuinfo_min_freq.attr,
625 &cpuinfo_max_freq.attr,
626 &cpuinfo_transition_latency.attr,
627 &scaling_min_freq.attr,
628 &scaling_max_freq.attr,
631 &scaling_governor.attr,
632 &scaling_driver.attr,
633 &scaling_available_governors.attr,
634 &scaling_setspeed.attr,
638 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
639 #define to_attr(a) container_of(a, struct freq_attr, attr)
641 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
643 struct cpufreq_policy *policy = to_policy(kobj);
644 struct freq_attr *fattr = to_attr(attr);
647 if (!down_read_trylock(&cpufreq_rwsem))
650 down_read(&policy->rwsem);
653 ret = fattr->show(policy, buf);
657 up_read(&policy->rwsem);
658 up_read(&cpufreq_rwsem);
663 static ssize_t store(struct kobject *kobj, struct attribute *attr,
664 const char *buf, size_t count)
666 struct cpufreq_policy *policy = to_policy(kobj);
667 struct freq_attr *fattr = to_attr(attr);
668 ssize_t ret = -EINVAL;
672 if (!cpu_online(policy->cpu))
675 if (!down_read_trylock(&cpufreq_rwsem))
678 down_write(&policy->rwsem);
681 ret = fattr->store(policy, buf, count);
685 up_write(&policy->rwsem);
687 up_read(&cpufreq_rwsem);
694 static void cpufreq_sysfs_release(struct kobject *kobj)
696 struct cpufreq_policy *policy = to_policy(kobj);
697 pr_debug("last reference is dropped\n");
698 complete(&policy->kobj_unregister);
701 static const struct sysfs_ops sysfs_ops = {
706 static struct kobj_type ktype_cpufreq = {
707 .sysfs_ops = &sysfs_ops,
708 .default_attrs = default_attrs,
709 .release = cpufreq_sysfs_release,
712 struct kobject *cpufreq_global_kobject;
713 EXPORT_SYMBOL(cpufreq_global_kobject);
715 static int cpufreq_global_kobject_usage;
717 int cpufreq_get_global_kobject(void)
719 if (!cpufreq_global_kobject_usage++)
720 return kobject_add(cpufreq_global_kobject,
721 &cpu_subsys.dev_root->kobj, "%s", "cpufreq");
725 EXPORT_SYMBOL(cpufreq_get_global_kobject);
727 void cpufreq_put_global_kobject(void)
729 if (!--cpufreq_global_kobject_usage)
730 kobject_del(cpufreq_global_kobject);
732 EXPORT_SYMBOL(cpufreq_put_global_kobject);
734 int cpufreq_sysfs_create_file(const struct attribute *attr)
736 int ret = cpufreq_get_global_kobject();
739 ret = sysfs_create_file(cpufreq_global_kobject, attr);
741 cpufreq_put_global_kobject();
746 EXPORT_SYMBOL(cpufreq_sysfs_create_file);
748 void cpufreq_sysfs_remove_file(const struct attribute *attr)
750 sysfs_remove_file(cpufreq_global_kobject, attr);
751 cpufreq_put_global_kobject();
753 EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
755 /* symlink affected CPUs */
756 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
761 for_each_cpu(j, policy->cpus) {
762 struct device *cpu_dev;
764 if (j == policy->cpu)
767 pr_debug("Adding link for CPU: %u\n", j);
768 cpu_dev = get_cpu_device(j);
769 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
777 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy,
780 struct freq_attr **drv_attr;
783 /* prepare interface data */
784 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
785 &dev->kobj, "cpufreq");
789 /* set up files for this cpu device */
790 drv_attr = cpufreq_driver->attr;
791 while ((drv_attr) && (*drv_attr)) {
792 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
794 goto err_out_kobj_put;
797 if (cpufreq_driver->get) {
798 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
800 goto err_out_kobj_put;
803 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
805 goto err_out_kobj_put;
807 if (cpufreq_driver->bios_limit) {
808 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
810 goto err_out_kobj_put;
813 ret = cpufreq_add_dev_symlink(policy);
815 goto err_out_kobj_put;
820 kobject_put(&policy->kobj);
821 wait_for_completion(&policy->kobj_unregister);
825 static void cpufreq_init_policy(struct cpufreq_policy *policy)
827 struct cpufreq_policy new_policy;
830 memcpy(&new_policy, policy, sizeof(*policy));
832 /* Use the default policy if its valid. */
833 if (cpufreq_driver->setpolicy)
834 cpufreq_parse_governor(policy->governor->name,
835 &new_policy.policy, NULL);
837 /* assure that the starting sequence is run in cpufreq_set_policy */
838 policy->governor = NULL;
840 /* set default policy */
841 ret = cpufreq_set_policy(policy, &new_policy);
842 policy->user_policy.policy = policy->policy;
843 policy->user_policy.governor = policy->governor;
846 pr_debug("setting policy failed\n");
847 if (cpufreq_driver->exit)
848 cpufreq_driver->exit(policy);
852 #ifdef CONFIG_HOTPLUG_CPU
853 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy,
854 unsigned int cpu, struct device *dev)
860 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
862 pr_err("%s: Failed to stop governor\n", __func__);
867 down_write(&policy->rwsem);
869 write_lock_irqsave(&cpufreq_driver_lock, flags);
871 cpumask_set_cpu(cpu, policy->cpus);
872 per_cpu(cpufreq_cpu_data, cpu) = policy;
873 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
875 up_write(&policy->rwsem);
878 if ((ret = __cpufreq_governor(policy, CPUFREQ_GOV_START)) ||
879 (ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))) {
880 pr_err("%s: Failed to start governor\n", __func__);
885 return sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
889 static struct cpufreq_policy *cpufreq_policy_restore(unsigned int cpu)
891 struct cpufreq_policy *policy;
894 read_lock_irqsave(&cpufreq_driver_lock, flags);
896 policy = per_cpu(cpufreq_cpu_data_fallback, cpu);
898 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
903 static struct cpufreq_policy *cpufreq_policy_alloc(void)
905 struct cpufreq_policy *policy;
907 policy = kzalloc(sizeof(*policy), GFP_KERNEL);
911 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
912 goto err_free_policy;
914 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
915 goto err_free_cpumask;
917 INIT_LIST_HEAD(&policy->policy_list);
918 init_rwsem(&policy->rwsem);
923 free_cpumask_var(policy->cpus);
930 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
932 struct kobject *kobj;
933 struct completion *cmp;
935 down_read(&policy->rwsem);
936 kobj = &policy->kobj;
937 cmp = &policy->kobj_unregister;
938 up_read(&policy->rwsem);
942 * We need to make sure that the underlying kobj is
943 * actually not referenced anymore by anybody before we
944 * proceed with unloading.
946 pr_debug("waiting for dropping of refcount\n");
947 wait_for_completion(cmp);
948 pr_debug("wait complete\n");
951 static void cpufreq_policy_free(struct cpufreq_policy *policy)
953 free_cpumask_var(policy->related_cpus);
954 free_cpumask_var(policy->cpus);
958 static void update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
960 if (WARN_ON(cpu == policy->cpu))
963 down_write(&policy->rwsem);
965 policy->last_cpu = policy->cpu;
968 up_write(&policy->rwsem);
970 cpufreq_frequency_table_update_policy_cpu(policy);
971 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
972 CPUFREQ_UPDATE_POLICY_CPU, policy);
975 static int __cpufreq_add_dev(struct device *dev, struct subsys_interface *sif,
978 unsigned int j, cpu = dev->id;
980 struct cpufreq_policy *policy;
982 #ifdef CONFIG_HOTPLUG_CPU
983 struct cpufreq_policy *tpolicy;
984 struct cpufreq_governor *gov;
987 if (cpu_is_offline(cpu))
990 pr_debug("adding CPU %u\n", cpu);
993 /* check whether a different CPU already registered this
994 * CPU because it is in the same boat. */
995 policy = cpufreq_cpu_get(cpu);
996 if (unlikely(policy)) {
997 cpufreq_cpu_put(policy);
1002 if (!down_read_trylock(&cpufreq_rwsem))
1005 #ifdef CONFIG_HOTPLUG_CPU
1006 /* Check if this cpu was hot-unplugged earlier and has siblings */
1007 read_lock_irqsave(&cpufreq_driver_lock, flags);
1008 list_for_each_entry(tpolicy, &cpufreq_policy_list, policy_list) {
1009 if (cpumask_test_cpu(cpu, tpolicy->related_cpus)) {
1010 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1011 ret = cpufreq_add_policy_cpu(tpolicy, cpu, dev);
1012 up_read(&cpufreq_rwsem);
1016 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1020 /* Restore the saved policy when doing light-weight init */
1021 policy = cpufreq_policy_restore(cpu);
1023 policy = cpufreq_policy_alloc();
1030 * In the resume path, since we restore a saved policy, the assignment
1031 * to policy->cpu is like an update of the existing policy, rather than
1032 * the creation of a brand new one. So we need to perform this update
1033 * by invoking update_policy_cpu().
1035 if (frozen && cpu != policy->cpu)
1036 update_policy_cpu(policy, cpu);
1040 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
1041 cpumask_copy(policy->cpus, cpumask_of(cpu));
1043 init_completion(&policy->kobj_unregister);
1044 INIT_WORK(&policy->update, handle_update);
1046 /* call driver. From then on the cpufreq must be able
1047 * to accept all calls to ->verify and ->setpolicy for this CPU
1049 ret = cpufreq_driver->init(policy);
1051 pr_debug("initialization failed\n");
1052 goto err_set_policy_cpu;
1055 if (cpufreq_driver->get) {
1056 policy->cur = cpufreq_driver->get(policy->cpu);
1058 pr_err("%s: ->get() failed\n", __func__);
1063 /* related cpus should atleast have policy->cpus */
1064 cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1067 * affected cpus must always be the one, which are online. We aren't
1068 * managing offline cpus here.
1070 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1072 policy->user_policy.min = policy->min;
1073 policy->user_policy.max = policy->max;
1075 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1076 CPUFREQ_START, policy);
1078 #ifdef CONFIG_HOTPLUG_CPU
1079 gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
1081 policy->governor = gov;
1082 pr_debug("Restoring governor %s for cpu %d\n",
1083 policy->governor->name, cpu);
1087 write_lock_irqsave(&cpufreq_driver_lock, flags);
1088 for_each_cpu(j, policy->cpus)
1089 per_cpu(cpufreq_cpu_data, j) = policy;
1090 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1093 ret = cpufreq_add_dev_interface(policy, dev);
1095 goto err_out_unregister;
1098 write_lock_irqsave(&cpufreq_driver_lock, flags);
1099 list_add(&policy->policy_list, &cpufreq_policy_list);
1100 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1102 cpufreq_init_policy(policy);
1104 kobject_uevent(&policy->kobj, KOBJ_ADD);
1105 up_read(&cpufreq_rwsem);
1107 pr_debug("initialization complete\n");
1112 write_lock_irqsave(&cpufreq_driver_lock, flags);
1113 for_each_cpu(j, policy->cpus)
1114 per_cpu(cpufreq_cpu_data, j) = NULL;
1115 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1118 if (cpufreq_driver->exit)
1119 cpufreq_driver->exit(policy);
1122 cpufreq_policy_put_kobj(policy);
1123 cpufreq_policy_free(policy);
1126 up_read(&cpufreq_rwsem);
1132 * cpufreq_add_dev - add a CPU device
1134 * Adds the cpufreq interface for a CPU device.
1136 * The Oracle says: try running cpufreq registration/unregistration concurrently
1137 * with with cpu hotplugging and all hell will break loose. Tried to clean this
1138 * mess up, but more thorough testing is needed. - Mathieu
1140 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1142 return __cpufreq_add_dev(dev, sif, false);
1145 static int cpufreq_nominate_new_policy_cpu(struct cpufreq_policy *policy,
1146 unsigned int old_cpu)
1148 struct device *cpu_dev;
1151 /* first sibling now owns the new sysfs dir */
1152 cpu_dev = get_cpu_device(cpumask_any_but(policy->cpus, old_cpu));
1154 sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1155 ret = kobject_move(&policy->kobj, &cpu_dev->kobj);
1157 pr_err("%s: Failed to move kobj: %d", __func__, ret);
1159 down_write(&policy->rwsem);
1160 cpumask_set_cpu(old_cpu, policy->cpus);
1161 up_write(&policy->rwsem);
1163 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
1172 static int __cpufreq_remove_dev_prepare(struct device *dev,
1173 struct subsys_interface *sif,
1176 unsigned int cpu = dev->id, cpus;
1178 unsigned long flags;
1179 struct cpufreq_policy *policy;
1181 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1183 write_lock_irqsave(&cpufreq_driver_lock, flags);
1185 policy = per_cpu(cpufreq_cpu_data, cpu);
1187 /* Save the policy somewhere when doing a light-weight tear-down */
1189 per_cpu(cpufreq_cpu_data_fallback, cpu) = policy;
1191 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1194 pr_debug("%s: No cpu_data found\n", __func__);
1199 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1201 pr_err("%s: Failed to stop governor\n", __func__);
1206 #ifdef CONFIG_HOTPLUG_CPU
1207 if (!cpufreq_driver->setpolicy)
1208 strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1209 policy->governor->name, CPUFREQ_NAME_LEN);
1212 down_read(&policy->rwsem);
1213 cpus = cpumask_weight(policy->cpus);
1214 up_read(&policy->rwsem);
1216 if (cpu != policy->cpu) {
1218 sysfs_remove_link(&dev->kobj, "cpufreq");
1219 } else if (cpus > 1) {
1220 new_cpu = cpufreq_nominate_new_policy_cpu(policy, cpu);
1222 update_policy_cpu(policy, new_cpu);
1225 pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1226 __func__, new_cpu, cpu);
1234 static int __cpufreq_remove_dev_finish(struct device *dev,
1235 struct subsys_interface *sif,
1238 unsigned int cpu = dev->id, cpus;
1240 unsigned long flags;
1241 struct cpufreq_policy *policy;
1243 read_lock_irqsave(&cpufreq_driver_lock, flags);
1244 policy = per_cpu(cpufreq_cpu_data, cpu);
1245 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1248 pr_debug("%s: No cpu_data found\n", __func__);
1252 down_write(&policy->rwsem);
1253 cpus = cpumask_weight(policy->cpus);
1256 cpumask_clear_cpu(cpu, policy->cpus);
1257 up_write(&policy->rwsem);
1259 /* If cpu is last user of policy, free policy */
1262 ret = __cpufreq_governor(policy,
1263 CPUFREQ_GOV_POLICY_EXIT);
1265 pr_err("%s: Failed to exit governor\n",
1272 cpufreq_policy_put_kobj(policy);
1275 * Perform the ->exit() even during light-weight tear-down,
1276 * since this is a core component, and is essential for the
1277 * subsequent light-weight ->init() to succeed.
1279 if (cpufreq_driver->exit)
1280 cpufreq_driver->exit(policy);
1282 /* Remove policy from list of active policies */
1283 write_lock_irqsave(&cpufreq_driver_lock, flags);
1284 list_del(&policy->policy_list);
1285 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1288 cpufreq_policy_free(policy);
1291 if ((ret = __cpufreq_governor(policy, CPUFREQ_GOV_START)) ||
1292 (ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))) {
1293 pr_err("%s: Failed to start governor\n",
1300 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1305 * cpufreq_remove_dev - remove a CPU device
1307 * Removes the cpufreq interface for a CPU device.
1309 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1311 unsigned int cpu = dev->id;
1314 if (cpu_is_offline(cpu))
1317 ret = __cpufreq_remove_dev_prepare(dev, sif, false);
1320 ret = __cpufreq_remove_dev_finish(dev, sif, false);
1325 static void handle_update(struct work_struct *work)
1327 struct cpufreq_policy *policy =
1328 container_of(work, struct cpufreq_policy, update);
1329 unsigned int cpu = policy->cpu;
1330 pr_debug("handle_update for cpu %u called\n", cpu);
1331 cpufreq_update_policy(cpu);
1335 * cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1338 * @old_freq: CPU frequency the kernel thinks the CPU runs at
1339 * @new_freq: CPU frequency the CPU actually runs at
1341 * We adjust to current frequency first, and need to clean up later.
1342 * So either call to cpufreq_update_policy() or schedule handle_update()).
1344 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1345 unsigned int new_freq)
1347 struct cpufreq_policy *policy;
1348 struct cpufreq_freqs freqs;
1349 unsigned long flags;
1351 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1352 "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1354 freqs.old = old_freq;
1355 freqs.new = new_freq;
1357 read_lock_irqsave(&cpufreq_driver_lock, flags);
1358 policy = per_cpu(cpufreq_cpu_data, cpu);
1359 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1361 cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
1362 cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
1366 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1369 * This is the last known freq, without actually getting it from the driver.
1370 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1372 unsigned int cpufreq_quick_get(unsigned int cpu)
1374 struct cpufreq_policy *policy;
1375 unsigned int ret_freq = 0;
1377 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1378 return cpufreq_driver->get(cpu);
1380 policy = cpufreq_cpu_get(cpu);
1382 ret_freq = policy->cur;
1383 cpufreq_cpu_put(policy);
1388 EXPORT_SYMBOL(cpufreq_quick_get);
1391 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1394 * Just return the max possible frequency for a given CPU.
1396 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1398 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1399 unsigned int ret_freq = 0;
1402 ret_freq = policy->max;
1403 cpufreq_cpu_put(policy);
1408 EXPORT_SYMBOL(cpufreq_quick_get_max);
1410 static unsigned int __cpufreq_get(unsigned int cpu)
1412 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1413 unsigned int ret_freq = 0;
1415 if (!cpufreq_driver->get)
1418 ret_freq = cpufreq_driver->get(cpu);
1420 if (ret_freq && policy->cur &&
1421 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1422 /* verify no discrepancy between actual and
1423 saved value exists */
1424 if (unlikely(ret_freq != policy->cur)) {
1425 cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1426 schedule_work(&policy->update);
1434 * cpufreq_get - get the current CPU frequency (in kHz)
1437 * Get the CPU current (static) CPU frequency
1439 unsigned int cpufreq_get(unsigned int cpu)
1441 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1442 unsigned int ret_freq = 0;
1444 if (cpufreq_disabled() || !cpufreq_driver)
1449 if (!down_read_trylock(&cpufreq_rwsem))
1452 down_read(&policy->rwsem);
1454 ret_freq = __cpufreq_get(cpu);
1456 up_read(&policy->rwsem);
1457 up_read(&cpufreq_rwsem);
1461 EXPORT_SYMBOL(cpufreq_get);
1463 static struct subsys_interface cpufreq_interface = {
1465 .subsys = &cpu_subsys,
1466 .add_dev = cpufreq_add_dev,
1467 .remove_dev = cpufreq_remove_dev,
1471 * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1473 * This function is only executed for the boot processor. The other CPUs
1474 * have been put offline by means of CPU hotplug.
1476 static int cpufreq_bp_suspend(void)
1480 int cpu = smp_processor_id();
1481 struct cpufreq_policy *policy;
1483 pr_debug("suspending cpu %u\n", cpu);
1485 /* If there's no policy for the boot CPU, we have nothing to do. */
1486 policy = cpufreq_cpu_get(cpu);
1490 if (cpufreq_driver->suspend) {
1491 ret = cpufreq_driver->suspend(policy);
1493 printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1494 "step on CPU %u\n", policy->cpu);
1497 cpufreq_cpu_put(policy);
1502 * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1504 * 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1505 * 2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1506 * restored. It will verify that the current freq is in sync with
1507 * what we believe it to be. This is a bit later than when it
1508 * should be, but nonethteless it's better than calling
1509 * cpufreq_driver->get() here which might re-enable interrupts...
1511 * This function is only executed for the boot CPU. The other CPUs have not
1512 * been turned on yet.
1514 static void cpufreq_bp_resume(void)
1518 int cpu = smp_processor_id();
1519 struct cpufreq_policy *policy;
1521 pr_debug("resuming cpu %u\n", cpu);
1523 /* If there's no policy for the boot CPU, we have nothing to do. */
1524 policy = cpufreq_cpu_get(cpu);
1528 if (cpufreq_driver->resume) {
1529 ret = cpufreq_driver->resume(policy);
1531 printk(KERN_ERR "cpufreq: resume failed in ->resume "
1532 "step on CPU %u\n", policy->cpu);
1537 schedule_work(&policy->update);
1540 cpufreq_cpu_put(policy);
1543 static struct syscore_ops cpufreq_syscore_ops = {
1544 .suspend = cpufreq_bp_suspend,
1545 .resume = cpufreq_bp_resume,
1549 * cpufreq_get_current_driver - return current driver's name
1551 * Return the name string of the currently loaded cpufreq driver
1554 const char *cpufreq_get_current_driver(void)
1557 return cpufreq_driver->name;
1561 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1563 /*********************************************************************
1564 * NOTIFIER LISTS INTERFACE *
1565 *********************************************************************/
1568 * cpufreq_register_notifier - register a driver with cpufreq
1569 * @nb: notifier function to register
1570 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1572 * Add a driver to one of two lists: either a list of drivers that
1573 * are notified about clock rate changes (once before and once after
1574 * the transition), or a list of drivers that are notified about
1575 * changes in cpufreq policy.
1577 * This function may sleep, and has the same return conditions as
1578 * blocking_notifier_chain_register.
1580 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1584 if (cpufreq_disabled())
1587 WARN_ON(!init_cpufreq_transition_notifier_list_called);
1590 case CPUFREQ_TRANSITION_NOTIFIER:
1591 ret = srcu_notifier_chain_register(
1592 &cpufreq_transition_notifier_list, nb);
1594 case CPUFREQ_POLICY_NOTIFIER:
1595 ret = blocking_notifier_chain_register(
1596 &cpufreq_policy_notifier_list, nb);
1604 EXPORT_SYMBOL(cpufreq_register_notifier);
1607 * cpufreq_unregister_notifier - unregister a driver with cpufreq
1608 * @nb: notifier block to be unregistered
1609 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1611 * Remove a driver from the CPU frequency notifier list.
1613 * This function may sleep, and has the same return conditions as
1614 * blocking_notifier_chain_unregister.
1616 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1620 if (cpufreq_disabled())
1624 case CPUFREQ_TRANSITION_NOTIFIER:
1625 ret = srcu_notifier_chain_unregister(
1626 &cpufreq_transition_notifier_list, nb);
1628 case CPUFREQ_POLICY_NOTIFIER:
1629 ret = blocking_notifier_chain_unregister(
1630 &cpufreq_policy_notifier_list, nb);
1638 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1641 /*********************************************************************
1643 *********************************************************************/
1645 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1646 unsigned int target_freq,
1647 unsigned int relation)
1649 int retval = -EINVAL;
1650 unsigned int old_target_freq = target_freq;
1652 if (cpufreq_disabled())
1655 /* Make sure that target_freq is within supported range */
1656 if (target_freq > policy->max)
1657 target_freq = policy->max;
1658 if (target_freq < policy->min)
1659 target_freq = policy->min;
1661 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1662 policy->cpu, target_freq, relation, old_target_freq);
1665 * This might look like a redundant call as we are checking it again
1666 * after finding index. But it is left intentionally for cases where
1667 * exactly same freq is called again and so we can save on few function
1670 if (target_freq == policy->cur)
1673 if (cpufreq_driver->target)
1674 retval = cpufreq_driver->target(policy, target_freq, relation);
1675 else if (cpufreq_driver->target_index) {
1676 struct cpufreq_frequency_table *freq_table;
1677 struct cpufreq_freqs freqs;
1681 freq_table = cpufreq_frequency_get_table(policy->cpu);
1682 if (unlikely(!freq_table)) {
1683 pr_err("%s: Unable to find freq_table\n", __func__);
1687 retval = cpufreq_frequency_table_target(policy, freq_table,
1688 target_freq, relation, &index);
1689 if (unlikely(retval)) {
1690 pr_err("%s: Unable to find matching freq\n", __func__);
1694 if (freq_table[index].frequency == policy->cur) {
1699 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1702 freqs.old = policy->cur;
1703 freqs.new = freq_table[index].frequency;
1706 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1707 __func__, policy->cpu, freqs.old,
1710 cpufreq_notify_transition(policy, &freqs,
1714 retval = cpufreq_driver->target_index(policy, index);
1716 pr_err("%s: Failed to change cpu frequency: %d\n",
1721 * Notify with old freq in case we failed to change
1725 freqs.new = freqs.old;
1727 cpufreq_notify_transition(policy, &freqs,
1728 CPUFREQ_POSTCHANGE);
1735 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1737 int cpufreq_driver_target(struct cpufreq_policy *policy,
1738 unsigned int target_freq,
1739 unsigned int relation)
1743 down_write(&policy->rwsem);
1745 ret = __cpufreq_driver_target(policy, target_freq, relation);
1747 up_write(&policy->rwsem);
1751 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1754 * when "event" is CPUFREQ_GOV_LIMITS
1757 static int __cpufreq_governor(struct cpufreq_policy *policy,
1762 /* Only must be defined when default governor is known to have latency
1763 restrictions, like e.g. conservative or ondemand.
1764 That this is the case is already ensured in Kconfig
1766 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1767 struct cpufreq_governor *gov = &cpufreq_gov_performance;
1769 struct cpufreq_governor *gov = NULL;
1772 if (policy->governor->max_transition_latency &&
1773 policy->cpuinfo.transition_latency >
1774 policy->governor->max_transition_latency) {
1778 printk(KERN_WARNING "%s governor failed, too long"
1779 " transition latency of HW, fallback"
1780 " to %s governor\n",
1781 policy->governor->name,
1783 policy->governor = gov;
1787 if (event == CPUFREQ_GOV_POLICY_INIT)
1788 if (!try_module_get(policy->governor->owner))
1791 pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1792 policy->cpu, event);
1794 mutex_lock(&cpufreq_governor_lock);
1795 if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
1796 || (!policy->governor_enabled
1797 && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
1798 mutex_unlock(&cpufreq_governor_lock);
1802 if (event == CPUFREQ_GOV_STOP)
1803 policy->governor_enabled = false;
1804 else if (event == CPUFREQ_GOV_START)
1805 policy->governor_enabled = true;
1807 mutex_unlock(&cpufreq_governor_lock);
1809 ret = policy->governor->governor(policy, event);
1812 if (event == CPUFREQ_GOV_POLICY_INIT)
1813 policy->governor->initialized++;
1814 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1815 policy->governor->initialized--;
1817 /* Restore original values */
1818 mutex_lock(&cpufreq_governor_lock);
1819 if (event == CPUFREQ_GOV_STOP)
1820 policy->governor_enabled = true;
1821 else if (event == CPUFREQ_GOV_START)
1822 policy->governor_enabled = false;
1823 mutex_unlock(&cpufreq_governor_lock);
1826 if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
1827 ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
1828 module_put(policy->governor->owner);
1833 int cpufreq_register_governor(struct cpufreq_governor *governor)
1840 if (cpufreq_disabled())
1843 mutex_lock(&cpufreq_governor_mutex);
1845 governor->initialized = 0;
1847 if (__find_governor(governor->name) == NULL) {
1849 list_add(&governor->governor_list, &cpufreq_governor_list);
1852 mutex_unlock(&cpufreq_governor_mutex);
1855 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1857 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1859 #ifdef CONFIG_HOTPLUG_CPU
1866 if (cpufreq_disabled())
1869 #ifdef CONFIG_HOTPLUG_CPU
1870 for_each_present_cpu(cpu) {
1871 if (cpu_online(cpu))
1873 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1874 strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1878 mutex_lock(&cpufreq_governor_mutex);
1879 list_del(&governor->governor_list);
1880 mutex_unlock(&cpufreq_governor_mutex);
1883 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1886 /*********************************************************************
1887 * POLICY INTERFACE *
1888 *********************************************************************/
1891 * cpufreq_get_policy - get the current cpufreq_policy
1892 * @policy: struct cpufreq_policy into which the current cpufreq_policy
1895 * Reads the current cpufreq policy.
1897 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1899 struct cpufreq_policy *cpu_policy;
1903 cpu_policy = cpufreq_cpu_get(cpu);
1907 memcpy(policy, cpu_policy, sizeof(*policy));
1909 cpufreq_cpu_put(cpu_policy);
1912 EXPORT_SYMBOL(cpufreq_get_policy);
1915 * policy : current policy.
1916 * new_policy: policy to be set.
1918 static int cpufreq_set_policy(struct cpufreq_policy *policy,
1919 struct cpufreq_policy *new_policy)
1921 int ret = 0, failed = 1;
1923 pr_debug("setting new policy for CPU %u: %u - %u kHz\n", new_policy->cpu,
1924 new_policy->min, new_policy->max);
1926 memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
1928 if (new_policy->min > policy->max || new_policy->max < policy->min) {
1933 /* verify the cpu speed can be set within this limit */
1934 ret = cpufreq_driver->verify(new_policy);
1938 /* adjust if necessary - all reasons */
1939 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1940 CPUFREQ_ADJUST, new_policy);
1942 /* adjust if necessary - hardware incompatibility*/
1943 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1944 CPUFREQ_INCOMPATIBLE, new_policy);
1947 * verify the cpu speed can be set within this limit, which might be
1948 * different to the first one
1950 ret = cpufreq_driver->verify(new_policy);
1954 /* notification of the new policy */
1955 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1956 CPUFREQ_NOTIFY, new_policy);
1958 policy->min = new_policy->min;
1959 policy->max = new_policy->max;
1961 pr_debug("new min and max freqs are %u - %u kHz\n",
1962 policy->min, policy->max);
1964 if (cpufreq_driver->setpolicy) {
1965 policy->policy = new_policy->policy;
1966 pr_debug("setting range\n");
1967 ret = cpufreq_driver->setpolicy(new_policy);
1969 if (new_policy->governor != policy->governor) {
1970 /* save old, working values */
1971 struct cpufreq_governor *old_gov = policy->governor;
1973 pr_debug("governor switch\n");
1975 /* end old governor */
1976 if (policy->governor) {
1977 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1978 up_write(&policy->rwsem);
1979 __cpufreq_governor(policy,
1980 CPUFREQ_GOV_POLICY_EXIT);
1981 down_write(&policy->rwsem);
1984 /* start new governor */
1985 policy->governor = new_policy->governor;
1986 if (!__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT)) {
1987 if (!__cpufreq_governor(policy, CPUFREQ_GOV_START)) {
1990 up_write(&policy->rwsem);
1991 __cpufreq_governor(policy,
1992 CPUFREQ_GOV_POLICY_EXIT);
1993 down_write(&policy->rwsem);
1998 /* new governor failed, so re-start old one */
1999 pr_debug("starting governor %s failed\n",
2000 policy->governor->name);
2002 policy->governor = old_gov;
2003 __cpufreq_governor(policy,
2004 CPUFREQ_GOV_POLICY_INIT);
2005 __cpufreq_governor(policy,
2011 /* might be a policy change, too, so fall through */
2013 pr_debug("governor: change or update limits\n");
2014 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2022 * cpufreq_update_policy - re-evaluate an existing cpufreq policy
2023 * @cpu: CPU which shall be re-evaluated
2025 * Useful for policy notifiers which have different necessities
2026 * at different times.
2028 int cpufreq_update_policy(unsigned int cpu)
2030 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2031 struct cpufreq_policy new_policy;
2039 down_write(&policy->rwsem);
2041 pr_debug("updating policy for CPU %u\n", cpu);
2042 memcpy(&new_policy, policy, sizeof(*policy));
2043 new_policy.min = policy->user_policy.min;
2044 new_policy.max = policy->user_policy.max;
2045 new_policy.policy = policy->user_policy.policy;
2046 new_policy.governor = policy->user_policy.governor;
2049 * BIOS might change freq behind our back
2050 * -> ask driver for current freq and notify governors about a change
2052 if (cpufreq_driver->get) {
2053 new_policy.cur = cpufreq_driver->get(cpu);
2055 pr_debug("Driver did not initialize current freq");
2056 policy->cur = new_policy.cur;
2058 if (policy->cur != new_policy.cur && has_target())
2059 cpufreq_out_of_sync(cpu, policy->cur,
2064 ret = cpufreq_set_policy(policy, &new_policy);
2066 up_write(&policy->rwsem);
2068 cpufreq_cpu_put(policy);
2072 EXPORT_SYMBOL(cpufreq_update_policy);
2074 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2075 unsigned long action, void *hcpu)
2077 unsigned int cpu = (unsigned long)hcpu;
2079 bool frozen = false;
2081 dev = get_cpu_device(cpu);
2084 if (action & CPU_TASKS_FROZEN)
2087 switch (action & ~CPU_TASKS_FROZEN) {
2089 __cpufreq_add_dev(dev, NULL, frozen);
2090 cpufreq_update_policy(cpu);
2093 case CPU_DOWN_PREPARE:
2094 __cpufreq_remove_dev_prepare(dev, NULL, frozen);
2098 __cpufreq_remove_dev_finish(dev, NULL, frozen);
2101 case CPU_DOWN_FAILED:
2102 __cpufreq_add_dev(dev, NULL, frozen);
2109 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2110 .notifier_call = cpufreq_cpu_callback,
2113 /*********************************************************************
2114 * REGISTER / UNREGISTER CPUFREQ DRIVER *
2115 *********************************************************************/
2118 * cpufreq_register_driver - register a CPU Frequency driver
2119 * @driver_data: A struct cpufreq_driver containing the values#
2120 * submitted by the CPU Frequency driver.
2122 * Registers a CPU Frequency driver to this core code. This code
2123 * returns zero on success, -EBUSY when another driver got here first
2124 * (and isn't unregistered in the meantime).
2127 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2129 unsigned long flags;
2132 if (cpufreq_disabled())
2135 if (!driver_data || !driver_data->verify || !driver_data->init ||
2136 !(driver_data->setpolicy || driver_data->target_index ||
2137 driver_data->target))
2140 pr_debug("trying to register driver %s\n", driver_data->name);
2142 if (driver_data->setpolicy)
2143 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2145 write_lock_irqsave(&cpufreq_driver_lock, flags);
2146 if (cpufreq_driver) {
2147 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2150 cpufreq_driver = driver_data;
2151 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2153 ret = subsys_interface_register(&cpufreq_interface);
2155 goto err_null_driver;
2157 if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
2161 /* check for at least one working CPU */
2162 for (i = 0; i < nr_cpu_ids; i++)
2163 if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
2168 /* if all ->init() calls failed, unregister */
2170 pr_debug("no CPU initialized for driver %s\n",
2176 register_hotcpu_notifier(&cpufreq_cpu_notifier);
2177 pr_debug("driver %s up and running\n", driver_data->name);
2181 subsys_interface_unregister(&cpufreq_interface);
2183 write_lock_irqsave(&cpufreq_driver_lock, flags);
2184 cpufreq_driver = NULL;
2185 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2188 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2191 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2193 * Unregister the current CPUFreq driver. Only call this if you have
2194 * the right to do so, i.e. if you have succeeded in initialising before!
2195 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2196 * currently not initialised.
2198 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2200 unsigned long flags;
2202 if (!cpufreq_driver || (driver != cpufreq_driver))
2205 pr_debug("unregistering driver %s\n", driver->name);
2207 subsys_interface_unregister(&cpufreq_interface);
2208 unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2210 down_write(&cpufreq_rwsem);
2211 write_lock_irqsave(&cpufreq_driver_lock, flags);
2213 cpufreq_driver = NULL;
2215 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2216 up_write(&cpufreq_rwsem);
2220 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2222 static int __init cpufreq_core_init(void)
2224 if (cpufreq_disabled())
2227 cpufreq_global_kobject = kobject_create();
2228 BUG_ON(!cpufreq_global_kobject);
2229 register_syscore_ops(&cpufreq_syscore_ops);
2233 core_initcall(cpufreq_core_init);