Loading...
1/*
2 * linux/drivers/cpufreq/cpufreq.c
3 *
4 * Copyright (C) 2001 Russell King
5 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6 *
7 * Oct 2005 - Ashok Raj <ashok.raj@intel.com>
8 * Added handling for CPU hotplug
9 * Feb 2006 - Jacob Shin <jacob.shin@amd.com>
10 * Fix handling for CPU hotplug -- affected CPUs
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 *
16 */
17
18#include <linux/kernel.h>
19#include <linux/module.h>
20#include <linux/init.h>
21#include <linux/notifier.h>
22#include <linux/cpufreq.h>
23#include <linux/delay.h>
24#include <linux/interrupt.h>
25#include <linux/spinlock.h>
26#include <linux/device.h>
27#include <linux/slab.h>
28#include <linux/cpu.h>
29#include <linux/completion.h>
30#include <linux/mutex.h>
31#include <linux/syscore_ops.h>
32
33#include <trace/events/power.h>
34
35/**
36 * The "cpufreq driver" - the arch- or hardware-dependent low
37 * level driver of CPUFreq support, and its spinlock. This lock
38 * also protects the cpufreq_cpu_data array.
39 */
40static struct cpufreq_driver *cpufreq_driver;
41static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
42#ifdef CONFIG_HOTPLUG_CPU
43/* This one keeps track of the previously set governor of a removed CPU */
44static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
45#endif
46static DEFINE_SPINLOCK(cpufreq_driver_lock);
47
48/*
49 * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
50 * all cpufreq/hotplug/workqueue/etc related lock issues.
51 *
52 * The rules for this semaphore:
53 * - Any routine that wants to read from the policy structure will
54 * do a down_read on this semaphore.
55 * - Any routine that will write to the policy structure and/or may take away
56 * the policy altogether (eg. CPU hotplug), will hold this lock in write
57 * mode before doing so.
58 *
59 * Additional rules:
60 * - All holders of the lock should check to make sure that the CPU they
61 * are concerned with are online after they get the lock.
62 * - Governor routines that can be called in cpufreq hotplug path should not
63 * take this sem as top level hotplug notifier handler takes this.
64 * - Lock should not be held across
65 * __cpufreq_governor(data, CPUFREQ_GOV_STOP);
66 */
67static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
68static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
69
70#define lock_policy_rwsem(mode, cpu) \
71static int lock_policy_rwsem_##mode \
72(int cpu) \
73{ \
74 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
75 BUG_ON(policy_cpu == -1); \
76 down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
77 if (unlikely(!cpu_online(cpu))) { \
78 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
79 return -1; \
80 } \
81 \
82 return 0; \
83}
84
85lock_policy_rwsem(read, cpu);
86
87lock_policy_rwsem(write, cpu);
88
89static void unlock_policy_rwsem_read(int cpu)
90{
91 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
92 BUG_ON(policy_cpu == -1);
93 up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
94}
95
96static void unlock_policy_rwsem_write(int cpu)
97{
98 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
99 BUG_ON(policy_cpu == -1);
100 up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
101}
102
103
104/* internal prototypes */
105static int __cpufreq_governor(struct cpufreq_policy *policy,
106 unsigned int event);
107static unsigned int __cpufreq_get(unsigned int cpu);
108static void handle_update(struct work_struct *work);
109
110/**
111 * Two notifier lists: the "policy" list is involved in the
112 * validation process for a new CPU frequency policy; the
113 * "transition" list for kernel code that needs to handle
114 * changes to devices when the CPU clock speed changes.
115 * The mutex locks both lists.
116 */
117static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
118static struct srcu_notifier_head cpufreq_transition_notifier_list;
119
120static bool init_cpufreq_transition_notifier_list_called;
121static int __init init_cpufreq_transition_notifier_list(void)
122{
123 srcu_init_notifier_head(&cpufreq_transition_notifier_list);
124 init_cpufreq_transition_notifier_list_called = true;
125 return 0;
126}
127pure_initcall(init_cpufreq_transition_notifier_list);
128
129static LIST_HEAD(cpufreq_governor_list);
130static DEFINE_MUTEX(cpufreq_governor_mutex);
131
132struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
133{
134 struct cpufreq_policy *data;
135 unsigned long flags;
136
137 if (cpu >= nr_cpu_ids)
138 goto err_out;
139
140 /* get the cpufreq driver */
141 spin_lock_irqsave(&cpufreq_driver_lock, flags);
142
143 if (!cpufreq_driver)
144 goto err_out_unlock;
145
146 if (!try_module_get(cpufreq_driver->owner))
147 goto err_out_unlock;
148
149
150 /* get the CPU */
151 data = per_cpu(cpufreq_cpu_data, cpu);
152
153 if (!data)
154 goto err_out_put_module;
155
156 if (!kobject_get(&data->kobj))
157 goto err_out_put_module;
158
159 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
160 return data;
161
162err_out_put_module:
163 module_put(cpufreq_driver->owner);
164err_out_unlock:
165 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
166err_out:
167 return NULL;
168}
169EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
170
171
172void cpufreq_cpu_put(struct cpufreq_policy *data)
173{
174 kobject_put(&data->kobj);
175 module_put(cpufreq_driver->owner);
176}
177EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
178
179
180/*********************************************************************
181 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
182 *********************************************************************/
183
184/**
185 * adjust_jiffies - adjust the system "loops_per_jiffy"
186 *
187 * This function alters the system "loops_per_jiffy" for the clock
188 * speed change. Note that loops_per_jiffy cannot be updated on SMP
189 * systems as each CPU might be scaled differently. So, use the arch
190 * per-CPU loops_per_jiffy value wherever possible.
191 */
192#ifndef CONFIG_SMP
193static unsigned long l_p_j_ref;
194static unsigned int l_p_j_ref_freq;
195
196static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
197{
198 if (ci->flags & CPUFREQ_CONST_LOOPS)
199 return;
200
201 if (!l_p_j_ref_freq) {
202 l_p_j_ref = loops_per_jiffy;
203 l_p_j_ref_freq = ci->old;
204 pr_debug("saving %lu as reference value for loops_per_jiffy; "
205 "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
206 }
207 if ((val == CPUFREQ_PRECHANGE && ci->old < ci->new) ||
208 (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
209 (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
210 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
211 ci->new);
212 pr_debug("scaling loops_per_jiffy to %lu "
213 "for frequency %u kHz\n", loops_per_jiffy, ci->new);
214 }
215}
216#else
217static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
218{
219 return;
220}
221#endif
222
223
224/**
225 * cpufreq_notify_transition - call notifier chain and adjust_jiffies
226 * on frequency transition.
227 *
228 * This function calls the transition notifiers and the "adjust_jiffies"
229 * function. It is called twice on all CPU frequency changes that have
230 * external effects.
231 */
232void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
233{
234 struct cpufreq_policy *policy;
235
236 BUG_ON(irqs_disabled());
237
238 freqs->flags = cpufreq_driver->flags;
239 pr_debug("notification %u of frequency transition to %u kHz\n",
240 state, freqs->new);
241
242 policy = per_cpu(cpufreq_cpu_data, freqs->cpu);
243 switch (state) {
244
245 case CPUFREQ_PRECHANGE:
246 /* detect if the driver reported a value as "old frequency"
247 * which is not equal to what the cpufreq core thinks is
248 * "old frequency".
249 */
250 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
251 if ((policy) && (policy->cpu == freqs->cpu) &&
252 (policy->cur) && (policy->cur != freqs->old)) {
253 pr_debug("Warning: CPU frequency is"
254 " %u, cpufreq assumed %u kHz.\n",
255 freqs->old, policy->cur);
256 freqs->old = policy->cur;
257 }
258 }
259 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
260 CPUFREQ_PRECHANGE, freqs);
261 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
262 break;
263
264 case CPUFREQ_POSTCHANGE:
265 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
266 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
267 (unsigned long)freqs->cpu);
268 trace_power_frequency(POWER_PSTATE, freqs->new, freqs->cpu);
269 trace_cpu_frequency(freqs->new, freqs->cpu);
270 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
271 CPUFREQ_POSTCHANGE, freqs);
272 if (likely(policy) && likely(policy->cpu == freqs->cpu))
273 policy->cur = freqs->new;
274 break;
275 }
276}
277EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
278
279
280
281/*********************************************************************
282 * SYSFS INTERFACE *
283 *********************************************************************/
284
285static struct cpufreq_governor *__find_governor(const char *str_governor)
286{
287 struct cpufreq_governor *t;
288
289 list_for_each_entry(t, &cpufreq_governor_list, governor_list)
290 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
291 return t;
292
293 return NULL;
294}
295
296/**
297 * cpufreq_parse_governor - parse a governor string
298 */
299static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
300 struct cpufreq_governor **governor)
301{
302 int err = -EINVAL;
303
304 if (!cpufreq_driver)
305 goto out;
306
307 if (cpufreq_driver->setpolicy) {
308 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
309 *policy = CPUFREQ_POLICY_PERFORMANCE;
310 err = 0;
311 } else if (!strnicmp(str_governor, "powersave",
312 CPUFREQ_NAME_LEN)) {
313 *policy = CPUFREQ_POLICY_POWERSAVE;
314 err = 0;
315 }
316 } else if (cpufreq_driver->target) {
317 struct cpufreq_governor *t;
318
319 mutex_lock(&cpufreq_governor_mutex);
320
321 t = __find_governor(str_governor);
322
323 if (t == NULL) {
324 int ret;
325
326 mutex_unlock(&cpufreq_governor_mutex);
327 ret = request_module("cpufreq_%s", str_governor);
328 mutex_lock(&cpufreq_governor_mutex);
329
330 if (ret == 0)
331 t = __find_governor(str_governor);
332 }
333
334 if (t != NULL) {
335 *governor = t;
336 err = 0;
337 }
338
339 mutex_unlock(&cpufreq_governor_mutex);
340 }
341out:
342 return err;
343}
344
345
346/**
347 * cpufreq_per_cpu_attr_read() / show_##file_name() -
348 * print out cpufreq information
349 *
350 * Write out information from cpufreq_driver->policy[cpu]; object must be
351 * "unsigned int".
352 */
353
354#define show_one(file_name, object) \
355static ssize_t show_##file_name \
356(struct cpufreq_policy *policy, char *buf) \
357{ \
358 return sprintf(buf, "%u\n", policy->object); \
359}
360
361show_one(cpuinfo_min_freq, cpuinfo.min_freq);
362show_one(cpuinfo_max_freq, cpuinfo.max_freq);
363show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
364show_one(scaling_min_freq, min);
365show_one(scaling_max_freq, max);
366show_one(scaling_cur_freq, cur);
367
368static int __cpufreq_set_policy(struct cpufreq_policy *data,
369 struct cpufreq_policy *policy);
370
371/**
372 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
373 */
374#define store_one(file_name, object) \
375static ssize_t store_##file_name \
376(struct cpufreq_policy *policy, const char *buf, size_t count) \
377{ \
378 unsigned int ret = -EINVAL; \
379 struct cpufreq_policy new_policy; \
380 \
381 ret = cpufreq_get_policy(&new_policy, policy->cpu); \
382 if (ret) \
383 return -EINVAL; \
384 \
385 ret = sscanf(buf, "%u", &new_policy.object); \
386 if (ret != 1) \
387 return -EINVAL; \
388 \
389 ret = __cpufreq_set_policy(policy, &new_policy); \
390 policy->user_policy.object = policy->object; \
391 \
392 return ret ? ret : count; \
393}
394
395store_one(scaling_min_freq, min);
396store_one(scaling_max_freq, max);
397
398/**
399 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
400 */
401static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
402 char *buf)
403{
404 unsigned int cur_freq = __cpufreq_get(policy->cpu);
405 if (!cur_freq)
406 return sprintf(buf, "<unknown>");
407 return sprintf(buf, "%u\n", cur_freq);
408}
409
410
411/**
412 * show_scaling_governor - show the current policy for the specified CPU
413 */
414static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
415{
416 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
417 return sprintf(buf, "powersave\n");
418 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
419 return sprintf(buf, "performance\n");
420 else if (policy->governor)
421 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n",
422 policy->governor->name);
423 return -EINVAL;
424}
425
426
427/**
428 * store_scaling_governor - store policy for the specified CPU
429 */
430static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
431 const char *buf, size_t count)
432{
433 unsigned int ret = -EINVAL;
434 char str_governor[16];
435 struct cpufreq_policy new_policy;
436
437 ret = cpufreq_get_policy(&new_policy, policy->cpu);
438 if (ret)
439 return ret;
440
441 ret = sscanf(buf, "%15s", str_governor);
442 if (ret != 1)
443 return -EINVAL;
444
445 if (cpufreq_parse_governor(str_governor, &new_policy.policy,
446 &new_policy.governor))
447 return -EINVAL;
448
449 /* Do not use cpufreq_set_policy here or the user_policy.max
450 will be wrongly overridden */
451 ret = __cpufreq_set_policy(policy, &new_policy);
452
453 policy->user_policy.policy = policy->policy;
454 policy->user_policy.governor = policy->governor;
455
456 if (ret)
457 return ret;
458 else
459 return count;
460}
461
462/**
463 * show_scaling_driver - show the cpufreq driver currently loaded
464 */
465static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
466{
467 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
468}
469
470/**
471 * show_scaling_available_governors - show the available CPUfreq governors
472 */
473static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
474 char *buf)
475{
476 ssize_t i = 0;
477 struct cpufreq_governor *t;
478
479 if (!cpufreq_driver->target) {
480 i += sprintf(buf, "performance powersave");
481 goto out;
482 }
483
484 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
485 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
486 - (CPUFREQ_NAME_LEN + 2)))
487 goto out;
488 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
489 }
490out:
491 i += sprintf(&buf[i], "\n");
492 return i;
493}
494
495static ssize_t show_cpus(const struct cpumask *mask, char *buf)
496{
497 ssize_t i = 0;
498 unsigned int cpu;
499
500 for_each_cpu(cpu, mask) {
501 if (i)
502 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
503 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
504 if (i >= (PAGE_SIZE - 5))
505 break;
506 }
507 i += sprintf(&buf[i], "\n");
508 return i;
509}
510
511/**
512 * show_related_cpus - show the CPUs affected by each transition even if
513 * hw coordination is in use
514 */
515static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
516{
517 if (cpumask_empty(policy->related_cpus))
518 return show_cpus(policy->cpus, buf);
519 return show_cpus(policy->related_cpus, buf);
520}
521
522/**
523 * show_affected_cpus - show the CPUs affected by each transition
524 */
525static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
526{
527 return show_cpus(policy->cpus, buf);
528}
529
530static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
531 const char *buf, size_t count)
532{
533 unsigned int freq = 0;
534 unsigned int ret;
535
536 if (!policy->governor || !policy->governor->store_setspeed)
537 return -EINVAL;
538
539 ret = sscanf(buf, "%u", &freq);
540 if (ret != 1)
541 return -EINVAL;
542
543 policy->governor->store_setspeed(policy, freq);
544
545 return count;
546}
547
548static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
549{
550 if (!policy->governor || !policy->governor->show_setspeed)
551 return sprintf(buf, "<unsupported>\n");
552
553 return policy->governor->show_setspeed(policy, buf);
554}
555
556/**
557 * show_scaling_driver - show the current cpufreq HW/BIOS limitation
558 */
559static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
560{
561 unsigned int limit;
562 int ret;
563 if (cpufreq_driver->bios_limit) {
564 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
565 if (!ret)
566 return sprintf(buf, "%u\n", limit);
567 }
568 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
569}
570
571cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
572cpufreq_freq_attr_ro(cpuinfo_min_freq);
573cpufreq_freq_attr_ro(cpuinfo_max_freq);
574cpufreq_freq_attr_ro(cpuinfo_transition_latency);
575cpufreq_freq_attr_ro(scaling_available_governors);
576cpufreq_freq_attr_ro(scaling_driver);
577cpufreq_freq_attr_ro(scaling_cur_freq);
578cpufreq_freq_attr_ro(bios_limit);
579cpufreq_freq_attr_ro(related_cpus);
580cpufreq_freq_attr_ro(affected_cpus);
581cpufreq_freq_attr_rw(scaling_min_freq);
582cpufreq_freq_attr_rw(scaling_max_freq);
583cpufreq_freq_attr_rw(scaling_governor);
584cpufreq_freq_attr_rw(scaling_setspeed);
585
586static struct attribute *default_attrs[] = {
587 &cpuinfo_min_freq.attr,
588 &cpuinfo_max_freq.attr,
589 &cpuinfo_transition_latency.attr,
590 &scaling_min_freq.attr,
591 &scaling_max_freq.attr,
592 &affected_cpus.attr,
593 &related_cpus.attr,
594 &scaling_governor.attr,
595 &scaling_driver.attr,
596 &scaling_available_governors.attr,
597 &scaling_setspeed.attr,
598 NULL
599};
600
601struct kobject *cpufreq_global_kobject;
602EXPORT_SYMBOL(cpufreq_global_kobject);
603
604#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
605#define to_attr(a) container_of(a, struct freq_attr, attr)
606
607static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
608{
609 struct cpufreq_policy *policy = to_policy(kobj);
610 struct freq_attr *fattr = to_attr(attr);
611 ssize_t ret = -EINVAL;
612 policy = cpufreq_cpu_get(policy->cpu);
613 if (!policy)
614 goto no_policy;
615
616 if (lock_policy_rwsem_read(policy->cpu) < 0)
617 goto fail;
618
619 if (fattr->show)
620 ret = fattr->show(policy, buf);
621 else
622 ret = -EIO;
623
624 unlock_policy_rwsem_read(policy->cpu);
625fail:
626 cpufreq_cpu_put(policy);
627no_policy:
628 return ret;
629}
630
631static ssize_t store(struct kobject *kobj, struct attribute *attr,
632 const char *buf, size_t count)
633{
634 struct cpufreq_policy *policy = to_policy(kobj);
635 struct freq_attr *fattr = to_attr(attr);
636 ssize_t ret = -EINVAL;
637 policy = cpufreq_cpu_get(policy->cpu);
638 if (!policy)
639 goto no_policy;
640
641 if (lock_policy_rwsem_write(policy->cpu) < 0)
642 goto fail;
643
644 if (fattr->store)
645 ret = fattr->store(policy, buf, count);
646 else
647 ret = -EIO;
648
649 unlock_policy_rwsem_write(policy->cpu);
650fail:
651 cpufreq_cpu_put(policy);
652no_policy:
653 return ret;
654}
655
656static void cpufreq_sysfs_release(struct kobject *kobj)
657{
658 struct cpufreq_policy *policy = to_policy(kobj);
659 pr_debug("last reference is dropped\n");
660 complete(&policy->kobj_unregister);
661}
662
663static const struct sysfs_ops sysfs_ops = {
664 .show = show,
665 .store = store,
666};
667
668static struct kobj_type ktype_cpufreq = {
669 .sysfs_ops = &sysfs_ops,
670 .default_attrs = default_attrs,
671 .release = cpufreq_sysfs_release,
672};
673
674/*
675 * Returns:
676 * Negative: Failure
677 * 0: Success
678 * Positive: When we have a managed CPU and the sysfs got symlinked
679 */
680static int cpufreq_add_dev_policy(unsigned int cpu,
681 struct cpufreq_policy *policy,
682 struct sys_device *sys_dev)
683{
684 int ret = 0;
685#ifdef CONFIG_SMP
686 unsigned long flags;
687 unsigned int j;
688#ifdef CONFIG_HOTPLUG_CPU
689 struct cpufreq_governor *gov;
690
691 gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
692 if (gov) {
693 policy->governor = gov;
694 pr_debug("Restoring governor %s for cpu %d\n",
695 policy->governor->name, cpu);
696 }
697#endif
698
699 for_each_cpu(j, policy->cpus) {
700 struct cpufreq_policy *managed_policy;
701
702 if (cpu == j)
703 continue;
704
705 /* Check for existing affected CPUs.
706 * They may not be aware of it due to CPU Hotplug.
707 * cpufreq_cpu_put is called when the device is removed
708 * in __cpufreq_remove_dev()
709 */
710 managed_policy = cpufreq_cpu_get(j);
711 if (unlikely(managed_policy)) {
712
713 /* Set proper policy_cpu */
714 unlock_policy_rwsem_write(cpu);
715 per_cpu(cpufreq_policy_cpu, cpu) = managed_policy->cpu;
716
717 if (lock_policy_rwsem_write(cpu) < 0) {
718 /* Should not go through policy unlock path */
719 if (cpufreq_driver->exit)
720 cpufreq_driver->exit(policy);
721 cpufreq_cpu_put(managed_policy);
722 return -EBUSY;
723 }
724
725 spin_lock_irqsave(&cpufreq_driver_lock, flags);
726 cpumask_copy(managed_policy->cpus, policy->cpus);
727 per_cpu(cpufreq_cpu_data, cpu) = managed_policy;
728 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
729
730 pr_debug("CPU already managed, adding link\n");
731 ret = sysfs_create_link(&sys_dev->kobj,
732 &managed_policy->kobj,
733 "cpufreq");
734 if (ret)
735 cpufreq_cpu_put(managed_policy);
736 /*
737 * Success. We only needed to be added to the mask.
738 * Call driver->exit() because only the cpu parent of
739 * the kobj needed to call init().
740 */
741 if (cpufreq_driver->exit)
742 cpufreq_driver->exit(policy);
743
744 if (!ret)
745 return 1;
746 else
747 return ret;
748 }
749 }
750#endif
751 return ret;
752}
753
754
755/* symlink affected CPUs */
756static int cpufreq_add_dev_symlink(unsigned int cpu,
757 struct cpufreq_policy *policy)
758{
759 unsigned int j;
760 int ret = 0;
761
762 for_each_cpu(j, policy->cpus) {
763 struct cpufreq_policy *managed_policy;
764 struct sys_device *cpu_sys_dev;
765
766 if (j == cpu)
767 continue;
768 if (!cpu_online(j))
769 continue;
770
771 pr_debug("CPU %u already managed, adding link\n", j);
772 managed_policy = cpufreq_cpu_get(cpu);
773 cpu_sys_dev = get_cpu_sysdev(j);
774 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
775 "cpufreq");
776 if (ret) {
777 cpufreq_cpu_put(managed_policy);
778 return ret;
779 }
780 }
781 return ret;
782}
783
784static int cpufreq_add_dev_interface(unsigned int cpu,
785 struct cpufreq_policy *policy,
786 struct sys_device *sys_dev)
787{
788 struct cpufreq_policy new_policy;
789 struct freq_attr **drv_attr;
790 unsigned long flags;
791 int ret = 0;
792 unsigned int j;
793
794 /* prepare interface data */
795 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
796 &sys_dev->kobj, "cpufreq");
797 if (ret)
798 return ret;
799
800 /* set up files for this cpu device */
801 drv_attr = cpufreq_driver->attr;
802 while ((drv_attr) && (*drv_attr)) {
803 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
804 if (ret)
805 goto err_out_kobj_put;
806 drv_attr++;
807 }
808 if (cpufreq_driver->get) {
809 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
810 if (ret)
811 goto err_out_kobj_put;
812 }
813 if (cpufreq_driver->target) {
814 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
815 if (ret)
816 goto err_out_kobj_put;
817 }
818 if (cpufreq_driver->bios_limit) {
819 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
820 if (ret)
821 goto err_out_kobj_put;
822 }
823
824 spin_lock_irqsave(&cpufreq_driver_lock, flags);
825 for_each_cpu(j, policy->cpus) {
826 if (!cpu_online(j))
827 continue;
828 per_cpu(cpufreq_cpu_data, j) = policy;
829 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
830 }
831 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
832
833 ret = cpufreq_add_dev_symlink(cpu, policy);
834 if (ret)
835 goto err_out_kobj_put;
836
837 memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
838 /* assure that the starting sequence is run in __cpufreq_set_policy */
839 policy->governor = NULL;
840
841 /* set default policy */
842 ret = __cpufreq_set_policy(policy, &new_policy);
843 policy->user_policy.policy = policy->policy;
844 policy->user_policy.governor = policy->governor;
845
846 if (ret) {
847 pr_debug("setting policy failed\n");
848 if (cpufreq_driver->exit)
849 cpufreq_driver->exit(policy);
850 }
851 return ret;
852
853err_out_kobj_put:
854 kobject_put(&policy->kobj);
855 wait_for_completion(&policy->kobj_unregister);
856 return ret;
857}
858
859
860/**
861 * cpufreq_add_dev - add a CPU device
862 *
863 * Adds the cpufreq interface for a CPU device.
864 *
865 * The Oracle says: try running cpufreq registration/unregistration concurrently
866 * with with cpu hotplugging and all hell will break loose. Tried to clean this
867 * mess up, but more thorough testing is needed. - Mathieu
868 */
869static int cpufreq_add_dev(struct sys_device *sys_dev)
870{
871 unsigned int cpu = sys_dev->id;
872 int ret = 0, found = 0;
873 struct cpufreq_policy *policy;
874 unsigned long flags;
875 unsigned int j;
876#ifdef CONFIG_HOTPLUG_CPU
877 int sibling;
878#endif
879
880 if (cpu_is_offline(cpu))
881 return 0;
882
883 pr_debug("adding CPU %u\n", cpu);
884
885#ifdef CONFIG_SMP
886 /* check whether a different CPU already registered this
887 * CPU because it is in the same boat. */
888 policy = cpufreq_cpu_get(cpu);
889 if (unlikely(policy)) {
890 cpufreq_cpu_put(policy);
891 return 0;
892 }
893#endif
894
895 if (!try_module_get(cpufreq_driver->owner)) {
896 ret = -EINVAL;
897 goto module_out;
898 }
899
900 ret = -ENOMEM;
901 policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
902 if (!policy)
903 goto nomem_out;
904
905 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
906 goto err_free_policy;
907
908 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
909 goto err_free_cpumask;
910
911 policy->cpu = cpu;
912 cpumask_copy(policy->cpus, cpumask_of(cpu));
913
914 /* Initially set CPU itself as the policy_cpu */
915 per_cpu(cpufreq_policy_cpu, cpu) = cpu;
916 ret = (lock_policy_rwsem_write(cpu) < 0);
917 WARN_ON(ret);
918
919 init_completion(&policy->kobj_unregister);
920 INIT_WORK(&policy->update, handle_update);
921
922 /* Set governor before ->init, so that driver could check it */
923#ifdef CONFIG_HOTPLUG_CPU
924 for_each_online_cpu(sibling) {
925 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
926 if (cp && cp->governor &&
927 (cpumask_test_cpu(cpu, cp->related_cpus))) {
928 policy->governor = cp->governor;
929 found = 1;
930 break;
931 }
932 }
933#endif
934 if (!found)
935 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
936 /* call driver. From then on the cpufreq must be able
937 * to accept all calls to ->verify and ->setpolicy for this CPU
938 */
939 ret = cpufreq_driver->init(policy);
940 if (ret) {
941 pr_debug("initialization failed\n");
942 goto err_unlock_policy;
943 }
944 policy->user_policy.min = policy->min;
945 policy->user_policy.max = policy->max;
946
947 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
948 CPUFREQ_START, policy);
949
950 ret = cpufreq_add_dev_policy(cpu, policy, sys_dev);
951 if (ret) {
952 if (ret > 0)
953 /* This is a managed cpu, symlink created,
954 exit with 0 */
955 ret = 0;
956 goto err_unlock_policy;
957 }
958
959 ret = cpufreq_add_dev_interface(cpu, policy, sys_dev);
960 if (ret)
961 goto err_out_unregister;
962
963 unlock_policy_rwsem_write(cpu);
964
965 kobject_uevent(&policy->kobj, KOBJ_ADD);
966 module_put(cpufreq_driver->owner);
967 pr_debug("initialization complete\n");
968
969 return 0;
970
971
972err_out_unregister:
973 spin_lock_irqsave(&cpufreq_driver_lock, flags);
974 for_each_cpu(j, policy->cpus)
975 per_cpu(cpufreq_cpu_data, j) = NULL;
976 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
977
978 kobject_put(&policy->kobj);
979 wait_for_completion(&policy->kobj_unregister);
980
981err_unlock_policy:
982 unlock_policy_rwsem_write(cpu);
983 free_cpumask_var(policy->related_cpus);
984err_free_cpumask:
985 free_cpumask_var(policy->cpus);
986err_free_policy:
987 kfree(policy);
988nomem_out:
989 module_put(cpufreq_driver->owner);
990module_out:
991 return ret;
992}
993
994
995/**
996 * __cpufreq_remove_dev - remove a CPU device
997 *
998 * Removes the cpufreq interface for a CPU device.
999 * Caller should already have policy_rwsem in write mode for this CPU.
1000 * This routine frees the rwsem before returning.
1001 */
1002static int __cpufreq_remove_dev(struct sys_device *sys_dev)
1003{
1004 unsigned int cpu = sys_dev->id;
1005 unsigned long flags;
1006 struct cpufreq_policy *data;
1007 struct kobject *kobj;
1008 struct completion *cmp;
1009#ifdef CONFIG_SMP
1010 struct sys_device *cpu_sys_dev;
1011 unsigned int j;
1012#endif
1013
1014 pr_debug("unregistering CPU %u\n", cpu);
1015
1016 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1017 data = per_cpu(cpufreq_cpu_data, cpu);
1018
1019 if (!data) {
1020 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1021 unlock_policy_rwsem_write(cpu);
1022 return -EINVAL;
1023 }
1024 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1025
1026
1027#ifdef CONFIG_SMP
1028 /* if this isn't the CPU which is the parent of the kobj, we
1029 * only need to unlink, put and exit
1030 */
1031 if (unlikely(cpu != data->cpu)) {
1032 pr_debug("removing link\n");
1033 cpumask_clear_cpu(cpu, data->cpus);
1034 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1035 kobj = &sys_dev->kobj;
1036 cpufreq_cpu_put(data);
1037 unlock_policy_rwsem_write(cpu);
1038 sysfs_remove_link(kobj, "cpufreq");
1039 return 0;
1040 }
1041#endif
1042
1043#ifdef CONFIG_SMP
1044
1045#ifdef CONFIG_HOTPLUG_CPU
1046 strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name,
1047 CPUFREQ_NAME_LEN);
1048#endif
1049
1050 /* if we have other CPUs still registered, we need to unlink them,
1051 * or else wait_for_completion below will lock up. Clean the
1052 * per_cpu(cpufreq_cpu_data) while holding the lock, and remove
1053 * the sysfs links afterwards.
1054 */
1055 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1056 for_each_cpu(j, data->cpus) {
1057 if (j == cpu)
1058 continue;
1059 per_cpu(cpufreq_cpu_data, j) = NULL;
1060 }
1061 }
1062
1063 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1064
1065 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1066 for_each_cpu(j, data->cpus) {
1067 if (j == cpu)
1068 continue;
1069 pr_debug("removing link for cpu %u\n", j);
1070#ifdef CONFIG_HOTPLUG_CPU
1071 strncpy(per_cpu(cpufreq_cpu_governor, j),
1072 data->governor->name, CPUFREQ_NAME_LEN);
1073#endif
1074 cpu_sys_dev = get_cpu_sysdev(j);
1075 kobj = &cpu_sys_dev->kobj;
1076 unlock_policy_rwsem_write(cpu);
1077 sysfs_remove_link(kobj, "cpufreq");
1078 lock_policy_rwsem_write(cpu);
1079 cpufreq_cpu_put(data);
1080 }
1081 }
1082#else
1083 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1084#endif
1085
1086 if (cpufreq_driver->target)
1087 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1088
1089 kobj = &data->kobj;
1090 cmp = &data->kobj_unregister;
1091 unlock_policy_rwsem_write(cpu);
1092 kobject_put(kobj);
1093
1094 /* we need to make sure that the underlying kobj is actually
1095 * not referenced anymore by anybody before we proceed with
1096 * unloading.
1097 */
1098 pr_debug("waiting for dropping of refcount\n");
1099 wait_for_completion(cmp);
1100 pr_debug("wait complete\n");
1101
1102 lock_policy_rwsem_write(cpu);
1103 if (cpufreq_driver->exit)
1104 cpufreq_driver->exit(data);
1105 unlock_policy_rwsem_write(cpu);
1106
1107#ifdef CONFIG_HOTPLUG_CPU
1108 /* when the CPU which is the parent of the kobj is hotplugged
1109 * offline, check for siblings, and create cpufreq sysfs interface
1110 * and symlinks
1111 */
1112 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1113 /* first sibling now owns the new sysfs dir */
1114 cpumask_clear_cpu(cpu, data->cpus);
1115 cpufreq_add_dev(get_cpu_sysdev(cpumask_first(data->cpus)));
1116
1117 /* finally remove our own symlink */
1118 lock_policy_rwsem_write(cpu);
1119 __cpufreq_remove_dev(sys_dev);
1120 }
1121#endif
1122
1123 free_cpumask_var(data->related_cpus);
1124 free_cpumask_var(data->cpus);
1125 kfree(data);
1126
1127 return 0;
1128}
1129
1130
1131static int cpufreq_remove_dev(struct sys_device *sys_dev)
1132{
1133 unsigned int cpu = sys_dev->id;
1134 int retval;
1135
1136 if (cpu_is_offline(cpu))
1137 return 0;
1138
1139 if (unlikely(lock_policy_rwsem_write(cpu)))
1140 BUG();
1141
1142 retval = __cpufreq_remove_dev(sys_dev);
1143 return retval;
1144}
1145
1146
1147static void handle_update(struct work_struct *work)
1148{
1149 struct cpufreq_policy *policy =
1150 container_of(work, struct cpufreq_policy, update);
1151 unsigned int cpu = policy->cpu;
1152 pr_debug("handle_update for cpu %u called\n", cpu);
1153 cpufreq_update_policy(cpu);
1154}
1155
1156/**
1157 * cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1158 * @cpu: cpu number
1159 * @old_freq: CPU frequency the kernel thinks the CPU runs at
1160 * @new_freq: CPU frequency the CPU actually runs at
1161 *
1162 * We adjust to current frequency first, and need to clean up later.
1163 * So either call to cpufreq_update_policy() or schedule handle_update()).
1164 */
1165static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1166 unsigned int new_freq)
1167{
1168 struct cpufreq_freqs freqs;
1169
1170 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1171 "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1172
1173 freqs.cpu = cpu;
1174 freqs.old = old_freq;
1175 freqs.new = new_freq;
1176 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1177 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1178}
1179
1180
1181/**
1182 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1183 * @cpu: CPU number
1184 *
1185 * This is the last known freq, without actually getting it from the driver.
1186 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1187 */
1188unsigned int cpufreq_quick_get(unsigned int cpu)
1189{
1190 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1191 unsigned int ret_freq = 0;
1192
1193 if (policy) {
1194 ret_freq = policy->cur;
1195 cpufreq_cpu_put(policy);
1196 }
1197
1198 return ret_freq;
1199}
1200EXPORT_SYMBOL(cpufreq_quick_get);
1201
1202/**
1203 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1204 * @cpu: CPU number
1205 *
1206 * Just return the max possible frequency for a given CPU.
1207 */
1208unsigned int cpufreq_quick_get_max(unsigned int cpu)
1209{
1210 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1211 unsigned int ret_freq = 0;
1212
1213 if (policy) {
1214 ret_freq = policy->max;
1215 cpufreq_cpu_put(policy);
1216 }
1217
1218 return ret_freq;
1219}
1220EXPORT_SYMBOL(cpufreq_quick_get_max);
1221
1222
1223static unsigned int __cpufreq_get(unsigned int cpu)
1224{
1225 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1226 unsigned int ret_freq = 0;
1227
1228 if (!cpufreq_driver->get)
1229 return ret_freq;
1230
1231 ret_freq = cpufreq_driver->get(cpu);
1232
1233 if (ret_freq && policy->cur &&
1234 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1235 /* verify no discrepancy between actual and
1236 saved value exists */
1237 if (unlikely(ret_freq != policy->cur)) {
1238 cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1239 schedule_work(&policy->update);
1240 }
1241 }
1242
1243 return ret_freq;
1244}
1245
1246/**
1247 * cpufreq_get - get the current CPU frequency (in kHz)
1248 * @cpu: CPU number
1249 *
1250 * Get the CPU current (static) CPU frequency
1251 */
1252unsigned int cpufreq_get(unsigned int cpu)
1253{
1254 unsigned int ret_freq = 0;
1255 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1256
1257 if (!policy)
1258 goto out;
1259
1260 if (unlikely(lock_policy_rwsem_read(cpu)))
1261 goto out_policy;
1262
1263 ret_freq = __cpufreq_get(cpu);
1264
1265 unlock_policy_rwsem_read(cpu);
1266
1267out_policy:
1268 cpufreq_cpu_put(policy);
1269out:
1270 return ret_freq;
1271}
1272EXPORT_SYMBOL(cpufreq_get);
1273
1274static struct sysdev_driver cpufreq_sysdev_driver = {
1275 .add = cpufreq_add_dev,
1276 .remove = cpufreq_remove_dev,
1277};
1278
1279
1280/**
1281 * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1282 *
1283 * This function is only executed for the boot processor. The other CPUs
1284 * have been put offline by means of CPU hotplug.
1285 */
1286static int cpufreq_bp_suspend(void)
1287{
1288 int ret = 0;
1289
1290 int cpu = smp_processor_id();
1291 struct cpufreq_policy *cpu_policy;
1292
1293 pr_debug("suspending cpu %u\n", cpu);
1294
1295 /* If there's no policy for the boot CPU, we have nothing to do. */
1296 cpu_policy = cpufreq_cpu_get(cpu);
1297 if (!cpu_policy)
1298 return 0;
1299
1300 if (cpufreq_driver->suspend) {
1301 ret = cpufreq_driver->suspend(cpu_policy);
1302 if (ret)
1303 printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1304 "step on CPU %u\n", cpu_policy->cpu);
1305 }
1306
1307 cpufreq_cpu_put(cpu_policy);
1308 return ret;
1309}
1310
1311/**
1312 * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1313 *
1314 * 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1315 * 2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1316 * restored. It will verify that the current freq is in sync with
1317 * what we believe it to be. This is a bit later than when it
1318 * should be, but nonethteless it's better than calling
1319 * cpufreq_driver->get() here which might re-enable interrupts...
1320 *
1321 * This function is only executed for the boot CPU. The other CPUs have not
1322 * been turned on yet.
1323 */
1324static void cpufreq_bp_resume(void)
1325{
1326 int ret = 0;
1327
1328 int cpu = smp_processor_id();
1329 struct cpufreq_policy *cpu_policy;
1330
1331 pr_debug("resuming cpu %u\n", cpu);
1332
1333 /* If there's no policy for the boot CPU, we have nothing to do. */
1334 cpu_policy = cpufreq_cpu_get(cpu);
1335 if (!cpu_policy)
1336 return;
1337
1338 if (cpufreq_driver->resume) {
1339 ret = cpufreq_driver->resume(cpu_policy);
1340 if (ret) {
1341 printk(KERN_ERR "cpufreq: resume failed in ->resume "
1342 "step on CPU %u\n", cpu_policy->cpu);
1343 goto fail;
1344 }
1345 }
1346
1347 schedule_work(&cpu_policy->update);
1348
1349fail:
1350 cpufreq_cpu_put(cpu_policy);
1351}
1352
1353static struct syscore_ops cpufreq_syscore_ops = {
1354 .suspend = cpufreq_bp_suspend,
1355 .resume = cpufreq_bp_resume,
1356};
1357
1358
1359/*********************************************************************
1360 * NOTIFIER LISTS INTERFACE *
1361 *********************************************************************/
1362
1363/**
1364 * cpufreq_register_notifier - register a driver with cpufreq
1365 * @nb: notifier function to register
1366 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1367 *
1368 * Add a driver to one of two lists: either a list of drivers that
1369 * are notified about clock rate changes (once before and once after
1370 * the transition), or a list of drivers that are notified about
1371 * changes in cpufreq policy.
1372 *
1373 * This function may sleep, and has the same return conditions as
1374 * blocking_notifier_chain_register.
1375 */
1376int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1377{
1378 int ret;
1379
1380 WARN_ON(!init_cpufreq_transition_notifier_list_called);
1381
1382 switch (list) {
1383 case CPUFREQ_TRANSITION_NOTIFIER:
1384 ret = srcu_notifier_chain_register(
1385 &cpufreq_transition_notifier_list, nb);
1386 break;
1387 case CPUFREQ_POLICY_NOTIFIER:
1388 ret = blocking_notifier_chain_register(
1389 &cpufreq_policy_notifier_list, nb);
1390 break;
1391 default:
1392 ret = -EINVAL;
1393 }
1394
1395 return ret;
1396}
1397EXPORT_SYMBOL(cpufreq_register_notifier);
1398
1399
1400/**
1401 * cpufreq_unregister_notifier - unregister a driver with cpufreq
1402 * @nb: notifier block to be unregistered
1403 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1404 *
1405 * Remove a driver from the CPU frequency notifier list.
1406 *
1407 * This function may sleep, and has the same return conditions as
1408 * blocking_notifier_chain_unregister.
1409 */
1410int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1411{
1412 int ret;
1413
1414 switch (list) {
1415 case CPUFREQ_TRANSITION_NOTIFIER:
1416 ret = srcu_notifier_chain_unregister(
1417 &cpufreq_transition_notifier_list, nb);
1418 break;
1419 case CPUFREQ_POLICY_NOTIFIER:
1420 ret = blocking_notifier_chain_unregister(
1421 &cpufreq_policy_notifier_list, nb);
1422 break;
1423 default:
1424 ret = -EINVAL;
1425 }
1426
1427 return ret;
1428}
1429EXPORT_SYMBOL(cpufreq_unregister_notifier);
1430
1431
1432/*********************************************************************
1433 * GOVERNORS *
1434 *********************************************************************/
1435
1436
1437int __cpufreq_driver_target(struct cpufreq_policy *policy,
1438 unsigned int target_freq,
1439 unsigned int relation)
1440{
1441 int retval = -EINVAL;
1442
1443 pr_debug("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1444 target_freq, relation);
1445 if (cpu_online(policy->cpu) && cpufreq_driver->target)
1446 retval = cpufreq_driver->target(policy, target_freq, relation);
1447
1448 return retval;
1449}
1450EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1451
1452int cpufreq_driver_target(struct cpufreq_policy *policy,
1453 unsigned int target_freq,
1454 unsigned int relation)
1455{
1456 int ret = -EINVAL;
1457
1458 policy = cpufreq_cpu_get(policy->cpu);
1459 if (!policy)
1460 goto no_policy;
1461
1462 if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1463 goto fail;
1464
1465 ret = __cpufreq_driver_target(policy, target_freq, relation);
1466
1467 unlock_policy_rwsem_write(policy->cpu);
1468
1469fail:
1470 cpufreq_cpu_put(policy);
1471no_policy:
1472 return ret;
1473}
1474EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1475
1476int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
1477{
1478 int ret = 0;
1479
1480 policy = cpufreq_cpu_get(policy->cpu);
1481 if (!policy)
1482 return -EINVAL;
1483
1484 if (cpu_online(cpu) && cpufreq_driver->getavg)
1485 ret = cpufreq_driver->getavg(policy, cpu);
1486
1487 cpufreq_cpu_put(policy);
1488 return ret;
1489}
1490EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1491
1492/*
1493 * when "event" is CPUFREQ_GOV_LIMITS
1494 */
1495
1496static int __cpufreq_governor(struct cpufreq_policy *policy,
1497 unsigned int event)
1498{
1499 int ret;
1500
1501 /* Only must be defined when default governor is known to have latency
1502 restrictions, like e.g. conservative or ondemand.
1503 That this is the case is already ensured in Kconfig
1504 */
1505#ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1506 struct cpufreq_governor *gov = &cpufreq_gov_performance;
1507#else
1508 struct cpufreq_governor *gov = NULL;
1509#endif
1510
1511 if (policy->governor->max_transition_latency &&
1512 policy->cpuinfo.transition_latency >
1513 policy->governor->max_transition_latency) {
1514 if (!gov)
1515 return -EINVAL;
1516 else {
1517 printk(KERN_WARNING "%s governor failed, too long"
1518 " transition latency of HW, fallback"
1519 " to %s governor\n",
1520 policy->governor->name,
1521 gov->name);
1522 policy->governor = gov;
1523 }
1524 }
1525
1526 if (!try_module_get(policy->governor->owner))
1527 return -EINVAL;
1528
1529 pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1530 policy->cpu, event);
1531 ret = policy->governor->governor(policy, event);
1532
1533 /* we keep one module reference alive for
1534 each CPU governed by this CPU */
1535 if ((event != CPUFREQ_GOV_START) || ret)
1536 module_put(policy->governor->owner);
1537 if ((event == CPUFREQ_GOV_STOP) && !ret)
1538 module_put(policy->governor->owner);
1539
1540 return ret;
1541}
1542
1543
1544int cpufreq_register_governor(struct cpufreq_governor *governor)
1545{
1546 int err;
1547
1548 if (!governor)
1549 return -EINVAL;
1550
1551 mutex_lock(&cpufreq_governor_mutex);
1552
1553 err = -EBUSY;
1554 if (__find_governor(governor->name) == NULL) {
1555 err = 0;
1556 list_add(&governor->governor_list, &cpufreq_governor_list);
1557 }
1558
1559 mutex_unlock(&cpufreq_governor_mutex);
1560 return err;
1561}
1562EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1563
1564
1565void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1566{
1567#ifdef CONFIG_HOTPLUG_CPU
1568 int cpu;
1569#endif
1570
1571 if (!governor)
1572 return;
1573
1574#ifdef CONFIG_HOTPLUG_CPU
1575 for_each_present_cpu(cpu) {
1576 if (cpu_online(cpu))
1577 continue;
1578 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1579 strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1580 }
1581#endif
1582
1583 mutex_lock(&cpufreq_governor_mutex);
1584 list_del(&governor->governor_list);
1585 mutex_unlock(&cpufreq_governor_mutex);
1586 return;
1587}
1588EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1589
1590
1591
1592/*********************************************************************
1593 * POLICY INTERFACE *
1594 *********************************************************************/
1595
1596/**
1597 * cpufreq_get_policy - get the current cpufreq_policy
1598 * @policy: struct cpufreq_policy into which the current cpufreq_policy
1599 * is written
1600 *
1601 * Reads the current cpufreq policy.
1602 */
1603int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1604{
1605 struct cpufreq_policy *cpu_policy;
1606 if (!policy)
1607 return -EINVAL;
1608
1609 cpu_policy = cpufreq_cpu_get(cpu);
1610 if (!cpu_policy)
1611 return -EINVAL;
1612
1613 memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1614
1615 cpufreq_cpu_put(cpu_policy);
1616 return 0;
1617}
1618EXPORT_SYMBOL(cpufreq_get_policy);
1619
1620
1621/*
1622 * data : current policy.
1623 * policy : policy to be set.
1624 */
1625static int __cpufreq_set_policy(struct cpufreq_policy *data,
1626 struct cpufreq_policy *policy)
1627{
1628 int ret = 0;
1629
1630 pr_debug("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1631 policy->min, policy->max);
1632
1633 memcpy(&policy->cpuinfo, &data->cpuinfo,
1634 sizeof(struct cpufreq_cpuinfo));
1635
1636 if (policy->min > data->max || policy->max < data->min) {
1637 ret = -EINVAL;
1638 goto error_out;
1639 }
1640
1641 /* verify the cpu speed can be set within this limit */
1642 ret = cpufreq_driver->verify(policy);
1643 if (ret)
1644 goto error_out;
1645
1646 /* adjust if necessary - all reasons */
1647 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1648 CPUFREQ_ADJUST, policy);
1649
1650 /* adjust if necessary - hardware incompatibility*/
1651 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1652 CPUFREQ_INCOMPATIBLE, policy);
1653
1654 /* verify the cpu speed can be set within this limit,
1655 which might be different to the first one */
1656 ret = cpufreq_driver->verify(policy);
1657 if (ret)
1658 goto error_out;
1659
1660 /* notification of the new policy */
1661 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1662 CPUFREQ_NOTIFY, policy);
1663
1664 data->min = policy->min;
1665 data->max = policy->max;
1666
1667 pr_debug("new min and max freqs are %u - %u kHz\n",
1668 data->min, data->max);
1669
1670 if (cpufreq_driver->setpolicy) {
1671 data->policy = policy->policy;
1672 pr_debug("setting range\n");
1673 ret = cpufreq_driver->setpolicy(policy);
1674 } else {
1675 if (policy->governor != data->governor) {
1676 /* save old, working values */
1677 struct cpufreq_governor *old_gov = data->governor;
1678
1679 pr_debug("governor switch\n");
1680
1681 /* end old governor */
1682 if (data->governor)
1683 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1684
1685 /* start new governor */
1686 data->governor = policy->governor;
1687 if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1688 /* new governor failed, so re-start old one */
1689 pr_debug("starting governor %s failed\n",
1690 data->governor->name);
1691 if (old_gov) {
1692 data->governor = old_gov;
1693 __cpufreq_governor(data,
1694 CPUFREQ_GOV_START);
1695 }
1696 ret = -EINVAL;
1697 goto error_out;
1698 }
1699 /* might be a policy change, too, so fall through */
1700 }
1701 pr_debug("governor: change or update limits\n");
1702 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1703 }
1704
1705error_out:
1706 return ret;
1707}
1708
1709/**
1710 * cpufreq_update_policy - re-evaluate an existing cpufreq policy
1711 * @cpu: CPU which shall be re-evaluated
1712 *
1713 * Useful for policy notifiers which have different necessities
1714 * at different times.
1715 */
1716int cpufreq_update_policy(unsigned int cpu)
1717{
1718 struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1719 struct cpufreq_policy policy;
1720 int ret;
1721
1722 if (!data) {
1723 ret = -ENODEV;
1724 goto no_policy;
1725 }
1726
1727 if (unlikely(lock_policy_rwsem_write(cpu))) {
1728 ret = -EINVAL;
1729 goto fail;
1730 }
1731
1732 pr_debug("updating policy for CPU %u\n", cpu);
1733 memcpy(&policy, data, sizeof(struct cpufreq_policy));
1734 policy.min = data->user_policy.min;
1735 policy.max = data->user_policy.max;
1736 policy.policy = data->user_policy.policy;
1737 policy.governor = data->user_policy.governor;
1738
1739 /* BIOS might change freq behind our back
1740 -> ask driver for current freq and notify governors about a change */
1741 if (cpufreq_driver->get) {
1742 policy.cur = cpufreq_driver->get(cpu);
1743 if (!data->cur) {
1744 pr_debug("Driver did not initialize current freq");
1745 data->cur = policy.cur;
1746 } else {
1747 if (data->cur != policy.cur)
1748 cpufreq_out_of_sync(cpu, data->cur,
1749 policy.cur);
1750 }
1751 }
1752
1753 ret = __cpufreq_set_policy(data, &policy);
1754
1755 unlock_policy_rwsem_write(cpu);
1756
1757fail:
1758 cpufreq_cpu_put(data);
1759no_policy:
1760 return ret;
1761}
1762EXPORT_SYMBOL(cpufreq_update_policy);
1763
1764static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1765 unsigned long action, void *hcpu)
1766{
1767 unsigned int cpu = (unsigned long)hcpu;
1768 struct sys_device *sys_dev;
1769
1770 sys_dev = get_cpu_sysdev(cpu);
1771 if (sys_dev) {
1772 switch (action) {
1773 case CPU_ONLINE:
1774 case CPU_ONLINE_FROZEN:
1775 cpufreq_add_dev(sys_dev);
1776 break;
1777 case CPU_DOWN_PREPARE:
1778 case CPU_DOWN_PREPARE_FROZEN:
1779 if (unlikely(lock_policy_rwsem_write(cpu)))
1780 BUG();
1781
1782 __cpufreq_remove_dev(sys_dev);
1783 break;
1784 case CPU_DOWN_FAILED:
1785 case CPU_DOWN_FAILED_FROZEN:
1786 cpufreq_add_dev(sys_dev);
1787 break;
1788 }
1789 }
1790 return NOTIFY_OK;
1791}
1792
1793static struct notifier_block __refdata cpufreq_cpu_notifier = {
1794 .notifier_call = cpufreq_cpu_callback,
1795};
1796
1797/*********************************************************************
1798 * REGISTER / UNREGISTER CPUFREQ DRIVER *
1799 *********************************************************************/
1800
1801/**
1802 * cpufreq_register_driver - register a CPU Frequency driver
1803 * @driver_data: A struct cpufreq_driver containing the values#
1804 * submitted by the CPU Frequency driver.
1805 *
1806 * Registers a CPU Frequency driver to this core code. This code
1807 * returns zero on success, -EBUSY when another driver got here first
1808 * (and isn't unregistered in the meantime).
1809 *
1810 */
1811int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1812{
1813 unsigned long flags;
1814 int ret;
1815
1816 if (!driver_data || !driver_data->verify || !driver_data->init ||
1817 ((!driver_data->setpolicy) && (!driver_data->target)))
1818 return -EINVAL;
1819
1820 pr_debug("trying to register driver %s\n", driver_data->name);
1821
1822 if (driver_data->setpolicy)
1823 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1824
1825 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1826 if (cpufreq_driver) {
1827 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1828 return -EBUSY;
1829 }
1830 cpufreq_driver = driver_data;
1831 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1832
1833 ret = sysdev_driver_register(&cpu_sysdev_class,
1834 &cpufreq_sysdev_driver);
1835 if (ret)
1836 goto err_null_driver;
1837
1838 if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1839 int i;
1840 ret = -ENODEV;
1841
1842 /* check for at least one working CPU */
1843 for (i = 0; i < nr_cpu_ids; i++)
1844 if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
1845 ret = 0;
1846 break;
1847 }
1848
1849 /* if all ->init() calls failed, unregister */
1850 if (ret) {
1851 pr_debug("no CPU initialized for driver %s\n",
1852 driver_data->name);
1853 goto err_sysdev_unreg;
1854 }
1855 }
1856
1857 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1858 pr_debug("driver %s up and running\n", driver_data->name);
1859
1860 return 0;
1861err_sysdev_unreg:
1862 sysdev_driver_unregister(&cpu_sysdev_class,
1863 &cpufreq_sysdev_driver);
1864err_null_driver:
1865 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1866 cpufreq_driver = NULL;
1867 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1868 return ret;
1869}
1870EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1871
1872
1873/**
1874 * cpufreq_unregister_driver - unregister the current CPUFreq driver
1875 *
1876 * Unregister the current CPUFreq driver. Only call this if you have
1877 * the right to do so, i.e. if you have succeeded in initialising before!
1878 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1879 * currently not initialised.
1880 */
1881int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1882{
1883 unsigned long flags;
1884
1885 if (!cpufreq_driver || (driver != cpufreq_driver))
1886 return -EINVAL;
1887
1888 pr_debug("unregistering driver %s\n", driver->name);
1889
1890 sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1891 unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1892
1893 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1894 cpufreq_driver = NULL;
1895 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1896
1897 return 0;
1898}
1899EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1900
1901static int __init cpufreq_core_init(void)
1902{
1903 int cpu;
1904
1905 for_each_possible_cpu(cpu) {
1906 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1907 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1908 }
1909
1910 cpufreq_global_kobject = kobject_create_and_add("cpufreq",
1911 &cpu_sysdev_class.kset.kobj);
1912 BUG_ON(!cpufreq_global_kobject);
1913 register_syscore_ops(&cpufreq_syscore_ops);
1914
1915 return 0;
1916}
1917core_initcall(cpufreq_core_init);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/drivers/cpufreq/cpufreq.c
4 *
5 * Copyright (C) 2001 Russell King
6 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
7 * (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
8 *
9 * Oct 2005 - Ashok Raj <ashok.raj@intel.com>
10 * Added handling for CPU hotplug
11 * Feb 2006 - Jacob Shin <jacob.shin@amd.com>
12 * Fix handling for CPU hotplug -- affected CPUs
13 */
14
15#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16
17#include <linux/cpu.h>
18#include <linux/cpufreq.h>
19#include <linux/cpu_cooling.h>
20#include <linux/delay.h>
21#include <linux/device.h>
22#include <linux/init.h>
23#include <linux/kernel_stat.h>
24#include <linux/module.h>
25#include <linux/mutex.h>
26#include <linux/pm_qos.h>
27#include <linux/slab.h>
28#include <linux/suspend.h>
29#include <linux/syscore_ops.h>
30#include <linux/tick.h>
31#include <trace/events/power.h>
32
33static LIST_HEAD(cpufreq_policy_list);
34
35/* Macros to iterate over CPU policies */
36#define for_each_suitable_policy(__policy, __active) \
37 list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
38 if ((__active) == !policy_is_inactive(__policy))
39
40#define for_each_active_policy(__policy) \
41 for_each_suitable_policy(__policy, true)
42#define for_each_inactive_policy(__policy) \
43 for_each_suitable_policy(__policy, false)
44
45#define for_each_policy(__policy) \
46 list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
47
48/* Iterate over governors */
49static LIST_HEAD(cpufreq_governor_list);
50#define for_each_governor(__governor) \
51 list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
52
53/**
54 * The "cpufreq driver" - the arch- or hardware-dependent low
55 * level driver of CPUFreq support, and its spinlock. This lock
56 * also protects the cpufreq_cpu_data array.
57 */
58static struct cpufreq_driver *cpufreq_driver;
59static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
60static DEFINE_RWLOCK(cpufreq_driver_lock);
61
62/* Flag to suspend/resume CPUFreq governors */
63static bool cpufreq_suspended;
64
65static inline bool has_target(void)
66{
67 return cpufreq_driver->target_index || cpufreq_driver->target;
68}
69
70/* internal prototypes */
71static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
72static int cpufreq_init_governor(struct cpufreq_policy *policy);
73static void cpufreq_exit_governor(struct cpufreq_policy *policy);
74static int cpufreq_start_governor(struct cpufreq_policy *policy);
75static void cpufreq_stop_governor(struct cpufreq_policy *policy);
76static void cpufreq_governor_limits(struct cpufreq_policy *policy);
77
78/**
79 * Two notifier lists: the "policy" list is involved in the
80 * validation process for a new CPU frequency policy; the
81 * "transition" list for kernel code that needs to handle
82 * changes to devices when the CPU clock speed changes.
83 * The mutex locks both lists.
84 */
85static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
86SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
87
88static int off __read_mostly;
89static int cpufreq_disabled(void)
90{
91 return off;
92}
93void disable_cpufreq(void)
94{
95 off = 1;
96}
97static DEFINE_MUTEX(cpufreq_governor_mutex);
98
99bool have_governor_per_policy(void)
100{
101 return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
102}
103EXPORT_SYMBOL_GPL(have_governor_per_policy);
104
105struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
106{
107 if (have_governor_per_policy())
108 return &policy->kobj;
109 else
110 return cpufreq_global_kobject;
111}
112EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
113
114static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
115{
116 u64 idle_time;
117 u64 cur_wall_time;
118 u64 busy_time;
119
120 cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
121
122 busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
123 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
124 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
125 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
126 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
127 busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
128
129 idle_time = cur_wall_time - busy_time;
130 if (wall)
131 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
132
133 return div_u64(idle_time, NSEC_PER_USEC);
134}
135
136u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
137{
138 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
139
140 if (idle_time == -1ULL)
141 return get_cpu_idle_time_jiffy(cpu, wall);
142 else if (!io_busy)
143 idle_time += get_cpu_iowait_time_us(cpu, wall);
144
145 return idle_time;
146}
147EXPORT_SYMBOL_GPL(get_cpu_idle_time);
148
149__weak void arch_set_freq_scale(struct cpumask *cpus, unsigned long cur_freq,
150 unsigned long max_freq)
151{
152}
153EXPORT_SYMBOL_GPL(arch_set_freq_scale);
154
155/*
156 * This is a generic cpufreq init() routine which can be used by cpufreq
157 * drivers of SMP systems. It will do following:
158 * - validate & show freq table passed
159 * - set policies transition latency
160 * - policy->cpus with all possible CPUs
161 */
162void cpufreq_generic_init(struct cpufreq_policy *policy,
163 struct cpufreq_frequency_table *table,
164 unsigned int transition_latency)
165{
166 policy->freq_table = table;
167 policy->cpuinfo.transition_latency = transition_latency;
168
169 /*
170 * The driver only supports the SMP configuration where all processors
171 * share the clock and voltage and clock.
172 */
173 cpumask_setall(policy->cpus);
174}
175EXPORT_SYMBOL_GPL(cpufreq_generic_init);
176
177struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
178{
179 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
180
181 return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
182}
183EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
184
185unsigned int cpufreq_generic_get(unsigned int cpu)
186{
187 struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
188
189 if (!policy || IS_ERR(policy->clk)) {
190 pr_err("%s: No %s associated to cpu: %d\n",
191 __func__, policy ? "clk" : "policy", cpu);
192 return 0;
193 }
194
195 return clk_get_rate(policy->clk) / 1000;
196}
197EXPORT_SYMBOL_GPL(cpufreq_generic_get);
198
199/**
200 * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
201 * @cpu: CPU to find the policy for.
202 *
203 * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
204 * the kobject reference counter of that policy. Return a valid policy on
205 * success or NULL on failure.
206 *
207 * The policy returned by this function has to be released with the help of
208 * cpufreq_cpu_put() to balance its kobject reference counter properly.
209 */
210struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
211{
212 struct cpufreq_policy *policy = NULL;
213 unsigned long flags;
214
215 if (WARN_ON(cpu >= nr_cpu_ids))
216 return NULL;
217
218 /* get the cpufreq driver */
219 read_lock_irqsave(&cpufreq_driver_lock, flags);
220
221 if (cpufreq_driver) {
222 /* get the CPU */
223 policy = cpufreq_cpu_get_raw(cpu);
224 if (policy)
225 kobject_get(&policy->kobj);
226 }
227
228 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
229
230 return policy;
231}
232EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
233
234/**
235 * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
236 * @policy: cpufreq policy returned by cpufreq_cpu_get().
237 */
238void cpufreq_cpu_put(struct cpufreq_policy *policy)
239{
240 kobject_put(&policy->kobj);
241}
242EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
243
244/**
245 * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
246 * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
247 */
248void cpufreq_cpu_release(struct cpufreq_policy *policy)
249{
250 if (WARN_ON(!policy))
251 return;
252
253 lockdep_assert_held(&policy->rwsem);
254
255 up_write(&policy->rwsem);
256
257 cpufreq_cpu_put(policy);
258}
259
260/**
261 * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
262 * @cpu: CPU to find the policy for.
263 *
264 * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
265 * if the policy returned by it is not NULL, acquire its rwsem for writing.
266 * Return the policy if it is active or release it and return NULL otherwise.
267 *
268 * The policy returned by this function has to be released with the help of
269 * cpufreq_cpu_release() in order to release its rwsem and balance its usage
270 * counter properly.
271 */
272struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
273{
274 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
275
276 if (!policy)
277 return NULL;
278
279 down_write(&policy->rwsem);
280
281 if (policy_is_inactive(policy)) {
282 cpufreq_cpu_release(policy);
283 return NULL;
284 }
285
286 return policy;
287}
288
289/*********************************************************************
290 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
291 *********************************************************************/
292
293/**
294 * adjust_jiffies - adjust the system "loops_per_jiffy"
295 *
296 * This function alters the system "loops_per_jiffy" for the clock
297 * speed change. Note that loops_per_jiffy cannot be updated on SMP
298 * systems as each CPU might be scaled differently. So, use the arch
299 * per-CPU loops_per_jiffy value wherever possible.
300 */
301static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
302{
303#ifndef CONFIG_SMP
304 static unsigned long l_p_j_ref;
305 static unsigned int l_p_j_ref_freq;
306
307 if (ci->flags & CPUFREQ_CONST_LOOPS)
308 return;
309
310 if (!l_p_j_ref_freq) {
311 l_p_j_ref = loops_per_jiffy;
312 l_p_j_ref_freq = ci->old;
313 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
314 l_p_j_ref, l_p_j_ref_freq);
315 }
316 if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
317 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
318 ci->new);
319 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
320 loops_per_jiffy, ci->new);
321 }
322#endif
323}
324
325/**
326 * cpufreq_notify_transition - Notify frequency transition and adjust_jiffies.
327 * @policy: cpufreq policy to enable fast frequency switching for.
328 * @freqs: contain details of the frequency update.
329 * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
330 *
331 * This function calls the transition notifiers and the "adjust_jiffies"
332 * function. It is called twice on all CPU frequency changes that have
333 * external effects.
334 */
335static void cpufreq_notify_transition(struct cpufreq_policy *policy,
336 struct cpufreq_freqs *freqs,
337 unsigned int state)
338{
339 int cpu;
340
341 BUG_ON(irqs_disabled());
342
343 if (cpufreq_disabled())
344 return;
345
346 freqs->policy = policy;
347 freqs->flags = cpufreq_driver->flags;
348 pr_debug("notification %u of frequency transition to %u kHz\n",
349 state, freqs->new);
350
351 switch (state) {
352 case CPUFREQ_PRECHANGE:
353 /*
354 * Detect if the driver reported a value as "old frequency"
355 * which is not equal to what the cpufreq core thinks is
356 * "old frequency".
357 */
358 if (policy->cur && policy->cur != freqs->old) {
359 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
360 freqs->old, policy->cur);
361 freqs->old = policy->cur;
362 }
363
364 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
365 CPUFREQ_PRECHANGE, freqs);
366
367 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
368 break;
369
370 case CPUFREQ_POSTCHANGE:
371 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
372 pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
373 cpumask_pr_args(policy->cpus));
374
375 for_each_cpu(cpu, policy->cpus)
376 trace_cpu_frequency(freqs->new, cpu);
377
378 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
379 CPUFREQ_POSTCHANGE, freqs);
380
381 cpufreq_stats_record_transition(policy, freqs->new);
382 policy->cur = freqs->new;
383 }
384}
385
386/* Do post notifications when there are chances that transition has failed */
387static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
388 struct cpufreq_freqs *freqs, int transition_failed)
389{
390 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
391 if (!transition_failed)
392 return;
393
394 swap(freqs->old, freqs->new);
395 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
396 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
397}
398
399void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
400 struct cpufreq_freqs *freqs)
401{
402
403 /*
404 * Catch double invocations of _begin() which lead to self-deadlock.
405 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
406 * doesn't invoke _begin() on their behalf, and hence the chances of
407 * double invocations are very low. Moreover, there are scenarios
408 * where these checks can emit false-positive warnings in these
409 * drivers; so we avoid that by skipping them altogether.
410 */
411 WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
412 && current == policy->transition_task);
413
414wait:
415 wait_event(policy->transition_wait, !policy->transition_ongoing);
416
417 spin_lock(&policy->transition_lock);
418
419 if (unlikely(policy->transition_ongoing)) {
420 spin_unlock(&policy->transition_lock);
421 goto wait;
422 }
423
424 policy->transition_ongoing = true;
425 policy->transition_task = current;
426
427 spin_unlock(&policy->transition_lock);
428
429 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
430}
431EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
432
433void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
434 struct cpufreq_freqs *freqs, int transition_failed)
435{
436 if (WARN_ON(!policy->transition_ongoing))
437 return;
438
439 cpufreq_notify_post_transition(policy, freqs, transition_failed);
440
441 policy->transition_ongoing = false;
442 policy->transition_task = NULL;
443
444 wake_up(&policy->transition_wait);
445}
446EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
447
448/*
449 * Fast frequency switching status count. Positive means "enabled", negative
450 * means "disabled" and 0 means "not decided yet".
451 */
452static int cpufreq_fast_switch_count;
453static DEFINE_MUTEX(cpufreq_fast_switch_lock);
454
455static void cpufreq_list_transition_notifiers(void)
456{
457 struct notifier_block *nb;
458
459 pr_info("Registered transition notifiers:\n");
460
461 mutex_lock(&cpufreq_transition_notifier_list.mutex);
462
463 for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
464 pr_info("%pS\n", nb->notifier_call);
465
466 mutex_unlock(&cpufreq_transition_notifier_list.mutex);
467}
468
469/**
470 * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
471 * @policy: cpufreq policy to enable fast frequency switching for.
472 *
473 * Try to enable fast frequency switching for @policy.
474 *
475 * The attempt will fail if there is at least one transition notifier registered
476 * at this point, as fast frequency switching is quite fundamentally at odds
477 * with transition notifiers. Thus if successful, it will make registration of
478 * transition notifiers fail going forward.
479 */
480void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
481{
482 lockdep_assert_held(&policy->rwsem);
483
484 if (!policy->fast_switch_possible)
485 return;
486
487 mutex_lock(&cpufreq_fast_switch_lock);
488 if (cpufreq_fast_switch_count >= 0) {
489 cpufreq_fast_switch_count++;
490 policy->fast_switch_enabled = true;
491 } else {
492 pr_warn("CPU%u: Fast frequency switching not enabled\n",
493 policy->cpu);
494 cpufreq_list_transition_notifiers();
495 }
496 mutex_unlock(&cpufreq_fast_switch_lock);
497}
498EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
499
500/**
501 * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
502 * @policy: cpufreq policy to disable fast frequency switching for.
503 */
504void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
505{
506 mutex_lock(&cpufreq_fast_switch_lock);
507 if (policy->fast_switch_enabled) {
508 policy->fast_switch_enabled = false;
509 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
510 cpufreq_fast_switch_count--;
511 }
512 mutex_unlock(&cpufreq_fast_switch_lock);
513}
514EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
515
516/**
517 * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
518 * one.
519 * @target_freq: target frequency to resolve.
520 *
521 * The target to driver frequency mapping is cached in the policy.
522 *
523 * Return: Lowest driver-supported frequency greater than or equal to the
524 * given target_freq, subject to policy (min/max) and driver limitations.
525 */
526unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
527 unsigned int target_freq)
528{
529 target_freq = clamp_val(target_freq, policy->min, policy->max);
530 policy->cached_target_freq = target_freq;
531
532 if (cpufreq_driver->target_index) {
533 int idx;
534
535 idx = cpufreq_frequency_table_target(policy, target_freq,
536 CPUFREQ_RELATION_L);
537 policy->cached_resolved_idx = idx;
538 return policy->freq_table[idx].frequency;
539 }
540
541 if (cpufreq_driver->resolve_freq)
542 return cpufreq_driver->resolve_freq(policy, target_freq);
543
544 return target_freq;
545}
546EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
547
548unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
549{
550 unsigned int latency;
551
552 if (policy->transition_delay_us)
553 return policy->transition_delay_us;
554
555 latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
556 if (latency) {
557 /*
558 * For platforms that can change the frequency very fast (< 10
559 * us), the above formula gives a decent transition delay. But
560 * for platforms where transition_latency is in milliseconds, it
561 * ends up giving unrealistic values.
562 *
563 * Cap the default transition delay to 10 ms, which seems to be
564 * a reasonable amount of time after which we should reevaluate
565 * the frequency.
566 */
567 return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
568 }
569
570 return LATENCY_MULTIPLIER;
571}
572EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
573
574/*********************************************************************
575 * SYSFS INTERFACE *
576 *********************************************************************/
577static ssize_t show_boost(struct kobject *kobj,
578 struct kobj_attribute *attr, char *buf)
579{
580 return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
581}
582
583static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
584 const char *buf, size_t count)
585{
586 int ret, enable;
587
588 ret = sscanf(buf, "%d", &enable);
589 if (ret != 1 || enable < 0 || enable > 1)
590 return -EINVAL;
591
592 if (cpufreq_boost_trigger_state(enable)) {
593 pr_err("%s: Cannot %s BOOST!\n",
594 __func__, enable ? "enable" : "disable");
595 return -EINVAL;
596 }
597
598 pr_debug("%s: cpufreq BOOST %s\n",
599 __func__, enable ? "enabled" : "disabled");
600
601 return count;
602}
603define_one_global_rw(boost);
604
605static struct cpufreq_governor *find_governor(const char *str_governor)
606{
607 struct cpufreq_governor *t;
608
609 for_each_governor(t)
610 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
611 return t;
612
613 return NULL;
614}
615
616static int cpufreq_parse_policy(char *str_governor,
617 struct cpufreq_policy *policy)
618{
619 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
620 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
621 return 0;
622 }
623 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
624 policy->policy = CPUFREQ_POLICY_POWERSAVE;
625 return 0;
626 }
627 return -EINVAL;
628}
629
630/**
631 * cpufreq_parse_governor - parse a governor string only for has_target()
632 */
633static int cpufreq_parse_governor(char *str_governor,
634 struct cpufreq_policy *policy)
635{
636 struct cpufreq_governor *t;
637
638 mutex_lock(&cpufreq_governor_mutex);
639
640 t = find_governor(str_governor);
641 if (!t) {
642 int ret;
643
644 mutex_unlock(&cpufreq_governor_mutex);
645
646 ret = request_module("cpufreq_%s", str_governor);
647 if (ret)
648 return -EINVAL;
649
650 mutex_lock(&cpufreq_governor_mutex);
651
652 t = find_governor(str_governor);
653 }
654 if (t && !try_module_get(t->owner))
655 t = NULL;
656
657 mutex_unlock(&cpufreq_governor_mutex);
658
659 if (t) {
660 policy->governor = t;
661 return 0;
662 }
663
664 return -EINVAL;
665}
666
667/**
668 * cpufreq_per_cpu_attr_read() / show_##file_name() -
669 * print out cpufreq information
670 *
671 * Write out information from cpufreq_driver->policy[cpu]; object must be
672 * "unsigned int".
673 */
674
675#define show_one(file_name, object) \
676static ssize_t show_##file_name \
677(struct cpufreq_policy *policy, char *buf) \
678{ \
679 return sprintf(buf, "%u\n", policy->object); \
680}
681
682show_one(cpuinfo_min_freq, cpuinfo.min_freq);
683show_one(cpuinfo_max_freq, cpuinfo.max_freq);
684show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
685show_one(scaling_min_freq, min);
686show_one(scaling_max_freq, max);
687
688__weak unsigned int arch_freq_get_on_cpu(int cpu)
689{
690 return 0;
691}
692
693static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
694{
695 ssize_t ret;
696 unsigned int freq;
697
698 freq = arch_freq_get_on_cpu(policy->cpu);
699 if (freq)
700 ret = sprintf(buf, "%u\n", freq);
701 else if (cpufreq_driver && cpufreq_driver->setpolicy &&
702 cpufreq_driver->get)
703 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
704 else
705 ret = sprintf(buf, "%u\n", policy->cur);
706 return ret;
707}
708
709/**
710 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
711 */
712#define store_one(file_name, object) \
713static ssize_t store_##file_name \
714(struct cpufreq_policy *policy, const char *buf, size_t count) \
715{ \
716 unsigned long val; \
717 int ret; \
718 \
719 ret = sscanf(buf, "%lu", &val); \
720 if (ret != 1) \
721 return -EINVAL; \
722 \
723 ret = freq_qos_update_request(policy->object##_freq_req, val);\
724 return ret >= 0 ? count : ret; \
725}
726
727store_one(scaling_min_freq, min);
728store_one(scaling_max_freq, max);
729
730/**
731 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
732 */
733static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
734 char *buf)
735{
736 unsigned int cur_freq = __cpufreq_get(policy);
737
738 if (cur_freq)
739 return sprintf(buf, "%u\n", cur_freq);
740
741 return sprintf(buf, "<unknown>\n");
742}
743
744/**
745 * show_scaling_governor - show the current policy for the specified CPU
746 */
747static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
748{
749 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
750 return sprintf(buf, "powersave\n");
751 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
752 return sprintf(buf, "performance\n");
753 else if (policy->governor)
754 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
755 policy->governor->name);
756 return -EINVAL;
757}
758
759/**
760 * store_scaling_governor - store policy for the specified CPU
761 */
762static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
763 const char *buf, size_t count)
764{
765 int ret;
766 char str_governor[16];
767 struct cpufreq_policy new_policy;
768
769 memcpy(&new_policy, policy, sizeof(*policy));
770
771 ret = sscanf(buf, "%15s", str_governor);
772 if (ret != 1)
773 return -EINVAL;
774
775 if (cpufreq_driver->setpolicy) {
776 if (cpufreq_parse_policy(str_governor, &new_policy))
777 return -EINVAL;
778 } else {
779 if (cpufreq_parse_governor(str_governor, &new_policy))
780 return -EINVAL;
781 }
782
783 ret = cpufreq_set_policy(policy, &new_policy);
784
785 if (new_policy.governor)
786 module_put(new_policy.governor->owner);
787
788 return ret ? ret : count;
789}
790
791/**
792 * show_scaling_driver - show the cpufreq driver currently loaded
793 */
794static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
795{
796 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
797}
798
799/**
800 * show_scaling_available_governors - show the available CPUfreq governors
801 */
802static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
803 char *buf)
804{
805 ssize_t i = 0;
806 struct cpufreq_governor *t;
807
808 if (!has_target()) {
809 i += sprintf(buf, "performance powersave");
810 goto out;
811 }
812
813 for_each_governor(t) {
814 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
815 - (CPUFREQ_NAME_LEN + 2)))
816 goto out;
817 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
818 }
819out:
820 i += sprintf(&buf[i], "\n");
821 return i;
822}
823
824ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
825{
826 ssize_t i = 0;
827 unsigned int cpu;
828
829 for_each_cpu(cpu, mask) {
830 if (i)
831 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
832 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
833 if (i >= (PAGE_SIZE - 5))
834 break;
835 }
836 i += sprintf(&buf[i], "\n");
837 return i;
838}
839EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
840
841/**
842 * show_related_cpus - show the CPUs affected by each transition even if
843 * hw coordination is in use
844 */
845static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
846{
847 return cpufreq_show_cpus(policy->related_cpus, buf);
848}
849
850/**
851 * show_affected_cpus - show the CPUs affected by each transition
852 */
853static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
854{
855 return cpufreq_show_cpus(policy->cpus, buf);
856}
857
858static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
859 const char *buf, size_t count)
860{
861 unsigned int freq = 0;
862 unsigned int ret;
863
864 if (!policy->governor || !policy->governor->store_setspeed)
865 return -EINVAL;
866
867 ret = sscanf(buf, "%u", &freq);
868 if (ret != 1)
869 return -EINVAL;
870
871 policy->governor->store_setspeed(policy, freq);
872
873 return count;
874}
875
876static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
877{
878 if (!policy->governor || !policy->governor->show_setspeed)
879 return sprintf(buf, "<unsupported>\n");
880
881 return policy->governor->show_setspeed(policy, buf);
882}
883
884/**
885 * show_bios_limit - show the current cpufreq HW/BIOS limitation
886 */
887static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
888{
889 unsigned int limit;
890 int ret;
891 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
892 if (!ret)
893 return sprintf(buf, "%u\n", limit);
894 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
895}
896
897cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
898cpufreq_freq_attr_ro(cpuinfo_min_freq);
899cpufreq_freq_attr_ro(cpuinfo_max_freq);
900cpufreq_freq_attr_ro(cpuinfo_transition_latency);
901cpufreq_freq_attr_ro(scaling_available_governors);
902cpufreq_freq_attr_ro(scaling_driver);
903cpufreq_freq_attr_ro(scaling_cur_freq);
904cpufreq_freq_attr_ro(bios_limit);
905cpufreq_freq_attr_ro(related_cpus);
906cpufreq_freq_attr_ro(affected_cpus);
907cpufreq_freq_attr_rw(scaling_min_freq);
908cpufreq_freq_attr_rw(scaling_max_freq);
909cpufreq_freq_attr_rw(scaling_governor);
910cpufreq_freq_attr_rw(scaling_setspeed);
911
912static struct attribute *default_attrs[] = {
913 &cpuinfo_min_freq.attr,
914 &cpuinfo_max_freq.attr,
915 &cpuinfo_transition_latency.attr,
916 &scaling_min_freq.attr,
917 &scaling_max_freq.attr,
918 &affected_cpus.attr,
919 &related_cpus.attr,
920 &scaling_governor.attr,
921 &scaling_driver.attr,
922 &scaling_available_governors.attr,
923 &scaling_setspeed.attr,
924 NULL
925};
926
927#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
928#define to_attr(a) container_of(a, struct freq_attr, attr)
929
930static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
931{
932 struct cpufreq_policy *policy = to_policy(kobj);
933 struct freq_attr *fattr = to_attr(attr);
934 ssize_t ret;
935
936 down_read(&policy->rwsem);
937 ret = fattr->show(policy, buf);
938 up_read(&policy->rwsem);
939
940 return ret;
941}
942
943static ssize_t store(struct kobject *kobj, struct attribute *attr,
944 const char *buf, size_t count)
945{
946 struct cpufreq_policy *policy = to_policy(kobj);
947 struct freq_attr *fattr = to_attr(attr);
948 ssize_t ret = -EINVAL;
949
950 /*
951 * cpus_read_trylock() is used here to work around a circular lock
952 * dependency problem with respect to the cpufreq_register_driver().
953 */
954 if (!cpus_read_trylock())
955 return -EBUSY;
956
957 if (cpu_online(policy->cpu)) {
958 down_write(&policy->rwsem);
959 ret = fattr->store(policy, buf, count);
960 up_write(&policy->rwsem);
961 }
962
963 cpus_read_unlock();
964
965 return ret;
966}
967
968static void cpufreq_sysfs_release(struct kobject *kobj)
969{
970 struct cpufreq_policy *policy = to_policy(kobj);
971 pr_debug("last reference is dropped\n");
972 complete(&policy->kobj_unregister);
973}
974
975static const struct sysfs_ops sysfs_ops = {
976 .show = show,
977 .store = store,
978};
979
980static struct kobj_type ktype_cpufreq = {
981 .sysfs_ops = &sysfs_ops,
982 .default_attrs = default_attrs,
983 .release = cpufreq_sysfs_release,
984};
985
986static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
987{
988 struct device *dev = get_cpu_device(cpu);
989
990 if (unlikely(!dev))
991 return;
992
993 if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
994 return;
995
996 dev_dbg(dev, "%s: Adding symlink\n", __func__);
997 if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
998 dev_err(dev, "cpufreq symlink creation failed\n");
999}
1000
1001static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
1002 struct device *dev)
1003{
1004 dev_dbg(dev, "%s: Removing symlink\n", __func__);
1005 sysfs_remove_link(&dev->kobj, "cpufreq");
1006}
1007
1008static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1009{
1010 struct freq_attr **drv_attr;
1011 int ret = 0;
1012
1013 /* set up files for this cpu device */
1014 drv_attr = cpufreq_driver->attr;
1015 while (drv_attr && *drv_attr) {
1016 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1017 if (ret)
1018 return ret;
1019 drv_attr++;
1020 }
1021 if (cpufreq_driver->get) {
1022 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1023 if (ret)
1024 return ret;
1025 }
1026
1027 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1028 if (ret)
1029 return ret;
1030
1031 if (cpufreq_driver->bios_limit) {
1032 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1033 if (ret)
1034 return ret;
1035 }
1036
1037 return 0;
1038}
1039
1040__weak struct cpufreq_governor *cpufreq_default_governor(void)
1041{
1042 return NULL;
1043}
1044
1045static int cpufreq_init_policy(struct cpufreq_policy *policy)
1046{
1047 struct cpufreq_governor *gov = NULL, *def_gov = NULL;
1048 struct cpufreq_policy new_policy;
1049
1050 memcpy(&new_policy, policy, sizeof(*policy));
1051
1052 def_gov = cpufreq_default_governor();
1053
1054 if (has_target()) {
1055 /*
1056 * Update governor of new_policy to the governor used before
1057 * hotplug
1058 */
1059 gov = find_governor(policy->last_governor);
1060 if (gov) {
1061 pr_debug("Restoring governor %s for cpu %d\n",
1062 policy->governor->name, policy->cpu);
1063 } else {
1064 if (!def_gov)
1065 return -ENODATA;
1066 gov = def_gov;
1067 }
1068 new_policy.governor = gov;
1069 } else {
1070 /* Use the default policy if there is no last_policy. */
1071 if (policy->last_policy) {
1072 new_policy.policy = policy->last_policy;
1073 } else {
1074 if (!def_gov)
1075 return -ENODATA;
1076 cpufreq_parse_policy(def_gov->name, &new_policy);
1077 }
1078 }
1079
1080 return cpufreq_set_policy(policy, &new_policy);
1081}
1082
1083static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1084{
1085 int ret = 0;
1086
1087 /* Has this CPU been taken care of already? */
1088 if (cpumask_test_cpu(cpu, policy->cpus))
1089 return 0;
1090
1091 down_write(&policy->rwsem);
1092 if (has_target())
1093 cpufreq_stop_governor(policy);
1094
1095 cpumask_set_cpu(cpu, policy->cpus);
1096
1097 if (has_target()) {
1098 ret = cpufreq_start_governor(policy);
1099 if (ret)
1100 pr_err("%s: Failed to start governor\n", __func__);
1101 }
1102 up_write(&policy->rwsem);
1103 return ret;
1104}
1105
1106void refresh_frequency_limits(struct cpufreq_policy *policy)
1107{
1108 struct cpufreq_policy new_policy;
1109
1110 if (!policy_is_inactive(policy)) {
1111 new_policy = *policy;
1112 pr_debug("updating policy for CPU %u\n", policy->cpu);
1113
1114 cpufreq_set_policy(policy, &new_policy);
1115 }
1116}
1117EXPORT_SYMBOL(refresh_frequency_limits);
1118
1119static void handle_update(struct work_struct *work)
1120{
1121 struct cpufreq_policy *policy =
1122 container_of(work, struct cpufreq_policy, update);
1123
1124 pr_debug("handle_update for cpu %u called\n", policy->cpu);
1125 down_write(&policy->rwsem);
1126 refresh_frequency_limits(policy);
1127 up_write(&policy->rwsem);
1128}
1129
1130static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1131 void *data)
1132{
1133 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1134
1135 schedule_work(&policy->update);
1136 return 0;
1137}
1138
1139static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1140 void *data)
1141{
1142 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1143
1144 schedule_work(&policy->update);
1145 return 0;
1146}
1147
1148static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1149{
1150 struct kobject *kobj;
1151 struct completion *cmp;
1152
1153 down_write(&policy->rwsem);
1154 cpufreq_stats_free_table(policy);
1155 kobj = &policy->kobj;
1156 cmp = &policy->kobj_unregister;
1157 up_write(&policy->rwsem);
1158 kobject_put(kobj);
1159
1160 /*
1161 * We need to make sure that the underlying kobj is
1162 * actually not referenced anymore by anybody before we
1163 * proceed with unloading.
1164 */
1165 pr_debug("waiting for dropping of refcount\n");
1166 wait_for_completion(cmp);
1167 pr_debug("wait complete\n");
1168}
1169
1170static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1171{
1172 struct cpufreq_policy *policy;
1173 struct device *dev = get_cpu_device(cpu);
1174 int ret;
1175
1176 if (!dev)
1177 return NULL;
1178
1179 policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1180 if (!policy)
1181 return NULL;
1182
1183 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1184 goto err_free_policy;
1185
1186 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1187 goto err_free_cpumask;
1188
1189 if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1190 goto err_free_rcpumask;
1191
1192 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1193 cpufreq_global_kobject, "policy%u", cpu);
1194 if (ret) {
1195 dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1196 /*
1197 * The entire policy object will be freed below, but the extra
1198 * memory allocated for the kobject name needs to be freed by
1199 * releasing the kobject.
1200 */
1201 kobject_put(&policy->kobj);
1202 goto err_free_real_cpus;
1203 }
1204
1205 freq_constraints_init(&policy->constraints);
1206
1207 policy->nb_min.notifier_call = cpufreq_notifier_min;
1208 policy->nb_max.notifier_call = cpufreq_notifier_max;
1209
1210 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1211 &policy->nb_min);
1212 if (ret) {
1213 dev_err(dev, "Failed to register MIN QoS notifier: %d (%*pbl)\n",
1214 ret, cpumask_pr_args(policy->cpus));
1215 goto err_kobj_remove;
1216 }
1217
1218 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1219 &policy->nb_max);
1220 if (ret) {
1221 dev_err(dev, "Failed to register MAX QoS notifier: %d (%*pbl)\n",
1222 ret, cpumask_pr_args(policy->cpus));
1223 goto err_min_qos_notifier;
1224 }
1225
1226 INIT_LIST_HEAD(&policy->policy_list);
1227 init_rwsem(&policy->rwsem);
1228 spin_lock_init(&policy->transition_lock);
1229 init_waitqueue_head(&policy->transition_wait);
1230 init_completion(&policy->kobj_unregister);
1231 INIT_WORK(&policy->update, handle_update);
1232
1233 policy->cpu = cpu;
1234 return policy;
1235
1236err_min_qos_notifier:
1237 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1238 &policy->nb_min);
1239err_kobj_remove:
1240 cpufreq_policy_put_kobj(policy);
1241err_free_real_cpus:
1242 free_cpumask_var(policy->real_cpus);
1243err_free_rcpumask:
1244 free_cpumask_var(policy->related_cpus);
1245err_free_cpumask:
1246 free_cpumask_var(policy->cpus);
1247err_free_policy:
1248 kfree(policy);
1249
1250 return NULL;
1251}
1252
1253static void cpufreq_policy_free(struct cpufreq_policy *policy)
1254{
1255 unsigned long flags;
1256 int cpu;
1257
1258 /* Remove policy from list */
1259 write_lock_irqsave(&cpufreq_driver_lock, flags);
1260 list_del(&policy->policy_list);
1261
1262 for_each_cpu(cpu, policy->related_cpus)
1263 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1264 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1265
1266 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1267 &policy->nb_max);
1268 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1269 &policy->nb_min);
1270
1271 /* Cancel any pending policy->update work before freeing the policy. */
1272 cancel_work_sync(&policy->update);
1273
1274 if (policy->max_freq_req) {
1275 /*
1276 * CPUFREQ_CREATE_POLICY notification is sent only after
1277 * successfully adding max_freq_req request.
1278 */
1279 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1280 CPUFREQ_REMOVE_POLICY, policy);
1281 freq_qos_remove_request(policy->max_freq_req);
1282 }
1283
1284 freq_qos_remove_request(policy->min_freq_req);
1285 kfree(policy->min_freq_req);
1286
1287 cpufreq_policy_put_kobj(policy);
1288 free_cpumask_var(policy->real_cpus);
1289 free_cpumask_var(policy->related_cpus);
1290 free_cpumask_var(policy->cpus);
1291 kfree(policy);
1292}
1293
1294static int cpufreq_online(unsigned int cpu)
1295{
1296 struct cpufreq_policy *policy;
1297 bool new_policy;
1298 unsigned long flags;
1299 unsigned int j;
1300 int ret;
1301
1302 pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1303
1304 /* Check if this CPU already has a policy to manage it */
1305 policy = per_cpu(cpufreq_cpu_data, cpu);
1306 if (policy) {
1307 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1308 if (!policy_is_inactive(policy))
1309 return cpufreq_add_policy_cpu(policy, cpu);
1310
1311 /* This is the only online CPU for the policy. Start over. */
1312 new_policy = false;
1313 down_write(&policy->rwsem);
1314 policy->cpu = cpu;
1315 policy->governor = NULL;
1316 up_write(&policy->rwsem);
1317 } else {
1318 new_policy = true;
1319 policy = cpufreq_policy_alloc(cpu);
1320 if (!policy)
1321 return -ENOMEM;
1322 }
1323
1324 if (!new_policy && cpufreq_driver->online) {
1325 ret = cpufreq_driver->online(policy);
1326 if (ret) {
1327 pr_debug("%s: %d: initialization failed\n", __func__,
1328 __LINE__);
1329 goto out_exit_policy;
1330 }
1331
1332 /* Recover policy->cpus using related_cpus */
1333 cpumask_copy(policy->cpus, policy->related_cpus);
1334 } else {
1335 cpumask_copy(policy->cpus, cpumask_of(cpu));
1336
1337 /*
1338 * Call driver. From then on the cpufreq must be able
1339 * to accept all calls to ->verify and ->setpolicy for this CPU.
1340 */
1341 ret = cpufreq_driver->init(policy);
1342 if (ret) {
1343 pr_debug("%s: %d: initialization failed\n", __func__,
1344 __LINE__);
1345 goto out_free_policy;
1346 }
1347
1348 ret = cpufreq_table_validate_and_sort(policy);
1349 if (ret)
1350 goto out_exit_policy;
1351
1352 /* related_cpus should at least include policy->cpus. */
1353 cpumask_copy(policy->related_cpus, policy->cpus);
1354 }
1355
1356 down_write(&policy->rwsem);
1357 /*
1358 * affected cpus must always be the one, which are online. We aren't
1359 * managing offline cpus here.
1360 */
1361 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1362
1363 if (new_policy) {
1364 for_each_cpu(j, policy->related_cpus) {
1365 per_cpu(cpufreq_cpu_data, j) = policy;
1366 add_cpu_dev_symlink(policy, j);
1367 }
1368
1369 policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1370 GFP_KERNEL);
1371 if (!policy->min_freq_req)
1372 goto out_destroy_policy;
1373
1374 ret = freq_qos_add_request(&policy->constraints,
1375 policy->min_freq_req, FREQ_QOS_MIN,
1376 policy->min);
1377 if (ret < 0) {
1378 /*
1379 * So we don't call freq_qos_remove_request() for an
1380 * uninitialized request.
1381 */
1382 kfree(policy->min_freq_req);
1383 policy->min_freq_req = NULL;
1384 goto out_destroy_policy;
1385 }
1386
1387 /*
1388 * This must be initialized right here to avoid calling
1389 * freq_qos_remove_request() on uninitialized request in case
1390 * of errors.
1391 */
1392 policy->max_freq_req = policy->min_freq_req + 1;
1393
1394 ret = freq_qos_add_request(&policy->constraints,
1395 policy->max_freq_req, FREQ_QOS_MAX,
1396 policy->max);
1397 if (ret < 0) {
1398 policy->max_freq_req = NULL;
1399 goto out_destroy_policy;
1400 }
1401
1402 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1403 CPUFREQ_CREATE_POLICY, policy);
1404 }
1405
1406 if (cpufreq_driver->get && has_target()) {
1407 policy->cur = cpufreq_driver->get(policy->cpu);
1408 if (!policy->cur) {
1409 pr_err("%s: ->get() failed\n", __func__);
1410 goto out_destroy_policy;
1411 }
1412 }
1413
1414 /*
1415 * Sometimes boot loaders set CPU frequency to a value outside of
1416 * frequency table present with cpufreq core. In such cases CPU might be
1417 * unstable if it has to run on that frequency for long duration of time
1418 * and so its better to set it to a frequency which is specified in
1419 * freq-table. This also makes cpufreq stats inconsistent as
1420 * cpufreq-stats would fail to register because current frequency of CPU
1421 * isn't found in freq-table.
1422 *
1423 * Because we don't want this change to effect boot process badly, we go
1424 * for the next freq which is >= policy->cur ('cur' must be set by now,
1425 * otherwise we will end up setting freq to lowest of the table as 'cur'
1426 * is initialized to zero).
1427 *
1428 * We are passing target-freq as "policy->cur - 1" otherwise
1429 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1430 * equal to target-freq.
1431 */
1432 if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1433 && has_target()) {
1434 /* Are we running at unknown frequency ? */
1435 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1436 if (ret == -EINVAL) {
1437 /* Warn user and fix it */
1438 pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1439 __func__, policy->cpu, policy->cur);
1440 ret = __cpufreq_driver_target(policy, policy->cur - 1,
1441 CPUFREQ_RELATION_L);
1442
1443 /*
1444 * Reaching here after boot in a few seconds may not
1445 * mean that system will remain stable at "unknown"
1446 * frequency for longer duration. Hence, a BUG_ON().
1447 */
1448 BUG_ON(ret);
1449 pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1450 __func__, policy->cpu, policy->cur);
1451 }
1452 }
1453
1454 if (new_policy) {
1455 ret = cpufreq_add_dev_interface(policy);
1456 if (ret)
1457 goto out_destroy_policy;
1458
1459 cpufreq_stats_create_table(policy);
1460
1461 write_lock_irqsave(&cpufreq_driver_lock, flags);
1462 list_add(&policy->policy_list, &cpufreq_policy_list);
1463 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1464 }
1465
1466 ret = cpufreq_init_policy(policy);
1467 if (ret) {
1468 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1469 __func__, cpu, ret);
1470 goto out_destroy_policy;
1471 }
1472
1473 up_write(&policy->rwsem);
1474
1475 kobject_uevent(&policy->kobj, KOBJ_ADD);
1476
1477 /* Callback for handling stuff after policy is ready */
1478 if (cpufreq_driver->ready)
1479 cpufreq_driver->ready(policy);
1480
1481 if (cpufreq_thermal_control_enabled(cpufreq_driver))
1482 policy->cdev = of_cpufreq_cooling_register(policy);
1483
1484 pr_debug("initialization complete\n");
1485
1486 return 0;
1487
1488out_destroy_policy:
1489 for_each_cpu(j, policy->real_cpus)
1490 remove_cpu_dev_symlink(policy, get_cpu_device(j));
1491
1492 up_write(&policy->rwsem);
1493
1494out_exit_policy:
1495 if (cpufreq_driver->exit)
1496 cpufreq_driver->exit(policy);
1497
1498out_free_policy:
1499 cpufreq_policy_free(policy);
1500 return ret;
1501}
1502
1503/**
1504 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1505 * @dev: CPU device.
1506 * @sif: Subsystem interface structure pointer (not used)
1507 */
1508static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1509{
1510 struct cpufreq_policy *policy;
1511 unsigned cpu = dev->id;
1512 int ret;
1513
1514 dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1515
1516 if (cpu_online(cpu)) {
1517 ret = cpufreq_online(cpu);
1518 if (ret)
1519 return ret;
1520 }
1521
1522 /* Create sysfs link on CPU registration */
1523 policy = per_cpu(cpufreq_cpu_data, cpu);
1524 if (policy)
1525 add_cpu_dev_symlink(policy, cpu);
1526
1527 return 0;
1528}
1529
1530static int cpufreq_offline(unsigned int cpu)
1531{
1532 struct cpufreq_policy *policy;
1533 int ret;
1534
1535 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1536
1537 policy = cpufreq_cpu_get_raw(cpu);
1538 if (!policy) {
1539 pr_debug("%s: No cpu_data found\n", __func__);
1540 return 0;
1541 }
1542
1543 down_write(&policy->rwsem);
1544 if (has_target())
1545 cpufreq_stop_governor(policy);
1546
1547 cpumask_clear_cpu(cpu, policy->cpus);
1548
1549 if (policy_is_inactive(policy)) {
1550 if (has_target())
1551 strncpy(policy->last_governor, policy->governor->name,
1552 CPUFREQ_NAME_LEN);
1553 else
1554 policy->last_policy = policy->policy;
1555 } else if (cpu == policy->cpu) {
1556 /* Nominate new CPU */
1557 policy->cpu = cpumask_any(policy->cpus);
1558 }
1559
1560 /* Start governor again for active policy */
1561 if (!policy_is_inactive(policy)) {
1562 if (has_target()) {
1563 ret = cpufreq_start_governor(policy);
1564 if (ret)
1565 pr_err("%s: Failed to start governor\n", __func__);
1566 }
1567
1568 goto unlock;
1569 }
1570
1571 if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1572 cpufreq_cooling_unregister(policy->cdev);
1573 policy->cdev = NULL;
1574 }
1575
1576 if (cpufreq_driver->stop_cpu)
1577 cpufreq_driver->stop_cpu(policy);
1578
1579 if (has_target())
1580 cpufreq_exit_governor(policy);
1581
1582 /*
1583 * Perform the ->offline() during light-weight tear-down, as
1584 * that allows fast recovery when the CPU comes back.
1585 */
1586 if (cpufreq_driver->offline) {
1587 cpufreq_driver->offline(policy);
1588 } else if (cpufreq_driver->exit) {
1589 cpufreq_driver->exit(policy);
1590 policy->freq_table = NULL;
1591 }
1592
1593unlock:
1594 up_write(&policy->rwsem);
1595 return 0;
1596}
1597
1598/**
1599 * cpufreq_remove_dev - remove a CPU device
1600 *
1601 * Removes the cpufreq interface for a CPU device.
1602 */
1603static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1604{
1605 unsigned int cpu = dev->id;
1606 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1607
1608 if (!policy)
1609 return;
1610
1611 if (cpu_online(cpu))
1612 cpufreq_offline(cpu);
1613
1614 cpumask_clear_cpu(cpu, policy->real_cpus);
1615 remove_cpu_dev_symlink(policy, dev);
1616
1617 if (cpumask_empty(policy->real_cpus)) {
1618 /* We did light-weight exit earlier, do full tear down now */
1619 if (cpufreq_driver->offline)
1620 cpufreq_driver->exit(policy);
1621
1622 cpufreq_policy_free(policy);
1623 }
1624}
1625
1626/**
1627 * cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1628 * in deep trouble.
1629 * @policy: policy managing CPUs
1630 * @new_freq: CPU frequency the CPU actually runs at
1631 *
1632 * We adjust to current frequency first, and need to clean up later.
1633 * So either call to cpufreq_update_policy() or schedule handle_update()).
1634 */
1635static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1636 unsigned int new_freq)
1637{
1638 struct cpufreq_freqs freqs;
1639
1640 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1641 policy->cur, new_freq);
1642
1643 freqs.old = policy->cur;
1644 freqs.new = new_freq;
1645
1646 cpufreq_freq_transition_begin(policy, &freqs);
1647 cpufreq_freq_transition_end(policy, &freqs, 0);
1648}
1649
1650static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1651{
1652 unsigned int new_freq;
1653
1654 new_freq = cpufreq_driver->get(policy->cpu);
1655 if (!new_freq)
1656 return 0;
1657
1658 /*
1659 * If fast frequency switching is used with the given policy, the check
1660 * against policy->cur is pointless, so skip it in that case.
1661 */
1662 if (policy->fast_switch_enabled || !has_target())
1663 return new_freq;
1664
1665 if (policy->cur != new_freq) {
1666 cpufreq_out_of_sync(policy, new_freq);
1667 if (update)
1668 schedule_work(&policy->update);
1669 }
1670
1671 return new_freq;
1672}
1673
1674/**
1675 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1676 * @cpu: CPU number
1677 *
1678 * This is the last known freq, without actually getting it from the driver.
1679 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1680 */
1681unsigned int cpufreq_quick_get(unsigned int cpu)
1682{
1683 struct cpufreq_policy *policy;
1684 unsigned int ret_freq = 0;
1685 unsigned long flags;
1686
1687 read_lock_irqsave(&cpufreq_driver_lock, flags);
1688
1689 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1690 ret_freq = cpufreq_driver->get(cpu);
1691 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1692 return ret_freq;
1693 }
1694
1695 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1696
1697 policy = cpufreq_cpu_get(cpu);
1698 if (policy) {
1699 ret_freq = policy->cur;
1700 cpufreq_cpu_put(policy);
1701 }
1702
1703 return ret_freq;
1704}
1705EXPORT_SYMBOL(cpufreq_quick_get);
1706
1707/**
1708 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1709 * @cpu: CPU number
1710 *
1711 * Just return the max possible frequency for a given CPU.
1712 */
1713unsigned int cpufreq_quick_get_max(unsigned int cpu)
1714{
1715 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1716 unsigned int ret_freq = 0;
1717
1718 if (policy) {
1719 ret_freq = policy->max;
1720 cpufreq_cpu_put(policy);
1721 }
1722
1723 return ret_freq;
1724}
1725EXPORT_SYMBOL(cpufreq_quick_get_max);
1726
1727static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1728{
1729 if (unlikely(policy_is_inactive(policy)))
1730 return 0;
1731
1732 return cpufreq_verify_current_freq(policy, true);
1733}
1734
1735/**
1736 * cpufreq_get - get the current CPU frequency (in kHz)
1737 * @cpu: CPU number
1738 *
1739 * Get the CPU current (static) CPU frequency
1740 */
1741unsigned int cpufreq_get(unsigned int cpu)
1742{
1743 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1744 unsigned int ret_freq = 0;
1745
1746 if (policy) {
1747 down_read(&policy->rwsem);
1748 if (cpufreq_driver->get)
1749 ret_freq = __cpufreq_get(policy);
1750 up_read(&policy->rwsem);
1751
1752 cpufreq_cpu_put(policy);
1753 }
1754
1755 return ret_freq;
1756}
1757EXPORT_SYMBOL(cpufreq_get);
1758
1759static struct subsys_interface cpufreq_interface = {
1760 .name = "cpufreq",
1761 .subsys = &cpu_subsys,
1762 .add_dev = cpufreq_add_dev,
1763 .remove_dev = cpufreq_remove_dev,
1764};
1765
1766/*
1767 * In case platform wants some specific frequency to be configured
1768 * during suspend..
1769 */
1770int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1771{
1772 int ret;
1773
1774 if (!policy->suspend_freq) {
1775 pr_debug("%s: suspend_freq not defined\n", __func__);
1776 return 0;
1777 }
1778
1779 pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1780 policy->suspend_freq);
1781
1782 ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1783 CPUFREQ_RELATION_H);
1784 if (ret)
1785 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1786 __func__, policy->suspend_freq, ret);
1787
1788 return ret;
1789}
1790EXPORT_SYMBOL(cpufreq_generic_suspend);
1791
1792/**
1793 * cpufreq_suspend() - Suspend CPUFreq governors
1794 *
1795 * Called during system wide Suspend/Hibernate cycles for suspending governors
1796 * as some platforms can't change frequency after this point in suspend cycle.
1797 * Because some of the devices (like: i2c, regulators, etc) they use for
1798 * changing frequency are suspended quickly after this point.
1799 */
1800void cpufreq_suspend(void)
1801{
1802 struct cpufreq_policy *policy;
1803
1804 if (!cpufreq_driver)
1805 return;
1806
1807 if (!has_target() && !cpufreq_driver->suspend)
1808 goto suspend;
1809
1810 pr_debug("%s: Suspending Governors\n", __func__);
1811
1812 for_each_active_policy(policy) {
1813 if (has_target()) {
1814 down_write(&policy->rwsem);
1815 cpufreq_stop_governor(policy);
1816 up_write(&policy->rwsem);
1817 }
1818
1819 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1820 pr_err("%s: Failed to suspend driver: %s\n", __func__,
1821 cpufreq_driver->name);
1822 }
1823
1824suspend:
1825 cpufreq_suspended = true;
1826}
1827
1828/**
1829 * cpufreq_resume() - Resume CPUFreq governors
1830 *
1831 * Called during system wide Suspend/Hibernate cycle for resuming governors that
1832 * are suspended with cpufreq_suspend().
1833 */
1834void cpufreq_resume(void)
1835{
1836 struct cpufreq_policy *policy;
1837 int ret;
1838
1839 if (!cpufreq_driver)
1840 return;
1841
1842 if (unlikely(!cpufreq_suspended))
1843 return;
1844
1845 cpufreq_suspended = false;
1846
1847 if (!has_target() && !cpufreq_driver->resume)
1848 return;
1849
1850 pr_debug("%s: Resuming Governors\n", __func__);
1851
1852 for_each_active_policy(policy) {
1853 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1854 pr_err("%s: Failed to resume driver: %p\n", __func__,
1855 policy);
1856 } else if (has_target()) {
1857 down_write(&policy->rwsem);
1858 ret = cpufreq_start_governor(policy);
1859 up_write(&policy->rwsem);
1860
1861 if (ret)
1862 pr_err("%s: Failed to start governor for policy: %p\n",
1863 __func__, policy);
1864 }
1865 }
1866}
1867
1868/**
1869 * cpufreq_get_current_driver - return current driver's name
1870 *
1871 * Return the name string of the currently loaded cpufreq driver
1872 * or NULL, if none.
1873 */
1874const char *cpufreq_get_current_driver(void)
1875{
1876 if (cpufreq_driver)
1877 return cpufreq_driver->name;
1878
1879 return NULL;
1880}
1881EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1882
1883/**
1884 * cpufreq_get_driver_data - return current driver data
1885 *
1886 * Return the private data of the currently loaded cpufreq
1887 * driver, or NULL if no cpufreq driver is loaded.
1888 */
1889void *cpufreq_get_driver_data(void)
1890{
1891 if (cpufreq_driver)
1892 return cpufreq_driver->driver_data;
1893
1894 return NULL;
1895}
1896EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1897
1898/*********************************************************************
1899 * NOTIFIER LISTS INTERFACE *
1900 *********************************************************************/
1901
1902/**
1903 * cpufreq_register_notifier - register a driver with cpufreq
1904 * @nb: notifier function to register
1905 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1906 *
1907 * Add a driver to one of two lists: either a list of drivers that
1908 * are notified about clock rate changes (once before and once after
1909 * the transition), or a list of drivers that are notified about
1910 * changes in cpufreq policy.
1911 *
1912 * This function may sleep, and has the same return conditions as
1913 * blocking_notifier_chain_register.
1914 */
1915int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1916{
1917 int ret;
1918
1919 if (cpufreq_disabled())
1920 return -EINVAL;
1921
1922 switch (list) {
1923 case CPUFREQ_TRANSITION_NOTIFIER:
1924 mutex_lock(&cpufreq_fast_switch_lock);
1925
1926 if (cpufreq_fast_switch_count > 0) {
1927 mutex_unlock(&cpufreq_fast_switch_lock);
1928 return -EBUSY;
1929 }
1930 ret = srcu_notifier_chain_register(
1931 &cpufreq_transition_notifier_list, nb);
1932 if (!ret)
1933 cpufreq_fast_switch_count--;
1934
1935 mutex_unlock(&cpufreq_fast_switch_lock);
1936 break;
1937 case CPUFREQ_POLICY_NOTIFIER:
1938 ret = blocking_notifier_chain_register(
1939 &cpufreq_policy_notifier_list, nb);
1940 break;
1941 default:
1942 ret = -EINVAL;
1943 }
1944
1945 return ret;
1946}
1947EXPORT_SYMBOL(cpufreq_register_notifier);
1948
1949/**
1950 * cpufreq_unregister_notifier - unregister a driver with cpufreq
1951 * @nb: notifier block to be unregistered
1952 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1953 *
1954 * Remove a driver from the CPU frequency notifier list.
1955 *
1956 * This function may sleep, and has the same return conditions as
1957 * blocking_notifier_chain_unregister.
1958 */
1959int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1960{
1961 int ret;
1962
1963 if (cpufreq_disabled())
1964 return -EINVAL;
1965
1966 switch (list) {
1967 case CPUFREQ_TRANSITION_NOTIFIER:
1968 mutex_lock(&cpufreq_fast_switch_lock);
1969
1970 ret = srcu_notifier_chain_unregister(
1971 &cpufreq_transition_notifier_list, nb);
1972 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1973 cpufreq_fast_switch_count++;
1974
1975 mutex_unlock(&cpufreq_fast_switch_lock);
1976 break;
1977 case CPUFREQ_POLICY_NOTIFIER:
1978 ret = blocking_notifier_chain_unregister(
1979 &cpufreq_policy_notifier_list, nb);
1980 break;
1981 default:
1982 ret = -EINVAL;
1983 }
1984
1985 return ret;
1986}
1987EXPORT_SYMBOL(cpufreq_unregister_notifier);
1988
1989
1990/*********************************************************************
1991 * GOVERNORS *
1992 *********************************************************************/
1993
1994/**
1995 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1996 * @policy: cpufreq policy to switch the frequency for.
1997 * @target_freq: New frequency to set (may be approximate).
1998 *
1999 * Carry out a fast frequency switch without sleeping.
2000 *
2001 * The driver's ->fast_switch() callback invoked by this function must be
2002 * suitable for being called from within RCU-sched read-side critical sections
2003 * and it is expected to select the minimum available frequency greater than or
2004 * equal to @target_freq (CPUFREQ_RELATION_L).
2005 *
2006 * This function must not be called if policy->fast_switch_enabled is unset.
2007 *
2008 * Governors calling this function must guarantee that it will never be invoked
2009 * twice in parallel for the same policy and that it will never be called in
2010 * parallel with either ->target() or ->target_index() for the same policy.
2011 *
2012 * Returns the actual frequency set for the CPU.
2013 *
2014 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2015 * error condition, the hardware configuration must be preserved.
2016 */
2017unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2018 unsigned int target_freq)
2019{
2020 target_freq = clamp_val(target_freq, policy->min, policy->max);
2021
2022 return cpufreq_driver->fast_switch(policy, target_freq);
2023}
2024EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2025
2026/* Must set freqs->new to intermediate frequency */
2027static int __target_intermediate(struct cpufreq_policy *policy,
2028 struct cpufreq_freqs *freqs, int index)
2029{
2030 int ret;
2031
2032 freqs->new = cpufreq_driver->get_intermediate(policy, index);
2033
2034 /* We don't need to switch to intermediate freq */
2035 if (!freqs->new)
2036 return 0;
2037
2038 pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2039 __func__, policy->cpu, freqs->old, freqs->new);
2040
2041 cpufreq_freq_transition_begin(policy, freqs);
2042 ret = cpufreq_driver->target_intermediate(policy, index);
2043 cpufreq_freq_transition_end(policy, freqs, ret);
2044
2045 if (ret)
2046 pr_err("%s: Failed to change to intermediate frequency: %d\n",
2047 __func__, ret);
2048
2049 return ret;
2050}
2051
2052static int __target_index(struct cpufreq_policy *policy, int index)
2053{
2054 struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2055 unsigned int intermediate_freq = 0;
2056 unsigned int newfreq = policy->freq_table[index].frequency;
2057 int retval = -EINVAL;
2058 bool notify;
2059
2060 if (newfreq == policy->cur)
2061 return 0;
2062
2063 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2064 if (notify) {
2065 /* Handle switching to intermediate frequency */
2066 if (cpufreq_driver->get_intermediate) {
2067 retval = __target_intermediate(policy, &freqs, index);
2068 if (retval)
2069 return retval;
2070
2071 intermediate_freq = freqs.new;
2072 /* Set old freq to intermediate */
2073 if (intermediate_freq)
2074 freqs.old = freqs.new;
2075 }
2076
2077 freqs.new = newfreq;
2078 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2079 __func__, policy->cpu, freqs.old, freqs.new);
2080
2081 cpufreq_freq_transition_begin(policy, &freqs);
2082 }
2083
2084 retval = cpufreq_driver->target_index(policy, index);
2085 if (retval)
2086 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2087 retval);
2088
2089 if (notify) {
2090 cpufreq_freq_transition_end(policy, &freqs, retval);
2091
2092 /*
2093 * Failed after setting to intermediate freq? Driver should have
2094 * reverted back to initial frequency and so should we. Check
2095 * here for intermediate_freq instead of get_intermediate, in
2096 * case we haven't switched to intermediate freq at all.
2097 */
2098 if (unlikely(retval && intermediate_freq)) {
2099 freqs.old = intermediate_freq;
2100 freqs.new = policy->restore_freq;
2101 cpufreq_freq_transition_begin(policy, &freqs);
2102 cpufreq_freq_transition_end(policy, &freqs, 0);
2103 }
2104 }
2105
2106 return retval;
2107}
2108
2109int __cpufreq_driver_target(struct cpufreq_policy *policy,
2110 unsigned int target_freq,
2111 unsigned int relation)
2112{
2113 unsigned int old_target_freq = target_freq;
2114 int index;
2115
2116 if (cpufreq_disabled())
2117 return -ENODEV;
2118
2119 /* Make sure that target_freq is within supported range */
2120 target_freq = clamp_val(target_freq, policy->min, policy->max);
2121
2122 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2123 policy->cpu, target_freq, relation, old_target_freq);
2124
2125 /*
2126 * This might look like a redundant call as we are checking it again
2127 * after finding index. But it is left intentionally for cases where
2128 * exactly same freq is called again and so we can save on few function
2129 * calls.
2130 */
2131 if (target_freq == policy->cur)
2132 return 0;
2133
2134 /* Save last value to restore later on errors */
2135 policy->restore_freq = policy->cur;
2136
2137 if (cpufreq_driver->target)
2138 return cpufreq_driver->target(policy, target_freq, relation);
2139
2140 if (!cpufreq_driver->target_index)
2141 return -EINVAL;
2142
2143 index = cpufreq_frequency_table_target(policy, target_freq, relation);
2144
2145 return __target_index(policy, index);
2146}
2147EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2148
2149int cpufreq_driver_target(struct cpufreq_policy *policy,
2150 unsigned int target_freq,
2151 unsigned int relation)
2152{
2153 int ret;
2154
2155 down_write(&policy->rwsem);
2156
2157 ret = __cpufreq_driver_target(policy, target_freq, relation);
2158
2159 up_write(&policy->rwsem);
2160
2161 return ret;
2162}
2163EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2164
2165__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2166{
2167 return NULL;
2168}
2169
2170static int cpufreq_init_governor(struct cpufreq_policy *policy)
2171{
2172 int ret;
2173
2174 /* Don't start any governor operations if we are entering suspend */
2175 if (cpufreq_suspended)
2176 return 0;
2177 /*
2178 * Governor might not be initiated here if ACPI _PPC changed
2179 * notification happened, so check it.
2180 */
2181 if (!policy->governor)
2182 return -EINVAL;
2183
2184 /* Platform doesn't want dynamic frequency switching ? */
2185 if (policy->governor->dynamic_switching &&
2186 cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2187 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2188
2189 if (gov) {
2190 pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2191 policy->governor->name, gov->name);
2192 policy->governor = gov;
2193 } else {
2194 return -EINVAL;
2195 }
2196 }
2197
2198 if (!try_module_get(policy->governor->owner))
2199 return -EINVAL;
2200
2201 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2202
2203 if (policy->governor->init) {
2204 ret = policy->governor->init(policy);
2205 if (ret) {
2206 module_put(policy->governor->owner);
2207 return ret;
2208 }
2209 }
2210
2211 return 0;
2212}
2213
2214static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2215{
2216 if (cpufreq_suspended || !policy->governor)
2217 return;
2218
2219 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2220
2221 if (policy->governor->exit)
2222 policy->governor->exit(policy);
2223
2224 module_put(policy->governor->owner);
2225}
2226
2227static int cpufreq_start_governor(struct cpufreq_policy *policy)
2228{
2229 int ret;
2230
2231 if (cpufreq_suspended)
2232 return 0;
2233
2234 if (!policy->governor)
2235 return -EINVAL;
2236
2237 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2238
2239 if (cpufreq_driver->get)
2240 cpufreq_verify_current_freq(policy, false);
2241
2242 if (policy->governor->start) {
2243 ret = policy->governor->start(policy);
2244 if (ret)
2245 return ret;
2246 }
2247
2248 if (policy->governor->limits)
2249 policy->governor->limits(policy);
2250
2251 return 0;
2252}
2253
2254static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2255{
2256 if (cpufreq_suspended || !policy->governor)
2257 return;
2258
2259 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2260
2261 if (policy->governor->stop)
2262 policy->governor->stop(policy);
2263}
2264
2265static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2266{
2267 if (cpufreq_suspended || !policy->governor)
2268 return;
2269
2270 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2271
2272 if (policy->governor->limits)
2273 policy->governor->limits(policy);
2274}
2275
2276int cpufreq_register_governor(struct cpufreq_governor *governor)
2277{
2278 int err;
2279
2280 if (!governor)
2281 return -EINVAL;
2282
2283 if (cpufreq_disabled())
2284 return -ENODEV;
2285
2286 mutex_lock(&cpufreq_governor_mutex);
2287
2288 err = -EBUSY;
2289 if (!find_governor(governor->name)) {
2290 err = 0;
2291 list_add(&governor->governor_list, &cpufreq_governor_list);
2292 }
2293
2294 mutex_unlock(&cpufreq_governor_mutex);
2295 return err;
2296}
2297EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2298
2299void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2300{
2301 struct cpufreq_policy *policy;
2302 unsigned long flags;
2303
2304 if (!governor)
2305 return;
2306
2307 if (cpufreq_disabled())
2308 return;
2309
2310 /* clear last_governor for all inactive policies */
2311 read_lock_irqsave(&cpufreq_driver_lock, flags);
2312 for_each_inactive_policy(policy) {
2313 if (!strcmp(policy->last_governor, governor->name)) {
2314 policy->governor = NULL;
2315 strcpy(policy->last_governor, "\0");
2316 }
2317 }
2318 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2319
2320 mutex_lock(&cpufreq_governor_mutex);
2321 list_del(&governor->governor_list);
2322 mutex_unlock(&cpufreq_governor_mutex);
2323}
2324EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2325
2326
2327/*********************************************************************
2328 * POLICY INTERFACE *
2329 *********************************************************************/
2330
2331/**
2332 * cpufreq_get_policy - get the current cpufreq_policy
2333 * @policy: struct cpufreq_policy into which the current cpufreq_policy
2334 * is written
2335 *
2336 * Reads the current cpufreq policy.
2337 */
2338int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2339{
2340 struct cpufreq_policy *cpu_policy;
2341 if (!policy)
2342 return -EINVAL;
2343
2344 cpu_policy = cpufreq_cpu_get(cpu);
2345 if (!cpu_policy)
2346 return -EINVAL;
2347
2348 memcpy(policy, cpu_policy, sizeof(*policy));
2349
2350 cpufreq_cpu_put(cpu_policy);
2351 return 0;
2352}
2353EXPORT_SYMBOL(cpufreq_get_policy);
2354
2355/**
2356 * cpufreq_set_policy - Modify cpufreq policy parameters.
2357 * @policy: Policy object to modify.
2358 * @new_policy: New policy data.
2359 *
2360 * Pass @new_policy to the cpufreq driver's ->verify() callback. Next, copy the
2361 * min and max parameters of @new_policy to @policy and either invoke the
2362 * driver's ->setpolicy() callback (if present) or carry out a governor update
2363 * for @policy. That is, run the current governor's ->limits() callback (if the
2364 * governor field in @new_policy points to the same object as the one in
2365 * @policy) or replace the governor for @policy with the new one stored in
2366 * @new_policy.
2367 *
2368 * The cpuinfo part of @policy is not updated by this function.
2369 */
2370int cpufreq_set_policy(struct cpufreq_policy *policy,
2371 struct cpufreq_policy *new_policy)
2372{
2373 struct cpufreq_governor *old_gov;
2374 int ret;
2375
2376 pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2377 new_policy->cpu, new_policy->min, new_policy->max);
2378
2379 memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2380
2381 /*
2382 * PM QoS framework collects all the requests from users and provide us
2383 * the final aggregated value here.
2384 */
2385 new_policy->min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2386 new_policy->max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2387
2388 /* verify the cpu speed can be set within this limit */
2389 ret = cpufreq_driver->verify(new_policy);
2390 if (ret)
2391 return ret;
2392
2393 policy->min = new_policy->min;
2394 policy->max = new_policy->max;
2395 trace_cpu_frequency_limits(policy);
2396
2397 policy->cached_target_freq = UINT_MAX;
2398
2399 pr_debug("new min and max freqs are %u - %u kHz\n",
2400 policy->min, policy->max);
2401
2402 if (cpufreq_driver->setpolicy) {
2403 policy->policy = new_policy->policy;
2404 pr_debug("setting range\n");
2405 return cpufreq_driver->setpolicy(policy);
2406 }
2407
2408 if (new_policy->governor == policy->governor) {
2409 pr_debug("governor limits update\n");
2410 cpufreq_governor_limits(policy);
2411 return 0;
2412 }
2413
2414 pr_debug("governor switch\n");
2415
2416 /* save old, working values */
2417 old_gov = policy->governor;
2418 /* end old governor */
2419 if (old_gov) {
2420 cpufreq_stop_governor(policy);
2421 cpufreq_exit_governor(policy);
2422 }
2423
2424 /* start new governor */
2425 policy->governor = new_policy->governor;
2426 ret = cpufreq_init_governor(policy);
2427 if (!ret) {
2428 ret = cpufreq_start_governor(policy);
2429 if (!ret) {
2430 pr_debug("governor change\n");
2431 sched_cpufreq_governor_change(policy, old_gov);
2432 return 0;
2433 }
2434 cpufreq_exit_governor(policy);
2435 }
2436
2437 /* new governor failed, so re-start old one */
2438 pr_debug("starting governor %s failed\n", policy->governor->name);
2439 if (old_gov) {
2440 policy->governor = old_gov;
2441 if (cpufreq_init_governor(policy))
2442 policy->governor = NULL;
2443 else
2444 cpufreq_start_governor(policy);
2445 }
2446
2447 return ret;
2448}
2449
2450/**
2451 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2452 * @cpu: CPU to re-evaluate the policy for.
2453 *
2454 * Update the current frequency for the cpufreq policy of @cpu and use
2455 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2456 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2457 * for the policy in question, among other things.
2458 */
2459void cpufreq_update_policy(unsigned int cpu)
2460{
2461 struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2462
2463 if (!policy)
2464 return;
2465
2466 /*
2467 * BIOS might change freq behind our back
2468 * -> ask driver for current freq and notify governors about a change
2469 */
2470 if (cpufreq_driver->get && has_target() &&
2471 (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2472 goto unlock;
2473
2474 refresh_frequency_limits(policy);
2475
2476unlock:
2477 cpufreq_cpu_release(policy);
2478}
2479EXPORT_SYMBOL(cpufreq_update_policy);
2480
2481/**
2482 * cpufreq_update_limits - Update policy limits for a given CPU.
2483 * @cpu: CPU to update the policy limits for.
2484 *
2485 * Invoke the driver's ->update_limits callback if present or call
2486 * cpufreq_update_policy() for @cpu.
2487 */
2488void cpufreq_update_limits(unsigned int cpu)
2489{
2490 if (cpufreq_driver->update_limits)
2491 cpufreq_driver->update_limits(cpu);
2492 else
2493 cpufreq_update_policy(cpu);
2494}
2495EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2496
2497/*********************************************************************
2498 * BOOST *
2499 *********************************************************************/
2500static int cpufreq_boost_set_sw(int state)
2501{
2502 struct cpufreq_policy *policy;
2503 int ret = -EINVAL;
2504
2505 for_each_active_policy(policy) {
2506 if (!policy->freq_table)
2507 continue;
2508
2509 ret = cpufreq_frequency_table_cpuinfo(policy,
2510 policy->freq_table);
2511 if (ret) {
2512 pr_err("%s: Policy frequency update failed\n",
2513 __func__);
2514 break;
2515 }
2516
2517 ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2518 if (ret < 0)
2519 break;
2520 }
2521
2522 return ret;
2523}
2524
2525int cpufreq_boost_trigger_state(int state)
2526{
2527 unsigned long flags;
2528 int ret = 0;
2529
2530 if (cpufreq_driver->boost_enabled == state)
2531 return 0;
2532
2533 write_lock_irqsave(&cpufreq_driver_lock, flags);
2534 cpufreq_driver->boost_enabled = state;
2535 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2536
2537 ret = cpufreq_driver->set_boost(state);
2538 if (ret) {
2539 write_lock_irqsave(&cpufreq_driver_lock, flags);
2540 cpufreq_driver->boost_enabled = !state;
2541 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2542
2543 pr_err("%s: Cannot %s BOOST\n",
2544 __func__, state ? "enable" : "disable");
2545 }
2546
2547 return ret;
2548}
2549
2550static bool cpufreq_boost_supported(void)
2551{
2552 return cpufreq_driver->set_boost;
2553}
2554
2555static int create_boost_sysfs_file(void)
2556{
2557 int ret;
2558
2559 ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2560 if (ret)
2561 pr_err("%s: cannot register global BOOST sysfs file\n",
2562 __func__);
2563
2564 return ret;
2565}
2566
2567static void remove_boost_sysfs_file(void)
2568{
2569 if (cpufreq_boost_supported())
2570 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2571}
2572
2573int cpufreq_enable_boost_support(void)
2574{
2575 if (!cpufreq_driver)
2576 return -EINVAL;
2577
2578 if (cpufreq_boost_supported())
2579 return 0;
2580
2581 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2582
2583 /* This will get removed on driver unregister */
2584 return create_boost_sysfs_file();
2585}
2586EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2587
2588int cpufreq_boost_enabled(void)
2589{
2590 return cpufreq_driver->boost_enabled;
2591}
2592EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2593
2594/*********************************************************************
2595 * REGISTER / UNREGISTER CPUFREQ DRIVER *
2596 *********************************************************************/
2597static enum cpuhp_state hp_online;
2598
2599static int cpuhp_cpufreq_online(unsigned int cpu)
2600{
2601 cpufreq_online(cpu);
2602
2603 return 0;
2604}
2605
2606static int cpuhp_cpufreq_offline(unsigned int cpu)
2607{
2608 cpufreq_offline(cpu);
2609
2610 return 0;
2611}
2612
2613/**
2614 * cpufreq_register_driver - register a CPU Frequency driver
2615 * @driver_data: A struct cpufreq_driver containing the values#
2616 * submitted by the CPU Frequency driver.
2617 *
2618 * Registers a CPU Frequency driver to this core code. This code
2619 * returns zero on success, -EEXIST when another driver got here first
2620 * (and isn't unregistered in the meantime).
2621 *
2622 */
2623int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2624{
2625 unsigned long flags;
2626 int ret;
2627
2628 if (cpufreq_disabled())
2629 return -ENODEV;
2630
2631 if (!driver_data || !driver_data->verify || !driver_data->init ||
2632 !(driver_data->setpolicy || driver_data->target_index ||
2633 driver_data->target) ||
2634 (driver_data->setpolicy && (driver_data->target_index ||
2635 driver_data->target)) ||
2636 (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2637 (!driver_data->online != !driver_data->offline))
2638 return -EINVAL;
2639
2640 pr_debug("trying to register driver %s\n", driver_data->name);
2641
2642 /* Protect against concurrent CPU online/offline. */
2643 cpus_read_lock();
2644
2645 write_lock_irqsave(&cpufreq_driver_lock, flags);
2646 if (cpufreq_driver) {
2647 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2648 ret = -EEXIST;
2649 goto out;
2650 }
2651 cpufreq_driver = driver_data;
2652 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2653
2654 if (driver_data->setpolicy)
2655 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2656
2657 if (cpufreq_boost_supported()) {
2658 ret = create_boost_sysfs_file();
2659 if (ret)
2660 goto err_null_driver;
2661 }
2662
2663 ret = subsys_interface_register(&cpufreq_interface);
2664 if (ret)
2665 goto err_boost_unreg;
2666
2667 if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2668 list_empty(&cpufreq_policy_list)) {
2669 /* if all ->init() calls failed, unregister */
2670 ret = -ENODEV;
2671 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2672 driver_data->name);
2673 goto err_if_unreg;
2674 }
2675
2676 ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2677 "cpufreq:online",
2678 cpuhp_cpufreq_online,
2679 cpuhp_cpufreq_offline);
2680 if (ret < 0)
2681 goto err_if_unreg;
2682 hp_online = ret;
2683 ret = 0;
2684
2685 pr_debug("driver %s up and running\n", driver_data->name);
2686 goto out;
2687
2688err_if_unreg:
2689 subsys_interface_unregister(&cpufreq_interface);
2690err_boost_unreg:
2691 remove_boost_sysfs_file();
2692err_null_driver:
2693 write_lock_irqsave(&cpufreq_driver_lock, flags);
2694 cpufreq_driver = NULL;
2695 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2696out:
2697 cpus_read_unlock();
2698 return ret;
2699}
2700EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2701
2702/**
2703 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2704 *
2705 * Unregister the current CPUFreq driver. Only call this if you have
2706 * the right to do so, i.e. if you have succeeded in initialising before!
2707 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2708 * currently not initialised.
2709 */
2710int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2711{
2712 unsigned long flags;
2713
2714 if (!cpufreq_driver || (driver != cpufreq_driver))
2715 return -EINVAL;
2716
2717 pr_debug("unregistering driver %s\n", driver->name);
2718
2719 /* Protect against concurrent cpu hotplug */
2720 cpus_read_lock();
2721 subsys_interface_unregister(&cpufreq_interface);
2722 remove_boost_sysfs_file();
2723 cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2724
2725 write_lock_irqsave(&cpufreq_driver_lock, flags);
2726
2727 cpufreq_driver = NULL;
2728
2729 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2730 cpus_read_unlock();
2731
2732 return 0;
2733}
2734EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2735
2736struct kobject *cpufreq_global_kobject;
2737EXPORT_SYMBOL(cpufreq_global_kobject);
2738
2739static int __init cpufreq_core_init(void)
2740{
2741 if (cpufreq_disabled())
2742 return -ENODEV;
2743
2744 cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2745 BUG_ON(!cpufreq_global_kobject);
2746
2747 return 0;
2748}
2749module_param(off, int, 0444);
2750core_initcall(cpufreq_core_init);