Loading...
1/*
2 * linux/drivers/cpufreq/cpufreq.c
3 *
4 * Copyright (C) 2001 Russell King
5 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6 *
7 * Oct 2005 - Ashok Raj <ashok.raj@intel.com>
8 * Added handling for CPU hotplug
9 * Feb 2006 - Jacob Shin <jacob.shin@amd.com>
10 * Fix handling for CPU hotplug -- affected CPUs
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 *
16 */
17
18#include <linux/kernel.h>
19#include <linux/module.h>
20#include <linux/init.h>
21#include <linux/notifier.h>
22#include <linux/cpufreq.h>
23#include <linux/delay.h>
24#include <linux/interrupt.h>
25#include <linux/spinlock.h>
26#include <linux/device.h>
27#include <linux/slab.h>
28#include <linux/cpu.h>
29#include <linux/completion.h>
30#include <linux/mutex.h>
31#include <linux/syscore_ops.h>
32
33#include <trace/events/power.h>
34
35/**
36 * The "cpufreq driver" - the arch- or hardware-dependent low
37 * level driver of CPUFreq support, and its spinlock. This lock
38 * also protects the cpufreq_cpu_data array.
39 */
40static struct cpufreq_driver *cpufreq_driver;
41static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
42#ifdef CONFIG_HOTPLUG_CPU
43/* This one keeps track of the previously set governor of a removed CPU */
44static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
45#endif
46static DEFINE_SPINLOCK(cpufreq_driver_lock);
47
48/*
49 * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
50 * all cpufreq/hotplug/workqueue/etc related lock issues.
51 *
52 * The rules for this semaphore:
53 * - Any routine that wants to read from the policy structure will
54 * do a down_read on this semaphore.
55 * - Any routine that will write to the policy structure and/or may take away
56 * the policy altogether (eg. CPU hotplug), will hold this lock in write
57 * mode before doing so.
58 *
59 * Additional rules:
60 * - All holders of the lock should check to make sure that the CPU they
61 * are concerned with are online after they get the lock.
62 * - Governor routines that can be called in cpufreq hotplug path should not
63 * take this sem as top level hotplug notifier handler takes this.
64 * - Lock should not be held across
65 * __cpufreq_governor(data, CPUFREQ_GOV_STOP);
66 */
67static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
68static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
69
70#define lock_policy_rwsem(mode, cpu) \
71static int lock_policy_rwsem_##mode \
72(int cpu) \
73{ \
74 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
75 BUG_ON(policy_cpu == -1); \
76 down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
77 if (unlikely(!cpu_online(cpu))) { \
78 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
79 return -1; \
80 } \
81 \
82 return 0; \
83}
84
85lock_policy_rwsem(read, cpu);
86
87lock_policy_rwsem(write, cpu);
88
89static void unlock_policy_rwsem_read(int cpu)
90{
91 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
92 BUG_ON(policy_cpu == -1);
93 up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
94}
95
96static void unlock_policy_rwsem_write(int cpu)
97{
98 int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
99 BUG_ON(policy_cpu == -1);
100 up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
101}
102
103
104/* internal prototypes */
105static int __cpufreq_governor(struct cpufreq_policy *policy,
106 unsigned int event);
107static unsigned int __cpufreq_get(unsigned int cpu);
108static void handle_update(struct work_struct *work);
109
110/**
111 * Two notifier lists: the "policy" list is involved in the
112 * validation process for a new CPU frequency policy; the
113 * "transition" list for kernel code that needs to handle
114 * changes to devices when the CPU clock speed changes.
115 * The mutex locks both lists.
116 */
117static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
118static struct srcu_notifier_head cpufreq_transition_notifier_list;
119
120static bool init_cpufreq_transition_notifier_list_called;
121static int __init init_cpufreq_transition_notifier_list(void)
122{
123 srcu_init_notifier_head(&cpufreq_transition_notifier_list);
124 init_cpufreq_transition_notifier_list_called = true;
125 return 0;
126}
127pure_initcall(init_cpufreq_transition_notifier_list);
128
129static LIST_HEAD(cpufreq_governor_list);
130static DEFINE_MUTEX(cpufreq_governor_mutex);
131
132struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
133{
134 struct cpufreq_policy *data;
135 unsigned long flags;
136
137 if (cpu >= nr_cpu_ids)
138 goto err_out;
139
140 /* get the cpufreq driver */
141 spin_lock_irqsave(&cpufreq_driver_lock, flags);
142
143 if (!cpufreq_driver)
144 goto err_out_unlock;
145
146 if (!try_module_get(cpufreq_driver->owner))
147 goto err_out_unlock;
148
149
150 /* get the CPU */
151 data = per_cpu(cpufreq_cpu_data, cpu);
152
153 if (!data)
154 goto err_out_put_module;
155
156 if (!kobject_get(&data->kobj))
157 goto err_out_put_module;
158
159 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
160 return data;
161
162err_out_put_module:
163 module_put(cpufreq_driver->owner);
164err_out_unlock:
165 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
166err_out:
167 return NULL;
168}
169EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
170
171
172void cpufreq_cpu_put(struct cpufreq_policy *data)
173{
174 kobject_put(&data->kobj);
175 module_put(cpufreq_driver->owner);
176}
177EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
178
179
180/*********************************************************************
181 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
182 *********************************************************************/
183
184/**
185 * adjust_jiffies - adjust the system "loops_per_jiffy"
186 *
187 * This function alters the system "loops_per_jiffy" for the clock
188 * speed change. Note that loops_per_jiffy cannot be updated on SMP
189 * systems as each CPU might be scaled differently. So, use the arch
190 * per-CPU loops_per_jiffy value wherever possible.
191 */
192#ifndef CONFIG_SMP
193static unsigned long l_p_j_ref;
194static unsigned int l_p_j_ref_freq;
195
196static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
197{
198 if (ci->flags & CPUFREQ_CONST_LOOPS)
199 return;
200
201 if (!l_p_j_ref_freq) {
202 l_p_j_ref = loops_per_jiffy;
203 l_p_j_ref_freq = ci->old;
204 pr_debug("saving %lu as reference value for loops_per_jiffy; "
205 "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
206 }
207 if ((val == CPUFREQ_PRECHANGE && ci->old < ci->new) ||
208 (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
209 (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
210 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
211 ci->new);
212 pr_debug("scaling loops_per_jiffy to %lu "
213 "for frequency %u kHz\n", loops_per_jiffy, ci->new);
214 }
215}
216#else
217static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
218{
219 return;
220}
221#endif
222
223
224/**
225 * cpufreq_notify_transition - call notifier chain and adjust_jiffies
226 * on frequency transition.
227 *
228 * This function calls the transition notifiers and the "adjust_jiffies"
229 * function. It is called twice on all CPU frequency changes that have
230 * external effects.
231 */
232void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
233{
234 struct cpufreq_policy *policy;
235
236 BUG_ON(irqs_disabled());
237
238 freqs->flags = cpufreq_driver->flags;
239 pr_debug("notification %u of frequency transition to %u kHz\n",
240 state, freqs->new);
241
242 policy = per_cpu(cpufreq_cpu_data, freqs->cpu);
243 switch (state) {
244
245 case CPUFREQ_PRECHANGE:
246 /* detect if the driver reported a value as "old frequency"
247 * which is not equal to what the cpufreq core thinks is
248 * "old frequency".
249 */
250 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
251 if ((policy) && (policy->cpu == freqs->cpu) &&
252 (policy->cur) && (policy->cur != freqs->old)) {
253 pr_debug("Warning: CPU frequency is"
254 " %u, cpufreq assumed %u kHz.\n",
255 freqs->old, policy->cur);
256 freqs->old = policy->cur;
257 }
258 }
259 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
260 CPUFREQ_PRECHANGE, freqs);
261 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
262 break;
263
264 case CPUFREQ_POSTCHANGE:
265 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
266 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
267 (unsigned long)freqs->cpu);
268 trace_power_frequency(POWER_PSTATE, freqs->new, freqs->cpu);
269 trace_cpu_frequency(freqs->new, freqs->cpu);
270 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
271 CPUFREQ_POSTCHANGE, freqs);
272 if (likely(policy) && likely(policy->cpu == freqs->cpu))
273 policy->cur = freqs->new;
274 break;
275 }
276}
277EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
278
279
280
281/*********************************************************************
282 * SYSFS INTERFACE *
283 *********************************************************************/
284
285static struct cpufreq_governor *__find_governor(const char *str_governor)
286{
287 struct cpufreq_governor *t;
288
289 list_for_each_entry(t, &cpufreq_governor_list, governor_list)
290 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
291 return t;
292
293 return NULL;
294}
295
296/**
297 * cpufreq_parse_governor - parse a governor string
298 */
299static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
300 struct cpufreq_governor **governor)
301{
302 int err = -EINVAL;
303
304 if (!cpufreq_driver)
305 goto out;
306
307 if (cpufreq_driver->setpolicy) {
308 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
309 *policy = CPUFREQ_POLICY_PERFORMANCE;
310 err = 0;
311 } else if (!strnicmp(str_governor, "powersave",
312 CPUFREQ_NAME_LEN)) {
313 *policy = CPUFREQ_POLICY_POWERSAVE;
314 err = 0;
315 }
316 } else if (cpufreq_driver->target) {
317 struct cpufreq_governor *t;
318
319 mutex_lock(&cpufreq_governor_mutex);
320
321 t = __find_governor(str_governor);
322
323 if (t == NULL) {
324 int ret;
325
326 mutex_unlock(&cpufreq_governor_mutex);
327 ret = request_module("cpufreq_%s", str_governor);
328 mutex_lock(&cpufreq_governor_mutex);
329
330 if (ret == 0)
331 t = __find_governor(str_governor);
332 }
333
334 if (t != NULL) {
335 *governor = t;
336 err = 0;
337 }
338
339 mutex_unlock(&cpufreq_governor_mutex);
340 }
341out:
342 return err;
343}
344
345
346/**
347 * cpufreq_per_cpu_attr_read() / show_##file_name() -
348 * print out cpufreq information
349 *
350 * Write out information from cpufreq_driver->policy[cpu]; object must be
351 * "unsigned int".
352 */
353
354#define show_one(file_name, object) \
355static ssize_t show_##file_name \
356(struct cpufreq_policy *policy, char *buf) \
357{ \
358 return sprintf(buf, "%u\n", policy->object); \
359}
360
361show_one(cpuinfo_min_freq, cpuinfo.min_freq);
362show_one(cpuinfo_max_freq, cpuinfo.max_freq);
363show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
364show_one(scaling_min_freq, min);
365show_one(scaling_max_freq, max);
366show_one(scaling_cur_freq, cur);
367
368static int __cpufreq_set_policy(struct cpufreq_policy *data,
369 struct cpufreq_policy *policy);
370
371/**
372 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
373 */
374#define store_one(file_name, object) \
375static ssize_t store_##file_name \
376(struct cpufreq_policy *policy, const char *buf, size_t count) \
377{ \
378 unsigned int ret = -EINVAL; \
379 struct cpufreq_policy new_policy; \
380 \
381 ret = cpufreq_get_policy(&new_policy, policy->cpu); \
382 if (ret) \
383 return -EINVAL; \
384 \
385 ret = sscanf(buf, "%u", &new_policy.object); \
386 if (ret != 1) \
387 return -EINVAL; \
388 \
389 ret = __cpufreq_set_policy(policy, &new_policy); \
390 policy->user_policy.object = policy->object; \
391 \
392 return ret ? ret : count; \
393}
394
395store_one(scaling_min_freq, min);
396store_one(scaling_max_freq, max);
397
398/**
399 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
400 */
401static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
402 char *buf)
403{
404 unsigned int cur_freq = __cpufreq_get(policy->cpu);
405 if (!cur_freq)
406 return sprintf(buf, "<unknown>");
407 return sprintf(buf, "%u\n", cur_freq);
408}
409
410
411/**
412 * show_scaling_governor - show the current policy for the specified CPU
413 */
414static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
415{
416 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
417 return sprintf(buf, "powersave\n");
418 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
419 return sprintf(buf, "performance\n");
420 else if (policy->governor)
421 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n",
422 policy->governor->name);
423 return -EINVAL;
424}
425
426
427/**
428 * store_scaling_governor - store policy for the specified CPU
429 */
430static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
431 const char *buf, size_t count)
432{
433 unsigned int ret = -EINVAL;
434 char str_governor[16];
435 struct cpufreq_policy new_policy;
436
437 ret = cpufreq_get_policy(&new_policy, policy->cpu);
438 if (ret)
439 return ret;
440
441 ret = sscanf(buf, "%15s", str_governor);
442 if (ret != 1)
443 return -EINVAL;
444
445 if (cpufreq_parse_governor(str_governor, &new_policy.policy,
446 &new_policy.governor))
447 return -EINVAL;
448
449 /* Do not use cpufreq_set_policy here or the user_policy.max
450 will be wrongly overridden */
451 ret = __cpufreq_set_policy(policy, &new_policy);
452
453 policy->user_policy.policy = policy->policy;
454 policy->user_policy.governor = policy->governor;
455
456 if (ret)
457 return ret;
458 else
459 return count;
460}
461
462/**
463 * show_scaling_driver - show the cpufreq driver currently loaded
464 */
465static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
466{
467 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
468}
469
470/**
471 * show_scaling_available_governors - show the available CPUfreq governors
472 */
473static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
474 char *buf)
475{
476 ssize_t i = 0;
477 struct cpufreq_governor *t;
478
479 if (!cpufreq_driver->target) {
480 i += sprintf(buf, "performance powersave");
481 goto out;
482 }
483
484 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
485 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
486 - (CPUFREQ_NAME_LEN + 2)))
487 goto out;
488 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
489 }
490out:
491 i += sprintf(&buf[i], "\n");
492 return i;
493}
494
495static ssize_t show_cpus(const struct cpumask *mask, char *buf)
496{
497 ssize_t i = 0;
498 unsigned int cpu;
499
500 for_each_cpu(cpu, mask) {
501 if (i)
502 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
503 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
504 if (i >= (PAGE_SIZE - 5))
505 break;
506 }
507 i += sprintf(&buf[i], "\n");
508 return i;
509}
510
511/**
512 * show_related_cpus - show the CPUs affected by each transition even if
513 * hw coordination is in use
514 */
515static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
516{
517 if (cpumask_empty(policy->related_cpus))
518 return show_cpus(policy->cpus, buf);
519 return show_cpus(policy->related_cpus, buf);
520}
521
522/**
523 * show_affected_cpus - show the CPUs affected by each transition
524 */
525static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
526{
527 return show_cpus(policy->cpus, buf);
528}
529
530static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
531 const char *buf, size_t count)
532{
533 unsigned int freq = 0;
534 unsigned int ret;
535
536 if (!policy->governor || !policy->governor->store_setspeed)
537 return -EINVAL;
538
539 ret = sscanf(buf, "%u", &freq);
540 if (ret != 1)
541 return -EINVAL;
542
543 policy->governor->store_setspeed(policy, freq);
544
545 return count;
546}
547
548static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
549{
550 if (!policy->governor || !policy->governor->show_setspeed)
551 return sprintf(buf, "<unsupported>\n");
552
553 return policy->governor->show_setspeed(policy, buf);
554}
555
556/**
557 * show_scaling_driver - show the current cpufreq HW/BIOS limitation
558 */
559static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
560{
561 unsigned int limit;
562 int ret;
563 if (cpufreq_driver->bios_limit) {
564 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
565 if (!ret)
566 return sprintf(buf, "%u\n", limit);
567 }
568 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
569}
570
571cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
572cpufreq_freq_attr_ro(cpuinfo_min_freq);
573cpufreq_freq_attr_ro(cpuinfo_max_freq);
574cpufreq_freq_attr_ro(cpuinfo_transition_latency);
575cpufreq_freq_attr_ro(scaling_available_governors);
576cpufreq_freq_attr_ro(scaling_driver);
577cpufreq_freq_attr_ro(scaling_cur_freq);
578cpufreq_freq_attr_ro(bios_limit);
579cpufreq_freq_attr_ro(related_cpus);
580cpufreq_freq_attr_ro(affected_cpus);
581cpufreq_freq_attr_rw(scaling_min_freq);
582cpufreq_freq_attr_rw(scaling_max_freq);
583cpufreq_freq_attr_rw(scaling_governor);
584cpufreq_freq_attr_rw(scaling_setspeed);
585
586static struct attribute *default_attrs[] = {
587 &cpuinfo_min_freq.attr,
588 &cpuinfo_max_freq.attr,
589 &cpuinfo_transition_latency.attr,
590 &scaling_min_freq.attr,
591 &scaling_max_freq.attr,
592 &affected_cpus.attr,
593 &related_cpus.attr,
594 &scaling_governor.attr,
595 &scaling_driver.attr,
596 &scaling_available_governors.attr,
597 &scaling_setspeed.attr,
598 NULL
599};
600
601struct kobject *cpufreq_global_kobject;
602EXPORT_SYMBOL(cpufreq_global_kobject);
603
604#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
605#define to_attr(a) container_of(a, struct freq_attr, attr)
606
607static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
608{
609 struct cpufreq_policy *policy = to_policy(kobj);
610 struct freq_attr *fattr = to_attr(attr);
611 ssize_t ret = -EINVAL;
612 policy = cpufreq_cpu_get(policy->cpu);
613 if (!policy)
614 goto no_policy;
615
616 if (lock_policy_rwsem_read(policy->cpu) < 0)
617 goto fail;
618
619 if (fattr->show)
620 ret = fattr->show(policy, buf);
621 else
622 ret = -EIO;
623
624 unlock_policy_rwsem_read(policy->cpu);
625fail:
626 cpufreq_cpu_put(policy);
627no_policy:
628 return ret;
629}
630
631static ssize_t store(struct kobject *kobj, struct attribute *attr,
632 const char *buf, size_t count)
633{
634 struct cpufreq_policy *policy = to_policy(kobj);
635 struct freq_attr *fattr = to_attr(attr);
636 ssize_t ret = -EINVAL;
637 policy = cpufreq_cpu_get(policy->cpu);
638 if (!policy)
639 goto no_policy;
640
641 if (lock_policy_rwsem_write(policy->cpu) < 0)
642 goto fail;
643
644 if (fattr->store)
645 ret = fattr->store(policy, buf, count);
646 else
647 ret = -EIO;
648
649 unlock_policy_rwsem_write(policy->cpu);
650fail:
651 cpufreq_cpu_put(policy);
652no_policy:
653 return ret;
654}
655
656static void cpufreq_sysfs_release(struct kobject *kobj)
657{
658 struct cpufreq_policy *policy = to_policy(kobj);
659 pr_debug("last reference is dropped\n");
660 complete(&policy->kobj_unregister);
661}
662
663static const struct sysfs_ops sysfs_ops = {
664 .show = show,
665 .store = store,
666};
667
668static struct kobj_type ktype_cpufreq = {
669 .sysfs_ops = &sysfs_ops,
670 .default_attrs = default_attrs,
671 .release = cpufreq_sysfs_release,
672};
673
674/*
675 * Returns:
676 * Negative: Failure
677 * 0: Success
678 * Positive: When we have a managed CPU and the sysfs got symlinked
679 */
680static int cpufreq_add_dev_policy(unsigned int cpu,
681 struct cpufreq_policy *policy,
682 struct sys_device *sys_dev)
683{
684 int ret = 0;
685#ifdef CONFIG_SMP
686 unsigned long flags;
687 unsigned int j;
688#ifdef CONFIG_HOTPLUG_CPU
689 struct cpufreq_governor *gov;
690
691 gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
692 if (gov) {
693 policy->governor = gov;
694 pr_debug("Restoring governor %s for cpu %d\n",
695 policy->governor->name, cpu);
696 }
697#endif
698
699 for_each_cpu(j, policy->cpus) {
700 struct cpufreq_policy *managed_policy;
701
702 if (cpu == j)
703 continue;
704
705 /* Check for existing affected CPUs.
706 * They may not be aware of it due to CPU Hotplug.
707 * cpufreq_cpu_put is called when the device is removed
708 * in __cpufreq_remove_dev()
709 */
710 managed_policy = cpufreq_cpu_get(j);
711 if (unlikely(managed_policy)) {
712
713 /* Set proper policy_cpu */
714 unlock_policy_rwsem_write(cpu);
715 per_cpu(cpufreq_policy_cpu, cpu) = managed_policy->cpu;
716
717 if (lock_policy_rwsem_write(cpu) < 0) {
718 /* Should not go through policy unlock path */
719 if (cpufreq_driver->exit)
720 cpufreq_driver->exit(policy);
721 cpufreq_cpu_put(managed_policy);
722 return -EBUSY;
723 }
724
725 spin_lock_irqsave(&cpufreq_driver_lock, flags);
726 cpumask_copy(managed_policy->cpus, policy->cpus);
727 per_cpu(cpufreq_cpu_data, cpu) = managed_policy;
728 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
729
730 pr_debug("CPU already managed, adding link\n");
731 ret = sysfs_create_link(&sys_dev->kobj,
732 &managed_policy->kobj,
733 "cpufreq");
734 if (ret)
735 cpufreq_cpu_put(managed_policy);
736 /*
737 * Success. We only needed to be added to the mask.
738 * Call driver->exit() because only the cpu parent of
739 * the kobj needed to call init().
740 */
741 if (cpufreq_driver->exit)
742 cpufreq_driver->exit(policy);
743
744 if (!ret)
745 return 1;
746 else
747 return ret;
748 }
749 }
750#endif
751 return ret;
752}
753
754
755/* symlink affected CPUs */
756static int cpufreq_add_dev_symlink(unsigned int cpu,
757 struct cpufreq_policy *policy)
758{
759 unsigned int j;
760 int ret = 0;
761
762 for_each_cpu(j, policy->cpus) {
763 struct cpufreq_policy *managed_policy;
764 struct sys_device *cpu_sys_dev;
765
766 if (j == cpu)
767 continue;
768 if (!cpu_online(j))
769 continue;
770
771 pr_debug("CPU %u already managed, adding link\n", j);
772 managed_policy = cpufreq_cpu_get(cpu);
773 cpu_sys_dev = get_cpu_sysdev(j);
774 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
775 "cpufreq");
776 if (ret) {
777 cpufreq_cpu_put(managed_policy);
778 return ret;
779 }
780 }
781 return ret;
782}
783
784static int cpufreq_add_dev_interface(unsigned int cpu,
785 struct cpufreq_policy *policy,
786 struct sys_device *sys_dev)
787{
788 struct cpufreq_policy new_policy;
789 struct freq_attr **drv_attr;
790 unsigned long flags;
791 int ret = 0;
792 unsigned int j;
793
794 /* prepare interface data */
795 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
796 &sys_dev->kobj, "cpufreq");
797 if (ret)
798 return ret;
799
800 /* set up files for this cpu device */
801 drv_attr = cpufreq_driver->attr;
802 while ((drv_attr) && (*drv_attr)) {
803 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
804 if (ret)
805 goto err_out_kobj_put;
806 drv_attr++;
807 }
808 if (cpufreq_driver->get) {
809 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
810 if (ret)
811 goto err_out_kobj_put;
812 }
813 if (cpufreq_driver->target) {
814 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
815 if (ret)
816 goto err_out_kobj_put;
817 }
818 if (cpufreq_driver->bios_limit) {
819 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
820 if (ret)
821 goto err_out_kobj_put;
822 }
823
824 spin_lock_irqsave(&cpufreq_driver_lock, flags);
825 for_each_cpu(j, policy->cpus) {
826 if (!cpu_online(j))
827 continue;
828 per_cpu(cpufreq_cpu_data, j) = policy;
829 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
830 }
831 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
832
833 ret = cpufreq_add_dev_symlink(cpu, policy);
834 if (ret)
835 goto err_out_kobj_put;
836
837 memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
838 /* assure that the starting sequence is run in __cpufreq_set_policy */
839 policy->governor = NULL;
840
841 /* set default policy */
842 ret = __cpufreq_set_policy(policy, &new_policy);
843 policy->user_policy.policy = policy->policy;
844 policy->user_policy.governor = policy->governor;
845
846 if (ret) {
847 pr_debug("setting policy failed\n");
848 if (cpufreq_driver->exit)
849 cpufreq_driver->exit(policy);
850 }
851 return ret;
852
853err_out_kobj_put:
854 kobject_put(&policy->kobj);
855 wait_for_completion(&policy->kobj_unregister);
856 return ret;
857}
858
859
860/**
861 * cpufreq_add_dev - add a CPU device
862 *
863 * Adds the cpufreq interface for a CPU device.
864 *
865 * The Oracle says: try running cpufreq registration/unregistration concurrently
866 * with with cpu hotplugging and all hell will break loose. Tried to clean this
867 * mess up, but more thorough testing is needed. - Mathieu
868 */
869static int cpufreq_add_dev(struct sys_device *sys_dev)
870{
871 unsigned int cpu = sys_dev->id;
872 int ret = 0, found = 0;
873 struct cpufreq_policy *policy;
874 unsigned long flags;
875 unsigned int j;
876#ifdef CONFIG_HOTPLUG_CPU
877 int sibling;
878#endif
879
880 if (cpu_is_offline(cpu))
881 return 0;
882
883 pr_debug("adding CPU %u\n", cpu);
884
885#ifdef CONFIG_SMP
886 /* check whether a different CPU already registered this
887 * CPU because it is in the same boat. */
888 policy = cpufreq_cpu_get(cpu);
889 if (unlikely(policy)) {
890 cpufreq_cpu_put(policy);
891 return 0;
892 }
893#endif
894
895 if (!try_module_get(cpufreq_driver->owner)) {
896 ret = -EINVAL;
897 goto module_out;
898 }
899
900 ret = -ENOMEM;
901 policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
902 if (!policy)
903 goto nomem_out;
904
905 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
906 goto err_free_policy;
907
908 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
909 goto err_free_cpumask;
910
911 policy->cpu = cpu;
912 cpumask_copy(policy->cpus, cpumask_of(cpu));
913
914 /* Initially set CPU itself as the policy_cpu */
915 per_cpu(cpufreq_policy_cpu, cpu) = cpu;
916 ret = (lock_policy_rwsem_write(cpu) < 0);
917 WARN_ON(ret);
918
919 init_completion(&policy->kobj_unregister);
920 INIT_WORK(&policy->update, handle_update);
921
922 /* Set governor before ->init, so that driver could check it */
923#ifdef CONFIG_HOTPLUG_CPU
924 for_each_online_cpu(sibling) {
925 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
926 if (cp && cp->governor &&
927 (cpumask_test_cpu(cpu, cp->related_cpus))) {
928 policy->governor = cp->governor;
929 found = 1;
930 break;
931 }
932 }
933#endif
934 if (!found)
935 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
936 /* call driver. From then on the cpufreq must be able
937 * to accept all calls to ->verify and ->setpolicy for this CPU
938 */
939 ret = cpufreq_driver->init(policy);
940 if (ret) {
941 pr_debug("initialization failed\n");
942 goto err_unlock_policy;
943 }
944 policy->user_policy.min = policy->min;
945 policy->user_policy.max = policy->max;
946
947 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
948 CPUFREQ_START, policy);
949
950 ret = cpufreq_add_dev_policy(cpu, policy, sys_dev);
951 if (ret) {
952 if (ret > 0)
953 /* This is a managed cpu, symlink created,
954 exit with 0 */
955 ret = 0;
956 goto err_unlock_policy;
957 }
958
959 ret = cpufreq_add_dev_interface(cpu, policy, sys_dev);
960 if (ret)
961 goto err_out_unregister;
962
963 unlock_policy_rwsem_write(cpu);
964
965 kobject_uevent(&policy->kobj, KOBJ_ADD);
966 module_put(cpufreq_driver->owner);
967 pr_debug("initialization complete\n");
968
969 return 0;
970
971
972err_out_unregister:
973 spin_lock_irqsave(&cpufreq_driver_lock, flags);
974 for_each_cpu(j, policy->cpus)
975 per_cpu(cpufreq_cpu_data, j) = NULL;
976 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
977
978 kobject_put(&policy->kobj);
979 wait_for_completion(&policy->kobj_unregister);
980
981err_unlock_policy:
982 unlock_policy_rwsem_write(cpu);
983 free_cpumask_var(policy->related_cpus);
984err_free_cpumask:
985 free_cpumask_var(policy->cpus);
986err_free_policy:
987 kfree(policy);
988nomem_out:
989 module_put(cpufreq_driver->owner);
990module_out:
991 return ret;
992}
993
994
995/**
996 * __cpufreq_remove_dev - remove a CPU device
997 *
998 * Removes the cpufreq interface for a CPU device.
999 * Caller should already have policy_rwsem in write mode for this CPU.
1000 * This routine frees the rwsem before returning.
1001 */
1002static int __cpufreq_remove_dev(struct sys_device *sys_dev)
1003{
1004 unsigned int cpu = sys_dev->id;
1005 unsigned long flags;
1006 struct cpufreq_policy *data;
1007 struct kobject *kobj;
1008 struct completion *cmp;
1009#ifdef CONFIG_SMP
1010 struct sys_device *cpu_sys_dev;
1011 unsigned int j;
1012#endif
1013
1014 pr_debug("unregistering CPU %u\n", cpu);
1015
1016 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1017 data = per_cpu(cpufreq_cpu_data, cpu);
1018
1019 if (!data) {
1020 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1021 unlock_policy_rwsem_write(cpu);
1022 return -EINVAL;
1023 }
1024 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1025
1026
1027#ifdef CONFIG_SMP
1028 /* if this isn't the CPU which is the parent of the kobj, we
1029 * only need to unlink, put and exit
1030 */
1031 if (unlikely(cpu != data->cpu)) {
1032 pr_debug("removing link\n");
1033 cpumask_clear_cpu(cpu, data->cpus);
1034 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1035 kobj = &sys_dev->kobj;
1036 cpufreq_cpu_put(data);
1037 unlock_policy_rwsem_write(cpu);
1038 sysfs_remove_link(kobj, "cpufreq");
1039 return 0;
1040 }
1041#endif
1042
1043#ifdef CONFIG_SMP
1044
1045#ifdef CONFIG_HOTPLUG_CPU
1046 strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name,
1047 CPUFREQ_NAME_LEN);
1048#endif
1049
1050 /* if we have other CPUs still registered, we need to unlink them,
1051 * or else wait_for_completion below will lock up. Clean the
1052 * per_cpu(cpufreq_cpu_data) while holding the lock, and remove
1053 * the sysfs links afterwards.
1054 */
1055 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1056 for_each_cpu(j, data->cpus) {
1057 if (j == cpu)
1058 continue;
1059 per_cpu(cpufreq_cpu_data, j) = NULL;
1060 }
1061 }
1062
1063 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1064
1065 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1066 for_each_cpu(j, data->cpus) {
1067 if (j == cpu)
1068 continue;
1069 pr_debug("removing link for cpu %u\n", j);
1070#ifdef CONFIG_HOTPLUG_CPU
1071 strncpy(per_cpu(cpufreq_cpu_governor, j),
1072 data->governor->name, CPUFREQ_NAME_LEN);
1073#endif
1074 cpu_sys_dev = get_cpu_sysdev(j);
1075 kobj = &cpu_sys_dev->kobj;
1076 unlock_policy_rwsem_write(cpu);
1077 sysfs_remove_link(kobj, "cpufreq");
1078 lock_policy_rwsem_write(cpu);
1079 cpufreq_cpu_put(data);
1080 }
1081 }
1082#else
1083 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1084#endif
1085
1086 if (cpufreq_driver->target)
1087 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1088
1089 kobj = &data->kobj;
1090 cmp = &data->kobj_unregister;
1091 unlock_policy_rwsem_write(cpu);
1092 kobject_put(kobj);
1093
1094 /* we need to make sure that the underlying kobj is actually
1095 * not referenced anymore by anybody before we proceed with
1096 * unloading.
1097 */
1098 pr_debug("waiting for dropping of refcount\n");
1099 wait_for_completion(cmp);
1100 pr_debug("wait complete\n");
1101
1102 lock_policy_rwsem_write(cpu);
1103 if (cpufreq_driver->exit)
1104 cpufreq_driver->exit(data);
1105 unlock_policy_rwsem_write(cpu);
1106
1107#ifdef CONFIG_HOTPLUG_CPU
1108 /* when the CPU which is the parent of the kobj is hotplugged
1109 * offline, check for siblings, and create cpufreq sysfs interface
1110 * and symlinks
1111 */
1112 if (unlikely(cpumask_weight(data->cpus) > 1)) {
1113 /* first sibling now owns the new sysfs dir */
1114 cpumask_clear_cpu(cpu, data->cpus);
1115 cpufreq_add_dev(get_cpu_sysdev(cpumask_first(data->cpus)));
1116
1117 /* finally remove our own symlink */
1118 lock_policy_rwsem_write(cpu);
1119 __cpufreq_remove_dev(sys_dev);
1120 }
1121#endif
1122
1123 free_cpumask_var(data->related_cpus);
1124 free_cpumask_var(data->cpus);
1125 kfree(data);
1126
1127 return 0;
1128}
1129
1130
1131static int cpufreq_remove_dev(struct sys_device *sys_dev)
1132{
1133 unsigned int cpu = sys_dev->id;
1134 int retval;
1135
1136 if (cpu_is_offline(cpu))
1137 return 0;
1138
1139 if (unlikely(lock_policy_rwsem_write(cpu)))
1140 BUG();
1141
1142 retval = __cpufreq_remove_dev(sys_dev);
1143 return retval;
1144}
1145
1146
1147static void handle_update(struct work_struct *work)
1148{
1149 struct cpufreq_policy *policy =
1150 container_of(work, struct cpufreq_policy, update);
1151 unsigned int cpu = policy->cpu;
1152 pr_debug("handle_update for cpu %u called\n", cpu);
1153 cpufreq_update_policy(cpu);
1154}
1155
1156/**
1157 * cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1158 * @cpu: cpu number
1159 * @old_freq: CPU frequency the kernel thinks the CPU runs at
1160 * @new_freq: CPU frequency the CPU actually runs at
1161 *
1162 * We adjust to current frequency first, and need to clean up later.
1163 * So either call to cpufreq_update_policy() or schedule handle_update()).
1164 */
1165static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1166 unsigned int new_freq)
1167{
1168 struct cpufreq_freqs freqs;
1169
1170 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1171 "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1172
1173 freqs.cpu = cpu;
1174 freqs.old = old_freq;
1175 freqs.new = new_freq;
1176 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1177 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1178}
1179
1180
1181/**
1182 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1183 * @cpu: CPU number
1184 *
1185 * This is the last known freq, without actually getting it from the driver.
1186 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1187 */
1188unsigned int cpufreq_quick_get(unsigned int cpu)
1189{
1190 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1191 unsigned int ret_freq = 0;
1192
1193 if (policy) {
1194 ret_freq = policy->cur;
1195 cpufreq_cpu_put(policy);
1196 }
1197
1198 return ret_freq;
1199}
1200EXPORT_SYMBOL(cpufreq_quick_get);
1201
1202/**
1203 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1204 * @cpu: CPU number
1205 *
1206 * Just return the max possible frequency for a given CPU.
1207 */
1208unsigned int cpufreq_quick_get_max(unsigned int cpu)
1209{
1210 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1211 unsigned int ret_freq = 0;
1212
1213 if (policy) {
1214 ret_freq = policy->max;
1215 cpufreq_cpu_put(policy);
1216 }
1217
1218 return ret_freq;
1219}
1220EXPORT_SYMBOL(cpufreq_quick_get_max);
1221
1222
1223static unsigned int __cpufreq_get(unsigned int cpu)
1224{
1225 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1226 unsigned int ret_freq = 0;
1227
1228 if (!cpufreq_driver->get)
1229 return ret_freq;
1230
1231 ret_freq = cpufreq_driver->get(cpu);
1232
1233 if (ret_freq && policy->cur &&
1234 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1235 /* verify no discrepancy between actual and
1236 saved value exists */
1237 if (unlikely(ret_freq != policy->cur)) {
1238 cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1239 schedule_work(&policy->update);
1240 }
1241 }
1242
1243 return ret_freq;
1244}
1245
1246/**
1247 * cpufreq_get - get the current CPU frequency (in kHz)
1248 * @cpu: CPU number
1249 *
1250 * Get the CPU current (static) CPU frequency
1251 */
1252unsigned int cpufreq_get(unsigned int cpu)
1253{
1254 unsigned int ret_freq = 0;
1255 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1256
1257 if (!policy)
1258 goto out;
1259
1260 if (unlikely(lock_policy_rwsem_read(cpu)))
1261 goto out_policy;
1262
1263 ret_freq = __cpufreq_get(cpu);
1264
1265 unlock_policy_rwsem_read(cpu);
1266
1267out_policy:
1268 cpufreq_cpu_put(policy);
1269out:
1270 return ret_freq;
1271}
1272EXPORT_SYMBOL(cpufreq_get);
1273
1274static struct sysdev_driver cpufreq_sysdev_driver = {
1275 .add = cpufreq_add_dev,
1276 .remove = cpufreq_remove_dev,
1277};
1278
1279
1280/**
1281 * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1282 *
1283 * This function is only executed for the boot processor. The other CPUs
1284 * have been put offline by means of CPU hotplug.
1285 */
1286static int cpufreq_bp_suspend(void)
1287{
1288 int ret = 0;
1289
1290 int cpu = smp_processor_id();
1291 struct cpufreq_policy *cpu_policy;
1292
1293 pr_debug("suspending cpu %u\n", cpu);
1294
1295 /* If there's no policy for the boot CPU, we have nothing to do. */
1296 cpu_policy = cpufreq_cpu_get(cpu);
1297 if (!cpu_policy)
1298 return 0;
1299
1300 if (cpufreq_driver->suspend) {
1301 ret = cpufreq_driver->suspend(cpu_policy);
1302 if (ret)
1303 printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1304 "step on CPU %u\n", cpu_policy->cpu);
1305 }
1306
1307 cpufreq_cpu_put(cpu_policy);
1308 return ret;
1309}
1310
1311/**
1312 * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1313 *
1314 * 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1315 * 2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1316 * restored. It will verify that the current freq is in sync with
1317 * what we believe it to be. This is a bit later than when it
1318 * should be, but nonethteless it's better than calling
1319 * cpufreq_driver->get() here which might re-enable interrupts...
1320 *
1321 * This function is only executed for the boot CPU. The other CPUs have not
1322 * been turned on yet.
1323 */
1324static void cpufreq_bp_resume(void)
1325{
1326 int ret = 0;
1327
1328 int cpu = smp_processor_id();
1329 struct cpufreq_policy *cpu_policy;
1330
1331 pr_debug("resuming cpu %u\n", cpu);
1332
1333 /* If there's no policy for the boot CPU, we have nothing to do. */
1334 cpu_policy = cpufreq_cpu_get(cpu);
1335 if (!cpu_policy)
1336 return;
1337
1338 if (cpufreq_driver->resume) {
1339 ret = cpufreq_driver->resume(cpu_policy);
1340 if (ret) {
1341 printk(KERN_ERR "cpufreq: resume failed in ->resume "
1342 "step on CPU %u\n", cpu_policy->cpu);
1343 goto fail;
1344 }
1345 }
1346
1347 schedule_work(&cpu_policy->update);
1348
1349fail:
1350 cpufreq_cpu_put(cpu_policy);
1351}
1352
1353static struct syscore_ops cpufreq_syscore_ops = {
1354 .suspend = cpufreq_bp_suspend,
1355 .resume = cpufreq_bp_resume,
1356};
1357
1358
1359/*********************************************************************
1360 * NOTIFIER LISTS INTERFACE *
1361 *********************************************************************/
1362
1363/**
1364 * cpufreq_register_notifier - register a driver with cpufreq
1365 * @nb: notifier function to register
1366 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1367 *
1368 * Add a driver to one of two lists: either a list of drivers that
1369 * are notified about clock rate changes (once before and once after
1370 * the transition), or a list of drivers that are notified about
1371 * changes in cpufreq policy.
1372 *
1373 * This function may sleep, and has the same return conditions as
1374 * blocking_notifier_chain_register.
1375 */
1376int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1377{
1378 int ret;
1379
1380 WARN_ON(!init_cpufreq_transition_notifier_list_called);
1381
1382 switch (list) {
1383 case CPUFREQ_TRANSITION_NOTIFIER:
1384 ret = srcu_notifier_chain_register(
1385 &cpufreq_transition_notifier_list, nb);
1386 break;
1387 case CPUFREQ_POLICY_NOTIFIER:
1388 ret = blocking_notifier_chain_register(
1389 &cpufreq_policy_notifier_list, nb);
1390 break;
1391 default:
1392 ret = -EINVAL;
1393 }
1394
1395 return ret;
1396}
1397EXPORT_SYMBOL(cpufreq_register_notifier);
1398
1399
1400/**
1401 * cpufreq_unregister_notifier - unregister a driver with cpufreq
1402 * @nb: notifier block to be unregistered
1403 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1404 *
1405 * Remove a driver from the CPU frequency notifier list.
1406 *
1407 * This function may sleep, and has the same return conditions as
1408 * blocking_notifier_chain_unregister.
1409 */
1410int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1411{
1412 int ret;
1413
1414 switch (list) {
1415 case CPUFREQ_TRANSITION_NOTIFIER:
1416 ret = srcu_notifier_chain_unregister(
1417 &cpufreq_transition_notifier_list, nb);
1418 break;
1419 case CPUFREQ_POLICY_NOTIFIER:
1420 ret = blocking_notifier_chain_unregister(
1421 &cpufreq_policy_notifier_list, nb);
1422 break;
1423 default:
1424 ret = -EINVAL;
1425 }
1426
1427 return ret;
1428}
1429EXPORT_SYMBOL(cpufreq_unregister_notifier);
1430
1431
1432/*********************************************************************
1433 * GOVERNORS *
1434 *********************************************************************/
1435
1436
1437int __cpufreq_driver_target(struct cpufreq_policy *policy,
1438 unsigned int target_freq,
1439 unsigned int relation)
1440{
1441 int retval = -EINVAL;
1442
1443 pr_debug("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1444 target_freq, relation);
1445 if (cpu_online(policy->cpu) && cpufreq_driver->target)
1446 retval = cpufreq_driver->target(policy, target_freq, relation);
1447
1448 return retval;
1449}
1450EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1451
1452int cpufreq_driver_target(struct cpufreq_policy *policy,
1453 unsigned int target_freq,
1454 unsigned int relation)
1455{
1456 int ret = -EINVAL;
1457
1458 policy = cpufreq_cpu_get(policy->cpu);
1459 if (!policy)
1460 goto no_policy;
1461
1462 if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1463 goto fail;
1464
1465 ret = __cpufreq_driver_target(policy, target_freq, relation);
1466
1467 unlock_policy_rwsem_write(policy->cpu);
1468
1469fail:
1470 cpufreq_cpu_put(policy);
1471no_policy:
1472 return ret;
1473}
1474EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1475
1476int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
1477{
1478 int ret = 0;
1479
1480 policy = cpufreq_cpu_get(policy->cpu);
1481 if (!policy)
1482 return -EINVAL;
1483
1484 if (cpu_online(cpu) && cpufreq_driver->getavg)
1485 ret = cpufreq_driver->getavg(policy, cpu);
1486
1487 cpufreq_cpu_put(policy);
1488 return ret;
1489}
1490EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1491
1492/*
1493 * when "event" is CPUFREQ_GOV_LIMITS
1494 */
1495
1496static int __cpufreq_governor(struct cpufreq_policy *policy,
1497 unsigned int event)
1498{
1499 int ret;
1500
1501 /* Only must be defined when default governor is known to have latency
1502 restrictions, like e.g. conservative or ondemand.
1503 That this is the case is already ensured in Kconfig
1504 */
1505#ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1506 struct cpufreq_governor *gov = &cpufreq_gov_performance;
1507#else
1508 struct cpufreq_governor *gov = NULL;
1509#endif
1510
1511 if (policy->governor->max_transition_latency &&
1512 policy->cpuinfo.transition_latency >
1513 policy->governor->max_transition_latency) {
1514 if (!gov)
1515 return -EINVAL;
1516 else {
1517 printk(KERN_WARNING "%s governor failed, too long"
1518 " transition latency of HW, fallback"
1519 " to %s governor\n",
1520 policy->governor->name,
1521 gov->name);
1522 policy->governor = gov;
1523 }
1524 }
1525
1526 if (!try_module_get(policy->governor->owner))
1527 return -EINVAL;
1528
1529 pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1530 policy->cpu, event);
1531 ret = policy->governor->governor(policy, event);
1532
1533 /* we keep one module reference alive for
1534 each CPU governed by this CPU */
1535 if ((event != CPUFREQ_GOV_START) || ret)
1536 module_put(policy->governor->owner);
1537 if ((event == CPUFREQ_GOV_STOP) && !ret)
1538 module_put(policy->governor->owner);
1539
1540 return ret;
1541}
1542
1543
1544int cpufreq_register_governor(struct cpufreq_governor *governor)
1545{
1546 int err;
1547
1548 if (!governor)
1549 return -EINVAL;
1550
1551 mutex_lock(&cpufreq_governor_mutex);
1552
1553 err = -EBUSY;
1554 if (__find_governor(governor->name) == NULL) {
1555 err = 0;
1556 list_add(&governor->governor_list, &cpufreq_governor_list);
1557 }
1558
1559 mutex_unlock(&cpufreq_governor_mutex);
1560 return err;
1561}
1562EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1563
1564
1565void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1566{
1567#ifdef CONFIG_HOTPLUG_CPU
1568 int cpu;
1569#endif
1570
1571 if (!governor)
1572 return;
1573
1574#ifdef CONFIG_HOTPLUG_CPU
1575 for_each_present_cpu(cpu) {
1576 if (cpu_online(cpu))
1577 continue;
1578 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1579 strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1580 }
1581#endif
1582
1583 mutex_lock(&cpufreq_governor_mutex);
1584 list_del(&governor->governor_list);
1585 mutex_unlock(&cpufreq_governor_mutex);
1586 return;
1587}
1588EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1589
1590
1591
1592/*********************************************************************
1593 * POLICY INTERFACE *
1594 *********************************************************************/
1595
1596/**
1597 * cpufreq_get_policy - get the current cpufreq_policy
1598 * @policy: struct cpufreq_policy into which the current cpufreq_policy
1599 * is written
1600 *
1601 * Reads the current cpufreq policy.
1602 */
1603int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1604{
1605 struct cpufreq_policy *cpu_policy;
1606 if (!policy)
1607 return -EINVAL;
1608
1609 cpu_policy = cpufreq_cpu_get(cpu);
1610 if (!cpu_policy)
1611 return -EINVAL;
1612
1613 memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1614
1615 cpufreq_cpu_put(cpu_policy);
1616 return 0;
1617}
1618EXPORT_SYMBOL(cpufreq_get_policy);
1619
1620
1621/*
1622 * data : current policy.
1623 * policy : policy to be set.
1624 */
1625static int __cpufreq_set_policy(struct cpufreq_policy *data,
1626 struct cpufreq_policy *policy)
1627{
1628 int ret = 0;
1629
1630 pr_debug("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1631 policy->min, policy->max);
1632
1633 memcpy(&policy->cpuinfo, &data->cpuinfo,
1634 sizeof(struct cpufreq_cpuinfo));
1635
1636 if (policy->min > data->max || policy->max < data->min) {
1637 ret = -EINVAL;
1638 goto error_out;
1639 }
1640
1641 /* verify the cpu speed can be set within this limit */
1642 ret = cpufreq_driver->verify(policy);
1643 if (ret)
1644 goto error_out;
1645
1646 /* adjust if necessary - all reasons */
1647 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1648 CPUFREQ_ADJUST, policy);
1649
1650 /* adjust if necessary - hardware incompatibility*/
1651 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1652 CPUFREQ_INCOMPATIBLE, policy);
1653
1654 /* verify the cpu speed can be set within this limit,
1655 which might be different to the first one */
1656 ret = cpufreq_driver->verify(policy);
1657 if (ret)
1658 goto error_out;
1659
1660 /* notification of the new policy */
1661 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1662 CPUFREQ_NOTIFY, policy);
1663
1664 data->min = policy->min;
1665 data->max = policy->max;
1666
1667 pr_debug("new min and max freqs are %u - %u kHz\n",
1668 data->min, data->max);
1669
1670 if (cpufreq_driver->setpolicy) {
1671 data->policy = policy->policy;
1672 pr_debug("setting range\n");
1673 ret = cpufreq_driver->setpolicy(policy);
1674 } else {
1675 if (policy->governor != data->governor) {
1676 /* save old, working values */
1677 struct cpufreq_governor *old_gov = data->governor;
1678
1679 pr_debug("governor switch\n");
1680
1681 /* end old governor */
1682 if (data->governor)
1683 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1684
1685 /* start new governor */
1686 data->governor = policy->governor;
1687 if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1688 /* new governor failed, so re-start old one */
1689 pr_debug("starting governor %s failed\n",
1690 data->governor->name);
1691 if (old_gov) {
1692 data->governor = old_gov;
1693 __cpufreq_governor(data,
1694 CPUFREQ_GOV_START);
1695 }
1696 ret = -EINVAL;
1697 goto error_out;
1698 }
1699 /* might be a policy change, too, so fall through */
1700 }
1701 pr_debug("governor: change or update limits\n");
1702 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1703 }
1704
1705error_out:
1706 return ret;
1707}
1708
1709/**
1710 * cpufreq_update_policy - re-evaluate an existing cpufreq policy
1711 * @cpu: CPU which shall be re-evaluated
1712 *
1713 * Useful for policy notifiers which have different necessities
1714 * at different times.
1715 */
1716int cpufreq_update_policy(unsigned int cpu)
1717{
1718 struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1719 struct cpufreq_policy policy;
1720 int ret;
1721
1722 if (!data) {
1723 ret = -ENODEV;
1724 goto no_policy;
1725 }
1726
1727 if (unlikely(lock_policy_rwsem_write(cpu))) {
1728 ret = -EINVAL;
1729 goto fail;
1730 }
1731
1732 pr_debug("updating policy for CPU %u\n", cpu);
1733 memcpy(&policy, data, sizeof(struct cpufreq_policy));
1734 policy.min = data->user_policy.min;
1735 policy.max = data->user_policy.max;
1736 policy.policy = data->user_policy.policy;
1737 policy.governor = data->user_policy.governor;
1738
1739 /* BIOS might change freq behind our back
1740 -> ask driver for current freq and notify governors about a change */
1741 if (cpufreq_driver->get) {
1742 policy.cur = cpufreq_driver->get(cpu);
1743 if (!data->cur) {
1744 pr_debug("Driver did not initialize current freq");
1745 data->cur = policy.cur;
1746 } else {
1747 if (data->cur != policy.cur)
1748 cpufreq_out_of_sync(cpu, data->cur,
1749 policy.cur);
1750 }
1751 }
1752
1753 ret = __cpufreq_set_policy(data, &policy);
1754
1755 unlock_policy_rwsem_write(cpu);
1756
1757fail:
1758 cpufreq_cpu_put(data);
1759no_policy:
1760 return ret;
1761}
1762EXPORT_SYMBOL(cpufreq_update_policy);
1763
1764static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1765 unsigned long action, void *hcpu)
1766{
1767 unsigned int cpu = (unsigned long)hcpu;
1768 struct sys_device *sys_dev;
1769
1770 sys_dev = get_cpu_sysdev(cpu);
1771 if (sys_dev) {
1772 switch (action) {
1773 case CPU_ONLINE:
1774 case CPU_ONLINE_FROZEN:
1775 cpufreq_add_dev(sys_dev);
1776 break;
1777 case CPU_DOWN_PREPARE:
1778 case CPU_DOWN_PREPARE_FROZEN:
1779 if (unlikely(lock_policy_rwsem_write(cpu)))
1780 BUG();
1781
1782 __cpufreq_remove_dev(sys_dev);
1783 break;
1784 case CPU_DOWN_FAILED:
1785 case CPU_DOWN_FAILED_FROZEN:
1786 cpufreq_add_dev(sys_dev);
1787 break;
1788 }
1789 }
1790 return NOTIFY_OK;
1791}
1792
1793static struct notifier_block __refdata cpufreq_cpu_notifier = {
1794 .notifier_call = cpufreq_cpu_callback,
1795};
1796
1797/*********************************************************************
1798 * REGISTER / UNREGISTER CPUFREQ DRIVER *
1799 *********************************************************************/
1800
1801/**
1802 * cpufreq_register_driver - register a CPU Frequency driver
1803 * @driver_data: A struct cpufreq_driver containing the values#
1804 * submitted by the CPU Frequency driver.
1805 *
1806 * Registers a CPU Frequency driver to this core code. This code
1807 * returns zero on success, -EBUSY when another driver got here first
1808 * (and isn't unregistered in the meantime).
1809 *
1810 */
1811int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1812{
1813 unsigned long flags;
1814 int ret;
1815
1816 if (!driver_data || !driver_data->verify || !driver_data->init ||
1817 ((!driver_data->setpolicy) && (!driver_data->target)))
1818 return -EINVAL;
1819
1820 pr_debug("trying to register driver %s\n", driver_data->name);
1821
1822 if (driver_data->setpolicy)
1823 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1824
1825 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1826 if (cpufreq_driver) {
1827 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1828 return -EBUSY;
1829 }
1830 cpufreq_driver = driver_data;
1831 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1832
1833 ret = sysdev_driver_register(&cpu_sysdev_class,
1834 &cpufreq_sysdev_driver);
1835 if (ret)
1836 goto err_null_driver;
1837
1838 if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1839 int i;
1840 ret = -ENODEV;
1841
1842 /* check for at least one working CPU */
1843 for (i = 0; i < nr_cpu_ids; i++)
1844 if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
1845 ret = 0;
1846 break;
1847 }
1848
1849 /* if all ->init() calls failed, unregister */
1850 if (ret) {
1851 pr_debug("no CPU initialized for driver %s\n",
1852 driver_data->name);
1853 goto err_sysdev_unreg;
1854 }
1855 }
1856
1857 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1858 pr_debug("driver %s up and running\n", driver_data->name);
1859
1860 return 0;
1861err_sysdev_unreg:
1862 sysdev_driver_unregister(&cpu_sysdev_class,
1863 &cpufreq_sysdev_driver);
1864err_null_driver:
1865 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1866 cpufreq_driver = NULL;
1867 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1868 return ret;
1869}
1870EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1871
1872
1873/**
1874 * cpufreq_unregister_driver - unregister the current CPUFreq driver
1875 *
1876 * Unregister the current CPUFreq driver. Only call this if you have
1877 * the right to do so, i.e. if you have succeeded in initialising before!
1878 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1879 * currently not initialised.
1880 */
1881int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1882{
1883 unsigned long flags;
1884
1885 if (!cpufreq_driver || (driver != cpufreq_driver))
1886 return -EINVAL;
1887
1888 pr_debug("unregistering driver %s\n", driver->name);
1889
1890 sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1891 unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1892
1893 spin_lock_irqsave(&cpufreq_driver_lock, flags);
1894 cpufreq_driver = NULL;
1895 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1896
1897 return 0;
1898}
1899EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1900
1901static int __init cpufreq_core_init(void)
1902{
1903 int cpu;
1904
1905 for_each_possible_cpu(cpu) {
1906 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1907 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1908 }
1909
1910 cpufreq_global_kobject = kobject_create_and_add("cpufreq",
1911 &cpu_sysdev_class.kset.kobj);
1912 BUG_ON(!cpufreq_global_kobject);
1913 register_syscore_ops(&cpufreq_syscore_ops);
1914
1915 return 0;
1916}
1917core_initcall(cpufreq_core_init);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/drivers/cpufreq/cpufreq.c
4 *
5 * Copyright (C) 2001 Russell King
6 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
7 * (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
8 *
9 * Oct 2005 - Ashok Raj <ashok.raj@intel.com>
10 * Added handling for CPU hotplug
11 * Feb 2006 - Jacob Shin <jacob.shin@amd.com>
12 * Fix handling for CPU hotplug -- affected CPUs
13 */
14
15#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16
17#include <linux/cpu.h>
18#include <linux/cpufreq.h>
19#include <linux/cpu_cooling.h>
20#include <linux/delay.h>
21#include <linux/device.h>
22#include <linux/init.h>
23#include <linux/kernel_stat.h>
24#include <linux/module.h>
25#include <linux/mutex.h>
26#include <linux/pm_qos.h>
27#include <linux/slab.h>
28#include <linux/suspend.h>
29#include <linux/syscore_ops.h>
30#include <linux/tick.h>
31#include <linux/units.h>
32#include <trace/events/power.h>
33
34static LIST_HEAD(cpufreq_policy_list);
35
36/* Macros to iterate over CPU policies */
37#define for_each_suitable_policy(__policy, __active) \
38 list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
39 if ((__active) == !policy_is_inactive(__policy))
40
41#define for_each_active_policy(__policy) \
42 for_each_suitable_policy(__policy, true)
43#define for_each_inactive_policy(__policy) \
44 for_each_suitable_policy(__policy, false)
45
46/* Iterate over governors */
47static LIST_HEAD(cpufreq_governor_list);
48#define for_each_governor(__governor) \
49 list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
50
51static char default_governor[CPUFREQ_NAME_LEN];
52
53/*
54 * The "cpufreq driver" - the arch- or hardware-dependent low
55 * level driver of CPUFreq support, and its spinlock. This lock
56 * also protects the cpufreq_cpu_data array.
57 */
58static struct cpufreq_driver *cpufreq_driver;
59static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
60static DEFINE_RWLOCK(cpufreq_driver_lock);
61
62static DEFINE_STATIC_KEY_FALSE(cpufreq_freq_invariance);
63bool cpufreq_supports_freq_invariance(void)
64{
65 return static_branch_likely(&cpufreq_freq_invariance);
66}
67
68/* Flag to suspend/resume CPUFreq governors */
69static bool cpufreq_suspended;
70
71static inline bool has_target(void)
72{
73 return cpufreq_driver->target_index || cpufreq_driver->target;
74}
75
76/* internal prototypes */
77static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
78static int cpufreq_init_governor(struct cpufreq_policy *policy);
79static void cpufreq_exit_governor(struct cpufreq_policy *policy);
80static void cpufreq_governor_limits(struct cpufreq_policy *policy);
81static int cpufreq_set_policy(struct cpufreq_policy *policy,
82 struct cpufreq_governor *new_gov,
83 unsigned int new_pol);
84
85/*
86 * Two notifier lists: the "policy" list is involved in the
87 * validation process for a new CPU frequency policy; the
88 * "transition" list for kernel code that needs to handle
89 * changes to devices when the CPU clock speed changes.
90 * The mutex locks both lists.
91 */
92static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
93SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
94
95static int off __read_mostly;
96static int cpufreq_disabled(void)
97{
98 return off;
99}
100void disable_cpufreq(void)
101{
102 off = 1;
103}
104static DEFINE_MUTEX(cpufreq_governor_mutex);
105
106bool have_governor_per_policy(void)
107{
108 return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
109}
110EXPORT_SYMBOL_GPL(have_governor_per_policy);
111
112static struct kobject *cpufreq_global_kobject;
113
114struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
115{
116 if (have_governor_per_policy())
117 return &policy->kobj;
118 else
119 return cpufreq_global_kobject;
120}
121EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
122
123static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
124{
125 struct kernel_cpustat kcpustat;
126 u64 cur_wall_time;
127 u64 idle_time;
128 u64 busy_time;
129
130 cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
131
132 kcpustat_cpu_fetch(&kcpustat, cpu);
133
134 busy_time = kcpustat.cpustat[CPUTIME_USER];
135 busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
136 busy_time += kcpustat.cpustat[CPUTIME_IRQ];
137 busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
138 busy_time += kcpustat.cpustat[CPUTIME_STEAL];
139 busy_time += kcpustat.cpustat[CPUTIME_NICE];
140
141 idle_time = cur_wall_time - busy_time;
142 if (wall)
143 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
144
145 return div_u64(idle_time, NSEC_PER_USEC);
146}
147
148u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
149{
150 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
151
152 if (idle_time == -1ULL)
153 return get_cpu_idle_time_jiffy(cpu, wall);
154 else if (!io_busy)
155 idle_time += get_cpu_iowait_time_us(cpu, wall);
156
157 return idle_time;
158}
159EXPORT_SYMBOL_GPL(get_cpu_idle_time);
160
161/*
162 * This is a generic cpufreq init() routine which can be used by cpufreq
163 * drivers of SMP systems. It will do following:
164 * - validate & show freq table passed
165 * - set policies transition latency
166 * - policy->cpus with all possible CPUs
167 */
168void cpufreq_generic_init(struct cpufreq_policy *policy,
169 struct cpufreq_frequency_table *table,
170 unsigned int transition_latency)
171{
172 policy->freq_table = table;
173 policy->cpuinfo.transition_latency = transition_latency;
174
175 /*
176 * The driver only supports the SMP configuration where all processors
177 * share the clock and voltage and clock.
178 */
179 cpumask_setall(policy->cpus);
180}
181EXPORT_SYMBOL_GPL(cpufreq_generic_init);
182
183struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
184{
185 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
186
187 return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
188}
189EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
190
191unsigned int cpufreq_generic_get(unsigned int cpu)
192{
193 struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
194
195 if (!policy || IS_ERR(policy->clk)) {
196 pr_err("%s: No %s associated to cpu: %d\n",
197 __func__, policy ? "clk" : "policy", cpu);
198 return 0;
199 }
200
201 return clk_get_rate(policy->clk) / 1000;
202}
203EXPORT_SYMBOL_GPL(cpufreq_generic_get);
204
205/**
206 * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
207 * @cpu: CPU to find the policy for.
208 *
209 * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
210 * the kobject reference counter of that policy. Return a valid policy on
211 * success or NULL on failure.
212 *
213 * The policy returned by this function has to be released with the help of
214 * cpufreq_cpu_put() to balance its kobject reference counter properly.
215 */
216struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
217{
218 struct cpufreq_policy *policy = NULL;
219 unsigned long flags;
220
221 if (WARN_ON(cpu >= nr_cpu_ids))
222 return NULL;
223
224 /* get the cpufreq driver */
225 read_lock_irqsave(&cpufreq_driver_lock, flags);
226
227 if (cpufreq_driver) {
228 /* get the CPU */
229 policy = cpufreq_cpu_get_raw(cpu);
230 if (policy)
231 kobject_get(&policy->kobj);
232 }
233
234 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
235
236 return policy;
237}
238EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
239
240/**
241 * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
242 * @policy: cpufreq policy returned by cpufreq_cpu_get().
243 */
244void cpufreq_cpu_put(struct cpufreq_policy *policy)
245{
246 kobject_put(&policy->kobj);
247}
248EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
249
250/**
251 * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
252 * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
253 */
254void cpufreq_cpu_release(struct cpufreq_policy *policy)
255{
256 if (WARN_ON(!policy))
257 return;
258
259 lockdep_assert_held(&policy->rwsem);
260
261 up_write(&policy->rwsem);
262
263 cpufreq_cpu_put(policy);
264}
265
266/**
267 * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
268 * @cpu: CPU to find the policy for.
269 *
270 * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
271 * if the policy returned by it is not NULL, acquire its rwsem for writing.
272 * Return the policy if it is active or release it and return NULL otherwise.
273 *
274 * The policy returned by this function has to be released with the help of
275 * cpufreq_cpu_release() in order to release its rwsem and balance its usage
276 * counter properly.
277 */
278struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
279{
280 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
281
282 if (!policy)
283 return NULL;
284
285 down_write(&policy->rwsem);
286
287 if (policy_is_inactive(policy)) {
288 cpufreq_cpu_release(policy);
289 return NULL;
290 }
291
292 return policy;
293}
294
295/*********************************************************************
296 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
297 *********************************************************************/
298
299/**
300 * adjust_jiffies - Adjust the system "loops_per_jiffy".
301 * @val: CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
302 * @ci: Frequency change information.
303 *
304 * This function alters the system "loops_per_jiffy" for the clock
305 * speed change. Note that loops_per_jiffy cannot be updated on SMP
306 * systems as each CPU might be scaled differently. So, use the arch
307 * per-CPU loops_per_jiffy value wherever possible.
308 */
309static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
310{
311#ifndef CONFIG_SMP
312 static unsigned long l_p_j_ref;
313 static unsigned int l_p_j_ref_freq;
314
315 if (ci->flags & CPUFREQ_CONST_LOOPS)
316 return;
317
318 if (!l_p_j_ref_freq) {
319 l_p_j_ref = loops_per_jiffy;
320 l_p_j_ref_freq = ci->old;
321 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
322 l_p_j_ref, l_p_j_ref_freq);
323 }
324 if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
325 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
326 ci->new);
327 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
328 loops_per_jiffy, ci->new);
329 }
330#endif
331}
332
333/**
334 * cpufreq_notify_transition - Notify frequency transition and adjust jiffies.
335 * @policy: cpufreq policy to enable fast frequency switching for.
336 * @freqs: contain details of the frequency update.
337 * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
338 *
339 * This function calls the transition notifiers and adjust_jiffies().
340 *
341 * It is called twice on all CPU frequency changes that have external effects.
342 */
343static void cpufreq_notify_transition(struct cpufreq_policy *policy,
344 struct cpufreq_freqs *freqs,
345 unsigned int state)
346{
347 int cpu;
348
349 BUG_ON(irqs_disabled());
350
351 if (cpufreq_disabled())
352 return;
353
354 freqs->policy = policy;
355 freqs->flags = cpufreq_driver->flags;
356 pr_debug("notification %u of frequency transition to %u kHz\n",
357 state, freqs->new);
358
359 switch (state) {
360 case CPUFREQ_PRECHANGE:
361 /*
362 * Detect if the driver reported a value as "old frequency"
363 * which is not equal to what the cpufreq core thinks is
364 * "old frequency".
365 */
366 if (policy->cur && policy->cur != freqs->old) {
367 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
368 freqs->old, policy->cur);
369 freqs->old = policy->cur;
370 }
371
372 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
373 CPUFREQ_PRECHANGE, freqs);
374
375 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
376 break;
377
378 case CPUFREQ_POSTCHANGE:
379 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
380 pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
381 cpumask_pr_args(policy->cpus));
382
383 for_each_cpu(cpu, policy->cpus)
384 trace_cpu_frequency(freqs->new, cpu);
385
386 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
387 CPUFREQ_POSTCHANGE, freqs);
388
389 cpufreq_stats_record_transition(policy, freqs->new);
390 policy->cur = freqs->new;
391 }
392}
393
394/* Do post notifications when there are chances that transition has failed */
395static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
396 struct cpufreq_freqs *freqs, int transition_failed)
397{
398 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
399 if (!transition_failed)
400 return;
401
402 swap(freqs->old, freqs->new);
403 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
404 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
405}
406
407void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
408 struct cpufreq_freqs *freqs)
409{
410
411 /*
412 * Catch double invocations of _begin() which lead to self-deadlock.
413 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
414 * doesn't invoke _begin() on their behalf, and hence the chances of
415 * double invocations are very low. Moreover, there are scenarios
416 * where these checks can emit false-positive warnings in these
417 * drivers; so we avoid that by skipping them altogether.
418 */
419 WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
420 && current == policy->transition_task);
421
422wait:
423 wait_event(policy->transition_wait, !policy->transition_ongoing);
424
425 spin_lock(&policy->transition_lock);
426
427 if (unlikely(policy->transition_ongoing)) {
428 spin_unlock(&policy->transition_lock);
429 goto wait;
430 }
431
432 policy->transition_ongoing = true;
433 policy->transition_task = current;
434
435 spin_unlock(&policy->transition_lock);
436
437 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
438}
439EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
440
441void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
442 struct cpufreq_freqs *freqs, int transition_failed)
443{
444 if (WARN_ON(!policy->transition_ongoing))
445 return;
446
447 cpufreq_notify_post_transition(policy, freqs, transition_failed);
448
449 arch_set_freq_scale(policy->related_cpus,
450 policy->cur,
451 policy->cpuinfo.max_freq);
452
453 policy->transition_ongoing = false;
454 policy->transition_task = NULL;
455
456 wake_up(&policy->transition_wait);
457}
458EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
459
460/*
461 * Fast frequency switching status count. Positive means "enabled", negative
462 * means "disabled" and 0 means "not decided yet".
463 */
464static int cpufreq_fast_switch_count;
465static DEFINE_MUTEX(cpufreq_fast_switch_lock);
466
467static void cpufreq_list_transition_notifiers(void)
468{
469 struct notifier_block *nb;
470
471 pr_info("Registered transition notifiers:\n");
472
473 mutex_lock(&cpufreq_transition_notifier_list.mutex);
474
475 for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
476 pr_info("%pS\n", nb->notifier_call);
477
478 mutex_unlock(&cpufreq_transition_notifier_list.mutex);
479}
480
481/**
482 * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
483 * @policy: cpufreq policy to enable fast frequency switching for.
484 *
485 * Try to enable fast frequency switching for @policy.
486 *
487 * The attempt will fail if there is at least one transition notifier registered
488 * at this point, as fast frequency switching is quite fundamentally at odds
489 * with transition notifiers. Thus if successful, it will make registration of
490 * transition notifiers fail going forward.
491 */
492void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
493{
494 lockdep_assert_held(&policy->rwsem);
495
496 if (!policy->fast_switch_possible)
497 return;
498
499 mutex_lock(&cpufreq_fast_switch_lock);
500 if (cpufreq_fast_switch_count >= 0) {
501 cpufreq_fast_switch_count++;
502 policy->fast_switch_enabled = true;
503 } else {
504 pr_warn("CPU%u: Fast frequency switching not enabled\n",
505 policy->cpu);
506 cpufreq_list_transition_notifiers();
507 }
508 mutex_unlock(&cpufreq_fast_switch_lock);
509}
510EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
511
512/**
513 * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
514 * @policy: cpufreq policy to disable fast frequency switching for.
515 */
516void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
517{
518 mutex_lock(&cpufreq_fast_switch_lock);
519 if (policy->fast_switch_enabled) {
520 policy->fast_switch_enabled = false;
521 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
522 cpufreq_fast_switch_count--;
523 }
524 mutex_unlock(&cpufreq_fast_switch_lock);
525}
526EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
527
528static unsigned int __resolve_freq(struct cpufreq_policy *policy,
529 unsigned int target_freq, unsigned int relation)
530{
531 unsigned int idx;
532
533 target_freq = clamp_val(target_freq, policy->min, policy->max);
534
535 if (!policy->freq_table)
536 return target_freq;
537
538 idx = cpufreq_frequency_table_target(policy, target_freq, relation);
539 policy->cached_resolved_idx = idx;
540 policy->cached_target_freq = target_freq;
541 return policy->freq_table[idx].frequency;
542}
543
544/**
545 * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
546 * one.
547 * @policy: associated policy to interrogate
548 * @target_freq: target frequency to resolve.
549 *
550 * The target to driver frequency mapping is cached in the policy.
551 *
552 * Return: Lowest driver-supported frequency greater than or equal to the
553 * given target_freq, subject to policy (min/max) and driver limitations.
554 */
555unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
556 unsigned int target_freq)
557{
558 return __resolve_freq(policy, target_freq, CPUFREQ_RELATION_LE);
559}
560EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
561
562unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
563{
564 unsigned int latency;
565
566 if (policy->transition_delay_us)
567 return policy->transition_delay_us;
568
569 latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
570 if (latency) {
571 /*
572 * For platforms that can change the frequency very fast (< 10
573 * us), the above formula gives a decent transition delay. But
574 * for platforms where transition_latency is in milliseconds, it
575 * ends up giving unrealistic values.
576 *
577 * Cap the default transition delay to 10 ms, which seems to be
578 * a reasonable amount of time after which we should reevaluate
579 * the frequency.
580 */
581 return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
582 }
583
584 return LATENCY_MULTIPLIER;
585}
586EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
587
588/*********************************************************************
589 * SYSFS INTERFACE *
590 *********************************************************************/
591static ssize_t show_boost(struct kobject *kobj,
592 struct kobj_attribute *attr, char *buf)
593{
594 return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
595}
596
597static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
598 const char *buf, size_t count)
599{
600 int ret, enable;
601
602 ret = sscanf(buf, "%d", &enable);
603 if (ret != 1 || enable < 0 || enable > 1)
604 return -EINVAL;
605
606 if (cpufreq_boost_trigger_state(enable)) {
607 pr_err("%s: Cannot %s BOOST!\n",
608 __func__, enable ? "enable" : "disable");
609 return -EINVAL;
610 }
611
612 pr_debug("%s: cpufreq BOOST %s\n",
613 __func__, enable ? "enabled" : "disabled");
614
615 return count;
616}
617define_one_global_rw(boost);
618
619static struct cpufreq_governor *find_governor(const char *str_governor)
620{
621 struct cpufreq_governor *t;
622
623 for_each_governor(t)
624 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
625 return t;
626
627 return NULL;
628}
629
630static struct cpufreq_governor *get_governor(const char *str_governor)
631{
632 struct cpufreq_governor *t;
633
634 mutex_lock(&cpufreq_governor_mutex);
635 t = find_governor(str_governor);
636 if (!t)
637 goto unlock;
638
639 if (!try_module_get(t->owner))
640 t = NULL;
641
642unlock:
643 mutex_unlock(&cpufreq_governor_mutex);
644
645 return t;
646}
647
648static unsigned int cpufreq_parse_policy(char *str_governor)
649{
650 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
651 return CPUFREQ_POLICY_PERFORMANCE;
652
653 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
654 return CPUFREQ_POLICY_POWERSAVE;
655
656 return CPUFREQ_POLICY_UNKNOWN;
657}
658
659/**
660 * cpufreq_parse_governor - parse a governor string only for has_target()
661 * @str_governor: Governor name.
662 */
663static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
664{
665 struct cpufreq_governor *t;
666
667 t = get_governor(str_governor);
668 if (t)
669 return t;
670
671 if (request_module("cpufreq_%s", str_governor))
672 return NULL;
673
674 return get_governor(str_governor);
675}
676
677/*
678 * cpufreq_per_cpu_attr_read() / show_##file_name() -
679 * print out cpufreq information
680 *
681 * Write out information from cpufreq_driver->policy[cpu]; object must be
682 * "unsigned int".
683 */
684
685#define show_one(file_name, object) \
686static ssize_t show_##file_name \
687(struct cpufreq_policy *policy, char *buf) \
688{ \
689 return sprintf(buf, "%u\n", policy->object); \
690}
691
692show_one(cpuinfo_min_freq, cpuinfo.min_freq);
693show_one(cpuinfo_max_freq, cpuinfo.max_freq);
694show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
695show_one(scaling_min_freq, min);
696show_one(scaling_max_freq, max);
697
698__weak unsigned int arch_freq_get_on_cpu(int cpu)
699{
700 return 0;
701}
702
703static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
704{
705 ssize_t ret;
706 unsigned int freq;
707
708 freq = arch_freq_get_on_cpu(policy->cpu);
709 if (freq)
710 ret = sprintf(buf, "%u\n", freq);
711 else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
712 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
713 else
714 ret = sprintf(buf, "%u\n", policy->cur);
715 return ret;
716}
717
718/*
719 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
720 */
721#define store_one(file_name, object) \
722static ssize_t store_##file_name \
723(struct cpufreq_policy *policy, const char *buf, size_t count) \
724{ \
725 unsigned long val; \
726 int ret; \
727 \
728 ret = sscanf(buf, "%lu", &val); \
729 if (ret != 1) \
730 return -EINVAL; \
731 \
732 ret = freq_qos_update_request(policy->object##_freq_req, val);\
733 return ret >= 0 ? count : ret; \
734}
735
736store_one(scaling_min_freq, min);
737store_one(scaling_max_freq, max);
738
739/*
740 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
741 */
742static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
743 char *buf)
744{
745 unsigned int cur_freq = __cpufreq_get(policy);
746
747 if (cur_freq)
748 return sprintf(buf, "%u\n", cur_freq);
749
750 return sprintf(buf, "<unknown>\n");
751}
752
753/*
754 * show_scaling_governor - show the current policy for the specified CPU
755 */
756static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
757{
758 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
759 return sprintf(buf, "powersave\n");
760 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
761 return sprintf(buf, "performance\n");
762 else if (policy->governor)
763 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
764 policy->governor->name);
765 return -EINVAL;
766}
767
768/*
769 * store_scaling_governor - store policy for the specified CPU
770 */
771static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
772 const char *buf, size_t count)
773{
774 char str_governor[16];
775 int ret;
776
777 ret = sscanf(buf, "%15s", str_governor);
778 if (ret != 1)
779 return -EINVAL;
780
781 if (cpufreq_driver->setpolicy) {
782 unsigned int new_pol;
783
784 new_pol = cpufreq_parse_policy(str_governor);
785 if (!new_pol)
786 return -EINVAL;
787
788 ret = cpufreq_set_policy(policy, NULL, new_pol);
789 } else {
790 struct cpufreq_governor *new_gov;
791
792 new_gov = cpufreq_parse_governor(str_governor);
793 if (!new_gov)
794 return -EINVAL;
795
796 ret = cpufreq_set_policy(policy, new_gov,
797 CPUFREQ_POLICY_UNKNOWN);
798
799 module_put(new_gov->owner);
800 }
801
802 return ret ? ret : count;
803}
804
805/*
806 * show_scaling_driver - show the cpufreq driver currently loaded
807 */
808static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
809{
810 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
811}
812
813/*
814 * show_scaling_available_governors - show the available CPUfreq governors
815 */
816static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
817 char *buf)
818{
819 ssize_t i = 0;
820 struct cpufreq_governor *t;
821
822 if (!has_target()) {
823 i += sprintf(buf, "performance powersave");
824 goto out;
825 }
826
827 mutex_lock(&cpufreq_governor_mutex);
828 for_each_governor(t) {
829 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
830 - (CPUFREQ_NAME_LEN + 2)))
831 break;
832 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
833 }
834 mutex_unlock(&cpufreq_governor_mutex);
835out:
836 i += sprintf(&buf[i], "\n");
837 return i;
838}
839
840ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
841{
842 ssize_t i = 0;
843 unsigned int cpu;
844
845 for_each_cpu(cpu, mask) {
846 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u ", cpu);
847 if (i >= (PAGE_SIZE - 5))
848 break;
849 }
850
851 /* Remove the extra space at the end */
852 i--;
853
854 i += sprintf(&buf[i], "\n");
855 return i;
856}
857EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
858
859/*
860 * show_related_cpus - show the CPUs affected by each transition even if
861 * hw coordination is in use
862 */
863static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
864{
865 return cpufreq_show_cpus(policy->related_cpus, buf);
866}
867
868/*
869 * show_affected_cpus - show the CPUs affected by each transition
870 */
871static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
872{
873 return cpufreq_show_cpus(policy->cpus, buf);
874}
875
876static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
877 const char *buf, size_t count)
878{
879 unsigned int freq = 0;
880 unsigned int ret;
881
882 if (!policy->governor || !policy->governor->store_setspeed)
883 return -EINVAL;
884
885 ret = sscanf(buf, "%u", &freq);
886 if (ret != 1)
887 return -EINVAL;
888
889 policy->governor->store_setspeed(policy, freq);
890
891 return count;
892}
893
894static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
895{
896 if (!policy->governor || !policy->governor->show_setspeed)
897 return sprintf(buf, "<unsupported>\n");
898
899 return policy->governor->show_setspeed(policy, buf);
900}
901
902/*
903 * show_bios_limit - show the current cpufreq HW/BIOS limitation
904 */
905static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
906{
907 unsigned int limit;
908 int ret;
909 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
910 if (!ret)
911 return sprintf(buf, "%u\n", limit);
912 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
913}
914
915cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
916cpufreq_freq_attr_ro(cpuinfo_min_freq);
917cpufreq_freq_attr_ro(cpuinfo_max_freq);
918cpufreq_freq_attr_ro(cpuinfo_transition_latency);
919cpufreq_freq_attr_ro(scaling_available_governors);
920cpufreq_freq_attr_ro(scaling_driver);
921cpufreq_freq_attr_ro(scaling_cur_freq);
922cpufreq_freq_attr_ro(bios_limit);
923cpufreq_freq_attr_ro(related_cpus);
924cpufreq_freq_attr_ro(affected_cpus);
925cpufreq_freq_attr_rw(scaling_min_freq);
926cpufreq_freq_attr_rw(scaling_max_freq);
927cpufreq_freq_attr_rw(scaling_governor);
928cpufreq_freq_attr_rw(scaling_setspeed);
929
930static struct attribute *cpufreq_attrs[] = {
931 &cpuinfo_min_freq.attr,
932 &cpuinfo_max_freq.attr,
933 &cpuinfo_transition_latency.attr,
934 &scaling_min_freq.attr,
935 &scaling_max_freq.attr,
936 &affected_cpus.attr,
937 &related_cpus.attr,
938 &scaling_governor.attr,
939 &scaling_driver.attr,
940 &scaling_available_governors.attr,
941 &scaling_setspeed.attr,
942 NULL
943};
944ATTRIBUTE_GROUPS(cpufreq);
945
946#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
947#define to_attr(a) container_of(a, struct freq_attr, attr)
948
949static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
950{
951 struct cpufreq_policy *policy = to_policy(kobj);
952 struct freq_attr *fattr = to_attr(attr);
953 ssize_t ret = -EBUSY;
954
955 if (!fattr->show)
956 return -EIO;
957
958 down_read(&policy->rwsem);
959 if (likely(!policy_is_inactive(policy)))
960 ret = fattr->show(policy, buf);
961 up_read(&policy->rwsem);
962
963 return ret;
964}
965
966static ssize_t store(struct kobject *kobj, struct attribute *attr,
967 const char *buf, size_t count)
968{
969 struct cpufreq_policy *policy = to_policy(kobj);
970 struct freq_attr *fattr = to_attr(attr);
971 ssize_t ret = -EBUSY;
972
973 if (!fattr->store)
974 return -EIO;
975
976 down_write(&policy->rwsem);
977 if (likely(!policy_is_inactive(policy)))
978 ret = fattr->store(policy, buf, count);
979 up_write(&policy->rwsem);
980
981 return ret;
982}
983
984static void cpufreq_sysfs_release(struct kobject *kobj)
985{
986 struct cpufreq_policy *policy = to_policy(kobj);
987 pr_debug("last reference is dropped\n");
988 complete(&policy->kobj_unregister);
989}
990
991static const struct sysfs_ops sysfs_ops = {
992 .show = show,
993 .store = store,
994};
995
996static struct kobj_type ktype_cpufreq = {
997 .sysfs_ops = &sysfs_ops,
998 .default_groups = cpufreq_groups,
999 .release = cpufreq_sysfs_release,
1000};
1001
1002static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
1003 struct device *dev)
1004{
1005 if (unlikely(!dev))
1006 return;
1007
1008 if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1009 return;
1010
1011 dev_dbg(dev, "%s: Adding symlink\n", __func__);
1012 if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1013 dev_err(dev, "cpufreq symlink creation failed\n");
1014}
1015
1016static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu,
1017 struct device *dev)
1018{
1019 dev_dbg(dev, "%s: Removing symlink\n", __func__);
1020 sysfs_remove_link(&dev->kobj, "cpufreq");
1021 cpumask_clear_cpu(cpu, policy->real_cpus);
1022}
1023
1024static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1025{
1026 struct freq_attr **drv_attr;
1027 int ret = 0;
1028
1029 /* set up files for this cpu device */
1030 drv_attr = cpufreq_driver->attr;
1031 while (drv_attr && *drv_attr) {
1032 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1033 if (ret)
1034 return ret;
1035 drv_attr++;
1036 }
1037 if (cpufreq_driver->get) {
1038 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1039 if (ret)
1040 return ret;
1041 }
1042
1043 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1044 if (ret)
1045 return ret;
1046
1047 if (cpufreq_driver->bios_limit) {
1048 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1049 if (ret)
1050 return ret;
1051 }
1052
1053 return 0;
1054}
1055
1056static int cpufreq_init_policy(struct cpufreq_policy *policy)
1057{
1058 struct cpufreq_governor *gov = NULL;
1059 unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1060 int ret;
1061
1062 if (has_target()) {
1063 /* Update policy governor to the one used before hotplug. */
1064 gov = get_governor(policy->last_governor);
1065 if (gov) {
1066 pr_debug("Restoring governor %s for cpu %d\n",
1067 gov->name, policy->cpu);
1068 } else {
1069 gov = get_governor(default_governor);
1070 }
1071
1072 if (!gov) {
1073 gov = cpufreq_default_governor();
1074 __module_get(gov->owner);
1075 }
1076
1077 } else {
1078
1079 /* Use the default policy if there is no last_policy. */
1080 if (policy->last_policy) {
1081 pol = policy->last_policy;
1082 } else {
1083 pol = cpufreq_parse_policy(default_governor);
1084 /*
1085 * In case the default governor is neither "performance"
1086 * nor "powersave", fall back to the initial policy
1087 * value set by the driver.
1088 */
1089 if (pol == CPUFREQ_POLICY_UNKNOWN)
1090 pol = policy->policy;
1091 }
1092 if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1093 pol != CPUFREQ_POLICY_POWERSAVE)
1094 return -ENODATA;
1095 }
1096
1097 ret = cpufreq_set_policy(policy, gov, pol);
1098 if (gov)
1099 module_put(gov->owner);
1100
1101 return ret;
1102}
1103
1104static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1105{
1106 int ret = 0;
1107
1108 /* Has this CPU been taken care of already? */
1109 if (cpumask_test_cpu(cpu, policy->cpus))
1110 return 0;
1111
1112 down_write(&policy->rwsem);
1113 if (has_target())
1114 cpufreq_stop_governor(policy);
1115
1116 cpumask_set_cpu(cpu, policy->cpus);
1117
1118 if (has_target()) {
1119 ret = cpufreq_start_governor(policy);
1120 if (ret)
1121 pr_err("%s: Failed to start governor\n", __func__);
1122 }
1123 up_write(&policy->rwsem);
1124 return ret;
1125}
1126
1127void refresh_frequency_limits(struct cpufreq_policy *policy)
1128{
1129 if (!policy_is_inactive(policy)) {
1130 pr_debug("updating policy for CPU %u\n", policy->cpu);
1131
1132 cpufreq_set_policy(policy, policy->governor, policy->policy);
1133 }
1134}
1135EXPORT_SYMBOL(refresh_frequency_limits);
1136
1137static void handle_update(struct work_struct *work)
1138{
1139 struct cpufreq_policy *policy =
1140 container_of(work, struct cpufreq_policy, update);
1141
1142 pr_debug("handle_update for cpu %u called\n", policy->cpu);
1143 down_write(&policy->rwsem);
1144 refresh_frequency_limits(policy);
1145 up_write(&policy->rwsem);
1146}
1147
1148static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1149 void *data)
1150{
1151 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1152
1153 schedule_work(&policy->update);
1154 return 0;
1155}
1156
1157static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1158 void *data)
1159{
1160 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1161
1162 schedule_work(&policy->update);
1163 return 0;
1164}
1165
1166static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1167{
1168 struct kobject *kobj;
1169 struct completion *cmp;
1170
1171 down_write(&policy->rwsem);
1172 cpufreq_stats_free_table(policy);
1173 kobj = &policy->kobj;
1174 cmp = &policy->kobj_unregister;
1175 up_write(&policy->rwsem);
1176 kobject_put(kobj);
1177
1178 /*
1179 * We need to make sure that the underlying kobj is
1180 * actually not referenced anymore by anybody before we
1181 * proceed with unloading.
1182 */
1183 pr_debug("waiting for dropping of refcount\n");
1184 wait_for_completion(cmp);
1185 pr_debug("wait complete\n");
1186}
1187
1188static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1189{
1190 struct cpufreq_policy *policy;
1191 struct device *dev = get_cpu_device(cpu);
1192 int ret;
1193
1194 if (!dev)
1195 return NULL;
1196
1197 policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1198 if (!policy)
1199 return NULL;
1200
1201 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1202 goto err_free_policy;
1203
1204 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1205 goto err_free_cpumask;
1206
1207 if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1208 goto err_free_rcpumask;
1209
1210 init_completion(&policy->kobj_unregister);
1211 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1212 cpufreq_global_kobject, "policy%u", cpu);
1213 if (ret) {
1214 dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1215 /*
1216 * The entire policy object will be freed below, but the extra
1217 * memory allocated for the kobject name needs to be freed by
1218 * releasing the kobject.
1219 */
1220 kobject_put(&policy->kobj);
1221 goto err_free_real_cpus;
1222 }
1223
1224 freq_constraints_init(&policy->constraints);
1225
1226 policy->nb_min.notifier_call = cpufreq_notifier_min;
1227 policy->nb_max.notifier_call = cpufreq_notifier_max;
1228
1229 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1230 &policy->nb_min);
1231 if (ret) {
1232 dev_err(dev, "Failed to register MIN QoS notifier: %d (%*pbl)\n",
1233 ret, cpumask_pr_args(policy->cpus));
1234 goto err_kobj_remove;
1235 }
1236
1237 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1238 &policy->nb_max);
1239 if (ret) {
1240 dev_err(dev, "Failed to register MAX QoS notifier: %d (%*pbl)\n",
1241 ret, cpumask_pr_args(policy->cpus));
1242 goto err_min_qos_notifier;
1243 }
1244
1245 INIT_LIST_HEAD(&policy->policy_list);
1246 init_rwsem(&policy->rwsem);
1247 spin_lock_init(&policy->transition_lock);
1248 init_waitqueue_head(&policy->transition_wait);
1249 INIT_WORK(&policy->update, handle_update);
1250
1251 policy->cpu = cpu;
1252 return policy;
1253
1254err_min_qos_notifier:
1255 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1256 &policy->nb_min);
1257err_kobj_remove:
1258 cpufreq_policy_put_kobj(policy);
1259err_free_real_cpus:
1260 free_cpumask_var(policy->real_cpus);
1261err_free_rcpumask:
1262 free_cpumask_var(policy->related_cpus);
1263err_free_cpumask:
1264 free_cpumask_var(policy->cpus);
1265err_free_policy:
1266 kfree(policy);
1267
1268 return NULL;
1269}
1270
1271static void cpufreq_policy_free(struct cpufreq_policy *policy)
1272{
1273 unsigned long flags;
1274 int cpu;
1275
1276 /*
1277 * The callers must ensure the policy is inactive by now, to avoid any
1278 * races with show()/store() callbacks.
1279 */
1280 if (unlikely(!policy_is_inactive(policy)))
1281 pr_warn("%s: Freeing active policy\n", __func__);
1282
1283 /* Remove policy from list */
1284 write_lock_irqsave(&cpufreq_driver_lock, flags);
1285 list_del(&policy->policy_list);
1286
1287 for_each_cpu(cpu, policy->related_cpus)
1288 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1289 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1290
1291 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1292 &policy->nb_max);
1293 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1294 &policy->nb_min);
1295
1296 /* Cancel any pending policy->update work before freeing the policy. */
1297 cancel_work_sync(&policy->update);
1298
1299 if (policy->max_freq_req) {
1300 /*
1301 * Remove max_freq_req after sending CPUFREQ_REMOVE_POLICY
1302 * notification, since CPUFREQ_CREATE_POLICY notification was
1303 * sent after adding max_freq_req earlier.
1304 */
1305 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1306 CPUFREQ_REMOVE_POLICY, policy);
1307 freq_qos_remove_request(policy->max_freq_req);
1308 }
1309
1310 freq_qos_remove_request(policy->min_freq_req);
1311 kfree(policy->min_freq_req);
1312
1313 cpufreq_policy_put_kobj(policy);
1314 free_cpumask_var(policy->real_cpus);
1315 free_cpumask_var(policy->related_cpus);
1316 free_cpumask_var(policy->cpus);
1317 kfree(policy);
1318}
1319
1320static int cpufreq_online(unsigned int cpu)
1321{
1322 struct cpufreq_policy *policy;
1323 bool new_policy;
1324 unsigned long flags;
1325 unsigned int j;
1326 int ret;
1327
1328 pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1329
1330 /* Check if this CPU already has a policy to manage it */
1331 policy = per_cpu(cpufreq_cpu_data, cpu);
1332 if (policy) {
1333 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1334 if (!policy_is_inactive(policy))
1335 return cpufreq_add_policy_cpu(policy, cpu);
1336
1337 /* This is the only online CPU for the policy. Start over. */
1338 new_policy = false;
1339 down_write(&policy->rwsem);
1340 policy->cpu = cpu;
1341 policy->governor = NULL;
1342 } else {
1343 new_policy = true;
1344 policy = cpufreq_policy_alloc(cpu);
1345 if (!policy)
1346 return -ENOMEM;
1347 down_write(&policy->rwsem);
1348 }
1349
1350 if (!new_policy && cpufreq_driver->online) {
1351 /* Recover policy->cpus using related_cpus */
1352 cpumask_copy(policy->cpus, policy->related_cpus);
1353
1354 ret = cpufreq_driver->online(policy);
1355 if (ret) {
1356 pr_debug("%s: %d: initialization failed\n", __func__,
1357 __LINE__);
1358 goto out_exit_policy;
1359 }
1360 } else {
1361 cpumask_copy(policy->cpus, cpumask_of(cpu));
1362
1363 /*
1364 * Call driver. From then on the cpufreq must be able
1365 * to accept all calls to ->verify and ->setpolicy for this CPU.
1366 */
1367 ret = cpufreq_driver->init(policy);
1368 if (ret) {
1369 pr_debug("%s: %d: initialization failed\n", __func__,
1370 __LINE__);
1371 goto out_free_policy;
1372 }
1373
1374 /*
1375 * The initialization has succeeded and the policy is online.
1376 * If there is a problem with its frequency table, take it
1377 * offline and drop it.
1378 */
1379 ret = cpufreq_table_validate_and_sort(policy);
1380 if (ret)
1381 goto out_offline_policy;
1382
1383 /* related_cpus should at least include policy->cpus. */
1384 cpumask_copy(policy->related_cpus, policy->cpus);
1385 }
1386
1387 /*
1388 * affected cpus must always be the one, which are online. We aren't
1389 * managing offline cpus here.
1390 */
1391 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1392
1393 if (new_policy) {
1394 for_each_cpu(j, policy->related_cpus) {
1395 per_cpu(cpufreq_cpu_data, j) = policy;
1396 add_cpu_dev_symlink(policy, j, get_cpu_device(j));
1397 }
1398
1399 policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1400 GFP_KERNEL);
1401 if (!policy->min_freq_req) {
1402 ret = -ENOMEM;
1403 goto out_destroy_policy;
1404 }
1405
1406 ret = freq_qos_add_request(&policy->constraints,
1407 policy->min_freq_req, FREQ_QOS_MIN,
1408 FREQ_QOS_MIN_DEFAULT_VALUE);
1409 if (ret < 0) {
1410 /*
1411 * So we don't call freq_qos_remove_request() for an
1412 * uninitialized request.
1413 */
1414 kfree(policy->min_freq_req);
1415 policy->min_freq_req = NULL;
1416 goto out_destroy_policy;
1417 }
1418
1419 /*
1420 * This must be initialized right here to avoid calling
1421 * freq_qos_remove_request() on uninitialized request in case
1422 * of errors.
1423 */
1424 policy->max_freq_req = policy->min_freq_req + 1;
1425
1426 ret = freq_qos_add_request(&policy->constraints,
1427 policy->max_freq_req, FREQ_QOS_MAX,
1428 FREQ_QOS_MAX_DEFAULT_VALUE);
1429 if (ret < 0) {
1430 policy->max_freq_req = NULL;
1431 goto out_destroy_policy;
1432 }
1433
1434 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1435 CPUFREQ_CREATE_POLICY, policy);
1436 }
1437
1438 if (cpufreq_driver->get && has_target()) {
1439 policy->cur = cpufreq_driver->get(policy->cpu);
1440 if (!policy->cur) {
1441 ret = -EIO;
1442 pr_err("%s: ->get() failed\n", __func__);
1443 goto out_destroy_policy;
1444 }
1445 }
1446
1447 /*
1448 * Sometimes boot loaders set CPU frequency to a value outside of
1449 * frequency table present with cpufreq core. In such cases CPU might be
1450 * unstable if it has to run on that frequency for long duration of time
1451 * and so its better to set it to a frequency which is specified in
1452 * freq-table. This also makes cpufreq stats inconsistent as
1453 * cpufreq-stats would fail to register because current frequency of CPU
1454 * isn't found in freq-table.
1455 *
1456 * Because we don't want this change to effect boot process badly, we go
1457 * for the next freq which is >= policy->cur ('cur' must be set by now,
1458 * otherwise we will end up setting freq to lowest of the table as 'cur'
1459 * is initialized to zero).
1460 *
1461 * We are passing target-freq as "policy->cur - 1" otherwise
1462 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1463 * equal to target-freq.
1464 */
1465 if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1466 && has_target()) {
1467 unsigned int old_freq = policy->cur;
1468
1469 /* Are we running at unknown frequency ? */
1470 ret = cpufreq_frequency_table_get_index(policy, old_freq);
1471 if (ret == -EINVAL) {
1472 ret = __cpufreq_driver_target(policy, old_freq - 1,
1473 CPUFREQ_RELATION_L);
1474
1475 /*
1476 * Reaching here after boot in a few seconds may not
1477 * mean that system will remain stable at "unknown"
1478 * frequency for longer duration. Hence, a BUG_ON().
1479 */
1480 BUG_ON(ret);
1481 pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
1482 __func__, policy->cpu, old_freq, policy->cur);
1483 }
1484 }
1485
1486 if (new_policy) {
1487 ret = cpufreq_add_dev_interface(policy);
1488 if (ret)
1489 goto out_destroy_policy;
1490
1491 cpufreq_stats_create_table(policy);
1492
1493 write_lock_irqsave(&cpufreq_driver_lock, flags);
1494 list_add(&policy->policy_list, &cpufreq_policy_list);
1495 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1496
1497 /*
1498 * Register with the energy model before
1499 * sched_cpufreq_governor_change() is called, which will result
1500 * in rebuilding of the sched domains, which should only be done
1501 * once the energy model is properly initialized for the policy
1502 * first.
1503 *
1504 * Also, this should be called before the policy is registered
1505 * with cooling framework.
1506 */
1507 if (cpufreq_driver->register_em)
1508 cpufreq_driver->register_em(policy);
1509 }
1510
1511 ret = cpufreq_init_policy(policy);
1512 if (ret) {
1513 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1514 __func__, cpu, ret);
1515 goto out_destroy_policy;
1516 }
1517
1518 up_write(&policy->rwsem);
1519
1520 kobject_uevent(&policy->kobj, KOBJ_ADD);
1521
1522 /* Callback for handling stuff after policy is ready */
1523 if (cpufreq_driver->ready)
1524 cpufreq_driver->ready(policy);
1525
1526 if (cpufreq_thermal_control_enabled(cpufreq_driver))
1527 policy->cdev = of_cpufreq_cooling_register(policy);
1528
1529 pr_debug("initialization complete\n");
1530
1531 return 0;
1532
1533out_destroy_policy:
1534 for_each_cpu(j, policy->real_cpus)
1535 remove_cpu_dev_symlink(policy, j, get_cpu_device(j));
1536
1537out_offline_policy:
1538 if (cpufreq_driver->offline)
1539 cpufreq_driver->offline(policy);
1540
1541out_exit_policy:
1542 if (cpufreq_driver->exit)
1543 cpufreq_driver->exit(policy);
1544
1545out_free_policy:
1546 cpumask_clear(policy->cpus);
1547 up_write(&policy->rwsem);
1548
1549 cpufreq_policy_free(policy);
1550 return ret;
1551}
1552
1553/**
1554 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1555 * @dev: CPU device.
1556 * @sif: Subsystem interface structure pointer (not used)
1557 */
1558static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1559{
1560 struct cpufreq_policy *policy;
1561 unsigned cpu = dev->id;
1562 int ret;
1563
1564 dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1565
1566 if (cpu_online(cpu)) {
1567 ret = cpufreq_online(cpu);
1568 if (ret)
1569 return ret;
1570 }
1571
1572 /* Create sysfs link on CPU registration */
1573 policy = per_cpu(cpufreq_cpu_data, cpu);
1574 if (policy)
1575 add_cpu_dev_symlink(policy, cpu, dev);
1576
1577 return 0;
1578}
1579
1580static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1581{
1582 int ret;
1583
1584 if (has_target())
1585 cpufreq_stop_governor(policy);
1586
1587 cpumask_clear_cpu(cpu, policy->cpus);
1588
1589 if (!policy_is_inactive(policy)) {
1590 /* Nominate a new CPU if necessary. */
1591 if (cpu == policy->cpu)
1592 policy->cpu = cpumask_any(policy->cpus);
1593
1594 /* Start the governor again for the active policy. */
1595 if (has_target()) {
1596 ret = cpufreq_start_governor(policy);
1597 if (ret)
1598 pr_err("%s: Failed to start governor\n", __func__);
1599 }
1600
1601 return;
1602 }
1603
1604 if (has_target())
1605 strncpy(policy->last_governor, policy->governor->name,
1606 CPUFREQ_NAME_LEN);
1607 else
1608 policy->last_policy = policy->policy;
1609
1610 if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1611 cpufreq_cooling_unregister(policy->cdev);
1612 policy->cdev = NULL;
1613 }
1614
1615 if (has_target())
1616 cpufreq_exit_governor(policy);
1617
1618 /*
1619 * Perform the ->offline() during light-weight tear-down, as
1620 * that allows fast recovery when the CPU comes back.
1621 */
1622 if (cpufreq_driver->offline) {
1623 cpufreq_driver->offline(policy);
1624 } else if (cpufreq_driver->exit) {
1625 cpufreq_driver->exit(policy);
1626 policy->freq_table = NULL;
1627 }
1628}
1629
1630static int cpufreq_offline(unsigned int cpu)
1631{
1632 struct cpufreq_policy *policy;
1633
1634 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1635
1636 policy = cpufreq_cpu_get_raw(cpu);
1637 if (!policy) {
1638 pr_debug("%s: No cpu_data found\n", __func__);
1639 return 0;
1640 }
1641
1642 down_write(&policy->rwsem);
1643
1644 __cpufreq_offline(cpu, policy);
1645
1646 up_write(&policy->rwsem);
1647 return 0;
1648}
1649
1650/*
1651 * cpufreq_remove_dev - remove a CPU device
1652 *
1653 * Removes the cpufreq interface for a CPU device.
1654 */
1655static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1656{
1657 unsigned int cpu = dev->id;
1658 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1659
1660 if (!policy)
1661 return;
1662
1663 down_write(&policy->rwsem);
1664
1665 if (cpu_online(cpu))
1666 __cpufreq_offline(cpu, policy);
1667
1668 remove_cpu_dev_symlink(policy, cpu, dev);
1669
1670 if (!cpumask_empty(policy->real_cpus)) {
1671 up_write(&policy->rwsem);
1672 return;
1673 }
1674
1675 /* We did light-weight exit earlier, do full tear down now */
1676 if (cpufreq_driver->offline)
1677 cpufreq_driver->exit(policy);
1678
1679 up_write(&policy->rwsem);
1680
1681 cpufreq_policy_free(policy);
1682}
1683
1684/**
1685 * cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1686 * @policy: Policy managing CPUs.
1687 * @new_freq: New CPU frequency.
1688 *
1689 * Adjust to the current frequency first and clean up later by either calling
1690 * cpufreq_update_policy(), or scheduling handle_update().
1691 */
1692static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1693 unsigned int new_freq)
1694{
1695 struct cpufreq_freqs freqs;
1696
1697 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1698 policy->cur, new_freq);
1699
1700 freqs.old = policy->cur;
1701 freqs.new = new_freq;
1702
1703 cpufreq_freq_transition_begin(policy, &freqs);
1704 cpufreq_freq_transition_end(policy, &freqs, 0);
1705}
1706
1707static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1708{
1709 unsigned int new_freq;
1710
1711 new_freq = cpufreq_driver->get(policy->cpu);
1712 if (!new_freq)
1713 return 0;
1714
1715 /*
1716 * If fast frequency switching is used with the given policy, the check
1717 * against policy->cur is pointless, so skip it in that case.
1718 */
1719 if (policy->fast_switch_enabled || !has_target())
1720 return new_freq;
1721
1722 if (policy->cur != new_freq) {
1723 /*
1724 * For some platforms, the frequency returned by hardware may be
1725 * slightly different from what is provided in the frequency
1726 * table, for example hardware may return 499 MHz instead of 500
1727 * MHz. In such cases it is better to avoid getting into
1728 * unnecessary frequency updates.
1729 */
1730 if (abs(policy->cur - new_freq) < HZ_PER_MHZ)
1731 return policy->cur;
1732
1733 cpufreq_out_of_sync(policy, new_freq);
1734 if (update)
1735 schedule_work(&policy->update);
1736 }
1737
1738 return new_freq;
1739}
1740
1741/**
1742 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1743 * @cpu: CPU number
1744 *
1745 * This is the last known freq, without actually getting it from the driver.
1746 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1747 */
1748unsigned int cpufreq_quick_get(unsigned int cpu)
1749{
1750 struct cpufreq_policy *policy;
1751 unsigned int ret_freq = 0;
1752 unsigned long flags;
1753
1754 read_lock_irqsave(&cpufreq_driver_lock, flags);
1755
1756 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1757 ret_freq = cpufreq_driver->get(cpu);
1758 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1759 return ret_freq;
1760 }
1761
1762 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1763
1764 policy = cpufreq_cpu_get(cpu);
1765 if (policy) {
1766 ret_freq = policy->cur;
1767 cpufreq_cpu_put(policy);
1768 }
1769
1770 return ret_freq;
1771}
1772EXPORT_SYMBOL(cpufreq_quick_get);
1773
1774/**
1775 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1776 * @cpu: CPU number
1777 *
1778 * Just return the max possible frequency for a given CPU.
1779 */
1780unsigned int cpufreq_quick_get_max(unsigned int cpu)
1781{
1782 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1783 unsigned int ret_freq = 0;
1784
1785 if (policy) {
1786 ret_freq = policy->max;
1787 cpufreq_cpu_put(policy);
1788 }
1789
1790 return ret_freq;
1791}
1792EXPORT_SYMBOL(cpufreq_quick_get_max);
1793
1794/**
1795 * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1796 * @cpu: CPU number
1797 *
1798 * The default return value is the max_freq field of cpuinfo.
1799 */
1800__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1801{
1802 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1803 unsigned int ret_freq = 0;
1804
1805 if (policy) {
1806 ret_freq = policy->cpuinfo.max_freq;
1807 cpufreq_cpu_put(policy);
1808 }
1809
1810 return ret_freq;
1811}
1812EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1813
1814static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1815{
1816 if (unlikely(policy_is_inactive(policy)))
1817 return 0;
1818
1819 return cpufreq_verify_current_freq(policy, true);
1820}
1821
1822/**
1823 * cpufreq_get - get the current CPU frequency (in kHz)
1824 * @cpu: CPU number
1825 *
1826 * Get the CPU current (static) CPU frequency
1827 */
1828unsigned int cpufreq_get(unsigned int cpu)
1829{
1830 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1831 unsigned int ret_freq = 0;
1832
1833 if (policy) {
1834 down_read(&policy->rwsem);
1835 if (cpufreq_driver->get)
1836 ret_freq = __cpufreq_get(policy);
1837 up_read(&policy->rwsem);
1838
1839 cpufreq_cpu_put(policy);
1840 }
1841
1842 return ret_freq;
1843}
1844EXPORT_SYMBOL(cpufreq_get);
1845
1846static struct subsys_interface cpufreq_interface = {
1847 .name = "cpufreq",
1848 .subsys = &cpu_subsys,
1849 .add_dev = cpufreq_add_dev,
1850 .remove_dev = cpufreq_remove_dev,
1851};
1852
1853/*
1854 * In case platform wants some specific frequency to be configured
1855 * during suspend..
1856 */
1857int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1858{
1859 int ret;
1860
1861 if (!policy->suspend_freq) {
1862 pr_debug("%s: suspend_freq not defined\n", __func__);
1863 return 0;
1864 }
1865
1866 pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1867 policy->suspend_freq);
1868
1869 ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1870 CPUFREQ_RELATION_H);
1871 if (ret)
1872 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1873 __func__, policy->suspend_freq, ret);
1874
1875 return ret;
1876}
1877EXPORT_SYMBOL(cpufreq_generic_suspend);
1878
1879/**
1880 * cpufreq_suspend() - Suspend CPUFreq governors.
1881 *
1882 * Called during system wide Suspend/Hibernate cycles for suspending governors
1883 * as some platforms can't change frequency after this point in suspend cycle.
1884 * Because some of the devices (like: i2c, regulators, etc) they use for
1885 * changing frequency are suspended quickly after this point.
1886 */
1887void cpufreq_suspend(void)
1888{
1889 struct cpufreq_policy *policy;
1890
1891 if (!cpufreq_driver)
1892 return;
1893
1894 if (!has_target() && !cpufreq_driver->suspend)
1895 goto suspend;
1896
1897 pr_debug("%s: Suspending Governors\n", __func__);
1898
1899 for_each_active_policy(policy) {
1900 if (has_target()) {
1901 down_write(&policy->rwsem);
1902 cpufreq_stop_governor(policy);
1903 up_write(&policy->rwsem);
1904 }
1905
1906 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1907 pr_err("%s: Failed to suspend driver: %s\n", __func__,
1908 cpufreq_driver->name);
1909 }
1910
1911suspend:
1912 cpufreq_suspended = true;
1913}
1914
1915/**
1916 * cpufreq_resume() - Resume CPUFreq governors.
1917 *
1918 * Called during system wide Suspend/Hibernate cycle for resuming governors that
1919 * are suspended with cpufreq_suspend().
1920 */
1921void cpufreq_resume(void)
1922{
1923 struct cpufreq_policy *policy;
1924 int ret;
1925
1926 if (!cpufreq_driver)
1927 return;
1928
1929 if (unlikely(!cpufreq_suspended))
1930 return;
1931
1932 cpufreq_suspended = false;
1933
1934 if (!has_target() && !cpufreq_driver->resume)
1935 return;
1936
1937 pr_debug("%s: Resuming Governors\n", __func__);
1938
1939 for_each_active_policy(policy) {
1940 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1941 pr_err("%s: Failed to resume driver: %p\n", __func__,
1942 policy);
1943 } else if (has_target()) {
1944 down_write(&policy->rwsem);
1945 ret = cpufreq_start_governor(policy);
1946 up_write(&policy->rwsem);
1947
1948 if (ret)
1949 pr_err("%s: Failed to start governor for policy: %p\n",
1950 __func__, policy);
1951 }
1952 }
1953}
1954
1955/**
1956 * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
1957 * @flags: Flags to test against the current cpufreq driver's flags.
1958 *
1959 * Assumes that the driver is there, so callers must ensure that this is the
1960 * case.
1961 */
1962bool cpufreq_driver_test_flags(u16 flags)
1963{
1964 return !!(cpufreq_driver->flags & flags);
1965}
1966
1967/**
1968 * cpufreq_get_current_driver - Return the current driver's name.
1969 *
1970 * Return the name string of the currently registered cpufreq driver or NULL if
1971 * none.
1972 */
1973const char *cpufreq_get_current_driver(void)
1974{
1975 if (cpufreq_driver)
1976 return cpufreq_driver->name;
1977
1978 return NULL;
1979}
1980EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1981
1982/**
1983 * cpufreq_get_driver_data - Return current driver data.
1984 *
1985 * Return the private data of the currently registered cpufreq driver, or NULL
1986 * if no cpufreq driver has been registered.
1987 */
1988void *cpufreq_get_driver_data(void)
1989{
1990 if (cpufreq_driver)
1991 return cpufreq_driver->driver_data;
1992
1993 return NULL;
1994}
1995EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1996
1997/*********************************************************************
1998 * NOTIFIER LISTS INTERFACE *
1999 *********************************************************************/
2000
2001/**
2002 * cpufreq_register_notifier - Register a notifier with cpufreq.
2003 * @nb: notifier function to register.
2004 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2005 *
2006 * Add a notifier to one of two lists: either a list of notifiers that run on
2007 * clock rate changes (once before and once after every transition), or a list
2008 * of notifiers that ron on cpufreq policy changes.
2009 *
2010 * This function may sleep and it has the same return values as
2011 * blocking_notifier_chain_register().
2012 */
2013int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
2014{
2015 int ret;
2016
2017 if (cpufreq_disabled())
2018 return -EINVAL;
2019
2020 switch (list) {
2021 case CPUFREQ_TRANSITION_NOTIFIER:
2022 mutex_lock(&cpufreq_fast_switch_lock);
2023
2024 if (cpufreq_fast_switch_count > 0) {
2025 mutex_unlock(&cpufreq_fast_switch_lock);
2026 return -EBUSY;
2027 }
2028 ret = srcu_notifier_chain_register(
2029 &cpufreq_transition_notifier_list, nb);
2030 if (!ret)
2031 cpufreq_fast_switch_count--;
2032
2033 mutex_unlock(&cpufreq_fast_switch_lock);
2034 break;
2035 case CPUFREQ_POLICY_NOTIFIER:
2036 ret = blocking_notifier_chain_register(
2037 &cpufreq_policy_notifier_list, nb);
2038 break;
2039 default:
2040 ret = -EINVAL;
2041 }
2042
2043 return ret;
2044}
2045EXPORT_SYMBOL(cpufreq_register_notifier);
2046
2047/**
2048 * cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2049 * @nb: notifier block to be unregistered.
2050 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2051 *
2052 * Remove a notifier from one of the cpufreq notifier lists.
2053 *
2054 * This function may sleep and it has the same return values as
2055 * blocking_notifier_chain_unregister().
2056 */
2057int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2058{
2059 int ret;
2060
2061 if (cpufreq_disabled())
2062 return -EINVAL;
2063
2064 switch (list) {
2065 case CPUFREQ_TRANSITION_NOTIFIER:
2066 mutex_lock(&cpufreq_fast_switch_lock);
2067
2068 ret = srcu_notifier_chain_unregister(
2069 &cpufreq_transition_notifier_list, nb);
2070 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2071 cpufreq_fast_switch_count++;
2072
2073 mutex_unlock(&cpufreq_fast_switch_lock);
2074 break;
2075 case CPUFREQ_POLICY_NOTIFIER:
2076 ret = blocking_notifier_chain_unregister(
2077 &cpufreq_policy_notifier_list, nb);
2078 break;
2079 default:
2080 ret = -EINVAL;
2081 }
2082
2083 return ret;
2084}
2085EXPORT_SYMBOL(cpufreq_unregister_notifier);
2086
2087
2088/*********************************************************************
2089 * GOVERNORS *
2090 *********************************************************************/
2091
2092/**
2093 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2094 * @policy: cpufreq policy to switch the frequency for.
2095 * @target_freq: New frequency to set (may be approximate).
2096 *
2097 * Carry out a fast frequency switch without sleeping.
2098 *
2099 * The driver's ->fast_switch() callback invoked by this function must be
2100 * suitable for being called from within RCU-sched read-side critical sections
2101 * and it is expected to select the minimum available frequency greater than or
2102 * equal to @target_freq (CPUFREQ_RELATION_L).
2103 *
2104 * This function must not be called if policy->fast_switch_enabled is unset.
2105 *
2106 * Governors calling this function must guarantee that it will never be invoked
2107 * twice in parallel for the same policy and that it will never be called in
2108 * parallel with either ->target() or ->target_index() for the same policy.
2109 *
2110 * Returns the actual frequency set for the CPU.
2111 *
2112 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2113 * error condition, the hardware configuration must be preserved.
2114 */
2115unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2116 unsigned int target_freq)
2117{
2118 unsigned int freq;
2119 int cpu;
2120
2121 target_freq = clamp_val(target_freq, policy->min, policy->max);
2122 freq = cpufreq_driver->fast_switch(policy, target_freq);
2123
2124 if (!freq)
2125 return 0;
2126
2127 policy->cur = freq;
2128 arch_set_freq_scale(policy->related_cpus, freq,
2129 policy->cpuinfo.max_freq);
2130 cpufreq_stats_record_transition(policy, freq);
2131
2132 if (trace_cpu_frequency_enabled()) {
2133 for_each_cpu(cpu, policy->cpus)
2134 trace_cpu_frequency(freq, cpu);
2135 }
2136
2137 return freq;
2138}
2139EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2140
2141/**
2142 * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2143 * @cpu: Target CPU.
2144 * @min_perf: Minimum (required) performance level (units of @capacity).
2145 * @target_perf: Target (desired) performance level (units of @capacity).
2146 * @capacity: Capacity of the target CPU.
2147 *
2148 * Carry out a fast performance level switch of @cpu without sleeping.
2149 *
2150 * The driver's ->adjust_perf() callback invoked by this function must be
2151 * suitable for being called from within RCU-sched read-side critical sections
2152 * and it is expected to select a suitable performance level equal to or above
2153 * @min_perf and preferably equal to or below @target_perf.
2154 *
2155 * This function must not be called if policy->fast_switch_enabled is unset.
2156 *
2157 * Governors calling this function must guarantee that it will never be invoked
2158 * twice in parallel for the same CPU and that it will never be called in
2159 * parallel with either ->target() or ->target_index() or ->fast_switch() for
2160 * the same CPU.
2161 */
2162void cpufreq_driver_adjust_perf(unsigned int cpu,
2163 unsigned long min_perf,
2164 unsigned long target_perf,
2165 unsigned long capacity)
2166{
2167 cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2168}
2169
2170/**
2171 * cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2172 *
2173 * Return 'true' if the ->adjust_perf callback is present for the
2174 * current driver or 'false' otherwise.
2175 */
2176bool cpufreq_driver_has_adjust_perf(void)
2177{
2178 return !!cpufreq_driver->adjust_perf;
2179}
2180
2181/* Must set freqs->new to intermediate frequency */
2182static int __target_intermediate(struct cpufreq_policy *policy,
2183 struct cpufreq_freqs *freqs, int index)
2184{
2185 int ret;
2186
2187 freqs->new = cpufreq_driver->get_intermediate(policy, index);
2188
2189 /* We don't need to switch to intermediate freq */
2190 if (!freqs->new)
2191 return 0;
2192
2193 pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2194 __func__, policy->cpu, freqs->old, freqs->new);
2195
2196 cpufreq_freq_transition_begin(policy, freqs);
2197 ret = cpufreq_driver->target_intermediate(policy, index);
2198 cpufreq_freq_transition_end(policy, freqs, ret);
2199
2200 if (ret)
2201 pr_err("%s: Failed to change to intermediate frequency: %d\n",
2202 __func__, ret);
2203
2204 return ret;
2205}
2206
2207static int __target_index(struct cpufreq_policy *policy, int index)
2208{
2209 struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2210 unsigned int restore_freq, intermediate_freq = 0;
2211 unsigned int newfreq = policy->freq_table[index].frequency;
2212 int retval = -EINVAL;
2213 bool notify;
2214
2215 if (newfreq == policy->cur)
2216 return 0;
2217
2218 /* Save last value to restore later on errors */
2219 restore_freq = policy->cur;
2220
2221 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2222 if (notify) {
2223 /* Handle switching to intermediate frequency */
2224 if (cpufreq_driver->get_intermediate) {
2225 retval = __target_intermediate(policy, &freqs, index);
2226 if (retval)
2227 return retval;
2228
2229 intermediate_freq = freqs.new;
2230 /* Set old freq to intermediate */
2231 if (intermediate_freq)
2232 freqs.old = freqs.new;
2233 }
2234
2235 freqs.new = newfreq;
2236 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2237 __func__, policy->cpu, freqs.old, freqs.new);
2238
2239 cpufreq_freq_transition_begin(policy, &freqs);
2240 }
2241
2242 retval = cpufreq_driver->target_index(policy, index);
2243 if (retval)
2244 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2245 retval);
2246
2247 if (notify) {
2248 cpufreq_freq_transition_end(policy, &freqs, retval);
2249
2250 /*
2251 * Failed after setting to intermediate freq? Driver should have
2252 * reverted back to initial frequency and so should we. Check
2253 * here for intermediate_freq instead of get_intermediate, in
2254 * case we haven't switched to intermediate freq at all.
2255 */
2256 if (unlikely(retval && intermediate_freq)) {
2257 freqs.old = intermediate_freq;
2258 freqs.new = restore_freq;
2259 cpufreq_freq_transition_begin(policy, &freqs);
2260 cpufreq_freq_transition_end(policy, &freqs, 0);
2261 }
2262 }
2263
2264 return retval;
2265}
2266
2267int __cpufreq_driver_target(struct cpufreq_policy *policy,
2268 unsigned int target_freq,
2269 unsigned int relation)
2270{
2271 unsigned int old_target_freq = target_freq;
2272
2273 if (cpufreq_disabled())
2274 return -ENODEV;
2275
2276 target_freq = __resolve_freq(policy, target_freq, relation);
2277
2278 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2279 policy->cpu, target_freq, relation, old_target_freq);
2280
2281 /*
2282 * This might look like a redundant call as we are checking it again
2283 * after finding index. But it is left intentionally for cases where
2284 * exactly same freq is called again and so we can save on few function
2285 * calls.
2286 */
2287 if (target_freq == policy->cur &&
2288 !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2289 return 0;
2290
2291 if (cpufreq_driver->target) {
2292 /*
2293 * If the driver hasn't setup a single inefficient frequency,
2294 * it's unlikely it knows how to decode CPUFREQ_RELATION_E.
2295 */
2296 if (!policy->efficiencies_available)
2297 relation &= ~CPUFREQ_RELATION_E;
2298
2299 return cpufreq_driver->target(policy, target_freq, relation);
2300 }
2301
2302 if (!cpufreq_driver->target_index)
2303 return -EINVAL;
2304
2305 return __target_index(policy, policy->cached_resolved_idx);
2306}
2307EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2308
2309int cpufreq_driver_target(struct cpufreq_policy *policy,
2310 unsigned int target_freq,
2311 unsigned int relation)
2312{
2313 int ret;
2314
2315 down_write(&policy->rwsem);
2316
2317 ret = __cpufreq_driver_target(policy, target_freq, relation);
2318
2319 up_write(&policy->rwsem);
2320
2321 return ret;
2322}
2323EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2324
2325__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2326{
2327 return NULL;
2328}
2329
2330static int cpufreq_init_governor(struct cpufreq_policy *policy)
2331{
2332 int ret;
2333
2334 /* Don't start any governor operations if we are entering suspend */
2335 if (cpufreq_suspended)
2336 return 0;
2337 /*
2338 * Governor might not be initiated here if ACPI _PPC changed
2339 * notification happened, so check it.
2340 */
2341 if (!policy->governor)
2342 return -EINVAL;
2343
2344 /* Platform doesn't want dynamic frequency switching ? */
2345 if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2346 cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2347 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2348
2349 if (gov) {
2350 pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2351 policy->governor->name, gov->name);
2352 policy->governor = gov;
2353 } else {
2354 return -EINVAL;
2355 }
2356 }
2357
2358 if (!try_module_get(policy->governor->owner))
2359 return -EINVAL;
2360
2361 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2362
2363 if (policy->governor->init) {
2364 ret = policy->governor->init(policy);
2365 if (ret) {
2366 module_put(policy->governor->owner);
2367 return ret;
2368 }
2369 }
2370
2371 policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2372
2373 return 0;
2374}
2375
2376static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2377{
2378 if (cpufreq_suspended || !policy->governor)
2379 return;
2380
2381 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2382
2383 if (policy->governor->exit)
2384 policy->governor->exit(policy);
2385
2386 module_put(policy->governor->owner);
2387}
2388
2389int cpufreq_start_governor(struct cpufreq_policy *policy)
2390{
2391 int ret;
2392
2393 if (cpufreq_suspended)
2394 return 0;
2395
2396 if (!policy->governor)
2397 return -EINVAL;
2398
2399 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2400
2401 if (cpufreq_driver->get)
2402 cpufreq_verify_current_freq(policy, false);
2403
2404 if (policy->governor->start) {
2405 ret = policy->governor->start(policy);
2406 if (ret)
2407 return ret;
2408 }
2409
2410 if (policy->governor->limits)
2411 policy->governor->limits(policy);
2412
2413 return 0;
2414}
2415
2416void cpufreq_stop_governor(struct cpufreq_policy *policy)
2417{
2418 if (cpufreq_suspended || !policy->governor)
2419 return;
2420
2421 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2422
2423 if (policy->governor->stop)
2424 policy->governor->stop(policy);
2425}
2426
2427static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2428{
2429 if (cpufreq_suspended || !policy->governor)
2430 return;
2431
2432 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2433
2434 if (policy->governor->limits)
2435 policy->governor->limits(policy);
2436}
2437
2438int cpufreq_register_governor(struct cpufreq_governor *governor)
2439{
2440 int err;
2441
2442 if (!governor)
2443 return -EINVAL;
2444
2445 if (cpufreq_disabled())
2446 return -ENODEV;
2447
2448 mutex_lock(&cpufreq_governor_mutex);
2449
2450 err = -EBUSY;
2451 if (!find_governor(governor->name)) {
2452 err = 0;
2453 list_add(&governor->governor_list, &cpufreq_governor_list);
2454 }
2455
2456 mutex_unlock(&cpufreq_governor_mutex);
2457 return err;
2458}
2459EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2460
2461void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2462{
2463 struct cpufreq_policy *policy;
2464 unsigned long flags;
2465
2466 if (!governor)
2467 return;
2468
2469 if (cpufreq_disabled())
2470 return;
2471
2472 /* clear last_governor for all inactive policies */
2473 read_lock_irqsave(&cpufreq_driver_lock, flags);
2474 for_each_inactive_policy(policy) {
2475 if (!strcmp(policy->last_governor, governor->name)) {
2476 policy->governor = NULL;
2477 strcpy(policy->last_governor, "\0");
2478 }
2479 }
2480 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2481
2482 mutex_lock(&cpufreq_governor_mutex);
2483 list_del(&governor->governor_list);
2484 mutex_unlock(&cpufreq_governor_mutex);
2485}
2486EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2487
2488
2489/*********************************************************************
2490 * POLICY INTERFACE *
2491 *********************************************************************/
2492
2493/**
2494 * cpufreq_get_policy - get the current cpufreq_policy
2495 * @policy: struct cpufreq_policy into which the current cpufreq_policy
2496 * is written
2497 * @cpu: CPU to find the policy for
2498 *
2499 * Reads the current cpufreq policy.
2500 */
2501int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2502{
2503 struct cpufreq_policy *cpu_policy;
2504 if (!policy)
2505 return -EINVAL;
2506
2507 cpu_policy = cpufreq_cpu_get(cpu);
2508 if (!cpu_policy)
2509 return -EINVAL;
2510
2511 memcpy(policy, cpu_policy, sizeof(*policy));
2512
2513 cpufreq_cpu_put(cpu_policy);
2514 return 0;
2515}
2516EXPORT_SYMBOL(cpufreq_get_policy);
2517
2518/**
2519 * cpufreq_set_policy - Modify cpufreq policy parameters.
2520 * @policy: Policy object to modify.
2521 * @new_gov: Policy governor pointer.
2522 * @new_pol: Policy value (for drivers with built-in governors).
2523 *
2524 * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2525 * limits to be set for the policy, update @policy with the verified limits
2526 * values and either invoke the driver's ->setpolicy() callback (if present) or
2527 * carry out a governor update for @policy. That is, run the current governor's
2528 * ->limits() callback (if @new_gov points to the same object as the one in
2529 * @policy) or replace the governor for @policy with @new_gov.
2530 *
2531 * The cpuinfo part of @policy is not updated by this function.
2532 */
2533static int cpufreq_set_policy(struct cpufreq_policy *policy,
2534 struct cpufreq_governor *new_gov,
2535 unsigned int new_pol)
2536{
2537 struct cpufreq_policy_data new_data;
2538 struct cpufreq_governor *old_gov;
2539 int ret;
2540
2541 memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2542 new_data.freq_table = policy->freq_table;
2543 new_data.cpu = policy->cpu;
2544 /*
2545 * PM QoS framework collects all the requests from users and provide us
2546 * the final aggregated value here.
2547 */
2548 new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2549 new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2550
2551 pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2552 new_data.cpu, new_data.min, new_data.max);
2553
2554 /*
2555 * Verify that the CPU speed can be set within these limits and make sure
2556 * that min <= max.
2557 */
2558 ret = cpufreq_driver->verify(&new_data);
2559 if (ret)
2560 return ret;
2561
2562 /*
2563 * Resolve policy min/max to available frequencies. It ensures
2564 * no frequency resolution will neither overshoot the requested maximum
2565 * nor undershoot the requested minimum.
2566 */
2567 policy->min = new_data.min;
2568 policy->max = new_data.max;
2569 policy->min = __resolve_freq(policy, policy->min, CPUFREQ_RELATION_L);
2570 policy->max = __resolve_freq(policy, policy->max, CPUFREQ_RELATION_H);
2571 trace_cpu_frequency_limits(policy);
2572
2573 policy->cached_target_freq = UINT_MAX;
2574
2575 pr_debug("new min and max freqs are %u - %u kHz\n",
2576 policy->min, policy->max);
2577
2578 if (cpufreq_driver->setpolicy) {
2579 policy->policy = new_pol;
2580 pr_debug("setting range\n");
2581 return cpufreq_driver->setpolicy(policy);
2582 }
2583
2584 if (new_gov == policy->governor) {
2585 pr_debug("governor limits update\n");
2586 cpufreq_governor_limits(policy);
2587 return 0;
2588 }
2589
2590 pr_debug("governor switch\n");
2591
2592 /* save old, working values */
2593 old_gov = policy->governor;
2594 /* end old governor */
2595 if (old_gov) {
2596 cpufreq_stop_governor(policy);
2597 cpufreq_exit_governor(policy);
2598 }
2599
2600 /* start new governor */
2601 policy->governor = new_gov;
2602 ret = cpufreq_init_governor(policy);
2603 if (!ret) {
2604 ret = cpufreq_start_governor(policy);
2605 if (!ret) {
2606 pr_debug("governor change\n");
2607 sched_cpufreq_governor_change(policy, old_gov);
2608 return 0;
2609 }
2610 cpufreq_exit_governor(policy);
2611 }
2612
2613 /* new governor failed, so re-start old one */
2614 pr_debug("starting governor %s failed\n", policy->governor->name);
2615 if (old_gov) {
2616 policy->governor = old_gov;
2617 if (cpufreq_init_governor(policy))
2618 policy->governor = NULL;
2619 else
2620 cpufreq_start_governor(policy);
2621 }
2622
2623 return ret;
2624}
2625
2626/**
2627 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2628 * @cpu: CPU to re-evaluate the policy for.
2629 *
2630 * Update the current frequency for the cpufreq policy of @cpu and use
2631 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2632 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2633 * for the policy in question, among other things.
2634 */
2635void cpufreq_update_policy(unsigned int cpu)
2636{
2637 struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2638
2639 if (!policy)
2640 return;
2641
2642 /*
2643 * BIOS might change freq behind our back
2644 * -> ask driver for current freq and notify governors about a change
2645 */
2646 if (cpufreq_driver->get && has_target() &&
2647 (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2648 goto unlock;
2649
2650 refresh_frequency_limits(policy);
2651
2652unlock:
2653 cpufreq_cpu_release(policy);
2654}
2655EXPORT_SYMBOL(cpufreq_update_policy);
2656
2657/**
2658 * cpufreq_update_limits - Update policy limits for a given CPU.
2659 * @cpu: CPU to update the policy limits for.
2660 *
2661 * Invoke the driver's ->update_limits callback if present or call
2662 * cpufreq_update_policy() for @cpu.
2663 */
2664void cpufreq_update_limits(unsigned int cpu)
2665{
2666 if (cpufreq_driver->update_limits)
2667 cpufreq_driver->update_limits(cpu);
2668 else
2669 cpufreq_update_policy(cpu);
2670}
2671EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2672
2673/*********************************************************************
2674 * BOOST *
2675 *********************************************************************/
2676static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2677{
2678 int ret;
2679
2680 if (!policy->freq_table)
2681 return -ENXIO;
2682
2683 ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2684 if (ret) {
2685 pr_err("%s: Policy frequency update failed\n", __func__);
2686 return ret;
2687 }
2688
2689 ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2690 if (ret < 0)
2691 return ret;
2692
2693 return 0;
2694}
2695
2696int cpufreq_boost_trigger_state(int state)
2697{
2698 struct cpufreq_policy *policy;
2699 unsigned long flags;
2700 int ret = 0;
2701
2702 if (cpufreq_driver->boost_enabled == state)
2703 return 0;
2704
2705 write_lock_irqsave(&cpufreq_driver_lock, flags);
2706 cpufreq_driver->boost_enabled = state;
2707 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2708
2709 cpus_read_lock();
2710 for_each_active_policy(policy) {
2711 ret = cpufreq_driver->set_boost(policy, state);
2712 if (ret)
2713 goto err_reset_state;
2714 }
2715 cpus_read_unlock();
2716
2717 return 0;
2718
2719err_reset_state:
2720 cpus_read_unlock();
2721
2722 write_lock_irqsave(&cpufreq_driver_lock, flags);
2723 cpufreq_driver->boost_enabled = !state;
2724 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2725
2726 pr_err("%s: Cannot %s BOOST\n",
2727 __func__, state ? "enable" : "disable");
2728
2729 return ret;
2730}
2731
2732static bool cpufreq_boost_supported(void)
2733{
2734 return cpufreq_driver->set_boost;
2735}
2736
2737static int create_boost_sysfs_file(void)
2738{
2739 int ret;
2740
2741 ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2742 if (ret)
2743 pr_err("%s: cannot register global BOOST sysfs file\n",
2744 __func__);
2745
2746 return ret;
2747}
2748
2749static void remove_boost_sysfs_file(void)
2750{
2751 if (cpufreq_boost_supported())
2752 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2753}
2754
2755int cpufreq_enable_boost_support(void)
2756{
2757 if (!cpufreq_driver)
2758 return -EINVAL;
2759
2760 if (cpufreq_boost_supported())
2761 return 0;
2762
2763 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2764
2765 /* This will get removed on driver unregister */
2766 return create_boost_sysfs_file();
2767}
2768EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2769
2770int cpufreq_boost_enabled(void)
2771{
2772 return cpufreq_driver->boost_enabled;
2773}
2774EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2775
2776/*********************************************************************
2777 * REGISTER / UNREGISTER CPUFREQ DRIVER *
2778 *********************************************************************/
2779static enum cpuhp_state hp_online;
2780
2781static int cpuhp_cpufreq_online(unsigned int cpu)
2782{
2783 cpufreq_online(cpu);
2784
2785 return 0;
2786}
2787
2788static int cpuhp_cpufreq_offline(unsigned int cpu)
2789{
2790 cpufreq_offline(cpu);
2791
2792 return 0;
2793}
2794
2795/**
2796 * cpufreq_register_driver - register a CPU Frequency driver
2797 * @driver_data: A struct cpufreq_driver containing the values#
2798 * submitted by the CPU Frequency driver.
2799 *
2800 * Registers a CPU Frequency driver to this core code. This code
2801 * returns zero on success, -EEXIST when another driver got here first
2802 * (and isn't unregistered in the meantime).
2803 *
2804 */
2805int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2806{
2807 unsigned long flags;
2808 int ret;
2809
2810 if (cpufreq_disabled())
2811 return -ENODEV;
2812
2813 /*
2814 * The cpufreq core depends heavily on the availability of device
2815 * structure, make sure they are available before proceeding further.
2816 */
2817 if (!get_cpu_device(0))
2818 return -EPROBE_DEFER;
2819
2820 if (!driver_data || !driver_data->verify || !driver_data->init ||
2821 !(driver_data->setpolicy || driver_data->target_index ||
2822 driver_data->target) ||
2823 (driver_data->setpolicy && (driver_data->target_index ||
2824 driver_data->target)) ||
2825 (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2826 (!driver_data->online != !driver_data->offline))
2827 return -EINVAL;
2828
2829 pr_debug("trying to register driver %s\n", driver_data->name);
2830
2831 /* Protect against concurrent CPU online/offline. */
2832 cpus_read_lock();
2833
2834 write_lock_irqsave(&cpufreq_driver_lock, flags);
2835 if (cpufreq_driver) {
2836 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2837 ret = -EEXIST;
2838 goto out;
2839 }
2840 cpufreq_driver = driver_data;
2841 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2842
2843 /*
2844 * Mark support for the scheduler's frequency invariance engine for
2845 * drivers that implement target(), target_index() or fast_switch().
2846 */
2847 if (!cpufreq_driver->setpolicy) {
2848 static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2849 pr_debug("supports frequency invariance");
2850 }
2851
2852 if (driver_data->setpolicy)
2853 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2854
2855 if (cpufreq_boost_supported()) {
2856 ret = create_boost_sysfs_file();
2857 if (ret)
2858 goto err_null_driver;
2859 }
2860
2861 ret = subsys_interface_register(&cpufreq_interface);
2862 if (ret)
2863 goto err_boost_unreg;
2864
2865 if (unlikely(list_empty(&cpufreq_policy_list))) {
2866 /* if all ->init() calls failed, unregister */
2867 ret = -ENODEV;
2868 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2869 driver_data->name);
2870 goto err_if_unreg;
2871 }
2872
2873 ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2874 "cpufreq:online",
2875 cpuhp_cpufreq_online,
2876 cpuhp_cpufreq_offline);
2877 if (ret < 0)
2878 goto err_if_unreg;
2879 hp_online = ret;
2880 ret = 0;
2881
2882 pr_debug("driver %s up and running\n", driver_data->name);
2883 goto out;
2884
2885err_if_unreg:
2886 subsys_interface_unregister(&cpufreq_interface);
2887err_boost_unreg:
2888 remove_boost_sysfs_file();
2889err_null_driver:
2890 write_lock_irqsave(&cpufreq_driver_lock, flags);
2891 cpufreq_driver = NULL;
2892 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2893out:
2894 cpus_read_unlock();
2895 return ret;
2896}
2897EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2898
2899/*
2900 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2901 *
2902 * Unregister the current CPUFreq driver. Only call this if you have
2903 * the right to do so, i.e. if you have succeeded in initialising before!
2904 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2905 * currently not initialised.
2906 */
2907int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2908{
2909 unsigned long flags;
2910
2911 if (!cpufreq_driver || (driver != cpufreq_driver))
2912 return -EINVAL;
2913
2914 pr_debug("unregistering driver %s\n", driver->name);
2915
2916 /* Protect against concurrent cpu hotplug */
2917 cpus_read_lock();
2918 subsys_interface_unregister(&cpufreq_interface);
2919 remove_boost_sysfs_file();
2920 static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
2921 cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2922
2923 write_lock_irqsave(&cpufreq_driver_lock, flags);
2924
2925 cpufreq_driver = NULL;
2926
2927 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2928 cpus_read_unlock();
2929
2930 return 0;
2931}
2932EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2933
2934static int __init cpufreq_core_init(void)
2935{
2936 struct cpufreq_governor *gov = cpufreq_default_governor();
2937
2938 if (cpufreq_disabled())
2939 return -ENODEV;
2940
2941 cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2942 BUG_ON(!cpufreq_global_kobject);
2943
2944 if (!strlen(default_governor))
2945 strncpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
2946
2947 return 0;
2948}
2949module_param(off, int, 0444);
2950module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
2951core_initcall(cpufreq_core_init);