Loading...
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
76bool has_target_index(void)
77{
78 return !!cpufreq_driver->target_index;
79}
80
81/* internal prototypes */
82static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
83static int cpufreq_init_governor(struct cpufreq_policy *policy);
84static void cpufreq_exit_governor(struct cpufreq_policy *policy);
85static void cpufreq_governor_limits(struct cpufreq_policy *policy);
86static int cpufreq_set_policy(struct cpufreq_policy *policy,
87 struct cpufreq_governor *new_gov,
88 unsigned int new_pol);
89static bool cpufreq_boost_supported(void);
90
91/*
92 * Two notifier lists: the "policy" list is involved in the
93 * validation process for a new CPU frequency policy; the
94 * "transition" list for kernel code that needs to handle
95 * changes to devices when the CPU clock speed changes.
96 * The mutex locks both lists.
97 */
98static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
99SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
100
101static int off __read_mostly;
102static int cpufreq_disabled(void)
103{
104 return off;
105}
106void disable_cpufreq(void)
107{
108 off = 1;
109}
110static DEFINE_MUTEX(cpufreq_governor_mutex);
111
112bool have_governor_per_policy(void)
113{
114 return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
115}
116EXPORT_SYMBOL_GPL(have_governor_per_policy);
117
118static struct kobject *cpufreq_global_kobject;
119
120struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
121{
122 if (have_governor_per_policy())
123 return &policy->kobj;
124 else
125 return cpufreq_global_kobject;
126}
127EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
128
129static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
130{
131 struct kernel_cpustat kcpustat;
132 u64 cur_wall_time;
133 u64 idle_time;
134 u64 busy_time;
135
136 cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
137
138 kcpustat_cpu_fetch(&kcpustat, cpu);
139
140 busy_time = kcpustat.cpustat[CPUTIME_USER];
141 busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
142 busy_time += kcpustat.cpustat[CPUTIME_IRQ];
143 busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
144 busy_time += kcpustat.cpustat[CPUTIME_STEAL];
145 busy_time += kcpustat.cpustat[CPUTIME_NICE];
146
147 idle_time = cur_wall_time - busy_time;
148 if (wall)
149 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
150
151 return div_u64(idle_time, NSEC_PER_USEC);
152}
153
154u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
155{
156 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
157
158 if (idle_time == -1ULL)
159 return get_cpu_idle_time_jiffy(cpu, wall);
160 else if (!io_busy)
161 idle_time += get_cpu_iowait_time_us(cpu, wall);
162
163 return idle_time;
164}
165EXPORT_SYMBOL_GPL(get_cpu_idle_time);
166
167/*
168 * This is a generic cpufreq init() routine which can be used by cpufreq
169 * drivers of SMP systems. It will do following:
170 * - validate & show freq table passed
171 * - set policies transition latency
172 * - policy->cpus with all possible CPUs
173 */
174void cpufreq_generic_init(struct cpufreq_policy *policy,
175 struct cpufreq_frequency_table *table,
176 unsigned int transition_latency)
177{
178 policy->freq_table = table;
179 policy->cpuinfo.transition_latency = transition_latency;
180
181 /*
182 * The driver only supports the SMP configuration where all processors
183 * share the clock and voltage and clock.
184 */
185 cpumask_setall(policy->cpus);
186}
187EXPORT_SYMBOL_GPL(cpufreq_generic_init);
188
189struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
190{
191 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
192
193 return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
194}
195EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
196
197unsigned int cpufreq_generic_get(unsigned int cpu)
198{
199 struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
200
201 if (!policy || IS_ERR(policy->clk)) {
202 pr_err("%s: No %s associated to cpu: %d\n",
203 __func__, policy ? "clk" : "policy", cpu);
204 return 0;
205 }
206
207 return clk_get_rate(policy->clk) / 1000;
208}
209EXPORT_SYMBOL_GPL(cpufreq_generic_get);
210
211/**
212 * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
213 * @cpu: CPU to find the policy for.
214 *
215 * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
216 * the kobject reference counter of that policy. Return a valid policy on
217 * success or NULL on failure.
218 *
219 * The policy returned by this function has to be released with the help of
220 * cpufreq_cpu_put() to balance its kobject reference counter properly.
221 */
222struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
223{
224 struct cpufreq_policy *policy = NULL;
225 unsigned long flags;
226
227 if (WARN_ON(cpu >= nr_cpu_ids))
228 return NULL;
229
230 /* get the cpufreq driver */
231 read_lock_irqsave(&cpufreq_driver_lock, flags);
232
233 if (cpufreq_driver) {
234 /* get the CPU */
235 policy = cpufreq_cpu_get_raw(cpu);
236 if (policy)
237 kobject_get(&policy->kobj);
238 }
239
240 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
241
242 return policy;
243}
244EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
245
246/**
247 * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
248 * @policy: cpufreq policy returned by cpufreq_cpu_get().
249 */
250void cpufreq_cpu_put(struct cpufreq_policy *policy)
251{
252 kobject_put(&policy->kobj);
253}
254EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
255
256/**
257 * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
258 * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
259 */
260void cpufreq_cpu_release(struct cpufreq_policy *policy)
261{
262 if (WARN_ON(!policy))
263 return;
264
265 lockdep_assert_held(&policy->rwsem);
266
267 up_write(&policy->rwsem);
268
269 cpufreq_cpu_put(policy);
270}
271
272/**
273 * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
274 * @cpu: CPU to find the policy for.
275 *
276 * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
277 * if the policy returned by it is not NULL, acquire its rwsem for writing.
278 * Return the policy if it is active or release it and return NULL otherwise.
279 *
280 * The policy returned by this function has to be released with the help of
281 * cpufreq_cpu_release() in order to release its rwsem and balance its usage
282 * counter properly.
283 */
284struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
285{
286 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
287
288 if (!policy)
289 return NULL;
290
291 down_write(&policy->rwsem);
292
293 if (policy_is_inactive(policy)) {
294 cpufreq_cpu_release(policy);
295 return NULL;
296 }
297
298 return policy;
299}
300
301/*********************************************************************
302 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
303 *********************************************************************/
304
305/**
306 * adjust_jiffies - Adjust the system "loops_per_jiffy".
307 * @val: CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
308 * @ci: Frequency change information.
309 *
310 * This function alters the system "loops_per_jiffy" for the clock
311 * speed change. Note that loops_per_jiffy cannot be updated on SMP
312 * systems as each CPU might be scaled differently. So, use the arch
313 * per-CPU loops_per_jiffy value wherever possible.
314 */
315static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
316{
317#ifndef CONFIG_SMP
318 static unsigned long l_p_j_ref;
319 static unsigned int l_p_j_ref_freq;
320
321 if (ci->flags & CPUFREQ_CONST_LOOPS)
322 return;
323
324 if (!l_p_j_ref_freq) {
325 l_p_j_ref = loops_per_jiffy;
326 l_p_j_ref_freq = ci->old;
327 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
328 l_p_j_ref, l_p_j_ref_freq);
329 }
330 if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
331 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
332 ci->new);
333 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
334 loops_per_jiffy, ci->new);
335 }
336#endif
337}
338
339/**
340 * cpufreq_notify_transition - Notify frequency transition and adjust jiffies.
341 * @policy: cpufreq policy to enable fast frequency switching for.
342 * @freqs: contain details of the frequency update.
343 * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
344 *
345 * This function calls the transition notifiers and adjust_jiffies().
346 *
347 * It is called twice on all CPU frequency changes that have external effects.
348 */
349static void cpufreq_notify_transition(struct cpufreq_policy *policy,
350 struct cpufreq_freqs *freqs,
351 unsigned int state)
352{
353 int cpu;
354
355 BUG_ON(irqs_disabled());
356
357 if (cpufreq_disabled())
358 return;
359
360 freqs->policy = policy;
361 freqs->flags = cpufreq_driver->flags;
362 pr_debug("notification %u of frequency transition to %u kHz\n",
363 state, freqs->new);
364
365 switch (state) {
366 case CPUFREQ_PRECHANGE:
367 /*
368 * Detect if the driver reported a value as "old frequency"
369 * which is not equal to what the cpufreq core thinks is
370 * "old frequency".
371 */
372 if (policy->cur && policy->cur != freqs->old) {
373 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
374 freqs->old, policy->cur);
375 freqs->old = policy->cur;
376 }
377
378 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
379 CPUFREQ_PRECHANGE, freqs);
380
381 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
382 break;
383
384 case CPUFREQ_POSTCHANGE:
385 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
386 pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
387 cpumask_pr_args(policy->cpus));
388
389 for_each_cpu(cpu, policy->cpus)
390 trace_cpu_frequency(freqs->new, cpu);
391
392 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
393 CPUFREQ_POSTCHANGE, freqs);
394
395 cpufreq_stats_record_transition(policy, freqs->new);
396 policy->cur = freqs->new;
397 }
398}
399
400/* Do post notifications when there are chances that transition has failed */
401static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
402 struct cpufreq_freqs *freqs, int transition_failed)
403{
404 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
405 if (!transition_failed)
406 return;
407
408 swap(freqs->old, freqs->new);
409 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
410 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
411}
412
413void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
414 struct cpufreq_freqs *freqs)
415{
416
417 /*
418 * Catch double invocations of _begin() which lead to self-deadlock.
419 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
420 * doesn't invoke _begin() on their behalf, and hence the chances of
421 * double invocations are very low. Moreover, there are scenarios
422 * where these checks can emit false-positive warnings in these
423 * drivers; so we avoid that by skipping them altogether.
424 */
425 WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
426 && current == policy->transition_task);
427
428wait:
429 wait_event(policy->transition_wait, !policy->transition_ongoing);
430
431 spin_lock(&policy->transition_lock);
432
433 if (unlikely(policy->transition_ongoing)) {
434 spin_unlock(&policy->transition_lock);
435 goto wait;
436 }
437
438 policy->transition_ongoing = true;
439 policy->transition_task = current;
440
441 spin_unlock(&policy->transition_lock);
442
443 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
444}
445EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
446
447void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
448 struct cpufreq_freqs *freqs, int transition_failed)
449{
450 if (WARN_ON(!policy->transition_ongoing))
451 return;
452
453 cpufreq_notify_post_transition(policy, freqs, transition_failed);
454
455 arch_set_freq_scale(policy->related_cpus,
456 policy->cur,
457 arch_scale_freq_ref(policy->cpu));
458
459 spin_lock(&policy->transition_lock);
460 policy->transition_ongoing = false;
461 policy->transition_task = NULL;
462 spin_unlock(&policy->transition_lock);
463
464 wake_up(&policy->transition_wait);
465}
466EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
467
468/*
469 * Fast frequency switching status count. Positive means "enabled", negative
470 * means "disabled" and 0 means "not decided yet".
471 */
472static int cpufreq_fast_switch_count;
473static DEFINE_MUTEX(cpufreq_fast_switch_lock);
474
475static void cpufreq_list_transition_notifiers(void)
476{
477 struct notifier_block *nb;
478
479 pr_info("Registered transition notifiers:\n");
480
481 mutex_lock(&cpufreq_transition_notifier_list.mutex);
482
483 for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
484 pr_info("%pS\n", nb->notifier_call);
485
486 mutex_unlock(&cpufreq_transition_notifier_list.mutex);
487}
488
489/**
490 * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
491 * @policy: cpufreq policy to enable fast frequency switching for.
492 *
493 * Try to enable fast frequency switching for @policy.
494 *
495 * The attempt will fail if there is at least one transition notifier registered
496 * at this point, as fast frequency switching is quite fundamentally at odds
497 * with transition notifiers. Thus if successful, it will make registration of
498 * transition notifiers fail going forward.
499 */
500void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
501{
502 lockdep_assert_held(&policy->rwsem);
503
504 if (!policy->fast_switch_possible)
505 return;
506
507 mutex_lock(&cpufreq_fast_switch_lock);
508 if (cpufreq_fast_switch_count >= 0) {
509 cpufreq_fast_switch_count++;
510 policy->fast_switch_enabled = true;
511 } else {
512 pr_warn("CPU%u: Fast frequency switching not enabled\n",
513 policy->cpu);
514 cpufreq_list_transition_notifiers();
515 }
516 mutex_unlock(&cpufreq_fast_switch_lock);
517}
518EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
519
520/**
521 * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
522 * @policy: cpufreq policy to disable fast frequency switching for.
523 */
524void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
525{
526 mutex_lock(&cpufreq_fast_switch_lock);
527 if (policy->fast_switch_enabled) {
528 policy->fast_switch_enabled = false;
529 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
530 cpufreq_fast_switch_count--;
531 }
532 mutex_unlock(&cpufreq_fast_switch_lock);
533}
534EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
535
536static unsigned int __resolve_freq(struct cpufreq_policy *policy,
537 unsigned int target_freq, unsigned int relation)
538{
539 unsigned int idx;
540
541 target_freq = clamp_val(target_freq, policy->min, policy->max);
542
543 if (!policy->freq_table)
544 return target_freq;
545
546 idx = cpufreq_frequency_table_target(policy, target_freq, relation);
547 policy->cached_resolved_idx = idx;
548 policy->cached_target_freq = target_freq;
549 return policy->freq_table[idx].frequency;
550}
551
552/**
553 * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
554 * one.
555 * @policy: associated policy to interrogate
556 * @target_freq: target frequency to resolve.
557 *
558 * The target to driver frequency mapping is cached in the policy.
559 *
560 * Return: Lowest driver-supported frequency greater than or equal to the
561 * given target_freq, subject to policy (min/max) and driver limitations.
562 */
563unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
564 unsigned int target_freq)
565{
566 return __resolve_freq(policy, target_freq, CPUFREQ_RELATION_LE);
567}
568EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
569
570unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
571{
572 unsigned int latency;
573
574 if (policy->transition_delay_us)
575 return policy->transition_delay_us;
576
577 latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
578 if (latency) {
579 /*
580 * For platforms that can change the frequency very fast (< 10
581 * us), the above formula gives a decent transition delay. But
582 * for platforms where transition_latency is in milliseconds, it
583 * ends up giving unrealistic values.
584 *
585 * Cap the default transition delay to 10 ms, which seems to be
586 * a reasonable amount of time after which we should reevaluate
587 * the frequency.
588 */
589 return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
590 }
591
592 return LATENCY_MULTIPLIER;
593}
594EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
595
596/*********************************************************************
597 * SYSFS INTERFACE *
598 *********************************************************************/
599static ssize_t show_boost(struct kobject *kobj,
600 struct kobj_attribute *attr, char *buf)
601{
602 return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
603}
604
605static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
606 const char *buf, size_t count)
607{
608 int ret, enable;
609
610 ret = sscanf(buf, "%d", &enable);
611 if (ret != 1 || enable < 0 || enable > 1)
612 return -EINVAL;
613
614 if (cpufreq_boost_trigger_state(enable)) {
615 pr_err("%s: Cannot %s BOOST!\n",
616 __func__, enable ? "enable" : "disable");
617 return -EINVAL;
618 }
619
620 pr_debug("%s: cpufreq BOOST %s\n",
621 __func__, enable ? "enabled" : "disabled");
622
623 return count;
624}
625define_one_global_rw(boost);
626
627static ssize_t show_local_boost(struct cpufreq_policy *policy, char *buf)
628{
629 return sysfs_emit(buf, "%d\n", policy->boost_enabled);
630}
631
632static ssize_t store_local_boost(struct cpufreq_policy *policy,
633 const char *buf, size_t count)
634{
635 int ret, enable;
636
637 ret = kstrtoint(buf, 10, &enable);
638 if (ret || enable < 0 || enable > 1)
639 return -EINVAL;
640
641 if (!cpufreq_driver->boost_enabled)
642 return -EINVAL;
643
644 if (policy->boost_enabled == enable)
645 return count;
646
647 cpus_read_lock();
648 ret = cpufreq_driver->set_boost(policy, enable);
649 cpus_read_unlock();
650
651 if (ret)
652 return ret;
653
654 policy->boost_enabled = enable;
655
656 return count;
657}
658
659static struct freq_attr local_boost = __ATTR(boost, 0644, show_local_boost, store_local_boost);
660
661static struct cpufreq_governor *find_governor(const char *str_governor)
662{
663 struct cpufreq_governor *t;
664
665 for_each_governor(t)
666 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
667 return t;
668
669 return NULL;
670}
671
672static struct cpufreq_governor *get_governor(const char *str_governor)
673{
674 struct cpufreq_governor *t;
675
676 mutex_lock(&cpufreq_governor_mutex);
677 t = find_governor(str_governor);
678 if (!t)
679 goto unlock;
680
681 if (!try_module_get(t->owner))
682 t = NULL;
683
684unlock:
685 mutex_unlock(&cpufreq_governor_mutex);
686
687 return t;
688}
689
690static unsigned int cpufreq_parse_policy(char *str_governor)
691{
692 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
693 return CPUFREQ_POLICY_PERFORMANCE;
694
695 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
696 return CPUFREQ_POLICY_POWERSAVE;
697
698 return CPUFREQ_POLICY_UNKNOWN;
699}
700
701/**
702 * cpufreq_parse_governor - parse a governor string only for has_target()
703 * @str_governor: Governor name.
704 */
705static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
706{
707 struct cpufreq_governor *t;
708
709 t = get_governor(str_governor);
710 if (t)
711 return t;
712
713 if (request_module("cpufreq_%s", str_governor))
714 return NULL;
715
716 return get_governor(str_governor);
717}
718
719/*
720 * cpufreq_per_cpu_attr_read() / show_##file_name() -
721 * print out cpufreq information
722 *
723 * Write out information from cpufreq_driver->policy[cpu]; object must be
724 * "unsigned int".
725 */
726
727#define show_one(file_name, object) \
728static ssize_t show_##file_name \
729(struct cpufreq_policy *policy, char *buf) \
730{ \
731 return sprintf(buf, "%u\n", policy->object); \
732}
733
734show_one(cpuinfo_min_freq, cpuinfo.min_freq);
735show_one(cpuinfo_max_freq, cpuinfo.max_freq);
736show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
737show_one(scaling_min_freq, min);
738show_one(scaling_max_freq, max);
739
740__weak unsigned int arch_freq_get_on_cpu(int cpu)
741{
742 return 0;
743}
744
745static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
746{
747 ssize_t ret;
748 unsigned int freq;
749
750 freq = arch_freq_get_on_cpu(policy->cpu);
751 if (freq)
752 ret = sprintf(buf, "%u\n", freq);
753 else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
754 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
755 else
756 ret = sprintf(buf, "%u\n", policy->cur);
757 return ret;
758}
759
760/*
761 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
762 */
763#define store_one(file_name, object) \
764static ssize_t store_##file_name \
765(struct cpufreq_policy *policy, const char *buf, size_t count) \
766{ \
767 unsigned long val; \
768 int ret; \
769 \
770 ret = kstrtoul(buf, 0, &val); \
771 if (ret) \
772 return ret; \
773 \
774 ret = freq_qos_update_request(policy->object##_freq_req, val);\
775 return ret >= 0 ? count : ret; \
776}
777
778store_one(scaling_min_freq, min);
779store_one(scaling_max_freq, max);
780
781/*
782 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
783 */
784static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
785 char *buf)
786{
787 unsigned int cur_freq = __cpufreq_get(policy);
788
789 if (cur_freq)
790 return sprintf(buf, "%u\n", cur_freq);
791
792 return sprintf(buf, "<unknown>\n");
793}
794
795/*
796 * show_scaling_governor - show the current policy for the specified CPU
797 */
798static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
799{
800 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
801 return sprintf(buf, "powersave\n");
802 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
803 return sprintf(buf, "performance\n");
804 else if (policy->governor)
805 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
806 policy->governor->name);
807 return -EINVAL;
808}
809
810/*
811 * store_scaling_governor - store policy for the specified CPU
812 */
813static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
814 const char *buf, size_t count)
815{
816 char str_governor[16];
817 int ret;
818
819 ret = sscanf(buf, "%15s", str_governor);
820 if (ret != 1)
821 return -EINVAL;
822
823 if (cpufreq_driver->setpolicy) {
824 unsigned int new_pol;
825
826 new_pol = cpufreq_parse_policy(str_governor);
827 if (!new_pol)
828 return -EINVAL;
829
830 ret = cpufreq_set_policy(policy, NULL, new_pol);
831 } else {
832 struct cpufreq_governor *new_gov;
833
834 new_gov = cpufreq_parse_governor(str_governor);
835 if (!new_gov)
836 return -EINVAL;
837
838 ret = cpufreq_set_policy(policy, new_gov,
839 CPUFREQ_POLICY_UNKNOWN);
840
841 module_put(new_gov->owner);
842 }
843
844 return ret ? ret : count;
845}
846
847/*
848 * show_scaling_driver - show the cpufreq driver currently loaded
849 */
850static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
851{
852 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
853}
854
855/*
856 * show_scaling_available_governors - show the available CPUfreq governors
857 */
858static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
859 char *buf)
860{
861 ssize_t i = 0;
862 struct cpufreq_governor *t;
863
864 if (!has_target()) {
865 i += sprintf(buf, "performance powersave");
866 goto out;
867 }
868
869 mutex_lock(&cpufreq_governor_mutex);
870 for_each_governor(t) {
871 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
872 - (CPUFREQ_NAME_LEN + 2)))
873 break;
874 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
875 }
876 mutex_unlock(&cpufreq_governor_mutex);
877out:
878 i += sprintf(&buf[i], "\n");
879 return i;
880}
881
882ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
883{
884 ssize_t i = 0;
885 unsigned int cpu;
886
887 for_each_cpu(cpu, mask) {
888 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u ", cpu);
889 if (i >= (PAGE_SIZE - 5))
890 break;
891 }
892
893 /* Remove the extra space at the end */
894 i--;
895
896 i += sprintf(&buf[i], "\n");
897 return i;
898}
899EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
900
901/*
902 * show_related_cpus - show the CPUs affected by each transition even if
903 * hw coordination is in use
904 */
905static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
906{
907 return cpufreq_show_cpus(policy->related_cpus, buf);
908}
909
910/*
911 * show_affected_cpus - show the CPUs affected by each transition
912 */
913static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
914{
915 return cpufreq_show_cpus(policy->cpus, buf);
916}
917
918static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
919 const char *buf, size_t count)
920{
921 unsigned int freq = 0;
922 unsigned int ret;
923
924 if (!policy->governor || !policy->governor->store_setspeed)
925 return -EINVAL;
926
927 ret = sscanf(buf, "%u", &freq);
928 if (ret != 1)
929 return -EINVAL;
930
931 policy->governor->store_setspeed(policy, freq);
932
933 return count;
934}
935
936static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
937{
938 if (!policy->governor || !policy->governor->show_setspeed)
939 return sprintf(buf, "<unsupported>\n");
940
941 return policy->governor->show_setspeed(policy, buf);
942}
943
944/*
945 * show_bios_limit - show the current cpufreq HW/BIOS limitation
946 */
947static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
948{
949 unsigned int limit;
950 int ret;
951 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
952 if (!ret)
953 return sprintf(buf, "%u\n", limit);
954 return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
955}
956
957cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
958cpufreq_freq_attr_ro(cpuinfo_min_freq);
959cpufreq_freq_attr_ro(cpuinfo_max_freq);
960cpufreq_freq_attr_ro(cpuinfo_transition_latency);
961cpufreq_freq_attr_ro(scaling_available_governors);
962cpufreq_freq_attr_ro(scaling_driver);
963cpufreq_freq_attr_ro(scaling_cur_freq);
964cpufreq_freq_attr_ro(bios_limit);
965cpufreq_freq_attr_ro(related_cpus);
966cpufreq_freq_attr_ro(affected_cpus);
967cpufreq_freq_attr_rw(scaling_min_freq);
968cpufreq_freq_attr_rw(scaling_max_freq);
969cpufreq_freq_attr_rw(scaling_governor);
970cpufreq_freq_attr_rw(scaling_setspeed);
971
972static struct attribute *cpufreq_attrs[] = {
973 &cpuinfo_min_freq.attr,
974 &cpuinfo_max_freq.attr,
975 &cpuinfo_transition_latency.attr,
976 &scaling_min_freq.attr,
977 &scaling_max_freq.attr,
978 &affected_cpus.attr,
979 &related_cpus.attr,
980 &scaling_governor.attr,
981 &scaling_driver.attr,
982 &scaling_available_governors.attr,
983 &scaling_setspeed.attr,
984 NULL
985};
986ATTRIBUTE_GROUPS(cpufreq);
987
988#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
989#define to_attr(a) container_of(a, struct freq_attr, attr)
990
991static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
992{
993 struct cpufreq_policy *policy = to_policy(kobj);
994 struct freq_attr *fattr = to_attr(attr);
995 ssize_t ret = -EBUSY;
996
997 if (!fattr->show)
998 return -EIO;
999
1000 down_read(&policy->rwsem);
1001 if (likely(!policy_is_inactive(policy)))
1002 ret = fattr->show(policy, buf);
1003 up_read(&policy->rwsem);
1004
1005 return ret;
1006}
1007
1008static ssize_t store(struct kobject *kobj, struct attribute *attr,
1009 const char *buf, size_t count)
1010{
1011 struct cpufreq_policy *policy = to_policy(kobj);
1012 struct freq_attr *fattr = to_attr(attr);
1013 ssize_t ret = -EBUSY;
1014
1015 if (!fattr->store)
1016 return -EIO;
1017
1018 down_write(&policy->rwsem);
1019 if (likely(!policy_is_inactive(policy)))
1020 ret = fattr->store(policy, buf, count);
1021 up_write(&policy->rwsem);
1022
1023 return ret;
1024}
1025
1026static void cpufreq_sysfs_release(struct kobject *kobj)
1027{
1028 struct cpufreq_policy *policy = to_policy(kobj);
1029 pr_debug("last reference is dropped\n");
1030 complete(&policy->kobj_unregister);
1031}
1032
1033static const struct sysfs_ops sysfs_ops = {
1034 .show = show,
1035 .store = store,
1036};
1037
1038static const struct kobj_type ktype_cpufreq = {
1039 .sysfs_ops = &sysfs_ops,
1040 .default_groups = cpufreq_groups,
1041 .release = cpufreq_sysfs_release,
1042};
1043
1044static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
1045 struct device *dev)
1046{
1047 if (unlikely(!dev))
1048 return;
1049
1050 if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1051 return;
1052
1053 dev_dbg(dev, "%s: Adding symlink\n", __func__);
1054 if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1055 dev_err(dev, "cpufreq symlink creation failed\n");
1056}
1057
1058static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu,
1059 struct device *dev)
1060{
1061 dev_dbg(dev, "%s: Removing symlink\n", __func__);
1062 sysfs_remove_link(&dev->kobj, "cpufreq");
1063 cpumask_clear_cpu(cpu, policy->real_cpus);
1064}
1065
1066static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1067{
1068 struct freq_attr **drv_attr;
1069 int ret = 0;
1070
1071 /* set up files for this cpu device */
1072 drv_attr = cpufreq_driver->attr;
1073 while (drv_attr && *drv_attr) {
1074 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1075 if (ret)
1076 return ret;
1077 drv_attr++;
1078 }
1079 if (cpufreq_driver->get) {
1080 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1081 if (ret)
1082 return ret;
1083 }
1084
1085 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1086 if (ret)
1087 return ret;
1088
1089 if (cpufreq_driver->bios_limit) {
1090 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1091 if (ret)
1092 return ret;
1093 }
1094
1095 if (cpufreq_boost_supported()) {
1096 ret = sysfs_create_file(&policy->kobj, &local_boost.attr);
1097 if (ret)
1098 return ret;
1099 }
1100
1101 return 0;
1102}
1103
1104static int cpufreq_init_policy(struct cpufreq_policy *policy)
1105{
1106 struct cpufreq_governor *gov = NULL;
1107 unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1108 int ret;
1109
1110 if (has_target()) {
1111 /* Update policy governor to the one used before hotplug. */
1112 gov = get_governor(policy->last_governor);
1113 if (gov) {
1114 pr_debug("Restoring governor %s for cpu %d\n",
1115 gov->name, policy->cpu);
1116 } else {
1117 gov = get_governor(default_governor);
1118 }
1119
1120 if (!gov) {
1121 gov = cpufreq_default_governor();
1122 __module_get(gov->owner);
1123 }
1124
1125 } else {
1126
1127 /* Use the default policy if there is no last_policy. */
1128 if (policy->last_policy) {
1129 pol = policy->last_policy;
1130 } else {
1131 pol = cpufreq_parse_policy(default_governor);
1132 /*
1133 * In case the default governor is neither "performance"
1134 * nor "powersave", fall back to the initial policy
1135 * value set by the driver.
1136 */
1137 if (pol == CPUFREQ_POLICY_UNKNOWN)
1138 pol = policy->policy;
1139 }
1140 if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1141 pol != CPUFREQ_POLICY_POWERSAVE)
1142 return -ENODATA;
1143 }
1144
1145 ret = cpufreq_set_policy(policy, gov, pol);
1146 if (gov)
1147 module_put(gov->owner);
1148
1149 return ret;
1150}
1151
1152static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1153{
1154 int ret = 0;
1155
1156 /* Has this CPU been taken care of already? */
1157 if (cpumask_test_cpu(cpu, policy->cpus))
1158 return 0;
1159
1160 down_write(&policy->rwsem);
1161 if (has_target())
1162 cpufreq_stop_governor(policy);
1163
1164 cpumask_set_cpu(cpu, policy->cpus);
1165
1166 if (has_target()) {
1167 ret = cpufreq_start_governor(policy);
1168 if (ret)
1169 pr_err("%s: Failed to start governor\n", __func__);
1170 }
1171 up_write(&policy->rwsem);
1172 return ret;
1173}
1174
1175void refresh_frequency_limits(struct cpufreq_policy *policy)
1176{
1177 if (!policy_is_inactive(policy)) {
1178 pr_debug("updating policy for CPU %u\n", policy->cpu);
1179
1180 cpufreq_set_policy(policy, policy->governor, policy->policy);
1181 }
1182}
1183EXPORT_SYMBOL(refresh_frequency_limits);
1184
1185static void handle_update(struct work_struct *work)
1186{
1187 struct cpufreq_policy *policy =
1188 container_of(work, struct cpufreq_policy, update);
1189
1190 pr_debug("handle_update for cpu %u called\n", policy->cpu);
1191 down_write(&policy->rwsem);
1192 refresh_frequency_limits(policy);
1193 up_write(&policy->rwsem);
1194}
1195
1196static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1197 void *data)
1198{
1199 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1200
1201 schedule_work(&policy->update);
1202 return 0;
1203}
1204
1205static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1206 void *data)
1207{
1208 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1209
1210 schedule_work(&policy->update);
1211 return 0;
1212}
1213
1214static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1215{
1216 struct kobject *kobj;
1217 struct completion *cmp;
1218
1219 down_write(&policy->rwsem);
1220 cpufreq_stats_free_table(policy);
1221 kobj = &policy->kobj;
1222 cmp = &policy->kobj_unregister;
1223 up_write(&policy->rwsem);
1224 kobject_put(kobj);
1225
1226 /*
1227 * We need to make sure that the underlying kobj is
1228 * actually not referenced anymore by anybody before we
1229 * proceed with unloading.
1230 */
1231 pr_debug("waiting for dropping of refcount\n");
1232 wait_for_completion(cmp);
1233 pr_debug("wait complete\n");
1234}
1235
1236static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1237{
1238 struct cpufreq_policy *policy;
1239 struct device *dev = get_cpu_device(cpu);
1240 int ret;
1241
1242 if (!dev)
1243 return NULL;
1244
1245 policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1246 if (!policy)
1247 return NULL;
1248
1249 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1250 goto err_free_policy;
1251
1252 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1253 goto err_free_cpumask;
1254
1255 if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1256 goto err_free_rcpumask;
1257
1258 init_completion(&policy->kobj_unregister);
1259 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1260 cpufreq_global_kobject, "policy%u", cpu);
1261 if (ret) {
1262 dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1263 /*
1264 * The entire policy object will be freed below, but the extra
1265 * memory allocated for the kobject name needs to be freed by
1266 * releasing the kobject.
1267 */
1268 kobject_put(&policy->kobj);
1269 goto err_free_real_cpus;
1270 }
1271
1272 freq_constraints_init(&policy->constraints);
1273
1274 policy->nb_min.notifier_call = cpufreq_notifier_min;
1275 policy->nb_max.notifier_call = cpufreq_notifier_max;
1276
1277 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1278 &policy->nb_min);
1279 if (ret) {
1280 dev_err(dev, "Failed to register MIN QoS notifier: %d (CPU%u)\n",
1281 ret, cpu);
1282 goto err_kobj_remove;
1283 }
1284
1285 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1286 &policy->nb_max);
1287 if (ret) {
1288 dev_err(dev, "Failed to register MAX QoS notifier: %d (CPU%u)\n",
1289 ret, cpu);
1290 goto err_min_qos_notifier;
1291 }
1292
1293 INIT_LIST_HEAD(&policy->policy_list);
1294 init_rwsem(&policy->rwsem);
1295 spin_lock_init(&policy->transition_lock);
1296 init_waitqueue_head(&policy->transition_wait);
1297 INIT_WORK(&policy->update, handle_update);
1298
1299 policy->cpu = cpu;
1300 return policy;
1301
1302err_min_qos_notifier:
1303 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1304 &policy->nb_min);
1305err_kobj_remove:
1306 cpufreq_policy_put_kobj(policy);
1307err_free_real_cpus:
1308 free_cpumask_var(policy->real_cpus);
1309err_free_rcpumask:
1310 free_cpumask_var(policy->related_cpus);
1311err_free_cpumask:
1312 free_cpumask_var(policy->cpus);
1313err_free_policy:
1314 kfree(policy);
1315
1316 return NULL;
1317}
1318
1319static void cpufreq_policy_free(struct cpufreq_policy *policy)
1320{
1321 unsigned long flags;
1322 int cpu;
1323
1324 /*
1325 * The callers must ensure the policy is inactive by now, to avoid any
1326 * races with show()/store() callbacks.
1327 */
1328 if (unlikely(!policy_is_inactive(policy)))
1329 pr_warn("%s: Freeing active policy\n", __func__);
1330
1331 /* Remove policy from list */
1332 write_lock_irqsave(&cpufreq_driver_lock, flags);
1333 list_del(&policy->policy_list);
1334
1335 for_each_cpu(cpu, policy->related_cpus)
1336 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1337 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1338
1339 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1340 &policy->nb_max);
1341 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1342 &policy->nb_min);
1343
1344 /* Cancel any pending policy->update work before freeing the policy. */
1345 cancel_work_sync(&policy->update);
1346
1347 if (policy->max_freq_req) {
1348 /*
1349 * Remove max_freq_req after sending CPUFREQ_REMOVE_POLICY
1350 * notification, since CPUFREQ_CREATE_POLICY notification was
1351 * sent after adding max_freq_req earlier.
1352 */
1353 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1354 CPUFREQ_REMOVE_POLICY, policy);
1355 freq_qos_remove_request(policy->max_freq_req);
1356 }
1357
1358 freq_qos_remove_request(policy->min_freq_req);
1359 kfree(policy->min_freq_req);
1360
1361 cpufreq_policy_put_kobj(policy);
1362 free_cpumask_var(policy->real_cpus);
1363 free_cpumask_var(policy->related_cpus);
1364 free_cpumask_var(policy->cpus);
1365 kfree(policy);
1366}
1367
1368static int cpufreq_online(unsigned int cpu)
1369{
1370 struct cpufreq_policy *policy;
1371 bool new_policy;
1372 unsigned long flags;
1373 unsigned int j;
1374 int ret;
1375
1376 pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1377
1378 /* Check if this CPU already has a policy to manage it */
1379 policy = per_cpu(cpufreq_cpu_data, cpu);
1380 if (policy) {
1381 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1382 if (!policy_is_inactive(policy))
1383 return cpufreq_add_policy_cpu(policy, cpu);
1384
1385 /* This is the only online CPU for the policy. Start over. */
1386 new_policy = false;
1387 down_write(&policy->rwsem);
1388 policy->cpu = cpu;
1389 policy->governor = NULL;
1390 } else {
1391 new_policy = true;
1392 policy = cpufreq_policy_alloc(cpu);
1393 if (!policy)
1394 return -ENOMEM;
1395 down_write(&policy->rwsem);
1396 }
1397
1398 if (!new_policy && cpufreq_driver->online) {
1399 /* Recover policy->cpus using related_cpus */
1400 cpumask_copy(policy->cpus, policy->related_cpus);
1401
1402 ret = cpufreq_driver->online(policy);
1403 if (ret) {
1404 pr_debug("%s: %d: initialization failed\n", __func__,
1405 __LINE__);
1406 goto out_exit_policy;
1407 }
1408 } else {
1409 cpumask_copy(policy->cpus, cpumask_of(cpu));
1410
1411 /*
1412 * Call driver. From then on the cpufreq must be able
1413 * to accept all calls to ->verify and ->setpolicy for this CPU.
1414 */
1415 ret = cpufreq_driver->init(policy);
1416 if (ret) {
1417 pr_debug("%s: %d: initialization failed\n", __func__,
1418 __LINE__);
1419 goto out_free_policy;
1420 }
1421
1422 /*
1423 * The initialization has succeeded and the policy is online.
1424 * If there is a problem with its frequency table, take it
1425 * offline and drop it.
1426 */
1427 ret = cpufreq_table_validate_and_sort(policy);
1428 if (ret)
1429 goto out_offline_policy;
1430
1431 /* related_cpus should at least include policy->cpus. */
1432 cpumask_copy(policy->related_cpus, policy->cpus);
1433 }
1434
1435 /*
1436 * affected cpus must always be the one, which are online. We aren't
1437 * managing offline cpus here.
1438 */
1439 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1440
1441 if (new_policy) {
1442 for_each_cpu(j, policy->related_cpus) {
1443 per_cpu(cpufreq_cpu_data, j) = policy;
1444 add_cpu_dev_symlink(policy, j, get_cpu_device(j));
1445 }
1446
1447 policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1448 GFP_KERNEL);
1449 if (!policy->min_freq_req) {
1450 ret = -ENOMEM;
1451 goto out_destroy_policy;
1452 }
1453
1454 ret = freq_qos_add_request(&policy->constraints,
1455 policy->min_freq_req, FREQ_QOS_MIN,
1456 FREQ_QOS_MIN_DEFAULT_VALUE);
1457 if (ret < 0) {
1458 /*
1459 * So we don't call freq_qos_remove_request() for an
1460 * uninitialized request.
1461 */
1462 kfree(policy->min_freq_req);
1463 policy->min_freq_req = NULL;
1464 goto out_destroy_policy;
1465 }
1466
1467 /*
1468 * This must be initialized right here to avoid calling
1469 * freq_qos_remove_request() on uninitialized request in case
1470 * of errors.
1471 */
1472 policy->max_freq_req = policy->min_freq_req + 1;
1473
1474 ret = freq_qos_add_request(&policy->constraints,
1475 policy->max_freq_req, FREQ_QOS_MAX,
1476 FREQ_QOS_MAX_DEFAULT_VALUE);
1477 if (ret < 0) {
1478 policy->max_freq_req = NULL;
1479 goto out_destroy_policy;
1480 }
1481
1482 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1483 CPUFREQ_CREATE_POLICY, policy);
1484 }
1485
1486 if (cpufreq_driver->get && has_target()) {
1487 policy->cur = cpufreq_driver->get(policy->cpu);
1488 if (!policy->cur) {
1489 ret = -EIO;
1490 pr_err("%s: ->get() failed\n", __func__);
1491 goto out_destroy_policy;
1492 }
1493 }
1494
1495 /*
1496 * Sometimes boot loaders set CPU frequency to a value outside of
1497 * frequency table present with cpufreq core. In such cases CPU might be
1498 * unstable if it has to run on that frequency for long duration of time
1499 * and so its better to set it to a frequency which is specified in
1500 * freq-table. This also makes cpufreq stats inconsistent as
1501 * cpufreq-stats would fail to register because current frequency of CPU
1502 * isn't found in freq-table.
1503 *
1504 * Because we don't want this change to effect boot process badly, we go
1505 * for the next freq which is >= policy->cur ('cur' must be set by now,
1506 * otherwise we will end up setting freq to lowest of the table as 'cur'
1507 * is initialized to zero).
1508 *
1509 * We are passing target-freq as "policy->cur - 1" otherwise
1510 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1511 * equal to target-freq.
1512 */
1513 if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1514 && has_target()) {
1515 unsigned int old_freq = policy->cur;
1516
1517 /* Are we running at unknown frequency ? */
1518 ret = cpufreq_frequency_table_get_index(policy, old_freq);
1519 if (ret == -EINVAL) {
1520 ret = __cpufreq_driver_target(policy, old_freq - 1,
1521 CPUFREQ_RELATION_L);
1522
1523 /*
1524 * Reaching here after boot in a few seconds may not
1525 * mean that system will remain stable at "unknown"
1526 * frequency for longer duration. Hence, a BUG_ON().
1527 */
1528 BUG_ON(ret);
1529 pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
1530 __func__, policy->cpu, old_freq, policy->cur);
1531 }
1532 }
1533
1534 if (new_policy) {
1535 ret = cpufreq_add_dev_interface(policy);
1536 if (ret)
1537 goto out_destroy_policy;
1538
1539 cpufreq_stats_create_table(policy);
1540
1541 write_lock_irqsave(&cpufreq_driver_lock, flags);
1542 list_add(&policy->policy_list, &cpufreq_policy_list);
1543 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1544
1545 /*
1546 * Register with the energy model before
1547 * sugov_eas_rebuild_sd() is called, which will result
1548 * in rebuilding of the sched domains, which should only be done
1549 * once the energy model is properly initialized for the policy
1550 * first.
1551 *
1552 * Also, this should be called before the policy is registered
1553 * with cooling framework.
1554 */
1555 if (cpufreq_driver->register_em)
1556 cpufreq_driver->register_em(policy);
1557 }
1558
1559 ret = cpufreq_init_policy(policy);
1560 if (ret) {
1561 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1562 __func__, cpu, ret);
1563 goto out_destroy_policy;
1564 }
1565
1566 up_write(&policy->rwsem);
1567
1568 kobject_uevent(&policy->kobj, KOBJ_ADD);
1569
1570 /* Callback for handling stuff after policy is ready */
1571 if (cpufreq_driver->ready)
1572 cpufreq_driver->ready(policy);
1573
1574 if (cpufreq_thermal_control_enabled(cpufreq_driver))
1575 policy->cdev = of_cpufreq_cooling_register(policy);
1576
1577 pr_debug("initialization complete\n");
1578
1579 return 0;
1580
1581out_destroy_policy:
1582 for_each_cpu(j, policy->real_cpus)
1583 remove_cpu_dev_symlink(policy, j, get_cpu_device(j));
1584
1585out_offline_policy:
1586 if (cpufreq_driver->offline)
1587 cpufreq_driver->offline(policy);
1588
1589out_exit_policy:
1590 if (cpufreq_driver->exit)
1591 cpufreq_driver->exit(policy);
1592
1593out_free_policy:
1594 cpumask_clear(policy->cpus);
1595 up_write(&policy->rwsem);
1596
1597 cpufreq_policy_free(policy);
1598 return ret;
1599}
1600
1601/**
1602 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1603 * @dev: CPU device.
1604 * @sif: Subsystem interface structure pointer (not used)
1605 */
1606static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1607{
1608 struct cpufreq_policy *policy;
1609 unsigned cpu = dev->id;
1610 int ret;
1611
1612 dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1613
1614 if (cpu_online(cpu)) {
1615 ret = cpufreq_online(cpu);
1616 if (ret)
1617 return ret;
1618 }
1619
1620 /* Create sysfs link on CPU registration */
1621 policy = per_cpu(cpufreq_cpu_data, cpu);
1622 if (policy)
1623 add_cpu_dev_symlink(policy, cpu, dev);
1624
1625 return 0;
1626}
1627
1628static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1629{
1630 int ret;
1631
1632 if (has_target())
1633 cpufreq_stop_governor(policy);
1634
1635 cpumask_clear_cpu(cpu, policy->cpus);
1636
1637 if (!policy_is_inactive(policy)) {
1638 /* Nominate a new CPU if necessary. */
1639 if (cpu == policy->cpu)
1640 policy->cpu = cpumask_any(policy->cpus);
1641
1642 /* Start the governor again for the active policy. */
1643 if (has_target()) {
1644 ret = cpufreq_start_governor(policy);
1645 if (ret)
1646 pr_err("%s: Failed to start governor\n", __func__);
1647 }
1648
1649 return;
1650 }
1651
1652 if (has_target())
1653 strscpy(policy->last_governor, policy->governor->name,
1654 CPUFREQ_NAME_LEN);
1655 else
1656 policy->last_policy = policy->policy;
1657
1658 if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1659 cpufreq_cooling_unregister(policy->cdev);
1660 policy->cdev = NULL;
1661 }
1662
1663 if (has_target())
1664 cpufreq_exit_governor(policy);
1665
1666 /*
1667 * Perform the ->offline() during light-weight tear-down, as
1668 * that allows fast recovery when the CPU comes back.
1669 */
1670 if (cpufreq_driver->offline) {
1671 cpufreq_driver->offline(policy);
1672 } else if (cpufreq_driver->exit) {
1673 cpufreq_driver->exit(policy);
1674 policy->freq_table = NULL;
1675 }
1676}
1677
1678static int cpufreq_offline(unsigned int cpu)
1679{
1680 struct cpufreq_policy *policy;
1681
1682 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1683
1684 policy = cpufreq_cpu_get_raw(cpu);
1685 if (!policy) {
1686 pr_debug("%s: No cpu_data found\n", __func__);
1687 return 0;
1688 }
1689
1690 down_write(&policy->rwsem);
1691
1692 __cpufreq_offline(cpu, policy);
1693
1694 up_write(&policy->rwsem);
1695 return 0;
1696}
1697
1698/*
1699 * cpufreq_remove_dev - remove a CPU device
1700 *
1701 * Removes the cpufreq interface for a CPU device.
1702 */
1703static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1704{
1705 unsigned int cpu = dev->id;
1706 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1707
1708 if (!policy)
1709 return;
1710
1711 down_write(&policy->rwsem);
1712
1713 if (cpu_online(cpu))
1714 __cpufreq_offline(cpu, policy);
1715
1716 remove_cpu_dev_symlink(policy, cpu, dev);
1717
1718 if (!cpumask_empty(policy->real_cpus)) {
1719 up_write(&policy->rwsem);
1720 return;
1721 }
1722
1723 /* We did light-weight exit earlier, do full tear down now */
1724 if (cpufreq_driver->offline)
1725 cpufreq_driver->exit(policy);
1726
1727 up_write(&policy->rwsem);
1728
1729 cpufreq_policy_free(policy);
1730}
1731
1732/**
1733 * cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1734 * @policy: Policy managing CPUs.
1735 * @new_freq: New CPU frequency.
1736 *
1737 * Adjust to the current frequency first and clean up later by either calling
1738 * cpufreq_update_policy(), or scheduling handle_update().
1739 */
1740static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1741 unsigned int new_freq)
1742{
1743 struct cpufreq_freqs freqs;
1744
1745 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1746 policy->cur, new_freq);
1747
1748 freqs.old = policy->cur;
1749 freqs.new = new_freq;
1750
1751 cpufreq_freq_transition_begin(policy, &freqs);
1752 cpufreq_freq_transition_end(policy, &freqs, 0);
1753}
1754
1755static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1756{
1757 unsigned int new_freq;
1758
1759 new_freq = cpufreq_driver->get(policy->cpu);
1760 if (!new_freq)
1761 return 0;
1762
1763 /*
1764 * If fast frequency switching is used with the given policy, the check
1765 * against policy->cur is pointless, so skip it in that case.
1766 */
1767 if (policy->fast_switch_enabled || !has_target())
1768 return new_freq;
1769
1770 if (policy->cur != new_freq) {
1771 /*
1772 * For some platforms, the frequency returned by hardware may be
1773 * slightly different from what is provided in the frequency
1774 * table, for example hardware may return 499 MHz instead of 500
1775 * MHz. In such cases it is better to avoid getting into
1776 * unnecessary frequency updates.
1777 */
1778 if (abs(policy->cur - new_freq) < KHZ_PER_MHZ)
1779 return policy->cur;
1780
1781 cpufreq_out_of_sync(policy, new_freq);
1782 if (update)
1783 schedule_work(&policy->update);
1784 }
1785
1786 return new_freq;
1787}
1788
1789/**
1790 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1791 * @cpu: CPU number
1792 *
1793 * This is the last known freq, without actually getting it from the driver.
1794 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1795 */
1796unsigned int cpufreq_quick_get(unsigned int cpu)
1797{
1798 struct cpufreq_policy *policy;
1799 unsigned int ret_freq = 0;
1800 unsigned long flags;
1801
1802 read_lock_irqsave(&cpufreq_driver_lock, flags);
1803
1804 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1805 ret_freq = cpufreq_driver->get(cpu);
1806 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1807 return ret_freq;
1808 }
1809
1810 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1811
1812 policy = cpufreq_cpu_get(cpu);
1813 if (policy) {
1814 ret_freq = policy->cur;
1815 cpufreq_cpu_put(policy);
1816 }
1817
1818 return ret_freq;
1819}
1820EXPORT_SYMBOL(cpufreq_quick_get);
1821
1822/**
1823 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1824 * @cpu: CPU number
1825 *
1826 * Just return the max possible frequency for a given CPU.
1827 */
1828unsigned int cpufreq_quick_get_max(unsigned int cpu)
1829{
1830 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1831 unsigned int ret_freq = 0;
1832
1833 if (policy) {
1834 ret_freq = policy->max;
1835 cpufreq_cpu_put(policy);
1836 }
1837
1838 return ret_freq;
1839}
1840EXPORT_SYMBOL(cpufreq_quick_get_max);
1841
1842/**
1843 * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1844 * @cpu: CPU number
1845 *
1846 * The default return value is the max_freq field of cpuinfo.
1847 */
1848__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1849{
1850 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1851 unsigned int ret_freq = 0;
1852
1853 if (policy) {
1854 ret_freq = policy->cpuinfo.max_freq;
1855 cpufreq_cpu_put(policy);
1856 }
1857
1858 return ret_freq;
1859}
1860EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1861
1862static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1863{
1864 if (unlikely(policy_is_inactive(policy)))
1865 return 0;
1866
1867 return cpufreq_verify_current_freq(policy, true);
1868}
1869
1870/**
1871 * cpufreq_get - get the current CPU frequency (in kHz)
1872 * @cpu: CPU number
1873 *
1874 * Get the CPU current (static) CPU frequency
1875 */
1876unsigned int cpufreq_get(unsigned int cpu)
1877{
1878 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1879 unsigned int ret_freq = 0;
1880
1881 if (policy) {
1882 down_read(&policy->rwsem);
1883 if (cpufreq_driver->get)
1884 ret_freq = __cpufreq_get(policy);
1885 up_read(&policy->rwsem);
1886
1887 cpufreq_cpu_put(policy);
1888 }
1889
1890 return ret_freq;
1891}
1892EXPORT_SYMBOL(cpufreq_get);
1893
1894static struct subsys_interface cpufreq_interface = {
1895 .name = "cpufreq",
1896 .subsys = &cpu_subsys,
1897 .add_dev = cpufreq_add_dev,
1898 .remove_dev = cpufreq_remove_dev,
1899};
1900
1901/*
1902 * In case platform wants some specific frequency to be configured
1903 * during suspend..
1904 */
1905int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1906{
1907 int ret;
1908
1909 if (!policy->suspend_freq) {
1910 pr_debug("%s: suspend_freq not defined\n", __func__);
1911 return 0;
1912 }
1913
1914 pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1915 policy->suspend_freq);
1916
1917 ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1918 CPUFREQ_RELATION_H);
1919 if (ret)
1920 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1921 __func__, policy->suspend_freq, ret);
1922
1923 return ret;
1924}
1925EXPORT_SYMBOL(cpufreq_generic_suspend);
1926
1927/**
1928 * cpufreq_suspend() - Suspend CPUFreq governors.
1929 *
1930 * Called during system wide Suspend/Hibernate cycles for suspending governors
1931 * as some platforms can't change frequency after this point in suspend cycle.
1932 * Because some of the devices (like: i2c, regulators, etc) they use for
1933 * changing frequency are suspended quickly after this point.
1934 */
1935void cpufreq_suspend(void)
1936{
1937 struct cpufreq_policy *policy;
1938
1939 if (!cpufreq_driver)
1940 return;
1941
1942 if (!has_target() && !cpufreq_driver->suspend)
1943 goto suspend;
1944
1945 pr_debug("%s: Suspending Governors\n", __func__);
1946
1947 for_each_active_policy(policy) {
1948 if (has_target()) {
1949 down_write(&policy->rwsem);
1950 cpufreq_stop_governor(policy);
1951 up_write(&policy->rwsem);
1952 }
1953
1954 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1955 pr_err("%s: Failed to suspend driver: %s\n", __func__,
1956 cpufreq_driver->name);
1957 }
1958
1959suspend:
1960 cpufreq_suspended = true;
1961}
1962
1963/**
1964 * cpufreq_resume() - Resume CPUFreq governors.
1965 *
1966 * Called during system wide Suspend/Hibernate cycle for resuming governors that
1967 * are suspended with cpufreq_suspend().
1968 */
1969void cpufreq_resume(void)
1970{
1971 struct cpufreq_policy *policy;
1972 int ret;
1973
1974 if (!cpufreq_driver)
1975 return;
1976
1977 if (unlikely(!cpufreq_suspended))
1978 return;
1979
1980 cpufreq_suspended = false;
1981
1982 if (!has_target() && !cpufreq_driver->resume)
1983 return;
1984
1985 pr_debug("%s: Resuming Governors\n", __func__);
1986
1987 for_each_active_policy(policy) {
1988 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1989 pr_err("%s: Failed to resume driver: %s\n", __func__,
1990 cpufreq_driver->name);
1991 } else if (has_target()) {
1992 down_write(&policy->rwsem);
1993 ret = cpufreq_start_governor(policy);
1994 up_write(&policy->rwsem);
1995
1996 if (ret)
1997 pr_err("%s: Failed to start governor for CPU%u's policy\n",
1998 __func__, policy->cpu);
1999 }
2000 }
2001}
2002
2003/**
2004 * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
2005 * @flags: Flags to test against the current cpufreq driver's flags.
2006 *
2007 * Assumes that the driver is there, so callers must ensure that this is the
2008 * case.
2009 */
2010bool cpufreq_driver_test_flags(u16 flags)
2011{
2012 return !!(cpufreq_driver->flags & flags);
2013}
2014
2015/**
2016 * cpufreq_get_current_driver - Return the current driver's name.
2017 *
2018 * Return the name string of the currently registered cpufreq driver or NULL if
2019 * none.
2020 */
2021const char *cpufreq_get_current_driver(void)
2022{
2023 if (cpufreq_driver)
2024 return cpufreq_driver->name;
2025
2026 return NULL;
2027}
2028EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
2029
2030/**
2031 * cpufreq_get_driver_data - Return current driver data.
2032 *
2033 * Return the private data of the currently registered cpufreq driver, or NULL
2034 * if no cpufreq driver has been registered.
2035 */
2036void *cpufreq_get_driver_data(void)
2037{
2038 if (cpufreq_driver)
2039 return cpufreq_driver->driver_data;
2040
2041 return NULL;
2042}
2043EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
2044
2045/*********************************************************************
2046 * NOTIFIER LISTS INTERFACE *
2047 *********************************************************************/
2048
2049/**
2050 * cpufreq_register_notifier - Register a notifier with cpufreq.
2051 * @nb: notifier function to register.
2052 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2053 *
2054 * Add a notifier to one of two lists: either a list of notifiers that run on
2055 * clock rate changes (once before and once after every transition), or a list
2056 * of notifiers that ron on cpufreq policy changes.
2057 *
2058 * This function may sleep and it has the same return values as
2059 * blocking_notifier_chain_register().
2060 */
2061int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
2062{
2063 int ret;
2064
2065 if (cpufreq_disabled())
2066 return -EINVAL;
2067
2068 switch (list) {
2069 case CPUFREQ_TRANSITION_NOTIFIER:
2070 mutex_lock(&cpufreq_fast_switch_lock);
2071
2072 if (cpufreq_fast_switch_count > 0) {
2073 mutex_unlock(&cpufreq_fast_switch_lock);
2074 return -EBUSY;
2075 }
2076 ret = srcu_notifier_chain_register(
2077 &cpufreq_transition_notifier_list, nb);
2078 if (!ret)
2079 cpufreq_fast_switch_count--;
2080
2081 mutex_unlock(&cpufreq_fast_switch_lock);
2082 break;
2083 case CPUFREQ_POLICY_NOTIFIER:
2084 ret = blocking_notifier_chain_register(
2085 &cpufreq_policy_notifier_list, nb);
2086 break;
2087 default:
2088 ret = -EINVAL;
2089 }
2090
2091 return ret;
2092}
2093EXPORT_SYMBOL(cpufreq_register_notifier);
2094
2095/**
2096 * cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2097 * @nb: notifier block to be unregistered.
2098 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2099 *
2100 * Remove a notifier from one of the cpufreq notifier lists.
2101 *
2102 * This function may sleep and it has the same return values as
2103 * blocking_notifier_chain_unregister().
2104 */
2105int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2106{
2107 int ret;
2108
2109 if (cpufreq_disabled())
2110 return -EINVAL;
2111
2112 switch (list) {
2113 case CPUFREQ_TRANSITION_NOTIFIER:
2114 mutex_lock(&cpufreq_fast_switch_lock);
2115
2116 ret = srcu_notifier_chain_unregister(
2117 &cpufreq_transition_notifier_list, nb);
2118 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2119 cpufreq_fast_switch_count++;
2120
2121 mutex_unlock(&cpufreq_fast_switch_lock);
2122 break;
2123 case CPUFREQ_POLICY_NOTIFIER:
2124 ret = blocking_notifier_chain_unregister(
2125 &cpufreq_policy_notifier_list, nb);
2126 break;
2127 default:
2128 ret = -EINVAL;
2129 }
2130
2131 return ret;
2132}
2133EXPORT_SYMBOL(cpufreq_unregister_notifier);
2134
2135
2136/*********************************************************************
2137 * GOVERNORS *
2138 *********************************************************************/
2139
2140/**
2141 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2142 * @policy: cpufreq policy to switch the frequency for.
2143 * @target_freq: New frequency to set (may be approximate).
2144 *
2145 * Carry out a fast frequency switch without sleeping.
2146 *
2147 * The driver's ->fast_switch() callback invoked by this function must be
2148 * suitable for being called from within RCU-sched read-side critical sections
2149 * and it is expected to select the minimum available frequency greater than or
2150 * equal to @target_freq (CPUFREQ_RELATION_L).
2151 *
2152 * This function must not be called if policy->fast_switch_enabled is unset.
2153 *
2154 * Governors calling this function must guarantee that it will never be invoked
2155 * twice in parallel for the same policy and that it will never be called in
2156 * parallel with either ->target() or ->target_index() for the same policy.
2157 *
2158 * Returns the actual frequency set for the CPU.
2159 *
2160 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2161 * error condition, the hardware configuration must be preserved.
2162 */
2163unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2164 unsigned int target_freq)
2165{
2166 unsigned int freq;
2167 int cpu;
2168
2169 target_freq = clamp_val(target_freq, policy->min, policy->max);
2170 freq = cpufreq_driver->fast_switch(policy, target_freq);
2171
2172 if (!freq)
2173 return 0;
2174
2175 policy->cur = freq;
2176 arch_set_freq_scale(policy->related_cpus, freq,
2177 arch_scale_freq_ref(policy->cpu));
2178 cpufreq_stats_record_transition(policy, freq);
2179
2180 if (trace_cpu_frequency_enabled()) {
2181 for_each_cpu(cpu, policy->cpus)
2182 trace_cpu_frequency(freq, cpu);
2183 }
2184
2185 return freq;
2186}
2187EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2188
2189/**
2190 * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2191 * @cpu: Target CPU.
2192 * @min_perf: Minimum (required) performance level (units of @capacity).
2193 * @target_perf: Target (desired) performance level (units of @capacity).
2194 * @capacity: Capacity of the target CPU.
2195 *
2196 * Carry out a fast performance level switch of @cpu without sleeping.
2197 *
2198 * The driver's ->adjust_perf() callback invoked by this function must be
2199 * suitable for being called from within RCU-sched read-side critical sections
2200 * and it is expected to select a suitable performance level equal to or above
2201 * @min_perf and preferably equal to or below @target_perf.
2202 *
2203 * This function must not be called if policy->fast_switch_enabled is unset.
2204 *
2205 * Governors calling this function must guarantee that it will never be invoked
2206 * twice in parallel for the same CPU and that it will never be called in
2207 * parallel with either ->target() or ->target_index() or ->fast_switch() for
2208 * the same CPU.
2209 */
2210void cpufreq_driver_adjust_perf(unsigned int cpu,
2211 unsigned long min_perf,
2212 unsigned long target_perf,
2213 unsigned long capacity)
2214{
2215 cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2216}
2217
2218/**
2219 * cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2220 *
2221 * Return 'true' if the ->adjust_perf callback is present for the
2222 * current driver or 'false' otherwise.
2223 */
2224bool cpufreq_driver_has_adjust_perf(void)
2225{
2226 return !!cpufreq_driver->adjust_perf;
2227}
2228
2229/* Must set freqs->new to intermediate frequency */
2230static int __target_intermediate(struct cpufreq_policy *policy,
2231 struct cpufreq_freqs *freqs, int index)
2232{
2233 int ret;
2234
2235 freqs->new = cpufreq_driver->get_intermediate(policy, index);
2236
2237 /* We don't need to switch to intermediate freq */
2238 if (!freqs->new)
2239 return 0;
2240
2241 pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2242 __func__, policy->cpu, freqs->old, freqs->new);
2243
2244 cpufreq_freq_transition_begin(policy, freqs);
2245 ret = cpufreq_driver->target_intermediate(policy, index);
2246 cpufreq_freq_transition_end(policy, freqs, ret);
2247
2248 if (ret)
2249 pr_err("%s: Failed to change to intermediate frequency: %d\n",
2250 __func__, ret);
2251
2252 return ret;
2253}
2254
2255static int __target_index(struct cpufreq_policy *policy, int index)
2256{
2257 struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2258 unsigned int restore_freq, intermediate_freq = 0;
2259 unsigned int newfreq = policy->freq_table[index].frequency;
2260 int retval = -EINVAL;
2261 bool notify;
2262
2263 if (newfreq == policy->cur)
2264 return 0;
2265
2266 /* Save last value to restore later on errors */
2267 restore_freq = policy->cur;
2268
2269 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2270 if (notify) {
2271 /* Handle switching to intermediate frequency */
2272 if (cpufreq_driver->get_intermediate) {
2273 retval = __target_intermediate(policy, &freqs, index);
2274 if (retval)
2275 return retval;
2276
2277 intermediate_freq = freqs.new;
2278 /* Set old freq to intermediate */
2279 if (intermediate_freq)
2280 freqs.old = freqs.new;
2281 }
2282
2283 freqs.new = newfreq;
2284 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2285 __func__, policy->cpu, freqs.old, freqs.new);
2286
2287 cpufreq_freq_transition_begin(policy, &freqs);
2288 }
2289
2290 retval = cpufreq_driver->target_index(policy, index);
2291 if (retval)
2292 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2293 retval);
2294
2295 if (notify) {
2296 cpufreq_freq_transition_end(policy, &freqs, retval);
2297
2298 /*
2299 * Failed after setting to intermediate freq? Driver should have
2300 * reverted back to initial frequency and so should we. Check
2301 * here for intermediate_freq instead of get_intermediate, in
2302 * case we haven't switched to intermediate freq at all.
2303 */
2304 if (unlikely(retval && intermediate_freq)) {
2305 freqs.old = intermediate_freq;
2306 freqs.new = restore_freq;
2307 cpufreq_freq_transition_begin(policy, &freqs);
2308 cpufreq_freq_transition_end(policy, &freqs, 0);
2309 }
2310 }
2311
2312 return retval;
2313}
2314
2315int __cpufreq_driver_target(struct cpufreq_policy *policy,
2316 unsigned int target_freq,
2317 unsigned int relation)
2318{
2319 unsigned int old_target_freq = target_freq;
2320
2321 if (cpufreq_disabled())
2322 return -ENODEV;
2323
2324 target_freq = __resolve_freq(policy, target_freq, relation);
2325
2326 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2327 policy->cpu, target_freq, relation, old_target_freq);
2328
2329 /*
2330 * This might look like a redundant call as we are checking it again
2331 * after finding index. But it is left intentionally for cases where
2332 * exactly same freq is called again and so we can save on few function
2333 * calls.
2334 */
2335 if (target_freq == policy->cur &&
2336 !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2337 return 0;
2338
2339 if (cpufreq_driver->target) {
2340 /*
2341 * If the driver hasn't setup a single inefficient frequency,
2342 * it's unlikely it knows how to decode CPUFREQ_RELATION_E.
2343 */
2344 if (!policy->efficiencies_available)
2345 relation &= ~CPUFREQ_RELATION_E;
2346
2347 return cpufreq_driver->target(policy, target_freq, relation);
2348 }
2349
2350 if (!cpufreq_driver->target_index)
2351 return -EINVAL;
2352
2353 return __target_index(policy, policy->cached_resolved_idx);
2354}
2355EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2356
2357int cpufreq_driver_target(struct cpufreq_policy *policy,
2358 unsigned int target_freq,
2359 unsigned int relation)
2360{
2361 int ret;
2362
2363 down_write(&policy->rwsem);
2364
2365 ret = __cpufreq_driver_target(policy, target_freq, relation);
2366
2367 up_write(&policy->rwsem);
2368
2369 return ret;
2370}
2371EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2372
2373__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2374{
2375 return NULL;
2376}
2377
2378static int cpufreq_init_governor(struct cpufreq_policy *policy)
2379{
2380 int ret;
2381
2382 /* Don't start any governor operations if we are entering suspend */
2383 if (cpufreq_suspended)
2384 return 0;
2385 /*
2386 * Governor might not be initiated here if ACPI _PPC changed
2387 * notification happened, so check it.
2388 */
2389 if (!policy->governor)
2390 return -EINVAL;
2391
2392 /* Platform doesn't want dynamic frequency switching ? */
2393 if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2394 cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2395 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2396
2397 if (gov) {
2398 pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2399 policy->governor->name, gov->name);
2400 policy->governor = gov;
2401 } else {
2402 return -EINVAL;
2403 }
2404 }
2405
2406 if (!try_module_get(policy->governor->owner))
2407 return -EINVAL;
2408
2409 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2410
2411 if (policy->governor->init) {
2412 ret = policy->governor->init(policy);
2413 if (ret) {
2414 module_put(policy->governor->owner);
2415 return ret;
2416 }
2417 }
2418
2419 policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2420
2421 return 0;
2422}
2423
2424static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2425{
2426 if (cpufreq_suspended || !policy->governor)
2427 return;
2428
2429 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2430
2431 if (policy->governor->exit)
2432 policy->governor->exit(policy);
2433
2434 module_put(policy->governor->owner);
2435}
2436
2437int cpufreq_start_governor(struct cpufreq_policy *policy)
2438{
2439 int ret;
2440
2441 if (cpufreq_suspended)
2442 return 0;
2443
2444 if (!policy->governor)
2445 return -EINVAL;
2446
2447 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2448
2449 if (cpufreq_driver->get)
2450 cpufreq_verify_current_freq(policy, false);
2451
2452 if (policy->governor->start) {
2453 ret = policy->governor->start(policy);
2454 if (ret)
2455 return ret;
2456 }
2457
2458 if (policy->governor->limits)
2459 policy->governor->limits(policy);
2460
2461 return 0;
2462}
2463
2464void cpufreq_stop_governor(struct cpufreq_policy *policy)
2465{
2466 if (cpufreq_suspended || !policy->governor)
2467 return;
2468
2469 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2470
2471 if (policy->governor->stop)
2472 policy->governor->stop(policy);
2473}
2474
2475static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2476{
2477 if (cpufreq_suspended || !policy->governor)
2478 return;
2479
2480 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2481
2482 if (policy->governor->limits)
2483 policy->governor->limits(policy);
2484}
2485
2486int cpufreq_register_governor(struct cpufreq_governor *governor)
2487{
2488 int err;
2489
2490 if (!governor)
2491 return -EINVAL;
2492
2493 if (cpufreq_disabled())
2494 return -ENODEV;
2495
2496 mutex_lock(&cpufreq_governor_mutex);
2497
2498 err = -EBUSY;
2499 if (!find_governor(governor->name)) {
2500 err = 0;
2501 list_add(&governor->governor_list, &cpufreq_governor_list);
2502 }
2503
2504 mutex_unlock(&cpufreq_governor_mutex);
2505 return err;
2506}
2507EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2508
2509void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2510{
2511 struct cpufreq_policy *policy;
2512 unsigned long flags;
2513
2514 if (!governor)
2515 return;
2516
2517 if (cpufreq_disabled())
2518 return;
2519
2520 /* clear last_governor for all inactive policies */
2521 read_lock_irqsave(&cpufreq_driver_lock, flags);
2522 for_each_inactive_policy(policy) {
2523 if (!strcmp(policy->last_governor, governor->name)) {
2524 policy->governor = NULL;
2525 strcpy(policy->last_governor, "\0");
2526 }
2527 }
2528 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2529
2530 mutex_lock(&cpufreq_governor_mutex);
2531 list_del(&governor->governor_list);
2532 mutex_unlock(&cpufreq_governor_mutex);
2533}
2534EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2535
2536
2537/*********************************************************************
2538 * POLICY INTERFACE *
2539 *********************************************************************/
2540
2541/**
2542 * cpufreq_get_policy - get the current cpufreq_policy
2543 * @policy: struct cpufreq_policy into which the current cpufreq_policy
2544 * is written
2545 * @cpu: CPU to find the policy for
2546 *
2547 * Reads the current cpufreq policy.
2548 */
2549int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2550{
2551 struct cpufreq_policy *cpu_policy;
2552 if (!policy)
2553 return -EINVAL;
2554
2555 cpu_policy = cpufreq_cpu_get(cpu);
2556 if (!cpu_policy)
2557 return -EINVAL;
2558
2559 memcpy(policy, cpu_policy, sizeof(*policy));
2560
2561 cpufreq_cpu_put(cpu_policy);
2562 return 0;
2563}
2564EXPORT_SYMBOL(cpufreq_get_policy);
2565
2566/**
2567 * cpufreq_set_policy - Modify cpufreq policy parameters.
2568 * @policy: Policy object to modify.
2569 * @new_gov: Policy governor pointer.
2570 * @new_pol: Policy value (for drivers with built-in governors).
2571 *
2572 * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2573 * limits to be set for the policy, update @policy with the verified limits
2574 * values and either invoke the driver's ->setpolicy() callback (if present) or
2575 * carry out a governor update for @policy. That is, run the current governor's
2576 * ->limits() callback (if @new_gov points to the same object as the one in
2577 * @policy) or replace the governor for @policy with @new_gov.
2578 *
2579 * The cpuinfo part of @policy is not updated by this function.
2580 */
2581static int cpufreq_set_policy(struct cpufreq_policy *policy,
2582 struct cpufreq_governor *new_gov,
2583 unsigned int new_pol)
2584{
2585 struct cpufreq_policy_data new_data;
2586 struct cpufreq_governor *old_gov;
2587 int ret;
2588
2589 memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2590 new_data.freq_table = policy->freq_table;
2591 new_data.cpu = policy->cpu;
2592 /*
2593 * PM QoS framework collects all the requests from users and provide us
2594 * the final aggregated value here.
2595 */
2596 new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2597 new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2598
2599 pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2600 new_data.cpu, new_data.min, new_data.max);
2601
2602 /*
2603 * Verify that the CPU speed can be set within these limits and make sure
2604 * that min <= max.
2605 */
2606 ret = cpufreq_driver->verify(&new_data);
2607 if (ret)
2608 return ret;
2609
2610 /*
2611 * Resolve policy min/max to available frequencies. It ensures
2612 * no frequency resolution will neither overshoot the requested maximum
2613 * nor undershoot the requested minimum.
2614 */
2615 policy->min = new_data.min;
2616 policy->max = new_data.max;
2617 policy->min = __resolve_freq(policy, policy->min, CPUFREQ_RELATION_L);
2618 policy->max = __resolve_freq(policy, policy->max, CPUFREQ_RELATION_H);
2619 trace_cpu_frequency_limits(policy);
2620
2621 policy->cached_target_freq = UINT_MAX;
2622
2623 pr_debug("new min and max freqs are %u - %u kHz\n",
2624 policy->min, policy->max);
2625
2626 if (cpufreq_driver->setpolicy) {
2627 policy->policy = new_pol;
2628 pr_debug("setting range\n");
2629 return cpufreq_driver->setpolicy(policy);
2630 }
2631
2632 if (new_gov == policy->governor) {
2633 pr_debug("governor limits update\n");
2634 cpufreq_governor_limits(policy);
2635 return 0;
2636 }
2637
2638 pr_debug("governor switch\n");
2639
2640 /* save old, working values */
2641 old_gov = policy->governor;
2642 /* end old governor */
2643 if (old_gov) {
2644 cpufreq_stop_governor(policy);
2645 cpufreq_exit_governor(policy);
2646 }
2647
2648 /* start new governor */
2649 policy->governor = new_gov;
2650 ret = cpufreq_init_governor(policy);
2651 if (!ret) {
2652 ret = cpufreq_start_governor(policy);
2653 if (!ret) {
2654 pr_debug("governor change\n");
2655 return 0;
2656 }
2657 cpufreq_exit_governor(policy);
2658 }
2659
2660 /* new governor failed, so re-start old one */
2661 pr_debug("starting governor %s failed\n", policy->governor->name);
2662 if (old_gov) {
2663 policy->governor = old_gov;
2664 if (cpufreq_init_governor(policy))
2665 policy->governor = NULL;
2666 else
2667 cpufreq_start_governor(policy);
2668 }
2669
2670 return ret;
2671}
2672
2673/**
2674 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2675 * @cpu: CPU to re-evaluate the policy for.
2676 *
2677 * Update the current frequency for the cpufreq policy of @cpu and use
2678 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2679 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2680 * for the policy in question, among other things.
2681 */
2682void cpufreq_update_policy(unsigned int cpu)
2683{
2684 struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2685
2686 if (!policy)
2687 return;
2688
2689 /*
2690 * BIOS might change freq behind our back
2691 * -> ask driver for current freq and notify governors about a change
2692 */
2693 if (cpufreq_driver->get && has_target() &&
2694 (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2695 goto unlock;
2696
2697 refresh_frequency_limits(policy);
2698
2699unlock:
2700 cpufreq_cpu_release(policy);
2701}
2702EXPORT_SYMBOL(cpufreq_update_policy);
2703
2704/**
2705 * cpufreq_update_limits - Update policy limits for a given CPU.
2706 * @cpu: CPU to update the policy limits for.
2707 *
2708 * Invoke the driver's ->update_limits callback if present or call
2709 * cpufreq_update_policy() for @cpu.
2710 */
2711void cpufreq_update_limits(unsigned int cpu)
2712{
2713 if (cpufreq_driver->update_limits)
2714 cpufreq_driver->update_limits(cpu);
2715 else
2716 cpufreq_update_policy(cpu);
2717}
2718EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2719
2720/*********************************************************************
2721 * BOOST *
2722 *********************************************************************/
2723static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2724{
2725 int ret;
2726
2727 if (!policy->freq_table)
2728 return -ENXIO;
2729
2730 ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2731 if (ret) {
2732 pr_err("%s: Policy frequency update failed\n", __func__);
2733 return ret;
2734 }
2735
2736 ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2737 if (ret < 0)
2738 return ret;
2739
2740 return 0;
2741}
2742
2743int cpufreq_boost_trigger_state(int state)
2744{
2745 struct cpufreq_policy *policy;
2746 unsigned long flags;
2747 int ret = 0;
2748
2749 if (cpufreq_driver->boost_enabled == state)
2750 return 0;
2751
2752 write_lock_irqsave(&cpufreq_driver_lock, flags);
2753 cpufreq_driver->boost_enabled = state;
2754 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2755
2756 cpus_read_lock();
2757 for_each_active_policy(policy) {
2758 ret = cpufreq_driver->set_boost(policy, state);
2759 if (ret)
2760 goto err_reset_state;
2761
2762 policy->boost_enabled = state;
2763 }
2764 cpus_read_unlock();
2765
2766 return 0;
2767
2768err_reset_state:
2769 cpus_read_unlock();
2770
2771 write_lock_irqsave(&cpufreq_driver_lock, flags);
2772 cpufreq_driver->boost_enabled = !state;
2773 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2774
2775 pr_err("%s: Cannot %s BOOST\n",
2776 __func__, state ? "enable" : "disable");
2777
2778 return ret;
2779}
2780
2781static bool cpufreq_boost_supported(void)
2782{
2783 return cpufreq_driver->set_boost;
2784}
2785
2786static int create_boost_sysfs_file(void)
2787{
2788 int ret;
2789
2790 ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2791 if (ret)
2792 pr_err("%s: cannot register global BOOST sysfs file\n",
2793 __func__);
2794
2795 return ret;
2796}
2797
2798static void remove_boost_sysfs_file(void)
2799{
2800 if (cpufreq_boost_supported())
2801 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2802}
2803
2804int cpufreq_enable_boost_support(void)
2805{
2806 if (!cpufreq_driver)
2807 return -EINVAL;
2808
2809 if (cpufreq_boost_supported())
2810 return 0;
2811
2812 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2813
2814 /* This will get removed on driver unregister */
2815 return create_boost_sysfs_file();
2816}
2817EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2818
2819int cpufreq_boost_enabled(void)
2820{
2821 return cpufreq_driver->boost_enabled;
2822}
2823EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2824
2825/*********************************************************************
2826 * REGISTER / UNREGISTER CPUFREQ DRIVER *
2827 *********************************************************************/
2828static enum cpuhp_state hp_online;
2829
2830static int cpuhp_cpufreq_online(unsigned int cpu)
2831{
2832 cpufreq_online(cpu);
2833
2834 return 0;
2835}
2836
2837static int cpuhp_cpufreq_offline(unsigned int cpu)
2838{
2839 cpufreq_offline(cpu);
2840
2841 return 0;
2842}
2843
2844/**
2845 * cpufreq_register_driver - register a CPU Frequency driver
2846 * @driver_data: A struct cpufreq_driver containing the values#
2847 * submitted by the CPU Frequency driver.
2848 *
2849 * Registers a CPU Frequency driver to this core code. This code
2850 * returns zero on success, -EEXIST when another driver got here first
2851 * (and isn't unregistered in the meantime).
2852 *
2853 */
2854int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2855{
2856 unsigned long flags;
2857 int ret;
2858
2859 if (cpufreq_disabled())
2860 return -ENODEV;
2861
2862 /*
2863 * The cpufreq core depends heavily on the availability of device
2864 * structure, make sure they are available before proceeding further.
2865 */
2866 if (!get_cpu_device(0))
2867 return -EPROBE_DEFER;
2868
2869 if (!driver_data || !driver_data->verify || !driver_data->init ||
2870 !(driver_data->setpolicy || driver_data->target_index ||
2871 driver_data->target) ||
2872 (driver_data->setpolicy && (driver_data->target_index ||
2873 driver_data->target)) ||
2874 (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2875 (!driver_data->online != !driver_data->offline) ||
2876 (driver_data->adjust_perf && !driver_data->fast_switch))
2877 return -EINVAL;
2878
2879 pr_debug("trying to register driver %s\n", driver_data->name);
2880
2881 /* Protect against concurrent CPU online/offline. */
2882 cpus_read_lock();
2883
2884 write_lock_irqsave(&cpufreq_driver_lock, flags);
2885 if (cpufreq_driver) {
2886 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2887 ret = -EEXIST;
2888 goto out;
2889 }
2890 cpufreq_driver = driver_data;
2891 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2892
2893 /*
2894 * Mark support for the scheduler's frequency invariance engine for
2895 * drivers that implement target(), target_index() or fast_switch().
2896 */
2897 if (!cpufreq_driver->setpolicy) {
2898 static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2899 pr_debug("supports frequency invariance");
2900 }
2901
2902 if (driver_data->setpolicy)
2903 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2904
2905 if (cpufreq_boost_supported()) {
2906 ret = create_boost_sysfs_file();
2907 if (ret)
2908 goto err_null_driver;
2909 }
2910
2911 ret = subsys_interface_register(&cpufreq_interface);
2912 if (ret)
2913 goto err_boost_unreg;
2914
2915 if (unlikely(list_empty(&cpufreq_policy_list))) {
2916 /* if all ->init() calls failed, unregister */
2917 ret = -ENODEV;
2918 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2919 driver_data->name);
2920 goto err_if_unreg;
2921 }
2922
2923 ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2924 "cpufreq:online",
2925 cpuhp_cpufreq_online,
2926 cpuhp_cpufreq_offline);
2927 if (ret < 0)
2928 goto err_if_unreg;
2929 hp_online = ret;
2930 ret = 0;
2931
2932 pr_debug("driver %s up and running\n", driver_data->name);
2933 goto out;
2934
2935err_if_unreg:
2936 subsys_interface_unregister(&cpufreq_interface);
2937err_boost_unreg:
2938 remove_boost_sysfs_file();
2939err_null_driver:
2940 write_lock_irqsave(&cpufreq_driver_lock, flags);
2941 cpufreq_driver = NULL;
2942 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2943out:
2944 cpus_read_unlock();
2945 return ret;
2946}
2947EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2948
2949/*
2950 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2951 *
2952 * Unregister the current CPUFreq driver. Only call this if you have
2953 * the right to do so, i.e. if you have succeeded in initialising before!
2954 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2955 * currently not initialised.
2956 */
2957void cpufreq_unregister_driver(struct cpufreq_driver *driver)
2958{
2959 unsigned long flags;
2960
2961 if (WARN_ON(!cpufreq_driver || (driver != cpufreq_driver)))
2962 return;
2963
2964 pr_debug("unregistering driver %s\n", driver->name);
2965
2966 /* Protect against concurrent cpu hotplug */
2967 cpus_read_lock();
2968 subsys_interface_unregister(&cpufreq_interface);
2969 remove_boost_sysfs_file();
2970 static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
2971 cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2972
2973 write_lock_irqsave(&cpufreq_driver_lock, flags);
2974
2975 cpufreq_driver = NULL;
2976
2977 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2978 cpus_read_unlock();
2979}
2980EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2981
2982static int __init cpufreq_core_init(void)
2983{
2984 struct cpufreq_governor *gov = cpufreq_default_governor();
2985 struct device *dev_root;
2986
2987 if (cpufreq_disabled())
2988 return -ENODEV;
2989
2990 dev_root = bus_get_dev_root(&cpu_subsys);
2991 if (dev_root) {
2992 cpufreq_global_kobject = kobject_create_and_add("cpufreq", &dev_root->kobj);
2993 put_device(dev_root);
2994 }
2995 BUG_ON(!cpufreq_global_kobject);
2996
2997 if (!strlen(default_governor))
2998 strscpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
2999
3000 return 0;
3001}
3002module_param(off, int, 0444);
3003module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
3004core_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);