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
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 /* Give a 50% breathing room between updates */
580 return latency + (latency >> 1);
581
582 return USEC_PER_MSEC;
583}
584EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
585
586/*********************************************************************
587 * SYSFS INTERFACE *
588 *********************************************************************/
589static ssize_t show_boost(struct kobject *kobj,
590 struct kobj_attribute *attr, char *buf)
591{
592 return sysfs_emit(buf, "%d\n", cpufreq_driver->boost_enabled);
593}
594
595static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
596 const char *buf, size_t count)
597{
598 bool enable;
599
600 if (kstrtobool(buf, &enable))
601 return -EINVAL;
602
603 if (cpufreq_boost_trigger_state(enable)) {
604 pr_err("%s: Cannot %s BOOST!\n",
605 __func__, enable ? "enable" : "disable");
606 return -EINVAL;
607 }
608
609 pr_debug("%s: cpufreq BOOST %s\n",
610 __func__, enable ? "enabled" : "disabled");
611
612 return count;
613}
614define_one_global_rw(boost);
615
616static ssize_t show_local_boost(struct cpufreq_policy *policy, char *buf)
617{
618 return sysfs_emit(buf, "%d\n", policy->boost_enabled);
619}
620
621static ssize_t store_local_boost(struct cpufreq_policy *policy,
622 const char *buf, size_t count)
623{
624 int ret;
625 bool enable;
626
627 if (kstrtobool(buf, &enable))
628 return -EINVAL;
629
630 if (!cpufreq_driver->boost_enabled)
631 return -EINVAL;
632
633 if (policy->boost_enabled == enable)
634 return count;
635
636 policy->boost_enabled = enable;
637
638 cpus_read_lock();
639 ret = cpufreq_driver->set_boost(policy, enable);
640 cpus_read_unlock();
641
642 if (ret) {
643 policy->boost_enabled = !policy->boost_enabled;
644 return ret;
645 }
646
647 return count;
648}
649
650static struct freq_attr local_boost = __ATTR(boost, 0644, show_local_boost, store_local_boost);
651
652static struct cpufreq_governor *find_governor(const char *str_governor)
653{
654 struct cpufreq_governor *t;
655
656 for_each_governor(t)
657 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
658 return t;
659
660 return NULL;
661}
662
663static struct cpufreq_governor *get_governor(const char *str_governor)
664{
665 struct cpufreq_governor *t;
666
667 mutex_lock(&cpufreq_governor_mutex);
668 t = find_governor(str_governor);
669 if (!t)
670 goto unlock;
671
672 if (!try_module_get(t->owner))
673 t = NULL;
674
675unlock:
676 mutex_unlock(&cpufreq_governor_mutex);
677
678 return t;
679}
680
681static unsigned int cpufreq_parse_policy(char *str_governor)
682{
683 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
684 return CPUFREQ_POLICY_PERFORMANCE;
685
686 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
687 return CPUFREQ_POLICY_POWERSAVE;
688
689 return CPUFREQ_POLICY_UNKNOWN;
690}
691
692/**
693 * cpufreq_parse_governor - parse a governor string only for has_target()
694 * @str_governor: Governor name.
695 */
696static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
697{
698 struct cpufreq_governor *t;
699
700 t = get_governor(str_governor);
701 if (t)
702 return t;
703
704 if (request_module("cpufreq_%s", str_governor))
705 return NULL;
706
707 return get_governor(str_governor);
708}
709
710/*
711 * cpufreq_per_cpu_attr_read() / show_##file_name() -
712 * print out cpufreq information
713 *
714 * Write out information from cpufreq_driver->policy[cpu]; object must be
715 * "unsigned int".
716 */
717
718#define show_one(file_name, object) \
719static ssize_t show_##file_name \
720(struct cpufreq_policy *policy, char *buf) \
721{ \
722 return sysfs_emit(buf, "%u\n", policy->object); \
723}
724
725show_one(cpuinfo_min_freq, cpuinfo.min_freq);
726show_one(cpuinfo_max_freq, cpuinfo.max_freq);
727show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
728show_one(scaling_min_freq, min);
729show_one(scaling_max_freq, max);
730
731__weak unsigned int arch_freq_get_on_cpu(int cpu)
732{
733 return 0;
734}
735
736static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
737{
738 ssize_t ret;
739 unsigned int freq;
740
741 freq = arch_freq_get_on_cpu(policy->cpu);
742 if (freq)
743 ret = sysfs_emit(buf, "%u\n", freq);
744 else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
745 ret = sysfs_emit(buf, "%u\n", cpufreq_driver->get(policy->cpu));
746 else
747 ret = sysfs_emit(buf, "%u\n", policy->cur);
748 return ret;
749}
750
751/*
752 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
753 */
754#define store_one(file_name, object) \
755static ssize_t store_##file_name \
756(struct cpufreq_policy *policy, const char *buf, size_t count) \
757{ \
758 unsigned long val; \
759 int ret; \
760 \
761 ret = kstrtoul(buf, 0, &val); \
762 if (ret) \
763 return ret; \
764 \
765 ret = freq_qos_update_request(policy->object##_freq_req, val);\
766 return ret >= 0 ? count : ret; \
767}
768
769store_one(scaling_min_freq, min);
770store_one(scaling_max_freq, max);
771
772/*
773 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
774 */
775static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
776 char *buf)
777{
778 unsigned int cur_freq = __cpufreq_get(policy);
779
780 if (cur_freq)
781 return sysfs_emit(buf, "%u\n", cur_freq);
782
783 return sysfs_emit(buf, "<unknown>\n");
784}
785
786/*
787 * show_scaling_governor - show the current policy for the specified CPU
788 */
789static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
790{
791 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
792 return sysfs_emit(buf, "powersave\n");
793 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
794 return sysfs_emit(buf, "performance\n");
795 else if (policy->governor)
796 return sysfs_emit(buf, "%s\n", policy->governor->name);
797 return -EINVAL;
798}
799
800/*
801 * store_scaling_governor - store policy for the specified CPU
802 */
803static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
804 const char *buf, size_t count)
805{
806 char str_governor[16];
807 int ret;
808
809 ret = sscanf(buf, "%15s", str_governor);
810 if (ret != 1)
811 return -EINVAL;
812
813 if (cpufreq_driver->setpolicy) {
814 unsigned int new_pol;
815
816 new_pol = cpufreq_parse_policy(str_governor);
817 if (!new_pol)
818 return -EINVAL;
819
820 ret = cpufreq_set_policy(policy, NULL, new_pol);
821 } else {
822 struct cpufreq_governor *new_gov;
823
824 new_gov = cpufreq_parse_governor(str_governor);
825 if (!new_gov)
826 return -EINVAL;
827
828 ret = cpufreq_set_policy(policy, new_gov,
829 CPUFREQ_POLICY_UNKNOWN);
830
831 module_put(new_gov->owner);
832 }
833
834 return ret ? ret : count;
835}
836
837/*
838 * show_scaling_driver - show the cpufreq driver currently loaded
839 */
840static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
841{
842 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
843}
844
845/*
846 * show_scaling_available_governors - show the available CPUfreq governors
847 */
848static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
849 char *buf)
850{
851 ssize_t i = 0;
852 struct cpufreq_governor *t;
853
854 if (!has_target()) {
855 i += sysfs_emit(buf, "performance powersave");
856 goto out;
857 }
858
859 mutex_lock(&cpufreq_governor_mutex);
860 for_each_governor(t) {
861 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
862 - (CPUFREQ_NAME_LEN + 2)))
863 break;
864 i += sysfs_emit_at(buf, i, "%s ", t->name);
865 }
866 mutex_unlock(&cpufreq_governor_mutex);
867out:
868 i += sysfs_emit_at(buf, i, "\n");
869 return i;
870}
871
872ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
873{
874 ssize_t i = 0;
875 unsigned int cpu;
876
877 for_each_cpu(cpu, mask) {
878 i += sysfs_emit_at(buf, i, "%u ", cpu);
879 if (i >= (PAGE_SIZE - 5))
880 break;
881 }
882
883 /* Remove the extra space at the end */
884 i--;
885
886 i += sysfs_emit_at(buf, i, "\n");
887 return i;
888}
889EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
890
891/*
892 * show_related_cpus - show the CPUs affected by each transition even if
893 * hw coordination is in use
894 */
895static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
896{
897 return cpufreq_show_cpus(policy->related_cpus, buf);
898}
899
900/*
901 * show_affected_cpus - show the CPUs affected by each transition
902 */
903static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
904{
905 return cpufreq_show_cpus(policy->cpus, buf);
906}
907
908static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
909 const char *buf, size_t count)
910{
911 unsigned int freq = 0;
912 unsigned int ret;
913
914 if (!policy->governor || !policy->governor->store_setspeed)
915 return -EINVAL;
916
917 ret = sscanf(buf, "%u", &freq);
918 if (ret != 1)
919 return -EINVAL;
920
921 policy->governor->store_setspeed(policy, freq);
922
923 return count;
924}
925
926static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
927{
928 if (!policy->governor || !policy->governor->show_setspeed)
929 return sysfs_emit(buf, "<unsupported>\n");
930
931 return policy->governor->show_setspeed(policy, buf);
932}
933
934/*
935 * show_bios_limit - show the current cpufreq HW/BIOS limitation
936 */
937static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
938{
939 unsigned int limit;
940 int ret;
941 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
942 if (!ret)
943 return sysfs_emit(buf, "%u\n", limit);
944 return sysfs_emit(buf, "%u\n", policy->cpuinfo.max_freq);
945}
946
947cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
948cpufreq_freq_attr_ro(cpuinfo_min_freq);
949cpufreq_freq_attr_ro(cpuinfo_max_freq);
950cpufreq_freq_attr_ro(cpuinfo_transition_latency);
951cpufreq_freq_attr_ro(scaling_available_governors);
952cpufreq_freq_attr_ro(scaling_driver);
953cpufreq_freq_attr_ro(scaling_cur_freq);
954cpufreq_freq_attr_ro(bios_limit);
955cpufreq_freq_attr_ro(related_cpus);
956cpufreq_freq_attr_ro(affected_cpus);
957cpufreq_freq_attr_rw(scaling_min_freq);
958cpufreq_freq_attr_rw(scaling_max_freq);
959cpufreq_freq_attr_rw(scaling_governor);
960cpufreq_freq_attr_rw(scaling_setspeed);
961
962static struct attribute *cpufreq_attrs[] = {
963 &cpuinfo_min_freq.attr,
964 &cpuinfo_max_freq.attr,
965 &cpuinfo_transition_latency.attr,
966 &scaling_min_freq.attr,
967 &scaling_max_freq.attr,
968 &affected_cpus.attr,
969 &related_cpus.attr,
970 &scaling_governor.attr,
971 &scaling_driver.attr,
972 &scaling_available_governors.attr,
973 &scaling_setspeed.attr,
974 NULL
975};
976ATTRIBUTE_GROUPS(cpufreq);
977
978#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
979#define to_attr(a) container_of(a, struct freq_attr, attr)
980
981static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
982{
983 struct cpufreq_policy *policy = to_policy(kobj);
984 struct freq_attr *fattr = to_attr(attr);
985 ssize_t ret = -EBUSY;
986
987 if (!fattr->show)
988 return -EIO;
989
990 down_read(&policy->rwsem);
991 if (likely(!policy_is_inactive(policy)))
992 ret = fattr->show(policy, buf);
993 up_read(&policy->rwsem);
994
995 return ret;
996}
997
998static ssize_t store(struct kobject *kobj, struct attribute *attr,
999 const char *buf, size_t count)
1000{
1001 struct cpufreq_policy *policy = to_policy(kobj);
1002 struct freq_attr *fattr = to_attr(attr);
1003 ssize_t ret = -EBUSY;
1004
1005 if (!fattr->store)
1006 return -EIO;
1007
1008 down_write(&policy->rwsem);
1009 if (likely(!policy_is_inactive(policy)))
1010 ret = fattr->store(policy, buf, count);
1011 up_write(&policy->rwsem);
1012
1013 return ret;
1014}
1015
1016static void cpufreq_sysfs_release(struct kobject *kobj)
1017{
1018 struct cpufreq_policy *policy = to_policy(kobj);
1019 pr_debug("last reference is dropped\n");
1020 complete(&policy->kobj_unregister);
1021}
1022
1023static const struct sysfs_ops sysfs_ops = {
1024 .show = show,
1025 .store = store,
1026};
1027
1028static const struct kobj_type ktype_cpufreq = {
1029 .sysfs_ops = &sysfs_ops,
1030 .default_groups = cpufreq_groups,
1031 .release = cpufreq_sysfs_release,
1032};
1033
1034static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
1035 struct device *dev)
1036{
1037 if (unlikely(!dev))
1038 return;
1039
1040 if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1041 return;
1042
1043 dev_dbg(dev, "%s: Adding symlink\n", __func__);
1044 if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1045 dev_err(dev, "cpufreq symlink creation failed\n");
1046}
1047
1048static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu,
1049 struct device *dev)
1050{
1051 dev_dbg(dev, "%s: Removing symlink\n", __func__);
1052 sysfs_remove_link(&dev->kobj, "cpufreq");
1053 cpumask_clear_cpu(cpu, policy->real_cpus);
1054}
1055
1056static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1057{
1058 struct freq_attr **drv_attr;
1059 int ret = 0;
1060
1061 /* set up files for this cpu device */
1062 drv_attr = cpufreq_driver->attr;
1063 while (drv_attr && *drv_attr) {
1064 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1065 if (ret)
1066 return ret;
1067 drv_attr++;
1068 }
1069 if (cpufreq_driver->get) {
1070 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1071 if (ret)
1072 return ret;
1073 }
1074
1075 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1076 if (ret)
1077 return ret;
1078
1079 if (cpufreq_driver->bios_limit) {
1080 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1081 if (ret)
1082 return ret;
1083 }
1084
1085 if (cpufreq_boost_supported()) {
1086 ret = sysfs_create_file(&policy->kobj, &local_boost.attr);
1087 if (ret)
1088 return ret;
1089 }
1090
1091 return 0;
1092}
1093
1094static int cpufreq_init_policy(struct cpufreq_policy *policy)
1095{
1096 struct cpufreq_governor *gov = NULL;
1097 unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1098 int ret;
1099
1100 if (has_target()) {
1101 /* Update policy governor to the one used before hotplug. */
1102 gov = get_governor(policy->last_governor);
1103 if (gov) {
1104 pr_debug("Restoring governor %s for cpu %d\n",
1105 gov->name, policy->cpu);
1106 } else {
1107 gov = get_governor(default_governor);
1108 }
1109
1110 if (!gov) {
1111 gov = cpufreq_default_governor();
1112 __module_get(gov->owner);
1113 }
1114
1115 } else {
1116
1117 /* Use the default policy if there is no last_policy. */
1118 if (policy->last_policy) {
1119 pol = policy->last_policy;
1120 } else {
1121 pol = cpufreq_parse_policy(default_governor);
1122 /*
1123 * In case the default governor is neither "performance"
1124 * nor "powersave", fall back to the initial policy
1125 * value set by the driver.
1126 */
1127 if (pol == CPUFREQ_POLICY_UNKNOWN)
1128 pol = policy->policy;
1129 }
1130 if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1131 pol != CPUFREQ_POLICY_POWERSAVE)
1132 return -ENODATA;
1133 }
1134
1135 ret = cpufreq_set_policy(policy, gov, pol);
1136 if (gov)
1137 module_put(gov->owner);
1138
1139 return ret;
1140}
1141
1142static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1143{
1144 int ret = 0;
1145
1146 /* Has this CPU been taken care of already? */
1147 if (cpumask_test_cpu(cpu, policy->cpus))
1148 return 0;
1149
1150 down_write(&policy->rwsem);
1151 if (has_target())
1152 cpufreq_stop_governor(policy);
1153
1154 cpumask_set_cpu(cpu, policy->cpus);
1155
1156 if (has_target()) {
1157 ret = cpufreq_start_governor(policy);
1158 if (ret)
1159 pr_err("%s: Failed to start governor\n", __func__);
1160 }
1161 up_write(&policy->rwsem);
1162 return ret;
1163}
1164
1165void refresh_frequency_limits(struct cpufreq_policy *policy)
1166{
1167 if (!policy_is_inactive(policy)) {
1168 pr_debug("updating policy for CPU %u\n", policy->cpu);
1169
1170 cpufreq_set_policy(policy, policy->governor, policy->policy);
1171 }
1172}
1173EXPORT_SYMBOL(refresh_frequency_limits);
1174
1175static void handle_update(struct work_struct *work)
1176{
1177 struct cpufreq_policy *policy =
1178 container_of(work, struct cpufreq_policy, update);
1179
1180 pr_debug("handle_update for cpu %u called\n", policy->cpu);
1181 down_write(&policy->rwsem);
1182 refresh_frequency_limits(policy);
1183 up_write(&policy->rwsem);
1184}
1185
1186static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1187 void *data)
1188{
1189 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1190
1191 schedule_work(&policy->update);
1192 return 0;
1193}
1194
1195static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1196 void *data)
1197{
1198 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1199
1200 schedule_work(&policy->update);
1201 return 0;
1202}
1203
1204static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1205{
1206 struct kobject *kobj;
1207 struct completion *cmp;
1208
1209 down_write(&policy->rwsem);
1210 cpufreq_stats_free_table(policy);
1211 kobj = &policy->kobj;
1212 cmp = &policy->kobj_unregister;
1213 up_write(&policy->rwsem);
1214 kobject_put(kobj);
1215
1216 /*
1217 * We need to make sure that the underlying kobj is
1218 * actually not referenced anymore by anybody before we
1219 * proceed with unloading.
1220 */
1221 pr_debug("waiting for dropping of refcount\n");
1222 wait_for_completion(cmp);
1223 pr_debug("wait complete\n");
1224}
1225
1226static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1227{
1228 struct cpufreq_policy *policy;
1229 struct device *dev = get_cpu_device(cpu);
1230 int ret;
1231
1232 if (!dev)
1233 return NULL;
1234
1235 policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1236 if (!policy)
1237 return NULL;
1238
1239 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1240 goto err_free_policy;
1241
1242 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1243 goto err_free_cpumask;
1244
1245 if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1246 goto err_free_rcpumask;
1247
1248 init_completion(&policy->kobj_unregister);
1249 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1250 cpufreq_global_kobject, "policy%u", cpu);
1251 if (ret) {
1252 dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1253 /*
1254 * The entire policy object will be freed below, but the extra
1255 * memory allocated for the kobject name needs to be freed by
1256 * releasing the kobject.
1257 */
1258 kobject_put(&policy->kobj);
1259 goto err_free_real_cpus;
1260 }
1261
1262 freq_constraints_init(&policy->constraints);
1263
1264 policy->nb_min.notifier_call = cpufreq_notifier_min;
1265 policy->nb_max.notifier_call = cpufreq_notifier_max;
1266
1267 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1268 &policy->nb_min);
1269 if (ret) {
1270 dev_err(dev, "Failed to register MIN QoS notifier: %d (CPU%u)\n",
1271 ret, cpu);
1272 goto err_kobj_remove;
1273 }
1274
1275 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1276 &policy->nb_max);
1277 if (ret) {
1278 dev_err(dev, "Failed to register MAX QoS notifier: %d (CPU%u)\n",
1279 ret, cpu);
1280 goto err_min_qos_notifier;
1281 }
1282
1283 INIT_LIST_HEAD(&policy->policy_list);
1284 init_rwsem(&policy->rwsem);
1285 spin_lock_init(&policy->transition_lock);
1286 init_waitqueue_head(&policy->transition_wait);
1287 INIT_WORK(&policy->update, handle_update);
1288
1289 policy->cpu = cpu;
1290 return policy;
1291
1292err_min_qos_notifier:
1293 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1294 &policy->nb_min);
1295err_kobj_remove:
1296 cpufreq_policy_put_kobj(policy);
1297err_free_real_cpus:
1298 free_cpumask_var(policy->real_cpus);
1299err_free_rcpumask:
1300 free_cpumask_var(policy->related_cpus);
1301err_free_cpumask:
1302 free_cpumask_var(policy->cpus);
1303err_free_policy:
1304 kfree(policy);
1305
1306 return NULL;
1307}
1308
1309static void cpufreq_policy_free(struct cpufreq_policy *policy)
1310{
1311 unsigned long flags;
1312 int cpu;
1313
1314 /*
1315 * The callers must ensure the policy is inactive by now, to avoid any
1316 * races with show()/store() callbacks.
1317 */
1318 if (unlikely(!policy_is_inactive(policy)))
1319 pr_warn("%s: Freeing active policy\n", __func__);
1320
1321 /* Remove policy from list */
1322 write_lock_irqsave(&cpufreq_driver_lock, flags);
1323 list_del(&policy->policy_list);
1324
1325 for_each_cpu(cpu, policy->related_cpus)
1326 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1327 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1328
1329 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1330 &policy->nb_max);
1331 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1332 &policy->nb_min);
1333
1334 /* Cancel any pending policy->update work before freeing the policy. */
1335 cancel_work_sync(&policy->update);
1336
1337 if (policy->max_freq_req) {
1338 /*
1339 * Remove max_freq_req after sending CPUFREQ_REMOVE_POLICY
1340 * notification, since CPUFREQ_CREATE_POLICY notification was
1341 * sent after adding max_freq_req earlier.
1342 */
1343 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1344 CPUFREQ_REMOVE_POLICY, policy);
1345 freq_qos_remove_request(policy->max_freq_req);
1346 }
1347
1348 freq_qos_remove_request(policy->min_freq_req);
1349 kfree(policy->min_freq_req);
1350
1351 cpufreq_policy_put_kobj(policy);
1352 free_cpumask_var(policy->real_cpus);
1353 free_cpumask_var(policy->related_cpus);
1354 free_cpumask_var(policy->cpus);
1355 kfree(policy);
1356}
1357
1358static int cpufreq_online(unsigned int cpu)
1359{
1360 struct cpufreq_policy *policy;
1361 bool new_policy;
1362 unsigned long flags;
1363 unsigned int j;
1364 int ret;
1365
1366 pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1367
1368 /* Check if this CPU already has a policy to manage it */
1369 policy = per_cpu(cpufreq_cpu_data, cpu);
1370 if (policy) {
1371 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1372 if (!policy_is_inactive(policy))
1373 return cpufreq_add_policy_cpu(policy, cpu);
1374
1375 /* This is the only online CPU for the policy. Start over. */
1376 new_policy = false;
1377 down_write(&policy->rwsem);
1378 policy->cpu = cpu;
1379 policy->governor = NULL;
1380 } else {
1381 new_policy = true;
1382 policy = cpufreq_policy_alloc(cpu);
1383 if (!policy)
1384 return -ENOMEM;
1385 down_write(&policy->rwsem);
1386 }
1387
1388 if (!new_policy && cpufreq_driver->online) {
1389 /* Recover policy->cpus using related_cpus */
1390 cpumask_copy(policy->cpus, policy->related_cpus);
1391
1392 ret = cpufreq_driver->online(policy);
1393 if (ret) {
1394 pr_debug("%s: %d: initialization failed\n", __func__,
1395 __LINE__);
1396 goto out_exit_policy;
1397 }
1398 } else {
1399 cpumask_copy(policy->cpus, cpumask_of(cpu));
1400
1401 /*
1402 * Call driver. From then on the cpufreq must be able
1403 * to accept all calls to ->verify and ->setpolicy for this CPU.
1404 */
1405 ret = cpufreq_driver->init(policy);
1406 if (ret) {
1407 pr_debug("%s: %d: initialization failed\n", __func__,
1408 __LINE__);
1409 goto out_free_policy;
1410 }
1411
1412 /* Let the per-policy boost flag mirror the cpufreq_driver boost during init */
1413 if (cpufreq_boost_enabled() && policy_has_boost_freq(policy))
1414 policy->boost_enabled = true;
1415
1416 /*
1417 * The initialization has succeeded and the policy is online.
1418 * If there is a problem with its frequency table, take it
1419 * offline and drop it.
1420 */
1421 ret = cpufreq_table_validate_and_sort(policy);
1422 if (ret)
1423 goto out_offline_policy;
1424
1425 /* related_cpus should at least include policy->cpus. */
1426 cpumask_copy(policy->related_cpus, policy->cpus);
1427 }
1428
1429 /*
1430 * affected cpus must always be the one, which are online. We aren't
1431 * managing offline cpus here.
1432 */
1433 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1434
1435 if (new_policy) {
1436 for_each_cpu(j, policy->related_cpus) {
1437 per_cpu(cpufreq_cpu_data, j) = policy;
1438 add_cpu_dev_symlink(policy, j, get_cpu_device(j));
1439 }
1440
1441 policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1442 GFP_KERNEL);
1443 if (!policy->min_freq_req) {
1444 ret = -ENOMEM;
1445 goto out_destroy_policy;
1446 }
1447
1448 ret = freq_qos_add_request(&policy->constraints,
1449 policy->min_freq_req, FREQ_QOS_MIN,
1450 FREQ_QOS_MIN_DEFAULT_VALUE);
1451 if (ret < 0) {
1452 /*
1453 * So we don't call freq_qos_remove_request() for an
1454 * uninitialized request.
1455 */
1456 kfree(policy->min_freq_req);
1457 policy->min_freq_req = NULL;
1458 goto out_destroy_policy;
1459 }
1460
1461 /*
1462 * This must be initialized right here to avoid calling
1463 * freq_qos_remove_request() on uninitialized request in case
1464 * of errors.
1465 */
1466 policy->max_freq_req = policy->min_freq_req + 1;
1467
1468 ret = freq_qos_add_request(&policy->constraints,
1469 policy->max_freq_req, FREQ_QOS_MAX,
1470 FREQ_QOS_MAX_DEFAULT_VALUE);
1471 if (ret < 0) {
1472 policy->max_freq_req = NULL;
1473 goto out_destroy_policy;
1474 }
1475
1476 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1477 CPUFREQ_CREATE_POLICY, policy);
1478 }
1479
1480 if (cpufreq_driver->get && has_target()) {
1481 policy->cur = cpufreq_driver->get(policy->cpu);
1482 if (!policy->cur) {
1483 ret = -EIO;
1484 pr_err("%s: ->get() failed\n", __func__);
1485 goto out_destroy_policy;
1486 }
1487 }
1488
1489 /*
1490 * Sometimes boot loaders set CPU frequency to a value outside of
1491 * frequency table present with cpufreq core. In such cases CPU might be
1492 * unstable if it has to run on that frequency for long duration of time
1493 * and so its better to set it to a frequency which is specified in
1494 * freq-table. This also makes cpufreq stats inconsistent as
1495 * cpufreq-stats would fail to register because current frequency of CPU
1496 * isn't found in freq-table.
1497 *
1498 * Because we don't want this change to effect boot process badly, we go
1499 * for the next freq which is >= policy->cur ('cur' must be set by now,
1500 * otherwise we will end up setting freq to lowest of the table as 'cur'
1501 * is initialized to zero).
1502 *
1503 * We are passing target-freq as "policy->cur - 1" otherwise
1504 * __cpufreq_driver_target() would simply fail, as policy->cur will be
1505 * equal to target-freq.
1506 */
1507 if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1508 && has_target()) {
1509 unsigned int old_freq = policy->cur;
1510
1511 /* Are we running at unknown frequency ? */
1512 ret = cpufreq_frequency_table_get_index(policy, old_freq);
1513 if (ret == -EINVAL) {
1514 ret = __cpufreq_driver_target(policy, old_freq - 1,
1515 CPUFREQ_RELATION_L);
1516
1517 /*
1518 * Reaching here after boot in a few seconds may not
1519 * mean that system will remain stable at "unknown"
1520 * frequency for longer duration. Hence, a BUG_ON().
1521 */
1522 BUG_ON(ret);
1523 pr_info("%s: CPU%d: Running at unlisted initial frequency: %u kHz, changing to: %u kHz\n",
1524 __func__, policy->cpu, old_freq, policy->cur);
1525 }
1526 }
1527
1528 if (new_policy) {
1529 ret = cpufreq_add_dev_interface(policy);
1530 if (ret)
1531 goto out_destroy_policy;
1532
1533 cpufreq_stats_create_table(policy);
1534
1535 write_lock_irqsave(&cpufreq_driver_lock, flags);
1536 list_add(&policy->policy_list, &cpufreq_policy_list);
1537 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1538
1539 /*
1540 * Register with the energy model before
1541 * sugov_eas_rebuild_sd() is called, which will result
1542 * in rebuilding of the sched domains, which should only be done
1543 * once the energy model is properly initialized for the policy
1544 * first.
1545 *
1546 * Also, this should be called before the policy is registered
1547 * with cooling framework.
1548 */
1549 if (cpufreq_driver->register_em)
1550 cpufreq_driver->register_em(policy);
1551 }
1552
1553 ret = cpufreq_init_policy(policy);
1554 if (ret) {
1555 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1556 __func__, cpu, ret);
1557 goto out_destroy_policy;
1558 }
1559
1560 up_write(&policy->rwsem);
1561
1562 kobject_uevent(&policy->kobj, KOBJ_ADD);
1563
1564 /* Callback for handling stuff after policy is ready */
1565 if (cpufreq_driver->ready)
1566 cpufreq_driver->ready(policy);
1567
1568 /* Register cpufreq cooling only for a new policy */
1569 if (new_policy && cpufreq_thermal_control_enabled(cpufreq_driver))
1570 policy->cdev = of_cpufreq_cooling_register(policy);
1571
1572 pr_debug("initialization complete\n");
1573
1574 return 0;
1575
1576out_destroy_policy:
1577 for_each_cpu(j, policy->real_cpus)
1578 remove_cpu_dev_symlink(policy, j, get_cpu_device(j));
1579
1580out_offline_policy:
1581 if (cpufreq_driver->offline)
1582 cpufreq_driver->offline(policy);
1583
1584out_exit_policy:
1585 if (cpufreq_driver->exit)
1586 cpufreq_driver->exit(policy);
1587
1588out_free_policy:
1589 cpumask_clear(policy->cpus);
1590 up_write(&policy->rwsem);
1591
1592 cpufreq_policy_free(policy);
1593 return ret;
1594}
1595
1596/**
1597 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1598 * @dev: CPU device.
1599 * @sif: Subsystem interface structure pointer (not used)
1600 */
1601static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1602{
1603 struct cpufreq_policy *policy;
1604 unsigned cpu = dev->id;
1605 int ret;
1606
1607 dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1608
1609 if (cpu_online(cpu)) {
1610 ret = cpufreq_online(cpu);
1611 if (ret)
1612 return ret;
1613 }
1614
1615 /* Create sysfs link on CPU registration */
1616 policy = per_cpu(cpufreq_cpu_data, cpu);
1617 if (policy)
1618 add_cpu_dev_symlink(policy, cpu, dev);
1619
1620 return 0;
1621}
1622
1623static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1624{
1625 int ret;
1626
1627 if (has_target())
1628 cpufreq_stop_governor(policy);
1629
1630 cpumask_clear_cpu(cpu, policy->cpus);
1631
1632 if (!policy_is_inactive(policy)) {
1633 /* Nominate a new CPU if necessary. */
1634 if (cpu == policy->cpu)
1635 policy->cpu = cpumask_any(policy->cpus);
1636
1637 /* Start the governor again for the active policy. */
1638 if (has_target()) {
1639 ret = cpufreq_start_governor(policy);
1640 if (ret)
1641 pr_err("%s: Failed to start governor\n", __func__);
1642 }
1643
1644 return;
1645 }
1646
1647 if (has_target())
1648 strscpy(policy->last_governor, policy->governor->name,
1649 CPUFREQ_NAME_LEN);
1650 else
1651 policy->last_policy = policy->policy;
1652
1653 if (has_target())
1654 cpufreq_exit_governor(policy);
1655
1656 /*
1657 * Perform the ->offline() during light-weight tear-down, as
1658 * that allows fast recovery when the CPU comes back.
1659 */
1660 if (cpufreq_driver->offline) {
1661 cpufreq_driver->offline(policy);
1662 return;
1663 }
1664
1665 if (cpufreq_driver->exit)
1666 cpufreq_driver->exit(policy);
1667
1668 policy->freq_table = NULL;
1669}
1670
1671static int cpufreq_offline(unsigned int cpu)
1672{
1673 struct cpufreq_policy *policy;
1674
1675 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1676
1677 policy = cpufreq_cpu_get_raw(cpu);
1678 if (!policy) {
1679 pr_debug("%s: No cpu_data found\n", __func__);
1680 return 0;
1681 }
1682
1683 down_write(&policy->rwsem);
1684
1685 __cpufreq_offline(cpu, policy);
1686
1687 up_write(&policy->rwsem);
1688 return 0;
1689}
1690
1691/*
1692 * cpufreq_remove_dev - remove a CPU device
1693 *
1694 * Removes the cpufreq interface for a CPU device.
1695 */
1696static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1697{
1698 unsigned int cpu = dev->id;
1699 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1700
1701 if (!policy)
1702 return;
1703
1704 down_write(&policy->rwsem);
1705
1706 if (cpu_online(cpu))
1707 __cpufreq_offline(cpu, policy);
1708
1709 remove_cpu_dev_symlink(policy, cpu, dev);
1710
1711 if (!cpumask_empty(policy->real_cpus)) {
1712 up_write(&policy->rwsem);
1713 return;
1714 }
1715
1716 /*
1717 * Unregister cpufreq cooling once all the CPUs of the policy are
1718 * removed.
1719 */
1720 if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1721 cpufreq_cooling_unregister(policy->cdev);
1722 policy->cdev = NULL;
1723 }
1724
1725 /* We did light-weight exit earlier, do full tear down now */
1726 if (cpufreq_driver->offline && cpufreq_driver->exit)
1727 cpufreq_driver->exit(policy);
1728
1729 up_write(&policy->rwsem);
1730
1731 cpufreq_policy_free(policy);
1732}
1733
1734/**
1735 * cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1736 * @policy: Policy managing CPUs.
1737 * @new_freq: New CPU frequency.
1738 *
1739 * Adjust to the current frequency first and clean up later by either calling
1740 * cpufreq_update_policy(), or scheduling handle_update().
1741 */
1742static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1743 unsigned int new_freq)
1744{
1745 struct cpufreq_freqs freqs;
1746
1747 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1748 policy->cur, new_freq);
1749
1750 freqs.old = policy->cur;
1751 freqs.new = new_freq;
1752
1753 cpufreq_freq_transition_begin(policy, &freqs);
1754 cpufreq_freq_transition_end(policy, &freqs, 0);
1755}
1756
1757static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1758{
1759 unsigned int new_freq;
1760
1761 new_freq = cpufreq_driver->get(policy->cpu);
1762 if (!new_freq)
1763 return 0;
1764
1765 /*
1766 * If fast frequency switching is used with the given policy, the check
1767 * against policy->cur is pointless, so skip it in that case.
1768 */
1769 if (policy->fast_switch_enabled || !has_target())
1770 return new_freq;
1771
1772 if (policy->cur != new_freq) {
1773 /*
1774 * For some platforms, the frequency returned by hardware may be
1775 * slightly different from what is provided in the frequency
1776 * table, for example hardware may return 499 MHz instead of 500
1777 * MHz. In such cases it is better to avoid getting into
1778 * unnecessary frequency updates.
1779 */
1780 if (abs(policy->cur - new_freq) < KHZ_PER_MHZ)
1781 return policy->cur;
1782
1783 cpufreq_out_of_sync(policy, new_freq);
1784 if (update)
1785 schedule_work(&policy->update);
1786 }
1787
1788 return new_freq;
1789}
1790
1791/**
1792 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1793 * @cpu: CPU number
1794 *
1795 * This is the last known freq, without actually getting it from the driver.
1796 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1797 */
1798unsigned int cpufreq_quick_get(unsigned int cpu)
1799{
1800 struct cpufreq_policy *policy;
1801 unsigned int ret_freq = 0;
1802 unsigned long flags;
1803
1804 read_lock_irqsave(&cpufreq_driver_lock, flags);
1805
1806 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1807 ret_freq = cpufreq_driver->get(cpu);
1808 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1809 return ret_freq;
1810 }
1811
1812 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1813
1814 policy = cpufreq_cpu_get(cpu);
1815 if (policy) {
1816 ret_freq = policy->cur;
1817 cpufreq_cpu_put(policy);
1818 }
1819
1820 return ret_freq;
1821}
1822EXPORT_SYMBOL(cpufreq_quick_get);
1823
1824/**
1825 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1826 * @cpu: CPU number
1827 *
1828 * Just return the max possible frequency for a given CPU.
1829 */
1830unsigned int cpufreq_quick_get_max(unsigned int cpu)
1831{
1832 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1833 unsigned int ret_freq = 0;
1834
1835 if (policy) {
1836 ret_freq = policy->max;
1837 cpufreq_cpu_put(policy);
1838 }
1839
1840 return ret_freq;
1841}
1842EXPORT_SYMBOL(cpufreq_quick_get_max);
1843
1844/**
1845 * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1846 * @cpu: CPU number
1847 *
1848 * The default return value is the max_freq field of cpuinfo.
1849 */
1850__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1851{
1852 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1853 unsigned int ret_freq = 0;
1854
1855 if (policy) {
1856 ret_freq = policy->cpuinfo.max_freq;
1857 cpufreq_cpu_put(policy);
1858 }
1859
1860 return ret_freq;
1861}
1862EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1863
1864static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1865{
1866 if (unlikely(policy_is_inactive(policy)))
1867 return 0;
1868
1869 return cpufreq_verify_current_freq(policy, true);
1870}
1871
1872/**
1873 * cpufreq_get - get the current CPU frequency (in kHz)
1874 * @cpu: CPU number
1875 *
1876 * Get the CPU current (static) CPU frequency
1877 */
1878unsigned int cpufreq_get(unsigned int cpu)
1879{
1880 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1881 unsigned int ret_freq = 0;
1882
1883 if (policy) {
1884 down_read(&policy->rwsem);
1885 if (cpufreq_driver->get)
1886 ret_freq = __cpufreq_get(policy);
1887 up_read(&policy->rwsem);
1888
1889 cpufreq_cpu_put(policy);
1890 }
1891
1892 return ret_freq;
1893}
1894EXPORT_SYMBOL(cpufreq_get);
1895
1896static struct subsys_interface cpufreq_interface = {
1897 .name = "cpufreq",
1898 .subsys = &cpu_subsys,
1899 .add_dev = cpufreq_add_dev,
1900 .remove_dev = cpufreq_remove_dev,
1901};
1902
1903/*
1904 * In case platform wants some specific frequency to be configured
1905 * during suspend..
1906 */
1907int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1908{
1909 int ret;
1910
1911 if (!policy->suspend_freq) {
1912 pr_debug("%s: suspend_freq not defined\n", __func__);
1913 return 0;
1914 }
1915
1916 pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1917 policy->suspend_freq);
1918
1919 ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1920 CPUFREQ_RELATION_H);
1921 if (ret)
1922 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1923 __func__, policy->suspend_freq, ret);
1924
1925 return ret;
1926}
1927EXPORT_SYMBOL(cpufreq_generic_suspend);
1928
1929/**
1930 * cpufreq_suspend() - Suspend CPUFreq governors.
1931 *
1932 * Called during system wide Suspend/Hibernate cycles for suspending governors
1933 * as some platforms can't change frequency after this point in suspend cycle.
1934 * Because some of the devices (like: i2c, regulators, etc) they use for
1935 * changing frequency are suspended quickly after this point.
1936 */
1937void cpufreq_suspend(void)
1938{
1939 struct cpufreq_policy *policy;
1940
1941 if (!cpufreq_driver)
1942 return;
1943
1944 if (!has_target() && !cpufreq_driver->suspend)
1945 goto suspend;
1946
1947 pr_debug("%s: Suspending Governors\n", __func__);
1948
1949 for_each_active_policy(policy) {
1950 if (has_target()) {
1951 down_write(&policy->rwsem);
1952 cpufreq_stop_governor(policy);
1953 up_write(&policy->rwsem);
1954 }
1955
1956 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1957 pr_err("%s: Failed to suspend driver: %s\n", __func__,
1958 cpufreq_driver->name);
1959 }
1960
1961suspend:
1962 cpufreq_suspended = true;
1963}
1964
1965/**
1966 * cpufreq_resume() - Resume CPUFreq governors.
1967 *
1968 * Called during system wide Suspend/Hibernate cycle for resuming governors that
1969 * are suspended with cpufreq_suspend().
1970 */
1971void cpufreq_resume(void)
1972{
1973 struct cpufreq_policy *policy;
1974 int ret;
1975
1976 if (!cpufreq_driver)
1977 return;
1978
1979 if (unlikely(!cpufreq_suspended))
1980 return;
1981
1982 cpufreq_suspended = false;
1983
1984 if (!has_target() && !cpufreq_driver->resume)
1985 return;
1986
1987 pr_debug("%s: Resuming Governors\n", __func__);
1988
1989 for_each_active_policy(policy) {
1990 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1991 pr_err("%s: Failed to resume driver: %s\n", __func__,
1992 cpufreq_driver->name);
1993 } else if (has_target()) {
1994 down_write(&policy->rwsem);
1995 ret = cpufreq_start_governor(policy);
1996 up_write(&policy->rwsem);
1997
1998 if (ret)
1999 pr_err("%s: Failed to start governor for CPU%u's policy\n",
2000 __func__, policy->cpu);
2001 }
2002 }
2003}
2004
2005/**
2006 * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
2007 * @flags: Flags to test against the current cpufreq driver's flags.
2008 *
2009 * Assumes that the driver is there, so callers must ensure that this is the
2010 * case.
2011 */
2012bool cpufreq_driver_test_flags(u16 flags)
2013{
2014 return !!(cpufreq_driver->flags & flags);
2015}
2016
2017/**
2018 * cpufreq_get_current_driver - Return the current driver's name.
2019 *
2020 * Return the name string of the currently registered cpufreq driver or NULL if
2021 * none.
2022 */
2023const char *cpufreq_get_current_driver(void)
2024{
2025 if (cpufreq_driver)
2026 return cpufreq_driver->name;
2027
2028 return NULL;
2029}
2030EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
2031
2032/**
2033 * cpufreq_get_driver_data - Return current driver data.
2034 *
2035 * Return the private data of the currently registered cpufreq driver, or NULL
2036 * if no cpufreq driver has been registered.
2037 */
2038void *cpufreq_get_driver_data(void)
2039{
2040 if (cpufreq_driver)
2041 return cpufreq_driver->driver_data;
2042
2043 return NULL;
2044}
2045EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
2046
2047/*********************************************************************
2048 * NOTIFIER LISTS INTERFACE *
2049 *********************************************************************/
2050
2051/**
2052 * cpufreq_register_notifier - Register a notifier with cpufreq.
2053 * @nb: notifier function to register.
2054 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2055 *
2056 * Add a notifier to one of two lists: either a list of notifiers that run on
2057 * clock rate changes (once before and once after every transition), or a list
2058 * of notifiers that ron on cpufreq policy changes.
2059 *
2060 * This function may sleep and it has the same return values as
2061 * blocking_notifier_chain_register().
2062 */
2063int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
2064{
2065 int ret;
2066
2067 if (cpufreq_disabled())
2068 return -EINVAL;
2069
2070 switch (list) {
2071 case CPUFREQ_TRANSITION_NOTIFIER:
2072 mutex_lock(&cpufreq_fast_switch_lock);
2073
2074 if (cpufreq_fast_switch_count > 0) {
2075 mutex_unlock(&cpufreq_fast_switch_lock);
2076 return -EBUSY;
2077 }
2078 ret = srcu_notifier_chain_register(
2079 &cpufreq_transition_notifier_list, nb);
2080 if (!ret)
2081 cpufreq_fast_switch_count--;
2082
2083 mutex_unlock(&cpufreq_fast_switch_lock);
2084 break;
2085 case CPUFREQ_POLICY_NOTIFIER:
2086 ret = blocking_notifier_chain_register(
2087 &cpufreq_policy_notifier_list, nb);
2088 break;
2089 default:
2090 ret = -EINVAL;
2091 }
2092
2093 return ret;
2094}
2095EXPORT_SYMBOL(cpufreq_register_notifier);
2096
2097/**
2098 * cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2099 * @nb: notifier block to be unregistered.
2100 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2101 *
2102 * Remove a notifier from one of the cpufreq notifier lists.
2103 *
2104 * This function may sleep and it has the same return values as
2105 * blocking_notifier_chain_unregister().
2106 */
2107int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2108{
2109 int ret;
2110
2111 if (cpufreq_disabled())
2112 return -EINVAL;
2113
2114 switch (list) {
2115 case CPUFREQ_TRANSITION_NOTIFIER:
2116 mutex_lock(&cpufreq_fast_switch_lock);
2117
2118 ret = srcu_notifier_chain_unregister(
2119 &cpufreq_transition_notifier_list, nb);
2120 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2121 cpufreq_fast_switch_count++;
2122
2123 mutex_unlock(&cpufreq_fast_switch_lock);
2124 break;
2125 case CPUFREQ_POLICY_NOTIFIER:
2126 ret = blocking_notifier_chain_unregister(
2127 &cpufreq_policy_notifier_list, nb);
2128 break;
2129 default:
2130 ret = -EINVAL;
2131 }
2132
2133 return ret;
2134}
2135EXPORT_SYMBOL(cpufreq_unregister_notifier);
2136
2137
2138/*********************************************************************
2139 * GOVERNORS *
2140 *********************************************************************/
2141
2142/**
2143 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2144 * @policy: cpufreq policy to switch the frequency for.
2145 * @target_freq: New frequency to set (may be approximate).
2146 *
2147 * Carry out a fast frequency switch without sleeping.
2148 *
2149 * The driver's ->fast_switch() callback invoked by this function must be
2150 * suitable for being called from within RCU-sched read-side critical sections
2151 * and it is expected to select the minimum available frequency greater than or
2152 * equal to @target_freq (CPUFREQ_RELATION_L).
2153 *
2154 * This function must not be called if policy->fast_switch_enabled is unset.
2155 *
2156 * Governors calling this function must guarantee that it will never be invoked
2157 * twice in parallel for the same policy and that it will never be called in
2158 * parallel with either ->target() or ->target_index() for the same policy.
2159 *
2160 * Returns the actual frequency set for the CPU.
2161 *
2162 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2163 * error condition, the hardware configuration must be preserved.
2164 */
2165unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2166 unsigned int target_freq)
2167{
2168 unsigned int freq;
2169 int cpu;
2170
2171 target_freq = clamp_val(target_freq, policy->min, policy->max);
2172 freq = cpufreq_driver->fast_switch(policy, target_freq);
2173
2174 if (!freq)
2175 return 0;
2176
2177 policy->cur = freq;
2178 arch_set_freq_scale(policy->related_cpus, freq,
2179 arch_scale_freq_ref(policy->cpu));
2180 cpufreq_stats_record_transition(policy, freq);
2181
2182 if (trace_cpu_frequency_enabled()) {
2183 for_each_cpu(cpu, policy->cpus)
2184 trace_cpu_frequency(freq, cpu);
2185 }
2186
2187 return freq;
2188}
2189EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2190
2191/**
2192 * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2193 * @cpu: Target CPU.
2194 * @min_perf: Minimum (required) performance level (units of @capacity).
2195 * @target_perf: Target (desired) performance level (units of @capacity).
2196 * @capacity: Capacity of the target CPU.
2197 *
2198 * Carry out a fast performance level switch of @cpu without sleeping.
2199 *
2200 * The driver's ->adjust_perf() callback invoked by this function must be
2201 * suitable for being called from within RCU-sched read-side critical sections
2202 * and it is expected to select a suitable performance level equal to or above
2203 * @min_perf and preferably equal to or below @target_perf.
2204 *
2205 * This function must not be called if policy->fast_switch_enabled is unset.
2206 *
2207 * Governors calling this function must guarantee that it will never be invoked
2208 * twice in parallel for the same CPU and that it will never be called in
2209 * parallel with either ->target() or ->target_index() or ->fast_switch() for
2210 * the same CPU.
2211 */
2212void cpufreq_driver_adjust_perf(unsigned int cpu,
2213 unsigned long min_perf,
2214 unsigned long target_perf,
2215 unsigned long capacity)
2216{
2217 cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2218}
2219
2220/**
2221 * cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2222 *
2223 * Return 'true' if the ->adjust_perf callback is present for the
2224 * current driver or 'false' otherwise.
2225 */
2226bool cpufreq_driver_has_adjust_perf(void)
2227{
2228 return !!cpufreq_driver->adjust_perf;
2229}
2230
2231/* Must set freqs->new to intermediate frequency */
2232static int __target_intermediate(struct cpufreq_policy *policy,
2233 struct cpufreq_freqs *freqs, int index)
2234{
2235 int ret;
2236
2237 freqs->new = cpufreq_driver->get_intermediate(policy, index);
2238
2239 /* We don't need to switch to intermediate freq */
2240 if (!freqs->new)
2241 return 0;
2242
2243 pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2244 __func__, policy->cpu, freqs->old, freqs->new);
2245
2246 cpufreq_freq_transition_begin(policy, freqs);
2247 ret = cpufreq_driver->target_intermediate(policy, index);
2248 cpufreq_freq_transition_end(policy, freqs, ret);
2249
2250 if (ret)
2251 pr_err("%s: Failed to change to intermediate frequency: %d\n",
2252 __func__, ret);
2253
2254 return ret;
2255}
2256
2257static int __target_index(struct cpufreq_policy *policy, int index)
2258{
2259 struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2260 unsigned int restore_freq, intermediate_freq = 0;
2261 unsigned int newfreq = policy->freq_table[index].frequency;
2262 int retval = -EINVAL;
2263 bool notify;
2264
2265 if (newfreq == policy->cur)
2266 return 0;
2267
2268 /* Save last value to restore later on errors */
2269 restore_freq = policy->cur;
2270
2271 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2272 if (notify) {
2273 /* Handle switching to intermediate frequency */
2274 if (cpufreq_driver->get_intermediate) {
2275 retval = __target_intermediate(policy, &freqs, index);
2276 if (retval)
2277 return retval;
2278
2279 intermediate_freq = freqs.new;
2280 /* Set old freq to intermediate */
2281 if (intermediate_freq)
2282 freqs.old = freqs.new;
2283 }
2284
2285 freqs.new = newfreq;
2286 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2287 __func__, policy->cpu, freqs.old, freqs.new);
2288
2289 cpufreq_freq_transition_begin(policy, &freqs);
2290 }
2291
2292 retval = cpufreq_driver->target_index(policy, index);
2293 if (retval)
2294 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2295 retval);
2296
2297 if (notify) {
2298 cpufreq_freq_transition_end(policy, &freqs, retval);
2299
2300 /*
2301 * Failed after setting to intermediate freq? Driver should have
2302 * reverted back to initial frequency and so should we. Check
2303 * here for intermediate_freq instead of get_intermediate, in
2304 * case we haven't switched to intermediate freq at all.
2305 */
2306 if (unlikely(retval && intermediate_freq)) {
2307 freqs.old = intermediate_freq;
2308 freqs.new = restore_freq;
2309 cpufreq_freq_transition_begin(policy, &freqs);
2310 cpufreq_freq_transition_end(policy, &freqs, 0);
2311 }
2312 }
2313
2314 return retval;
2315}
2316
2317int __cpufreq_driver_target(struct cpufreq_policy *policy,
2318 unsigned int target_freq,
2319 unsigned int relation)
2320{
2321 unsigned int old_target_freq = target_freq;
2322
2323 if (cpufreq_disabled())
2324 return -ENODEV;
2325
2326 target_freq = __resolve_freq(policy, target_freq, relation);
2327
2328 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2329 policy->cpu, target_freq, relation, old_target_freq);
2330
2331 /*
2332 * This might look like a redundant call as we are checking it again
2333 * after finding index. But it is left intentionally for cases where
2334 * exactly same freq is called again and so we can save on few function
2335 * calls.
2336 */
2337 if (target_freq == policy->cur &&
2338 !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2339 return 0;
2340
2341 if (cpufreq_driver->target) {
2342 /*
2343 * If the driver hasn't setup a single inefficient frequency,
2344 * it's unlikely it knows how to decode CPUFREQ_RELATION_E.
2345 */
2346 if (!policy->efficiencies_available)
2347 relation &= ~CPUFREQ_RELATION_E;
2348
2349 return cpufreq_driver->target(policy, target_freq, relation);
2350 }
2351
2352 if (!cpufreq_driver->target_index)
2353 return -EINVAL;
2354
2355 return __target_index(policy, policy->cached_resolved_idx);
2356}
2357EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2358
2359int cpufreq_driver_target(struct cpufreq_policy *policy,
2360 unsigned int target_freq,
2361 unsigned int relation)
2362{
2363 int ret;
2364
2365 down_write(&policy->rwsem);
2366
2367 ret = __cpufreq_driver_target(policy, target_freq, relation);
2368
2369 up_write(&policy->rwsem);
2370
2371 return ret;
2372}
2373EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2374
2375__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2376{
2377 return NULL;
2378}
2379
2380static int cpufreq_init_governor(struct cpufreq_policy *policy)
2381{
2382 int ret;
2383
2384 /* Don't start any governor operations if we are entering suspend */
2385 if (cpufreq_suspended)
2386 return 0;
2387 /*
2388 * Governor might not be initiated here if ACPI _PPC changed
2389 * notification happened, so check it.
2390 */
2391 if (!policy->governor)
2392 return -EINVAL;
2393
2394 /* Platform doesn't want dynamic frequency switching ? */
2395 if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2396 cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2397 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2398
2399 if (gov) {
2400 pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2401 policy->governor->name, gov->name);
2402 policy->governor = gov;
2403 } else {
2404 return -EINVAL;
2405 }
2406 }
2407
2408 if (!try_module_get(policy->governor->owner))
2409 return -EINVAL;
2410
2411 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2412
2413 if (policy->governor->init) {
2414 ret = policy->governor->init(policy);
2415 if (ret) {
2416 module_put(policy->governor->owner);
2417 return ret;
2418 }
2419 }
2420
2421 policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2422
2423 return 0;
2424}
2425
2426static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2427{
2428 if (cpufreq_suspended || !policy->governor)
2429 return;
2430
2431 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2432
2433 if (policy->governor->exit)
2434 policy->governor->exit(policy);
2435
2436 module_put(policy->governor->owner);
2437}
2438
2439int cpufreq_start_governor(struct cpufreq_policy *policy)
2440{
2441 int ret;
2442
2443 if (cpufreq_suspended)
2444 return 0;
2445
2446 if (!policy->governor)
2447 return -EINVAL;
2448
2449 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2450
2451 if (cpufreq_driver->get)
2452 cpufreq_verify_current_freq(policy, false);
2453
2454 if (policy->governor->start) {
2455 ret = policy->governor->start(policy);
2456 if (ret)
2457 return ret;
2458 }
2459
2460 if (policy->governor->limits)
2461 policy->governor->limits(policy);
2462
2463 return 0;
2464}
2465
2466void cpufreq_stop_governor(struct cpufreq_policy *policy)
2467{
2468 if (cpufreq_suspended || !policy->governor)
2469 return;
2470
2471 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2472
2473 if (policy->governor->stop)
2474 policy->governor->stop(policy);
2475}
2476
2477static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2478{
2479 if (cpufreq_suspended || !policy->governor)
2480 return;
2481
2482 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2483
2484 if (policy->governor->limits)
2485 policy->governor->limits(policy);
2486}
2487
2488int cpufreq_register_governor(struct cpufreq_governor *governor)
2489{
2490 int err;
2491
2492 if (!governor)
2493 return -EINVAL;
2494
2495 if (cpufreq_disabled())
2496 return -ENODEV;
2497
2498 mutex_lock(&cpufreq_governor_mutex);
2499
2500 err = -EBUSY;
2501 if (!find_governor(governor->name)) {
2502 err = 0;
2503 list_add(&governor->governor_list, &cpufreq_governor_list);
2504 }
2505
2506 mutex_unlock(&cpufreq_governor_mutex);
2507 return err;
2508}
2509EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2510
2511void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2512{
2513 struct cpufreq_policy *policy;
2514 unsigned long flags;
2515
2516 if (!governor)
2517 return;
2518
2519 if (cpufreq_disabled())
2520 return;
2521
2522 /* clear last_governor for all inactive policies */
2523 read_lock_irqsave(&cpufreq_driver_lock, flags);
2524 for_each_inactive_policy(policy) {
2525 if (!strcmp(policy->last_governor, governor->name)) {
2526 policy->governor = NULL;
2527 strcpy(policy->last_governor, "\0");
2528 }
2529 }
2530 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2531
2532 mutex_lock(&cpufreq_governor_mutex);
2533 list_del(&governor->governor_list);
2534 mutex_unlock(&cpufreq_governor_mutex);
2535}
2536EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2537
2538
2539/*********************************************************************
2540 * POLICY INTERFACE *
2541 *********************************************************************/
2542
2543/**
2544 * cpufreq_get_policy - get the current cpufreq_policy
2545 * @policy: struct cpufreq_policy into which the current cpufreq_policy
2546 * is written
2547 * @cpu: CPU to find the policy for
2548 *
2549 * Reads the current cpufreq policy.
2550 */
2551int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2552{
2553 struct cpufreq_policy *cpu_policy;
2554 if (!policy)
2555 return -EINVAL;
2556
2557 cpu_policy = cpufreq_cpu_get(cpu);
2558 if (!cpu_policy)
2559 return -EINVAL;
2560
2561 memcpy(policy, cpu_policy, sizeof(*policy));
2562
2563 cpufreq_cpu_put(cpu_policy);
2564 return 0;
2565}
2566EXPORT_SYMBOL(cpufreq_get_policy);
2567
2568DEFINE_PER_CPU(unsigned long, cpufreq_pressure);
2569
2570/**
2571 * cpufreq_update_pressure() - Update cpufreq pressure for CPUs
2572 * @policy: cpufreq policy of the CPUs.
2573 *
2574 * Update the value of cpufreq pressure for all @cpus in the policy.
2575 */
2576static void cpufreq_update_pressure(struct cpufreq_policy *policy)
2577{
2578 unsigned long max_capacity, capped_freq, pressure;
2579 u32 max_freq;
2580 int cpu;
2581
2582 cpu = cpumask_first(policy->related_cpus);
2583 max_freq = arch_scale_freq_ref(cpu);
2584 capped_freq = policy->max;
2585
2586 /*
2587 * Handle properly the boost frequencies, which should simply clean
2588 * the cpufreq pressure value.
2589 */
2590 if (max_freq <= capped_freq) {
2591 pressure = 0;
2592 } else {
2593 max_capacity = arch_scale_cpu_capacity(cpu);
2594 pressure = max_capacity -
2595 mult_frac(max_capacity, capped_freq, max_freq);
2596 }
2597
2598 for_each_cpu(cpu, policy->related_cpus)
2599 WRITE_ONCE(per_cpu(cpufreq_pressure, cpu), pressure);
2600}
2601
2602/**
2603 * cpufreq_set_policy - Modify cpufreq policy parameters.
2604 * @policy: Policy object to modify.
2605 * @new_gov: Policy governor pointer.
2606 * @new_pol: Policy value (for drivers with built-in governors).
2607 *
2608 * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2609 * limits to be set for the policy, update @policy with the verified limits
2610 * values and either invoke the driver's ->setpolicy() callback (if present) or
2611 * carry out a governor update for @policy. That is, run the current governor's
2612 * ->limits() callback (if @new_gov points to the same object as the one in
2613 * @policy) or replace the governor for @policy with @new_gov.
2614 *
2615 * The cpuinfo part of @policy is not updated by this function.
2616 */
2617static int cpufreq_set_policy(struct cpufreq_policy *policy,
2618 struct cpufreq_governor *new_gov,
2619 unsigned int new_pol)
2620{
2621 struct cpufreq_policy_data new_data;
2622 struct cpufreq_governor *old_gov;
2623 int ret;
2624
2625 memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2626 new_data.freq_table = policy->freq_table;
2627 new_data.cpu = policy->cpu;
2628 /*
2629 * PM QoS framework collects all the requests from users and provide us
2630 * the final aggregated value here.
2631 */
2632 new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2633 new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2634
2635 pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2636 new_data.cpu, new_data.min, new_data.max);
2637
2638 /*
2639 * Verify that the CPU speed can be set within these limits and make sure
2640 * that min <= max.
2641 */
2642 ret = cpufreq_driver->verify(&new_data);
2643 if (ret)
2644 return ret;
2645
2646 /*
2647 * Resolve policy min/max to available frequencies. It ensures
2648 * no frequency resolution will neither overshoot the requested maximum
2649 * nor undershoot the requested minimum.
2650 */
2651 policy->min = new_data.min;
2652 policy->max = new_data.max;
2653 policy->min = __resolve_freq(policy, policy->min, CPUFREQ_RELATION_L);
2654 policy->max = __resolve_freq(policy, policy->max, CPUFREQ_RELATION_H);
2655 trace_cpu_frequency_limits(policy);
2656
2657 cpufreq_update_pressure(policy);
2658
2659 policy->cached_target_freq = UINT_MAX;
2660
2661 pr_debug("new min and max freqs are %u - %u kHz\n",
2662 policy->min, policy->max);
2663
2664 if (cpufreq_driver->setpolicy) {
2665 policy->policy = new_pol;
2666 pr_debug("setting range\n");
2667 return cpufreq_driver->setpolicy(policy);
2668 }
2669
2670 if (new_gov == policy->governor) {
2671 pr_debug("governor limits update\n");
2672 cpufreq_governor_limits(policy);
2673 return 0;
2674 }
2675
2676 pr_debug("governor switch\n");
2677
2678 /* save old, working values */
2679 old_gov = policy->governor;
2680 /* end old governor */
2681 if (old_gov) {
2682 cpufreq_stop_governor(policy);
2683 cpufreq_exit_governor(policy);
2684 }
2685
2686 /* start new governor */
2687 policy->governor = new_gov;
2688 ret = cpufreq_init_governor(policy);
2689 if (!ret) {
2690 ret = cpufreq_start_governor(policy);
2691 if (!ret) {
2692 pr_debug("governor change\n");
2693 return 0;
2694 }
2695 cpufreq_exit_governor(policy);
2696 }
2697
2698 /* new governor failed, so re-start old one */
2699 pr_debug("starting governor %s failed\n", policy->governor->name);
2700 if (old_gov) {
2701 policy->governor = old_gov;
2702 if (cpufreq_init_governor(policy))
2703 policy->governor = NULL;
2704 else
2705 cpufreq_start_governor(policy);
2706 }
2707
2708 return ret;
2709}
2710
2711/**
2712 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2713 * @cpu: CPU to re-evaluate the policy for.
2714 *
2715 * Update the current frequency for the cpufreq policy of @cpu and use
2716 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2717 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2718 * for the policy in question, among other things.
2719 */
2720void cpufreq_update_policy(unsigned int cpu)
2721{
2722 struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2723
2724 if (!policy)
2725 return;
2726
2727 /*
2728 * BIOS might change freq behind our back
2729 * -> ask driver for current freq and notify governors about a change
2730 */
2731 if (cpufreq_driver->get && has_target() &&
2732 (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2733 goto unlock;
2734
2735 refresh_frequency_limits(policy);
2736
2737unlock:
2738 cpufreq_cpu_release(policy);
2739}
2740EXPORT_SYMBOL(cpufreq_update_policy);
2741
2742/**
2743 * cpufreq_update_limits - Update policy limits for a given CPU.
2744 * @cpu: CPU to update the policy limits for.
2745 *
2746 * Invoke the driver's ->update_limits callback if present or call
2747 * cpufreq_update_policy() for @cpu.
2748 */
2749void cpufreq_update_limits(unsigned int cpu)
2750{
2751 if (cpufreq_driver->update_limits)
2752 cpufreq_driver->update_limits(cpu);
2753 else
2754 cpufreq_update_policy(cpu);
2755}
2756EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2757
2758/*********************************************************************
2759 * BOOST *
2760 *********************************************************************/
2761static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2762{
2763 int ret;
2764
2765 if (!policy->freq_table)
2766 return -ENXIO;
2767
2768 ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2769 if (ret) {
2770 pr_err("%s: Policy frequency update failed\n", __func__);
2771 return ret;
2772 }
2773
2774 ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2775 if (ret < 0)
2776 return ret;
2777
2778 return 0;
2779}
2780
2781int cpufreq_boost_trigger_state(int state)
2782{
2783 struct cpufreq_policy *policy;
2784 unsigned long flags;
2785 int ret = 0;
2786
2787 if (cpufreq_driver->boost_enabled == state)
2788 return 0;
2789
2790 write_lock_irqsave(&cpufreq_driver_lock, flags);
2791 cpufreq_driver->boost_enabled = state;
2792 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2793
2794 cpus_read_lock();
2795 for_each_active_policy(policy) {
2796 policy->boost_enabled = state;
2797 ret = cpufreq_driver->set_boost(policy, state);
2798 if (ret) {
2799 policy->boost_enabled = !policy->boost_enabled;
2800 goto err_reset_state;
2801 }
2802 }
2803 cpus_read_unlock();
2804
2805 return 0;
2806
2807err_reset_state:
2808 cpus_read_unlock();
2809
2810 write_lock_irqsave(&cpufreq_driver_lock, flags);
2811 cpufreq_driver->boost_enabled = !state;
2812 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2813
2814 pr_err("%s: Cannot %s BOOST\n",
2815 __func__, state ? "enable" : "disable");
2816
2817 return ret;
2818}
2819
2820static bool cpufreq_boost_supported(void)
2821{
2822 return cpufreq_driver->set_boost;
2823}
2824
2825static int create_boost_sysfs_file(void)
2826{
2827 int ret;
2828
2829 ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2830 if (ret)
2831 pr_err("%s: cannot register global BOOST sysfs file\n",
2832 __func__);
2833
2834 return ret;
2835}
2836
2837static void remove_boost_sysfs_file(void)
2838{
2839 if (cpufreq_boost_supported())
2840 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2841}
2842
2843int cpufreq_enable_boost_support(void)
2844{
2845 if (!cpufreq_driver)
2846 return -EINVAL;
2847
2848 if (cpufreq_boost_supported())
2849 return 0;
2850
2851 cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2852
2853 /* This will get removed on driver unregister */
2854 return create_boost_sysfs_file();
2855}
2856EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2857
2858bool cpufreq_boost_enabled(void)
2859{
2860 return cpufreq_driver->boost_enabled;
2861}
2862EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2863
2864/*********************************************************************
2865 * REGISTER / UNREGISTER CPUFREQ DRIVER *
2866 *********************************************************************/
2867static enum cpuhp_state hp_online;
2868
2869static int cpuhp_cpufreq_online(unsigned int cpu)
2870{
2871 cpufreq_online(cpu);
2872
2873 return 0;
2874}
2875
2876static int cpuhp_cpufreq_offline(unsigned int cpu)
2877{
2878 cpufreq_offline(cpu);
2879
2880 return 0;
2881}
2882
2883/**
2884 * cpufreq_register_driver - register a CPU Frequency driver
2885 * @driver_data: A struct cpufreq_driver containing the values#
2886 * submitted by the CPU Frequency driver.
2887 *
2888 * Registers a CPU Frequency driver to this core code. This code
2889 * returns zero on success, -EEXIST when another driver got here first
2890 * (and isn't unregistered in the meantime).
2891 *
2892 */
2893int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2894{
2895 unsigned long flags;
2896 int ret;
2897
2898 if (cpufreq_disabled())
2899 return -ENODEV;
2900
2901 /*
2902 * The cpufreq core depends heavily on the availability of device
2903 * structure, make sure they are available before proceeding further.
2904 */
2905 if (!get_cpu_device(0))
2906 return -EPROBE_DEFER;
2907
2908 if (!driver_data || !driver_data->verify || !driver_data->init ||
2909 !(driver_data->setpolicy || driver_data->target_index ||
2910 driver_data->target) ||
2911 (driver_data->setpolicy && (driver_data->target_index ||
2912 driver_data->target)) ||
2913 (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2914 (!driver_data->online != !driver_data->offline) ||
2915 (driver_data->adjust_perf && !driver_data->fast_switch))
2916 return -EINVAL;
2917
2918 pr_debug("trying to register driver %s\n", driver_data->name);
2919
2920 /* Protect against concurrent CPU online/offline. */
2921 cpus_read_lock();
2922
2923 write_lock_irqsave(&cpufreq_driver_lock, flags);
2924 if (cpufreq_driver) {
2925 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2926 ret = -EEXIST;
2927 goto out;
2928 }
2929 cpufreq_driver = driver_data;
2930 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2931
2932 /*
2933 * Mark support for the scheduler's frequency invariance engine for
2934 * drivers that implement target(), target_index() or fast_switch().
2935 */
2936 if (!cpufreq_driver->setpolicy) {
2937 static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2938 pr_debug("supports frequency invariance");
2939 }
2940
2941 if (driver_data->setpolicy)
2942 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2943
2944 if (cpufreq_boost_supported()) {
2945 ret = create_boost_sysfs_file();
2946 if (ret)
2947 goto err_null_driver;
2948 }
2949
2950 ret = subsys_interface_register(&cpufreq_interface);
2951 if (ret)
2952 goto err_boost_unreg;
2953
2954 if (unlikely(list_empty(&cpufreq_policy_list))) {
2955 /* if all ->init() calls failed, unregister */
2956 ret = -ENODEV;
2957 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2958 driver_data->name);
2959 goto err_if_unreg;
2960 }
2961
2962 ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2963 "cpufreq:online",
2964 cpuhp_cpufreq_online,
2965 cpuhp_cpufreq_offline);
2966 if (ret < 0)
2967 goto err_if_unreg;
2968 hp_online = ret;
2969 ret = 0;
2970
2971 pr_debug("driver %s up and running\n", driver_data->name);
2972 goto out;
2973
2974err_if_unreg:
2975 subsys_interface_unregister(&cpufreq_interface);
2976err_boost_unreg:
2977 remove_boost_sysfs_file();
2978err_null_driver:
2979 write_lock_irqsave(&cpufreq_driver_lock, flags);
2980 cpufreq_driver = NULL;
2981 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2982out:
2983 cpus_read_unlock();
2984 return ret;
2985}
2986EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2987
2988/*
2989 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2990 *
2991 * Unregister the current CPUFreq driver. Only call this if you have
2992 * the right to do so, i.e. if you have succeeded in initialising before!
2993 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2994 * currently not initialised.
2995 */
2996void cpufreq_unregister_driver(struct cpufreq_driver *driver)
2997{
2998 unsigned long flags;
2999
3000 if (WARN_ON(!cpufreq_driver || (driver != cpufreq_driver)))
3001 return;
3002
3003 pr_debug("unregistering driver %s\n", driver->name);
3004
3005 /* Protect against concurrent cpu hotplug */
3006 cpus_read_lock();
3007 subsys_interface_unregister(&cpufreq_interface);
3008 remove_boost_sysfs_file();
3009 static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
3010 cpuhp_remove_state_nocalls_cpuslocked(hp_online);
3011
3012 write_lock_irqsave(&cpufreq_driver_lock, flags);
3013
3014 cpufreq_driver = NULL;
3015
3016 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
3017 cpus_read_unlock();
3018}
3019EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
3020
3021static int __init cpufreq_core_init(void)
3022{
3023 struct cpufreq_governor *gov = cpufreq_default_governor();
3024 struct device *dev_root;
3025
3026 if (cpufreq_disabled())
3027 return -ENODEV;
3028
3029 dev_root = bus_get_dev_root(&cpu_subsys);
3030 if (dev_root) {
3031 cpufreq_global_kobject = kobject_create_and_add("cpufreq", &dev_root->kobj);
3032 put_device(dev_root);
3033 }
3034 BUG_ON(!cpufreq_global_kobject);
3035
3036 if (!strlen(default_governor))
3037 strscpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
3038
3039 return 0;
3040}
3041module_param(off, int, 0444);
3042module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
3043core_initcall(cpufreq_core_init);