projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'for-6.16-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
[linux-block.git] / drivers / cpufreq / cpufreq.c
... / ...
CommitLineData
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/string_choices.h>
29#include <linux/suspend.h>
30#include <linux/syscore_ops.h>
31#include <linux/tick.h>
32#include <linux/units.h>
33#include <trace/events/power.h>
34
35static LIST_HEAD(cpufreq_policy_list);
36
37/* Macros to iterate over CPU policies */
38#define for_each_suitable_policy(__policy, __active) \
39 list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
40 if ((__active) == !policy_is_inactive(__policy))
41
42#define for_each_active_policy(__policy) \
43 for_each_suitable_policy(__policy, true)
44#define for_each_inactive_policy(__policy) \
45 for_each_suitable_policy(__policy, false)
46
47/* Iterate over governors */
48static LIST_HEAD(cpufreq_governor_list);
49#define for_each_governor(__governor) \
50 list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
51
52static char default_governor[CPUFREQ_NAME_LEN];
53
54/*
55 * The "cpufreq driver" - the arch- or hardware-dependent low
56 * level driver of CPUFreq support, and its spinlock. This lock
57 * also protects the cpufreq_cpu_data array.
58 */
59static struct cpufreq_driver *cpufreq_driver;
60static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
61static DEFINE_RWLOCK(cpufreq_driver_lock);
62
63static DEFINE_STATIC_KEY_FALSE(cpufreq_freq_invariance);
64bool cpufreq_supports_freq_invariance(void)
65{
66 return static_branch_likely(&cpufreq_freq_invariance);
67}
68
69/* Flag to suspend/resume CPUFreq governors */
70static bool cpufreq_suspended;
71
72static inline bool has_target(void)
73{
74 return cpufreq_driver->target_index || cpufreq_driver->target;
75}
76
77bool has_target_index(void)
78{
79 return !!cpufreq_driver->target_index;
80}
81
82/* internal prototypes */
83static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
84static int cpufreq_init_governor(struct cpufreq_policy *policy);
85static void cpufreq_exit_governor(struct cpufreq_policy *policy);
86static void cpufreq_governor_limits(struct cpufreq_policy *policy);
87static int cpufreq_set_policy(struct cpufreq_policy *policy,
88 struct cpufreq_governor *new_gov,
89 unsigned int new_pol);
90static bool cpufreq_boost_supported(void);
91static int cpufreq_boost_trigger_state(int state);
92
93/*
94 * Two notifier lists: the "policy" list is involved in the
95 * validation process for a new CPU frequency policy; the
96 * "transition" list for kernel code that needs to handle
97 * changes to devices when the CPU clock speed changes.
98 * The mutex locks both lists.
99 */
100static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
101SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
102
103static int off __read_mostly;
104static int cpufreq_disabled(void)
105{
106 return off;
107}
108void disable_cpufreq(void)
109{
110 off = 1;
111}
112static DEFINE_MUTEX(cpufreq_governor_mutex);
113
114bool have_governor_per_policy(void)
115{
116 return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
117}
118EXPORT_SYMBOL_GPL(have_governor_per_policy);
119
120static struct kobject *cpufreq_global_kobject;
121
122struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
123{
124 if (have_governor_per_policy())
125 return &policy->kobj;
126 else
127 return cpufreq_global_kobject;
128}
129EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
130
131static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
132{
133 struct kernel_cpustat kcpustat;
134 u64 cur_wall_time;
135 u64 idle_time;
136 u64 busy_time;
137
138 cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
139
140 kcpustat_cpu_fetch(&kcpustat, cpu);
141
142 busy_time = kcpustat.cpustat[CPUTIME_USER];
143 busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
144 busy_time += kcpustat.cpustat[CPUTIME_IRQ];
145 busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
146 busy_time += kcpustat.cpustat[CPUTIME_STEAL];
147 busy_time += kcpustat.cpustat[CPUTIME_NICE];
148
149 idle_time = cur_wall_time - busy_time;
150 if (wall)
151 *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
152
153 return div_u64(idle_time, NSEC_PER_USEC);
154}
155
156u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
157{
158 u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
159
160 if (idle_time == -1ULL)
161 return get_cpu_idle_time_jiffy(cpu, wall);
162 else if (!io_busy)
163 idle_time += get_cpu_iowait_time_us(cpu, wall);
164
165 return idle_time;
166}
167EXPORT_SYMBOL_GPL(get_cpu_idle_time);
168
169/*
170 * This is a generic cpufreq init() routine which can be used by cpufreq
171 * drivers of SMP systems. It will do following:
172 * - validate & show freq table passed
173 * - set policies transition latency
174 * - policy->cpus with all possible CPUs
175 */
176void cpufreq_generic_init(struct cpufreq_policy *policy,
177 struct cpufreq_frequency_table *table,
178 unsigned int transition_latency)
179{
180 policy->freq_table = table;
181 policy->cpuinfo.transition_latency = transition_latency;
182
183 /*
184 * The driver only supports the SMP configuration where all processors
185 * share the clock and voltage and clock.
186 */
187 cpumask_setall(policy->cpus);
188}
189EXPORT_SYMBOL_GPL(cpufreq_generic_init);
190
191struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
192{
193 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
194
195 return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
196}
197EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
198
199unsigned int cpufreq_generic_get(unsigned int cpu)
200{
201 struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
202
203 if (!policy || IS_ERR(policy->clk)) {
204 pr_err("%s: No %s associated to cpu: %d\n",
205 __func__, policy ? "clk" : "policy", cpu);
206 return 0;
207 }
208
209 return clk_get_rate(policy->clk) / 1000;
210}
211EXPORT_SYMBOL_GPL(cpufreq_generic_get);
212
213/**
214 * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
215 * @cpu: CPU to find the policy for.
216 *
217 * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
218 * the kobject reference counter of that policy. Return a valid policy on
219 * success or NULL on failure.
220 *
221 * The policy returned by this function has to be released with the help of
222 * cpufreq_cpu_put() to balance its kobject reference counter properly.
223 */
224struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
225{
226 struct cpufreq_policy *policy = NULL;
227 unsigned long flags;
228
229 if (WARN_ON(cpu >= nr_cpu_ids))
230 return NULL;
231
232 /* get the cpufreq driver */
233 read_lock_irqsave(&cpufreq_driver_lock, flags);
234
235 if (cpufreq_driver) {
236 /* get the CPU */
237 policy = cpufreq_cpu_get_raw(cpu);
238 if (policy)
239 kobject_get(&policy->kobj);
240 }
241
242 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
243
244 return policy;
245}
246EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
247
248/**
249 * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
250 * @policy: cpufreq policy returned by cpufreq_cpu_get().
251 */
252void cpufreq_cpu_put(struct cpufreq_policy *policy)
253{
254 kobject_put(&policy->kobj);
255}
256EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
257
258/*********************************************************************
259 * EXTERNALLY AFFECTING FREQUENCY CHANGES *
260 *********************************************************************/
261
262/**
263 * adjust_jiffies - Adjust the system "loops_per_jiffy".
264 * @val: CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
265 * @ci: Frequency change information.
266 *
267 * This function alters the system "loops_per_jiffy" for the clock
268 * speed change. Note that loops_per_jiffy cannot be updated on SMP
269 * systems as each CPU might be scaled differently. So, use the arch
270 * per-CPU loops_per_jiffy value wherever possible.
271 */
272static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
273{
274#ifndef CONFIG_SMP
275 static unsigned long l_p_j_ref;
276 static unsigned int l_p_j_ref_freq;
277
278 if (ci->flags & CPUFREQ_CONST_LOOPS)
279 return;
280
281 if (!l_p_j_ref_freq) {
282 l_p_j_ref = loops_per_jiffy;
283 l_p_j_ref_freq = ci->old;
284 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
285 l_p_j_ref, l_p_j_ref_freq);
286 }
287 if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
288 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
289 ci->new);
290 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
291 loops_per_jiffy, ci->new);
292 }
293#endif
294}
295
296/**
297 * cpufreq_notify_transition - Notify frequency transition and adjust jiffies.
298 * @policy: cpufreq policy to enable fast frequency switching for.
299 * @freqs: contain details of the frequency update.
300 * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
301 *
302 * This function calls the transition notifiers and adjust_jiffies().
303 *
304 * It is called twice on all CPU frequency changes that have external effects.
305 */
306static void cpufreq_notify_transition(struct cpufreq_policy *policy,
307 struct cpufreq_freqs *freqs,
308 unsigned int state)
309{
310 int cpu;
311
312 BUG_ON(irqs_disabled());
313
314 if (cpufreq_disabled())
315 return;
316
317 freqs->policy = policy;
318 freqs->flags = cpufreq_driver->flags;
319 pr_debug("notification %u of frequency transition to %u kHz\n",
320 state, freqs->new);
321
322 switch (state) {
323 case CPUFREQ_PRECHANGE:
324 /*
325 * Detect if the driver reported a value as "old frequency"
326 * which is not equal to what the cpufreq core thinks is
327 * "old frequency".
328 */
329 if (policy->cur && policy->cur != freqs->old) {
330 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
331 freqs->old, policy->cur);
332 freqs->old = policy->cur;
333 }
334
335 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
336 CPUFREQ_PRECHANGE, freqs);
337
338 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
339 break;
340
341 case CPUFREQ_POSTCHANGE:
342 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
343 pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
344 cpumask_pr_args(policy->cpus));
345
346 for_each_cpu(cpu, policy->cpus)
347 trace_cpu_frequency(freqs->new, cpu);
348
349 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
350 CPUFREQ_POSTCHANGE, freqs);
351
352 cpufreq_stats_record_transition(policy, freqs->new);
353 policy->cur = freqs->new;
354 }
355}
356
357/* Do post notifications when there are chances that transition has failed */
358static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
359 struct cpufreq_freqs *freqs, int transition_failed)
360{
361 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
362 if (!transition_failed)
363 return;
364
365 swap(freqs->old, freqs->new);
366 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
367 cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
368}
369
370void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
371 struct cpufreq_freqs *freqs)
372{
373
374 /*
375 * Catch double invocations of _begin() which lead to self-deadlock.
376 * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
377 * doesn't invoke _begin() on their behalf, and hence the chances of
378 * double invocations are very low. Moreover, there are scenarios
379 * where these checks can emit false-positive warnings in these
380 * drivers; so we avoid that by skipping them altogether.
381 */
382 WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
383 && current == policy->transition_task);
384
385wait:
386 wait_event(policy->transition_wait, !policy->transition_ongoing);
387
388 spin_lock(&policy->transition_lock);
389
390 if (unlikely(policy->transition_ongoing)) {
391 spin_unlock(&policy->transition_lock);
392 goto wait;
393 }
394
395 policy->transition_ongoing = true;
396 policy->transition_task = current;
397
398 spin_unlock(&policy->transition_lock);
399
400 cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
401}
402EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
403
404void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
405 struct cpufreq_freqs *freqs, int transition_failed)
406{
407 if (WARN_ON(!policy->transition_ongoing))
408 return;
409
410 cpufreq_notify_post_transition(policy, freqs, transition_failed);
411
412 arch_set_freq_scale(policy->related_cpus,
413 policy->cur,
414 arch_scale_freq_ref(policy->cpu));
415
416 spin_lock(&policy->transition_lock);
417 policy->transition_ongoing = false;
418 policy->transition_task = NULL;
419 spin_unlock(&policy->transition_lock);
420
421 wake_up(&policy->transition_wait);
422}
423EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
424
425/*
426 * Fast frequency switching status count. Positive means "enabled", negative
427 * means "disabled" and 0 means "not decided yet".
428 */
429static int cpufreq_fast_switch_count;
430static DEFINE_MUTEX(cpufreq_fast_switch_lock);
431
432static void cpufreq_list_transition_notifiers(void)
433{
434 struct notifier_block *nb;
435
436 pr_info("Registered transition notifiers:\n");
437
438 mutex_lock(&cpufreq_transition_notifier_list.mutex);
439
440 for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
441 pr_info("%pS\n", nb->notifier_call);
442
443 mutex_unlock(&cpufreq_transition_notifier_list.mutex);
444}
445
446/**
447 * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
448 * @policy: cpufreq policy to enable fast frequency switching for.
449 *
450 * Try to enable fast frequency switching for @policy.
451 *
452 * The attempt will fail if there is at least one transition notifier registered
453 * at this point, as fast frequency switching is quite fundamentally at odds
454 * with transition notifiers. Thus if successful, it will make registration of
455 * transition notifiers fail going forward.
456 */
457void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
458{
459 lockdep_assert_held(&policy->rwsem);
460
461 if (!policy->fast_switch_possible)
462 return;
463
464 mutex_lock(&cpufreq_fast_switch_lock);
465 if (cpufreq_fast_switch_count >= 0) {
466 cpufreq_fast_switch_count++;
467 policy->fast_switch_enabled = true;
468 } else {
469 pr_warn("CPU%u: Fast frequency switching not enabled\n",
470 policy->cpu);
471 cpufreq_list_transition_notifiers();
472 }
473 mutex_unlock(&cpufreq_fast_switch_lock);
474}
475EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
476
477/**
478 * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
479 * @policy: cpufreq policy to disable fast frequency switching for.
480 */
481void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
482{
483 mutex_lock(&cpufreq_fast_switch_lock);
484 if (policy->fast_switch_enabled) {
485 policy->fast_switch_enabled = false;
486 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
487 cpufreq_fast_switch_count--;
488 }
489 mutex_unlock(&cpufreq_fast_switch_lock);
490}
491EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
492
493static unsigned int __resolve_freq(struct cpufreq_policy *policy,
494 unsigned int target_freq,
495 unsigned int min, unsigned int max,
496 unsigned int relation)
497{
498 unsigned int idx;
499
500 target_freq = clamp_val(target_freq, min, max);
501
502 if (!policy->freq_table)
503 return target_freq;
504
505 idx = cpufreq_frequency_table_target(policy, target_freq, min, max, relation);
506 policy->cached_resolved_idx = idx;
507 policy->cached_target_freq = target_freq;
508 return policy->freq_table[idx].frequency;
509}
510
511/**
512 * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
513 * one.
514 * @policy: associated policy to interrogate
515 * @target_freq: target frequency to resolve.
516 *
517 * The target to driver frequency mapping is cached in the policy.
518 *
519 * Return: Lowest driver-supported frequency greater than or equal to the
520 * given target_freq, subject to policy (min/max) and driver limitations.
521 */
522unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
523 unsigned int target_freq)
524{
525 unsigned int min = READ_ONCE(policy->min);
526 unsigned int max = READ_ONCE(policy->max);
527
528 /*
529 * If this function runs in parallel with cpufreq_set_policy(), it may
530 * read policy->min before the update and policy->max after the update
531 * or the other way around, so there is no ordering guarantee.
532 *
533 * Resolve this by always honoring the max (in case it comes from
534 * thermal throttling or similar).
535 */
536 if (unlikely(min > max))
537 min = max;
538
539 return __resolve_freq(policy, target_freq, min, max, CPUFREQ_RELATION_LE);
540}
541EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
542
543unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
544{
545 unsigned int latency;
546
547 if (policy->transition_delay_us)
548 return policy->transition_delay_us;
549
550 latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
551 if (latency)
552 /* Give a 50% breathing room between updates */
553 return latency + (latency >> 1);
554
555 return USEC_PER_MSEC;
556}
557EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
558
559/*********************************************************************
560 * SYSFS INTERFACE *
561 *********************************************************************/
562static ssize_t show_boost(struct kobject *kobj,
563 struct kobj_attribute *attr, char *buf)
564{
565 return sysfs_emit(buf, "%d\n", cpufreq_driver->boost_enabled);
566}
567
568static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
569 const char *buf, size_t count)
570{
571 bool enable;
572
573 if (kstrtobool(buf, &enable))
574 return -EINVAL;
575
576 if (cpufreq_boost_trigger_state(enable)) {
577 pr_err("%s: Cannot %s BOOST!\n",
578 __func__, str_enable_disable(enable));
579 return -EINVAL;
580 }
581
582 pr_debug("%s: cpufreq BOOST %s\n",
583 __func__, str_enabled_disabled(enable));
584
585 return count;
586}
587define_one_global_rw(boost);
588
589static ssize_t show_local_boost(struct cpufreq_policy *policy, char *buf)
590{
591 return sysfs_emit(buf, "%d\n", policy->boost_enabled);
592}
593
594static int policy_set_boost(struct cpufreq_policy *policy, bool enable)
595{
596 int ret;
597
598 if (policy->boost_enabled == enable)
599 return 0;
600
601 policy->boost_enabled = enable;
602
603 ret = cpufreq_driver->set_boost(policy, enable);
604 if (ret)
605 policy->boost_enabled = !policy->boost_enabled;
606
607 return ret;
608}
609
610static ssize_t store_local_boost(struct cpufreq_policy *policy,
611 const char *buf, size_t count)
612{
613 int ret;
614 bool enable;
615
616 if (kstrtobool(buf, &enable))
617 return -EINVAL;
618
619 if (!cpufreq_driver->boost_enabled)
620 return -EINVAL;
621
622 if (!policy->boost_supported)
623 return -EINVAL;
624
625 ret = policy_set_boost(policy, enable);
626 if (!ret)
627 return count;
628
629 return ret;
630}
631
632static struct freq_attr local_boost = __ATTR(boost, 0644, show_local_boost, store_local_boost);
633
634static struct cpufreq_governor *find_governor(const char *str_governor)
635{
636 struct cpufreq_governor *t;
637
638 for_each_governor(t)
639 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
640 return t;
641
642 return NULL;
643}
644
645static struct cpufreq_governor *get_governor(const char *str_governor)
646{
647 struct cpufreq_governor *t;
648
649 mutex_lock(&cpufreq_governor_mutex);
650 t = find_governor(str_governor);
651 if (!t)
652 goto unlock;
653
654 if (!try_module_get(t->owner))
655 t = NULL;
656
657unlock:
658 mutex_unlock(&cpufreq_governor_mutex);
659
660 return t;
661}
662
663static unsigned int cpufreq_parse_policy(char *str_governor)
664{
665 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
666 return CPUFREQ_POLICY_PERFORMANCE;
667
668 if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
669 return CPUFREQ_POLICY_POWERSAVE;
670
671 return CPUFREQ_POLICY_UNKNOWN;
672}
673
674/**
675 * cpufreq_parse_governor - parse a governor string only for has_target()
676 * @str_governor: Governor name.
677 */
678static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
679{
680 struct cpufreq_governor *t;
681
682 t = get_governor(str_governor);
683 if (t)
684 return t;
685
686 if (request_module("cpufreq_%s", str_governor))
687 return NULL;
688
689 return get_governor(str_governor);
690}
691
692/*
693 * cpufreq_per_cpu_attr_read() / show_##file_name() -
694 * print out cpufreq information
695 *
696 * Write out information from cpufreq_driver->policy[cpu]; object must be
697 * "unsigned int".
698 */
699
700#define show_one(file_name, object) \
701static ssize_t show_##file_name \
702(struct cpufreq_policy *policy, char *buf) \
703{ \
704 return sysfs_emit(buf, "%u\n", policy->object); \
705}
706
707show_one(cpuinfo_min_freq, cpuinfo.min_freq);
708show_one(cpuinfo_max_freq, cpuinfo.max_freq);
709show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
710show_one(scaling_min_freq, min);
711show_one(scaling_max_freq, max);
712
713__weak int arch_freq_get_on_cpu(int cpu)
714{
715 return -EOPNOTSUPP;
716}
717
718static inline bool cpufreq_avg_freq_supported(struct cpufreq_policy *policy)
719{
720 return arch_freq_get_on_cpu(policy->cpu) != -EOPNOTSUPP;
721}
722
723static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
724{
725 ssize_t ret;
726 int freq;
727
728 freq = IS_ENABLED(CONFIG_CPUFREQ_ARCH_CUR_FREQ)
729 ? arch_freq_get_on_cpu(policy->cpu)
730 : 0;
731
732 if (freq > 0)
733 ret = sysfs_emit(buf, "%u\n", freq);
734 else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
735 ret = sysfs_emit(buf, "%u\n", cpufreq_driver->get(policy->cpu));
736 else
737 ret = sysfs_emit(buf, "%u\n", policy->cur);
738 return ret;
739}
740
741/*
742 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
743 */
744#define store_one(file_name, object) \
745static ssize_t store_##file_name \
746(struct cpufreq_policy *policy, const char *buf, size_t count) \
747{ \
748 unsigned long val; \
749 int ret; \
750 \
751 ret = kstrtoul(buf, 0, &val); \
752 if (ret) \
753 return ret; \
754 \
755 ret = freq_qos_update_request(policy->object##_freq_req, val);\
756 return ret >= 0 ? count : ret; \
757}
758
759store_one(scaling_min_freq, min);
760store_one(scaling_max_freq, max);
761
762/*
763 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
764 */
765static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
766 char *buf)
767{
768 unsigned int cur_freq = __cpufreq_get(policy);
769
770 if (cur_freq)
771 return sysfs_emit(buf, "%u\n", cur_freq);
772
773 return sysfs_emit(buf, "<unknown>\n");
774}
775
776/*
777 * show_cpuinfo_avg_freq - average CPU frequency as detected by hardware
778 */
779static ssize_t show_cpuinfo_avg_freq(struct cpufreq_policy *policy,
780 char *buf)
781{
782 int avg_freq = arch_freq_get_on_cpu(policy->cpu);
783
784 if (avg_freq > 0)
785 return sysfs_emit(buf, "%u\n", avg_freq);
786 return avg_freq != 0 ? avg_freq : -EINVAL;
787}
788
789/*
790 * show_scaling_governor - show the current policy for the specified CPU
791 */
792static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
793{
794 if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
795 return sysfs_emit(buf, "powersave\n");
796 else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
797 return sysfs_emit(buf, "performance\n");
798 else if (policy->governor)
799 return sysfs_emit(buf, "%s\n", policy->governor->name);
800 return -EINVAL;
801}
802
803/*
804 * store_scaling_governor - store policy for the specified CPU
805 */
806static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
807 const char *buf, size_t count)
808{
809 char str_governor[CPUFREQ_NAME_LEN];
810 int ret;
811
812 ret = sscanf(buf, "%15s", str_governor);
813 if (ret != 1)
814 return -EINVAL;
815
816 if (cpufreq_driver->setpolicy) {
817 unsigned int new_pol;
818
819 new_pol = cpufreq_parse_policy(str_governor);
820 if (!new_pol)
821 return -EINVAL;
822
823 ret = cpufreq_set_policy(policy, NULL, new_pol);
824 } else {
825 struct cpufreq_governor *new_gov;
826
827 new_gov = cpufreq_parse_governor(str_governor);
828 if (!new_gov)
829 return -EINVAL;
830
831 ret = cpufreq_set_policy(policy, new_gov,
832 CPUFREQ_POLICY_UNKNOWN);
833
834 module_put(new_gov->owner);
835 }
836
837 return ret ? ret : count;
838}
839
840/*
841 * show_scaling_driver - show the cpufreq driver currently loaded
842 */
843static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
844{
845 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
846}
847
848/*
849 * show_scaling_available_governors - show the available CPUfreq governors
850 */
851static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
852 char *buf)
853{
854 ssize_t i = 0;
855 struct cpufreq_governor *t;
856
857 if (!has_target()) {
858 i += sysfs_emit(buf, "performance powersave");
859 goto out;
860 }
861
862 mutex_lock(&cpufreq_governor_mutex);
863 for_each_governor(t) {
864 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
865 - (CPUFREQ_NAME_LEN + 2)))
866 break;
867 i += sysfs_emit_at(buf, i, "%s ", t->name);
868 }
869 mutex_unlock(&cpufreq_governor_mutex);
870out:
871 i += sysfs_emit_at(buf, i, "\n");
872 return i;
873}
874
875ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
876{
877 ssize_t i = 0;
878 unsigned int cpu;
879
880 for_each_cpu(cpu, mask) {
881 i += sysfs_emit_at(buf, i, "%u ", cpu);
882 if (i >= (PAGE_SIZE - 5))
883 break;
884 }
885
886 /* Remove the extra space at the end */
887 i--;
888
889 i += sysfs_emit_at(buf, i, "\n");
890 return i;
891}
892EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
893
894/*
895 * show_related_cpus - show the CPUs affected by each transition even if
896 * hw coordination is in use
897 */
898static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
899{
900 return cpufreq_show_cpus(policy->related_cpus, buf);
901}
902
903/*
904 * show_affected_cpus - show the CPUs affected by each transition
905 */
906static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
907{
908 return cpufreq_show_cpus(policy->cpus, buf);
909}
910
911static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
912 const char *buf, size_t count)
913{
914 unsigned int freq = 0;
915 unsigned int ret;
916
917 if (!policy->governor || !policy->governor->store_setspeed)
918 return -EINVAL;
919
920 ret = kstrtouint(buf, 0, &freq);
921 if (ret)
922 return ret;
923
924 policy->governor->store_setspeed(policy, freq);
925
926 return count;
927}
928
929static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
930{
931 if (!policy->governor || !policy->governor->show_setspeed)
932 return sysfs_emit(buf, "<unsupported>\n");
933
934 return policy->governor->show_setspeed(policy, buf);
935}
936
937/*
938 * show_bios_limit - show the current cpufreq HW/BIOS limitation
939 */
940static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
941{
942 unsigned int limit;
943 int ret;
944 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
945 if (!ret)
946 return sysfs_emit(buf, "%u\n", limit);
947 return sysfs_emit(buf, "%u\n", policy->cpuinfo.max_freq);
948}
949
950cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
951cpufreq_freq_attr_ro(cpuinfo_avg_freq);
952cpufreq_freq_attr_ro(cpuinfo_min_freq);
953cpufreq_freq_attr_ro(cpuinfo_max_freq);
954cpufreq_freq_attr_ro(cpuinfo_transition_latency);
955cpufreq_freq_attr_ro(scaling_available_governors);
956cpufreq_freq_attr_ro(scaling_driver);
957cpufreq_freq_attr_ro(scaling_cur_freq);
958cpufreq_freq_attr_ro(bios_limit);
959cpufreq_freq_attr_ro(related_cpus);
960cpufreq_freq_attr_ro(affected_cpus);
961cpufreq_freq_attr_rw(scaling_min_freq);
962cpufreq_freq_attr_rw(scaling_max_freq);
963cpufreq_freq_attr_rw(scaling_governor);
964cpufreq_freq_attr_rw(scaling_setspeed);
965
966static struct attribute *cpufreq_attrs[] = {
967 &cpuinfo_min_freq.attr,
968 &cpuinfo_max_freq.attr,
969 &cpuinfo_transition_latency.attr,
970 &scaling_min_freq.attr,
971 &scaling_max_freq.attr,
972 &affected_cpus.attr,
973 &related_cpus.attr,
974 &scaling_governor.attr,
975 &scaling_driver.attr,
976 &scaling_available_governors.attr,
977 &scaling_setspeed.attr,
978 NULL
979};
980ATTRIBUTE_GROUPS(cpufreq);
981
982#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
983#define to_attr(a) container_of(a, struct freq_attr, attr)
984
985static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
986{
987 struct cpufreq_policy *policy = to_policy(kobj);
988 struct freq_attr *fattr = to_attr(attr);
989
990 if (!fattr->show)
991 return -EIO;
992
993 guard(cpufreq_policy_read)(policy);
994
995 if (likely(!policy_is_inactive(policy)))
996 return fattr->show(policy, buf);
997
998 return -EBUSY;
999}
1000
1001static ssize_t store(struct kobject *kobj, struct attribute *attr,
1002 const char *buf, size_t count)
1003{
1004 struct cpufreq_policy *policy = to_policy(kobj);
1005 struct freq_attr *fattr = to_attr(attr);
1006
1007 if (!fattr->store)
1008 return -EIO;
1009
1010 guard(cpufreq_policy_write)(policy);
1011
1012 if (likely(!policy_is_inactive(policy)))
1013 return fattr->store(policy, buf, count);
1014
1015 return -EBUSY;
1016}
1017
1018static void cpufreq_sysfs_release(struct kobject *kobj)
1019{
1020 struct cpufreq_policy *policy = to_policy(kobj);
1021 pr_debug("last reference is dropped\n");
1022 complete(&policy->kobj_unregister);
1023}
1024
1025static const struct sysfs_ops sysfs_ops = {
1026 .show = show,
1027 .store = store,
1028};
1029
1030static const struct kobj_type ktype_cpufreq = {
1031 .sysfs_ops = &sysfs_ops,
1032 .default_groups = cpufreq_groups,
1033 .release = cpufreq_sysfs_release,
1034};
1035
1036static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
1037 struct device *dev)
1038{
1039 if (unlikely(!dev))
1040 return;
1041
1042 if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1043 return;
1044
1045 dev_dbg(dev, "%s: Adding symlink\n", __func__);
1046 if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1047 dev_err(dev, "cpufreq symlink creation failed\n");
1048}
1049
1050static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu,
1051 struct device *dev)
1052{
1053 dev_dbg(dev, "%s: Removing symlink\n", __func__);
1054 sysfs_remove_link(&dev->kobj, "cpufreq");
1055 cpumask_clear_cpu(cpu, policy->real_cpus);
1056}
1057
1058static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1059{
1060 struct freq_attr **drv_attr;
1061 int ret = 0;
1062
1063 /* Attributes that need freq_table */
1064 if (policy->freq_table) {
1065 ret = sysfs_create_file(&policy->kobj,
1066 &cpufreq_freq_attr_scaling_available_freqs.attr);
1067 if (ret)
1068 return ret;
1069
1070 if (cpufreq_boost_supported()) {
1071 ret = sysfs_create_file(&policy->kobj,
1072 &cpufreq_freq_attr_scaling_boost_freqs.attr);
1073 if (ret)
1074 return ret;
1075 }
1076 }
1077
1078 /* set up files for this cpu device */
1079 drv_attr = cpufreq_driver->attr;
1080 while (drv_attr && *drv_attr) {
1081 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1082 if (ret)
1083 return ret;
1084 drv_attr++;
1085 }
1086 if (cpufreq_driver->get) {
1087 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1088 if (ret)
1089 return ret;
1090 }
1091
1092 if (cpufreq_avg_freq_supported(policy)) {
1093 ret = sysfs_create_file(&policy->kobj, &cpuinfo_avg_freq.attr);
1094 if (ret)
1095 return ret;
1096 }
1097
1098 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1099 if (ret)
1100 return ret;
1101
1102 if (cpufreq_driver->bios_limit) {
1103 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1104 if (ret)
1105 return ret;
1106 }
1107
1108 if (cpufreq_boost_supported()) {
1109 ret = sysfs_create_file(&policy->kobj, &local_boost.attr);
1110 if (ret)
1111 return ret;
1112 }
1113
1114 return 0;
1115}
1116
1117static int cpufreq_init_policy(struct cpufreq_policy *policy)
1118{
1119 struct cpufreq_governor *gov = NULL;
1120 unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1121 int ret;
1122
1123 if (has_target()) {
1124 /* Update policy governor to the one used before hotplug. */
1125 gov = get_governor(policy->last_governor);
1126 if (gov) {
1127 pr_debug("Restoring governor %s for cpu %d\n",
1128 gov->name, policy->cpu);
1129 } else {
1130 gov = get_governor(default_governor);
1131 }
1132
1133 if (!gov) {
1134 gov = cpufreq_default_governor();
1135 __module_get(gov->owner);
1136 }
1137
1138 } else {
1139
1140 /* Use the default policy if there is no last_policy. */
1141 if (policy->last_policy) {
1142 pol = policy->last_policy;
1143 } else {
1144 pol = cpufreq_parse_policy(default_governor);
1145 /*
1146 * In case the default governor is neither "performance"
1147 * nor "powersave", fall back to the initial policy
1148 * value set by the driver.
1149 */
1150 if (pol == CPUFREQ_POLICY_UNKNOWN)
1151 pol = policy->policy;
1152 }
1153 if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1154 pol != CPUFREQ_POLICY_POWERSAVE)
1155 return -ENODATA;
1156 }
1157
1158 ret = cpufreq_set_policy(policy, gov, pol);
1159 if (gov)
1160 module_put(gov->owner);
1161
1162 return ret;
1163}
1164
1165static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1166{
1167 int ret = 0;
1168
1169 /* Has this CPU been taken care of already? */
1170 if (cpumask_test_cpu(cpu, policy->cpus))
1171 return 0;
1172
1173 guard(cpufreq_policy_write)(policy);
1174
1175 if (has_target())
1176 cpufreq_stop_governor(policy);
1177
1178 cpumask_set_cpu(cpu, policy->cpus);
1179
1180 if (has_target()) {
1181 ret = cpufreq_start_governor(policy);
1182 if (ret)
1183 pr_err("%s: Failed to start governor\n", __func__);
1184 }
1185
1186 return ret;
1187}
1188
1189void refresh_frequency_limits(struct cpufreq_policy *policy)
1190{
1191 if (!policy_is_inactive(policy)) {
1192 pr_debug("updating policy for CPU %u\n", policy->cpu);
1193
1194 cpufreq_set_policy(policy, policy->governor, policy->policy);
1195 }
1196}
1197EXPORT_SYMBOL(refresh_frequency_limits);
1198
1199static void handle_update(struct work_struct *work)
1200{
1201 struct cpufreq_policy *policy =
1202 container_of(work, struct cpufreq_policy, update);
1203
1204 pr_debug("handle_update for cpu %u called\n", policy->cpu);
1205
1206 guard(cpufreq_policy_write)(policy);
1207
1208 refresh_frequency_limits(policy);
1209}
1210
1211static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1212 void *data)
1213{
1214 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1215
1216 schedule_work(&policy->update);
1217 return 0;
1218}
1219
1220static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1221 void *data)
1222{
1223 struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1224
1225 schedule_work(&policy->update);
1226 return 0;
1227}
1228
1229static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1230{
1231 struct kobject *kobj;
1232 struct completion *cmp;
1233
1234 scoped_guard(cpufreq_policy_write, policy) {
1235 cpufreq_stats_free_table(policy);
1236 kobj = &policy->kobj;
1237 cmp = &policy->kobj_unregister;
1238 }
1239 kobject_put(kobj);
1240
1241 /*
1242 * We need to make sure that the underlying kobj is
1243 * actually not referenced anymore by anybody before we
1244 * proceed with unloading.
1245 */
1246 pr_debug("waiting for dropping of refcount\n");
1247 wait_for_completion(cmp);
1248 pr_debug("wait complete\n");
1249}
1250
1251static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1252{
1253 struct cpufreq_policy *policy;
1254 struct device *dev = get_cpu_device(cpu);
1255 int ret;
1256
1257 if (!dev)
1258 return NULL;
1259
1260 policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1261 if (!policy)
1262 return NULL;
1263
1264 if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1265 goto err_free_policy;
1266
1267 if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1268 goto err_free_cpumask;
1269
1270 if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1271 goto err_free_rcpumask;
1272
1273 init_completion(&policy->kobj_unregister);
1274 ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1275 cpufreq_global_kobject, "policy%u", cpu);
1276 if (ret) {
1277 dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1278 /*
1279 * The entire policy object will be freed below, but the extra
1280 * memory allocated for the kobject name needs to be freed by
1281 * releasing the kobject.
1282 */
1283 kobject_put(&policy->kobj);
1284 goto err_free_real_cpus;
1285 }
1286
1287 freq_constraints_init(&policy->constraints);
1288
1289 policy->nb_min.notifier_call = cpufreq_notifier_min;
1290 policy->nb_max.notifier_call = cpufreq_notifier_max;
1291
1292 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1293 &policy->nb_min);
1294 if (ret) {
1295 dev_err(dev, "Failed to register MIN QoS notifier: %d (CPU%u)\n",
1296 ret, cpu);
1297 goto err_kobj_remove;
1298 }
1299
1300 ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1301 &policy->nb_max);
1302 if (ret) {
1303 dev_err(dev, "Failed to register MAX QoS notifier: %d (CPU%u)\n",
1304 ret, cpu);
1305 goto err_min_qos_notifier;
1306 }
1307
1308 INIT_LIST_HEAD(&policy->policy_list);
1309 init_rwsem(&policy->rwsem);
1310 spin_lock_init(&policy->transition_lock);
1311 init_waitqueue_head(&policy->transition_wait);
1312 INIT_WORK(&policy->update, handle_update);
1313
1314 return policy;
1315
1316err_min_qos_notifier:
1317 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1318 &policy->nb_min);
1319err_kobj_remove:
1320 cpufreq_policy_put_kobj(policy);
1321err_free_real_cpus:
1322 free_cpumask_var(policy->real_cpus);
1323err_free_rcpumask:
1324 free_cpumask_var(policy->related_cpus);
1325err_free_cpumask:
1326 free_cpumask_var(policy->cpus);
1327err_free_policy:
1328 kfree(policy);
1329
1330 return NULL;
1331}
1332
1333static void cpufreq_policy_free(struct cpufreq_policy *policy)
1334{
1335 unsigned long flags;
1336 int cpu;
1337
1338 /*
1339 * The callers must ensure the policy is inactive by now, to avoid any
1340 * races with show()/store() callbacks.
1341 */
1342 if (unlikely(!policy_is_inactive(policy)))
1343 pr_warn("%s: Freeing active policy\n", __func__);
1344
1345 /* Remove policy from list */
1346 write_lock_irqsave(&cpufreq_driver_lock, flags);
1347 list_del(&policy->policy_list);
1348
1349 for_each_cpu(cpu, policy->related_cpus)
1350 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1351 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1352
1353 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1354 &policy->nb_max);
1355 freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1356 &policy->nb_min);
1357
1358 /* Cancel any pending policy->update work before freeing the policy. */
1359 cancel_work_sync(&policy->update);
1360
1361 if (policy->max_freq_req) {
1362 /*
1363 * Remove max_freq_req after sending CPUFREQ_REMOVE_POLICY
1364 * notification, since CPUFREQ_CREATE_POLICY notification was
1365 * sent after adding max_freq_req earlier.
1366 */
1367 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1368 CPUFREQ_REMOVE_POLICY, policy);
1369 freq_qos_remove_request(policy->max_freq_req);
1370 }
1371
1372 freq_qos_remove_request(policy->min_freq_req);
1373 kfree(policy->min_freq_req);
1374
1375 cpufreq_policy_put_kobj(policy);
1376 free_cpumask_var(policy->real_cpus);
1377 free_cpumask_var(policy->related_cpus);
1378 free_cpumask_var(policy->cpus);
1379 kfree(policy);
1380}
1381
1382static int cpufreq_policy_online(struct cpufreq_policy *policy,
1383 unsigned int cpu, bool new_policy)
1384{
1385 unsigned long flags;
1386 unsigned int j;
1387 int ret;
1388
1389 guard(cpufreq_policy_write)(policy);
1390
1391 policy->cpu = cpu;
1392 policy->governor = NULL;
1393
1394 if (!new_policy && cpufreq_driver->online) {
1395 /* Recover policy->cpus using related_cpus */
1396 cpumask_copy(policy->cpus, policy->related_cpus);
1397
1398 ret = cpufreq_driver->online(policy);
1399 if (ret) {
1400 pr_debug("%s: %d: initialization failed\n", __func__,
1401 __LINE__);
1402 goto out_exit_policy;
1403 }
1404 } else {
1405 cpumask_copy(policy->cpus, cpumask_of(cpu));
1406
1407 /*
1408 * Call driver. From then on the cpufreq must be able
1409 * to accept all calls to ->verify and ->setpolicy for this CPU.
1410 */
1411 ret = cpufreq_driver->init(policy);
1412 if (ret) {
1413 pr_debug("%s: %d: initialization failed\n", __func__,
1414 __LINE__);
1415 goto out_clear_policy;
1416 }
1417
1418 /*
1419 * The initialization has succeeded and the policy is online.
1420 * If there is a problem with its frequency table, take it
1421 * offline and drop it.
1422 */
1423 ret = cpufreq_table_validate_and_sort(policy);
1424 if (ret)
1425 goto out_offline_policy;
1426
1427 /* related_cpus should at least include policy->cpus. */
1428 cpumask_copy(policy->related_cpus, policy->cpus);
1429 }
1430
1431 /*
1432 * affected cpus must always be the one, which are online. We aren't
1433 * managing offline cpus here.
1434 */
1435 cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1436
1437 if (new_policy) {
1438 for_each_cpu(j, policy->related_cpus) {
1439 per_cpu(cpufreq_cpu_data, j) = policy;
1440 add_cpu_dev_symlink(policy, j, get_cpu_device(j));
1441 }
1442
1443 policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1444 GFP_KERNEL);
1445 if (!policy->min_freq_req) {
1446 ret = -ENOMEM;
1447 goto out_destroy_policy;
1448 }
1449
1450 ret = freq_qos_add_request(&policy->constraints,
1451 policy->min_freq_req, FREQ_QOS_MIN,
1452 FREQ_QOS_MIN_DEFAULT_VALUE);
1453 if (ret < 0) {
1454 /*
1455 * So we don't call freq_qos_remove_request() for an
1456 * uninitialized request.
1457 */
1458 kfree(policy->min_freq_req);
1459 policy->min_freq_req = NULL;
1460 goto out_destroy_policy;
1461 }
1462
1463 /*
1464 * This must be initialized right here to avoid calling
1465 * freq_qos_remove_request() on uninitialized request in case
1466 * of errors.
1467 */
1468 policy->max_freq_req = policy->min_freq_req + 1;
1469
1470 ret = freq_qos_add_request(&policy->constraints,
1471 policy->max_freq_req, FREQ_QOS_MAX,
1472 FREQ_QOS_MAX_DEFAULT_VALUE);
1473 if (ret < 0) {
1474 policy->max_freq_req = NULL;
1475 goto out_destroy_policy;
1476 }
1477
1478 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1479 CPUFREQ_CREATE_POLICY, policy);
1480 } else {
1481 ret = freq_qos_update_request(policy->max_freq_req, policy->max);
1482 if (ret < 0)
1483 goto out_destroy_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 * em_rebuild_sched_domains() 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 return 0;
1567
1568out_destroy_policy:
1569 for_each_cpu(j, policy->real_cpus)
1570 remove_cpu_dev_symlink(policy, j, get_cpu_device(j));
1571
1572out_offline_policy:
1573 if (cpufreq_driver->offline)
1574 cpufreq_driver->offline(policy);
1575
1576out_exit_policy:
1577 if (cpufreq_driver->exit)
1578 cpufreq_driver->exit(policy);
1579
1580out_clear_policy:
1581 cpumask_clear(policy->cpus);
1582
1583 return ret;
1584}
1585
1586static int cpufreq_online(unsigned int cpu)
1587{
1588 struct cpufreq_policy *policy;
1589 bool new_policy;
1590 int ret;
1591
1592 pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1593
1594 /* Check if this CPU already has a policy to manage it */
1595 policy = per_cpu(cpufreq_cpu_data, cpu);
1596 if (policy) {
1597 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1598 if (!policy_is_inactive(policy))
1599 return cpufreq_add_policy_cpu(policy, cpu);
1600
1601 /* This is the only online CPU for the policy. Start over. */
1602 new_policy = false;
1603 } else {
1604 new_policy = true;
1605 policy = cpufreq_policy_alloc(cpu);
1606 if (!policy)
1607 return -ENOMEM;
1608 }
1609
1610 ret = cpufreq_policy_online(policy, cpu, new_policy);
1611 if (ret) {
1612 cpufreq_policy_free(policy);
1613 return ret;
1614 }
1615
1616 kobject_uevent(&policy->kobj, KOBJ_ADD);
1617
1618 /* Callback for handling stuff after policy is ready */
1619 if (cpufreq_driver->ready)
1620 cpufreq_driver->ready(policy);
1621
1622 /* Register cpufreq cooling only for a new policy */
1623 if (new_policy && cpufreq_thermal_control_enabled(cpufreq_driver))
1624 policy->cdev = of_cpufreq_cooling_register(policy);
1625
1626 /*
1627 * Let the per-policy boost flag mirror the cpufreq_driver boost during
1628 * initialization for a new policy. For an existing policy, maintain the
1629 * previous boost value unless global boost is disabled.
1630 */
1631 if (cpufreq_driver->set_boost && policy->boost_supported &&
1632 (new_policy || !cpufreq_boost_enabled())) {
1633 ret = policy_set_boost(policy, cpufreq_boost_enabled());
1634 if (ret) {
1635 /* If the set_boost fails, the online operation is not affected */
1636 pr_info("%s: CPU%d: Cannot %s BOOST\n", __func__, policy->cpu,
1637 str_enable_disable(cpufreq_boost_enabled()));
1638 }
1639 }
1640
1641 pr_debug("initialization complete\n");
1642
1643 return 0;
1644}
1645
1646/**
1647 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1648 * @dev: CPU device.
1649 * @sif: Subsystem interface structure pointer (not used)
1650 */
1651static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1652{
1653 struct cpufreq_policy *policy;
1654 unsigned cpu = dev->id;
1655 int ret;
1656
1657 dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1658
1659 if (cpu_online(cpu)) {
1660 ret = cpufreq_online(cpu);
1661 if (ret)
1662 return ret;
1663 }
1664
1665 /* Create sysfs link on CPU registration */
1666 policy = per_cpu(cpufreq_cpu_data, cpu);
1667 if (policy)
1668 add_cpu_dev_symlink(policy, cpu, dev);
1669
1670 return 0;
1671}
1672
1673static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
1674{
1675 int ret;
1676
1677 if (has_target())
1678 cpufreq_stop_governor(policy);
1679
1680 cpumask_clear_cpu(cpu, policy->cpus);
1681
1682 if (!policy_is_inactive(policy)) {
1683 /* Nominate a new CPU if necessary. */
1684 if (cpu == policy->cpu)
1685 policy->cpu = cpumask_any(policy->cpus);
1686
1687 /* Start the governor again for the active policy. */
1688 if (has_target()) {
1689 ret = cpufreq_start_governor(policy);
1690 if (ret)
1691 pr_err("%s: Failed to start governor\n", __func__);
1692 }
1693
1694 return;
1695 }
1696
1697 if (has_target())
1698 strscpy(policy->last_governor, policy->governor->name,
1699 CPUFREQ_NAME_LEN);
1700 else
1701 policy->last_policy = policy->policy;
1702
1703 if (has_target())
1704 cpufreq_exit_governor(policy);
1705
1706 /*
1707 * Perform the ->offline() during light-weight tear-down, as
1708 * that allows fast recovery when the CPU comes back.
1709 */
1710 if (cpufreq_driver->offline) {
1711 cpufreq_driver->offline(policy);
1712 return;
1713 }
1714
1715 if (cpufreq_driver->exit)
1716 cpufreq_driver->exit(policy);
1717
1718 policy->freq_table = NULL;
1719}
1720
1721static int cpufreq_offline(unsigned int cpu)
1722{
1723 struct cpufreq_policy *policy;
1724
1725 pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1726
1727 policy = cpufreq_cpu_get_raw(cpu);
1728 if (!policy) {
1729 pr_debug("%s: No cpu_data found\n", __func__);
1730 return 0;
1731 }
1732
1733 guard(cpufreq_policy_write)(policy);
1734
1735 __cpufreq_offline(cpu, policy);
1736
1737 return 0;
1738}
1739
1740/*
1741 * cpufreq_remove_dev - remove a CPU device
1742 *
1743 * Removes the cpufreq interface for a CPU device.
1744 */
1745static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1746{
1747 unsigned int cpu = dev->id;
1748 struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1749
1750 if (!policy)
1751 return;
1752
1753 scoped_guard(cpufreq_policy_write, policy) {
1754 if (cpu_online(cpu))
1755 __cpufreq_offline(cpu, policy);
1756
1757 remove_cpu_dev_symlink(policy, cpu, dev);
1758
1759 if (!cpumask_empty(policy->real_cpus))
1760 return;
1761
1762 /*
1763 * Unregister cpufreq cooling once all the CPUs of the policy
1764 * are removed.
1765 */
1766 if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1767 cpufreq_cooling_unregister(policy->cdev);
1768 policy->cdev = NULL;
1769 }
1770
1771 /* We did light-weight exit earlier, do full tear down now */
1772 if (cpufreq_driver->offline && cpufreq_driver->exit)
1773 cpufreq_driver->exit(policy);
1774 }
1775
1776 cpufreq_policy_free(policy);
1777}
1778
1779/**
1780 * cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1781 * @policy: Policy managing CPUs.
1782 * @new_freq: New CPU frequency.
1783 *
1784 * Adjust to the current frequency first and clean up later by either calling
1785 * cpufreq_update_policy(), or scheduling handle_update().
1786 */
1787static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1788 unsigned int new_freq)
1789{
1790 struct cpufreq_freqs freqs;
1791
1792 pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1793 policy->cur, new_freq);
1794
1795 freqs.old = policy->cur;
1796 freqs.new = new_freq;
1797
1798 cpufreq_freq_transition_begin(policy, &freqs);
1799 cpufreq_freq_transition_end(policy, &freqs, 0);
1800}
1801
1802static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1803{
1804 unsigned int new_freq;
1805
1806 new_freq = cpufreq_driver->get(policy->cpu);
1807 if (!new_freq)
1808 return 0;
1809
1810 /*
1811 * If fast frequency switching is used with the given policy, the check
1812 * against policy->cur is pointless, so skip it in that case.
1813 */
1814 if (policy->fast_switch_enabled || !has_target())
1815 return new_freq;
1816
1817 if (policy->cur != new_freq) {
1818 /*
1819 * For some platforms, the frequency returned by hardware may be
1820 * slightly different from what is provided in the frequency
1821 * table, for example hardware may return 499 MHz instead of 500
1822 * MHz. In such cases it is better to avoid getting into
1823 * unnecessary frequency updates.
1824 */
1825 if (abs(policy->cur - new_freq) < KHZ_PER_MHZ)
1826 return policy->cur;
1827
1828 cpufreq_out_of_sync(policy, new_freq);
1829 if (update)
1830 schedule_work(&policy->update);
1831 }
1832
1833 return new_freq;
1834}
1835
1836/**
1837 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1838 * @cpu: CPU number
1839 *
1840 * This is the last known freq, without actually getting it from the driver.
1841 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1842 */
1843unsigned int cpufreq_quick_get(unsigned int cpu)
1844{
1845 struct cpufreq_policy *policy __free(put_cpufreq_policy) = NULL;
1846 unsigned long flags;
1847
1848 read_lock_irqsave(&cpufreq_driver_lock, flags);
1849
1850 if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1851 unsigned int ret_freq = cpufreq_driver->get(cpu);
1852
1853 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1854
1855 return ret_freq;
1856 }
1857
1858 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1859
1860 policy = cpufreq_cpu_get(cpu);
1861 if (policy)
1862 return policy->cur;
1863
1864 return 0;
1865}
1866EXPORT_SYMBOL(cpufreq_quick_get);
1867
1868/**
1869 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1870 * @cpu: CPU number
1871 *
1872 * Just return the max possible frequency for a given CPU.
1873 */
1874unsigned int cpufreq_quick_get_max(unsigned int cpu)
1875{
1876 struct cpufreq_policy *policy __free(put_cpufreq_policy);
1877
1878 policy = cpufreq_cpu_get(cpu);
1879 if (policy)
1880 return policy->max;
1881
1882 return 0;
1883}
1884EXPORT_SYMBOL(cpufreq_quick_get_max);
1885
1886/**
1887 * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1888 * @cpu: CPU number
1889 *
1890 * The default return value is the max_freq field of cpuinfo.
1891 */
1892__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1893{
1894 struct cpufreq_policy *policy __free(put_cpufreq_policy);
1895
1896 policy = cpufreq_cpu_get(cpu);
1897 if (policy)
1898 return policy->cpuinfo.max_freq;
1899
1900 return 0;
1901}
1902EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1903
1904static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1905{
1906 if (unlikely(policy_is_inactive(policy)))
1907 return 0;
1908
1909 return cpufreq_verify_current_freq(policy, true);
1910}
1911
1912/**
1913 * cpufreq_get - get the current CPU frequency (in kHz)
1914 * @cpu: CPU number
1915 *
1916 * Get the CPU current (static) CPU frequency
1917 */
1918unsigned int cpufreq_get(unsigned int cpu)
1919{
1920 struct cpufreq_policy *policy __free(put_cpufreq_policy);
1921
1922 policy = cpufreq_cpu_get(cpu);
1923 if (!policy)
1924 return 0;
1925
1926 guard(cpufreq_policy_read)(policy);
1927
1928 if (cpufreq_driver->get)
1929 return __cpufreq_get(policy);
1930
1931 return 0;
1932}
1933EXPORT_SYMBOL(cpufreq_get);
1934
1935static struct subsys_interface cpufreq_interface = {
1936 .name = "cpufreq",
1937 .subsys = &cpu_subsys,
1938 .add_dev = cpufreq_add_dev,
1939 .remove_dev = cpufreq_remove_dev,
1940};
1941
1942/*
1943 * In case platform wants some specific frequency to be configured
1944 * during suspend..
1945 */
1946int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1947{
1948 int ret;
1949
1950 if (!policy->suspend_freq) {
1951 pr_debug("%s: suspend_freq not defined\n", __func__);
1952 return 0;
1953 }
1954
1955 pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1956 policy->suspend_freq);
1957
1958 ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1959 CPUFREQ_RELATION_H);
1960 if (ret)
1961 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1962 __func__, policy->suspend_freq, ret);
1963
1964 return ret;
1965}
1966EXPORT_SYMBOL(cpufreq_generic_suspend);
1967
1968/**
1969 * cpufreq_suspend() - Suspend CPUFreq governors.
1970 *
1971 * Called during system wide Suspend/Hibernate cycles for suspending governors
1972 * as some platforms can't change frequency after this point in suspend cycle.
1973 * Because some of the devices (like: i2c, regulators, etc) they use for
1974 * changing frequency are suspended quickly after this point.
1975 */
1976void cpufreq_suspend(void)
1977{
1978 struct cpufreq_policy *policy;
1979
1980 if (!cpufreq_driver)
1981 return;
1982
1983 if (!has_target() && !cpufreq_driver->suspend)
1984 goto suspend;
1985
1986 pr_debug("%s: Suspending Governors\n", __func__);
1987
1988 for_each_active_policy(policy) {
1989 if (has_target()) {
1990 scoped_guard(cpufreq_policy_write, policy) {
1991 cpufreq_stop_governor(policy);
1992 }
1993 }
1994
1995 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1996 pr_err("%s: Failed to suspend driver: %s\n", __func__,
1997 cpufreq_driver->name);
1998 }
1999
2000suspend:
2001 cpufreq_suspended = true;
2002}
2003
2004/**
2005 * cpufreq_resume() - Resume CPUFreq governors.
2006 *
2007 * Called during system wide Suspend/Hibernate cycle for resuming governors that
2008 * are suspended with cpufreq_suspend().
2009 */
2010void cpufreq_resume(void)
2011{
2012 struct cpufreq_policy *policy;
2013 int ret;
2014
2015 if (!cpufreq_driver)
2016 return;
2017
2018 if (unlikely(!cpufreq_suspended))
2019 return;
2020
2021 cpufreq_suspended = false;
2022
2023 if (!has_target() && !cpufreq_driver->resume)
2024 return;
2025
2026 pr_debug("%s: Resuming Governors\n", __func__);
2027
2028 for_each_active_policy(policy) {
2029 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
2030 pr_err("%s: Failed to resume driver: %s\n", __func__,
2031 cpufreq_driver->name);
2032 } else if (has_target()) {
2033 scoped_guard(cpufreq_policy_write, policy) {
2034 ret = cpufreq_start_governor(policy);
2035 }
2036
2037 if (ret)
2038 pr_err("%s: Failed to start governor for CPU%u's policy\n",
2039 __func__, policy->cpu);
2040 }
2041 }
2042}
2043
2044/**
2045 * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
2046 * @flags: Flags to test against the current cpufreq driver's flags.
2047 *
2048 * Assumes that the driver is there, so callers must ensure that this is the
2049 * case.
2050 */
2051bool cpufreq_driver_test_flags(u16 flags)
2052{
2053 return !!(cpufreq_driver->flags & flags);
2054}
2055
2056/**
2057 * cpufreq_get_current_driver - Return the current driver's name.
2058 *
2059 * Return the name string of the currently registered cpufreq driver or NULL if
2060 * none.
2061 */
2062const char *cpufreq_get_current_driver(void)
2063{
2064 if (cpufreq_driver)
2065 return cpufreq_driver->name;
2066
2067 return NULL;
2068}
2069EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
2070
2071/**
2072 * cpufreq_get_driver_data - Return current driver data.
2073 *
2074 * Return the private data of the currently registered cpufreq driver, or NULL
2075 * if no cpufreq driver has been registered.
2076 */
2077void *cpufreq_get_driver_data(void)
2078{
2079 if (cpufreq_driver)
2080 return cpufreq_driver->driver_data;
2081
2082 return NULL;
2083}
2084EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
2085
2086/*********************************************************************
2087 * NOTIFIER LISTS INTERFACE *
2088 *********************************************************************/
2089
2090/**
2091 * cpufreq_register_notifier - Register a notifier with cpufreq.
2092 * @nb: notifier function to register.
2093 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2094 *
2095 * Add a notifier to one of two lists: either a list of notifiers that run on
2096 * clock rate changes (once before and once after every transition), or a list
2097 * of notifiers that ron on cpufreq policy changes.
2098 *
2099 * This function may sleep and it has the same return values as
2100 * blocking_notifier_chain_register().
2101 */
2102int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
2103{
2104 int ret;
2105
2106 if (cpufreq_disabled())
2107 return -EINVAL;
2108
2109 switch (list) {
2110 case CPUFREQ_TRANSITION_NOTIFIER:
2111 mutex_lock(&cpufreq_fast_switch_lock);
2112
2113 if (cpufreq_fast_switch_count > 0) {
2114 mutex_unlock(&cpufreq_fast_switch_lock);
2115 return -EBUSY;
2116 }
2117 ret = srcu_notifier_chain_register(
2118 &cpufreq_transition_notifier_list, nb);
2119 if (!ret)
2120 cpufreq_fast_switch_count--;
2121
2122 mutex_unlock(&cpufreq_fast_switch_lock);
2123 break;
2124 case CPUFREQ_POLICY_NOTIFIER:
2125 ret = blocking_notifier_chain_register(
2126 &cpufreq_policy_notifier_list, nb);
2127 break;
2128 default:
2129 ret = -EINVAL;
2130 }
2131
2132 return ret;
2133}
2134EXPORT_SYMBOL(cpufreq_register_notifier);
2135
2136/**
2137 * cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2138 * @nb: notifier block to be unregistered.
2139 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2140 *
2141 * Remove a notifier from one of the cpufreq notifier lists.
2142 *
2143 * This function may sleep and it has the same return values as
2144 * blocking_notifier_chain_unregister().
2145 */
2146int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2147{
2148 int ret;
2149
2150 if (cpufreq_disabled())
2151 return -EINVAL;
2152
2153 switch (list) {
2154 case CPUFREQ_TRANSITION_NOTIFIER:
2155 mutex_lock(&cpufreq_fast_switch_lock);
2156
2157 ret = srcu_notifier_chain_unregister(
2158 &cpufreq_transition_notifier_list, nb);
2159 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2160 cpufreq_fast_switch_count++;
2161
2162 mutex_unlock(&cpufreq_fast_switch_lock);
2163 break;
2164 case CPUFREQ_POLICY_NOTIFIER:
2165 ret = blocking_notifier_chain_unregister(
2166 &cpufreq_policy_notifier_list, nb);
2167 break;
2168 default:
2169 ret = -EINVAL;
2170 }
2171
2172 return ret;
2173}
2174EXPORT_SYMBOL(cpufreq_unregister_notifier);
2175
2176
2177/*********************************************************************
2178 * GOVERNORS *
2179 *********************************************************************/
2180
2181/**
2182 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2183 * @policy: cpufreq policy to switch the frequency for.
2184 * @target_freq: New frequency to set (may be approximate).
2185 *
2186 * Carry out a fast frequency switch without sleeping.
2187 *
2188 * The driver's ->fast_switch() callback invoked by this function must be
2189 * suitable for being called from within RCU-sched read-side critical sections
2190 * and it is expected to select the minimum available frequency greater than or
2191 * equal to @target_freq (CPUFREQ_RELATION_L).
2192 *
2193 * This function must not be called if policy->fast_switch_enabled is unset.
2194 *
2195 * Governors calling this function must guarantee that it will never be invoked
2196 * twice in parallel for the same policy and that it will never be called in
2197 * parallel with either ->target() or ->target_index() for the same policy.
2198 *
2199 * Returns the actual frequency set for the CPU.
2200 *
2201 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2202 * error condition, the hardware configuration must be preserved.
2203 */
2204unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2205 unsigned int target_freq)
2206{
2207 unsigned int freq;
2208 int cpu;
2209
2210 target_freq = clamp_val(target_freq, policy->min, policy->max);
2211 freq = cpufreq_driver->fast_switch(policy, target_freq);
2212
2213 if (!freq)
2214 return 0;
2215
2216 policy->cur = freq;
2217 arch_set_freq_scale(policy->related_cpus, freq,
2218 arch_scale_freq_ref(policy->cpu));
2219 cpufreq_stats_record_transition(policy, freq);
2220
2221 if (trace_cpu_frequency_enabled()) {
2222 for_each_cpu(cpu, policy->cpus)
2223 trace_cpu_frequency(freq, cpu);
2224 }
2225
2226 return freq;
2227}
2228EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2229
2230/**
2231 * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2232 * @cpu: Target CPU.
2233 * @min_perf: Minimum (required) performance level (units of @capacity).
2234 * @target_perf: Target (desired) performance level (units of @capacity).
2235 * @capacity: Capacity of the target CPU.
2236 *
2237 * Carry out a fast performance level switch of @cpu without sleeping.
2238 *
2239 * The driver's ->adjust_perf() callback invoked by this function must be
2240 * suitable for being called from within RCU-sched read-side critical sections
2241 * and it is expected to select a suitable performance level equal to or above
2242 * @min_perf and preferably equal to or below @target_perf.
2243 *
2244 * This function must not be called if policy->fast_switch_enabled is unset.
2245 *
2246 * Governors calling this function must guarantee that it will never be invoked
2247 * twice in parallel for the same CPU and that it will never be called in
2248 * parallel with either ->target() or ->target_index() or ->fast_switch() for
2249 * the same CPU.
2250 */
2251void cpufreq_driver_adjust_perf(unsigned int cpu,
2252 unsigned long min_perf,
2253 unsigned long target_perf,
2254 unsigned long capacity)
2255{
2256 cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2257}
2258
2259/**
2260 * cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2261 *
2262 * Return 'true' if the ->adjust_perf callback is present for the
2263 * current driver or 'false' otherwise.
2264 */
2265bool cpufreq_driver_has_adjust_perf(void)
2266{
2267 return !!cpufreq_driver->adjust_perf;
2268}
2269
2270/* Must set freqs->new to intermediate frequency */
2271static int __target_intermediate(struct cpufreq_policy *policy,
2272 struct cpufreq_freqs *freqs, int index)
2273{
2274 int ret;
2275
2276 freqs->new = cpufreq_driver->get_intermediate(policy, index);
2277
2278 /* We don't need to switch to intermediate freq */
2279 if (!freqs->new)
2280 return 0;
2281
2282 pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2283 __func__, policy->cpu, freqs->old, freqs->new);
2284
2285 cpufreq_freq_transition_begin(policy, freqs);
2286 ret = cpufreq_driver->target_intermediate(policy, index);
2287 cpufreq_freq_transition_end(policy, freqs, ret);
2288
2289 if (ret)
2290 pr_err("%s: Failed to change to intermediate frequency: %d\n",
2291 __func__, ret);
2292
2293 return ret;
2294}
2295
2296static int __target_index(struct cpufreq_policy *policy, int index)
2297{
2298 struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2299 unsigned int restore_freq, intermediate_freq = 0;
2300 unsigned int newfreq = policy->freq_table[index].frequency;
2301 int retval = -EINVAL;
2302 bool notify;
2303
2304 if (newfreq == policy->cur)
2305 return 0;
2306
2307 /* Save last value to restore later on errors */
2308 restore_freq = policy->cur;
2309
2310 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2311 if (notify) {
2312 /* Handle switching to intermediate frequency */
2313 if (cpufreq_driver->get_intermediate) {
2314 retval = __target_intermediate(policy, &freqs, index);
2315 if (retval)
2316 return retval;
2317
2318 intermediate_freq = freqs.new;
2319 /* Set old freq to intermediate */
2320 if (intermediate_freq)
2321 freqs.old = freqs.new;
2322 }
2323
2324 freqs.new = newfreq;
2325 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2326 __func__, policy->cpu, freqs.old, freqs.new);
2327
2328 cpufreq_freq_transition_begin(policy, &freqs);
2329 }
2330
2331 retval = cpufreq_driver->target_index(policy, index);
2332 if (retval)
2333 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2334 retval);
2335
2336 if (notify) {
2337 cpufreq_freq_transition_end(policy, &freqs, retval);
2338
2339 /*
2340 * Failed after setting to intermediate freq? Driver should have
2341 * reverted back to initial frequency and so should we. Check
2342 * here for intermediate_freq instead of get_intermediate, in
2343 * case we haven't switched to intermediate freq at all.
2344 */
2345 if (unlikely(retval && intermediate_freq)) {
2346 freqs.old = intermediate_freq;
2347 freqs.new = restore_freq;
2348 cpufreq_freq_transition_begin(policy, &freqs);
2349 cpufreq_freq_transition_end(policy, &freqs, 0);
2350 }
2351 }
2352
2353 return retval;
2354}
2355
2356int __cpufreq_driver_target(struct cpufreq_policy *policy,
2357 unsigned int target_freq,
2358 unsigned int relation)
2359{
2360 unsigned int old_target_freq = target_freq;
2361
2362 if (cpufreq_disabled())
2363 return -ENODEV;
2364
2365 target_freq = __resolve_freq(policy, target_freq, policy->min,
2366 policy->max, relation);
2367
2368 pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2369 policy->cpu, target_freq, relation, old_target_freq);
2370
2371 /*
2372 * This might look like a redundant call as we are checking it again
2373 * after finding index. But it is left intentionally for cases where
2374 * exactly same freq is called again and so we can save on few function
2375 * calls.
2376 */
2377 if (target_freq == policy->cur &&
2378 !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2379 return 0;
2380
2381 if (cpufreq_driver->target) {
2382 /*
2383 * If the driver hasn't setup a single inefficient frequency,
2384 * it's unlikely it knows how to decode CPUFREQ_RELATION_E.
2385 */
2386 if (!policy->efficiencies_available)
2387 relation &= ~CPUFREQ_RELATION_E;
2388
2389 return cpufreq_driver->target(policy, target_freq, relation);
2390 }
2391
2392 if (!cpufreq_driver->target_index)
2393 return -EINVAL;
2394
2395 return __target_index(policy, policy->cached_resolved_idx);
2396}
2397EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2398
2399int cpufreq_driver_target(struct cpufreq_policy *policy,
2400 unsigned int target_freq,
2401 unsigned int relation)
2402{
2403 guard(cpufreq_policy_write)(policy);
2404
2405 return __cpufreq_driver_target(policy, target_freq, relation);
2406}
2407EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2408
2409__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2410{
2411 return NULL;
2412}
2413
2414static int cpufreq_init_governor(struct cpufreq_policy *policy)
2415{
2416 int ret;
2417
2418 /* Don't start any governor operations if we are entering suspend */
2419 if (cpufreq_suspended)
2420 return 0;
2421 /*
2422 * Governor might not be initiated here if ACPI _PPC changed
2423 * notification happened, so check it.
2424 */
2425 if (!policy->governor)
2426 return -EINVAL;
2427
2428 /* Platform doesn't want dynamic frequency switching ? */
2429 if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2430 cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2431 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2432
2433 if (gov) {
2434 pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2435 policy->governor->name, gov->name);
2436 policy->governor = gov;
2437 } else {
2438 return -EINVAL;
2439 }
2440 }
2441
2442 if (!try_module_get(policy->governor->owner))
2443 return -EINVAL;
2444
2445 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2446
2447 if (policy->governor->init) {
2448 ret = policy->governor->init(policy);
2449 if (ret) {
2450 module_put(policy->governor->owner);
2451 return ret;
2452 }
2453 }
2454
2455 policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2456
2457 return 0;
2458}
2459
2460static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2461{
2462 if (cpufreq_suspended || !policy->governor)
2463 return;
2464
2465 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2466
2467 if (policy->governor->exit)
2468 policy->governor->exit(policy);
2469
2470 module_put(policy->governor->owner);
2471}
2472
2473int cpufreq_start_governor(struct cpufreq_policy *policy)
2474{
2475 int ret;
2476
2477 if (cpufreq_suspended)
2478 return 0;
2479
2480 if (!policy->governor)
2481 return -EINVAL;
2482
2483 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2484
2485 if (cpufreq_driver->get)
2486 cpufreq_verify_current_freq(policy, false);
2487
2488 if (policy->governor->start) {
2489 ret = policy->governor->start(policy);
2490 if (ret)
2491 return ret;
2492 }
2493
2494 if (policy->governor->limits)
2495 policy->governor->limits(policy);
2496
2497 return 0;
2498}
2499
2500void cpufreq_stop_governor(struct cpufreq_policy *policy)
2501{
2502 if (cpufreq_suspended || !policy->governor)
2503 return;
2504
2505 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2506
2507 if (policy->governor->stop)
2508 policy->governor->stop(policy);
2509}
2510
2511static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2512{
2513 if (cpufreq_suspended || !policy->governor)
2514 return;
2515
2516 pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2517
2518 if (policy->governor->limits)
2519 policy->governor->limits(policy);
2520}
2521
2522int cpufreq_register_governor(struct cpufreq_governor *governor)
2523{
2524 int err;
2525
2526 if (!governor)
2527 return -EINVAL;
2528
2529 if (cpufreq_disabled())
2530 return -ENODEV;
2531
2532 mutex_lock(&cpufreq_governor_mutex);
2533
2534 err = -EBUSY;
2535 if (!find_governor(governor->name)) {
2536 err = 0;
2537 list_add(&governor->governor_list, &cpufreq_governor_list);
2538 }
2539
2540 mutex_unlock(&cpufreq_governor_mutex);
2541 return err;
2542}
2543EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2544
2545void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2546{
2547 struct cpufreq_policy *policy;
2548 unsigned long flags;
2549
2550 if (!governor)
2551 return;
2552
2553 if (cpufreq_disabled())
2554 return;
2555
2556 /* clear last_governor for all inactive policies */
2557 read_lock_irqsave(&cpufreq_driver_lock, flags);
2558 for_each_inactive_policy(policy) {
2559 if (!strcmp(policy->last_governor, governor->name)) {
2560 policy->governor = NULL;
2561 strcpy(policy->last_governor, "\0");
2562 }
2563 }
2564 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2565
2566 mutex_lock(&cpufreq_governor_mutex);
2567 list_del(&governor->governor_list);
2568 mutex_unlock(&cpufreq_governor_mutex);
2569}
2570EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2571
2572
2573/*********************************************************************
2574 * POLICY INTERFACE *
2575 *********************************************************************/
2576
2577DEFINE_PER_CPU(unsigned long, cpufreq_pressure);
2578
2579/**
2580 * cpufreq_update_pressure() - Update cpufreq pressure for CPUs
2581 * @policy: cpufreq policy of the CPUs.
2582 *
2583 * Update the value of cpufreq pressure for all @cpus in the policy.
2584 */
2585static void cpufreq_update_pressure(struct cpufreq_policy *policy)
2586{
2587 unsigned long max_capacity, capped_freq, pressure;
2588 u32 max_freq;
2589 int cpu;
2590
2591 cpu = cpumask_first(policy->related_cpus);
2592 max_freq = arch_scale_freq_ref(cpu);
2593 capped_freq = policy->max;
2594
2595 /*
2596 * Handle properly the boost frequencies, which should simply clean
2597 * the cpufreq pressure value.
2598 */
2599 if (max_freq <= capped_freq) {
2600 pressure = 0;
2601 } else {
2602 max_capacity = arch_scale_cpu_capacity(cpu);
2603 pressure = max_capacity -
2604 mult_frac(max_capacity, capped_freq, max_freq);
2605 }
2606
2607 for_each_cpu(cpu, policy->related_cpus)
2608 WRITE_ONCE(per_cpu(cpufreq_pressure, cpu), pressure);
2609}
2610
2611/**
2612 * cpufreq_set_policy - Modify cpufreq policy parameters.
2613 * @policy: Policy object to modify.
2614 * @new_gov: Policy governor pointer.
2615 * @new_pol: Policy value (for drivers with built-in governors).
2616 *
2617 * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2618 * limits to be set for the policy, update @policy with the verified limits
2619 * values and either invoke the driver's ->setpolicy() callback (if present) or
2620 * carry out a governor update for @policy. That is, run the current governor's
2621 * ->limits() callback (if @new_gov points to the same object as the one in
2622 * @policy) or replace the governor for @policy with @new_gov.
2623 *
2624 * The cpuinfo part of @policy is not updated by this function.
2625 */
2626static int cpufreq_set_policy(struct cpufreq_policy *policy,
2627 struct cpufreq_governor *new_gov,
2628 unsigned int new_pol)
2629{
2630 struct cpufreq_policy_data new_data;
2631 struct cpufreq_governor *old_gov;
2632 int ret;
2633
2634 memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2635 new_data.freq_table = policy->freq_table;
2636 new_data.cpu = policy->cpu;
2637 /*
2638 * PM QoS framework collects all the requests from users and provide us
2639 * the final aggregated value here.
2640 */
2641 new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2642 new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2643
2644 pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2645 new_data.cpu, new_data.min, new_data.max);
2646
2647 /*
2648 * Verify that the CPU speed can be set within these limits and make sure
2649 * that min <= max.
2650 */
2651 ret = cpufreq_driver->verify(&new_data);
2652 if (ret)
2653 return ret;
2654
2655 /*
2656 * Resolve policy min/max to available frequencies. It ensures
2657 * no frequency resolution will neither overshoot the requested maximum
2658 * nor undershoot the requested minimum.
2659 *
2660 * Avoid storing intermediate values in policy->max or policy->min and
2661 * compiler optimizations around them because they may be accessed
2662 * concurrently by cpufreq_driver_resolve_freq() during the update.
2663 */
2664 WRITE_ONCE(policy->max, __resolve_freq(policy, new_data.max,
2665 new_data.min, new_data.max,
2666 CPUFREQ_RELATION_H));
2667 new_data.min = __resolve_freq(policy, new_data.min, new_data.min,
2668 new_data.max, CPUFREQ_RELATION_L);
2669 WRITE_ONCE(policy->min, new_data.min > policy->max ? policy->max : new_data.min);
2670
2671 trace_cpu_frequency_limits(policy);
2672
2673 cpufreq_update_pressure(policy);
2674
2675 policy->cached_target_freq = UINT_MAX;
2676
2677 pr_debug("new min and max freqs are %u - %u kHz\n",
2678 policy->min, policy->max);
2679
2680 if (cpufreq_driver->setpolicy) {
2681 policy->policy = new_pol;
2682 pr_debug("setting range\n");
2683 return cpufreq_driver->setpolicy(policy);
2684 }
2685
2686 if (new_gov == policy->governor) {
2687 pr_debug("governor limits update\n");
2688 cpufreq_governor_limits(policy);
2689 return 0;
2690 }
2691
2692 pr_debug("governor switch\n");
2693
2694 /* save old, working values */
2695 old_gov = policy->governor;
2696 /* end old governor */
2697 if (old_gov) {
2698 cpufreq_stop_governor(policy);
2699 cpufreq_exit_governor(policy);
2700 }
2701
2702 /* start new governor */
2703 policy->governor = new_gov;
2704 ret = cpufreq_init_governor(policy);
2705 if (!ret) {
2706 ret = cpufreq_start_governor(policy);
2707 if (!ret) {
2708 pr_debug("governor change\n");
2709 return 0;
2710 }
2711 cpufreq_exit_governor(policy);
2712 }
2713
2714 /* new governor failed, so re-start old one */
2715 pr_debug("starting governor %s failed\n", policy->governor->name);
2716 if (old_gov) {
2717 policy->governor = old_gov;
2718 if (cpufreq_init_governor(policy))
2719 policy->governor = NULL;
2720 else
2721 cpufreq_start_governor(policy);
2722 }
2723
2724 return ret;
2725}
2726
2727static void cpufreq_policy_refresh(struct cpufreq_policy *policy)
2728{
2729 guard(cpufreq_policy_write)(policy);
2730
2731 /*
2732 * BIOS might change freq behind our back
2733 * -> ask driver for current freq and notify governors about a change
2734 */
2735 if (cpufreq_driver->get && has_target() &&
2736 (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2737 return;
2738
2739 refresh_frequency_limits(policy);
2740}
2741
2742/**
2743 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2744 * @cpu: CPU to re-evaluate the policy for.
2745 *
2746 * Update the current frequency for the cpufreq policy of @cpu and use
2747 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2748 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2749 * for the policy in question, among other things.
2750 */
2751void cpufreq_update_policy(unsigned int cpu)
2752{
2753 struct cpufreq_policy *policy __free(put_cpufreq_policy);
2754
2755 policy = cpufreq_cpu_get(cpu);
2756 if (!policy)
2757 return;
2758
2759 cpufreq_policy_refresh(policy);
2760}
2761EXPORT_SYMBOL(cpufreq_update_policy);
2762
2763/**
2764 * cpufreq_update_limits - Update policy limits for a given CPU.
2765 * @cpu: CPU to update the policy limits for.
2766 *
2767 * Invoke the driver's ->update_limits callback if present or call
2768 * cpufreq_policy_refresh() for @cpu.
2769 */
2770void cpufreq_update_limits(unsigned int cpu)
2771{
2772 struct cpufreq_policy *policy __free(put_cpufreq_policy);
2773
2774 policy = cpufreq_cpu_get(cpu);
2775 if (!policy)
2776 return;
2777
2778 if (cpufreq_driver->update_limits)
2779 cpufreq_driver->update_limits(policy);
2780 else
2781 cpufreq_policy_refresh(policy);
2782}
2783EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2784
2785/*********************************************************************
2786 * BOOST *
2787 *********************************************************************/
2788int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2789{
2790 int ret;
2791
2792 if (!policy->freq_table)
2793 return -ENXIO;
2794
2795 ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2796 if (ret) {
2797 pr_err("%s: Policy frequency update failed\n", __func__);
2798 return ret;
2799 }
2800
2801 ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2802 if (ret < 0)
2803 return ret;
2804
2805 return 0;
2806}
2807EXPORT_SYMBOL_GPL(cpufreq_boost_set_sw);
2808
2809static int cpufreq_boost_trigger_state(int state)
2810{
2811 struct cpufreq_policy *policy;
2812 unsigned long flags;
2813 int ret = 0;
2814
2815 /*
2816 * Don't compare 'cpufreq_driver->boost_enabled' with 'state' here to
2817 * make sure all policies are in sync with global boost flag.
2818 */
2819
2820 write_lock_irqsave(&cpufreq_driver_lock, flags);
2821 cpufreq_driver->boost_enabled = state;
2822 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2823
2824 cpus_read_lock();
2825 for_each_active_policy(policy) {
2826 if (!policy->boost_supported)
2827 continue;
2828
2829 ret = policy_set_boost(policy, state);
2830 if (ret)
2831 goto err_reset_state;
2832 }
2833 cpus_read_unlock();
2834
2835 return 0;
2836
2837err_reset_state:
2838 cpus_read_unlock();
2839
2840 write_lock_irqsave(&cpufreq_driver_lock, flags);
2841 cpufreq_driver->boost_enabled = !state;
2842 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2843
2844 pr_err("%s: Cannot %s BOOST\n",
2845 __func__, str_enable_disable(state));
2846
2847 return ret;
2848}
2849
2850static bool cpufreq_boost_supported(void)
2851{
2852 return cpufreq_driver->set_boost;
2853}
2854
2855static int create_boost_sysfs_file(void)
2856{
2857 int ret;
2858
2859 ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2860 if (ret)
2861 pr_err("%s: cannot register global BOOST sysfs file\n",
2862 __func__);
2863
2864 return ret;
2865}
2866
2867static void remove_boost_sysfs_file(void)
2868{
2869 if (cpufreq_boost_supported())
2870 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2871}
2872
2873bool cpufreq_boost_enabled(void)
2874{
2875 return cpufreq_driver->boost_enabled;
2876}
2877EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2878
2879/*********************************************************************
2880 * REGISTER / UNREGISTER CPUFREQ DRIVER *
2881 *********************************************************************/
2882static enum cpuhp_state hp_online;
2883
2884static int cpuhp_cpufreq_online(unsigned int cpu)
2885{
2886 cpufreq_online(cpu);
2887
2888 return 0;
2889}
2890
2891static int cpuhp_cpufreq_offline(unsigned int cpu)
2892{
2893 cpufreq_offline(cpu);
2894
2895 return 0;
2896}
2897
2898/**
2899 * cpufreq_register_driver - register a CPU Frequency driver
2900 * @driver_data: A struct cpufreq_driver containing the values#
2901 * submitted by the CPU Frequency driver.
2902 *
2903 * Registers a CPU Frequency driver to this core code. This code
2904 * returns zero on success, -EEXIST when another driver got here first
2905 * (and isn't unregistered in the meantime).
2906 *
2907 */
2908int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2909{
2910 unsigned long flags;
2911 int ret;
2912
2913 if (cpufreq_disabled())
2914 return -ENODEV;
2915
2916 /*
2917 * The cpufreq core depends heavily on the availability of device
2918 * structure, make sure they are available before proceeding further.
2919 */
2920 if (!get_cpu_device(0))
2921 return -EPROBE_DEFER;
2922
2923 if (!driver_data || !driver_data->verify || !driver_data->init ||
2924 !(driver_data->setpolicy || driver_data->target_index ||
2925 driver_data->target) ||
2926 (driver_data->setpolicy && (driver_data->target_index ||
2927 driver_data->target)) ||
2928 (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2929 (!driver_data->online != !driver_data->offline) ||
2930 (driver_data->adjust_perf && !driver_data->fast_switch))
2931 return -EINVAL;
2932
2933 pr_debug("trying to register driver %s\n", driver_data->name);
2934
2935 /* Protect against concurrent CPU online/offline. */
2936 cpus_read_lock();
2937
2938 write_lock_irqsave(&cpufreq_driver_lock, flags);
2939 if (cpufreq_driver) {
2940 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2941 ret = -EEXIST;
2942 goto out;
2943 }
2944 cpufreq_driver = driver_data;
2945 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2946
2947 /*
2948 * Mark support for the scheduler's frequency invariance engine for
2949 * drivers that implement target(), target_index() or fast_switch().
2950 */
2951 if (!cpufreq_driver->setpolicy) {
2952 static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2953 pr_debug("supports frequency invariance");
2954 }
2955
2956 if (driver_data->setpolicy)
2957 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2958
2959 if (cpufreq_boost_supported()) {
2960 ret = create_boost_sysfs_file();
2961 if (ret)
2962 goto err_null_driver;
2963 }
2964
2965 ret = subsys_interface_register(&cpufreq_interface);
2966 if (ret)
2967 goto err_boost_unreg;
2968
2969 if (unlikely(list_empty(&cpufreq_policy_list))) {
2970 /* if all ->init() calls failed, unregister */
2971 ret = -ENODEV;
2972 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2973 driver_data->name);
2974 goto err_if_unreg;
2975 }
2976
2977 ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2978 "cpufreq:online",
2979 cpuhp_cpufreq_online,
2980 cpuhp_cpufreq_offline);
2981 if (ret < 0)
2982 goto err_if_unreg;
2983 hp_online = ret;
2984 ret = 0;
2985
2986 pr_debug("driver %s up and running\n", driver_data->name);
2987 goto out;
2988
2989err_if_unreg:
2990 subsys_interface_unregister(&cpufreq_interface);
2991err_boost_unreg:
2992 remove_boost_sysfs_file();
2993err_null_driver:
2994 write_lock_irqsave(&cpufreq_driver_lock, flags);
2995 cpufreq_driver = NULL;
2996 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2997out:
2998 cpus_read_unlock();
2999 return ret;
3000}
3001EXPORT_SYMBOL_GPL(cpufreq_register_driver);
3002
3003/*
3004 * cpufreq_unregister_driver - unregister the current CPUFreq driver
3005 *
3006 * Unregister the current CPUFreq driver. Only call this if you have
3007 * the right to do so, i.e. if you have succeeded in initialising before!
3008 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
3009 * currently not initialised.
3010 */
3011void cpufreq_unregister_driver(struct cpufreq_driver *driver)
3012{
3013 unsigned long flags;
3014
3015 if (WARN_ON(!cpufreq_driver || (driver != cpufreq_driver)))
3016 return;
3017
3018 pr_debug("unregistering driver %s\n", driver->name);
3019
3020 /* Protect against concurrent cpu hotplug */
3021 cpus_read_lock();
3022 subsys_interface_unregister(&cpufreq_interface);
3023 remove_boost_sysfs_file();
3024 static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
3025 cpuhp_remove_state_nocalls_cpuslocked(hp_online);
3026
3027 write_lock_irqsave(&cpufreq_driver_lock, flags);
3028
3029 cpufreq_driver = NULL;
3030
3031 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
3032 cpus_read_unlock();
3033}
3034EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
3035
3036static int __init cpufreq_core_init(void)
3037{
3038 struct cpufreq_governor *gov = cpufreq_default_governor();
3039 struct device *dev_root;
3040
3041 if (cpufreq_disabled())
3042 return -ENODEV;
3043
3044 dev_root = bus_get_dev_root(&cpu_subsys);
3045 if (dev_root) {
3046 cpufreq_global_kobject = kobject_create_and_add("cpufreq", &dev_root->kobj);
3047 put_device(dev_root);
3048 }
3049 BUG_ON(!cpufreq_global_kobject);
3050
3051 if (!strlen(default_governor))
3052 strscpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
3053
3054 return 0;
3055}
3056
3057static bool cpufreq_policy_is_good_for_eas(unsigned int cpu)
3058{
3059 struct cpufreq_policy *policy __free(put_cpufreq_policy);
3060
3061 policy = cpufreq_cpu_get(cpu);
3062 if (!policy) {
3063 pr_debug("cpufreq policy not set for CPU: %d\n", cpu);
3064 return false;
3065 }
3066
3067 return sugov_is_governor(policy);
3068}
3069
3070bool cpufreq_ready_for_eas(const struct cpumask *cpu_mask)
3071{
3072 unsigned int cpu;
3073
3074 /* Do not attempt EAS if schedutil is not being used. */
3075 for_each_cpu(cpu, cpu_mask) {
3076 if (!cpufreq_policy_is_good_for_eas(cpu)) {
3077 pr_debug("rd %*pbl: schedutil is mandatory for EAS\n",
3078 cpumask_pr_args(cpu_mask));
3079 return false;
3080 }
3081 }
3082
3083 return true;
3084}
3085
3086module_param(off, int, 0444);
3087module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
3088core_initcall(cpufreq_core_init);
Linux 6.x block layer and io_uring tree(s)