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