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