1 // SPDX-License-Identifier: GPL-2.0
3 * CPUFreq governor based on scheduler-provided CPU utilization data.
5 * Copyright (C) 2016, Intel Corporation
6 * Author: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/sched/cpufreq.h>
14 #include <trace/events/power.h>
16 #define IOWAIT_BOOST_MIN (SCHED_CAPACITY_SCALE / 8)
18 struct sugov_tunables {
19 struct gov_attr_set attr_set;
20 unsigned int rate_limit_us;
24 struct cpufreq_policy *policy;
26 struct sugov_tunables *tunables;
27 struct list_head tunables_hook;
29 raw_spinlock_t update_lock; /* For shared policies */
30 u64 last_freq_update_time;
31 s64 freq_update_delay_ns;
32 unsigned int next_freq;
33 unsigned int cached_raw_freq;
35 /* The next fields are only needed if fast switch cannot be used: */
36 struct irq_work irq_work;
37 struct kthread_work work;
38 struct mutex work_lock;
39 struct kthread_worker worker;
40 struct task_struct *thread;
41 bool work_in_progress;
44 bool need_freq_update;
48 struct update_util_data update_util;
49 struct sugov_policy *sg_policy;
52 bool iowait_boost_pending;
53 unsigned int iowait_boost;
60 /* The field below is for single-CPU policies only: */
61 #ifdef CONFIG_NO_HZ_COMMON
62 unsigned long saved_idle_calls;
66 static DEFINE_PER_CPU(struct sugov_cpu, sugov_cpu);
68 /************************ Governor internals ***********************/
70 static bool sugov_should_update_freq(struct sugov_policy *sg_policy, u64 time)
75 * Since cpufreq_update_util() is called with rq->lock held for
76 * the @target_cpu, our per-CPU data is fully serialized.
78 * However, drivers cannot in general deal with cross-CPU
79 * requests, so while get_next_freq() will work, our
80 * sugov_update_commit() call may not for the fast switching platforms.
82 * Hence stop here for remote requests if they aren't supported
83 * by the hardware, as calculating the frequency is pointless if
84 * we cannot in fact act on it.
86 * This is needed on the slow switching platforms too to prevent CPUs
87 * going offline from leaving stale IRQ work items behind.
89 if (!cpufreq_this_cpu_can_update(sg_policy->policy))
92 if (unlikely(sg_policy->limits_changed)) {
93 sg_policy->limits_changed = false;
94 sg_policy->need_freq_update = true;
98 delta_ns = time - sg_policy->last_freq_update_time;
100 return delta_ns >= sg_policy->freq_update_delay_ns;
103 static bool sugov_update_next_freq(struct sugov_policy *sg_policy, u64 time,
104 unsigned int next_freq)
106 if (sg_policy->need_freq_update)
107 sg_policy->need_freq_update = cpufreq_driver_test_flags(CPUFREQ_NEED_UPDATE_LIMITS);
108 else if (sg_policy->next_freq == next_freq)
111 sg_policy->next_freq = next_freq;
112 sg_policy->last_freq_update_time = time;
117 static void sugov_fast_switch(struct sugov_policy *sg_policy, u64 time,
118 unsigned int next_freq)
120 if (sugov_update_next_freq(sg_policy, time, next_freq))
121 cpufreq_driver_fast_switch(sg_policy->policy, next_freq);
124 static void sugov_deferred_update(struct sugov_policy *sg_policy, u64 time,
125 unsigned int next_freq)
127 if (!sugov_update_next_freq(sg_policy, time, next_freq))
130 if (!sg_policy->work_in_progress) {
131 sg_policy->work_in_progress = true;
132 irq_work_queue(&sg_policy->irq_work);
137 * get_next_freq - Compute a new frequency for a given cpufreq policy.
138 * @sg_policy: schedutil policy object to compute the new frequency for.
139 * @util: Current CPU utilization.
140 * @max: CPU capacity.
142 * If the utilization is frequency-invariant, choose the new frequency to be
143 * proportional to it, that is
145 * next_freq = C * max_freq * util / max
147 * Otherwise, approximate the would-be frequency-invariant utilization by
148 * util_raw * (curr_freq / max_freq) which leads to
150 * next_freq = C * curr_freq * util_raw / max
152 * Take C = 1.25 for the frequency tipping point at (util / max) = 0.8.
154 * The lowest driver-supported frequency which is equal or greater than the raw
155 * next_freq (as calculated above) is returned, subject to policy min/max and
156 * cpufreq driver limitations.
158 static unsigned int get_next_freq(struct sugov_policy *sg_policy,
159 unsigned long util, unsigned long max)
161 struct cpufreq_policy *policy = sg_policy->policy;
162 unsigned int freq = arch_scale_freq_invariant() ?
163 policy->cpuinfo.max_freq : policy->cur;
165 freq = map_util_freq(util, freq, max);
167 if (freq == sg_policy->cached_raw_freq && !sg_policy->need_freq_update)
168 return sg_policy->next_freq;
170 sg_policy->cached_raw_freq = freq;
171 return cpufreq_driver_resolve_freq(policy, freq);
174 static void sugov_get_util(struct sugov_cpu *sg_cpu)
176 struct rq *rq = cpu_rq(sg_cpu->cpu);
177 unsigned long max = arch_scale_cpu_capacity(sg_cpu->cpu);
180 sg_cpu->bw_dl = cpu_bw_dl(rq);
181 sg_cpu->util = effective_cpu_util(sg_cpu->cpu, cpu_util_cfs(rq), max,
182 FREQUENCY_UTIL, NULL);
186 * sugov_iowait_reset() - Reset the IO boost status of a CPU.
187 * @sg_cpu: the sugov data for the CPU to boost
188 * @time: the update time from the caller
189 * @set_iowait_boost: true if an IO boost has been requested
191 * The IO wait boost of a task is disabled after a tick since the last update
192 * of a CPU. If a new IO wait boost is requested after more then a tick, then
193 * we enable the boost starting from IOWAIT_BOOST_MIN, which improves energy
194 * efficiency by ignoring sporadic wakeups from IO.
196 static bool sugov_iowait_reset(struct sugov_cpu *sg_cpu, u64 time,
197 bool set_iowait_boost)
199 s64 delta_ns = time - sg_cpu->last_update;
201 /* Reset boost only if a tick has elapsed since last request */
202 if (delta_ns <= TICK_NSEC)
205 sg_cpu->iowait_boost = set_iowait_boost ? IOWAIT_BOOST_MIN : 0;
206 sg_cpu->iowait_boost_pending = set_iowait_boost;
212 * sugov_iowait_boost() - Updates the IO boost status of a CPU.
213 * @sg_cpu: the sugov data for the CPU to boost
214 * @time: the update time from the caller
215 * @flags: SCHED_CPUFREQ_IOWAIT if the task is waking up after an IO wait
217 * Each time a task wakes up after an IO operation, the CPU utilization can be
218 * boosted to a certain utilization which doubles at each "frequent and
219 * successive" wakeup from IO, ranging from IOWAIT_BOOST_MIN to the utilization
220 * of the maximum OPP.
222 * To keep doubling, an IO boost has to be requested at least once per tick,
223 * otherwise we restart from the utilization of the minimum OPP.
225 static void sugov_iowait_boost(struct sugov_cpu *sg_cpu, u64 time,
228 bool set_iowait_boost = flags & SCHED_CPUFREQ_IOWAIT;
230 /* Reset boost if the CPU appears to have been idle enough */
231 if (sg_cpu->iowait_boost &&
232 sugov_iowait_reset(sg_cpu, time, set_iowait_boost))
235 /* Boost only tasks waking up after IO */
236 if (!set_iowait_boost)
239 /* Ensure boost doubles only one time at each request */
240 if (sg_cpu->iowait_boost_pending)
242 sg_cpu->iowait_boost_pending = true;
244 /* Double the boost at each request */
245 if (sg_cpu->iowait_boost) {
246 sg_cpu->iowait_boost =
247 min_t(unsigned int, sg_cpu->iowait_boost << 1, SCHED_CAPACITY_SCALE);
251 /* First wakeup after IO: start with minimum boost */
252 sg_cpu->iowait_boost = IOWAIT_BOOST_MIN;
256 * sugov_iowait_apply() - Apply the IO boost to a CPU.
257 * @sg_cpu: the sugov data for the cpu to boost
258 * @time: the update time from the caller
260 * A CPU running a task which woken up after an IO operation can have its
261 * utilization boosted to speed up the completion of those IO operations.
262 * The IO boost value is increased each time a task wakes up from IO, in
263 * sugov_iowait_apply(), and it's instead decreased by this function,
264 * each time an increase has not been requested (!iowait_boost_pending).
266 * A CPU which also appears to have been idle for at least one tick has also
267 * its IO boost utilization reset.
269 * This mechanism is designed to boost high frequently IO waiting tasks, while
270 * being more conservative on tasks which does sporadic IO operations.
272 static void sugov_iowait_apply(struct sugov_cpu *sg_cpu, u64 time)
276 /* No boost currently required */
277 if (!sg_cpu->iowait_boost)
280 /* Reset boost if the CPU appears to have been idle enough */
281 if (sugov_iowait_reset(sg_cpu, time, false))
284 if (!sg_cpu->iowait_boost_pending) {
286 * No boost pending; reduce the boost value.
288 sg_cpu->iowait_boost >>= 1;
289 if (sg_cpu->iowait_boost < IOWAIT_BOOST_MIN) {
290 sg_cpu->iowait_boost = 0;
295 sg_cpu->iowait_boost_pending = false;
298 * sg_cpu->util is already in capacity scale; convert iowait_boost
299 * into the same scale so we can compare.
301 boost = (sg_cpu->iowait_boost * sg_cpu->max) >> SCHED_CAPACITY_SHIFT;
302 if (sg_cpu->util < boost)
303 sg_cpu->util = boost;
306 #ifdef CONFIG_NO_HZ_COMMON
307 static bool sugov_cpu_is_busy(struct sugov_cpu *sg_cpu)
309 unsigned long idle_calls = tick_nohz_get_idle_calls_cpu(sg_cpu->cpu);
310 bool ret = idle_calls == sg_cpu->saved_idle_calls;
312 sg_cpu->saved_idle_calls = idle_calls;
316 static inline bool sugov_cpu_is_busy(struct sugov_cpu *sg_cpu) { return false; }
317 #endif /* CONFIG_NO_HZ_COMMON */
320 * Make sugov_should_update_freq() ignore the rate limit when DL
321 * has increased the utilization.
323 static inline void ignore_dl_rate_limit(struct sugov_cpu *sg_cpu, struct sugov_policy *sg_policy)
325 if (cpu_bw_dl(cpu_rq(sg_cpu->cpu)) > sg_cpu->bw_dl)
326 sg_policy->limits_changed = true;
329 static inline bool sugov_update_single_common(struct sugov_cpu *sg_cpu,
330 u64 time, unsigned int flags)
332 struct sugov_policy *sg_policy = sg_cpu->sg_policy;
334 sugov_iowait_boost(sg_cpu, time, flags);
335 sg_cpu->last_update = time;
337 ignore_dl_rate_limit(sg_cpu, sg_policy);
339 if (!sugov_should_update_freq(sg_policy, time))
342 sugov_get_util(sg_cpu);
343 sugov_iowait_apply(sg_cpu, time);
348 static void sugov_update_single_freq(struct update_util_data *hook, u64 time,
351 struct sugov_cpu *sg_cpu = container_of(hook, struct sugov_cpu, update_util);
352 struct sugov_policy *sg_policy = sg_cpu->sg_policy;
353 unsigned int cached_freq = sg_policy->cached_raw_freq;
356 if (!sugov_update_single_common(sg_cpu, time, flags))
359 next_f = get_next_freq(sg_policy, sg_cpu->util, sg_cpu->max);
361 * Do not reduce the frequency if the CPU has not been idle
362 * recently, as the reduction is likely to be premature then.
364 if (sugov_cpu_is_busy(sg_cpu) && next_f < sg_policy->next_freq) {
365 next_f = sg_policy->next_freq;
367 /* Restore cached freq as next_freq has changed */
368 sg_policy->cached_raw_freq = cached_freq;
372 * This code runs under rq->lock for the target CPU, so it won't run
373 * concurrently on two different CPUs for the same target and it is not
374 * necessary to acquire the lock in the fast switch case.
376 if (sg_policy->policy->fast_switch_enabled) {
377 sugov_fast_switch(sg_policy, time, next_f);
379 raw_spin_lock(&sg_policy->update_lock);
380 sugov_deferred_update(sg_policy, time, next_f);
381 raw_spin_unlock(&sg_policy->update_lock);
385 static void sugov_update_single_perf(struct update_util_data *hook, u64 time,
388 struct sugov_cpu *sg_cpu = container_of(hook, struct sugov_cpu, update_util);
389 unsigned long prev_util = sg_cpu->util;
392 * Fall back to the "frequency" path if frequency invariance is not
393 * supported, because the direct mapping between the utilization and
394 * the performance levels depends on the frequency invariance.
396 if (!arch_scale_freq_invariant()) {
397 sugov_update_single_freq(hook, time, flags);
401 if (!sugov_update_single_common(sg_cpu, time, flags))
405 * Do not reduce the target performance level if the CPU has not been
406 * idle recently, as the reduction is likely to be premature then.
408 if (sugov_cpu_is_busy(sg_cpu) && sg_cpu->util < prev_util)
409 sg_cpu->util = prev_util;
411 cpufreq_driver_adjust_perf(sg_cpu->cpu, map_util_perf(sg_cpu->bw_dl),
412 map_util_perf(sg_cpu->util), sg_cpu->max);
414 sg_cpu->sg_policy->last_freq_update_time = time;
417 static unsigned int sugov_next_freq_shared(struct sugov_cpu *sg_cpu, u64 time)
419 struct sugov_policy *sg_policy = sg_cpu->sg_policy;
420 struct cpufreq_policy *policy = sg_policy->policy;
421 unsigned long util = 0, max = 1;
424 for_each_cpu(j, policy->cpus) {
425 struct sugov_cpu *j_sg_cpu = &per_cpu(sugov_cpu, j);
426 unsigned long j_util, j_max;
428 sugov_get_util(j_sg_cpu);
429 sugov_iowait_apply(j_sg_cpu, time);
430 j_util = j_sg_cpu->util;
431 j_max = j_sg_cpu->max;
433 if (j_util * max > j_max * util) {
439 return get_next_freq(sg_policy, util, max);
443 sugov_update_shared(struct update_util_data *hook, u64 time, unsigned int flags)
445 struct sugov_cpu *sg_cpu = container_of(hook, struct sugov_cpu, update_util);
446 struct sugov_policy *sg_policy = sg_cpu->sg_policy;
449 raw_spin_lock(&sg_policy->update_lock);
451 sugov_iowait_boost(sg_cpu, time, flags);
452 sg_cpu->last_update = time;
454 ignore_dl_rate_limit(sg_cpu, sg_policy);
456 if (sugov_should_update_freq(sg_policy, time)) {
457 next_f = sugov_next_freq_shared(sg_cpu, time);
459 if (sg_policy->policy->fast_switch_enabled)
460 sugov_fast_switch(sg_policy, time, next_f);
462 sugov_deferred_update(sg_policy, time, next_f);
465 raw_spin_unlock(&sg_policy->update_lock);
468 static void sugov_work(struct kthread_work *work)
470 struct sugov_policy *sg_policy = container_of(work, struct sugov_policy, work);
475 * Hold sg_policy->update_lock shortly to handle the case where:
476 * incase sg_policy->next_freq is read here, and then updated by
477 * sugov_deferred_update() just before work_in_progress is set to false
478 * here, we may miss queueing the new update.
480 * Note: If a work was queued after the update_lock is released,
481 * sugov_work() will just be called again by kthread_work code; and the
482 * request will be proceed before the sugov thread sleeps.
484 raw_spin_lock_irqsave(&sg_policy->update_lock, flags);
485 freq = sg_policy->next_freq;
486 sg_policy->work_in_progress = false;
487 raw_spin_unlock_irqrestore(&sg_policy->update_lock, flags);
489 mutex_lock(&sg_policy->work_lock);
490 __cpufreq_driver_target(sg_policy->policy, freq, CPUFREQ_RELATION_L);
491 mutex_unlock(&sg_policy->work_lock);
494 static void sugov_irq_work(struct irq_work *irq_work)
496 struct sugov_policy *sg_policy;
498 sg_policy = container_of(irq_work, struct sugov_policy, irq_work);
500 kthread_queue_work(&sg_policy->worker, &sg_policy->work);
503 /************************** sysfs interface ************************/
505 static struct sugov_tunables *global_tunables;
506 static DEFINE_MUTEX(global_tunables_lock);
508 static inline struct sugov_tunables *to_sugov_tunables(struct gov_attr_set *attr_set)
510 return container_of(attr_set, struct sugov_tunables, attr_set);
513 static ssize_t rate_limit_us_show(struct gov_attr_set *attr_set, char *buf)
515 struct sugov_tunables *tunables = to_sugov_tunables(attr_set);
517 return sprintf(buf, "%u\n", tunables->rate_limit_us);
521 rate_limit_us_store(struct gov_attr_set *attr_set, const char *buf, size_t count)
523 struct sugov_tunables *tunables = to_sugov_tunables(attr_set);
524 struct sugov_policy *sg_policy;
525 unsigned int rate_limit_us;
527 if (kstrtouint(buf, 10, &rate_limit_us))
530 tunables->rate_limit_us = rate_limit_us;
532 list_for_each_entry(sg_policy, &attr_set->policy_list, tunables_hook)
533 sg_policy->freq_update_delay_ns = rate_limit_us * NSEC_PER_USEC;
538 static struct governor_attr rate_limit_us = __ATTR_RW(rate_limit_us);
540 static struct attribute *sugov_attrs[] = {
544 ATTRIBUTE_GROUPS(sugov);
546 static struct kobj_type sugov_tunables_ktype = {
547 .default_groups = sugov_groups,
548 .sysfs_ops = &governor_sysfs_ops,
551 /********************** cpufreq governor interface *********************/
553 struct cpufreq_governor schedutil_gov;
555 static struct sugov_policy *sugov_policy_alloc(struct cpufreq_policy *policy)
557 struct sugov_policy *sg_policy;
559 sg_policy = kzalloc(sizeof(*sg_policy), GFP_KERNEL);
563 sg_policy->policy = policy;
564 raw_spin_lock_init(&sg_policy->update_lock);
568 static void sugov_policy_free(struct sugov_policy *sg_policy)
573 static int sugov_kthread_create(struct sugov_policy *sg_policy)
575 struct task_struct *thread;
576 struct sched_attr attr = {
577 .size = sizeof(struct sched_attr),
578 .sched_policy = SCHED_DEADLINE,
579 .sched_flags = SCHED_FLAG_SUGOV,
583 * Fake (unused) bandwidth; workaround to "fix"
584 * priority inheritance.
586 .sched_runtime = 1000000,
587 .sched_deadline = 10000000,
588 .sched_period = 10000000,
590 struct cpufreq_policy *policy = sg_policy->policy;
593 /* kthread only required for slow path */
594 if (policy->fast_switch_enabled)
597 kthread_init_work(&sg_policy->work, sugov_work);
598 kthread_init_worker(&sg_policy->worker);
599 thread = kthread_create(kthread_worker_fn, &sg_policy->worker,
601 cpumask_first(policy->related_cpus));
602 if (IS_ERR(thread)) {
603 pr_err("failed to create sugov thread: %ld\n", PTR_ERR(thread));
604 return PTR_ERR(thread);
607 ret = sched_setattr_nocheck(thread, &attr);
609 kthread_stop(thread);
610 pr_warn("%s: failed to set SCHED_DEADLINE\n", __func__);
614 sg_policy->thread = thread;
615 kthread_bind_mask(thread, policy->related_cpus);
616 init_irq_work(&sg_policy->irq_work, sugov_irq_work);
617 mutex_init(&sg_policy->work_lock);
619 wake_up_process(thread);
624 static void sugov_kthread_stop(struct sugov_policy *sg_policy)
626 /* kthread only required for slow path */
627 if (sg_policy->policy->fast_switch_enabled)
630 kthread_flush_worker(&sg_policy->worker);
631 kthread_stop(sg_policy->thread);
632 mutex_destroy(&sg_policy->work_lock);
635 static struct sugov_tunables *sugov_tunables_alloc(struct sugov_policy *sg_policy)
637 struct sugov_tunables *tunables;
639 tunables = kzalloc(sizeof(*tunables), GFP_KERNEL);
641 gov_attr_set_init(&tunables->attr_set, &sg_policy->tunables_hook);
642 if (!have_governor_per_policy())
643 global_tunables = tunables;
648 static void sugov_tunables_free(struct sugov_tunables *tunables)
650 if (!have_governor_per_policy())
651 global_tunables = NULL;
656 static int sugov_init(struct cpufreq_policy *policy)
658 struct sugov_policy *sg_policy;
659 struct sugov_tunables *tunables;
662 /* State should be equivalent to EXIT */
663 if (policy->governor_data)
666 cpufreq_enable_fast_switch(policy);
668 sg_policy = sugov_policy_alloc(policy);
671 goto disable_fast_switch;
674 ret = sugov_kthread_create(sg_policy);
678 mutex_lock(&global_tunables_lock);
680 if (global_tunables) {
681 if (WARN_ON(have_governor_per_policy())) {
685 policy->governor_data = sg_policy;
686 sg_policy->tunables = global_tunables;
688 gov_attr_set_get(&global_tunables->attr_set, &sg_policy->tunables_hook);
692 tunables = sugov_tunables_alloc(sg_policy);
698 tunables->rate_limit_us = cpufreq_policy_transition_delay_us(policy);
700 policy->governor_data = sg_policy;
701 sg_policy->tunables = tunables;
703 ret = kobject_init_and_add(&tunables->attr_set.kobj, &sugov_tunables_ktype,
704 get_governor_parent_kobj(policy), "%s",
710 mutex_unlock(&global_tunables_lock);
714 kobject_put(&tunables->attr_set.kobj);
715 policy->governor_data = NULL;
716 sugov_tunables_free(tunables);
719 sugov_kthread_stop(sg_policy);
720 mutex_unlock(&global_tunables_lock);
723 sugov_policy_free(sg_policy);
726 cpufreq_disable_fast_switch(policy);
728 pr_err("initialization failed (error %d)\n", ret);
732 static void sugov_exit(struct cpufreq_policy *policy)
734 struct sugov_policy *sg_policy = policy->governor_data;
735 struct sugov_tunables *tunables = sg_policy->tunables;
738 mutex_lock(&global_tunables_lock);
740 count = gov_attr_set_put(&tunables->attr_set, &sg_policy->tunables_hook);
741 policy->governor_data = NULL;
743 sugov_tunables_free(tunables);
745 mutex_unlock(&global_tunables_lock);
747 sugov_kthread_stop(sg_policy);
748 sugov_policy_free(sg_policy);
749 cpufreq_disable_fast_switch(policy);
752 static int sugov_start(struct cpufreq_policy *policy)
754 struct sugov_policy *sg_policy = policy->governor_data;
755 void (*uu)(struct update_util_data *data, u64 time, unsigned int flags);
758 sg_policy->freq_update_delay_ns = sg_policy->tunables->rate_limit_us * NSEC_PER_USEC;
759 sg_policy->last_freq_update_time = 0;
760 sg_policy->next_freq = 0;
761 sg_policy->work_in_progress = false;
762 sg_policy->limits_changed = false;
763 sg_policy->cached_raw_freq = 0;
765 sg_policy->need_freq_update = cpufreq_driver_test_flags(CPUFREQ_NEED_UPDATE_LIMITS);
767 for_each_cpu(cpu, policy->cpus) {
768 struct sugov_cpu *sg_cpu = &per_cpu(sugov_cpu, cpu);
770 memset(sg_cpu, 0, sizeof(*sg_cpu));
772 sg_cpu->sg_policy = sg_policy;
775 if (policy_is_shared(policy))
776 uu = sugov_update_shared;
777 else if (policy->fast_switch_enabled && cpufreq_driver_has_adjust_perf())
778 uu = sugov_update_single_perf;
780 uu = sugov_update_single_freq;
782 for_each_cpu(cpu, policy->cpus) {
783 struct sugov_cpu *sg_cpu = &per_cpu(sugov_cpu, cpu);
785 cpufreq_add_update_util_hook(cpu, &sg_cpu->update_util, uu);
790 static void sugov_stop(struct cpufreq_policy *policy)
792 struct sugov_policy *sg_policy = policy->governor_data;
795 for_each_cpu(cpu, policy->cpus)
796 cpufreq_remove_update_util_hook(cpu);
800 if (!policy->fast_switch_enabled) {
801 irq_work_sync(&sg_policy->irq_work);
802 kthread_cancel_work_sync(&sg_policy->work);
806 static void sugov_limits(struct cpufreq_policy *policy)
808 struct sugov_policy *sg_policy = policy->governor_data;
810 if (!policy->fast_switch_enabled) {
811 mutex_lock(&sg_policy->work_lock);
812 cpufreq_policy_apply_limits(policy);
813 mutex_unlock(&sg_policy->work_lock);
816 sg_policy->limits_changed = true;
819 struct cpufreq_governor schedutil_gov = {
821 .owner = THIS_MODULE,
822 .flags = CPUFREQ_GOV_DYNAMIC_SWITCHING,
825 .start = sugov_start,
827 .limits = sugov_limits,
830 #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_SCHEDUTIL
831 struct cpufreq_governor *cpufreq_default_governor(void)
833 return &schedutil_gov;
837 cpufreq_governor_init(schedutil_gov);
839 #ifdef CONFIG_ENERGY_MODEL
840 static void rebuild_sd_workfn(struct work_struct *work)
842 rebuild_sched_domains_energy();
844 static DECLARE_WORK(rebuild_sd_work, rebuild_sd_workfn);
847 * EAS shouldn't be attempted without sugov, so rebuild the sched_domains
848 * on governor changes to make sure the scheduler knows about it.
850 void sched_cpufreq_governor_change(struct cpufreq_policy *policy,
851 struct cpufreq_governor *old_gov)
853 if (old_gov == &schedutil_gov || policy->governor == &schedutil_gov) {
855 * When called from the cpufreq_register_driver() path, the
856 * cpu_hotplug_lock is already held, so use a work item to
857 * avoid nested locking in rebuild_sched_domains().
859 schedule_work(&rebuild_sd_work);