/* * drivers/cpufreq/cpufreq_ondemand.c * * Copyright (C) 2001 Russell King * (C) 2003 Venkatesh Pallipadi . * Jun Nakajima * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as * published by the Free Software Foundation. */ #include #include #include #include #include #include #include #include #include #include #include #include #include /* * dbs is used in this file as a shortform for demandbased switching * It helps to keep variable names smaller, simpler */ #define DEF_FREQUENCY_DOWN_DIFFERENTIAL (10) #define DEF_FREQUENCY_UP_THRESHOLD (80) #define DEF_SAMPLING_DOWN_FACTOR (1) #define MAX_SAMPLING_DOWN_FACTOR (100000) #if defined(CONFIG_MACH_SLP_PQ) #define MICRO_FREQUENCY_DOWN_DIFFERENTIAL (5) #define MICRO_FREQUENCY_UP_THRESHOLD (85) #else #define MICRO_FREQUENCY_DOWN_DIFFERENTIAL (3) #define MICRO_FREQUENCY_UP_THRESHOLD (95) #endif #define MICRO_FREQUENCY_MIN_SAMPLE_RATE (10000) #define MIN_FREQUENCY_UP_THRESHOLD (11) #define MAX_FREQUENCY_UP_THRESHOLD (100) /* * The polling frequency of this governor depends on the capability of * the processor. Default polling frequency is 1000 times the transition * latency of the processor. The governor will work on any processor with * transition latency <= 10mS, using appropriate sampling * rate. * For CPUs with transition latency > 10mS (mostly drivers with CPUFREQ_ETERNAL) * this governor will not work. * All times here are in uS. */ #define MIN_SAMPLING_RATE_RATIO (2) static unsigned int min_sampling_rate; #define LATENCY_MULTIPLIER (1000) #define MIN_LATENCY_MULTIPLIER (100) #define TRANSITION_LATENCY_LIMIT (10 * 1000 * 1000) static void do_dbs_timer(struct work_struct *work); static int cpufreq_governor_dbs(struct cpufreq_policy *policy, unsigned int event); #ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND static #endif struct cpufreq_governor cpufreq_gov_ondemand = { .name = "ondemand", .governor = cpufreq_governor_dbs, .max_transition_latency = TRANSITION_LATENCY_LIMIT, .owner = THIS_MODULE, }; /* Sampling types */ enum {DBS_NORMAL_SAMPLE, DBS_SUB_SAMPLE}; struct cpu_dbs_info_s { cputime64_t prev_cpu_idle; cputime64_t prev_cpu_iowait; cputime64_t prev_cpu_wall; cputime64_t prev_cpu_nice; struct cpufreq_policy *cur_policy; struct delayed_work work; struct cpufreq_frequency_table *freq_table; unsigned int freq_lo; unsigned int freq_lo_jiffies; unsigned int freq_hi_jiffies; unsigned int rate_mult; int cpu; unsigned int sample_type:1; /* * percpu mutex that serializes governor limit change with * do_dbs_timer invocation. We do not want do_dbs_timer to run * when user is changing the governor or limits. */ struct mutex timer_mutex; bool activated; /* dbs_timer_init is in effect */ #ifdef CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE unsigned int flex_duration; #endif }; static DEFINE_PER_CPU(struct cpu_dbs_info_s, od_cpu_dbs_info); static unsigned int dbs_enable; /* number of CPUs using this policy */ /* * dbs_mutex protects dbs_enable in governor start/stop. */ static DEFINE_MUTEX(dbs_mutex); #ifdef CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE static DEFINE_MUTEX(flex_mutex); #endif static struct dbs_tuners { unsigned int sampling_rate; unsigned int up_threshold; unsigned int down_differential; unsigned int ignore_nice; unsigned int sampling_down_factor; unsigned int powersave_bias; unsigned int io_is_busy; struct notifier_block dvfs_lat_qos_db; unsigned int dvfs_lat_qos_wants; unsigned int freq_step; #ifdef CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE unsigned int flex_sampling_rate; unsigned int flex_duration; #endif } dbs_tuners_ins = { .up_threshold = DEF_FREQUENCY_UP_THRESHOLD, .sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR, .down_differential = DEF_FREQUENCY_DOWN_DIFFERENTIAL, .ignore_nice = 0, .powersave_bias = 0, .freq_step = 100, }; static inline cputime64_t get_cpu_idle_time_jiffy(unsigned int cpu, cputime64_t *wall) { cputime64_t idle_time; cputime64_t cur_wall_time; cputime64_t busy_time; cur_wall_time = jiffies64_to_cputime64(get_jiffies_64()); busy_time = cputime64_add(kstat_cpu(cpu).cpustat.user, kstat_cpu(cpu).cpustat.system); busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.irq); busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.softirq); busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.steal); busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.nice); idle_time = cputime64_sub(cur_wall_time, busy_time); if (wall) *wall = (cputime64_t)jiffies_to_usecs(cur_wall_time); return (cputime64_t)jiffies_to_usecs(idle_time); } static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall) { u64 idle_time = get_cpu_idle_time_us(cpu, wall); if (idle_time == -1ULL) return get_cpu_idle_time_jiffy(cpu, wall); return idle_time; } static inline cputime64_t get_cpu_iowait_time(unsigned int cpu, cputime64_t *wall) { u64 iowait_time = get_cpu_iowait_time_us(cpu, wall); if (iowait_time == -1ULL) return 0; return iowait_time; } /* * Find right sampling rate based on sampling_rate and * QoS requests on dvfs latency. */ static unsigned int effective_sampling_rate(void) { unsigned int effective; if (dbs_tuners_ins.dvfs_lat_qos_wants) effective = min(dbs_tuners_ins.dvfs_lat_qos_wants, dbs_tuners_ins.sampling_rate); else effective = dbs_tuners_ins.sampling_rate; return max(effective, min_sampling_rate); } /* * Find right freq to be set now with powersave_bias on. * Returns the freq_hi to be used right now and will set freq_hi_jiffies, * freq_lo, and freq_lo_jiffies in percpu area for averaging freqs. */ static unsigned int powersave_bias_target(struct cpufreq_policy *policy, unsigned int freq_next, unsigned int relation) { unsigned int freq_req, freq_reduc, freq_avg; unsigned int freq_hi, freq_lo; unsigned int index = 0; unsigned int jiffies_total, jiffies_hi, jiffies_lo; struct cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, policy->cpu); if (!dbs_info->freq_table) { dbs_info->freq_lo = 0; dbs_info->freq_lo_jiffies = 0; return freq_next; } cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_next, relation, &index); freq_req = dbs_info->freq_table[index].frequency; freq_reduc = freq_req * dbs_tuners_ins.powersave_bias / 1000; freq_avg = freq_req - freq_reduc; /* Find freq bounds for freq_avg in freq_table */ index = 0; cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg, CPUFREQ_RELATION_H, &index); freq_lo = dbs_info->freq_table[index].frequency; index = 0; cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg, CPUFREQ_RELATION_L, &index); freq_hi = dbs_info->freq_table[index].frequency; /* Find out how long we have to be in hi and lo freqs */ if (freq_hi == freq_lo) { dbs_info->freq_lo = 0; dbs_info->freq_lo_jiffies = 0; return freq_lo; } jiffies_total = usecs_to_jiffies(effective_sampling_rate()); jiffies_hi = (freq_avg - freq_lo) * jiffies_total; jiffies_hi += ((freq_hi - freq_lo) / 2); jiffies_hi /= (freq_hi - freq_lo); jiffies_lo = jiffies_total - jiffies_hi; dbs_info->freq_lo = freq_lo; dbs_info->freq_lo_jiffies = jiffies_lo; dbs_info->freq_hi_jiffies = jiffies_hi; return freq_hi; } static void ondemand_powersave_bias_init_cpu(int cpu) { struct cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, cpu); dbs_info->freq_table = cpufreq_frequency_get_table(cpu); dbs_info->freq_lo = 0; } static void ondemand_powersave_bias_init(void) { int i; for_each_online_cpu(i) { ondemand_powersave_bias_init_cpu(i); } } /************************** sysfs interface ************************/ static ssize_t show_sampling_rate_min(struct kobject *kobj, struct attribute *attr, char *buf) { return sprintf(buf, "%u\n", min_sampling_rate); } define_one_global_ro(sampling_rate_min); /* cpufreq_ondemand Governor Tunables */ #define show_one(file_name, object) \ static ssize_t show_##file_name \ (struct kobject *kobj, struct attribute *attr, char *buf) \ { \ return sprintf(buf, "%u\n", dbs_tuners_ins.object); \ } show_one(sampling_rate, sampling_rate); show_one(io_is_busy, io_is_busy); show_one(up_threshold, up_threshold); show_one(sampling_down_factor, sampling_down_factor); show_one(ignore_nice_load, ignore_nice); show_one(powersave_bias, powersave_bias); show_one(down_differential, down_differential); show_one(freq_step, freq_step); /** * update_sampling_rate - update sampling rate effective immediately if needed. * @new_rate: new sampling rate. If it is 0, regard sampling rate is not * changed and assume that qos request value is changed. * * If new rate is smaller than the old, simply updaing * dbs_tuners_int.sampling_rate might not be appropriate. For example, * if the original sampling_rate was 1 second and the requested new sampling * rate is 10 ms because the user needs immediate reaction from ondemand * governor, but not sure if higher frequency will be required or not, * then, the governor may change the sampling rate too late; up to 1 second * later. Thus, if we are reducing the sampling rate, we need to make the * new value effective immediately. */ static void update_sampling_rate(unsigned int new_rate) { int cpu; unsigned int effective; if (new_rate) dbs_tuners_ins.sampling_rate = max(new_rate, min_sampling_rate); effective = effective_sampling_rate(); for_each_online_cpu(cpu) { struct cpufreq_policy *policy; struct cpu_dbs_info_s *dbs_info; unsigned long next_sampling, appointed_at; /* * mutex_destory(&dbs_info->timer_mutex) should not happen * in this context. dbs_mutex is locked/unlocked at GOV_START * and GOV_STOP context only other than here. */ mutex_lock(&dbs_mutex); policy = cpufreq_cpu_get(cpu); if (!policy) { mutex_unlock(&dbs_mutex); continue; } dbs_info = &per_cpu(od_cpu_dbs_info, policy->cpu); cpufreq_cpu_put(policy); /* timer_mutex is destroyed or will be destroyed soon */ if (!dbs_info->activated) { mutex_unlock(&dbs_mutex); continue; } mutex_lock(&dbs_info->timer_mutex); if (!delayed_work_pending(&dbs_info->work)) { mutex_unlock(&dbs_info->timer_mutex); mutex_unlock(&dbs_mutex); continue; } next_sampling = jiffies + usecs_to_jiffies(new_rate); appointed_at = dbs_info->work.timer.expires; if (time_before(next_sampling, appointed_at)) { mutex_unlock(&dbs_info->timer_mutex); cancel_delayed_work_sync(&dbs_info->work); mutex_lock(&dbs_info->timer_mutex); schedule_delayed_work_on(dbs_info->cpu, &dbs_info->work, usecs_to_jiffies(effective)); } mutex_unlock(&dbs_info->timer_mutex); /* * For the little possiblity that dbs_timer_exit() has been * called after checking dbs_info->activated above. * If cancel_delayed_work_syn() has been calld by * dbs_timer_exit() before schedule_delayed_work_on() of this * function, it should be revoked by calling cancel again * before releasing dbs_mutex, which will trigger mutex_destroy * to be called. */ if (!dbs_info->activated) cancel_delayed_work_sync(&dbs_info->work); mutex_unlock(&dbs_mutex); } } static ssize_t store_sampling_rate(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1) return -EINVAL; update_sampling_rate(input); return count; } static ssize_t store_io_is_busy(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1) return -EINVAL; dbs_tuners_ins.io_is_busy = !!input; return count; } static ssize_t store_up_threshold(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD || input < MIN_FREQUENCY_UP_THRESHOLD) { return -EINVAL; } dbs_tuners_ins.up_threshold = input; return count; } static ssize_t store_sampling_down_factor(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input, j; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1) return -EINVAL; dbs_tuners_ins.sampling_down_factor = input; /* Reset down sampling multiplier in case it was active */ for_each_online_cpu(j) { struct cpu_dbs_info_s *dbs_info; dbs_info = &per_cpu(od_cpu_dbs_info, j); dbs_info->rate_mult = 1; } return count; } static ssize_t store_ignore_nice_load(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; unsigned int j; ret = sscanf(buf, "%u", &input); if (ret != 1) return -EINVAL; if (input > 1) input = 1; if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */ return count; } dbs_tuners_ins.ignore_nice = input; /* we need to re-evaluate prev_cpu_idle */ for_each_online_cpu(j) { struct cpu_dbs_info_s *dbs_info; dbs_info = &per_cpu(od_cpu_dbs_info, j); dbs_info->prev_cpu_idle = get_cpu_idle_time(j, &dbs_info->prev_cpu_wall); if (dbs_tuners_ins.ignore_nice) dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; } return count; } static ssize_t store_powersave_bias(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1) return -EINVAL; if (input > 1000) input = 1000; dbs_tuners_ins.powersave_bias = input; ondemand_powersave_bias_init(); return count; } static ssize_t store_down_differential(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1) return -EINVAL; dbs_tuners_ins.down_differential = min(input, 100u); return count; } static ssize_t store_freq_step(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1) return -EINVAL; dbs_tuners_ins.freq_step = min(input, 100u); return count; } define_one_global_rw(sampling_rate); define_one_global_rw(io_is_busy); define_one_global_rw(up_threshold); define_one_global_rw(sampling_down_factor); define_one_global_rw(ignore_nice_load); define_one_global_rw(powersave_bias); define_one_global_rw(down_differential); define_one_global_rw(freq_step); #ifdef CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE static struct global_attr flexrate_request; static struct global_attr flexrate_duration; static struct global_attr flexrate_enable; static struct global_attr flexrate_forcerate; static struct global_attr flexrate_num_effective_usage; #endif static struct attribute *dbs_attributes[] = { &sampling_rate_min.attr, &sampling_rate.attr, &up_threshold.attr, &sampling_down_factor.attr, &ignore_nice_load.attr, &powersave_bias.attr, &io_is_busy.attr, &down_differential.attr, &freq_step.attr, #ifdef CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE &flexrate_request.attr, &flexrate_duration.attr, &flexrate_enable.attr, &flexrate_forcerate.attr, &flexrate_num_effective_usage.attr, #endif NULL }; static struct attribute_group dbs_attr_group = { .attrs = dbs_attributes, .name = "ondemand", }; /************************** sysfs end ************************/ static void dbs_freq_increase(struct cpufreq_policy *p, unsigned int freq) { if (dbs_tuners_ins.powersave_bias) freq = powersave_bias_target(p, freq, CPUFREQ_RELATION_H); #if !defined(CONFIG_ARCH_EXYNOS4) && !defined(CONFIG_ARCH_EXYNOS5) else if (p->cur == p->max) return; #endif __cpufreq_driver_target(p, freq, dbs_tuners_ins.powersave_bias ? CPUFREQ_RELATION_L : CPUFREQ_RELATION_H); } static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info) { unsigned int max_load_freq; struct cpufreq_policy *policy; unsigned int j; this_dbs_info->freq_lo = 0; policy = this_dbs_info->cur_policy; /* * Every sampling_rate, we check, if current idle time is less * than 20% (default), then we try to increase frequency * Every sampling_rate, we look for a the lowest * frequency which can sustain the load while keeping idle time over * 30%. If such a frequency exist, we try to decrease to this frequency. * * Any frequency increase takes it to the maximum frequency. * Frequency reduction happens at minimum steps of * 5% (default) of current frequency */ /* Get Absolute Load - in terms of freq */ max_load_freq = 0; for_each_cpu(j, policy->cpus) { struct cpu_dbs_info_s *j_dbs_info; cputime64_t cur_wall_time, cur_idle_time, cur_iowait_time; unsigned int idle_time, wall_time, iowait_time; unsigned int load, load_freq; int freq_avg; j_dbs_info = &per_cpu(od_cpu_dbs_info, j); cur_idle_time = get_cpu_idle_time(j, &cur_wall_time); cur_iowait_time = get_cpu_iowait_time(j, &cur_wall_time); wall_time = (unsigned int) cputime64_sub(cur_wall_time, j_dbs_info->prev_cpu_wall); j_dbs_info->prev_cpu_wall = cur_wall_time; idle_time = (unsigned int) cputime64_sub(cur_idle_time, j_dbs_info->prev_cpu_idle); j_dbs_info->prev_cpu_idle = cur_idle_time; iowait_time = (unsigned int) cputime64_sub(cur_iowait_time, j_dbs_info->prev_cpu_iowait); j_dbs_info->prev_cpu_iowait = cur_iowait_time; if (dbs_tuners_ins.ignore_nice) { cputime64_t cur_nice; unsigned long cur_nice_jiffies; cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice, j_dbs_info->prev_cpu_nice); /* * Assumption: nice time between sampling periods will * be less than 2^32 jiffies for 32 bit sys */ cur_nice_jiffies = (unsigned long) cputime64_to_jiffies64(cur_nice); j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; idle_time += jiffies_to_usecs(cur_nice_jiffies); } /* * For the purpose of ondemand, waiting for disk IO is an * indication that you're performance critical, and not that * the system is actually idle. So subtract the iowait time * from the cpu idle time. */ if (dbs_tuners_ins.io_is_busy && idle_time >= iowait_time) idle_time -= iowait_time; if (unlikely(!wall_time || wall_time < idle_time)) continue; load = 100 * (wall_time - idle_time) / wall_time; freq_avg = __cpufreq_driver_getavg(policy, j); if (freq_avg <= 0) freq_avg = policy->cur; load_freq = load * freq_avg; if (load_freq > max_load_freq) max_load_freq = load_freq; } /* Check for frequency increase */ if (max_load_freq > dbs_tuners_ins.up_threshold * policy->cur) { int inc = (policy->max * dbs_tuners_ins.freq_step) / 100; int target = min(policy->max, policy->cur + inc); /* If switching to max speed, apply sampling_down_factor */ if (policy->cur < policy->max && target == policy->max) this_dbs_info->rate_mult = dbs_tuners_ins.sampling_down_factor; dbs_freq_increase(policy, target); return; } /* Check for frequency decrease */ #if !defined(CONFIG_ARCH_EXYNOS4) && !defined(CONFIG_ARCH_EXYNOS5) /* if we cannot reduce the frequency anymore, break out early */ if (policy->cur == policy->min) return; #endif /* * The optimal frequency is the frequency that is the lowest that * can support the current CPU usage without triggering the up * policy. To be safe, we focus 10 points under the threshold. */ if (max_load_freq < (dbs_tuners_ins.up_threshold - dbs_tuners_ins.down_differential) * policy->cur) { unsigned int freq_next; freq_next = max_load_freq / (dbs_tuners_ins.up_threshold - dbs_tuners_ins.down_differential); /* No longer fully busy, reset rate_mult */ this_dbs_info->rate_mult = 1; if (freq_next < policy->min) freq_next = policy->min; if (!dbs_tuners_ins.powersave_bias) { __cpufreq_driver_target(policy, freq_next, CPUFREQ_RELATION_L); } else { int freq = powersave_bias_target(policy, freq_next, CPUFREQ_RELATION_L); __cpufreq_driver_target(policy, freq, CPUFREQ_RELATION_L); } } } static void do_dbs_timer(struct work_struct *work) { struct cpu_dbs_info_s *dbs_info = container_of(work, struct cpu_dbs_info_s, work.work); unsigned int cpu = dbs_info->cpu; int sample_type = dbs_info->sample_type; int delay; mutex_lock(&dbs_info->timer_mutex); /* Common NORMAL_SAMPLE setup */ dbs_info->sample_type = DBS_NORMAL_SAMPLE; if (!dbs_tuners_ins.powersave_bias || sample_type == DBS_NORMAL_SAMPLE) { dbs_check_cpu(dbs_info); if (dbs_info->freq_lo) { /* Setup timer for SUB_SAMPLE */ dbs_info->sample_type = DBS_SUB_SAMPLE; delay = dbs_info->freq_hi_jiffies; } else { /* We want all CPUs to do sampling nearly on * same jiffy */ delay = usecs_to_jiffies(effective_sampling_rate() * dbs_info->rate_mult); if (num_online_cpus() > 1) delay -= jiffies % delay; } } else { __cpufreq_driver_target(dbs_info->cur_policy, dbs_info->freq_lo, CPUFREQ_RELATION_H); delay = dbs_info->freq_lo_jiffies; } #ifdef CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE if (dbs_info->flex_duration) { struct cpufreq_policy *policy = dbs_info->cur_policy; mutex_lock(&flex_mutex); delay = usecs_to_jiffies(dbs_tuners_ins.flex_sampling_rate); /* If it's already max, we don't need to iterate fast */ if (policy->cur >= policy->max) dbs_info->flex_duration = 1; if (--dbs_info->flex_duration < dbs_tuners_ins.flex_duration) { dbs_tuners_ins.flex_duration = dbs_info->flex_duration; } mutex_unlock(&flex_mutex); } #endif /* CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE */ schedule_delayed_work_on(cpu, &dbs_info->work, delay); mutex_unlock(&dbs_info->timer_mutex); } static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info) { /* We want all CPUs to do sampling nearly on same jiffy */ int delay = usecs_to_jiffies(effective_sampling_rate()); if (num_online_cpus() > 1) delay -= jiffies % delay; dbs_info->sample_type = DBS_NORMAL_SAMPLE; INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer); schedule_delayed_work_on(dbs_info->cpu, &dbs_info->work, 10 * delay); dbs_info->activated = true; } static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info) { dbs_info->activated = false; cancel_delayed_work_sync(&dbs_info->work); } /* * Not all CPUs want IO time to be accounted as busy; this dependson how * efficient idling at a higher frequency/voltage is. * Pavel Machek says this is not so for various generations of AMD and old * Intel systems. * Mike Chan (androidlcom) calis this is also not true for ARM. * Because of this, whitelist specific known (series) of CPUs by default, and * leave all others up to the user. */ static int should_io_be_busy(void) { #if defined(CONFIG_X86) /* * For Intel, Core 2 (model 15) andl later have an efficient idle. */ if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model >= 15) return 1; #endif return 0; } static int cpufreq_governor_dbs(struct cpufreq_policy *policy, unsigned int event) { unsigned int cpu = policy->cpu; struct cpu_dbs_info_s *this_dbs_info; unsigned int j; int rc; this_dbs_info = &per_cpu(od_cpu_dbs_info, cpu); switch (event) { case CPUFREQ_GOV_START: if ((!cpu_online(cpu)) || (!policy->cur)) return -EINVAL; mutex_lock(&dbs_mutex); dbs_enable++; for_each_cpu(j, policy->cpus) { struct cpu_dbs_info_s *j_dbs_info; j_dbs_info = &per_cpu(od_cpu_dbs_info, j); j_dbs_info->cur_policy = policy; j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j, &j_dbs_info->prev_cpu_wall); if (dbs_tuners_ins.ignore_nice) { j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; } } this_dbs_info->cpu = cpu; this_dbs_info->rate_mult = 1; ondemand_powersave_bias_init_cpu(cpu); /* * Start the timerschedule work, when this governor * is used for first time */ if (dbs_enable == 1) { unsigned int latency; rc = sysfs_create_group(cpufreq_global_kobject, &dbs_attr_group); if (rc) { mutex_unlock(&dbs_mutex); return rc; } /* policy latency is in nS. Convert it to uS first */ latency = policy->cpuinfo.transition_latency / 1000; if (latency == 0) latency = 1; /* Bring kernel and HW constraints together */ min_sampling_rate = max(min_sampling_rate, MIN_LATENCY_MULTIPLIER * latency); dbs_tuners_ins.sampling_rate = max(min_sampling_rate, latency * LATENCY_MULTIPLIER); dbs_tuners_ins.io_is_busy = should_io_be_busy(); } mutex_unlock(&dbs_mutex); mutex_init(&this_dbs_info->timer_mutex); dbs_timer_init(this_dbs_info); break; case CPUFREQ_GOV_STOP: dbs_timer_exit(this_dbs_info); mutex_lock(&dbs_mutex); mutex_destroy(&this_dbs_info->timer_mutex); dbs_enable--; mutex_unlock(&dbs_mutex); if (!dbs_enable) sysfs_remove_group(cpufreq_global_kobject, &dbs_attr_group); break; case CPUFREQ_GOV_LIMITS: mutex_lock(&this_dbs_info->timer_mutex); if (policy->max < this_dbs_info->cur_policy->cur) __cpufreq_driver_target(this_dbs_info->cur_policy, policy->max, CPUFREQ_RELATION_H); else if (policy->min > this_dbs_info->cur_policy->cur) __cpufreq_driver_target(this_dbs_info->cur_policy, policy->min, CPUFREQ_RELATION_L); mutex_unlock(&this_dbs_info->timer_mutex); break; } return 0; } /** * qos_dvfs_lat_notify - PM QoS Notifier for DVFS_LATENCY QoS Request * @nb notifier block struct * @value QoS value * @dummy */ static int qos_dvfs_lat_notify(struct notifier_block *nb, unsigned long value, void *dummy) { /* * In the worst case, with a continuous up-treshold + e cpu load * from up-threshold - e load, the ondemand governor will react * sampling_rate * 2. * * Thus, based on the worst case scenario, we use value / 2; */ dbs_tuners_ins.dvfs_lat_qos_wants = value / 2; /* Update sampling rate */ update_sampling_rate(0); return NOTIFY_OK; } static struct notifier_block ondemand_qos_dvfs_lat_nb = { .notifier_call = qos_dvfs_lat_notify, }; static int __init cpufreq_gov_dbs_init(void) { cputime64_t wall; u64 idle_time; int cpu = get_cpu(); int err = 0; idle_time = get_cpu_idle_time_us(cpu, &wall); put_cpu(); if (idle_time != -1ULL) { /* Idle micro accounting is supported. Use finer thresholds */ dbs_tuners_ins.up_threshold = MICRO_FREQUENCY_UP_THRESHOLD; dbs_tuners_ins.down_differential = MICRO_FREQUENCY_DOWN_DIFFERENTIAL; /* * In no_hz/micro accounting case we set the minimum frequency * not depending on HZ, but fixed (very low). The deferred * timer might skip some samples if idle/sleeping as needed. */ min_sampling_rate = MICRO_FREQUENCY_MIN_SAMPLE_RATE; } else { /* For correct statistics, we need 10 ticks for each measure */ min_sampling_rate = MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10); } err = pm_qos_add_notifier(PM_QOS_DVFS_RESPONSE_LATENCY, &ondemand_qos_dvfs_lat_nb); if (err) return err; err = cpufreq_register_governor(&cpufreq_gov_ondemand); if (err) { pm_qos_remove_notifier(PM_QOS_DVFS_RESPONSE_LATENCY, &ondemand_qos_dvfs_lat_nb); } return err; } static void __exit cpufreq_gov_dbs_exit(void) { pm_qos_remove_notifier(PM_QOS_DVFS_RESPONSE_LATENCY, &ondemand_qos_dvfs_lat_nb); cpufreq_unregister_governor(&cpufreq_gov_ondemand); } #ifdef CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE static unsigned int max_duration = (CONFIG_CPU_FREQ_GOV_ONDEMAND_FLEXRATE_MAX_DURATION); #define DEFAULT_DURATION (5) static unsigned int sysfs_duration = DEFAULT_DURATION; static bool flexrate_enabled = true; static unsigned int forced_rate; static unsigned int flexrate_num_effective; static int cpufreq_ondemand_flexrate_do(struct cpufreq_policy *policy, bool now) { unsigned int cpu = policy->cpu; bool using_ondemand; struct cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, cpu); WARN(!mutex_is_locked(&flex_mutex), "flex_mutex not locked\n"); dbs_info->flex_duration = dbs_tuners_ins.flex_duration; if (now) { flexrate_num_effective++; mutex_lock(&dbs_mutex); using_ondemand = dbs_enable && !strncmp(policy->governor->name, "ondemand", 8); mutex_unlock(&dbs_mutex); if (!using_ondemand) return 0; mutex_unlock(&flex_mutex); mutex_lock(&dbs_info->timer_mutex); /* Do It! */ cancel_delayed_work_sync(&dbs_info->work); schedule_delayed_work_on(cpu, &dbs_info->work, 1); mutex_unlock(&dbs_info->timer_mutex); mutex_lock(&flex_mutex); } return 0; } int cpufreq_ondemand_flexrate_request(unsigned int rate_us, unsigned int duration) { int err = 0; if (!flexrate_enabled) return 0; if (forced_rate) rate_us = forced_rate; mutex_lock(&flex_mutex); /* Unnecessary requests are dropped */ if (rate_us >= dbs_tuners_ins.sampling_rate) goto out; if (rate_us >= dbs_tuners_ins.flex_sampling_rate && duration <= dbs_tuners_ins.flex_duration) goto out; duration = min(max_duration, duration); if (rate_us > 0 && rate_us < min_sampling_rate) rate_us = min_sampling_rate; err = 1; /* Need update */ /* Cancel the active flexrate requests */ if (rate_us == 0 || duration == 0) { dbs_tuners_ins.flex_duration = 0; dbs_tuners_ins.flex_sampling_rate = 0; goto out; } if (dbs_tuners_ins.flex_sampling_rate == 0 || dbs_tuners_ins.flex_sampling_rate > rate_us) err = 2; /* Need to poll faster */ /* Set new flexrate per the request */ dbs_tuners_ins.flex_sampling_rate = min(dbs_tuners_ins.flex_sampling_rate, rate_us); dbs_tuners_ins.flex_duration = max(dbs_tuners_ins.flex_duration, duration); out: /* Apply new flexrate */ if (err > 0) { bool now = (err == 2); int cpu = 0; /* TODO: For every CPU using ONDEMAND */ err = cpufreq_ondemand_flexrate_do(cpufreq_cpu_get(cpu), now); } mutex_unlock(&flex_mutex); return err; } EXPORT_SYMBOL_GPL(cpufreq_ondemand_flexrate_request); static ssize_t store_flexrate_request(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int rate; int ret; ret = sscanf(buf, "%u", &rate); if (ret != 1) return -EINVAL; ret = cpufreq_ondemand_flexrate_request(rate, sysfs_duration); if (ret) return ret; return count; } static ssize_t show_flexrate_request(struct kobject *a, struct attribute *b, char *buf) { return sprintf(buf, "Flexrate decreases CPUFreq Ondemand governor's polling rate temporaily.\n" "Usage Example:\n" "# echo 8 > flexrate_duration\n" "# echo 10000 > flexrate_request\n" "With the second statement, Ondemand polls with 10ms(10000us) interval 8 times.\n" "run \"echo flexrate_duration\" to see the currecnt duration setting.\n"); } static ssize_t store_flexrate_duration(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int duration; int ret; /* mutex not needed for flexrate_sysfs_duration */ ret = sscanf(buf, "%u", &duration); if (ret != 1) return -EINVAL; if (duration == 0) duration = DEFAULT_DURATION; if (duration > max_duration) duration = max_duration; sysfs_duration = duration; return count; } static ssize_t show_flexrate_duration(struct kobject *a, struct attribute *b, char *buf) { return sprintf(buf, "%d\n", sysfs_duration); } static ssize_t store_flexrate_enable(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int input; int ret; ret = sscanf(buf, "%u", &input); if (ret != 1) return -EINVAL; if (input > 0) flexrate_enabled = true; else flexrate_enabled = false; return count; } static ssize_t show_flexrate_enable(struct kobject *a, struct attribute *b, char *buf) { return sprintf(buf, "%d\n", !!flexrate_enabled); } static ssize_t store_flexrate_forcerate(struct kobject *a, struct attribute *b, const char *buf, size_t count) { unsigned int rate; int ret; ret = sscanf(buf, "%u", &rate); if (ret != 1) return -EINVAL; forced_rate = rate; pr_info("CAUTION: flexrate_forcerate is for debugging/benchmarking only.\n"); return count; } static ssize_t show_flexrate_forcerate(struct kobject *a, struct attribute *b, char *buf) { return sprintf(buf, "%u\n", forced_rate); } static ssize_t show_flexrate_num_effective_usage(struct kobject *a, struct attribute *b, char *buf) { return sprintf(buf, "%u\n", flexrate_num_effective); } define_one_global_rw(flexrate_request); define_one_global_rw(flexrate_duration); define_one_global_rw(flexrate_enable); define_one_global_rw(flexrate_forcerate); define_one_global_ro(flexrate_num_effective_usage); #endif MODULE_AUTHOR("Venkatesh Pallipadi "); MODULE_AUTHOR("Alexey Starikovskiy "); MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for " "Low Latency Frequency Transition capable processors"); MODULE_LICENSE("GPL"); #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND fs_initcall(cpufreq_gov_dbs_init); #else module_init(cpufreq_gov_dbs_init); #endif module_exit(cpufreq_gov_dbs_exit);