From 94209cc2c3436f73fbf0f13527ab8dba220acf74 Mon Sep 17 00:00:00 2001 From: Sam Shih Date: Mon, 27 Mar 2023 17:34:38 +0800 Subject: [PATCH 1/5] pm_devfreq_add_cpu_based_scaling_support_to_passive_governor --- drivers/devfreq/devfreq.c | 20 +- drivers/devfreq/governor.h | 27 ++ drivers/devfreq/governor_passive.c | 403 ++++++++++++++++++++++++----- include/linux/devfreq.h | 17 +- 4 files changed, 389 insertions(+), 78 deletions(-) diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c index a473b64..c225234 100644 --- a/drivers/devfreq/devfreq.c +++ b/drivers/devfreq/devfreq.c @@ -112,16 +112,16 @@ static unsigned long find_available_max_freq(struct devfreq *devfreq) } /** - * get_freq_range() - Get the current freq range + * devfreq_get_freq_range() - Get the current freq range * @devfreq: the devfreq instance * @min_freq: the min frequency * @max_freq: the max frequency * * This takes into consideration all constraints. */ -static void get_freq_range(struct devfreq *devfreq, - unsigned long *min_freq, - unsigned long *max_freq) +void devfreq_get_freq_range(struct devfreq *devfreq, + unsigned long *min_freq, + unsigned long *max_freq) { unsigned long *freq_table = devfreq->profile->freq_table; s32 qos_min_freq, qos_max_freq; @@ -158,6 +158,7 @@ static void get_freq_range(struct devfreq *devfreq, if (*min_freq > *max_freq) *min_freq = *max_freq; } +EXPORT_SYMBOL(devfreq_get_freq_range); /** * devfreq_get_freq_level() - Lookup freq_table for the frequency @@ -418,7 +419,7 @@ int devfreq_update_target(struct devfreq *devfreq, unsigned long freq) err = devfreq->governor->get_target_freq(devfreq, &freq); if (err) return err; - get_freq_range(devfreq, &min_freq, &max_freq); + devfreq_get_freq_range(devfreq, &min_freq, &max_freq); if (freq < min_freq) { freq = min_freq; @@ -784,6 +785,7 @@ struct devfreq *devfreq_add_device(struct device *dev, { struct devfreq *devfreq; struct devfreq_governor *governor; + unsigned long min_freq, max_freq; int err = 0; if (!dev || !profile || !governor_name) { @@ -848,6 +850,8 @@ struct devfreq *devfreq_add_device(struct device *dev, goto err_dev; } + devfreq_get_freq_range(devfreq, &min_freq, &max_freq); + devfreq->suspend_freq = dev_pm_opp_get_suspend_opp_freq(dev); devfreq->opp_table = dev_pm_opp_get_opp_table(dev); if (IS_ERR(devfreq->opp_table)) @@ -1559,7 +1563,7 @@ static ssize_t min_freq_show(struct device *dev, struct device_attribute *attr, unsigned long min_freq, max_freq; mutex_lock(&df->lock); - get_freq_range(df, &min_freq, &max_freq); + devfreq_get_freq_range(df, &min_freq, &max_freq); mutex_unlock(&df->lock); return sprintf(buf, "%lu\n", min_freq); @@ -1613,7 +1617,7 @@ static ssize_t max_freq_show(struct device *dev, struct device_attribute *attr, unsigned long min_freq, max_freq; mutex_lock(&df->lock); - get_freq_range(df, &min_freq, &max_freq); + devfreq_get_freq_range(df, &min_freq, &max_freq); mutex_unlock(&df->lock); return sprintf(buf, "%lu\n", max_freq); @@ -1927,7 +1931,7 @@ static int devfreq_summary_show(struct seq_file *s, void *data) mutex_lock(&devfreq->lock); cur_freq = devfreq->previous_freq; - get_freq_range(devfreq, &min_freq, &max_freq); + devfreq_get_freq_range(devfreq, &min_freq, &max_freq); timer = devfreq->profile->timer; if (IS_SUPPORTED_ATTR(devfreq->governor->attrs, POLLING_INTERVAL)) diff --git a/drivers/devfreq/governor.h b/drivers/devfreq/governor.h index 2d69a0c..07008dd 100644 --- a/drivers/devfreq/governor.h +++ b/drivers/devfreq/governor.h @@ -47,6 +47,31 @@ #define DEVFREQ_GOV_ATTR_POLLING_INTERVAL BIT(0) #define DEVFREQ_GOV_ATTR_TIMER BIT(1) +/** + * struct devfreq_cpu_data - Hold the per-cpu data + * @node: list node + * @dev: reference to cpu device. + * @first_cpu: the cpumask of the first cpu of a policy. + * @opp_table: reference to cpu opp table. + * @cur_freq: the current frequency of the cpu. + * @min_freq: the min frequency of the cpu. + * @max_freq: the max frequency of the cpu. + * + * This structure stores the required cpu_data of a cpu. + * This is auto-populated by the governor. + */ +struct devfreq_cpu_data { + struct list_head node; + + struct device *dev; + unsigned int first_cpu; + + struct opp_table *opp_table; + unsigned int cur_freq; + unsigned int min_freq; + unsigned int max_freq; +}; + /** * struct devfreq_governor - Devfreq policy governor * @node: list node - contains registered devfreq governors @@ -86,6 +111,8 @@ int devfreq_remove_governor(struct devfreq_governor *governor); int devfreq_update_status(struct devfreq *devfreq, unsigned long freq); int devfreq_update_target(struct devfreq *devfreq, unsigned long freq); +void devfreq_get_freq_range(struct devfreq *devfreq, unsigned long *min_freq, + unsigned long *max_freq); static inline int devfreq_update_stats(struct devfreq *df) { diff --git a/drivers/devfreq/governor_passive.c b/drivers/devfreq/governor_passive.c index fc09324..7306e94 100644 --- a/drivers/devfreq/governor_passive.c +++ b/drivers/devfreq/governor_passive.c @@ -1,4 +1,4 @@ -// SPDX-License-Identifier: GPL-2.0-only + // SPDX-License-Identifier: GPL-2.0-only /* * linux/drivers/devfreq/governor_passive.c * @@ -8,76 +8,129 @@ */ #include +#include +#include +#include +#include #include #include #include "governor.h" -static int devfreq_passive_get_target_freq(struct devfreq *devfreq, - unsigned long *freq) +#define HZ_PER_KHZ 1000 + +static struct devfreq_cpu_data * +get_parent_cpu_data(struct devfreq_passive_data *p_data, + struct cpufreq_policy *policy) { - struct devfreq_passive_data *p_data - = (struct devfreq_passive_data *)devfreq->data; - struct devfreq *parent_devfreq = (struct devfreq *)p_data->parent; - unsigned long child_freq = ULONG_MAX; - struct dev_pm_opp *opp, *p_opp; - int i, count; + struct devfreq_cpu_data *parent_cpu_data; - /* - * If the devfreq device with passive governor has the specific method - * to determine the next frequency, should use the get_target_freq() - * of struct devfreq_passive_data. - */ - if (p_data->get_target_freq) - return p_data->get_target_freq(devfreq, freq); + if (!p_data || !policy) + return NULL; - /* - * If the parent and passive devfreq device uses the OPP table, - * get the next frequency by using the OPP table. - */ + list_for_each_entry(parent_cpu_data, &p_data->cpu_data_list, node) + if (parent_cpu_data->first_cpu == cpumask_first(policy->related_cpus)) + return parent_cpu_data; - /* - * - parent devfreq device uses the governors except for passive. - * - passive devfreq device uses the passive governor. - * - * Each devfreq has the OPP table. After deciding the new frequency - * from the governor of parent devfreq device, the passive governor - * need to get the index of new frequency on OPP table of parent - * device. And then the index is used for getting the suitable - * new frequency for passive devfreq device. - */ - if (!devfreq->profile || !devfreq->profile->freq_table - || devfreq->profile->max_state <= 0) - return -EINVAL; + return NULL; +} - /* - * The passive governor have to get the correct frequency from OPP - * list of parent device. Because in this case, *freq is temporary - * value which is decided by ondemand governor. - */ - if (devfreq->opp_table && parent_devfreq->opp_table) { - p_opp = devfreq_recommended_opp(parent_devfreq->dev.parent, - freq, 0); - if (IS_ERR(p_opp)) - return PTR_ERR(p_opp); +static unsigned long get_target_freq_by_required_opp(struct device *p_dev, + struct opp_table *p_opp_table, + struct opp_table *opp_table, + unsigned long *freq) +{ + struct dev_pm_opp *opp = NULL, *p_opp = NULL; + unsigned long target_freq; - opp = dev_pm_opp_xlate_required_opp(parent_devfreq->opp_table, - devfreq->opp_table, p_opp); - dev_pm_opp_put(p_opp); + if (!p_dev || !p_opp_table || !opp_table || !freq) + return 0; - if (IS_ERR(opp)) - goto no_required_opp; + p_opp = devfreq_recommended_opp(p_dev, freq, 0); + if (IS_ERR(p_opp)) + return 0; - *freq = dev_pm_opp_get_freq(opp); - dev_pm_opp_put(opp); + opp = dev_pm_opp_xlate_required_opp(p_opp_table, opp_table, p_opp); + dev_pm_opp_put(p_opp); + if (IS_ERR(opp)) return 0; + + target_freq = dev_pm_opp_get_freq(opp); + dev_pm_opp_put(opp); + + return target_freq; +} + +static int get_target_freq_with_cpufreq(struct devfreq *devfreq, + unsigned long *target_freq) +{ + struct devfreq_passive_data *p_data = + (struct devfreq_passive_data *)devfreq->data; + struct devfreq_cpu_data *parent_cpu_data; + struct cpufreq_policy *policy; + unsigned long cpu, cpu_cur, cpu_min, cpu_max, cpu_percent; + unsigned long dev_min, dev_max; + unsigned long freq = 0; + int ret = 0; + + for_each_online_cpu(cpu) { + policy = cpufreq_cpu_get(cpu); + if (!policy) { + ret = -EINVAL; + continue; + } + + parent_cpu_data = get_parent_cpu_data(p_data, policy); + if (!parent_cpu_data) { + cpufreq_cpu_put(policy); + continue; + } + + /* Get target freq via required opps */ + cpu_cur = parent_cpu_data->cur_freq * HZ_PER_KHZ; + freq = get_target_freq_by_required_opp(parent_cpu_data->dev, + parent_cpu_data->opp_table, + devfreq->opp_table, &cpu_cur); + if (freq) { + *target_freq = max(freq, *target_freq); + cpufreq_cpu_put(policy); + continue; + } + + /* Use interpolation if required opps is not available */ + devfreq_get_freq_range(devfreq, &dev_min, &dev_max); + + cpu_min = parent_cpu_data->min_freq; + cpu_max = parent_cpu_data->max_freq; + cpu_cur = parent_cpu_data->cur_freq; + + cpu_percent = ((cpu_cur - cpu_min) * 100) / (cpu_max - cpu_min); + freq = dev_min + mult_frac(dev_max - dev_min, cpu_percent, 100); + + *target_freq = max(freq, *target_freq); + cpufreq_cpu_put(policy); } -no_required_opp: - /* - * Get the OPP table's index of decided frequency by governor - * of parent device. - */ + return ret; +} + +static int get_target_freq_with_devfreq(struct devfreq *devfreq, + unsigned long *freq) +{ + struct devfreq_passive_data *p_data + = (struct devfreq_passive_data *)devfreq->data; + struct devfreq *parent_devfreq = (struct devfreq *)p_data->parent; + unsigned long child_freq = ULONG_MAX; + int i, count; + + /* Get target freq via required opps */ + child_freq = get_target_freq_by_required_opp(parent_devfreq->dev.parent, + parent_devfreq->opp_table, + devfreq->opp_table, freq); + if (child_freq) + goto out; + + /* Use interpolation if required opps is not available */ for (i = 0; i < parent_devfreq->profile->max_state; i++) if (parent_devfreq->profile->freq_table[i] == *freq) break; @@ -85,7 +138,6 @@ static int devfreq_passive_get_target_freq(struct devfreq *devfreq, if (i == parent_devfreq->profile->max_state) return -EINVAL; - /* Get the suitable frequency by using index of parent device. */ if (i < devfreq->profile->max_state) { child_freq = devfreq->profile->freq_table[i]; } else { @@ -93,12 +145,202 @@ static int devfreq_passive_get_target_freq(struct devfreq *devfreq, child_freq = devfreq->profile->freq_table[count - 1]; } - /* Return the suitable frequency for passive device. */ +out: *freq = child_freq; return 0; } +static int devfreq_passive_get_target_freq(struct devfreq *devfreq, + unsigned long *freq) +{ + struct devfreq_passive_data *p_data = + (struct devfreq_passive_data *)devfreq->data; + int ret; + + if (!p_data) + return -EINVAL; + + /* + * If the devfreq device with passive governor has the specific method + * to determine the next frequency, should use the get_target_freq() + * of struct devfreq_passive_data. + */ + if (p_data->get_target_freq) + return p_data->get_target_freq(devfreq, freq); + + switch (p_data->parent_type) { + case DEVFREQ_PARENT_DEV: + ret = get_target_freq_with_devfreq(devfreq, freq); + break; + case CPUFREQ_PARENT_DEV: + ret = get_target_freq_with_cpufreq(devfreq, freq); + break; + default: + ret = -EINVAL; + dev_err(&devfreq->dev, "Invalid parent type\n"); + break; + } + + return ret; +} + +static int cpufreq_passive_notifier_call(struct notifier_block *nb, + unsigned long event, void *ptr) +{ + struct devfreq_passive_data *p_data = + container_of(nb, struct devfreq_passive_data, nb); + struct devfreq *devfreq = (struct devfreq *)p_data->this; + struct devfreq_cpu_data *parent_cpu_data; + struct cpufreq_freqs *freqs = ptr; + unsigned int cur_freq; + int ret; + + if (event != CPUFREQ_POSTCHANGE || !freqs) + return 0; + + parent_cpu_data = get_parent_cpu_data(p_data, freqs->policy); + if (!parent_cpu_data || parent_cpu_data->cur_freq == freqs->new) + return 0; + + cur_freq = parent_cpu_data->cur_freq; + parent_cpu_data->cur_freq = freqs->new; + + mutex_lock(&devfreq->lock); + ret = devfreq_update_target(devfreq, freqs->new); + mutex_unlock(&devfreq->lock); + if (ret) { + parent_cpu_data->cur_freq = cur_freq; + dev_err(&devfreq->dev, "failed to update the frequency.\n"); + return ret; + } + + return 0; +} + +static int cpufreq_passive_unregister_notifier(struct devfreq *devfreq) +{ + struct devfreq_passive_data *p_data + = (struct devfreq_passive_data *)devfreq->data; + struct devfreq_cpu_data *parent_cpu_data; + int cpu, ret = 0; + + if (p_data->nb.notifier_call) { + ret = cpufreq_unregister_notifier(&p_data->nb, + CPUFREQ_TRANSITION_NOTIFIER); + if (ret < 0) + return ret; + } + + for_each_possible_cpu(cpu) { + struct cpufreq_policy *policy = cpufreq_cpu_get(cpu); + if (!policy) { + ret = -EINVAL; + continue; + } + + parent_cpu_data = get_parent_cpu_data(p_data, policy); + if (!parent_cpu_data) { + cpufreq_cpu_put(policy); + continue; + } + + list_del(&parent_cpu_data->node); + if (parent_cpu_data->opp_table) + dev_pm_opp_put_opp_table(parent_cpu_data->opp_table); + kfree(parent_cpu_data); + cpufreq_cpu_put(policy); + } + + return ret; +} + +static int cpufreq_passive_register_notifier(struct devfreq *devfreq) +{ + struct devfreq_passive_data *p_data + = (struct devfreq_passive_data *)devfreq->data; + struct device *dev = devfreq->dev.parent; + struct opp_table *opp_table = NULL; + struct devfreq_cpu_data *parent_cpu_data; + struct cpufreq_policy *policy; + struct device *cpu_dev; + unsigned int cpu; + int ret; + + p_data->cpu_data_list + = (struct list_head)LIST_HEAD_INIT(p_data->cpu_data_list); + + p_data->nb.notifier_call = cpufreq_passive_notifier_call; + ret = cpufreq_register_notifier(&p_data->nb, CPUFREQ_TRANSITION_NOTIFIER); + if (ret) { + dev_err(dev, "failed to register cpufreq notifier\n"); + p_data->nb.notifier_call = NULL; + goto err; + } + + for_each_possible_cpu(cpu) { + policy = cpufreq_cpu_get(cpu); + if (!policy) { + ret = -EPROBE_DEFER; + goto err; + } + + parent_cpu_data = get_parent_cpu_data(p_data, policy); + if (parent_cpu_data) { + cpufreq_cpu_put(policy); + continue; + } + + parent_cpu_data = kzalloc(sizeof(*parent_cpu_data), + GFP_KERNEL); + if (!parent_cpu_data) { + ret = -ENOMEM; + goto err_put_policy; + } + + cpu_dev = get_cpu_device(cpu); + if (!cpu_dev) { + dev_err(dev, "failed to get cpu device\n"); + ret = -ENODEV; + goto err_free_cpu_data; + } + + opp_table = dev_pm_opp_get_opp_table(cpu_dev); + if (IS_ERR(opp_table)) { + dev_err(dev, "failed to get opp_table of cpu%d\n", cpu); + ret = PTR_ERR(opp_table); + goto err_free_cpu_data; + } + + parent_cpu_data->dev = cpu_dev; + parent_cpu_data->opp_table = opp_table; + parent_cpu_data->first_cpu = cpumask_first(policy->related_cpus); + parent_cpu_data->cur_freq = policy->cur; + parent_cpu_data->min_freq = policy->cpuinfo.min_freq; + parent_cpu_data->max_freq = policy->cpuinfo.max_freq; + + list_add_tail(&parent_cpu_data->node, &p_data->cpu_data_list); + cpufreq_cpu_put(policy); + } + + mutex_lock(&devfreq->lock); + ret = devfreq_update_target(devfreq, 0L); + mutex_unlock(&devfreq->lock); + if (ret) + dev_err(dev, "failed to update the frequency\n"); + + return ret; + +err_free_cpu_data: + kfree(parent_cpu_data); +err_put_policy: + cpufreq_cpu_put(policy); +err: + WARN_ON(cpufreq_passive_unregister_notifier(devfreq)); + + return ret; +} + static int devfreq_passive_notifier_call(struct notifier_block *nb, unsigned long event, void *ptr) { @@ -131,30 +373,55 @@ static int devfreq_passive_notifier_call(struct notifier_block *nb, return NOTIFY_DONE; } -static int devfreq_passive_event_handler(struct devfreq *devfreq, - unsigned int event, void *data) +static int devfreq_passive_unregister_notifier(struct devfreq *devfreq) +{ + struct devfreq_passive_data *p_data + = (struct devfreq_passive_data *)devfreq->data; + struct devfreq *parent = (struct devfreq *)p_data->parent; + struct notifier_block *nb = &p_data->nb; + + return devfreq_unregister_notifier(parent, nb, DEVFREQ_TRANSITION_NOTIFIER); +} + +static int devfreq_passive_register_notifier(struct devfreq *devfreq) { struct devfreq_passive_data *p_data = (struct devfreq_passive_data *)devfreq->data; struct devfreq *parent = (struct devfreq *)p_data->parent; struct notifier_block *nb = &p_data->nb; - int ret = 0; if (!parent) return -EPROBE_DEFER; + nb->notifier_call = devfreq_passive_notifier_call; + return devfreq_register_notifier(parent, nb, DEVFREQ_TRANSITION_NOTIFIER); +} + +static int devfreq_passive_event_handler(struct devfreq *devfreq, + unsigned int event, void *data) +{ + struct devfreq_passive_data *p_data + = (struct devfreq_passive_data *)devfreq->data; + int ret = -EINVAL; + + if (!p_data) + return -EINVAL; + + if (!p_data->this) + p_data->this = devfreq; + switch (event) { case DEVFREQ_GOV_START: - if (!p_data->this) - p_data->this = devfreq; - - nb->notifier_call = devfreq_passive_notifier_call; - ret = devfreq_register_notifier(parent, nb, - DEVFREQ_TRANSITION_NOTIFIER); + if (p_data->parent_type == DEVFREQ_PARENT_DEV) + ret = devfreq_passive_register_notifier(devfreq); + else if (p_data->parent_type == CPUFREQ_PARENT_DEV) + ret = cpufreq_passive_register_notifier(devfreq); break; case DEVFREQ_GOV_STOP: - WARN_ON(devfreq_unregister_notifier(parent, nb, - DEVFREQ_TRANSITION_NOTIFIER)); + if (p_data->parent_type == DEVFREQ_PARENT_DEV) + WARN_ON(devfreq_passive_unregister_notifier(devfreq)); + else if (p_data->parent_type == CPUFREQ_PARENT_DEV) + WARN_ON(cpufreq_passive_unregister_notifier(devfreq)); break; default: break; diff --git a/include/linux/devfreq.h b/include/linux/devfreq.h index d94b9ed..9a5246d 100644 --- a/include/linux/devfreq.h +++ b/include/linux/devfreq.h @@ -38,6 +38,7 @@ enum devfreq_timer { struct devfreq; struct devfreq_governor; +struct devfreq_cpu_data; struct thermal_cooling_device; /** @@ -289,6 +290,11 @@ struct devfreq_simple_ondemand_data { #endif #if IS_ENABLED(CONFIG_DEVFREQ_GOV_PASSIVE) +enum devfreq_parent_dev_type { + DEVFREQ_PARENT_DEV, + CPUFREQ_PARENT_DEV, +}; + /** * struct devfreq_passive_data - ``void *data`` fed to struct devfreq * and devfreq_add_device @@ -300,8 +306,11 @@ struct devfreq_simple_ondemand_data { * using governors except for passive governor. * If the devfreq device has the specific method to decide * the next frequency, should use this callback. - * @this: the devfreq instance of own device. - * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER list + * @parent_type: the parent type of the device. + * @this: the devfreq instance of own device. + * @nb: the notifier block for DEVFREQ_TRANSITION_NOTIFIER or + * CPUFREQ_TRANSITION_NOTIFIER list. + * @cpu_data_list: the list of cpu frequency data for all cpufreq_policy. * * The devfreq_passive_data have to set the devfreq instance of parent * device with governors except for the passive governor. But, don't need to @@ -315,9 +324,13 @@ struct devfreq_passive_data { /* Optional callback to decide the next frequency of passvice device */ int (*get_target_freq)(struct devfreq *this, unsigned long *freq); + /* Should set the type of parent device */ + enum devfreq_parent_dev_type parent_type; + /* For passive governor's internal use. Don't need to set them */ struct devfreq *this; struct notifier_block nb; + struct list_head cpu_data_list; }; #endif -- 2.18.0