hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I7ZBSR CVE: NA
----------------------------------------
In smart_grid zone, the system is divided into multiple zones. The typical configuration is one hot zone and the other warm zone.
This commit can configure an independent cpufreq governor for each zone. After the zone range in the system changes dynamically, it can Automatically configure the corresponding cpufreq governor.
The default configuration of hot zone is performance governor, and the default configuration of warm zone is powersave governor.
-------- -------- -------- | group0 | | group1 | | group2 | -------- -------- -------- | | | v v v --------------------------- ----------------- | | | | | hot zone | | warm zone | | | | | --------------------------- ----------------- ^ ^ | | ------------- ---------- | performance | <- cpufreq governor -> |powersave | ------------ ----------
Introduce two attributes in /sys/devices/system/cpu/cpufreq:
1. smart_grid_governor_enable
For smart_grid governor, set 1 for enable, 0 for disable, read to get current status.
For Notice, if smart_grid governor is enabled the governor of cpu may auto switch by smart_grid, so if user wants to config governor by itself smart_grid_governor shoudle be disabled.
2. smart_grid_governor
Only can set when smart_grid_governor has been enabled.
Set {level}-{governor name} formate to change each zone's governor:
0-performance will set all hot zone cpu's governor to performance. 1-powersave will set all warm zone cpu's governor to powersave.
Signed-off-by: Yipeng Zou zouyipeng@huawei.com --- drivers/cpufreq/cpufreq.c | 236 ++++++++++++++++++++++++++++++++++++++ include/linux/cpufreq.h | 12 ++ kernel/sched/grid/qos.c | 10 +- 3 files changed, 256 insertions(+), 2 deletions(-)
diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c index 4bd9fc287c82..7656f45bb851 100644 --- a/drivers/cpufreq/cpufreq.c +++ b/drivers/cpufreq/cpufreq.c @@ -2444,6 +2444,239 @@ int cpufreq_boost_enabled(void) } EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
+#ifdef CONFIG_QOS_SCHED_SMART_GRID + +struct smart_grid_zone { + char governor_name[SMART_GRID_ZONE_NR][CPUFREQ_NAME_LEN]; + unsigned int enable; + struct irq_work irq_work; + struct work_struct work; + unsigned int is_init; +}; + +static struct smart_grid_zone sg_zone; +static DEFINE_MUTEX(sg_zone_lock); + +#define SG_WRITE_BUFF_LEN 30 + +void cpufreq_smart_grid_start_sync(void) +{ + if (likely(sg_zone.is_init)) + irq_work_queue(&sg_zone.irq_work); +} + +static ssize_t show_smart_grid_governor(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + int len = 0; + int gov_index; + + mutex_lock(&sg_zone_lock); + if (!sg_zone.enable) { + mutex_unlock(&sg_zone_lock); + return sprintf(buf, "smart_grid governor disable\n"); + } + + for (gov_index = 0; gov_index < SMART_GRID_ZONE_NR; gov_index++) + len += sprintf(buf + len, "smart_grid-%d: %s\n", gov_index, + sg_zone.governor_name[gov_index]); + + mutex_unlock(&sg_zone_lock); + return len; +} + +static ssize_t store_smart_grid_governor(struct kobject *kobj, struct kobj_attribute *attr, + const char *buf, size_t count) +{ + unsigned int current_level; + char *level_string = NULL; + char buf_string[SG_WRITE_BUFF_LEN]; + char *gov_string = buf_string; + char save_string[CPUFREQ_NAME_LEN]; + int ret; + struct cpufreq_policy new_policy; + + mutex_lock(&sg_zone_lock); + if (!sg_zone.enable) { + ret = -EINVAL; + goto fail; + } + + if (strscpy(buf_string, buf, SG_WRITE_BUFF_LEN) <= 0) { + ret = -EINVAL; + goto fail; + } + + level_string = strsep(&gov_string, "-"); + if (level_string == NULL) { + ret = -EINVAL; + goto fail; + } + + if (kstrtouint(level_string, 10, ¤t_level)) { + ret = -EINVAL; + goto fail; + } + + if (current_level >= SMART_GRID_ZONE_NR) { + ret = -EINVAL; + goto fail; + } + + if (sscanf(gov_string, "%15s", save_string) != 1) { + ret = -EINVAL; + goto fail; + } + + if (cpufreq_parse_governor(save_string, &new_policy)) { + ret = -EINVAL; + goto fail; + } + if (new_policy.governor) + module_put(new_policy.governor->owner); + + strscpy(sg_zone.governor_name[current_level], save_string, CPUFREQ_NAME_LEN); + cpufreq_smart_grid_start_sync(); + mutex_unlock(&sg_zone_lock); + return count; + +fail: + mutex_unlock(&sg_zone_lock); + return ret; +} +define_one_global_rw(smart_grid_governor); + +static ssize_t show_smart_grid_governor_enable(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + return sprintf(buf, "%u\n", sg_zone.enable); +} + +static void smart_grid_irq_work(struct irq_work *irq_work) +{ + struct smart_grid_zone *zone; + + zone = container_of(irq_work, struct smart_grid_zone, irq_work); + schedule_work_on(smp_processor_id(), &zone->work); +} + +static void smart_grid_work_handler(struct work_struct *work) +{ + struct smart_grid_zone *zone; + struct cpufreq_policy *policy = NULL; + unsigned int cpu; + int gov_index; + struct cpufreq_policy new_policy; + + zone = container_of(work, struct smart_grid_zone, work); + + mutex_lock(&sg_zone_lock); + if (!sg_zone.enable) { + mutex_unlock(&sg_zone_lock); + return; + } + + /* + * Because of the policy may be shared between hot and warm zone. + * We need to make sure hot zone have the highest priority. + */ + for (gov_index = SMART_GRID_ZONE_NR - 1; gov_index >= 0; gov_index--) { + if (cpufreq_parse_governor(sg_zone.governor_name[gov_index], &new_policy)) + continue; + + for_each_cpu(cpu, sched_grid_zone_cpumask(gov_index)) { + if (cpu_is_offline(cpu)) + continue; + + policy = cpufreq_cpu_get(cpu); + if (policy == NULL) + continue; + + if (policy->governor == new_policy.governor) { + cpufreq_cpu_put(policy); + continue; + } + /*Try to switch governor */ + store_scaling_governor(policy, sg_zone.governor_name[gov_index], + CPUFREQ_NAME_LEN); + cpufreq_cpu_put(policy); + } + if (new_policy.governor) + module_put(new_policy.governor->owner); + } + mutex_unlock(&sg_zone_lock); +} + +static void sg_zone_set_enable(void) +{ + int gov_index; + + /* Set default smart_grid governor */ + for (gov_index = 0; gov_index < SMART_GRID_ZONE_NR; gov_index++) { + if (!gov_index) + strscpy(sg_zone.governor_name[gov_index], "performance", CPUFREQ_NAME_LEN); + else + strscpy(sg_zone.governor_name[gov_index], "powersave", CPUFREQ_NAME_LEN); + } + + sg_zone.enable = 1; + cpufreq_smart_grid_start_sync(); +} + +static void sg_zone_set_disable(void) +{ + sg_zone.enable = 0; +} + +static ssize_t store_smart_grid_governor_enable(struct kobject *kobj, struct kobj_attribute *attr, + const char *buf, size_t count) +{ + unsigned int enable; + + if (kstrtouint(buf, 10, &enable)) + return -EINVAL; + + if (enable > 1) + return -EINVAL; + + mutex_lock(&sg_zone_lock); + if (sg_zone.enable == enable) { + mutex_unlock(&sg_zone_lock); + return -EINVAL; + } + + if (enable) + sg_zone_set_enable(); + else + sg_zone_set_disable(); + + mutex_unlock(&sg_zone_lock); + return count; +} +define_one_global_rw(smart_grid_governor_enable); + +static int create_smart_grid_sysfs_file(void) +{ + int ret; + + ret = sysfs_create_file(cpufreq_global_kobject, &smart_grid_governor.attr); + if (ret) + pr_err("%s: cannot register global smart_grid_governor sysfs file\n", + __func__); + + ret = sysfs_create_file(cpufreq_global_kobject, &smart_grid_governor_enable.attr); + if (ret) + pr_err("%s: cannot register global smart_grid_governor_enable sysfs file\n", + __func__); + + init_irq_work(&sg_zone.irq_work, smart_grid_irq_work); + INIT_WORK(&sg_zone.work, smart_grid_work_handler); + sg_zone.enable = 0; + sg_zone.is_init = 1; + return ret; +} +#endif + /********************************************************************* * REGISTER / UNREGISTER CPUFREQ DRIVER * *********************************************************************/ @@ -2603,6 +2836,9 @@ static int __init cpufreq_core_init(void) cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj); BUG_ON(!cpufreq_global_kobject);
+#ifdef CONFIG_QOS_SCHED_SMART_GRID + create_smart_grid_sysfs_file(); +#endif return 0; } module_param(off, int, 0444); diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h index 3361663144a1..07df0924bced 100644 --- a/include/linux/cpufreq.h +++ b/include/linux/cpufreq.h @@ -18,6 +18,9 @@ #include <linux/notifier.h> #include <linux/spinlock.h> #include <linux/sysfs.h> +#ifdef CONFIG_QOS_SCHED_SMART_GRID +#include <linux/sched/grid_qos.h> +#endif
/********************************************************************* * CPUFREQ INTERFACE * @@ -529,6 +532,15 @@ unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy); int cpufreq_register_governor(struct cpufreq_governor *governor); void cpufreq_unregister_governor(struct cpufreq_governor *governor);
+#ifdef CONFIG_QOS_SCHED_SMART_GRID +/* Implement in cpufreq.c */ +#ifdef CONFIG_CPU_FREQ +void cpufreq_smart_grid_start_sync(void); +#else +static inline void cpufreq_smart_grid_start_sync(void) { return; } +#endif +#endif + struct cpufreq_governor *cpufreq_default_governor(void); struct cpufreq_governor *cpufreq_fallback_governor(void);
diff --git a/kernel/sched/grid/qos.c b/kernel/sched/grid/qos.c index 87ebd5003640..dbf070c8cc58 100644 --- a/kernel/sched/grid/qos.c +++ b/kernel/sched/grid/qos.c @@ -174,6 +174,7 @@ int sched_grid_zone_update(bool is_locked) raw_spin_lock_irqsave(&sg_zone.lock, flags);
cpumask_clear(&sg_zone.cpus[SMART_GRID_ZONE_HOT]); + cpumask_clear(&sg_zone.cpus[SMART_GRID_ZONE_WARM]);
list_for_each(pos, &sg_zone.af_list_head) { af_pos = list_entry(pos, struct auto_affinity, af_list); @@ -185,14 +186,19 @@ int sched_grid_zone_update(bool is_locked)
cpumask_or(&sg_zone.cpus[SMART_GRID_ZONE_HOT], &sg_zone.cpus[SMART_GRID_ZONE_HOT], af_pos->ad.domains[af_pos->ad.curr_level]); + /* Update warm zone CPUs to max level first */ + cpumask_or(&sg_zone.cpus[SMART_GRID_ZONE_WARM], &sg_zone.cpus[SMART_GRID_ZONE_WARM], + af_pos->ad.domains[af_pos->ad.dcount - 1]); }
- cpumask_complement(&sg_zone.cpus[SMART_GRID_ZONE_WARM], - &sg_zone.cpus[SMART_GRID_ZONE_HOT]); + /* Then reset warm zone CPUs without hot zone CPUs */ + cpumask_andnot(&sg_zone.cpus[SMART_GRID_ZONE_WARM], &sg_zone.cpus[SMART_GRID_ZONE_WARM], + &sg_zone.cpus[SMART_GRID_ZONE_HOT]);
if (!is_locked) raw_spin_unlock_irqrestore(&sg_zone.lock, flags);
+ cpufreq_smart_grid_start_sync(); return 0; }