[PATCH OLK-5.10 1/2] sched/dynamic_affinity: fix preffered_cpu offline problem
From: He Yujie <coka.heyujie@huawei.com> Offering: HULK hulk inclusion category: bugfix bugzilla: 190373 -------------------------------- After the preferred_cpu goes offline, the core selection of dynamic affinity does not check whether the preferred_cpu is valid. As a result, related dynamic affinity processes are concentrated on a shared_cpu. This patch resolves this problem to checks whether the preferred_cpu is valid and compares only the usage threshold of the valid preferred_cpu. Fixes: 2e1dfc02d115 ("sched: Adjust wakeup cpu range according CPU util dynamicly") Signed-off-by: He Yujie <coka.heyujie@huawei.com> Signed-off-by: Xiangwei Li <liwei728@huawei.com> --- kernel/sched/fair.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 5fc8d6a25b9a..459a1b174f08 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -8141,7 +8141,7 @@ static void set_task_select_cpus(struct task_struct *p, int *idlest_cpu, /* manual mode */ tg = task_group(p); - for_each_cpu(cpu, p->prefer_cpus) { + for_each_cpu_and(cpu, p->prefer_cpus, cpu_online_mask) { if (idlest_cpu && (available_idle_cpu(cpu) || sched_idle_cpu(cpu))) { *idlest_cpu = cpu; } else if (idlest_cpu) { @@ -8163,10 +8163,17 @@ static void set_task_select_cpus(struct task_struct *p, int *idlest_cpu, util_avg_sum += taskgroup_cpu_util(tg, cpu); tg_capacity += capacity_of(cpu); + nr_cpus_valid++; } rcu_read_unlock(); - if (tg_capacity > cpumask_weight(p->prefer_cpus) && + /* + * Follow cases should select cpus_ptr, checking by condition of + * tg_capacity > nr_cpus_valid: + * 1. all prefer_cpus offline; + * 2. all prefer_cpus has no cfs capaicity(tg_capacity = nr_cpus_valid * 1) + */ + if (tg_capacity > nr_cpus_valid && util_avg_sum * 100 <= tg_capacity * sysctl_sched_util_low_pct) { p->select_cpus = p->prefer_cpus; if (sd_flag & SD_BALANCE_WAKE) -- 2.30.0
反馈: 您发送到kernel@openeuler.org的补丁/补丁集，转换为PR失败！ 邮件列表地址：https://mailweb.openeuler.org/archives/list/kernel@openeuler.org/message/T4J... 失败原因：补丁集缺失封面信息 建议解决方法：请提供补丁集并重新发送您的补丁集到邮件列表 FeedBack: The patch(es) which you have sent to kernel@openeuler.org has been converted to PR failed! Mailing list address: https://mailweb.openeuler.org/archives/list/kernel@openeuler.org/message/T4J... Failed Reason: the cover of the patches is missing Suggest Solution: please checkout and apply the patches' cover and send all again
From: He Yujie <coka.heyujie@huawei.com> Offering: HULK hulk inclusion category: bugfix bugzilla: 190429 -------------------------------- The task_rq selection of dynamic affinity use cpu capacity to determine select_cpus range. When realtime tasks are running on the cpu all the time, cfs tasks and the thread of softirq is suppressed, and the cpu capacity is not updated in time. As a result, the select_cpus range is always selected for preferred_cpus. then cfs task will never be able to run because realtime tasks has been running. Therefore, if realtime tasks is running during the task_rq selection of dynamic affinity, the cpu capacity should be calculated to solve such a problem. Fixes: bb9b25afa69f ("[Huawei] sched: Adjust wakeup cpu range according CPU util dynamicly") Signed-off-by: He Yujie <coka.heyujie@huawei.com> Signed-off-by: Xiangwei Li <liwei728@huawei.com> --- kernel/sched/fair.c | 19 ++++++++++++++++++- 1 file changed, 18 insertions(+), 1 deletion(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 459a1b174f08..c8cde4072674 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -8102,6 +8102,19 @@ static inline unsigned long taskgroup_cpu_util(struct task_group *tg, return cpu_util(cpu); } +static unsigned long scale_rt_capacity(int cpu); + +static inline unsigned long calc_cpu_capacity(int cpu) +{ + unsigned long capacity; + + capacity = scale_rt_capacity(cpu); + if (!capacity) + capacity = 1; + + return capacity; +} + /* * set_task_select_cpus: select the cpu range for task * @p: the task whose available cpu range will to set @@ -8162,8 +8175,12 @@ static void set_task_select_cpus(struct task_struct *p, int *idlest_cpu, } util_avg_sum += taskgroup_cpu_util(tg, cpu); - tg_capacity += capacity_of(cpu); nr_cpus_valid++; + + if (cpu_rq(cpu)->rt.rt_nr_running) + tg_capacity += calc_cpu_capacity(cpu); + else + tg_capacity += capacity_of(cpu); } rcu_read_unlock(); -- 2.30.0
participants (2)
-
patchwork bot -
Xiangwei Li