From: Zhang Qiao <zhangqiao22@huawei.com> hulk inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I4WOPM CVE: NA -------------------------------- cfs_bandwidth_usage_inc() need hold jump_label_mutex and might sleep, so we can not call it in atomic context. Fix this by moving cfs_bandwidth_usage_{inc,dec}() out of rcu read critical section. Fixes: f7b390cd929 ("sched: Change cgroup task scheduler policy") Signed-off-by: Zhang Qiao <zhangqiao22@huawei.com> Reviewed-by: Chen Hui <judy.chenhui@huawei.com> Reviewed-by: Wang Weiyang <wangweiyang2@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Laibin Qiu <qiulaibin@huawei.com> --- kernel/sched/core.c | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-) diff --git a/kernel/sched/core.c b/kernel/sched/core.c index b36a3b4c60e9..496ce71f93a7 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -6981,13 +6981,10 @@ static int tg_change_scheduler(struct task_group *tg, void *data) struct cgroup_subsys_state *css = &tg->css; tg->qos_level = qos_level; - if (qos_level == -1) { + if (qos_level == -1) policy = SCHED_IDLE; - cfs_bandwidth_usage_inc(); - } else { + else policy = SCHED_NORMAL; - cfs_bandwidth_usage_dec(); - } param.sched_priority = 0; css_task_iter_start(css, 0, &it); @@ -7015,6 +7012,13 @@ static int cpu_qos_write(struct cgroup_subsys_state *css, if (tg->qos_level == -1 && qos_level == 0) return -EINVAL; + cpus_read_lock(); + if (qos_level == -1) + cfs_bandwidth_usage_inc(); + else + cfs_bandwidth_usage_dec(); + cpus_read_unlock(); + rcu_read_lock(); walk_tg_tree_from(tg, tg_change_scheduler, tg_nop, (void *)(&qos_level)); rcu_read_unlock(); -- 2.22.0