!640 sched/fair: Introduce QOS_SMT_EXPELL priority reversion mechanism

Merge Pull Request from: @NNNNicole 
 
Here is the typical case that priority inversion will caused occasionally
by SMT expelling: Assuming that there are two SMT cores-cA and cB,
online tasks are running on cA while offline tasks on cB. With SMT
expelling, online task will drives off offline tasks to occupy all SMT
cores exclusively, which, in turn, will starve the offline task to release
the related resources other tasks with higher priority need.

Hence, this patch will introduce another mechanism to alleviate this
situation. For all offline tasks, one metric of profiling the maximum task
expelling duration is set up and the default value is 5 seconds, if
such offline task exsits, all offline tasks will be allowed to run into
one small sleep(msleep) loop in kernel before they goes into
usermode; and further, if the two SMT cores(such as cA and cB) are
idle or don't get any online tasks to run, for these offline tasks, they
will continue to run in usermode for the next schedule.

kernel options:
CONFIG_QOS_SCHED_SMT_EXPELLER=y 
 
Link:https://gitee.com/openeuler/kernel/pulls/640

 

Reviewed-by: Zucheng Zheng <zhengzucheng@huawei.com>
Reviewed-by: Liu Chao <liuchao173@huawei.com>
Signed-off-by: Jialin Zhang <zhangjialin11@huawei.com>