On 01/06/16 21:10, Peter Zijlstra wrote: > On Wed, Jun 01, 2016 at 08:39:19PM +0100, Dietmar Eggemann wrote: >> This is an alternative approach to '[RFC PATCH v2] sched: reflect >> sched_entity movement into task_group's utilization' which requires >> '[RFC PATCH] sched: fix hierarchical order in rq->leaf_cfs_rq_list' > > Awesome, a 3rd alternative :-)
Yeah, sorry, but at least it runs op tip/sched/core w/o any other PELT related changes. Moreover, it's less code changes compared to the reflect-thing. And Yuyang's implementation depended on the optimization of _update_load_avg() (1us->1ms) which IMHO turned out to be not working.