On Wed, Aug 19, 2015 at 12:32:43AM +0800, T. Zhou wrote: > Hi, > > On Mon, Aug 17, 2015 at 04:45:50PM +0900, byungchul.p...@lge.com wrote: > > From: Byungchul Park <byungchul.p...@lge.com> > > > > current code is wrong with cfs_rq's avg loads when changing a task's > > cfs_rq to another. i tested with "echo pid > cgroup" and found that > > e.g. cfs_rq->avg.load_avg became larger and larger whenever i changed > > a cgroup to another again and again. we have to sync se's avg loads > > with both *prev* cfs_rq and next cfs_rq when changing its group. > > > > my simple think about above, may be nothing or wrong, just ignore it. > > if a load balance migration happened just before cgroup change, prev > cfs_rq and next cfs_rq will be on different cpu. migrate_task_rq_fair()
hello, two oerations, migration and cgroup change, are protected by lock. therefore it would never happen. :) thanks, byungchul > and update_cfs_rq_load_avg() will sync and remove se's load avg from > prev cfs_rq. whether or not queued, well done. dequeue_task() decay se > and pre_cfs before calling task_move_group_fair(). after set cfs_rq in > task_move_group_fair(), if queued, se's load avg do not add to next > cfs_rq(try set last_update_time to 0 like migration to add), if !queued, > also need to add se's load avg to next cfs_rq. > > if no load balance migration happened when change cgroup. prev cfs_rq > and next cfs_rq may be on same cpu(not sure), this time, need to remove > se's load avg by ourself, also need to add se's load avg on next cfs_rq. > > thinks, > -- > Tao > -- > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > Please read the FAQ at http://www.tux.org/lkml/ -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/