On Thu, May 26, 2022 at 12:40:17PM +0800, 史思远 <shisiyuan19870131@xxxxxxxxx> wrote: > echo 1413 > /dev/cpuset/foreground/cgroup.procs > echo 1413 > /dev/cpuset/foreground/cgroup.procs > echo 1413 > /dev/cpuset/foreground/cgroup.procs Thank you. > In this situation, since the second echo, dst_cgrp equals to src_cgrp(both > cgroups are the 'foreground' cpuset). > cgroup1_procs_write（）-> __cgroup1_procs_write() > -> cgrp = cgroup_kn_lock_live() > -> task = cgroup_procs_write_start(buf...) > -> cgroup_attach_task(cgrp, task, threadgroup); > -> 1) cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx); // > detect the (src_cgrp == dst_cgrp), don't need to continue the following > migration flow. > -> 2) cgroup_migrate_prepare_dst(&mgctx) Here, I'd like to let you know about [1] > -> 3) cgroup_migrate(leader, threadgroup, &mgctx) > -> 4) cgroup_migrate_finish(&mgctx) // cleanup > -> cgroup_procs_write_finish(); // cleanup > >> The main purpose for this patch is to handle the situation above, and > just noticed that cgroup_update_dfl_csses() also > calls cgroup_migrate_add_src(). So I assume it's a similar situation. We > can discuss more about it. So I think the patch doesn't save that much and identical css_set is better condition to check on (covers cgroup_update_dfl_csses too). Michal [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/cgroup/cgroup.c?h=v5.18#n2710
