summaryrefslogtreecommitdiffstats
path: root/security/device_cgroup.c
diff options
context:
space:
mode:
authorTejun Heo <tj@kernel.org>2014-05-16 19:22:48 +0200
committerTejun Heo <tj@kernel.org>2014-05-16 19:22:48 +0200
commitea280e7b408ca0dad195ce9836feccdd1dc32131 (patch)
tree524a648df41fbd7dc270c225212d96d10b5a7815 /security/device_cgroup.c
parentmemcg: remove tasks/children test from mem_cgroup_force_empty() (diff)
downloadlinux-ea280e7b408ca0dad195ce9836feccdd1dc32131.tar.xz
linux-ea280e7b408ca0dad195ce9836feccdd1dc32131.zip
memcg: update memcg_has_children() to use css_next_child()
Currently, memcg_has_children() and mem_cgroup_hierarchy_write() directly test cgroup->children for list emptiness. It's semantically correct in traditional hierarchies as it actually wants to test for any children dead or alive; however, cgroup->children is not a published field and scheduled to go away. This patch moves out .use_hierarchy test out of memcg_has_children() and updates it to use css_next_child() to test whether there exists any children. With .use_hierarchy test moved out, it can also be used by mem_cgroup_hierarchy_write(). A side note: As .use_hierarchy is going away, it doesn't really matter but I'm not sure about how it's used in __memcg_activate_kmem(). The condition tested by memcg_has_children() is mushy when seen from userland as its result is affected by dead csses which aren't visible from userland. I think the rule would be a lot clearer if we have a dedicated "freshly minted" flag which gets cleared when the first task is migrated into it or the first child is created and then gate activation with that. v2: Added comment noting that testing use_hierarchy is the responsibility of the callers of memcg_has_children() as suggested by Michal Hocko. Signed-off-by: Tejun Heo <tj@kernel.org> Acked-by: Michal Hocko <mhocko@suse.cz> Acked-by: Li Zefan <lizefan@huawei.com> Cc: Johannes Weiner <hannes@cmpxchg.org>
Diffstat (limited to 'security/device_cgroup.c')
0 files changed, 0 insertions, 0 deletions