[RFC][ only for review ] memory controller bacground reclaim [5/5]

Paul Menage menage at google.com
Fri Nov 30 23:16:26 PST 2007


On Nov 28, 2007 12:57 AM, KAMEZAWA Hiroyuki
<kamezawa.hiroyu at jp.fujitsu.com> wrote:
> @@ -1226,6 +1303,7 @@
>  {
>         struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
>         mem_cgroup_force_empty(mem);
> +       kthread_stop(mem->daemon.thread);
>  }
>

This means that if the user tries to delete the cgroup (hence calling
the predestroy handlers) but fails (e.g. because some other subsystem
is still busy) then the cgroup's reclaim daemon will have exited but
the cgroup will still exist. I think you should wait until the actual
destroy call, and synchronize with the exiting reclaim daemon.

Paul


More information about the Containers mailing list