Messages in this thread | | | Date | Fri, 16 Jan 2009 10:58:28 +0900 | From | KAMEZAWA Hiroyuki <> | Subject | Re: [PATCH 3/4] memcg: hierarchical reclaim by CSS ID |
| |
On Fri, 16 Jan 2009 09:49:05 +0800 Li Zefan <lizf@cn.fujitsu.com> wrote:
> KAMEZAWA Hiroyuki wrote: > > On Fri, 16 Jan 2009 09:29:48 +0800 > > Li Zefan <lizf@cn.fujitsu.com> wrote: > > > >>> /* > >>> - * Dance down the hierarchy if needed to reclaim memory. We remember the > >>> - * last child we reclaimed from, so that we don't end up penalizing > >>> - * one child extensively based on its position in the children list. > >>> + * Visit the first child (need not be the first child as per the ordering > >>> + * of the cgroup list, since we track last_scanned_child) of @mem and use > >>> + * that to reclaim free pages from. > >>> + */ > >>> +static struct mem_cgroup * > >>> +mem_cgroup_select_victim(struct mem_cgroup *root_mem) > >>> +{ > >>> + struct mem_cgroup *ret = NULL; > >>> + struct cgroup_subsys_state *css; > >>> + int nextid, found; > >>> + > >>> + if (!root_mem->use_hierarchy) { > >>> + spin_lock(&root_mem->reclaim_param_lock); > >>> + root_mem->scan_age++; > >>> + spin_unlock(&root_mem->reclaim_param_lock); > >>> + css_get(&root_mem->css); > >>> + ret = root_mem; > >>> + } > >>> + > >>> + while (!ret) { > >>> + rcu_read_lock(); > >>> + nextid = root_mem->last_scanned_child + 1; > >>> + css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, > >>> + &found); > >>> + if (css && css_is_populated(css) && css_tryget(css)) > >> I don't see why you need to check css_is_populated(css) ? > >> > > > > Main reason is for sanity. I don't like to hold css->refcnt of not populated css. > > I think this is a rare case. It's just a very short period when a cgroup is > being created but not yet fully created. > I don't think so. When the cgroup is mounted with several subsystems, it can call create() and populate() several times. So, memory allocation occurs between create() and populate(), it can call try_to_free_page() (of global LRU). More than that, if CONFIG_PREEMPT=y, any "short" race doesn't mean safe.
Thanks, -Kame
| |