Messages in this thread | | | Date | Tue, 23 Jun 2009 13:54:05 +0900 | From | Daisuke Nishimura <> | Subject | Re: [RFC][PATCH] cgroup: fix permanent wait in rmdir |
| |
On Tue, 23 Jun 2009 13:44:20 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote: > On Tue, 23 Jun 2009 13:13:33 +0900 > Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote: > > > On Tue, 23 Jun 2009 09:22:23 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote: > > > On Mon, 22 Jun 2009 18:37:07 +0900 > > > KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote: > > > > > > > previous discussion was this => http://marc.info/?t=124478543600001&r=1&w=2 > > > > > > > > I think this is a minimum fix (in code size and behavior) and because > > > > we can take a BIG LOCK, this kind of check is necessary, anyway. > > > > Any comments are welcome. > > > > > > I'll split this into 2 patches...and I found I should check page-migration, too. > > I'll wait a new version, but can you explain in advance this page-migration case ? > > > > Not far from swap-in case. > > Assume cgroup "A" which includes file caches. A task in other group mmap file caches > and do page migration and rmdir against "A" is called at the same time. > > In mem_cgroup_prepare_migration(), following check is used. > > == > lock_page_cgroup(pc); > if (PageCgroupUsed(pc)) { > mem = pc->mem_cgroup; > css_get(&mem->css); > } > unlock_page_cgroup(pc); > <======================================(*) > if (mem) { > <==============================(**) > try_charge(); > ... > } > == > > At (*), we grab css refcnt which can be under pre_destroy() and > At (**), pre_destroy may returns 0 but charge may be done after the end of pre_destroy(). > Ah I see, you're right.
Thank you for your clarification.
Daisuke Nishimura.
> > > > > +static int mem_cgroup_retry_rmdir(struct cgroup_subsys *ss, > > > > + struct cgroup *cont) > > > > +{ > > > > + struct mem_cgroup *mem = mem_cgroup_from_cont(cont); > > > > + > > > > + if (res_counter_read_u64(&memcg->res, RES_USAGE)) > > It should be &mem->res. > > > yes. > too many typos in my patches in these days.. > > Thanks, > -Kame >
| |