On Mon, 26 May 2008 09:57:06 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> On Fri, 23 May 2008 20:52:29 +0900
> Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote:
> 
> > On 2008/05/22 16:37 +0900, KAMEZAWA Hiroyuki wrote:
> > > On Thu, 22 May 2008 15:20:05 +0900
> > > Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote:
> > > 
> > >> +#ifdef CONFIG_CGROUP_SWAP_RES_CTLR
> > >> +int swap_cgroup_charge(struct page *page,
> > >> +			struct swap_info_struct *si,
> > >> +			unsigned long offset)
> > >> +{
> > >> +	int ret;
> > >> +	struct page_cgroup *pc;
> > >> +	struct mem_cgroup *mem;
> > >> +
> > >> +	lock_page_cgroup(page);
> > >> +	pc = page_get_page_cgroup(page);
> > >> +	if (unlikely(!pc))
> > >> +		mem = &init_mem_cgroup;
> > >> +	else
> > >> +		mem = pc->mem_cgroup;
> > >> +	unlock_page_cgroup(page);
> > > 
> > > If !pc, the page is used before memory controller is available. But is it
> > > good to be charged to init_mem_cgroup() ?
> > I'm sorry, but I can't understand this situation.
> > memory controller is initialized at kernel initialization,
> > so aren't processes created after it is initialized?
> > 
> I think add_to_page_cache() may be called before late_init..I'll check again.
> (Because I saw some panics related to it, but I noticed this is _swap_ controller
>  ...)

_Now_, force_empty() will create a page which is used but
page->page_cgroup is NULL page. I'm now writing a  workaround (1/4 in my newest set)
but it's better to check page->page_cgroup is NULL or not.

Thanks,
-Kame




_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers