From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Wed, 20 Feb 2008 13:37:42 +0900 From: KAMEZAWA Hiroyuki Subject: Re: [RFC][PATCH] Clarify mem_cgroup lock handling and avoid races. Message-Id: <20080220133742.94a0b1b6.kamezawa.hiroyu@jp.fujitsu.com> In-Reply-To: References: <20080219215431.1aa9fa8a.kamezawa.hiroyu@jp.fujitsu.com> <20080220100333.a014083c.kamezawa.hiroyu@jp.fujitsu.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-linux-mm@kvack.org Return-Path: To: Hugh Dickins Cc: "linux-mm@kvack.org" , "balbir@linux.vnet.ibm.com" , "yamamoto@valinux.co.jp" , "riel@redhat.com" List-ID: On Wed, 20 Feb 2008 04:14:58 +0000 (GMT) Hugh Dickins wrote: > On Wed, 20 Feb 2008, KAMEZAWA Hiroyuki wrote: > > On Tue, 19 Feb 2008 15:40:45 +0000 (GMT) > > Hugh Dickins wrote: > > > > > A lot in common with yours, a lot not. (And none of it addressing > > > that issue of opt-out I raise in the last paragraph: haven't begun > > > to go into that one, hoped you and Balbir would look into it.) > > > > > I have some trial patches for reducing atomic_ops by do_it_lazy method. > > Now, I'm afraid that performence is too bad when there is *no* memory > > pressure. > > But it isn't just the atomic ops, it's the whole business of > mem_cgroup_charge_common plus mem_cgroup_uncharge_page per page. > > The existence of force_empty indicates that the system can get along > without the charge on the page. Yes. > What's needed, I think, is something in struct mm, a flag or a reserved value > in mm->mem_cgroup, to say don't do any of this mem_cgroup stuff on me; and a cgroup > fs interface to set that, in the same way as force_empty is done. I agree here. I believe we need "no charge" flag at least to the root group. For root group, it's better to have boot option if not complicated. Thanks, -Kame -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org