From mboxrd@z Thu Jan 1 00:00:00 1970 From: Balbir Singh Subject: Re: [-mm] Add an owner to the mm_struct (v7) Date: Thu, 03 Apr 2008 23:41:51 +0530 Message-ID: <47F51DE7.7010204@linux.vnet.ibm.com> References: <20080403174433.26356.42121.sendpatchset@localhost.localdomain> <6599ad830804031058l1e2a7ad9p56cff47dca738d79@mail.gmail.com> Reply-To: balbir@linux.vnet.ibm.com Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <6599ad830804031058l1e2a7ad9p56cff47dca738d79@mail.gmail.com> Sender: linux-kernel-owner@vger.kernel.org To: Paul Menage Cc: Pavel Emelianov , Hugh Dickins , Sudhir Kumar , YAMAMOTO Takashi , lizf@cn.fujitsu.com, linux-kernel@vger.kernel.org, taka@valinux.co.jp, linux-mm@kvack.org, David Rientjes , Andrew Morton , KAMEZAWA Hiroyuki List-Id: linux-mm.kvack.org Paul Menage wrote: > On Thu, Apr 3, 2008 at 10:44 AM, Balbir Singh wrote: >> + >> + /* >> + * If there are other users of the mm and the owner (us) is exiting >> + * we need to find a new owner to take on the responsibility. >> + * When we use thread groups (CLONE_THREAD), the thread group >> + * leader is kept around in zombie state, even after it exits. >> + * delay_group_leader() ensures that if the group leader is around >> + * we need not select a new owner. >> + */ > > Hmm, is this new check for delay_group_leader() safe? Won't we have > called exit_cgroup() by this point, and hence be reassigned to the > root cgroup? And so mm->owner->cgroups won't point to the right place? > cgroup_exit() comes in much later after exit_mm(). Moreover delay_group_leader() is a function that checks to see if We are the group leader and the thread group is not empty. -- Warm Regards, Balbir Singh Linux Technology Center IBM, ISTL