From: KAMEZAWA Hiroyuki Subject: Re: [PATCH 1/7] res_counter: add a function res_counter_move_parent(). Date: Fri, 13 Apr 2012 09:57:03 +0900 Message-ID: <4F8779DF.3080307@jp.fujitsu.com> References: <4F86B9BE.8000105@jp.fujitsu.com> <4F86BA66.2010503@jp.fujitsu.com> <4F86D733.50809@parallels.com> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: Sender: cgroups-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-ID: Content-Type: text/plain; charset="us-ascii" To: Frederic Weisbecker Cc: Glauber Costa , "linux-mm-Bw31MaZKKs3YtjvyW6yDsg@public.gmane.org" , "cgroups-u79uwXL29TY76Z2rM5mHXA@public.gmane.org" , Michal Hocko , Johannes Weiner , Tejun Heo , Hugh Dickins , Andrew Morton (2012/04/12 23:30), Frederic Weisbecker wrote: > 2012/4/12 Glauber Costa : >> On 04/12/2012 08:20 AM, KAMEZAWA Hiroyuki wrote: >>> >>> This function is used for moving accounting information to its >>> parent in the hierarchy of res_counter. >>> >>> Signed-off-by: KAMEZAWA Hiroyuki >> >> Frederic has a patch in his fork cgroup series, that allows you to >> uncharge a counter until you reach a specific ancestor. >> You pass the parent as a parameter, and then only you gets uncharged. > > I'm missing the referring patchset from Kamezawa. Ok I'm going to > subscribe to the > cgroup mailing list. Meanwhile perhaps would it be nice to keep Cc > LKML for cgroup patches? > Ah, sorry. I will do next time. > Some comments below: > >> >> I think that is a much better interface than this you are proposing. >> We should probably merge that patch and use it. >> >>> --- >>> include/linux/res_counter.h | 3 +++ >>> kernel/res_counter.c | 13 +++++++++++++ >>> 2 files changed, 16 insertions(+), 0 deletions(-) >>> >>> diff --git a/include/linux/res_counter.h b/include/linux/res_counter.h >>> index da81af0..8919d3c 100644 >>> --- a/include/linux/res_counter.h >>> +++ b/include/linux/res_counter.h >>> @@ -135,6 +135,9 @@ int __must_check res_counter_charge_nofail(struct res_counter *counter, >>> void res_counter_uncharge_locked(struct res_counter *counter, unsigned long val); >>> void res_counter_uncharge(struct res_counter *counter, unsigned long val); >>> >>> +/* move resource to parent counter...i.e. just forget accounting in a child */ >>> +void res_counter_move_parent(struct res_counter *counter, unsigned long val); >>> + >>> /** >>> * res_counter_margin - calculate chargeable space of a counter >>> * @cnt: the counter >>> diff --git a/kernel/res_counter.c b/kernel/res_counter.c >>> index d508363..fafebf0 100644 >>> --- a/kernel/res_counter.c >>> +++ b/kernel/res_counter.c >>> @@ -113,6 +113,19 @@ void res_counter_uncharge(struct res_counter *counter, unsigned long val) >>> local_irq_restore(flags); >>> } >>> >>> +/* >>> + * In hierarchical accounting, child's usage is accounted into ancestors. >>> + * To move local usage to its parent, just forget current level usage. > > The way I understand this comment and the changelog matches the opposite > of what the below function is doing. > > The function charges a child and ignore all its parents. The comments says it > charges the parents but not the child. > Sure, I'll fix...and look into your code first. "uncharge a counter until you reach a specific ancestor"... Is it in linux-next ? Thanks, -Kame >>> + */ >>> +void res_counter_move_parent(struct res_counter *counter, unsigned long val) >>> +{ >>> + unsigned long flags; >>> + >>> + BUG_ON(!counter->parent); >>> + spin_lock_irqsave(&counter->lock, flags); >>> + res_counter_uncharge_locked(counter, val); >>> + spin_unlock_irqrestore(&counter->lock, flags); >>> +} >>> >>> static inline unsigned long long * >>> res_counter_member(struct res_counter *counter, int member) >> > >