From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755533Ab3GKDDG (ORCPT ); Wed, 10 Jul 2013 23:03:06 -0400 Received: from mail.linuxfoundation.org ([140.211.169.12]:57220 "EHLO mail.linuxfoundation.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755229Ab3GKDDE (ORCPT ); Wed, 10 Jul 2013 23:03:04 -0400 Date: Wed, 10 Jul 2013 20:03:37 -0700 From: Andrew Morton To: Dave Chinner Cc: Michal Hocko , Glauber Costa , linux-mm@kvack.org, LKML Subject: Re: linux-next: slab shrinkers: BUG at mm/list_lru.c:92 Message-Id: <20130710200337.cd9a05d6.akpm@linux-foundation.org> In-Reply-To: <20130711022634.GZ3438@dastard> References: <20130701075005.GA28765@dhcp22.suse.cz> <20130701081056.GA4072@dastard> <20130702092200.GB16815@dhcp22.suse.cz> <20130702121947.GE14996@dastard> <20130702124427.GG16815@dhcp22.suse.cz> <20130703112403.GP14996@dastard> <20130704163643.GF7833@dhcp22.suse.cz> <20130708125352.GC20149@dhcp22.suse.cz> <20130710023138.GO3438@dastard> <20130710080605.GC4437@dhcp22.suse.cz> <20130711022634.GZ3438@dastard> X-Mailer: Sylpheed 2.7.1 (GTK+ 2.18.9; x86_64-redhat-linux-gnu) Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 11 Jul 2013 12:26:34 +1000 Dave Chinner wrote: > > Just for reference. wait_on_page_writeback is issued only for memcg > > reclaim because there is no other throttling mechanism to prevent from > > too many dirty pages on the list, thus pre-mature OOM killer. See > > e62e384e9d (memcg: prevent OOM with too many dirty pages) for more > > details. The original patch relied on may_enter_fs but that check > > disappeared by later changes by c3b94f44fc (memcg: further prevent OOM > > with too many dirty pages). > > Aye. That's the exact code I was looking at yesterday and wondering > "how the hell is waiting on page writeback valid in GFP_NOFS > context?". It seems that memcg reclaim is intentionally ignoring > GFP_NOFS to avoid OOM issues. That's a memcg implementation problem, > not a filesystem or LRU infrastructure problem.... Yup, c3b94f44fc shouldn't have done that. Throttling by waiting on a specific page is indeed prone to deadlocks and has a number of efficiency problems as well: if 1,000,000 pages came clean while you're waiting for *this* page to come clean, you're left looking pretty stupid. Hence congestion_wait(), which perhaps can save us here. I'm not sure how the wait_on_page_writeback() got back in there - I must have been asleep at the time.