From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753701Ab3FDEln (ORCPT ); Tue, 4 Jun 2013 00:41:43 -0400 Received: from LGEMRELSE6Q.lge.com ([156.147.1.121]:50406 "EHLO LGEMRELSE6Q.lge.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751044Ab3FDElm (ORCPT ); Tue, 4 Jun 2013 00:41:42 -0400 X-AuditID: 9c930179-b7c07ae000000e33-db-51ad7003b85e Date: Tue, 4 Jun 2013 13:41:39 +0900 From: Minchan Kim To: Dave Hansen Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org, akpm@linux-foundation.org, mgorman@suse.de, tim.c.chen@linux.intel.com Subject: Re: [v4][PATCH 1/6] mm: swap: defer clearing of page_private() for swap cache pages Message-ID: <20130604044139.GB14719@blaptop> References: <20130531183855.44DDF928@viggo.jf.intel.com> <20130531183856.1D7D75AD@viggo.jf.intel.com> <20130603054048.GA27858@blaptop> <51ACADCD.70904@sr71.net> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <51ACADCD.70904@sr71.net> User-Agent: Mutt/1.5.21 (2010-09-15) X-Brightmail-Tracker: AAAAAA== Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, Jun 03, 2013 at 07:53:01AM -0700, Dave Hansen wrote: > On 06/02/2013 10:40 PM, Minchan Kim wrote: > >> > diff -puN mm/vmscan.c~__delete_from_swap_cache-dont-clear-page-private mm/vmscan.c > >> > --- linux.git/mm/vmscan.c~__delete_from_swap_cache-dont-clear-page-private 2013-05-30 16:07:50.632079492 -0700 > >> > +++ linux.git-davehans/mm/vmscan.c 2013-05-30 16:07:50.637079712 -0700 > >> > @@ -494,6 +494,8 @@ static int __remove_mapping(struct addre > >> > __delete_from_swap_cache(page); > >> > spin_unlock_irq(&mapping->tree_lock); > >> > swapcache_free(swap, page); > >> > + set_page_private(page, 0); > >> > + ClearPageSwapCache(page); > > It it worth to support non-atomic version of ClearPageSwapCache? > > Just for this, probably not. > > It does look like a site where it would be theoretically safe to use > non-atomic flag operations since the page is on a one-way trip to the > allocator at this point and the __clear_page_locked() now happens _just_ > after this code. True. > > But, personally, I'm happy to leave it as-is. The atomic vs. non-atomic > flags look to me like a micro-optimization that we should use when we > _know_ there will be some tangible benefit. Otherwise, they're just > something extra for developers to trip over and cause very subtle bugs. I just asked it because when I read the description of patchset, I felt you were very sensitive to the atomic operation on many CPU system with several sockets. Anyway, if you don't want it, I don't mind it, either. -- Kind regards, Minchan Kim