From mboxrd@z Thu Jan 1 00:00:00 1970 From: KOSAKI Motohiro Subject: Re: [PATCH v3] mm: make expand_downwards symmetrical to expand_upwards Date: Wed, 20 Apr 2011 16:15:23 +0900 (JST) Message-ID: <20110420161615.462D.A69D9226@jp.fujitsu.com> References: <20110420102314.4604.A69D9226@jp.fujitsu.com> Mime-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Cc: kosaki.motohiro@jp.fujitsu.com, Christoph Lameter , James Bottomley , Michal Hocko , Andrew Morton , Hugh Dickins , linux-mm@kvack.org, LKML , linux-parisc@vger.kernel.org, David Rientjes To: Pekka Enberg Return-path: In-Reply-To: List-ID: List-Id: linux-parisc.vger.kernel.org > On Wed, Apr 20, 2011 at 4:23 AM, KOSAKI Motohiro > wrote: > > I'm worry about this patch. A lot of mm code assume !NUMA systems > > only have node 0. Not only SLUB. > > So is that a valid assumption or not? Christoph seems to think it is > and James seems to think it's not. Which way should we aim to fix it? > Would be nice if other people chimed in as we already know what James > and Christoph think. I'm sorry. I don't know it really. The fact was gone into historical myst. ;-) Now, CONFIG_NUMA has mainly five meanings. 1) system may has !0 node id. 2) compile mm/mempolicy.c (ie enable mempolicy APIs) 3) Allocator (kmalloc, vmalloc, alloc_page, et al) awake NUMA topology. 4) enable zone-reclaim feature 5) scheduler makes per-node load balancing scheduler domain Anyway, we have to fix this issue. I'm digging which fixing way has least risk. btw, x86 don't have an issue. Probably it's a reason why this issue was neglected long time. arch/x86/Kconfig ------------------------------------- config ARCH_DISCONTIGMEM_ENABLE def_bool y depends on NUMA && X86_32