From mboxrd@z Thu Jan 1 00:00:00 1970 From: Alexander Graf Subject: Re: [PATCH v2] powernv: kvm: make _PAGE_NUMA take effect Date: Thu, 10 Apr 2014 12:02:34 +0200 Message-ID: <53466C3A.5070204@suse.de> References: <1390292129-15871-1-git-send-email-pingfank@linux.vnet.ibm.com> <8761pdk6x5.fsf@linux.vnet.ibm.com> <533D47CD.906@suse.com> <87bnwdshzu.fsf@linux.vnet.ibm.com> <53426385.2010102@suse.de> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Cc: "Aneesh Kumar K.V" , Alexander Graf , linuxppc-dev@lists.ozlabs.org, kvm-ppc , kvm-devel , Ben Herrenschmidt , Paul Mackerras To: Liu ping fan Return-path: In-Reply-To: Sender: kvm-ppc-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On 10.04.14 05:28, Liu ping fan wrote: > On Mon, Apr 7, 2014 at 4:36 PM, Alexander Graf wrote: >> On 07.04.14 09:42, Aneesh Kumar K.V wrote: >>> Alexander Graf writes: >>> >>>> On 03.04.14 04:36, Liu ping fan wrote: >>>>> Hi Alex, could you help to pick up this patch? since v3.14 kernel can >>>>> enable numa fault for powerpc. >>>> What bad happens without this patch? We map a page even though it was >>>> declared to get NUMA migrated? What happens next? >>> Nothing much, we won't be properly accounting the numa access in the >>> host. What we want to achieve is to convert a guest access of the page to >>> a host fault so that we can do proper numa access accounting in the >>> host. This would enable us to migrate the page to the correct numa >>> node. >> >> Ok, so no breakages, just less performance. I wouldn't consider it stable >> material then :). >> > Sorry to reply late, since I am half out of office during this period. > I am puzzling about you reply. Without this patch, the guest can > NOT sense the numa changes and ask host to put the pages in right > place. So the pages which is used by guest will be always misplaced. > The numa-fault method is inspired by real requirement to improve > performance, so we should also consider the performance drop of guest. > Right? The patch will get into Linux, I just consider a non-working new feature not a regression that warrants us to CC stable@vger :). After all performance shouldn't be worse than without the numa migration feature, correct? Alex