From mboxrd@z Thu Jan 1 00:00:00 1970 From: Elena Ufimtseva Subject: Re: [PATCH v9 1/9] xen: vnuma topology and subop hypercalls Date: Thu, 4 Sep 2014 23:27:13 -0400 Message-ID: References: <1409281448-30498-1-git-send-email-ufimtseva@gmail.com> <1409281448-30498-2-git-send-email-ufimtseva@gmail.com> <54007426020000780002F043@mail.emea.novell.com> <5406ED620200007800030150@mail.emea.novell.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <5406ED620200007800030150@mail.emea.novell.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Jan Beulich Cc: Keir Fraser , Ian Campbell , Li Yechen , George Dunlap , Matt Wilson , Dario Faggioli , Stefano Stabellini , Ian Jackson , "xen-devel@lists.xen.org" List-Id: xen-devel@lists.xenproject.org On Wed, Sep 3, 2014 at 4:28 AM, Jan Beulich wrote: >>>> On 03.09.14 at 02:46, wrote: >> On Fri, Aug 29, 2014 at 6:37 AM, Jan Beulich wrote: >>>>>> On 29.08.14 at 05:04, wrote: >>>> +static struct vnuma_info *vnuma_init(const struct xen_domctl_vnuma *uinfo, >>>> + const struct domain *d) >>>> +{ >>>> + unsigned int nr_vnodes; >>>> + int i, ret = -EINVAL; >>>> + struct vnuma_info *info; >>>> + >>>> + nr_vnodes = uinfo->nr_vnodes; >>>> + >>>> + if ( nr_vnodes == 0 || nr_vnodes > uinfo->nr_vmemranges || >>> >>> Is that really a necessary check? I.e. does code elsewhere rely on >>> that? I ask because memory-less nodes are possible on real >>> hardware. >> >> That is true. But taking into account that there are no buses support >> yet added, absence of memory and buses for a vNUMA node >> seem to be useless. And vNUMA can mimic hardware NUMA as close as >> possible, but I think the degree of this is pretty much our choice. >> With further extension of vNUMA to include buses I think this check >> will naturally disappear. > > I have to admit that I struggle with the references to "buses" in your > reply. Could you perhaps give some context (not the least because > already known future extensions may call for making provisions for > them in the public interface)? > > Jan > Sorry Jan, looks like I have missed that one. I meant NUMA I/O when was referring to buses. For example from here/ http://en.community.dell.com/cfs-file.ashx/__key/telligent-evolution-components-attachments/13-4491-00-00-20-26-69-46/NUMA-for-Dell-PowerEdge-12G-Servers.pdf Or here NUMA I/O = PCIe : http://events.linuxfoundation.org/sites/events/files/eeus13_shelton.pdf In earlier discussions of vNUMA, numa I/O was mentioned but never really looked at yet from implementation point of view. -- Elena