From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [PATCH 0/7] Support for GB pages in KVM Date: Sun, 29 Mar 2009 15:03:28 +0300 Message-ID: <49CF6390.1020009@redhat.com> References: <1238164319-16092-1-git-send-email-joerg.roedel@amd.com> <20090328214008.GB4694@amt.cnet> <20090328214900.GE31080@8bytes.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: Marcelo Tosatti , Joerg Roedel , kvm@vger.kernel.org, linux-kernel@vger.kernel.org To: Joerg Roedel Return-path: Received: from mx2.redhat.com ([66.187.237.31]:39621 "EHLO mx2.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756877AbZC2MDe (ORCPT ); Sun, 29 Mar 2009 08:03:34 -0400 In-Reply-To: <20090328214900.GE31080@8bytes.org> Sender: kvm-owner@vger.kernel.org List-ID: Joerg Roedel wrote: > On Sat, Mar 28, 2009 at 06:40:08PM -0300, Marcelo Tosatti wrote: > >> On Fri, Mar 27, 2009 at 03:31:52PM +0100, Joerg Roedel wrote: >> >>> Hi, >>> >>> this patchset extends the KVM MMU implementation to support 1GB pages as >>> supported by AMD family 16 processors. These patches enable support for >>> 1 GB pages with Nested Paging. Support for these pages in the shadow >>> paging code was also developed but does not run stable yet. The patch >>> for shadow-paging support is not included in this series and will be >>> sent out seperatly. >>> >> Looks generally sane. I'm not sure its even worthwhile to support >> GBpages with softmmu, because the chance of finding an area without >> shadowed (write protected) pages is much smaller than with 2MB pages. >> > > Thanks for your review. > > The idea behind GB pages in softmmu code was to provide GB pages to the > guest even if hardware does not support it. This would work better with > live migration (Only case where we wouldn't have gbpages then would be > vmx with ept enabled). > > >> Have any numbers to share? >> > > No numbers I fully trust by now. I measured a 32% improvement in > kernbench using nested pages backed with gb pages. I will do some more > measurements and share some more solid numbers. > > Compared to 2M pages? But we're already close to native here. -- error compiling committee.c: too many arguments to function