From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753528Ab3A3DyA (ORCPT ); Tue, 29 Jan 2013 22:54:00 -0500 Received: from e23smtp01.au.ibm.com ([202.81.31.143]:53667 "EHLO e23smtp01.au.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753033Ab3A3Dx7 (ORCPT ); Tue, 29 Jan 2013 22:53:59 -0500 Message-ID: <5108994E.6060506@linux.vnet.ibm.com> Date: Wed, 30 Jan 2013 11:53:50 +0800 From: Xiao Guangrong User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/17.0 Thunderbird/17.0 MIME-Version: 1.0 To: Marcelo Tosatti CC: Avi Kivity , Gleb Natapov , LKML , KVM Subject: Re: [PATCH v2 06/12] KVM: MMU: introduce a static table to map guest access to spte access References: <50FFB5A1.5090708@linux.vnet.ibm.com> <50FFB658.6040205@linux.vnet.ibm.com> <20130125001550.GB22875@amt.cnet> <5101F207.6040603@linux.vnet.ibm.com> <20130129000715.GA10814@amt.cnet> <20130129010758.GA12132@amt.cnet> In-Reply-To: <20130129010758.GA12132@amt.cnet> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit X-Content-Scanned: Fidelis XPS MAILER x-cbid: 13013003-1618-0000-0000-000003439B24 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 01/29/2013 09:07 AM, Marcelo Tosatti wrote: > On Mon, Jan 28, 2013 at 10:07:15PM -0200, Marcelo Tosatti wrote: >> On Fri, Jan 25, 2013 at 10:46:31AM +0800, Xiao Guangrong wrote: >>> On 01/25/2013 08:15 AM, Marcelo Tosatti wrote: >>>> On Wed, Jan 23, 2013 at 06:07:20PM +0800, Xiao Guangrong wrote: >>>>> It makes set_spte more clean and reduces branch prediction >>>>> >>>>> Signed-off-by: Xiao Guangrong >>>>> --- >>>>> arch/x86/kvm/mmu.c | 37 ++++++++++++++++++++++++++----------- >>>>> 1 files changed, 26 insertions(+), 11 deletions(-) >>>> >>>> Don't see set_spte as being a performance problem? >>>> IMO the current code is quite simple. >>> >>> Yes, this is not a performance problem. >>> >>> I just dislike this many continuous "if"-s in the function: >>> >>> if (xxx) >>> xxx >>> if (xxx) >>> xxx >>> .... >>> >>> Totally, it has 7 "if"-s before this patch. >>> >>> Okay, if you think this is unnecessary, i will drop this patch. :) >> >> Yes, please (unless you can show set_spte is a performance problem). > > Same thing for spte fast drop: is it a performance problem? It does not fix a performance problem. The patch does optimization on the noncrucial path and cleanup for the previous patch. > > Please try to group changes into smaller, less controversial sets with > a clear goal: Ah. I thought they are simple and most of them have been reviewed, so i putted them into one group, now i know i made a mistake. ;) > > - Debated performance improvement. > - Cleanups (eg mmu_set_spte argument removal). > - Bug fixes. > - Performance improvements. Okay. I will split the long series into these set: Set 1 for improvement: [PATCH v2 01/12] KVM: MMU: lazily drop large spte Set 2 for cleanups: [PATCH v2 02/12] KVM: MMU: cleanup mapping-level [PATCH v2 07/12] KVM: MMU: remove pt_access in mmu_set_spte [PATCH v2 08/12] KVM: MMU: cleanup __direct_map Set 3 for simplifying set spte path: [PATCH v2 03/12] KVM: MMU: simplify mmu_set_spte [PATCH v2 04/12] KVM: MMU: simplify set_spte [PATCH v2 05/12] KVM: MMU: introduce vcpu_adjust_access Set 4 for fixing and unifying rmap walking [PATCH v2 11/12] KVM: MMU: fix spte assertion [PATCH v2 09/12] KVM: MMU: introduce mmu_spte_establish [PATCH v2 10/12] KVM: MMU: unify the code of walking pte list [PATCH v2 12/12] KVM: MMU: fast drop all spte on the pte_list Thanks!