From mboxrd@z Thu Jan 1 00:00:00 1970 From: Gleb Natapov Subject: Re: [PATCH v3 1/7] KVM: MMU: fix release noslot pfn Date: Mon, 24 Sep 2012 14:04:24 +0200 Message-ID: <20120924120424.GB23096@redhat.com> References: <505C0FA8.5070007@linux.vnet.ibm.com> <505C0FCF.2070308@linux.vnet.ibm.com> <20120923091335.GA20907@redhat.com> <505FE8B4.8030309@linux.vnet.ibm.com> <20120924112416.GA23096@redhat.com> <506048D1.70403@linux.vnet.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Avi Kivity , Marcelo Tosatti , LKML , KVM To: Xiao Guangrong Return-path: Received: from mx1.redhat.com ([209.132.183.28]:6645 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754906Ab2IXME3 (ORCPT ); Mon, 24 Sep 2012 08:04:29 -0400 Content-Disposition: inline In-Reply-To: <506048D1.70403@linux.vnet.ibm.com> Sender: kvm-owner@vger.kernel.org List-ID: On Mon, Sep 24, 2012 at 07:49:37PM +0800, Xiao Guangrong wrote: > On 09/24/2012 07:24 PM, Gleb Natapov wrote: > > On Mon, Sep 24, 2012 at 12:59:32PM +0800, Xiao Guangrong wrote: > >> On 09/23/2012 05:13 PM, Gleb Natapov wrote: > >>> On Fri, Sep 21, 2012 at 02:57:19PM +0800, Xiao Guangrong wrote: > >>>> We can not directly call kvm_release_pfn_clean to release the pfn > >>>> since we can meet noslot pfn which is used to cache mmio info into > >>>> spte > >>>> > >>> Wouldn't it be better to move the check into kvm_release_pfn_clean()? > >> > >> I think there is no reason for us to prefer to adding this branch in > >> the common code. :) > > > > Is the function performance critical? Is function called without the check > > on a hot path? The function already contains much heavier kvm_is_mmio_pfn() > > check. If most/all function invocation require check before call it's > > better to move it inside. > > It is not most/all functions need do this check - it is only needed on x86 mmu > page-fault/prefetch path. At least on x86 there 7 calls to kvm_release_pfn_clean(), 5 of them are guarded by is_noslot_pfn() (after this patch) and one by even stronger is_error_pfn(). I guess when/if other architectures will add MMIO MMU caching they will need to guard kvm_release_pfn_clean() by is_noslot_pfn() too in most cases. I am not insisting, but as this patch shows it is easy to miss the check before calling the function. -- Gleb.