From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeremy Fitzhardinge Subject: Re: Really need to pin page tables? Date: Tue, 27 May 2008 11:49:33 +0100 Message-ID: <483BE73D.3090403@goop.org> References: <20080527101704.GN4335@implementation.uk.xensource.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20080527101704.GN4335@implementation.uk.xensource.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Samuel Thibault , Xen-devel@lists.xensource.com List-Id: xen-devel@lists.xenproject.org Samuel Thibault wrote: > Hello, > > In extras/mini-os/arch/x86/mm.c:new_pt_frame, Mini-OS pins its L1, > L2, and L3 page tables. Does that really make a difference from the > Hypervisor point of view? I mean, once L4 is pinned, pointing to these, > and thus their content has been checked, is there any performance > difference? > Shouldn't be. Pinning an L4 implicitly pins everything else below it. The only reason to pin the leafy parts of a pagetable is if you want to play games with incrementally pinning the pagetable, or if you want to pull them apart and rearrange the pieces for some reason. For example, I do incremental pagetable pins in the Xen/pvops kernel to limit the number of pte locks I need to hold at any one time. J