From mboxrd@z Thu Jan 1 00:00:00 1970 From: Marcelo Tosatti Subject: Re: [patch 4/4] KVM: allow direct access to PMTimer port Date: Sun, 25 May 2008 13:12:56 -0300 Message-ID: <20080525161256.GB16012@dmt> References: <20080524234342.983197667@localhost.localdomain> <20080525000036.710955693@localhost.localdomain> <48395C07.1050005@qumranet.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Chris Wright , kvm-devel , Len Brown To: Avi Kivity Return-path: Received: from mx1.redhat.com ([66.187.233.31]:34839 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755650AbYEYQNW (ORCPT ); Sun, 25 May 2008 12:13:22 -0400 Content-Disposition: inline In-Reply-To: <48395C07.1050005@qumranet.com> Sender: kvm-owner@vger.kernel.org List-ID: On Sun, May 25, 2008 at 03:31:03PM +0300, Avi Kivity wrote: > Marcelo Tosatti wrote: > >There's not much point in exiting for pmtimer reads, since it runs at a > >fixed clock rate and its start value is undefined. > > > > On a migration farm with lots of different hardware, eventually all > guests will have migrated and none of them will be able to use the > passthrough port. > > What about emulating pmtimer in the host kernel instead of userspace? > It increases the hit from 1 exit to 3 exits, but they're much faster exits. I agree, even began writing it, but I think the non-passthrough case could be optimized later.