From mboxrd@z Thu Jan 1 00:00:00 1970 From: Chris Friesen Subject: looking for info on TSC virtualization with kvm Date: Thu, 07 Aug 2014 12:27:50 -0600 Message-ID: <53E3C526.1060007@mail.usask.ca> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit To: kvm@vger.kernel.org Return-path: Received: from idcmail-mo2no.shaw.ca ([64.59.134.9]:16949 "EHLO idcmail-mo2no.shaw.ca" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755050AbaHGS1w (ORCPT ); Thu, 7 Aug 2014 14:27:52 -0400 Sender: kvm-owner@vger.kernel.org List-ID: I'm trying to find out some hard data on TSC virtualization when using qemu-kvm to run linux guests on Intel-based linux 3.4 hosts. I've read "https://www.kernel.org/doc/Documentation/virtual/kvm/timekeeping.txt" and looked at the code somewhat, but it's a bit tricky to figure out exactly what will happen. I'm using kvm_clock for the guest kernel clocksource, but I'm concerned about applications in the guest accessing the TSC directly. My main concern is what happens during live migration to a host with a faster TSC frequency (even though both hosts may have nonstop constant TSCs). It appears that VMX provides hardware virtualization of the TSC to allow an offset to be applied, but no support for frequency scaling. Under what circumstances would KVM use the hardware virtualized TSC? What would happen on migration to a host with faster TSC? Does KVM support a fully-emulated software TSC? If so, under what circumstances would it be used and can it handle frequency shifts and offsets over live migration? Lastly...I found a patch series by Zachary Amsden at "http://thread.gmane.org/gmane.linux.kernel/1157689" that doesn't seem to have ever made it in to the mainline kernel...why not? Thanks, Chris